summaryrefslogtreecommitdiffstats
path: root/fs/ubifs/gc.c
blob: bad3339a800d6aad2762818c284301f464800ffb (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
/*
 * This file is part of UBIFS.
 *
 * Copyright (C) 2006-2008 Nokia Corporation.
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 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., 51
 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 *
 * Authors: Adrian Hunter
 *          Artem Bityutskiy (Битюцкий Артём)
 */

/*
 * This file implements garbage collection. The procedure for garbage collection
 * is different depending on whether a LEB as an index LEB (contains index
 * nodes) or not. For non-index LEBs, garbage collection finds a LEB which
 * contains a lot of dirty space (obsolete nodes), and copies the non-obsolete
 * nodes to the journal, at which point the garbage-collected LEB is free to be
 * reused. For index LEBs, garbage collection marks the non-obsolete index nodes
 * dirty in the TNC, and after the next commit, the garbage-collected LEB is
 * to be reused. Garbage collection will cause the number of dirty index nodes
 * to grow, however sufficient space is reserved for the index to ensure the
 * commit will never run out of space.
 *
 * Notes about dead watermark. At current UBIFS implementation we assume that
 * LEBs which have less than @c->dead_wm bytes of free + dirty space are full
 * and not worth garbage-collecting. The dead watermark is one min. I/O unit
 * size, or min. UBIFS node size, depending on what is greater. Indeed, UBIFS
 * Garbage Collector has to synchronize the GC head's write buffer before
 * returning, so this is about wasting one min. I/O unit. However, UBIFS GC can
 * actually reclaim even very small pieces of dirty space by garbage collecting
 * enough dirty LEBs, but we do not bother doing this at this implementation.
 *
 * Notes about dark watermark. The results of GC work depends on how big are
 * the UBIFS nodes GC deals with. Large nodes make GC waste more space. Indeed,
 * if GC move data from LEB A to LEB B and nodes in LEB A are large, GC would
 * have to waste large pieces of free space at the end of LEB B, because nodes
 * from LEB A would not fit. And the worst situation is when all nodes are of
 * maximum size. So dark watermark is the amount of free + dirty space in LEB
 * which are guaranteed to be reclaimable. If LEB has less space, the GC migh
 * be unable to reclaim it. So, LEBs with free + dirty greater than dark
 * watermark are "good" LEBs from GC's point of few. The other LEBs are not so
 * good, and GC takes extra care when moving them.
 */

#include <linux/pagemap.h>
#include "ubifs.h"

/*
 * GC tries to optimize the way it fit nodes to available space, and it sorts
 * nodes a little. The below constants are watermarks which define "large",
 * "medium", and "small" nodes.
 */
#define MEDIUM_NODE_WM (UBIFS_BLOCK_SIZE / 4)
#define SMALL_NODE_WM  UBIFS_MAX_DENT_NODE_SZ

/*
 * GC may need to move more than one LEB to make progress. The below constants
 * define "soft" and "hard" limits on the number of LEBs the garbage collector
 * may move.
 */
#define SOFT_LEBS_LIMIT 4
#define HARD_LEBS_LIMIT 32

/**
 * switch_gc_head - switch the garbage collection journal head.
 * @c: UBIFS file-system description object
 * @buf: buffer to write
 * @len: length of the buffer to write
 * @lnum: LEB number written is returned here
 * @offs: offset written is returned here
 *
 * This function switch the GC head to the next LEB which is reserved in
 * @c->gc_lnum. Returns %0 in case of success, %-EAGAIN if commit is required,
 * and other negative error code in case of failures.
 */
static int switch_gc_head(struct ubifs_info *c)
{
	int err, gc_lnum = c->gc_lnum;
	struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;

	ubifs_assert(gc_lnum != -1);
	dbg_gc("switch GC head from LEB %d:%d to LEB %d (waste %d bytes)",
	       wbuf->lnum, wbuf->offs + wbuf->used, gc_lnum,
	       c->leb_size - wbuf->offs - wbuf->used);

	err = ubifs_wbuf_sync_nolock(wbuf);
	if (err)
		return err;

	/*
	 * The GC write-buffer was synchronized, we may safely unmap
	 * 'c->gc_lnum'.
	 */
	err = ubifs_leb_unmap(c, gc_lnum);
	if (err)
		return err;

	err = ubifs_add_bud_to_log(c, GCHD, gc_lnum, 0);
	if (err)
		return err;

	c->gc_lnum = -1;
	err = ubifs_wbuf_seek_nolock(wbuf, gc_lnum, 0, UBI_LONGTERM);
	return err;
}

/**
 * joinup - bring data nodes for an inode together.
 * @c: UBIFS file-system description object
 * @sleb: describes scanned LEB
 * @inum: inode number
 * @blk: block number
 * @data: list to which to add data nodes
 *
 * This function looks at the first few nodes in the scanned LEB @sleb and adds
 * them to @data if they are data nodes from @inum and have a larger block
 * number than @blk. This function returns %0 on success and a negative error
 * code on failure.
 */
static int joinup(struct ubifs_info *c, struct ubifs_scan_leb *sleb, ino_t inum,
		  unsigned int blk, struct list_head *data)
{
	int err, cnt = 6, lnum = sleb->lnum, offs;
	struct ubifs_scan_node *snod, *tmp;
	union ubifs_key *key;

	list_for_each_entry_safe(snod, tmp, &sleb->nodes, list) {
		key = &snod->key;
		if (key_inum(c, key) == inum &&
		    key_type(c, key) == UBIFS_DATA_KEY &&
		    key_block(c, key) > blk) {
			offs = snod->offs;
			err = ubifs_tnc_has_node(c, key, 0, lnum, offs, 0);
			if (err < 0)
				return err;
			list_del(&snod->list);
			if (err) {
				list_add_tail(&snod->list, data);
				blk = key_block(c, key);
			} else
				kfree(snod);
			cnt = 6;
		} else if (--cnt == 0)
			break;
	}
	return 0;
}

/**
 * move_nodes - move nodes.
 * @c: UBIFS file-system description object
 * @sleb: describes nodes to move
 *
 * This function moves valid nodes from data LEB described by @sleb to the GC
 * journal head. The obsolete nodes are dropped.
 *
 * When moving nodes we have to deal with classical bin-packing problem: the
 * space in the current GC journal head LEB and in @c->gc_lnum are the "bins",
 * where the nodes in the @sleb->nodes list are the elements which should be
 * fit optimally to the bins. This function uses the "first fit decreasing"
 * strategy, although it does not really sort the nodes but just split them on
 * 3 classes - large, medium, and small, so they are roughly sorted.
 *
 * This function returns zero in case of success, %-EAGAIN if commit is
 * required, and other negative error codes in case of other failures.
 */
static int move_nodes(struct ubifs_info *c, struct ubifs_scan_leb *sleb)
{
	struct ubifs_scan_node *snod, *tmp;
	struct list_head data, large, medium, small;
	struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
	int avail, err, min = INT_MAX;
	unsigned int blk = 0;
	ino_t inum = 0;

	INIT_LIST_HEAD(&data);
	INIT_LIST_HEAD(&large);
	INIT_LIST_HEAD(&medium);
	INIT_LIST_HEAD(&small);

	while (!list_empty(&sleb->nodes)) {
		struct list_head *lst = sleb->nodes.next;

		snod = list_entry(lst, struct ubifs_scan_node, list);

		ubifs_assert(snod->type != UBIFS_IDX_NODE);
		ubifs_assert(snod->type != UBIFS_REF_NODE);
		ubifs_assert(snod->type != UBIFS_CS_NODE);

		err = ubifs_tnc_has_node(c, &snod->key, 0, sleb->lnum,
					 snod->offs, 0);
		if (err < 0)
			goto out;

		list_del(lst);
		if (!err) {
			/* The node is obsolete, remove it from the list */
			kfree(snod);
			continue;
		}

		/*
		 * Sort the list of nodes so that data nodes go first, large
		 * nodes go second, and small nodes go last.
		 */
		if (key_type(c, &snod->key) == UBIFS_DATA_KEY) {
			if (inum != key_inum(c, &snod->key)) {
				if (inum) {
					/*
					 * Try to move data nodes from the same
					 * inode together.
					 */
					err = joinup(c, sleb, inum, blk, &data);
					if (err)
						goto out;
				}
				inum = key_inum(c, &snod->key);
				blk = key_block(c, &snod->key);
			}
			list_add_tail(lst, &data);
		} else if (snod->len > MEDIUM_NODE_WM)
			list_add_tail(lst, &large);
		else if (snod->len > SMALL_NODE_WM)
			list_add_tail(lst, &medium);
		else
			list_add_tail(lst, &small);

		/* And find the smallest node */
		if (snod->len < min)
			min = snod->len;
	}

	/*
	 * Join the tree lists so that we'd have one roughly sorted list
	 * ('large' will be the head of the joined list).
	 */
	list_splice(&data, &large);
	list_splice(&medium, large.prev);
	list_splice(&small, large.prev);

	if (wbuf->lnum == -1) {
		/*
		 * The GC journal head is not set, because it is the first GC
		 * invocation since mount.
		 */
		err = switch_gc_head(c);
		if (err)
			goto out;
	}

	/* Write nodes to their new location. Use the first-fit strategy */
	while (1) {
		avail = c->leb_size - wbuf->offs - wbuf->used;
		list_for_each_entry_safe(snod, tmp, &large, list) {
			int new_lnum, new_offs;

			if (avail < min)
				break;

			if (snod->len > avail)
				/* This node does not fit */
				continue;

			cond_resched();

			new_lnum = wbuf->lnum;
			new_offs = wbuf->offs + wbuf->used;
			err = ubifs_wbuf_write_nolock(wbuf, snod->node,
						      snod->len);
			if (err)
				goto out;
			err = ubifs_tnc_replace(c, &snod->key, sleb->lnum,
						snod->offs, new_lnum, new_offs,
						snod->len);
			if (err)
				goto out;

			avail = c->leb_size - wbuf->offs - wbuf->used;
			list_del(&snod->list);
			kfree(snod);
		}

		if (list_empty(&large))
			break;

		/*
		 * Waste the rest of the space in the LEB and switch to the
		 * next LEB.
		 */
		err = switch_gc_head(c);
		if (err)
			goto out;
	}

	return 0;

out:
	list_for_each_entry_safe(snod, tmp, &large, list) {
		list_del(&snod->list);
		kfree(snod);
	}
	return err;
}

/**
 * gc_sync_wbufs - sync write-buffers for GC.
 * @c: UBIFS file-system description object
 *
 * We must guarantee that obsoleting nodes are on flash. Unfortunately they may
 * be in a write-buffer instead. That is, a node could be written to a
 * write-buffer, obsoleting another node in a LEB that is GC'd. If that LEB is
 * erased before the write-buffer is sync'd and then there is an unclean
 * unmount, then an existing node is lost. To avoid this, we sync all
 * write-buffers.
 *
 * This function returns %0 on success or a negative error code on failure.
 */
static int gc_sync_wbufs(struct ubifs_info *c)
{
	int err, i;

	for (i = 0; i < c->jhead_cnt; i++) {
		if (i == GCHD)
			continue;
		err = ubifs_wbuf_sync(&c->jheads[i].wbuf);
		if (err)
			return err;
	}
	return 0;
}

/**
 * ubifs_garbage_collect_leb - garbage-collect a logical eraseblock.
 * @c: UBIFS file-system description object
 * @lp: describes the LEB to garbage collect
 *
 * This function garbage-collects an LEB and returns one of the @LEB_FREED,
 * @LEB_RETAINED, etc positive codes in case of success, %-EAGAIN if commit is
 * required, and other negative error codes in case of failures.
 */
int ubifs_garbage_collect_leb(struct ubifs_info *c, struct ubifs_lprops *lp)
{
	struct ubifs_scan_leb *sleb;
	struct ubifs_scan_node *snod;
	struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
	int err = 0, lnum = lp->lnum;

	ubifs_assert(c->gc_lnum != -1 || wbuf->offs + wbuf->used == 0 ||
		     c->need_recovery);
	ubifs_assert(c->gc_lnum != lnum);
	ubifs_assert(wbuf->lnum != lnum);

	/*
	 * We scan the entire LEB even though we only really need to scan up to
	 * (c->leb_size - lp->free).
	 */
	sleb = ubifs_scan(c, lnum, 0, c->sbuf);
	if (IS_ERR(sleb))
		return PTR_ERR(sleb);

	ubifs_assert(!list_empty(&sleb->nodes));
	snod = list_entry(sleb->nodes.next, struct ubifs_scan_node, list);

	if (snod->type == UBIFS_IDX_NODE) {
		struct ubifs_gced_idx_leb *idx_gc;

		dbg_gc("indexing LEB %d (free %d, dirty %d)",
		       lnum, lp->free, lp->dirty);
		list_for_each_entry(snod, &sleb->nodes, list) {
			struct ubifs_idx_node *idx = snod->node;
			int level = le16_to_cpu(idx->level);

			ubifs_assert(snod->type == UBIFS_IDX_NODE);
			key_read(c, ubifs_idx_key(c, idx), &snod->key);
			err = ubifs_dirty_idx_node(c, &snod->key, level, lnum,
						   snod->offs);
			if (err)
				goto out;
		}

		idx_gc = kmalloc(sizeof(struct ubifs_gced_idx_leb), GFP_NOFS);
		if (!idx_gc) {
			err = -ENOMEM;
			goto out;
		}

		idx_gc->lnum = lnum;
		idx_gc->unmap = 0;
		list_add(&idx_gc->list, &c->idx_gc);

		/*
		 * Don't release the LEB until after the next commit, because
		 * it may contain data which is needed for recovery. So
		 * although we freed this LEB, it will become usable only after
		 * the commit.
		 */
		err = ubifs_change_one_lp(c, lnum, c->leb_size, 0, 0,
					  LPROPS_INDEX, 1);
		if (err)
			goto out;
		err = LEB_FREED_IDX;
	} else {
		dbg_gc("data LEB %d (free %d, dirty %d)",
		       lnum, lp->free, lp->dirty);

		err = move_nodes(c, sleb);
		if (err)
			goto out_inc_seq;

		err = gc_sync_wbufs(c);
		if (err)
			goto out_inc_seq;

		err = ubifs_change_one_lp(c, lnum, c->leb_size, 0, 0, 0, 0);
		if (err)
			goto out_inc_seq;

		/* Allow for races with TNC */
		c->gced_lnum = lnum;
		smp_wmb();
		c->gc_seq += 1;
		smp_wmb();

		if (c->gc_lnum == -1) {
			c->gc_lnum = lnum;
			err = LEB_RETAINED;
		} else {
			err = ubifs_wbuf_sync_nolock(wbuf);
			if (err)
				goto out;

			err = ubifs_leb_unmap(c, lnum);
			if (err)
				goto out;

			err = LEB_FREED;
		}
	}

out:
	ubifs_scan_destroy(sleb);
	return err;

out_inc_seq:
	/* We may have moved at least some nodes so allow for races with TNC */
	c->gced_lnum = lnum;
	smp_wmb();
	c->gc_seq += 1;
	smp_wmb();
	goto out;
}

/**
 * ubifs_garbage_collect - UBIFS garbage collector.
 * @c: UBIFS file-system description object
 * @anyway: do GC even if there are free LEBs
 *
 * This function does out-of-place garbage collection. The return codes are:
 *   o positive LEB number if the LEB has been freed and may be used;
 *   o %-EAGAIN if the caller has to run commit;
 *   o %-ENOSPC if GC failed to make any progress;
 *   o other negative error codes in case of other errors.
 *
 * Garbage collector writes data to the journal when GC'ing data LEBs, and just
 * marking indexing nodes dirty when GC'ing indexing LEBs. Thus, at some point
 * commit may be required. But commit cannot be run from inside GC, because the
 * caller might be holding the commit lock, so %-EAGAIN is returned instead;
 * And this error code means that the caller has to run commit, and re-run GC
 * if there is still no free space.
 *
 * There are many reasons why this function may return %-EAGAIN:
 * o the log is full and there is no space to write an LEB reference for
 *   @c->gc_lnum;
 * o the journal is too large and exceeds size limitations;
 * o GC moved indexing LEBs, but they can be used only after the commit;
 * o the shrinker fails to find clean znodes to free and requests the commit;
 * o etc.
 *
 * Note, if the file-system is close to be full, this function may return
 * %-EAGAIN infinitely, so the caller has to limit amount of re-invocations of
 * the function. E.g., this happens if the limits on the journal size are too
 * tough and GC writes too much to the journal before an LEB is freed. This
 * might also mean that the journal is too large, and the TNC becomes to big,
 * so that the shrinker is constantly called, finds not clean znodes to free,
 * and requests commit. Well, this may also happen if the journal is all right,
 * but another kernel process consumes too much memory. Anyway, infinite
 * %-EAGAIN may happen, but in some extreme/misconfiguration cases.
 */
int ubifs_garbage_collect(struct ubifs_info *c, int anyway)
{
	int i, err, ret, min_space = c->dead_wm;
	struct ubifs_lprops lp;
	struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;

	ubifs_assert_cmt_locked(c);

	if (ubifs_gc_should_commit(c))
		return -EAGAIN;

	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);

	if (c->ro_media) {
		ret = -EROFS;
		goto out_unlock;
	}

	/* We expect the write-buffer to be empty on entry */
	ubifs_assert(!wbuf->used);

	for (i = 0; ; i++) {
		int space_before = c->leb_size - wbuf->offs - wbuf->used;
		int space_after;

		cond_resched();

		/* Give the commit an opportunity to run */
		if (ubifs_gc_should_commit(c)) {
			ret = -EAGAIN;
			break;
		}

		if (i > SOFT_LEBS_LIMIT && !list_empty(&c->idx_gc)) {
			/*
			 * We've done enough iterations. Indexing LEBs were
			 * moved and will be available after the commit.
			 */
			dbg_gc("soft limit, some index LEBs GC'ed, -EAGAIN");
			ubifs_commit_required(c);
			ret = -EAGAIN;
			break;
		}

		if (i > HARD_LEBS_LIMIT) {
			/*
			 * We've moved too many LEBs and have not made
			 * progress, give up.
			 */
			dbg_gc("hard limit, -ENOSPC");
			ret = -ENOSPC;
			break;
		}

		/*
		 * Empty and freeable LEBs can turn up while we waited for
		 * the wbuf lock, or while we have been running GC. In that
		 * case, we should just return one of those instead of
		 * continuing to GC dirty LEBs. Hence we request
		 * 'ubifs_find_dirty_leb()' to return an empty LEB if it can.
		 */
		ret = ubifs_find_dirty_leb(c, &lp, min_space, anyway ? 0 : 1);
		if (ret) {
			if (ret == -ENOSPC)
				dbg_gc("no more dirty LEBs");
			break;
		}

		dbg_gc("found LEB %d: free %d, dirty %d, sum %d "
		       "(min. space %d)", lp.lnum, lp.free, lp.dirty,
		       lp.free + lp.dirty, min_space);

		if (lp.free + lp.dirty == c->leb_size) {
			/* An empty LEB was returned */
			dbg_gc("LEB %d is free, return it", lp.lnum);
			/*
			 * ubifs_find_dirty_leb() doesn't return freeable index
			 * LEBs.
			 */
			ubifs_assert(!(lp.flags & LPROPS_INDEX));
			if (lp.free != c->leb_size) {
				/*
				 * Write buffers must be sync'd before
				 * unmapping freeable LEBs, because one of them
				 * may contain data which obsoletes something
				 * in 'lp.pnum'.
				 */
				ret = gc_sync_wbufs(c);
				if (ret)
					goto out;
				ret = ubifs_change_one_lp(c, lp.lnum,
							  c->leb_size, 0, 0, 0,
							  0);
				if (ret)
					goto out;
			}
			ret = ubifs_leb_unmap(c, lp.lnum);
			if (ret)
				goto out;
			ret = lp.lnum;
			break;
		}

		space_before = c->leb_size - wbuf->offs - wbuf->used;
		if (wbuf->lnum == -1)
			space_before = 0;

		ret = ubifs_garbage_collect_leb(c, &lp);
		if (ret < 0) {
			if (ret == -EAGAIN || ret == -ENOSPC) {
				/*
				 * These codes are not errors, so we have to
				 * return the LEB to lprops. But if the
				 * 'ubifs_return_leb()' function fails, its
				 * failure code is propagated to the caller
				 * instead of the original '-EAGAIN' or
				 * '-ENOSPC'.
				 */
				err = ubifs_return_leb(c, lp.lnum);
				if (err)
					ret = err;
				break;
			}
			goto out;
		}

		if (ret == LEB_FREED) {
			/* An LEB has been freed and is ready for use */
			dbg_gc("LEB %d freed, return", lp.lnum);
			ret = lp.lnum;
			break;
		}

		if (ret == LEB_FREED_IDX) {
			/*
			 * This was an indexing LEB and it cannot be
			 * immediately used. And instead of requesting the
			 * commit straight away, we try to garbage collect some
			 * more.
			 */
			dbg_gc("indexing LEB %d freed, continue", lp.lnum);
			continue;
		}

		ubifs_assert(ret == LEB_RETAINED);
		space_after = c->leb_size - wbuf->offs - wbuf->used;
		dbg_gc("LEB %d retained, freed %d bytes", lp.lnum,
		       space_after - space_before);

		if (space_after > space_before) {
			/* GC makes progress, keep working */
			min_space >>= 1;
			if (min_space < c->dead_wm)
				min_space = c->dead_wm;
			continue;
		}

		dbg_gc("did not make progress");

		/*
		 * GC moved an LEB bud have not done any progress. This means
		 * that the previous GC head LEB contained too few free space
		 * and the LEB which was GC'ed contained only large nodes which
		 * did not fit that space.
		 *
		 * We can do 2 things:
		 * 1. pick another LEB in a hope it'll contain a small node
		 *    which will fit the space we have at the end of current GC
		 *    head LEB, but there is no guarantee, so we try this out
		 *    unless we have already been working for too long;
		 * 2. request an LEB with more dirty space, which will force
		 *    'ubifs_find_dirty_leb()' to start scanning the lprops
		 *    table, instead of just picking one from the heap
		 *    (previously it already picked the dirtiest LEB).
		 */
		if (i < SOFT_LEBS_LIMIT) {
			dbg_gc("try again");
			continue;
		}

		min_space <<= 1;
		if (min_space > c->dark_wm)
			min_space = c->dark_wm;
		dbg_gc("set min. space to %d", min_space);
	}

	if (ret == -ENOSPC && !list_empty(&c->idx_gc)) {
		dbg_gc("no space, some index LEBs GC'ed, -EAGAIN");
		ubifs_commit_required(c);
		ret = -EAGAIN;
	}

	err = ubifs_wbuf_sync_nolock(wbuf);
	if (!err)
		err = ubifs_leb_unmap(c, c->gc_lnum);
	if (err) {
		ret = err;
		goto out;
	}
out_unlock:
	mutex_unlock(&wbuf->io_mutex);
	return ret;

out:
	ubifs_assert(ret < 0);
	ubifs_assert(ret != -ENOSPC && ret != -EAGAIN);
	ubifs_ro_mode(c, ret);
	ubifs_wbuf_sync_nolock(wbuf);
	mutex_unlock(&wbuf->io_mutex);
	ubifs_return_leb(c, lp.lnum);
	return ret;
}

/**
 * ubifs_gc_start_commit - garbage collection at start of commit.
 * @c: UBIFS file-system description object
 *
 * If a LEB has only dirty and free space, then we may safely unmap it and make
 * it free.  Note, we cannot do this with indexing LEBs because dirty space may
 * correspond index nodes that are required for recovery.  In that case, the
 * LEB cannot be unmapped until after the next commit.
 *
 * This function returns %0 upon success and a negative error code upon failure.
 */
int ubifs_gc_start_commit(struct ubifs_info *c)
{
	struct ubifs_gced_idx_leb *idx_gc;
	const struct ubifs_lprops *lp;
	int err = 0, flags;

	ubifs_get_lprops(c);

	/*
	 * Unmap (non-index) freeable LEBs. Note that recovery requires that all
	 * wbufs are sync'd before this, which is done in 'do_commit()'.
	 */
	while (1) {
		lp = ubifs_fast_find_freeable(c);
		if (IS_ERR(lp)) {
			err = PTR_ERR(lp);
			goto out;
		}
		if (!lp)
			break;
		ubifs_assert(!(lp->flags & LPROPS_TAKEN));
		ubifs_assert(!(lp->flags & LPROPS_INDEX));
		err = ubifs_leb_unmap(c, lp->lnum);
		if (err)
			goto out;
		lp = ubifs_change_lp(c, lp, c->leb_size, 0, lp->flags, 0);
		if (IS_ERR(lp)) {
			err = PTR_ERR(lp);
			goto out;
		}
		ubifs_assert(!(lp->flags & LPROPS_TAKEN));
		ubifs_assert(!(lp->flags & LPROPS_INDEX));
	}

	/* Mark GC'd index LEBs OK to unmap after this commit finishes */
	list_for_each_entry(idx_gc, &c->idx_gc, list)
		idx_gc->unmap = 1;

	/* Record index freeable LEBs for unmapping after commit */
	while (1) {
		lp = ubifs_fast_find_frdi_idx(c);
		if (IS_ERR(lp)) {
			err = PTR_ERR(lp);
			goto out;
		}
		if (!lp)
			break;
		idx_gc = kmalloc(sizeof(struct ubifs_gced_idx_leb), GFP_NOFS);
		if (!idx_gc) {
			err = -ENOMEM;
			goto out;
		}
		ubifs_assert(!(lp->flags & LPROPS_TAKEN));
		ubifs_assert(lp->flags & LPROPS_INDEX);
		/* Don't release the LEB until after the next commit */
		flags = (lp->flags | LPROPS_TAKEN) ^ LPROPS_INDEX;
		lp = ubifs_change_lp(c, lp, c->leb_size, 0, flags, 1);
		if (IS_ERR(lp)) {
			err = PTR_ERR(lp);
			kfree(idx_gc);
			goto out;
		}
		ubifs_assert(lp->flags & LPROPS_TAKEN);
		ubifs_assert(!(lp->flags & LPROPS_INDEX));
		idx_gc->lnum = lp->lnum;
		idx_gc->unmap = 1;
		list_add(&idx_gc->list, &c->idx_gc);
	}
out:
	ubifs_release_lprops(c);
	return err;
}

/**
 * ubifs_gc_end_commit - garbage collection at end of commit.
 * @c: UBIFS file-system description object
 *
 * This function completes out-of-place garbage collection of index LEBs.
 */
int ubifs_gc_end_commit(struct ubifs_info *c)
{
	struct ubifs_gced_idx_leb *idx_gc, *tmp;
	struct ubifs_wbuf *wbuf;
	int err = 0;

	wbuf = &c->jheads[GCHD].wbuf;
	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
	list_for_each_entry_safe(idx_gc, tmp, &c->idx_gc, list)
		if (idx_gc->unmap) {
			dbg_gc("LEB %d", idx_gc->lnum);
			err = ubifs_leb_unmap(c, idx_gc->lnum);
			if (err)
				goto out;
			err = ubifs_change_one_lp(c, idx_gc->lnum, LPROPS_NC,
					  LPROPS_NC, 0, LPROPS_TAKEN, -1);
			if (err)
				goto out;
			list_del(&idx_gc->list);
			kfree(idx_gc);
		}
out:
	mutex_unlock(&wbuf->io_mutex);
	return err;
}

/**
 * ubifs_destroy_idx_gc - destroy idx_gc list.
 * @c: UBIFS file-system description object
 *
 * This function destroys the @c->idx_gc list. It is called when unmounting or
 * remounting read-only so locks are not needed. Returns zero in case of
 * success and a negative error code in case of failure.
 */
int ubifs_destroy_idx_gc(struct ubifs_info *c)
{
	int ret = 0;

	while (!list_empty(&c->idx_gc)) {
		int err;
		struct ubifs_gced_idx_leb *idx_gc;

		idx_gc = list_entry(c->idx_gc.next, struct ubifs_gced_idx_leb,
				    list);
		err = ubifs_change_one_lp(c, idx_gc->lnum, LPROPS_NC,
					  LPROPS_NC, 0, LPROPS_TAKEN, -1);
		if (err && !ret)
			ret = err;
		list_del(&idx_gc->list);
		kfree(idx_gc);
	}

	return ret;
}

/**
 * ubifs_get_idx_gc_leb - get a LEB from GC'd index LEB list.
 * @c: UBIFS file-system description object
 *
 * Called during start commit so locks are not needed.
 */
int ubifs_get_idx_gc_leb(struct ubifs_info *c)
{
	struct ubifs_gced_idx_leb *idx_gc;
	int lnum;

	if (list_empty(&c->idx_gc))
		return -ENOSPC;
	idx_gc = list_entry(c->idx_gc.next, struct ubifs_gced_idx_leb, list);
	lnum = idx_gc->lnum;
	/* c->idx_gc_cnt is updated by the caller when lprops are updated */
	list_del(&idx_gc->list);
	kfree(idx_gc);
	return lnum;
}
OpenPOWER on IntegriCloud