Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
53b381b3 DW |
2 | /* |
3 | * Copyright (C) 2012 Fusion-io All rights reserved. | |
4 | * Copyright (C) 2012 Intel Corp. All rights reserved. | |
53b381b3 | 5 | */ |
c1d7c514 | 6 | |
53b381b3 | 7 | #include <linux/sched.h> |
53b381b3 DW |
8 | #include <linux/bio.h> |
9 | #include <linux/slab.h> | |
53b381b3 | 10 | #include <linux/blkdev.h> |
53b381b3 DW |
11 | #include <linux/raid/pq.h> |
12 | #include <linux/hash.h> | |
13 | #include <linux/list_sort.h> | |
14 | #include <linux/raid/xor.h> | |
818e010b | 15 | #include <linux/mm.h> |
cea62800 | 16 | #include "misc.h" |
53b381b3 | 17 | #include "ctree.h" |
53b381b3 | 18 | #include "disk-io.h" |
53b381b3 DW |
19 | #include "volumes.h" |
20 | #include "raid56.h" | |
21 | #include "async-thread.h" | |
53b381b3 DW |
22 | |
23 | /* set when additional merges to this rbio are not allowed */ | |
24 | #define RBIO_RMW_LOCKED_BIT 1 | |
25 | ||
4ae10b3a CM |
26 | /* |
27 | * set when this rbio is sitting in the hash, but it is just a cache | |
28 | * of past RMW | |
29 | */ | |
30 | #define RBIO_CACHE_BIT 2 | |
31 | ||
32 | /* | |
33 | * set when it is safe to trust the stripe_pages for caching | |
34 | */ | |
35 | #define RBIO_CACHE_READY_BIT 3 | |
36 | ||
4ae10b3a CM |
37 | #define RBIO_CACHE_SIZE 1024 |
38 | ||
8a953348 DS |
39 | #define BTRFS_STRIPE_HASH_TABLE_BITS 11 |
40 | ||
41 | /* Used by the raid56 code to lock stripes for read/modify/write */ | |
42 | struct btrfs_stripe_hash { | |
43 | struct list_head hash_list; | |
44 | spinlock_t lock; | |
45 | }; | |
46 | ||
47 | /* Used by the raid56 code to lock stripes for read/modify/write */ | |
48 | struct btrfs_stripe_hash_table { | |
49 | struct list_head stripe_cache; | |
50 | spinlock_t cache_lock; | |
51 | int cache_size; | |
52 | struct btrfs_stripe_hash table[]; | |
53 | }; | |
54 | ||
1b94b556 | 55 | enum btrfs_rbio_ops { |
b4ee1782 OS |
56 | BTRFS_RBIO_WRITE, |
57 | BTRFS_RBIO_READ_REBUILD, | |
58 | BTRFS_RBIO_PARITY_SCRUB, | |
59 | BTRFS_RBIO_REBUILD_MISSING, | |
1b94b556 MX |
60 | }; |
61 | ||
53b381b3 | 62 | struct btrfs_raid_bio { |
4c664611 | 63 | struct btrfs_io_context *bioc; |
53b381b3 | 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); |
6a258d72 | 194 | btrfs_queue_work(rbio->bioc->fs_info->rmw_workers, &rbio->work); |
ac638859 DS |
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; | |
209 | ||
210 | if (info->stripe_hash_table) | |
211 | return 0; | |
212 | ||
83c8266a DS |
213 | /* |
214 | * The table is large, starting with order 4 and can go as high as | |
215 | * order 7 in case lock debugging is turned on. | |
216 | * | |
217 | * Try harder to allocate and fallback to vmalloc to lower the chance | |
218 | * of a failing mount. | |
219 | */ | |
ee787f95 | 220 | table = kvzalloc(struct_size(table, table, num_entries), GFP_KERNEL); |
818e010b DS |
221 | if (!table) |
222 | return -ENOMEM; | |
53b381b3 | 223 | |
4ae10b3a CM |
224 | spin_lock_init(&table->cache_lock); |
225 | INIT_LIST_HEAD(&table->stripe_cache); | |
226 | ||
53b381b3 DW |
227 | h = table->table; |
228 | ||
229 | for (i = 0; i < num_entries; i++) { | |
230 | cur = h + i; | |
231 | INIT_LIST_HEAD(&cur->hash_list); | |
232 | spin_lock_init(&cur->lock); | |
53b381b3 DW |
233 | } |
234 | ||
235 | x = cmpxchg(&info->stripe_hash_table, NULL, table); | |
fe3b7bb0 | 236 | kvfree(x); |
53b381b3 DW |
237 | return 0; |
238 | } | |
239 | ||
4ae10b3a CM |
240 | /* |
241 | * caching an rbio means to copy anything from the | |
242 | * bio_pages array into the stripe_pages array. We | |
243 | * use the page uptodate bit in the stripe cache array | |
244 | * to indicate if it has valid data | |
245 | * | |
246 | * once the caching is done, we set the cache ready | |
247 | * bit. | |
248 | */ | |
249 | static void cache_rbio_pages(struct btrfs_raid_bio *rbio) | |
250 | { | |
251 | int i; | |
4ae10b3a CM |
252 | int ret; |
253 | ||
254 | ret = alloc_rbio_pages(rbio); | |
255 | if (ret) | |
256 | return; | |
257 | ||
258 | for (i = 0; i < rbio->nr_pages; i++) { | |
259 | if (!rbio->bio_pages[i]) | |
260 | continue; | |
261 | ||
80cc8384 | 262 | copy_highpage(rbio->stripe_pages[i], rbio->bio_pages[i]); |
4ae10b3a CM |
263 | SetPageUptodate(rbio->stripe_pages[i]); |
264 | } | |
265 | set_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
266 | } | |
267 | ||
53b381b3 DW |
268 | /* |
269 | * we hash on the first logical address of the stripe | |
270 | */ | |
271 | static int rbio_bucket(struct btrfs_raid_bio *rbio) | |
272 | { | |
4c664611 | 273 | u64 num = rbio->bioc->raid_map[0]; |
53b381b3 DW |
274 | |
275 | /* | |
276 | * we shift down quite a bit. We're using byte | |
277 | * addressing, and most of the lower bits are zeros. | |
278 | * This tends to upset hash_64, and it consistently | |
279 | * returns just one or two different values. | |
280 | * | |
281 | * shifting off the lower bits fixes things. | |
282 | */ | |
283 | return hash_64(num >> 16, BTRFS_STRIPE_HASH_TABLE_BITS); | |
284 | } | |
285 | ||
4ae10b3a CM |
286 | /* |
287 | * stealing an rbio means taking all the uptodate pages from the stripe | |
288 | * array in the source rbio and putting them into the destination rbio | |
289 | */ | |
290 | static void steal_rbio(struct btrfs_raid_bio *src, struct btrfs_raid_bio *dest) | |
291 | { | |
292 | int i; | |
293 | struct page *s; | |
294 | struct page *d; | |
295 | ||
296 | if (!test_bit(RBIO_CACHE_READY_BIT, &src->flags)) | |
297 | return; | |
298 | ||
299 | for (i = 0; i < dest->nr_pages; i++) { | |
300 | s = src->stripe_pages[i]; | |
301 | if (!s || !PageUptodate(s)) { | |
302 | continue; | |
303 | } | |
304 | ||
305 | d = dest->stripe_pages[i]; | |
306 | if (d) | |
307 | __free_page(d); | |
308 | ||
309 | dest->stripe_pages[i] = s; | |
310 | src->stripe_pages[i] = NULL; | |
311 | } | |
312 | } | |
313 | ||
53b381b3 DW |
314 | /* |
315 | * merging means we take the bio_list from the victim and | |
316 | * splice it into the destination. The victim should | |
317 | * be discarded afterwards. | |
318 | * | |
319 | * must be called with dest->rbio_list_lock held | |
320 | */ | |
321 | static void merge_rbio(struct btrfs_raid_bio *dest, | |
322 | struct btrfs_raid_bio *victim) | |
323 | { | |
324 | bio_list_merge(&dest->bio_list, &victim->bio_list); | |
325 | dest->bio_list_bytes += victim->bio_list_bytes; | |
4245215d | 326 | dest->generic_bio_cnt += victim->generic_bio_cnt; |
53b381b3 DW |
327 | bio_list_init(&victim->bio_list); |
328 | } | |
329 | ||
330 | /* | |
4ae10b3a CM |
331 | * used to prune items that are in the cache. The caller |
332 | * must hold the hash table lock. | |
333 | */ | |
334 | static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio) | |
335 | { | |
336 | int bucket = rbio_bucket(rbio); | |
337 | struct btrfs_stripe_hash_table *table; | |
338 | struct btrfs_stripe_hash *h; | |
339 | int freeit = 0; | |
340 | ||
341 | /* | |
342 | * check the bit again under the hash table lock. | |
343 | */ | |
344 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
345 | return; | |
346 | ||
6a258d72 | 347 | table = rbio->bioc->fs_info->stripe_hash_table; |
4ae10b3a CM |
348 | h = table->table + bucket; |
349 | ||
350 | /* hold the lock for the bucket because we may be | |
351 | * removing it from the hash table | |
352 | */ | |
353 | spin_lock(&h->lock); | |
354 | ||
355 | /* | |
356 | * hold the lock for the bio list because we need | |
357 | * to make sure the bio list is empty | |
358 | */ | |
359 | spin_lock(&rbio->bio_list_lock); | |
360 | ||
361 | if (test_and_clear_bit(RBIO_CACHE_BIT, &rbio->flags)) { | |
362 | list_del_init(&rbio->stripe_cache); | |
363 | table->cache_size -= 1; | |
364 | freeit = 1; | |
365 | ||
366 | /* if the bio list isn't empty, this rbio is | |
367 | * still involved in an IO. We take it out | |
368 | * of the cache list, and drop the ref that | |
369 | * was held for the list. | |
370 | * | |
371 | * If the bio_list was empty, we also remove | |
372 | * the rbio from the hash_table, and drop | |
373 | * the corresponding ref | |
374 | */ | |
375 | if (bio_list_empty(&rbio->bio_list)) { | |
376 | if (!list_empty(&rbio->hash_list)) { | |
377 | list_del_init(&rbio->hash_list); | |
dec95574 | 378 | refcount_dec(&rbio->refs); |
4ae10b3a CM |
379 | BUG_ON(!list_empty(&rbio->plug_list)); |
380 | } | |
381 | } | |
382 | } | |
383 | ||
384 | spin_unlock(&rbio->bio_list_lock); | |
385 | spin_unlock(&h->lock); | |
386 | ||
387 | if (freeit) | |
388 | __free_raid_bio(rbio); | |
389 | } | |
390 | ||
391 | /* | |
392 | * prune a given rbio from the cache | |
393 | */ | |
394 | static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio) | |
395 | { | |
396 | struct btrfs_stripe_hash_table *table; | |
397 | unsigned long flags; | |
398 | ||
399 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
400 | return; | |
401 | ||
6a258d72 | 402 | table = rbio->bioc->fs_info->stripe_hash_table; |
4ae10b3a CM |
403 | |
404 | spin_lock_irqsave(&table->cache_lock, flags); | |
405 | __remove_rbio_from_cache(rbio); | |
406 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
407 | } | |
408 | ||
409 | /* | |
410 | * remove everything in the cache | |
411 | */ | |
48a3b636 | 412 | static void btrfs_clear_rbio_cache(struct btrfs_fs_info *info) |
4ae10b3a CM |
413 | { |
414 | struct btrfs_stripe_hash_table *table; | |
415 | unsigned long flags; | |
416 | struct btrfs_raid_bio *rbio; | |
417 | ||
418 | table = info->stripe_hash_table; | |
419 | ||
420 | spin_lock_irqsave(&table->cache_lock, flags); | |
421 | while (!list_empty(&table->stripe_cache)) { | |
422 | rbio = list_entry(table->stripe_cache.next, | |
423 | struct btrfs_raid_bio, | |
424 | stripe_cache); | |
425 | __remove_rbio_from_cache(rbio); | |
426 | } | |
427 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
428 | } | |
429 | ||
430 | /* | |
431 | * remove all cached entries and free the hash table | |
432 | * used by unmount | |
53b381b3 DW |
433 | */ |
434 | void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info) | |
435 | { | |
436 | if (!info->stripe_hash_table) | |
437 | return; | |
4ae10b3a | 438 | btrfs_clear_rbio_cache(info); |
f749303b | 439 | kvfree(info->stripe_hash_table); |
53b381b3 DW |
440 | info->stripe_hash_table = NULL; |
441 | } | |
442 | ||
4ae10b3a CM |
443 | /* |
444 | * insert an rbio into the stripe cache. It | |
445 | * must have already been prepared by calling | |
446 | * cache_rbio_pages | |
447 | * | |
448 | * If this rbio was already cached, it gets | |
449 | * moved to the front of the lru. | |
450 | * | |
451 | * If the size of the rbio cache is too big, we | |
452 | * prune an item. | |
453 | */ | |
454 | static void cache_rbio(struct btrfs_raid_bio *rbio) | |
455 | { | |
456 | struct btrfs_stripe_hash_table *table; | |
457 | unsigned long flags; | |
458 | ||
459 | if (!test_bit(RBIO_CACHE_READY_BIT, &rbio->flags)) | |
460 | return; | |
461 | ||
6a258d72 | 462 | table = rbio->bioc->fs_info->stripe_hash_table; |
4ae10b3a CM |
463 | |
464 | spin_lock_irqsave(&table->cache_lock, flags); | |
465 | spin_lock(&rbio->bio_list_lock); | |
466 | ||
467 | /* bump our ref if we were not in the list before */ | |
468 | if (!test_and_set_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
dec95574 | 469 | refcount_inc(&rbio->refs); |
4ae10b3a CM |
470 | |
471 | if (!list_empty(&rbio->stripe_cache)){ | |
472 | list_move(&rbio->stripe_cache, &table->stripe_cache); | |
473 | } else { | |
474 | list_add(&rbio->stripe_cache, &table->stripe_cache); | |
475 | table->cache_size += 1; | |
476 | } | |
477 | ||
478 | spin_unlock(&rbio->bio_list_lock); | |
479 | ||
480 | if (table->cache_size > RBIO_CACHE_SIZE) { | |
481 | struct btrfs_raid_bio *found; | |
482 | ||
483 | found = list_entry(table->stripe_cache.prev, | |
484 | struct btrfs_raid_bio, | |
485 | stripe_cache); | |
486 | ||
487 | if (found != rbio) | |
488 | __remove_rbio_from_cache(found); | |
489 | } | |
490 | ||
491 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
4ae10b3a CM |
492 | } |
493 | ||
53b381b3 DW |
494 | /* |
495 | * helper function to run the xor_blocks api. It is only | |
496 | * able to do MAX_XOR_BLOCKS at a time, so we need to | |
497 | * loop through. | |
498 | */ | |
499 | static void run_xor(void **pages, int src_cnt, ssize_t len) | |
500 | { | |
501 | int src_off = 0; | |
502 | int xor_src_cnt = 0; | |
503 | void *dest = pages[src_cnt]; | |
504 | ||
505 | while(src_cnt > 0) { | |
506 | xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS); | |
507 | xor_blocks(xor_src_cnt, len, dest, pages + src_off); | |
508 | ||
509 | src_cnt -= xor_src_cnt; | |
510 | src_off += xor_src_cnt; | |
511 | } | |
512 | } | |
513 | ||
514 | /* | |
176571a1 DS |
515 | * Returns true if the bio list inside this rbio covers an entire stripe (no |
516 | * rmw required). | |
53b381b3 | 517 | */ |
176571a1 | 518 | static int rbio_is_full(struct btrfs_raid_bio *rbio) |
53b381b3 | 519 | { |
176571a1 | 520 | unsigned long flags; |
53b381b3 DW |
521 | unsigned long size = rbio->bio_list_bytes; |
522 | int ret = 1; | |
523 | ||
176571a1 | 524 | spin_lock_irqsave(&rbio->bio_list_lock, flags); |
53b381b3 DW |
525 | if (size != rbio->nr_data * rbio->stripe_len) |
526 | ret = 0; | |
53b381b3 | 527 | BUG_ON(size > rbio->nr_data * rbio->stripe_len); |
53b381b3 | 528 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); |
176571a1 | 529 | |
53b381b3 DW |
530 | return ret; |
531 | } | |
532 | ||
533 | /* | |
534 | * returns 1 if it is safe to merge two rbios together. | |
535 | * The merging is safe if the two rbios correspond to | |
536 | * the same stripe and if they are both going in the same | |
537 | * direction (read vs write), and if neither one is | |
538 | * locked for final IO | |
539 | * | |
540 | * The caller is responsible for locking such that | |
541 | * rmw_locked is safe to test | |
542 | */ | |
543 | static int rbio_can_merge(struct btrfs_raid_bio *last, | |
544 | struct btrfs_raid_bio *cur) | |
545 | { | |
546 | if (test_bit(RBIO_RMW_LOCKED_BIT, &last->flags) || | |
547 | test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) | |
548 | return 0; | |
549 | ||
4ae10b3a CM |
550 | /* |
551 | * we can't merge with cached rbios, since the | |
552 | * idea is that when we merge the destination | |
553 | * rbio is going to run our IO for us. We can | |
01327610 | 554 | * steal from cached rbios though, other functions |
4ae10b3a CM |
555 | * handle that. |
556 | */ | |
557 | if (test_bit(RBIO_CACHE_BIT, &last->flags) || | |
558 | test_bit(RBIO_CACHE_BIT, &cur->flags)) | |
559 | return 0; | |
560 | ||
4c664611 | 561 | if (last->bioc->raid_map[0] != cur->bioc->raid_map[0]) |
53b381b3 DW |
562 | return 0; |
563 | ||
5a6ac9ea MX |
564 | /* we can't merge with different operations */ |
565 | if (last->operation != cur->operation) | |
566 | return 0; | |
567 | /* | |
568 | * We've need read the full stripe from the drive. | |
569 | * check and repair the parity and write the new results. | |
570 | * | |
571 | * We're not allowed to add any new bios to the | |
572 | * bio list here, anyone else that wants to | |
573 | * change this stripe needs to do their own rmw. | |
574 | */ | |
db34be19 | 575 | if (last->operation == BTRFS_RBIO_PARITY_SCRUB) |
53b381b3 | 576 | return 0; |
53b381b3 | 577 | |
db34be19 | 578 | if (last->operation == BTRFS_RBIO_REBUILD_MISSING) |
b4ee1782 OS |
579 | return 0; |
580 | ||
cc54ff62 LB |
581 | if (last->operation == BTRFS_RBIO_READ_REBUILD) { |
582 | int fa = last->faila; | |
583 | int fb = last->failb; | |
584 | int cur_fa = cur->faila; | |
585 | int cur_fb = cur->failb; | |
586 | ||
587 | if (last->faila >= last->failb) { | |
588 | fa = last->failb; | |
589 | fb = last->faila; | |
590 | } | |
591 | ||
592 | if (cur->faila >= cur->failb) { | |
593 | cur_fa = cur->failb; | |
594 | cur_fb = cur->faila; | |
595 | } | |
596 | ||
597 | if (fa != cur_fa || fb != cur_fb) | |
598 | return 0; | |
599 | } | |
53b381b3 DW |
600 | return 1; |
601 | } | |
602 | ||
b7178a5f ZL |
603 | static int rbio_stripe_page_index(struct btrfs_raid_bio *rbio, int stripe, |
604 | int index) | |
605 | { | |
606 | return stripe * rbio->stripe_npages + index; | |
607 | } | |
608 | ||
609 | /* | |
610 | * these are just the pages from the rbio array, not from anything | |
611 | * the FS sent down to us | |
612 | */ | |
613 | static struct page *rbio_stripe_page(struct btrfs_raid_bio *rbio, int stripe, | |
614 | int index) | |
615 | { | |
616 | return rbio->stripe_pages[rbio_stripe_page_index(rbio, stripe, index)]; | |
617 | } | |
618 | ||
53b381b3 DW |
619 | /* |
620 | * helper to index into the pstripe | |
621 | */ | |
622 | static struct page *rbio_pstripe_page(struct btrfs_raid_bio *rbio, int index) | |
623 | { | |
b7178a5f | 624 | return rbio_stripe_page(rbio, rbio->nr_data, index); |
53b381b3 DW |
625 | } |
626 | ||
627 | /* | |
628 | * helper to index into the qstripe, returns null | |
629 | * if there is no qstripe | |
630 | */ | |
631 | static struct page *rbio_qstripe_page(struct btrfs_raid_bio *rbio, int index) | |
632 | { | |
2c8cdd6e | 633 | if (rbio->nr_data + 1 == rbio->real_stripes) |
53b381b3 | 634 | return NULL; |
b7178a5f | 635 | return rbio_stripe_page(rbio, rbio->nr_data + 1, index); |
53b381b3 DW |
636 | } |
637 | ||
638 | /* | |
639 | * The first stripe in the table for a logical address | |
640 | * has the lock. rbios are added in one of three ways: | |
641 | * | |
642 | * 1) Nobody has the stripe locked yet. The rbio is given | |
643 | * the lock and 0 is returned. The caller must start the IO | |
644 | * themselves. | |
645 | * | |
646 | * 2) Someone has the stripe locked, but we're able to merge | |
647 | * with the lock owner. The rbio is freed and the IO will | |
648 | * start automatically along with the existing rbio. 1 is returned. | |
649 | * | |
650 | * 3) Someone has the stripe locked, but we're not able to merge. | |
651 | * The rbio is added to the lock owner's plug list, or merged into | |
652 | * an rbio already on the plug list. When the lock owner unlocks, | |
653 | * the next rbio on the list is run and the IO is started automatically. | |
654 | * 1 is returned | |
655 | * | |
656 | * If we return 0, the caller still owns the rbio and must continue with | |
657 | * IO submission. If we return 1, the caller must assume the rbio has | |
658 | * already been freed. | |
659 | */ | |
660 | static noinline int lock_stripe_add(struct btrfs_raid_bio *rbio) | |
661 | { | |
721860d5 | 662 | struct btrfs_stripe_hash *h; |
53b381b3 DW |
663 | struct btrfs_raid_bio *cur; |
664 | struct btrfs_raid_bio *pending; | |
665 | unsigned long flags; | |
53b381b3 | 666 | struct btrfs_raid_bio *freeit = NULL; |
4ae10b3a | 667 | struct btrfs_raid_bio *cache_drop = NULL; |
53b381b3 | 668 | int ret = 0; |
53b381b3 | 669 | |
6a258d72 | 670 | h = rbio->bioc->fs_info->stripe_hash_table->table + rbio_bucket(rbio); |
721860d5 | 671 | |
53b381b3 DW |
672 | spin_lock_irqsave(&h->lock, flags); |
673 | list_for_each_entry(cur, &h->hash_list, hash_list) { | |
4c664611 | 674 | if (cur->bioc->raid_map[0] != rbio->bioc->raid_map[0]) |
9d6cb1b0 | 675 | continue; |
4ae10b3a | 676 | |
9d6cb1b0 | 677 | spin_lock(&cur->bio_list_lock); |
4ae10b3a | 678 | |
9d6cb1b0 JT |
679 | /* Can we steal this cached rbio's pages? */ |
680 | if (bio_list_empty(&cur->bio_list) && | |
681 | list_empty(&cur->plug_list) && | |
682 | test_bit(RBIO_CACHE_BIT, &cur->flags) && | |
683 | !test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) { | |
684 | list_del_init(&cur->hash_list); | |
685 | refcount_dec(&cur->refs); | |
53b381b3 | 686 | |
9d6cb1b0 JT |
687 | steal_rbio(cur, rbio); |
688 | cache_drop = cur; | |
689 | spin_unlock(&cur->bio_list_lock); | |
4ae10b3a | 690 | |
9d6cb1b0 JT |
691 | goto lockit; |
692 | } | |
53b381b3 | 693 | |
9d6cb1b0 JT |
694 | /* Can we merge into the lock owner? */ |
695 | if (rbio_can_merge(cur, rbio)) { | |
696 | merge_rbio(cur, rbio); | |
53b381b3 | 697 | spin_unlock(&cur->bio_list_lock); |
9d6cb1b0 | 698 | freeit = rbio; |
53b381b3 DW |
699 | ret = 1; |
700 | goto out; | |
701 | } | |
9d6cb1b0 JT |
702 | |
703 | ||
704 | /* | |
705 | * We couldn't merge with the running rbio, see if we can merge | |
706 | * with the pending ones. We don't have to check for rmw_locked | |
707 | * because there is no way they are inside finish_rmw right now | |
708 | */ | |
709 | list_for_each_entry(pending, &cur->plug_list, plug_list) { | |
710 | if (rbio_can_merge(pending, rbio)) { | |
711 | merge_rbio(pending, rbio); | |
712 | spin_unlock(&cur->bio_list_lock); | |
713 | freeit = rbio; | |
714 | ret = 1; | |
715 | goto out; | |
716 | } | |
717 | } | |
718 | ||
719 | /* | |
720 | * No merging, put us on the tail of the plug list, our rbio | |
721 | * will be started with the currently running rbio unlocks | |
722 | */ | |
723 | list_add_tail(&rbio->plug_list, &cur->plug_list); | |
724 | spin_unlock(&cur->bio_list_lock); | |
725 | ret = 1; | |
726 | goto out; | |
53b381b3 | 727 | } |
4ae10b3a | 728 | lockit: |
dec95574 | 729 | refcount_inc(&rbio->refs); |
53b381b3 DW |
730 | list_add(&rbio->hash_list, &h->hash_list); |
731 | out: | |
732 | spin_unlock_irqrestore(&h->lock, flags); | |
4ae10b3a CM |
733 | if (cache_drop) |
734 | remove_rbio_from_cache(cache_drop); | |
53b381b3 DW |
735 | if (freeit) |
736 | __free_raid_bio(freeit); | |
737 | return ret; | |
738 | } | |
739 | ||
740 | /* | |
741 | * called as rmw or parity rebuild is completed. If the plug list has more | |
742 | * rbios waiting for this stripe, the next one on the list will be started | |
743 | */ | |
744 | static noinline void unlock_stripe(struct btrfs_raid_bio *rbio) | |
745 | { | |
746 | int bucket; | |
747 | struct btrfs_stripe_hash *h; | |
748 | unsigned long flags; | |
4ae10b3a | 749 | int keep_cache = 0; |
53b381b3 DW |
750 | |
751 | bucket = rbio_bucket(rbio); | |
6a258d72 | 752 | h = rbio->bioc->fs_info->stripe_hash_table->table + bucket; |
53b381b3 | 753 | |
4ae10b3a CM |
754 | if (list_empty(&rbio->plug_list)) |
755 | cache_rbio(rbio); | |
756 | ||
53b381b3 DW |
757 | spin_lock_irqsave(&h->lock, flags); |
758 | spin_lock(&rbio->bio_list_lock); | |
759 | ||
760 | if (!list_empty(&rbio->hash_list)) { | |
4ae10b3a CM |
761 | /* |
762 | * if we're still cached and there is no other IO | |
763 | * to perform, just leave this rbio here for others | |
764 | * to steal from later | |
765 | */ | |
766 | if (list_empty(&rbio->plug_list) && | |
767 | test_bit(RBIO_CACHE_BIT, &rbio->flags)) { | |
768 | keep_cache = 1; | |
769 | clear_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
770 | BUG_ON(!bio_list_empty(&rbio->bio_list)); | |
771 | goto done; | |
772 | } | |
53b381b3 DW |
773 | |
774 | list_del_init(&rbio->hash_list); | |
dec95574 | 775 | refcount_dec(&rbio->refs); |
53b381b3 DW |
776 | |
777 | /* | |
778 | * we use the plug list to hold all the rbios | |
779 | * waiting for the chance to lock this stripe. | |
780 | * hand the lock over to one of them. | |
781 | */ | |
782 | if (!list_empty(&rbio->plug_list)) { | |
783 | struct btrfs_raid_bio *next; | |
784 | struct list_head *head = rbio->plug_list.next; | |
785 | ||
786 | next = list_entry(head, struct btrfs_raid_bio, | |
787 | plug_list); | |
788 | ||
789 | list_del_init(&rbio->plug_list); | |
790 | ||
791 | list_add(&next->hash_list, &h->hash_list); | |
dec95574 | 792 | refcount_inc(&next->refs); |
53b381b3 DW |
793 | spin_unlock(&rbio->bio_list_lock); |
794 | spin_unlock_irqrestore(&h->lock, flags); | |
795 | ||
1b94b556 | 796 | if (next->operation == BTRFS_RBIO_READ_REBUILD) |
e66d8d5a | 797 | start_async_work(next, read_rebuild_work); |
b4ee1782 OS |
798 | else if (next->operation == BTRFS_RBIO_REBUILD_MISSING) { |
799 | steal_rbio(rbio, next); | |
e66d8d5a | 800 | start_async_work(next, read_rebuild_work); |
b4ee1782 | 801 | } else if (next->operation == BTRFS_RBIO_WRITE) { |
4ae10b3a | 802 | steal_rbio(rbio, next); |
cf6a4a75 | 803 | start_async_work(next, rmw_work); |
5a6ac9ea MX |
804 | } else if (next->operation == BTRFS_RBIO_PARITY_SCRUB) { |
805 | steal_rbio(rbio, next); | |
a81b747d | 806 | start_async_work(next, scrub_parity_work); |
4ae10b3a | 807 | } |
53b381b3 DW |
808 | |
809 | goto done_nolock; | |
53b381b3 DW |
810 | } |
811 | } | |
4ae10b3a | 812 | done: |
53b381b3 DW |
813 | spin_unlock(&rbio->bio_list_lock); |
814 | spin_unlock_irqrestore(&h->lock, flags); | |
815 | ||
816 | done_nolock: | |
4ae10b3a CM |
817 | if (!keep_cache) |
818 | remove_rbio_from_cache(rbio); | |
53b381b3 DW |
819 | } |
820 | ||
821 | static void __free_raid_bio(struct btrfs_raid_bio *rbio) | |
822 | { | |
823 | int i; | |
824 | ||
dec95574 | 825 | if (!refcount_dec_and_test(&rbio->refs)) |
53b381b3 DW |
826 | return; |
827 | ||
4ae10b3a | 828 | WARN_ON(!list_empty(&rbio->stripe_cache)); |
53b381b3 DW |
829 | WARN_ON(!list_empty(&rbio->hash_list)); |
830 | WARN_ON(!bio_list_empty(&rbio->bio_list)); | |
831 | ||
832 | for (i = 0; i < rbio->nr_pages; i++) { | |
833 | if (rbio->stripe_pages[i]) { | |
834 | __free_page(rbio->stripe_pages[i]); | |
835 | rbio->stripe_pages[i] = NULL; | |
836 | } | |
837 | } | |
af8e2d1d | 838 | |
4c664611 | 839 | btrfs_put_bioc(rbio->bioc); |
53b381b3 DW |
840 | kfree(rbio); |
841 | } | |
842 | ||
7583d8d0 | 843 | static void rbio_endio_bio_list(struct bio *cur, blk_status_t err) |
53b381b3 | 844 | { |
7583d8d0 LB |
845 | struct bio *next; |
846 | ||
847 | while (cur) { | |
848 | next = cur->bi_next; | |
849 | cur->bi_next = NULL; | |
850 | cur->bi_status = err; | |
851 | bio_endio(cur); | |
852 | cur = next; | |
853 | } | |
53b381b3 DW |
854 | } |
855 | ||
856 | /* | |
857 | * this frees the rbio and runs through all the bios in the | |
858 | * bio_list and calls end_io on them | |
859 | */ | |
4e4cbee9 | 860 | static void rbio_orig_end_io(struct btrfs_raid_bio *rbio, blk_status_t err) |
53b381b3 DW |
861 | { |
862 | struct bio *cur = bio_list_get(&rbio->bio_list); | |
7583d8d0 | 863 | struct bio *extra; |
4245215d MX |
864 | |
865 | if (rbio->generic_bio_cnt) | |
6a258d72 | 866 | btrfs_bio_counter_sub(rbio->bioc->fs_info, rbio->generic_bio_cnt); |
4245215d | 867 | |
7583d8d0 LB |
868 | /* |
869 | * At this moment, rbio->bio_list is empty, however since rbio does not | |
870 | * always have RBIO_RMW_LOCKED_BIT set and rbio is still linked on the | |
871 | * hash list, rbio may be merged with others so that rbio->bio_list | |
872 | * becomes non-empty. | |
873 | * Once unlock_stripe() is done, rbio->bio_list will not be updated any | |
874 | * more and we can call bio_endio() on all queued bios. | |
875 | */ | |
876 | unlock_stripe(rbio); | |
877 | extra = bio_list_get(&rbio->bio_list); | |
878 | __free_raid_bio(rbio); | |
53b381b3 | 879 | |
7583d8d0 LB |
880 | rbio_endio_bio_list(cur, err); |
881 | if (extra) | |
882 | rbio_endio_bio_list(extra, err); | |
53b381b3 DW |
883 | } |
884 | ||
885 | /* | |
886 | * end io function used by finish_rmw. When we finally | |
887 | * get here, we've written a full stripe | |
888 | */ | |
4246a0b6 | 889 | static void raid_write_end_io(struct bio *bio) |
53b381b3 DW |
890 | { |
891 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
4e4cbee9 | 892 | blk_status_t err = bio->bi_status; |
a6111d11 | 893 | int max_errors; |
53b381b3 DW |
894 | |
895 | if (err) | |
896 | fail_bio_stripe(rbio, bio); | |
897 | ||
898 | bio_put(bio); | |
899 | ||
b89e1b01 | 900 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
901 | return; |
902 | ||
58efbc9f | 903 | err = BLK_STS_OK; |
53b381b3 DW |
904 | |
905 | /* OK, we have read all the stripes we need to. */ | |
a6111d11 | 906 | max_errors = (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) ? |
4c664611 | 907 | 0 : rbio->bioc->max_errors; |
a6111d11 | 908 | if (atomic_read(&rbio->error) > max_errors) |
4e4cbee9 | 909 | err = BLK_STS_IOERR; |
53b381b3 | 910 | |
4246a0b6 | 911 | rbio_orig_end_io(rbio, err); |
53b381b3 DW |
912 | } |
913 | ||
914 | /* | |
915 | * the read/modify/write code wants to use the original bio for | |
916 | * any pages it included, and then use the rbio for everything | |
917 | * else. This function decides if a given index (stripe number) | |
918 | * and page number in that stripe fall inside the original bio | |
919 | * or the rbio. | |
920 | * | |
921 | * if you set bio_list_only, you'll get a NULL back for any ranges | |
922 | * that are outside the bio_list | |
923 | * | |
924 | * This doesn't take any refs on anything, you get a bare page pointer | |
925 | * and the caller must bump refs as required. | |
926 | * | |
927 | * You must call index_rbio_pages once before you can trust | |
928 | * the answers from this function. | |
929 | */ | |
930 | static struct page *page_in_rbio(struct btrfs_raid_bio *rbio, | |
931 | int index, int pagenr, int bio_list_only) | |
932 | { | |
933 | int chunk_page; | |
934 | struct page *p = NULL; | |
935 | ||
936 | chunk_page = index * (rbio->stripe_len >> PAGE_SHIFT) + pagenr; | |
937 | ||
938 | spin_lock_irq(&rbio->bio_list_lock); | |
939 | p = rbio->bio_pages[chunk_page]; | |
940 | spin_unlock_irq(&rbio->bio_list_lock); | |
941 | ||
942 | if (p || bio_list_only) | |
943 | return p; | |
944 | ||
945 | return rbio->stripe_pages[chunk_page]; | |
946 | } | |
947 | ||
948 | /* | |
949 | * number of pages we need for the entire stripe across all the | |
950 | * drives | |
951 | */ | |
952 | static unsigned long rbio_nr_pages(unsigned long stripe_len, int nr_stripes) | |
953 | { | |
09cbfeaf | 954 | return DIV_ROUND_UP(stripe_len, PAGE_SIZE) * nr_stripes; |
53b381b3 DW |
955 | } |
956 | ||
957 | /* | |
958 | * allocation and initial setup for the btrfs_raid_bio. Not | |
959 | * this does not allocate any pages for rbio->pages. | |
960 | */ | |
2ff7e61e | 961 | static struct btrfs_raid_bio *alloc_rbio(struct btrfs_fs_info *fs_info, |
4c664611 | 962 | struct btrfs_io_context *bioc, |
2ff7e61e | 963 | u64 stripe_len) |
53b381b3 DW |
964 | { |
965 | struct btrfs_raid_bio *rbio; | |
966 | int nr_data = 0; | |
4c664611 | 967 | int real_stripes = bioc->num_stripes - bioc->num_tgtdevs; |
2c8cdd6e | 968 | int num_pages = rbio_nr_pages(stripe_len, real_stripes); |
5a6ac9ea | 969 | int stripe_npages = DIV_ROUND_UP(stripe_len, PAGE_SIZE); |
53b381b3 DW |
970 | void *p; |
971 | ||
1389053e KC |
972 | rbio = kzalloc(sizeof(*rbio) + |
973 | sizeof(*rbio->stripe_pages) * num_pages + | |
974 | sizeof(*rbio->bio_pages) * num_pages + | |
975 | sizeof(*rbio->finish_pointers) * real_stripes + | |
976 | sizeof(*rbio->dbitmap) * BITS_TO_LONGS(stripe_npages) + | |
977 | sizeof(*rbio->finish_pbitmap) * | |
978 | BITS_TO_LONGS(stripe_npages), | |
979 | GFP_NOFS); | |
af8e2d1d | 980 | if (!rbio) |
53b381b3 | 981 | return ERR_PTR(-ENOMEM); |
53b381b3 DW |
982 | |
983 | bio_list_init(&rbio->bio_list); | |
984 | INIT_LIST_HEAD(&rbio->plug_list); | |
985 | spin_lock_init(&rbio->bio_list_lock); | |
4ae10b3a | 986 | INIT_LIST_HEAD(&rbio->stripe_cache); |
53b381b3 | 987 | INIT_LIST_HEAD(&rbio->hash_list); |
4c664611 | 988 | rbio->bioc = bioc; |
53b381b3 DW |
989 | rbio->stripe_len = stripe_len; |
990 | rbio->nr_pages = num_pages; | |
2c8cdd6e | 991 | rbio->real_stripes = real_stripes; |
5a6ac9ea | 992 | rbio->stripe_npages = stripe_npages; |
53b381b3 DW |
993 | rbio->faila = -1; |
994 | rbio->failb = -1; | |
dec95574 | 995 | refcount_set(&rbio->refs, 1); |
b89e1b01 MX |
996 | atomic_set(&rbio->error, 0); |
997 | atomic_set(&rbio->stripes_pending, 0); | |
53b381b3 DW |
998 | |
999 | /* | |
1389053e | 1000 | * the stripe_pages, bio_pages, etc arrays point to the extra |
53b381b3 DW |
1001 | * memory we allocated past the end of the rbio |
1002 | */ | |
1003 | p = rbio + 1; | |
1389053e KC |
1004 | #define CONSUME_ALLOC(ptr, count) do { \ |
1005 | ptr = p; \ | |
1006 | p = (unsigned char *)p + sizeof(*(ptr)) * (count); \ | |
1007 | } while (0) | |
1008 | CONSUME_ALLOC(rbio->stripe_pages, num_pages); | |
1009 | CONSUME_ALLOC(rbio->bio_pages, num_pages); | |
1010 | CONSUME_ALLOC(rbio->finish_pointers, real_stripes); | |
1011 | CONSUME_ALLOC(rbio->dbitmap, BITS_TO_LONGS(stripe_npages)); | |
1012 | CONSUME_ALLOC(rbio->finish_pbitmap, BITS_TO_LONGS(stripe_npages)); | |
1013 | #undef CONSUME_ALLOC | |
53b381b3 | 1014 | |
4c664611 | 1015 | if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID5) |
10f11900 | 1016 | nr_data = real_stripes - 1; |
4c664611 | 1017 | else if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID6) |
2c8cdd6e | 1018 | nr_data = real_stripes - 2; |
53b381b3 | 1019 | else |
10f11900 | 1020 | BUG(); |
53b381b3 DW |
1021 | |
1022 | rbio->nr_data = nr_data; | |
1023 | return rbio; | |
1024 | } | |
1025 | ||
1026 | /* allocate pages for all the stripes in the bio, including parity */ | |
1027 | static int alloc_rbio_pages(struct btrfs_raid_bio *rbio) | |
1028 | { | |
dd137dd1 | 1029 | return btrfs_alloc_page_array(rbio->nr_pages, rbio->stripe_pages); |
53b381b3 DW |
1030 | } |
1031 | ||
b7178a5f | 1032 | /* only allocate pages for p/q stripes */ |
53b381b3 DW |
1033 | static int alloc_rbio_parity_pages(struct btrfs_raid_bio *rbio) |
1034 | { | |
dd137dd1 | 1035 | int data_pages = rbio_stripe_page_index(rbio, rbio->nr_data, 0); |
53b381b3 | 1036 | |
dd137dd1 STD |
1037 | return btrfs_alloc_page_array(rbio->nr_pages - data_pages, |
1038 | rbio->stripe_pages + data_pages); | |
53b381b3 DW |
1039 | } |
1040 | ||
1041 | /* | |
1042 | * add a single page from a specific stripe into our list of bios for IO | |
1043 | * this will try to merge into existing bios if possible, and returns | |
1044 | * zero if all went well. | |
1045 | */ | |
48a3b636 ES |
1046 | static int rbio_add_io_page(struct btrfs_raid_bio *rbio, |
1047 | struct bio_list *bio_list, | |
1048 | struct page *page, | |
1049 | int stripe_nr, | |
1050 | unsigned long page_index, | |
1051 | unsigned long bio_max_len) | |
53b381b3 DW |
1052 | { |
1053 | struct bio *last = bio_list->tail; | |
53b381b3 DW |
1054 | int ret; |
1055 | struct bio *bio; | |
4c664611 | 1056 | struct btrfs_io_stripe *stripe; |
53b381b3 DW |
1057 | u64 disk_start; |
1058 | ||
4c664611 | 1059 | stripe = &rbio->bioc->stripes[stripe_nr]; |
09cbfeaf | 1060 | disk_start = stripe->physical + (page_index << PAGE_SHIFT); |
53b381b3 DW |
1061 | |
1062 | /* if the device is missing, just fail this stripe */ | |
1063 | if (!stripe->dev->bdev) | |
1064 | return fail_rbio_index(rbio, stripe_nr); | |
1065 | ||
1066 | /* see if we can add this page onto our existing bio */ | |
1067 | if (last) { | |
1201b58b | 1068 | u64 last_end = last->bi_iter.bi_sector << 9; |
4f024f37 | 1069 | last_end += last->bi_iter.bi_size; |
53b381b3 DW |
1070 | |
1071 | /* | |
1072 | * we can't merge these if they are from different | |
1073 | * devices or if they are not contiguous | |
1074 | */ | |
f90ae76a | 1075 | if (last_end == disk_start && !last->bi_status && |
309dca30 | 1076 | last->bi_bdev == stripe->dev->bdev) { |
09cbfeaf KS |
1077 | ret = bio_add_page(last, page, PAGE_SIZE, 0); |
1078 | if (ret == PAGE_SIZE) | |
53b381b3 DW |
1079 | return 0; |
1080 | } | |
1081 | } | |
1082 | ||
1083 | /* put a new bio on the list */ | |
c3a3b19b QW |
1084 | bio = btrfs_bio_alloc(bio_max_len >> PAGE_SHIFT ?: 1); |
1085 | btrfs_bio(bio)->device = stripe->dev; | |
4f024f37 | 1086 | bio->bi_iter.bi_size = 0; |
74d46992 | 1087 | bio_set_dev(bio, stripe->dev->bdev); |
4f024f37 | 1088 | bio->bi_iter.bi_sector = disk_start >> 9; |
53b381b3 | 1089 | |
09cbfeaf | 1090 | bio_add_page(bio, page, PAGE_SIZE, 0); |
53b381b3 DW |
1091 | bio_list_add(bio_list, bio); |
1092 | return 0; | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * while we're doing the read/modify/write cycle, we could | |
1097 | * have errors in reading pages off the disk. This checks | |
1098 | * for errors and if we're not able to read the page it'll | |
1099 | * trigger parity reconstruction. The rmw will be finished | |
1100 | * after we've reconstructed the failed stripes | |
1101 | */ | |
1102 | static void validate_rbio_for_rmw(struct btrfs_raid_bio *rbio) | |
1103 | { | |
1104 | if (rbio->faila >= 0 || rbio->failb >= 0) { | |
2c8cdd6e | 1105 | BUG_ON(rbio->faila == rbio->real_stripes - 1); |
53b381b3 DW |
1106 | __raid56_parity_recover(rbio); |
1107 | } else { | |
1108 | finish_rmw(rbio); | |
1109 | } | |
1110 | } | |
1111 | ||
53b381b3 DW |
1112 | /* |
1113 | * helper function to walk our bio list and populate the bio_pages array with | |
1114 | * the result. This seems expensive, but it is faster than constantly | |
1115 | * searching through the bio list as we setup the IO in finish_rmw or stripe | |
1116 | * reconstruction. | |
1117 | * | |
1118 | * This must be called before you trust the answers from page_in_rbio | |
1119 | */ | |
1120 | static void index_rbio_pages(struct btrfs_raid_bio *rbio) | |
1121 | { | |
1122 | struct bio *bio; | |
1123 | u64 start; | |
1124 | unsigned long stripe_offset; | |
1125 | unsigned long page_index; | |
53b381b3 DW |
1126 | |
1127 | spin_lock_irq(&rbio->bio_list_lock); | |
1128 | bio_list_for_each(bio, &rbio->bio_list) { | |
6592e58c FM |
1129 | struct bio_vec bvec; |
1130 | struct bvec_iter iter; | |
1131 | int i = 0; | |
1132 | ||
1201b58b | 1133 | start = bio->bi_iter.bi_sector << 9; |
4c664611 | 1134 | stripe_offset = start - rbio->bioc->raid_map[0]; |
09cbfeaf | 1135 | page_index = stripe_offset >> PAGE_SHIFT; |
53b381b3 | 1136 | |
6592e58c | 1137 | if (bio_flagged(bio, BIO_CLONED)) |
c3a3b19b | 1138 | bio->bi_iter = btrfs_bio(bio)->iter; |
6592e58c FM |
1139 | |
1140 | bio_for_each_segment(bvec, bio, iter) { | |
1141 | rbio->bio_pages[page_index + i] = bvec.bv_page; | |
1142 | i++; | |
1143 | } | |
53b381b3 DW |
1144 | } |
1145 | spin_unlock_irq(&rbio->bio_list_lock); | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * this is called from one of two situations. We either | |
1150 | * have a full stripe from the higher layers, or we've read all | |
1151 | * the missing bits off disk. | |
1152 | * | |
1153 | * This will calculate the parity and then send down any | |
1154 | * changed blocks. | |
1155 | */ | |
1156 | static noinline void finish_rmw(struct btrfs_raid_bio *rbio) | |
1157 | { | |
4c664611 | 1158 | struct btrfs_io_context *bioc = rbio->bioc; |
1389053e | 1159 | void **pointers = rbio->finish_pointers; |
53b381b3 DW |
1160 | int nr_data = rbio->nr_data; |
1161 | int stripe; | |
1162 | int pagenr; | |
c17af965 | 1163 | bool has_qstripe; |
53b381b3 DW |
1164 | struct bio_list bio_list; |
1165 | struct bio *bio; | |
53b381b3 DW |
1166 | int ret; |
1167 | ||
1168 | bio_list_init(&bio_list); | |
1169 | ||
c17af965 DS |
1170 | if (rbio->real_stripes - rbio->nr_data == 1) |
1171 | has_qstripe = false; | |
1172 | else if (rbio->real_stripes - rbio->nr_data == 2) | |
1173 | has_qstripe = true; | |
1174 | else | |
53b381b3 | 1175 | BUG(); |
53b381b3 DW |
1176 | |
1177 | /* at this point we either have a full stripe, | |
1178 | * or we've read the full stripe from the drive. | |
1179 | * recalculate the parity and write the new results. | |
1180 | * | |
1181 | * We're not allowed to add any new bios to the | |
1182 | * bio list here, anyone else that wants to | |
1183 | * change this stripe needs to do their own rmw. | |
1184 | */ | |
1185 | spin_lock_irq(&rbio->bio_list_lock); | |
1186 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
1187 | spin_unlock_irq(&rbio->bio_list_lock); | |
1188 | ||
b89e1b01 | 1189 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
1190 | |
1191 | /* | |
1192 | * now that we've set rmw_locked, run through the | |
1193 | * bio list one last time and map the page pointers | |
4ae10b3a CM |
1194 | * |
1195 | * We don't cache full rbios because we're assuming | |
1196 | * the higher layers are unlikely to use this area of | |
1197 | * the disk again soon. If they do use it again, | |
1198 | * hopefully they will send another full bio. | |
53b381b3 DW |
1199 | */ |
1200 | index_rbio_pages(rbio); | |
4ae10b3a CM |
1201 | if (!rbio_is_full(rbio)) |
1202 | cache_rbio_pages(rbio); | |
1203 | else | |
1204 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
53b381b3 | 1205 | |
915e2290 | 1206 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
1207 | struct page *p; |
1208 | /* first collect one page from each data stripe */ | |
1209 | for (stripe = 0; stripe < nr_data; stripe++) { | |
1210 | p = page_in_rbio(rbio, stripe, pagenr, 0); | |
94a0b58d | 1211 | pointers[stripe] = kmap_local_page(p); |
53b381b3 DW |
1212 | } |
1213 | ||
1214 | /* then add the parity stripe */ | |
1215 | p = rbio_pstripe_page(rbio, pagenr); | |
1216 | SetPageUptodate(p); | |
94a0b58d | 1217 | pointers[stripe++] = kmap_local_page(p); |
53b381b3 | 1218 | |
c17af965 | 1219 | if (has_qstripe) { |
53b381b3 DW |
1220 | |
1221 | /* | |
1222 | * raid6, add the qstripe and call the | |
1223 | * library function to fill in our p/q | |
1224 | */ | |
1225 | p = rbio_qstripe_page(rbio, pagenr); | |
1226 | SetPageUptodate(p); | |
94a0b58d | 1227 | pointers[stripe++] = kmap_local_page(p); |
53b381b3 | 1228 | |
2c8cdd6e | 1229 | raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE, |
53b381b3 DW |
1230 | pointers); |
1231 | } else { | |
1232 | /* raid5 */ | |
69d24804 | 1233 | copy_page(pointers[nr_data], pointers[0]); |
09cbfeaf | 1234 | run_xor(pointers + 1, nr_data - 1, PAGE_SIZE); |
53b381b3 | 1235 | } |
94a0b58d IW |
1236 | for (stripe = stripe - 1; stripe >= 0; stripe--) |
1237 | kunmap_local(pointers[stripe]); | |
53b381b3 DW |
1238 | } |
1239 | ||
1240 | /* | |
1241 | * time to start writing. Make bios for everything from the | |
1242 | * higher layers (the bio_list in our rbio) and our p/q. Ignore | |
1243 | * everything else. | |
1244 | */ | |
2c8cdd6e | 1245 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
915e2290 | 1246 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
1247 | struct page *page; |
1248 | if (stripe < rbio->nr_data) { | |
1249 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
1250 | if (!page) | |
1251 | continue; | |
1252 | } else { | |
1253 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
1254 | } | |
1255 | ||
1256 | ret = rbio_add_io_page(rbio, &bio_list, | |
1257 | page, stripe, pagenr, rbio->stripe_len); | |
1258 | if (ret) | |
1259 | goto cleanup; | |
1260 | } | |
1261 | } | |
1262 | ||
4c664611 | 1263 | if (likely(!bioc->num_tgtdevs)) |
2c8cdd6e MX |
1264 | goto write_data; |
1265 | ||
1266 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { | |
4c664611 | 1267 | if (!bioc->tgtdev_map[stripe]) |
2c8cdd6e MX |
1268 | continue; |
1269 | ||
915e2290 | 1270 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
2c8cdd6e MX |
1271 | struct page *page; |
1272 | if (stripe < rbio->nr_data) { | |
1273 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
1274 | if (!page) | |
1275 | continue; | |
1276 | } else { | |
1277 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
1278 | } | |
1279 | ||
1280 | ret = rbio_add_io_page(rbio, &bio_list, page, | |
4c664611 | 1281 | rbio->bioc->tgtdev_map[stripe], |
2c8cdd6e MX |
1282 | pagenr, rbio->stripe_len); |
1283 | if (ret) | |
1284 | goto cleanup; | |
1285 | } | |
1286 | } | |
1287 | ||
1288 | write_data: | |
b89e1b01 MX |
1289 | atomic_set(&rbio->stripes_pending, bio_list_size(&bio_list)); |
1290 | BUG_ON(atomic_read(&rbio->stripes_pending) == 0); | |
53b381b3 | 1291 | |
bf28a605 | 1292 | while ((bio = bio_list_pop(&bio_list))) { |
53b381b3 DW |
1293 | bio->bi_private = rbio; |
1294 | bio->bi_end_io = raid_write_end_io; | |
ebcc3263 | 1295 | bio->bi_opf = REQ_OP_WRITE; |
4e49ea4a MC |
1296 | |
1297 | submit_bio(bio); | |
53b381b3 DW |
1298 | } |
1299 | return; | |
1300 | ||
1301 | cleanup: | |
58efbc9f | 1302 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
1303 | |
1304 | while ((bio = bio_list_pop(&bio_list))) | |
1305 | bio_put(bio); | |
53b381b3 DW |
1306 | } |
1307 | ||
1308 | /* | |
1309 | * helper to find the stripe number for a given bio. Used to figure out which | |
1310 | * stripe has failed. This expects the bio to correspond to a physical disk, | |
1311 | * so it looks up based on physical sector numbers. | |
1312 | */ | |
1313 | static int find_bio_stripe(struct btrfs_raid_bio *rbio, | |
1314 | struct bio *bio) | |
1315 | { | |
4f024f37 | 1316 | u64 physical = bio->bi_iter.bi_sector; |
53b381b3 | 1317 | int i; |
4c664611 | 1318 | struct btrfs_io_stripe *stripe; |
53b381b3 DW |
1319 | |
1320 | physical <<= 9; | |
1321 | ||
4c664611 QW |
1322 | for (i = 0; i < rbio->bioc->num_stripes; i++) { |
1323 | stripe = &rbio->bioc->stripes[i]; | |
83025863 | 1324 | if (in_range(physical, stripe->physical, rbio->stripe_len) && |
309dca30 | 1325 | stripe->dev->bdev && bio->bi_bdev == stripe->dev->bdev) { |
53b381b3 DW |
1326 | return i; |
1327 | } | |
1328 | } | |
1329 | return -1; | |
1330 | } | |
1331 | ||
1332 | /* | |
1333 | * helper to find the stripe number for a given | |
1334 | * bio (before mapping). Used to figure out which stripe has | |
1335 | * failed. This looks up based on logical block numbers. | |
1336 | */ | |
1337 | static int find_logical_bio_stripe(struct btrfs_raid_bio *rbio, | |
1338 | struct bio *bio) | |
1339 | { | |
1201b58b | 1340 | u64 logical = bio->bi_iter.bi_sector << 9; |
53b381b3 DW |
1341 | int i; |
1342 | ||
53b381b3 | 1343 | for (i = 0; i < rbio->nr_data; i++) { |
4c664611 | 1344 | u64 stripe_start = rbio->bioc->raid_map[i]; |
83025863 NB |
1345 | |
1346 | if (in_range(logical, stripe_start, rbio->stripe_len)) | |
53b381b3 | 1347 | return i; |
53b381b3 DW |
1348 | } |
1349 | return -1; | |
1350 | } | |
1351 | ||
1352 | /* | |
1353 | * returns -EIO if we had too many failures | |
1354 | */ | |
1355 | static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed) | |
1356 | { | |
1357 | unsigned long flags; | |
1358 | int ret = 0; | |
1359 | ||
1360 | spin_lock_irqsave(&rbio->bio_list_lock, flags); | |
1361 | ||
1362 | /* we already know this stripe is bad, move on */ | |
1363 | if (rbio->faila == failed || rbio->failb == failed) | |
1364 | goto out; | |
1365 | ||
1366 | if (rbio->faila == -1) { | |
1367 | /* first failure on this rbio */ | |
1368 | rbio->faila = failed; | |
b89e1b01 | 1369 | atomic_inc(&rbio->error); |
53b381b3 DW |
1370 | } else if (rbio->failb == -1) { |
1371 | /* second failure on this rbio */ | |
1372 | rbio->failb = failed; | |
b89e1b01 | 1373 | atomic_inc(&rbio->error); |
53b381b3 DW |
1374 | } else { |
1375 | ret = -EIO; | |
1376 | } | |
1377 | out: | |
1378 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); | |
1379 | ||
1380 | return ret; | |
1381 | } | |
1382 | ||
1383 | /* | |
1384 | * helper to fail a stripe based on a physical disk | |
1385 | * bio. | |
1386 | */ | |
1387 | static int fail_bio_stripe(struct btrfs_raid_bio *rbio, | |
1388 | struct bio *bio) | |
1389 | { | |
1390 | int failed = find_bio_stripe(rbio, bio); | |
1391 | ||
1392 | if (failed < 0) | |
1393 | return -EIO; | |
1394 | ||
1395 | return fail_rbio_index(rbio, failed); | |
1396 | } | |
1397 | ||
1398 | /* | |
1399 | * this sets each page in the bio uptodate. It should only be used on private | |
1400 | * rbio pages, nothing that comes in from the higher layers | |
1401 | */ | |
1402 | static void set_bio_pages_uptodate(struct bio *bio) | |
1403 | { | |
0198e5b7 | 1404 | struct bio_vec *bvec; |
6dc4f100 | 1405 | struct bvec_iter_all iter_all; |
6592e58c | 1406 | |
0198e5b7 | 1407 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
53b381b3 | 1408 | |
2b070cfe | 1409 | bio_for_each_segment_all(bvec, bio, iter_all) |
0198e5b7 | 1410 | SetPageUptodate(bvec->bv_page); |
53b381b3 DW |
1411 | } |
1412 | ||
1413 | /* | |
1414 | * end io for the read phase of the rmw cycle. All the bios here are physical | |
1415 | * stripe bios we've read from the disk so we can recalculate the parity of the | |
1416 | * stripe. | |
1417 | * | |
1418 | * This will usually kick off finish_rmw once all the bios are read in, but it | |
1419 | * may trigger parity reconstruction if we had any errors along the way | |
1420 | */ | |
4246a0b6 | 1421 | static void raid_rmw_end_io(struct bio *bio) |
53b381b3 DW |
1422 | { |
1423 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
1424 | ||
4e4cbee9 | 1425 | if (bio->bi_status) |
53b381b3 DW |
1426 | fail_bio_stripe(rbio, bio); |
1427 | else | |
1428 | set_bio_pages_uptodate(bio); | |
1429 | ||
1430 | bio_put(bio); | |
1431 | ||
b89e1b01 | 1432 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
1433 | return; |
1434 | ||
4c664611 | 1435 | if (atomic_read(&rbio->error) > rbio->bioc->max_errors) |
53b381b3 DW |
1436 | goto cleanup; |
1437 | ||
1438 | /* | |
1439 | * this will normally call finish_rmw to start our write | |
1440 | * but if there are any failed stripes we'll reconstruct | |
1441 | * from parity first | |
1442 | */ | |
1443 | validate_rbio_for_rmw(rbio); | |
1444 | return; | |
1445 | ||
1446 | cleanup: | |
1447 | ||
58efbc9f | 1448 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
53b381b3 DW |
1449 | } |
1450 | ||
53b381b3 DW |
1451 | /* |
1452 | * the stripe must be locked by the caller. It will | |
1453 | * unlock after all the writes are done | |
1454 | */ | |
1455 | static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio) | |
1456 | { | |
1457 | int bios_to_read = 0; | |
53b381b3 DW |
1458 | struct bio_list bio_list; |
1459 | int ret; | |
53b381b3 DW |
1460 | int pagenr; |
1461 | int stripe; | |
1462 | struct bio *bio; | |
1463 | ||
1464 | bio_list_init(&bio_list); | |
1465 | ||
1466 | ret = alloc_rbio_pages(rbio); | |
1467 | if (ret) | |
1468 | goto cleanup; | |
1469 | ||
1470 | index_rbio_pages(rbio); | |
1471 | ||
b89e1b01 | 1472 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
1473 | /* |
1474 | * build a list of bios to read all the missing parts of this | |
1475 | * stripe | |
1476 | */ | |
1477 | for (stripe = 0; stripe < rbio->nr_data; stripe++) { | |
915e2290 | 1478 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
1479 | struct page *page; |
1480 | /* | |
1481 | * we want to find all the pages missing from | |
1482 | * the rbio and read them from the disk. If | |
1483 | * page_in_rbio finds a page in the bio list | |
1484 | * we don't need to read it off the stripe. | |
1485 | */ | |
1486 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
1487 | if (page) | |
1488 | continue; | |
1489 | ||
1490 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
4ae10b3a CM |
1491 | /* |
1492 | * the bio cache may have handed us an uptodate | |
1493 | * page. If so, be happy and use it | |
1494 | */ | |
1495 | if (PageUptodate(page)) | |
1496 | continue; | |
1497 | ||
53b381b3 DW |
1498 | ret = rbio_add_io_page(rbio, &bio_list, page, |
1499 | stripe, pagenr, rbio->stripe_len); | |
1500 | if (ret) | |
1501 | goto cleanup; | |
1502 | } | |
1503 | } | |
1504 | ||
1505 | bios_to_read = bio_list_size(&bio_list); | |
1506 | if (!bios_to_read) { | |
1507 | /* | |
1508 | * this can happen if others have merged with | |
1509 | * us, it means there is nothing left to read. | |
1510 | * But if there are missing devices it may not be | |
1511 | * safe to do the full stripe write yet. | |
1512 | */ | |
1513 | goto finish; | |
1514 | } | |
1515 | ||
1516 | /* | |
4c664611 QW |
1517 | * The bioc may be freed once we submit the last bio. Make sure not to |
1518 | * touch it after that. | |
53b381b3 | 1519 | */ |
b89e1b01 | 1520 | atomic_set(&rbio->stripes_pending, bios_to_read); |
bf28a605 | 1521 | while ((bio = bio_list_pop(&bio_list))) { |
53b381b3 DW |
1522 | bio->bi_private = rbio; |
1523 | bio->bi_end_io = raid_rmw_end_io; | |
ebcc3263 | 1524 | bio->bi_opf = REQ_OP_READ; |
53b381b3 | 1525 | |
6a258d72 | 1526 | btrfs_bio_wq_end_io(rbio->bioc->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
53b381b3 | 1527 | |
4e49ea4a | 1528 | submit_bio(bio); |
53b381b3 DW |
1529 | } |
1530 | /* the actual write will happen once the reads are done */ | |
1531 | return 0; | |
1532 | ||
1533 | cleanup: | |
58efbc9f | 1534 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
1535 | |
1536 | while ((bio = bio_list_pop(&bio_list))) | |
1537 | bio_put(bio); | |
1538 | ||
53b381b3 DW |
1539 | return -EIO; |
1540 | ||
1541 | finish: | |
1542 | validate_rbio_for_rmw(rbio); | |
1543 | return 0; | |
1544 | } | |
1545 | ||
1546 | /* | |
1547 | * if the upper layers pass in a full stripe, we thank them by only allocating | |
1548 | * enough pages to hold the parity, and sending it all down quickly. | |
1549 | */ | |
1550 | static int full_stripe_write(struct btrfs_raid_bio *rbio) | |
1551 | { | |
1552 | int ret; | |
1553 | ||
1554 | ret = alloc_rbio_parity_pages(rbio); | |
3cd846d1 MX |
1555 | if (ret) { |
1556 | __free_raid_bio(rbio); | |
53b381b3 | 1557 | return ret; |
3cd846d1 | 1558 | } |
53b381b3 DW |
1559 | |
1560 | ret = lock_stripe_add(rbio); | |
1561 | if (ret == 0) | |
1562 | finish_rmw(rbio); | |
1563 | return 0; | |
1564 | } | |
1565 | ||
1566 | /* | |
1567 | * partial stripe writes get handed over to async helpers. | |
1568 | * We're really hoping to merge a few more writes into this | |
1569 | * rbio before calculating new parity | |
1570 | */ | |
1571 | static int partial_stripe_write(struct btrfs_raid_bio *rbio) | |
1572 | { | |
1573 | int ret; | |
1574 | ||
1575 | ret = lock_stripe_add(rbio); | |
1576 | if (ret == 0) | |
cf6a4a75 | 1577 | start_async_work(rbio, rmw_work); |
53b381b3 DW |
1578 | return 0; |
1579 | } | |
1580 | ||
1581 | /* | |
1582 | * sometimes while we were reading from the drive to | |
1583 | * recalculate parity, enough new bios come into create | |
1584 | * a full stripe. So we do a check here to see if we can | |
1585 | * go directly to finish_rmw | |
1586 | */ | |
1587 | static int __raid56_parity_write(struct btrfs_raid_bio *rbio) | |
1588 | { | |
1589 | /* head off into rmw land if we don't have a full stripe */ | |
1590 | if (!rbio_is_full(rbio)) | |
1591 | return partial_stripe_write(rbio); | |
1592 | return full_stripe_write(rbio); | |
1593 | } | |
1594 | ||
6ac0f488 CM |
1595 | /* |
1596 | * We use plugging call backs to collect full stripes. | |
1597 | * Any time we get a partial stripe write while plugged | |
1598 | * we collect it into a list. When the unplug comes down, | |
1599 | * we sort the list by logical block number and merge | |
1600 | * everything we can into the same rbios | |
1601 | */ | |
1602 | struct btrfs_plug_cb { | |
1603 | struct blk_plug_cb cb; | |
1604 | struct btrfs_fs_info *info; | |
1605 | struct list_head rbio_list; | |
1606 | struct btrfs_work work; | |
1607 | }; | |
1608 | ||
1609 | /* | |
1610 | * rbios on the plug list are sorted for easier merging. | |
1611 | */ | |
4f0f586b ST |
1612 | static int plug_cmp(void *priv, const struct list_head *a, |
1613 | const struct list_head *b) | |
6ac0f488 | 1614 | { |
214cc184 DS |
1615 | const struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio, |
1616 | plug_list); | |
1617 | const struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio, | |
1618 | plug_list); | |
4f024f37 KO |
1619 | u64 a_sector = ra->bio_list.head->bi_iter.bi_sector; |
1620 | u64 b_sector = rb->bio_list.head->bi_iter.bi_sector; | |
6ac0f488 CM |
1621 | |
1622 | if (a_sector < b_sector) | |
1623 | return -1; | |
1624 | if (a_sector > b_sector) | |
1625 | return 1; | |
1626 | return 0; | |
1627 | } | |
1628 | ||
1629 | static void run_plug(struct btrfs_plug_cb *plug) | |
1630 | { | |
1631 | struct btrfs_raid_bio *cur; | |
1632 | struct btrfs_raid_bio *last = NULL; | |
1633 | ||
1634 | /* | |
1635 | * sort our plug list then try to merge | |
1636 | * everything we can in hopes of creating full | |
1637 | * stripes. | |
1638 | */ | |
1639 | list_sort(NULL, &plug->rbio_list, plug_cmp); | |
1640 | while (!list_empty(&plug->rbio_list)) { | |
1641 | cur = list_entry(plug->rbio_list.next, | |
1642 | struct btrfs_raid_bio, plug_list); | |
1643 | list_del_init(&cur->plug_list); | |
1644 | ||
1645 | if (rbio_is_full(cur)) { | |
c7b562c5 DS |
1646 | int ret; |
1647 | ||
6ac0f488 | 1648 | /* we have a full stripe, send it down */ |
c7b562c5 DS |
1649 | ret = full_stripe_write(cur); |
1650 | BUG_ON(ret); | |
6ac0f488 CM |
1651 | continue; |
1652 | } | |
1653 | if (last) { | |
1654 | if (rbio_can_merge(last, cur)) { | |
1655 | merge_rbio(last, cur); | |
1656 | __free_raid_bio(cur); | |
1657 | continue; | |
1658 | ||
1659 | } | |
1660 | __raid56_parity_write(last); | |
1661 | } | |
1662 | last = cur; | |
1663 | } | |
1664 | if (last) { | |
1665 | __raid56_parity_write(last); | |
1666 | } | |
1667 | kfree(plug); | |
1668 | } | |
1669 | ||
1670 | /* | |
1671 | * if the unplug comes from schedule, we have to push the | |
1672 | * work off to a helper thread | |
1673 | */ | |
1674 | static void unplug_work(struct btrfs_work *work) | |
1675 | { | |
1676 | struct btrfs_plug_cb *plug; | |
1677 | plug = container_of(work, struct btrfs_plug_cb, work); | |
1678 | run_plug(plug); | |
1679 | } | |
1680 | ||
1681 | static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1682 | { | |
1683 | struct btrfs_plug_cb *plug; | |
1684 | plug = container_of(cb, struct btrfs_plug_cb, cb); | |
1685 | ||
1686 | if (from_schedule) { | |
a0cac0ec | 1687 | btrfs_init_work(&plug->work, unplug_work, NULL, NULL); |
d05a33ac QW |
1688 | btrfs_queue_work(plug->info->rmw_workers, |
1689 | &plug->work); | |
6ac0f488 CM |
1690 | return; |
1691 | } | |
1692 | run_plug(plug); | |
1693 | } | |
1694 | ||
53b381b3 DW |
1695 | /* |
1696 | * our main entry point for writes from the rest of the FS. | |
1697 | */ | |
6a258d72 QW |
1698 | int raid56_parity_write(struct bio *bio, struct btrfs_io_context *bioc, |
1699 | u64 stripe_len) | |
53b381b3 | 1700 | { |
6a258d72 | 1701 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
53b381b3 | 1702 | struct btrfs_raid_bio *rbio; |
6ac0f488 CM |
1703 | struct btrfs_plug_cb *plug = NULL; |
1704 | struct blk_plug_cb *cb; | |
4245215d | 1705 | int ret; |
53b381b3 | 1706 | |
4c664611 | 1707 | rbio = alloc_rbio(fs_info, bioc, stripe_len); |
af8e2d1d | 1708 | if (IS_ERR(rbio)) { |
4c664611 | 1709 | btrfs_put_bioc(bioc); |
53b381b3 | 1710 | return PTR_ERR(rbio); |
af8e2d1d | 1711 | } |
53b381b3 | 1712 | bio_list_add(&rbio->bio_list, bio); |
4f024f37 | 1713 | rbio->bio_list_bytes = bio->bi_iter.bi_size; |
1b94b556 | 1714 | rbio->operation = BTRFS_RBIO_WRITE; |
6ac0f488 | 1715 | |
0b246afa | 1716 | btrfs_bio_counter_inc_noblocked(fs_info); |
4245215d MX |
1717 | rbio->generic_bio_cnt = 1; |
1718 | ||
6ac0f488 CM |
1719 | /* |
1720 | * don't plug on full rbios, just get them out the door | |
1721 | * as quickly as we can | |
1722 | */ | |
4245215d MX |
1723 | if (rbio_is_full(rbio)) { |
1724 | ret = full_stripe_write(rbio); | |
1725 | if (ret) | |
0b246afa | 1726 | btrfs_bio_counter_dec(fs_info); |
4245215d MX |
1727 | return ret; |
1728 | } | |
6ac0f488 | 1729 | |
0b246afa | 1730 | cb = blk_check_plugged(btrfs_raid_unplug, fs_info, sizeof(*plug)); |
6ac0f488 CM |
1731 | if (cb) { |
1732 | plug = container_of(cb, struct btrfs_plug_cb, cb); | |
1733 | if (!plug->info) { | |
0b246afa | 1734 | plug->info = fs_info; |
6ac0f488 CM |
1735 | INIT_LIST_HEAD(&plug->rbio_list); |
1736 | } | |
1737 | list_add_tail(&rbio->plug_list, &plug->rbio_list); | |
4245215d | 1738 | ret = 0; |
6ac0f488 | 1739 | } else { |
4245215d MX |
1740 | ret = __raid56_parity_write(rbio); |
1741 | if (ret) | |
0b246afa | 1742 | btrfs_bio_counter_dec(fs_info); |
6ac0f488 | 1743 | } |
4245215d | 1744 | return ret; |
53b381b3 DW |
1745 | } |
1746 | ||
1747 | /* | |
1748 | * all parity reconstruction happens here. We've read in everything | |
1749 | * we can find from the drives and this does the heavy lifting of | |
1750 | * sorting the good from the bad. | |
1751 | */ | |
1752 | static void __raid_recover_end_io(struct btrfs_raid_bio *rbio) | |
1753 | { | |
1754 | int pagenr, stripe; | |
1755 | void **pointers; | |
94a0b58d | 1756 | void **unmap_array; |
53b381b3 | 1757 | int faila = -1, failb = -1; |
53b381b3 | 1758 | struct page *page; |
58efbc9f | 1759 | blk_status_t err; |
53b381b3 DW |
1760 | int i; |
1761 | ||
31e818fe | 1762 | pointers = kcalloc(rbio->real_stripes, sizeof(void *), GFP_NOFS); |
53b381b3 | 1763 | if (!pointers) { |
58efbc9f | 1764 | err = BLK_STS_RESOURCE; |
53b381b3 DW |
1765 | goto cleanup_io; |
1766 | } | |
1767 | ||
94a0b58d IW |
1768 | /* |
1769 | * Store copy of pointers that does not get reordered during | |
1770 | * reconstruction so that kunmap_local works. | |
1771 | */ | |
1772 | unmap_array = kcalloc(rbio->real_stripes, sizeof(void *), GFP_NOFS); | |
1773 | if (!unmap_array) { | |
1774 | err = BLK_STS_RESOURCE; | |
1775 | goto cleanup_pointers; | |
1776 | } | |
1777 | ||
53b381b3 DW |
1778 | faila = rbio->faila; |
1779 | failb = rbio->failb; | |
1780 | ||
b4ee1782 OS |
1781 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1782 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { | |
53b381b3 DW |
1783 | spin_lock_irq(&rbio->bio_list_lock); |
1784 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
1785 | spin_unlock_irq(&rbio->bio_list_lock); | |
1786 | } | |
1787 | ||
1788 | index_rbio_pages(rbio); | |
1789 | ||
915e2290 | 1790 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
5a6ac9ea MX |
1791 | /* |
1792 | * Now we just use bitmap to mark the horizontal stripes in | |
1793 | * which we have data when doing parity scrub. | |
1794 | */ | |
1795 | if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB && | |
1796 | !test_bit(pagenr, rbio->dbitmap)) | |
1797 | continue; | |
1798 | ||
94a0b58d IW |
1799 | /* |
1800 | * Setup our array of pointers with pages from each stripe | |
1801 | * | |
1802 | * NOTE: store a duplicate array of pointers to preserve the | |
1803 | * pointer order | |
53b381b3 | 1804 | */ |
2c8cdd6e | 1805 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
53b381b3 DW |
1806 | /* |
1807 | * if we're rebuilding a read, we have to use | |
1808 | * pages from the bio list | |
1809 | */ | |
b4ee1782 OS |
1810 | if ((rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1811 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) && | |
53b381b3 DW |
1812 | (stripe == faila || stripe == failb)) { |
1813 | page = page_in_rbio(rbio, stripe, pagenr, 0); | |
1814 | } else { | |
1815 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
1816 | } | |
94a0b58d IW |
1817 | pointers[stripe] = kmap_local_page(page); |
1818 | unmap_array[stripe] = pointers[stripe]; | |
53b381b3 DW |
1819 | } |
1820 | ||
1821 | /* all raid6 handling here */ | |
4c664611 | 1822 | if (rbio->bioc->map_type & BTRFS_BLOCK_GROUP_RAID6) { |
53b381b3 DW |
1823 | /* |
1824 | * single failure, rebuild from parity raid5 | |
1825 | * style | |
1826 | */ | |
1827 | if (failb < 0) { | |
1828 | if (faila == rbio->nr_data) { | |
1829 | /* | |
1830 | * Just the P stripe has failed, without | |
1831 | * a bad data or Q stripe. | |
1832 | * TODO, we should redo the xor here. | |
1833 | */ | |
58efbc9f | 1834 | err = BLK_STS_IOERR; |
53b381b3 DW |
1835 | goto cleanup; |
1836 | } | |
1837 | /* | |
1838 | * a single failure in raid6 is rebuilt | |
1839 | * in the pstripe code below | |
1840 | */ | |
1841 | goto pstripe; | |
1842 | } | |
1843 | ||
1844 | /* make sure our ps and qs are in order */ | |
b7d2083a NB |
1845 | if (faila > failb) |
1846 | swap(faila, failb); | |
53b381b3 DW |
1847 | |
1848 | /* if the q stripe is failed, do a pstripe reconstruction | |
1849 | * from the xors. | |
1850 | * If both the q stripe and the P stripe are failed, we're | |
1851 | * here due to a crc mismatch and we can't give them the | |
1852 | * data they want | |
1853 | */ | |
4c664611 QW |
1854 | if (rbio->bioc->raid_map[failb] == RAID6_Q_STRIPE) { |
1855 | if (rbio->bioc->raid_map[faila] == | |
8e5cfb55 | 1856 | RAID5_P_STRIPE) { |
58efbc9f | 1857 | err = BLK_STS_IOERR; |
53b381b3 DW |
1858 | goto cleanup; |
1859 | } | |
1860 | /* | |
1861 | * otherwise we have one bad data stripe and | |
1862 | * a good P stripe. raid5! | |
1863 | */ | |
1864 | goto pstripe; | |
1865 | } | |
1866 | ||
4c664611 | 1867 | if (rbio->bioc->raid_map[failb] == RAID5_P_STRIPE) { |
2c8cdd6e | 1868 | raid6_datap_recov(rbio->real_stripes, |
53b381b3 DW |
1869 | PAGE_SIZE, faila, pointers); |
1870 | } else { | |
2c8cdd6e | 1871 | raid6_2data_recov(rbio->real_stripes, |
53b381b3 DW |
1872 | PAGE_SIZE, faila, failb, |
1873 | pointers); | |
1874 | } | |
1875 | } else { | |
1876 | void *p; | |
1877 | ||
1878 | /* rebuild from P stripe here (raid5 or raid6) */ | |
1879 | BUG_ON(failb != -1); | |
1880 | pstripe: | |
1881 | /* Copy parity block into failed block to start with */ | |
69d24804 | 1882 | copy_page(pointers[faila], pointers[rbio->nr_data]); |
53b381b3 DW |
1883 | |
1884 | /* rearrange the pointer array */ | |
1885 | p = pointers[faila]; | |
1886 | for (stripe = faila; stripe < rbio->nr_data - 1; stripe++) | |
1887 | pointers[stripe] = pointers[stripe + 1]; | |
1888 | pointers[rbio->nr_data - 1] = p; | |
1889 | ||
1890 | /* xor in the rest */ | |
09cbfeaf | 1891 | run_xor(pointers, rbio->nr_data - 1, PAGE_SIZE); |
53b381b3 DW |
1892 | } |
1893 | /* if we're doing this rebuild as part of an rmw, go through | |
1894 | * and set all of our private rbio pages in the | |
1895 | * failed stripes as uptodate. This way finish_rmw will | |
1896 | * know they can be trusted. If this was a read reconstruction, | |
1897 | * other endio functions will fiddle the uptodate bits | |
1898 | */ | |
1b94b556 | 1899 | if (rbio->operation == BTRFS_RBIO_WRITE) { |
915e2290 | 1900 | for (i = 0; i < rbio->stripe_npages; i++) { |
53b381b3 DW |
1901 | if (faila != -1) { |
1902 | page = rbio_stripe_page(rbio, faila, i); | |
1903 | SetPageUptodate(page); | |
1904 | } | |
1905 | if (failb != -1) { | |
1906 | page = rbio_stripe_page(rbio, failb, i); | |
1907 | SetPageUptodate(page); | |
1908 | } | |
1909 | } | |
1910 | } | |
94a0b58d IW |
1911 | for (stripe = rbio->real_stripes - 1; stripe >= 0; stripe--) |
1912 | kunmap_local(unmap_array[stripe]); | |
53b381b3 DW |
1913 | } |
1914 | ||
58efbc9f | 1915 | err = BLK_STS_OK; |
53b381b3 | 1916 | cleanup: |
94a0b58d IW |
1917 | kfree(unmap_array); |
1918 | cleanup_pointers: | |
53b381b3 DW |
1919 | kfree(pointers); |
1920 | ||
1921 | cleanup_io: | |
580c6efa LB |
1922 | /* |
1923 | * Similar to READ_REBUILD, REBUILD_MISSING at this point also has a | |
1924 | * valid rbio which is consistent with ondisk content, thus such a | |
1925 | * valid rbio can be cached to avoid further disk reads. | |
1926 | */ | |
1927 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || | |
1928 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { | |
44ac474d LB |
1929 | /* |
1930 | * - In case of two failures, where rbio->failb != -1: | |
1931 | * | |
1932 | * Do not cache this rbio since the above read reconstruction | |
1933 | * (raid6_datap_recov() or raid6_2data_recov()) may have | |
1934 | * changed some content of stripes which are not identical to | |
1935 | * on-disk content any more, otherwise, a later write/recover | |
1936 | * may steal stripe_pages from this rbio and end up with | |
1937 | * corruptions or rebuild failures. | |
1938 | * | |
1939 | * - In case of single failure, where rbio->failb == -1: | |
1940 | * | |
1941 | * Cache this rbio iff the above read reconstruction is | |
52042d8e | 1942 | * executed without problems. |
44ac474d LB |
1943 | */ |
1944 | if (err == BLK_STS_OK && rbio->failb < 0) | |
4ae10b3a CM |
1945 | cache_rbio_pages(rbio); |
1946 | else | |
1947 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
1948 | ||
4246a0b6 | 1949 | rbio_orig_end_io(rbio, err); |
58efbc9f | 1950 | } else if (err == BLK_STS_OK) { |
53b381b3 DW |
1951 | rbio->faila = -1; |
1952 | rbio->failb = -1; | |
5a6ac9ea MX |
1953 | |
1954 | if (rbio->operation == BTRFS_RBIO_WRITE) | |
1955 | finish_rmw(rbio); | |
1956 | else if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) | |
1957 | finish_parity_scrub(rbio, 0); | |
1958 | else | |
1959 | BUG(); | |
53b381b3 | 1960 | } else { |
4246a0b6 | 1961 | rbio_orig_end_io(rbio, err); |
53b381b3 DW |
1962 | } |
1963 | } | |
1964 | ||
1965 | /* | |
1966 | * This is called only for stripes we've read from disk to | |
1967 | * reconstruct the parity. | |
1968 | */ | |
4246a0b6 | 1969 | static void raid_recover_end_io(struct bio *bio) |
53b381b3 DW |
1970 | { |
1971 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
1972 | ||
1973 | /* | |
1974 | * we only read stripe pages off the disk, set them | |
1975 | * up to date if there were no errors | |
1976 | */ | |
4e4cbee9 | 1977 | if (bio->bi_status) |
53b381b3 DW |
1978 | fail_bio_stripe(rbio, bio); |
1979 | else | |
1980 | set_bio_pages_uptodate(bio); | |
1981 | bio_put(bio); | |
1982 | ||
b89e1b01 | 1983 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
1984 | return; |
1985 | ||
4c664611 | 1986 | if (atomic_read(&rbio->error) > rbio->bioc->max_errors) |
58efbc9f | 1987 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
53b381b3 DW |
1988 | else |
1989 | __raid_recover_end_io(rbio); | |
1990 | } | |
1991 | ||
1992 | /* | |
1993 | * reads everything we need off the disk to reconstruct | |
1994 | * the parity. endio handlers trigger final reconstruction | |
1995 | * when the IO is done. | |
1996 | * | |
1997 | * This is used both for reads from the higher layers and for | |
1998 | * parity construction required to finish a rmw cycle. | |
1999 | */ | |
2000 | static int __raid56_parity_recover(struct btrfs_raid_bio *rbio) | |
2001 | { | |
2002 | int bios_to_read = 0; | |
53b381b3 DW |
2003 | struct bio_list bio_list; |
2004 | int ret; | |
53b381b3 DW |
2005 | int pagenr; |
2006 | int stripe; | |
2007 | struct bio *bio; | |
2008 | ||
2009 | bio_list_init(&bio_list); | |
2010 | ||
2011 | ret = alloc_rbio_pages(rbio); | |
2012 | if (ret) | |
2013 | goto cleanup; | |
2014 | ||
b89e1b01 | 2015 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
2016 | |
2017 | /* | |
4ae10b3a CM |
2018 | * read everything that hasn't failed. Thanks to the |
2019 | * stripe cache, it is possible that some or all of these | |
2020 | * pages are going to be uptodate. | |
53b381b3 | 2021 | */ |
2c8cdd6e | 2022 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
5588383e | 2023 | if (rbio->faila == stripe || rbio->failb == stripe) { |
b89e1b01 | 2024 | atomic_inc(&rbio->error); |
53b381b3 | 2025 | continue; |
5588383e | 2026 | } |
53b381b3 | 2027 | |
915e2290 | 2028 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
2029 | struct page *p; |
2030 | ||
2031 | /* | |
2032 | * the rmw code may have already read this | |
2033 | * page in | |
2034 | */ | |
2035 | p = rbio_stripe_page(rbio, stripe, pagenr); | |
2036 | if (PageUptodate(p)) | |
2037 | continue; | |
2038 | ||
2039 | ret = rbio_add_io_page(rbio, &bio_list, | |
2040 | rbio_stripe_page(rbio, stripe, pagenr), | |
2041 | stripe, pagenr, rbio->stripe_len); | |
2042 | if (ret < 0) | |
2043 | goto cleanup; | |
2044 | } | |
2045 | } | |
2046 | ||
2047 | bios_to_read = bio_list_size(&bio_list); | |
2048 | if (!bios_to_read) { | |
2049 | /* | |
2050 | * we might have no bios to read just because the pages | |
2051 | * were up to date, or we might have no bios to read because | |
2052 | * the devices were gone. | |
2053 | */ | |
4c664611 | 2054 | if (atomic_read(&rbio->error) <= rbio->bioc->max_errors) { |
53b381b3 | 2055 | __raid_recover_end_io(rbio); |
813f8a0e | 2056 | return 0; |
53b381b3 DW |
2057 | } else { |
2058 | goto cleanup; | |
2059 | } | |
2060 | } | |
2061 | ||
2062 | /* | |
4c664611 QW |
2063 | * The bioc may be freed once we submit the last bio. Make sure not to |
2064 | * touch it after that. | |
53b381b3 | 2065 | */ |
b89e1b01 | 2066 | atomic_set(&rbio->stripes_pending, bios_to_read); |
bf28a605 | 2067 | while ((bio = bio_list_pop(&bio_list))) { |
53b381b3 DW |
2068 | bio->bi_private = rbio; |
2069 | bio->bi_end_io = raid_recover_end_io; | |
ebcc3263 | 2070 | bio->bi_opf = REQ_OP_READ; |
53b381b3 | 2071 | |
6a258d72 | 2072 | btrfs_bio_wq_end_io(rbio->bioc->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
53b381b3 | 2073 | |
4e49ea4a | 2074 | submit_bio(bio); |
53b381b3 | 2075 | } |
813f8a0e | 2076 | |
53b381b3 DW |
2077 | return 0; |
2078 | ||
2079 | cleanup: | |
b4ee1782 OS |
2080 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
2081 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) | |
58efbc9f | 2082 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2083 | |
2084 | while ((bio = bio_list_pop(&bio_list))) | |
2085 | bio_put(bio); | |
2086 | ||
53b381b3 DW |
2087 | return -EIO; |
2088 | } | |
2089 | ||
2090 | /* | |
2091 | * the main entry point for reads from the higher layers. This | |
2092 | * is really only called when the normal read path had a failure, | |
2093 | * so we assume the bio they send down corresponds to a failed part | |
2094 | * of the drive. | |
2095 | */ | |
6a258d72 QW |
2096 | int raid56_parity_recover(struct bio *bio, struct btrfs_io_context *bioc, |
2097 | u64 stripe_len, int mirror_num, int generic_io) | |
53b381b3 | 2098 | { |
6a258d72 | 2099 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
53b381b3 DW |
2100 | struct btrfs_raid_bio *rbio; |
2101 | int ret; | |
2102 | ||
abad60c6 | 2103 | if (generic_io) { |
4c664611 | 2104 | ASSERT(bioc->mirror_num == mirror_num); |
c3a3b19b | 2105 | btrfs_bio(bio)->mirror_num = mirror_num; |
abad60c6 LB |
2106 | } |
2107 | ||
4c664611 | 2108 | rbio = alloc_rbio(fs_info, bioc, stripe_len); |
af8e2d1d | 2109 | if (IS_ERR(rbio)) { |
6e9606d2 | 2110 | if (generic_io) |
4c664611 | 2111 | btrfs_put_bioc(bioc); |
53b381b3 | 2112 | return PTR_ERR(rbio); |
af8e2d1d | 2113 | } |
53b381b3 | 2114 | |
1b94b556 | 2115 | rbio->operation = BTRFS_RBIO_READ_REBUILD; |
53b381b3 | 2116 | bio_list_add(&rbio->bio_list, bio); |
4f024f37 | 2117 | rbio->bio_list_bytes = bio->bi_iter.bi_size; |
53b381b3 DW |
2118 | |
2119 | rbio->faila = find_logical_bio_stripe(rbio, bio); | |
2120 | if (rbio->faila == -1) { | |
0b246afa | 2121 | btrfs_warn(fs_info, |
4c664611 | 2122 | "%s could not find the bad stripe in raid56 so that we cannot recover any more (bio has logical %llu len %llu, bioc has map_type %llu)", |
1201b58b | 2123 | __func__, bio->bi_iter.bi_sector << 9, |
4c664611 | 2124 | (u64)bio->bi_iter.bi_size, bioc->map_type); |
6e9606d2 | 2125 | if (generic_io) |
4c664611 | 2126 | btrfs_put_bioc(bioc); |
53b381b3 DW |
2127 | kfree(rbio); |
2128 | return -EIO; | |
2129 | } | |
2130 | ||
4245215d | 2131 | if (generic_io) { |
0b246afa | 2132 | btrfs_bio_counter_inc_noblocked(fs_info); |
4245215d MX |
2133 | rbio->generic_bio_cnt = 1; |
2134 | } else { | |
4c664611 | 2135 | btrfs_get_bioc(bioc); |
4245215d MX |
2136 | } |
2137 | ||
53b381b3 | 2138 | /* |
8810f751 LB |
2139 | * Loop retry: |
2140 | * for 'mirror == 2', reconstruct from all other stripes. | |
2141 | * for 'mirror_num > 2', select a stripe to fail on every retry. | |
53b381b3 | 2142 | */ |
8810f751 LB |
2143 | if (mirror_num > 2) { |
2144 | /* | |
2145 | * 'mirror == 3' is to fail the p stripe and | |
2146 | * reconstruct from the q stripe. 'mirror > 3' is to | |
2147 | * fail a data stripe and reconstruct from p+q stripe. | |
2148 | */ | |
2149 | rbio->failb = rbio->real_stripes - (mirror_num - 1); | |
2150 | ASSERT(rbio->failb > 0); | |
2151 | if (rbio->failb <= rbio->faila) | |
2152 | rbio->failb--; | |
2153 | } | |
53b381b3 DW |
2154 | |
2155 | ret = lock_stripe_add(rbio); | |
2156 | ||
2157 | /* | |
2158 | * __raid56_parity_recover will end the bio with | |
2159 | * any errors it hits. We don't want to return | |
2160 | * its error value up the stack because our caller | |
2161 | * will end up calling bio_endio with any nonzero | |
2162 | * return | |
2163 | */ | |
2164 | if (ret == 0) | |
2165 | __raid56_parity_recover(rbio); | |
2166 | /* | |
2167 | * our rbio has been added to the list of | |
2168 | * rbios that will be handled after the | |
2169 | * currently lock owner is done | |
2170 | */ | |
2171 | return 0; | |
2172 | ||
2173 | } | |
2174 | ||
2175 | static void rmw_work(struct btrfs_work *work) | |
2176 | { | |
2177 | struct btrfs_raid_bio *rbio; | |
2178 | ||
2179 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2180 | raid56_rmw_stripe(rbio); | |
2181 | } | |
2182 | ||
2183 | static void read_rebuild_work(struct btrfs_work *work) | |
2184 | { | |
2185 | struct btrfs_raid_bio *rbio; | |
2186 | ||
2187 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2188 | __raid56_parity_recover(rbio); | |
2189 | } | |
5a6ac9ea MX |
2190 | |
2191 | /* | |
2192 | * The following code is used to scrub/replace the parity stripe | |
2193 | * | |
4c664611 | 2194 | * Caller must have already increased bio_counter for getting @bioc. |
ae6529c3 | 2195 | * |
5a6ac9ea MX |
2196 | * Note: We need make sure all the pages that add into the scrub/replace |
2197 | * raid bio are correct and not be changed during the scrub/replace. That | |
2198 | * is those pages just hold metadata or file data with checksum. | |
2199 | */ | |
2200 | ||
6a258d72 QW |
2201 | struct btrfs_raid_bio *raid56_parity_alloc_scrub_rbio(struct bio *bio, |
2202 | struct btrfs_io_context *bioc, | |
2203 | u64 stripe_len, struct btrfs_device *scrub_dev, | |
2204 | unsigned long *dbitmap, int stripe_nsectors) | |
5a6ac9ea | 2205 | { |
6a258d72 | 2206 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
5a6ac9ea MX |
2207 | struct btrfs_raid_bio *rbio; |
2208 | int i; | |
2209 | ||
4c664611 | 2210 | rbio = alloc_rbio(fs_info, bioc, stripe_len); |
5a6ac9ea MX |
2211 | if (IS_ERR(rbio)) |
2212 | return NULL; | |
2213 | bio_list_add(&rbio->bio_list, bio); | |
2214 | /* | |
2215 | * This is a special bio which is used to hold the completion handler | |
2216 | * and make the scrub rbio is similar to the other types | |
2217 | */ | |
2218 | ASSERT(!bio->bi_iter.bi_size); | |
2219 | rbio->operation = BTRFS_RBIO_PARITY_SCRUB; | |
2220 | ||
9cd3a7eb | 2221 | /* |
4c664611 | 2222 | * After mapping bioc with BTRFS_MAP_WRITE, parities have been sorted |
9cd3a7eb LB |
2223 | * to the end position, so this search can start from the first parity |
2224 | * stripe. | |
2225 | */ | |
2226 | for (i = rbio->nr_data; i < rbio->real_stripes; i++) { | |
4c664611 | 2227 | if (bioc->stripes[i].dev == scrub_dev) { |
5a6ac9ea MX |
2228 | rbio->scrubp = i; |
2229 | break; | |
2230 | } | |
2231 | } | |
9cd3a7eb | 2232 | ASSERT(i < rbio->real_stripes); |
5a6ac9ea MX |
2233 | |
2234 | /* Now we just support the sectorsize equals to page size */ | |
0b246afa | 2235 | ASSERT(fs_info->sectorsize == PAGE_SIZE); |
5a6ac9ea MX |
2236 | ASSERT(rbio->stripe_npages == stripe_nsectors); |
2237 | bitmap_copy(rbio->dbitmap, dbitmap, stripe_nsectors); | |
2238 | ||
ae6529c3 | 2239 | /* |
4c664611 | 2240 | * We have already increased bio_counter when getting bioc, record it |
ae6529c3 QW |
2241 | * so we can free it at rbio_orig_end_io(). |
2242 | */ | |
2243 | rbio->generic_bio_cnt = 1; | |
2244 | ||
5a6ac9ea MX |
2245 | return rbio; |
2246 | } | |
2247 | ||
b4ee1782 OS |
2248 | /* Used for both parity scrub and missing. */ |
2249 | void raid56_add_scrub_pages(struct btrfs_raid_bio *rbio, struct page *page, | |
2250 | u64 logical) | |
5a6ac9ea MX |
2251 | { |
2252 | int stripe_offset; | |
2253 | int index; | |
2254 | ||
4c664611 QW |
2255 | ASSERT(logical >= rbio->bioc->raid_map[0]); |
2256 | ASSERT(logical + PAGE_SIZE <= rbio->bioc->raid_map[0] + | |
5a6ac9ea | 2257 | rbio->stripe_len * rbio->nr_data); |
4c664611 | 2258 | stripe_offset = (int)(logical - rbio->bioc->raid_map[0]); |
09cbfeaf | 2259 | index = stripe_offset >> PAGE_SHIFT; |
5a6ac9ea MX |
2260 | rbio->bio_pages[index] = page; |
2261 | } | |
2262 | ||
2263 | /* | |
2264 | * We just scrub the parity that we have correct data on the same horizontal, | |
2265 | * so we needn't allocate all pages for all the stripes. | |
2266 | */ | |
2267 | static int alloc_rbio_essential_pages(struct btrfs_raid_bio *rbio) | |
2268 | { | |
2269 | int i; | |
2270 | int bit; | |
2271 | int index; | |
2272 | struct page *page; | |
2273 | ||
2274 | for_each_set_bit(bit, rbio->dbitmap, rbio->stripe_npages) { | |
2c8cdd6e | 2275 | for (i = 0; i < rbio->real_stripes; i++) { |
5a6ac9ea MX |
2276 | index = i * rbio->stripe_npages + bit; |
2277 | if (rbio->stripe_pages[index]) | |
2278 | continue; | |
2279 | ||
b0ee5e1e | 2280 | page = alloc_page(GFP_NOFS); |
5a6ac9ea MX |
2281 | if (!page) |
2282 | return -ENOMEM; | |
2283 | rbio->stripe_pages[index] = page; | |
5a6ac9ea MX |
2284 | } |
2285 | } | |
2286 | return 0; | |
2287 | } | |
2288 | ||
5a6ac9ea MX |
2289 | static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, |
2290 | int need_check) | |
2291 | { | |
4c664611 | 2292 | struct btrfs_io_context *bioc = rbio->bioc; |
1389053e KC |
2293 | void **pointers = rbio->finish_pointers; |
2294 | unsigned long *pbitmap = rbio->finish_pbitmap; | |
5a6ac9ea MX |
2295 | int nr_data = rbio->nr_data; |
2296 | int stripe; | |
2297 | int pagenr; | |
c17af965 | 2298 | bool has_qstripe; |
5a6ac9ea MX |
2299 | struct page *p_page = NULL; |
2300 | struct page *q_page = NULL; | |
2301 | struct bio_list bio_list; | |
2302 | struct bio *bio; | |
76035976 | 2303 | int is_replace = 0; |
5a6ac9ea MX |
2304 | int ret; |
2305 | ||
2306 | bio_list_init(&bio_list); | |
2307 | ||
c17af965 DS |
2308 | if (rbio->real_stripes - rbio->nr_data == 1) |
2309 | has_qstripe = false; | |
2310 | else if (rbio->real_stripes - rbio->nr_data == 2) | |
2311 | has_qstripe = true; | |
2312 | else | |
5a6ac9ea | 2313 | BUG(); |
5a6ac9ea | 2314 | |
4c664611 | 2315 | if (bioc->num_tgtdevs && bioc->tgtdev_map[rbio->scrubp]) { |
76035976 MX |
2316 | is_replace = 1; |
2317 | bitmap_copy(pbitmap, rbio->dbitmap, rbio->stripe_npages); | |
2318 | } | |
2319 | ||
5a6ac9ea MX |
2320 | /* |
2321 | * Because the higher layers(scrubber) are unlikely to | |
2322 | * use this area of the disk again soon, so don't cache | |
2323 | * it. | |
2324 | */ | |
2325 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
2326 | ||
2327 | if (!need_check) | |
2328 | goto writeback; | |
2329 | ||
b0ee5e1e | 2330 | p_page = alloc_page(GFP_NOFS); |
5a6ac9ea MX |
2331 | if (!p_page) |
2332 | goto cleanup; | |
2333 | SetPageUptodate(p_page); | |
2334 | ||
c17af965 | 2335 | if (has_qstripe) { |
d70cef0d | 2336 | /* RAID6, allocate and map temp space for the Q stripe */ |
b0ee5e1e | 2337 | q_page = alloc_page(GFP_NOFS); |
5a6ac9ea MX |
2338 | if (!q_page) { |
2339 | __free_page(p_page); | |
2340 | goto cleanup; | |
2341 | } | |
2342 | SetPageUptodate(q_page); | |
94a0b58d | 2343 | pointers[rbio->real_stripes - 1] = kmap_local_page(q_page); |
5a6ac9ea MX |
2344 | } |
2345 | ||
2346 | atomic_set(&rbio->error, 0); | |
2347 | ||
d70cef0d | 2348 | /* Map the parity stripe just once */ |
94a0b58d | 2349 | pointers[nr_data] = kmap_local_page(p_page); |
d70cef0d | 2350 | |
5a6ac9ea MX |
2351 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { |
2352 | struct page *p; | |
2353 | void *parity; | |
2354 | /* first collect one page from each data stripe */ | |
2355 | for (stripe = 0; stripe < nr_data; stripe++) { | |
2356 | p = page_in_rbio(rbio, stripe, pagenr, 0); | |
94a0b58d | 2357 | pointers[stripe] = kmap_local_page(p); |
5a6ac9ea MX |
2358 | } |
2359 | ||
c17af965 | 2360 | if (has_qstripe) { |
d70cef0d | 2361 | /* RAID6, call the library function to fill in our P/Q */ |
2c8cdd6e | 2362 | raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE, |
5a6ac9ea MX |
2363 | pointers); |
2364 | } else { | |
2365 | /* raid5 */ | |
69d24804 | 2366 | copy_page(pointers[nr_data], pointers[0]); |
09cbfeaf | 2367 | run_xor(pointers + 1, nr_data - 1, PAGE_SIZE); |
5a6ac9ea MX |
2368 | } |
2369 | ||
01327610 | 2370 | /* Check scrubbing parity and repair it */ |
5a6ac9ea | 2371 | p = rbio_stripe_page(rbio, rbio->scrubp, pagenr); |
58c1a35c | 2372 | parity = kmap_local_page(p); |
09cbfeaf | 2373 | if (memcmp(parity, pointers[rbio->scrubp], PAGE_SIZE)) |
69d24804 | 2374 | copy_page(parity, pointers[rbio->scrubp]); |
5a6ac9ea MX |
2375 | else |
2376 | /* Parity is right, needn't writeback */ | |
2377 | bitmap_clear(rbio->dbitmap, pagenr, 1); | |
58c1a35c | 2378 | kunmap_local(parity); |
5a6ac9ea | 2379 | |
94a0b58d IW |
2380 | for (stripe = nr_data - 1; stripe >= 0; stripe--) |
2381 | kunmap_local(pointers[stripe]); | |
5a6ac9ea MX |
2382 | } |
2383 | ||
94a0b58d | 2384 | kunmap_local(pointers[nr_data]); |
5a6ac9ea | 2385 | __free_page(p_page); |
d70cef0d | 2386 | if (q_page) { |
94a0b58d | 2387 | kunmap_local(pointers[rbio->real_stripes - 1]); |
5a6ac9ea | 2388 | __free_page(q_page); |
d70cef0d | 2389 | } |
5a6ac9ea MX |
2390 | |
2391 | writeback: | |
2392 | /* | |
2393 | * time to start writing. Make bios for everything from the | |
2394 | * higher layers (the bio_list in our rbio) and our p/q. Ignore | |
2395 | * everything else. | |
2396 | */ | |
2397 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { | |
2398 | struct page *page; | |
2399 | ||
2400 | page = rbio_stripe_page(rbio, rbio->scrubp, pagenr); | |
2401 | ret = rbio_add_io_page(rbio, &bio_list, | |
2402 | page, rbio->scrubp, pagenr, rbio->stripe_len); | |
2403 | if (ret) | |
2404 | goto cleanup; | |
2405 | } | |
2406 | ||
76035976 MX |
2407 | if (!is_replace) |
2408 | goto submit_write; | |
2409 | ||
2410 | for_each_set_bit(pagenr, pbitmap, rbio->stripe_npages) { | |
2411 | struct page *page; | |
2412 | ||
2413 | page = rbio_stripe_page(rbio, rbio->scrubp, pagenr); | |
2414 | ret = rbio_add_io_page(rbio, &bio_list, page, | |
4c664611 | 2415 | bioc->tgtdev_map[rbio->scrubp], |
76035976 MX |
2416 | pagenr, rbio->stripe_len); |
2417 | if (ret) | |
2418 | goto cleanup; | |
2419 | } | |
2420 | ||
2421 | submit_write: | |
5a6ac9ea MX |
2422 | nr_data = bio_list_size(&bio_list); |
2423 | if (!nr_data) { | |
2424 | /* Every parity is right */ | |
58efbc9f | 2425 | rbio_orig_end_io(rbio, BLK_STS_OK); |
5a6ac9ea MX |
2426 | return; |
2427 | } | |
2428 | ||
2429 | atomic_set(&rbio->stripes_pending, nr_data); | |
2430 | ||
bf28a605 | 2431 | while ((bio = bio_list_pop(&bio_list))) { |
5a6ac9ea | 2432 | bio->bi_private = rbio; |
a6111d11 | 2433 | bio->bi_end_io = raid_write_end_io; |
ebcc3263 | 2434 | bio->bi_opf = REQ_OP_WRITE; |
4e49ea4a MC |
2435 | |
2436 | submit_bio(bio); | |
5a6ac9ea MX |
2437 | } |
2438 | return; | |
2439 | ||
2440 | cleanup: | |
58efbc9f | 2441 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2442 | |
2443 | while ((bio = bio_list_pop(&bio_list))) | |
2444 | bio_put(bio); | |
5a6ac9ea MX |
2445 | } |
2446 | ||
2447 | static inline int is_data_stripe(struct btrfs_raid_bio *rbio, int stripe) | |
2448 | { | |
2449 | if (stripe >= 0 && stripe < rbio->nr_data) | |
2450 | return 1; | |
2451 | return 0; | |
2452 | } | |
2453 | ||
2454 | /* | |
2455 | * While we're doing the parity check and repair, we could have errors | |
2456 | * in reading pages off the disk. This checks for errors and if we're | |
2457 | * not able to read the page it'll trigger parity reconstruction. The | |
2458 | * parity scrub will be finished after we've reconstructed the failed | |
2459 | * stripes | |
2460 | */ | |
2461 | static void validate_rbio_for_parity_scrub(struct btrfs_raid_bio *rbio) | |
2462 | { | |
4c664611 | 2463 | if (atomic_read(&rbio->error) > rbio->bioc->max_errors) |
5a6ac9ea MX |
2464 | goto cleanup; |
2465 | ||
2466 | if (rbio->faila >= 0 || rbio->failb >= 0) { | |
2467 | int dfail = 0, failp = -1; | |
2468 | ||
2469 | if (is_data_stripe(rbio, rbio->faila)) | |
2470 | dfail++; | |
2471 | else if (is_parity_stripe(rbio->faila)) | |
2472 | failp = rbio->faila; | |
2473 | ||
2474 | if (is_data_stripe(rbio, rbio->failb)) | |
2475 | dfail++; | |
2476 | else if (is_parity_stripe(rbio->failb)) | |
2477 | failp = rbio->failb; | |
2478 | ||
2479 | /* | |
2480 | * Because we can not use a scrubbing parity to repair | |
2481 | * the data, so the capability of the repair is declined. | |
2482 | * (In the case of RAID5, we can not repair anything) | |
2483 | */ | |
4c664611 | 2484 | if (dfail > rbio->bioc->max_errors - 1) |
5a6ac9ea MX |
2485 | goto cleanup; |
2486 | ||
2487 | /* | |
2488 | * If all data is good, only parity is correctly, just | |
2489 | * repair the parity. | |
2490 | */ | |
2491 | if (dfail == 0) { | |
2492 | finish_parity_scrub(rbio, 0); | |
2493 | return; | |
2494 | } | |
2495 | ||
2496 | /* | |
2497 | * Here means we got one corrupted data stripe and one | |
2498 | * corrupted parity on RAID6, if the corrupted parity | |
01327610 | 2499 | * is scrubbing parity, luckily, use the other one to repair |
5a6ac9ea MX |
2500 | * the data, or we can not repair the data stripe. |
2501 | */ | |
2502 | if (failp != rbio->scrubp) | |
2503 | goto cleanup; | |
2504 | ||
2505 | __raid_recover_end_io(rbio); | |
2506 | } else { | |
2507 | finish_parity_scrub(rbio, 1); | |
2508 | } | |
2509 | return; | |
2510 | ||
2511 | cleanup: | |
58efbc9f | 2512 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
5a6ac9ea MX |
2513 | } |
2514 | ||
2515 | /* | |
2516 | * end io for the read phase of the rmw cycle. All the bios here are physical | |
2517 | * stripe bios we've read from the disk so we can recalculate the parity of the | |
2518 | * stripe. | |
2519 | * | |
2520 | * This will usually kick off finish_rmw once all the bios are read in, but it | |
2521 | * may trigger parity reconstruction if we had any errors along the way | |
2522 | */ | |
4246a0b6 | 2523 | static void raid56_parity_scrub_end_io(struct bio *bio) |
5a6ac9ea MX |
2524 | { |
2525 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
2526 | ||
4e4cbee9 | 2527 | if (bio->bi_status) |
5a6ac9ea MX |
2528 | fail_bio_stripe(rbio, bio); |
2529 | else | |
2530 | set_bio_pages_uptodate(bio); | |
2531 | ||
2532 | bio_put(bio); | |
2533 | ||
2534 | if (!atomic_dec_and_test(&rbio->stripes_pending)) | |
2535 | return; | |
2536 | ||
2537 | /* | |
2538 | * this will normally call finish_rmw to start our write | |
2539 | * but if there are any failed stripes we'll reconstruct | |
2540 | * from parity first | |
2541 | */ | |
2542 | validate_rbio_for_parity_scrub(rbio); | |
2543 | } | |
2544 | ||
2545 | static void raid56_parity_scrub_stripe(struct btrfs_raid_bio *rbio) | |
2546 | { | |
2547 | int bios_to_read = 0; | |
5a6ac9ea MX |
2548 | struct bio_list bio_list; |
2549 | int ret; | |
2550 | int pagenr; | |
2551 | int stripe; | |
2552 | struct bio *bio; | |
2553 | ||
785884fc LB |
2554 | bio_list_init(&bio_list); |
2555 | ||
5a6ac9ea MX |
2556 | ret = alloc_rbio_essential_pages(rbio); |
2557 | if (ret) | |
2558 | goto cleanup; | |
2559 | ||
5a6ac9ea MX |
2560 | atomic_set(&rbio->error, 0); |
2561 | /* | |
2562 | * build a list of bios to read all the missing parts of this | |
2563 | * stripe | |
2564 | */ | |
2c8cdd6e | 2565 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
5a6ac9ea MX |
2566 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { |
2567 | struct page *page; | |
2568 | /* | |
2569 | * we want to find all the pages missing from | |
2570 | * the rbio and read them from the disk. If | |
2571 | * page_in_rbio finds a page in the bio list | |
2572 | * we don't need to read it off the stripe. | |
2573 | */ | |
2574 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
2575 | if (page) | |
2576 | continue; | |
2577 | ||
2578 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
2579 | /* | |
2580 | * the bio cache may have handed us an uptodate | |
2581 | * page. If so, be happy and use it | |
2582 | */ | |
2583 | if (PageUptodate(page)) | |
2584 | continue; | |
2585 | ||
2586 | ret = rbio_add_io_page(rbio, &bio_list, page, | |
2587 | stripe, pagenr, rbio->stripe_len); | |
2588 | if (ret) | |
2589 | goto cleanup; | |
2590 | } | |
2591 | } | |
2592 | ||
2593 | bios_to_read = bio_list_size(&bio_list); | |
2594 | if (!bios_to_read) { | |
2595 | /* | |
2596 | * this can happen if others have merged with | |
2597 | * us, it means there is nothing left to read. | |
2598 | * But if there are missing devices it may not be | |
2599 | * safe to do the full stripe write yet. | |
2600 | */ | |
2601 | goto finish; | |
2602 | } | |
2603 | ||
2604 | /* | |
4c664611 QW |
2605 | * The bioc may be freed once we submit the last bio. Make sure not to |
2606 | * touch it after that. | |
5a6ac9ea MX |
2607 | */ |
2608 | atomic_set(&rbio->stripes_pending, bios_to_read); | |
bf28a605 | 2609 | while ((bio = bio_list_pop(&bio_list))) { |
5a6ac9ea MX |
2610 | bio->bi_private = rbio; |
2611 | bio->bi_end_io = raid56_parity_scrub_end_io; | |
ebcc3263 | 2612 | bio->bi_opf = REQ_OP_READ; |
5a6ac9ea | 2613 | |
6a258d72 | 2614 | btrfs_bio_wq_end_io(rbio->bioc->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
5a6ac9ea | 2615 | |
4e49ea4a | 2616 | submit_bio(bio); |
5a6ac9ea MX |
2617 | } |
2618 | /* the actual write will happen once the reads are done */ | |
2619 | return; | |
2620 | ||
2621 | cleanup: | |
58efbc9f | 2622 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2623 | |
2624 | while ((bio = bio_list_pop(&bio_list))) | |
2625 | bio_put(bio); | |
2626 | ||
5a6ac9ea MX |
2627 | return; |
2628 | ||
2629 | finish: | |
2630 | validate_rbio_for_parity_scrub(rbio); | |
2631 | } | |
2632 | ||
2633 | static void scrub_parity_work(struct btrfs_work *work) | |
2634 | { | |
2635 | struct btrfs_raid_bio *rbio; | |
2636 | ||
2637 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2638 | raid56_parity_scrub_stripe(rbio); | |
2639 | } | |
2640 | ||
5a6ac9ea MX |
2641 | void raid56_parity_submit_scrub_rbio(struct btrfs_raid_bio *rbio) |
2642 | { | |
2643 | if (!lock_stripe_add(rbio)) | |
a81b747d | 2644 | start_async_work(rbio, scrub_parity_work); |
5a6ac9ea | 2645 | } |
b4ee1782 OS |
2646 | |
2647 | /* The following code is used for dev replace of a missing RAID 5/6 device. */ | |
2648 | ||
2649 | struct btrfs_raid_bio * | |
6a258d72 QW |
2650 | raid56_alloc_missing_rbio(struct bio *bio, struct btrfs_io_context *bioc, |
2651 | u64 length) | |
b4ee1782 | 2652 | { |
6a258d72 | 2653 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
b4ee1782 OS |
2654 | struct btrfs_raid_bio *rbio; |
2655 | ||
4c664611 | 2656 | rbio = alloc_rbio(fs_info, bioc, length); |
b4ee1782 OS |
2657 | if (IS_ERR(rbio)) |
2658 | return NULL; | |
2659 | ||
2660 | rbio->operation = BTRFS_RBIO_REBUILD_MISSING; | |
2661 | bio_list_add(&rbio->bio_list, bio); | |
2662 | /* | |
2663 | * This is a special bio which is used to hold the completion handler | |
2664 | * and make the scrub rbio is similar to the other types | |
2665 | */ | |
2666 | ASSERT(!bio->bi_iter.bi_size); | |
2667 | ||
2668 | rbio->faila = find_logical_bio_stripe(rbio, bio); | |
2669 | if (rbio->faila == -1) { | |
2670 | BUG(); | |
2671 | kfree(rbio); | |
2672 | return NULL; | |
2673 | } | |
2674 | ||
ae6529c3 | 2675 | /* |
4c664611 | 2676 | * When we get bioc, we have already increased bio_counter, record it |
ae6529c3 QW |
2677 | * so we can free it at rbio_orig_end_io() |
2678 | */ | |
2679 | rbio->generic_bio_cnt = 1; | |
2680 | ||
b4ee1782 OS |
2681 | return rbio; |
2682 | } | |
2683 | ||
b4ee1782 OS |
2684 | void raid56_submit_missing_rbio(struct btrfs_raid_bio *rbio) |
2685 | { | |
2686 | if (!lock_stripe_add(rbio)) | |
e66d8d5a | 2687 | start_async_work(rbio, read_rebuild_work); |
b4ee1782 | 2688 | } |