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

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

Version: ~ [ linux-5.9 ] ~ [ linux-5.8.14 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.70 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.150 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.200 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.238 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.238 ] ~ [ 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-3.16.85 ] ~ [ 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-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 // SPDX-License-Identifier: GPL-2.0
  2 #include <linux/bitops.h>
  3 #include <linux/slab.h>
  4 #include <linux/bio.h>
  5 #include <linux/mm.h>
  6 #include <linux/pagemap.h>
  7 #include <linux/page-flags.h>
  8 #include <linux/spinlock.h>
  9 #include <linux/blkdev.h>
 10 #include <linux/swap.h>
 11 #include <linux/writeback.h>
 12 #include <linux/pagevec.h>
 13 #include <linux/prefetch.h>
 14 #include <linux/cleancache.h>
 15 #include "extent_io.h"
 16 #include "extent_map.h"
 17 #include "ctree.h"
 18 #include "btrfs_inode.h"
 19 #include "volumes.h"
 20 #include "check-integrity.h"
 21 #include "locking.h"
 22 #include "rcu-string.h"
 23 #include "backref.h"
 24 
 25 static struct kmem_cache *extent_state_cache;
 26 static struct kmem_cache *extent_buffer_cache;
 27 static struct bio_set *btrfs_bioset;
 28 
 29 static inline bool extent_state_in_tree(const struct extent_state *state)
 30 {
 31         return !RB_EMPTY_NODE(&state->rb_node);
 32 }
 33 
 34 #ifdef CONFIG_BTRFS_DEBUG
 35 static LIST_HEAD(buffers);
 36 static LIST_HEAD(states);
 37 
 38 static DEFINE_SPINLOCK(leak_lock);
 39 
 40 static inline
 41 void btrfs_leak_debug_add(struct list_head *new, struct list_head *head)
 42 {
 43         unsigned long flags;
 44 
 45         spin_lock_irqsave(&leak_lock, flags);
 46         list_add(new, head);
 47         spin_unlock_irqrestore(&leak_lock, flags);
 48 }
 49 
 50 static inline
 51 void btrfs_leak_debug_del(struct list_head *entry)
 52 {
 53         unsigned long flags;
 54 
 55         spin_lock_irqsave(&leak_lock, flags);
 56         list_del(entry);
 57         spin_unlock_irqrestore(&leak_lock, flags);
 58 }
 59 
 60 static inline
 61 void btrfs_leak_debug_check(void)
 62 {
 63         struct extent_state *state;
 64         struct extent_buffer *eb;
 65 
 66         while (!list_empty(&states)) {
 67                 state = list_entry(states.next, struct extent_state, leak_list);
 68                 pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n",
 69                        state->start, state->end, state->state,
 70                        extent_state_in_tree(state),
 71                        refcount_read(&state->refs));
 72                 list_del(&state->leak_list);
 73                 kmem_cache_free(extent_state_cache, state);
 74         }
 75 
 76         while (!list_empty(&buffers)) {
 77                 eb = list_entry(buffers.next, struct extent_buffer, leak_list);
 78                 pr_err("BTRFS: buffer leak start %llu len %lu 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         if (tree->ops && tree->ops->check_extent_io_range)
 91                 tree->ops->check_extent_io_range(tree->private_data, caller,
 92                                                  start, end);
 93 }
 94 #else
 95 #define btrfs_leak_debug_add(new, head) do {} while (0)
 96 #define btrfs_leak_debug_del(entry)     do {} while (0)
 97 #define btrfs_leak_debug_check()        do {} while (0)
 98 #define btrfs_debug_check_extent_io_range(c, s, e)      do {} while (0)
 99 #endif
100 
101 #define BUFFER_LRU_MAX 64
102 
103 struct tree_entry {
104         u64 start;
105         u64 end;
106         struct rb_node rb_node;
107 };
108 
109 struct extent_page_data {
110         struct bio *bio;
111         struct extent_io_tree *tree;
112         get_extent_t *get_extent;
113 
114         /* tells writepage not to lock the state bits for this range
115          * it still does the unlocking
116          */
117         unsigned int extent_locked:1;
118 
119         /* tells the submit_bio code to use REQ_SYNC */
120         unsigned int sync_io:1;
121 };
122 
123 static void add_extent_changeset(struct extent_state *state, unsigned bits,
124                                  struct extent_changeset *changeset,
125                                  int set)
126 {
127         int ret;
128 
129         if (!changeset)
130                 return;
131         if (set && (state->state & bits) == bits)
132                 return;
133         if (!set && (state->state & bits) == 0)
134                 return;
135         changeset->bytes_changed += state->end - state->start + 1;
136         ret = ulist_add(&changeset->range_changed, state->start, state->end,
137                         GFP_ATOMIC);
138         /* ENOMEM */
139         BUG_ON(ret < 0);
140 }
141 
142 static noinline void flush_write_bio(void *data);
143 static inline struct btrfs_fs_info *
144 tree_fs_info(struct extent_io_tree *tree)
145 {
146         if (tree->ops)
147                 return tree->ops->tree_fs_info(tree->private_data);
148         return NULL;
149 }
150 
151 int __init extent_io_init(void)
152 {
153         extent_state_cache = kmem_cache_create("btrfs_extent_state",
154                         sizeof(struct extent_state), 0,
155                         SLAB_MEM_SPREAD, NULL);
156         if (!extent_state_cache)
157                 return -ENOMEM;
158 
159         extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
160                         sizeof(struct extent_buffer), 0,
161                         SLAB_MEM_SPREAD, NULL);
162         if (!extent_buffer_cache)
163                 goto free_state_cache;
164 
165         btrfs_bioset = bioset_create(BIO_POOL_SIZE,
166                                      offsetof(struct btrfs_io_bio, bio),
167                                      BIOSET_NEED_BVECS);
168         if (!btrfs_bioset)
169                 goto free_buffer_cache;
170 
171         if (bioset_integrity_create(btrfs_bioset, BIO_POOL_SIZE))
172                 goto free_bioset;
173 
174         return 0;
175 
176 free_bioset:
177         bioset_free(btrfs_bioset);
178         btrfs_bioset = NULL;
179 
180 free_buffer_cache:
181         kmem_cache_destroy(extent_buffer_cache);
182         extent_buffer_cache = NULL;
183 
184 free_state_cache:
185         kmem_cache_destroy(extent_state_cache);
186         extent_state_cache = NULL;
187         return -ENOMEM;
188 }
189 
190 void extent_io_exit(void)
191 {
192         btrfs_leak_debug_check();
193 
194         /*
195          * Make sure all delayed rcu free are flushed before we
196          * destroy caches.
197          */
198         rcu_barrier();
199         kmem_cache_destroy(extent_state_cache);
200         kmem_cache_destroy(extent_buffer_cache);
201         if (btrfs_bioset)
202                 bioset_free(btrfs_bioset);
203 }
204 
205 void extent_io_tree_init(struct extent_io_tree *tree,
206                          void *private_data)
207 {
208         tree->state = RB_ROOT;
209         tree->ops = NULL;
210         tree->dirty_bytes = 0;
211         spin_lock_init(&tree->lock);
212         tree->private_data = private_data;
213 }
214 
215 static struct extent_state *alloc_extent_state(gfp_t mask)
216 {
217         struct extent_state *state;
218 
219         /*
220          * The given mask might be not appropriate for the slab allocator,
221          * drop the unsupported bits
222          */
223         mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);
224         state = kmem_cache_alloc(extent_state_cache, mask);
225         if (!state)
226                 return state;
227         state->state = 0;
228         state->failrec = NULL;
229         RB_CLEAR_NODE(&state->rb_node);
230         btrfs_leak_debug_add(&state->leak_list, &states);
231         refcount_set(&state->refs, 1);
232         init_waitqueue_head(&state->wq);
233         trace_alloc_extent_state(state, mask, _RET_IP_);
234         return state;
235 }
236 
237 void free_extent_state(struct extent_state *state)
238 {
239         if (!state)
240                 return;
241         if (refcount_dec_and_test(&state->refs)) {
242                 WARN_ON(extent_state_in_tree(state));
243                 btrfs_leak_debug_del(&state->leak_list);
244                 trace_free_extent_state(state, _RET_IP_);
245                 kmem_cache_free(extent_state_cache, state);
246         }
247 }
248 
249 static struct rb_node *tree_insert(struct rb_root *root,
250                                    struct rb_node *search_start,
251                                    u64 offset,
252                                    struct rb_node *node,
253                                    struct rb_node ***p_in,
254                                    struct rb_node **parent_in)
255 {
256         struct rb_node **p;
257         struct rb_node *parent = NULL;
258         struct tree_entry *entry;
259 
260         if (p_in && parent_in) {
261                 p = *p_in;
262                 parent = *parent_in;
263                 goto do_insert;
264         }
265 
266         p = search_start ? &search_start : &root->rb_node;
267         while (*p) {
268                 parent = *p;
269                 entry = rb_entry(parent, struct tree_entry, rb_node);
270 
271                 if (offset < entry->start)
272                         p = &(*p)->rb_left;
273                 else if (offset > entry->end)
274                         p = &(*p)->rb_right;
275                 else
276                         return parent;
277         }
278 
279 do_insert:
280         rb_link_node(node, parent, p);
281         rb_insert_color(node, root);
282         return NULL;
283 }
284 
285 static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset,
286                                       struct rb_node **prev_ret,
287                                       struct rb_node **next_ret,
288                                       struct rb_node ***p_ret,
289                                       struct rb_node **parent_ret)
290 {
291         struct rb_root *root = &tree->state;
292         struct rb_node **n = &root->rb_node;
293         struct rb_node *prev = NULL;
294         struct rb_node *orig_prev = NULL;
295         struct tree_entry *entry;
296         struct tree_entry *prev_entry = NULL;
297 
298         while (*n) {
299                 prev = *n;
300                 entry = rb_entry(prev, struct tree_entry, rb_node);
301                 prev_entry = entry;
302 
303                 if (offset < entry->start)
304                         n = &(*n)->rb_left;
305                 else if (offset > entry->end)
306                         n = &(*n)->rb_right;
307                 else
308                         return *n;
309         }
310 
311         if (p_ret)
312                 *p_ret = n;
313         if (parent_ret)
314                 *parent_ret = prev;
315 
316         if (prev_ret) {
317                 orig_prev = prev;
318                 while (prev && offset > prev_entry->end) {
319                         prev = rb_next(prev);
320                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
321                 }
322                 *prev_ret = prev;
323                 prev = orig_prev;
324         }
325 
326         if (next_ret) {
327                 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
328                 while (prev && offset < prev_entry->start) {
329                         prev = rb_prev(prev);
330                         prev_entry = rb_entry(prev, struct tree_entry, rb_node);
331                 }
332                 *next_ret = prev;
333         }
334         return NULL;
335 }
336 
337 static inline struct rb_node *
338 tree_search_for_insert(struct extent_io_tree *tree,
339                        u64 offset,
340                        struct rb_node ***p_ret,
341                        struct rb_node **parent_ret)
342 {
343         struct rb_node *prev = NULL;
344         struct rb_node *ret;
345 
346         ret = __etree_search(tree, offset, &prev, NULL, p_ret, parent_ret);
347         if (!ret)
348                 return prev;
349         return ret;
350 }
351 
352 static inline struct rb_node *tree_search(struct extent_io_tree *tree,
353                                           u64 offset)
354 {
355         return tree_search_for_insert(tree, offset, NULL, NULL);
356 }
357 
358 static void merge_cb(struct extent_io_tree *tree, struct extent_state *new,
359                      struct extent_state *other)
360 {
361         if (tree->ops && tree->ops->merge_extent_hook)
362                 tree->ops->merge_extent_hook(tree->private_data, new, other);
363 }
364 
365 /*
366  * utility function to look for merge candidates inside a given range.
367  * Any extents with matching state are merged together into a single
368  * extent in the tree.  Extents with EXTENT_IO in their state field
369  * are not merged because the end_io handlers need to be able to do
370  * operations on them without sleeping (or doing allocations/splits).
371  *
372  * This should be called with the tree lock held.
373  */
374 static void merge_state(struct extent_io_tree *tree,
375                         struct extent_state *state)
376 {
377         struct extent_state *other;
378         struct rb_node *other_node;
379 
380         if (state->state & (EXTENT_IOBITS | EXTENT_BOUNDARY))
381                 return;
382 
383         other_node = rb_prev(&state->rb_node);
384         if (other_node) {
385                 other = rb_entry(other_node, struct extent_state, rb_node);
386                 if (other->end == state->start - 1 &&
387                     other->state == state->state) {
388                         merge_cb(tree, state, other);
389                         state->start = other->start;
390                         rb_erase(&other->rb_node, &tree->state);
391                         RB_CLEAR_NODE(&other->rb_node);
392                         free_extent_state(other);
393                 }
394         }
395         other_node = rb_next(&state->rb_node);
396         if (other_node) {
397                 other = rb_entry(other_node, struct extent_state, rb_node);
398                 if (other->start == state->end + 1 &&
399                     other->state == state->state) {
400                         merge_cb(tree, state, other);
401                         state->end = other->end;
402                         rb_erase(&other->rb_node, &tree->state);
403                         RB_CLEAR_NODE(&other->rb_node);
404                         free_extent_state(other);
405                 }
406         }
407 }
408 
409 static void set_state_cb(struct extent_io_tree *tree,
410                          struct extent_state *state, unsigned *bits)
411 {
412         if (tree->ops && tree->ops->set_bit_hook)
413                 tree->ops->set_bit_hook(tree->private_data, state, bits);
414 }
415 
416 static void clear_state_cb(struct extent_io_tree *tree,
417                            struct extent_state *state, unsigned *bits)
418 {
419         if (tree->ops && tree->ops->clear_bit_hook)
420                 tree->ops->clear_bit_hook(tree->private_data, state, bits);
421 }
422 
423 static void set_state_bits(struct extent_io_tree *tree,
424                            struct extent_state *state, unsigned *bits,
425                            struct extent_changeset *changeset);
426 
427 /*
428  * insert an extent_state struct into the tree.  'bits' are set on the
429  * struct before it is inserted.
430  *
431  * This may return -EEXIST if the extent is already there, in which case the
432  * state struct is freed.
433  *
434  * The tree lock is not taken internally.  This is a utility function and
435  * probably isn't what you want to call (see set/clear_extent_bit).
436  */
437 static int insert_state(struct extent_io_tree *tree,
438                         struct extent_state *state, u64 start, u64 end,
439                         struct rb_node ***p,
440                         struct rb_node **parent,
441                         unsigned *bits, struct extent_changeset *changeset)
442 {
443         struct rb_node *node;
444 
445         if (end < start)
446                 WARN(1, KERN_ERR "BTRFS: end < start %llu %llu\n",
447                        end, start);
448         state->start = start;
449         state->end = end;
450 
451         set_state_bits(tree, state, bits, changeset);
452 
453         node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent);
454         if (node) {
455                 struct extent_state *found;
456                 found = rb_entry(node, struct extent_state, rb_node);
457                 pr_err("BTRFS: found node %llu %llu on insert of %llu %llu\n",
458                        found->start, found->end, start, end);
459                 return -EEXIST;
460         }
461         merge_state(tree, state);
462         return 0;
463 }
464 
465 static void split_cb(struct extent_io_tree *tree, struct extent_state *orig,
466                      u64 split)
467 {
468         if (tree->ops && tree->ops->split_extent_hook)
469                 tree->ops->split_extent_hook(tree->private_data, orig, split);
470 }
471 
472 /*
473  * split a given extent state struct in two, inserting the preallocated
474  * struct 'prealloc' as the newly created second half.  'split' indicates an
475  * offset inside 'orig' where it should be split.
476  *
477  * Before calling,
478  * the tree has 'orig' at [orig->start, orig->end].  After calling, there
479  * are two extent state structs in the tree:
480  * prealloc: [orig->start, split - 1]
481  * orig: [ split, orig->end ]
482  *
483  * The tree locks are not taken by this function. They need to be held
484  * by the caller.
485  */
486 static int split_state(struct extent_io_tree *tree, struct extent_state *orig,
487                        struct extent_state *prealloc, u64 split)
488 {
489         struct rb_node *node;
490 
491         split_cb(tree, orig, split);
492 
493         prealloc->start = orig->start;
494         prealloc->end = split - 1;
495         prealloc->state = orig->state;
496         orig->start = split;
497 
498         node = tree_insert(&tree->state, &orig->rb_node, prealloc->end,
499                            &prealloc->rb_node, NULL, NULL);
500         if (node) {
501                 free_extent_state(prealloc);
502                 return -EEXIST;
503         }
504         return 0;
505 }
506 
507 static struct extent_state *next_state(struct extent_state *state)
508 {
509         struct rb_node *next = rb_next(&state->rb_node);
510         if (next)
511                 return rb_entry(next, struct extent_state, rb_node);
512         else
513                 return NULL;
514 }
515 
516 /*
517  * utility function to clear some bits in an extent state struct.
518  * it will optionally wake up any one waiting on this state (wake == 1).
519  *
520  * If no bits are set on the state struct after clearing things, the
521  * struct is freed and removed from the tree
522  */
523 static struct extent_state *clear_state_bit(struct extent_io_tree *tree,
524                                             struct extent_state *state,
525                                             unsigned *bits, int wake,
526                                             struct extent_changeset *changeset)
527 {
528         struct extent_state *next;
529         unsigned bits_to_clear = *bits & ~EXTENT_CTLBITS;
530 
531         if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) {
532                 u64 range = state->end - state->start + 1;
533                 WARN_ON(range > tree->dirty_bytes);
534                 tree->dirty_bytes -= range;
535         }
536         clear_state_cb(tree, state, bits);
537         add_extent_changeset(state, bits_to_clear, changeset, 0);
538         state->state &= ~bits_to_clear;
539         if (wake)
540                 wake_up(&state->wq);
541         if (state->state == 0) {
542                 next = next_state(state);
543                 if (extent_state_in_tree(state)) {
544                         rb_erase(&state->rb_node, &tree->state);
545                         RB_CLEAR_NODE(&state->rb_node);
546                         free_extent_state(state);
547                 } else {
548                         WARN_ON(1);
549                 }
550         } else {
551                 merge_state(tree, state);
552                 next = next_state(state);
553         }
554         return next;
555 }
556 
557 static struct extent_state *
558 alloc_extent_state_atomic(struct extent_state *prealloc)
559 {
560         if (!prealloc)
561                 prealloc = alloc_extent_state(GFP_ATOMIC);
562 
563         return prealloc;
564 }
565 
566 static void extent_io_tree_panic(struct extent_io_tree *tree, int err)
567 {
568         btrfs_panic(tree_fs_info(tree), err,
569                     "Locking error: Extent tree was modified by another thread while locked.");
570 }
571 
572 /*
573  * clear some bits on a range in the tree.  This may require splitting
574  * or inserting elements in the tree, so the gfp mask is used to
575  * indicate which allocations or sleeping are allowed.
576  *
577  * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
578  * the given range from the tree regardless of state (ie for truncate).
579  *
580  * the range [start, end] is inclusive.
581  *
582  * This takes the tree lock, and returns 0 on success and < 0 on error.
583  */
584 static int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
585                               unsigned bits, int wake, int delete,
586                               struct extent_state **cached_state,
587                               gfp_t mask, struct extent_changeset *changeset)
588 {
589         struct extent_state *state;
590         struct extent_state *cached;
591         struct extent_state *prealloc = NULL;
592         struct rb_node *node;
593         u64 last_end;
594         int err;
595         int clear = 0;
596 
597         btrfs_debug_check_extent_io_range(tree, start, end);
598 
599         if (bits & EXTENT_DELALLOC)
600                 bits |= EXTENT_NORESERVE;
601 
602         if (delete)
603                 bits |= ~EXTENT_CTLBITS;
604         bits |= EXTENT_FIRST_DELALLOC;
605 
606         if (bits & (EXTENT_IOBITS | EXTENT_BOUNDARY))
607                 clear = 1;
608 again:
609         if (!prealloc && gfpflags_allow_blocking(mask)) {
610                 /*
611                  * Don't care for allocation failure here because we might end
612                  * up not needing the pre-allocated extent state at all, which
613                  * is the case if we only have in the tree extent states that
614                  * cover our input range and don't cover too any other range.
615                  * If we end up needing a new extent state we allocate it later.
616                  */
617                 prealloc = alloc_extent_state(mask);
618         }
619 
620         spin_lock(&tree->lock);
621         if (cached_state) {
622                 cached = *cached_state;
623 
624                 if (clear) {
625                         *cached_state = NULL;
626                         cached_state = NULL;
627                 }
628 
629                 if (cached && extent_state_in_tree(cached) &&
630                     cached->start <= start && cached->end > start) {
631                         if (clear)
632                                 refcount_dec(&cached->refs);
633                         state = cached;
634                         goto hit_next;
635                 }
636                 if (clear)
637                         free_extent_state(cached);
638         }
639         /*
640          * this search will find the extents that end after
641          * our range starts
642          */
643         node = tree_search(tree, start);
644         if (!node)
645                 goto out;
646         state = rb_entry(node, struct extent_state, rb_node);
647 hit_next:
648         if (state->start > end)
649                 goto out;
650         WARN_ON(state->end < start);
651         last_end = state->end;
652 
653         /* the state doesn't have the wanted bits, go ahead */
654         if (!(state->state & bits)) {
655                 state = next_state(state);
656                 goto next;
657         }
658 
659         /*
660          *     | ---- desired range ---- |
661          *  | state | or
662          *  | ------------- state -------------- |
663          *
664          * We need to split the extent we found, and may flip
665          * bits on second half.
666          *
667          * If the extent we found extends past our range, we
668          * just split and search again.  It'll get split again
669          * the next time though.
670          *
671          * If the extent we found is inside our range, we clear
672          * the desired bit on it.
673          */
674 
675         if (state->start < start) {
676                 prealloc = alloc_extent_state_atomic(prealloc);
677                 BUG_ON(!prealloc);
678                 err = split_state(tree, state, prealloc, start);
679                 if (err)
680                         extent_io_tree_panic(tree, err);
681 
682                 prealloc = NULL;
683                 if (err)
684                         goto out;
685                 if (state->end <= end) {
686                         state = clear_state_bit(tree, state, &bits, wake,
687                                                 changeset);
688                         goto next;
689                 }
690                 goto search_again;
691         }
692         /*
693          * | ---- desired range ---- |
694          *                        | state |
695          * We need to split the extent, and clear the bit
696          * on the first half
697          */
698         if (state->start <= end && state->end > end) {
699                 prealloc = alloc_extent_state_atomic(prealloc);
700                 BUG_ON(!prealloc);
701                 err = split_state(tree, state, prealloc, end + 1);
702                 if (err)
703                         extent_io_tree_panic(tree, err);
704 
705                 if (wake)
706                         wake_up(&state->wq);
707 
708                 clear_state_bit(tree, prealloc, &bits, wake, changeset);
709 
710                 prealloc = NULL;
711                 goto out;
712         }
713 
714         state = clear_state_bit(tree, state, &bits, wake, changeset);
715 next:
716         if (last_end == (u64)-1)
717                 goto out;
718         start = last_end + 1;
719         if (start <= end && state && !need_resched())
720                 goto hit_next;
721 
722 search_again:
723         if (start > end)
724                 goto out;
725         spin_unlock(&tree->lock);
726         if (gfpflags_allow_blocking(mask))
727                 cond_resched();
728         goto again;
729 
730 out:
731         spin_unlock(&tree->lock);
732         if (prealloc)
733                 free_extent_state(prealloc);
734 
735         return 0;
736 
737 }
738 
739 static void wait_on_state(struct extent_io_tree *tree,
740                           struct extent_state *state)
741                 __releases(tree->lock)
742                 __acquires(tree->lock)
743 {
744         DEFINE_WAIT(wait);
745         prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
746         spin_unlock(&tree->lock);
747         schedule();
748         spin_lock(&tree->lock);
749         finish_wait(&state->wq, &wait);
750 }
751 
752 /*
753  * waits for one or more bits to clear on a range in the state tree.
754  * The range [start, end] is inclusive.
755  * The tree lock is taken by this function
756  */
757 static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
758                             unsigned long bits)
759 {
760         struct extent_state *state;
761         struct rb_node *node;
762 
763         btrfs_debug_check_extent_io_range(tree, start, end);
764 
765         spin_lock(&tree->lock);
766 again:
767         while (1) {
768                 /*
769                  * this search will find all the extents that end after
770                  * our range starts
771                  */
772                 node = tree_search(tree, start);
773 process_node:
774                 if (!node)
775                         break;
776 
777                 state = rb_entry(node, struct extent_state, rb_node);
778 
779                 if (state->start > end)
780                         goto out;
781 
782                 if (state->state & bits) {
783                         start = state->start;
784                         refcount_inc(&state->refs);
785                         wait_on_state(tree, state);
786                         free_extent_state(state);
787                         goto again;
788                 }
789                 start = state->end + 1;
790 
791                 if (start > end)
792                         break;
793 
794                 if (!cond_resched_lock(&tree->lock)) {
795                         node = rb_next(node);
796                         goto process_node;
797                 }
798         }
799 out:
800         spin_unlock(&tree->lock);
801 }
802 
803 static void set_state_bits(struct extent_io_tree *tree,
804                            struct extent_state *state,
805                            unsigned *bits, struct extent_changeset *changeset)
806 {
807         unsigned bits_to_set = *bits & ~EXTENT_CTLBITS;
808 
809         set_state_cb(tree, state, bits);
810         if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) {
811                 u64 range = state->end - state->start + 1;
812                 tree->dirty_bytes += range;
813         }
814         add_extent_changeset(state, bits_to_set, changeset, 1);
815         state->state |= bits_to_set;
816 }
817 
818 static void cache_state_if_flags(struct extent_state *state,
819                                  struct extent_state **cached_ptr,
820                                  unsigned flags)
821 {
822         if (cached_ptr && !(*cached_ptr)) {
823                 if (!flags || (state->state & flags)) {
824                         *cached_ptr = state;
825                         refcount_inc(&state->refs);
826                 }
827         }
828 }
829 
830 static void cache_state(struct extent_state *state,
831                         struct extent_state **cached_ptr)
832 {
833         return cache_state_if_flags(state, cached_ptr,
834                                     EXTENT_IOBITS | EXTENT_BOUNDARY);
835 }
836 
837 /*
838  * set some bits on a range in the tree.  This may require allocations or
839  * sleeping, so the gfp mask is used to indicate what is allowed.
840  *
841  * If any of the exclusive bits are set, this will fail with -EEXIST if some
842  * part of the range already has the desired bits set.  The start of the
843  * existing range is returned in failed_start in this case.
844  *
845  * [start, end] is inclusive This takes the tree lock.
846  */
847 
848 static int __must_check
849 __set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
850                  unsigned bits, unsigned exclusive_bits,
851                  u64 *failed_start, struct extent_state **cached_state,
852                  gfp_t mask, struct extent_changeset *changeset)
853 {
854         struct extent_state *state;
855         struct extent_state *prealloc = NULL;
856         struct rb_node *node;
857         struct rb_node **p;
858         struct rb_node *parent;
859         int err = 0;
860         u64 last_start;
861         u64 last_end;
862 
863         btrfs_debug_check_extent_io_range(tree, start, end);
864 
865         bits |= EXTENT_FIRST_DELALLOC;
866 again:
867         if (!prealloc && gfpflags_allow_blocking(mask)) {
868                 /*
869                  * Don't care for allocation failure here because we might end
870                  * up not needing the pre-allocated extent state at all, which
871                  * is the case if we only have in the tree extent states that
872                  * cover our input range and don't cover too any other range.
873                  * If we end up needing a new extent state we allocate it later.
874                  */
875                 prealloc = alloc_extent_state(mask);
876         }
877 
878         spin_lock(&tree->lock);
879         if (cached_state && *cached_state) {
880                 state = *cached_state;
881                 if (state->start <= start && state->end > start &&
882                     extent_state_in_tree(state)) {
883                         node = &state->rb_node;
884                         goto hit_next;
885                 }
886         }
887         /*
888          * this search will find all the extents that end after
889          * our range starts.
890          */
891         node = tree_search_for_insert(tree, start, &p, &parent);
892         if (!node) {
893                 prealloc = alloc_extent_state_atomic(prealloc);
894                 BUG_ON(!prealloc);
895                 err = insert_state(tree, prealloc, start, end,
896                                    &p, &parent, &bits, changeset);
897                 if (err)
898                         extent_io_tree_panic(tree, err);
899 
900                 cache_state(prealloc, cached_state);
901                 prealloc = NULL;
902                 goto out;
903         }
904         state = rb_entry(node, struct extent_state, rb_node);
905 hit_next:
906         last_start = state->start;
907         last_end = state->end;
908 
909         /*
910          * | ---- desired range ---- |
911          * | state |
912          *
913          * Just lock what we found and keep going
914          */
915         if (state->start == start && state->end <= end) {
916                 if (state->state & exclusive_bits) {
917                         *failed_start = state->start;
918                         err = -EEXIST;
919                         goto out;
920                 }
921 
922                 set_state_bits(tree, state, &bits, changeset);
923                 cache_state(state, cached_state);
924                 merge_state(tree, state);
925                 if (last_end == (u64)-1)
926                         goto out;
927                 start = last_end + 1;
928                 state = next_state(state);
929                 if (start < end && state && state->start == start &&
930                     !need_resched())
931                         goto hit_next;
932                 goto search_again;
933         }
934 
935         /*
936          *     | ---- desired range ---- |
937          * | state |
938          *   or
939          * | ------------- state -------------- |
940          *
941          * We need to split the extent we found, and may flip bits on
942          * second half.
943          *
944          * If the extent we found extends past our
945          * range, we just split and search again.  It'll get split
946          * again the next time though.
947          *
948          * If the extent we found is inside our range, we set the
949          * desired bit on it.
950          */
951         if (state->start < start) {
952                 if (state->state & exclusive_bits) {
953                         *failed_start = start;
954                         err = -EEXIST;
955                         goto out;
956                 }
957 
958                 prealloc = alloc_extent_state_atomic(prealloc);
959                 BUG_ON(!prealloc);
960                 err = split_state(tree, state, prealloc, start);
961                 if (err)
962                         extent_io_tree_panic(tree, err);
963 
964                 prealloc = NULL;
965                 if (err)
966                         goto out;
967                 if (state->end <= end) {
968                         set_state_bits(tree, state, &bits, changeset);
969                         cache_state(state, cached_state);
970                         merge_state(tree, state);
971                         if (last_end == (u64)-1)
972                                 goto out;
973                         start = last_end + 1;
974                         state = next_state(state);
975                         if (start < end && state && state->start == start &&
976                             !need_resched())
977                                 goto hit_next;
978                 }
979                 goto search_again;
980         }
981         /*
982          * | ---- desired range ---- |
983          *     | state | or               | state |
984          *
985          * There's a hole, we need to insert something in it and
986          * ignore the extent we found.
987          */
988         if (state->start > start) {
989                 u64 this_end;
990                 if (end < last_start)
991                         this_end = end;
992                 else
993                         this_end = last_start - 1;
994 
995                 prealloc = alloc_extent_state_atomic(prealloc);
996                 BUG_ON(!prealloc);
997 
998                 /*
999                  * Avoid to free 'prealloc' if it can be merged with
1000                  * the later extent.
1001                  */
1002                 err = insert_state(tree, prealloc, start, this_end,
1003                                    NULL, NULL, &bits, changeset);
1004                 if (err)
1005                         extent_io_tree_panic(tree, err);
1006 
1007                 cache_state(prealloc, cached_state);
1008                 prealloc = NULL;
1009                 start = this_end + 1;
1010                 goto search_again;
1011         }
1012         /*
1013          * | ---- desired range ---- |
1014          *                        | state |
1015          * We need to split the extent, and set the bit
1016          * on the first half
1017          */
1018         if (state->start <= end && state->end > end) {
1019                 if (state->state & exclusive_bits) {
1020                         *failed_start = start;
1021                         err = -EEXIST;
1022                         goto out;
1023                 }
1024 
1025                 prealloc = alloc_extent_state_atomic(prealloc);
1026                 BUG_ON(!prealloc);
1027                 err = split_state(tree, state, prealloc, end + 1);
1028                 if (err)
1029                         extent_io_tree_panic(tree, err);
1030 
1031                 set_state_bits(tree, prealloc, &bits, changeset);
1032                 cache_state(prealloc, cached_state);
1033                 merge_state(tree, prealloc);
1034                 prealloc = NULL;
1035                 goto out;
1036         }
1037 
1038 search_again:
1039         if (start > end)
1040                 goto out;
1041         spin_unlock(&tree->lock);
1042         if (gfpflags_allow_blocking(mask))
1043                 cond_resched();
1044         goto again;
1045 
1046 out:
1047         spin_unlock(&tree->lock);
1048         if (prealloc)
1049                 free_extent_state(prealloc);
1050 
1051         return err;
1052 
1053 }
1054 
1055 int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1056                    unsigned bits, u64 * failed_start,
1057                    struct extent_state **cached_state, gfp_t mask)
1058 {
1059         return __set_extent_bit(tree, start, end, bits, 0, failed_start,
1060                                 cached_state, mask, NULL);
1061 }
1062 
1063 
1064 /**
1065  * convert_extent_bit - convert all bits in a given range from one bit to
1066  *                      another
1067  * @tree:       the io tree to search
1068  * @start:      the start offset in bytes
1069  * @end:        the end offset in bytes (inclusive)
1070  * @bits:       the bits to set in this range
1071  * @clear_bits: the bits to clear in this range
1072  * @cached_state:       state that we're going to cache
1073  *
1074  * This will go through and set bits for the given range.  If any states exist
1075  * already in this range they are set with the given bit and cleared of the
1076  * clear_bits.  This is only meant to be used by things that are mergeable, ie
1077  * converting from say DELALLOC to DIRTY.  This is not meant to be used with
1078  * boundary bits like LOCK.
1079  *
1080  * All allocations are done with GFP_NOFS.
1081  */
1082 int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1083                        unsigned bits, unsigned clear_bits,
1084                        struct extent_state **cached_state)
1085 {
1086         struct extent_state *state;
1087         struct extent_state *prealloc = NULL;
1088         struct rb_node *node;
1089         struct rb_node **p;
1090         struct rb_node *parent;
1091         int err = 0;
1092         u64 last_start;
1093         u64 last_end;
1094         bool first_iteration = true;
1095 
1096         btrfs_debug_check_extent_io_range(tree, start, end);
1097 
1098 again:
1099         if (!prealloc) {
1100                 /*
1101                  * Best effort, don't worry if extent state allocation fails
1102                  * here for the first iteration. We might have a cached state
1103                  * that matches exactly the target range, in which case no
1104                  * extent state allocations are needed. We'll only know this
1105                  * after locking the tree.
1106                  */
1107                 prealloc = alloc_extent_state(GFP_NOFS);
1108                 if (!prealloc && !first_iteration)
1109                         return -ENOMEM;
1110         }
1111 
1112         spin_lock(&tree->lock);
1113         if (cached_state && *cached_state) {
1114                 state = *cached_state;
1115                 if (state->start <= start && state->end > start &&
1116                     extent_state_in_tree(state)) {
1117                         node = &state->rb_node;
1118                         goto hit_next;
1119                 }
1120         }
1121 
1122         /*
1123          * this search will find all the extents that end after
1124          * our range starts.
1125          */
1126         node = tree_search_for_insert(tree, start, &p, &parent);
1127         if (!node) {
1128                 prealloc = alloc_extent_state_atomic(prealloc);
1129                 if (!prealloc) {
1130                         err = -ENOMEM;
1131                         goto out;
1132                 }
1133                 err = insert_state(tree, prealloc, start, end,
1134                                    &p, &parent, &bits, NULL);
1135                 if (err)
1136                         extent_io_tree_panic(tree, err);
1137                 cache_state(prealloc, cached_state);
1138                 prealloc = NULL;
1139                 goto out;
1140         }
1141         state = rb_entry(node, struct extent_state, rb_node);
1142 hit_next:
1143         last_start = state->start;
1144         last_end = state->end;
1145 
1146         /*
1147          * | ---- desired range ---- |
1148          * | state |
1149          *
1150          * Just lock what we found and keep going
1151          */
1152         if (state->start == start && state->end <= end) {
1153                 set_state_bits(tree, state, &bits, NULL);
1154                 cache_state(state, cached_state);
1155                 state = clear_state_bit(tree, state, &clear_bits, 0, NULL);
1156                 if (last_end == (u64)-1)
1157                         goto out;
1158                 start = last_end + 1;
1159                 if (start < end && state && state->start == start &&
1160                     !need_resched())
1161                         goto hit_next;
1162                 goto search_again;
1163         }
1164 
1165         /*
1166          *     | ---- desired range ---- |
1167          * | state |
1168          *   or
1169          * | ------------- state -------------- |
1170          *
1171          * We need to split the extent we found, and may flip bits on
1172          * second half.
1173          *
1174          * If the extent we found extends past our
1175          * range, we just split and search again.  It'll get split
1176          * again the next time though.
1177          *
1178          * If the extent we found is inside our range, we set the
1179          * desired bit on it.
1180          */
1181         if (state->start < start) {
1182                 prealloc = alloc_extent_state_atomic(prealloc);
1183                 if (!prealloc) {
1184                         err = -ENOMEM;
1185                         goto out;
1186                 }
1187                 err = split_state(tree, state, prealloc, start);
1188                 if (err)
1189                         extent_io_tree_panic(tree, err);
1190                 prealloc = NULL;
1191                 if (err)
1192                         goto out;
1193                 if (state->end <= end) {
1194                         set_state_bits(tree, state, &bits, NULL);
1195                         cache_state(state, cached_state);
1196                         state = clear_state_bit(tree, state, &clear_bits, 0,
1197                                                 NULL);
1198                         if (last_end == (u64)-1)
1199                                 goto out;
1200                         start = last_end + 1;
1201                         if (start < end && state && state->start == start &&
1202                             !need_resched())
1203                                 goto hit_next;
1204                 }
1205                 goto search_again;
1206         }
1207         /*
1208          * | ---- desired range ---- |
1209          *     | state | or               | state |
1210          *
1211          * There's a hole, we need to insert something in it and
1212          * ignore the extent we found.
1213          */
1214         if (state->start > start) {
1215                 u64 this_end;
1216                 if (end < last_start)
1217                         this_end = end;
1218                 else
1219                         this_end = last_start - 1;
1220 
1221                 prealloc = alloc_extent_state_atomic(prealloc);
1222                 if (!prealloc) {
1223                         err = -ENOMEM;
1224                         goto out;
1225                 }
1226 
1227                 /*
1228                  * Avoid to free 'prealloc' if it can be merged with
1229                  * the later extent.
1230                  */
1231                 err = insert_state(tree, prealloc, start, this_end,
1232                                    NULL, NULL, &bits, NULL);
1233                 if (err)
1234                         extent_io_tree_panic(tree, err);
1235                 cache_state(prealloc, cached_state);
1236                 prealloc = NULL;
1237                 start = this_end + 1;
1238                 goto search_again;
1239         }
1240         /*
1241          * | ---- desired range ---- |
1242          *                        | state |
1243          * We need to split the extent, and set the bit
1244          * on the first half
1245          */
1246         if (state->start <= end && state->end > end) {
1247                 prealloc = alloc_extent_state_atomic(prealloc);
1248                 if (!prealloc) {
1249                         err = -ENOMEM;
1250                         goto out;
1251                 }
1252 
1253                 err = split_state(tree, state, prealloc, end + 1);
1254                 if (err)
1255                         extent_io_tree_panic(tree, err);
1256 
1257                 set_state_bits(tree, prealloc, &bits, NULL);
1258                 cache_state(prealloc, cached_state);
1259                 clear_state_bit(tree, prealloc, &clear_bits, 0, NULL);
1260                 prealloc = NULL;
1261                 goto out;
1262         }
1263 
1264 search_again:
1265         if (start > end)
1266                 goto out;
1267         spin_unlock(&tree->lock);
1268         cond_resched();
1269         first_iteration = false;
1270         goto again;
1271 
1272 out:
1273         spin_unlock(&tree->lock);
1274         if (prealloc)
1275                 free_extent_state(prealloc);
1276 
1277         return err;
1278 }
1279 
1280 /* wrappers around set/clear extent bit */
1281 int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1282                            unsigned bits, struct extent_changeset *changeset)
1283 {
1284         /*
1285          * We don't support EXTENT_LOCKED yet, as current changeset will
1286          * record any bits changed, so for EXTENT_LOCKED case, it will
1287          * either fail with -EEXIST or changeset will record the whole
1288          * range.
1289          */
1290         BUG_ON(bits & EXTENT_LOCKED);
1291 
1292         return __set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS,
1293                                 changeset);
1294 }
1295 
1296 int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
1297                      unsigned bits, int wake, int delete,
1298                      struct extent_state **cached, gfp_t mask)
1299 {
1300         return __clear_extent_bit(tree, start, end, bits, wake, delete,
1301                                   cached, mask, NULL);
1302 }
1303 
1304 int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1305                 unsigned bits, struct extent_changeset *changeset)
1306 {
1307         /*
1308          * Don't support EXTENT_LOCKED case, same reason as
1309          * set_record_extent_bits().
1310          */
1311         BUG_ON(bits & EXTENT_LOCKED);
1312 
1313         return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS,
1314                                   changeset);
1315 }
1316 
1317 /*
1318  * either insert or lock state struct between start and end use mask to tell
1319  * us if waiting is desired.
1320  */
1321 int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end,
1322                      struct extent_state **cached_state)
1323 {
1324         int err;
1325         u64 failed_start;
1326 
1327         while (1) {
1328                 err = __set_extent_bit(tree, start, end, EXTENT_LOCKED,
1329                                        EXTENT_LOCKED, &failed_start,
1330                                        cached_state, GFP_NOFS, NULL);
1331                 if (err == -EEXIST) {
1332                         wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
1333                         start = failed_start;
1334                 } else
1335                         break;
1336                 WARN_ON(start > end);
1337         }
1338         return err;
1339 }
1340 
1341 int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end)
1342 {
1343         int err;
1344         u64 failed_start;
1345 
1346         err = __set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED,
1347                                &failed_start, NULL, GFP_NOFS, NULL);
1348         if (err == -EEXIST) {
1349                 if (failed_start > start)
1350                         clear_extent_bit(tree, start, failed_start - 1,
1351                                          EXTENT_LOCKED, 1, 0, NULL, GFP_NOFS);
1352                 return 0;
1353         }
1354         return 1;
1355 }
1356 
1357 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
1358 {
1359         unsigned long index = start >> PAGE_SHIFT;
1360         unsigned long end_index = end >> PAGE_SHIFT;
1361         struct page *page;
1362 
1363         while (index <= end_index) {
1364                 page = find_get_page(inode->i_mapping, index);
1365                 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1366                 clear_page_dirty_for_io(page);
1367                 put_page(page);
1368                 index++;
1369         }
1370 }
1371 
1372 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
1373 {
1374         unsigned long index = start >> PAGE_SHIFT;
1375         unsigned long end_index = end >> PAGE_SHIFT;
1376         struct page *page;
1377 
1378         while (index <= end_index) {
1379                 page = find_get_page(inode->i_mapping, index);
1380                 BUG_ON(!page); /* Pages should be in the extent_io_tree */
1381                 __set_page_dirty_nobuffers(page);
1382                 account_page_redirty(page);
1383                 put_page(page);
1384                 index++;
1385         }
1386 }
1387 
1388 /*
1389  * helper function to set both pages and extents in the tree writeback
1390  */
1391 static void set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
1392 {
1393         tree->ops->set_range_writeback(tree->private_data, start, end);
1394 }
1395 
1396 /* find the first state struct with 'bits' set after 'start', and
1397  * return it.  tree->lock must be held.  NULL will returned if
1398  * nothing was found after 'start'
1399  */
1400 static struct extent_state *
1401 find_first_extent_bit_state(struct extent_io_tree *tree,
1402                             u64 start, unsigned bits)
1403 {
1404         struct rb_node *node;
1405         struct extent_state *state;
1406 
1407         /*
1408          * this search will find all the extents that end after
1409          * our range starts.
1410          */
1411         node = tree_search(tree, start);
1412         if (!node)
1413                 goto out;
1414 
1415         while (1) {
1416                 state = rb_entry(node, struct extent_state, rb_node);
1417                 if (state->end >= start && (state->state & bits))
1418                         return state;
1419 
1420                 node = rb_next(node);
1421                 if (!node)
1422                         break;
1423         }
1424 out:
1425         return NULL;
1426 }
1427 
1428 /*
1429  * find the first offset in the io tree with 'bits' set. zero is
1430  * returned if we find something, and *start_ret and *end_ret are
1431  * set to reflect the state struct that was found.
1432  *
1433  * If nothing was found, 1 is returned. If found something, return 0.
1434  */
1435 int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
1436                           u64 *start_ret, u64 *end_ret, unsigned bits,
1437                           struct extent_state **cached_state)
1438 {
1439         struct extent_state *state;
1440         struct rb_node *n;
1441         int ret = 1;
1442 
1443         spin_lock(&tree->lock);
1444         if (cached_state && *cached_state) {
1445                 state = *cached_state;
1446                 if (state->end == start - 1 && extent_state_in_tree(state)) {
1447                         n = rb_next(&state->rb_node);
1448                         while (n) {
1449                                 state = rb_entry(n, struct extent_state,
1450                                                  rb_node);
1451                                 if (state->state & bits)
1452                                         goto got_it;
1453                                 n = rb_next(n);
1454                         }
1455                         free_extent_state(*cached_state);
1456                         *cached_state = NULL;
1457                         goto out;
1458                 }
1459                 free_extent_state(*cached_state);
1460                 *cached_state = NULL;
1461         }
1462 
1463         state = find_first_extent_bit_state(tree, start, bits);
1464 got_it:
1465         if (state) {
1466                 cache_state_if_flags(state, cached_state, 0);
1467                 *start_ret = state->start;
1468                 *end_ret = state->end;
1469                 ret = 0;
1470         }
1471 out:
1472         spin_unlock(&tree->lock);
1473         return ret;
1474 }
1475 
1476 /*
1477  * find a contiguous range of bytes in the file marked as delalloc, not
1478  * more than 'max_bytes'.  start and end are used to return the range,
1479  *
1480  * 1 is returned if we find something, 0 if nothing was in the tree
1481  */
1482 static noinline u64 find_delalloc_range(struct extent_io_tree *tree,
1483                                         u64 *start, u64 *end, u64 max_bytes,
1484                                         struct extent_state **cached_state)
1485 {
1486         struct rb_node *node;
1487         struct extent_state *state;
1488         u64 cur_start = *start;
1489         u64 found = 0;
1490         u64 total_bytes = 0;
1491 
1492         spin_lock(&tree->lock);
1493 
1494         /*
1495          * this search will find all the extents that end after
1496          * our range starts.
1497          */
1498         node = tree_search(tree, cur_start);
1499         if (!node) {
1500                 if (!found)
1501                         *end = (u64)-1;
1502                 goto out;
1503         }
1504 
1505         while (1) {
1506                 state = rb_entry(node, struct extent_state, rb_node);
1507                 if (found && (state->start != cur_start ||
1508                               (state->state & EXTENT_BOUNDARY))) {
1509                         goto out;
1510                 }
1511                 if (!(state->state & EXTENT_DELALLOC)) {
1512                         if (!found)
1513                                 *end = state->end;
1514                         goto out;
1515                 }
1516                 if (!found) {
1517                         *start = state->start;
1518                         *cached_state = state;
1519                         refcount_inc(&state->refs);
1520                 }
1521                 found++;
1522                 *end = state->end;
1523                 cur_start = state->end + 1;
1524                 node = rb_next(node);
1525                 total_bytes += state->end - state->start + 1;
1526                 if (total_bytes >= max_bytes)
1527                         break;
1528                 if (!node)
1529                         break;
1530         }
1531 out:
1532         spin_unlock(&tree->lock);
1533         return found;
1534 }
1535 
1536 static int __process_pages_contig(struct address_space *mapping,
1537                                   struct page *locked_page,
1538                                   pgoff_t start_index, pgoff_t end_index,
1539                                   unsigned long page_ops, pgoff_t *index_ret);
1540 
1541 static noinline void __unlock_for_delalloc(struct inode *inode,
1542                                            struct page *locked_page,
1543                                            u64 start, u64 end)
1544 {
1545         unsigned long index = start >> PAGE_SHIFT;
1546         unsigned long end_index = end >> PAGE_SHIFT;
1547 
1548         ASSERT(locked_page);
1549         if (index == locked_page->index && end_index == index)
1550                 return;
1551 
1552         __process_pages_contig(inode->i_mapping, locked_page, index, end_index,
1553                                PAGE_UNLOCK, NULL);
1554 }
1555 
1556 static noinline int lock_delalloc_pages(struct inode *inode,
1557                                         struct page *locked_page,
1558                                         u64 delalloc_start,
1559                                         u64 delalloc_end)
1560 {
1561         unsigned long index = delalloc_start >> PAGE_SHIFT;
1562         unsigned long index_ret = index;
1563         unsigned long end_index = delalloc_end >> PAGE_SHIFT;
1564         int ret;
1565 
1566         ASSERT(locked_page);
1567         if (index == locked_page->index && index == end_index)
1568                 return 0;
1569 
1570         ret = __process_pages_contig(inode->i_mapping, locked_page, index,
1571                                      end_index, PAGE_LOCK, &index_ret);
1572         if (ret == -EAGAIN)
1573                 __unlock_for_delalloc(inode, locked_page, delalloc_start,
1574                                       (u64)index_ret << PAGE_SHIFT);
1575         return ret;
1576 }
1577 
1578 /*
1579  * find a contiguous range of bytes in the file marked as delalloc, not
1580  * more than 'max_bytes'.  start and end are used to return the range,
1581  *
1582  * 1 is returned if we find something, 0 if nothing was in the tree
1583  */
1584 STATIC u64 find_lock_delalloc_range(struct inode *inode,
1585                                     struct extent_io_tree *tree,
1586                                     struct page *locked_page, u64 *start,
1587                                     u64 *end, u64 max_bytes)
1588 {
1589         u64 delalloc_start;
1590         u64 delalloc_end;
1591         u64 found;
1592         struct extent_state *cached_state = NULL;
1593         int ret;
1594         int loops = 0;
1595 
1596 again:
1597         /* step one, find a bunch of delalloc bytes starting at start */
1598         delalloc_start = *start;
1599         delalloc_end = 0;
1600         found = find_delalloc_range(tree, &delalloc_start, &delalloc_end,
1601                                     max_bytes, &cached_state);
1602         if (!found || delalloc_end <= *start) {
1603                 *start = delalloc_start;
1604                 *end = delalloc_end;
1605                 free_extent_state(cached_state);
1606                 return 0;
1607         }
1608 
1609         /*
1610          * start comes from the offset of locked_page.  We have to lock
1611          * pages in order, so we can't process delalloc bytes before
1612          * locked_page
1613          */
1614         if (delalloc_start < *start)
1615                 delalloc_start = *start;
1616 
1617         /*
1618          * make sure to limit the number of pages we try to lock down
1619          */
1620         if (delalloc_end + 1 - delalloc_start > max_bytes)
1621                 delalloc_end = delalloc_start + max_bytes - 1;
1622 
1623         /* step two, lock all the pages after the page that has start */
1624         ret = lock_delalloc_pages(inode, locked_page,
1625                                   delalloc_start, delalloc_end);
1626         if (ret == -EAGAIN) {
1627                 /* some of the pages are gone, lets avoid looping by
1628                  * shortening the size of the delalloc range we're searching
1629                  */
1630                 free_extent_state(cached_state);
1631                 cached_state = NULL;
1632                 if (!loops) {
1633                         max_bytes = PAGE_SIZE;
1634                         loops = 1;
1635                         goto again;
1636                 } else {
1637                         found = 0;
1638                         goto out_failed;
1639                 }
1640         }
1641         BUG_ON(ret); /* Only valid values are 0 and -EAGAIN */
1642 
1643         /* step three, lock the state bits for the whole range */
1644         lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state);
1645 
1646         /* then test to make sure it is all still delalloc */
1647         ret = test_range_bit(tree, delalloc_start, delalloc_end,
1648                              EXTENT_DELALLOC, 1, cached_state);
1649         if (!ret) {
1650                 unlock_extent_cached(tree, delalloc_start, delalloc_end,
1651                                      &cached_state, GFP_NOFS);
1652                 __unlock_for_delalloc(inode, locked_page,
1653                               delalloc_start, delalloc_end);
1654                 cond_resched();
1655                 goto again;
1656         }
1657         free_extent_state(cached_state);
1658         *start = delalloc_start;
1659         *end = delalloc_end;
1660 out_failed:
1661         return found;
1662 }
1663 
1664 static int __process_pages_contig(struct address_space *mapping,
1665                                   struct page *locked_page,
1666                                   pgoff_t start_index, pgoff_t end_index,
1667                                   unsigned long page_ops, pgoff_t *index_ret)
1668 {
1669         unsigned long nr_pages = end_index - start_index + 1;
1670         unsigned long pages_locked = 0;
1671         pgoff_t index = start_index;
1672         struct page *pages[16];
1673         unsigned ret;
1674         int err = 0;
1675         int i;
1676 
1677         if (page_ops & PAGE_LOCK) {
1678                 ASSERT(page_ops == PAGE_LOCK);
1679                 ASSERT(index_ret && *index_ret == start_index);
1680         }
1681 
1682         if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0)
1683                 mapping_set_error(mapping, -EIO);
1684 
1685         while (nr_pages > 0) {
1686                 ret = find_get_pages_contig(mapping, index,
1687                                      min_t(unsigned long,
1688                                      nr_pages, ARRAY_SIZE(pages)), pages);
1689                 if (ret == 0) {
1690                         /*
1691                          * Only if we're going to lock these pages,
1692                          * can we find nothing at @index.
1693                          */
1694                         ASSERT(page_ops & PAGE_LOCK);
1695                         err = -EAGAIN;
1696                         goto out;
1697                 }
1698 
1699                 for (i = 0; i < ret; i++) {
1700                         if (page_ops & PAGE_SET_PRIVATE2)
1701                                 SetPagePrivate2(pages[i]);
1702 
1703                         if (pages[i] == locked_page) {
1704                                 put_page(pages[i]);
1705                                 pages_locked++;
1706                                 continue;
1707                         }
1708                         if (page_ops & PAGE_CLEAR_DIRTY)
1709                                 clear_page_dirty_for_io(pages[i]);
1710                         if (page_ops & PAGE_SET_WRITEBACK)
1711                                 set_page_writeback(pages[i]);
1712                         if (page_ops & PAGE_SET_ERROR)
1713                                 SetPageError(pages[i]);
1714                         if (page_ops & PAGE_END_WRITEBACK)
1715                                 end_page_writeback(pages[i]);
1716                         if (page_ops & PAGE_UNLOCK)
1717                                 unlock_page(pages[i]);
1718                         if (page_ops & PAGE_LOCK) {
1719                                 lock_page(pages[i]);
1720                                 if (!PageDirty(pages[i]) ||
1721                                     pages[i]->mapping != mapping) {
1722                                         unlock_page(pages[i]);
1723                                         put_page(pages[i]);
1724                                         err = -EAGAIN;
1725                                         goto out;
1726                                 }
1727                         }
1728                         put_page(pages[i]);
1729                         pages_locked++;
1730                 }
1731                 nr_pages -= ret;
1732                 index += ret;
1733                 cond_resched();
1734         }
1735 out:
1736         if (err && index_ret)
1737                 *index_ret = start_index + pages_locked - 1;
1738         return err;
1739 }
1740 
1741 void extent_clear_unlock_delalloc(struct inode *inode, u64 start, u64 end,
1742                                  u64 delalloc_end, struct page *locked_page,
1743                                  unsigned clear_bits,
1744                                  unsigned long page_ops)
1745 {
1746         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits, 1, 0,
1747                          NULL, GFP_NOFS);
1748 
1749         __process_pages_contig(inode->i_mapping, locked_page,
1750                                start >> PAGE_SHIFT, end >> PAGE_SHIFT,
1751                                page_ops, NULL);
1752 }
1753 
1754 /*
1755  * count the number of bytes in the tree that have a given bit(s)
1756  * set.  This can be fairly slow, except for EXTENT_DIRTY which is
1757  * cached.  The total number found is returned.
1758  */
1759 u64 count_range_bits(struct extent_io_tree *tree,
1760                      u64 *start, u64 search_end, u64 max_bytes,
1761                      unsigned bits, int contig)
1762 {
1763         struct rb_node *node;
1764         struct extent_state *state;
1765         u64 cur_start = *start;
1766         u64 total_bytes = 0;
1767         u64 last = 0;
1768         int found = 0;
1769 
1770         if (WARN_ON(search_end <= cur_start))
1771                 return 0;
1772 
1773         spin_lock(&tree->lock);
1774         if (cur_start == 0 && bits == EXTENT_DIRTY) {
1775                 total_bytes = tree->dirty_bytes;
1776                 goto out;
1777         }
1778         /*
1779          * this search will find all the extents that end after
1780          * our range starts.
1781          */
1782         node = tree_search(tree, cur_start);
1783         if (!node)
1784                 goto out;
1785 
1786         while (1) {
1787                 state = rb_entry(node, struct extent_state, rb_node);
1788                 if (state->start > search_end)
1789                         break;
1790                 if (contig && found && state->start > last + 1)
1791                         break;
1792                 if (state->end >= cur_start && (state->state & bits) == bits) {
1793                         total_bytes += min(search_end, state->end) + 1 -
1794                                        max(cur_start, state->start);
1795                         if (total_bytes >= max_bytes)
1796                                 break;
1797                         if (!found) {
1798                                 *start = max(cur_start, state->start);
1799                                 found = 1;
1800                         }
1801                         last = state->end;
1802                 } else if (contig && found) {
1803                         break;
1804                 }
1805                 node = rb_next(node);
1806                 if (!node)
1807                         break;
1808         }
1809 out:
1810         spin_unlock(&tree->lock);
1811         return total_bytes;
1812 }
1813 
1814 /*
1815  * set the private field for a given byte offset in the tree.  If there isn't
1816  * an extent_state there already, this does nothing.
1817  */
1818 static noinline int set_state_failrec(struct extent_io_tree *tree, u64 start,
1819                 struct io_failure_record *failrec)
1820 {
1821         struct rb_node *node;
1822         struct extent_state *state;
1823         int ret = 0;
1824 
1825         spin_lock(&tree->lock);
1826         /*
1827          * this search will find all the extents that end after
1828          * our range starts.
1829          */
1830         node = tree_search(tree, start);
1831         if (!node) {
1832                 ret = -ENOENT;
1833                 goto out;
1834         }
1835         state = rb_entry(node, struct extent_state, rb_node);
1836         if (state->start != start) {
1837                 ret = -ENOENT;
1838                 goto out;
1839         }
1840         state->failrec = failrec;
1841 out:
1842         spin_unlock(&tree->lock);
1843         return ret;
1844 }
1845 
1846 static noinline int get_state_failrec(struct extent_io_tree *tree, u64 start,
1847                 struct io_failure_record **failrec)
1848 {
1849         struct rb_node *node;
1850         struct extent_state *state;
1851         int ret = 0;
1852 
1853         spin_lock(&tree->lock);
1854         /*
1855          * this search will find all the extents that end after
1856          * our range starts.
1857          */
1858         node = tree_search(tree, start);
1859         if (!node) {
1860                 ret = -ENOENT;
1861                 goto out;
1862         }
1863         state = rb_entry(node, struct extent_state, rb_node);
1864         if (state->start != start) {
1865                 ret = -ENOENT;
1866                 goto out;
1867         }
1868         *failrec = state->failrec;
1869 out:
1870         spin_unlock(&tree->lock);
1871         return ret;
1872 }
1873 
1874 /*
1875  * searches a range in the state tree for a given mask.
1876  * If 'filled' == 1, this returns 1 only if every extent in the tree
1877  * has the bits set.  Otherwise, 1 is returned if any bit in the
1878  * range is found set.
1879  */
1880 int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end,
1881                    unsigned bits, int filled, struct extent_state *cached)
1882 {
1883         struct extent_state *state = NULL;
1884         struct rb_node *node;
1885         int bitset = 0;
1886 
1887         spin_lock(&tree->lock);
1888         if (cached && extent_state_in_tree(cached) && cached->start <= start &&
1889             cached->end > start)
1890                 node = &cached->rb_node;
1891         else
1892                 node = tree_search(tree, start);
1893         while (node && start <= end) {
1894                 state = rb_entry(node, struct extent_state, rb_node);
1895 
1896                 if (filled && state->start > start) {
1897                         bitset = 0;
1898                         break;
1899                 }
1900 
1901                 if (state->start > end)
1902                         break;
1903 
1904                 if (state->state & bits) {
1905                         bitset = 1;
1906                         if (!filled)
1907                                 break;
1908                 } else if (filled) {
1909                         bitset = 0;
1910                         break;
1911                 }
1912 
1913                 if (state->end == (u64)-1)
1914                         break;
1915 
1916                 start = state->end + 1;
1917                 if (start > end)
1918                         break;
1919                 node = rb_next(node);
1920                 if (!node) {
1921                         if (filled)
1922                                 bitset = 0;
1923                         break;
1924                 }
1925         }
1926         spin_unlock(&tree->lock);
1927         return bitset;
1928 }
1929 
1930 /*
1931  * helper function to set a given page up to date if all the
1932  * extents in the tree for that page are up to date
1933  */
1934 static void check_page_uptodate(struct extent_io_tree *tree, struct page *page)
1935 {
1936         u64 start = page_offset(page);
1937         u64 end = start + PAGE_SIZE - 1;
1938         if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL))
1939                 SetPageUptodate(page);
1940 }
1941 
1942 int free_io_failure(struct extent_io_tree *failure_tree,
1943                     struct extent_io_tree *io_tree,
1944                     struct io_failure_record *rec)
1945 {
1946         int ret;
1947         int err = 0;
1948 
1949         set_state_failrec(failure_tree, rec->start, NULL);
1950         ret = clear_extent_bits(failure_tree, rec->start,
1951                                 rec->start + rec->len - 1,
1952                                 EXTENT_LOCKED | EXTENT_DIRTY);
1953         if (ret)
1954                 err = ret;
1955 
1956         ret = clear_extent_bits(io_tree, rec->start,
1957                                 rec->start + rec->len - 1,
1958                                 EXTENT_DAMAGED);
1959         if (ret && !err)
1960                 err = ret;
1961 
1962         kfree(rec);
1963         return err;
1964 }
1965 
1966 /*
1967  * this bypasses the standard btrfs submit functions deliberately, as
1968  * the standard behavior is to write all copies in a raid setup. here we only
1969  * want to write the one bad copy. so we do the mapping for ourselves and issue
1970  * submit_bio directly.
1971  * to avoid any synchronization issues, wait for the data after writing, which
1972  * actually prevents the read that triggered the error from finishing.
1973  * currently, there can be no more than two copies of every data bit. thus,
1974  * exactly one rewrite is required.
1975  */
1976 int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
1977                       u64 length, u64 logical, struct page *page,
1978                       unsigned int pg_offset, int mirror_num)
1979 {
1980         struct bio *bio;
1981         struct btrfs_device *dev;
1982         u64 map_length = 0;
1983         u64 sector;
1984         struct btrfs_bio *bbio = NULL;
1985         int ret;
1986 
1987         ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
1988         BUG_ON(!mirror_num);
1989 
1990         bio = btrfs_io_bio_alloc(1);
1991         bio->bi_iter.bi_size = 0;
1992         map_length = length;
1993 
1994         /*
1995          * Avoid races with device replace and make sure our bbio has devices
1996          * associated to its stripes that don't go away while we are doing the
1997          * read repair operation.
1998          */
1999         btrfs_bio_counter_inc_blocked(fs_info);
2000         if (btrfs_is_parity_mirror(fs_info, logical, length)) {
2001                 /*
2002                  * Note that we don't use BTRFS_MAP_WRITE because it's supposed
2003                  * to update all raid stripes, but here we just want to correct
2004                  * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
2005                  * stripe's dev and sector.
2006                  */
2007                 ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
2008                                       &map_length, &bbio, 0);
2009                 if (ret) {
2010                         btrfs_bio_counter_dec(fs_info);
2011                         bio_put(bio);
2012                         return -EIO;
2013                 }
2014                 ASSERT(bbio->mirror_num == 1);
2015         } else {
2016                 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
2017                                       &map_length, &bbio, mirror_num);
2018                 if (ret) {
2019                         btrfs_bio_counter_dec(fs_info);
2020                         bio_put(bio);
2021                         return -EIO;
2022                 }
2023                 BUG_ON(mirror_num != bbio->mirror_num);
2024         }
2025 
2026         sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9;
2027         bio->bi_iter.bi_sector = sector;
2028         dev = bbio->stripes[bbio->mirror_num - 1].dev;
2029         btrfs_put_bbio(bbio);
2030         if (!dev || !dev->bdev || !dev->writeable) {
2031                 btrfs_bio_counter_dec(fs_info);
2032                 bio_put(bio);
2033                 return -EIO;
2034         }
2035         bio_set_dev(bio, dev->bdev);
2036         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
2037         bio_add_page(bio, page, length, pg_offset);
2038 
2039         if (btrfsic_submit_bio_wait(bio)) {
2040                 /* try to remap that extent elsewhere? */
2041                 btrfs_bio_counter_dec(fs_info);
2042                 bio_put(bio);
2043                 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
2044                 return -EIO;
2045         }
2046 
2047         btrfs_info_rl_in_rcu(fs_info,
2048                 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
2049                                   ino, start,
2050                                   rcu_str_deref(dev->name), sector);
2051         btrfs_bio_counter_dec(fs_info);
2052         bio_put(bio);
2053         return 0;
2054 }
2055 
2056 int repair_eb_io_failure(struct btrfs_fs_info *fs_info,
2057                          struct extent_buffer *eb, int mirror_num)
2058 {
2059         u64 start = eb->start;
2060         unsigned long i, num_pages = num_extent_pages(eb->start, eb->len);
2061         int ret = 0;
2062 
2063         if (sb_rdonly(fs_info->sb))
2064                 return -EROFS;
2065 
2066         for (i = 0; i < num_pages; i++) {
2067                 struct page *p = eb->pages[i];
2068 
2069                 ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p,
2070                                         start - page_offset(p), mirror_num);
2071                 if (ret)
2072                         break;
2073                 start += PAGE_SIZE;
2074         }
2075 
2076         return ret;
2077 }
2078 
2079 /*
2080  * each time an IO finishes, we do a fast check in the IO failure tree
2081  * to see if we need to process or clean up an io_failure_record
2082  */
2083 int clean_io_failure(struct btrfs_fs_info *fs_info,
2084                      struct extent_io_tree *failure_tree,
2085                      struct extent_io_tree *io_tree, u64 start,
2086                      struct page *page, u64 ino, unsigned int pg_offset)
2087 {
2088         u64 private;
2089         struct io_failure_record *failrec;
2090         struct extent_state *state;
2091         int num_copies;
2092         int ret;
2093 
2094         private = 0;
2095         ret = count_range_bits(failure_tree, &private, (u64)-1, 1,
2096                                EXTENT_DIRTY, 0);
2097         if (!ret)
2098                 return 0;
2099 
2100         ret = get_state_failrec(failure_tree, start, &failrec);
2101         if (ret)
2102                 return 0;
2103 
2104         BUG_ON(!failrec->this_mirror);
2105 
2106         if (failrec->in_validation) {
2107                 /* there was no real error, just free the record */
2108                 btrfs_debug(fs_info,
2109                         "clean_io_failure: freeing dummy error at %llu",
2110                         failrec->start);
2111                 goto out;
2112         }
2113         if (sb_rdonly(fs_info->sb))
2114                 goto out;
2115 
2116         spin_lock(&io_tree->lock);
2117         state = find_first_extent_bit_state(io_tree,
2118                                             failrec->start,
2119                                             EXTENT_LOCKED);
2120         spin_unlock(&io_tree->lock);
2121 
2122         if (state && state->start <= failrec->start &&
2123             state->end >= failrec->start + failrec->len - 1) {
2124                 num_copies = btrfs_num_copies(fs_info, failrec->logical,
2125                                               failrec->len);
2126                 if (num_copies > 1)  {
2127                         repair_io_failure(fs_info, ino, start, failrec->len,
2128                                           failrec->logical, page, pg_offset,
2129                                           failrec->failed_mirror);
2130                 }
2131         }
2132 
2133 out:
2134         free_io_failure(failure_tree, io_tree, failrec);
2135 
2136         return 0;
2137 }
2138 
2139 /*
2140  * Can be called when
2141  * - hold extent lock
2142  * - under ordered extent
2143  * - the inode is freeing
2144  */
2145 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
2146 {
2147         struct extent_io_tree *failure_tree = &inode->io_failure_tree;
2148         struct io_failure_record *failrec;
2149         struct extent_state *state, *next;
2150 
2151         if (RB_EMPTY_ROOT(&failure_tree->state))
2152                 return;
2153 
2154         spin_lock(&failure_tree->lock);
2155         state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY);
2156         while (state) {
2157                 if (state->start > end)
2158                         break;
2159 
2160                 ASSERT(state->end <= end);
2161 
2162                 next = next_state(state);
2163 
2164                 failrec = state->failrec;
2165                 free_extent_state(state);
2166                 kfree(failrec);
2167 
2168                 state = next;
2169         }
2170         spin_unlock(&failure_tree->lock);
2171 }
2172 
2173 int btrfs_get_io_failure_record(struct inode *inode, u64 start, u64 end,
2174                 struct io_failure_record **failrec_ret)
2175 {
2176         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2177         struct io_failure_record *failrec;
2178         struct extent_map *em;
2179         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2180         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2181         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2182         int ret;
2183         u64 logical;
2184 
2185         ret = get_state_failrec(failure_tree, start, &failrec);
2186         if (ret) {
2187                 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
2188                 if (!failrec)
2189                         return -ENOMEM;
2190 
2191                 failrec->start = start;
2192                 failrec->len = end - start + 1;
2193                 failrec->this_mirror = 0;
2194                 failrec->bio_flags = 0;
2195                 failrec->in_validation = 0;
2196 
2197                 read_lock(&em_tree->lock);
2198                 em = lookup_extent_mapping(em_tree, start, failrec->len);
2199                 if (!em) {
2200                         read_unlock(&em_tree->lock);
2201                         kfree(failrec);
2202                         return -EIO;
2203                 }
2204 
2205                 if (em->start > start || em->start + em->len <= start) {
2206                         free_extent_map(em);
2207                         em = NULL;
2208                 }
2209                 read_unlock(&em_tree->lock);
2210                 if (!em) {
2211                         kfree(failrec);
2212                         return -EIO;
2213                 }
2214 
2215                 logical = start - em->start;
2216                 logical = em->block_start + logical;
2217                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2218                         logical = em->block_start;
2219                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
2220                         extent_set_compress_type(&failrec->bio_flags,
2221                                                  em->compress_type);
2222                 }
2223 
2224                 btrfs_debug(fs_info,
2225                         "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu",
2226                         logical, start, failrec->len);
2227 
2228                 failrec->logical = logical;
2229                 free_extent_map(em);
2230 
2231                 /* set the bits in the private failure tree */
2232                 ret = set_extent_bits(failure_tree, start, end,
2233                                         EXTENT_LOCKED | EXTENT_DIRTY);
2234                 if (ret >= 0)
2235                         ret = set_state_failrec(failure_tree, start, failrec);
2236                 /* set the bits in the inode's tree */
2237                 if (ret >= 0)
2238                         ret = set_extent_bits(tree, start, end, EXTENT_DAMAGED);
2239                 if (ret < 0) {
2240                         kfree(failrec);
2241                         return ret;
2242                 }
2243         } else {
2244                 btrfs_debug(fs_info,
2245                         "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu, validation=%d",
2246                         failrec->logical, failrec->start, failrec->len,
2247                         failrec->in_validation);
2248                 /*
2249                  * when data can be on disk more than twice, add to failrec here
2250                  * (e.g. with a list for failed_mirror) to make
2251                  * clean_io_failure() clean all those errors at once.
2252                  */
2253         }
2254 
2255         *failrec_ret = failrec;
2256 
2257         return 0;
2258 }
2259 
2260 bool btrfs_check_repairable(struct inode *inode, struct bio *failed_bio,
2261                            struct io_failure_record *failrec, int failed_mirror)
2262 {
2263         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2264         int num_copies;
2265 
2266         num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
2267         if (num_copies == 1) {
2268                 /*
2269                  * we only have a single copy of the data, so don't bother with
2270                  * all the retry and error correction code that follows. no
2271                  * matter what the error is, it is very likely to persist.
2272                  */
2273                 btrfs_debug(fs_info,
2274                         "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
2275                         num_copies, failrec->this_mirror, failed_mirror);
2276                 return false;
2277         }
2278 
2279         /*
2280          * there are two premises:
2281          *      a) deliver good data to the caller
2282          *      b) correct the bad sectors on disk
2283          */
2284         if (failed_bio->bi_vcnt > 1) {
2285                 /*
2286                  * to fulfill b), we need to know the exact failing sectors, as
2287                  * we don't want to rewrite any more than the failed ones. thus,
2288                  * we need separate read requests for the failed bio
2289                  *
2290                  * if the following BUG_ON triggers, our validation request got
2291                  * merged. we need separate requests for our algorithm to work.
2292                  */
2293                 BUG_ON(failrec->in_validation);
2294                 failrec->in_validation = 1;
2295                 failrec->this_mirror = failed_mirror;
2296         } else {
2297                 /*
2298                  * we're ready to fulfill a) and b) alongside. get a good copy
2299                  * of the failed sector and if we succeed, we have setup
2300                  * everything for repair_io_failure to do the rest for us.
2301                  */
2302                 if (failrec->in_validation) {
2303                         BUG_ON(failrec->this_mirror != failed_mirror);
2304                         failrec->in_validation = 0;
2305                         failrec->this_mirror = 0;
2306                 }
2307                 failrec->failed_mirror = failed_mirror;
2308                 failrec->this_mirror++;
2309                 if (failrec->this_mirror == failed_mirror)
2310                         failrec->this_mirror++;
2311         }
2312 
2313         if (failrec->this_mirror > num_copies) {
2314                 btrfs_debug(fs_info,
2315                         "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
2316                         num_copies, failrec->this_mirror, failed_mirror);
2317                 return false;
2318         }
2319 
2320         return true;
2321 }
2322 
2323 
2324 struct bio *btrfs_create_repair_bio(struct inode *inode, struct bio *failed_bio,
2325                                     struct io_failure_record *failrec,
2326                                     struct page *page, int pg_offset, int icsum,
2327                                     bio_end_io_t *endio_func, void *data)
2328 {
2329         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2330         struct bio *bio;
2331         struct btrfs_io_bio *btrfs_failed_bio;
2332         struct btrfs_io_bio *btrfs_bio;
2333 
2334         bio = btrfs_io_bio_alloc(1);
2335         bio->bi_end_io = endio_func;
2336         bio->bi_iter.bi_sector = failrec->logical >> 9;
2337         bio_set_dev(bio, fs_info->fs_devices->latest_bdev);
2338         bio->bi_iter.bi_size = 0;
2339         bio->bi_private = data;
2340 
2341         btrfs_failed_bio = btrfs_io_bio(failed_bio);
2342         if (btrfs_failed_bio->csum) {
2343                 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
2344 
2345                 btrfs_bio = btrfs_io_bio(bio);
2346                 btrfs_bio->csum = btrfs_bio->csum_inline;
2347                 icsum *= csum_size;
2348                 memcpy(btrfs_bio->csum, btrfs_failed_bio->csum + icsum,
2349                        csum_size);
2350         }
2351 
2352         bio_add_page(bio, page, failrec->len, pg_offset);
2353 
2354         return bio;
2355 }
2356 
2357 /*
2358  * this is a generic handler for readpage errors (default
2359  * readpage_io_failed_hook). if other copies exist, read those and write back
2360  * good data to the failed position. does not investigate in remapping the
2361  * failed extent elsewhere, hoping the device will be smart enough to do this as
2362  * needed
2363  */
2364 
2365 static int bio_readpage_error(struct bio *failed_bio, u64 phy_offset,
2366                               struct page *page, u64 start, u64 end,
2367                               int failed_mirror)
2368 {
2369         struct io_failure_record *failrec;
2370         struct inode *inode = page->mapping->host;
2371         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
2372         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
2373         struct bio *bio;
2374         int read_mode = 0;
2375         blk_status_t status;
2376         int ret;
2377 
2378         BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
2379 
2380         ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
2381         if (ret)
2382                 return ret;
2383 
2384         if (!btrfs_check_repairable(inode, failed_bio, failrec,
2385                                     failed_mirror)) {
2386                 free_io_failure(failure_tree, tree, failrec);
2387                 return -EIO;
2388         }
2389 
2390         if (failed_bio->bi_vcnt > 1)
2391                 read_mode |= REQ_FAILFAST_DEV;
2392 
2393         phy_offset >>= inode->i_sb->s_blocksize_bits;
2394         bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
2395                                       start - page_offset(page),
2396                                       (int)phy_offset, failed_bio->bi_end_io,
2397                                       NULL);
2398         bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
2399 
2400         btrfs_debug(btrfs_sb(inode->i_sb),
2401                 "Repair Read Error: submitting new read[%#x] to this_mirror=%d, in_validation=%d",
2402                 read_mode, failrec->this_mirror, failrec->in_validation);
2403 
2404         status = tree->ops->submit_bio_hook(tree->private_data, bio, failrec->this_mirror,
2405                                          failrec->bio_flags, 0);
2406         if (status) {
2407                 free_io_failure(failure_tree, tree, failrec);
2408                 bio_put(bio);
2409                 ret = blk_status_to_errno(status);
2410         }
2411 
2412         return ret;
2413 }
2414 
2415 /* lots and lots of room for performance fixes in the end_bio funcs */
2416 
2417 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
2418 {
2419         int uptodate = (err == 0);
2420         struct extent_io_tree *tree;
2421         int ret = 0;
2422 
2423         tree = &BTRFS_I(page->mapping->host)->io_tree;
2424 
2425         if (tree->ops && tree->ops->writepage_end_io_hook)
2426                 tree->ops->writepage_end_io_hook(page, start, end, NULL,
2427                                 uptodate);
2428 
2429         if (!uptodate) {
2430                 ClearPageUptodate(page);
2431                 SetPageError(page);
2432                 ret = err < 0 ? err : -EIO;
2433                 mapping_set_error(page->mapping, ret);
2434         }
2435 }
2436 
2437 /*
2438  * after a writepage IO is done, we need to:
2439  * clear the uptodate bits on error
2440  * clear the writeback bits in the extent tree for this IO
2441  * end_page_writeback if the page has no more pending IO
2442  *
2443  * Scheduling is not allowed, so the extent state tree is expected
2444  * to have one and only one object corresponding to this IO.
2445  */
2446 static void end_bio_extent_writepage(struct bio *bio)
2447 {
2448         int error = blk_status_to_errno(bio->bi_status);
2449         struct bio_vec *bvec;
2450         u64 start;
2451         u64 end;
2452         int i;
2453 
2454         ASSERT(!bio_flagged(bio, BIO_CLONED));
2455         bio_for_each_segment_all(bvec, bio, i) {
2456                 struct page *page = bvec->bv_page;
2457                 struct inode *inode = page->mapping->host;
2458                 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2459 
2460                 /* We always issue full-page reads, but if some block
2461                  * in a page fails to read, blk_update_request() will
2462                  * advance bv_offset and adjust bv_len to compensate.
2463                  * Print a warning for nonzero offsets, and an error
2464                  * if they don't add up to a full page.  */
2465                 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2466                         if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2467                                 btrfs_err(fs_info,
2468                                    "partial page write in btrfs with offset %u and length %u",
2469                                         bvec->bv_offset, bvec->bv_len);
2470                         else
2471                                 btrfs_info(fs_info,
2472                                    "incomplete page write in btrfs with offset %u and length %u",
2473                                         bvec->bv_offset, bvec->bv_len);
2474                 }
2475 
2476                 start = page_offset(page);
2477                 end = start + bvec->bv_offset + bvec->bv_len - 1;
2478 
2479                 end_extent_writepage(page, error, start, end);
2480                 end_page_writeback(page);
2481         }
2482 
2483         bio_put(bio);
2484 }
2485 
2486 static void
2487 endio_readpage_release_extent(struct extent_io_tree *tree, u64 start, u64 len,
2488                               int uptodate)
2489 {
2490         struct extent_state *cached = NULL;
2491         u64 end = start + len - 1;
2492 
2493         if (uptodate && tree->track_uptodate)
2494                 set_extent_uptodate(tree, start, end, &cached, GFP_ATOMIC);
2495         unlock_extent_cached(tree, start, end, &cached, GFP_ATOMIC);
2496 }
2497 
2498 /*
2499  * after a readpage IO is done, we need to:
2500  * clear the uptodate bits on error
2501  * set the uptodate bits if things worked
2502  * set the page up to date if all extents in the tree are uptodate
2503  * clear the lock bit in the extent tree
2504  * unlock the page if there are no other extents locked for it
2505  *
2506  * Scheduling is not allowed, so the extent state tree is expected
2507  * to have one and only one object corresponding to this IO.
2508  */
2509 static void end_bio_extent_readpage(struct bio *bio)
2510 {
2511         struct bio_vec *bvec;
2512         int uptodate = !bio->bi_status;
2513         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
2514         struct extent_io_tree *tree, *failure_tree;
2515         u64 offset = 0;
2516         u64 start;
2517         u64 end;
2518         u64 len;
2519         u64 extent_start = 0;
2520         u64 extent_len = 0;
2521         int mirror;
2522         int ret;
2523         int i;
2524 
2525         ASSERT(!bio_flagged(bio, BIO_CLONED));
2526         bio_for_each_segment_all(bvec, bio, i) {
2527                 struct page *page = bvec->bv_page;
2528                 struct inode *inode = page->mapping->host;
2529                 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2530 
2531                 btrfs_debug(fs_info,
2532                         "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
2533                         (u64)bio->bi_iter.bi_sector, bio->bi_status,
2534                         io_bio->mirror_num);
2535                 tree = &BTRFS_I(inode)->io_tree;
2536                 failure_tree = &BTRFS_I(inode)->io_failure_tree;
2537 
2538                 /* We always issue full-page reads, but if some block
2539                  * in a page fails to read, blk_update_request() will
2540                  * advance bv_offset and adjust bv_len to compensate.
2541                  * Print a warning for nonzero offsets, and an error
2542                  * if they don't add up to a full page.  */
2543                 if (bvec->bv_offset || bvec->bv_len != PAGE_SIZE) {
2544                         if (bvec->bv_offset + bvec->bv_len != PAGE_SIZE)
2545                                 btrfs_err(fs_info,
2546                                         "partial page read in btrfs with offset %u and length %u",
2547                                         bvec->bv_offset, bvec->bv_len);
2548                         else
2549                                 btrfs_info(fs_info,
2550                                         "incomplete page read in btrfs with offset %u and length %u",
2551                                         bvec->bv_offset, bvec->bv_len);
2552                 }
2553 
2554                 start = page_offset(page);
2555                 end = start + bvec->bv_offset + bvec->bv_len - 1;
2556                 len = bvec->bv_len;
2557 
2558                 mirror = io_bio->mirror_num;
2559                 if (likely(uptodate && tree->ops)) {
2560                         ret = tree->ops->readpage_end_io_hook(io_bio, offset,
2561                                                               page, start, end,
2562                                                               mirror);
2563                         if (ret)
2564                                 uptodate = 0;
2565                         else
2566                                 clean_io_failure(BTRFS_I(inode)->root->fs_info,
2567                                                  failure_tree, tree, start,
2568                                                  page,
2569                                                  btrfs_ino(BTRFS_I(inode)), 0);
2570                 }
2571 
2572                 if (likely(uptodate))
2573                         goto readpage_ok;
2574 
2575                 if (tree->ops) {
2576                         ret = tree->ops->readpage_io_failed_hook(page, mirror);
2577                         if (ret == -EAGAIN) {
2578                                 /*
2579                                  * Data inode's readpage_io_failed_hook() always
2580                                  * returns -EAGAIN.
2581                                  *
2582                                  * The generic bio_readpage_error handles errors
2583                                  * the following way: If possible, new read
2584                                  * requests are created and submitted and will
2585                                  * end up in end_bio_extent_readpage as well (if
2586                                  * we're lucky, not in the !uptodate case). In
2587                                  * that case it returns 0 and we just go on with
2588                                  * the next page in our bio. If it can't handle
2589                                  * the error it will return -EIO and we remain
2590                                  * responsible for that page.
2591                                  */
2592                                 ret = bio_readpage_error(bio, offset, page,
2593                                                          start, end, mirror);
2594                                 if (ret == 0) {
2595                                         uptodate = !bio->bi_status;
2596                                         offset += len;
2597                                         continue;
2598                                 }
2599                         }
2600 
2601                         /*
2602                          * metadata's readpage_io_failed_hook() always returns
2603                          * -EIO and fixes nothing.  -EIO is also returned if
2604                          * data inode error could not be fixed.
2605                          */
2606                         ASSERT(ret == -EIO);
2607                 }
2608 readpage_ok:
2609                 if (likely(uptodate)) {
2610                         loff_t i_size = i_size_read(inode);
2611                         pgoff_t end_index = i_size >> PAGE_SHIFT;
2612                         unsigned off;
2613 
2614                         /* Zero out the end if this page straddles i_size */
2615                         off = i_size & (PAGE_SIZE-1);
2616                         if (page->index == end_index && off)
2617                                 zero_user_segment(page, off, PAGE_SIZE);
2618                         SetPageUptodate(page);
2619                 } else {
2620                         ClearPageUptodate(page);
2621                         SetPageError(page);
2622                 }
2623                 unlock_page(page);
2624                 offset += len;
2625 
2626                 if (unlikely(!uptodate)) {
2627                         if (extent_len) {
2628                                 endio_readpage_release_extent(tree,
2629                                                               extent_start,
2630                                                               extent_len, 1);
2631                                 extent_start = 0;
2632                                 extent_len = 0;
2633                         }
2634                         endio_readpage_release_extent(tree, start,
2635                                                       end - start + 1, 0);
2636                 } else if (!extent_len) {
2637                         extent_start = start;
2638                         extent_len = end + 1 - start;
2639                 } else if (extent_start + extent_len == start) {
2640                         extent_len += end + 1 - start;
2641                 } else {
2642                         endio_readpage_release_extent(tree, extent_start,
2643                                                       extent_len, uptodate);
2644                         extent_start = start;
2645                         extent_len = end + 1 - start;
2646                 }
2647         }
2648 
2649         if (extent_len)
2650                 endio_readpage_release_extent(tree, extent_start, extent_len,
2651                                               uptodate);
2652         if (io_bio->end_io)
2653                 io_bio->end_io(io_bio, blk_status_to_errno(bio->bi_status));
2654         bio_put(bio);
2655 }
2656 
2657 /*
2658  * Initialize the members up to but not including 'bio'. Use after allocating a
2659  * new bio by bio_alloc_bioset as it does not initialize the bytes outside of
2660  * 'bio' because use of __GFP_ZERO is not supported.
2661  */
2662 static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio)
2663 {
2664         memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio));
2665 }
2666 
2667 /*
2668  * The following helpers allocate a bio. As it's backed by a bioset, it'll
2669  * never fail.  We're returning a bio right now but you can call btrfs_io_bio
2670  * for the appropriate container_of magic
2671  */
2672 struct bio *btrfs_bio_alloc(struct block_device *bdev, u64 first_byte)
2673 {
2674         struct bio *bio;
2675 
2676         bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, btrfs_bioset);
2677         bio_set_dev(bio, bdev);
2678         bio->bi_iter.bi_sector = first_byte >> 9;
2679         btrfs_io_bio_init(btrfs_io_bio(bio));
2680         return bio;
2681 }
2682 
2683 struct bio *btrfs_bio_clone(struct bio *bio)
2684 {
2685         struct btrfs_io_bio *btrfs_bio;
2686         struct bio *new;
2687 
2688         /* Bio allocation backed by a bioset does not fail */
2689         new = bio_clone_fast(bio, GFP_NOFS, btrfs_bioset);
2690         btrfs_bio = btrfs_io_bio(new);
2691         btrfs_io_bio_init(btrfs_bio);
2692         btrfs_bio->iter = bio->bi_iter;
2693         return new;
2694 }
2695 
2696 struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs)
2697 {
2698         struct bio *bio;
2699 
2700         /* Bio allocation backed by a bioset does not fail */
2701         bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, btrfs_bioset);
2702         btrfs_io_bio_init(btrfs_io_bio(bio));
2703         return bio;
2704 }
2705 
2706 struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size)
2707 {
2708         struct bio *bio;
2709         struct btrfs_io_bio *btrfs_bio;
2710 
2711         /* this will never fail when it's backed by a bioset */
2712         bio = bio_clone_fast(orig, GFP_NOFS, btrfs_bioset);
2713         ASSERT(bio);
2714 
2715         btrfs_bio = btrfs_io_bio(bio);
2716         btrfs_io_bio_init(btrfs_bio);
2717 
2718         bio_trim(bio, offset >> 9, size >> 9);
2719         btrfs_bio->iter = bio->bi_iter;
2720         return bio;
2721 }
2722 
2723 static int __must_check submit_one_bio(struct bio *bio, int mirror_num,
2724                                        unsigned long bio_flags)
2725 {
2726         blk_status_t ret = 0;
2727         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
2728         struct page *page = bvec->bv_page;
2729         struct extent_io_tree *tree = bio->bi_private;
2730         u64 start;
2731 
2732         start = page_offset(page) + bvec->bv_offset;
2733 
2734         bio->bi_private = NULL;
2735         bio_get(bio);
2736 
2737         if (tree->ops)
2738                 ret = tree->ops->submit_bio_hook(tree->private_data, bio,
2739                                            mirror_num, bio_flags, start);
2740         else
2741                 btrfsic_submit_bio(bio);
2742 
2743         bio_put(bio);
2744         return blk_status_to_errno(ret);
2745 }
2746 
2747 static int merge_bio(struct extent_io_tree *tree, struct page *page,
2748                      unsigned long offset, size_t size, struct bio *bio,
2749                      unsigned long bio_flags)
2750 {
2751         int ret = 0;
2752         if (tree->ops)
2753                 ret = tree->ops->merge_bio_hook(page, offset, size, bio,
2754                                                 bio_flags);
2755         return ret;
2756 
2757 }
2758 
2759 /*
2760  * @opf:        bio REQ_OP_* and REQ_* flags as one value
2761  */
2762 static int submit_extent_page(unsigned int opf, struct extent_io_tree *tree,
2763                               struct writeback_control *wbc,
2764                               struct page *page, u64 offset,
2765                               size_t size, unsigned long pg_offset,
2766                               struct block_device *bdev,
2767                               struct bio **bio_ret,
2768                               bio_end_io_t end_io_func,
2769                               int mirror_num,
2770                               unsigned long prev_bio_flags,
2771                               unsigned long bio_flags,
2772                               bool force_bio_submit)
2773 {
2774         int ret = 0;
2775         struct bio *bio;
2776         int contig = 0;
2777         int old_compressed = prev_bio_flags & EXTENT_BIO_COMPRESSED;
2778         size_t page_size = min_t(size_t, size, PAGE_SIZE);
2779         sector_t sector = offset >> 9;
2780 
2781         if (bio_ret && *bio_ret) {
2782                 bio = *bio_ret;
2783                 if (old_compressed)
2784                         contig = bio->bi_iter.bi_sector == sector;
2785                 else
2786                         contig = bio_end_sector(bio) == sector;
2787 
2788                 if (prev_bio_flags != bio_flags || !contig ||
2789                     force_bio_submit ||
2790                     merge_bio(tree, page, pg_offset, page_size, bio, bio_flags) ||
2791                     bio_add_page(bio, page, page_size, pg_offset) < page_size) {
2792                         ret = submit_one_bio(bio, mirror_num, prev_bio_flags);
2793                         if (ret < 0) {
2794                                 *bio_ret = NULL;
2795                                 return ret;
2796                         }
2797                         bio = NULL;
2798                 } else {
2799                         if (wbc)
2800                                 wbc_account_io(wbc, page, page_size);
2801                         return 0;
2802                 }
2803         }
2804 
2805         bio = btrfs_bio_alloc(bdev, offset);
2806         bio_add_page(bio, page, page_size, pg_offset);
2807         bio->bi_end_io = end_io_func;
2808         bio->bi_private = tree;
2809         bio->bi_write_hint = page->mapping->host->i_write_hint;
2810         bio->bi_opf = opf;
2811         if (wbc) {
2812                 wbc_init_bio(wbc, bio);
2813                 wbc_account_io(wbc, page, page_size);
2814         }
2815 
2816         if (bio_ret)
2817                 *bio_ret = bio;
2818         else
2819                 ret = submit_one_bio(bio, mirror_num, bio_flags);
2820 
2821         return ret;
2822 }
2823 
2824 static void attach_extent_buffer_page(struct extent_buffer *eb,
2825                                       struct page *page)
2826 {
2827         if (!PagePrivate(page)) {
2828                 SetPagePrivate(page);
2829                 get_page(page);
2830                 set_page_private(page, (unsigned long)eb);
2831         } else {
2832                 WARN_ON(page->private != (unsigned long)eb);
2833         }
2834 }
2835 
2836 void set_page_extent_mapped(struct page *page)
2837 {
2838         if (!PagePrivate(page)) {
2839                 SetPagePrivate(page);
2840                 get_page(page);
2841                 set_page_private(page, EXTENT_PAGE_PRIVATE);
2842         }
2843 }
2844 
2845 static struct extent_map *
2846 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
2847                  u64 start, u64 len, get_extent_t *get_extent,
2848                  struct extent_map **em_cached)
2849 {
2850         struct extent_map *em;
2851 
2852         if (em_cached && *em_cached) {
2853                 em = *em_cached;
2854                 if (extent_map_in_tree(em) && start >= em->start &&
2855                     start < extent_map_end(em)) {
2856                         refcount_inc(&em->refs);
2857                         return em;
2858                 }
2859 
2860                 free_extent_map(em);
2861                 *em_cached = NULL;
2862         }
2863 
2864         em = get_extent(BTRFS_I(inode), page, pg_offset, start, len, 0);
2865         if (em_cached && !IS_ERR_OR_NULL(em)) {
2866                 BUG_ON(*em_cached);
2867                 refcount_inc(&em->refs);
2868                 *em_cached = em;
2869         }
2870         return em;
2871 }
2872 /*
2873  * basic readpage implementation.  Locked extent state structs are inserted
2874  * into the tree that are removed when the IO is done (by the end_io
2875  * handlers)
2876  * XXX JDM: This needs looking at to ensure proper page locking
2877  * return 0 on success, otherwise return error
2878  */
2879 static int __do_readpage(struct extent_io_tree *tree,
2880                          struct page *page,
2881                          get_extent_t *get_extent,
2882                          struct extent_map **em_cached,
2883                          struct bio **bio, int mirror_num,
2884                          unsigned long *bio_flags, unsigned int read_flags,
2885                          u64 *prev_em_start)
2886 {
2887         struct inode *inode = page->mapping->host;
2888         u64 start = page_offset(page);
2889         u64 page_end = start + PAGE_SIZE - 1;
2890         u64 end;
2891         u64 cur = start;
2892         u64 extent_offset;
2893         u64 last_byte = i_size_read(inode);
2894         u64 block_start;
2895         u64 cur_end;
2896         struct extent_map *em;
2897         struct block_device *bdev;
2898         int ret = 0;
2899         int nr = 0;
2900         size_t pg_offset = 0;
2901         size_t iosize;
2902         size_t disk_io_size;
2903         size_t blocksize = inode->i_sb->s_blocksize;
2904         unsigned long this_bio_flag = 0;
2905 
2906         set_page_extent_mapped(page);
2907 
2908         end = page_end;
2909         if (!PageUptodate(page)) {
2910                 if (cleancache_get_page(page) == 0) {
2911                         BUG_ON(blocksize != PAGE_SIZE);
2912                         unlock_extent(tree, start, end);
2913                         goto out;
2914                 }
2915         }
2916 
2917         if (page->index == last_byte >> PAGE_SHIFT) {
2918                 char *userpage;
2919                 size_t zero_offset = last_byte & (PAGE_SIZE - 1);
2920 
2921                 if (zero_offset) {
2922                         iosize = PAGE_SIZE - zero_offset;
2923                         userpage = kmap_atomic(page);
2924                         memset(userpage + zero_offset, 0, iosize);
2925                         flush_dcache_page(page);
2926                         kunmap_atomic(userpage);
2927                 }
2928         }
2929         while (cur <= end) {
2930                 bool force_bio_submit = false;
2931                 u64 offset;
2932 
2933                 if (cur >= last_byte) {
2934                         char *userpage;
2935                         struct extent_state *cached = NULL;
2936 
2937                         iosize = PAGE_SIZE - pg_offset;
2938                         userpage = kmap_atomic(page);
2939                         memset(userpage + pg_offset, 0, iosize);
2940                         flush_dcache_page(page);
2941                         kunmap_atomic(userpage);
2942                         set_extent_uptodate(tree, cur, cur + iosize - 1,
2943                                             &cached, GFP_NOFS);
2944                         unlock_extent_cached(tree, cur,
2945                                              cur + iosize - 1,
2946                                              &cached, GFP_NOFS);
2947                         break;
2948                 }
2949                 em = __get_extent_map(inode, page, pg_offset, cur,
2950                                       end - cur + 1, get_extent, em_cached);
2951                 if (IS_ERR_OR_NULL(em)) {
2952                         SetPageError(page);
2953                         unlock_extent(tree, cur, end);
2954                         break;
2955                 }
2956                 extent_offset = cur - em->start;
2957                 BUG_ON(extent_map_end(em) <= cur);
2958                 BUG_ON(end < cur);
2959 
2960                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
2961                         this_bio_flag |= EXTENT_BIO_COMPRESSED;
2962                         extent_set_compress_type(&this_bio_flag,
2963                                                  em->compress_type);
2964                 }
2965 
2966                 iosize = min(extent_map_end(em) - cur, end - cur + 1);
2967                 cur_end = min(extent_map_end(em) - 1, end);
2968                 iosize = ALIGN(iosize, blocksize);
2969                 if (this_bio_flag & EXTENT_BIO_COMPRESSED) {
2970                         disk_io_size = em->block_len;
2971                         offset = em->block_start;
2972                 } else {
2973                         offset = em->block_start + extent_offset;
2974                         disk_io_size = iosize;
2975                 }
2976                 bdev = em->bdev;
2977                 block_start = em->block_start;
2978                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2979                         block_start = EXTENT_MAP_HOLE;
2980 
2981                 /*
2982                  * If we have a file range that points to a compressed extent
2983                  * and it's followed by a consecutive file range that points to
2984                  * to the same compressed extent (possibly with a different
2985                  * offset and/or length, so it either points to the whole extent
2986                  * or only part of it), we must make sure we do not submit a
2987                  * single bio to populate the pages for the 2 ranges because
2988                  * this makes the compressed extent read zero out the pages
2989                  * belonging to the 2nd range. Imagine the following scenario:
2990                  *
2991                  *  File layout
2992                  *  [0 - 8K]                     [8K - 24K]
2993                  *    |                               |
2994                  *    |                               |
2995                  * points to extent X,         points to extent X,
2996                  * offset 4K, length of 8K     offset 0, length 16K
2997                  *
2998                  * [extent X, compressed length = 4K uncompressed length = 16K]
2999                  *
3000                  * If the bio to read the compressed extent covers both ranges,
3001                  * it will decompress extent X into the pages belonging to the
3002                  * first range and then it will stop, zeroing out the remaining
3003                  * pages that belong to the other range that points to extent X.
3004                  * So here we make sure we submit 2 bios, one for the first
3005                  * range and another one for the third range. Both will target
3006                  * the same physical extent from disk, but we can't currently
3007                  * make the compressed bio endio callback populate the pages
3008                  * for both ranges because each compressed bio is tightly
3009                  * coupled with a single extent map, and each range can have
3010                  * an extent map with a different offset value relative to the
3011                  * uncompressed data of our extent and different lengths. This
3012                  * is a corner case so we prioritize correctness over
3013                  * non-optimal behavior (submitting 2 bios for the same extent).
3014                  */
3015                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
3016                     prev_em_start && *prev_em_start != (u64)-1 &&
3017                     *prev_em_start != em->orig_start)
3018                         force_bio_submit = true;
3019 
3020                 if (prev_em_start)
3021                         *prev_em_start = em->orig_start;
3022 
3023                 free_extent_map(em);
3024                 em = NULL;
3025 
3026                 /* we've found a hole, just zero and go on */
3027                 if (block_start == EXTENT_MAP_HOLE) {
3028                         char *userpage;
3029                         struct extent_state *cached = NULL;
3030 
3031                         userpage = kmap_atomic(page);
3032                         memset(userpage + pg_offset, 0, iosize);
3033                         flush_dcache_page(page);
3034                         kunmap_atomic(userpage);
3035 
3036                         set_extent_uptodate(tree, cur, cur + iosize - 1,
3037                                             &cached, GFP_NOFS);
3038                         unlock_extent_cached(tree, cur,
3039                                              cur + iosize - 1,
3040                                              &cached, GFP_NOFS);
3041                         cur = cur + iosize;
3042                         pg_offset += iosize;
3043                         continue;
3044                 }
3045                 /* the get_extent function already copied into the page */
3046                 if (test_range_bit(tree, cur, cur_end,
3047                                    EXTENT_UPTODATE, 1, NULL)) {
3048                         check_page_uptodate(tree, page);
3049                         unlock_extent(tree, cur, cur + iosize - 1);
3050                         cur = cur + iosize;
3051                         pg_offset += iosize;
3052                         continue;
3053                 }
3054                 /* we have an inline extent but it didn't get marked up
3055                  * to date.  Error out
3056                  */
3057                 if (block_start == EXTENT_MAP_INLINE) {
3058                         SetPageError(page);
3059                         unlock_extent(tree, cur, cur + iosize - 1);
3060                         cur = cur + iosize;
3061                         pg_offset += iosize;
3062                         continue;
3063                 }
3064 
3065                 ret = submit_extent_page(REQ_OP_READ | read_flags, tree, NULL,
3066                                          page, offset, disk_io_size,
3067                                          pg_offset, bdev, bio,
3068                                          end_bio_extent_readpage, mirror_num,
3069                                          *bio_flags,
3070                                          this_bio_flag,
3071                                          force_bio_submit);
3072                 if (!ret) {
3073                         nr++;
3074                         *bio_flags = this_bio_flag;
3075                 } else {
3076                         SetPageError(page);
3077                         unlock_extent(tree, cur, cur + iosize - 1);
3078                         goto out;
3079                 }
3080                 cur = cur + iosize;
3081                 pg_offset += iosize;
3082         }
3083 out:
3084         if (!nr) {
3085                 if (!PageError(page))
3086                         SetPageUptodate(page);
3087                 unlock_page(page);
3088         }
3089         return ret;
3090 }
3091 
3092 static inline void __do_contiguous_readpages(struct extent_io_tree *tree,
3093                                              struct page *pages[], int nr_pages,
3094                                              u64 start, u64 end,
3095                                              get_extent_t *get_extent,
3096                                              struct extent_map **em_cached,
3097                                              struct bio **bio, int mirror_num,
3098                                              unsigned long *bio_flags,
3099                                              u64 *prev_em_start)
3100 {
3101         struct inode *inode;
3102         struct btrfs_ordered_extent *ordered;
3103         int index;
3104 
3105         inode = pages[0]->mapping->host;
3106         while (1) {
3107                 lock_extent(tree, start, end);
3108                 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3109                                                      end - start + 1);
3110                 if (!ordered)
3111                         break;
3112                 unlock_extent(tree, start, end);
3113                 btrfs_start_ordered_extent(inode, ordered, 1);
3114                 btrfs_put_ordered_extent(ordered);
3115         }
3116 
3117         for (index = 0; index < nr_pages; index++) {
3118                 __do_readpage(tree, pages[index], get_extent, em_cached, bio,
3119                               mirror_num, bio_flags, 0, prev_em_start);
3120                 put_page(pages[index]);
3121         }
3122 }
3123 
3124 static void __extent_readpages(struct extent_io_tree *tree,
3125                                struct page *pages[],
3126                                int nr_pages, get_extent_t *get_extent,
3127                                struct extent_map **em_cached,
3128                                struct bio **bio, int mirror_num,
3129                                unsigned long *bio_flags,
3130                                u64 *prev_em_start)
3131 {
3132         u64 start = 0;
3133         u64 end = 0;
3134         u64 page_start;
3135         int index;
3136         int first_index = 0;
3137 
3138         for (index = 0; index < nr_pages; index++) {
3139                 page_start = page_offset(pages[index]);
3140                 if (!end) {
3141                         start = page_start;
3142                         end = start + PAGE_SIZE - 1;
3143                         first_index = index;
3144                 } else if (end + 1 == page_start) {
3145                         end += PAGE_SIZE;
3146                 } else {
3147                         __do_contiguous_readpages(tree, &pages[first_index],
3148                                                   index - first_index, start,
3149                                                   end, get_extent, em_cached,
3150                                                   bio, mirror_num, bio_flags,
3151                                                   prev_em_start);
3152                         start = page_start;
3153                         end = start + PAGE_SIZE - 1;
3154                         first_index = index;
3155                 }
3156         }
3157 
3158         if (end)
3159                 __do_contiguous_readpages(tree, &pages[first_index],
3160                                           index - first_index, start,
3161                                           end, get_extent, em_cached, bio,
3162                                           mirror_num, bio_flags,
3163                                           prev_em_start);
3164 }
3165 
3166 static int __extent_read_full_page(struct extent_io_tree *tree,
3167                                    struct page *page,
3168                                    get_extent_t *get_extent,
3169                                    struct bio **bio, int mirror_num,
3170                                    unsigned long *bio_flags,
3171                                    unsigned int read_flags)
3172 {
3173         struct inode *inode = page->mapping->host;
3174         struct btrfs_ordered_extent *ordered;
3175         u64 start = page_offset(page);
3176         u64 end = start + PAGE_SIZE - 1;
3177         int ret;
3178 
3179         while (1) {
3180                 lock_extent(tree, start, end);
3181                 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
3182                                                 PAGE_SIZE);
3183                 if (!ordered)
3184                         break;
3185                 unlock_extent(tree, start, end);
3186                 btrfs_start_ordered_extent(inode, ordered, 1);
3187                 btrfs_put_ordered_extent(ordered);
3188         }
3189 
3190         ret = __do_readpage(tree, page, get_extent, NULL, bio, mirror_num,
3191                             bio_flags, read_flags, NULL);
3192         return ret;
3193 }
3194 
3195 int extent_read_full_page(struct extent_io_tree *tree, struct page *page,
3196                             get_extent_t *get_extent, int mirror_num)
3197 {
3198         struct bio *bio = NULL;
3199         unsigned long bio_flags = 0;
3200         int ret;
3201 
3202         ret = __extent_read_full_page(tree, page, get_extent, &bio, mirror_num,
3203                                       &bio_flags, 0);
3204         if (bio)
3205                 ret = submit_one_bio(bio, mirror_num, bio_flags);
3206         return ret;
3207 }
3208 
3209 static void update_nr_written(struct writeback_control *wbc,
3210                               unsigned long nr_written)
3211 {
3212         wbc->nr_to_write -= nr_written;
3213 }
3214 
3215 /*
3216  * helper for __extent_writepage, doing all of the delayed allocation setup.
3217  *
3218  * This returns 1 if our fill_delalloc function did all the work required
3219  * to write the page (copy into inline extent).  In this case the IO has
3220  * been started and the page is already unlocked.
3221  *
3222  * This returns 0 if all went well (page still locked)
3223  * This returns < 0 if there were errors (page still locked)
3224  */
3225 static noinline_for_stack int writepage_delalloc(struct inode *inode,
3226                               struct page *page, struct writeback_control *wbc,
3227                               struct extent_page_data *epd,
3228                               u64 delalloc_start,
3229                               unsigned long *nr_written)
3230 {
3231         struct extent_io_tree *tree = epd->tree;
3232         u64 page_end = delalloc_start + PAGE_SIZE - 1;
3233         u64 nr_delalloc;
3234         u64 delalloc_to_write = 0;
3235         u64 delalloc_end = 0;
3236         int ret;
3237         int page_started = 0;
3238 
3239         if (epd->extent_locked || !tree->ops || !tree->ops->fill_delalloc)
3240                 return 0;
3241 
3242         while (delalloc_end < page_end) {
3243                 nr_delalloc = find_lock_delalloc_range(inode, tree,
3244                                                page,
3245                                                &delalloc_start,
3246                                                &delalloc_end,
3247                                                BTRFS_MAX_EXTENT_SIZE);
3248                 if (nr_delalloc == 0) {
3249                         delalloc_start = delalloc_end + 1;
3250                         continue;
3251                 }
3252                 ret = tree->ops->fill_delalloc(inode, page,
3253                                                delalloc_start,
3254                                                delalloc_end,
3255                                                &page_started,
3256                                                nr_written, wbc);
3257                 /* File system has been set read-only */
3258                 if (ret) {
3259                         SetPageError(page);
3260                         /* fill_delalloc should be return < 0 for error
3261                          * but just in case, we use > 0 here meaning the
3262                          * IO is started, so we don't want to return > 0
3263                          * unless things are going well.
3264                          */
3265                         ret = ret < 0 ? ret : -EIO;
3266                         goto done;
3267                 }
3268                 /*
3269                  * delalloc_end is already one less than the total length, so
3270                  * we don't subtract one from PAGE_SIZE
3271                  */
3272                 delalloc_to_write += (delalloc_end - delalloc_start +
3273                                       PAGE_SIZE) >> PAGE_SHIFT;
3274                 delalloc_start = delalloc_end + 1;
3275         }
3276         if (wbc->nr_to_write < delalloc_to_write) {
3277                 int thresh = 8192;
3278 
3279                 if (delalloc_to_write < thresh * 2)
3280                         thresh = delalloc_to_write;
3281                 wbc->nr_to_write = min_t(u64, delalloc_to_write,
3282                                          thresh);
3283         }
3284 
3285         /* did the fill delalloc function already unlock and start
3286          * the IO?
3287          */
3288         if (page_started) {
3289                 /*
3290                  * we've unlocked the page, so we can't update
3291                  * the mapping's writeback index, just update
3292                  * nr_to_write.
3293                  */
3294                 wbc->nr_to_write -= *nr_written;
3295                 return 1;
3296         }
3297 
3298         ret = 0;
3299 
3300 done:
3301         return ret;
3302 }
3303 
3304 /*
3305  * helper for __extent_writepage.  This calls the writepage start hooks,
3306  * and does the loop to map the page into extents and bios.
3307  *
3308  * We return 1 if the IO is started and the page is unlocked,
3309  * 0 if all went well (page still locked)
3310  * < 0 if there were errors (page still locked)
3311  */
3312 static noinline_for_stack int __extent_writepage_io(struct inode *inode,
3313                                  struct page *page,
3314                                  struct writeback_control *wbc,
3315                                  struct extent_page_data *epd,
3316                                  loff_t i_size,
3317                                  unsigned long nr_written,
3318                                  unsigned int write_flags, int *nr_ret)
3319 {
3320         struct extent_io_tree *tree = epd->tree;
3321         u64 start = page_offset(page);
3322         u64 page_end = start + PAGE_SIZE - 1;
3323         u64 end;
3324         u64 cur = start;
3325         u64 extent_offset;
3326         u64 block_start;
3327         u64 iosize;
3328         struct extent_map *em;
3329         struct block_device *bdev;
3330         size_t pg_offset = 0;
3331         size_t blocksize;
3332         int ret = 0;
3333         int nr = 0;
3334         bool compressed;
3335 
3336         if (tree->ops && tree->ops->writepage_start_hook) {
3337                 ret = tree->ops->writepage_start_hook(page, start,
3338                                                       page_end);
3339                 if (ret) {
3340                         /* Fixup worker will requeue */
3341                         if (ret == -EBUSY)
3342                                 wbc->pages_skipped++;
3343                         else
3344                                 redirty_page_for_writepage(wbc, page);
3345 
3346                         update_nr_written(wbc, nr_written);
3347                         unlock_page(page);
3348                         return 1;
3349                 }
3350         }
3351 
3352         /*
3353          * we don't want to touch the inode after unlocking the page,
3354          * so we update the mapping writeback index now
3355          */
3356         update_nr_written(wbc, nr_written + 1);
3357 
3358         end = page_end;
3359         if (i_size <= start) {
3360                 if (tree->ops && tree->ops->writepage_end_io_hook)
3361                         tree->ops->writepage_end_io_hook(page, start,
3362                                                          page_end, NULL, 1);
3363                 goto done;
3364         }
3365 
3366         blocksize = inode->i_sb->s_blocksize;
3367 
3368         while (cur <= end) {
3369                 u64 em_end;
3370                 u64 offset;
3371 
3372                 if (cur >= i_size) {
3373                         if (tree->ops && tree->ops->writepage_end_io_hook)
3374                                 tree->ops->writepage_end_io_hook(page, cur,
3375                                                          page_end, NULL, 1);
3376                         break;
3377                 }
3378                 em = epd->get_extent(BTRFS_I(inode), page, pg_offset, cur,
3379                                      end - cur + 1, 1);
3380                 if (IS_ERR_OR_NULL(em)) {
3381                         SetPageError(page);
3382                         ret = PTR_ERR_OR_ZERO(em);
3383                         break;
3384                 }
3385 
3386                 extent_offset = cur - em->start;
3387                 em_end = extent_map_end(em);
3388                 BUG_ON(em_end <= cur);
3389                 BUG_ON(end < cur);
3390                 iosize = min(em_end - cur, end - cur + 1);
3391                 iosize = ALIGN(iosize, blocksize);
3392                 offset = em->block_start + extent_offset;
3393                 bdev = em->bdev;
3394                 block_start = em->block_start;
3395                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3396                 free_extent_map(em);
3397                 em = NULL;
3398 
3399                 /*
3400                  * compressed and inline extents are written through other
3401                  * paths in the FS
3402                  */
3403                 if (compressed || block_start == EXTENT_MAP_HOLE ||
3404                     block_start == EXTENT_MAP_INLINE) {
3405                         /*
3406                          * end_io notification does not happen here for
3407                          * compressed extents
3408                          */
3409                         if (!compressed && tree->ops &&
3410                             tree->ops->writepage_end_io_hook)
3411                                 tree->ops->writepage_end_io_hook(page, cur,
3412                                                          cur + iosize - 1,
3413                                                          NULL, 1);
3414                         else if (compressed) {
3415                                 /* we don't want to end_page_writeback on
3416                                  * a compressed extent.  this happens
3417                                  * elsewhere
3418                                  */
3419                                 nr++;
3420                         }
3421 
3422                         cur += iosize;
3423                         pg_offset += iosize;
3424                         continue;
3425                 }
3426 
3427                 set_range_writeback(tree, cur, cur + iosize - 1);
3428                 if (!PageWriteback(page)) {
3429                         btrfs_err(BTRFS_I(inode)->root->fs_info,
3430                                    "page %lu not writeback, cur %llu end %llu",
3431                                page->index, cur, end);
3432                 }
3433 
3434                 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3435                                          page, offset, iosize, pg_offset,
3436                                          bdev, &epd->bio,
3437                                          end_bio_extent_writepage,
3438                                          0, 0, 0, false);
3439                 if (ret) {
3440                         SetPageError(page);
3441                         if (PageWriteback(page))
3442                                 end_page_writeback(page);
3443                 }
3444 
3445                 cur = cur + iosize;
3446                 pg_offset += iosize;
3447                 nr++;
3448         }
3449 done:
3450         *nr_ret = nr;
3451         return ret;
3452 }
3453 
3454 /*
3455  * the writepage semantics are similar to regular writepage.  extent
3456  * records are inserted to lock ranges in the tree, and as dirty areas
3457  * are found, they are marked writeback.  Then the lock bits are removed
3458  * and the end_io handler clears the writeback ranges
3459  */
3460 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
3461                               void *data)
3462 {
3463         struct inode *inode = page->mapping->host;
3464         struct extent_page_data *epd = data;
3465         u64 start = page_offset(page);
3466         u64 page_end = start + PAGE_SIZE - 1;
3467         int ret;
3468         int nr = 0;
3469         size_t pg_offset = 0;
3470         loff_t i_size = i_size_read(inode);
3471         unsigned long end_index = i_size >> PAGE_SHIFT;
3472         unsigned int write_flags = 0;
3473         unsigned long nr_written = 0;
3474 
3475         write_flags = wbc_to_write_flags(wbc);
3476 
3477         trace___extent_writepage(page, inode, wbc);
3478 
3479         WARN_ON(!PageLocked(page));
3480 
3481         ClearPageError(page);
3482 
3483         pg_offset = i_size & (PAGE_SIZE - 1);
3484         if (page->index > end_index ||
3485            (page->index == end_index && !pg_offset)) {
3486                 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
3487                 unlock_page(page);
3488                 return 0;
3489         }
3490 
3491         if (page->index == end_index) {
3492                 char *userpage;
3493 
3494                 userpage = kmap_atomic(page);
3495                 memset(userpage + pg_offset, 0,
3496                        PAGE_SIZE - pg_offset);
3497                 kunmap_atomic(userpage);
3498                 flush_dcache_page(page);
3499         }
3500 
3501         pg_offset = 0;
3502 
3503         set_page_extent_mapped(page);
3504 
3505         ret = writepage_delalloc(inode, page, wbc, epd, start, &nr_written);
3506         if (ret == 1)
3507                 goto done_unlocked;
3508         if (ret)
3509                 goto done;
3510 
3511         ret = __extent_writepage_io(inode, page, wbc, epd,
3512                                     i_size, nr_written, write_flags, &nr);
3513         if (ret == 1)
3514                 goto done_unlocked;
3515 
3516 done:
3517         if (nr == 0) {
3518                 /* make sure the mapping tag for page dirty gets cleared */
3519                 set_page_writeback(page);
3520                 end_page_writeback(page);
3521         }
3522         if (PageError(page)) {
3523                 ret = ret < 0 ? ret : -EIO;
3524                 end_extent_writepage(page, ret, start, page_end);
3525         }
3526         unlock_page(page);
3527         return ret;
3528 
3529 done_unlocked:
3530         return 0;
3531 }
3532 
3533 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
3534 {
3535         wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
3536                        TASK_UNINTERRUPTIBLE);
3537 }
3538 
3539 static noinline_for_stack int
3540 lock_extent_buffer_for_io(struct extent_buffer *eb,
3541                           struct btrfs_fs_info *fs_info,
3542                           struct extent_page_data *epd)
3543 {
3544         unsigned long i, num_pages;
3545         int flush = 0;
3546         int ret = 0;
3547 
3548         if (!btrfs_try_tree_write_lock(eb)) {
3549                 flush = 1;
3550                 flush_write_bio(epd);
3551                 btrfs_tree_lock(eb);
3552         }
3553 
3554         if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
3555                 btrfs_tree_unlock(eb);
3556                 if (!epd->sync_io)
3557                         return 0;
3558                 if (!flush) {
3559                         flush_write_bio(epd);
3560                         flush = 1;
3561                 }
3562                 while (1) {
3563                         wait_on_extent_buffer_writeback(eb);
3564                         btrfs_tree_lock(eb);
3565                         if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
3566                                 break;
3567                         btrfs_tree_unlock(eb);
3568                 }
3569         }
3570 
3571         /*
3572          * We need to do this to prevent races in people who check if the eb is
3573          * under IO since we can end up having no IO bits set for a short period
3574          * of time.
3575          */
3576         spin_lock(&eb->refs_lock);
3577         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3578                 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3579                 spin_unlock(&eb->refs_lock);
3580                 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3581                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
3582                                          -eb->len,
3583                                          fs_info->dirty_metadata_batch);
3584                 ret = 1;
3585         } else {
3586                 spin_unlock(&eb->refs_lock);
3587         }
3588 
3589         btrfs_tree_unlock(eb);
3590 
3591         if (!ret)
3592                 return ret;
3593 
3594         num_pages = num_extent_pages(eb->start, eb->len);
3595         for (i = 0; i < num_pages; i++) {
3596                 struct page *p = eb->pages[i];
3597 
3598                 if (!trylock_page(p)) {
3599                         if (!flush) {
3600                                 flush_write_bio(epd);
3601                                 flush = 1;
3602                         }
3603                         lock_page(p);
3604                 }
3605         }
3606 
3607         return ret;
3608 }
3609 
3610 static void end_extent_buffer_writeback(struct extent_buffer *eb)
3611 {
3612         clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
3613         smp_mb__after_atomic();
3614         wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
3615 }
3616 
3617 static void set_btree_ioerr(struct page *page)
3618 {
3619         struct extent_buffer *eb = (struct extent_buffer *)page->private;
3620 
3621         SetPageError(page);
3622         if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
3623                 return;
3624 
3625         /*
3626          * If writeback for a btree extent that doesn't belong to a log tree
3627          * failed, increment the counter transaction->eb_write_errors.
3628          * We do this because while the transaction is running and before it's
3629          * committing (when we call filemap_fdata[write|wait]_range against
3630          * the btree inode), we might have
3631          * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
3632          * returns an error or an error happens during writeback, when we're
3633          * committing the transaction we wouldn't know about it, since the pages
3634          * can be no longer dirty nor marked anymore for writeback (if a
3635          * subsequent modification to the extent buffer didn't happen before the
3636          * transaction commit), which makes filemap_fdata[write|wait]_range not
3637          * able to find the pages tagged with SetPageError at transaction
3638          * commit time. So if this happens we must abort the transaction,
3639          * otherwise we commit a super block with btree roots that point to
3640          * btree nodes/leafs whose content on disk is invalid - either garbage
3641          * or the content of some node/leaf from a past generation that got
3642          * cowed or deleted and is no longer valid.
3643          *
3644          * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
3645          * not be enough - we need to distinguish between log tree extents vs
3646          * non-log tree extents, and the next filemap_fdatawait_range() call
3647          * will catch and clear such errors in the mapping - and that call might
3648          * be from a log sync and not from a transaction commit. Also, checking
3649          * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
3650          * not done and would not be reliable - the eb might have been released
3651          * from memory and reading it back again means that flag would not be
3652          * set (since it's a runtime flag, not persisted on disk).
3653          *
3654          * Using the flags below in the btree inode also makes us achieve the
3655          * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
3656          * writeback for all dirty pages and before filemap_fdatawait_range()
3657          * is called, the writeback for all dirty pages had already finished
3658          * with errors - because we were not using AS_EIO/AS_ENOSPC,
3659          * filemap_fdatawait_range() would return success, as it could not know
3660          * that writeback errors happened (the pages were no longer tagged for
3661          * writeback).
3662          */
3663         switch (eb->log_index) {
3664         case -1:
3665                 set_bit(BTRFS_FS_BTREE_ERR, &eb->fs_info->flags);
3666                 break;
3667         case 0:
3668                 set_bit(BTRFS_FS_LOG1_ERR, &eb->fs_info->flags);
3669                 break;
3670         case 1:
3671                 set_bit(BTRFS_FS_LOG2_ERR, &eb->fs_info->flags);
3672                 break;
3673         default:
3674                 BUG(); /* unexpected, logic error */
3675         }
3676 }
3677 
3678 static void end_bio_extent_buffer_writepage(struct bio *bio)
3679 {
3680         struct bio_vec *bvec;
3681         struct extent_buffer *eb;
3682         int i, done;
3683 
3684         ASSERT(!bio_flagged(bio, BIO_CLONED));
3685         bio_for_each_segment_all(bvec, bio, i) {
3686                 struct page *page = bvec->bv_page;
3687 
3688                 eb = (struct extent_buffer *)page->private;
3689                 BUG_ON(!eb);
3690                 done = atomic_dec_and_test(&eb->io_pages);
3691 
3692                 if (bio->bi_status ||
3693                     test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
3694                         ClearPageUptodate(page);
3695                         set_btree_ioerr(page);
3696                 }
3697 
3698                 end_page_writeback(page);
3699 
3700                 if (!done)
3701                         continue;
3702 
3703                 end_extent_buffer_writeback(eb);
3704         }
3705 
3706         bio_put(bio);
3707 }
3708 
3709 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
3710                         struct btrfs_fs_info *fs_info,
3711                         struct writeback_control *wbc,
3712                         struct extent_page_data *epd)
3713 {
3714         struct block_device *bdev = fs_info->fs_devices->latest_bdev;
3715         struct extent_io_tree *tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
3716         u64 offset = eb->start;
3717         u32 nritems;
3718         unsigned long i, num_pages;
3719         unsigned long start, end;
3720         unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META;
3721         int ret = 0;
3722 
3723         clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
3724         num_pages = num_extent_pages(eb->start, eb->len);
3725         atomic_set(&eb->io_pages, num_pages);
3726 
3727         /* set btree blocks beyond nritems with 0 to avoid stale content. */
3728         nritems = btrfs_header_nritems(eb);
3729         if (btrfs_header_level(eb) > 0) {
3730                 end = btrfs_node_key_ptr_offset(nritems);
3731 
3732                 memzero_extent_buffer(eb, end, eb->len - end);
3733         } else {
3734                 /*
3735                  * leaf:
3736                  * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
3737                  */
3738                 start = btrfs_item_nr_offset(nritems);
3739                 end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, eb);
3740                 memzero_extent_buffer(eb, start, end - start);
3741         }
3742 
3743         for (i = 0; i < num_pages; i++) {
3744                 struct page *p = eb->pages[i];
3745 
3746                 clear_page_dirty_for_io(p);
3747                 set_page_writeback(p);
3748                 ret = submit_extent_page(REQ_OP_WRITE | write_flags, tree, wbc,
3749                                          p, offset, PAGE_SIZE, 0, bdev,
3750                                          &epd->bio,
3751                                          end_bio_extent_buffer_writepage,
3752                                          0, 0, 0, false);
3753                 if (ret) {
3754                         set_btree_ioerr(p);
3755                         if (PageWriteback(p))
3756                                 end_page_writeback(p);
3757                         if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
3758                                 end_extent_buffer_writeback(eb);
3759                         ret = -EIO;
3760                         break;
3761                 }
3762                 offset += PAGE_SIZE;
3763                 update_nr_written(wbc, 1);
3764                 unlock_page(p);
3765         }
3766 
3767         if (unlikely(ret)) {
3768                 for (; i < num_pages; i++) {
3769                         struct page *p = eb->pages[i];
3770                         clear_page_dirty_for_io(p);
3771                         unlock_page(p);
3772                 }
3773         }
3774 
3775         return ret;
3776 }
3777 
3778 int btree_write_cache_pages(struct address_space *mapping,
3779                                    struct writeback_control *wbc)
3780 {
3781         struct extent_io_tree *tree = &BTRFS_I(mapping->host)->io_tree;
3782         struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
3783         struct extent_buffer *eb, *prev_eb = NULL;
3784         struct extent_page_data epd = {
3785                 .bio = NULL,
3786                 .tree = tree,
3787                 .extent_locked = 0,
3788                 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
3789         };
3790         int ret = 0;
3791         int done = 0;
3792         int nr_to_write_done = 0;
3793         struct pagevec pvec;
3794         int nr_pages;
3795         pgoff_t index;
3796         pgoff_t end;            /* Inclusive */
3797         int scanned = 0;
3798         int tag;
3799 
3800         pagevec_init(&pvec);
3801         if (wbc->range_cyclic) {
3802                 index = mapping->writeback_index; /* Start from prev offset */
3803                 end = -1;
3804         } else {
3805                 index = wbc->range_start >> PAGE_SHIFT;
3806                 end = wbc->range_end >> PAGE_SHIFT;
3807                 scanned = 1;
3808         }
3809         if (wbc->sync_mode == WB_SYNC_ALL)
3810                 tag = PAGECACHE_TAG_TOWRITE;
3811         else
3812                 tag = PAGECACHE_TAG_DIRTY;
3813 retry:
3814         if (wbc->sync_mode == WB_SYNC_ALL)
3815                 tag_pages_for_writeback(mapping, index, end);
3816         while (!done && !nr_to_write_done && (index <= end) &&
3817                (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
3818                         tag))) {
3819                 unsigned i;
3820 
3821                 scanned = 1;
3822                 for (i = 0; i < nr_pages; i++) {
3823                         struct page *page = pvec.pages[i];
3824 
3825                         if (!PagePrivate(page))
3826                                 continue;
3827 
3828                         spin_lock(&mapping->private_lock);
3829                         if (!PagePrivate(page)) {
3830                                 spin_unlock(&mapping->private_lock);
3831                                 continue;
3832                         }
3833 
3834                         eb = (struct extent_buffer *)page->private;
3835 
3836                         /*
3837                          * Shouldn't happen and normally this would be a BUG_ON
3838                          * but no sense in crashing the users box for something
3839                          * we can survive anyway.
3840                          */
3841                         if (WARN_ON(!eb)) {
3842                                 spin_unlock(&mapping->private_lock);
3843                                 continue;
3844                         }
3845 
3846                         if (eb == prev_eb) {
3847                                 spin_unlock(&mapping->private_lock);
3848                                 continue;
3849                         }
3850 
3851                         ret = atomic_inc_not_zero(&eb->refs);
3852                         spin_unlock(&mapping->private_lock);
3853                         if (!ret)
3854                                 continue;
3855 
3856                         prev_eb = eb;
3857                         ret = lock_extent_buffer_for_io(eb, fs_info, &epd);
3858                         if (!ret) {
3859                                 free_extent_buffer(eb);
3860                                 continue;
3861                         }
3862 
3863                         ret = write_one_eb(eb, fs_info, wbc, &epd);
3864                         if (ret) {
3865                                 done = 1;
3866                                 free_extent_buffer(eb);
3867                                 break;
3868                         }
3869                         free_extent_buffer(eb);
3870 
3871                         /*
3872                          * the filesystem may choose to bump up nr_to_write.
3873                          * We have to make sure to honor the new nr_to_write
3874                          * at any time
3875                          */
3876                         nr_to_write_done = wbc->nr_to_write <= 0;
3877                 }
3878                 pagevec_release(&pvec);
3879                 cond_resched();
3880         }
3881         if (!scanned && !done) {
3882                 /*
3883                  * We hit the last page and there is more work to be done: wrap
3884                  * back to the start of the file
3885                  */
3886                 scanned = 1;
3887                 index = 0;
3888                 goto retry;
3889         }
3890         flush_write_bio(&epd);
3891         return ret;
3892 }
3893 
3894 /**
3895  * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
3896  * @mapping: address space structure to write
3897  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
3898  * @writepage: function called for each page
3899  * @data: data passed to writepage function
3900  *
3901  * If a page is already under I/O, write_cache_pages() skips it, even
3902  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
3903  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
3904  * and msync() need to guarantee that all the data which was dirty at the time
3905  * the call was made get new I/O started against them.  If wbc->sync_mode is
3906  * WB_SYNC_ALL then we were called for data integrity and we must wait for
3907  * existing IO to complete.
3908  */
3909 static int extent_write_cache_pages(struct address_space *mapping,
3910                              struct writeback_control *wbc,
3911                              writepage_t writepage, void *data,
3912                              void (*flush_fn)(void *))
3913 {
3914         struct inode *inode = mapping->host;
3915         int ret = 0;
3916         int done = 0;
3917         int nr_to_write_done = 0;
3918         struct pagevec pvec;
3919         int nr_pages;
3920         pgoff_t index;
3921         pgoff_t end;            /* Inclusive */
3922         pgoff_t done_index;
3923         int range_whole = 0;
3924         int scanned = 0;
3925         int tag;
3926 
3927         /*
3928          * We have to hold onto the inode so that ordered extents can do their
3929          * work when the IO finishes.  The alternative to this is failing to add
3930          * an ordered extent if the igrab() fails there and that is a huge pain
3931          * to deal with, so instead just hold onto the inode throughout the
3932          * writepages operation.  If it fails here we are freeing up the inode
3933          * anyway and we'd rather not waste our time writing out stuff that is
3934          * going to be truncated anyway.
3935          */
3936         if (!igrab(inode))
3937                 return 0;
3938 
3939         pagevec_init(&pvec);
3940         if (wbc->range_cyclic) {
3941                 index = mapping->writeback_index; /* Start from prev offset */
3942                 end = -1;
3943         } else {
3944                 index = wbc->range_start >> PAGE_SHIFT;
3945                 end = wbc->range_end >> PAGE_SHIFT;
3946                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3947                         range_whole = 1;
3948                 scanned = 1;
3949         }
3950         if (wbc->sync_mode == WB_SYNC_ALL)
3951                 tag = PAGECACHE_TAG_TOWRITE;
3952         else
3953                 tag = PAGECACHE_TAG_DIRTY;
3954 retry:
3955         if (wbc->sync_mode == WB_SYNC_ALL)
3956                 tag_pages_for_writeback(mapping, index, end);
3957         done_index = index;
3958         while (!done && !nr_to_write_done && (index <= end) &&
3959                         (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
3960                                                 &index, end, tag))) {
3961                 unsigned i;
3962 
3963                 scanned = 1;
3964                 for (i = 0; i < nr_pages; i++) {
3965                         struct page *page = pvec.pages[i];
3966 
3967                         done_index = page->index;
3968                         /*
3969                          * At this point we hold neither mapping->tree_lock nor
3970                          * lock on the page itself: the page may be truncated or
3971                          * invalidated (changing page->mapping to NULL), or even
3972                          * swizzled back from swapper_space to tmpfs file
3973                          * mapping
3974                          */
3975                         if (!trylock_page(page)) {
3976                                 flush_fn(data);
3977                                 lock_page(page);
3978                         }
3979 
3980                         if (unlikely(page->mapping != mapping)) {
3981                                 unlock_page(page);
3982                                 continue;
3983                         }
3984 
3985                         if (wbc->sync_mode != WB_SYNC_NONE) {
3986                                 if (PageWriteback(page))
3987                                         flush_fn(data);
3988                                 wait_on_page_writeback(page);
3989                         }
3990 
3991                         if (PageWriteback(page) ||
3992                             !clear_page_dirty_for_io(page)) {
3993                                 unlock_page(page);
3994                                 continue;
3995                         }
3996 
3997                         ret = (*writepage)(page, wbc, data);
3998 
3999                         if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE)) {
4000                                 unlock_page(page);
4001                                 ret = 0;
4002                         }
4003                         if (ret < 0) {
4004                                 /*
4005                                  * done_index is set past this page,
4006                                  * so media errors will not choke
4007                                  * background writeout for the entire
4008                                  * file. This has consequences for
4009                                  * range_cyclic semantics (ie. it may
4010                                  * not be suitable for data integrity
4011                                  * writeout).
4012                                  */
4013                                 done_index = page->index + 1;
4014                                 done = 1;
4015                                 break;
4016                         }
4017 
4018                         /*
4019                          * the filesystem may choose to bump up nr_to_write.
4020                          * We have to make sure to honor the new nr_to_write
4021                          * at any time
4022                          */
4023                         nr_to_write_done = wbc->nr_to_write <= 0;
4024                 }
4025                 pagevec_release(&pvec);
4026                 cond_resched();
4027         }
4028         if (!scanned && !done) {
4029                 /*
4030                  * We hit the last page and there is more work to be done: wrap
4031                  * back to the start of the file
4032                  */
4033                 scanned = 1;
4034                 index = 0;
4035                 goto retry;
4036         }
4037 
4038         if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
4039                 mapping->writeback_index = done_index;
4040 
4041         btrfs_add_delayed_iput(inode);
4042         return ret;
4043 }
4044 
4045 static void flush_epd_write_bio(struct extent_page_data *epd)
4046 {
4047         if (epd->bio) {
4048                 int ret;
4049 
4050                 ret = submit_one_bio(epd->bio, 0, 0);
4051                 BUG_ON(ret < 0); /* -ENOMEM */
4052                 epd->bio = NULL;
4053         }
4054 }
4055 
4056 static noinline void flush_write_bio(void *data)
4057 {
4058         struct extent_page_data *epd = data;
4059         flush_epd_write_bio(epd);
4060 }
4061 
4062 int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
4063                           get_extent_t *get_extent,
4064                           struct writeback_control *wbc)
4065 {
4066         int ret;
4067         struct extent_page_data epd = {
4068                 .bio = NULL,
4069                 .tree = tree,
4070                 .get_extent = get_extent,
4071                 .extent_locked = 0,
4072                 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4073         };
4074 
4075         ret = __extent_writepage(page, wbc, &epd);
4076 
4077         flush_epd_write_bio(&epd);
4078         return ret;
4079 }
4080 
4081 int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
4082                               u64 start, u64 end, get_extent_t *get_extent,
4083                               int mode)
4084 {
4085         int ret = 0;
4086         struct address_space *mapping = inode->i_mapping;
4087         struct page *page;
4088         unsigned long nr_pages = (end - start + PAGE_SIZE) >>
4089                 PAGE_SHIFT;
4090 
4091         struct extent_page_data epd = {
4092                 .bio = NULL,
4093                 .tree = tree,
4094                 .get_extent = get_extent,
4095                 .extent_locked = 1,
4096                 .sync_io = mode == WB_SYNC_ALL,
4097         };
4098         struct writeback_control wbc_writepages = {
4099                 .sync_mode      = mode,
4100                 .nr_to_write    = nr_pages * 2,
4101                 .range_start    = start,
4102                 .range_end      = end + 1,
4103         };
4104 
4105         while (start <= end) {
4106                 page = find_get_page(mapping, start >> PAGE_SHIFT);
4107                 if (clear_page_dirty_for_io(page))
4108                         ret = __extent_writepage(page, &wbc_writepages, &epd);
4109                 else {
4110                         if (tree->ops && tree->ops->writepage_end_io_hook)
4111                                 tree->ops->writepage_end_io_hook(page, start,
4112                                                  start + PAGE_SIZE - 1,
4113                                                  NULL, 1);
4114                         unlock_page(page);
4115                 }
4116                 put_page(page);
4117                 start += PAGE_SIZE;
4118         }
4119 
4120         flush_epd_write_bio(&epd);
4121         return ret;
4122 }
4123 
4124 int extent_writepages(struct extent_io_tree *tree,
4125                       struct address_space *mapping,
4126                       get_extent_t *get_extent,
4127                       struct writeback_control *wbc)
4128 {
4129         int ret = 0;
4130         struct extent_page_data epd = {
4131                 .bio = NULL,
4132                 .tree = tree,
4133                 .get_extent = get_extent,
4134                 .extent_locked = 0,
4135                 .sync_io = wbc->sync_mode == WB_SYNC_ALL,
4136         };
4137 
4138         ret = extent_write_cache_pages(mapping, wbc, __extent_writepage, &epd,
4139                                        flush_write_bio);
4140         flush_epd_write_bio(&epd);
4141         return ret;
4142 }
4143 
4144 int extent_readpages(struct extent_io_tree *tree,
4145                      struct address_space *mapping,
4146                      struct list_head *pages, unsigned nr_pages,
4147                      get_extent_t get_extent)
4148 {
4149         struct bio *bio = NULL;
4150         unsigned page_idx;
4151         unsigned long bio_flags = 0;
4152         struct page *pagepool[16];
4153         struct page *page;
4154         struct extent_map *em_cached = NULL;
4155         int nr = 0;
4156         u64 prev_em_start = (u64)-1;
4157 
4158         for (page_idx = 0; page_idx < nr_pages; page_idx++) {
4159                 page = list_entry(pages->prev, struct page, lru);
4160 
4161                 prefetchw(&page->flags);
4162                 list_del(&page->lru);
4163                 if (add_to_page_cache_lru(page, mapping,
4164                                         page->index,
4165                                         readahead_gfp_mask(mapping))) {
4166                         put_page(page);
4167                         continue;
4168                 }
4169 
4170                 pagepool[nr++] = page;
4171                 if (nr < ARRAY_SIZE(pagepool))
4172                         continue;
4173                 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4174                                    &bio, 0, &bio_flags, &prev_em_start);
4175                 nr = 0;
4176         }
4177         if (nr)
4178                 __extent_readpages(tree, pagepool, nr, get_extent, &em_cached,
4179                                    &bio, 0, &bio_flags, &prev_em_start);
4180 
4181         if (em_cached)
4182                 free_extent_map(em_cached);
4183 
4184         BUG_ON(!list_empty(pages));
4185         if (bio)
4186                 return submit_one_bio(bio, 0, bio_flags);
4187         return 0;
4188 }
4189 
4190 /*
4191  * basic invalidatepage code, this waits on any locked or writeback
4192  * ranges corresponding to the page, and then deletes any extent state
4193  * records from the tree
4194  */
4195 int extent_invalidatepage(struct extent_io_tree *tree,
4196                           struct page *page, unsigned long offset)
4197 {
4198         struct extent_state *cached_state = NULL;
4199         u64 start = page_offset(page);
4200         u64 end = start + PAGE_SIZE - 1;
4201         size_t blocksize = page->mapping->host->i_sb->s_blocksize;
4202 
4203         start += ALIGN(offset, blocksize);
4204         if (start > end)
4205                 return 0;
4206 
4207         lock_extent_bits(tree, start, end, &cached_state);
4208         wait_on_page_writeback(page);
4209         clear_extent_bit(tree, start, end,
4210                          EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4211                          EXTENT_DO_ACCOUNTING,
4212                          1, 1, &cached_state, GFP_NOFS);
4213         return 0;
4214 }
4215 
4216 /*
4217  * a helper for releasepage, this tests for areas of the page that
4218  * are locked or under IO and drops the related state bits if it is safe
4219  * to drop the page.
4220  */
4221 static int try_release_extent_state(struct extent_map_tree *map,
4222                                     struct extent_io_tree *tree,
4223                                     struct page *page, gfp_t mask)
4224 {
4225         u64 start = page_offset(page);
4226         u64 end = start + PAGE_SIZE - 1;
4227         int ret = 1;
4228 
4229         if (test_range_bit(tree, start, end,
4230                            EXTENT_IOBITS, 0, NULL))
4231                 ret = 0;
4232         else {
4233                 /*
4234                  * at this point we can safely clear everything except the
4235                  * locked bit and the nodatasum bit
4236                  */
4237                 ret = clear_extent_bit(tree, start, end,
4238                                  ~(EXTENT_LOCKED | EXTENT_NODATASUM),
4239                                  0, 0, NULL, mask);
4240 
4241                 /* if clear_extent_bit failed for enomem reasons,
4242                  * we can't allow the release to continue.
4243                  */
4244                 if (ret < 0)
4245                         ret = 0;
4246                 else
4247                         ret = 1;
4248         }
4249         return ret;
4250 }
4251 
4252 /*
4253  * a helper for releasepage.  As long as there are no locked extents
4254  * in the range corresponding to the page, both state records and extent
4255  * map records are removed
4256  */
4257 int try_release_extent_mapping(struct extent_map_tree *map,
4258                                struct extent_io_tree *tree, struct page *page,
4259                                gfp_t mask)
4260 {
4261         struct extent_map *em;
4262         u64 start = page_offset(page);
4263         u64 end = start + PAGE_SIZE - 1;
4264 
4265         if (gfpflags_allow_blocking(mask) &&
4266             page->mapping->host->i_size > SZ_16M) {
4267                 u64 len;
4268                 while (start <= end) {
4269                         len = end - start + 1;
4270                         write_lock(&map->lock);
4271                         em = lookup_extent_mapping(map, start, len);
4272                         if (!em) {
4273                                 write_unlock(&map->lock);
4274                                 break;
4275                         }
4276                         if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
4277                             em->start != start) {
4278                                 write_unlock(&map->lock);
4279                                 free_extent_map(em);
4280                                 break;
4281                         }
4282                         if (!test_range_bit(tree, em->start,
4283                                             extent_map_end(em) - 1,
4284                                             EXTENT_LOCKED | EXTENT_WRITEBACK,
4285                                             0, NULL)) {
4286                                 remove_extent_mapping(map, em);
4287                                 /* once for the rb tree */
4288                                 free_extent_map(em);
4289                         }
4290                         start = extent_map_end(em);
4291                         write_unlock(&map->lock);
4292 
4293                         /* once for us */
4294                         free_extent_map(em);
4295                 }
4296         }
4297         return try_release_extent_state(map, tree, page, mask);
4298 }
4299 
4300 /*
4301  * helper function for fiemap, which doesn't want to see any holes.
4302  * This maps until we find something past 'last'
4303  */
4304 static struct extent_map *get_extent_skip_holes(struct inode *inode,
4305                                                 u64 offset,
4306                                                 u64 last,
4307                                                 get_extent_t *get_extent)
4308 {
4309         u64 sectorsize = btrfs_inode_sectorsize(inode);
4310         struct extent_map *em;
4311         u64 len;
4312 
4313         if (offset >= last)
4314                 return NULL;
4315 
4316         while (1) {
4317                 len = last - offset;
4318                 if (len == 0)
4319                         break;
4320                 len = ALIGN(len, sectorsize);
4321                 em = get_extent(BTRFS_I(inode), NULL, 0, offset, len, 0);
4322                 if (IS_ERR_OR_NULL(em))
4323                         return em;
4324 
4325                 /* if this isn't a hole return it */
4326                 if (!test_bit(EXTENT_FLAG_VACANCY, &em->flags) &&
4327                     em->block_start != EXTENT_MAP_HOLE) {
4328                         return em;
4329                 }
4330 
4331                 /* this is a hole, advance to the next extent */
4332                 offset = extent_map_end(em);
4333                 free_extent_map(em);
4334                 if (offset >= last)
4335                         break;
4336         }
4337         return NULL;
4338 }
4339 
4340 /*
4341  * To cache previous fiemap extent
4342  *
4343  * Will be used for merging fiemap extent
4344  */
4345 struct fiemap_cache {
4346         u64 offset;
4347         u64 phys;
4348         u64 len;
4349         u32 flags;
4350         bool cached;
4351 };
4352 
4353 /*
4354  * Helper to submit fiemap extent.
4355  *
4356  * Will try to merge current fiemap extent specified by @offset, @phys,
4357  * @len and @flags with cached one.
4358  * And only when we fails to merge, cached one will be submitted as
4359  * fiemap extent.
4360  *
4361  * Return value is the same as fiemap_fill_next_extent().
4362  */
4363 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
4364                                 struct fiemap_cache *cache,
4365                                 u64 offset, u64 phys, u64 len, u32 flags)
4366 {
4367         int ret = 0;
4368 
4369         if (!cache->cached)
4370                 goto assign;
4371 
4372         /*
4373          * Sanity check, extent_fiemap() should have ensured that new
4374          * fiemap extent won't overlap with cahced one.
4375          * Not recoverable.
4376          *
4377          * NOTE: Physical address can overlap, due to compression
4378          */
4379         if (cache->offset + cache->len > offset) {
4380                 WARN_ON(1);
4381                 return -EINVAL;
4382         }
4383 
4384         /*
4385          * Only merges fiemap extents if
4386          * 1) Their logical addresses are continuous
4387          *
4388          * 2) Their physical addresses are continuous
4389          *    So truly compressed (physical size smaller than logical size)
4390          *    extents won't get merged with each other
4391          *
4392          * 3) Share same flags except FIEMAP_EXTENT_LAST
4393          *    So regular extent won't get merged with prealloc extent
4394          */
4395         if (cache->offset + cache->len  == offset &&
4396             cache->phys + cache->len == phys  &&
4397             (cache->flags & ~FIEMAP_EXTENT_LAST) ==
4398                         (flags & ~FIEMAP_EXTENT_LAST)) {
4399                 cache->len += len;
4400                 cache->flags |= flags;
4401                 goto try_submit_last;
4402         }
4403 
4404         /* Not mergeable, need to submit cached one */
4405         ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4406                                       cache->len, cache->flags);
4407         cache->cached = false;
4408         if (ret)
4409                 return ret;
4410 assign:
4411         cache->cached = true;
4412         cache->offset = offset;
4413         cache->phys = phys;
4414         cache->len = len;
4415         cache->flags = flags;
4416 try_submit_last:
4417         if (cache->flags & FIEMAP_EXTENT_LAST) {
4418                 ret = fiemap_fill_next_extent(fieinfo, cache->offset,
4419                                 cache->phys, cache->len, cache->flags);
4420                 cache->cached = false;
4421         }
4422         return ret;
4423 }
4424 
4425 /*
4426  * Emit last fiemap cache
4427  *
4428  * The last fiemap cache may still be cached in the following case:
4429  * 0                  4k                    8k
4430  * |<- Fiemap range ->|
4431  * |<------------  First extent ----------->|
4432  *
4433  * In this case, the first extent range will be cached but not emitted.
4434  * So we must emit it before ending extent_fiemap().
4435  */
4436 static int emit_last_fiemap_cache(struct btrfs_fs_info *fs_info,
4437                                   struct fiemap_extent_info *fieinfo,
4438                                   struct fiemap_cache *cache)
4439 {
4440         int ret;
4441 
4442         if (!cache->cached)
4443                 return 0;
4444 
4445         ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
4446                                       cache->len, cache->flags);
4447         cache->cached = false;
4448         if (ret > 0)
4449                 ret = 0;
4450         return ret;
4451 }
4452 
4453 int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4454                 __u64 start, __u64 len, get_extent_t *get_extent)
4455 {
4456         int ret = 0;
4457         u64 off = start;
4458         u64 max = start + len;
4459         u32 flags = 0;
4460         u32 found_type;
4461         u64 last;
4462         u64 last_for_get_extent = 0;
4463         u64 disko = 0;
4464         u64 isize = i_size_read(inode);
4465         struct btrfs_key found_key;
4466         struct extent_map *em = NULL;
4467         struct extent_state *cached_state = NULL;
4468         struct btrfs_path *path;
4469         struct btrfs_root *root = BTRFS_I(inode)->root;
4470         struct fiemap_cache cache = { 0 };
4471         int end = 0;
4472         u64 em_start = 0;
4473         u64 em_len = 0;
4474         u64 em_end = 0;
4475 
4476         if (len == 0)
4477                 return -EINVAL;
4478 
4479         path = btrfs_alloc_path();
4480         if (!path)
4481                 return -ENOMEM;
4482         path->leave_spinning = 1;
4483 
4484         start = round_down(start, btrfs_inode_sectorsize(inode));
4485         len = round_up(max, btrfs_inode_sectorsize(inode)) - start;
4486 
4487         /*
4488          * lookup the last file extent.  We're not using i_size here
4489          * because there might be preallocation past i_size
4490          */
4491         ret = btrfs_lookup_file_extent(NULL, root, path,
4492                         btrfs_ino(BTRFS_I(inode)), -1, 0);
4493         if (ret < 0) {
4494                 btrfs_free_path(path);
4495                 return ret;
4496         } else {
4497                 WARN_ON(!ret);
4498                 if (ret == 1)
4499                         ret = 0;
4500         }
4501 
4502         path->slots[0]--;
4503         btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
4504         found_type = found_key.type;
4505 
4506         /* No extents, but there might be delalloc bits */
4507         if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) ||
4508             found_type != BTRFS_EXTENT_DATA_KEY) {
4509                 /* have to trust i_size as the end */
4510                 last = (u64)-1;
4511                 last_for_get_extent = isize;
4512         } else {
4513                 /*
4514                  * remember the start of the last extent.  There are a
4515                  * bunch of different factors that go into the length of the
4516                  * extent, so its much less complex to remember where it started
4517                  */
4518                 last = found_key.offset;
4519                 last_for_get_extent = last + 1;
4520         }
4521         btrfs_release_path(path);
4522 
4523         /*
4524          * we might have some extents allocated but more delalloc past those
4525          * extents.  so, we trust isize unless the start of the last extent is
4526          * beyond isize
4527          */
4528         if (last < isize) {
4529                 last = (u64)-1;
4530                 last_for_get_extent = isize;
4531         }
4532 
4533         lock_extent_bits(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4534                          &cached_state);
4535 
4536         em = get_extent_skip_holes(inode, start, last_for_get_extent,
4537                                    get_extent);
4538         if (!em)
4539                 goto out;
4540         if (IS_ERR(em)) {
4541                 ret = PTR_ERR(em);
4542                 goto out;
4543         }
4544 
4545         while (!end) {
4546                 u64 offset_in_extent = 0;
4547 
4548                 /* break if the extent we found is outside the range */
4549                 if (em->start >= max || extent_map_end(em) < off)
4550                         break;
4551 
4552                 /*
4553                  * get_extent may return an extent that starts before our
4554                  * requested range.  We have to make sure the ranges
4555                  * we return to fiemap always move forward and don't
4556                  * overlap, so adjust the offsets here
4557                  */
4558                 em_start = max(em->start, off);
4559 
4560                 /*
4561                  * record the offset from the start of the extent
4562                  * for adjusting the disk offset below.  Only do this if the
4563                  * extent isn't compressed since our in ram offset may be past
4564                  * what we have actually allocated on disk.
4565                  */
4566                 if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4567                         offset_in_extent = em_start - em->start;
4568                 em_end = extent_map_end(em);
4569                 em_len = em_end - em_start;
4570                 disko = 0;
4571                 flags = 0;
4572 
4573                 /*
4574                  * bump off for our next call to get_extent
4575                  */
4576                 off = extent_map_end(em);
4577                 if (off >= max)
4578                         end = 1;
4579 
4580                 if (em->block_start == EXTENT_MAP_LAST_BYTE) {
4581                         end = 1;
4582                         flags |= FIEMAP_EXTENT_LAST;
4583                 } else if (em->block_start == EXTENT_MAP_INLINE) {
4584                         flags |= (FIEMAP_EXTENT_DATA_INLINE |
4585                                   FIEMAP_EXTENT_NOT_ALIGNED);
4586                 } else if (em->block_start == EXTENT_MAP_DELALLOC) {
4587                         flags |= (FIEMAP_EXTENT_DELALLOC |
4588                                   FIEMAP_EXTENT_UNKNOWN);
4589                 } else if (fieinfo->fi_extents_max) {
4590                         u64 bytenr = em->block_start -
4591                                 (em->start - em->orig_start);
4592 
4593                         disko = em->block_start + offset_in_extent;
4594 
4595                         /*
4596                          * As btrfs supports shared space, this information
4597                          * can be exported to userspace tools via
4598                          * flag FIEMAP_EXTENT_SHARED.  If fi_extents_max == 0
4599                          * then we're just getting a count and we can skip the
4600                          * lookup stuff.
4601                          */
4602                         ret = btrfs_check_shared(root,
4603                                                  btrfs_ino(BTRFS_I(inode)),
4604                                                  bytenr);
4605                         if (ret < 0)
4606                                 goto out_free;
4607                         if (ret)
4608                                 flags |= FIEMAP_EXTENT_SHARED;
4609                         ret = 0;
4610                 }
4611                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
4612                         flags |= FIEMAP_EXTENT_ENCODED;
4613                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4614                         flags |= FIEMAP_EXTENT_UNWRITTEN;
4615 
4616                 free_extent_map(em);
4617                 em = NULL;
4618                 if ((em_start >= last) || em_len == (u64)-1 ||
4619                    (last == (u64)-1 && isize <= em_end)) {
4620                         flags |= FIEMAP_EXTENT_LAST;
4621                         end = 1;
4622                 }
4623 
4624                 /* now scan forward to see if this is really the last extent. */
4625                 em = get_extent_skip_holes(inode, off, last_for_get_extent,
4626                                            get_extent);
4627                 if (IS_ERR(em)) {
4628                         ret = PTR_ERR(em);
4629                         goto out;
4630                 }
4631                 if (!em) {
4632                         flags |= FIEMAP_EXTENT_LAST;
4633                         end = 1;
4634                 }
4635                 ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko,
4636                                            em_len, flags);
4637                 if (ret) {
4638                         if (ret == 1)
4639                                 ret = 0;
4640                         goto out_free;
4641                 }
4642         }
4643 out_free:
4644         if (!ret)
4645                 ret = emit_last_fiemap_cache(root->fs_info, fieinfo, &cache);
4646         free_extent_map(em);
4647 out:
4648         btrfs_free_path(path);
4649         unlock_extent_cached(&BTRFS_I(inode)->io_tree, start, start + len - 1,
4650                              &cached_state, GFP_NOFS);
4651         return ret;
4652 }
4653 
4654 static void __free_extent_buffer(struct extent_buffer *eb)
4655 {
4656         btrfs_leak_debug_del(&eb->leak_list);
4657         kmem_cache_free(extent_buffer_cache, eb);
4658 }
4659 
4660 int extent_buffer_under_io(struct extent_buffer *eb)
4661 {
4662         return (atomic_read(&eb->io_pages) ||
4663                 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4664                 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4665 }
4666 
4667 /*
4668  * Helper for releasing extent buffer page.
4669  */
4670 static void btrfs_release_extent_buffer_page(struct extent_buffer *eb)
4671 {
4672         unsigned long index;
4673         struct page *page;
4674         int mapped = !test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4675 
4676         BUG_ON(extent_buffer_under_io(eb));
4677 
4678         index = num_extent_pages(eb->start, eb->len);
4679         if (index == 0)
4680                 return;
4681 
4682         do {
4683                 index--;
4684                 page = eb->pages[index];
4685                 if (!page)
4686                         continue;
4687                 if (mapped)
4688                         spin_lock(&page->mapping->private_lock);
4689                 /*
4690                  * We do this since we'll remove the pages after we've
4691                  * removed the eb from the radix tree, so we could race
4692                  * and have this page now attached to the new eb.  So
4693                  * only clear page_private if it's still connected to
4694                  * this eb.
4695                  */
4696                 if (PagePrivate(page) &&
4697                     page->private == (unsigned long)eb) {
4698                         BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4699                         BUG_ON(PageDirty(page));
4700                         BUG_ON(PageWriteback(page));
4701                         /*
4702                          * We need to make sure we haven't be attached
4703                          * to a new eb.
4704                          */
4705                         ClearPagePrivate(page);
4706                         set_page_private(page, 0);
4707                         /* One for the page private */
4708                         put_page(page);
4709                 }
4710 
4711                 if (mapped)
4712                         spin_unlock(&page->mapping->private_lock);
4713 
4714                 /* One for when we allocated the page */
4715                 put_page(page);
4716         } while (index != 0);
4717 }
4718 
4719 /*
4720  * Helper for releasing the extent buffer.
4721  */
4722 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4723 {
4724         btrfs_release_extent_buffer_page(eb);
4725         __free_extent_buffer(eb);
4726 }
4727 
4728 static struct extent_buffer *
4729 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4730                       unsigned long len)
4731 {
4732         struct extent_buffer *eb = NULL;
4733 
4734         eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4735         eb->start = start;
4736         eb->len = len;
4737         eb->fs_info = fs_info;
4738         eb->bflags = 0;
4739         rwlock_init(&eb->lock);
4740         atomic_set(&eb->write_locks, 0);
4741         atomic_set(&eb->read_locks, 0);
4742         atomic_set(&eb->blocking_readers, 0);
4743         atomic_set(&eb->blocking_writers, 0);
4744         atomic_set(&eb->spinning_readers, 0);
4745         atomic_set(&eb->spinning_writers, 0);
4746         eb->lock_nested = 0;
4747         init_waitqueue_head(&eb->write_lock_wq);
4748         init_waitqueue_head(&eb->read_lock_wq);
4749 
4750         btrfs_leak_debug_add(&eb->leak_list, &buffers);
4751 
4752         spin_lock_init(&eb->refs_lock);
4753         atomic_set(&eb->refs, 1);
4754         atomic_set(&eb->io_pages, 0);
4755 
4756         /*
4757          * Sanity checks, currently the maximum is 64k covered by 16x 4k pages
4758          */
4759         BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE
4760                 > MAX_INLINE_EXTENT_BUFFER_SIZE);
4761         BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE);
4762 
4763         return eb;
4764 }
4765 
4766 struct extent_buffer *btrfs_clone_extent_buffer(struct extent_buffer *src)
4767 {
4768         unsigned long i;
4769         struct page *p;
4770         struct extent_buffer *new;
4771         unsigned long num_pages = num_extent_pages(src->start, src->len);
4772 
4773         new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4774         if (new == NULL)
4775                 return NULL;
4776 
4777         for (i = 0; i < num_pages; i++) {
4778                 p = alloc_page(GFP_NOFS);
4779                 if (!p) {
4780                         btrfs_release_extent_buffer(new);
4781                         return NULL;
4782                 }
4783                 attach_extent_buffer_page(new, p);
4784                 WARN_ON(PageDirty(p));
4785                 SetPageUptodate(p);
4786                 new->pages[i] = p;
4787                 copy_page(page_address(p), page_address(src->pages[i]));
4788         }
4789 
4790         set_bit(EXTENT_BUFFER_UPTODATE, &new->bflags);
4791         set_bit(EXTENT_BUFFER_DUMMY, &new->bflags);
4792 
4793         return new;
4794 }
4795 
4796 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4797                                                   u64 start, unsigned long len)
4798 {
4799         struct extent_buffer *eb;
4800         unsigned long num_pages;
4801         unsigned long i;
4802 
4803         num_pages = num_extent_pages(start, len);
4804 
4805         eb = __alloc_extent_buffer(fs_info, start, len);
4806         if (!eb)
4807                 return NULL;
4808 
4809         for (i = 0; i < num_pages; i++) {
4810                 eb->pages[i] = alloc_page(GFP_NOFS);
4811                 if (!eb->pages[i])
4812                         goto err;
4813         }
4814         set_extent_buffer_uptodate(eb);
4815         btrfs_set_header_nritems(eb, 0);
4816         set_bit(EXTENT_BUFFER_DUMMY, &eb->bflags);
4817 
4818         return eb;
4819 err:
4820         for (; i > 0; i--)
4821                 __free_page(eb->pages[i - 1]);
4822         __free_extent_buffer(eb);
4823         return NULL;
4824 }
4825 
4826 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4827                                                 u64 start)
4828 {
4829         return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4830 }
4831 
4832 static void check_buffer_tree_ref(struct extent_buffer *eb)
4833 {
4834         int refs;
4835         /* the ref bit is tricky.  We have to make sure it is set
4836          * if we have the buffer dirty.   Otherwise the
4837          * code to free a buffer can end up dropping a dirty
4838          * page
4839          *
4840          * Once the ref bit is set, it won't go away while the
4841          * buffer is dirty or in writeback, and it also won't
4842          * go away while we have the reference count on the
4843          * eb bumped.
4844          *
4845          * We can't just set the ref bit without bumping the
4846          * ref on the eb because free_extent_buffer might
4847          * see the ref bit and try to clear it.  If this happens
4848          * free_extent_buffer might end up dropping our original
4849          * ref by mistake and freeing the page before we are able
4850          * to add one more ref.
4851          *
4852          * So bump the ref count first, then set the bit.  If someone
4853          * beat us to it, drop the ref we added.
4854          */
4855         refs = atomic_read(&eb->refs);
4856         if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4857                 return;
4858 
4859         spin_lock(&eb->refs_lock);
4860         if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4861                 atomic_inc(&eb->refs);
4862         spin_unlock(&eb->refs_lock);
4863 }
4864 
4865 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4866                 struct page *accessed)
4867 {
4868         unsigned long num_pages, i;
4869 
4870         check_buffer_tree_ref(eb);
4871 
4872         num_pages = num_extent_pages(eb->start, eb->len);
4873         for (i = 0; i < num_pages; i++) {
4874                 struct page *p = eb->pages[i];
4875 
4876                 if (p != accessed)
4877                         mark_page_accessed(p);
4878         }
4879 }
4880 
4881 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4882                                          u64 start)
4883 {
4884         struct extent_buffer *eb;
4885 
4886         rcu_read_lock();
4887         eb = radix_tree_lookup(&fs_info->buffer_radix,
4888                                start >> PAGE_SHIFT);
4889         if (eb && atomic_inc_not_zero(&eb->refs)) {
4890                 rcu_read_unlock();
4891                 /*
4892                  * Lock our eb's refs_lock to avoid races with
4893                  * free_extent_buffer. When we get our eb it might be flagged
4894                  * with EXTENT_BUFFER_STALE and another task running
4895                  * free_extent_buffer might have seen that flag set,
4896                  * eb->refs == 2, that the buffer isn't under IO (dirty and
4897                  * writeback flags not set) and it's still in the tree (flag
4898                  * EXTENT_BUFFER_TREE_REF set), therefore being in the process
4899                  * of decrementing the extent buffer's reference count twice.
4900                  * So here we could race and increment the eb's reference count,
4901                  * clear its stale flag, mark it as dirty and drop our reference
4902                  * before the other task finishes executing free_extent_buffer,
4903                  * which would later result in an attempt to free an extent
4904                  * buffer that is dirty.
4905                  */
4906                 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4907                         spin_lock(&eb->refs_lock);
4908                         spin_unlock(&eb->refs_lock);
4909                 }
4910                 mark_extent_buffer_accessed(eb, NULL);
4911                 return eb;
4912         }
4913         rcu_read_unlock();
4914 
4915         return NULL;
4916 }
4917 
4918 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4919 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4920                                         u64 start)
4921 {
4922         struct extent_buffer *eb, *exists = NULL;
4923         int ret;
4924 
4925         eb = find_extent_buffer(fs_info, start);
4926         if (eb)
4927                 return eb;
4928         eb = alloc_dummy_extent_buffer(fs_info, start);
4929         if (!eb)
4930                 return NULL;
4931         eb->fs_info = fs_info;
4932 again:
4933         ret = radix_tree_preload(GFP_NOFS);
4934         if (ret)
4935                 goto free_eb;
4936         spin_lock(&fs_info->buffer_lock);
4937         ret = radix_tree_insert(&fs_info->buffer_radix,
4938                                 start >> PAGE_SHIFT, eb);
4939         spin_unlock(&fs_info->buffer_lock);
4940         radix_tree_preload_end();
4941         if (ret == -EEXIST) {
4942                 exists = find_extent_buffer(fs_info, start);
4943                 if (exists)
4944                         goto free_eb;
4945                 else
4946                         goto again;
4947         }
4948         check_buffer_tree_ref(eb);
4949         set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4950 
4951         /*
4952          * We will free dummy extent buffer's if they come into
4953          * free_extent_buffer with a ref count of 2, but if we are using this we
4954          * want the buffers to stay in memory until we're done with them, so
4955          * bump the ref count again.
4956          */
4957         atomic_inc(&eb->refs);
4958         return eb;
4959 free_eb:
4960         btrfs_release_extent_buffer(eb);
4961         return exists;
4962 }
4963 #endif
4964 
4965 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4966                                           u64 start)
4967 {
4968         unsigned long len = fs_info->nodesize;
4969         unsigned long num_pages = num_extent_pages(start, len);
4970         unsigned long i;
4971         unsigned long index = start >> PAGE_SHIFT;
4972         struct extent_buffer *eb;
4973         struct extent_buffer *exists = NULL;
4974         struct page *p;
4975         struct address_space *mapping = fs_info->btree_inode->i_mapping;
4976         int uptodate = 1;
4977         int ret;
4978 
4979         if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4980                 btrfs_err(fs_info, "bad tree block start %llu", start);
4981                 return ERR_PTR(-EINVAL);
4982         }
4983 
4984         eb = find_extent_buffer(fs_info, start);
4985         if (eb)
4986                 return eb;
4987 
4988         eb = __alloc_extent_buffer(fs_info, start, len);
4989         if (!eb)
4990                 return ERR_PTR(-ENOMEM);
4991 
4992         for (i = 0; i < num_pages; i++, index++) {
4993                 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4994                 if (!p) {
4995                         exists = ERR_PTR(-ENOMEM);
4996                         goto free_eb;
4997                 }
4998 
4999                 spin_lock(&mapping->private_lock);
5000                 if (PagePrivate(p)) {
5001                         /*
5002                          * We could have already allocated an eb for this page
5003                          * and attached one so lets see if we can get a ref on
5004                          * the existing eb, and if we can we know it's good and
5005                          * we can just return that one, else we know we can just
5006                          * overwrite page->private.
5007                          */
5008                         exists = (struct extent_buffer *)p->private;
5009                         if (atomic_inc_not_zero(&exists->refs)) {
5010                                 spin_unlock(&mapping->private_lock);
5011                                 unlock_page(p);
5012                                 put_page(p);
5013                                 mark_extent_buffer_accessed(exists, p);
5014                                 goto free_eb;
5015                         }
5016                         exists = NULL;
5017 
5018                         /*
5019                          * Do this so attach doesn't complain and we need to
5020                          * drop the ref the old guy had.
5021                          */
5022                         ClearPagePrivate(p);
5023                         WARN_ON(PageDirty(p));
5024                         put_page(p);
5025                 }
5026                 attach_extent_buffer_page(eb, p);
5027                 spin_unlock(&mapping->private_lock);
5028                 WARN_ON(PageDirty(p));
5029                 eb->pages[i] = p;
5030                 if (!PageUptodate(p))
5031                         uptodate = 0;
5032 
5033                 /*
5034                  * see below about how we avoid a nasty race with release page
5035                  * and why we unlock later
5036                  */
5037         }
5038         if (uptodate)
5039                 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5040 again:
5041         ret = radix_tree_preload(GFP_NOFS);
5042         if (ret) {
5043                 exists = ERR_PTR(ret);
5044                 goto free_eb;
5045         }
5046 
5047         spin_lock(&fs_info->buffer_lock);
5048         ret = radix_tree_insert(&fs_info->buffer_radix,
5049                                 start >> PAGE_SHIFT, eb);
5050         spin_unlock(&fs_info->buffer_lock);
5051         radix_tree_preload_end();
5052         if (ret == -EEXIST) {
5053                 exists = find_extent_buffer(fs_info, start);
5054                 if (exists)
5055                         goto free_eb;
5056                 else
5057                         goto again;
5058         }
5059         /* add one reference for the tree */
5060         check_buffer_tree_ref(eb);
5061         set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
5062 
5063         /*
5064          * there is a race where release page may have
5065          * tried to find this extent buffer in the radix
5066          * but failed.  It will tell the VM it is safe to
5067          * reclaim the, and it will clear the page private bit.
5068          * We must make sure to set the page private bit properly
5069          * after the extent buffer is in the radix tree so
5070          * it doesn't get lost
5071          */
5072         SetPageChecked(eb->pages[0]);
5073         for (i = 1; i < num_pages; i++) {
5074                 p = eb->pages[i];
5075                 ClearPageChecked(p);
5076                 unlock_page(p);
5077         }
5078         unlock_page(eb->pages[0]);
5079         return eb;
5080 
5081 free_eb:
5082         WARN_ON(!atomic_dec_and_test(&eb->refs));
5083         for (i = 0; i < num_pages; i++) {
5084                 if (eb->pages[i])
5085                         unlock_page(eb->pages[i]);
5086         }
5087 
5088         btrfs_release_extent_buffer(eb);
5089         return exists;
5090 }
5091 
5092 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
5093 {
5094         struct extent_buffer *eb =
5095                         container_of(head, struct extent_buffer, rcu_head);
5096 
5097         __free_extent_buffer(eb);
5098 }
5099 
5100 /* Expects to have eb->eb_lock already held */
5101 static int release_extent_buffer(struct extent_buffer *eb)
5102 {
5103         WARN_ON(atomic_read(&eb->refs) == 0);
5104         if (atomic_dec_and_test(&eb->refs)) {
5105                 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
5106                         struct btrfs_fs_info *fs_info = eb->fs_info;
5107 
5108                         spin_unlock(&eb->refs_lock);
5109 
5110                         spin_lock(&fs_info->buffer_lock);
5111                         radix_tree_delete(&fs_info->buffer_radix,
5112                                           eb->start >> PAGE_SHIFT);
5113                         spin_unlock(&fs_info->buffer_lock);
5114                 } else {
5115                         spin_unlock(&eb->refs_lock);
5116                 }
5117 
5118                 /* Should be safe to release our pages at this point */
5119                 btrfs_release_extent_buffer_page(eb);
5120 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5121                 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))) {
5122                         __free_extent_buffer(eb);
5123                         return 1;
5124                 }
5125 #endif
5126                 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
5127                 return 1;
5128         }
5129         spin_unlock(&eb->refs_lock);
5130 
5131         return 0;
5132 }
5133 
5134 void free_extent_buffer(struct extent_buffer *eb)
5135 {
5136         int refs;
5137         int old;
5138         if (!eb)
5139                 return;
5140 
5141         while (1) {
5142                 refs = atomic_read(&eb->refs);
5143                 if (refs <= 3)
5144                         break;
5145                 old = atomic_cmpxchg(&eb->refs, refs, refs - 1);
5146                 if (old == refs)
5147                         return;
5148         }
5149 
5150         spin_lock(&eb->refs_lock);
5151         if (atomic_read(&eb->refs) == 2 &&
5152             test_bit(EXTENT_BUFFER_DUMMY, &eb->bflags))
5153                 atomic_dec(&eb->refs);
5154 
5155         if (atomic_read(&eb->refs) == 2 &&
5156             test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
5157             !extent_buffer_under_io(eb) &&
5158             test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5159                 atomic_dec(&eb->refs);
5160 
5161         /*
5162          * I know this is terrible, but it's temporary until we stop tracking
5163          * the uptodate bits and such for the extent buffers.
5164          */
5165         release_extent_buffer(eb);
5166 }
5167 
5168 void free_extent_buffer_stale(struct extent_buffer *eb)
5169 {
5170         if (!eb)
5171                 return;
5172 
5173         spin_lock(&eb->refs_lock);
5174         set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
5175 
5176         if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
5177             test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
5178                 atomic_dec(&eb->refs);
5179         release_extent_buffer(eb);
5180 }
5181 
5182 void clear_extent_buffer_dirty(struct extent_buffer *eb)
5183 {
5184         unsigned long i;
5185         unsigned long num_pages;
5186         struct page *page;
5187 
5188         num_pages = num_extent_pages(eb->start, eb->len);
5189 
5190         for (i = 0; i < num_pages; i++) {
5191                 page = eb->pages[i];
5192                 if (!PageDirty(page))
5193                         continue;
5194 
5195                 lock_page(page);
5196                 WARN_ON(!PagePrivate(page));
5197 
5198                 clear_page_dirty_for_io(page);
5199                 spin_lock_irq(&page->mapping->tree_lock);
5200                 if (!PageDirty(page)) {
5201                         radix_tree_tag_clear(&page->mapping->page_tree,
5202                                                 page_index(page),
5203                                                 PAGECACHE_TAG_DIRTY);
5204                 }
5205                 spin_unlock_irq(&page->mapping->tree_lock);
5206                 ClearPageError(page);
5207                 unlock_page(page);
5208         }
5209         WARN_ON(atomic_read(&eb->refs) == 0);
5210 }
5211 
5212 int set_extent_buffer_dirty(struct extent_buffer *eb)
5213 {
5214         unsigned long i;
5215         unsigned long num_pages;
5216         int was_dirty = 0;
5217 
5218         check_buffer_tree_ref(eb);
5219 
5220         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
5221 
5222         num_pages = num_extent_pages(eb->start, eb->len);
5223         WARN_ON(atomic_read(&eb->refs) == 0);
5224         WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
5225 
5226         for (i = 0; i < num_pages; i++)
5227                 set_page_dirty(eb->pages[i]);
5228         return was_dirty;
5229 }
5230 
5231 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
5232 {
5233         unsigned long i;
5234         struct page *page;
5235         unsigned long num_pages;
5236 
5237         clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5238         num_pages = num_extent_pages(eb->start, eb->len);
5239         for (i = 0; i < num_pages; i++) {
5240                 page = eb->pages[i];
5241                 if (page)
5242                         ClearPageUptodate(page);
5243         }
5244 }
5245 
5246 void set_extent_buffer_uptodate(struct extent_buffer *eb)
5247 {
5248         unsigned long i;
5249         struct page *page;
5250         unsigned long num_pages;
5251 
5252         set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5253         num_pages = num_extent_pages(eb->start, eb->len);
5254         for (i = 0; i < num_pages; i++) {
5255                 page = eb->pages[i];
5256                 SetPageUptodate(page);
5257         }
5258 }
5259 
5260 int extent_buffer_uptodate(struct extent_buffer *eb)
5261 {
5262         return test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5263 }
5264 
5265 int read_extent_buffer_pages(struct extent_io_tree *tree,
5266                              struct extent_buffer *eb, int wait,
5267                              get_extent_t *get_extent, int mirror_num)
5268 {
5269         unsigned long i;
5270         struct page *page;
5271         int err;
5272         int ret = 0;
5273         int locked_pages = 0;
5274         int all_uptodate = 1;
5275         unsigned long num_pages;
5276         unsigned long num_reads = 0;
5277         struct bio *bio = NULL;
5278         unsigned long bio_flags = 0;
5279 
5280         if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
5281                 return 0;
5282 
5283         num_pages = num_extent_pages(eb->start, eb->len);
5284         for (i = 0; i < num_pages; i++) {
5285                 page = eb->pages[i];
5286                 if (wait == WAIT_NONE) {
5287                         if (!trylock_page(page))
5288                                 goto unlock_exit;
5289                 } else {
5290                         lock_page(page);
5291                 }
5292                 locked_pages++;
5293         }
5294         /*
5295          * We need to firstly lock all pages to make sure that
5296          * the uptodate bit of our pages won't be affected by
5297          * clear_extent_buffer_uptodate().
5298          */
5299         for (i = 0; i < num_pages; i++) {
5300                 page = eb->pages[i];
5301                 if (!PageUptodate(page)) {
5302                         num_reads++;
5303                         all_uptodate = 0;
5304                 }
5305         }
5306 
5307         if (all_uptodate) {
5308                 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
5309                 goto unlock_exit;
5310         }
5311 
5312         clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
5313         eb->read_mirror = 0;
5314         atomic_set(&eb->io_pages, num_reads);
5315         for (i = 0; i < num_pages; i++) {
5316                 page = eb->pages[i];
5317 
5318                 if (!PageUptodate(page)) {
5319                         if (ret) {
5320                                 atomic_dec(&eb->io_pages);
5321                                 unlock_page(page);
5322                                 continue;
5323                         }
5324 
5325                         ClearPageError(page);
5326                         err = __extent_read_full_page(tree, page,
5327                                                       get_extent, &bio,
5328                                                       mirror_num, &bio_flags,
5329                                                       REQ_META);
5330                         if (err) {
5331                                 ret = err;
5332                                 /*
5333                                  * We use &bio in above __extent_read_full_page,
5334                                  * so we ensure that if it returns error, the
5335                                  * current page fails to add itself to bio and
5336                                  * it's been unlocked.
5337                                  *
5338                                  * We must dec io_pages by ourselves.
5339                                  */
5340                                 atomic_dec(&eb->io_pages);
5341                         }
5342                 } else {
5343                         unlock_page(page);
5344                 }
5345         }
5346 
5347         if (bio) {
5348                 err = submit_one_bio(bio, mirror_num, bio_flags);
5349                 if (err)
5350                         return err;
5351         }
5352 
5353         if (ret || wait != WAIT_COMPLETE)
5354                 return ret;
5355 
5356         for (i = 0; i < num_pages; i++) {
5357                 page = eb->pages[i];
5358                 wait_on_page_locked(page);
5359                 if (!PageUptodate(page))
5360                         ret = -EIO;
5361         }
5362 
5363         return ret;
5364 
5365 unlock_exit:
5366         while (locked_pages > 0) {
5367                 locked_pages--;
5368                 page = eb->pages[locked_pages];
5369                 unlock_page(page);
5370         }
5371         return ret;
5372 }
5373 
5374 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5375                         unsigned long start, unsigned long len)
5376 {
5377         size_t cur;
5378         size_t offset;
5379         struct page *page;
5380         char *kaddr;
5381         char *dst = (char *)dstv;
5382         size_t start_offset = eb->start & ((u64)PAGE_SIZE - 1);
5383         unsigned long i = (start_offset + start) >> PAGE_SHIFT;
5384 
5385         if (start + len > eb->len) {
5386                 WARN(1, KERN_ERR "btrfs bad mapping eb start %llu len %lu, wanted %lu %lu\n",
5387                      eb->start, eb->len, start, len);
5388                 memset(dst, 0, len);
5389                 return;
5390         }
5391 
5392         offset = (start_offset + start) & (PAGE_SIZE - 1);
5393 
5394         while (len > 0) {
5395                 page = eb->pages[i];
5396 
5397                 cur = min(len, (PAGE_SIZE - offset));
5398                 kaddr = page_address(page);
5399                 memcpy(dst, kaddr + offset, cur);
5400 
5401                 dst += cur;
5402