?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / drivers / md / dm-zoned-metadata.c
blob5c2bbdf67f25d7bb9d2f16e92e6ec9abe439df53
1 /*
2 * Copyright (C) 2017 Western Digital Corporation or its affiliates.
3 *
4 * This file is released under the GPL.
5 */
6
7 #include "dm-zoned.h"
8
9 #include <linux/module.h>
10 #include <linux/crc32.h>
11
12 #define DM_MSG_PREFIX "zoned metadata"
13
14 /*
15 * Metadata version.
16 */
17 #define DMZ_META_VER 1
18
19 /*
20 * On-disk super block magic.
21 */
22 #define DMZ_MAGIC ((((unsigned int)('D')) << 24) | \
23 (((unsigned int)('Z')) << 16) | \
24 (((unsigned int)('B')) << 8) | \
25 ((unsigned int)('D')))
26
27 /*
28 * On disk super block.
29 * This uses only 512 B but uses on disk a full 4KB block. This block is
30 * followed on disk by the mapping table of chunks to zones and the bitmap
31 * blocks indicating zone block validity.
32 * The overall resulting metadata format is:
33 * (1) Super block (1 block)
34 * (2) Chunk mapping table (nr_map_blocks)
35 * (3) Bitmap blocks (nr_bitmap_blocks)
36 * All metadata blocks are stored in conventional zones, starting from the
37 * the first conventional zone found on disk.
38 */
39 struct dmz_super {
40 /* Magic number */
41 __le32 magic; /* 4 */
42
43 /* Metadata version number */
44 __le32 version; /* 8 */
45
46 /* Generation number */
47 __le64 gen; /* 16 */
48
49 /* This block number */
50 __le64 sb_block; /* 24 */
51
52 /* The number of metadata blocks, including this super block */
53 __le32 nr_meta_blocks; /* 28 */
54
55 /* The number of sequential zones reserved for reclaim */
56 __le32 nr_reserved_seq; /* 32 */
57
58 /* The number of entries in the mapping table */
59 __le32 nr_chunks; /* 36 */
60
61 /* The number of blocks used for the chunk mapping table */
62 __le32 nr_map_blocks; /* 40 */
63
64 /* The number of blocks used for the block bitmaps */
65 __le32 nr_bitmap_blocks; /* 44 */
66
67 /* Checksum */
68 __le32 crc; /* 48 */
69
70 /* Padding to full 512B sector */
71 u8 reserved[464]; /* 512 */
72 };
73
74 /*
75 * Chunk mapping entry: entries are indexed by chunk number
76 * and give the zone ID (dzone_id) mapping the chunk on disk.
77 * This zone may be sequential or random. If it is a sequential
78 * zone, a second zone (bzone_id) used as a write buffer may
79 * also be specified. This second zone will always be a randomly
80 * writeable zone.
81 */
82 struct dmz_map {
83 __le32 dzone_id;
84 __le32 bzone_id;
85 };
86
87 /*
88 * Chunk mapping table metadata: 512 8-bytes entries per 4KB block.
89 */
90 #define DMZ_MAP_ENTRIES (DMZ_BLOCK_SIZE / sizeof(struct dmz_map))
91 #define DMZ_MAP_ENTRIES_SHIFT (ilog2(DMZ_MAP_ENTRIES))
92 #define DMZ_MAP_ENTRIES_MASK (DMZ_MAP_ENTRIES - 1)
93 #define DMZ_MAP_UNMAPPED UINT_MAX
94
95 /*
96 * Meta data block descriptor (for cached metadata blocks).
97 */
98 struct dmz_mblock {
99 struct rb_node node;
100 struct list_head link;
101 sector_t no;
102 unsigned int ref;
103 unsigned long state;
104 struct page *page;
105 void *data;
106 };
107
108 /*
109 * Metadata block state flags.
110 */
111 enum {
112 DMZ_META_DIRTY,
113 DMZ_META_READING,
114 DMZ_META_WRITING,
115 DMZ_META_ERROR,
116 };
117
118 /*
119 * Super block information (one per metadata set).
120 */
121 struct dmz_sb {
122 sector_t block;
123 struct dmz_mblock *mblk;
124 struct dmz_super *sb;
125 };
126
127 /*
128 * In-memory metadata.
129 */
130 struct dmz_metadata {
131 struct dmz_dev *dev;
132
133 sector_t zone_bitmap_size;
134 unsigned int zone_nr_bitmap_blocks;
135 unsigned int zone_bits_per_mblk;
136
137 unsigned int nr_bitmap_blocks;
138 unsigned int nr_map_blocks;
139
140 unsigned int nr_useable_zones;
141 unsigned int nr_meta_blocks;
142 unsigned int nr_meta_zones;
143 unsigned int nr_data_zones;
144 unsigned int nr_rnd_zones;
145 unsigned int nr_reserved_seq;
146 unsigned int nr_chunks;
147
148 /* Zone information array */
149 struct dm_zone *zones;
150
151 struct dm_zone *sb_zone;
152 struct dmz_sb sb[2];
153 unsigned int mblk_primary;
154 u64 sb_gen;
155 unsigned int min_nr_mblks;
156 unsigned int max_nr_mblks;
157 atomic_t nr_mblks;
158 struct rw_semaphore mblk_sem;
159 struct mutex mblk_flush_lock;
160 spinlock_t mblk_lock;
161 struct rb_root mblk_rbtree;
162 struct list_head mblk_lru_list;
163 struct list_head mblk_dirty_list;
164 struct shrinker mblk_shrinker;
165
166 /* Zone allocation management */
167 struct mutex map_lock;
168 struct dmz_mblock **map_mblk;
169 unsigned int nr_rnd;
170 atomic_t unmap_nr_rnd;
171 struct list_head unmap_rnd_list;
172 struct list_head map_rnd_list;
173
174 unsigned int nr_seq;
175 atomic_t unmap_nr_seq;
176 struct list_head unmap_seq_list;
177 struct list_head map_seq_list;
178
179 atomic_t nr_reserved_seq_zones;
180 struct list_head reserved_seq_zones_list;
181
182 wait_queue_head_t free_wq;
183 };
184
185 /*
186 * Various accessors
187 */
188 unsigned int dmz_id(struct dmz_metadata *zmd, struct dm_zone *zone)
189 {
190 return ((unsigned int)(zone - zmd->zones));
191 }
192
193 sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone)
194 {
195 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_sectors_shift;
196 }
197
198 sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone)
199 {
200 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_blocks_shift;
201 }
202
203 unsigned int dmz_nr_chunks(struct dmz_metadata *zmd)
204 {
205 return zmd->nr_chunks;
206 }
207
208 unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd)
209 {
210 return zmd->nr_rnd;
211 }
212
213 unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd)
214 {
215 return atomic_read(&zmd->unmap_nr_rnd);
216 }
217
218 /*
219 * Lock/unlock mapping table.
220 * The map lock also protects all the zone lists.
221 */
222 void dmz_lock_map(struct dmz_metadata *zmd)
223 {
224 mutex_lock(&zmd->map_lock);
225 }
226
227 void dmz_unlock_map(struct dmz_metadata *zmd)
228 {
229 mutex_unlock(&zmd->map_lock);
230 }
231
232 /*
233 * Lock/unlock metadata access. This is a "read" lock on a semaphore
234 * that prevents metadata flush from running while metadata are being
235 * modified. The actual metadata write mutual exclusion is achieved with
236 * the map lock and zone styate management (active and reclaim state are
237 * mutually exclusive).
238 */
239 void dmz_lock_metadata(struct dmz_metadata *zmd)
240 {
241 down_read(&zmd->mblk_sem);
242 }
243
244 void dmz_unlock_metadata(struct dmz_metadata *zmd)
245 {
246 up_read(&zmd->mblk_sem);
247 }
248
249 /*
250 * Lock/unlock flush: prevent concurrent executions
251 * of dmz_flush_metadata as well as metadata modification in reclaim
252 * while flush is being executed.
253 */
254 void dmz_lock_flush(struct dmz_metadata *zmd)
255 {
256 mutex_lock(&zmd->mblk_flush_lock);
257 }
258
259 void dmz_unlock_flush(struct dmz_metadata *zmd)
260 {
261 mutex_unlock(&zmd->mblk_flush_lock);
262 }
263
264 /*
265 * Allocate a metadata block.
266 */
267 static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd,
268 sector_t mblk_no)
269 {
270 struct dmz_mblock *mblk = NULL;
271
272 /* See if we can reuse cached blocks */
273 if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) {
274 spin_lock(&zmd->mblk_lock);
275 mblk = list_first_entry_or_null(&zmd->mblk_lru_list,
276 struct dmz_mblock, link);
277 if (mblk) {
278 list_del_init(&mblk->link);
279 rb_erase(&mblk->node, &zmd->mblk_rbtree);
280 mblk->no = mblk_no;
281 }
282 spin_unlock(&zmd->mblk_lock);
283 if (mblk)
284 return mblk;
285 }
286
287 /* Allocate a new block */
288 mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO);
289 if (!mblk)
290 return NULL;
291
292 mblk->page = alloc_page(GFP_NOIO);
293 if (!mblk->page) {
294 kfree(mblk);
295 return NULL;
296 }
297
298 RB_CLEAR_NODE(&mblk->node);
299 INIT_LIST_HEAD(&mblk->link);
300 mblk->ref = 0;
301 mblk->state = 0;
302 mblk->no = mblk_no;
303 mblk->data = page_address(mblk->page);
304
305 atomic_inc(&zmd->nr_mblks);
306
307 return mblk;
308 }
309
310 /*
311 * Free a metadata block.
312 */
313 static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
314 {
315 __free_pages(mblk->page, 0);
316 kfree(mblk);
317
318 atomic_dec(&zmd->nr_mblks);
319 }
320
321 /*
322 * Insert a metadata block in the rbtree.
323 */
324 static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
325 {
326 struct rb_root *root = &zmd->mblk_rbtree;
327 struct rb_node **new = &(root->rb_node), *parent = NULL;
328 struct dmz_mblock *b;
329
330 /* Figure out where to put the new node */
331 while (*new) {
332 b = container_of(*new, struct dmz_mblock, node);
333 parent = *new;
334 new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right);
335 }
336
337 /* Add new node and rebalance tree */
338 rb_link_node(&mblk->node, parent, new);
339 rb_insert_color(&mblk->node, root);
340 }
341
342 /*
343 * Lookup a metadata block in the rbtree. If the block is found, increment
344 * its reference count.
345 */
346 static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd,
347 sector_t mblk_no)
348 {
349 struct rb_root *root = &zmd->mblk_rbtree;
350 struct rb_node *node = root->rb_node;
351 struct dmz_mblock *mblk;
352
353 while (node) {
354 mblk = container_of(node, struct dmz_mblock, node);
355 if (mblk->no == mblk_no) {
356 /*
357 * If this is the first reference to the block,
358 * remove it from the LRU list.
359 */
360 mblk->ref++;
361 if (mblk->ref == 1 &&
362 !test_bit(DMZ_META_DIRTY, &mblk->state))
363 list_del_init(&mblk->link);
364 return mblk;
365 }
366 node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right;
367 }
368
369 return NULL;
370 }
371
372 /*
373 * Metadata block BIO end callback.
374 */
375 static void dmz_mblock_bio_end_io(struct bio *bio)
376 {
377 struct dmz_mblock *mblk = bio->bi_private;
378 int flag;
379
380 if (bio->bi_status)
381 set_bit(DMZ_META_ERROR, &mblk->state);
382
383 if (bio_op(bio) == REQ_OP_WRITE)
384 flag = DMZ_META_WRITING;
385 else
386 flag = DMZ_META_READING;
387
388 clear_bit_unlock(flag, &mblk->state);
389 smp_mb__after_atomic();
390 wake_up_bit(&mblk->state, flag);
391
392 bio_put(bio);
393 }
394
395 /*
396 * Read an uncached metadata block from disk and add it to the cache.
397 */
398 static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd,
399 sector_t mblk_no)
400 {
401 struct dmz_mblock *mblk, *m;
402 sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no;
403 struct bio *bio;
404
405 if (dmz_bdev_is_dying(zmd->dev))
406 return ERR_PTR(-EIO);
407
408 /* Get a new block and a BIO to read it */
409 mblk = dmz_alloc_mblock(zmd, mblk_no);
410 if (!mblk)
411 return ERR_PTR(-ENOMEM);
412
413 bio = bio_alloc(GFP_NOIO, 1);
414 if (!bio) {
415 dmz_free_mblock(zmd, mblk);
416 return ERR_PTR(-ENOMEM);
417 }
418
419 spin_lock(&zmd->mblk_lock);
420
421 /*
422 * Make sure that another context did not start reading
423 * the block already.
424 */
425 m = dmz_get_mblock_fast(zmd, mblk_no);
426 if (m) {
427 spin_unlock(&zmd->mblk_lock);
428 dmz_free_mblock(zmd, mblk);
429 bio_put(bio);
430 return m;
431 }
432
433 mblk->ref++;
434 set_bit(DMZ_META_READING, &mblk->state);
435 dmz_insert_mblock(zmd, mblk);
436
437 spin_unlock(&zmd->mblk_lock);
438
439 /* Submit read BIO */
440 bio->bi_iter.bi_sector = dmz_blk2sect(block);
441 bio_set_dev(bio, zmd->dev->bdev);
442 bio->bi_private = mblk;
443 bio->bi_end_io = dmz_mblock_bio_end_io;
444 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO);
445 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
446 submit_bio(bio);
447
448 return mblk;
449 }
450
451 /*
452 * Free metadata blocks.
453 */
454 static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd,
455 unsigned long limit)
456 {
457 struct dmz_mblock *mblk;
458 unsigned long count = 0;
459
460 if (!zmd->max_nr_mblks)
461 return 0;
462
463 while (!list_empty(&zmd->mblk_lru_list) &&
464 atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks &&
465 count < limit) {
466 mblk = list_first_entry(&zmd->mblk_lru_list,
467 struct dmz_mblock, link);
468 list_del_init(&mblk->link);
469 rb_erase(&mblk->node, &zmd->mblk_rbtree);
470 dmz_free_mblock(zmd, mblk);
471 count++;
472 }
473
474 return count;
475 }
476
477 /*
478 * For mblock shrinker: get the number of unused metadata blocks in the cache.
479 */
480 static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink,
481 struct shrink_control *sc)
482 {
483 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
484
485 return atomic_read(&zmd->nr_mblks);
486 }
487
488 /*
489 * For mblock shrinker: scan unused metadata blocks and shrink the cache.
490 */
491 static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink,
492 struct shrink_control *sc)
493 {
494 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
495 unsigned long count;
496
497 spin_lock(&zmd->mblk_lock);
498 count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan);
499 spin_unlock(&zmd->mblk_lock);
500
501 return count ? count : SHRINK_STOP;
502 }
503
504 /*
505 * Release a metadata block.
506 */
507 static void dmz_release_mblock(struct dmz_metadata *zmd,
508 struct dmz_mblock *mblk)
509 {
510
511 if (!mblk)
512 return;
513
514 spin_lock(&zmd->mblk_lock);
515
516 mblk->ref--;
517 if (mblk->ref == 0) {
518 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
519 rb_erase(&mblk->node, &zmd->mblk_rbtree);
520 dmz_free_mblock(zmd, mblk);
521 } else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) {
522 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
523 dmz_shrink_mblock_cache(zmd, 1);
524 }
525 }
526
527 spin_unlock(&zmd->mblk_lock);
528 }
529
530 /*
531 * Get a metadata block from the rbtree. If the block
532 * is not present, read it from disk.
533 */
534 static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
535 sector_t mblk_no)
536 {
537 struct dmz_mblock *mblk;
538
539 /* Check rbtree */
540 spin_lock(&zmd->mblk_lock);
541 mblk = dmz_get_mblock_fast(zmd, mblk_no);
542 spin_unlock(&zmd->mblk_lock);
543
544 if (!mblk) {
545 /* Cache miss: read the block from disk */
546 mblk = dmz_get_mblock_slow(zmd, mblk_no);
547 if (IS_ERR(mblk))
548 return mblk;
549 }
550
551 /* Wait for on-going read I/O and check for error */
552 wait_on_bit_io(&mblk->state, DMZ_META_READING,
553 TASK_UNINTERRUPTIBLE);
554 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
555 dmz_release_mblock(zmd, mblk);
556 dmz_check_bdev(zmd->dev);
557 return ERR_PTR(-EIO);
558 }
559
560 return mblk;
561 }
562
563 /*
564 * Mark a metadata block dirty.
565 */
566 static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
567 {
568 spin_lock(&zmd->mblk_lock);
569 if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state))
570 list_add_tail(&mblk->link, &zmd->mblk_dirty_list);
571 spin_unlock(&zmd->mblk_lock);
572 }
573
574 /*
575 * Issue a metadata block write BIO.
576 */
577 static int dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk,
578 unsigned int set)
579 {
580 sector_t block = zmd->sb[set].block + mblk->no;
581 struct bio *bio;
582
583 if (dmz_bdev_is_dying(zmd->dev))
584 return -EIO;
585
586 bio = bio_alloc(GFP_NOIO, 1);
587 if (!bio) {
588 set_bit(DMZ_META_ERROR, &mblk->state);
589 return -ENOMEM;
590 }
591
592 set_bit(DMZ_META_WRITING, &mblk->state);
593
594 bio->bi_iter.bi_sector = dmz_blk2sect(block);
595 bio_set_dev(bio, zmd->dev->bdev);
596 bio->bi_private = mblk;
597 bio->bi_end_io = dmz_mblock_bio_end_io;
598 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO);
599 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
600 submit_bio(bio);
601
602 return 0;
603 }
604
605 /*
606 * Read/write a metadata block.
607 */
608 static int dmz_rdwr_block(struct dmz_metadata *zmd, int op, sector_t block,
609 struct page *page)
610 {
611 struct bio *bio;
612 int ret;
613
614 if (dmz_bdev_is_dying(zmd->dev))
615 return -EIO;
616
617 bio = bio_alloc(GFP_NOIO, 1);
618 if (!bio)
619 return -ENOMEM;
620
621 bio->bi_iter.bi_sector = dmz_blk2sect(block);
622 bio_set_dev(bio, zmd->dev->bdev);
623 bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO);
624 bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0);
625 ret = submit_bio_wait(bio);
626 bio_put(bio);
627
628 if (ret)
629 dmz_check_bdev(zmd->dev);
630 return ret;
631 }
632
633 /*
634 * Write super block of the specified metadata set.
635 */
636 static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set)
637 {
638 sector_t block = zmd->sb[set].block;
639 struct dmz_mblock *mblk = zmd->sb[set].mblk;
640 struct dmz_super *sb = zmd->sb[set].sb;
641 u64 sb_gen = zmd->sb_gen + 1;
642 int ret;
643
644 sb->magic = cpu_to_le32(DMZ_MAGIC);
645 sb->version = cpu_to_le32(DMZ_META_VER);
646
647 sb->gen = cpu_to_le64(sb_gen);
648
649 sb->sb_block = cpu_to_le64(block);
650 sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks);
651 sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq);
652 sb->nr_chunks = cpu_to_le32(zmd->nr_chunks);
653
654 sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks);
655 sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks);
656
657 sb->crc = 0;
658 sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE));
659
660 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE, block, mblk->page);
661 if (ret == 0)
662 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
663
664 return ret;
665 }
666
667 /*
668 * Write dirty metadata blocks to the specified set.
669 */
670 static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
671 struct list_head *write_list,
672 unsigned int set)
673 {
674 struct dmz_mblock *mblk;
675 struct blk_plug plug;
676 int ret = 0, nr_mblks_submitted = 0;
677
678 /* Issue writes */
679 blk_start_plug(&plug);
680 list_for_each_entry(mblk, write_list, link) {
681 ret = dmz_write_mblock(zmd, mblk, set);
682 if (ret)
683 break;
684 nr_mblks_submitted++;
685 }
686 blk_finish_plug(&plug);
687
688 /* Wait for completion */
689 list_for_each_entry(mblk, write_list, link) {
690 if (!nr_mblks_submitted)
691 break;
692 wait_on_bit_io(&mblk->state, DMZ_META_WRITING,
693 TASK_UNINTERRUPTIBLE);
694 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
695 clear_bit(DMZ_META_ERROR, &mblk->state);
696 dmz_check_bdev(zmd->dev);
697 ret = -EIO;
698 }
699 nr_mblks_submitted--;
700 }
701
702 /* Flush drive cache (this will also sync data) */
703 if (ret == 0)
704 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
705
706 return ret;
707 }
708
709 /*
710 * Log dirty metadata blocks.
711 */
712 static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd,
713 struct list_head *write_list)
714 {
715 unsigned int log_set = zmd->mblk_primary ^ 0x1;
716 int ret;
717
718 /* Write dirty blocks to the log */
719 ret = dmz_write_dirty_mblocks(zmd, write_list, log_set);
720 if (ret)
721 return ret;
722
723 /*
724 * No error so far: now validate the log by updating the
725 * log index super block generation.
726 */
727 ret = dmz_write_sb(zmd, log_set);
728 if (ret)
729 return ret;
730
731 return 0;
732 }
733
734 /*
735 * Flush dirty metadata blocks.
736 */
737 int dmz_flush_metadata(struct dmz_metadata *zmd)
738 {
739 struct dmz_mblock *mblk;
740 struct list_head write_list;
741 int ret;
742
743 if (WARN_ON(!zmd))
744 return 0;
745
746 INIT_LIST_HEAD(&write_list);
747
748 /*
749 * Make sure that metadata blocks are stable before logging: take
750 * the write lock on the metadata semaphore to prevent target BIOs
751 * from modifying metadata.
752 */
753 down_write(&zmd->mblk_sem);
754
755 /*
756 * This is called from the target flush work and reclaim work.
757 * Concurrent execution is not allowed.
758 */
759 dmz_lock_flush(zmd);
760
761 if (dmz_bdev_is_dying(zmd->dev)) {
762 ret = -EIO;
763 goto out;
764 }
765
766 /* Get dirty blocks */
767 spin_lock(&zmd->mblk_lock);
768 list_splice_init(&zmd->mblk_dirty_list, &write_list);
769 spin_unlock(&zmd->mblk_lock);
770
771 /* If there are no dirty metadata blocks, just flush the device cache */
772 if (list_empty(&write_list)) {
773 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
774 goto err;
775 }
776
777 /*
778 * The primary metadata set is still clean. Keep it this way until
779 * all updates are successful in the secondary set. That is, use
780 * the secondary set as a log.
781 */
782 ret = dmz_log_dirty_mblocks(zmd, &write_list);
783 if (ret)
784 goto err;
785
786 /*
787 * The log is on disk. It is now safe to update in place
788 * in the primary metadata set.
789 */
790 ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
791 if (ret)
792 goto err;
793
794 ret = dmz_write_sb(zmd, zmd->mblk_primary);
795 if (ret)
796 goto err;
797
798 while (!list_empty(&write_list)) {
799 mblk = list_first_entry(&write_list, struct dmz_mblock, link);
800 list_del_init(&mblk->link);
801
802 spin_lock(&zmd->mblk_lock);
803 clear_bit(DMZ_META_DIRTY, &mblk->state);
804 if (mblk->ref == 0)
805 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
806 spin_unlock(&zmd->mblk_lock);
807 }
808
809 zmd->sb_gen++;
810 out:
811 dmz_unlock_flush(zmd);
812 up_write(&zmd->mblk_sem);
813
814 return ret;
815
816 err:
817 if (!list_empty(&write_list)) {
818 spin_lock(&zmd->mblk_lock);
819 list_splice(&write_list, &zmd->mblk_dirty_list);
820 spin_unlock(&zmd->mblk_lock);
821 }
822 if (!dmz_check_bdev(zmd->dev))
823 ret = -EIO;
824 goto out;
825 }
826
827 /*
828 * Check super block.
829 */
830 static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_super *sb)
831 {
832 unsigned int nr_meta_zones, nr_data_zones;
833 struct dmz_dev *dev = zmd->dev;
834 u32 crc, stored_crc;
835 u64 gen;
836
837 gen = le64_to_cpu(sb->gen);
838 stored_crc = le32_to_cpu(sb->crc);
839 sb->crc = 0;
840 crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
841 if (crc != stored_crc) {
842 dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
843 crc, stored_crc);
844 return -ENXIO;
845 }
846
847 if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
848 dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
849 DMZ_MAGIC, le32_to_cpu(sb->magic));
850 return -ENXIO;
851 }
852
853 if (le32_to_cpu(sb->version) != DMZ_META_VER) {
854 dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
855 DMZ_META_VER, le32_to_cpu(sb->version));
856 return -ENXIO;
857 }
858
859 nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + dev->zone_nr_blocks - 1)
860 >> dev->zone_nr_blocks_shift;
861 if (!nr_meta_zones ||
862 nr_meta_zones >= zmd->nr_rnd_zones) {
863 dmz_dev_err(dev, "Invalid number of metadata blocks");
864 return -ENXIO;
865 }
866
867 if (!le32_to_cpu(sb->nr_reserved_seq) ||
868 le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
869 dmz_dev_err(dev, "Invalid number of reserved sequential zones");
870 return -ENXIO;
871 }
872
873 nr_data_zones = zmd->nr_useable_zones -
874 (nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
875 if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
876 dmz_dev_err(dev, "Invalid number of chunks %u / %u",
877 le32_to_cpu(sb->nr_chunks), nr_data_zones);
878 return -ENXIO;
879 }
880
881 /* OK */
882 zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
883 zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
884 zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
885 zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
886 zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
887 zmd->nr_meta_zones = nr_meta_zones;
888 zmd->nr_data_zones = nr_data_zones;
889
890 return 0;
891 }
892
893 /*
894 * Read the first or second super block from disk.
895 */
896 static int dmz_read_sb(struct dmz_metadata *zmd, unsigned int set)
897 {
898 return dmz_rdwr_block(zmd, REQ_OP_READ, zmd->sb[set].block,
899 zmd->sb[set].mblk->page);
900 }
901
902 /*
903 * Determine the position of the secondary super blocks on disk.
904 * This is used only if a corruption of the primary super block
905 * is detected.
906 */
907 static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
908 {
909 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
910 struct dmz_mblock *mblk;
911 int i;
912
913 /* Allocate a block */
914 mblk = dmz_alloc_mblock(zmd, 0);
915 if (!mblk)
916 return -ENOMEM;
917
918 zmd->sb[1].mblk = mblk;
919 zmd->sb[1].sb = mblk->data;
920
921 /* Bad first super block: search for the second one */
922 zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
923 for (i = 0; i < zmd->nr_rnd_zones - 1; i++) {
924 if (dmz_read_sb(zmd, 1) != 0)
925 break;
926 if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
927 return 0;
928 zmd->sb[1].block += zone_nr_blocks;
929 }
930
931 dmz_free_mblock(zmd, mblk);
932 zmd->sb[1].mblk = NULL;
933
934 return -EIO;
935 }
936
937 /*
938 * Read the first or second super block from disk.
939 */
940 static int dmz_get_sb(struct dmz_metadata *zmd, unsigned int set)
941 {
942 struct dmz_mblock *mblk;
943 int ret;
944
945 /* Allocate a block */
946 mblk = dmz_alloc_mblock(zmd, 0);
947 if (!mblk)
948 return -ENOMEM;
949
950 zmd->sb[set].mblk = mblk;
951 zmd->sb[set].sb = mblk->data;
952
953 /* Read super block */
954 ret = dmz_read_sb(zmd, set);
955 if (ret) {
956 dmz_free_mblock(zmd, mblk);
957 zmd->sb[set].mblk = NULL;
958 return ret;
959 }
960
961 return 0;
962 }
963
964 /*
965 * Recover a metadata set.
966 */
967 static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
968 {
969 unsigned int src_set = dst_set ^ 0x1;
970 struct page *page;
971 int i, ret;
972
973 dmz_dev_warn(zmd->dev, "Metadata set %u invalid: recovering", dst_set);
974
975 if (dst_set == 0)
976 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
977 else {
978 zmd->sb[1].block = zmd->sb[0].block +
979 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
980 }
981
982 page = alloc_page(GFP_NOIO);
983 if (!page)
984 return -ENOMEM;
985
986 /* Copy metadata blocks */
987 for (i = 1; i < zmd->nr_meta_blocks; i++) {
988 ret = dmz_rdwr_block(zmd, REQ_OP_READ,
989 zmd->sb[src_set].block + i, page);
990 if (ret)
991 goto out;
992 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE,
993 zmd->sb[dst_set].block + i, page);
994 if (ret)
995 goto out;
996 }
997
998 /* Finalize with the super block */
999 if (!zmd->sb[dst_set].mblk) {
1000 zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
1001 if (!zmd->sb[dst_set].mblk) {
1002 ret = -ENOMEM;
1003 goto out;
1004 }
1005 zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
1006 }
1007
1008 ret = dmz_write_sb(zmd, dst_set);
1009 out:
1010 __free_pages(page, 0);
1011
1012 return ret;
1013 }
1014
1015 /*
1016 * Get super block from disk.
1017 */
1018 static int dmz_load_sb(struct dmz_metadata *zmd)
1019 {
1020 bool sb_good[2] = {false, false};
1021 u64 sb_gen[2] = {0, 0};
1022 int ret;
1023
1024 /* Read and check the primary super block */
1025 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
1026 ret = dmz_get_sb(zmd, 0);
1027 if (ret) {
1028 dmz_dev_err(zmd->dev, "Read primary super block failed");
1029 return ret;
1030 }
1031
1032 ret = dmz_check_sb(zmd, zmd->sb[0].sb);
1033
1034 /* Read and check secondary super block */
1035 if (ret == 0) {
1036 sb_good[0] = true;
1037 zmd->sb[1].block = zmd->sb[0].block +
1038 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
1039 ret = dmz_get_sb(zmd, 1);
1040 } else
1041 ret = dmz_lookup_secondary_sb(zmd);
1042
1043 if (ret) {
1044 dmz_dev_err(zmd->dev, "Read secondary super block failed");
1045 return ret;
1046 }
1047
1048 ret = dmz_check_sb(zmd, zmd->sb[1].sb);
1049 if (ret == 0)
1050 sb_good[1] = true;
1051
1052 /* Use highest generation sb first */
1053 if (!sb_good[0] && !sb_good[1]) {
1054 dmz_dev_err(zmd->dev, "No valid super block found");
1055 return -EIO;
1056 }
1057
1058 if (sb_good[0])
1059 sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1060 else
1061 ret = dmz_recover_mblocks(zmd, 0);
1062
1063 if (sb_good[1])
1064 sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1065 else
1066 ret = dmz_recover_mblocks(zmd, 1);
1067
1068 if (ret) {
1069 dmz_dev_err(zmd->dev, "Recovery failed");
1070 return -EIO;
1071 }
1072
1073 if (sb_gen[0] >= sb_gen[1]) {
1074 zmd->sb_gen = sb_gen[0];
1075 zmd->mblk_primary = 0;
1076 } else {
1077 zmd->sb_gen = sb_gen[1];
1078 zmd->mblk_primary = 1;
1079 }
1080
1081 dmz_dev_debug(zmd->dev, "Using super block %u (gen %llu)",
1082 zmd->mblk_primary, zmd->sb_gen);
1083
1084 return 0;
1085 }
1086
1087 /*
1088 * Initialize a zone descriptor.
1089 */
1090 static int dmz_init_zone(struct dmz_metadata *zmd, struct dm_zone *zone,
1091 struct blk_zone *blkz)
1092 {
1093 struct dmz_dev *dev = zmd->dev;
1094
1095 /* Ignore the eventual last runt (smaller) zone */
1096 if (blkz->len != dev->zone_nr_sectors) {
1097 if (blkz->start + blkz->len == dev->capacity)
1098 return 0;
1099 return -ENXIO;
1100 }
1101
1102 INIT_LIST_HEAD(&zone->link);
1103 atomic_set(&zone->refcount, 0);
1104 zone->chunk = DMZ_MAP_UNMAPPED;
1105
1106 if (blkz->type == BLK_ZONE_TYPE_CONVENTIONAL) {
1107 set_bit(DMZ_RND, &zone->flags);
1108 } else if (blkz->type == BLK_ZONE_TYPE_SEQWRITE_REQ ||
1109 blkz->type == BLK_ZONE_TYPE_SEQWRITE_PREF) {
1110 set_bit(DMZ_SEQ, &zone->flags);
1111 } else
1112 return -ENXIO;
1113
1114 if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1115 set_bit(DMZ_OFFLINE, &zone->flags);
1116 else if (blkz->cond == BLK_ZONE_COND_READONLY)
1117 set_bit(DMZ_READ_ONLY, &zone->flags);
1118
1119 if (dmz_is_rnd(zone))
1120 zone->wp_block = 0;
1121 else
1122 zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1123
1124 if (!dmz_is_offline(zone) && !dmz_is_readonly(zone)) {
1125 zmd->nr_useable_zones++;
1126 if (dmz_is_rnd(zone)) {
1127 zmd->nr_rnd_zones++;
1128 if (!zmd->sb_zone) {
1129 /* Super block zone */
1130 zmd->sb_zone = zone;
1131 }
1132 }
1133 }
1134
1135 return 0;
1136 }
1137
1138 /*
1139 * Free zones descriptors.
1140 */
1141 static void dmz_drop_zones(struct dmz_metadata *zmd)
1142 {
1143 kfree(zmd->zones);
1144 zmd->zones = NULL;
1145 }
1146
1147 /*
1148 * The size of a zone report in number of zones.
1149 * This results in 4096*64B=256KB report zones commands.
1150 */
1151 #define DMZ_REPORT_NR_ZONES 4096
1152
1153 /*
1154 * Allocate and initialize zone descriptors using the zone
1155 * information from disk.
1156 */
1157 static int dmz_init_zones(struct dmz_metadata *zmd)
1158 {
1159 struct dmz_dev *dev = zmd->dev;
1160 struct dm_zone *zone;
1161 struct blk_zone *blkz;
1162 unsigned int nr_blkz;
1163 sector_t sector = 0;
1164 int i, ret = 0;
1165
1166 /* Init */
1167 zmd->zone_bitmap_size = dev->zone_nr_blocks >> 3;
1168 zmd->zone_nr_bitmap_blocks =
1169 max_t(sector_t, 1, zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT);
1170 zmd->zone_bits_per_mblk = min_t(sector_t, dev->zone_nr_blocks,
1171 DMZ_BLOCK_SIZE_BITS);
1172
1173 /* Allocate zone array */
1174 zmd->zones = kcalloc(dev->nr_zones, sizeof(struct dm_zone), GFP_KERNEL);
1175 if (!zmd->zones)
1176 return -ENOMEM;
1177
1178 dmz_dev_info(dev, "Using %zu B for zone information",
1179 sizeof(struct dm_zone) * dev->nr_zones);
1180
1181 /* Get zone information */
1182 nr_blkz = DMZ_REPORT_NR_ZONES;
1183 blkz = kcalloc(nr_blkz, sizeof(struct blk_zone), GFP_KERNEL);
1184 if (!blkz) {
1185 ret = -ENOMEM;
1186 goto out;
1187 }
1188
1189 /*
1190 * Get zone information and initialize zone descriptors.
1191 * At the same time, determine where the super block
1192 * should be: first block of the first randomly writable
1193 * zone.
1194 */
1195 zone = zmd->zones;
1196 while (sector < dev->capacity) {
1197 /* Get zone information */
1198 nr_blkz = DMZ_REPORT_NR_ZONES;
1199 ret = blkdev_report_zones(dev->bdev, sector, blkz,
1200 &nr_blkz, GFP_KERNEL);
1201 if (ret) {
1202 dmz_dev_err(dev, "Report zones failed %d", ret);
1203 goto out;
1204 }
1205
1206 if (!nr_blkz)
1207 break;
1208
1209 /* Process report */
1210 for (i = 0; i < nr_blkz; i++) {
1211 ret = dmz_init_zone(zmd, zone, &blkz[i]);
1212 if (ret)
1213 goto out;
1214 sector += dev->zone_nr_sectors;
1215 zone++;
1216 }
1217 }
1218
1219 /* The entire zone configuration of the disk should now be known */
1220 if (sector < dev->capacity) {
1221 dmz_dev_err(dev, "Failed to get correct zone information");
1222 ret = -ENXIO;
1223 }
1224 out:
1225 kfree(blkz);
1226 if (ret)
1227 dmz_drop_zones(zmd);
1228
1229 return ret;
1230 }
1231
1232 /*
1233 * Update a zone information.
1234 */
1235 static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1236 {
1237 unsigned int nr_blkz = 1;
1238 struct blk_zone blkz;
1239 int ret;
1240
1241 /* Get zone information from disk */
1242 ret = blkdev_report_zones(zmd->dev->bdev, dmz_start_sect(zmd, zone),
1243 &blkz, &nr_blkz, GFP_NOIO);
1244 if (!nr_blkz)
1245 ret = -EIO;
1246 if (ret) {
1247 dmz_dev_err(zmd->dev, "Get zone %u report failed",
1248 dmz_id(zmd, zone));
1249 dmz_check_bdev(zmd->dev);
1250 return ret;
1251 }
1252
1253 clear_bit(DMZ_OFFLINE, &zone->flags);
1254 clear_bit(DMZ_READ_ONLY, &zone->flags);
1255 if (blkz.cond == BLK_ZONE_COND_OFFLINE)
1256 set_bit(DMZ_OFFLINE, &zone->flags);
1257 else if (blkz.cond == BLK_ZONE_COND_READONLY)
1258 set_bit(DMZ_READ_ONLY, &zone->flags);
1259
1260 if (dmz_is_seq(zone))
1261 zone->wp_block = dmz_sect2blk(blkz.wp - blkz.start);
1262 else
1263 zone->wp_block = 0;
1264
1265 return 0;
1266 }
1267
1268 /*
1269 * Check a zone write pointer position when the zone is marked
1270 * with the sequential write error flag.
1271 */
1272 static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1273 struct dm_zone *zone)
1274 {
1275 unsigned int wp = 0;
1276 int ret;
1277
1278 wp = zone->wp_block;
1279 ret = dmz_update_zone(zmd, zone);
1280 if (ret)
1281 return ret;
1282
1283 dmz_dev_warn(zmd->dev, "Processing zone %u write error (zone wp %u/%u)",
1284 dmz_id(zmd, zone), zone->wp_block, wp);
1285
1286 if (zone->wp_block < wp) {
1287 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1288 wp - zone->wp_block);
1289 }
1290
1291 return 0;
1292 }
1293
1294 static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
1295 {
1296 return &zmd->zones[zone_id];
1297 }
1298
1299 /*
1300 * Reset a zone write pointer.
1301 */
1302 static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1303 {
1304 int ret;
1305
1306 /*
1307 * Ignore offline zones, read only zones,
1308 * and conventional zones.
1309 */
1310 if (dmz_is_offline(zone) ||
1311 dmz_is_readonly(zone) ||
1312 dmz_is_rnd(zone))
1313 return 0;
1314
1315 if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1316 struct dmz_dev *dev = zmd->dev;
1317
1318 ret = blkdev_reset_zones(dev->bdev,
1319 dmz_start_sect(zmd, zone),
1320 dev->zone_nr_sectors, GFP_NOIO);
1321 if (ret) {
1322 dmz_dev_err(dev, "Reset zone %u failed %d",
1323 dmz_id(zmd, zone), ret);
1324 return ret;
1325 }
1326 }
1327
1328 /* Clear write error bit and rewind write pointer position */
1329 clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1330 zone->wp_block = 0;
1331
1332 return 0;
1333 }
1334
1335 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1336
1337 /*
1338 * Initialize chunk mapping.
1339 */
1340 static int dmz_load_mapping(struct dmz_metadata *zmd)
1341 {
1342 struct dmz_dev *dev = zmd->dev;
1343 struct dm_zone *dzone, *bzone;
1344 struct dmz_mblock *dmap_mblk = NULL;
1345 struct dmz_map *dmap;
1346 unsigned int i = 0, e = 0, chunk = 0;
1347 unsigned int dzone_id;
1348 unsigned int bzone_id;
1349
1350 /* Metadata block array for the chunk mapping table */
1351 zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1352 sizeof(struct dmz_mblk *), GFP_KERNEL);
1353 if (!zmd->map_mblk)
1354 return -ENOMEM;
1355
1356 /* Get chunk mapping table blocks and initialize zone mapping */
1357 while (chunk < zmd->nr_chunks) {
1358 if (!dmap_mblk) {
1359 /* Get mapping block */
1360 dmap_mblk = dmz_get_mblock(zmd, i + 1);
1361 if (IS_ERR(dmap_mblk))
1362 return PTR_ERR(dmap_mblk);
1363 zmd->map_mblk[i] = dmap_mblk;
1364 dmap = (struct dmz_map *) dmap_mblk->data;
1365 i++;
1366 e = 0;
1367 }
1368
1369 /* Check data zone */
1370 dzone_id = le32_to_cpu(dmap[e].dzone_id);
1371 if (dzone_id == DMZ_MAP_UNMAPPED)
1372 goto next;
1373
1374 if (dzone_id >= dev->nr_zones) {
1375 dmz_dev_err(dev, "Chunk %u mapping: invalid data zone ID %u",
1376 chunk, dzone_id);
1377 return -EIO;
1378 }
1379
1380 dzone = dmz_get(zmd, dzone_id);
1381 set_bit(DMZ_DATA, &dzone->flags);
1382 dzone->chunk = chunk;
1383 dmz_get_zone_weight(zmd, dzone);
1384
1385 if (dmz_is_rnd(dzone))
1386 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1387 else
1388 list_add_tail(&dzone->link, &zmd->map_seq_list);
1389
1390 /* Check buffer zone */
1391 bzone_id = le32_to_cpu(dmap[e].bzone_id);
1392 if (bzone_id == DMZ_MAP_UNMAPPED)
1393 goto next;
1394
1395 if (bzone_id >= dev->nr_zones) {
1396 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone ID %u",
1397 chunk, bzone_id);
1398 return -EIO;
1399 }
1400
1401 bzone = dmz_get(zmd, bzone_id);
1402 if (!dmz_is_rnd(bzone)) {
1403 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone %u",
1404 chunk, bzone_id);
1405 return -EIO;
1406 }
1407
1408 set_bit(DMZ_DATA, &bzone->flags);
1409 set_bit(DMZ_BUF, &bzone->flags);
1410 bzone->chunk = chunk;
1411 bzone->bzone = dzone;
1412 dzone->bzone = bzone;
1413 dmz_get_zone_weight(zmd, bzone);
1414 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1415 next:
1416 chunk++;
1417 e++;
1418 if (e >= DMZ_MAP_ENTRIES)
1419 dmap_mblk = NULL;
1420 }
1421
1422 /*
1423 * At this point, only meta zones and mapped data zones were
1424 * fully initialized. All remaining zones are unmapped data
1425 * zones. Finish initializing those here.
1426 */
1427 for (i = 0; i < dev->nr_zones; i++) {
1428 dzone = dmz_get(zmd, i);
1429 if (dmz_is_meta(dzone))
1430 continue;
1431
1432 if (dmz_is_rnd(dzone))
1433 zmd->nr_rnd++;
1434 else
1435 zmd->nr_seq++;
1436
1437 if (dmz_is_data(dzone)) {
1438 /* Already initialized */
1439 continue;
1440 }
1441
1442 /* Unmapped data zone */
1443 set_bit(DMZ_DATA, &dzone->flags);
1444 dzone->chunk = DMZ_MAP_UNMAPPED;
1445 if (dmz_is_rnd(dzone)) {
1446 list_add_tail(&dzone->link, &zmd->unmap_rnd_list);
1447 atomic_inc(&zmd->unmap_nr_rnd);
1448 } else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1449 list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1450 atomic_inc(&zmd->nr_reserved_seq_zones);
1451 zmd->nr_seq--;
1452 } else {
1453 list_add_tail(&dzone->link, &zmd->unmap_seq_list);
1454 atomic_inc(&zmd->unmap_nr_seq);
1455 }
1456 }
1457
1458 return 0;
1459 }
1460
1461 /*
1462 * Set a data chunk mapping.
1463 */
1464 static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1465 unsigned int dzone_id, unsigned int bzone_id)
1466 {
1467 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1468 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1469 int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1470
1471 dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1472 dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1473 dmz_dirty_mblock(zmd, dmap_mblk);
1474 }
1475
1476 /*
1477 * The list of mapped zones is maintained in LRU order.
1478 * This rotates a zone at the end of its map list.
1479 */
1480 static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1481 {
1482 if (list_empty(&zone->link))
1483 return;
1484
1485 list_del_init(&zone->link);
1486 if (dmz_is_seq(zone)) {
1487 /* LRU rotate sequential zone */
1488 list_add_tail(&zone->link, &zmd->map_seq_list);
1489 } else {
1490 /* LRU rotate random zone */
1491 list_add_tail(&zone->link, &zmd->map_rnd_list);
1492 }
1493 }
1494
1495 /*
1496 * The list of mapped random zones is maintained
1497 * in LRU order. This rotates a zone at the end of the list.
1498 */
1499 static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1500 {
1501 __dmz_lru_zone(zmd, zone);
1502 if (zone->bzone)
1503 __dmz_lru_zone(zmd, zone->bzone);
1504 }
1505
1506 /*
1507 * Wait for any zone to be freed.
1508 */
1509 static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1510 {
1511 DEFINE_WAIT(wait);
1512
1513 prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1514 dmz_unlock_map(zmd);
1515 dmz_unlock_metadata(zmd);
1516
1517 io_schedule_timeout(HZ);
1518
1519 dmz_lock_metadata(zmd);
1520 dmz_lock_map(zmd);
1521 finish_wait(&zmd->free_wq, &wait);
1522 }
1523
1524 /*
1525 * Lock a zone for reclaim (set the zone RECLAIM bit).
1526 * Returns false if the zone cannot be locked or if it is already locked
1527 * and 1 otherwise.
1528 */
1529 int dmz_lock_zone_reclaim(struct dm_zone *zone)
1530 {
1531 /* Active zones cannot be reclaimed */
1532 if (dmz_is_active(zone))
1533 return 0;
1534
1535 return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1536 }
1537
1538 /*
1539 * Clear a zone reclaim flag.
1540 */
1541 void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1542 {
1543 WARN_ON(dmz_is_active(zone));
1544 WARN_ON(!dmz_in_reclaim(zone));
1545
1546 clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1547 smp_mb__after_atomic();
1548 wake_up_bit(&zone->flags, DMZ_RECLAIM);
1549 }
1550
1551 /*
1552 * Wait for a zone reclaim to complete.
1553 */
1554 static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1555 {
1556 dmz_unlock_map(zmd);
1557 dmz_unlock_metadata(zmd);
1558 wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1559 dmz_lock_metadata(zmd);
1560 dmz_lock_map(zmd);
1561 }
1562
1563 /*
1564 * Select a random write zone for reclaim.
1565 */
1566 static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd)
1567 {
1568 struct dm_zone *dzone = NULL;
1569 struct dm_zone *zone;
1570
1571 if (list_empty(&zmd->map_rnd_list))
1572 return ERR_PTR(-EBUSY);
1573
1574 list_for_each_entry(zone, &zmd->map_rnd_list, link) {
1575 if (dmz_is_buf(zone))
1576 dzone = zone->bzone;
1577 else
1578 dzone = zone;
1579 if (dmz_lock_zone_reclaim(dzone))
1580 return dzone;
1581 }
1582
1583 return NULL;
1584 }
1585
1586 /*
1587 * Select a buffered sequential zone for reclaim.
1588 */
1589 static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd)
1590 {
1591 struct dm_zone *zone;
1592
1593 if (list_empty(&zmd->map_seq_list))
1594 return ERR_PTR(-EBUSY);
1595
1596 list_for_each_entry(zone, &zmd->map_seq_list, link) {
1597 if (!zone->bzone)
1598 continue;
1599 if (dmz_lock_zone_reclaim(zone))
1600 return zone;
1601 }
1602
1603 return NULL;
1604 }
1605
1606 /*
1607 * Select a zone for reclaim.
1608 */
1609 struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd)
1610 {
1611 struct dm_zone *zone;
1612
1613 /*
1614 * Search for a zone candidate to reclaim: 2 cases are possible.
1615 * (1) There is no free sequential zones. Then a random data zone
1616 * cannot be reclaimed. So choose a sequential zone to reclaim so
1617 * that afterward a random zone can be reclaimed.
1618 * (2) At least one free sequential zone is available, then choose
1619 * the oldest random zone (data or buffer) that can be locked.
1620 */
1621 dmz_lock_map(zmd);
1622 if (list_empty(&zmd->reserved_seq_zones_list))
1623 zone = dmz_get_seq_zone_for_reclaim(zmd);
1624 else
1625 zone = dmz_get_rnd_zone_for_reclaim(zmd);
1626 dmz_unlock_map(zmd);
1627
1628 return zone;
1629 }
1630
1631 /*
1632 * Get the zone mapping a chunk, if the chunk is mapped already.
1633 * If no mapping exist and the operation is WRITE, a zone is
1634 * allocated and used to map the chunk.
1635 * The zone returned will be set to the active state.
1636 */
1637 struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op)
1638 {
1639 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1640 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1641 int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1642 unsigned int dzone_id;
1643 struct dm_zone *dzone = NULL;
1644 int ret = 0;
1645
1646 dmz_lock_map(zmd);
1647 again:
1648 /* Get the chunk mapping */
1649 dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
1650 if (dzone_id == DMZ_MAP_UNMAPPED) {
1651 /*
1652 * Read or discard in unmapped chunks are fine. But for
1653 * writes, we need a mapping, so get one.
1654 */
1655 if (op != REQ_OP_WRITE)
1656 goto out;
1657
1658 /* Alloate a random zone */
1659 dzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1660 if (!dzone) {
1661 if (dmz_bdev_is_dying(zmd->dev)) {
1662 dzone = ERR_PTR(-EIO);
1663 goto out;
1664 }
1665 dmz_wait_for_free_zones(zmd);
1666 goto again;
1667 }
1668
1669 dmz_map_zone(zmd, dzone, chunk);
1670
1671 } else {
1672 /* The chunk is already mapped: get the mapping zone */
1673 dzone = dmz_get(zmd, dzone_id);
1674 if (dzone->chunk != chunk) {
1675 dzone = ERR_PTR(-EIO);
1676 goto out;
1677 }
1678
1679 /* Repair write pointer if the sequential dzone has error */
1680 if (dmz_seq_write_err(dzone)) {
1681 ret = dmz_handle_seq_write_err(zmd, dzone);
1682 if (ret) {
1683 dzone = ERR_PTR(-EIO);
1684 goto out;
1685 }
1686 clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
1687 }
1688 }
1689
1690 /*
1691 * If the zone is being reclaimed, the chunk mapping may change
1692 * to a different zone. So wait for reclaim and retry. Otherwise,
1693 * activate the zone (this will prevent reclaim from touching it).
1694 */
1695 if (dmz_in_reclaim(dzone)) {
1696 dmz_wait_for_reclaim(zmd, dzone);
1697 goto again;
1698 }
1699 dmz_activate_zone(dzone);
1700 dmz_lru_zone(zmd, dzone);
1701 out:
1702 dmz_unlock_map(zmd);
1703
1704 return dzone;
1705 }
1706
1707 /*
1708 * Write and discard change the block validity of data zones and their buffer
1709 * zones. Check here that valid blocks are still present. If all blocks are
1710 * invalid, the zones can be unmapped on the fly without waiting for reclaim
1711 * to do it.
1712 */
1713 void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
1714 {
1715 struct dm_zone *bzone;
1716
1717 dmz_lock_map(zmd);
1718
1719 bzone = dzone->bzone;
1720 if (bzone) {
1721 if (dmz_weight(bzone))
1722 dmz_lru_zone(zmd, bzone);
1723 else {
1724 /* Empty buffer zone: reclaim it */
1725 dmz_unmap_zone(zmd, bzone);
1726 dmz_free_zone(zmd, bzone);
1727 bzone = NULL;
1728 }
1729 }
1730
1731 /* Deactivate the data zone */
1732 dmz_deactivate_zone(dzone);
1733 if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
1734 dmz_lru_zone(zmd, dzone);
1735 else {
1736 /* Unbuffered inactive empty data zone: reclaim it */
1737 dmz_unmap_zone(zmd, dzone);
1738 dmz_free_zone(zmd, dzone);
1739 }
1740
1741 dmz_unlock_map(zmd);
1742 }
1743
1744 /*
1745 * Allocate and map a random zone to buffer a chunk
1746 * already mapped to a sequential zone.
1747 */
1748 struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
1749 struct dm_zone *dzone)
1750 {
1751 struct dm_zone *bzone;
1752
1753 dmz_lock_map(zmd);
1754 again:
1755 bzone = dzone->bzone;
1756 if (bzone)
1757 goto out;
1758
1759 /* Alloate a random zone */
1760 bzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1761 if (!bzone) {
1762 if (dmz_bdev_is_dying(zmd->dev)) {
1763 bzone = ERR_PTR(-EIO);
1764 goto out;
1765 }
1766 dmz_wait_for_free_zones(zmd);
1767 goto again;
1768 }
1769
1770 /* Update the chunk mapping */
1771 dmz_set_chunk_mapping(zmd, dzone->chunk, dmz_id(zmd, dzone),
1772 dmz_id(zmd, bzone));
1773
1774 set_bit(DMZ_BUF, &bzone->flags);
1775 bzone->chunk = dzone->chunk;
1776 bzone->bzone = dzone;
1777 dzone->bzone = bzone;
1778 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1779 out:
1780 dmz_unlock_map(zmd);
1781
1782 return bzone;
1783 }
1784
1785 /*
1786 * Get an unmapped (free) zone.
1787 * This must be called with the mapping lock held.
1788 */
1789 struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags)
1790 {
1791 struct list_head *list;
1792 struct dm_zone *zone;
1793
1794 if (flags & DMZ_ALLOC_RND)
1795 list = &zmd->unmap_rnd_list;
1796 else
1797 list = &zmd->unmap_seq_list;
1798 again:
1799 if (list_empty(list)) {
1800 /*
1801 * No free zone: if this is for reclaim, allow using the
1802 * reserved sequential zones.
1803 */
1804 if (!(flags & DMZ_ALLOC_RECLAIM) ||
1805 list_empty(&zmd->reserved_seq_zones_list))
1806 return NULL;
1807
1808 zone = list_first_entry(&zmd->reserved_seq_zones_list,
1809 struct dm_zone, link);
1810 list_del_init(&zone->link);
1811 atomic_dec(&zmd->nr_reserved_seq_zones);
1812 return zone;
1813 }
1814
1815 zone = list_first_entry(list, struct dm_zone, link);
1816 list_del_init(&zone->link);
1817
1818 if (dmz_is_rnd(zone))
1819 atomic_dec(&zmd->unmap_nr_rnd);
1820 else
1821 atomic_dec(&zmd->unmap_nr_seq);
1822
1823 if (dmz_is_offline(zone)) {
1824 dmz_dev_warn(zmd->dev, "Zone %u is offline", dmz_id(zmd, zone));
1825 zone = NULL;
1826 goto again;
1827 }
1828
1829 return zone;
1830 }
1831
1832 /*
1833 * Free a zone.
1834 * This must be called with the mapping lock held.
1835 */
1836 void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1837 {
1838 /* If this is a sequential zone, reset it */
1839 if (dmz_is_seq(zone))
1840 dmz_reset_zone(zmd, zone);
1841
1842 /* Return the zone to its type unmap list */
1843 if (dmz_is_rnd(zone)) {
1844 list_add_tail(&zone->link, &zmd->unmap_rnd_list);
1845 atomic_inc(&zmd->unmap_nr_rnd);
1846 } else if (atomic_read(&zmd->nr_reserved_seq_zones) <
1847 zmd->nr_reserved_seq) {
1848 list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
1849 atomic_inc(&zmd->nr_reserved_seq_zones);
1850 } else {
1851 list_add_tail(&zone->link, &zmd->unmap_seq_list);
1852 atomic_inc(&zmd->unmap_nr_seq);
1853 }
1854
1855 wake_up_all(&zmd->free_wq);
1856 }
1857
1858 /*
1859 * Map a chunk to a zone.
1860 * This must be called with the mapping lock held.
1861 */
1862 void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
1863 unsigned int chunk)
1864 {
1865 /* Set the chunk mapping */
1866 dmz_set_chunk_mapping(zmd, chunk, dmz_id(zmd, dzone),
1867 DMZ_MAP_UNMAPPED);
1868 dzone->chunk = chunk;
1869 if (dmz_is_rnd(dzone))
1870 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1871 else
1872 list_add_tail(&dzone->link, &zmd->map_seq_list);
1873 }
1874
1875 /*
1876 * Unmap a zone.
1877 * This must be called with the mapping lock held.
1878 */
1879 void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1880 {
1881 unsigned int chunk = zone->chunk;
1882 unsigned int dzone_id;
1883
1884 if (chunk == DMZ_MAP_UNMAPPED) {
1885 /* Already unmapped */
1886 return;
1887 }
1888
1889 if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
1890 /*
1891 * Unmapping the chunk buffer zone: clear only
1892 * the chunk buffer mapping
1893 */
1894 dzone_id = dmz_id(zmd, zone->bzone);
1895 zone->bzone->bzone = NULL;
1896 zone->bzone = NULL;
1897
1898 } else {
1899 /*
1900 * Unmapping the chunk data zone: the zone must
1901 * not be buffered.
1902 */
1903 if (WARN_ON(zone->bzone)) {
1904 zone->bzone->bzone = NULL;
1905 zone->bzone = NULL;
1906 }
1907 dzone_id = DMZ_MAP_UNMAPPED;
1908 }
1909
1910 dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
1911
1912 zone->chunk = DMZ_MAP_UNMAPPED;
1913 list_del_init(&zone->link);
1914 }
1915
1916 /*
1917 * Set @nr_bits bits in @bitmap starting from @bit.
1918 * Return the number of bits changed from 0 to 1.
1919 */
1920 static unsigned int dmz_set_bits(unsigned long *bitmap,
1921 unsigned int bit, unsigned int nr_bits)
1922 {
1923 unsigned long *addr;
1924 unsigned int end = bit + nr_bits;
1925 unsigned int n = 0;
1926
1927 while (bit < end) {
1928 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
1929 ((end - bit) >= BITS_PER_LONG)) {
1930 /* Try to set the whole word at once */
1931 addr = bitmap + BIT_WORD(bit);
1932 if (*addr == 0) {
1933 *addr = ULONG_MAX;
1934 n += BITS_PER_LONG;
1935 bit += BITS_PER_LONG;
1936 continue;
1937 }
1938 }
1939
1940 if (!test_and_set_bit(bit, bitmap))
1941 n++;
1942 bit++;
1943 }
1944
1945 return n;
1946 }
1947
1948 /*
1949 * Get the bitmap block storing the bit for chunk_block in zone.
1950 */
1951 static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
1952 struct dm_zone *zone,
1953 sector_t chunk_block)
1954 {
1955 sector_t bitmap_block = 1 + zmd->nr_map_blocks +
1956 (sector_t)(dmz_id(zmd, zone) * zmd->zone_nr_bitmap_blocks) +
1957 (chunk_block >> DMZ_BLOCK_SHIFT_BITS);
1958
1959 return dmz_get_mblock(zmd, bitmap_block);
1960 }
1961
1962 /*
1963 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
1964 */
1965 int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1966 struct dm_zone *to_zone)
1967 {
1968 struct dmz_mblock *from_mblk, *to_mblk;
1969 sector_t chunk_block = 0;
1970
1971 /* Get the zones bitmap blocks */
1972 while (chunk_block < zmd->dev->zone_nr_blocks) {
1973 from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
1974 if (IS_ERR(from_mblk))
1975 return PTR_ERR(from_mblk);
1976 to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
1977 if (IS_ERR(to_mblk)) {
1978 dmz_release_mblock(zmd, from_mblk);
1979 return PTR_ERR(to_mblk);
1980 }
1981
1982 memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
1983 dmz_dirty_mblock(zmd, to_mblk);
1984
1985 dmz_release_mblock(zmd, to_mblk);
1986 dmz_release_mblock(zmd, from_mblk);
1987
1988 chunk_block += zmd->zone_bits_per_mblk;
1989 }
1990
1991 to_zone->weight = from_zone->weight;
1992
1993 return 0;
1994 }
1995
1996 /*
1997 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
1998 * starting from chunk_block.
1999 */
2000 int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
2001 struct dm_zone *to_zone, sector_t chunk_block)
2002 {
2003 unsigned int nr_blocks;
2004 int ret;
2005
2006 /* Get the zones bitmap blocks */
2007 while (chunk_block < zmd->dev->zone_nr_blocks) {
2008 /* Get a valid region from the source zone */
2009 ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
2010 if (ret <= 0)
2011 return ret;
2012
2013 nr_blocks = ret;
2014 ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
2015 if (ret)
2016 return ret;
2017
2018 chunk_block += nr_blocks;
2019 }
2020
2021 return 0;
2022 }
2023
2024 /*
2025 * Validate all the blocks in the range [block..block+nr_blocks-1].
2026 */
2027 int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2028 sector_t chunk_block, unsigned int nr_blocks)
2029 {
2030 unsigned int count, bit, nr_bits;
2031 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
2032 struct dmz_mblock *mblk;
2033 unsigned int n = 0;
2034
2035 dmz_dev_debug(zmd->dev, "=> VALIDATE zone %u, block %llu, %u blocks",
2036 dmz_id(zmd, zone), (unsigned long long)chunk_block,
2037 nr_blocks);
2038
2039 WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2040
2041 while (nr_blocks) {
2042 /* Get bitmap block */
2043 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2044 if (IS_ERR(mblk))
2045 return PTR_ERR(mblk);
2046
2047 /* Set bits */
2048 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2049 nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2050
2051 count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2052 if (count) {
2053 dmz_dirty_mblock(zmd, mblk);
2054 n += count;
2055 }
2056 dmz_release_mblock(zmd, mblk);
2057
2058 nr_blocks -= nr_bits;
2059 chunk_block += nr_bits;
2060 }
2061
2062 if (likely(zone->weight + n <= zone_nr_blocks))
2063 zone->weight += n;
2064 else {
2065 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be <= %u",
2066 dmz_id(zmd, zone), zone->weight,
2067 zone_nr_blocks - n);
2068 zone->weight = zone_nr_blocks;
2069 }
2070
2071 return 0;
2072 }
2073
2074 /*
2075 * Clear nr_bits bits in bitmap starting from bit.
2076 * Return the number of bits cleared.
2077 */
2078 static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2079 {
2080 unsigned long *addr;
2081 int end = bit + nr_bits;
2082 int n = 0;
2083
2084 while (bit < end) {
2085 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2086 ((end - bit) >= BITS_PER_LONG)) {
2087 /* Try to clear whole word at once */
2088 addr = bitmap + BIT_WORD(bit);
2089 if (*addr == ULONG_MAX) {
2090 *addr = 0;
2091 n += BITS_PER_LONG;
2092 bit += BITS_PER_LONG;
2093 continue;
2094 }
2095 }
2096
2097 if (test_and_clear_bit(bit, bitmap))
2098 n++;
2099 bit++;
2100 }
2101
2102 return n;
2103 }
2104
2105 /*
2106 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2107 */
2108 int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2109 sector_t chunk_block, unsigned int nr_blocks)
2110 {
2111 unsigned int count, bit, nr_bits;
2112 struct dmz_mblock *mblk;
2113 unsigned int n = 0;
2114
2115 dmz_dev_debug(zmd->dev, "=> INVALIDATE zone %u, block %llu, %u blocks",
2116 dmz_id(zmd, zone), (u64)chunk_block, nr_blocks);
2117
2118 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2119
2120 while (nr_blocks) {
2121 /* Get bitmap block */
2122 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2123 if (IS_ERR(mblk))
2124 return PTR_ERR(mblk);
2125
2126 /* Clear bits */
2127 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2128 nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2129
2130 count = dmz_clear_bits((unsigned long *)mblk->data,
2131 bit, nr_bits);
2132 if (count) {
2133 dmz_dirty_mblock(zmd, mblk);
2134 n += count;
2135 }
2136 dmz_release_mblock(zmd, mblk);
2137
2138 nr_blocks -= nr_bits;
2139 chunk_block += nr_bits;
2140 }
2141
2142 if (zone->weight >= n)
2143 zone->weight -= n;
2144 else {
2145 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be >= %u",
2146 dmz_id(zmd, zone), zone->weight, n);
2147 zone->weight = 0;
2148 }
2149
2150 return 0;
2151 }
2152
2153 /*
2154 * Get a block bit value.
2155 */
2156 static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2157 sector_t chunk_block)
2158 {
2159 struct dmz_mblock *mblk;
2160 int ret;
2161
2162 WARN_ON(chunk_block >= zmd->dev->zone_nr_blocks);
2163
2164 /* Get bitmap block */
2165 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2166 if (IS_ERR(mblk))
2167 return PTR_ERR(mblk);
2168
2169 /* Get offset */
2170 ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2171 (unsigned long *) mblk->data) != 0;
2172
2173 dmz_release_mblock(zmd, mblk);
2174
2175 return ret;
2176 }
2177
2178 /*
2179 * Return the number of blocks from chunk_block to the first block with a bit
2180 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2181 */
2182 static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2183 sector_t chunk_block, unsigned int nr_blocks,
2184 int set)
2185 {
2186 struct dmz_mblock *mblk;
2187 unsigned int bit, set_bit, nr_bits;
2188 unsigned int zone_bits = zmd->zone_bits_per_mblk;
2189 unsigned long *bitmap;
2190 int n = 0;
2191
2192 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2193
2194 while (nr_blocks) {
2195 /* Get bitmap block */
2196 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2197 if (IS_ERR(mblk))
2198 return PTR_ERR(mblk);
2199
2200 /* Get offset */
2201 bitmap = (unsigned long *) mblk->data;
2202 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2203 nr_bits = min(nr_blocks, zone_bits - bit);
2204 if (set)
2205 set_bit = find_next_bit(bitmap, zone_bits, bit);
2206 else
2207 set_bit = find_next_zero_bit(bitmap, zone_bits, bit);
2208 dmz_release_mblock(zmd, mblk);
2209
2210 n += set_bit - bit;
2211 if (set_bit < zone_bits)
2212 break;
2213
2214 nr_blocks -= nr_bits;
2215 chunk_block += nr_bits;
2216 }
2217
2218 return n;
2219 }
2220
2221 /*
2222 * Test if chunk_block is valid. If it is, the number of consecutive
2223 * valid blocks from chunk_block will be returned.
2224 */
2225 int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2226 sector_t chunk_block)
2227 {
2228 int valid;
2229
2230 valid = dmz_test_block(zmd, zone, chunk_block);
2231 if (valid <= 0)
2232 return valid;
2233
2234 /* The block is valid: get the number of valid blocks from block */
2235 return dmz_to_next_set_block(zmd, zone, chunk_block,
2236 zmd->dev->zone_nr_blocks - chunk_block, 0);
2237 }
2238
2239 /*
2240 * Find the first valid block from @chunk_block in @zone.
2241 * If such a block is found, its number is returned using
2242 * @chunk_block and the total number of valid blocks from @chunk_block
2243 * is returned.
2244 */
2245 int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2246 sector_t *chunk_block)
2247 {
2248 sector_t start_block = *chunk_block;
2249 int ret;
2250
2251 ret = dmz_to_next_set_block(zmd, zone, start_block,
2252 zmd->dev->zone_nr_blocks - start_block, 1);
2253 if (ret < 0)
2254 return ret;
2255
2256 start_block += ret;
2257 *chunk_block = start_block;
2258
2259 return dmz_to_next_set_block(zmd, zone, start_block,
2260 zmd->dev->zone_nr_blocks - start_block, 0);
2261 }
2262
2263 /*
2264 * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2265 */
2266 static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2267 {
2268 unsigned long *addr;
2269 int end = bit + nr_bits;
2270 int n = 0;
2271
2272 while (bit < end) {
2273 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2274 ((end - bit) >= BITS_PER_LONG)) {
2275 addr = (unsigned long *)bitmap + BIT_WORD(bit);
2276 if (*addr == ULONG_MAX) {
2277 n += BITS_PER_LONG;
2278 bit += BITS_PER_LONG;
2279 continue;
2280 }
2281 }
2282
2283 if (test_bit(bit, bitmap))
2284 n++;
2285 bit++;
2286 }
2287
2288 return n;
2289 }
2290
2291 /*
2292 * Get a zone weight.
2293 */
2294 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2295 {
2296 struct dmz_mblock *mblk;
2297 sector_t chunk_block = 0;
2298 unsigned int bit, nr_bits;
2299 unsigned int nr_blocks = zmd->dev->zone_nr_blocks;
2300 void *bitmap;
2301 int n = 0;
2302
2303 while (nr_blocks) {
2304 /* Get bitmap block */
2305 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2306 if (IS_ERR(mblk)) {
2307 n = 0;
2308 break;
2309 }
2310
2311 /* Count bits in this block */
2312 bitmap = mblk->data;
2313 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2314 nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2315 n += dmz_count_bits(bitmap, bit, nr_bits);
2316
2317 dmz_release_mblock(zmd, mblk);
2318
2319 nr_blocks -= nr_bits;
2320 chunk_block += nr_bits;
2321 }
2322
2323 zone->weight = n;
2324 }
2325
2326 /*
2327 * Cleanup the zoned metadata resources.
2328 */
2329 static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2330 {
2331 struct rb_root *root;
2332 struct dmz_mblock *mblk, *next;
2333 int i;
2334
2335 /* Release zone mapping resources */
2336 if (zmd->map_mblk) {
2337 for (i = 0; i < zmd->nr_map_blocks; i++)
2338 dmz_release_mblock(zmd, zmd->map_mblk[i]);
2339 kfree(zmd->map_mblk);
2340 zmd->map_mblk = NULL;
2341 }
2342
2343 /* Release super blocks */
2344 for (i = 0; i < 2; i++) {
2345 if (zmd->sb[i].mblk) {
2346 dmz_free_mblock(zmd, zmd->sb[i].mblk);
2347 zmd->sb[i].mblk = NULL;
2348 }
2349 }
2350
2351 /* Free cached blocks */
2352 while (!list_empty(&zmd->mblk_dirty_list)) {
2353 mblk = list_first_entry(&zmd->mblk_dirty_list,
2354 struct dmz_mblock, link);
2355 dmz_dev_warn(zmd->dev, "mblock %llu still in dirty list (ref %u)",
2356 (u64)mblk->no, mblk->ref);
2357 list_del_init(&mblk->link);
2358 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2359 dmz_free_mblock(zmd, mblk);
2360 }
2361
2362 while (!list_empty(&zmd->mblk_lru_list)) {
2363 mblk = list_first_entry(&zmd->mblk_lru_list,
2364 struct dmz_mblock, link);
2365 list_del_init(&mblk->link);
2366 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2367 dmz_free_mblock(zmd, mblk);
2368 }
2369
2370 /* Sanity checks: the mblock rbtree should now be empty */
2371 root = &zmd->mblk_rbtree;
2372 rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2373 dmz_dev_warn(zmd->dev, "mblock %llu ref %u still in rbtree",
2374 (u64)mblk->no, mblk->ref);
2375 mblk->ref = 0;
2376 dmz_free_mblock(zmd, mblk);
2377 }
2378
2379 /* Free the zone descriptors */
2380 dmz_drop_zones(zmd);
2381
2382 mutex_destroy(&zmd->mblk_flush_lock);
2383 mutex_destroy(&zmd->map_lock);
2384 }
2385
2386 /*
2387 * Initialize the zoned metadata.
2388 */
2389 int dmz_ctr_metadata(struct dmz_dev *dev, struct dmz_metadata **metadata)
2390 {
2391 struct dmz_metadata *zmd;
2392 unsigned int i, zid;
2393 struct dm_zone *zone;
2394 int ret;
2395
2396 zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2397 if (!zmd)
2398 return -ENOMEM;
2399
2400 zmd->dev = dev;
2401 zmd->mblk_rbtree = RB_ROOT;
2402 init_rwsem(&zmd->mblk_sem);
2403 mutex_init(&zmd->mblk_flush_lock);
2404 spin_lock_init(&zmd->mblk_lock);
2405 INIT_LIST_HEAD(&zmd->mblk_lru_list);
2406 INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2407
2408 mutex_init(&zmd->map_lock);
2409 atomic_set(&zmd->unmap_nr_rnd, 0);
2410 INIT_LIST_HEAD(&zmd->unmap_rnd_list);
2411 INIT_LIST_HEAD(&zmd->map_rnd_list);
2412
2413 atomic_set(&zmd->unmap_nr_seq, 0);
2414 INIT_LIST_HEAD(&zmd->unmap_seq_list);
2415 INIT_LIST_HEAD(&zmd->map_seq_list);
2416
2417 atomic_set(&zmd->nr_reserved_seq_zones, 0);
2418 INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2419
2420 init_waitqueue_head(&zmd->free_wq);
2421
2422 /* Initialize zone descriptors */
2423 ret = dmz_init_zones(zmd);
2424 if (ret)
2425 goto err;
2426
2427 /* Get super block */
2428 ret = dmz_load_sb(zmd);
2429 if (ret)
2430 goto err;
2431
2432 /* Set metadata zones starting from sb_zone */
2433 zid = dmz_id(zmd, zmd->sb_zone);
2434 for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2435 zone = dmz_get(zmd, zid + i);
2436 if (!dmz_is_rnd(zone))
2437 goto err;
2438 set_bit(DMZ_META, &zone->flags);
2439 }
2440
2441 /* Load mapping table */
2442 ret = dmz_load_mapping(zmd);
2443 if (ret)
2444 goto err;
2445
2446 /*
2447 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2448 * blocks and enough blocks to be able to cache the bitmap blocks of
2449 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2450 * the cache to add 512 more metadata blocks.
2451 */
2452 zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2453 zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2454 zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2455 zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2456 zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2457
2458 /* Metadata cache shrinker */
2459 ret = register_shrinker(&zmd->mblk_shrinker);
2460 if (ret) {
2461 dmz_dev_err(dev, "Register metadata cache shrinker failed");
2462 goto err;
2463 }
2464
2465 dmz_dev_info(dev, "Host-%s zoned block device",
2466 bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2467 "aware" : "managed");
2468 dmz_dev_info(dev, " %llu 512-byte logical sectors",
2469 (u64)dev->capacity);
2470 dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors",
2471 dev->nr_zones, (u64)dev->zone_nr_sectors);
2472 dmz_dev_info(dev, " %u metadata zones",
2473 zmd->nr_meta_zones * 2);
2474 dmz_dev_info(dev, " %u data zones for %u chunks",
2475 zmd->nr_data_zones, zmd->nr_chunks);
2476 dmz_dev_info(dev, " %u random zones (%u unmapped)",
2477 zmd->nr_rnd, atomic_read(&zmd->unmap_nr_rnd));
2478 dmz_dev_info(dev, " %u sequential zones (%u unmapped)",
2479 zmd->nr_seq, atomic_read(&zmd->unmap_nr_seq));
2480 dmz_dev_info(dev, " %u reserved sequential data zones",
2481 zmd->nr_reserved_seq);
2482
2483 dmz_dev_debug(dev, "Format:");
2484 dmz_dev_debug(dev, "%u metadata blocks per set (%u max cache)",
2485 zmd->nr_meta_blocks, zmd->max_nr_mblks);
2486 dmz_dev_debug(dev, " %u data zone mapping blocks",
2487 zmd->nr_map_blocks);
2488 dmz_dev_debug(dev, " %u bitmap blocks",
2489 zmd->nr_bitmap_blocks);
2490
2491 *metadata = zmd;
2492
2493 return 0;
2494 err:
2495 dmz_cleanup_metadata(zmd);
2496 kfree(zmd);
2497 *metadata = NULL;
2498
2499 return ret;
2500 }
2501
2502 /*
2503 * Cleanup the zoned metadata resources.
2504 */
2505 void dmz_dtr_metadata(struct dmz_metadata *zmd)
2506 {
2507 unregister_shrinker(&zmd->mblk_shrinker);
2508 dmz_cleanup_metadata(zmd);
2509 kfree(zmd);
2510 }
2511
2512 /*
2513 * Check zone information on resume.
2514 */
2515 int dmz_resume_metadata(struct dmz_metadata *zmd)
2516 {
2517 struct dmz_dev *dev = zmd->dev;
2518 struct dm_zone *zone;
2519 sector_t wp_block;
2520 unsigned int i;
2521 int ret;
2522
2523 /* Check zones */
2524 for (i = 0; i < dev->nr_zones; i++) {
2525 zone = dmz_get(zmd, i);
2526 if (!zone) {
2527 dmz_dev_err(dev, "Unable to get zone %u", i);
2528 return -EIO;
2529 }
2530
2531 wp_block = zone->wp_block;
2532
2533 ret = dmz_update_zone(zmd, zone);
2534 if (ret) {
2535 dmz_dev_err(dev, "Broken zone %u", i);
2536 return ret;
2537 }
2538
2539 if (dmz_is_offline(zone)) {
2540 dmz_dev_warn(dev, "Zone %u is offline", i);
2541 continue;
2542 }
2543
2544 /* Check write pointer */
2545 if (!dmz_is_seq(zone))
2546 zone->wp_block = 0;
2547 else if (zone->wp_block != wp_block) {
2548 dmz_dev_err(dev, "Zone %u: Invalid wp (%llu / %llu)",
2549 i, (u64)zone->wp_block, (u64)wp_block);
2550 zone->wp_block = wp_block;
2551 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
2552 dev->zone_nr_blocks - zone->wp_block);
2553 }
2554 }
2555
2556 return 0;
2557 }
Linux with added FPC-III support