Commit | Line | Data |
---|---|---|
53b381b3 DW |
1 | /* |
2 | * Copyright (C) 2012 Fusion-io All rights reserved. | |
3 | * Copyright (C) 2012 Intel Corp. All rights reserved. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or | |
6 | * modify it under the terms of the GNU General Public | |
7 | * License v2 as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
12 | * General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public | |
15 | * License along with this program; if not, write to the | |
16 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
17 | * Boston, MA 021110-1307, USA. | |
18 | */ | |
19 | #include <linux/sched.h> | |
20 | #include <linux/wait.h> | |
21 | #include <linux/bio.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/buffer_head.h> | |
24 | #include <linux/blkdev.h> | |
25 | #include <linux/random.h> | |
26 | #include <linux/iocontext.h> | |
27 | #include <linux/capability.h> | |
28 | #include <linux/ratelimit.h> | |
29 | #include <linux/kthread.h> | |
30 | #include <linux/raid/pq.h> | |
31 | #include <linux/hash.h> | |
32 | #include <linux/list_sort.h> | |
33 | #include <linux/raid/xor.h> | |
818e010b | 34 | #include <linux/mm.h> |
53b381b3 | 35 | #include <asm/div64.h> |
53b381b3 DW |
36 | #include "ctree.h" |
37 | #include "extent_map.h" | |
38 | #include "disk-io.h" | |
39 | #include "transaction.h" | |
40 | #include "print-tree.h" | |
41 | #include "volumes.h" | |
42 | #include "raid56.h" | |
43 | #include "async-thread.h" | |
44 | #include "check-integrity.h" | |
45 | #include "rcu-string.h" | |
46 | ||
47 | /* set when additional merges to this rbio are not allowed */ | |
48 | #define RBIO_RMW_LOCKED_BIT 1 | |
49 | ||
4ae10b3a CM |
50 | /* |
51 | * set when this rbio is sitting in the hash, but it is just a cache | |
52 | * of past RMW | |
53 | */ | |
54 | #define RBIO_CACHE_BIT 2 | |
55 | ||
56 | /* | |
57 | * set when it is safe to trust the stripe_pages for caching | |
58 | */ | |
59 | #define RBIO_CACHE_READY_BIT 3 | |
60 | ||
4ae10b3a CM |
61 | #define RBIO_CACHE_SIZE 1024 |
62 | ||
1b94b556 | 63 | enum btrfs_rbio_ops { |
b4ee1782 OS |
64 | BTRFS_RBIO_WRITE, |
65 | BTRFS_RBIO_READ_REBUILD, | |
66 | BTRFS_RBIO_PARITY_SCRUB, | |
67 | BTRFS_RBIO_REBUILD_MISSING, | |
1b94b556 MX |
68 | }; |
69 | ||
53b381b3 DW |
70 | struct btrfs_raid_bio { |
71 | struct btrfs_fs_info *fs_info; | |
72 | struct btrfs_bio *bbio; | |
73 | ||
53b381b3 DW |
74 | /* while we're doing rmw on a stripe |
75 | * we put it into a hash table so we can | |
76 | * lock the stripe and merge more rbios | |
77 | * into it. | |
78 | */ | |
79 | struct list_head hash_list; | |
80 | ||
4ae10b3a CM |
81 | /* |
82 | * LRU list for the stripe cache | |
83 | */ | |
84 | struct list_head stripe_cache; | |
85 | ||
53b381b3 DW |
86 | /* |
87 | * for scheduling work in the helper threads | |
88 | */ | |
89 | struct btrfs_work work; | |
90 | ||
91 | /* | |
92 | * bio list and bio_list_lock are used | |
93 | * to add more bios into the stripe | |
94 | * in hopes of avoiding the full rmw | |
95 | */ | |
96 | struct bio_list bio_list; | |
97 | spinlock_t bio_list_lock; | |
98 | ||
6ac0f488 CM |
99 | /* also protected by the bio_list_lock, the |
100 | * plug list is used by the plugging code | |
101 | * to collect partial bios while plugged. The | |
102 | * stripe locking code also uses it to hand off | |
53b381b3 DW |
103 | * the stripe lock to the next pending IO |
104 | */ | |
105 | struct list_head plug_list; | |
106 | ||
107 | /* | |
108 | * flags that tell us if it is safe to | |
109 | * merge with this bio | |
110 | */ | |
111 | unsigned long flags; | |
112 | ||
113 | /* size of each individual stripe on disk */ | |
114 | int stripe_len; | |
115 | ||
116 | /* number of data stripes (no p/q) */ | |
117 | int nr_data; | |
118 | ||
2c8cdd6e MX |
119 | int real_stripes; |
120 | ||
5a6ac9ea | 121 | int stripe_npages; |
53b381b3 DW |
122 | /* |
123 | * set if we're doing a parity rebuild | |
124 | * for a read from higher up, which is handled | |
125 | * differently from a parity rebuild as part of | |
126 | * rmw | |
127 | */ | |
1b94b556 | 128 | enum btrfs_rbio_ops operation; |
53b381b3 DW |
129 | |
130 | /* first bad stripe */ | |
131 | int faila; | |
132 | ||
133 | /* second bad stripe (for raid6 use) */ | |
134 | int failb; | |
135 | ||
5a6ac9ea | 136 | int scrubp; |
53b381b3 DW |
137 | /* |
138 | * number of pages needed to represent the full | |
139 | * stripe | |
140 | */ | |
141 | int nr_pages; | |
142 | ||
143 | /* | |
144 | * size of all the bios in the bio_list. This | |
145 | * helps us decide if the rbio maps to a full | |
146 | * stripe or not | |
147 | */ | |
148 | int bio_list_bytes; | |
149 | ||
4245215d MX |
150 | int generic_bio_cnt; |
151 | ||
dec95574 | 152 | refcount_t refs; |
53b381b3 | 153 | |
b89e1b01 MX |
154 | atomic_t stripes_pending; |
155 | ||
156 | atomic_t error; | |
53b381b3 DW |
157 | /* |
158 | * these are two arrays of pointers. We allocate the | |
159 | * rbio big enough to hold them both and setup their | |
160 | * locations when the rbio is allocated | |
161 | */ | |
162 | ||
163 | /* pointers to pages that we allocated for | |
164 | * reading/writing stripes directly from the disk (including P/Q) | |
165 | */ | |
166 | struct page **stripe_pages; | |
167 | ||
168 | /* | |
169 | * pointers to the pages in the bio_list. Stored | |
170 | * here for faster lookup | |
171 | */ | |
172 | struct page **bio_pages; | |
5a6ac9ea MX |
173 | |
174 | /* | |
175 | * bitmap to record which horizontal stripe has data | |
176 | */ | |
177 | unsigned long *dbitmap; | |
53b381b3 DW |
178 | }; |
179 | ||
180 | static int __raid56_parity_recover(struct btrfs_raid_bio *rbio); | |
181 | static noinline void finish_rmw(struct btrfs_raid_bio *rbio); | |
182 | static void rmw_work(struct btrfs_work *work); | |
183 | static void read_rebuild_work(struct btrfs_work *work); | |
184 | static void async_rmw_stripe(struct btrfs_raid_bio *rbio); | |
185 | static void async_read_rebuild(struct btrfs_raid_bio *rbio); | |
186 | static int fail_bio_stripe(struct btrfs_raid_bio *rbio, struct bio *bio); | |
187 | static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed); | |
188 | static void __free_raid_bio(struct btrfs_raid_bio *rbio); | |
189 | static void index_rbio_pages(struct btrfs_raid_bio *rbio); | |
190 | static int alloc_rbio_pages(struct btrfs_raid_bio *rbio); | |
191 | ||
5a6ac9ea MX |
192 | static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, |
193 | int need_check); | |
194 | static void async_scrub_parity(struct btrfs_raid_bio *rbio); | |
195 | ||
53b381b3 DW |
196 | /* |
197 | * the stripe hash table is used for locking, and to collect | |
198 | * bios in hopes of making a full stripe | |
199 | */ | |
200 | int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info) | |
201 | { | |
202 | struct btrfs_stripe_hash_table *table; | |
203 | struct btrfs_stripe_hash_table *x; | |
204 | struct btrfs_stripe_hash *cur; | |
205 | struct btrfs_stripe_hash *h; | |
206 | int num_entries = 1 << BTRFS_STRIPE_HASH_TABLE_BITS; | |
207 | int i; | |
83c8266a | 208 | int table_size; |
53b381b3 DW |
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 | */ | |
220 | table_size = sizeof(*table) + sizeof(*h) * num_entries; | |
818e010b DS |
221 | table = kvzalloc(table_size, GFP_KERNEL); |
222 | if (!table) | |
223 | return -ENOMEM; | |
53b381b3 | 224 | |
4ae10b3a CM |
225 | spin_lock_init(&table->cache_lock); |
226 | INIT_LIST_HEAD(&table->stripe_cache); | |
227 | ||
53b381b3 DW |
228 | h = table->table; |
229 | ||
230 | for (i = 0; i < num_entries; i++) { | |
231 | cur = h + i; | |
232 | INIT_LIST_HEAD(&cur->hash_list); | |
233 | spin_lock_init(&cur->lock); | |
53b381b3 DW |
234 | } |
235 | ||
236 | x = cmpxchg(&info->stripe_hash_table, NULL, table); | |
f749303b WS |
237 | if (x) |
238 | kvfree(x); | |
53b381b3 DW |
239 | return 0; |
240 | } | |
241 | ||
4ae10b3a CM |
242 | /* |
243 | * caching an rbio means to copy anything from the | |
244 | * bio_pages array into the stripe_pages array. We | |
245 | * use the page uptodate bit in the stripe cache array | |
246 | * to indicate if it has valid data | |
247 | * | |
248 | * once the caching is done, we set the cache ready | |
249 | * bit. | |
250 | */ | |
251 | static void cache_rbio_pages(struct btrfs_raid_bio *rbio) | |
252 | { | |
253 | int i; | |
254 | char *s; | |
255 | char *d; | |
256 | int ret; | |
257 | ||
258 | ret = alloc_rbio_pages(rbio); | |
259 | if (ret) | |
260 | return; | |
261 | ||
262 | for (i = 0; i < rbio->nr_pages; i++) { | |
263 | if (!rbio->bio_pages[i]) | |
264 | continue; | |
265 | ||
266 | s = kmap(rbio->bio_pages[i]); | |
267 | d = kmap(rbio->stripe_pages[i]); | |
268 | ||
09cbfeaf | 269 | memcpy(d, s, PAGE_SIZE); |
4ae10b3a CM |
270 | |
271 | kunmap(rbio->bio_pages[i]); | |
272 | kunmap(rbio->stripe_pages[i]); | |
273 | SetPageUptodate(rbio->stripe_pages[i]); | |
274 | } | |
275 | set_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
276 | } | |
277 | ||
53b381b3 DW |
278 | /* |
279 | * we hash on the first logical address of the stripe | |
280 | */ | |
281 | static int rbio_bucket(struct btrfs_raid_bio *rbio) | |
282 | { | |
8e5cfb55 | 283 | u64 num = rbio->bbio->raid_map[0]; |
53b381b3 DW |
284 | |
285 | /* | |
286 | * we shift down quite a bit. We're using byte | |
287 | * addressing, and most of the lower bits are zeros. | |
288 | * This tends to upset hash_64, and it consistently | |
289 | * returns just one or two different values. | |
290 | * | |
291 | * shifting off the lower bits fixes things. | |
292 | */ | |
293 | return hash_64(num >> 16, BTRFS_STRIPE_HASH_TABLE_BITS); | |
294 | } | |
295 | ||
4ae10b3a CM |
296 | /* |
297 | * stealing an rbio means taking all the uptodate pages from the stripe | |
298 | * array in the source rbio and putting them into the destination rbio | |
299 | */ | |
300 | static void steal_rbio(struct btrfs_raid_bio *src, struct btrfs_raid_bio *dest) | |
301 | { | |
302 | int i; | |
303 | struct page *s; | |
304 | struct page *d; | |
305 | ||
306 | if (!test_bit(RBIO_CACHE_READY_BIT, &src->flags)) | |
307 | return; | |
308 | ||
309 | for (i = 0; i < dest->nr_pages; i++) { | |
310 | s = src->stripe_pages[i]; | |
311 | if (!s || !PageUptodate(s)) { | |
312 | continue; | |
313 | } | |
314 | ||
315 | d = dest->stripe_pages[i]; | |
316 | if (d) | |
317 | __free_page(d); | |
318 | ||
319 | dest->stripe_pages[i] = s; | |
320 | src->stripe_pages[i] = NULL; | |
321 | } | |
322 | } | |
323 | ||
53b381b3 DW |
324 | /* |
325 | * merging means we take the bio_list from the victim and | |
326 | * splice it into the destination. The victim should | |
327 | * be discarded afterwards. | |
328 | * | |
329 | * must be called with dest->rbio_list_lock held | |
330 | */ | |
331 | static void merge_rbio(struct btrfs_raid_bio *dest, | |
332 | struct btrfs_raid_bio *victim) | |
333 | { | |
334 | bio_list_merge(&dest->bio_list, &victim->bio_list); | |
335 | dest->bio_list_bytes += victim->bio_list_bytes; | |
4245215d | 336 | dest->generic_bio_cnt += victim->generic_bio_cnt; |
53b381b3 DW |
337 | bio_list_init(&victim->bio_list); |
338 | } | |
339 | ||
340 | /* | |
4ae10b3a CM |
341 | * used to prune items that are in the cache. The caller |
342 | * must hold the hash table lock. | |
343 | */ | |
344 | static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio) | |
345 | { | |
346 | int bucket = rbio_bucket(rbio); | |
347 | struct btrfs_stripe_hash_table *table; | |
348 | struct btrfs_stripe_hash *h; | |
349 | int freeit = 0; | |
350 | ||
351 | /* | |
352 | * check the bit again under the hash table lock. | |
353 | */ | |
354 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
355 | return; | |
356 | ||
357 | table = rbio->fs_info->stripe_hash_table; | |
358 | h = table->table + bucket; | |
359 | ||
360 | /* hold the lock for the bucket because we may be | |
361 | * removing it from the hash table | |
362 | */ | |
363 | spin_lock(&h->lock); | |
364 | ||
365 | /* | |
366 | * hold the lock for the bio list because we need | |
367 | * to make sure the bio list is empty | |
368 | */ | |
369 | spin_lock(&rbio->bio_list_lock); | |
370 | ||
371 | if (test_and_clear_bit(RBIO_CACHE_BIT, &rbio->flags)) { | |
372 | list_del_init(&rbio->stripe_cache); | |
373 | table->cache_size -= 1; | |
374 | freeit = 1; | |
375 | ||
376 | /* if the bio list isn't empty, this rbio is | |
377 | * still involved in an IO. We take it out | |
378 | * of the cache list, and drop the ref that | |
379 | * was held for the list. | |
380 | * | |
381 | * If the bio_list was empty, we also remove | |
382 | * the rbio from the hash_table, and drop | |
383 | * the corresponding ref | |
384 | */ | |
385 | if (bio_list_empty(&rbio->bio_list)) { | |
386 | if (!list_empty(&rbio->hash_list)) { | |
387 | list_del_init(&rbio->hash_list); | |
dec95574 | 388 | refcount_dec(&rbio->refs); |
4ae10b3a CM |
389 | BUG_ON(!list_empty(&rbio->plug_list)); |
390 | } | |
391 | } | |
392 | } | |
393 | ||
394 | spin_unlock(&rbio->bio_list_lock); | |
395 | spin_unlock(&h->lock); | |
396 | ||
397 | if (freeit) | |
398 | __free_raid_bio(rbio); | |
399 | } | |
400 | ||
401 | /* | |
402 | * prune a given rbio from the cache | |
403 | */ | |
404 | static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio) | |
405 | { | |
406 | struct btrfs_stripe_hash_table *table; | |
407 | unsigned long flags; | |
408 | ||
409 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
410 | return; | |
411 | ||
412 | table = rbio->fs_info->stripe_hash_table; | |
413 | ||
414 | spin_lock_irqsave(&table->cache_lock, flags); | |
415 | __remove_rbio_from_cache(rbio); | |
416 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
417 | } | |
418 | ||
419 | /* | |
420 | * remove everything in the cache | |
421 | */ | |
48a3b636 | 422 | static void btrfs_clear_rbio_cache(struct btrfs_fs_info *info) |
4ae10b3a CM |
423 | { |
424 | struct btrfs_stripe_hash_table *table; | |
425 | unsigned long flags; | |
426 | struct btrfs_raid_bio *rbio; | |
427 | ||
428 | table = info->stripe_hash_table; | |
429 | ||
430 | spin_lock_irqsave(&table->cache_lock, flags); | |
431 | while (!list_empty(&table->stripe_cache)) { | |
432 | rbio = list_entry(table->stripe_cache.next, | |
433 | struct btrfs_raid_bio, | |
434 | stripe_cache); | |
435 | __remove_rbio_from_cache(rbio); | |
436 | } | |
437 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
438 | } | |
439 | ||
440 | /* | |
441 | * remove all cached entries and free the hash table | |
442 | * used by unmount | |
53b381b3 DW |
443 | */ |
444 | void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info) | |
445 | { | |
446 | if (!info->stripe_hash_table) | |
447 | return; | |
4ae10b3a | 448 | btrfs_clear_rbio_cache(info); |
f749303b | 449 | kvfree(info->stripe_hash_table); |
53b381b3 DW |
450 | info->stripe_hash_table = NULL; |
451 | } | |
452 | ||
4ae10b3a CM |
453 | /* |
454 | * insert an rbio into the stripe cache. It | |
455 | * must have already been prepared by calling | |
456 | * cache_rbio_pages | |
457 | * | |
458 | * If this rbio was already cached, it gets | |
459 | * moved to the front of the lru. | |
460 | * | |
461 | * If the size of the rbio cache is too big, we | |
462 | * prune an item. | |
463 | */ | |
464 | static void cache_rbio(struct btrfs_raid_bio *rbio) | |
465 | { | |
466 | struct btrfs_stripe_hash_table *table; | |
467 | unsigned long flags; | |
468 | ||
469 | if (!test_bit(RBIO_CACHE_READY_BIT, &rbio->flags)) | |
470 | return; | |
471 | ||
472 | table = rbio->fs_info->stripe_hash_table; | |
473 | ||
474 | spin_lock_irqsave(&table->cache_lock, flags); | |
475 | spin_lock(&rbio->bio_list_lock); | |
476 | ||
477 | /* bump our ref if we were not in the list before */ | |
478 | if (!test_and_set_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
dec95574 | 479 | refcount_inc(&rbio->refs); |
4ae10b3a CM |
480 | |
481 | if (!list_empty(&rbio->stripe_cache)){ | |
482 | list_move(&rbio->stripe_cache, &table->stripe_cache); | |
483 | } else { | |
484 | list_add(&rbio->stripe_cache, &table->stripe_cache); | |
485 | table->cache_size += 1; | |
486 | } | |
487 | ||
488 | spin_unlock(&rbio->bio_list_lock); | |
489 | ||
490 | if (table->cache_size > RBIO_CACHE_SIZE) { | |
491 | struct btrfs_raid_bio *found; | |
492 | ||
493 | found = list_entry(table->stripe_cache.prev, | |
494 | struct btrfs_raid_bio, | |
495 | stripe_cache); | |
496 | ||
497 | if (found != rbio) | |
498 | __remove_rbio_from_cache(found); | |
499 | } | |
500 | ||
501 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
4ae10b3a CM |
502 | } |
503 | ||
53b381b3 DW |
504 | /* |
505 | * helper function to run the xor_blocks api. It is only | |
506 | * able to do MAX_XOR_BLOCKS at a time, so we need to | |
507 | * loop through. | |
508 | */ | |
509 | static void run_xor(void **pages, int src_cnt, ssize_t len) | |
510 | { | |
511 | int src_off = 0; | |
512 | int xor_src_cnt = 0; | |
513 | void *dest = pages[src_cnt]; | |
514 | ||
515 | while(src_cnt > 0) { | |
516 | xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS); | |
517 | xor_blocks(xor_src_cnt, len, dest, pages + src_off); | |
518 | ||
519 | src_cnt -= xor_src_cnt; | |
520 | src_off += xor_src_cnt; | |
521 | } | |
522 | } | |
523 | ||
524 | /* | |
525 | * returns true if the bio list inside this rbio | |
526 | * covers an entire stripe (no rmw required). | |
527 | * Must be called with the bio list lock held, or | |
528 | * at a time when you know it is impossible to add | |
529 | * new bios into the list | |
530 | */ | |
531 | static int __rbio_is_full(struct btrfs_raid_bio *rbio) | |
532 | { | |
533 | unsigned long size = rbio->bio_list_bytes; | |
534 | int ret = 1; | |
535 | ||
536 | if (size != rbio->nr_data * rbio->stripe_len) | |
537 | ret = 0; | |
538 | ||
539 | BUG_ON(size > rbio->nr_data * rbio->stripe_len); | |
540 | return ret; | |
541 | } | |
542 | ||
543 | static int rbio_is_full(struct btrfs_raid_bio *rbio) | |
544 | { | |
545 | unsigned long flags; | |
546 | int ret; | |
547 | ||
548 | spin_lock_irqsave(&rbio->bio_list_lock, flags); | |
549 | ret = __rbio_is_full(rbio); | |
550 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); | |
551 | return ret; | |
552 | } | |
553 | ||
554 | /* | |
555 | * returns 1 if it is safe to merge two rbios together. | |
556 | * The merging is safe if the two rbios correspond to | |
557 | * the same stripe and if they are both going in the same | |
558 | * direction (read vs write), and if neither one is | |
559 | * locked for final IO | |
560 | * | |
561 | * The caller is responsible for locking such that | |
562 | * rmw_locked is safe to test | |
563 | */ | |
564 | static int rbio_can_merge(struct btrfs_raid_bio *last, | |
565 | struct btrfs_raid_bio *cur) | |
566 | { | |
567 | if (test_bit(RBIO_RMW_LOCKED_BIT, &last->flags) || | |
568 | test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) | |
569 | return 0; | |
570 | ||
4ae10b3a CM |
571 | /* |
572 | * we can't merge with cached rbios, since the | |
573 | * idea is that when we merge the destination | |
574 | * rbio is going to run our IO for us. We can | |
01327610 | 575 | * steal from cached rbios though, other functions |
4ae10b3a CM |
576 | * handle that. |
577 | */ | |
578 | if (test_bit(RBIO_CACHE_BIT, &last->flags) || | |
579 | test_bit(RBIO_CACHE_BIT, &cur->flags)) | |
580 | return 0; | |
581 | ||
8e5cfb55 ZL |
582 | if (last->bbio->raid_map[0] != |
583 | cur->bbio->raid_map[0]) | |
53b381b3 DW |
584 | return 0; |
585 | ||
5a6ac9ea MX |
586 | /* we can't merge with different operations */ |
587 | if (last->operation != cur->operation) | |
588 | return 0; | |
589 | /* | |
590 | * We've need read the full stripe from the drive. | |
591 | * check and repair the parity and write the new results. | |
592 | * | |
593 | * We're not allowed to add any new bios to the | |
594 | * bio list here, anyone else that wants to | |
595 | * change this stripe needs to do their own rmw. | |
596 | */ | |
db34be19 | 597 | if (last->operation == BTRFS_RBIO_PARITY_SCRUB) |
53b381b3 | 598 | return 0; |
53b381b3 | 599 | |
db34be19 | 600 | if (last->operation == BTRFS_RBIO_REBUILD_MISSING) |
b4ee1782 OS |
601 | return 0; |
602 | ||
cc54ff62 LB |
603 | if (last->operation == BTRFS_RBIO_READ_REBUILD) { |
604 | int fa = last->faila; | |
605 | int fb = last->failb; | |
606 | int cur_fa = cur->faila; | |
607 | int cur_fb = cur->failb; | |
608 | ||
609 | if (last->faila >= last->failb) { | |
610 | fa = last->failb; | |
611 | fb = last->faila; | |
612 | } | |
613 | ||
614 | if (cur->faila >= cur->failb) { | |
615 | cur_fa = cur->failb; | |
616 | cur_fb = cur->faila; | |
617 | } | |
618 | ||
619 | if (fa != cur_fa || fb != cur_fb) | |
620 | return 0; | |
621 | } | |
53b381b3 DW |
622 | return 1; |
623 | } | |
624 | ||
b7178a5f ZL |
625 | static int rbio_stripe_page_index(struct btrfs_raid_bio *rbio, int stripe, |
626 | int index) | |
627 | { | |
628 | return stripe * rbio->stripe_npages + index; | |
629 | } | |
630 | ||
631 | /* | |
632 | * these are just the pages from the rbio array, not from anything | |
633 | * the FS sent down to us | |
634 | */ | |
635 | static struct page *rbio_stripe_page(struct btrfs_raid_bio *rbio, int stripe, | |
636 | int index) | |
637 | { | |
638 | return rbio->stripe_pages[rbio_stripe_page_index(rbio, stripe, index)]; | |
639 | } | |
640 | ||
53b381b3 DW |
641 | /* |
642 | * helper to index into the pstripe | |
643 | */ | |
644 | static struct page *rbio_pstripe_page(struct btrfs_raid_bio *rbio, int index) | |
645 | { | |
b7178a5f | 646 | return rbio_stripe_page(rbio, rbio->nr_data, index); |
53b381b3 DW |
647 | } |
648 | ||
649 | /* | |
650 | * helper to index into the qstripe, returns null | |
651 | * if there is no qstripe | |
652 | */ | |
653 | static struct page *rbio_qstripe_page(struct btrfs_raid_bio *rbio, int index) | |
654 | { | |
2c8cdd6e | 655 | if (rbio->nr_data + 1 == rbio->real_stripes) |
53b381b3 | 656 | return NULL; |
b7178a5f | 657 | return rbio_stripe_page(rbio, rbio->nr_data + 1, index); |
53b381b3 DW |
658 | } |
659 | ||
660 | /* | |
661 | * The first stripe in the table for a logical address | |
662 | * has the lock. rbios are added in one of three ways: | |
663 | * | |
664 | * 1) Nobody has the stripe locked yet. The rbio is given | |
665 | * the lock and 0 is returned. The caller must start the IO | |
666 | * themselves. | |
667 | * | |
668 | * 2) Someone has the stripe locked, but we're able to merge | |
669 | * with the lock owner. The rbio is freed and the IO will | |
670 | * start automatically along with the existing rbio. 1 is returned. | |
671 | * | |
672 | * 3) Someone has the stripe locked, but we're not able to merge. | |
673 | * The rbio is added to the lock owner's plug list, or merged into | |
674 | * an rbio already on the plug list. When the lock owner unlocks, | |
675 | * the next rbio on the list is run and the IO is started automatically. | |
676 | * 1 is returned | |
677 | * | |
678 | * If we return 0, the caller still owns the rbio and must continue with | |
679 | * IO submission. If we return 1, the caller must assume the rbio has | |
680 | * already been freed. | |
681 | */ | |
682 | static noinline int lock_stripe_add(struct btrfs_raid_bio *rbio) | |
683 | { | |
684 | int bucket = rbio_bucket(rbio); | |
685 | struct btrfs_stripe_hash *h = rbio->fs_info->stripe_hash_table->table + bucket; | |
686 | struct btrfs_raid_bio *cur; | |
687 | struct btrfs_raid_bio *pending; | |
688 | unsigned long flags; | |
53b381b3 | 689 | struct btrfs_raid_bio *freeit = NULL; |
4ae10b3a | 690 | struct btrfs_raid_bio *cache_drop = NULL; |
53b381b3 | 691 | int ret = 0; |
53b381b3 DW |
692 | |
693 | spin_lock_irqsave(&h->lock, flags); | |
694 | list_for_each_entry(cur, &h->hash_list, hash_list) { | |
8e5cfb55 | 695 | if (cur->bbio->raid_map[0] == rbio->bbio->raid_map[0]) { |
53b381b3 DW |
696 | spin_lock(&cur->bio_list_lock); |
697 | ||
4ae10b3a CM |
698 | /* can we steal this cached rbio's pages? */ |
699 | if (bio_list_empty(&cur->bio_list) && | |
700 | list_empty(&cur->plug_list) && | |
701 | test_bit(RBIO_CACHE_BIT, &cur->flags) && | |
702 | !test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) { | |
703 | list_del_init(&cur->hash_list); | |
dec95574 | 704 | refcount_dec(&cur->refs); |
4ae10b3a CM |
705 | |
706 | steal_rbio(cur, rbio); | |
707 | cache_drop = cur; | |
708 | spin_unlock(&cur->bio_list_lock); | |
709 | ||
710 | goto lockit; | |
711 | } | |
712 | ||
53b381b3 DW |
713 | /* can we merge into the lock owner? */ |
714 | if (rbio_can_merge(cur, rbio)) { | |
715 | merge_rbio(cur, rbio); | |
716 | spin_unlock(&cur->bio_list_lock); | |
717 | freeit = rbio; | |
718 | ret = 1; | |
719 | goto out; | |
720 | } | |
721 | ||
4ae10b3a | 722 | |
53b381b3 DW |
723 | /* |
724 | * we couldn't merge with the running | |
725 | * rbio, see if we can merge with the | |
726 | * pending ones. We don't have to | |
727 | * check for rmw_locked because there | |
728 | * is no way they are inside finish_rmw | |
729 | * right now | |
730 | */ | |
731 | list_for_each_entry(pending, &cur->plug_list, | |
732 | plug_list) { | |
733 | if (rbio_can_merge(pending, rbio)) { | |
734 | merge_rbio(pending, rbio); | |
735 | spin_unlock(&cur->bio_list_lock); | |
736 | freeit = rbio; | |
737 | ret = 1; | |
738 | goto out; | |
739 | } | |
740 | } | |
741 | ||
742 | /* no merging, put us on the tail of the plug list, | |
743 | * our rbio will be started with the currently | |
744 | * running rbio unlocks | |
745 | */ | |
746 | list_add_tail(&rbio->plug_list, &cur->plug_list); | |
747 | spin_unlock(&cur->bio_list_lock); | |
748 | ret = 1; | |
749 | goto out; | |
750 | } | |
751 | } | |
4ae10b3a | 752 | lockit: |
dec95574 | 753 | refcount_inc(&rbio->refs); |
53b381b3 DW |
754 | list_add(&rbio->hash_list, &h->hash_list); |
755 | out: | |
756 | spin_unlock_irqrestore(&h->lock, flags); | |
4ae10b3a CM |
757 | if (cache_drop) |
758 | remove_rbio_from_cache(cache_drop); | |
53b381b3 DW |
759 | if (freeit) |
760 | __free_raid_bio(freeit); | |
761 | return ret; | |
762 | } | |
763 | ||
764 | /* | |
765 | * called as rmw or parity rebuild is completed. If the plug list has more | |
766 | * rbios waiting for this stripe, the next one on the list will be started | |
767 | */ | |
768 | static noinline void unlock_stripe(struct btrfs_raid_bio *rbio) | |
769 | { | |
770 | int bucket; | |
771 | struct btrfs_stripe_hash *h; | |
772 | unsigned long flags; | |
4ae10b3a | 773 | int keep_cache = 0; |
53b381b3 DW |
774 | |
775 | bucket = rbio_bucket(rbio); | |
776 | h = rbio->fs_info->stripe_hash_table->table + bucket; | |
777 | ||
4ae10b3a CM |
778 | if (list_empty(&rbio->plug_list)) |
779 | cache_rbio(rbio); | |
780 | ||
53b381b3 DW |
781 | spin_lock_irqsave(&h->lock, flags); |
782 | spin_lock(&rbio->bio_list_lock); | |
783 | ||
784 | if (!list_empty(&rbio->hash_list)) { | |
4ae10b3a CM |
785 | /* |
786 | * if we're still cached and there is no other IO | |
787 | * to perform, just leave this rbio here for others | |
788 | * to steal from later | |
789 | */ | |
790 | if (list_empty(&rbio->plug_list) && | |
791 | test_bit(RBIO_CACHE_BIT, &rbio->flags)) { | |
792 | keep_cache = 1; | |
793 | clear_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
794 | BUG_ON(!bio_list_empty(&rbio->bio_list)); | |
795 | goto done; | |
796 | } | |
53b381b3 DW |
797 | |
798 | list_del_init(&rbio->hash_list); | |
dec95574 | 799 | refcount_dec(&rbio->refs); |
53b381b3 DW |
800 | |
801 | /* | |
802 | * we use the plug list to hold all the rbios | |
803 | * waiting for the chance to lock this stripe. | |
804 | * hand the lock over to one of them. | |
805 | */ | |
806 | if (!list_empty(&rbio->plug_list)) { | |
807 | struct btrfs_raid_bio *next; | |
808 | struct list_head *head = rbio->plug_list.next; | |
809 | ||
810 | next = list_entry(head, struct btrfs_raid_bio, | |
811 | plug_list); | |
812 | ||
813 | list_del_init(&rbio->plug_list); | |
814 | ||
815 | list_add(&next->hash_list, &h->hash_list); | |
dec95574 | 816 | refcount_inc(&next->refs); |
53b381b3 DW |
817 | spin_unlock(&rbio->bio_list_lock); |
818 | spin_unlock_irqrestore(&h->lock, flags); | |
819 | ||
1b94b556 | 820 | if (next->operation == BTRFS_RBIO_READ_REBUILD) |
53b381b3 | 821 | async_read_rebuild(next); |
b4ee1782 OS |
822 | else if (next->operation == BTRFS_RBIO_REBUILD_MISSING) { |
823 | steal_rbio(rbio, next); | |
824 | async_read_rebuild(next); | |
825 | } else if (next->operation == BTRFS_RBIO_WRITE) { | |
4ae10b3a | 826 | steal_rbio(rbio, next); |
53b381b3 | 827 | async_rmw_stripe(next); |
5a6ac9ea MX |
828 | } else if (next->operation == BTRFS_RBIO_PARITY_SCRUB) { |
829 | steal_rbio(rbio, next); | |
830 | async_scrub_parity(next); | |
4ae10b3a | 831 | } |
53b381b3 DW |
832 | |
833 | goto done_nolock; | |
53b381b3 DW |
834 | } |
835 | } | |
4ae10b3a | 836 | done: |
53b381b3 DW |
837 | spin_unlock(&rbio->bio_list_lock); |
838 | spin_unlock_irqrestore(&h->lock, flags); | |
839 | ||
840 | done_nolock: | |
4ae10b3a CM |
841 | if (!keep_cache) |
842 | remove_rbio_from_cache(rbio); | |
53b381b3 DW |
843 | } |
844 | ||
845 | static void __free_raid_bio(struct btrfs_raid_bio *rbio) | |
846 | { | |
847 | int i; | |
848 | ||
dec95574 | 849 | if (!refcount_dec_and_test(&rbio->refs)) |
53b381b3 DW |
850 | return; |
851 | ||
4ae10b3a | 852 | WARN_ON(!list_empty(&rbio->stripe_cache)); |
53b381b3 DW |
853 | WARN_ON(!list_empty(&rbio->hash_list)); |
854 | WARN_ON(!bio_list_empty(&rbio->bio_list)); | |
855 | ||
856 | for (i = 0; i < rbio->nr_pages; i++) { | |
857 | if (rbio->stripe_pages[i]) { | |
858 | __free_page(rbio->stripe_pages[i]); | |
859 | rbio->stripe_pages[i] = NULL; | |
860 | } | |
861 | } | |
af8e2d1d | 862 | |
6e9606d2 | 863 | btrfs_put_bbio(rbio->bbio); |
53b381b3 DW |
864 | kfree(rbio); |
865 | } | |
866 | ||
7583d8d0 | 867 | static void rbio_endio_bio_list(struct bio *cur, blk_status_t err) |
53b381b3 | 868 | { |
7583d8d0 LB |
869 | struct bio *next; |
870 | ||
871 | while (cur) { | |
872 | next = cur->bi_next; | |
873 | cur->bi_next = NULL; | |
874 | cur->bi_status = err; | |
875 | bio_endio(cur); | |
876 | cur = next; | |
877 | } | |
53b381b3 DW |
878 | } |
879 | ||
880 | /* | |
881 | * this frees the rbio and runs through all the bios in the | |
882 | * bio_list and calls end_io on them | |
883 | */ | |
4e4cbee9 | 884 | static void rbio_orig_end_io(struct btrfs_raid_bio *rbio, blk_status_t err) |
53b381b3 DW |
885 | { |
886 | struct bio *cur = bio_list_get(&rbio->bio_list); | |
7583d8d0 | 887 | struct bio *extra; |
4245215d MX |
888 | |
889 | if (rbio->generic_bio_cnt) | |
890 | btrfs_bio_counter_sub(rbio->fs_info, rbio->generic_bio_cnt); | |
891 | ||
7583d8d0 LB |
892 | /* |
893 | * At this moment, rbio->bio_list is empty, however since rbio does not | |
894 | * always have RBIO_RMW_LOCKED_BIT set and rbio is still linked on the | |
895 | * hash list, rbio may be merged with others so that rbio->bio_list | |
896 | * becomes non-empty. | |
897 | * Once unlock_stripe() is done, rbio->bio_list will not be updated any | |
898 | * more and we can call bio_endio() on all queued bios. | |
899 | */ | |
900 | unlock_stripe(rbio); | |
901 | extra = bio_list_get(&rbio->bio_list); | |
902 | __free_raid_bio(rbio); | |
53b381b3 | 903 | |
7583d8d0 LB |
904 | rbio_endio_bio_list(cur, err); |
905 | if (extra) | |
906 | rbio_endio_bio_list(extra, err); | |
53b381b3 DW |
907 | } |
908 | ||
909 | /* | |
910 | * end io function used by finish_rmw. When we finally | |
911 | * get here, we've written a full stripe | |
912 | */ | |
4246a0b6 | 913 | static void raid_write_end_io(struct bio *bio) |
53b381b3 DW |
914 | { |
915 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
4e4cbee9 | 916 | blk_status_t err = bio->bi_status; |
a6111d11 | 917 | int max_errors; |
53b381b3 DW |
918 | |
919 | if (err) | |
920 | fail_bio_stripe(rbio, bio); | |
921 | ||
922 | bio_put(bio); | |
923 | ||
b89e1b01 | 924 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
925 | return; |
926 | ||
58efbc9f | 927 | err = BLK_STS_OK; |
53b381b3 DW |
928 | |
929 | /* OK, we have read all the stripes we need to. */ | |
a6111d11 ZL |
930 | max_errors = (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) ? |
931 | 0 : rbio->bbio->max_errors; | |
932 | if (atomic_read(&rbio->error) > max_errors) | |
4e4cbee9 | 933 | err = BLK_STS_IOERR; |
53b381b3 | 934 | |
4246a0b6 | 935 | rbio_orig_end_io(rbio, err); |
53b381b3 DW |
936 | } |
937 | ||
938 | /* | |
939 | * the read/modify/write code wants to use the original bio for | |
940 | * any pages it included, and then use the rbio for everything | |
941 | * else. This function decides if a given index (stripe number) | |
942 | * and page number in that stripe fall inside the original bio | |
943 | * or the rbio. | |
944 | * | |
945 | * if you set bio_list_only, you'll get a NULL back for any ranges | |
946 | * that are outside the bio_list | |
947 | * | |
948 | * This doesn't take any refs on anything, you get a bare page pointer | |
949 | * and the caller must bump refs as required. | |
950 | * | |
951 | * You must call index_rbio_pages once before you can trust | |
952 | * the answers from this function. | |
953 | */ | |
954 | static struct page *page_in_rbio(struct btrfs_raid_bio *rbio, | |
955 | int index, int pagenr, int bio_list_only) | |
956 | { | |
957 | int chunk_page; | |
958 | struct page *p = NULL; | |
959 | ||
960 | chunk_page = index * (rbio->stripe_len >> PAGE_SHIFT) + pagenr; | |
961 | ||
962 | spin_lock_irq(&rbio->bio_list_lock); | |
963 | p = rbio->bio_pages[chunk_page]; | |
964 | spin_unlock_irq(&rbio->bio_list_lock); | |
965 | ||
966 | if (p || bio_list_only) | |
967 | return p; | |
968 | ||
969 | return rbio->stripe_pages[chunk_page]; | |
970 | } | |
971 | ||
972 | /* | |
973 | * number of pages we need for the entire stripe across all the | |
974 | * drives | |
975 | */ | |
976 | static unsigned long rbio_nr_pages(unsigned long stripe_len, int nr_stripes) | |
977 | { | |
09cbfeaf | 978 | return DIV_ROUND_UP(stripe_len, PAGE_SIZE) * nr_stripes; |
53b381b3 DW |
979 | } |
980 | ||
981 | /* | |
982 | * allocation and initial setup for the btrfs_raid_bio. Not | |
983 | * this does not allocate any pages for rbio->pages. | |
984 | */ | |
2ff7e61e JM |
985 | static struct btrfs_raid_bio *alloc_rbio(struct btrfs_fs_info *fs_info, |
986 | struct btrfs_bio *bbio, | |
987 | u64 stripe_len) | |
53b381b3 DW |
988 | { |
989 | struct btrfs_raid_bio *rbio; | |
990 | int nr_data = 0; | |
2c8cdd6e MX |
991 | int real_stripes = bbio->num_stripes - bbio->num_tgtdevs; |
992 | int num_pages = rbio_nr_pages(stripe_len, real_stripes); | |
5a6ac9ea | 993 | int stripe_npages = DIV_ROUND_UP(stripe_len, PAGE_SIZE); |
53b381b3 DW |
994 | void *p; |
995 | ||
5a6ac9ea | 996 | rbio = kzalloc(sizeof(*rbio) + num_pages * sizeof(struct page *) * 2 + |
bfca9a6d ZL |
997 | DIV_ROUND_UP(stripe_npages, BITS_PER_LONG) * |
998 | sizeof(long), GFP_NOFS); | |
af8e2d1d | 999 | if (!rbio) |
53b381b3 | 1000 | return ERR_PTR(-ENOMEM); |
53b381b3 DW |
1001 | |
1002 | bio_list_init(&rbio->bio_list); | |
1003 | INIT_LIST_HEAD(&rbio->plug_list); | |
1004 | spin_lock_init(&rbio->bio_list_lock); | |
4ae10b3a | 1005 | INIT_LIST_HEAD(&rbio->stripe_cache); |
53b381b3 DW |
1006 | INIT_LIST_HEAD(&rbio->hash_list); |
1007 | rbio->bbio = bbio; | |
2ff7e61e | 1008 | rbio->fs_info = fs_info; |
53b381b3 DW |
1009 | rbio->stripe_len = stripe_len; |
1010 | rbio->nr_pages = num_pages; | |
2c8cdd6e | 1011 | rbio->real_stripes = real_stripes; |
5a6ac9ea | 1012 | rbio->stripe_npages = stripe_npages; |
53b381b3 DW |
1013 | rbio->faila = -1; |
1014 | rbio->failb = -1; | |
dec95574 | 1015 | refcount_set(&rbio->refs, 1); |
b89e1b01 MX |
1016 | atomic_set(&rbio->error, 0); |
1017 | atomic_set(&rbio->stripes_pending, 0); | |
53b381b3 DW |
1018 | |
1019 | /* | |
1020 | * the stripe_pages and bio_pages array point to the extra | |
1021 | * memory we allocated past the end of the rbio | |
1022 | */ | |
1023 | p = rbio + 1; | |
1024 | rbio->stripe_pages = p; | |
1025 | rbio->bio_pages = p + sizeof(struct page *) * num_pages; | |
5a6ac9ea | 1026 | rbio->dbitmap = p + sizeof(struct page *) * num_pages * 2; |
53b381b3 | 1027 | |
10f11900 ZL |
1028 | if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5) |
1029 | nr_data = real_stripes - 1; | |
1030 | else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6) | |
2c8cdd6e | 1031 | nr_data = real_stripes - 2; |
53b381b3 | 1032 | else |
10f11900 | 1033 | BUG(); |
53b381b3 DW |
1034 | |
1035 | rbio->nr_data = nr_data; | |
1036 | return rbio; | |
1037 | } | |
1038 | ||
1039 | /* allocate pages for all the stripes in the bio, including parity */ | |
1040 | static int alloc_rbio_pages(struct btrfs_raid_bio *rbio) | |
1041 | { | |
1042 | int i; | |
1043 | struct page *page; | |
1044 | ||
1045 | for (i = 0; i < rbio->nr_pages; i++) { | |
1046 | if (rbio->stripe_pages[i]) | |
1047 | continue; | |
1048 | page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
1049 | if (!page) | |
1050 | return -ENOMEM; | |
1051 | rbio->stripe_pages[i] = page; | |
53b381b3 DW |
1052 | } |
1053 | return 0; | |
1054 | } | |
1055 | ||
b7178a5f | 1056 | /* only allocate pages for p/q stripes */ |
53b381b3 DW |
1057 | static int alloc_rbio_parity_pages(struct btrfs_raid_bio *rbio) |
1058 | { | |
1059 | int i; | |
1060 | struct page *page; | |
1061 | ||
b7178a5f | 1062 | i = rbio_stripe_page_index(rbio, rbio->nr_data, 0); |
53b381b3 DW |
1063 | |
1064 | for (; i < rbio->nr_pages; i++) { | |
1065 | if (rbio->stripe_pages[i]) | |
1066 | continue; | |
1067 | page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
1068 | if (!page) | |
1069 | return -ENOMEM; | |
1070 | rbio->stripe_pages[i] = page; | |
1071 | } | |
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * add a single page from a specific stripe into our list of bios for IO | |
1077 | * this will try to merge into existing bios if possible, and returns | |
1078 | * zero if all went well. | |
1079 | */ | |
48a3b636 ES |
1080 | static int rbio_add_io_page(struct btrfs_raid_bio *rbio, |
1081 | struct bio_list *bio_list, | |
1082 | struct page *page, | |
1083 | int stripe_nr, | |
1084 | unsigned long page_index, | |
1085 | unsigned long bio_max_len) | |
53b381b3 DW |
1086 | { |
1087 | struct bio *last = bio_list->tail; | |
1088 | u64 last_end = 0; | |
1089 | int ret; | |
1090 | struct bio *bio; | |
1091 | struct btrfs_bio_stripe *stripe; | |
1092 | u64 disk_start; | |
1093 | ||
1094 | stripe = &rbio->bbio->stripes[stripe_nr]; | |
09cbfeaf | 1095 | disk_start = stripe->physical + (page_index << PAGE_SHIFT); |
53b381b3 DW |
1096 | |
1097 | /* if the device is missing, just fail this stripe */ | |
1098 | if (!stripe->dev->bdev) | |
1099 | return fail_rbio_index(rbio, stripe_nr); | |
1100 | ||
1101 | /* see if we can add this page onto our existing bio */ | |
1102 | if (last) { | |
4f024f37 KO |
1103 | last_end = (u64)last->bi_iter.bi_sector << 9; |
1104 | last_end += last->bi_iter.bi_size; | |
53b381b3 DW |
1105 | |
1106 | /* | |
1107 | * we can't merge these if they are from different | |
1108 | * devices or if they are not contiguous | |
1109 | */ | |
1110 | if (last_end == disk_start && stripe->dev->bdev && | |
4e4cbee9 | 1111 | !last->bi_status && |
74d46992 CH |
1112 | last->bi_disk == stripe->dev->bdev->bd_disk && |
1113 | last->bi_partno == stripe->dev->bdev->bd_partno) { | |
09cbfeaf KS |
1114 | ret = bio_add_page(last, page, PAGE_SIZE, 0); |
1115 | if (ret == PAGE_SIZE) | |
53b381b3 DW |
1116 | return 0; |
1117 | } | |
1118 | } | |
1119 | ||
1120 | /* put a new bio on the list */ | |
c5e4c3d7 | 1121 | bio = btrfs_io_bio_alloc(bio_max_len >> PAGE_SHIFT ?: 1); |
4f024f37 | 1122 | bio->bi_iter.bi_size = 0; |
74d46992 | 1123 | bio_set_dev(bio, stripe->dev->bdev); |
4f024f37 | 1124 | bio->bi_iter.bi_sector = disk_start >> 9; |
53b381b3 | 1125 | |
09cbfeaf | 1126 | bio_add_page(bio, page, PAGE_SIZE, 0); |
53b381b3 DW |
1127 | bio_list_add(bio_list, bio); |
1128 | return 0; | |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * while we're doing the read/modify/write cycle, we could | |
1133 | * have errors in reading pages off the disk. This checks | |
1134 | * for errors and if we're not able to read the page it'll | |
1135 | * trigger parity reconstruction. The rmw will be finished | |
1136 | * after we've reconstructed the failed stripes | |
1137 | */ | |
1138 | static void validate_rbio_for_rmw(struct btrfs_raid_bio *rbio) | |
1139 | { | |
1140 | if (rbio->faila >= 0 || rbio->failb >= 0) { | |
2c8cdd6e | 1141 | BUG_ON(rbio->faila == rbio->real_stripes - 1); |
53b381b3 DW |
1142 | __raid56_parity_recover(rbio); |
1143 | } else { | |
1144 | finish_rmw(rbio); | |
1145 | } | |
1146 | } | |
1147 | ||
53b381b3 DW |
1148 | /* |
1149 | * helper function to walk our bio list and populate the bio_pages array with | |
1150 | * the result. This seems expensive, but it is faster than constantly | |
1151 | * searching through the bio list as we setup the IO in finish_rmw or stripe | |
1152 | * reconstruction. | |
1153 | * | |
1154 | * This must be called before you trust the answers from page_in_rbio | |
1155 | */ | |
1156 | static void index_rbio_pages(struct btrfs_raid_bio *rbio) | |
1157 | { | |
1158 | struct bio *bio; | |
1159 | u64 start; | |
1160 | unsigned long stripe_offset; | |
1161 | unsigned long page_index; | |
53b381b3 DW |
1162 | |
1163 | spin_lock_irq(&rbio->bio_list_lock); | |
1164 | bio_list_for_each(bio, &rbio->bio_list) { | |
6592e58c FM |
1165 | struct bio_vec bvec; |
1166 | struct bvec_iter iter; | |
1167 | int i = 0; | |
1168 | ||
4f024f37 | 1169 | start = (u64)bio->bi_iter.bi_sector << 9; |
8e5cfb55 | 1170 | stripe_offset = start - rbio->bbio->raid_map[0]; |
09cbfeaf | 1171 | page_index = stripe_offset >> PAGE_SHIFT; |
53b381b3 | 1172 | |
6592e58c FM |
1173 | if (bio_flagged(bio, BIO_CLONED)) |
1174 | bio->bi_iter = btrfs_io_bio(bio)->iter; | |
1175 | ||
1176 | bio_for_each_segment(bvec, bio, iter) { | |
1177 | rbio->bio_pages[page_index + i] = bvec.bv_page; | |
1178 | i++; | |
1179 | } | |
53b381b3 DW |
1180 | } |
1181 | spin_unlock_irq(&rbio->bio_list_lock); | |
1182 | } | |
1183 | ||
1184 | /* | |
1185 | * this is called from one of two situations. We either | |
1186 | * have a full stripe from the higher layers, or we've read all | |
1187 | * the missing bits off disk. | |
1188 | * | |
1189 | * This will calculate the parity and then send down any | |
1190 | * changed blocks. | |
1191 | */ | |
1192 | static noinline void finish_rmw(struct btrfs_raid_bio *rbio) | |
1193 | { | |
1194 | struct btrfs_bio *bbio = rbio->bbio; | |
2c8cdd6e | 1195 | void *pointers[rbio->real_stripes]; |
53b381b3 DW |
1196 | int nr_data = rbio->nr_data; |
1197 | int stripe; | |
1198 | int pagenr; | |
1199 | int p_stripe = -1; | |
1200 | int q_stripe = -1; | |
1201 | struct bio_list bio_list; | |
1202 | struct bio *bio; | |
53b381b3 DW |
1203 | int ret; |
1204 | ||
1205 | bio_list_init(&bio_list); | |
1206 | ||
2c8cdd6e MX |
1207 | if (rbio->real_stripes - rbio->nr_data == 1) { |
1208 | p_stripe = rbio->real_stripes - 1; | |
1209 | } else if (rbio->real_stripes - rbio->nr_data == 2) { | |
1210 | p_stripe = rbio->real_stripes - 2; | |
1211 | q_stripe = rbio->real_stripes - 1; | |
53b381b3 DW |
1212 | } else { |
1213 | BUG(); | |
1214 | } | |
1215 | ||
1216 | /* at this point we either have a full stripe, | |
1217 | * or we've read the full stripe from the drive. | |
1218 | * recalculate the parity and write the new results. | |
1219 | * | |
1220 | * We're not allowed to add any new bios to the | |
1221 | * bio list here, anyone else that wants to | |
1222 | * change this stripe needs to do their own rmw. | |
1223 | */ | |
1224 | spin_lock_irq(&rbio->bio_list_lock); | |
1225 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
1226 | spin_unlock_irq(&rbio->bio_list_lock); | |
1227 | ||
b89e1b01 | 1228 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
1229 | |
1230 | /* | |
1231 | * now that we've set rmw_locked, run through the | |
1232 | * bio list one last time and map the page pointers | |
4ae10b3a CM |
1233 | * |
1234 | * We don't cache full rbios because we're assuming | |
1235 | * the higher layers are unlikely to use this area of | |
1236 | * the disk again soon. If they do use it again, | |
1237 | * hopefully they will send another full bio. | |
53b381b3 DW |
1238 | */ |
1239 | index_rbio_pages(rbio); | |
4ae10b3a CM |
1240 | if (!rbio_is_full(rbio)) |
1241 | cache_rbio_pages(rbio); | |
1242 | else | |
1243 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
53b381b3 | 1244 | |
915e2290 | 1245 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
1246 | struct page *p; |
1247 | /* first collect one page from each data stripe */ | |
1248 | for (stripe = 0; stripe < nr_data; stripe++) { | |
1249 | p = page_in_rbio(rbio, stripe, pagenr, 0); | |
1250 | pointers[stripe] = kmap(p); | |
1251 | } | |
1252 | ||
1253 | /* then add the parity stripe */ | |
1254 | p = rbio_pstripe_page(rbio, pagenr); | |
1255 | SetPageUptodate(p); | |
1256 | pointers[stripe++] = kmap(p); | |
1257 | ||
1258 | if (q_stripe != -1) { | |
1259 | ||
1260 | /* | |
1261 | * raid6, add the qstripe and call the | |
1262 | * library function to fill in our p/q | |
1263 | */ | |
1264 | p = rbio_qstripe_page(rbio, pagenr); | |
1265 | SetPageUptodate(p); | |
1266 | pointers[stripe++] = kmap(p); | |
1267 | ||
2c8cdd6e | 1268 | raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE, |
53b381b3 DW |
1269 | pointers); |
1270 | } else { | |
1271 | /* raid5 */ | |
1272 | memcpy(pointers[nr_data], pointers[0], PAGE_SIZE); | |
09cbfeaf | 1273 | run_xor(pointers + 1, nr_data - 1, PAGE_SIZE); |
53b381b3 DW |
1274 | } |
1275 | ||
1276 | ||
2c8cdd6e | 1277 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) |
53b381b3 DW |
1278 | kunmap(page_in_rbio(rbio, stripe, pagenr, 0)); |
1279 | } | |
1280 | ||
1281 | /* | |
1282 | * time to start writing. Make bios for everything from the | |
1283 | * higher layers (the bio_list in our rbio) and our p/q. Ignore | |
1284 | * everything else. | |
1285 | */ | |
2c8cdd6e | 1286 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
915e2290 | 1287 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
1288 | struct page *page; |
1289 | if (stripe < rbio->nr_data) { | |
1290 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
1291 | if (!page) | |
1292 | continue; | |
1293 | } else { | |
1294 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
1295 | } | |
1296 | ||
1297 | ret = rbio_add_io_page(rbio, &bio_list, | |
1298 | page, stripe, pagenr, rbio->stripe_len); | |
1299 | if (ret) | |
1300 | goto cleanup; | |
1301 | } | |
1302 | } | |
1303 | ||
2c8cdd6e MX |
1304 | if (likely(!bbio->num_tgtdevs)) |
1305 | goto write_data; | |
1306 | ||
1307 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { | |
1308 | if (!bbio->tgtdev_map[stripe]) | |
1309 | continue; | |
1310 | ||
915e2290 | 1311 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
2c8cdd6e MX |
1312 | struct page *page; |
1313 | if (stripe < rbio->nr_data) { | |
1314 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
1315 | if (!page) | |
1316 | continue; | |
1317 | } else { | |
1318 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
1319 | } | |
1320 | ||
1321 | ret = rbio_add_io_page(rbio, &bio_list, page, | |
1322 | rbio->bbio->tgtdev_map[stripe], | |
1323 | pagenr, rbio->stripe_len); | |
1324 | if (ret) | |
1325 | goto cleanup; | |
1326 | } | |
1327 | } | |
1328 | ||
1329 | write_data: | |
b89e1b01 MX |
1330 | atomic_set(&rbio->stripes_pending, bio_list_size(&bio_list)); |
1331 | BUG_ON(atomic_read(&rbio->stripes_pending) == 0); | |
53b381b3 DW |
1332 | |
1333 | while (1) { | |
1334 | bio = bio_list_pop(&bio_list); | |
1335 | if (!bio) | |
1336 | break; | |
1337 | ||
1338 | bio->bi_private = rbio; | |
1339 | bio->bi_end_io = raid_write_end_io; | |
37226b21 | 1340 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4e49ea4a MC |
1341 | |
1342 | submit_bio(bio); | |
53b381b3 DW |
1343 | } |
1344 | return; | |
1345 | ||
1346 | cleanup: | |
58efbc9f | 1347 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
1348 | |
1349 | while ((bio = bio_list_pop(&bio_list))) | |
1350 | bio_put(bio); | |
53b381b3 DW |
1351 | } |
1352 | ||
1353 | /* | |
1354 | * helper to find the stripe number for a given bio. Used to figure out which | |
1355 | * stripe has failed. This expects the bio to correspond to a physical disk, | |
1356 | * so it looks up based on physical sector numbers. | |
1357 | */ | |
1358 | static int find_bio_stripe(struct btrfs_raid_bio *rbio, | |
1359 | struct bio *bio) | |
1360 | { | |
4f024f37 | 1361 | u64 physical = bio->bi_iter.bi_sector; |
53b381b3 DW |
1362 | u64 stripe_start; |
1363 | int i; | |
1364 | struct btrfs_bio_stripe *stripe; | |
1365 | ||
1366 | physical <<= 9; | |
1367 | ||
1368 | for (i = 0; i < rbio->bbio->num_stripes; i++) { | |
1369 | stripe = &rbio->bbio->stripes[i]; | |
1370 | stripe_start = stripe->physical; | |
1371 | if (physical >= stripe_start && | |
2c8cdd6e | 1372 | physical < stripe_start + rbio->stripe_len && |
047fdea6 | 1373 | stripe->dev->bdev && |
74d46992 CH |
1374 | bio->bi_disk == stripe->dev->bdev->bd_disk && |
1375 | bio->bi_partno == stripe->dev->bdev->bd_partno) { | |
53b381b3 DW |
1376 | return i; |
1377 | } | |
1378 | } | |
1379 | return -1; | |
1380 | } | |
1381 | ||
1382 | /* | |
1383 | * helper to find the stripe number for a given | |
1384 | * bio (before mapping). Used to figure out which stripe has | |
1385 | * failed. This looks up based on logical block numbers. | |
1386 | */ | |
1387 | static int find_logical_bio_stripe(struct btrfs_raid_bio *rbio, | |
1388 | struct bio *bio) | |
1389 | { | |
4f024f37 | 1390 | u64 logical = bio->bi_iter.bi_sector; |
53b381b3 DW |
1391 | u64 stripe_start; |
1392 | int i; | |
1393 | ||
1394 | logical <<= 9; | |
1395 | ||
1396 | for (i = 0; i < rbio->nr_data; i++) { | |
8e5cfb55 | 1397 | stripe_start = rbio->bbio->raid_map[i]; |
53b381b3 DW |
1398 | if (logical >= stripe_start && |
1399 | logical < stripe_start + rbio->stripe_len) { | |
1400 | return i; | |
1401 | } | |
1402 | } | |
1403 | return -1; | |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * returns -EIO if we had too many failures | |
1408 | */ | |
1409 | static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed) | |
1410 | { | |
1411 | unsigned long flags; | |
1412 | int ret = 0; | |
1413 | ||
1414 | spin_lock_irqsave(&rbio->bio_list_lock, flags); | |
1415 | ||
1416 | /* we already know this stripe is bad, move on */ | |
1417 | if (rbio->faila == failed || rbio->failb == failed) | |
1418 | goto out; | |
1419 | ||
1420 | if (rbio->faila == -1) { | |
1421 | /* first failure on this rbio */ | |
1422 | rbio->faila = failed; | |
b89e1b01 | 1423 | atomic_inc(&rbio->error); |
53b381b3 DW |
1424 | } else if (rbio->failb == -1) { |
1425 | /* second failure on this rbio */ | |
1426 | rbio->failb = failed; | |
b89e1b01 | 1427 | atomic_inc(&rbio->error); |
53b381b3 DW |
1428 | } else { |
1429 | ret = -EIO; | |
1430 | } | |
1431 | out: | |
1432 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); | |
1433 | ||
1434 | return ret; | |
1435 | } | |
1436 | ||
1437 | /* | |
1438 | * helper to fail a stripe based on a physical disk | |
1439 | * bio. | |
1440 | */ | |
1441 | static int fail_bio_stripe(struct btrfs_raid_bio *rbio, | |
1442 | struct bio *bio) | |
1443 | { | |
1444 | int failed = find_bio_stripe(rbio, bio); | |
1445 | ||
1446 | if (failed < 0) | |
1447 | return -EIO; | |
1448 | ||
1449 | return fail_rbio_index(rbio, failed); | |
1450 | } | |
1451 | ||
1452 | /* | |
1453 | * this sets each page in the bio uptodate. It should only be used on private | |
1454 | * rbio pages, nothing that comes in from the higher layers | |
1455 | */ | |
1456 | static void set_bio_pages_uptodate(struct bio *bio) | |
1457 | { | |
0198e5b7 LB |
1458 | struct bio_vec *bvec; |
1459 | int i; | |
6592e58c | 1460 | |
0198e5b7 | 1461 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
53b381b3 | 1462 | |
0198e5b7 LB |
1463 | bio_for_each_segment_all(bvec, bio, i) |
1464 | SetPageUptodate(bvec->bv_page); | |
53b381b3 DW |
1465 | } |
1466 | ||
1467 | /* | |
1468 | * end io for the read phase of the rmw cycle. All the bios here are physical | |
1469 | * stripe bios we've read from the disk so we can recalculate the parity of the | |
1470 | * stripe. | |
1471 | * | |
1472 | * This will usually kick off finish_rmw once all the bios are read in, but it | |
1473 | * may trigger parity reconstruction if we had any errors along the way | |
1474 | */ | |
4246a0b6 | 1475 | static void raid_rmw_end_io(struct bio *bio) |
53b381b3 DW |
1476 | { |
1477 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
1478 | ||
4e4cbee9 | 1479 | if (bio->bi_status) |
53b381b3 DW |
1480 | fail_bio_stripe(rbio, bio); |
1481 | else | |
1482 | set_bio_pages_uptodate(bio); | |
1483 | ||
1484 | bio_put(bio); | |
1485 | ||
b89e1b01 | 1486 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
1487 | return; |
1488 | ||
b89e1b01 | 1489 | if (atomic_read(&rbio->error) > rbio->bbio->max_errors) |
53b381b3 DW |
1490 | goto cleanup; |
1491 | ||
1492 | /* | |
1493 | * this will normally call finish_rmw to start our write | |
1494 | * but if there are any failed stripes we'll reconstruct | |
1495 | * from parity first | |
1496 | */ | |
1497 | validate_rbio_for_rmw(rbio); | |
1498 | return; | |
1499 | ||
1500 | cleanup: | |
1501 | ||
58efbc9f | 1502 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
53b381b3 DW |
1503 | } |
1504 | ||
1505 | static void async_rmw_stripe(struct btrfs_raid_bio *rbio) | |
1506 | { | |
0b246afa JM |
1507 | btrfs_init_work(&rbio->work, btrfs_rmw_helper, rmw_work, NULL, NULL); |
1508 | btrfs_queue_work(rbio->fs_info->rmw_workers, &rbio->work); | |
53b381b3 DW |
1509 | } |
1510 | ||
1511 | static void async_read_rebuild(struct btrfs_raid_bio *rbio) | |
1512 | { | |
9e0af237 LB |
1513 | btrfs_init_work(&rbio->work, btrfs_rmw_helper, |
1514 | read_rebuild_work, NULL, NULL); | |
53b381b3 | 1515 | |
0b246afa | 1516 | btrfs_queue_work(rbio->fs_info->rmw_workers, &rbio->work); |
53b381b3 DW |
1517 | } |
1518 | ||
1519 | /* | |
1520 | * the stripe must be locked by the caller. It will | |
1521 | * unlock after all the writes are done | |
1522 | */ | |
1523 | static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio) | |
1524 | { | |
1525 | int bios_to_read = 0; | |
53b381b3 DW |
1526 | struct bio_list bio_list; |
1527 | int ret; | |
53b381b3 DW |
1528 | int pagenr; |
1529 | int stripe; | |
1530 | struct bio *bio; | |
1531 | ||
1532 | bio_list_init(&bio_list); | |
1533 | ||
1534 | ret = alloc_rbio_pages(rbio); | |
1535 | if (ret) | |
1536 | goto cleanup; | |
1537 | ||
1538 | index_rbio_pages(rbio); | |
1539 | ||
b89e1b01 | 1540 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
1541 | /* |
1542 | * build a list of bios to read all the missing parts of this | |
1543 | * stripe | |
1544 | */ | |
1545 | for (stripe = 0; stripe < rbio->nr_data; stripe++) { | |
915e2290 | 1546 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
1547 | struct page *page; |
1548 | /* | |
1549 | * we want to find all the pages missing from | |
1550 | * the rbio and read them from the disk. If | |
1551 | * page_in_rbio finds a page in the bio list | |
1552 | * we don't need to read it off the stripe. | |
1553 | */ | |
1554 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
1555 | if (page) | |
1556 | continue; | |
1557 | ||
1558 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
4ae10b3a CM |
1559 | /* |
1560 | * the bio cache may have handed us an uptodate | |
1561 | * page. If so, be happy and use it | |
1562 | */ | |
1563 | if (PageUptodate(page)) | |
1564 | continue; | |
1565 | ||
53b381b3 DW |
1566 | ret = rbio_add_io_page(rbio, &bio_list, page, |
1567 | stripe, pagenr, rbio->stripe_len); | |
1568 | if (ret) | |
1569 | goto cleanup; | |
1570 | } | |
1571 | } | |
1572 | ||
1573 | bios_to_read = bio_list_size(&bio_list); | |
1574 | if (!bios_to_read) { | |
1575 | /* | |
1576 | * this can happen if others have merged with | |
1577 | * us, it means there is nothing left to read. | |
1578 | * But if there are missing devices it may not be | |
1579 | * safe to do the full stripe write yet. | |
1580 | */ | |
1581 | goto finish; | |
1582 | } | |
1583 | ||
1584 | /* | |
1585 | * the bbio may be freed once we submit the last bio. Make sure | |
1586 | * not to touch it after that | |
1587 | */ | |
b89e1b01 | 1588 | atomic_set(&rbio->stripes_pending, bios_to_read); |
53b381b3 DW |
1589 | while (1) { |
1590 | bio = bio_list_pop(&bio_list); | |
1591 | if (!bio) | |
1592 | break; | |
1593 | ||
1594 | bio->bi_private = rbio; | |
1595 | bio->bi_end_io = raid_rmw_end_io; | |
37226b21 | 1596 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
53b381b3 | 1597 | |
0b246afa | 1598 | btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
53b381b3 | 1599 | |
4e49ea4a | 1600 | submit_bio(bio); |
53b381b3 DW |
1601 | } |
1602 | /* the actual write will happen once the reads are done */ | |
1603 | return 0; | |
1604 | ||
1605 | cleanup: | |
58efbc9f | 1606 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
1607 | |
1608 | while ((bio = bio_list_pop(&bio_list))) | |
1609 | bio_put(bio); | |
1610 | ||
53b381b3 DW |
1611 | return -EIO; |
1612 | ||
1613 | finish: | |
1614 | validate_rbio_for_rmw(rbio); | |
1615 | return 0; | |
1616 | } | |
1617 | ||
1618 | /* | |
1619 | * if the upper layers pass in a full stripe, we thank them by only allocating | |
1620 | * enough pages to hold the parity, and sending it all down quickly. | |
1621 | */ | |
1622 | static int full_stripe_write(struct btrfs_raid_bio *rbio) | |
1623 | { | |
1624 | int ret; | |
1625 | ||
1626 | ret = alloc_rbio_parity_pages(rbio); | |
3cd846d1 MX |
1627 | if (ret) { |
1628 | __free_raid_bio(rbio); | |
53b381b3 | 1629 | return ret; |
3cd846d1 | 1630 | } |
53b381b3 DW |
1631 | |
1632 | ret = lock_stripe_add(rbio); | |
1633 | if (ret == 0) | |
1634 | finish_rmw(rbio); | |
1635 | return 0; | |
1636 | } | |
1637 | ||
1638 | /* | |
1639 | * partial stripe writes get handed over to async helpers. | |
1640 | * We're really hoping to merge a few more writes into this | |
1641 | * rbio before calculating new parity | |
1642 | */ | |
1643 | static int partial_stripe_write(struct btrfs_raid_bio *rbio) | |
1644 | { | |
1645 | int ret; | |
1646 | ||
1647 | ret = lock_stripe_add(rbio); | |
1648 | if (ret == 0) | |
1649 | async_rmw_stripe(rbio); | |
1650 | return 0; | |
1651 | } | |
1652 | ||
1653 | /* | |
1654 | * sometimes while we were reading from the drive to | |
1655 | * recalculate parity, enough new bios come into create | |
1656 | * a full stripe. So we do a check here to see if we can | |
1657 | * go directly to finish_rmw | |
1658 | */ | |
1659 | static int __raid56_parity_write(struct btrfs_raid_bio *rbio) | |
1660 | { | |
1661 | /* head off into rmw land if we don't have a full stripe */ | |
1662 | if (!rbio_is_full(rbio)) | |
1663 | return partial_stripe_write(rbio); | |
1664 | return full_stripe_write(rbio); | |
1665 | } | |
1666 | ||
6ac0f488 CM |
1667 | /* |
1668 | * We use plugging call backs to collect full stripes. | |
1669 | * Any time we get a partial stripe write while plugged | |
1670 | * we collect it into a list. When the unplug comes down, | |
1671 | * we sort the list by logical block number and merge | |
1672 | * everything we can into the same rbios | |
1673 | */ | |
1674 | struct btrfs_plug_cb { | |
1675 | struct blk_plug_cb cb; | |
1676 | struct btrfs_fs_info *info; | |
1677 | struct list_head rbio_list; | |
1678 | struct btrfs_work work; | |
1679 | }; | |
1680 | ||
1681 | /* | |
1682 | * rbios on the plug list are sorted for easier merging. | |
1683 | */ | |
1684 | static int plug_cmp(void *priv, struct list_head *a, struct list_head *b) | |
1685 | { | |
1686 | struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio, | |
1687 | plug_list); | |
1688 | struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio, | |
1689 | plug_list); | |
4f024f37 KO |
1690 | u64 a_sector = ra->bio_list.head->bi_iter.bi_sector; |
1691 | u64 b_sector = rb->bio_list.head->bi_iter.bi_sector; | |
6ac0f488 CM |
1692 | |
1693 | if (a_sector < b_sector) | |
1694 | return -1; | |
1695 | if (a_sector > b_sector) | |
1696 | return 1; | |
1697 | return 0; | |
1698 | } | |
1699 | ||
1700 | static void run_plug(struct btrfs_plug_cb *plug) | |
1701 | { | |
1702 | struct btrfs_raid_bio *cur; | |
1703 | struct btrfs_raid_bio *last = NULL; | |
1704 | ||
1705 | /* | |
1706 | * sort our plug list then try to merge | |
1707 | * everything we can in hopes of creating full | |
1708 | * stripes. | |
1709 | */ | |
1710 | list_sort(NULL, &plug->rbio_list, plug_cmp); | |
1711 | while (!list_empty(&plug->rbio_list)) { | |
1712 | cur = list_entry(plug->rbio_list.next, | |
1713 | struct btrfs_raid_bio, plug_list); | |
1714 | list_del_init(&cur->plug_list); | |
1715 | ||
1716 | if (rbio_is_full(cur)) { | |
1717 | /* we have a full stripe, send it down */ | |
1718 | full_stripe_write(cur); | |
1719 | continue; | |
1720 | } | |
1721 | if (last) { | |
1722 | if (rbio_can_merge(last, cur)) { | |
1723 | merge_rbio(last, cur); | |
1724 | __free_raid_bio(cur); | |
1725 | continue; | |
1726 | ||
1727 | } | |
1728 | __raid56_parity_write(last); | |
1729 | } | |
1730 | last = cur; | |
1731 | } | |
1732 | if (last) { | |
1733 | __raid56_parity_write(last); | |
1734 | } | |
1735 | kfree(plug); | |
1736 | } | |
1737 | ||
1738 | /* | |
1739 | * if the unplug comes from schedule, we have to push the | |
1740 | * work off to a helper thread | |
1741 | */ | |
1742 | static void unplug_work(struct btrfs_work *work) | |
1743 | { | |
1744 | struct btrfs_plug_cb *plug; | |
1745 | plug = container_of(work, struct btrfs_plug_cb, work); | |
1746 | run_plug(plug); | |
1747 | } | |
1748 | ||
1749 | static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1750 | { | |
1751 | struct btrfs_plug_cb *plug; | |
1752 | plug = container_of(cb, struct btrfs_plug_cb, cb); | |
1753 | ||
1754 | if (from_schedule) { | |
9e0af237 LB |
1755 | btrfs_init_work(&plug->work, btrfs_rmw_helper, |
1756 | unplug_work, NULL, NULL); | |
d05a33ac QW |
1757 | btrfs_queue_work(plug->info->rmw_workers, |
1758 | &plug->work); | |
6ac0f488 CM |
1759 | return; |
1760 | } | |
1761 | run_plug(plug); | |
1762 | } | |
1763 | ||
53b381b3 DW |
1764 | /* |
1765 | * our main entry point for writes from the rest of the FS. | |
1766 | */ | |
2ff7e61e | 1767 | int raid56_parity_write(struct btrfs_fs_info *fs_info, struct bio *bio, |
8e5cfb55 | 1768 | struct btrfs_bio *bbio, u64 stripe_len) |
53b381b3 DW |
1769 | { |
1770 | struct btrfs_raid_bio *rbio; | |
6ac0f488 CM |
1771 | struct btrfs_plug_cb *plug = NULL; |
1772 | struct blk_plug_cb *cb; | |
4245215d | 1773 | int ret; |
53b381b3 | 1774 | |
2ff7e61e | 1775 | rbio = alloc_rbio(fs_info, bbio, stripe_len); |
af8e2d1d | 1776 | if (IS_ERR(rbio)) { |
6e9606d2 | 1777 | btrfs_put_bbio(bbio); |
53b381b3 | 1778 | return PTR_ERR(rbio); |
af8e2d1d | 1779 | } |
53b381b3 | 1780 | bio_list_add(&rbio->bio_list, bio); |
4f024f37 | 1781 | rbio->bio_list_bytes = bio->bi_iter.bi_size; |
1b94b556 | 1782 | rbio->operation = BTRFS_RBIO_WRITE; |
6ac0f488 | 1783 | |
0b246afa | 1784 | btrfs_bio_counter_inc_noblocked(fs_info); |
4245215d MX |
1785 | rbio->generic_bio_cnt = 1; |
1786 | ||
6ac0f488 CM |
1787 | /* |
1788 | * don't plug on full rbios, just get them out the door | |
1789 | * as quickly as we can | |
1790 | */ | |
4245215d MX |
1791 | if (rbio_is_full(rbio)) { |
1792 | ret = full_stripe_write(rbio); | |
1793 | if (ret) | |
0b246afa | 1794 | btrfs_bio_counter_dec(fs_info); |
4245215d MX |
1795 | return ret; |
1796 | } | |
6ac0f488 | 1797 | |
0b246afa | 1798 | cb = blk_check_plugged(btrfs_raid_unplug, fs_info, sizeof(*plug)); |
6ac0f488 CM |
1799 | if (cb) { |
1800 | plug = container_of(cb, struct btrfs_plug_cb, cb); | |
1801 | if (!plug->info) { | |
0b246afa | 1802 | plug->info = fs_info; |
6ac0f488 CM |
1803 | INIT_LIST_HEAD(&plug->rbio_list); |
1804 | } | |
1805 | list_add_tail(&rbio->plug_list, &plug->rbio_list); | |
4245215d | 1806 | ret = 0; |
6ac0f488 | 1807 | } else { |
4245215d MX |
1808 | ret = __raid56_parity_write(rbio); |
1809 | if (ret) | |
0b246afa | 1810 | btrfs_bio_counter_dec(fs_info); |
6ac0f488 | 1811 | } |
4245215d | 1812 | return ret; |
53b381b3 DW |
1813 | } |
1814 | ||
1815 | /* | |
1816 | * all parity reconstruction happens here. We've read in everything | |
1817 | * we can find from the drives and this does the heavy lifting of | |
1818 | * sorting the good from the bad. | |
1819 | */ | |
1820 | static void __raid_recover_end_io(struct btrfs_raid_bio *rbio) | |
1821 | { | |
1822 | int pagenr, stripe; | |
1823 | void **pointers; | |
1824 | int faila = -1, failb = -1; | |
53b381b3 | 1825 | struct page *page; |
58efbc9f | 1826 | blk_status_t err; |
53b381b3 DW |
1827 | int i; |
1828 | ||
31e818fe | 1829 | pointers = kcalloc(rbio->real_stripes, sizeof(void *), GFP_NOFS); |
53b381b3 | 1830 | if (!pointers) { |
58efbc9f | 1831 | err = BLK_STS_RESOURCE; |
53b381b3 DW |
1832 | goto cleanup_io; |
1833 | } | |
1834 | ||
1835 | faila = rbio->faila; | |
1836 | failb = rbio->failb; | |
1837 | ||
b4ee1782 OS |
1838 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1839 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { | |
53b381b3 DW |
1840 | spin_lock_irq(&rbio->bio_list_lock); |
1841 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
1842 | spin_unlock_irq(&rbio->bio_list_lock); | |
1843 | } | |
1844 | ||
1845 | index_rbio_pages(rbio); | |
1846 | ||
915e2290 | 1847 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
5a6ac9ea MX |
1848 | /* |
1849 | * Now we just use bitmap to mark the horizontal stripes in | |
1850 | * which we have data when doing parity scrub. | |
1851 | */ | |
1852 | if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB && | |
1853 | !test_bit(pagenr, rbio->dbitmap)) | |
1854 | continue; | |
1855 | ||
53b381b3 DW |
1856 | /* setup our array of pointers with pages |
1857 | * from each stripe | |
1858 | */ | |
2c8cdd6e | 1859 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
53b381b3 DW |
1860 | /* |
1861 | * if we're rebuilding a read, we have to use | |
1862 | * pages from the bio list | |
1863 | */ | |
b4ee1782 OS |
1864 | if ((rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1865 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) && | |
53b381b3 DW |
1866 | (stripe == faila || stripe == failb)) { |
1867 | page = page_in_rbio(rbio, stripe, pagenr, 0); | |
1868 | } else { | |
1869 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
1870 | } | |
1871 | pointers[stripe] = kmap(page); | |
1872 | } | |
1873 | ||
1874 | /* all raid6 handling here */ | |
10f11900 | 1875 | if (rbio->bbio->map_type & BTRFS_BLOCK_GROUP_RAID6) { |
53b381b3 DW |
1876 | /* |
1877 | * single failure, rebuild from parity raid5 | |
1878 | * style | |
1879 | */ | |
1880 | if (failb < 0) { | |
1881 | if (faila == rbio->nr_data) { | |
1882 | /* | |
1883 | * Just the P stripe has failed, without | |
1884 | * a bad data or Q stripe. | |
1885 | * TODO, we should redo the xor here. | |
1886 | */ | |
58efbc9f | 1887 | err = BLK_STS_IOERR; |
53b381b3 DW |
1888 | goto cleanup; |
1889 | } | |
1890 | /* | |
1891 | * a single failure in raid6 is rebuilt | |
1892 | * in the pstripe code below | |
1893 | */ | |
1894 | goto pstripe; | |
1895 | } | |
1896 | ||
1897 | /* make sure our ps and qs are in order */ | |
1898 | if (faila > failb) { | |
1899 | int tmp = failb; | |
1900 | failb = faila; | |
1901 | faila = tmp; | |
1902 | } | |
1903 | ||
1904 | /* if the q stripe is failed, do a pstripe reconstruction | |
1905 | * from the xors. | |
1906 | * If both the q stripe and the P stripe are failed, we're | |
1907 | * here due to a crc mismatch and we can't give them the | |
1908 | * data they want | |
1909 | */ | |
8e5cfb55 ZL |
1910 | if (rbio->bbio->raid_map[failb] == RAID6_Q_STRIPE) { |
1911 | if (rbio->bbio->raid_map[faila] == | |
1912 | RAID5_P_STRIPE) { | |
58efbc9f | 1913 | err = BLK_STS_IOERR; |
53b381b3 DW |
1914 | goto cleanup; |
1915 | } | |
1916 | /* | |
1917 | * otherwise we have one bad data stripe and | |
1918 | * a good P stripe. raid5! | |
1919 | */ | |
1920 | goto pstripe; | |
1921 | } | |
1922 | ||
8e5cfb55 | 1923 | if (rbio->bbio->raid_map[failb] == RAID5_P_STRIPE) { |
2c8cdd6e | 1924 | raid6_datap_recov(rbio->real_stripes, |
53b381b3 DW |
1925 | PAGE_SIZE, faila, pointers); |
1926 | } else { | |
2c8cdd6e | 1927 | raid6_2data_recov(rbio->real_stripes, |
53b381b3 DW |
1928 | PAGE_SIZE, faila, failb, |
1929 | pointers); | |
1930 | } | |
1931 | } else { | |
1932 | void *p; | |
1933 | ||
1934 | /* rebuild from P stripe here (raid5 or raid6) */ | |
1935 | BUG_ON(failb != -1); | |
1936 | pstripe: | |
1937 | /* Copy parity block into failed block to start with */ | |
1938 | memcpy(pointers[faila], | |
1939 | pointers[rbio->nr_data], | |
09cbfeaf | 1940 | PAGE_SIZE); |
53b381b3 DW |
1941 | |
1942 | /* rearrange the pointer array */ | |
1943 | p = pointers[faila]; | |
1944 | for (stripe = faila; stripe < rbio->nr_data - 1; stripe++) | |
1945 | pointers[stripe] = pointers[stripe + 1]; | |
1946 | pointers[rbio->nr_data - 1] = p; | |
1947 | ||
1948 | /* xor in the rest */ | |
09cbfeaf | 1949 | run_xor(pointers, rbio->nr_data - 1, PAGE_SIZE); |
53b381b3 DW |
1950 | } |
1951 | /* if we're doing this rebuild as part of an rmw, go through | |
1952 | * and set all of our private rbio pages in the | |
1953 | * failed stripes as uptodate. This way finish_rmw will | |
1954 | * know they can be trusted. If this was a read reconstruction, | |
1955 | * other endio functions will fiddle the uptodate bits | |
1956 | */ | |
1b94b556 | 1957 | if (rbio->operation == BTRFS_RBIO_WRITE) { |
915e2290 | 1958 | for (i = 0; i < rbio->stripe_npages; i++) { |
53b381b3 DW |
1959 | if (faila != -1) { |
1960 | page = rbio_stripe_page(rbio, faila, i); | |
1961 | SetPageUptodate(page); | |
1962 | } | |
1963 | if (failb != -1) { | |
1964 | page = rbio_stripe_page(rbio, failb, i); | |
1965 | SetPageUptodate(page); | |
1966 | } | |
1967 | } | |
1968 | } | |
2c8cdd6e | 1969 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
53b381b3 DW |
1970 | /* |
1971 | * if we're rebuilding a read, we have to use | |
1972 | * pages from the bio list | |
1973 | */ | |
b4ee1782 OS |
1974 | if ((rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1975 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) && | |
53b381b3 DW |
1976 | (stripe == faila || stripe == failb)) { |
1977 | page = page_in_rbio(rbio, stripe, pagenr, 0); | |
1978 | } else { | |
1979 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
1980 | } | |
1981 | kunmap(page); | |
1982 | } | |
1983 | } | |
1984 | ||
58efbc9f | 1985 | err = BLK_STS_OK; |
53b381b3 DW |
1986 | cleanup: |
1987 | kfree(pointers); | |
1988 | ||
1989 | cleanup_io: | |
580c6efa LB |
1990 | /* |
1991 | * Similar to READ_REBUILD, REBUILD_MISSING at this point also has a | |
1992 | * valid rbio which is consistent with ondisk content, thus such a | |
1993 | * valid rbio can be cached to avoid further disk reads. | |
1994 | */ | |
1995 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || | |
1996 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { | |
44ac474d LB |
1997 | /* |
1998 | * - In case of two failures, where rbio->failb != -1: | |
1999 | * | |
2000 | * Do not cache this rbio since the above read reconstruction | |
2001 | * (raid6_datap_recov() or raid6_2data_recov()) may have | |
2002 | * changed some content of stripes which are not identical to | |
2003 | * on-disk content any more, otherwise, a later write/recover | |
2004 | * may steal stripe_pages from this rbio and end up with | |
2005 | * corruptions or rebuild failures. | |
2006 | * | |
2007 | * - In case of single failure, where rbio->failb == -1: | |
2008 | * | |
2009 | * Cache this rbio iff the above read reconstruction is | |
2010 | * excuted without problems. | |
2011 | */ | |
2012 | if (err == BLK_STS_OK && rbio->failb < 0) | |
4ae10b3a CM |
2013 | cache_rbio_pages(rbio); |
2014 | else | |
2015 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
2016 | ||
4246a0b6 | 2017 | rbio_orig_end_io(rbio, err); |
58efbc9f | 2018 | } else if (err == BLK_STS_OK) { |
53b381b3 DW |
2019 | rbio->faila = -1; |
2020 | rbio->failb = -1; | |
5a6ac9ea MX |
2021 | |
2022 | if (rbio->operation == BTRFS_RBIO_WRITE) | |
2023 | finish_rmw(rbio); | |
2024 | else if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) | |
2025 | finish_parity_scrub(rbio, 0); | |
2026 | else | |
2027 | BUG(); | |
53b381b3 | 2028 | } else { |
4246a0b6 | 2029 | rbio_orig_end_io(rbio, err); |
53b381b3 DW |
2030 | } |
2031 | } | |
2032 | ||
2033 | /* | |
2034 | * This is called only for stripes we've read from disk to | |
2035 | * reconstruct the parity. | |
2036 | */ | |
4246a0b6 | 2037 | static void raid_recover_end_io(struct bio *bio) |
53b381b3 DW |
2038 | { |
2039 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
2040 | ||
2041 | /* | |
2042 | * we only read stripe pages off the disk, set them | |
2043 | * up to date if there were no errors | |
2044 | */ | |
4e4cbee9 | 2045 | if (bio->bi_status) |
53b381b3 DW |
2046 | fail_bio_stripe(rbio, bio); |
2047 | else | |
2048 | set_bio_pages_uptodate(bio); | |
2049 | bio_put(bio); | |
2050 | ||
b89e1b01 | 2051 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
2052 | return; |
2053 | ||
b89e1b01 | 2054 | if (atomic_read(&rbio->error) > rbio->bbio->max_errors) |
58efbc9f | 2055 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
53b381b3 DW |
2056 | else |
2057 | __raid_recover_end_io(rbio); | |
2058 | } | |
2059 | ||
2060 | /* | |
2061 | * reads everything we need off the disk to reconstruct | |
2062 | * the parity. endio handlers trigger final reconstruction | |
2063 | * when the IO is done. | |
2064 | * | |
2065 | * This is used both for reads from the higher layers and for | |
2066 | * parity construction required to finish a rmw cycle. | |
2067 | */ | |
2068 | static int __raid56_parity_recover(struct btrfs_raid_bio *rbio) | |
2069 | { | |
2070 | int bios_to_read = 0; | |
53b381b3 DW |
2071 | struct bio_list bio_list; |
2072 | int ret; | |
53b381b3 DW |
2073 | int pagenr; |
2074 | int stripe; | |
2075 | struct bio *bio; | |
2076 | ||
2077 | bio_list_init(&bio_list); | |
2078 | ||
2079 | ret = alloc_rbio_pages(rbio); | |
2080 | if (ret) | |
2081 | goto cleanup; | |
2082 | ||
b89e1b01 | 2083 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
2084 | |
2085 | /* | |
4ae10b3a CM |
2086 | * read everything that hasn't failed. Thanks to the |
2087 | * stripe cache, it is possible that some or all of these | |
2088 | * pages are going to be uptodate. | |
53b381b3 | 2089 | */ |
2c8cdd6e | 2090 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
5588383e | 2091 | if (rbio->faila == stripe || rbio->failb == stripe) { |
b89e1b01 | 2092 | atomic_inc(&rbio->error); |
53b381b3 | 2093 | continue; |
5588383e | 2094 | } |
53b381b3 | 2095 | |
915e2290 | 2096 | for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { |
53b381b3 DW |
2097 | struct page *p; |
2098 | ||
2099 | /* | |
2100 | * the rmw code may have already read this | |
2101 | * page in | |
2102 | */ | |
2103 | p = rbio_stripe_page(rbio, stripe, pagenr); | |
2104 | if (PageUptodate(p)) | |
2105 | continue; | |
2106 | ||
2107 | ret = rbio_add_io_page(rbio, &bio_list, | |
2108 | rbio_stripe_page(rbio, stripe, pagenr), | |
2109 | stripe, pagenr, rbio->stripe_len); | |
2110 | if (ret < 0) | |
2111 | goto cleanup; | |
2112 | } | |
2113 | } | |
2114 | ||
2115 | bios_to_read = bio_list_size(&bio_list); | |
2116 | if (!bios_to_read) { | |
2117 | /* | |
2118 | * we might have no bios to read just because the pages | |
2119 | * were up to date, or we might have no bios to read because | |
2120 | * the devices were gone. | |
2121 | */ | |
b89e1b01 | 2122 | if (atomic_read(&rbio->error) <= rbio->bbio->max_errors) { |
53b381b3 DW |
2123 | __raid_recover_end_io(rbio); |
2124 | goto out; | |
2125 | } else { | |
2126 | goto cleanup; | |
2127 | } | |
2128 | } | |
2129 | ||
2130 | /* | |
2131 | * the bbio may be freed once we submit the last bio. Make sure | |
2132 | * not to touch it after that | |
2133 | */ | |
b89e1b01 | 2134 | atomic_set(&rbio->stripes_pending, bios_to_read); |
53b381b3 DW |
2135 | while (1) { |
2136 | bio = bio_list_pop(&bio_list); | |
2137 | if (!bio) | |
2138 | break; | |
2139 | ||
2140 | bio->bi_private = rbio; | |
2141 | bio->bi_end_io = raid_recover_end_io; | |
37226b21 | 2142 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
53b381b3 | 2143 | |
0b246afa | 2144 | btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
53b381b3 | 2145 | |
4e49ea4a | 2146 | submit_bio(bio); |
53b381b3 DW |
2147 | } |
2148 | out: | |
2149 | return 0; | |
2150 | ||
2151 | cleanup: | |
b4ee1782 OS |
2152 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
2153 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) | |
58efbc9f | 2154 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2155 | |
2156 | while ((bio = bio_list_pop(&bio_list))) | |
2157 | bio_put(bio); | |
2158 | ||
53b381b3 DW |
2159 | return -EIO; |
2160 | } | |
2161 | ||
2162 | /* | |
2163 | * the main entry point for reads from the higher layers. This | |
2164 | * is really only called when the normal read path had a failure, | |
2165 | * so we assume the bio they send down corresponds to a failed part | |
2166 | * of the drive. | |
2167 | */ | |
2ff7e61e | 2168 | int raid56_parity_recover(struct btrfs_fs_info *fs_info, struct bio *bio, |
8e5cfb55 ZL |
2169 | struct btrfs_bio *bbio, u64 stripe_len, |
2170 | int mirror_num, int generic_io) | |
53b381b3 DW |
2171 | { |
2172 | struct btrfs_raid_bio *rbio; | |
2173 | int ret; | |
2174 | ||
abad60c6 LB |
2175 | if (generic_io) { |
2176 | ASSERT(bbio->mirror_num == mirror_num); | |
2177 | btrfs_io_bio(bio)->mirror_num = mirror_num; | |
2178 | } | |
2179 | ||
2ff7e61e | 2180 | rbio = alloc_rbio(fs_info, bbio, stripe_len); |
af8e2d1d | 2181 | if (IS_ERR(rbio)) { |
6e9606d2 ZL |
2182 | if (generic_io) |
2183 | btrfs_put_bbio(bbio); | |
53b381b3 | 2184 | return PTR_ERR(rbio); |
af8e2d1d | 2185 | } |
53b381b3 | 2186 | |
1b94b556 | 2187 | rbio->operation = BTRFS_RBIO_READ_REBUILD; |
53b381b3 | 2188 | bio_list_add(&rbio->bio_list, bio); |
4f024f37 | 2189 | rbio->bio_list_bytes = bio->bi_iter.bi_size; |
53b381b3 DW |
2190 | |
2191 | rbio->faila = find_logical_bio_stripe(rbio, bio); | |
2192 | if (rbio->faila == -1) { | |
0b246afa | 2193 | btrfs_warn(fs_info, |
e46a28ca LB |
2194 | "%s could not find the bad stripe in raid56 so that we cannot recover any more (bio has logical %llu len %llu, bbio has map_type %llu)", |
2195 | __func__, (u64)bio->bi_iter.bi_sector << 9, | |
2196 | (u64)bio->bi_iter.bi_size, bbio->map_type); | |
6e9606d2 ZL |
2197 | if (generic_io) |
2198 | btrfs_put_bbio(bbio); | |
53b381b3 DW |
2199 | kfree(rbio); |
2200 | return -EIO; | |
2201 | } | |
2202 | ||
4245215d | 2203 | if (generic_io) { |
0b246afa | 2204 | btrfs_bio_counter_inc_noblocked(fs_info); |
4245215d MX |
2205 | rbio->generic_bio_cnt = 1; |
2206 | } else { | |
6e9606d2 | 2207 | btrfs_get_bbio(bbio); |
4245215d MX |
2208 | } |
2209 | ||
53b381b3 | 2210 | /* |
8810f751 LB |
2211 | * Loop retry: |
2212 | * for 'mirror == 2', reconstruct from all other stripes. | |
2213 | * for 'mirror_num > 2', select a stripe to fail on every retry. | |
53b381b3 | 2214 | */ |
8810f751 LB |
2215 | if (mirror_num > 2) { |
2216 | /* | |
2217 | * 'mirror == 3' is to fail the p stripe and | |
2218 | * reconstruct from the q stripe. 'mirror > 3' is to | |
2219 | * fail a data stripe and reconstruct from p+q stripe. | |
2220 | */ | |
2221 | rbio->failb = rbio->real_stripes - (mirror_num - 1); | |
2222 | ASSERT(rbio->failb > 0); | |
2223 | if (rbio->failb <= rbio->faila) | |
2224 | rbio->failb--; | |
2225 | } | |
53b381b3 DW |
2226 | |
2227 | ret = lock_stripe_add(rbio); | |
2228 | ||
2229 | /* | |
2230 | * __raid56_parity_recover will end the bio with | |
2231 | * any errors it hits. We don't want to return | |
2232 | * its error value up the stack because our caller | |
2233 | * will end up calling bio_endio with any nonzero | |
2234 | * return | |
2235 | */ | |
2236 | if (ret == 0) | |
2237 | __raid56_parity_recover(rbio); | |
2238 | /* | |
2239 | * our rbio has been added to the list of | |
2240 | * rbios that will be handled after the | |
2241 | * currently lock owner is done | |
2242 | */ | |
2243 | return 0; | |
2244 | ||
2245 | } | |
2246 | ||
2247 | static void rmw_work(struct btrfs_work *work) | |
2248 | { | |
2249 | struct btrfs_raid_bio *rbio; | |
2250 | ||
2251 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2252 | raid56_rmw_stripe(rbio); | |
2253 | } | |
2254 | ||
2255 | static void read_rebuild_work(struct btrfs_work *work) | |
2256 | { | |
2257 | struct btrfs_raid_bio *rbio; | |
2258 | ||
2259 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2260 | __raid56_parity_recover(rbio); | |
2261 | } | |
5a6ac9ea MX |
2262 | |
2263 | /* | |
2264 | * The following code is used to scrub/replace the parity stripe | |
2265 | * | |
ae6529c3 QW |
2266 | * Caller must have already increased bio_counter for getting @bbio. |
2267 | * | |
5a6ac9ea MX |
2268 | * Note: We need make sure all the pages that add into the scrub/replace |
2269 | * raid bio are correct and not be changed during the scrub/replace. That | |
2270 | * is those pages just hold metadata or file data with checksum. | |
2271 | */ | |
2272 | ||
2273 | struct btrfs_raid_bio * | |
2ff7e61e | 2274 | raid56_parity_alloc_scrub_rbio(struct btrfs_fs_info *fs_info, struct bio *bio, |
8e5cfb55 ZL |
2275 | struct btrfs_bio *bbio, u64 stripe_len, |
2276 | struct btrfs_device *scrub_dev, | |
5a6ac9ea MX |
2277 | unsigned long *dbitmap, int stripe_nsectors) |
2278 | { | |
2279 | struct btrfs_raid_bio *rbio; | |
2280 | int i; | |
2281 | ||
2ff7e61e | 2282 | rbio = alloc_rbio(fs_info, bbio, stripe_len); |
5a6ac9ea MX |
2283 | if (IS_ERR(rbio)) |
2284 | return NULL; | |
2285 | bio_list_add(&rbio->bio_list, bio); | |
2286 | /* | |
2287 | * This is a special bio which is used to hold the completion handler | |
2288 | * and make the scrub rbio is similar to the other types | |
2289 | */ | |
2290 | ASSERT(!bio->bi_iter.bi_size); | |
2291 | rbio->operation = BTRFS_RBIO_PARITY_SCRUB; | |
2292 | ||
9cd3a7eb LB |
2293 | /* |
2294 | * After mapping bbio with BTRFS_MAP_WRITE, parities have been sorted | |
2295 | * to the end position, so this search can start from the first parity | |
2296 | * stripe. | |
2297 | */ | |
2298 | for (i = rbio->nr_data; i < rbio->real_stripes; i++) { | |
5a6ac9ea MX |
2299 | if (bbio->stripes[i].dev == scrub_dev) { |
2300 | rbio->scrubp = i; | |
2301 | break; | |
2302 | } | |
2303 | } | |
9cd3a7eb | 2304 | ASSERT(i < rbio->real_stripes); |
5a6ac9ea MX |
2305 | |
2306 | /* Now we just support the sectorsize equals to page size */ | |
0b246afa | 2307 | ASSERT(fs_info->sectorsize == PAGE_SIZE); |
5a6ac9ea MX |
2308 | ASSERT(rbio->stripe_npages == stripe_nsectors); |
2309 | bitmap_copy(rbio->dbitmap, dbitmap, stripe_nsectors); | |
2310 | ||
ae6529c3 QW |
2311 | /* |
2312 | * We have already increased bio_counter when getting bbio, record it | |
2313 | * so we can free it at rbio_orig_end_io(). | |
2314 | */ | |
2315 | rbio->generic_bio_cnt = 1; | |
2316 | ||
5a6ac9ea MX |
2317 | return rbio; |
2318 | } | |
2319 | ||
b4ee1782 OS |
2320 | /* Used for both parity scrub and missing. */ |
2321 | void raid56_add_scrub_pages(struct btrfs_raid_bio *rbio, struct page *page, | |
2322 | u64 logical) | |
5a6ac9ea MX |
2323 | { |
2324 | int stripe_offset; | |
2325 | int index; | |
2326 | ||
8e5cfb55 ZL |
2327 | ASSERT(logical >= rbio->bbio->raid_map[0]); |
2328 | ASSERT(logical + PAGE_SIZE <= rbio->bbio->raid_map[0] + | |
5a6ac9ea | 2329 | rbio->stripe_len * rbio->nr_data); |
8e5cfb55 | 2330 | stripe_offset = (int)(logical - rbio->bbio->raid_map[0]); |
09cbfeaf | 2331 | index = stripe_offset >> PAGE_SHIFT; |
5a6ac9ea MX |
2332 | rbio->bio_pages[index] = page; |
2333 | } | |
2334 | ||
2335 | /* | |
2336 | * We just scrub the parity that we have correct data on the same horizontal, | |
2337 | * so we needn't allocate all pages for all the stripes. | |
2338 | */ | |
2339 | static int alloc_rbio_essential_pages(struct btrfs_raid_bio *rbio) | |
2340 | { | |
2341 | int i; | |
2342 | int bit; | |
2343 | int index; | |
2344 | struct page *page; | |
2345 | ||
2346 | for_each_set_bit(bit, rbio->dbitmap, rbio->stripe_npages) { | |
2c8cdd6e | 2347 | for (i = 0; i < rbio->real_stripes; i++) { |
5a6ac9ea MX |
2348 | index = i * rbio->stripe_npages + bit; |
2349 | if (rbio->stripe_pages[index]) | |
2350 | continue; | |
2351 | ||
2352 | page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
2353 | if (!page) | |
2354 | return -ENOMEM; | |
2355 | rbio->stripe_pages[index] = page; | |
5a6ac9ea MX |
2356 | } |
2357 | } | |
2358 | return 0; | |
2359 | } | |
2360 | ||
5a6ac9ea MX |
2361 | static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, |
2362 | int need_check) | |
2363 | { | |
76035976 | 2364 | struct btrfs_bio *bbio = rbio->bbio; |
2c8cdd6e | 2365 | void *pointers[rbio->real_stripes]; |
76035976 | 2366 | DECLARE_BITMAP(pbitmap, rbio->stripe_npages); |
5a6ac9ea MX |
2367 | int nr_data = rbio->nr_data; |
2368 | int stripe; | |
2369 | int pagenr; | |
2370 | int p_stripe = -1; | |
2371 | int q_stripe = -1; | |
2372 | struct page *p_page = NULL; | |
2373 | struct page *q_page = NULL; | |
2374 | struct bio_list bio_list; | |
2375 | struct bio *bio; | |
76035976 | 2376 | int is_replace = 0; |
5a6ac9ea MX |
2377 | int ret; |
2378 | ||
2379 | bio_list_init(&bio_list); | |
2380 | ||
2c8cdd6e MX |
2381 | if (rbio->real_stripes - rbio->nr_data == 1) { |
2382 | p_stripe = rbio->real_stripes - 1; | |
2383 | } else if (rbio->real_stripes - rbio->nr_data == 2) { | |
2384 | p_stripe = rbio->real_stripes - 2; | |
2385 | q_stripe = rbio->real_stripes - 1; | |
5a6ac9ea MX |
2386 | } else { |
2387 | BUG(); | |
2388 | } | |
2389 | ||
76035976 MX |
2390 | if (bbio->num_tgtdevs && bbio->tgtdev_map[rbio->scrubp]) { |
2391 | is_replace = 1; | |
2392 | bitmap_copy(pbitmap, rbio->dbitmap, rbio->stripe_npages); | |
2393 | } | |
2394 | ||
5a6ac9ea MX |
2395 | /* |
2396 | * Because the higher layers(scrubber) are unlikely to | |
2397 | * use this area of the disk again soon, so don't cache | |
2398 | * it. | |
2399 | */ | |
2400 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
2401 | ||
2402 | if (!need_check) | |
2403 | goto writeback; | |
2404 | ||
2405 | p_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
2406 | if (!p_page) | |
2407 | goto cleanup; | |
2408 | SetPageUptodate(p_page); | |
2409 | ||
2410 | if (q_stripe != -1) { | |
2411 | q_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
2412 | if (!q_page) { | |
2413 | __free_page(p_page); | |
2414 | goto cleanup; | |
2415 | } | |
2416 | SetPageUptodate(q_page); | |
2417 | } | |
2418 | ||
2419 | atomic_set(&rbio->error, 0); | |
2420 | ||
2421 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { | |
2422 | struct page *p; | |
2423 | void *parity; | |
2424 | /* first collect one page from each data stripe */ | |
2425 | for (stripe = 0; stripe < nr_data; stripe++) { | |
2426 | p = page_in_rbio(rbio, stripe, pagenr, 0); | |
2427 | pointers[stripe] = kmap(p); | |
2428 | } | |
2429 | ||
2430 | /* then add the parity stripe */ | |
2431 | pointers[stripe++] = kmap(p_page); | |
2432 | ||
2433 | if (q_stripe != -1) { | |
2434 | ||
2435 | /* | |
2436 | * raid6, add the qstripe and call the | |
2437 | * library function to fill in our p/q | |
2438 | */ | |
2439 | pointers[stripe++] = kmap(q_page); | |
2440 | ||
2c8cdd6e | 2441 | raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE, |
5a6ac9ea MX |
2442 | pointers); |
2443 | } else { | |
2444 | /* raid5 */ | |
2445 | memcpy(pointers[nr_data], pointers[0], PAGE_SIZE); | |
09cbfeaf | 2446 | run_xor(pointers + 1, nr_data - 1, PAGE_SIZE); |
5a6ac9ea MX |
2447 | } |
2448 | ||
01327610 | 2449 | /* Check scrubbing parity and repair it */ |
5a6ac9ea MX |
2450 | p = rbio_stripe_page(rbio, rbio->scrubp, pagenr); |
2451 | parity = kmap(p); | |
09cbfeaf KS |
2452 | if (memcmp(parity, pointers[rbio->scrubp], PAGE_SIZE)) |
2453 | memcpy(parity, pointers[rbio->scrubp], PAGE_SIZE); | |
5a6ac9ea MX |
2454 | else |
2455 | /* Parity is right, needn't writeback */ | |
2456 | bitmap_clear(rbio->dbitmap, pagenr, 1); | |
2457 | kunmap(p); | |
2458 | ||
2c8cdd6e | 2459 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) |
5a6ac9ea MX |
2460 | kunmap(page_in_rbio(rbio, stripe, pagenr, 0)); |
2461 | } | |
2462 | ||
2463 | __free_page(p_page); | |
2464 | if (q_page) | |
2465 | __free_page(q_page); | |
2466 | ||
2467 | writeback: | |
2468 | /* | |
2469 | * time to start writing. Make bios for everything from the | |
2470 | * higher layers (the bio_list in our rbio) and our p/q. Ignore | |
2471 | * everything else. | |
2472 | */ | |
2473 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { | |
2474 | struct page *page; | |
2475 | ||
2476 | page = rbio_stripe_page(rbio, rbio->scrubp, pagenr); | |
2477 | ret = rbio_add_io_page(rbio, &bio_list, | |
2478 | page, rbio->scrubp, pagenr, rbio->stripe_len); | |
2479 | if (ret) | |
2480 | goto cleanup; | |
2481 | } | |
2482 | ||
76035976 MX |
2483 | if (!is_replace) |
2484 | goto submit_write; | |
2485 | ||
2486 | for_each_set_bit(pagenr, pbitmap, rbio->stripe_npages) { | |
2487 | struct page *page; | |
2488 | ||
2489 | page = rbio_stripe_page(rbio, rbio->scrubp, pagenr); | |
2490 | ret = rbio_add_io_page(rbio, &bio_list, page, | |
2491 | bbio->tgtdev_map[rbio->scrubp], | |
2492 | pagenr, rbio->stripe_len); | |
2493 | if (ret) | |
2494 | goto cleanup; | |
2495 | } | |
2496 | ||
2497 | submit_write: | |
5a6ac9ea MX |
2498 | nr_data = bio_list_size(&bio_list); |
2499 | if (!nr_data) { | |
2500 | /* Every parity is right */ | |
58efbc9f | 2501 | rbio_orig_end_io(rbio, BLK_STS_OK); |
5a6ac9ea MX |
2502 | return; |
2503 | } | |
2504 | ||
2505 | atomic_set(&rbio->stripes_pending, nr_data); | |
2506 | ||
2507 | while (1) { | |
2508 | bio = bio_list_pop(&bio_list); | |
2509 | if (!bio) | |
2510 | break; | |
2511 | ||
2512 | bio->bi_private = rbio; | |
a6111d11 | 2513 | bio->bi_end_io = raid_write_end_io; |
37226b21 | 2514 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4e49ea4a MC |
2515 | |
2516 | submit_bio(bio); | |
5a6ac9ea MX |
2517 | } |
2518 | return; | |
2519 | ||
2520 | cleanup: | |
58efbc9f | 2521 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2522 | |
2523 | while ((bio = bio_list_pop(&bio_list))) | |
2524 | bio_put(bio); | |
5a6ac9ea MX |
2525 | } |
2526 | ||
2527 | static inline int is_data_stripe(struct btrfs_raid_bio *rbio, int stripe) | |
2528 | { | |
2529 | if (stripe >= 0 && stripe < rbio->nr_data) | |
2530 | return 1; | |
2531 | return 0; | |
2532 | } | |
2533 | ||
2534 | /* | |
2535 | * While we're doing the parity check and repair, we could have errors | |
2536 | * in reading pages off the disk. This checks for errors and if we're | |
2537 | * not able to read the page it'll trigger parity reconstruction. The | |
2538 | * parity scrub will be finished after we've reconstructed the failed | |
2539 | * stripes | |
2540 | */ | |
2541 | static void validate_rbio_for_parity_scrub(struct btrfs_raid_bio *rbio) | |
2542 | { | |
2543 | if (atomic_read(&rbio->error) > rbio->bbio->max_errors) | |
2544 | goto cleanup; | |
2545 | ||
2546 | if (rbio->faila >= 0 || rbio->failb >= 0) { | |
2547 | int dfail = 0, failp = -1; | |
2548 | ||
2549 | if (is_data_stripe(rbio, rbio->faila)) | |
2550 | dfail++; | |
2551 | else if (is_parity_stripe(rbio->faila)) | |
2552 | failp = rbio->faila; | |
2553 | ||
2554 | if (is_data_stripe(rbio, rbio->failb)) | |
2555 | dfail++; | |
2556 | else if (is_parity_stripe(rbio->failb)) | |
2557 | failp = rbio->failb; | |
2558 | ||
2559 | /* | |
2560 | * Because we can not use a scrubbing parity to repair | |
2561 | * the data, so the capability of the repair is declined. | |
2562 | * (In the case of RAID5, we can not repair anything) | |
2563 | */ | |
2564 | if (dfail > rbio->bbio->max_errors - 1) | |
2565 | goto cleanup; | |
2566 | ||
2567 | /* | |
2568 | * If all data is good, only parity is correctly, just | |
2569 | * repair the parity. | |
2570 | */ | |
2571 | if (dfail == 0) { | |
2572 | finish_parity_scrub(rbio, 0); | |
2573 | return; | |
2574 | } | |
2575 | ||
2576 | /* | |
2577 | * Here means we got one corrupted data stripe and one | |
2578 | * corrupted parity on RAID6, if the corrupted parity | |
01327610 | 2579 | * is scrubbing parity, luckily, use the other one to repair |
5a6ac9ea MX |
2580 | * the data, or we can not repair the data stripe. |
2581 | */ | |
2582 | if (failp != rbio->scrubp) | |
2583 | goto cleanup; | |
2584 | ||
2585 | __raid_recover_end_io(rbio); | |
2586 | } else { | |
2587 | finish_parity_scrub(rbio, 1); | |
2588 | } | |
2589 | return; | |
2590 | ||
2591 | cleanup: | |
58efbc9f | 2592 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
5a6ac9ea MX |
2593 | } |
2594 | ||
2595 | /* | |
2596 | * end io for the read phase of the rmw cycle. All the bios here are physical | |
2597 | * stripe bios we've read from the disk so we can recalculate the parity of the | |
2598 | * stripe. | |
2599 | * | |
2600 | * This will usually kick off finish_rmw once all the bios are read in, but it | |
2601 | * may trigger parity reconstruction if we had any errors along the way | |
2602 | */ | |
4246a0b6 | 2603 | static void raid56_parity_scrub_end_io(struct bio *bio) |
5a6ac9ea MX |
2604 | { |
2605 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
2606 | ||
4e4cbee9 | 2607 | if (bio->bi_status) |
5a6ac9ea MX |
2608 | fail_bio_stripe(rbio, bio); |
2609 | else | |
2610 | set_bio_pages_uptodate(bio); | |
2611 | ||
2612 | bio_put(bio); | |
2613 | ||
2614 | if (!atomic_dec_and_test(&rbio->stripes_pending)) | |
2615 | return; | |
2616 | ||
2617 | /* | |
2618 | * this will normally call finish_rmw to start our write | |
2619 | * but if there are any failed stripes we'll reconstruct | |
2620 | * from parity first | |
2621 | */ | |
2622 | validate_rbio_for_parity_scrub(rbio); | |
2623 | } | |
2624 | ||
2625 | static void raid56_parity_scrub_stripe(struct btrfs_raid_bio *rbio) | |
2626 | { | |
2627 | int bios_to_read = 0; | |
5a6ac9ea MX |
2628 | struct bio_list bio_list; |
2629 | int ret; | |
2630 | int pagenr; | |
2631 | int stripe; | |
2632 | struct bio *bio; | |
2633 | ||
785884fc LB |
2634 | bio_list_init(&bio_list); |
2635 | ||
5a6ac9ea MX |
2636 | ret = alloc_rbio_essential_pages(rbio); |
2637 | if (ret) | |
2638 | goto cleanup; | |
2639 | ||
5a6ac9ea MX |
2640 | atomic_set(&rbio->error, 0); |
2641 | /* | |
2642 | * build a list of bios to read all the missing parts of this | |
2643 | * stripe | |
2644 | */ | |
2c8cdd6e | 2645 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
5a6ac9ea MX |
2646 | for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { |
2647 | struct page *page; | |
2648 | /* | |
2649 | * we want to find all the pages missing from | |
2650 | * the rbio and read them from the disk. If | |
2651 | * page_in_rbio finds a page in the bio list | |
2652 | * we don't need to read it off the stripe. | |
2653 | */ | |
2654 | page = page_in_rbio(rbio, stripe, pagenr, 1); | |
2655 | if (page) | |
2656 | continue; | |
2657 | ||
2658 | page = rbio_stripe_page(rbio, stripe, pagenr); | |
2659 | /* | |
2660 | * the bio cache may have handed us an uptodate | |
2661 | * page. If so, be happy and use it | |
2662 | */ | |
2663 | if (PageUptodate(page)) | |
2664 | continue; | |
2665 | ||
2666 | ret = rbio_add_io_page(rbio, &bio_list, page, | |
2667 | stripe, pagenr, rbio->stripe_len); | |
2668 | if (ret) | |
2669 | goto cleanup; | |
2670 | } | |
2671 | } | |
2672 | ||
2673 | bios_to_read = bio_list_size(&bio_list); | |
2674 | if (!bios_to_read) { | |
2675 | /* | |
2676 | * this can happen if others have merged with | |
2677 | * us, it means there is nothing left to read. | |
2678 | * But if there are missing devices it may not be | |
2679 | * safe to do the full stripe write yet. | |
2680 | */ | |
2681 | goto finish; | |
2682 | } | |
2683 | ||
2684 | /* | |
2685 | * the bbio may be freed once we submit the last bio. Make sure | |
2686 | * not to touch it after that | |
2687 | */ | |
2688 | atomic_set(&rbio->stripes_pending, bios_to_read); | |
2689 | while (1) { | |
2690 | bio = bio_list_pop(&bio_list); | |
2691 | if (!bio) | |
2692 | break; | |
2693 | ||
2694 | bio->bi_private = rbio; | |
2695 | bio->bi_end_io = raid56_parity_scrub_end_io; | |
37226b21 | 2696 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
5a6ac9ea | 2697 | |
0b246afa | 2698 | btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
5a6ac9ea | 2699 | |
4e49ea4a | 2700 | submit_bio(bio); |
5a6ac9ea MX |
2701 | } |
2702 | /* the actual write will happen once the reads are done */ | |
2703 | return; | |
2704 | ||
2705 | cleanup: | |
58efbc9f | 2706 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2707 | |
2708 | while ((bio = bio_list_pop(&bio_list))) | |
2709 | bio_put(bio); | |
2710 | ||
5a6ac9ea MX |
2711 | return; |
2712 | ||
2713 | finish: | |
2714 | validate_rbio_for_parity_scrub(rbio); | |
2715 | } | |
2716 | ||
2717 | static void scrub_parity_work(struct btrfs_work *work) | |
2718 | { | |
2719 | struct btrfs_raid_bio *rbio; | |
2720 | ||
2721 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2722 | raid56_parity_scrub_stripe(rbio); | |
2723 | } | |
2724 | ||
2725 | static void async_scrub_parity(struct btrfs_raid_bio *rbio) | |
2726 | { | |
2727 | btrfs_init_work(&rbio->work, btrfs_rmw_helper, | |
2728 | scrub_parity_work, NULL, NULL); | |
2729 | ||
0b246afa | 2730 | btrfs_queue_work(rbio->fs_info->rmw_workers, &rbio->work); |
5a6ac9ea MX |
2731 | } |
2732 | ||
2733 | void raid56_parity_submit_scrub_rbio(struct btrfs_raid_bio *rbio) | |
2734 | { | |
2735 | if (!lock_stripe_add(rbio)) | |
2736 | async_scrub_parity(rbio); | |
2737 | } | |
b4ee1782 OS |
2738 | |
2739 | /* The following code is used for dev replace of a missing RAID 5/6 device. */ | |
2740 | ||
2741 | struct btrfs_raid_bio * | |
2ff7e61e | 2742 | raid56_alloc_missing_rbio(struct btrfs_fs_info *fs_info, struct bio *bio, |
b4ee1782 OS |
2743 | struct btrfs_bio *bbio, u64 length) |
2744 | { | |
2745 | struct btrfs_raid_bio *rbio; | |
2746 | ||
2ff7e61e | 2747 | rbio = alloc_rbio(fs_info, bbio, length); |
b4ee1782 OS |
2748 | if (IS_ERR(rbio)) |
2749 | return NULL; | |
2750 | ||
2751 | rbio->operation = BTRFS_RBIO_REBUILD_MISSING; | |
2752 | bio_list_add(&rbio->bio_list, bio); | |
2753 | /* | |
2754 | * This is a special bio which is used to hold the completion handler | |
2755 | * and make the scrub rbio is similar to the other types | |
2756 | */ | |
2757 | ASSERT(!bio->bi_iter.bi_size); | |
2758 | ||
2759 | rbio->faila = find_logical_bio_stripe(rbio, bio); | |
2760 | if (rbio->faila == -1) { | |
2761 | BUG(); | |
2762 | kfree(rbio); | |
2763 | return NULL; | |
2764 | } | |
2765 | ||
ae6529c3 QW |
2766 | /* |
2767 | * When we get bbio, we have already increased bio_counter, record it | |
2768 | * so we can free it at rbio_orig_end_io() | |
2769 | */ | |
2770 | rbio->generic_bio_cnt = 1; | |
2771 | ||
b4ee1782 OS |
2772 | return rbio; |
2773 | } | |
2774 | ||
b4ee1782 OS |
2775 | void raid56_submit_missing_rbio(struct btrfs_raid_bio *rbio) |
2776 | { | |
2777 | if (!lock_stripe_add(rbio)) | |
d6a69135 | 2778 | async_read_rebuild(rbio); |
b4ee1782 | 2779 | } |