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;
102 unsigned int ref;
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{
194 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_sectors_shift;
195}
196
197sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone)
198{
199 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_blocks_shift;
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);
299 mblk->ref = 0;
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/*
342 * Lookup a metadata block in the rbtree. If the block is found, increment
343 * its reference count.
344 */
345static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd,
346 sector_t mblk_no)
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);
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);
363 return mblk;
364 }
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/*
395 * Read an uncached metadata block from disk and add it to the cache.
396 */
397static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd,
398 sector_t mblk_no)
399{
400 struct dmz_mblock *mblk, *m;
401 sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no;
402 struct bio *bio;
403
404 /* Get a new block and a BIO to read it */
405 mblk = dmz_alloc_mblock(zmd, mblk_no);
406 if (!mblk)
407 return NULL;
408
409 bio = bio_alloc(GFP_NOIO, 1);
410 if (!bio) {
411 dmz_free_mblock(zmd, mblk);
412 return NULL;
413 }
414
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 */
436 bio->bi_iter.bi_sector = dmz_blk2sect(block);
437 bio_set_dev(bio, zmd->dev->bdev);
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
512 mblk->ref--;
513 if (mblk->ref == 0) {
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);
537 mblk = dmz_get_mblock_fast(zmd, mblk_no);
538 spin_unlock(&zmd->mblk_lock);
539
540 if (!mblk) {
541 /* Cache miss: read the block from disk */
542 mblk = dmz_get_mblock_slow(zmd, mblk_no);
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);
587 bio_set_dev(bio, zmd->dev->bdev);
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);
609 bio_set_dev(bio, zmd->dev->bdev);
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)
647 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
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)
681 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
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)) {
745 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
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);
776 if (mblk->ref == 0)
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
950 page = alloc_page(GFP_NOIO);
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
1172 /* Process report */
1173 for (i = 0; i < nr_blkz; i++) {
1174 ret = dmz_init_zone(zmd, zone, &blkz[i]);
1175 if (ret)
1176 goto out;
1177 sector += dev->zone_nr_sectors;
1178 zone++;
1179 }
1180 }
1181
1182 /* The entire zone configuration of the disk should now be known */
1183 if (sector < dev->capacity) {
1184 dmz_dev_err(dev, "Failed to get correct zone information");
1185 ret = -ENXIO;
1186 }
1187out:
1188 kfree(blkz);
1189 if (ret)
1190 dmz_drop_zones(zmd);
1191
1192 return ret;
1193}
1194
1195/*
1196 * Update a zone information.
1197 */
1198static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1199{
1200 unsigned int nr_blkz = 1;
1201 struct blk_zone blkz;
1202 int ret;
1203
1204 /* Get zone information from disk */
1205 ret = blkdev_report_zones(zmd->dev->bdev, dmz_start_sect(zmd, zone),
1206 &blkz, &nr_blkz, GFP_NOIO);
1207 if (ret) {
1208 dmz_dev_err(zmd->dev, "Get zone %u report failed",
1209 dmz_id(zmd, zone));
1210 return ret;
1211 }
1212
1213 clear_bit(DMZ_OFFLINE, &zone->flags);
1214 clear_bit(DMZ_READ_ONLY, &zone->flags);
1215 if (blkz.cond == BLK_ZONE_COND_OFFLINE)
1216 set_bit(DMZ_OFFLINE, &zone->flags);
1217 else if (blkz.cond == BLK_ZONE_COND_READONLY)
1218 set_bit(DMZ_READ_ONLY, &zone->flags);
1219
1220 if (dmz_is_seq(zone))
1221 zone->wp_block = dmz_sect2blk(blkz.wp - blkz.start);
1222 else
1223 zone->wp_block = 0;
1224
1225 return 0;
1226}
1227
1228/*
1229 * Check a zone write pointer position when the zone is marked
1230 * with the sequential write error flag.
1231 */
1232static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1233 struct dm_zone *zone)
1234{
1235 unsigned int wp = 0;
1236 int ret;
1237
1238 wp = zone->wp_block;
1239 ret = dmz_update_zone(zmd, zone);
1240 if (ret)
1241 return ret;
1242
1243 dmz_dev_warn(zmd->dev, "Processing zone %u write error (zone wp %u/%u)",
1244 dmz_id(zmd, zone), zone->wp_block, wp);
1245
1246 if (zone->wp_block < wp) {
1247 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1248 wp - zone->wp_block);
1249 }
1250
1251 return 0;
1252}
1253
1254static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
1255{
1256 return &zmd->zones[zone_id];
1257}
1258
1259/*
1260 * Reset a zone write pointer.
1261 */
1262static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1263{
1264 int ret;
1265
1266 /*
1267 * Ignore offline zones, read only zones,
1268 * and conventional zones.
1269 */
1270 if (dmz_is_offline(zone) ||
1271 dmz_is_readonly(zone) ||
1272 dmz_is_rnd(zone))
1273 return 0;
1274
1275 if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1276 struct dmz_dev *dev = zmd->dev;
1277
1278 ret = blkdev_reset_zones(dev->bdev,
1279 dmz_start_sect(zmd, zone),
1280 dev->zone_nr_sectors, GFP_NOIO);
1281 if (ret) {
1282 dmz_dev_err(dev, "Reset zone %u failed %d",
1283 dmz_id(zmd, zone), ret);
1284 return ret;
1285 }
1286 }
1287
1288 /* Clear write error bit and rewind write pointer position */
1289 clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1290 zone->wp_block = 0;
1291
1292 return 0;
1293}
1294
1295static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1296
1297/*
1298 * Initialize chunk mapping.
1299 */
1300static int dmz_load_mapping(struct dmz_metadata *zmd)
1301{
1302 struct dmz_dev *dev = zmd->dev;
1303 struct dm_zone *dzone, *bzone;
1304 struct dmz_mblock *dmap_mblk = NULL;
1305 struct dmz_map *dmap;
1306 unsigned int i = 0, e = 0, chunk = 0;
1307 unsigned int dzone_id;
1308 unsigned int bzone_id;
1309
1310 /* Metadata block array for the chunk mapping table */
1311 zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1312 sizeof(struct dmz_mblk *), GFP_KERNEL);
1313 if (!zmd->map_mblk)
1314 return -ENOMEM;
1315
1316 /* Get chunk mapping table blocks and initialize zone mapping */
1317 while (chunk < zmd->nr_chunks) {
1318 if (!dmap_mblk) {
1319 /* Get mapping block */
1320 dmap_mblk = dmz_get_mblock(zmd, i + 1);
1321 if (IS_ERR(dmap_mblk))
1322 return PTR_ERR(dmap_mblk);
1323 zmd->map_mblk[i] = dmap_mblk;
1324 dmap = (struct dmz_map *) dmap_mblk->data;
1325 i++;
1326 e = 0;
1327 }
1328
1329 /* Check data zone */
1330 dzone_id = le32_to_cpu(dmap[e].dzone_id);
1331 if (dzone_id == DMZ_MAP_UNMAPPED)
1332 goto next;
1333
1334 if (dzone_id >= dev->nr_zones) {
1335 dmz_dev_err(dev, "Chunk %u mapping: invalid data zone ID %u",
1336 chunk, dzone_id);
1337 return -EIO;
1338 }
1339
1340 dzone = dmz_get(zmd, dzone_id);
1341 set_bit(DMZ_DATA, &dzone->flags);
1342 dzone->chunk = chunk;
1343 dmz_get_zone_weight(zmd, dzone);
1344
1345 if (dmz_is_rnd(dzone))
1346 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1347 else
1348 list_add_tail(&dzone->link, &zmd->map_seq_list);
1349
1350 /* Check buffer zone */
1351 bzone_id = le32_to_cpu(dmap[e].bzone_id);
1352 if (bzone_id == DMZ_MAP_UNMAPPED)
1353 goto next;
1354
1355 if (bzone_id >= dev->nr_zones) {
1356 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone ID %u",
1357 chunk, bzone_id);
1358 return -EIO;
1359 }
1360
1361 bzone = dmz_get(zmd, bzone_id);
1362 if (!dmz_is_rnd(bzone)) {
1363 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone %u",
1364 chunk, bzone_id);
1365 return -EIO;
1366 }
1367
1368 set_bit(DMZ_DATA, &bzone->flags);
1369 set_bit(DMZ_BUF, &bzone->flags);
1370 bzone->chunk = chunk;
1371 bzone->bzone = dzone;
1372 dzone->bzone = bzone;
1373 dmz_get_zone_weight(zmd, bzone);
1374 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1375next:
1376 chunk++;
1377 e++;
1378 if (e >= DMZ_MAP_ENTRIES)
1379 dmap_mblk = NULL;
1380 }
1381
1382 /*
1383 * At this point, only meta zones and mapped data zones were
1384 * fully initialized. All remaining zones are unmapped data
1385 * zones. Finish initializing those here.
1386 */
1387 for (i = 0; i < dev->nr_zones; i++) {
1388 dzone = dmz_get(zmd, i);
1389 if (dmz_is_meta(dzone))
1390 continue;
1391
1392 if (dmz_is_rnd(dzone))
1393 zmd->nr_rnd++;
1394 else
1395 zmd->nr_seq++;
1396
1397 if (dmz_is_data(dzone)) {
1398 /* Already initialized */
1399 continue;
1400 }
1401
1402 /* Unmapped data zone */
1403 set_bit(DMZ_DATA, &dzone->flags);
1404 dzone->chunk = DMZ_MAP_UNMAPPED;
1405 if (dmz_is_rnd(dzone)) {
1406 list_add_tail(&dzone->link, &zmd->unmap_rnd_list);
1407 atomic_inc(&zmd->unmap_nr_rnd);
1408 } else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1409 list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1410 atomic_inc(&zmd->nr_reserved_seq_zones);
1411 zmd->nr_seq--;
1412 } else {
1413 list_add_tail(&dzone->link, &zmd->unmap_seq_list);
1414 atomic_inc(&zmd->unmap_nr_seq);
1415 }
1416 }
1417
1418 return 0;
1419}
1420
1421/*
1422 * Set a data chunk mapping.
1423 */
1424static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1425 unsigned int dzone_id, unsigned int bzone_id)
1426{
1427 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1428 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1429 int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1430
1431 dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1432 dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1433 dmz_dirty_mblock(zmd, dmap_mblk);
1434}
1435
1436/*
1437 * The list of mapped zones is maintained in LRU order.
1438 * This rotates a zone at the end of its map list.
1439 */
1440static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1441{
1442 if (list_empty(&zone->link))
1443 return;
1444
1445 list_del_init(&zone->link);
1446 if (dmz_is_seq(zone)) {
1447 /* LRU rotate sequential zone */
1448 list_add_tail(&zone->link, &zmd->map_seq_list);
1449 } else {
1450 /* LRU rotate random zone */
1451 list_add_tail(&zone->link, &zmd->map_rnd_list);
1452 }
1453}
1454
1455/*
1456 * The list of mapped random zones is maintained
1457 * in LRU order. This rotates a zone at the end of the list.
1458 */
1459static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1460{
1461 __dmz_lru_zone(zmd, zone);
1462 if (zone->bzone)
1463 __dmz_lru_zone(zmd, zone->bzone);
1464}
1465
1466/*
1467 * Wait for any zone to be freed.
1468 */
1469static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1470{
1471 DEFINE_WAIT(wait);
1472
1473 prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1474 dmz_unlock_map(zmd);
1475 dmz_unlock_metadata(zmd);
1476
1477 io_schedule_timeout(HZ);
1478
1479 dmz_lock_metadata(zmd);
1480 dmz_lock_map(zmd);
1481 finish_wait(&zmd->free_wq, &wait);
1482}
1483
1484/*
1485 * Lock a zone for reclaim (set the zone RECLAIM bit).
1486 * Returns false if the zone cannot be locked or if it is already locked
1487 * and 1 otherwise.
1488 */
1489int dmz_lock_zone_reclaim(struct dm_zone *zone)
1490{
1491 /* Active zones cannot be reclaimed */
1492 if (dmz_is_active(zone))
1493 return 0;
1494
1495 return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1496}
1497
1498/*
1499 * Clear a zone reclaim flag.
1500 */
1501void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1502{
1503 WARN_ON(dmz_is_active(zone));
1504 WARN_ON(!dmz_in_reclaim(zone));
1505
1506 clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1507 smp_mb__after_atomic();
1508 wake_up_bit(&zone->flags, DMZ_RECLAIM);
1509}
1510
1511/*
1512 * Wait for a zone reclaim to complete.
1513 */
1514static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1515{
1516 dmz_unlock_map(zmd);
1517 dmz_unlock_metadata(zmd);
1518 wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1519 dmz_lock_metadata(zmd);
1520 dmz_lock_map(zmd);
1521}
1522
1523/*
1524 * Select a random write zone for reclaim.
1525 */
1526static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd)
1527{
1528 struct dm_zone *dzone = NULL;
1529 struct dm_zone *zone;
1530
1531 if (list_empty(&zmd->map_rnd_list))
1532 return NULL;
1533
1534 list_for_each_entry(zone, &zmd->map_rnd_list, link) {
1535 if (dmz_is_buf(zone))
1536 dzone = zone->bzone;
1537 else
1538 dzone = zone;
1539 if (dmz_lock_zone_reclaim(dzone))
1540 return dzone;
1541 }
1542
1543 return NULL;
1544}
1545
1546/*
1547 * Select a buffered sequential zone for reclaim.
1548 */
1549static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd)
1550{
1551 struct dm_zone *zone;
1552
1553 if (list_empty(&zmd->map_seq_list))
1554 return NULL;
1555
1556 list_for_each_entry(zone, &zmd->map_seq_list, link) {
1557 if (!zone->bzone)
1558 continue;
1559 if (dmz_lock_zone_reclaim(zone))
1560 return zone;
1561 }
1562
1563 return NULL;
1564}
1565
1566/*
1567 * Select a zone for reclaim.
1568 */
1569struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd)
1570{
1571 struct dm_zone *zone;
1572
1573 /*
1574 * Search for a zone candidate to reclaim: 2 cases are possible.
1575 * (1) There is no free sequential zones. Then a random data zone
1576 * cannot be reclaimed. So choose a sequential zone to reclaim so
1577 * that afterward a random zone can be reclaimed.
1578 * (2) At least one free sequential zone is available, then choose
1579 * the oldest random zone (data or buffer) that can be locked.
1580 */
1581 dmz_lock_map(zmd);
1582 if (list_empty(&zmd->reserved_seq_zones_list))
1583 zone = dmz_get_seq_zone_for_reclaim(zmd);
1584 else
1585 zone = dmz_get_rnd_zone_for_reclaim(zmd);
1586 dmz_unlock_map(zmd);
1587
1588 return zone;
1589}
1590
1591/*
1592 * Activate a zone (increment its reference count).
1593 */
1594void dmz_activate_zone(struct dm_zone *zone)
1595{
1596 set_bit(DMZ_ACTIVE, &zone->flags);
1597 atomic_inc(&zone->refcount);
1598}
1599
1600/*
1601 * Deactivate a zone. This decrement the zone reference counter
1602 * and clears the active state of the zone once the count reaches 0,
1603 * indicating that all BIOs to the zone have completed. Returns
1604 * true if the zone was deactivated.
1605 */
1606void dmz_deactivate_zone(struct dm_zone *zone)
1607{
1608 if (atomic_dec_and_test(&zone->refcount)) {
1609 WARN_ON(!test_bit(DMZ_ACTIVE, &zone->flags));
1610 clear_bit_unlock(DMZ_ACTIVE, &zone->flags);
1611 smp_mb__after_atomic();
1612 }
1613}
1614
1615/*
1616 * Get the zone mapping a chunk, if the chunk is mapped already.
1617 * If no mapping exist and the operation is WRITE, a zone is
1618 * allocated and used to map the chunk.
1619 * The zone returned will be set to the active state.
1620 */
1621struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op)
1622{
1623 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1624 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1625 int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1626 unsigned int dzone_id;
1627 struct dm_zone *dzone = NULL;
1628 int ret = 0;
1629
1630 dmz_lock_map(zmd);
1631again:
1632 /* Get the chunk mapping */
1633 dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
1634 if (dzone_id == DMZ_MAP_UNMAPPED) {
1635 /*
1636 * Read or discard in unmapped chunks are fine. But for
1637 * writes, we need a mapping, so get one.
1638 */
1639 if (op != REQ_OP_WRITE)
1640 goto out;
1641
1642 /* Alloate a random zone */
1643 dzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1644 if (!dzone) {
1645 dmz_wait_for_free_zones(zmd);
1646 goto again;
1647 }
1648
1649 dmz_map_zone(zmd, dzone, chunk);
1650
1651 } else {
1652 /* The chunk is already mapped: get the mapping zone */
1653 dzone = dmz_get(zmd, dzone_id);
1654 if (dzone->chunk != chunk) {
1655 dzone = ERR_PTR(-EIO);
1656 goto out;
1657 }
1658
1659 /* Repair write pointer if the sequential dzone has error */
1660 if (dmz_seq_write_err(dzone)) {
1661 ret = dmz_handle_seq_write_err(zmd, dzone);
1662 if (ret) {
1663 dzone = ERR_PTR(-EIO);
1664 goto out;
1665 }
1666 clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
1667 }
1668 }
1669
1670 /*
1671 * If the zone is being reclaimed, the chunk mapping may change
1672 * to a different zone. So wait for reclaim and retry. Otherwise,
1673 * activate the zone (this will prevent reclaim from touching it).
1674 */
1675 if (dmz_in_reclaim(dzone)) {
1676 dmz_wait_for_reclaim(zmd, dzone);
1677 goto again;
1678 }
1679 dmz_activate_zone(dzone);
1680 dmz_lru_zone(zmd, dzone);
1681out:
1682 dmz_unlock_map(zmd);
1683
1684 return dzone;
1685}
1686
1687/*
1688 * Write and discard change the block validity of data zones and their buffer
1689 * zones. Check here that valid blocks are still present. If all blocks are
1690 * invalid, the zones can be unmapped on the fly without waiting for reclaim
1691 * to do it.
1692 */
1693void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
1694{
1695 struct dm_zone *bzone;
1696
1697 dmz_lock_map(zmd);
1698
1699 bzone = dzone->bzone;
1700 if (bzone) {
1701 if (dmz_weight(bzone))
1702 dmz_lru_zone(zmd, bzone);
1703 else {
1704 /* Empty buffer zone: reclaim it */
1705 dmz_unmap_zone(zmd, bzone);
1706 dmz_free_zone(zmd, bzone);
1707 bzone = NULL;
1708 }
1709 }
1710
1711 /* Deactivate the data zone */
1712 dmz_deactivate_zone(dzone);
1713 if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
1714 dmz_lru_zone(zmd, dzone);
1715 else {
1716 /* Unbuffered inactive empty data zone: reclaim it */
1717 dmz_unmap_zone(zmd, dzone);
1718 dmz_free_zone(zmd, dzone);
1719 }
1720
1721 dmz_unlock_map(zmd);
1722}
1723
1724/*
1725 * Allocate and map a random zone to buffer a chunk
1726 * already mapped to a sequential zone.
1727 */
1728struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
1729 struct dm_zone *dzone)
1730{
1731 struct dm_zone *bzone;
1732
1733 dmz_lock_map(zmd);
1734again:
1735 bzone = dzone->bzone;
1736 if (bzone)
1737 goto out;
1738
1739 /* Alloate a random zone */
1740 bzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1741 if (!bzone) {
1742 dmz_wait_for_free_zones(zmd);
1743 goto again;
1744 }
1745
1746 /* Update the chunk mapping */
1747 dmz_set_chunk_mapping(zmd, dzone->chunk, dmz_id(zmd, dzone),
1748 dmz_id(zmd, bzone));
1749
1750 set_bit(DMZ_BUF, &bzone->flags);
1751 bzone->chunk = dzone->chunk;
1752 bzone->bzone = dzone;
1753 dzone->bzone = bzone;
1754 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1755out:
1756 dmz_unlock_map(zmd);
1757
1758 return bzone;
1759}
1760
1761/*
1762 * Get an unmapped (free) zone.
1763 * This must be called with the mapping lock held.
1764 */
1765struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags)
1766{
1767 struct list_head *list;
1768 struct dm_zone *zone;
1769
1770 if (flags & DMZ_ALLOC_RND)
1771 list = &zmd->unmap_rnd_list;
1772 else
1773 list = &zmd->unmap_seq_list;
1774again:
1775 if (list_empty(list)) {
1776 /*
1777 * No free zone: if this is for reclaim, allow using the
1778 * reserved sequential zones.
1779 */
1780 if (!(flags & DMZ_ALLOC_RECLAIM) ||
1781 list_empty(&zmd->reserved_seq_zones_list))
1782 return NULL;
1783
1784 zone = list_first_entry(&zmd->reserved_seq_zones_list,
1785 struct dm_zone, link);
1786 list_del_init(&zone->link);
1787 atomic_dec(&zmd->nr_reserved_seq_zones);
1788 return zone;
1789 }
1790
1791 zone = list_first_entry(list, struct dm_zone, link);
1792 list_del_init(&zone->link);
1793
1794 if (dmz_is_rnd(zone))
1795 atomic_dec(&zmd->unmap_nr_rnd);
1796 else
1797 atomic_dec(&zmd->unmap_nr_seq);
1798
1799 if (dmz_is_offline(zone)) {
1800 dmz_dev_warn(zmd->dev, "Zone %u is offline", dmz_id(zmd, zone));
1801 zone = NULL;
1802 goto again;
1803 }
1804
1805 return zone;
1806}
1807
1808/*
1809 * Free a zone.
1810 * This must be called with the mapping lock held.
1811 */
1812void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1813{
1814 /* If this is a sequential zone, reset it */
1815 if (dmz_is_seq(zone))
1816 dmz_reset_zone(zmd, zone);
1817
1818 /* Return the zone to its type unmap list */
1819 if (dmz_is_rnd(zone)) {
1820 list_add_tail(&zone->link, &zmd->unmap_rnd_list);
1821 atomic_inc(&zmd->unmap_nr_rnd);
1822 } else if (atomic_read(&zmd->nr_reserved_seq_zones) <
1823 zmd->nr_reserved_seq) {
1824 list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
1825 atomic_inc(&zmd->nr_reserved_seq_zones);
1826 } else {
1827 list_add_tail(&zone->link, &zmd->unmap_seq_list);
1828 atomic_inc(&zmd->unmap_nr_seq);
1829 }
1830
1831 wake_up_all(&zmd->free_wq);
1832}
1833
1834/*
1835 * Map a chunk to a zone.
1836 * This must be called with the mapping lock held.
1837 */
1838void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
1839 unsigned int chunk)
1840{
1841 /* Set the chunk mapping */
1842 dmz_set_chunk_mapping(zmd, chunk, dmz_id(zmd, dzone),
1843 DMZ_MAP_UNMAPPED);
1844 dzone->chunk = chunk;
1845 if (dmz_is_rnd(dzone))
1846 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1847 else
1848 list_add_tail(&dzone->link, &zmd->map_seq_list);
1849}
1850
1851/*
1852 * Unmap a zone.
1853 * This must be called with the mapping lock held.
1854 */
1855void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1856{
1857 unsigned int chunk = zone->chunk;
1858 unsigned int dzone_id;
1859
1860 if (chunk == DMZ_MAP_UNMAPPED) {
1861 /* Already unmapped */
1862 return;
1863 }
1864
1865 if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
1866 /*
1867 * Unmapping the chunk buffer zone: clear only
1868 * the chunk buffer mapping
1869 */
1870 dzone_id = dmz_id(zmd, zone->bzone);
1871 zone->bzone->bzone = NULL;
1872 zone->bzone = NULL;
1873
1874 } else {
1875 /*
1876 * Unmapping the chunk data zone: the zone must
1877 * not be buffered.
1878 */
1879 if (WARN_ON(zone->bzone)) {
1880 zone->bzone->bzone = NULL;
1881 zone->bzone = NULL;
1882 }
1883 dzone_id = DMZ_MAP_UNMAPPED;
1884 }
1885
1886 dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
1887
1888 zone->chunk = DMZ_MAP_UNMAPPED;
1889 list_del_init(&zone->link);
1890}
1891
1892/*
1893 * Set @nr_bits bits in @bitmap starting from @bit.
1894 * Return the number of bits changed from 0 to 1.
1895 */
1896static unsigned int dmz_set_bits(unsigned long *bitmap,
1897 unsigned int bit, unsigned int nr_bits)
1898{
1899 unsigned long *addr;
1900 unsigned int end = bit + nr_bits;
1901 unsigned int n = 0;
1902
1903 while (bit < end) {
1904 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
1905 ((end - bit) >= BITS_PER_LONG)) {
1906 /* Try to set the whole word at once */
1907 addr = bitmap + BIT_WORD(bit);
1908 if (*addr == 0) {
1909 *addr = ULONG_MAX;
1910 n += BITS_PER_LONG;
1911 bit += BITS_PER_LONG;
1912 continue;
1913 }
1914 }
1915
1916 if (!test_and_set_bit(bit, bitmap))
1917 n++;
1918 bit++;
1919 }
1920
1921 return n;
1922}
1923
1924/*
1925 * Get the bitmap block storing the bit for chunk_block in zone.
1926 */
1927static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
1928 struct dm_zone *zone,
1929 sector_t chunk_block)
1930{
1931 sector_t bitmap_block = 1 + zmd->nr_map_blocks +
1932 (sector_t)(dmz_id(zmd, zone) * zmd->zone_nr_bitmap_blocks) +
1933 (chunk_block >> DMZ_BLOCK_SHIFT_BITS);
1934
1935 return dmz_get_mblock(zmd, bitmap_block);
1936}
1937
1938/*
1939 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
1940 */
1941int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1942 struct dm_zone *to_zone)
1943{
1944 struct dmz_mblock *from_mblk, *to_mblk;
1945 sector_t chunk_block = 0;
1946
1947 /* Get the zones bitmap blocks */
1948 while (chunk_block < zmd->dev->zone_nr_blocks) {
1949 from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
1950 if (IS_ERR(from_mblk))
1951 return PTR_ERR(from_mblk);
1952 to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
1953 if (IS_ERR(to_mblk)) {
1954 dmz_release_mblock(zmd, from_mblk);
1955 return PTR_ERR(to_mblk);
1956 }
1957
1958 memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
1959 dmz_dirty_mblock(zmd, to_mblk);
1960
1961 dmz_release_mblock(zmd, to_mblk);
1962 dmz_release_mblock(zmd, from_mblk);
1963
1964 chunk_block += DMZ_BLOCK_SIZE_BITS;
1965 }
1966
1967 to_zone->weight = from_zone->weight;
1968
1969 return 0;
1970}
1971
1972/*
1973 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
1974 * starting from chunk_block.
1975 */
1976int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1977 struct dm_zone *to_zone, sector_t chunk_block)
1978{
1979 unsigned int nr_blocks;
1980 int ret;
1981
1982 /* Get the zones bitmap blocks */
1983 while (chunk_block < zmd->dev->zone_nr_blocks) {
1984 /* Get a valid region from the source zone */
1985 ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
1986 if (ret <= 0)
1987 return ret;
1988
1989 nr_blocks = ret;
1990 ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
1991 if (ret)
1992 return ret;
1993
1994 chunk_block += nr_blocks;
1995 }
1996
1997 return 0;
1998}
1999
2000/*
2001 * Validate all the blocks in the range [block..block+nr_blocks-1].
2002 */
2003int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2004 sector_t chunk_block, unsigned int nr_blocks)
2005{
2006 unsigned int count, bit, nr_bits;
2007 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
2008 struct dmz_mblock *mblk;
2009 unsigned int n = 0;
2010
2011 dmz_dev_debug(zmd->dev, "=> VALIDATE zone %u, block %llu, %u blocks",
2012 dmz_id(zmd, zone), (unsigned long long)chunk_block,
2013 nr_blocks);
2014
2015 WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2016
2017 while (nr_blocks) {
2018 /* Get bitmap block */
2019 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2020 if (IS_ERR(mblk))
2021 return PTR_ERR(mblk);
2022
2023 /* Set bits */
2024 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2025 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2026
2027 count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2028 if (count) {
2029 dmz_dirty_mblock(zmd, mblk);
2030 n += count;
2031 }
2032 dmz_release_mblock(zmd, mblk);
2033
2034 nr_blocks -= nr_bits;
2035 chunk_block += nr_bits;
2036 }
2037
2038 if (likely(zone->weight + n <= zone_nr_blocks))
2039 zone->weight += n;
2040 else {
2041 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be <= %u",
2042 dmz_id(zmd, zone), zone->weight,
2043 zone_nr_blocks - n);
2044 zone->weight = zone_nr_blocks;
2045 }
2046
2047 return 0;
2048}
2049
2050/*
2051 * Clear nr_bits bits in bitmap starting from bit.
2052 * Return the number of bits cleared.
2053 */
2054static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2055{
2056 unsigned long *addr;
2057 int end = bit + nr_bits;
2058 int n = 0;
2059
2060 while (bit < end) {
2061 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2062 ((end - bit) >= BITS_PER_LONG)) {
2063 /* Try to clear whole word at once */
2064 addr = bitmap + BIT_WORD(bit);
2065 if (*addr == ULONG_MAX) {
2066 *addr = 0;
2067 n += BITS_PER_LONG;
2068 bit += BITS_PER_LONG;
2069 continue;
2070 }
2071 }
2072
2073 if (test_and_clear_bit(bit, bitmap))
2074 n++;
2075 bit++;
2076 }
2077
2078 return n;
2079}
2080
2081/*
2082 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2083 */
2084int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2085 sector_t chunk_block, unsigned int nr_blocks)
2086{
2087 unsigned int count, bit, nr_bits;
2088 struct dmz_mblock *mblk;
2089 unsigned int n = 0;
2090
2091 dmz_dev_debug(zmd->dev, "=> INVALIDATE zone %u, block %llu, %u blocks",
2092 dmz_id(zmd, zone), (u64)chunk_block, nr_blocks);
2093
2094 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2095
2096 while (nr_blocks) {
2097 /* Get bitmap block */
2098 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2099 if (IS_ERR(mblk))
2100 return PTR_ERR(mblk);
2101
2102 /* Clear bits */
2103 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2104 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2105
2106 count = dmz_clear_bits((unsigned long *)mblk->data,
2107 bit, nr_bits);
2108 if (count) {
2109 dmz_dirty_mblock(zmd, mblk);
2110 n += count;
2111 }
2112 dmz_release_mblock(zmd, mblk);
2113
2114 nr_blocks -= nr_bits;
2115 chunk_block += nr_bits;
2116 }
2117
2118 if (zone->weight >= n)
2119 zone->weight -= n;
2120 else {
2121 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be >= %u",
2122 dmz_id(zmd, zone), zone->weight, n);
2123 zone->weight = 0;
2124 }
2125
2126 return 0;
2127}
2128
2129/*
2130 * Get a block bit value.
2131 */
2132static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2133 sector_t chunk_block)
2134{
2135 struct dmz_mblock *mblk;
2136 int ret;
2137
2138 WARN_ON(chunk_block >= zmd->dev->zone_nr_blocks);
2139
2140 /* Get bitmap block */
2141 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2142 if (IS_ERR(mblk))
2143 return PTR_ERR(mblk);
2144
2145 /* Get offset */
2146 ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2147 (unsigned long *) mblk->data) != 0;
2148
2149 dmz_release_mblock(zmd, mblk);
2150
2151 return ret;
2152}
2153
2154/*
2155 * Return the number of blocks from chunk_block to the first block with a bit
2156 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2157 */
2158static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2159 sector_t chunk_block, unsigned int nr_blocks,
2160 int set)
2161{
2162 struct dmz_mblock *mblk;
2163 unsigned int bit, set_bit, nr_bits;
2164 unsigned long *bitmap;
2165 int n = 0;
2166
2167 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2168
2169 while (nr_blocks) {
2170 /* Get bitmap block */
2171 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2172 if (IS_ERR(mblk))
2173 return PTR_ERR(mblk);
2174
2175 /* Get offset */
2176 bitmap = (unsigned long *) mblk->data;
2177 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2178 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2179 if (set)
2180 set_bit = find_next_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2181 else
2182 set_bit = find_next_zero_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2183 dmz_release_mblock(zmd, mblk);
2184
2185 n += set_bit - bit;
2186 if (set_bit < DMZ_BLOCK_SIZE_BITS)
2187 break;
2188
2189 nr_blocks -= nr_bits;
2190 chunk_block += nr_bits;
2191 }
2192
2193 return n;
2194}
2195
2196/*
2197 * Test if chunk_block is valid. If it is, the number of consecutive
2198 * valid blocks from chunk_block will be returned.
2199 */
2200int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2201 sector_t chunk_block)
2202{
2203 int valid;
2204
2205 valid = dmz_test_block(zmd, zone, chunk_block);
2206 if (valid <= 0)
2207 return valid;
2208
2209 /* The block is valid: get the number of valid blocks from block */
2210 return dmz_to_next_set_block(zmd, zone, chunk_block,
2211 zmd->dev->zone_nr_blocks - chunk_block, 0);
2212}
2213
2214/*
2215 * Find the first valid block from @chunk_block in @zone.
2216 * If such a block is found, its number is returned using
2217 * @chunk_block and the total number of valid blocks from @chunk_block
2218 * is returned.
2219 */
2220int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2221 sector_t *chunk_block)
2222{
2223 sector_t start_block = *chunk_block;
2224 int ret;
2225
2226 ret = dmz_to_next_set_block(zmd, zone, start_block,
2227 zmd->dev->zone_nr_blocks - start_block, 1);
2228 if (ret < 0)
2229 return ret;
2230
2231 start_block += ret;
2232 *chunk_block = start_block;
2233
2234 return dmz_to_next_set_block(zmd, zone, start_block,
2235 zmd->dev->zone_nr_blocks - start_block, 0);
2236}
2237
2238/*
2239 * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2240 */
2241static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2242{
2243 unsigned long *addr;
2244 int end = bit + nr_bits;
2245 int n = 0;
2246
2247 while (bit < end) {
2248 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2249 ((end - bit) >= BITS_PER_LONG)) {
2250 addr = (unsigned long *)bitmap + BIT_WORD(bit);
2251 if (*addr == ULONG_MAX) {
2252 n += BITS_PER_LONG;
2253 bit += BITS_PER_LONG;
2254 continue;
2255 }
2256 }
2257
2258 if (test_bit(bit, bitmap))
2259 n++;
2260 bit++;
2261 }
2262
2263 return n;
2264}
2265
2266/*
2267 * Get a zone weight.
2268 */
2269static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2270{
2271 struct dmz_mblock *mblk;
2272 sector_t chunk_block = 0;
2273 unsigned int bit, nr_bits;
2274 unsigned int nr_blocks = zmd->dev->zone_nr_blocks;
2275 void *bitmap;
2276 int n = 0;
2277
2278 while (nr_blocks) {
2279 /* Get bitmap block */
2280 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2281 if (IS_ERR(mblk)) {
2282 n = 0;
2283 break;
2284 }
2285
2286 /* Count bits in this block */
2287 bitmap = mblk->data;
2288 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2289 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2290 n += dmz_count_bits(bitmap, bit, nr_bits);
2291
2292 dmz_release_mblock(zmd, mblk);
2293
2294 nr_blocks -= nr_bits;
2295 chunk_block += nr_bits;
2296 }
2297
2298 zone->weight = n;
2299}
2300
2301/*
2302 * Cleanup the zoned metadata resources.
2303 */
2304static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2305{
2306 struct rb_root *root;
2307 struct dmz_mblock *mblk, *next;
2308 int i;
2309
2310 /* Release zone mapping resources */
2311 if (zmd->map_mblk) {
2312 for (i = 0; i < zmd->nr_map_blocks; i++)
2313 dmz_release_mblock(zmd, zmd->map_mblk[i]);
2314 kfree(zmd->map_mblk);
2315 zmd->map_mblk = NULL;
2316 }
2317
2318 /* Release super blocks */
2319 for (i = 0; i < 2; i++) {
2320 if (zmd->sb[i].mblk) {
2321 dmz_free_mblock(zmd, zmd->sb[i].mblk);
2322 zmd->sb[i].mblk = NULL;
2323 }
2324 }
2325
2326 /* Free cached blocks */
2327 while (!list_empty(&zmd->mblk_dirty_list)) {
2328 mblk = list_first_entry(&zmd->mblk_dirty_list,
2329 struct dmz_mblock, link);
2330 dmz_dev_warn(zmd->dev, "mblock %llu still in dirty list (ref %u)",
2331 (u64)mblk->no, mblk->ref);
2332 list_del_init(&mblk->link);
2333 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2334 dmz_free_mblock(zmd, mblk);
2335 }
2336
2337 while (!list_empty(&zmd->mblk_lru_list)) {
2338 mblk = list_first_entry(&zmd->mblk_lru_list,
2339 struct dmz_mblock, link);
2340 list_del_init(&mblk->link);
2341 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2342 dmz_free_mblock(zmd, mblk);
2343 }
2344
2345 /* Sanity checks: the mblock rbtree should now be empty */
2346 root = &zmd->mblk_rbtree;
2347 rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2348 dmz_dev_warn(zmd->dev, "mblock %llu ref %u still in rbtree",
2349 (u64)mblk->no, mblk->ref);
2350 mblk->ref = 0;
2351 dmz_free_mblock(zmd, mblk);
2352 }
2353
2354 /* Free the zone descriptors */
2355 dmz_drop_zones(zmd);
2356
2357 mutex_destroy(&zmd->mblk_flush_lock);
2358 mutex_destroy(&zmd->map_lock);
2359}
2360
2361/*
2362 * Initialize the zoned metadata.
2363 */
2364int dmz_ctr_metadata(struct dmz_dev *dev, struct dmz_metadata **metadata)
2365{
2366 struct dmz_metadata *zmd;
2367 unsigned int i, zid;
2368 struct dm_zone *zone;
2369 int ret;
2370
2371 zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2372 if (!zmd)
2373 return -ENOMEM;
2374
2375 zmd->dev = dev;
2376 zmd->mblk_rbtree = RB_ROOT;
2377 init_rwsem(&zmd->mblk_sem);
2378 mutex_init(&zmd->mblk_flush_lock);
2379 spin_lock_init(&zmd->mblk_lock);
2380 INIT_LIST_HEAD(&zmd->mblk_lru_list);
2381 INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2382
2383 mutex_init(&zmd->map_lock);
2384 atomic_set(&zmd->unmap_nr_rnd, 0);
2385 INIT_LIST_HEAD(&zmd->unmap_rnd_list);
2386 INIT_LIST_HEAD(&zmd->map_rnd_list);
2387
2388 atomic_set(&zmd->unmap_nr_seq, 0);
2389 INIT_LIST_HEAD(&zmd->unmap_seq_list);
2390 INIT_LIST_HEAD(&zmd->map_seq_list);
2391
2392 atomic_set(&zmd->nr_reserved_seq_zones, 0);
2393 INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2394
2395 init_waitqueue_head(&zmd->free_wq);
2396
2397 /* Initialize zone descriptors */
2398 ret = dmz_init_zones(zmd);
2399 if (ret)
2400 goto err;
2401
2402 /* Get super block */
2403 ret = dmz_load_sb(zmd);
2404 if (ret)
2405 goto err;
2406
2407 /* Set metadata zones starting from sb_zone */
2408 zid = dmz_id(zmd, zmd->sb_zone);
2409 for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2410 zone = dmz_get(zmd, zid + i);
2411 if (!dmz_is_rnd(zone))
2412 goto err;
2413 set_bit(DMZ_META, &zone->flags);
2414 }
2415
2416 /* Load mapping table */
2417 ret = dmz_load_mapping(zmd);
2418 if (ret)
2419 goto err;
2420
2421 /*
2422 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2423 * blocks and enough blocks to be able to cache the bitmap blocks of
2424 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2425 * the cache to add 512 more metadata blocks.
2426 */
2427 zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2428 zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2429 zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2430 zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2431 zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2432
2433 /* Metadata cache shrinker */
2434 ret = register_shrinker(&zmd->mblk_shrinker);
2435 if (ret) {
2436 dmz_dev_err(dev, "Register metadata cache shrinker failed");
2437 goto err;
2438 }
2439
2440 dmz_dev_info(dev, "Host-%s zoned block device",
2441 bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2442 "aware" : "managed");
2443 dmz_dev_info(dev, " %llu 512-byte logical sectors",
2444 (u64)dev->capacity);
2445 dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors",
2446 dev->nr_zones, (u64)dev->zone_nr_sectors);
2447 dmz_dev_info(dev, " %u metadata zones",
2448 zmd->nr_meta_zones * 2);
2449 dmz_dev_info(dev, " %u data zones for %u chunks",
2450 zmd->nr_data_zones, zmd->nr_chunks);
2451 dmz_dev_info(dev, " %u random zones (%u unmapped)",
2452 zmd->nr_rnd, atomic_read(&zmd->unmap_nr_rnd));
2453 dmz_dev_info(dev, " %u sequential zones (%u unmapped)",
2454 zmd->nr_seq, atomic_read(&zmd->unmap_nr_seq));
2455 dmz_dev_info(dev, " %u reserved sequential data zones",
2456 zmd->nr_reserved_seq);
2457
2458 dmz_dev_debug(dev, "Format:");
2459 dmz_dev_debug(dev, "%u metadata blocks per set (%u max cache)",
2460 zmd->nr_meta_blocks, zmd->max_nr_mblks);
2461 dmz_dev_debug(dev, " %u data zone mapping blocks",
2462 zmd->nr_map_blocks);
2463 dmz_dev_debug(dev, " %u bitmap blocks",
2464 zmd->nr_bitmap_blocks);
2465
2466 *metadata = zmd;
2467
2468 return 0;
2469err:
2470 dmz_cleanup_metadata(zmd);
2471 kfree(zmd);
2472 *metadata = NULL;
2473
2474 return ret;
2475}
2476
2477/*
2478 * Cleanup the zoned metadata resources.
2479 */
2480void dmz_dtr_metadata(struct dmz_metadata *zmd)
2481{
2482 unregister_shrinker(&zmd->mblk_shrinker);
2483 dmz_cleanup_metadata(zmd);
2484 kfree(zmd);
2485}
2486
2487/*
2488 * Check zone information on resume.
2489 */
2490int dmz_resume_metadata(struct dmz_metadata *zmd)
2491{
2492 struct dmz_dev *dev = zmd->dev;
2493 struct dm_zone *zone;
2494 sector_t wp_block;
2495 unsigned int i;
2496 int ret;
2497
2498 /* Check zones */
2499 for (i = 0; i < dev->nr_zones; i++) {
2500 zone = dmz_get(zmd, i);
2501 if (!zone) {
2502 dmz_dev_err(dev, "Unable to get zone %u", i);
2503 return -EIO;
2504 }
2505
2506 wp_block = zone->wp_block;
2507
2508 ret = dmz_update_zone(zmd, zone);
2509 if (ret) {
2510 dmz_dev_err(dev, "Broken zone %u", i);
2511 return ret;
2512 }
2513
2514 if (dmz_is_offline(zone)) {
2515 dmz_dev_warn(dev, "Zone %u is offline", i);
2516 continue;
2517 }
2518
2519 /* Check write pointer */
2520 if (!dmz_is_seq(zone))
2521 zone->wp_block = 0;
2522 else if (zone->wp_block != wp_block) {
2523 dmz_dev_err(dev, "Zone %u: Invalid wp (%llu / %llu)",
2524 i, (u64)zone->wp_block, (u64)wp_block);
2525 zone->wp_block = wp_block;
2526 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
2527 dev->zone_nr_blocks - zone->wp_block);
2528 }
2529 }
2530
2531 return 0;
2532}
2533