Commit | Line | Data |
---|---|---|
bae9a0aa | 1 | // SPDX-License-Identifier: GPL-2.0-only |
3b1a94c8 DLM |
2 | /* |
3 | * Copyright (C) 2017 Western Digital Corporation or its affiliates. | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm-zoned.h" | |
9 | ||
10 | #include <linux/module.h> | |
11 | #include <linux/crc32.h> | |
bd976e52 | 12 | #include <linux/sched/mm.h> |
3b1a94c8 DLM |
13 | |
14 | #define DM_MSG_PREFIX "zoned metadata" | |
15 | ||
16 | /* | |
17 | * Metadata version. | |
18 | */ | |
bd5c4031 | 19 | #define DMZ_META_VER 2 |
3b1a94c8 DLM |
20 | |
21 | /* | |
22 | * On-disk super block magic. | |
23 | */ | |
24 | #define DMZ_MAGIC ((((unsigned int)('D')) << 24) | \ | |
25 | (((unsigned int)('Z')) << 16) | \ | |
26 | (((unsigned int)('B')) << 8) | \ | |
27 | ((unsigned int)('D'))) | |
28 | ||
29 | /* | |
30 | * On disk super block. | |
31 | * This uses only 512 B but uses on disk a full 4KB block. This block is | |
32 | * followed on disk by the mapping table of chunks to zones and the bitmap | |
33 | * blocks indicating zone block validity. | |
34 | * The overall resulting metadata format is: | |
35 | * (1) Super block (1 block) | |
36 | * (2) Chunk mapping table (nr_map_blocks) | |
37 | * (3) Bitmap blocks (nr_bitmap_blocks) | |
ad1bd578 | 38 | * All metadata blocks are stored in conventional zones, starting from |
3b1a94c8 DLM |
39 | * the first conventional zone found on disk. |
40 | */ | |
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 | ||
bd5c4031 HR |
72 | /* DM-Zoned label */ |
73 | u8 dmz_label[32]; /* 80 */ | |
74 | ||
75 | /* DM-Zoned UUID */ | |
76 | u8 dmz_uuid[16]; /* 96 */ | |
77 | ||
78 | /* Device UUID */ | |
79 | u8 dev_uuid[16]; /* 112 */ | |
80 | ||
3b1a94c8 | 81 | /* Padding to full 512B sector */ |
bd5c4031 | 82 | u8 reserved[400]; /* 512 */ |
3b1a94c8 DLM |
83 | }; |
84 | ||
85 | /* | |
86 | * Chunk mapping entry: entries are indexed by chunk number | |
87 | * and give the zone ID (dzone_id) mapping the chunk on disk. | |
88 | * This zone may be sequential or random. If it is a sequential | |
89 | * zone, a second zone (bzone_id) used as a write buffer may | |
90 | * also be specified. This second zone will always be a randomly | |
91 | * writeable zone. | |
92 | */ | |
93 | struct dmz_map { | |
94 | __le32 dzone_id; | |
95 | __le32 bzone_id; | |
96 | }; | |
97 | ||
98 | /* | |
99 | * Chunk mapping table metadata: 512 8-bytes entries per 4KB block. | |
100 | */ | |
101 | #define DMZ_MAP_ENTRIES (DMZ_BLOCK_SIZE / sizeof(struct dmz_map)) | |
102 | #define DMZ_MAP_ENTRIES_SHIFT (ilog2(DMZ_MAP_ENTRIES)) | |
103 | #define DMZ_MAP_ENTRIES_MASK (DMZ_MAP_ENTRIES - 1) | |
104 | #define DMZ_MAP_UNMAPPED UINT_MAX | |
105 | ||
106 | /* | |
107 | * Meta data block descriptor (for cached metadata blocks). | |
108 | */ | |
109 | struct dmz_mblock { | |
110 | struct rb_node node; | |
111 | struct list_head link; | |
112 | sector_t no; | |
33c2865f | 113 | unsigned int ref; |
3b1a94c8 DLM |
114 | unsigned long state; |
115 | struct page *page; | |
116 | void *data; | |
117 | }; | |
118 | ||
119 | /* | |
120 | * Metadata block state flags. | |
121 | */ | |
122 | enum { | |
123 | DMZ_META_DIRTY, | |
124 | DMZ_META_READING, | |
125 | DMZ_META_WRITING, | |
126 | DMZ_META_ERROR, | |
127 | }; | |
128 | ||
129 | /* | |
130 | * Super block information (one per metadata set). | |
131 | */ | |
132 | struct dmz_sb { | |
133 | sector_t block; | |
bf28a3ba | 134 | struct dmz_dev *dev; |
3b1a94c8 DLM |
135 | struct dmz_mblock *mblk; |
136 | struct dmz_super *sb; | |
735bd7e4 | 137 | struct dm_zone *zone; |
3b1a94c8 DLM |
138 | }; |
139 | ||
140 | /* | |
141 | * In-memory metadata. | |
142 | */ | |
143 | struct dmz_metadata { | |
144 | struct dmz_dev *dev; | |
bd5c4031 | 145 | unsigned int nr_devs; |
3b1a94c8 | 146 | |
2234e732 | 147 | char devname[BDEVNAME_SIZE]; |
bd5c4031 HR |
148 | char label[BDEVNAME_SIZE]; |
149 | uuid_t uuid; | |
2234e732 | 150 | |
3b1a94c8 DLM |
151 | sector_t zone_bitmap_size; |
152 | unsigned int zone_nr_bitmap_blocks; | |
b3996295 | 153 | unsigned int zone_bits_per_mblk; |
3b1a94c8 | 154 | |
36820560 HR |
155 | sector_t zone_nr_blocks; |
156 | sector_t zone_nr_blocks_shift; | |
157 | ||
158 | sector_t zone_nr_sectors; | |
159 | sector_t zone_nr_sectors_shift; | |
160 | ||
3b1a94c8 DLM |
161 | unsigned int nr_bitmap_blocks; |
162 | unsigned int nr_map_blocks; | |
163 | ||
36820560 | 164 | unsigned int nr_zones; |
3b1a94c8 DLM |
165 | unsigned int nr_useable_zones; |
166 | unsigned int nr_meta_blocks; | |
167 | unsigned int nr_meta_zones; | |
168 | unsigned int nr_data_zones; | |
34f5affd | 169 | unsigned int nr_cache_zones; |
3b1a94c8 DLM |
170 | unsigned int nr_rnd_zones; |
171 | unsigned int nr_reserved_seq; | |
172 | unsigned int nr_chunks; | |
173 | ||
174 | /* Zone information array */ | |
175 | struct dm_zone *zones; | |
176 | ||
bd5c4031 | 177 | struct dmz_sb sb[3]; |
3b1a94c8 | 178 | unsigned int mblk_primary; |
bd5c4031 | 179 | unsigned int sb_version; |
3b1a94c8 DLM |
180 | u64 sb_gen; |
181 | unsigned int min_nr_mblks; | |
182 | unsigned int max_nr_mblks; | |
183 | atomic_t nr_mblks; | |
184 | struct rw_semaphore mblk_sem; | |
185 | struct mutex mblk_flush_lock; | |
186 | spinlock_t mblk_lock; | |
187 | struct rb_root mblk_rbtree; | |
188 | struct list_head mblk_lru_list; | |
189 | struct list_head mblk_dirty_list; | |
190 | struct shrinker mblk_shrinker; | |
191 | ||
192 | /* Zone allocation management */ | |
193 | struct mutex map_lock; | |
194 | struct dmz_mblock **map_mblk; | |
195 | unsigned int nr_rnd; | |
196 | atomic_t unmap_nr_rnd; | |
197 | struct list_head unmap_rnd_list; | |
198 | struct list_head map_rnd_list; | |
199 | ||
34f5affd HR |
200 | unsigned int nr_cache; |
201 | atomic_t unmap_nr_cache; | |
202 | struct list_head unmap_cache_list; | |
203 | struct list_head map_cache_list; | |
204 | ||
3b1a94c8 DLM |
205 | unsigned int nr_seq; |
206 | atomic_t unmap_nr_seq; | |
207 | struct list_head unmap_seq_list; | |
208 | struct list_head map_seq_list; | |
209 | ||
210 | atomic_t nr_reserved_seq_zones; | |
211 | struct list_head reserved_seq_zones_list; | |
212 | ||
213 | wait_queue_head_t free_wq; | |
214 | }; | |
215 | ||
ca1a7045 | 216 | #define dmz_zmd_info(zmd, format, args...) \ |
bd5c4031 | 217 | DMINFO("(%s): " format, (zmd)->label, ## args) |
ca1a7045 HR |
218 | |
219 | #define dmz_zmd_err(zmd, format, args...) \ | |
bd5c4031 | 220 | DMERR("(%s): " format, (zmd)->label, ## args) |
ca1a7045 HR |
221 | |
222 | #define dmz_zmd_warn(zmd, format, args...) \ | |
bd5c4031 | 223 | DMWARN("(%s): " format, (zmd)->label, ## args) |
ca1a7045 HR |
224 | |
225 | #define dmz_zmd_debug(zmd, format, args...) \ | |
bd5c4031 | 226 | DMDEBUG("(%s): " format, (zmd)->label, ## args) |
3b1a94c8 DLM |
227 | /* |
228 | * Various accessors | |
229 | */ | |
bd5c4031 HR |
230 | static unsigned int dmz_dev_zone_id(struct dmz_metadata *zmd, struct dm_zone *zone) |
231 | { | |
232 | unsigned int zone_id; | |
233 | ||
234 | if (WARN_ON(!zone)) | |
235 | return 0; | |
236 | ||
237 | zone_id = zone->id; | |
238 | if (zmd->nr_devs > 1 && | |
239 | (zone_id >= zmd->dev[1].zone_offset)) | |
240 | zone_id -= zmd->dev[1].zone_offset; | |
241 | return zone_id; | |
242 | } | |
243 | ||
3b1a94c8 DLM |
244 | sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone) |
245 | { | |
bd5c4031 HR |
246 | unsigned int zone_id = dmz_dev_zone_id(zmd, zone); |
247 | ||
248 | return (sector_t)zone_id << zmd->zone_nr_sectors_shift; | |
3b1a94c8 DLM |
249 | } |
250 | ||
251 | sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone) | |
252 | { | |
bd5c4031 HR |
253 | unsigned int zone_id = dmz_dev_zone_id(zmd, zone); |
254 | ||
255 | return (sector_t)zone_id << zmd->zone_nr_blocks_shift; | |
3b1a94c8 DLM |
256 | } |
257 | ||
bf28a3ba HR |
258 | struct dmz_dev *dmz_zone_to_dev(struct dmz_metadata *zmd, struct dm_zone *zone) |
259 | { | |
bd5c4031 HR |
260 | if (WARN_ON(!zone)) |
261 | return &zmd->dev[0]; | |
262 | ||
263 | if (zmd->nr_devs > 1 && | |
264 | zone->id >= zmd->dev[1].zone_offset) | |
265 | return &zmd->dev[1]; | |
266 | ||
bf28a3ba HR |
267 | return &zmd->dev[0]; |
268 | } | |
269 | ||
36820560 HR |
270 | unsigned int dmz_zone_nr_blocks(struct dmz_metadata *zmd) |
271 | { | |
272 | return zmd->zone_nr_blocks; | |
273 | } | |
274 | ||
275 | unsigned int dmz_zone_nr_blocks_shift(struct dmz_metadata *zmd) | |
276 | { | |
277 | return zmd->zone_nr_blocks_shift; | |
278 | } | |
279 | ||
280 | unsigned int dmz_zone_nr_sectors(struct dmz_metadata *zmd) | |
281 | { | |
282 | return zmd->zone_nr_sectors; | |
283 | } | |
284 | ||
285 | unsigned int dmz_zone_nr_sectors_shift(struct dmz_metadata *zmd) | |
286 | { | |
287 | return zmd->zone_nr_sectors_shift; | |
288 | } | |
289 | ||
bc3d5717 HR |
290 | unsigned int dmz_nr_zones(struct dmz_metadata *zmd) |
291 | { | |
36820560 | 292 | return zmd->nr_zones; |
bc3d5717 HR |
293 | } |
294 | ||
3b1a94c8 DLM |
295 | unsigned int dmz_nr_chunks(struct dmz_metadata *zmd) |
296 | { | |
297 | return zmd->nr_chunks; | |
298 | } | |
299 | ||
300 | unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd) | |
301 | { | |
302 | return zmd->nr_rnd; | |
303 | } | |
304 | ||
305 | unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd) | |
306 | { | |
307 | return atomic_read(&zmd->unmap_nr_rnd); | |
308 | } | |
309 | ||
34f5affd HR |
310 | unsigned int dmz_nr_cache_zones(struct dmz_metadata *zmd) |
311 | { | |
312 | return zmd->nr_cache; | |
313 | } | |
314 | ||
315 | unsigned int dmz_nr_unmap_cache_zones(struct dmz_metadata *zmd) | |
316 | { | |
317 | return atomic_read(&zmd->unmap_nr_cache); | |
318 | } | |
319 | ||
bc3d5717 HR |
320 | unsigned int dmz_nr_seq_zones(struct dmz_metadata *zmd) |
321 | { | |
322 | return zmd->nr_seq; | |
323 | } | |
324 | ||
325 | unsigned int dmz_nr_unmap_seq_zones(struct dmz_metadata *zmd) | |
326 | { | |
327 | return atomic_read(&zmd->unmap_nr_seq); | |
328 | } | |
329 | ||
2234e732 HR |
330 | const char *dmz_metadata_label(struct dmz_metadata *zmd) |
331 | { | |
bd5c4031 | 332 | return (const char *)zmd->label; |
2234e732 HR |
333 | } |
334 | ||
d0e21ce4 HR |
335 | bool dmz_check_dev(struct dmz_metadata *zmd) |
336 | { | |
bd5c4031 HR |
337 | unsigned int i; |
338 | ||
339 | for (i = 0; i < zmd->nr_devs; i++) { | |
340 | if (!dmz_check_bdev(&zmd->dev[i])) | |
341 | return false; | |
342 | } | |
343 | return true; | |
d0e21ce4 HR |
344 | } |
345 | ||
346 | bool dmz_dev_is_dying(struct dmz_metadata *zmd) | |
347 | { | |
bd5c4031 HR |
348 | unsigned int i; |
349 | ||
350 | for (i = 0; i < zmd->nr_devs; i++) { | |
351 | if (dmz_bdev_is_dying(&zmd->dev[i])) | |
352 | return true; | |
353 | } | |
354 | return false; | |
d0e21ce4 HR |
355 | } |
356 | ||
3b1a94c8 DLM |
357 | /* |
358 | * Lock/unlock mapping table. | |
359 | * The map lock also protects all the zone lists. | |
360 | */ | |
361 | void dmz_lock_map(struct dmz_metadata *zmd) | |
362 | { | |
363 | mutex_lock(&zmd->map_lock); | |
364 | } | |
365 | ||
366 | void dmz_unlock_map(struct dmz_metadata *zmd) | |
367 | { | |
368 | mutex_unlock(&zmd->map_lock); | |
369 | } | |
370 | ||
371 | /* | |
372 | * Lock/unlock metadata access. This is a "read" lock on a semaphore | |
373 | * that prevents metadata flush from running while metadata are being | |
374 | * modified. The actual metadata write mutual exclusion is achieved with | |
ad1bd578 | 375 | * the map lock and zone state management (active and reclaim state are |
3b1a94c8 DLM |
376 | * mutually exclusive). |
377 | */ | |
378 | void dmz_lock_metadata(struct dmz_metadata *zmd) | |
379 | { | |
380 | down_read(&zmd->mblk_sem); | |
381 | } | |
382 | ||
383 | void dmz_unlock_metadata(struct dmz_metadata *zmd) | |
384 | { | |
385 | up_read(&zmd->mblk_sem); | |
386 | } | |
387 | ||
388 | /* | |
389 | * Lock/unlock flush: prevent concurrent executions | |
390 | * of dmz_flush_metadata as well as metadata modification in reclaim | |
391 | * while flush is being executed. | |
392 | */ | |
393 | void dmz_lock_flush(struct dmz_metadata *zmd) | |
394 | { | |
395 | mutex_lock(&zmd->mblk_flush_lock); | |
396 | } | |
397 | ||
398 | void dmz_unlock_flush(struct dmz_metadata *zmd) | |
399 | { | |
400 | mutex_unlock(&zmd->mblk_flush_lock); | |
401 | } | |
402 | ||
403 | /* | |
404 | * Allocate a metadata block. | |
405 | */ | |
406 | static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd, | |
407 | sector_t mblk_no) | |
408 | { | |
409 | struct dmz_mblock *mblk = NULL; | |
410 | ||
411 | /* See if we can reuse cached blocks */ | |
412 | if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) { | |
413 | spin_lock(&zmd->mblk_lock); | |
414 | mblk = list_first_entry_or_null(&zmd->mblk_lru_list, | |
415 | struct dmz_mblock, link); | |
416 | if (mblk) { | |
417 | list_del_init(&mblk->link); | |
418 | rb_erase(&mblk->node, &zmd->mblk_rbtree); | |
419 | mblk->no = mblk_no; | |
420 | } | |
421 | spin_unlock(&zmd->mblk_lock); | |
422 | if (mblk) | |
423 | return mblk; | |
424 | } | |
425 | ||
426 | /* Allocate a new block */ | |
427 | mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO); | |
428 | if (!mblk) | |
429 | return NULL; | |
430 | ||
431 | mblk->page = alloc_page(GFP_NOIO); | |
432 | if (!mblk->page) { | |
433 | kfree(mblk); | |
434 | return NULL; | |
435 | } | |
436 | ||
437 | RB_CLEAR_NODE(&mblk->node); | |
438 | INIT_LIST_HEAD(&mblk->link); | |
33c2865f | 439 | mblk->ref = 0; |
3b1a94c8 DLM |
440 | mblk->state = 0; |
441 | mblk->no = mblk_no; | |
442 | mblk->data = page_address(mblk->page); | |
443 | ||
444 | atomic_inc(&zmd->nr_mblks); | |
445 | ||
446 | return mblk; | |
447 | } | |
448 | ||
449 | /* | |
450 | * Free a metadata block. | |
451 | */ | |
452 | static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk) | |
453 | { | |
454 | __free_pages(mblk->page, 0); | |
455 | kfree(mblk); | |
456 | ||
457 | atomic_dec(&zmd->nr_mblks); | |
458 | } | |
459 | ||
460 | /* | |
461 | * Insert a metadata block in the rbtree. | |
462 | */ | |
463 | static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk) | |
464 | { | |
465 | struct rb_root *root = &zmd->mblk_rbtree; | |
466 | struct rb_node **new = &(root->rb_node), *parent = NULL; | |
467 | struct dmz_mblock *b; | |
468 | ||
469 | /* Figure out where to put the new node */ | |
470 | while (*new) { | |
471 | b = container_of(*new, struct dmz_mblock, node); | |
472 | parent = *new; | |
473 | new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right); | |
474 | } | |
475 | ||
476 | /* Add new node and rebalance tree */ | |
477 | rb_link_node(&mblk->node, parent, new); | |
478 | rb_insert_color(&mblk->node, root); | |
479 | } | |
480 | ||
481 | /* | |
3d4e7383 DLM |
482 | * Lookup a metadata block in the rbtree. If the block is found, increment |
483 | * its reference count. | |
3b1a94c8 | 484 | */ |
3d4e7383 DLM |
485 | static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd, |
486 | sector_t mblk_no) | |
3b1a94c8 DLM |
487 | { |
488 | struct rb_root *root = &zmd->mblk_rbtree; | |
489 | struct rb_node *node = root->rb_node; | |
490 | struct dmz_mblock *mblk; | |
491 | ||
492 | while (node) { | |
493 | mblk = container_of(node, struct dmz_mblock, node); | |
3d4e7383 DLM |
494 | if (mblk->no == mblk_no) { |
495 | /* | |
496 | * If this is the first reference to the block, | |
497 | * remove it from the LRU list. | |
498 | */ | |
499 | mblk->ref++; | |
500 | if (mblk->ref == 1 && | |
501 | !test_bit(DMZ_META_DIRTY, &mblk->state)) | |
502 | list_del_init(&mblk->link); | |
3b1a94c8 | 503 | return mblk; |
3d4e7383 | 504 | } |
3b1a94c8 DLM |
505 | node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right; |
506 | } | |
507 | ||
508 | return NULL; | |
509 | } | |
510 | ||
511 | /* | |
512 | * Metadata block BIO end callback. | |
513 | */ | |
514 | static void dmz_mblock_bio_end_io(struct bio *bio) | |
515 | { | |
516 | struct dmz_mblock *mblk = bio->bi_private; | |
517 | int flag; | |
518 | ||
519 | if (bio->bi_status) | |
520 | set_bit(DMZ_META_ERROR, &mblk->state); | |
521 | ||
522 | if (bio_op(bio) == REQ_OP_WRITE) | |
523 | flag = DMZ_META_WRITING; | |
524 | else | |
525 | flag = DMZ_META_READING; | |
526 | ||
527 | clear_bit_unlock(flag, &mblk->state); | |
528 | smp_mb__after_atomic(); | |
529 | wake_up_bit(&mblk->state, flag); | |
530 | ||
531 | bio_put(bio); | |
532 | } | |
533 | ||
534 | /* | |
3d4e7383 | 535 | * Read an uncached metadata block from disk and add it to the cache. |
3b1a94c8 | 536 | */ |
3d4e7383 DLM |
537 | static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd, |
538 | sector_t mblk_no) | |
3b1a94c8 | 539 | { |
3d4e7383 | 540 | struct dmz_mblock *mblk, *m; |
3b1a94c8 | 541 | sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no; |
bf28a3ba | 542 | struct dmz_dev *dev = zmd->sb[zmd->mblk_primary].dev; |
3b1a94c8 DLM |
543 | struct bio *bio; |
544 | ||
bf28a3ba | 545 | if (dmz_bdev_is_dying(dev)) |
75d66ffb DF |
546 | return ERR_PTR(-EIO); |
547 | ||
3d4e7383 | 548 | /* Get a new block and a BIO to read it */ |
3b1a94c8 DLM |
549 | mblk = dmz_alloc_mblock(zmd, mblk_no); |
550 | if (!mblk) | |
75d66ffb | 551 | return ERR_PTR(-ENOMEM); |
3b1a94c8 | 552 | |
3b1a94c8 DLM |
553 | bio = bio_alloc(GFP_NOIO, 1); |
554 | if (!bio) { | |
555 | dmz_free_mblock(zmd, mblk); | |
75d66ffb | 556 | return ERR_PTR(-ENOMEM); |
3b1a94c8 DLM |
557 | } |
558 | ||
3d4e7383 DLM |
559 | spin_lock(&zmd->mblk_lock); |
560 | ||
561 | /* | |
562 | * Make sure that another context did not start reading | |
563 | * the block already. | |
564 | */ | |
565 | m = dmz_get_mblock_fast(zmd, mblk_no); | |
566 | if (m) { | |
567 | spin_unlock(&zmd->mblk_lock); | |
568 | dmz_free_mblock(zmd, mblk); | |
569 | bio_put(bio); | |
570 | return m; | |
571 | } | |
572 | ||
573 | mblk->ref++; | |
574 | set_bit(DMZ_META_READING, &mblk->state); | |
575 | dmz_insert_mblock(zmd, mblk); | |
576 | ||
577 | spin_unlock(&zmd->mblk_lock); | |
578 | ||
579 | /* Submit read BIO */ | |
3b1a94c8 | 580 | bio->bi_iter.bi_sector = dmz_blk2sect(block); |
bf28a3ba | 581 | bio_set_dev(bio, dev->bdev); |
3b1a94c8 DLM |
582 | bio->bi_private = mblk; |
583 | bio->bi_end_io = dmz_mblock_bio_end_io; | |
584 | bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO); | |
585 | bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0); | |
586 | submit_bio(bio); | |
587 | ||
588 | return mblk; | |
589 | } | |
590 | ||
591 | /* | |
592 | * Free metadata blocks. | |
593 | */ | |
594 | static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd, | |
595 | unsigned long limit) | |
596 | { | |
597 | struct dmz_mblock *mblk; | |
598 | unsigned long count = 0; | |
599 | ||
600 | if (!zmd->max_nr_mblks) | |
601 | return 0; | |
602 | ||
603 | while (!list_empty(&zmd->mblk_lru_list) && | |
604 | atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks && | |
605 | count < limit) { | |
606 | mblk = list_first_entry(&zmd->mblk_lru_list, | |
607 | struct dmz_mblock, link); | |
608 | list_del_init(&mblk->link); | |
609 | rb_erase(&mblk->node, &zmd->mblk_rbtree); | |
610 | dmz_free_mblock(zmd, mblk); | |
611 | count++; | |
612 | } | |
613 | ||
614 | return count; | |
615 | } | |
616 | ||
617 | /* | |
618 | * For mblock shrinker: get the number of unused metadata blocks in the cache. | |
619 | */ | |
620 | static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink, | |
621 | struct shrink_control *sc) | |
622 | { | |
623 | struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker); | |
624 | ||
625 | return atomic_read(&zmd->nr_mblks); | |
626 | } | |
627 | ||
628 | /* | |
629 | * For mblock shrinker: scan unused metadata blocks and shrink the cache. | |
630 | */ | |
631 | static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink, | |
632 | struct shrink_control *sc) | |
633 | { | |
634 | struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker); | |
635 | unsigned long count; | |
636 | ||
637 | spin_lock(&zmd->mblk_lock); | |
638 | count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan); | |
639 | spin_unlock(&zmd->mblk_lock); | |
640 | ||
641 | return count ? count : SHRINK_STOP; | |
642 | } | |
643 | ||
644 | /* | |
645 | * Release a metadata block. | |
646 | */ | |
647 | static void dmz_release_mblock(struct dmz_metadata *zmd, | |
648 | struct dmz_mblock *mblk) | |
649 | { | |
650 | ||
651 | if (!mblk) | |
652 | return; | |
653 | ||
654 | spin_lock(&zmd->mblk_lock); | |
655 | ||
33c2865f DLM |
656 | mblk->ref--; |
657 | if (mblk->ref == 0) { | |
3b1a94c8 DLM |
658 | if (test_bit(DMZ_META_ERROR, &mblk->state)) { |
659 | rb_erase(&mblk->node, &zmd->mblk_rbtree); | |
660 | dmz_free_mblock(zmd, mblk); | |
661 | } else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) { | |
662 | list_add_tail(&mblk->link, &zmd->mblk_lru_list); | |
663 | dmz_shrink_mblock_cache(zmd, 1); | |
664 | } | |
665 | } | |
666 | ||
667 | spin_unlock(&zmd->mblk_lock); | |
668 | } | |
669 | ||
670 | /* | |
671 | * Get a metadata block from the rbtree. If the block | |
672 | * is not present, read it from disk. | |
673 | */ | |
674 | static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd, | |
675 | sector_t mblk_no) | |
676 | { | |
677 | struct dmz_mblock *mblk; | |
bf28a3ba | 678 | struct dmz_dev *dev = zmd->sb[zmd->mblk_primary].dev; |
3b1a94c8 DLM |
679 | |
680 | /* Check rbtree */ | |
681 | spin_lock(&zmd->mblk_lock); | |
3d4e7383 | 682 | mblk = dmz_get_mblock_fast(zmd, mblk_no); |
3b1a94c8 DLM |
683 | spin_unlock(&zmd->mblk_lock); |
684 | ||
685 | if (!mblk) { | |
686 | /* Cache miss: read the block from disk */ | |
3d4e7383 | 687 | mblk = dmz_get_mblock_slow(zmd, mblk_no); |
75d66ffb DF |
688 | if (IS_ERR(mblk)) |
689 | return mblk; | |
3b1a94c8 DLM |
690 | } |
691 | ||
692 | /* Wait for on-going read I/O and check for error */ | |
693 | wait_on_bit_io(&mblk->state, DMZ_META_READING, | |
694 | TASK_UNINTERRUPTIBLE); | |
695 | if (test_bit(DMZ_META_ERROR, &mblk->state)) { | |
696 | dmz_release_mblock(zmd, mblk); | |
bf28a3ba | 697 | dmz_check_bdev(dev); |
3b1a94c8 DLM |
698 | return ERR_PTR(-EIO); |
699 | } | |
700 | ||
701 | return mblk; | |
702 | } | |
703 | ||
704 | /* | |
705 | * Mark a metadata block dirty. | |
706 | */ | |
707 | static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk) | |
708 | { | |
709 | spin_lock(&zmd->mblk_lock); | |
710 | if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state)) | |
711 | list_add_tail(&mblk->link, &zmd->mblk_dirty_list); | |
712 | spin_unlock(&zmd->mblk_lock); | |
713 | } | |
714 | ||
715 | /* | |
716 | * Issue a metadata block write BIO. | |
717 | */ | |
75d66ffb DF |
718 | static int dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk, |
719 | unsigned int set) | |
3b1a94c8 | 720 | { |
bf28a3ba | 721 | struct dmz_dev *dev = zmd->sb[set].dev; |
3b1a94c8 DLM |
722 | sector_t block = zmd->sb[set].block + mblk->no; |
723 | struct bio *bio; | |
724 | ||
bf28a3ba | 725 | if (dmz_bdev_is_dying(dev)) |
75d66ffb DF |
726 | return -EIO; |
727 | ||
3b1a94c8 DLM |
728 | bio = bio_alloc(GFP_NOIO, 1); |
729 | if (!bio) { | |
730 | set_bit(DMZ_META_ERROR, &mblk->state); | |
75d66ffb | 731 | return -ENOMEM; |
3b1a94c8 DLM |
732 | } |
733 | ||
734 | set_bit(DMZ_META_WRITING, &mblk->state); | |
735 | ||
736 | bio->bi_iter.bi_sector = dmz_blk2sect(block); | |
bf28a3ba | 737 | bio_set_dev(bio, dev->bdev); |
3b1a94c8 DLM |
738 | bio->bi_private = mblk; |
739 | bio->bi_end_io = dmz_mblock_bio_end_io; | |
740 | bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO); | |
741 | bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0); | |
742 | submit_bio(bio); | |
75d66ffb DF |
743 | |
744 | return 0; | |
3b1a94c8 DLM |
745 | } |
746 | ||
747 | /* | |
748 | * Read/write a metadata block. | |
749 | */ | |
bf28a3ba HR |
750 | static int dmz_rdwr_block(struct dmz_dev *dev, int op, |
751 | sector_t block, struct page *page) | |
3b1a94c8 DLM |
752 | { |
753 | struct bio *bio; | |
754 | int ret; | |
755 | ||
bd5c4031 HR |
756 | if (WARN_ON(!dev)) |
757 | return -EIO; | |
758 | ||
bf28a3ba | 759 | if (dmz_bdev_is_dying(dev)) |
75d66ffb DF |
760 | return -EIO; |
761 | ||
3b1a94c8 DLM |
762 | bio = bio_alloc(GFP_NOIO, 1); |
763 | if (!bio) | |
764 | return -ENOMEM; | |
765 | ||
766 | bio->bi_iter.bi_sector = dmz_blk2sect(block); | |
bf28a3ba | 767 | bio_set_dev(bio, dev->bdev); |
3b1a94c8 DLM |
768 | bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO); |
769 | bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0); | |
770 | ret = submit_bio_wait(bio); | |
771 | bio_put(bio); | |
772 | ||
e7fad909 | 773 | if (ret) |
bf28a3ba | 774 | dmz_check_bdev(dev); |
3b1a94c8 DLM |
775 | return ret; |
776 | } | |
777 | ||
778 | /* | |
779 | * Write super block of the specified metadata set. | |
780 | */ | |
781 | static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set) | |
782 | { | |
3b1a94c8 DLM |
783 | struct dmz_mblock *mblk = zmd->sb[set].mblk; |
784 | struct dmz_super *sb = zmd->sb[set].sb; | |
bf28a3ba | 785 | struct dmz_dev *dev = zmd->sb[set].dev; |
bd5c4031 | 786 | sector_t sb_block; |
3b1a94c8 DLM |
787 | u64 sb_gen = zmd->sb_gen + 1; |
788 | int ret; | |
789 | ||
790 | sb->magic = cpu_to_le32(DMZ_MAGIC); | |
bd5c4031 HR |
791 | |
792 | sb->version = cpu_to_le32(zmd->sb_version); | |
793 | if (zmd->sb_version > 1) { | |
794 | BUILD_BUG_ON(UUID_SIZE != 16); | |
795 | export_uuid(sb->dmz_uuid, &zmd->uuid); | |
796 | memcpy(sb->dmz_label, zmd->label, BDEVNAME_SIZE); | |
797 | export_uuid(sb->dev_uuid, &dev->uuid); | |
798 | } | |
3b1a94c8 DLM |
799 | |
800 | sb->gen = cpu_to_le64(sb_gen); | |
801 | ||
bd5c4031 HR |
802 | /* |
803 | * The metadata always references the absolute block address, | |
804 | * ie relative to the entire block range, not the per-device | |
805 | * block address. | |
806 | */ | |
807 | sb_block = zmd->sb[set].zone->id << zmd->zone_nr_blocks_shift; | |
808 | sb->sb_block = cpu_to_le64(sb_block); | |
3b1a94c8 DLM |
809 | sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks); |
810 | sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq); | |
811 | sb->nr_chunks = cpu_to_le32(zmd->nr_chunks); | |
812 | ||
813 | sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks); | |
814 | sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks); | |
815 | ||
816 | sb->crc = 0; | |
817 | sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE)); | |
818 | ||
bd5c4031 HR |
819 | ret = dmz_rdwr_block(dev, REQ_OP_WRITE, zmd->sb[set].block, |
820 | mblk->page); | |
3b1a94c8 | 821 | if (ret == 0) |
bf28a3ba | 822 | ret = blkdev_issue_flush(dev->bdev, GFP_NOIO, NULL); |
3b1a94c8 DLM |
823 | |
824 | return ret; | |
825 | } | |
826 | ||
827 | /* | |
828 | * Write dirty metadata blocks to the specified set. | |
829 | */ | |
830 | static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd, | |
831 | struct list_head *write_list, | |
832 | unsigned int set) | |
833 | { | |
834 | struct dmz_mblock *mblk; | |
bf28a3ba | 835 | struct dmz_dev *dev = zmd->sb[set].dev; |
3b1a94c8 | 836 | struct blk_plug plug; |
75d66ffb | 837 | int ret = 0, nr_mblks_submitted = 0; |
3b1a94c8 DLM |
838 | |
839 | /* Issue writes */ | |
840 | blk_start_plug(&plug); | |
75d66ffb DF |
841 | list_for_each_entry(mblk, write_list, link) { |
842 | ret = dmz_write_mblock(zmd, mblk, set); | |
843 | if (ret) | |
844 | break; | |
845 | nr_mblks_submitted++; | |
846 | } | |
3b1a94c8 DLM |
847 | blk_finish_plug(&plug); |
848 | ||
849 | /* Wait for completion */ | |
850 | list_for_each_entry(mblk, write_list, link) { | |
75d66ffb DF |
851 | if (!nr_mblks_submitted) |
852 | break; | |
3b1a94c8 DLM |
853 | wait_on_bit_io(&mblk->state, DMZ_META_WRITING, |
854 | TASK_UNINTERRUPTIBLE); | |
855 | if (test_bit(DMZ_META_ERROR, &mblk->state)) { | |
856 | clear_bit(DMZ_META_ERROR, &mblk->state); | |
bf28a3ba | 857 | dmz_check_bdev(dev); |
3b1a94c8 DLM |
858 | ret = -EIO; |
859 | } | |
75d66ffb | 860 | nr_mblks_submitted--; |
3b1a94c8 DLM |
861 | } |
862 | ||
863 | /* Flush drive cache (this will also sync data) */ | |
864 | if (ret == 0) | |
bf28a3ba | 865 | ret = blkdev_issue_flush(dev->bdev, GFP_NOIO, NULL); |
3b1a94c8 DLM |
866 | |
867 | return ret; | |
868 | } | |
869 | ||
870 | /* | |
871 | * Log dirty metadata blocks. | |
872 | */ | |
873 | static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd, | |
874 | struct list_head *write_list) | |
875 | { | |
876 | unsigned int log_set = zmd->mblk_primary ^ 0x1; | |
877 | int ret; | |
878 | ||
879 | /* Write dirty blocks to the log */ | |
880 | ret = dmz_write_dirty_mblocks(zmd, write_list, log_set); | |
881 | if (ret) | |
882 | return ret; | |
883 | ||
884 | /* | |
885 | * No error so far: now validate the log by updating the | |
886 | * log index super block generation. | |
887 | */ | |
888 | ret = dmz_write_sb(zmd, log_set); | |
889 | if (ret) | |
890 | return ret; | |
891 | ||
892 | return 0; | |
893 | } | |
894 | ||
895 | /* | |
896 | * Flush dirty metadata blocks. | |
897 | */ | |
898 | int dmz_flush_metadata(struct dmz_metadata *zmd) | |
899 | { | |
900 | struct dmz_mblock *mblk; | |
901 | struct list_head write_list; | |
bf28a3ba | 902 | struct dmz_dev *dev; |
3b1a94c8 DLM |
903 | int ret; |
904 | ||
905 | if (WARN_ON(!zmd)) | |
906 | return 0; | |
907 | ||
908 | INIT_LIST_HEAD(&write_list); | |
909 | ||
910 | /* | |
911 | * Make sure that metadata blocks are stable before logging: take | |
912 | * the write lock on the metadata semaphore to prevent target BIOs | |
913 | * from modifying metadata. | |
914 | */ | |
915 | down_write(&zmd->mblk_sem); | |
bf28a3ba | 916 | dev = zmd->sb[zmd->mblk_primary].dev; |
3b1a94c8 DLM |
917 | |
918 | /* | |
919 | * This is called from the target flush work and reclaim work. | |
920 | * Concurrent execution is not allowed. | |
921 | */ | |
922 | dmz_lock_flush(zmd); | |
923 | ||
bf28a3ba | 924 | if (dmz_bdev_is_dying(dev)) { |
75d66ffb DF |
925 | ret = -EIO; |
926 | goto out; | |
927 | } | |
928 | ||
3b1a94c8 DLM |
929 | /* Get dirty blocks */ |
930 | spin_lock(&zmd->mblk_lock); | |
931 | list_splice_init(&zmd->mblk_dirty_list, &write_list); | |
932 | spin_unlock(&zmd->mblk_lock); | |
933 | ||
934 | /* If there are no dirty metadata blocks, just flush the device cache */ | |
935 | if (list_empty(&write_list)) { | |
bf28a3ba | 936 | ret = blkdev_issue_flush(dev->bdev, GFP_NOIO, NULL); |
e7fad909 | 937 | goto err; |
3b1a94c8 DLM |
938 | } |
939 | ||
940 | /* | |
941 | * The primary metadata set is still clean. Keep it this way until | |
942 | * all updates are successful in the secondary set. That is, use | |
943 | * the secondary set as a log. | |
944 | */ | |
945 | ret = dmz_log_dirty_mblocks(zmd, &write_list); | |
946 | if (ret) | |
e7fad909 | 947 | goto err; |
3b1a94c8 DLM |
948 | |
949 | /* | |
950 | * The log is on disk. It is now safe to update in place | |
951 | * in the primary metadata set. | |
952 | */ | |
953 | ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary); | |
954 | if (ret) | |
e7fad909 | 955 | goto err; |
3b1a94c8 DLM |
956 | |
957 | ret = dmz_write_sb(zmd, zmd->mblk_primary); | |
958 | if (ret) | |
e7fad909 | 959 | goto err; |
3b1a94c8 DLM |
960 | |
961 | while (!list_empty(&write_list)) { | |
962 | mblk = list_first_entry(&write_list, struct dmz_mblock, link); | |
963 | list_del_init(&mblk->link); | |
964 | ||
965 | spin_lock(&zmd->mblk_lock); | |
966 | clear_bit(DMZ_META_DIRTY, &mblk->state); | |
33c2865f | 967 | if (mblk->ref == 0) |
3b1a94c8 DLM |
968 | list_add_tail(&mblk->link, &zmd->mblk_lru_list); |
969 | spin_unlock(&zmd->mblk_lock); | |
970 | } | |
971 | ||
972 | zmd->sb_gen++; | |
973 | out: | |
3b1a94c8 DLM |
974 | dmz_unlock_flush(zmd); |
975 | up_write(&zmd->mblk_sem); | |
976 | ||
977 | return ret; | |
e7fad909 DF |
978 | |
979 | err: | |
980 | if (!list_empty(&write_list)) { | |
981 | spin_lock(&zmd->mblk_lock); | |
982 | list_splice(&write_list, &zmd->mblk_dirty_list); | |
983 | spin_unlock(&zmd->mblk_lock); | |
984 | } | |
bf28a3ba | 985 | if (!dmz_check_bdev(dev)) |
e7fad909 DF |
986 | ret = -EIO; |
987 | goto out; | |
3b1a94c8 DLM |
988 | } |
989 | ||
990 | /* | |
991 | * Check super block. | |
992 | */ | |
735bd7e4 | 993 | static int dmz_check_sb(struct dmz_metadata *zmd, unsigned int set) |
3b1a94c8 | 994 | { |
735bd7e4 | 995 | struct dmz_super *sb = zmd->sb[set].sb; |
bf28a3ba | 996 | struct dmz_dev *dev = zmd->sb[set].dev; |
3b1a94c8 | 997 | unsigned int nr_meta_zones, nr_data_zones; |
3b1a94c8 DLM |
998 | u32 crc, stored_crc; |
999 | u64 gen; | |
1000 | ||
bd5c4031 HR |
1001 | if (le32_to_cpu(sb->magic) != DMZ_MAGIC) { |
1002 | dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)", | |
1003 | DMZ_MAGIC, le32_to_cpu(sb->magic)); | |
1004 | return -ENXIO; | |
1005 | } | |
1006 | ||
1007 | zmd->sb_version = le32_to_cpu(sb->version); | |
1008 | if (zmd->sb_version > DMZ_META_VER) { | |
1009 | dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)", | |
1010 | DMZ_META_VER, zmd->sb_version); | |
1011 | return -EINVAL; | |
1012 | } | |
1013 | if ((zmd->sb_version < 1) && (set == 2)) { | |
1014 | dmz_dev_err(dev, "Tertiary superblocks are not supported"); | |
1015 | return -EINVAL; | |
1016 | } | |
1017 | ||
3b1a94c8 DLM |
1018 | gen = le64_to_cpu(sb->gen); |
1019 | stored_crc = le32_to_cpu(sb->crc); | |
1020 | sb->crc = 0; | |
1021 | crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE); | |
1022 | if (crc != stored_crc) { | |
1023 | dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)", | |
1024 | crc, stored_crc); | |
1025 | return -ENXIO; | |
1026 | } | |
1027 | ||
bd5c4031 HR |
1028 | if (zmd->sb_version > 1) { |
1029 | uuid_t sb_uuid; | |
1030 | ||
1031 | import_uuid(&sb_uuid, sb->dmz_uuid); | |
1032 | if (uuid_is_null(&sb_uuid)) { | |
1033 | dmz_dev_err(dev, "NULL DM-Zoned uuid"); | |
1034 | return -ENXIO; | |
1035 | } else if (uuid_is_null(&zmd->uuid)) { | |
1036 | uuid_copy(&zmd->uuid, &sb_uuid); | |
1037 | } else if (!uuid_equal(&zmd->uuid, &sb_uuid)) { | |
1038 | dmz_dev_err(dev, "mismatching DM-Zoned uuid, " | |
1039 | "is %pUl expected %pUl", | |
1040 | &sb_uuid, &zmd->uuid); | |
1041 | return -ENXIO; | |
1042 | } | |
1043 | if (!strlen(zmd->label)) | |
1044 | memcpy(zmd->label, sb->dmz_label, BDEVNAME_SIZE); | |
1045 | else if (memcmp(zmd->label, sb->dmz_label, BDEVNAME_SIZE)) { | |
1046 | dmz_dev_err(dev, "mismatching DM-Zoned label, " | |
1047 | "is %s expected %s", | |
1048 | sb->dmz_label, zmd->label); | |
1049 | return -ENXIO; | |
1050 | } | |
1051 | import_uuid(&dev->uuid, sb->dev_uuid); | |
1052 | if (uuid_is_null(&dev->uuid)) { | |
1053 | dmz_dev_err(dev, "NULL device uuid"); | |
1054 | return -ENXIO; | |
1055 | } | |
3b1a94c8 | 1056 | |
bd5c4031 HR |
1057 | if (set == 2) { |
1058 | /* | |
1059 | * Generation number should be 0, but it doesn't | |
1060 | * really matter if it isn't. | |
1061 | */ | |
1062 | if (gen != 0) | |
1063 | dmz_dev_warn(dev, "Invalid generation %llu", | |
1064 | gen); | |
1065 | return 0; | |
1066 | } | |
3b1a94c8 DLM |
1067 | } |
1068 | ||
36820560 HR |
1069 | nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + zmd->zone_nr_blocks - 1) |
1070 | >> zmd->zone_nr_blocks_shift; | |
3b1a94c8 DLM |
1071 | if (!nr_meta_zones || |
1072 | nr_meta_zones >= zmd->nr_rnd_zones) { | |
1073 | dmz_dev_err(dev, "Invalid number of metadata blocks"); | |
1074 | return -ENXIO; | |
1075 | } | |
1076 | ||
1077 | if (!le32_to_cpu(sb->nr_reserved_seq) || | |
1078 | le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) { | |
1079 | dmz_dev_err(dev, "Invalid number of reserved sequential zones"); | |
1080 | return -ENXIO; | |
1081 | } | |
1082 | ||
1083 | nr_data_zones = zmd->nr_useable_zones - | |
1084 | (nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq)); | |
1085 | if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) { | |
1086 | dmz_dev_err(dev, "Invalid number of chunks %u / %u", | |
1087 | le32_to_cpu(sb->nr_chunks), nr_data_zones); | |
1088 | return -ENXIO; | |
1089 | } | |
1090 | ||
1091 | /* OK */ | |
1092 | zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks); | |
1093 | zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq); | |
1094 | zmd->nr_chunks = le32_to_cpu(sb->nr_chunks); | |
1095 | zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks); | |
1096 | zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks); | |
1097 | zmd->nr_meta_zones = nr_meta_zones; | |
1098 | zmd->nr_data_zones = nr_data_zones; | |
1099 | ||
1100 | return 0; | |
1101 | } | |
1102 | ||
1103 | /* | |
1104 | * Read the first or second super block from disk. | |
1105 | */ | |
1106 | static int dmz_read_sb(struct dmz_metadata *zmd, unsigned int set) | |
1107 | { | |
35d0c96e HR |
1108 | dmz_zmd_debug(zmd, "read superblock set %d dev %s block %llu", |
1109 | set, zmd->sb[set].dev->name, | |
1110 | zmd->sb[set].block); | |
1111 | ||
bf28a3ba HR |
1112 | return dmz_rdwr_block(zmd->sb[set].dev, REQ_OP_READ, |
1113 | zmd->sb[set].block, zmd->sb[set].mblk->page); | |
3b1a94c8 DLM |
1114 | } |
1115 | ||
1116 | /* | |
1117 | * Determine the position of the secondary super blocks on disk. | |
1118 | * This is used only if a corruption of the primary super block | |
1119 | * is detected. | |
1120 | */ | |
1121 | static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd) | |
1122 | { | |
36820560 | 1123 | unsigned int zone_nr_blocks = zmd->zone_nr_blocks; |
3b1a94c8 DLM |
1124 | struct dmz_mblock *mblk; |
1125 | int i; | |
1126 | ||
1127 | /* Allocate a block */ | |
1128 | mblk = dmz_alloc_mblock(zmd, 0); | |
1129 | if (!mblk) | |
1130 | return -ENOMEM; | |
1131 | ||
1132 | zmd->sb[1].mblk = mblk; | |
1133 | zmd->sb[1].sb = mblk->data; | |
1134 | ||
1135 | /* Bad first super block: search for the second one */ | |
1136 | zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks; | |
735bd7e4 | 1137 | zmd->sb[1].zone = zmd->sb[0].zone + 1; |
1565929b | 1138 | zmd->sb[1].dev = zmd->sb[0].dev; |
3b1a94c8 DLM |
1139 | for (i = 0; i < zmd->nr_rnd_zones - 1; i++) { |
1140 | if (dmz_read_sb(zmd, 1) != 0) | |
1141 | break; | |
735bd7e4 HR |
1142 | if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC) { |
1143 | zmd->sb[1].zone += i; | |
3b1a94c8 | 1144 | return 0; |
735bd7e4 | 1145 | } |
3b1a94c8 DLM |
1146 | zmd->sb[1].block += zone_nr_blocks; |
1147 | } | |
1148 | ||
1149 | dmz_free_mblock(zmd, mblk); | |
1150 | zmd->sb[1].mblk = NULL; | |
735bd7e4 | 1151 | zmd->sb[1].zone = NULL; |
bf28a3ba | 1152 | zmd->sb[1].dev = NULL; |
3b1a94c8 DLM |
1153 | |
1154 | return -EIO; | |
1155 | } | |
1156 | ||
1157 | /* | |
1158 | * Read the first or second super block from disk. | |
1159 | */ | |
1160 | static int dmz_get_sb(struct dmz_metadata *zmd, unsigned int set) | |
1161 | { | |
1162 | struct dmz_mblock *mblk; | |
1163 | int ret; | |
1164 | ||
1165 | /* Allocate a block */ | |
1166 | mblk = dmz_alloc_mblock(zmd, 0); | |
1167 | if (!mblk) | |
1168 | return -ENOMEM; | |
1169 | ||
1170 | zmd->sb[set].mblk = mblk; | |
1171 | zmd->sb[set].sb = mblk->data; | |
1172 | ||
1173 | /* Read super block */ | |
1174 | ret = dmz_read_sb(zmd, set); | |
1175 | if (ret) { | |
1176 | dmz_free_mblock(zmd, mblk); | |
1177 | zmd->sb[set].mblk = NULL; | |
1178 | return ret; | |
1179 | } | |
1180 | ||
1181 | return 0; | |
1182 | } | |
1183 | ||
1184 | /* | |
1185 | * Recover a metadata set. | |
1186 | */ | |
1187 | static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set) | |
1188 | { | |
1189 | unsigned int src_set = dst_set ^ 0x1; | |
1190 | struct page *page; | |
1191 | int i, ret; | |
1192 | ||
bf28a3ba HR |
1193 | dmz_dev_warn(zmd->sb[dst_set].dev, |
1194 | "Metadata set %u invalid: recovering", dst_set); | |
3b1a94c8 DLM |
1195 | |
1196 | if (dst_set == 0) | |
735bd7e4 HR |
1197 | zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone); |
1198 | else | |
1199 | zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone); | |
3b1a94c8 | 1200 | |
4218a955 | 1201 | page = alloc_page(GFP_NOIO); |
3b1a94c8 DLM |
1202 | if (!page) |
1203 | return -ENOMEM; | |
1204 | ||
1205 | /* Copy metadata blocks */ | |
1206 | for (i = 1; i < zmd->nr_meta_blocks; i++) { | |
bf28a3ba | 1207 | ret = dmz_rdwr_block(zmd->sb[src_set].dev, REQ_OP_READ, |
3b1a94c8 DLM |
1208 | zmd->sb[src_set].block + i, page); |
1209 | if (ret) | |
1210 | goto out; | |
bf28a3ba | 1211 | ret = dmz_rdwr_block(zmd->sb[dst_set].dev, REQ_OP_WRITE, |
3b1a94c8 DLM |
1212 | zmd->sb[dst_set].block + i, page); |
1213 | if (ret) | |
1214 | goto out; | |
1215 | } | |
1216 | ||
1217 | /* Finalize with the super block */ | |
1218 | if (!zmd->sb[dst_set].mblk) { | |
1219 | zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0); | |
1220 | if (!zmd->sb[dst_set].mblk) { | |
1221 | ret = -ENOMEM; | |
1222 | goto out; | |
1223 | } | |
1224 | zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data; | |
1225 | } | |
1226 | ||
1227 | ret = dmz_write_sb(zmd, dst_set); | |
1228 | out: | |
1229 | __free_pages(page, 0); | |
1230 | ||
1231 | return ret; | |
1232 | } | |
1233 | ||
1234 | /* | |
1235 | * Get super block from disk. | |
1236 | */ | |
1237 | static int dmz_load_sb(struct dmz_metadata *zmd) | |
1238 | { | |
1239 | bool sb_good[2] = {false, false}; | |
1240 | u64 sb_gen[2] = {0, 0}; | |
1241 | int ret; | |
1242 | ||
735bd7e4 | 1243 | if (!zmd->sb[0].zone) { |
ca1a7045 | 1244 | dmz_zmd_err(zmd, "Primary super block zone not set"); |
735bd7e4 HR |
1245 | return -ENXIO; |
1246 | } | |
1247 | ||
3b1a94c8 | 1248 | /* Read and check the primary super block */ |
735bd7e4 | 1249 | zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone); |
bf28a3ba | 1250 | zmd->sb[0].dev = dmz_zone_to_dev(zmd, zmd->sb[0].zone); |
3b1a94c8 DLM |
1251 | ret = dmz_get_sb(zmd, 0); |
1252 | if (ret) { | |
bf28a3ba | 1253 | dmz_dev_err(zmd->sb[0].dev, "Read primary super block failed"); |
3b1a94c8 DLM |
1254 | return ret; |
1255 | } | |
1256 | ||
735bd7e4 | 1257 | ret = dmz_check_sb(zmd, 0); |
3b1a94c8 DLM |
1258 | |
1259 | /* Read and check secondary super block */ | |
1260 | if (ret == 0) { | |
1261 | sb_good[0] = true; | |
735bd7e4 HR |
1262 | if (!zmd->sb[1].zone) |
1263 | zmd->sb[1].zone = zmd->sb[0].zone + zmd->nr_meta_zones; | |
1264 | zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone); | |
1565929b | 1265 | zmd->sb[1].dev = zmd->sb[0].dev; |
3b1a94c8 DLM |
1266 | ret = dmz_get_sb(zmd, 1); |
1267 | } else | |
1268 | ret = dmz_lookup_secondary_sb(zmd); | |
1269 | ||
1270 | if (ret) { | |
bf28a3ba | 1271 | dmz_dev_err(zmd->sb[1].dev, "Read secondary super block failed"); |
3b1a94c8 DLM |
1272 | return ret; |
1273 | } | |
1274 | ||
735bd7e4 | 1275 | ret = dmz_check_sb(zmd, 1); |
3b1a94c8 DLM |
1276 | if (ret == 0) |
1277 | sb_good[1] = true; | |
1278 | ||
1279 | /* Use highest generation sb first */ | |
1280 | if (!sb_good[0] && !sb_good[1]) { | |
ca1a7045 | 1281 | dmz_zmd_err(zmd, "No valid super block found"); |
3b1a94c8 DLM |
1282 | return -EIO; |
1283 | } | |
1284 | ||
1285 | if (sb_good[0]) | |
1286 | sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen); | |
bf28a3ba | 1287 | else { |
3b1a94c8 | 1288 | ret = dmz_recover_mblocks(zmd, 0); |
bf28a3ba HR |
1289 | if (ret) { |
1290 | dmz_dev_err(zmd->sb[0].dev, | |
1291 | "Recovery of superblock 0 failed"); | |
1292 | return -EIO; | |
1293 | } | |
1294 | } | |
3b1a94c8 DLM |
1295 | |
1296 | if (sb_good[1]) | |
1297 | sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen); | |
bf28a3ba | 1298 | else { |
3b1a94c8 DLM |
1299 | ret = dmz_recover_mblocks(zmd, 1); |
1300 | ||
bf28a3ba HR |
1301 | if (ret) { |
1302 | dmz_dev_err(zmd->sb[1].dev, | |
1303 | "Recovery of superblock 1 failed"); | |
1304 | return -EIO; | |
1305 | } | |
3b1a94c8 DLM |
1306 | } |
1307 | ||
1308 | if (sb_gen[0] >= sb_gen[1]) { | |
1309 | zmd->sb_gen = sb_gen[0]; | |
1310 | zmd->mblk_primary = 0; | |
1311 | } else { | |
1312 | zmd->sb_gen = sb_gen[1]; | |
1313 | zmd->mblk_primary = 1; | |
1314 | } | |
1315 | ||
bf28a3ba HR |
1316 | dmz_dev_debug(zmd->sb[zmd->mblk_primary].dev, |
1317 | "Using super block %u (gen %llu)", | |
3b1a94c8 DLM |
1318 | zmd->mblk_primary, zmd->sb_gen); |
1319 | ||
bd5c4031 HR |
1320 | if ((zmd->sb_version > 1) && zmd->sb[2].zone) { |
1321 | zmd->sb[2].block = dmz_start_block(zmd, zmd->sb[2].zone); | |
1322 | zmd->sb[2].dev = dmz_zone_to_dev(zmd, zmd->sb[2].zone); | |
1323 | ret = dmz_get_sb(zmd, 2); | |
1324 | if (ret) { | |
1325 | dmz_dev_err(zmd->sb[2].dev, | |
1326 | "Read tertiary super block failed"); | |
1327 | return ret; | |
1328 | } | |
1329 | ret = dmz_check_sb(zmd, 2); | |
1330 | if (ret == -EINVAL) | |
1331 | return ret; | |
1332 | } | |
3b1a94c8 DLM |
1333 | return 0; |
1334 | } | |
1335 | ||
1336 | /* | |
1337 | * Initialize a zone descriptor. | |
1338 | */ | |
bd5c4031 | 1339 | static int dmz_init_zone(struct blk_zone *blkz, unsigned int num, void *data) |
3b1a94c8 | 1340 | { |
d4100351 | 1341 | struct dmz_metadata *zmd = data; |
bd5c4031 HR |
1342 | struct dmz_dev *dev = zmd->nr_devs > 1 ? &zmd->dev[1] : &zmd->dev[0]; |
1343 | int idx = num + dev->zone_offset; | |
d4100351 | 1344 | struct dm_zone *zone = &zmd->zones[idx]; |
3b1a94c8 | 1345 | |
36820560 | 1346 | if (blkz->len != zmd->zone_nr_sectors) { |
bd5c4031 HR |
1347 | if (zmd->sb_version > 1) { |
1348 | /* Ignore the eventual runt (smaller) zone */ | |
1349 | set_bit(DMZ_OFFLINE, &zone->flags); | |
1350 | return 0; | |
1351 | } else if (blkz->start + blkz->len == dev->capacity) | |
3b1a94c8 DLM |
1352 | return 0; |
1353 | return -ENXIO; | |
1354 | } | |
1355 | ||
1356 | INIT_LIST_HEAD(&zone->link); | |
1357 | atomic_set(&zone->refcount, 0); | |
b7122873 | 1358 | zone->id = idx; |
3b1a94c8 DLM |
1359 | zone->chunk = DMZ_MAP_UNMAPPED; |
1360 | ||
d4100351 CH |
1361 | switch (blkz->type) { |
1362 | case BLK_ZONE_TYPE_CONVENTIONAL: | |
3b1a94c8 | 1363 | set_bit(DMZ_RND, &zone->flags); |
d4100351 CH |
1364 | break; |
1365 | case BLK_ZONE_TYPE_SEQWRITE_REQ: | |
1366 | case BLK_ZONE_TYPE_SEQWRITE_PREF: | |
3b1a94c8 | 1367 | set_bit(DMZ_SEQ, &zone->flags); |
d4100351 CH |
1368 | break; |
1369 | default: | |
3b1a94c8 | 1370 | return -ENXIO; |
d4100351 | 1371 | } |
3b1a94c8 DLM |
1372 | |
1373 | if (dmz_is_rnd(zone)) | |
1374 | zone->wp_block = 0; | |
1375 | else | |
1376 | zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start); | |
1377 | ||
d4100351 CH |
1378 | if (blkz->cond == BLK_ZONE_COND_OFFLINE) |
1379 | set_bit(DMZ_OFFLINE, &zone->flags); | |
1380 | else if (blkz->cond == BLK_ZONE_COND_READONLY) | |
1381 | set_bit(DMZ_READ_ONLY, &zone->flags); | |
1382 | else { | |
3b1a94c8 DLM |
1383 | zmd->nr_useable_zones++; |
1384 | if (dmz_is_rnd(zone)) { | |
1385 | zmd->nr_rnd_zones++; | |
bd5c4031 HR |
1386 | if (zmd->nr_devs == 1 && !zmd->sb[0].zone) { |
1387 | /* Primary super block zone */ | |
735bd7e4 | 1388 | zmd->sb[0].zone = zone; |
3b1a94c8 DLM |
1389 | } |
1390 | } | |
bd5c4031 HR |
1391 | if (zmd->nr_devs > 1 && !zmd->sb[2].zone) { |
1392 | /* Tertiary superblock zone */ | |
1393 | zmd->sb[2].zone = zone; | |
1394 | } | |
3b1a94c8 DLM |
1395 | } |
1396 | ||
1397 | return 0; | |
1398 | } | |
1399 | ||
bd5c4031 HR |
1400 | static void dmz_emulate_zones(struct dmz_metadata *zmd, struct dmz_dev *dev) |
1401 | { | |
1402 | int idx; | |
1403 | sector_t zone_offset = 0; | |
1404 | ||
1405 | for(idx = 0; idx < dev->nr_zones; idx++) { | |
1406 | struct dm_zone *zone = &zmd->zones[idx]; | |
1407 | ||
1408 | INIT_LIST_HEAD(&zone->link); | |
1409 | atomic_set(&zone->refcount, 0); | |
1410 | zone->id = idx; | |
1411 | zone->chunk = DMZ_MAP_UNMAPPED; | |
34f5affd | 1412 | set_bit(DMZ_CACHE, &zone->flags); |
bd5c4031 | 1413 | zone->wp_block = 0; |
34f5affd | 1414 | zmd->nr_cache_zones++; |
bd5c4031 HR |
1415 | zmd->nr_useable_zones++; |
1416 | if (dev->capacity - zone_offset < zmd->zone_nr_sectors) { | |
1417 | /* Disable runt zone */ | |
1418 | set_bit(DMZ_OFFLINE, &zone->flags); | |
1419 | break; | |
1420 | } | |
1421 | zone_offset += zmd->zone_nr_sectors; | |
1422 | } | |
1423 | } | |
1424 | ||
3b1a94c8 DLM |
1425 | /* |
1426 | * Free zones descriptors. | |
1427 | */ | |
1428 | static void dmz_drop_zones(struct dmz_metadata *zmd) | |
1429 | { | |
1430 | kfree(zmd->zones); | |
1431 | zmd->zones = NULL; | |
1432 | } | |
1433 | ||
3b1a94c8 DLM |
1434 | /* |
1435 | * Allocate and initialize zone descriptors using the zone | |
1436 | * information from disk. | |
1437 | */ | |
1438 | static int dmz_init_zones(struct dmz_metadata *zmd) | |
1439 | { | |
bd5c4031 HR |
1440 | int i, ret; |
1441 | struct dmz_dev *zoned_dev = &zmd->dev[0]; | |
3b1a94c8 DLM |
1442 | |
1443 | /* Init */ | |
bd5c4031 | 1444 | zmd->zone_nr_sectors = zmd->dev[0].zone_nr_sectors; |
36820560 HR |
1445 | zmd->zone_nr_sectors_shift = ilog2(zmd->zone_nr_sectors); |
1446 | zmd->zone_nr_blocks = dmz_sect2blk(zmd->zone_nr_sectors); | |
1447 | zmd->zone_nr_blocks_shift = ilog2(zmd->zone_nr_blocks); | |
1448 | zmd->zone_bitmap_size = zmd->zone_nr_blocks >> 3; | |
b3996295 DF |
1449 | zmd->zone_nr_bitmap_blocks = |
1450 | max_t(sector_t, 1, zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT); | |
36820560 | 1451 | zmd->zone_bits_per_mblk = min_t(sector_t, zmd->zone_nr_blocks, |
b3996295 | 1452 | DMZ_BLOCK_SIZE_BITS); |
3b1a94c8 DLM |
1453 | |
1454 | /* Allocate zone array */ | |
bd5c4031 HR |
1455 | zmd->nr_zones = 0; |
1456 | for (i = 0; i < zmd->nr_devs; i++) | |
1457 | zmd->nr_zones += zmd->dev[i].nr_zones; | |
1458 | ||
1459 | if (!zmd->nr_zones) { | |
1460 | DMERR("(%s): No zones found", zmd->devname); | |
1461 | return -ENXIO; | |
1462 | } | |
36820560 | 1463 | zmd->zones = kcalloc(zmd->nr_zones, sizeof(struct dm_zone), GFP_KERNEL); |
3b1a94c8 DLM |
1464 | if (!zmd->zones) |
1465 | return -ENOMEM; | |
1466 | ||
ae3c1f11 HR |
1467 | DMDEBUG("(%s): Using %zu B for zone information", |
1468 | zmd->devname, sizeof(struct dm_zone) * zmd->nr_zones); | |
3b1a94c8 | 1469 | |
bd5c4031 HR |
1470 | if (zmd->nr_devs > 1) { |
1471 | dmz_emulate_zones(zmd, &zmd->dev[0]); | |
1472 | /* | |
1473 | * Primary superblock zone is always at zone 0 when multiple | |
1474 | * drives are present. | |
1475 | */ | |
1476 | zmd->sb[0].zone = &zmd->zones[0]; | |
1477 | ||
1478 | zoned_dev = &zmd->dev[1]; | |
1479 | } | |
1480 | ||
3b1a94c8 | 1481 | /* |
d4100351 CH |
1482 | * Get zone information and initialize zone descriptors. At the same |
1483 | * time, determine where the super block should be: first block of the | |
1484 | * first randomly writable zone. | |
3b1a94c8 | 1485 | */ |
bd5c4031 HR |
1486 | ret = blkdev_report_zones(zoned_dev->bdev, 0, BLK_ALL_ZONES, |
1487 | dmz_init_zone, zmd); | |
d4100351 | 1488 | if (ret < 0) { |
bd5c4031 HR |
1489 | DMDEBUG("(%s): Failed to report zones, error %d", |
1490 | zmd->devname, ret); | |
d4100351 CH |
1491 | dmz_drop_zones(zmd); |
1492 | return ret; | |
1493 | } | |
3b1a94c8 | 1494 | |
d4100351 CH |
1495 | return 0; |
1496 | } | |
7aedf75f | 1497 | |
d4100351 CH |
1498 | static int dmz_update_zone_cb(struct blk_zone *blkz, unsigned int idx, |
1499 | void *data) | |
1500 | { | |
1501 | struct dm_zone *zone = data; | |
3b1a94c8 | 1502 | |
d4100351 CH |
1503 | clear_bit(DMZ_OFFLINE, &zone->flags); |
1504 | clear_bit(DMZ_READ_ONLY, &zone->flags); | |
1505 | if (blkz->cond == BLK_ZONE_COND_OFFLINE) | |
1506 | set_bit(DMZ_OFFLINE, &zone->flags); | |
1507 | else if (blkz->cond == BLK_ZONE_COND_READONLY) | |
1508 | set_bit(DMZ_READ_ONLY, &zone->flags); | |
3b1a94c8 | 1509 | |
d4100351 CH |
1510 | if (dmz_is_seq(zone)) |
1511 | zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start); | |
1512 | else | |
1513 | zone->wp_block = 0; | |
1514 | return 0; | |
3b1a94c8 DLM |
1515 | } |
1516 | ||
1517 | /* | |
1518 | * Update a zone information. | |
1519 | */ | |
1520 | static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone) | |
1521 | { | |
aa821c8d | 1522 | struct dmz_dev *dev = dmz_zone_to_dev(zmd, zone); |
bd976e52 | 1523 | unsigned int noio_flag; |
3b1a94c8 DLM |
1524 | int ret; |
1525 | ||
bd5c4031 HR |
1526 | if (dev->flags & DMZ_BDEV_REGULAR) |
1527 | return 0; | |
1528 | ||
bd976e52 DLM |
1529 | /* |
1530 | * Get zone information from disk. Since blkdev_report_zones() uses | |
1531 | * GFP_KERNEL by default for memory allocations, set the per-task | |
1532 | * PF_MEMALLOC_NOIO flag so that all allocations are done as if | |
1533 | * GFP_NOIO was specified. | |
1534 | */ | |
1535 | noio_flag = memalloc_noio_save(); | |
aa821c8d | 1536 | ret = blkdev_report_zones(dev->bdev, dmz_start_sect(zmd, zone), 1, |
d4100351 | 1537 | dmz_update_zone_cb, zone); |
bd976e52 | 1538 | memalloc_noio_restore(noio_flag); |
d4100351 CH |
1539 | |
1540 | if (ret == 0) | |
7aedf75f | 1541 | ret = -EIO; |
d4100351 | 1542 | if (ret < 0) { |
aa821c8d | 1543 | dmz_dev_err(dev, "Get zone %u report failed", |
b7122873 | 1544 | zone->id); |
aa821c8d | 1545 | dmz_check_bdev(dev); |
3b1a94c8 DLM |
1546 | return ret; |
1547 | } | |
1548 | ||
3b1a94c8 DLM |
1549 | return 0; |
1550 | } | |
1551 | ||
1552 | /* | |
1553 | * Check a zone write pointer position when the zone is marked | |
1554 | * with the sequential write error flag. | |
1555 | */ | |
1556 | static int dmz_handle_seq_write_err(struct dmz_metadata *zmd, | |
1557 | struct dm_zone *zone) | |
1558 | { | |
aa821c8d | 1559 | struct dmz_dev *dev = dmz_zone_to_dev(zmd, zone); |
3b1a94c8 DLM |
1560 | unsigned int wp = 0; |
1561 | int ret; | |
1562 | ||
1563 | wp = zone->wp_block; | |
1564 | ret = dmz_update_zone(zmd, zone); | |
1565 | if (ret) | |
1566 | return ret; | |
1567 | ||
aa821c8d | 1568 | dmz_dev_warn(dev, "Processing zone %u write error (zone wp %u/%u)", |
b7122873 | 1569 | zone->id, zone->wp_block, wp); |
3b1a94c8 DLM |
1570 | |
1571 | if (zone->wp_block < wp) { | |
1572 | dmz_invalidate_blocks(zmd, zone, zone->wp_block, | |
1573 | wp - zone->wp_block); | |
1574 | } | |
1575 | ||
1576 | return 0; | |
1577 | } | |
1578 | ||
1579 | static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id) | |
1580 | { | |
1581 | return &zmd->zones[zone_id]; | |
1582 | } | |
1583 | ||
1584 | /* | |
1585 | * Reset a zone write pointer. | |
1586 | */ | |
1587 | static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone) | |
1588 | { | |
1589 | int ret; | |
1590 | ||
1591 | /* | |
1592 | * Ignore offline zones, read only zones, | |
1593 | * and conventional zones. | |
1594 | */ | |
1595 | if (dmz_is_offline(zone) || | |
1596 | dmz_is_readonly(zone) || | |
1597 | dmz_is_rnd(zone)) | |
1598 | return 0; | |
1599 | ||
1600 | if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) { | |
aa821c8d | 1601 | struct dmz_dev *dev = dmz_zone_to_dev(zmd, zone); |
3b1a94c8 | 1602 | |
6c1b1da5 AJ |
1603 | ret = blkdev_zone_mgmt(dev->bdev, REQ_OP_ZONE_RESET, |
1604 | dmz_start_sect(zmd, zone), | |
36820560 | 1605 | zmd->zone_nr_sectors, GFP_NOIO); |
3b1a94c8 DLM |
1606 | if (ret) { |
1607 | dmz_dev_err(dev, "Reset zone %u failed %d", | |
b7122873 | 1608 | zone->id, ret); |
3b1a94c8 DLM |
1609 | return ret; |
1610 | } | |
1611 | } | |
1612 | ||
1613 | /* Clear write error bit and rewind write pointer position */ | |
1614 | clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags); | |
1615 | zone->wp_block = 0; | |
1616 | ||
1617 | return 0; | |
1618 | } | |
1619 | ||
1620 | static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone); | |
1621 | ||
1622 | /* | |
1623 | * Initialize chunk mapping. | |
1624 | */ | |
1625 | static int dmz_load_mapping(struct dmz_metadata *zmd) | |
1626 | { | |
3b1a94c8 DLM |
1627 | struct dm_zone *dzone, *bzone; |
1628 | struct dmz_mblock *dmap_mblk = NULL; | |
1629 | struct dmz_map *dmap; | |
1630 | unsigned int i = 0, e = 0, chunk = 0; | |
1631 | unsigned int dzone_id; | |
1632 | unsigned int bzone_id; | |
1633 | ||
1634 | /* Metadata block array for the chunk mapping table */ | |
1635 | zmd->map_mblk = kcalloc(zmd->nr_map_blocks, | |
1636 | sizeof(struct dmz_mblk *), GFP_KERNEL); | |
1637 | if (!zmd->map_mblk) | |
1638 | return -ENOMEM; | |
1639 | ||
1640 | /* Get chunk mapping table blocks and initialize zone mapping */ | |
1641 | while (chunk < zmd->nr_chunks) { | |
1642 | if (!dmap_mblk) { | |
1643 | /* Get mapping block */ | |
1644 | dmap_mblk = dmz_get_mblock(zmd, i + 1); | |
1645 | if (IS_ERR(dmap_mblk)) | |
1646 | return PTR_ERR(dmap_mblk); | |
1647 | zmd->map_mblk[i] = dmap_mblk; | |
1648 | dmap = (struct dmz_map *) dmap_mblk->data; | |
1649 | i++; | |
1650 | e = 0; | |
1651 | } | |
1652 | ||
1653 | /* Check data zone */ | |
1654 | dzone_id = le32_to_cpu(dmap[e].dzone_id); | |
1655 | if (dzone_id == DMZ_MAP_UNMAPPED) | |
1656 | goto next; | |
1657 | ||
36820560 | 1658 | if (dzone_id >= zmd->nr_zones) { |
ca1a7045 | 1659 | dmz_zmd_err(zmd, "Chunk %u mapping: invalid data zone ID %u", |
3b1a94c8 DLM |
1660 | chunk, dzone_id); |
1661 | return -EIO; | |
1662 | } | |
1663 | ||
1664 | dzone = dmz_get(zmd, dzone_id); | |
1665 | set_bit(DMZ_DATA, &dzone->flags); | |
1666 | dzone->chunk = chunk; | |
1667 | dmz_get_zone_weight(zmd, dzone); | |
1668 | ||
34f5affd HR |
1669 | if (dmz_is_cache(dzone)) |
1670 | list_add_tail(&dzone->link, &zmd->map_cache_list); | |
1671 | else if (dmz_is_rnd(dzone)) | |
3b1a94c8 DLM |
1672 | list_add_tail(&dzone->link, &zmd->map_rnd_list); |
1673 | else | |
1674 | list_add_tail(&dzone->link, &zmd->map_seq_list); | |
1675 | ||
1676 | /* Check buffer zone */ | |
1677 | bzone_id = le32_to_cpu(dmap[e].bzone_id); | |
1678 | if (bzone_id == DMZ_MAP_UNMAPPED) | |
1679 | goto next; | |
1680 | ||
36820560 | 1681 | if (bzone_id >= zmd->nr_zones) { |
ca1a7045 | 1682 | dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone ID %u", |
3b1a94c8 DLM |
1683 | chunk, bzone_id); |
1684 | return -EIO; | |
1685 | } | |
1686 | ||
1687 | bzone = dmz_get(zmd, bzone_id); | |
34f5affd | 1688 | if (!dmz_is_rnd(bzone) && !dmz_is_cache(bzone)) { |
ca1a7045 | 1689 | dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone %u", |
3b1a94c8 DLM |
1690 | chunk, bzone_id); |
1691 | return -EIO; | |
1692 | } | |
1693 | ||
1694 | set_bit(DMZ_DATA, &bzone->flags); | |
1695 | set_bit(DMZ_BUF, &bzone->flags); | |
1696 | bzone->chunk = chunk; | |
1697 | bzone->bzone = dzone; | |
1698 | dzone->bzone = bzone; | |
1699 | dmz_get_zone_weight(zmd, bzone); | |
34f5affd HR |
1700 | if (dmz_is_cache(bzone)) |
1701 | list_add_tail(&bzone->link, &zmd->map_cache_list); | |
1702 | else | |
1703 | list_add_tail(&bzone->link, &zmd->map_rnd_list); | |
3b1a94c8 DLM |
1704 | next: |
1705 | chunk++; | |
1706 | e++; | |
1707 | if (e >= DMZ_MAP_ENTRIES) | |
1708 | dmap_mblk = NULL; | |
1709 | } | |
1710 | ||
1711 | /* | |
1712 | * At this point, only meta zones and mapped data zones were | |
1713 | * fully initialized. All remaining zones are unmapped data | |
1714 | * zones. Finish initializing those here. | |
1715 | */ | |
36820560 | 1716 | for (i = 0; i < zmd->nr_zones; i++) { |
3b1a94c8 DLM |
1717 | dzone = dmz_get(zmd, i); |
1718 | if (dmz_is_meta(dzone)) | |
1719 | continue; | |
34f5affd HR |
1720 | if (dmz_is_offline(dzone)) |
1721 | continue; | |
3b1a94c8 | 1722 | |
34f5affd HR |
1723 | if (dmz_is_cache(dzone)) |
1724 | zmd->nr_cache++; | |
1725 | else if (dmz_is_rnd(dzone)) | |
3b1a94c8 DLM |
1726 | zmd->nr_rnd++; |
1727 | else | |
1728 | zmd->nr_seq++; | |
1729 | ||
1730 | if (dmz_is_data(dzone)) { | |
1731 | /* Already initialized */ | |
1732 | continue; | |
1733 | } | |
1734 | ||
1735 | /* Unmapped data zone */ | |
1736 | set_bit(DMZ_DATA, &dzone->flags); | |
1737 | dzone->chunk = DMZ_MAP_UNMAPPED; | |
34f5affd HR |
1738 | if (dmz_is_cache(dzone)) { |
1739 | list_add_tail(&dzone->link, &zmd->unmap_cache_list); | |
1740 | atomic_inc(&zmd->unmap_nr_cache); | |
1741 | } else if (dmz_is_rnd(dzone)) { | |
3b1a94c8 DLM |
1742 | list_add_tail(&dzone->link, &zmd->unmap_rnd_list); |
1743 | atomic_inc(&zmd->unmap_nr_rnd); | |
1744 | } else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) { | |
1745 | list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list); | |
1746 | atomic_inc(&zmd->nr_reserved_seq_zones); | |
1747 | zmd->nr_seq--; | |
1748 | } else { | |
1749 | list_add_tail(&dzone->link, &zmd->unmap_seq_list); | |
1750 | atomic_inc(&zmd->unmap_nr_seq); | |
1751 | } | |
1752 | } | |
1753 | ||
1754 | return 0; | |
1755 | } | |
1756 | ||
1757 | /* | |
1758 | * Set a data chunk mapping. | |
1759 | */ | |
1760 | static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, | |
1761 | unsigned int dzone_id, unsigned int bzone_id) | |
1762 | { | |
1763 | struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT]; | |
1764 | struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data; | |
1765 | int map_idx = chunk & DMZ_MAP_ENTRIES_MASK; | |
1766 | ||
1767 | dmap[map_idx].dzone_id = cpu_to_le32(dzone_id); | |
1768 | dmap[map_idx].bzone_id = cpu_to_le32(bzone_id); | |
1769 | dmz_dirty_mblock(zmd, dmap_mblk); | |
1770 | } | |
1771 | ||
1772 | /* | |
1773 | * The list of mapped zones is maintained in LRU order. | |
1774 | * This rotates a zone at the end of its map list. | |
1775 | */ | |
1776 | static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone) | |
1777 | { | |
1778 | if (list_empty(&zone->link)) | |
1779 | return; | |
1780 | ||
1781 | list_del_init(&zone->link); | |
1782 | if (dmz_is_seq(zone)) { | |
1783 | /* LRU rotate sequential zone */ | |
1784 | list_add_tail(&zone->link, &zmd->map_seq_list); | |
34f5affd HR |
1785 | } else if (dmz_is_cache(zone)) { |
1786 | /* LRU rotate cache zone */ | |
1787 | list_add_tail(&zone->link, &zmd->map_cache_list); | |
3b1a94c8 DLM |
1788 | } else { |
1789 | /* LRU rotate random zone */ | |
1790 | list_add_tail(&zone->link, &zmd->map_rnd_list); | |
1791 | } | |
1792 | } | |
1793 | ||
1794 | /* | |
1795 | * The list of mapped random zones is maintained | |
1796 | * in LRU order. This rotates a zone at the end of the list. | |
1797 | */ | |
1798 | static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone) | |
1799 | { | |
1800 | __dmz_lru_zone(zmd, zone); | |
1801 | if (zone->bzone) | |
1802 | __dmz_lru_zone(zmd, zone->bzone); | |
1803 | } | |
1804 | ||
1805 | /* | |
1806 | * Wait for any zone to be freed. | |
1807 | */ | |
1808 | static void dmz_wait_for_free_zones(struct dmz_metadata *zmd) | |
1809 | { | |
1810 | DEFINE_WAIT(wait); | |
1811 | ||
1812 | prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE); | |
1813 | dmz_unlock_map(zmd); | |
1814 | dmz_unlock_metadata(zmd); | |
1815 | ||
1816 | io_schedule_timeout(HZ); | |
1817 | ||
1818 | dmz_lock_metadata(zmd); | |
1819 | dmz_lock_map(zmd); | |
1820 | finish_wait(&zmd->free_wq, &wait); | |
1821 | } | |
1822 | ||
1823 | /* | |
1824 | * Lock a zone for reclaim (set the zone RECLAIM bit). | |
1825 | * Returns false if the zone cannot be locked or if it is already locked | |
1826 | * and 1 otherwise. | |
1827 | */ | |
1828 | int dmz_lock_zone_reclaim(struct dm_zone *zone) | |
1829 | { | |
1830 | /* Active zones cannot be reclaimed */ | |
1831 | if (dmz_is_active(zone)) | |
1832 | return 0; | |
1833 | ||
1834 | return !test_and_set_bit(DMZ_RECLAIM, &zone->flags); | |
1835 | } | |
1836 | ||
1837 | /* | |
1838 | * Clear a zone reclaim flag. | |
1839 | */ | |
1840 | void dmz_unlock_zone_reclaim(struct dm_zone *zone) | |
1841 | { | |
1842 | WARN_ON(dmz_is_active(zone)); | |
1843 | WARN_ON(!dmz_in_reclaim(zone)); | |
1844 | ||
1845 | clear_bit_unlock(DMZ_RECLAIM, &zone->flags); | |
1846 | smp_mb__after_atomic(); | |
1847 | wake_up_bit(&zone->flags, DMZ_RECLAIM); | |
1848 | } | |
1849 | ||
1850 | /* | |
1851 | * Wait for a zone reclaim to complete. | |
1852 | */ | |
1853 | static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone) | |
1854 | { | |
1855 | dmz_unlock_map(zmd); | |
1856 | dmz_unlock_metadata(zmd); | |
a16b7dee | 1857 | set_bit(DMZ_RECLAIM_TERMINATE, &zone->flags); |
3b1a94c8 | 1858 | wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ); |
a16b7dee | 1859 | clear_bit(DMZ_RECLAIM_TERMINATE, &zone->flags); |
3b1a94c8 DLM |
1860 | dmz_lock_metadata(zmd); |
1861 | dmz_lock_map(zmd); | |
1862 | } | |
1863 | ||
1864 | /* | |
34f5affd | 1865 | * Select a cache or random write zone for reclaim. |
3b1a94c8 | 1866 | */ |
90a9b869 HR |
1867 | static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd, |
1868 | bool idle) | |
3b1a94c8 DLM |
1869 | { |
1870 | struct dm_zone *dzone = NULL; | |
1871 | struct dm_zone *zone; | |
34f5affd | 1872 | struct list_head *zone_list = &zmd->map_rnd_list; |
3b1a94c8 | 1873 | |
34f5affd | 1874 | /* If we have cache zones select from the cache zone list */ |
90a9b869 | 1875 | if (zmd->nr_cache) { |
34f5affd | 1876 | zone_list = &zmd->map_cache_list; |
90a9b869 HR |
1877 | /* Try to relaim random zones, too, when idle */ |
1878 | if (idle && list_empty(zone_list)) | |
1879 | zone_list = &zmd->map_rnd_list; | |
1880 | } | |
3b1a94c8 | 1881 | |
34f5affd | 1882 | list_for_each_entry(zone, zone_list, link) { |
3b1a94c8 DLM |
1883 | if (dmz_is_buf(zone)) |
1884 | dzone = zone->bzone; | |
1885 | else | |
1886 | dzone = zone; | |
1887 | if (dmz_lock_zone_reclaim(dzone)) | |
1888 | return dzone; | |
1889 | } | |
1890 | ||
489dc0f0 | 1891 | return NULL; |
3b1a94c8 DLM |
1892 | } |
1893 | ||
1894 | /* | |
1895 | * Select a buffered sequential zone for reclaim. | |
1896 | */ | |
1897 | static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd) | |
1898 | { | |
1899 | struct dm_zone *zone; | |
1900 | ||
3b1a94c8 DLM |
1901 | list_for_each_entry(zone, &zmd->map_seq_list, link) { |
1902 | if (!zone->bzone) | |
1903 | continue; | |
1904 | if (dmz_lock_zone_reclaim(zone)) | |
1905 | return zone; | |
1906 | } | |
1907 | ||
489dc0f0 | 1908 | return NULL; |
3b1a94c8 DLM |
1909 | } |
1910 | ||
1911 | /* | |
1912 | * Select a zone for reclaim. | |
1913 | */ | |
90a9b869 | 1914 | struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd, bool idle) |
3b1a94c8 DLM |
1915 | { |
1916 | struct dm_zone *zone; | |
1917 | ||
1918 | /* | |
1919 | * Search for a zone candidate to reclaim: 2 cases are possible. | |
1920 | * (1) There is no free sequential zones. Then a random data zone | |
1921 | * cannot be reclaimed. So choose a sequential zone to reclaim so | |
1922 | * that afterward a random zone can be reclaimed. | |
1923 | * (2) At least one free sequential zone is available, then choose | |
1924 | * the oldest random zone (data or buffer) that can be locked. | |
1925 | */ | |
1926 | dmz_lock_map(zmd); | |
1927 | if (list_empty(&zmd->reserved_seq_zones_list)) | |
1928 | zone = dmz_get_seq_zone_for_reclaim(zmd); | |
1929 | else | |
90a9b869 | 1930 | zone = dmz_get_rnd_zone_for_reclaim(zmd, idle); |
3b1a94c8 DLM |
1931 | dmz_unlock_map(zmd); |
1932 | ||
1933 | return zone; | |
1934 | } | |
1935 | ||
3b1a94c8 DLM |
1936 | /* |
1937 | * Get the zone mapping a chunk, if the chunk is mapped already. | |
1938 | * If no mapping exist and the operation is WRITE, a zone is | |
1939 | * allocated and used to map the chunk. | |
1940 | * The zone returned will be set to the active state. | |
1941 | */ | |
1942 | struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op) | |
1943 | { | |
1944 | struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT]; | |
1945 | struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data; | |
1946 | int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK; | |
1947 | unsigned int dzone_id; | |
1948 | struct dm_zone *dzone = NULL; | |
1949 | int ret = 0; | |
34f5affd | 1950 | int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND; |
3b1a94c8 DLM |
1951 | |
1952 | dmz_lock_map(zmd); | |
1953 | again: | |
1954 | /* Get the chunk mapping */ | |
1955 | dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id); | |
1956 | if (dzone_id == DMZ_MAP_UNMAPPED) { | |
1957 | /* | |
1958 | * Read or discard in unmapped chunks are fine. But for | |
1959 | * writes, we need a mapping, so get one. | |
1960 | */ | |
1961 | if (op != REQ_OP_WRITE) | |
1962 | goto out; | |
1963 | ||
ad1bd578 | 1964 | /* Allocate a random zone */ |
34f5affd | 1965 | dzone = dmz_alloc_zone(zmd, alloc_flags); |
3b1a94c8 | 1966 | if (!dzone) { |
d0e21ce4 | 1967 | if (dmz_dev_is_dying(zmd)) { |
75d66ffb DF |
1968 | dzone = ERR_PTR(-EIO); |
1969 | goto out; | |
1970 | } | |
3b1a94c8 DLM |
1971 | dmz_wait_for_free_zones(zmd); |
1972 | goto again; | |
1973 | } | |
1974 | ||
1975 | dmz_map_zone(zmd, dzone, chunk); | |
1976 | ||
1977 | } else { | |
1978 | /* The chunk is already mapped: get the mapping zone */ | |
1979 | dzone = dmz_get(zmd, dzone_id); | |
1980 | if (dzone->chunk != chunk) { | |
1981 | dzone = ERR_PTR(-EIO); | |
1982 | goto out; | |
1983 | } | |
1984 | ||
1985 | /* Repair write pointer if the sequential dzone has error */ | |
1986 | if (dmz_seq_write_err(dzone)) { | |
1987 | ret = dmz_handle_seq_write_err(zmd, dzone); | |
1988 | if (ret) { | |
1989 | dzone = ERR_PTR(-EIO); | |
1990 | goto out; | |
1991 | } | |
1992 | clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags); | |
1993 | } | |
1994 | } | |
1995 | ||
1996 | /* | |
1997 | * If the zone is being reclaimed, the chunk mapping may change | |
1998 | * to a different zone. So wait for reclaim and retry. Otherwise, | |
1999 | * activate the zone (this will prevent reclaim from touching it). | |
2000 | */ | |
2001 | if (dmz_in_reclaim(dzone)) { | |
2002 | dmz_wait_for_reclaim(zmd, dzone); | |
2003 | goto again; | |
2004 | } | |
2005 | dmz_activate_zone(dzone); | |
2006 | dmz_lru_zone(zmd, dzone); | |
2007 | out: | |
2008 | dmz_unlock_map(zmd); | |
2009 | ||
2010 | return dzone; | |
2011 | } | |
2012 | ||
2013 | /* | |
2014 | * Write and discard change the block validity of data zones and their buffer | |
2015 | * zones. Check here that valid blocks are still present. If all blocks are | |
2016 | * invalid, the zones can be unmapped on the fly without waiting for reclaim | |
2017 | * to do it. | |
2018 | */ | |
2019 | void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone) | |
2020 | { | |
2021 | struct dm_zone *bzone; | |
2022 | ||
2023 | dmz_lock_map(zmd); | |
2024 | ||
2025 | bzone = dzone->bzone; | |
2026 | if (bzone) { | |
2027 | if (dmz_weight(bzone)) | |
2028 | dmz_lru_zone(zmd, bzone); | |
2029 | else { | |
2030 | /* Empty buffer zone: reclaim it */ | |
2031 | dmz_unmap_zone(zmd, bzone); | |
2032 | dmz_free_zone(zmd, bzone); | |
2033 | bzone = NULL; | |
2034 | } | |
2035 | } | |
2036 | ||
2037 | /* Deactivate the data zone */ | |
2038 | dmz_deactivate_zone(dzone); | |
2039 | if (dmz_is_active(dzone) || bzone || dmz_weight(dzone)) | |
2040 | dmz_lru_zone(zmd, dzone); | |
2041 | else { | |
2042 | /* Unbuffered inactive empty data zone: reclaim it */ | |
2043 | dmz_unmap_zone(zmd, dzone); | |
2044 | dmz_free_zone(zmd, dzone); | |
2045 | } | |
2046 | ||
2047 | dmz_unlock_map(zmd); | |
2048 | } | |
2049 | ||
2050 | /* | |
2051 | * Allocate and map a random zone to buffer a chunk | |
2052 | * already mapped to a sequential zone. | |
2053 | */ | |
2054 | struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd, | |
2055 | struct dm_zone *dzone) | |
2056 | { | |
2057 | struct dm_zone *bzone; | |
34f5affd | 2058 | int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND; |
3b1a94c8 DLM |
2059 | |
2060 | dmz_lock_map(zmd); | |
2061 | again: | |
2062 | bzone = dzone->bzone; | |
2063 | if (bzone) | |
2064 | goto out; | |
2065 | ||
ad1bd578 | 2066 | /* Allocate a random zone */ |
34f5affd | 2067 | bzone = dmz_alloc_zone(zmd, alloc_flags); |
3b1a94c8 | 2068 | if (!bzone) { |
d0e21ce4 | 2069 | if (dmz_dev_is_dying(zmd)) { |
75d66ffb DF |
2070 | bzone = ERR_PTR(-EIO); |
2071 | goto out; | |
2072 | } | |
3b1a94c8 DLM |
2073 | dmz_wait_for_free_zones(zmd); |
2074 | goto again; | |
2075 | } | |
2076 | ||
2077 | /* Update the chunk mapping */ | |
b7122873 | 2078 | dmz_set_chunk_mapping(zmd, dzone->chunk, dzone->id, bzone->id); |
3b1a94c8 DLM |
2079 | |
2080 | set_bit(DMZ_BUF, &bzone->flags); | |
2081 | bzone->chunk = dzone->chunk; | |
2082 | bzone->bzone = dzone; | |
2083 | dzone->bzone = bzone; | |
34f5affd HR |
2084 | if (dmz_is_cache(bzone)) |
2085 | list_add_tail(&bzone->link, &zmd->map_cache_list); | |
2086 | else | |
2087 | list_add_tail(&bzone->link, &zmd->map_rnd_list); | |
3b1a94c8 DLM |
2088 | out: |
2089 | dmz_unlock_map(zmd); | |
2090 | ||
2091 | return bzone; | |
2092 | } | |
2093 | ||
2094 | /* | |
2095 | * Get an unmapped (free) zone. | |
2096 | * This must be called with the mapping lock held. | |
2097 | */ | |
2098 | struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags) | |
2099 | { | |
2100 | struct list_head *list; | |
2101 | struct dm_zone *zone; | |
2102 | ||
34f5affd HR |
2103 | if (flags & DMZ_ALLOC_CACHE) |
2104 | list = &zmd->unmap_cache_list; | |
2105 | else if (flags & DMZ_ALLOC_RND) | |
3b1a94c8 DLM |
2106 | list = &zmd->unmap_rnd_list; |
2107 | else | |
2108 | list = &zmd->unmap_seq_list; | |
34f5affd | 2109 | |
3b1a94c8 DLM |
2110 | again: |
2111 | if (list_empty(list)) { | |
2112 | /* | |
34f5affd | 2113 | * No free zone: return NULL if this is for not reclaim. |
3b1a94c8 | 2114 | */ |
34f5affd | 2115 | if (!(flags & DMZ_ALLOC_RECLAIM)) |
3b1a94c8 | 2116 | return NULL; |
34f5affd HR |
2117 | /* |
2118 | * Fallback to the reserved sequential zones | |
2119 | */ | |
2120 | zone = list_first_entry_or_null(&zmd->reserved_seq_zones_list, | |
2121 | struct dm_zone, link); | |
2122 | if (zone) { | |
2123 | list_del_init(&zone->link); | |
2124 | atomic_dec(&zmd->nr_reserved_seq_zones); | |
2125 | } | |
3b1a94c8 DLM |
2126 | return zone; |
2127 | } | |
2128 | ||
2129 | zone = list_first_entry(list, struct dm_zone, link); | |
2130 | list_del_init(&zone->link); | |
2131 | ||
34f5affd HR |
2132 | if (dmz_is_cache(zone)) |
2133 | atomic_dec(&zmd->unmap_nr_cache); | |
2134 | else if (dmz_is_rnd(zone)) | |
3b1a94c8 DLM |
2135 | atomic_dec(&zmd->unmap_nr_rnd); |
2136 | else | |
2137 | atomic_dec(&zmd->unmap_nr_seq); | |
2138 | ||
2139 | if (dmz_is_offline(zone)) { | |
ca1a7045 | 2140 | dmz_zmd_warn(zmd, "Zone %u is offline", zone->id); |
3b1a94c8 DLM |
2141 | zone = NULL; |
2142 | goto again; | |
2143 | } | |
dc076c83 HR |
2144 | if (dmz_is_meta(zone)) { |
2145 | struct dmz_dev *dev = dmz_zone_to_dev(zmd, zone); | |
3b1a94c8 | 2146 | |
dc076c83 HR |
2147 | dmz_dev_warn(dev, "Zone %u has metadata", zone->id); |
2148 | zone = NULL; | |
2149 | goto again; | |
2150 | } | |
3b1a94c8 DLM |
2151 | return zone; |
2152 | } | |
2153 | ||
2154 | /* | |
2155 | * Free a zone. | |
2156 | * This must be called with the mapping lock held. | |
2157 | */ | |
2158 | void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone) | |
2159 | { | |
2160 | /* If this is a sequential zone, reset it */ | |
2161 | if (dmz_is_seq(zone)) | |
2162 | dmz_reset_zone(zmd, zone); | |
2163 | ||
2164 | /* Return the zone to its type unmap list */ | |
34f5affd HR |
2165 | if (dmz_is_cache(zone)) { |
2166 | list_add_tail(&zone->link, &zmd->unmap_cache_list); | |
2167 | atomic_inc(&zmd->unmap_nr_cache); | |
2168 | } else if (dmz_is_rnd(zone)) { | |
3b1a94c8 DLM |
2169 | list_add_tail(&zone->link, &zmd->unmap_rnd_list); |
2170 | atomic_inc(&zmd->unmap_nr_rnd); | |
2171 | } else if (atomic_read(&zmd->nr_reserved_seq_zones) < | |
2172 | zmd->nr_reserved_seq) { | |
2173 | list_add_tail(&zone->link, &zmd->reserved_seq_zones_list); | |
2174 | atomic_inc(&zmd->nr_reserved_seq_zones); | |
2175 | } else { | |
2176 | list_add_tail(&zone->link, &zmd->unmap_seq_list); | |
2177 | atomic_inc(&zmd->unmap_nr_seq); | |
2178 | } | |
2179 | ||
2180 | wake_up_all(&zmd->free_wq); | |
2181 | } | |
2182 | ||
2183 | /* | |
2184 | * Map a chunk to a zone. | |
2185 | * This must be called with the mapping lock held. | |
2186 | */ | |
2187 | void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone, | |
2188 | unsigned int chunk) | |
2189 | { | |
2190 | /* Set the chunk mapping */ | |
b7122873 | 2191 | dmz_set_chunk_mapping(zmd, chunk, dzone->id, |
3b1a94c8 DLM |
2192 | DMZ_MAP_UNMAPPED); |
2193 | dzone->chunk = chunk; | |
34f5affd HR |
2194 | if (dmz_is_cache(dzone)) |
2195 | list_add_tail(&dzone->link, &zmd->map_cache_list); | |
2196 | else if (dmz_is_rnd(dzone)) | |
3b1a94c8 DLM |
2197 | list_add_tail(&dzone->link, &zmd->map_rnd_list); |
2198 | else | |
2199 | list_add_tail(&dzone->link, &zmd->map_seq_list); | |
2200 | } | |
2201 | ||
2202 | /* | |
2203 | * Unmap a zone. | |
2204 | * This must be called with the mapping lock held. | |
2205 | */ | |
2206 | void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone) | |
2207 | { | |
2208 | unsigned int chunk = zone->chunk; | |
2209 | unsigned int dzone_id; | |
2210 | ||
2211 | if (chunk == DMZ_MAP_UNMAPPED) { | |
2212 | /* Already unmapped */ | |
2213 | return; | |
2214 | } | |
2215 | ||
2216 | if (test_and_clear_bit(DMZ_BUF, &zone->flags)) { | |
2217 | /* | |
2218 | * Unmapping the chunk buffer zone: clear only | |
2219 | * the chunk buffer mapping | |
2220 | */ | |
b7122873 | 2221 | dzone_id = zone->bzone->id; |
3b1a94c8 DLM |
2222 | zone->bzone->bzone = NULL; |
2223 | zone->bzone = NULL; | |
2224 | ||
2225 | } else { | |
2226 | /* | |
2227 | * Unmapping the chunk data zone: the zone must | |
2228 | * not be buffered. | |
2229 | */ | |
2230 | if (WARN_ON(zone->bzone)) { | |
2231 | zone->bzone->bzone = NULL; | |
2232 | zone->bzone = NULL; | |
2233 | } | |
2234 | dzone_id = DMZ_MAP_UNMAPPED; | |
2235 | } | |
2236 | ||
2237 | dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED); | |
2238 | ||
2239 | zone->chunk = DMZ_MAP_UNMAPPED; | |
2240 | list_del_init(&zone->link); | |
2241 | } | |
2242 | ||
2243 | /* | |
2244 | * Set @nr_bits bits in @bitmap starting from @bit. | |
2245 | * Return the number of bits changed from 0 to 1. | |
2246 | */ | |
2247 | static unsigned int dmz_set_bits(unsigned long *bitmap, | |
2248 | unsigned int bit, unsigned int nr_bits) | |
2249 | { | |
2250 | unsigned long *addr; | |
2251 | unsigned int end = bit + nr_bits; | |
2252 | unsigned int n = 0; | |
2253 | ||
2254 | while (bit < end) { | |
2255 | if (((bit & (BITS_PER_LONG - 1)) == 0) && | |
2256 | ((end - bit) >= BITS_PER_LONG)) { | |
2257 | /* Try to set the whole word at once */ | |
2258 | addr = bitmap + BIT_WORD(bit); | |
2259 | if (*addr == 0) { | |
2260 | *addr = ULONG_MAX; | |
2261 | n += BITS_PER_LONG; | |
2262 | bit += BITS_PER_LONG; | |
2263 | continue; | |
2264 | } | |
2265 | } | |
2266 | ||
2267 | if (!test_and_set_bit(bit, bitmap)) | |
2268 | n++; | |
2269 | bit++; | |
2270 | } | |
2271 | ||
2272 | return n; | |
2273 | } | |
2274 | ||
2275 | /* | |
2276 | * Get the bitmap block storing the bit for chunk_block in zone. | |
2277 | */ | |
2278 | static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd, | |
2279 | struct dm_zone *zone, | |
2280 | sector_t chunk_block) | |
2281 | { | |
2282 | sector_t bitmap_block = 1 + zmd->nr_map_blocks + | |
b7122873 | 2283 | (sector_t)(zone->id * zmd->zone_nr_bitmap_blocks) + |
3b1a94c8 DLM |
2284 | (chunk_block >> DMZ_BLOCK_SHIFT_BITS); |
2285 | ||
2286 | return dmz_get_mblock(zmd, bitmap_block); | |
2287 | } | |
2288 | ||
2289 | /* | |
2290 | * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone. | |
2291 | */ | |
2292 | int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone, | |
2293 | struct dm_zone *to_zone) | |
2294 | { | |
2295 | struct dmz_mblock *from_mblk, *to_mblk; | |
2296 | sector_t chunk_block = 0; | |
2297 | ||
2298 | /* Get the zones bitmap blocks */ | |
36820560 | 2299 | while (chunk_block < zmd->zone_nr_blocks) { |
3b1a94c8 DLM |
2300 | from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block); |
2301 | if (IS_ERR(from_mblk)) | |
2302 | return PTR_ERR(from_mblk); | |
2303 | to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block); | |
2304 | if (IS_ERR(to_mblk)) { | |
2305 | dmz_release_mblock(zmd, from_mblk); | |
2306 | return PTR_ERR(to_mblk); | |
2307 | } | |
2308 | ||
2309 | memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE); | |
2310 | dmz_dirty_mblock(zmd, to_mblk); | |
2311 | ||
2312 | dmz_release_mblock(zmd, to_mblk); | |
2313 | dmz_release_mblock(zmd, from_mblk); | |
2314 | ||
b3996295 | 2315 | chunk_block += zmd->zone_bits_per_mblk; |
3b1a94c8 DLM |
2316 | } |
2317 | ||
2318 | to_zone->weight = from_zone->weight; | |
2319 | ||
2320 | return 0; | |
2321 | } | |
2322 | ||
2323 | /* | |
2324 | * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone, | |
2325 | * starting from chunk_block. | |
2326 | */ | |
2327 | int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone, | |
2328 | struct dm_zone *to_zone, sector_t chunk_block) | |
2329 | { | |
2330 | unsigned int nr_blocks; | |
2331 | int ret; | |
2332 | ||
2333 | /* Get the zones bitmap blocks */ | |
36820560 | 2334 | while (chunk_block < zmd->zone_nr_blocks) { |
3b1a94c8 DLM |
2335 | /* Get a valid region from the source zone */ |
2336 | ret = dmz_first_valid_block(zmd, from_zone, &chunk_block); | |
2337 | if (ret <= 0) | |
2338 | return ret; | |
2339 | ||
2340 | nr_blocks = ret; | |
2341 | ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks); | |
2342 | if (ret) | |
2343 | return ret; | |
2344 | ||
2345 | chunk_block += nr_blocks; | |
2346 | } | |
2347 | ||
2348 | return 0; | |
2349 | } | |
2350 | ||
2351 | /* | |
2352 | * Validate all the blocks in the range [block..block+nr_blocks-1]. | |
2353 | */ | |
2354 | int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone, | |
2355 | sector_t chunk_block, unsigned int nr_blocks) | |
2356 | { | |
2357 | unsigned int count, bit, nr_bits; | |
36820560 | 2358 | unsigned int zone_nr_blocks = zmd->zone_nr_blocks; |
3b1a94c8 DLM |
2359 | struct dmz_mblock *mblk; |
2360 | unsigned int n = 0; | |
2361 | ||
ca1a7045 | 2362 | dmz_zmd_debug(zmd, "=> VALIDATE zone %u, block %llu, %u blocks", |
b7122873 | 2363 | zone->id, (unsigned long long)chunk_block, |
3b1a94c8 DLM |
2364 | nr_blocks); |
2365 | ||
2366 | WARN_ON(chunk_block + nr_blocks > zone_nr_blocks); | |
2367 | ||
2368 | while (nr_blocks) { | |
2369 | /* Get bitmap block */ | |
2370 | mblk = dmz_get_bitmap(zmd, zone, chunk_block); | |
2371 | if (IS_ERR(mblk)) | |
2372 | return PTR_ERR(mblk); | |
2373 | ||
2374 | /* Set bits */ | |
2375 | bit = chunk_block & DMZ_BLOCK_MASK_BITS; | |
b3996295 | 2376 | nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit); |
3b1a94c8 DLM |
2377 | |
2378 | count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits); | |
2379 | if (count) { | |
2380 | dmz_dirty_mblock(zmd, mblk); | |
2381 | n += count; | |
2382 | } | |
2383 | dmz_release_mblock(zmd, mblk); | |
2384 | ||
2385 | nr_blocks -= nr_bits; | |
2386 | chunk_block += nr_bits; | |
2387 | } | |
2388 | ||
2389 | if (likely(zone->weight + n <= zone_nr_blocks)) | |
2390 | zone->weight += n; | |
2391 | else { | |
ca1a7045 | 2392 | dmz_zmd_warn(zmd, "Zone %u: weight %u should be <= %u", |
b7122873 | 2393 | zone->id, zone->weight, |
3b1a94c8 DLM |
2394 | zone_nr_blocks - n); |
2395 | zone->weight = zone_nr_blocks; | |
2396 | } | |
2397 | ||
2398 | return 0; | |
2399 | } | |
2400 | ||
2401 | /* | |
2402 | * Clear nr_bits bits in bitmap starting from bit. | |
2403 | * Return the number of bits cleared. | |
2404 | */ | |
2405 | static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits) | |
2406 | { | |
2407 | unsigned long *addr; | |
2408 | int end = bit + nr_bits; | |
2409 | int n = 0; | |
2410 | ||
2411 | while (bit < end) { | |
2412 | if (((bit & (BITS_PER_LONG - 1)) == 0) && | |
2413 | ((end - bit) >= BITS_PER_LONG)) { | |
2414 | /* Try to clear whole word at once */ | |
2415 | addr = bitmap + BIT_WORD(bit); | |
2416 | if (*addr == ULONG_MAX) { | |
2417 | *addr = 0; | |
2418 | n += BITS_PER_LONG; | |
2419 | bit += BITS_PER_LONG; | |
2420 | continue; | |
2421 | } | |
2422 | } | |
2423 | ||
2424 | if (test_and_clear_bit(bit, bitmap)) | |
2425 | n++; | |
2426 | bit++; | |
2427 | } | |
2428 | ||
2429 | return n; | |
2430 | } | |
2431 | ||
2432 | /* | |
2433 | * Invalidate all the blocks in the range [block..block+nr_blocks-1]. | |
2434 | */ | |
2435 | int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone, | |
2436 | sector_t chunk_block, unsigned int nr_blocks) | |
2437 | { | |
2438 | unsigned int count, bit, nr_bits; | |
2439 | struct dmz_mblock *mblk; | |
2440 | unsigned int n = 0; | |
2441 | ||
ca1a7045 | 2442 | dmz_zmd_debug(zmd, "=> INVALIDATE zone %u, block %llu, %u blocks", |
b7122873 | 2443 | zone->id, (u64)chunk_block, nr_blocks); |
3b1a94c8 | 2444 | |
36820560 | 2445 | WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks); |
3b1a94c8 DLM |
2446 | |
2447 | while (nr_blocks) { | |
2448 | /* Get bitmap block */ | |
2449 | mblk = dmz_get_bitmap(zmd, zone, chunk_block); | |
2450 | if (IS_ERR(mblk)) | |
2451 | return PTR_ERR(mblk); | |
2452 | ||
2453 | /* Clear bits */ | |
2454 | bit = chunk_block & DMZ_BLOCK_MASK_BITS; | |
b3996295 | 2455 | nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit); |
3b1a94c8 DLM |
2456 | |
2457 | count = dmz_clear_bits((unsigned long *)mblk->data, | |
2458 | bit, nr_bits); | |
2459 | if (count) { | |
2460 | dmz_dirty_mblock(zmd, mblk); | |
2461 | n += count; | |
2462 | } | |
2463 | dmz_release_mblock(zmd, mblk); | |
2464 | ||
2465 | nr_blocks -= nr_bits; | |
2466 | chunk_block += nr_bits; | |
2467 | } | |
2468 | ||
2469 | if (zone->weight >= n) | |
2470 | zone->weight -= n; | |
2471 | else { | |
ca1a7045 | 2472 | dmz_zmd_warn(zmd, "Zone %u: weight %u should be >= %u", |
b7122873 | 2473 | zone->id, zone->weight, n); |
3b1a94c8 DLM |
2474 | zone->weight = 0; |
2475 | } | |
2476 | ||
2477 | return 0; | |
2478 | } | |
2479 | ||
2480 | /* | |
2481 | * Get a block bit value. | |
2482 | */ | |
2483 | static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone, | |
2484 | sector_t chunk_block) | |
2485 | { | |
2486 | struct dmz_mblock *mblk; | |
2487 | int ret; | |
2488 | ||
36820560 | 2489 | WARN_ON(chunk_block >= zmd->zone_nr_blocks); |
3b1a94c8 DLM |
2490 | |
2491 | /* Get bitmap block */ | |
2492 | mblk = dmz_get_bitmap(zmd, zone, chunk_block); | |
2493 | if (IS_ERR(mblk)) | |
2494 | return PTR_ERR(mblk); | |
2495 | ||
2496 | /* Get offset */ | |
2497 | ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS, | |
2498 | (unsigned long *) mblk->data) != 0; | |
2499 | ||
2500 | dmz_release_mblock(zmd, mblk); | |
2501 | ||
2502 | return ret; | |
2503 | } | |
2504 | ||
2505 | /* | |
2506 | * Return the number of blocks from chunk_block to the first block with a bit | |
2507 | * value specified by set. Search at most nr_blocks blocks from chunk_block. | |
2508 | */ | |
2509 | static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone, | |
2510 | sector_t chunk_block, unsigned int nr_blocks, | |
2511 | int set) | |
2512 | { | |
2513 | struct dmz_mblock *mblk; | |
2514 | unsigned int bit, set_bit, nr_bits; | |
b3996295 | 2515 | unsigned int zone_bits = zmd->zone_bits_per_mblk; |
3b1a94c8 DLM |
2516 | unsigned long *bitmap; |
2517 | int n = 0; | |
2518 | ||
36820560 | 2519 | WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks); |
3b1a94c8 DLM |
2520 | |
2521 | while (nr_blocks) { | |
2522 | /* Get bitmap block */ | |
2523 | mblk = dmz_get_bitmap(zmd, zone, chunk_block); | |
2524 | if (IS_ERR(mblk)) | |
2525 | return PTR_ERR(mblk); | |
2526 | ||
2527 | /* Get offset */ | |
2528 | bitmap = (unsigned long *) mblk->data; | |
2529 | bit = chunk_block & DMZ_BLOCK_MASK_BITS; | |
b3996295 | 2530 | nr_bits = min(nr_blocks, zone_bits - bit); |
3b1a94c8 | 2531 | if (set) |
b3996295 | 2532 | set_bit = find_next_bit(bitmap, zone_bits, bit); |
3b1a94c8 | 2533 | else |
b3996295 | 2534 | set_bit = find_next_zero_bit(bitmap, zone_bits, bit); |
3b1a94c8 DLM |
2535 | dmz_release_mblock(zmd, mblk); |
2536 | ||
2537 | n += set_bit - bit; | |
b3996295 | 2538 | if (set_bit < zone_bits) |
3b1a94c8 DLM |
2539 | break; |
2540 | ||
2541 | nr_blocks -= nr_bits; | |
2542 | chunk_block += nr_bits; | |
2543 | } | |
2544 | ||
2545 | return n; | |
2546 | } | |
2547 | ||
2548 | /* | |
2549 | * Test if chunk_block is valid. If it is, the number of consecutive | |
2550 | * valid blocks from chunk_block will be returned. | |
2551 | */ | |
2552 | int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone, | |
2553 | sector_t chunk_block) | |
2554 | { | |
2555 | int valid; | |
2556 | ||
2557 | valid = dmz_test_block(zmd, zone, chunk_block); | |
2558 | if (valid <= 0) | |
2559 | return valid; | |
2560 | ||
2561 | /* The block is valid: get the number of valid blocks from block */ | |
2562 | return dmz_to_next_set_block(zmd, zone, chunk_block, | |
36820560 | 2563 | zmd->zone_nr_blocks - chunk_block, 0); |
3b1a94c8 DLM |
2564 | } |
2565 | ||
2566 | /* | |
2567 | * Find the first valid block from @chunk_block in @zone. | |
2568 | * If such a block is found, its number is returned using | |
2569 | * @chunk_block and the total number of valid blocks from @chunk_block | |
2570 | * is returned. | |
2571 | */ | |
2572 | int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone, | |
2573 | sector_t *chunk_block) | |
2574 | { | |
2575 | sector_t start_block = *chunk_block; | |
2576 | int ret; | |
2577 | ||
2578 | ret = dmz_to_next_set_block(zmd, zone, start_block, | |
36820560 | 2579 | zmd->zone_nr_blocks - start_block, 1); |
3b1a94c8 DLM |
2580 | if (ret < 0) |
2581 | return ret; | |
2582 | ||
2583 | start_block += ret; | |
2584 | *chunk_block = start_block; | |
2585 | ||
2586 | return dmz_to_next_set_block(zmd, zone, start_block, | |
36820560 | 2587 | zmd->zone_nr_blocks - start_block, 0); |
3b1a94c8 DLM |
2588 | } |
2589 | ||
2590 | /* | |
2591 | * Count the number of bits set starting from bit up to bit + nr_bits - 1. | |
2592 | */ | |
2593 | static int dmz_count_bits(void *bitmap, int bit, int nr_bits) | |
2594 | { | |
2595 | unsigned long *addr; | |
2596 | int end = bit + nr_bits; | |
2597 | int n = 0; | |
2598 | ||
2599 | while (bit < end) { | |
2600 | if (((bit & (BITS_PER_LONG - 1)) == 0) && | |
2601 | ((end - bit) >= BITS_PER_LONG)) { | |
2602 | addr = (unsigned long *)bitmap + BIT_WORD(bit); | |
2603 | if (*addr == ULONG_MAX) { | |
2604 | n += BITS_PER_LONG; | |
2605 | bit += BITS_PER_LONG; | |
2606 | continue; | |
2607 | } | |
2608 | } | |
2609 | ||
2610 | if (test_bit(bit, bitmap)) | |
2611 | n++; | |
2612 | bit++; | |
2613 | } | |
2614 | ||
2615 | return n; | |
2616 | } | |
2617 | ||
2618 | /* | |
2619 | * Get a zone weight. | |
2620 | */ | |
2621 | static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone) | |
2622 | { | |
2623 | struct dmz_mblock *mblk; | |
2624 | sector_t chunk_block = 0; | |
2625 | unsigned int bit, nr_bits; | |
36820560 | 2626 | unsigned int nr_blocks = zmd->zone_nr_blocks; |
3b1a94c8 DLM |
2627 | void *bitmap; |
2628 | int n = 0; | |
2629 | ||
2630 | while (nr_blocks) { | |
2631 | /* Get bitmap block */ | |
2632 | mblk = dmz_get_bitmap(zmd, zone, chunk_block); | |
2633 | if (IS_ERR(mblk)) { | |
2634 | n = 0; | |
2635 | break; | |
2636 | } | |
2637 | ||
2638 | /* Count bits in this block */ | |
2639 | bitmap = mblk->data; | |
2640 | bit = chunk_block & DMZ_BLOCK_MASK_BITS; | |
b3996295 | 2641 | nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit); |
3b1a94c8 DLM |
2642 | n += dmz_count_bits(bitmap, bit, nr_bits); |
2643 | ||
2644 | dmz_release_mblock(zmd, mblk); | |
2645 | ||
2646 | nr_blocks -= nr_bits; | |
2647 | chunk_block += nr_bits; | |
2648 | } | |
2649 | ||
2650 | zone->weight = n; | |
2651 | } | |
2652 | ||
2653 | /* | |
2654 | * Cleanup the zoned metadata resources. | |
2655 | */ | |
2656 | static void dmz_cleanup_metadata(struct dmz_metadata *zmd) | |
2657 | { | |
2658 | struct rb_root *root; | |
2659 | struct dmz_mblock *mblk, *next; | |
2660 | int i; | |
2661 | ||
2662 | /* Release zone mapping resources */ | |
2663 | if (zmd->map_mblk) { | |
2664 | for (i = 0; i < zmd->nr_map_blocks; i++) | |
2665 | dmz_release_mblock(zmd, zmd->map_mblk[i]); | |
2666 | kfree(zmd->map_mblk); | |
2667 | zmd->map_mblk = NULL; | |
2668 | } | |
2669 | ||
2670 | /* Release super blocks */ | |
2671 | for (i = 0; i < 2; i++) { | |
2672 | if (zmd->sb[i].mblk) { | |
2673 | dmz_free_mblock(zmd, zmd->sb[i].mblk); | |
2674 | zmd->sb[i].mblk = NULL; | |
2675 | } | |
2676 | } | |
2677 | ||
2678 | /* Free cached blocks */ | |
2679 | while (!list_empty(&zmd->mblk_dirty_list)) { | |
2680 | mblk = list_first_entry(&zmd->mblk_dirty_list, | |
2681 | struct dmz_mblock, link); | |
ca1a7045 | 2682 | dmz_zmd_warn(zmd, "mblock %llu still in dirty list (ref %u)", |
33c2865f | 2683 | (u64)mblk->no, mblk->ref); |
3b1a94c8 DLM |
2684 | list_del_init(&mblk->link); |
2685 | rb_erase(&mblk->node, &zmd->mblk_rbtree); | |
2686 | dmz_free_mblock(zmd, mblk); | |
2687 | } | |
2688 | ||
2689 | while (!list_empty(&zmd->mblk_lru_list)) { | |
2690 | mblk = list_first_entry(&zmd->mblk_lru_list, | |
2691 | struct dmz_mblock, link); | |
2692 | list_del_init(&mblk->link); | |
2693 | rb_erase(&mblk->node, &zmd->mblk_rbtree); | |
2694 | dmz_free_mblock(zmd, mblk); | |
2695 | } | |
2696 | ||
2697 | /* Sanity checks: the mblock rbtree should now be empty */ | |
2698 | root = &zmd->mblk_rbtree; | |
2699 | rbtree_postorder_for_each_entry_safe(mblk, next, root, node) { | |
ca1a7045 | 2700 | dmz_zmd_warn(zmd, "mblock %llu ref %u still in rbtree", |
33c2865f DLM |
2701 | (u64)mblk->no, mblk->ref); |
2702 | mblk->ref = 0; | |
3b1a94c8 DLM |
2703 | dmz_free_mblock(zmd, mblk); |
2704 | } | |
2705 | ||
2706 | /* Free the zone descriptors */ | |
2707 | dmz_drop_zones(zmd); | |
d5ffebdd MS |
2708 | |
2709 | mutex_destroy(&zmd->mblk_flush_lock); | |
2710 | mutex_destroy(&zmd->map_lock); | |
3b1a94c8 DLM |
2711 | } |
2712 | ||
ca1a7045 HR |
2713 | static void dmz_print_dev(struct dmz_metadata *zmd, int num) |
2714 | { | |
2715 | struct dmz_dev *dev = &zmd->dev[num]; | |
2716 | ||
bd5c4031 HR |
2717 | if (bdev_zoned_model(dev->bdev) == BLK_ZONED_NONE) |
2718 | dmz_dev_info(dev, "Regular block device"); | |
2719 | else | |
2720 | dmz_dev_info(dev, "Host-%s zoned block device", | |
2721 | bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ? | |
2722 | "aware" : "managed"); | |
2723 | if (zmd->sb_version > 1) { | |
2724 | sector_t sector_offset = | |
2725 | dev->zone_offset << zmd->zone_nr_sectors_shift; | |
2726 | ||
2727 | dmz_dev_info(dev, " %llu 512-byte logical sectors (offset %llu)", | |
2728 | (u64)dev->capacity, (u64)sector_offset); | |
2729 | dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors (offset %llu)", | |
2730 | dev->nr_zones, (u64)zmd->zone_nr_sectors, | |
2731 | (u64)dev->zone_offset); | |
2732 | } else { | |
2733 | dmz_dev_info(dev, " %llu 512-byte logical sectors", | |
2734 | (u64)dev->capacity); | |
2735 | dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors", | |
2736 | dev->nr_zones, (u64)zmd->zone_nr_sectors); | |
2737 | } | |
ca1a7045 HR |
2738 | } |
2739 | ||
3b1a94c8 DLM |
2740 | /* |
2741 | * Initialize the zoned metadata. | |
2742 | */ | |
bd5c4031 HR |
2743 | int dmz_ctr_metadata(struct dmz_dev *dev, int num_dev, |
2744 | struct dmz_metadata **metadata, | |
2234e732 | 2745 | const char *devname) |
3b1a94c8 DLM |
2746 | { |
2747 | struct dmz_metadata *zmd; | |
b7122873 | 2748 | unsigned int i; |
3b1a94c8 DLM |
2749 | struct dm_zone *zone; |
2750 | int ret; | |
2751 | ||
2752 | zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL); | |
2753 | if (!zmd) | |
2754 | return -ENOMEM; | |
2755 | ||
2234e732 | 2756 | strcpy(zmd->devname, devname); |
3b1a94c8 | 2757 | zmd->dev = dev; |
bd5c4031 | 2758 | zmd->nr_devs = num_dev; |
3b1a94c8 DLM |
2759 | zmd->mblk_rbtree = RB_ROOT; |
2760 | init_rwsem(&zmd->mblk_sem); | |
2761 | mutex_init(&zmd->mblk_flush_lock); | |
2762 | spin_lock_init(&zmd->mblk_lock); | |
2763 | INIT_LIST_HEAD(&zmd->mblk_lru_list); | |
2764 | INIT_LIST_HEAD(&zmd->mblk_dirty_list); | |
2765 | ||
2766 | mutex_init(&zmd->map_lock); | |
2767 | atomic_set(&zmd->unmap_nr_rnd, 0); | |
2768 | INIT_LIST_HEAD(&zmd->unmap_rnd_list); | |
2769 | INIT_LIST_HEAD(&zmd->map_rnd_list); | |
2770 | ||
34f5affd HR |
2771 | atomic_set(&zmd->unmap_nr_cache, 0); |
2772 | INIT_LIST_HEAD(&zmd->unmap_cache_list); | |
2773 | INIT_LIST_HEAD(&zmd->map_cache_list); | |
2774 | ||
3b1a94c8 DLM |
2775 | atomic_set(&zmd->unmap_nr_seq, 0); |
2776 | INIT_LIST_HEAD(&zmd->unmap_seq_list); | |
2777 | INIT_LIST_HEAD(&zmd->map_seq_list); | |
2778 | ||
2779 | atomic_set(&zmd->nr_reserved_seq_zones, 0); | |
2780 | INIT_LIST_HEAD(&zmd->reserved_seq_zones_list); | |
2781 | ||
2782 | init_waitqueue_head(&zmd->free_wq); | |
2783 | ||
2784 | /* Initialize zone descriptors */ | |
2785 | ret = dmz_init_zones(zmd); | |
2786 | if (ret) | |
2787 | goto err; | |
2788 | ||
2789 | /* Get super block */ | |
2790 | ret = dmz_load_sb(zmd); | |
2791 | if (ret) | |
2792 | goto err; | |
2793 | ||
2794 | /* Set metadata zones starting from sb_zone */ | |
3b1a94c8 | 2795 | for (i = 0; i < zmd->nr_meta_zones << 1; i++) { |
735bd7e4 | 2796 | zone = dmz_get(zmd, zmd->sb[0].zone->id + i); |
34f5affd | 2797 | if (!dmz_is_rnd(zone) && !dmz_is_cache(zone)) { |
bd5c4031 HR |
2798 | dmz_zmd_err(zmd, |
2799 | "metadata zone %d is not random", i); | |
2800 | ret = -ENXIO; | |
3b1a94c8 | 2801 | goto err; |
bd5c4031 HR |
2802 | } |
2803 | set_bit(DMZ_META, &zone->flags); | |
2804 | } | |
2805 | if (zmd->sb[2].zone) { | |
2806 | zone = dmz_get(zmd, zmd->sb[2].zone->id); | |
2807 | if (!zone) { | |
2808 | dmz_zmd_err(zmd, | |
2809 | "Tertiary metadata zone not present"); | |
2810 | ret = -ENXIO; | |
2811 | goto err; | |
2812 | } | |
3b1a94c8 DLM |
2813 | set_bit(DMZ_META, &zone->flags); |
2814 | } | |
3b1a94c8 DLM |
2815 | /* Load mapping table */ |
2816 | ret = dmz_load_mapping(zmd); | |
2817 | if (ret) | |
2818 | goto err; | |
2819 | ||
2820 | /* | |
2821 | * Cache size boundaries: allow at least 2 super blocks, the chunk map | |
2822 | * blocks and enough blocks to be able to cache the bitmap blocks of | |
2823 | * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow | |
2824 | * the cache to add 512 more metadata blocks. | |
2825 | */ | |
2826 | zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16; | |
2827 | zmd->max_nr_mblks = zmd->min_nr_mblks + 512; | |
2828 | zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count; | |
2829 | zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan; | |
2830 | zmd->mblk_shrinker.seeks = DEFAULT_SEEKS; | |
2831 | ||
2832 | /* Metadata cache shrinker */ | |
2833 | ret = register_shrinker(&zmd->mblk_shrinker); | |
2834 | if (ret) { | |
ca1a7045 | 2835 | dmz_zmd_err(zmd, "Register metadata cache shrinker failed"); |
3b1a94c8 DLM |
2836 | goto err; |
2837 | } | |
2838 | ||
bd5c4031 HR |
2839 | dmz_zmd_info(zmd, "DM-Zoned metadata version %d", zmd->sb_version); |
2840 | for (i = 0; i < zmd->nr_devs; i++) | |
2841 | dmz_print_dev(zmd, i); | |
ca1a7045 HR |
2842 | |
2843 | dmz_zmd_info(zmd, " %u zones of %llu 512-byte logical sectors", | |
36820560 | 2844 | zmd->nr_zones, (u64)zmd->zone_nr_sectors); |
ae3c1f11 HR |
2845 | dmz_zmd_debug(zmd, " %u metadata zones", |
2846 | zmd->nr_meta_zones * 2); | |
2847 | dmz_zmd_debug(zmd, " %u data zones for %u chunks", | |
2848 | zmd->nr_data_zones, zmd->nr_chunks); | |
34f5affd HR |
2849 | dmz_zmd_debug(zmd, " %u cache zones (%u unmapped)", |
2850 | zmd->nr_cache, atomic_read(&zmd->unmap_nr_cache)); | |
ae3c1f11 HR |
2851 | dmz_zmd_debug(zmd, " %u random zones (%u unmapped)", |
2852 | zmd->nr_rnd, atomic_read(&zmd->unmap_nr_rnd)); | |
2853 | dmz_zmd_debug(zmd, " %u sequential zones (%u unmapped)", | |
2854 | zmd->nr_seq, atomic_read(&zmd->unmap_nr_seq)); | |
2855 | dmz_zmd_debug(zmd, " %u reserved sequential data zones", | |
2856 | zmd->nr_reserved_seq); | |
ca1a7045 HR |
2857 | dmz_zmd_debug(zmd, "Format:"); |
2858 | dmz_zmd_debug(zmd, "%u metadata blocks per set (%u max cache)", | |
3b1a94c8 | 2859 | zmd->nr_meta_blocks, zmd->max_nr_mblks); |
ca1a7045 | 2860 | dmz_zmd_debug(zmd, " %u data zone mapping blocks", |
3b1a94c8 | 2861 | zmd->nr_map_blocks); |
ca1a7045 | 2862 | dmz_zmd_debug(zmd, " %u bitmap blocks", |
3b1a94c8 DLM |
2863 | zmd->nr_bitmap_blocks); |
2864 | ||
2865 | *metadata = zmd; | |
2866 | ||
2867 | return 0; | |
2868 | err: | |
2869 | dmz_cleanup_metadata(zmd); | |
2870 | kfree(zmd); | |
2871 | *metadata = NULL; | |
2872 | ||
2873 | return ret; | |
2874 | } | |
2875 | ||
2876 | /* | |
2877 | * Cleanup the zoned metadata resources. | |
2878 | */ | |
2879 | void dmz_dtr_metadata(struct dmz_metadata *zmd) | |
2880 | { | |
2881 | unregister_shrinker(&zmd->mblk_shrinker); | |
2882 | dmz_cleanup_metadata(zmd); | |
2883 | kfree(zmd); | |
2884 | } | |
2885 | ||
2886 | /* | |
2887 | * Check zone information on resume. | |
2888 | */ | |
2889 | int dmz_resume_metadata(struct dmz_metadata *zmd) | |
2890 | { | |
3b1a94c8 DLM |
2891 | struct dm_zone *zone; |
2892 | sector_t wp_block; | |
2893 | unsigned int i; | |
2894 | int ret; | |
2895 | ||
2896 | /* Check zones */ | |
36820560 | 2897 | for (i = 0; i < zmd->nr_zones; i++) { |
3b1a94c8 DLM |
2898 | zone = dmz_get(zmd, i); |
2899 | if (!zone) { | |
ca1a7045 | 2900 | dmz_zmd_err(zmd, "Unable to get zone %u", i); |
3b1a94c8 DLM |
2901 | return -EIO; |
2902 | } | |
3b1a94c8 DLM |
2903 | wp_block = zone->wp_block; |
2904 | ||
2905 | ret = dmz_update_zone(zmd, zone); | |
2906 | if (ret) { | |
ca1a7045 | 2907 | dmz_zmd_err(zmd, "Broken zone %u", i); |
3b1a94c8 DLM |
2908 | return ret; |
2909 | } | |
2910 | ||
2911 | if (dmz_is_offline(zone)) { | |
ca1a7045 | 2912 | dmz_zmd_warn(zmd, "Zone %u is offline", i); |
3b1a94c8 DLM |
2913 | continue; |
2914 | } | |
2915 | ||
2916 | /* Check write pointer */ | |
2917 | if (!dmz_is_seq(zone)) | |
2918 | zone->wp_block = 0; | |
2919 | else if (zone->wp_block != wp_block) { | |
ca1a7045 | 2920 | dmz_zmd_err(zmd, "Zone %u: Invalid wp (%llu / %llu)", |
3b1a94c8 DLM |
2921 | i, (u64)zone->wp_block, (u64)wp_block); |
2922 | zone->wp_block = wp_block; | |
2923 | dmz_invalidate_blocks(zmd, zone, zone->wp_block, | |
36820560 | 2924 | zmd->zone_nr_blocks - zone->wp_block); |
3b1a94c8 DLM |
2925 | } |
2926 | } | |
2927 | ||
2928 | return 0; | |
2929 | } |