]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blame - drivers/md/dm-zoned-metadata.c
projects / mirror_ubuntu-hirsute-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'powerpc-5.2-2' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
[mirror_ubuntu-hirsute-kernel.git] / drivers / md / dm-zoned-metadata.c
CommitLineData
3b1a94c8
DLM		 1/*
2 * Copyright (C) 2017 Western Digital Corporation or its affiliates.
3 *
4 * This file is released under the GPL.
5 */
6
7#include "dm-zoned.h"
8
9#include <linux/module.h>
10#include <linux/crc32.h>
11
12#define DM_MSG_PREFIX "zoned metadata"
13
14/*
15 * Metadata version.
16 */
17#define DMZ_META_VER 1
18
19/*
20 * On-disk super block magic.
21 */
22#define DMZ_MAGIC ((((unsigned int)('D')) << 24) | \
23 (((unsigned int)('Z')) << 16) | \
24 (((unsigned int)('B')) << 8) | \
25 ((unsigned int)('D')))
26
27/*
28 * On disk super block.
29 * This uses only 512 B but uses on disk a full 4KB block. This block is
30 * followed on disk by the mapping table of chunks to zones and the bitmap
31 * blocks indicating zone block validity.
32 * The overall resulting metadata format is:
33 * (1) Super block (1 block)
34 * (2) Chunk mapping table (nr_map_blocks)
35 * (3) Bitmap blocks (nr_bitmap_blocks)
36 * All metadata blocks are stored in conventional zones, starting from the
37 * the first conventional zone found on disk.
38 */
39struct dmz_super {
40 /* Magic number */
41 __le32 magic; /* 4 */
42
43 /* Metadata version number */
44 __le32 version; /* 8 */
45
46 /* Generation number */
47 __le64 gen; /* 16 */
48
49 /* This block number */
50 __le64 sb_block; /* 24 */
51
52 /* The number of metadata blocks, including this super block */
53 __le32 nr_meta_blocks; /* 28 */
54
55 /* The number of sequential zones reserved for reclaim */
56 __le32 nr_reserved_seq; /* 32 */
57
58 /* The number of entries in the mapping table */
59 __le32 nr_chunks; /* 36 */
60
61 /* The number of blocks used for the chunk mapping table */
62 __le32 nr_map_blocks; /* 40 */
63
64 /* The number of blocks used for the block bitmaps */
65 __le32 nr_bitmap_blocks; /* 44 */
66
67 /* Checksum */
68 __le32 crc; /* 48 */
69
70 /* Padding to full 512B sector */
71 u8 reserved[464]; /* 512 */
72};
73
74/*
75 * Chunk mapping entry: entries are indexed by chunk number
76 * and give the zone ID (dzone_id) mapping the chunk on disk.
77 * This zone may be sequential or random. If it is a sequential
78 * zone, a second zone (bzone_id) used as a write buffer may
79 * also be specified. This second zone will always be a randomly
80 * writeable zone.
81 */
82struct dmz_map {
83 __le32 dzone_id;
84 __le32 bzone_id;
85};
86
87/*
88 * Chunk mapping table metadata: 512 8-bytes entries per 4KB block.
89 */
90#define DMZ_MAP_ENTRIES (DMZ_BLOCK_SIZE / sizeof(struct dmz_map))
91#define DMZ_MAP_ENTRIES_SHIFT (ilog2(DMZ_MAP_ENTRIES))
92#define DMZ_MAP_ENTRIES_MASK (DMZ_MAP_ENTRIES - 1)
93#define DMZ_MAP_UNMAPPED UINT_MAX
94
95/*
96 * Meta data block descriptor (for cached metadata blocks).
97 */
98struct dmz_mblock {
99 struct rb_node node;
100 struct list_head link;
101 sector_t no;
33c2865f 102 unsigned int ref;
3b1a94c8
DLM		 103 unsigned long state;
104 struct page *page;
105 void *data;
106};
107
108/*
109 * Metadata block state flags.
110 */
111enum {
112 DMZ_META_DIRTY,
113 DMZ_META_READING,
114 DMZ_META_WRITING,
115 DMZ_META_ERROR,
116};
117
118/*
119 * Super block information (one per metadata set).
120 */
121struct dmz_sb {
122 sector_t block;
123 struct dmz_mblock *mblk;
124 struct dmz_super *sb;
125};
126
127/*
128 * In-memory metadata.
129 */
130struct dmz_metadata {
131 struct dmz_dev *dev;
132
133 sector_t zone_bitmap_size;
134 unsigned int zone_nr_bitmap_blocks;
135
136 unsigned int nr_bitmap_blocks;
137 unsigned int nr_map_blocks;
138
139 unsigned int nr_useable_zones;
140 unsigned int nr_meta_blocks;
141 unsigned int nr_meta_zones;
142 unsigned int nr_data_zones;
143 unsigned int nr_rnd_zones;
144 unsigned int nr_reserved_seq;
145 unsigned int nr_chunks;
146
147 /* Zone information array */
148 struct dm_zone *zones;
149
150 struct dm_zone *sb_zone;
151 struct dmz_sb sb[2];
152 unsigned int mblk_primary;
153 u64 sb_gen;
154 unsigned int min_nr_mblks;
155 unsigned int max_nr_mblks;
156 atomic_t nr_mblks;
157 struct rw_semaphore mblk_sem;
158 struct mutex mblk_flush_lock;
159 spinlock_t mblk_lock;
160 struct rb_root mblk_rbtree;
161 struct list_head mblk_lru_list;
162 struct list_head mblk_dirty_list;
163 struct shrinker mblk_shrinker;
164
165 /* Zone allocation management */
166 struct mutex map_lock;
167 struct dmz_mblock **map_mblk;
168 unsigned int nr_rnd;
169 atomic_t unmap_nr_rnd;
170 struct list_head unmap_rnd_list;
171 struct list_head map_rnd_list;
172
173 unsigned int nr_seq;
174 atomic_t unmap_nr_seq;
175 struct list_head unmap_seq_list;
176 struct list_head map_seq_list;
177
178 atomic_t nr_reserved_seq_zones;
179 struct list_head reserved_seq_zones_list;
180
181 wait_queue_head_t free_wq;
182};
183
184/*
185 * Various accessors
186 */
187unsigned int dmz_id(struct dmz_metadata *zmd, struct dm_zone *zone)
188{
189 return ((unsigned int)(zone - zmd->zones));
190}
191
192sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone)
193{
3908c983 194 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_sectors_shift;
3b1a94c8
DLM		 195}
196
197sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone)
198{
3908c983 199 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_blocks_shift;
3b1a94c8
DLM		 200}
201
202unsigned int dmz_nr_chunks(struct dmz_metadata *zmd)
203{
204 return zmd->nr_chunks;
205}
206
207unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd)
208{
209 return zmd->nr_rnd;
210}
211
212unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd)
213{
214 return atomic_read(&zmd->unmap_nr_rnd);
215}
216
217/*
218 * Lock/unlock mapping table.
219 * The map lock also protects all the zone lists.
220 */
221void dmz_lock_map(struct dmz_metadata *zmd)
222{
223 mutex_lock(&zmd->map_lock);
224}
225
226void dmz_unlock_map(struct dmz_metadata *zmd)
227{
228 mutex_unlock(&zmd->map_lock);
229}
230
231/*
232 * Lock/unlock metadata access. This is a "read" lock on a semaphore
233 * that prevents metadata flush from running while metadata are being
234 * modified. The actual metadata write mutual exclusion is achieved with
235 * the map lock and zone styate management (active and reclaim state are
236 * mutually exclusive).
237 */
238void dmz_lock_metadata(struct dmz_metadata *zmd)
239{
240 down_read(&zmd->mblk_sem);
241}
242
243void dmz_unlock_metadata(struct dmz_metadata *zmd)
244{
245 up_read(&zmd->mblk_sem);
246}
247
248/*
249 * Lock/unlock flush: prevent concurrent executions
250 * of dmz_flush_metadata as well as metadata modification in reclaim
251 * while flush is being executed.
252 */
253void dmz_lock_flush(struct dmz_metadata *zmd)
254{
255 mutex_lock(&zmd->mblk_flush_lock);
256}
257
258void dmz_unlock_flush(struct dmz_metadata *zmd)
259{
260 mutex_unlock(&zmd->mblk_flush_lock);
261}
262
263/*
264 * Allocate a metadata block.
265 */
266static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd,
267 sector_t mblk_no)
268{
269 struct dmz_mblock *mblk = NULL;
270
271 /* See if we can reuse cached blocks */
272 if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) {
273 spin_lock(&zmd->mblk_lock);
274 mblk = list_first_entry_or_null(&zmd->mblk_lru_list,
275 struct dmz_mblock, link);
276 if (mblk) {
277 list_del_init(&mblk->link);
278 rb_erase(&mblk->node, &zmd->mblk_rbtree);
279 mblk->no = mblk_no;
280 }
281 spin_unlock(&zmd->mblk_lock);
282 if (mblk)
283 return mblk;
284 }
285
286 /* Allocate a new block */
287 mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO);
288 if (!mblk)
289 return NULL;
290
291 mblk->page = alloc_page(GFP_NOIO);
292 if (!mblk->page) {
293 kfree(mblk);
294 return NULL;
295 }
296
297 RB_CLEAR_NODE(&mblk->node);
298 INIT_LIST_HEAD(&mblk->link);
33c2865f 299 mblk->ref = 0;
3b1a94c8
DLM		 300 mblk->state = 0;
301 mblk->no = mblk_no;
302 mblk->data = page_address(mblk->page);
303
304 atomic_inc(&zmd->nr_mblks);
305
306 return mblk;
307}
308
309/*
310 * Free a metadata block.
311 */
312static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
313{
314 __free_pages(mblk->page, 0);
315 kfree(mblk);
316
317 atomic_dec(&zmd->nr_mblks);
318}
319
320/*
321 * Insert a metadata block in the rbtree.
322 */
323static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
324{
325 struct rb_root *root = &zmd->mblk_rbtree;
326 struct rb_node **new = &(root->rb_node), *parent = NULL;
327 struct dmz_mblock *b;
328
329 /* Figure out where to put the new node */
330 while (*new) {
331 b = container_of(*new, struct dmz_mblock, node);
332 parent = *new;
333 new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right);
334 }
335
336 /* Add new node and rebalance tree */
337 rb_link_node(&mblk->node, parent, new);
338 rb_insert_color(&mblk->node, root);
339}
340
341/*
3d4e7383
DLM		 342 * Lookup a metadata block in the rbtree. If the block is found, increment
343 * its reference count.
3b1a94c8 344 */
3d4e7383
DLM		 345static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd,
346 sector_t mblk_no)
3b1a94c8
DLM		 347{
348 struct rb_root *root = &zmd->mblk_rbtree;
349 struct rb_node *node = root->rb_node;
350 struct dmz_mblock *mblk;
351
352 while (node) {
353 mblk = container_of(node, struct dmz_mblock, node);
3d4e7383
DLM		 354 if (mblk->no == mblk_no) {
355 /*
356 * If this is the first reference to the block,
357 * remove it from the LRU list.
358 */
359 mblk->ref++;
360 if (mblk->ref == 1 &&
361 !test_bit(DMZ_META_DIRTY, &mblk->state))
362 list_del_init(&mblk->link);
3b1a94c8 363 return mblk;
3d4e7383 364 }
3b1a94c8
DLM		 365 node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right;
366 }
367
368 return NULL;
369}
370
371/*
372 * Metadata block BIO end callback.
373 */
374static void dmz_mblock_bio_end_io(struct bio *bio)
375{
376 struct dmz_mblock *mblk = bio->bi_private;
377 int flag;
378
379 if (bio->bi_status)
380 set_bit(DMZ_META_ERROR, &mblk->state);
381
382 if (bio_op(bio) == REQ_OP_WRITE)
383 flag = DMZ_META_WRITING;
384 else
385 flag = DMZ_META_READING;
386
387 clear_bit_unlock(flag, &mblk->state);
388 smp_mb__after_atomic();
389 wake_up_bit(&mblk->state, flag);
390
391 bio_put(bio);
392}
393
394/*
3d4e7383 395 * Read an uncached metadata block from disk and add it to the cache.
3b1a94c8 396 */
3d4e7383
DLM		 397static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd,
398 sector_t mblk_no)
3b1a94c8 399{
3d4e7383 400 struct dmz_mblock *mblk, *m;
3b1a94c8
DLM		 401 sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no;
402 struct bio *bio;
403
3d4e7383 404 /* Get a new block and a BIO to read it */
3b1a94c8
DLM		 405 mblk = dmz_alloc_mblock(zmd, mblk_no);
406 if (!mblk)
407 return NULL;
408
3b1a94c8
DLM		 409 bio = bio_alloc(GFP_NOIO, 1);
410 if (!bio) {
411 dmz_free_mblock(zmd, mblk);
412 return NULL;
413 }
414
3d4e7383
DLM		 415 spin_lock(&zmd->mblk_lock);
416
417 /*
418 * Make sure that another context did not start reading
419 * the block already.
420 */
421 m = dmz_get_mblock_fast(zmd, mblk_no);
422 if (m) {
423 spin_unlock(&zmd->mblk_lock);
424 dmz_free_mblock(zmd, mblk);
425 bio_put(bio);
426 return m;
427 }
428
429 mblk->ref++;
430 set_bit(DMZ_META_READING, &mblk->state);
431 dmz_insert_mblock(zmd, mblk);
432
433 spin_unlock(&zmd->mblk_lock);
434
435 /* Submit read BIO */
3b1a94c8 436 bio->bi_iter.bi_sector = dmz_blk2sect(block);
74d46992 437 bio_set_dev(bio, zmd->dev->bdev);
3b1a94c8
DLM		 438 bio->bi_private = mblk;
439 bio->bi_end_io = dmz_mblock_bio_end_io;
440 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO);
441 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
442 submit_bio(bio);
443
444 return mblk;
445}
446
447/*
448 * Free metadata blocks.
449 */
450static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd,
451 unsigned long limit)
452{
453 struct dmz_mblock *mblk;
454 unsigned long count = 0;
455
456 if (!zmd->max_nr_mblks)
457 return 0;
458
459 while (!list_empty(&zmd->mblk_lru_list) &&
460 atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks &&
461 count < limit) {
462 mblk = list_first_entry(&zmd->mblk_lru_list,
463 struct dmz_mblock, link);
464 list_del_init(&mblk->link);
465 rb_erase(&mblk->node, &zmd->mblk_rbtree);
466 dmz_free_mblock(zmd, mblk);
467 count++;
468 }
469
470 return count;
471}
472
473/*
474 * For mblock shrinker: get the number of unused metadata blocks in the cache.
475 */
476static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink,
477 struct shrink_control *sc)
478{
479 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
480
481 return atomic_read(&zmd->nr_mblks);
482}
483
484/*
485 * For mblock shrinker: scan unused metadata blocks and shrink the cache.
486 */
487static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink,
488 struct shrink_control *sc)
489{
490 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
491 unsigned long count;
492
493 spin_lock(&zmd->mblk_lock);
494 count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan);
495 spin_unlock(&zmd->mblk_lock);
496
497 return count ? count : SHRINK_STOP;
498}
499
500/*
501 * Release a metadata block.
502 */
503static void dmz_release_mblock(struct dmz_metadata *zmd,
504 struct dmz_mblock *mblk)
505{
506
507 if (!mblk)
508 return;
509
510 spin_lock(&zmd->mblk_lock);
511
33c2865f
DLM		 512 mblk->ref--;
513 if (mblk->ref == 0) {
3b1a94c8
DLM		 514 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
515 rb_erase(&mblk->node, &zmd->mblk_rbtree);
516 dmz_free_mblock(zmd, mblk);
517 } else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) {
518 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
519 dmz_shrink_mblock_cache(zmd, 1);
520 }
521 }
522
523 spin_unlock(&zmd->mblk_lock);
524}
525
526/*
527 * Get a metadata block from the rbtree. If the block
528 * is not present, read it from disk.
529 */
530static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
531 sector_t mblk_no)
532{
533 struct dmz_mblock *mblk;
534
535 /* Check rbtree */
536 spin_lock(&zmd->mblk_lock);
3d4e7383 537 mblk = dmz_get_mblock_fast(zmd, mblk_no);
3b1a94c8
DLM		 538 spin_unlock(&zmd->mblk_lock);
539
540 if (!mblk) {
541 /* Cache miss: read the block from disk */
3d4e7383 542 mblk = dmz_get_mblock_slow(zmd, mblk_no);
3b1a94c8
DLM		 543 if (!mblk)
544 return ERR_PTR(-ENOMEM);
545 }
546
547 /* Wait for on-going read I/O and check for error */
548 wait_on_bit_io(&mblk->state, DMZ_META_READING,
549 TASK_UNINTERRUPTIBLE);
550 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
551 dmz_release_mblock(zmd, mblk);
552 return ERR_PTR(-EIO);
553 }
554
555 return mblk;
556}
557
558/*
559 * Mark a metadata block dirty.
560 */
561static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
562{
563 spin_lock(&zmd->mblk_lock);
564 if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state))
565 list_add_tail(&mblk->link, &zmd->mblk_dirty_list);
566 spin_unlock(&zmd->mblk_lock);
567}
568
569/*
570 * Issue a metadata block write BIO.
571 */
572static void dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk,
573 unsigned int set)
574{
575 sector_t block = zmd->sb[set].block + mblk->no;
576 struct bio *bio;
577
578 bio = bio_alloc(GFP_NOIO, 1);
579 if (!bio) {
580 set_bit(DMZ_META_ERROR, &mblk->state);
581 return;
582 }
583
584 set_bit(DMZ_META_WRITING, &mblk->state);
585
586 bio->bi_iter.bi_sector = dmz_blk2sect(block);
74d46992 587 bio_set_dev(bio, zmd->dev->bdev);
3b1a94c8
DLM		 588 bio->bi_private = mblk;
589 bio->bi_end_io = dmz_mblock_bio_end_io;
590 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO);
591 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
592 submit_bio(bio);
593}
594
595/*
596 * Read/write a metadata block.
597 */
598static int dmz_rdwr_block(struct dmz_metadata *zmd, int op, sector_t block,
599 struct page *page)
600{
601 struct bio *bio;
602 int ret;
603
604 bio = bio_alloc(GFP_NOIO, 1);
605 if (!bio)
606 return -ENOMEM;
607
608 bio->bi_iter.bi_sector = dmz_blk2sect(block);
74d46992 609 bio_set_dev(bio, zmd->dev->bdev);
3b1a94c8
DLM		 610 bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO);
611 bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0);
612 ret = submit_bio_wait(bio);
613 bio_put(bio);
614
615 return ret;
616}
617
618/*
619 * Write super block of the specified metadata set.
620 */
621static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set)
622{
623 sector_t block = zmd->sb[set].block;
624 struct dmz_mblock *mblk = zmd->sb[set].mblk;
625 struct dmz_super *sb = zmd->sb[set].sb;
626 u64 sb_gen = zmd->sb_gen + 1;
627 int ret;
628
629 sb->magic = cpu_to_le32(DMZ_MAGIC);
630 sb->version = cpu_to_le32(DMZ_META_VER);
631
632 sb->gen = cpu_to_le64(sb_gen);
633
634 sb->sb_block = cpu_to_le64(block);
635 sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks);
636 sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq);
637 sb->nr_chunks = cpu_to_le32(zmd->nr_chunks);
638
639 sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks);
640 sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks);
641
642 sb->crc = 0;
643 sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE));
644
645 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE, block, mblk->page);
646 if (ret == 0)
4218a955 647 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
3b1a94c8
DLM		 648
649 return ret;
650}
651
652/*
653 * Write dirty metadata blocks to the specified set.
654 */
655static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
656 struct list_head *write_list,
657 unsigned int set)
658{
659 struct dmz_mblock *mblk;
660 struct blk_plug plug;
661 int ret = 0;
662
663 /* Issue writes */
664 blk_start_plug(&plug);
665 list_for_each_entry(mblk, write_list, link)
666 dmz_write_mblock(zmd, mblk, set);
667 blk_finish_plug(&plug);
668
669 /* Wait for completion */
670 list_for_each_entry(mblk, write_list, link) {
671 wait_on_bit_io(&mblk->state, DMZ_META_WRITING,
672 TASK_UNINTERRUPTIBLE);
673 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
674 clear_bit(DMZ_META_ERROR, &mblk->state);
675 ret = -EIO;
676 }
677 }
678
679 /* Flush drive cache (this will also sync data) */
680 if (ret == 0)
4218a955 681 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
3b1a94c8
DLM		 682
683 return ret;
684}
685
686/*
687 * Log dirty metadata blocks.
688 */
689static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd,
690 struct list_head *write_list)
691{
692 unsigned int log_set = zmd->mblk_primary ^ 0x1;
693 int ret;
694
695 /* Write dirty blocks to the log */
696 ret = dmz_write_dirty_mblocks(zmd, write_list, log_set);
697 if (ret)
698 return ret;
699
700 /*
701 * No error so far: now validate the log by updating the
702 * log index super block generation.
703 */
704 ret = dmz_write_sb(zmd, log_set);
705 if (ret)
706 return ret;
707
708 return 0;
709}
710
711/*
712 * Flush dirty metadata blocks.
713 */
714int dmz_flush_metadata(struct dmz_metadata *zmd)
715{
716 struct dmz_mblock *mblk;
717 struct list_head write_list;
718 int ret;
719
720 if (WARN_ON(!zmd))
721 return 0;
722
723 INIT_LIST_HEAD(&write_list);
724
725 /*
726 * Make sure that metadata blocks are stable before logging: take
727 * the write lock on the metadata semaphore to prevent target BIOs
728 * from modifying metadata.
729 */
730 down_write(&zmd->mblk_sem);
731
732 /*
733 * This is called from the target flush work and reclaim work.
734 * Concurrent execution is not allowed.
735 */
736 dmz_lock_flush(zmd);
737
738 /* Get dirty blocks */
739 spin_lock(&zmd->mblk_lock);
740 list_splice_init(&zmd->mblk_dirty_list, &write_list);
741 spin_unlock(&zmd->mblk_lock);
742
743 /* If there are no dirty metadata blocks, just flush the device cache */
744 if (list_empty(&write_list)) {
4218a955 745 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
3b1a94c8
DLM		 746 goto out;
747 }
748
749 /*
750 * The primary metadata set is still clean. Keep it this way until
751 * all updates are successful in the secondary set. That is, use
752 * the secondary set as a log.
753 */
754 ret = dmz_log_dirty_mblocks(zmd, &write_list);
755 if (ret)
756 goto out;
757
758 /*
759 * The log is on disk. It is now safe to update in place
760 * in the primary metadata set.
761 */
762 ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
763 if (ret)
764 goto out;
765
766 ret = dmz_write_sb(zmd, zmd->mblk_primary);
767 if (ret)
768 goto out;
769
770 while (!list_empty(&write_list)) {
771 mblk = list_first_entry(&write_list, struct dmz_mblock, link);
772 list_del_init(&mblk->link);
773
774 spin_lock(&zmd->mblk_lock);
775 clear_bit(DMZ_META_DIRTY, &mblk->state);
33c2865f 776 if (mblk->ref == 0)
3b1a94c8
DLM		 777 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
778 spin_unlock(&zmd->mblk_lock);
779 }
780
781 zmd->sb_gen++;
782out:
783 if (ret && !list_empty(&write_list)) {
784 spin_lock(&zmd->mblk_lock);
785 list_splice(&write_list, &zmd->mblk_dirty_list);
786 spin_unlock(&zmd->mblk_lock);
787 }
788
789 dmz_unlock_flush(zmd);
790 up_write(&zmd->mblk_sem);
791
792 return ret;
793}
794
795/*
796 * Check super block.
797 */
798static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_super *sb)
799{
800 unsigned int nr_meta_zones, nr_data_zones;
801 struct dmz_dev *dev = zmd->dev;
802 u32 crc, stored_crc;
803 u64 gen;
804
805 gen = le64_to_cpu(sb->gen);
806 stored_crc = le32_to_cpu(sb->crc);
807 sb->crc = 0;
808 crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
809 if (crc != stored_crc) {
810 dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
811 crc, stored_crc);
812 return -ENXIO;
813 }
814
815 if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
816 dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
817 DMZ_MAGIC, le32_to_cpu(sb->magic));
818 return -ENXIO;
819 }
820
821 if (le32_to_cpu(sb->version) != DMZ_META_VER) {
822 dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
823 DMZ_META_VER, le32_to_cpu(sb->version));
824 return -ENXIO;
825 }
826
827 nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + dev->zone_nr_blocks - 1)
828 >> dev->zone_nr_blocks_shift;
829 if (!nr_meta_zones ||
830 nr_meta_zones >= zmd->nr_rnd_zones) {
831 dmz_dev_err(dev, "Invalid number of metadata blocks");
832 return -ENXIO;
833 }
834
835 if (!le32_to_cpu(sb->nr_reserved_seq) ||
836 le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
837 dmz_dev_err(dev, "Invalid number of reserved sequential zones");
838 return -ENXIO;
839 }
840
841 nr_data_zones = zmd->nr_useable_zones -
842 (nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
843 if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
844 dmz_dev_err(dev, "Invalid number of chunks %u / %u",
845 le32_to_cpu(sb->nr_chunks), nr_data_zones);
846 return -ENXIO;
847 }
848
849 /* OK */
850 zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
851 zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
852 zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
853 zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
854 zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
855 zmd->nr_meta_zones = nr_meta_zones;
856 zmd->nr_data_zones = nr_data_zones;
857
858 return 0;
859}
860
861/*
862 * Read the first or second super block from disk.
863 */
864static int dmz_read_sb(struct dmz_metadata *zmd, unsigned int set)
865{
866 return dmz_rdwr_block(zmd, REQ_OP_READ, zmd->sb[set].block,
867 zmd->sb[set].mblk->page);
868}
869
870/*
871 * Determine the position of the secondary super blocks on disk.
872 * This is used only if a corruption of the primary super block
873 * is detected.
874 */
875static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
876{
877 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
878 struct dmz_mblock *mblk;
879 int i;
880
881 /* Allocate a block */
882 mblk = dmz_alloc_mblock(zmd, 0);
883 if (!mblk)
884 return -ENOMEM;
885
886 zmd->sb[1].mblk = mblk;
887 zmd->sb[1].sb = mblk->data;
888
889 /* Bad first super block: search for the second one */
890 zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
891 for (i = 0; i < zmd->nr_rnd_zones - 1; i++) {
892 if (dmz_read_sb(zmd, 1) != 0)
893 break;
894 if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
895 return 0;
896 zmd->sb[1].block += zone_nr_blocks;
897 }
898
899 dmz_free_mblock(zmd, mblk);
900 zmd->sb[1].mblk = NULL;
901
902 return -EIO;
903}
904
905/*
906 * Read the first or second super block from disk.
907 */
908static int dmz_get_sb(struct dmz_metadata *zmd, unsigned int set)
909{
910 struct dmz_mblock *mblk;
911 int ret;
912
913 /* Allocate a block */
914 mblk = dmz_alloc_mblock(zmd, 0);
915 if (!mblk)
916 return -ENOMEM;
917
918 zmd->sb[set].mblk = mblk;
919 zmd->sb[set].sb = mblk->data;
920
921 /* Read super block */
922 ret = dmz_read_sb(zmd, set);
923 if (ret) {
924 dmz_free_mblock(zmd, mblk);
925 zmd->sb[set].mblk = NULL;
926 return ret;
927 }
928
929 return 0;
930}
931
932/*
933 * Recover a metadata set.
934 */
935static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
936{
937 unsigned int src_set = dst_set ^ 0x1;
938 struct page *page;
939 int i, ret;
940
941 dmz_dev_warn(zmd->dev, "Metadata set %u invalid: recovering", dst_set);
942
943 if (dst_set == 0)
944 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
945 else {
946 zmd->sb[1].block = zmd->sb[0].block +
947 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
948 }
949
4218a955 950 page = alloc_page(GFP_NOIO);
3b1a94c8
DLM		 951 if (!page)
952 return -ENOMEM;
953
954 /* Copy metadata blocks */
955 for (i = 1; i < zmd->nr_meta_blocks; i++) {
956 ret = dmz_rdwr_block(zmd, REQ_OP_READ,
957 zmd->sb[src_set].block + i, page);
958 if (ret)
959 goto out;
960 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE,
961 zmd->sb[dst_set].block + i, page);
962 if (ret)
963 goto out;
964 }
965
966 /* Finalize with the super block */
967 if (!zmd->sb[dst_set].mblk) {
968 zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
969 if (!zmd->sb[dst_set].mblk) {
970 ret = -ENOMEM;
971 goto out;
972 }
973 zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
974 }
975
976 ret = dmz_write_sb(zmd, dst_set);
977out:
978 __free_pages(page, 0);
979
980 return ret;
981}
982
983/*
984 * Get super block from disk.
985 */
986static int dmz_load_sb(struct dmz_metadata *zmd)
987{
988 bool sb_good[2] = {false, false};
989 u64 sb_gen[2] = {0, 0};
990 int ret;
991
992 /* Read and check the primary super block */
993 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
994 ret = dmz_get_sb(zmd, 0);
995 if (ret) {
996 dmz_dev_err(zmd->dev, "Read primary super block failed");
997 return ret;
998 }
999
1000 ret = dmz_check_sb(zmd, zmd->sb[0].sb);
1001
1002 /* Read and check secondary super block */
1003 if (ret == 0) {
1004 sb_good[0] = true;
1005 zmd->sb[1].block = zmd->sb[0].block +
1006 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
1007 ret = dmz_get_sb(zmd, 1);
1008 } else
1009 ret = dmz_lookup_secondary_sb(zmd);
1010
1011 if (ret) {
1012 dmz_dev_err(zmd->dev, "Read secondary super block failed");
1013 return ret;
1014 }
1015
1016 ret = dmz_check_sb(zmd, zmd->sb[1].sb);
1017 if (ret == 0)
1018 sb_good[1] = true;
1019
1020 /* Use highest generation sb first */
1021 if (!sb_good[0] && !sb_good[1]) {
1022 dmz_dev_err(zmd->dev, "No valid super block found");
1023 return -EIO;
1024 }
1025
1026 if (sb_good[0])
1027 sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1028 else
1029 ret = dmz_recover_mblocks(zmd, 0);
1030
1031 if (sb_good[1])
1032 sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1033 else
1034 ret = dmz_recover_mblocks(zmd, 1);
1035
1036 if (ret) {
1037 dmz_dev_err(zmd->dev, "Recovery failed");
1038 return -EIO;
1039 }
1040
1041 if (sb_gen[0] >= sb_gen[1]) {
1042 zmd->sb_gen = sb_gen[0];
1043 zmd->mblk_primary = 0;
1044 } else {
1045 zmd->sb_gen = sb_gen[1];
1046 zmd->mblk_primary = 1;
1047 }
1048
1049 dmz_dev_debug(zmd->dev, "Using super block %u (gen %llu)",
1050 zmd->mblk_primary, zmd->sb_gen);
1051
1052 return 0;
1053}
1054
1055/*
1056 * Initialize a zone descriptor.
1057 */
1058static int dmz_init_zone(struct dmz_metadata *zmd, struct dm_zone *zone,
1059 struct blk_zone *blkz)
1060{
1061 struct dmz_dev *dev = zmd->dev;
1062
1063 /* Ignore the eventual last runt (smaller) zone */
1064 if (blkz->len != dev->zone_nr_sectors) {
1065 if (blkz->start + blkz->len == dev->capacity)
1066 return 0;
1067 return -ENXIO;
1068 }
1069
1070 INIT_LIST_HEAD(&zone->link);
1071 atomic_set(&zone->refcount, 0);
1072 zone->chunk = DMZ_MAP_UNMAPPED;
1073
1074 if (blkz->type == BLK_ZONE_TYPE_CONVENTIONAL) {
1075 set_bit(DMZ_RND, &zone->flags);
1076 zmd->nr_rnd_zones++;
1077 } else if (blkz->type == BLK_ZONE_TYPE_SEQWRITE_REQ ||
1078 blkz->type == BLK_ZONE_TYPE_SEQWRITE_PREF) {
1079 set_bit(DMZ_SEQ, &zone->flags);
1080 } else
1081 return -ENXIO;
1082
1083 if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1084 set_bit(DMZ_OFFLINE, &zone->flags);
1085 else if (blkz->cond == BLK_ZONE_COND_READONLY)
1086 set_bit(DMZ_READ_ONLY, &zone->flags);
1087
1088 if (dmz_is_rnd(zone))
1089 zone->wp_block = 0;
1090 else
1091 zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1092
1093 if (!dmz_is_offline(zone) && !dmz_is_readonly(zone)) {
1094 zmd->nr_useable_zones++;
1095 if (dmz_is_rnd(zone)) {
1096 zmd->nr_rnd_zones++;
1097 if (!zmd->sb_zone) {
1098 /* Super block zone */
1099 zmd->sb_zone = zone;
1100 }
1101 }
1102 }
1103
1104 return 0;
1105}
1106
1107/*
1108 * Free zones descriptors.
1109 */
1110static void dmz_drop_zones(struct dmz_metadata *zmd)
1111{
1112 kfree(zmd->zones);
1113 zmd->zones = NULL;
1114}
1115
1116/*
1117 * The size of a zone report in number of zones.
1118 * This results in 4096*64B=256KB report zones commands.
1119 */
1120#define DMZ_REPORT_NR_ZONES 4096
1121
1122/*
1123 * Allocate and initialize zone descriptors using the zone
1124 * information from disk.
1125 */
1126static int dmz_init_zones(struct dmz_metadata *zmd)
1127{
1128 struct dmz_dev *dev = zmd->dev;
1129 struct dm_zone *zone;
1130 struct blk_zone *blkz;
1131 unsigned int nr_blkz;
1132 sector_t sector = 0;
1133 int i, ret = 0;
1134
1135 /* Init */
1136 zmd->zone_bitmap_size = dev->zone_nr_blocks >> 3;
1137 zmd->zone_nr_bitmap_blocks = zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT;
1138
1139 /* Allocate zone array */
1140 zmd->zones = kcalloc(dev->nr_zones, sizeof(struct dm_zone), GFP_KERNEL);
1141 if (!zmd->zones)
1142 return -ENOMEM;
1143
1144 dmz_dev_info(dev, "Using %zu B for zone information",
1145 sizeof(struct dm_zone) * dev->nr_zones);
1146
1147 /* Get zone information */
1148 nr_blkz = DMZ_REPORT_NR_ZONES;
1149 blkz = kcalloc(nr_blkz, sizeof(struct blk_zone), GFP_KERNEL);
1150 if (!blkz) {
1151 ret = -ENOMEM;
1152 goto out;
1153 }
1154
1155 /*
1156 * Get zone information and initialize zone descriptors.
1157 * At the same time, determine where the super block
1158 * should be: first block of the first randomly writable
1159 * zone.
1160 */
1161 zone = zmd->zones;
1162 while (sector < dev->capacity) {
1163 /* Get zone information */
1164 nr_blkz = DMZ_REPORT_NR_ZONES;
1165 ret = blkdev_report_zones(dev->bdev, sector, blkz,
1166 &nr_blkz, GFP_KERNEL);
1167 if (ret) {
1168 dmz_dev_err(dev, "Report zones failed %d", ret);
1169 goto out;
1170 }
1171
7aedf75f
DLM		 1172 if (!nr_blkz)
1173 break;
1174
3b1a94c8
DLM		 1175 /* Process report */
1176 for (i = 0; i < nr_blkz; i++) {
1177 ret = dmz_init_zone(zmd, zone, &blkz[i]);
1178 if (ret)
1179 goto out;
1180 sector += dev->zone_nr_sectors;
1181 zone++;
1182 }
1183 }
1184
1185 /* The entire zone configuration of the disk should now be known */
1186 if (sector < dev->capacity) {
1187 dmz_dev_err(dev, "Failed to get correct zone information");
1188 ret = -ENXIO;
1189 }
1190out:
1191 kfree(blkz);
1192 if (ret)
1193 dmz_drop_zones(zmd);
1194
1195 return ret;
1196}
1197
1198/*
1199 * Update a zone information.
1200 */
1201static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1202{
1203 unsigned int nr_blkz = 1;
1204 struct blk_zone blkz;
1205 int ret;
1206
1207 /* Get zone information from disk */
1208 ret = blkdev_report_zones(zmd->dev->bdev, dmz_start_sect(zmd, zone),
4218a955 1209 &blkz, &nr_blkz, GFP_NOIO);
7aedf75f
DLM		 1210 if (!nr_blkz)
1211 ret = -EIO;
3b1a94c8
DLM		 1212 if (ret) {
1213 dmz_dev_err(zmd->dev, "Get zone %u report failed",
1214 dmz_id(zmd, zone));
1215 return ret;
1216 }
1217
1218 clear_bit(DMZ_OFFLINE, &zone->flags);
1219 clear_bit(DMZ_READ_ONLY, &zone->flags);
1220 if (blkz.cond == BLK_ZONE_COND_OFFLINE)
1221 set_bit(DMZ_OFFLINE, &zone->flags);
1222 else if (blkz.cond == BLK_ZONE_COND_READONLY)
1223 set_bit(DMZ_READ_ONLY, &zone->flags);
1224
1225 if (dmz_is_seq(zone))
1226 zone->wp_block = dmz_sect2blk(blkz.wp - blkz.start);
1227 else
1228 zone->wp_block = 0;
1229
1230 return 0;
1231}
1232
1233/*
1234 * Check a zone write pointer position when the zone is marked
1235 * with the sequential write error flag.
1236 */
1237static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1238 struct dm_zone *zone)
1239{
1240 unsigned int wp = 0;
1241 int ret;
1242
1243 wp = zone->wp_block;
1244 ret = dmz_update_zone(zmd, zone);
1245 if (ret)
1246 return ret;
1247
1248 dmz_dev_warn(zmd->dev, "Processing zone %u write error (zone wp %u/%u)",
1249 dmz_id(zmd, zone), zone->wp_block, wp);
1250
1251 if (zone->wp_block < wp) {
1252 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1253 wp - zone->wp_block);
1254 }
1255
1256 return 0;
1257}
1258
1259static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
1260{
1261 return &zmd->zones[zone_id];
1262}
1263
1264/*
1265 * Reset a zone write pointer.
1266 */
1267static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1268{
1269 int ret;
1270
1271 /*
1272 * Ignore offline zones, read only zones,
1273 * and conventional zones.
1274 */
1275 if (dmz_is_offline(zone) ||
1276 dmz_is_readonly(zone) ||
1277 dmz_is_rnd(zone))
1278 return 0;
1279
1280 if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1281 struct dmz_dev *dev = zmd->dev;
1282
1283 ret = blkdev_reset_zones(dev->bdev,
1284 dmz_start_sect(zmd, zone),
4218a955 1285 dev->zone_nr_sectors, GFP_NOIO);
3b1a94c8
DLM		 1286 if (ret) {
1287 dmz_dev_err(dev, "Reset zone %u failed %d",
1288 dmz_id(zmd, zone), ret);
1289 return ret;
1290 }
1291 }
1292
1293 /* Clear write error bit and rewind write pointer position */
1294 clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1295 zone->wp_block = 0;
1296
1297 return 0;
1298}
1299
1300static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1301
1302/*
1303 * Initialize chunk mapping.
1304 */
1305static int dmz_load_mapping(struct dmz_metadata *zmd)
1306{
1307 struct dmz_dev *dev = zmd->dev;
1308 struct dm_zone *dzone, *bzone;
1309 struct dmz_mblock *dmap_mblk = NULL;
1310 struct dmz_map *dmap;
1311 unsigned int i = 0, e = 0, chunk = 0;
1312 unsigned int dzone_id;
1313 unsigned int bzone_id;
1314
1315 /* Metadata block array for the chunk mapping table */
1316 zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1317 sizeof(struct dmz_mblk *), GFP_KERNEL);
1318 if (!zmd->map_mblk)
1319 return -ENOMEM;
1320
1321 /* Get chunk mapping table blocks and initialize zone mapping */
1322 while (chunk < zmd->nr_chunks) {
1323 if (!dmap_mblk) {
1324 /* Get mapping block */
1325 dmap_mblk = dmz_get_mblock(zmd, i + 1);
1326 if (IS_ERR(dmap_mblk))
1327 return PTR_ERR(dmap_mblk);
1328 zmd->map_mblk[i] = dmap_mblk;
1329 dmap = (struct dmz_map *) dmap_mblk->data;
1330 i++;
1331 e = 0;
1332 }
1333
1334 /* Check data zone */
1335 dzone_id = le32_to_cpu(dmap[e].dzone_id);
1336 if (dzone_id == DMZ_MAP_UNMAPPED)
1337 goto next;
1338
1339 if (dzone_id >= dev->nr_zones) {
1340 dmz_dev_err(dev, "Chunk %u mapping: invalid data zone ID %u",
1341 chunk, dzone_id);
1342 return -EIO;
1343 }
1344
1345 dzone = dmz_get(zmd, dzone_id);
1346 set_bit(DMZ_DATA, &dzone->flags);
1347 dzone->chunk = chunk;
1348 dmz_get_zone_weight(zmd, dzone);
1349
1350 if (dmz_is_rnd(dzone))
1351 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1352 else
1353 list_add_tail(&dzone->link, &zmd->map_seq_list);
1354
1355 /* Check buffer zone */
1356 bzone_id = le32_to_cpu(dmap[e].bzone_id);
1357 if (bzone_id == DMZ_MAP_UNMAPPED)
1358 goto next;
1359
1360 if (bzone_id >= dev->nr_zones) {
1361 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone ID %u",
1362 chunk, bzone_id);
1363 return -EIO;
1364 }
1365
1366 bzone = dmz_get(zmd, bzone_id);
1367 if (!dmz_is_rnd(bzone)) {
1368 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone %u",
1369 chunk, bzone_id);
1370 return -EIO;
1371 }
1372
1373 set_bit(DMZ_DATA, &bzone->flags);
1374 set_bit(DMZ_BUF, &bzone->flags);
1375 bzone->chunk = chunk;
1376 bzone->bzone = dzone;
1377 dzone->bzone = bzone;
1378 dmz_get_zone_weight(zmd, bzone);
1379 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1380next:
1381 chunk++;
1382 e++;
1383 if (e >= DMZ_MAP_ENTRIES)
1384 dmap_mblk = NULL;
1385 }
1386
1387 /*
1388 * At this point, only meta zones and mapped data zones were
1389 * fully initialized. All remaining zones are unmapped data
1390 * zones. Finish initializing those here.
1391 */
1392 for (i = 0; i < dev->nr_zones; i++) {
1393 dzone = dmz_get(zmd, i);
1394 if (dmz_is_meta(dzone))
1395 continue;
1396
1397 if (dmz_is_rnd(dzone))
1398 zmd->nr_rnd++;
1399 else
1400 zmd->nr_seq++;
1401
1402 if (dmz_is_data(dzone)) {
1403 /* Already initialized */
1404 continue;
1405 }
1406
1407 /* Unmapped data zone */
1408 set_bit(DMZ_DATA, &dzone->flags);
1409 dzone->chunk = DMZ_MAP_UNMAPPED;
1410 if (dmz_is_rnd(dzone)) {
1411 list_add_tail(&dzone->link, &zmd->unmap_rnd_list);
1412 atomic_inc(&zmd->unmap_nr_rnd);
1413 } else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1414 list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1415 atomic_inc(&zmd->nr_reserved_seq_zones);
1416 zmd->nr_seq--;
1417 } else {
1418 list_add_tail(&dzone->link, &zmd->unmap_seq_list);
1419 atomic_inc(&zmd->unmap_nr_seq);
1420 }
1421 }
1422
1423 return 0;
1424}
1425
1426/*
1427 * Set a data chunk mapping.
1428 */
1429static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1430 unsigned int dzone_id, unsigned int bzone_id)
1431{
1432 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1433 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1434 int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1435
1436 dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1437 dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1438 dmz_dirty_mblock(zmd, dmap_mblk);
1439}
1440
1441/*
1442 * The list of mapped zones is maintained in LRU order.
1443 * This rotates a zone at the end of its map list.
1444 */
1445static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1446{
1447 if (list_empty(&zone->link))
1448 return;
1449
1450 list_del_init(&zone->link);
1451 if (dmz_is_seq(zone)) {
1452 /* LRU rotate sequential zone */
1453 list_add_tail(&zone->link, &zmd->map_seq_list);
1454 } else {
1455 /* LRU rotate random zone */
1456 list_add_tail(&zone->link, &zmd->map_rnd_list);
1457 }
1458}
1459
1460/*
1461 * The list of mapped random zones is maintained
1462 * in LRU order. This rotates a zone at the end of the list.
1463 */
1464static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1465{
1466 __dmz_lru_zone(zmd, zone);
1467 if (zone->bzone)
1468 __dmz_lru_zone(zmd, zone->bzone);
1469}
1470
1471/*
1472 * Wait for any zone to be freed.
1473 */
1474static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1475{
1476 DEFINE_WAIT(wait);
1477
1478 prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1479 dmz_unlock_map(zmd);
1480 dmz_unlock_metadata(zmd);
1481
1482 io_schedule_timeout(HZ);
1483
1484 dmz_lock_metadata(zmd);
1485 dmz_lock_map(zmd);
1486 finish_wait(&zmd->free_wq, &wait);
1487}
1488
1489/*
1490 * Lock a zone for reclaim (set the zone RECLAIM bit).
1491 * Returns false if the zone cannot be locked or if it is already locked
1492 * and 1 otherwise.
1493 */
1494int dmz_lock_zone_reclaim(struct dm_zone *zone)
1495{
1496 /* Active zones cannot be reclaimed */
1497 if (dmz_is_active(zone))
1498 return 0;
1499
1500 return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1501}
1502
1503/*
1504 * Clear a zone reclaim flag.
1505 */
1506void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1507{
1508 WARN_ON(dmz_is_active(zone));
1509 WARN_ON(!dmz_in_reclaim(zone));
1510
1511 clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1512 smp_mb__after_atomic();
1513 wake_up_bit(&zone->flags, DMZ_RECLAIM);
1514}
1515
1516/*
1517 * Wait for a zone reclaim to complete.
1518 */
1519static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1520{
1521 dmz_unlock_map(zmd);
1522 dmz_unlock_metadata(zmd);
1523 wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1524 dmz_lock_metadata(zmd);
1525 dmz_lock_map(zmd);
1526}
1527
1528/*
1529 * Select a random write zone for reclaim.
1530 */
1531static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd)
1532{
1533 struct dm_zone *dzone = NULL;
1534 struct dm_zone *zone;
1535
1536 if (list_empty(&zmd->map_rnd_list))
1537 return NULL;
1538
1539 list_for_each_entry(zone, &zmd->map_rnd_list, link) {
1540 if (dmz_is_buf(zone))
1541 dzone = zone->bzone;
1542 else
1543 dzone = zone;
1544 if (dmz_lock_zone_reclaim(dzone))
1545 return dzone;
1546 }
1547
1548 return NULL;
1549}
1550
1551/*
1552 * Select a buffered sequential zone for reclaim.
1553 */
1554static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd)
1555{
1556 struct dm_zone *zone;
1557
1558 if (list_empty(&zmd->map_seq_list))
1559 return NULL;
1560
1561 list_for_each_entry(zone, &zmd->map_seq_list, link) {
1562 if (!zone->bzone)
1563 continue;
1564 if (dmz_lock_zone_reclaim(zone))
1565 return zone;
1566 }
1567
1568 return NULL;
1569}
1570
1571/*
1572 * Select a zone for reclaim.
1573 */
1574struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd)
1575{
1576 struct dm_zone *zone;
1577
1578 /*
1579 * Search for a zone candidate to reclaim: 2 cases are possible.
1580 * (1) There is no free sequential zones. Then a random data zone
1581 * cannot be reclaimed. So choose a sequential zone to reclaim so
1582 * that afterward a random zone can be reclaimed.
1583 * (2) At least one free sequential zone is available, then choose
1584 * the oldest random zone (data or buffer) that can be locked.
1585 */
1586 dmz_lock_map(zmd);
1587 if (list_empty(&zmd->reserved_seq_zones_list))
1588 zone = dmz_get_seq_zone_for_reclaim(zmd);
1589 else
1590 zone = dmz_get_rnd_zone_for_reclaim(zmd);
1591 dmz_unlock_map(zmd);
1592
1593 return zone;
1594}
1595
1596/*
1597 * Activate a zone (increment its reference count).
1598 */
1599void dmz_activate_zone(struct dm_zone *zone)
1600{
1601 set_bit(DMZ_ACTIVE, &zone->flags);
1602 atomic_inc(&zone->refcount);
1603}
1604
1605/*
1606 * Deactivate a zone. This decrement the zone reference counter
1607 * and clears the active state of the zone once the count reaches 0,
1608 * indicating that all BIOs to the zone have completed. Returns
1609 * true if the zone was deactivated.
1610 */
1611void dmz_deactivate_zone(struct dm_zone *zone)
1612{
1613 if (atomic_dec_and_test(&zone->refcount)) {
1614 WARN_ON(!test_bit(DMZ_ACTIVE, &zone->flags));
1615 clear_bit_unlock(DMZ_ACTIVE, &zone->flags);
1616 smp_mb__after_atomic();
1617 }
1618}
1619
1620/*
1621 * Get the zone mapping a chunk, if the chunk is mapped already.
1622 * If no mapping exist and the operation is WRITE, a zone is
1623 * allocated and used to map the chunk.
1624 * The zone returned will be set to the active state.
1625 */
1626struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op)
1627{
1628 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1629 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1630 int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1631 unsigned int dzone_id;
1632 struct dm_zone *dzone = NULL;
1633 int ret = 0;
1634
1635 dmz_lock_map(zmd);
1636again:
1637 /* Get the chunk mapping */
1638 dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
1639 if (dzone_id == DMZ_MAP_UNMAPPED) {
1640 /*
1641 * Read or discard in unmapped chunks are fine. But for
1642 * writes, we need a mapping, so get one.
1643 */
1644 if (op != REQ_OP_WRITE)
1645 goto out;
1646
1647 /* Alloate a random zone */
1648 dzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1649 if (!dzone) {
1650 dmz_wait_for_free_zones(zmd);
1651 goto again;
1652 }
1653
1654 dmz_map_zone(zmd, dzone, chunk);
1655
1656 } else {
1657 /* The chunk is already mapped: get the mapping zone */
1658 dzone = dmz_get(zmd, dzone_id);
1659 if (dzone->chunk != chunk) {
1660 dzone = ERR_PTR(-EIO);
1661 goto out;
1662 }
1663
1664 /* Repair write pointer if the sequential dzone has error */
1665 if (dmz_seq_write_err(dzone)) {
1666 ret = dmz_handle_seq_write_err(zmd, dzone);
1667 if (ret) {
1668 dzone = ERR_PTR(-EIO);
1669 goto out;
1670 }
1671 clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
1672 }
1673 }
1674
1675 /*
1676 * If the zone is being reclaimed, the chunk mapping may change
1677 * to a different zone. So wait for reclaim and retry. Otherwise,
1678 * activate the zone (this will prevent reclaim from touching it).
1679 */
1680 if (dmz_in_reclaim(dzone)) {
1681 dmz_wait_for_reclaim(zmd, dzone);
1682 goto again;
1683 }
1684 dmz_activate_zone(dzone);
1685 dmz_lru_zone(zmd, dzone);
1686out:
1687 dmz_unlock_map(zmd);
1688
1689 return dzone;
1690}
1691
1692/*
1693 * Write and discard change the block validity of data zones and their buffer
1694 * zones. Check here that valid blocks are still present. If all blocks are
1695 * invalid, the zones can be unmapped on the fly without waiting for reclaim
1696 * to do it.
1697 */
1698void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
1699{
1700 struct dm_zone *bzone;
1701
1702 dmz_lock_map(zmd);
1703
1704 bzone = dzone->bzone;
1705 if (bzone) {
1706 if (dmz_weight(bzone))
1707 dmz_lru_zone(zmd, bzone);
1708 else {
1709 /* Empty buffer zone: reclaim it */
1710 dmz_unmap_zone(zmd, bzone);
1711 dmz_free_zone(zmd, bzone);
1712 bzone = NULL;
1713 }
1714 }
1715
1716 /* Deactivate the data zone */
1717 dmz_deactivate_zone(dzone);
1718 if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
1719 dmz_lru_zone(zmd, dzone);
1720 else {
1721 /* Unbuffered inactive empty data zone: reclaim it */
1722 dmz_unmap_zone(zmd, dzone);
1723 dmz_free_zone(zmd, dzone);
1724 }
1725
1726 dmz_unlock_map(zmd);
1727}
1728
1729/*
1730 * Allocate and map a random zone to buffer a chunk
1731 * already mapped to a sequential zone.
1732 */
1733struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
1734 struct dm_zone *dzone)
1735{
1736 struct dm_zone *bzone;
1737
1738 dmz_lock_map(zmd);
1739again:
1740 bzone = dzone->bzone;
1741 if (bzone)
1742 goto out;
1743
1744 /* Alloate a random zone */
1745 bzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1746 if (!bzone) {
1747 dmz_wait_for_free_zones(zmd);
1748 goto again;
1749 }
1750
1751 /* Update the chunk mapping */
1752 dmz_set_chunk_mapping(zmd, dzone->chunk, dmz_id(zmd, dzone),
1753 dmz_id(zmd, bzone));
1754
1755 set_bit(DMZ_BUF, &bzone->flags);
1756 bzone->chunk = dzone->chunk;
1757 bzone->bzone = dzone;
1758 dzone->bzone = bzone;
1759 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1760out:
1761 dmz_unlock_map(zmd);
1762
1763 return bzone;
1764}
1765
1766/*
1767 * Get an unmapped (free) zone.
1768 * This must be called with the mapping lock held.
1769 */
1770struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags)
1771{
1772 struct list_head *list;
1773 struct dm_zone *zone;
1774
1775 if (flags & DMZ_ALLOC_RND)
1776 list = &zmd->unmap_rnd_list;
1777 else
1778 list = &zmd->unmap_seq_list;
1779again:
1780 if (list_empty(list)) {
1781 /*
1782 * No free zone: if this is for reclaim, allow using the
1783 * reserved sequential zones.
1784 */
1785 if (!(flags & DMZ_ALLOC_RECLAIM) ||
1786 list_empty(&zmd->reserved_seq_zones_list))
1787 return NULL;
1788
1789 zone = list_first_entry(&zmd->reserved_seq_zones_list,
1790 struct dm_zone, link);
1791 list_del_init(&zone->link);
1792 atomic_dec(&zmd->nr_reserved_seq_zones);
1793 return zone;
1794 }
1795
1796 zone = list_first_entry(list, struct dm_zone, link);
1797 list_del_init(&zone->link);
1798
1799 if (dmz_is_rnd(zone))
1800 atomic_dec(&zmd->unmap_nr_rnd);
1801 else
1802 atomic_dec(&zmd->unmap_nr_seq);
1803
1804 if (dmz_is_offline(zone)) {
1805 dmz_dev_warn(zmd->dev, "Zone %u is offline", dmz_id(zmd, zone));
1806 zone = NULL;
1807 goto again;
1808 }
1809
1810 return zone;
1811}
1812
1813/*
1814 * Free a zone.
1815 * This must be called with the mapping lock held.
1816 */
1817void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1818{
1819 /* If this is a sequential zone, reset it */
1820 if (dmz_is_seq(zone))
1821 dmz_reset_zone(zmd, zone);
1822
1823 /* Return the zone to its type unmap list */
1824 if (dmz_is_rnd(zone)) {
1825 list_add_tail(&zone->link, &zmd->unmap_rnd_list);
1826 atomic_inc(&zmd->unmap_nr_rnd);
1827 } else if (atomic_read(&zmd->nr_reserved_seq_zones) <
1828 zmd->nr_reserved_seq) {
1829 list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
1830 atomic_inc(&zmd->nr_reserved_seq_zones);
1831 } else {
1832 list_add_tail(&zone->link, &zmd->unmap_seq_list);
1833 atomic_inc(&zmd->unmap_nr_seq);
1834 }
1835
1836 wake_up_all(&zmd->free_wq);
1837}
1838
1839/*
1840 * Map a chunk to a zone.
1841 * This must be called with the mapping lock held.
1842 */
1843void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
1844 unsigned int chunk)
1845{
1846 /* Set the chunk mapping */
1847 dmz_set_chunk_mapping(zmd, chunk, dmz_id(zmd, dzone),
1848 DMZ_MAP_UNMAPPED);
1849 dzone->chunk = chunk;
1850 if (dmz_is_rnd(dzone))
1851 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1852 else
1853 list_add_tail(&dzone->link, &zmd->map_seq_list);
1854}
1855
1856/*
1857 * Unmap a zone.
1858 * This must be called with the mapping lock held.
1859 */
1860void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1861{
1862 unsigned int chunk = zone->chunk;
1863 unsigned int dzone_id;
1864
1865 if (chunk == DMZ_MAP_UNMAPPED) {
1866 /* Already unmapped */
1867 return;
1868 }
1869
1870 if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
1871 /*
1872 * Unmapping the chunk buffer zone: clear only
1873 * the chunk buffer mapping
1874 */
1875 dzone_id = dmz_id(zmd, zone->bzone);
1876 zone->bzone->bzone = NULL;
1877 zone->bzone = NULL;
1878
1879 } else {
1880 /*
1881 * Unmapping the chunk data zone: the zone must
1882 * not be buffered.
1883 */
1884 if (WARN_ON(zone->bzone)) {
1885 zone->bzone->bzone = NULL;
1886 zone->bzone = NULL;
1887 }
1888 dzone_id = DMZ_MAP_UNMAPPED;
1889 }
1890
1891 dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
1892
1893 zone->chunk = DMZ_MAP_UNMAPPED;
1894 list_del_init(&zone->link);
1895}
1896
1897/*
1898 * Set @nr_bits bits in @bitmap starting from @bit.
1899 * Return the number of bits changed from 0 to 1.
1900 */
1901static unsigned int dmz_set_bits(unsigned long *bitmap,
1902 unsigned int bit, unsigned int nr_bits)
1903{
1904 unsigned long *addr;
1905 unsigned int end = bit + nr_bits;
1906 unsigned int n = 0;
1907
1908 while (bit < end) {
1909 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
1910 ((end - bit) >= BITS_PER_LONG)) {
1911 /* Try to set the whole word at once */
1912 addr = bitmap + BIT_WORD(bit);
1913 if (*addr == 0) {
1914 *addr = ULONG_MAX;
1915 n += BITS_PER_LONG;
1916 bit += BITS_PER_LONG;
1917 continue;
1918 }
1919 }
1920
1921 if (!test_and_set_bit(bit, bitmap))
1922 n++;
1923 bit++;
1924 }
1925
1926 return n;
1927}
1928
1929/*
1930 * Get the bitmap block storing the bit for chunk_block in zone.
1931 */
1932static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
1933 struct dm_zone *zone,
1934 sector_t chunk_block)
1935{
1936 sector_t bitmap_block = 1 + zmd->nr_map_blocks +
1937 (sector_t)(dmz_id(zmd, zone) * zmd->zone_nr_bitmap_blocks) +
1938 (chunk_block >> DMZ_BLOCK_SHIFT_BITS);
1939
1940 return dmz_get_mblock(zmd, bitmap_block);
1941}
1942
1943/*
1944 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
1945 */
1946int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1947 struct dm_zone *to_zone)
1948{
1949 struct dmz_mblock *from_mblk, *to_mblk;
1950 sector_t chunk_block = 0;
1951
1952 /* Get the zones bitmap blocks */
1953 while (chunk_block < zmd->dev->zone_nr_blocks) {
1954 from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
1955 if (IS_ERR(from_mblk))
1956 return PTR_ERR(from_mblk);
1957 to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
1958 if (IS_ERR(to_mblk)) {
1959 dmz_release_mblock(zmd, from_mblk);
1960 return PTR_ERR(to_mblk);
1961 }
1962
1963 memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
1964 dmz_dirty_mblock(zmd, to_mblk);
1965
1966 dmz_release_mblock(zmd, to_mblk);
1967 dmz_release_mblock(zmd, from_mblk);
1968
1969 chunk_block += DMZ_BLOCK_SIZE_BITS;
1970 }
1971
1972 to_zone->weight = from_zone->weight;
1973
1974 return 0;
1975}
1976
1977/*
1978 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
1979 * starting from chunk_block.
1980 */
1981int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1982 struct dm_zone *to_zone, sector_t chunk_block)
1983{
1984 unsigned int nr_blocks;
1985 int ret;
1986
1987 /* Get the zones bitmap blocks */
1988 while (chunk_block < zmd->dev->zone_nr_blocks) {
1989 /* Get a valid region from the source zone */
1990 ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
1991 if (ret <= 0)
1992 return ret;
1993
1994 nr_blocks = ret;
1995 ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
1996 if (ret)
1997 return ret;
1998
1999 chunk_block += nr_blocks;
2000 }
2001
2002 return 0;
2003}
2004
2005/*
2006 * Validate all the blocks in the range [block..block+nr_blocks-1].
2007 */
2008int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2009 sector_t chunk_block, unsigned int nr_blocks)
2010{
2011 unsigned int count, bit, nr_bits;
2012 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
2013 struct dmz_mblock *mblk;
2014 unsigned int n = 0;
2015
2016 dmz_dev_debug(zmd->dev, "=> VALIDATE zone %u, block %llu, %u blocks",
2017 dmz_id(zmd, zone), (unsigned long long)chunk_block,
2018 nr_blocks);
2019
2020 WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2021
2022 while (nr_blocks) {
2023 /* Get bitmap block */
2024 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2025 if (IS_ERR(mblk))
2026 return PTR_ERR(mblk);
2027
2028 /* Set bits */
2029 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2030 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2031
2032 count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2033 if (count) {
2034 dmz_dirty_mblock(zmd, mblk);
2035 n += count;
2036 }
2037 dmz_release_mblock(zmd, mblk);
2038
2039 nr_blocks -= nr_bits;
2040 chunk_block += nr_bits;
2041 }
2042
2043 if (likely(zone->weight + n <= zone_nr_blocks))
2044 zone->weight += n;
2045 else {
2046 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be <= %u",
2047 dmz_id(zmd, zone), zone->weight,
2048 zone_nr_blocks - n);
2049 zone->weight = zone_nr_blocks;
2050 }
2051
2052 return 0;
2053}
2054
2055/*
2056 * Clear nr_bits bits in bitmap starting from bit.
2057 * Return the number of bits cleared.
2058 */
2059static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2060{
2061 unsigned long *addr;
2062 int end = bit + nr_bits;
2063 int n = 0;
2064
2065 while (bit < end) {
2066 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2067 ((end - bit) >= BITS_PER_LONG)) {
2068 /* Try to clear whole word at once */
2069 addr = bitmap + BIT_WORD(bit);
2070 if (*addr == ULONG_MAX) {
2071 *addr = 0;
2072 n += BITS_PER_LONG;
2073 bit += BITS_PER_LONG;
2074 continue;
2075 }
2076 }
2077
2078 if (test_and_clear_bit(bit, bitmap))
2079 n++;
2080 bit++;
2081 }
2082
2083 return n;
2084}
2085
2086/*
2087 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2088 */
2089int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2090 sector_t chunk_block, unsigned int nr_blocks)
2091{
2092 unsigned int count, bit, nr_bits;
2093 struct dmz_mblock *mblk;
2094 unsigned int n = 0;
2095
2096 dmz_dev_debug(zmd->dev, "=> INVALIDATE zone %u, block %llu, %u blocks",
2097 dmz_id(zmd, zone), (u64)chunk_block, nr_blocks);
2098
2099 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2100
2101 while (nr_blocks) {
2102 /* Get bitmap block */
2103 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2104 if (IS_ERR(mblk))
2105 return PTR_ERR(mblk);
2106
2107 /* Clear bits */
2108 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2109 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2110
2111 count = dmz_clear_bits((unsigned long *)mblk->data,
2112 bit, nr_bits);
2113 if (count) {
2114 dmz_dirty_mblock(zmd, mblk);
2115 n += count;
2116 }
2117 dmz_release_mblock(zmd, mblk);
2118
2119 nr_blocks -= nr_bits;
2120 chunk_block += nr_bits;
2121 }
2122
2123 if (zone->weight >= n)
2124 zone->weight -= n;
2125 else {
2126 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be >= %u",
2127 dmz_id(zmd, zone), zone->weight, n);
2128 zone->weight = 0;
2129 }
2130
2131 return 0;
2132}
2133
2134/*
2135 * Get a block bit value.
2136 */
2137static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2138 sector_t chunk_block)
2139{
2140 struct dmz_mblock *mblk;
2141 int ret;
2142
2143 WARN_ON(chunk_block >= zmd->dev->zone_nr_blocks);
2144
2145 /* Get bitmap block */
2146 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2147 if (IS_ERR(mblk))
2148 return PTR_ERR(mblk);
2149
2150 /* Get offset */
2151 ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2152 (unsigned long *) mblk->data) != 0;
2153
2154 dmz_release_mblock(zmd, mblk);
2155
2156 return ret;
2157}
2158
2159/*
2160 * Return the number of blocks from chunk_block to the first block with a bit
2161 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2162 */
2163static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2164 sector_t chunk_block, unsigned int nr_blocks,
2165 int set)
2166{
2167 struct dmz_mblock *mblk;
2168 unsigned int bit, set_bit, nr_bits;
2169 unsigned long *bitmap;
2170 int n = 0;
2171
2172 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2173
2174 while (nr_blocks) {
2175 /* Get bitmap block */
2176 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2177 if (IS_ERR(mblk))
2178 return PTR_ERR(mblk);
2179
2180 /* Get offset */
2181 bitmap = (unsigned long *) mblk->data;
2182 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2183 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2184 if (set)
2185 set_bit = find_next_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2186 else
2187 set_bit = find_next_zero_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2188 dmz_release_mblock(zmd, mblk);
2189
2190 n += set_bit - bit;
2191 if (set_bit < DMZ_BLOCK_SIZE_BITS)
2192 break;
2193
2194 nr_blocks -= nr_bits;
2195 chunk_block += nr_bits;
2196 }
2197
2198 return n;
2199}
2200
2201/*
2202 * Test if chunk_block is valid. If it is, the number of consecutive
2203 * valid blocks from chunk_block will be returned.
2204 */
2205int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2206 sector_t chunk_block)
2207{
2208 int valid;
2209
2210 valid = dmz_test_block(zmd, zone, chunk_block);
2211 if (valid <= 0)
2212 return valid;
2213
2214 /* The block is valid: get the number of valid blocks from block */
2215 return dmz_to_next_set_block(zmd, zone, chunk_block,
2216 zmd->dev->zone_nr_blocks - chunk_block, 0);
2217}
2218
2219/*
2220 * Find the first valid block from @chunk_block in @zone.
2221 * If such a block is found, its number is returned using
2222 * @chunk_block and the total number of valid blocks from @chunk_block
2223 * is returned.
2224 */
2225int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2226 sector_t *chunk_block)
2227{
2228 sector_t start_block = *chunk_block;
2229 int ret;
2230
2231 ret = dmz_to_next_set_block(zmd, zone, start_block,
2232 zmd->dev->zone_nr_blocks - start_block, 1);
2233 if (ret < 0)
2234 return ret;
2235
2236 start_block += ret;
2237 *chunk_block = start_block;
2238
2239 return dmz_to_next_set_block(zmd, zone, start_block,
2240 zmd->dev->zone_nr_blocks - start_block, 0);
2241}
2242
2243/*
2244 * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2245 */
2246static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2247{
2248 unsigned long *addr;
2249 int end = bit + nr_bits;
2250 int n = 0;
2251
2252 while (bit < end) {
2253 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2254 ((end - bit) >= BITS_PER_LONG)) {
2255 addr = (unsigned long *)bitmap + BIT_WORD(bit);
2256 if (*addr == ULONG_MAX) {
2257 n += BITS_PER_LONG;
2258 bit += BITS_PER_LONG;
2259 continue;
2260 }
2261 }
2262
2263 if (test_bit(bit, bitmap))
2264 n++;
2265 bit++;
2266 }
2267
2268 return n;
2269}
2270
2271/*
2272 * Get a zone weight.
2273 */
2274static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2275{
2276 struct dmz_mblock *mblk;
2277 sector_t chunk_block = 0;
2278 unsigned int bit, nr_bits;
2279 unsigned int nr_blocks = zmd->dev->zone_nr_blocks;
2280 void *bitmap;
2281 int n = 0;
2282
2283 while (nr_blocks) {
2284 /* Get bitmap block */
2285 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2286 if (IS_ERR(mblk)) {
2287 n = 0;
2288 break;
2289 }
2290
2291 /* Count bits in this block */
2292 bitmap = mblk->data;
2293 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2294 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2295 n += dmz_count_bits(bitmap, bit, nr_bits);
2296
2297 dmz_release_mblock(zmd, mblk);
2298
2299 nr_blocks -= nr_bits;
2300 chunk_block += nr_bits;
2301 }
2302
2303 zone->weight = n;
2304}
2305
2306/*
2307 * Cleanup the zoned metadata resources.
2308 */
2309static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2310{
2311 struct rb_root *root;
2312 struct dmz_mblock *mblk, *next;
2313 int i;
2314
2315 /* Release zone mapping resources */
2316 if (zmd->map_mblk) {
2317 for (i = 0; i < zmd->nr_map_blocks; i++)
2318 dmz_release_mblock(zmd, zmd->map_mblk[i]);
2319 kfree(zmd->map_mblk);
2320 zmd->map_mblk = NULL;
2321 }
2322
2323 /* Release super blocks */
2324 for (i = 0; i < 2; i++) {
2325 if (zmd->sb[i].mblk) {
2326 dmz_free_mblock(zmd, zmd->sb[i].mblk);
2327 zmd->sb[i].mblk = NULL;
2328 }
2329 }
2330
2331 /* Free cached blocks */
2332 while (!list_empty(&zmd->mblk_dirty_list)) {
2333 mblk = list_first_entry(&zmd->mblk_dirty_list,
2334 struct dmz_mblock, link);
2335 dmz_dev_warn(zmd->dev, "mblock %llu still in dirty list (ref %u)",
33c2865f 2336 (u64)mblk->no, mblk->ref);
3b1a94c8
DLM		 2337 list_del_init(&mblk->link);
2338 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2339 dmz_free_mblock(zmd, mblk);
2340 }
2341
2342 while (!list_empty(&zmd->mblk_lru_list)) {
2343 mblk = list_first_entry(&zmd->mblk_lru_list,
2344 struct dmz_mblock, link);
2345 list_del_init(&mblk->link);
2346 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2347 dmz_free_mblock(zmd, mblk);
2348 }
2349
2350 /* Sanity checks: the mblock rbtree should now be empty */
2351 root = &zmd->mblk_rbtree;
2352 rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2353 dmz_dev_warn(zmd->dev, "mblock %llu ref %u still in rbtree",
33c2865f
DLM		 2354 (u64)mblk->no, mblk->ref);
2355 mblk->ref = 0;
3b1a94c8
DLM		 2356 dmz_free_mblock(zmd, mblk);
2357 }
2358
2359 /* Free the zone descriptors */
2360 dmz_drop_zones(zmd);
d5ffebdd
MS		 2361
2362 mutex_destroy(&zmd->mblk_flush_lock);
2363 mutex_destroy(&zmd->map_lock);
3b1a94c8
DLM		 2364}
2365
2366/*
2367 * Initialize the zoned metadata.
2368 */
2369int dmz_ctr_metadata(struct dmz_dev *dev, struct dmz_metadata **metadata)
2370{
2371 struct dmz_metadata *zmd;
2372 unsigned int i, zid;
2373 struct dm_zone *zone;
2374 int ret;
2375
2376 zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2377 if (!zmd)
2378 return -ENOMEM;
2379
2380 zmd->dev = dev;
2381 zmd->mblk_rbtree = RB_ROOT;
2382 init_rwsem(&zmd->mblk_sem);
2383 mutex_init(&zmd->mblk_flush_lock);
2384 spin_lock_init(&zmd->mblk_lock);
2385 INIT_LIST_HEAD(&zmd->mblk_lru_list);
2386 INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2387
2388 mutex_init(&zmd->map_lock);
2389 atomic_set(&zmd->unmap_nr_rnd, 0);
2390 INIT_LIST_HEAD(&zmd->unmap_rnd_list);
2391 INIT_LIST_HEAD(&zmd->map_rnd_list);
2392
2393 atomic_set(&zmd->unmap_nr_seq, 0);
2394 INIT_LIST_HEAD(&zmd->unmap_seq_list);
2395 INIT_LIST_HEAD(&zmd->map_seq_list);
2396
2397 atomic_set(&zmd->nr_reserved_seq_zones, 0);
2398 INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2399
2400 init_waitqueue_head(&zmd->free_wq);
2401
2402 /* Initialize zone descriptors */
2403 ret = dmz_init_zones(zmd);
2404 if (ret)
2405 goto err;
2406
2407 /* Get super block */
2408 ret = dmz_load_sb(zmd);
2409 if (ret)
2410 goto err;
2411
2412 /* Set metadata zones starting from sb_zone */
2413 zid = dmz_id(zmd, zmd->sb_zone);
2414 for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2415 zone = dmz_get(zmd, zid + i);
2416 if (!dmz_is_rnd(zone))
2417 goto err;
2418 set_bit(DMZ_META, &zone->flags);
2419 }
2420
2421 /* Load mapping table */
2422 ret = dmz_load_mapping(zmd);
2423 if (ret)
2424 goto err;
2425
2426 /*
2427 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2428 * blocks and enough blocks to be able to cache the bitmap blocks of
2429 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2430 * the cache to add 512 more metadata blocks.
2431 */
2432 zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2433 zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2434 zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2435 zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2436 zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2437
2438 /* Metadata cache shrinker */
2439 ret = register_shrinker(&zmd->mblk_shrinker);
2440 if (ret) {
2441 dmz_dev_err(dev, "Register metadata cache shrinker failed");
2442 goto err;
2443 }
2444
2445 dmz_dev_info(dev, "Host-%s zoned block device",
2446 bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2447 "aware" : "managed");
2448 dmz_dev_info(dev, " %llu 512-byte logical sectors",
2449 (u64)dev->capacity);
2450 dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors",
2451 dev->nr_zones, (u64)dev->zone_nr_sectors);
2452 dmz_dev_info(dev, " %u metadata zones",
2453 zmd->nr_meta_zones * 2);
2454 dmz_dev_info(dev, " %u data zones for %u chunks",
2455 zmd->nr_data_zones, zmd->nr_chunks);
2456 dmz_dev_info(dev, " %u random zones (%u unmapped)",
2457 zmd->nr_rnd, atomic_read(&zmd->unmap_nr_rnd));
2458 dmz_dev_info(dev, " %u sequential zones (%u unmapped)",
2459 zmd->nr_seq, atomic_read(&zmd->unmap_nr_seq));
2460 dmz_dev_info(dev, " %u reserved sequential data zones",
2461 zmd->nr_reserved_seq);
2462
2463 dmz_dev_debug(dev, "Format:");
2464 dmz_dev_debug(dev, "%u metadata blocks per set (%u max cache)",
2465 zmd->nr_meta_blocks, zmd->max_nr_mblks);
2466 dmz_dev_debug(dev, " %u data zone mapping blocks",
2467 zmd->nr_map_blocks);
2468 dmz_dev_debug(dev, " %u bitmap blocks",
2469 zmd->nr_bitmap_blocks);
2470
2471 *metadata = zmd;
2472
2473 return 0;
2474err:
2475 dmz_cleanup_metadata(zmd);
2476 kfree(zmd);
2477 *metadata = NULL;
2478
2479 return ret;
2480}
2481
2482/*
2483 * Cleanup the zoned metadata resources.
2484 */
2485void dmz_dtr_metadata(struct dmz_metadata *zmd)
2486{
2487 unregister_shrinker(&zmd->mblk_shrinker);
2488 dmz_cleanup_metadata(zmd);
2489 kfree(zmd);
2490}
2491
2492/*
2493 * Check zone information on resume.
2494 */
2495int dmz_resume_metadata(struct dmz_metadata *zmd)
2496{
2497 struct dmz_dev *dev = zmd->dev;
2498 struct dm_zone *zone;
2499 sector_t wp_block;
2500 unsigned int i;
2501 int ret;
2502
2503 /* Check zones */
2504 for (i = 0; i < dev->nr_zones; i++) {
2505 zone = dmz_get(zmd, i);
2506 if (!zone) {
2507 dmz_dev_err(dev, "Unable to get zone %u", i);
2508 return -EIO;
2509 }
2510
2511 wp_block = zone->wp_block;
2512
2513 ret = dmz_update_zone(zmd, zone);
2514 if (ret) {
2515 dmz_dev_err(dev, "Broken zone %u", i);
2516 return ret;
2517 }
2518
2519 if (dmz_is_offline(zone)) {
2520 dmz_dev_warn(dev, "Zone %u is offline", i);
2521 continue;
2522 }
2523
2524 /* Check write pointer */
2525 if (!dmz_is_seq(zone))
2526 zone->wp_block = 0;
2527 else if (zone->wp_block != wp_block) {
2528 dmz_dev_err(dev, "Zone %u: Invalid wp (%llu / %llu)",
2529 i, (u64)zone->wp_block, (u64)wp_block);
2530 zone->wp_block = wp_block;
2531 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
2532 dev->zone_nr_blocks - zone->wp_block);
2533 }
2534 }
2535
2536 return 0;
2537}
Ubuntu Hirsute Linux kernel mirror