Loading...
Note: File does not exist in v3.1.
1/*
2 * Copyright (C) 2012 Fusion-io All rights reserved.
3 * Copyright (C) 2012 Intel Corp. All rights reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public
7 * License v2 as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public
15 * License along with this program; if not, write to the
16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
17 * Boston, MA 021110-1307, USA.
18 */
19#include <linux/sched.h>
20#include <linux/wait.h>
21#include <linux/bio.h>
22#include <linux/slab.h>
23#include <linux/buffer_head.h>
24#include <linux/blkdev.h>
25#include <linux/random.h>
26#include <linux/iocontext.h>
27#include <linux/capability.h>
28#include <linux/ratelimit.h>
29#include <linux/kthread.h>
30#include <linux/raid/pq.h>
31#include <linux/hash.h>
32#include <linux/list_sort.h>
33#include <linux/raid/xor.h>
34#include <linux/vmalloc.h>
35#include <asm/div64.h>
36#include "ctree.h"
37#include "extent_map.h"
38#include "disk-io.h"
39#include "transaction.h"
40#include "print-tree.h"
41#include "volumes.h"
42#include "raid56.h"
43#include "async-thread.h"
44#include "check-integrity.h"
45#include "rcu-string.h"
46
47/* set when additional merges to this rbio are not allowed */
48#define RBIO_RMW_LOCKED_BIT 1
49
50/*
51 * set when this rbio is sitting in the hash, but it is just a cache
52 * of past RMW
53 */
54#define RBIO_CACHE_BIT 2
55
56/*
57 * set when it is safe to trust the stripe_pages for caching
58 */
59#define RBIO_CACHE_READY_BIT 3
60
61
62#define RBIO_CACHE_SIZE 1024
63
64struct btrfs_raid_bio {
65 struct btrfs_fs_info *fs_info;
66 struct btrfs_bio *bbio;
67
68 /*
69 * logical block numbers for the start of each stripe
70 * The last one or two are p/q. These are sorted,
71 * so raid_map[0] is the start of our full stripe
72 */
73 u64 *raid_map;
74
75 /* while we're doing rmw on a stripe
76 * we put it into a hash table so we can
77 * lock the stripe and merge more rbios
78 * into it.
79 */
80 struct list_head hash_list;
81
82 /*
83 * LRU list for the stripe cache
84 */
85 struct list_head stripe_cache;
86
87 /*
88 * for scheduling work in the helper threads
89 */
90 struct btrfs_work work;
91
92 /*
93 * bio list and bio_list_lock are used
94 * to add more bios into the stripe
95 * in hopes of avoiding the full rmw
96 */
97 struct bio_list bio_list;
98 spinlock_t bio_list_lock;
99
100 /* also protected by the bio_list_lock, the
101 * plug list is used by the plugging code
102 * to collect partial bios while plugged. The
103 * stripe locking code also uses it to hand off
104 * the stripe lock to the next pending IO
105 */
106 struct list_head plug_list;
107
108 /*
109 * flags that tell us if it is safe to
110 * merge with this bio
111 */
112 unsigned long flags;
113
114 /* size of each individual stripe on disk */
115 int stripe_len;
116
117 /* number of data stripes (no p/q) */
118 int nr_data;
119
120 /*
121 * set if we're doing a parity rebuild
122 * for a read from higher up, which is handled
123 * differently from a parity rebuild as part of
124 * rmw
125 */
126 int read_rebuild;
127
128 /* first bad stripe */
129 int faila;
130
131 /* second bad stripe (for raid6 use) */
132 int failb;
133
134 /*
135 * number of pages needed to represent the full
136 * stripe
137 */
138 int nr_pages;
139
140 /*
141 * size of all the bios in the bio_list. This
142 * helps us decide if the rbio maps to a full
143 * stripe or not
144 */
145 int bio_list_bytes;
146
147 atomic_t refs;
148
149 /*
150 * these are two arrays of pointers. We allocate the
151 * rbio big enough to hold them both and setup their
152 * locations when the rbio is allocated
153 */
154
155 /* pointers to pages that we allocated for
156 * reading/writing stripes directly from the disk (including P/Q)
157 */
158 struct page **stripe_pages;
159
160 /*
161 * pointers to the pages in the bio_list. Stored
162 * here for faster lookup
163 */
164 struct page **bio_pages;
165};
166
167static int __raid56_parity_recover(struct btrfs_raid_bio *rbio);
168static noinline void finish_rmw(struct btrfs_raid_bio *rbio);
169static void rmw_work(struct btrfs_work *work);
170static void read_rebuild_work(struct btrfs_work *work);
171static void async_rmw_stripe(struct btrfs_raid_bio *rbio);
172static void async_read_rebuild(struct btrfs_raid_bio *rbio);
173static int fail_bio_stripe(struct btrfs_raid_bio *rbio, struct bio *bio);
174static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed);
175static void __free_raid_bio(struct btrfs_raid_bio *rbio);
176static void index_rbio_pages(struct btrfs_raid_bio *rbio);
177static int alloc_rbio_pages(struct btrfs_raid_bio *rbio);
178
179/*
180 * the stripe hash table is used for locking, and to collect
181 * bios in hopes of making a full stripe
182 */
183int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info)
184{
185 struct btrfs_stripe_hash_table *table;
186 struct btrfs_stripe_hash_table *x;
187 struct btrfs_stripe_hash *cur;
188 struct btrfs_stripe_hash *h;
189 int num_entries = 1 << BTRFS_STRIPE_HASH_TABLE_BITS;
190 int i;
191 int table_size;
192
193 if (info->stripe_hash_table)
194 return 0;
195
196 /*
197 * The table is large, starting with order 4 and can go as high as
198 * order 7 in case lock debugging is turned on.
199 *
200 * Try harder to allocate and fallback to vmalloc to lower the chance
201 * of a failing mount.
202 */
203 table_size = sizeof(*table) + sizeof(*h) * num_entries;
204 table = kzalloc(table_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
205 if (!table) {
206 table = vzalloc(table_size);
207 if (!table)
208 return -ENOMEM;
209 }
210
211 spin_lock_init(&table->cache_lock);
212 INIT_LIST_HEAD(&table->stripe_cache);
213
214 h = table->table;
215
216 for (i = 0; i < num_entries; i++) {
217 cur = h + i;
218 INIT_LIST_HEAD(&cur->hash_list);
219 spin_lock_init(&cur->lock);
220 init_waitqueue_head(&cur->wait);
221 }
222
223 x = cmpxchg(&info->stripe_hash_table, NULL, table);
224 if (x) {
225 if (is_vmalloc_addr(x))
226 vfree(x);
227 else
228 kfree(x);
229 }
230 return 0;
231}
232
233/*
234 * caching an rbio means to copy anything from the
235 * bio_pages array into the stripe_pages array. We
236 * use the page uptodate bit in the stripe cache array
237 * to indicate if it has valid data
238 *
239 * once the caching is done, we set the cache ready
240 * bit.
241 */
242static void cache_rbio_pages(struct btrfs_raid_bio *rbio)
243{
244 int i;
245 char *s;
246 char *d;
247 int ret;
248
249 ret = alloc_rbio_pages(rbio);
250 if (ret)
251 return;
252
253 for (i = 0; i < rbio->nr_pages; i++) {
254 if (!rbio->bio_pages[i])
255 continue;
256
257 s = kmap(rbio->bio_pages[i]);
258 d = kmap(rbio->stripe_pages[i]);
259
260 memcpy(d, s, PAGE_CACHE_SIZE);
261
262 kunmap(rbio->bio_pages[i]);
263 kunmap(rbio->stripe_pages[i]);
264 SetPageUptodate(rbio->stripe_pages[i]);
265 }
266 set_bit(RBIO_CACHE_READY_BIT, &rbio->flags);
267}
268
269/*
270 * we hash on the first logical address of the stripe
271 */
272static int rbio_bucket(struct btrfs_raid_bio *rbio)
273{
274 u64 num = rbio->raid_map[0];
275
276 /*
277 * we shift down quite a bit. We're using byte
278 * addressing, and most of the lower bits are zeros.
279 * This tends to upset hash_64, and it consistently
280 * returns just one or two different values.
281 *
282 * shifting off the lower bits fixes things.
283 */
284 return hash_64(num >> 16, BTRFS_STRIPE_HASH_TABLE_BITS);
285}
286
287/*
288 * stealing an rbio means taking all the uptodate pages from the stripe
289 * array in the source rbio and putting them into the destination rbio
290 */
291static void steal_rbio(struct btrfs_raid_bio *src, struct btrfs_raid_bio *dest)
292{
293 int i;
294 struct page *s;
295 struct page *d;
296
297 if (!test_bit(RBIO_CACHE_READY_BIT, &src->flags))
298 return;
299
300 for (i = 0; i < dest->nr_pages; i++) {
301 s = src->stripe_pages[i];
302 if (!s || !PageUptodate(s)) {
303 continue;
304 }
305
306 d = dest->stripe_pages[i];
307 if (d)
308 __free_page(d);
309
310 dest->stripe_pages[i] = s;
311 src->stripe_pages[i] = NULL;
312 }
313}
314
315/*
316 * merging means we take the bio_list from the victim and
317 * splice it into the destination. The victim should
318 * be discarded afterwards.
319 *
320 * must be called with dest->rbio_list_lock held
321 */
322static void merge_rbio(struct btrfs_raid_bio *dest,
323 struct btrfs_raid_bio *victim)
324{
325 bio_list_merge(&dest->bio_list, &victim->bio_list);
326 dest->bio_list_bytes += victim->bio_list_bytes;
327 bio_list_init(&victim->bio_list);
328}
329
330/*
331 * used to prune items that are in the cache. The caller
332 * must hold the hash table lock.
333 */
334static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio)
335{
336 int bucket = rbio_bucket(rbio);
337 struct btrfs_stripe_hash_table *table;
338 struct btrfs_stripe_hash *h;
339 int freeit = 0;
340
341 /*
342 * check the bit again under the hash table lock.
343 */
344 if (!test_bit(RBIO_CACHE_BIT, &rbio->flags))
345 return;
346
347 table = rbio->fs_info->stripe_hash_table;
348 h = table->table + bucket;
349
350 /* hold the lock for the bucket because we may be
351 * removing it from the hash table
352 */
353 spin_lock(&h->lock);
354
355 /*
356 * hold the lock for the bio list because we need
357 * to make sure the bio list is empty
358 */
359 spin_lock(&rbio->bio_list_lock);
360
361 if (test_and_clear_bit(RBIO_CACHE_BIT, &rbio->flags)) {
362 list_del_init(&rbio->stripe_cache);
363 table->cache_size -= 1;
364 freeit = 1;
365
366 /* if the bio list isn't empty, this rbio is
367 * still involved in an IO. We take it out
368 * of the cache list, and drop the ref that
369 * was held for the list.
370 *
371 * If the bio_list was empty, we also remove
372 * the rbio from the hash_table, and drop
373 * the corresponding ref
374 */
375 if (bio_list_empty(&rbio->bio_list)) {
376 if (!list_empty(&rbio->hash_list)) {
377 list_del_init(&rbio->hash_list);
378 atomic_dec(&rbio->refs);
379 BUG_ON(!list_empty(&rbio->plug_list));
380 }
381 }
382 }
383
384 spin_unlock(&rbio->bio_list_lock);
385 spin_unlock(&h->lock);
386
387 if (freeit)
388 __free_raid_bio(rbio);
389}
390
391/*
392 * prune a given rbio from the cache
393 */
394static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio)
395{
396 struct btrfs_stripe_hash_table *table;
397 unsigned long flags;
398
399 if (!test_bit(RBIO_CACHE_BIT, &rbio->flags))
400 return;
401
402 table = rbio->fs_info->stripe_hash_table;
403
404 spin_lock_irqsave(&table->cache_lock, flags);
405 __remove_rbio_from_cache(rbio);
406 spin_unlock_irqrestore(&table->cache_lock, flags);
407}
408
409/*
410 * remove everything in the cache
411 */
412static void btrfs_clear_rbio_cache(struct btrfs_fs_info *info)
413{
414 struct btrfs_stripe_hash_table *table;
415 unsigned long flags;
416 struct btrfs_raid_bio *rbio;
417
418 table = info->stripe_hash_table;
419
420 spin_lock_irqsave(&table->cache_lock, flags);
421 while (!list_empty(&table->stripe_cache)) {
422 rbio = list_entry(table->stripe_cache.next,
423 struct btrfs_raid_bio,
424 stripe_cache);
425 __remove_rbio_from_cache(rbio);
426 }
427 spin_unlock_irqrestore(&table->cache_lock, flags);
428}
429
430/*
431 * remove all cached entries and free the hash table
432 * used by unmount
433 */
434void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info)
435{
436 if (!info->stripe_hash_table)
437 return;
438 btrfs_clear_rbio_cache(info);
439 if (is_vmalloc_addr(info->stripe_hash_table))
440 vfree(info->stripe_hash_table);
441 else
442 kfree(info->stripe_hash_table);
443 info->stripe_hash_table = NULL;
444}
445
446/*
447 * insert an rbio into the stripe cache. It
448 * must have already been prepared by calling
449 * cache_rbio_pages
450 *
451 * If this rbio was already cached, it gets
452 * moved to the front of the lru.
453 *
454 * If the size of the rbio cache is too big, we
455 * prune an item.
456 */
457static void cache_rbio(struct btrfs_raid_bio *rbio)
458{
459 struct btrfs_stripe_hash_table *table;
460 unsigned long flags;
461
462 if (!test_bit(RBIO_CACHE_READY_BIT, &rbio->flags))
463 return;
464
465 table = rbio->fs_info->stripe_hash_table;
466
467 spin_lock_irqsave(&table->cache_lock, flags);
468 spin_lock(&rbio->bio_list_lock);
469
470 /* bump our ref if we were not in the list before */
471 if (!test_and_set_bit(RBIO_CACHE_BIT, &rbio->flags))
472 atomic_inc(&rbio->refs);
473
474 if (!list_empty(&rbio->stripe_cache)){
475 list_move(&rbio->stripe_cache, &table->stripe_cache);
476 } else {
477 list_add(&rbio->stripe_cache, &table->stripe_cache);
478 table->cache_size += 1;
479 }
480
481 spin_unlock(&rbio->bio_list_lock);
482
483 if (table->cache_size > RBIO_CACHE_SIZE) {
484 struct btrfs_raid_bio *found;
485
486 found = list_entry(table->stripe_cache.prev,
487 struct btrfs_raid_bio,
488 stripe_cache);
489
490 if (found != rbio)
491 __remove_rbio_from_cache(found);
492 }
493
494 spin_unlock_irqrestore(&table->cache_lock, flags);
495 return;
496}
497
498/*
499 * helper function to run the xor_blocks api. It is only
500 * able to do MAX_XOR_BLOCKS at a time, so we need to
501 * loop through.
502 */
503static void run_xor(void **pages, int src_cnt, ssize_t len)
504{
505 int src_off = 0;
506 int xor_src_cnt = 0;
507 void *dest = pages[src_cnt];
508
509 while(src_cnt > 0) {
510 xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS);
511 xor_blocks(xor_src_cnt, len, dest, pages + src_off);
512
513 src_cnt -= xor_src_cnt;
514 src_off += xor_src_cnt;
515 }
516}
517
518/*
519 * returns true if the bio list inside this rbio
520 * covers an entire stripe (no rmw required).
521 * Must be called with the bio list lock held, or
522 * at a time when you know it is impossible to add
523 * new bios into the list
524 */
525static int __rbio_is_full(struct btrfs_raid_bio *rbio)
526{
527 unsigned long size = rbio->bio_list_bytes;
528 int ret = 1;
529
530 if (size != rbio->nr_data * rbio->stripe_len)
531 ret = 0;
532
533 BUG_ON(size > rbio->nr_data * rbio->stripe_len);
534 return ret;
535}
536
537static int rbio_is_full(struct btrfs_raid_bio *rbio)
538{
539 unsigned long flags;
540 int ret;
541
542 spin_lock_irqsave(&rbio->bio_list_lock, flags);
543 ret = __rbio_is_full(rbio);
544 spin_unlock_irqrestore(&rbio->bio_list_lock, flags);
545 return ret;
546}
547
548/*
549 * returns 1 if it is safe to merge two rbios together.
550 * The merging is safe if the two rbios correspond to
551 * the same stripe and if they are both going in the same
552 * direction (read vs write), and if neither one is
553 * locked for final IO
554 *
555 * The caller is responsible for locking such that
556 * rmw_locked is safe to test
557 */
558static int rbio_can_merge(struct btrfs_raid_bio *last,
559 struct btrfs_raid_bio *cur)
560{
561 if (test_bit(RBIO_RMW_LOCKED_BIT, &last->flags) ||
562 test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags))
563 return 0;
564
565 /*
566 * we can't merge with cached rbios, since the
567 * idea is that when we merge the destination
568 * rbio is going to run our IO for us. We can
569 * steal from cached rbio's though, other functions
570 * handle that.
571 */
572 if (test_bit(RBIO_CACHE_BIT, &last->flags) ||
573 test_bit(RBIO_CACHE_BIT, &cur->flags))
574 return 0;
575
576 if (last->raid_map[0] !=
577 cur->raid_map[0])
578 return 0;
579
580 /* reads can't merge with writes */
581 if (last->read_rebuild !=
582 cur->read_rebuild) {
583 return 0;
584 }
585
586 return 1;
587}
588
589/*
590 * helper to index into the pstripe
591 */
592static struct page *rbio_pstripe_page(struct btrfs_raid_bio *rbio, int index)
593{
594 index += (rbio->nr_data * rbio->stripe_len) >> PAGE_CACHE_SHIFT;
595 return rbio->stripe_pages[index];
596}
597
598/*
599 * helper to index into the qstripe, returns null
600 * if there is no qstripe
601 */
602static struct page *rbio_qstripe_page(struct btrfs_raid_bio *rbio, int index)
603{
604 if (rbio->nr_data + 1 == rbio->bbio->num_stripes)
605 return NULL;
606
607 index += ((rbio->nr_data + 1) * rbio->stripe_len) >>
608 PAGE_CACHE_SHIFT;
609 return rbio->stripe_pages[index];
610}
611
612/*
613 * The first stripe in the table for a logical address
614 * has the lock. rbios are added in one of three ways:
615 *
616 * 1) Nobody has the stripe locked yet. The rbio is given
617 * the lock and 0 is returned. The caller must start the IO
618 * themselves.
619 *
620 * 2) Someone has the stripe locked, but we're able to merge
621 * with the lock owner. The rbio is freed and the IO will
622 * start automatically along with the existing rbio. 1 is returned.
623 *
624 * 3) Someone has the stripe locked, but we're not able to merge.
625 * The rbio is added to the lock owner's plug list, or merged into
626 * an rbio already on the plug list. When the lock owner unlocks,
627 * the next rbio on the list is run and the IO is started automatically.
628 * 1 is returned
629 *
630 * If we return 0, the caller still owns the rbio and must continue with
631 * IO submission. If we return 1, the caller must assume the rbio has
632 * already been freed.
633 */
634static noinline int lock_stripe_add(struct btrfs_raid_bio *rbio)
635{
636 int bucket = rbio_bucket(rbio);
637 struct btrfs_stripe_hash *h = rbio->fs_info->stripe_hash_table->table + bucket;
638 struct btrfs_raid_bio *cur;
639 struct btrfs_raid_bio *pending;
640 unsigned long flags;
641 DEFINE_WAIT(wait);
642 struct btrfs_raid_bio *freeit = NULL;
643 struct btrfs_raid_bio *cache_drop = NULL;
644 int ret = 0;
645 int walk = 0;
646
647 spin_lock_irqsave(&h->lock, flags);
648 list_for_each_entry(cur, &h->hash_list, hash_list) {
649 walk++;
650 if (cur->raid_map[0] == rbio->raid_map[0]) {
651 spin_lock(&cur->bio_list_lock);
652
653 /* can we steal this cached rbio's pages? */
654 if (bio_list_empty(&cur->bio_list) &&
655 list_empty(&cur->plug_list) &&
656 test_bit(RBIO_CACHE_BIT, &cur->flags) &&
657 !test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) {
658 list_del_init(&cur->hash_list);
659 atomic_dec(&cur->refs);
660
661 steal_rbio(cur, rbio);
662 cache_drop = cur;
663 spin_unlock(&cur->bio_list_lock);
664
665 goto lockit;
666 }
667
668 /* can we merge into the lock owner? */
669 if (rbio_can_merge(cur, rbio)) {
670 merge_rbio(cur, rbio);
671 spin_unlock(&cur->bio_list_lock);
672 freeit = rbio;
673 ret = 1;
674 goto out;
675 }
676
677
678 /*
679 * we couldn't merge with the running
680 * rbio, see if we can merge with the
681 * pending ones. We don't have to
682 * check for rmw_locked because there
683 * is no way they are inside finish_rmw
684 * right now
685 */
686 list_for_each_entry(pending, &cur->plug_list,
687 plug_list) {
688 if (rbio_can_merge(pending, rbio)) {
689 merge_rbio(pending, rbio);
690 spin_unlock(&cur->bio_list_lock);
691 freeit = rbio;
692 ret = 1;
693 goto out;
694 }
695 }
696
697 /* no merging, put us on the tail of the plug list,
698 * our rbio will be started with the currently
699 * running rbio unlocks
700 */
701 list_add_tail(&rbio->plug_list, &cur->plug_list);
702 spin_unlock(&cur->bio_list_lock);
703 ret = 1;
704 goto out;
705 }
706 }
707lockit:
708 atomic_inc(&rbio->refs);
709 list_add(&rbio->hash_list, &h->hash_list);
710out:
711 spin_unlock_irqrestore(&h->lock, flags);
712 if (cache_drop)
713 remove_rbio_from_cache(cache_drop);
714 if (freeit)
715 __free_raid_bio(freeit);
716 return ret;
717}
718
719/*
720 * called as rmw or parity rebuild is completed. If the plug list has more
721 * rbios waiting for this stripe, the next one on the list will be started
722 */
723static noinline void unlock_stripe(struct btrfs_raid_bio *rbio)
724{
725 int bucket;
726 struct btrfs_stripe_hash *h;
727 unsigned long flags;
728 int keep_cache = 0;
729
730 bucket = rbio_bucket(rbio);
731 h = rbio->fs_info->stripe_hash_table->table + bucket;
732
733 if (list_empty(&rbio->plug_list))
734 cache_rbio(rbio);
735
736 spin_lock_irqsave(&h->lock, flags);
737 spin_lock(&rbio->bio_list_lock);
738
739 if (!list_empty(&rbio->hash_list)) {
740 /*
741 * if we're still cached and there is no other IO
742 * to perform, just leave this rbio here for others
743 * to steal from later
744 */
745 if (list_empty(&rbio->plug_list) &&
746 test_bit(RBIO_CACHE_BIT, &rbio->flags)) {
747 keep_cache = 1;
748 clear_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags);
749 BUG_ON(!bio_list_empty(&rbio->bio_list));
750 goto done;
751 }
752
753 list_del_init(&rbio->hash_list);
754 atomic_dec(&rbio->refs);
755
756 /*
757 * we use the plug list to hold all the rbios
758 * waiting for the chance to lock this stripe.
759 * hand the lock over to one of them.
760 */
761 if (!list_empty(&rbio->plug_list)) {
762 struct btrfs_raid_bio *next;
763 struct list_head *head = rbio->plug_list.next;
764
765 next = list_entry(head, struct btrfs_raid_bio,
766 plug_list);
767
768 list_del_init(&rbio->plug_list);
769
770 list_add(&next->hash_list, &h->hash_list);
771 atomic_inc(&next->refs);
772 spin_unlock(&rbio->bio_list_lock);
773 spin_unlock_irqrestore(&h->lock, flags);
774
775 if (next->read_rebuild)
776 async_read_rebuild(next);
777 else {
778 steal_rbio(rbio, next);
779 async_rmw_stripe(next);
780 }
781
782 goto done_nolock;
783 } else if (waitqueue_active(&h->wait)) {
784 spin_unlock(&rbio->bio_list_lock);
785 spin_unlock_irqrestore(&h->lock, flags);
786 wake_up(&h->wait);
787 goto done_nolock;
788 }
789 }
790done:
791 spin_unlock(&rbio->bio_list_lock);
792 spin_unlock_irqrestore(&h->lock, flags);
793
794done_nolock:
795 if (!keep_cache)
796 remove_rbio_from_cache(rbio);
797}
798
799static void __free_raid_bio(struct btrfs_raid_bio *rbio)
800{
801 int i;
802
803 WARN_ON(atomic_read(&rbio->refs) < 0);
804 if (!atomic_dec_and_test(&rbio->refs))
805 return;
806
807 WARN_ON(!list_empty(&rbio->stripe_cache));
808 WARN_ON(!list_empty(&rbio->hash_list));
809 WARN_ON(!bio_list_empty(&rbio->bio_list));
810
811 for (i = 0; i < rbio->nr_pages; i++) {
812 if (rbio->stripe_pages[i]) {
813 __free_page(rbio->stripe_pages[i]);
814 rbio->stripe_pages[i] = NULL;
815 }
816 }
817 kfree(rbio->raid_map);
818 kfree(rbio->bbio);
819 kfree(rbio);
820}
821
822static void free_raid_bio(struct btrfs_raid_bio *rbio)
823{
824 unlock_stripe(rbio);
825 __free_raid_bio(rbio);
826}
827
828/*
829 * this frees the rbio and runs through all the bios in the
830 * bio_list and calls end_io on them
831 */
832static void rbio_orig_end_io(struct btrfs_raid_bio *rbio, int err, int uptodate)
833{
834 struct bio *cur = bio_list_get(&rbio->bio_list);
835 struct bio *next;
836 free_raid_bio(rbio);
837
838 while (cur) {
839 next = cur->bi_next;
840 cur->bi_next = NULL;
841 if (uptodate)
842 set_bit(BIO_UPTODATE, &cur->bi_flags);
843 bio_endio(cur, err);
844 cur = next;
845 }
846}
847
848/*
849 * end io function used by finish_rmw. When we finally
850 * get here, we've written a full stripe
851 */
852static void raid_write_end_io(struct bio *bio, int err)
853{
854 struct btrfs_raid_bio *rbio = bio->bi_private;
855
856 if (err)
857 fail_bio_stripe(rbio, bio);
858
859 bio_put(bio);
860
861 if (!atomic_dec_and_test(&rbio->bbio->stripes_pending))
862 return;
863
864 err = 0;
865
866 /* OK, we have read all the stripes we need to. */
867 if (atomic_read(&rbio->bbio->error) > rbio->bbio->max_errors)
868 err = -EIO;
869
870 rbio_orig_end_io(rbio, err, 0);
871 return;
872}
873
874/*
875 * the read/modify/write code wants to use the original bio for
876 * any pages it included, and then use the rbio for everything
877 * else. This function decides if a given index (stripe number)
878 * and page number in that stripe fall inside the original bio
879 * or the rbio.
880 *
881 * if you set bio_list_only, you'll get a NULL back for any ranges
882 * that are outside the bio_list
883 *
884 * This doesn't take any refs on anything, you get a bare page pointer
885 * and the caller must bump refs as required.
886 *
887 * You must call index_rbio_pages once before you can trust
888 * the answers from this function.
889 */
890static struct page *page_in_rbio(struct btrfs_raid_bio *rbio,
891 int index, int pagenr, int bio_list_only)
892{
893 int chunk_page;
894 struct page *p = NULL;
895
896 chunk_page = index * (rbio->stripe_len >> PAGE_SHIFT) + pagenr;
897
898 spin_lock_irq(&rbio->bio_list_lock);
899 p = rbio->bio_pages[chunk_page];
900 spin_unlock_irq(&rbio->bio_list_lock);
901
902 if (p || bio_list_only)
903 return p;
904
905 return rbio->stripe_pages[chunk_page];
906}
907
908/*
909 * number of pages we need for the entire stripe across all the
910 * drives
911 */
912static unsigned long rbio_nr_pages(unsigned long stripe_len, int nr_stripes)
913{
914 unsigned long nr = stripe_len * nr_stripes;
915 return (nr + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
916}
917
918/*
919 * allocation and initial setup for the btrfs_raid_bio. Not
920 * this does not allocate any pages for rbio->pages.
921 */
922static struct btrfs_raid_bio *alloc_rbio(struct btrfs_root *root,
923 struct btrfs_bio *bbio, u64 *raid_map,
924 u64 stripe_len)
925{
926 struct btrfs_raid_bio *rbio;
927 int nr_data = 0;
928 int num_pages = rbio_nr_pages(stripe_len, bbio->num_stripes);
929 void *p;
930
931 rbio = kzalloc(sizeof(*rbio) + num_pages * sizeof(struct page *) * 2,
932 GFP_NOFS);
933 if (!rbio) {
934 kfree(raid_map);
935 kfree(bbio);
936 return ERR_PTR(-ENOMEM);
937 }
938
939 bio_list_init(&rbio->bio_list);
940 INIT_LIST_HEAD(&rbio->plug_list);
941 spin_lock_init(&rbio->bio_list_lock);
942 INIT_LIST_HEAD(&rbio->stripe_cache);
943 INIT_LIST_HEAD(&rbio->hash_list);
944 rbio->bbio = bbio;
945 rbio->raid_map = raid_map;
946 rbio->fs_info = root->fs_info;
947 rbio->stripe_len = stripe_len;
948 rbio->nr_pages = num_pages;
949 rbio->faila = -1;
950 rbio->failb = -1;
951 atomic_set(&rbio->refs, 1);
952
953 /*
954 * the stripe_pages and bio_pages array point to the extra
955 * memory we allocated past the end of the rbio
956 */
957 p = rbio + 1;
958 rbio->stripe_pages = p;
959 rbio->bio_pages = p + sizeof(struct page *) * num_pages;
960
961 if (raid_map[bbio->num_stripes - 1] == RAID6_Q_STRIPE)
962 nr_data = bbio->num_stripes - 2;
963 else
964 nr_data = bbio->num_stripes - 1;
965
966 rbio->nr_data = nr_data;
967 return rbio;
968}
969
970/* allocate pages for all the stripes in the bio, including parity */
971static int alloc_rbio_pages(struct btrfs_raid_bio *rbio)
972{
973 int i;
974 struct page *page;
975
976 for (i = 0; i < rbio->nr_pages; i++) {
977 if (rbio->stripe_pages[i])
978 continue;
979 page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
980 if (!page)
981 return -ENOMEM;
982 rbio->stripe_pages[i] = page;
983 ClearPageUptodate(page);
984 }
985 return 0;
986}
987
988/* allocate pages for just the p/q stripes */
989static int alloc_rbio_parity_pages(struct btrfs_raid_bio *rbio)
990{
991 int i;
992 struct page *page;
993
994 i = (rbio->nr_data * rbio->stripe_len) >> PAGE_CACHE_SHIFT;
995
996 for (; i < rbio->nr_pages; i++) {
997 if (rbio->stripe_pages[i])
998 continue;
999 page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1000 if (!page)
1001 return -ENOMEM;
1002 rbio->stripe_pages[i] = page;
1003 }
1004 return 0;
1005}
1006
1007/*
1008 * add a single page from a specific stripe into our list of bios for IO
1009 * this will try to merge into existing bios if possible, and returns
1010 * zero if all went well.
1011 */
1012static int rbio_add_io_page(struct btrfs_raid_bio *rbio,
1013 struct bio_list *bio_list,
1014 struct page *page,
1015 int stripe_nr,
1016 unsigned long page_index,
1017 unsigned long bio_max_len)
1018{
1019 struct bio *last = bio_list->tail;
1020 u64 last_end = 0;
1021 int ret;
1022 struct bio *bio;
1023 struct btrfs_bio_stripe *stripe;
1024 u64 disk_start;
1025
1026 stripe = &rbio->bbio->stripes[stripe_nr];
1027 disk_start = stripe->physical + (page_index << PAGE_CACHE_SHIFT);
1028
1029 /* if the device is missing, just fail this stripe */
1030 if (!stripe->dev->bdev)
1031 return fail_rbio_index(rbio, stripe_nr);
1032
1033 /* see if we can add this page onto our existing bio */
1034 if (last) {
1035 last_end = (u64)last->bi_iter.bi_sector << 9;
1036 last_end += last->bi_iter.bi_size;
1037
1038 /*
1039 * we can't merge these if they are from different
1040 * devices or if they are not contiguous
1041 */
1042 if (last_end == disk_start && stripe->dev->bdev &&
1043 test_bit(BIO_UPTODATE, &last->bi_flags) &&
1044 last->bi_bdev == stripe->dev->bdev) {
1045 ret = bio_add_page(last, page, PAGE_CACHE_SIZE, 0);
1046 if (ret == PAGE_CACHE_SIZE)
1047 return 0;
1048 }
1049 }
1050
1051 /* put a new bio on the list */
1052 bio = btrfs_io_bio_alloc(GFP_NOFS, bio_max_len >> PAGE_SHIFT?:1);
1053 if (!bio)
1054 return -ENOMEM;
1055
1056 bio->bi_iter.bi_size = 0;
1057 bio->bi_bdev = stripe->dev->bdev;
1058 bio->bi_iter.bi_sector = disk_start >> 9;
1059 set_bit(BIO_UPTODATE, &bio->bi_flags);
1060
1061 bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
1062 bio_list_add(bio_list, bio);
1063 return 0;
1064}
1065
1066/*
1067 * while we're doing the read/modify/write cycle, we could
1068 * have errors in reading pages off the disk. This checks
1069 * for errors and if we're not able to read the page it'll
1070 * trigger parity reconstruction. The rmw will be finished
1071 * after we've reconstructed the failed stripes
1072 */
1073static void validate_rbio_for_rmw(struct btrfs_raid_bio *rbio)
1074{
1075 if (rbio->faila >= 0 || rbio->failb >= 0) {
1076 BUG_ON(rbio->faila == rbio->bbio->num_stripes - 1);
1077 __raid56_parity_recover(rbio);
1078 } else {
1079 finish_rmw(rbio);
1080 }
1081}
1082
1083/*
1084 * these are just the pages from the rbio array, not from anything
1085 * the FS sent down to us
1086 */
1087static struct page *rbio_stripe_page(struct btrfs_raid_bio *rbio, int stripe, int page)
1088{
1089 int index;
1090 index = stripe * (rbio->stripe_len >> PAGE_CACHE_SHIFT);
1091 index += page;
1092 return rbio->stripe_pages[index];
1093}
1094
1095/*
1096 * helper function to walk our bio list and populate the bio_pages array with
1097 * the result. This seems expensive, but it is faster than constantly
1098 * searching through the bio list as we setup the IO in finish_rmw or stripe
1099 * reconstruction.
1100 *
1101 * This must be called before you trust the answers from page_in_rbio
1102 */
1103static void index_rbio_pages(struct btrfs_raid_bio *rbio)
1104{
1105 struct bio *bio;
1106 u64 start;
1107 unsigned long stripe_offset;
1108 unsigned long page_index;
1109 struct page *p;
1110 int i;
1111
1112 spin_lock_irq(&rbio->bio_list_lock);
1113 bio_list_for_each(bio, &rbio->bio_list) {
1114 start = (u64)bio->bi_iter.bi_sector << 9;
1115 stripe_offset = start - rbio->raid_map[0];
1116 page_index = stripe_offset >> PAGE_CACHE_SHIFT;
1117
1118 for (i = 0; i < bio->bi_vcnt; i++) {
1119 p = bio->bi_io_vec[i].bv_page;
1120 rbio->bio_pages[page_index + i] = p;
1121 }
1122 }
1123 spin_unlock_irq(&rbio->bio_list_lock);
1124}
1125
1126/*
1127 * this is called from one of two situations. We either
1128 * have a full stripe from the higher layers, or we've read all
1129 * the missing bits off disk.
1130 *
1131 * This will calculate the parity and then send down any
1132 * changed blocks.
1133 */
1134static noinline void finish_rmw(struct btrfs_raid_bio *rbio)
1135{
1136 struct btrfs_bio *bbio = rbio->bbio;
1137 void *pointers[bbio->num_stripes];
1138 int stripe_len = rbio->stripe_len;
1139 int nr_data = rbio->nr_data;
1140 int stripe;
1141 int pagenr;
1142 int p_stripe = -1;
1143 int q_stripe = -1;
1144 struct bio_list bio_list;
1145 struct bio *bio;
1146 int pages_per_stripe = stripe_len >> PAGE_CACHE_SHIFT;
1147 int ret;
1148
1149 bio_list_init(&bio_list);
1150
1151 if (bbio->num_stripes - rbio->nr_data == 1) {
1152 p_stripe = bbio->num_stripes - 1;
1153 } else if (bbio->num_stripes - rbio->nr_data == 2) {
1154 p_stripe = bbio->num_stripes - 2;
1155 q_stripe = bbio->num_stripes - 1;
1156 } else {
1157 BUG();
1158 }
1159
1160 /* at this point we either have a full stripe,
1161 * or we've read the full stripe from the drive.
1162 * recalculate the parity and write the new results.
1163 *
1164 * We're not allowed to add any new bios to the
1165 * bio list here, anyone else that wants to
1166 * change this stripe needs to do their own rmw.
1167 */
1168 spin_lock_irq(&rbio->bio_list_lock);
1169 set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags);
1170 spin_unlock_irq(&rbio->bio_list_lock);
1171
1172 atomic_set(&rbio->bbio->error, 0);
1173
1174 /*
1175 * now that we've set rmw_locked, run through the
1176 * bio list one last time and map the page pointers
1177 *
1178 * We don't cache full rbios because we're assuming
1179 * the higher layers are unlikely to use this area of
1180 * the disk again soon. If they do use it again,
1181 * hopefully they will send another full bio.
1182 */
1183 index_rbio_pages(rbio);
1184 if (!rbio_is_full(rbio))
1185 cache_rbio_pages(rbio);
1186 else
1187 clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags);
1188
1189 for (pagenr = 0; pagenr < pages_per_stripe; pagenr++) {
1190 struct page *p;
1191 /* first collect one page from each data stripe */
1192 for (stripe = 0; stripe < nr_data; stripe++) {
1193 p = page_in_rbio(rbio, stripe, pagenr, 0);
1194 pointers[stripe] = kmap(p);
1195 }
1196
1197 /* then add the parity stripe */
1198 p = rbio_pstripe_page(rbio, pagenr);
1199 SetPageUptodate(p);
1200 pointers[stripe++] = kmap(p);
1201
1202 if (q_stripe != -1) {
1203
1204 /*
1205 * raid6, add the qstripe and call the
1206 * library function to fill in our p/q
1207 */
1208 p = rbio_qstripe_page(rbio, pagenr);
1209 SetPageUptodate(p);
1210 pointers[stripe++] = kmap(p);
1211
1212 raid6_call.gen_syndrome(bbio->num_stripes, PAGE_SIZE,
1213 pointers);
1214 } else {
1215 /* raid5 */
1216 memcpy(pointers[nr_data], pointers[0], PAGE_SIZE);
1217 run_xor(pointers + 1, nr_data - 1, PAGE_CACHE_SIZE);
1218 }
1219
1220
1221 for (stripe = 0; stripe < bbio->num_stripes; stripe++)
1222 kunmap(page_in_rbio(rbio, stripe, pagenr, 0));
1223 }
1224
1225 /*
1226 * time to start writing. Make bios for everything from the
1227 * higher layers (the bio_list in our rbio) and our p/q. Ignore
1228 * everything else.
1229 */
1230 for (stripe = 0; stripe < bbio->num_stripes; stripe++) {
1231 for (pagenr = 0; pagenr < pages_per_stripe; pagenr++) {
1232 struct page *page;
1233 if (stripe < rbio->nr_data) {
1234 page = page_in_rbio(rbio, stripe, pagenr, 1);
1235 if (!page)
1236 continue;
1237 } else {
1238 page = rbio_stripe_page(rbio, stripe, pagenr);
1239 }
1240
1241 ret = rbio_add_io_page(rbio, &bio_list,
1242 page, stripe, pagenr, rbio->stripe_len);
1243 if (ret)
1244 goto cleanup;
1245 }
1246 }
1247
1248 atomic_set(&bbio->stripes_pending, bio_list_size(&bio_list));
1249 BUG_ON(atomic_read(&bbio->stripes_pending) == 0);
1250
1251 while (1) {
1252 bio = bio_list_pop(&bio_list);
1253 if (!bio)
1254 break;
1255
1256 bio->bi_private = rbio;
1257 bio->bi_end_io = raid_write_end_io;
1258 BUG_ON(!test_bit(BIO_UPTODATE, &bio->bi_flags));
1259 submit_bio(WRITE, bio);
1260 }
1261 return;
1262
1263cleanup:
1264 rbio_orig_end_io(rbio, -EIO, 0);
1265}
1266
1267/*
1268 * helper to find the stripe number for a given bio. Used to figure out which
1269 * stripe has failed. This expects the bio to correspond to a physical disk,
1270 * so it looks up based on physical sector numbers.
1271 */
1272static int find_bio_stripe(struct btrfs_raid_bio *rbio,
1273 struct bio *bio)
1274{
1275 u64 physical = bio->bi_iter.bi_sector;
1276 u64 stripe_start;
1277 int i;
1278 struct btrfs_bio_stripe *stripe;
1279
1280 physical <<= 9;
1281
1282 for (i = 0; i < rbio->bbio->num_stripes; i++) {
1283 stripe = &rbio->bbio->stripes[i];
1284 stripe_start = stripe->physical;
1285 if (physical >= stripe_start &&
1286 physical < stripe_start + rbio->stripe_len) {
1287 return i;
1288 }
1289 }
1290 return -1;
1291}
1292
1293/*
1294 * helper to find the stripe number for a given
1295 * bio (before mapping). Used to figure out which stripe has
1296 * failed. This looks up based on logical block numbers.
1297 */
1298static int find_logical_bio_stripe(struct btrfs_raid_bio *rbio,
1299 struct bio *bio)
1300{
1301 u64 logical = bio->bi_iter.bi_sector;
1302 u64 stripe_start;
1303 int i;
1304
1305 logical <<= 9;
1306
1307 for (i = 0; i < rbio->nr_data; i++) {
1308 stripe_start = rbio->raid_map[i];
1309 if (logical >= stripe_start &&
1310 logical < stripe_start + rbio->stripe_len) {
1311 return i;
1312 }
1313 }
1314 return -1;
1315}
1316
1317/*
1318 * returns -EIO if we had too many failures
1319 */
1320static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed)
1321{
1322 unsigned long flags;
1323 int ret = 0;
1324
1325 spin_lock_irqsave(&rbio->bio_list_lock, flags);
1326
1327 /* we already know this stripe is bad, move on */
1328 if (rbio->faila == failed || rbio->failb == failed)
1329 goto out;
1330
1331 if (rbio->faila == -1) {
1332 /* first failure on this rbio */
1333 rbio->faila = failed;
1334 atomic_inc(&rbio->bbio->error);
1335 } else if (rbio->failb == -1) {
1336 /* second failure on this rbio */
1337 rbio->failb = failed;
1338 atomic_inc(&rbio->bbio->error);
1339 } else {
1340 ret = -EIO;
1341 }
1342out:
1343 spin_unlock_irqrestore(&rbio->bio_list_lock, flags);
1344
1345 return ret;
1346}
1347
1348/*
1349 * helper to fail a stripe based on a physical disk
1350 * bio.
1351 */
1352static int fail_bio_stripe(struct btrfs_raid_bio *rbio,
1353 struct bio *bio)
1354{
1355 int failed = find_bio_stripe(rbio, bio);
1356
1357 if (failed < 0)
1358 return -EIO;
1359
1360 return fail_rbio_index(rbio, failed);
1361}
1362
1363/*
1364 * this sets each page in the bio uptodate. It should only be used on private
1365 * rbio pages, nothing that comes in from the higher layers
1366 */
1367static void set_bio_pages_uptodate(struct bio *bio)
1368{
1369 int i;
1370 struct page *p;
1371
1372 for (i = 0; i < bio->bi_vcnt; i++) {
1373 p = bio->bi_io_vec[i].bv_page;
1374 SetPageUptodate(p);
1375 }
1376}
1377
1378/*
1379 * end io for the read phase of the rmw cycle. All the bios here are physical
1380 * stripe bios we've read from the disk so we can recalculate the parity of the
1381 * stripe.
1382 *
1383 * This will usually kick off finish_rmw once all the bios are read in, but it
1384 * may trigger parity reconstruction if we had any errors along the way
1385 */
1386static void raid_rmw_end_io(struct bio *bio, int err)
1387{
1388 struct btrfs_raid_bio *rbio = bio->bi_private;
1389
1390 if (err)
1391 fail_bio_stripe(rbio, bio);
1392 else
1393 set_bio_pages_uptodate(bio);
1394
1395 bio_put(bio);
1396
1397 if (!atomic_dec_and_test(&rbio->bbio->stripes_pending))
1398 return;
1399
1400 err = 0;
1401 if (atomic_read(&rbio->bbio->error) > rbio->bbio->max_errors)
1402 goto cleanup;
1403
1404 /*
1405 * this will normally call finish_rmw to start our write
1406 * but if there are any failed stripes we'll reconstruct
1407 * from parity first
1408 */
1409 validate_rbio_for_rmw(rbio);
1410 return;
1411
1412cleanup:
1413
1414 rbio_orig_end_io(rbio, -EIO, 0);
1415}
1416
1417static void async_rmw_stripe(struct btrfs_raid_bio *rbio)
1418{
1419 btrfs_init_work(&rbio->work, rmw_work, NULL, NULL);
1420
1421 btrfs_queue_work(rbio->fs_info->rmw_workers,
1422 &rbio->work);
1423}
1424
1425static void async_read_rebuild(struct btrfs_raid_bio *rbio)
1426{
1427 btrfs_init_work(&rbio->work, read_rebuild_work, NULL, NULL);
1428
1429 btrfs_queue_work(rbio->fs_info->rmw_workers,
1430 &rbio->work);
1431}
1432
1433/*
1434 * the stripe must be locked by the caller. It will
1435 * unlock after all the writes are done
1436 */
1437static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio)
1438{
1439 int bios_to_read = 0;
1440 struct btrfs_bio *bbio = rbio->bbio;
1441 struct bio_list bio_list;
1442 int ret;
1443 int nr_pages = (rbio->stripe_len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1444 int pagenr;
1445 int stripe;
1446 struct bio *bio;
1447
1448 bio_list_init(&bio_list);
1449
1450 ret = alloc_rbio_pages(rbio);
1451 if (ret)
1452 goto cleanup;
1453
1454 index_rbio_pages(rbio);
1455
1456 atomic_set(&rbio->bbio->error, 0);
1457 /*
1458 * build a list of bios to read all the missing parts of this
1459 * stripe
1460 */
1461 for (stripe = 0; stripe < rbio->nr_data; stripe++) {
1462 for (pagenr = 0; pagenr < nr_pages; pagenr++) {
1463 struct page *page;
1464 /*
1465 * we want to find all the pages missing from
1466 * the rbio and read them from the disk. If
1467 * page_in_rbio finds a page in the bio list
1468 * we don't need to read it off the stripe.
1469 */
1470 page = page_in_rbio(rbio, stripe, pagenr, 1);
1471 if (page)
1472 continue;
1473
1474 page = rbio_stripe_page(rbio, stripe, pagenr);
1475 /*
1476 * the bio cache may have handed us an uptodate
1477 * page. If so, be happy and use it
1478 */
1479 if (PageUptodate(page))
1480 continue;
1481
1482 ret = rbio_add_io_page(rbio, &bio_list, page,
1483 stripe, pagenr, rbio->stripe_len);
1484 if (ret)
1485 goto cleanup;
1486 }
1487 }
1488
1489 bios_to_read = bio_list_size(&bio_list);
1490 if (!bios_to_read) {
1491 /*
1492 * this can happen if others have merged with
1493 * us, it means there is nothing left to read.
1494 * But if there are missing devices it may not be
1495 * safe to do the full stripe write yet.
1496 */
1497 goto finish;
1498 }
1499
1500 /*
1501 * the bbio may be freed once we submit the last bio. Make sure
1502 * not to touch it after that
1503 */
1504 atomic_set(&bbio->stripes_pending, bios_to_read);
1505 while (1) {
1506 bio = bio_list_pop(&bio_list);
1507 if (!bio)
1508 break;
1509
1510 bio->bi_private = rbio;
1511 bio->bi_end_io = raid_rmw_end_io;
1512
1513 btrfs_bio_wq_end_io(rbio->fs_info, bio,
1514 BTRFS_WQ_ENDIO_RAID56);
1515
1516 BUG_ON(!test_bit(BIO_UPTODATE, &bio->bi_flags));
1517 submit_bio(READ, bio);
1518 }
1519 /* the actual write will happen once the reads are done */
1520 return 0;
1521
1522cleanup:
1523 rbio_orig_end_io(rbio, -EIO, 0);
1524 return -EIO;
1525
1526finish:
1527 validate_rbio_for_rmw(rbio);
1528 return 0;
1529}
1530
1531/*
1532 * if the upper layers pass in a full stripe, we thank them by only allocating
1533 * enough pages to hold the parity, and sending it all down quickly.
1534 */
1535static int full_stripe_write(struct btrfs_raid_bio *rbio)
1536{
1537 int ret;
1538
1539 ret = alloc_rbio_parity_pages(rbio);
1540 if (ret) {
1541 __free_raid_bio(rbio);
1542 return ret;
1543 }
1544
1545 ret = lock_stripe_add(rbio);
1546 if (ret == 0)
1547 finish_rmw(rbio);
1548 return 0;
1549}
1550
1551/*
1552 * partial stripe writes get handed over to async helpers.
1553 * We're really hoping to merge a few more writes into this
1554 * rbio before calculating new parity
1555 */
1556static int partial_stripe_write(struct btrfs_raid_bio *rbio)
1557{
1558 int ret;
1559
1560 ret = lock_stripe_add(rbio);
1561 if (ret == 0)
1562 async_rmw_stripe(rbio);
1563 return 0;
1564}
1565
1566/*
1567 * sometimes while we were reading from the drive to
1568 * recalculate parity, enough new bios come into create
1569 * a full stripe. So we do a check here to see if we can
1570 * go directly to finish_rmw
1571 */
1572static int __raid56_parity_write(struct btrfs_raid_bio *rbio)
1573{
1574 /* head off into rmw land if we don't have a full stripe */
1575 if (!rbio_is_full(rbio))
1576 return partial_stripe_write(rbio);
1577 return full_stripe_write(rbio);
1578}
1579
1580/*
1581 * We use plugging call backs to collect full stripes.
1582 * Any time we get a partial stripe write while plugged
1583 * we collect it into a list. When the unplug comes down,
1584 * we sort the list by logical block number and merge
1585 * everything we can into the same rbios
1586 */
1587struct btrfs_plug_cb {
1588 struct blk_plug_cb cb;
1589 struct btrfs_fs_info *info;
1590 struct list_head rbio_list;
1591 struct btrfs_work work;
1592};
1593
1594/*
1595 * rbios on the plug list are sorted for easier merging.
1596 */
1597static int plug_cmp(void *priv, struct list_head *a, struct list_head *b)
1598{
1599 struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio,
1600 plug_list);
1601 struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio,
1602 plug_list);
1603 u64 a_sector = ra->bio_list.head->bi_iter.bi_sector;
1604 u64 b_sector = rb->bio_list.head->bi_iter.bi_sector;
1605
1606 if (a_sector < b_sector)
1607 return -1;
1608 if (a_sector > b_sector)
1609 return 1;
1610 return 0;
1611}
1612
1613static void run_plug(struct btrfs_plug_cb *plug)
1614{
1615 struct btrfs_raid_bio *cur;
1616 struct btrfs_raid_bio *last = NULL;
1617
1618 /*
1619 * sort our plug list then try to merge
1620 * everything we can in hopes of creating full
1621 * stripes.
1622 */
1623 list_sort(NULL, &plug->rbio_list, plug_cmp);
1624 while (!list_empty(&plug->rbio_list)) {
1625 cur = list_entry(plug->rbio_list.next,
1626 struct btrfs_raid_bio, plug_list);
1627 list_del_init(&cur->plug_list);
1628
1629 if (rbio_is_full(cur)) {
1630 /* we have a full stripe, send it down */
1631 full_stripe_write(cur);
1632 continue;
1633 }
1634 if (last) {
1635 if (rbio_can_merge(last, cur)) {
1636 merge_rbio(last, cur);
1637 __free_raid_bio(cur);
1638 continue;
1639
1640 }
1641 __raid56_parity_write(last);
1642 }
1643 last = cur;
1644 }
1645 if (last) {
1646 __raid56_parity_write(last);
1647 }
1648 kfree(plug);
1649}
1650
1651/*
1652 * if the unplug comes from schedule, we have to push the
1653 * work off to a helper thread
1654 */
1655static void unplug_work(struct btrfs_work *work)
1656{
1657 struct btrfs_plug_cb *plug;
1658 plug = container_of(work, struct btrfs_plug_cb, work);
1659 run_plug(plug);
1660}
1661
1662static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule)
1663{
1664 struct btrfs_plug_cb *plug;
1665 plug = container_of(cb, struct btrfs_plug_cb, cb);
1666
1667 if (from_schedule) {
1668 btrfs_init_work(&plug->work, unplug_work, NULL, NULL);
1669 btrfs_queue_work(plug->info->rmw_workers,
1670 &plug->work);
1671 return;
1672 }
1673 run_plug(plug);
1674}
1675
1676/*
1677 * our main entry point for writes from the rest of the FS.
1678 */
1679int raid56_parity_write(struct btrfs_root *root, struct bio *bio,
1680 struct btrfs_bio *bbio, u64 *raid_map,
1681 u64 stripe_len)
1682{
1683 struct btrfs_raid_bio *rbio;
1684 struct btrfs_plug_cb *plug = NULL;
1685 struct blk_plug_cb *cb;
1686
1687 rbio = alloc_rbio(root, bbio, raid_map, stripe_len);
1688 if (IS_ERR(rbio))
1689 return PTR_ERR(rbio);
1690 bio_list_add(&rbio->bio_list, bio);
1691 rbio->bio_list_bytes = bio->bi_iter.bi_size;
1692
1693 /*
1694 * don't plug on full rbios, just get them out the door
1695 * as quickly as we can
1696 */
1697 if (rbio_is_full(rbio))
1698 return full_stripe_write(rbio);
1699
1700 cb = blk_check_plugged(btrfs_raid_unplug, root->fs_info,
1701 sizeof(*plug));
1702 if (cb) {
1703 plug = container_of(cb, struct btrfs_plug_cb, cb);
1704 if (!plug->info) {
1705 plug->info = root->fs_info;
1706 INIT_LIST_HEAD(&plug->rbio_list);
1707 }
1708 list_add_tail(&rbio->plug_list, &plug->rbio_list);
1709 } else {
1710 return __raid56_parity_write(rbio);
1711 }
1712 return 0;
1713}
1714
1715/*
1716 * all parity reconstruction happens here. We've read in everything
1717 * we can find from the drives and this does the heavy lifting of
1718 * sorting the good from the bad.
1719 */
1720static void __raid_recover_end_io(struct btrfs_raid_bio *rbio)
1721{
1722 int pagenr, stripe;
1723 void **pointers;
1724 int faila = -1, failb = -1;
1725 int nr_pages = (rbio->stripe_len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1726 struct page *page;
1727 int err;
1728 int i;
1729
1730 pointers = kzalloc(rbio->bbio->num_stripes * sizeof(void *),
1731 GFP_NOFS);
1732 if (!pointers) {
1733 err = -ENOMEM;
1734 goto cleanup_io;
1735 }
1736
1737 faila = rbio->faila;
1738 failb = rbio->failb;
1739
1740 if (rbio->read_rebuild) {
1741 spin_lock_irq(&rbio->bio_list_lock);
1742 set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags);
1743 spin_unlock_irq(&rbio->bio_list_lock);
1744 }
1745
1746 index_rbio_pages(rbio);
1747
1748 for (pagenr = 0; pagenr < nr_pages; pagenr++) {
1749 /* setup our array of pointers with pages
1750 * from each stripe
1751 */
1752 for (stripe = 0; stripe < rbio->bbio->num_stripes; stripe++) {
1753 /*
1754 * if we're rebuilding a read, we have to use
1755 * pages from the bio list
1756 */
1757 if (rbio->read_rebuild &&
1758 (stripe == faila || stripe == failb)) {
1759 page = page_in_rbio(rbio, stripe, pagenr, 0);
1760 } else {
1761 page = rbio_stripe_page(rbio, stripe, pagenr);
1762 }
1763 pointers[stripe] = kmap(page);
1764 }
1765
1766 /* all raid6 handling here */
1767 if (rbio->raid_map[rbio->bbio->num_stripes - 1] ==
1768 RAID6_Q_STRIPE) {
1769
1770 /*
1771 * single failure, rebuild from parity raid5
1772 * style
1773 */
1774 if (failb < 0) {
1775 if (faila == rbio->nr_data) {
1776 /*
1777 * Just the P stripe has failed, without
1778 * a bad data or Q stripe.
1779 * TODO, we should redo the xor here.
1780 */
1781 err = -EIO;
1782 goto cleanup;
1783 }
1784 /*
1785 * a single failure in raid6 is rebuilt
1786 * in the pstripe code below
1787 */
1788 goto pstripe;
1789 }
1790
1791 /* make sure our ps and qs are in order */
1792 if (faila > failb) {
1793 int tmp = failb;
1794 failb = faila;
1795 faila = tmp;
1796 }
1797
1798 /* if the q stripe is failed, do a pstripe reconstruction
1799 * from the xors.
1800 * If both the q stripe and the P stripe are failed, we're
1801 * here due to a crc mismatch and we can't give them the
1802 * data they want
1803 */
1804 if (rbio->raid_map[failb] == RAID6_Q_STRIPE) {
1805 if (rbio->raid_map[faila] == RAID5_P_STRIPE) {
1806 err = -EIO;
1807 goto cleanup;
1808 }
1809 /*
1810 * otherwise we have one bad data stripe and
1811 * a good P stripe. raid5!
1812 */
1813 goto pstripe;
1814 }
1815
1816 if (rbio->raid_map[failb] == RAID5_P_STRIPE) {
1817 raid6_datap_recov(rbio->bbio->num_stripes,
1818 PAGE_SIZE, faila, pointers);
1819 } else {
1820 raid6_2data_recov(rbio->bbio->num_stripes,
1821 PAGE_SIZE, faila, failb,
1822 pointers);
1823 }
1824 } else {
1825 void *p;
1826
1827 /* rebuild from P stripe here (raid5 or raid6) */
1828 BUG_ON(failb != -1);
1829pstripe:
1830 /* Copy parity block into failed block to start with */
1831 memcpy(pointers[faila],
1832 pointers[rbio->nr_data],
1833 PAGE_CACHE_SIZE);
1834
1835 /* rearrange the pointer array */
1836 p = pointers[faila];
1837 for (stripe = faila; stripe < rbio->nr_data - 1; stripe++)
1838 pointers[stripe] = pointers[stripe + 1];
1839 pointers[rbio->nr_data - 1] = p;
1840
1841 /* xor in the rest */
1842 run_xor(pointers, rbio->nr_data - 1, PAGE_CACHE_SIZE);
1843 }
1844 /* if we're doing this rebuild as part of an rmw, go through
1845 * and set all of our private rbio pages in the
1846 * failed stripes as uptodate. This way finish_rmw will
1847 * know they can be trusted. If this was a read reconstruction,
1848 * other endio functions will fiddle the uptodate bits
1849 */
1850 if (!rbio->read_rebuild) {
1851 for (i = 0; i < nr_pages; i++) {
1852 if (faila != -1) {
1853 page = rbio_stripe_page(rbio, faila, i);
1854 SetPageUptodate(page);
1855 }
1856 if (failb != -1) {
1857 page = rbio_stripe_page(rbio, failb, i);
1858 SetPageUptodate(page);
1859 }
1860 }
1861 }
1862 for (stripe = 0; stripe < rbio->bbio->num_stripes; stripe++) {
1863 /*
1864 * if we're rebuilding a read, we have to use
1865 * pages from the bio list
1866 */
1867 if (rbio->read_rebuild &&
1868 (stripe == faila || stripe == failb)) {
1869 page = page_in_rbio(rbio, stripe, pagenr, 0);
1870 } else {
1871 page = rbio_stripe_page(rbio, stripe, pagenr);
1872 }
1873 kunmap(page);
1874 }
1875 }
1876
1877 err = 0;
1878cleanup:
1879 kfree(pointers);
1880
1881cleanup_io:
1882
1883 if (rbio->read_rebuild) {
1884 if (err == 0)
1885 cache_rbio_pages(rbio);
1886 else
1887 clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags);
1888
1889 rbio_orig_end_io(rbio, err, err == 0);
1890 } else if (err == 0) {
1891 rbio->faila = -1;
1892 rbio->failb = -1;
1893 finish_rmw(rbio);
1894 } else {
1895 rbio_orig_end_io(rbio, err, 0);
1896 }
1897}
1898
1899/*
1900 * This is called only for stripes we've read from disk to
1901 * reconstruct the parity.
1902 */
1903static void raid_recover_end_io(struct bio *bio, int err)
1904{
1905 struct btrfs_raid_bio *rbio = bio->bi_private;
1906
1907 /*
1908 * we only read stripe pages off the disk, set them
1909 * up to date if there were no errors
1910 */
1911 if (err)
1912 fail_bio_stripe(rbio, bio);
1913 else
1914 set_bio_pages_uptodate(bio);
1915 bio_put(bio);
1916
1917 if (!atomic_dec_and_test(&rbio->bbio->stripes_pending))
1918 return;
1919
1920 if (atomic_read(&rbio->bbio->error) > rbio->bbio->max_errors)
1921 rbio_orig_end_io(rbio, -EIO, 0);
1922 else
1923 __raid_recover_end_io(rbio);
1924}
1925
1926/*
1927 * reads everything we need off the disk to reconstruct
1928 * the parity. endio handlers trigger final reconstruction
1929 * when the IO is done.
1930 *
1931 * This is used both for reads from the higher layers and for
1932 * parity construction required to finish a rmw cycle.
1933 */
1934static int __raid56_parity_recover(struct btrfs_raid_bio *rbio)
1935{
1936 int bios_to_read = 0;
1937 struct btrfs_bio *bbio = rbio->bbio;
1938 struct bio_list bio_list;
1939 int ret;
1940 int nr_pages = (rbio->stripe_len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1941 int pagenr;
1942 int stripe;
1943 struct bio *bio;
1944
1945 bio_list_init(&bio_list);
1946
1947 ret = alloc_rbio_pages(rbio);
1948 if (ret)
1949 goto cleanup;
1950
1951 atomic_set(&rbio->bbio->error, 0);
1952
1953 /*
1954 * read everything that hasn't failed. Thanks to the
1955 * stripe cache, it is possible that some or all of these
1956 * pages are going to be uptodate.
1957 */
1958 for (stripe = 0; stripe < bbio->num_stripes; stripe++) {
1959 if (rbio->faila == stripe ||
1960 rbio->failb == stripe)
1961 continue;
1962
1963 for (pagenr = 0; pagenr < nr_pages; pagenr++) {
1964 struct page *p;
1965
1966 /*
1967 * the rmw code may have already read this
1968 * page in
1969 */
1970 p = rbio_stripe_page(rbio, stripe, pagenr);
1971 if (PageUptodate(p))
1972 continue;
1973
1974 ret = rbio_add_io_page(rbio, &bio_list,
1975 rbio_stripe_page(rbio, stripe, pagenr),
1976 stripe, pagenr, rbio->stripe_len);
1977 if (ret < 0)
1978 goto cleanup;
1979 }
1980 }
1981
1982 bios_to_read = bio_list_size(&bio_list);
1983 if (!bios_to_read) {
1984 /*
1985 * we might have no bios to read just because the pages
1986 * were up to date, or we might have no bios to read because
1987 * the devices were gone.
1988 */
1989 if (atomic_read(&rbio->bbio->error) <= rbio->bbio->max_errors) {
1990 __raid_recover_end_io(rbio);
1991 goto out;
1992 } else {
1993 goto cleanup;
1994 }
1995 }
1996
1997 /*
1998 * the bbio may be freed once we submit the last bio. Make sure
1999 * not to touch it after that
2000 */
2001 atomic_set(&bbio->stripes_pending, bios_to_read);
2002 while (1) {
2003 bio = bio_list_pop(&bio_list);
2004 if (!bio)
2005 break;
2006
2007 bio->bi_private = rbio;
2008 bio->bi_end_io = raid_recover_end_io;
2009
2010 btrfs_bio_wq_end_io(rbio->fs_info, bio,
2011 BTRFS_WQ_ENDIO_RAID56);
2012
2013 BUG_ON(!test_bit(BIO_UPTODATE, &bio->bi_flags));
2014 submit_bio(READ, bio);
2015 }
2016out:
2017 return 0;
2018
2019cleanup:
2020 if (rbio->read_rebuild)
2021 rbio_orig_end_io(rbio, -EIO, 0);
2022 return -EIO;
2023}
2024
2025/*
2026 * the main entry point for reads from the higher layers. This
2027 * is really only called when the normal read path had a failure,
2028 * so we assume the bio they send down corresponds to a failed part
2029 * of the drive.
2030 */
2031int raid56_parity_recover(struct btrfs_root *root, struct bio *bio,
2032 struct btrfs_bio *bbio, u64 *raid_map,
2033 u64 stripe_len, int mirror_num)
2034{
2035 struct btrfs_raid_bio *rbio;
2036 int ret;
2037
2038 rbio = alloc_rbio(root, bbio, raid_map, stripe_len);
2039 if (IS_ERR(rbio))
2040 return PTR_ERR(rbio);
2041
2042 rbio->read_rebuild = 1;
2043 bio_list_add(&rbio->bio_list, bio);
2044 rbio->bio_list_bytes = bio->bi_iter.bi_size;
2045
2046 rbio->faila = find_logical_bio_stripe(rbio, bio);
2047 if (rbio->faila == -1) {
2048 BUG();
2049 kfree(raid_map);
2050 kfree(bbio);
2051 kfree(rbio);
2052 return -EIO;
2053 }
2054
2055 /*
2056 * reconstruct from the q stripe if they are
2057 * asking for mirror 3
2058 */
2059 if (mirror_num == 3)
2060 rbio->failb = bbio->num_stripes - 2;
2061
2062 ret = lock_stripe_add(rbio);
2063
2064 /*
2065 * __raid56_parity_recover will end the bio with
2066 * any errors it hits. We don't want to return
2067 * its error value up the stack because our caller
2068 * will end up calling bio_endio with any nonzero
2069 * return
2070 */
2071 if (ret == 0)
2072 __raid56_parity_recover(rbio);
2073 /*
2074 * our rbio has been added to the list of
2075 * rbios that will be handled after the
2076 * currently lock owner is done
2077 */
2078 return 0;
2079
2080}
2081
2082static void rmw_work(struct btrfs_work *work)
2083{
2084 struct btrfs_raid_bio *rbio;
2085
2086 rbio = container_of(work, struct btrfs_raid_bio, work);
2087 raid56_rmw_stripe(rbio);
2088}
2089
2090static void read_rebuild_work(struct btrfs_work *work)
2091{
2092 struct btrfs_raid_bio *rbio;
2093
2094 rbio = container_of(work, struct btrfs_raid_bio, work);
2095 __raid56_parity_recover(rbio);
2096}