xref: /openbmc/linux/fs/btrfs/raid56.c (revision 39b6f3aa)
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 "compat.h"
37 #include "ctree.h"
38 #include "extent_map.h"
39 #include "disk-io.h"
40 #include "transaction.h"
41 #include "print-tree.h"
42 #include "volumes.h"
43 #include "raid56.h"
44 #include "async-thread.h"
45 #include "check-integrity.h"
46 #include "rcu-string.h"
47 
48 /* set when additional merges to this rbio are not allowed */
49 #define RBIO_RMW_LOCKED_BIT	1
50 
51 /*
52  * set when this rbio is sitting in the hash, but it is just a cache
53  * of past RMW
54  */
55 #define RBIO_CACHE_BIT		2
56 
57 /*
58  * set when it is safe to trust the stripe_pages for caching
59  */
60 #define RBIO_CACHE_READY_BIT	3
61 
62 
63 #define RBIO_CACHE_SIZE 1024
64 
65 struct btrfs_raid_bio {
66 	struct btrfs_fs_info *fs_info;
67 	struct btrfs_bio *bbio;
68 
69 	/*
70 	 * logical block numbers for the start of each stripe
71 	 * The last one or two are p/q.  These are sorted,
72 	 * so raid_map[0] is the start of our full stripe
73 	 */
74 	u64 *raid_map;
75 
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 
83 	/*
84 	 * LRU list for the stripe cache
85 	 */
86 	struct list_head stripe_cache;
87 
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 
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
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 
115 	/* size of each individual stripe on disk */
116 	int stripe_len;
117 
118 	/* number of data stripes (no p/q) */
119 	int nr_data;
120 
121 	/*
122 	 * set if we're doing a parity rebuild
123 	 * for a read from higher up, which is handled
124 	 * differently from a parity rebuild as part of
125 	 * rmw
126 	 */
127 	int read_rebuild;
128 
129 	/* first bad stripe */
130 	int faila;
131 
132 	/* second bad stripe (for raid6 use) */
133 	int failb;
134 
135 	/*
136 	 * number of pages needed to represent the full
137 	 * stripe
138 	 */
139 	int nr_pages;
140 
141 	/*
142 	 * size of all the bios in the bio_list.  This
143 	 * helps us decide if the rbio maps to a full
144 	 * stripe or not
145 	 */
146 	int bio_list_bytes;
147 
148 	atomic_t refs;
149 
150 	/*
151 	 * these are two arrays of pointers.  We allocate the
152 	 * rbio big enough to hold them both and setup their
153 	 * locations when the rbio is allocated
154 	 */
155 
156 	/* pointers to pages that we allocated for
157 	 * reading/writing stripes directly from the disk (including P/Q)
158 	 */
159 	struct page **stripe_pages;
160 
161 	/*
162 	 * pointers to the pages in the bio_list.  Stored
163 	 * here for faster lookup
164 	 */
165 	struct page **bio_pages;
166 };
167 
168 static int __raid56_parity_recover(struct btrfs_raid_bio *rbio);
169 static noinline void finish_rmw(struct btrfs_raid_bio *rbio);
170 static void rmw_work(struct btrfs_work *work);
171 static void read_rebuild_work(struct btrfs_work *work);
172 static void async_rmw_stripe(struct btrfs_raid_bio *rbio);
173 static void async_read_rebuild(struct btrfs_raid_bio *rbio);
174 static int fail_bio_stripe(struct btrfs_raid_bio *rbio, struct bio *bio);
175 static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed);
176 static void __free_raid_bio(struct btrfs_raid_bio *rbio);
177 static void index_rbio_pages(struct btrfs_raid_bio *rbio);
178 static int alloc_rbio_pages(struct btrfs_raid_bio *rbio);
179 
180 /*
181  * the stripe hash table is used for locking, and to collect
182  * bios in hopes of making a full stripe
183  */
184 int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info)
185 {
186 	struct btrfs_stripe_hash_table *table;
187 	struct btrfs_stripe_hash_table *x;
188 	struct btrfs_stripe_hash *cur;
189 	struct btrfs_stripe_hash *h;
190 	int num_entries = 1 << BTRFS_STRIPE_HASH_TABLE_BITS;
191 	int i;
192 	int table_size;
193 
194 	if (info->stripe_hash_table)
195 		return 0;
196 
197 	/*
198 	 * The table is large, starting with order 4 and can go as high as
199 	 * order 7 in case lock debugging is turned on.
200 	 *
201 	 * Try harder to allocate and fallback to vmalloc to lower the chance
202 	 * of a failing mount.
203 	 */
204 	table_size = sizeof(*table) + sizeof(*h) * num_entries;
205 	table = kzalloc(table_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
206 	if (!table) {
207 		table = vzalloc(table_size);
208 		if (!table)
209 			return -ENOMEM;
210 	}
211 
212 	spin_lock_init(&table->cache_lock);
213 	INIT_LIST_HEAD(&table->stripe_cache);
214 
215 	h = table->table;
216 
217 	for (i = 0; i < num_entries; i++) {
218 		cur = h + i;
219 		INIT_LIST_HEAD(&cur->hash_list);
220 		spin_lock_init(&cur->lock);
221 		init_waitqueue_head(&cur->wait);
222 	}
223 
224 	x = cmpxchg(&info->stripe_hash_table, NULL, table);
225 	if (x) {
226 		if (is_vmalloc_addr(x))
227 			vfree(x);
228 		else
229 			kfree(x);
230 	}
231 	return 0;
232 }
233 
234 /*
235  * caching an rbio means to copy anything from the
236  * bio_pages array into the stripe_pages array.  We
237  * use the page uptodate bit in the stripe cache array
238  * to indicate if it has valid data
239  *
240  * once the caching is done, we set the cache ready
241  * bit.
242  */
243 static void cache_rbio_pages(struct btrfs_raid_bio *rbio)
244 {
245 	int i;
246 	char *s;
247 	char *d;
248 	int ret;
249 
250 	ret = alloc_rbio_pages(rbio);
251 	if (ret)
252 		return;
253 
254 	for (i = 0; i < rbio->nr_pages; i++) {
255 		if (!rbio->bio_pages[i])
256 			continue;
257 
258 		s = kmap(rbio->bio_pages[i]);
259 		d = kmap(rbio->stripe_pages[i]);
260 
261 		memcpy(d, s, PAGE_CACHE_SIZE);
262 
263 		kunmap(rbio->bio_pages[i]);
264 		kunmap(rbio->stripe_pages[i]);
265 		SetPageUptodate(rbio->stripe_pages[i]);
266 	}
267 	set_bit(RBIO_CACHE_READY_BIT, &rbio->flags);
268 }
269 
270 /*
271  * we hash on the first logical address of the stripe
272  */
273 static int rbio_bucket(struct btrfs_raid_bio *rbio)
274 {
275 	u64 num = rbio->raid_map[0];
276 
277 	/*
278 	 * we shift down quite a bit.  We're using byte
279 	 * addressing, and most of the lower bits are zeros.
280 	 * This tends to upset hash_64, and it consistently
281 	 * returns just one or two different values.
282 	 *
283 	 * shifting off the lower bits fixes things.
284 	 */
285 	return hash_64(num >> 16, BTRFS_STRIPE_HASH_TABLE_BITS);
286 }
287 
288 /*
289  * stealing an rbio means taking all the uptodate pages from the stripe
290  * array in the source rbio and putting them into the destination rbio
291  */
292 static void steal_rbio(struct btrfs_raid_bio *src, struct btrfs_raid_bio *dest)
293 {
294 	int i;
295 	struct page *s;
296 	struct page *d;
297 
298 	if (!test_bit(RBIO_CACHE_READY_BIT, &src->flags))
299 		return;
300 
301 	for (i = 0; i < dest->nr_pages; i++) {
302 		s = src->stripe_pages[i];
303 		if (!s || !PageUptodate(s)) {
304 			continue;
305 		}
306 
307 		d = dest->stripe_pages[i];
308 		if (d)
309 			__free_page(d);
310 
311 		dest->stripe_pages[i] = s;
312 		src->stripe_pages[i] = NULL;
313 	}
314 }
315 
316 /*
317  * merging means we take the bio_list from the victim and
318  * splice it into the destination.  The victim should
319  * be discarded afterwards.
320  *
321  * must be called with dest->rbio_list_lock held
322  */
323 static void merge_rbio(struct btrfs_raid_bio *dest,
324 		       struct btrfs_raid_bio *victim)
325 {
326 	bio_list_merge(&dest->bio_list, &victim->bio_list);
327 	dest->bio_list_bytes += victim->bio_list_bytes;
328 	bio_list_init(&victim->bio_list);
329 }
330 
331 /*
332  * used to prune items that are in the cache.  The caller
333  * must hold the hash table lock.
334  */
335 static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio)
336 {
337 	int bucket = rbio_bucket(rbio);
338 	struct btrfs_stripe_hash_table *table;
339 	struct btrfs_stripe_hash *h;
340 	int freeit = 0;
341 
342 	/*
343 	 * check the bit again under the hash table lock.
344 	 */
345 	if (!test_bit(RBIO_CACHE_BIT, &rbio->flags))
346 		return;
347 
348 	table = rbio->fs_info->stripe_hash_table;
349 	h = table->table + bucket;
350 
351 	/* hold the lock for the bucket because we may be
352 	 * removing it from the hash table
353 	 */
354 	spin_lock(&h->lock);
355 
356 	/*
357 	 * hold the lock for the bio list because we need
358 	 * to make sure the bio list is empty
359 	 */
360 	spin_lock(&rbio->bio_list_lock);
361 
362 	if (test_and_clear_bit(RBIO_CACHE_BIT, &rbio->flags)) {
363 		list_del_init(&rbio->stripe_cache);
364 		table->cache_size -= 1;
365 		freeit = 1;
366 
367 		/* if the bio list isn't empty, this rbio is
368 		 * still involved in an IO.  We take it out
369 		 * of the cache list, and drop the ref that
370 		 * was held for the list.
371 		 *
372 		 * If the bio_list was empty, we also remove
373 		 * the rbio from the hash_table, and drop
374 		 * the corresponding ref
375 		 */
376 		if (bio_list_empty(&rbio->bio_list)) {
377 			if (!list_empty(&rbio->hash_list)) {
378 				list_del_init(&rbio->hash_list);
379 				atomic_dec(&rbio->refs);
380 				BUG_ON(!list_empty(&rbio->plug_list));
381 			}
382 		}
383 	}
384 
385 	spin_unlock(&rbio->bio_list_lock);
386 	spin_unlock(&h->lock);
387 
388 	if (freeit)
389 		__free_raid_bio(rbio);
390 }
391 
392 /*
393  * prune a given rbio from the cache
394  */
395 static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio)
396 {
397 	struct btrfs_stripe_hash_table *table;
398 	unsigned long flags;
399 
400 	if (!test_bit(RBIO_CACHE_BIT, &rbio->flags))
401 		return;
402 
403 	table = rbio->fs_info->stripe_hash_table;
404 
405 	spin_lock_irqsave(&table->cache_lock, flags);
406 	__remove_rbio_from_cache(rbio);
407 	spin_unlock_irqrestore(&table->cache_lock, flags);
408 }
409 
410 /*
411  * remove everything in the cache
412  */
413 static void btrfs_clear_rbio_cache(struct btrfs_fs_info *info)
414 {
415 	struct btrfs_stripe_hash_table *table;
416 	unsigned long flags;
417 	struct btrfs_raid_bio *rbio;
418 
419 	table = info->stripe_hash_table;
420 
421 	spin_lock_irqsave(&table->cache_lock, flags);
422 	while (!list_empty(&table->stripe_cache)) {
423 		rbio = list_entry(table->stripe_cache.next,
424 				  struct btrfs_raid_bio,
425 				  stripe_cache);
426 		__remove_rbio_from_cache(rbio);
427 	}
428 	spin_unlock_irqrestore(&table->cache_lock, flags);
429 }
430 
431 /*
432  * remove all cached entries and free the hash table
433  * used by unmount
434  */
435 void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info)
436 {
437 	if (!info->stripe_hash_table)
438 		return;
439 	btrfs_clear_rbio_cache(info);
440 	if (is_vmalloc_addr(info->stripe_hash_table))
441 		vfree(info->stripe_hash_table);
442 	else
443 		kfree(info->stripe_hash_table);
444 	info->stripe_hash_table = NULL;
445 }
446 
447 /*
448  * insert an rbio into the stripe cache.  It
449  * must have already been prepared by calling
450  * cache_rbio_pages
451  *
452  * If this rbio was already cached, it gets
453  * moved to the front of the lru.
454  *
455  * If the size of the rbio cache is too big, we
456  * prune an item.
457  */
458 static void cache_rbio(struct btrfs_raid_bio *rbio)
459 {
460 	struct btrfs_stripe_hash_table *table;
461 	unsigned long flags;
462 
463 	if (!test_bit(RBIO_CACHE_READY_BIT, &rbio->flags))
464 		return;
465 
466 	table = rbio->fs_info->stripe_hash_table;
467 
468 	spin_lock_irqsave(&table->cache_lock, flags);
469 	spin_lock(&rbio->bio_list_lock);
470 
471 	/* bump our ref if we were not in the list before */
472 	if (!test_and_set_bit(RBIO_CACHE_BIT, &rbio->flags))
473 		atomic_inc(&rbio->refs);
474 
475 	if (!list_empty(&rbio->stripe_cache)){
476 		list_move(&rbio->stripe_cache, &table->stripe_cache);
477 	} else {
478 		list_add(&rbio->stripe_cache, &table->stripe_cache);
479 		table->cache_size += 1;
480 	}
481 
482 	spin_unlock(&rbio->bio_list_lock);
483 
484 	if (table->cache_size > RBIO_CACHE_SIZE) {
485 		struct btrfs_raid_bio *found;
486 
487 		found = list_entry(table->stripe_cache.prev,
488 				  struct btrfs_raid_bio,
489 				  stripe_cache);
490 
491 		if (found != rbio)
492 			__remove_rbio_from_cache(found);
493 	}
494 
495 	spin_unlock_irqrestore(&table->cache_lock, flags);
496 	return;
497 }
498 
499 /*
500  * helper function to run the xor_blocks api.  It is only
501  * able to do MAX_XOR_BLOCKS at a time, so we need to
502  * loop through.
503  */
504 static void run_xor(void **pages, int src_cnt, ssize_t len)
505 {
506 	int src_off = 0;
507 	int xor_src_cnt = 0;
508 	void *dest = pages[src_cnt];
509 
510 	while(src_cnt > 0) {
511 		xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS);
512 		xor_blocks(xor_src_cnt, len, dest, pages + src_off);
513 
514 		src_cnt -= xor_src_cnt;
515 		src_off += xor_src_cnt;
516 	}
517 }
518 
519 /*
520  * returns true if the bio list inside this rbio
521  * covers an entire stripe (no rmw required).
522  * Must be called with the bio list lock held, or
523  * at a time when you know it is impossible to add
524  * new bios into the list
525  */
526 static int __rbio_is_full(struct btrfs_raid_bio *rbio)
527 {
528 	unsigned long size = rbio->bio_list_bytes;
529 	int ret = 1;
530 
531 	if (size != rbio->nr_data * rbio->stripe_len)
532 		ret = 0;
533 
534 	BUG_ON(size > rbio->nr_data * rbio->stripe_len);
535 	return ret;
536 }
537 
538 static int rbio_is_full(struct btrfs_raid_bio *rbio)
539 {
540 	unsigned long flags;
541 	int ret;
542 
543 	spin_lock_irqsave(&rbio->bio_list_lock, flags);
544 	ret = __rbio_is_full(rbio);
545 	spin_unlock_irqrestore(&rbio->bio_list_lock, flags);
546 	return ret;
547 }
548 
549 /*
550  * returns 1 if it is safe to merge two rbios together.
551  * The merging is safe if the two rbios correspond to
552  * the same stripe and if they are both going in the same
553  * direction (read vs write), and if neither one is
554  * locked for final IO
555  *
556  * The caller is responsible for locking such that
557  * rmw_locked is safe to test
558  */
559 static int rbio_can_merge(struct btrfs_raid_bio *last,
560 			  struct btrfs_raid_bio *cur)
561 {
562 	if (test_bit(RBIO_RMW_LOCKED_BIT, &last->flags) ||
563 	    test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags))
564 		return 0;
565 
566 	/*
567 	 * we can't merge with cached rbios, since the
568 	 * idea is that when we merge the destination
569 	 * rbio is going to run our IO for us.  We can
570 	 * steal from cached rbio's though, other functions
571 	 * handle that.
572 	 */
573 	if (test_bit(RBIO_CACHE_BIT, &last->flags) ||
574 	    test_bit(RBIO_CACHE_BIT, &cur->flags))
575 		return 0;
576 
577 	if (last->raid_map[0] !=
578 	    cur->raid_map[0])
579 		return 0;
580 
581 	/* reads can't merge with writes */
582 	if (last->read_rebuild !=
583 	    cur->read_rebuild) {
584 		return 0;
585 	}
586 
587 	return 1;
588 }
589 
590 /*
591  * helper to index into the pstripe
592  */
593 static struct page *rbio_pstripe_page(struct btrfs_raid_bio *rbio, int index)
594 {
595 	index += (rbio->nr_data * rbio->stripe_len) >> PAGE_CACHE_SHIFT;
596 	return rbio->stripe_pages[index];
597 }
598 
599 /*
600  * helper to index into the qstripe, returns null
601  * if there is no qstripe
602  */
603 static struct page *rbio_qstripe_page(struct btrfs_raid_bio *rbio, int index)
604 {
605 	if (rbio->nr_data + 1 == rbio->bbio->num_stripes)
606 		return NULL;
607 
608 	index += ((rbio->nr_data + 1) * rbio->stripe_len) >>
609 		PAGE_CACHE_SHIFT;
610 	return rbio->stripe_pages[index];
611 }
612 
613 /*
614  * The first stripe in the table for a logical address
615  * has the lock.  rbios are added in one of three ways:
616  *
617  * 1) Nobody has the stripe locked yet.  The rbio is given
618  * the lock and 0 is returned.  The caller must start the IO
619  * themselves.
620  *
621  * 2) Someone has the stripe locked, but we're able to merge
622  * with the lock owner.  The rbio is freed and the IO will
623  * start automatically along with the existing rbio.  1 is returned.
624  *
625  * 3) Someone has the stripe locked, but we're not able to merge.
626  * The rbio is added to the lock owner's plug list, or merged into
627  * an rbio already on the plug list.  When the lock owner unlocks,
628  * the next rbio on the list is run and the IO is started automatically.
629  * 1 is returned
630  *
631  * If we return 0, the caller still owns the rbio and must continue with
632  * IO submission.  If we return 1, the caller must assume the rbio has
633  * already been freed.
634  */
635 static noinline int lock_stripe_add(struct btrfs_raid_bio *rbio)
636 {
637 	int bucket = rbio_bucket(rbio);
638 	struct btrfs_stripe_hash *h = rbio->fs_info->stripe_hash_table->table + bucket;
639 	struct btrfs_raid_bio *cur;
640 	struct btrfs_raid_bio *pending;
641 	unsigned long flags;
642 	DEFINE_WAIT(wait);
643 	struct btrfs_raid_bio *freeit = NULL;
644 	struct btrfs_raid_bio *cache_drop = NULL;
645 	int ret = 0;
646 	int walk = 0;
647 
648 	spin_lock_irqsave(&h->lock, flags);
649 	list_for_each_entry(cur, &h->hash_list, hash_list) {
650 		walk++;
651 		if (cur->raid_map[0] == rbio->raid_map[0]) {
652 			spin_lock(&cur->bio_list_lock);
653 
654 			/* can we steal this cached rbio's pages? */
655 			if (bio_list_empty(&cur->bio_list) &&
656 			    list_empty(&cur->plug_list) &&
657 			    test_bit(RBIO_CACHE_BIT, &cur->flags) &&
658 			    !test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) {
659 				list_del_init(&cur->hash_list);
660 				atomic_dec(&cur->refs);
661 
662 				steal_rbio(cur, rbio);
663 				cache_drop = cur;
664 				spin_unlock(&cur->bio_list_lock);
665 
666 				goto lockit;
667 			}
668 
669 			/* can we merge into the lock owner? */
670 			if (rbio_can_merge(cur, rbio)) {
671 				merge_rbio(cur, rbio);
672 				spin_unlock(&cur->bio_list_lock);
673 				freeit = rbio;
674 				ret = 1;
675 				goto out;
676 			}
677 
678 
679 			/*
680 			 * we couldn't merge with the running
681 			 * rbio, see if we can merge with the
682 			 * pending ones.  We don't have to
683 			 * check for rmw_locked because there
684 			 * is no way they are inside finish_rmw
685 			 * right now
686 			 */
687 			list_for_each_entry(pending, &cur->plug_list,
688 					    plug_list) {
689 				if (rbio_can_merge(pending, rbio)) {
690 					merge_rbio(pending, rbio);
691 					spin_unlock(&cur->bio_list_lock);
692 					freeit = rbio;
693 					ret = 1;
694 					goto out;
695 				}
696 			}
697 
698 			/* no merging, put us on the tail of the plug list,
699 			 * our rbio will be started with the currently
700 			 * running rbio unlocks
701 			 */
702 			list_add_tail(&rbio->plug_list, &cur->plug_list);
703 			spin_unlock(&cur->bio_list_lock);
704 			ret = 1;
705 			goto out;
706 		}
707 	}
708 lockit:
709 	atomic_inc(&rbio->refs);
710 	list_add(&rbio->hash_list, &h->hash_list);
711 out:
712 	spin_unlock_irqrestore(&h->lock, flags);
713 	if (cache_drop)
714 		remove_rbio_from_cache(cache_drop);
715 	if (freeit)
716 		__free_raid_bio(freeit);
717 	return ret;
718 }
719 
720 /*
721  * called as rmw or parity rebuild is completed.  If the plug list has more
722  * rbios waiting for this stripe, the next one on the list will be started
723  */
724 static noinline void unlock_stripe(struct btrfs_raid_bio *rbio)
725 {
726 	int bucket;
727 	struct btrfs_stripe_hash *h;
728 	unsigned long flags;
729 	int keep_cache = 0;
730 
731 	bucket = rbio_bucket(rbio);
732 	h = rbio->fs_info->stripe_hash_table->table + bucket;
733 
734 	if (list_empty(&rbio->plug_list))
735 		cache_rbio(rbio);
736 
737 	spin_lock_irqsave(&h->lock, flags);
738 	spin_lock(&rbio->bio_list_lock);
739 
740 	if (!list_empty(&rbio->hash_list)) {
741 		/*
742 		 * if we're still cached and there is no other IO
743 		 * to perform, just leave this rbio here for others
744 		 * to steal from later
745 		 */
746 		if (list_empty(&rbio->plug_list) &&
747 		    test_bit(RBIO_CACHE_BIT, &rbio->flags)) {
748 			keep_cache = 1;
749 			clear_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags);
750 			BUG_ON(!bio_list_empty(&rbio->bio_list));
751 			goto done;
752 		}
753 
754 		list_del_init(&rbio->hash_list);
755 		atomic_dec(&rbio->refs);
756 
757 		/*
758 		 * we use the plug list to hold all the rbios
759 		 * waiting for the chance to lock this stripe.
760 		 * hand the lock over to one of them.
761 		 */
762 		if (!list_empty(&rbio->plug_list)) {
763 			struct btrfs_raid_bio *next;
764 			struct list_head *head = rbio->plug_list.next;
765 
766 			next = list_entry(head, struct btrfs_raid_bio,
767 					  plug_list);
768 
769 			list_del_init(&rbio->plug_list);
770 
771 			list_add(&next->hash_list, &h->hash_list);
772 			atomic_inc(&next->refs);
773 			spin_unlock(&rbio->bio_list_lock);
774 			spin_unlock_irqrestore(&h->lock, flags);
775 
776 			if (next->read_rebuild)
777 				async_read_rebuild(next);
778 			else {
779 				steal_rbio(rbio, next);
780 				async_rmw_stripe(next);
781 			}
782 
783 			goto done_nolock;
784 		} else  if (waitqueue_active(&h->wait)) {
785 			spin_unlock(&rbio->bio_list_lock);
786 			spin_unlock_irqrestore(&h->lock, flags);
787 			wake_up(&h->wait);
788 			goto done_nolock;
789 		}
790 	}
791 done:
792 	spin_unlock(&rbio->bio_list_lock);
793 	spin_unlock_irqrestore(&h->lock, flags);
794 
795 done_nolock:
796 	if (!keep_cache)
797 		remove_rbio_from_cache(rbio);
798 }
799 
800 static void __free_raid_bio(struct btrfs_raid_bio *rbio)
801 {
802 	int i;
803 
804 	WARN_ON(atomic_read(&rbio->refs) < 0);
805 	if (!atomic_dec_and_test(&rbio->refs))
806 		return;
807 
808 	WARN_ON(!list_empty(&rbio->stripe_cache));
809 	WARN_ON(!list_empty(&rbio->hash_list));
810 	WARN_ON(!bio_list_empty(&rbio->bio_list));
811 
812 	for (i = 0; i < rbio->nr_pages; i++) {
813 		if (rbio->stripe_pages[i]) {
814 			__free_page(rbio->stripe_pages[i]);
815 			rbio->stripe_pages[i] = NULL;
816 		}
817 	}
818 	kfree(rbio->raid_map);
819 	kfree(rbio->bbio);
820 	kfree(rbio);
821 }
822 
823 static void free_raid_bio(struct btrfs_raid_bio *rbio)
824 {
825 	unlock_stripe(rbio);
826 	__free_raid_bio(rbio);
827 }
828 
829 /*
830  * this frees the rbio and runs through all the bios in the
831  * bio_list and calls end_io on them
832  */
833 static void rbio_orig_end_io(struct btrfs_raid_bio *rbio, int err, int uptodate)
834 {
835 	struct bio *cur = bio_list_get(&rbio->bio_list);
836 	struct bio *next;
837 	free_raid_bio(rbio);
838 
839 	while (cur) {
840 		next = cur->bi_next;
841 		cur->bi_next = NULL;
842 		if (uptodate)
843 			set_bit(BIO_UPTODATE, &cur->bi_flags);
844 		bio_endio(cur, err);
845 		cur = next;
846 	}
847 }
848 
849 /*
850  * end io function used by finish_rmw.  When we finally
851  * get here, we've written a full stripe
852  */
853 static void raid_write_end_io(struct bio *bio, int err)
854 {
855 	struct btrfs_raid_bio *rbio = bio->bi_private;
856 
857 	if (err)
858 		fail_bio_stripe(rbio, bio);
859 
860 	bio_put(bio);
861 
862 	if (!atomic_dec_and_test(&rbio->bbio->stripes_pending))
863 		return;
864 
865 	err = 0;
866 
867 	/* OK, we have read all the stripes we need to. */
868 	if (atomic_read(&rbio->bbio->error) > rbio->bbio->max_errors)
869 		err = -EIO;
870 
871 	rbio_orig_end_io(rbio, err, 0);
872 	return;
873 }
874 
875 /*
876  * the read/modify/write code wants to use the original bio for
877  * any pages it included, and then use the rbio for everything
878  * else.  This function decides if a given index (stripe number)
879  * and page number in that stripe fall inside the original bio
880  * or the rbio.
881  *
882  * if you set bio_list_only, you'll get a NULL back for any ranges
883  * that are outside the bio_list
884  *
885  * This doesn't take any refs on anything, you get a bare page pointer
886  * and the caller must bump refs as required.
887  *
888  * You must call index_rbio_pages once before you can trust
889  * the answers from this function.
890  */
891 static struct page *page_in_rbio(struct btrfs_raid_bio *rbio,
892 				 int index, int pagenr, int bio_list_only)
893 {
894 	int chunk_page;
895 	struct page *p = NULL;
896 
897 	chunk_page = index * (rbio->stripe_len >> PAGE_SHIFT) + pagenr;
898 
899 	spin_lock_irq(&rbio->bio_list_lock);
900 	p = rbio->bio_pages[chunk_page];
901 	spin_unlock_irq(&rbio->bio_list_lock);
902 
903 	if (p || bio_list_only)
904 		return p;
905 
906 	return rbio->stripe_pages[chunk_page];
907 }
908 
909 /*
910  * number of pages we need for the entire stripe across all the
911  * drives
912  */
913 static unsigned long rbio_nr_pages(unsigned long stripe_len, int nr_stripes)
914 {
915 	unsigned long nr = stripe_len * nr_stripes;
916 	return (nr + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
917 }
918 
919 /*
920  * allocation and initial setup for the btrfs_raid_bio.  Not
921  * this does not allocate any pages for rbio->pages.
922  */
923 static struct btrfs_raid_bio *alloc_rbio(struct btrfs_root *root,
924 			  struct btrfs_bio *bbio, u64 *raid_map,
925 			  u64 stripe_len)
926 {
927 	struct btrfs_raid_bio *rbio;
928 	int nr_data = 0;
929 	int num_pages = rbio_nr_pages(stripe_len, bbio->num_stripes);
930 	void *p;
931 
932 	rbio = kzalloc(sizeof(*rbio) + num_pages * sizeof(struct page *) * 2,
933 			GFP_NOFS);
934 	if (!rbio) {
935 		kfree(raid_map);
936 		kfree(bbio);
937 		return ERR_PTR(-ENOMEM);
938 	}
939 
940 	bio_list_init(&rbio->bio_list);
941 	INIT_LIST_HEAD(&rbio->plug_list);
942 	spin_lock_init(&rbio->bio_list_lock);
943 	INIT_LIST_HEAD(&rbio->stripe_cache);
944 	INIT_LIST_HEAD(&rbio->hash_list);
945 	rbio->bbio = bbio;
946 	rbio->raid_map = raid_map;
947 	rbio->fs_info = root->fs_info;
948 	rbio->stripe_len = stripe_len;
949 	rbio->nr_pages = num_pages;
950 	rbio->faila = -1;
951 	rbio->failb = -1;
952 	atomic_set(&rbio->refs, 1);
953 
954 	/*
955 	 * the stripe_pages and bio_pages array point to the extra
956 	 * memory we allocated past the end of the rbio
957 	 */
958 	p = rbio + 1;
959 	rbio->stripe_pages = p;
960 	rbio->bio_pages = p + sizeof(struct page *) * num_pages;
961 
962 	if (raid_map[bbio->num_stripes - 1] == RAID6_Q_STRIPE)
963 		nr_data = bbio->num_stripes - 2;
964 	else
965 		nr_data = bbio->num_stripes - 1;
966 
967 	rbio->nr_data = nr_data;
968 	return rbio;
969 }
970 
971 /* allocate pages for all the stripes in the bio, including parity */
972 static int alloc_rbio_pages(struct btrfs_raid_bio *rbio)
973 {
974 	int i;
975 	struct page *page;
976 
977 	for (i = 0; i < rbio->nr_pages; i++) {
978 		if (rbio->stripe_pages[i])
979 			continue;
980 		page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
981 		if (!page)
982 			return -ENOMEM;
983 		rbio->stripe_pages[i] = page;
984 		ClearPageUptodate(page);
985 	}
986 	return 0;
987 }
988 
989 /* allocate pages for just the p/q stripes */
990 static int alloc_rbio_parity_pages(struct btrfs_raid_bio *rbio)
991 {
992 	int i;
993 	struct page *page;
994 
995 	i = (rbio->nr_data * rbio->stripe_len) >> PAGE_CACHE_SHIFT;
996 
997 	for (; i < rbio->nr_pages; i++) {
998 		if (rbio->stripe_pages[i])
999 			continue;
1000 		page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1001 		if (!page)
1002 			return -ENOMEM;
1003 		rbio->stripe_pages[i] = page;
1004 	}
1005 	return 0;
1006 }
1007 
1008 /*
1009  * add a single page from a specific stripe into our list of bios for IO
1010  * this will try to merge into existing bios if possible, and returns
1011  * zero if all went well.
1012  */
1013 static int rbio_add_io_page(struct btrfs_raid_bio *rbio,
1014 			    struct bio_list *bio_list,
1015 			    struct page *page,
1016 			    int stripe_nr,
1017 			    unsigned long page_index,
1018 			    unsigned long bio_max_len)
1019 {
1020 	struct bio *last = bio_list->tail;
1021 	u64 last_end = 0;
1022 	int ret;
1023 	struct bio *bio;
1024 	struct btrfs_bio_stripe *stripe;
1025 	u64 disk_start;
1026 
1027 	stripe = &rbio->bbio->stripes[stripe_nr];
1028 	disk_start = stripe->physical + (page_index << PAGE_CACHE_SHIFT);
1029 
1030 	/* if the device is missing, just fail this stripe */
1031 	if (!stripe->dev->bdev)
1032 		return fail_rbio_index(rbio, stripe_nr);
1033 
1034 	/* see if we can add this page onto our existing bio */
1035 	if (last) {
1036 		last_end = (u64)last->bi_sector << 9;
1037 		last_end += last->bi_size;
1038 
1039 		/*
1040 		 * we can't merge these if they are from different
1041 		 * devices or if they are not contiguous
1042 		 */
1043 		if (last_end == disk_start && stripe->dev->bdev &&
1044 		    test_bit(BIO_UPTODATE, &last->bi_flags) &&
1045 		    last->bi_bdev == stripe->dev->bdev) {
1046 			ret = bio_add_page(last, page, PAGE_CACHE_SIZE, 0);
1047 			if (ret == PAGE_CACHE_SIZE)
1048 				return 0;
1049 		}
1050 	}
1051 
1052 	/* put a new bio on the list */
1053 	bio = btrfs_io_bio_alloc(GFP_NOFS, bio_max_len >> PAGE_SHIFT?:1);
1054 	if (!bio)
1055 		return -ENOMEM;
1056 
1057 	bio->bi_size = 0;
1058 	bio->bi_bdev = stripe->dev->bdev;
1059 	bio->bi_sector = disk_start >> 9;
1060 	set_bit(BIO_UPTODATE, &bio->bi_flags);
1061 
1062 	bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
1063 	bio_list_add(bio_list, bio);
1064 	return 0;
1065 }
1066 
1067 /*
1068  * while we're doing the read/modify/write cycle, we could
1069  * have errors in reading pages off the disk.  This checks
1070  * for errors and if we're not able to read the page it'll
1071  * trigger parity reconstruction.  The rmw will be finished
1072  * after we've reconstructed the failed stripes
1073  */
1074 static void validate_rbio_for_rmw(struct btrfs_raid_bio *rbio)
1075 {
1076 	if (rbio->faila >= 0 || rbio->failb >= 0) {
1077 		BUG_ON(rbio->faila == rbio->bbio->num_stripes - 1);
1078 		__raid56_parity_recover(rbio);
1079 	} else {
1080 		finish_rmw(rbio);
1081 	}
1082 }
1083 
1084 /*
1085  * these are just the pages from the rbio array, not from anything
1086  * the FS sent down to us
1087  */
1088 static struct page *rbio_stripe_page(struct btrfs_raid_bio *rbio, int stripe, int page)
1089 {
1090 	int index;
1091 	index = stripe * (rbio->stripe_len >> PAGE_CACHE_SHIFT);
1092 	index += page;
1093 	return rbio->stripe_pages[index];
1094 }
1095 
1096 /*
1097  * helper function to walk our bio list and populate the bio_pages array with
1098  * the result.  This seems expensive, but it is faster than constantly
1099  * searching through the bio list as we setup the IO in finish_rmw or stripe
1100  * reconstruction.
1101  *
1102  * This must be called before you trust the answers from page_in_rbio
1103  */
1104 static void index_rbio_pages(struct btrfs_raid_bio *rbio)
1105 {
1106 	struct bio *bio;
1107 	u64 start;
1108 	unsigned long stripe_offset;
1109 	unsigned long page_index;
1110 	struct page *p;
1111 	int i;
1112 
1113 	spin_lock_irq(&rbio->bio_list_lock);
1114 	bio_list_for_each(bio, &rbio->bio_list) {
1115 		start = (u64)bio->bi_sector << 9;
1116 		stripe_offset = start - rbio->raid_map[0];
1117 		page_index = stripe_offset >> PAGE_CACHE_SHIFT;
1118 
1119 		for (i = 0; i < bio->bi_vcnt; i++) {
1120 			p = bio->bi_io_vec[i].bv_page;
1121 			rbio->bio_pages[page_index + i] = p;
1122 		}
1123 	}
1124 	spin_unlock_irq(&rbio->bio_list_lock);
1125 }
1126 
1127 /*
1128  * this is called from one of two situations.  We either
1129  * have a full stripe from the higher layers, or we've read all
1130  * the missing bits off disk.
1131  *
1132  * This will calculate the parity and then send down any
1133  * changed blocks.
1134  */
1135 static noinline void finish_rmw(struct btrfs_raid_bio *rbio)
1136 {
1137 	struct btrfs_bio *bbio = rbio->bbio;
1138 	void *pointers[bbio->num_stripes];
1139 	int stripe_len = rbio->stripe_len;
1140 	int nr_data = rbio->nr_data;
1141 	int stripe;
1142 	int pagenr;
1143 	int p_stripe = -1;
1144 	int q_stripe = -1;
1145 	struct bio_list bio_list;
1146 	struct bio *bio;
1147 	int pages_per_stripe = stripe_len >> PAGE_CACHE_SHIFT;
1148 	int ret;
1149 
1150 	bio_list_init(&bio_list);
1151 
1152 	if (bbio->num_stripes - rbio->nr_data == 1) {
1153 		p_stripe = bbio->num_stripes - 1;
1154 	} else if (bbio->num_stripes - rbio->nr_data == 2) {
1155 		p_stripe = bbio->num_stripes - 2;
1156 		q_stripe = bbio->num_stripes - 1;
1157 	} else {
1158 		BUG();
1159 	}
1160 
1161 	/* at this point we either have a full stripe,
1162 	 * or we've read the full stripe from the drive.
1163 	 * recalculate the parity and write the new results.
1164 	 *
1165 	 * We're not allowed to add any new bios to the
1166 	 * bio list here, anyone else that wants to
1167 	 * change this stripe needs to do their own rmw.
1168 	 */
1169 	spin_lock_irq(&rbio->bio_list_lock);
1170 	set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags);
1171 	spin_unlock_irq(&rbio->bio_list_lock);
1172 
1173 	atomic_set(&rbio->bbio->error, 0);
1174 
1175 	/*
1176 	 * now that we've set rmw_locked, run through the
1177 	 * bio list one last time and map the page pointers
1178 	 *
1179 	 * We don't cache full rbios because we're assuming
1180 	 * the higher layers are unlikely to use this area of
1181 	 * the disk again soon.  If they do use it again,
1182 	 * hopefully they will send another full bio.
1183 	 */
1184 	index_rbio_pages(rbio);
1185 	if (!rbio_is_full(rbio))
1186 		cache_rbio_pages(rbio);
1187 	else
1188 		clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags);
1189 
1190 	for (pagenr = 0; pagenr < pages_per_stripe; pagenr++) {
1191 		struct page *p;
1192 		/* first collect one page from each data stripe */
1193 		for (stripe = 0; stripe < nr_data; stripe++) {
1194 			p = page_in_rbio(rbio, stripe, pagenr, 0);
1195 			pointers[stripe] = kmap(p);
1196 		}
1197 
1198 		/* then add the parity stripe */
1199 		p = rbio_pstripe_page(rbio, pagenr);
1200 		SetPageUptodate(p);
1201 		pointers[stripe++] = kmap(p);
1202 
1203 		if (q_stripe != -1) {
1204 
1205 			/*
1206 			 * raid6, add the qstripe and call the
1207 			 * library function to fill in our p/q
1208 			 */
1209 			p = rbio_qstripe_page(rbio, pagenr);
1210 			SetPageUptodate(p);
1211 			pointers[stripe++] = kmap(p);
1212 
1213 			raid6_call.gen_syndrome(bbio->num_stripes, PAGE_SIZE,
1214 						pointers);
1215 		} else {
1216 			/* raid5 */
1217 			memcpy(pointers[nr_data], pointers[0], PAGE_SIZE);
1218 			run_xor(pointers + 1, nr_data - 1, PAGE_CACHE_SIZE);
1219 		}
1220 
1221 
1222 		for (stripe = 0; stripe < bbio->num_stripes; stripe++)
1223 			kunmap(page_in_rbio(rbio, stripe, pagenr, 0));
1224 	}
1225 
1226 	/*
1227 	 * time to start writing.  Make bios for everything from the
1228 	 * higher layers (the bio_list in our rbio) and our p/q.  Ignore
1229 	 * everything else.
1230 	 */
1231 	for (stripe = 0; stripe < bbio->num_stripes; stripe++) {
1232 		for (pagenr = 0; pagenr < pages_per_stripe; pagenr++) {
1233 			struct page *page;
1234 			if (stripe < rbio->nr_data) {
1235 				page = page_in_rbio(rbio, stripe, pagenr, 1);
1236 				if (!page)
1237 					continue;
1238 			} else {
1239 			       page = rbio_stripe_page(rbio, stripe, pagenr);
1240 			}
1241 
1242 			ret = rbio_add_io_page(rbio, &bio_list,
1243 				       page, stripe, pagenr, rbio->stripe_len);
1244 			if (ret)
1245 				goto cleanup;
1246 		}
1247 	}
1248 
1249 	atomic_set(&bbio->stripes_pending, bio_list_size(&bio_list));
1250 	BUG_ON(atomic_read(&bbio->stripes_pending) == 0);
1251 
1252 	while (1) {
1253 		bio = bio_list_pop(&bio_list);
1254 		if (!bio)
1255 			break;
1256 
1257 		bio->bi_private = rbio;
1258 		bio->bi_end_io = raid_write_end_io;
1259 		BUG_ON(!test_bit(BIO_UPTODATE, &bio->bi_flags));
1260 		submit_bio(WRITE, bio);
1261 	}
1262 	return;
1263 
1264 cleanup:
1265 	rbio_orig_end_io(rbio, -EIO, 0);
1266 }
1267 
1268 /*
1269  * helper to find the stripe number for a given bio.  Used to figure out which
1270  * stripe has failed.  This expects the bio to correspond to a physical disk,
1271  * so it looks up based on physical sector numbers.
1272  */
1273 static int find_bio_stripe(struct btrfs_raid_bio *rbio,
1274 			   struct bio *bio)
1275 {
1276 	u64 physical = bio->bi_sector;
1277 	u64 stripe_start;
1278 	int i;
1279 	struct btrfs_bio_stripe *stripe;
1280 
1281 	physical <<= 9;
1282 
1283 	for (i = 0; i < rbio->bbio->num_stripes; i++) {
1284 		stripe = &rbio->bbio->stripes[i];
1285 		stripe_start = stripe->physical;
1286 		if (physical >= stripe_start &&
1287 		    physical < stripe_start + rbio->stripe_len) {
1288 			return i;
1289 		}
1290 	}
1291 	return -1;
1292 }
1293 
1294 /*
1295  * helper to find the stripe number for a given
1296  * bio (before mapping).  Used to figure out which stripe has
1297  * failed.  This looks up based on logical block numbers.
1298  */
1299 static int find_logical_bio_stripe(struct btrfs_raid_bio *rbio,
1300 				   struct bio *bio)
1301 {
1302 	u64 logical = bio->bi_sector;
1303 	u64 stripe_start;
1304 	int i;
1305 
1306 	logical <<= 9;
1307 
1308 	for (i = 0; i < rbio->nr_data; i++) {
1309 		stripe_start = rbio->raid_map[i];
1310 		if (logical >= stripe_start &&
1311 		    logical < stripe_start + rbio->stripe_len) {
1312 			return i;
1313 		}
1314 	}
1315 	return -1;
1316 }
1317 
1318 /*
1319  * returns -EIO if we had too many failures
1320  */
1321 static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed)
1322 {
1323 	unsigned long flags;
1324 	int ret = 0;
1325 
1326 	spin_lock_irqsave(&rbio->bio_list_lock, flags);
1327 
1328 	/* we already know this stripe is bad, move on */
1329 	if (rbio->faila == failed || rbio->failb == failed)
1330 		goto out;
1331 
1332 	if (rbio->faila == -1) {
1333 		/* first failure on this rbio */
1334 		rbio->faila = failed;
1335 		atomic_inc(&rbio->bbio->error);
1336 	} else if (rbio->failb == -1) {
1337 		/* second failure on this rbio */
1338 		rbio->failb = failed;
1339 		atomic_inc(&rbio->bbio->error);
1340 	} else {
1341 		ret = -EIO;
1342 	}
1343 out:
1344 	spin_unlock_irqrestore(&rbio->bio_list_lock, flags);
1345 
1346 	return ret;
1347 }
1348 
1349 /*
1350  * helper to fail a stripe based on a physical disk
1351  * bio.
1352  */
1353 static int fail_bio_stripe(struct btrfs_raid_bio *rbio,
1354 			   struct bio *bio)
1355 {
1356 	int failed = find_bio_stripe(rbio, bio);
1357 
1358 	if (failed < 0)
1359 		return -EIO;
1360 
1361 	return fail_rbio_index(rbio, failed);
1362 }
1363 
1364 /*
1365  * this sets each page in the bio uptodate.  It should only be used on private
1366  * rbio pages, nothing that comes in from the higher layers
1367  */
1368 static void set_bio_pages_uptodate(struct bio *bio)
1369 {
1370 	int i;
1371 	struct page *p;
1372 
1373 	for (i = 0; i < bio->bi_vcnt; i++) {
1374 		p = bio->bi_io_vec[i].bv_page;
1375 		SetPageUptodate(p);
1376 	}
1377 }
1378 
1379 /*
1380  * end io for the read phase of the rmw cycle.  All the bios here are physical
1381  * stripe bios we've read from the disk so we can recalculate the parity of the
1382  * stripe.
1383  *
1384  * This will usually kick off finish_rmw once all the bios are read in, but it
1385  * may trigger parity reconstruction if we had any errors along the way
1386  */
1387 static void raid_rmw_end_io(struct bio *bio, int err)
1388 {
1389 	struct btrfs_raid_bio *rbio = bio->bi_private;
1390 
1391 	if (err)
1392 		fail_bio_stripe(rbio, bio);
1393 	else
1394 		set_bio_pages_uptodate(bio);
1395 
1396 	bio_put(bio);
1397 
1398 	if (!atomic_dec_and_test(&rbio->bbio->stripes_pending))
1399 		return;
1400 
1401 	err = 0;
1402 	if (atomic_read(&rbio->bbio->error) > rbio->bbio->max_errors)
1403 		goto cleanup;
1404 
1405 	/*
1406 	 * this will normally call finish_rmw to start our write
1407 	 * but if there are any failed stripes we'll reconstruct
1408 	 * from parity first
1409 	 */
1410 	validate_rbio_for_rmw(rbio);
1411 	return;
1412 
1413 cleanup:
1414 
1415 	rbio_orig_end_io(rbio, -EIO, 0);
1416 }
1417 
1418 static void async_rmw_stripe(struct btrfs_raid_bio *rbio)
1419 {
1420 	rbio->work.flags = 0;
1421 	rbio->work.func = rmw_work;
1422 
1423 	btrfs_queue_worker(&rbio->fs_info->rmw_workers,
1424 			   &rbio->work);
1425 }
1426 
1427 static void async_read_rebuild(struct btrfs_raid_bio *rbio)
1428 {
1429 	rbio->work.flags = 0;
1430 	rbio->work.func = read_rebuild_work;
1431 
1432 	btrfs_queue_worker(&rbio->fs_info->rmw_workers,
1433 			   &rbio->work);
1434 }
1435 
1436 /*
1437  * the stripe must be locked by the caller.  It will
1438  * unlock after all the writes are done
1439  */
1440 static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio)
1441 {
1442 	int bios_to_read = 0;
1443 	struct btrfs_bio *bbio = rbio->bbio;
1444 	struct bio_list bio_list;
1445 	int ret;
1446 	int nr_pages = (rbio->stripe_len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1447 	int pagenr;
1448 	int stripe;
1449 	struct bio *bio;
1450 
1451 	bio_list_init(&bio_list);
1452 
1453 	ret = alloc_rbio_pages(rbio);
1454 	if (ret)
1455 		goto cleanup;
1456 
1457 	index_rbio_pages(rbio);
1458 
1459 	atomic_set(&rbio->bbio->error, 0);
1460 	/*
1461 	 * build a list of bios to read all the missing parts of this
1462 	 * stripe
1463 	 */
1464 	for (stripe = 0; stripe < rbio->nr_data; stripe++) {
1465 		for (pagenr = 0; pagenr < nr_pages; pagenr++) {
1466 			struct page *page;
1467 			/*
1468 			 * we want to find all the pages missing from
1469 			 * the rbio and read them from the disk.  If
1470 			 * page_in_rbio finds a page in the bio list
1471 			 * we don't need to read it off the stripe.
1472 			 */
1473 			page = page_in_rbio(rbio, stripe, pagenr, 1);
1474 			if (page)
1475 				continue;
1476 
1477 			page = rbio_stripe_page(rbio, stripe, pagenr);
1478 			/*
1479 			 * the bio cache may have handed us an uptodate
1480 			 * page.  If so, be happy and use it
1481 			 */
1482 			if (PageUptodate(page))
1483 				continue;
1484 
1485 			ret = rbio_add_io_page(rbio, &bio_list, page,
1486 				       stripe, pagenr, rbio->stripe_len);
1487 			if (ret)
1488 				goto cleanup;
1489 		}
1490 	}
1491 
1492 	bios_to_read = bio_list_size(&bio_list);
1493 	if (!bios_to_read) {
1494 		/*
1495 		 * this can happen if others have merged with
1496 		 * us, it means there is nothing left to read.
1497 		 * But if there are missing devices it may not be
1498 		 * safe to do the full stripe write yet.
1499 		 */
1500 		goto finish;
1501 	}
1502 
1503 	/*
1504 	 * the bbio may be freed once we submit the last bio.  Make sure
1505 	 * not to touch it after that
1506 	 */
1507 	atomic_set(&bbio->stripes_pending, bios_to_read);
1508 	while (1) {
1509 		bio = bio_list_pop(&bio_list);
1510 		if (!bio)
1511 			break;
1512 
1513 		bio->bi_private = rbio;
1514 		bio->bi_end_io = raid_rmw_end_io;
1515 
1516 		btrfs_bio_wq_end_io(rbio->fs_info, bio,
1517 				    BTRFS_WQ_ENDIO_RAID56);
1518 
1519 		BUG_ON(!test_bit(BIO_UPTODATE, &bio->bi_flags));
1520 		submit_bio(READ, bio);
1521 	}
1522 	/* the actual write will happen once the reads are done */
1523 	return 0;
1524 
1525 cleanup:
1526 	rbio_orig_end_io(rbio, -EIO, 0);
1527 	return -EIO;
1528 
1529 finish:
1530 	validate_rbio_for_rmw(rbio);
1531 	return 0;
1532 }
1533 
1534 /*
1535  * if the upper layers pass in a full stripe, we thank them by only allocating
1536  * enough pages to hold the parity, and sending it all down quickly.
1537  */
1538 static int full_stripe_write(struct btrfs_raid_bio *rbio)
1539 {
1540 	int ret;
1541 
1542 	ret = alloc_rbio_parity_pages(rbio);
1543 	if (ret)
1544 		return ret;
1545 
1546 	ret = lock_stripe_add(rbio);
1547 	if (ret == 0)
1548 		finish_rmw(rbio);
1549 	return 0;
1550 }
1551 
1552 /*
1553  * partial stripe writes get handed over to async helpers.
1554  * We're really hoping to merge a few more writes into this
1555  * rbio before calculating new parity
1556  */
1557 static int partial_stripe_write(struct btrfs_raid_bio *rbio)
1558 {
1559 	int ret;
1560 
1561 	ret = lock_stripe_add(rbio);
1562 	if (ret == 0)
1563 		async_rmw_stripe(rbio);
1564 	return 0;
1565 }
1566 
1567 /*
1568  * sometimes while we were reading from the drive to
1569  * recalculate parity, enough new bios come into create
1570  * a full stripe.  So we do a check here to see if we can
1571  * go directly to finish_rmw
1572  */
1573 static int __raid56_parity_write(struct btrfs_raid_bio *rbio)
1574 {
1575 	/* head off into rmw land if we don't have a full stripe */
1576 	if (!rbio_is_full(rbio))
1577 		return partial_stripe_write(rbio);
1578 	return full_stripe_write(rbio);
1579 }
1580 
1581 /*
1582  * We use plugging call backs to collect full stripes.
1583  * Any time we get a partial stripe write while plugged
1584  * we collect it into a list.  When the unplug comes down,
1585  * we sort the list by logical block number and merge
1586  * everything we can into the same rbios
1587  */
1588 struct btrfs_plug_cb {
1589 	struct blk_plug_cb cb;
1590 	struct btrfs_fs_info *info;
1591 	struct list_head rbio_list;
1592 	struct btrfs_work work;
1593 };
1594 
1595 /*
1596  * rbios on the plug list are sorted for easier merging.
1597  */
1598 static int plug_cmp(void *priv, struct list_head *a, struct list_head *b)
1599 {
1600 	struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio,
1601 						 plug_list);
1602 	struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio,
1603 						 plug_list);
1604 	u64 a_sector = ra->bio_list.head->bi_sector;
1605 	u64 b_sector = rb->bio_list.head->bi_sector;
1606 
1607 	if (a_sector < b_sector)
1608 		return -1;
1609 	if (a_sector > b_sector)
1610 		return 1;
1611 	return 0;
1612 }
1613 
1614 static void run_plug(struct btrfs_plug_cb *plug)
1615 {
1616 	struct btrfs_raid_bio *cur;
1617 	struct btrfs_raid_bio *last = NULL;
1618 
1619 	/*
1620 	 * sort our plug list then try to merge
1621 	 * everything we can in hopes of creating full
1622 	 * stripes.
1623 	 */
1624 	list_sort(NULL, &plug->rbio_list, plug_cmp);
1625 	while (!list_empty(&plug->rbio_list)) {
1626 		cur = list_entry(plug->rbio_list.next,
1627 				 struct btrfs_raid_bio, plug_list);
1628 		list_del_init(&cur->plug_list);
1629 
1630 		if (rbio_is_full(cur)) {
1631 			/* we have a full stripe, send it down */
1632 			full_stripe_write(cur);
1633 			continue;
1634 		}
1635 		if (last) {
1636 			if (rbio_can_merge(last, cur)) {
1637 				merge_rbio(last, cur);
1638 				__free_raid_bio(cur);
1639 				continue;
1640 
1641 			}
1642 			__raid56_parity_write(last);
1643 		}
1644 		last = cur;
1645 	}
1646 	if (last) {
1647 		__raid56_parity_write(last);
1648 	}
1649 	kfree(plug);
1650 }
1651 
1652 /*
1653  * if the unplug comes from schedule, we have to push the
1654  * work off to a helper thread
1655  */
1656 static void unplug_work(struct btrfs_work *work)
1657 {
1658 	struct btrfs_plug_cb *plug;
1659 	plug = container_of(work, struct btrfs_plug_cb, work);
1660 	run_plug(plug);
1661 }
1662 
1663 static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule)
1664 {
1665 	struct btrfs_plug_cb *plug;
1666 	plug = container_of(cb, struct btrfs_plug_cb, cb);
1667 
1668 	if (from_schedule) {
1669 		plug->work.flags = 0;
1670 		plug->work.func = unplug_work;
1671 		btrfs_queue_worker(&plug->info->rmw_workers,
1672 				   &plug->work);
1673 		return;
1674 	}
1675 	run_plug(plug);
1676 }
1677 
1678 /*
1679  * our main entry point for writes from the rest of the FS.
1680  */
1681 int raid56_parity_write(struct btrfs_root *root, struct bio *bio,
1682 			struct btrfs_bio *bbio, u64 *raid_map,
1683 			u64 stripe_len)
1684 {
1685 	struct btrfs_raid_bio *rbio;
1686 	struct btrfs_plug_cb *plug = NULL;
1687 	struct blk_plug_cb *cb;
1688 
1689 	rbio = alloc_rbio(root, bbio, raid_map, stripe_len);
1690 	if (IS_ERR(rbio)) {
1691 		kfree(raid_map);
1692 		kfree(bbio);
1693 		return PTR_ERR(rbio);
1694 	}
1695 	bio_list_add(&rbio->bio_list, bio);
1696 	rbio->bio_list_bytes = bio->bi_size;
1697 
1698 	/*
1699 	 * don't plug on full rbios, just get them out the door
1700 	 * as quickly as we can
1701 	 */
1702 	if (rbio_is_full(rbio))
1703 		return full_stripe_write(rbio);
1704 
1705 	cb = blk_check_plugged(btrfs_raid_unplug, root->fs_info,
1706 			       sizeof(*plug));
1707 	if (cb) {
1708 		plug = container_of(cb, struct btrfs_plug_cb, cb);
1709 		if (!plug->info) {
1710 			plug->info = root->fs_info;
1711 			INIT_LIST_HEAD(&plug->rbio_list);
1712 		}
1713 		list_add_tail(&rbio->plug_list, &plug->rbio_list);
1714 	} else {
1715 		return __raid56_parity_write(rbio);
1716 	}
1717 	return 0;
1718 }
1719 
1720 /*
1721  * all parity reconstruction happens here.  We've read in everything
1722  * we can find from the drives and this does the heavy lifting of
1723  * sorting the good from the bad.
1724  */
1725 static void __raid_recover_end_io(struct btrfs_raid_bio *rbio)
1726 {
1727 	int pagenr, stripe;
1728 	void **pointers;
1729 	int faila = -1, failb = -1;
1730 	int nr_pages = (rbio->stripe_len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1731 	struct page *page;
1732 	int err;
1733 	int i;
1734 
1735 	pointers = kzalloc(rbio->bbio->num_stripes * sizeof(void *),
1736 			   GFP_NOFS);
1737 	if (!pointers) {
1738 		err = -ENOMEM;
1739 		goto cleanup_io;
1740 	}
1741 
1742 	faila = rbio->faila;
1743 	failb = rbio->failb;
1744 
1745 	if (rbio->read_rebuild) {
1746 		spin_lock_irq(&rbio->bio_list_lock);
1747 		set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags);
1748 		spin_unlock_irq(&rbio->bio_list_lock);
1749 	}
1750 
1751 	index_rbio_pages(rbio);
1752 
1753 	for (pagenr = 0; pagenr < nr_pages; pagenr++) {
1754 		/* setup our array of pointers with pages
1755 		 * from each stripe
1756 		 */
1757 		for (stripe = 0; stripe < rbio->bbio->num_stripes; stripe++) {
1758 			/*
1759 			 * if we're rebuilding a read, we have to use
1760 			 * pages from the bio list
1761 			 */
1762 			if (rbio->read_rebuild &&
1763 			    (stripe == faila || stripe == failb)) {
1764 				page = page_in_rbio(rbio, stripe, pagenr, 0);
1765 			} else {
1766 				page = rbio_stripe_page(rbio, stripe, pagenr);
1767 			}
1768 			pointers[stripe] = kmap(page);
1769 		}
1770 
1771 		/* all raid6 handling here */
1772 		if (rbio->raid_map[rbio->bbio->num_stripes - 1] ==
1773 		    RAID6_Q_STRIPE) {
1774 
1775 			/*
1776 			 * single failure, rebuild from parity raid5
1777 			 * style
1778 			 */
1779 			if (failb < 0) {
1780 				if (faila == rbio->nr_data) {
1781 					/*
1782 					 * Just the P stripe has failed, without
1783 					 * a bad data or Q stripe.
1784 					 * TODO, we should redo the xor here.
1785 					 */
1786 					err = -EIO;
1787 					goto cleanup;
1788 				}
1789 				/*
1790 				 * a single failure in raid6 is rebuilt
1791 				 * in the pstripe code below
1792 				 */
1793 				goto pstripe;
1794 			}
1795 
1796 			/* make sure our ps and qs are in order */
1797 			if (faila > failb) {
1798 				int tmp = failb;
1799 				failb = faila;
1800 				faila = tmp;
1801 			}
1802 
1803 			/* if the q stripe is failed, do a pstripe reconstruction
1804 			 * from the xors.
1805 			 * If both the q stripe and the P stripe are failed, we're
1806 			 * here due to a crc mismatch and we can't give them the
1807 			 * data they want
1808 			 */
1809 			if (rbio->raid_map[failb] == RAID6_Q_STRIPE) {
1810 				if (rbio->raid_map[faila] == RAID5_P_STRIPE) {
1811 					err = -EIO;
1812 					goto cleanup;
1813 				}
1814 				/*
1815 				 * otherwise we have one bad data stripe and
1816 				 * a good P stripe.  raid5!
1817 				 */
1818 				goto pstripe;
1819 			}
1820 
1821 			if (rbio->raid_map[failb] == RAID5_P_STRIPE) {
1822 				raid6_datap_recov(rbio->bbio->num_stripes,
1823 						  PAGE_SIZE, faila, pointers);
1824 			} else {
1825 				raid6_2data_recov(rbio->bbio->num_stripes,
1826 						  PAGE_SIZE, faila, failb,
1827 						  pointers);
1828 			}
1829 		} else {
1830 			void *p;
1831 
1832 			/* rebuild from P stripe here (raid5 or raid6) */
1833 			BUG_ON(failb != -1);
1834 pstripe:
1835 			/* Copy parity block into failed block to start with */
1836 			memcpy(pointers[faila],
1837 			       pointers[rbio->nr_data],
1838 			       PAGE_CACHE_SIZE);
1839 
1840 			/* rearrange the pointer array */
1841 			p = pointers[faila];
1842 			for (stripe = faila; stripe < rbio->nr_data - 1; stripe++)
1843 				pointers[stripe] = pointers[stripe + 1];
1844 			pointers[rbio->nr_data - 1] = p;
1845 
1846 			/* xor in the rest */
1847 			run_xor(pointers, rbio->nr_data - 1, PAGE_CACHE_SIZE);
1848 		}
1849 		/* if we're doing this rebuild as part of an rmw, go through
1850 		 * and set all of our private rbio pages in the
1851 		 * failed stripes as uptodate.  This way finish_rmw will
1852 		 * know they can be trusted.  If this was a read reconstruction,
1853 		 * other endio functions will fiddle the uptodate bits
1854 		 */
1855 		if (!rbio->read_rebuild) {
1856 			for (i = 0;  i < nr_pages; i++) {
1857 				if (faila != -1) {
1858 					page = rbio_stripe_page(rbio, faila, i);
1859 					SetPageUptodate(page);
1860 				}
1861 				if (failb != -1) {
1862 					page = rbio_stripe_page(rbio, failb, i);
1863 					SetPageUptodate(page);
1864 				}
1865 			}
1866 		}
1867 		for (stripe = 0; stripe < rbio->bbio->num_stripes; stripe++) {
1868 			/*
1869 			 * if we're rebuilding a read, we have to use
1870 			 * pages from the bio list
1871 			 */
1872 			if (rbio->read_rebuild &&
1873 			    (stripe == faila || stripe == failb)) {
1874 				page = page_in_rbio(rbio, stripe, pagenr, 0);
1875 			} else {
1876 				page = rbio_stripe_page(rbio, stripe, pagenr);
1877 			}
1878 			kunmap(page);
1879 		}
1880 	}
1881 
1882 	err = 0;
1883 cleanup:
1884 	kfree(pointers);
1885 
1886 cleanup_io:
1887 
1888 	if (rbio->read_rebuild) {
1889 		if (err == 0)
1890 			cache_rbio_pages(rbio);
1891 		else
1892 			clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags);
1893 
1894 		rbio_orig_end_io(rbio, err, err == 0);
1895 	} else if (err == 0) {
1896 		rbio->faila = -1;
1897 		rbio->failb = -1;
1898 		finish_rmw(rbio);
1899 	} else {
1900 		rbio_orig_end_io(rbio, err, 0);
1901 	}
1902 }
1903 
1904 /*
1905  * This is called only for stripes we've read from disk to
1906  * reconstruct the parity.
1907  */
1908 static void raid_recover_end_io(struct bio *bio, int err)
1909 {
1910 	struct btrfs_raid_bio *rbio = bio->bi_private;
1911 
1912 	/*
1913 	 * we only read stripe pages off the disk, set them
1914 	 * up to date if there were no errors
1915 	 */
1916 	if (err)
1917 		fail_bio_stripe(rbio, bio);
1918 	else
1919 		set_bio_pages_uptodate(bio);
1920 	bio_put(bio);
1921 
1922 	if (!atomic_dec_and_test(&rbio->bbio->stripes_pending))
1923 		return;
1924 
1925 	if (atomic_read(&rbio->bbio->error) > rbio->bbio->max_errors)
1926 		rbio_orig_end_io(rbio, -EIO, 0);
1927 	else
1928 		__raid_recover_end_io(rbio);
1929 }
1930 
1931 /*
1932  * reads everything we need off the disk to reconstruct
1933  * the parity. endio handlers trigger final reconstruction
1934  * when the IO is done.
1935  *
1936  * This is used both for reads from the higher layers and for
1937  * parity construction required to finish a rmw cycle.
1938  */
1939 static int __raid56_parity_recover(struct btrfs_raid_bio *rbio)
1940 {
1941 	int bios_to_read = 0;
1942 	struct btrfs_bio *bbio = rbio->bbio;
1943 	struct bio_list bio_list;
1944 	int ret;
1945 	int nr_pages = (rbio->stripe_len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1946 	int pagenr;
1947 	int stripe;
1948 	struct bio *bio;
1949 
1950 	bio_list_init(&bio_list);
1951 
1952 	ret = alloc_rbio_pages(rbio);
1953 	if (ret)
1954 		goto cleanup;
1955 
1956 	atomic_set(&rbio->bbio->error, 0);
1957 
1958 	/*
1959 	 * read everything that hasn't failed.  Thanks to the
1960 	 * stripe cache, it is possible that some or all of these
1961 	 * pages are going to be uptodate.
1962 	 */
1963 	for (stripe = 0; stripe < bbio->num_stripes; stripe++) {
1964 		if (rbio->faila == stripe ||
1965 		    rbio->failb == stripe)
1966 			continue;
1967 
1968 		for (pagenr = 0; pagenr < nr_pages; pagenr++) {
1969 			struct page *p;
1970 
1971 			/*
1972 			 * the rmw code may have already read this
1973 			 * page in
1974 			 */
1975 			p = rbio_stripe_page(rbio, stripe, pagenr);
1976 			if (PageUptodate(p))
1977 				continue;
1978 
1979 			ret = rbio_add_io_page(rbio, &bio_list,
1980 				       rbio_stripe_page(rbio, stripe, pagenr),
1981 				       stripe, pagenr, rbio->stripe_len);
1982 			if (ret < 0)
1983 				goto cleanup;
1984 		}
1985 	}
1986 
1987 	bios_to_read = bio_list_size(&bio_list);
1988 	if (!bios_to_read) {
1989 		/*
1990 		 * we might have no bios to read just because the pages
1991 		 * were up to date, or we might have no bios to read because
1992 		 * the devices were gone.
1993 		 */
1994 		if (atomic_read(&rbio->bbio->error) <= rbio->bbio->max_errors) {
1995 			__raid_recover_end_io(rbio);
1996 			goto out;
1997 		} else {
1998 			goto cleanup;
1999 		}
2000 	}
2001 
2002 	/*
2003 	 * the bbio may be freed once we submit the last bio.  Make sure
2004 	 * not to touch it after that
2005 	 */
2006 	atomic_set(&bbio->stripes_pending, bios_to_read);
2007 	while (1) {
2008 		bio = bio_list_pop(&bio_list);
2009 		if (!bio)
2010 			break;
2011 
2012 		bio->bi_private = rbio;
2013 		bio->bi_end_io = raid_recover_end_io;
2014 
2015 		btrfs_bio_wq_end_io(rbio->fs_info, bio,
2016 				    BTRFS_WQ_ENDIO_RAID56);
2017 
2018 		BUG_ON(!test_bit(BIO_UPTODATE, &bio->bi_flags));
2019 		submit_bio(READ, bio);
2020 	}
2021 out:
2022 	return 0;
2023 
2024 cleanup:
2025 	if (rbio->read_rebuild)
2026 		rbio_orig_end_io(rbio, -EIO, 0);
2027 	return -EIO;
2028 }
2029 
2030 /*
2031  * the main entry point for reads from the higher layers.  This
2032  * is really only called when the normal read path had a failure,
2033  * so we assume the bio they send down corresponds to a failed part
2034  * of the drive.
2035  */
2036 int raid56_parity_recover(struct btrfs_root *root, struct bio *bio,
2037 			  struct btrfs_bio *bbio, u64 *raid_map,
2038 			  u64 stripe_len, int mirror_num)
2039 {
2040 	struct btrfs_raid_bio *rbio;
2041 	int ret;
2042 
2043 	rbio = alloc_rbio(root, bbio, raid_map, stripe_len);
2044 	if (IS_ERR(rbio)) {
2045 		return PTR_ERR(rbio);
2046 	}
2047 
2048 	rbio->read_rebuild = 1;
2049 	bio_list_add(&rbio->bio_list, bio);
2050 	rbio->bio_list_bytes = bio->bi_size;
2051 
2052 	rbio->faila = find_logical_bio_stripe(rbio, bio);
2053 	if (rbio->faila == -1) {
2054 		BUG();
2055 		kfree(rbio);
2056 		return -EIO;
2057 	}
2058 
2059 	/*
2060 	 * reconstruct from the q stripe if they are
2061 	 * asking for mirror 3
2062 	 */
2063 	if (mirror_num == 3)
2064 		rbio->failb = bbio->num_stripes - 2;
2065 
2066 	ret = lock_stripe_add(rbio);
2067 
2068 	/*
2069 	 * __raid56_parity_recover will end the bio with
2070 	 * any errors it hits.  We don't want to return
2071 	 * its error value up the stack because our caller
2072 	 * will end up calling bio_endio with any nonzero
2073 	 * return
2074 	 */
2075 	if (ret == 0)
2076 		__raid56_parity_recover(rbio);
2077 	/*
2078 	 * our rbio has been added to the list of
2079 	 * rbios that will be handled after the
2080 	 * currently lock owner is done
2081 	 */
2082 	return 0;
2083 
2084 }
2085 
2086 static void rmw_work(struct btrfs_work *work)
2087 {
2088 	struct btrfs_raid_bio *rbio;
2089 
2090 	rbio = container_of(work, struct btrfs_raid_bio, work);
2091 	raid56_rmw_stripe(rbio);
2092 }
2093 
2094 static void read_rebuild_work(struct btrfs_work *work)
2095 {
2096 	struct btrfs_raid_bio *rbio;
2097 
2098 	rbio = container_of(work, struct btrfs_raid_bio, work);
2099 	__raid56_parity_recover(rbio);
2100 }
2101