GNU Linux-libre 4.19.286-gnu1
[releases.git] / drivers / md / persistent-data / dm-space-map-metadata.c
1 /*
2  * Copyright (C) 2011 Red Hat, Inc.
3  *
4  * This file is released under the GPL.
5  */
6
7 #include "dm-space-map.h"
8 #include "dm-space-map-common.h"
9 #include "dm-space-map-metadata.h"
10
11 #include <linux/list.h>
12 #include <linux/slab.h>
13 #include <linux/device-mapper.h>
14 #include <linux/kernel.h>
15
16 #define DM_MSG_PREFIX "space map metadata"
17
18 /*----------------------------------------------------------------*/
19
20 /*
21  * An edge triggered threshold.
22  */
23 struct threshold {
24         bool threshold_set;
25         bool value_set;
26         dm_block_t threshold;
27         dm_block_t current_value;
28         dm_sm_threshold_fn fn;
29         void *context;
30 };
31
32 static void threshold_init(struct threshold *t)
33 {
34         t->threshold_set = false;
35         t->value_set = false;
36 }
37
38 static void set_threshold(struct threshold *t, dm_block_t value,
39                           dm_sm_threshold_fn fn, void *context)
40 {
41         t->threshold_set = true;
42         t->threshold = value;
43         t->fn = fn;
44         t->context = context;
45 }
46
47 static bool below_threshold(struct threshold *t, dm_block_t value)
48 {
49         return t->threshold_set && value <= t->threshold;
50 }
51
52 static bool threshold_already_triggered(struct threshold *t)
53 {
54         return t->value_set && below_threshold(t, t->current_value);
55 }
56
57 static void check_threshold(struct threshold *t, dm_block_t value)
58 {
59         if (below_threshold(t, value) &&
60             !threshold_already_triggered(t))
61                 t->fn(t->context);
62
63         t->value_set = true;
64         t->current_value = value;
65 }
66
67 /*----------------------------------------------------------------*/
68
69 /*
70  * Space map interface.
71  *
72  * The low level disk format is written using the standard btree and
73  * transaction manager.  This means that performing disk operations may
74  * cause us to recurse into the space map in order to allocate new blocks.
75  * For this reason we have a pool of pre-allocated blocks large enough to
76  * service any metadata_ll_disk operation.
77  */
78
79 /*
80  * FIXME: we should calculate this based on the size of the device.
81  * Only the metadata space map needs this functionality.
82  */
83 #define MAX_RECURSIVE_ALLOCATIONS 1024
84
85 enum block_op_type {
86         BOP_INC,
87         BOP_DEC
88 };
89
90 struct block_op {
91         enum block_op_type type;
92         dm_block_t block;
93 };
94
95 struct bop_ring_buffer {
96         unsigned begin;
97         unsigned end;
98         struct block_op bops[MAX_RECURSIVE_ALLOCATIONS + 1];
99 };
100
101 static void brb_init(struct bop_ring_buffer *brb)
102 {
103         brb->begin = 0;
104         brb->end = 0;
105 }
106
107 static bool brb_empty(struct bop_ring_buffer *brb)
108 {
109         return brb->begin == brb->end;
110 }
111
112 static unsigned brb_next(struct bop_ring_buffer *brb, unsigned old)
113 {
114         unsigned r = old + 1;
115         return r >= ARRAY_SIZE(brb->bops) ? 0 : r;
116 }
117
118 static int brb_push(struct bop_ring_buffer *brb,
119                     enum block_op_type type, dm_block_t b)
120 {
121         struct block_op *bop;
122         unsigned next = brb_next(brb, brb->end);
123
124         /*
125          * We don't allow the last bop to be filled, this way we can
126          * differentiate between full and empty.
127          */
128         if (next == brb->begin)
129                 return -ENOMEM;
130
131         bop = brb->bops + brb->end;
132         bop->type = type;
133         bop->block = b;
134
135         brb->end = next;
136
137         return 0;
138 }
139
140 static int brb_peek(struct bop_ring_buffer *brb, struct block_op *result)
141 {
142         struct block_op *bop;
143
144         if (brb_empty(brb))
145                 return -ENODATA;
146
147         bop = brb->bops + brb->begin;
148         result->type = bop->type;
149         result->block = bop->block;
150
151         return 0;
152 }
153
154 static int brb_pop(struct bop_ring_buffer *brb)
155 {
156         if (brb_empty(brb))
157                 return -ENODATA;
158
159         brb->begin = brb_next(brb, brb->begin);
160
161         return 0;
162 }
163
164 /*----------------------------------------------------------------*/
165
166 struct sm_metadata {
167         struct dm_space_map sm;
168
169         struct ll_disk ll;
170         struct ll_disk old_ll;
171
172         dm_block_t begin;
173
174         unsigned recursion_count;
175         unsigned allocated_this_transaction;
176         struct bop_ring_buffer uncommitted;
177
178         struct threshold threshold;
179 };
180
181 static int add_bop(struct sm_metadata *smm, enum block_op_type type, dm_block_t b)
182 {
183         int r = brb_push(&smm->uncommitted, type, b);
184
185         if (r) {
186                 DMERR("too many recursive allocations");
187                 return -ENOMEM;
188         }
189
190         return 0;
191 }
192
193 static int commit_bop(struct sm_metadata *smm, struct block_op *op)
194 {
195         int r = 0;
196         enum allocation_event ev;
197
198         switch (op->type) {
199         case BOP_INC:
200                 r = sm_ll_inc(&smm->ll, op->block, &ev);
201                 break;
202
203         case BOP_DEC:
204                 r = sm_ll_dec(&smm->ll, op->block, &ev);
205                 break;
206         }
207
208         return r;
209 }
210
211 static void in(struct sm_metadata *smm)
212 {
213         smm->recursion_count++;
214 }
215
216 static int apply_bops(struct sm_metadata *smm)
217 {
218         int r = 0;
219
220         while (!brb_empty(&smm->uncommitted)) {
221                 struct block_op bop;
222
223                 r = brb_peek(&smm->uncommitted, &bop);
224                 if (r) {
225                         DMERR("bug in bop ring buffer");
226                         break;
227                 }
228
229                 r = commit_bop(smm, &bop);
230                 if (r)
231                         break;
232
233                 brb_pop(&smm->uncommitted);
234         }
235
236         return r;
237 }
238
239 static int out(struct sm_metadata *smm)
240 {
241         int r = 0;
242
243         /*
244          * If we're not recursing then very bad things are happening.
245          */
246         if (!smm->recursion_count) {
247                 DMERR("lost track of recursion depth");
248                 return -ENOMEM;
249         }
250
251         if (smm->recursion_count == 1)
252                 r = apply_bops(smm);
253
254         smm->recursion_count--;
255
256         return r;
257 }
258
259 /*
260  * When using the out() function above, we often want to combine an error
261  * code for the operation run in the recursive context with that from
262  * out().
263  */
264 static int combine_errors(int r1, int r2)
265 {
266         return r1 ? r1 : r2;
267 }
268
269 static int recursing(struct sm_metadata *smm)
270 {
271         return smm->recursion_count;
272 }
273
274 static void sm_metadata_destroy(struct dm_space_map *sm)
275 {
276         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
277
278         kfree(smm);
279 }
280
281 static int sm_metadata_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
282 {
283         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
284
285         *count = smm->ll.nr_blocks;
286
287         return 0;
288 }
289
290 static int sm_metadata_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
291 {
292         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
293
294         *count = smm->old_ll.nr_blocks - smm->old_ll.nr_allocated -
295                  smm->allocated_this_transaction;
296
297         return 0;
298 }
299
300 static int sm_metadata_get_count(struct dm_space_map *sm, dm_block_t b,
301                                  uint32_t *result)
302 {
303         int r;
304         unsigned i;
305         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
306         unsigned adjustment = 0;
307
308         /*
309          * We may have some uncommitted adjustments to add.  This list
310          * should always be really short.
311          */
312         for (i = smm->uncommitted.begin;
313              i != smm->uncommitted.end;
314              i = brb_next(&smm->uncommitted, i)) {
315                 struct block_op *op = smm->uncommitted.bops + i;
316
317                 if (op->block != b)
318                         continue;
319
320                 switch (op->type) {
321                 case BOP_INC:
322                         adjustment++;
323                         break;
324
325                 case BOP_DEC:
326                         adjustment--;
327                         break;
328                 }
329         }
330
331         r = sm_ll_lookup(&smm->ll, b, result);
332         if (r)
333                 return r;
334
335         *result += adjustment;
336
337         return 0;
338 }
339
340 static int sm_metadata_count_is_more_than_one(struct dm_space_map *sm,
341                                               dm_block_t b, int *result)
342 {
343         int r, adjustment = 0;
344         unsigned i;
345         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
346         uint32_t rc;
347
348         /*
349          * We may have some uncommitted adjustments to add.  This list
350          * should always be really short.
351          */
352         for (i = smm->uncommitted.begin;
353              i != smm->uncommitted.end;
354              i = brb_next(&smm->uncommitted, i)) {
355
356                 struct block_op *op = smm->uncommitted.bops + i;
357
358                 if (op->block != b)
359                         continue;
360
361                 switch (op->type) {
362                 case BOP_INC:
363                         adjustment++;
364                         break;
365
366                 case BOP_DEC:
367                         adjustment--;
368                         break;
369                 }
370         }
371
372         if (adjustment > 1) {
373                 *result = 1;
374                 return 0;
375         }
376
377         r = sm_ll_lookup_bitmap(&smm->ll, b, &rc);
378         if (r)
379                 return r;
380
381         if (rc == 3)
382                 /*
383                  * We err on the side of caution, and always return true.
384                  */
385                 *result = 1;
386         else
387                 *result = rc + adjustment > 1;
388
389         return 0;
390 }
391
392 static int sm_metadata_set_count(struct dm_space_map *sm, dm_block_t b,
393                                  uint32_t count)
394 {
395         int r, r2;
396         enum allocation_event ev;
397         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
398
399         if (smm->recursion_count) {
400                 DMERR("cannot recurse set_count()");
401                 return -EINVAL;
402         }
403
404         in(smm);
405         r = sm_ll_insert(&smm->ll, b, count, &ev);
406         r2 = out(smm);
407
408         return combine_errors(r, r2);
409 }
410
411 static int sm_metadata_inc_block(struct dm_space_map *sm, dm_block_t b)
412 {
413         int r, r2 = 0;
414         enum allocation_event ev;
415         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
416
417         if (recursing(smm))
418                 r = add_bop(smm, BOP_INC, b);
419         else {
420                 in(smm);
421                 r = sm_ll_inc(&smm->ll, b, &ev);
422                 r2 = out(smm);
423         }
424
425         return combine_errors(r, r2);
426 }
427
428 static int sm_metadata_dec_block(struct dm_space_map *sm, dm_block_t b)
429 {
430         int r, r2 = 0;
431         enum allocation_event ev;
432         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
433
434         if (recursing(smm))
435                 r = add_bop(smm, BOP_DEC, b);
436         else {
437                 in(smm);
438                 r = sm_ll_dec(&smm->ll, b, &ev);
439                 r2 = out(smm);
440         }
441
442         return combine_errors(r, r2);
443 }
444
445 static int sm_metadata_new_block_(struct dm_space_map *sm, dm_block_t *b)
446 {
447         int r, r2 = 0;
448         enum allocation_event ev;
449         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
450
451         /*
452          * Any block we allocate has to be free in both the old and current ll.
453          */
454         r = sm_ll_find_common_free_block(&smm->old_ll, &smm->ll, smm->begin, smm->ll.nr_blocks, b);
455         if (r == -ENOSPC) {
456                 /*
457                  * There's no free block between smm->begin and the end of the metadata device.
458                  * We search before smm->begin in case something has been freed.
459                  */
460                 r = sm_ll_find_common_free_block(&smm->old_ll, &smm->ll, 0, smm->begin, b);
461         }
462
463         if (r)
464                 return r;
465
466         smm->begin = *b + 1;
467
468         if (recursing(smm))
469                 r = add_bop(smm, BOP_INC, *b);
470         else {
471                 in(smm);
472                 r = sm_ll_inc(&smm->ll, *b, &ev);
473                 r2 = out(smm);
474         }
475
476         if (!r)
477                 smm->allocated_this_transaction++;
478
479         return combine_errors(r, r2);
480 }
481
482 static int sm_metadata_new_block(struct dm_space_map *sm, dm_block_t *b)
483 {
484         dm_block_t count;
485         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
486
487         int r = sm_metadata_new_block_(sm, b);
488         if (r) {
489                 DMERR_LIMIT("unable to allocate new metadata block");
490                 return r;
491         }
492
493         r = sm_metadata_get_nr_free(sm, &count);
494         if (r) {
495                 DMERR_LIMIT("couldn't get free block count");
496                 return r;
497         }
498
499         check_threshold(&smm->threshold, count);
500
501         return r;
502 }
503
504 static int sm_metadata_commit(struct dm_space_map *sm)
505 {
506         int r;
507         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
508
509         r = sm_ll_commit(&smm->ll);
510         if (r)
511                 return r;
512
513         memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
514         smm->allocated_this_transaction = 0;
515
516         return 0;
517 }
518
519 static int sm_metadata_register_threshold_callback(struct dm_space_map *sm,
520                                                    dm_block_t threshold,
521                                                    dm_sm_threshold_fn fn,
522                                                    void *context)
523 {
524         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
525
526         set_threshold(&smm->threshold, threshold, fn, context);
527
528         return 0;
529 }
530
531 static int sm_metadata_root_size(struct dm_space_map *sm, size_t *result)
532 {
533         *result = sizeof(struct disk_sm_root);
534
535         return 0;
536 }
537
538 static int sm_metadata_copy_root(struct dm_space_map *sm, void *where_le, size_t max)
539 {
540         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
541         struct disk_sm_root root_le;
542
543         root_le.nr_blocks = cpu_to_le64(smm->ll.nr_blocks);
544         root_le.nr_allocated = cpu_to_le64(smm->ll.nr_allocated);
545         root_le.bitmap_root = cpu_to_le64(smm->ll.bitmap_root);
546         root_le.ref_count_root = cpu_to_le64(smm->ll.ref_count_root);
547
548         if (max < sizeof(root_le))
549                 return -ENOSPC;
550
551         memcpy(where_le, &root_le, sizeof(root_le));
552
553         return 0;
554 }
555
556 static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks);
557
558 static const struct dm_space_map ops = {
559         .destroy = sm_metadata_destroy,
560         .extend = sm_metadata_extend,
561         .get_nr_blocks = sm_metadata_get_nr_blocks,
562         .get_nr_free = sm_metadata_get_nr_free,
563         .get_count = sm_metadata_get_count,
564         .count_is_more_than_one = sm_metadata_count_is_more_than_one,
565         .set_count = sm_metadata_set_count,
566         .inc_block = sm_metadata_inc_block,
567         .dec_block = sm_metadata_dec_block,
568         .new_block = sm_metadata_new_block,
569         .commit = sm_metadata_commit,
570         .root_size = sm_metadata_root_size,
571         .copy_root = sm_metadata_copy_root,
572         .register_threshold_callback = sm_metadata_register_threshold_callback
573 };
574
575 /*----------------------------------------------------------------*/
576
577 /*
578  * When a new space map is created that manages its own space.  We use
579  * this tiny bootstrap allocator.
580  */
581 static void sm_bootstrap_destroy(struct dm_space_map *sm)
582 {
583 }
584
585 static int sm_bootstrap_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
586 {
587         DMERR("bootstrap doesn't support extend");
588
589         return -EINVAL;
590 }
591
592 static int sm_bootstrap_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
593 {
594         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
595
596         *count = smm->ll.nr_blocks;
597
598         return 0;
599 }
600
601 static int sm_bootstrap_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
602 {
603         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
604
605         *count = smm->ll.nr_blocks - smm->begin;
606
607         return 0;
608 }
609
610 static int sm_bootstrap_get_count(struct dm_space_map *sm, dm_block_t b,
611                                   uint32_t *result)
612 {
613         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
614
615         *result = (b < smm->begin) ? 1 : 0;
616
617         return 0;
618 }
619
620 static int sm_bootstrap_count_is_more_than_one(struct dm_space_map *sm,
621                                                dm_block_t b, int *result)
622 {
623         *result = 0;
624
625         return 0;
626 }
627
628 static int sm_bootstrap_set_count(struct dm_space_map *sm, dm_block_t b,
629                                   uint32_t count)
630 {
631         DMERR("bootstrap doesn't support set_count");
632
633         return -EINVAL;
634 }
635
636 static int sm_bootstrap_new_block(struct dm_space_map *sm, dm_block_t *b)
637 {
638         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
639
640         /*
641          * We know the entire device is unused.
642          */
643         if (smm->begin == smm->ll.nr_blocks)
644                 return -ENOSPC;
645
646         *b = smm->begin++;
647
648         return 0;
649 }
650
651 static int sm_bootstrap_inc_block(struct dm_space_map *sm, dm_block_t b)
652 {
653         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
654
655         return add_bop(smm, BOP_INC, b);
656 }
657
658 static int sm_bootstrap_dec_block(struct dm_space_map *sm, dm_block_t b)
659 {
660         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
661
662         return add_bop(smm, BOP_DEC, b);
663 }
664
665 static int sm_bootstrap_commit(struct dm_space_map *sm)
666 {
667         return 0;
668 }
669
670 static int sm_bootstrap_root_size(struct dm_space_map *sm, size_t *result)
671 {
672         DMERR("bootstrap doesn't support root_size");
673
674         return -EINVAL;
675 }
676
677 static int sm_bootstrap_copy_root(struct dm_space_map *sm, void *where,
678                                   size_t max)
679 {
680         DMERR("bootstrap doesn't support copy_root");
681
682         return -EINVAL;
683 }
684
685 static const struct dm_space_map bootstrap_ops = {
686         .destroy = sm_bootstrap_destroy,
687         .extend = sm_bootstrap_extend,
688         .get_nr_blocks = sm_bootstrap_get_nr_blocks,
689         .get_nr_free = sm_bootstrap_get_nr_free,
690         .get_count = sm_bootstrap_get_count,
691         .count_is_more_than_one = sm_bootstrap_count_is_more_than_one,
692         .set_count = sm_bootstrap_set_count,
693         .inc_block = sm_bootstrap_inc_block,
694         .dec_block = sm_bootstrap_dec_block,
695         .new_block = sm_bootstrap_new_block,
696         .commit = sm_bootstrap_commit,
697         .root_size = sm_bootstrap_root_size,
698         .copy_root = sm_bootstrap_copy_root,
699         .register_threshold_callback = NULL
700 };
701
702 /*----------------------------------------------------------------*/
703
704 static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
705 {
706         int r, i;
707         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
708         dm_block_t old_len = smm->ll.nr_blocks;
709
710         /*
711          * Flick into a mode where all blocks get allocated in the new area.
712          */
713         smm->begin = old_len;
714         memcpy(sm, &bootstrap_ops, sizeof(*sm));
715
716         /*
717          * Extend.
718          */
719         r = sm_ll_extend(&smm->ll, extra_blocks);
720         if (r)
721                 goto out;
722
723         /*
724          * We repeatedly increment then commit until the commit doesn't
725          * allocate any new blocks.
726          */
727         do {
728                 for (i = old_len; !r && i < smm->begin; i++)
729                         r = add_bop(smm, BOP_INC, i);
730
731                 if (r)
732                         goto out;
733
734                 old_len = smm->begin;
735
736                 r = apply_bops(smm);
737                 if (r) {
738                         DMERR("%s: apply_bops failed", __func__);
739                         goto out;
740                 }
741
742                 r = sm_ll_commit(&smm->ll);
743                 if (r)
744                         goto out;
745
746         } while (old_len != smm->begin);
747
748 out:
749         /*
750          * Switch back to normal behaviour.
751          */
752         memcpy(sm, &ops, sizeof(*sm));
753         return r;
754 }
755
756 /*----------------------------------------------------------------*/
757
758 struct dm_space_map *dm_sm_metadata_init(void)
759 {
760         struct sm_metadata *smm;
761
762         smm = kmalloc(sizeof(*smm), GFP_KERNEL);
763         if (!smm)
764                 return ERR_PTR(-ENOMEM);
765
766         memcpy(&smm->sm, &ops, sizeof(smm->sm));
767
768         return &smm->sm;
769 }
770
771 int dm_sm_metadata_create(struct dm_space_map *sm,
772                           struct dm_transaction_manager *tm,
773                           dm_block_t nr_blocks,
774                           dm_block_t superblock)
775 {
776         int r;
777         dm_block_t i;
778         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
779
780         smm->begin = superblock + 1;
781         smm->recursion_count = 0;
782         smm->allocated_this_transaction = 0;
783         brb_init(&smm->uncommitted);
784         threshold_init(&smm->threshold);
785
786         memcpy(&smm->sm, &bootstrap_ops, sizeof(smm->sm));
787
788         r = sm_ll_new_metadata(&smm->ll, tm);
789         if (!r) {
790                 if (nr_blocks > DM_SM_METADATA_MAX_BLOCKS)
791                         nr_blocks = DM_SM_METADATA_MAX_BLOCKS;
792                 r = sm_ll_extend(&smm->ll, nr_blocks);
793         }
794         memcpy(&smm->sm, &ops, sizeof(smm->sm));
795         if (r)
796                 return r;
797
798         /*
799          * Now we need to update the newly created data structures with the
800          * allocated blocks that they were built from.
801          */
802         for (i = superblock; !r && i < smm->begin; i++)
803                 r = add_bop(smm, BOP_INC, i);
804
805         if (r)
806                 return r;
807
808         r = apply_bops(smm);
809         if (r) {
810                 DMERR("%s: apply_bops failed", __func__);
811                 return r;
812         }
813
814         return sm_metadata_commit(sm);
815 }
816
817 int dm_sm_metadata_open(struct dm_space_map *sm,
818                         struct dm_transaction_manager *tm,
819                         void *root_le, size_t len)
820 {
821         int r;
822         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
823
824         r = sm_ll_open_metadata(&smm->ll, tm, root_le, len);
825         if (r)
826                 return r;
827
828         smm->begin = 0;
829         smm->recursion_count = 0;
830         smm->allocated_this_transaction = 0;
831         brb_init(&smm->uncommitted);
832         threshold_init(&smm->threshold);
833
834         memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
835         return 0;
836 }