aboutsummaryrefslogtreecommitdiffstats
path: root/module/zfs/zio_inject.c
blob: 3598351c499dc3ff2dba886793b6ab9199fcb498 (plain)
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
/*
 * 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 https://opensource.org/licenses/CDDL-1.0.
 * 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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
 * Copyright (c) 2017, Intel Corporation.
 */

/*
 * ZFS fault injection
 *
 * To handle fault injection, we keep track of a series of zinject_record_t
 * structures which describe which logical block(s) should be injected with a
 * fault.  These are kept in a global list.  Each record corresponds to a given
 * spa_t and maintains a special hold on the spa_t so that it cannot be deleted
 * or exported while the injection record exists.
 *
 * Device level injection is done using the 'zi_guid' field.  If this is set, it
 * means that the error is destined for a particular device, not a piece of
 * data.
 *
 * This is a rather poor data structure and algorithm, but we don't expect more
 * than a few faults at any one time, so it should be sufficient for our needs.
 */

#include <sys/arc.h>
#include <sys/zio.h>
#include <sys/zfs_ioctl.h>
#include <sys/vdev_impl.h>
#include <sys/dmu_objset.h>
#include <sys/dsl_dataset.h>
#include <sys/fs/zfs.h>

uint32_t zio_injection_enabled = 0;

/*
 * Data describing each zinject handler registered on the system, and
 * contains the list node linking the handler in the global zinject
 * handler list.
 */
typedef struct inject_handler {
	int			zi_id;
	spa_t			*zi_spa;
	zinject_record_t	zi_record;
	uint64_t		*zi_lanes;
	int			zi_next_lane;
	list_node_t		zi_link;
} inject_handler_t;

/*
 * List of all zinject handlers registered on the system, protected by
 * the inject_lock defined below.
 */
static list_t inject_handlers;

/*
 * This protects insertion into, and traversal of, the inject handler
 * list defined above; as well as the inject_delay_count. Any time a
 * handler is inserted or removed from the list, this lock should be
 * taken as a RW_WRITER; and any time traversal is done over the list
 * (without modification to it) this lock should be taken as a RW_READER.
 */
static krwlock_t inject_lock;

/*
 * This holds the number of zinject delay handlers that have been
 * registered on the system. It is protected by the inject_lock defined
 * above. Thus modifications to this count must be a RW_WRITER of the
 * inject_lock, and reads of this count must be (at least) a RW_READER
 * of the lock.
 */
static int inject_delay_count = 0;

/*
 * This lock is used only in zio_handle_io_delay(), refer to the comment
 * in that function for more details.
 */
static kmutex_t inject_delay_mtx;

/*
 * Used to assign unique identifying numbers to each new zinject handler.
 */
static int inject_next_id = 1;

/*
 * Test if the requested frequency was triggered
 */
static boolean_t
freq_triggered(uint32_t frequency)
{
	/*
	 * zero implies always (100%)
	 */
	if (frequency == 0)
		return (B_TRUE);

	/*
	 * Note: we still handle legacy (unscaled) frequency values
	 */
	uint32_t maximum = (frequency <= 100) ? 100 : ZI_PERCENTAGE_MAX;

	return (random_in_range(maximum) < frequency);
}

/*
 * Returns true if the given record matches the I/O in progress.
 */
static boolean_t
zio_match_handler(const zbookmark_phys_t *zb, uint64_t type, int dva,
    zinject_record_t *record, int error)
{
	/*
	 * Check for a match against the MOS, which is based on type
	 */
	if (zb->zb_objset == DMU_META_OBJSET &&
	    record->zi_objset == DMU_META_OBJSET &&
	    record->zi_object == DMU_META_DNODE_OBJECT) {
		if (record->zi_type == DMU_OT_NONE ||
		    type == record->zi_type)
			return (freq_triggered(record->zi_freq));
		else
			return (B_FALSE);
	}

	/*
	 * Check for an exact match.
	 */
	if (zb->zb_objset == record->zi_objset &&
	    zb->zb_object == record->zi_object &&
	    zb->zb_level == record->zi_level &&
	    zb->zb_blkid >= record->zi_start &&
	    zb->zb_blkid <= record->zi_end &&
	    (record->zi_dvas == 0 ||
	    (dva != ZI_NO_DVA && (record->zi_dvas & (1ULL << dva)))) &&
	    error == record->zi_error) {
		return (freq_triggered(record->zi_freq));
	}

	return (B_FALSE);
}

/*
 * Panic the system when a config change happens in the function
 * specified by tag.
 */
void
zio_handle_panic_injection(spa_t *spa, const char *tag, uint64_t type)
{
	inject_handler_t *handler;

	rw_enter(&inject_lock, RW_READER);

	for (handler = list_head(&inject_handlers); handler != NULL;
	    handler = list_next(&inject_handlers, handler)) {

		if (spa != handler->zi_spa)
			continue;

		if (handler->zi_record.zi_type == type &&
		    strcmp(tag, handler->zi_record.zi_func) == 0)
			panic("Panic requested in function %s\n", tag);
	}

	rw_exit(&inject_lock);
}

/*
 * Inject a decryption failure. Decryption failures can occur in
 * both the ARC and the ZIO layers.
 */
int
zio_handle_decrypt_injection(spa_t *spa, const zbookmark_phys_t *zb,
    uint64_t type, int error)
{
	int ret = 0;
	inject_handler_t *handler;

	rw_enter(&inject_lock, RW_READER);

	for (handler = list_head(&inject_handlers); handler != NULL;
	    handler = list_next(&inject_handlers, handler)) {

		if (spa != handler->zi_spa ||
		    handler->zi_record.zi_cmd != ZINJECT_DECRYPT_FAULT)
			continue;

		if (zio_match_handler(zb, type, ZI_NO_DVA,
		    &handler->zi_record, error)) {
			ret = error;
			break;
		}
	}

	rw_exit(&inject_lock);
	return (ret);
}

/*
 * If this is a physical I/O for a vdev child determine which DVA it is
 * for. We iterate backwards through the DVAs matching on the offset so
 * that we end up with ZI_NO_DVA (-1) if we don't find a match.
 */
static int
zio_match_dva(zio_t *zio)
{
	int i = ZI_NO_DVA;

	if (zio->io_bp != NULL && zio->io_vd != NULL &&
	    zio->io_child_type == ZIO_CHILD_VDEV) {
		for (i = BP_GET_NDVAS(zio->io_bp) - 1; i >= 0; i--) {
			dva_t *dva = &zio->io_bp->blk_dva[i];
			uint64_t off = DVA_GET_OFFSET(dva);
			vdev_t *vd = vdev_lookup_top(zio->io_spa,
			    DVA_GET_VDEV(dva));

			/* Compensate for vdev label added to leaves */
			if (zio->io_vd->vdev_ops->vdev_op_leaf)
				off += VDEV_LABEL_START_SIZE;

			if (zio->io_vd == vd && zio->io_offset == off)
				break;
		}
	}

	return (i);
}


/*
 * Determine if the I/O in question should return failure.  Returns the errno
 * to be returned to the caller.
 */
int
zio_handle_fault_injection(zio_t *zio, int error)
{
	int ret = 0;
	inject_handler_t *handler;

	/*
	 * Ignore I/O not associated with any logical data.
	 */
	if (zio->io_logical == NULL)
		return (0);

	/*
	 * Currently, we only support fault injection on reads.
	 */
	if (zio->io_type != ZIO_TYPE_READ)
		return (0);

	/*
	 * A rebuild I/O has no checksum to verify.
	 */
	if (zio->io_priority == ZIO_PRIORITY_REBUILD && error == ECKSUM)
		return (0);

	rw_enter(&inject_lock, RW_READER);

	for (handler = list_head(&inject_handlers); handler != NULL;
	    handler = list_next(&inject_handlers, handler)) {
		if (zio->io_spa != handler->zi_spa ||
		    handler->zi_record.zi_cmd != ZINJECT_DATA_FAULT)
			continue;

		/* If this handler matches, return the specified error */
		if (zio_match_handler(&zio->io_logical->io_bookmark,
		    zio->io_bp ? BP_GET_TYPE(zio->io_bp) : DMU_OT_NONE,
		    zio_match_dva(zio), &handler->zi_record, error)) {
			ret = error;
			break;
		}
	}

	rw_exit(&inject_lock);

	return (ret);
}

/*
 * Determine if the zio is part of a label update and has an injection
 * handler associated with that portion of the label. Currently, we
 * allow error injection in either the nvlist or the uberblock region of
 * of the vdev label.
 */
int
zio_handle_label_injection(zio_t *zio, int error)
{
	inject_handler_t *handler;
	vdev_t *vd = zio->io_vd;
	uint64_t offset = zio->io_offset;
	int label;
	int ret = 0;

	if (offset >= VDEV_LABEL_START_SIZE &&
	    offset < vd->vdev_psize - VDEV_LABEL_END_SIZE)
		return (0);

	rw_enter(&inject_lock, RW_READER);

	for (handler = list_head(&inject_handlers); handler != NULL;
	    handler = list_next(&inject_handlers, handler)) {
		uint64_t start = handler->zi_record.zi_start;
		uint64_t end = handler->zi_record.zi_end;

		if (handler->zi_record.zi_cmd != ZINJECT_LABEL_FAULT)
			continue;

		/*
		 * The injection region is the relative offsets within a
		 * vdev label. We must determine the label which is being
		 * updated and adjust our region accordingly.
		 */
		label = vdev_label_number(vd->vdev_psize, offset);
		start = vdev_label_offset(vd->vdev_psize, label, start);
		end = vdev_label_offset(vd->vdev_psize, label, end);

		if (zio->io_vd->vdev_guid == handler->zi_record.zi_guid &&
		    (offset >= start && offset <= end)) {
			ret = error;
			break;
		}
	}
	rw_exit(&inject_lock);
	return (ret);
}

static int
zio_inject_bitflip_cb(void *data, size_t len, void *private)
{
	zio_t *zio = private;
	uint8_t *buffer = data;
	uint_t byte = random_in_range(len);

	ASSERT3U(zio->io_type, ==, ZIO_TYPE_READ);

	/* flip a single random bit in an abd data buffer */
	buffer[byte] ^= 1 << random_in_range(8);

	return (1);	/* stop after first flip */
}

static int
zio_handle_device_injection_impl(vdev_t *vd, zio_t *zio, int err1, int err2)
{
	inject_handler_t *handler;
	int ret = 0;

	/*
	 * We skip over faults in the labels unless it's during
	 * device open (i.e. zio == NULL).
	 */
	if (zio != NULL) {
		uint64_t offset = zio->io_offset;

		if (offset < VDEV_LABEL_START_SIZE ||
		    offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE)
			return (0);
	}

	rw_enter(&inject_lock, RW_READER);

	for (handler = list_head(&inject_handlers); handler != NULL;
	    handler = list_next(&inject_handlers, handler)) {

		if (handler->zi_record.zi_cmd != ZINJECT_DEVICE_FAULT)
			continue;

		if (vd->vdev_guid == handler->zi_record.zi_guid) {
			if (handler->zi_record.zi_failfast &&
			    (zio == NULL || (zio->io_flags &
			    (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))) {
				continue;
			}

			/* Handle type specific I/O failures */
			if (zio != NULL &&
			    handler->zi_record.zi_iotype != ZIO_TYPES &&
			    handler->zi_record.zi_iotype != zio->io_type)
				continue;

			if (handler->zi_record.zi_error == err1 ||
			    handler->zi_record.zi_error == err2) {
				/*
				 * limit error injection if requested
				 */
				if (!freq_triggered(handler->zi_record.zi_freq))
					continue;

				/*
				 * For a failed open, pretend like the device
				 * has gone away.
				 */
				if (err1 == ENXIO)
					vd->vdev_stat.vs_aux =
					    VDEV_AUX_OPEN_FAILED;

				/*
				 * Treat these errors as if they had been
				 * retried so that all the appropriate stats
				 * and FMA events are generated.
				 */
				if (!handler->zi_record.zi_failfast &&
				    zio != NULL)
					zio->io_flags |= ZIO_FLAG_IO_RETRY;

				/*
				 * EILSEQ means flip a bit after a read
				 */
				if (handler->zi_record.zi_error == EILSEQ) {
					if (zio == NULL)
						break;

					/* locate buffer data and flip a bit */
					(void) abd_iterate_func(zio->io_abd, 0,
					    zio->io_size, zio_inject_bitflip_cb,
					    zio);
					break;
				}

				ret = handler->zi_record.zi_error;
				break;
			}
			if (handler->zi_record.zi_error == ENXIO) {
				ret = SET_ERROR(EIO);
				break;
			}
		}
	}

	rw_exit(&inject_lock);

	return (ret);
}

int
zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error)
{
	return (zio_handle_device_injection_impl(vd, zio, error, INT_MAX));
}

int
zio_handle_device_injections(vdev_t *vd, zio_t *zio, int err1, int err2)
{
	return (zio_handle_device_injection_impl(vd, zio, err1, err2));
}

/*
 * Simulate hardware that ignores cache flushes.  For requested number
 * of seconds nix the actual writing to disk.
 */
void
zio_handle_ignored_writes(zio_t *zio)
{
	inject_handler_t *handler;

	rw_enter(&inject_lock, RW_READER);

	for (handler = list_head(&inject_handlers); handler != NULL;
	    handler = list_next(&inject_handlers, handler)) {

		/* Ignore errors not destined for this pool */
		if (zio->io_spa != handler->zi_spa ||
		    handler->zi_record.zi_cmd != ZINJECT_IGNORED_WRITES)
			continue;

		/*
		 * Positive duration implies # of seconds, negative
		 * a number of txgs
		 */
		if (handler->zi_record.zi_timer == 0) {
			if (handler->zi_record.zi_duration > 0)
				handler->zi_record.zi_timer = ddi_get_lbolt64();
			else
				handler->zi_record.zi_timer = zio->io_txg;
		}

		/* Have a "problem" writing 60% of the time */
		if (random_in_range(100) < 60)
			zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES;
		break;
	}

	rw_exit(&inject_lock);
}

void
spa_handle_ignored_writes(spa_t *spa)
{
	inject_handler_t *handler;

	if (zio_injection_enabled == 0)
		return;

	rw_enter(&inject_lock, RW_READER);

	for (handler = list_head(&inject_handlers); handler != NULL;
	    handler = list_next(&inject_handlers, handler)) {

		if (spa != handler->zi_spa ||
		    handler->zi_record.zi_cmd != ZINJECT_IGNORED_WRITES)
			continue;

		if (handler->zi_record.zi_duration > 0) {
			VERIFY(handler->zi_record.zi_timer == 0 ||
			    ddi_time_after64(
			    (int64_t)handler->zi_record.zi_timer +
			    handler->zi_record.zi_duration * hz,
			    ddi_get_lbolt64()));
		} else {
			/* duration is negative so the subtraction here adds */
			VERIFY(handler->zi_record.zi_timer == 0 ||
			    handler->zi_record.zi_timer -
			    handler->zi_record.zi_duration >=
			    spa_syncing_txg(spa));
		}
	}

	rw_exit(&inject_lock);
}

hrtime_t
zio_handle_io_delay(zio_t *zio)
{
	vdev_t *vd = zio->io_vd;
	inject_handler_t *min_handler = NULL;
	hrtime_t min_target = 0;

	rw_enter(&inject_lock, RW_READER);

	/*
	 * inject_delay_count is a subset of zio_injection_enabled that
	 * is only incremented for delay handlers. These checks are
	 * mainly added to remind the reader why we're not explicitly
	 * checking zio_injection_enabled like the other functions.
	 */
	IMPLY(inject_delay_count > 0, zio_injection_enabled > 0);
	IMPLY(zio_injection_enabled == 0, inject_delay_count == 0);

	/*
	 * If there aren't any inject delay handlers registered, then we
	 * can short circuit and simply return 0 here. A value of zero
	 * informs zio_delay_interrupt() that this request should not be
	 * delayed. This short circuit keeps us from acquiring the
	 * inject_delay_mutex unnecessarily.
	 */
	if (inject_delay_count == 0) {
		rw_exit(&inject_lock);
		return (0);
	}

	/*
	 * Each inject handler has a number of "lanes" associated with
	 * it. Each lane is able to handle requests independently of one
	 * another, and at a latency defined by the inject handler
	 * record's zi_timer field. Thus if a handler in configured with
	 * a single lane with a 10ms latency, it will delay requests
	 * such that only a single request is completed every 10ms. So,
	 * if more than one request is attempted per each 10ms interval,
	 * the average latency of the requests will be greater than
	 * 10ms; but if only a single request is submitted each 10ms
	 * interval the average latency will be 10ms.
	 *
	 * We need to acquire this mutex to prevent multiple concurrent
	 * threads being assigned to the same lane of a given inject
	 * handler. The mutex allows us to perform the following two
	 * operations atomically:
	 *
	 *	1. determine the minimum handler and minimum target
	 *	   value of all the possible handlers
	 *	2. update that minimum handler's lane array
	 *
	 * Without atomicity, two (or more) threads could pick the same
	 * lane in step (1), and then conflict with each other in step
	 * (2). This could allow a single lane handler to process
	 * multiple requests simultaneously, which shouldn't be possible.
	 */
	mutex_enter(&inject_delay_mtx);

	for (inject_handler_t *handler = list_head(&inject_handlers);
	    handler != NULL; handler = list_next(&inject_handlers, handler)) {
		if (handler->zi_record.zi_cmd != ZINJECT_DELAY_IO)
			continue;

		if (!freq_triggered(handler->zi_record.zi_freq))
			continue;

		if (vd->vdev_guid != handler->zi_record.zi_guid)
			continue;

		/*
		 * Defensive; should never happen as the array allocation
		 * occurs prior to inserting this handler on the list.
		 */
		ASSERT3P(handler->zi_lanes, !=, NULL);

		/*
		 * This should never happen, the zinject command should
		 * prevent a user from setting an IO delay with zero lanes.
		 */
		ASSERT3U(handler->zi_record.zi_nlanes, !=, 0);

		ASSERT3U(handler->zi_record.zi_nlanes, >,
		    handler->zi_next_lane);

		/*
		 * We want to issue this IO to the lane that will become
		 * idle the soonest, so we compare the soonest this
		 * specific handler can complete the IO with all other
		 * handlers, to find the lowest value of all possible
		 * lanes. We then use this lane to submit the request.
		 *
		 * Since each handler has a constant value for its
		 * delay, we can just use the "next" lane for that
		 * handler; as it will always be the lane with the
		 * lowest value for that particular handler (i.e. the
		 * lane that will become idle the soonest). This saves a
		 * scan of each handler's lanes array.
		 *
		 * There's two cases to consider when determining when
		 * this specific IO request should complete. If this
		 * lane is idle, we want to "submit" the request now so
		 * it will complete after zi_timer milliseconds. Thus,
		 * we set the target to now + zi_timer.
		 *
		 * If the lane is busy, we want this request to complete
		 * zi_timer milliseconds after the lane becomes idle.
		 * Since the 'zi_lanes' array holds the time at which
		 * each lane will become idle, we use that value to
		 * determine when this request should complete.
		 */
		hrtime_t idle = handler->zi_record.zi_timer + gethrtime();
		hrtime_t busy = handler->zi_record.zi_timer +
		    handler->zi_lanes[handler->zi_next_lane];
		hrtime_t target = MAX(idle, busy);

		if (min_handler == NULL) {
			min_handler = handler;
			min_target = target;
			continue;
		}

		ASSERT3P(min_handler, !=, NULL);
		ASSERT3U(min_target, !=, 0);

		/*
		 * We don't yet increment the "next lane" variable since
		 * we still might find a lower value lane in another
		 * handler during any remaining iterations. Once we're
		 * sure we've selected the absolute minimum, we'll claim
		 * the lane and increment the handler's "next lane"
		 * field below.
		 */

		if (target < min_target) {
			min_handler = handler;
			min_target = target;
		}
	}

	/*
	 * 'min_handler' will be NULL if no IO delays are registered for
	 * this vdev, otherwise it will point to the handler containing
	 * the lane that will become idle the soonest.
	 */
	if (min_handler != NULL) {
		ASSERT3U(min_target, !=, 0);
		min_handler->zi_lanes[min_handler->zi_next_lane] = min_target;

		/*
		 * If we've used all possible lanes for this handler,
		 * loop back and start using the first lane again;
		 * otherwise, just increment the lane index.
		 */
		min_handler->zi_next_lane = (min_handler->zi_next_lane + 1) %
		    min_handler->zi_record.zi_nlanes;
	}

	mutex_exit(&inject_delay_mtx);
	rw_exit(&inject_lock);

	return (min_target);
}

static int
zio_calculate_range(const char *pool, zinject_record_t *record)
{
	dsl_pool_t *dp;
	dsl_dataset_t *ds;
	objset_t *os = NULL;
	dnode_t *dn = NULL;
	int error;

	/*
	 * Obtain the dnode for object using pool, objset, and object
	 */
	error = dsl_pool_hold(pool, FTAG, &dp);
	if (error)
		return (error);

	error = dsl_dataset_hold_obj(dp, record->zi_objset, FTAG, &ds);
	dsl_pool_rele(dp, FTAG);
	if (error)
		return (error);

	error = dmu_objset_from_ds(ds, &os);
	dsl_dataset_rele(ds, FTAG);
	if (error)
		return (error);

	error = dnode_hold(os, record->zi_object, FTAG, &dn);
	if (error)
		return (error);

	/*
	 * Translate the range into block IDs
	 */
	if (record->zi_start != 0 || record->zi_end != -1ULL) {
		record->zi_start >>= dn->dn_datablkshift;
		record->zi_end >>= dn->dn_datablkshift;
	}
	if (record->zi_level > 0) {
		if (record->zi_level >= dn->dn_nlevels) {
			dnode_rele(dn, FTAG);
			return (SET_ERROR(EDOM));
		}

		if (record->zi_start != 0 || record->zi_end != 0) {
			int shift = dn->dn_indblkshift - SPA_BLKPTRSHIFT;

			for (int level = record->zi_level; level > 0; level--) {
				record->zi_start >>= shift;
				record->zi_end >>= shift;
			}
		}
	}

	dnode_rele(dn, FTAG);
	return (0);
}

/*
 * Create a new handler for the given record.  We add it to the list, adding
 * a reference to the spa_t in the process.  We increment zio_injection_enabled,
 * which is the switch to trigger all fault injection.
 */
int
zio_inject_fault(char *name, int flags, int *id, zinject_record_t *record)
{
	inject_handler_t *handler;
	int error;
	spa_t *spa;

	/*
	 * If this is pool-wide metadata, make sure we unload the corresponding
	 * spa_t, so that the next attempt to load it will trigger the fault.
	 * We call spa_reset() to unload the pool appropriately.
	 */
	if (flags & ZINJECT_UNLOAD_SPA)
		if ((error = spa_reset(name)) != 0)
			return (error);

	if (record->zi_cmd == ZINJECT_DELAY_IO) {
		/*
		 * A value of zero for the number of lanes or for the
		 * delay time doesn't make sense.
		 */
		if (record->zi_timer == 0 || record->zi_nlanes == 0)
			return (SET_ERROR(EINVAL));

		/*
		 * The number of lanes is directly mapped to the size of
		 * an array used by the handler. Thus, to ensure the
		 * user doesn't trigger an allocation that's "too large"
		 * we cap the number of lanes here.
		 */
		if (record->zi_nlanes >= UINT16_MAX)
			return (SET_ERROR(EINVAL));
	}

	/*
	 * If the supplied range was in bytes -- calculate the actual blkid
	 */
	if (flags & ZINJECT_CALC_RANGE) {
		error = zio_calculate_range(name, record);
		if (error != 0)
			return (error);
	}

	if (!(flags & ZINJECT_NULL)) {
		/*
		 * spa_inject_ref() will add an injection reference, which will
		 * prevent the pool from being removed from the namespace while
		 * still allowing it to be unloaded.
		 */
		if ((spa = spa_inject_addref(name)) == NULL)
			return (SET_ERROR(ENOENT));

		handler = kmem_alloc(sizeof (inject_handler_t), KM_SLEEP);

		handler->zi_spa = spa;
		handler->zi_record = *record;

		if (handler->zi_record.zi_cmd == ZINJECT_DELAY_IO) {
			handler->zi_lanes = kmem_zalloc(
			    sizeof (*handler->zi_lanes) *
			    handler->zi_record.zi_nlanes, KM_SLEEP);
			handler->zi_next_lane = 0;
		} else {
			handler->zi_lanes = NULL;
			handler->zi_next_lane = 0;
		}

		rw_enter(&inject_lock, RW_WRITER);

		/*
		 * We can't move this increment into the conditional
		 * above because we need to hold the RW_WRITER lock of
		 * inject_lock, and we don't want to hold that while
		 * allocating the handler's zi_lanes array.
		 */
		if (handler->zi_record.zi_cmd == ZINJECT_DELAY_IO) {
			ASSERT3S(inject_delay_count, >=, 0);
			inject_delay_count++;
			ASSERT3S(inject_delay_count, >, 0);
		}

		*id = handler->zi_id = inject_next_id++;
		list_insert_tail(&inject_handlers, handler);
		atomic_inc_32(&zio_injection_enabled);

		rw_exit(&inject_lock);
	}

	/*
	 * Flush the ARC, so that any attempts to read this data will end up
	 * going to the ZIO layer.  Note that this is a little overkill, but
	 * we don't have the necessary ARC interfaces to do anything else, and
	 * fault injection isn't a performance critical path.
	 */
	if (flags & ZINJECT_FLUSH_ARC)
		/*
		 * We must use FALSE to ensure arc_flush returns, since
		 * we're not preventing concurrent ARC insertions.
		 */
		arc_flush(NULL, FALSE);

	return (0);
}

/*
 * Returns the next record with an ID greater than that supplied to the
 * function.  Used to iterate over all handlers in the system.
 */
int
zio_inject_list_next(int *id, char *name, size_t buflen,
    zinject_record_t *record)
{
	inject_handler_t *handler;
	int ret;

	mutex_enter(&spa_namespace_lock);
	rw_enter(&inject_lock, RW_READER);

	for (handler = list_head(&inject_handlers); handler != NULL;
	    handler = list_next(&inject_handlers, handler))
		if (handler->zi_id > *id)
			break;

	if (handler) {
		*record = handler->zi_record;
		*id = handler->zi_id;
		(void) strlcpy(name, spa_name(handler->zi_spa), buflen);
		ret = 0;
	} else {
		ret = SET_ERROR(ENOENT);
	}

	rw_exit(&inject_lock);
	mutex_exit(&spa_namespace_lock);

	return (ret);
}

/*
 * Clear the fault handler with the given identifier, or return ENOENT if none
 * exists.
 */
int
zio_clear_fault(int id)
{
	inject_handler_t *handler;

	rw_enter(&inject_lock, RW_WRITER);

	for (handler = list_head(&inject_handlers); handler != NULL;
	    handler = list_next(&inject_handlers, handler))
		if (handler->zi_id == id)
			break;

	if (handler == NULL) {
		rw_exit(&inject_lock);
		return (SET_ERROR(ENOENT));
	}

	if (handler->zi_record.zi_cmd == ZINJECT_DELAY_IO) {
		ASSERT3S(inject_delay_count, >, 0);
		inject_delay_count--;
		ASSERT3S(inject_delay_count, >=, 0);
	}

	list_remove(&inject_handlers, handler);
	rw_exit(&inject_lock);

	if (handler->zi_record.zi_cmd == ZINJECT_DELAY_IO) {
		ASSERT3P(handler->zi_lanes, !=, NULL);
		kmem_free(handler->zi_lanes, sizeof (*handler->zi_lanes) *
		    handler->zi_record.zi_nlanes);
	} else {
		ASSERT3P(handler->zi_lanes, ==, NULL);
	}

	spa_inject_delref(handler->zi_spa);
	kmem_free(handler, sizeof (inject_handler_t));
	atomic_dec_32(&zio_injection_enabled);

	return (0);
}

void
zio_inject_init(void)
{
	rw_init(&inject_lock, NULL, RW_DEFAULT, NULL);
	mutex_init(&inject_delay_mtx, NULL, MUTEX_DEFAULT, NULL);
	list_create(&inject_handlers, sizeof (inject_handler_t),
	    offsetof(inject_handler_t, zi_link));
}

void
zio_inject_fini(void)
{
	list_destroy(&inject_handlers);
	mutex_destroy(&inject_delay_mtx);
	rw_destroy(&inject_lock);
}

#if defined(_KERNEL)
EXPORT_SYMBOL(zio_injection_enabled);
EXPORT_SYMBOL(zio_inject_fault);
EXPORT_SYMBOL(zio_inject_list_next);
EXPORT_SYMBOL(zio_clear_fault);
EXPORT_SYMBOL(zio_handle_fault_injection);
EXPORT_SYMBOL(zio_handle_device_injection);
EXPORT_SYMBOL(zio_handle_label_injection);
#endif