GNU Linux-libre 4.4.294-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
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_peek(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         return 0;
151 }
152
153 static int brb_pop(struct bop_ring_buffer *brb)
154 {
155         if (brb_empty(brb))
156                 return -ENODATA;
157
158         brb->begin = brb_next(brb, brb->begin);
159
160         return 0;
161 }
162
163 /*----------------------------------------------------------------*/
164
165 struct sm_metadata {
166         struct dm_space_map sm;
167
168         struct ll_disk ll;
169         struct ll_disk old_ll;
170
171         dm_block_t begin;
172
173         unsigned recursion_count;
174         unsigned allocated_this_transaction;
175         struct bop_ring_buffer uncommitted;
176
177         struct threshold threshold;
178 };
179
180 static int add_bop(struct sm_metadata *smm, enum block_op_type type, dm_block_t b)
181 {
182         int r = brb_push(&smm->uncommitted, type, b);
183
184         if (r) {
185                 DMERR("too many recursive allocations");
186                 return -ENOMEM;
187         }
188
189         return 0;
190 }
191
192 static int commit_bop(struct sm_metadata *smm, struct block_op *op)
193 {
194         int r = 0;
195         enum allocation_event ev;
196
197         switch (op->type) {
198         case BOP_INC:
199                 r = sm_ll_inc(&smm->ll, op->block, &ev);
200                 break;
201
202         case BOP_DEC:
203                 r = sm_ll_dec(&smm->ll, op->block, &ev);
204                 break;
205         }
206
207         return r;
208 }
209
210 static void in(struct sm_metadata *smm)
211 {
212         smm->recursion_count++;
213 }
214
215 static int apply_bops(struct sm_metadata *smm)
216 {
217         int r = 0;
218
219         while (!brb_empty(&smm->uncommitted)) {
220                 struct block_op bop;
221
222                 r = brb_peek(&smm->uncommitted, &bop);
223                 if (r) {
224                         DMERR("bug in bop ring buffer");
225                         break;
226                 }
227
228                 r = commit_bop(smm, &bop);
229                 if (r)
230                         break;
231
232                 brb_pop(&smm->uncommitted);
233         }
234
235         return r;
236 }
237
238 static int out(struct sm_metadata *smm)
239 {
240         int r = 0;
241
242         /*
243          * If we're not recursing then very bad things are happening.
244          */
245         if (!smm->recursion_count) {
246                 DMERR("lost track of recursion depth");
247                 return -ENOMEM;
248         }
249
250         if (smm->recursion_count == 1)
251                 r = apply_bops(smm);
252
253         smm->recursion_count--;
254
255         return r;
256 }
257
258 /*
259  * When using the out() function above, we often want to combine an error
260  * code for the operation run in the recursive context with that from
261  * out().
262  */
263 static int combine_errors(int r1, int r2)
264 {
265         return r1 ? r1 : r2;
266 }
267
268 static int recursing(struct sm_metadata *smm)
269 {
270         return smm->recursion_count;
271 }
272
273 static void sm_metadata_destroy(struct dm_space_map *sm)
274 {
275         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
276
277         kfree(smm);
278 }
279
280 static int sm_metadata_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
281 {
282         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
283
284         *count = smm->ll.nr_blocks;
285
286         return 0;
287 }
288
289 static int sm_metadata_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
290 {
291         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
292
293         *count = smm->old_ll.nr_blocks - smm->old_ll.nr_allocated -
294                  smm->allocated_this_transaction;
295
296         return 0;
297 }
298
299 static int sm_metadata_get_count(struct dm_space_map *sm, dm_block_t b,
300                                  uint32_t *result)
301 {
302         int r;
303         unsigned i;
304         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
305         unsigned adjustment = 0;
306
307         /*
308          * We may have some uncommitted adjustments to add.  This list
309          * should always be really short.
310          */
311         for (i = smm->uncommitted.begin;
312              i != smm->uncommitted.end;
313              i = brb_next(&smm->uncommitted, i)) {
314                 struct block_op *op = smm->uncommitted.bops + i;
315
316                 if (op->block != b)
317                         continue;
318
319                 switch (op->type) {
320                 case BOP_INC:
321                         adjustment++;
322                         break;
323
324                 case BOP_DEC:
325                         adjustment--;
326                         break;
327                 }
328         }
329
330         r = sm_ll_lookup(&smm->ll, b, result);
331         if (r)
332                 return r;
333
334         *result += adjustment;
335
336         return 0;
337 }
338
339 static int sm_metadata_count_is_more_than_one(struct dm_space_map *sm,
340                                               dm_block_t b, int *result)
341 {
342         int r, adjustment = 0;
343         unsigned i;
344         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
345         uint32_t rc;
346
347         /*
348          * We may have some uncommitted adjustments to add.  This list
349          * should always be really short.
350          */
351         for (i = smm->uncommitted.begin;
352              i != smm->uncommitted.end;
353              i = brb_next(&smm->uncommitted, i)) {
354
355                 struct block_op *op = smm->uncommitted.bops + i;
356
357                 if (op->block != b)
358                         continue;
359
360                 switch (op->type) {
361                 case BOP_INC:
362                         adjustment++;
363                         break;
364
365                 case BOP_DEC:
366                         adjustment--;
367                         break;
368                 }
369         }
370
371         if (adjustment > 1) {
372                 *result = 1;
373                 return 0;
374         }
375
376         r = sm_ll_lookup_bitmap(&smm->ll, b, &rc);
377         if (r)
378                 return r;
379
380         if (rc == 3)
381                 /*
382                  * We err on the side of caution, and always return true.
383                  */
384                 *result = 1;
385         else
386                 *result = rc + adjustment > 1;
387
388         return 0;
389 }
390
391 static int sm_metadata_set_count(struct dm_space_map *sm, dm_block_t b,
392                                  uint32_t count)
393 {
394         int r, r2;
395         enum allocation_event ev;
396         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
397
398         if (smm->recursion_count) {
399                 DMERR("cannot recurse set_count()");
400                 return -EINVAL;
401         }
402
403         in(smm);
404         r = sm_ll_insert(&smm->ll, b, count, &ev);
405         r2 = out(smm);
406
407         return combine_errors(r, r2);
408 }
409
410 static int sm_metadata_inc_block(struct dm_space_map *sm, dm_block_t b)
411 {
412         int r, r2 = 0;
413         enum allocation_event ev;
414         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
415
416         if (recursing(smm))
417                 r = add_bop(smm, BOP_INC, b);
418         else {
419                 in(smm);
420                 r = sm_ll_inc(&smm->ll, b, &ev);
421                 r2 = out(smm);
422         }
423
424         return combine_errors(r, r2);
425 }
426
427 static int sm_metadata_dec_block(struct dm_space_map *sm, dm_block_t b)
428 {
429         int r, r2 = 0;
430         enum allocation_event ev;
431         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
432
433         if (recursing(smm))
434                 r = add_bop(smm, BOP_DEC, b);
435         else {
436                 in(smm);
437                 r = sm_ll_dec(&smm->ll, b, &ev);
438                 r2 = out(smm);
439         }
440
441         return combine_errors(r, r2);
442 }
443
444 static int sm_metadata_new_block_(struct dm_space_map *sm, dm_block_t *b)
445 {
446         int r, r2 = 0;
447         enum allocation_event ev;
448         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
449
450         /*
451          * Any block we allocate has to be free in both the old and current ll.
452          */
453         r = sm_ll_find_common_free_block(&smm->old_ll, &smm->ll, smm->begin, smm->ll.nr_blocks, b);
454         if (r == -ENOSPC) {
455                 /*
456                  * There's no free block between smm->begin and the end of the metadata device.
457                  * We search before smm->begin in case something has been freed.
458                  */
459                 r = sm_ll_find_common_free_block(&smm->old_ll, &smm->ll, 0, smm->begin, b);
460         }
461
462         if (r)
463                 return r;
464
465         smm->begin = *b + 1;
466
467         if (recursing(smm))
468                 r = add_bop(smm, BOP_INC, *b);
469         else {
470                 in(smm);
471                 r = sm_ll_inc(&smm->ll, *b, &ev);
472                 r2 = out(smm);
473         }
474
475         if (!r)
476                 smm->allocated_this_transaction++;
477
478         return combine_errors(r, r2);
479 }
480
481 static int sm_metadata_new_block(struct dm_space_map *sm, dm_block_t *b)
482 {
483         dm_block_t count;
484         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
485
486         int r = sm_metadata_new_block_(sm, b);
487         if (r) {
488                 DMERR_LIMIT("unable to allocate new metadata block");
489                 return r;
490         }
491
492         r = sm_metadata_get_nr_free(sm, &count);
493         if (r) {
494                 DMERR_LIMIT("couldn't get free block count");
495                 return r;
496         }
497
498         check_threshold(&smm->threshold, count);
499
500         return r;
501 }
502
503 static int sm_metadata_commit(struct dm_space_map *sm)
504 {
505         int r;
506         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
507
508         r = sm_ll_commit(&smm->ll);
509         if (r)
510                 return r;
511
512         memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
513         smm->allocated_this_transaction = 0;
514
515         return 0;
516 }
517
518 static int sm_metadata_register_threshold_callback(struct dm_space_map *sm,
519                                                    dm_block_t threshold,
520                                                    dm_sm_threshold_fn fn,
521                                                    void *context)
522 {
523         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
524
525         set_threshold(&smm->threshold, threshold, fn, context);
526
527         return 0;
528 }
529
530 static int sm_metadata_root_size(struct dm_space_map *sm, size_t *result)
531 {
532         *result = sizeof(struct disk_sm_root);
533
534         return 0;
535 }
536
537 static int sm_metadata_copy_root(struct dm_space_map *sm, void *where_le, size_t max)
538 {
539         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
540         struct disk_sm_root root_le;
541
542         root_le.nr_blocks = cpu_to_le64(smm->ll.nr_blocks);
543         root_le.nr_allocated = cpu_to_le64(smm->ll.nr_allocated);
544         root_le.bitmap_root = cpu_to_le64(smm->ll.bitmap_root);
545         root_le.ref_count_root = cpu_to_le64(smm->ll.ref_count_root);
546
547         if (max < sizeof(root_le))
548                 return -ENOSPC;
549
550         memcpy(where_le, &root_le, sizeof(root_le));
551
552         return 0;
553 }
554
555 static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks);
556
557 static struct dm_space_map ops = {
558         .destroy = sm_metadata_destroy,
559         .extend = sm_metadata_extend,
560         .get_nr_blocks = sm_metadata_get_nr_blocks,
561         .get_nr_free = sm_metadata_get_nr_free,
562         .get_count = sm_metadata_get_count,
563         .count_is_more_than_one = sm_metadata_count_is_more_than_one,
564         .set_count = sm_metadata_set_count,
565         .inc_block = sm_metadata_inc_block,
566         .dec_block = sm_metadata_dec_block,
567         .new_block = sm_metadata_new_block,
568         .commit = sm_metadata_commit,
569         .root_size = sm_metadata_root_size,
570         .copy_root = sm_metadata_copy_root,
571         .register_threshold_callback = sm_metadata_register_threshold_callback
572 };
573
574 /*----------------------------------------------------------------*/
575
576 /*
577  * When a new space map is created that manages its own space.  We use
578  * this tiny bootstrap allocator.
579  */
580 static void sm_bootstrap_destroy(struct dm_space_map *sm)
581 {
582 }
583
584 static int sm_bootstrap_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
585 {
586         DMERR("bootstrap doesn't support extend");
587
588         return -EINVAL;
589 }
590
591 static int sm_bootstrap_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
592 {
593         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
594
595         *count = smm->ll.nr_blocks;
596
597         return 0;
598 }
599
600 static int sm_bootstrap_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
601 {
602         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
603
604         *count = smm->ll.nr_blocks - smm->begin;
605
606         return 0;
607 }
608
609 static int sm_bootstrap_get_count(struct dm_space_map *sm, dm_block_t b,
610                                   uint32_t *result)
611 {
612         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
613
614         *result = (b < smm->begin) ? 1 : 0;
615
616         return 0;
617 }
618
619 static int sm_bootstrap_count_is_more_than_one(struct dm_space_map *sm,
620                                                dm_block_t b, int *result)
621 {
622         *result = 0;
623
624         return 0;
625 }
626
627 static int sm_bootstrap_set_count(struct dm_space_map *sm, dm_block_t b,
628                                   uint32_t count)
629 {
630         DMERR("bootstrap doesn't support set_count");
631
632         return -EINVAL;
633 }
634
635 static int sm_bootstrap_new_block(struct dm_space_map *sm, dm_block_t *b)
636 {
637         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
638
639         /*
640          * We know the entire device is unused.
641          */
642         if (smm->begin == smm->ll.nr_blocks)
643                 return -ENOSPC;
644
645         *b = smm->begin++;
646
647         return 0;
648 }
649
650 static int sm_bootstrap_inc_block(struct dm_space_map *sm, dm_block_t b)
651 {
652         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
653
654         return add_bop(smm, BOP_INC, b);
655 }
656
657 static int sm_bootstrap_dec_block(struct dm_space_map *sm, dm_block_t b)
658 {
659         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
660
661         return add_bop(smm, BOP_DEC, b);
662 }
663
664 static int sm_bootstrap_commit(struct dm_space_map *sm)
665 {
666         return 0;
667 }
668
669 static int sm_bootstrap_root_size(struct dm_space_map *sm, size_t *result)
670 {
671         DMERR("bootstrap doesn't support root_size");
672
673         return -EINVAL;
674 }
675
676 static int sm_bootstrap_copy_root(struct dm_space_map *sm, void *where,
677                                   size_t max)
678 {
679         DMERR("bootstrap doesn't support copy_root");
680
681         return -EINVAL;
682 }
683
684 static struct dm_space_map bootstrap_ops = {
685         .destroy = sm_bootstrap_destroy,
686         .extend = sm_bootstrap_extend,
687         .get_nr_blocks = sm_bootstrap_get_nr_blocks,
688         .get_nr_free = sm_bootstrap_get_nr_free,
689         .get_count = sm_bootstrap_get_count,
690         .count_is_more_than_one = sm_bootstrap_count_is_more_than_one,
691         .set_count = sm_bootstrap_set_count,
692         .inc_block = sm_bootstrap_inc_block,
693         .dec_block = sm_bootstrap_dec_block,
694         .new_block = sm_bootstrap_new_block,
695         .commit = sm_bootstrap_commit,
696         .root_size = sm_bootstrap_root_size,
697         .copy_root = sm_bootstrap_copy_root,
698         .register_threshold_callback = NULL
699 };
700
701 /*----------------------------------------------------------------*/
702
703 static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
704 {
705         int r, i;
706         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
707         dm_block_t old_len = smm->ll.nr_blocks;
708
709         /*
710          * Flick into a mode where all blocks get allocated in the new area.
711          */
712         smm->begin = old_len;
713         memcpy(sm, &bootstrap_ops, sizeof(*sm));
714
715         /*
716          * Extend.
717          */
718         r = sm_ll_extend(&smm->ll, extra_blocks);
719         if (r)
720                 goto out;
721
722         /*
723          * We repeatedly increment then commit until the commit doesn't
724          * allocate any new blocks.
725          */
726         do {
727                 for (i = old_len; !r && i < smm->begin; i++)
728                         r = add_bop(smm, BOP_INC, i);
729
730                 if (r)
731                         goto out;
732
733                 old_len = smm->begin;
734
735                 r = apply_bops(smm);
736                 if (r) {
737                         DMERR("%s: apply_bops failed", __func__);
738                         goto out;
739                 }
740
741                 r = sm_ll_commit(&smm->ll);
742                 if (r)
743                         goto out;
744
745         } while (old_len != smm->begin);
746
747 out:
748         /*
749          * Switch back to normal behaviour.
750          */
751         memcpy(sm, &ops, sizeof(*sm));
752         return r;
753 }
754
755 /*----------------------------------------------------------------*/
756
757 struct dm_space_map *dm_sm_metadata_init(void)
758 {
759         struct sm_metadata *smm;
760
761         smm = kmalloc(sizeof(*smm), GFP_KERNEL);
762         if (!smm)
763                 return ERR_PTR(-ENOMEM);
764
765         memcpy(&smm->sm, &ops, sizeof(smm->sm));
766
767         return &smm->sm;
768 }
769
770 int dm_sm_metadata_create(struct dm_space_map *sm,
771                           struct dm_transaction_manager *tm,
772                           dm_block_t nr_blocks,
773                           dm_block_t superblock)
774 {
775         int r;
776         dm_block_t i;
777         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
778
779         smm->begin = superblock + 1;
780         smm->recursion_count = 0;
781         smm->allocated_this_transaction = 0;
782         brb_init(&smm->uncommitted);
783         threshold_init(&smm->threshold);
784
785         memcpy(&smm->sm, &bootstrap_ops, sizeof(smm->sm));
786
787         r = sm_ll_new_metadata(&smm->ll, tm);
788         if (!r) {
789                 if (nr_blocks > DM_SM_METADATA_MAX_BLOCKS)
790                         nr_blocks = DM_SM_METADATA_MAX_BLOCKS;
791                 r = sm_ll_extend(&smm->ll, nr_blocks);
792         }
793         memcpy(&smm->sm, &ops, sizeof(smm->sm));
794         if (r)
795                 return r;
796
797         /*
798          * Now we need to update the newly created data structures with the
799          * allocated blocks that they were built from.
800          */
801         for (i = superblock; !r && i < smm->begin; i++)
802                 r = add_bop(smm, BOP_INC, i);
803
804         if (r)
805                 return r;
806
807         r = apply_bops(smm);
808         if (r) {
809                 DMERR("%s: apply_bops failed", __func__);
810                 return r;
811         }
812
813         return sm_metadata_commit(sm);
814 }
815
816 int dm_sm_metadata_open(struct dm_space_map *sm,
817                         struct dm_transaction_manager *tm,
818                         void *root_le, size_t len)
819 {
820         int r;
821         struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
822
823         r = sm_ll_open_metadata(&smm->ll, tm, root_le, len);
824         if (r)
825                 return r;
826
827         smm->begin = 0;
828         smm->recursion_count = 0;
829         smm->allocated_this_transaction = 0;
830         brb_init(&smm->uncommitted);
831         threshold_init(&smm->threshold);
832
833         memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
834         return 0;
835 }