projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
dm zoned: add 'message' callback
[linux-block.git] / drivers / md / dm-zoned-metadata.c
CommitLineData
bae9a0aa 1// SPDX-License-Identifier: GPL-2.0-only
3b1a94c8
DLM		 2/*
3 * Copyright (C) 2017 Western Digital Corporation or its affiliates.
4 *
5 * This file is released under the GPL.
6 */
7
8#include "dm-zoned.h"
9
10#include <linux/module.h>
11#include <linux/crc32.h>
bd976e52 12#include <linux/sched/mm.h>
3b1a94c8
DLM		 13
14#define DM_MSG_PREFIX "zoned metadata"
15
16/*
17 * Metadata version.
18 */
19#define DMZ_META_VER 1
20
21/*
22 * On-disk super block magic.
23 */
24#define DMZ_MAGIC ((((unsigned int)('D')) << 24) | \
25 (((unsigned int)('Z')) << 16) | \
26 (((unsigned int)('B')) << 8) | \
27 ((unsigned int)('D')))
28
29/*
30 * On disk super block.
31 * This uses only 512 B but uses on disk a full 4KB block. This block is
32 * followed on disk by the mapping table of chunks to zones and the bitmap
33 * blocks indicating zone block validity.
34 * The overall resulting metadata format is:
35 * (1) Super block (1 block)
36 * (2) Chunk mapping table (nr_map_blocks)
37 * (3) Bitmap blocks (nr_bitmap_blocks)
ad1bd578 38 * All metadata blocks are stored in conventional zones, starting from
3b1a94c8
DLM		 39 * the first conventional zone found on disk.
40 */
41struct dmz_super {
42 /* Magic number */
43 __le32 magic; /* 4 */
44
45 /* Metadata version number */
46 __le32 version; /* 8 */
47
48 /* Generation number */
49 __le64 gen; /* 16 */
50
51 /* This block number */
52 __le64 sb_block; /* 24 */
53
54 /* The number of metadata blocks, including this super block */
55 __le32 nr_meta_blocks; /* 28 */
56
57 /* The number of sequential zones reserved for reclaim */
58 __le32 nr_reserved_seq; /* 32 */
59
60 /* The number of entries in the mapping table */
61 __le32 nr_chunks; /* 36 */
62
63 /* The number of blocks used for the chunk mapping table */
64 __le32 nr_map_blocks; /* 40 */
65
66 /* The number of blocks used for the block bitmaps */
67 __le32 nr_bitmap_blocks; /* 44 */
68
69 /* Checksum */
70 __le32 crc; /* 48 */
71
72 /* Padding to full 512B sector */
73 u8 reserved[464]; /* 512 */
74};
75
76/*
77 * Chunk mapping entry: entries are indexed by chunk number
78 * and give the zone ID (dzone_id) mapping the chunk on disk.
79 * This zone may be sequential or random. If it is a sequential
80 * zone, a second zone (bzone_id) used as a write buffer may
81 * also be specified. This second zone will always be a randomly
82 * writeable zone.
83 */
84struct dmz_map {
85 __le32 dzone_id;
86 __le32 bzone_id;
87};
88
89/*
90 * Chunk mapping table metadata: 512 8-bytes entries per 4KB block.
91 */
92#define DMZ_MAP_ENTRIES (DMZ_BLOCK_SIZE / sizeof(struct dmz_map))
93#define DMZ_MAP_ENTRIES_SHIFT (ilog2(DMZ_MAP_ENTRIES))
94#define DMZ_MAP_ENTRIES_MASK (DMZ_MAP_ENTRIES - 1)
95#define DMZ_MAP_UNMAPPED UINT_MAX
96
97/*
98 * Meta data block descriptor (for cached metadata blocks).
99 */
100struct dmz_mblock {
101 struct rb_node node;
102 struct list_head link;
103 sector_t no;
33c2865f 104 unsigned int ref;
3b1a94c8
DLM		 105 unsigned long state;
106 struct page *page;
107 void *data;
108};
109
110/*
111 * Metadata block state flags.
112 */
113enum {
114 DMZ_META_DIRTY,
115 DMZ_META_READING,
116 DMZ_META_WRITING,
117 DMZ_META_ERROR,
118};
119
120/*
121 * Super block information (one per metadata set).
122 */
123struct dmz_sb {
124 sector_t block;
125 struct dmz_mblock *mblk;
126 struct dmz_super *sb;
127};
128
129/*
130 * In-memory metadata.
131 */
132struct dmz_metadata {
133 struct dmz_dev *dev;
134
135 sector_t zone_bitmap_size;
136 unsigned int zone_nr_bitmap_blocks;
b3996295 137 unsigned int zone_bits_per_mblk;
3b1a94c8
DLM		 138
139 unsigned int nr_bitmap_blocks;
140 unsigned int nr_map_blocks;
141
142 unsigned int nr_useable_zones;
143 unsigned int nr_meta_blocks;
144 unsigned int nr_meta_zones;
145 unsigned int nr_data_zones;
146 unsigned int nr_rnd_zones;
147 unsigned int nr_reserved_seq;
148 unsigned int nr_chunks;
149
150 /* Zone information array */
151 struct dm_zone *zones;
152
153 struct dm_zone *sb_zone;
154 struct dmz_sb sb[2];
155 unsigned int mblk_primary;
156 u64 sb_gen;
157 unsigned int min_nr_mblks;
158 unsigned int max_nr_mblks;
159 atomic_t nr_mblks;
160 struct rw_semaphore mblk_sem;
161 struct mutex mblk_flush_lock;
162 spinlock_t mblk_lock;
163 struct rb_root mblk_rbtree;
164 struct list_head mblk_lru_list;
165 struct list_head mblk_dirty_list;
166 struct shrinker mblk_shrinker;
167
168 /* Zone allocation management */
169 struct mutex map_lock;
170 struct dmz_mblock **map_mblk;
171 unsigned int nr_rnd;
172 atomic_t unmap_nr_rnd;
173 struct list_head unmap_rnd_list;
174 struct list_head map_rnd_list;
175
176 unsigned int nr_seq;
177 atomic_t unmap_nr_seq;
178 struct list_head unmap_seq_list;
179 struct list_head map_seq_list;
180
181 atomic_t nr_reserved_seq_zones;
182 struct list_head reserved_seq_zones_list;
183
184 wait_queue_head_t free_wq;
185};
186
187/*
188 * Various accessors
189 */
190unsigned int dmz_id(struct dmz_metadata *zmd, struct dm_zone *zone)
191{
192 return ((unsigned int)(zone - zmd->zones));
193}
194
195sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone)
196{
3908c983 197 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_sectors_shift;
3b1a94c8
DLM		 198}
199
200sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone)
201{
3908c983 202 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_blocks_shift;
3b1a94c8
DLM		 203}
204
bc3d5717
HR		 205unsigned int dmz_nr_zones(struct dmz_metadata *zmd)
206{
207 return zmd->dev->nr_zones;
208}
209
3b1a94c8
DLM		 210unsigned int dmz_nr_chunks(struct dmz_metadata *zmd)
211{
212 return zmd->nr_chunks;
213}
214
215unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd)
216{
217 return zmd->nr_rnd;
218}
219
220unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd)
221{
222 return atomic_read(&zmd->unmap_nr_rnd);
223}
224
bc3d5717
HR		 225unsigned int dmz_nr_seq_zones(struct dmz_metadata *zmd)
226{
227 return zmd->nr_seq;
228}
229
230unsigned int dmz_nr_unmap_seq_zones(struct dmz_metadata *zmd)
231{
232 return atomic_read(&zmd->unmap_nr_seq);
233}
234
3b1a94c8
DLM		 235/*
236 * Lock/unlock mapping table.
237 * The map lock also protects all the zone lists.
238 */
239void dmz_lock_map(struct dmz_metadata *zmd)
240{
241 mutex_lock(&zmd->map_lock);
242}
243
244void dmz_unlock_map(struct dmz_metadata *zmd)
245{
246 mutex_unlock(&zmd->map_lock);
247}
248
249/*
250 * Lock/unlock metadata access. This is a "read" lock on a semaphore
251 * that prevents metadata flush from running while metadata are being
252 * modified. The actual metadata write mutual exclusion is achieved with
ad1bd578 253 * the map lock and zone state management (active and reclaim state are
3b1a94c8
DLM		 254 * mutually exclusive).
255 */
256void dmz_lock_metadata(struct dmz_metadata *zmd)
257{
258 down_read(&zmd->mblk_sem);
259}
260
261void dmz_unlock_metadata(struct dmz_metadata *zmd)
262{
263 up_read(&zmd->mblk_sem);
264}
265
266/*
267 * Lock/unlock flush: prevent concurrent executions
268 * of dmz_flush_metadata as well as metadata modification in reclaim
269 * while flush is being executed.
270 */
271void dmz_lock_flush(struct dmz_metadata *zmd)
272{
273 mutex_lock(&zmd->mblk_flush_lock);
274}
275
276void dmz_unlock_flush(struct dmz_metadata *zmd)
277{
278 mutex_unlock(&zmd->mblk_flush_lock);
279}
280
281/*
282 * Allocate a metadata block.
283 */
284static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd,
285 sector_t mblk_no)
286{
287 struct dmz_mblock *mblk = NULL;
288
289 /* See if we can reuse cached blocks */
290 if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) {
291 spin_lock(&zmd->mblk_lock);
292 mblk = list_first_entry_or_null(&zmd->mblk_lru_list,
293 struct dmz_mblock, link);
294 if (mblk) {
295 list_del_init(&mblk->link);
296 rb_erase(&mblk->node, &zmd->mblk_rbtree);
297 mblk->no = mblk_no;
298 }
299 spin_unlock(&zmd->mblk_lock);
300 if (mblk)
301 return mblk;
302 }
303
304 /* Allocate a new block */
305 mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO);
306 if (!mblk)
307 return NULL;
308
309 mblk->page = alloc_page(GFP_NOIO);
310 if (!mblk->page) {
311 kfree(mblk);
312 return NULL;
313 }
314
315 RB_CLEAR_NODE(&mblk->node);
316 INIT_LIST_HEAD(&mblk->link);
33c2865f 317 mblk->ref = 0;
3b1a94c8
DLM		 318 mblk->state = 0;
319 mblk->no = mblk_no;
320 mblk->data = page_address(mblk->page);
321
322 atomic_inc(&zmd->nr_mblks);
323
324 return mblk;
325}
326
327/*
328 * Free a metadata block.
329 */
330static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
331{
332 __free_pages(mblk->page, 0);
333 kfree(mblk);
334
335 atomic_dec(&zmd->nr_mblks);
336}
337
338/*
339 * Insert a metadata block in the rbtree.
340 */
341static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
342{
343 struct rb_root *root = &zmd->mblk_rbtree;
344 struct rb_node **new = &(root->rb_node), *parent = NULL;
345 struct dmz_mblock *b;
346
347 /* Figure out where to put the new node */
348 while (*new) {
349 b = container_of(*new, struct dmz_mblock, node);
350 parent = *new;
351 new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right);
352 }
353
354 /* Add new node and rebalance tree */
355 rb_link_node(&mblk->node, parent, new);
356 rb_insert_color(&mblk->node, root);
357}
358
359/*
3d4e7383
DLM		 360 * Lookup a metadata block in the rbtree. If the block is found, increment
361 * its reference count.
3b1a94c8 362 */
3d4e7383
DLM		 363static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd,
364 sector_t mblk_no)
3b1a94c8
DLM		 365{
366 struct rb_root *root = &zmd->mblk_rbtree;
367 struct rb_node *node = root->rb_node;
368 struct dmz_mblock *mblk;
369
370 while (node) {
371 mblk = container_of(node, struct dmz_mblock, node);
3d4e7383
DLM		 372 if (mblk->no == mblk_no) {
373 /*
374 * If this is the first reference to the block,
375 * remove it from the LRU list.
376 */
377 mblk->ref++;
378 if (mblk->ref == 1 &&
379 !test_bit(DMZ_META_DIRTY, &mblk->state))
380 list_del_init(&mblk->link);
3b1a94c8 381 return mblk;
3d4e7383 382 }
3b1a94c8
DLM		 383 node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right;
384 }
385
386 return NULL;
387}
388
389/*
390 * Metadata block BIO end callback.
391 */
392static void dmz_mblock_bio_end_io(struct bio *bio)
393{
394 struct dmz_mblock *mblk = bio->bi_private;
395 int flag;
396
397 if (bio->bi_status)
398 set_bit(DMZ_META_ERROR, &mblk->state);
399
400 if (bio_op(bio) == REQ_OP_WRITE)
401 flag = DMZ_META_WRITING;
402 else
403 flag = DMZ_META_READING;
404
405 clear_bit_unlock(flag, &mblk->state);
406 smp_mb__after_atomic();
407 wake_up_bit(&mblk->state, flag);
408
409 bio_put(bio);
410}
411
412/*
3d4e7383 413 * Read an uncached metadata block from disk and add it to the cache.
3b1a94c8 414 */
3d4e7383
DLM		 415static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd,
416 sector_t mblk_no)
3b1a94c8 417{
3d4e7383 418 struct dmz_mblock *mblk, *m;
3b1a94c8
DLM		 419 sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no;
420 struct bio *bio;
421
75d66ffb
DF		 422 if (dmz_bdev_is_dying(zmd->dev))
423 return ERR_PTR(-EIO);
424
3d4e7383 425 /* Get a new block and a BIO to read it */
3b1a94c8
DLM		 426 mblk = dmz_alloc_mblock(zmd, mblk_no);
427 if (!mblk)
75d66ffb 428 return ERR_PTR(-ENOMEM);
3b1a94c8 429
3b1a94c8
DLM		 430 bio = bio_alloc(GFP_NOIO, 1);
431 if (!bio) {
432 dmz_free_mblock(zmd, mblk);
75d66ffb 433 return ERR_PTR(-ENOMEM);
3b1a94c8
DLM		 434 }
435
3d4e7383
DLM		 436 spin_lock(&zmd->mblk_lock);
437
438 /*
439 * Make sure that another context did not start reading
440 * the block already.
441 */
442 m = dmz_get_mblock_fast(zmd, mblk_no);
443 if (m) {
444 spin_unlock(&zmd->mblk_lock);
445 dmz_free_mblock(zmd, mblk);
446 bio_put(bio);
447 return m;
448 }
449
450 mblk->ref++;
451 set_bit(DMZ_META_READING, &mblk->state);
452 dmz_insert_mblock(zmd, mblk);
453
454 spin_unlock(&zmd->mblk_lock);
455
456 /* Submit read BIO */
3b1a94c8 457 bio->bi_iter.bi_sector = dmz_blk2sect(block);
74d46992 458 bio_set_dev(bio, zmd->dev->bdev);
3b1a94c8
DLM		 459 bio->bi_private = mblk;
460 bio->bi_end_io = dmz_mblock_bio_end_io;
461 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO);
462 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
463 submit_bio(bio);
464
465 return mblk;
466}
467
468/*
469 * Free metadata blocks.
470 */
471static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd,
472 unsigned long limit)
473{
474 struct dmz_mblock *mblk;
475 unsigned long count = 0;
476
477 if (!zmd->max_nr_mblks)
478 return 0;
479
480 while (!list_empty(&zmd->mblk_lru_list) &&
481 atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks &&
482 count < limit) {
483 mblk = list_first_entry(&zmd->mblk_lru_list,
484 struct dmz_mblock, link);
485 list_del_init(&mblk->link);
486 rb_erase(&mblk->node, &zmd->mblk_rbtree);
487 dmz_free_mblock(zmd, mblk);
488 count++;
489 }
490
491 return count;
492}
493
494/*
495 * For mblock shrinker: get the number of unused metadata blocks in the cache.
496 */
497static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink,
498 struct shrink_control *sc)
499{
500 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
501
502 return atomic_read(&zmd->nr_mblks);
503}
504
505/*
506 * For mblock shrinker: scan unused metadata blocks and shrink the cache.
507 */
508static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink,
509 struct shrink_control *sc)
510{
511 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
512 unsigned long count;
513
514 spin_lock(&zmd->mblk_lock);
515 count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan);
516 spin_unlock(&zmd->mblk_lock);
517
518 return count ? count : SHRINK_STOP;
519}
520
521/*
522 * Release a metadata block.
523 */
524static void dmz_release_mblock(struct dmz_metadata *zmd,
525 struct dmz_mblock *mblk)
526{
527
528 if (!mblk)
529 return;
530
531 spin_lock(&zmd->mblk_lock);
532
33c2865f
DLM		 533 mblk->ref--;
534 if (mblk->ref == 0) {
3b1a94c8
DLM		 535 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
536 rb_erase(&mblk->node, &zmd->mblk_rbtree);
537 dmz_free_mblock(zmd, mblk);
538 } else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) {
539 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
540 dmz_shrink_mblock_cache(zmd, 1);
541 }
542 }
543
544 spin_unlock(&zmd->mblk_lock);
545}
546
547/*
548 * Get a metadata block from the rbtree. If the block
549 * is not present, read it from disk.
550 */
551static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
552 sector_t mblk_no)
553{
554 struct dmz_mblock *mblk;
555
556 /* Check rbtree */
557 spin_lock(&zmd->mblk_lock);
3d4e7383 558 mblk = dmz_get_mblock_fast(zmd, mblk_no);
3b1a94c8
DLM		 559 spin_unlock(&zmd->mblk_lock);
560
561 if (!mblk) {
562 /* Cache miss: read the block from disk */
3d4e7383 563 mblk = dmz_get_mblock_slow(zmd, mblk_no);
75d66ffb
DF		 564 if (IS_ERR(mblk))
565 return mblk;
3b1a94c8
DLM		 566 }
567
568 /* Wait for on-going read I/O and check for error */
569 wait_on_bit_io(&mblk->state, DMZ_META_READING,
570 TASK_UNINTERRUPTIBLE);
571 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
572 dmz_release_mblock(zmd, mblk);
e7fad909 573 dmz_check_bdev(zmd->dev);
3b1a94c8
DLM		 574 return ERR_PTR(-EIO);
575 }
576
577 return mblk;
578}
579
580/*
581 * Mark a metadata block dirty.
582 */
583static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
584{
585 spin_lock(&zmd->mblk_lock);
586 if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state))
587 list_add_tail(&mblk->link, &zmd->mblk_dirty_list);
588 spin_unlock(&zmd->mblk_lock);
589}
590
591/*
592 * Issue a metadata block write BIO.
593 */
75d66ffb
DF		 594static int dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk,
595 unsigned int set)
3b1a94c8
DLM		 596{
597 sector_t block = zmd->sb[set].block + mblk->no;
598 struct bio *bio;
599
75d66ffb
DF		 600 if (dmz_bdev_is_dying(zmd->dev))
601 return -EIO;
602
3b1a94c8
DLM		 603 bio = bio_alloc(GFP_NOIO, 1);
604 if (!bio) {
605 set_bit(DMZ_META_ERROR, &mblk->state);
75d66ffb 606 return -ENOMEM;
3b1a94c8
DLM		 607 }
608
609 set_bit(DMZ_META_WRITING, &mblk->state);
610
611 bio->bi_iter.bi_sector = dmz_blk2sect(block);
74d46992 612 bio_set_dev(bio, zmd->dev->bdev);
3b1a94c8
DLM		 613 bio->bi_private = mblk;
614 bio->bi_end_io = dmz_mblock_bio_end_io;
615 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO);
616 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
617 submit_bio(bio);
75d66ffb
DF		 618
619 return 0;
3b1a94c8
DLM		 620}
621
622/*
623 * Read/write a metadata block.
624 */
625static int dmz_rdwr_block(struct dmz_metadata *zmd, int op, sector_t block,
626 struct page *page)
627{
628 struct bio *bio;
629 int ret;
630
75d66ffb
DF		 631 if (dmz_bdev_is_dying(zmd->dev))
632 return -EIO;
633
3b1a94c8
DLM		 634 bio = bio_alloc(GFP_NOIO, 1);
635 if (!bio)
636 return -ENOMEM;
637
638 bio->bi_iter.bi_sector = dmz_blk2sect(block);
74d46992 639 bio_set_dev(bio, zmd->dev->bdev);
3b1a94c8
DLM		 640 bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO);
641 bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0);
642 ret = submit_bio_wait(bio);
643 bio_put(bio);
644
e7fad909
DF		 645 if (ret)
646 dmz_check_bdev(zmd->dev);
3b1a94c8
DLM		 647 return ret;
648}
649
650/*
651 * Write super block of the specified metadata set.
652 */
653static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set)
654{
655 sector_t block = zmd->sb[set].block;
656 struct dmz_mblock *mblk = zmd->sb[set].mblk;
657 struct dmz_super *sb = zmd->sb[set].sb;
658 u64 sb_gen = zmd->sb_gen + 1;
659 int ret;
660
661 sb->magic = cpu_to_le32(DMZ_MAGIC);
662 sb->version = cpu_to_le32(DMZ_META_VER);
663
664 sb->gen = cpu_to_le64(sb_gen);
665
666 sb->sb_block = cpu_to_le64(block);
667 sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks);
668 sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq);
669 sb->nr_chunks = cpu_to_le32(zmd->nr_chunks);
670
671 sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks);
672 sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks);
673
674 sb->crc = 0;
675 sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE));
676
677 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE, block, mblk->page);
678 if (ret == 0)
4218a955 679 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
3b1a94c8
DLM		 680
681 return ret;
682}
683
684/*
685 * Write dirty metadata blocks to the specified set.
686 */
687static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
688 struct list_head *write_list,
689 unsigned int set)
690{
691 struct dmz_mblock *mblk;
692 struct blk_plug plug;
75d66ffb 693 int ret = 0, nr_mblks_submitted = 0;
3b1a94c8
DLM		 694
695 /* Issue writes */
696 blk_start_plug(&plug);
75d66ffb
DF		 697 list_for_each_entry(mblk, write_list, link) {
698 ret = dmz_write_mblock(zmd, mblk, set);
699 if (ret)
700 break;
701 nr_mblks_submitted++;
702 }
3b1a94c8
DLM		 703 blk_finish_plug(&plug);
704
705 /* Wait for completion */
706 list_for_each_entry(mblk, write_list, link) {
75d66ffb
DF		 707 if (!nr_mblks_submitted)
708 break;
3b1a94c8
DLM		 709 wait_on_bit_io(&mblk->state, DMZ_META_WRITING,
710 TASK_UNINTERRUPTIBLE);
711 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
712 clear_bit(DMZ_META_ERROR, &mblk->state);
e7fad909 713 dmz_check_bdev(zmd->dev);
3b1a94c8
DLM		 714 ret = -EIO;
715 }
75d66ffb 716 nr_mblks_submitted--;
3b1a94c8
DLM		 717 }
718
719 /* Flush drive cache (this will also sync data) */
720 if (ret == 0)
4218a955 721 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
3b1a94c8
DLM		 722
723 return ret;
724}
725
726/*
727 * Log dirty metadata blocks.
728 */
729static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd,
730 struct list_head *write_list)
731{
732 unsigned int log_set = zmd->mblk_primary ^ 0x1;
733 int ret;
734
735 /* Write dirty blocks to the log */
736 ret = dmz_write_dirty_mblocks(zmd, write_list, log_set);
737 if (ret)
738 return ret;
739
740 /*
741 * No error so far: now validate the log by updating the
742 * log index super block generation.
743 */
744 ret = dmz_write_sb(zmd, log_set);
745 if (ret)
746 return ret;
747
748 return 0;
749}
750
751/*
752 * Flush dirty metadata blocks.
753 */
754int dmz_flush_metadata(struct dmz_metadata *zmd)
755{
756 struct dmz_mblock *mblk;
757 struct list_head write_list;
758 int ret;
759
760 if (WARN_ON(!zmd))
761 return 0;
762
763 INIT_LIST_HEAD(&write_list);
764
765 /*
766 * Make sure that metadata blocks are stable before logging: take
767 * the write lock on the metadata semaphore to prevent target BIOs
768 * from modifying metadata.
769 */
770 down_write(&zmd->mblk_sem);
771
772 /*
773 * This is called from the target flush work and reclaim work.
774 * Concurrent execution is not allowed.
775 */
776 dmz_lock_flush(zmd);
777
75d66ffb
DF		 778 if (dmz_bdev_is_dying(zmd->dev)) {
779 ret = -EIO;
780 goto out;
781 }
782
3b1a94c8
DLM		 783 /* Get dirty blocks */
784 spin_lock(&zmd->mblk_lock);
785 list_splice_init(&zmd->mblk_dirty_list, &write_list);
786 spin_unlock(&zmd->mblk_lock);
787
788 /* If there are no dirty metadata blocks, just flush the device cache */
789 if (list_empty(&write_list)) {
4218a955 790 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
e7fad909 791 goto err;
3b1a94c8
DLM		 792 }
793
794 /*
795 * The primary metadata set is still clean. Keep it this way until
796 * all updates are successful in the secondary set. That is, use
797 * the secondary set as a log.
798 */
799 ret = dmz_log_dirty_mblocks(zmd, &write_list);
800 if (ret)
e7fad909 801 goto err;
3b1a94c8
DLM		 802
803 /*
804 * The log is on disk. It is now safe to update in place
805 * in the primary metadata set.
806 */
807 ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
808 if (ret)
e7fad909 809 goto err;
3b1a94c8
DLM		 810
811 ret = dmz_write_sb(zmd, zmd->mblk_primary);
812 if (ret)
e7fad909 813 goto err;
3b1a94c8
DLM		 814
815 while (!list_empty(&write_list)) {
816 mblk = list_first_entry(&write_list, struct dmz_mblock, link);
817 list_del_init(&mblk->link);
818
819 spin_lock(&zmd->mblk_lock);
820 clear_bit(DMZ_META_DIRTY, &mblk->state);
33c2865f 821 if (mblk->ref == 0)
3b1a94c8
DLM		 822 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
823 spin_unlock(&zmd->mblk_lock);
824 }
825
826 zmd->sb_gen++;
827out:
3b1a94c8
DLM		 828 dmz_unlock_flush(zmd);
829 up_write(&zmd->mblk_sem);
830
831 return ret;
e7fad909
DF		 832
833err:
834 if (!list_empty(&write_list)) {
835 spin_lock(&zmd->mblk_lock);
836 list_splice(&write_list, &zmd->mblk_dirty_list);
837 spin_unlock(&zmd->mblk_lock);
838 }
839 if (!dmz_check_bdev(zmd->dev))
840 ret = -EIO;
841 goto out;
3b1a94c8
DLM		 842}
843
844/*
845 * Check super block.
846 */
847static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_super *sb)
848{
849 unsigned int nr_meta_zones, nr_data_zones;
850 struct dmz_dev *dev = zmd->dev;
851 u32 crc, stored_crc;
852 u64 gen;
853
854 gen = le64_to_cpu(sb->gen);
855 stored_crc = le32_to_cpu(sb->crc);
856 sb->crc = 0;
857 crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
858 if (crc != stored_crc) {
859 dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
860 crc, stored_crc);
861 return -ENXIO;
862 }
863
864 if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
865 dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
866 DMZ_MAGIC, le32_to_cpu(sb->magic));
867 return -ENXIO;
868 }
869
870 if (le32_to_cpu(sb->version) != DMZ_META_VER) {
871 dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
872 DMZ_META_VER, le32_to_cpu(sb->version));
873 return -ENXIO;
874 }
875
876 nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + dev->zone_nr_blocks - 1)
877 >> dev->zone_nr_blocks_shift;
878 if (!nr_meta_zones ||
879 nr_meta_zones >= zmd->nr_rnd_zones) {
880 dmz_dev_err(dev, "Invalid number of metadata blocks");
881 return -ENXIO;
882 }
883
884 if (!le32_to_cpu(sb->nr_reserved_seq) ||
885 le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
886 dmz_dev_err(dev, "Invalid number of reserved sequential zones");
887 return -ENXIO;
888 }
889
890 nr_data_zones = zmd->nr_useable_zones -
891 (nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
892 if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
893 dmz_dev_err(dev, "Invalid number of chunks %u / %u",
894 le32_to_cpu(sb->nr_chunks), nr_data_zones);
895 return -ENXIO;
896 }
897
898 /* OK */
899 zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
900 zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
901 zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
902 zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
903 zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
904 zmd->nr_meta_zones = nr_meta_zones;
905 zmd->nr_data_zones = nr_data_zones;
906
907 return 0;
908}
909
910/*
911 * Read the first or second super block from disk.
912 */
913static int dmz_read_sb(struct dmz_metadata *zmd, unsigned int set)
914{
915 return dmz_rdwr_block(zmd, REQ_OP_READ, zmd->sb[set].block,
916 zmd->sb[set].mblk->page);
917}
918
919/*
920 * Determine the position of the secondary super blocks on disk.
921 * This is used only if a corruption of the primary super block
922 * is detected.
923 */
924static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
925{
926 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
927 struct dmz_mblock *mblk;
928 int i;
929
930 /* Allocate a block */
931 mblk = dmz_alloc_mblock(zmd, 0);
932 if (!mblk)
933 return -ENOMEM;
934
935 zmd->sb[1].mblk = mblk;
936 zmd->sb[1].sb = mblk->data;
937
938 /* Bad first super block: search for the second one */
939 zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
940 for (i = 0; i < zmd->nr_rnd_zones - 1; i++) {
941 if (dmz_read_sb(zmd, 1) != 0)
942 break;
943 if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
944 return 0;
945 zmd->sb[1].block += zone_nr_blocks;
946 }
947
948 dmz_free_mblock(zmd, mblk);
949 zmd->sb[1].mblk = NULL;
950
951 return -EIO;
952}
953
954/*
955 * Read the first or second super block from disk.
956 */
957static int dmz_get_sb(struct dmz_metadata *zmd, unsigned int set)
958{
959 struct dmz_mblock *mblk;
960 int ret;
961
962 /* Allocate a block */
963 mblk = dmz_alloc_mblock(zmd, 0);
964 if (!mblk)
965 return -ENOMEM;
966
967 zmd->sb[set].mblk = mblk;
968 zmd->sb[set].sb = mblk->data;
969
970 /* Read super block */
971 ret = dmz_read_sb(zmd, set);
972 if (ret) {
973 dmz_free_mblock(zmd, mblk);
974 zmd->sb[set].mblk = NULL;
975 return ret;
976 }
977
978 return 0;
979}
980
981/*
982 * Recover a metadata set.
983 */
984static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
985{
986 unsigned int src_set = dst_set ^ 0x1;
987 struct page *page;
988 int i, ret;
989
990 dmz_dev_warn(zmd->dev, "Metadata set %u invalid: recovering", dst_set);
991
992 if (dst_set == 0)
993 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
994 else {
995 zmd->sb[1].block = zmd->sb[0].block +
996 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
997 }
998
4218a955 999 page = alloc_page(GFP_NOIO);
3b1a94c8
DLM		 1000 if (!page)
1001 return -ENOMEM;
1002
1003 /* Copy metadata blocks */
1004 for (i = 1; i < zmd->nr_meta_blocks; i++) {
1005 ret = dmz_rdwr_block(zmd, REQ_OP_READ,
1006 zmd->sb[src_set].block + i, page);
1007 if (ret)
1008 goto out;
1009 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE,
1010 zmd->sb[dst_set].block + i, page);
1011 if (ret)
1012 goto out;
1013 }
1014
1015 /* Finalize with the super block */
1016 if (!zmd->sb[dst_set].mblk) {
1017 zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
1018 if (!zmd->sb[dst_set].mblk) {
1019 ret = -ENOMEM;
1020 goto out;
1021 }
1022 zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
1023 }
1024
1025 ret = dmz_write_sb(zmd, dst_set);
1026out:
1027 __free_pages(page, 0);
1028
1029 return ret;
1030}
1031
1032/*
1033 * Get super block from disk.
1034 */
1035static int dmz_load_sb(struct dmz_metadata *zmd)
1036{
1037 bool sb_good[2] = {false, false};
1038 u64 sb_gen[2] = {0, 0};
1039 int ret;
1040
1041 /* Read and check the primary super block */
1042 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
1043 ret = dmz_get_sb(zmd, 0);
1044 if (ret) {
1045 dmz_dev_err(zmd->dev, "Read primary super block failed");
1046 return ret;
1047 }
1048
1049 ret = dmz_check_sb(zmd, zmd->sb[0].sb);
1050
1051 /* Read and check secondary super block */
1052 if (ret == 0) {
1053 sb_good[0] = true;
1054 zmd->sb[1].block = zmd->sb[0].block +
1055 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
1056 ret = dmz_get_sb(zmd, 1);
1057 } else
1058 ret = dmz_lookup_secondary_sb(zmd);
1059
1060 if (ret) {
1061 dmz_dev_err(zmd->dev, "Read secondary super block failed");
1062 return ret;
1063 }
1064
1065 ret = dmz_check_sb(zmd, zmd->sb[1].sb);
1066 if (ret == 0)
1067 sb_good[1] = true;
1068
1069 /* Use highest generation sb first */
1070 if (!sb_good[0] && !sb_good[1]) {
1071 dmz_dev_err(zmd->dev, "No valid super block found");
1072 return -EIO;
1073 }
1074
1075 if (sb_good[0])
1076 sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1077 else
1078 ret = dmz_recover_mblocks(zmd, 0);
1079
1080 if (sb_good[1])
1081 sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1082 else
1083 ret = dmz_recover_mblocks(zmd, 1);
1084
1085 if (ret) {
1086 dmz_dev_err(zmd->dev, "Recovery failed");
1087 return -EIO;
1088 }
1089
1090 if (sb_gen[0] >= sb_gen[1]) {
1091 zmd->sb_gen = sb_gen[0];
1092 zmd->mblk_primary = 0;
1093 } else {
1094 zmd->sb_gen = sb_gen[1];
1095 zmd->mblk_primary = 1;
1096 }
1097
1098 dmz_dev_debug(zmd->dev, "Using super block %u (gen %llu)",
1099 zmd->mblk_primary, zmd->sb_gen);
1100
1101 return 0;
1102}
1103
1104/*
1105 * Initialize a zone descriptor.
1106 */
d4100351 1107static int dmz_init_zone(struct blk_zone *blkz, unsigned int idx, void *data)
3b1a94c8 1108{
d4100351
CH		 1109 struct dmz_metadata *zmd = data;
1110 struct dm_zone *zone = &zmd->zones[idx];
3b1a94c8
DLM		 1111 struct dmz_dev *dev = zmd->dev;
1112
1113 /* Ignore the eventual last runt (smaller) zone */
1114 if (blkz->len != dev->zone_nr_sectors) {
1115 if (blkz->start + blkz->len == dev->capacity)
1116 return 0;
1117 return -ENXIO;
1118 }
1119
1120 INIT_LIST_HEAD(&zone->link);
1121 atomic_set(&zone->refcount, 0);
1122 zone->chunk = DMZ_MAP_UNMAPPED;
1123
d4100351
CH		 1124 switch (blkz->type) {
1125 case BLK_ZONE_TYPE_CONVENTIONAL:
3b1a94c8 1126 set_bit(DMZ_RND, &zone->flags);
d4100351
CH		 1127 break;
1128 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1129 case BLK_ZONE_TYPE_SEQWRITE_PREF:
3b1a94c8 1130 set_bit(DMZ_SEQ, &zone->flags);
d4100351
CH		 1131 break;
1132 default:
3b1a94c8 1133 return -ENXIO;
d4100351 1134 }
3b1a94c8
DLM		 1135
1136 if (dmz_is_rnd(zone))
1137 zone->wp_block = 0;
1138 else
1139 zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1140
d4100351
CH		 1141 if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1142 set_bit(DMZ_OFFLINE, &zone->flags);
1143 else if (blkz->cond == BLK_ZONE_COND_READONLY)
1144 set_bit(DMZ_READ_ONLY, &zone->flags);
1145 else {
3b1a94c8
DLM		 1146 zmd->nr_useable_zones++;
1147 if (dmz_is_rnd(zone)) {
1148 zmd->nr_rnd_zones++;
1149 if (!zmd->sb_zone) {
1150 /* Super block zone */
1151 zmd->sb_zone = zone;
1152 }
1153 }
1154 }
1155
1156 return 0;
1157}
1158
1159/*
1160 * Free zones descriptors.
1161 */
1162static void dmz_drop_zones(struct dmz_metadata *zmd)
1163{
1164 kfree(zmd->zones);
1165 zmd->zones = NULL;
1166}
1167
3b1a94c8
DLM		 1168/*
1169 * Allocate and initialize zone descriptors using the zone
1170 * information from disk.
1171 */
1172static int dmz_init_zones(struct dmz_metadata *zmd)
1173{
1174 struct dmz_dev *dev = zmd->dev;
d4100351 1175 int ret;
3b1a94c8
DLM		 1176
1177 /* Init */
1178 zmd->zone_bitmap_size = dev->zone_nr_blocks >> 3;
b3996295
DF		 1179 zmd->zone_nr_bitmap_blocks =
1180 max_t(sector_t, 1, zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT);
1181 zmd->zone_bits_per_mblk = min_t(sector_t, dev->zone_nr_blocks,
1182 DMZ_BLOCK_SIZE_BITS);
3b1a94c8
DLM		 1183
1184 /* Allocate zone array */
1185 zmd->zones = kcalloc(dev->nr_zones, sizeof(struct dm_zone), GFP_KERNEL);
1186 if (!zmd->zones)
1187 return -ENOMEM;
1188
1189 dmz_dev_info(dev, "Using %zu B for zone information",
1190 sizeof(struct dm_zone) * dev->nr_zones);
1191
3b1a94c8 1192 /*
d4100351
CH		 1193 * Get zone information and initialize zone descriptors. At the same
1194 * time, determine where the super block should be: first block of the
1195 * first randomly writable zone.
3b1a94c8 1196 */
d4100351
CH		 1197 ret = blkdev_report_zones(dev->bdev, 0, BLK_ALL_ZONES, dmz_init_zone,
1198 zmd);
1199 if (ret < 0) {
1200 dmz_drop_zones(zmd);
1201 return ret;
1202 }
3b1a94c8 1203
d4100351
CH		 1204 return 0;
1205}
7aedf75f 1206
d4100351
CH		 1207static int dmz_update_zone_cb(struct blk_zone *blkz, unsigned int idx,
1208 void *data)
1209{
1210 struct dm_zone *zone = data;
3b1a94c8 1211
d4100351
CH		 1212 clear_bit(DMZ_OFFLINE, &zone->flags);
1213 clear_bit(DMZ_READ_ONLY, &zone->flags);
1214 if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1215 set_bit(DMZ_OFFLINE, &zone->flags);
1216 else if (blkz->cond == BLK_ZONE_COND_READONLY)
1217 set_bit(DMZ_READ_ONLY, &zone->flags);
3b1a94c8 1218
d4100351
CH		 1219 if (dmz_is_seq(zone))
1220 zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1221 else
1222 zone->wp_block = 0;
1223 return 0;
3b1a94c8
DLM		 1224}
1225
1226/*
1227 * Update a zone information.
1228 */
1229static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1230{
bd976e52 1231 unsigned int noio_flag;
3b1a94c8
DLM		 1232 int ret;
1233
bd976e52
DLM		 1234 /*
1235 * Get zone information from disk. Since blkdev_report_zones() uses
1236 * GFP_KERNEL by default for memory allocations, set the per-task
1237 * PF_MEMALLOC_NOIO flag so that all allocations are done as if
1238 * GFP_NOIO was specified.
1239 */
1240 noio_flag = memalloc_noio_save();
d4100351
CH		 1241 ret = blkdev_report_zones(zmd->dev->bdev, dmz_start_sect(zmd, zone), 1,
1242 dmz_update_zone_cb, zone);
bd976e52 1243 memalloc_noio_restore(noio_flag);
d4100351
CH		 1244
1245 if (ret == 0)
7aedf75f 1246 ret = -EIO;
d4100351 1247 if (ret < 0) {
3b1a94c8
DLM		 1248 dmz_dev_err(zmd->dev, "Get zone %u report failed",
1249 dmz_id(zmd, zone));
e7fad909 1250 dmz_check_bdev(zmd->dev);
3b1a94c8
DLM		 1251 return ret;
1252 }
1253
3b1a94c8
DLM		 1254 return 0;
1255}
1256
1257/*
1258 * Check a zone write pointer position when the zone is marked
1259 * with the sequential write error flag.
1260 */
1261static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1262 struct dm_zone *zone)
1263{
1264 unsigned int wp = 0;
1265 int ret;
1266
1267 wp = zone->wp_block;
1268 ret = dmz_update_zone(zmd, zone);
1269 if (ret)
1270 return ret;
1271
1272 dmz_dev_warn(zmd->dev, "Processing zone %u write error (zone wp %u/%u)",
1273 dmz_id(zmd, zone), zone->wp_block, wp);
1274
1275 if (zone->wp_block < wp) {
1276 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1277 wp - zone->wp_block);
1278 }
1279
1280 return 0;
1281}
1282
1283static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
1284{
1285 return &zmd->zones[zone_id];
1286}
1287
1288/*
1289 * Reset a zone write pointer.
1290 */
1291static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1292{
1293 int ret;
1294
1295 /*
1296 * Ignore offline zones, read only zones,
1297 * and conventional zones.
1298 */
1299 if (dmz_is_offline(zone) ||
1300 dmz_is_readonly(zone) ||
1301 dmz_is_rnd(zone))
1302 return 0;
1303
1304 if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1305 struct dmz_dev *dev = zmd->dev;
1306
6c1b1da5
AJ		 1307 ret = blkdev_zone_mgmt(dev->bdev, REQ_OP_ZONE_RESET,
1308 dmz_start_sect(zmd, zone),
1309 dev->zone_nr_sectors, GFP_NOIO);
3b1a94c8
DLM		 1310 if (ret) {
1311 dmz_dev_err(dev, "Reset zone %u failed %d",
1312 dmz_id(zmd, zone), ret);
1313 return ret;
1314 }
1315 }
1316
1317 /* Clear write error bit and rewind write pointer position */
1318 clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1319 zone->wp_block = 0;
1320
1321 return 0;
1322}
1323
1324static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1325
1326/*
1327 * Initialize chunk mapping.
1328 */
1329static int dmz_load_mapping(struct dmz_metadata *zmd)
1330{
1331 struct dmz_dev *dev = zmd->dev;
1332 struct dm_zone *dzone, *bzone;
1333 struct dmz_mblock *dmap_mblk = NULL;
1334 struct dmz_map *dmap;
1335 unsigned int i = 0, e = 0, chunk = 0;
1336 unsigned int dzone_id;
1337 unsigned int bzone_id;
1338
1339 /* Metadata block array for the chunk mapping table */
1340 zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1341 sizeof(struct dmz_mblk *), GFP_KERNEL);
1342 if (!zmd->map_mblk)
1343 return -ENOMEM;
1344
1345 /* Get chunk mapping table blocks and initialize zone mapping */
1346 while (chunk < zmd->nr_chunks) {
1347 if (!dmap_mblk) {
1348 /* Get mapping block */
1349 dmap_mblk = dmz_get_mblock(zmd, i + 1);
1350 if (IS_ERR(dmap_mblk))
1351 return PTR_ERR(dmap_mblk);
1352 zmd->map_mblk[i] = dmap_mblk;
1353 dmap = (struct dmz_map *) dmap_mblk->data;
1354 i++;
1355 e = 0;
1356 }
1357
1358 /* Check data zone */
1359 dzone_id = le32_to_cpu(dmap[e].dzone_id);
1360 if (dzone_id == DMZ_MAP_UNMAPPED)
1361 goto next;
1362
1363 if (dzone_id >= dev->nr_zones) {
1364 dmz_dev_err(dev, "Chunk %u mapping: invalid data zone ID %u",
1365 chunk, dzone_id);
1366 return -EIO;
1367 }
1368
1369 dzone = dmz_get(zmd, dzone_id);
1370 set_bit(DMZ_DATA, &dzone->flags);
1371 dzone->chunk = chunk;
1372 dmz_get_zone_weight(zmd, dzone);
1373
1374 if (dmz_is_rnd(dzone))
1375 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1376 else
1377 list_add_tail(&dzone->link, &zmd->map_seq_list);
1378
1379 /* Check buffer zone */
1380 bzone_id = le32_to_cpu(dmap[e].bzone_id);
1381 if (bzone_id == DMZ_MAP_UNMAPPED)
1382 goto next;
1383
1384 if (bzone_id >= dev->nr_zones) {
1385 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone ID %u",
1386 chunk, bzone_id);
1387 return -EIO;
1388 }
1389
1390 bzone = dmz_get(zmd, bzone_id);
1391 if (!dmz_is_rnd(bzone)) {
1392 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone %u",
1393 chunk, bzone_id);
1394 return -EIO;
1395 }
1396
1397 set_bit(DMZ_DATA, &bzone->flags);
1398 set_bit(DMZ_BUF, &bzone->flags);
1399 bzone->chunk = chunk;
1400 bzone->bzone = dzone;
1401 dzone->bzone = bzone;
1402 dmz_get_zone_weight(zmd, bzone);
1403 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1404next:
1405 chunk++;
1406 e++;
1407 if (e >= DMZ_MAP_ENTRIES)
1408 dmap_mblk = NULL;
1409 }
1410
1411 /*
1412 * At this point, only meta zones and mapped data zones were
1413 * fully initialized. All remaining zones are unmapped data
1414 * zones. Finish initializing those here.
1415 */
1416 for (i = 0; i < dev->nr_zones; i++) {
1417 dzone = dmz_get(zmd, i);
1418 if (dmz_is_meta(dzone))
1419 continue;
1420
1421 if (dmz_is_rnd(dzone))
1422 zmd->nr_rnd++;
1423 else
1424 zmd->nr_seq++;
1425
1426 if (dmz_is_data(dzone)) {
1427 /* Already initialized */
1428 continue;
1429 }
1430
1431 /* Unmapped data zone */
1432 set_bit(DMZ_DATA, &dzone->flags);
1433 dzone->chunk = DMZ_MAP_UNMAPPED;
1434 if (dmz_is_rnd(dzone)) {
1435 list_add_tail(&dzone->link, &zmd->unmap_rnd_list);
1436 atomic_inc(&zmd->unmap_nr_rnd);
1437 } else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1438 list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1439 atomic_inc(&zmd->nr_reserved_seq_zones);
1440 zmd->nr_seq--;
1441 } else {
1442 list_add_tail(&dzone->link, &zmd->unmap_seq_list);
1443 atomic_inc(&zmd->unmap_nr_seq);
1444 }
1445 }
1446
1447 return 0;
1448}
1449
1450/*
1451 * Set a data chunk mapping.
1452 */
1453static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1454 unsigned int dzone_id, unsigned int bzone_id)
1455{
1456 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1457 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1458 int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1459
1460 dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1461 dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1462 dmz_dirty_mblock(zmd, dmap_mblk);
1463}
1464
1465/*
1466 * The list of mapped zones is maintained in LRU order.
1467 * This rotates a zone at the end of its map list.
1468 */
1469static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1470{
1471 if (list_empty(&zone->link))
1472 return;
1473
1474 list_del_init(&zone->link);
1475 if (dmz_is_seq(zone)) {
1476 /* LRU rotate sequential zone */
1477 list_add_tail(&zone->link, &zmd->map_seq_list);
1478 } else {
1479 /* LRU rotate random zone */
1480 list_add_tail(&zone->link, &zmd->map_rnd_list);
1481 }
1482}
1483
1484/*
1485 * The list of mapped random zones is maintained
1486 * in LRU order. This rotates a zone at the end of the list.
1487 */
1488static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1489{
1490 __dmz_lru_zone(zmd, zone);
1491 if (zone->bzone)
1492 __dmz_lru_zone(zmd, zone->bzone);
1493}
1494
1495/*
1496 * Wait for any zone to be freed.
1497 */
1498static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1499{
1500 DEFINE_WAIT(wait);
1501
1502 prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1503 dmz_unlock_map(zmd);
1504 dmz_unlock_metadata(zmd);
1505
1506 io_schedule_timeout(HZ);
1507
1508 dmz_lock_metadata(zmd);
1509 dmz_lock_map(zmd);
1510 finish_wait(&zmd->free_wq, &wait);
1511}
1512
1513/*
1514 * Lock a zone for reclaim (set the zone RECLAIM bit).
1515 * Returns false if the zone cannot be locked or if it is already locked
1516 * and 1 otherwise.
1517 */
1518int dmz_lock_zone_reclaim(struct dm_zone *zone)
1519{
1520 /* Active zones cannot be reclaimed */
1521 if (dmz_is_active(zone))
1522 return 0;
1523
1524 return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1525}
1526
1527/*
1528 * Clear a zone reclaim flag.
1529 */
1530void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1531{
1532 WARN_ON(dmz_is_active(zone));
1533 WARN_ON(!dmz_in_reclaim(zone));
1534
1535 clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1536 smp_mb__after_atomic();
1537 wake_up_bit(&zone->flags, DMZ_RECLAIM);
1538}
1539
1540/*
1541 * Wait for a zone reclaim to complete.
1542 */
1543static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1544{
1545 dmz_unlock_map(zmd);
1546 dmz_unlock_metadata(zmd);
1547 wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1548 dmz_lock_metadata(zmd);
1549 dmz_lock_map(zmd);
1550}
1551
1552/*
1553 * Select a random write zone for reclaim.
1554 */
1555static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd)
1556{
1557 struct dm_zone *dzone = NULL;
1558 struct dm_zone *zone;
1559
1560 if (list_empty(&zmd->map_rnd_list))
b234c6d7 1561 return ERR_PTR(-EBUSY);
3b1a94c8
DLM		 1562
1563 list_for_each_entry(zone, &zmd->map_rnd_list, link) {
1564 if (dmz_is_buf(zone))
1565 dzone = zone->bzone;
1566 else
1567 dzone = zone;
1568 if (dmz_lock_zone_reclaim(dzone))
1569 return dzone;
1570 }
1571
b234c6d7 1572 return ERR_PTR(-EBUSY);
3b1a94c8
DLM		 1573}
1574
1575/*
1576 * Select a buffered sequential zone for reclaim.
1577 */
1578static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd)
1579{
1580 struct dm_zone *zone;
1581
1582 if (list_empty(&zmd->map_seq_list))
e0702d90 1583 return ERR_PTR(-EBUSY);
3b1a94c8
DLM		 1584
1585 list_for_each_entry(zone, &zmd->map_seq_list, link) {
1586 if (!zone->bzone)
1587 continue;
1588 if (dmz_lock_zone_reclaim(zone))
1589 return zone;
1590 }
1591
e0702d90 1592 return ERR_PTR(-EBUSY);
3b1a94c8
DLM		 1593}
1594
1595/*
1596 * Select a zone for reclaim.
1597 */
1598struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd)
1599{
1600 struct dm_zone *zone;
1601
1602 /*
1603 * Search for a zone candidate to reclaim: 2 cases are possible.
1604 * (1) There is no free sequential zones. Then a random data zone
1605 * cannot be reclaimed. So choose a sequential zone to reclaim so
1606 * that afterward a random zone can be reclaimed.
1607 * (2) At least one free sequential zone is available, then choose
1608 * the oldest random zone (data or buffer) that can be locked.
1609 */
1610 dmz_lock_map(zmd);
1611 if (list_empty(&zmd->reserved_seq_zones_list))
1612 zone = dmz_get_seq_zone_for_reclaim(zmd);
1613 else
1614 zone = dmz_get_rnd_zone_for_reclaim(zmd);
1615 dmz_unlock_map(zmd);
1616
1617 return zone;
1618}
1619
3b1a94c8
DLM		 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
ad1bd578 1647 /* Allocate a random zone */
3b1a94c8
DLM		 1648 dzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1649 if (!dzone) {
75d66ffb
DF		 1650 if (dmz_bdev_is_dying(zmd->dev)) {
1651 dzone = ERR_PTR(-EIO);
1652 goto out;
1653 }
3b1a94c8
DLM		 1654 dmz_wait_for_free_zones(zmd);
1655 goto again;
1656 }
1657
1658 dmz_map_zone(zmd, dzone, chunk);
1659
1660 } else {
1661 /* The chunk is already mapped: get the mapping zone */
1662 dzone = dmz_get(zmd, dzone_id);
1663 if (dzone->chunk != chunk) {
1664 dzone = ERR_PTR(-EIO);
1665 goto out;
1666 }
1667
1668 /* Repair write pointer if the sequential dzone has error */
1669 if (dmz_seq_write_err(dzone)) {
1670 ret = dmz_handle_seq_write_err(zmd, dzone);
1671 if (ret) {
1672 dzone = ERR_PTR(-EIO);
1673 goto out;
1674 }
1675 clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
1676 }
1677 }
1678
1679 /*
1680 * If the zone is being reclaimed, the chunk mapping may change
1681 * to a different zone. So wait for reclaim and retry. Otherwise,
1682 * activate the zone (this will prevent reclaim from touching it).
1683 */
1684 if (dmz_in_reclaim(dzone)) {
1685 dmz_wait_for_reclaim(zmd, dzone);
1686 goto again;
1687 }
1688 dmz_activate_zone(dzone);
1689 dmz_lru_zone(zmd, dzone);
1690out:
1691 dmz_unlock_map(zmd);
1692
1693 return dzone;
1694}
1695
1696/*
1697 * Write and discard change the block validity of data zones and their buffer
1698 * zones. Check here that valid blocks are still present. If all blocks are
1699 * invalid, the zones can be unmapped on the fly without waiting for reclaim
1700 * to do it.
1701 */
1702void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
1703{
1704 struct dm_zone *bzone;
1705
1706 dmz_lock_map(zmd);
1707
1708 bzone = dzone->bzone;
1709 if (bzone) {
1710 if (dmz_weight(bzone))
1711 dmz_lru_zone(zmd, bzone);
1712 else {
1713 /* Empty buffer zone: reclaim it */
1714 dmz_unmap_zone(zmd, bzone);
1715 dmz_free_zone(zmd, bzone);
1716 bzone = NULL;
1717 }
1718 }
1719
1720 /* Deactivate the data zone */
1721 dmz_deactivate_zone(dzone);
1722 if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
1723 dmz_lru_zone(zmd, dzone);
1724 else {
1725 /* Unbuffered inactive empty data zone: reclaim it */
1726 dmz_unmap_zone(zmd, dzone);
1727 dmz_free_zone(zmd, dzone);
1728 }
1729
1730 dmz_unlock_map(zmd);
1731}
1732
1733/*
1734 * Allocate and map a random zone to buffer a chunk
1735 * already mapped to a sequential zone.
1736 */
1737struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
1738 struct dm_zone *dzone)
1739{
1740 struct dm_zone *bzone;
1741
1742 dmz_lock_map(zmd);
1743again:
1744 bzone = dzone->bzone;
1745 if (bzone)
1746 goto out;
1747
ad1bd578 1748 /* Allocate a random zone */
3b1a94c8
DLM		 1749 bzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1750 if (!bzone) {
75d66ffb
DF		 1751 if (dmz_bdev_is_dying(zmd->dev)) {
1752 bzone = ERR_PTR(-EIO);
1753 goto out;
1754 }
3b1a94c8
DLM		 1755 dmz_wait_for_free_zones(zmd);
1756 goto again;
1757 }
1758
1759 /* Update the chunk mapping */
1760 dmz_set_chunk_mapping(zmd, dzone->chunk, dmz_id(zmd, dzone),
1761 dmz_id(zmd, bzone));
1762
1763 set_bit(DMZ_BUF, &bzone->flags);
1764 bzone->chunk = dzone->chunk;
1765 bzone->bzone = dzone;
1766 dzone->bzone = bzone;
1767 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1768out:
1769 dmz_unlock_map(zmd);
1770
1771 return bzone;
1772}
1773
1774/*
1775 * Get an unmapped (free) zone.
1776 * This must be called with the mapping lock held.
1777 */
1778struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags)
1779{
1780 struct list_head *list;
1781 struct dm_zone *zone;
1782
1783 if (flags & DMZ_ALLOC_RND)
1784 list = &zmd->unmap_rnd_list;
1785 else
1786 list = &zmd->unmap_seq_list;
1787again:
1788 if (list_empty(list)) {
1789 /*
1790 * No free zone: if this is for reclaim, allow using the
1791 * reserved sequential zones.
1792 */
1793 if (!(flags & DMZ_ALLOC_RECLAIM) ||
1794 list_empty(&zmd->reserved_seq_zones_list))
1795 return NULL;
1796
1797 zone = list_first_entry(&zmd->reserved_seq_zones_list,
1798 struct dm_zone, link);
1799 list_del_init(&zone->link);
1800 atomic_dec(&zmd->nr_reserved_seq_zones);
1801 return zone;
1802 }
1803
1804 zone = list_first_entry(list, struct dm_zone, link);
1805 list_del_init(&zone->link);
1806
1807 if (dmz_is_rnd(zone))
1808 atomic_dec(&zmd->unmap_nr_rnd);
1809 else
1810 atomic_dec(&zmd->unmap_nr_seq);
1811
1812 if (dmz_is_offline(zone)) {
1813 dmz_dev_warn(zmd->dev, "Zone %u is offline", dmz_id(zmd, zone));
1814 zone = NULL;
1815 goto again;
1816 }
1817
1818 return zone;
1819}
1820
1821/*
1822 * Free a zone.
1823 * This must be called with the mapping lock held.
1824 */
1825void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1826{
1827 /* If this is a sequential zone, reset it */
1828 if (dmz_is_seq(zone))
1829 dmz_reset_zone(zmd, zone);
1830
1831 /* Return the zone to its type unmap list */
1832 if (dmz_is_rnd(zone)) {
1833 list_add_tail(&zone->link, &zmd->unmap_rnd_list);
1834 atomic_inc(&zmd->unmap_nr_rnd);
1835 } else if (atomic_read(&zmd->nr_reserved_seq_zones) <
1836 zmd->nr_reserved_seq) {
1837 list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
1838 atomic_inc(&zmd->nr_reserved_seq_zones);
1839 } else {
1840 list_add_tail(&zone->link, &zmd->unmap_seq_list);
1841 atomic_inc(&zmd->unmap_nr_seq);
1842 }
1843
1844 wake_up_all(&zmd->free_wq);
1845}
1846
1847/*
1848 * Map a chunk to a zone.
1849 * This must be called with the mapping lock held.
1850 */
1851void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
1852 unsigned int chunk)
1853{
1854 /* Set the chunk mapping */
1855 dmz_set_chunk_mapping(zmd, chunk, dmz_id(zmd, dzone),
1856 DMZ_MAP_UNMAPPED);
1857 dzone->chunk = chunk;
1858 if (dmz_is_rnd(dzone))
1859 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1860 else
1861 list_add_tail(&dzone->link, &zmd->map_seq_list);
1862}
1863
1864/*
1865 * Unmap a zone.
1866 * This must be called with the mapping lock held.
1867 */
1868void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1869{
1870 unsigned int chunk = zone->chunk;
1871 unsigned int dzone_id;
1872
1873 if (chunk == DMZ_MAP_UNMAPPED) {
1874 /* Already unmapped */
1875 return;
1876 }
1877
1878 if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
1879 /*
1880 * Unmapping the chunk buffer zone: clear only
1881 * the chunk buffer mapping
1882 */
1883 dzone_id = dmz_id(zmd, zone->bzone);
1884 zone->bzone->bzone = NULL;
1885 zone->bzone = NULL;
1886
1887 } else {
1888 /*
1889 * Unmapping the chunk data zone: the zone must
1890 * not be buffered.
1891 */
1892 if (WARN_ON(zone->bzone)) {
1893 zone->bzone->bzone = NULL;
1894 zone->bzone = NULL;
1895 }
1896 dzone_id = DMZ_MAP_UNMAPPED;
1897 }
1898
1899 dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
1900
1901 zone->chunk = DMZ_MAP_UNMAPPED;
1902 list_del_init(&zone->link);
1903}
1904
1905/*
1906 * Set @nr_bits bits in @bitmap starting from @bit.
1907 * Return the number of bits changed from 0 to 1.
1908 */
1909static unsigned int dmz_set_bits(unsigned long *bitmap,
1910 unsigned int bit, unsigned int nr_bits)
1911{
1912 unsigned long *addr;
1913 unsigned int end = bit + nr_bits;
1914 unsigned int n = 0;
1915
1916 while (bit < end) {
1917 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
1918 ((end - bit) >= BITS_PER_LONG)) {
1919 /* Try to set the whole word at once */
1920 addr = bitmap + BIT_WORD(bit);
1921 if (*addr == 0) {
1922 *addr = ULONG_MAX;
1923 n += BITS_PER_LONG;
1924 bit += BITS_PER_LONG;
1925 continue;
1926 }
1927 }
1928
1929 if (!test_and_set_bit(bit, bitmap))
1930 n++;
1931 bit++;
1932 }
1933
1934 return n;
1935}
1936
1937/*
1938 * Get the bitmap block storing the bit for chunk_block in zone.
1939 */
1940static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
1941 struct dm_zone *zone,
1942 sector_t chunk_block)
1943{
1944 sector_t bitmap_block = 1 + zmd->nr_map_blocks +
1945 (sector_t)(dmz_id(zmd, zone) * zmd->zone_nr_bitmap_blocks) +
1946 (chunk_block >> DMZ_BLOCK_SHIFT_BITS);
1947
1948 return dmz_get_mblock(zmd, bitmap_block);
1949}
1950
1951/*
1952 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
1953 */
1954int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1955 struct dm_zone *to_zone)
1956{
1957 struct dmz_mblock *from_mblk, *to_mblk;
1958 sector_t chunk_block = 0;
1959
1960 /* Get the zones bitmap blocks */
1961 while (chunk_block < zmd->dev->zone_nr_blocks) {
1962 from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
1963 if (IS_ERR(from_mblk))
1964 return PTR_ERR(from_mblk);
1965 to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
1966 if (IS_ERR(to_mblk)) {
1967 dmz_release_mblock(zmd, from_mblk);
1968 return PTR_ERR(to_mblk);
1969 }
1970
1971 memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
1972 dmz_dirty_mblock(zmd, to_mblk);
1973
1974 dmz_release_mblock(zmd, to_mblk);
1975 dmz_release_mblock(zmd, from_mblk);
1976
b3996295 1977 chunk_block += zmd->zone_bits_per_mblk;
3b1a94c8
DLM		 1978 }
1979
1980 to_zone->weight = from_zone->weight;
1981
1982 return 0;
1983}
1984
1985/*
1986 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
1987 * starting from chunk_block.
1988 */
1989int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1990 struct dm_zone *to_zone, sector_t chunk_block)
1991{
1992 unsigned int nr_blocks;
1993 int ret;
1994
1995 /* Get the zones bitmap blocks */
1996 while (chunk_block < zmd->dev->zone_nr_blocks) {
1997 /* Get a valid region from the source zone */
1998 ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
1999 if (ret <= 0)
2000 return ret;
2001
2002 nr_blocks = ret;
2003 ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
2004 if (ret)
2005 return ret;
2006
2007 chunk_block += nr_blocks;
2008 }
2009
2010 return 0;
2011}
2012
2013/*
2014 * Validate all the blocks in the range [block..block+nr_blocks-1].
2015 */
2016int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2017 sector_t chunk_block, unsigned int nr_blocks)
2018{
2019 unsigned int count, bit, nr_bits;
2020 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
2021 struct dmz_mblock *mblk;
2022 unsigned int n = 0;
2023
2024 dmz_dev_debug(zmd->dev, "=> VALIDATE zone %u, block %llu, %u blocks",
2025 dmz_id(zmd, zone), (unsigned long long)chunk_block,
2026 nr_blocks);
2027
2028 WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2029
2030 while (nr_blocks) {
2031 /* Get bitmap block */
2032 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2033 if (IS_ERR(mblk))
2034 return PTR_ERR(mblk);
2035
2036 /* Set bits */
2037 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
b3996295 2038 nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
3b1a94c8
DLM		 2039
2040 count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2041 if (count) {
2042 dmz_dirty_mblock(zmd, mblk);
2043 n += count;
2044 }
2045 dmz_release_mblock(zmd, mblk);
2046
2047 nr_blocks -= nr_bits;
2048 chunk_block += nr_bits;
2049 }
2050
2051 if (likely(zone->weight + n <= zone_nr_blocks))
2052 zone->weight += n;
2053 else {
2054 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be <= %u",
2055 dmz_id(zmd, zone), zone->weight,
2056 zone_nr_blocks - n);
2057 zone->weight = zone_nr_blocks;
2058 }
2059
2060 return 0;
2061}
2062
2063/*
2064 * Clear nr_bits bits in bitmap starting from bit.
2065 * Return the number of bits cleared.
2066 */
2067static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2068{
2069 unsigned long *addr;
2070 int end = bit + nr_bits;
2071 int n = 0;
2072
2073 while (bit < end) {
2074 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2075 ((end - bit) >= BITS_PER_LONG)) {
2076 /* Try to clear whole word at once */
2077 addr = bitmap + BIT_WORD(bit);
2078 if (*addr == ULONG_MAX) {
2079 *addr = 0;
2080 n += BITS_PER_LONG;
2081 bit += BITS_PER_LONG;
2082 continue;
2083 }
2084 }
2085
2086 if (test_and_clear_bit(bit, bitmap))
2087 n++;
2088 bit++;
2089 }
2090
2091 return n;
2092}
2093
2094/*
2095 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2096 */
2097int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2098 sector_t chunk_block, unsigned int nr_blocks)
2099{
2100 unsigned int count, bit, nr_bits;
2101 struct dmz_mblock *mblk;
2102 unsigned int n = 0;
2103
2104 dmz_dev_debug(zmd->dev, "=> INVALIDATE zone %u, block %llu, %u blocks",
2105 dmz_id(zmd, zone), (u64)chunk_block, nr_blocks);
2106
2107 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2108
2109 while (nr_blocks) {
2110 /* Get bitmap block */
2111 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2112 if (IS_ERR(mblk))
2113 return PTR_ERR(mblk);
2114
2115 /* Clear bits */
2116 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
b3996295 2117 nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
3b1a94c8
DLM		 2118
2119 count = dmz_clear_bits((unsigned long *)mblk->data,
2120 bit, nr_bits);
2121 if (count) {
2122 dmz_dirty_mblock(zmd, mblk);
2123 n += count;
2124 }
2125 dmz_release_mblock(zmd, mblk);
2126
2127 nr_blocks -= nr_bits;
2128 chunk_block += nr_bits;
2129 }
2130
2131 if (zone->weight >= n)
2132 zone->weight -= n;
2133 else {
2134 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be >= %u",
2135 dmz_id(zmd, zone), zone->weight, n);
2136 zone->weight = 0;
2137 }
2138
2139 return 0;
2140}
2141
2142/*
2143 * Get a block bit value.
2144 */
2145static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2146 sector_t chunk_block)
2147{
2148 struct dmz_mblock *mblk;
2149 int ret;
2150
2151 WARN_ON(chunk_block >= zmd->dev->zone_nr_blocks);
2152
2153 /* Get bitmap block */
2154 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2155 if (IS_ERR(mblk))
2156 return PTR_ERR(mblk);
2157
2158 /* Get offset */
2159 ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2160 (unsigned long *) mblk->data) != 0;
2161
2162 dmz_release_mblock(zmd, mblk);
2163
2164 return ret;
2165}
2166
2167/*
2168 * Return the number of blocks from chunk_block to the first block with a bit
2169 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2170 */
2171static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2172 sector_t chunk_block, unsigned int nr_blocks,
2173 int set)
2174{
2175 struct dmz_mblock *mblk;
2176 unsigned int bit, set_bit, nr_bits;
b3996295 2177 unsigned int zone_bits = zmd->zone_bits_per_mblk;
3b1a94c8
DLM		 2178 unsigned long *bitmap;
2179 int n = 0;
2180
2181 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2182
2183 while (nr_blocks) {
2184 /* Get bitmap block */
2185 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2186 if (IS_ERR(mblk))
2187 return PTR_ERR(mblk);
2188
2189 /* Get offset */
2190 bitmap = (unsigned long *) mblk->data;
2191 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
b3996295 2192 nr_bits = min(nr_blocks, zone_bits - bit);
3b1a94c8 2193 if (set)
b3996295 2194 set_bit = find_next_bit(bitmap, zone_bits, bit);
3b1a94c8 2195 else
b3996295 2196 set_bit = find_next_zero_bit(bitmap, zone_bits, bit);
3b1a94c8
DLM		 2197 dmz_release_mblock(zmd, mblk);
2198
2199 n += set_bit - bit;
b3996295 2200 if (set_bit < zone_bits)
3b1a94c8
DLM		 2201 break;
2202
2203 nr_blocks -= nr_bits;
2204 chunk_block += nr_bits;
2205 }
2206
2207 return n;
2208}
2209
2210/*
2211 * Test if chunk_block is valid. If it is, the number of consecutive
2212 * valid blocks from chunk_block will be returned.
2213 */
2214int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2215 sector_t chunk_block)
2216{
2217 int valid;
2218
2219 valid = dmz_test_block(zmd, zone, chunk_block);
2220 if (valid <= 0)
2221 return valid;
2222
2223 /* The block is valid: get the number of valid blocks from block */
2224 return dmz_to_next_set_block(zmd, zone, chunk_block,
2225 zmd->dev->zone_nr_blocks - chunk_block, 0);
2226}
2227
2228/*
2229 * Find the first valid block from @chunk_block in @zone.
2230 * If such a block is found, its number is returned using
2231 * @chunk_block and the total number of valid blocks from @chunk_block
2232 * is returned.
2233 */
2234int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2235 sector_t *chunk_block)
2236{
2237 sector_t start_block = *chunk_block;
2238 int ret;
2239
2240 ret = dmz_to_next_set_block(zmd, zone, start_block,
2241 zmd->dev->zone_nr_blocks - start_block, 1);
2242 if (ret < 0)
2243 return ret;
2244
2245 start_block += ret;
2246 *chunk_block = start_block;
2247
2248 return dmz_to_next_set_block(zmd, zone, start_block,
2249 zmd->dev->zone_nr_blocks - start_block, 0);
2250}
2251
2252/*
2253 * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2254 */
2255static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2256{
2257 unsigned long *addr;
2258 int end = bit + nr_bits;
2259 int n = 0;
2260
2261 while (bit < end) {
2262 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2263 ((end - bit) >= BITS_PER_LONG)) {
2264 addr = (unsigned long *)bitmap + BIT_WORD(bit);
2265 if (*addr == ULONG_MAX) {
2266 n += BITS_PER_LONG;
2267 bit += BITS_PER_LONG;
2268 continue;
2269 }
2270 }
2271
2272 if (test_bit(bit, bitmap))
2273 n++;
2274 bit++;
2275 }
2276
2277 return n;
2278}
2279
2280/*
2281 * Get a zone weight.
2282 */
2283static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2284{
2285 struct dmz_mblock *mblk;
2286 sector_t chunk_block = 0;
2287 unsigned int bit, nr_bits;
2288 unsigned int nr_blocks = zmd->dev->zone_nr_blocks;
2289 void *bitmap;
2290 int n = 0;
2291
2292 while (nr_blocks) {
2293 /* Get bitmap block */
2294 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2295 if (IS_ERR(mblk)) {
2296 n = 0;
2297 break;
2298 }
2299
2300 /* Count bits in this block */
2301 bitmap = mblk->data;
2302 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
b3996295 2303 nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
3b1a94c8
DLM		 2304 n += dmz_count_bits(bitmap, bit, nr_bits);
2305
2306 dmz_release_mblock(zmd, mblk);
2307
2308 nr_blocks -= nr_bits;
2309 chunk_block += nr_bits;
2310 }
2311
2312 zone->weight = n;
2313}
2314
2315/*
2316 * Cleanup the zoned metadata resources.
2317 */
2318static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2319{
2320 struct rb_root *root;
2321 struct dmz_mblock *mblk, *next;
2322 int i;
2323
2324 /* Release zone mapping resources */
2325 if (zmd->map_mblk) {
2326 for (i = 0; i < zmd->nr_map_blocks; i++)
2327 dmz_release_mblock(zmd, zmd->map_mblk[i]);
2328 kfree(zmd->map_mblk);
2329 zmd->map_mblk = NULL;
2330 }
2331
2332 /* Release super blocks */
2333 for (i = 0; i < 2; i++) {
2334 if (zmd->sb[i].mblk) {
2335 dmz_free_mblock(zmd, zmd->sb[i].mblk);
2336 zmd->sb[i].mblk = NULL;
2337 }
2338 }
2339
2340 /* Free cached blocks */
2341 while (!list_empty(&zmd->mblk_dirty_list)) {
2342 mblk = list_first_entry(&zmd->mblk_dirty_list,
2343 struct dmz_mblock, link);
2344 dmz_dev_warn(zmd->dev, "mblock %llu still in dirty list (ref %u)",
33c2865f 2345 (u64)mblk->no, mblk->ref);
3b1a94c8
DLM		 2346 list_del_init(&mblk->link);
2347 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2348 dmz_free_mblock(zmd, mblk);
2349 }
2350
2351 while (!list_empty(&zmd->mblk_lru_list)) {
2352 mblk = list_first_entry(&zmd->mblk_lru_list,
2353 struct dmz_mblock, link);
2354 list_del_init(&mblk->link);
2355 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2356 dmz_free_mblock(zmd, mblk);
2357 }
2358
2359 /* Sanity checks: the mblock rbtree should now be empty */
2360 root = &zmd->mblk_rbtree;
2361 rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2362 dmz_dev_warn(zmd->dev, "mblock %llu ref %u still in rbtree",
33c2865f
DLM		 2363 (u64)mblk->no, mblk->ref);
2364 mblk->ref = 0;
3b1a94c8
DLM		 2365 dmz_free_mblock(zmd, mblk);
2366 }
2367
2368 /* Free the zone descriptors */
2369 dmz_drop_zones(zmd);
d5ffebdd
MS		 2370
2371 mutex_destroy(&zmd->mblk_flush_lock);
2372 mutex_destroy(&zmd->map_lock);
3b1a94c8
DLM		 2373}
2374
2375/*
2376 * Initialize the zoned metadata.
2377 */
2378int dmz_ctr_metadata(struct dmz_dev *dev, struct dmz_metadata **metadata)
2379{
2380 struct dmz_metadata *zmd;
2381 unsigned int i, zid;
2382 struct dm_zone *zone;
2383 int ret;
2384
2385 zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2386 if (!zmd)
2387 return -ENOMEM;
2388
2389 zmd->dev = dev;
2390 zmd->mblk_rbtree = RB_ROOT;
2391 init_rwsem(&zmd->mblk_sem);
2392 mutex_init(&zmd->mblk_flush_lock);
2393 spin_lock_init(&zmd->mblk_lock);
2394 INIT_LIST_HEAD(&zmd->mblk_lru_list);
2395 INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2396
2397 mutex_init(&zmd->map_lock);
2398 atomic_set(&zmd->unmap_nr_rnd, 0);
2399 INIT_LIST_HEAD(&zmd->unmap_rnd_list);
2400 INIT_LIST_HEAD(&zmd->map_rnd_list);
2401
2402 atomic_set(&zmd->unmap_nr_seq, 0);
2403 INIT_LIST_HEAD(&zmd->unmap_seq_list);
2404 INIT_LIST_HEAD(&zmd->map_seq_list);
2405
2406 atomic_set(&zmd->nr_reserved_seq_zones, 0);
2407 INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2408
2409 init_waitqueue_head(&zmd->free_wq);
2410
2411 /* Initialize zone descriptors */
2412 ret = dmz_init_zones(zmd);
2413 if (ret)
2414 goto err;
2415
2416 /* Get super block */
2417 ret = dmz_load_sb(zmd);
2418 if (ret)
2419 goto err;
2420
2421 /* Set metadata zones starting from sb_zone */
2422 zid = dmz_id(zmd, zmd->sb_zone);
2423 for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2424 zone = dmz_get(zmd, zid + i);
2425 if (!dmz_is_rnd(zone))
2426 goto err;
2427 set_bit(DMZ_META, &zone->flags);
2428 }
2429
2430 /* Load mapping table */
2431 ret = dmz_load_mapping(zmd);
2432 if (ret)
2433 goto err;
2434
2435 /*
2436 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2437 * blocks and enough blocks to be able to cache the bitmap blocks of
2438 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2439 * the cache to add 512 more metadata blocks.
2440 */
2441 zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2442 zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2443 zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2444 zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2445 zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2446
2447 /* Metadata cache shrinker */
2448 ret = register_shrinker(&zmd->mblk_shrinker);
2449 if (ret) {
2450 dmz_dev_err(dev, "Register metadata cache shrinker failed");
2451 goto err;
2452 }
2453
2454 dmz_dev_info(dev, "Host-%s zoned block device",
2455 bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2456 "aware" : "managed");
2457 dmz_dev_info(dev, " %llu 512-byte logical sectors",
2458 (u64)dev->capacity);
2459 dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors",
2460 dev->nr_zones, (u64)dev->zone_nr_sectors);
2461 dmz_dev_info(dev, " %u metadata zones",
2462 zmd->nr_meta_zones * 2);
2463 dmz_dev_info(dev, " %u data zones for %u chunks",
2464 zmd->nr_data_zones, zmd->nr_chunks);
2465 dmz_dev_info(dev, " %u random zones (%u unmapped)",
2466 zmd->nr_rnd, atomic_read(&zmd->unmap_nr_rnd));
2467 dmz_dev_info(dev, " %u sequential zones (%u unmapped)",
2468 zmd->nr_seq, atomic_read(&zmd->unmap_nr_seq));
2469 dmz_dev_info(dev, " %u reserved sequential data zones",
2470 zmd->nr_reserved_seq);
2471
2472 dmz_dev_debug(dev, "Format:");
2473 dmz_dev_debug(dev, "%u metadata blocks per set (%u max cache)",
2474 zmd->nr_meta_blocks, zmd->max_nr_mblks);
2475 dmz_dev_debug(dev, " %u data zone mapping blocks",
2476 zmd->nr_map_blocks);
2477 dmz_dev_debug(dev, " %u bitmap blocks",
2478 zmd->nr_bitmap_blocks);
2479
2480 *metadata = zmd;
2481
2482 return 0;
2483err:
2484 dmz_cleanup_metadata(zmd);
2485 kfree(zmd);
2486 *metadata = NULL;
2487
2488 return ret;
2489}
2490
2491/*
2492 * Cleanup the zoned metadata resources.
2493 */
2494void dmz_dtr_metadata(struct dmz_metadata *zmd)
2495{
2496 unregister_shrinker(&zmd->mblk_shrinker);
2497 dmz_cleanup_metadata(zmd);
2498 kfree(zmd);
2499}
2500
2501/*
2502 * Check zone information on resume.
2503 */
2504int dmz_resume_metadata(struct dmz_metadata *zmd)
2505{
2506 struct dmz_dev *dev = zmd->dev;
2507 struct dm_zone *zone;
2508 sector_t wp_block;
2509 unsigned int i;
2510 int ret;
2511
2512 /* Check zones */
2513 for (i = 0; i < dev->nr_zones; i++) {
2514 zone = dmz_get(zmd, i);
2515 if (!zone) {
2516 dmz_dev_err(dev, "Unable to get zone %u", i);
2517 return -EIO;
2518 }
2519
2520 wp_block = zone->wp_block;
2521
2522 ret = dmz_update_zone(zmd, zone);
2523 if (ret) {
2524 dmz_dev_err(dev, "Broken zone %u", i);
2525 return ret;
2526 }
2527
2528 if (dmz_is_offline(zone)) {
2529 dmz_dev_warn(dev, "Zone %u is offline", i);
2530 continue;
2531 }
2532
2533 /* Check write pointer */
2534 if (!dmz_is_seq(zone))
2535 zone->wp_block = 0;
2536 else if (zone->wp_block != wp_block) {
2537 dmz_dev_err(dev, "Zone %u: Invalid wp (%llu / %llu)",
2538 i, (u64)zone->wp_block, (u64)wp_block);
2539 zone->wp_block = wp_block;
2540 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
2541 dev->zone_nr_blocks - zone->wp_block);
2542 }
2543 }
2544
2545 return 0;
2546}
Linux 6.x block layer and io_uring tree(s)