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

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

Version: ~ [ linux-5.4-rc7 ] ~ [ linux-5.3.11 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.84 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.154 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.201 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.201 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.77 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ 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 "ctree.h"
 17 #include "btrfs_inode.h"
 18 #include "volumes.h"
 19 #include "check-integrity.h"
 20 #include "locking.h"
 21 #include "rcu-string.h"
 22 #include "backref.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 static inline bool extent_state_in_tree(const struct extent_state *state)
 29 {
 30         return !RB_EMPTY_NODE(&state->rb_node);
 31 }
 32 
 33 #ifdef CONFIG_BTRFS_DEBUG
 34 static LIST_HEAD(buffers);
 35 static LIST_HEAD(states);
 36 
 37 static DEFINE_SPINLOCK(leak_lock);
 38 
 39 static inline
 40 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
 41 {
 42         unsigned long flags;
 43 
 44         spin_lock_irqsave(&leak_lock, flags);
 45         list_add(new, head);
 46         spin_unlock_irqrestore(&leak_lock, flags);
 47 }
 48 
 49 static inline
 50 void btrfs_leak_debug_del(struct list_head *entry)
 51 {
 52         unsigned long flags;
 53 
 54         spin_lock_irqsave(&leak_lock, flags);
 55         list_del(entry);
 56         spin_unlock_irqrestore(&leak_lock, flags);
 57 }
 58 
 59 static inline
 60 void btrfs_leak_debug_check(void)
 61 {
 62         struct extent_state *state;
 63         struct extent_buffer *eb;
 64 
 65         while (!list_empty(&states)) {
 66                 state = list_entry(states.next, struct extent_state, leak_list);
 67                 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
 68                        state->start, state->end, state->state,
 69                        extent_state_in_tree(state),
 70                        atomic_read(&state->refs));
 71                 list_del(&state->leak_list);
 72                 kmem_cache_free(extent_state_cache, state);
 73         }
 74 
 75         while (!list_empty(&buffers)) {
 76                 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
 77                 printk(KERN_ERR "BTRFS: buffer leak start %llu len %lu "
 78                        "refs %d\n",
 79                        eb->start, eb->len, atomic_read(&eb->refs));
 80                 list_del(&eb->leak_list);
 81                 kmem_cache_free(extent_buffer_cache, eb);
 82         }
 83 }
 84 
 85 #define btrfs_debug_check_extent_io_range(tree, start, end)             \
 86         __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end))
 87 static inline void __btrfs_debug_check_extent_io_range(const char *caller,
 88                 struct extent_io_tree *tree, u64 start, u64 end)
 89 {
 90         struct inode *inode;
 91         u64 isize;
 92 
 93         if (!tree->mapping)
 94                 return;
 95 
 96         inode = tree->mapping->host;
 97         isize = i_size_read(inode);
 98         if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
 99                 printk_ratelimited(KERN_DEBUG
100                     "BTRFS: %s: ino %llu isize %llu odd range [%llu,%llu]\n",
101                                 caller, btrfs_ino(inode), isize, start, end);
102         }
103 }
104 #else
105 #define btrfs_leak_debug_add(new, head) do {} while (0)
106 #define btrfs_leak_debug_del(entry)     do {} while (0)
107 #define btrfs_leak_debug_check()        do {} while (0)
108 #define btrfs_debug_check_extent_io_range(c, s, e)      do {} while (0)
109 #endif
110 
111 #define BUFFER_LRU_MAX 64
112 
113 struct tree_entry {
114         u64 start;
115         u64 end;
116         struct rb_node rb_node;
117 };
118 
119 struct extent_page_data {
120         struct bio *bio;
121         struct extent_io_tree *tree;
122         get_extent_t *get_extent;
123         unsigned long bio_flags;
124 
125         /* tells writepage not to lock the state bits for this range
126          * it still does the unlocking
127          */
128         unsigned int extent_locked:1;
129 
130         /* tells the submit_bio code to use a WRITE_SYNC */
131         unsigned int sync_io:1;
132 };
133 
134 static noinline void flush_write_bio(void *data);
135 static inline struct btrfs_fs_info *
136 tree_fs_info(struct extent_io_tree *tree)
137 {
138         if (!tree->mapping)
139                 return NULL;
140         return btrfs_sb(tree->mapping->host->i_sb);
141 }
142 
143 int __init extent_io_init(void)
144 {
145         extent_state_cache = kmem_cache_create("btrfs_extent_state",
146                         sizeof(struct extent_state), 0,
147                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
148         if (!extent_state_cache)
149                 return -ENOMEM;
150 
151         extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
152                         sizeof(struct extent_buffer), 0,
153                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
154         if (!extent_buffer_cache)
155                 goto free_state_cache;
156 
157         btrfs_bioset = bioset_create(BIO_POOL_SIZE,
158                                      offsetof(struct btrfs_io_bio, bio));
159         if (!btrfs_bioset)
160                 goto free_buffer_cache;
161 
162         if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
163                 goto free_bioset;
164 
165         return 0;
166 
167 free_bioset:
168         bioset_free(btrfs_bioset);
169         btrfs_bioset = NULL;
170 
171 free_buffer_cache:
172         kmem_cache_destroy(extent_buffer_cache);
173         extent_buffer_cache = NULL;
174 
175 free_state_cache:
176         kmem_cache_destroy(extent_state_cache);
177         extent_state_cache = NULL;
178         return -ENOMEM;
179 }
180 
181 void extent_io_exit(void)
182 {
183         btrfs_leak_debug_check();
184 
185         /*
186          * Make sure all delayed rcu free are flushed before we
187          * destroy caches.
188          */
189         rcu_barrier();
190         if (extent_state_cache)
191                 kmem_cache_destroy(extent_state_cache);
192         if (extent_buffer_cache)
193                 kmem_cache_destroy(extent_buffer_cache);
194         if (btrfs_bioset)
195                 bioset_free(btrfs_bioset);
196 }
197 
198 void extent_io_tree_init(struct extent_io_tree *tree,
199                          struct address_space *mapping)
200 {
201         tree->state = RB_ROOT;
202         tree->ops = NULL;
203         tree->dirty_bytes = 0;
204         spin_lock_init(&tree->lock);
205         tree->mapping = mapping;
206 }
207 
208 static struct extent_state *alloc_extent_state(gfp_t mask)
209 {
210         struct extent_state *state;
211 
212         state = kmem_cache_alloc(extent_state_cache, mask);
213         if (!state)
214                 return state;
215         state->state = 0;
216         state->private = 0;
217         RB_CLEAR_NODE(&state->rb_node);
218         btrfs_leak_debug_add(&state->leak_list, &states);
219         atomic_set(&state->refs, 1);
220         init_waitqueue_head(&state->wq);
221         trace_alloc_extent_state(state, mask, _RET_IP_);
222         return state;
223 }
224 
225 void free_extent_state(struct extent_state *state)
226 {
227         if (!state)
228                 return;
229         if (atomic_dec_and_test(&state->refs)) {
230                 WARN_ON(extent_state_in_tree(state));
231                 btrfs_leak_debug_del(&state->leak_list);
232                 trace_free_extent_state(state, _RET_IP_);
233                 kmem_cache_free(extent_state_cache, state);
234         }
235 }
236 
237 static struct rb_node *tree_insert(struct rb_root *root,
238                                    struct rb_node *search_start,
239                                    u64 offset,
240                                    struct rb_node *node,
241                                    struct rb_node ***p_in,
242                                    struct rb_node **parent_in)
243 {
244         struct rb_node **p;
245         struct rb_node *parent = NULL;
246         struct tree_entry *entry;
247 
248         if (p_in && parent_in) {
249                 p = *p_in;
250                 parent = *parent_in;
251                 goto do_insert;
252         }
253 
254         p = search_start ? &search_start : &root->rb_node;
255         while (*p) {
256                 parent = *p;
257                 entry = rb_entry(parent, struct tree_entry, rb_node);
258 
259                 if (offset < entry->start)
260                         p = &(*p)->rb_left;
261                 else if (offset > entry->end)
262                         p = &(*p)->rb_right;
263                 else
264                         return parent;
265         }
266 
267 do_insert:
268         rb_link_node(node, parent, p);
269         rb_insert_color(node, root);
270         return NULL;
271 }
272 
273 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
274                                       struct rb_node **prev_ret,
275                                       struct rb_node **next_ret,
276                                       struct rb_node ***p_ret,
277                                       struct rb_node **parent_ret)
278 {
279         struct rb_root *root = &tree->state;
280         struct rb_node **n = &root->rb_node;
281         struct rb_node *prev = NULL;
282         struct rb_node *orig_prev = NULL;
283         struct tree_entry *entry;
284         struct tree_entry *prev_entry = NULL;
285 
286         while (*n) {
287                 prev = *n;
288                 entry = rb_entry(prev, struct tree_entry, rb_node);
289                 prev_entry = entry;
290 
291                 if (offset < entry->start)
292                         n = &(*n)->rb_left;
293                 else if (offset > entry->end)
294                         n = &(*n)->rb_right;
295                 else
296                         return *n;
297         }
298 
299         if (p_ret)
300                 *p_ret = n;
301         if (parent_ret)
302                 *parent_ret = prev;
303 
304         if (prev_ret) {
305                 orig_prev = prev;
306                 while (prev && offset > prev_entry->end) {
307                         prev = rb_next(prev);
308                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
309                 }
310                 *prev_ret = prev;
311                 prev = orig_prev;
312         }
313 
314         if (next_ret) {
315                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
316                 while (prev && offset < prev_entry->start) {
317                         prev = rb_prev(prev);
318                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
319                 }
320                 *next_ret = prev;
321         }
322         return NULL;
323 }
324 
325 static inline struct rb_node *
326 tree_search_for_insert(struct extent_io_tree *tree,
327                        u64 offset,
328                        struct rb_node ***p_ret,
329                        struct rb_node **parent_ret)
330 {
331         struct rb_node *prev = NULL;
332         struct rb_node *ret;
333 
334         ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
335         if (!ret)
336                 return prev;
337         return ret;
338 }
339 
340 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
341                                           u64 offset)
342 {
343         return tree_search_for_insert(tree, offset, NULL, NULL);
344 }
345 
346 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
347                      struct extent_state *other)
348 {
349         if (tree->ops && tree->ops->merge_extent_hook)
350                 tree->ops->merge_extent_hook(tree->mapping->host, new,
351                                              other);
352 }
353 
354 /*
355  * utility function to look for merge candidates inside a given range.
356  * Any extents with matching state are merged together into a single
357  * extent in the tree.  Extents with EXTENT_IO in their state field
358  * are not merged because the end_io handlers need to be able to do
359  * operations on them without sleeping (or doing allocations/splits).
360  *
361  * This should be called with the tree lock held.
362  */
363 static void merge_state(struct extent_io_tree *tree,
364                         struct extent_state *state)
365 {
366         struct extent_state *other;
367         struct rb_node *other_node;
368 
369         if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
370                 return;
371 
372         other_node = rb_prev(&state->rb_node);
373         if (other_node) {
374                 other = rb_entry(other_node, struct extent_state, rb_node);
375                 if (other->end == state->start - 1 &&
376                     other->state == state->state) {
377                         merge_cb(tree, state, other);
378                         state->start = other->start;
379                         rb_erase(&other->rb_node, &tree->state);
380                         RB_CLEAR_NODE(&other->rb_node);
381                         free_extent_state(other);
382                 }
383         }
384         other_node = rb_next(&state->rb_node);
385         if (other_node) {
386                 other = rb_entry(other_node, struct extent_state, rb_node);
387                 if (other->start == state->end + 1 &&
388                     other->state == state->state) {
389                         merge_cb(tree, state, other);
390                         state->end = other->end;
391                         rb_erase(&other->rb_node, &tree->state);
392                         RB_CLEAR_NODE(&other->rb_node);
393                         free_extent_state(other);
394                 }
395         }
396 }
397 
398 static void set_state_cb(struct extent_io_tree *tree,
399                          struct extent_state *state, unsigned *bits)
400 {
401         if (tree->ops && tree->ops->set_bit_hook)
402                 tree->ops->set_bit_hook(tree->mapping->host, state, bits);
403 }
404 
405 static void clear_state_cb(struct extent_io_tree *tree,
406                            struct extent_state *state, unsigned *bits)
407 {
408         if (tree->ops && tree->ops->clear_bit_hook)
409                 tree->ops->clear_bit_hook(tree->mapping->host, state, bits);
410 }
411 
412 static void set_state_bits(struct extent_io_tree *tree,
413                            struct extent_state *state, unsigned *bits);
414 
415 /*
416  * insert an extent_state struct into the tree.  'bits' are set on the
417  * struct before it is inserted.
418  *
419  * This may return -EEXIST if the extent is already there, in which case the
420  * state struct is freed.
421  *
422  * The tree lock is not taken internally.  This is a utility function and
423  * probably isn't what you want to call (see set/clear_extent_bit).
424  */
425 static int insert_state(struct extent_io_tree *tree,
426                         struct extent_state *state, u64 start, u64 end,
427                         struct rb_node ***p,
428                         struct rb_node **parent,
429                         unsigned *bits)
430 {
431         struct rb_node *node;
432 
433         if (end < start)
434                 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
435                        end, start);
436         state->start = start;
437         state->end = end;
438 
439         set_state_bits(tree, state, bits);
440 
441         node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
442         if (node) {
443                 struct extent_state *found;
444                 found = rb_entry(node, struct extent_state, rb_node);
445                 printk(KERN_ERR "BTRFS: found node %llu %llu on insert of "
446                        "%llu %llu\n",
447                        found->start, found->end, start, end);
448                 return -EEXIST;
449         }
450         merge_state(tree, state);
451         return 0;
452 }
453 
454 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
455                      u64 split)
456 {
457         if (tree->ops && tree->ops->split_extent_hook)
458                 tree->ops->split_extent_hook(tree->mapping->host, orig, split);
459 }
460 
461 /*
462  * split a given extent state struct in two, inserting the preallocated
463  * struct 'prealloc' as the newly created second half.  'split' indicates an
464  * offset inside 'orig' where it should be split.
465  *
466  * Before calling,
467  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
468  * are two extent state structs in the tree:
469  * prealloc: [orig->start, split - 1]
470  * orig: [ split, orig->end ]
471  *
472  * The tree locks are not taken by this function. They need to be held
473  * by the caller.
474  */
475 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
476                        struct extent_state *prealloc, u64 split)
477 {
478         struct rb_node *node;
479 
480         split_cb(tree, orig, split);
481 
482         prealloc->start = orig->start;
483         prealloc->end = split - 1;
484         prealloc->state = orig->state;
485         orig->start = split;
486 
487         node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
488                            &prealloc->rb_node, NULL, NULL);
489         if (node) {
490                 free_extent_state(prealloc);
491                 return -EEXIST;
492         }
493         return 0;
494 }
495 
496 static struct extent_state *next_state(struct extent_state *state)
497 {
498         struct rb_node *next = rb_next(&state->rb_node);
499         if (next)
500                 return rb_entry(next, struct extent_state, rb_node);
501         else
502                 return NULL;
503 }
504 
505 /*
506  * utility function to clear some bits in an extent state struct.
507  * it will optionally wake up any one waiting on this state (wake == 1).
508  *
509  * If no bits are set on the state struct after clearing things, the
510  * struct is freed and removed from the tree
511  */
512 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
513                                             struct extent_state *state,
514                                             unsigned *bits, int wake)
515 {
516         struct extent_state *next;
517         unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
518 
519         if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
520                 u64 range = state->end - state->start + 1;
521                 WARN_ON(range > tree->dirty_bytes);
522                 tree->dirty_bytes -= range;
523         }
524         clear_state_cb(tree, state, bits);
525         state->state &= ~bits_to_clear;
526         if (wake)
527                 wake_up(&state->wq);
528         if (state->state == 0) {
529                 next = next_state(state);
530                 if (extent_state_in_tree(state)) {
531                         rb_erase(&state->rb_node, &tree->state);
532                         RB_CLEAR_NODE(&state->rb_node);
533                         free_extent_state(state);
534                 } else {
535                         WARN_ON(1);
536                 }
537         } else {
538                 merge_state(tree, state);
539                 next = next_state(state);
540         }
541         return next;
542 }
543 
544 static struct extent_state *
545 alloc_extent_state_atomic(struct extent_state *prealloc)
546 {
547         if (!prealloc)
548                 prealloc = alloc_extent_state(GFP_ATOMIC);
549 
550         return prealloc;
551 }
552 
553 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
554 {
555         btrfs_panic(tree_fs_info(tree), err, "Locking error: "
556                     "Extent tree was modified by another "
557                     "thread while locked.");
558 }
559 
560 /*
561  * clear some bits on a range in the tree.  This may require splitting
562  * or inserting elements in the tree, so the gfp mask is used to
563  * indicate which allocations or sleeping are allowed.
564  *
565  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
566  * the given range from the tree regardless of state (ie for truncate).
567  *
568  * the range [start, end] is inclusive.
569  *
570  * This takes the tree lock, and returns 0 on success and < 0 on error.
571  */
572 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
573                      unsigned bits, int wake, int delete,
574                      struct extent_state **cached_state,
575                      gfp_t mask)
576 {
577         struct extent_state *state;
578         struct extent_state *cached;
579         struct extent_state *prealloc = NULL;
580         struct rb_node *node;
581         u64 last_end;
582         int err;
583         int clear = 0;
584 
585         btrfs_debug_check_extent_io_range(tree, start, end);
586 
587         if (bits & EXTENT_DELALLOC)
588                 bits |= EXTENT_NORESERVE;
589 
590         if (delete)
591                 bits |= ~EXTENT_CTLBITS;
592         bits |= EXTENT_FIRST_DELALLOC;
593 
594         if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
595                 clear = 1;
596 again:
597         if (!prealloc && (mask & __GFP_WAIT)) {
598                 /*
599                  * Don't care for allocation failure here because we might end
600                  * up not needing the pre-allocated extent state at all, which
601                  * is the case if we only have in the tree extent states that
602                  * cover our input range and don't cover too any other range.
603                  * If we end up needing a new extent state we allocate it later.
604                  */
605                 prealloc = alloc_extent_state(mask);
606         }
607 
608         spin_lock(&tree->lock);
609         if (cached_state) {
610                 cached = *cached_state;
611 
612                 if (clear) {
613                         *cached_state = NULL;
614                         cached_state = NULL;
615                 }
616 
617                 if (cached && extent_state_in_tree(cached) &&
618                     cached->start <= start && cached->end > start) {
619                         if (clear)
620                                 atomic_dec(&cached->refs);
621                         state = cached;
622                         goto hit_next;
623                 }
624                 if (clear)
625                         free_extent_state(cached);
626         }
627         /*
628          * this search will find the extents that end after
629          * our range starts
630          */
631         node = tree_search(tree, start);
632         if (!node)
633                 goto out;
634         state = rb_entry(node, struct extent_state, rb_node);
635 hit_next:
636         if (state->start > end)
637                 goto out;
638         WARN_ON(state->end < start);
639         last_end = state->end;
640 
641         /* the state doesn't have the wanted bits, go ahead */
642         if (!(state->state & bits)) {
643                 state = next_state(state);
644                 goto next;
645         }
646 
647         /*
648          *     | ---- desired range ---- |
649          *  | state | or
650          *  | ------------- state -------------- |
651          *
652          * We need to split the extent we found, and may flip
653          * bits on second half.
654          *
655          * If the extent we found extends past our range, we
656          * just split and search again.  It'll get split again
657          * the next time though.
658          *
659          * If the extent we found is inside our range, we clear
660          * the desired bit on it.
661          */
662 
663         if (state->start < start) {
664                 prealloc = alloc_extent_state_atomic(prealloc);
665                 BUG_ON(!prealloc);
666                 err = split_state(tree, state, prealloc, start);
667                 if (err)
668                         extent_io_tree_panic(tree, err);
669 
670                 prealloc = NULL;
671                 if (err)
672                         goto out;
673                 if (state->end <= end) {
674                         state = clear_state_bit(tree, state, &bits, wake);
675                         goto next;
676                 }
677                 goto search_again;
678         }
679         /*
680          * | ---- desired range ---- |
681          *                        | state |
682          * We need to split the extent, and clear the bit
683          * on the first half
684          */
685         if (state->start <= end && state->end > end) {
686                 prealloc = alloc_extent_state_atomic(prealloc);
687                 BUG_ON(!prealloc);
688                 err = split_state(tree, state, prealloc, end + 1);
689                 if (err)
690                         extent_io_tree_panic(tree, err);
691 
692                 if (wake)
693                         wake_up(&state->wq);
694 
695                 clear_state_bit(tree, prealloc, &bits, wake);
696 
697                 prealloc = NULL;
698                 goto out;
699         }
700 
701         state = clear_state_bit(tree, state, &bits, wake);
702 next:
703         if (last_end == (u64)-1)
704                 goto out;
705         start = last_end + 1;
706         if (start <= end && state && !need_resched())
707                 goto hit_next;
708         goto search_again;
709 
710 out:
711         spin_unlock(&tree->lock);
712         if (prealloc)
713                 free_extent_state(prealloc);
714 
715         return 0;
716 
717 search_again:
718         if (start > end)
719                 goto out;
720         spin_unlock(&tree->lock);
721         if (mask & __GFP_WAIT)
722                 cond_resched();
723         goto again;
724 }
725 
726 static void wait_on_state(struct extent_io_tree *tree,
727                           struct extent_state *state)
728                 __releases(tree->lock)
729                 __acquires(tree->lock)
730 {
731         DEFINE_WAIT(wait);
732         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
733         spin_unlock(&tree->lock);
734         schedule();
735         spin_lock(&tree->lock);
736         finish_wait(&state->wq, &wait);
737 }
738 
739 /*
740  * waits for one or more bits to clear on a range in the state tree.
741  * The range [start, end] is inclusive.
742  * The tree lock is taken by this function
743  */
744 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
745                             unsigned long bits)
746 {
747         struct extent_state *state;
748         struct rb_node *node;
749 
750         btrfs_debug_check_extent_io_range(tree, start, end);
751 
752         spin_lock(&tree->lock);
753 again:
754         while (1) {
755                 /*
756                  * this search will find all the extents that end after
757                  * our range starts
758                  */
759                 node = tree_search(tree, start);
760 process_node:
761                 if (!node)
762                         break;
763 
764                 state = rb_entry(node, struct extent_state, rb_node);
765 
766                 if (state->start > end)
767                         goto out;
768 
769                 if (state->state & bits) {
770                         start = state->start;
771                         atomic_inc(&state->refs);
772                         wait_on_state(tree, state);
773                         free_extent_state(state);
774                         goto again;
775                 }
776                 start = state->end + 1;
777 
778                 if (start > end)
779                         break;
780 
781                 if (!cond_resched_lock(&tree->lock)) {
782                         node = rb_next(node);
783                         goto process_node;
784                 }
785         }
786 out:
787         spin_unlock(&tree->lock);
788 }
789 
790 static void set_state_bits(struct extent_io_tree *tree,
791                            struct extent_state *state,
792                            unsigned *bits)
793 {
794         unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
795 
796         set_state_cb(tree, state, bits);
797         if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
798                 u64 range = state->end - state->start + 1;
799                 tree->dirty_bytes += range;
800         }
801         state->state |= bits_to_set;
802 }
803 
804 static void cache_state_if_flags(struct extent_state *state,
805                                  struct extent_state **cached_ptr,
806                                  unsigned flags)
807 {
808         if (cached_ptr && !(*cached_ptr)) {
809                 if (!flags || (state->state & flags)) {
810                         *cached_ptr = state;
811                         atomic_inc(&state->refs);
812                 }
813         }
814 }
815 
816 static void cache_state(struct extent_state *state,
817                         struct extent_state **cached_ptr)
818 {
819         return cache_state_if_flags(state, cached_ptr,
820                                     EXTENT_IOBITS | EXTENT_BOUNDARY);
821 }
822 
823 /*
824  * set some bits on a range in the tree.  This may require allocations or
825  * sleeping, so the gfp mask is used to indicate what is allowed.
826  *
827  * If any of the exclusive bits are set, this will fail with -EEXIST if some
828  * part of the range already has the desired bits set.  The start of the
829  * existing range is returned in failed_start in this case.
830  *
831  * [start, end] is inclusive This takes the tree lock.
832  */
833 
834 static int __must_check
835 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
836                  unsigned bits, unsigned exclusive_bits,
837                  u64 *failed_start, struct extent_state **cached_state,
838                  gfp_t mask)
839 {
840         struct extent_state *state;
841         struct extent_state *prealloc = NULL;
842         struct rb_node *node;
843         struct rb_node **p;
844         struct rb_node *parent;
845         int err = 0;
846         u64 last_start;
847         u64 last_end;
848 
849         btrfs_debug_check_extent_io_range(tree, start, end);
850 
851         bits |= EXTENT_FIRST_DELALLOC;
852 again:
853         if (!prealloc && (mask & __GFP_WAIT)) {
854                 prealloc = alloc_extent_state(mask);
855                 BUG_ON(!prealloc);
856         }
857 
858         spin_lock(&tree->lock);
859         if (cached_state && *cached_state) {
860                 state = *cached_state;
861                 if (state->start <= start && state->end > start &&
862                     extent_state_in_tree(state)) {
863                         node = &state->rb_node;
864                         goto hit_next;
865                 }
866         }
867         /*
868          * this search will find all the extents that end after
869          * our range starts.
870          */
871         node = tree_search_for_insert(tree, start, &p, &parent);
872         if (!node) {
873                 prealloc = alloc_extent_state_atomic(prealloc);
874                 BUG_ON(!prealloc);
875                 err = insert_state(tree, prealloc, start, end,
876                                    &p, &parent, &bits);
877                 if (err)
878                         extent_io_tree_panic(tree, err);
879 
880                 cache_state(prealloc, cached_state);
881                 prealloc = NULL;
882                 goto out;
883         }
884         state = rb_entry(node, struct extent_state, rb_node);
885 hit_next:
886         last_start = state->start;
887         last_end = state->end;
888 
889         /*
890          * | ---- desired range ---- |
891          * | state |
892          *
893          * Just lock what we found and keep going
894          */
895         if (state->start == start && state->end <= end) {
896                 if (state->state & exclusive_bits) {
897                         *failed_start = state->start;
898                         err = -EEXIST;
899                         goto out;
900                 }
901 
902                 set_state_bits(tree, state, &bits);
903                 cache_state(state, cached_state);
904                 merge_state(tree, state);
905                 if (last_end == (u64)-1)
906                         goto out;
907                 start = last_end + 1;
908                 state = next_state(state);
909                 if (start < end && state && state->start == start &&
910                     !need_resched())
911                         goto hit_next;
912                 goto search_again;
913         }
914 
915         /*
916          *     | ---- desired range ---- |
917          * | state |
918          *   or
919          * | ------------- state -------------- |
920          *
921          * We need to split the extent we found, and may flip bits on
922          * second half.
923          *
924          * If the extent we found extends past our
925          * range, we just split and search again.  It'll get split
926          * again the next time though.
927          *
928          * If the extent we found is inside our range, we set the
929          * desired bit on it.
930          */
931         if (state->start < start) {
932                 if (state->state & exclusive_bits) {
933                         *failed_start = start;
934                         err = -EEXIST;
935                         goto out;
936                 }
937 
938                 prealloc = alloc_extent_state_atomic(prealloc);
939                 BUG_ON(!prealloc);
940                 err = split_state(tree, state, prealloc, start);
941                 if (err)
942                         extent_io_tree_panic(tree, err);
943 
944                 prealloc = NULL;
945                 if (err)
946                         goto out;
947                 if (state->end <= end) {
948                         set_state_bits(tree, state, &bits);
949                         cache_state(state, cached_state);
950                         merge_state(tree, state);
951                         if (last_end == (u64)-1)
952                                 goto out;
953                         start = last_end + 1;
954                         state = next_state(state);
955                         if (start < end && state && state->start == start &&
956                             !need_resched())
957                                 goto hit_next;
958                 }
959                 goto search_again;
960         }
961         /*
962          * | ---- desired range ---- |
963          *     | state | or               | state |
964          *
965          * There's a hole, we need to insert something in it and
966          * ignore the extent we found.
967          */
968         if (state->start > start) {
969                 u64 this_end;
970                 if (end < last_start)
971                         this_end = end;
972                 else
973                         this_end = last_start - 1;
974 
975                 prealloc = alloc_extent_state_atomic(prealloc);
976                 BUG_ON(!prealloc);
977 
978                 /*
979                  * Avoid to free 'prealloc' if it can be merged with
980                  * the later extent.
981                  */
982                 err = insert_state(tree, prealloc, start, this_end,
983                                    NULL, NULL, &bits);
984                 if (err)
985                         extent_io_tree_panic(tree, err);
986 
987                 cache_state(prealloc, cached_state);
988                 prealloc = NULL;
989                 start = this_end + 1;
990                 goto search_again;
991         }
992         /*
993          * | ---- desired range ---- |
994          *                        | state |
995          * We need to split the extent, and set the bit
996          * on the first half
997          */
998         if (state->start <= end && state->end > end) {
999                 if (state->state & exclusive_bits) {
1000                         *failed_start = start;
1001                         err = -EEXIST;
1002                         goto out;
1003                 }
1004 
1005                 prealloc = alloc_extent_state_atomic(prealloc);
1006                 BUG_ON(!prealloc);
1007                 err = split_state(tree, state, prealloc, end + 1);
1008                 if (err)
1009                         extent_io_tree_panic(tree, err);
1010 
1011                 set_state_bits(tree, prealloc, &bits);
1012                 cache_state(prealloc, cached_state);
1013                 merge_state(tree, prealloc);
1014                 prealloc = NULL;
1015                 goto out;
1016         }
1017 
1018         goto search_again;
1019 
1020 out:
1021         spin_unlock(&tree->lock);
1022         if (prealloc)
1023                 free_extent_state(prealloc);
1024 
1025         return err;
1026 
1027 search_again:
1028         if (start > end)
1029                 goto out;
1030         spin_unlock(&tree->lock);
1031         if (mask & __GFP_WAIT)
1032                 cond_resched();
1033         goto again;
1034 }
1035 
1036 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1037                    unsigned bits, u64 * failed_start,
1038                    struct extent_state **cached_state, gfp_t mask)
1039 {
1040         return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1041                                 cached_state, mask);
1042 }
1043 
1044 
1045 /**
1046  * convert_extent_bit - convert all bits in a given range from one bit to
1047  *                      another
1048  * @tree:       the io tree to search
1049  * @start:      the start offset in bytes
1050  * @end:        the end offset in bytes (inclusive)
1051  * @bits:       the bits to set in this range
1052  * @clear_bits: the bits to clear in this range
1053  * @cached_state:       state that we're going to cache
1054  * @mask:       the allocation mask
1055  *
1056  * This will go through and set bits for the given range.  If any states exist
1057  * already in this range they are set with the given bit and cleared of the
1058  * clear_bits.  This is only meant to be used by things that are mergeable, ie
1059  * converting from say DELALLOC to DIRTY.  This is not meant to be used with
1060  * boundary bits like LOCK.
1061  */
1062 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1063                        unsigned bits, unsigned clear_bits,
1064                        struct extent_state **cached_state, gfp_t mask)
1065 {
1066         struct extent_state *state;
1067         struct extent_state *prealloc = NULL;
1068         struct rb_node *node;
1069         struct rb_node **p;
1070         struct rb_node *parent;
1071         int err = 0;
1072         u64 last_start;
1073         u64 last_end;
1074         bool first_iteration = true;
1075 
1076         btrfs_debug_check_extent_io_range(tree, start, end);
1077 
1078 again:
1079         if (!prealloc && (mask & __GFP_WAIT)) {
1080                 /*
1081                  * Best effort, don't worry if extent state allocation fails
1082                  * here for the first iteration. We might have a cached state
1083                  * that matches exactly the target range, in which case no
1084                  * extent state allocations are needed. We'll only know this
1085                  * after locking the tree.
1086                  */
1087                 prealloc = alloc_extent_state(mask);
1088                 if (!prealloc && !first_iteration)
1089                         return -ENOMEM;
1090         }
1091 
1092         spin_lock(&tree->lock);
1093         if (cached_state && *cached_state) {
1094                 state = *cached_state;
1095                 if (state->start <= start && state->end > start &&
1096                     extent_state_in_tree(state)) {
1097                         node = &state->rb_node;
1098                         goto hit_next;
1099                 }
1100         }
1101 
1102         /*
1103          * this search will find all the extents that end after
1104          * our range starts.
1105          */
1106         node = tree_search_for_insert(tree, start, &p, &parent);
1107         if (!node) {
1108                 prealloc = alloc_extent_state_atomic(prealloc);
1109                 if (!prealloc) {
1110                         err = -ENOMEM;
1111                         goto out;
1112                 }
1113                 err = insert_state(tree, prealloc, start, end,
1114                                    &p, &parent, &bits);
1115                 if (err)
1116                         extent_io_tree_panic(tree, err);
1117                 cache_state(prealloc, cached_state);
1118                 prealloc = NULL;
1119                 goto out;
1120         }
1121         state = rb_entry(node, struct extent_state, rb_node);
1122 hit_next:
1123         last_start = state->start;
1124         last_end = state->end;
1125 
1126         /*
1127          * | ---- desired range ---- |
1128          * | state |
1129          *
1130          * Just lock what we found and keep going
1131          */
1132         if (state->start == start && state->end <= end) {
1133                 set_state_bits(tree, state, &bits);
1134                 cache_state(state, cached_state);
1135                 state = clear_state_bit(tree, state, &clear_bits, 0);
1136                 if (last_end == (u64)-1)
1137                         goto out;
1138                 start = last_end + 1;
1139                 if (start < end && state && state->start == start &&
1140                     !need_resched())
1141                         goto hit_next;
1142                 goto search_again;
1143         }
1144 
1145         /*
1146          *     | ---- desired range ---- |
1147          * | state |
1148          *   or
1149          * | ------------- state -------------- |
1150          *
1151          * We need to split the extent we found, and may flip bits on
1152          * second half.
1153          *
1154          * If the extent we found extends past our
1155          * range, we just split and search again.  It'll get split
1156          * again the next time though.
1157          *
1158          * If the extent we found is inside our range, we set the
1159          * desired bit on it.
1160          */
1161         if (state->start < start) {
1162                 prealloc = alloc_extent_state_atomic(prealloc);
1163                 if (!prealloc) {
1164                         err = -ENOMEM;
1165                         goto out;
1166                 }
1167                 err = split_state(tree, state, prealloc, start);
1168                 if (err)
1169                         extent_io_tree_panic(tree, err);
1170                 prealloc = NULL;
1171                 if (err)
1172                         goto out;
1173                 if (state->end <= end) {
1174                         set_state_bits(tree, state, &bits);
1175                         cache_state(state, cached_state);
1176                         state = clear_state_bit(tree, state, &clear_bits, 0);
1177                         if (last_end == (u64)-1)
1178                                 goto out;
1179                         start = last_end + 1;
1180                         if (start < end && state && state->start == start &&
1181                             !need_resched())
1182                                 goto hit_next;
1183                 }
1184                 goto search_again;
1185         }
1186         /*
1187          * | ---- desired range ---- |
1188          *     | state | or               | state |
1189          *
1190          * There's a hole, we need to insert something in it and
1191          * ignore the extent we found.
1192          */
1193         if (state->start > start) {
1194                 u64 this_end;
1195                 if (end < last_start)
1196                         this_end = end;
1197                 else
1198                         this_end = last_start - 1;
1199 
1200                 prealloc = alloc_extent_state_atomic(prealloc);
1201                 if (!prealloc) {
1202                         err = -ENOMEM;
1203                         goto out;
1204                 }
1205 
1206                 /*
1207                  * Avoid to free 'prealloc' if it can be merged with
1208                  * the later extent.
1209                  */
1210                 err = insert_state(tree, prealloc, start, this_end,
1211                                    NULL, NULL, &bits);
1212                 if (err)
1213                         extent_io_tree_panic(tree, err);
1214                 cache_state(prealloc, cached_state);
1215                 prealloc = NULL;
1216                 start = this_end + 1;
1217                 goto search_again;
1218         }
1219         /*
1220          * | ---- desired range ---- |
1221          *                        | state |
1222          * We need to split the extent, and set the bit
1223          * on the first half
1224          */
1225         if (state->start <= end && state->end > end) {
1226                 prealloc = alloc_extent_state_atomic(prealloc);
1227                 if (!prealloc) {
1228                         err = -ENOMEM;
1229                         goto out;
1230                 }
1231 
1232                 err = split_state(tree, state, prealloc, end + 1);
1233                 if (err)
1234                         extent_io_tree_panic(tree, err);
1235 
1236                 set_state_bits(tree, prealloc, &bits);
1237                 cache_state(prealloc, cached_state);
1238                 clear_state_bit(tree, prealloc, &clear_bits, 0);
1239                 prealloc = NULL;
1240                 goto out;
1241         }
1242 
1243         goto search_again;
1244 
1245 out:
1246         spin_unlock(&tree->lock);
1247         if (prealloc)
1248                 free_extent_state(prealloc);
1249 
1250         return err;
1251 
1252 search_again:
1253         if (start > end)
1254                 goto out;
1255         spin_unlock(&tree->lock);
1256         if (mask & __GFP_WAIT)
1257                 cond_resched();
1258         first_iteration = false;
1259         goto again;
1260 }
1261 
1262 /* wrappers around set/clear extent bit */
1263 int set_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1264                      gfp_t mask)
1265 {
1266         return set_extent_bit(tree, start, end, EXTENT_DIRTY, NULL,
1267                               NULL, mask);
1268 }
1269 
1270 int set_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1271                     unsigned bits, gfp_t mask)
1272 {
1273         return set_extent_bit(tree, start, end, bits, NULL,
1274                               NULL, mask);
1275 }
1276 
1277 int clear_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1278                       unsigned bits, gfp_t mask)
1279 {
1280         return clear_extent_bit(tree, start, end, bits, 0, 0, NULL, mask);
1281 }
1282 
1283 int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
1284                         struct extent_state **cached_state, gfp_t mask)
1285 {
1286         return set_extent_bit(tree, start, end,
1287                               EXTENT_DELALLOC | EXTENT_UPTODATE,
1288                               NULL, cached_state, mask);
1289 }
1290 
1291 int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
1292                       struct extent_state **cached_state, gfp_t mask)
1293 {
1294         return set_extent_bit(tree, start, end,
1295                               EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
1296                               NULL, cached_state, mask);
1297 }
1298 
1299 int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
1300                        gfp_t mask)
1301 {
1302         return clear_extent_bit(tree, start, end,
1303                                 EXTENT_DIRTY | EXTENT_DELALLOC |
1304                                 EXTENT_DO_ACCOUNTING, 0, 0, NULL, mask);
1305 }
1306 
1307 int set_extent_new(struct extent_io_tree *tree, u64 start, u64 end,
1308                      gfp_t mask)
1309 {
1310         return set_extent_bit(tree, start, end, EXTENT_NEW, NULL,
1311                               NULL, mask);
1312 }
1313 
1314 int set_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1315                         struct extent_state **cached_state, gfp_t mask)
1316 {
1317         return set_extent_bit(tree, start, end, EXTENT_UPTODATE, NULL,
1318                               cached_state, mask);
1319 }
1320 
1321 int clear_extent_uptodate(struct extent_io_tree *tree, u64 start, u64 end,
1322                           struct extent_state **cached_state, gfp_t mask)
1323 {
1324         return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0,
1325                                 cached_state, mask);
1326 }
1327 
1328 /*
1329  * either insert or lock state struct between start and end use mask to tell
1330  * us if waiting is desired.
1331  */
1332 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1333                      unsigned bits, struct extent_state **cached_state)
1334 {
1335         int err;
1336         u64 failed_start;
1337 
1338         while (1) {
1339                 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED | bits,
1340                                        EXTENT_LOCKED, &failed_start,
1341                                        cached_state, GFP_NOFS);
1342                 if (err == -EEXIST) {
1343                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1344                         start = failed_start;
1345                 } else
1346                         break;
1347                 WARN_ON(start > end);
1348         }
1349         return err;
1350 }
1351 
1352 int lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1353 {
1354         return lock_extent_bits(tree, start, end, 0, NULL);
1355 }
1356 
1357 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1358 {
1359         int err;
1360         u64 failed_start;
1361 
1362         err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1363                                &failed_start, NULL, GFP_NOFS);
1364         if (err == -EEXIST) {
1365                 if (failed_start > start)
1366                         clear_extent_bit(tree, start, failed_start - 1,
1367                                          EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1368                 return 0;
1369         }
1370         return 1;
1371 }
1372 
1373 int unlock_extent_cached(struct extent_io_tree *tree, u64 start, u64 end,
1374                          struct extent_state **cached, gfp_t mask)
1375 {
1376         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, cached,
1377                                 mask);
1378 }
1379 
1380 int unlock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1381 {
1382         return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, NULL,
1383                                 GFP_NOFS);
1384 }
1385 
1386 int extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1387 {
1388         unsigned long index = start >> PAGE_CACHE_SHIFT;
1389         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1390         struct page *page;
1391 
1392         while (index <= end_index) {
1393                 page = find_get_page(inode->i_mapping, index);
1394                 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1395                 clear_page_dirty_for_io(page);
1396                 page_cache_release(page);
1397                 index++;
1398         }
1399         return 0;
1400 }
1401 
1402 int extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1403 {
1404         unsigned long index = start >> PAGE_CACHE_SHIFT;
1405         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1406         struct page *page;
1407 
1408         while (index <= end_index) {
1409                 page = find_get_page(inode->i_mapping, index);
1410                 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1411                 __set_page_dirty_nobuffers(page);
1412                 account_page_redirty(page);
1413                 page_cache_release(page);
1414                 index++;
1415         }
1416         return 0;
1417 }
1418 
1419 /*
1420  * helper function to set both pages and extents in the tree writeback
1421  */
1422 static int set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1423 {
1424         unsigned long index = start >> PAGE_CACHE_SHIFT;
1425         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1426         struct page *page;
1427 
1428         while (index <= end_index) {
1429                 page = find_get_page(tree->mapping, index);
1430                 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1431                 set_page_writeback(page);
1432                 page_cache_release(page);
1433                 index++;
1434         }
1435         return 0;
1436 }
1437 
1438 /* find the first state struct with 'bits' set after 'start', and
1439  * return it.  tree->lock must be held.  NULL will returned if
1440  * nothing was found after 'start'
1441  */
1442 static struct extent_state *
1443 find_first_extent_bit_state(struct extent_io_tree *tree,
1444                             u64 start, unsigned bits)
1445 {
1446         struct rb_node *node;
1447         struct extent_state *state;
1448 
1449         /*
1450          * this search will find all the extents that end after
1451          * our range starts.
1452          */
1453         node = tree_search(tree, start);
1454         if (!node)
1455                 goto out;
1456 
1457         while (1) {
1458                 state = rb_entry(node, struct extent_state, rb_node);
1459                 if (state->end >= start && (state->state & bits))
1460                         return state;
1461 
1462                 node = rb_next(node);
1463                 if (!node)
1464                         break;
1465         }
1466 out:
1467         return NULL;
1468 }
1469 
1470 /*
1471  * find the first offset in the io tree with 'bits' set. zero is
1472  * returned if we find something, and *start_ret and *end_ret are
1473  * set to reflect the state struct that was found.
1474  *
1475  * If nothing was found, 1 is returned. If found something, return 0.
1476  */
1477 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1478                           u64 *start_ret, u64 *end_ret, unsigned bits,
1479                           struct extent_state **cached_state)
1480 {
1481         struct extent_state *state;
1482         struct rb_node *n;
1483         int ret = 1;
1484 
1485         spin_lock(&tree->lock);
1486         if (cached_state && *cached_state) {
1487                 state = *cached_state;
1488                 if (state->end == start - 1 && extent_state_in_tree(state)) {
1489                         n = rb_next(&state->rb_node);
1490                         while (n) {
1491                                 state = rb_entry(n, struct extent_state,
1492                                                  rb_node);
1493                                 if (state->state & bits)
1494                                         goto got_it;
1495                                 n = rb_next(n);
1496                         }
1497                         free_extent_state(*cached_state);
1498                         *cached_state = NULL;
1499                         goto out;
1500                 }
1501                 free_extent_state(*cached_state);
1502                 *cached_state = NULL;
1503         }
1504 
1505         state = find_first_extent_bit_state(tree, start, bits);
1506 got_it:
1507         if (state) {
1508                 cache_state_if_flags(state, cached_state, 0);
1509                 *start_ret = state->start;
1510                 *end_ret = state->end;
1511                 ret = 0;
1512         }
1513 out:
1514         spin_unlock(&tree->lock);
1515         return ret;
1516 }
1517 
1518 /*
1519  * find a contiguous range of bytes in the file marked as delalloc, not
1520  * more than 'max_bytes'.  start and end are used to return the range,
1521  *
1522  * 1 is returned if we find something, 0 if nothing was in the tree
1523  */
1524 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1525                                         u64 *start, u64 *end, u64 max_bytes,
1526                                         struct extent_state **cached_state)
1527 {
1528         struct rb_node *node;
1529         struct extent_state *state;
1530         u64 cur_start = *start;
1531         u64 found = 0;
1532         u64 total_bytes = 0;
1533 
1534         spin_lock(&tree->lock);
1535 
1536         /*
1537          * this search will find all the extents that end after
1538          * our range starts.
1539          */
1540         node = tree_search(tree, cur_start);
1541         if (!node) {
1542                 if (!found)
1543                         *end = (u64)-1;
1544                 goto out;
1545         }
1546 
1547         while (1) {
1548                 state = rb_entry(node, struct extent_state, rb_node);
1549                 if (found && (state->start != cur_start ||
1550                               (state->state & EXTENT_BOUNDARY))) {
1551                         goto out;
1552                 }
1553                 if (!(state->state & EXTENT_DELALLOC)) {
1554                         if (!found)
1555                                 *end = state->end;
1556                         goto out;
1557                 }
1558                 if (!found) {
1559                         *start = state->start;
1560                         *cached_state = state;
1561                         atomic_inc(&state->refs);
1562                 }
1563                 found++;
1564                 *end = state->end;
1565                 cur_start = state->end + 1;
1566                 node = rb_next(node);
1567                 total_bytes += state->end - state->start + 1;
1568                 if (total_bytes >= max_bytes)
1569                         break;
1570                 if (!node)
1571                         break;
1572         }
1573 out:
1574         spin_unlock(&tree->lock);
1575         return found;
1576 }
1577 
1578 static noinline void __unlock_for_delalloc(struct inode *inode,
1579                                            struct page *locked_page,
1580                                            u64 start, u64 end)
1581 {
1582         int ret;
1583         struct page *pages[16];
1584         unsigned long index = start >> PAGE_CACHE_SHIFT;
1585         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1586         unsigned long nr_pages = end_index - index + 1;
1587         int i;
1588 
1589         if (index == locked_page->index && end_index == index)
1590                 return;
1591 
1592         while (nr_pages > 0) {
1593                 ret = find_get_pages_contig(inode->i_mapping, index,
1594                                      min_t(unsigned long, nr_pages,
1595                                      ARRAY_SIZE(pages)), pages);
1596                 for (i = 0; i < ret; i++) {
1597                         if (pages[i] != locked_page)
1598                                 unlock_page(pages[i]);
1599                         page_cache_release(pages[i]);
1600                 }
1601                 nr_pages -= ret;
1602                 index += ret;
1603                 cond_resched();
1604         }
1605 }
1606 
1607 static noinline int lock_delalloc_pages(struct inode *inode,
1608                                         struct page *locked_page,
1609                                         u64 delalloc_start,
1610                                         u64 delalloc_end)
1611 {
1612         unsigned long index = delalloc_start >> PAGE_CACHE_SHIFT;
1613         unsigned long start_index = index;
1614         unsigned long end_index = delalloc_end >> PAGE_CACHE_SHIFT;
1615         unsigned long pages_locked = 0;
1616         struct page *pages[16];
1617         unsigned long nrpages;
1618         int ret;
1619         int i;
1620 
1621         /* the caller is responsible for locking the start index */
1622         if (index == locked_page->index && index == end_index)
1623                 return 0;
1624 
1625         /* skip the page at the start index */
1626         nrpages = end_index - index + 1;
1627         while (nrpages > 0) {
1628                 ret = find_get_pages_contig(inode->i_mapping, index,
1629                                      min_t(unsigned long,
1630                                      nrpages, ARRAY_SIZE(pages)), pages);
1631                 if (ret == 0) {
1632                         ret = -EAGAIN;
1633                         goto done;
1634                 }
1635                 /* now we have an array of pages, lock them all */
1636                 for (i = 0; i < ret; i++) {
1637                         /*
1638                          * the caller is taking responsibility for
1639                          * locked_page
1640                          */
1641                         if (pages[i] != locked_page) {
1642                                 lock_page(pages[i]);
1643                                 if (!PageDirty(pages[i]) ||
1644                                     pages[i]->mapping != inode->i_mapping) {
1645                                         ret = -EAGAIN;
1646                                         unlock_page(pages[i]);
1647                                         page_cache_release(pages[i]);
1648                                         goto done;
1649                                 }
1650                         }
1651                         page_cache_release(pages[i]);
1652                         pages_locked++;
1653                 }
1654                 nrpages -= ret;
1655                 index += ret;
1656                 cond_resched();
1657         }
1658         ret = 0;
1659 done:
1660         if (ret && pages_locked) {
1661                 __unlock_for_delalloc(inode, locked_page,
1662                               delalloc_start,
1663                               ((u64)(start_index + pages_locked - 1)) <<
1664                               PAGE_CACHE_SHIFT);
1665         }
1666         return ret;
1667 }
1668 
1669 /*
1670  * find a contiguous range of bytes in the file marked as delalloc, not
1671  * more than 'max_bytes'.  start and end are used to return the range,
1672  *
1673  * 1 is returned if we find something, 0 if nothing was in the tree
1674  */
1675 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1676                                     struct extent_io_tree *tree,
1677                                     struct page *locked_page, u64 *start,
1678                                     u64 *end, u64 max_bytes)
1679 {
1680         u64 delalloc_start;
1681         u64 delalloc_end;
1682         u64 found;
1683         struct extent_state *cached_state = NULL;
1684         int ret;
1685         int loops = 0;
1686 
1687 again:
1688         /* step one, find a bunch of delalloc bytes starting at start */
1689         delalloc_start = *start;
1690         delalloc_end = 0;
1691         found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1692                                     max_bytes, &cached_state);
1693         if (!found || delalloc_end <= *start) {
1694                 *start = delalloc_start;
1695                 *end = delalloc_end;
1696                 free_extent_state(cached_state);
1697                 return 0;
1698         }
1699 
1700         /*
1701          * start comes from the offset of locked_page.  We have to lock
1702          * pages in order, so we can't process delalloc bytes before
1703          * locked_page
1704          */
1705         if (delalloc_start < *start)
1706                 delalloc_start = *start;
1707 
1708         /*
1709          * make sure to limit the number of pages we try to lock down
1710          */
1711         if (delalloc_end + 1 - delalloc_start > max_bytes)
1712                 delalloc_end = delalloc_start + max_bytes - 1;
1713 
1714         /* step two, lock all the pages after the page that has start */
1715         ret = lock_delalloc_pages(inode, locked_page,
1716                                   delalloc_start, delalloc_end);
1717         if (ret == -EAGAIN) {
1718                 /* some of the pages are gone, lets avoid looping by
1719                  * shortening the size of the delalloc range we're searching
1720                  */
1721                 free_extent_state(cached_state);
1722                 cached_state = NULL;
1723                 if (!loops) {
1724                         max_bytes = PAGE_CACHE_SIZE;
1725                         loops = 1;
1726                         goto again;
1727                 } else {
1728                         found = 0;
1729                         goto out_failed;
1730                 }
1731         }
1732         BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1733 
1734         /* step three, lock the state bits for the whole range */
1735         lock_extent_bits(tree, delalloc_start, delalloc_end, 0, &cached_state);
1736 
1737         /* then test to make sure it is all still delalloc */
1738         ret = test_range_bit(tree, delalloc_start, delalloc_end,
1739                              EXTENT_DELALLOC, 1, cached_state);
1740         if (!ret) {
1741                 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1742                                      &cached_state, GFP_NOFS);
1743                 __unlock_for_delalloc(inode, locked_page,
1744                               delalloc_start, delalloc_end);
1745                 cond_resched();
1746                 goto again;
1747         }
1748         free_extent_state(cached_state);
1749         *start = delalloc_start;
1750         *end = delalloc_end;
1751 out_failed:
1752         return found;
1753 }
1754 
1755 int extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1756                                  struct page *locked_page,
1757                                  unsigned clear_bits,
1758                                  unsigned long page_ops)
1759 {
1760         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1761         int ret;
1762         struct page *pages[16];
1763         unsigned long index = start >> PAGE_CACHE_SHIFT;
1764         unsigned long end_index = end >> PAGE_CACHE_SHIFT;
1765         unsigned long nr_pages = end_index - index + 1;
1766         int i;
1767 
1768         clear_extent_bit(tree, start, end, clear_bits, 1, 0, NULL, GFP_NOFS);
1769         if (page_ops == 0)
1770                 return 0;
1771 
1772         if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1773                 mapping_set_error(inode->i_mapping, -EIO);
1774 
1775         while (nr_pages > 0) {
1776                 ret = find_get_pages_contig(inode->i_mapping, index,
1777                                      min_t(unsigned long,
1778                                      nr_pages, ARRAY_SIZE(pages)), pages);
1779                 for (i = 0; i < ret; i++) {
1780 
1781                         if (page_ops & PAGE_SET_PRIVATE2)
1782                                 SetPagePrivate2(pages[i]);
1783 
1784                         if (pages[i] == locked_page) {
1785                                 page_cache_release(pages[i]);
1786                                 continue;
1787                         }
1788                         if (page_ops & PAGE_CLEAR_DIRTY)
1789                                 clear_page_dirty_for_io(pages[i]);
1790                         if (page_ops & PAGE_SET_WRITEBACK)
1791                                 set_page_writeback(pages[i]);
1792                         if (page_ops & PAGE_SET_ERROR)
1793                                 SetPageError(pages[i]);
1794                         if (page_ops & PAGE_END_WRITEBACK)
1795                                 end_page_writeback(pages[i]);
1796                         if (page_ops & PAGE_UNLOCK)
1797                                 unlock_page(pages[i]);
1798                         page_cache_release(pages[i]);
1799                 }
1800                 nr_pages -= ret;
1801                 index += ret;
1802                 cond_resched();
1803         }
1804         return 0;
1805 }
1806 
1807 /*
1808  * count the number of bytes in the tree that have a given bit(s)
1809  * set.  This can be fairly slow, except for EXTENT_DIRTY which is
1810  * cached.  The total number found is returned.
1811  */
1812 u64 count_range_bits(struct extent_io_tree *tree,
1813                      u64 *start, u64 search_end, u64 max_bytes,
1814                      unsigned bits, int contig)
1815 {
1816         struct rb_node *node;
1817         struct extent_state *state;
1818         u64 cur_start = *start;
1819         u64 total_bytes = 0;
1820         u64 last = 0;
1821         int found = 0;
1822 
1823         if (WARN_ON(search_end <= cur_start))
1824                 return 0;
1825 
1826         spin_lock(&tree->lock);
1827         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1828                 total_bytes = tree->dirty_bytes;
1829                 goto out;
1830         }
1831         /*
1832          * this search will find all the extents that end after
1833          * our range starts.
1834          */
1835         node = tree_search(tree, cur_start);
1836         if (!node)
1837                 goto out;
1838 
1839         while (1) {
1840                 state = rb_entry(node, struct extent_state, rb_node);
1841                 if (state->start > search_end)
1842                         break;
1843                 if (contig && found && state->start > last + 1)
1844                         break;
1845                 if (state->end >= cur_start && (state->state & bits) == bits) {
1846                         total_bytes += min(search_end, state->end) + 1 -
1847                                        max(cur_start, state->start);
1848                         if (total_bytes >= max_bytes)
1849                                 break;
1850                         if (!found) {
1851                                 *start = max(cur_start, state->start);
1852                                 found = 1;
1853                         }
1854                         last = state->end;
1855                 } else if (contig && found) {
1856                         break;
1857                 }
1858                 node = rb_next(node);
1859                 if (!node)
1860                         break;
1861         }
1862 out:
1863         spin_unlock(&tree->lock);
1864         return total_bytes;
1865 }
1866 
1867 /*
1868  * set the private field for a given byte offset in the tree.  If there isn't
1869  * an extent_state there already, this does nothing.
1870  */
1871 static int set_state_private(struct extent_io_tree *tree, u64 start, u64 private)
1872 {
1873         struct rb_node *node;
1874         struct extent_state *state;
1875         int ret = 0;
1876 
1877         spin_lock(&tree->lock);
1878         /*
1879          * this search will find all the extents that end after
1880          * our range starts.
1881          */
1882         node = tree_search(tree, start);
1883         if (!node) {
1884                 ret = -ENOENT;
1885                 goto out;
1886         }
1887         state = rb_entry(node, struct extent_state, rb_node);
1888         if (state->start != start) {
1889                 ret = -ENOENT;
1890                 goto out;
1891         }
1892         state->private = private;
1893 out:
1894         spin_unlock(&tree->lock);
1895         return ret;
1896 }
1897 
1898 int get_state_private(struct extent_io_tree *tree, u64 start, u64 *private)
1899 {
1900         struct rb_node *node;
1901         struct extent_state *state;
1902         int ret = 0;
1903 
1904         spin_lock(&tree->lock);
1905         /*
1906          * this search will find all the extents that end after
1907          * our range starts.
1908          */
1909         node = tree_search(tree, start);
1910         if (!node) {
1911                 ret = -ENOENT;
1912                 goto out;
1913         }
1914         state = rb_entry(node, struct extent_state, rb_node);
1915         if (state->start != start) {
1916                 ret = -ENOENT;
1917                 goto out;
1918         }
1919         *private = state->private;
1920 out:
1921         spin_unlock(&tree->lock);
1922         return ret;
1923 }
1924 
1925 /*
1926  * searches a range in the state tree for a given mask.
1927  * If 'filled' == 1, this returns 1 only if every extent in the tree
1928  * has the bits set.  Otherwise, 1 is returned if any bit in the
1929  * range is found set.
1930  */
1931 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1932                    unsigned bits, int filled, struct extent_state *cached)
1933 {
1934         struct extent_state *state = NULL;
1935         struct rb_node *node;
1936         int bitset = 0;
1937 
1938         spin_lock(&tree->lock);
1939         if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1940             cached->end > start)
1941                 node = &cached->rb_node;
1942         else
1943                 node = tree_search(tree, start);
1944         while (node && start <= end) {
1945                 state = rb_entry(node, struct extent_state, rb_node);
1946 
1947                 if (filled && state->start > start) {
1948                         bitset = 0;
1949                         break;
1950                 }
1951 
1952                 if (state->start > end)
1953                         break;
1954 
1955                 if (state->state & bits) {
1956                         bitset = 1;
1957                         if (!filled)
1958                                 break;
1959                 } else if (filled) {
1960                         bitset = 0;
1961                         break;
1962                 }
1963 
1964                 if (state->end == (u64)-1)
1965                         break;
1966 
1967                 start = state->end + 1;
1968                 if (start > end)
1969                         break;
1970                 node = rb_next(node);
1971                 if (!node) {
1972                         if (filled)
1973                                 bitset = 0;
1974                         break;
1975                 }
1976         }
1977         spin_unlock(&tree->lock);
1978         return bitset;
1979 }
1980 
1981 /*
1982  * helper function to set a given page up to date if all the
1983  * extents in the tree for that page are up to date
1984  */
1985 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1986 {
1987         u64 start = page_offset(page);
1988         u64 end = start + PAGE_CACHE_SIZE - 1;
1989         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1990                 SetPageUptodate(page);
1991 }
1992 
1993 int free_io_failure(struct inode *inode, struct io_failure_record *rec)
1994 {
1995         int ret;
1996         int err = 0;
1997         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1998 
1999         set_state_private(failure_tree, rec->start, 0);
2000         ret = clear_extent_bits(failure_tree, rec->start,
2001                                 rec->start + rec->len - 1,
2002                                 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2003         if (ret)
2004                 err = ret;
2005 
2006         ret = clear_extent_bits(&BTRFS_I(inode)->io_tree, rec->start,
2007                                 rec->start + rec->len - 1,
2008                                 EXTENT_DAMAGED, GFP_NOFS);
2009         if (ret && !err)
2010                 err = ret;
2011 
2012         kfree(rec);
2013         return err;
2014 }
2015 
2016 /*
2017  * this bypasses the standard btrfs submit functions deliberately, as
2018  * the standard behavior is to write all copies in a raid setup. here we only
2019  * want to write the one bad copy. so we do the mapping for ourselves and issue
2020  * submit_bio directly.
2021  * to avoid any synchronization issues, wait for the data after writing, which
2022  * actually prevents the read that triggered the error from finishing.
2023  * currently, there can be no more than two copies of every data bit. thus,
2024  * exactly one rewrite is required.
2025  */
2026 int repair_io_failure(struct inode *inode, u64 start, u64 length, u64 logical,
2027                       struct page *page, unsigned int pg_offset, int mirror_num)
2028 {
2029         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2030         struct bio *bio;
2031         struct btrfs_device *dev;
2032         u64 map_length = 0;
2033         u64 sector;
2034         struct btrfs_bio *bbio = NULL;
2035         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
2036         int ret;
2037 
2038         ASSERT(!(fs_info->sb->s_flags & MS_RDONLY));
2039         BUG_ON(!mirror_num);
2040 
2041         /* we can't repair anything in raid56 yet */
2042         if (btrfs_is_parity_mirror(map_tree, logical, length, mirror_num))
2043                 return 0;
2044 
2045         bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2046         if (!bio)
2047                 return -EIO;
2048         bio->bi_iter.bi_size = 0;
2049         map_length = length;
2050 
2051         ret = btrfs_map_block(fs_info, WRITE, logical,
2052                               &map_length, &bbio, mirror_num);
2053         if (ret) {
2054                 bio_put(bio);
2055                 return -EIO;
2056         }
2057         BUG_ON(mirror_num != bbio->mirror_num);
2058         sector = bbio->stripes[mirror_num-1].physical >> 9;
2059         bio->bi_iter.bi_sector = sector;
2060         dev = bbio->stripes[mirror_num-1].dev;
2061         btrfs_put_bbio(bbio);
2062         if (!dev || !dev->bdev || !dev->writeable) {
2063                 bio_put(bio);
2064                 return -EIO;
2065         }
2066         bio->bi_bdev = dev->bdev;
2067         bio_add_page(bio, page, length, pg_offset);
2068 
2069         if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) {
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
2077                                   "BTRFS: read error corrected: ino %llu off %llu (dev %s sector %llu)\n",
2078                                   btrfs_ino(inode), start,
2079                                   rcu_str_deref(dev->name), sector);
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         if (root->fs_info->sb->s_flags & MS_RDONLY)
2092                 return -EROFS;
2093 
2094         for (i = 0; i < num_pages; i++) {
2095                 struct page *p = eb->pages[i];
2096 
2097                 ret = repair_io_failure(root->fs_info->btree_inode, start,
2098                                         PAGE_CACHE_SIZE, start, p,
2099                                         start - page_offset(p), mirror_num);
2100                 if (ret)
2101                         break;
2102                 start += PAGE_CACHE_SIZE;
2103         }
2104 
2105         return ret;
2106 }
2107 
2108 /*
2109  * each time an IO finishes, we do a fast check in the IO failure tree
2110  * to see if we need to process or clean up an io_failure_record
2111  */
2112 int clean_io_failure(struct inode *inode, u64 start, struct page *page,
2113                      unsigned int pg_offset)
2114 {
2115         u64 private;
2116         u64 private_failure;
2117         struct io_failure_record *failrec;
2118         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2119         struct extent_state *state;
2120         int num_copies;
2121         int ret;
2122 
2123         private = 0;
2124         ret = count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
2125                                 (u64)-1, 1, EXTENT_DIRTY, 0);
2126         if (!ret)
2127                 return 0;
2128 
2129         ret = get_state_private(&BTRFS_I(inode)->io_failure_tree, start,
2130                                 &private_failure);
2131         if (ret)
2132                 return 0;
2133 
2134         failrec = (struct io_failure_record *)(unsigned long) private_failure;
2135         BUG_ON(!failrec->this_mirror);
2136 
2137         if (failrec->in_validation) {
2138                 /* there was no real error, just free the record */
2139                 pr_debug("clean_io_failure: freeing dummy error at %llu\n",
2140                          failrec->start);
2141                 goto out;
2142         }
2143         if (fs_info->sb->s_flags & MS_RDONLY)
2144                 goto out;
2145 
2146         spin_lock(&BTRFS_I(inode)->io_tree.lock);
2147         state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
2148                                             failrec->start,
2149                                             EXTENT_LOCKED);
2150         spin_unlock(&BTRFS_I(inode)->io_tree.lock);
2151 
2152         if (state && state->start <= failrec->start &&
2153             state->end >= failrec->start + failrec->len - 1) {
2154                 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2155                                               failrec->len);
2156                 if (num_copies > 1)  {
2157                         repair_io_failure(inode, start, failrec->len,
2158                                           failrec->logical, page,
2159                                           pg_offset, failrec->failed_mirror);
2160                 }
2161         }
2162 
2163 out:
2164         free_io_failure(inode, failrec);
2165 
2166         return 0;
2167 }
2168 
2169 /*
2170  * Can be called when
2171  * - hold extent lock
2172  * - under ordered extent
2173  * - the inode is freeing
2174  */
2175 void btrfs_free_io_failure_record(struct inode *inode, u64 start, u64 end)
2176 {
2177         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2178         struct io_failure_record *failrec;
2179         struct extent_state *state, *next;
2180 
2181         if (RB_EMPTY_ROOT(&failure_tree->state))
2182                 return;
2183 
2184         spin_lock(&failure_tree->lock);
2185         state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2186         while (state) {
2187                 if (state->start > end)
2188                         break;
2189 
2190                 ASSERT(state->end <= end);
2191 
2192                 next = next_state(state);
2193 
2194                 failrec = (struct io_failure_record *)(unsigned long)state->private;
2195                 free_extent_state(state);
2196                 kfree(failrec);
2197 
2198                 state = next;
2199         }
2200         spin_unlock(&failure_tree->lock);
2201 }
2202 
2203 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2204                                 struct io_failure_record **failrec_ret)
2205 {
2206         struct io_failure_record *failrec;
2207         u64 private;
2208         struct extent_map *em;
2209         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2210         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2211         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2212         int ret;
2213         u64 logical;
2214 
2215         ret = get_state_private(failure_tree, start, &private);
2216         if (ret) {
2217                 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2218                 if (!failrec)
2219                         return -ENOMEM;
2220 
2221                 failrec->start = start;
2222                 failrec->len = end - start + 1;
2223                 failrec->this_mirror = 0;
2224                 failrec->bio_flags = 0;
2225                 failrec->in_validation = 0;
2226 
2227                 read_lock(&em_tree->lock);
2228                 em = lookup_extent_mapping(em_tree, start, failrec->len);
2229                 if (!em) {
2230                         read_unlock(&em_tree->lock);
2231                         kfree(failrec);
2232                         return -EIO;
2233                 }
2234 
2235                 if (em->start > start || em->start + em->len <= start) {
2236                         free_extent_map(em);
2237                         em = NULL;
2238                 }
2239                 read_unlock(&em_tree->lock);
2240                 if (!em) {
2241                         kfree(failrec);
2242                         return -EIO;
2243                 }
2244 
2245                 logical = start - em->start;
2246                 logical = em->block_start + logical;
2247                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2248                         logical = em->block_start;
2249                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2250                         extent_set_compress_type(&failrec->bio_flags,
2251                                                  em->compress_type);
2252                 }
2253 
2254                 pr_debug("Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu\n",
2255                          logical, start, failrec->len);
2256 
2257                 failrec->logical = logical;
2258                 free_extent_map(em);
2259 
2260                 /* set the bits in the private failure tree */
2261                 ret = set_extent_bits(failure_tree, start, end,
2262                                         EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
2263                 if (ret >= 0)
2264                         ret = set_state_private(failure_tree, start,
2265                                                 (u64)(unsigned long)failrec);
2266                 /* set the bits in the inode's tree */
2267                 if (ret >= 0)
2268                         ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED,
2269                                                 GFP_NOFS);
2270                 if (ret < 0) {
2271                         kfree(failrec);
2272                         return ret;
2273                 }
2274         } else {
2275                 failrec = (struct io_failure_record *)(unsigned long)private;
2276                 pr_debug("Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d\n",
2277                          failrec->logical, failrec->start, failrec->len,
2278                          failrec->in_validation);
2279                 /*
2280                  * when data can be on disk more than twice, add to failrec here
2281                  * (e.g. with a list for failed_mirror) to make
2282                  * clean_io_failure() clean all those errors at once.
2283                  */
2284         }
2285 
2286         *failrec_ret = failrec;
2287 
2288         return 0;
2289 }
2290 
2291 int btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2292                            struct io_failure_record *failrec, int failed_mirror)
2293 {
2294         int num_copies;
2295 
2296         num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
2297                                       failrec->logical, failrec->len);
2298         if (num_copies == 1) {
2299                 /*
2300                  * we only have a single copy of the data, so don't bother with
2301                  * all the retry and error correction code that follows. no
2302                  * matter what the error is, it is very likely to persist.
2303                  */
2304                 pr_debug("Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
2305                          num_copies, failrec->this_mirror, failed_mirror);
2306                 return 0;
2307         }
2308 
2309         /*
2310          * there are two premises:
2311          *      a) deliver good data to the caller
2312          *      b) correct the bad sectors on disk
2313          */
2314         if (failed_bio->bi_vcnt > 1) {
2315                 /*
2316                  * to fulfill b), we need to know the exact failing sectors, as
2317                  * we don't want to rewrite any more than the failed ones. thus,
2318                  * we need separate read requests for the failed bio
2319                  *
2320                  * if the following BUG_ON triggers, our validation request got
2321                  * merged. we need separate requests for our algorithm to work.
2322                  */
2323                 BUG_ON(failrec->in_validation);
2324                 failrec->in_validation = 1;
2325                 failrec->this_mirror = failed_mirror;
2326         } else {
2327                 /*
2328                  * we're ready to fulfill a) and b) alongside. get a good copy
2329                  * of the failed sector and if we succeed, we have setup
2330                  * everything for repair_io_failure to do the rest for us.
2331                  */
2332                 if (failrec->in_validation) {
2333                         BUG_ON(failrec->this_mirror != failed_mirror);
2334                         failrec->in_validation = 0;
2335                         failrec->this_mirror = 0;
2336                 }
2337                 failrec->failed_mirror = failed_mirror;
2338                 failrec->this_mirror++;
2339                 if (failrec->this_mirror == failed_mirror)
2340                         failrec->this_mirror++;
2341         }
2342 
2343         if (failrec->this_mirror > num_copies) {
2344                 pr_debug("Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
2345                          num_copies, failrec->this_mirror, failed_mirror);
2346                 return 0;
2347         }
2348 
2349         return 1;
2350 }
2351 
2352 
2353 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2354                                     struct io_failure_record *failrec,
2355                                     struct page *page, int pg_offset, int icsum,
2356                                     bio_end_io_t *endio_func, void *data)
2357 {
2358         struct bio *bio;
2359         struct btrfs_io_bio *btrfs_failed_bio;
2360         struct btrfs_io_bio *btrfs_bio;
2361 
2362         bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
2363         if (!bio)
2364                 return NULL;
2365 
2366         bio->bi_end_io = endio_func;
2367         bio->bi_iter.bi_sector = failrec->logical >> 9;
2368         bio->bi_bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
2369         bio->bi_iter.bi_size = 0;
2370         bio->bi_private = data;
2371 
2372         btrfs_failed_bio = btrfs_io_bio(failed_bio);
2373         if (btrfs_failed_bio->csum) {
2374                 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2375                 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2376 
2377                 btrfs_bio = btrfs_io_bio(bio);
2378                 btrfs_bio->csum = btrfs_bio->csum_inline;
2379                 icsum *= csum_size;
2380                 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2381                        csum_size);
2382         }
2383 
2384         bio_add_page(bio, page, failrec->len, pg_offset);
2385 
2386         return bio;
2387 }
2388 
2389 /*
2390  * this is a generic handler for readpage errors (default
2391  * readpage_io_failed_hook). if other copies exist, read those and write back
2392  * good data to the failed position. does not investigate in remapping the
2393  * failed extent elsewhere, hoping the device will be smart enough to do this as
2394  * needed
2395  */
2396 
2397 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2398                               struct page *page, u64 start, u64 end,
2399                               int failed_mirror)
2400 {
2401         struct io_failure_record *failrec;
2402         struct inode *inode = page->mapping->host;
2403         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2404         struct bio *bio;
2405         int read_mode;
2406         int ret;
2407 
2408         BUG_ON(failed_bio->bi_rw & REQ_WRITE);
2409 
2410         ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2411         if (ret)
2412                 return ret;
2413 
2414         ret = btrfs_check_repairable(inode, failed_bio, failrec, failed_mirror);
2415         if (!ret) {
2416                 free_io_failure(inode, failrec);
2417                 return -EIO;
2418         }
2419 
2420         if (failed_bio->bi_vcnt > 1)
2421                 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
2422         else
2423                 read_mode = READ_SYNC;
2424 
2425         phy_offset >>= inode->i_sb->s_blocksize_bits;
2426         bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2427                                       start - page_offset(page),
2428                                       (int)phy_offset, failed_bio->bi_end_io,
2429                                       NULL);
2430         if (!bio) {
2431                 free_io_failure(inode, failrec);
2432                 return -EIO;
2433         }
2434 
2435         pr_debug("Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d\n",
2436                  read_mode, failrec->this_mirror, failrec->in_validation);
2437 
2438         ret = tree->ops->submit_bio_hook(inode, read_mode, bio,
2439                                          failrec->this_mirror,
2440                                          failrec->bio_flags, 0);
2441         if (ret) {
2442                 free_io_failure(inode, failrec);
2443                 bio_put(bio);
2444         }
2445 
2446         return ret;
2447 }
2448 
2449 /* lots and lots of room for performance fixes in the end_bio funcs */
2450 
2451 int end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2452 {
2453         int uptodate = (err == 0);
2454         struct extent_io_tree *tree;
2455         int ret = 0;
2456 
2457         tree = &BTRFS_I(page->mapping->host)->io_tree;
2458 
2459         if (tree->ops && tree->ops->writepage_end_io_hook) {
2460                 ret = tree->ops->writepage_end_io_hook(page, start,
2461                                                end, NULL, uptodate);
2462                 if (ret)
2463                         uptodate = 0;
2464         }
2465 
2466         if (!uptodate) {
2467                 ClearPageUptodate(page);
2468                 SetPageError(page);
2469                 ret = err < 0 ? err : -EIO;
2470                 mapping_set_error(page->mapping, ret);
2471         }
2472         return 0;
2473 }
2474 
2475 /*
2476  * after a writepage IO is done, we need to:
2477  * clear the uptodate bits on error
2478  * clear the writeback bits in the extent tree for this IO
2479  * end_page_writeback if the page has no more pending IO
2480  *
2481  * Scheduling is not allowed, so the extent state tree is expected
2482  * to have one and only one object corresponding to this IO.
2483  */
2484 static void end_bio_extent_writepage(struct bio *bio, int err)
2485 {
2486         struct bio_vec *bvec;
2487         u64 start;
2488         u64 end;
2489         int i;
2490 
2491         bio_for_each_segment_all(bvec, bio, i) {
2492                 struct page *page = bvec->bv_page;
2493 
2494                 /* We always issue full-page reads, but if some block
2495                  * in a page fails to read, blk_update_request() will
2496                  * advance bv_offset and adjust bv_len to compensate.
2497                  * Print a warning for nonzero offsets, and an error
2498                  * if they don't add up to a full page.  */
2499                 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2500                         if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2501                                 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2502                                    "partial page write in btrfs with offset %u and length %u",
2503                                         bvec->bv_offset, bvec->bv_len);
2504                         else
2505                                 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2506                                    "incomplete page write in btrfs with offset %u and "
2507                                    "length %u",
2508                                         bvec->bv_offset, bvec->bv_len);
2509                 }
2510 
2511                 start = page_offset(page);
2512                 end = start + bvec->bv_offset + bvec->bv_len - 1;
2513 
2514                 if (end_extent_writepage(page, err, start, end))
2515                         continue;
2516 
2517                 end_page_writeback(page);
2518         }
2519 
2520         bio_put(bio);
2521 }
2522 
2523 static void
2524 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2525                               int uptodate)
2526 {
2527         struct extent_state *cached = NULL;
2528         u64 end = start + len - 1;
2529 
2530         if (uptodate && tree->track_uptodate)
2531                 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2532         unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2533 }
2534 
2535 /*
2536  * after a readpage IO is done, we need to:
2537  * clear the uptodate bits on error
2538  * set the uptodate bits if things worked
2539  * set the page up to date if all extents in the tree are uptodate
2540  * clear the lock bit in the extent tree
2541  * unlock the page if there are no other extents locked for it
2542  *
2543  * Scheduling is not allowed, so the extent state tree is expected
2544  * to have one and only one object corresponding to this IO.
2545  */
2546 static void end_bio_extent_readpage(struct bio *bio, int err)
2547 {
2548         struct bio_vec *bvec;
2549         int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
2550         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2551         struct extent_io_tree *tree;
2552         u64 offset = 0;
2553         u64 start;
2554         u64 end;
2555         u64 len;
2556         u64 extent_start = 0;
2557         u64 extent_len = 0;
2558         int mirror;
2559         int ret;
2560         int i;
2561 
2562         if (err)
2563                 uptodate = 0;
2564 
2565         bio_for_each_segment_all(bvec, bio, i) {
2566                 struct page *page = bvec->bv_page;
2567                 struct inode *inode = page->mapping->host;
2568 
2569                 pr_debug("end_bio_extent_readpage: bi_sector=%llu, err=%d, "
2570                          "mirror=%u\n", (u64)bio->bi_iter.bi_sector, err,
2571                          io_bio->mirror_num);
2572                 tree = &BTRFS_I(inode)->io_tree;
2573 
2574                 /* We always issue full-page reads, but if some block
2575                  * in a page fails to read, blk_update_request() will
2576                  * advance bv_offset and adjust bv_len to compensate.
2577                  * Print a warning for nonzero offsets, and an error
2578                  * if they don't add up to a full page.  */
2579                 if (bvec->bv_offset || bvec->bv_len != PAGE_CACHE_SIZE) {
2580                         if (bvec->bv_offset + bvec->bv_len != PAGE_CACHE_SIZE)
2581                                 btrfs_err(BTRFS_I(page->mapping->host)->root->fs_info,
2582                                    "partial page read in btrfs with offset %u and length %u",
2583                                         bvec->bv_offset, bvec->bv_len);
2584                         else
2585                                 btrfs_info(BTRFS_I(page->mapping->host)->root->fs_info,
2586                                    "incomplete page read in btrfs with offset %u and "
2587                                    "length %u",
2588                                         bvec->bv_offset, bvec->bv_len);
2589                 }
2590 
2591                 start = page_offset(page);
2592                 end = start + bvec->bv_offset + bvec->bv_len - 1;
2593                 len = bvec->bv_len;
2594 
2595                 mirror = io_bio->mirror_num;
2596                 if (likely(uptodate && tree->ops &&
2597                            tree->ops->readpage_end_io_hook)) {
2598                         ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2599                                                               page, start, end,
2600                                                               mirror);
2601                         if (ret)
2602                                 uptodate = 0;
2603                         else
2604                                 clean_io_failure(inode, start, page, 0);
2605                 }
2606 
2607                 if (likely(uptodate))
2608                         goto readpage_ok;
2609 
2610                 if (tree->ops && tree->ops->readpage_io_failed_hook) {
2611                         ret = tree->ops->readpage_io_failed_hook(page, mirror);
2612                         if (!ret && !err &&
2613                             test_bit(BIO_UPTODATE, &bio->bi_flags))
2614                                 uptodate = 1;
2615                 } else {
2616                         /*
2617                          * The generic bio_readpage_error handles errors the
2618                          * following way: If possible, new read requests are
2619                          * created and submitted and will end up in
2620                          * end_bio_extent_readpage as well (if we're lucky, not
2621                          * in the !uptodate case). In that case it returns 0 and
2622                          * we just go on with the next page in our bio. If it
2623                          * can't handle the error it will return -EIO and we
2624                          * remain responsible for that page.
2625                          */
2626                         ret = bio_readpage_error(bio, offset, page, start, end,
2627                                                  mirror);
2628                         if (ret == 0) {
2629                                 uptodate =
2630                                         test_bit(BIO_UPTODATE, &bio->bi_flags);
2631                                 if (err)
2632                                         uptodate = 0;
2633                                 offset += len;
2634                                 continue;
2635                         }
2636                 }
2637 readpage_ok:
2638                 if (likely(uptodate)) {
2639                         loff_t i_size = i_size_read(inode);
2640                         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2641                         unsigned off;
2642 
2643                         /* Zero out the end if this page straddles i_size */
2644                         off = i_size & (PAGE_CACHE_SIZE-1);
2645                         if (page->index == end_index && off)
2646                                 zero_user_segment(page, off, PAGE_CACHE_SIZE);
2647                         SetPageUptodate(page);
2648                 } else {
2649                         ClearPageUptodate(page);
2650                         SetPageError(page);
2651                 }
2652                 unlock_page(page);
2653                 offset += len;
2654 
2655                 if (unlikely(!uptodate)) {
2656                         if (extent_len) {
2657                                 endio_readpage_release_extent(tree,
2658                                                               extent_start,
2659                                                               extent_len, 1);
2660                                 extent_start = 0;
2661                                 extent_len = 0;
2662                         }
2663                         endio_readpage_release_extent(tree, start,
2664                                                       end - start + 1, 0);
2665                 } else if (!extent_len) {
2666                         extent_start = start;
2667                         extent_len = end + 1 - start;
2668                 } else if (extent_start + extent_len == start) {
2669                         extent_len += end + 1 - start;
2670                 } else {
2671                         endio_readpage_release_extent(tree, extent_start,
2672                                                       extent_len, uptodate);
2673                         extent_start = start;
2674                         extent_len = end + 1 - start;
2675                 }
2676         }
2677 
2678         if (extent_len)
2679                 endio_readpage_release_extent(tree, extent_start, extent_len,
2680                                               uptodate);
2681         if (io_bio->end_io)
2682                 io_bio->end_io(io_bio, err);
2683         bio_put(bio);
2684 }
2685 
2686 /*
2687  * this allocates from the btrfs_bioset.  We're returning a bio right now
2688  * but you can call btrfs_io_bio for the appropriate container_of magic
2689  */
2690 struct bio *
2691 btrfs_bio_alloc(struct block_device *bdev, u64 first_sector, int nr_vecs,
2692                 gfp_t gfp_flags)
2693 {
2694         struct btrfs_io_bio *btrfs_bio;
2695         struct bio *bio;
2696 
2697         bio = bio_alloc_bioset(gfp_flags, nr_vecs, btrfs_bioset);
2698 
2699         if (bio == NULL && (current->flags & PF_MEMALLOC)) {
2700                 while (!bio && (nr_vecs /= 2)) {
2701                         bio = bio_alloc_bioset(gfp_flags,
2702                                                nr_vecs, btrfs_bioset);
2703                 }
2704         }
2705 
2706         if (bio) {
2707                 bio->bi_bdev = bdev;
2708                 bio->bi_iter.bi_sector = first_sector;
2709                 btrfs_bio = btrfs_io_bio(bio);
2710                 btrfs_bio->csum = NULL;
2711                 btrfs_bio->csum_allocated = NULL;
2712                 btrfs_bio->end_io = NULL;
2713         }
2714         return bio;
2715 }
2716 
2717 struct bio *btrfs_bio_clone(struct bio *bio, gfp_t gfp_mask)
2718 {
2719         struct btrfs_io_bio *btrfs_bio;
2720         struct bio *new;
2721 
2722         new = bio_clone_bioset(bio, gfp_mask, btrfs_bioset);
2723         if (new) {
2724                 btrfs_bio = btrfs_io_bio(new);
2725                 btrfs_bio->csum = NULL;
2726                 btrfs_bio->csum_allocated = NULL;
2727                 btrfs_bio->end_io = NULL;
2728         }
2729         return new;
2730 }
2731 
2732 /* this also allocates from the btrfs_bioset */
2733 struct bio *btrfs_io_bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
2734 {
2735         struct btrfs_io_bio *btrfs_bio;
2736         struct bio *bio;
2737 
2738         bio = bio_alloc_bioset(gfp_mask, nr_iovecs, btrfs_bioset);
2739         if (bio) {
2740                 btrfs_bio = btrfs_io_bio(bio);
2741                 btrfs_bio->csum = NULL;
2742                 btrfs_bio->csum_allocated = NULL;
2743                 btrfs_bio->end_io = NULL;
2744         }
2745         return bio;
2746 }
2747 
2748 
2749 static int __must_check submit_one_bio(int rw, struct bio *bio,
2750                                        int mirror_num, unsigned long bio_flags)
2751 {
2752         int ret = 0;
2753         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2754         struct page *page = bvec->bv_page;
2755         struct extent_io_tree *tree = bio->bi_private;
2756         u64 start;
2757 
2758         start = page_offset(page) + bvec->bv_offset;
2759 
2760         bio->bi_private = NULL;
2761 
2762         bio_get(bio);
2763 
2764         if (tree->ops && tree->ops->submit_bio_hook)
2765                 ret = tree->ops->submit_bio_hook(page->mapping->host, rw, bio,
2766                                            mirror_num, bio_flags, start);
2767         else
2768                 btrfsic_submit_bio(rw, bio);
2769 
2770         if (bio_flagged(bio, BIO_EOPNOTSUPP))
2771                 ret = -EOPNOTSUPP;
2772         bio_put(bio);
2773         return ret;
2774 }
2775 
2776 static int merge_bio(int rw, struct extent_io_tree *tree, struct page *page,
2777                      unsigned long offset, size_t size, struct bio *bio,
2778                      unsigned long bio_flags)
2779 {
2780         int ret = 0;
2781         if (tree->ops && tree->ops->merge_bio_hook)
2782                 ret = tree->ops->merge_bio_hook(rw, page, offset, size, bio,
2783                                                 bio_flags);
2784         BUG_ON(ret < 0);
2785         return ret;
2786 
2787 }
2788 
2789 static int submit_extent_page(int rw, struct extent_io_tree *tree,
2790                               struct page *page, sector_t sector,
2791                               size_t size, unsigned long offset,
2792                               struct block_device *bdev,
2793                               struct bio **bio_ret,
2794                               unsigned long max_pages,
2795                               bio_end_io_t end_io_func,
2796                               int mirror_num,
2797                               unsigned long prev_bio_flags,
2798                               unsigned long bio_flags,
2799                               bool force_bio_submit)
2800 {
2801         int ret = 0;
2802         struct bio *bio;
2803         int nr;
2804         int contig = 0;
2805         int this_compressed = bio_flags & EXTENT_BIO_COMPRESSED;
2806         int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2807         size_t page_size = min_t(size_t, size, PAGE_CACHE_SIZE);
2808 
2809         if (bio_ret && *bio_ret) {
2810                 bio = *bio_ret;
2811                 if (old_compressed)
2812                         contig = bio->bi_iter.bi_sector == sector;
2813                 else
2814                         contig = bio_end_sector(bio) == sector;
2815 
2816                 if (prev_bio_flags != bio_flags || !contig ||
2817                     force_bio_submit ||
2818                     merge_bio(rw, tree, page, offset, page_size, bio, bio_flags) ||
2819                     bio_add_page(bio, page, page_size, offset) < page_size) {
2820                         ret = submit_one_bio(rw, bio, mirror_num,
2821                                              prev_bio_flags);
2822                         if (ret < 0) {
2823                                 *bio_ret = NULL;
2824                                 return ret;
2825                         }
2826                         bio = NULL;
2827                 } else {
2828                         return 0;
2829                 }
2830         }
2831         if (this_compressed)
2832                 nr = BIO_MAX_PAGES;
2833         else
2834                 nr = bio_get_nr_vecs(bdev);
2835 
2836         bio = btrfs_bio_alloc(bdev, sector, nr, GFP_NOFS | __GFP_HIGH);
2837         if (!bio)
2838                 return -ENOMEM;
2839 
2840         bio_add_page(bio, page, page_size, offset);
2841         bio->bi_end_io = end_io_func;
2842         bio->bi_private = tree;
2843 
2844         if (bio_ret)
2845                 *bio_ret = bio;
2846         else
2847                 ret = submit_one_bio(rw, bio, mirror_num, bio_flags);
2848 
2849         return ret;
2850 }
2851 
2852 static void attach_extent_buffer_page(struct extent_buffer *eb,
2853                                       struct page *page)
2854 {
2855         if (!PagePrivate(page)) {
2856                 SetPagePrivate(page);
2857                 page_cache_get(page);
2858                 set_page_private(page, (unsigned long)eb);
2859         } else {
2860                 WARN_ON(page->private != (unsigned long)eb);
2861         }
2862 }
2863 
2864 void set_page_extent_mapped(struct page *page)
2865 {
2866         if (!PagePrivate(page)) {
2867                 SetPagePrivate(page);
2868                 page_cache_get(page);
2869                 set_page_private(page, EXTENT_PAGE_PRIVATE);
2870         }
2871 }
2872 
2873 static struct extent_map *
2874 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2875                  u64 start, u64 len, get_extent_t *get_extent,
2876                  struct extent_map **em_cached)
2877 {
2878         struct extent_map *em;
2879 
2880         if (em_cached && *em_cached) {
2881                 em = *em_cached;
2882                 if (extent_map_in_tree(em) && start >= em->start &&
2883                     start < extent_map_end(em)) {
2884                         atomic_inc(&em->refs);
2885                         return em;
2886                 }
2887 
2888                 free_extent_map(em);
2889                 *em_cached = NULL;
2890         }
2891 
2892         em = get_extent(inode, page, pg_offset, start, len, 0);
2893         if (em_cached && !IS_ERR_OR_NULL(em)) {
2894                 BUG_ON(*em_cached);
2895                 atomic_inc(&em->refs);
2896                 *em_cached = em;
2897         }
2898         return em;
2899 }
2900 /*
2901  * basic readpage implementation.  Locked extent state structs are inserted
2902  * into the tree that are removed when the IO is done (by the end_io
2903  * handlers)
2904  * XXX JDM: This needs looking at to ensure proper page locking
2905  */
2906 static int __do_readpage(struct extent_io_tree *tree,
2907                          struct page *page,
2908                          get_extent_t *get_extent,
2909                          struct extent_map **em_cached,
2910                          struct bio **bio, int mirror_num,
2911                          unsigned long *bio_flags, int rw,
2912                          u64 *prev_em_start)
2913 {
2914         struct inode *inode = page->mapping->host;
2915         u64 start = page_offset(page);
2916         u64 page_end = start + PAGE_CACHE_SIZE - 1;
2917         u64 end;
2918         u64 cur = start;
2919         u64 extent_offset;
2920         u64 last_byte = i_size_read(inode);
2921         u64 block_start;
2922         u64 cur_end;
2923         sector_t sector;
2924         struct extent_map *em;
2925         struct block_device *bdev;
2926         int ret;
2927         int nr = 0;
2928         int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2929         size_t pg_offset = 0;
2930         size_t iosize;
2931         size_t disk_io_size;
2932         size_t blocksize = inode->i_sb->s_blocksize;
2933         unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED;
2934 
2935         set_page_extent_mapped(page);
2936 
2937         end = page_end;
2938         if (!PageUptodate(page)) {
2939                 if (cleancache_get_page(page) == 0) {
2940                         BUG_ON(blocksize != PAGE_SIZE);
2941                         unlock_extent(tree, start, end);
2942                         goto out;
2943                 }
2944         }
2945 
2946         if (page->index == last_byte >> PAGE_CACHE_SHIFT) {
2947                 char *userpage;
2948                 size_t zero_offset = last_byte & (PAGE_CACHE_SIZE - 1);
2949 
2950                 if (zero_offset) {
2951                         iosize = PAGE_CACHE_SIZE - zero_offset;
2952                         userpage = kmap_atomic(page);
2953                         memset(userpage + zero_offset, 0, iosize);
2954                         flush_dcache_page(page);
2955                         kunmap_atomic(userpage);
2956                 }
2957         }
2958         while (cur <= end) {
2959                 unsigned long pnr = (last_byte >> PAGE_CACHE_SHIFT) + 1;
2960                 bool force_bio_submit = false;
2961 
2962                 if (cur >= last_byte) {
2963                         char *userpage;
2964                         struct extent_state *cached = NULL;
2965 
2966                         iosize = PAGE_CACHE_SIZE - pg_offset;
2967                         userpage = kmap_atomic(page);
2968                         memset(userpage + pg_offset, 0, iosize);
2969                         flush_dcache_page(page);
2970                         kunmap_atomic(userpage);
2971                         set_extent_uptodate(tree, cur, cur + iosize - 1,
2972                                             &cached, GFP_NOFS);
2973                         if (!parent_locked)
2974                                 unlock_extent_cached(tree, cur,
2975                                                      cur + iosize - 1,
2976                                                      &cached, GFP_NOFS);
2977                         break;
2978                 }
2979                 em = __get_extent_map(inode, page, pg_offset, cur,
2980                                       end - cur + 1, get_extent, em_cached);
2981                 if (IS_ERR_OR_NULL(em)) {
2982                         SetPageError(page);
2983                         if (!parent_locked)
2984                                 unlock_extent(tree, cur, end);
2985                         break;
2986                 }
2987                 extent_offset = cur - em->start;
2988                 BUG_ON(extent_map_end(em) <= cur);
2989                 BUG_ON(end < cur);
2990 
2991                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2992                         this_bio_flag |= EXTENT_BIO_COMPRESSED;
2993                         extent_set_compress_type(&this_bio_flag,
2994                                                  em->compress_type);
2995                 }
2996 
2997                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2998                 cur_end = min(extent_map_end(em) - 1, end);
2999                 iosize = ALIGN(iosize, blocksize);
3000                 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
3001                         disk_io_size = em->block_len;
3002                         sector = em->block_start >> 9;
3003                 } else {
3004                         sector = (em->block_start + extent_offset) >> 9;
3005                         disk_io_size = iosize;
3006                 }
3007                 bdev = em->bdev;
3008                 block_start = em->block_start;
3009                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3010                         block_start = EXTENT_MAP_HOLE;
3011 
3012                 /*
3013                  * If we have a file range that points to a compressed extent
3014                  * and it's followed by a consecutive file range that points to
3015                  * to the same compressed extent (possibly with a different
3016                  * offset and/or length, so it either points to the whole extent
3017                  * or only part of it), we must make sure we do not submit a
3018                  * single bio to populate the pages for the 2 ranges because
3019                  * this makes the compressed extent read zero out the pages
3020                  * belonging to the 2nd range. Imagine the following scenario:
3021                  *
3022                  *  File layout
3023                  *  [0 - 8K]                     [8K - 24K]
3024                  *    |                               |
3025                  *    |                               |
3026                  * points to extent X,         points to extent X,
3027                  * offset 4K, length of 8K     offset 0, length 16K
3028                  *
3029                  * [extent X, compressed length = 4K uncompressed length = 16K]
3030                  *
3031                  * If the bio to read the compressed extent covers both ranges,
3032                  * it will decompress extent X into the pages belonging to the
3033                  * first range and then it will stop, zeroing out the remaining
3034                  * pages that belong to the other range that points to extent X.
3035                  * So here we make sure we submit 2 bios, one for the first
3036                  * range and another one for the third range. Both will target
3037                  * the same physical extent from disk, but we can't currently
3038                  * make the compressed bio endio callback populate the pages
3039                  * for both ranges because each compressed bio is tightly
3040                  * coupled with a single extent map, and each range can have
3041                  * an extent map with a different offset value relative to the
3042                  * uncompressed data of our extent and different lengths. This
3043                  * is a corner case so we prioritize correctness over
3044                  * non-optimal behavior (submitting 2 bios for the same extent).
3045                  */
3046                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3047                     prev_em_start && *prev_em_start != (u64)-1 &&
3048                     *prev_em_start != em->orig_start)
3049                         force_bio_submit = true;
3050 
3051                 if (prev_em_start)
3052                         *prev_em_start = em->orig_start;
3053 
3054                 free_extent_map(em);
3055                 em = NULL;
3056 
3057                 /* we've found a hole, just zero and go on */
3058                 if (block_start == EXTENT_MAP_HOLE) {
3059                         char *userpage;
3060                         struct extent_state *cached = NULL;
3061 
3062                         userpage = kmap_atomic(page);
3063                         memset(userpage + pg_offset, 0, iosize);
3064                         flush_dcache_page(page);
3065                         kunmap_atomic(userpage);
3066 
3067                         set_extent_uptodate(tree, cur, cur + iosize - 1,
3068                                             &cached, GFP_NOFS);
3069                         unlock_extent_cached(tree, cur, cur + iosize - 1,
3070                                              &cached, GFP_NOFS);
3071                         cur = cur + iosize;
3072                         pg_offset += iosize;
3073                         continue;
3074                 }
3075                 /* the get_extent function already copied into the page */
3076                 if (test_range_bit(tree, cur, cur_end,
3077                                    EXTENT_UPTODATE, 1, NULL)) {
3078                         check_page_uptodate(tree, page);
3079                         if (!parent_locked)
3080                                 unlock_extent(tree, cur, cur + iosize - 1);
3081                         cur = cur + iosize;
3082                         pg_offset += iosize;
3083                         continue;
3084                 }
3085                 /* we have an inline extent but it didn't get marked up
3086                  * to date.  Error out
3087                  */
3088                 if (block_start == EXTENT_MAP_INLINE) {
3089                         SetPageError(page);
3090                         if (!parent_locked)
3091                                 unlock_extent(tree, cur, cur + iosize - 1);
3092                         cur = cur + iosize;
3093                         pg_offset += iosize;
3094                         continue;
3095                 }
3096 
3097                 pnr -= page->index;
3098                 ret = submit_extent_page(rw, tree, page,
3099                                          sector, disk_io_size, pg_offset,
3100                                          bdev, bio, pnr,
3101                                          end_bio_extent_readpage, mirror_num,
3102                                          *bio_flags,
3103                                          this_bio_flag,
3104                                          force_bio_submit);
3105                 if (!ret) {
3106                         nr++;
3107                         *bio_flags = this_bio_flag;
3108                 } else {
3109                         SetPageError(page);
3110                         if (!parent_locked)
3111                                 unlock_extent(tree, cur, cur + iosize - 1);
3112                 }
3113                 cur = cur + iosize;
3114                 pg_offset += iosize;
3115         }
3116 out:
3117         if (!nr) {
3118                 if (!PageError(page))
3119                         SetPageUptodate(page);
3120                 unlock_page(page);
3121         }
3122         return 0;
3123 }
3124 
3125 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3126                                              struct page *pages[], int nr_pages,
3127                                              u64 start, u64 end,
3128                                              get_extent_t *get_extent,
3129                                              struct extent_map **em_cached,
3130                                              struct bio **bio, int mirror_num,
3131                                              unsigned long *bio_flags, int rw,
3132                                              u64 *prev_em_start)
3133 {
3134         struct inode *inode;
3135         struct btrfs_ordered_extent *ordered;
3136         int index;
3137 
3138         inode = pages[0]->mapping->host;
3139         while (1) {
3140                 lock_extent(tree, start, end);
3141                 ordered = btrfs_lookup_ordered_range(inode, start,
3142                                                      end - start + 1);
3143                 if (!ordered)
3144                         break;
3145                 unlock_extent(tree, start, end);
3146                 btrfs_start_ordered_extent(inode, ordered, 1);
3147                 btrfs_put_ordered_extent(ordered);
3148         }
3149 
3150         for (index = 0; index < nr_pages; index++) {
3151                 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3152                               mirror_num, bio_flags, rw, prev_em_start);
3153                 page_cache_release(pages[index]);
3154         }
3155 }
3156 
3157 static void __extent_readpages(struct extent_io_tree *tree,
3158                                struct page *pages[],
3159                                int nr_pages, get_extent_t *get_extent,
3160                                struct extent_map **em_cached,
3161                                struct bio **bio, int mirror_num,
3162                                unsigned long *bio_flags, int rw,
3163                                u64 *prev_em_start)
3164 {
3165         u64 start = 0;
3166         u64 end = 0;
3167         u64 page_start;
3168         int index;
3169         int first_index = 0;
3170 
3171         for (index = 0; index < nr_pages; index++) {
3172                 page_start = page_offset(pages[index]);
3173                 if (!end) {
3174                         start = page_start;
3175                         end = start + PAGE_CACHE_SIZE - 1;
3176                         first_index = index;
3177                 } else if (end + 1 == page_start) {
3178                         end += PAGE_CACHE_SIZE;
3179                 } else {
3180                         __do_contiguous_readpages(tree, &pages[first_index],
3181                                                   index - first_index, start,
3182                                                   end, get_extent, em_cached,
3183                                                   bio, mirror_num, bio_flags,
3184                                                   rw, prev_em_start);
3185                         start = page_start;
3186                         end = start + PAGE_CACHE_SIZE - 1;
3187                         first_index = index;
3188                 }
3189         }
3190 
3191         if (end)
3192                 __do_contiguous_readpages(tree, &pages[first_index],
3193                                           index - first_index, start,
3194                                           end, get_extent, em_cached, bio,
3195                                           mirror_num, bio_flags, rw,
3196                                           prev_em_start);
3197 }
3198 
3199 static int __extent_read_full_page(struct extent_io_tree *tree,
3200                                    struct page *page,
3201                                    get_extent_t *get_extent,
3202                                    struct bio **bio, int mirror_num,
3203                                    unsigned long *bio_flags, int rw)
3204 {
3205         struct inode *inode = page->mapping->host;
3206         struct btrfs_ordered_extent *ordered;
3207         u64 start = page_offset(page);
3208         u64 end = start + PAGE_CACHE_SIZE - 1;
3209         int ret;
3210 
3211         while (1) {
3212                 lock_extent(tree, start, end);
3213                 ordered = btrfs_lookup_ordered_extent(inode, start);
3214                 if (!ordered)
3215                         break;
3216                 unlock_extent(tree, start, end);
3217                 btrfs_start_ordered_extent(inode, ordered, 1);
3218                 btrfs_put_ordered_extent(ordered);
3219         }
3220 
3221         ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3222                             bio_flags, rw, NULL);
3223         return ret;
3224 }
3225 
3226 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3227                             get_extent_t *get_extent, int mirror_num)
3228 {
3229         struct bio *bio = NULL;
3230         unsigned long bio_flags = 0;
3231         int ret;
3232 
3233         ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3234                                       &bio_flags, READ);
3235         if (bio)
3236                 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3237         return ret;
3238 }
3239 
3240 int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page,
3241                                  get_extent_t *get_extent, int mirror_num)
3242 {
3243         struct bio *bio = NULL;
3244         unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED;
3245         int ret;
3246 
3247         ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num,
3248                             &bio_flags, READ, NULL);
3249         if (bio)
3250                 ret = submit_one_bio(READ, bio, mirror_num, bio_flags);
3251         return ret;
3252 }
3253 
3254 static noinline void update_nr_written(struct page *page,
3255                                       struct writeback_control *wbc,
3256                                       unsigned long nr_written)
3257 {
3258         wbc->nr_to_write -= nr_written;
3259         if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
3260             wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
3261                 page->mapping->writeback_index = page->index + nr_written;
3262 }
3263 
3264 /*
3265  * helper for __extent_writepage, doing all of the delayed allocation setup.
3266  *
3267  * This returns 1 if our fill_delalloc function did all the work required
3268  * to write the page (copy into inline extent).  In this case the IO has
3269  * been started and the page is already unlocked.
3270  *
3271  * This returns 0 if all went well (page still locked)
3272  * This returns < 0 if there were errors (page still locked)
3273  */
3274 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3275                               struct page *page, struct writeback_control *wbc,
3276                               struct extent_page_data *epd,
3277                               u64 delalloc_start,
3278                               unsigned long *nr_written)
3279 {
3280         struct extent_io_tree *tree = epd->tree;
3281         u64 page_end = delalloc_start + PAGE_CACHE_SIZE - 1;
3282         u64 nr_delalloc;
3283         u64 delalloc_to_write = 0;
3284         u64 delalloc_end = 0;
3285         int ret;
3286         int page_started = 0;
3287 
3288         if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3289                 return 0;
3290 
3291         while (delalloc_end < page_end) {
3292                 nr_delalloc = find_lock_delalloc_range(inode, tree,
3293                                                page,
3294                                                &delalloc_start,
3295                                                &delalloc_end,
3296                                                BTRFS_MAX_EXTENT_SIZE);
3297                 if (nr_delalloc == 0) {
3298                         delalloc_start = delalloc_end + 1;
3299                         continue;
3300                 }
3301                 ret = tree->ops->fill_delalloc(inode, page,
3302                                                delalloc_start,
3303                                                delalloc_end,
3304                                                &page_started,
3305                                                nr_written);
3306                 /* File system has been set read-only */
3307                 if (ret) {
3308                         SetPageError(page);
3309                         /* fill_delalloc should be return < 0 for error
3310                          * but just in case, we use > 0 here meaning the
3311                          * IO is started, so we don't want to return > 0
3312                          * unless things are going well.
3313                          */
3314                         ret = ret < 0 ? ret : -EIO;
3315                         goto done;
3316                 }
3317                 /*
3318                  * delalloc_end is already one less than the total
3319                  * length, so we don't subtract one from
3320                  * PAGE_CACHE_SIZE
3321                  */
3322                 delalloc_to_write += (delalloc_end - delalloc_start +
3323                                       PAGE_CACHE_SIZE) >>
3324                                       PAGE_CACHE_SHIFT;
3325                 delalloc_start = delalloc_end + 1;
3326         }
3327         if (wbc->nr_to_write < delalloc_to_write) {
3328                 int thresh = 8192;
3329 
3330                 if (delalloc_to_write < thresh * 2)
3331                         thresh = delalloc_to_write;
3332                 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3333                                          thresh);
3334         }
3335 
3336         /* did the fill delalloc function already unlock and start
3337          * the IO?
3338          */
3339         if (page_started) {
3340                 /*
3341                  * we've unlocked the page, so we can't update
3342                  * the mapping's writeback index, just update
3343                  * nr_to_write.
3344                  */
3345                 wbc->nr_to_write -= *nr_written;
3346                 return 1;
3347         }
3348 
3349         ret = 0;
3350 
3351 done:
3352         return ret;
3353 }
3354 
3355 /*
3356  * helper for __extent_writepage.  This calls the writepage start hooks,
3357  * and does the loop to map the page into extents and bios.
3358  *
3359  * We return 1 if the IO is started and the page is unlocked,
3360  * 0 if all went well (page still locked)
3361  * < 0 if there were errors (page still locked)
3362  */
3363 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3364                                  struct page *page,
3365                                  struct writeback_control *wbc,
3366                                  struct extent_page_data *epd,
3367                                  loff_t i_size,
3368                                  unsigned long nr_written,
3369                                  int write_flags, int *nr_ret)
3370 {
3371         struct extent_io_tree *tree = epd->tree;
3372         u64 start = page_offset(page);
3373         u64 page_end = start + PAGE_CACHE_SIZE - 1;
3374         u64 end;
3375         u64 cur = start;
3376         u64 extent_offset;
3377         u64 block_start;
3378         u64 iosize;
3379         sector_t sector;
3380         struct extent_state *cached_state = NULL;
3381         struct extent_map *em;
3382         struct block_device *bdev;
3383         size_t pg_offset = 0;
3384         size_t blocksize;
3385         int ret = 0;
3386         int nr = 0;
3387         bool compressed;
3388 
3389         if (tree->ops && tree->ops->writepage_start_hook) {
3390                 ret = tree->ops->writepage_start_hook(page, start,
3391                                                       page_end);
3392                 if (ret) {
3393                         /* Fixup worker will requeue */
3394                         if (ret == -EBUSY)
3395                                 wbc->pages_skipped++;
3396                         else
3397                                 redirty_page_for_writepage(wbc, page);
3398 
3399                         update_nr_written(page, wbc, nr_written);
3400                         unlock_page(page);
3401                         ret = 1;
3402                         goto done_unlocked;
3403                 }
3404         }
3405 
3406         /*
3407          * we don't want to touch the inode after unlocking the page,
3408          * so we update the mapping writeback index now
3409          */
3410         update_nr_written(page, wbc, nr_written + 1);
3411 
3412         end = page_end;
3413         if (i_size <= start) {
3414                 if (tree->ops && tree->ops->writepage_end_io_hook)
3415                         tree->ops->writepage_end_io_hook(page, start,
3416                                                          page_end, NULL, 1);
3417                 goto done;
3418         }
3419 
3420         blocksize = inode->i_sb->s_blocksize;
3421 
3422         while (cur <= end) {
3423                 u64 em_end;
3424                 if (cur >= i_size) {
3425                         if (tree->ops && tree->ops->writepage_end_io_hook)
3426                                 tree->ops->writepage_end_io_hook(page, cur,
3427                                                          page_end, NULL, 1);
3428                         break;
3429                 }
3430                 em = epd->get_extent(inode, page, pg_offset, cur,
3431                                      end - cur + 1, 1);
3432                 if (IS_ERR_OR_NULL(em)) {
3433                         SetPageError(page);
3434                         ret = PTR_ERR_OR_ZERO(em);
3435                         break;
3436                 }
3437 
3438                 extent_offset = cur - em->start;
3439                 em_end = extent_map_end(em);
3440                 BUG_ON(em_end <= cur);
3441                 BUG_ON(end < cur);
3442                 iosize = min(em_end - cur, end - cur + 1);
3443                 iosize = ALIGN(iosize, blocksize);
3444                 sector = (em->block_start + extent_offset) >> 9;
3445                 bdev = em->bdev;
3446                 block_start = em->block_start;
3447                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3448                 free_extent_map(em);
3449                 em = NULL;
3450 
3451                 /*
3452                  * compressed and inline extents are written through other
3453                  * paths in the FS
3454                  */
3455                 if (compressed || block_start == EXTENT_MAP_HOLE ||
3456                     block_start == EXTENT_MAP_INLINE) {
3457                         /*
3458                          * end_io notification does not happen here for
3459                          * compressed extents
3460                          */
3461                         if (!compressed && tree->ops &&
3462                             tree->ops->writepage_end_io_hook)
3463                                 tree->ops->writepage_end_io_hook(page, cur,
3464                                                          cur + iosize - 1,
3465                                                          NULL, 1);
3466                         else if (compressed) {
3467                                 /* we don't want to end_page_writeback on
3468                                  * a compressed extent.  this happens
3469                                  * elsewhere
3470                                  */
3471                                 nr++;
3472                         }
3473 
3474                         cur += iosize;
3475                         pg_offset += iosize;
3476                         continue;
3477                 }
3478 
3479                 if (tree->ops && tree->ops->writepage_io_hook) {
3480                         ret = tree->ops->writepage_io_hook(page, cur,
3481                                                 cur + iosize - 1);
3482                 } else {
3483                         ret = 0;
3484                 }
3485                 if (ret) {
3486                         SetPageError(page);
3487                 } else {
3488                         unsigned long max_nr = (i_size >> PAGE_CACHE_SHIFT) + 1;
3489 
3490                         set_range_writeback(tree, cur, cur + iosize - 1);
3491                         if (!PageWriteback(page)) {
3492                                 btrfs_err(BTRFS_I(inode)->root->fs_info,
3493                                            "page %lu not writeback, cur %llu end %llu",
3494                                        page->index, cur, end);
3495                         }
3496 
3497                         ret = submit_extent_page(write_flags, tree, page,
3498                                                  sector, iosize, pg_offset,
3499                                                  bdev, &epd->bio, max_nr,
3500                                                  end_bio_extent_writepage,
3501                                                  0, 0, 0, false);
3502                         if (ret)
3503                                 SetPageError(page);
3504                 }
3505                 cur = cur + iosize;
3506                 pg_offset += iosize;
3507                 nr++;
3508         }
3509 done:
3510         *nr_ret = nr;
3511 
3512 done_unlocked:
3513 
3514         /* drop our reference on any cached states */
3515         free_extent_state(cached_state);
3516         return ret;
3517 }
3518 
3519 /*
3520  * the writepage semantics are similar to regular writepage.  extent
3521  * records are inserted to lock ranges in the tree, and as dirty areas
3522  * are found, they are marked writeback.  Then the lock bits are removed
3523  * and the end_io handler clears the writeback ranges
3524  */
3525 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3526                               void *data)
3527 {
3528         struct inode *inode = page->mapping->host;
3529         struct extent_page_data *epd = data;
3530         u64 start = page_offset(page);
3531         u64 page_end = start + PAGE_CACHE_SIZE - 1;
3532         int ret;
3533         int nr = 0;
3534         size_t pg_offset = 0;
3535         loff_t i_size = i_size_read(inode);
3536         unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
3537         int write_flags;
3538         unsigned long nr_written = 0;
3539 
3540         if (wbc->sync_mode == WB_SYNC_ALL)
3541                 write_flags = WRITE_SYNC;
3542         else
3543                 write_flags = WRITE;
3544 
3545         trace___extent_writepage(page, inode, wbc);
3546 
3547         WARN_ON(!PageLocked(page));
3548 
3549         ClearPageError(page);
3550 
3551         pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
3552         if (page->index > end_index ||
3553            (page->index == end_index && !pg_offset)) {
3554                 page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE);
3555                 unlock_page(page);
3556                 return 0;
3557         }
3558 
3559         if (page->index == end_index) {
3560                 char *userpage;
3561 
3562                 userpage = kmap_atomic(page);
3563                 memset(userpage + pg_offset, 0,
3564                        PAGE_CACHE_SIZE - pg_offset);
3565                 kunmap_atomic(userpage);
3566                 flush_dcache_page(page);
3567         }
3568 
3569         pg_offset = 0;
3570 
3571         set_page_extent_mapped(page);
3572 
3573         ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3574         if (ret == 1)
3575                 goto done_unlocked;
3576         if (ret)
3577                 goto done;
3578 
3579         ret = __extent_writepage_io(inode, page, wbc, epd,
3580                                     i_size, nr_written, write_flags, &nr);
3581         if (ret == 1)
3582                 goto done_unlocked;
3583 
3584 done:
3585         if (nr == 0) {
3586                 /* make sure the mapping tag for page dirty gets cleared */
3587                 set_page_writeback(page);
3588                 end_page_writeback(page);
3589         }
3590         if (PageError(page)) {
3591                 ret = ret < 0 ? ret : -EIO;
3592                 end_extent_writepage(page, ret, start, page_end);
3593         }
3594         unlock_page(page);
3595         return ret;
3596 
3597 done_unlocked:
3598         return 0;
3599 }
3600 
3601 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3602 {
3603         wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3604                        TASK_UNINTERRUPTIBLE);
3605 }
3606 
3607 static noinline_for_stack int
3608 lock_extent_buffer_for_io(struct extent_buffer *eb,
3609                           struct btrfs_fs_info *fs_info,
3610                           struct extent_page_data *epd)
3611 {
3612         unsigned long i, num_pages;
3613         int flush = 0;
3614         int ret = 0;
3615 
3616         if (!btrfs_try_tree_write_lock(eb)) {
3617                 flush = 1;
3618                 flush_write_bio(epd);
3619                 btrfs_tree_lock(eb);
3620         }
3621 
3622         if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3623                 btrfs_tree_unlock(eb);
3624                 if (!epd->sync_io)
3625                         return 0;
3626                 if (!flush) {
3627                         flush_write_bio(epd);
3628                         flush = 1;
3629                 }
3630                 while (1) {
3631                         wait_on_extent_buffer_writeback(eb);
3632                         btrfs_tree_lock(eb);
3633                         if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3634                                 break;
3635                         btrfs_tree_unlock(eb);
3636                 }
3637         }
3638 
3639         /*
3640          * We need to do this to prevent races in people who check if the eb is
3641          * under IO since we can end up having no IO bits set for a short period
3642          * of time.
3643          */
3644         spin_lock(&eb->refs_lock);
3645         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3646                 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3647                 spin_unlock(&eb->refs_lock);
3648                 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3649                 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
3650                                      -eb->len,
3651                                      fs_info->dirty_metadata_batch);
3652                 ret = 1;
3653         } else {
3654                 spin_unlock(&eb->refs_lock);
3655         }
3656 
3657         btrfs_tree_unlock(eb);
3658 
3659         if (!ret)
3660                 return ret;
3661 
3662         num_pages = num_extent_pages(eb->start, eb->len);
3663         for (i = 0; i < num_pages; i++) {
3664                 struct page *p = eb->pages[i];
3665 
3666                 if (!trylock_page(p)) {
3667                         if (!flush) {
3668                                 flush_write_bio(epd);
3669                                 flush = 1;
3670                         }
3671                         lock_page(p);
3672                 }
3673         }
3674 
3675         return ret;
3676 }
3677 
3678 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3679 {
3680         clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3681         smp_mb__after_atomic();
3682         wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3683 }
3684 
3685 static void set_btree_ioerr(struct page *page)
3686 {
3687         struct extent_buffer *eb = (struct extent_buffer *)page->private;
3688         struct btrfs_inode *btree_ino = BTRFS_I(eb->fs_info->btree_inode);
3689 
3690         SetPageError(page);
3691         if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3692                 return;
3693 
3694         /*
3695          * If writeback for a btree extent that doesn't belong to a log tree
3696          * failed, increment the counter transaction->eb_write_errors.
3697          * We do this because while the transaction is running and before it's
3698          * committing (when we call filemap_fdata[write|wait]_range against
3699          * the btree inode), we might have
3700          * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3701          * returns an error or an error happens during writeback, when we're
3702          * committing the transaction we wouldn't know about it, since the pages
3703          * can be no longer dirty nor marked anymore for writeback (if a
3704          * subsequent modification to the extent buffer didn't happen before the
3705          * transaction commit), which makes filemap_fdata[write|wait]_range not
3706          * able to find the pages tagged with SetPageError at transaction
3707          * commit time. So if this happens we must abort the transaction,
3708          * otherwise we commit a super block with btree roots that point to
3709          * btree nodes/leafs whose content on disk is invalid - either garbage
3710          * or the content of some node/leaf from a past generation that got
3711          * cowed or deleted and is no longer valid.
3712          *
3713          * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3714          * not be enough - we need to distinguish between log tree extents vs
3715          * non-log tree extents, and the next filemap_fdatawait_range() call
3716          * will catch and clear such errors in the mapping - and that call might
3717          * be from a log sync and not from a transaction commit. Also, checking
3718          * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3719          * not done and would not be reliable - the eb might have been released
3720          * from memory and reading it back again means that flag would not be
3721          * set (since it's a runtime flag, not persisted on disk).
3722          *
3723          * Using the flags below in the btree inode also makes us achieve the
3724          * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3725          * writeback for all dirty pages and before filemap_fdatawait_range()
3726          * is called, the writeback for all dirty pages had already finished
3727          * with errors - because we were not using AS_EIO/AS_ENOSPC,
3728          * filemap_fdatawait_range() would return success, as it could not know
3729          * that writeback errors happened (the pages were no longer tagged for
3730          * writeback).
3731          */
3732         switch (eb->log_index) {
3733         case -1:
3734                 set_bit(BTRFS_INODE_BTREE_ERR, &btree_ino->runtime_flags);
3735                 break;
3736         case 0:
3737                 set_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags);
3738                 break;
3739         case 1:
3740                 set_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags);
3741                 break;
3742         default:
3743                 BUG(); /* unexpected, logic error */
3744         }
3745 }
3746 
3747 static void end_bio_extent_buffer_writepage(struct bio *bio, int err)
3748 {
3749         struct bio_vec *bvec;
3750         struct extent_buffer *eb;
3751         int i, done;
3752 
3753         bio_for_each_segment_all(bvec, bio, i) {
3754                 struct page *page = bvec->bv_page;
3755 
3756                 eb = (struct extent_buffer *)page->private;
3757                 BUG_ON(!eb);
3758                 done = atomic_dec_and_test(&eb->io_pages);
3759 
3760                 if (err || test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3761                         ClearPageUptodate(page);
3762                         set_btree_ioerr(page);
3763                 }
3764 
3765                 end_page_writeback(page);
3766 
3767                 if (!done)
3768                         continue;
3769 
3770                 end_extent_buffer_writeback(eb);
3771         }
3772 
3773         bio_put(bio);
3774 }
3775 
3776 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3777                         struct btrfs_fs_info *fs_info,
3778                         struct writeback_control *wbc,
3779                         struct extent_page_data *epd)
3780 {
3781         struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3782         struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3783         u64 offset = eb->start;
3784         unsigned long i, num_pages;
3785         unsigned long bio_flags = 0;
3786         int rw = (epd->sync_io ? WRITE_SYNC : WRITE) | REQ_META;
3787         int ret = 0;
3788 
3789         clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3790         num_pages = num_extent_pages(eb->start, eb->len);
3791         atomic_set(&eb->io_pages, num_pages);
3792         if (btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID)
3793                 bio_flags = EXTENT_BIO_TREE_LOG;
3794 
3795         for (i = 0; i < num_pages; i++) {
3796                 struct page *p = eb->pages[i];
3797 
3798                 clear_page_dirty_for_io(p);
3799                 set_page_writeback(p);
3800                 ret = submit_extent_page(rw, tree, p, offset >> 9,
3801                                          PAGE_CACHE_SIZE, 0, bdev, &epd->bio,
3802                                          -1, end_bio_extent_buffer_writepage,
3803                                          0, epd->bio_flags, bio_flags, false);
3804                 epd->bio_flags = bio_flags;
3805                 if (ret) {
3806                         set_btree_ioerr(p);
3807                         end_page_writeback(p);
3808                         if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3809                                 end_extent_buffer_writeback(eb);
3810                         ret = -EIO;
3811                         break;
3812                 }
3813                 offset += PAGE_CACHE_SIZE;
3814                 update_nr_written(p, wbc, 1);
3815                 unlock_page(p);
3816         }
3817 
3818         if (unlikely(ret)) {
3819                 for (; i < num_pages; i++) {
3820                         struct page *p = eb->pages[i];
3821                         clear_page_dirty_for_io(p);
3822                         unlock_page(p);
3823                 }
3824         }
3825 
3826         return ret;
3827 }
3828 
3829 int btree_write_cache_pages(struct address_space *mapping,
3830                                    struct writeback_control *wbc)
3831 {
3832         struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3833         struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3834         struct extent_buffer *eb, *prev_eb = NULL;
3835         struct extent_page_data epd = {
3836                 .bio = NULL,
3837                 .tree = tree,
3838                 .extent_locked = 0,
3839                 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3840                 .bio_flags = 0,
3841         };
3842         int ret = 0;
3843         int done = 0;
3844         int nr_to_write_done = 0;
3845         struct pagevec pvec;
3846         int nr_pages;
3847         pgoff_t index;
3848         pgoff_t end;            /* Inclusive */
3849         int scanned = 0;
3850         int tag;
3851 
3852         pagevec_init(&pvec, 0);
3853         if (wbc->range_cyclic) {
3854                 index = mapping->writeback_index; /* Start from prev offset */
3855                 end = -1;
3856         } else {
3857                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
3858                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
3859                 scanned = 1;
3860         }
3861         if (wbc->sync_mode == WB_SYNC_ALL)
3862                 tag = PAGECACHE_TAG_TOWRITE;
3863         else
3864                 tag = PAGECACHE_TAG_DIRTY;
3865 retry:
3866         if (wbc->sync_mode == WB_SYNC_ALL)
3867                 tag_pages_for_writeback(mapping, index, end);
3868         while (!done && !nr_to_write_done && (index <= end) &&
3869                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
3870                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
3871                 unsigned i;
3872 
3873                 scanned = 1;
3874                 for (i = 0; i < nr_pages; i++) {
3875                         struct page *page = pvec.pages[i];
3876 
3877                         if (!PagePrivate(page))
3878                                 continue;
3879 
3880                         if (!wbc->range_cyclic && page->index > end) {
3881                                 done = 1;
3882                                 break;
3883                         }
3884 
3885                         spin_lock(&mapping->private_lock);
3886                         if (!PagePrivate(page)) {
3887                                 spin_unlock(&mapping->private_lock);
3888                                 continue;
3889                         }
3890 
3891                         eb = (struct extent_buffer *)page->private;
3892 
3893                         /*
3894                          * Shouldn't happen and normally this would be a BUG_ON
3895                          * but no sense in crashing the users box for something
3896                          * we can survive anyway.
3897                          */
3898                         if (WARN_ON(!eb)) {
3899                                 spin_unlock(&mapping->private_lock);
3900                                 continue;
3901                         }
3902 
3903                         if (eb == prev_eb) {
3904                                 spin_unlock(&mapping->private_lock);
3905                                 continue;
3906                         }
3907 
3908                         ret = atomic_inc_not_zero(&eb->refs);
3909                         spin_unlock(&mapping->private_lock);
3910                         if (!ret)
3911                                 continue;
3912 
3913                         prev_eb = eb;
3914                         ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3915                         if (!ret) {
3916                                 free_extent_buffer(eb);
3917                                 continue;
3918                         }
3919 
3920                         ret = write_one_eb(eb, fs_info, wbc, &epd);
3921                         if (ret) {
3922                                 done = 1;
3923                                 free_extent_buffer(eb);
3924                                 break;
3925                         }
3926                         free_extent_buffer(eb);
3927 
3928                         /*
3929                          * the filesystem may choose to bump up nr_to_write.
3930                          * We have to make sure to honor the new nr_to_write
3931                          * at any time
3932                          */
3933                         nr_to_write_done = wbc->nr_to_write <= 0;
3934                 }
3935                 pagevec_release(&pvec);
3936                 cond_resched();
3937         }
3938         if (!scanned && !done) {
3939                 /*
3940                  * We hit the last page and there is more work to be done: wrap
3941                  * back to the start of the file
3942                  */
3943                 scanned = 1;
3944                 index = 0;
3945                 goto retry;
3946         }
3947         flush_write_bio(&epd);
3948         return ret;
3949 }
3950 
3951 /**
3952  * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3953  * @mapping: address space structure to write
3954  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3955  * @writepage: function called for each page
3956  * @data: data passed to writepage function
3957  *
3958  * If a page is already under I/O, write_cache_pages() skips it, even
3959  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
3960  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
3961  * and msync() need to guarantee that all the data which was dirty at the time
3962  * the call was made get new I/O started against them.  If wbc->sync_mode is
3963  * WB_SYNC_ALL then we were called for data integrity and we must wait for
3964  * existing IO to complete.
3965  */
3966 static int extent_write_cache_pages(struct extent_io_tree *tree,
3967                              struct address_space *mapping,
3968                              struct writeback_control *wbc,
3969                              writepage_t writepage, void *data,
3970                              void (*flush_fn)(void *))
3971 {
3972         struct inode *inode = mapping->host;
3973         int ret = 0;
3974         int done = 0;
3975         int err = 0;
3976         int nr_to_write_done = 0;
3977         struct pagevec pvec;
3978         int nr_pages;
3979         pgoff_t index;
3980         pgoff_t end;            /* Inclusive */
3981         int scanned = 0;
3982         int tag;
3983 
3984         /*
3985          * We have to hold onto the inode so that ordered extents can do their
3986          * work when the IO finishes.  The alternative to this is failing to add
3987          * an ordered extent if the igrab() fails there and that is a huge pain
3988          * to deal with, so instead just hold onto the inode throughout the
3989          * writepages operation.  If it fails here we are freeing up the inode
3990          * anyway and we'd rather not waste our time writing out stuff that is
3991          * going to be truncated anyway.
3992          */
3993         if (!igrab(inode))
3994                 return 0;
3995 
3996         pagevec_init(&pvec, 0);
3997         if (wbc->range_cyclic) {
3998                 index = mapping->writeback_index; /* Start from prev offset */
3999                 end = -1;
4000         } else {
4001                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
4002                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
4003                 scanned = 1;
4004         }
4005         if (wbc->sync_mode == WB_SYNC_ALL)
4006                 tag = PAGECACHE_TAG_TOWRITE;
4007         else
4008                 tag = PAGECACHE_TAG_DIRTY;
4009 retry:
4010         if (wbc->sync_mode == WB_SYNC_ALL)
4011                 tag_pages_for_writeback(mapping, index, end);
4012         while (!done && !nr_to_write_done && (index <= end) &&
4013                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
4014                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
4015                 unsigned i;
4016 
4017                 scanned = 1;
4018                 for (i = 0; i < nr_pages; i++) {
4019                         struct page *page = pvec.pages[i];
4020 
4021                         /*
4022                          * At this point we hold neither mapping->tree_lock nor
4023                          * lock on the page itself: the page may be truncated or
4024                          * invalidated (changing page->mapping to NULL), or even
4025                          * swizzled back from swapper_space to tmpfs file
4026                          * mapping
4027                          */
4028                         if (!trylock_page(page)) {
4029                                 flush_fn(data);
4030                                 lock_page(page);
4031                         }
4032 
4033                         if (unlikely(page->mapping != mapping)) {
4034                                 unlock_page(page);
4035                                 continue;
4036                         }
4037 
4038                         if (!wbc->range_cyclic && page->index > end) {
4039                                 done = 1;
4040                                 unlock_page(page);
4041                                 continue;
4042                         }
4043 
4044                         if (wbc->sync_mode != WB_SYNC_NONE) {
4045                                 if (PageWriteback(page))
4046                                         flush_fn(data);
4047                                 wait_on_page_writeback(page);
4048                         }
4049 
4050                         if (PageWriteback(page) ||
4051                             !clear_page_dirty_for_io(page)) {
4052                                 unlock_page(page);
4053                                 continue;
4054                         }
4055 
4056                         ret = (*writepage)(page, wbc, data);
4057 
4058                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4059                                 unlock_page(page);
4060                                 ret = 0;
4061                         }
4062                         if (!err && ret < 0)
4063                                 err = ret;
4064 
4065                         /*
4066                          * the filesystem may choose to bump up nr_to_write.
4067                          * We have to make sure to honor the new nr_to_write
4068                          * at any time
4069                          */
4070                         nr_to_write_done = wbc->nr_to_write <= 0;
4071                 }
4072                 pagevec_release(&pvec);
4073                 cond_resched();
4074         }
4075         if (!scanned && !done && !err) {
4076                 /*
4077                  * We hit the last page and there is more work to be done: wrap
4078                  * back to the start of the file
4079                  */
4080                 scanned = 1;
4081                 index = 0;
4082                 goto retry;
4083         }
4084         btrfs_add_delayed_iput(inode);
4085         return err;
4086 }
4087 
4088 static void flush_epd_write_bio(struct extent_page_data *epd)
4089 {
4090         if (epd->bio) {
4091                 int rw = WRITE;
4092                 int ret;
4093 
4094                 if (epd->sync_io)
4095                         rw = WRITE_SYNC;
4096 
4097                 ret = submit_one_bio(rw, epd->bio, 0, epd->bio_flags);
4098                 BUG_ON(ret < 0); /* -ENOMEM */
4099                 epd->bio = NULL;
4100         }
4101 }
4102 
4103 static noinline void flush_write_bio(void *data)
4104 {
4105         struct extent_page_data *epd = data;
4106         flush_epd_write_bio(epd);
4107 }
4108 
4109 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4110                           get_extent_t *get_extent,
4111                           struct writeback_control *wbc)
4112 {
4113         int ret;
4114         struct extent_page_data epd = {
4115                 .bio = NULL,
4116                 .tree = tree,
4117                 .get_extent = get_extent,
4118                 .extent_locked = 0,
4119                 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4120                 .bio_flags = 0,
4121         };
4122 
4123         ret = __extent_writepage(page, wbc, &epd);
4124 
4125         flush_epd_write_bio(&epd);
4126         return ret;
4127 }
4128 
4129 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4130                               u64 start, u64 end, get_extent_t *get_extent,
4131                               int mode)
4132 {
4133         int ret = 0;
4134         struct address_space *mapping = inode->i_mapping;
4135         struct page *page;
4136         unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
4137                 PAGE_CACHE_SHIFT;
4138 
4139         struct extent_page_data epd = {
4140                 .bio = NULL,
4141                 .tree = tree,
4142                 .get_extent = get_extent,
4143                 .extent_locked = 1,
4144                 .sync_io = mode == WB_SYNC_ALL,
4145                 .bio_flags = 0,
4146         };
4147         struct writeback_control wbc_writepages = {
4148                 .sync_mode      = mode,
4149                 .nr_to_write    = nr_pages * 2,
4150                 .range_start    = start,
4151                 .range_end      = end + 1,
4152         };
4153 
4154         while (start <= end) {
4155                 page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
4156                 if (clear_page_dirty_for_io(page))
4157                         ret = __extent_writepage(page, &wbc_writepages, &epd);
4158                 else {
4159                         if (tree->ops && tree->ops->writepage_end_io_hook)
4160                                 tree->ops->writepage_end_io_hook(page, start,
4161                                                  start + PAGE_CACHE_SIZE - 1,
4162                                                  NULL, 1);
4163                         unlock_page(page);
4164                 }
4165                 page_cache_release(page);
4166                 start += PAGE_CACHE_SIZE;
4167         }
4168 
4169         flush_epd_write_bio(&epd);
4170         return ret;
4171 }
4172 
4173 int extent_writepages(struct extent_io_tree *tree,
4174                       struct address_space *mapping,
4175                       get_extent_t *get_extent,
4176                       struct writeback_control *wbc)
4177 {
4178         int ret = 0;
4179         struct extent_page_data epd = {
4180                 .bio = NULL,
4181                 .tree = tree,
4182                 .get_extent = get_extent,
4183                 .extent_locked = 0,
4184                 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4185                 .bio_flags = 0,
4186         };
4187 
4188         ret = extent_write_cache_pages(tree, mapping, wbc,
4189                                        __extent_writepage, &epd,
4190                                        flush_write_bio);
4191         flush_epd_write_bio(&epd);
4192         return ret;
4193 }
4194 
4195 int extent_readpages(struct extent_io_tree *tree,
4196                      struct address_space *mapping,
4197                      struct list_head *pages, unsigned nr_pages,
4198                      get_extent_t get_extent)
4199 {
4200         struct bio *bio = NULL;
4201         unsigned page_idx;
4202         unsigned long bio_flags = 0;
4203         struct page *pagepool[16];
4204         struct page *page;
4205         struct extent_map *em_cached = NULL;
4206         int nr = 0;
4207         u64 prev_em_start = (u64)-1;
4208 
4209         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4210                 page = list_entry(pages->prev, struct page, lru);
4211 
4212                 prefetchw(&page->flags);
4213                 list_del(&page->lru);
4214                 if (add_to_page_cache_lru(page, mapping,
4215                                         page->index, GFP_NOFS)) {
4216                         page_cache_release(page);
4217                         continue;
4218                 }
4219 
4220                 pagepool[nr++] = page;
4221                 if (nr < ARRAY_SIZE(pagepool))
4222                         continue;
4223                 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4224                                    &bio, 0, &bio_flags, READ, &prev_em_start);
4225                 nr = 0;
4226         }
4227         if (nr)
4228                 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4229                                    &bio, 0, &bio_flags, READ, &prev_em_start);
4230 
4231         if (em_cached)
4232                 free_extent_map(em_cached);
4233 
4234         BUG_ON(!list_empty(pages));
4235         if (bio)
4236                 return submit_one_bio(READ, bio, 0, bio_flags);
4237         return 0;
4238 }
4239 
4240 /*
4241  * basic invalidatepage code, this waits on any locked or writeback
4242  * ranges corresponding to the page, and then deletes any extent state
4243  * records from the tree
4244  */
4245 int extent_invalidatepage(struct extent_io_tree *tree,
4246                           struct page *page, unsigned long offset)
4247 {
4248         struct extent_state *cached_state = NULL;
4249         u64 start = page_offset(page);
4250         u64 end = start + PAGE_CACHE_SIZE - 1;
4251         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4252 
4253         start += ALIGN(offset, blocksize);
4254         if (start > end)
4255                 return 0;
4256 
4257         lock_extent_bits(tree, start, end, 0, &cached_state);
4258         wait_on_page_writeback(page);
4259         clear_extent_bit(tree, start, end,
4260                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4261                          EXTENT_DO_ACCOUNTING,
4262                          1, 1, &cached_state, GFP_NOFS);
4263         return 0;
4264 }
4265 
4266 /*
4267  * a helper for releasepage, this tests for areas of the page that
4268  * are locked or under IO and drops the related state bits if it is safe
4269  * to drop the page.
4270  */
4271 static int try_release_extent_state(struct extent_map_tree *map,
4272                                     struct extent_io_tree *tree,
4273                                     struct page *page, gfp_t mask)
4274 {
4275         u64 start = page_offset(page);
4276         u64 end = start + PAGE_CACHE_SIZE - 1;
4277         int ret = 1;
4278 
4279         if (test_range_bit(tree, start, end,
4280                            EXTENT_IOBITS, 0, NULL))
4281                 ret = 0;
4282         else {
4283                 if ((mask & GFP_NOFS) == GFP_NOFS)
4284                         mask = GFP_NOFS;
4285                 /*
4286                  * at this point we can safely clear everything except the
4287                  * locked bit and the nodatasum bit
4288                  */
4289                 ret = clear_extent_bit(tree, start, end,
4290                                  ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4291                                  0, 0, NULL, mask);
4292 
4293                 /* if clear_extent_bit failed for enomem reasons,
4294                  * we can't allow the release to continue.
4295                  */
4296                 if (ret < 0)
4297                         ret = 0;
4298                 else
4299                         ret = 1;
4300         }
4301         return ret;
4302 }
4303 
4304 /*
4305  * a helper for releasepage.  As long as there are no locked extents
4306  * in the range corresponding to the page, both state records and extent
4307  * map records are removed
4308  */
4309 int try_release_extent_mapping(struct extent_map_tree *map,
4310                                struct extent_io_tree *tree, struct page *page,
4311                                gfp_t mask)
4312 {
4313         struct extent_map *em;
4314         u64 start = page_offset(page);
4315         u64 end = start + PAGE_CACHE_SIZE - 1;
4316 
4317         if ((mask & __GFP_WAIT) &&
4318             page->mapping->host->i_size > 16 * 1024 * 1024) {
4319                 u64 len;
4320                 while (start <= end) {
4321                         len = end - start + 1;
4322                         write_lock(&map->lock);
4323                         em = lookup_extent_mapping(map, start, len);
4324                         if (!em) {
4325                                 write_unlock(&map->lock);
4326                                 break;
4327                         }
4328                         if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4329                             em->start != start) {
4330                                 write_unlock(&map->lock);
4331                                 free_extent_map(em);
4332                                 break;
4333                         }
4334                         if (!test_range_bit(tree, em->start,
4335                                             extent_map_end(em) - 1,
4336                                             EXTENT_LOCKED | EXTENT_WRITEBACK,
4337                                             0, NULL)) {
4338                                 remove_extent_mapping(map, em);
4339                                 /* once for the rb tree */
4340                                 free_extent_map(em);
4341                         }
4342                         start = extent_map_end(em);
4343                         write_unlock(&map->lock);
4344 
4345                         /* once for us */
4346                         free_extent_map(em);
4347                 }
4348         }
4349         return try_release_extent_state(map, tree, page, mask);
4350 }
4351 
4352 /*
4353  * helper function for fiemap, which doesn't want to see any holes.
4354  * This maps until we find something past 'last'
4355  */
4356 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4357                                                 u64 offset,
4358                                                 u64 last,
4359                                                 get_extent_t *get_extent)
4360 {
4361         u64 sectorsize = BTRFS_I(inode)->root->sectorsize;
4362         struct extent_map *em;
4363         u64 len;
4364 
4365         if (offset >= last)
4366                 return NULL;
4367 
4368         while (1) {
4369                 len = last - offset;
4370                 if (len == 0)
4371                         break;
4372                 len = ALIGN(len, sectorsize);
4373                 em = get_extent(inode, NULL, 0, offset, len, 0);
4374                 if (IS_ERR_OR_NULL(em))
4375                         return em;
4376 
4377                 /* if this isn't a hole return it */
4378                 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4379                     em->block_start != EXTENT_MAP_HOLE) {
4380                         return em;
4381                 }
4382 
4383                 /* this is a hole, advance to the next extent */
4384                 offset = extent_map_end(em);
4385                 free_extent_map(em);
4386                 if (offset >= last)
4387                         break;
4388         }
4389         return NULL;
4390 }
4391 
4392 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4393                 __u64 start, __u64 len, get_extent_t *get_extent)
4394 {
4395         int ret = 0;
4396         u64 off = start;
4397         u64 max = start + len;
4398         u32 flags = 0;
4399         u32 found_type;
4400         u64 last;
4401         u64 last_for_get_extent = 0;
4402         u64 disko = 0;
4403         u64 isize = i_size_read(inode);
4404         struct btrfs_key found_key;
4405         struct extent_map *em = NULL;
4406         struct extent_state *cached_state = NULL;
4407         struct btrfs_path *path;
4408         struct btrfs_root *root = BTRFS_I(inode)->root;
4409         int end = 0;
4410         u64 em_start = 0;
4411         u64 em_len = 0;
4412         u64 em_end = 0;
4413 
4414         if (len == 0)
4415                 return -EINVAL;
4416 
4417         path = btrfs_alloc_path();
4418         if (!path)
4419                 return -ENOMEM;
4420         path->leave_spinning = 1;
4421 
4422         start = round_down(start, BTRFS_I(inode)->root->sectorsize);
4423         len = round_up(max, BTRFS_I(inode)->root->sectorsize) - start;
4424 
4425         /*
4426          * lookup the last file extent.  We're not using i_size here
4427          * because there might be preallocation past i_size
4428          */
4429         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1,
4430                                        0);
4431         if (ret < 0) {
4432                 btrfs_free_path(path);
4433                 return ret;
4434         }
4435         WARN_ON(!ret);
4436         path->slots[0]--;
4437         btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4438         found_type = found_key.type;
4439 
4440         /* No extents, but there might be delalloc bits */
4441         if (found_key.objectid != btrfs_ino(inode) ||
4442             found_type != BTRFS_EXTENT_DATA_KEY) {
4443                 /* have to trust i_size as the end */
4444                 last = (u64)-1;
4445                 last_for_get_extent = isize;
4446         } else {
4447                 /*
4448                  * remember the start of the last extent.  There are a
4449                  * bunch of different factors that go into the length of the
4450                  * extent, so its much less complex to remember where it started
4451                  */
4452                 last = found_key.offset;
4453                 last_for_get_extent = last + 1;
4454         }
4455         btrfs_release_path(path);
4456 
4457         /*
4458          * we might have some extents allocated but more delalloc past those
4459          * extents.  so, we trust isize unless the start of the last extent is
4460          * beyond isize
4461          */
4462         if (last < isize) {
4463                 last = (u64)-1;
4464                 last_for_get_extent = isize;
4465         }
4466 
4467         lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1, 0,
4468                          &cached_state);
4469 
4470         em = get_extent_skip_holes(inode, start, last_for_get_extent,
4471                                    get_extent);
4472         if (!em)
4473                 goto out;
4474         if (IS_ERR(em)) {
4475                 ret = PTR_ERR(em);
4476                 goto out;
4477         }
4478 
4479         while (!end) {
4480                 u64 offset_in_extent = 0;
4481 
4482                 /* break if the extent we found is outside the range */
4483                 if (em->start >= max || extent_map_end(em) < off)
4484                         break;
4485 
4486                 /*
4487                  * get_extent may return an extent that starts before our
4488                  * requested range.  We have to make sure the ranges
4489                  * we return to fiemap always move forward and don't
4490                  * overlap, so adjust the offsets here
4491                  */
4492                 em_start = max(em->start, off);
4493 
4494                 /*
4495                  * record the offset from the start of the extent
4496                  * for adjusting the disk offset below.  Only do this if the
4497                  * extent isn't compressed since our in ram offset may be past
4498                  * what we have actually allocated on disk.
4499                  */
4500                 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4501                         offset_in_extent = em_start - em->start;
4502                 em_end = extent_map_end(em);
4503                 em_len = em_end - em_start;
4504                 disko = 0;
4505                 flags = 0;
4506 
4507                 /*
4508                  * bump off for our next call to get_extent
4509                  */
4510                 off = extent_map_end(em);
4511                 if (off >= max)
4512                         end = 1;
4513 
4514                 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4515                         end = 1;
4516                         flags |= FIEMAP_EXTENT_LAST;
4517                 } else if (em->block_start == EXTENT_MAP_INLINE) {
4518                         flags |= (FIEMAP_EXTENT_DATA_INLINE |
4519                                   FIEMAP_EXTENT_NOT_ALIGNED);
4520                 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4521                         flags |= (FIEMAP_EXTENT_DELALLOC |
4522                                   FIEMAP_EXTENT_UNKNOWN);
4523                 } else if (fieinfo->fi_extents_max) {
4524                         u64 bytenr = em->block_start -
4525                                 (em->start - em->orig_start);
4526 
4527                         disko = em->block_start + offset_in_extent;
4528 
4529                         /*
4530                          * As btrfs supports shared space, this information
4531                          * can be exported to userspace tools via
4532                          * flag FIEMAP_EXTENT_SHARED.  If fi_extents_max == 0
4533                          * then we're just getting a count and we can skip the
4534                          * lookup stuff.
4535                          */
4536                         ret = btrfs_check_shared(NULL, root->fs_info,
4537                                                  root->objectid,
4538                                                  btrfs_ino(inode), bytenr);
4539                         if (ret < 0)
4540                                 goto out_free;
4541                         if (ret)
4542                                 flags |= FIEMAP_EXTENT_SHARED;
4543                         ret = 0;
4544                 }
4545                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4546                         flags |= FIEMAP_EXTENT_ENCODED;
4547 
4548                 free_extent_map(em);
4549                 em = NULL;
4550                 if ((em_start >= last) || em_len == (u64)-1 ||
4551                    (last == (u64)-1 && isize <= em_end)) {
4552                         flags |= FIEMAP_EXTENT_LAST;
4553                         end = 1;
4554                 }
4555 
4556                 /* now scan forward to see if this is really the last extent. */
4557                 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4558                                            get_extent);
4559                 if (IS_ERR(em)) {
4560                         ret = PTR_ERR(em);
4561                         goto out;
4562                 }
4563                 if (!em) {
4564                         flags |= FIEMAP_EXTENT_LAST;
4565                         end = 1;
4566                 }
4567                 ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
4568                                               em_len, flags);
4569                 if (ret) {
4570                         if (ret == 1)
4571                                 ret = 0;
4572                         goto out_free;
4573                 }
4574         }
4575 out_free:
4576         free_extent_map(em);
4577 out:
4578         btrfs_free_path(path);
4579         unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4580                              &cached_state, GFP_NOFS);
4581         return ret;
4582 }
4583 
4584 static void __free_extent_buffer(struct extent_buffer *eb)
4585 {
4586         btrfs_leak_debug_del(&eb->leak_list);
4587         kmem_cache_free(extent_buffer_cache, eb);
4588 }
4589 
4590 int extent_buffer_under_io(struct extent_buffer *eb)
4591 {
4592         return (atomic_read(&eb->io_pages) ||
4593                 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4594                 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4595 }
4596 
4597 /*
4598  * Helper for releasing extent buffer page.
4599  */
4600 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4601 {
4602         unsigned long index;
4603         struct page *page;
4604         int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4605 
4606         BUG_ON(extent_buffer_under_io(eb));
4607 
4608         index = num_extent_pages(eb->start, eb->len);
4609         if (index == 0)
4610                 return;
4611 
4612         do {
4613                 index--;
4614                 page = eb->pages[index];
4615                 if (!page)
4616                         continue;
4617                 if (mapped)
4618                         spin_lock(&page->mapping->private_lock);
4619                 /*
4620                  * We do this since we'll remove the pages after we've
4621                  * removed the eb from the radix tree, so we could race
4622                  * and have this page now attached to the new eb.  So
4623                  * only clear page_private if it's still connected to
4624                  * this eb.
4625                  */
4626                 if (PagePrivate(page) &&
4627                     page->private == (unsigned long)eb) {
4628                         BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4629                         BUG_ON(PageDirty(page));
4630                         BUG_ON(PageWriteback(page));
4631                         /*
4632                          * We need to make sure we haven't be attached
4633                          * to a new eb.
4634                          */
4635                         ClearPagePrivate(page);
4636                         set_page_private(page, 0);
4637                         /* One for the page private */
4638                         page_cache_release(page);
4639                 }
4640 
4641                 if (mapped)
4642                         spin_unlock(&page->mapping->private_lock);
4643 
4644                 /* One for when we alloced the page */
4645                 page_cache_release(page);
4646         } while (index != 0);
4647 }
4648 
4649 /*
4650  * Helper for releasing the extent buffer.
4651  */
4652 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4653 {
4654         btrfs_release_extent_buffer_page(eb);
4655         __free_extent_buffer(eb);
4656 }
4657 
4658 static struct extent_buffer *
4659 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4660                       unsigned long len)
4661 {
4662         struct extent_buffer *eb = NULL;
4663 
4664         eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS);
4665         if (eb == NULL)
4666                 return NULL;
4667         eb->start = start;
4668         eb->len = len;
4669         eb->fs_info = fs_info;
4670         eb->bflags = 0;
4671         rwlock_init(&eb->lock);
4672         atomic_set(&eb->write_locks, 0);
4673         atomic_set(&eb->read_locks, 0);
4674         atomic_set(&eb->blocking_readers, 0);
4675         atomic_set(&eb->blocking_writers, 0);
4676         atomic_set(&eb->spinning_readers, 0);
4677         atomic_set(&eb->spinning_writers, 0);
4678         eb->lock_nested = 0;
4679         init_waitqueue_head(&eb->write_lock_wq);
4680         init_waitqueue_head(&eb->read_lock_wq);
4681 
4682         btrfs_leak_debug_add(&eb->leak_list, &buffers);
4683 
4684         spin_lock_init(&eb->refs_lock);
4685         atomic_set(&eb->refs, 1);
4686         atomic_set(&eb->io_pages, 0);
4687 
4688         /*
4689          * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4690          */
4691         BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4692                 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4693         BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4694 
4695         return eb;
4696 }
4697 
4698 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4699 {
4700         unsigned long i;
4701         struct page *p;
4702         struct extent_buffer *new;
4703         unsigned long num_pages = num_extent_pages(src->start, src->len);
4704 
4705         new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4706         if (new == NULL)
4707                 return NULL;
4708 
4709         for (i = 0; i < num_pages; i++) {
4710                 p = alloc_page(GFP_NOFS);
4711                 if (!p) {
4712                         btrfs_release_extent_buffer(new);
4713                         return NULL;
4714                 }
4715                 attach_extent_buffer_page(new, p);
4716                 WARN_ON(PageDirty(p));
4717                 SetPageUptodate(p);
4718                 new->pages[i] = p;
4719         }
4720 
4721         copy_extent_buffer(new, src, 0, 0, src->len);
4722         set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4723         set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4724 
4725         return new;
4726 }
4727 
4728 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4729                                                 u64 start)
4730 {
4731         struct extent_buffer *eb;
4732         unsigned long len;
4733         unsigned long num_pages;
4734         unsigned long i;
4735 
4736         if (!fs_info) {
4737                 /*
4738                  * Called only from tests that don't always have a fs_info
4739                  * available, but we know that nodesize is 4096
4740                  */
4741                 len = 4096;
4742         } else {
4743                 len = fs_info->tree_root->nodesize;
4744         }
4745         num_pages = num_extent_pages(0, len);
4746 
4747         eb = __alloc_extent_buffer(fs_info, start, len);
4748         if (!eb)
4749                 return NULL;
4750 
4751         for (i = 0; i < num_pages; i++) {
4752                 eb->pages[i] = alloc_page(GFP_NOFS);
4753                 if (!eb->pages[i])
4754                         goto err;
4755         }
4756         set_extent_buffer_uptodate(eb);
4757         btrfs_set_header_nritems(eb, 0);
4758         set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4759 
4760         return eb;
4761 err:
4762         for (; i > 0; i--)
4763                 __free_page(eb->pages[i - 1]);
4764         __free_extent_buffer(eb);
4765         return NULL;
4766 }
4767 
4768 static void check_buffer_tree_ref(struct extent_buffer *eb)
4769 {
4770         int refs;
4771         /* the ref bit is tricky.  We have to make sure it is set
4772          * if we have the buffer dirty.   Otherwise the
4773          * code to free a buffer can end up dropping a dirty
4774          * page
4775          *
4776          * Once the ref bit is set, it won't go away while the
4777          * buffer is dirty or in writeback, and it also won't
4778          * go away while we have the reference count on the
4779          * eb bumped.
4780          *
4781          * We can't just set the ref bit without bumping the
4782          * ref on the eb because free_extent_buffer might
4783          * see the ref bit and try to clear it.  If this happens
4784          * free_extent_buffer might end up dropping our original
4785          * ref by mistake and freeing the page before we are able
4786          * to add one more ref.
4787          *
4788          * So bump the ref count first, then set the bit.  If someone
4789          * beat us to it, drop the ref we added.
4790          */
4791         refs = atomic_read(&eb->refs);
4792         if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4793                 return;
4794 
4795         spin_lock(&eb->refs_lock);
4796         if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4797                 atomic_inc(&eb->refs);
4798         spin_unlock(&eb->refs_lock);
4799 }
4800 
4801 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4802                 struct page *accessed)
4803 {
4804         unsigned long num_pages, i;
4805 
4806         check_buffer_tree_ref(eb);
4807 
4808         num_pages = num_extent_pages(eb->start, eb->len);
4809         for (i = 0; i < num_pages; i++) {
4810                 struct page *p = eb->pages[i];
4811 
4812                 if (p != accessed)
4813                         mark_page_accessed(p);
4814         }
4815 }
4816 
4817 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4818                                          u64 start)
4819 {
4820         struct extent_buffer *eb;
4821 
4822         rcu_read_lock();
4823         eb = radix_tree_lookup(&fs_info->buffer_radix,
4824                                start >> PAGE_CACHE_SHIFT);
4825         if (eb && atomic_inc_not_zero(&eb->refs)) {
4826                 rcu_read_unlock();
4827                 /*
4828                  * Lock our eb's refs_lock to avoid races with
4829                  * free_extent_buffer. When we get our eb it might be flagged
4830                  * with EXTENT_BUFFER_STALE and another task running
4831                  * free_extent_buffer might have seen that flag set,
4832                  * eb->refs == 2, that the buffer isn't under IO (dirty and
4833                  * writeback flags not set) and it's still in the tree (flag
4834                  * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4835                  * of decrementing the extent buffer's reference count twice.
4836                  * So here we could race and increment the eb's reference count,
4837                  * clear its stale flag, mark it as dirty and drop our reference
4838                  * before the other task finishes executing free_extent_buffer,
4839                  * which would later result in an attempt to free an extent
4840                  * buffer that is dirty.
4841                  */
4842                 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4843                         spin_lock(&eb->refs_lock);
4844                         spin_unlock(&eb->refs_lock);
4845                 }
4846                 mark_extent_buffer_accessed(eb, NULL);
4847                 return eb;
4848         }
4849         rcu_read_unlock();
4850 
4851         return NULL;
4852 }
4853 
4854 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4855 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4856                                                u64 start)
4857 {
4858         struct extent_buffer *eb, *exists = NULL;
4859         int ret;
4860 
4861         eb = find_extent_buffer(fs_info, start);
4862         if (eb)
4863                 return eb;
4864         eb = alloc_dummy_extent_buffer(fs_info, start);
4865         if (!eb)
4866                 return NULL;
4867         eb->fs_info = fs_info;
4868 again:
4869         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4870         if (ret)
4871                 goto free_eb;
4872         spin_lock(&fs_info->buffer_lock);
4873         ret = radix_tree_insert(&fs_info->buffer_radix,
4874                                 start >> PAGE_CACHE_SHIFT, eb);
4875         spin_unlock(&fs_info->buffer_lock);
4876         radix_tree_preload_end();
4877         if (ret == -EEXIST) {
4878                 exists = find_extent_buffer(fs_info, start);
4879                 if (exists)
4880                         goto free_eb;
4881                 else
4882                         goto again;
4883         }
4884         check_buffer_tree_ref(eb);
4885         set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4886 
4887         /*
4888          * We will free dummy extent buffer's if they come into
4889          * free_extent_buffer with a ref count of 2, but if we are using this we
4890          * want the buffers to stay in memory until we're done with them, so
4891          * bump the ref count again.
4892          */
4893         atomic_inc(&eb->refs);
4894         return eb;
4895 free_eb:
4896         btrfs_release_extent_buffer(eb);
4897         return exists;
4898 }
4899 #endif
4900 
4901 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4902                                           u64 start)
4903 {
4904         unsigned long len = fs_info->tree_root->nodesize;
4905         unsigned long num_pages = num_extent_pages(start, len);
4906         unsigned long i;
4907         unsigned long index = start >> PAGE_CACHE_SHIFT;
4908         struct extent_buffer *eb;
4909         struct extent_buffer *exists = NULL;
4910         struct page *p;
4911         struct address_space *mapping = fs_info->btree_inode->i_mapping;
4912         int uptodate = 1;
4913         int ret;
4914 
4915         eb = find_extent_buffer(fs_info, start);
4916         if (eb)
4917                 return eb;
4918 
4919         eb = __alloc_extent_buffer(fs_info, start, len);
4920         if (!eb)
4921                 return NULL;
4922 
4923         for (i = 0; i < num_pages; i++, index++) {
4924                 p = find_or_create_page(mapping, index, GFP_NOFS);
4925                 if (!p)
4926                         goto free_eb;
4927 
4928                 spin_lock(&mapping->private_lock);
4929                 if (PagePrivate(p)) {
4930                         /*
4931                          * We could have already allocated an eb for this page
4932                          * and attached one so lets see if we can get a ref on
4933                          * the existing eb, and if we can we know it's good and
4934                          * we can just return that one, else we know we can just
4935                          * overwrite page->private.
4936                          */
4937                         exists = (struct extent_buffer *)p->private;
4938                         if (atomic_inc_not_zero(&exists->refs)) {
4939                                 spin_unlock(&mapping->private_lock);
4940                                 unlock_page(p);
4941                                 page_cache_release(p);
4942                                 mark_extent_buffer_accessed(exists, p);
4943                                 goto free_eb;
4944                         }
4945                         exists = NULL;
4946 
4947                         /*
4948                          * Do this so attach doesn't complain and we need to
4949                          * drop the ref the old guy had.
4950                          */
4951                         ClearPagePrivate(p);
4952                         WARN_ON(PageDirty(p));
4953                         page_cache_release(p);
4954                 }
4955                 attach_extent_buffer_page(eb, p);
4956                 spin_unlock(&mapping->private_lock);
4957                 WARN_ON(PageDirty(p));
4958                 eb->pages[i] = p;
4959                 if (!PageUptodate(p))
4960                         uptodate = 0;
4961 
4962                 /*
4963                  * see below about how we avoid a nasty race with release page
4964                  * and why we unlock later
4965                  */
4966         }
4967         if (uptodate)
4968                 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4969 again:
4970         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
4971         if (ret)
4972                 goto free_eb;
4973 
4974         spin_lock(&fs_info->buffer_lock);
4975         ret = radix_tree_insert(&fs_info->buffer_radix,
4976                                 start >> PAGE_CACHE_SHIFT, eb);
4977         spin_unlock(&fs_info->buffer_lock);
4978         radix_tree_preload_end();
4979         if (ret == -EEXIST) {
4980                 exists = find_extent_buffer(fs_info, start);
4981                 if (exists)
4982                         goto free_eb;
4983                 else
4984                         goto again;
4985         }
4986         /* add one reference for the tree */
4987         check_buffer_tree_ref(eb);
4988         set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4989 
4990         /*
4991          * there is a race where release page may have
4992          * tried to find this extent buffer in the radix
4993          * but failed.  It will tell the VM it is safe to
4994          * reclaim the, and it will clear the page private bit.
4995          * We must make sure to set the page private bit properly
4996          * after the extent buffer is in the radix tree so
4997          * it doesn't get lost
4998          */
4999         SetPageChecked(eb->pages[0]);
5000         for (i = 1; i < num_pages; i++) {
5001                 p = eb->pages[i];
5002                 ClearPageChecked(p);
5003                 unlock_page(p);
5004         }
5005         unlock_page(eb->pages[0]);
5006         return eb;
5007 
5008 free_eb:
5009         WARN_ON(!atomic_dec_and_test(&eb->refs));
5010         for (i = 0; i < num_pages; i++) {
5011                 if (eb->pages[i])
5012                         unlock_page(eb->pages[i]);
5013         }
5014 
5015         btrfs_release_extent_buffer(eb);
5016         return exists;
5017 }
5018 
5019 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5020 {
5021         struct extent_buffer *eb =
5022                         container_of(head, struct extent_buffer, rcu_head);
5023 
5024         __free_extent_buffer(eb);
5025 }
5026 
5027 /* Expects to have eb->eb_lock already held */
5028 static int release_extent_buffer(struct extent_buffer *eb)
5029 {
5030         WARN_ON(atomic_read(&eb->refs) == 0);
5031         if (atomic_dec_and_test(&eb->refs)) {
5032                 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5033                         struct btrfs_fs_info *fs_info = eb->fs_info;
5034 
5035                         spin_unlock(&eb->refs_lock);
5036 
5037                         spin_lock(&fs_info->buffer_lock);
5038                         radix_tree_delete(&fs_info->buffer_radix,
5039                                           eb->start >> PAGE_CACHE_SHIFT);
5040                         spin_unlock(&fs_info->buffer_lock);
5041                 } else {
5042                         spin_unlock(&eb->refs_lock);
5043                 }
5044 
5045                 /* Should be safe to release our pages at this point */
5046                 btrfs_release_extent_buffer_page(eb);
5047 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5048                 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5049                         __free_extent_buffer(eb);
5050                         return 1;
5051                 }
5052 #endif
5053                 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5054                 return 1;
5055         }
5056         spin_unlock(&eb->refs_lock);
5057 
5058         return 0;
5059 }
5060 
5061 void free_extent_buffer(struct extent_buffer *eb)
5062 {
5063         int refs;
5064         int old;
5065         if (!eb)
5066                 return;
5067 
5068         while (1) {
5069                 refs = atomic_read(&eb->refs);
5070                 if (refs <= 3)
5071                         break;
5072                 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5073                 if (old == refs)
5074                         return;
5075         }
5076 
5077         spin_lock(&eb->refs_lock);
5078         if (atomic_read(&eb->refs) == 2 &&
5079             test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5080                 atomic_dec(&eb->refs);
5081 
5082         if (atomic_read(&eb->refs) == 2 &&
5083             test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5084             !extent_buffer_under_io(eb) &&
5085             test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5086                 atomic_dec(&eb->refs);
5087 
5088         /*
5089          * I know this is terrible, but it's temporary until we stop tracking
5090          * the uptodate bits and such for the extent buffers.
5091          */
5092         release_extent_buffer(eb);
5093 }
5094 
5095 void free_extent_buffer_stale(struct extent_buffer *eb)
5096 {
5097         if (!eb)
5098                 return;
5099 
5100         spin_lock(&eb->refs_lock);
5101         set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5102 
5103         if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5104             test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5105                 atomic_dec(&eb->refs);
5106         release_extent_buffer(eb);
5107 }
5108 
5109 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5110 {
5111         unsigned long i;
5112         unsigned long num_pages;
5113         struct page *page;
5114 
5115         num_pages = num_extent_pages(eb->start, eb->len);
5116 
5117         for (i = 0; i < num_pages; i++) {
5118                 page = eb->pages[i];
5119                 if (!PageDirty(page))
5120                         continue;
5121 
5122                 lock_page(page);
5123                 WARN_ON(!PagePrivate(page));
5124 
5125                 clear_page_dirty_for_io(page);
5126                 spin_lock_irq(&page->mapping->tree_lock);
5127                 if (!PageDirty(page)) {
5128                         radix_tree_tag_clear(&page->mapping->page_tree,
5129                                                 page_index(page),
5130                                                 PAGECACHE_TAG_DIRTY);
5131                 }
5132                 spin_unlock_irq(&page->mapping->tree_lock);
5133                 ClearPageError(page);
5134                 unlock_page(page);
5135         }
5136         WARN_ON(atomic_read(&eb->refs) == 0);
5137 }
5138 
5139 int set_extent_buffer_dirty(struct extent_buffer *eb)
5140 {
5141         unsigned long i;
5142         unsigned long num_pages;
5143         int was_dirty = 0;
5144 
5145         check_buffer_tree_ref(eb);
5146 
5147         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5148 
5149         num_pages = num_extent_pages(eb->start, eb->len);
5150         WARN_ON(atomic_read(&eb->refs) == 0);
5151         WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5152 
5153         for (i = 0; i < num_pages; i++)
5154                 set_page_dirty(eb->pages[i]);
5155         return was_dirty;
5156 }
5157 
5158 int clear_extent_buffer_uptodate(struct extent_buffer *eb)
5159 {
5160         unsigned long i;
5161         struct page *page;
5162         unsigned long num_pages;
5163 
5164         clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5165         num_pages = num_extent_pages(eb->start, eb->len);
5166         for (i = 0; i < num_pages; i++) {
5167                 page = eb->pages[i];
5168                 if (page)
5169                         ClearPageUptodate(page);
5170         }
5171         return 0;
5172 }
5173 
5174 int set_extent_buffer_uptodate(struct extent_buffer *eb)
5175 {
5176         unsigned long i;
5177         struct page *page;
5178         unsigned long num_pages;
5179 
5180         set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5181         num_pages = num_extent_pages(eb->start, eb->len);
5182         for (i = 0; i < num_pages; i++) {
5183                 page = eb->pages[i];
5184                 SetPageUptodate(page);
5185         }
5186         return 0;
5187 }
5188 
5189 int extent_buffer_uptodate(struct extent_buffer *eb)
5190 {
5191         return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5192 }
5193 
5194 int read_extent_buffer_pages(struct extent_io_tree *tree,
5195                              struct extent_buffer *eb, u64 start, int wait,
5196                              get_extent_t *get_extent, int mirror_num)
5197 {
5198         unsigned long i;
5199         unsigned long start_i;
5200         struct page *page;
5201         int err;
5202         int ret = 0;
5203         int locked_pages = 0;
5204         int all_uptodate = 1;
5205         unsigned long num_pages;
5206         unsigned long num_reads = 0;
5207         struct bio *bio = NULL;
5208         unsigned long bio_flags = 0;
5209 
5210         if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5211                 return 0;
5212 
5213         if (start) {
5214                 WARN_ON(start < eb->start);
5215                 start_i = (start >> PAGE_CACHE_SHIFT) -
5216                         (eb->start >> PAGE_CACHE_SHIFT);
5217         } else {
5218                 start_i = 0;
5219         }
5220 
5221         num_pages = num_extent_pages(eb->start, eb->len);
5222         for (i = start_i; i < num_pages; i++) {
5223                 page = eb->pages[i];
5224                 if (wait == WAIT_NONE) {
5225                         if (!trylock_page(page))
5226                                 goto unlock_exit;
5227                 } else {
5228                         lock_page(page);
5229                 }
5230                 locked_pages++;
5231                 if (!PageUptodate(page)) {
5232                         num_reads++;
5233                         all_uptodate = 0;
5234                 }
5235         }
5236         if (all_uptodate) {
5237                 if (start_i == 0)
5238                         set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5239                 goto unlock_exit;
5240         }
5241 
5242         clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5243         eb->read_mirror = 0;
5244         atomic_set(&eb->io_pages, num_reads);
5245         for (i = start_i; i < num_pages; i++) {
5246                 page = eb->pages[i];
5247                 if (!PageUptodate(page)) {
5248                         ClearPageError(page);
5249                         err = __extent_read_full_page(tree, page,
5250                                                       get_extent, &bio,
5251                                                       mirror_num, &bio_flags,
5252                                                       READ | REQ_META);
5253                         if (err)
5254                                 ret = err;
5255                 } else {
5256                         unlock_page(page);
5257                 }
5258         }
5259 
5260         if (bio) {
5261                 err = submit_one_bio(READ | REQ_META, bio, mirror_num,
5262                                      bio_flags);
5263                 if (err)
5264                         return err;
5265         }
5266 
5267         if (ret || wait != WAIT_COMPLETE)
5268                 return ret;
5269 
5270         for (i = start_i; i < num_pages; i++) {
5271                 page = eb->pages[i];
5272                 wait_on_page_locked(page);
5273                 if (!PageUptodate(page))
5274                         ret = -EIO;
5275         }
5276 
5277         return ret;
5278 
5279 unlock_exit:
5280         i = start_i;
5281         while (locked_pages > 0) {
5282                 page = eb->pages[i];
5283                 i++;
5284                 unlock_page(page);
5285                 locked_pages--;
5286         }
5287         return ret;
5288 }
5289 
5290 void read_extent_buffer(struct extent_buffer *eb, void *dstv,
5291                         unsigned long start,
5292                         unsigned long len)
5293 {
5294         size_t cur;
5295         size_t offset;
5296         struct page *page;
5297         char *kaddr;
5298         char *dst = (char *)dstv;
5299         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5300         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5301 
5302         WARN_ON(start > eb->len);
5303         WARN_ON(start + len > eb->start + eb->len);
5304 
5305         offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5306 
5307         while (len > 0) {
5308                 page = eb->pages[i];
5309 
5310                 cur = min(len, (PAGE_CACHE_SIZE - offset));
5311                 kaddr = page_address(page);
5312                 memcpy(dst, kaddr + offset, cur);
5313 
5314                 dst += cur;
5315                 len -= cur;
5316                 offset = 0;
5317                 i++;
5318         }
5319 }
5320 
5321 int read_extent_buffer_to_user(struct extent_buffer *eb, void __user *dstv,
5322                         unsigned long start,
5323                         unsigned long len)
5324 {
5325         size_t cur;
5326         size_t offset;
5327         struct page *page;
5328         char *kaddr;
5329         char __user *dst = (char __user *)dstv;
5330         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5331         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5332         int ret = 0;
5333 
5334         WARN_ON(start > eb->len);
5335         WARN_ON(start + len > eb->start + eb->len);
5336 
5337         offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5338 
5339         while (len > 0) {
5340                 page = eb->pages[i];
5341 
5342                 cur = min(len, (PAGE_CACHE_SIZE - offset));
5343                 kaddr = page_address(page);
5344                 if (copy_to_user(dst, kaddr + offset, cur)) {
5345                         ret = -EFAULT;
5346                         break;
5347                 }
5348 
5349                 dst += cur;
5350                 len -= cur;
5351                 offset = 0;
5352                 i++;
5353         }
5354 
5355         return ret;
5356 }
5357 
5358 int map_private_extent_buffer(struct extent_buffer *eb, unsigned long start,
5359                                unsigned long min_len, char **map,
5360                                unsigned long *map_start,
5361                                unsigned long *map_len)
5362 {
5363         size_t offset = start & (PAGE_CACHE_SIZE - 1);
5364         char *kaddr;
5365         struct page *p;
5366         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5367         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5368         unsigned long end_i = (start_offset + start + min_len - 1) >>
5369                 PAGE_CACHE_SHIFT;
5370 
5371         if (i != end_i)
5372                 return -EINVAL;
5373 
5374         if (i == 0) {
5375                 offset = start_offset;
5376                 *map_start = 0;
5377         } else {
5378                 offset = 0;
5379                 *map_start = ((u64)i << PAGE_CACHE_SHIFT) - start_offset;
5380         }
5381 
5382         if (start + min_len > eb->len) {
5383                 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, "
5384                        "wanted %lu %lu\n",
5385                        eb->start, eb->len, start, min_len);
5386                 return -EINVAL;
5387         }
5388 
5389         p = eb->pages[i];
5390         kaddr = page_address(p);
5391         *map = kaddr + offset;
5392         *map_len = PAGE_CACHE_SIZE - offset;
5393         return 0;
5394 }
5395 
5396 int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
5397                           unsigned long start,
5398                           unsigned long len)
5399 {
5400         size_t cur;
5401         size_t offset;
5402         struct page *page;
5403         char *kaddr;
5404         char *ptr = (char *)ptrv;
5405         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5406         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5407         int ret = 0;
5408 
5409         WARN_ON(start > eb->len);
5410         WARN_ON(start + len > eb->start + eb->len);
5411 
5412         offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5413 
5414         while (len > 0) {
5415                 page = eb->pages[i];
5416 
5417                 cur = min(len, (PAGE_CACHE_SIZE - offset));
5418 
5419                 kaddr = page_address(page);
5420                 ret = memcmp(ptr, kaddr + offset, cur);
5421                 if (ret)
5422                         break;
5423 
5424                 ptr += cur;
5425                 len -= cur;
5426                 offset = 0;
5427                 i++;
5428         }
5429         return ret;
5430 }
5431 
5432 void write_extent_buffer(struct extent_buffer *eb, const void *srcv,
5433                          unsigned long start, unsigned long len)
5434 {
5435         size_t cur;
5436         size_t offset;
5437         struct page *page;
5438         char *kaddr;
5439         char *src = (char *)srcv;
5440         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5441         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5442 
5443         WARN_ON(start > eb->len);
5444         WARN_ON(start + len > eb->start + eb->len);
5445 
5446         offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5447 
5448         while (len > 0) {
5449                 page = eb->pages[i];
5450                 WARN_ON(!PageUptodate(page));
5451 
5452                 cur = min(len, PAGE_CACHE_SIZE - offset);
5453                 kaddr = page_address(page);
5454                 memcpy(kaddr + offset, src, cur);
5455 
5456                 src += cur;
5457                 len -= cur;
5458                 offset = 0;
5459                 i++;
5460         }
5461 }
5462 
5463 void memset_extent_buffer(struct extent_buffer *eb, char c,
5464                           unsigned long start, unsigned long len)
5465 {
5466         size_t cur;
5467         size_t offset;
5468         struct page *page;
5469         char *kaddr;
5470         size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
5471         unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
5472 
5473         WARN_ON(start > eb->len);
5474         WARN_ON(start + len > eb->start + eb->len);
5475 
5476         offset = (start_offset + start) & (PAGE_CACHE_SIZE - 1);
5477 
5478         while (len > 0) {
5479                 page = eb->pages[i];
5480                 WARN_ON(!PageUptodate(page));
5481 
5482                 cur = min(len, PAGE_CACHE_SIZE - offset);
5483                 kaddr = page_address(page);
5484                 memset(kaddr + offset, c, cur);
5485 
5486                 len -= cur;
5487                 offset = 0;
5488                 i++;
5489         }
5490 }
5491 
5492 void copy_extent_buffer(struct extent_buffer *dst, struct extent_buffer *src,
5493                         unsigned long dst_offset, unsigned long src_offset,
5494                         unsigned long len)
5495 {
5496         u64 dst_len = dst->len;
5497         size_t cur;
5498         size_t offset;
5499         struct page *page;
5500         char *kaddr;
5501         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5502         unsigned long i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5503 
5504         WARN_ON(src->len != dst_len);
5505 
5506         offset = (start_offset + dst_offset) &
5507                 (PAGE_CACHE_SIZE - 1);
5508 
5509         while (len > 0) {
5510                 page = dst->pages[i];
5511                 WARN_ON(!PageUptodate(page));
5512 
5513                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE - offset));
5514 
5515                 kaddr = page_address(page);
5516                 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5517 
5518                 src_offset += cur;
5519                 len -= cur;
5520                 offset = 0;
5521                 i++;
5522         }
5523 }
5524 
5525 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5526 {
5527         unsigned long distance = (src > dst) ? src - dst : dst - src;
5528         return distance < len;
5529 }
5530 
5531 static void copy_pages(struct page *dst_page, struct page *src_page,
5532                        unsigned long dst_off, unsigned long src_off,
5533                        unsigned long len)
5534 {
5535         char *dst_kaddr = page_address(dst_page);
5536         char *src_kaddr;
5537         int must_memmove = 0;
5538 
5539         if (dst_page != src_page) {
5540                 src_kaddr = page_address(src_page);
5541         } else {
5542                 src_kaddr = dst_kaddr;
5543                 if (areas_overlap(src_off, dst_off, len))
5544                         must_memmove = 1;
5545         }
5546 
5547         if (must_memmove)
5548                 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5549         else
5550                 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5551 }
5552 
5553 void memcpy_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5554                            unsigned long src_offset, unsigned long len)
5555 {
5556         size_t cur;
5557         size_t dst_off_in_page;
5558         size_t src_off_in_page;
5559         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5560         unsigned long dst_i;
5561         unsigned long src_i;
5562 
5563         if (src_offset + len > dst->len) {
5564                 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5565                        "len %lu dst len %lu\n", src_offset, len, dst->len);
5566                 BUG_ON(1);
5567         }
5568         if (dst_offset + len > dst->len) {
5569                 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5570                        "len %lu dst len %lu\n", dst_offset, len, dst->len);
5571                 BUG_ON(1);
5572         }
5573 
5574         while (len > 0) {
5575                 dst_off_in_page = (start_offset + dst_offset) &
5576                         (PAGE_CACHE_SIZE - 1);
5577                 src_off_in_page = (start_offset + src_offset) &
5578                         (PAGE_CACHE_SIZE - 1);
5579 
5580                 dst_i = (start_offset + dst_offset) >> PAGE_CACHE_SHIFT;
5581                 src_i = (start_offset + src_offset) >> PAGE_CACHE_SHIFT;
5582 
5583                 cur = min(len, (unsigned long)(PAGE_CACHE_SIZE -
5584                                                src_off_in_page));
5585                 cur = min_t(unsigned long, cur,
5586                         (unsigned long)(PAGE_CACHE_SIZE - dst_off_in_page));
5587 
5588                 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5589                            dst_off_in_page, src_off_in_page, cur);
5590 
5591                 src_offset += cur;
5592                 dst_offset += cur;
5593                 len -= cur;
5594         }
5595 }
5596 
5597 void memmove_extent_buffer(struct extent_buffer *dst, unsigned long dst_offset,
5598                            unsigned long src_offset, unsigned long len)
5599 {
5600         size_t cur;
5601         size_t dst_off_in_page;
5602         size_t src_off_in_page;
5603         unsigned long dst_end = dst_offset + len - 1;
5604         unsigned long src_end = src_offset + len - 1;
5605         size_t start_offset = dst->start & ((u64)PAGE_CACHE_SIZE - 1);
5606         unsigned long dst_i;
5607         unsigned long src_i;
5608 
5609         if (src_offset + len > dst->len) {
5610                 printk(KERN_ERR "BTRFS: memmove bogus src_offset %lu move "
5611                        "len %lu len %lu\n", src_offset, len, dst->len);
5612                 BUG_ON(1);
5613         }
5614         if (dst_offset + len > dst->len) {
5615                 printk(KERN_ERR "BTRFS: memmove bogus dst_offset %lu move "
5616                        "len %lu len %lu\n", dst_offset, len, dst->len);
5617                 BUG_ON(1);
5618         }
5619         if (dst_offset < src_offset) {
5620                 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5621                 return;
5622         }
5623         while (len > 0) {
5624                 dst_i = (start_offset + dst_end) >> PAGE_CACHE_SHIFT;
5625                 src_i = (start_offset + src_end) >> PAGE_CACHE_SHIFT;
5626 
5627                 dst_off_in_page = (start_offset + dst_end) &
5628                         (PAGE_CACHE_SIZE - 1);
5629                 src_off_in_page = (start_offset + src_end) &
5630                         (PAGE_CACHE_SIZE - 1);
5631 
5632                 cur = min_t(unsigned long, len, src_off_in_page + 1);
5633                 cur = min(cur, dst_off_in_page + 1);
5634                 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5635                            dst_off_in_page - cur + 1,
5636                            src_off_in_page - cur + 1, cur);
5637 
5638                 dst_end -= cur;
5639                 src_end -= cur;
5640                 len -= cur;
5641         }
5642 }
5643 
5644 int try_release_extent_buffer(struct page *page)
5645 {
5646         struct extent_buffer *eb;
5647 
5648         /*
5649          * We need to make sure noboody is attaching this page to an eb right
5650          * now.
5651          */
5652         spin_lock(&page->mapping->private_lock);
5653         if (!PagePrivate(page)) {
5654                 spin_unlock(&page->mapping->private_lock);
5655                 return 1;
5656         }
5657 
5658         eb = (struct extent_buffer *)page->private;
5659         BUG_ON(!eb);
5660 
5661         /*
5662          * This is a little awful but should be ok, we need to make sure that
5663          * the eb doesn't disappear out from under us while we're looking at
5664          * this page.
5665          */
5666         spin_lock(&eb->refs_lock);
5667         if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5668                 spin_unlock(&eb->refs_lock);
5669                 spin_unlock(&page->mapping->private_lock);
5670                 return 0;
5671         }
5672         spin_unlock(&page->mapping->private_lock);
5673 
5674         /*
5675          * If tree ref isn't set then we know the ref on this eb is a real ref,
5676          * so just return, this page will likely be freed soon anyway.
5677          */
5678         if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5679                 spin_unlock(&eb->refs_lock);
5680                 return 0;
5681         }
5682 
5683         return release_extent_buffer(eb);
5684 }
5685 

~ [ 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