Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
53b381b3 DW |
2 | /* |
3 | * Copyright (C) 2012 Fusion-io All rights reserved. | |
4 | * Copyright (C) 2012 Intel Corp. All rights reserved. | |
53b381b3 | 5 | */ |
c1d7c514 | 6 | |
53b381b3 | 7 | #include <linux/sched.h> |
53b381b3 DW |
8 | #include <linux/bio.h> |
9 | #include <linux/slab.h> | |
53b381b3 | 10 | #include <linux/blkdev.h> |
53b381b3 DW |
11 | #include <linux/raid/pq.h> |
12 | #include <linux/hash.h> | |
13 | #include <linux/list_sort.h> | |
14 | #include <linux/raid/xor.h> | |
818e010b | 15 | #include <linux/mm.h> |
cea62800 | 16 | #include "misc.h" |
53b381b3 | 17 | #include "ctree.h" |
53b381b3 | 18 | #include "disk-io.h" |
53b381b3 DW |
19 | #include "volumes.h" |
20 | #include "raid56.h" | |
21 | #include "async-thread.h" | |
53b381b3 DW |
22 | |
23 | /* set when additional merges to this rbio are not allowed */ | |
24 | #define RBIO_RMW_LOCKED_BIT 1 | |
25 | ||
4ae10b3a CM |
26 | /* |
27 | * set when this rbio is sitting in the hash, but it is just a cache | |
28 | * of past RMW | |
29 | */ | |
30 | #define RBIO_CACHE_BIT 2 | |
31 | ||
32 | /* | |
33 | * set when it is safe to trust the stripe_pages for caching | |
34 | */ | |
35 | #define RBIO_CACHE_READY_BIT 3 | |
36 | ||
4ae10b3a CM |
37 | #define RBIO_CACHE_SIZE 1024 |
38 | ||
8a953348 DS |
39 | #define BTRFS_STRIPE_HASH_TABLE_BITS 11 |
40 | ||
41 | /* Used by the raid56 code to lock stripes for read/modify/write */ | |
42 | struct btrfs_stripe_hash { | |
43 | struct list_head hash_list; | |
44 | spinlock_t lock; | |
45 | }; | |
46 | ||
47 | /* Used by the raid56 code to lock stripes for read/modify/write */ | |
48 | struct btrfs_stripe_hash_table { | |
49 | struct list_head stripe_cache; | |
50 | spinlock_t cache_lock; | |
51 | int cache_size; | |
52 | struct btrfs_stripe_hash table[]; | |
53 | }; | |
54 | ||
eb357060 QW |
55 | /* |
56 | * A bvec like structure to present a sector inside a page. | |
57 | * | |
58 | * Unlike bvec we don't need bvlen, as it's fixed to sectorsize. | |
59 | */ | |
60 | struct sector_ptr { | |
61 | struct page *page; | |
00425dd9 QW |
62 | unsigned int pgoff:24; |
63 | unsigned int uptodate:8; | |
eb357060 QW |
64 | }; |
65 | ||
1b94b556 | 66 | enum btrfs_rbio_ops { |
b4ee1782 OS |
67 | BTRFS_RBIO_WRITE, |
68 | BTRFS_RBIO_READ_REBUILD, | |
69 | BTRFS_RBIO_PARITY_SCRUB, | |
70 | BTRFS_RBIO_REBUILD_MISSING, | |
1b94b556 MX |
71 | }; |
72 | ||
53b381b3 | 73 | struct btrfs_raid_bio { |
4c664611 | 74 | struct btrfs_io_context *bioc; |
53b381b3 | 75 | |
53b381b3 DW |
76 | /* while we're doing rmw on a stripe |
77 | * we put it into a hash table so we can | |
78 | * lock the stripe and merge more rbios | |
79 | * into it. | |
80 | */ | |
81 | struct list_head hash_list; | |
82 | ||
4ae10b3a CM |
83 | /* |
84 | * LRU list for the stripe cache | |
85 | */ | |
86 | struct list_head stripe_cache; | |
87 | ||
53b381b3 DW |
88 | /* |
89 | * for scheduling work in the helper threads | |
90 | */ | |
91 | struct btrfs_work work; | |
92 | ||
93 | /* | |
94 | * bio list and bio_list_lock are used | |
95 | * to add more bios into the stripe | |
96 | * in hopes of avoiding the full rmw | |
97 | */ | |
98 | struct bio_list bio_list; | |
99 | spinlock_t bio_list_lock; | |
100 | ||
6ac0f488 CM |
101 | /* also protected by the bio_list_lock, the |
102 | * plug list is used by the plugging code | |
103 | * to collect partial bios while plugged. The | |
104 | * stripe locking code also uses it to hand off | |
53b381b3 DW |
105 | * the stripe lock to the next pending IO |
106 | */ | |
107 | struct list_head plug_list; | |
108 | ||
109 | /* | |
110 | * flags that tell us if it is safe to | |
111 | * merge with this bio | |
112 | */ | |
113 | unsigned long flags; | |
114 | ||
53b381b3 DW |
115 | /* |
116 | * set if we're doing a parity rebuild | |
117 | * for a read from higher up, which is handled | |
118 | * differently from a parity rebuild as part of | |
119 | * rmw | |
120 | */ | |
1b94b556 | 121 | enum btrfs_rbio_ops operation; |
53b381b3 | 122 | |
29b06838 QW |
123 | /* Size of each individual stripe on disk */ |
124 | u32 stripe_len; | |
53b381b3 | 125 | |
29b06838 QW |
126 | /* How many pages there are for the full stripe including P/Q */ |
127 | u16 nr_pages; | |
53b381b3 | 128 | |
94efbe19 QW |
129 | /* How many sectors there are for the full stripe including P/Q */ |
130 | u16 nr_sectors; | |
131 | ||
29b06838 QW |
132 | /* Number of data stripes (no p/q) */ |
133 | u8 nr_data; | |
134 | ||
135 | /* Numer of all stripes (including P/Q) */ | |
136 | u8 real_stripes; | |
137 | ||
138 | /* How many pages there are for each stripe */ | |
139 | u8 stripe_npages; | |
140 | ||
94efbe19 QW |
141 | /* How many sectors there are for each stripe */ |
142 | u8 stripe_nsectors; | |
143 | ||
29b06838 QW |
144 | /* First bad stripe, -1 means no corruption */ |
145 | s8 faila; | |
146 | ||
147 | /* Second bad stripe (for RAID6 use) */ | |
148 | s8 failb; | |
149 | ||
150 | /* Stripe number that we're scrubbing */ | |
151 | u8 scrubp; | |
53b381b3 DW |
152 | |
153 | /* | |
154 | * size of all the bios in the bio_list. This | |
155 | * helps us decide if the rbio maps to a full | |
156 | * stripe or not | |
157 | */ | |
158 | int bio_list_bytes; | |
159 | ||
4245215d MX |
160 | int generic_bio_cnt; |
161 | ||
dec95574 | 162 | refcount_t refs; |
53b381b3 | 163 | |
b89e1b01 MX |
164 | atomic_t stripes_pending; |
165 | ||
166 | atomic_t error; | |
53b381b3 DW |
167 | /* |
168 | * these are two arrays of pointers. We allocate the | |
169 | * rbio big enough to hold them both and setup their | |
170 | * locations when the rbio is allocated | |
171 | */ | |
172 | ||
173 | /* pointers to pages that we allocated for | |
174 | * reading/writing stripes directly from the disk (including P/Q) | |
175 | */ | |
176 | struct page **stripe_pages; | |
177 | ||
00425dd9 QW |
178 | /* Pointers to the sectors in the bio_list, for faster lookup */ |
179 | struct sector_ptr *bio_sectors; | |
180 | ||
5a6ac9ea | 181 | /* |
eb357060 QW |
182 | * For subpage support, we need to map each sector to above |
183 | * stripe_pages. | |
5a6ac9ea | 184 | */ |
eb357060 QW |
185 | struct sector_ptr *stripe_sectors; |
186 | ||
187 | /* Bitmap to record which horizontal stripe has data */ | |
5a6ac9ea | 188 | unsigned long *dbitmap; |
1389053e KC |
189 | |
190 | /* allocated with real_stripes-many pointers for finish_*() calls */ | |
191 | void **finish_pointers; | |
192 | ||
94efbe19 | 193 | /* Allocated with stripe_nsectors-many bits for finish_*() calls */ |
1389053e | 194 | unsigned long *finish_pbitmap; |
53b381b3 DW |
195 | }; |
196 | ||
197 | static int __raid56_parity_recover(struct btrfs_raid_bio *rbio); | |
198 | static noinline void finish_rmw(struct btrfs_raid_bio *rbio); | |
199 | static void rmw_work(struct btrfs_work *work); | |
200 | static void read_rebuild_work(struct btrfs_work *work); | |
53b381b3 DW |
201 | static int fail_bio_stripe(struct btrfs_raid_bio *rbio, struct bio *bio); |
202 | static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed); | |
203 | static void __free_raid_bio(struct btrfs_raid_bio *rbio); | |
204 | static void index_rbio_pages(struct btrfs_raid_bio *rbio); | |
205 | static int alloc_rbio_pages(struct btrfs_raid_bio *rbio); | |
206 | ||
5a6ac9ea MX |
207 | static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, |
208 | int need_check); | |
a81b747d | 209 | static void scrub_parity_work(struct btrfs_work *work); |
5a6ac9ea | 210 | |
ac638859 DS |
211 | static void start_async_work(struct btrfs_raid_bio *rbio, btrfs_func_t work_func) |
212 | { | |
a0cac0ec | 213 | btrfs_init_work(&rbio->work, work_func, NULL, NULL); |
6a258d72 | 214 | btrfs_queue_work(rbio->bioc->fs_info->rmw_workers, &rbio->work); |
ac638859 DS |
215 | } |
216 | ||
53b381b3 DW |
217 | /* |
218 | * the stripe hash table is used for locking, and to collect | |
219 | * bios in hopes of making a full stripe | |
220 | */ | |
221 | int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info) | |
222 | { | |
223 | struct btrfs_stripe_hash_table *table; | |
224 | struct btrfs_stripe_hash_table *x; | |
225 | struct btrfs_stripe_hash *cur; | |
226 | struct btrfs_stripe_hash *h; | |
227 | int num_entries = 1 << BTRFS_STRIPE_HASH_TABLE_BITS; | |
228 | int i; | |
229 | ||
230 | if (info->stripe_hash_table) | |
231 | return 0; | |
232 | ||
83c8266a DS |
233 | /* |
234 | * The table is large, starting with order 4 and can go as high as | |
235 | * order 7 in case lock debugging is turned on. | |
236 | * | |
237 | * Try harder to allocate and fallback to vmalloc to lower the chance | |
238 | * of a failing mount. | |
239 | */ | |
ee787f95 | 240 | table = kvzalloc(struct_size(table, table, num_entries), GFP_KERNEL); |
818e010b DS |
241 | if (!table) |
242 | return -ENOMEM; | |
53b381b3 | 243 | |
4ae10b3a CM |
244 | spin_lock_init(&table->cache_lock); |
245 | INIT_LIST_HEAD(&table->stripe_cache); | |
246 | ||
53b381b3 DW |
247 | h = table->table; |
248 | ||
249 | for (i = 0; i < num_entries; i++) { | |
250 | cur = h + i; | |
251 | INIT_LIST_HEAD(&cur->hash_list); | |
252 | spin_lock_init(&cur->lock); | |
53b381b3 DW |
253 | } |
254 | ||
255 | x = cmpxchg(&info->stripe_hash_table, NULL, table); | |
fe3b7bb0 | 256 | kvfree(x); |
53b381b3 DW |
257 | return 0; |
258 | } | |
259 | ||
4ae10b3a CM |
260 | /* |
261 | * caching an rbio means to copy anything from the | |
ac26df8b | 262 | * bio_sectors array into the stripe_pages array. We |
4ae10b3a CM |
263 | * use the page uptodate bit in the stripe cache array |
264 | * to indicate if it has valid data | |
265 | * | |
266 | * once the caching is done, we set the cache ready | |
267 | * bit. | |
268 | */ | |
269 | static void cache_rbio_pages(struct btrfs_raid_bio *rbio) | |
270 | { | |
271 | int i; | |
4ae10b3a CM |
272 | int ret; |
273 | ||
274 | ret = alloc_rbio_pages(rbio); | |
275 | if (ret) | |
276 | return; | |
277 | ||
00425dd9 QW |
278 | for (i = 0; i < rbio->nr_sectors; i++) { |
279 | /* Some range not covered by bio (partial write), skip it */ | |
280 | if (!rbio->bio_sectors[i].page) | |
281 | continue; | |
282 | ||
283 | ASSERT(rbio->stripe_sectors[i].page); | |
284 | memcpy_page(rbio->stripe_sectors[i].page, | |
285 | rbio->stripe_sectors[i].pgoff, | |
286 | rbio->bio_sectors[i].page, | |
287 | rbio->bio_sectors[i].pgoff, | |
288 | rbio->bioc->fs_info->sectorsize); | |
289 | rbio->stripe_sectors[i].uptodate = 1; | |
290 | } | |
4ae10b3a CM |
291 | set_bit(RBIO_CACHE_READY_BIT, &rbio->flags); |
292 | } | |
293 | ||
53b381b3 DW |
294 | /* |
295 | * we hash on the first logical address of the stripe | |
296 | */ | |
297 | static int rbio_bucket(struct btrfs_raid_bio *rbio) | |
298 | { | |
4c664611 | 299 | u64 num = rbio->bioc->raid_map[0]; |
53b381b3 DW |
300 | |
301 | /* | |
302 | * we shift down quite a bit. We're using byte | |
303 | * addressing, and most of the lower bits are zeros. | |
304 | * This tends to upset hash_64, and it consistently | |
305 | * returns just one or two different values. | |
306 | * | |
307 | * shifting off the lower bits fixes things. | |
308 | */ | |
309 | return hash_64(num >> 16, BTRFS_STRIPE_HASH_TABLE_BITS); | |
310 | } | |
311 | ||
d4e28d9b QW |
312 | static bool full_page_sectors_uptodate(struct btrfs_raid_bio *rbio, |
313 | unsigned int page_nr) | |
314 | { | |
315 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; | |
316 | const u32 sectors_per_page = PAGE_SIZE / sectorsize; | |
317 | int i; | |
318 | ||
319 | ASSERT(page_nr < rbio->nr_pages); | |
320 | ||
321 | for (i = sectors_per_page * page_nr; | |
322 | i < sectors_per_page * page_nr + sectors_per_page; | |
323 | i++) { | |
324 | if (!rbio->stripe_sectors[i].uptodate) | |
325 | return false; | |
326 | } | |
327 | return true; | |
328 | } | |
329 | ||
eb357060 QW |
330 | /* |
331 | * Update the stripe_sectors[] array to use correct page and pgoff | |
332 | * | |
333 | * Should be called every time any page pointer in stripes_pages[] got modified. | |
334 | */ | |
335 | static void index_stripe_sectors(struct btrfs_raid_bio *rbio) | |
336 | { | |
337 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; | |
338 | u32 offset; | |
339 | int i; | |
340 | ||
341 | for (i = 0, offset = 0; i < rbio->nr_sectors; i++, offset += sectorsize) { | |
342 | int page_index = offset >> PAGE_SHIFT; | |
343 | ||
344 | ASSERT(page_index < rbio->nr_pages); | |
345 | rbio->stripe_sectors[i].page = rbio->stripe_pages[page_index]; | |
346 | rbio->stripe_sectors[i].pgoff = offset_in_page(offset); | |
347 | } | |
348 | } | |
349 | ||
4ae10b3a | 350 | /* |
d4e28d9b QW |
351 | * Stealing an rbio means taking all the uptodate pages from the stripe array |
352 | * in the source rbio and putting them into the destination rbio. | |
353 | * | |
354 | * This will also update the involved stripe_sectors[] which are referring to | |
355 | * the old pages. | |
4ae10b3a CM |
356 | */ |
357 | static void steal_rbio(struct btrfs_raid_bio *src, struct btrfs_raid_bio *dest) | |
358 | { | |
359 | int i; | |
360 | struct page *s; | |
361 | struct page *d; | |
362 | ||
363 | if (!test_bit(RBIO_CACHE_READY_BIT, &src->flags)) | |
364 | return; | |
365 | ||
366 | for (i = 0; i < dest->nr_pages; i++) { | |
367 | s = src->stripe_pages[i]; | |
d4e28d9b | 368 | if (!s || !full_page_sectors_uptodate(src, i)) |
4ae10b3a | 369 | continue; |
4ae10b3a CM |
370 | |
371 | d = dest->stripe_pages[i]; | |
372 | if (d) | |
373 | __free_page(d); | |
374 | ||
375 | dest->stripe_pages[i] = s; | |
376 | src->stripe_pages[i] = NULL; | |
377 | } | |
eb357060 QW |
378 | index_stripe_sectors(dest); |
379 | index_stripe_sectors(src); | |
4ae10b3a CM |
380 | } |
381 | ||
53b381b3 DW |
382 | /* |
383 | * merging means we take the bio_list from the victim and | |
384 | * splice it into the destination. The victim should | |
385 | * be discarded afterwards. | |
386 | * | |
387 | * must be called with dest->rbio_list_lock held | |
388 | */ | |
389 | static void merge_rbio(struct btrfs_raid_bio *dest, | |
390 | struct btrfs_raid_bio *victim) | |
391 | { | |
392 | bio_list_merge(&dest->bio_list, &victim->bio_list); | |
393 | dest->bio_list_bytes += victim->bio_list_bytes; | |
4245215d | 394 | dest->generic_bio_cnt += victim->generic_bio_cnt; |
53b381b3 DW |
395 | bio_list_init(&victim->bio_list); |
396 | } | |
397 | ||
398 | /* | |
4ae10b3a CM |
399 | * used to prune items that are in the cache. The caller |
400 | * must hold the hash table lock. | |
401 | */ | |
402 | static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio) | |
403 | { | |
404 | int bucket = rbio_bucket(rbio); | |
405 | struct btrfs_stripe_hash_table *table; | |
406 | struct btrfs_stripe_hash *h; | |
407 | int freeit = 0; | |
408 | ||
409 | /* | |
410 | * check the bit again under the hash table lock. | |
411 | */ | |
412 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
413 | return; | |
414 | ||
6a258d72 | 415 | table = rbio->bioc->fs_info->stripe_hash_table; |
4ae10b3a CM |
416 | h = table->table + bucket; |
417 | ||
418 | /* hold the lock for the bucket because we may be | |
419 | * removing it from the hash table | |
420 | */ | |
421 | spin_lock(&h->lock); | |
422 | ||
423 | /* | |
424 | * hold the lock for the bio list because we need | |
425 | * to make sure the bio list is empty | |
426 | */ | |
427 | spin_lock(&rbio->bio_list_lock); | |
428 | ||
429 | if (test_and_clear_bit(RBIO_CACHE_BIT, &rbio->flags)) { | |
430 | list_del_init(&rbio->stripe_cache); | |
431 | table->cache_size -= 1; | |
432 | freeit = 1; | |
433 | ||
434 | /* if the bio list isn't empty, this rbio is | |
435 | * still involved in an IO. We take it out | |
436 | * of the cache list, and drop the ref that | |
437 | * was held for the list. | |
438 | * | |
439 | * If the bio_list was empty, we also remove | |
440 | * the rbio from the hash_table, and drop | |
441 | * the corresponding ref | |
442 | */ | |
443 | if (bio_list_empty(&rbio->bio_list)) { | |
444 | if (!list_empty(&rbio->hash_list)) { | |
445 | list_del_init(&rbio->hash_list); | |
dec95574 | 446 | refcount_dec(&rbio->refs); |
4ae10b3a CM |
447 | BUG_ON(!list_empty(&rbio->plug_list)); |
448 | } | |
449 | } | |
450 | } | |
451 | ||
452 | spin_unlock(&rbio->bio_list_lock); | |
453 | spin_unlock(&h->lock); | |
454 | ||
455 | if (freeit) | |
456 | __free_raid_bio(rbio); | |
457 | } | |
458 | ||
459 | /* | |
460 | * prune a given rbio from the cache | |
461 | */ | |
462 | static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio) | |
463 | { | |
464 | struct btrfs_stripe_hash_table *table; | |
465 | unsigned long flags; | |
466 | ||
467 | if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
468 | return; | |
469 | ||
6a258d72 | 470 | table = rbio->bioc->fs_info->stripe_hash_table; |
4ae10b3a CM |
471 | |
472 | spin_lock_irqsave(&table->cache_lock, flags); | |
473 | __remove_rbio_from_cache(rbio); | |
474 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
475 | } | |
476 | ||
477 | /* | |
478 | * remove everything in the cache | |
479 | */ | |
48a3b636 | 480 | static void btrfs_clear_rbio_cache(struct btrfs_fs_info *info) |
4ae10b3a CM |
481 | { |
482 | struct btrfs_stripe_hash_table *table; | |
483 | unsigned long flags; | |
484 | struct btrfs_raid_bio *rbio; | |
485 | ||
486 | table = info->stripe_hash_table; | |
487 | ||
488 | spin_lock_irqsave(&table->cache_lock, flags); | |
489 | while (!list_empty(&table->stripe_cache)) { | |
490 | rbio = list_entry(table->stripe_cache.next, | |
491 | struct btrfs_raid_bio, | |
492 | stripe_cache); | |
493 | __remove_rbio_from_cache(rbio); | |
494 | } | |
495 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
496 | } | |
497 | ||
498 | /* | |
499 | * remove all cached entries and free the hash table | |
500 | * used by unmount | |
53b381b3 DW |
501 | */ |
502 | void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info) | |
503 | { | |
504 | if (!info->stripe_hash_table) | |
505 | return; | |
4ae10b3a | 506 | btrfs_clear_rbio_cache(info); |
f749303b | 507 | kvfree(info->stripe_hash_table); |
53b381b3 DW |
508 | info->stripe_hash_table = NULL; |
509 | } | |
510 | ||
4ae10b3a CM |
511 | /* |
512 | * insert an rbio into the stripe cache. It | |
513 | * must have already been prepared by calling | |
514 | * cache_rbio_pages | |
515 | * | |
516 | * If this rbio was already cached, it gets | |
517 | * moved to the front of the lru. | |
518 | * | |
519 | * If the size of the rbio cache is too big, we | |
520 | * prune an item. | |
521 | */ | |
522 | static void cache_rbio(struct btrfs_raid_bio *rbio) | |
523 | { | |
524 | struct btrfs_stripe_hash_table *table; | |
525 | unsigned long flags; | |
526 | ||
527 | if (!test_bit(RBIO_CACHE_READY_BIT, &rbio->flags)) | |
528 | return; | |
529 | ||
6a258d72 | 530 | table = rbio->bioc->fs_info->stripe_hash_table; |
4ae10b3a CM |
531 | |
532 | spin_lock_irqsave(&table->cache_lock, flags); | |
533 | spin_lock(&rbio->bio_list_lock); | |
534 | ||
535 | /* bump our ref if we were not in the list before */ | |
536 | if (!test_and_set_bit(RBIO_CACHE_BIT, &rbio->flags)) | |
dec95574 | 537 | refcount_inc(&rbio->refs); |
4ae10b3a CM |
538 | |
539 | if (!list_empty(&rbio->stripe_cache)){ | |
540 | list_move(&rbio->stripe_cache, &table->stripe_cache); | |
541 | } else { | |
542 | list_add(&rbio->stripe_cache, &table->stripe_cache); | |
543 | table->cache_size += 1; | |
544 | } | |
545 | ||
546 | spin_unlock(&rbio->bio_list_lock); | |
547 | ||
548 | if (table->cache_size > RBIO_CACHE_SIZE) { | |
549 | struct btrfs_raid_bio *found; | |
550 | ||
551 | found = list_entry(table->stripe_cache.prev, | |
552 | struct btrfs_raid_bio, | |
553 | stripe_cache); | |
554 | ||
555 | if (found != rbio) | |
556 | __remove_rbio_from_cache(found); | |
557 | } | |
558 | ||
559 | spin_unlock_irqrestore(&table->cache_lock, flags); | |
4ae10b3a CM |
560 | } |
561 | ||
53b381b3 DW |
562 | /* |
563 | * helper function to run the xor_blocks api. It is only | |
564 | * able to do MAX_XOR_BLOCKS at a time, so we need to | |
565 | * loop through. | |
566 | */ | |
567 | static void run_xor(void **pages, int src_cnt, ssize_t len) | |
568 | { | |
569 | int src_off = 0; | |
570 | int xor_src_cnt = 0; | |
571 | void *dest = pages[src_cnt]; | |
572 | ||
573 | while(src_cnt > 0) { | |
574 | xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS); | |
575 | xor_blocks(xor_src_cnt, len, dest, pages + src_off); | |
576 | ||
577 | src_cnt -= xor_src_cnt; | |
578 | src_off += xor_src_cnt; | |
579 | } | |
580 | } | |
581 | ||
582 | /* | |
176571a1 DS |
583 | * Returns true if the bio list inside this rbio covers an entire stripe (no |
584 | * rmw required). | |
53b381b3 | 585 | */ |
176571a1 | 586 | static int rbio_is_full(struct btrfs_raid_bio *rbio) |
53b381b3 | 587 | { |
176571a1 | 588 | unsigned long flags; |
53b381b3 DW |
589 | unsigned long size = rbio->bio_list_bytes; |
590 | int ret = 1; | |
591 | ||
176571a1 | 592 | spin_lock_irqsave(&rbio->bio_list_lock, flags); |
53b381b3 DW |
593 | if (size != rbio->nr_data * rbio->stripe_len) |
594 | ret = 0; | |
53b381b3 | 595 | BUG_ON(size > rbio->nr_data * rbio->stripe_len); |
53b381b3 | 596 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); |
176571a1 | 597 | |
53b381b3 DW |
598 | return ret; |
599 | } | |
600 | ||
601 | /* | |
602 | * returns 1 if it is safe to merge two rbios together. | |
603 | * The merging is safe if the two rbios correspond to | |
604 | * the same stripe and if they are both going in the same | |
605 | * direction (read vs write), and if neither one is | |
606 | * locked for final IO | |
607 | * | |
608 | * The caller is responsible for locking such that | |
609 | * rmw_locked is safe to test | |
610 | */ | |
611 | static int rbio_can_merge(struct btrfs_raid_bio *last, | |
612 | struct btrfs_raid_bio *cur) | |
613 | { | |
614 | if (test_bit(RBIO_RMW_LOCKED_BIT, &last->flags) || | |
615 | test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) | |
616 | return 0; | |
617 | ||
4ae10b3a CM |
618 | /* |
619 | * we can't merge with cached rbios, since the | |
620 | * idea is that when we merge the destination | |
621 | * rbio is going to run our IO for us. We can | |
01327610 | 622 | * steal from cached rbios though, other functions |
4ae10b3a CM |
623 | * handle that. |
624 | */ | |
625 | if (test_bit(RBIO_CACHE_BIT, &last->flags) || | |
626 | test_bit(RBIO_CACHE_BIT, &cur->flags)) | |
627 | return 0; | |
628 | ||
4c664611 | 629 | if (last->bioc->raid_map[0] != cur->bioc->raid_map[0]) |
53b381b3 DW |
630 | return 0; |
631 | ||
5a6ac9ea MX |
632 | /* we can't merge with different operations */ |
633 | if (last->operation != cur->operation) | |
634 | return 0; | |
635 | /* | |
636 | * We've need read the full stripe from the drive. | |
637 | * check and repair the parity and write the new results. | |
638 | * | |
639 | * We're not allowed to add any new bios to the | |
640 | * bio list here, anyone else that wants to | |
641 | * change this stripe needs to do their own rmw. | |
642 | */ | |
db34be19 | 643 | if (last->operation == BTRFS_RBIO_PARITY_SCRUB) |
53b381b3 | 644 | return 0; |
53b381b3 | 645 | |
db34be19 | 646 | if (last->operation == BTRFS_RBIO_REBUILD_MISSING) |
b4ee1782 OS |
647 | return 0; |
648 | ||
cc54ff62 LB |
649 | if (last->operation == BTRFS_RBIO_READ_REBUILD) { |
650 | int fa = last->faila; | |
651 | int fb = last->failb; | |
652 | int cur_fa = cur->faila; | |
653 | int cur_fb = cur->failb; | |
654 | ||
655 | if (last->faila >= last->failb) { | |
656 | fa = last->failb; | |
657 | fb = last->faila; | |
658 | } | |
659 | ||
660 | if (cur->faila >= cur->failb) { | |
661 | cur_fa = cur->failb; | |
662 | cur_fb = cur->faila; | |
663 | } | |
664 | ||
665 | if (fa != cur_fa || fb != cur_fb) | |
666 | return 0; | |
667 | } | |
53b381b3 DW |
668 | return 1; |
669 | } | |
670 | ||
3e77605d QW |
671 | static unsigned int rbio_stripe_sector_index(const struct btrfs_raid_bio *rbio, |
672 | unsigned int stripe_nr, | |
673 | unsigned int sector_nr) | |
674 | { | |
675 | ASSERT(stripe_nr < rbio->real_stripes); | |
676 | ASSERT(sector_nr < rbio->stripe_nsectors); | |
677 | ||
678 | return stripe_nr * rbio->stripe_nsectors + sector_nr; | |
679 | } | |
680 | ||
681 | /* Return a sector from rbio->stripe_sectors, not from the bio list */ | |
682 | static struct sector_ptr *rbio_stripe_sector(const struct btrfs_raid_bio *rbio, | |
683 | unsigned int stripe_nr, | |
684 | unsigned int sector_nr) | |
685 | { | |
686 | return &rbio->stripe_sectors[rbio_stripe_sector_index(rbio, stripe_nr, | |
687 | sector_nr)]; | |
688 | } | |
689 | ||
1145059a QW |
690 | /* Grab a sector inside P stripe */ |
691 | static struct sector_ptr *rbio_pstripe_sector(const struct btrfs_raid_bio *rbio, | |
692 | unsigned int sector_nr) | |
b7178a5f | 693 | { |
1145059a | 694 | return rbio_stripe_sector(rbio, rbio->nr_data, sector_nr); |
b7178a5f ZL |
695 | } |
696 | ||
1145059a QW |
697 | /* Grab a sector inside Q stripe, return NULL if not RAID6 */ |
698 | static struct sector_ptr *rbio_qstripe_sector(const struct btrfs_raid_bio *rbio, | |
699 | unsigned int sector_nr) | |
53b381b3 | 700 | { |
1145059a QW |
701 | if (rbio->nr_data + 1 == rbio->real_stripes) |
702 | return NULL; | |
703 | return rbio_stripe_sector(rbio, rbio->nr_data + 1, sector_nr); | |
53b381b3 DW |
704 | } |
705 | ||
53b381b3 DW |
706 | /* |
707 | * The first stripe in the table for a logical address | |
708 | * has the lock. rbios are added in one of three ways: | |
709 | * | |
710 | * 1) Nobody has the stripe locked yet. The rbio is given | |
711 | * the lock and 0 is returned. The caller must start the IO | |
712 | * themselves. | |
713 | * | |
714 | * 2) Someone has the stripe locked, but we're able to merge | |
715 | * with the lock owner. The rbio is freed and the IO will | |
716 | * start automatically along with the existing rbio. 1 is returned. | |
717 | * | |
718 | * 3) Someone has the stripe locked, but we're not able to merge. | |
719 | * The rbio is added to the lock owner's plug list, or merged into | |
720 | * an rbio already on the plug list. When the lock owner unlocks, | |
721 | * the next rbio on the list is run and the IO is started automatically. | |
722 | * 1 is returned | |
723 | * | |
724 | * If we return 0, the caller still owns the rbio and must continue with | |
725 | * IO submission. If we return 1, the caller must assume the rbio has | |
726 | * already been freed. | |
727 | */ | |
728 | static noinline int lock_stripe_add(struct btrfs_raid_bio *rbio) | |
729 | { | |
721860d5 | 730 | struct btrfs_stripe_hash *h; |
53b381b3 DW |
731 | struct btrfs_raid_bio *cur; |
732 | struct btrfs_raid_bio *pending; | |
733 | unsigned long flags; | |
53b381b3 | 734 | struct btrfs_raid_bio *freeit = NULL; |
4ae10b3a | 735 | struct btrfs_raid_bio *cache_drop = NULL; |
53b381b3 | 736 | int ret = 0; |
53b381b3 | 737 | |
6a258d72 | 738 | h = rbio->bioc->fs_info->stripe_hash_table->table + rbio_bucket(rbio); |
721860d5 | 739 | |
53b381b3 DW |
740 | spin_lock_irqsave(&h->lock, flags); |
741 | list_for_each_entry(cur, &h->hash_list, hash_list) { | |
4c664611 | 742 | if (cur->bioc->raid_map[0] != rbio->bioc->raid_map[0]) |
9d6cb1b0 | 743 | continue; |
4ae10b3a | 744 | |
9d6cb1b0 | 745 | spin_lock(&cur->bio_list_lock); |
4ae10b3a | 746 | |
9d6cb1b0 JT |
747 | /* Can we steal this cached rbio's pages? */ |
748 | if (bio_list_empty(&cur->bio_list) && | |
749 | list_empty(&cur->plug_list) && | |
750 | test_bit(RBIO_CACHE_BIT, &cur->flags) && | |
751 | !test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) { | |
752 | list_del_init(&cur->hash_list); | |
753 | refcount_dec(&cur->refs); | |
53b381b3 | 754 | |
9d6cb1b0 JT |
755 | steal_rbio(cur, rbio); |
756 | cache_drop = cur; | |
757 | spin_unlock(&cur->bio_list_lock); | |
4ae10b3a | 758 | |
9d6cb1b0 JT |
759 | goto lockit; |
760 | } | |
53b381b3 | 761 | |
9d6cb1b0 JT |
762 | /* Can we merge into the lock owner? */ |
763 | if (rbio_can_merge(cur, rbio)) { | |
764 | merge_rbio(cur, rbio); | |
53b381b3 | 765 | spin_unlock(&cur->bio_list_lock); |
9d6cb1b0 | 766 | freeit = rbio; |
53b381b3 DW |
767 | ret = 1; |
768 | goto out; | |
769 | } | |
9d6cb1b0 JT |
770 | |
771 | ||
772 | /* | |
773 | * We couldn't merge with the running rbio, see if we can merge | |
774 | * with the pending ones. We don't have to check for rmw_locked | |
775 | * because there is no way they are inside finish_rmw right now | |
776 | */ | |
777 | list_for_each_entry(pending, &cur->plug_list, plug_list) { | |
778 | if (rbio_can_merge(pending, rbio)) { | |
779 | merge_rbio(pending, rbio); | |
780 | spin_unlock(&cur->bio_list_lock); | |
781 | freeit = rbio; | |
782 | ret = 1; | |
783 | goto out; | |
784 | } | |
785 | } | |
786 | ||
787 | /* | |
788 | * No merging, put us on the tail of the plug list, our rbio | |
789 | * will be started with the currently running rbio unlocks | |
790 | */ | |
791 | list_add_tail(&rbio->plug_list, &cur->plug_list); | |
792 | spin_unlock(&cur->bio_list_lock); | |
793 | ret = 1; | |
794 | goto out; | |
53b381b3 | 795 | } |
4ae10b3a | 796 | lockit: |
dec95574 | 797 | refcount_inc(&rbio->refs); |
53b381b3 DW |
798 | list_add(&rbio->hash_list, &h->hash_list); |
799 | out: | |
800 | spin_unlock_irqrestore(&h->lock, flags); | |
4ae10b3a CM |
801 | if (cache_drop) |
802 | remove_rbio_from_cache(cache_drop); | |
53b381b3 DW |
803 | if (freeit) |
804 | __free_raid_bio(freeit); | |
805 | return ret; | |
806 | } | |
807 | ||
808 | /* | |
809 | * called as rmw or parity rebuild is completed. If the plug list has more | |
810 | * rbios waiting for this stripe, the next one on the list will be started | |
811 | */ | |
812 | static noinline void unlock_stripe(struct btrfs_raid_bio *rbio) | |
813 | { | |
814 | int bucket; | |
815 | struct btrfs_stripe_hash *h; | |
816 | unsigned long flags; | |
4ae10b3a | 817 | int keep_cache = 0; |
53b381b3 DW |
818 | |
819 | bucket = rbio_bucket(rbio); | |
6a258d72 | 820 | h = rbio->bioc->fs_info->stripe_hash_table->table + bucket; |
53b381b3 | 821 | |
4ae10b3a CM |
822 | if (list_empty(&rbio->plug_list)) |
823 | cache_rbio(rbio); | |
824 | ||
53b381b3 DW |
825 | spin_lock_irqsave(&h->lock, flags); |
826 | spin_lock(&rbio->bio_list_lock); | |
827 | ||
828 | if (!list_empty(&rbio->hash_list)) { | |
4ae10b3a CM |
829 | /* |
830 | * if we're still cached and there is no other IO | |
831 | * to perform, just leave this rbio here for others | |
832 | * to steal from later | |
833 | */ | |
834 | if (list_empty(&rbio->plug_list) && | |
835 | test_bit(RBIO_CACHE_BIT, &rbio->flags)) { | |
836 | keep_cache = 1; | |
837 | clear_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
838 | BUG_ON(!bio_list_empty(&rbio->bio_list)); | |
839 | goto done; | |
840 | } | |
53b381b3 DW |
841 | |
842 | list_del_init(&rbio->hash_list); | |
dec95574 | 843 | refcount_dec(&rbio->refs); |
53b381b3 DW |
844 | |
845 | /* | |
846 | * we use the plug list to hold all the rbios | |
847 | * waiting for the chance to lock this stripe. | |
848 | * hand the lock over to one of them. | |
849 | */ | |
850 | if (!list_empty(&rbio->plug_list)) { | |
851 | struct btrfs_raid_bio *next; | |
852 | struct list_head *head = rbio->plug_list.next; | |
853 | ||
854 | next = list_entry(head, struct btrfs_raid_bio, | |
855 | plug_list); | |
856 | ||
857 | list_del_init(&rbio->plug_list); | |
858 | ||
859 | list_add(&next->hash_list, &h->hash_list); | |
dec95574 | 860 | refcount_inc(&next->refs); |
53b381b3 DW |
861 | spin_unlock(&rbio->bio_list_lock); |
862 | spin_unlock_irqrestore(&h->lock, flags); | |
863 | ||
1b94b556 | 864 | if (next->operation == BTRFS_RBIO_READ_REBUILD) |
e66d8d5a | 865 | start_async_work(next, read_rebuild_work); |
b4ee1782 OS |
866 | else if (next->operation == BTRFS_RBIO_REBUILD_MISSING) { |
867 | steal_rbio(rbio, next); | |
e66d8d5a | 868 | start_async_work(next, read_rebuild_work); |
b4ee1782 | 869 | } else if (next->operation == BTRFS_RBIO_WRITE) { |
4ae10b3a | 870 | steal_rbio(rbio, next); |
cf6a4a75 | 871 | start_async_work(next, rmw_work); |
5a6ac9ea MX |
872 | } else if (next->operation == BTRFS_RBIO_PARITY_SCRUB) { |
873 | steal_rbio(rbio, next); | |
a81b747d | 874 | start_async_work(next, scrub_parity_work); |
4ae10b3a | 875 | } |
53b381b3 DW |
876 | |
877 | goto done_nolock; | |
53b381b3 DW |
878 | } |
879 | } | |
4ae10b3a | 880 | done: |
53b381b3 DW |
881 | spin_unlock(&rbio->bio_list_lock); |
882 | spin_unlock_irqrestore(&h->lock, flags); | |
883 | ||
884 | done_nolock: | |
4ae10b3a CM |
885 | if (!keep_cache) |
886 | remove_rbio_from_cache(rbio); | |
53b381b3 DW |
887 | } |
888 | ||
889 | static void __free_raid_bio(struct btrfs_raid_bio *rbio) | |
890 | { | |
891 | int i; | |
892 | ||
dec95574 | 893 | if (!refcount_dec_and_test(&rbio->refs)) |
53b381b3 DW |
894 | return; |
895 | ||
4ae10b3a | 896 | WARN_ON(!list_empty(&rbio->stripe_cache)); |
53b381b3 DW |
897 | WARN_ON(!list_empty(&rbio->hash_list)); |
898 | WARN_ON(!bio_list_empty(&rbio->bio_list)); | |
899 | ||
900 | for (i = 0; i < rbio->nr_pages; i++) { | |
901 | if (rbio->stripe_pages[i]) { | |
902 | __free_page(rbio->stripe_pages[i]); | |
903 | rbio->stripe_pages[i] = NULL; | |
904 | } | |
905 | } | |
af8e2d1d | 906 | |
4c664611 | 907 | btrfs_put_bioc(rbio->bioc); |
53b381b3 DW |
908 | kfree(rbio); |
909 | } | |
910 | ||
7583d8d0 | 911 | static void rbio_endio_bio_list(struct bio *cur, blk_status_t err) |
53b381b3 | 912 | { |
7583d8d0 LB |
913 | struct bio *next; |
914 | ||
915 | while (cur) { | |
916 | next = cur->bi_next; | |
917 | cur->bi_next = NULL; | |
918 | cur->bi_status = err; | |
919 | bio_endio(cur); | |
920 | cur = next; | |
921 | } | |
53b381b3 DW |
922 | } |
923 | ||
924 | /* | |
925 | * this frees the rbio and runs through all the bios in the | |
926 | * bio_list and calls end_io on them | |
927 | */ | |
4e4cbee9 | 928 | static void rbio_orig_end_io(struct btrfs_raid_bio *rbio, blk_status_t err) |
53b381b3 DW |
929 | { |
930 | struct bio *cur = bio_list_get(&rbio->bio_list); | |
7583d8d0 | 931 | struct bio *extra; |
4245215d MX |
932 | |
933 | if (rbio->generic_bio_cnt) | |
6a258d72 | 934 | btrfs_bio_counter_sub(rbio->bioc->fs_info, rbio->generic_bio_cnt); |
4245215d | 935 | |
7583d8d0 LB |
936 | /* |
937 | * At this moment, rbio->bio_list is empty, however since rbio does not | |
938 | * always have RBIO_RMW_LOCKED_BIT set and rbio is still linked on the | |
939 | * hash list, rbio may be merged with others so that rbio->bio_list | |
940 | * becomes non-empty. | |
941 | * Once unlock_stripe() is done, rbio->bio_list will not be updated any | |
942 | * more and we can call bio_endio() on all queued bios. | |
943 | */ | |
944 | unlock_stripe(rbio); | |
945 | extra = bio_list_get(&rbio->bio_list); | |
946 | __free_raid_bio(rbio); | |
53b381b3 | 947 | |
7583d8d0 LB |
948 | rbio_endio_bio_list(cur, err); |
949 | if (extra) | |
950 | rbio_endio_bio_list(extra, err); | |
53b381b3 DW |
951 | } |
952 | ||
953 | /* | |
954 | * end io function used by finish_rmw. When we finally | |
955 | * get here, we've written a full stripe | |
956 | */ | |
4246a0b6 | 957 | static void raid_write_end_io(struct bio *bio) |
53b381b3 DW |
958 | { |
959 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
4e4cbee9 | 960 | blk_status_t err = bio->bi_status; |
a6111d11 | 961 | int max_errors; |
53b381b3 DW |
962 | |
963 | if (err) | |
964 | fail_bio_stripe(rbio, bio); | |
965 | ||
966 | bio_put(bio); | |
967 | ||
b89e1b01 | 968 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
969 | return; |
970 | ||
58efbc9f | 971 | err = BLK_STS_OK; |
53b381b3 DW |
972 | |
973 | /* OK, we have read all the stripes we need to. */ | |
a6111d11 | 974 | max_errors = (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) ? |
4c664611 | 975 | 0 : rbio->bioc->max_errors; |
a6111d11 | 976 | if (atomic_read(&rbio->error) > max_errors) |
4e4cbee9 | 977 | err = BLK_STS_IOERR; |
53b381b3 | 978 | |
4246a0b6 | 979 | rbio_orig_end_io(rbio, err); |
53b381b3 DW |
980 | } |
981 | ||
3e77605d QW |
982 | /** |
983 | * Get a sector pointer specified by its @stripe_nr and @sector_nr | |
984 | * | |
985 | * @rbio: The raid bio | |
986 | * @stripe_nr: Stripe number, valid range [0, real_stripe) | |
987 | * @sector_nr: Sector number inside the stripe, | |
988 | * valid range [0, stripe_nsectors) | |
989 | * @bio_list_only: Whether to use sectors inside the bio list only. | |
990 | * | |
991 | * The read/modify/write code wants to reuse the original bio page as much | |
992 | * as possible, and only use stripe_sectors as fallback. | |
993 | */ | |
994 | static struct sector_ptr *sector_in_rbio(struct btrfs_raid_bio *rbio, | |
995 | int stripe_nr, int sector_nr, | |
996 | bool bio_list_only) | |
997 | { | |
998 | struct sector_ptr *sector; | |
999 | int index; | |
1000 | ||
1001 | ASSERT(stripe_nr >= 0 && stripe_nr < rbio->real_stripes); | |
1002 | ASSERT(sector_nr >= 0 && sector_nr < rbio->stripe_nsectors); | |
1003 | ||
1004 | index = stripe_nr * rbio->stripe_nsectors + sector_nr; | |
1005 | ASSERT(index >= 0 && index < rbio->nr_sectors); | |
1006 | ||
1007 | spin_lock_irq(&rbio->bio_list_lock); | |
1008 | sector = &rbio->bio_sectors[index]; | |
1009 | if (sector->page || bio_list_only) { | |
1010 | /* Don't return sector without a valid page pointer */ | |
1011 | if (!sector->page) | |
1012 | sector = NULL; | |
1013 | spin_unlock_irq(&rbio->bio_list_lock); | |
1014 | return sector; | |
1015 | } | |
1016 | spin_unlock_irq(&rbio->bio_list_lock); | |
1017 | ||
1018 | return &rbio->stripe_sectors[index]; | |
1019 | } | |
1020 | ||
53b381b3 DW |
1021 | /* |
1022 | * allocation and initial setup for the btrfs_raid_bio. Not | |
1023 | * this does not allocate any pages for rbio->pages. | |
1024 | */ | |
2ff7e61e | 1025 | static struct btrfs_raid_bio *alloc_rbio(struct btrfs_fs_info *fs_info, |
4c664611 | 1026 | struct btrfs_io_context *bioc, |
cc353a8b | 1027 | u32 stripe_len) |
53b381b3 | 1028 | { |
843de58b QW |
1029 | const unsigned int real_stripes = bioc->num_stripes - bioc->num_tgtdevs; |
1030 | const unsigned int stripe_npages = stripe_len >> PAGE_SHIFT; | |
1031 | const unsigned int num_pages = stripe_npages * real_stripes; | |
94efbe19 QW |
1032 | const unsigned int stripe_nsectors = stripe_len >> fs_info->sectorsize_bits; |
1033 | const unsigned int num_sectors = stripe_nsectors * real_stripes; | |
53b381b3 DW |
1034 | struct btrfs_raid_bio *rbio; |
1035 | int nr_data = 0; | |
53b381b3 DW |
1036 | void *p; |
1037 | ||
843de58b | 1038 | ASSERT(IS_ALIGNED(stripe_len, PAGE_SIZE)); |
94efbe19 QW |
1039 | /* PAGE_SIZE must also be aligned to sectorsize for subpage support */ |
1040 | ASSERT(IS_ALIGNED(PAGE_SIZE, fs_info->sectorsize)); | |
843de58b | 1041 | |
1389053e KC |
1042 | rbio = kzalloc(sizeof(*rbio) + |
1043 | sizeof(*rbio->stripe_pages) * num_pages + | |
00425dd9 | 1044 | sizeof(*rbio->bio_sectors) * num_sectors + |
eb357060 | 1045 | sizeof(*rbio->stripe_sectors) * num_sectors + |
1389053e | 1046 | sizeof(*rbio->finish_pointers) * real_stripes + |
94efbe19 QW |
1047 | sizeof(*rbio->dbitmap) * BITS_TO_LONGS(stripe_nsectors) + |
1048 | sizeof(*rbio->finish_pbitmap) * BITS_TO_LONGS(stripe_nsectors), | |
1389053e | 1049 | GFP_NOFS); |
af8e2d1d | 1050 | if (!rbio) |
53b381b3 | 1051 | return ERR_PTR(-ENOMEM); |
53b381b3 DW |
1052 | |
1053 | bio_list_init(&rbio->bio_list); | |
1054 | INIT_LIST_HEAD(&rbio->plug_list); | |
1055 | spin_lock_init(&rbio->bio_list_lock); | |
4ae10b3a | 1056 | INIT_LIST_HEAD(&rbio->stripe_cache); |
53b381b3 | 1057 | INIT_LIST_HEAD(&rbio->hash_list); |
4c664611 | 1058 | rbio->bioc = bioc; |
53b381b3 DW |
1059 | rbio->stripe_len = stripe_len; |
1060 | rbio->nr_pages = num_pages; | |
94efbe19 | 1061 | rbio->nr_sectors = num_sectors; |
2c8cdd6e | 1062 | rbio->real_stripes = real_stripes; |
5a6ac9ea | 1063 | rbio->stripe_npages = stripe_npages; |
94efbe19 | 1064 | rbio->stripe_nsectors = stripe_nsectors; |
53b381b3 DW |
1065 | rbio->faila = -1; |
1066 | rbio->failb = -1; | |
dec95574 | 1067 | refcount_set(&rbio->refs, 1); |
b89e1b01 MX |
1068 | atomic_set(&rbio->error, 0); |
1069 | atomic_set(&rbio->stripes_pending, 0); | |
53b381b3 DW |
1070 | |
1071 | /* | |
ac26df8b QW |
1072 | * The stripe_pages, bio_sectors, etc arrays point to the extra memory |
1073 | * we allocated past the end of the rbio. | |
53b381b3 DW |
1074 | */ |
1075 | p = rbio + 1; | |
1389053e KC |
1076 | #define CONSUME_ALLOC(ptr, count) do { \ |
1077 | ptr = p; \ | |
1078 | p = (unsigned char *)p + sizeof(*(ptr)) * (count); \ | |
1079 | } while (0) | |
1080 | CONSUME_ALLOC(rbio->stripe_pages, num_pages); | |
00425dd9 | 1081 | CONSUME_ALLOC(rbio->bio_sectors, num_sectors); |
eb357060 | 1082 | CONSUME_ALLOC(rbio->stripe_sectors, num_sectors); |
1389053e | 1083 | CONSUME_ALLOC(rbio->finish_pointers, real_stripes); |
94efbe19 QW |
1084 | CONSUME_ALLOC(rbio->dbitmap, BITS_TO_LONGS(stripe_nsectors)); |
1085 | CONSUME_ALLOC(rbio->finish_pbitmap, BITS_TO_LONGS(stripe_nsectors)); | |
1389053e | 1086 | #undef CONSUME_ALLOC |
53b381b3 | 1087 | |
4c664611 | 1088 | if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID5) |
10f11900 | 1089 | nr_data = real_stripes - 1; |
4c664611 | 1090 | else if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID6) |
2c8cdd6e | 1091 | nr_data = real_stripes - 2; |
53b381b3 | 1092 | else |
10f11900 | 1093 | BUG(); |
53b381b3 DW |
1094 | |
1095 | rbio->nr_data = nr_data; | |
1096 | return rbio; | |
1097 | } | |
1098 | ||
1099 | /* allocate pages for all the stripes in the bio, including parity */ | |
1100 | static int alloc_rbio_pages(struct btrfs_raid_bio *rbio) | |
1101 | { | |
eb357060 QW |
1102 | int ret; |
1103 | ||
1104 | ret = btrfs_alloc_page_array(rbio->nr_pages, rbio->stripe_pages); | |
1105 | if (ret < 0) | |
1106 | return ret; | |
1107 | /* Mapping all sectors */ | |
1108 | index_stripe_sectors(rbio); | |
1109 | return 0; | |
53b381b3 DW |
1110 | } |
1111 | ||
b7178a5f | 1112 | /* only allocate pages for p/q stripes */ |
53b381b3 DW |
1113 | static int alloc_rbio_parity_pages(struct btrfs_raid_bio *rbio) |
1114 | { | |
f77183dc | 1115 | const int data_pages = rbio->nr_data * rbio->stripe_npages; |
eb357060 | 1116 | int ret; |
53b381b3 | 1117 | |
eb357060 QW |
1118 | ret = btrfs_alloc_page_array(rbio->nr_pages - data_pages, |
1119 | rbio->stripe_pages + data_pages); | |
1120 | if (ret < 0) | |
1121 | return ret; | |
1122 | ||
1123 | index_stripe_sectors(rbio); | |
1124 | return 0; | |
53b381b3 DW |
1125 | } |
1126 | ||
1127 | /* | |
3e77605d QW |
1128 | * Add a single sector @sector into our list of bios for IO. |
1129 | * | |
1130 | * Return 0 if everything went well. | |
1131 | * Return <0 for error. | |
53b381b3 | 1132 | */ |
3e77605d QW |
1133 | static int rbio_add_io_sector(struct btrfs_raid_bio *rbio, |
1134 | struct bio_list *bio_list, | |
1135 | struct sector_ptr *sector, | |
1136 | unsigned int stripe_nr, | |
1137 | unsigned int sector_nr, | |
1138 | unsigned long bio_max_len, | |
1139 | unsigned int opf) | |
53b381b3 | 1140 | { |
3e77605d | 1141 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; |
53b381b3 | 1142 | struct bio *last = bio_list->tail; |
53b381b3 DW |
1143 | int ret; |
1144 | struct bio *bio; | |
4c664611 | 1145 | struct btrfs_io_stripe *stripe; |
53b381b3 DW |
1146 | u64 disk_start; |
1147 | ||
3e77605d QW |
1148 | /* |
1149 | * Note: here stripe_nr has taken device replace into consideration, | |
1150 | * thus it can be larger than rbio->real_stripe. | |
1151 | * So here we check against bioc->num_stripes, not rbio->real_stripes. | |
1152 | */ | |
1153 | ASSERT(stripe_nr >= 0 && stripe_nr < rbio->bioc->num_stripes); | |
1154 | ASSERT(sector_nr >= 0 && sector_nr < rbio->stripe_nsectors); | |
1155 | ASSERT(sector->page); | |
1156 | ||
1157 | /* We don't yet support subpage, thus pgoff should always be 0 */ | |
1158 | ASSERT(sector->pgoff == 0); | |
1159 | ||
4c664611 | 1160 | stripe = &rbio->bioc->stripes[stripe_nr]; |
3e77605d | 1161 | disk_start = stripe->physical + sector_nr * sectorsize; |
53b381b3 DW |
1162 | |
1163 | /* if the device is missing, just fail this stripe */ | |
1164 | if (!stripe->dev->bdev) | |
1165 | return fail_rbio_index(rbio, stripe_nr); | |
1166 | ||
1167 | /* see if we can add this page onto our existing bio */ | |
1168 | if (last) { | |
1201b58b | 1169 | u64 last_end = last->bi_iter.bi_sector << 9; |
4f024f37 | 1170 | last_end += last->bi_iter.bi_size; |
53b381b3 DW |
1171 | |
1172 | /* | |
1173 | * we can't merge these if they are from different | |
1174 | * devices or if they are not contiguous | |
1175 | */ | |
f90ae76a | 1176 | if (last_end == disk_start && !last->bi_status && |
309dca30 | 1177 | last->bi_bdev == stripe->dev->bdev) { |
3e77605d QW |
1178 | ret = bio_add_page(last, sector->page, sectorsize, |
1179 | sector->pgoff); | |
1180 | if (ret == sectorsize) | |
53b381b3 DW |
1181 | return 0; |
1182 | } | |
1183 | } | |
1184 | ||
1185 | /* put a new bio on the list */ | |
e1b4b44e CH |
1186 | bio = bio_alloc(stripe->dev->bdev, max(bio_max_len >> PAGE_SHIFT, 1UL), |
1187 | opf, GFP_NOFS); | |
4f024f37 | 1188 | bio->bi_iter.bi_sector = disk_start >> 9; |
e01bf588 | 1189 | bio->bi_private = rbio; |
53b381b3 | 1190 | |
3e77605d | 1191 | bio_add_page(bio, sector->page, sectorsize, sector->pgoff); |
53b381b3 DW |
1192 | bio_list_add(bio_list, bio); |
1193 | return 0; | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * while we're doing the read/modify/write cycle, we could | |
1198 | * have errors in reading pages off the disk. This checks | |
1199 | * for errors and if we're not able to read the page it'll | |
1200 | * trigger parity reconstruction. The rmw will be finished | |
1201 | * after we've reconstructed the failed stripes | |
1202 | */ | |
1203 | static void validate_rbio_for_rmw(struct btrfs_raid_bio *rbio) | |
1204 | { | |
1205 | if (rbio->faila >= 0 || rbio->failb >= 0) { | |
2c8cdd6e | 1206 | BUG_ON(rbio->faila == rbio->real_stripes - 1); |
53b381b3 DW |
1207 | __raid56_parity_recover(rbio); |
1208 | } else { | |
1209 | finish_rmw(rbio); | |
1210 | } | |
1211 | } | |
1212 | ||
00425dd9 QW |
1213 | static void index_one_bio(struct btrfs_raid_bio *rbio, struct bio *bio) |
1214 | { | |
1215 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; | |
1216 | struct bio_vec bvec; | |
1217 | struct bvec_iter iter; | |
1218 | u32 offset = (bio->bi_iter.bi_sector << SECTOR_SHIFT) - | |
1219 | rbio->bioc->raid_map[0]; | |
1220 | ||
1221 | if (bio_flagged(bio, BIO_CLONED)) | |
1222 | bio->bi_iter = btrfs_bio(bio)->iter; | |
1223 | ||
1224 | bio_for_each_segment(bvec, bio, iter) { | |
1225 | u32 bvec_offset; | |
1226 | ||
1227 | for (bvec_offset = 0; bvec_offset < bvec.bv_len; | |
1228 | bvec_offset += sectorsize, offset += sectorsize) { | |
1229 | int index = offset / sectorsize; | |
1230 | struct sector_ptr *sector = &rbio->bio_sectors[index]; | |
1231 | ||
1232 | sector->page = bvec.bv_page; | |
1233 | sector->pgoff = bvec.bv_offset + bvec_offset; | |
1234 | ASSERT(sector->pgoff < PAGE_SIZE); | |
1235 | } | |
1236 | } | |
1237 | } | |
1238 | ||
53b381b3 DW |
1239 | /* |
1240 | * helper function to walk our bio list and populate the bio_pages array with | |
1241 | * the result. This seems expensive, but it is faster than constantly | |
1242 | * searching through the bio list as we setup the IO in finish_rmw or stripe | |
1243 | * reconstruction. | |
1244 | * | |
1245 | * This must be called before you trust the answers from page_in_rbio | |
1246 | */ | |
1247 | static void index_rbio_pages(struct btrfs_raid_bio *rbio) | |
1248 | { | |
1249 | struct bio *bio; | |
53b381b3 DW |
1250 | |
1251 | spin_lock_irq(&rbio->bio_list_lock); | |
00425dd9 QW |
1252 | bio_list_for_each(bio, &rbio->bio_list) |
1253 | index_one_bio(rbio, bio); | |
1254 | ||
53b381b3 DW |
1255 | spin_unlock_irq(&rbio->bio_list_lock); |
1256 | } | |
1257 | ||
1258 | /* | |
1259 | * this is called from one of two situations. We either | |
1260 | * have a full stripe from the higher layers, or we've read all | |
1261 | * the missing bits off disk. | |
1262 | * | |
1263 | * This will calculate the parity and then send down any | |
1264 | * changed blocks. | |
1265 | */ | |
1266 | static noinline void finish_rmw(struct btrfs_raid_bio *rbio) | |
1267 | { | |
4c664611 | 1268 | struct btrfs_io_context *bioc = rbio->bioc; |
1145059a | 1269 | const u32 sectorsize = bioc->fs_info->sectorsize; |
1389053e | 1270 | void **pointers = rbio->finish_pointers; |
53b381b3 DW |
1271 | int nr_data = rbio->nr_data; |
1272 | int stripe; | |
3e77605d | 1273 | int sectornr; |
c17af965 | 1274 | bool has_qstripe; |
53b381b3 DW |
1275 | struct bio_list bio_list; |
1276 | struct bio *bio; | |
53b381b3 DW |
1277 | int ret; |
1278 | ||
1279 | bio_list_init(&bio_list); | |
1280 | ||
c17af965 DS |
1281 | if (rbio->real_stripes - rbio->nr_data == 1) |
1282 | has_qstripe = false; | |
1283 | else if (rbio->real_stripes - rbio->nr_data == 2) | |
1284 | has_qstripe = true; | |
1285 | else | |
53b381b3 | 1286 | BUG(); |
53b381b3 DW |
1287 | |
1288 | /* at this point we either have a full stripe, | |
1289 | * or we've read the full stripe from the drive. | |
1290 | * recalculate the parity and write the new results. | |
1291 | * | |
1292 | * We're not allowed to add any new bios to the | |
1293 | * bio list here, anyone else that wants to | |
1294 | * change this stripe needs to do their own rmw. | |
1295 | */ | |
1296 | spin_lock_irq(&rbio->bio_list_lock); | |
1297 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
1298 | spin_unlock_irq(&rbio->bio_list_lock); | |
1299 | ||
b89e1b01 | 1300 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
1301 | |
1302 | /* | |
1303 | * now that we've set rmw_locked, run through the | |
1304 | * bio list one last time and map the page pointers | |
4ae10b3a CM |
1305 | * |
1306 | * We don't cache full rbios because we're assuming | |
1307 | * the higher layers are unlikely to use this area of | |
1308 | * the disk again soon. If they do use it again, | |
1309 | * hopefully they will send another full bio. | |
53b381b3 DW |
1310 | */ |
1311 | index_rbio_pages(rbio); | |
4ae10b3a CM |
1312 | if (!rbio_is_full(rbio)) |
1313 | cache_rbio_pages(rbio); | |
1314 | else | |
1315 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
53b381b3 | 1316 | |
3e77605d | 1317 | for (sectornr = 0; sectornr < rbio->stripe_nsectors; sectornr++) { |
1145059a QW |
1318 | struct sector_ptr *sector; |
1319 | ||
1320 | /* First collect one sector from each data stripe */ | |
53b381b3 | 1321 | for (stripe = 0; stripe < nr_data; stripe++) { |
1145059a QW |
1322 | sector = sector_in_rbio(rbio, stripe, sectornr, 0); |
1323 | pointers[stripe] = kmap_local_page(sector->page) + | |
1324 | sector->pgoff; | |
53b381b3 DW |
1325 | } |
1326 | ||
1145059a QW |
1327 | /* Then add the parity stripe */ |
1328 | sector = rbio_pstripe_sector(rbio, sectornr); | |
1329 | sector->uptodate = 1; | |
1330 | pointers[stripe++] = kmap_local_page(sector->page) + sector->pgoff; | |
53b381b3 | 1331 | |
c17af965 | 1332 | if (has_qstripe) { |
53b381b3 | 1333 | /* |
1145059a QW |
1334 | * RAID6, add the qstripe and call the library function |
1335 | * to fill in our p/q | |
53b381b3 | 1336 | */ |
1145059a QW |
1337 | sector = rbio_qstripe_sector(rbio, sectornr); |
1338 | sector->uptodate = 1; | |
1339 | pointers[stripe++] = kmap_local_page(sector->page) + | |
1340 | sector->pgoff; | |
53b381b3 | 1341 | |
1145059a | 1342 | raid6_call.gen_syndrome(rbio->real_stripes, sectorsize, |
53b381b3 DW |
1343 | pointers); |
1344 | } else { | |
1345 | /* raid5 */ | |
1145059a QW |
1346 | memcpy(pointers[nr_data], pointers[0], sectorsize); |
1347 | run_xor(pointers + 1, nr_data - 1, sectorsize); | |
53b381b3 | 1348 | } |
94a0b58d IW |
1349 | for (stripe = stripe - 1; stripe >= 0; stripe--) |
1350 | kunmap_local(pointers[stripe]); | |
53b381b3 DW |
1351 | } |
1352 | ||
1353 | /* | |
1354 | * time to start writing. Make bios for everything from the | |
1355 | * higher layers (the bio_list in our rbio) and our p/q. Ignore | |
1356 | * everything else. | |
1357 | */ | |
2c8cdd6e | 1358 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
3e77605d QW |
1359 | for (sectornr = 0; sectornr < rbio->stripe_nsectors; sectornr++) { |
1360 | struct sector_ptr *sector; | |
1361 | ||
53b381b3 | 1362 | if (stripe < rbio->nr_data) { |
3e77605d QW |
1363 | sector = sector_in_rbio(rbio, stripe, sectornr, 1); |
1364 | if (!sector) | |
53b381b3 DW |
1365 | continue; |
1366 | } else { | |
3e77605d | 1367 | sector = rbio_stripe_sector(rbio, stripe, sectornr); |
53b381b3 DW |
1368 | } |
1369 | ||
3e77605d QW |
1370 | ret = rbio_add_io_sector(rbio, &bio_list, sector, stripe, |
1371 | sectornr, rbio->stripe_len, | |
1372 | REQ_OP_WRITE); | |
53b381b3 DW |
1373 | if (ret) |
1374 | goto cleanup; | |
1375 | } | |
1376 | } | |
1377 | ||
4c664611 | 1378 | if (likely(!bioc->num_tgtdevs)) |
2c8cdd6e MX |
1379 | goto write_data; |
1380 | ||
1381 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { | |
4c664611 | 1382 | if (!bioc->tgtdev_map[stripe]) |
2c8cdd6e MX |
1383 | continue; |
1384 | ||
3e77605d QW |
1385 | for (sectornr = 0; sectornr < rbio->stripe_nsectors; sectornr++) { |
1386 | struct sector_ptr *sector; | |
1387 | ||
2c8cdd6e | 1388 | if (stripe < rbio->nr_data) { |
3e77605d QW |
1389 | sector = sector_in_rbio(rbio, stripe, sectornr, 1); |
1390 | if (!sector) | |
2c8cdd6e MX |
1391 | continue; |
1392 | } else { | |
3e77605d | 1393 | sector = rbio_stripe_sector(rbio, stripe, sectornr); |
2c8cdd6e MX |
1394 | } |
1395 | ||
3e77605d | 1396 | ret = rbio_add_io_sector(rbio, &bio_list, sector, |
4c664611 | 1397 | rbio->bioc->tgtdev_map[stripe], |
3e77605d | 1398 | sectornr, rbio->stripe_len, |
e01bf588 | 1399 | REQ_OP_WRITE); |
2c8cdd6e MX |
1400 | if (ret) |
1401 | goto cleanup; | |
1402 | } | |
1403 | } | |
1404 | ||
1405 | write_data: | |
b89e1b01 MX |
1406 | atomic_set(&rbio->stripes_pending, bio_list_size(&bio_list)); |
1407 | BUG_ON(atomic_read(&rbio->stripes_pending) == 0); | |
53b381b3 | 1408 | |
bf28a605 | 1409 | while ((bio = bio_list_pop(&bio_list))) { |
53b381b3 | 1410 | bio->bi_end_io = raid_write_end_io; |
4e49ea4a MC |
1411 | |
1412 | submit_bio(bio); | |
53b381b3 DW |
1413 | } |
1414 | return; | |
1415 | ||
1416 | cleanup: | |
58efbc9f | 1417 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
1418 | |
1419 | while ((bio = bio_list_pop(&bio_list))) | |
1420 | bio_put(bio); | |
53b381b3 DW |
1421 | } |
1422 | ||
1423 | /* | |
1424 | * helper to find the stripe number for a given bio. Used to figure out which | |
1425 | * stripe has failed. This expects the bio to correspond to a physical disk, | |
1426 | * so it looks up based on physical sector numbers. | |
1427 | */ | |
1428 | static int find_bio_stripe(struct btrfs_raid_bio *rbio, | |
1429 | struct bio *bio) | |
1430 | { | |
4f024f37 | 1431 | u64 physical = bio->bi_iter.bi_sector; |
53b381b3 | 1432 | int i; |
4c664611 | 1433 | struct btrfs_io_stripe *stripe; |
53b381b3 DW |
1434 | |
1435 | physical <<= 9; | |
1436 | ||
4c664611 QW |
1437 | for (i = 0; i < rbio->bioc->num_stripes; i++) { |
1438 | stripe = &rbio->bioc->stripes[i]; | |
83025863 | 1439 | if (in_range(physical, stripe->physical, rbio->stripe_len) && |
309dca30 | 1440 | stripe->dev->bdev && bio->bi_bdev == stripe->dev->bdev) { |
53b381b3 DW |
1441 | return i; |
1442 | } | |
1443 | } | |
1444 | return -1; | |
1445 | } | |
1446 | ||
1447 | /* | |
1448 | * helper to find the stripe number for a given | |
1449 | * bio (before mapping). Used to figure out which stripe has | |
1450 | * failed. This looks up based on logical block numbers. | |
1451 | */ | |
1452 | static int find_logical_bio_stripe(struct btrfs_raid_bio *rbio, | |
1453 | struct bio *bio) | |
1454 | { | |
1201b58b | 1455 | u64 logical = bio->bi_iter.bi_sector << 9; |
53b381b3 DW |
1456 | int i; |
1457 | ||
53b381b3 | 1458 | for (i = 0; i < rbio->nr_data; i++) { |
4c664611 | 1459 | u64 stripe_start = rbio->bioc->raid_map[i]; |
83025863 NB |
1460 | |
1461 | if (in_range(logical, stripe_start, rbio->stripe_len)) | |
53b381b3 | 1462 | return i; |
53b381b3 DW |
1463 | } |
1464 | return -1; | |
1465 | } | |
1466 | ||
1467 | /* | |
1468 | * returns -EIO if we had too many failures | |
1469 | */ | |
1470 | static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed) | |
1471 | { | |
1472 | unsigned long flags; | |
1473 | int ret = 0; | |
1474 | ||
1475 | spin_lock_irqsave(&rbio->bio_list_lock, flags); | |
1476 | ||
1477 | /* we already know this stripe is bad, move on */ | |
1478 | if (rbio->faila == failed || rbio->failb == failed) | |
1479 | goto out; | |
1480 | ||
1481 | if (rbio->faila == -1) { | |
1482 | /* first failure on this rbio */ | |
1483 | rbio->faila = failed; | |
b89e1b01 | 1484 | atomic_inc(&rbio->error); |
53b381b3 DW |
1485 | } else if (rbio->failb == -1) { |
1486 | /* second failure on this rbio */ | |
1487 | rbio->failb = failed; | |
b89e1b01 | 1488 | atomic_inc(&rbio->error); |
53b381b3 DW |
1489 | } else { |
1490 | ret = -EIO; | |
1491 | } | |
1492 | out: | |
1493 | spin_unlock_irqrestore(&rbio->bio_list_lock, flags); | |
1494 | ||
1495 | return ret; | |
1496 | } | |
1497 | ||
1498 | /* | |
1499 | * helper to fail a stripe based on a physical disk | |
1500 | * bio. | |
1501 | */ | |
1502 | static int fail_bio_stripe(struct btrfs_raid_bio *rbio, | |
1503 | struct bio *bio) | |
1504 | { | |
1505 | int failed = find_bio_stripe(rbio, bio); | |
1506 | ||
1507 | if (failed < 0) | |
1508 | return -EIO; | |
1509 | ||
1510 | return fail_rbio_index(rbio, failed); | |
1511 | } | |
1512 | ||
5fdb7afc QW |
1513 | /* |
1514 | * For subpage case, we can no longer set page Uptodate directly for | |
1515 | * stripe_pages[], thus we need to locate the sector. | |
1516 | */ | |
1517 | static struct sector_ptr *find_stripe_sector(struct btrfs_raid_bio *rbio, | |
1518 | struct page *page, | |
1519 | unsigned int pgoff) | |
1520 | { | |
1521 | int i; | |
1522 | ||
1523 | for (i = 0; i < rbio->nr_sectors; i++) { | |
1524 | struct sector_ptr *sector = &rbio->stripe_sectors[i]; | |
1525 | ||
1526 | if (sector->page == page && sector->pgoff == pgoff) | |
1527 | return sector; | |
1528 | } | |
1529 | return NULL; | |
1530 | } | |
1531 | ||
53b381b3 DW |
1532 | /* |
1533 | * this sets each page in the bio uptodate. It should only be used on private | |
1534 | * rbio pages, nothing that comes in from the higher layers | |
1535 | */ | |
5fdb7afc | 1536 | static void set_bio_pages_uptodate(struct btrfs_raid_bio *rbio, struct bio *bio) |
53b381b3 | 1537 | { |
5fdb7afc | 1538 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; |
0198e5b7 | 1539 | struct bio_vec *bvec; |
6dc4f100 | 1540 | struct bvec_iter_all iter_all; |
6592e58c | 1541 | |
0198e5b7 | 1542 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
53b381b3 | 1543 | |
5fdb7afc QW |
1544 | bio_for_each_segment_all(bvec, bio, iter_all) { |
1545 | struct sector_ptr *sector; | |
1546 | int pgoff; | |
1547 | ||
1548 | for (pgoff = bvec->bv_offset; pgoff - bvec->bv_offset < bvec->bv_len; | |
1549 | pgoff += sectorsize) { | |
1550 | sector = find_stripe_sector(rbio, bvec->bv_page, pgoff); | |
1551 | ASSERT(sector); | |
1552 | if (sector) | |
1553 | sector->uptodate = 1; | |
1554 | } | |
1555 | } | |
53b381b3 DW |
1556 | } |
1557 | ||
1558 | /* | |
1559 | * end io for the read phase of the rmw cycle. All the bios here are physical | |
1560 | * stripe bios we've read from the disk so we can recalculate the parity of the | |
1561 | * stripe. | |
1562 | * | |
1563 | * This will usually kick off finish_rmw once all the bios are read in, but it | |
1564 | * may trigger parity reconstruction if we had any errors along the way | |
1565 | */ | |
4246a0b6 | 1566 | static void raid_rmw_end_io(struct bio *bio) |
53b381b3 DW |
1567 | { |
1568 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
1569 | ||
4e4cbee9 | 1570 | if (bio->bi_status) |
53b381b3 DW |
1571 | fail_bio_stripe(rbio, bio); |
1572 | else | |
5fdb7afc | 1573 | set_bio_pages_uptodate(rbio, bio); |
53b381b3 DW |
1574 | |
1575 | bio_put(bio); | |
1576 | ||
b89e1b01 | 1577 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
1578 | return; |
1579 | ||
4c664611 | 1580 | if (atomic_read(&rbio->error) > rbio->bioc->max_errors) |
53b381b3 DW |
1581 | goto cleanup; |
1582 | ||
1583 | /* | |
1584 | * this will normally call finish_rmw to start our write | |
1585 | * but if there are any failed stripes we'll reconstruct | |
1586 | * from parity first | |
1587 | */ | |
1588 | validate_rbio_for_rmw(rbio); | |
1589 | return; | |
1590 | ||
1591 | cleanup: | |
1592 | ||
58efbc9f | 1593 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
53b381b3 DW |
1594 | } |
1595 | ||
53b381b3 DW |
1596 | /* |
1597 | * the stripe must be locked by the caller. It will | |
1598 | * unlock after all the writes are done | |
1599 | */ | |
1600 | static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio) | |
1601 | { | |
1602 | int bios_to_read = 0; | |
53b381b3 DW |
1603 | struct bio_list bio_list; |
1604 | int ret; | |
3e77605d | 1605 | int sectornr; |
53b381b3 DW |
1606 | int stripe; |
1607 | struct bio *bio; | |
1608 | ||
1609 | bio_list_init(&bio_list); | |
1610 | ||
1611 | ret = alloc_rbio_pages(rbio); | |
1612 | if (ret) | |
1613 | goto cleanup; | |
1614 | ||
1615 | index_rbio_pages(rbio); | |
1616 | ||
b89e1b01 | 1617 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
1618 | /* |
1619 | * build a list of bios to read all the missing parts of this | |
1620 | * stripe | |
1621 | */ | |
1622 | for (stripe = 0; stripe < rbio->nr_data; stripe++) { | |
3e77605d QW |
1623 | for (sectornr = 0; sectornr < rbio->stripe_nsectors; sectornr++) { |
1624 | struct sector_ptr *sector; | |
1625 | ||
53b381b3 | 1626 | /* |
3e77605d QW |
1627 | * We want to find all the sectors missing from the |
1628 | * rbio and read them from the disk. If * sector_in_rbio() | |
1629 | * finds a page in the bio list we don't need to read | |
1630 | * it off the stripe. | |
53b381b3 | 1631 | */ |
3e77605d QW |
1632 | sector = sector_in_rbio(rbio, stripe, sectornr, 1); |
1633 | if (sector) | |
53b381b3 DW |
1634 | continue; |
1635 | ||
3e77605d | 1636 | sector = rbio_stripe_sector(rbio, stripe, sectornr); |
4ae10b3a | 1637 | /* |
3e77605d QW |
1638 | * The bio cache may have handed us an uptodate page. |
1639 | * If so, be happy and use it. | |
4ae10b3a | 1640 | */ |
3e77605d | 1641 | if (sector->uptodate) |
4ae10b3a CM |
1642 | continue; |
1643 | ||
3e77605d QW |
1644 | ret = rbio_add_io_sector(rbio, &bio_list, sector, |
1645 | stripe, sectornr, rbio->stripe_len, | |
e01bf588 | 1646 | REQ_OP_READ); |
53b381b3 DW |
1647 | if (ret) |
1648 | goto cleanup; | |
1649 | } | |
1650 | } | |
1651 | ||
1652 | bios_to_read = bio_list_size(&bio_list); | |
1653 | if (!bios_to_read) { | |
1654 | /* | |
1655 | * this can happen if others have merged with | |
1656 | * us, it means there is nothing left to read. | |
1657 | * But if there are missing devices it may not be | |
1658 | * safe to do the full stripe write yet. | |
1659 | */ | |
1660 | goto finish; | |
1661 | } | |
1662 | ||
1663 | /* | |
4c664611 QW |
1664 | * The bioc may be freed once we submit the last bio. Make sure not to |
1665 | * touch it after that. | |
53b381b3 | 1666 | */ |
b89e1b01 | 1667 | atomic_set(&rbio->stripes_pending, bios_to_read); |
bf28a605 | 1668 | while ((bio = bio_list_pop(&bio_list))) { |
53b381b3 DW |
1669 | bio->bi_end_io = raid_rmw_end_io; |
1670 | ||
6a258d72 | 1671 | btrfs_bio_wq_end_io(rbio->bioc->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
53b381b3 | 1672 | |
4e49ea4a | 1673 | submit_bio(bio); |
53b381b3 DW |
1674 | } |
1675 | /* the actual write will happen once the reads are done */ | |
1676 | return 0; | |
1677 | ||
1678 | cleanup: | |
58efbc9f | 1679 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
1680 | |
1681 | while ((bio = bio_list_pop(&bio_list))) | |
1682 | bio_put(bio); | |
1683 | ||
53b381b3 DW |
1684 | return -EIO; |
1685 | ||
1686 | finish: | |
1687 | validate_rbio_for_rmw(rbio); | |
1688 | return 0; | |
1689 | } | |
1690 | ||
1691 | /* | |
1692 | * if the upper layers pass in a full stripe, we thank them by only allocating | |
1693 | * enough pages to hold the parity, and sending it all down quickly. | |
1694 | */ | |
1695 | static int full_stripe_write(struct btrfs_raid_bio *rbio) | |
1696 | { | |
1697 | int ret; | |
1698 | ||
1699 | ret = alloc_rbio_parity_pages(rbio); | |
3cd846d1 MX |
1700 | if (ret) { |
1701 | __free_raid_bio(rbio); | |
53b381b3 | 1702 | return ret; |
3cd846d1 | 1703 | } |
53b381b3 DW |
1704 | |
1705 | ret = lock_stripe_add(rbio); | |
1706 | if (ret == 0) | |
1707 | finish_rmw(rbio); | |
1708 | return 0; | |
1709 | } | |
1710 | ||
1711 | /* | |
1712 | * partial stripe writes get handed over to async helpers. | |
1713 | * We're really hoping to merge a few more writes into this | |
1714 | * rbio before calculating new parity | |
1715 | */ | |
1716 | static int partial_stripe_write(struct btrfs_raid_bio *rbio) | |
1717 | { | |
1718 | int ret; | |
1719 | ||
1720 | ret = lock_stripe_add(rbio); | |
1721 | if (ret == 0) | |
cf6a4a75 | 1722 | start_async_work(rbio, rmw_work); |
53b381b3 DW |
1723 | return 0; |
1724 | } | |
1725 | ||
1726 | /* | |
1727 | * sometimes while we were reading from the drive to | |
1728 | * recalculate parity, enough new bios come into create | |
1729 | * a full stripe. So we do a check here to see if we can | |
1730 | * go directly to finish_rmw | |
1731 | */ | |
1732 | static int __raid56_parity_write(struct btrfs_raid_bio *rbio) | |
1733 | { | |
1734 | /* head off into rmw land if we don't have a full stripe */ | |
1735 | if (!rbio_is_full(rbio)) | |
1736 | return partial_stripe_write(rbio); | |
1737 | return full_stripe_write(rbio); | |
1738 | } | |
1739 | ||
6ac0f488 CM |
1740 | /* |
1741 | * We use plugging call backs to collect full stripes. | |
1742 | * Any time we get a partial stripe write while plugged | |
1743 | * we collect it into a list. When the unplug comes down, | |
1744 | * we sort the list by logical block number and merge | |
1745 | * everything we can into the same rbios | |
1746 | */ | |
1747 | struct btrfs_plug_cb { | |
1748 | struct blk_plug_cb cb; | |
1749 | struct btrfs_fs_info *info; | |
1750 | struct list_head rbio_list; | |
1751 | struct btrfs_work work; | |
1752 | }; | |
1753 | ||
1754 | /* | |
1755 | * rbios on the plug list are sorted for easier merging. | |
1756 | */ | |
4f0f586b ST |
1757 | static int plug_cmp(void *priv, const struct list_head *a, |
1758 | const struct list_head *b) | |
6ac0f488 | 1759 | { |
214cc184 DS |
1760 | const struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio, |
1761 | plug_list); | |
1762 | const struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio, | |
1763 | plug_list); | |
4f024f37 KO |
1764 | u64 a_sector = ra->bio_list.head->bi_iter.bi_sector; |
1765 | u64 b_sector = rb->bio_list.head->bi_iter.bi_sector; | |
6ac0f488 CM |
1766 | |
1767 | if (a_sector < b_sector) | |
1768 | return -1; | |
1769 | if (a_sector > b_sector) | |
1770 | return 1; | |
1771 | return 0; | |
1772 | } | |
1773 | ||
1774 | static void run_plug(struct btrfs_plug_cb *plug) | |
1775 | { | |
1776 | struct btrfs_raid_bio *cur; | |
1777 | struct btrfs_raid_bio *last = NULL; | |
1778 | ||
1779 | /* | |
1780 | * sort our plug list then try to merge | |
1781 | * everything we can in hopes of creating full | |
1782 | * stripes. | |
1783 | */ | |
1784 | list_sort(NULL, &plug->rbio_list, plug_cmp); | |
1785 | while (!list_empty(&plug->rbio_list)) { | |
1786 | cur = list_entry(plug->rbio_list.next, | |
1787 | struct btrfs_raid_bio, plug_list); | |
1788 | list_del_init(&cur->plug_list); | |
1789 | ||
1790 | if (rbio_is_full(cur)) { | |
c7b562c5 DS |
1791 | int ret; |
1792 | ||
6ac0f488 | 1793 | /* we have a full stripe, send it down */ |
c7b562c5 DS |
1794 | ret = full_stripe_write(cur); |
1795 | BUG_ON(ret); | |
6ac0f488 CM |
1796 | continue; |
1797 | } | |
1798 | if (last) { | |
1799 | if (rbio_can_merge(last, cur)) { | |
1800 | merge_rbio(last, cur); | |
1801 | __free_raid_bio(cur); | |
1802 | continue; | |
1803 | ||
1804 | } | |
1805 | __raid56_parity_write(last); | |
1806 | } | |
1807 | last = cur; | |
1808 | } | |
1809 | if (last) { | |
1810 | __raid56_parity_write(last); | |
1811 | } | |
1812 | kfree(plug); | |
1813 | } | |
1814 | ||
1815 | /* | |
1816 | * if the unplug comes from schedule, we have to push the | |
1817 | * work off to a helper thread | |
1818 | */ | |
1819 | static void unplug_work(struct btrfs_work *work) | |
1820 | { | |
1821 | struct btrfs_plug_cb *plug; | |
1822 | plug = container_of(work, struct btrfs_plug_cb, work); | |
1823 | run_plug(plug); | |
1824 | } | |
1825 | ||
1826 | static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1827 | { | |
1828 | struct btrfs_plug_cb *plug; | |
1829 | plug = container_of(cb, struct btrfs_plug_cb, cb); | |
1830 | ||
1831 | if (from_schedule) { | |
a0cac0ec | 1832 | btrfs_init_work(&plug->work, unplug_work, NULL, NULL); |
d05a33ac QW |
1833 | btrfs_queue_work(plug->info->rmw_workers, |
1834 | &plug->work); | |
6ac0f488 CM |
1835 | return; |
1836 | } | |
1837 | run_plug(plug); | |
1838 | } | |
1839 | ||
53b381b3 DW |
1840 | /* |
1841 | * our main entry point for writes from the rest of the FS. | |
1842 | */ | |
cc353a8b | 1843 | int raid56_parity_write(struct bio *bio, struct btrfs_io_context *bioc, u32 stripe_len) |
53b381b3 | 1844 | { |
6a258d72 | 1845 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
53b381b3 | 1846 | struct btrfs_raid_bio *rbio; |
6ac0f488 CM |
1847 | struct btrfs_plug_cb *plug = NULL; |
1848 | struct blk_plug_cb *cb; | |
4245215d | 1849 | int ret; |
53b381b3 | 1850 | |
4c664611 | 1851 | rbio = alloc_rbio(fs_info, bioc, stripe_len); |
af8e2d1d | 1852 | if (IS_ERR(rbio)) { |
4c664611 | 1853 | btrfs_put_bioc(bioc); |
53b381b3 | 1854 | return PTR_ERR(rbio); |
af8e2d1d | 1855 | } |
53b381b3 | 1856 | bio_list_add(&rbio->bio_list, bio); |
4f024f37 | 1857 | rbio->bio_list_bytes = bio->bi_iter.bi_size; |
1b94b556 | 1858 | rbio->operation = BTRFS_RBIO_WRITE; |
6ac0f488 | 1859 | |
0b246afa | 1860 | btrfs_bio_counter_inc_noblocked(fs_info); |
4245215d MX |
1861 | rbio->generic_bio_cnt = 1; |
1862 | ||
6ac0f488 CM |
1863 | /* |
1864 | * don't plug on full rbios, just get them out the door | |
1865 | * as quickly as we can | |
1866 | */ | |
4245215d MX |
1867 | if (rbio_is_full(rbio)) { |
1868 | ret = full_stripe_write(rbio); | |
1869 | if (ret) | |
0b246afa | 1870 | btrfs_bio_counter_dec(fs_info); |
4245215d MX |
1871 | return ret; |
1872 | } | |
6ac0f488 | 1873 | |
0b246afa | 1874 | cb = blk_check_plugged(btrfs_raid_unplug, fs_info, sizeof(*plug)); |
6ac0f488 CM |
1875 | if (cb) { |
1876 | plug = container_of(cb, struct btrfs_plug_cb, cb); | |
1877 | if (!plug->info) { | |
0b246afa | 1878 | plug->info = fs_info; |
6ac0f488 CM |
1879 | INIT_LIST_HEAD(&plug->rbio_list); |
1880 | } | |
1881 | list_add_tail(&rbio->plug_list, &plug->rbio_list); | |
4245215d | 1882 | ret = 0; |
6ac0f488 | 1883 | } else { |
4245215d MX |
1884 | ret = __raid56_parity_write(rbio); |
1885 | if (ret) | |
0b246afa | 1886 | btrfs_bio_counter_dec(fs_info); |
6ac0f488 | 1887 | } |
4245215d | 1888 | return ret; |
53b381b3 DW |
1889 | } |
1890 | ||
1891 | /* | |
1892 | * all parity reconstruction happens here. We've read in everything | |
1893 | * we can find from the drives and this does the heavy lifting of | |
1894 | * sorting the good from the bad. | |
1895 | */ | |
1896 | static void __raid_recover_end_io(struct btrfs_raid_bio *rbio) | |
1897 | { | |
07e4d380 QW |
1898 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; |
1899 | int sectornr, stripe; | |
53b381b3 | 1900 | void **pointers; |
94a0b58d | 1901 | void **unmap_array; |
53b381b3 | 1902 | int faila = -1, failb = -1; |
58efbc9f | 1903 | blk_status_t err; |
53b381b3 DW |
1904 | int i; |
1905 | ||
07e4d380 QW |
1906 | /* |
1907 | * This array stores the pointer for each sector, thus it has the extra | |
1908 | * pgoff value added from each sector | |
1909 | */ | |
31e818fe | 1910 | pointers = kcalloc(rbio->real_stripes, sizeof(void *), GFP_NOFS); |
53b381b3 | 1911 | if (!pointers) { |
58efbc9f | 1912 | err = BLK_STS_RESOURCE; |
53b381b3 DW |
1913 | goto cleanup_io; |
1914 | } | |
1915 | ||
94a0b58d IW |
1916 | /* |
1917 | * Store copy of pointers that does not get reordered during | |
1918 | * reconstruction so that kunmap_local works. | |
1919 | */ | |
1920 | unmap_array = kcalloc(rbio->real_stripes, sizeof(void *), GFP_NOFS); | |
1921 | if (!unmap_array) { | |
1922 | err = BLK_STS_RESOURCE; | |
1923 | goto cleanup_pointers; | |
1924 | } | |
1925 | ||
53b381b3 DW |
1926 | faila = rbio->faila; |
1927 | failb = rbio->failb; | |
1928 | ||
b4ee1782 OS |
1929 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1930 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { | |
53b381b3 DW |
1931 | spin_lock_irq(&rbio->bio_list_lock); |
1932 | set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); | |
1933 | spin_unlock_irq(&rbio->bio_list_lock); | |
1934 | } | |
1935 | ||
1936 | index_rbio_pages(rbio); | |
1937 | ||
07e4d380 QW |
1938 | for (sectornr = 0; sectornr < rbio->stripe_nsectors; sectornr++) { |
1939 | struct sector_ptr *sector; | |
1940 | ||
5a6ac9ea MX |
1941 | /* |
1942 | * Now we just use bitmap to mark the horizontal stripes in | |
1943 | * which we have data when doing parity scrub. | |
1944 | */ | |
1945 | if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB && | |
07e4d380 | 1946 | !test_bit(sectornr, rbio->dbitmap)) |
5a6ac9ea MX |
1947 | continue; |
1948 | ||
94a0b58d | 1949 | /* |
07e4d380 | 1950 | * Setup our array of pointers with sectors from each stripe |
94a0b58d IW |
1951 | * |
1952 | * NOTE: store a duplicate array of pointers to preserve the | |
1953 | * pointer order | |
53b381b3 | 1954 | */ |
2c8cdd6e | 1955 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
53b381b3 | 1956 | /* |
07e4d380 | 1957 | * If we're rebuilding a read, we have to use |
53b381b3 DW |
1958 | * pages from the bio list |
1959 | */ | |
b4ee1782 OS |
1960 | if ((rbio->operation == BTRFS_RBIO_READ_REBUILD || |
1961 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) && | |
53b381b3 | 1962 | (stripe == faila || stripe == failb)) { |
07e4d380 | 1963 | sector = sector_in_rbio(rbio, stripe, sectornr, 0); |
53b381b3 | 1964 | } else { |
07e4d380 | 1965 | sector = rbio_stripe_sector(rbio, stripe, sectornr); |
53b381b3 | 1966 | } |
07e4d380 QW |
1967 | ASSERT(sector->page); |
1968 | pointers[stripe] = kmap_local_page(sector->page) + | |
1969 | sector->pgoff; | |
94a0b58d | 1970 | unmap_array[stripe] = pointers[stripe]; |
53b381b3 DW |
1971 | } |
1972 | ||
07e4d380 | 1973 | /* All raid6 handling here */ |
4c664611 | 1974 | if (rbio->bioc->map_type & BTRFS_BLOCK_GROUP_RAID6) { |
07e4d380 | 1975 | /* Single failure, rebuild from parity raid5 style */ |
53b381b3 DW |
1976 | if (failb < 0) { |
1977 | if (faila == rbio->nr_data) { | |
1978 | /* | |
1979 | * Just the P stripe has failed, without | |
1980 | * a bad data or Q stripe. | |
1981 | * TODO, we should redo the xor here. | |
1982 | */ | |
58efbc9f | 1983 | err = BLK_STS_IOERR; |
53b381b3 DW |
1984 | goto cleanup; |
1985 | } | |
1986 | /* | |
1987 | * a single failure in raid6 is rebuilt | |
1988 | * in the pstripe code below | |
1989 | */ | |
1990 | goto pstripe; | |
1991 | } | |
1992 | ||
1993 | /* make sure our ps and qs are in order */ | |
b7d2083a NB |
1994 | if (faila > failb) |
1995 | swap(faila, failb); | |
53b381b3 DW |
1996 | |
1997 | /* if the q stripe is failed, do a pstripe reconstruction | |
1998 | * from the xors. | |
1999 | * If both the q stripe and the P stripe are failed, we're | |
2000 | * here due to a crc mismatch and we can't give them the | |
2001 | * data they want | |
2002 | */ | |
4c664611 QW |
2003 | if (rbio->bioc->raid_map[failb] == RAID6_Q_STRIPE) { |
2004 | if (rbio->bioc->raid_map[faila] == | |
8e5cfb55 | 2005 | RAID5_P_STRIPE) { |
58efbc9f | 2006 | err = BLK_STS_IOERR; |
53b381b3 DW |
2007 | goto cleanup; |
2008 | } | |
2009 | /* | |
2010 | * otherwise we have one bad data stripe and | |
2011 | * a good P stripe. raid5! | |
2012 | */ | |
2013 | goto pstripe; | |
2014 | } | |
2015 | ||
4c664611 | 2016 | if (rbio->bioc->raid_map[failb] == RAID5_P_STRIPE) { |
2c8cdd6e | 2017 | raid6_datap_recov(rbio->real_stripes, |
07e4d380 | 2018 | sectorsize, faila, pointers); |
53b381b3 | 2019 | } else { |
2c8cdd6e | 2020 | raid6_2data_recov(rbio->real_stripes, |
07e4d380 | 2021 | sectorsize, faila, failb, |
53b381b3 DW |
2022 | pointers); |
2023 | } | |
2024 | } else { | |
2025 | void *p; | |
2026 | ||
2027 | /* rebuild from P stripe here (raid5 or raid6) */ | |
2028 | BUG_ON(failb != -1); | |
2029 | pstripe: | |
2030 | /* Copy parity block into failed block to start with */ | |
07e4d380 | 2031 | memcpy(pointers[faila], pointers[rbio->nr_data], sectorsize); |
53b381b3 DW |
2032 | |
2033 | /* rearrange the pointer array */ | |
2034 | p = pointers[faila]; | |
2035 | for (stripe = faila; stripe < rbio->nr_data - 1; stripe++) | |
2036 | pointers[stripe] = pointers[stripe + 1]; | |
2037 | pointers[rbio->nr_data - 1] = p; | |
2038 | ||
2039 | /* xor in the rest */ | |
07e4d380 | 2040 | run_xor(pointers, rbio->nr_data - 1, sectorsize); |
53b381b3 DW |
2041 | } |
2042 | /* if we're doing this rebuild as part of an rmw, go through | |
2043 | * and set all of our private rbio pages in the | |
2044 | * failed stripes as uptodate. This way finish_rmw will | |
2045 | * know they can be trusted. If this was a read reconstruction, | |
2046 | * other endio functions will fiddle the uptodate bits | |
2047 | */ | |
1b94b556 | 2048 | if (rbio->operation == BTRFS_RBIO_WRITE) { |
07e4d380 | 2049 | for (i = 0; i < rbio->stripe_nsectors; i++) { |
53b381b3 | 2050 | if (faila != -1) { |
07e4d380 QW |
2051 | sector = rbio_stripe_sector(rbio, faila, i); |
2052 | sector->uptodate = 1; | |
53b381b3 DW |
2053 | } |
2054 | if (failb != -1) { | |
07e4d380 QW |
2055 | sector = rbio_stripe_sector(rbio, failb, i); |
2056 | sector->uptodate = 1; | |
53b381b3 DW |
2057 | } |
2058 | } | |
2059 | } | |
94a0b58d IW |
2060 | for (stripe = rbio->real_stripes - 1; stripe >= 0; stripe--) |
2061 | kunmap_local(unmap_array[stripe]); | |
53b381b3 DW |
2062 | } |
2063 | ||
58efbc9f | 2064 | err = BLK_STS_OK; |
53b381b3 | 2065 | cleanup: |
94a0b58d IW |
2066 | kfree(unmap_array); |
2067 | cleanup_pointers: | |
53b381b3 DW |
2068 | kfree(pointers); |
2069 | ||
2070 | cleanup_io: | |
580c6efa LB |
2071 | /* |
2072 | * Similar to READ_REBUILD, REBUILD_MISSING at this point also has a | |
2073 | * valid rbio which is consistent with ondisk content, thus such a | |
2074 | * valid rbio can be cached to avoid further disk reads. | |
2075 | */ | |
2076 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || | |
2077 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { | |
44ac474d LB |
2078 | /* |
2079 | * - In case of two failures, where rbio->failb != -1: | |
2080 | * | |
2081 | * Do not cache this rbio since the above read reconstruction | |
2082 | * (raid6_datap_recov() or raid6_2data_recov()) may have | |
2083 | * changed some content of stripes which are not identical to | |
2084 | * on-disk content any more, otherwise, a later write/recover | |
2085 | * may steal stripe_pages from this rbio and end up with | |
2086 | * corruptions or rebuild failures. | |
2087 | * | |
2088 | * - In case of single failure, where rbio->failb == -1: | |
2089 | * | |
2090 | * Cache this rbio iff the above read reconstruction is | |
52042d8e | 2091 | * executed without problems. |
44ac474d LB |
2092 | */ |
2093 | if (err == BLK_STS_OK && rbio->failb < 0) | |
4ae10b3a CM |
2094 | cache_rbio_pages(rbio); |
2095 | else | |
2096 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
2097 | ||
4246a0b6 | 2098 | rbio_orig_end_io(rbio, err); |
58efbc9f | 2099 | } else if (err == BLK_STS_OK) { |
53b381b3 DW |
2100 | rbio->faila = -1; |
2101 | rbio->failb = -1; | |
5a6ac9ea MX |
2102 | |
2103 | if (rbio->operation == BTRFS_RBIO_WRITE) | |
2104 | finish_rmw(rbio); | |
2105 | else if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) | |
2106 | finish_parity_scrub(rbio, 0); | |
2107 | else | |
2108 | BUG(); | |
53b381b3 | 2109 | } else { |
4246a0b6 | 2110 | rbio_orig_end_io(rbio, err); |
53b381b3 DW |
2111 | } |
2112 | } | |
2113 | ||
2114 | /* | |
2115 | * This is called only for stripes we've read from disk to | |
2116 | * reconstruct the parity. | |
2117 | */ | |
4246a0b6 | 2118 | static void raid_recover_end_io(struct bio *bio) |
53b381b3 DW |
2119 | { |
2120 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
2121 | ||
2122 | /* | |
2123 | * we only read stripe pages off the disk, set them | |
2124 | * up to date if there were no errors | |
2125 | */ | |
4e4cbee9 | 2126 | if (bio->bi_status) |
53b381b3 DW |
2127 | fail_bio_stripe(rbio, bio); |
2128 | else | |
5fdb7afc | 2129 | set_bio_pages_uptodate(rbio, bio); |
53b381b3 DW |
2130 | bio_put(bio); |
2131 | ||
b89e1b01 | 2132 | if (!atomic_dec_and_test(&rbio->stripes_pending)) |
53b381b3 DW |
2133 | return; |
2134 | ||
4c664611 | 2135 | if (atomic_read(&rbio->error) > rbio->bioc->max_errors) |
58efbc9f | 2136 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
53b381b3 DW |
2137 | else |
2138 | __raid_recover_end_io(rbio); | |
2139 | } | |
2140 | ||
2141 | /* | |
2142 | * reads everything we need off the disk to reconstruct | |
2143 | * the parity. endio handlers trigger final reconstruction | |
2144 | * when the IO is done. | |
2145 | * | |
2146 | * This is used both for reads from the higher layers and for | |
2147 | * parity construction required to finish a rmw cycle. | |
2148 | */ | |
2149 | static int __raid56_parity_recover(struct btrfs_raid_bio *rbio) | |
2150 | { | |
2151 | int bios_to_read = 0; | |
53b381b3 DW |
2152 | struct bio_list bio_list; |
2153 | int ret; | |
3e77605d | 2154 | int sectornr; |
53b381b3 DW |
2155 | int stripe; |
2156 | struct bio *bio; | |
2157 | ||
2158 | bio_list_init(&bio_list); | |
2159 | ||
2160 | ret = alloc_rbio_pages(rbio); | |
2161 | if (ret) | |
2162 | goto cleanup; | |
2163 | ||
b89e1b01 | 2164 | atomic_set(&rbio->error, 0); |
53b381b3 DW |
2165 | |
2166 | /* | |
4ae10b3a CM |
2167 | * read everything that hasn't failed. Thanks to the |
2168 | * stripe cache, it is possible that some or all of these | |
2169 | * pages are going to be uptodate. | |
53b381b3 | 2170 | */ |
2c8cdd6e | 2171 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
5588383e | 2172 | if (rbio->faila == stripe || rbio->failb == stripe) { |
b89e1b01 | 2173 | atomic_inc(&rbio->error); |
53b381b3 | 2174 | continue; |
5588383e | 2175 | } |
53b381b3 | 2176 | |
3e77605d QW |
2177 | for (sectornr = 0; sectornr < rbio->stripe_nsectors; sectornr++) { |
2178 | struct sector_ptr *sector; | |
53b381b3 DW |
2179 | |
2180 | /* | |
2181 | * the rmw code may have already read this | |
2182 | * page in | |
2183 | */ | |
3e77605d QW |
2184 | sector = rbio_stripe_sector(rbio, stripe, sectornr); |
2185 | if (sector->uptodate) | |
53b381b3 DW |
2186 | continue; |
2187 | ||
3e77605d QW |
2188 | ret = rbio_add_io_sector(rbio, &bio_list, sector, |
2189 | stripe, sectornr, rbio->stripe_len, | |
2190 | REQ_OP_READ); | |
53b381b3 DW |
2191 | if (ret < 0) |
2192 | goto cleanup; | |
2193 | } | |
2194 | } | |
2195 | ||
2196 | bios_to_read = bio_list_size(&bio_list); | |
2197 | if (!bios_to_read) { | |
2198 | /* | |
2199 | * we might have no bios to read just because the pages | |
2200 | * were up to date, or we might have no bios to read because | |
2201 | * the devices were gone. | |
2202 | */ | |
4c664611 | 2203 | if (atomic_read(&rbio->error) <= rbio->bioc->max_errors) { |
53b381b3 | 2204 | __raid_recover_end_io(rbio); |
813f8a0e | 2205 | return 0; |
53b381b3 DW |
2206 | } else { |
2207 | goto cleanup; | |
2208 | } | |
2209 | } | |
2210 | ||
2211 | /* | |
4c664611 QW |
2212 | * The bioc may be freed once we submit the last bio. Make sure not to |
2213 | * touch it after that. | |
53b381b3 | 2214 | */ |
b89e1b01 | 2215 | atomic_set(&rbio->stripes_pending, bios_to_read); |
bf28a605 | 2216 | while ((bio = bio_list_pop(&bio_list))) { |
53b381b3 DW |
2217 | bio->bi_end_io = raid_recover_end_io; |
2218 | ||
6a258d72 | 2219 | btrfs_bio_wq_end_io(rbio->bioc->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
53b381b3 | 2220 | |
4e49ea4a | 2221 | submit_bio(bio); |
53b381b3 | 2222 | } |
813f8a0e | 2223 | |
53b381b3 DW |
2224 | return 0; |
2225 | ||
2226 | cleanup: | |
b4ee1782 OS |
2227 | if (rbio->operation == BTRFS_RBIO_READ_REBUILD || |
2228 | rbio->operation == BTRFS_RBIO_REBUILD_MISSING) | |
58efbc9f | 2229 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2230 | |
2231 | while ((bio = bio_list_pop(&bio_list))) | |
2232 | bio_put(bio); | |
2233 | ||
53b381b3 DW |
2234 | return -EIO; |
2235 | } | |
2236 | ||
2237 | /* | |
2238 | * the main entry point for reads from the higher layers. This | |
2239 | * is really only called when the normal read path had a failure, | |
2240 | * so we assume the bio they send down corresponds to a failed part | |
2241 | * of the drive. | |
2242 | */ | |
6a258d72 | 2243 | int raid56_parity_recover(struct bio *bio, struct btrfs_io_context *bioc, |
cc353a8b | 2244 | u32 stripe_len, int mirror_num, int generic_io) |
53b381b3 | 2245 | { |
6a258d72 | 2246 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
53b381b3 DW |
2247 | struct btrfs_raid_bio *rbio; |
2248 | int ret; | |
2249 | ||
abad60c6 | 2250 | if (generic_io) { |
4c664611 | 2251 | ASSERT(bioc->mirror_num == mirror_num); |
c3a3b19b | 2252 | btrfs_bio(bio)->mirror_num = mirror_num; |
abad60c6 LB |
2253 | } |
2254 | ||
4c664611 | 2255 | rbio = alloc_rbio(fs_info, bioc, stripe_len); |
af8e2d1d | 2256 | if (IS_ERR(rbio)) { |
6e9606d2 | 2257 | if (generic_io) |
4c664611 | 2258 | btrfs_put_bioc(bioc); |
53b381b3 | 2259 | return PTR_ERR(rbio); |
af8e2d1d | 2260 | } |
53b381b3 | 2261 | |
1b94b556 | 2262 | rbio->operation = BTRFS_RBIO_READ_REBUILD; |
53b381b3 | 2263 | bio_list_add(&rbio->bio_list, bio); |
4f024f37 | 2264 | rbio->bio_list_bytes = bio->bi_iter.bi_size; |
53b381b3 DW |
2265 | |
2266 | rbio->faila = find_logical_bio_stripe(rbio, bio); | |
2267 | if (rbio->faila == -1) { | |
0b246afa | 2268 | btrfs_warn(fs_info, |
4c664611 | 2269 | "%s could not find the bad stripe in raid56 so that we cannot recover any more (bio has logical %llu len %llu, bioc has map_type %llu)", |
1201b58b | 2270 | __func__, bio->bi_iter.bi_sector << 9, |
4c664611 | 2271 | (u64)bio->bi_iter.bi_size, bioc->map_type); |
6e9606d2 | 2272 | if (generic_io) |
4c664611 | 2273 | btrfs_put_bioc(bioc); |
53b381b3 DW |
2274 | kfree(rbio); |
2275 | return -EIO; | |
2276 | } | |
2277 | ||
4245215d | 2278 | if (generic_io) { |
0b246afa | 2279 | btrfs_bio_counter_inc_noblocked(fs_info); |
4245215d MX |
2280 | rbio->generic_bio_cnt = 1; |
2281 | } else { | |
4c664611 | 2282 | btrfs_get_bioc(bioc); |
4245215d MX |
2283 | } |
2284 | ||
53b381b3 | 2285 | /* |
8810f751 LB |
2286 | * Loop retry: |
2287 | * for 'mirror == 2', reconstruct from all other stripes. | |
2288 | * for 'mirror_num > 2', select a stripe to fail on every retry. | |
53b381b3 | 2289 | */ |
8810f751 LB |
2290 | if (mirror_num > 2) { |
2291 | /* | |
2292 | * 'mirror == 3' is to fail the p stripe and | |
2293 | * reconstruct from the q stripe. 'mirror > 3' is to | |
2294 | * fail a data stripe and reconstruct from p+q stripe. | |
2295 | */ | |
2296 | rbio->failb = rbio->real_stripes - (mirror_num - 1); | |
2297 | ASSERT(rbio->failb > 0); | |
2298 | if (rbio->failb <= rbio->faila) | |
2299 | rbio->failb--; | |
2300 | } | |
53b381b3 DW |
2301 | |
2302 | ret = lock_stripe_add(rbio); | |
2303 | ||
2304 | /* | |
2305 | * __raid56_parity_recover will end the bio with | |
2306 | * any errors it hits. We don't want to return | |
2307 | * its error value up the stack because our caller | |
2308 | * will end up calling bio_endio with any nonzero | |
2309 | * return | |
2310 | */ | |
2311 | if (ret == 0) | |
2312 | __raid56_parity_recover(rbio); | |
2313 | /* | |
2314 | * our rbio has been added to the list of | |
2315 | * rbios that will be handled after the | |
2316 | * currently lock owner is done | |
2317 | */ | |
2318 | return 0; | |
2319 | ||
2320 | } | |
2321 | ||
2322 | static void rmw_work(struct btrfs_work *work) | |
2323 | { | |
2324 | struct btrfs_raid_bio *rbio; | |
2325 | ||
2326 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2327 | raid56_rmw_stripe(rbio); | |
2328 | } | |
2329 | ||
2330 | static void read_rebuild_work(struct btrfs_work *work) | |
2331 | { | |
2332 | struct btrfs_raid_bio *rbio; | |
2333 | ||
2334 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2335 | __raid56_parity_recover(rbio); | |
2336 | } | |
5a6ac9ea MX |
2337 | |
2338 | /* | |
2339 | * The following code is used to scrub/replace the parity stripe | |
2340 | * | |
4c664611 | 2341 | * Caller must have already increased bio_counter for getting @bioc. |
ae6529c3 | 2342 | * |
5a6ac9ea MX |
2343 | * Note: We need make sure all the pages that add into the scrub/replace |
2344 | * raid bio are correct and not be changed during the scrub/replace. That | |
2345 | * is those pages just hold metadata or file data with checksum. | |
2346 | */ | |
2347 | ||
6a258d72 QW |
2348 | struct btrfs_raid_bio *raid56_parity_alloc_scrub_rbio(struct bio *bio, |
2349 | struct btrfs_io_context *bioc, | |
cc353a8b | 2350 | u32 stripe_len, struct btrfs_device *scrub_dev, |
6a258d72 | 2351 | unsigned long *dbitmap, int stripe_nsectors) |
5a6ac9ea | 2352 | { |
6a258d72 | 2353 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
5a6ac9ea MX |
2354 | struct btrfs_raid_bio *rbio; |
2355 | int i; | |
2356 | ||
4c664611 | 2357 | rbio = alloc_rbio(fs_info, bioc, stripe_len); |
5a6ac9ea MX |
2358 | if (IS_ERR(rbio)) |
2359 | return NULL; | |
2360 | bio_list_add(&rbio->bio_list, bio); | |
2361 | /* | |
2362 | * This is a special bio which is used to hold the completion handler | |
2363 | * and make the scrub rbio is similar to the other types | |
2364 | */ | |
2365 | ASSERT(!bio->bi_iter.bi_size); | |
2366 | rbio->operation = BTRFS_RBIO_PARITY_SCRUB; | |
2367 | ||
9cd3a7eb | 2368 | /* |
4c664611 | 2369 | * After mapping bioc with BTRFS_MAP_WRITE, parities have been sorted |
9cd3a7eb LB |
2370 | * to the end position, so this search can start from the first parity |
2371 | * stripe. | |
2372 | */ | |
2373 | for (i = rbio->nr_data; i < rbio->real_stripes; i++) { | |
4c664611 | 2374 | if (bioc->stripes[i].dev == scrub_dev) { |
5a6ac9ea MX |
2375 | rbio->scrubp = i; |
2376 | break; | |
2377 | } | |
2378 | } | |
9cd3a7eb | 2379 | ASSERT(i < rbio->real_stripes); |
5a6ac9ea MX |
2380 | |
2381 | /* Now we just support the sectorsize equals to page size */ | |
0b246afa | 2382 | ASSERT(fs_info->sectorsize == PAGE_SIZE); |
5a6ac9ea MX |
2383 | ASSERT(rbio->stripe_npages == stripe_nsectors); |
2384 | bitmap_copy(rbio->dbitmap, dbitmap, stripe_nsectors); | |
2385 | ||
ae6529c3 | 2386 | /* |
4c664611 | 2387 | * We have already increased bio_counter when getting bioc, record it |
ae6529c3 QW |
2388 | * so we can free it at rbio_orig_end_io(). |
2389 | */ | |
2390 | rbio->generic_bio_cnt = 1; | |
2391 | ||
5a6ac9ea MX |
2392 | return rbio; |
2393 | } | |
2394 | ||
b4ee1782 OS |
2395 | /* Used for both parity scrub and missing. */ |
2396 | void raid56_add_scrub_pages(struct btrfs_raid_bio *rbio, struct page *page, | |
6346f6bf | 2397 | unsigned int pgoff, u64 logical) |
5a6ac9ea | 2398 | { |
6346f6bf | 2399 | const u32 sectorsize = rbio->bioc->fs_info->sectorsize; |
5a6ac9ea MX |
2400 | int stripe_offset; |
2401 | int index; | |
2402 | ||
4c664611 | 2403 | ASSERT(logical >= rbio->bioc->raid_map[0]); |
6346f6bf | 2404 | ASSERT(logical + sectorsize <= rbio->bioc->raid_map[0] + |
5a6ac9ea | 2405 | rbio->stripe_len * rbio->nr_data); |
4c664611 | 2406 | stripe_offset = (int)(logical - rbio->bioc->raid_map[0]); |
6346f6bf QW |
2407 | index = stripe_offset / sectorsize; |
2408 | rbio->bio_sectors[index].page = page; | |
2409 | rbio->bio_sectors[index].pgoff = pgoff; | |
5a6ac9ea MX |
2410 | } |
2411 | ||
2412 | /* | |
2413 | * We just scrub the parity that we have correct data on the same horizontal, | |
2414 | * so we needn't allocate all pages for all the stripes. | |
2415 | */ | |
2416 | static int alloc_rbio_essential_pages(struct btrfs_raid_bio *rbio) | |
2417 | { | |
2418 | int i; | |
2419 | int bit; | |
2420 | int index; | |
2421 | struct page *page; | |
2422 | ||
2423 | for_each_set_bit(bit, rbio->dbitmap, rbio->stripe_npages) { | |
2c8cdd6e | 2424 | for (i = 0; i < rbio->real_stripes; i++) { |
5a6ac9ea MX |
2425 | index = i * rbio->stripe_npages + bit; |
2426 | if (rbio->stripe_pages[index]) | |
2427 | continue; | |
2428 | ||
b0ee5e1e | 2429 | page = alloc_page(GFP_NOFS); |
5a6ac9ea MX |
2430 | if (!page) |
2431 | return -ENOMEM; | |
2432 | rbio->stripe_pages[index] = page; | |
5a6ac9ea MX |
2433 | } |
2434 | } | |
eb357060 | 2435 | index_stripe_sectors(rbio); |
5a6ac9ea MX |
2436 | return 0; |
2437 | } | |
2438 | ||
5a6ac9ea MX |
2439 | static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, |
2440 | int need_check) | |
2441 | { | |
4c664611 | 2442 | struct btrfs_io_context *bioc = rbio->bioc; |
46900662 | 2443 | const u32 sectorsize = bioc->fs_info->sectorsize; |
1389053e KC |
2444 | void **pointers = rbio->finish_pointers; |
2445 | unsigned long *pbitmap = rbio->finish_pbitmap; | |
5a6ac9ea MX |
2446 | int nr_data = rbio->nr_data; |
2447 | int stripe; | |
3e77605d | 2448 | int sectornr; |
c17af965 | 2449 | bool has_qstripe; |
46900662 QW |
2450 | struct sector_ptr p_sector = { 0 }; |
2451 | struct sector_ptr q_sector = { 0 }; | |
5a6ac9ea MX |
2452 | struct bio_list bio_list; |
2453 | struct bio *bio; | |
76035976 | 2454 | int is_replace = 0; |
5a6ac9ea MX |
2455 | int ret; |
2456 | ||
2457 | bio_list_init(&bio_list); | |
2458 | ||
c17af965 DS |
2459 | if (rbio->real_stripes - rbio->nr_data == 1) |
2460 | has_qstripe = false; | |
2461 | else if (rbio->real_stripes - rbio->nr_data == 2) | |
2462 | has_qstripe = true; | |
2463 | else | |
5a6ac9ea | 2464 | BUG(); |
5a6ac9ea | 2465 | |
4c664611 | 2466 | if (bioc->num_tgtdevs && bioc->tgtdev_map[rbio->scrubp]) { |
76035976 | 2467 | is_replace = 1; |
3e77605d | 2468 | bitmap_copy(pbitmap, rbio->dbitmap, rbio->stripe_nsectors); |
76035976 MX |
2469 | } |
2470 | ||
5a6ac9ea MX |
2471 | /* |
2472 | * Because the higher layers(scrubber) are unlikely to | |
2473 | * use this area of the disk again soon, so don't cache | |
2474 | * it. | |
2475 | */ | |
2476 | clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); | |
2477 | ||
2478 | if (!need_check) | |
2479 | goto writeback; | |
2480 | ||
46900662 QW |
2481 | p_sector.page = alloc_page(GFP_NOFS); |
2482 | if (!p_sector.page) | |
5a6ac9ea | 2483 | goto cleanup; |
46900662 QW |
2484 | p_sector.pgoff = 0; |
2485 | p_sector.uptodate = 1; | |
5a6ac9ea | 2486 | |
c17af965 | 2487 | if (has_qstripe) { |
d70cef0d | 2488 | /* RAID6, allocate and map temp space for the Q stripe */ |
46900662 QW |
2489 | q_sector.page = alloc_page(GFP_NOFS); |
2490 | if (!q_sector.page) { | |
2491 | __free_page(p_sector.page); | |
2492 | p_sector.page = NULL; | |
5a6ac9ea MX |
2493 | goto cleanup; |
2494 | } | |
46900662 QW |
2495 | q_sector.pgoff = 0; |
2496 | q_sector.uptodate = 1; | |
2497 | pointers[rbio->real_stripes - 1] = kmap_local_page(q_sector.page); | |
5a6ac9ea MX |
2498 | } |
2499 | ||
2500 | atomic_set(&rbio->error, 0); | |
2501 | ||
d70cef0d | 2502 | /* Map the parity stripe just once */ |
46900662 | 2503 | pointers[nr_data] = kmap_local_page(p_sector.page); |
d70cef0d | 2504 | |
3e77605d | 2505 | for_each_set_bit(sectornr, rbio->dbitmap, rbio->stripe_nsectors) { |
46900662 | 2506 | struct sector_ptr *sector; |
5a6ac9ea | 2507 | void *parity; |
46900662 | 2508 | |
5a6ac9ea MX |
2509 | /* first collect one page from each data stripe */ |
2510 | for (stripe = 0; stripe < nr_data; stripe++) { | |
46900662 QW |
2511 | sector = sector_in_rbio(rbio, stripe, sectornr, 0); |
2512 | pointers[stripe] = kmap_local_page(sector->page) + | |
2513 | sector->pgoff; | |
5a6ac9ea MX |
2514 | } |
2515 | ||
c17af965 | 2516 | if (has_qstripe) { |
d70cef0d | 2517 | /* RAID6, call the library function to fill in our P/Q */ |
46900662 | 2518 | raid6_call.gen_syndrome(rbio->real_stripes, sectorsize, |
5a6ac9ea MX |
2519 | pointers); |
2520 | } else { | |
2521 | /* raid5 */ | |
46900662 QW |
2522 | memcpy(pointers[nr_data], pointers[0], sectorsize); |
2523 | run_xor(pointers + 1, nr_data - 1, sectorsize); | |
5a6ac9ea MX |
2524 | } |
2525 | ||
01327610 | 2526 | /* Check scrubbing parity and repair it */ |
46900662 QW |
2527 | sector = rbio_stripe_sector(rbio, rbio->scrubp, sectornr); |
2528 | parity = kmap_local_page(sector->page) + sector->pgoff; | |
2529 | if (memcmp(parity, pointers[rbio->scrubp], sectorsize) != 0) | |
2530 | memcpy(parity, pointers[rbio->scrubp], sectorsize); | |
5a6ac9ea MX |
2531 | else |
2532 | /* Parity is right, needn't writeback */ | |
3e77605d | 2533 | bitmap_clear(rbio->dbitmap, sectornr, 1); |
58c1a35c | 2534 | kunmap_local(parity); |
5a6ac9ea | 2535 | |
94a0b58d IW |
2536 | for (stripe = nr_data - 1; stripe >= 0; stripe--) |
2537 | kunmap_local(pointers[stripe]); | |
5a6ac9ea MX |
2538 | } |
2539 | ||
94a0b58d | 2540 | kunmap_local(pointers[nr_data]); |
46900662 QW |
2541 | __free_page(p_sector.page); |
2542 | p_sector.page = NULL; | |
2543 | if (q_sector.page) { | |
94a0b58d | 2544 | kunmap_local(pointers[rbio->real_stripes - 1]); |
46900662 QW |
2545 | __free_page(q_sector.page); |
2546 | q_sector.page = NULL; | |
d70cef0d | 2547 | } |
5a6ac9ea MX |
2548 | |
2549 | writeback: | |
2550 | /* | |
2551 | * time to start writing. Make bios for everything from the | |
2552 | * higher layers (the bio_list in our rbio) and our p/q. Ignore | |
2553 | * everything else. | |
2554 | */ | |
3e77605d QW |
2555 | for_each_set_bit(sectornr, rbio->dbitmap, rbio->stripe_nsectors) { |
2556 | struct sector_ptr *sector; | |
5a6ac9ea | 2557 | |
3e77605d QW |
2558 | sector = rbio_stripe_sector(rbio, rbio->scrubp, sectornr); |
2559 | ret = rbio_add_io_sector(rbio, &bio_list, sector, rbio->scrubp, | |
2560 | sectornr, rbio->stripe_len, REQ_OP_WRITE); | |
5a6ac9ea MX |
2561 | if (ret) |
2562 | goto cleanup; | |
2563 | } | |
2564 | ||
76035976 MX |
2565 | if (!is_replace) |
2566 | goto submit_write; | |
2567 | ||
3e77605d QW |
2568 | for_each_set_bit(sectornr, pbitmap, rbio->stripe_nsectors) { |
2569 | struct sector_ptr *sector; | |
76035976 | 2570 | |
3e77605d QW |
2571 | sector = rbio_stripe_sector(rbio, rbio->scrubp, sectornr); |
2572 | ret = rbio_add_io_sector(rbio, &bio_list, sector, | |
4c664611 | 2573 | bioc->tgtdev_map[rbio->scrubp], |
3e77605d | 2574 | sectornr, rbio->stripe_len, REQ_OP_WRITE); |
76035976 MX |
2575 | if (ret) |
2576 | goto cleanup; | |
2577 | } | |
2578 | ||
2579 | submit_write: | |
5a6ac9ea MX |
2580 | nr_data = bio_list_size(&bio_list); |
2581 | if (!nr_data) { | |
2582 | /* Every parity is right */ | |
58efbc9f | 2583 | rbio_orig_end_io(rbio, BLK_STS_OK); |
5a6ac9ea MX |
2584 | return; |
2585 | } | |
2586 | ||
2587 | atomic_set(&rbio->stripes_pending, nr_data); | |
2588 | ||
bf28a605 | 2589 | while ((bio = bio_list_pop(&bio_list))) { |
a6111d11 | 2590 | bio->bi_end_io = raid_write_end_io; |
4e49ea4a MC |
2591 | |
2592 | submit_bio(bio); | |
5a6ac9ea MX |
2593 | } |
2594 | return; | |
2595 | ||
2596 | cleanup: | |
58efbc9f | 2597 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2598 | |
2599 | while ((bio = bio_list_pop(&bio_list))) | |
2600 | bio_put(bio); | |
5a6ac9ea MX |
2601 | } |
2602 | ||
2603 | static inline int is_data_stripe(struct btrfs_raid_bio *rbio, int stripe) | |
2604 | { | |
2605 | if (stripe >= 0 && stripe < rbio->nr_data) | |
2606 | return 1; | |
2607 | return 0; | |
2608 | } | |
2609 | ||
2610 | /* | |
2611 | * While we're doing the parity check and repair, we could have errors | |
2612 | * in reading pages off the disk. This checks for errors and if we're | |
2613 | * not able to read the page it'll trigger parity reconstruction. The | |
2614 | * parity scrub will be finished after we've reconstructed the failed | |
2615 | * stripes | |
2616 | */ | |
2617 | static void validate_rbio_for_parity_scrub(struct btrfs_raid_bio *rbio) | |
2618 | { | |
4c664611 | 2619 | if (atomic_read(&rbio->error) > rbio->bioc->max_errors) |
5a6ac9ea MX |
2620 | goto cleanup; |
2621 | ||
2622 | if (rbio->faila >= 0 || rbio->failb >= 0) { | |
2623 | int dfail = 0, failp = -1; | |
2624 | ||
2625 | if (is_data_stripe(rbio, rbio->faila)) | |
2626 | dfail++; | |
2627 | else if (is_parity_stripe(rbio->faila)) | |
2628 | failp = rbio->faila; | |
2629 | ||
2630 | if (is_data_stripe(rbio, rbio->failb)) | |
2631 | dfail++; | |
2632 | else if (is_parity_stripe(rbio->failb)) | |
2633 | failp = rbio->failb; | |
2634 | ||
2635 | /* | |
2636 | * Because we can not use a scrubbing parity to repair | |
2637 | * the data, so the capability of the repair is declined. | |
2638 | * (In the case of RAID5, we can not repair anything) | |
2639 | */ | |
4c664611 | 2640 | if (dfail > rbio->bioc->max_errors - 1) |
5a6ac9ea MX |
2641 | goto cleanup; |
2642 | ||
2643 | /* | |
2644 | * If all data is good, only parity is correctly, just | |
2645 | * repair the parity. | |
2646 | */ | |
2647 | if (dfail == 0) { | |
2648 | finish_parity_scrub(rbio, 0); | |
2649 | return; | |
2650 | } | |
2651 | ||
2652 | /* | |
2653 | * Here means we got one corrupted data stripe and one | |
2654 | * corrupted parity on RAID6, if the corrupted parity | |
01327610 | 2655 | * is scrubbing parity, luckily, use the other one to repair |
5a6ac9ea MX |
2656 | * the data, or we can not repair the data stripe. |
2657 | */ | |
2658 | if (failp != rbio->scrubp) | |
2659 | goto cleanup; | |
2660 | ||
2661 | __raid_recover_end_io(rbio); | |
2662 | } else { | |
2663 | finish_parity_scrub(rbio, 1); | |
2664 | } | |
2665 | return; | |
2666 | ||
2667 | cleanup: | |
58efbc9f | 2668 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
5a6ac9ea MX |
2669 | } |
2670 | ||
2671 | /* | |
2672 | * end io for the read phase of the rmw cycle. All the bios here are physical | |
2673 | * stripe bios we've read from the disk so we can recalculate the parity of the | |
2674 | * stripe. | |
2675 | * | |
2676 | * This will usually kick off finish_rmw once all the bios are read in, but it | |
2677 | * may trigger parity reconstruction if we had any errors along the way | |
2678 | */ | |
4246a0b6 | 2679 | static void raid56_parity_scrub_end_io(struct bio *bio) |
5a6ac9ea MX |
2680 | { |
2681 | struct btrfs_raid_bio *rbio = bio->bi_private; | |
2682 | ||
4e4cbee9 | 2683 | if (bio->bi_status) |
5a6ac9ea MX |
2684 | fail_bio_stripe(rbio, bio); |
2685 | else | |
5fdb7afc | 2686 | set_bio_pages_uptodate(rbio, bio); |
5a6ac9ea MX |
2687 | |
2688 | bio_put(bio); | |
2689 | ||
2690 | if (!atomic_dec_and_test(&rbio->stripes_pending)) | |
2691 | return; | |
2692 | ||
2693 | /* | |
2694 | * this will normally call finish_rmw to start our write | |
2695 | * but if there are any failed stripes we'll reconstruct | |
2696 | * from parity first | |
2697 | */ | |
2698 | validate_rbio_for_parity_scrub(rbio); | |
2699 | } | |
2700 | ||
2701 | static void raid56_parity_scrub_stripe(struct btrfs_raid_bio *rbio) | |
2702 | { | |
2703 | int bios_to_read = 0; | |
5a6ac9ea MX |
2704 | struct bio_list bio_list; |
2705 | int ret; | |
3e77605d | 2706 | int sectornr; |
5a6ac9ea MX |
2707 | int stripe; |
2708 | struct bio *bio; | |
2709 | ||
785884fc LB |
2710 | bio_list_init(&bio_list); |
2711 | ||
5a6ac9ea MX |
2712 | ret = alloc_rbio_essential_pages(rbio); |
2713 | if (ret) | |
2714 | goto cleanup; | |
2715 | ||
5a6ac9ea MX |
2716 | atomic_set(&rbio->error, 0); |
2717 | /* | |
2718 | * build a list of bios to read all the missing parts of this | |
2719 | * stripe | |
2720 | */ | |
2c8cdd6e | 2721 | for (stripe = 0; stripe < rbio->real_stripes; stripe++) { |
3e77605d QW |
2722 | for_each_set_bit(sectornr , rbio->dbitmap, rbio->stripe_nsectors) { |
2723 | struct sector_ptr *sector; | |
5a6ac9ea | 2724 | /* |
3e77605d QW |
2725 | * We want to find all the sectors missing from the |
2726 | * rbio and read them from the disk. If * sector_in_rbio() | |
2727 | * finds a sector in the bio list we don't need to read | |
2728 | * it off the stripe. | |
5a6ac9ea | 2729 | */ |
3e77605d QW |
2730 | sector = sector_in_rbio(rbio, stripe, sectornr, 1); |
2731 | if (sector) | |
5a6ac9ea MX |
2732 | continue; |
2733 | ||
3e77605d | 2734 | sector = rbio_stripe_sector(rbio, stripe, sectornr); |
5a6ac9ea | 2735 | /* |
3e77605d QW |
2736 | * The bio cache may have handed us an uptodate sector. |
2737 | * If so, be happy and use it. | |
5a6ac9ea | 2738 | */ |
3e77605d | 2739 | if (sector->uptodate) |
5a6ac9ea MX |
2740 | continue; |
2741 | ||
3e77605d QW |
2742 | ret = rbio_add_io_sector(rbio, &bio_list, sector, |
2743 | stripe, sectornr, rbio->stripe_len, | |
2744 | REQ_OP_READ); | |
5a6ac9ea MX |
2745 | if (ret) |
2746 | goto cleanup; | |
2747 | } | |
2748 | } | |
2749 | ||
2750 | bios_to_read = bio_list_size(&bio_list); | |
2751 | if (!bios_to_read) { | |
2752 | /* | |
2753 | * this can happen if others have merged with | |
2754 | * us, it means there is nothing left to read. | |
2755 | * But if there are missing devices it may not be | |
2756 | * safe to do the full stripe write yet. | |
2757 | */ | |
2758 | goto finish; | |
2759 | } | |
2760 | ||
2761 | /* | |
4c664611 QW |
2762 | * The bioc may be freed once we submit the last bio. Make sure not to |
2763 | * touch it after that. | |
5a6ac9ea MX |
2764 | */ |
2765 | atomic_set(&rbio->stripes_pending, bios_to_read); | |
bf28a605 | 2766 | while ((bio = bio_list_pop(&bio_list))) { |
5a6ac9ea MX |
2767 | bio->bi_end_io = raid56_parity_scrub_end_io; |
2768 | ||
6a258d72 | 2769 | btrfs_bio_wq_end_io(rbio->bioc->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); |
5a6ac9ea | 2770 | |
4e49ea4a | 2771 | submit_bio(bio); |
5a6ac9ea MX |
2772 | } |
2773 | /* the actual write will happen once the reads are done */ | |
2774 | return; | |
2775 | ||
2776 | cleanup: | |
58efbc9f | 2777 | rbio_orig_end_io(rbio, BLK_STS_IOERR); |
785884fc LB |
2778 | |
2779 | while ((bio = bio_list_pop(&bio_list))) | |
2780 | bio_put(bio); | |
2781 | ||
5a6ac9ea MX |
2782 | return; |
2783 | ||
2784 | finish: | |
2785 | validate_rbio_for_parity_scrub(rbio); | |
2786 | } | |
2787 | ||
2788 | static void scrub_parity_work(struct btrfs_work *work) | |
2789 | { | |
2790 | struct btrfs_raid_bio *rbio; | |
2791 | ||
2792 | rbio = container_of(work, struct btrfs_raid_bio, work); | |
2793 | raid56_parity_scrub_stripe(rbio); | |
2794 | } | |
2795 | ||
5a6ac9ea MX |
2796 | void raid56_parity_submit_scrub_rbio(struct btrfs_raid_bio *rbio) |
2797 | { | |
2798 | if (!lock_stripe_add(rbio)) | |
a81b747d | 2799 | start_async_work(rbio, scrub_parity_work); |
5a6ac9ea | 2800 | } |
b4ee1782 OS |
2801 | |
2802 | /* The following code is used for dev replace of a missing RAID 5/6 device. */ | |
2803 | ||
2804 | struct btrfs_raid_bio * | |
6a258d72 QW |
2805 | raid56_alloc_missing_rbio(struct bio *bio, struct btrfs_io_context *bioc, |
2806 | u64 length) | |
b4ee1782 | 2807 | { |
6a258d72 | 2808 | struct btrfs_fs_info *fs_info = bioc->fs_info; |
b4ee1782 OS |
2809 | struct btrfs_raid_bio *rbio; |
2810 | ||
4c664611 | 2811 | rbio = alloc_rbio(fs_info, bioc, length); |
b4ee1782 OS |
2812 | if (IS_ERR(rbio)) |
2813 | return NULL; | |
2814 | ||
2815 | rbio->operation = BTRFS_RBIO_REBUILD_MISSING; | |
2816 | bio_list_add(&rbio->bio_list, bio); | |
2817 | /* | |
2818 | * This is a special bio which is used to hold the completion handler | |
2819 | * and make the scrub rbio is similar to the other types | |
2820 | */ | |
2821 | ASSERT(!bio->bi_iter.bi_size); | |
2822 | ||
2823 | rbio->faila = find_logical_bio_stripe(rbio, bio); | |
2824 | if (rbio->faila == -1) { | |
2825 | BUG(); | |
2826 | kfree(rbio); | |
2827 | return NULL; | |
2828 | } | |
2829 | ||
ae6529c3 | 2830 | /* |
4c664611 | 2831 | * When we get bioc, we have already increased bio_counter, record it |
ae6529c3 QW |
2832 | * so we can free it at rbio_orig_end_io() |
2833 | */ | |
2834 | rbio->generic_bio_cnt = 1; | |
2835 | ||
b4ee1782 OS |
2836 | return rbio; |
2837 | } | |
2838 | ||
b4ee1782 OS |
2839 | void raid56_submit_missing_rbio(struct btrfs_raid_bio *rbio) |
2840 | { | |
2841 | if (!lock_stripe_add(rbio)) | |
e66d8d5a | 2842 | start_async_work(rbio, read_rebuild_work); |
b4ee1782 | 2843 | } |