~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/btrfs/extent_io.c

Version: ~ [ linux-5.16 ] ~ [ linux-5.15.13 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.90 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.170 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.224 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.261 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.296 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.298 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 #include <linux/bitops.h>
  2 #include <linux/slab.h>
  3 #include <linux/bio.h>
  4 #include <linux/mm.h>
  5 #include <linux/pagemap.h>
  6 #include <linux/page-flags.h>
  7 #include <linux/spinlock.h>
  8 #include <linux/blkdev.h>
  9 #include <linux/swap.h>
 10 #include <linux/writeback.h>
 11 #include <linux/pagevec.h>
 12 #include <linux/prefetch.h>
 13 #include <linux/cleancache.h>
 14 #include "extent_io.h"
 15 #include "extent_map.h"
 16 #include "compat.h"
 17 #include "ctree.h"
 18 #include "btrfs_inode.h"
 19 #include "volumes.h"
 20 #include "check-integrity.h"
 21 #include "locking.h"
 22 #include "rcu-string.h"
 23 
 24 static struct kmem_cache *extent_state_cache;
 25 static struct kmem_cache *extent_buffer_cache;
 26 static struct bio_set *btrfs_bioset;
 27 
 28 #ifdef CONFIG_BTRFS_DEBUG
 29 static LIST_HEAD(buffers);
 30 static LIST_HEAD(states);
 31 
 32 static DEFINE_SPINLOCK(leak_lock);
 33 
 34 static inline
 35 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
 36 {
 37         unsigned long flags;
 38 
 39         spin_lock_irqsave(&leak_lock, flags);
 40         list_add(new, head);
 41         spin_unlock_irqrestore(&leak_lock, flags);
 42 }
 43 
 44 static inline
 45 void btrfs_leak_debug_del(struct list_head *entry)
 46 {
 47         unsigned long flags;
 48 
 49         spin_lock_irqsave(&leak_lock, flags);
 50         list_del(entry);
 51         spin_unlock_irqrestore(&leak_lock, flags);
 52 }
 53 
 54 static inline
 55 void btrfs_leak_debug_check(void)
 56 {
 57         struct extent_state *state;
 58         struct extent_buffer *eb;
 59 
 60         while (!list_empty(&states)) {
 61                 state = list_entry(states.next, struct extent_state, leak_list);
 62                 printk(KERN_ERR "btrfs state leak: start %llu end %llu "
 63                        "state %lu in tree %p refs %d\n",
 64                        (unsigned long long)state->start,
 65                        (unsigned long long)state->end,
 66                        state->state, state->tree, atomic_read(&state->refs));
 67                 list_del(&state->leak_list);
 68                 kmem_cache_free(extent_state_cache, state);
 69         }
 70 
 71         while (!list_empty(&buffers)) {
 72                 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
 73                 printk(KERN_ERR "btrfs buffer leak start %llu len %lu "
 74                        "refs %d\n", (unsigned long long)eb->start,
 75                        eb->len, atomic_read(&eb->refs));
 76                 list_del(&eb->leak_list);
 77                 kmem_cache_free(extent_buffer_cache, eb);
 78         }
 79 }
 80 #else
 81 #define btrfs_leak_debug_add(new, head) do {} while (0)
 82 #define btrfs_leak_debug_del(entry)     do {} while (0)
 83 #define btrfs_leak_debug_check()        do {} while (0)
 84 #endif
 85 
 86 #define BUFFER_LRU_MAX 64
 87 
 88 struct tree_entry {
 89         u64 start;
 90         u64 end;
 91         struct rb_node rb_node;
 92 };
 93 
 94 struct extent_page_data {
 95         struct bio *bio;
 96         struct extent_io_tree *tree;
 97         get_extent_t *get_extent;
 98         unsigned long bio_flags;
 99 
100         /* tells writepage not to lock the state bits for this range
101          * it still does the unlocking
102          */
103         unsigned int extent_locked:1;
104 
105         /* tells the submit_bio code to use a WRITE_SYNC */
106         unsigned int sync_io:1;
107 };
108 
109 static noinline void flush_write_bio(void *data);
110 static inline struct btrfs_fs_info *
111 tree_fs_info(struct extent_io_tree *tree)
112 {
113         return btrfs_sb(tree->mapping->host->i_sb);
114 }
115 
116 int __init extent_io_init(void)
117 {
118         extent_state_cache = kmem_cache_create("btrfs_extent_state",
119                         sizeof(struct extent_state), 0,
120                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
121         if (!extent_state_cache)
122                 return -ENOMEM;
123 
124         extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
125                         sizeof(struct extent_buffer), 0,
126                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
127         if (!extent_buffer_cache)
128                 goto free_state_cache;
129 
130         btrfs_bioset = bioset_create(BIO_POOL_SIZE,
131                                      offsetof(struct btrfs_io_bio, bio));
132         if (!btrfs_bioset)
133                 goto free_buffer_cache;
134         return 0;
135 
136 free_buffer_cache:
137         kmem_cache_destroy(extent_buffer_cache);
138         extent_buffer_cache = NULL;
139 
140 free_state_cache:
141         kmem_cache_destroy(extent_state_cache);
142         extent_state_cache = NULL;
143         return -ENOMEM;
144 }
145 
146 void extent_io_exit(void)
147 {
148         btrfs_leak_debug_check();
149 
150         /*
151          * Make sure all delayed rcu free are flushed before we
152          * destroy caches.
153          */
154         rcu_barrier();
155         if (extent_state_cache)
156                 kmem_cache_destroy(extent_state_cache);
157         if (extent_buffer_cache)
158                 kmem_cache_destroy(extent_buffer_cache);
159         if (btrfs_bioset)
160                 bioset_free(btrfs_bioset);
161 }
162 
163 void extent_io_tree_init(struct extent_io_tree *tree,
164                          struct address_space *mapping)
165 {
166         tree->state = RB_ROOT;
167         INIT_RADIX_TREE(&tree->buffer, GFP_ATOMIC);
168         tree->ops = NULL;
169         tree->dirty_bytes = 0;
170         spin_lock_init(&tree->lock);
171         spin_lock_init(&tree->buffer_lock);
172         tree->mapping = mapping;
173 }
174 
175 static struct extent_state *alloc_extent_state(gfp_t mask)
176 {
177         struct extent_state *state;
178 
179         state = kmem_cache_alloc(extent_state_cache, mask);
180         if (!state)
181                 return state;
182         state->state = 0;
183         state->private = 0;
184         state->tree = NULL;
185         btrfs_leak_debug_add(&state->leak_list, &states);
186         atomic_set(&state->refs, 1);
187         init_waitqueue_head(&state->wq);
188         trace_alloc_extent_state(state, mask, _RET_IP_);
189         return state;
190 }
191 
192 void free_extent_state(struct extent_state *state)
193 {
194         if (!state)
195                 return;
196         if (atomic_dec_and_test(&state->refs)) {
197                 WARN_ON(state->tree);
198                 btrfs_leak_debug_del(&state->leak_list);
199                 trace_free_extent_state(state, _RET_IP_);
200                 kmem_cache_free(extent_state_cache, state);
201         }
202 }
203 
204 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
205                                    struct rb_node *node)
206 {
207         struct rb_node **p = &root->rb_node;
208         struct rb_node *parent = NULL;
209         struct tree_entry *entry;
210 
211         while (*p) {
212                 parent = *p;
213                 entry = rb_entry(parent, struct tree_entry, rb_node);
214 
215                 if (offset < entry->start)
216                         p = &(*p)->rb_left;
217                 else if (offset > entry->end)
218                         p = &(*p)->rb_right;
219                 else
220                         return parent;
221         }
222 
223         rb_link_node(node, parent, p);
224         rb_insert_color(node, root);
225         return NULL;
226 }
227 
228 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
229                                      struct rb_node **prev_ret,
230                                      struct rb_node **next_ret)
231 {
232         struct rb_root *root = &tree->state;
233         struct rb_node *n = root->rb_node;
234         struct rb_node *prev = NULL;
235         struct rb_node *orig_prev = NULL;
236         struct tree_entry *entry;
237         struct tree_entry *prev_entry = NULL;
238 
239         while (n) {
240                 entry = rb_entry(n, struct tree_entry, rb_node);
241                 prev = n;
242                 prev_entry = entry;
243 
244                 if (offset < entry->start)
245                         n = n->rb_left;
246                 else if (offset > entry->end)
247                         n = n->rb_right;
248                 else
249                         return n;
250         }
251 
252         if (prev_ret) {
253                 orig_prev = prev;
254                 while (prev && offset > prev_entry->end) {
255                         prev = rb_next(prev);
256                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
257                 }
258                 *prev_ret = prev;
259                 prev = orig_prev;
260         }
261 
262         if (next_ret) {
263                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
264                 while (prev && offset < prev_entry->start) {
265                         prev = rb_prev(prev);
266                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
267                 }
268                 *next_ret = prev;
269         }
270         return NULL;
271 }
272 
273 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
274                                           u64 offset)
275 {
276         struct rb_node *prev = NULL;
277         struct rb_node *ret;
278 
279         ret = __etree_search(tree, offset, &prev, NULL);
280         if (!ret)
281                 return prev;
282         return ret;
283 }
284 
285 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
286                      struct extent_state *other)
287 {
288         if (tree->ops && tree->ops->merge_extent_hook)
289                 tree->ops->merge_extent_hook(tree->mapping->host, new,
290                                              other);
291 }
292 
293 /*
294  * utility function to look for merge candidates inside a given range.
295  * Any extents with matching state are merged together into a single
296  * extent in the tree.  Extents with EXTENT_IO in their state field
297  * are not merged because the end_io handlers need to be able to do
298  * operations on them without sleeping (or doing allocations/splits).
299  *
300  * This should be called with the tree lock held.
301  */
302 static void merge_state(struct extent_io_tree *tree,
303                         struct extent_state *state)
304 {
305         struct extent_state *other;
306         struct rb_node *other_node;
307 
308         if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
309                 return;
310 
311         other_node = rb_prev(&state->rb_node);
312         if (other_node) {
313                 other = rb_entry(other_node, struct extent_state, rb_node);
314                 if (other->end == state->start - 1 &&
315                     other->state == state->state) {
316                         merge_cb(tree, state, other);
317                         state->start = other->start;
318                         other->tree = NULL;
319                         rb_erase(&other->rb_node, &tree->state);
320                         free_extent_state(other);
321                 }
322         }
323         other_node = rb_next(&state->rb_node);
324         if (other_node) {
325                 other = rb_entry(other_node, struct extent_state, rb_node);
326                 if (other->start == state->end + 1 &&
327                     other->state == state->state) {
328                         merge_cb(tree, state, other);
329                         state->end = other->end;
330                         other->tree = NULL;
331                         rb_erase(&other->rb_node, &tree->state);
332                         free_extent_state(other);
333                 }
334         }
335 }
336 
337 static void set_state_cb(struct extent_io_tree *tree,
338                          struct extent_state *state, unsigned long *bits)
339 {
340         if (tree->ops && tree->ops->set_bit_hook)
341                 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
342 }
343 
344 static void clear_state_cb(struct extent_io_tree *tree,
345                            struct extent_state *state, unsigned long *bits)
346 {
347         if (tree->ops && tree->ops->clear_bit_hook)
348                 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
349 }
350 
351 static void set_state_bits(struct extent_io_tree *tree,
352                            struct extent_state *state, unsigned long *bits);
353 
354 /*
355  * insert an extent_state struct into the tree.  'bits' are set on the
356  * struct before it is inserted.
357  *
358  * This may return -EEXIST if the extent is already there, in which case the
359  * state struct is freed.
360  *
361  * The tree lock is not taken internally.  This is a utility function and
362  * probably isn't what you want to call (see set/clear_extent_bit).
363  */
364 static int insert_state(struct extent_io_tree *tree,
365                         struct extent_state *state, u64 start, u64 end,
366                         unsigned long *bits)
367 {
368         struct rb_node *node;
369 
370         if (end < start)
371                 WARN(1, KERN_ERR "btrfs end < start %llu %llu\n",
372                        (unsigned long long)end,
373                        (unsigned long long)start);
374         state->start = start;
375         state->end = end;
376 
377         set_state_bits(tree, state, bits);
378 
379         node = tree_insert(&tree->state, end, &state->rb_node);
380         if (node) {
381                 struct extent_state *found;
382                 found = rb_entry(node, struct extent_state, rb_node);
383                 printk(KERN_ERR "btrfs found node %llu %llu on insert of "
384                        "%llu %llu\n", (unsigned long long)found->start,
385                        (unsigned long long)found->end,
386                        (unsigned long long)start, (unsigned long long)end);
387                 return -EEXIST;
388         }
389         state->tree = tree;
390         merge_state(tree, state);
391         return 0;
392 }
393 
394 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
395                      u64 split)
396 {
397         if (tree->ops && tree->ops->split_extent_hook)
398                 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
399 }
400 
401 /*
402  * split a given extent state struct in two, inserting the preallocated
403  * struct 'prealloc' as the newly created second half.  'split' indicates an
404  * offset inside 'orig' where it should be split.
405  *
406  * Before calling,
407  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
408  * are two extent state structs in the tree:
409  * prealloc: [orig->start, split - 1]
410  * orig: [ split, orig->end ]
411  *
412  * The tree locks are not taken by this function. They need to be held
413  * by the caller.
414  */
415 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
416                        struct extent_state *prealloc, u64 split)
417 {
418         struct rb_node *node;
419 
420         split_cb(tree, orig, split);
421 
422         prealloc->start = orig->start;
423         prealloc->end = split - 1;
424         prealloc->state = orig->state;
425         orig->start = split;
426 
427         node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
428         if (node) {
429                 free_extent_state(prealloc);
430                 return -EEXIST;
431         }
432         prealloc->tree = tree;
433         return 0;
434 }
435 
436 static struct extent_state *next_state(struct extent_state *state)
437 {
438         struct rb_node *next = rb_next(&state->rb_node);
439         if (next)
440                 return rb_entry(next, struct extent_state, rb_node);
441         else
442                 return NULL;
443 }
444 
445 /*
446  * utility function to clear some bits in an extent state struct.
447  * it will optionally wake up any one waiting on this state (wake == 1).
448  *
449  * If no bits are set on the state struct after clearing things, the
450  * struct is freed and removed from the tree
451  */
452 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
453                                             struct extent_state *state,
454                                             unsigned long *bits, int wake)
455 {
456         struct extent_state *next;
457         unsigned long bits_to_clear = *bits & ~EXTENT_CTLBITS;
458 
459         if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
460                 u64 range = state->end - state->start + 1;
461                 WARN_ON(range > tree->dirty_bytes);
462                 tree->dirty_bytes -= range;
463         }
464         clear_state_cb(tree, state, bits);
465         state->state &= ~bits_to_clear;
466         if (wake)
467                 wake_up(&state->wq);
468         if (state->state == 0) {
469                 next = next_state(state);
470                 if (state->tree) {
471                         rb_erase(&state->rb_node, &tree->state);
472                         state->tree = NULL;
473                         free_extent_state(state);
474                 } else {
475                         WARN_ON(1);
476                 }
477         } else {
478                 merge_state(tree, state);
479                 next = next_state(state);
480         }
481         return next;
482 }
483 
484 static struct extent_state *
485 alloc_extent_state_atomic(struct extent_state *prealloc)
486 {
487         if (!prealloc)
488                 prealloc = alloc_extent_state(GFP_ATOMIC);
489 
490         return prealloc;
491 }
492 
493 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
494 {
495         btrfs_panic(tree_fs_info(tree), err, "Locking error: "
496                     "Extent tree was modified by another "
497                     "thread while locked.");
498 }
499 
500 /*
501  * clear some bits on a range in the tree.  This may require splitting
502  * or inserting elements in the tree, so the gfp mask is used to
503  * indicate which allocations or sleeping are allowed.
504  *
505  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
506  * the given range from the tree regardless of state (ie for truncate).
507  *
508  * the range [start, end] is inclusive.
509  *
510  * This takes the tree lock, and returns 0 on success and < 0 on error.
511  */
512 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
513                      unsigned long bits, int wake, int delete,
514                      struct extent_state **cached_state,
515                      gfp_t mask)
516 {
517         struct extent_state *state;
518         struct extent_state *cached;
519         struct extent_state *prealloc = NULL;
520         struct rb_node *node;
521         u64 last_end;
522         int err;
523         int clear = 0;
524 
525         if (delete)
526                 bits |= ~EXTENT_CTLBITS;
527         bits |= EXTENT_FIRST_DELALLOC;
528 
529         if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
530                 clear = 1;
531 again:
532         if (!prealloc && (mask & __GFP_WAIT)) {
533                 prealloc = alloc_extent_state(mask);
534                 if (!prealloc)
535                         return -ENOMEM;
536         }
537 
538         spin_lock(&tree->lock);
539         if (cached_state) {
540                 cached = *cached_state;
541 
542                 if (clear) {
543                         *cached_state = NULL;
544                         cached_state = NULL;
545                 }
546 
547                 if (cached && cached->tree && cached->start <= start &&
548                     cached->end > start) {
549                         if (clear)
550                                 atomic_dec(&cached->refs);
551                         state = cached;
552                         goto hit_next;
553                 }
554                 if (clear)
555                         free_extent_state(cached);
556         }
557         /*
558          * this search will find the extents that end after
559          * our range starts
560          */
561         node = tree_search(tree, start);
562         if (!node)
563                 goto out;
564         state = rb_entry(node, struct extent_state, rb_node);
565 hit_next:
566         if (state->start > end)
567                 goto out;
568         WARN_ON(state->end < start);
569         last_end = state->end;
570 
571         /* the state doesn't have the wanted bits, go ahead */
572         if (!(state->state & bits)) {
573                 state = next_state(state);
574                 goto next;
575         }
576 
577         /*
578          *     | ---- desired range ---- |
579          *  | state | or
580          *  | ------------- state -------------- |
581          *
582          * We need to split the extent we found, and may flip
583          * bits on second half.
584          *
585          * If the extent we found extends past our range, we
586          * just split and search again.  It'll get split again
587          * the next time though.
588          *
589          * If the extent we found is inside our range, we clear
590          * the desired bit on it.
591          */
592 
593         if (state->start < start) {
594                 prealloc = alloc_extent_state_atomic(prealloc);
595                 BUG_ON(!prealloc);
596                 err = split_state(tree, state, prealloc, start);
597                 if (err)
598                         extent_io_tree_panic(tree, err);
599 
600                 prealloc = NULL;
601                 if (err)
602                         goto out;
603                 if (state->end <= end) {
604                         state = clear_state_bit(tree, state, &bits, wake);
605                         goto next;
606                 }
607                 goto search_again;
608         }
609         /*
610          * | ---- desired range ---- |
611          *                        | state |
612          * We need to split the extent, and clear the bit
613          * on the first half
614          */
615         if (state->start <= end && state->end > end) {
616                 prealloc = alloc_extent_state_atomic(prealloc);
617                 BUG_ON(!prealloc);
618                 err = split_state(tree, state, prealloc, end + 1);
619                 if (err)
620                         extent_io_tree_panic(tree, err);
621 
622                 if (wake)
623                         wake_up(&state->wq);
624 
625                 clear_state_bit(tree, prealloc, &bits, wake);
626 
627                 prealloc = NULL;
628                 goto out;
629         }
630 
631         state = clear_state_bit(tree, state, &bits, wake);
632 next:
633         if (last_end == (u64)-1)
634                 goto out;
635         start = last_end + 1;
636         if (start <= end && state && !need_resched())
637                 goto hit_next;
638         goto search_again;
639 
640 out:
641         spin_unlock(&tree->lock);
642         if (prealloc)
643                 free_extent_state(prealloc);
644 
645         return 0;
646 
647 search_again:
648         if (start > end)
649                 goto out;
650         spin_unlock(&tree->lock);
651         if (mask & __GFP_WAIT)
652                 cond_resched();
653         goto again;
654 }
655 
656 static void wait_on_state(struct extent_io_tree *tree,
657                           struct extent_state *state)
658                 __releases(tree->lock)
659                 __acquires(tree->lock)
660 {
661         DEFINE_WAIT(wait);
662         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
663         spin_unlock(&tree->lock);
664         schedule();
665         spin_lock(&tree->lock);
666         finish_wait(&state->wq, &wait);
667 }
668 
669 /*
670  * waits for one or more bits to clear on a range in the state tree.
671  * The range [start, end] is inclusive.
672  * The tree lock is taken by this function
673  */
674 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
675                             unsigned long bits)
676 {
677         struct extent_state *state;
678         struct rb_node *node;
679 
680         spin_lock(&tree->lock);
681 again:
682         while (1) {
683                 /*
684                  * this search will find all the extents that end after
685                  * our range starts
686                  */
687                 node = tree_search(tree, start);
688                 if (!node)
689                         break;
690 
691                 state = rb_entry(node, struct extent_state, rb_node);
692 
693                 if (state->start > end)
694                         goto out;
695 
696                 if (state->state & bits) {
697                         start = state->start;
698                         atomic_inc(&state->refs);
699                         wait_on_state(tree, state);
700                         free_extent_state(state);
701                         goto again;
702                 }
703                 start = state->end + 1;
704 
705                 if (start > end)
706                         break;
707 
708                 cond_resched_lock(&tree->lock);
709         }
710 out:
711         spin_unlock(&tree->lock);
712 }
713 
714 static void set_state_bits(struct extent_io_tree *tree,
715                            struct extent_state *state,
716                            unsigned long *bits)
717 {
718         unsigned long bits_to_set = *bits & ~EXTENT_CTLBITS;
719 
720         set_state_cb(tree, state, bits);
721         if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
722                 u64 range = state->end - state->start + 1;
723                 tree->dirty_bytes += range;
724         }
725         state->state |= bits_to_set;
726 }
727 
728 static void cache_state(struct extent_state *state,
729                         struct extent_state **cached_ptr)
730 {
731         if (cached_ptr && !(*cached_ptr)) {
732                 if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY)) {
733                         *cached_ptr = state;
734                         atomic_inc(&state->refs);
735                 }
736         }
737 }
738 
739 static void uncache_state(struct extent_state **cached_ptr)
740 {
741         if (cached_ptr && (*cached_ptr)) {
742                 struct extent_state *state = *cached_ptr;
743                 *cached_ptr = NULL;
744                 free_extent_state(state);
745         }
746 }
747 
748 /*
749  * set some bits on a range in the tree.  This may require allocations or
750  * sleeping, so the gfp mask is used to indicate what is allowed.
751  *
752  * If any of the exclusive bits are set, this will fail with -EEXIST if some
753  * part of the range already has the desired bits set.  The start of the
754  * existing range is returned in failed_start in this case.
755  *
756  * [start, end] is inclusive This takes the tree lock.
757  */
758 
759 static int __must_check
760 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
761                  unsigned long bits, unsigned long exclusive_bits,
762                  u64 *failed_start, struct extent_state **cached_state,
763                  gfp_t mask)
764 {
765         struct extent_state *state;
766         struct extent_state *prealloc = NULL;
767         struct rb_node *node;
768         int err = 0;
769         u64 last_start;
770         u64 last_end;
771 
772         bits |= EXTENT_FIRST_DELALLOC;
773 again:
774         if (!prealloc && (mask & __GFP_WAIT)) {
775                 prealloc = alloc_extent_state(mask);
776                 BUG_ON(!prealloc);
777         }
778 
779         spin_lock(&tree->lock);
780         if (cached_state && *cached_state) {
781                 state = *cached_state;
782                 if (state->start <= start && state->end > start &&
783                     state->tree) {
784                         node = &state->rb_node;
785                         goto hit_next;
786                 }
787         }
788         /*
789          * this search will find all the extents that end after
790          * our range starts.
791          */
792         node = tree_search(tree, start);
793         if (!node) {
794                 prealloc = alloc_extent_state_atomic(prealloc);
795                 BUG_ON(!prealloc);
796                 err = insert_state(tree, prealloc, start, end, &bits);
797                 if (err)
798                         extent_io_tree_panic(tree, err);
799 
800                 prealloc = NULL;
801                 goto out;
802         }
803         state = rb_entry(node, struct extent_state, rb_node);
804 hit_next:
805         last_start = state->start;
806         last_end = state->end;
807 
808         /*
809          * | ---- desired range ---- |
810          * | state |
811          *
812          * Just lock what we found and keep going
813          */
814         if (state->start == start && state->end <= end) {
815                 if (state->state & exclusive_bits) {
816                         *failed_start = state->start;
817                         err = -EEXIST;
818                         goto out;
819                 }
820 
821                 set_state_bits(tree, state, &bits);
822                 cache_state(state, cached_state);
823                 merge_state(tree, state);
824                 if (last_end == (u64)-1)
825                         goto out;
826                 start = last_end + 1;
827                 state = next_state(state);
828                 if (start < end && state && state->start == start &&
829                     !need_resched())
830                         goto hit_next;
831                 goto search_again;
832         }
833 
834         /*
835          *     | ---- desired range ---- |
836          * | state |
837          *   or
838          * | ------------- state -------------- |
839          *
840          * We need to split the extent we found, and may flip bits on
841          * second half.
842          *
843          * If the extent we found extends past our
844          * range, we just split and search again.  It'll get split
845          * again the next time though.
846          *
847          * If the extent we found is inside our range, we set the
848          * desired bit on it.
849          */
850         if (state->start < start) {
851                 if (state->state & exclusive_bits) {
852                         *failed_start = start;
853                         err = -EEXIST;
854                         goto out;
855                 }
856 
857                 prealloc = alloc_extent_state_atomic(prealloc);
858                 BUG_ON(!prealloc);
859                 err = split_state(tree, state, prealloc, start);
860                 if (err)
861                         extent_io_tree_panic(tree, err);
862 
863                 prealloc = NULL;
864                 if (err)
865                         goto out;
866                 if (state->end <= end) {
867                         set_state_bits(tree, state, &bits);
868                         cache_state(state, cached_state);
869                         merge_state(tree, state);
870                         if (last_end == (u64)-1)
871                                 goto out;
872                         start = last_end + 1;
873                         state = next_state(state);
874                         if (start < end && state && state->start == start &&
875                             !need_resched())
876                                 goto hit_next;
877                 }
878                 goto search_again;
879         }
880         /*
881          * | ---- desired range ---- |
882          *     | state | or               | state |
883          *
884          * There's a hole, we need to insert something in it and
885          * ignore the extent we found.
886          */
887         if (state->start > start) {
888                 u64 this_end;
889                 if (end < last_start)
890                         this_end = end;
891                 else
892                         this_end = last_start - 1;
893 
894                 prealloc = alloc_extent_state_atomic(prealloc);
895                 BUG_ON(!prealloc);
896 
897                 /*
898                  * Avoid to free 'prealloc' if it can be merged with
899                  * the later extent.
900                  */
901                 err = insert_state(tree, prealloc, start, this_end,
902                                    &bits);
903                 if (err)
904                         extent_io_tree_panic(tree, err);
905 
906                 cache_state(prealloc, cached_state);
907                 prealloc = NULL;
908                 start = this_end + 1;
909                 goto search_again;
910         }
911         /*
912          * | ---- desired range ---- |
913          *                        | state |
914          * We need to split the extent, and set the bit
915          * on the first half
916          */
917         if (state->start <= end && state->end > end) {
918                 if (state->state & exclusive_bits) {
919                         *failed_start = start;
920                         err = -EEXIST;
921                         goto out;
922                 }
923 
924                 prealloc = alloc_extent_state_atomic(prealloc);
925                 BUG_ON(!prealloc);
926                 err = split_state(tree, state, prealloc, end + 1);
927                 if (err)
928                         extent_io_tree_panic(tree, err);
929 
930                 set_state_bits(tree, prealloc, &bits);
931                 cache_state(prealloc, cached_state);
932                 merge_state(tree, prealloc);
933                 prealloc = NULL;
934                 goto out;
935         }
936 
937         goto search_again;
938 
939 out:
940         spin_unlock(&tree->lock);
941         if (prealloc)
942                 free_extent_state(prealloc);
943 
944         return err;
945 
946 search_again:
947         if (start > end)
948                 goto out;
949         spin_unlock(&tree->lock);
950         if (mask & __GFP_WAIT)
951                 cond_resched();
952         goto again;
953 }
954 
955 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
956                    unsigned long bits, u64 * failed_start,
957                    struct extent_state **cached_state, gfp_t mask)
958 {
959         return __set_extent_bit(tree, start, end, bits, 0, failed_start,
960                                 cached_state, mask);
961 }
962 
963 
964 /**
965  * convert_extent_bit - convert all bits in a given range from one bit to
966  *                      another
967  * @tree:       the io tree to search
968  * @start:      the start offset in bytes
969  * @end:        the end offset in bytes (inclusive)
970  * @bits:       the bits to set in this range
971  * @clear_bits: the bits to clear in this range
972  * @cached_state:       state that we're going to cache
973  * @mask:       the allocation mask
974  *
975  * This will go through and set bits for the given range.  If any states exist
976  * already in this range they are set with the given bit and cleared of the
977  * clear_bits.  This is only meant to be used by things that are mergeable, ie
978  * converting from say DELALLOC to DIRTY.  This is not meant to be used with
979  * boundary bits like LOCK.
980  */
981 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
982                        unsigned long bits, unsigned long clear_bits,
983                        struct extent_state **cached_state, gfp_t mask)
984 {
985         struct extent_state *state;
986         struct extent_state *prealloc = NULL;
987         struct rb_node *node;
988         int err = 0;
989         u64 last_start;
990         u64 last_end;
991 
992 again:
993         if (!prealloc && (mask & __GFP_WAIT)) {
994                 prealloc = alloc_extent_state(mask);
995                 if (!prealloc)
996                         return -ENOMEM;
997         }
998 
999         spin_lock(&tree->lock);
1000         if (cached_state && *cached_state) {
1001                 state = *cached_state;
1002                 if (state->start <= start && state->end > start &&
1003                     state->tree) {
1004                         node = &state->rb_node;
1005                         goto hit_next;
1006                 }
1007         }
1008 
1009         /*
1010          * this search will find all the extents that end after
1011          * our range starts.
1012          */
1013         node = tree_search(tree, start);
1014         if (!node) {
1015                 prealloc = alloc_extent_state_atomic(prealloc);
1016                 if (!prealloc) {
1017                         err = -ENOMEM;
1018                         goto out;
1019                 }
1020                 err = insert_state(tree, prealloc, start, end, &bits);
1021                 prealloc = NULL;
1022                 if (err)
1023                         extent_io_tree_panic(tree, err);
1024                 goto out;
1025         }
1026         state = rb_entry(node, struct extent_state, rb_node);
1027 hit_next:
1028         last_start = state->start;
1029         last_end = state->end;
1030 
1031         /*
1032          * | ---- desired range ---- |
1033          * | state |
1034          *
1035          * Just lock what we found and keep going
1036          */
1037         if (state->start == start && state->end <= end) {
1038                 set_state_bits(tree, state, &bits);
1039                 cache_state(state, cached_state);
1040                 state = clear_state_bit(tree, state, &clear_bits, 0);
1041                 if (last_end == (u64)-1)
1042                         goto out;
1043                 start = last_end + 1;
1044                 if (start < end && state && state->start == start &&
1045                     !need_resched())
1046                         goto hit_next;
1047                 goto search_again;
1048         }
1049 
1050         /*
1051          *     | ---- desired range ---- |
1052          * | state |
1053          *   or
1054          * | ------------- state -------------- |
1055          *
1056          * We need to split the extent we found, and may flip bits on
1057          * second half.
1058          *
1059          * If the extent we found extends past our
1060          * range, we just split and search again.  It'll get split
1061          * again the next time though.
1062          *
1063          * If the extent we found is inside our range, we set the
1064          * desired bit on it.
1065          */
1066         if (state->start < start) {
1067                 prealloc = alloc_extent_state_atomic(prealloc);
1068                 if (!prealloc) {
1069                         err = -ENOMEM;
1070                         goto out;
1071                 }
1072                 err = split_state(tree, state, prealloc, start);
1073                 if (err)
1074                         extent_io_tree_panic(tree, err);
1075                 prealloc = NULL;
1076                 if (err)
1077                         goto out;
1078                 if (state->end <= end) {
1079                         set_state_bits(tree, state, &bits);
1080                         cache_state(state, cached_state);
1081                         state = clear_state_bit(tree, state, &clear_bits, 0);
1082                         if (last_end == (u64)-1)
1083                                 goto out;
1084                         start = last_end + 1;
1085                         if (start < end && state && state->start == start &&
1086                             !need_resched())
1087                                 goto hit_next;
1088                 }
1089                 goto search_again;
1090         }
1091         /*
1092          * | ---- desired range ---- |
1093          *     | state | or               | state |
1094          *
1095          * There's a hole, we need to insert something in it and
1096          * ignore the extent we found.
1097          */
1098         if (state->start > start) {
1099                 u64 this_end;
1100                 if (end < last_start)
1101                         this_end = end;
1102                 else
1103                         this_end = last_start - 1;
1104 
1105                 prealloc = alloc_extent_state_atomic(prealloc);
1106                 if (!prealloc) {
1107                         err = -ENOMEM;
1108                         goto out;
1109                 }
1110 
1111                 /*
1112                  * Avoid to free 'prealloc' if it can be merged with
1113                  * the later extent.
1114                  */
1115                 err = insert_state(tree, prealloc, start, this_end,
1116                                    &bits);
1117                 if (err)
1118                         extent_io_tree_panic(tree, err);
1119                 cache_state(prealloc, cached_state);
1120                 prealloc = NULL;
1121                 start = this_end + 1;
1122                 goto search_again;
1123         }
1124         /*
1125          * | ---- desired range ---- |
1126          *                        | state |
1127          * We need to split the extent, and set the bit
1128          * on the first half
1129          */
1130         if (state->start <= end && state->end > end) {
1131                 prealloc = alloc_extent_state_atomic(prealloc);
1132                 if (!prealloc) {
1133                         err = -ENOMEM;
1134                         goto out;
1135                 }
1136 
1137                 err = split_state(tree, state, prealloc, end + 1);
1138                 if (err)
1139                         extent_io_tree_panic(tree, err);
1140 
1141                 set_state_bits(tree, prealloc, &bits);
1142                 cache_state(prealloc, cached_state);
1143                 clear_state_bit(tree, prealloc, &clear_bits, 0);
1144                 prealloc = NULL;
1145                 goto out;
1146         }
1147 
1148         goto search_again;
1149 
1150 out:
1151         spin_unlock(&tree->lock);
1152         if (prealloc)
1153                 free_extent_state(prealloc);
1154 
1155         return err;
1156 
1157 search_again:
1158         if (start > end)
1159                 goto out;
1160         spin_unlock(&tree->lock);
1161         if (mask & __GFP_WAIT)
1162                 cond_resched();
1163         goto again;
1164 }
1165 
1166 /* wrappers around set/clear extent bit */
1167 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1168                      gfp_t mask)
1169 {
1170         return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1171                               NULL, mask);
1172 }
1173 
1174 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1175                     unsigned long bits, gfp_t mask)
1176 {
1177         return set_extent_bit(tree, start, end, bits, NULL,
1178                               NULL, mask);
1179 }
1180 
1181 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1182                       unsigned long bits, gfp_t mask)
1183 {
1184         return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1185 }
1186 
1187 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1188                         struct extent_state **cached_state, gfp_t mask)
1189 {
1190         return set_extent_bit(tree, start, end,
1191                               EXTENT_DELALLOC | EXTENT_UPTODATE,
1192                               NULL, cached_state, mask);
1193 }
1194 
1195 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1196                       struct extent_state **cached_state, gfp_t mask)
1197 {
1198         return set_extent_bit(tree, start, end,
1199                               EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1200                               NULL, cached_state, mask);
1201 }
1202 
1203 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1204                        gfp_t mask)
1205 {
1206         return clear_extent_bit(tree, start, end,
1207                                 EXTENT_DIRTY | EXTENT_DELALLOC |
1208                                 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1209 }
1210 
1211 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1212                      gfp_t mask)
1213 {
1214         return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1215                               NULL, mask);
1216 }
1217 
1218 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1219                         struct extent_state **cached_state, gfp_t mask)
1220 {
1221         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1222                               cached_state, mask);
1223 }
1224 
1225 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1226                           struct extent_state **cached_state, gfp_t mask)
1227 {
1228         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1229                                 cached_state, mask);
1230 }
1231 
1232 /*
1233  * either insert or lock state struct between start and end use mask to tell
1234  * us if waiting is desired.
1235  */
1236 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1237                      unsigned long bits, struct extent_state **cached_state)
1238 {
1239         int err;
1240         u64 failed_start;
1241         while (1) {
1242                 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1243                                        EXTENT_LOCKED, &failed_start,
1244                                        cached_state, GFP_NOFS);
1245                 if (err == -EEXIST) {
1246                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1247                         start = failed_start;
1248                 } else
1249                         break;
1250                 WARN_ON(start > end);
1251         }
1252         return err;
1253 }
1254 
1255 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1256 {
1257         return lock_extent_bits(tree, start, end, 0, NULL);
1258 }
1259 
1260 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1261 {
1262         int err;
1263         u64 failed_start;
1264 
1265         err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1266                                &failed_start, NULL, GFP_NOFS);
1267         if (err == -EEXIST) {
1268                 if (failed_start > start)
1269                         clear_extent_bit(tree, start, failed_start - 1,
1270                                          EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1271                 return 0;
1272         }
1273         return 1;
1274 }
1275 
1276 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1277                          struct extent_state **cached, gfp_t mask)
1278 {
1279         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1280                                 mask);
1281 }
1282 
1283 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1284 {
1285         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1286                                 GFP_NOFS);
1287 }
1288 
1289 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1290 {
1291         unsigned long index = start >> PAGE_CACHE_SHIFT;
1292         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1293         struct page *page;
1294 
1295         while (index <= end_index) {
1296                 page = find_get_page(inode->i_mapping, index);
1297                 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1298                 clear_page_dirty_for_io(page);
1299                 page_cache_release(page);
1300                 index++;
1301         }
1302         return 0;
1303 }
1304 
1305 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1306 {
1307         unsigned long index = start >> PAGE_CACHE_SHIFT;
1308         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1309         struct page *page;
1310 
1311         while (index <= end_index) {
1312                 page = find_get_page(inode->i_mapping, index);
1313                 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1314                 account_page_redirty(page);
1315                 __set_page_dirty_nobuffers(page);
1316                 page_cache_release(page);
1317                 index++;
1318         }
1319         return 0;
1320 }
1321 
1322 /*
1323  * helper function to set both pages and extents in the tree writeback
1324  */
1325 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1326 {
1327         unsigned long index = start >> PAGE_CACHE_SHIFT;
1328         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1329         struct page *page;
1330 
1331         while (index <= end_index) {
1332                 page = find_get_page(tree->mapping, index);
1333                 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1334                 set_page_writeback(page);
1335                 page_cache_release(page);
1336                 index++;
1337         }
1338         return 0;
1339 }
1340 
1341 /* find the first state struct with 'bits' set after 'start', and
1342  * return it.  tree->lock must be held.  NULL will returned if
1343  * nothing was found after 'start'
1344  */
1345 static struct extent_state *
1346 find_first_extent_bit_state(struct extent_io_tree *tree,
1347                             u64 start, unsigned long bits)
1348 {
1349         struct rb_node *node;
1350         struct extent_state *state;
1351 
1352         /*
1353          * this search will find all the extents that end after
1354          * our range starts.
1355          */
1356         node = tree_search(tree, start);
1357         if (!node)
1358                 goto out;
1359 
1360         while (1) {
1361                 state = rb_entry(node, struct extent_state, rb_node);
1362                 if (state->end >= start && (state->state & bits))
1363                         return state;
1364 
1365                 node = rb_next(node);
1366                 if (!node)
1367                         break;
1368         }
1369 out:
1370         return NULL;
1371 }
1372 
1373 /*
1374  * find the first offset in the io tree with 'bits' set. zero is
1375  * returned if we find something, and *start_ret and *end_ret are
1376  * set to reflect the state struct that was found.
1377  *
1378  * If nothing was found, 1 is returned. If found something, return 0.
1379  */
1380 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1381                           u64 *start_ret, u64 *end_ret, unsigned long bits,
1382                           struct extent_state **cached_state)
1383 {
1384         struct extent_state *state;
1385         struct rb_node *n;
1386         int ret = 1;
1387 
1388         spin_lock(&tree->lock);
1389         if (cached_state && *cached_state) {
1390                 state = *cached_state;
1391                 if (state->end == start - 1 && state->tree) {
1392                         n = rb_next(&state->rb_node);
1393                         while (n) {
1394                                 state = rb_entry(n, struct extent_state,
1395                                                  rb_node);
1396                                 if (state->state & bits)
1397                                         goto got_it;
1398                                 n = rb_next(n);
1399                         }
1400                         free_extent_state(*cached_state);
1401                         *cached_state = NULL;
1402                         goto out;
1403                 }
1404                 free_extent_state(*cached_state);
1405                 *cached_state = NULL;
1406         }
1407 
1408         state = find_first_extent_bit_state(tree, start, bits);
1409 got_it:
1410         if (state) {
1411                 cache_state(state, cached_state);
1412                 *start_ret = state->start;
1413                 *end_ret = state->end;
1414                 ret = 0;
1415         }
1416 out:
1417         spin_unlock(&tree->lock);
1418         return ret;
1419 }
1420 
1421 /*
1422  * find a contiguous range of bytes in the file marked as delalloc, not
1423  * more than 'max_bytes'.  start and end are used to return the range,
1424  *
1425  * 1 is returned if we find something, 0 if nothing was in the tree
1426  */
1427 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1428                                         u64 *start, u64 *end, u64 max_bytes,
1429                                         struct extent_state **cached_state)
1430 {
1431         struct rb_node *node;
1432         struct extent_state *state;
1433         u64 cur_start = *start;
1434         u64 found = 0;
1435         u64 total_bytes = 0;
1436 
1437         spin_lock(&tree->lock);
1438 
1439         /*
1440          * this search will find all the extents that end after
1441          * our range starts.
1442          */
1443         node = tree_search(tree, cur_start);
1444         if (!node) {
1445                 if (!found)
1446                         *end = (u64)-1;
1447                 goto out;
1448         }
1449 
1450         while (1) {
1451                 state = rb_entry(node, struct extent_state, rb_node);
1452                 if (found && (state->start != cur_start ||
1453                               (state->state & EXTENT_BOUNDARY))) {
1454                         goto out;
1455                 }
1456                 if (!(state->state & EXTENT_DELALLOC)) {
1457                         if (!found)
1458                                 *end = state->end;
1459                         goto out;
1460                 }
1461                 if (!found) {
1462                         *start = state->start;
1463                         *cached_state = state;
1464                         atomic_inc(&state->refs);
1465                 }
1466                 found++;
1467                 *end = state->end;
1468                 cur_start = state->end + 1;
1469                 node = rb_next(node);
1470                 if (!node)
1471                         break;
1472                 total_bytes += state->end - state->start + 1;
1473                 if (total_bytes >= max_bytes)
1474                         break;
1475         }
1476 out:
1477         spin_unlock(&tree->lock);
1478         return found;
1479 }
1480 
1481 static noinline void __unlock_for_delalloc(struct inode *inode,
1482                                            struct page *locked_page,
1483                                            u64 start, u64 end)
1484 {
1485         int ret;
1486         struct page *pages[16];
1487         unsigned long index = start >> PAGE_CACHE_SHIFT;
1488         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1489         unsigned long nr_pages = end_index - index + 1;
1490         int i;
1491 
1492         if (index == locked_page->index && end_index == index)
1493                 return;
1494 
1495         while (nr_pages > 0) {
1496                 ret = find_get_pages_contig(inode->i_mapping, index,
1497                                      min_t(unsigned long, nr_pages,
1498                                      ARRAY_SIZE(pages)), pages);
1499                 for (i = 0; i < ret; i++) {
1500                         if (pages[i] != locked_page)
1501                                 unlock_page(pages[i]);
1502                         page_cache_release(pages[i]);
1503                 }
1504                 nr_pages -= ret;
1505                 index += ret;
1506                 cond_resched();
1507         }
1508 }
1509 
1510 static noinline int lock_delalloc_pages(struct inode *inode,
1511                                         struct page *locked_page,
1512                                         u64 delalloc_start,
1513                                         u64 delalloc_end)
1514 {
1515         unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1516         unsigned long start_index = index;
1517         unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1518         unsigned long pages_locked = 0;
1519         struct page *pages[16];
1520         unsigned long nrpages;
1521         int ret;
1522         int i;
1523 
1524         /* the caller is responsible for locking the start index */
1525         if (index == locked_page->index && index == end_index)
1526                 return 0;
1527 
1528         /* skip the page at the start index */
1529         nrpages = end_index - index + 1;
1530         while (nrpages > 0) {
1531                 ret = find_get_pages_contig(inode->i_mapping, index,
1532                                      min_t(unsigned long,
1533                                      nrpages, ARRAY_SIZE(pages)), pages);
1534                 if (ret == 0) {
1535                         ret = -EAGAIN;
1536                         goto done;
1537                 }
1538                 /* now we have an array of pages, lock them all */
1539                 for (i = 0; i < ret; i++) {
1540                         /*
1541                          * the caller is taking responsibility for
1542                          * locked_page
1543                          */
1544                         if (pages[i] != locked_page) {
1545                                 lock_page(pages[i]);
1546                                 if (!PageDirty(pages[i]) ||
1547                                     pages[i]->mapping != inode->i_mapping) {
1548                                         ret = -EAGAIN;
1549                                         unlock_page(pages[i]);
1550                                         page_cache_release(pages[i]);
1551                                         goto done;
1552                                 }
1553                         }
1554                         page_cache_release(pages[i]);
1555                         pages_locked++;
1556                 }
1557                 nrpages -= ret;
1558                 index += ret;
1559                 cond_resched();
1560         }
1561         ret = 0;
1562 done:
1563         if (ret && pages_locked) {
1564                 __unlock_for_delalloc(inode, locked_page,
1565                               delalloc_start,
1566                               ((u64)(start_index + pages_locked - 1)) <<
1567                               PAGE_CACHE_SHIFT);
1568         }
1569         return ret;
1570 }
1571 
1572 /*
1573  * find a contiguous range of bytes in the file marked as delalloc, not
1574  * more than 'max_bytes'.  start and end are used to return the range,
1575  *
1576  * 1 is returned if we find something, 0 if nothing was in the tree
1577  */
1578 static noinline u64 find_lock_delalloc_range(struct inode *inode,
1579                                              struct extent_io_tree *tree,
1580                                              struct page *locked_page,
1581                                              u64 *start, u64 *end,
1582                                              u64 max_bytes)
1583 {
1584         u64 delalloc_start;
1585         u64 delalloc_end;
1586         u64 found;
1587         struct extent_state *cached_state = NULL;
1588         int ret;
1589         int loops = 0;
1590 
1591 again:
1592         /* step one, find a bunch of delalloc bytes starting at start */
1593         delalloc_start = *start;
1594         delalloc_end = 0;
1595         found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1596                                     max_bytes, &cached_state);
1597         if (!found || delalloc_end <= *start) {
1598                 *start = delalloc_start;
1599                 *end = delalloc_end;
1600                 free_extent_state(cached_state);
1601                 return found;
1602         }
1603 
1604         /*
1605          * start comes from the offset of locked_page.  We have to lock
1606          * pages in order, so we can't process delalloc bytes before
1607          * locked_page
1608          */
1609         if (delalloc_start < *start)
1610                 delalloc_start = *start;
1611 
1612         /*
1613          * make sure to limit the number of pages we try to lock down
1614          * if we're looping.
1615          */
1616         if (delalloc_end + 1 - delalloc_start > max_bytes && loops)
1617                 delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
1618 
1619         /* step two, lock all the pages after the page that has start */
1620         ret = lock_delalloc_pages(inode, locked_page,
1621                                   delalloc_start, delalloc_end);
1622         if (ret == -EAGAIN) {
1623                 /* some of the pages are gone, lets avoid looping by
1624                  * shortening the size of the delalloc range we're searching
1625                  */
1626                 free_extent_state(cached_state);
1627                 cached_state = NULL;
1628                 if (!loops) {
1629                         unsigned long offset = (*start) & (PAGE_CACHE_SIZE - 1);
1630                         max_bytes = PAGE_CACHE_SIZE - offset;
1631                         loops = 1;
1632                         goto again;
1633                 } else {
1634                         found = 0;
1635                         goto out_failed;
1636                 }
1637         }
1638         BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1639 
1640         /* step three, lock the state bits for the whole range */
1641         lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1642 
1643         /* then test to make sure it is all still delalloc */
1644         ret = test_range_bit(tree, delalloc_start, delalloc_end,
1645                              EXTENT_DELALLOC, 1, cached_state);
1646         if (!ret) {
1647                 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1648                                      &cached_state, GFP_NOFS);
1649                 __unlock_for_delalloc(inode, locked_page,
1650                               delalloc_start, delalloc_end);
1651                 cond_resched();
1652                 goto again;
1653         }
1654         free_extent_state(cached_state);
1655         *start = delalloc_start;
1656         *end = delalloc_end;
1657 out_failed:
1658         return found;
1659 }
1660 
1661 int extent_clear_unlock_delalloc(struct inode *inode,
1662                                 struct extent_io_tree *tree,
1663                                 u64 start, u64 end, struct page *locked_page,
1664                                 unsigned long op)
1665 {
1666         int ret;
1667         struct page *pages[16];
1668         unsigned long index = start >> PAGE_CACHE_SHIFT;
1669         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1670         unsigned long nr_pages = end_index - index + 1;
1671         int i;
1672         unsigned long clear_bits = 0;
1673 
1674         if (op & EXTENT_CLEAR_UNLOCK)
1675                 clear_bits |= EXTENT_LOCKED;
1676         if (op & EXTENT_CLEAR_DIRTY)
1677                 clear_bits |= EXTENT_DIRTY;
1678 
1679         if (op & EXTENT_CLEAR_DELALLOC)
1680                 clear_bits |= EXTENT_DELALLOC;
1681 
1682         clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1683         if (!(op & (EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
1684                     EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK |
1685                     EXTENT_SET_PRIVATE2)))
1686                 return 0;
1687 
1688         while (nr_pages > 0) {
1689                 ret = find_get_pages_contig(inode->i_mapping, index,
1690                                      min_t(unsigned long,
1691                                      nr_pages, ARRAY_SIZE(pages)), pages);
1692                 for (i = 0; i < ret; i++) {
1693 
1694                         if (op & EXTENT_SET_PRIVATE2)
1695                                 SetPagePrivate2(pages[i]);
1696 
1697                         if (pages[i] == locked_page) {
1698                                 page_cache_release(pages[i]);
1699                                 continue;
1700                         }
1701                         if (op & EXTENT_CLEAR_DIRTY)
1702                                 clear_page_dirty_for_io(pages[i]);
1703                         if (op & EXTENT_SET_WRITEBACK)
1704                                 set_page_writeback(pages[i]);
1705                         if (op & EXTENT_END_WRITEBACK)
1706                                 end_page_writeback(pages[i]);
1707                         if (op & EXTENT_CLEAR_UNLOCK_PAGE)
1708                                 unlock_page(pages[i]);
1709                         page_cache_release(pages[i]);
1710                 }
1711                 nr_pages -= ret;
1712                 index += ret;
1713                 cond_resched();
1714         }
1715         return 0;
1716 }
1717 
1718 /*
1719  * count the number of bytes in the tree that have a given bit(s)
1720  * set.  This can be fairly slow, except for EXTENT_DIRTY which is
1721  * cached.  The total number found is returned.
1722  */
1723 u64 count_range_bits(struct extent_io_tree *tree,
1724                      u64 *start, u64 search_end, u64 max_bytes,
1725                      unsigned long bits, int contig)
1726 {
1727         struct rb_node *node;
1728         struct extent_state *state;
1729         u64 cur_start = *start;
1730         u64 total_bytes = 0;
1731         u64 last = 0;
1732         int found = 0;
1733 
1734         if (search_end <= cur_start) {
1735                 WARN_ON(1);
1736                 return 0;
1737         }
1738 
1739         spin_lock(&tree->lock);
1740         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1741                 total_bytes = tree->dirty_bytes;
1742                 goto out;
1743         }
1744         /*
1745          * this search will find all the extents that end after
1746          * our range starts.
1747          */
1748         node = tree_search(tree, cur_start);
1749         if (!node)
1750                 goto out;
1751 
1752         while (1) {
1753                 state = rb_entry(node, struct extent_state, rb_node);
1754                 if (state->start > search_end)
1755                         break;
1756                 if (contig && found && state->start > last + 1)
1757                         break;
1758                 if (state->end >= cur_start && (state->state & bits) == bits) {
1759                         total_bytes += min(search_end, state->end) + 1 -
1760                                        max(cur_start, state->start);
1761                         if (total_bytes >= max_bytes)
1762                                 break;
1763                         if (!found) {
1764                                 *start = max(cur_start, state->start);
1765                                 found = 1;
1766                         }
1767                         last = state->end;
1768                 } else if (contig && found) {
1769                         break;
1770                 }
1771                 node = rb_next(node);
1772                 if (!node)
1773                         break;
1774         }
1775 out:
1776         spin_unlock(&tree->lock);
1777         return total_bytes;
1778 }
1779 
1780 /*
1781  * set the private field for a given byte offset in the tree.  If there isn't
1782  * an extent_state there already, this does nothing.
1783  */
1784 int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1785 {
1786         struct rb_node *node;
1787         struct extent_state *state;
1788         int ret = 0;
1789 
1790         spin_lock(&tree->lock);
1791         /*
1792          * this search will find all the extents that end after
1793          * our range starts.
1794          */
1795         node = tree_search(tree, start);
1796         if (!node) {
1797                 ret = -ENOENT;
1798                 goto out;
1799         }
1800         state = rb_entry(node, struct extent_state, rb_node);
1801         if (state->start != start) {
1802                 ret = -ENOENT;
1803                 goto out;
1804         }
1805         state->private = private;
1806 out:
1807         spin_unlock(&tree->lock);
1808         return ret;
1809 }
1810 
1811 void extent_cache_csums_dio(struct extent_io_tree *tree, u64 start, u32 csums[],
1812                             int count)
1813 {
1814         struct rb_node *node;
1815         struct extent_state *state;
1816 
1817         spin_lock(&tree->lock);
1818         /*
1819          * this search will find all the extents that end after
1820          * our range starts.
1821          */
1822         node = tree_search(tree, start);
1823         BUG_ON(!node);
1824 
1825         state = rb_entry(node, struct extent_state, rb_node);
1826         BUG_ON(state->start != start);
1827 
1828         while (count) {
1829                 state->private = *csums++;
1830                 count--;
1831                 state = next_state(state);
1832         }
1833         spin_unlock(&tree->lock);
1834 }
1835 
1836 static inline u64 __btrfs_get_bio_offset(struct bio *bio, int bio_index)
1837 {
1838         struct bio_vec *bvec = bio->bi_io_vec + bio_index;
1839 
1840         return page_offset(bvec->bv_page) + bvec->bv_offset;
1841 }
1842 
1843 void extent_cache_csums(struct extent_io_tree *tree, struct bio *bio, int bio_index,
1844                         u32 csums[], int count)
1845 {
1846         struct rb_node *node;
1847         struct extent_state *state = NULL;
1848         u64 start;
1849 
1850         spin_lock(&tree->lock);
1851         do {
1852                 start = __btrfs_get_bio_offset(bio, bio_index);
1853                 if (state == NULL || state->start != start) {
1854                         node = tree_search(tree, start);
1855                         BUG_ON(!node);
1856 
1857                         state = rb_entry(node, struct extent_state, rb_node);
1858                         BUG_ON(state->start != start);
1859                 }
1860                 state->private = *csums++;
1861                 count--;
1862                 bio_index++;
1863 
1864                 state = next_state(state);
1865         } while (count);
1866         spin_unlock(&tree->lock);
1867 }
1868 
1869 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1870 {
1871         struct rb_node *node;
1872         struct extent_state *state;
1873         int ret = 0;
1874 
1875         spin_lock(&tree->lock);
1876         /*
1877          * this search will find all the extents that end after
1878          * our range starts.
1879          */
1880         node = tree_search(tree, start);
1881         if (!node) {
1882                 ret = -ENOENT;
1883                 goto out;
1884         }
1885         state = rb_entry(node, struct extent_state, rb_node);
1886         if (state->start != start) {
1887                 ret = -ENOENT;
1888                 goto out;
1889         }
1890         *private = state->private;
1891 out:
1892         spin_unlock(&tree->lock);
1893         return ret;
1894 }
1895 
1896 /*
1897  * searches a range in the state tree for a given mask.
1898  * If 'filled' == 1, this returns 1 only if every extent in the tree
1899  * has the bits set.  Otherwise, 1 is returned if any bit in the
1900  * range is found set.
1901  */
1902 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1903                    unsigned long bits, int filled, struct extent_state *cached)
1904 {
1905         struct extent_state *state = NULL;
1906         struct rb_node *node;
1907         int bitset = 0;
1908 
1909         spin_lock(&tree->lock);
1910         if (cached && cached->tree && cached->start <= start &&
1911             cached->end > start)
1912                 node = &cached->rb_node;
1913         else
1914                 node = tree_search(tree, start);
1915         while (node && start <= end) {
1916                 state = rb_entry(node, struct extent_state, rb_node);
1917 
1918                 if (filled && state->start > start) {
1919                         bitset = 0;
1920                         break;
1921                 }
1922 
1923                 if (state->start > end)
1924                         break;
1925 
1926                 if (state->state & bits) {
1927                         bitset = 1;
1928                         if (!filled)
1929                                 break;
1930                 } else if (filled) {
1931                         bitset = 0;
1932                         break;
1933                 }
1934 
1935                 if (state->end == (u64)-1)
1936                         break;
1937 
1938                 start = state->end + 1;
1939                 if (start > end)
1940                         break;
1941                 node = rb_next(node);
1942                 if (!node) {
1943                         if (filled)
1944                                 bitset = 0;
1945                         break;
1946                 }
1947         }
1948         spin_unlock(&tree->lock);
1949         return bitset;
1950 }
1951 
1952 /*
1953  * helper function to set a given page up to date if all the
1954  * extents in the tree for that page are up to date
1955  */
1956 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1957 {
1958         u64 start = page_offset(page);
1959         u64 end = start + PAGE_CACHE_SIZE - 1;
1960         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1961                 SetPageUptodate(page);
1962 }
1963 
1964 /*
1965  * When IO fails, either with EIO or csum verification fails, we
1966  * try other mirrors that might have a good copy of the data.  This
1967  * io_failure_record is used to record state as we go through all the
1968  * mirrors.  If another mirror has good data, the page is set up to date
1969  * and things continue.  If a good mirror can't be found, the original
1970  * bio end_io callback is called to indicate things have failed.
1971  */
1972 struct io_failure_record {
1973         struct page *page;
1974         u64 start;
1975         u64 len;
1976         u64 logical;
1977         unsigned long bio_flags;
1978         int this_mirror;
1979         int failed_mirror;
1980         int in_validation;
1981 };
1982 
1983 static int free_io_failure(struct inode *inode, struct io_failure_record *rec,
1984                                 int did_repair)
1985 {
1986         int ret;
1987         int err = 0;
1988         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1989 
1990         set_state_private(failure_tree, rec->start, 0);
1991         ret = clear_extent_bits(failure_tree, rec->start,
1992                                 rec->start + rec->len - 1,
1993                                 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1994         if (ret)
1995                 err = ret;
1996 
1997         ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
1998                                 rec->start + rec->len - 1,
1999                                 EXTENT_DAMAGED, GFP_NOFS);
2000         if (ret && !err)
2001                 err = ret;
2002 
2003         kfree(rec);
2004         return err;
2005 }
2006 
2007 static void repair_io_failure_callback(struct bio *bio, int err)
2008 {
2009         complete(bio->bi_private);
2010 }
2011 
2012 /*
2013  * this bypasses the standard btrfs submit functions deliberately, as
2014  * the standard behavior is to write all copies in a raid setup. here we only
2015  * want to write the one bad copy. so we do the mapping for ourselves and issue
2016  * submit_bio directly.
2017  * to avoid any synchronization issues, wait for the data after writing, which
2018  * actually prevents the read that triggered the error from finishing.
2019  * currently, there can be no more than two copies of every data bit. thus,
2020  * exactly one rewrite is required.
2021  */
2022 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 start,
2023                         u64 length, u64 logical, struct page *page,
2024                         int mirror_num)
2025 {
2026         struct bio *bio;
2027         struct btrfs_device *dev;
2028         DECLARE_COMPLETION_ONSTACK(compl);
2029         u64 map_length = 0;
2030         u64 sector;
2031         struct btrfs_bio *bbio = NULL;
2032         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2033         int ret;
2034 
2035         BUG_ON(!mirror_num);
2036 
2037         /* we can't repair anything in raid56 yet */
2038         if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2039                 return 0;
2040 
2041         bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2042         if (!bio)
2043                 return -EIO;
2044         bio->bi_private = &compl;
2045         bio->bi_end_io = repair_io_failure_callback;
2046         bio->bi_size = 0;
2047         map_length = length;
2048 
2049         ret = btrfs_map_block(fs_info, WRITE, logical,
2050                               &map_length, &bbio, mirror_num);
2051         if (ret) {
2052                 bio_put(bio);
2053                 return -EIO;
2054         }
2055         BUG_ON(mirror_num != bbio->mirror_num);
2056         sector = bbio->stripes[mirror_num-1].physical >> 9;
2057         bio->bi_sector = sector;
2058         dev = bbio->stripes[mirror_num-1].dev;
2059         kfree(bbio);
2060         if (!dev || !dev->bdev || !dev->writeable) {
2061                 bio_put(bio);
2062                 return -EIO;
2063         }
2064         bio->bi_bdev = dev->bdev;
2065         bio_add_page(bio, page, length, start - page_offset(page));
2066         btrfsic_submit_bio(WRITE_SYNC, bio);
2067         wait_for_completion(&compl);
2068 
2069         if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2070                 /* try to remap that extent elsewhere? */
2071                 bio_put(bio);
2072                 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2073                 return -EIO;
2074         }
2075 
2076         printk_ratelimited_in_rcu(KERN_INFO "btrfs read error corrected: ino %lu off %llu "
2077                       "(dev %s sector %llu)\n", page->mapping->host->i_ino,
2078                       start, rcu_str_deref(dev->name), sector);
2079 
2080         bio_put(bio);
2081         return 0;
2082 }
2083 
2084 int repair_eb_io_failure(struct btrfs_root *root, struct extent_buffer *eb,
2085                          int mirror_num)
2086 {
2087         u64 start = eb->start;
2088         unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2089         int ret = 0;
2090 
2091         for (i = 0; i < num_pages; i++) {
2092                 struct page *p = extent_buffer_page(eb, i);
2093                 ret = repair_io_failure(root->fs_info, start, PAGE_CACHE_SIZE,
2094                                         start, p, mirror_num);
2095                 if (ret)
2096                         break;
2097                 start += PAGE_CACHE_SIZE;
2098         }
2099 
2100         return ret;
2101 }
2102 
2103 /*
2104  * each time an IO finishes, we do a fast check in the IO failure tree
2105  * to see if we need to process or clean up an io_failure_record
2106  */
2107 static int clean_io_failure(u64 start, struct page *page)
2108 {
2109         u64 private;
2110         u64 private_failure;
2111         struct io_failure_record *failrec;
2112         struct btrfs_fs_info *fs_info;
2113         struct extent_state *state;
2114         int num_copies;
2115         int did_repair = 0;
2116         int ret;
2117         struct inode *inode = page->mapping->host;
2118 
2119         private = 0;
2120         ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2121                                 (u64)-1, 1, EXTENT_DIRTY, 0);
2122         if (!ret)
2123                 return 0;
2124 
2125         ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2126                                 &private_failure);
2127         if (ret)
2128                 return 0;
2129 
2130         failrec = (struct io_failure_record *)(unsigned long) private_failure;
2131         BUG_ON(!failrec->this_mirror);
2132 
2133         if (failrec->in_validation) {
2134                 /* there was no real error, just free the record */
2135                 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2136                          failrec->start);
2137                 did_repair = 1;
2138                 goto out;
2139         }
2140 
2141         spin_lock(&BTRFS_I(inode)->io_tree.lock);
2142         state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2143                                             failrec->start,
2144                                             EXTENT_LOCKED);
2145         spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2146 
2147         if (state && state->start == failrec->start) {
2148                 fs_info = BTRFS_I(inode)->root->fs_info;
2149                 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2150                                               failrec->len);
2151                 if (num_copies > 1)  {
2152                         ret = repair_io_failure(fs_info, start, failrec->len,
2153                                                 failrec->logical, page,
2154                                                 failrec->failed_mirror);
2155                         did_repair = !ret;
2156                 }
2157                 ret = 0;
2158         }
2159 
2160 out:
2161         if (!ret)
2162                 ret = free_io_failure(inode, failrec, did_repair);
2163 
2164         return ret;
2165 }
2166 
2167 /*
2168  * this is a generic handler for readpage errors (default
2169  * readpage_io_failed_hook). if other copies exist, read those and write back
2170  * good data to the failed position. does not investigate in remapping the
2171  * failed extent elsewhere, hoping the device will be smart enough to do this as
2172  * needed
2173  */
2174 
2175 static int bio_readpage_error(struct bio *failed_bio, struct page *page,
2176                                 u64 start, u64 end, int failed_mirror,
2177                                 struct extent_state *state)
2178 {
2179         struct io_failure_record *failrec = NULL;
2180         u64 private;
2181         struct extent_map *em;
2182         struct inode *inode = page->mapping->host;
2183         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2184         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2185         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2186         struct bio *bio;
2187         int num_copies;
2188         int ret;
2189         int read_mode;
2190         u64 logical;
2191 
2192         BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2193 
2194         ret = get_state_private(failure_tree, start, &private);
2195         if (ret) {
2196                 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2197                 if (!failrec)
2198                         return -ENOMEM;
2199                 failrec->start = start;
2200                 failrec->len = end - start + 1;
2201                 failrec->this_mirror = 0;
2202                 failrec->bio_flags = 0;
2203                 failrec->in_validation = 0;
2204 
2205                 read_lock(&em_tree->lock);
2206                 em = lookup_extent_mapping(em_tree, start, failrec->len);
2207                 if (!em) {
2208                         read_unlock(&em_tree->lock);
2209                         kfree(failrec);
2210                         return -EIO;
2211                 }
2212 
2213                 if (em->start > start || em->start + em->len < start) {
2214                         free_extent_map(em);
2215                         em = NULL;
2216                 }
2217                 read_unlock(&em_tree->lock);
2218 
2219                 if (!em) {
2220                         kfree(failrec);
2221                         return -EIO;
2222                 }
2223                 logical = start - em->start;
2224                 logical = em->block_start + logical;
2225                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2226                         logical = em->block_start;
2227                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2228                         extent_set_compress_type(&failrec->bio_flags,
2229                                                  em->compress_type);
2230                 }
2231                 pr_debug("bio_readpage_error: (new) logical=%llu, start=%llu, "
2232                          "len=%llu\n", logical, start, failrec->len);
2233                 failrec->logical = logical;
2234                 free_extent_map(em);
2235 
2236                 /* set the bits in the private failure tree */
2237                 ret = set_extent_bits(failure_tree, start, end,
2238                                         EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2239                 if (ret >= 0)
2240                         ret = set_state_private(failure_tree, start,
2241                                                 (u64)(unsigned long)failrec);
2242                 /* set the bits in the inode's tree */
2243                 if (ret >= 0)
2244                         ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2245                                                 GFP_NOFS);
2246                 if (ret < 0) {
2247                         kfree(failrec);
2248                         return ret;
2249                 }
2250         } else {
2251                 failrec = (struct io_failure_record *)(unsigned long)private;
2252                 pr_debug("bio_readpage_error: (found) logical=%llu, "
2253                          "start=%llu, len=%llu, validation=%d\n",
2254                          failrec->logical, failrec->start, failrec->len,
2255                          failrec->in_validation);
2256                 /*
2257                  * when data can be on disk more than twice, add to failrec here
2258                  * (e.g. with a list for failed_mirror) to make
2259                  * clean_io_failure() clean all those errors at once.
2260                  */
2261         }
2262         num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2263                                       failrec->logical, failrec->len);
2264         if (num_copies == 1) {
2265                 /*
2266                  * we only have a single copy of the data, so don't bother with
2267                  * all the retry and error correction code that follows. no
2268                  * matter what the error is, it is very likely to persist.
2269                  */
2270                 pr_debug("bio_readpage_error: cannot repair, num_copies == 1. "
2271                          "state=%p, num_copies=%d, next_mirror %d, "
2272                          "failed_mirror %d\n", state, num_copies,
2273                          failrec->this_mirror, failed_mirror);
2274                 free_io_failure(inode, failrec, 0);
2275                 return -EIO;
2276         }
2277 
2278         if (!state) {
2279                 spin_lock(&tree->lock);
2280                 state = find_first_extent_bit_state(tree, failrec->start,
2281                                                     EXTENT_LOCKED);
2282                 if (state && state->start != failrec->start)
2283                         state = NULL;
2284                 spin_unlock(&tree->lock);
2285         }
2286 
2287         /*
2288          * there are two premises:
2289          *      a) deliver good data to the caller
2290          *      b) correct the bad sectors on disk
2291          */
2292         if (failed_bio->bi_vcnt > 1) {
2293                 /*
2294                  * to fulfill b), we need to know the exact failing sectors, as
2295                  * we don't want to rewrite any more than the failed ones. thus,
2296                  * we need separate read requests for the failed bio
2297                  *
2298                  * if the following BUG_ON triggers, our validation request got
2299                  * merged. we need separate requests for our algorithm to work.
2300                  */
2301                 BUG_ON(failrec->in_validation);
2302                 failrec->in_validation = 1;
2303                 failrec->this_mirror = failed_mirror;
2304                 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2305         } else {
2306                 /*
2307                  * we're ready to fulfill a) and b) alongside. get a good copy
2308                  * of the failed sector and if we succeed, we have setup
2309                  * everything for repair_io_failure to do the rest for us.
2310                  */
2311                 if (failrec->in_validation) {
2312                         BUG_ON(failrec->this_mirror != failed_mirror);
2313                         failrec->in_validation = 0;
2314                         failrec->this_mirror = 0;
2315                 }
2316                 failrec->failed_mirror = failed_mirror;
2317                 failrec->this_mirror++;
2318                 if (failrec->this_mirror == failed_mirror)
2319                         failrec->this_mirror++;
2320                 read_mode = READ_SYNC;
2321         }
2322 
2323         if (!state || failrec->this_mirror > num_copies) {
2324                 pr_debug("bio_readpage_error: (fail) state=%p, num_copies=%d, "
2325                          "next_mirror %d, failed_mirror %d\n", state,
2326                          num_copies, failrec->this_mirror, failed_mirror);
2327                 free_io_failure(inode, failrec, 0);
2328                 return -EIO;
2329         }
2330 
2331         bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2332         if (!bio) {
2333                 free_io_failure(inode, failrec, 0);
2334                 return -EIO;
2335         }
2336         bio->bi_private = state;
2337         bio->bi_end_io = failed_bio->bi_end_io;
2338         bio->bi_sector = failrec->logical >> 9;
2339         bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2340         bio->bi_size = 0;
2341 
2342         bio_add_page(bio, page, failrec->len, start - page_offset(page));
2343 
2344         pr_debug("bio_readpage_error: submitting new read[%#x] to "
2345                  "this_mirror=%d, num_copies=%d, in_validation=%d\n", read_mode,
2346                  failrec->this_mirror, num_copies, failrec->in_validation);
2347 
2348         ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2349                                          failrec->this_mirror,
2350                                          failrec->bio_flags, 0);
2351         return ret;
2352 }
2353 
2354 /* lots and lots of room for performance fixes in the end_bio funcs */
2355 
2356 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2357 {
2358         int uptodate = (err == 0);
2359         struct extent_io_tree *tree;
2360         int ret = 0;
2361 
2362         tree = &BTRFS_I(page->mapping->host)->io_tree;
2363 
2364         if (tree->ops && tree->ops->writepage_end_io_hook) {
2365                 ret = tree->ops->writepage_end_io_hook(page, start,
2366                                                end, NULL, uptodate);
2367                 if (ret)
2368                         uptodate = 0;
2369         }
2370 
2371         if (!uptodate) {
2372                 ClearPageUptodate(page);
2373                 SetPageError(page);
2374                 ret = ret < 0 ? ret : -EIO;
2375                 mapping_set_error(page->mapping, ret);
2376         }
2377         return 0;
2378 }
2379 
2380 /*
2381  * after a writepage IO is done, we need to:
2382  * clear the uptodate bits on error
2383  * clear the writeback bits in the extent tree for this IO
2384  * end_page_writeback if the page has no more pending IO
2385  *
2386  * Scheduling is not allowed, so the extent state tree is expected
2387  * to have one and only one object corresponding to this IO.
2388  */
2389 static void end_bio_extent_writepage(struct bio *bio, int err)
2390 {
2391         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2392         struct extent_io_tree *tree;
2393         u64 start;
2394         u64 end;
2395 
2396         do {
2397                 struct page *page = bvec->bv_page;
2398                 tree = &BTRFS_I(page->mapping->host)->io_tree;
2399 
2400                 /* We always issue full-page reads, but if some block
2401                  * in a page fails to read, blk_update_request() will
2402                  * advance bv_offset and adjust bv_len to compensate.
2403                  * Print a warning for nonzero offsets, and an error
2404                  * if they don't add up to a full page.  */
2405                 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2406                         printk("%s page write in btrfs with offset %u and length %u\n",
2407                                bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2408                                ? KERN_ERR "partial" : KERN_INFO "incomplete",
2409                                bvec->bv_offset, bvec->bv_len);
2410 
2411                 start = page_offset(page);
2412                 end = start + bvec->bv_offset + bvec->bv_len - 1;
2413 
2414                 if (--bvec >= bio->bi_io_vec)
2415                         prefetchw(&bvec->bv_page->flags);
2416 
2417                 if (end_extent_writepage(page, err, start, end))
2418                         continue;
2419 
2420                 end_page_writeback(page);
2421         } while (bvec >= bio->bi_io_vec);
2422 
2423         bio_put(bio);
2424 }
2425 
2426 /*
2427  * after a readpage IO is done, we need to:
2428  * clear the uptodate bits on error
2429  * set the uptodate bits if things worked
2430  * set the page up to date if all extents in the tree are uptodate
2431  * clear the lock bit in the extent tree
2432  * unlock the page if there are no other extents locked for it
2433  *
2434  * Scheduling is not allowed, so the extent state tree is expected
2435  * to have one and only one object corresponding to this IO.
2436  */
2437 static void end_bio_extent_readpage(struct bio *bio, int err)
2438 {
2439         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2440         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
2441         struct bio_vec *bvec = bio->bi_io_vec;
2442         struct extent_io_tree *tree;
2443         u64 start;
2444         u64 end;
2445         int mirror;
2446         int ret;
2447 
2448         if (err)
2449                 uptodate = 0;
2450 
2451         do {
2452                 struct page *page = bvec->bv_page;
2453                 struct extent_state *cached = NULL;
2454                 struct extent_state *state;
2455                 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2456 
2457                 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2458                          "mirror=%lu\n", (u64)bio->bi_sector, err,
2459                          io_bio->mirror_num);
2460                 tree = &BTRFS_I(page->mapping->host)->io_tree;
2461 
2462                 /* We always issue full-page reads, but if some block
2463                  * in a page fails to read, blk_update_request() will
2464                  * advance bv_offset and adjust bv_len to compensate.
2465                  * Print a warning for nonzero offsets, and an error
2466                  * if they don't add up to a full page.  */
2467                 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE)
2468                         printk("%s page read in btrfs with offset %u and length %u\n",
2469                                bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE
2470                                ? KERN_ERR "partial" : KERN_INFO "incomplete",
2471                                bvec->bv_offset, bvec->bv_len);
2472 
2473                 start = page_offset(page);
2474                 end = start + bvec->bv_offset + bvec->bv_len - 1;
2475 
2476                 if (++bvec <= bvec_end)
2477                         prefetchw(&bvec->bv_page->flags);
2478 
2479                 spin_lock(&tree->lock);
2480                 state = find_first_extent_bit_state(tree, start, EXTENT_LOCKED);
2481                 if (state && state->start == start) {
2482                         /*
2483                          * take a reference on the state, unlock will drop
2484                          * the ref
2485                          */
2486                         cache_state(state, &cached);
2487                 }
2488                 spin_unlock(&tree->lock);
2489 
2490                 mirror = io_bio->mirror_num;
2491                 if (uptodate && tree->ops && tree->ops->readpage_end_io_hook) {
2492                         ret = tree->ops->readpage_end_io_hook(page, start, end,
2493                                                               state, mirror);
2494                         if (ret)
2495                                 uptodate = 0;
2496                         else
2497                                 clean_io_failure(start, page);
2498                 }
2499 
2500                 if (!uptodate && tree->ops && tree->ops->readpage_io_failed_hook) {
2501                         ret = tree->ops->readpage_io_failed_hook(page, mirror);
2502                         if (!ret && !err &&
2503                             test_bit(BIO_UPTODATE, &bio->bi_flags))
2504                                 uptodate = 1;
2505                 } else if (!uptodate) {
2506                         /*
2507                          * The generic bio_readpage_error handles errors the
2508                          * following way: If possible, new read requests are
2509                          * created and submitted and will end up in
2510                          * end_bio_extent_readpage as well (if we're lucky, not
2511                          * in the !uptodate case). In that case it returns 0 and
2512                          * we just go on with the next page in our bio. If it
2513                          * can't handle the error it will return -EIO and we
2514                          * remain responsible for that page.
2515                          */
2516                         ret = bio_readpage_error(bio, page, start, end, mirror, NULL);
2517                         if (ret == 0) {
2518                                 uptodate =
2519                                         test_bit(BIO_UPTODATE, &bio->bi_flags);
2520                                 if (err)
2521                                         uptodate = 0;
2522                                 uncache_state(&cached);
2523                                 continue;
2524                         }
2525                 }
2526 
2527                 if (uptodate && tree->track_uptodate) {
2528                         set_extent_uptodate(tree, start, end, &cached,
2529                                             GFP_ATOMIC);
2530                 }
2531                 unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2532 
2533                 if (uptodate) {
2534                         SetPageUptodate(page);
2535                 } else {
2536                         ClearPageUptodate(page);
2537                         SetPageError(page);
2538                 }
2539                 unlock_page(page);
2540         } while (bvec <= bvec_end);
2541 
2542         bio_put(bio);
2543 }
2544 
2545 /*
2546  * this allocates from the btrfs_bioset.  We're returning a bio right now
2547  * but you can call btrfs_io_bio for the appropriate container_of magic
2548  */
2549 struct bio *
2550 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2551                 gfp_t gfp_flags)
2552 {
2553         struct bio *bio;
2554 
2555         bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2556 
2557         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2558                 while (!bio && (nr_vecs /= 2)) {
2559                         bio = bio_alloc_bioset(gfp_flags,
2560                                                nr_vecs, btrfs_bioset);
2561                 }
2562         }
2563 
2564         if (bio) {
2565                 bio->bi_size = 0;
2566                 bio->bi_bdev = bdev;
2567                 bio->bi_sector = first_sector;
2568         }
2569         return bio;
2570 }
2571 
2572 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2573 {
2574         return bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2575 }
2576 
2577 
2578 /* this also allocates from the btrfs_bioset */
2579 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2580 {
2581         return bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2582 }
2583 
2584 
2585 static int __must_check submit_one_bio(int rw, struct bio *bio,
2586                                        int mirror_num, unsigned long bio_flags)
2587 {
2588         int ret = 0;
2589         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2590         struct page *page = bvec->bv_page;
2591         struct extent_io_tree *tree = bio->bi_private;
2592         u64 start;
2593 
2594         start = page_offset(page) + bvec->bv_offset;
2595 
2596         bio->bi_private = NULL;
2597 
2598         bio_get(bio);
2599 
2600         if (tree->ops && tree->ops->submit_bio_hook)
2601                 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2602                                            mirror_num, bio_flags, start);
2603         else
2604                 btrfsic_submit_bio(rw, bio);
2605 
2606         if (bio_flagged(bio, BIO_EOPNOTSUPP))
2607                 ret = -EOPNOTSUPP;
2608         bio_put(bio);
2609         return ret;
2610 }
2611 
2612 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2613                      unsigned long offset, size_t size, struct bio *bio,
2614                      unsigned long bio_flags)
2615 {
2616         int ret = 0;
2617         if (tree->ops && tree->ops->merge_bio_hook)
2618                 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2619                                                 bio_flags);
2620         BUG_ON(ret < 0);
2621         return ret;
2622 
2623 }
2624 
2625 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2626                               struct page *page, sector_t sector,
2627                               size_t size, unsigned long offset,
2628                               struct block_device *bdev,
2629                               struct bio **bio_ret,
2630                               unsigned long max_pages,
2631                               bio_end_io_t end_io_func,
2632                               int mirror_num,
2633                               unsigned long prev_bio_flags,
2634                               unsigned long bio_flags)
2635 {
2636         int ret = 0;
2637         struct bio *bio;
2638         int nr;
2639         int contig = 0;
2640         int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2641         int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2642         size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2643 
2644         if (bio_ret && *bio_ret) {
2645                 bio = *bio_ret;
2646                 if (old_compressed)
2647                         contig = bio->bi_sector == sector;
2648                 else
2649                         contig = bio_end_sector(bio) == sector;
2650 
2651                 if (prev_bio_flags != bio_flags || !contig ||
2652                     merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2653                     bio_add_page(bio, page, page_size, offset) < page_size) {
2654                         ret = submit_one_bio(rw, bio, mirror_num,
2655                                              prev_bio_flags);
2656                         if (ret < 0)
2657                                 return ret;
2658                         bio = NULL;
2659                 } else {
2660                         return 0;
2661                 }
2662         }
2663         if (this_compressed)
2664                 nr = BIO_MAX_PAGES;
2665         else
2666                 nr = bio_get_nr_vecs(bdev);
2667 
2668         bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2669         if (!bio)
2670                 return -ENOMEM;
2671 
2672         bio_add_page(bio, page, page_size, offset);
2673         bio->bi_end_io = end_io_func;
2674         bio->bi_private = tree;
2675 
2676         if (bio_ret)
2677                 *bio_ret = bio;
2678         else
2679                 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2680 
2681         return ret;
2682 }
2683 
2684 static void attach_extent_buffer_page(struct extent_buffer *eb,
2685                                       struct page *page)
2686 {
2687         if (!PagePrivate(page)) {
2688                 SetPagePrivate(page);
2689                 page_cache_get(page);
2690                 set_page_private(page, (unsigned long)eb);
2691         } else {
2692                 WARN_ON(page->private != (unsigned long)eb);
2693         }
2694 }
2695 
2696 void set_page_extent_mapped(struct page *page)
2697 {
2698         if (!PagePrivate(page)) {
2699                 SetPagePrivate(page);
2700                 page_cache_get(page);
2701                 set_page_private(page, EXTENT_PAGE_PRIVATE);
2702         }
2703 }
2704 
2705 /*
2706  * basic readpage implementation.  Locked extent state structs are inserted
2707  * into the tree that are removed when the IO is done (by the end_io
2708  * handlers)
2709  * XXX JDM: This needs looking at to ensure proper page locking
2710  */
2711 static int __extent_read_full_page(struct extent_io_tree *tree,
2712                                    struct page *page,
2713                                    get_extent_t *get_extent,
2714                                    struct bio **bio, int mirror_num,
2715                                    unsigned long *bio_flags, int rw)
2716 {
2717         struct inode *inode = page->mapping->host;
2718         u64 start = page_offset(page);
2719         u64 page_end = start + PAGE_CACHE_SIZE - 1;
2720         u64 end;
2721         u64 cur = start;
2722         u64 extent_offset;
2723         u64 last_byte = i_size_read(inode);
2724         u64 block_start;
2725         u64 cur_end;
2726         sector_t sector;
2727         struct extent_map *em;
2728         struct block_device *bdev;
2729         struct btrfs_ordered_extent *ordered;
2730         int ret;
2731         int nr = 0;
2732         size_t pg_offset = 0;
2733         size_t iosize;
2734         size_t disk_io_size;
2735         size_t blocksize = inode->i_sb->s_blocksize;
2736         unsigned long this_bio_flag = 0;
2737 
2738         set_page_extent_mapped(page);
2739 
2740         if (!PageUptodate(page)) {
2741                 if (cleancache_get_page(page) == 0) {
2742                         BUG_ON(blocksize != PAGE_SIZE);
2743                         goto out;
2744                 }
2745         }
2746 
2747         end = page_end;
2748         while (1) {
2749                 lock_extent(tree, start, end);
2750                 ordered = btrfs_lookup_ordered_extent(inode, start);
2751                 if (!ordered)
2752                         break;
2753                 unlock_extent(tree, start, end);
2754                 btrfs_start_ordered_extent(inode, ordered, 1);
2755                 btrfs_put_ordered_extent(ordered);
2756         }
2757 
2758         if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2759                 char *userpage;
2760                 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2761 
2762                 if (zero_offset) {
2763                         iosize = PAGE_CACHE_SIZE - zero_offset;
2764                         userpage = kmap_atomic(page);
2765                         memset(userpage + zero_offset, 0, iosize);
2766                         flush_dcache_page(page);
2767                         kunmap_atomic(userpage);
2768                 }
2769         }
2770         while (cur <= end) {
2771                 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2772 
2773                 if (cur >= last_byte) {
2774                         char *userpage;
2775                         struct extent_state *cached = NULL;
2776 
2777                         iosize = PAGE_CACHE_SIZE - pg_offset;
2778                         userpage = kmap_atomic(page);
2779                         memset(userpage + pg_offset, 0, iosize);
2780                         flush_dcache_page(page);
2781                         kunmap_atomic(userpage);
2782                         set_extent_uptodate(tree, cur, cur + iosize - 1,
2783                                             &cached, GFP_NOFS);
2784                         unlock_extent_cached(tree, cur, cur + iosize - 1,
2785                                              &cached, GFP_NOFS);
2786                         break;
2787                 }
2788                 em = get_extent(inode, page, pg_offset, cur,
2789                                 end - cur + 1, 0);
2790                 if (IS_ERR_OR_NULL(em)) {
2791                         SetPageError(page);
2792                         unlock_extent(tree, cur, end);
2793                         break;
2794                 }
2795                 extent_offset = cur - em->start;
2796                 BUG_ON(extent_map_end(em) <= cur);
2797                 BUG_ON(end < cur);
2798 
2799                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2800                         this_bio_flag = EXTENT_BIO_COMPRESSED;
2801                         extent_set_compress_type(&this_bio_flag,
2802                                                  em->compress_type);
2803                 }
2804 
2805                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2806                 cur_end = min(extent_map_end(em) - 1, end);
2807                 iosize = ALIGN(iosize, blocksize);
2808                 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2809                         disk_io_size = em->block_len;
2810                         sector = em->block_start >> 9;
2811                 } else {
2812                         sector = (em->block_start + extent_offset) >> 9;
2813                         disk_io_size = iosize;
2814                 }
2815                 bdev = em->bdev;
2816                 block_start = em->block_start;
2817                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2818                         block_start = EXTENT_MAP_HOLE;
2819                 free_extent_map(em);
2820                 em = NULL;
2821 
2822                 /* we've found a hole, just zero and go on */
2823                 if (block_start == EXTENT_MAP_HOLE) {
2824                         char *userpage;
2825                         struct extent_state *cached = NULL;
2826 
2827                         userpage = kmap_atomic(page);
2828                         memset(userpage + pg_offset, 0, iosize);
2829                         flush_dcache_page(page);
2830                         kunmap_atomic(userpage);
2831 
2832                         set_extent_uptodate(tree, cur, cur + iosize - 1,
2833                                             &cached, GFP_NOFS);
2834                         unlock_extent_cached(tree, cur, cur + iosize - 1,
2835                                              &cached, GFP_NOFS);
2836                         cur = cur + iosize;
2837                         pg_offset += iosize;
2838                         continue;
2839                 }
2840                 /* the get_extent function already copied into the page */
2841                 if (test_range_bit(tree, cur, cur_end,
2842                                    EXTENT_UPTODATE, 1, NULL)) {
2843                         check_page_uptodate(tree, page);
2844                         unlock_extent(tree, cur, cur + iosize - 1);
2845                         cur = cur + iosize;
2846                         pg_offset += iosize;
2847                         continue;
2848                 }
2849                 /* we have an inline extent but it didn't get marked up
2850                  * to date.  Error out
2851                  */
2852                 if (block_start == EXTENT_MAP_INLINE) {
2853                         SetPageError(page);
2854                         unlock_extent(tree, cur, cur + iosize - 1);
2855                         cur = cur + iosize;
2856                         pg_offset += iosize;
2857                         continue;
2858                 }
2859 
2860                 pnr -= page->index;
2861                 ret = submit_extent_page(rw, tree, page,
2862                                          sector, disk_io_size, pg_offset,
2863                                          bdev, bio, pnr,
2864                                          end_bio_extent_readpage, mirror_num,
2865                                          *bio_flags,
2866                                          this_bio_flag);
2867                 if (!ret) {
2868                         nr++;
2869                         *bio_flags = this_bio_flag;
2870                 } else {
2871                         SetPageError(page);
2872                         unlock_extent(tree, cur, cur + iosize - 1);
2873                 }
2874                 cur = cur + iosize;
2875                 pg_offset += iosize;
2876         }
2877 out:
2878         if (!nr) {
2879                 if (!PageError(page))
2880                         SetPageUptodate(page);
2881                 unlock_page(page);
2882         }
2883         return 0;
2884 }
2885 
2886 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
2887                             get_extent_t *get_extent, int mirror_num)
2888 {
2889         struct bio *bio = NULL;
2890         unsigned long bio_flags = 0;
2891         int ret;
2892 
2893         ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
2894                                       &bio_flags, READ);
2895         if (bio)
2896                 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
2897         return ret;
2898 }
2899 
2900 static noinline void update_nr_written(struct page *page,
2901                                       struct writeback_control *wbc,
2902                                       unsigned long nr_written)
2903 {
2904         wbc->nr_to_write -= nr_written;
2905         if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
2906             wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
2907                 page->mapping->writeback_index = page->index + nr_written;
2908 }
2909 
2910 /*
2911  * the writepage semantics are similar to regular writepage.  extent
2912  * records are inserted to lock ranges in the tree, and as dirty areas
2913  * are found, they are marked writeback.  Then the lock bits are removed
2914  * and the end_io handler clears the writeback ranges
2915  */
2916 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2917                               void *data)
2918 {
2919         struct inode *inode = page->mapping->host;
2920         struct extent_page_data *epd = data;
2921         struct extent_io_tree *tree = epd->tree;
2922         u64 start = page_offset(page);
2923         u64 delalloc_start;
2924         u64 page_end = start + PAGE_CACHE_SIZE - 1;
2925         u64 end;
2926         u64 cur = start;
2927         u64 extent_offset;
2928         u64 last_byte = i_size_read(inode);
2929         u64 block_start;
2930         u64 iosize;
2931         sector_t sector;
2932         struct extent_state *cached_state = NULL;
2933         struct extent_map *em;
2934         struct block_device *bdev;
2935         int ret;
2936         int nr = 0;
2937         size_t pg_offset = 0;
2938         size_t blocksize;
2939         loff_t i_size = i_size_read(inode);
2940         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
2941         u64 nr_delalloc;
2942         u64 delalloc_end;
2943         int page_started;
2944         int compressed;
2945         int write_flags;
2946         unsigned long nr_written = 0;
2947         bool fill_delalloc = true;
2948 
2949         if (wbc->sync_mode == WB_SYNC_ALL)
2950                 write_flags = WRITE_SYNC;
2951         else
2952                 write_flags = WRITE;
2953 
2954         trace___extent_writepage(page, inode, wbc);
2955 
2956         WARN_ON(!PageLocked(page));
2957 
2958         ClearPageError(page);
2959 
2960         pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
2961         if (page->index > end_index ||
2962            (page->index == end_index && !pg_offset)) {
2963                 page->mapping->a_ops->invalidatepage(page, 0);
2964                 unlock_page(page);
2965                 return 0;
2966         }
2967 
2968         if (page->index == end_index) {
2969                 char *userpage;
2970 
2971                 userpage = kmap_atomic(page);
2972                 memset(userpage + pg_offset, 0,
2973                        PAGE_CACHE_SIZE - pg_offset);
2974                 kunmap_atomic(userpage);
2975                 flush_dcache_page(page);
2976         }
2977         pg_offset = 0;
2978 
2979         set_page_extent_mapped(page);
2980 
2981         if (!tree->ops || !tree->ops->fill_delalloc)
2982                 fill_delalloc = false;
2983 
2984         delalloc_start = start;
2985         delalloc_end = 0;
2986         page_started = 0;
2987         if (!epd->extent_locked && fill_delalloc) {
2988                 u64 delalloc_to_write = 0;
2989                 /*
2990                  * make sure the wbc mapping index is at least updated
2991                  * to this page.
2992                  */
2993                 update_nr_written(page, wbc, 0);
2994 
2995                 while (delalloc_end < page_end) {
2996                         nr_delalloc = find_lock_delalloc_range(inode, tree,
2997                                                        page,
2998                                                        &delalloc_start,
2999                                                        &delalloc_end,
3000                                                        128 * 1024 * 1024);
3001                         if (nr_delalloc == 0) {
3002                                 delalloc_start = delalloc_end + 1;
3003                                 continue;
3004                         }
3005                         ret = tree->ops->fill_delalloc(inode, page,
3006                                                        delalloc_start,
3007                                                        delalloc_end,
3008                                                        &page_started,
3009                                                        &nr_written);
3010                         /* File system has been set read-only */
3011                         if (ret) {
3012                                 SetPageError(page);
3013                                 goto done;
3014                         }
3015                         /*
3016                          * delalloc_end is already one less than the total
3017                          * length, so we don't subtract one from
3018                          * PAGE_CACHE_SIZE
3019                          */
3020                         delalloc_to_write += (delalloc_end - delalloc_start +
3021                                               PAGE_CACHE_SIZE) >>
3022                                               PAGE_CACHE_SHIFT;
3023                         delalloc_start = delalloc_end + 1;
3024                 }
3025                 if (wbc->nr_to_write < delalloc_to_write) {
3026                         int thresh = 8192;
3027 
3028                         if (delalloc_to_write < thresh * 2)
3029                                 thresh = delalloc_to_write;
3030                         wbc->nr_to_write = min_t(u64, delalloc_to_write,
3031                                                  thresh);
3032                 }
3033 
3034                 /* did the fill delalloc function already unlock and start
3035                  * the IO?
3036                  */
3037                 if (page_started) {
3038                         ret = 0;
3039                         /*
3040                          * we've unlocked the page, so we can't update
3041                          * the mapping's writeback index, just update
3042                          * nr_to_write.
3043                          */
3044                         wbc->nr_to_write -= nr_written;
3045                         goto done_unlocked;
3046                 }
3047         }
3048         if (tree->ops && tree->ops->writepage_start_hook) {
3049                 ret = tree->ops->writepage_start_hook(page, start,
3050                                                       page_end);
3051                 if (ret) {
3052                         /* Fixup worker will requeue */
3053                         if (ret == -EBUSY)
3054                                 wbc->pages_skipped++;
3055                         else
3056                                 redirty_page_for_writepage(wbc, page);
3057                         update_nr_written(page, wbc, nr_written);
3058                         unlock_page(page);
3059                         ret = 0;
3060                         goto done_unlocked;
3061                 }
3062         }
3063 
3064         /*
3065          * we don't want to touch the inode after unlocking the page,
3066          * so we update the mapping writeback index now
3067          */
3068         update_nr_written(page, wbc, nr_written + 1);
3069 
3070         end = page_end;
3071         if (last_byte <= start) {
3072                 if (tree->ops && tree->ops->writepage_end_io_hook)
3073                         tree->ops->writepage_end_io_hook(page, start,
3074                                                          page_end, NULL, 1);
3075                 goto done;
3076         }
3077 
3078         blocksize = inode->i_sb->s_blocksize;
3079 
3080         while (cur <= end) {
3081                 if (cur >= last_byte) {
3082                         if (tree->ops && tree->ops->writepage_end_io_hook)
3083                                 tree->ops->writepage_end_io_hook(page, cur,
3084                                                          page_end, NULL, 1);
3085                         break;
3086                 }
3087                 em = epd->get_extent(inode, page, pg_offset, cur,
3088                                      end - cur + 1, 1);
3089                 if (IS_ERR_OR_NULL(em)) {
3090                         SetPageError(page);
3091                         break;
3092                 }
3093 
3094                 extent_offset = cur - em->start;
3095                 BUG_ON(extent_map_end(em) <= cur);
3096                 BUG_ON(end < cur);
3097                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
3098                 iosize = ALIGN(iosize, blocksize);
3099                 sector = (em->block_start + extent_offset) >> 9;
3100                 bdev = em->bdev;
3101                 block_start = em->block_start;
3102                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3103                 free_extent_map(em);
3104                 em = NULL;
3105 
3106                 /*
3107                  * compressed and inline extents are written through other
3108                  * paths in the FS
3109                  */
3110                 if (compressed || block_start == EXTENT_MAP_HOLE ||
3111                     block_start == EXTENT_MAP_INLINE) {
3112                         /*
3113                          * end_io notification does not happen here for
3114                          * compressed extents
3115                          */
3116                         if (!compressed && tree->ops &&
3117                             tree->ops->writepage_end_io_hook)
3118                                 tree->ops->writepage_end_io_hook(page, cur,
3119                                                          cur + iosize - 1,
3120                                                          NULL, 1);
3121                         else if (compressed) {
3122                                 /* we don't want to end_page_writeback on
3123                                  * a compressed extent.  this happens
3124                                  * elsewhere
3125                                  */
3126                                 nr++;
3127                         }
3128 
3129                         cur += iosize;
3130                         pg_offset += iosize;
3131                         continue;
3132                 }
3133                 /* leave this out until we have a page_mkwrite call */
3134                 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
3135                                    EXTENT_DIRTY, 0, NULL)) {
3136                         cur = cur + iosize;
3137                         pg_offset += iosize;
3138                         continue;
3139                 }
3140 
3141                 if (tree->ops && tree->ops->writepage_io_hook) {
3142                         ret = tree->ops->writepage_io_hook(page, cur,
3143                                                 cur + iosize - 1);
3144                 } else {
3145                         ret = 0;
3146                 }
3147                 if (ret) {
3148                         SetPageError(page);
3149                 } else {
3150                         unsigned long max_nr = end_index + 1;
3151 
3152                         set_range_writeback(tree, cur, cur + iosize - 1);
3153                         if (!PageWriteback(page)) {
3154                                 printk(KERN_ERR "btrfs warning page %lu not "
3155                                        "writeback, cur %llu end %llu\n",
3156                                        page->index, (unsigned long long)cur,
3157                                        (unsigned long long)end);
3158                         }
3159 
3160                         ret = submit_extent_page(write_flags, tree, page,
3161                                                  sector, iosize, pg_offset,
3162                                                  bdev, &epd->bio, max_nr,
3163                                                  end_bio_extent_writepage,
3164                                                  0, 0, 0);
3165                         if (ret)
3166                                 SetPageError(page);
3167                 }
3168                 cur = cur + iosize;
3169                 pg_offset += iosize;
3170                 nr++;
3171         }
3172 done:
3173         if (nr == 0) {
3174                 /* make sure the mapping tag for page dirty gets cleared */
3175                 set_page_writeback(page);
3176                 end_page_writeback(page);
3177         }
3178         unlock_page(page);
3179 
3180 done_unlocked:
3181 
3182         /* drop our reference on any cached states */
3183         free_extent_state(cached_state);
3184         return 0;
3185 }
3186 
3187 static int eb_wait(void *word)
3188 {
3189         io_schedule();
3190         return 0;
3191 }
3192 
3193 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3194 {
3195         wait_on_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK, eb_wait,
3196                     TASK_UNINTERRUPTIBLE);
3197 }
3198 
3199 static int lock_extent_buffer_for_io(struct extent_buffer *eb,
3200                                      struct btrfs_fs_info *fs_info,
3201                                      struct extent_page_data *epd)
3202 {
3203         unsigned long i, num_pages;
3204         int flush = 0;
3205         int ret = 0;
3206 
3207         if (!btrfs_try_tree_write_lock(eb)) {
3208                 flush = 1;
3209                 flush_write_bio(epd);
3210                 btrfs_tree_lock(eb);
3211         }
3212 
3213         if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3214                 btrfs_tree_unlock(eb);
3215                 if (!epd->sync_io)
3216                         return 0;
3217                 if (!flush) {
3218                         flush_write_bio(epd);
3219                         flush = 1;
3220                 }
3221                 while (1) {
3222                         wait_on_extent_buffer_writeback(eb);
3223                         btrfs_tree_lock(eb);
3224                         if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3225                                 break;
3226                         btrfs_tree_unlock(eb);
3227                 }
3228         }
3229 
3230         /*
3231          * We need to do this to prevent races in people who check if the eb is
3232          * under IO since we can end up having no IO bits set for a short period
3233          * of time.
3234          */
3235         spin_lock(&eb->refs_lock);
3236         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3237                 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3238                 spin_unlock(&eb->refs_lock);
3239                 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3240                 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3241                                      -eb->len,
3242                                      fs_info->dirty_metadata_batch);
3243                 ret = 1;
3244         } else {
3245                 spin_unlock(&eb->refs_lock);
3246         }
3247 
3248         btrfs_tree_unlock(eb);
3249 
3250         if (!ret)
3251                 return ret;
3252 
3253         num_pages = num_extent_pages(eb->start, eb->len);
3254         for (i = 0; i < num_pages; i++) {
3255                 struct page *p = extent_buffer_page(eb, i);
3256 
3257                 if (!trylock_page(p)) {
3258                         if (!flush) {
3259                                 flush_write_bio(epd);
3260                                 flush = 1;
3261                         }
3262                         lock_page(p);
3263                 }
3264         }
3265 
3266         return ret;
3267 }
3268 
3269 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3270 {
3271         clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3272         smp_mb__after_clear_bit();
3273         wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3274 }
3275 
3276 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3277 {
3278         int uptodate = err == 0;
3279         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
3280         struct extent_buffer *eb;
3281         int done;
3282 
3283         do {
3284                 struct page *page = bvec->bv_page;
3285 
3286                 bvec--;
3287                 eb = (struct extent_buffer *)page->private;
3288                 BUG_ON(!eb);
3289                 done = atomic_dec_and_test(&eb->io_pages);
3290 
3291                 if (!uptodate || test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
3292                         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3293                         ClearPageUptodate(page);
3294                         SetPageError(page);
3295                 }
3296 
3297                 end_page_writeback(page);
3298 
3299                 if (!done)
3300                         continue;
3301 
3302                 end_extent_buffer_writeback(eb);
3303         } while (bvec >= bio->bi_io_vec);
3304 
3305         bio_put(bio);
3306 
3307 }
3308 
3309 static int write_one_eb(struct extent_buffer *eb,
3310                         struct btrfs_fs_info *fs_info,
3311                         struct writeback_control *wbc,
3312                         struct extent_page_data *epd)
3313 {
3314         struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3315         u64 offset = eb->start;
3316         unsigned long i, num_pages;
3317         unsigned long bio_flags = 0;
3318         int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3319         int ret = 0;
3320 
3321         clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3322         num_pages = num_extent_pages(eb->start, eb->len);
3323         atomic_set(&eb->io_pages, num_pages);
3324         if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3325                 bio_flags = EXTENT_BIO_TREE_LOG;
3326 
3327         for (i = 0; i < num_pages; i++) {
3328                 struct page *p = extent_buffer_page(eb, i);
3329 
3330                 clear_page_dirty_for_io(p);
3331                 set_page_writeback(p);
3332                 ret = submit_extent_page(rw, eb->tree, p, offset >> 9,
3333                                          PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3334                                          -1, end_bio_extent_buffer_writepage,
3335                                          0, epd->bio_flags, bio_flags);
3336                 epd->bio_flags = bio_flags;
3337                 if (ret) {
3338                         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
3339                         SetPageError(p);
3340                         if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3341                                 end_extent_buffer_writeback(eb);
3342                         ret = -EIO;
3343                         break;
3344                 }
3345                 offset += PAGE_CACHE_SIZE;
3346                 update_nr_written(p, wbc, 1);
3347                 unlock_page(p);
3348         }
3349 
3350         if (unlikely(ret)) {
3351                 for (; i < num_pages; i++) {
3352                         struct page *p = extent_buffer_page(eb, i);
3353                         unlock_page(p);
3354                 }
3355         }
3356 
3357         return ret;
3358 }
3359 
3360 int btree_write_cache_pages(struct address_space *mapping,
3361                                    struct writeback_control *wbc)
3362 {
3363         struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3364         struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3365         struct extent_buffer *eb, *prev_eb = NULL;
3366         struct extent_page_data epd = {
3367                 .bio = NULL,
3368                 .tree = tree,
3369                 .extent_locked = 0,
3370                 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3371                 .bio_flags = 0,
3372         };
3373         int ret = 0;
3374         int done = 0;
3375         int nr_to_write_done = 0;
3376         struct pagevec pvec;
3377         int nr_pages;
3378         pgoff_t index;
3379         pgoff_t end;            /* Inclusive */
3380         int scanned = 0;
3381         int tag;
3382 
3383         pagevec_init(&pvec, 0);
3384         if (wbc->range_cyclic) {
3385                 index = mapping->writeback_index; /* Start from prev offset */
3386                 end = -1;
3387         } else {
3388                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3389                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3390                 scanned = 1;
3391         }
3392         if (wbc->sync_mode == WB_SYNC_ALL)
3393                 tag = PAGECACHE_TAG_TOWRITE;
3394         else
3395                 tag = PAGECACHE_TAG_DIRTY;
3396 retry:
3397         if (wbc->sync_mode == WB_SYNC_ALL)
3398                 tag_pages_for_writeback(mapping, index, end);
3399         while (!done && !nr_to_write_done && (index <= end) &&
3400                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3401                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3402                 unsigned i;
3403 
3404                 scanned = 1;
3405                 for (i = 0; i < nr_pages; i++) {
3406                         struct page *page = pvec.pages[i];
3407 
3408                         if (!PagePrivate(page))
3409                                 continue;
3410 
3411                         if (!wbc->range_cyclic && page->index > end) {
3412                                 done = 1;
3413                                 break;
3414                         }
3415 
3416                         spin_lock(&mapping->private_lock);
3417                         if (!PagePrivate(page)) {
3418                                 spin_unlock(&mapping->private_lock);
3419                                 continue;
3420                         }
3421 
3422                         eb = (struct extent_buffer *)page->private;
3423 
3424                         /*
3425                          * Shouldn't happen and normally this would be a BUG_ON
3426                          * but no sense in crashing the users box for something
3427                          * we can survive anyway.
3428                          */
3429                         if (!eb) {
3430                                 spin_unlock(&mapping->private_lock);
3431                                 WARN_ON(1);
3432                                 continue;
3433                         }
3434 
3435                         if (eb == prev_eb) {
3436                                 spin_unlock(&mapping->private_lock);
3437                                 continue;
3438                         }
3439 
3440                         ret = atomic_inc_not_zero(&eb->refs);
3441                         spin_unlock(&mapping->private_lock);
3442                         if (!ret)
3443                                 continue;
3444 
3445                         prev_eb = eb;
3446                         ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3447                         if (!ret) {
3448                                 free_extent_buffer(eb);
3449                                 continue;
3450                         }
3451 
3452                         ret = write_one_eb(eb, fs_info, wbc, &epd);
3453                         if (ret) {
3454                                 done = 1;
3455                                 free_extent_buffer(eb);
3456                                 break;
3457                         }
3458                         free_extent_buffer(eb);
3459 
3460                         /*
3461                          * the filesystem may choose to bump up nr_to_write.
3462                          * We have to make sure to honor the new nr_to_write
3463                          * at any time
3464                          */
3465                         nr_to_write_done = wbc->nr_to_write <= 0;
3466                 }
3467                 pagevec_release(&pvec);
3468                 cond_resched();
3469         }
3470         if (!scanned && !done) {
3471                 /*
3472                  * We hit the last page and there is more work to be done: wrap
3473                  * back to the start of the file
3474                  */
3475                 scanned = 1;
3476                 index = 0;
3477                 goto retry;
3478         }
3479         flush_write_bio(&epd);
3480         return ret;
3481 }
3482 
3483 /**
3484  * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3485  * @mapping: address space structure to write
3486  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3487  * @writepage: function called for each page
3488  * @data: data passed to writepage function
3489  *
3490  * If a page is already under I/O, write_cache_pages() skips it, even
3491  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
3492  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
3493  * and msync() need to guarantee that all the data which was dirty at the time
3494  * the call was made get new I/O started against them.  If wbc->sync_mode is
3495  * WB_SYNC_ALL then we were called for data integrity and we must wait for
3496  * existing IO to complete.
3497  */
3498 static int extent_write_cache_pages(struct extent_io_tree *tree,
3499                              struct address_space *mapping,
3500                              struct writeback_control *wbc,
3501                              writepage_t writepage, void *data,
3502                              void (*flush_fn)(void *))
3503 {
3504         struct inode *inode = mapping->host;
3505         int ret = 0;
3506         int done = 0;
3507         int nr_to_write_done = 0;
3508         struct pagevec pvec;
3509         int nr_pages;
3510         pgoff_t index;
3511         pgoff_t end;            /* Inclusive */
3512         int scanned = 0;
3513         int tag;
3514 
3515         /*
3516          * We have to hold onto the inode so that ordered extents can do their
3517          * work when the IO finishes.  The alternative to this is failing to add
3518          * an ordered extent if the igrab() fails there and that is a huge pain
3519          * to deal with, so instead just hold onto the inode throughout the
3520          * writepages operation.  If it fails here we are freeing up the inode
3521          * anyway and we'd rather not waste our time writing out stuff that is
3522          * going to be truncated anyway.
3523          */
3524         if (!igrab(inode))
3525                 return 0;
3526 
3527         pagevec_init(&pvec, 0);
3528         if (wbc->range_cyclic) {
3529                 index = mapping->writeback_index; /* Start from prev offset */
3530                 end = -1;
3531         } else {
3532                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3533                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3534                 scanned = 1;
3535         }
3536         if (wbc->sync_mode == WB_SYNC_ALL)
3537                 tag = PAGECACHE_TAG_TOWRITE;
3538         else
3539                 tag = PAGECACHE_TAG_DIRTY;
3540 retry:
3541         if (wbc->sync_mode == WB_SYNC_ALL)
3542                 tag_pages_for_writeback(mapping, index, end);
3543         while (!done && !nr_to_write_done && (index <= end) &&
3544                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3545                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3546                 unsigned i;
3547 
3548                 scanned = 1;
3549                 for (i = 0; i < nr_pages; i++) {
3550                         struct page *page = pvec.pages[i];
3551 
3552                         /*
3553                          * At this point we hold neither mapping->tree_lock nor
3554                          * lock on the page itself: the page may be truncated or
3555                          * invalidated (changing page->mapping to NULL), or even
3556                          * swizzled back from swapper_space to tmpfs file
3557                          * mapping
3558                          */
3559                         if (!trylock_page(page)) {
3560                                 flush_fn(data);
3561                                 lock_page(page);
3562                         }
3563 
3564                         if (unlikely(page->mapping != mapping)) {
3565                                 unlock_page(page);
3566                                 continue;
3567                         }
3568 
3569                         if (!wbc->range_cyclic && page->index > end) {
3570                                 done = 1;
3571                                 unlock_page(page);
3572                                 continue;
3573                         }
3574 
3575                         if (wbc->sync_mode != WB_SYNC_NONE) {
3576                                 if (PageWriteback(page))
3577                                         flush_fn(data);
3578                                 wait_on_page_writeback(page);
3579                         }
3580 
3581                         if (PageWriteback(page) ||
3582                             !clear_page_dirty_for_io(page)) {
3583                                 unlock_page(page);
3584                                 continue;
3585                         }
3586 
3587                         ret = (*writepage)(page, wbc, data);
3588 
3589                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
3590                                 unlock_page(page);
3591                                 ret = 0;
3592                         }
3593                         if (ret)
3594                                 done = 1;
3595 
3596                         /*
3597                          * the filesystem may choose to bump up nr_to_write.
3598                          * We have to make sure to honor the new nr_to_write
3599                          * at any time
3600                          */
3601                         nr_to_write_done = wbc->nr_to_write <= 0;
3602                 }
3603                 pagevec_release(&pvec);
3604                 cond_resched();
3605         }
3606         if (!scanned && !done) {
3607                 /*
3608                  * We hit the last page and there is more work to be done: wrap
3609                  * back to the start of the file
3610                  */
3611                 scanned = 1;
3612                 index = 0;
3613                 goto retry;
3614         }
3615         btrfs_add_delayed_iput(inode);
3616         return ret;
3617 }
3618 
3619 static void flush_epd_write_bio(struct extent_page_data *epd)
3620 {
3621         if (epd->bio) {
3622                 int rw = WRITE;
3623                 int ret;
3624 
3625                 if (epd->sync_io)
3626                         rw = WRITE_SYNC;
3627 
3628                 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
3629                 BUG_ON(ret < 0); /* -ENOMEM */
3630                 epd->bio = NULL;
3631         }
3632 }
3633 
3634 static noinline void flush_write_bio(void *data)
3635 {
3636         struct extent_page_data *epd = data;
3637         flush_epd_write_bio(epd);
3638 }
3639 
3640 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
3641                           get_extent_t *get_extent,
3642                           struct writeback_control *wbc)
3643 {
3644         int ret;
3645         struct extent_page_data epd = {
3646                 .bio = NULL,
3647                 .tree = tree,
3648                 .get_extent = get_extent,
3649                 .extent_locked = 0,
3650                 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3651                 .bio_flags = 0,
3652         };
3653 
3654         ret = __extent_writepage(page, wbc, &epd);
3655 
3656         flush_epd_write_bio(&epd);
3657         return ret;
3658 }
3659 
3660 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
3661                               u64 start, u64 end, get_extent_t *get_extent,
3662                               int mode)
3663 {
3664         int ret = 0;
3665         struct address_space *mapping = inode->i_mapping;
3666         struct page *page;
3667         unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
3668                 PAGE_CACHE_SHIFT;
3669 
3670         struct extent_page_data epd = {
3671                 .bio = NULL,
3672                 .tree = tree,
3673                 .get_extent = get_extent,
3674                 .extent_locked = 1,
3675                 .sync_io = mode == WB_SYNC_ALL,
3676                 .bio_flags = 0,
3677         };
3678         struct writeback_control wbc_writepages = {
3679                 .sync_mode      = mode,
3680                 .nr_to_write    = nr_pages * 2,
3681                 .range_start    = start,
3682                 .range_end      = end + 1,
3683         };
3684 
3685         while (start <= end) {
3686                 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
3687                 if (clear_page_dirty_for_io(page))
3688                         ret = __extent_writepage(page, &wbc_writepages, &epd);
3689                 else {
3690                         if (tree->ops && tree->ops->writepage_end_io_hook)
3691                                 tree->ops->writepage_end_io_hook(page, start,
3692                                                  start + PAGE_CACHE_SIZE - 1,
3693                                                  NULL, 1);
3694                         unlock_page(page);
3695                 }
3696                 page_cache_release(page);
3697                 start += PAGE_CACHE_SIZE;
3698         }
3699 
3700         flush_epd_write_bio(&epd);
3701         return ret;
3702 }
3703 
3704 int extent_writepages(struct extent_io_tree *tree,
3705                       struct address_space *mapping,
3706                       get_extent_t *get_extent,
3707                       struct writeback_control *wbc)
3708 {
3709         int ret = 0;
3710         struct extent_page_data epd = {
3711                 .bio = NULL,
3712                 .tree = tree,
3713                 .get_extent = get_extent,
3714                 .extent_locked = 0,
3715                 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3716                 .bio_flags = 0,
3717         };
3718 
3719         ret = extent_write_cache_pages(tree, mapping, wbc,
3720                                        __extent_writepage, &epd,
3721                                        flush_write_bio);
3722         flush_epd_write_bio(&epd);
3723         return ret;
3724 }
3725 
3726 int extent_readpages(struct extent_io_tree *tree,
3727                      struct address_space *mapping,
3728                      struct list_head *pages, unsigned nr_pages,
3729                      get_extent_t get_extent)
3730 {
3731         struct bio *bio = NULL;
3732         unsigned page_idx;
3733         unsigned long bio_flags = 0;
3734         struct page *pagepool[16];
3735         struct page *page;
3736         int i = 0;
3737         int nr = 0;
3738 
3739         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
3740                 page = list_entry(pages->prev, struct page, lru);
3741 
3742                 prefetchw(&page->flags);
3743                 list_del(&page->lru);
3744                 if (add_to_page_cache_lru(page, mapping,
3745                                         page->index, GFP_NOFS)) {
3746                         page_cache_release(page);
3747                         continue;
3748                 }
3749 
3750                 pagepool[nr++] = page;
3751                 if (nr < ARRAY_SIZE(pagepool))
3752                         continue;
3753                 for (i = 0; i < nr; i++) {
3754                         __extent_read_full_page(tree, pagepool[i], get_extent,
3755                                         &bio, 0, &bio_flags, READ);
3756                         page_cache_release(pagepool[i]);
3757                 }
3758                 nr = 0;
3759         }
3760         for (i = 0; i < nr; i++) {
3761                 __extent_read_full_page(tree, pagepool[i], get_extent,
3762                                         &bio, 0, &bio_flags, READ);
3763                 page_cache_release(pagepool[i]);
3764         }
3765 
3766         BUG_ON(!list_empty(pages));
3767         if (bio)
3768                 return submit_one_bio(READ, bio, 0, bio_flags);
3769         return 0;
3770 }
3771 
3772 /*
3773  * basic invalidatepage code, this waits on any locked or writeback
3774  * ranges corresponding to the page, and then deletes any extent state
3775  * records from the tree
3776  */
3777 int extent_invalidatepage(struct extent_io_tree *tree,
3778                           struct page *page, unsigned long offset)
3779 {
3780         struct extent_state *cached_state = NULL;
3781         u64 start = page_offset(page);
3782         u64 end = start + PAGE_CACHE_SIZE - 1;
3783         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
3784 
3785         start += ALIGN(offset, blocksize);
3786         if (start > end)
3787                 return 0;
3788 
3789         lock_extent_bits(tree, start, end, 0, &cached_state);
3790         wait_on_page_writeback(page);
3791         clear_extent_bit(tree, start, end,
3792                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3793                          EXTENT_DO_ACCOUNTING,
3794                          1, 1, &cached_state, GFP_NOFS);
3795         return 0;
3796 }
3797 
3798 /*
3799  * a helper for releasepage, this tests for areas of the page that
3800  * are locked or under IO and drops the related state bits if it is safe
3801  * to drop the page.
3802  */
3803 static int try_release_extent_state(struct extent_map_tree *map,
3804                                     struct extent_io_tree *tree,
3805                                     struct page *page, gfp_t mask)
3806 {
3807         u64 start = page_offset(page);
3808         u64 end = start + PAGE_CACHE_SIZE - 1;
3809         int ret = 1;
3810 
3811         if (test_range_bit(tree, start, end,
3812                            EXTENT_IOBITS, 0, NULL))
3813                 ret = 0;
3814         else {
3815                 if ((mask & GFP_NOFS) == GFP_NOFS)
3816                         mask = GFP_NOFS;
3817                 /*
3818                  * at this point we can safely clear everything except the
3819                  * locked bit and the nodatasum bit
3820                  */
3821                 ret = clear_extent_bit(tree, start, end,
3822                                  ~(EXTENT_LOCKED | EXTENT_NODATASUM),
3823                                  0, 0, NULL, mask);
3824 
3825                 /* if clear_extent_bit failed for enomem reasons,
3826                  * we can't allow the release to continue.
3827                  */
3828                 if (ret < 0)
3829                         ret = 0;
3830                 else
3831                         ret = 1;
3832         }
3833         return ret;
3834 }
3835 
3836 /*
3837  * a helper for releasepage.  As long as there are no locked extents
3838  * in the range corresponding to the page, both state records and extent
3839  * map records are removed
3840  */
3841 int try_release_extent_mapping(struct extent_map_tree *map,
3842                                struct extent_io_tree *tree, struct page *page,
3843                                gfp_t mask)
3844 {
3845         struct extent_map *em;
3846         u64 start = page_offset(page);
3847         u64 end = start + PAGE_CACHE_SIZE - 1;
3848 
3849         if ((mask & __GFP_WAIT) &&
3850             page->mapping->host->i_size > 16 * 1024 * 1024) {
3851                 u64 len;
3852                 while (start <= end) {
3853                         len = end - start + 1;
3854                         write_lock(&map->lock);
3855                         em = lookup_extent_mapping(map, start, len);
3856                         if (!em) {
3857                                 write_unlock(&map->lock);
3858                                 break;
3859                         }
3860                         if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3861                             em->start != start) {
3862                                 write_unlock(&map->lock);
3863                                 free_extent_map(em);
3864                                 break;
3865                         }
3866                         if (!test_range_bit(tree, em->start,
3867                                             extent_map_end(em) - 1,
3868                                             EXTENT_LOCKED | EXTENT_WRITEBACK,
3869                                             0, NULL)) {
3870                                 remove_extent_mapping(map, em);
3871                                 /* once for the rb tree */
3872                                 free_extent_map(em);
3873                         }
3874                         start = extent_map_end(em);
3875                         write_unlock(&map->lock);
3876 
3877                         /* once for us */
3878                         free_extent_map(em);
3879                 }
3880         }
3881         return try_release_extent_state(map, tree, page, mask);
3882 }
3883 
3884 /*
3885  * helper function for fiemap, which doesn't want to see any holes.
3886  * This maps until we find something past 'last'
3887  */
3888 static struct extent_map *get_extent_skip_holes(struct inode *inode,
3889                                                 u64 offset,
3890                                                 u64 last,
3891                                                 get_extent_t *get_extent)
3892 {
3893         u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
3894         struct extent_map *em;
3895         u64 len;
3896 
3897         if (offset >= last)
3898                 return NULL;
3899 
3900         while(1) {
3901                 len = last - offset;
3902                 if (len == 0)
3903                         break;
3904                 len = ALIGN(len, sectorsize);
3905                 em = get_extent(inode, NULL, 0, offset, len, 0);
3906                 if (IS_ERR_OR_NULL(em))
3907                         return em;
3908 
3909                 /* if this isn't a hole return it */
3910                 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
3911                     em->block_start != EXTENT_MAP_HOLE) {
3912                         return em;
3913                 }
3914 
3915                 /* this is a hole, advance to the next extent */
3916                 offset = extent_map_end(em);
3917                 free_extent_map(em);
3918                 if (offset >= last)
3919                         break;
3920         }
3921         return NULL;
3922 }
3923 
3924 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3925                 __u64 start, __u64 len, get_extent_t *get_extent)
3926 {
3927         int ret = 0;
3928         u64 off = start;
3929         u64 max = start + len;
3930         u32 flags = 0;
3931         u32 found_type;
3932         u64 last;
3933         u64 last_for_get_extent = 0;
3934         u64 disko = 0;
3935         u64 isize = i_size_read(inode);
3936         struct btrfs_key found_key;
3937         struct extent_map *em = NULL;
3938         struct extent_state *cached_state = NULL;
3939         struct btrfs_path *path;
3940         struct btrfs_file_extent_item *item;
3941         int end = 0;
3942         u64 em_start = 0;
3943         u64 em_len = 0;
3944         u64 em_end = 0;
3945         unsigned long emflags;
3946 
3947         if (len == 0)
3948                 return -EINVAL;
3949 
3950         path = btrfs_alloc_path();
3951         if (!path)
3952                 return -ENOMEM;
3953         path->leave_spinning = 1;
3954 
3955         start = ALIGN(start, BTRFS_I(inode)->root->sectorsize);
3956         len = ALIGN(len, BTRFS_I(inode)->root->sectorsize);
3957 
3958         /*
3959          * lookup the last file extent.  We're not using i_size here
3960          * because there might be preallocation past i_size
3961          */
3962         ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root,
3963                                        path, btrfs_ino(inode), -1, 0);
3964         if (ret < 0) {
3965                 btrfs_free_path(path);
3966                 return ret;
3967         }
3968         WARN_ON(!ret);
3969         path->slots[0]--;
3970         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3971                               struct btrfs_file_extent_item);
3972         btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
3973         found_type = btrfs_key_type(&found_key);
3974 
3975         /* No extents, but there might be delalloc bits */
3976         if (found_key.objectid != btrfs_ino(inode) ||
3977             found_type != BTRFS_EXTENT_DATA_KEY) {
3978                 /* have to trust i_size as the end */
3979                 last = (u64)-1;
3980                 last_for_get_extent = isize;
3981         } else {
3982                 /*
3983                  * remember the start of the last extent.  There are a
3984                  * bunch of different factors that go into the length of the
3985                  * extent, so its much less complex to remember where it started
3986                  */
3987                 last = found_key.offset;
3988                 last_for_get_extent = last + 1;
3989         }
3990         btrfs_free_path(path);
3991 
3992         /*
3993          * we might have some extents allocated but more delalloc past those
3994          * extents.  so, we trust isize unless the start of the last extent is
3995          * beyond isize
3996          */
3997         if (last < isize) {
3998                 last = (u64)-1;
3999                 last_for_get_extent = isize;
4000         }
4001 
4002         lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4003                          &cached_state);
4004 
4005         em = get_extent_skip_holes(inode, start, last_for_get_extent,
4006                                    get_extent);
4007         if (!em)
4008                 goto out;
4009         if (IS_ERR(em)) {
4010                 ret = PTR_ERR(em);
4011                 goto out;
4012         }
4013 
4014         while (!end) {
4015                 u64 offset_in_extent;
4016 
4017                 /* break if the extent we found is outside the range */
4018                 if (em->start >= max || extent_map_end(em) < off)
4019                         break;
4020 
4021                 /*
4022                  * get_extent may return an extent that starts before our
4023                  * requested range.  We have to make sure the ranges
4024                  * we return to fiemap always move forward and don't
4025                  * overlap, so adjust the offsets here
4026                  */
4027                 em_start = max(em->start, off);
4028 
4029                 /*
4030                  * record the offset from the start of the extent
4031                  * for adjusting the disk offset below
4032                  */
4033                 offset_in_extent = em_start - em->start;
4034                 em_end = extent_map_end(em);
4035                 em_len = em_end - em_start;
4036                 emflags = em->flags;
4037                 disko = 0;
4038                 flags = 0;
4039 
4040                 /*
4041                  * bump off for our next call to get_extent
4042                  */
4043                 off = extent_map_end(em);
4044                 if (off >= max)
4045                         end = 1;
4046 
4047                 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4048                         end = 1;
4049                         flags |= FIEMAP_EXTENT_LAST;
4050                 } else if (em->block_start == EXTENT_MAP_INLINE) {
4051                         flags |= (FIEMAP_EXTENT_DATA_INLINE |
4052                                   FIEMAP_EXTENT_NOT_ALIGNED);
4053                 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4054                         flags |= (FIEMAP_EXTENT_DELALLOC |
4055                                   FIEMAP_EXTENT_UNKNOWN);
4056                 } else {
4057                         disko = em->block_start + offset_in_extent;
4058                 }
4059                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4060                         flags |= FIEMAP_EXTENT_ENCODED;
4061 
4062                 free_extent_map(em);
4063                 em = NULL;
4064                 if ((em_start >= last) || em_len == (u64)-1 ||
4065                    (last == (u64)-1 && isize <= em_end)) {
4066                         flags |= FIEMAP_EXTENT_LAST;
4067                         end = 1;
4068                 }
4069 
4070                 /* now scan forward to see if this is really the last extent. */
4071                 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4072                                            get_extent);
4073                 if (IS_ERR(em)) {
4074                         ret = PTR_ERR(em);
4075                         goto out;
4076                 }
4077                 if (!em) {
4078                         flags |= FIEMAP_EXTENT_LAST;
4079                         end = 1;
4080                 }
4081                 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4082                                               em_len, flags);
4083                 if (ret) {
4084                         if (ret == 1)
4085                                 ret = 0;
4086                         goto out_free;
4087                 }
4088         }
4089 out_free:
4090         free_extent_map(em);
4091 out:
4092         unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4093                              &cached_state, GFP_NOFS);
4094         return ret;
4095 }
4096 
4097 static void __free_extent_buffer(struct extent_buffer *eb)
4098 {
4099         btrfs_leak_debug_del(&eb->leak_list);
4100         kmem_cache_free(extent_buffer_cache, eb);
4101 }
4102 
4103 static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
4104                                                    u64 start,
4105                                                    unsigned long len,
4106                                                    gfp_t mask)
4107 {
4108         struct extent_buffer *eb = NULL;
4109 
4110         eb = kmem_cache_zalloc(extent_buffer_cache, mask);
4111         if (eb == NULL)
4112                 return NULL;
4113         eb->start = start;
4114         eb->len = len;
4115         eb->tree = tree;
4116         eb->bflags = 0;
4117         rwlock_init(&eb->lock);
4118         atomic_set(&eb->write_locks, 0);
4119         atomic_set(&eb->read_locks, 0);
4120         atomic_set(&eb->blocking_readers, 0);
4121         atomic_set(&eb->blocking_writers, 0);
4122         atomic_set(&eb->spinning_readers, 0);
4123         atomic_set(&eb->spinning_writers, 0);
4124         eb->lock_nested = 0;
4125         init_waitqueue_head(&eb->write_lock_wq);
4126         init_waitqueue_head(&eb->read_lock_wq);
4127 
4128         btrfs_leak_debug_add(&eb->leak_list, &buffers);
4129 
4130         spin_lock_init(&eb->refs_lock);
4131         atomic_set(&eb->refs, 1);
4132         atomic_set(&eb->io_pages, 0);
4133 
4134         /*
4135          * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4136          */
4137         BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4138                 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4139         BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4140 
4141         return eb;
4142 }
4143 
4144 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4145 {
4146         unsigned long i;
4147         struct page *p;
4148         struct extent_buffer *new;
4149         unsigned long num_pages = num_extent_pages(src->start, src->len);
4150 
4151         new = __alloc_extent_buffer(NULL, src->start, src->len, GFP_ATOMIC);
4152         if (new == NULL)
4153                 return NULL;
4154 
4155         for (i = 0; i < num_pages; i++) {
4156                 p = alloc_page(GFP_ATOMIC);
4157                 BUG_ON(!p);
4158                 attach_extent_buffer_page(new, p);
4159                 WARN_ON(PageDirty(p));
4160                 SetPageUptodate(p);
4161                 new->pages[i] = p;
4162         }
4163 
4164         copy_extent_buffer(new, src, 0, 0, src->len);
4165         set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4166         set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4167 
4168         return new;
4169 }
4170 
4171 struct extent_buffer *alloc_dummy_extent_buffer(u64 start, unsigned long len)
4172 {
4173         struct extent_buffer *eb;
4174         unsigned long num_pages = num_extent_pages(0, len);
4175         unsigned long i;
4176 
4177         eb = __alloc_extent_buffer(NULL, start, len, GFP_ATOMIC);
4178         if (!eb)
4179                 return NULL;
4180 
4181         for (i = 0; i < num_pages; i++) {
4182                 eb->pages[i] = alloc_page(GFP_ATOMIC);
4183                 if (!eb->pages[i])
4184                         goto err;
4185         }
4186         set_extent_buffer_uptodate(eb);
4187         btrfs_set_header_nritems(eb, 0);
4188         set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4189 
4190         return eb;
4191 err:
4192         for (; i > 0; i--)
4193                 __free_page(eb->pages[i - 1]);
4194         __free_extent_buffer(eb);
4195         return NULL;
4196 }
4197 
4198 static int extent_buffer_under_io(struct extent_buffer *eb)
4199 {
4200         return (atomic_read(&eb->io_pages) ||
4201                 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4202                 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4203 }
4204 
4205 /*
4206  * Helper for releasing extent buffer page.
4207  */
4208 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb,
4209                                                 unsigned long start_idx)
4210 {
4211         unsigned long index;
4212         unsigned long num_pages;
4213         struct page *page;
4214         int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4215 
4216         BUG_ON(extent_buffer_under_io(eb));
4217 
4218         num_pages = num_extent_pages(eb->start, eb->len);
4219         index = start_idx + num_pages;
4220         if (start_idx >= index)
4221                 return;
4222 
4223         do {
4224                 index--;
4225                 page = extent_buffer_page(eb, index);
4226                 if (page && mapped) {
4227                         spin_lock(&page->mapping->private_lock);
4228                         /*
4229                          * We do this since we'll remove the pages after we've
4230                          * removed the eb from the radix tree, so we could race
4231                          * and have this page now attached to the new eb.  So
4232                          * only clear page_private if it's still connected to
4233                          * this eb.
4234                          */
4235                         if (PagePrivate(page) &&
4236                             page->private == (unsigned long)eb) {
4237                                 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4238                                 BUG_ON(PageDirty(page));
4239                                 BUG_ON(PageWriteback(page));
4240                                 /*
4241                                  * We need to make sure we haven't be attached
4242                                  * to a new eb.
4243                                  */
4244                                 ClearPagePrivate(page);
4245                                 set_page_private(page, 0);
4246                                 /* One for the page private */
4247                                 page_cache_release(page);
4248                         }
4249                         spin_unlock(&page->mapping->private_lock);
4250 
4251                 }
4252                 if (page) {
4253                         /* One for when we alloced the page */
4254                         page_cache_release(page);
4255                 }
4256         } while (index != start_idx);
4257 }
4258 
4259 /*
4260  * Helper for releasing the extent buffer.
4261  */
4262 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4263 {
4264         btrfs_release_extent_buffer_page(eb, 0);
4265         __free_extent_buffer(eb);
4266 }
4267 
4268 static void check_buffer_tree_ref(struct extent_buffer *eb)
4269 {
4270         int refs;
4271         /* the ref bit is tricky.  We have to make sure it is set
4272          * if we have the buffer dirty.   Otherwise the
4273          * code to free a buffer can end up dropping a dirty
4274          * page
4275          *
4276          * Once the ref bit is set, it won't go away while the
4277          * buffer is dirty or in writeback, and it also won't
4278          * go away while we have the reference count on the
4279          * eb bumped.
4280          *
4281          * We can't just set the ref bit without bumping the
4282          * ref on the eb because free_extent_buffer might
4283          * see the ref bit and try to clear it.  If this happens
4284          * free_extent_buffer might end up dropping our original
4285          * ref by mistake and freeing the page before we are able
4286          * to add one more ref.
4287          *
4288          * So bump the ref count first, then set the bit.  If someone
4289          * beat us to it, drop the ref we added.
4290          */
4291         refs = atomic_read(&eb->refs);
4292         if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4293                 return;
4294 
4295         spin_lock(&eb->refs_lock);
4296         if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4297                 atomic_inc(&eb->refs);
4298         spin_unlock(&eb->refs_lock);
4299 }
4300 
4301 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
4302 {
4303         unsigned long num_pages, i;
4304 
4305         check_buffer_tree_ref(eb);
4306 
4307         num_pages = num_extent_pages(eb->start, eb->len);
4308         for (i = 0; i < num_pages; i++) {
4309                 struct page *p = extent_buffer_page(eb, i);
4310                 mark_page_accessed(p);
4311         }
4312 }
4313 
4314 struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
4315                                           u64 start, unsigned long len)
4316 {
4317         unsigned long num_pages = num_extent_pages(start, len);
4318         unsigned long i;
4319         unsigned long index = start >> PAGE_CACHE_SHIFT;
4320         struct extent_buffer *eb;
4321         struct extent_buffer *exists = NULL;
4322         struct page *p;
4323         struct address_space *mapping = tree->mapping;
4324         int uptodate = 1;
4325         int ret;
4326 
4327         rcu_read_lock();
4328         eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4329         if (eb && atomic_inc_not_zero(&eb->refs)) {
4330                 rcu_read_unlock();
4331                 mark_extent_buffer_accessed(eb);
4332                 return eb;
4333         }
4334         rcu_read_unlock();
4335 
4336         eb = __alloc_extent_buffer(tree, start, len, GFP_NOFS);
4337         if (!eb)
4338                 return NULL;
4339 
4340         for (i = 0; i < num_pages; i++, index++) {
4341                 p = find_or_create_page(mapping, index, GFP_NOFS);
4342                 if (!p)
4343                         goto free_eb;
4344 
4345                 spin_lock(&mapping->private_lock);
4346                 if (PagePrivate(p)) {
4347                         /*
4348                          * We could have already allocated an eb for this page
4349                          * and attached one so lets see if we can get a ref on
4350                          * the existing eb, and if we can we know it's good and
4351                          * we can just return that one, else we know we can just
4352                          * overwrite page->private.
4353                          */
4354                         exists = (struct extent_buffer *)p->private;
4355                         if (atomic_inc_not_zero(&exists->refs)) {
4356                                 spin_unlock(&mapping->private_lock);
4357                                 unlock_page(p);
4358                                 page_cache_release(p);
4359                                 mark_extent_buffer_accessed(exists);
4360                                 goto free_eb;
4361                         }
4362 
4363                         /*
4364                          * Do this so attach doesn't complain and we need to
4365                          * drop the ref the old guy had.
4366                          */
4367                         ClearPagePrivate(p);
4368                         WARN_ON(PageDirty(p));
4369                         page_cache_release(p);
4370                 }
4371                 attach_extent_buffer_page(eb, p);
4372                 spin_unlock(&mapping->private_lock);
4373                 WARN_ON(PageDirty(p));
4374                 mark_page_accessed(p);
4375                 eb->pages[i] = p;
4376                 if (!PageUptodate(p))
4377                         uptodate = 0;
4378 
4379                 /*
4380                  * see below about how we avoid a nasty race with release page
4381                  * and why we unlock later
4382                  */
4383         }
4384         if (uptodate)
4385                 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4386 again:
4387         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4388         if (ret)
4389                 goto free_eb;
4390 
4391         spin_lock(&tree->buffer_lock);
4392         ret = radix_tree_insert(&tree->buffer, start >> PAGE_CACHE_SHIFT, eb);
4393         if (ret == -EEXIST) {
4394                 exists = radix_tree_lookup(&tree->buffer,
4395                                                 start >> PAGE_CACHE_SHIFT);
4396                 if (!atomic_inc_not_zero(&exists->refs)) {
4397                         spin_unlock(&tree->buffer_lock);
4398                         radix_tree_preload_end();
4399                         exists = NULL;
4400                         goto again;
4401                 }
4402                 spin_unlock(&tree->buffer_lock);
4403                 radix_tree_preload_end();
4404                 mark_extent_buffer_accessed(exists);
4405                 goto free_eb;
4406         }
4407         /* add one reference for the tree */
4408         check_buffer_tree_ref(eb);
4409         spin_unlock(&tree->buffer_lock);
4410         radix_tree_preload_end();
4411 
4412         /*
4413          * there is a race where release page may have
4414          * tried to find this extent buffer in the radix
4415          * but failed.  It will tell the VM it is safe to
4416          * reclaim the, and it will clear the page private bit.
4417          * We must make sure to set the page private bit properly
4418          * after the extent buffer is in the radix tree so
4419          * it doesn't get lost
4420          */
4421         SetPageChecked(eb->pages[0]);
4422         for (i = 1; i < num_pages; i++) {
4423                 p = extent_buffer_page(eb, i);
4424                 ClearPageChecked(p);
4425                 unlock_page(p);
4426         }
4427         unlock_page(eb->pages[0]);
4428         return eb;
4429 
4430 free_eb:
4431         for (i = 0; i < num_pages; i++) {
4432                 if (eb->pages[i])
4433                         unlock_page(eb->pages[i]);
4434         }
4435 
4436         WARN_ON(!atomic_dec_and_test(&eb->refs));
4437         btrfs_release_extent_buffer(eb);
4438         return exists;
4439 }
4440 
4441 struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
4442                                          u64 start, unsigned long len)
4443 {
4444         struct extent_buffer *eb;
4445 
4446         rcu_read_lock();
4447         eb = radix_tree_lookup(&tree->buffer, start >> PAGE_CACHE_SHIFT);
4448         if (eb && atomic_inc_not_zero(&eb->refs)) {
4449                 rcu_read_unlock();
4450                 mark_extent_buffer_accessed(eb);
4451                 return eb;
4452         }
4453         rcu_read_unlock();
4454 
4455         return NULL;
4456 }
4457 
4458 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4459 {
4460         struct extent_buffer *eb =
4461                         container_of(head, struct extent_buffer, rcu_head);
4462 
4463         __free_extent_buffer(eb);
4464 }
4465 
4466 /* Expects to have eb->eb_lock already held */
4467 static int release_extent_buffer(struct extent_buffer *eb)
4468 {
4469         WARN_ON(atomic_read(&eb->refs) == 0);
4470         if (atomic_dec_and_test(&eb->refs)) {
4471                 if (test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags)) {
4472                         spin_unlock(&eb->refs_lock);
4473                 } else {
4474                         struct extent_io_tree *tree = eb->tree;
4475 
4476                         spin_unlock(&eb->refs_lock);
4477 
4478                         spin_lock(&tree->buffer_lock);
4479                         radix_tree_delete(&tree->buffer,
4480                                           eb->start >> PAGE_CACHE_SHIFT);
4481                         spin_unlock(&tree->buffer_lock);
4482                 }
4483 
4484                 /* Should be safe to release our pages at this point */
4485                 btrfs_release_extent_buffer_page(eb, 0);
4486                 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4487                 return 1;
4488         }
4489         spin_unlock(&eb->refs_lock);
4490 
4491         return 0;
4492 }
4493 
4494 void free_extent_buffer(struct extent_buffer *eb)
4495 {
4496         int refs;
4497         int old;
4498         if (!eb)
4499                 return;
4500 
4501         while (1) {
4502                 refs = atomic_read(&eb->refs);
4503                 if (refs <= 3)
4504                         break;
4505                 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
4506                 if (old == refs)
4507                         return;
4508         }
4509 
4510         spin_lock(&eb->refs_lock);
4511         if (atomic_read(&eb->refs) == 2 &&
4512             test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
4513                 atomic_dec(&eb->refs);
4514 
4515         if (atomic_read(&eb->refs) == 2 &&
4516             test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4517             !extent_buffer_under_io(eb) &&
4518             test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4519                 atomic_dec(&eb->refs);
4520 
4521         /*
4522          * I know this is terrible, but it's temporary until we stop tracking
4523          * the uptodate bits and such for the extent buffers.
4524          */
4525         release_extent_buffer(eb);
4526 }
4527 
4528 void free_extent_buffer_stale(struct extent_buffer *eb)
4529 {
4530         if (!eb)
4531                 return;
4532 
4533         spin_lock(&eb->refs_lock);
4534         set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4535 
4536         if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4537             test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4538                 atomic_dec(&eb->refs);
4539         release_extent_buffer(eb);
4540 }
4541 
4542 void clear_extent_buffer_dirty(struct extent_buffer *eb)
4543 {
4544         unsigned long i;
4545         unsigned long num_pages;
4546         struct page *page;
4547 
4548         num_pages = num_extent_pages(eb->start, eb->len);
4549 
4550         for (i = 0; i < num_pages; i++) {
4551                 page = extent_buffer_page(eb, i);
4552                 if (!PageDirty(page))
4553                         continue;
4554 
4555                 lock_page(page);
4556                 WARN_ON(!PagePrivate(page));
4557 
4558                 clear_page_dirty_for_io(page);
4559                 spin_lock_irq(&page->mapping->tree_lock);
4560                 if (!PageDirty(page)) {
4561                         radix_tree_tag_clear(&page->mapping->page_tree,
4562                                                 page_index(page),
4563                                                 PAGECACHE_TAG_DIRTY);
4564                 }
4565                 spin_unlock_irq(&page->mapping->tree_lock);
4566                 ClearPageError(page);
4567                 unlock_page(page);
4568         }
4569         WARN_ON(atomic_read(&eb->refs) == 0);
4570 }
4571 
4572 int set_extent_buffer_dirty(struct extent_buffer *eb)
4573 {
4574         unsigned long i;
4575         unsigned long num_pages;
4576         int was_dirty = 0;
4577 
4578         check_buffer_tree_ref(eb);
4579 
4580         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4581 
4582         num_pages = num_extent_pages(eb->start, eb->len);
4583         WARN_ON(atomic_read(&eb->refs) == 0);
4584         WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4585 
4586         for (i = 0; i < num_pages; i++)
4587                 set_page_dirty(extent_buffer_page(eb, i));
4588         return was_dirty;
4589 }
4590 
4591 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
4592 {
4593         unsigned long i;
4594         struct page *page;
4595         unsigned long num_pages;
4596 
4597         clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4598         num_pages = num_extent_pages(eb->start, eb->len);
4599         for (i = 0; i < num_pages; i++) {
4600                 page = extent_buffer_page(eb, i);
4601                 if (page)
4602                         ClearPageUptodate(page);
4603         }
4604         return 0;
4605 }
4606 
4607 int set_extent_buffer_uptodate(struct extent_buffer *eb)
4608 {
4609         unsigned long i;
4610         struct page *page;
4611         unsigned long num_pages;
4612 
4613         set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4614         num_pages = num_extent_pages(eb->start, eb->len);
4615         for (i = 0; i < num_pages; i++) {
4616                 page = extent_buffer_page(eb, i);
4617                 SetPageUptodate(page);
4618         }
4619         return 0;
4620 }
4621 
4622 int extent_buffer_uptodate(struct extent_buffer *eb)
4623 {
4624         return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4625 }
4626 
4627 int read_extent_buffer_pages(struct extent_io_tree *tree,
4628                              struct extent_buffer *eb, u64 start, int wait,
4629                              get_extent_t *get_extent, int mirror_num)
4630 {
4631         unsigned long i;
4632         unsigned long start_i;
4633         struct page *page;
4634         int err;
4635         int ret = 0;
4636         int locked_pages = 0;
4637         int all_uptodate = 1;
4638         unsigned long num_pages;
4639         unsigned long num_reads = 0;
4640         struct bio *bio = NULL;
4641         unsigned long bio_flags = 0;
4642 
4643         if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4644                 return 0;
4645 
4646         if (start) {
4647                 WARN_ON(start < eb->start);
4648                 start_i = (start >> PAGE_CACHE_SHIFT) -
4649                         (eb->start >> PAGE_CACHE_SHIFT);
4650         } else {
4651                 start_i = 0;
4652         }
4653 
4654         num_pages = num_extent_pages(eb->start, eb->len);
4655         for (i = start_i; i < num_pages; i++) {
4656                 page = extent_buffer_page(eb, i);
4657                 if (wait == WAIT_NONE) {
4658                         if (!trylock_page(page))
4659                                 goto unlock_exit;
4660                 } else {
4661                         lock_page(page);
4662                 }
4663                 locked_pages++;
4664         }
4665         /*
4666          * We need to firstly lock all pages to make sure that
4667          * the uptodate bit of our pages won't be affected by
4668          * clear_extent_buffer_uptodate().
4669          */
4670         for (i = start_i; i < num_pages; i++) {
4671                 page = eb->pages[i];
4672                 if (!PageUptodate(page)) {
4673                         num_reads++;
4674                         all_uptodate = 0;
4675                 }
4676         }
4677 
4678         if (all_uptodate) {
4679                 if (start_i == 0)
4680                         set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4681                 goto unlock_exit;
4682         }
4683 
4684         clear_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
4685         eb->read_mirror = 0;
4686         atomic_set(&eb->io_pages, num_reads);
4687         for (i = start_i; i < num_pages; i++) {
4688                 page = extent_buffer_page(eb, i);
4689                 if (!PageUptodate(page)) {
4690                         ClearPageError(page);
4691                         err = __extent_read_full_page(tree, page,
4692                                                       get_extent, &bio,
4693                                                       mirror_num, &bio_flags,
4694                                                       READ | REQ_META);
4695                         if (err)
4696                                 ret = err;
4697                 } else {
4698                         unlock_page(page);
4699                 }
4700         }
4701 
4702         if (bio) {
4703                 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
4704                                      bio_flags);
4705                 if (err)
4706                         return err;
4707         }
4708 
4709         if (ret || wait != WAIT_COMPLETE)
4710                 return ret;
4711 
4712         for (i = start_i; i < num_pages; i++) {
4713                 page = extent_buffer_page(eb, i);
4714                 wait_on_page_locked(page);
4715                 if (!PageUptodate(page))
4716                         ret = -EIO;
4717         }
4718 
4719         return ret;
4720 
4721 unlock_exit:
4722         i = start_i;
4723         while (locked_pages > 0) {
4724                 page = extent_buffer_page(eb, i);
4725                 i++;
4726                 unlock_page(page);
4727                 locked_pages--;
4728         }
4729         return ret;
4730 }
4731 
4732 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
4733                         unsigned long start,
4734                         unsigned long len)
4735 {
4736         size_t cur;
4737         size_t offset;
4738         struct page *page;
4739         char *kaddr;
4740         char *dst = (char *)dstv;
4741         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4742         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4743 
4744         WARN_ON(start > eb->len);
4745         WARN_ON(start + len > eb->start + eb->len);
4746 
4747         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4748 
4749         while (len > 0) {
4750                 page = extent_buffer_page(eb, i);
4751 
4752                 cur = min(len, (PAGE_CACHE_SIZE - offset));
4753                 kaddr = page_address(page);
4754                 memcpy(dst, kaddr + offset, cur);
4755 
4756                 dst += cur;
4757                 len -= cur;
4758                 offset = 0;
4759                 i++;
4760         }
4761 }
4762 
4763 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
4764                                unsigned long min_len, char **map,
4765                                unsigned long *map_start,
4766                                unsigned long *map_len)
4767 {
4768         size_t offset = start & (PAGE_CACHE_SIZE - 1);
4769         char *kaddr;
4770         struct page *p;
4771         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4772         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4773         unsigned long end_i = (start_offset + start + min_len - 1) >>
4774                 PAGE_CACHE_SHIFT;
4775 
4776         if (i != end_i)
4777                 return -EINVAL;
4778 
4779         if (i == 0) {
4780                 offset = start_offset;
4781                 *map_start = 0;
4782         } else {
4783                 offset = 0;
4784                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
4785         }
4786 
4787         if (start + min_len > eb->len) {
4788                 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
4789                        "wanted %lu %lu\n", (unsigned long long)eb->start,
4790                        eb->len, start, min_len);
4791                 return -EINVAL;
4792         }
4793 
4794         p = extent_buffer_page(eb, i);
4795         kaddr = page_address(p);
4796         *map = kaddr + offset;
4797         *map_len = PAGE_CACHE_SIZE - offset;
4798         return 0;
4799 }
4800 
4801 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
4802                           unsigned long start,
4803                           unsigned long len)
4804 {
4805         size_t cur;
4806         size_t offset;
4807         struct page *page;
4808         char *kaddr;
4809         char *ptr = (char *)ptrv;
4810         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4811         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4812         int ret = 0;
4813 
4814         WARN_ON(start > eb->len);
4815         WARN_ON(start + len > eb->start + eb->len);
4816 
4817         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4818 
4819         while (len > 0) {
4820                 page = extent_buffer_page(eb, i);
4821 
4822                 cur = min(len, (PAGE_CACHE_SIZE - offset));
4823 
4824                 kaddr = page_address(page);
4825                 ret = memcmp(ptr, kaddr + offset, cur);
4826                 if (ret)
4827                         break;
4828 
4829                 ptr += cur;
4830                 len -= cur;
4831                 offset = 0;
4832                 i++;
4833         }
4834         return ret;
4835 }
4836 
4837 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
4838                          unsigned long start, unsigned long len)
4839 {
4840         size_t cur;
4841         size_t offset;
4842         struct page *page;
4843         char *kaddr;
4844         char *src = (char *)srcv;
4845         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4846         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4847 
4848         WARN_ON(start > eb->len);
4849         WARN_ON(start + len > eb->start + eb->len);
4850 
4851         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4852 
4853         while (len > 0) {
4854                 page = extent_buffer_page(eb, i);
4855                 WARN_ON(!PageUptodate(page));
4856 
4857                 cur = min(len, PAGE_CACHE_SIZE - offset);
4858                 kaddr = page_address(page);
4859                 memcpy(kaddr + offset, src, cur);
4860 
4861                 src += cur;
4862                 len -= cur;
4863                 offset = 0;
4864                 i++;
4865         }
4866 }
4867 
4868 void memset_extent_buffer(struct extent_buffer *eb, char c,
4869                           unsigned long start, unsigned long len)
4870 {
4871         size_t cur;
4872         size_t offset;
4873         struct page *page;
4874         char *kaddr;
4875         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
4876         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
4877 
4878         WARN_ON(start > eb->len);
4879         WARN_ON(start + len > eb->start + eb->len);
4880 
4881         offset = (start_offset + start) & ((unsigned long)PAGE_CACHE_SIZE - 1);
4882 
4883         while (len > 0) {
4884                 page = extent_buffer_page(eb, i);
4885                 WARN_ON(!PageUptodate(page));
4886 
4887                 cur = min(len, PAGE_CACHE_SIZE - offset);
4888                 kaddr = page_address(page);
4889                 memset(kaddr + offset, c, cur);
4890 
4891                 len -= cur;
4892                 offset = 0;
4893                 i++;
4894         }
4895 }
4896 
4897 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
4898                         unsigned long dst_offset, unsigned long src_offset,
4899                         unsigned long len)
4900 {
4901         u64 dst_len = dst->len;
4902         size_t cur;
4903         size_t offset;
4904         struct page *page;
4905         char *kaddr;
4906         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4907         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
4908 
4909         WARN_ON(src->len != dst_len);
4910 
4911         offset = (start_offset + dst_offset) &
4912                 ((unsigned long)PAGE_CACHE_SIZE - 1);
4913 
4914         while (len > 0) {
4915                 page = extent_buffer_page(dst, i);
4916                 WARN_ON(!PageUptodate(page));
4917 
4918                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
4919 
4920                 kaddr = page_address(page);
4921                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4922 
4923                 src_offset += cur;
4924                 len -= cur;
4925                 offset = 0;
4926                 i++;
4927         }
4928 }
4929 
4930 static void move_pages(struct page *dst_page, struct page *src_page,
4931                        unsigned long dst_off, unsigned long src_off,
4932                        unsigned long len)
4933 {
4934         char *dst_kaddr = page_address(dst_page);
4935         if (dst_page == src_page) {
4936                 memmove(dst_kaddr + dst_off, dst_kaddr + src_off, len);
4937         } else {
4938                 char *src_kaddr = page_address(src_page);
4939                 char *p = dst_kaddr + dst_off + len;
4940                 char *s = src_kaddr + src_off + len;
4941 
4942                 while (len--)
4943                         *--p = *--s;
4944         }
4945 }
4946 
4947 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4948 {
4949         unsigned long distance = (src > dst) ? src - dst : dst - src;
4950         return distance < len;
4951 }
4952 
4953 static void copy_pages(struct page *dst_page, struct page *src_page,
4954                        unsigned long dst_off, unsigned long src_off,
4955                        unsigned long len)
4956 {
4957         char *dst_kaddr = page_address(dst_page);
4958         char *src_kaddr;
4959         int must_memmove = 0;
4960 
4961         if (dst_page != src_page) {
4962                 src_kaddr = page_address(src_page);
4963         } else {
4964                 src_kaddr = dst_kaddr;
4965                 if (areas_overlap(src_off, dst_off, len))
4966                         must_memmove = 1;
4967         }
4968 
4969         if (must_memmove)
4970                 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4971         else
4972                 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4973 }
4974 
4975 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
4976                            unsigned long src_offset, unsigned long len)
4977 {
4978         size_t cur;
4979         size_t dst_off_in_page;
4980         size_t src_off_in_page;
4981         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
4982         unsigned long dst_i;
4983         unsigned long src_i;
4984 
4985         if (src_offset + len > dst->len) {
4986                 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
4987                        "len %lu dst len %lu\n", src_offset, len, dst->len);
4988                 BUG_ON(1);
4989         }
4990         if (dst_offset + len > dst->len) {
4991                 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
4992                        "len %lu dst len %lu\n", dst_offset, len, dst->len);
4993                 BUG_ON(1);
4994         }
4995 
4996         while (len > 0) {
4997                 dst_off_in_page = (start_offset + dst_offset) &
4998                         ((unsigned long)PAGE_CACHE_SIZE - 1);
4999                 src_off_in_page = (start_offset + src_offset) &
5000                         ((unsigned long)PAGE_CACHE_SIZE - 1);
5001 
5002                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5003                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5004 
5005                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5006                                                src_off_in_page));
5007                 cur = min_t(unsigned long, cur,
5008                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5009 
5010                 copy_pages(extent_buffer_page(dst, dst_i),
5011                            extent_buffer_page(dst, src_i),
5012                            dst_off_in_page, src_off_in_page, cur);
5013 
5014                 src_offset += cur;
5015                 dst_offset += cur;
5016                 len -= cur;
5017         }
5018 }
5019 
5020 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5021                            unsigned long src_offset, unsigned long len)
5022 {
5023         size_t cur;
5024         size_t dst_off_in_page;
5025         size_t src_off_in_page;
5026         unsigned long dst_end = dst_offset + len - 1;
5027         unsigned long src_end = src_offset + len - 1;
5028         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5029         unsigned long dst_i;
5030         unsigned long src_i;
5031 
5032         if (src_offset + len > dst->len) {
5033                 printk(KERN_ERR "btrfs memmove bogus src_offset %lu move "
5034                        "len %lu len %lu\n", src_offset, len, dst->len);
5035                 BUG_ON(1);
5036         }
5037         if (dst_offset + len > dst->len) {
5038                 printk(KERN_ERR "btrfs memmove bogus dst_offset %lu move "
5039                        "len %lu len %lu\n", dst_offset, len, dst->len);
5040                 BUG_ON(1);
5041         }
5042         if (dst_offset < src_offset) {
5043                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5044                 return;
5045         }
5046         while (len > 0) {
5047                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5048                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5049 
5050                 dst_off_in_page = (start_offset + dst_end) &
5051                         ((unsigned long)PAGE_CACHE_SIZE - 1);
5052                 src_off_in_page = (start_offset + src_end) &
5053                         ((unsigned long)PAGE_CACHE_SIZE - 1);
5054 
5055                 cur = min_t(unsigned long, len, src_off_in_page + 1);
5056                 cur = min(cur, dst_off_in_page + 1);
5057                 move_pages(extent_buffer_page(dst, dst_i),
5058                            extent_buffer_page(dst, src_i),
5059                            dst_off_in_page - cur + 1,
5060                            src_off_in_page - cur + 1, cur);
5061 
5062                 dst_end -= cur;
5063                 src_end -= cur;
5064                 len -= cur;
5065         }
5066 }
5067 
5068 int try_release_extent_buffer(struct page *page)
5069 {
5070         struct extent_buffer *eb;
5071 
5072         /*
5073          * We need to make sure noboody is attaching this page to an eb right
5074          * now.
5075          */
5076         spin_lock(&page->mapping->private_lock);
5077         if (!PagePrivate(page)) {
5078                 spin_unlock(&page->mapping->private_lock);
5079                 return 1;
5080         }
5081 
5082         eb = (struct extent_buffer *)page->private;
5083         BUG_ON(!eb);
5084 
5085         /*
5086          * This is a little awful but should be ok, we need to make sure that
5087          * the eb doesn't disappear out from under us while we're looking at
5088          * this page.
5089          */
5090         spin_lock(&eb->refs_lock);
5091         if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5092                 spin_unlock(&eb->refs_lock);
5093                 spin_unlock(&page->mapping->private_lock);
5094                 return 0;
5095         }
5096         spin_unlock(&page->mapping->private_lock);
5097 
5098         /*
5099          * If tree ref isn't set then we know the ref on this eb is a real ref,
5100          * so just return, this page will likely be freed soon anyway.
5101          */
5102         if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5103                 spin_unlock(&eb->refs_lock);
5104                 return 0;
5105         }
5106 
5107         return release_extent_buffer(eb);
5108 }
5109 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp