/
dsl_scan.c
4140 lines (3585 loc) · 125 KB
/
dsl_scan.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2011, 2017 by Delphix. All rights reserved.
* Copyright 2016 Gary Mills
* Copyright (c) 2017 Datto Inc.
* Copyright 2017 Joyent, Inc.
*/
#include <sys/dsl_scan.h>
#include <sys/dsl_pool.h>
#include <sys/dsl_dataset.h>
#include <sys/dsl_prop.h>
#include <sys/dsl_dir.h>
#include <sys/dsl_synctask.h>
#include <sys/dnode.h>
#include <sys/dmu_tx.h>
#include <sys/dmu_objset.h>
#include <sys/arc.h>
#include <sys/zap.h>
#include <sys/zio.h>
#include <sys/zfs_context.h>
#include <sys/fs/zfs.h>
#include <sys/zfs_znode.h>
#include <sys/spa_impl.h>
#include <sys/vdev_impl.h>
#include <sys/zil_impl.h>
#include <sys/zio_checksum.h>
#include <sys/ddt.h>
#include <sys/sa.h>
#include <sys/sa_impl.h>
#include <sys/zfeature.h>
#include <sys/abd.h>
#include <sys/range_tree.h>
#ifdef _KERNEL
#include <sys/zfs_vfsops.h>
#endif
/*
* Grand theory statement on scan queue sorting
*
* Scanning is implemented by recursively traversing all indirection levels
* in an object and reading all blocks referenced from said objects. This
* results in us approximately traversing the object from lowest logical
* offset to the highest. For best performance, we would want the logical
* blocks to be physically contiguous. However, this is frequently not the
* case with pools given the allocation patterns of copy-on-write filesystems.
* So instead, we put the I/Os into a reordering queue and issue them in a
* way that will most benefit physical disks (LBA-order).
*
* Queue management:
*
* Ideally, we would want to scan all metadata and queue up all block I/O
* prior to starting to issue it, because that allows us to do an optimal
* sorting job. This can however consume large amounts of memory. Therefore
* we continuously monitor the size of the queues and constrain them to 5%
* (zfs_scan_mem_lim_fact) of physmem. If the queues grow larger than this
* limit, we clear out a few of the largest extents at the head of the queues
* to make room for more scanning. Hopefully, these extents will be fairly
* large and contiguous, allowing us to approach sequential I/O throughput
* even without a fully sorted tree.
*
* Metadata scanning takes place in dsl_scan_visit(), which is called from
* dsl_scan_sync() every spa_sync(). If we have either fully scanned all
* metadata on the pool, or we need to make room in memory because our
* queues are too large, dsl_scan_visit() is postponed and
* scan_io_queues_run() is called from dsl_scan_sync() instead. This implies
* that metadata scanning and queued I/O issuing are mutually exclusive. This
* allows us to provide maximum sequential I/O throughput for the majority of
* I/O's issued since sequential I/O performance is significantly negatively
* impacted if it is interleaved with random I/O.
*
* Implementation Notes
*
* One side effect of the queued scanning algorithm is that the scanning code
* needs to be notified whenever a block is freed. This is needed to allow
* the scanning code to remove these I/Os from the issuing queue. Additionally,
* we do not attempt to queue gang blocks to be issued sequentially since this
* is very hard to do and would have an extremely limited performance benefit.
* Instead, we simply issue gang I/Os as soon as we find them using the legacy
* algorithm.
*
* Backwards compatibility
*
* This new algorithm is backwards compatible with the legacy on-disk data
* structures (and therefore does not require a new feature flag).
* Periodically during scanning (see zfs_scan_checkpoint_intval), the scan
* will stop scanning metadata (in logical order) and wait for all outstanding
* sorted I/O to complete. Once this is done, we write out a checkpoint
* bookmark, indicating that we have scanned everything logically before it.
* If the pool is imported on a machine without the new sorting algorithm,
* the scan simply resumes from the last checkpoint using the legacy algorithm.
*/
typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *,
const zbookmark_phys_t *);
static scan_cb_t dsl_scan_scrub_cb;
static int scan_ds_queue_compare(const void *a, const void *b);
static int scan_prefetch_queue_compare(const void *a, const void *b);
static void scan_ds_queue_clear(dsl_scan_t *scn);
static void scan_ds_prefetch_queue_clear(dsl_scan_t *scn);
static boolean_t scan_ds_queue_contains(dsl_scan_t *scn, uint64_t dsobj,
uint64_t *txg);
static void scan_ds_queue_insert(dsl_scan_t *scn, uint64_t dsobj, uint64_t txg);
static void scan_ds_queue_remove(dsl_scan_t *scn, uint64_t dsobj);
static void scan_ds_queue_sync(dsl_scan_t *scn, dmu_tx_t *tx);
static uint64_t dsl_scan_count_leaves(vdev_t *vd);
extern int zfs_vdev_async_write_active_min_dirty_percent;
/*
* By default zfs will check to ensure it is not over the hard memory
* limit before each txg. If finer-grained control of this is needed
* this value can be set to 1 to enable checking before scanning each
* block.
*/
int zfs_scan_strict_mem_lim = B_FALSE;
/*
* Maximum number of parallelly executed bytes per leaf vdev. We attempt
* to strike a balance here between keeping the vdev queues full of I/Os
* at all times and not overflowing the queues to cause long latency,
* which would cause long txg sync times. No matter what, we will not
* overload the drives with I/O, since that is protected by
* zfs_vdev_scrub_max_active.
*/
unsigned long zfs_scan_vdev_limit = 4 << 20;
int zfs_scan_issue_strategy = 0;
int zfs_scan_legacy = B_FALSE; /* don't queue & sort zios, go direct */
unsigned long zfs_scan_max_ext_gap = 2 << 20; /* in bytes */
/*
* fill_weight is non-tunable at runtime, so we copy it at module init from
* zfs_scan_fill_weight. Runtime adjustments to zfs_scan_fill_weight would
* break queue sorting.
*/
int zfs_scan_fill_weight = 3;
static uint64_t fill_weight;
/* See dsl_scan_should_clear() for details on the memory limit tunables */
uint64_t zfs_scan_mem_lim_min = 16 << 20; /* bytes */
uint64_t zfs_scan_mem_lim_soft_max = 128 << 20; /* bytes */
int zfs_scan_mem_lim_fact = 20; /* fraction of physmem */
int zfs_scan_mem_lim_soft_fact = 20; /* fraction of mem lim above */
int zfs_scrub_min_time_ms = 1000; /* min millisecs to scrub per txg */
int zfs_obsolete_min_time_ms = 500; /* min millisecs to obsolete per txg */
int zfs_free_min_time_ms = 1000; /* min millisecs to free per txg */
int zfs_resilver_min_time_ms = 3000; /* min millisecs to resilver per txg */
int zfs_scan_checkpoint_intval = 7200; /* in seconds */
int zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */
int zfs_no_scrub_prefetch = B_FALSE; /* set to disable scrub prefetch */
enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE;
/* max number of blocks to free in a single TXG */
unsigned long zfs_async_block_max_blocks = 100000;
int zfs_resilver_disable_defer = 0; /* set to disable resilver deferring */
/*
* We wait a few txgs after importing a pool to begin scanning so that
* the import / mounting code isn't held up by scrub / resilver IO.
* Unfortunately, it is a bit difficult to determine exactly how long
* this will take since userspace will trigger fs mounts asynchronously
* and the kernel will create zvol minors asynchronously. As a result,
* the value provided here is a bit arbitrary, but represents a
* reasonable estimate of how many txgs it will take to finish fully
* importing a pool
*/
#define SCAN_IMPORT_WAIT_TXGS 5
#define DSL_SCAN_IS_SCRUB_RESILVER(scn) \
((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \
(scn)->scn_phys.scn_func == POOL_SCAN_RESILVER)
/*
* Enable/disable the processing of the free_bpobj object.
*/
int zfs_free_bpobj_enabled = 1;
/* the order has to match pool_scan_type */
static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = {
NULL,
dsl_scan_scrub_cb, /* POOL_SCAN_SCRUB */
dsl_scan_scrub_cb, /* POOL_SCAN_RESILVER */
};
/* In core node for the scn->scn_queue. Represents a dataset to be scanned */
typedef struct {
uint64_t sds_dsobj;
uint64_t sds_txg;
avl_node_t sds_node;
} scan_ds_t;
/*
* This controls what conditions are placed on dsl_scan_sync_state():
* SYNC_OPTIONAL) write out scn_phys iff scn_bytes_pending == 0
* SYNC_MANDATORY) write out scn_phys always. scn_bytes_pending must be 0.
* SYNC_CACHED) if scn_bytes_pending == 0, write out scn_phys. Otherwise
* write out the scn_phys_cached version.
* See dsl_scan_sync_state for details.
*/
typedef enum {
SYNC_OPTIONAL,
SYNC_MANDATORY,
SYNC_CACHED
} state_sync_type_t;
/*
* This struct represents the minimum information needed to reconstruct a
* zio for sequential scanning. This is useful because many of these will
* accumulate in the sequential IO queues before being issued, so saving
* memory matters here.
*/
typedef struct scan_io {
/* fields from blkptr_t */
uint64_t sio_offset;
uint64_t sio_blk_prop;
uint64_t sio_phys_birth;
uint64_t sio_birth;
zio_cksum_t sio_cksum;
uint32_t sio_asize;
/* fields from zio_t */
int sio_flags;
zbookmark_phys_t sio_zb;
/* members for queue sorting */
union {
avl_node_t sio_addr_node; /* link into issueing queue */
list_node_t sio_list_node; /* link for issuing to disk */
} sio_nodes;
} scan_io_t;
struct dsl_scan_io_queue {
dsl_scan_t *q_scn; /* associated dsl_scan_t */
vdev_t *q_vd; /* top-level vdev that this queue represents */
/* trees used for sorting I/Os and extents of I/Os */
range_tree_t *q_exts_by_addr;
avl_tree_t q_exts_by_size;
avl_tree_t q_sios_by_addr;
/* members for zio rate limiting */
uint64_t q_maxinflight_bytes;
uint64_t q_inflight_bytes;
kcondvar_t q_zio_cv; /* used under vd->vdev_scan_io_queue_lock */
/* per txg statistics */
uint64_t q_total_seg_size_this_txg;
uint64_t q_segs_this_txg;
uint64_t q_total_zio_size_this_txg;
uint64_t q_zios_this_txg;
};
/* private data for dsl_scan_prefetch_cb() */
typedef struct scan_prefetch_ctx {
zfs_refcount_t spc_refcnt; /* refcount for memory management */
dsl_scan_t *spc_scn; /* dsl_scan_t for the pool */
boolean_t spc_root; /* is this prefetch for an objset? */
uint8_t spc_indblkshift; /* dn_indblkshift of current dnode */
uint16_t spc_datablkszsec; /* dn_idatablkszsec of current dnode */
} scan_prefetch_ctx_t;
/* private data for dsl_scan_prefetch() */
typedef struct scan_prefetch_issue_ctx {
avl_node_t spic_avl_node; /* link into scn->scn_prefetch_queue */
scan_prefetch_ctx_t *spic_spc; /* spc for the callback */
blkptr_t spic_bp; /* bp to prefetch */
zbookmark_phys_t spic_zb; /* bookmark to prefetch */
} scan_prefetch_issue_ctx_t;
static void scan_exec_io(dsl_pool_t *dp, const blkptr_t *bp, int zio_flags,
const zbookmark_phys_t *zb, dsl_scan_io_queue_t *queue);
static void scan_io_queue_insert_impl(dsl_scan_io_queue_t *queue,
scan_io_t *sio);
static dsl_scan_io_queue_t *scan_io_queue_create(vdev_t *vd);
static void scan_io_queues_destroy(dsl_scan_t *scn);
static kmem_cache_t *sio_cache;
void
scan_init(void)
{
/*
* This is used in ext_size_compare() to weight segments
* based on how sparse they are. This cannot be changed
* mid-scan and the tree comparison functions don't currently
* have a mechanism for passing additional context to the
* compare functions. Thus we store this value globally and
* we only allow it to be set at module initialization time
*/
fill_weight = zfs_scan_fill_weight;
sio_cache = kmem_cache_create("sio_cache",
sizeof (scan_io_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
}
void
scan_fini(void)
{
kmem_cache_destroy(sio_cache);
}
static inline boolean_t
dsl_scan_is_running(const dsl_scan_t *scn)
{
return (scn->scn_phys.scn_state == DSS_SCANNING);
}
boolean_t
dsl_scan_resilvering(dsl_pool_t *dp)
{
return (dsl_scan_is_running(dp->dp_scan) &&
dp->dp_scan->scn_phys.scn_func == POOL_SCAN_RESILVER);
}
static inline void
sio2bp(const scan_io_t *sio, blkptr_t *bp, uint64_t vdev_id)
{
bzero(bp, sizeof (*bp));
DVA_SET_ASIZE(&bp->blk_dva[0], sio->sio_asize);
DVA_SET_VDEV(&bp->blk_dva[0], vdev_id);
DVA_SET_OFFSET(&bp->blk_dva[0], sio->sio_offset);
bp->blk_prop = sio->sio_blk_prop;
bp->blk_phys_birth = sio->sio_phys_birth;
bp->blk_birth = sio->sio_birth;
bp->blk_fill = 1; /* we always only work with data pointers */
bp->blk_cksum = sio->sio_cksum;
}
static inline void
bp2sio(const blkptr_t *bp, scan_io_t *sio, int dva_i)
{
/* we discard the vdev id, since we can deduce it from the queue */
sio->sio_offset = DVA_GET_OFFSET(&bp->blk_dva[dva_i]);
sio->sio_asize = DVA_GET_ASIZE(&bp->blk_dva[dva_i]);
sio->sio_blk_prop = bp->blk_prop;
sio->sio_phys_birth = bp->blk_phys_birth;
sio->sio_birth = bp->blk_birth;
sio->sio_cksum = bp->blk_cksum;
}
int
dsl_scan_init(dsl_pool_t *dp, uint64_t txg)
{
int err;
dsl_scan_t *scn;
spa_t *spa = dp->dp_spa;
uint64_t f;
scn = dp->dp_scan = kmem_zalloc(sizeof (dsl_scan_t), KM_SLEEP);
scn->scn_dp = dp;
/*
* It's possible that we're resuming a scan after a reboot so
* make sure that the scan_async_destroying flag is initialized
* appropriately.
*/
ASSERT(!scn->scn_async_destroying);
scn->scn_async_destroying = spa_feature_is_active(dp->dp_spa,
SPA_FEATURE_ASYNC_DESTROY);
/*
* Calculate the max number of in-flight bytes for pool-wide
* scanning operations (minimum 1MB). Limits for the issuing
* phase are done per top-level vdev and are handled separately.
*/
scn->scn_maxinflight_bytes = MAX(zfs_scan_vdev_limit *
dsl_scan_count_leaves(spa->spa_root_vdev), 1ULL << 20);
avl_create(&scn->scn_queue, scan_ds_queue_compare, sizeof (scan_ds_t),
offsetof(scan_ds_t, sds_node));
avl_create(&scn->scn_prefetch_queue, scan_prefetch_queue_compare,
sizeof (scan_prefetch_issue_ctx_t),
offsetof(scan_prefetch_issue_ctx_t, spic_avl_node));
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
"scrub_func", sizeof (uint64_t), 1, &f);
if (err == 0) {
/*
* There was an old-style scrub in progress. Restart a
* new-style scrub from the beginning.
*/
scn->scn_restart_txg = txg;
zfs_dbgmsg("old-style scrub was in progress; "
"restarting new-style scrub in txg %llu",
(longlong_t)scn->scn_restart_txg);
/*
* Load the queue obj from the old location so that it
* can be freed by dsl_scan_done().
*/
(void) zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
"scrub_queue", sizeof (uint64_t), 1,
&scn->scn_phys.scn_queue_obj);
} else {
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
&scn->scn_phys);
/*
* Detect if the pool contains the signature of #2094. If it
* does properly update the scn->scn_phys structure and notify
* the administrator by setting an errata for the pool.
*/
if (err == EOVERFLOW) {
uint64_t zaptmp[SCAN_PHYS_NUMINTS + 1];
VERIFY3S(SCAN_PHYS_NUMINTS, ==, 24);
VERIFY3S(offsetof(dsl_scan_phys_t, scn_flags), ==,
(23 * sizeof (uint64_t)));
err = zap_lookup(dp->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN,
sizeof (uint64_t), SCAN_PHYS_NUMINTS + 1, &zaptmp);
if (err == 0) {
uint64_t overflow = zaptmp[SCAN_PHYS_NUMINTS];
if (overflow & ~DSL_SCAN_FLAGS_MASK ||
scn->scn_async_destroying) {
spa->spa_errata =
ZPOOL_ERRATA_ZOL_2094_ASYNC_DESTROY;
return (EOVERFLOW);
}
bcopy(zaptmp, &scn->scn_phys,
SCAN_PHYS_NUMINTS * sizeof (uint64_t));
scn->scn_phys.scn_flags = overflow;
/* Required scrub already in progress. */
if (scn->scn_phys.scn_state == DSS_FINISHED ||
scn->scn_phys.scn_state == DSS_CANCELED)
spa->spa_errata =
ZPOOL_ERRATA_ZOL_2094_SCRUB;
}
}
if (err == ENOENT)
return (0);
else if (err)
return (err);
/*
* We might be restarting after a reboot, so jump the issued
* counter to how far we've scanned. We know we're consistent
* up to here.
*/
scn->scn_issued_before_pass = scn->scn_phys.scn_examined;
if (dsl_scan_is_running(scn) &&
spa_prev_software_version(dp->dp_spa) < SPA_VERSION_SCAN) {
/*
* A new-type scrub was in progress on an old
* pool, and the pool was accessed by old
* software. Restart from the beginning, since
* the old software may have changed the pool in
* the meantime.
*/
scn->scn_restart_txg = txg;
zfs_dbgmsg("new-style scrub was modified "
"by old software; restarting in txg %llu",
(longlong_t)scn->scn_restart_txg);
}
}
bcopy(&scn->scn_phys, &scn->scn_phys_cached, sizeof (scn->scn_phys));
/* reload the queue into the in-core state */
if (scn->scn_phys.scn_queue_obj != 0) {
zap_cursor_t zc;
zap_attribute_t za;
for (zap_cursor_init(&zc, dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj);
zap_cursor_retrieve(&zc, &za) == 0;
(void) zap_cursor_advance(&zc)) {
scan_ds_queue_insert(scn,
zfs_strtonum(za.za_name, NULL),
za.za_first_integer);
}
zap_cursor_fini(&zc);
}
spa_scan_stat_init(spa);
return (0);
}
void
dsl_scan_fini(dsl_pool_t *dp)
{
if (dp->dp_scan != NULL) {
dsl_scan_t *scn = dp->dp_scan;
if (scn->scn_taskq != NULL)
taskq_destroy(scn->scn_taskq);
scan_ds_queue_clear(scn);
avl_destroy(&scn->scn_queue);
scan_ds_prefetch_queue_clear(scn);
avl_destroy(&scn->scn_prefetch_queue);
kmem_free(dp->dp_scan, sizeof (dsl_scan_t));
dp->dp_scan = NULL;
}
}
static boolean_t
dsl_scan_restarting(dsl_scan_t *scn, dmu_tx_t *tx)
{
return (scn->scn_restart_txg != 0 &&
scn->scn_restart_txg <= tx->tx_txg);
}
boolean_t
dsl_scan_scrubbing(const dsl_pool_t *dp)
{
dsl_scan_phys_t *scn_phys = &dp->dp_scan->scn_phys;
return (scn_phys->scn_state == DSS_SCANNING &&
scn_phys->scn_func == POOL_SCAN_SCRUB);
}
boolean_t
dsl_scan_is_paused_scrub(const dsl_scan_t *scn)
{
return (dsl_scan_scrubbing(scn->scn_dp) &&
scn->scn_phys.scn_flags & DSF_SCRUB_PAUSED);
}
/*
* Writes out a persistent dsl_scan_phys_t record to the pool directory.
* Because we can be running in the block sorting algorithm, we do not always
* want to write out the record, only when it is "safe" to do so. This safety
* condition is achieved by making sure that the sorting queues are empty
* (scn_bytes_pending == 0). When this condition is not true, the sync'd state
* is inconsistent with how much actual scanning progress has been made. The
* kind of sync to be performed is specified by the sync_type argument. If the
* sync is optional, we only sync if the queues are empty. If the sync is
* mandatory, we do a hard ASSERT to make sure that the queues are empty. The
* third possible state is a "cached" sync. This is done in response to:
* 1) The dataset that was in the last sync'd dsl_scan_phys_t having been
* destroyed, so we wouldn't be able to restart scanning from it.
* 2) The snapshot that was in the last sync'd dsl_scan_phys_t having been
* superseded by a newer snapshot.
* 3) The dataset that was in the last sync'd dsl_scan_phys_t having been
* swapped with its clone.
* In all cases, a cached sync simply rewrites the last record we've written,
* just slightly modified. For the modifications that are performed to the
* last written dsl_scan_phys_t, see dsl_scan_ds_destroyed,
* dsl_scan_ds_snapshotted and dsl_scan_ds_clone_swapped.
*/
static void
dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx, state_sync_type_t sync_type)
{
int i;
spa_t *spa = scn->scn_dp->dp_spa;
ASSERT(sync_type != SYNC_MANDATORY || scn->scn_bytes_pending == 0);
if (scn->scn_bytes_pending == 0) {
for (i = 0; i < spa->spa_root_vdev->vdev_children; i++) {
vdev_t *vd = spa->spa_root_vdev->vdev_child[i];
dsl_scan_io_queue_t *q = vd->vdev_scan_io_queue;
if (q == NULL)
continue;
mutex_enter(&vd->vdev_scan_io_queue_lock);
ASSERT3P(avl_first(&q->q_sios_by_addr), ==, NULL);
ASSERT3P(avl_first(&q->q_exts_by_size), ==, NULL);
ASSERT3P(range_tree_first(q->q_exts_by_addr), ==, NULL);
mutex_exit(&vd->vdev_scan_io_queue_lock);
}
if (scn->scn_phys.scn_queue_obj != 0)
scan_ds_queue_sync(scn, tx);
VERIFY0(zap_update(scn->scn_dp->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
&scn->scn_phys, tx));
bcopy(&scn->scn_phys, &scn->scn_phys_cached,
sizeof (scn->scn_phys));
if (scn->scn_checkpointing)
zfs_dbgmsg("finish scan checkpoint");
scn->scn_checkpointing = B_FALSE;
scn->scn_last_checkpoint = ddi_get_lbolt();
} else if (sync_type == SYNC_CACHED) {
VERIFY0(zap_update(scn->scn_dp->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
&scn->scn_phys_cached, tx));
}
}
/* ARGSUSED */
static int
dsl_scan_setup_check(void *arg, dmu_tx_t *tx)
{
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
if (dsl_scan_is_running(scn))
return (SET_ERROR(EBUSY));
return (0);
}
static void
dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
{
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
pool_scan_func_t *funcp = arg;
dmu_object_type_t ot = 0;
dsl_pool_t *dp = scn->scn_dp;
spa_t *spa = dp->dp_spa;
ASSERT(!dsl_scan_is_running(scn));
ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS);
bzero(&scn->scn_phys, sizeof (scn->scn_phys));
scn->scn_phys.scn_func = *funcp;
scn->scn_phys.scn_state = DSS_SCANNING;
scn->scn_phys.scn_min_txg = 0;
scn->scn_phys.scn_max_txg = tx->tx_txg;
scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */
scn->scn_phys.scn_start_time = gethrestime_sec();
scn->scn_phys.scn_errors = 0;
scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc;
scn->scn_issued_before_pass = 0;
scn->scn_restart_txg = 0;
scn->scn_done_txg = 0;
scn->scn_last_checkpoint = 0;
scn->scn_checkpointing = B_FALSE;
spa_scan_stat_init(spa);
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max;
/* rewrite all disk labels */
vdev_config_dirty(spa->spa_root_vdev);
if (vdev_resilver_needed(spa->spa_root_vdev,
&scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) {
spa_event_notify(spa, NULL, NULL,
ESC_ZFS_RESILVER_START);
} else {
spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_START);
}
spa->spa_scrub_started = B_TRUE;
/*
* If this is an incremental scrub, limit the DDT scrub phase
* to just the auto-ditto class (for correctness); the rest
* of the scrub should go faster using top-down pruning.
*/
if (scn->scn_phys.scn_min_txg > TXG_INITIAL)
scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO;
}
/* back to the generic stuff */
if (dp->dp_blkstats == NULL) {
dp->dp_blkstats =
vmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP);
mutex_init(&dp->dp_blkstats->zab_lock, NULL,
MUTEX_DEFAULT, NULL);
}
bzero(&dp->dp_blkstats->zab_type, sizeof (dp->dp_blkstats->zab_type));
if (spa_version(spa) < SPA_VERSION_DSL_SCRUB)
ot = DMU_OT_ZAP_OTHER;
scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset,
ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx);
bcopy(&scn->scn_phys, &scn->scn_phys_cached, sizeof (scn->scn_phys));
dsl_scan_sync_state(scn, tx, SYNC_MANDATORY);
spa_history_log_internal(spa, "scan setup", tx,
"func=%u mintxg=%llu maxtxg=%llu",
*funcp, scn->scn_phys.scn_min_txg, scn->scn_phys.scn_max_txg);
}
/*
* Called by the ZFS_IOC_POOL_SCAN ioctl to start a scrub or resilver.
* Can also be called to resume a paused scrub.
*/
int
dsl_scan(dsl_pool_t *dp, pool_scan_func_t func)
{
spa_t *spa = dp->dp_spa;
dsl_scan_t *scn = dp->dp_scan;
/*
* Purge all vdev caches and probe all devices. We do this here
* rather than in sync context because this requires a writer lock
* on the spa_config lock, which we can't do from sync context. The
* spa_scrub_reopen flag indicates that vdev_open() should not
* attempt to start another scrub.
*/
spa_vdev_state_enter(spa, SCL_NONE);
spa->spa_scrub_reopen = B_TRUE;
vdev_reopen(spa->spa_root_vdev);
spa->spa_scrub_reopen = B_FALSE;
(void) spa_vdev_state_exit(spa, NULL, 0);
if (func == POOL_SCAN_RESILVER) {
dsl_resilver_restart(spa->spa_dsl_pool, 0);
return (0);
}
if (func == POOL_SCAN_SCRUB && dsl_scan_is_paused_scrub(scn)) {
/* got scrub start cmd, resume paused scrub */
int err = dsl_scrub_set_pause_resume(scn->scn_dp,
POOL_SCRUB_NORMAL);
if (err == 0) {
spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_RESUME);
return (ECANCELED);
}
return (SET_ERROR(err));
}
return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check,
dsl_scan_setup_sync, &func, 0, ZFS_SPACE_CHECK_EXTRA_RESERVED));
}
/*
* Sets the resilver defer flag to B_FALSE on all leaf devs under vd. Returns
* B_TRUE if we have devices that need to be resilvered and are available to
* accept resilver I/Os.
*/
static boolean_t
dsl_scan_clear_deferred(vdev_t *vd, dmu_tx_t *tx)
{
boolean_t resilver_needed = B_FALSE;
spa_t *spa = vd->vdev_spa;
for (int c = 0; c < vd->vdev_children; c++) {
resilver_needed |=
dsl_scan_clear_deferred(vd->vdev_child[c], tx);
}
if (vd == spa->spa_root_vdev &&
spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) {
spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx);
vdev_config_dirty(vd);
spa->spa_resilver_deferred = B_FALSE;
return (resilver_needed);
}
if (!vdev_is_concrete(vd) || vd->vdev_aux ||
!vd->vdev_ops->vdev_op_leaf)
return (resilver_needed);
if (vd->vdev_resilver_deferred)
vd->vdev_resilver_deferred = B_FALSE;
return (!vdev_is_dead(vd) && !vd->vdev_offline &&
vdev_resilver_needed(vd, NULL, NULL));
}
/* ARGSUSED */
static void
dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx)
{
static const char *old_names[] = {
"scrub_bookmark",
"scrub_ddt_bookmark",
"scrub_ddt_class_max",
"scrub_queue",
"scrub_min_txg",
"scrub_max_txg",
"scrub_func",
"scrub_errors",
NULL
};
dsl_pool_t *dp = scn->scn_dp;
spa_t *spa = dp->dp_spa;
int i;
/* Remove any remnants of an old-style scrub. */
for (i = 0; old_names[i]; i++) {
(void) zap_remove(dp->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT, old_names[i], tx);
}
if (scn->scn_phys.scn_queue_obj != 0) {
VERIFY0(dmu_object_free(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, tx));
scn->scn_phys.scn_queue_obj = 0;
}
scan_ds_queue_clear(scn);
scan_ds_prefetch_queue_clear(scn);
scn->scn_phys.scn_flags &= ~DSF_SCRUB_PAUSED;
/*
* If we were "restarted" from a stopped state, don't bother
* with anything else.
*/
if (!dsl_scan_is_running(scn)) {
ASSERT(!scn->scn_is_sorted);
return;
}
if (scn->scn_is_sorted) {
scan_io_queues_destroy(scn);
scn->scn_is_sorted = B_FALSE;
if (scn->scn_taskq != NULL) {
taskq_destroy(scn->scn_taskq);
scn->scn_taskq = NULL;
}
}
scn->scn_phys.scn_state = complete ? DSS_FINISHED : DSS_CANCELED;
if (dsl_scan_restarting(scn, tx))
spa_history_log_internal(spa, "scan aborted, restarting", tx,
"errors=%llu", spa_get_errlog_size(spa));
else if (!complete)
spa_history_log_internal(spa, "scan cancelled", tx,
"errors=%llu", spa_get_errlog_size(spa));
else
spa_history_log_internal(spa, "scan done", tx,
"errors=%llu", spa_get_errlog_size(spa));
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
spa->spa_scrub_started = B_FALSE;
spa->spa_scrub_active = B_FALSE;
/*
* If the scrub/resilver completed, update all DTLs to
* reflect this. Whether it succeeded or not, vacate
* all temporary scrub DTLs.
*
* As the scrub does not currently support traversing
* data that have been freed but are part of a checkpoint,
* we don't mark the scrub as done in the DTLs as faults
* may still exist in those vdevs.
*/
if (complete &&
!spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg,
scn->scn_phys.scn_max_txg, B_TRUE);
spa_event_notify(spa, NULL, NULL,
scn->scn_phys.scn_min_txg ?
ESC_ZFS_RESILVER_FINISH : ESC_ZFS_SCRUB_FINISH);
} else {
vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg,
0, B_TRUE);
}
spa_errlog_rotate(spa);
/*
* We may have finished replacing a device.
* Let the async thread assess this and handle the detach.
*/
spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
/*
* Clear any deferred_resilver flags in the config.
* If there are drives that need resilvering, kick
* off an asynchronous request to start resilver.
* dsl_scan_clear_deferred() may update the config
* before the resilver can restart. In the event of
* a crash during this period, the spa loading code
* will find the drives that need to be resilvered
* when the machine reboots and start the resilver then.
*/
boolean_t resilver_needed =
dsl_scan_clear_deferred(spa->spa_root_vdev, tx);
if (resilver_needed) {
spa_history_log_internal(spa,
"starting deferred resilver", tx,
"errors=%llu", spa_get_errlog_size(spa));
spa_async_request(spa, SPA_ASYNC_RESILVER);
}
}
scn->scn_phys.scn_end_time = gethrestime_sec();
if (spa->spa_errata == ZPOOL_ERRATA_ZOL_2094_SCRUB)
spa->spa_errata = 0;
ASSERT(!dsl_scan_is_running(scn));
}
/* ARGSUSED */
static int
dsl_scan_cancel_check(void *arg, dmu_tx_t *tx)
{
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
if (!dsl_scan_is_running(scn))
return (SET_ERROR(ENOENT));
return (0);
}
/* ARGSUSED */
static void
dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx)
{
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
dsl_scan_done(scn, B_FALSE, tx);
dsl_scan_sync_state(scn, tx, SYNC_MANDATORY);
spa_event_notify(scn->scn_dp->dp_spa, NULL, NULL, ESC_ZFS_SCRUB_ABORT);
}
int
dsl_scan_cancel(dsl_pool_t *dp)
{
return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check,
dsl_scan_cancel_sync, NULL, 3, ZFS_SPACE_CHECK_RESERVED));
}
static int
dsl_scrub_pause_resume_check(void *arg, dmu_tx_t *tx)
{
pool_scrub_cmd_t *cmd = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_scan_t *scn = dp->dp_scan;
if (*cmd == POOL_SCRUB_PAUSE) {
/* can't pause a scrub when there is no in-progress scrub */
if (!dsl_scan_scrubbing(dp))
return (SET_ERROR(ENOENT));
/* can't pause a paused scrub */
if (dsl_scan_is_paused_scrub(scn))
return (SET_ERROR(EBUSY));
} else if (*cmd != POOL_SCRUB_NORMAL) {
return (SET_ERROR(ENOTSUP));
}
return (0);
}
static void
dsl_scrub_pause_resume_sync(void *arg, dmu_tx_t *tx)
{
pool_scrub_cmd_t *cmd = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
spa_t *spa = dp->dp_spa;
dsl_scan_t *scn = dp->dp_scan;
if (*cmd == POOL_SCRUB_PAUSE) {
/* can't pause a scrub when there is no in-progress scrub */
spa->spa_scan_pass_scrub_pause = gethrestime_sec();
scn->scn_phys.scn_flags |= DSF_SCRUB_PAUSED;
scn->scn_phys_cached.scn_flags |= DSF_SCRUB_PAUSED;
dsl_scan_sync_state(scn, tx, SYNC_CACHED);
spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_PAUSED);
} else {
ASSERT3U(*cmd, ==, POOL_SCRUB_NORMAL);
if (dsl_scan_is_paused_scrub(scn)) {
/*
* We need to keep track of how much time we spend
* paused per pass so that we can adjust the scrub rate
* shown in the output of 'zpool status'
*/
spa->spa_scan_pass_scrub_spent_paused +=
gethrestime_sec() - spa->spa_scan_pass_scrub_pause;
spa->spa_scan_pass_scrub_pause = 0;
scn->scn_phys.scn_flags &= ~DSF_SCRUB_PAUSED;
scn->scn_phys_cached.scn_flags &= ~DSF_SCRUB_PAUSED;
dsl_scan_sync_state(scn, tx, SYNC_CACHED);
}
}
}
/*
* Set scrub pause/resume state if it makes sense to do so
*/