xref: /openbmc/linux/fs/btrfs/raid56.c (revision 609e478b)
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 
64 struct 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 
167 static int __raid56_parity_recover(struct btrfs_raid_bio *rbio);
168 static noinline void finish_rmw(struct btrfs_raid_bio *rbio);
169 static void rmw_work(struct btrfs_work *work);
170 static void read_rebuild_work(struct btrfs_work *work);
171 static void async_rmw_stripe(struct btrfs_raid_bio *rbio);
172 static void async_read_rebuild(struct btrfs_raid_bio *rbio);
173 static int fail_bio_stripe(struct btrfs_raid_bio *rbio, struct bio *bio);
174 static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed);
175 static void __free_raid_bio(struct btrfs_raid_bio *rbio);
176 static void index_rbio_pages(struct btrfs_raid_bio *rbio);
177 static 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  */
183 int 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  */
242 static 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  */
272 static 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  */
291 static 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  */
322 static 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  */
334 static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio)
335 {
336 	int bucket = rbio_bucket(rbio);
337 	struct btrfs_stripe_hash_table *table;
338 	struct btrfs_stripe_hash *h;
339 	int freeit = 0;
340 
341 	/*
342 	 * check the bit again under the hash table lock.
343 	 */
344 	if (!test_bit(RBIO_CACHE_BIT, &rbio->flags))
345 		return;
346 
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  */
394 static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio)
395 {
396 	struct btrfs_stripe_hash_table *table;
397 	unsigned long flags;
398 
399 	if (!test_bit(RBIO_CACHE_BIT, &rbio->flags))
400 		return;
401 
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  */
412 static 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  */
434 void 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  */
457 static 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  */
503 static 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  */
525 static 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 
537 static 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  */
558 static 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  */
592 static 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  */
602 static 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  */
634 static 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 	}
707 lockit:
708 	atomic_inc(&rbio->refs);
709 	list_add(&rbio->hash_list, &h->hash_list);
710 out:
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  */
723 static 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 	}
790 done:
791 	spin_unlock(&rbio->bio_list_lock);
792 	spin_unlock_irqrestore(&h->lock, flags);
793 
794 done_nolock:
795 	if (!keep_cache)
796 		remove_rbio_from_cache(rbio);
797 }
798 
799 static 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 
822 static 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  */
832 static 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  */
852 static 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  */
890 static 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  */
912 static unsigned long rbio_nr_pages(unsigned long stripe_len, int nr_stripes)
913 {
914 	unsigned long nr = stripe_len * nr_stripes;
915 	return DIV_ROUND_UP(nr, PAGE_CACHE_SIZE);
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  */
922 static 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 */
971 static 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 */
989 static 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  */
1012 static 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  */
1073 static 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  */
1087 static 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  */
1103 static 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  */
1134 static 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 
1263 cleanup:
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  */
1272 static 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  */
1298 static 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  */
1320 static 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 	}
1342 out:
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  */
1352 static 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  */
1367 static 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  */
1386 static 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 
1412 cleanup:
1413 
1414 	rbio_orig_end_io(rbio, -EIO, 0);
1415 }
1416 
1417 static void async_rmw_stripe(struct btrfs_raid_bio *rbio)
1418 {
1419 	btrfs_init_work(&rbio->work, btrfs_rmw_helper,
1420 			rmw_work, NULL, NULL);
1421 
1422 	btrfs_queue_work(rbio->fs_info->rmw_workers,
1423 			 &rbio->work);
1424 }
1425 
1426 static void async_read_rebuild(struct btrfs_raid_bio *rbio)
1427 {
1428 	btrfs_init_work(&rbio->work, btrfs_rmw_helper,
1429 			read_rebuild_work, NULL, NULL);
1430 
1431 	btrfs_queue_work(rbio->fs_info->rmw_workers,
1432 			 &rbio->work);
1433 }
1434 
1435 /*
1436  * the stripe must be locked by the caller.  It will
1437  * unlock after all the writes are done
1438  */
1439 static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio)
1440 {
1441 	int bios_to_read = 0;
1442 	struct btrfs_bio *bbio = rbio->bbio;
1443 	struct bio_list bio_list;
1444 	int ret;
1445 	int nr_pages = DIV_ROUND_UP(rbio->stripe_len, PAGE_CACHE_SIZE);
1446 	int pagenr;
1447 	int stripe;
1448 	struct bio *bio;
1449 
1450 	bio_list_init(&bio_list);
1451 
1452 	ret = alloc_rbio_pages(rbio);
1453 	if (ret)
1454 		goto cleanup;
1455 
1456 	index_rbio_pages(rbio);
1457 
1458 	atomic_set(&rbio->bbio->error, 0);
1459 	/*
1460 	 * build a list of bios to read all the missing parts of this
1461 	 * stripe
1462 	 */
1463 	for (stripe = 0; stripe < rbio->nr_data; stripe++) {
1464 		for (pagenr = 0; pagenr < nr_pages; pagenr++) {
1465 			struct page *page;
1466 			/*
1467 			 * we want to find all the pages missing from
1468 			 * the rbio and read them from the disk.  If
1469 			 * page_in_rbio finds a page in the bio list
1470 			 * we don't need to read it off the stripe.
1471 			 */
1472 			page = page_in_rbio(rbio, stripe, pagenr, 1);
1473 			if (page)
1474 				continue;
1475 
1476 			page = rbio_stripe_page(rbio, stripe, pagenr);
1477 			/*
1478 			 * the bio cache may have handed us an uptodate
1479 			 * page.  If so, be happy and use it
1480 			 */
1481 			if (PageUptodate(page))
1482 				continue;
1483 
1484 			ret = rbio_add_io_page(rbio, &bio_list, page,
1485 				       stripe, pagenr, rbio->stripe_len);
1486 			if (ret)
1487 				goto cleanup;
1488 		}
1489 	}
1490 
1491 	bios_to_read = bio_list_size(&bio_list);
1492 	if (!bios_to_read) {
1493 		/*
1494 		 * this can happen if others have merged with
1495 		 * us, it means there is nothing left to read.
1496 		 * But if there are missing devices it may not be
1497 		 * safe to do the full stripe write yet.
1498 		 */
1499 		goto finish;
1500 	}
1501 
1502 	/*
1503 	 * the bbio may be freed once we submit the last bio.  Make sure
1504 	 * not to touch it after that
1505 	 */
1506 	atomic_set(&bbio->stripes_pending, bios_to_read);
1507 	while (1) {
1508 		bio = bio_list_pop(&bio_list);
1509 		if (!bio)
1510 			break;
1511 
1512 		bio->bi_private = rbio;
1513 		bio->bi_end_io = raid_rmw_end_io;
1514 
1515 		btrfs_bio_wq_end_io(rbio->fs_info, bio,
1516 				    BTRFS_WQ_ENDIO_RAID56);
1517 
1518 		BUG_ON(!test_bit(BIO_UPTODATE, &bio->bi_flags));
1519 		submit_bio(READ, bio);
1520 	}
1521 	/* the actual write will happen once the reads are done */
1522 	return 0;
1523 
1524 cleanup:
1525 	rbio_orig_end_io(rbio, -EIO, 0);
1526 	return -EIO;
1527 
1528 finish:
1529 	validate_rbio_for_rmw(rbio);
1530 	return 0;
1531 }
1532 
1533 /*
1534  * if the upper layers pass in a full stripe, we thank them by only allocating
1535  * enough pages to hold the parity, and sending it all down quickly.
1536  */
1537 static int full_stripe_write(struct btrfs_raid_bio *rbio)
1538 {
1539 	int ret;
1540 
1541 	ret = alloc_rbio_parity_pages(rbio);
1542 	if (ret) {
1543 		__free_raid_bio(rbio);
1544 		return ret;
1545 	}
1546 
1547 	ret = lock_stripe_add(rbio);
1548 	if (ret == 0)
1549 		finish_rmw(rbio);
1550 	return 0;
1551 }
1552 
1553 /*
1554  * partial stripe writes get handed over to async helpers.
1555  * We're really hoping to merge a few more writes into this
1556  * rbio before calculating new parity
1557  */
1558 static int partial_stripe_write(struct btrfs_raid_bio *rbio)
1559 {
1560 	int ret;
1561 
1562 	ret = lock_stripe_add(rbio);
1563 	if (ret == 0)
1564 		async_rmw_stripe(rbio);
1565 	return 0;
1566 }
1567 
1568 /*
1569  * sometimes while we were reading from the drive to
1570  * recalculate parity, enough new bios come into create
1571  * a full stripe.  So we do a check here to see if we can
1572  * go directly to finish_rmw
1573  */
1574 static int __raid56_parity_write(struct btrfs_raid_bio *rbio)
1575 {
1576 	/* head off into rmw land if we don't have a full stripe */
1577 	if (!rbio_is_full(rbio))
1578 		return partial_stripe_write(rbio);
1579 	return full_stripe_write(rbio);
1580 }
1581 
1582 /*
1583  * We use plugging call backs to collect full stripes.
1584  * Any time we get a partial stripe write while plugged
1585  * we collect it into a list.  When the unplug comes down,
1586  * we sort the list by logical block number and merge
1587  * everything we can into the same rbios
1588  */
1589 struct btrfs_plug_cb {
1590 	struct blk_plug_cb cb;
1591 	struct btrfs_fs_info *info;
1592 	struct list_head rbio_list;
1593 	struct btrfs_work work;
1594 };
1595 
1596 /*
1597  * rbios on the plug list are sorted for easier merging.
1598  */
1599 static int plug_cmp(void *priv, struct list_head *a, struct list_head *b)
1600 {
1601 	struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio,
1602 						 plug_list);
1603 	struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio,
1604 						 plug_list);
1605 	u64 a_sector = ra->bio_list.head->bi_iter.bi_sector;
1606 	u64 b_sector = rb->bio_list.head->bi_iter.bi_sector;
1607 
1608 	if (a_sector < b_sector)
1609 		return -1;
1610 	if (a_sector > b_sector)
1611 		return 1;
1612 	return 0;
1613 }
1614 
1615 static void run_plug(struct btrfs_plug_cb *plug)
1616 {
1617 	struct btrfs_raid_bio *cur;
1618 	struct btrfs_raid_bio *last = NULL;
1619 
1620 	/*
1621 	 * sort our plug list then try to merge
1622 	 * everything we can in hopes of creating full
1623 	 * stripes.
1624 	 */
1625 	list_sort(NULL, &plug->rbio_list, plug_cmp);
1626 	while (!list_empty(&plug->rbio_list)) {
1627 		cur = list_entry(plug->rbio_list.next,
1628 				 struct btrfs_raid_bio, plug_list);
1629 		list_del_init(&cur->plug_list);
1630 
1631 		if (rbio_is_full(cur)) {
1632 			/* we have a full stripe, send it down */
1633 			full_stripe_write(cur);
1634 			continue;
1635 		}
1636 		if (last) {
1637 			if (rbio_can_merge(last, cur)) {
1638 				merge_rbio(last, cur);
1639 				__free_raid_bio(cur);
1640 				continue;
1641 
1642 			}
1643 			__raid56_parity_write(last);
1644 		}
1645 		last = cur;
1646 	}
1647 	if (last) {
1648 		__raid56_parity_write(last);
1649 	}
1650 	kfree(plug);
1651 }
1652 
1653 /*
1654  * if the unplug comes from schedule, we have to push the
1655  * work off to a helper thread
1656  */
1657 static void unplug_work(struct btrfs_work *work)
1658 {
1659 	struct btrfs_plug_cb *plug;
1660 	plug = container_of(work, struct btrfs_plug_cb, work);
1661 	run_plug(plug);
1662 }
1663 
1664 static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule)
1665 {
1666 	struct btrfs_plug_cb *plug;
1667 	plug = container_of(cb, struct btrfs_plug_cb, cb);
1668 
1669 	if (from_schedule) {
1670 		btrfs_init_work(&plug->work, btrfs_rmw_helper,
1671 				unplug_work, NULL, NULL);
1672 		btrfs_queue_work(plug->info->rmw_workers,
1673 				 &plug->work);
1674 		return;
1675 	}
1676 	run_plug(plug);
1677 }
1678 
1679 /*
1680  * our main entry point for writes from the rest of the FS.
1681  */
1682 int raid56_parity_write(struct btrfs_root *root, struct bio *bio,
1683 			struct btrfs_bio *bbio, u64 *raid_map,
1684 			u64 stripe_len)
1685 {
1686 	struct btrfs_raid_bio *rbio;
1687 	struct btrfs_plug_cb *plug = NULL;
1688 	struct blk_plug_cb *cb;
1689 
1690 	rbio = alloc_rbio(root, bbio, raid_map, stripe_len);
1691 	if (IS_ERR(rbio))
1692 		return PTR_ERR(rbio);
1693 	bio_list_add(&rbio->bio_list, bio);
1694 	rbio->bio_list_bytes = bio->bi_iter.bi_size;
1695 
1696 	/*
1697 	 * don't plug on full rbios, just get them out the door
1698 	 * as quickly as we can
1699 	 */
1700 	if (rbio_is_full(rbio))
1701 		return full_stripe_write(rbio);
1702 
1703 	cb = blk_check_plugged(btrfs_raid_unplug, root->fs_info,
1704 			       sizeof(*plug));
1705 	if (cb) {
1706 		plug = container_of(cb, struct btrfs_plug_cb, cb);
1707 		if (!plug->info) {
1708 			plug->info = root->fs_info;
1709 			INIT_LIST_HEAD(&plug->rbio_list);
1710 		}
1711 		list_add_tail(&rbio->plug_list, &plug->rbio_list);
1712 	} else {
1713 		return __raid56_parity_write(rbio);
1714 	}
1715 	return 0;
1716 }
1717 
1718 /*
1719  * all parity reconstruction happens here.  We've read in everything
1720  * we can find from the drives and this does the heavy lifting of
1721  * sorting the good from the bad.
1722  */
1723 static void __raid_recover_end_io(struct btrfs_raid_bio *rbio)
1724 {
1725 	int pagenr, stripe;
1726 	void **pointers;
1727 	int faila = -1, failb = -1;
1728 	int nr_pages = DIV_ROUND_UP(rbio->stripe_len, PAGE_CACHE_SIZE);
1729 	struct page *page;
1730 	int err;
1731 	int i;
1732 
1733 	pointers = kzalloc(rbio->bbio->num_stripes * sizeof(void *),
1734 			   GFP_NOFS);
1735 	if (!pointers) {
1736 		err = -ENOMEM;
1737 		goto cleanup_io;
1738 	}
1739 
1740 	faila = rbio->faila;
1741 	failb = rbio->failb;
1742 
1743 	if (rbio->read_rebuild) {
1744 		spin_lock_irq(&rbio->bio_list_lock);
1745 		set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags);
1746 		spin_unlock_irq(&rbio->bio_list_lock);
1747 	}
1748 
1749 	index_rbio_pages(rbio);
1750 
1751 	for (pagenr = 0; pagenr < nr_pages; pagenr++) {
1752 		/* setup our array of pointers with pages
1753 		 * from each stripe
1754 		 */
1755 		for (stripe = 0; stripe < rbio->bbio->num_stripes; stripe++) {
1756 			/*
1757 			 * if we're rebuilding a read, we have to use
1758 			 * pages from the bio list
1759 			 */
1760 			if (rbio->read_rebuild &&
1761 			    (stripe == faila || stripe == failb)) {
1762 				page = page_in_rbio(rbio, stripe, pagenr, 0);
1763 			} else {
1764 				page = rbio_stripe_page(rbio, stripe, pagenr);
1765 			}
1766 			pointers[stripe] = kmap(page);
1767 		}
1768 
1769 		/* all raid6 handling here */
1770 		if (rbio->raid_map[rbio->bbio->num_stripes - 1] ==
1771 		    RAID6_Q_STRIPE) {
1772 
1773 			/*
1774 			 * single failure, rebuild from parity raid5
1775 			 * style
1776 			 */
1777 			if (failb < 0) {
1778 				if (faila == rbio->nr_data) {
1779 					/*
1780 					 * Just the P stripe has failed, without
1781 					 * a bad data or Q stripe.
1782 					 * TODO, we should redo the xor here.
1783 					 */
1784 					err = -EIO;
1785 					goto cleanup;
1786 				}
1787 				/*
1788 				 * a single failure in raid6 is rebuilt
1789 				 * in the pstripe code below
1790 				 */
1791 				goto pstripe;
1792 			}
1793 
1794 			/* make sure our ps and qs are in order */
1795 			if (faila > failb) {
1796 				int tmp = failb;
1797 				failb = faila;
1798 				faila = tmp;
1799 			}
1800 
1801 			/* if the q stripe is failed, do a pstripe reconstruction
1802 			 * from the xors.
1803 			 * If both the q stripe and the P stripe are failed, we're
1804 			 * here due to a crc mismatch and we can't give them the
1805 			 * data they want
1806 			 */
1807 			if (rbio->raid_map[failb] == RAID6_Q_STRIPE) {
1808 				if (rbio->raid_map[faila] == RAID5_P_STRIPE) {
1809 					err = -EIO;
1810 					goto cleanup;
1811 				}
1812 				/*
1813 				 * otherwise we have one bad data stripe and
1814 				 * a good P stripe.  raid5!
1815 				 */
1816 				goto pstripe;
1817 			}
1818 
1819 			if (rbio->raid_map[failb] == RAID5_P_STRIPE) {
1820 				raid6_datap_recov(rbio->bbio->num_stripes,
1821 						  PAGE_SIZE, faila, pointers);
1822 			} else {
1823 				raid6_2data_recov(rbio->bbio->num_stripes,
1824 						  PAGE_SIZE, faila, failb,
1825 						  pointers);
1826 			}
1827 		} else {
1828 			void *p;
1829 
1830 			/* rebuild from P stripe here (raid5 or raid6) */
1831 			BUG_ON(failb != -1);
1832 pstripe:
1833 			/* Copy parity block into failed block to start with */
1834 			memcpy(pointers[faila],
1835 			       pointers[rbio->nr_data],
1836 			       PAGE_CACHE_SIZE);
1837 
1838 			/* rearrange the pointer array */
1839 			p = pointers[faila];
1840 			for (stripe = faila; stripe < rbio->nr_data - 1; stripe++)
1841 				pointers[stripe] = pointers[stripe + 1];
1842 			pointers[rbio->nr_data - 1] = p;
1843 
1844 			/* xor in the rest */
1845 			run_xor(pointers, rbio->nr_data - 1, PAGE_CACHE_SIZE);
1846 		}
1847 		/* if we're doing this rebuild as part of an rmw, go through
1848 		 * and set all of our private rbio pages in the
1849 		 * failed stripes as uptodate.  This way finish_rmw will
1850 		 * know they can be trusted.  If this was a read reconstruction,
1851 		 * other endio functions will fiddle the uptodate bits
1852 		 */
1853 		if (!rbio->read_rebuild) {
1854 			for (i = 0;  i < nr_pages; i++) {
1855 				if (faila != -1) {
1856 					page = rbio_stripe_page(rbio, faila, i);
1857 					SetPageUptodate(page);
1858 				}
1859 				if (failb != -1) {
1860 					page = rbio_stripe_page(rbio, failb, i);
1861 					SetPageUptodate(page);
1862 				}
1863 			}
1864 		}
1865 		for (stripe = 0; stripe < rbio->bbio->num_stripes; stripe++) {
1866 			/*
1867 			 * if we're rebuilding a read, we have to use
1868 			 * pages from the bio list
1869 			 */
1870 			if (rbio->read_rebuild &&
1871 			    (stripe == faila || stripe == failb)) {
1872 				page = page_in_rbio(rbio, stripe, pagenr, 0);
1873 			} else {
1874 				page = rbio_stripe_page(rbio, stripe, pagenr);
1875 			}
1876 			kunmap(page);
1877 		}
1878 	}
1879 
1880 	err = 0;
1881 cleanup:
1882 	kfree(pointers);
1883 
1884 cleanup_io:
1885 
1886 	if (rbio->read_rebuild) {
1887 		if (err == 0)
1888 			cache_rbio_pages(rbio);
1889 		else
1890 			clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags);
1891 
1892 		rbio_orig_end_io(rbio, err, err == 0);
1893 	} else if (err == 0) {
1894 		rbio->faila = -1;
1895 		rbio->failb = -1;
1896 		finish_rmw(rbio);
1897 	} else {
1898 		rbio_orig_end_io(rbio, err, 0);
1899 	}
1900 }
1901 
1902 /*
1903  * This is called only for stripes we've read from disk to
1904  * reconstruct the parity.
1905  */
1906 static void raid_recover_end_io(struct bio *bio, int err)
1907 {
1908 	struct btrfs_raid_bio *rbio = bio->bi_private;
1909 
1910 	/*
1911 	 * we only read stripe pages off the disk, set them
1912 	 * up to date if there were no errors
1913 	 */
1914 	if (err)
1915 		fail_bio_stripe(rbio, bio);
1916 	else
1917 		set_bio_pages_uptodate(bio);
1918 	bio_put(bio);
1919 
1920 	if (!atomic_dec_and_test(&rbio->bbio->stripes_pending))
1921 		return;
1922 
1923 	if (atomic_read(&rbio->bbio->error) > rbio->bbio->max_errors)
1924 		rbio_orig_end_io(rbio, -EIO, 0);
1925 	else
1926 		__raid_recover_end_io(rbio);
1927 }
1928 
1929 /*
1930  * reads everything we need off the disk to reconstruct
1931  * the parity. endio handlers trigger final reconstruction
1932  * when the IO is done.
1933  *
1934  * This is used both for reads from the higher layers and for
1935  * parity construction required to finish a rmw cycle.
1936  */
1937 static int __raid56_parity_recover(struct btrfs_raid_bio *rbio)
1938 {
1939 	int bios_to_read = 0;
1940 	struct btrfs_bio *bbio = rbio->bbio;
1941 	struct bio_list bio_list;
1942 	int ret;
1943 	int nr_pages = DIV_ROUND_UP(rbio->stripe_len, PAGE_CACHE_SIZE);
1944 	int pagenr;
1945 	int stripe;
1946 	struct bio *bio;
1947 
1948 	bio_list_init(&bio_list);
1949 
1950 	ret = alloc_rbio_pages(rbio);
1951 	if (ret)
1952 		goto cleanup;
1953 
1954 	atomic_set(&rbio->bbio->error, 0);
1955 
1956 	/*
1957 	 * read everything that hasn't failed.  Thanks to the
1958 	 * stripe cache, it is possible that some or all of these
1959 	 * pages are going to be uptodate.
1960 	 */
1961 	for (stripe = 0; stripe < bbio->num_stripes; stripe++) {
1962 		if (rbio->faila == stripe || rbio->failb == stripe) {
1963 			atomic_inc(&rbio->bbio->error);
1964 			continue;
1965 		}
1966 
1967 		for (pagenr = 0; pagenr < nr_pages; pagenr++) {
1968 			struct page *p;
1969 
1970 			/*
1971 			 * the rmw code may have already read this
1972 			 * page in
1973 			 */
1974 			p = rbio_stripe_page(rbio, stripe, pagenr);
1975 			if (PageUptodate(p))
1976 				continue;
1977 
1978 			ret = rbio_add_io_page(rbio, &bio_list,
1979 				       rbio_stripe_page(rbio, stripe, pagenr),
1980 				       stripe, pagenr, rbio->stripe_len);
1981 			if (ret < 0)
1982 				goto cleanup;
1983 		}
1984 	}
1985 
1986 	bios_to_read = bio_list_size(&bio_list);
1987 	if (!bios_to_read) {
1988 		/*
1989 		 * we might have no bios to read just because the pages
1990 		 * were up to date, or we might have no bios to read because
1991 		 * the devices were gone.
1992 		 */
1993 		if (atomic_read(&rbio->bbio->error) <= rbio->bbio->max_errors) {
1994 			__raid_recover_end_io(rbio);
1995 			goto out;
1996 		} else {
1997 			goto cleanup;
1998 		}
1999 	}
2000 
2001 	/*
2002 	 * the bbio may be freed once we submit the last bio.  Make sure
2003 	 * not to touch it after that
2004 	 */
2005 	atomic_set(&bbio->stripes_pending, bios_to_read);
2006 	while (1) {
2007 		bio = bio_list_pop(&bio_list);
2008 		if (!bio)
2009 			break;
2010 
2011 		bio->bi_private = rbio;
2012 		bio->bi_end_io = raid_recover_end_io;
2013 
2014 		btrfs_bio_wq_end_io(rbio->fs_info, bio,
2015 				    BTRFS_WQ_ENDIO_RAID56);
2016 
2017 		BUG_ON(!test_bit(BIO_UPTODATE, &bio->bi_flags));
2018 		submit_bio(READ, bio);
2019 	}
2020 out:
2021 	return 0;
2022 
2023 cleanup:
2024 	if (rbio->read_rebuild)
2025 		rbio_orig_end_io(rbio, -EIO, 0);
2026 	return -EIO;
2027 }
2028 
2029 /*
2030  * the main entry point for reads from the higher layers.  This
2031  * is really only called when the normal read path had a failure,
2032  * so we assume the bio they send down corresponds to a failed part
2033  * of the drive.
2034  */
2035 int raid56_parity_recover(struct btrfs_root *root, struct bio *bio,
2036 			  struct btrfs_bio *bbio, u64 *raid_map,
2037 			  u64 stripe_len, int mirror_num)
2038 {
2039 	struct btrfs_raid_bio *rbio;
2040 	int ret;
2041 
2042 	rbio = alloc_rbio(root, bbio, raid_map, stripe_len);
2043 	if (IS_ERR(rbio))
2044 		return PTR_ERR(rbio);
2045 
2046 	rbio->read_rebuild = 1;
2047 	bio_list_add(&rbio->bio_list, bio);
2048 	rbio->bio_list_bytes = bio->bi_iter.bi_size;
2049 
2050 	rbio->faila = find_logical_bio_stripe(rbio, bio);
2051 	if (rbio->faila == -1) {
2052 		BUG();
2053 		kfree(raid_map);
2054 		kfree(bbio);
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