]>
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> |
53b381b3 | 16 | #include "ctree.h" |
53b381b3 | 17 | #include "disk-io.h" |
53b381b3 DW |
18 | #include "volumes.h" |
19 | #include "raid56.h" | |
20 | #include "async-thread.h" | |
53b381b3 DW |
21 | |
22 | /* set when additional merges to this rbio are not allowed */ | |
23 | #define RBIO_RMW_LOCKED_BIT 1 | |
24 | ||
4ae10b3a CM |
25 | /* |
26 | * set when this rbio is sitting in the hash, but it is just a cache | |
27 | * of past RMW | |
28 | */ | |
29 | #define RBIO_CACHE_BIT 2 | |
30 | ||
31 | /* | |
32 | * set when it is safe to trust the stripe_pages for caching | |
33 | */ | |
34 | #define RBIO_CACHE_READY_BIT 3 | |
35 | ||
4ae10b3a CM |
36 | #define RBIO_CACHE_SIZE 1024 |
37 | ||
8a953348 DS |
38 | #define BTRFS_STRIPE_HASH_TABLE_BITS 11 |
39 | ||
40 | /* Used by the raid56 code to lock stripes for read/modify/write */ | |
41 | struct btrfs_stripe_hash { | |
42 | struct list_head hash_list; | |
43 | spinlock_t lock; | |
44 | }; | |
45 | ||
46 | /* Used by the raid56 code to lock stripes for read/modify/write */ | |
47 | struct btrfs_stripe_hash_table { | |
48 | struct list_head stripe_cache; | |
49 | spinlock_t cache_lock; | |
50 | int cache_size; | |
51 | struct btrfs_stripe_hash table[]; | |
52 | }; | |
53 | ||
1b94b556 | 54 | enum btrfs_rbio_ops { |
b4ee1782 OS |
55 | BTRFS_RBIO_WRITE, |
56 | BTRFS_RBIO_READ_REBUILD, | |
57 | BTRFS_RBIO_PARITY_SCRUB, | |
58 | BTRFS_RBIO_REBUILD_MISSING, | |
1b94b556 MX |
59 | }; |
60 | ||
53b381b3 DW |
61 | struct btrfs_raid_bio { |
62 | struct btrfs_fs_info *fs_info; | |
63 | struct btrfs_bio *bbio; | |
64 | ||
53b381b3 DW |
65 | /* while we're doing rmw on a stripe |
66 | * we put it into a hash table so we can | |
67 | * lock the stripe and merge more rbios | |
68 | * into it. | |
69 | */ | |
70 | struct list_head hash_list; | |
71 | ||
4ae10b3a CM |
72 | /* |
73 | * LRU list for the stripe cache | |
74 | */ | |
75 | struct list_head stripe_cache; | |
76 | ||
53b381b3 DW |
77 | /* |
78 | * for scheduling work in the helper threads | |
79 | */ | |
80 | struct btrfs_work work; | |
81 | ||
82 | /* | |
83 | * bio list and bio_list_lock are used | |
84 | * to add more bios into the stripe | |
85 | * in hopes of avoiding the full rmw | |
86 | */ | |
87 | struct bio_list bio_list; | |
88 | spinlock_t bio_list_lock; | |
89 | ||
6ac0f488 CM |
90 | /* also protected by the bio_list_lock, the |
91 | * plug list is used by the plugging code | |
92 | * to collect partial bios while plugged. The | |
93 | * stripe locking code also uses it to hand off | |
53b381b3 DW |
94 | * the stripe lock to the next pending IO |
95 | */ | |
96 | struct list_head plug_list; | |
97 | ||
98 | /* | |
99 | * flags that tell us if it is safe to | |
100 | * merge with this bio | |
101 | */ | |
102 | unsigned long flags; | |
103 | ||
104 | /* size of each individual stripe on disk */ | |
105 | int stripe_len; | |
106 | ||
107 | /* number of data stripes (no p/q) */ | |
108 | int nr_data; | |
109 | ||
2c8cdd6e MX |
110 | int real_stripes; |
111 | ||
5a6ac9ea | 112 | int stripe_npages; |
53b381b3 DW |
113 | /* |
114 | * set if we're doing a parity rebuild | |
115 | * for a read from higher up, which is handled | |
116 | * differently from a parity rebuild as part of | |
117 | * rmw | |
118 | */ | |
1b94b556 | 119 | enum btrfs_rbio_ops operation; |
53b381b3 DW |
120 | |
121 | /* first bad stripe */ | |
122 | int faila; | |
123 | ||
124 | /* second bad stripe (for raid6 use) */ | |
125 | int failb; | |
126 | ||
5a6ac9ea | 127 | int scrubp; |
53b381b3 DW |
128 | /* |
129 | * number of pages needed to represent the full | |
130 | * stripe | |
131 | */ | |
132 | int nr_pages; | |
133 | ||
134 | /* | |
135 | * size of all the bios in the bio_list. This | |
136 | * helps us decide if the rbio maps to a full | |
137 | * stripe or not | |
138 | */ | |
139 | int bio_list_bytes; | |
140 | ||
4245215d MX |
141 | int generic_bio_cnt; |
142 | ||
dec95574 | 143 | refcount_t refs; |
53b381b3 | 144 | |
b89e1b01 MX |
145 | atomic_t stripes_pending; |
146 | ||
147 | atomic_t error; | |
53b381b3 DW |
148 | /* |
149 | * these are two arrays of pointers. We allocate the | |
150 | * rbio big enough to hold them both and setup their | |
151 | * locations when the rbio is allocated | |
152 | */ | |
153 | ||
154 | /* pointers to pages that we allocated for | |
155 | * reading/writing stripes directly from the disk (including P/Q) | |
156 | */ | |
157 | struct page **stripe_pages; | |
158 | ||
159 | /* | |
160 | * pointers to the pages in the bio_list. Stored | |
161 | * here for faster lookup | |
162 | */ | |
163 | struct page **bio_pages; | |
5a6ac9ea MX |
164 | |
165 | /* | |
166 | * bitmap to record which horizontal stripe has data | |
167 | */ | |
168 | unsigned long *dbitmap; | |
1389053e KC |
169 | |
170 | /* allocated with real_stripes-many pointers for finish_*() calls */ | |
171 | void **finish_pointers; | |
172 | ||
173 | /* allocated with stripe_npages-many bits for finish_*() calls */ | |
174 | unsigned long *finish_pbitmap; | |
53b381b3 DW |
175 | }; |
176 | ||
177 | static int __raid56_parity_recover(struct btrfs_raid_bio *rbio); | |
178 | static noinline void finish_rmw(struct btrfs_raid_bio *rbio); | |
179 | static void rmw_work(struct btrfs_work *work); | |
180 | static void read_rebuild_work(struct btrfs_work *work); | |
53b381b3 DW |
181 | static int fail_bio_stripe(struct btrfs_raid_bio *rbio, struct bio *bio); |
182 | static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed); | |
183 | static void __free_raid_bio(struct btrfs_raid_bio *rbio); | |
184 | static void index_rbio_pages(struct btrfs_raid_bio *rbio); | |
185 | static int alloc_rbio_pages(struct btrfs_raid_bio *rbio); | |
186 | ||
5a6ac9ea MX |
187 | static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, |
188 | int need_check); | |
a81b747d | 189 | static void scrub_parity_work(struct btrfs_work *work); |
5a6ac9ea | 190 | |
ac638859 DS |
191 | static void start_async_work(struct btrfs_raid_bio *rbio, btrfs_func_t work_func) |
192 | { | |
a0cac0ec | 193 | btrfs_init_work(&rbio->work, work_func, NULL, NULL); |
ac638859 DS |
194 | btrfs_queue_work(rbio->fs_info->rmw_workers, &rbio->work); |
195 | } | |
196 | ||
53b381b3 DW |
197 | /* |
198 | * the stripe hash table is used for locking, and to collect | |
199 | * bios in hopes of making a full stripe | |
200 | */ | |
201 | int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info) | |
202 | { | |
203 | struct btrfs_stripe_hash_table *table; | |
204 | struct btrfs_stripe_hash_table *x; | |
205 | struct btrfs_stripe_hash *cur; | |
206 | struct btrfs_stripe_hash *h; | |
207 | int num_entries = 1 << BTRFS_STRIPE_HASH_TABLE_BITS; | |
208 | int i; | |
83c8266a | 209 | int table_size; |
53b381b3 DW |
210 | |
211 | if (info->stripe_hash_table) | |
212 | return 0; | |
213 | ||
83c8266a DS |
214 | /* |
215 | * The table is large, starting with order 4 and can go as high as | |
216 | * order 7 in case lock debugging is turned on. | |
217 | * | |
218 | * Try harder to allocate and fallback to vmalloc to lower the chance | |
219 | * of a failing mount. | |
220 | */ | |
221 | table_size = sizeof(*table) + sizeof(*h) * num_entries; | |
818e010b DS |
222 | table = kvzalloc(table_size, GFP_KERNEL); |
223 | if (!table) | |
224 | return -ENOMEM; | |
53b381b3 | 225 | |
4ae10b3a CM |
226 | spin_lock_init(&table->cache_lock); |
227 | INIT_LIST_HEAD(&table->stripe_cache); | |
228 | ||
53b381b3 DW |
229 | h = table->table; |
230 | ||
231 | for (i = 0; i < num_entries; i++) { | |
232 | cur = h + i; | |
233 | INIT_LIST_HEAD(&cur->hash_list); | |
234 | spin_lock_init(&cur->lock); | |
53b381b3 DW |
235 | } |
236 | ||
237 | x = cmpxchg(&info->stripe_hash_table, NULL, table); | |
f749303b WS |
238 | if (x) |
239 | kvfree(x); | |
53b381b3 DW |
240 | return 0; |
241 | } | |
242 | ||
4ae10b3a CM |
243 | /* |
244 | * caching an rbio means to copy anything from the | |
245 | * bio_pages array into the stripe_pages array. We | |
246 | * use the page uptodate bit in the stripe cache array | |
247 | * to indicate if it has valid data | |
248 | * | |
249 | * once the caching is done, we set the cache ready | |
250 | * bit. | |
251 | */ | |
252 | static void cache_rbio_pages(struct btrfs_raid_bio *rbio) | |
253 | { | |
254 | int i; | |
255 | char *s; | |
256 | char *d; | |
257 | int ret; | |
258 | ||
259 | ret = alloc_rbio_pages(rbio); | |
260 | if (ret) | |
261 | return; | |
262 | ||
263 | for (i = 0; i < rbio->nr_pages; i++) { | |
264 | if (!rbio->bio_pages[i]) | |
265 | continue; | |
266 | ||
267 | s = kmap(rbio->bio_pages[i]); | |
268 | d = kmap(rbio->stripe_pages[i]); | |
269 | ||
69d24804 | 270 | copy_page(d, s); |
4ae10b3a CM |
271 | |
272 | kunmap(rbio->bio_pages[i]); | |
273 | kunmap(rbio->stripe_pages[i]); | |
274 | SetPageUptodate(rbio->stripe_pages[i]); | |
275 | } | |
276 | set_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
277 | } | |
278 | ||
53b381b3 DW |
279 | /* |
280 | * we hash on the first logical address of the stripe | |
281 | */ | |
282 | static int rbio_bucket(struct btrfs_raid_bio *rbio) | |
283 | { | |
8e5cfb55 | 284 | u64 num = rbio->bbio->raid_map[0]; |
53b381b3 DW |
285 | |
286 | /* | |
287 | * we shift down quite a bit. We're using byte | |
288 | * addressing, and most of the lower bits are zeros. | |
289 | * This tends to upset hash_64, and it consistently | |
290 | * returns just one or two different values. | |
291 | * | |
292 | * shifting off the lower bits fixes things. | |
293 | */ | |
294 | return hash_64(num >> 16, BTRFS_STRIPE_HASH_TABLE_BITS); | |
295 | } | |
296 | ||
4ae10b3a CM |
297 | /* |
298 | * stealing an rbio means taking all the uptodate pages from the stripe | |
299 | * array in the source rbio and putting them into the destination rbio | |
300 | */ | |
301 | static void steal_rbio(struct btrfs_raid_bio *src, struct btrfs_raid_bio *dest) | |
302 | { | |
303 | int i; | |
304 | struct page *s; | |
305 | struct page *d; | |
306 | ||
307 | if (!test_bit(RBIO_CACHE_READY_BIT, &src->flags)) | |
308 | return; | |
309 | ||
310 | for (i = 0; i < dest->nr_pages; i++) { | |
311 | s = src->stripe_pages[i]; | |
312 | if (!s || !PageUptodate(s)) { | |
313 | continue; | |
314 | } | |
315 | ||
316 | d = dest->stripe_pages[i]; | |
317 | if (d) | |
318 | __free_page(d); | |
319 | ||
320 | dest->stripe_pages[i] = s; | |
321 | src->stripe_pages[i] = NULL; | |
322 | } | |
323 | } | |
324 | ||
53b381b3 DW |
325 | /* |
326 | * merging means we take the bio_list from the victim and | |
327 | * splice it into the destination. The victim should | |
328 | * be discarded afterwards. | |
329 | * | |
330 | * must be called with dest->rbio_list_lock held | |
331 | */ | |
332 | static void merge_rbio(struct btrfs_raid_bio *dest, | |
333 | struct btrfs_raid_bio *victim) | |
334 | { | |
335 | bio_list_merge(&dest->bio_list, &victim->bio_list); | |
336 | dest->bio_list_bytes += victim->bio_list_bytes; | |
4245215d | 337 | dest->generic_bio_cnt += victim->generic_bio_cnt; |
53b381b3 DW |
338 | bio_list_init(&victim->bio_list); |
339 | } | |
340 | ||
341 | /* | |
4ae10b3a CM |
342 | * used to prune items that are in the cache. The caller |
343 | * must hold the hash table lock. | |
344 | */ | |
345 | static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio) | |
346 | { | |
347 | int bucket = rbio_bucket(rbio); | |
348 | struct btrfs_stripe_hash_table *table; | |
349 | struct btrfs_stripe_hash *h; | |
350 | int freeit = 0; | |
351 | ||
352 | /* | |
353 | * check the bit again under the hash table lock. | |
354 | */ | |
355 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
356 | return; | |
357 | ||
358 | table = rbio->fs_info->stripe_hash_table; | |
359 | h = table->table + bucket; | |
360 | ||
361 | /* hold the lock for the bucket because we may be | |
362 | * removing it from the hash table | |
363 | */ | |
364 | spin_lock(&h->lock); | |
365 | ||
366 | /* | |
367 | * hold the lock for the bio list because we need | |
368 | * to make sure the bio list is empty | |
369 | */ | |
370 | spin_lock(&rbio->bio_list_lock); | |
371 | ||
372 | if (test_and_clear_bit(RBIO_CACHE_BIT, &rbio->flags)) { | |
373 | list_del_init(&rbio->stripe_cache); | |
374 | table->cache_size -= 1; | |
375 | freeit = 1; | |
376 | ||
377 | /* if the bio list isn't empty, this rbio is | |
378 | * still involved in an IO. We take it out | |
379 | * of the cache list, and drop the ref that | |
380 | * was held for the list. | |
381 | * | |
382 | * If the bio_list was empty, we also remove | |
383 | * the rbio from the hash_table, and drop | |
384 | * the corresponding ref | |
385 | */ | |
386 | if (bio_list_empty(&rbio->bio_list)) { | |
387 | if (!list_empty(&rbio->hash_list)) { | |
388 | list_del_init(&rbio->hash_list); | |
dec95574 | 389 | refcount_dec(&rbio->refs); |
4ae10b3a CM |
390 | BUG_ON(!list_empty(&rbio->plug_list)); |
391 | } | |
392 | } | |
393 | } | |
394 | ||
395 | spin_unlock(&rbio->bio_list_lock); | |
396 | spin_unlock(&h->lock); | |
397 | ||
398 | if (freeit) | |
399 | __free_raid_bio(rbio); | |
400 | } | |
401 | ||
402 | /* | |
403 | * prune a given rbio from the cache | |
404 | */ | |
405 | static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio) | |
406 | { | |
407 | struct btrfs_stripe_hash_table *table; | |
408 | unsigned long flags; | |
409 | ||
410 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
411 | return; | |
412 | ||
413 | table = rbio->fs_info->stripe_hash_table; | |
414 | ||
415 | spin_lock_irqsave(&table->cache_lock, flags); | |
416 | __remove_rbio_from_cache(rbio); | |
417 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
418 | } | |
419 | ||
420 | /* | |
421 | * remove everything in the cache | |
422 | */ | |
48a3b636 | 423 | static void btrfs_clear_rbio_cache(struct btrfs_fs_info *info) |
4ae10b3a CM |
424 | { |
425 | struct btrfs_stripe_hash_table *table; | |
426 | unsigned long flags; | |
427 | struct btrfs_raid_bio *rbio; | |
428 | ||
429 | table = info->stripe_hash_table; | |
430 | ||
431 | spin_lock_irqsave(&table->cache_lock, flags); | |
432 | while (!list_empty(&table->stripe_cache)) { | |
433 | rbio = list_entry(table->stripe_cache.next, | |
434 | struct btrfs_raid_bio, | |
435 | stripe_cache); | |
436 | __remove_rbio_from_cache(rbio); | |
437 | } | |
438 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
439 | } | |
440 | ||
441 | /* | |
442 | * remove all cached entries and free the hash table | |
443 | * used by unmount | |
53b381b3 DW |
444 | */ |
445 | void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info) | |
446 | { | |
447 | if (!info->stripe_hash_table) | |
448 | return; | |
4ae10b3a | 449 | btrfs_clear_rbio_cache(info); |
f749303b | 450 | kvfree(info->stripe_hash_table); |
53b381b3 DW |
451 | info->stripe_hash_table = NULL; |
452 | } | |
453 | ||
4ae10b3a CM |
454 | /* |
455 | * insert an rbio into the stripe cache. It | |
456 | * must have already been prepared by calling | |
457 | * cache_rbio_pages | |
458 | * | |
459 | * If this rbio was already cached, it gets | |
460 | * moved to the front of the lru. | |
461 | * | |
462 | * If the size of the rbio cache is too big, we | |
463 | * prune an item. | |
464 | */ | |
465 | static void cache_rbio(struct btrfs_raid_bio *rbio) | |
466 | { | |
467 | struct btrfs_stripe_hash_table *table; | |
468 | unsigned long flags; | |
469 | ||
470 | if (!test_bit(RBIO_CACHE_READY_BIT, &rbio->flags)) | |
471 | return; | |
472 | ||
473 | table = rbio->fs_info->stripe_hash_table; | |
474 | ||
475 | spin_lock_irqsave(&table->cache_lock, flags); | |
476 | spin_lock(&rbio->bio_list_lock); | |
477 | ||
478 | /* bump our ref if we were not in the list before */ | |
479 | if (!test_and_set_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
dec95574 | 480 | refcount_inc(&rbio->refs); |
4ae10b3a CM |
481 | |
482 | if (!list_empty(&rbio->stripe_cache)){ | |
483 | list_move(&rbio->stripe_cache, &table->stripe_cache); | |
484 | } else { | |
485 | list_add(&rbio->stripe_cache, &table->stripe_cache); | |
486 | table->cache_size += 1; | |
487 | } | |
488 | ||
489 | spin_unlock(&rbio->bio_list_lock); | |
490 | ||
491 | if (table->cache_size > RBIO_CACHE_SIZE) { | |
492 | struct btrfs_raid_bio *found; | |
493 | ||
494 | found = list_entry(table->stripe_cache.prev, | |
495 | struct btrfs_raid_bio, | |
496 | stripe_cache); | |
497 | ||
498 | if (found != rbio) | |
499 | __remove_rbio_from_cache(found); | |
500 | } | |
501 | ||
502 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
4ae10b3a CM |
503 | } |
504 | ||
53b381b3 DW |
505 | /* |
506 | * helper function to run the xor_blocks api. It is only | |
507 | * able to do MAX_XOR_BLOCKS at a time, so we need to | |
508 | * loop through. | |
509 | */ | |
510 | static void run_xor(void **pages, int src_cnt, ssize_t len) | |
511 | { | |
512 | int src_off = 0; | |
513 | int xor_src_cnt = 0; | |
514 | void *dest = pages[src_cnt]; | |
515 | ||
516 | while(src_cnt > 0) { | |
517 | xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS); | |
518 | xor_blocks(xor_src_cnt, len, dest, pages + src_off); | |
519 | ||
520 | src_cnt -= xor_src_cnt; | |
521 | src_off += xor_src_cnt; | |
522 | } | |
523 | } | |
524 | ||
525 | /* | |
176571a1 DS |
526 | * Returns true if the bio list inside this rbio covers an entire stripe (no |
527 | * rmw required). | |
53b381b3 | 528 | */ |
176571a1 | 529 | static int rbio_is_full(struct btrfs_raid_bio *rbio) |
53b381b3 | 530 | { |
176571a1 | 531 | unsigned long flags; |
53b381b3 DW |
532 | unsigned long size = rbio->bio_list_bytes; |
533 | int ret = 1; | |
534 | ||
176571a1 | 535 | spin_lock_irqsave(&rbio->bio_list_lock, flags); |
53b381b3 DW |
536 | if (size != rbio->nr_data * rbio->stripe_len) |
537 | ret = 0; | |
53b381b3 | 538 | BUG_ON(size > rbio->nr_data * rbio->stripe_len); |
53b381b3 | 539 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); |
176571a1 | 540 | |
53b381b3 DW |
541 | return ret; |
542 | } | |
543 | ||
544 | /* | |
545 | * returns 1 if it is safe to merge two rbios together. | |
546 | * The merging is safe if the two rbios correspond to | |
547 | * the same stripe and if they are both going in the same | |
548 | * direction (read vs write), and if neither one is | |
549 | * locked for final IO | |
550 | * | |
551 | * The caller is responsible for locking such that | |
552 | * rmw_locked is safe to test | |
553 | */ | |
554 | static int rbio_can_merge(struct btrfs_raid_bio *last, | |
555 | struct btrfs_raid_bio *cur) | |
556 | { | |
557 | if (test_bit(RBIO_RMW_LOCKED_BIT, &last->flags) || | |
558 | test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) | |
559 | return 0; | |
560 | ||
4ae10b3a CM |
561 | /* |
562 | * we can't merge with cached rbios, since the | |
563 | * idea is that when we merge the destination | |
564 | * rbio is going to run our IO for us. We can | |
01327610 | 565 | * steal from cached rbios though, other functions |
4ae10b3a CM |
566 | * handle that. |
567 | */ | |
568 | if (test_bit(RBIO_CACHE_BIT, &last->flags) || | |
569 | test_bit(RBIO_CACHE_BIT, &cur->flags)) | |
570 | return 0; | |
571 | ||
8e5cfb55 ZL |
572 | if (last->bbio->raid_map[0] != |
573 | cur->bbio->raid_map[0]) | |
53b381b3 DW |
574 | return 0; |
575 | ||
5a6ac9ea MX |
576 | /* we can't merge with different operations */ |
577 | if (last->operation != cur->operation) | |
578 | return 0; | |
579 | /* | |
580 | * We've need read the full stripe from the drive. | |
581 | * check and repair the parity and write the new results. | |
582 | * | |
583 | * We're not allowed to add any new bios to the | |
584 | * bio list here, anyone else that wants to | |
585 | * change this stripe needs to do their own rmw. | |
586 | */ | |
db34be19 | 587 | if (last->operation == BTRFS_RBIO_PARITY_SCRUB) |
53b381b3 | 588 | return 0; |
53b381b3 | 589 | |
db34be19 | 590 | if (last->operation == BTRFS_RBIO_REBUILD_MISSING) |
b4ee1782 OS |
591 | return 0; |
592 | ||
cc54ff62 LB |
593 | if (last->operation == BTRFS_RBIO_READ_REBUILD) { |
594 | int fa = last->faila; | |
595 | int fb = last->failb; | |
596 | int cur_fa = cur->faila; | |
597 | int cur_fb = cur->failb; | |
598 | ||
599 | if (last->faila >= last->failb) { | |
600 | fa = last->failb; | |
601 | fb = last->faila; | |
602 | } | |
603 | ||
604 | if (cur->faila >= cur->failb) { | |
605 | cur_fa = cur->failb; | |
606 | cur_fb = cur->faila; | |
607 | } | |
608 | ||
609 | if (fa != cur_fa || fb != cur_fb) | |
610 | return 0; | |
611 | } | |
53b381b3 DW |
612 | return 1; |
613 | } | |
614 | ||
b7178a5f ZL |
615 | static int rbio_stripe_page_index(struct btrfs_raid_bio *rbio, int stripe, |
616 | int index) | |
617 | { | |
618 | return stripe * rbio->stripe_npages + index; | |
619 | } | |
620 | ||
621 | /* | |
622 | * these are just the pages from the rbio array, not from anything | |
623 | * the FS sent down to us | |
624 | */ | |
625 | static struct page *rbio_stripe_page(struct btrfs_raid_bio *rbio, int stripe, | |
626 | int index) | |
627 | { | |
628 | return rbio->stripe_pages[rbio_stripe_page_index(rbio, stripe, index)]; | |
629 | } | |
630 | ||
53b381b3 DW |
631 | /* |
632 | * helper to index into the pstripe | |
633 | */ | |
634 | static struct page *rbio_pstripe_page(struct btrfs_raid_bio *rbio, int index) | |
635 | { | |
b7178a5f | 636 | return rbio_stripe_page(rbio, rbio->nr_data, index); |
53b381b3 DW |
637 | } |
638 | ||
639 | /* | |
640 | * helper to index into the qstripe, returns null | |
641 | * if there is no qstripe | |
642 | */ | |
643 | static struct page *rbio_qstripe_page(struct btrfs_raid_bio *rbio, int index) | |
644 | { | |
2c8cdd6e | 645 | if (rbio->nr_data + 1 == rbio->real_stripes) |
53b381b3 | 646 | return NULL; |
b7178a5f | 647 | return rbio_stripe_page(rbio, rbio->nr_data + 1, index); |
53b381b3 DW |
648 | } |
649 | ||
650 | /* | |
651 | * The first stripe in the table for a logical address | |
652 | * has the lock. rbios are added in one of three ways: | |
653 | * | |
654 | * 1) Nobody has the stripe locked yet. The rbio is given | |
655 | * the lock and 0 is returned. The caller must start the IO | |
656 | * themselves. | |
657 | * | |
658 | * 2) Someone has the stripe locked, but we're able to merge | |
659 | * with the lock owner. The rbio is freed and the IO will | |
660 | * start automatically along with the existing rbio. 1 is returned. | |
661 | * | |
662 | * 3) Someone has the stripe locked, but we're not able to merge. | |
663 | * The rbio is added to the lock owner's plug list, or merged into | |
664 | * an rbio already on the plug list. When the lock owner unlocks, | |
665 | * the next rbio on the list is run and the IO is started automatically. | |
666 | * 1 is returned | |
667 | * | |
668 | * If we return 0, the caller still owns the rbio and must continue with | |
669 | * IO submission. If we return 1, the caller must assume the rbio has | |
670 | * already been freed. | |
671 | */ | |
672 | static noinline int lock_stripe_add(struct btrfs_raid_bio *rbio) | |
673 | { | |
721860d5 | 674 | struct btrfs_stripe_hash *h; |
53b381b3 DW |
675 | struct btrfs_raid_bio *cur; |
676 | struct btrfs_raid_bio *pending; | |
677 | unsigned long flags; | |
53b381b3 | 678 | struct btrfs_raid_bio *freeit = NULL; |
4ae10b3a | 679 | struct btrfs_raid_bio *cache_drop = NULL; |
53b381b3 | 680 | int ret = 0; |
53b381b3 | 681 | |
721860d5 JT |
682 | h = rbio->fs_info->stripe_hash_table->table + rbio_bucket(rbio); |
683 | ||
53b381b3 DW |
684 | spin_lock_irqsave(&h->lock, flags); |
685 | list_for_each_entry(cur, &h->hash_list, hash_list) { | |
9d6cb1b0 JT |
686 | if (cur->bbio->raid_map[0] != rbio->bbio->raid_map[0]) |
687 | continue; | |
4ae10b3a | 688 | |
9d6cb1b0 | 689 | spin_lock(&cur->bio_list_lock); |
4ae10b3a | 690 | |
9d6cb1b0 JT |
691 | /* Can we steal this cached rbio's pages? */ |
692 | if (bio_list_empty(&cur->bio_list) && | |
693 | list_empty(&cur->plug_list) && | |
694 | test_bit(RBIO_CACHE_BIT, &cur->flags) && | |
695 | !test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) { | |
696 | list_del_init(&cur->hash_list); | |
697 | refcount_dec(&cur->refs); | |
53b381b3 | 698 | |
9d6cb1b0 JT |
699 | steal_rbio(cur, rbio); |
700 | cache_drop = cur; | |
701 | spin_unlock(&cur->bio_list_lock); | |
4ae10b3a | 702 | |
9d6cb1b0 JT |
703 | goto lockit; |
704 | } | |
53b381b3 | 705 | |
9d6cb1b0 JT |
706 | /* Can we merge into the lock owner? */ |
707 | if (rbio_can_merge(cur, rbio)) { | |
708 | merge_rbio(cur, rbio); | |
53b381b3 | 709 | spin_unlock(&cur->bio_list_lock); |
9d6cb1b0 | 710 | freeit = rbio; |
53b381b3 DW |
711 | ret = 1; |
712 | goto out; | |
713 | } | |
9d6cb1b0 JT |
714 | |
715 | ||
716 | /* | |
717 | * We couldn't merge with the running rbio, see if we can merge | |
718 | * with the pending ones. We don't have to check for rmw_locked | |
719 | * because there is no way they are inside finish_rmw right now | |
720 | */ | |
721 | list_for_each_entry(pending, &cur->plug_list, plug_list) { | |
722 | if (rbio_can_merge(pending, rbio)) { | |
723 | merge_rbio(pending, rbio); | |
724 | spin_unlock(&cur->bio_list_lock); | |
725 | freeit = rbio; | |
726 | ret = 1; | |
727 | goto out; | |
728 | } | |
729 | } | |
730 | ||
731 | /* | |
732 | * No merging, put us on the tail of the plug list, our rbio | |
733 | * will be started with the currently running rbio unlocks | |
734 | */ | |
735 | list_add_tail(&rbio->plug_list, &cur->plug_list); | |
736 | spin_unlock(&cur->bio_list_lock); | |
737 | ret = 1; | |
738 | goto out; | |
53b381b3 | 739 | } |
4ae10b3a | 740 | lockit: |
dec95574 | 741 | refcount_inc(&rbio->refs); |
53b381b3 DW |
742 | list_add(&rbio->hash_list, &h->hash_list); |
743 | out: | |
744 | spin_unlock_irqrestore(&h->lock, flags); | |
4ae10b3a CM |
745 | if (cache_drop) |
746 | remove_rbio_from_cache(cache_drop); | |
53b381b3 DW |
747 | if (freeit) |
748 | __free_raid_bio(freeit); | |
749 | return ret; | |
750 | } | |
751 | ||
752 | /* | |
753 | * called as rmw or parity rebuild is completed. If the plug list has more | |
754 | * rbios waiting for this stripe, the next one on the list will be started | |
755 | */ | |
756 | static noinline void unlock_stripe(struct btrfs_raid_bio *rbio) | |
757 | { | |
758 | int bucket; | |
759 | struct btrfs_stripe_hash *h; | |
760 | unsigned long flags; | |
4ae10b3a | 761 | int keep_cache = 0; |
53b381b3 DW |
762 | |
763 | bucket = rbio_bucket(rbio); | |
764 | h = rbio->fs_info->stripe_hash_table->table + bucket; | |
765 | ||
4ae10b3a CM |
766 | if (list_empty(&rbio->plug_list)) |
767 | cache_rbio(rbio); | |
768 | ||
53b381b3 DW |
769 | spin_lock_irqsave(&h->lock, flags); |
770 | spin_lock(&rbio->bio_list_lock); | |
771 | ||
772 | if (!list_empty(&rbio->hash_list)) { | |
4ae10b3a CM |
773 | /* |
774 | * if we're still cached and there is no other IO | |
775 | * to perform, just leave this rbio here for others | |
776 | * to steal from later | |
777 | */ | |
778 | if (list_empty(&rbio->plug_list) && | |
779 | test_bit(RBIO_CACHE_BIT, &rbio->flags)) { | |
780 | keep_cache = 1; | |
781 | clear_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
782 | BUG_ON(!bio_list_empty(&rbio->bio_list)); | |
783 | goto done; | |
784 | } | |
53b381b3 DW |
785 | |
786 | list_del_init(&rbio->hash_list); | |
dec95574 | 787 | refcount_dec(&rbio->refs); |
53b381b3 DW |
788 | |
789 | /* | |
790 | * we use the plug list to hold all the rbios | |
791 | * waiting for the chance to lock this stripe. | |
792 | * hand the lock over to one of them. | |
793 | */ | |
794 | if (!list_empty(&rbio->plug_list)) { | |
795 | struct btrfs_raid_bio *next; | |
796 | struct list_head *head = rbio->plug_list.next; | |
797 | ||
798 | next = list_entry(head, struct btrfs_raid_bio, | |
799 | plug_list); | |
800 | ||
801 | list_del_init(&rbio->plug_list); | |
802 | ||
803 | list_add(&next->hash_list, &h->hash_list); | |
dec95574 | 804 | refcount_inc(&next->refs); |
53b381b3 DW |
805 | spin_unlock(&rbio->bio_list_lock); |
806 | spin_unlock_irqrestore(&h->lock, flags); | |
807 | ||
1b94b556 | 808 | if (next->operation == BTRFS_RBIO_READ_REBUILD) |
e66d8d5a | 809 | start_async_work(next, read_rebuild_work); |
b4ee1782 OS |
810 | else if (next->operation == BTRFS_RBIO_REBUILD_MISSING) { |
811 | steal_rbio(rbio, next); | |
e66d8d5a | 812 | start_async_work(next, read_rebuild_work); |
b4ee1782 | 813 | } else if (next->operation == BTRFS_RBIO_WRITE) { |
4ae10b3a | 814 | steal_rbio(rbio, next); |
cf6a4a75 | 815 | start_async_work(next, rmw_work); |
5a6ac9ea MX |
816 | } else if (next->operation == BTRFS_RBIO_PARITY_SCRUB) { |
817 | steal_rbio(rbio, next); | |
a81b747d | 818 | start_async_work(next, scrub_parity_work); |
4ae10b3a | 819 | } |
53b381b3 DW |
820 | |
821 | goto done_nolock; | |
53b381b3 DW |
822 | } |
823 | } | |
4ae10b3a | 824 | done: |
53b381b3 DW |
825 | spin_unlock(&rbio->bio_list_lock); |
826 | spin_unlock_irqrestore(&h->lock, flags); | |
827 | ||
828 | done_nolock: | |
4ae10b3a CM |
829 | if (!keep_cache) |
830 | remove_rbio_from_cache(rbio); | |
53b381b3 DW |
831 | } |
832 | ||
833 | static void __free_raid_bio(struct btrfs_raid_bio *rbio) | |
834 | { | |
835 | int i; | |
836 | ||
dec95574 | 837 | if (!refcount_dec_and_test(&rbio->refs)) |
53b381b3 DW |
838 | return; |
839 | ||
4ae10b3a | 840 | WARN_ON(!list_empty(&rbio->stripe_cache)); |
53b381b3 DW |
841 | WARN_ON(!list_empty(&rbio->hash_list)); |
842 | WARN_ON(!bio_list_empty(&rbio->bio_list)); | |
843 | ||
844 | for (i = 0; i < rbio->nr_pages; i++) { | |
845 | if (rbio->stripe_pages[i]) { | |
846 | __free_page(rbio->stripe_pages[i]); | |
847 | rbio->stripe_pages[i] = NULL; | |
848 | } | |
849 | } | |
af8e2d1d | 850 | |
6e9606d2 | 851 | btrfs_put_bbio(rbio->bbio); |
53b381b3 DW |
852 | kfree(rbio); |
853 | } | |
854 | ||
7583d8d0 | 855 | static void rbio_endio_bio_list(struct bio *cur, blk_status_t err) |
53b381b3 | 856 | { |
7583d8d0 LB |
857 | struct bio *next; |
858 | ||
859 | while (cur) { | |
860 | next = cur->bi_next; | |
861 | cur->bi_next = NULL; | |
862 | cur->bi_status = err; | |
863 | bio_endio(cur); | |
864 | cur = next; | |
865 | } | |
53b381b3 DW |
866 | } |
867 | ||
868 | /* | |
869 | * this frees the rbio and runs through all the bios in the | |
870 | * bio_list and calls end_io on them | |
871 | */ | |
4e4cbee9 | 872 | static void rbio_orig_end_io(struct btrfs_raid_bio *rbio, blk_status_t err) |
53b381b3 DW |
873 | { |
874 | struct bio *cur = bio_list_get(&rbio->bio_list); | |
7583d8d0 | 875 | struct bio *extra; |
4245215d MX |
876 | |
877 | if (rbio->generic_bio_cnt) | |
878 | btrfs_bio_counter_sub(rbio->fs_info, rbio->generic_bio_cnt); | |
879 | ||
7583d8d0 LB |
880 | /* |
881 | * At this moment, rbio->bio_list is empty, however since rbio does not | |
882 | * always have RBIO_RMW_LOCKED_BIT set and rbio is still linked on the | |
883 | * hash list, rbio may be merged with others so that rbio->bio_list | |
884 | * becomes non-empty. | |
885 | * Once unlock_stripe() is done, rbio->bio_list will not be updated any | |
886 | * more and we can call bio_endio() on all queued bios. | |
887 | */ | |
888 | unlock_stripe(rbio); | |
889 | extra = bio_list_get(&rbio->bio_list); | |
890 | __free_raid_bio(rbio); | |
53b381b3 | 891 | |
7583d8d0 LB |
892 | rbio_endio_bio_list(cur, err); |
893 | if (extra) | |
894 | rbio_endio_bio_list(extra, err); | |
53b381b3 DW |
895 | } |
896 | ||
897 | /* | |
898 | * end io function used by finish_rmw. When we finally | |
899 | * get here, we've written a full stripe | |
900 | */ | |
4246a0b6 | 901 | static void raid_write_end_io(struct bio *bio) |
53b381b3 DW |
902 | { |
903 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
4e4cbee9 | 904 | blk_status_t err = bio->bi_status; |
a6111d11 | 905 | int max_errors; |
53b381b3 DW |
906 | |
907 | if (err) | |
908 | fail_bio_stripe(rbio, bio); | |
909 | ||
910 | bio_put(bio); | |
911 | ||
b89e1b01 | 912 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
913 | return; |
914 | ||
58efbc9f | 915 | err = BLK_STS_OK; |
53b381b3 DW |
916 | |
917 | /* OK, we have read all the stripes we need to. */ | |
a6111d11 ZL |
918 | max_errors = (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) ? |
919 | 0 : rbio->bbio->max_errors; | |
920 | if (atomic_read(&rbio->error) > max_errors) | |
4e4cbee9 | 921 | err = BLK_STS_IOERR; |
53b381b3 | 922 | |
4246a0b6 | 923 | rbio_orig_end_io(rbio, err); |
53b381b3 DW |
924 | } |
925 | ||
926 | /* | |
927 | * the read/modify/write code wants to use the original bio for | |
928 | * any pages it included, and then use the rbio for everything | |
929 | * else. This function decides if a given index (stripe number) | |
930 | * and page number in that stripe fall inside the original bio | |
931 | * or the rbio. | |
932 | * | |
933 | * if you set bio_list_only, you'll get a NULL back for any ranges | |
934 | * that are outside the bio_list | |
935 | * | |
936 | * This doesn't take any refs on anything, you get a bare page pointer | |
937 | * and the caller must bump refs as required. | |
938 | * | |
939 | * You must call index_rbio_pages once before you can trust | |
940 | * the answers from this function. | |
941 | */ | |
942 | static struct page *page_in_rbio(struct btrfs_raid_bio *rbio, | |
943 | int index, int pagenr, int bio_list_only) | |
944 | { | |
945 | int chunk_page; | |
946 | struct page *p = NULL; | |
947 | ||
948 | chunk_page = index * (rbio->stripe_len >> PAGE_SHIFT) + pagenr; | |
949 | ||
950 | spin_lock_irq(&rbio->bio_list_lock); | |
951 | p = rbio->bio_pages[chunk_page]; | |
952 | spin_unlock_irq(&rbio->bio_list_lock); | |
953 | ||
954 | if (p || bio_list_only) | |
955 | return p; | |
956 | ||
957 | return rbio->stripe_pages[chunk_page]; | |
958 | } | |
959 | ||
960 | /* | |
961 | * number of pages we need for the entire stripe across all the | |
962 | * drives | |
963 | */ | |
964 | static unsigned long rbio_nr_pages(unsigned long stripe_len, int nr_stripes) | |
965 | { | |
09cbfeaf | 966 | return DIV_ROUND_UP(stripe_len, PAGE_SIZE) * nr_stripes; |
53b381b3 DW |
967 | } |
968 | ||
969 | /* | |
970 | * allocation and initial setup for the btrfs_raid_bio. Not | |
971 | * this does not allocate any pages for rbio->pages. | |
972 | */ | |
2ff7e61e JM |
973 | static struct btrfs_raid_bio *alloc_rbio(struct btrfs_fs_info *fs_info, |
974 | struct btrfs_bio *bbio, | |
975 | u64 stripe_len) | |
53b381b3 DW |
976 | { |
977 | struct btrfs_raid_bio *rbio; | |
978 | int nr_data = 0; | |
2c8cdd6e MX |
979 | int real_stripes = bbio->num_stripes - bbio->num_tgtdevs; |
980 | int num_pages = rbio_nr_pages(stripe_len, real_stripes); | |
5a6ac9ea | 981 | int stripe_npages = DIV_ROUND_UP(stripe_len, PAGE_SIZE); |
53b381b3 DW |
982 | void *p; |
983 | ||
1389053e KC |
984 | rbio = kzalloc(sizeof(*rbio) + |
985 | sizeof(*rbio->stripe_pages) * num_pages + | |
986 | sizeof(*rbio->bio_pages) * num_pages + | |
987 | sizeof(*rbio->finish_pointers) * real_stripes + | |
988 | sizeof(*rbio->dbitmap) * BITS_TO_LONGS(stripe_npages) + | |
989 | sizeof(*rbio->finish_pbitmap) * | |
990 | BITS_TO_LONGS(stripe_npages), | |
991 | GFP_NOFS); | |
af8e2d1d | 992 | if (!rbio) |
53b381b3 | 993 | return ERR_PTR(-ENOMEM); |
53b381b3 DW |
994 | |
995 | bio_list_init(&rbio->bio_list); | |
996 | INIT_LIST_HEAD(&rbio->plug_list); | |
997 | spin_lock_init(&rbio->bio_list_lock); | |
4ae10b3a | 998 | INIT_LIST_HEAD(&rbio->stripe_cache); |
53b381b3 DW |
999 | INIT_LIST_HEAD(&rbio->hash_list); |
1000 | rbio->bbio = bbio; | |
2ff7e61e | 1001 | rbio->fs_info = fs_info; |
53b381b3 DW |
1002 | rbio->stripe_len = stripe_len; |
1003 | rbio->nr_pages = num_pages; | |
2c8cdd6e | 1004 | rbio->real_stripes = real_stripes; |
5a6ac9ea | 1005 | rbio->stripe_npages = stripe_npages; |
53b381b3 DW |
1006 | rbio->faila = -1; |
1007 | rbio->failb = -1; | |
dec95574 | 1008 | refcount_set(&rbio->refs, 1); |
b89e1b01 MX |
1009 | atomic_set(&rbio->error, 0); |
1010 | atomic_set(&rbio->stripes_pending, 0); | |
53b381b3 DW |
1011 | |
1012 | /* | |
1389053e | 1013 | * the stripe_pages, bio_pages, etc arrays point to the extra |
53b381b3 DW |
1014 | * memory we allocated past the end of the rbio |
1015 | */ | |
1016 | p = rbio + 1; | |
1389053e KC |
1017 | #define CONSUME_ALLOC(ptr, count) do { \ |
1018 | ptr = p; \ | |
1019 | p = (unsigned char *)p + sizeof(*(ptr)) * (count); \ | |
1020 | } while (0) | |
1021 | CONSUME_ALLOC(rbio->stripe_pages, num_pages); | |
1022 | CONSUME_ALLOC(rbio->bio_pages, num_pages); | |
1023 | CONSUME_ALLOC(rbio->finish_pointers, real_stripes); | |
1024 | CONSUME_ALLOC(rbio->dbitmap, BITS_TO_LONGS(stripe_npages)); | |
1025 | CONSUME_ALLOC(rbio->finish_pbitmap, BITS_TO_LONGS(stripe_npages)); | |
1026 | #undef CONSUME_ALLOC | |
53b381b3 | 1027 | |
10f11900 ZL |
1028 | if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5) |
1029 | nr_data = real_stripes - 1; | |
1030 | else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6) | |
2c8cdd6e | 1031 | nr_data = real_stripes - 2; |
53b381b3 | 1032 | else |
10f11900 | 1033 | BUG(); |
53b381b3 DW |
1034 | |
1035 | rbio->nr_data = nr_data; | |
1036 | return rbio; | |
1037 | } | |
1038 | ||
1039 | /* allocate pages for all the stripes in the bio, including parity */ | |
1040 | static int alloc_rbio_pages(struct btrfs_raid_bio *rbio) | |
1041 | { | |
1042 | int i; | |
1043 | struct page *page; | |
1044 | ||
1045 | for (i = 0; i < rbio->nr_pages; i++) { | |
1046 | if (rbio->stripe_pages[i]) | |
1047 | continue; | |
1048 | page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
1049 | if (!page) | |
1050 | return -ENOMEM; | |
1051 | rbio->stripe_pages[i] = page; | |
53b381b3 DW |
1052 | } |
1053 | return 0; | |
1054 | } | |
1055 | ||
b7178a5f | 1056 | /* only allocate pages for p/q stripes */ |
53b381b3 DW |
1057 | static int alloc_rbio_parity_pages(struct btrfs_raid_bio *rbio) |
1058 | { | |
1059 | int i; | |
1060 | struct page *page; | |
1061 | ||
b7178a5f | 1062 | i = rbio_stripe_page_index(rbio, rbio->nr_data, 0); |
53b381b3 DW |
1063 | |
1064 | for (; i < rbio->nr_pages; i++) { | |
1065 | if (rbio->stripe_pages[i]) | |
1066 | continue; | |
1067 | page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
1068 | if (!page) | |
1069 | return -ENOMEM; | |
1070 | rbio->stripe_pages[i] = page; | |
1071 | } | |
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * add a single page from a specific stripe into our list of bios for IO | |
1077 | * this will try to merge into existing bios if possible, and returns | |
1078 | * zero if all went well. | |
1079 | */ | |
48a3b636 ES |
1080 | static int rbio_add_io_page(struct btrfs_raid_bio *rbio, |
1081 | struct bio_list *bio_list, | |
1082 | struct page *page, | |
1083 | int stripe_nr, | |
1084 | unsigned long page_index, | |
1085 | unsigned long bio_max_len) | |
53b381b3 DW |
1086 | { |
1087 | struct bio *last = bio_list->tail; | |
1088 | u64 last_end = 0; | |
1089 | int ret; | |
1090 | struct bio *bio; | |
1091 | struct btrfs_bio_stripe *stripe; | |
1092 | u64 disk_start; | |
1093 | ||
1094 | stripe = &rbio->bbio->stripes[stripe_nr]; | |
09cbfeaf | 1095 | disk_start = stripe->physical + (page_index << PAGE_SHIFT); |
53b381b3 DW |
1096 | |
1097 | /* if the device is missing, just fail this stripe */ | |
1098 | if (!stripe->dev->bdev) | |
1099 | return fail_rbio_index(rbio, stripe_nr); | |
1100 | ||
1101 | /* see if we can add this page onto our existing bio */ | |
1102 | if (last) { | |
4f024f37 KO |
1103 | last_end = (u64)last->bi_iter.bi_sector << 9; |
1104 | last_end += last->bi_iter.bi_size; | |
53b381b3 DW |
1105 | |
1106 | /* | |
1107 | * we can't merge these if they are from different | |
1108 | * devices or if they are not contiguous | |
1109 | */ | |
1110 | if (last_end == disk_start && stripe->dev->bdev && | |
4e4cbee9 | 1111 | !last->bi_status && |
74d46992 CH |
1112 | last->bi_disk == stripe->dev->bdev->bd_disk && |
1113 | last->bi_partno == stripe->dev->bdev->bd_partno) { | |
09cbfeaf KS |
1114 | ret = bio_add_page(last, page, PAGE_SIZE, 0); |
1115 | if (ret == PAGE_SIZE) | |
53b381b3 DW |
1116 | return 0; |
1117 | } | |
1118 | } | |
1119 | ||
1120 | /* put a new bio on the list */ | |
c5e4c3d7 | 1121 | bio = btrfs_io_bio_alloc(bio_max_len >> PAGE_SHIFT ?: 1); |
4f024f37 | 1122 | bio->bi_iter.bi_size = 0; |
74d46992 | 1123 | bio_set_dev(bio, stripe->dev->bdev); |
4f024f37 | 1124 | bio->bi_iter.bi_sector = disk_start >> 9; |
53b381b3 | 1125 | |
09cbfeaf | 1126 | bio_add_page(bio, page, PAGE_SIZE, 0); |
53b381b3 DW |
1127 | bio_list_add(bio_list, bio); |
1128 | return 0; | |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * while we're doing the read/modify/write cycle, we could | |
1133 | * have errors in reading pages off the disk. This checks | |
1134 | * for errors and if we're not able to read the page it'll | |
1135 | * trigger parity reconstruction. The rmw will be finished | |
1136 | * after we've reconstructed the failed stripes | |
1137 | */ | |
1138 | static void validate_rbio_for_rmw(struct btrfs_raid_bio *rbio) | |
1139 | { | |
1140 | if (rbio->faila >= 0 || rbio->failb >= 0) { | |
2c8cdd6e | 1141 | BUG_ON(rbio->faila == rbio->real_stripes - 1); |
53b381b3 DW |
1142 | __raid56_parity_recover(rbio); |
1143 | } else { | |
1144 | finish_rmw(rbio); | |
1145 | } | |
1146 | } | |
1147 | ||
53b381b3 DW |
1148 | /* |
1149 | * helper function to walk our bio list and populate the bio_pages array with | |
1150 | * the result. This seems expensive, but it is faster than constantly | |
1151 | * searching through the bio list as we setup the IO in finish_rmw or stripe | |
1152 | * reconstruction. | |
1153 | * | |
1154 | * This must be called before you trust the answers from page_in_rbio | |
1155 | */ | |
1156 | static void index_rbio_pages(struct btrfs_raid_bio *rbio) | |
1157 | { | |
1158 | struct bio *bio; | |
1159 | u64 start; | |
1160 | unsigned long stripe_offset; | |
1161 | unsigned long page_index; | |
53b381b3 DW |
1162 | |
1163 | spin_lock_irq(&rbio->bio_list_lock); | |
1164 | bio_list_for_each(bio, &rbio->bio_list) { | |
6592e58c FM |
1165 | struct bio_vec bvec; |
1166 | struct bvec_iter iter; | |
1167 | int i = 0; | |
1168 | ||
4f024f37 | 1169 | start = (u64)bio->bi_iter.bi_sector << 9; |
8e5cfb55 | 1170 | stripe_offset = start - rbio->bbio->raid_map[0]; |
09cbfeaf | 1171 | page_index = stripe_offset >> PAGE_SHIFT; |
53b381b3 | 1172 | |
6592e58c FM |
1173 | if (bio_flagged(bio, BIO_CLONED)) |
1174 | bio->bi_iter = btrfs_io_bio(bio)->iter; | |
1175 | ||
1176 | bio_for_each_segment(bvec, bio, iter) { | |
1177 | rbio->bio_pages[page_index + i] = bvec.bv_page; | |
1178 | i++; | |
1179 | } | |
53b381b3 DW |
1180 | } |
1181 | spin_unlock_irq(&rbio->bio_list_lock); | |
1182 | } | |
1183 | ||
1184 | /* | |
1185 | * this is called from one of two situations. We either | |
1186 | * have a full stripe from the higher layers, or we've read all | |
1187 | * the missing bits off disk. | |
1188 | * | |
1189 | * This will calculate the parity and then send down any | |
1190 | * changed blocks. | |
1191 | */ | |
1192 | static noinline void finish_rmw(struct btrfs_raid_bio *rbio) | |
1193 | { | |
1194 | struct btrfs_bio *bbio = rbio->bbio; | |
1389053e | 1195 | void **pointers = rbio->finish_pointers; |
53b381b3 DW |
1196 | int nr_data = rbio->nr_data; |
1197 | int stripe; | |
1198 | int pagenr; | |
1199 | int p_stripe = -1; | |
1200 | int q_stripe = -1; | |
1201 | struct bio_list bio_list; | |
1202 | struct bio *bio; | |
53b381b3 DW |
1203 | int ret; |
1204 | ||
1205 | bio_list_init(&bio_list); | |
1206 | ||
2c8cdd6e MX |
1207 | if (rbio->real_stripes - rbio->nr_data == 1) { |
1208 | p_stripe = rbio->real_stripes - 1; | |
1209 | } else if (rbio->real_stripes - rbio->nr_data == 2) { | |
1210 | p_stripe = rbio->real_stripes - 2; | |
1211 | q_stripe = rbio->real_stripes - 1; | |
53b381b3 DW |
1212 | } else { |
1213 | BUG(); | |
1214 | } | |
1215 | ||
1216 | /* at this point we either have a full stripe, | |
1217 | * or we've read the full stripe from the drive. | |
1218 | * recalculate the parity and write the new results. | |
1219 | * | |
1220 | * We're not allowed to add any new bios to the | |
1221 | * bio list here, anyone else that wants to | |
1222 | * change this stripe needs to do their own rmw. | |
1223 | */ | |
1224 | spin_lock_irq(&rbio->bio_list_lock); | |
1225 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
1226 | spin_unlock_irq(&rbio->bio_list_lock); | |
1227 | ||
b89e1b01 | 1228 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
1229 | |
1230 | /* | |
1231 | * now that we've set rmw_locked, run through the | |
1232 | * bio list one last time and map the page pointers | |
4ae10b3a CM |
1233 | * |
1234 | * We don't cache full rbios because we're assuming | |
1235 | * the higher layers are unlikely to use this area of | |
1236 | * the disk again soon. If they do use it again, | |
1237 | * hopefully they will send another full bio. | |
53b381b3 DW |
1238 | */ |
1239 | index_rbio_pages(rbio); | |
4ae10b3a CM |
1240 | if (!rbio_is_full(rbio)) |
1241 | cache_rbio_pages(rbio); | |
1242 | else | |
1243 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
53b381b3 | 1244 | |
915e2290 | 1245 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
1246 | struct page *p; |
1247 | /* first collect one page from each data stripe */ | |
1248 | for (stripe = 0; stripe < nr_data; stripe++) { | |
1249 | p = page_in_rbio(rbio, stripe, pagenr, 0); | |
1250 | pointers[stripe] = kmap(p); | |
1251 | } | |
1252 | ||
1253 | /* then add the parity stripe */ | |
1254 | p = rbio_pstripe_page(rbio, pagenr); | |
1255 | SetPageUptodate(p); | |
1256 | pointers[stripe++] = kmap(p); | |
1257 | ||
1258 | if (q_stripe != -1) { | |
1259 | ||
1260 | /* | |
1261 | * raid6, add the qstripe and call the | |
1262 | * library function to fill in our p/q | |
1263 | */ | |
1264 | p = rbio_qstripe_page(rbio, pagenr); | |
1265 | SetPageUptodate(p); | |
1266 | pointers[stripe++] = kmap(p); | |
1267 | ||
2c8cdd6e | 1268 | raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE, |
53b381b3 DW |
1269 | pointers); |
1270 | } else { | |
1271 | /* raid5 */ | |
69d24804 | 1272 | copy_page(pointers[nr_data], pointers[0]); |
09cbfeaf | 1273 | run_xor(pointers + 1, nr_data - 1, PAGE_SIZE); |
53b381b3 DW |
1274 | } |
1275 | ||
1276 | ||
2c8cdd6e | 1277 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) |
53b381b3 DW |
1278 | kunmap(page_in_rbio(rbio, stripe, pagenr, 0)); |
1279 | } | |
1280 | ||
1281 | /* | |
1282 | * time to start writing. Make bios for everything from the | |
1283 | * higher layers (the bio_list in our rbio) and our p/q. Ignore | |
1284 | * everything else. | |
1285 | */ | |
2c8cdd6e | 1286 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
915e2290 | 1287 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
1288 | struct page *page; |
1289 | if (stripe < rbio->nr_data) { | |
1290 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
1291 | if (!page) | |
1292 | continue; | |
1293 | } else { | |
1294 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
1295 | } | |
1296 | ||
1297 | ret = rbio_add_io_page(rbio, &bio_list, | |
1298 | page, stripe, pagenr, rbio->stripe_len); | |
1299 | if (ret) | |
1300 | goto cleanup; | |
1301 | } | |
1302 | } | |
1303 | ||
2c8cdd6e MX |
1304 | if (likely(!bbio->num_tgtdevs)) |
1305 | goto write_data; | |
1306 | ||
1307 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { | |
1308 | if (!bbio->tgtdev_map[stripe]) | |
1309 | continue; | |
1310 | ||
915e2290 | 1311 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
2c8cdd6e MX |
1312 | struct page *page; |
1313 | if (stripe < rbio->nr_data) { | |
1314 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
1315 | if (!page) | |
1316 | continue; | |
1317 | } else { | |
1318 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
1319 | } | |
1320 | ||
1321 | ret = rbio_add_io_page(rbio, &bio_list, page, | |
1322 | rbio->bbio->tgtdev_map[stripe], | |
1323 | pagenr, rbio->stripe_len); | |
1324 | if (ret) | |
1325 | goto cleanup; | |
1326 | } | |
1327 | } | |
1328 | ||
1329 | write_data: | |
b89e1b01 MX |
1330 | atomic_set(&rbio->stripes_pending, bio_list_size(&bio_list)); |
1331 | BUG_ON(atomic_read(&rbio->stripes_pending) == 0); | |
53b381b3 DW |
1332 | |
1333 | while (1) { | |
1334 | bio = bio_list_pop(&bio_list); | |
1335 | if (!bio) | |
1336 | break; | |
1337 | ||
1338 | bio->bi_private = rbio; | |
1339 | bio->bi_end_io = raid_write_end_io; | |
ebcc3263 | 1340 | bio->bi_opf = REQ_OP_WRITE; |
4e49ea4a MC |
1341 | |
1342 | submit_bio(bio); | |
53b381b3 DW |
1343 | } |
1344 | return; | |
1345 | ||
1346 | cleanup: | |
58efbc9f | 1347 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
1348 | |
1349 | while ((bio = bio_list_pop(&bio_list))) | |
1350 | bio_put(bio); | |
53b381b3 DW |
1351 | } |
1352 | ||
1353 | /* | |
1354 | * helper to find the stripe number for a given bio. Used to figure out which | |
1355 | * stripe has failed. This expects the bio to correspond to a physical disk, | |
1356 | * so it looks up based on physical sector numbers. | |
1357 | */ | |
1358 | static int find_bio_stripe(struct btrfs_raid_bio *rbio, | |
1359 | struct bio *bio) | |
1360 | { | |
4f024f37 | 1361 | u64 physical = bio->bi_iter.bi_sector; |
53b381b3 DW |
1362 | u64 stripe_start; |
1363 | int i; | |
1364 | struct btrfs_bio_stripe *stripe; | |
1365 | ||
1366 | physical <<= 9; | |
1367 | ||
1368 | for (i = 0; i < rbio->bbio->num_stripes; i++) { | |
1369 | stripe = &rbio->bbio->stripes[i]; | |
1370 | stripe_start = stripe->physical; | |
1371 | if (physical >= stripe_start && | |
2c8cdd6e | 1372 | physical < stripe_start + rbio->stripe_len && |
047fdea6 | 1373 | stripe->dev->bdev && |
74d46992 CH |
1374 | bio->bi_disk == stripe->dev->bdev->bd_disk && |
1375 | bio->bi_partno == stripe->dev->bdev->bd_partno) { | |
53b381b3 DW |
1376 | return i; |
1377 | } | |
1378 | } | |
1379 | return -1; | |
1380 | } | |
1381 | ||
1382 | /* | |
1383 | * helper to find the stripe number for a given | |
1384 | * bio (before mapping). Used to figure out which stripe has | |
1385 | * failed. This looks up based on logical block numbers. | |
1386 | */ | |
1387 | static int find_logical_bio_stripe(struct btrfs_raid_bio *rbio, | |
1388 | struct bio *bio) | |
1389 | { | |
4f024f37 | 1390 | u64 logical = bio->bi_iter.bi_sector; |
53b381b3 DW |
1391 | u64 stripe_start; |
1392 | int i; | |
1393 | ||
1394 | logical <<= 9; | |
1395 | ||
1396 | for (i = 0; i < rbio->nr_data; i++) { | |
8e5cfb55 | 1397 | stripe_start = rbio->bbio->raid_map[i]; |
53b381b3 DW |
1398 | if (logical >= stripe_start && |
1399 | logical < stripe_start + rbio->stripe_len) { | |
1400 | return i; | |
1401 | } | |
1402 | } | |
1403 | return -1; | |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * returns -EIO if we had too many failures | |
1408 | */ | |
1409 | static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed) | |
1410 | { | |
1411 | unsigned long flags; | |
1412 | int ret = 0; | |
1413 | ||
1414 | spin_lock_irqsave(&rbio->bio_list_lock, flags); | |
1415 | ||
1416 | /* we already know this stripe is bad, move on */ | |
1417 | if (rbio->faila == failed || rbio->failb == failed) | |
1418 | goto out; | |
1419 | ||
1420 | if (rbio->faila == -1) { | |
1421 | /* first failure on this rbio */ | |
1422 | rbio->faila = failed; | |
b89e1b01 | 1423 | atomic_inc(&rbio->error); |
53b381b3 DW |
1424 | } else if (rbio->failb == -1) { |
1425 | /* second failure on this rbio */ | |
1426 | rbio->failb = failed; | |
b89e1b01 | 1427 | atomic_inc(&rbio->error); |
53b381b3 DW |
1428 | } else { |
1429 | ret = -EIO; | |
1430 | } | |
1431 | out: | |
1432 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); | |
1433 | ||
1434 | return ret; | |
1435 | } | |
1436 | ||
1437 | /* | |
1438 | * helper to fail a stripe based on a physical disk | |
1439 | * bio. | |
1440 | */ | |
1441 | static int fail_bio_stripe(struct btrfs_raid_bio *rbio, | |
1442 | struct bio *bio) | |
1443 | { | |
1444 | int failed = find_bio_stripe(rbio, bio); | |
1445 | ||
1446 | if (failed < 0) | |
1447 | return -EIO; | |
1448 | ||
1449 | return fail_rbio_index(rbio, failed); | |
1450 | } | |
1451 | ||
1452 | /* | |
1453 | * this sets each page in the bio uptodate. It should only be used on private | |
1454 | * rbio pages, nothing that comes in from the higher layers | |
1455 | */ | |
1456 | static void set_bio_pages_uptodate(struct bio *bio) | |
1457 | { | |
0198e5b7 | 1458 | struct bio_vec *bvec; |
6dc4f100 | 1459 | struct bvec_iter_all iter_all; |
6592e58c | 1460 | |
0198e5b7 | 1461 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
53b381b3 | 1462 | |
2b070cfe | 1463 | bio_for_each_segment_all(bvec, bio, iter_all) |
0198e5b7 | 1464 | SetPageUptodate(bvec->bv_page); |
53b381b3 DW |
1465 | } |
1466 | ||
1467 | /* | |
1468 | * end io for the read phase of the rmw cycle. All the bios here are physical | |
1469 | * stripe bios we've read from the disk so we can recalculate the parity of the | |
1470 | * stripe. | |
1471 | * | |
1472 | * This will usually kick off finish_rmw once all the bios are read in, but it | |
1473 | * may trigger parity reconstruction if we had any errors along the way | |
1474 | */ | |
4246a0b6 | 1475 | static void raid_rmw_end_io(struct bio *bio) |
53b381b3 DW |
1476 | { |
1477 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
1478 | ||
4e4cbee9 | 1479 | if (bio->bi_status) |
53b381b3 DW |
1480 | fail_bio_stripe(rbio, bio); |
1481 | else | |
1482 | set_bio_pages_uptodate(bio); | |
1483 | ||
1484 | bio_put(bio); | |
1485 | ||
b89e1b01 | 1486 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
1487 | return; |
1488 | ||
b89e1b01 | 1489 | if (atomic_read(&rbio->error) > rbio->bbio->max_errors) |
53b381b3 DW |
1490 | goto cleanup; |
1491 | ||
1492 | /* | |
1493 | * this will normally call finish_rmw to start our write | |
1494 | * but if there are any failed stripes we'll reconstruct | |
1495 | * from parity first | |
1496 | */ | |
1497 | validate_rbio_for_rmw(rbio); | |
1498 | return; | |
1499 | ||
1500 | cleanup: | |
1501 | ||
58efbc9f | 1502 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
53b381b3 DW |
1503 | } |
1504 | ||
53b381b3 DW |
1505 | /* |
1506 | * the stripe must be locked by the caller. It will | |
1507 | * unlock after all the writes are done | |
1508 | */ | |
1509 | static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio) | |
1510 | { | |
1511 | int bios_to_read = 0; | |
53b381b3 DW |
1512 | struct bio_list bio_list; |
1513 | int ret; | |
53b381b3 DW |
1514 | int pagenr; |
1515 | int stripe; | |
1516 | struct bio *bio; | |
1517 | ||
1518 | bio_list_init(&bio_list); | |
1519 | ||
1520 | ret = alloc_rbio_pages(rbio); | |
1521 | if (ret) | |
1522 | goto cleanup; | |
1523 | ||
1524 | index_rbio_pages(rbio); | |
1525 | ||
b89e1b01 | 1526 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
1527 | /* |
1528 | * build a list of bios to read all the missing parts of this | |
1529 | * stripe | |
1530 | */ | |
1531 | for (stripe = 0; stripe < rbio->nr_data; stripe++) { | |
915e2290 | 1532 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
1533 | struct page *page; |
1534 | /* | |
1535 | * we want to find all the pages missing from | |
1536 | * the rbio and read them from the disk. If | |
1537 | * page_in_rbio finds a page in the bio list | |
1538 | * we don't need to read it off the stripe. | |
1539 | */ | |
1540 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
1541 | if (page) | |
1542 | continue; | |
1543 | ||
1544 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
4ae10b3a CM |
1545 | /* |
1546 | * the bio cache may have handed us an uptodate | |
1547 | * page. If so, be happy and use it | |
1548 | */ | |
1549 | if (PageUptodate(page)) | |
1550 | continue; | |
1551 | ||
53b381b3 DW |
1552 | ret = rbio_add_io_page(rbio, &bio_list, page, |
1553 | stripe, pagenr, rbio->stripe_len); | |
1554 | if (ret) | |
1555 | goto cleanup; | |
1556 | } | |
1557 | } | |
1558 | ||
1559 | bios_to_read = bio_list_size(&bio_list); | |
1560 | if (!bios_to_read) { | |
1561 | /* | |
1562 | * this can happen if others have merged with | |
1563 | * us, it means there is nothing left to read. | |
1564 | * But if there are missing devices it may not be | |
1565 | * safe to do the full stripe write yet. | |
1566 | */ | |
1567 | goto finish; | |
1568 | } | |
1569 | ||
1570 | /* | |
1571 | * the bbio may be freed once we submit the last bio. Make sure | |
1572 | * not to touch it after that | |
1573 | */ | |
b89e1b01 | 1574 | atomic_set(&rbio->stripes_pending, bios_to_read); |
53b381b3 DW |
1575 | while (1) { |
1576 | bio = bio_list_pop(&bio_list); | |
1577 | if (!bio) | |
1578 | break; | |
1579 | ||
1580 | bio->bi_private = rbio; | |
1581 | bio->bi_end_io = raid_rmw_end_io; | |
ebcc3263 | 1582 | bio->bi_opf = REQ_OP_READ; |
53b381b3 | 1583 | |
0b246afa | 1584 | btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
53b381b3 | 1585 | |
4e49ea4a | 1586 | submit_bio(bio); |
53b381b3 DW |
1587 | } |
1588 | /* the actual write will happen once the reads are done */ | |
1589 | return 0; | |
1590 | ||
1591 | cleanup: | |
58efbc9f | 1592 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
1593 | |
1594 | while ((bio = bio_list_pop(&bio_list))) | |
1595 | bio_put(bio); | |
1596 | ||
53b381b3 DW |
1597 | return -EIO; |
1598 | ||
1599 | finish: | |
1600 | validate_rbio_for_rmw(rbio); | |
1601 | return 0; | |
1602 | } | |
1603 | ||
1604 | /* | |
1605 | * if the upper layers pass in a full stripe, we thank them by only allocating | |
1606 | * enough pages to hold the parity, and sending it all down quickly. | |
1607 | */ | |
1608 | static int full_stripe_write(struct btrfs_raid_bio *rbio) | |
1609 | { | |
1610 | int ret; | |
1611 | ||
1612 | ret = alloc_rbio_parity_pages(rbio); | |
3cd846d1 MX |
1613 | if (ret) { |
1614 | __free_raid_bio(rbio); | |
53b381b3 | 1615 | return ret; |
3cd846d1 | 1616 | } |
53b381b3 DW |
1617 | |
1618 | ret = lock_stripe_add(rbio); | |
1619 | if (ret == 0) | |
1620 | finish_rmw(rbio); | |
1621 | return 0; | |
1622 | } | |
1623 | ||
1624 | /* | |
1625 | * partial stripe writes get handed over to async helpers. | |
1626 | * We're really hoping to merge a few more writes into this | |
1627 | * rbio before calculating new parity | |
1628 | */ | |
1629 | static int partial_stripe_write(struct btrfs_raid_bio *rbio) | |
1630 | { | |
1631 | int ret; | |
1632 | ||
1633 | ret = lock_stripe_add(rbio); | |
1634 | if (ret == 0) | |
cf6a4a75 | 1635 | start_async_work(rbio, rmw_work); |
53b381b3 DW |
1636 | return 0; |
1637 | } | |
1638 | ||
1639 | /* | |
1640 | * sometimes while we were reading from the drive to | |
1641 | * recalculate parity, enough new bios come into create | |
1642 | * a full stripe. So we do a check here to see if we can | |
1643 | * go directly to finish_rmw | |
1644 | */ | |
1645 | static int __raid56_parity_write(struct btrfs_raid_bio *rbio) | |
1646 | { | |
1647 | /* head off into rmw land if we don't have a full stripe */ | |
1648 | if (!rbio_is_full(rbio)) | |
1649 | return partial_stripe_write(rbio); | |
1650 | return full_stripe_write(rbio); | |
1651 | } | |
1652 | ||
6ac0f488 CM |
1653 | /* |
1654 | * We use plugging call backs to collect full stripes. | |
1655 | * Any time we get a partial stripe write while plugged | |
1656 | * we collect it into a list. When the unplug comes down, | |
1657 | * we sort the list by logical block number and merge | |
1658 | * everything we can into the same rbios | |
1659 | */ | |
1660 | struct btrfs_plug_cb { | |
1661 | struct blk_plug_cb cb; | |
1662 | struct btrfs_fs_info *info; | |
1663 | struct list_head rbio_list; | |
1664 | struct btrfs_work work; | |
1665 | }; | |
1666 | ||
1667 | /* | |
1668 | * rbios on the plug list are sorted for easier merging. | |
1669 | */ | |
1670 | static int plug_cmp(void *priv, struct list_head *a, struct list_head *b) | |
1671 | { | |
1672 | struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio, | |
1673 | plug_list); | |
1674 | struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio, | |
1675 | plug_list); | |
4f024f37 KO |
1676 | u64 a_sector = ra->bio_list.head->bi_iter.bi_sector; |
1677 | u64 b_sector = rb->bio_list.head->bi_iter.bi_sector; | |
6ac0f488 CM |
1678 | |
1679 | if (a_sector < b_sector) | |
1680 | return -1; | |
1681 | if (a_sector > b_sector) | |
1682 | return 1; | |
1683 | return 0; | |
1684 | } | |
1685 | ||
1686 | static void run_plug(struct btrfs_plug_cb *plug) | |
1687 | { | |
1688 | struct btrfs_raid_bio *cur; | |
1689 | struct btrfs_raid_bio *last = NULL; | |
1690 | ||
1691 | /* | |
1692 | * sort our plug list then try to merge | |
1693 | * everything we can in hopes of creating full | |
1694 | * stripes. | |
1695 | */ | |
1696 | list_sort(NULL, &plug->rbio_list, plug_cmp); | |
1697 | while (!list_empty(&plug->rbio_list)) { | |
1698 | cur = list_entry(plug->rbio_list.next, | |
1699 | struct btrfs_raid_bio, plug_list); | |
1700 | list_del_init(&cur->plug_list); | |
1701 | ||
1702 | if (rbio_is_full(cur)) { | |
c7b562c5 DS |
1703 | int ret; |
1704 | ||
6ac0f488 | 1705 | /* we have a full stripe, send it down */ |
c7b562c5 DS |
1706 | ret = full_stripe_write(cur); |
1707 | BUG_ON(ret); | |
6ac0f488 CM |
1708 | continue; |
1709 | } | |
1710 | if (last) { | |
1711 | if (rbio_can_merge(last, cur)) { | |
1712 | merge_rbio(last, cur); | |
1713 | __free_raid_bio(cur); | |
1714 | continue; | |
1715 | ||
1716 | } | |
1717 | __raid56_parity_write(last); | |
1718 | } | |
1719 | last = cur; | |
1720 | } | |
1721 | if (last) { | |
1722 | __raid56_parity_write(last); | |
1723 | } | |
1724 | kfree(plug); | |
1725 | } | |
1726 | ||
1727 | /* | |
1728 | * if the unplug comes from schedule, we have to push the | |
1729 | * work off to a helper thread | |
1730 | */ | |
1731 | static void unplug_work(struct btrfs_work *work) | |
1732 | { | |
1733 | struct btrfs_plug_cb *plug; | |
1734 | plug = container_of(work, struct btrfs_plug_cb, work); | |
1735 | run_plug(plug); | |
1736 | } | |
1737 | ||
1738 | static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1739 | { | |
1740 | struct btrfs_plug_cb *plug; | |
1741 | plug = container_of(cb, struct btrfs_plug_cb, cb); | |
1742 | ||
1743 | if (from_schedule) { | |
a0cac0ec | 1744 | btrfs_init_work(&plug->work, unplug_work, NULL, NULL); |
d05a33ac QW |
1745 | btrfs_queue_work(plug->info->rmw_workers, |
1746 | &plug->work); | |
6ac0f488 CM |
1747 | return; |
1748 | } | |
1749 | run_plug(plug); | |
1750 | } | |
1751 | ||
53b381b3 DW |
1752 | /* |
1753 | * our main entry point for writes from the rest of the FS. | |
1754 | */ | |
2ff7e61e | 1755 | int raid56_parity_write(struct btrfs_fs_info *fs_info, struct bio *bio, |
8e5cfb55 | 1756 | struct btrfs_bio *bbio, u64 stripe_len) |
53b381b3 DW |
1757 | { |
1758 | struct btrfs_raid_bio *rbio; | |
6ac0f488 CM |
1759 | struct btrfs_plug_cb *plug = NULL; |
1760 | struct blk_plug_cb *cb; | |
4245215d | 1761 | int ret; |
53b381b3 | 1762 | |
2ff7e61e | 1763 | rbio = alloc_rbio(fs_info, bbio, stripe_len); |
af8e2d1d | 1764 | if (IS_ERR(rbio)) { |
6e9606d2 | 1765 | btrfs_put_bbio(bbio); |
53b381b3 | 1766 | return PTR_ERR(rbio); |
af8e2d1d | 1767 | } |
53b381b3 | 1768 | bio_list_add(&rbio->bio_list, bio); |
4f024f37 | 1769 | rbio->bio_list_bytes = bio->bi_iter.bi_size; |
1b94b556 | 1770 | rbio->operation = BTRFS_RBIO_WRITE; |
6ac0f488 | 1771 | |
0b246afa | 1772 | btrfs_bio_counter_inc_noblocked(fs_info); |
4245215d MX |
1773 | rbio->generic_bio_cnt = 1; |
1774 | ||
6ac0f488 CM |
1775 | /* |
1776 | * don't plug on full rbios, just get them out the door | |
1777 | * as quickly as we can | |
1778 | */ | |
4245215d MX |
1779 | if (rbio_is_full(rbio)) { |
1780 | ret = full_stripe_write(rbio); | |
1781 | if (ret) | |
0b246afa | 1782 | btrfs_bio_counter_dec(fs_info); |
4245215d MX |
1783 | return ret; |
1784 | } | |
6ac0f488 | 1785 | |
0b246afa | 1786 | cb = blk_check_plugged(btrfs_raid_unplug, fs_info, sizeof(*plug)); |
6ac0f488 CM |
1787 | if (cb) { |
1788 | plug = container_of(cb, struct btrfs_plug_cb, cb); | |
1789 | if (!plug->info) { | |
0b246afa | 1790 | plug->info = fs_info; |
6ac0f488 CM |
1791 | INIT_LIST_HEAD(&plug->rbio_list); |
1792 | } | |
1793 | list_add_tail(&rbio->plug_list, &plug->rbio_list); | |
4245215d | 1794 | ret = 0; |
6ac0f488 | 1795 | } else { |
4245215d MX |
1796 | ret = __raid56_parity_write(rbio); |
1797 | if (ret) | |
0b246afa | 1798 | btrfs_bio_counter_dec(fs_info); |
6ac0f488 | 1799 | } |
4245215d | 1800 | return ret; |
53b381b3 DW |
1801 | } |
1802 | ||
1803 | /* | |
1804 | * all parity reconstruction happens here. We've read in everything | |
1805 | * we can find from the drives and this does the heavy lifting of | |
1806 | * sorting the good from the bad. | |
1807 | */ | |
1808 | static void __raid_recover_end_io(struct btrfs_raid_bio *rbio) | |
1809 | { | |
1810 | int pagenr, stripe; | |
1811 | void **pointers; | |
1812 | int faila = -1, failb = -1; | |
53b381b3 | 1813 | struct page *page; |
58efbc9f | 1814 | blk_status_t err; |
53b381b3 DW |
1815 | int i; |
1816 | ||
31e818fe | 1817 | pointers = kcalloc(rbio->real_stripes, sizeof(void *), GFP_NOFS); |
53b381b3 | 1818 | if (!pointers) { |
58efbc9f | 1819 | err = BLK_STS_RESOURCE; |
53b381b3 DW |
1820 | goto cleanup_io; |
1821 | } | |
1822 | ||
1823 | faila = rbio->faila; | |
1824 | failb = rbio->failb; | |
1825 | ||
b4ee1782 OS |
1826 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1827 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { | |
53b381b3 DW |
1828 | spin_lock_irq(&rbio->bio_list_lock); |
1829 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
1830 | spin_unlock_irq(&rbio->bio_list_lock); | |
1831 | } | |
1832 | ||
1833 | index_rbio_pages(rbio); | |
1834 | ||
915e2290 | 1835 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
5a6ac9ea MX |
1836 | /* |
1837 | * Now we just use bitmap to mark the horizontal stripes in | |
1838 | * which we have data when doing parity scrub. | |
1839 | */ | |
1840 | if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB && | |
1841 | !test_bit(pagenr, rbio->dbitmap)) | |
1842 | continue; | |
1843 | ||
53b381b3 DW |
1844 | /* setup our array of pointers with pages |
1845 | * from each stripe | |
1846 | */ | |
2c8cdd6e | 1847 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
53b381b3 DW |
1848 | /* |
1849 | * if we're rebuilding a read, we have to use | |
1850 | * pages from the bio list | |
1851 | */ | |
b4ee1782 OS |
1852 | if ((rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1853 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) && | |
53b381b3 DW |
1854 | (stripe == faila || stripe == failb)) { |
1855 | page = page_in_rbio(rbio, stripe, pagenr, 0); | |
1856 | } else { | |
1857 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
1858 | } | |
1859 | pointers[stripe] = kmap(page); | |
1860 | } | |
1861 | ||
1862 | /* all raid6 handling here */ | |
10f11900 | 1863 | if (rbio->bbio->map_type & BTRFS_BLOCK_GROUP_RAID6) { |
53b381b3 DW |
1864 | /* |
1865 | * single failure, rebuild from parity raid5 | |
1866 | * style | |
1867 | */ | |
1868 | if (failb < 0) { | |
1869 | if (faila == rbio->nr_data) { | |
1870 | /* | |
1871 | * Just the P stripe has failed, without | |
1872 | * a bad data or Q stripe. | |
1873 | * TODO, we should redo the xor here. | |
1874 | */ | |
58efbc9f | 1875 | err = BLK_STS_IOERR; |
53b381b3 DW |
1876 | goto cleanup; |
1877 | } | |
1878 | /* | |
1879 | * a single failure in raid6 is rebuilt | |
1880 | * in the pstripe code below | |
1881 | */ | |
1882 | goto pstripe; | |
1883 | } | |
1884 | ||
1885 | /* make sure our ps and qs are in order */ | |
1886 | if (faila > failb) { | |
1887 | int tmp = failb; | |
1888 | failb = faila; | |
1889 | faila = tmp; | |
1890 | } | |
1891 | ||
1892 | /* if the q stripe is failed, do a pstripe reconstruction | |
1893 | * from the xors. | |
1894 | * If both the q stripe and the P stripe are failed, we're | |
1895 | * here due to a crc mismatch and we can't give them the | |
1896 | * data they want | |
1897 | */ | |
8e5cfb55 ZL |
1898 | if (rbio->bbio->raid_map[failb] == RAID6_Q_STRIPE) { |
1899 | if (rbio->bbio->raid_map[faila] == | |
1900 | RAID5_P_STRIPE) { | |
58efbc9f | 1901 | err = BLK_STS_IOERR; |
53b381b3 DW |
1902 | goto cleanup; |
1903 | } | |
1904 | /* | |
1905 | * otherwise we have one bad data stripe and | |
1906 | * a good P stripe. raid5! | |
1907 | */ | |
1908 | goto pstripe; | |
1909 | } | |
1910 | ||
8e5cfb55 | 1911 | if (rbio->bbio->raid_map[failb] == RAID5_P_STRIPE) { |
2c8cdd6e | 1912 | raid6_datap_recov(rbio->real_stripes, |
53b381b3 DW |
1913 | PAGE_SIZE, faila, pointers); |
1914 | } else { | |
2c8cdd6e | 1915 | raid6_2data_recov(rbio->real_stripes, |
53b381b3 DW |
1916 | PAGE_SIZE, faila, failb, |
1917 | pointers); | |
1918 | } | |
1919 | } else { | |
1920 | void *p; | |
1921 | ||
1922 | /* rebuild from P stripe here (raid5 or raid6) */ | |
1923 | BUG_ON(failb != -1); | |
1924 | pstripe: | |
1925 | /* Copy parity block into failed block to start with */ | |
69d24804 | 1926 | copy_page(pointers[faila], pointers[rbio->nr_data]); |
53b381b3 DW |
1927 | |
1928 | /* rearrange the pointer array */ | |
1929 | p = pointers[faila]; | |
1930 | for (stripe = faila; stripe < rbio->nr_data - 1; stripe++) | |
1931 | pointers[stripe] = pointers[stripe + 1]; | |
1932 | pointers[rbio->nr_data - 1] = p; | |
1933 | ||
1934 | /* xor in the rest */ | |
09cbfeaf | 1935 | run_xor(pointers, rbio->nr_data - 1, PAGE_SIZE); |
53b381b3 DW |
1936 | } |
1937 | /* if we're doing this rebuild as part of an rmw, go through | |
1938 | * and set all of our private rbio pages in the | |
1939 | * failed stripes as uptodate. This way finish_rmw will | |
1940 | * know they can be trusted. If this was a read reconstruction, | |
1941 | * other endio functions will fiddle the uptodate bits | |
1942 | */ | |
1b94b556 | 1943 | if (rbio->operation == BTRFS_RBIO_WRITE) { |
915e2290 | 1944 | for (i = 0; i < rbio->stripe_npages; i++) { |
53b381b3 DW |
1945 | if (faila != -1) { |
1946 | page = rbio_stripe_page(rbio, faila, i); | |
1947 | SetPageUptodate(page); | |
1948 | } | |
1949 | if (failb != -1) { | |
1950 | page = rbio_stripe_page(rbio, failb, i); | |
1951 | SetPageUptodate(page); | |
1952 | } | |
1953 | } | |
1954 | } | |
2c8cdd6e | 1955 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
53b381b3 DW |
1956 | /* |
1957 | * if we're rebuilding a read, we have to use | |
1958 | * pages from the bio list | |
1959 | */ | |
b4ee1782 OS |
1960 | if ((rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1961 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) && | |
53b381b3 DW |
1962 | (stripe == faila || stripe == failb)) { |
1963 | page = page_in_rbio(rbio, stripe, pagenr, 0); | |
1964 | } else { | |
1965 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
1966 | } | |
1967 | kunmap(page); | |
1968 | } | |
1969 | } | |
1970 | ||
58efbc9f | 1971 | err = BLK_STS_OK; |
53b381b3 DW |
1972 | cleanup: |
1973 | kfree(pointers); | |
1974 | ||
1975 | cleanup_io: | |
580c6efa LB |
1976 | /* |
1977 | * Similar to READ_REBUILD, REBUILD_MISSING at this point also has a | |
1978 | * valid rbio which is consistent with ondisk content, thus such a | |
1979 | * valid rbio can be cached to avoid further disk reads. | |
1980 | */ | |
1981 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || | |
1982 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { | |
44ac474d LB |
1983 | /* |
1984 | * - In case of two failures, where rbio->failb != -1: | |
1985 | * | |
1986 | * Do not cache this rbio since the above read reconstruction | |
1987 | * (raid6_datap_recov() or raid6_2data_recov()) may have | |
1988 | * changed some content of stripes which are not identical to | |
1989 | * on-disk content any more, otherwise, a later write/recover | |
1990 | * may steal stripe_pages from this rbio and end up with | |
1991 | * corruptions or rebuild failures. | |
1992 | * | |
1993 | * - In case of single failure, where rbio->failb == -1: | |
1994 | * | |
1995 | * Cache this rbio iff the above read reconstruction is | |
52042d8e | 1996 | * executed without problems. |
44ac474d LB |
1997 | */ |
1998 | if (err == BLK_STS_OK && rbio->failb < 0) | |
4ae10b3a CM |
1999 | cache_rbio_pages(rbio); |
2000 | else | |
2001 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
2002 | ||
4246a0b6 | 2003 | rbio_orig_end_io(rbio, err); |
58efbc9f | 2004 | } else if (err == BLK_STS_OK) { |
53b381b3 DW |
2005 | rbio->faila = -1; |
2006 | rbio->failb = -1; | |
5a6ac9ea MX |
2007 | |
2008 | if (rbio->operation == BTRFS_RBIO_WRITE) | |
2009 | finish_rmw(rbio); | |
2010 | else if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) | |
2011 | finish_parity_scrub(rbio, 0); | |
2012 | else | |
2013 | BUG(); | |
53b381b3 | 2014 | } else { |
4246a0b6 | 2015 | rbio_orig_end_io(rbio, err); |
53b381b3 DW |
2016 | } |
2017 | } | |
2018 | ||
2019 | /* | |
2020 | * This is called only for stripes we've read from disk to | |
2021 | * reconstruct the parity. | |
2022 | */ | |
4246a0b6 | 2023 | static void raid_recover_end_io(struct bio *bio) |
53b381b3 DW |
2024 | { |
2025 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
2026 | ||
2027 | /* | |
2028 | * we only read stripe pages off the disk, set them | |
2029 | * up to date if there were no errors | |
2030 | */ | |
4e4cbee9 | 2031 | if (bio->bi_status) |
53b381b3 DW |
2032 | fail_bio_stripe(rbio, bio); |
2033 | else | |
2034 | set_bio_pages_uptodate(bio); | |
2035 | bio_put(bio); | |
2036 | ||
b89e1b01 | 2037 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
2038 | return; |
2039 | ||
b89e1b01 | 2040 | if (atomic_read(&rbio->error) > rbio->bbio->max_errors) |
58efbc9f | 2041 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
53b381b3 DW |
2042 | else |
2043 | __raid_recover_end_io(rbio); | |
2044 | } | |
2045 | ||
2046 | /* | |
2047 | * reads everything we need off the disk to reconstruct | |
2048 | * the parity. endio handlers trigger final reconstruction | |
2049 | * when the IO is done. | |
2050 | * | |
2051 | * This is used both for reads from the higher layers and for | |
2052 | * parity construction required to finish a rmw cycle. | |
2053 | */ | |
2054 | static int __raid56_parity_recover(struct btrfs_raid_bio *rbio) | |
2055 | { | |
2056 | int bios_to_read = 0; | |
53b381b3 DW |
2057 | struct bio_list bio_list; |
2058 | int ret; | |
53b381b3 DW |
2059 | int pagenr; |
2060 | int stripe; | |
2061 | struct bio *bio; | |
2062 | ||
2063 | bio_list_init(&bio_list); | |
2064 | ||
2065 | ret = alloc_rbio_pages(rbio); | |
2066 | if (ret) | |
2067 | goto cleanup; | |
2068 | ||
b89e1b01 | 2069 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
2070 | |
2071 | /* | |
4ae10b3a CM |
2072 | * read everything that hasn't failed. Thanks to the |
2073 | * stripe cache, it is possible that some or all of these | |
2074 | * pages are going to be uptodate. | |
53b381b3 | 2075 | */ |
2c8cdd6e | 2076 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
5588383e | 2077 | if (rbio->faila == stripe || rbio->failb == stripe) { |
b89e1b01 | 2078 | atomic_inc(&rbio->error); |
53b381b3 | 2079 | continue; |
5588383e | 2080 | } |
53b381b3 | 2081 | |
915e2290 | 2082 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
2083 | struct page *p; |
2084 | ||
2085 | /* | |
2086 | * the rmw code may have already read this | |
2087 | * page in | |
2088 | */ | |
2089 | p = rbio_stripe_page(rbio, stripe, pagenr); | |
2090 | if (PageUptodate(p)) | |
2091 | continue; | |
2092 | ||
2093 | ret = rbio_add_io_page(rbio, &bio_list, | |
2094 | rbio_stripe_page(rbio, stripe, pagenr), | |
2095 | stripe, pagenr, rbio->stripe_len); | |
2096 | if (ret < 0) | |
2097 | goto cleanup; | |
2098 | } | |
2099 | } | |
2100 | ||
2101 | bios_to_read = bio_list_size(&bio_list); | |
2102 | if (!bios_to_read) { | |
2103 | /* | |
2104 | * we might have no bios to read just because the pages | |
2105 | * were up to date, or we might have no bios to read because | |
2106 | * the devices were gone. | |
2107 | */ | |
b89e1b01 | 2108 | if (atomic_read(&rbio->error) <= rbio->bbio->max_errors) { |
53b381b3 DW |
2109 | __raid_recover_end_io(rbio); |
2110 | goto out; | |
2111 | } else { | |
2112 | goto cleanup; | |
2113 | } | |
2114 | } | |
2115 | ||
2116 | /* | |
2117 | * the bbio may be freed once we submit the last bio. Make sure | |
2118 | * not to touch it after that | |
2119 | */ | |
b89e1b01 | 2120 | atomic_set(&rbio->stripes_pending, bios_to_read); |
53b381b3 DW |
2121 | while (1) { |
2122 | bio = bio_list_pop(&bio_list); | |
2123 | if (!bio) | |
2124 | break; | |
2125 | ||
2126 | bio->bi_private = rbio; | |
2127 | bio->bi_end_io = raid_recover_end_io; | |
ebcc3263 | 2128 | bio->bi_opf = REQ_OP_READ; |
53b381b3 | 2129 | |
0b246afa | 2130 | btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
53b381b3 | 2131 | |
4e49ea4a | 2132 | submit_bio(bio); |
53b381b3 DW |
2133 | } |
2134 | out: | |
2135 | return 0; | |
2136 | ||
2137 | cleanup: | |
b4ee1782 OS |
2138 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
2139 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) | |
58efbc9f | 2140 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2141 | |
2142 | while ((bio = bio_list_pop(&bio_list))) | |
2143 | bio_put(bio); | |
2144 | ||
53b381b3 DW |
2145 | return -EIO; |
2146 | } | |
2147 | ||
2148 | /* | |
2149 | * the main entry point for reads from the higher layers. This | |
2150 | * is really only called when the normal read path had a failure, | |
2151 | * so we assume the bio they send down corresponds to a failed part | |
2152 | * of the drive. | |
2153 | */ | |
2ff7e61e | 2154 | int raid56_parity_recover(struct btrfs_fs_info *fs_info, struct bio *bio, |
8e5cfb55 ZL |
2155 | struct btrfs_bio *bbio, u64 stripe_len, |
2156 | int mirror_num, int generic_io) | |
53b381b3 DW |
2157 | { |
2158 | struct btrfs_raid_bio *rbio; | |
2159 | int ret; | |
2160 | ||
abad60c6 LB |
2161 | if (generic_io) { |
2162 | ASSERT(bbio->mirror_num == mirror_num); | |
2163 | btrfs_io_bio(bio)->mirror_num = mirror_num; | |
2164 | } | |
2165 | ||
2ff7e61e | 2166 | rbio = alloc_rbio(fs_info, bbio, stripe_len); |
af8e2d1d | 2167 | if (IS_ERR(rbio)) { |
6e9606d2 ZL |
2168 | if (generic_io) |
2169 | btrfs_put_bbio(bbio); | |
53b381b3 | 2170 | return PTR_ERR(rbio); |
af8e2d1d | 2171 | } |
53b381b3 | 2172 | |
1b94b556 | 2173 | rbio->operation = BTRFS_RBIO_READ_REBUILD; |
53b381b3 | 2174 | bio_list_add(&rbio->bio_list, bio); |
4f024f37 | 2175 | rbio->bio_list_bytes = bio->bi_iter.bi_size; |
53b381b3 DW |
2176 | |
2177 | rbio->faila = find_logical_bio_stripe(rbio, bio); | |
2178 | if (rbio->faila == -1) { | |
0b246afa | 2179 | btrfs_warn(fs_info, |
e46a28ca LB |
2180 | "%s could not find the bad stripe in raid56 so that we cannot recover any more (bio has logical %llu len %llu, bbio has map_type %llu)", |
2181 | __func__, (u64)bio->bi_iter.bi_sector << 9, | |
2182 | (u64)bio->bi_iter.bi_size, bbio->map_type); | |
6e9606d2 ZL |
2183 | if (generic_io) |
2184 | btrfs_put_bbio(bbio); | |
53b381b3 DW |
2185 | kfree(rbio); |
2186 | return -EIO; | |
2187 | } | |
2188 | ||
4245215d | 2189 | if (generic_io) { |
0b246afa | 2190 | btrfs_bio_counter_inc_noblocked(fs_info); |
4245215d MX |
2191 | rbio->generic_bio_cnt = 1; |
2192 | } else { | |
6e9606d2 | 2193 | btrfs_get_bbio(bbio); |
4245215d MX |
2194 | } |
2195 | ||
53b381b3 | 2196 | /* |
8810f751 LB |
2197 | * Loop retry: |
2198 | * for 'mirror == 2', reconstruct from all other stripes. | |
2199 | * for 'mirror_num > 2', select a stripe to fail on every retry. | |
53b381b3 | 2200 | */ |
8810f751 LB |
2201 | if (mirror_num > 2) { |
2202 | /* | |
2203 | * 'mirror == 3' is to fail the p stripe and | |
2204 | * reconstruct from the q stripe. 'mirror > 3' is to | |
2205 | * fail a data stripe and reconstruct from p+q stripe. | |
2206 | */ | |
2207 | rbio->failb = rbio->real_stripes - (mirror_num - 1); | |
2208 | ASSERT(rbio->failb > 0); | |
2209 | if (rbio->failb <= rbio->faila) | |
2210 | rbio->failb--; | |
2211 | } | |
53b381b3 DW |
2212 | |
2213 | ret = lock_stripe_add(rbio); | |
2214 | ||
2215 | /* | |
2216 | * __raid56_parity_recover will end the bio with | |
2217 | * any errors it hits. We don't want to return | |
2218 | * its error value up the stack because our caller | |
2219 | * will end up calling bio_endio with any nonzero | |
2220 | * return | |
2221 | */ | |
2222 | if (ret == 0) | |
2223 | __raid56_parity_recover(rbio); | |
2224 | /* | |
2225 | * our rbio has been added to the list of | |
2226 | * rbios that will be handled after the | |
2227 | * currently lock owner is done | |
2228 | */ | |
2229 | return 0; | |
2230 | ||
2231 | } | |
2232 | ||
2233 | static void rmw_work(struct btrfs_work *work) | |
2234 | { | |
2235 | struct btrfs_raid_bio *rbio; | |
2236 | ||
2237 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2238 | raid56_rmw_stripe(rbio); | |
2239 | } | |
2240 | ||
2241 | static void read_rebuild_work(struct btrfs_work *work) | |
2242 | { | |
2243 | struct btrfs_raid_bio *rbio; | |
2244 | ||
2245 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2246 | __raid56_parity_recover(rbio); | |
2247 | } | |
5a6ac9ea MX |
2248 | |
2249 | /* | |
2250 | * The following code is used to scrub/replace the parity stripe | |
2251 | * | |
ae6529c3 QW |
2252 | * Caller must have already increased bio_counter for getting @bbio. |
2253 | * | |
5a6ac9ea MX |
2254 | * Note: We need make sure all the pages that add into the scrub/replace |
2255 | * raid bio are correct and not be changed during the scrub/replace. That | |
2256 | * is those pages just hold metadata or file data with checksum. | |
2257 | */ | |
2258 | ||
2259 | struct btrfs_raid_bio * | |
2ff7e61e | 2260 | raid56_parity_alloc_scrub_rbio(struct btrfs_fs_info *fs_info, struct bio *bio, |
8e5cfb55 ZL |
2261 | struct btrfs_bio *bbio, u64 stripe_len, |
2262 | struct btrfs_device *scrub_dev, | |
5a6ac9ea MX |
2263 | unsigned long *dbitmap, int stripe_nsectors) |
2264 | { | |
2265 | struct btrfs_raid_bio *rbio; | |
2266 | int i; | |
2267 | ||
2ff7e61e | 2268 | rbio = alloc_rbio(fs_info, bbio, stripe_len); |
5a6ac9ea MX |
2269 | if (IS_ERR(rbio)) |
2270 | return NULL; | |
2271 | bio_list_add(&rbio->bio_list, bio); | |
2272 | /* | |
2273 | * This is a special bio which is used to hold the completion handler | |
2274 | * and make the scrub rbio is similar to the other types | |
2275 | */ | |
2276 | ASSERT(!bio->bi_iter.bi_size); | |
2277 | rbio->operation = BTRFS_RBIO_PARITY_SCRUB; | |
2278 | ||
9cd3a7eb LB |
2279 | /* |
2280 | * After mapping bbio with BTRFS_MAP_WRITE, parities have been sorted | |
2281 | * to the end position, so this search can start from the first parity | |
2282 | * stripe. | |
2283 | */ | |
2284 | for (i = rbio->nr_data; i < rbio->real_stripes; i++) { | |
5a6ac9ea MX |
2285 | if (bbio->stripes[i].dev == scrub_dev) { |
2286 | rbio->scrubp = i; | |
2287 | break; | |
2288 | } | |
2289 | } | |
9cd3a7eb | 2290 | ASSERT(i < rbio->real_stripes); |
5a6ac9ea MX |
2291 | |
2292 | /* Now we just support the sectorsize equals to page size */ | |
0b246afa | 2293 | ASSERT(fs_info->sectorsize == PAGE_SIZE); |
5a6ac9ea MX |
2294 | ASSERT(rbio->stripe_npages == stripe_nsectors); |
2295 | bitmap_copy(rbio->dbitmap, dbitmap, stripe_nsectors); | |
2296 | ||
ae6529c3 QW |
2297 | /* |
2298 | * We have already increased bio_counter when getting bbio, record it | |
2299 | * so we can free it at rbio_orig_end_io(). | |
2300 | */ | |
2301 | rbio->generic_bio_cnt = 1; | |
2302 | ||
5a6ac9ea MX |
2303 | return rbio; |
2304 | } | |
2305 | ||
b4ee1782 OS |
2306 | /* Used for both parity scrub and missing. */ |
2307 | void raid56_add_scrub_pages(struct btrfs_raid_bio *rbio, struct page *page, | |
2308 | u64 logical) | |
5a6ac9ea MX |
2309 | { |
2310 | int stripe_offset; | |
2311 | int index; | |
2312 | ||
8e5cfb55 ZL |
2313 | ASSERT(logical >= rbio->bbio->raid_map[0]); |
2314 | ASSERT(logical + PAGE_SIZE <= rbio->bbio->raid_map[0] + | |
5a6ac9ea | 2315 | rbio->stripe_len * rbio->nr_data); |
8e5cfb55 | 2316 | stripe_offset = (int)(logical - rbio->bbio->raid_map[0]); |
09cbfeaf | 2317 | index = stripe_offset >> PAGE_SHIFT; |
5a6ac9ea MX |
2318 | rbio->bio_pages[index] = page; |
2319 | } | |
2320 | ||
2321 | /* | |
2322 | * We just scrub the parity that we have correct data on the same horizontal, | |
2323 | * so we needn't allocate all pages for all the stripes. | |
2324 | */ | |
2325 | static int alloc_rbio_essential_pages(struct btrfs_raid_bio *rbio) | |
2326 | { | |
2327 | int i; | |
2328 | int bit; | |
2329 | int index; | |
2330 | struct page *page; | |
2331 | ||
2332 | for_each_set_bit(bit, rbio->dbitmap, rbio->stripe_npages) { | |
2c8cdd6e | 2333 | for (i = 0; i < rbio->real_stripes; i++) { |
5a6ac9ea MX |
2334 | index = i * rbio->stripe_npages + bit; |
2335 | if (rbio->stripe_pages[index]) | |
2336 | continue; | |
2337 | ||
2338 | page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
2339 | if (!page) | |
2340 | return -ENOMEM; | |
2341 | rbio->stripe_pages[index] = page; | |
5a6ac9ea MX |
2342 | } |
2343 | } | |
2344 | return 0; | |
2345 | } | |
2346 | ||
5a6ac9ea MX |
2347 | static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, |
2348 | int need_check) | |
2349 | { | |
76035976 | 2350 | struct btrfs_bio *bbio = rbio->bbio; |
1389053e KC |
2351 | void **pointers = rbio->finish_pointers; |
2352 | unsigned long *pbitmap = rbio->finish_pbitmap; | |
5a6ac9ea MX |
2353 | int nr_data = rbio->nr_data; |
2354 | int stripe; | |
2355 | int pagenr; | |
2356 | int p_stripe = -1; | |
2357 | int q_stripe = -1; | |
2358 | struct page *p_page = NULL; | |
2359 | struct page *q_page = NULL; | |
2360 | struct bio_list bio_list; | |
2361 | struct bio *bio; | |
76035976 | 2362 | int is_replace = 0; |
5a6ac9ea MX |
2363 | int ret; |
2364 | ||
2365 | bio_list_init(&bio_list); | |
2366 | ||
2c8cdd6e MX |
2367 | if (rbio->real_stripes - rbio->nr_data == 1) { |
2368 | p_stripe = rbio->real_stripes - 1; | |
2369 | } else if (rbio->real_stripes - rbio->nr_data == 2) { | |
2370 | p_stripe = rbio->real_stripes - 2; | |
2371 | q_stripe = rbio->real_stripes - 1; | |
5a6ac9ea MX |
2372 | } else { |
2373 | BUG(); | |
2374 | } | |
2375 | ||
76035976 MX |
2376 | if (bbio->num_tgtdevs && bbio->tgtdev_map[rbio->scrubp]) { |
2377 | is_replace = 1; | |
2378 | bitmap_copy(pbitmap, rbio->dbitmap, rbio->stripe_npages); | |
2379 | } | |
2380 | ||
5a6ac9ea MX |
2381 | /* |
2382 | * Because the higher layers(scrubber) are unlikely to | |
2383 | * use this area of the disk again soon, so don't cache | |
2384 | * it. | |
2385 | */ | |
2386 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
2387 | ||
2388 | if (!need_check) | |
2389 | goto writeback; | |
2390 | ||
2391 | p_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
2392 | if (!p_page) | |
2393 | goto cleanup; | |
2394 | SetPageUptodate(p_page); | |
2395 | ||
2396 | if (q_stripe != -1) { | |
2397 | q_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
2398 | if (!q_page) { | |
2399 | __free_page(p_page); | |
2400 | goto cleanup; | |
2401 | } | |
2402 | SetPageUptodate(q_page); | |
2403 | } | |
2404 | ||
2405 | atomic_set(&rbio->error, 0); | |
2406 | ||
2407 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { | |
2408 | struct page *p; | |
2409 | void *parity; | |
2410 | /* first collect one page from each data stripe */ | |
2411 | for (stripe = 0; stripe < nr_data; stripe++) { | |
2412 | p = page_in_rbio(rbio, stripe, pagenr, 0); | |
2413 | pointers[stripe] = kmap(p); | |
2414 | } | |
2415 | ||
2416 | /* then add the parity stripe */ | |
2417 | pointers[stripe++] = kmap(p_page); | |
2418 | ||
2419 | if (q_stripe != -1) { | |
2420 | ||
2421 | /* | |
2422 | * raid6, add the qstripe and call the | |
2423 | * library function to fill in our p/q | |
2424 | */ | |
2425 | pointers[stripe++] = kmap(q_page); | |
2426 | ||
2c8cdd6e | 2427 | raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE, |
5a6ac9ea MX |
2428 | pointers); |
2429 | } else { | |
2430 | /* raid5 */ | |
69d24804 | 2431 | copy_page(pointers[nr_data], pointers[0]); |
09cbfeaf | 2432 | run_xor(pointers + 1, nr_data - 1, PAGE_SIZE); |
5a6ac9ea MX |
2433 | } |
2434 | ||
01327610 | 2435 | /* Check scrubbing parity and repair it */ |
5a6ac9ea MX |
2436 | p = rbio_stripe_page(rbio, rbio->scrubp, pagenr); |
2437 | parity = kmap(p); | |
09cbfeaf | 2438 | if (memcmp(parity, pointers[rbio->scrubp], PAGE_SIZE)) |
69d24804 | 2439 | copy_page(parity, pointers[rbio->scrubp]); |
5a6ac9ea MX |
2440 | else |
2441 | /* Parity is right, needn't writeback */ | |
2442 | bitmap_clear(rbio->dbitmap, pagenr, 1); | |
2443 | kunmap(p); | |
2444 | ||
3897b6f0 | 2445 | for (stripe = 0; stripe < nr_data; stripe++) |
5a6ac9ea | 2446 | kunmap(page_in_rbio(rbio, stripe, pagenr, 0)); |
3897b6f0 | 2447 | kunmap(p_page); |
5a6ac9ea MX |
2448 | } |
2449 | ||
2450 | __free_page(p_page); | |
2451 | if (q_page) | |
2452 | __free_page(q_page); | |
2453 | ||
2454 | writeback: | |
2455 | /* | |
2456 | * time to start writing. Make bios for everything from the | |
2457 | * higher layers (the bio_list in our rbio) and our p/q. Ignore | |
2458 | * everything else. | |
2459 | */ | |
2460 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { | |
2461 | struct page *page; | |
2462 | ||
2463 | page = rbio_stripe_page(rbio, rbio->scrubp, pagenr); | |
2464 | ret = rbio_add_io_page(rbio, &bio_list, | |
2465 | page, rbio->scrubp, pagenr, rbio->stripe_len); | |
2466 | if (ret) | |
2467 | goto cleanup; | |
2468 | } | |
2469 | ||
76035976 MX |
2470 | if (!is_replace) |
2471 | goto submit_write; | |
2472 | ||
2473 | for_each_set_bit(pagenr, pbitmap, rbio->stripe_npages) { | |
2474 | struct page *page; | |
2475 | ||
2476 | page = rbio_stripe_page(rbio, rbio->scrubp, pagenr); | |
2477 | ret = rbio_add_io_page(rbio, &bio_list, page, | |
2478 | bbio->tgtdev_map[rbio->scrubp], | |
2479 | pagenr, rbio->stripe_len); | |
2480 | if (ret) | |
2481 | goto cleanup; | |
2482 | } | |
2483 | ||
2484 | submit_write: | |
5a6ac9ea MX |
2485 | nr_data = bio_list_size(&bio_list); |
2486 | if (!nr_data) { | |
2487 | /* Every parity is right */ | |
58efbc9f | 2488 | rbio_orig_end_io(rbio, BLK_STS_OK); |
5a6ac9ea MX |
2489 | return; |
2490 | } | |
2491 | ||
2492 | atomic_set(&rbio->stripes_pending, nr_data); | |
2493 | ||
2494 | while (1) { | |
2495 | bio = bio_list_pop(&bio_list); | |
2496 | if (!bio) | |
2497 | break; | |
2498 | ||
2499 | bio->bi_private = rbio; | |
a6111d11 | 2500 | bio->bi_end_io = raid_write_end_io; |
ebcc3263 | 2501 | bio->bi_opf = REQ_OP_WRITE; |
4e49ea4a MC |
2502 | |
2503 | submit_bio(bio); | |
5a6ac9ea MX |
2504 | } |
2505 | return; | |
2506 | ||
2507 | cleanup: | |
58efbc9f | 2508 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2509 | |
2510 | while ((bio = bio_list_pop(&bio_list))) | |
2511 | bio_put(bio); | |
5a6ac9ea MX |
2512 | } |
2513 | ||
2514 | static inline int is_data_stripe(struct btrfs_raid_bio *rbio, int stripe) | |
2515 | { | |
2516 | if (stripe >= 0 && stripe < rbio->nr_data) | |
2517 | return 1; | |
2518 | return 0; | |
2519 | } | |
2520 | ||
2521 | /* | |
2522 | * While we're doing the parity check and repair, we could have errors | |
2523 | * in reading pages off the disk. This checks for errors and if we're | |
2524 | * not able to read the page it'll trigger parity reconstruction. The | |
2525 | * parity scrub will be finished after we've reconstructed the failed | |
2526 | * stripes | |
2527 | */ | |
2528 | static void validate_rbio_for_parity_scrub(struct btrfs_raid_bio *rbio) | |
2529 | { | |
2530 | if (atomic_read(&rbio->error) > rbio->bbio->max_errors) | |
2531 | goto cleanup; | |
2532 | ||
2533 | if (rbio->faila >= 0 || rbio->failb >= 0) { | |
2534 | int dfail = 0, failp = -1; | |
2535 | ||
2536 | if (is_data_stripe(rbio, rbio->faila)) | |
2537 | dfail++; | |
2538 | else if (is_parity_stripe(rbio->faila)) | |
2539 | failp = rbio->faila; | |
2540 | ||
2541 | if (is_data_stripe(rbio, rbio->failb)) | |
2542 | dfail++; | |
2543 | else if (is_parity_stripe(rbio->failb)) | |
2544 | failp = rbio->failb; | |
2545 | ||
2546 | /* | |
2547 | * Because we can not use a scrubbing parity to repair | |
2548 | * the data, so the capability of the repair is declined. | |
2549 | * (In the case of RAID5, we can not repair anything) | |
2550 | */ | |
2551 | if (dfail > rbio->bbio->max_errors - 1) | |
2552 | goto cleanup; | |
2553 | ||
2554 | /* | |
2555 | * If all data is good, only parity is correctly, just | |
2556 | * repair the parity. | |
2557 | */ | |
2558 | if (dfail == 0) { | |
2559 | finish_parity_scrub(rbio, 0); | |
2560 | return; | |
2561 | } | |
2562 | ||
2563 | /* | |
2564 | * Here means we got one corrupted data stripe and one | |
2565 | * corrupted parity on RAID6, if the corrupted parity | |
01327610 | 2566 | * is scrubbing parity, luckily, use the other one to repair |
5a6ac9ea MX |
2567 | * the data, or we can not repair the data stripe. |
2568 | */ | |
2569 | if (failp != rbio->scrubp) | |
2570 | goto cleanup; | |
2571 | ||
2572 | __raid_recover_end_io(rbio); | |
2573 | } else { | |
2574 | finish_parity_scrub(rbio, 1); | |
2575 | } | |
2576 | return; | |
2577 | ||
2578 | cleanup: | |
58efbc9f | 2579 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
5a6ac9ea MX |
2580 | } |
2581 | ||
2582 | /* | |
2583 | * end io for the read phase of the rmw cycle. All the bios here are physical | |
2584 | * stripe bios we've read from the disk so we can recalculate the parity of the | |
2585 | * stripe. | |
2586 | * | |
2587 | * This will usually kick off finish_rmw once all the bios are read in, but it | |
2588 | * may trigger parity reconstruction if we had any errors along the way | |
2589 | */ | |
4246a0b6 | 2590 | static void raid56_parity_scrub_end_io(struct bio *bio) |
5a6ac9ea MX |
2591 | { |
2592 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
2593 | ||
4e4cbee9 | 2594 | if (bio->bi_status) |
5a6ac9ea MX |
2595 | fail_bio_stripe(rbio, bio); |
2596 | else | |
2597 | set_bio_pages_uptodate(bio); | |
2598 | ||
2599 | bio_put(bio); | |
2600 | ||
2601 | if (!atomic_dec_and_test(&rbio->stripes_pending)) | |
2602 | return; | |
2603 | ||
2604 | /* | |
2605 | * this will normally call finish_rmw to start our write | |
2606 | * but if there are any failed stripes we'll reconstruct | |
2607 | * from parity first | |
2608 | */ | |
2609 | validate_rbio_for_parity_scrub(rbio); | |
2610 | } | |
2611 | ||
2612 | static void raid56_parity_scrub_stripe(struct btrfs_raid_bio *rbio) | |
2613 | { | |
2614 | int bios_to_read = 0; | |
5a6ac9ea MX |
2615 | struct bio_list bio_list; |
2616 | int ret; | |
2617 | int pagenr; | |
2618 | int stripe; | |
2619 | struct bio *bio; | |
2620 | ||
785884fc LB |
2621 | bio_list_init(&bio_list); |
2622 | ||
5a6ac9ea MX |
2623 | ret = alloc_rbio_essential_pages(rbio); |
2624 | if (ret) | |
2625 | goto cleanup; | |
2626 | ||
5a6ac9ea MX |
2627 | atomic_set(&rbio->error, 0); |
2628 | /* | |
2629 | * build a list of bios to read all the missing parts of this | |
2630 | * stripe | |
2631 | */ | |
2c8cdd6e | 2632 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
5a6ac9ea MX |
2633 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { |
2634 | struct page *page; | |
2635 | /* | |
2636 | * we want to find all the pages missing from | |
2637 | * the rbio and read them from the disk. If | |
2638 | * page_in_rbio finds a page in the bio list | |
2639 | * we don't need to read it off the stripe. | |
2640 | */ | |
2641 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
2642 | if (page) | |
2643 | continue; | |
2644 | ||
2645 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
2646 | /* | |
2647 | * the bio cache may have handed us an uptodate | |
2648 | * page. If so, be happy and use it | |
2649 | */ | |
2650 | if (PageUptodate(page)) | |
2651 | continue; | |
2652 | ||
2653 | ret = rbio_add_io_page(rbio, &bio_list, page, | |
2654 | stripe, pagenr, rbio->stripe_len); | |
2655 | if (ret) | |
2656 | goto cleanup; | |
2657 | } | |
2658 | } | |
2659 | ||
2660 | bios_to_read = bio_list_size(&bio_list); | |
2661 | if (!bios_to_read) { | |
2662 | /* | |
2663 | * this can happen if others have merged with | |
2664 | * us, it means there is nothing left to read. | |
2665 | * But if there are missing devices it may not be | |
2666 | * safe to do the full stripe write yet. | |
2667 | */ | |
2668 | goto finish; | |
2669 | } | |
2670 | ||
2671 | /* | |
2672 | * the bbio may be freed once we submit the last bio. Make sure | |
2673 | * not to touch it after that | |
2674 | */ | |
2675 | atomic_set(&rbio->stripes_pending, bios_to_read); | |
2676 | while (1) { | |
2677 | bio = bio_list_pop(&bio_list); | |
2678 | if (!bio) | |
2679 | break; | |
2680 | ||
2681 | bio->bi_private = rbio; | |
2682 | bio->bi_end_io = raid56_parity_scrub_end_io; | |
ebcc3263 | 2683 | bio->bi_opf = REQ_OP_READ; |
5a6ac9ea | 2684 | |
0b246afa | 2685 | btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
5a6ac9ea | 2686 | |
4e49ea4a | 2687 | submit_bio(bio); |
5a6ac9ea MX |
2688 | } |
2689 | /* the actual write will happen once the reads are done */ | |
2690 | return; | |
2691 | ||
2692 | cleanup: | |
58efbc9f | 2693 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2694 | |
2695 | while ((bio = bio_list_pop(&bio_list))) | |
2696 | bio_put(bio); | |
2697 | ||
5a6ac9ea MX |
2698 | return; |
2699 | ||
2700 | finish: | |
2701 | validate_rbio_for_parity_scrub(rbio); | |
2702 | } | |
2703 | ||
2704 | static void scrub_parity_work(struct btrfs_work *work) | |
2705 | { | |
2706 | struct btrfs_raid_bio *rbio; | |
2707 | ||
2708 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2709 | raid56_parity_scrub_stripe(rbio); | |
2710 | } | |
2711 | ||
5a6ac9ea MX |
2712 | void raid56_parity_submit_scrub_rbio(struct btrfs_raid_bio *rbio) |
2713 | { | |
2714 | if (!lock_stripe_add(rbio)) | |
a81b747d | 2715 | start_async_work(rbio, scrub_parity_work); |
5a6ac9ea | 2716 | } |
b4ee1782 OS |
2717 | |
2718 | /* The following code is used for dev replace of a missing RAID 5/6 device. */ | |
2719 | ||
2720 | struct btrfs_raid_bio * | |
2ff7e61e | 2721 | raid56_alloc_missing_rbio(struct btrfs_fs_info *fs_info, struct bio *bio, |
b4ee1782 OS |
2722 | struct btrfs_bio *bbio, u64 length) |
2723 | { | |
2724 | struct btrfs_raid_bio *rbio; | |
2725 | ||
2ff7e61e | 2726 | rbio = alloc_rbio(fs_info, bbio, length); |
b4ee1782 OS |
2727 | if (IS_ERR(rbio)) |
2728 | return NULL; | |
2729 | ||
2730 | rbio->operation = BTRFS_RBIO_REBUILD_MISSING; | |
2731 | bio_list_add(&rbio->bio_list, bio); | |
2732 | /* | |
2733 | * This is a special bio which is used to hold the completion handler | |
2734 | * and make the scrub rbio is similar to the other types | |
2735 | */ | |
2736 | ASSERT(!bio->bi_iter.bi_size); | |
2737 | ||
2738 | rbio->faila = find_logical_bio_stripe(rbio, bio); | |
2739 | if (rbio->faila == -1) { | |
2740 | BUG(); | |
2741 | kfree(rbio); | |
2742 | return NULL; | |
2743 | } | |
2744 | ||
ae6529c3 QW |
2745 | /* |
2746 | * When we get bbio, we have already increased bio_counter, record it | |
2747 | * so we can free it at rbio_orig_end_io() | |
2748 | */ | |
2749 | rbio->generic_bio_cnt = 1; | |
2750 | ||
b4ee1782 OS |
2751 | return rbio; |
2752 | } | |
2753 | ||
b4ee1782 OS |
2754 | void raid56_submit_missing_rbio(struct btrfs_raid_bio *rbio) |
2755 | { | |
2756 | if (!lock_stripe_add(rbio)) | |
e66d8d5a | 2757 | start_async_work(rbio, read_rebuild_work); |
b4ee1782 | 2758 | } |