summaryrefslogtreecommitdiffstats
path: root/fs/btrfs/reada.c
blob: e88bca87f5d275c7c25b2ee6279fe4cdc1482413 (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
973
974
975
976
977
978
979
980
981
982
983
984
/*
 * Copyright (C) 2011 STRATO.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

#include <linux/sched.h>
#include <linux/pagemap.h>
#include <linux/writeback.h>
#include <linux/blkdev.h>
#include <linux/rbtree.h>
#include <linux/slab.h>
#include <linux/workqueue.h>
#include "ctree.h"
#include "volumes.h"
#include "disk-io.h"
#include "transaction.h"
#include "dev-replace.h"

#undef DEBUG

/*
 * This is the implementation for the generic read ahead framework.
 *
 * To trigger a readahead, btrfs_reada_add must be called. It will start
 * a read ahead for the given range [start, end) on tree root. The returned
 * handle can either be used to wait on the readahead to finish
 * (btrfs_reada_wait), or to send it to the background (btrfs_reada_detach).
 *
 * The read ahead works as follows:
 * On btrfs_reada_add, the root of the tree is inserted into a radix_tree.
 * reada_start_machine will then search for extents to prefetch and trigger
 * some reads. When a read finishes for a node, all contained node/leaf
 * pointers that lie in the given range will also be enqueued. The reads will
 * be triggered in sequential order, thus giving a big win over a naive
 * enumeration. It will also make use of multi-device layouts. Each disk
 * will have its on read pointer and all disks will by utilized in parallel.
 * Also will no two disks read both sides of a mirror simultaneously, as this
 * would waste seeking capacity. Instead both disks will read different parts
 * of the filesystem.
 * Any number of readaheads can be started in parallel. The read order will be
 * determined globally, i.e. 2 parallel readaheads will normally finish faster
 * than the 2 started one after another.
 */

#define MAX_IN_FLIGHT 6

struct reada_extctl {
	struct list_head	list;
	struct reada_control	*rc;
	u64			generation;
};

struct reada_extent {
	u64			logical;
	struct btrfs_key	top;
	int			err;
	struct list_head	extctl;
	int 			refcnt;
	spinlock_t		lock;
	struct reada_zone	*zones[BTRFS_MAX_MIRRORS];
	int			nzones;
	int			scheduled;
};

struct reada_zone {
	u64			start;
	u64			end;
	u64			elems;
	struct list_head	list;
	spinlock_t		lock;
	int			locked;
	struct btrfs_device	*device;
	struct btrfs_device	*devs[BTRFS_MAX_MIRRORS]; /* full list, incl
							   * self */
	int			ndevs;
	struct kref		refcnt;
};

struct reada_machine_work {
	struct btrfs_work	work;
	struct btrfs_fs_info	*fs_info;
};

static void reada_extent_put(struct btrfs_fs_info *, struct reada_extent *);
static void reada_control_release(struct kref *kref);
static void reada_zone_release(struct kref *kref);
static void reada_start_machine(struct btrfs_fs_info *fs_info);
static void __reada_start_machine(struct btrfs_fs_info *fs_info);

static int reada_add_block(struct reada_control *rc, u64 logical,
			   struct btrfs_key *top, u64 generation);

/* recurses */
/* in case of err, eb might be NULL */
static void __readahead_hook(struct btrfs_fs_info *fs_info,
			     struct reada_extent *re, struct extent_buffer *eb,
			     int err)
{
	int nritems;
	int i;
	u64 bytenr;
	u64 generation;
	struct list_head list;

	spin_lock(&re->lock);
	/*
	 * just take the full list from the extent. afterwards we
	 * don't need the lock anymore
	 */
	list_replace_init(&re->extctl, &list);
	re->scheduled = 0;
	spin_unlock(&re->lock);

	/*
	 * this is the error case, the extent buffer has not been
	 * read correctly. We won't access anything from it and
	 * just cleanup our data structures. Effectively this will
	 * cut the branch below this node from read ahead.
	 */
	if (err)
		goto cleanup;

	/*
	 * FIXME: currently we just set nritems to 0 if this is a leaf,
	 * effectively ignoring the content. In a next step we could
	 * trigger more readahead depending from the content, e.g.
	 * fetch the checksums for the extents in the leaf.
	 */
	if (!btrfs_header_level(eb))
		goto cleanup;

	nritems = btrfs_header_nritems(eb);
	generation = btrfs_header_generation(eb);
	for (i = 0; i < nritems; i++) {
		struct reada_extctl *rec;
		u64 n_gen;
		struct btrfs_key key;
		struct btrfs_key next_key;

		btrfs_node_key_to_cpu(eb, &key, i);
		if (i + 1 < nritems)
			btrfs_node_key_to_cpu(eb, &next_key, i + 1);
		else
			next_key = re->top;
		bytenr = btrfs_node_blockptr(eb, i);
		n_gen = btrfs_node_ptr_generation(eb, i);

		list_for_each_entry(rec, &list, list) {
			struct reada_control *rc = rec->rc;

			/*
			 * if the generation doesn't match, just ignore this
			 * extctl. This will probably cut off a branch from
			 * prefetch. Alternatively one could start a new (sub-)
			 * prefetch for this branch, starting again from root.
			 * FIXME: move the generation check out of this loop
			 */
#ifdef DEBUG
			if (rec->generation != generation) {
				btrfs_debug(fs_info,
					    "generation mismatch for (%llu,%d,%llu) %llu != %llu",
					    key.objectid, key.type, key.offset,
					    rec->generation, generation);
			}
#endif
			if (rec->generation == generation &&
			    btrfs_comp_cpu_keys(&key, &rc->key_end) < 0 &&
			    btrfs_comp_cpu_keys(&next_key, &rc->key_start) > 0)
				reada_add_block(rc, bytenr, &next_key, n_gen);
		}
	}

cleanup:
	/*
	 * free extctl records
	 */
	while (!list_empty(&list)) {
		struct reada_control *rc;
		struct reada_extctl *rec;

		rec = list_first_entry(&list, struct reada_extctl, list);
		list_del(&rec->list);
		rc = rec->rc;
		kfree(rec);

		kref_get(&rc->refcnt);
		if (atomic_dec_and_test(&rc->elems)) {
			kref_put(&rc->refcnt, reada_control_release);
			wake_up(&rc->wait);
		}
		kref_put(&rc->refcnt, reada_control_release);

		reada_extent_put(fs_info, re);	/* one ref for each entry */
	}

	return;
}

int btree_readahead_hook(struct btrfs_fs_info *fs_info,
			 struct extent_buffer *eb, int err)
{
	int ret = 0;
	struct reada_extent *re;

	/* find extent */
	spin_lock(&fs_info->reada_lock);
	re = radix_tree_lookup(&fs_info->reada_tree,
			       eb->start >> PAGE_SHIFT);
	if (re)
		re->refcnt++;
	spin_unlock(&fs_info->reada_lock);
	if (!re) {
		ret = -1;
		goto start_machine;
	}

	__readahead_hook(fs_info, re, eb, err);
	reada_extent_put(fs_info, re);	/* our ref */

start_machine:
	reada_start_machine(fs_info);
	return ret;
}

static struct reada_zone *reada_find_zone(struct btrfs_fs_info *fs_info,
					  struct btrfs_device *dev, u64 logical,
					  struct btrfs_bio *bbio)
{
	int ret;
	struct reada_zone *zone;
	struct btrfs_block_group_cache *cache = NULL;
	u64 start;
	u64 end;
	int i;

	zone = NULL;
	spin_lock(&fs_info->reada_lock);
	ret = radix_tree_gang_lookup(&dev->reada_zones, (void **)&zone,
				     logical >> PAGE_SHIFT, 1);
	if (ret == 1 && logical >= zone->start && logical <= zone->end) {
		kref_get(&zone->refcnt);
		spin_unlock(&fs_info->reada_lock);
		return zone;
	}

	spin_unlock(&fs_info->reada_lock);

	cache = btrfs_lookup_block_group(fs_info, logical);
	if (!cache)
		return NULL;

	start = cache->key.objectid;
	end = start + cache->key.offset - 1;
	btrfs_put_block_group(cache);

	zone = kzalloc(sizeof(*zone), GFP_KERNEL);
	if (!zone)
		return NULL;

	zone->start = start;
	zone->end = end;
	INIT_LIST_HEAD(&zone->list);
	spin_lock_init(&zone->lock);
	zone->locked = 0;
	kref_init(&zone->refcnt);
	zone->elems = 0;
	zone->device = dev; /* our device always sits at index 0 */
	for (i = 0; i < bbio->num_stripes; ++i) {
		/* bounds have already been checked */
		zone->devs[i] = bbio->stripes[i].dev;
	}
	zone->ndevs = bbio->num_stripes;

	spin_lock(&fs_info->reada_lock);
	ret = radix_tree_insert(&dev->reada_zones,
				(unsigned long)(zone->end >> PAGE_SHIFT),
				zone);

	if (ret == -EEXIST) {
		kfree(zone);
		ret = radix_tree_gang_lookup(&dev->reada_zones, (void **)&zone,
					     logical >> PAGE_SHIFT, 1);
		if (ret == 1 && logical >= zone->start && logical <= zone->end)
			kref_get(&zone->refcnt);
		else
			zone = NULL;
	}
	spin_unlock(&fs_info->reada_lock);

	return zone;
}

static struct reada_extent *reada_find_extent(struct btrfs_fs_info *fs_info,
					      u64 logical,
					      struct btrfs_key *top)
{
	int ret;
	struct reada_extent *re = NULL;
	struct reada_extent *re_exist = NULL;
	struct btrfs_bio *bbio = NULL;
	struct btrfs_device *dev;
	struct btrfs_device *prev_dev;
	u32 blocksize;
	u64 length;
	int real_stripes;
	int nzones = 0;
	unsigned long index = logical >> PAGE_SHIFT;
	int dev_replace_is_ongoing;
	int have_zone = 0;

	spin_lock(&fs_info->reada_lock);
	re = radix_tree_lookup(&fs_info->reada_tree, index);
	if (re)
		re->refcnt++;
	spin_unlock(&fs_info->reada_lock);

	if (re)
		return re;

	re = kzalloc(sizeof(*re), GFP_KERNEL);
	if (!re)
		return NULL;

	blocksize = fs_info->nodesize;
	re->logical = logical;
	re->top = *top;
	INIT_LIST_HEAD(&re->extctl);
	spin_lock_init(&re->lock);
	re->refcnt = 1;

	/*
	 * map block
	 */
	length = blocksize;
	ret = btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
			&length, &bbio, 0);
	if (ret || !bbio || length < blocksize)
		goto error;

	if (bbio->num_stripes > BTRFS_MAX_MIRRORS) {
		btrfs_err(fs_info,
			   "readahead: more than %d copies not supported",
			   BTRFS_MAX_MIRRORS);
		goto error;
	}

	real_stripes = bbio->num_stripes - bbio->num_tgtdevs;
	for (nzones = 0; nzones < real_stripes; ++nzones) {
		struct reada_zone *zone;

		dev = bbio->stripes[nzones].dev;

		/* cannot read ahead on missing device. */
		 if (!dev->bdev)
			continue;

		zone = reada_find_zone(fs_info, dev, logical, bbio);
		if (!zone)
			continue;

		re->zones[re->nzones++] = zone;
		spin_lock(&zone->lock);
		if (!zone->elems)
			kref_get(&zone->refcnt);
		++zone->elems;
		spin_unlock(&zone->lock);
		spin_lock(&fs_info->reada_lock);
		kref_put(&zone->refcnt, reada_zone_release);
		spin_unlock(&fs_info->reada_lock);
	}
	if (re->nzones == 0) {
		/* not a single zone found, error and out */
		goto error;
	}

	/* insert extent in reada_tree + all per-device trees, all or nothing */
	btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
	spin_lock(&fs_info->reada_lock);
	ret = radix_tree_insert(&fs_info->reada_tree, index, re);
	if (ret == -EEXIST) {
		re_exist = radix_tree_lookup(&fs_info->reada_tree, index);
		re_exist->refcnt++;
		spin_unlock(&fs_info->reada_lock);
		btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
		goto error;
	}
	if (ret) {
		spin_unlock(&fs_info->reada_lock);
		btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
		goto error;
	}
	prev_dev = NULL;
	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(
			&fs_info->dev_replace);
	for (nzones = 0; nzones < re->nzones; ++nzones) {
		dev = re->zones[nzones]->device;

		if (dev == prev_dev) {
			/*
			 * in case of DUP, just add the first zone. As both
			 * are on the same device, there's nothing to gain
			 * from adding both.
			 * Also, it wouldn't work, as the tree is per device
			 * and adding would fail with EEXIST
			 */
			continue;
		}
		if (!dev->bdev)
			continue;

		if (dev_replace_is_ongoing &&
		    dev == fs_info->dev_replace.tgtdev) {
			/*
			 * as this device is selected for reading only as
			 * a last resort, skip it for read ahead.
			 */
			continue;
		}
		prev_dev = dev;
		ret = radix_tree_insert(&dev->reada_extents, index, re);
		if (ret) {
			while (--nzones >= 0) {
				dev = re->zones[nzones]->device;
				BUG_ON(dev == NULL);
				/* ignore whether the entry was inserted */
				radix_tree_delete(&dev->reada_extents, index);
			}
			radix_tree_delete(&fs_info->reada_tree, index);
			spin_unlock(&fs_info->reada_lock);
			btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
			goto error;
		}
		have_zone = 1;
	}
	spin_unlock(&fs_info->reada_lock);
	btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);

	if (!have_zone)
		goto error;

	btrfs_put_bbio(bbio);
	return re;

error:
	for (nzones = 0; nzones < re->nzones; ++nzones) {
		struct reada_zone *zone;

		zone = re->zones[nzones];
		kref_get(&zone->refcnt);
		spin_lock(&zone->lock);
		--zone->elems;
		if (zone->elems == 0) {
			/*
			 * no fs_info->reada_lock needed, as this can't be
			 * the last ref
			 */
			kref_put(&zone->refcnt, reada_zone_release);
		}
		spin_unlock(&zone->lock);

		spin_lock(&fs_info->reada_lock);
		kref_put(&zone->refcnt, reada_zone_release);
		spin_unlock(&fs_info->reada_lock);
	}
	btrfs_put_bbio(bbio);
	kfree(re);
	return re_exist;
}

static void reada_extent_put(struct btrfs_fs_info *fs_info,
			     struct reada_extent *re)
{
	int i;
	unsigned long index = re->logical >> PAGE_SHIFT;

	spin_lock(&fs_info->reada_lock);
	if (--re->refcnt) {
		spin_unlock(&fs_info->reada_lock);
		return;
	}

	radix_tree_delete(&fs_info->reada_tree, index);
	for (i = 0; i < re->nzones; ++i) {
		struct reada_zone *zone = re->zones[i];

		radix_tree_delete(&zone->device->reada_extents, index);
	}

	spin_unlock(&fs_info->reada_lock);

	for (i = 0; i < re->nzones; ++i) {
		struct reada_zone *zone = re->zones[i];

		kref_get(&zone->refcnt);
		spin_lock(&zone->lock);
		--zone->elems;
		if (zone->elems == 0) {
			/* no fs_info->reada_lock needed, as this can't be
			 * the last ref */
			kref_put(&zone->refcnt, reada_zone_release);
		}
		spin_unlock(&zone->lock);

		spin_lock(&fs_info->reada_lock);
		kref_put(&zone->refcnt, reada_zone_release);
		spin_unlock(&fs_info->reada_lock);
	}

	kfree(re);
}

static void reada_zone_release(struct kref *kref)
{
	struct reada_zone *zone = container_of(kref, struct reada_zone, refcnt);

	radix_tree_delete(&zone->device->reada_zones,
			  zone->end >> PAGE_SHIFT);

	kfree(zone);
}

static void reada_control_release(struct kref *kref)
{
	struct reada_control *rc = container_of(kref, struct reada_control,
						refcnt);

	kfree(rc);
}

static int reada_add_block(struct reada_control *rc, u64 logical,
			   struct btrfs_key *top, u64 generation)
{
	struct btrfs_fs_info *fs_info = rc->fs_info;
	struct reada_extent *re;
	struct reada_extctl *rec;

	/* takes one ref */
	re = reada_find_extent(fs_info, logical, top);
	if (!re)
		return -1;

	rec = kzalloc(sizeof(*rec), GFP_KERNEL);
	if (!rec) {
		reada_extent_put(fs_info, re);
		return -ENOMEM;
	}

	rec->rc = rc;
	rec->generation = generation;
	atomic_inc(&rc->elems);

	spin_lock(&re->lock);
	list_add_tail(&rec->list, &re->extctl);
	spin_unlock(&re->lock);

	/* leave the ref on the extent */

	return 0;
}

/*
 * called with fs_info->reada_lock held
 */
static void reada_peer_zones_set_lock(struct reada_zone *zone, int lock)
{
	int i;
	unsigned long index = zone->end >> PAGE_SHIFT;

	for (i = 0; i < zone->ndevs; ++i) {
		struct reada_zone *peer;
		peer = radix_tree_lookup(&zone->devs[i]->reada_zones, index);
		if (peer && peer->device != zone->device)
			peer->locked = lock;
	}
}

/*
 * called with fs_info->reada_lock held
 */
static int reada_pick_zone(struct btrfs_device *dev)
{
	struct reada_zone *top_zone = NULL;
	struct reada_zone *top_locked_zone = NULL;
	u64 top_elems = 0;
	u64 top_locked_elems = 0;
	unsigned long index = 0;
	int ret;

	if (dev->reada_curr_zone) {
		reada_peer_zones_set_lock(dev->reada_curr_zone, 0);
		kref_put(&dev->reada_curr_zone->refcnt, reada_zone_release);
		dev->reada_curr_zone = NULL;
	}
	/* pick the zone with the most elements */
	while (1) {
		struct reada_zone *zone;

		ret = radix_tree_gang_lookup(&dev->reada_zones,
					     (void **)&zone, index, 1);
		if (ret == 0)
			break;
		index = (zone->end >> PAGE_SHIFT) + 1;
		if (zone->locked) {
			if (zone->elems > top_locked_elems) {
				top_locked_elems = zone->elems;
				top_locked_zone = zone;
			}
		} else {
			if (zone->elems > top_elems) {
				top_elems = zone->elems;
				top_zone = zone;
			}
		}
	}
	if (top_zone)
		dev->reada_curr_zone = top_zone;
	else if (top_locked_zone)
		dev->reada_curr_zone = top_locked_zone;
	else
		return 0;

	dev->reada_next = dev->reada_curr_zone->start;
	kref_get(&dev->reada_curr_zone->refcnt);
	reada_peer_zones_set_lock(dev->reada_curr_zone, 1);

	return 1;
}

static int reada_start_machine_dev(struct btrfs_fs_info *fs_info,
				   struct btrfs_device *dev)
{
	struct reada_extent *re = NULL;
	int mirror_num = 0;
	struct extent_buffer *eb = NULL;
	u64 logical;
	int ret;
	int i;

	spin_lock(&fs_info->reada_lock);
	if (dev->reada_curr_zone == NULL) {
		ret = reada_pick_zone(dev);
		if (!ret) {
			spin_unlock(&fs_info->reada_lock);
			return 0;
		}
	}
	/*
	 * FIXME currently we issue the reads one extent at a time. If we have
	 * a contiguous block of extents, we could also coagulate them or use
	 * plugging to speed things up
	 */
	ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
				     dev->reada_next >> PAGE_SHIFT, 1);
	if (ret == 0 || re->logical > dev->reada_curr_zone->end) {
		ret = reada_pick_zone(dev);
		if (!ret) {
			spin_unlock(&fs_info->reada_lock);
			return 0;
		}
		re = NULL;
		ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
					dev->reada_next >> PAGE_SHIFT, 1);
	}
	if (ret == 0) {
		spin_unlock(&fs_info->reada_lock);
		return 0;
	}
	dev->reada_next = re->logical + fs_info->nodesize;
	re->refcnt++;

	spin_unlock(&fs_info->reada_lock);

	spin_lock(&re->lock);
	if (re->scheduled || list_empty(&re->extctl)) {
		spin_unlock(&re->lock);
		reada_extent_put(fs_info, re);
		return 0;
	}
	re->scheduled = 1;
	spin_unlock(&re->lock);

	/*
	 * find mirror num
	 */
	for (i = 0; i < re->nzones; ++i) {
		if (re->zones[i]->device == dev) {
			mirror_num = i + 1;
			break;
		}
	}
	logical = re->logical;

	atomic_inc(&dev->reada_in_flight);
	ret = reada_tree_block_flagged(fs_info, logical, mirror_num, &eb);
	if (ret)
		__readahead_hook(fs_info, re, NULL, ret);
	else if (eb)
		__readahead_hook(fs_info, re, eb, ret);

	if (eb)
		free_extent_buffer(eb);

	atomic_dec(&dev->reada_in_flight);
	reada_extent_put(fs_info, re);

	return 1;

}

static void reada_start_machine_worker(struct btrfs_work *work)
{
	struct reada_machine_work *rmw;
	struct btrfs_fs_info *fs_info;
	int old_ioprio;

	rmw = container_of(work, struct reada_machine_work, work);
	fs_info = rmw->fs_info;

	kfree(rmw);

	old_ioprio = IOPRIO_PRIO_VALUE(task_nice_ioclass(current),
				       task_nice_ioprio(current));
	set_task_ioprio(current, BTRFS_IOPRIO_READA);
	__reada_start_machine(fs_info);
	set_task_ioprio(current, old_ioprio);

	atomic_dec(&fs_info->reada_works_cnt);
}

static void __reada_start_machine(struct btrfs_fs_info *fs_info)
{
	struct btrfs_device *device;
	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
	u64 enqueued;
	u64 total = 0;
	int i;

	do {
		enqueued = 0;
		mutex_lock(&fs_devices->device_list_mutex);
		list_for_each_entry(device, &fs_devices->devices, dev_list) {
			if (atomic_read(&device->reada_in_flight) <
			    MAX_IN_FLIGHT)
				enqueued += reada_start_machine_dev(fs_info,
								    device);
		}
		mutex_unlock(&fs_devices->device_list_mutex);
		total += enqueued;
	} while (enqueued && total < 10000);

	if (enqueued == 0)
		return;

	/*
	 * If everything is already in the cache, this is effectively single
	 * threaded. To a) not hold the caller for too long and b) to utilize
	 * more cores, we broke the loop above after 10000 iterations and now
	 * enqueue to workers to finish it. This will distribute the load to
	 * the cores.
	 */
	for (i = 0; i < 2; ++i) {
		reada_start_machine(fs_info);
		if (atomic_read(&fs_info->reada_works_cnt) >
		    BTRFS_MAX_MIRRORS * 2)
			break;
	}
}

static void reada_start_machine(struct btrfs_fs_info *fs_info)
{
	struct reada_machine_work *rmw;

	rmw = kzalloc(sizeof(*rmw), GFP_KERNEL);
	if (!rmw) {
		/* FIXME we cannot handle this properly right now */
		BUG();
	}
	btrfs_init_work(&rmw->work, btrfs_readahead_helper,
			reada_start_machine_worker, NULL, NULL);
	rmw->fs_info = fs_info;

	btrfs_queue_work(fs_info->readahead_workers, &rmw->work);
	atomic_inc(&fs_info->reada_works_cnt);
}

#ifdef DEBUG
static void dump_devs(struct btrfs_fs_info *fs_info, int all)
{
	struct btrfs_device *device;
	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
	unsigned long index;
	int ret;
	int i;
	int j;
	int cnt;

	spin_lock(&fs_info->reada_lock);
	list_for_each_entry(device, &fs_devices->devices, dev_list) {
		btrfs_debug(fs_info, "dev %lld has %d in flight", device->devid,
			atomic_read(&device->reada_in_flight));
		index = 0;
		while (1) {
			struct reada_zone *zone;
			ret = radix_tree_gang_lookup(&device->reada_zones,
						     (void **)&zone, index, 1);
			if (ret == 0)
				break;
			pr_debug("  zone %llu-%llu elems %llu locked %d devs",
				    zone->start, zone->end, zone->elems,
				    zone->locked);
			for (j = 0; j < zone->ndevs; ++j) {
				pr_cont(" %lld",
					zone->devs[j]->devid);
			}
			if (device->reada_curr_zone == zone)
				pr_cont(" curr off %llu",
					device->reada_next - zone->start);
			pr_cont("\n");
			index = (zone->end >> PAGE_SHIFT) + 1;
		}
		cnt = 0;
		index = 0;
		while (all) {
			struct reada_extent *re = NULL;

			ret = radix_tree_gang_lookup(&device->reada_extents,
						     (void **)&re, index, 1);
			if (ret == 0)
				break;
			pr_debug("  re: logical %llu size %u empty %d scheduled %d",
				re->logical, fs_info->nodesize,
				list_empty(&re->extctl), re->scheduled);

			for (i = 0; i < re->nzones; ++i) {
				pr_cont(" zone %llu-%llu devs",
					re->zones[i]->start,
					re->zones[i]->end);
				for (j = 0; j < re->zones[i]->ndevs; ++j) {
					pr_cont(" %lld",
						re->zones[i]->devs[j]->devid);
				}
			}
			pr_cont("\n");
			index = (re->logical >> PAGE_SHIFT) + 1;
			if (++cnt > 15)
				break;
		}
	}

	index = 0;
	cnt = 0;
	while (all) {
		struct reada_extent *re = NULL;

		ret = radix_tree_gang_lookup(&fs_info->reada_tree, (void **)&re,
					     index, 1);
		if (ret == 0)
			break;
		if (!re->scheduled) {
			index = (re->logical >> PAGE_SHIFT) + 1;
			continue;
		}
		pr_debug("re: logical %llu size %u list empty %d scheduled %d",
			re->logical, fs_info->nodesize,
			list_empty(&re->extctl), re->scheduled);
		for (i = 0; i < re->nzones; ++i) {
			pr_cont(" zone %llu-%llu devs",
				re->zones[i]->start,
				re->zones[i]->end);
			for (j = 0; j < re->zones[i]->ndevs; ++j) {
				pr_cont(" %lld",
				       re->zones[i]->devs[j]->devid);
			}
		}
		pr_cont("\n");
		index = (re->logical >> PAGE_SHIFT) + 1;
	}
	spin_unlock(&fs_info->reada_lock);
}
#endif

/*
 * interface
 */
struct reada_control *btrfs_reada_add(struct btrfs_root *root,
			struct btrfs_key *key_start, struct btrfs_key *key_end)
{
	struct reada_control *rc;
	u64 start;
	u64 generation;
	int ret;
	struct extent_buffer *node;
	static struct btrfs_key max_key = {
		.objectid = (u64)-1,
		.type = (u8)-1,
		.offset = (u64)-1
	};

	rc = kzalloc(sizeof(*rc), GFP_KERNEL);
	if (!rc)
		return ERR_PTR(-ENOMEM);

	rc->fs_info = root->fs_info;
	rc->key_start = *key_start;
	rc->key_end = *key_end;
	atomic_set(&rc->elems, 0);
	init_waitqueue_head(&rc->wait);
	kref_init(&rc->refcnt);
	kref_get(&rc->refcnt); /* one ref for having elements */

	node = btrfs_root_node(root);
	start = node->start;
	generation = btrfs_header_generation(node);
	free_extent_buffer(node);

	ret = reada_add_block(rc, start, &max_key, generation);
	if (ret) {
		kfree(rc);
		return ERR_PTR(ret);
	}

	reada_start_machine(root->fs_info);

	return rc;
}

#ifdef DEBUG
int btrfs_reada_wait(void *handle)
{
	struct reada_control *rc = handle;
	struct btrfs_fs_info *fs_info = rc->fs_info;

	while (atomic_read(&rc->elems)) {
		if (!atomic_read(&fs_info->reada_works_cnt))
			reada_start_machine(fs_info);
		wait_event_timeout(rc->wait, atomic_read(&rc->elems) == 0,
				   5 * HZ);
		dump_devs(fs_info, atomic_read(&rc->elems) < 10 ? 1 : 0);
	}

	dump_devs(fs_info, atomic_read(&rc->elems) < 10 ? 1 : 0);

	kref_put(&rc->refcnt, reada_control_release);

	return 0;
}
#else
int btrfs_reada_wait(void *handle)
{
	struct reada_control *rc = handle;
	struct btrfs_fs_info *fs_info = rc->fs_info;

	while (atomic_read(&rc->elems)) {
		if (!atomic_read(&fs_info->reada_works_cnt))
			reada_start_machine(fs_info);
		wait_event_timeout(rc->wait, atomic_read(&rc->elems) == 0,
				   (HZ + 9) / 10);
	}

	kref_put(&rc->refcnt, reada_control_release);

	return 0;
}
#endif

void btrfs_reada_detach(void *handle)
{
	struct reada_control *rc = handle;

	kref_put(&rc->refcnt, reada_control_release);
}