e8a9042988871c2fcfb12b8a85c20b8c8a4629c3
[kvmfornfv.git] / kernel / 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
15 #define DM_MSG_PREFIX "space map metadata"
16
17 /*----------------------------------------------------------------*/
18
19 /*
20  * An edge triggered threshold.
21  */
22 struct threshold {
23         bool threshold_set;
24         bool value_set;
25         dm_block_t threshold;
26         dm_block_t current_value;
27         dm_sm_threshold_fn fn;
28         void *context;
29 };
30
31 static void threshold_init(struct threshold *t)
32 {
33         t->threshold_set = false;
34         t->value_set = false;
35 }
36
37 static void set_threshold(struct threshold *t, dm_block_t value,
38                           dm_sm_threshold_fn fn, void *context)
39 {
40         t->threshold_set = true;
41         t->threshold = value;
42         t->fn = fn;
43         t->context = context;
44 }
45
46 static bool below_threshold(struct threshold *t, dm_block_t value)
47 {
48         return t->threshold_set && value <= t->threshold;
49 }
50
51 static bool threshold_already_triggered(struct threshold *t)
52 {
53         return t->value_set && below_threshold(t, t->current_value);
54 }
55
56 static void check_threshold(struct threshold *t, dm_block_t value)
57 {
58         if (below_threshold(t, value) &&
59             !threshold_already_triggered(t))
60                 t->fn(t->context);
61
62         t->value_set = true;
63         t->current_value = value;
64 }
65
66 /*----------------------------------------------------------------*/
67
68 /*
69  * Space map interface.
70  *
71  * The low level disk format is written using the standard btree and
72  * transaction manager.  This means that performing disk operations may
73  * cause us to recurse into the space map in order to allocate new blocks.
74  * For this reason we have a pool of pre-allocated blocks large enough to
75  * service any metadata_ll_disk operation.
76  */
77
78 /*
79  * FIXME: we should calculate this based on the size of the device.
80  * Only the metadata space map needs this functionality.
81  */
82 #define MAX_RECURSIVE_ALLOCATIONS 1024
83
84 enum block_op_type {
85         BOP_INC,
86         BOP_DEC
87 };
88
89 struct block_op {
90         enum block_op_type type;
91         dm_block_t block;
92 };
93
94 struct bop_ring_buffer {
95         unsigned begin;
96         unsigned end;
97         struct block_op bops[MAX_RECURSIVE_ALLOCATIONS + 1];
98 };
99
100 static void brb_init(struct bop_ring_buffer *brb)
101 {
102         brb->begin = 0;
103         brb->end = 0;
104 }
105
106 static bool brb_empty(struct bop_ring_buffer *brb)
107 {
108         return brb->begin == brb->end;
109 }
110
111 static unsigned brb_next(struct bop_ring_buffer *brb, unsigned old)
112 {
113         unsigned r = old + 1;
114         return (r >= (sizeof(brb->bops) / sizeof(*brb->bops))) ? 0 : r;
115 }
116
117 static int brb_push(struct bop_ring_buffer *brb,
118                     enum block_op_type type, dm_block_t b)
119 {
120         struct block_op *bop;
121         unsigned next = brb_next(brb, brb->end);
122
123         /*
124          * We don't allow the last bop to be filled, this way we can
125          * differentiate between full and empty.
126          */
127         if (next == brb->begin)
128                 return -ENOMEM;
129
130         bop = brb->bops + brb->end;
131         bop->type = type;
132         bop->block = b;
133
134         brb->end = next;
135
136         return 0;
137 }
138
139 static int brb_pop(struct bop_ring_buffer *brb, struct block_op *result)
140 {
141         struct block_op *bop;
142
143         if (brb_empty(brb))
144                 return -ENODATA;
145
146         bop = brb->bops + brb->begin;
147         result->type = bop->type;
148         result->block = bop->block;
149
150         brb->begin = brb_next(brb, brb->begin);
151
152         return 0;
153 }
154
155 /*----------------------------------------------------------------*/
156
157 struct sm_metadata {
158         struct dm_space_map sm;
159
160         struct ll_disk ll;
161         struct ll_disk old_ll;
162
163         dm_block_t begin;
164
165         unsigned recursion_count;
166         unsigned allocated_this_transaction;
167         struct bop_ring_buffer uncommitted;
168
169         struct threshold threshold;
170 };
171
172 static int add_bop(struct sm_metadata *smm, enum block_op_type type, dm_block_t b)
173 {
174         int r = brb_push(&smm->uncommitted, type, b);
175
176         if (r) {
177                 DMERR("too many recursive allocations");
178                 return -ENOMEM;
179         }
180
181         return 0;
182 }
183
184 static int commit_bop(struct sm_metadata *smm, struct block_op *op)
185 {
186         int r = 0;
187         enum allocation_event ev;
188
189         switch (op->type) {
190         case BOP_INC:
191                 r = sm_ll_inc(&smm->ll, op->block, &ev);
192                 break;
193
194         case BOP_DEC:
195                 r = sm_ll_dec(&smm->ll, op->block, &ev);
196                 break;
197         }
198
199         return r;
200 }
201
202 static void in(struct sm_metadata *smm)
203 {
204         smm->recursion_count++;
205 }
206
207 static int out(struct sm_metadata *smm)
208 {
209         int r = 0;
210
211         /*
212          * If we're not recursing then very bad things are happening.
213          */
214         if (!smm->recursion_count) {
215                 DMERR("lost track of recursion depth");
216                 return -ENOMEM;
217         }
218
219         if (smm->recursion_count == 1) {
220                 while (!brb_empty(&smm->uncommitted)) {
221                         struct block_op bop;
222
223                         r = brb_pop(&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         }
234
235         smm->recursion_count--;
236
237         return r;
238 }
239
240 /*
241  * When using the out() function above, we often want to combine an error
242  * code for the operation run in the recursive context with that from
243  * out().
244  */
245 static int combine_errors(int r1, int r2)
246 {
247         return r1 ? r1 : r2;
248 }
249
250 static int recursing(struct sm_metadata *smm)
251 {
252         return smm->recursion_count;
253 }
254
255 static void sm_metadata_destroy(struct dm_space_map *sm)
256 {
257         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
258
259         kfree(smm);
260 }
261
262 static int sm_metadata_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
263 {
264         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
265
266         *count = smm->ll.nr_blocks;
267
268         return 0;
269 }
270
271 static int sm_metadata_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
272 {
273         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
274
275         *count = smm->old_ll.nr_blocks - smm->old_ll.nr_allocated -
276                  smm->allocated_this_transaction;
277
278         return 0;
279 }
280
281 static int sm_metadata_get_count(struct dm_space_map *sm, dm_block_t b,
282                                  uint32_t *result)
283 {
284         int r;
285         unsigned i;
286         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
287         unsigned adjustment = 0;
288
289         /*
290          * We may have some uncommitted adjustments to add.  This list
291          * should always be really short.
292          */
293         for (i = smm->uncommitted.begin;
294              i != smm->uncommitted.end;
295              i = brb_next(&smm->uncommitted, i)) {
296                 struct block_op *op = smm->uncommitted.bops + i;
297
298                 if (op->block != b)
299                         continue;
300
301                 switch (op->type) {
302                 case BOP_INC:
303                         adjustment++;
304                         break;
305
306                 case BOP_DEC:
307                         adjustment--;
308                         break;
309                 }
310         }
311
312         r = sm_ll_lookup(&smm->ll, b, result);
313         if (r)
314                 return r;
315
316         *result += adjustment;
317
318         return 0;
319 }
320
321 static int sm_metadata_count_is_more_than_one(struct dm_space_map *sm,
322                                               dm_block_t b, int *result)
323 {
324         int r, adjustment = 0;
325         unsigned i;
326         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
327         uint32_t rc;
328
329         /*
330          * We may have some uncommitted adjustments to add.  This list
331          * should always be really short.
332          */
333         for (i = smm->uncommitted.begin;
334              i != smm->uncommitted.end;
335              i = brb_next(&smm->uncommitted, i)) {
336
337                 struct block_op *op = smm->uncommitted.bops + i;
338
339                 if (op->block != b)
340                         continue;
341
342                 switch (op->type) {
343                 case BOP_INC:
344                         adjustment++;
345                         break;
346
347                 case BOP_DEC:
348                         adjustment--;
349                         break;
350                 }
351         }
352
353         if (adjustment > 1) {
354                 *result = 1;
355                 return 0;
356         }
357
358         r = sm_ll_lookup_bitmap(&smm->ll, b, &rc);
359         if (r)
360                 return r;
361
362         if (rc == 3)
363                 /*
364                  * We err on the side of caution, and always return true.
365                  */
366                 *result = 1;
367         else
368                 *result = rc + adjustment > 1;
369
370         return 0;
371 }
372
373 static int sm_metadata_set_count(struct dm_space_map *sm, dm_block_t b,
374                                  uint32_t count)
375 {
376         int r, r2;
377         enum allocation_event ev;
378         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
379
380         if (smm->recursion_count) {
381                 DMERR("cannot recurse set_count()");
382                 return -EINVAL;
383         }
384
385         in(smm);
386         r = sm_ll_insert(&smm->ll, b, count, &ev);
387         r2 = out(smm);
388
389         return combine_errors(r, r2);
390 }
391
392 static int sm_metadata_inc_block(struct dm_space_map *sm, dm_block_t b)
393 {
394         int r, r2 = 0;
395         enum allocation_event ev;
396         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
397
398         if (recursing(smm))
399                 r = add_bop(smm, BOP_INC, b);
400         else {
401                 in(smm);
402                 r = sm_ll_inc(&smm->ll, b, &ev);
403                 r2 = out(smm);
404         }
405
406         return combine_errors(r, r2);
407 }
408
409 static int sm_metadata_dec_block(struct dm_space_map *sm, dm_block_t b)
410 {
411         int r, r2 = 0;
412         enum allocation_event ev;
413         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
414
415         if (recursing(smm))
416                 r = add_bop(smm, BOP_DEC, b);
417         else {
418                 in(smm);
419                 r = sm_ll_dec(&smm->ll, b, &ev);
420                 r2 = out(smm);
421         }
422
423         return combine_errors(r, r2);
424 }
425
426 static int sm_metadata_new_block_(struct dm_space_map *sm, dm_block_t *b)
427 {
428         int r, r2 = 0;
429         enum allocation_event ev;
430         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
431
432         r = sm_ll_find_free_block(&smm->old_ll, smm->begin, smm->old_ll.nr_blocks, b);
433         if (r)
434                 return r;
435
436         smm->begin = *b + 1;
437
438         if (recursing(smm))
439                 r = add_bop(smm, BOP_INC, *b);
440         else {
441                 in(smm);
442                 r = sm_ll_inc(&smm->ll, *b, &ev);
443                 r2 = out(smm);
444         }
445
446         if (!r)
447                 smm->allocated_this_transaction++;
448
449         return combine_errors(r, r2);
450 }
451
452 static int sm_metadata_new_block(struct dm_space_map *sm, dm_block_t *b)
453 {
454         dm_block_t count;
455         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
456
457         int r = sm_metadata_new_block_(sm, b);
458         if (r) {
459                 DMERR_LIMIT("unable to allocate new metadata block");
460                 return r;
461         }
462
463         r = sm_metadata_get_nr_free(sm, &count);
464         if (r) {
465                 DMERR_LIMIT("couldn't get free block count");
466                 return r;
467         }
468
469         check_threshold(&smm->threshold, count);
470
471         return r;
472 }
473
474 static int sm_metadata_commit(struct dm_space_map *sm)
475 {
476         int r;
477         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
478
479         r = sm_ll_commit(&smm->ll);
480         if (r)
481                 return r;
482
483         memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
484         smm->begin = 0;
485         smm->allocated_this_transaction = 0;
486
487         return 0;
488 }
489
490 static int sm_metadata_register_threshold_callback(struct dm_space_map *sm,
491                                                    dm_block_t threshold,
492                                                    dm_sm_threshold_fn fn,
493                                                    void *context)
494 {
495         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
496
497         set_threshold(&smm->threshold, threshold, fn, context);
498
499         return 0;
500 }
501
502 static int sm_metadata_root_size(struct dm_space_map *sm, size_t *result)
503 {
504         *result = sizeof(struct disk_sm_root);
505
506         return 0;
507 }
508
509 static int sm_metadata_copy_root(struct dm_space_map *sm, void *where_le, size_t max)
510 {
511         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
512         struct disk_sm_root root_le;
513
514         root_le.nr_blocks = cpu_to_le64(smm->ll.nr_blocks);
515         root_le.nr_allocated = cpu_to_le64(smm->ll.nr_allocated);
516         root_le.bitmap_root = cpu_to_le64(smm->ll.bitmap_root);
517         root_le.ref_count_root = cpu_to_le64(smm->ll.ref_count_root);
518
519         if (max < sizeof(root_le))
520                 return -ENOSPC;
521
522         memcpy(where_le, &root_le, sizeof(root_le));
523
524         return 0;
525 }
526
527 static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks);
528
529 static struct dm_space_map ops = {
530         .destroy = sm_metadata_destroy,
531         .extend = sm_metadata_extend,
532         .get_nr_blocks = sm_metadata_get_nr_blocks,
533         .get_nr_free = sm_metadata_get_nr_free,
534         .get_count = sm_metadata_get_count,
535         .count_is_more_than_one = sm_metadata_count_is_more_than_one,
536         .set_count = sm_metadata_set_count,
537         .inc_block = sm_metadata_inc_block,
538         .dec_block = sm_metadata_dec_block,
539         .new_block = sm_metadata_new_block,
540         .commit = sm_metadata_commit,
541         .root_size = sm_metadata_root_size,
542         .copy_root = sm_metadata_copy_root,
543         .register_threshold_callback = sm_metadata_register_threshold_callback
544 };
545
546 /*----------------------------------------------------------------*/
547
548 /*
549  * When a new space map is created that manages its own space.  We use
550  * this tiny bootstrap allocator.
551  */
552 static void sm_bootstrap_destroy(struct dm_space_map *sm)
553 {
554 }
555
556 static int sm_bootstrap_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
557 {
558         DMERR("bootstrap doesn't support extend");
559
560         return -EINVAL;
561 }
562
563 static int sm_bootstrap_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
564 {
565         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
566
567         *count = smm->ll.nr_blocks;
568
569         return 0;
570 }
571
572 static int sm_bootstrap_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
573 {
574         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
575
576         *count = smm->ll.nr_blocks - smm->begin;
577
578         return 0;
579 }
580
581 static int sm_bootstrap_get_count(struct dm_space_map *sm, dm_block_t b,
582                                   uint32_t *result)
583 {
584         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
585
586         *result = (b < smm->begin) ? 1 : 0;
587
588         return 0;
589 }
590
591 static int sm_bootstrap_count_is_more_than_one(struct dm_space_map *sm,
592                                                dm_block_t b, int *result)
593 {
594         *result = 0;
595
596         return 0;
597 }
598
599 static int sm_bootstrap_set_count(struct dm_space_map *sm, dm_block_t b,
600                                   uint32_t count)
601 {
602         DMERR("bootstrap doesn't support set_count");
603
604         return -EINVAL;
605 }
606
607 static int sm_bootstrap_new_block(struct dm_space_map *sm, dm_block_t *b)
608 {
609         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
610
611         /*
612          * We know the entire device is unused.
613          */
614         if (smm->begin == smm->ll.nr_blocks)
615                 return -ENOSPC;
616
617         *b = smm->begin++;
618
619         return 0;
620 }
621
622 static int sm_bootstrap_inc_block(struct dm_space_map *sm, dm_block_t b)
623 {
624         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
625
626         return add_bop(smm, BOP_INC, b);
627 }
628
629 static int sm_bootstrap_dec_block(struct dm_space_map *sm, dm_block_t b)
630 {
631         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
632
633         return add_bop(smm, BOP_DEC, b);
634 }
635
636 static int sm_bootstrap_commit(struct dm_space_map *sm)
637 {
638         return 0;
639 }
640
641 static int sm_bootstrap_root_size(struct dm_space_map *sm, size_t *result)
642 {
643         DMERR("bootstrap doesn't support root_size");
644
645         return -EINVAL;
646 }
647
648 static int sm_bootstrap_copy_root(struct dm_space_map *sm, void *where,
649                                   size_t max)
650 {
651         DMERR("bootstrap doesn't support copy_root");
652
653         return -EINVAL;
654 }
655
656 static struct dm_space_map bootstrap_ops = {
657         .destroy = sm_bootstrap_destroy,
658         .extend = sm_bootstrap_extend,
659         .get_nr_blocks = sm_bootstrap_get_nr_blocks,
660         .get_nr_free = sm_bootstrap_get_nr_free,
661         .get_count = sm_bootstrap_get_count,
662         .count_is_more_than_one = sm_bootstrap_count_is_more_than_one,
663         .set_count = sm_bootstrap_set_count,
664         .inc_block = sm_bootstrap_inc_block,
665         .dec_block = sm_bootstrap_dec_block,
666         .new_block = sm_bootstrap_new_block,
667         .commit = sm_bootstrap_commit,
668         .root_size = sm_bootstrap_root_size,
669         .copy_root = sm_bootstrap_copy_root,
670         .register_threshold_callback = NULL
671 };
672
673 /*----------------------------------------------------------------*/
674
675 static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
676 {
677         int r, i;
678         enum allocation_event ev;
679         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
680         dm_block_t old_len = smm->ll.nr_blocks;
681
682         /*
683          * Flick into a mode where all blocks get allocated in the new area.
684          */
685         smm->begin = old_len;
686         memcpy(sm, &bootstrap_ops, sizeof(*sm));
687
688         /*
689          * Extend.
690          */
691         r = sm_ll_extend(&smm->ll, extra_blocks);
692         if (r)
693                 goto out;
694
695         /*
696          * We repeatedly increment then commit until the commit doesn't
697          * allocate any new blocks.
698          */
699         do {
700                 for (i = old_len; !r && i < smm->begin; i++) {
701                         r = sm_ll_inc(&smm->ll, i, &ev);
702                         if (r)
703                                 goto out;
704                 }
705                 old_len = smm->begin;
706
707                 r = sm_ll_commit(&smm->ll);
708                 if (r)
709                         goto out;
710
711         } while (old_len != smm->begin);
712
713 out:
714         /*
715          * Switch back to normal behaviour.
716          */
717         memcpy(sm, &ops, sizeof(*sm));
718         return r;
719 }
720
721 /*----------------------------------------------------------------*/
722
723 struct dm_space_map *dm_sm_metadata_init(void)
724 {
725         struct sm_metadata *smm;
726
727         smm = kmalloc(sizeof(*smm), GFP_KERNEL);
728         if (!smm)
729                 return ERR_PTR(-ENOMEM);
730
731         memcpy(&smm->sm, &ops, sizeof(smm->sm));
732
733         return &smm->sm;
734 }
735
736 int dm_sm_metadata_create(struct dm_space_map *sm,
737                           struct dm_transaction_manager *tm,
738                           dm_block_t nr_blocks,
739                           dm_block_t superblock)
740 {
741         int r;
742         dm_block_t i;
743         enum allocation_event ev;
744         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
745
746         smm->begin = superblock + 1;
747         smm->recursion_count = 0;
748         smm->allocated_this_transaction = 0;
749         brb_init(&smm->uncommitted);
750         threshold_init(&smm->threshold);
751
752         memcpy(&smm->sm, &bootstrap_ops, sizeof(smm->sm));
753
754         r = sm_ll_new_metadata(&smm->ll, tm);
755         if (r)
756                 return r;
757
758         if (nr_blocks > DM_SM_METADATA_MAX_BLOCKS)
759                 nr_blocks = DM_SM_METADATA_MAX_BLOCKS;
760         r = sm_ll_extend(&smm->ll, nr_blocks);
761         if (r)
762                 return r;
763
764         memcpy(&smm->sm, &ops, sizeof(smm->sm));
765
766         /*
767          * Now we need to update the newly created data structures with the
768          * allocated blocks that they were built from.
769          */
770         for (i = superblock; !r && i < smm->begin; i++)
771                 r = sm_ll_inc(&smm->ll, i, &ev);
772
773         if (r)
774                 return r;
775
776         return sm_metadata_commit(sm);
777 }
778
779 int dm_sm_metadata_open(struct dm_space_map *sm,
780                         struct dm_transaction_manager *tm,
781                         void *root_le, size_t len)
782 {
783         int r;
784         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
785
786         r = sm_ll_open_metadata(&smm->ll, tm, root_le, len);
787         if (r)
788                 return r;
789
790         smm->begin = 0;
791         smm->recursion_count = 0;
792         smm->allocated_this_transaction = 0;
793         brb_init(&smm->uncommitted);
794         threshold_init(&smm->threshold);
795
796         memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
797         return 0;
798 }