?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
treewide: remove redundant IS_ERR() before error code check
[linux/fpc-iii.git] / drivers / md / dm-zoned-metadata.c
blob516c7b671d257e71c663b11b56177e59be7455a4
1 // SPDX-License-Identifier: GPL-2.0-only
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>
12 #include <linux/sched/mm.h>
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)
38 * All metadata blocks are stored in conventional zones, starting from
39 * the first conventional zone found on disk.
40 */
41 struct 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 */
84 struct 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 */
100 struct dmz_mblock {
101 struct rb_node node;
102 struct list_head link;
103 sector_t no;
104 unsigned int ref;
105 unsigned long state;
106 struct page *page;
107 void *data;
108 };
109
110 /*
111 * Metadata block state flags.
112 */
113 enum {
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 */
123 struct dmz_sb {
124 sector_t block;
125 struct dmz_mblock *mblk;
126 struct dmz_super *sb;
127 };
128
129 /*
130 * In-memory metadata.
131 */
132 struct dmz_metadata {
133 struct dmz_dev *dev;
134
135 sector_t zone_bitmap_size;
136 unsigned int zone_nr_bitmap_blocks;
137 unsigned int zone_bits_per_mblk;
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 */
190 unsigned int dmz_id(struct dmz_metadata *zmd, struct dm_zone *zone)
191 {
192 return ((unsigned int)(zone - zmd->zones));
193 }
194
195 sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone)
196 {
197 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_sectors_shift;
198 }
199
200 sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone)
201 {
202 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_blocks_shift;
203 }
204
205 unsigned int dmz_nr_chunks(struct dmz_metadata *zmd)
206 {
207 return zmd->nr_chunks;
208 }
209
210 unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd)
211 {
212 return zmd->nr_rnd;
213 }
214
215 unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd)
216 {
217 return atomic_read(&zmd->unmap_nr_rnd);
218 }
219
220 /*
221 * Lock/unlock mapping table.
222 * The map lock also protects all the zone lists.
223 */
224 void dmz_lock_map(struct dmz_metadata *zmd)
225 {
226 mutex_lock(&zmd->map_lock);
227 }
228
229 void dmz_unlock_map(struct dmz_metadata *zmd)
230 {
231 mutex_unlock(&zmd->map_lock);
232 }
233
234 /*
235 * Lock/unlock metadata access. This is a "read" lock on a semaphore
236 * that prevents metadata flush from running while metadata are being
237 * modified. The actual metadata write mutual exclusion is achieved with
238 * the map lock and zone state management (active and reclaim state are
239 * mutually exclusive).
240 */
241 void dmz_lock_metadata(struct dmz_metadata *zmd)
242 {
243 down_read(&zmd->mblk_sem);
244 }
245
246 void dmz_unlock_metadata(struct dmz_metadata *zmd)
247 {
248 up_read(&zmd->mblk_sem);
249 }
250
251 /*
252 * Lock/unlock flush: prevent concurrent executions
253 * of dmz_flush_metadata as well as metadata modification in reclaim
254 * while flush is being executed.
255 */
256 void dmz_lock_flush(struct dmz_metadata *zmd)
257 {
258 mutex_lock(&zmd->mblk_flush_lock);
259 }
260
261 void dmz_unlock_flush(struct dmz_metadata *zmd)
262 {
263 mutex_unlock(&zmd->mblk_flush_lock);
264 }
265
266 /*
267 * Allocate a metadata block.
268 */
269 static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd,
270 sector_t mblk_no)
271 {
272 struct dmz_mblock *mblk = NULL;
273
274 /* See if we can reuse cached blocks */
275 if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) {
276 spin_lock(&zmd->mblk_lock);
277 mblk = list_first_entry_or_null(&zmd->mblk_lru_list,
278 struct dmz_mblock, link);
279 if (mblk) {
280 list_del_init(&mblk->link);
281 rb_erase(&mblk->node, &zmd->mblk_rbtree);
282 mblk->no = mblk_no;
283 }
284 spin_unlock(&zmd->mblk_lock);
285 if (mblk)
286 return mblk;
287 }
288
289 /* Allocate a new block */
290 mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO);
291 if (!mblk)
292 return NULL;
293
294 mblk->page = alloc_page(GFP_NOIO);
295 if (!mblk->page) {
296 kfree(mblk);
297 return NULL;
298 }
299
300 RB_CLEAR_NODE(&mblk->node);
301 INIT_LIST_HEAD(&mblk->link);
302 mblk->ref = 0;
303 mblk->state = 0;
304 mblk->no = mblk_no;
305 mblk->data = page_address(mblk->page);
306
307 atomic_inc(&zmd->nr_mblks);
308
309 return mblk;
310 }
311
312 /*
313 * Free a metadata block.
314 */
315 static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
316 {
317 __free_pages(mblk->page, 0);
318 kfree(mblk);
319
320 atomic_dec(&zmd->nr_mblks);
321 }
322
323 /*
324 * Insert a metadata block in the rbtree.
325 */
326 static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
327 {
328 struct rb_root *root = &zmd->mblk_rbtree;
329 struct rb_node **new = &(root->rb_node), *parent = NULL;
330 struct dmz_mblock *b;
331
332 /* Figure out where to put the new node */
333 while (*new) {
334 b = container_of(*new, struct dmz_mblock, node);
335 parent = *new;
336 new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right);
337 }
338
339 /* Add new node and rebalance tree */
340 rb_link_node(&mblk->node, parent, new);
341 rb_insert_color(&mblk->node, root);
342 }
343
344 /*
345 * Lookup a metadata block in the rbtree. If the block is found, increment
346 * its reference count.
347 */
348 static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd,
349 sector_t mblk_no)
350 {
351 struct rb_root *root = &zmd->mblk_rbtree;
352 struct rb_node *node = root->rb_node;
353 struct dmz_mblock *mblk;
354
355 while (node) {
356 mblk = container_of(node, struct dmz_mblock, node);
357 if (mblk->no == mblk_no) {
358 /*
359 * If this is the first reference to the block,
360 * remove it from the LRU list.
361 */
362 mblk->ref++;
363 if (mblk->ref == 1 &&
364 !test_bit(DMZ_META_DIRTY, &mblk->state))
365 list_del_init(&mblk->link);
366 return mblk;
367 }
368 node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right;
369 }
370
371 return NULL;
372 }
373
374 /*
375 * Metadata block BIO end callback.
376 */
377 static void dmz_mblock_bio_end_io(struct bio *bio)
378 {
379 struct dmz_mblock *mblk = bio->bi_private;
380 int flag;
381
382 if (bio->bi_status)
383 set_bit(DMZ_META_ERROR, &mblk->state);
384
385 if (bio_op(bio) == REQ_OP_WRITE)
386 flag = DMZ_META_WRITING;
387 else
388 flag = DMZ_META_READING;
389
390 clear_bit_unlock(flag, &mblk->state);
391 smp_mb__after_atomic();
392 wake_up_bit(&mblk->state, flag);
393
394 bio_put(bio);
395 }
396
397 /*
398 * Read an uncached metadata block from disk and add it to the cache.
399 */
400 static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd,
401 sector_t mblk_no)
402 {
403 struct dmz_mblock *mblk, *m;
404 sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no;
405 struct bio *bio;
406
407 if (dmz_bdev_is_dying(zmd->dev))
408 return ERR_PTR(-EIO);
409
410 /* Get a new block and a BIO to read it */
411 mblk = dmz_alloc_mblock(zmd, mblk_no);
412 if (!mblk)
413 return ERR_PTR(-ENOMEM);
414
415 bio = bio_alloc(GFP_NOIO, 1);
416 if (!bio) {
417 dmz_free_mblock(zmd, mblk);
418 return ERR_PTR(-ENOMEM);
419 }
420
421 spin_lock(&zmd->mblk_lock);
422
423 /*
424 * Make sure that another context did not start reading
425 * the block already.
426 */
427 m = dmz_get_mblock_fast(zmd, mblk_no);
428 if (m) {
429 spin_unlock(&zmd->mblk_lock);
430 dmz_free_mblock(zmd, mblk);
431 bio_put(bio);
432 return m;
433 }
434
435 mblk->ref++;
436 set_bit(DMZ_META_READING, &mblk->state);
437 dmz_insert_mblock(zmd, mblk);
438
439 spin_unlock(&zmd->mblk_lock);
440
441 /* Submit read BIO */
442 bio->bi_iter.bi_sector = dmz_blk2sect(block);
443 bio_set_dev(bio, zmd->dev->bdev);
444 bio->bi_private = mblk;
445 bio->bi_end_io = dmz_mblock_bio_end_io;
446 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO);
447 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
448 submit_bio(bio);
449
450 return mblk;
451 }
452
453 /*
454 * Free metadata blocks.
455 */
456 static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd,
457 unsigned long limit)
458 {
459 struct dmz_mblock *mblk;
460 unsigned long count = 0;
461
462 if (!zmd->max_nr_mblks)
463 return 0;
464
465 while (!list_empty(&zmd->mblk_lru_list) &&
466 atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks &&
467 count < limit) {
468 mblk = list_first_entry(&zmd->mblk_lru_list,
469 struct dmz_mblock, link);
470 list_del_init(&mblk->link);
471 rb_erase(&mblk->node, &zmd->mblk_rbtree);
472 dmz_free_mblock(zmd, mblk);
473 count++;
474 }
475
476 return count;
477 }
478
479 /*
480 * For mblock shrinker: get the number of unused metadata blocks in the cache.
481 */
482 static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink,
483 struct shrink_control *sc)
484 {
485 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
486
487 return atomic_read(&zmd->nr_mblks);
488 }
489
490 /*
491 * For mblock shrinker: scan unused metadata blocks and shrink the cache.
492 */
493 static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink,
494 struct shrink_control *sc)
495 {
496 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
497 unsigned long count;
498
499 spin_lock(&zmd->mblk_lock);
500 count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan);
501 spin_unlock(&zmd->mblk_lock);
502
503 return count ? count : SHRINK_STOP;
504 }
505
506 /*
507 * Release a metadata block.
508 */
509 static void dmz_release_mblock(struct dmz_metadata *zmd,
510 struct dmz_mblock *mblk)
511 {
512
513 if (!mblk)
514 return;
515
516 spin_lock(&zmd->mblk_lock);
517
518 mblk->ref--;
519 if (mblk->ref == 0) {
520 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
521 rb_erase(&mblk->node, &zmd->mblk_rbtree);
522 dmz_free_mblock(zmd, mblk);
523 } else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) {
524 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
525 dmz_shrink_mblock_cache(zmd, 1);
526 }
527 }
528
529 spin_unlock(&zmd->mblk_lock);
530 }
531
532 /*
533 * Get a metadata block from the rbtree. If the block
534 * is not present, read it from disk.
535 */
536 static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
537 sector_t mblk_no)
538 {
539 struct dmz_mblock *mblk;
540
541 /* Check rbtree */
542 spin_lock(&zmd->mblk_lock);
543 mblk = dmz_get_mblock_fast(zmd, mblk_no);
544 spin_unlock(&zmd->mblk_lock);
545
546 if (!mblk) {
547 /* Cache miss: read the block from disk */
548 mblk = dmz_get_mblock_slow(zmd, mblk_no);
549 if (IS_ERR(mblk))
550 return mblk;
551 }
552
553 /* Wait for on-going read I/O and check for error */
554 wait_on_bit_io(&mblk->state, DMZ_META_READING,
555 TASK_UNINTERRUPTIBLE);
556 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
557 dmz_release_mblock(zmd, mblk);
558 dmz_check_bdev(zmd->dev);
559 return ERR_PTR(-EIO);
560 }
561
562 return mblk;
563 }
564
565 /*
566 * Mark a metadata block dirty.
567 */
568 static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
569 {
570 spin_lock(&zmd->mblk_lock);
571 if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state))
572 list_add_tail(&mblk->link, &zmd->mblk_dirty_list);
573 spin_unlock(&zmd->mblk_lock);
574 }
575
576 /*
577 * Issue a metadata block write BIO.
578 */
579 static int dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk,
580 unsigned int set)
581 {
582 sector_t block = zmd->sb[set].block + mblk->no;
583 struct bio *bio;
584
585 if (dmz_bdev_is_dying(zmd->dev))
586 return -EIO;
587
588 bio = bio_alloc(GFP_NOIO, 1);
589 if (!bio) {
590 set_bit(DMZ_META_ERROR, &mblk->state);
591 return -ENOMEM;
592 }
593
594 set_bit(DMZ_META_WRITING, &mblk->state);
595
596 bio->bi_iter.bi_sector = dmz_blk2sect(block);
597 bio_set_dev(bio, zmd->dev->bdev);
598 bio->bi_private = mblk;
599 bio->bi_end_io = dmz_mblock_bio_end_io;
600 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO);
601 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
602 submit_bio(bio);
603
604 return 0;
605 }
606
607 /*
608 * Read/write a metadata block.
609 */
610 static int dmz_rdwr_block(struct dmz_metadata *zmd, int op, sector_t block,
611 struct page *page)
612 {
613 struct bio *bio;
614 int ret;
615
616 if (dmz_bdev_is_dying(zmd->dev))
617 return -EIO;
618
619 bio = bio_alloc(GFP_NOIO, 1);
620 if (!bio)
621 return -ENOMEM;
622
623 bio->bi_iter.bi_sector = dmz_blk2sect(block);
624 bio_set_dev(bio, zmd->dev->bdev);
625 bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO);
626 bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0);
627 ret = submit_bio_wait(bio);
628 bio_put(bio);
629
630 if (ret)
631 dmz_check_bdev(zmd->dev);
632 return ret;
633 }
634
635 /*
636 * Write super block of the specified metadata set.
637 */
638 static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set)
639 {
640 sector_t block = zmd->sb[set].block;
641 struct dmz_mblock *mblk = zmd->sb[set].mblk;
642 struct dmz_super *sb = zmd->sb[set].sb;
643 u64 sb_gen = zmd->sb_gen + 1;
644 int ret;
645
646 sb->magic = cpu_to_le32(DMZ_MAGIC);
647 sb->version = cpu_to_le32(DMZ_META_VER);
648
649 sb->gen = cpu_to_le64(sb_gen);
650
651 sb->sb_block = cpu_to_le64(block);
652 sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks);
653 sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq);
654 sb->nr_chunks = cpu_to_le32(zmd->nr_chunks);
655
656 sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks);
657 sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks);
658
659 sb->crc = 0;
660 sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE));
661
662 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE, block, mblk->page);
663 if (ret == 0)
664 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
665
666 return ret;
667 }
668
669 /*
670 * Write dirty metadata blocks to the specified set.
671 */
672 static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
673 struct list_head *write_list,
674 unsigned int set)
675 {
676 struct dmz_mblock *mblk;
677 struct blk_plug plug;
678 int ret = 0, nr_mblks_submitted = 0;
679
680 /* Issue writes */
681 blk_start_plug(&plug);
682 list_for_each_entry(mblk, write_list, link) {
683 ret = dmz_write_mblock(zmd, mblk, set);
684 if (ret)
685 break;
686 nr_mblks_submitted++;
687 }
688 blk_finish_plug(&plug);
689
690 /* Wait for completion */
691 list_for_each_entry(mblk, write_list, link) {
692 if (!nr_mblks_submitted)
693 break;
694 wait_on_bit_io(&mblk->state, DMZ_META_WRITING,
695 TASK_UNINTERRUPTIBLE);
696 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
697 clear_bit(DMZ_META_ERROR, &mblk->state);
698 dmz_check_bdev(zmd->dev);
699 ret = -EIO;
700 }
701 nr_mblks_submitted--;
702 }
703
704 /* Flush drive cache (this will also sync data) */
705 if (ret == 0)
706 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
707
708 return ret;
709 }
710
711 /*
712 * Log dirty metadata blocks.
713 */
714 static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd,
715 struct list_head *write_list)
716 {
717 unsigned int log_set = zmd->mblk_primary ^ 0x1;
718 int ret;
719
720 /* Write dirty blocks to the log */
721 ret = dmz_write_dirty_mblocks(zmd, write_list, log_set);
722 if (ret)
723 return ret;
724
725 /*
726 * No error so far: now validate the log by updating the
727 * log index super block generation.
728 */
729 ret = dmz_write_sb(zmd, log_set);
730 if (ret)
731 return ret;
732
733 return 0;
734 }
735
736 /*
737 * Flush dirty metadata blocks.
738 */
739 int dmz_flush_metadata(struct dmz_metadata *zmd)
740 {
741 struct dmz_mblock *mblk;
742 struct list_head write_list;
743 int ret;
744
745 if (WARN_ON(!zmd))
746 return 0;
747
748 INIT_LIST_HEAD(&write_list);
749
750 /*
751 * Make sure that metadata blocks are stable before logging: take
752 * the write lock on the metadata semaphore to prevent target BIOs
753 * from modifying metadata.
754 */
755 down_write(&zmd->mblk_sem);
756
757 /*
758 * This is called from the target flush work and reclaim work.
759 * Concurrent execution is not allowed.
760 */
761 dmz_lock_flush(zmd);
762
763 if (dmz_bdev_is_dying(zmd->dev)) {
764 ret = -EIO;
765 goto out;
766 }
767
768 /* Get dirty blocks */
769 spin_lock(&zmd->mblk_lock);
770 list_splice_init(&zmd->mblk_dirty_list, &write_list);
771 spin_unlock(&zmd->mblk_lock);
772
773 /* If there are no dirty metadata blocks, just flush the device cache */
774 if (list_empty(&write_list)) {
775 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
776 goto err;
777 }
778
779 /*
780 * The primary metadata set is still clean. Keep it this way until
781 * all updates are successful in the secondary set. That is, use
782 * the secondary set as a log.
783 */
784 ret = dmz_log_dirty_mblocks(zmd, &write_list);
785 if (ret)
786 goto err;
787
788 /*
789 * The log is on disk. It is now safe to update in place
790 * in the primary metadata set.
791 */
792 ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
793 if (ret)
794 goto err;
795
796 ret = dmz_write_sb(zmd, zmd->mblk_primary);
797 if (ret)
798 goto err;
799
800 while (!list_empty(&write_list)) {
801 mblk = list_first_entry(&write_list, struct dmz_mblock, link);
802 list_del_init(&mblk->link);
803
804 spin_lock(&zmd->mblk_lock);
805 clear_bit(DMZ_META_DIRTY, &mblk->state);
806 if (mblk->ref == 0)
807 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
808 spin_unlock(&zmd->mblk_lock);
809 }
810
811 zmd->sb_gen++;
812 out:
813 dmz_unlock_flush(zmd);
814 up_write(&zmd->mblk_sem);
815
816 return ret;
817
818 err:
819 if (!list_empty(&write_list)) {
820 spin_lock(&zmd->mblk_lock);
821 list_splice(&write_list, &zmd->mblk_dirty_list);
822 spin_unlock(&zmd->mblk_lock);
823 }
824 if (!dmz_check_bdev(zmd->dev))
825 ret = -EIO;
826 goto out;
827 }
828
829 /*
830 * Check super block.
831 */
832 static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_super *sb)
833 {
834 unsigned int nr_meta_zones, nr_data_zones;
835 struct dmz_dev *dev = zmd->dev;
836 u32 crc, stored_crc;
837 u64 gen;
838
839 gen = le64_to_cpu(sb->gen);
840 stored_crc = le32_to_cpu(sb->crc);
841 sb->crc = 0;
842 crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
843 if (crc != stored_crc) {
844 dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
845 crc, stored_crc);
846 return -ENXIO;
847 }
848
849 if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
850 dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
851 DMZ_MAGIC, le32_to_cpu(sb->magic));
852 return -ENXIO;
853 }
854
855 if (le32_to_cpu(sb->version) != DMZ_META_VER) {
856 dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
857 DMZ_META_VER, le32_to_cpu(sb->version));
858 return -ENXIO;
859 }
860
861 nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + dev->zone_nr_blocks - 1)
862 >> dev->zone_nr_blocks_shift;
863 if (!nr_meta_zones ||
864 nr_meta_zones >= zmd->nr_rnd_zones) {
865 dmz_dev_err(dev, "Invalid number of metadata blocks");
866 return -ENXIO;
867 }
868
869 if (!le32_to_cpu(sb->nr_reserved_seq) ||
870 le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
871 dmz_dev_err(dev, "Invalid number of reserved sequential zones");
872 return -ENXIO;
873 }
874
875 nr_data_zones = zmd->nr_useable_zones -
876 (nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
877 if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
878 dmz_dev_err(dev, "Invalid number of chunks %u / %u",
879 le32_to_cpu(sb->nr_chunks), nr_data_zones);
880 return -ENXIO;
881 }
882
883 /* OK */
884 zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
885 zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
886 zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
887 zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
888 zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
889 zmd->nr_meta_zones = nr_meta_zones;
890 zmd->nr_data_zones = nr_data_zones;
891
892 return 0;
893 }
894
895 /*
896 * Read the first or second super block from disk.
897 */
898 static int dmz_read_sb(struct dmz_metadata *zmd, unsigned int set)
899 {
900 return dmz_rdwr_block(zmd, REQ_OP_READ, zmd->sb[set].block,
901 zmd->sb[set].mblk->page);
902 }
903
904 /*
905 * Determine the position of the secondary super blocks on disk.
906 * This is used only if a corruption of the primary super block
907 * is detected.
908 */
909 static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
910 {
911 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
912 struct dmz_mblock *mblk;
913 int i;
914
915 /* Allocate a block */
916 mblk = dmz_alloc_mblock(zmd, 0);
917 if (!mblk)
918 return -ENOMEM;
919
920 zmd->sb[1].mblk = mblk;
921 zmd->sb[1].sb = mblk->data;
922
923 /* Bad first super block: search for the second one */
924 zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
925 for (i = 0; i < zmd->nr_rnd_zones - 1; i++) {
926 if (dmz_read_sb(zmd, 1) != 0)
927 break;
928 if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
929 return 0;
930 zmd->sb[1].block += zone_nr_blocks;
931 }
932
933 dmz_free_mblock(zmd, mblk);
934 zmd->sb[1].mblk = NULL;
935
936 return -EIO;
937 }
938
939 /*
940 * Read the first or second super block from disk.
941 */
942 static int dmz_get_sb(struct dmz_metadata *zmd, unsigned int set)
943 {
944 struct dmz_mblock *mblk;
945 int ret;
946
947 /* Allocate a block */
948 mblk = dmz_alloc_mblock(zmd, 0);
949 if (!mblk)
950 return -ENOMEM;
951
952 zmd->sb[set].mblk = mblk;
953 zmd->sb[set].sb = mblk->data;
954
955 /* Read super block */
956 ret = dmz_read_sb(zmd, set);
957 if (ret) {
958 dmz_free_mblock(zmd, mblk);
959 zmd->sb[set].mblk = NULL;
960 return ret;
961 }
962
963 return 0;
964 }
965
966 /*
967 * Recover a metadata set.
968 */
969 static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
970 {
971 unsigned int src_set = dst_set ^ 0x1;
972 struct page *page;
973 int i, ret;
974
975 dmz_dev_warn(zmd->dev, "Metadata set %u invalid: recovering", dst_set);
976
977 if (dst_set == 0)
978 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
979 else {
980 zmd->sb[1].block = zmd->sb[0].block +
981 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
982 }
983
984 page = alloc_page(GFP_NOIO);
985 if (!page)
986 return -ENOMEM;
987
988 /* Copy metadata blocks */
989 for (i = 1; i < zmd->nr_meta_blocks; i++) {
990 ret = dmz_rdwr_block(zmd, REQ_OP_READ,
991 zmd->sb[src_set].block + i, page);
992 if (ret)
993 goto out;
994 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE,
995 zmd->sb[dst_set].block + i, page);
996 if (ret)
997 goto out;
998 }
999
1000 /* Finalize with the super block */
1001 if (!zmd->sb[dst_set].mblk) {
1002 zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
1003 if (!zmd->sb[dst_set].mblk) {
1004 ret = -ENOMEM;
1005 goto out;
1006 }
1007 zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
1008 }
1009
1010 ret = dmz_write_sb(zmd, dst_set);
1011 out:
1012 __free_pages(page, 0);
1013
1014 return ret;
1015 }
1016
1017 /*
1018 * Get super block from disk.
1019 */
1020 static int dmz_load_sb(struct dmz_metadata *zmd)
1021 {
1022 bool sb_good[2] = {false, false};
1023 u64 sb_gen[2] = {0, 0};
1024 int ret;
1025
1026 /* Read and check the primary super block */
1027 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
1028 ret = dmz_get_sb(zmd, 0);
1029 if (ret) {
1030 dmz_dev_err(zmd->dev, "Read primary super block failed");
1031 return ret;
1032 }
1033
1034 ret = dmz_check_sb(zmd, zmd->sb[0].sb);
1035
1036 /* Read and check secondary super block */
1037 if (ret == 0) {
1038 sb_good[0] = true;
1039 zmd->sb[1].block = zmd->sb[0].block +
1040 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
1041 ret = dmz_get_sb(zmd, 1);
1042 } else
1043 ret = dmz_lookup_secondary_sb(zmd);
1044
1045 if (ret) {
1046 dmz_dev_err(zmd->dev, "Read secondary super block failed");
1047 return ret;
1048 }
1049
1050 ret = dmz_check_sb(zmd, zmd->sb[1].sb);
1051 if (ret == 0)
1052 sb_good[1] = true;
1053
1054 /* Use highest generation sb first */
1055 if (!sb_good[0] && !sb_good[1]) {
1056 dmz_dev_err(zmd->dev, "No valid super block found");
1057 return -EIO;
1058 }
1059
1060 if (sb_good[0])
1061 sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1062 else
1063 ret = dmz_recover_mblocks(zmd, 0);
1064
1065 if (sb_good[1])
1066 sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1067 else
1068 ret = dmz_recover_mblocks(zmd, 1);
1069
1070 if (ret) {
1071 dmz_dev_err(zmd->dev, "Recovery failed");
1072 return -EIO;
1073 }
1074
1075 if (sb_gen[0] >= sb_gen[1]) {
1076 zmd->sb_gen = sb_gen[0];
1077 zmd->mblk_primary = 0;
1078 } else {
1079 zmd->sb_gen = sb_gen[1];
1080 zmd->mblk_primary = 1;
1081 }
1082
1083 dmz_dev_debug(zmd->dev, "Using super block %u (gen %llu)",
1084 zmd->mblk_primary, zmd->sb_gen);
1085
1086 return 0;
1087 }
1088
1089 /*
1090 * Initialize a zone descriptor.
1091 */
1092 static int dmz_init_zone(struct blk_zone *blkz, unsigned int idx, void *data)
1093 {
1094 struct dmz_metadata *zmd = data;
1095 struct dm_zone *zone = &zmd->zones[idx];
1096 struct dmz_dev *dev = zmd->dev;
1097
1098 /* Ignore the eventual last runt (smaller) zone */
1099 if (blkz->len != dev->zone_nr_sectors) {
1100 if (blkz->start + blkz->len == dev->capacity)
1101 return 0;
1102 return -ENXIO;
1103 }
1104
1105 INIT_LIST_HEAD(&zone->link);
1106 atomic_set(&zone->refcount, 0);
1107 zone->chunk = DMZ_MAP_UNMAPPED;
1108
1109 switch (blkz->type) {
1110 case BLK_ZONE_TYPE_CONVENTIONAL:
1111 set_bit(DMZ_RND, &zone->flags);
1112 zmd->nr_rnd_zones++;
1113 break;
1114 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1115 case BLK_ZONE_TYPE_SEQWRITE_PREF:
1116 set_bit(DMZ_SEQ, &zone->flags);
1117 break;
1118 default:
1119 return -ENXIO;
1120 }
1121
1122 if (dmz_is_rnd(zone))
1123 zone->wp_block = 0;
1124 else
1125 zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1126
1127 if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1128 set_bit(DMZ_OFFLINE, &zone->flags);
1129 else if (blkz->cond == BLK_ZONE_COND_READONLY)
1130 set_bit(DMZ_READ_ONLY, &zone->flags);
1131 else {
1132 zmd->nr_useable_zones++;
1133 if (dmz_is_rnd(zone)) {
1134 zmd->nr_rnd_zones++;
1135 if (!zmd->sb_zone) {
1136 /* Super block zone */
1137 zmd->sb_zone = zone;
1138 }
1139 }
1140 }
1141
1142 return 0;
1143 }
1144
1145 /*
1146 * Free zones descriptors.
1147 */
1148 static void dmz_drop_zones(struct dmz_metadata *zmd)
1149 {
1150 kfree(zmd->zones);
1151 zmd->zones = NULL;
1152 }
1153
1154 /*
1155 * Allocate and initialize zone descriptors using the zone
1156 * information from disk.
1157 */
1158 static int dmz_init_zones(struct dmz_metadata *zmd)
1159 {
1160 struct dmz_dev *dev = zmd->dev;
1161 int ret;
1162
1163 /* Init */
1164 zmd->zone_bitmap_size = dev->zone_nr_blocks >> 3;
1165 zmd->zone_nr_bitmap_blocks =
1166 max_t(sector_t, 1, zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT);
1167 zmd->zone_bits_per_mblk = min_t(sector_t, dev->zone_nr_blocks,
1168 DMZ_BLOCK_SIZE_BITS);
1169
1170 /* Allocate zone array */
1171 zmd->zones = kcalloc(dev->nr_zones, sizeof(struct dm_zone), GFP_KERNEL);
1172 if (!zmd->zones)
1173 return -ENOMEM;
1174
1175 dmz_dev_info(dev, "Using %zu B for zone information",
1176 sizeof(struct dm_zone) * dev->nr_zones);
1177
1178 /*
1179 * Get zone information and initialize zone descriptors. At the same
1180 * time, determine where the super block should be: first block of the
1181 * first randomly writable zone.
1182 */
1183 ret = blkdev_report_zones(dev->bdev, 0, BLK_ALL_ZONES, dmz_init_zone,
1184 zmd);
1185 if (ret < 0) {
1186 dmz_drop_zones(zmd);
1187 return ret;
1188 }
1189
1190 return 0;
1191 }
1192
1193 static int dmz_update_zone_cb(struct blk_zone *blkz, unsigned int idx,
1194 void *data)
1195 {
1196 struct dm_zone *zone = data;
1197
1198 clear_bit(DMZ_OFFLINE, &zone->flags);
1199 clear_bit(DMZ_READ_ONLY, &zone->flags);
1200 if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1201 set_bit(DMZ_OFFLINE, &zone->flags);
1202 else if (blkz->cond == BLK_ZONE_COND_READONLY)
1203 set_bit(DMZ_READ_ONLY, &zone->flags);
1204
1205 if (dmz_is_seq(zone))
1206 zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1207 else
1208 zone->wp_block = 0;
1209 return 0;
1210 }
1211
1212 /*
1213 * Update a zone information.
1214 */
1215 static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1216 {
1217 unsigned int noio_flag;
1218 int ret;
1219
1220 /*
1221 * Get zone information from disk. Since blkdev_report_zones() uses
1222 * GFP_KERNEL by default for memory allocations, set the per-task
1223 * PF_MEMALLOC_NOIO flag so that all allocations are done as if
1224 * GFP_NOIO was specified.
1225 */
1226 noio_flag = memalloc_noio_save();
1227 ret = blkdev_report_zones(zmd->dev->bdev, dmz_start_sect(zmd, zone), 1,
1228 dmz_update_zone_cb, zone);
1229 memalloc_noio_restore(noio_flag);
1230
1231 if (ret == 0)
1232 ret = -EIO;
1233 if (ret < 0) {
1234 dmz_dev_err(zmd->dev, "Get zone %u report failed",
1235 dmz_id(zmd, zone));
1236 dmz_check_bdev(zmd->dev);
1237 return ret;
1238 }
1239
1240 return 0;
1241 }
1242
1243 /*
1244 * Check a zone write pointer position when the zone is marked
1245 * with the sequential write error flag.
1246 */
1247 static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1248 struct dm_zone *zone)
1249 {
1250 unsigned int wp = 0;
1251 int ret;
1252
1253 wp = zone->wp_block;
1254 ret = dmz_update_zone(zmd, zone);
1255 if (ret)
1256 return ret;
1257
1258 dmz_dev_warn(zmd->dev, "Processing zone %u write error (zone wp %u/%u)",
1259 dmz_id(zmd, zone), zone->wp_block, wp);
1260
1261 if (zone->wp_block < wp) {
1262 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1263 wp - zone->wp_block);
1264 }
1265
1266 return 0;
1267 }
1268
1269 static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
1270 {
1271 return &zmd->zones[zone_id];
1272 }
1273
1274 /*
1275 * Reset a zone write pointer.
1276 */
1277 static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1278 {
1279 int ret;
1280
1281 /*
1282 * Ignore offline zones, read only zones,
1283 * and conventional zones.
1284 */
1285 if (dmz_is_offline(zone) ||
1286 dmz_is_readonly(zone) ||
1287 dmz_is_rnd(zone))
1288 return 0;
1289
1290 if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1291 struct dmz_dev *dev = zmd->dev;
1292
1293 ret = blkdev_zone_mgmt(dev->bdev, REQ_OP_ZONE_RESET,
1294 dmz_start_sect(zmd, zone),
1295 dev->zone_nr_sectors, GFP_NOIO);
1296 if (ret) {
1297 dmz_dev_err(dev, "Reset zone %u failed %d",
1298 dmz_id(zmd, zone), ret);
1299 return ret;
1300 }
1301 }
1302
1303 /* Clear write error bit and rewind write pointer position */
1304 clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1305 zone->wp_block = 0;
1306
1307 return 0;
1308 }
1309
1310 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1311
1312 /*
1313 * Initialize chunk mapping.
1314 */
1315 static int dmz_load_mapping(struct dmz_metadata *zmd)
1316 {
1317 struct dmz_dev *dev = zmd->dev;
1318 struct dm_zone *dzone, *bzone;
1319 struct dmz_mblock *dmap_mblk = NULL;
1320 struct dmz_map *dmap;
1321 unsigned int i = 0, e = 0, chunk = 0;
1322 unsigned int dzone_id;
1323 unsigned int bzone_id;
1324
1325 /* Metadata block array for the chunk mapping table */
1326 zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1327 sizeof(struct dmz_mblk *), GFP_KERNEL);
1328 if (!zmd->map_mblk)
1329 return -ENOMEM;
1330
1331 /* Get chunk mapping table blocks and initialize zone mapping */
1332 while (chunk < zmd->nr_chunks) {
1333 if (!dmap_mblk) {
1334 /* Get mapping block */
1335 dmap_mblk = dmz_get_mblock(zmd, i + 1);
1336 if (IS_ERR(dmap_mblk))
1337 return PTR_ERR(dmap_mblk);
1338 zmd->map_mblk[i] = dmap_mblk;
1339 dmap = (struct dmz_map *) dmap_mblk->data;
1340 i++;
1341 e = 0;
1342 }
1343
1344 /* Check data zone */
1345 dzone_id = le32_to_cpu(dmap[e].dzone_id);
1346 if (dzone_id == DMZ_MAP_UNMAPPED)
1347 goto next;
1348
1349 if (dzone_id >= dev->nr_zones) {
1350 dmz_dev_err(dev, "Chunk %u mapping: invalid data zone ID %u",
1351 chunk, dzone_id);
1352 return -EIO;
1353 }
1354
1355 dzone = dmz_get(zmd, dzone_id);
1356 set_bit(DMZ_DATA, &dzone->flags);
1357 dzone->chunk = chunk;
1358 dmz_get_zone_weight(zmd, dzone);
1359
1360 if (dmz_is_rnd(dzone))
1361 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1362 else
1363 list_add_tail(&dzone->link, &zmd->map_seq_list);
1364
1365 /* Check buffer zone */
1366 bzone_id = le32_to_cpu(dmap[e].bzone_id);
1367 if (bzone_id == DMZ_MAP_UNMAPPED)
1368 goto next;
1369
1370 if (bzone_id >= dev->nr_zones) {
1371 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone ID %u",
1372 chunk, bzone_id);
1373 return -EIO;
1374 }
1375
1376 bzone = dmz_get(zmd, bzone_id);
1377 if (!dmz_is_rnd(bzone)) {
1378 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone %u",
1379 chunk, bzone_id);
1380 return -EIO;
1381 }
1382
1383 set_bit(DMZ_DATA, &bzone->flags);
1384 set_bit(DMZ_BUF, &bzone->flags);
1385 bzone->chunk = chunk;
1386 bzone->bzone = dzone;
1387 dzone->bzone = bzone;
1388 dmz_get_zone_weight(zmd, bzone);
1389 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1390 next:
1391 chunk++;
1392 e++;
1393 if (e >= DMZ_MAP_ENTRIES)
1394 dmap_mblk = NULL;
1395 }
1396
1397 /*
1398 * At this point, only meta zones and mapped data zones were
1399 * fully initialized. All remaining zones are unmapped data
1400 * zones. Finish initializing those here.
1401 */
1402 for (i = 0; i < dev->nr_zones; i++) {
1403 dzone = dmz_get(zmd, i);
1404 if (dmz_is_meta(dzone))
1405 continue;
1406
1407 if (dmz_is_rnd(dzone))
1408 zmd->nr_rnd++;
1409 else
1410 zmd->nr_seq++;
1411
1412 if (dmz_is_data(dzone)) {
1413 /* Already initialized */
1414 continue;
1415 }
1416
1417 /* Unmapped data zone */
1418 set_bit(DMZ_DATA, &dzone->flags);
1419 dzone->chunk = DMZ_MAP_UNMAPPED;
1420 if (dmz_is_rnd(dzone)) {
1421 list_add_tail(&dzone->link, &zmd->unmap_rnd_list);
1422 atomic_inc(&zmd->unmap_nr_rnd);
1423 } else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1424 list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1425 atomic_inc(&zmd->nr_reserved_seq_zones);
1426 zmd->nr_seq--;
1427 } else {
1428 list_add_tail(&dzone->link, &zmd->unmap_seq_list);
1429 atomic_inc(&zmd->unmap_nr_seq);
1430 }
1431 }
1432
1433 return 0;
1434 }
1435
1436 /*
1437 * Set a data chunk mapping.
1438 */
1439 static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1440 unsigned int dzone_id, unsigned int bzone_id)
1441 {
1442 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1443 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1444 int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1445
1446 dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1447 dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1448 dmz_dirty_mblock(zmd, dmap_mblk);
1449 }
1450
1451 /*
1452 * The list of mapped zones is maintained in LRU order.
1453 * This rotates a zone at the end of its map list.
1454 */
1455 static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1456 {
1457 if (list_empty(&zone->link))
1458 return;
1459
1460 list_del_init(&zone->link);
1461 if (dmz_is_seq(zone)) {
1462 /* LRU rotate sequential zone */
1463 list_add_tail(&zone->link, &zmd->map_seq_list);
1464 } else {
1465 /* LRU rotate random zone */
1466 list_add_tail(&zone->link, &zmd->map_rnd_list);
1467 }
1468 }
1469
1470 /*
1471 * The list of mapped random zones is maintained
1472 * in LRU order. This rotates a zone at the end of the list.
1473 */
1474 static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1475 {
1476 __dmz_lru_zone(zmd, zone);
1477 if (zone->bzone)
1478 __dmz_lru_zone(zmd, zone->bzone);
1479 }
1480
1481 /*
1482 * Wait for any zone to be freed.
1483 */
1484 static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1485 {
1486 DEFINE_WAIT(wait);
1487
1488 prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1489 dmz_unlock_map(zmd);
1490 dmz_unlock_metadata(zmd);
1491
1492 io_schedule_timeout(HZ);
1493
1494 dmz_lock_metadata(zmd);
1495 dmz_lock_map(zmd);
1496 finish_wait(&zmd->free_wq, &wait);
1497 }
1498
1499 /*
1500 * Lock a zone for reclaim (set the zone RECLAIM bit).
1501 * Returns false if the zone cannot be locked or if it is already locked
1502 * and 1 otherwise.
1503 */
1504 int dmz_lock_zone_reclaim(struct dm_zone *zone)
1505 {
1506 /* Active zones cannot be reclaimed */
1507 if (dmz_is_active(zone))
1508 return 0;
1509
1510 return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1511 }
1512
1513 /*
1514 * Clear a zone reclaim flag.
1515 */
1516 void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1517 {
1518 WARN_ON(dmz_is_active(zone));
1519 WARN_ON(!dmz_in_reclaim(zone));
1520
1521 clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1522 smp_mb__after_atomic();
1523 wake_up_bit(&zone->flags, DMZ_RECLAIM);
1524 }
1525
1526 /*
1527 * Wait for a zone reclaim to complete.
1528 */
1529 static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1530 {
1531 dmz_unlock_map(zmd);
1532 dmz_unlock_metadata(zmd);
1533 wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1534 dmz_lock_metadata(zmd);
1535 dmz_lock_map(zmd);
1536 }
1537
1538 /*
1539 * Select a random write zone for reclaim.
1540 */
1541 static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd)
1542 {
1543 struct dm_zone *dzone = NULL;
1544 struct dm_zone *zone;
1545
1546 if (list_empty(&zmd->map_rnd_list))
1547 return ERR_PTR(-EBUSY);
1548
1549 list_for_each_entry(zone, &zmd->map_rnd_list, link) {
1550 if (dmz_is_buf(zone))
1551 dzone = zone->bzone;
1552 else
1553 dzone = zone;
1554 if (dmz_lock_zone_reclaim(dzone))
1555 return dzone;
1556 }
1557
1558 return ERR_PTR(-EBUSY);
1559 }
1560
1561 /*
1562 * Select a buffered sequential zone for reclaim.
1563 */
1564 static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd)
1565 {
1566 struct dm_zone *zone;
1567
1568 if (list_empty(&zmd->map_seq_list))
1569 return ERR_PTR(-EBUSY);
1570
1571 list_for_each_entry(zone, &zmd->map_seq_list, link) {
1572 if (!zone->bzone)
1573 continue;
1574 if (dmz_lock_zone_reclaim(zone))
1575 return zone;
1576 }
1577
1578 return ERR_PTR(-EBUSY);
1579 }
1580
1581 /*
1582 * Select a zone for reclaim.
1583 */
1584 struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd)
1585 {
1586 struct dm_zone *zone;
1587
1588 /*
1589 * Search for a zone candidate to reclaim: 2 cases are possible.
1590 * (1) There is no free sequential zones. Then a random data zone
1591 * cannot be reclaimed. So choose a sequential zone to reclaim so
1592 * that afterward a random zone can be reclaimed.
1593 * (2) At least one free sequential zone is available, then choose
1594 * the oldest random zone (data or buffer) that can be locked.
1595 */
1596 dmz_lock_map(zmd);
1597 if (list_empty(&zmd->reserved_seq_zones_list))
1598 zone = dmz_get_seq_zone_for_reclaim(zmd);
1599 else
1600 zone = dmz_get_rnd_zone_for_reclaim(zmd);
1601 dmz_unlock_map(zmd);
1602
1603 return zone;
1604 }
1605
1606 /*
1607 * Get the zone mapping a chunk, if the chunk is mapped already.
1608 * If no mapping exist and the operation is WRITE, a zone is
1609 * allocated and used to map the chunk.
1610 * The zone returned will be set to the active state.
1611 */
1612 struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op)
1613 {
1614 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1615 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1616 int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1617 unsigned int dzone_id;
1618 struct dm_zone *dzone = NULL;
1619 int ret = 0;
1620
1621 dmz_lock_map(zmd);
1622 again:
1623 /* Get the chunk mapping */
1624 dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
1625 if (dzone_id == DMZ_MAP_UNMAPPED) {
1626 /*
1627 * Read or discard in unmapped chunks are fine. But for
1628 * writes, we need a mapping, so get one.
1629 */
1630 if (op != REQ_OP_WRITE)
1631 goto out;
1632
1633 /* Allocate a random zone */
1634 dzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1635 if (!dzone) {
1636 if (dmz_bdev_is_dying(zmd->dev)) {
1637 dzone = ERR_PTR(-EIO);
1638 goto out;
1639 }
1640 dmz_wait_for_free_zones(zmd);
1641 goto again;
1642 }
1643
1644 dmz_map_zone(zmd, dzone, chunk);
1645
1646 } else {
1647 /* The chunk is already mapped: get the mapping zone */
1648 dzone = dmz_get(zmd, dzone_id);
1649 if (dzone->chunk != chunk) {
1650 dzone = ERR_PTR(-EIO);
1651 goto out;
1652 }
1653
1654 /* Repair write pointer if the sequential dzone has error */
1655 if (dmz_seq_write_err(dzone)) {
1656 ret = dmz_handle_seq_write_err(zmd, dzone);
1657 if (ret) {
1658 dzone = ERR_PTR(-EIO);
1659 goto out;
1660 }
1661 clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
1662 }
1663 }
1664
1665 /*
1666 * If the zone is being reclaimed, the chunk mapping may change
1667 * to a different zone. So wait for reclaim and retry. Otherwise,
1668 * activate the zone (this will prevent reclaim from touching it).
1669 */
1670 if (dmz_in_reclaim(dzone)) {
1671 dmz_wait_for_reclaim(zmd, dzone);
1672 goto again;
1673 }
1674 dmz_activate_zone(dzone);
1675 dmz_lru_zone(zmd, dzone);
1676 out:
1677 dmz_unlock_map(zmd);
1678
1679 return dzone;
1680 }
1681
1682 /*
1683 * Write and discard change the block validity of data zones and their buffer
1684 * zones. Check here that valid blocks are still present. If all blocks are
1685 * invalid, the zones can be unmapped on the fly without waiting for reclaim
1686 * to do it.
1687 */
1688 void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
1689 {
1690 struct dm_zone *bzone;
1691
1692 dmz_lock_map(zmd);
1693
1694 bzone = dzone->bzone;
1695 if (bzone) {
1696 if (dmz_weight(bzone))
1697 dmz_lru_zone(zmd, bzone);
1698 else {
1699 /* Empty buffer zone: reclaim it */
1700 dmz_unmap_zone(zmd, bzone);
1701 dmz_free_zone(zmd, bzone);
1702 bzone = NULL;
1703 }
1704 }
1705
1706 /* Deactivate the data zone */
1707 dmz_deactivate_zone(dzone);
1708 if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
1709 dmz_lru_zone(zmd, dzone);
1710 else {
1711 /* Unbuffered inactive empty data zone: reclaim it */
1712 dmz_unmap_zone(zmd, dzone);
1713 dmz_free_zone(zmd, dzone);
1714 }
1715
1716 dmz_unlock_map(zmd);
1717 }
1718
1719 /*
1720 * Allocate and map a random zone to buffer a chunk
1721 * already mapped to a sequential zone.
1722 */
1723 struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
1724 struct dm_zone *dzone)
1725 {
1726 struct dm_zone *bzone;
1727
1728 dmz_lock_map(zmd);
1729 again:
1730 bzone = dzone->bzone;
1731 if (bzone)
1732 goto out;
1733
1734 /* Allocate a random zone */
1735 bzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1736 if (!bzone) {
1737 if (dmz_bdev_is_dying(zmd->dev)) {
1738 bzone = ERR_PTR(-EIO);
1739 goto out;
1740 }
1741 dmz_wait_for_free_zones(zmd);
1742 goto again;
1743 }
1744
1745 /* Update the chunk mapping */
1746 dmz_set_chunk_mapping(zmd, dzone->chunk, dmz_id(zmd, dzone),
1747 dmz_id(zmd, bzone));
1748
1749 set_bit(DMZ_BUF, &bzone->flags);
1750 bzone->chunk = dzone->chunk;
1751 bzone->bzone = dzone;
1752 dzone->bzone = bzone;
1753 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1754 out:
1755 dmz_unlock_map(zmd);
1756
1757 return bzone;
1758 }
1759
1760 /*
1761 * Get an unmapped (free) zone.
1762 * This must be called with the mapping lock held.
1763 */
1764 struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags)
1765 {
1766 struct list_head *list;
1767 struct dm_zone *zone;
1768
1769 if (flags & DMZ_ALLOC_RND)
1770 list = &zmd->unmap_rnd_list;
1771 else
1772 list = &zmd->unmap_seq_list;
1773 again:
1774 if (list_empty(list)) {
1775 /*
1776 * No free zone: if this is for reclaim, allow using the
1777 * reserved sequential zones.
1778 */
1779 if (!(flags & DMZ_ALLOC_RECLAIM) ||
1780 list_empty(&zmd->reserved_seq_zones_list))
1781 return NULL;
1782
1783 zone = list_first_entry(&zmd->reserved_seq_zones_list,
1784 struct dm_zone, link);
1785 list_del_init(&zone->link);
1786 atomic_dec(&zmd->nr_reserved_seq_zones);
1787 return zone;
1788 }
1789
1790 zone = list_first_entry(list, struct dm_zone, link);
1791 list_del_init(&zone->link);
1792
1793 if (dmz_is_rnd(zone))
1794 atomic_dec(&zmd->unmap_nr_rnd);
1795 else
1796 atomic_dec(&zmd->unmap_nr_seq);
1797
1798 if (dmz_is_offline(zone)) {
1799 dmz_dev_warn(zmd->dev, "Zone %u is offline", dmz_id(zmd, zone));
1800 zone = NULL;
1801 goto again;
1802 }
1803
1804 return zone;
1805 }
1806
1807 /*
1808 * Free a zone.
1809 * This must be called with the mapping lock held.
1810 */
1811 void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1812 {
1813 /* If this is a sequential zone, reset it */
1814 if (dmz_is_seq(zone))
1815 dmz_reset_zone(zmd, zone);
1816
1817 /* Return the zone to its type unmap list */
1818 if (dmz_is_rnd(zone)) {
1819 list_add_tail(&zone->link, &zmd->unmap_rnd_list);
1820 atomic_inc(&zmd->unmap_nr_rnd);
1821 } else if (atomic_read(&zmd->nr_reserved_seq_zones) <
1822 zmd->nr_reserved_seq) {
1823 list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
1824 atomic_inc(&zmd->nr_reserved_seq_zones);
1825 } else {
1826 list_add_tail(&zone->link, &zmd->unmap_seq_list);
1827 atomic_inc(&zmd->unmap_nr_seq);
1828 }
1829
1830 wake_up_all(&zmd->free_wq);
1831 }
1832
1833 /*
1834 * Map a chunk to a zone.
1835 * This must be called with the mapping lock held.
1836 */
1837 void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
1838 unsigned int chunk)
1839 {
1840 /* Set the chunk mapping */
1841 dmz_set_chunk_mapping(zmd, chunk, dmz_id(zmd, dzone),
1842 DMZ_MAP_UNMAPPED);
1843 dzone->chunk = chunk;
1844 if (dmz_is_rnd(dzone))
1845 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1846 else
1847 list_add_tail(&dzone->link, &zmd->map_seq_list);
1848 }
1849
1850 /*
1851 * Unmap a zone.
1852 * This must be called with the mapping lock held.
1853 */
1854 void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1855 {
1856 unsigned int chunk = zone->chunk;
1857 unsigned int dzone_id;
1858
1859 if (chunk == DMZ_MAP_UNMAPPED) {
1860 /* Already unmapped */
1861 return;
1862 }
1863
1864 if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
1865 /*
1866 * Unmapping the chunk buffer zone: clear only
1867 * the chunk buffer mapping
1868 */
1869 dzone_id = dmz_id(zmd, zone->bzone);
1870 zone->bzone->bzone = NULL;
1871 zone->bzone = NULL;
1872
1873 } else {
1874 /*
1875 * Unmapping the chunk data zone: the zone must
1876 * not be buffered.
1877 */
1878 if (WARN_ON(zone->bzone)) {
1879 zone->bzone->bzone = NULL;
1880 zone->bzone = NULL;
1881 }
1882 dzone_id = DMZ_MAP_UNMAPPED;
1883 }
1884
1885 dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
1886
1887 zone->chunk = DMZ_MAP_UNMAPPED;
1888 list_del_init(&zone->link);
1889 }
1890
1891 /*
1892 * Set @nr_bits bits in @bitmap starting from @bit.
1893 * Return the number of bits changed from 0 to 1.
1894 */
1895 static unsigned int dmz_set_bits(unsigned long *bitmap,
1896 unsigned int bit, unsigned int nr_bits)
1897 {
1898 unsigned long *addr;
1899 unsigned int end = bit + nr_bits;
1900 unsigned int n = 0;
1901
1902 while (bit < end) {
1903 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
1904 ((end - bit) >= BITS_PER_LONG)) {
1905 /* Try to set the whole word at once */
1906 addr = bitmap + BIT_WORD(bit);
1907 if (*addr == 0) {
1908 *addr = ULONG_MAX;
1909 n += BITS_PER_LONG;
1910 bit += BITS_PER_LONG;
1911 continue;
1912 }
1913 }
1914
1915 if (!test_and_set_bit(bit, bitmap))
1916 n++;
1917 bit++;
1918 }
1919
1920 return n;
1921 }
1922
1923 /*
1924 * Get the bitmap block storing the bit for chunk_block in zone.
1925 */
1926 static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
1927 struct dm_zone *zone,
1928 sector_t chunk_block)
1929 {
1930 sector_t bitmap_block = 1 + zmd->nr_map_blocks +
1931 (sector_t)(dmz_id(zmd, zone) * zmd->zone_nr_bitmap_blocks) +
1932 (chunk_block >> DMZ_BLOCK_SHIFT_BITS);
1933
1934 return dmz_get_mblock(zmd, bitmap_block);
1935 }
1936
1937 /*
1938 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
1939 */
1940 int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1941 struct dm_zone *to_zone)
1942 {
1943 struct dmz_mblock *from_mblk, *to_mblk;
1944 sector_t chunk_block = 0;
1945
1946 /* Get the zones bitmap blocks */
1947 while (chunk_block < zmd->dev->zone_nr_blocks) {
1948 from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
1949 if (IS_ERR(from_mblk))
1950 return PTR_ERR(from_mblk);
1951 to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
1952 if (IS_ERR(to_mblk)) {
1953 dmz_release_mblock(zmd, from_mblk);
1954 return PTR_ERR(to_mblk);
1955 }
1956
1957 memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
1958 dmz_dirty_mblock(zmd, to_mblk);
1959
1960 dmz_release_mblock(zmd, to_mblk);
1961 dmz_release_mblock(zmd, from_mblk);
1962
1963 chunk_block += zmd->zone_bits_per_mblk;
1964 }
1965
1966 to_zone->weight = from_zone->weight;
1967
1968 return 0;
1969 }
1970
1971 /*
1972 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
1973 * starting from chunk_block.
1974 */
1975 int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1976 struct dm_zone *to_zone, sector_t chunk_block)
1977 {
1978 unsigned int nr_blocks;
1979 int ret;
1980
1981 /* Get the zones bitmap blocks */
1982 while (chunk_block < zmd->dev->zone_nr_blocks) {
1983 /* Get a valid region from the source zone */
1984 ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
1985 if (ret <= 0)
1986 return ret;
1987
1988 nr_blocks = ret;
1989 ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
1990 if (ret)
1991 return ret;
1992
1993 chunk_block += nr_blocks;
1994 }
1995
1996 return 0;
1997 }
1998
1999 /*
2000 * Validate all the blocks in the range [block..block+nr_blocks-1].
2001 */
2002 int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2003 sector_t chunk_block, unsigned int nr_blocks)
2004 {
2005 unsigned int count, bit, nr_bits;
2006 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
2007 struct dmz_mblock *mblk;
2008 unsigned int n = 0;
2009
2010 dmz_dev_debug(zmd->dev, "=> VALIDATE zone %u, block %llu, %u blocks",
2011 dmz_id(zmd, zone), (unsigned long long)chunk_block,
2012 nr_blocks);
2013
2014 WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2015
2016 while (nr_blocks) {
2017 /* Get bitmap block */
2018 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2019 if (IS_ERR(mblk))
2020 return PTR_ERR(mblk);
2021
2022 /* Set bits */
2023 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2024 nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2025
2026 count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2027 if (count) {
2028 dmz_dirty_mblock(zmd, mblk);
2029 n += count;
2030 }
2031 dmz_release_mblock(zmd, mblk);
2032
2033 nr_blocks -= nr_bits;
2034 chunk_block += nr_bits;
2035 }
2036
2037 if (likely(zone->weight + n <= zone_nr_blocks))
2038 zone->weight += n;
2039 else {
2040 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be <= %u",
2041 dmz_id(zmd, zone), zone->weight,
2042 zone_nr_blocks - n);
2043 zone->weight = zone_nr_blocks;
2044 }
2045
2046 return 0;
2047 }
2048
2049 /*
2050 * Clear nr_bits bits in bitmap starting from bit.
2051 * Return the number of bits cleared.
2052 */
2053 static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2054 {
2055 unsigned long *addr;
2056 int end = bit + nr_bits;
2057 int n = 0;
2058
2059 while (bit < end) {
2060 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2061 ((end - bit) >= BITS_PER_LONG)) {
2062 /* Try to clear whole word at once */
2063 addr = bitmap + BIT_WORD(bit);
2064 if (*addr == ULONG_MAX) {
2065 *addr = 0;
2066 n += BITS_PER_LONG;
2067 bit += BITS_PER_LONG;
2068 continue;
2069 }
2070 }
2071
2072 if (test_and_clear_bit(bit, bitmap))
2073 n++;
2074 bit++;
2075 }
2076
2077 return n;
2078 }
2079
2080 /*
2081 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2082 */
2083 int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2084 sector_t chunk_block, unsigned int nr_blocks)
2085 {
2086 unsigned int count, bit, nr_bits;
2087 struct dmz_mblock *mblk;
2088 unsigned int n = 0;
2089
2090 dmz_dev_debug(zmd->dev, "=> INVALIDATE zone %u, block %llu, %u blocks",
2091 dmz_id(zmd, zone), (u64)chunk_block, nr_blocks);
2092
2093 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2094
2095 while (nr_blocks) {
2096 /* Get bitmap block */
2097 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2098 if (IS_ERR(mblk))
2099 return PTR_ERR(mblk);
2100
2101 /* Clear bits */
2102 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2103 nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2104
2105 count = dmz_clear_bits((unsigned long *)mblk->data,
2106 bit, nr_bits);
2107 if (count) {
2108 dmz_dirty_mblock(zmd, mblk);
2109 n += count;
2110 }
2111 dmz_release_mblock(zmd, mblk);
2112
2113 nr_blocks -= nr_bits;
2114 chunk_block += nr_bits;
2115 }
2116
2117 if (zone->weight >= n)
2118 zone->weight -= n;
2119 else {
2120 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be >= %u",
2121 dmz_id(zmd, zone), zone->weight, n);
2122 zone->weight = 0;
2123 }
2124
2125 return 0;
2126 }
2127
2128 /*
2129 * Get a block bit value.
2130 */
2131 static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2132 sector_t chunk_block)
2133 {
2134 struct dmz_mblock *mblk;
2135 int ret;
2136
2137 WARN_ON(chunk_block >= zmd->dev->zone_nr_blocks);
2138
2139 /* Get bitmap block */
2140 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2141 if (IS_ERR(mblk))
2142 return PTR_ERR(mblk);
2143
2144 /* Get offset */
2145 ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2146 (unsigned long *) mblk->data) != 0;
2147
2148 dmz_release_mblock(zmd, mblk);
2149
2150 return ret;
2151 }
2152
2153 /*
2154 * Return the number of blocks from chunk_block to the first block with a bit
2155 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2156 */
2157 static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2158 sector_t chunk_block, unsigned int nr_blocks,
2159 int set)
2160 {
2161 struct dmz_mblock *mblk;
2162 unsigned int bit, set_bit, nr_bits;
2163 unsigned int zone_bits = zmd->zone_bits_per_mblk;
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, zone_bits - bit);
2179 if (set)
2180 set_bit = find_next_bit(bitmap, zone_bits, bit);
2181 else
2182 set_bit = find_next_zero_bit(bitmap, zone_bits, bit);
2183 dmz_release_mblock(zmd, mblk);
2184
2185 n += set_bit - bit;
2186 if (set_bit < zone_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 */
2200 int 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 */
2220 int 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 */
2241 static 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 */
2269 static 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, zmd->zone_bits_per_mblk - 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 */
2304 static 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 */
2364 int 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;
2469 err:
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 */
2480 void 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 */
2490 int 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 }
Linux with added FPC-III support