Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
53b381b3 DW |
2 | /* |
3 | * Copyright (C) 2012 Fusion-io All rights reserved. | |
4 | * Copyright (C) 2012 Intel Corp. All rights reserved. | |
53b381b3 | 5 | */ |
c1d7c514 | 6 | |
53b381b3 | 7 | #include <linux/sched.h> |
53b381b3 DW |
8 | #include <linux/bio.h> |
9 | #include <linux/slab.h> | |
53b381b3 | 10 | #include <linux/blkdev.h> |
53b381b3 DW |
11 | #include <linux/raid/pq.h> |
12 | #include <linux/hash.h> | |
13 | #include <linux/list_sort.h> | |
14 | #include <linux/raid/xor.h> | |
818e010b | 15 | #include <linux/mm.h> |
cea62800 | 16 | #include "misc.h" |
53b381b3 | 17 | #include "ctree.h" |
53b381b3 | 18 | #include "disk-io.h" |
53b381b3 DW |
19 | #include "volumes.h" |
20 | #include "raid56.h" | |
21 | #include "async-thread.h" | |
53b381b3 DW |
22 | |
23 | /* set when additional merges to this rbio are not allowed */ | |
24 | #define RBIO_RMW_LOCKED_BIT 1 | |
25 | ||
4ae10b3a CM |
26 | /* |
27 | * set when this rbio is sitting in the hash, but it is just a cache | |
28 | * of past RMW | |
29 | */ | |
30 | #define RBIO_CACHE_BIT 2 | |
31 | ||
32 | /* | |
33 | * set when it is safe to trust the stripe_pages for caching | |
34 | */ | |
35 | #define RBIO_CACHE_READY_BIT 3 | |
36 | ||
4ae10b3a CM |
37 | #define RBIO_CACHE_SIZE 1024 |
38 | ||
8a953348 DS |
39 | #define BTRFS_STRIPE_HASH_TABLE_BITS 11 |
40 | ||
41 | /* Used by the raid56 code to lock stripes for read/modify/write */ | |
42 | struct btrfs_stripe_hash { | |
43 | struct list_head hash_list; | |
44 | spinlock_t lock; | |
45 | }; | |
46 | ||
47 | /* Used by the raid56 code to lock stripes for read/modify/write */ | |
48 | struct btrfs_stripe_hash_table { | |
49 | struct list_head stripe_cache; | |
50 | spinlock_t cache_lock; | |
51 | int cache_size; | |
52 | struct btrfs_stripe_hash table[]; | |
53 | }; | |
54 | ||
eb357060 QW |
55 | /* |
56 | * A bvec like structure to present a sector inside a page. | |
57 | * | |
58 | * Unlike bvec we don't need bvlen, as it's fixed to sectorsize. | |
59 | */ | |
60 | struct sector_ptr { | |
61 | struct page *page; | |
00425dd9 QW |
62 | unsigned int pgoff:24; |
63 | unsigned int uptodate:8; | |
eb357060 QW |
64 | }; |
65 | ||
53b381b3 DW |
66 | static int __raid56_parity_recover(struct btrfs_raid_bio *rbio); |
67 | static noinline void finish_rmw(struct btrfs_raid_bio *rbio); | |
385de0ef CH |
68 | static void rmw_work(struct work_struct *work); |
69 | static void read_rebuild_work(struct work_struct *work); | |
53b381b3 DW |
70 | static int fail_bio_stripe(struct btrfs_raid_bio *rbio, struct bio *bio); |
71 | static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed); | |
53b381b3 DW |
72 | static void index_rbio_pages(struct btrfs_raid_bio *rbio); |
73 | static int alloc_rbio_pages(struct btrfs_raid_bio *rbio); | |
74 | ||
5a6ac9ea MX |
75 | static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, |
76 | int need_check); | |
385de0ef | 77 | static void scrub_parity_work(struct work_struct *work); |
5a6ac9ea | 78 | |
797d74b7 QW |
79 | static void free_raid_bio_pointers(struct btrfs_raid_bio *rbio) |
80 | { | |
81 | kfree(rbio->stripe_pages); | |
82 | kfree(rbio->bio_sectors); | |
83 | kfree(rbio->stripe_sectors); | |
84 | kfree(rbio->finish_pointers); | |
85 | } | |
86 | ||
ff2b64a2 QW |
87 | static void free_raid_bio(struct btrfs_raid_bio *rbio) |
88 | { | |
89 | int i; | |
90 | ||
91 | if (!refcount_dec_and_test(&rbio->refs)) | |
92 | return; | |
93 | ||
94 | WARN_ON(!list_empty(&rbio->stripe_cache)); | |
95 | WARN_ON(!list_empty(&rbio->hash_list)); | |
96 | WARN_ON(!bio_list_empty(&rbio->bio_list)); | |
97 | ||
98 | for (i = 0; i < rbio->nr_pages; i++) { | |
99 | if (rbio->stripe_pages[i]) { | |
100 | __free_page(rbio->stripe_pages[i]); | |
101 | rbio->stripe_pages[i] = NULL; | |
102 | } | |
103 | } | |
104 | ||
105 | btrfs_put_bioc(rbio->bioc); | |
797d74b7 | 106 | free_raid_bio_pointers(rbio); |
ff2b64a2 QW |
107 | kfree(rbio); |
108 | } | |
109 | ||
385de0ef | 110 | static void start_async_work(struct btrfs_raid_bio *rbio, work_func_t work_func) |
ac638859 | 111 | { |
385de0ef CH |
112 | INIT_WORK(&rbio->work, work_func); |
113 | queue_work(rbio->bioc->fs_info->rmw_workers, &rbio->work); | |
ac638859 DS |
114 | } |
115 | ||
53b381b3 DW |
116 | /* |
117 | * the stripe hash table is used for locking, and to collect | |
118 | * bios in hopes of making a full stripe | |
119 | */ | |
120 | int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info) | |
121 | { | |
122 | struct btrfs_stripe_hash_table *table; | |
123 | struct btrfs_stripe_hash_table *x; | |
124 | struct btrfs_stripe_hash *cur; | |
125 | struct btrfs_stripe_hash *h; | |
126 | int num_entries = 1 << BTRFS_STRIPE_HASH_TABLE_BITS; | |
127 | int i; | |
128 | ||
129 | if (info->stripe_hash_table) | |
130 | return 0; | |
131 | ||
83c8266a DS |
132 | /* |
133 | * The table is large, starting with order 4 and can go as high as | |
134 | * order 7 in case lock debugging is turned on. | |
135 | * | |
136 | * Try harder to allocate and fallback to vmalloc to lower the chance | |
137 | * of a failing mount. | |
138 | */ | |
ee787f95 | 139 | table = kvzalloc(struct_size(table, table, num_entries), GFP_KERNEL); |
818e010b DS |
140 | if (!table) |
141 | return -ENOMEM; | |
53b381b3 | 142 | |
4ae10b3a CM |
143 | spin_lock_init(&table->cache_lock); |
144 | INIT_LIST_HEAD(&table->stripe_cache); | |
145 | ||
53b381b3 DW |
146 | h = table->table; |
147 | ||
148 | for (i = 0; i < num_entries; i++) { | |
149 | cur = h + i; | |
150 | INIT_LIST_HEAD(&cur->hash_list); | |
151 | spin_lock_init(&cur->lock); | |
53b381b3 DW |
152 | } |
153 | ||
154 | x = cmpxchg(&info->stripe_hash_table, NULL, table); | |
fe3b7bb0 | 155 | kvfree(x); |
53b381b3 DW |
156 | return 0; |
157 | } | |
158 | ||
4ae10b3a CM |
159 | /* |
160 | * caching an rbio means to copy anything from the | |
ac26df8b | 161 | * bio_sectors array into the stripe_pages array. We |
4ae10b3a CM |
162 | * use the page uptodate bit in the stripe cache array |
163 | * to indicate if it has valid data | |
164 | * | |
165 | * once the caching is done, we set the cache ready | |
166 | * bit. | |
167 | */ | |
168 | static void cache_rbio_pages(struct btrfs_raid_bio *rbio) | |
169 | { | |
170 | int i; | |
4ae10b3a CM |
171 | int ret; |
172 | ||
173 | ret = alloc_rbio_pages(rbio); | |
174 | if (ret) | |
175 | return; | |
176 | ||
00425dd9 QW |
177 | for (i = 0; i < rbio->nr_sectors; i++) { |
178 | /* Some range not covered by bio (partial write), skip it */ | |
179 | if (!rbio->bio_sectors[i].page) | |
180 | continue; | |
181 | ||
182 | ASSERT(rbio->stripe_sectors[i].page); | |
183 | memcpy_page(rbio->stripe_sectors[i].page, | |
184 | rbio->stripe_sectors[i].pgoff, | |
185 | rbio->bio_sectors[i].page, | |
186 | rbio->bio_sectors[i].pgoff, | |
187 | rbio->bioc->fs_info->sectorsize); | |
188 | rbio->stripe_sectors[i].uptodate = 1; | |
189 | } | |
4ae10b3a CM |
190 | set_bit(RBIO_CACHE_READY_BIT, &rbio->flags); |
191 | } | |
192 | ||
53b381b3 DW |
193 | /* |
194 | * we hash on the first logical address of the stripe | |
195 | */ | |
196 | static int rbio_bucket(struct btrfs_raid_bio *rbio) | |
197 | { | |
4c664611 | 198 | u64 num = rbio->bioc->raid_map[0]; |
53b381b3 DW |
199 | |
200 | /* | |
201 | * we shift down quite a bit. We're using byte | |
202 | * addressing, and most of the lower bits are zeros. | |
203 | * This tends to upset hash_64, and it consistently | |
204 | * returns just one or two different values. | |
205 | * | |
206 | * shifting off the lower bits fixes things. | |
207 | */ | |
208 | return hash_64(num >> 16, BTRFS_STRIPE_HASH_TABLE_BITS); | |
209 | } | |
210 | ||
d4e28d9b QW |
211 | static bool full_page_sectors_uptodate(struct btrfs_raid_bio *rbio, |
212 | unsigned int page_nr) | |
213 | { | |
214 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; | |
215 | const u32 sectors_per_page = PAGE_SIZE / sectorsize; | |
216 | int i; | |
217 | ||
218 | ASSERT(page_nr < rbio->nr_pages); | |
219 | ||
220 | for (i = sectors_per_page * page_nr; | |
221 | i < sectors_per_page * page_nr + sectors_per_page; | |
222 | i++) { | |
223 | if (!rbio->stripe_sectors[i].uptodate) | |
224 | return false; | |
225 | } | |
226 | return true; | |
227 | } | |
228 | ||
eb357060 QW |
229 | /* |
230 | * Update the stripe_sectors[] array to use correct page and pgoff | |
231 | * | |
232 | * Should be called every time any page pointer in stripes_pages[] got modified. | |
233 | */ | |
234 | static void index_stripe_sectors(struct btrfs_raid_bio *rbio) | |
235 | { | |
236 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; | |
237 | u32 offset; | |
238 | int i; | |
239 | ||
240 | for (i = 0, offset = 0; i < rbio->nr_sectors; i++, offset += sectorsize) { | |
241 | int page_index = offset >> PAGE_SHIFT; | |
242 | ||
243 | ASSERT(page_index < rbio->nr_pages); | |
244 | rbio->stripe_sectors[i].page = rbio->stripe_pages[page_index]; | |
245 | rbio->stripe_sectors[i].pgoff = offset_in_page(offset); | |
246 | } | |
247 | } | |
248 | ||
4d100466 QW |
249 | static void steal_rbio_page(struct btrfs_raid_bio *src, |
250 | struct btrfs_raid_bio *dest, int page_nr) | |
251 | { | |
252 | const u32 sectorsize = src->bioc->fs_info->sectorsize; | |
253 | const u32 sectors_per_page = PAGE_SIZE / sectorsize; | |
254 | int i; | |
255 | ||
256 | if (dest->stripe_pages[page_nr]) | |
257 | __free_page(dest->stripe_pages[page_nr]); | |
258 | dest->stripe_pages[page_nr] = src->stripe_pages[page_nr]; | |
259 | src->stripe_pages[page_nr] = NULL; | |
260 | ||
261 | /* Also update the sector->uptodate bits. */ | |
262 | for (i = sectors_per_page * page_nr; | |
263 | i < sectors_per_page * page_nr + sectors_per_page; i++) | |
264 | dest->stripe_sectors[i].uptodate = true; | |
265 | } | |
266 | ||
4ae10b3a | 267 | /* |
d4e28d9b QW |
268 | * Stealing an rbio means taking all the uptodate pages from the stripe array |
269 | * in the source rbio and putting them into the destination rbio. | |
270 | * | |
271 | * This will also update the involved stripe_sectors[] which are referring to | |
272 | * the old pages. | |
4ae10b3a CM |
273 | */ |
274 | static void steal_rbio(struct btrfs_raid_bio *src, struct btrfs_raid_bio *dest) | |
275 | { | |
276 | int i; | |
277 | struct page *s; | |
4ae10b3a CM |
278 | |
279 | if (!test_bit(RBIO_CACHE_READY_BIT, &src->flags)) | |
280 | return; | |
281 | ||
282 | for (i = 0; i < dest->nr_pages; i++) { | |
283 | s = src->stripe_pages[i]; | |
d4e28d9b | 284 | if (!s || !full_page_sectors_uptodate(src, i)) |
4ae10b3a | 285 | continue; |
4ae10b3a | 286 | |
4d100466 | 287 | steal_rbio_page(src, dest, i); |
4ae10b3a | 288 | } |
eb357060 QW |
289 | index_stripe_sectors(dest); |
290 | index_stripe_sectors(src); | |
4ae10b3a CM |
291 | } |
292 | ||
53b381b3 DW |
293 | /* |
294 | * merging means we take the bio_list from the victim and | |
295 | * splice it into the destination. The victim should | |
296 | * be discarded afterwards. | |
297 | * | |
298 | * must be called with dest->rbio_list_lock held | |
299 | */ | |
300 | static void merge_rbio(struct btrfs_raid_bio *dest, | |
301 | struct btrfs_raid_bio *victim) | |
302 | { | |
303 | bio_list_merge(&dest->bio_list, &victim->bio_list); | |
304 | dest->bio_list_bytes += victim->bio_list_bytes; | |
bd8f7e62 QW |
305 | /* Also inherit the bitmaps from @victim. */ |
306 | bitmap_or(&dest->dbitmap, &victim->dbitmap, &dest->dbitmap, | |
307 | dest->stripe_nsectors); | |
53b381b3 DW |
308 | bio_list_init(&victim->bio_list); |
309 | } | |
310 | ||
311 | /* | |
4ae10b3a CM |
312 | * used to prune items that are in the cache. The caller |
313 | * must hold the hash table lock. | |
314 | */ | |
315 | static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio) | |
316 | { | |
317 | int bucket = rbio_bucket(rbio); | |
318 | struct btrfs_stripe_hash_table *table; | |
319 | struct btrfs_stripe_hash *h; | |
320 | int freeit = 0; | |
321 | ||
322 | /* | |
323 | * check the bit again under the hash table lock. | |
324 | */ | |
325 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
326 | return; | |
327 | ||
6a258d72 | 328 | table = rbio->bioc->fs_info->stripe_hash_table; |
4ae10b3a CM |
329 | h = table->table + bucket; |
330 | ||
331 | /* hold the lock for the bucket because we may be | |
332 | * removing it from the hash table | |
333 | */ | |
334 | spin_lock(&h->lock); | |
335 | ||
336 | /* | |
337 | * hold the lock for the bio list because we need | |
338 | * to make sure the bio list is empty | |
339 | */ | |
340 | spin_lock(&rbio->bio_list_lock); | |
341 | ||
342 | if (test_and_clear_bit(RBIO_CACHE_BIT, &rbio->flags)) { | |
343 | list_del_init(&rbio->stripe_cache); | |
344 | table->cache_size -= 1; | |
345 | freeit = 1; | |
346 | ||
347 | /* if the bio list isn't empty, this rbio is | |
348 | * still involved in an IO. We take it out | |
349 | * of the cache list, and drop the ref that | |
350 | * was held for the list. | |
351 | * | |
352 | * If the bio_list was empty, we also remove | |
353 | * the rbio from the hash_table, and drop | |
354 | * the corresponding ref | |
355 | */ | |
356 | if (bio_list_empty(&rbio->bio_list)) { | |
357 | if (!list_empty(&rbio->hash_list)) { | |
358 | list_del_init(&rbio->hash_list); | |
dec95574 | 359 | refcount_dec(&rbio->refs); |
4ae10b3a CM |
360 | BUG_ON(!list_empty(&rbio->plug_list)); |
361 | } | |
362 | } | |
363 | } | |
364 | ||
365 | spin_unlock(&rbio->bio_list_lock); | |
366 | spin_unlock(&h->lock); | |
367 | ||
368 | if (freeit) | |
ff2b64a2 | 369 | free_raid_bio(rbio); |
4ae10b3a CM |
370 | } |
371 | ||
372 | /* | |
373 | * prune a given rbio from the cache | |
374 | */ | |
375 | static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio) | |
376 | { | |
377 | struct btrfs_stripe_hash_table *table; | |
378 | unsigned long flags; | |
379 | ||
380 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
381 | return; | |
382 | ||
6a258d72 | 383 | table = rbio->bioc->fs_info->stripe_hash_table; |
4ae10b3a CM |
384 | |
385 | spin_lock_irqsave(&table->cache_lock, flags); | |
386 | __remove_rbio_from_cache(rbio); | |
387 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
388 | } | |
389 | ||
390 | /* | |
391 | * remove everything in the cache | |
392 | */ | |
48a3b636 | 393 | static void btrfs_clear_rbio_cache(struct btrfs_fs_info *info) |
4ae10b3a CM |
394 | { |
395 | struct btrfs_stripe_hash_table *table; | |
396 | unsigned long flags; | |
397 | struct btrfs_raid_bio *rbio; | |
398 | ||
399 | table = info->stripe_hash_table; | |
400 | ||
401 | spin_lock_irqsave(&table->cache_lock, flags); | |
402 | while (!list_empty(&table->stripe_cache)) { | |
403 | rbio = list_entry(table->stripe_cache.next, | |
404 | struct btrfs_raid_bio, | |
405 | stripe_cache); | |
406 | __remove_rbio_from_cache(rbio); | |
407 | } | |
408 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
409 | } | |
410 | ||
411 | /* | |
412 | * remove all cached entries and free the hash table | |
413 | * used by unmount | |
53b381b3 DW |
414 | */ |
415 | void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info) | |
416 | { | |
417 | if (!info->stripe_hash_table) | |
418 | return; | |
4ae10b3a | 419 | btrfs_clear_rbio_cache(info); |
f749303b | 420 | kvfree(info->stripe_hash_table); |
53b381b3 DW |
421 | info->stripe_hash_table = NULL; |
422 | } | |
423 | ||
4ae10b3a CM |
424 | /* |
425 | * insert an rbio into the stripe cache. It | |
426 | * must have already been prepared by calling | |
427 | * cache_rbio_pages | |
428 | * | |
429 | * If this rbio was already cached, it gets | |
430 | * moved to the front of the lru. | |
431 | * | |
432 | * If the size of the rbio cache is too big, we | |
433 | * prune an item. | |
434 | */ | |
435 | static void cache_rbio(struct btrfs_raid_bio *rbio) | |
436 | { | |
437 | struct btrfs_stripe_hash_table *table; | |
438 | unsigned long flags; | |
439 | ||
440 | if (!test_bit(RBIO_CACHE_READY_BIT, &rbio->flags)) | |
441 | return; | |
442 | ||
6a258d72 | 443 | table = rbio->bioc->fs_info->stripe_hash_table; |
4ae10b3a CM |
444 | |
445 | spin_lock_irqsave(&table->cache_lock, flags); | |
446 | spin_lock(&rbio->bio_list_lock); | |
447 | ||
448 | /* bump our ref if we were not in the list before */ | |
449 | if (!test_and_set_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
dec95574 | 450 | refcount_inc(&rbio->refs); |
4ae10b3a CM |
451 | |
452 | if (!list_empty(&rbio->stripe_cache)){ | |
453 | list_move(&rbio->stripe_cache, &table->stripe_cache); | |
454 | } else { | |
455 | list_add(&rbio->stripe_cache, &table->stripe_cache); | |
456 | table->cache_size += 1; | |
457 | } | |
458 | ||
459 | spin_unlock(&rbio->bio_list_lock); | |
460 | ||
461 | if (table->cache_size > RBIO_CACHE_SIZE) { | |
462 | struct btrfs_raid_bio *found; | |
463 | ||
464 | found = list_entry(table->stripe_cache.prev, | |
465 | struct btrfs_raid_bio, | |
466 | stripe_cache); | |
467 | ||
468 | if (found != rbio) | |
469 | __remove_rbio_from_cache(found); | |
470 | } | |
471 | ||
472 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
4ae10b3a CM |
473 | } |
474 | ||
53b381b3 DW |
475 | /* |
476 | * helper function to run the xor_blocks api. It is only | |
477 | * able to do MAX_XOR_BLOCKS at a time, so we need to | |
478 | * loop through. | |
479 | */ | |
480 | static void run_xor(void **pages, int src_cnt, ssize_t len) | |
481 | { | |
482 | int src_off = 0; | |
483 | int xor_src_cnt = 0; | |
484 | void *dest = pages[src_cnt]; | |
485 | ||
486 | while(src_cnt > 0) { | |
487 | xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS); | |
488 | xor_blocks(xor_src_cnt, len, dest, pages + src_off); | |
489 | ||
490 | src_cnt -= xor_src_cnt; | |
491 | src_off += xor_src_cnt; | |
492 | } | |
493 | } | |
494 | ||
495 | /* | |
176571a1 DS |
496 | * Returns true if the bio list inside this rbio covers an entire stripe (no |
497 | * rmw required). | |
53b381b3 | 498 | */ |
176571a1 | 499 | static int rbio_is_full(struct btrfs_raid_bio *rbio) |
53b381b3 | 500 | { |
176571a1 | 501 | unsigned long flags; |
53b381b3 DW |
502 | unsigned long size = rbio->bio_list_bytes; |
503 | int ret = 1; | |
504 | ||
176571a1 | 505 | spin_lock_irqsave(&rbio->bio_list_lock, flags); |
ff18a4af | 506 | if (size != rbio->nr_data * BTRFS_STRIPE_LEN) |
53b381b3 | 507 | ret = 0; |
ff18a4af | 508 | BUG_ON(size > rbio->nr_data * BTRFS_STRIPE_LEN); |
53b381b3 | 509 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); |
176571a1 | 510 | |
53b381b3 DW |
511 | return ret; |
512 | } | |
513 | ||
514 | /* | |
515 | * returns 1 if it is safe to merge two rbios together. | |
516 | * The merging is safe if the two rbios correspond to | |
517 | * the same stripe and if they are both going in the same | |
518 | * direction (read vs write), and if neither one is | |
519 | * locked for final IO | |
520 | * | |
521 | * The caller is responsible for locking such that | |
522 | * rmw_locked is safe to test | |
523 | */ | |
524 | static int rbio_can_merge(struct btrfs_raid_bio *last, | |
525 | struct btrfs_raid_bio *cur) | |
526 | { | |
527 | if (test_bit(RBIO_RMW_LOCKED_BIT, &last->flags) || | |
528 | test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) | |
529 | return 0; | |
530 | ||
4ae10b3a CM |
531 | /* |
532 | * we can't merge with cached rbios, since the | |
533 | * idea is that when we merge the destination | |
534 | * rbio is going to run our IO for us. We can | |
01327610 | 535 | * steal from cached rbios though, other functions |
4ae10b3a CM |
536 | * handle that. |
537 | */ | |
538 | if (test_bit(RBIO_CACHE_BIT, &last->flags) || | |
539 | test_bit(RBIO_CACHE_BIT, &cur->flags)) | |
540 | return 0; | |
541 | ||
4c664611 | 542 | if (last->bioc->raid_map[0] != cur->bioc->raid_map[0]) |
53b381b3 DW |
543 | return 0; |
544 | ||
5a6ac9ea MX |
545 | /* we can't merge with different operations */ |
546 | if (last->operation != cur->operation) | |
547 | return 0; | |
548 | /* | |
549 | * We've need read the full stripe from the drive. | |
550 | * check and repair the parity and write the new results. | |
551 | * | |
552 | * We're not allowed to add any new bios to the | |
553 | * bio list here, anyone else that wants to | |
554 | * change this stripe needs to do their own rmw. | |
555 | */ | |
db34be19 | 556 | if (last->operation == BTRFS_RBIO_PARITY_SCRUB) |
53b381b3 | 557 | return 0; |
53b381b3 | 558 | |
db34be19 | 559 | if (last->operation == BTRFS_RBIO_REBUILD_MISSING) |
b4ee1782 OS |
560 | return 0; |
561 | ||
cc54ff62 LB |
562 | if (last->operation == BTRFS_RBIO_READ_REBUILD) { |
563 | int fa = last->faila; | |
564 | int fb = last->failb; | |
565 | int cur_fa = cur->faila; | |
566 | int cur_fb = cur->failb; | |
567 | ||
568 | if (last->faila >= last->failb) { | |
569 | fa = last->failb; | |
570 | fb = last->faila; | |
571 | } | |
572 | ||
573 | if (cur->faila >= cur->failb) { | |
574 | cur_fa = cur->failb; | |
575 | cur_fb = cur->faila; | |
576 | } | |
577 | ||
578 | if (fa != cur_fa || fb != cur_fb) | |
579 | return 0; | |
580 | } | |
53b381b3 DW |
581 | return 1; |
582 | } | |
583 | ||
3e77605d QW |
584 | static unsigned int rbio_stripe_sector_index(const struct btrfs_raid_bio *rbio, |
585 | unsigned int stripe_nr, | |
586 | unsigned int sector_nr) | |
587 | { | |
588 | ASSERT(stripe_nr < rbio->real_stripes); | |
589 | ASSERT(sector_nr < rbio->stripe_nsectors); | |
590 | ||
591 | return stripe_nr * rbio->stripe_nsectors + sector_nr; | |
592 | } | |
593 | ||
594 | /* Return a sector from rbio->stripe_sectors, not from the bio list */ | |
595 | static struct sector_ptr *rbio_stripe_sector(const struct btrfs_raid_bio *rbio, | |
596 | unsigned int stripe_nr, | |
597 | unsigned int sector_nr) | |
598 | { | |
599 | return &rbio->stripe_sectors[rbio_stripe_sector_index(rbio, stripe_nr, | |
600 | sector_nr)]; | |
601 | } | |
602 | ||
1145059a QW |
603 | /* Grab a sector inside P stripe */ |
604 | static struct sector_ptr *rbio_pstripe_sector(const struct btrfs_raid_bio *rbio, | |
605 | unsigned int sector_nr) | |
b7178a5f | 606 | { |
1145059a | 607 | return rbio_stripe_sector(rbio, rbio->nr_data, sector_nr); |
b7178a5f ZL |
608 | } |
609 | ||
1145059a QW |
610 | /* Grab a sector inside Q stripe, return NULL if not RAID6 */ |
611 | static struct sector_ptr *rbio_qstripe_sector(const struct btrfs_raid_bio *rbio, | |
612 | unsigned int sector_nr) | |
53b381b3 | 613 | { |
1145059a QW |
614 | if (rbio->nr_data + 1 == rbio->real_stripes) |
615 | return NULL; | |
616 | return rbio_stripe_sector(rbio, rbio->nr_data + 1, sector_nr); | |
53b381b3 DW |
617 | } |
618 | ||
53b381b3 DW |
619 | /* |
620 | * The first stripe in the table for a logical address | |
621 | * has the lock. rbios are added in one of three ways: | |
622 | * | |
623 | * 1) Nobody has the stripe locked yet. The rbio is given | |
624 | * the lock and 0 is returned. The caller must start the IO | |
625 | * themselves. | |
626 | * | |
627 | * 2) Someone has the stripe locked, but we're able to merge | |
628 | * with the lock owner. The rbio is freed and the IO will | |
629 | * start automatically along with the existing rbio. 1 is returned. | |
630 | * | |
631 | * 3) Someone has the stripe locked, but we're not able to merge. | |
632 | * The rbio is added to the lock owner's plug list, or merged into | |
633 | * an rbio already on the plug list. When the lock owner unlocks, | |
634 | * the next rbio on the list is run and the IO is started automatically. | |
635 | * 1 is returned | |
636 | * | |
637 | * If we return 0, the caller still owns the rbio and must continue with | |
638 | * IO submission. If we return 1, the caller must assume the rbio has | |
639 | * already been freed. | |
640 | */ | |
641 | static noinline int lock_stripe_add(struct btrfs_raid_bio *rbio) | |
642 | { | |
721860d5 | 643 | struct btrfs_stripe_hash *h; |
53b381b3 DW |
644 | struct btrfs_raid_bio *cur; |
645 | struct btrfs_raid_bio *pending; | |
646 | unsigned long flags; | |
53b381b3 | 647 | struct btrfs_raid_bio *freeit = NULL; |
4ae10b3a | 648 | struct btrfs_raid_bio *cache_drop = NULL; |
53b381b3 | 649 | int ret = 0; |
53b381b3 | 650 | |
6a258d72 | 651 | h = rbio->bioc->fs_info->stripe_hash_table->table + rbio_bucket(rbio); |
721860d5 | 652 | |
53b381b3 DW |
653 | spin_lock_irqsave(&h->lock, flags); |
654 | list_for_each_entry(cur, &h->hash_list, hash_list) { | |
4c664611 | 655 | if (cur->bioc->raid_map[0] != rbio->bioc->raid_map[0]) |
9d6cb1b0 | 656 | continue; |
4ae10b3a | 657 | |
9d6cb1b0 | 658 | spin_lock(&cur->bio_list_lock); |
4ae10b3a | 659 | |
9d6cb1b0 JT |
660 | /* Can we steal this cached rbio's pages? */ |
661 | if (bio_list_empty(&cur->bio_list) && | |
662 | list_empty(&cur->plug_list) && | |
663 | test_bit(RBIO_CACHE_BIT, &cur->flags) && | |
664 | !test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) { | |
665 | list_del_init(&cur->hash_list); | |
666 | refcount_dec(&cur->refs); | |
53b381b3 | 667 | |
9d6cb1b0 JT |
668 | steal_rbio(cur, rbio); |
669 | cache_drop = cur; | |
670 | spin_unlock(&cur->bio_list_lock); | |
4ae10b3a | 671 | |
9d6cb1b0 JT |
672 | goto lockit; |
673 | } | |
53b381b3 | 674 | |
9d6cb1b0 JT |
675 | /* Can we merge into the lock owner? */ |
676 | if (rbio_can_merge(cur, rbio)) { | |
677 | merge_rbio(cur, rbio); | |
53b381b3 | 678 | spin_unlock(&cur->bio_list_lock); |
9d6cb1b0 | 679 | freeit = rbio; |
53b381b3 DW |
680 | ret = 1; |
681 | goto out; | |
682 | } | |
9d6cb1b0 JT |
683 | |
684 | ||
685 | /* | |
686 | * We couldn't merge with the running rbio, see if we can merge | |
687 | * with the pending ones. We don't have to check for rmw_locked | |
688 | * because there is no way they are inside finish_rmw right now | |
689 | */ | |
690 | list_for_each_entry(pending, &cur->plug_list, plug_list) { | |
691 | if (rbio_can_merge(pending, rbio)) { | |
692 | merge_rbio(pending, rbio); | |
693 | spin_unlock(&cur->bio_list_lock); | |
694 | freeit = rbio; | |
695 | ret = 1; | |
696 | goto out; | |
697 | } | |
698 | } | |
699 | ||
700 | /* | |
701 | * No merging, put us on the tail of the plug list, our rbio | |
702 | * will be started with the currently running rbio unlocks | |
703 | */ | |
704 | list_add_tail(&rbio->plug_list, &cur->plug_list); | |
705 | spin_unlock(&cur->bio_list_lock); | |
706 | ret = 1; | |
707 | goto out; | |
53b381b3 | 708 | } |
4ae10b3a | 709 | lockit: |
dec95574 | 710 | refcount_inc(&rbio->refs); |
53b381b3 DW |
711 | list_add(&rbio->hash_list, &h->hash_list); |
712 | out: | |
713 | spin_unlock_irqrestore(&h->lock, flags); | |
4ae10b3a CM |
714 | if (cache_drop) |
715 | remove_rbio_from_cache(cache_drop); | |
53b381b3 | 716 | if (freeit) |
ff2b64a2 | 717 | free_raid_bio(freeit); |
53b381b3 DW |
718 | return ret; |
719 | } | |
720 | ||
721 | /* | |
722 | * called as rmw or parity rebuild is completed. If the plug list has more | |
723 | * rbios waiting for this stripe, the next one on the list will be started | |
724 | */ | |
725 | static noinline void unlock_stripe(struct btrfs_raid_bio *rbio) | |
726 | { | |
727 | int bucket; | |
728 | struct btrfs_stripe_hash *h; | |
729 | unsigned long flags; | |
4ae10b3a | 730 | int keep_cache = 0; |
53b381b3 DW |
731 | |
732 | bucket = rbio_bucket(rbio); | |
6a258d72 | 733 | h = rbio->bioc->fs_info->stripe_hash_table->table + bucket; |
53b381b3 | 734 | |
4ae10b3a CM |
735 | if (list_empty(&rbio->plug_list)) |
736 | cache_rbio(rbio); | |
737 | ||
53b381b3 DW |
738 | spin_lock_irqsave(&h->lock, flags); |
739 | spin_lock(&rbio->bio_list_lock); | |
740 | ||
741 | if (!list_empty(&rbio->hash_list)) { | |
4ae10b3a CM |
742 | /* |
743 | * if we're still cached and there is no other IO | |
744 | * to perform, just leave this rbio here for others | |
745 | * to steal from later | |
746 | */ | |
747 | if (list_empty(&rbio->plug_list) && | |
748 | test_bit(RBIO_CACHE_BIT, &rbio->flags)) { | |
749 | keep_cache = 1; | |
750 | clear_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
751 | BUG_ON(!bio_list_empty(&rbio->bio_list)); | |
752 | goto done; | |
753 | } | |
53b381b3 DW |
754 | |
755 | list_del_init(&rbio->hash_list); | |
dec95574 | 756 | refcount_dec(&rbio->refs); |
53b381b3 DW |
757 | |
758 | /* | |
759 | * we use the plug list to hold all the rbios | |
760 | * waiting for the chance to lock this stripe. | |
761 | * hand the lock over to one of them. | |
762 | */ | |
763 | if (!list_empty(&rbio->plug_list)) { | |
764 | struct btrfs_raid_bio *next; | |
765 | struct list_head *head = rbio->plug_list.next; | |
766 | ||
767 | next = list_entry(head, struct btrfs_raid_bio, | |
768 | plug_list); | |
769 | ||
770 | list_del_init(&rbio->plug_list); | |
771 | ||
772 | list_add(&next->hash_list, &h->hash_list); | |
dec95574 | 773 | refcount_inc(&next->refs); |
53b381b3 DW |
774 | spin_unlock(&rbio->bio_list_lock); |
775 | spin_unlock_irqrestore(&h->lock, flags); | |
776 | ||
1b94b556 | 777 | if (next->operation == BTRFS_RBIO_READ_REBUILD) |
e66d8d5a | 778 | start_async_work(next, read_rebuild_work); |
b4ee1782 OS |
779 | else if (next->operation == BTRFS_RBIO_REBUILD_MISSING) { |
780 | steal_rbio(rbio, next); | |
e66d8d5a | 781 | start_async_work(next, read_rebuild_work); |
b4ee1782 | 782 | } else if (next->operation == BTRFS_RBIO_WRITE) { |
4ae10b3a | 783 | steal_rbio(rbio, next); |
cf6a4a75 | 784 | start_async_work(next, rmw_work); |
5a6ac9ea MX |
785 | } else if (next->operation == BTRFS_RBIO_PARITY_SCRUB) { |
786 | steal_rbio(rbio, next); | |
a81b747d | 787 | start_async_work(next, scrub_parity_work); |
4ae10b3a | 788 | } |
53b381b3 DW |
789 | |
790 | goto done_nolock; | |
53b381b3 DW |
791 | } |
792 | } | |
4ae10b3a | 793 | done: |
53b381b3 DW |
794 | spin_unlock(&rbio->bio_list_lock); |
795 | spin_unlock_irqrestore(&h->lock, flags); | |
796 | ||
797 | done_nolock: | |
4ae10b3a CM |
798 | if (!keep_cache) |
799 | remove_rbio_from_cache(rbio); | |
53b381b3 DW |
800 | } |
801 | ||
7583d8d0 | 802 | static void rbio_endio_bio_list(struct bio *cur, blk_status_t err) |
53b381b3 | 803 | { |
7583d8d0 LB |
804 | struct bio *next; |
805 | ||
806 | while (cur) { | |
807 | next = cur->bi_next; | |
808 | cur->bi_next = NULL; | |
809 | cur->bi_status = err; | |
810 | bio_endio(cur); | |
811 | cur = next; | |
812 | } | |
53b381b3 DW |
813 | } |
814 | ||
815 | /* | |
816 | * this frees the rbio and runs through all the bios in the | |
817 | * bio_list and calls end_io on them | |
818 | */ | |
4e4cbee9 | 819 | static void rbio_orig_end_io(struct btrfs_raid_bio *rbio, blk_status_t err) |
53b381b3 DW |
820 | { |
821 | struct bio *cur = bio_list_get(&rbio->bio_list); | |
7583d8d0 | 822 | struct bio *extra; |
4245215d | 823 | |
bd8f7e62 QW |
824 | /* |
825 | * Clear the data bitmap, as the rbio may be cached for later usage. | |
826 | * do this before before unlock_stripe() so there will be no new bio | |
827 | * for this bio. | |
828 | */ | |
829 | bitmap_clear(&rbio->dbitmap, 0, rbio->stripe_nsectors); | |
4245215d | 830 | |
7583d8d0 LB |
831 | /* |
832 | * At this moment, rbio->bio_list is empty, however since rbio does not | |
833 | * always have RBIO_RMW_LOCKED_BIT set and rbio is still linked on the | |
834 | * hash list, rbio may be merged with others so that rbio->bio_list | |
835 | * becomes non-empty. | |
836 | * Once unlock_stripe() is done, rbio->bio_list will not be updated any | |
837 | * more and we can call bio_endio() on all queued bios. | |
838 | */ | |
839 | unlock_stripe(rbio); | |
840 | extra = bio_list_get(&rbio->bio_list); | |
ff2b64a2 | 841 | free_raid_bio(rbio); |
53b381b3 | 842 | |
7583d8d0 LB |
843 | rbio_endio_bio_list(cur, err); |
844 | if (extra) | |
845 | rbio_endio_bio_list(extra, err); | |
53b381b3 DW |
846 | } |
847 | ||
848 | /* | |
849 | * end io function used by finish_rmw. When we finally | |
850 | * get here, we've written a full stripe | |
851 | */ | |
4246a0b6 | 852 | static void raid_write_end_io(struct bio *bio) |
53b381b3 DW |
853 | { |
854 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
4e4cbee9 | 855 | blk_status_t err = bio->bi_status; |
a6111d11 | 856 | int max_errors; |
53b381b3 DW |
857 | |
858 | if (err) | |
859 | fail_bio_stripe(rbio, bio); | |
860 | ||
861 | bio_put(bio); | |
862 | ||
b89e1b01 | 863 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
864 | return; |
865 | ||
58efbc9f | 866 | err = BLK_STS_OK; |
53b381b3 DW |
867 | |
868 | /* OK, we have read all the stripes we need to. */ | |
a6111d11 | 869 | max_errors = (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) ? |
4c664611 | 870 | 0 : rbio->bioc->max_errors; |
a6111d11 | 871 | if (atomic_read(&rbio->error) > max_errors) |
4e4cbee9 | 872 | err = BLK_STS_IOERR; |
53b381b3 | 873 | |
4246a0b6 | 874 | rbio_orig_end_io(rbio, err); |
53b381b3 DW |
875 | } |
876 | ||
3e77605d QW |
877 | /** |
878 | * Get a sector pointer specified by its @stripe_nr and @sector_nr | |
879 | * | |
880 | * @rbio: The raid bio | |
881 | * @stripe_nr: Stripe number, valid range [0, real_stripe) | |
882 | * @sector_nr: Sector number inside the stripe, | |
883 | * valid range [0, stripe_nsectors) | |
884 | * @bio_list_only: Whether to use sectors inside the bio list only. | |
885 | * | |
886 | * The read/modify/write code wants to reuse the original bio page as much | |
887 | * as possible, and only use stripe_sectors as fallback. | |
888 | */ | |
889 | static struct sector_ptr *sector_in_rbio(struct btrfs_raid_bio *rbio, | |
890 | int stripe_nr, int sector_nr, | |
891 | bool bio_list_only) | |
892 | { | |
893 | struct sector_ptr *sector; | |
894 | int index; | |
895 | ||
896 | ASSERT(stripe_nr >= 0 && stripe_nr < rbio->real_stripes); | |
897 | ASSERT(sector_nr >= 0 && sector_nr < rbio->stripe_nsectors); | |
898 | ||
899 | index = stripe_nr * rbio->stripe_nsectors + sector_nr; | |
900 | ASSERT(index >= 0 && index < rbio->nr_sectors); | |
901 | ||
902 | spin_lock_irq(&rbio->bio_list_lock); | |
903 | sector = &rbio->bio_sectors[index]; | |
904 | if (sector->page || bio_list_only) { | |
905 | /* Don't return sector without a valid page pointer */ | |
906 | if (!sector->page) | |
907 | sector = NULL; | |
908 | spin_unlock_irq(&rbio->bio_list_lock); | |
909 | return sector; | |
910 | } | |
911 | spin_unlock_irq(&rbio->bio_list_lock); | |
912 | ||
913 | return &rbio->stripe_sectors[index]; | |
914 | } | |
915 | ||
53b381b3 DW |
916 | /* |
917 | * allocation and initial setup for the btrfs_raid_bio. Not | |
918 | * this does not allocate any pages for rbio->pages. | |
919 | */ | |
2ff7e61e | 920 | static struct btrfs_raid_bio *alloc_rbio(struct btrfs_fs_info *fs_info, |
ff18a4af | 921 | struct btrfs_io_context *bioc) |
53b381b3 | 922 | { |
843de58b | 923 | const unsigned int real_stripes = bioc->num_stripes - bioc->num_tgtdevs; |
ff18a4af | 924 | const unsigned int stripe_npages = BTRFS_STRIPE_LEN >> PAGE_SHIFT; |
843de58b | 925 | const unsigned int num_pages = stripe_npages * real_stripes; |
ff18a4af CH |
926 | const unsigned int stripe_nsectors = |
927 | BTRFS_STRIPE_LEN >> fs_info->sectorsize_bits; | |
94efbe19 | 928 | const unsigned int num_sectors = stripe_nsectors * real_stripes; |
53b381b3 | 929 | struct btrfs_raid_bio *rbio; |
53b381b3 | 930 | |
94efbe19 QW |
931 | /* PAGE_SIZE must also be aligned to sectorsize for subpage support */ |
932 | ASSERT(IS_ALIGNED(PAGE_SIZE, fs_info->sectorsize)); | |
c67c68eb QW |
933 | /* |
934 | * Our current stripe len should be fixed to 64k thus stripe_nsectors | |
935 | * (at most 16) should be no larger than BITS_PER_LONG. | |
936 | */ | |
937 | ASSERT(stripe_nsectors <= BITS_PER_LONG); | |
843de58b | 938 | |
797d74b7 | 939 | rbio = kzalloc(sizeof(*rbio), GFP_NOFS); |
af8e2d1d | 940 | if (!rbio) |
53b381b3 | 941 | return ERR_PTR(-ENOMEM); |
797d74b7 QW |
942 | rbio->stripe_pages = kcalloc(num_pages, sizeof(struct page *), |
943 | GFP_NOFS); | |
944 | rbio->bio_sectors = kcalloc(num_sectors, sizeof(struct sector_ptr), | |
945 | GFP_NOFS); | |
946 | rbio->stripe_sectors = kcalloc(num_sectors, sizeof(struct sector_ptr), | |
947 | GFP_NOFS); | |
948 | rbio->finish_pointers = kcalloc(real_stripes, sizeof(void *), GFP_NOFS); | |
949 | ||
950 | if (!rbio->stripe_pages || !rbio->bio_sectors || !rbio->stripe_sectors || | |
951 | !rbio->finish_pointers) { | |
952 | free_raid_bio_pointers(rbio); | |
953 | kfree(rbio); | |
954 | return ERR_PTR(-ENOMEM); | |
955 | } | |
53b381b3 DW |
956 | |
957 | bio_list_init(&rbio->bio_list); | |
958 | INIT_LIST_HEAD(&rbio->plug_list); | |
959 | spin_lock_init(&rbio->bio_list_lock); | |
4ae10b3a | 960 | INIT_LIST_HEAD(&rbio->stripe_cache); |
53b381b3 | 961 | INIT_LIST_HEAD(&rbio->hash_list); |
f1c29379 | 962 | btrfs_get_bioc(bioc); |
4c664611 | 963 | rbio->bioc = bioc; |
53b381b3 | 964 | rbio->nr_pages = num_pages; |
94efbe19 | 965 | rbio->nr_sectors = num_sectors; |
2c8cdd6e | 966 | rbio->real_stripes = real_stripes; |
5a6ac9ea | 967 | rbio->stripe_npages = stripe_npages; |
94efbe19 | 968 | rbio->stripe_nsectors = stripe_nsectors; |
53b381b3 DW |
969 | rbio->faila = -1; |
970 | rbio->failb = -1; | |
dec95574 | 971 | refcount_set(&rbio->refs, 1); |
b89e1b01 MX |
972 | atomic_set(&rbio->error, 0); |
973 | atomic_set(&rbio->stripes_pending, 0); | |
53b381b3 | 974 | |
0b30f719 QW |
975 | ASSERT(btrfs_nr_parity_stripes(bioc->map_type)); |
976 | rbio->nr_data = real_stripes - btrfs_nr_parity_stripes(bioc->map_type); | |
53b381b3 | 977 | |
53b381b3 DW |
978 | return rbio; |
979 | } | |
980 | ||
981 | /* allocate pages for all the stripes in the bio, including parity */ | |
982 | static int alloc_rbio_pages(struct btrfs_raid_bio *rbio) | |
983 | { | |
eb357060 QW |
984 | int ret; |
985 | ||
986 | ret = btrfs_alloc_page_array(rbio->nr_pages, rbio->stripe_pages); | |
987 | if (ret < 0) | |
988 | return ret; | |
989 | /* Mapping all sectors */ | |
990 | index_stripe_sectors(rbio); | |
991 | return 0; | |
53b381b3 DW |
992 | } |
993 | ||
b7178a5f | 994 | /* only allocate pages for p/q stripes */ |
53b381b3 DW |
995 | static int alloc_rbio_parity_pages(struct btrfs_raid_bio *rbio) |
996 | { | |
f77183dc | 997 | const int data_pages = rbio->nr_data * rbio->stripe_npages; |
eb357060 | 998 | int ret; |
53b381b3 | 999 | |
eb357060 QW |
1000 | ret = btrfs_alloc_page_array(rbio->nr_pages - data_pages, |
1001 | rbio->stripe_pages + data_pages); | |
1002 | if (ret < 0) | |
1003 | return ret; | |
1004 | ||
1005 | index_stripe_sectors(rbio); | |
1006 | return 0; | |
53b381b3 DW |
1007 | } |
1008 | ||
1009 | /* | |
3e77605d QW |
1010 | * Add a single sector @sector into our list of bios for IO. |
1011 | * | |
1012 | * Return 0 if everything went well. | |
1013 | * Return <0 for error. | |
53b381b3 | 1014 | */ |
3e77605d QW |
1015 | static int rbio_add_io_sector(struct btrfs_raid_bio *rbio, |
1016 | struct bio_list *bio_list, | |
1017 | struct sector_ptr *sector, | |
1018 | unsigned int stripe_nr, | |
1019 | unsigned int sector_nr, | |
bf9486d6 | 1020 | enum req_op op) |
53b381b3 | 1021 | { |
3e77605d | 1022 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; |
53b381b3 | 1023 | struct bio *last = bio_list->tail; |
53b381b3 DW |
1024 | int ret; |
1025 | struct bio *bio; | |
4c664611 | 1026 | struct btrfs_io_stripe *stripe; |
53b381b3 DW |
1027 | u64 disk_start; |
1028 | ||
3e77605d QW |
1029 | /* |
1030 | * Note: here stripe_nr has taken device replace into consideration, | |
1031 | * thus it can be larger than rbio->real_stripe. | |
1032 | * So here we check against bioc->num_stripes, not rbio->real_stripes. | |
1033 | */ | |
1034 | ASSERT(stripe_nr >= 0 && stripe_nr < rbio->bioc->num_stripes); | |
1035 | ASSERT(sector_nr >= 0 && sector_nr < rbio->stripe_nsectors); | |
1036 | ASSERT(sector->page); | |
1037 | ||
4c664611 | 1038 | stripe = &rbio->bioc->stripes[stripe_nr]; |
3e77605d | 1039 | disk_start = stripe->physical + sector_nr * sectorsize; |
53b381b3 DW |
1040 | |
1041 | /* if the device is missing, just fail this stripe */ | |
1042 | if (!stripe->dev->bdev) | |
1043 | return fail_rbio_index(rbio, stripe_nr); | |
1044 | ||
1045 | /* see if we can add this page onto our existing bio */ | |
1046 | if (last) { | |
1201b58b | 1047 | u64 last_end = last->bi_iter.bi_sector << 9; |
4f024f37 | 1048 | last_end += last->bi_iter.bi_size; |
53b381b3 DW |
1049 | |
1050 | /* | |
1051 | * we can't merge these if they are from different | |
1052 | * devices or if they are not contiguous | |
1053 | */ | |
f90ae76a | 1054 | if (last_end == disk_start && !last->bi_status && |
309dca30 | 1055 | last->bi_bdev == stripe->dev->bdev) { |
3e77605d QW |
1056 | ret = bio_add_page(last, sector->page, sectorsize, |
1057 | sector->pgoff); | |
1058 | if (ret == sectorsize) | |
53b381b3 DW |
1059 | return 0; |
1060 | } | |
1061 | } | |
1062 | ||
1063 | /* put a new bio on the list */ | |
ff18a4af CH |
1064 | bio = bio_alloc(stripe->dev->bdev, |
1065 | max(BTRFS_STRIPE_LEN >> PAGE_SHIFT, 1), | |
bf9486d6 | 1066 | op, GFP_NOFS); |
4f024f37 | 1067 | bio->bi_iter.bi_sector = disk_start >> 9; |
e01bf588 | 1068 | bio->bi_private = rbio; |
53b381b3 | 1069 | |
3e77605d | 1070 | bio_add_page(bio, sector->page, sectorsize, sector->pgoff); |
53b381b3 DW |
1071 | bio_list_add(bio_list, bio); |
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * while we're doing the read/modify/write cycle, we could | |
1077 | * have errors in reading pages off the disk. This checks | |
1078 | * for errors and if we're not able to read the page it'll | |
1079 | * trigger parity reconstruction. The rmw will be finished | |
1080 | * after we've reconstructed the failed stripes | |
1081 | */ | |
1082 | static void validate_rbio_for_rmw(struct btrfs_raid_bio *rbio) | |
1083 | { | |
1084 | if (rbio->faila >= 0 || rbio->failb >= 0) { | |
2c8cdd6e | 1085 | BUG_ON(rbio->faila == rbio->real_stripes - 1); |
53b381b3 DW |
1086 | __raid56_parity_recover(rbio); |
1087 | } else { | |
1088 | finish_rmw(rbio); | |
1089 | } | |
1090 | } | |
1091 | ||
00425dd9 QW |
1092 | static void index_one_bio(struct btrfs_raid_bio *rbio, struct bio *bio) |
1093 | { | |
1094 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; | |
1095 | struct bio_vec bvec; | |
1096 | struct bvec_iter iter; | |
1097 | u32 offset = (bio->bi_iter.bi_sector << SECTOR_SHIFT) - | |
1098 | rbio->bioc->raid_map[0]; | |
1099 | ||
00425dd9 QW |
1100 | bio_for_each_segment(bvec, bio, iter) { |
1101 | u32 bvec_offset; | |
1102 | ||
1103 | for (bvec_offset = 0; bvec_offset < bvec.bv_len; | |
1104 | bvec_offset += sectorsize, offset += sectorsize) { | |
1105 | int index = offset / sectorsize; | |
1106 | struct sector_ptr *sector = &rbio->bio_sectors[index]; | |
1107 | ||
1108 | sector->page = bvec.bv_page; | |
1109 | sector->pgoff = bvec.bv_offset + bvec_offset; | |
1110 | ASSERT(sector->pgoff < PAGE_SIZE); | |
1111 | } | |
1112 | } | |
1113 | } | |
1114 | ||
53b381b3 DW |
1115 | /* |
1116 | * helper function to walk our bio list and populate the bio_pages array with | |
1117 | * the result. This seems expensive, but it is faster than constantly | |
1118 | * searching through the bio list as we setup the IO in finish_rmw or stripe | |
1119 | * reconstruction. | |
1120 | * | |
1121 | * This must be called before you trust the answers from page_in_rbio | |
1122 | */ | |
1123 | static void index_rbio_pages(struct btrfs_raid_bio *rbio) | |
1124 | { | |
1125 | struct bio *bio; | |
53b381b3 DW |
1126 | |
1127 | spin_lock_irq(&rbio->bio_list_lock); | |
00425dd9 QW |
1128 | bio_list_for_each(bio, &rbio->bio_list) |
1129 | index_one_bio(rbio, bio); | |
1130 | ||
53b381b3 DW |
1131 | spin_unlock_irq(&rbio->bio_list_lock); |
1132 | } | |
1133 | ||
b8bea09a QW |
1134 | static void bio_get_trace_info(struct btrfs_raid_bio *rbio, struct bio *bio, |
1135 | struct raid56_bio_trace_info *trace_info) | |
1136 | { | |
1137 | const struct btrfs_io_context *bioc = rbio->bioc; | |
1138 | int i; | |
1139 | ||
1140 | ASSERT(bioc); | |
1141 | ||
1142 | /* We rely on bio->bi_bdev to find the stripe number. */ | |
1143 | if (!bio->bi_bdev) | |
1144 | goto not_found; | |
1145 | ||
1146 | for (i = 0; i < bioc->num_stripes; i++) { | |
1147 | if (bio->bi_bdev != bioc->stripes[i].dev->bdev) | |
1148 | continue; | |
1149 | trace_info->stripe_nr = i; | |
1150 | trace_info->devid = bioc->stripes[i].dev->devid; | |
1151 | trace_info->offset = (bio->bi_iter.bi_sector << SECTOR_SHIFT) - | |
1152 | bioc->stripes[i].physical; | |
1153 | return; | |
1154 | } | |
1155 | ||
1156 | not_found: | |
1157 | trace_info->devid = -1; | |
1158 | trace_info->offset = -1; | |
1159 | trace_info->stripe_nr = -1; | |
1160 | } | |
1161 | ||
53b381b3 DW |
1162 | /* |
1163 | * this is called from one of two situations. We either | |
1164 | * have a full stripe from the higher layers, or we've read all | |
1165 | * the missing bits off disk. | |
1166 | * | |
1167 | * This will calculate the parity and then send down any | |
1168 | * changed blocks. | |
1169 | */ | |
1170 | static noinline void finish_rmw(struct btrfs_raid_bio *rbio) | |
1171 | { | |
4c664611 | 1172 | struct btrfs_io_context *bioc = rbio->bioc; |
1145059a | 1173 | const u32 sectorsize = bioc->fs_info->sectorsize; |
1389053e | 1174 | void **pointers = rbio->finish_pointers; |
53b381b3 | 1175 | int nr_data = rbio->nr_data; |
36920044 QW |
1176 | /* The total sector number inside the full stripe. */ |
1177 | int total_sector_nr; | |
53b381b3 | 1178 | int stripe; |
36920044 | 1179 | /* Sector number inside a stripe. */ |
3e77605d | 1180 | int sectornr; |
c17af965 | 1181 | bool has_qstripe; |
53b381b3 DW |
1182 | struct bio_list bio_list; |
1183 | struct bio *bio; | |
53b381b3 DW |
1184 | int ret; |
1185 | ||
1186 | bio_list_init(&bio_list); | |
1187 | ||
c17af965 DS |
1188 | if (rbio->real_stripes - rbio->nr_data == 1) |
1189 | has_qstripe = false; | |
1190 | else if (rbio->real_stripes - rbio->nr_data == 2) | |
1191 | has_qstripe = true; | |
1192 | else | |
53b381b3 | 1193 | BUG(); |
53b381b3 | 1194 | |
bd8f7e62 QW |
1195 | /* We should have at least one data sector. */ |
1196 | ASSERT(bitmap_weight(&rbio->dbitmap, rbio->stripe_nsectors)); | |
1197 | ||
53b381b3 DW |
1198 | /* at this point we either have a full stripe, |
1199 | * or we've read the full stripe from the drive. | |
1200 | * recalculate the parity and write the new results. | |
1201 | * | |
1202 | * We're not allowed to add any new bios to the | |
1203 | * bio list here, anyone else that wants to | |
1204 | * change this stripe needs to do their own rmw. | |
1205 | */ | |
1206 | spin_lock_irq(&rbio->bio_list_lock); | |
1207 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
1208 | spin_unlock_irq(&rbio->bio_list_lock); | |
1209 | ||
b89e1b01 | 1210 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
1211 | |
1212 | /* | |
1213 | * now that we've set rmw_locked, run through the | |
1214 | * bio list one last time and map the page pointers | |
4ae10b3a CM |
1215 | * |
1216 | * We don't cache full rbios because we're assuming | |
1217 | * the higher layers are unlikely to use this area of | |
1218 | * the disk again soon. If they do use it again, | |
1219 | * hopefully they will send another full bio. | |
53b381b3 DW |
1220 | */ |
1221 | index_rbio_pages(rbio); | |
4ae10b3a CM |
1222 | if (!rbio_is_full(rbio)) |
1223 | cache_rbio_pages(rbio); | |
1224 | else | |
1225 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
53b381b3 | 1226 | |
3e77605d | 1227 | for (sectornr = 0; sectornr < rbio->stripe_nsectors; sectornr++) { |
1145059a QW |
1228 | struct sector_ptr *sector; |
1229 | ||
1230 | /* First collect one sector from each data stripe */ | |
53b381b3 | 1231 | for (stripe = 0; stripe < nr_data; stripe++) { |
1145059a QW |
1232 | sector = sector_in_rbio(rbio, stripe, sectornr, 0); |
1233 | pointers[stripe] = kmap_local_page(sector->page) + | |
1234 | sector->pgoff; | |
53b381b3 DW |
1235 | } |
1236 | ||
1145059a QW |
1237 | /* Then add the parity stripe */ |
1238 | sector = rbio_pstripe_sector(rbio, sectornr); | |
1239 | sector->uptodate = 1; | |
1240 | pointers[stripe++] = kmap_local_page(sector->page) + sector->pgoff; | |
53b381b3 | 1241 | |
c17af965 | 1242 | if (has_qstripe) { |
53b381b3 | 1243 | /* |
1145059a QW |
1244 | * RAID6, add the qstripe and call the library function |
1245 | * to fill in our p/q | |
53b381b3 | 1246 | */ |
1145059a QW |
1247 | sector = rbio_qstripe_sector(rbio, sectornr); |
1248 | sector->uptodate = 1; | |
1249 | pointers[stripe++] = kmap_local_page(sector->page) + | |
1250 | sector->pgoff; | |
53b381b3 | 1251 | |
1145059a | 1252 | raid6_call.gen_syndrome(rbio->real_stripes, sectorsize, |
53b381b3 DW |
1253 | pointers); |
1254 | } else { | |
1255 | /* raid5 */ | |
1145059a QW |
1256 | memcpy(pointers[nr_data], pointers[0], sectorsize); |
1257 | run_xor(pointers + 1, nr_data - 1, sectorsize); | |
53b381b3 | 1258 | } |
94a0b58d IW |
1259 | for (stripe = stripe - 1; stripe >= 0; stripe--) |
1260 | kunmap_local(pointers[stripe]); | |
53b381b3 DW |
1261 | } |
1262 | ||
1263 | /* | |
36920044 QW |
1264 | * Start writing. Make bios for everything from the higher layers (the |
1265 | * bio_list in our rbio) and our P/Q. Ignore everything else. | |
53b381b3 | 1266 | */ |
36920044 QW |
1267 | for (total_sector_nr = 0; total_sector_nr < rbio->nr_sectors; |
1268 | total_sector_nr++) { | |
1269 | struct sector_ptr *sector; | |
3e77605d | 1270 | |
36920044 QW |
1271 | stripe = total_sector_nr / rbio->stripe_nsectors; |
1272 | sectornr = total_sector_nr % rbio->stripe_nsectors; | |
53b381b3 | 1273 | |
36920044 QW |
1274 | /* This vertical stripe has no data, skip it. */ |
1275 | if (!test_bit(sectornr, &rbio->dbitmap)) | |
1276 | continue; | |
53b381b3 | 1277 | |
36920044 QW |
1278 | if (stripe < rbio->nr_data) { |
1279 | sector = sector_in_rbio(rbio, stripe, sectornr, 1); | |
1280 | if (!sector) | |
1281 | continue; | |
1282 | } else { | |
1283 | sector = rbio_stripe_sector(rbio, stripe, sectornr); | |
53b381b3 | 1284 | } |
36920044 QW |
1285 | |
1286 | ret = rbio_add_io_sector(rbio, &bio_list, sector, stripe, | |
ff18a4af | 1287 | sectornr, REQ_OP_WRITE); |
36920044 QW |
1288 | if (ret) |
1289 | goto cleanup; | |
53b381b3 DW |
1290 | } |
1291 | ||
4c664611 | 1292 | if (likely(!bioc->num_tgtdevs)) |
2c8cdd6e MX |
1293 | goto write_data; |
1294 | ||
36920044 QW |
1295 | for (total_sector_nr = 0; total_sector_nr < rbio->nr_sectors; |
1296 | total_sector_nr++) { | |
1297 | struct sector_ptr *sector; | |
2c8cdd6e | 1298 | |
36920044 QW |
1299 | stripe = total_sector_nr / rbio->stripe_nsectors; |
1300 | sectornr = total_sector_nr % rbio->stripe_nsectors; | |
3e77605d | 1301 | |
36920044 QW |
1302 | if (!bioc->tgtdev_map[stripe]) { |
1303 | /* | |
1304 | * We can skip the whole stripe completely, note | |
1305 | * total_sector_nr will be increased by one anyway. | |
1306 | */ | |
1307 | ASSERT(sectornr == 0); | |
1308 | total_sector_nr += rbio->stripe_nsectors - 1; | |
1309 | continue; | |
1310 | } | |
2c8cdd6e | 1311 | |
36920044 QW |
1312 | /* This vertical stripe has no data, skip it. */ |
1313 | if (!test_bit(sectornr, &rbio->dbitmap)) | |
1314 | continue; | |
2c8cdd6e | 1315 | |
36920044 QW |
1316 | if (stripe < rbio->nr_data) { |
1317 | sector = sector_in_rbio(rbio, stripe, sectornr, 1); | |
1318 | if (!sector) | |
1319 | continue; | |
1320 | } else { | |
1321 | sector = rbio_stripe_sector(rbio, stripe, sectornr); | |
2c8cdd6e | 1322 | } |
36920044 QW |
1323 | |
1324 | ret = rbio_add_io_sector(rbio, &bio_list, sector, | |
1325 | rbio->bioc->tgtdev_map[stripe], | |
ff18a4af | 1326 | sectornr, REQ_OP_WRITE); |
36920044 QW |
1327 | if (ret) |
1328 | goto cleanup; | |
2c8cdd6e MX |
1329 | } |
1330 | ||
1331 | write_data: | |
b89e1b01 MX |
1332 | atomic_set(&rbio->stripes_pending, bio_list_size(&bio_list)); |
1333 | BUG_ON(atomic_read(&rbio->stripes_pending) == 0); | |
53b381b3 | 1334 | |
bf28a605 | 1335 | while ((bio = bio_list_pop(&bio_list))) { |
53b381b3 | 1336 | bio->bi_end_io = raid_write_end_io; |
4e49ea4a | 1337 | |
b8bea09a QW |
1338 | if (trace_raid56_write_stripe_enabled()) { |
1339 | struct raid56_bio_trace_info trace_info = { 0 }; | |
1340 | ||
1341 | bio_get_trace_info(rbio, bio, &trace_info); | |
1342 | trace_raid56_write_stripe(rbio, bio, &trace_info); | |
1343 | } | |
4e49ea4a | 1344 | submit_bio(bio); |
53b381b3 DW |
1345 | } |
1346 | return; | |
1347 | ||
1348 | cleanup: | |
58efbc9f | 1349 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
1350 | |
1351 | while ((bio = bio_list_pop(&bio_list))) | |
1352 | bio_put(bio); | |
53b381b3 DW |
1353 | } |
1354 | ||
1355 | /* | |
1356 | * helper to find the stripe number for a given bio. Used to figure out which | |
1357 | * stripe has failed. This expects the bio to correspond to a physical disk, | |
1358 | * so it looks up based on physical sector numbers. | |
1359 | */ | |
1360 | static int find_bio_stripe(struct btrfs_raid_bio *rbio, | |
1361 | struct bio *bio) | |
1362 | { | |
4f024f37 | 1363 | u64 physical = bio->bi_iter.bi_sector; |
53b381b3 | 1364 | int i; |
4c664611 | 1365 | struct btrfs_io_stripe *stripe; |
53b381b3 DW |
1366 | |
1367 | physical <<= 9; | |
1368 | ||
4c664611 QW |
1369 | for (i = 0; i < rbio->bioc->num_stripes; i++) { |
1370 | stripe = &rbio->bioc->stripes[i]; | |
ff18a4af | 1371 | if (in_range(physical, stripe->physical, BTRFS_STRIPE_LEN) && |
309dca30 | 1372 | stripe->dev->bdev && bio->bi_bdev == stripe->dev->bdev) { |
53b381b3 DW |
1373 | return i; |
1374 | } | |
1375 | } | |
1376 | return -1; | |
1377 | } | |
1378 | ||
1379 | /* | |
1380 | * helper to find the stripe number for a given | |
1381 | * bio (before mapping). Used to figure out which stripe has | |
1382 | * failed. This looks up based on logical block numbers. | |
1383 | */ | |
1384 | static int find_logical_bio_stripe(struct btrfs_raid_bio *rbio, | |
1385 | struct bio *bio) | |
1386 | { | |
1201b58b | 1387 | u64 logical = bio->bi_iter.bi_sector << 9; |
53b381b3 DW |
1388 | int i; |
1389 | ||
53b381b3 | 1390 | for (i = 0; i < rbio->nr_data; i++) { |
4c664611 | 1391 | u64 stripe_start = rbio->bioc->raid_map[i]; |
83025863 | 1392 | |
ff18a4af | 1393 | if (in_range(logical, stripe_start, BTRFS_STRIPE_LEN)) |
53b381b3 | 1394 | return i; |
53b381b3 DW |
1395 | } |
1396 | return -1; | |
1397 | } | |
1398 | ||
1399 | /* | |
1400 | * returns -EIO if we had too many failures | |
1401 | */ | |
1402 | static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed) | |
1403 | { | |
1404 | unsigned long flags; | |
1405 | int ret = 0; | |
1406 | ||
1407 | spin_lock_irqsave(&rbio->bio_list_lock, flags); | |
1408 | ||
1409 | /* we already know this stripe is bad, move on */ | |
1410 | if (rbio->faila == failed || rbio->failb == failed) | |
1411 | goto out; | |
1412 | ||
1413 | if (rbio->faila == -1) { | |
1414 | /* first failure on this rbio */ | |
1415 | rbio->faila = failed; | |
b89e1b01 | 1416 | atomic_inc(&rbio->error); |
53b381b3 DW |
1417 | } else if (rbio->failb == -1) { |
1418 | /* second failure on this rbio */ | |
1419 | rbio->failb = failed; | |
b89e1b01 | 1420 | atomic_inc(&rbio->error); |
53b381b3 DW |
1421 | } else { |
1422 | ret = -EIO; | |
1423 | } | |
1424 | out: | |
1425 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); | |
1426 | ||
1427 | return ret; | |
1428 | } | |
1429 | ||
1430 | /* | |
1431 | * helper to fail a stripe based on a physical disk | |
1432 | * bio. | |
1433 | */ | |
1434 | static int fail_bio_stripe(struct btrfs_raid_bio *rbio, | |
1435 | struct bio *bio) | |
1436 | { | |
1437 | int failed = find_bio_stripe(rbio, bio); | |
1438 | ||
1439 | if (failed < 0) | |
1440 | return -EIO; | |
1441 | ||
1442 | return fail_rbio_index(rbio, failed); | |
1443 | } | |
1444 | ||
5fdb7afc QW |
1445 | /* |
1446 | * For subpage case, we can no longer set page Uptodate directly for | |
1447 | * stripe_pages[], thus we need to locate the sector. | |
1448 | */ | |
1449 | static struct sector_ptr *find_stripe_sector(struct btrfs_raid_bio *rbio, | |
1450 | struct page *page, | |
1451 | unsigned int pgoff) | |
1452 | { | |
1453 | int i; | |
1454 | ||
1455 | for (i = 0; i < rbio->nr_sectors; i++) { | |
1456 | struct sector_ptr *sector = &rbio->stripe_sectors[i]; | |
1457 | ||
1458 | if (sector->page == page && sector->pgoff == pgoff) | |
1459 | return sector; | |
1460 | } | |
1461 | return NULL; | |
1462 | } | |
1463 | ||
53b381b3 DW |
1464 | /* |
1465 | * this sets each page in the bio uptodate. It should only be used on private | |
1466 | * rbio pages, nothing that comes in from the higher layers | |
1467 | */ | |
5fdb7afc | 1468 | static void set_bio_pages_uptodate(struct btrfs_raid_bio *rbio, struct bio *bio) |
53b381b3 | 1469 | { |
5fdb7afc | 1470 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; |
0198e5b7 | 1471 | struct bio_vec *bvec; |
6dc4f100 | 1472 | struct bvec_iter_all iter_all; |
6592e58c | 1473 | |
0198e5b7 | 1474 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
53b381b3 | 1475 | |
5fdb7afc QW |
1476 | bio_for_each_segment_all(bvec, bio, iter_all) { |
1477 | struct sector_ptr *sector; | |
1478 | int pgoff; | |
1479 | ||
1480 | for (pgoff = bvec->bv_offset; pgoff - bvec->bv_offset < bvec->bv_len; | |
1481 | pgoff += sectorsize) { | |
1482 | sector = find_stripe_sector(rbio, bvec->bv_page, pgoff); | |
1483 | ASSERT(sector); | |
1484 | if (sector) | |
1485 | sector->uptodate = 1; | |
1486 | } | |
1487 | } | |
53b381b3 DW |
1488 | } |
1489 | ||
d34e123d | 1490 | static void raid56_bio_end_io(struct bio *bio) |
53b381b3 DW |
1491 | { |
1492 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
1493 | ||
4e4cbee9 | 1494 | if (bio->bi_status) |
53b381b3 DW |
1495 | fail_bio_stripe(rbio, bio); |
1496 | else | |
5fdb7afc | 1497 | set_bio_pages_uptodate(rbio, bio); |
53b381b3 DW |
1498 | |
1499 | bio_put(bio); | |
1500 | ||
d34e123d CH |
1501 | if (atomic_dec_and_test(&rbio->stripes_pending)) |
1502 | queue_work(rbio->bioc->fs_info->endio_raid56_workers, | |
1503 | &rbio->end_io_work); | |
1504 | } | |
53b381b3 | 1505 | |
d34e123d CH |
1506 | /* |
1507 | * End io handler for the read phase of the RMW cycle. All the bios here are | |
1508 | * physical stripe bios we've read from the disk so we can recalculate the | |
1509 | * parity of the stripe. | |
1510 | * | |
1511 | * This will usually kick off finish_rmw once all the bios are read in, but it | |
1512 | * may trigger parity reconstruction if we had any errors along the way | |
1513 | */ | |
1514 | static void raid56_rmw_end_io_work(struct work_struct *work) | |
1515 | { | |
1516 | struct btrfs_raid_bio *rbio = | |
1517 | container_of(work, struct btrfs_raid_bio, end_io_work); | |
1518 | ||
1519 | if (atomic_read(&rbio->error) > rbio->bioc->max_errors) { | |
1520 | rbio_orig_end_io(rbio, BLK_STS_IOERR); | |
1521 | return; | |
1522 | } | |
53b381b3 DW |
1523 | |
1524 | /* | |
d34e123d CH |
1525 | * This will normally call finish_rmw to start our write but if there |
1526 | * are any failed stripes we'll reconstruct from parity first. | |
53b381b3 DW |
1527 | */ |
1528 | validate_rbio_for_rmw(rbio); | |
53b381b3 DW |
1529 | } |
1530 | ||
53b381b3 DW |
1531 | /* |
1532 | * the stripe must be locked by the caller. It will | |
1533 | * unlock after all the writes are done | |
1534 | */ | |
1535 | static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio) | |
1536 | { | |
1537 | int bios_to_read = 0; | |
53b381b3 | 1538 | struct bio_list bio_list; |
550cdeb3 | 1539 | const int nr_data_sectors = rbio->stripe_nsectors * rbio->nr_data; |
53b381b3 | 1540 | int ret; |
550cdeb3 | 1541 | int total_sector_nr; |
53b381b3 DW |
1542 | struct bio *bio; |
1543 | ||
1544 | bio_list_init(&bio_list); | |
1545 | ||
1546 | ret = alloc_rbio_pages(rbio); | |
1547 | if (ret) | |
1548 | goto cleanup; | |
1549 | ||
1550 | index_rbio_pages(rbio); | |
1551 | ||
b89e1b01 | 1552 | atomic_set(&rbio->error, 0); |
550cdeb3 QW |
1553 | /* Build a list of bios to read all the missing data sectors. */ |
1554 | for (total_sector_nr = 0; total_sector_nr < nr_data_sectors; | |
1555 | total_sector_nr++) { | |
1556 | struct sector_ptr *sector; | |
1557 | int stripe = total_sector_nr / rbio->stripe_nsectors; | |
1558 | int sectornr = total_sector_nr % rbio->stripe_nsectors; | |
3e77605d | 1559 | |
550cdeb3 QW |
1560 | /* |
1561 | * We want to find all the sectors missing from the rbio and | |
1562 | * read them from the disk. If sector_in_rbio() finds a page | |
1563 | * in the bio list we don't need to read it off the stripe. | |
1564 | */ | |
1565 | sector = sector_in_rbio(rbio, stripe, sectornr, 1); | |
1566 | if (sector) | |
1567 | continue; | |
53b381b3 | 1568 | |
550cdeb3 QW |
1569 | sector = rbio_stripe_sector(rbio, stripe, sectornr); |
1570 | /* | |
1571 | * The bio cache may have handed us an uptodate page. If so, | |
1572 | * use it. | |
1573 | */ | |
1574 | if (sector->uptodate) | |
1575 | continue; | |
4ae10b3a | 1576 | |
550cdeb3 | 1577 | ret = rbio_add_io_sector(rbio, &bio_list, sector, |
ff18a4af | 1578 | stripe, sectornr, REQ_OP_READ); |
550cdeb3 QW |
1579 | if (ret) |
1580 | goto cleanup; | |
53b381b3 DW |
1581 | } |
1582 | ||
1583 | bios_to_read = bio_list_size(&bio_list); | |
1584 | if (!bios_to_read) { | |
1585 | /* | |
1586 | * this can happen if others have merged with | |
1587 | * us, it means there is nothing left to read. | |
1588 | * But if there are missing devices it may not be | |
1589 | * safe to do the full stripe write yet. | |
1590 | */ | |
1591 | goto finish; | |
1592 | } | |
1593 | ||
1594 | /* | |
4c664611 QW |
1595 | * The bioc may be freed once we submit the last bio. Make sure not to |
1596 | * touch it after that. | |
53b381b3 | 1597 | */ |
b89e1b01 | 1598 | atomic_set(&rbio->stripes_pending, bios_to_read); |
d34e123d | 1599 | INIT_WORK(&rbio->end_io_work, raid56_rmw_end_io_work); |
bf28a605 | 1600 | while ((bio = bio_list_pop(&bio_list))) { |
d34e123d | 1601 | bio->bi_end_io = raid56_bio_end_io; |
53b381b3 | 1602 | |
b8bea09a QW |
1603 | if (trace_raid56_read_partial_enabled()) { |
1604 | struct raid56_bio_trace_info trace_info = { 0 }; | |
53b381b3 | 1605 | |
b8bea09a QW |
1606 | bio_get_trace_info(rbio, bio, &trace_info); |
1607 | trace_raid56_read_partial(rbio, bio, &trace_info); | |
1608 | } | |
4e49ea4a | 1609 | submit_bio(bio); |
53b381b3 DW |
1610 | } |
1611 | /* the actual write will happen once the reads are done */ | |
1612 | return 0; | |
1613 | ||
1614 | cleanup: | |
58efbc9f | 1615 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
1616 | |
1617 | while ((bio = bio_list_pop(&bio_list))) | |
1618 | bio_put(bio); | |
1619 | ||
53b381b3 DW |
1620 | return -EIO; |
1621 | ||
1622 | finish: | |
1623 | validate_rbio_for_rmw(rbio); | |
1624 | return 0; | |
1625 | } | |
1626 | ||
1627 | /* | |
1628 | * if the upper layers pass in a full stripe, we thank them by only allocating | |
1629 | * enough pages to hold the parity, and sending it all down quickly. | |
1630 | */ | |
1631 | static int full_stripe_write(struct btrfs_raid_bio *rbio) | |
1632 | { | |
1633 | int ret; | |
1634 | ||
1635 | ret = alloc_rbio_parity_pages(rbio); | |
ab4c54c6 | 1636 | if (ret) |
53b381b3 DW |
1637 | return ret; |
1638 | ||
1639 | ret = lock_stripe_add(rbio); | |
1640 | if (ret == 0) | |
1641 | finish_rmw(rbio); | |
1642 | return 0; | |
1643 | } | |
1644 | ||
1645 | /* | |
1646 | * partial stripe writes get handed over to async helpers. | |
1647 | * We're really hoping to merge a few more writes into this | |
1648 | * rbio before calculating new parity | |
1649 | */ | |
1650 | static int partial_stripe_write(struct btrfs_raid_bio *rbio) | |
1651 | { | |
1652 | int ret; | |
1653 | ||
1654 | ret = lock_stripe_add(rbio); | |
1655 | if (ret == 0) | |
cf6a4a75 | 1656 | start_async_work(rbio, rmw_work); |
53b381b3 DW |
1657 | return 0; |
1658 | } | |
1659 | ||
1660 | /* | |
1661 | * sometimes while we were reading from the drive to | |
1662 | * recalculate parity, enough new bios come into create | |
1663 | * a full stripe. So we do a check here to see if we can | |
1664 | * go directly to finish_rmw | |
1665 | */ | |
1666 | static int __raid56_parity_write(struct btrfs_raid_bio *rbio) | |
1667 | { | |
1668 | /* head off into rmw land if we don't have a full stripe */ | |
1669 | if (!rbio_is_full(rbio)) | |
1670 | return partial_stripe_write(rbio); | |
1671 | return full_stripe_write(rbio); | |
1672 | } | |
1673 | ||
6ac0f488 CM |
1674 | /* |
1675 | * We use plugging call backs to collect full stripes. | |
1676 | * Any time we get a partial stripe write while plugged | |
1677 | * we collect it into a list. When the unplug comes down, | |
1678 | * we sort the list by logical block number and merge | |
1679 | * everything we can into the same rbios | |
1680 | */ | |
1681 | struct btrfs_plug_cb { | |
1682 | struct blk_plug_cb cb; | |
1683 | struct btrfs_fs_info *info; | |
1684 | struct list_head rbio_list; | |
385de0ef | 1685 | struct work_struct work; |
6ac0f488 CM |
1686 | }; |
1687 | ||
1688 | /* | |
1689 | * rbios on the plug list are sorted for easier merging. | |
1690 | */ | |
4f0f586b ST |
1691 | static int plug_cmp(void *priv, const struct list_head *a, |
1692 | const struct list_head *b) | |
6ac0f488 | 1693 | { |
214cc184 DS |
1694 | const struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio, |
1695 | plug_list); | |
1696 | const struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio, | |
1697 | plug_list); | |
4f024f37 KO |
1698 | u64 a_sector = ra->bio_list.head->bi_iter.bi_sector; |
1699 | u64 b_sector = rb->bio_list.head->bi_iter.bi_sector; | |
6ac0f488 CM |
1700 | |
1701 | if (a_sector < b_sector) | |
1702 | return -1; | |
1703 | if (a_sector > b_sector) | |
1704 | return 1; | |
1705 | return 0; | |
1706 | } | |
1707 | ||
1708 | static void run_plug(struct btrfs_plug_cb *plug) | |
1709 | { | |
1710 | struct btrfs_raid_bio *cur; | |
1711 | struct btrfs_raid_bio *last = NULL; | |
1712 | ||
1713 | /* | |
1714 | * sort our plug list then try to merge | |
1715 | * everything we can in hopes of creating full | |
1716 | * stripes. | |
1717 | */ | |
1718 | list_sort(NULL, &plug->rbio_list, plug_cmp); | |
1719 | while (!list_empty(&plug->rbio_list)) { | |
1720 | cur = list_entry(plug->rbio_list.next, | |
1721 | struct btrfs_raid_bio, plug_list); | |
1722 | list_del_init(&cur->plug_list); | |
1723 | ||
1724 | if (rbio_is_full(cur)) { | |
c7b562c5 DS |
1725 | int ret; |
1726 | ||
6ac0f488 | 1727 | /* we have a full stripe, send it down */ |
c7b562c5 DS |
1728 | ret = full_stripe_write(cur); |
1729 | BUG_ON(ret); | |
6ac0f488 CM |
1730 | continue; |
1731 | } | |
1732 | if (last) { | |
1733 | if (rbio_can_merge(last, cur)) { | |
1734 | merge_rbio(last, cur); | |
ff2b64a2 | 1735 | free_raid_bio(cur); |
6ac0f488 CM |
1736 | continue; |
1737 | ||
1738 | } | |
1739 | __raid56_parity_write(last); | |
1740 | } | |
1741 | last = cur; | |
1742 | } | |
1743 | if (last) { | |
1744 | __raid56_parity_write(last); | |
1745 | } | |
1746 | kfree(plug); | |
1747 | } | |
1748 | ||
1749 | /* | |
1750 | * if the unplug comes from schedule, we have to push the | |
1751 | * work off to a helper thread | |
1752 | */ | |
385de0ef | 1753 | static void unplug_work(struct work_struct *work) |
6ac0f488 CM |
1754 | { |
1755 | struct btrfs_plug_cb *plug; | |
1756 | plug = container_of(work, struct btrfs_plug_cb, work); | |
1757 | run_plug(plug); | |
1758 | } | |
1759 | ||
1760 | static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1761 | { | |
1762 | struct btrfs_plug_cb *plug; | |
1763 | plug = container_of(cb, struct btrfs_plug_cb, cb); | |
1764 | ||
1765 | if (from_schedule) { | |
385de0ef CH |
1766 | INIT_WORK(&plug->work, unplug_work); |
1767 | queue_work(plug->info->rmw_workers, &plug->work); | |
6ac0f488 CM |
1768 | return; |
1769 | } | |
1770 | run_plug(plug); | |
1771 | } | |
1772 | ||
bd8f7e62 QW |
1773 | /* Add the original bio into rbio->bio_list, and update rbio::dbitmap. */ |
1774 | static void rbio_add_bio(struct btrfs_raid_bio *rbio, struct bio *orig_bio) | |
1775 | { | |
1776 | const struct btrfs_fs_info *fs_info = rbio->bioc->fs_info; | |
1777 | const u64 orig_logical = orig_bio->bi_iter.bi_sector << SECTOR_SHIFT; | |
1778 | const u64 full_stripe_start = rbio->bioc->raid_map[0]; | |
1779 | const u32 orig_len = orig_bio->bi_iter.bi_size; | |
1780 | const u32 sectorsize = fs_info->sectorsize; | |
1781 | u64 cur_logical; | |
1782 | ||
1783 | ASSERT(orig_logical >= full_stripe_start && | |
1784 | orig_logical + orig_len <= full_stripe_start + | |
ff18a4af | 1785 | rbio->nr_data * BTRFS_STRIPE_LEN); |
bd8f7e62 QW |
1786 | |
1787 | bio_list_add(&rbio->bio_list, orig_bio); | |
1788 | rbio->bio_list_bytes += orig_bio->bi_iter.bi_size; | |
1789 | ||
1790 | /* Update the dbitmap. */ | |
1791 | for (cur_logical = orig_logical; cur_logical < orig_logical + orig_len; | |
1792 | cur_logical += sectorsize) { | |
1793 | int bit = ((u32)(cur_logical - full_stripe_start) >> | |
1794 | fs_info->sectorsize_bits) % rbio->stripe_nsectors; | |
1795 | ||
1796 | set_bit(bit, &rbio->dbitmap); | |
1797 | } | |
1798 | } | |
1799 | ||
53b381b3 DW |
1800 | /* |
1801 | * our main entry point for writes from the rest of the FS. | |
1802 | */ | |
31683f4a | 1803 | void raid56_parity_write(struct bio *bio, struct btrfs_io_context *bioc) |
53b381b3 | 1804 | { |
6a258d72 | 1805 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
53b381b3 | 1806 | struct btrfs_raid_bio *rbio; |
6ac0f488 CM |
1807 | struct btrfs_plug_cb *plug = NULL; |
1808 | struct blk_plug_cb *cb; | |
31683f4a | 1809 | int ret = 0; |
53b381b3 | 1810 | |
ff18a4af | 1811 | rbio = alloc_rbio(fs_info, bioc); |
af8e2d1d | 1812 | if (IS_ERR(rbio)) { |
31683f4a | 1813 | ret = PTR_ERR(rbio); |
f1c29379 | 1814 | goto fail; |
af8e2d1d | 1815 | } |
1b94b556 | 1816 | rbio->operation = BTRFS_RBIO_WRITE; |
bd8f7e62 | 1817 | rbio_add_bio(rbio, bio); |
6ac0f488 CM |
1818 | |
1819 | /* | |
1820 | * don't plug on full rbios, just get them out the door | |
1821 | * as quickly as we can | |
1822 | */ | |
4245215d MX |
1823 | if (rbio_is_full(rbio)) { |
1824 | ret = full_stripe_write(rbio); | |
ab4c54c6 | 1825 | if (ret) { |
ff2b64a2 | 1826 | free_raid_bio(rbio); |
f1c29379 | 1827 | goto fail; |
ab4c54c6 | 1828 | } |
31683f4a | 1829 | return; |
4245215d | 1830 | } |
6ac0f488 | 1831 | |
0b246afa | 1832 | cb = blk_check_plugged(btrfs_raid_unplug, fs_info, sizeof(*plug)); |
6ac0f488 CM |
1833 | if (cb) { |
1834 | plug = container_of(cb, struct btrfs_plug_cb, cb); | |
1835 | if (!plug->info) { | |
0b246afa | 1836 | plug->info = fs_info; |
6ac0f488 CM |
1837 | INIT_LIST_HEAD(&plug->rbio_list); |
1838 | } | |
1839 | list_add_tail(&rbio->plug_list, &plug->rbio_list); | |
1840 | } else { | |
4245215d | 1841 | ret = __raid56_parity_write(rbio); |
ab4c54c6 | 1842 | if (ret) { |
ff2b64a2 | 1843 | free_raid_bio(rbio); |
f1c29379 | 1844 | goto fail; |
ab4c54c6 | 1845 | } |
6ac0f488 | 1846 | } |
31683f4a CH |
1847 | |
1848 | return; | |
1849 | ||
f1c29379 | 1850 | fail: |
31683f4a CH |
1851 | bio->bi_status = errno_to_blk_status(ret); |
1852 | bio_endio(bio); | |
53b381b3 DW |
1853 | } |
1854 | ||
1855 | /* | |
1856 | * all parity reconstruction happens here. We've read in everything | |
1857 | * we can find from the drives and this does the heavy lifting of | |
1858 | * sorting the good from the bad. | |
1859 | */ | |
1860 | static void __raid_recover_end_io(struct btrfs_raid_bio *rbio) | |
1861 | { | |
07e4d380 QW |
1862 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; |
1863 | int sectornr, stripe; | |
53b381b3 | 1864 | void **pointers; |
94a0b58d | 1865 | void **unmap_array; |
53b381b3 | 1866 | int faila = -1, failb = -1; |
58efbc9f | 1867 | blk_status_t err; |
53b381b3 DW |
1868 | int i; |
1869 | ||
07e4d380 QW |
1870 | /* |
1871 | * This array stores the pointer for each sector, thus it has the extra | |
1872 | * pgoff value added from each sector | |
1873 | */ | |
31e818fe | 1874 | pointers = kcalloc(rbio->real_stripes, sizeof(void *), GFP_NOFS); |
53b381b3 | 1875 | if (!pointers) { |
58efbc9f | 1876 | err = BLK_STS_RESOURCE; |
53b381b3 DW |
1877 | goto cleanup_io; |
1878 | } | |
1879 | ||
94a0b58d IW |
1880 | /* |
1881 | * Store copy of pointers that does not get reordered during | |
1882 | * reconstruction so that kunmap_local works. | |
1883 | */ | |
1884 | unmap_array = kcalloc(rbio->real_stripes, sizeof(void *), GFP_NOFS); | |
1885 | if (!unmap_array) { | |
1886 | err = BLK_STS_RESOURCE; | |
1887 | goto cleanup_pointers; | |
1888 | } | |
1889 | ||
53b381b3 DW |
1890 | faila = rbio->faila; |
1891 | failb = rbio->failb; | |
1892 | ||
b4ee1782 OS |
1893 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1894 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { | |
53b381b3 DW |
1895 | spin_lock_irq(&rbio->bio_list_lock); |
1896 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
1897 | spin_unlock_irq(&rbio->bio_list_lock); | |
1898 | } | |
1899 | ||
1900 | index_rbio_pages(rbio); | |
1901 | ||
07e4d380 QW |
1902 | for (sectornr = 0; sectornr < rbio->stripe_nsectors; sectornr++) { |
1903 | struct sector_ptr *sector; | |
1904 | ||
5a6ac9ea MX |
1905 | /* |
1906 | * Now we just use bitmap to mark the horizontal stripes in | |
1907 | * which we have data when doing parity scrub. | |
1908 | */ | |
1909 | if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB && | |
c67c68eb | 1910 | !test_bit(sectornr, &rbio->dbitmap)) |
5a6ac9ea MX |
1911 | continue; |
1912 | ||
94a0b58d | 1913 | /* |
07e4d380 | 1914 | * Setup our array of pointers with sectors from each stripe |
94a0b58d IW |
1915 | * |
1916 | * NOTE: store a duplicate array of pointers to preserve the | |
1917 | * pointer order | |
53b381b3 | 1918 | */ |
2c8cdd6e | 1919 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
53b381b3 | 1920 | /* |
07e4d380 | 1921 | * If we're rebuilding a read, we have to use |
53b381b3 DW |
1922 | * pages from the bio list |
1923 | */ | |
b4ee1782 OS |
1924 | if ((rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1925 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) && | |
53b381b3 | 1926 | (stripe == faila || stripe == failb)) { |
07e4d380 | 1927 | sector = sector_in_rbio(rbio, stripe, sectornr, 0); |
53b381b3 | 1928 | } else { |
07e4d380 | 1929 | sector = rbio_stripe_sector(rbio, stripe, sectornr); |
53b381b3 | 1930 | } |
07e4d380 QW |
1931 | ASSERT(sector->page); |
1932 | pointers[stripe] = kmap_local_page(sector->page) + | |
1933 | sector->pgoff; | |
94a0b58d | 1934 | unmap_array[stripe] = pointers[stripe]; |
53b381b3 DW |
1935 | } |
1936 | ||
07e4d380 | 1937 | /* All raid6 handling here */ |
4c664611 | 1938 | if (rbio->bioc->map_type & BTRFS_BLOCK_GROUP_RAID6) { |
07e4d380 | 1939 | /* Single failure, rebuild from parity raid5 style */ |
53b381b3 DW |
1940 | if (failb < 0) { |
1941 | if (faila == rbio->nr_data) { | |
1942 | /* | |
1943 | * Just the P stripe has failed, without | |
1944 | * a bad data or Q stripe. | |
1945 | * TODO, we should redo the xor here. | |
1946 | */ | |
58efbc9f | 1947 | err = BLK_STS_IOERR; |
53b381b3 DW |
1948 | goto cleanup; |
1949 | } | |
1950 | /* | |
1951 | * a single failure in raid6 is rebuilt | |
1952 | * in the pstripe code below | |
1953 | */ | |
1954 | goto pstripe; | |
1955 | } | |
1956 | ||
1957 | /* make sure our ps and qs are in order */ | |
b7d2083a NB |
1958 | if (faila > failb) |
1959 | swap(faila, failb); | |
53b381b3 DW |
1960 | |
1961 | /* if the q stripe is failed, do a pstripe reconstruction | |
1962 | * from the xors. | |
1963 | * If both the q stripe and the P stripe are failed, we're | |
1964 | * here due to a crc mismatch and we can't give them the | |
1965 | * data they want | |
1966 | */ | |
4c664611 QW |
1967 | if (rbio->bioc->raid_map[failb] == RAID6_Q_STRIPE) { |
1968 | if (rbio->bioc->raid_map[faila] == | |
8e5cfb55 | 1969 | RAID5_P_STRIPE) { |
58efbc9f | 1970 | err = BLK_STS_IOERR; |
53b381b3 DW |
1971 | goto cleanup; |
1972 | } | |
1973 | /* | |
1974 | * otherwise we have one bad data stripe and | |
1975 | * a good P stripe. raid5! | |
1976 | */ | |
1977 | goto pstripe; | |
1978 | } | |
1979 | ||
4c664611 | 1980 | if (rbio->bioc->raid_map[failb] == RAID5_P_STRIPE) { |
2c8cdd6e | 1981 | raid6_datap_recov(rbio->real_stripes, |
07e4d380 | 1982 | sectorsize, faila, pointers); |
53b381b3 | 1983 | } else { |
2c8cdd6e | 1984 | raid6_2data_recov(rbio->real_stripes, |
07e4d380 | 1985 | sectorsize, faila, failb, |
53b381b3 DW |
1986 | pointers); |
1987 | } | |
1988 | } else { | |
1989 | void *p; | |
1990 | ||
1991 | /* rebuild from P stripe here (raid5 or raid6) */ | |
1992 | BUG_ON(failb != -1); | |
1993 | pstripe: | |
1994 | /* Copy parity block into failed block to start with */ | |
07e4d380 | 1995 | memcpy(pointers[faila], pointers[rbio->nr_data], sectorsize); |
53b381b3 DW |
1996 | |
1997 | /* rearrange the pointer array */ | |
1998 | p = pointers[faila]; | |
1999 | for (stripe = faila; stripe < rbio->nr_data - 1; stripe++) | |
2000 | pointers[stripe] = pointers[stripe + 1]; | |
2001 | pointers[rbio->nr_data - 1] = p; | |
2002 | ||
2003 | /* xor in the rest */ | |
07e4d380 | 2004 | run_xor(pointers, rbio->nr_data - 1, sectorsize); |
53b381b3 DW |
2005 | } |
2006 | /* if we're doing this rebuild as part of an rmw, go through | |
2007 | * and set all of our private rbio pages in the | |
2008 | * failed stripes as uptodate. This way finish_rmw will | |
2009 | * know they can be trusted. If this was a read reconstruction, | |
2010 | * other endio functions will fiddle the uptodate bits | |
2011 | */ | |
1b94b556 | 2012 | if (rbio->operation == BTRFS_RBIO_WRITE) { |
07e4d380 | 2013 | for (i = 0; i < rbio->stripe_nsectors; i++) { |
53b381b3 | 2014 | if (faila != -1) { |
07e4d380 QW |
2015 | sector = rbio_stripe_sector(rbio, faila, i); |
2016 | sector->uptodate = 1; | |
53b381b3 DW |
2017 | } |
2018 | if (failb != -1) { | |
07e4d380 QW |
2019 | sector = rbio_stripe_sector(rbio, failb, i); |
2020 | sector->uptodate = 1; | |
53b381b3 DW |
2021 | } |
2022 | } | |
2023 | } | |
94a0b58d IW |
2024 | for (stripe = rbio->real_stripes - 1; stripe >= 0; stripe--) |
2025 | kunmap_local(unmap_array[stripe]); | |
53b381b3 DW |
2026 | } |
2027 | ||
58efbc9f | 2028 | err = BLK_STS_OK; |
53b381b3 | 2029 | cleanup: |
94a0b58d IW |
2030 | kfree(unmap_array); |
2031 | cleanup_pointers: | |
53b381b3 DW |
2032 | kfree(pointers); |
2033 | ||
2034 | cleanup_io: | |
580c6efa LB |
2035 | /* |
2036 | * Similar to READ_REBUILD, REBUILD_MISSING at this point also has a | |
2037 | * valid rbio which is consistent with ondisk content, thus such a | |
2038 | * valid rbio can be cached to avoid further disk reads. | |
2039 | */ | |
2040 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || | |
2041 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { | |
44ac474d LB |
2042 | /* |
2043 | * - In case of two failures, where rbio->failb != -1: | |
2044 | * | |
2045 | * Do not cache this rbio since the above read reconstruction | |
2046 | * (raid6_datap_recov() or raid6_2data_recov()) may have | |
2047 | * changed some content of stripes which are not identical to | |
2048 | * on-disk content any more, otherwise, a later write/recover | |
2049 | * may steal stripe_pages from this rbio and end up with | |
2050 | * corruptions or rebuild failures. | |
2051 | * | |
2052 | * - In case of single failure, where rbio->failb == -1: | |
2053 | * | |
2054 | * Cache this rbio iff the above read reconstruction is | |
52042d8e | 2055 | * executed without problems. |
44ac474d LB |
2056 | */ |
2057 | if (err == BLK_STS_OK && rbio->failb < 0) | |
4ae10b3a CM |
2058 | cache_rbio_pages(rbio); |
2059 | else | |
2060 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
2061 | ||
4246a0b6 | 2062 | rbio_orig_end_io(rbio, err); |
58efbc9f | 2063 | } else if (err == BLK_STS_OK) { |
53b381b3 DW |
2064 | rbio->faila = -1; |
2065 | rbio->failb = -1; | |
5a6ac9ea MX |
2066 | |
2067 | if (rbio->operation == BTRFS_RBIO_WRITE) | |
2068 | finish_rmw(rbio); | |
2069 | else if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) | |
2070 | finish_parity_scrub(rbio, 0); | |
2071 | else | |
2072 | BUG(); | |
53b381b3 | 2073 | } else { |
4246a0b6 | 2074 | rbio_orig_end_io(rbio, err); |
53b381b3 DW |
2075 | } |
2076 | } | |
2077 | ||
2078 | /* | |
d34e123d CH |
2079 | * This is called only for stripes we've read from disk to reconstruct the |
2080 | * parity. | |
53b381b3 | 2081 | */ |
d34e123d | 2082 | static void raid_recover_end_io_work(struct work_struct *work) |
53b381b3 | 2083 | { |
d34e123d CH |
2084 | struct btrfs_raid_bio *rbio = |
2085 | container_of(work, struct btrfs_raid_bio, end_io_work); | |
53b381b3 | 2086 | |
4c664611 | 2087 | if (atomic_read(&rbio->error) > rbio->bioc->max_errors) |
58efbc9f | 2088 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
53b381b3 DW |
2089 | else |
2090 | __raid_recover_end_io(rbio); | |
2091 | } | |
2092 | ||
2093 | /* | |
2094 | * reads everything we need off the disk to reconstruct | |
2095 | * the parity. endio handlers trigger final reconstruction | |
2096 | * when the IO is done. | |
2097 | * | |
2098 | * This is used both for reads from the higher layers and for | |
2099 | * parity construction required to finish a rmw cycle. | |
2100 | */ | |
2101 | static int __raid56_parity_recover(struct btrfs_raid_bio *rbio) | |
2102 | { | |
2103 | int bios_to_read = 0; | |
53b381b3 DW |
2104 | struct bio_list bio_list; |
2105 | int ret; | |
ef340fcc | 2106 | int total_sector_nr; |
53b381b3 DW |
2107 | struct bio *bio; |
2108 | ||
2109 | bio_list_init(&bio_list); | |
2110 | ||
2111 | ret = alloc_rbio_pages(rbio); | |
2112 | if (ret) | |
2113 | goto cleanup; | |
2114 | ||
b89e1b01 | 2115 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
2116 | |
2117 | /* | |
f6065f8e QW |
2118 | * Read everything that hasn't failed. However this time we will |
2119 | * not trust any cached sector. | |
2120 | * As we may read out some stale data but higher layer is not reading | |
2121 | * that stale part. | |
2122 | * | |
2123 | * So here we always re-read everything in recovery path. | |
53b381b3 | 2124 | */ |
ef340fcc QW |
2125 | for (total_sector_nr = 0; total_sector_nr < rbio->nr_sectors; |
2126 | total_sector_nr++) { | |
2127 | int stripe = total_sector_nr / rbio->stripe_nsectors; | |
2128 | int sectornr = total_sector_nr % rbio->stripe_nsectors; | |
2129 | struct sector_ptr *sector; | |
2130 | ||
5588383e | 2131 | if (rbio->faila == stripe || rbio->failb == stripe) { |
b89e1b01 | 2132 | atomic_inc(&rbio->error); |
ef340fcc QW |
2133 | /* Skip the current stripe. */ |
2134 | ASSERT(sectornr == 0); | |
2135 | total_sector_nr += rbio->stripe_nsectors - 1; | |
53b381b3 | 2136 | continue; |
5588383e | 2137 | } |
ef340fcc | 2138 | sector = rbio_stripe_sector(rbio, stripe, sectornr); |
ef340fcc | 2139 | ret = rbio_add_io_sector(rbio, &bio_list, sector, stripe, |
ff18a4af | 2140 | sectornr, REQ_OP_READ); |
ef340fcc QW |
2141 | if (ret < 0) |
2142 | goto cleanup; | |
53b381b3 DW |
2143 | } |
2144 | ||
2145 | bios_to_read = bio_list_size(&bio_list); | |
2146 | if (!bios_to_read) { | |
2147 | /* | |
2148 | * we might have no bios to read just because the pages | |
2149 | * were up to date, or we might have no bios to read because | |
2150 | * the devices were gone. | |
2151 | */ | |
4c664611 | 2152 | if (atomic_read(&rbio->error) <= rbio->bioc->max_errors) { |
53b381b3 | 2153 | __raid_recover_end_io(rbio); |
813f8a0e | 2154 | return 0; |
53b381b3 DW |
2155 | } else { |
2156 | goto cleanup; | |
2157 | } | |
2158 | } | |
2159 | ||
2160 | /* | |
4c664611 QW |
2161 | * The bioc may be freed once we submit the last bio. Make sure not to |
2162 | * touch it after that. | |
53b381b3 | 2163 | */ |
b89e1b01 | 2164 | atomic_set(&rbio->stripes_pending, bios_to_read); |
d34e123d | 2165 | INIT_WORK(&rbio->end_io_work, raid_recover_end_io_work); |
bf28a605 | 2166 | while ((bio = bio_list_pop(&bio_list))) { |
d34e123d | 2167 | bio->bi_end_io = raid56_bio_end_io; |
53b381b3 | 2168 | |
b8bea09a QW |
2169 | if (trace_raid56_scrub_read_recover_enabled()) { |
2170 | struct raid56_bio_trace_info trace_info = { 0 }; | |
53b381b3 | 2171 | |
b8bea09a QW |
2172 | bio_get_trace_info(rbio, bio, &trace_info); |
2173 | trace_raid56_scrub_read_recover(rbio, bio, &trace_info); | |
2174 | } | |
4e49ea4a | 2175 | submit_bio(bio); |
53b381b3 | 2176 | } |
813f8a0e | 2177 | |
53b381b3 DW |
2178 | return 0; |
2179 | ||
2180 | cleanup: | |
b4ee1782 OS |
2181 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
2182 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) | |
58efbc9f | 2183 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2184 | |
2185 | while ((bio = bio_list_pop(&bio_list))) | |
2186 | bio_put(bio); | |
2187 | ||
53b381b3 DW |
2188 | return -EIO; |
2189 | } | |
2190 | ||
2191 | /* | |
2192 | * the main entry point for reads from the higher layers. This | |
2193 | * is really only called when the normal read path had a failure, | |
2194 | * so we assume the bio they send down corresponds to a failed part | |
2195 | * of the drive. | |
2196 | */ | |
6065fd95 | 2197 | void raid56_parity_recover(struct bio *bio, struct btrfs_io_context *bioc, |
f1c29379 | 2198 | int mirror_num) |
53b381b3 | 2199 | { |
6a258d72 | 2200 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
53b381b3 | 2201 | struct btrfs_raid_bio *rbio; |
53b381b3 | 2202 | |
ff18a4af | 2203 | rbio = alloc_rbio(fs_info, bioc); |
af8e2d1d | 2204 | if (IS_ERR(rbio)) { |
6065fd95 CH |
2205 | bio->bi_status = errno_to_blk_status(PTR_ERR(rbio)); |
2206 | goto out_end_bio; | |
af8e2d1d | 2207 | } |
53b381b3 | 2208 | |
1b94b556 | 2209 | rbio->operation = BTRFS_RBIO_READ_REBUILD; |
bd8f7e62 | 2210 | rbio_add_bio(rbio, bio); |
53b381b3 DW |
2211 | |
2212 | rbio->faila = find_logical_bio_stripe(rbio, bio); | |
2213 | if (rbio->faila == -1) { | |
0b246afa | 2214 | btrfs_warn(fs_info, |
4c664611 | 2215 | "%s could not find the bad stripe in raid56 so that we cannot recover any more (bio has logical %llu len %llu, bioc has map_type %llu)", |
1201b58b | 2216 | __func__, bio->bi_iter.bi_sector << 9, |
4c664611 | 2217 | (u64)bio->bi_iter.bi_size, bioc->map_type); |
ff2b64a2 | 2218 | free_raid_bio(rbio); |
6065fd95 CH |
2219 | bio->bi_status = BLK_STS_IOERR; |
2220 | goto out_end_bio; | |
53b381b3 DW |
2221 | } |
2222 | ||
2223 | /* | |
8810f751 LB |
2224 | * Loop retry: |
2225 | * for 'mirror == 2', reconstruct from all other stripes. | |
2226 | * for 'mirror_num > 2', select a stripe to fail on every retry. | |
53b381b3 | 2227 | */ |
8810f751 LB |
2228 | if (mirror_num > 2) { |
2229 | /* | |
2230 | * 'mirror == 3' is to fail the p stripe and | |
2231 | * reconstruct from the q stripe. 'mirror > 3' is to | |
2232 | * fail a data stripe and reconstruct from p+q stripe. | |
2233 | */ | |
2234 | rbio->failb = rbio->real_stripes - (mirror_num - 1); | |
2235 | ASSERT(rbio->failb > 0); | |
2236 | if (rbio->failb <= rbio->faila) | |
2237 | rbio->failb--; | |
2238 | } | |
53b381b3 | 2239 | |
6065fd95 CH |
2240 | if (lock_stripe_add(rbio)) |
2241 | return; | |
53b381b3 DW |
2242 | |
2243 | /* | |
6065fd95 CH |
2244 | * This adds our rbio to the list of rbios that will be handled after |
2245 | * the current lock owner is done. | |
53b381b3 | 2246 | */ |
6065fd95 CH |
2247 | __raid56_parity_recover(rbio); |
2248 | return; | |
53b381b3 | 2249 | |
6065fd95 | 2250 | out_end_bio: |
6065fd95 | 2251 | bio_endio(bio); |
53b381b3 DW |
2252 | } |
2253 | ||
385de0ef | 2254 | static void rmw_work(struct work_struct *work) |
53b381b3 DW |
2255 | { |
2256 | struct btrfs_raid_bio *rbio; | |
2257 | ||
2258 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2259 | raid56_rmw_stripe(rbio); | |
2260 | } | |
2261 | ||
385de0ef | 2262 | static void read_rebuild_work(struct work_struct *work) |
53b381b3 DW |
2263 | { |
2264 | struct btrfs_raid_bio *rbio; | |
2265 | ||
2266 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2267 | __raid56_parity_recover(rbio); | |
2268 | } | |
5a6ac9ea MX |
2269 | |
2270 | /* | |
2271 | * The following code is used to scrub/replace the parity stripe | |
2272 | * | |
4c664611 | 2273 | * Caller must have already increased bio_counter for getting @bioc. |
ae6529c3 | 2274 | * |
5a6ac9ea MX |
2275 | * Note: We need make sure all the pages that add into the scrub/replace |
2276 | * raid bio are correct and not be changed during the scrub/replace. That | |
2277 | * is those pages just hold metadata or file data with checksum. | |
2278 | */ | |
2279 | ||
6a258d72 QW |
2280 | struct btrfs_raid_bio *raid56_parity_alloc_scrub_rbio(struct bio *bio, |
2281 | struct btrfs_io_context *bioc, | |
ff18a4af | 2282 | struct btrfs_device *scrub_dev, |
6a258d72 | 2283 | unsigned long *dbitmap, int stripe_nsectors) |
5a6ac9ea | 2284 | { |
6a258d72 | 2285 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
5a6ac9ea MX |
2286 | struct btrfs_raid_bio *rbio; |
2287 | int i; | |
2288 | ||
ff18a4af | 2289 | rbio = alloc_rbio(fs_info, bioc); |
5a6ac9ea MX |
2290 | if (IS_ERR(rbio)) |
2291 | return NULL; | |
2292 | bio_list_add(&rbio->bio_list, bio); | |
2293 | /* | |
2294 | * This is a special bio which is used to hold the completion handler | |
2295 | * and make the scrub rbio is similar to the other types | |
2296 | */ | |
2297 | ASSERT(!bio->bi_iter.bi_size); | |
2298 | rbio->operation = BTRFS_RBIO_PARITY_SCRUB; | |
2299 | ||
9cd3a7eb | 2300 | /* |
4c664611 | 2301 | * After mapping bioc with BTRFS_MAP_WRITE, parities have been sorted |
9cd3a7eb LB |
2302 | * to the end position, so this search can start from the first parity |
2303 | * stripe. | |
2304 | */ | |
2305 | for (i = rbio->nr_data; i < rbio->real_stripes; i++) { | |
4c664611 | 2306 | if (bioc->stripes[i].dev == scrub_dev) { |
5a6ac9ea MX |
2307 | rbio->scrubp = i; |
2308 | break; | |
2309 | } | |
2310 | } | |
9cd3a7eb | 2311 | ASSERT(i < rbio->real_stripes); |
5a6ac9ea | 2312 | |
c67c68eb | 2313 | bitmap_copy(&rbio->dbitmap, dbitmap, stripe_nsectors); |
5a6ac9ea MX |
2314 | return rbio; |
2315 | } | |
2316 | ||
b4ee1782 OS |
2317 | /* Used for both parity scrub and missing. */ |
2318 | void raid56_add_scrub_pages(struct btrfs_raid_bio *rbio, struct page *page, | |
6346f6bf | 2319 | unsigned int pgoff, u64 logical) |
5a6ac9ea | 2320 | { |
6346f6bf | 2321 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; |
5a6ac9ea MX |
2322 | int stripe_offset; |
2323 | int index; | |
2324 | ||
4c664611 | 2325 | ASSERT(logical >= rbio->bioc->raid_map[0]); |
6346f6bf | 2326 | ASSERT(logical + sectorsize <= rbio->bioc->raid_map[0] + |
ff18a4af | 2327 | BTRFS_STRIPE_LEN * rbio->nr_data); |
4c664611 | 2328 | stripe_offset = (int)(logical - rbio->bioc->raid_map[0]); |
6346f6bf QW |
2329 | index = stripe_offset / sectorsize; |
2330 | rbio->bio_sectors[index].page = page; | |
2331 | rbio->bio_sectors[index].pgoff = pgoff; | |
5a6ac9ea MX |
2332 | } |
2333 | ||
2334 | /* | |
2335 | * We just scrub the parity that we have correct data on the same horizontal, | |
2336 | * so we needn't allocate all pages for all the stripes. | |
2337 | */ | |
2338 | static int alloc_rbio_essential_pages(struct btrfs_raid_bio *rbio) | |
2339 | { | |
3907ce29 | 2340 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; |
aee35e4b | 2341 | int total_sector_nr; |
5a6ac9ea | 2342 | |
aee35e4b QW |
2343 | for (total_sector_nr = 0; total_sector_nr < rbio->nr_sectors; |
2344 | total_sector_nr++) { | |
2345 | struct page *page; | |
2346 | int sectornr = total_sector_nr % rbio->stripe_nsectors; | |
2347 | int index = (total_sector_nr * sectorsize) >> PAGE_SHIFT; | |
5a6ac9ea | 2348 | |
aee35e4b QW |
2349 | if (!test_bit(sectornr, &rbio->dbitmap)) |
2350 | continue; | |
2351 | if (rbio->stripe_pages[index]) | |
2352 | continue; | |
2353 | page = alloc_page(GFP_NOFS); | |
2354 | if (!page) | |
2355 | return -ENOMEM; | |
2356 | rbio->stripe_pages[index] = page; | |
5a6ac9ea | 2357 | } |
eb357060 | 2358 | index_stripe_sectors(rbio); |
5a6ac9ea MX |
2359 | return 0; |
2360 | } | |
2361 | ||
5a6ac9ea MX |
2362 | static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, |
2363 | int need_check) | |
2364 | { | |
4c664611 | 2365 | struct btrfs_io_context *bioc = rbio->bioc; |
46900662 | 2366 | const u32 sectorsize = bioc->fs_info->sectorsize; |
1389053e | 2367 | void **pointers = rbio->finish_pointers; |
c67c68eb | 2368 | unsigned long *pbitmap = &rbio->finish_pbitmap; |
5a6ac9ea MX |
2369 | int nr_data = rbio->nr_data; |
2370 | int stripe; | |
3e77605d | 2371 | int sectornr; |
c17af965 | 2372 | bool has_qstripe; |
46900662 QW |
2373 | struct sector_ptr p_sector = { 0 }; |
2374 | struct sector_ptr q_sector = { 0 }; | |
5a6ac9ea MX |
2375 | struct bio_list bio_list; |
2376 | struct bio *bio; | |
76035976 | 2377 | int is_replace = 0; |
5a6ac9ea MX |
2378 | int ret; |
2379 | ||
2380 | bio_list_init(&bio_list); | |
2381 | ||
c17af965 DS |
2382 | if (rbio->real_stripes - rbio->nr_data == 1) |
2383 | has_qstripe = false; | |
2384 | else if (rbio->real_stripes - rbio->nr_data == 2) | |
2385 | has_qstripe = true; | |
2386 | else | |
5a6ac9ea | 2387 | BUG(); |
5a6ac9ea | 2388 | |
4c664611 | 2389 | if (bioc->num_tgtdevs && bioc->tgtdev_map[rbio->scrubp]) { |
76035976 | 2390 | is_replace = 1; |
c67c68eb | 2391 | bitmap_copy(pbitmap, &rbio->dbitmap, rbio->stripe_nsectors); |
76035976 MX |
2392 | } |
2393 | ||
5a6ac9ea MX |
2394 | /* |
2395 | * Because the higher layers(scrubber) are unlikely to | |
2396 | * use this area of the disk again soon, so don't cache | |
2397 | * it. | |
2398 | */ | |
2399 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
2400 | ||
2401 | if (!need_check) | |
2402 | goto writeback; | |
2403 | ||
46900662 QW |
2404 | p_sector.page = alloc_page(GFP_NOFS); |
2405 | if (!p_sector.page) | |
5a6ac9ea | 2406 | goto cleanup; |
46900662 QW |
2407 | p_sector.pgoff = 0; |
2408 | p_sector.uptodate = 1; | |
5a6ac9ea | 2409 | |
c17af965 | 2410 | if (has_qstripe) { |
d70cef0d | 2411 | /* RAID6, allocate and map temp space for the Q stripe */ |
46900662 QW |
2412 | q_sector.page = alloc_page(GFP_NOFS); |
2413 | if (!q_sector.page) { | |
2414 | __free_page(p_sector.page); | |
2415 | p_sector.page = NULL; | |
5a6ac9ea MX |
2416 | goto cleanup; |
2417 | } | |
46900662 QW |
2418 | q_sector.pgoff = 0; |
2419 | q_sector.uptodate = 1; | |
2420 | pointers[rbio->real_stripes - 1] = kmap_local_page(q_sector.page); | |
5a6ac9ea MX |
2421 | } |
2422 | ||
2423 | atomic_set(&rbio->error, 0); | |
2424 | ||
d70cef0d | 2425 | /* Map the parity stripe just once */ |
46900662 | 2426 | pointers[nr_data] = kmap_local_page(p_sector.page); |
d70cef0d | 2427 | |
c67c68eb | 2428 | for_each_set_bit(sectornr, &rbio->dbitmap, rbio->stripe_nsectors) { |
46900662 | 2429 | struct sector_ptr *sector; |
5a6ac9ea | 2430 | void *parity; |
46900662 | 2431 | |
5a6ac9ea MX |
2432 | /* first collect one page from each data stripe */ |
2433 | for (stripe = 0; stripe < nr_data; stripe++) { | |
46900662 QW |
2434 | sector = sector_in_rbio(rbio, stripe, sectornr, 0); |
2435 | pointers[stripe] = kmap_local_page(sector->page) + | |
2436 | sector->pgoff; | |
5a6ac9ea MX |
2437 | } |
2438 | ||
c17af965 | 2439 | if (has_qstripe) { |
d70cef0d | 2440 | /* RAID6, call the library function to fill in our P/Q */ |
46900662 | 2441 | raid6_call.gen_syndrome(rbio->real_stripes, sectorsize, |
5a6ac9ea MX |
2442 | pointers); |
2443 | } else { | |
2444 | /* raid5 */ | |
46900662 QW |
2445 | memcpy(pointers[nr_data], pointers[0], sectorsize); |
2446 | run_xor(pointers + 1, nr_data - 1, sectorsize); | |
5a6ac9ea MX |
2447 | } |
2448 | ||
01327610 | 2449 | /* Check scrubbing parity and repair it */ |
46900662 QW |
2450 | sector = rbio_stripe_sector(rbio, rbio->scrubp, sectornr); |
2451 | parity = kmap_local_page(sector->page) + sector->pgoff; | |
2452 | if (memcmp(parity, pointers[rbio->scrubp], sectorsize) != 0) | |
2453 | memcpy(parity, pointers[rbio->scrubp], sectorsize); | |
5a6ac9ea MX |
2454 | else |
2455 | /* Parity is right, needn't writeback */ | |
c67c68eb | 2456 | bitmap_clear(&rbio->dbitmap, sectornr, 1); |
58c1a35c | 2457 | kunmap_local(parity); |
5a6ac9ea | 2458 | |
94a0b58d IW |
2459 | for (stripe = nr_data - 1; stripe >= 0; stripe--) |
2460 | kunmap_local(pointers[stripe]); | |
5a6ac9ea MX |
2461 | } |
2462 | ||
94a0b58d | 2463 | kunmap_local(pointers[nr_data]); |
46900662 QW |
2464 | __free_page(p_sector.page); |
2465 | p_sector.page = NULL; | |
2466 | if (q_sector.page) { | |
94a0b58d | 2467 | kunmap_local(pointers[rbio->real_stripes - 1]); |
46900662 QW |
2468 | __free_page(q_sector.page); |
2469 | q_sector.page = NULL; | |
d70cef0d | 2470 | } |
5a6ac9ea MX |
2471 | |
2472 | writeback: | |
2473 | /* | |
2474 | * time to start writing. Make bios for everything from the | |
2475 | * higher layers (the bio_list in our rbio) and our p/q. Ignore | |
2476 | * everything else. | |
2477 | */ | |
c67c68eb | 2478 | for_each_set_bit(sectornr, &rbio->dbitmap, rbio->stripe_nsectors) { |
3e77605d | 2479 | struct sector_ptr *sector; |
5a6ac9ea | 2480 | |
3e77605d QW |
2481 | sector = rbio_stripe_sector(rbio, rbio->scrubp, sectornr); |
2482 | ret = rbio_add_io_sector(rbio, &bio_list, sector, rbio->scrubp, | |
ff18a4af | 2483 | sectornr, REQ_OP_WRITE); |
5a6ac9ea MX |
2484 | if (ret) |
2485 | goto cleanup; | |
2486 | } | |
2487 | ||
76035976 MX |
2488 | if (!is_replace) |
2489 | goto submit_write; | |
2490 | ||
3e77605d QW |
2491 | for_each_set_bit(sectornr, pbitmap, rbio->stripe_nsectors) { |
2492 | struct sector_ptr *sector; | |
76035976 | 2493 | |
3e77605d QW |
2494 | sector = rbio_stripe_sector(rbio, rbio->scrubp, sectornr); |
2495 | ret = rbio_add_io_sector(rbio, &bio_list, sector, | |
4c664611 | 2496 | bioc->tgtdev_map[rbio->scrubp], |
ff18a4af | 2497 | sectornr, REQ_OP_WRITE); |
76035976 MX |
2498 | if (ret) |
2499 | goto cleanup; | |
2500 | } | |
2501 | ||
2502 | submit_write: | |
5a6ac9ea MX |
2503 | nr_data = bio_list_size(&bio_list); |
2504 | if (!nr_data) { | |
2505 | /* Every parity is right */ | |
58efbc9f | 2506 | rbio_orig_end_io(rbio, BLK_STS_OK); |
5a6ac9ea MX |
2507 | return; |
2508 | } | |
2509 | ||
2510 | atomic_set(&rbio->stripes_pending, nr_data); | |
2511 | ||
bf28a605 | 2512 | while ((bio = bio_list_pop(&bio_list))) { |
a6111d11 | 2513 | bio->bi_end_io = raid_write_end_io; |
4e49ea4a | 2514 | |
b8bea09a QW |
2515 | if (trace_raid56_scrub_write_stripe_enabled()) { |
2516 | struct raid56_bio_trace_info trace_info = { 0 }; | |
2517 | ||
2518 | bio_get_trace_info(rbio, bio, &trace_info); | |
2519 | trace_raid56_scrub_write_stripe(rbio, bio, &trace_info); | |
2520 | } | |
4e49ea4a | 2521 | submit_bio(bio); |
5a6ac9ea MX |
2522 | } |
2523 | return; | |
2524 | ||
2525 | cleanup: | |
58efbc9f | 2526 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2527 | |
2528 | while ((bio = bio_list_pop(&bio_list))) | |
2529 | bio_put(bio); | |
5a6ac9ea MX |
2530 | } |
2531 | ||
2532 | static inline int is_data_stripe(struct btrfs_raid_bio *rbio, int stripe) | |
2533 | { | |
2534 | if (stripe >= 0 && stripe < rbio->nr_data) | |
2535 | return 1; | |
2536 | return 0; | |
2537 | } | |
2538 | ||
2539 | /* | |
2540 | * While we're doing the parity check and repair, we could have errors | |
2541 | * in reading pages off the disk. This checks for errors and if we're | |
2542 | * not able to read the page it'll trigger parity reconstruction. The | |
2543 | * parity scrub will be finished after we've reconstructed the failed | |
2544 | * stripes | |
2545 | */ | |
2546 | static void validate_rbio_for_parity_scrub(struct btrfs_raid_bio *rbio) | |
2547 | { | |
4c664611 | 2548 | if (atomic_read(&rbio->error) > rbio->bioc->max_errors) |
5a6ac9ea MX |
2549 | goto cleanup; |
2550 | ||
2551 | if (rbio->faila >= 0 || rbio->failb >= 0) { | |
2552 | int dfail = 0, failp = -1; | |
2553 | ||
2554 | if (is_data_stripe(rbio, rbio->faila)) | |
2555 | dfail++; | |
2556 | else if (is_parity_stripe(rbio->faila)) | |
2557 | failp = rbio->faila; | |
2558 | ||
2559 | if (is_data_stripe(rbio, rbio->failb)) | |
2560 | dfail++; | |
2561 | else if (is_parity_stripe(rbio->failb)) | |
2562 | failp = rbio->failb; | |
2563 | ||
2564 | /* | |
2565 | * Because we can not use a scrubbing parity to repair | |
2566 | * the data, so the capability of the repair is declined. | |
2567 | * (In the case of RAID5, we can not repair anything) | |
2568 | */ | |
4c664611 | 2569 | if (dfail > rbio->bioc->max_errors - 1) |
5a6ac9ea MX |
2570 | goto cleanup; |
2571 | ||
2572 | /* | |
2573 | * If all data is good, only parity is correctly, just | |
2574 | * repair the parity. | |
2575 | */ | |
2576 | if (dfail == 0) { | |
2577 | finish_parity_scrub(rbio, 0); | |
2578 | return; | |
2579 | } | |
2580 | ||
2581 | /* | |
2582 | * Here means we got one corrupted data stripe and one | |
2583 | * corrupted parity on RAID6, if the corrupted parity | |
01327610 | 2584 | * is scrubbing parity, luckily, use the other one to repair |
5a6ac9ea MX |
2585 | * the data, or we can not repair the data stripe. |
2586 | */ | |
2587 | if (failp != rbio->scrubp) | |
2588 | goto cleanup; | |
2589 | ||
2590 | __raid_recover_end_io(rbio); | |
2591 | } else { | |
2592 | finish_parity_scrub(rbio, 1); | |
2593 | } | |
2594 | return; | |
2595 | ||
2596 | cleanup: | |
58efbc9f | 2597 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
5a6ac9ea MX |
2598 | } |
2599 | ||
2600 | /* | |
2601 | * end io for the read phase of the rmw cycle. All the bios here are physical | |
2602 | * stripe bios we've read from the disk so we can recalculate the parity of the | |
2603 | * stripe. | |
2604 | * | |
2605 | * This will usually kick off finish_rmw once all the bios are read in, but it | |
2606 | * may trigger parity reconstruction if we had any errors along the way | |
2607 | */ | |
d34e123d | 2608 | static void raid56_parity_scrub_end_io_work(struct work_struct *work) |
5a6ac9ea | 2609 | { |
d34e123d CH |
2610 | struct btrfs_raid_bio *rbio = |
2611 | container_of(work, struct btrfs_raid_bio, end_io_work); | |
5a6ac9ea MX |
2612 | |
2613 | /* | |
d34e123d CH |
2614 | * This will normally call finish_rmw to start our write, but if there |
2615 | * are any failed stripes we'll reconstruct from parity first | |
5a6ac9ea MX |
2616 | */ |
2617 | validate_rbio_for_parity_scrub(rbio); | |
2618 | } | |
2619 | ||
2620 | static void raid56_parity_scrub_stripe(struct btrfs_raid_bio *rbio) | |
2621 | { | |
2622 | int bios_to_read = 0; | |
5a6ac9ea MX |
2623 | struct bio_list bio_list; |
2624 | int ret; | |
1c10702e | 2625 | int total_sector_nr; |
5a6ac9ea MX |
2626 | struct bio *bio; |
2627 | ||
785884fc LB |
2628 | bio_list_init(&bio_list); |
2629 | ||
5a6ac9ea MX |
2630 | ret = alloc_rbio_essential_pages(rbio); |
2631 | if (ret) | |
2632 | goto cleanup; | |
2633 | ||
5a6ac9ea | 2634 | atomic_set(&rbio->error, 0); |
1c10702e QW |
2635 | /* Build a list of bios to read all the missing parts. */ |
2636 | for (total_sector_nr = 0; total_sector_nr < rbio->nr_sectors; | |
2637 | total_sector_nr++) { | |
2638 | int sectornr = total_sector_nr % rbio->stripe_nsectors; | |
2639 | int stripe = total_sector_nr / rbio->stripe_nsectors; | |
2640 | struct sector_ptr *sector; | |
5a6ac9ea | 2641 | |
1c10702e QW |
2642 | /* No data in the vertical stripe, no need to read. */ |
2643 | if (!test_bit(sectornr, &rbio->dbitmap)) | |
2644 | continue; | |
5a6ac9ea | 2645 | |
1c10702e QW |
2646 | /* |
2647 | * We want to find all the sectors missing from the rbio and | |
2648 | * read them from the disk. If sector_in_rbio() finds a sector | |
2649 | * in the bio list we don't need to read it off the stripe. | |
2650 | */ | |
2651 | sector = sector_in_rbio(rbio, stripe, sectornr, 1); | |
2652 | if (sector) | |
2653 | continue; | |
2654 | ||
2655 | sector = rbio_stripe_sector(rbio, stripe, sectornr); | |
2656 | /* | |
2657 | * The bio cache may have handed us an uptodate sector. If so, | |
2658 | * use it. | |
2659 | */ | |
2660 | if (sector->uptodate) | |
2661 | continue; | |
2662 | ||
2663 | ret = rbio_add_io_sector(rbio, &bio_list, sector, stripe, | |
ff18a4af | 2664 | sectornr, REQ_OP_READ); |
1c10702e QW |
2665 | if (ret) |
2666 | goto cleanup; | |
5a6ac9ea MX |
2667 | } |
2668 | ||
2669 | bios_to_read = bio_list_size(&bio_list); | |
2670 | if (!bios_to_read) { | |
2671 | /* | |
2672 | * this can happen if others have merged with | |
2673 | * us, it means there is nothing left to read. | |
2674 | * But if there are missing devices it may not be | |
2675 | * safe to do the full stripe write yet. | |
2676 | */ | |
2677 | goto finish; | |
2678 | } | |
2679 | ||
2680 | /* | |
4c664611 QW |
2681 | * The bioc may be freed once we submit the last bio. Make sure not to |
2682 | * touch it after that. | |
5a6ac9ea MX |
2683 | */ |
2684 | atomic_set(&rbio->stripes_pending, bios_to_read); | |
d34e123d | 2685 | INIT_WORK(&rbio->end_io_work, raid56_parity_scrub_end_io_work); |
bf28a605 | 2686 | while ((bio = bio_list_pop(&bio_list))) { |
d34e123d | 2687 | bio->bi_end_io = raid56_bio_end_io; |
5a6ac9ea | 2688 | |
b8bea09a QW |
2689 | if (trace_raid56_scrub_read_enabled()) { |
2690 | struct raid56_bio_trace_info trace_info = { 0 }; | |
5a6ac9ea | 2691 | |
b8bea09a QW |
2692 | bio_get_trace_info(rbio, bio, &trace_info); |
2693 | trace_raid56_scrub_read(rbio, bio, &trace_info); | |
2694 | } | |
4e49ea4a | 2695 | submit_bio(bio); |
5a6ac9ea MX |
2696 | } |
2697 | /* the actual write will happen once the reads are done */ | |
2698 | return; | |
2699 | ||
2700 | cleanup: | |
58efbc9f | 2701 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2702 | |
2703 | while ((bio = bio_list_pop(&bio_list))) | |
2704 | bio_put(bio); | |
2705 | ||
5a6ac9ea MX |
2706 | return; |
2707 | ||
2708 | finish: | |
2709 | validate_rbio_for_parity_scrub(rbio); | |
2710 | } | |
2711 | ||
385de0ef | 2712 | static void scrub_parity_work(struct work_struct *work) |
5a6ac9ea MX |
2713 | { |
2714 | struct btrfs_raid_bio *rbio; | |
2715 | ||
2716 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2717 | raid56_parity_scrub_stripe(rbio); | |
2718 | } | |
2719 | ||
5a6ac9ea MX |
2720 | void raid56_parity_submit_scrub_rbio(struct btrfs_raid_bio *rbio) |
2721 | { | |
2722 | if (!lock_stripe_add(rbio)) | |
a81b747d | 2723 | start_async_work(rbio, scrub_parity_work); |
5a6ac9ea | 2724 | } |
b4ee1782 OS |
2725 | |
2726 | /* The following code is used for dev replace of a missing RAID 5/6 device. */ | |
2727 | ||
2728 | struct btrfs_raid_bio * | |
ff18a4af | 2729 | raid56_alloc_missing_rbio(struct bio *bio, struct btrfs_io_context *bioc) |
b4ee1782 | 2730 | { |
6a258d72 | 2731 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
b4ee1782 OS |
2732 | struct btrfs_raid_bio *rbio; |
2733 | ||
ff18a4af | 2734 | rbio = alloc_rbio(fs_info, bioc); |
b4ee1782 OS |
2735 | if (IS_ERR(rbio)) |
2736 | return NULL; | |
2737 | ||
2738 | rbio->operation = BTRFS_RBIO_REBUILD_MISSING; | |
2739 | bio_list_add(&rbio->bio_list, bio); | |
2740 | /* | |
2741 | * This is a special bio which is used to hold the completion handler | |
2742 | * and make the scrub rbio is similar to the other types | |
2743 | */ | |
2744 | ASSERT(!bio->bi_iter.bi_size); | |
2745 | ||
2746 | rbio->faila = find_logical_bio_stripe(rbio, bio); | |
2747 | if (rbio->faila == -1) { | |
f15fb2cd QW |
2748 | btrfs_warn_rl(fs_info, |
2749 | "can not determine the failed stripe number for full stripe %llu", | |
2750 | bioc->raid_map[0]); | |
ff2b64a2 | 2751 | free_raid_bio(rbio); |
b4ee1782 OS |
2752 | return NULL; |
2753 | } | |
2754 | ||
2755 | return rbio; | |
2756 | } | |
2757 | ||
b4ee1782 OS |
2758 | void raid56_submit_missing_rbio(struct btrfs_raid_bio *rbio) |
2759 | { | |
2760 | if (!lock_stripe_add(rbio)) | |
e66d8d5a | 2761 | start_async_work(rbio, read_rebuild_work); |
b4ee1782 | 2762 | } |