aboutsummaryrefslogtreecommitdiffstats
path: root/fs
diff options
context:
space:
mode:
authorChris Mason <chris.mason@oracle.com>2007-08-27 16:49:44 -0400
committerDavid Woodhouse <dwmw2@hera.kernel.org>2007-08-27 16:49:44 -0400
commita52d9a8033c454cd9b4697cfafb467509fc1693f (patch)
tree83d8bbf2df29598ea9c60498fadf5990262a11c8 /fs
parent83df7c1d8b885e216c68c3f0469bec9b93666b74 (diff)
Btrfs: Extent based page cache code. This uses an rbtree of extents and tests
instead of buffer heads. Signed-off-by: Chris Mason <chris.mason@oracle.com>
Diffstat (limited to 'fs')
-rw-r--r--fs/btrfs/Makefile3
-rw-r--r--fs/btrfs/btrfs_inode.h3
-rw-r--r--fs/btrfs/ctree.h7
-rw-r--r--fs/btrfs/disk-io.h1
-rw-r--r--fs/btrfs/extent_map.c1656
-rw-r--r--fs/btrfs/extent_map.h89
-rw-r--r--fs/btrfs/file.c295
-rw-r--r--fs/btrfs/inode.c816
-rw-r--r--fs/btrfs/super.c2
9 files changed, 2203 insertions, 669 deletions
diff --git a/fs/btrfs/Makefile b/fs/btrfs/Makefile
index 9321438c5eef..ea7a22870350 100644
--- a/fs/btrfs/Makefile
+++ b/fs/btrfs/Makefile
@@ -4,7 +4,8 @@ ifneq ($(KERNELRELEASE),)
4obj-m := btrfs.o 4obj-m := btrfs.o
5btrfs-y := super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.o \ 5btrfs-y := super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.o \
6 hash.o file-item.o inode-item.o inode-map.o disk-io.o \ 6 hash.o file-item.o inode-item.o inode-map.o disk-io.o \
7 transaction.o bit-radix.o inode.o file.o tree-defrag.o 7 transaction.o bit-radix.o inode.o file.o tree-defrag.o \
8 extent_map.o
8 9
9#btrfs-y := ctree.o disk-io.o radix-tree.o extent-tree.o print-tree.o \ 10#btrfs-y := ctree.o disk-io.o radix-tree.o extent-tree.o print-tree.o \
10# root-tree.o dir-item.o hash.o file-item.o inode-item.o \ 11# root-tree.o dir-item.o hash.o file-item.o inode-item.o \
diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index f446f161f736..d1d5af471c3c 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -19,11 +19,14 @@
19#ifndef __BTRFS_I__ 19#ifndef __BTRFS_I__
20#define __BTRFS_I__ 20#define __BTRFS_I__
21 21
22#include "extent_map.h"
23
22/* in memory btrfs inode */ 24/* in memory btrfs inode */
23struct btrfs_inode { 25struct btrfs_inode {
24 struct btrfs_root *root; 26 struct btrfs_root *root;
25 struct btrfs_block_group_cache *block_group; 27 struct btrfs_block_group_cache *block_group;
26 struct btrfs_key location; 28 struct btrfs_key location;
29 struct extent_map_tree extent_tree;
27 struct inode vfs_inode; 30 struct inode vfs_inode;
28 31
29 /* 32 /*
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index c7f5161271d3..947f061ed118 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -1176,7 +1176,14 @@ int btrfs_commit_write(struct file *file, struct page *page,
1176 unsigned from, unsigned to); 1176 unsigned from, unsigned to);
1177int btrfs_get_block(struct inode *inode, sector_t iblock, 1177int btrfs_get_block(struct inode *inode, sector_t iblock,
1178 struct buffer_head *result, int create); 1178 struct buffer_head *result, int create);
1179struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1180 size_t page_offset, u64 start, u64 end,
1181 int create);
1182int btrfs_update_inode(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct inode *inode);
1179/* file.c */ 1185/* file.c */
1186int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end);
1180extern struct file_operations btrfs_file_operations; 1187extern struct file_operations btrfs_file_operations;
1181int btrfs_drop_extents(struct btrfs_trans_handle *trans, 1188int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1182 struct btrfs_root *root, struct inode *inode, 1189 struct btrfs_root *root, struct inode *inode,
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index 81fd18cbd824..5261733b8735 100644
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -77,7 +77,6 @@ int btrfs_insert_dev_radix(struct btrfs_root *root,
77 u64 num_blocks); 77 u64 num_blocks);
78int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh, 78int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
79 u64 logical); 79 u64 logical);
80int btrfs_releasepage(struct page *page, gfp_t flags);
81void btrfs_btree_balance_dirty(struct btrfs_root *root); 80void btrfs_btree_balance_dirty(struct btrfs_root *root);
82int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root); 81int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root);
83void btrfs_mark_buffer_dirty(struct buffer_head *bh); 82void btrfs_mark_buffer_dirty(struct buffer_head *bh);
diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c
new file mode 100644
index 000000000000..d378edf0964e
--- /dev/null
+++ b/fs/btrfs/extent_map.c
@@ -0,0 +1,1656 @@
1#include <linux/bitops.h>
2#include <linux/slab.h>
3#include <linux/bio.h>
4#include <linux/mm.h>
5#include <linux/gfp.h>
6#include <linux/pagemap.h>
7#include <linux/page-flags.h>
8#include <linux/module.h>
9#include <linux/spinlock.h>
10#include <linux/blkdev.h>
11#include "extent_map.h"
12
13static struct kmem_cache *extent_map_cache;
14static struct kmem_cache *extent_state_cache;
15
16struct tree_entry {
17 u64 start;
18 u64 end;
19 int in_tree;
20 struct rb_node rb_node;
21};
22
23/* bits for the extent state */
24#define EXTENT_DIRTY 1
25#define EXTENT_WRITEBACK (1 << 1)
26#define EXTENT_UPTODATE (1 << 2)
27#define EXTENT_LOCKED (1 << 3)
28#define EXTENT_NEW (1 << 4)
29#define EXTENT_DELALLOC (1 << 5)
30
31#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK)
32
33static LIST_HEAD(all_states);
34spinlock_t state_lock = SPIN_LOCK_UNLOCKED;
35
36void __init extent_map_init(void)
37{
38 extent_map_cache = kmem_cache_create("extent_map",
39 sizeof(struct extent_map), 0,
40 SLAB_RECLAIM_ACCOUNT |
41 SLAB_DESTROY_BY_RCU,
42 NULL);
43 extent_state_cache = kmem_cache_create("extent_state",
44 sizeof(struct extent_state), 0,
45 SLAB_RECLAIM_ACCOUNT |
46 SLAB_DESTROY_BY_RCU,
47 NULL);
48}
49
50void __exit extent_map_exit(void)
51{
52 while(!list_empty(&all_states)) {
53 struct extent_state *state;
54 struct list_head *cur = all_states.next;
55 state = list_entry(cur, struct extent_state, list);
56 printk("found leaked state %Lu %Lu state %d in_tree %d\n",
57 state->start, state->end, state->state, state->in_tree);
58 list_del(&state->list);
59 kfree(state);
60 }
61 if (extent_map_cache)
62 kmem_cache_destroy(extent_map_cache);
63 if (extent_state_cache)
64 kmem_cache_destroy(extent_state_cache);
65}
66
67void extent_map_tree_init(struct extent_map_tree *tree,
68 struct address_space *mapping, gfp_t mask)
69{
70 tree->map.rb_node = NULL;
71 tree->state.rb_node = NULL;
72 rwlock_init(&tree->lock);
73 tree->mapping = mapping;
74}
75EXPORT_SYMBOL(extent_map_tree_init);
76
77struct extent_map *alloc_extent_map(gfp_t mask)
78{
79 struct extent_map *em;
80 em = kmem_cache_alloc(extent_map_cache, mask);
81 if (!em || IS_ERR(em))
82 return em;
83 em->in_tree = 0;
84 atomic_set(&em->refs, 1);
85 return em;
86}
87EXPORT_SYMBOL(alloc_extent_map);
88
89void free_extent_map(struct extent_map *em)
90{
91 if (atomic_dec_and_test(&em->refs)) {
92 WARN_ON(em->in_tree);
93 kmem_cache_free(extent_map_cache, em);
94 }
95}
96EXPORT_SYMBOL(free_extent_map);
97
98
99struct extent_state *alloc_extent_state(gfp_t mask)
100{
101 struct extent_state *state;
102 state = kmem_cache_alloc(extent_state_cache, mask);
103 if (!state || IS_ERR(state))
104 return state;
105 state->state = 0;
106 state->in_tree = 0;
107 atomic_set(&state->refs, 1);
108 init_waitqueue_head(&state->wq);
109 spin_lock_irq(&state_lock);
110 list_add(&state->list, &all_states);
111 spin_unlock_irq(&state_lock);
112 return state;
113}
114EXPORT_SYMBOL(alloc_extent_state);
115
116void free_extent_state(struct extent_state *state)
117{
118 if (atomic_dec_and_test(&state->refs)) {
119 WARN_ON(state->in_tree);
120 spin_lock_irq(&state_lock);
121 list_del_init(&state->list);
122 spin_unlock_irq(&state_lock);
123 kmem_cache_free(extent_state_cache, state);
124 }
125}
126EXPORT_SYMBOL(free_extent_state);
127
128static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
129 struct rb_node *node)
130{
131 struct rb_node ** p = &root->rb_node;
132 struct rb_node * parent = NULL;
133 struct tree_entry *entry;
134
135 while(*p) {
136 parent = *p;
137 entry = rb_entry(parent, struct tree_entry, rb_node);
138
139 if (offset < entry->start)
140 p = &(*p)->rb_left;
141 else if (offset > entry->end)
142 p = &(*p)->rb_right;
143 else
144 return parent;
145 }
146
147 entry = rb_entry(node, struct tree_entry, rb_node);
148 entry->in_tree = 1;
149 rb_link_node(node, parent, p);
150 rb_insert_color(node, root);
151 return NULL;
152}
153
154static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
155 struct rb_node **prev_ret)
156{
157 struct rb_node * n = root->rb_node;
158 struct rb_node *prev = NULL;
159 struct tree_entry *entry;
160 struct tree_entry *prev_entry = NULL;
161
162 while(n) {
163 entry = rb_entry(n, struct tree_entry, rb_node);
164 prev = n;
165 prev_entry = entry;
166
167 if (offset < entry->start)
168 n = n->rb_left;
169 else if (offset > entry->end)
170 n = n->rb_right;
171 else
172 return n;
173 }
174 if (!prev_ret)
175 return NULL;
176 while(prev && offset > prev_entry->end) {
177 prev = rb_next(prev);
178 prev_entry = rb_entry(prev, struct tree_entry, rb_node);
179 }
180 *prev_ret = prev;
181 return NULL;
182}
183
184static inline struct rb_node *tree_search(struct rb_root *root, u64 offset)
185{
186 struct rb_node *prev;
187 struct rb_node *ret;
188 ret = __tree_search(root, offset, &prev);
189 if (!ret)
190 return prev;
191 return ret;
192}
193
194static int tree_delete(struct rb_root *root, u64 offset)
195{
196 struct rb_node *node;
197 struct tree_entry *entry;
198
199 node = __tree_search(root, offset, NULL);
200 if (!node)
201 return -ENOENT;
202 entry = rb_entry(node, struct tree_entry, rb_node);
203 entry->in_tree = 0;
204 rb_erase(node, root);
205 return 0;
206}
207
208/*
209 * add_extent_mapping tries a simple backward merge with existing
210 * mappings. The extent_map struct passed in will be inserted into
211 * the tree directly (no copies made, just a reference taken).
212 */
213int add_extent_mapping(struct extent_map_tree *tree,
214 struct extent_map *em)
215{
216 int ret = 0;
217 struct extent_map *prev = NULL;
218 struct rb_node *rb;
219
220 write_lock_irq(&tree->lock);
221 rb = tree_insert(&tree->map, em->end, &em->rb_node);
222 if (rb) {
223 prev = rb_entry(rb, struct extent_map, rb_node);
224 printk("found extent map %Lu %Lu on insert of %Lu %Lu\n", prev->start, prev->end, em->start, em->end);
225 ret = -EEXIST;
226 goto out;
227 }
228 atomic_inc(&em->refs);
229 if (em->start != 0) {
230 rb = rb_prev(&em->rb_node);
231 if (rb)
232 prev = rb_entry(rb, struct extent_map, rb_node);
233 if (prev && prev->end + 1 == em->start &&
234 ((em->block_start == 0 && prev->block_start == 0) ||
235 (em->block_start == prev->block_end + 1))) {
236 em->start = prev->start;
237 em->block_start = prev->block_start;
238 rb_erase(&prev->rb_node, &tree->map);
239 prev->in_tree = 0;
240 free_extent_map(prev);
241 }
242 }
243out:
244 write_unlock_irq(&tree->lock);
245 return ret;
246}
247EXPORT_SYMBOL(add_extent_mapping);
248
249/*
250 * lookup_extent_mapping returns the first extent_map struct in the
251 * tree that intersects the [start, end] (inclusive) range. There may
252 * be additional objects in the tree that intersect, so check the object
253 * returned carefully to make sure you don't need additional lookups.
254 */
255struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
256 u64 start, u64 end)
257{
258 struct extent_map *em;
259 struct rb_node *rb_node;
260
261 read_lock_irq(&tree->lock);
262 rb_node = tree_search(&tree->map, start);
263 if (!rb_node) {
264 em = NULL;
265 goto out;
266 }
267 if (IS_ERR(rb_node)) {
268 em = ERR_PTR(PTR_ERR(rb_node));
269 goto out;
270 }
271 em = rb_entry(rb_node, struct extent_map, rb_node);
272 if (em->end < start || em->start > end) {
273 em = NULL;
274 goto out;
275 }
276 atomic_inc(&em->refs);
277out:
278 read_unlock_irq(&tree->lock);
279 return em;
280}
281EXPORT_SYMBOL(lookup_extent_mapping);
282
283/*
284 * removes an extent_map struct from the tree. No reference counts are
285 * dropped, and no checks are done to see if the range is in use
286 */
287int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
288{
289 int ret;
290
291 write_lock_irq(&tree->lock);
292 ret = tree_delete(&tree->map, em->end);
293 write_unlock_irq(&tree->lock);
294 return ret;
295}
296EXPORT_SYMBOL(remove_extent_mapping);
297
298/*
299 * utility function to look for merge candidates inside a given range.
300 * Any extents with matching state are merged together into a single
301 * extent in the tree. Extents with EXTENT_IO in their state field
302 * are not merged because the end_io handlers need to be able to do
303 * operations on them without sleeping (or doing allocations/splits).
304 *
305 * This should be called with the tree lock held.
306 */
307static int merge_state(struct extent_map_tree *tree,
308 struct extent_state *state)
309{
310 struct extent_state *other;
311 struct rb_node *other_node;
312
313 if (state->state & EXTENT_IOBITS)
314 return 0;
315
316 other_node = rb_prev(&state->rb_node);
317 if (other_node) {
318 other = rb_entry(other_node, struct extent_state, rb_node);
319 if (other->end == state->start - 1 &&
320 other->state == state->state) {
321 state->start = other->start;
322 other->in_tree = 0;
323 rb_erase(&other->rb_node, &tree->state);
324 free_extent_state(other);
325 }
326 }
327 other_node = rb_next(&state->rb_node);
328 if (other_node) {
329 other = rb_entry(other_node, struct extent_state, rb_node);
330 if (other->start == state->end + 1 &&
331 other->state == state->state) {
332 other->start = state->start;
333 state->in_tree = 0;
334 rb_erase(&state->rb_node, &tree->state);
335 free_extent_state(state);
336 }
337 }
338 return 0;
339}
340
341/*
342 * insert an extent_state struct into the tree. 'bits' are set on the
343 * struct before it is inserted.
344 *
345 * This may return -EEXIST if the extent is already there, in which case the
346 * state struct is freed.
347 *
348 * The tree lock is not taken internally. This is a utility function and
349 * probably isn't what you want to call (see set/clear_extent_bit).
350 */
351static int insert_state(struct extent_map_tree *tree,
352 struct extent_state *state, u64 start, u64 end,
353 int bits)
354{
355 struct rb_node *node;
356
357 if (end < start) {
358 printk("end < start %Lu %Lu\n", end, start);
359 WARN_ON(1);
360 }
361 state->state |= bits;
362 state->start = start;
363 state->end = end;
364 if ((end & 4095) == 0) {
365 printk("insert state %Lu %Lu strange end\n", start, end);
366 WARN_ON(1);
367 }
368 node = tree_insert(&tree->state, end, &state->rb_node);
369 if (node) {
370 struct extent_state *found;
371 found = rb_entry(node, struct extent_state, rb_node);
372printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, start, end);
373 free_extent_state(state);
374 return -EEXIST;
375 }
376 merge_state(tree, state);
377 return 0;
378}
379
380/*
381 * split a given extent state struct in two, inserting the preallocated
382 * struct 'prealloc' as the newly created second half. 'split' indicates an
383 * offset inside 'orig' where it should be split.
384 *
385 * Before calling,
386 * the tree has 'orig' at [orig->start, orig->end]. After calling, there
387 * are two extent state structs in the tree:
388 * prealloc: [orig->start, split - 1]
389 * orig: [ split, orig->end ]
390 *
391 * The tree locks are not taken by this function. They need to be held
392 * by the caller.
393 */
394static int split_state(struct extent_map_tree *tree, struct extent_state *orig,
395 struct extent_state *prealloc, u64 split)
396{
397 struct rb_node *node;
398 prealloc->start = orig->start;
399 prealloc->end = split - 1;
400 prealloc->state = orig->state;
401 orig->start = split;
402 if ((prealloc->end & 4095) == 0) {
403 printk("insert state %Lu %Lu strange end\n", prealloc->start,
404 prealloc->end);
405 WARN_ON(1);
406 }
407 node = tree_insert(&tree->state, prealloc->end, &prealloc->rb_node);
408 if (node) {
409 struct extent_state *found;
410 found = rb_entry(node, struct extent_state, rb_node);
411printk("found node %Lu %Lu on insert of %Lu %Lu\n", found->start, found->end, prealloc->start, prealloc->end);
412 free_extent_state(prealloc);
413 return -EEXIST;
414 }
415 return 0;
416}
417
418/*
419 * utility function to clear some bits in an extent state struct.
420 * it will optionally wake up any one waiting on this state (wake == 1), or
421 * forcibly remove the state from the tree (delete == 1).
422 *
423 * If no bits are set on the state struct after clearing things, the
424 * struct is freed and removed from the tree
425 */
426static int clear_state_bit(struct extent_map_tree *tree,
427 struct extent_state *state, int bits, int wake,
428 int delete)
429{
430 int ret = state->state & bits;
431 state->state &= ~bits;
432 if (wake)
433 wake_up(&state->wq);
434 if (delete || state->state == 0) {
435 if (state->in_tree) {
436 rb_erase(&state->rb_node, &tree->state);
437 state->in_tree = 0;
438 free_extent_state(state);
439 } else {
440 WARN_ON(1);
441 }
442 } else {
443 merge_state(tree, state);
444 }
445 return ret;
446}
447
448/*
449 * clear some bits on a range in the tree. This may require splitting
450 * or inserting elements in the tree, so the gfp mask is used to
451 * indicate which allocations or sleeping are allowed.
452 *
453 * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove
454 * the given range from the tree regardless of state (ie for truncate).
455 *
456 * the range [start, end] is inclusive.
457 *
458 * This takes the tree lock, and returns < 0 on error, > 0 if any of the
459 * bits were already set, or zero if none of the bits were already set.
460 */
461int clear_extent_bit(struct extent_map_tree *tree, u64 start, u64 end,
462 int bits, int wake, int delete, gfp_t mask)
463{
464 struct extent_state *state;
465 struct extent_state *prealloc = NULL;
466 struct rb_node *node;
467 int err;
468 int set = 0;
469
470again:
471 if (!prealloc && (mask & __GFP_WAIT)) {
472 prealloc = alloc_extent_state(mask);
473 if (!prealloc)
474 return -ENOMEM;
475 }
476
477 write_lock_irq(&tree->lock);
478 /*
479 * this search will find the extents that end after
480 * our range starts
481 */
482 node = tree_search(&tree->state, start);
483 if (!node)
484 goto out;
485 state = rb_entry(node, struct extent_state, rb_node);
486 if (state->start > end)
487 goto out;
488 WARN_ON(state->end < start);
489
490 /*
491 * | ---- desired range ---- |
492 * | state | or
493 * | ------------- state -------------- |
494 *
495 * We need to split the extent we found, and may flip
496 * bits on second half.
497 *
498 * If the extent we found extends past our range, we
499 * just split and search again. It'll get split again
500 * the next time though.
501 *
502 * If the extent we found is inside our range, we clear
503 * the desired bit on it.
504 */
505
506 if (state->start < start) {
507 err = split_state(tree, state, prealloc, start);
508 BUG_ON(err == -EEXIST);
509 prealloc = NULL;
510 if (err)
511 goto out;
512 if (state->end <= end) {
513 start = state->end + 1;
514 set |= clear_state_bit(tree, state, bits,
515 wake, delete);
516 } else {
517 start = state->start;
518 }
519 goto search_again;
520 }
521 /*
522 * | ---- desired range ---- |
523 * | state |
524 * We need to split the extent, and clear the bit
525 * on the first half
526 */
527 if (state->start <= end && state->end > end) {
528 err = split_state(tree, state, prealloc, end + 1);
529 BUG_ON(err == -EEXIST);
530
531 if (wake)
532 wake_up(&state->wq);
533 set |= clear_state_bit(tree, prealloc, bits,
534 wake, delete);
535 prealloc = NULL;
536 goto out;
537 }
538
539 start = state->end + 1;
540 set |= clear_state_bit(tree, state, bits, wake, delete);
541 goto search_again;
542
543out:
544 write_unlock_irq(&tree->lock);
545 if (prealloc)
546 free_extent_state(prealloc);
547
548 return set;
549
550search_again:
551 if (start >= end)
552 goto out;
553 write_unlock_irq(&tree->lock);
554 if (mask & __GFP_WAIT)
555 cond_resched();
556 goto again;
557}
558EXPORT_SYMBOL(clear_extent_bit);
559
560static int wait_on_state(struct extent_map_tree *tree,
561 struct extent_state *state)
562{
563 DEFINE_WAIT(wait);
564 prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE);
565 read_unlock_irq(&tree->lock);
566 schedule();
567 read_lock_irq(&tree->lock);
568 finish_wait(&state->wq, &wait);
569 return 0;
570}
571
572/*
573 * waits for one or more bits to clear on a range in the state tree.
574 * The range [start, end] is inclusive.
575 * The tree lock is taken by this function
576 */
577int wait_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits)
578{
579 struct extent_state *state;
580 struct rb_node *node;
581
582 read_lock_irq(&tree->lock);
583again:
584 while (1) {
585 /*
586 * this search will find all the extents that end after
587 * our range starts
588 */
589 node = tree_search(&tree->state, start);
590 if (!node)
591 break;
592
593 state = rb_entry(node, struct extent_state, rb_node);
594
595 if (state->start > end)
596 goto out;
597
598 if (state->state & bits) {
599 start = state->start;
600 atomic_inc(&state->refs);
601 wait_on_state(tree, state);
602 free_extent_state(state);
603 goto again;
604 }
605 start = state->end + 1;
606
607 if (start > end)
608 break;
609
610 if (need_resched()) {
611 read_unlock_irq(&tree->lock);
612 cond_resched();
613 read_lock_irq(&tree->lock);
614 }
615 }
616out:
617 read_unlock_irq(&tree->lock);
618 return 0;
619}
620EXPORT_SYMBOL(wait_extent_bit);
621
622/*
623 * set some bits on a range in the tree. This may require allocations
624 * or sleeping, so the gfp mask is used to indicate what is allowed.
625 *
626 * If 'exclusive' == 1, this will fail with -EEXIST if some part of the
627 * range already has the desired bits set. The start of the existing
628 * range is returned in failed_start in this case.
629 *
630 * [start, end] is inclusive
631 * This takes the tree lock.
632 */
633int set_extent_bit(struct extent_map_tree *tree, u64 start, u64 end, int bits,
634 int exclusive, u64 *failed_start, gfp_t mask)
635{
636 struct extent_state *state;
637 struct extent_state *prealloc = NULL;
638 struct rb_node *node;
639 int err = 0;
640 int set;
641 u64 last_start;
642 u64 last_end;
643again:
644 if (!prealloc && (mask & __GFP_WAIT)) {
645 prealloc = alloc_extent_state(mask);
646 if (!prealloc)
647 return -ENOMEM;
648 }
649
650 write_lock_irq(&tree->lock);
651 /*
652 * this search will find all the extents that end after
653 * our range starts.
654 */
655 node = tree_search(&tree->state, start);
656 if (!node) {
657 err = insert_state(tree, prealloc, start, end, bits);
658 prealloc = NULL;
659 BUG_ON(err == -EEXIST);
660 goto out;
661 }
662
663 state = rb_entry(node, struct extent_state, rb_node);
664 last_start = state->start;
665 last_end = state->end;
666
667 /*
668 * | ---- desired range ---- |
669 * | state |
670 *
671 * Just lock what we found and keep going
672 */
673 if (state->start == start && state->end <= end) {
674 set = state->state & bits;
675 if (set && exclusive) {
676 *failed_start = state->start;
677 err = -EEXIST;
678 goto out;
679 }
680 state->state |= bits;
681 start = state->end + 1;
682 merge_state(tree, state);
683 goto search_again;
684 }
685
686 /*
687 * | ---- desired range ---- |
688 * | state |
689 * or
690 * | ------------- state -------------- |
691 *
692 * We need to split the extent we found, and may flip bits on
693 * second half.
694 *
695 * If the extent we found extends past our
696 * range, we just split and search again. It'll get split
697 * again the next time though.
698 *
699 * If the extent we found is inside our range, we set the
700 * desired bit on it.
701 */
702 if (state->start < start) {
703 set = state->state & bits;
704 if (exclusive && set) {
705 *failed_start = start;
706 err = -EEXIST;
707 goto out;
708 }
709 err = split_state(tree, state, prealloc, start);
710 BUG_ON(err == -EEXIST);
711 prealloc = NULL;
712 if (err)
713 goto out;
714 if (state->end <= end) {
715 state->state |= bits;
716 start = state->end + 1;
717 merge_state(tree, state);
718 } else {
719 start = state->start;
720 }
721 goto search_again;
722 }
723 /*
724 * | ---- desired range ---- |
725 * | state |
726 * We need to split the extent, and set the bit
727 * on the first half
728 */
729 if (state->start <= end && state->end > end) {
730 set = state->state & bits;
731 if (exclusive && set) {
732 *failed_start = start;
733 err = -EEXIST;
734 goto out;
735 }
736 err = split_state(tree, state, prealloc, end + 1);
737 BUG_ON(err == -EEXIST);
738
739 prealloc->state |= bits;
740 merge_state(tree, prealloc);
741 prealloc = NULL;
742 goto out;
743 }
744
745 /*
746 * | ---- desired range ---- |
747 * | state | or | state |
748 *
749 * There's a hole, we need to insert something in it and
750 * ignore the extent we found.
751 */
752 if (state->start > start) {
753 u64 this_end;
754 if (end < last_start)
755 this_end = end;
756 else
757 this_end = last_start -1;
758 err = insert_state(tree, prealloc, start, this_end,
759 bits);
760 prealloc = NULL;
761 BUG_ON(err == -EEXIST);
762 if (err)
763 goto out;
764 start = this_end + 1;
765 goto search_again;
766 }
767 goto search_again;
768
769out:
770 write_unlock_irq(&tree->lock);
771 if (prealloc)
772 free_extent_state(prealloc);
773
774 return err;
775
776search_again:
777 if (start > end)
778 goto out;
779 write_unlock_irq(&tree->lock);
780 if (mask & __GFP_WAIT)
781 cond_resched();
782 goto again;
783}
784EXPORT_SYMBOL(set_extent_bit);
785
786/* wrappers around set/clear extent bit */
787int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
788 gfp_t mask)
789{
790 return set_extent_bit(tree, start, end, EXTENT_DIRTY, 0, NULL,
791 mask);
792}
793EXPORT_SYMBOL(set_extent_dirty);
794
795int clear_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
796 gfp_t mask)
797{
798 return clear_extent_bit(tree, start, end, EXTENT_DIRTY, 0, 0, mask);
799}
800EXPORT_SYMBOL(clear_extent_dirty);
801
802int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
803 gfp_t mask)
804{
805 return set_extent_bit(tree, start, end, EXTENT_NEW, 0, NULL,
806 mask);
807}
808EXPORT_SYMBOL(set_extent_new);
809
810int clear_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
811 gfp_t mask)
812{
813 return clear_extent_bit(tree, start, end, EXTENT_NEW, 0, 0, mask);
814}
815EXPORT_SYMBOL(clear_extent_new);
816
817int set_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
818 gfp_t mask)
819{
820 return set_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, NULL,
821 mask);
822}
823EXPORT_SYMBOL(set_extent_uptodate);
824
825int clear_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
826 gfp_t mask)
827{
828 return clear_extent_bit(tree, start, end, EXTENT_UPTODATE, 0, 0, mask);
829}
830EXPORT_SYMBOL(clear_extent_uptodate);
831
832int set_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
833 gfp_t mask)
834{
835 return set_extent_bit(tree, start, end, EXTENT_WRITEBACK,
836 0, NULL, mask);
837}
838EXPORT_SYMBOL(set_extent_writeback);
839
840int clear_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end,
841 gfp_t mask)
842{
843 return clear_extent_bit(tree, start, end, EXTENT_WRITEBACK, 1, 0, mask);
844}
845EXPORT_SYMBOL(clear_extent_writeback);
846
847int wait_on_extent_writeback(struct extent_map_tree *tree, u64 start, u64 end)
848{
849 return wait_extent_bit(tree, start, end, EXTENT_WRITEBACK);
850}
851EXPORT_SYMBOL(wait_on_extent_writeback);
852
853/*
854 * locks a range in ascending order, waiting for any locked regions
855 * it hits on the way. [start,end] are inclusive, and this will sleep.
856 */
857int lock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask)
858{
859 int err;
860 u64 failed_start;
861 while (1) {
862 err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
863 &failed_start, mask);
864 if (err == -EEXIST && (mask & __GFP_WAIT)) {
865 wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED);
866 start = failed_start;
867 } else {
868 break;
869 }
870 WARN_ON(start > end);
871 }
872 return err;
873}
874EXPORT_SYMBOL(lock_extent);
875
876int unlock_extent(struct extent_map_tree *tree, u64 start, u64 end,
877 gfp_t mask)
878{
879 return clear_extent_bit(tree, start, end, EXTENT_LOCKED, 1, 0, mask);
880}
881EXPORT_SYMBOL(unlock_extent);
882
883/*
884 * helper function to set pages and extents in the tree dirty
885 */
886int set_range_dirty(struct extent_map_tree *tree, u64 start, u64 end)
887{
888 unsigned long index = start >> PAGE_CACHE_SHIFT;
889 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
890 struct page *page;
891
892 while (index <= end_index) {
893 page = find_get_page(tree->mapping, index);
894 BUG_ON(!page);
895 __set_page_dirty_nobuffers(page);
896 page_cache_release(page);
897 index++;
898 }
899 set_extent_dirty(tree, start, end, GFP_NOFS);
900 return 0;
901}
902EXPORT_SYMBOL(set_range_dirty);
903
904/*
905 * helper function to set both pages and extents in the tree writeback
906 */
907int set_range_writeback(struct extent_map_tree *tree, u64 start, u64 end)
908{
909 unsigned long index = start >> PAGE_CACHE_SHIFT;
910 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
911 struct page *page;
912
913 while (index <= end_index) {
914 page = find_get_page(tree->mapping, index);
915 BUG_ON(!page);
916 set_page_writeback(page);
917 page_cache_release(page);
918 index++;
919 }
920 set_extent_writeback(tree, start, end, GFP_NOFS);
921 return 0;
922}
923EXPORT_SYMBOL(set_range_writeback);
924
925/*
926 * helper function to lock both pages and extents in the tree.
927 * pages must be locked first.
928 */
929int lock_range(struct extent_map_tree *tree, u64 start, u64 end)
930{
931 unsigned long index = start >> PAGE_CACHE_SHIFT;
932 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
933 struct page *page;
934 int err;
935
936 while (index <= end_index) {
937 page = grab_cache_page(tree->mapping, index);
938 if (!page) {
939 err = -ENOMEM;
940 goto failed;
941 }
942 if (IS_ERR(page)) {
943 err = PTR_ERR(page);
944 goto failed;
945 }
946 index++;
947 }
948 lock_extent(tree, start, end, GFP_NOFS);
949 return 0;
950
951failed:
952 /*
953 * we failed above in getting the page at 'index', so we undo here
954 * up to but not including the page at 'index'
955 */
956 end_index = index;
957 index = start >> PAGE_CACHE_SHIFT;
958 while (index < end_index) {
959 page = find_get_page(tree->mapping, index);
960 unlock_page(page);
961 page_cache_release(page);
962 index++;
963 }
964 return err;
965}
966EXPORT_SYMBOL(lock_range);
967
968/*
969 * helper function to unlock both pages and extents in the tree.
970 */
971int unlock_range(struct extent_map_tree *tree, u64 start, u64 end)
972{
973 unsigned long index = start >> PAGE_CACHE_SHIFT;
974 unsigned long end_index = end >> PAGE_CACHE_SHIFT;
975 struct page *page;
976
977 while (index <= end_index) {
978 page = find_get_page(tree->mapping, index);
979 unlock_page(page);
980 page_cache_release(page);
981 index++;
982 }
983 unlock_extent(tree, start, end, GFP_NOFS);
984 return 0;
985}
986EXPORT_SYMBOL(unlock_range);
987
988/*
989 * searches a range in the state tree for a given mask.
990 * If 'filled' == 1, this returns 1 only if ever extent in the tree
991 * has the bits set. Otherwise, 1 is returned if any bit in the
992 * range is found set.
993 */
994static int test_range_bit(struct extent_map_tree *tree, u64 start, u64 end,
995 int bits, int filled)
996{
997 struct extent_state *state = NULL;
998 struct rb_node *node;
999 int bitset = 0;
1000
1001 read_lock_irq(&tree->lock);
1002 node = tree_search(&tree->state, start);
1003 while (node && start <= end) {
1004 state = rb_entry(node, struct extent_state, rb_node);
1005 if (state->start > end)
1006 break;
1007
1008 if (filled && state->start > start) {
1009 bitset = 0;
1010 break;
1011 }
1012 if (state->state & bits) {
1013 bitset = 1;
1014 if (!filled)
1015 break;
1016 } else if (filled) {
1017 bitset = 0;
1018 break;
1019 }
1020 start = state->end + 1;
1021 if (start > end)
1022 break;
1023 node = rb_next(node);
1024 }
1025 read_unlock_irq(&tree->lock);
1026 return bitset;
1027}
1028
1029/*
1030 * helper function to set a given page up to date if all the
1031 * extents in the tree for that page are up to date
1032 */
1033static int check_page_uptodate(struct extent_map_tree *tree,
1034 struct page *page)
1035{
1036 u64 start = page->index << PAGE_CACHE_SHIFT;
1037 u64 end = start + PAGE_CACHE_SIZE - 1;
1038 if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1))
1039 SetPageUptodate(page);
1040 return 0;
1041}
1042
1043/*
1044 * helper function to unlock a page if all the extents in the tree
1045 * for that page are unlocked
1046 */
1047static int check_page_locked(struct extent_map_tree *tree,
1048 struct page *page)
1049{
1050 u64 start = page->index << PAGE_CACHE_SHIFT;
1051 u64 end = start + PAGE_CACHE_SIZE - 1;
1052 if (!test_range_bit(tree, start, end, EXTENT_LOCKED, 0))
1053 unlock_page(page);
1054 return 0;
1055}
1056
1057/*
1058 * helper function to end page writeback if all the extents
1059 * in the tree for that page are done with writeback
1060 */
1061static int check_page_writeback(struct extent_map_tree *tree,
1062 struct page *page)
1063{
1064 u64 start = page->index << PAGE_CACHE_SHIFT;
1065 u64 end = start + PAGE_CACHE_SIZE - 1;
1066 if (!test_range_bit(tree, start, end, EXTENT_WRITEBACK, 0))
1067 end_page_writeback(page);
1068 return 0;
1069}
1070
1071/* lots and lots of room for performance fixes in the end_bio funcs */
1072
1073/*
1074 * after a writepage IO is done, we need to:
1075 * clear the uptodate bits on error
1076 * clear the writeback bits in the extent tree for this IO
1077 * end_page_writeback if the page has no more pending IO
1078 *
1079 * Scheduling is not allowed, so the extent state tree is expected
1080 * to have one and only one object corresponding to this IO.
1081 */
1082static int end_bio_extent_writepage(struct bio *bio,
1083 unsigned int bytes_done, int err)
1084{
1085 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1086 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1087 struct extent_map_tree *tree = bio->bi_private;
1088 u64 start;
1089 u64 end;
1090 int whole_page;
1091
1092 if (bio->bi_size)
1093 return 1;
1094
1095 do {
1096 struct page *page = bvec->bv_page;
1097 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1098 end = start + bvec->bv_len - 1;
1099
1100 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1101 whole_page = 1;
1102 else
1103 whole_page = 0;
1104
1105 if (--bvec >= bio->bi_io_vec)
1106 prefetchw(&bvec->bv_page->flags);
1107
1108 if (!uptodate) {
1109 clear_extent_uptodate(tree, start, end, GFP_ATOMIC);
1110 ClearPageUptodate(page);
1111 SetPageError(page);
1112 }
1113 clear_extent_writeback(tree, start, end, GFP_ATOMIC);
1114
1115 if (whole_page)
1116 end_page_writeback(page);
1117 else
1118 check_page_writeback(tree, page);
1119 } while (bvec >= bio->bi_io_vec);
1120
1121 bio_put(bio);
1122 return 0;
1123}
1124
1125/*
1126 * after a readpage IO is done, we need to:
1127 * clear the uptodate bits on error
1128 * set the uptodate bits if things worked
1129 * set the page up to date if all extents in the tree are uptodate
1130 * clear the lock bit in the extent tree
1131 * unlock the page if there are no other extents locked for it
1132 *
1133 * Scheduling is not allowed, so the extent state tree is expected
1134 * to have one and only one object corresponding to this IO.
1135 */
1136static int end_bio_extent_readpage(struct bio *bio,
1137 unsigned int bytes_done, int err)
1138{
1139 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1140 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1141 struct extent_map_tree *tree = bio->bi_private;
1142 u64 start;
1143 u64 end;
1144 int whole_page;
1145
1146 if (bio->bi_size)
1147 return 1;
1148
1149 do {
1150 struct page *page = bvec->bv_page;
1151 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1152 end = start + bvec->bv_len - 1;
1153
1154 if (bvec->bv_offset == 0 && bvec->bv_len == PAGE_CACHE_SIZE)
1155 whole_page = 1;
1156 else
1157 whole_page = 0;
1158
1159 if (--bvec >= bio->bi_io_vec)
1160 prefetchw(&bvec->bv_page->flags);
1161
1162 if (uptodate) {
1163 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1164 if (whole_page)
1165 SetPageUptodate(page);
1166 else
1167 check_page_uptodate(tree, page);
1168 } else {
1169 ClearPageUptodate(page);
1170 SetPageError(page);
1171 }
1172
1173 unlock_extent(tree, start, end, GFP_ATOMIC);
1174
1175 if (whole_page)
1176 unlock_page(page);
1177 else
1178 check_page_locked(tree, page);
1179 } while (bvec >= bio->bi_io_vec);
1180
1181 bio_put(bio);
1182 return 0;
1183}
1184
1185/*
1186 * IO done from prepare_write is pretty simple, we just unlock
1187 * the structs in the extent tree when done, and set the uptodate bits
1188 * as appropriate.
1189 */
1190static int end_bio_extent_preparewrite(struct bio *bio,
1191 unsigned int bytes_done, int err)
1192{
1193 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1194 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
1195 struct extent_map_tree *tree = bio->bi_private;
1196 u64 start;
1197 u64 end;
1198
1199 if (bio->bi_size)
1200 return 1;
1201
1202 do {
1203 struct page *page = bvec->bv_page;
1204 start = (page->index << PAGE_CACHE_SHIFT) + bvec->bv_offset;
1205 end = start + bvec->bv_len - 1;
1206
1207 if (--bvec >= bio->bi_io_vec)
1208 prefetchw(&bvec->bv_page->flags);
1209
1210 if (uptodate) {
1211 set_extent_uptodate(tree, start, end, GFP_ATOMIC);
1212 } else {
1213 ClearPageUptodate(page);
1214 SetPageError(page);
1215 }
1216
1217 unlock_extent(tree, start, end, GFP_ATOMIC);
1218
1219 } while (bvec >= bio->bi_io_vec);
1220
1221 bio_put(bio);
1222 return 0;
1223}
1224
1225static int submit_extent_page(int rw, struct extent_map_tree *tree,
1226 struct page *page, sector_t sector,
1227 size_t size, unsigned long offset,
1228 struct block_device *bdev,
1229 bio_end_io_t end_io_func)
1230{
1231 struct bio *bio;
1232 int ret = 0;
1233
1234 bio = bio_alloc(GFP_NOIO, 1);
1235
1236 bio->bi_sector = sector;
1237 bio->bi_bdev = bdev;
1238 bio->bi_io_vec[0].bv_page = page;
1239 bio->bi_io_vec[0].bv_len = size;
1240 bio->bi_io_vec[0].bv_offset = offset;
1241
1242 bio->bi_vcnt = 1;
1243 bio->bi_idx = 0;
1244 bio->bi_size = size;
1245
1246 bio->bi_end_io = end_io_func;
1247 bio->bi_private = tree;
1248
1249 bio_get(bio);
1250 submit_bio(rw, bio);
1251
1252 if (bio_flagged(bio, BIO_EOPNOTSUPP))
1253 ret = -EOPNOTSUPP;
1254
1255 bio_put(bio);
1256 return ret;
1257}
1258
1259/*
1260 * basic readpage implementation. Locked extent state structs are inserted
1261 * into the tree that are removed when the IO is done (by the end_io
1262 * handlers)
1263 */
1264int extent_read_full_page(struct extent_map_tree *tree, struct page *page,
1265 get_extent_t *get_extent)
1266{
1267 struct inode *inode = page->mapping->host;
1268 u64 start = page->index << PAGE_CACHE_SHIFT;
1269 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1270 u64 end;
1271 u64 cur = start;
1272 u64 extent_offset;
1273 u64 last_byte = i_size_read(inode);
1274 u64 block_start;
1275 u64 cur_end;
1276 sector_t sector;
1277 struct extent_map *em;
1278 struct block_device *bdev;
1279 int ret;
1280 int nr = 0;
1281 size_t page_offset = 0;
1282 size_t iosize;
1283 size_t blocksize = inode->i_sb->s_blocksize;
1284
1285 if (!PagePrivate(page)) {
1286 SetPagePrivate(page);
1287 set_page_private(page, 1);
1288 page_cache_get(page);
1289 }
1290
1291 end = page_end;
1292 lock_extent(tree, start, end, GFP_NOFS);
1293
1294 while (cur <= end) {
1295 if (cur >= last_byte) {
1296 iosize = PAGE_CACHE_SIZE - page_offset;
1297 zero_user_page(page, page_offset, iosize, KM_USER0);
1298 set_extent_uptodate(tree, cur, cur + iosize - 1,
1299 GFP_NOFS);
1300 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1301 break;
1302 }
1303 em = get_extent(inode, page, page_offset, cur, end, 0);
1304 if (IS_ERR(em) || !em) {
1305 SetPageError(page);
1306 unlock_extent(tree, cur, end, GFP_NOFS);
1307 break;
1308 }
1309
1310 extent_offset = cur - em->start;
1311 BUG_ON(em->end < cur);
1312 BUG_ON(end < cur);
1313
1314 iosize = min(em->end - cur, end - cur) + 1;
1315 cur_end = min(em->end, end);
1316 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1317 sector = (em->block_start + extent_offset) >> 9;
1318 bdev = em->bdev;
1319 block_start = em->block_start;
1320 free_extent_map(em);
1321 em = NULL;
1322
1323 /* we've found a hole, just zero and go on */
1324 if (block_start == 0) {
1325 zero_user_page(page, page_offset, iosize, KM_USER0);
1326 set_extent_uptodate(tree, cur, cur + iosize - 1,
1327 GFP_NOFS);
1328 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1329 cur = cur + iosize;
1330 page_offset += iosize;
1331 continue;
1332 }
1333 /* the get_extent function already copied into the page */
1334 if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1)) {
1335 unlock_extent(tree, cur, cur + iosize - 1, GFP_NOFS);
1336 cur = cur + iosize;
1337 page_offset += iosize;
1338 continue;
1339 }
1340
1341 ret = submit_extent_page(READ, tree, page,
1342 sector, iosize, page_offset, bdev,
1343 end_bio_extent_readpage);
1344 if (ret)
1345 SetPageError(page);
1346 cur = cur + iosize;
1347 page_offset += iosize;
1348 nr++;
1349 }
1350 if (!nr) {
1351 if (!PageError(page))
1352 SetPageUptodate(page);
1353 unlock_page(page);
1354 }
1355 return 0;
1356}
1357EXPORT_SYMBOL(extent_read_full_page);
1358
1359/*
1360 * the writepage semantics are similar to regular writepage. extent
1361 * records are inserted to lock ranges in the tree, and as dirty areas
1362 * are found, they are marked writeback. Then the lock bits are removed
1363 * and the end_io handler clears the writeback ranges
1364 */
1365int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
1366 get_extent_t *get_extent,
1367 struct writeback_control *wbc)
1368{
1369 struct inode *inode = page->mapping->host;
1370 u64 start = page->index << PAGE_CACHE_SHIFT;
1371 u64 page_end = start + PAGE_CACHE_SIZE - 1;
1372 u64 end;
1373 u64 cur = start;
1374 u64 extent_offset;
1375 u64 last_byte = i_size_read(inode);
1376 u64 block_start;
1377 sector_t sector;
1378 struct extent_map *em;
1379 struct block_device *bdev;
1380 int ret;
1381 int nr = 0;
1382 size_t page_offset = 0;
1383 size_t iosize;
1384 size_t blocksize;
1385 loff_t i_size = i_size_read(inode);
1386 unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
1387
1388 if (page->index > end_index) {
1389 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1390 unlock_page(page);
1391 return 0;
1392 }
1393
1394 if (page->index == end_index) {
1395 size_t offset = i_size & (PAGE_CACHE_SIZE - 1);
1396 zero_user_page(page, offset,
1397 PAGE_CACHE_SIZE - offset, KM_USER0);
1398 }
1399
1400 if (!PagePrivate(page)) {
1401 SetPagePrivate(page);
1402 set_page_private(page, 1);
1403 page_cache_get(page);
1404 }
1405
1406 end = page_end;
1407 lock_extent(tree, start, page_end, GFP_NOFS);
1408
1409 if (last_byte <= start) {
1410 clear_extent_dirty(tree, start, page_end, GFP_NOFS);
1411 goto done;
1412 }
1413
1414 set_extent_uptodate(tree, start, page_end, GFP_NOFS);
1415 blocksize = inode->i_sb->s_blocksize;
1416
1417 while (cur <= end) {
1418 if (cur >= last_byte) {
1419 clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
1420 break;
1421 }
1422 em = get_extent(inode, page, page_offset, cur, end, 1);
1423 if (IS_ERR(em) || !em) {
1424 SetPageError(page);
1425 break;
1426 }
1427
1428 extent_offset = cur - em->start;
1429 BUG_ON(em->end < cur);
1430 BUG_ON(end < cur);
1431 iosize = min(em->end - cur, end - cur) + 1;
1432 iosize = (iosize + blocksize - 1) & ~((u64)blocksize - 1);
1433 sector = (em->block_start + extent_offset) >> 9;
1434 bdev = em->bdev;
1435 block_start = em->block_start;
1436 free_extent_map(em);
1437 em = NULL;
1438
1439 if (block_start == 0 || block_start == EXTENT_MAP_INLINE) {
1440 clear_extent_dirty(tree, cur,
1441 cur + iosize - 1, GFP_NOFS);
1442 cur = cur + iosize;
1443 page_offset += iosize;
1444 continue;
1445 }
1446
1447 /* leave this out until we have a page_mkwrite call */
1448 if (0 && !test_range_bit(tree, cur, cur + iosize - 1,
1449 EXTENT_DIRTY, 0)) {
1450 cur = cur + iosize;
1451 page_offset += iosize;
1452 continue;
1453 }
1454 clear_extent_dirty(tree, cur, cur + iosize - 1, GFP_NOFS);
1455 set_range_writeback(tree, cur, cur + iosize - 1);
1456 ret = submit_extent_page(WRITE, tree, page,
1457 sector, iosize, page_offset, bdev,
1458 end_bio_extent_writepage);
1459 if (ret)
1460 SetPageError(page);
1461 cur = cur + iosize;
1462 page_offset += iosize;
1463 nr++;
1464 }
1465done:
1466 WARN_ON(test_range_bit(tree, start, page_end, EXTENT_DIRTY, 0));
1467 unlock_extent(tree, start, page_end, GFP_NOFS);
1468 unlock_page(page);
1469 return 0;
1470}
1471EXPORT_SYMBOL(extent_write_full_page);
1472
1473/*
1474 * basic invalidatepage code, this waits on any locked or writeback
1475 * ranges corresponding to the page, and then deletes any extent state
1476 * records from the tree
1477 */
1478int extent_invalidatepage(struct extent_map_tree *tree,
1479 struct page *page, unsigned long offset)
1480{
1481 u64 start = (page->index << PAGE_CACHE_SHIFT);
1482 u64 end = start + PAGE_CACHE_SIZE - 1;
1483 size_t blocksize = page->mapping->host->i_sb->s_blocksize;
1484
1485 start += (offset + blocksize -1) & ~(blocksize - 1);
1486 if (start > end)
1487 return 0;
1488
1489 lock_extent(tree, start, end, GFP_NOFS);
1490 wait_on_extent_writeback(tree, start, end);
1491 clear_extent_bit(tree, start, end, EXTENT_LOCKED | EXTENT_DIRTY,
1492 1, 1, GFP_NOFS);
1493 return 0;
1494}
1495EXPORT_SYMBOL(extent_invalidatepage);
1496
1497/*
1498 * simple commit_write call, set_range_dirty is used to mark both
1499 * the pages and the extent records as dirty
1500 */
1501int extent_commit_write(struct extent_map_tree *tree,
1502 struct inode *inode, struct page *page,
1503 unsigned from, unsigned to)
1504{
1505 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1506
1507 if (!PagePrivate(page)) {
1508 SetPagePrivate(page);
1509 set_page_private(page, 1);
1510 page_cache_get(page);
1511 }
1512
1513 set_page_dirty(page);
1514
1515 if (pos > inode->i_size) {
1516 i_size_write(inode, pos);
1517 mark_inode_dirty(inode);
1518 }
1519 return 0;
1520}
1521EXPORT_SYMBOL(extent_commit_write);
1522
1523int extent_prepare_write(struct extent_map_tree *tree,
1524 struct inode *inode, struct page *page,
1525 unsigned from, unsigned to, get_extent_t *get_extent)
1526{
1527 u64 page_start = page->index << PAGE_CACHE_SHIFT;
1528 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
1529 u64 block_start;
1530 u64 orig_block_start;
1531 u64 block_end;
1532 u64 cur_end;
1533 struct extent_map *em;
1534 unsigned blocksize = 1 << inode->i_blkbits;
1535 size_t page_offset = 0;
1536 size_t block_off_start;
1537 size_t block_off_end;
1538 int err = 0;
1539 int iocount = 0;
1540 int ret = 0;
1541 int isnew;
1542
1543 if (!PagePrivate(page)) {
1544 SetPagePrivate(page);
1545 set_page_private(page, 1);
1546 page_cache_get(page);
1547 }
1548 block_start = (page_start + from) & ~((u64)blocksize - 1);
1549 block_end = (page_start + to - 1) | (blocksize - 1);
1550 orig_block_start = block_start;
1551
1552 lock_extent(tree, page_start, page_end, GFP_NOFS);
1553 while(block_start <= block_end) {
1554 em = get_extent(inode, page, page_offset, block_start,
1555 block_end, 1);
1556 if (IS_ERR(em) || !em) {
1557 goto err;
1558 }
1559 cur_end = min(block_end, em->end);
1560 block_off_start = block_start & (PAGE_CACHE_SIZE - 1);
1561 block_off_end = block_off_start + blocksize;
1562 isnew = clear_extent_new(tree, block_start, cur_end, GFP_NOFS);
1563
1564 if (!PageUptodate(page) && isnew &&
1565 (block_off_end > to || block_off_start < from)) {
1566 void *kaddr;
1567
1568 kaddr = kmap_atomic(page, KM_USER0);
1569 if (block_off_end > to)
1570 memset(kaddr + to, 0, block_off_end - to);
1571 if (block_off_start < from)
1572 memset(kaddr + block_off_start, 0,
1573 from - block_off_start);
1574 flush_dcache_page(page);
1575 kunmap_atomic(kaddr, KM_USER0);
1576 }
1577 if (!isnew && !PageUptodate(page) &&
1578 (block_off_end > to || block_off_start < from) &&
1579 !test_range_bit(tree, block_start, cur_end,
1580 EXTENT_UPTODATE, 1)) {
1581 u64 sector;
1582 u64 extent_offset = block_start - em->start;
1583 size_t iosize;
1584 sector = (em->block_start + extent_offset) >> 9;
1585 iosize = (cur_end - block_start + blocksize - 1) &
1586 ~((u64)blocksize - 1);
1587 /*
1588 * we've already got the extent locked, but we
1589 * need to split the state such that our end_bio
1590 * handler can clear the lock.
1591 */
1592 set_extent_bit(tree, block_start,
1593 block_start + iosize - 1,
1594 EXTENT_LOCKED, 0, NULL, GFP_NOFS);
1595 ret = submit_extent_page(READ, tree, page,
1596 sector, iosize, page_offset, em->bdev,
1597 end_bio_extent_preparewrite);
1598 iocount++;
1599 block_start = block_start + iosize;
1600 } else {
1601 set_extent_uptodate(tree, block_start, cur_end,
1602 GFP_NOFS);
1603 unlock_extent(tree, block_start, cur_end, GFP_NOFS);
1604 block_start = cur_end + 1;
1605 }
1606 page_offset = block_start & (PAGE_CACHE_SIZE - 1);
1607 free_extent_map(em);
1608 }
1609 if (iocount) {
1610 wait_extent_bit(tree, orig_block_start,
1611 block_end, EXTENT_LOCKED);
1612 }
1613 check_page_uptodate(tree, page);
1614err:
1615 /* FIXME, zero out newly allocated blocks on error */
1616 return err;
1617}
1618EXPORT_SYMBOL(extent_prepare_write);
1619
1620/*
1621 * a helper for releasepage. As long as there are no locked extents
1622 * in the range corresponding to the page, both state records and extent
1623 * map records are removed
1624 */
1625int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page)
1626{
1627 struct extent_map *em;
1628 u64 start = page->index << PAGE_CACHE_SHIFT;
1629 u64 end = start + PAGE_CACHE_SIZE - 1;
1630 u64 orig_start = start;
1631
1632 while (start <= end) {
1633 em = lookup_extent_mapping(tree, start, end);
1634 if (!em || IS_ERR(em))
1635 break;
1636 if (test_range_bit(tree, em->start, em->end,
1637 EXTENT_LOCKED, 0)) {
1638 free_extent_map(em);
1639 start = em->end + 1;
1640printk("range still locked %Lu %Lu\n", em->start, em->end);
1641 break;
1642 }
1643 remove_extent_mapping(tree, em);
1644 start = em->end + 1;
1645 /* once for the rb tree */
1646 free_extent_map(em);
1647 /* once for us */
1648 free_extent_map(em);
1649 }
1650 WARN_ON(test_range_bit(tree, orig_start, end, EXTENT_WRITEBACK, 0));
1651 clear_extent_bit(tree, orig_start, end, EXTENT_UPTODATE,
1652 1, 1, GFP_NOFS);
1653 return 1;
1654}
1655EXPORT_SYMBOL(try_release_extent_mapping);
1656
diff --git a/fs/btrfs/extent_map.h b/fs/btrfs/extent_map.h
new file mode 100644
index 000000000000..108944aab4bd
--- /dev/null
+++ b/fs/btrfs/extent_map.h
@@ -0,0 +1,89 @@
1#ifndef __EXTENTMAP__
2#define __EXTENTMAP__
3
4#include <linux/rbtree.h>
5
6#define EXTENT_MAP_INLINE (u64)-2
7#define EXTENT_MAP_DELALLOC (u64)-1
8
9struct extent_map_tree {
10 struct rb_root map;
11 struct rb_root state;
12 struct address_space *mapping;
13 rwlock_t lock;
14};
15
16/* note, this must start with the same fields as fs/extent_map.c:tree_entry */
17struct extent_map {
18 u64 start;
19 u64 end; /* inclusive */
20 int in_tree;
21 struct rb_node rb_node;
22 /* block_start and block_end are in bytes */
23 u64 block_start;
24 u64 block_end; /* inclusive */
25 struct block_device *bdev;
26 atomic_t refs;
27};
28
29/* note, this must start with the same fields as fs/extent_map.c:tree_entry */
30struct extent_state {
31 u64 start;
32 u64 end; /* inclusive */
33 int in_tree;
34 struct rb_node rb_node;
35 wait_queue_head_t wq;
36 atomic_t refs;
37 unsigned long state;
38 struct list_head list;
39};
40
41struct extent_buffer {
42 u64 start;
43 u64 end; /* inclusive */
44 char *addr;
45 struct page *pages[];
46};
47
48typedef struct extent_map *(get_extent_t)(struct inode *inode,
49 struct page *page,
50 size_t page_offset,
51 u64 start, u64 end,
52 int create);
53
54void extent_map_tree_init(struct extent_map_tree *tree,
55 struct address_space *mapping, gfp_t mask);
56struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
57 u64 start, u64 end);
58int add_extent_mapping(struct extent_map_tree *tree,
59 struct extent_map *em);
60int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em);
61int try_release_extent_mapping(struct extent_map_tree *tree, struct page *page);
62int lock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask);
63int unlock_extent(struct extent_map_tree *tree, u64 start, u64 end, gfp_t mask);
64struct extent_map *alloc_extent_map(gfp_t mask);
65void free_extent_map(struct extent_map *em);
66int extent_read_full_page(struct extent_map_tree *tree, struct page *page,
67 get_extent_t *get_extent);
68void __init extent_map_init(void);
69void __exit extent_map_exit(void);
70int extent_clean_all_trees(struct extent_map_tree *tree);
71int set_extent_uptodate(struct extent_map_tree *tree, u64 start, u64 end,
72 gfp_t mask);
73int set_extent_new(struct extent_map_tree *tree, u64 start, u64 end,
74 gfp_t mask);
75int set_extent_dirty(struct extent_map_tree *tree, u64 start, u64 end,
76 gfp_t mask);
77int extent_invalidatepage(struct extent_map_tree *tree,
78 struct page *page, unsigned long offset);
79int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
80 get_extent_t *get_extent,
81 struct writeback_control *wbc);
82int extent_prepare_write(struct extent_map_tree *tree,
83 struct inode *inode, struct page *page,
84 unsigned from, unsigned to, get_extent_t *get_extent);
85int extent_commit_write(struct extent_map_tree *tree,
86 struct inode *inode, struct page *page,
87 unsigned from, unsigned to);
88int set_range_dirty(struct extent_map_tree *tree, u64 start, u64 end);
89#endif
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 6933ab11a5cd..71a481894ab6 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -81,14 +81,14 @@ static void btrfs_drop_pages(struct page **pages, size_t num_pages)
81 } 81 }
82} 82}
83 83
84static int insert_inline_extent(struct btrfs_root *root, struct inode *inode, 84static int insert_inline_extent(struct btrfs_trans_handle *trans,
85 struct btrfs_root *root, struct inode *inode,
85 u64 offset, ssize_t size, 86 u64 offset, ssize_t size,
86 struct buffer_head *bh) 87 struct page *page, size_t page_offset)
87{ 88{
88 struct btrfs_key key; 89 struct btrfs_key key;
89 struct btrfs_path *path; 90 struct btrfs_path *path;
90 char *ptr, *kaddr; 91 char *ptr, *kaddr;
91 struct btrfs_trans_handle *trans;
92 struct btrfs_file_extent_item *ei; 92 struct btrfs_file_extent_item *ei;
93 u32 datasize; 93 u32 datasize;
94 int err = 0; 94 int err = 0;
@@ -98,8 +98,6 @@ static int insert_inline_extent(struct btrfs_root *root, struct inode *inode,
98 if (!path) 98 if (!path)
99 return -ENOMEM; 99 return -ENOMEM;
100 100
101 mutex_lock(&root->fs_info->fs_mutex);
102 trans = btrfs_start_transaction(root, 1);
103 btrfs_set_trans_block_group(trans, inode); 101 btrfs_set_trans_block_group(trans, inode);
104 102
105 key.objectid = inode->i_ino; 103 key.objectid = inode->i_ino;
@@ -122,18 +120,13 @@ static int insert_inline_extent(struct btrfs_root *root, struct inode *inode,
122 BTRFS_FILE_EXTENT_INLINE); 120 BTRFS_FILE_EXTENT_INLINE);
123 ptr = btrfs_file_extent_inline_start(ei); 121 ptr = btrfs_file_extent_inline_start(ei);
124 122
125 kaddr = kmap_atomic(bh->b_page, KM_USER0); 123 kaddr = kmap_atomic(page, KM_USER0);
126 btrfs_memcpy(root, path->nodes[0]->b_data, 124 btrfs_memcpy(root, path->nodes[0]->b_data,
127 ptr, kaddr + bh_offset(bh), 125 ptr, kaddr + page_offset, size);
128 size);
129 kunmap_atomic(kaddr, KM_USER0); 126 kunmap_atomic(kaddr, KM_USER0);
130 btrfs_mark_buffer_dirty(path->nodes[0]); 127 btrfs_mark_buffer_dirty(path->nodes[0]);
131fail: 128fail:
132 btrfs_free_path(path); 129 btrfs_free_path(path);
133 ret = btrfs_end_transaction(trans, root);
134 if (ret && !err)
135 err = ret;
136 mutex_unlock(&root->fs_info->fs_mutex);
137 return err; 130 return err;
138} 131}
139 132
@@ -145,45 +138,143 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
145 loff_t pos, 138 loff_t pos,
146 size_t write_bytes) 139 size_t write_bytes)
147{ 140{
148 int i;
149 int offset;
150 int err = 0; 141 int err = 0;
151 int ret; 142 int i;
152 int this_write;
153 struct inode *inode = file->f_path.dentry->d_inode; 143 struct inode *inode = file->f_path.dentry->d_inode;
154 struct buffer_head *bh; 144 struct extent_map *em;
145 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
146 struct btrfs_key ins;
147 u64 hint_block;
148 u64 num_blocks;
149 u64 start_pos;
150 u64 end_of_last_block;
151 u64 end_pos = pos + write_bytes;
152 loff_t isize = i_size_read(inode);
155 153
156 for (i = 0; i < num_pages; i++) { 154 em = alloc_extent_map(GFP_NOFS);
157 offset = pos & (PAGE_CACHE_SIZE -1); 155 if (!em)
158 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes); 156 return -ENOMEM;
159 157
160 /* FIXME, one block at a time */ 158 em->bdev = inode->i_sb->s_bdev;
161 bh = page_buffers(pages[i]);
162 159
163 if (buffer_mapped(bh) && bh->b_blocknr == 0) { 160 start_pos = pos & ~((u64)root->blocksize - 1);
164 ret = insert_inline_extent(root, inode, 161 num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
165 pages[i]->index << PAGE_CACHE_SHIFT, 162 inode->i_blkbits;
166 offset + this_write, bh);
167 if (ret) {
168 err = ret;
169 goto failed;
170 }
171 }
172 163
173 ret = btrfs_commit_write(file, pages[i], offset, 164 end_of_last_block = start_pos + (num_blocks << inode->i_blkbits) - 1;
174 offset + this_write); 165 mutex_lock(&root->fs_info->fs_mutex);
175 pos += this_write; 166 trans = btrfs_start_transaction(root, 1);
176 if (ret) { 167 if (!trans) {
177 err = ret; 168 err = -ENOMEM;
169 goto out_unlock;
170 }
171 btrfs_set_trans_block_group(trans, inode);
172 inode->i_blocks += num_blocks << 3;
173 hint_block = 0;
174
175 if ((end_of_last_block & 4095) == 0) {
176 printk("strange end of last %Lu %lu %Lu\n", start_pos, write_bytes, end_of_last_block);
177 }
178 set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
179
180 /* FIXME...EIEIO, ENOSPC and more */
181
182 /* step one, delete the existing extents in this range */
183 /* FIXME blocksize != pagesize */
184 if (start_pos < inode->i_size) {
185 err = btrfs_drop_extents(trans, root, inode,
186 start_pos, (pos + write_bytes + root->blocksize -1) &
187 ~((u64)root->blocksize - 1), &hint_block);
188 if (err)
189 goto failed;
190 }
191
192 /* insert any holes we need to create */
193 if (inode->i_size < start_pos) {
194 u64 last_pos_in_file;
195 u64 hole_size;
196 u64 mask = root->blocksize - 1;
197 last_pos_in_file = (isize + mask) & ~mask;
198 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
199 hole_size >>= inode->i_blkbits;
200 if (last_pos_in_file < start_pos) {
201 err = btrfs_insert_file_extent(trans, root,
202 inode->i_ino,
203 last_pos_in_file,
204 0, 0, hole_size);
205 }
206 if (err)
178 goto failed; 207 goto failed;
208 }
209
210 /*
211 * either allocate an extent for the new bytes or setup the key
212 * to show we are doing inline data in the extent
213 */
214 if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
215 pos + write_bytes - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
216 err = btrfs_alloc_extent(trans, root, inode->i_ino,
217 num_blocks, 0, hint_block, (u64)-1,
218 &ins, 1);
219 BUG_ON(err);
220 err = btrfs_insert_file_extent(trans, root, inode->i_ino,
221 start_pos, ins.objectid, ins.offset,
222 ins.offset);
223 BUG_ON(err);
224 em->start = start_pos;
225 em->end = end_of_last_block;
226 em->block_start = ins.objectid << inode->i_blkbits;
227 em->block_end = em->block_start +
228 (ins.offset << inode->i_blkbits) - 1;
229 set_extent_dirty(em_tree, start_pos, end_of_last_block,
230 GFP_NOFS);
231 err = add_extent_mapping(em_tree, em);
232 for (i = 0; i < num_pages; i++) {
233 struct page *p = pages[i];
234 SetPageUptodate(p);
235 __set_page_dirty_nobuffers(p);
179 } 236 }
180 WARN_ON(this_write > write_bytes); 237 } else {
181 write_bytes -= this_write; 238 struct page *p = pages[0];
239 err = insert_inline_extent(trans, root, inode, start_pos,
240 end_pos - start_pos, p, 0);
241 BUG_ON(err);
242 em->start = start_pos;
243 em->end = end_pos;
244 em->block_start = EXTENT_MAP_INLINE;
245 em->block_end = EXTENT_MAP_INLINE;
246 add_extent_mapping(em_tree, em);
247 }
248 if (end_pos > isize) {
249 i_size_write(inode, end_pos);
250 btrfs_update_inode(trans, root, inode);
182 } 251 }
183failed: 252failed:
253 err = btrfs_end_transaction(trans, root);
254out_unlock:
255 mutex_unlock(&root->fs_info->fs_mutex);
256 free_extent_map(em);
184 return err; 257 return err;
185} 258}
186 259
260int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
261{
262 struct extent_map *em;
263 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
264
265 while(1) {
266 em = lookup_extent_mapping(em_tree, start, end);
267 if (!em)
268 break;
269 remove_extent_mapping(em_tree, em);
270 /* once for us */
271 free_extent_map(em);
272 /* once for the tree*/
273 free_extent_map(em);
274 }
275 return 0;
276}
277
187/* 278/*
188 * this is very complex, but the basic idea is to drop all extents 279 * this is very complex, but the basic idea is to drop all extents
189 * in the range start - end. hint_block is filled in with a block number 280 * in the range start - end. hint_block is filled in with a block number
@@ -213,6 +304,8 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
213 int found_inline; 304 int found_inline;
214 int recow; 305 int recow;
215 306
307 btrfs_drop_extent_cache(inode, start, end - 1);
308
216 path = btrfs_alloc_path(); 309 path = btrfs_alloc_path();
217 if (!path) 310 if (!path)
218 return -ENOMEM; 311 return -ENOMEM;
@@ -434,18 +527,9 @@ static int prepare_pages(struct btrfs_root *root,
434 int i; 527 int i;
435 unsigned long index = pos >> PAGE_CACHE_SHIFT; 528 unsigned long index = pos >> PAGE_CACHE_SHIFT;
436 struct inode *inode = file->f_path.dentry->d_inode; 529 struct inode *inode = file->f_path.dentry->d_inode;
437 int offset;
438 int err = 0; 530 int err = 0;
439 int this_write;
440 struct buffer_head *bh;
441 struct buffer_head *head;
442 loff_t isize = i_size_read(inode);
443 struct btrfs_trans_handle *trans;
444 u64 hint_block;
445 u64 num_blocks; 531 u64 num_blocks;
446 u64 alloc_extent_start;
447 u64 start_pos; 532 u64 start_pos;
448 struct btrfs_key ins;
449 533
450 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1); 534 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
451 num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >> 535 num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
@@ -457,119 +541,17 @@ static int prepare_pages(struct btrfs_root *root,
457 pages[i] = grab_cache_page(inode->i_mapping, index + i); 541 pages[i] = grab_cache_page(inode->i_mapping, index + i);
458 if (!pages[i]) { 542 if (!pages[i]) {
459 err = -ENOMEM; 543 err = -ENOMEM;
460 goto failed_release; 544 BUG_ON(1);
461 } 545 }
462 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE); 546 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
463 wait_on_page_writeback(pages[i]); 547 wait_on_page_writeback(pages[i]);
464 } 548 if (!PagePrivate(pages[i])) {
465 549 SetPagePrivate(pages[i]);
466 mutex_lock(&root->fs_info->fs_mutex); 550 set_page_private(pages[i], 1);
467 trans = btrfs_start_transaction(root, 1); 551 page_cache_get(pages[i]);
468 if (!trans) {
469 err = -ENOMEM;
470 mutex_unlock(&root->fs_info->fs_mutex);
471 goto out_unlock;
472 }
473 btrfs_set_trans_block_group(trans, inode);
474 /* FIXME blocksize != 4096 */
475 inode->i_blocks += num_blocks << 3;
476 hint_block = 0;
477
478 /* FIXME...EIEIO, ENOSPC and more */
479
480 /* step one, delete the existing extents in this range */
481 /* FIXME blocksize != pagesize */
482 if (start_pos < inode->i_size) {
483 err = btrfs_drop_extents(trans, root, inode,
484 start_pos, (pos + write_bytes + root->blocksize -1) &
485 ~((u64)root->blocksize - 1), &hint_block);
486 if (err)
487 goto failed_release;
488 }
489
490 /* insert any holes we need to create */
491 if (inode->i_size < start_pos) {
492 u64 last_pos_in_file;
493 u64 hole_size;
494 u64 mask = root->blocksize - 1;
495 last_pos_in_file = (isize + mask) & ~mask;
496 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
497 hole_size >>= inode->i_blkbits;
498 if (last_pos_in_file < start_pos) {
499 err = btrfs_insert_file_extent(trans, root,
500 inode->i_ino,
501 last_pos_in_file,
502 0, 0, hole_size);
503 }
504 if (err)
505 goto failed_release;
506 }
507
508 /*
509 * either allocate an extent for the new bytes or setup the key
510 * to show we are doing inline data in the extent
511 */
512 if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
513 pos + write_bytes - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
514 err = btrfs_alloc_extent(trans, root, inode->i_ino,
515 num_blocks, 0, hint_block, (u64)-1,
516 &ins, 1);
517 if (err)
518 goto failed_truncate;
519 err = btrfs_insert_file_extent(trans, root, inode->i_ino,
520 start_pos, ins.objectid, ins.offset,
521 ins.offset);
522 if (err)
523 goto failed_truncate;
524 } else {
525 ins.offset = 0;
526 ins.objectid = 0;
527 }
528 BUG_ON(err);
529 alloc_extent_start = ins.objectid;
530 err = btrfs_end_transaction(trans, root);
531 mutex_unlock(&root->fs_info->fs_mutex);
532
533 for (i = 0; i < num_pages; i++) {
534 offset = pos & (PAGE_CACHE_SIZE -1);
535 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
536 if (!page_has_buffers(pages[i])) {
537 create_empty_buffers(pages[i],
538 root->fs_info->sb->s_blocksize,
539 (1 << BH_Uptodate));
540 } 552 }
541 head = page_buffers(pages[i]);
542 bh = head;
543 do {
544 err = btrfs_map_bh_to_logical(root, bh,
545 alloc_extent_start);
546 BUG_ON(err);
547 if (err)
548 goto failed_truncate;
549 bh = bh->b_this_page;
550 if (alloc_extent_start)
551 alloc_extent_start++;
552 } while (bh != head);
553 pos += this_write;
554 WARN_ON(this_write > write_bytes);
555 write_bytes -= this_write;
556 } 553 }
557 return 0; 554 return 0;
558
559failed_release:
560 btrfs_drop_pages(pages, num_pages);
561 return err;
562
563failed_truncate:
564 btrfs_drop_pages(pages, num_pages);
565 if (pos > isize)
566 vmtruncate(inode, isize);
567 return err;
568
569out_unlock:
570 mutex_unlock(&root->fs_info->fs_mutex);
571 goto failed_release;
572
573} 555}
574 556
575static ssize_t btrfs_file_write(struct file *file, const char __user *buf, 557static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
@@ -685,7 +667,6 @@ out:
685 page_cache_release(pinned[1]); 667 page_cache_release(pinned[1]);
686 *ppos = pos; 668 *ppos = pos;
687 current->backing_dev_info = NULL; 669 current->backing_dev_info = NULL;
688 mark_inode_dirty(inode);
689 return num_written ? num_written : err; 670 return num_written ? num_written : err;
690} 671}
691 672
@@ -714,8 +695,8 @@ static int btrfs_sync_file(struct file *file,
714 mutex_unlock(&root->fs_info->trans_mutex); 695 mutex_unlock(&root->fs_info->trans_mutex);
715 696
716 /* 697 /*
717 * ok we haven't committed the transaction yet, lets do a commit 698 * ok we haven't committed the transaction yet, lets do a commit
718 */ 699 */
719 trans = btrfs_start_transaction(root, 1); 700 trans = btrfs_start_transaction(root, 1);
720 if (!trans) { 701 if (!trans) {
721 ret = -ENOMEM; 702 ret = -ENOMEM;
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 504d37dfa6c9..fd4cb65fcf1f 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -121,6 +121,8 @@ void btrfs_read_locked_inode(struct inode *inode)
121 switch (inode->i_mode & S_IFMT) { 121 switch (inode->i_mode & S_IFMT) {
122 case S_IFREG: 122 case S_IFREG:
123 inode->i_mapping->a_ops = &btrfs_aops; 123 inode->i_mapping->a_ops = &btrfs_aops;
124 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
125 inode->i_mapping, GFP_NOFS);
124 inode->i_fop = &btrfs_file_operations; 126 inode->i_fop = &btrfs_file_operations;
125 inode->i_op = &btrfs_file_inode_operations; 127 inode->i_op = &btrfs_file_inode_operations;
126 break; 128 break;
@@ -169,7 +171,7 @@ static void fill_inode_item(struct btrfs_inode_item *item,
169 BTRFS_I(inode)->block_group->key.objectid); 171 BTRFS_I(inode)->block_group->key.objectid);
170} 172}
171 173
172static int btrfs_update_inode(struct btrfs_trans_handle *trans, 174int btrfs_update_inode(struct btrfs_trans_handle *trans,
173 struct btrfs_root *root, 175 struct btrfs_root *root,
174 struct inode *inode) 176 struct inode *inode)
175{ 177{
@@ -400,6 +402,7 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
400 int found_extent; 402 int found_extent;
401 int del_item; 403 int del_item;
402 404
405 btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
403 path = btrfs_alloc_path(); 406 path = btrfs_alloc_path();
404 path->reada = -1; 407 path->reada = -1;
405 BUG_ON(!path); 408 BUG_ON(!path);
@@ -511,6 +514,50 @@ error:
511 return ret; 514 return ret;
512} 515}
513 516
517static int btrfs_cow_one_page(struct btrfs_trans_handle *trans,
518 struct inode *inode, struct page *page,
519 size_t zero_start)
520{
521 char *kaddr;
522 int ret = 0;
523 struct btrfs_root *root = BTRFS_I(inode)->root;
524 u64 alloc_hint = 0;
525 u64 page_start = page->index << PAGE_CACHE_SHIFT;
526 struct btrfs_key ins;
527
528 if (!PagePrivate(page)) {
529 SetPagePrivate(page);
530 set_page_private(page, 1);
531 page_cache_get(page);
532 }
533
534 btrfs_set_trans_block_group(trans, inode);
535
536 ret = btrfs_drop_extents(trans, root, inode,
537 page_start, page_start + PAGE_CACHE_SIZE,
538 &alloc_hint);
539 if (ret)
540 goto out;
541 ret = btrfs_alloc_extent(trans, root, inode->i_ino, 1, 0,
542 alloc_hint, (u64)-1, &ins, 1);
543 if (ret)
544 goto out;
545 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
546 page_start, ins.objectid, 1, 1);
547 if (ret)
548 goto out;
549 SetPageChecked(page);
550 kaddr = kmap(page);
551 if (zero_start != PAGE_CACHE_SIZE) {
552 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
553 flush_dcache_page(page);
554 }
555 kunmap(page);
556
557out:
558 return ret;
559}
560
514/* 561/*
515 * taken from block_truncate_page, but does cow as it zeros out 562 * taken from block_truncate_page, but does cow as it zeros out
516 * any bytes left in the last page in the file. 563 * any bytes left in the last page in the file.
@@ -518,16 +565,14 @@ error:
518static int btrfs_truncate_page(struct address_space *mapping, loff_t from) 565static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
519{ 566{
520 struct inode *inode = mapping->host; 567 struct inode *inode = mapping->host;
568 struct btrfs_root *root = BTRFS_I(inode)->root;
569 struct btrfs_trans_handle *trans;
521 unsigned blocksize = 1 << inode->i_blkbits; 570 unsigned blocksize = 1 << inode->i_blkbits;
522 pgoff_t index = from >> PAGE_CACHE_SHIFT; 571 pgoff_t index = from >> PAGE_CACHE_SHIFT;
523 unsigned offset = from & (PAGE_CACHE_SIZE-1); 572 unsigned offset = from & (PAGE_CACHE_SIZE-1);
524 struct page *page; 573 struct page *page;
525 char *kaddr;
526 int ret = 0; 574 int ret = 0;
527 struct btrfs_root *root = BTRFS_I(inode)->root; 575 u64 page_start;
528 u64 alloc_hint = 0;
529 struct btrfs_key ins;
530 struct btrfs_trans_handle *trans;
531 576
532 if ((offset & (blocksize - 1)) == 0) 577 if ((offset & (blocksize - 1)) == 0)
533 goto out; 578 goto out;
@@ -536,7 +581,6 @@ static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
536 page = grab_cache_page(mapping, index); 581 page = grab_cache_page(mapping, index);
537 if (!page) 582 if (!page)
538 goto out; 583 goto out;
539
540 if (!PageUptodate(page)) { 584 if (!PageUptodate(page)) {
541 ret = btrfs_readpage(NULL, page); 585 ret = btrfs_readpage(NULL, page);
542 lock_page(page); 586 lock_page(page);
@@ -545,37 +589,24 @@ static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
545 goto out; 589 goto out;
546 } 590 }
547 } 591 }
592 page_start = page->index << PAGE_CACHE_SHIFT;
593
548 mutex_lock(&root->fs_info->fs_mutex); 594 mutex_lock(&root->fs_info->fs_mutex);
549 trans = btrfs_start_transaction(root, 1); 595 trans = btrfs_start_transaction(root, 1);
550 btrfs_set_trans_block_group(trans, inode); 596 ret = btrfs_cow_one_page(trans, inode, page, offset);
551 597 if (!ret) {
552 ret = btrfs_drop_extents(trans, root, inode, 598 char *kaddr = kmap(page);
553 page->index << PAGE_CACHE_SHIFT, 599 ret = btrfs_csum_file_block(trans, root, inode->i_ino,
554 (page->index + 1) << PAGE_CACHE_SHIFT, 600 page_start, kaddr, PAGE_CACHE_SIZE);
555 &alloc_hint); 601 kunmap(page);
556 if (ret) 602 }
557 goto out; 603 set_extent_dirty(&BTRFS_I(inode)->extent_tree,
558 ret = btrfs_alloc_extent(trans, root, inode->i_ino, 1, 0, 604 page_start, page_start + PAGE_CACHE_SIZE - 1,
559 alloc_hint, (u64)-1, &ins, 1); 605 GFP_NOFS);
560 if (ret) 606 set_page_dirty(page);
561 goto out;
562 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
563 page->index << PAGE_CACHE_SHIFT,
564 ins.objectid, 1, 1);
565 if (ret)
566 goto out;
567 SetPageChecked(page);
568 kaddr = kmap(page);
569 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
570 flush_dcache_page(page);
571 ret = btrfs_csum_file_block(trans, root, inode->i_ino,
572 page->index << PAGE_CACHE_SHIFT,
573 kaddr, PAGE_CACHE_SIZE);
574 kunmap(page);
575 btrfs_end_transaction(trans, root); 607 btrfs_end_transaction(trans, root);
576 mutex_unlock(&root->fs_info->fs_mutex); 608 mutex_unlock(&root->fs_info->fs_mutex);
577 609
578 set_page_dirty(page);
579 unlock_page(page); 610 unlock_page(page);
580 page_cache_release(page); 611 page_cache_release(page);
581out: 612out:
@@ -1095,6 +1126,8 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
1095 inode->i_mapping->a_ops = &btrfs_aops; 1126 inode->i_mapping->a_ops = &btrfs_aops;
1096 inode->i_fop = &btrfs_file_operations; 1127 inode->i_fop = &btrfs_file_operations;
1097 inode->i_op = &btrfs_file_inode_operations; 1128 inode->i_op = &btrfs_file_inode_operations;
1129 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1130 inode->i_mapping, GFP_NOFS);
1098 } 1131 }
1099 dir->i_sb->s_dirt = 1; 1132 dir->i_sb->s_dirt = 1;
1100 btrfs_update_inode_block_group(trans, inode); 1133 btrfs_update_inode_block_group(trans, inode);
@@ -1238,6 +1271,182 @@ out_unlock:
1238 return err; 1271 return err;
1239} 1272}
1240 1273
1274struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1275 size_t page_offset, u64 start, u64 end,
1276 int create)
1277{
1278 int ret;
1279 int err = 0;
1280 u64 blocknr;
1281 u64 extent_start = 0;
1282 u64 extent_end = 0;
1283 u64 objectid = inode->i_ino;
1284 u32 found_type;
1285 int failed_insert = 0;
1286 struct btrfs_path *path;
1287 struct btrfs_root *root = BTRFS_I(inode)->root;
1288 struct btrfs_file_extent_item *item;
1289 struct btrfs_leaf *leaf;
1290 struct btrfs_disk_key *found_key;
1291 struct extent_map *em = NULL;
1292 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1293 struct btrfs_trans_handle *trans = NULL;
1294
1295 path = btrfs_alloc_path();
1296 BUG_ON(!path);
1297 mutex_lock(&root->fs_info->fs_mutex);
1298
1299again:
1300 em = lookup_extent_mapping(em_tree, start, end);
1301 if (em) {
1302 goto out;
1303 }
1304 if (!em) {
1305 em = alloc_extent_map(GFP_NOFS);
1306 if (!em) {
1307 err = -ENOMEM;
1308 goto out;
1309 }
1310 em->start = 0;
1311 em->end = 0;
1312 }
1313 em->bdev = inode->i_sb->s_bdev;
1314 ret = btrfs_lookup_file_extent(NULL, root, path,
1315 objectid, start, 0);
1316 if (ret < 0) {
1317 err = ret;
1318 goto out;
1319 }
1320
1321 if (ret != 0) {
1322 if (path->slots[0] == 0)
1323 goto not_found;
1324 path->slots[0]--;
1325 }
1326
1327 item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
1328 struct btrfs_file_extent_item);
1329 leaf = btrfs_buffer_leaf(path->nodes[0]);
1330 blocknr = btrfs_file_extent_disk_blocknr(item);
1331 blocknr += btrfs_file_extent_offset(item);
1332
1333 /* are we inside the extent that was found? */
1334 found_key = &leaf->items[path->slots[0]].key;
1335 found_type = btrfs_disk_key_type(found_key);
1336 if (btrfs_disk_key_objectid(found_key) != objectid ||
1337 found_type != BTRFS_EXTENT_DATA_KEY) {
1338 goto not_found;
1339 }
1340
1341 found_type = btrfs_file_extent_type(item);
1342 extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
1343 if (found_type == BTRFS_FILE_EXTENT_REG) {
1344 extent_end = extent_start +
1345 (btrfs_file_extent_num_blocks(item) << inode->i_blkbits);
1346 err = 0;
1347 if (start < extent_start || start > extent_end) {
1348 em->start = start;
1349 if (start < extent_start) {
1350 em->end = extent_end - 1;
1351 } else {
1352 em->end = end;
1353 }
1354 goto not_found_em;
1355 }
1356 if (btrfs_file_extent_disk_blocknr(item) == 0) {
1357 em->start = extent_start;
1358 em->end = extent_end - 1;
1359 em->block_start = 0;
1360 em->block_end = 0;
1361 goto insert;
1362 }
1363 em->block_start = blocknr << inode->i_blkbits;
1364 em->block_end = em->block_start +
1365 (btrfs_file_extent_num_blocks(item) <<
1366 inode->i_blkbits) - 1;
1367 em->start = extent_start;
1368 em->end = extent_end - 1;
1369 goto insert;
1370 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1371 char *ptr;
1372 char *map;
1373 u32 size;
1374
1375 size = btrfs_file_extent_inline_len(leaf->items +
1376 path->slots[0]);
1377 extent_end = extent_start + size;
1378 if (start < extent_start || start > extent_end) {
1379 em->start = start;
1380 if (start < extent_start) {
1381 em->end = extent_end - 1;
1382 } else {
1383 em->end = end;
1384 }
1385 goto not_found_em;
1386 }
1387 em->block_start = EXTENT_MAP_INLINE;
1388 em->block_end = EXTENT_MAP_INLINE;
1389 em->start = extent_start;
1390 em->end = extent_end - 1;
1391 if (!page) {
1392 goto insert;
1393 }
1394 ptr = btrfs_file_extent_inline_start(item);
1395 map = kmap(page);
1396 memcpy(map + page_offset, ptr, size);
1397 flush_dcache_page(result->b_page);
1398 kunmap(page);
1399 set_extent_uptodate(em_tree, extent_start,
1400 extent_end, GFP_NOFS);
1401 goto insert;
1402 } else {
1403 printk("unkknown found_type %d\n", found_type);
1404 WARN_ON(1);
1405 }
1406not_found:
1407 em->start = start;
1408 em->end = end;
1409not_found_em:
1410 em->block_start = 0;
1411 em->block_end = 0;
1412insert:
1413 btrfs_release_path(root, path);
1414 if (em->start > start || em->end < start) {
1415 printk("bad extent! %Lu %Lu start %Lu end %Lu\n", em->start, em->end, start, end);
1416 WARN_ON(1);
1417 err = -EIO;
1418 goto out;
1419 }
1420 ret = add_extent_mapping(em_tree, em);
1421 if (ret == -EEXIST) {
1422 free_extent_map(em);
1423 failed_insert++;
1424 if (failed_insert > 5) {
1425 printk("failing to insert %Lu %Lu\n", start, end);
1426 err = -EIO;
1427 goto out;
1428 }
1429 em = NULL;
1430 goto again;
1431 }
1432 err = 0;
1433out:
1434 btrfs_free_path(path);
1435 if (trans) {
1436 ret = btrfs_end_transaction(trans, root);
1437 if (!err)
1438 err = ret;
1439 }
1440 mutex_unlock(&root->fs_info->fs_mutex);
1441 if (err) {
1442 free_extent_map(em);
1443 WARN_ON(1);
1444 return ERR_PTR(err);
1445 }
1446 return em;
1447}
1448
1449
1241/* 1450/*
1242 * FIBMAP and others want to pass in a fake buffer head. They need to 1451 * FIBMAP and others want to pass in a fake buffer head. They need to
1243 * use BTRFS_GET_BLOCK_NO_DIRECT to make sure we don't try to memcpy 1452 * use BTRFS_GET_BLOCK_NO_DIRECT to make sure we don't try to memcpy
@@ -1398,46 +1607,22 @@ int btrfs_get_block(struct inode *inode, sector_t iblock,
1398 return err; 1607 return err;
1399} 1608}
1400 1609
1401static int btrfs_get_block_csum(struct inode *inode, sector_t iblock,
1402 struct buffer_head *result, int create)
1403{
1404 int ret;
1405 struct btrfs_root *root = BTRFS_I(inode)->root;
1406 struct page *page = result->b_page;
1407 u64 offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(result);
1408 struct btrfs_csum_item *item;
1409 struct btrfs_path *path = NULL;
1410
1411 mutex_lock(&root->fs_info->fs_mutex);
1412 ret = btrfs_get_block_lock(inode, iblock, result, create);
1413 if (ret)
1414 goto out;
1415
1416 path = btrfs_alloc_path();
1417 item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, offset, 0);
1418 if (IS_ERR(item)) {
1419 ret = PTR_ERR(item);
1420 /* a csum that isn't present is a preallocated region. */
1421 if (ret == -ENOENT || ret == -EFBIG)
1422 ret = 0;
1423 result->b_private = NULL;
1424 goto out;
1425 }
1426 memcpy((char *)&result->b_private, &item->csum, BTRFS_CRC32_SIZE);
1427out:
1428 if (path)
1429 btrfs_free_path(path);
1430 mutex_unlock(&root->fs_info->fs_mutex);
1431 return ret;
1432}
1433
1434static int btrfs_get_block_bmap(struct inode *inode, sector_t iblock, 1610static int btrfs_get_block_bmap(struct inode *inode, sector_t iblock,
1435 struct buffer_head *result, int create) 1611 struct buffer_head *result, int create)
1436{ 1612{
1437 struct btrfs_root *root = BTRFS_I(inode)->root; 1613 struct btrfs_root *root = BTRFS_I(inode)->root;
1438 mutex_lock(&root->fs_info->fs_mutex); 1614 u64 start = iblock << inode->i_blkbits;
1439 btrfs_get_block_lock(inode, iblock, result, BTRFS_GET_BLOCK_NO_DIRECT); 1615 u64 end = start + root->blocksize -1;
1440 mutex_unlock(&root->fs_info->fs_mutex); 1616 struct extent_map *em;
1617
1618 em = btrfs_get_extent(inode, NULL, 0, start, end, 0);
1619 if (em && !IS_ERR(em) && em->block_start != EXTENT_MAP_INLINE &&
1620 em->block_start != 0) {
1621 u64 offset;
1622 offset = start - em->start;
1623 start = (em->block_start + offset) >> inode->i_blkbits;
1624 btrfs_map_bh_to_logical(root, result, start);
1625 }
1441 return 0; 1626 return 0;
1442} 1627}
1443 1628
@@ -1449,442 +1634,50 @@ static sector_t btrfs_bmap(struct address_space *as, sector_t block)
1449static int btrfs_prepare_write(struct file *file, struct page *page, 1634static int btrfs_prepare_write(struct file *file, struct page *page,
1450 unsigned from, unsigned to) 1635 unsigned from, unsigned to)
1451{ 1636{
1452 return block_prepare_write(page, from, to, btrfs_get_block); 1637 return extent_prepare_write(&BTRFS_I(page->mapping->host)->extent_tree,
1638 page->mapping->host, page, from, to,
1639 btrfs_get_extent);
1453} 1640}
1454 1641
1455static void buffer_io_error(struct buffer_head *bh) 1642int btrfs_readpage(struct file *file, struct page *page)
1456{ 1643{
1457 char b[BDEVNAME_SIZE]; 1644 struct extent_map_tree *tree;
1458 1645 tree = &BTRFS_I(page->mapping->host)->extent_tree;
1459 printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n", 1646 return extent_read_full_page(tree, page, btrfs_get_extent);
1460 bdevname(bh->b_bdev, b),
1461 (unsigned long long)bh->b_blocknr);
1462} 1647}
1463 1648static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1464/*
1465 * I/O completion handler for block_read_full_page() - pages
1466 * which come unlocked at the end of I/O.
1467 */
1468static void btrfs_end_buffer_async_read(struct buffer_head *bh, int uptodate)
1469{ 1649{
1470 unsigned long flags; 1650 struct extent_map_tree *tree;
1471 struct buffer_head *first; 1651 tree = &BTRFS_I(page->mapping->host)->extent_tree;
1472 struct buffer_head *tmp; 1652 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
1473 struct page *page;
1474 int page_uptodate = 1;
1475 struct inode *inode;
1476 int ret;
1477
1478 BUG_ON(!buffer_async_read(bh));
1479
1480 page = bh->b_page;
1481 inode = page->mapping->host;
1482 if (uptodate) {
1483 void *kaddr;
1484 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1485 if (bh->b_private) {
1486 char csum[BTRFS_CRC32_SIZE];
1487 kaddr = kmap_atomic(page, KM_IRQ0);
1488 ret = btrfs_csum_data(root, kaddr + bh_offset(bh),
1489 bh->b_size, csum);
1490 BUG_ON(ret);
1491 if (memcmp(csum, &bh->b_private, BTRFS_CRC32_SIZE)) {
1492 u64 offset;
1493 offset = (page->index << PAGE_CACHE_SHIFT) +
1494 bh_offset(bh);
1495 printk("btrfs csum failed ino %lu off %llu\n",
1496 page->mapping->host->i_ino,
1497 (unsigned long long)offset);
1498 memset(kaddr + bh_offset(bh), 1, bh->b_size);
1499 flush_dcache_page(page);
1500 }
1501 kunmap_atomic(kaddr, KM_IRQ0);
1502 }
1503 set_buffer_uptodate(bh);
1504 } else {
1505 clear_buffer_uptodate(bh);
1506 if (printk_ratelimit())
1507 buffer_io_error(bh);
1508 SetPageError(page);
1509 }
1510
1511 /*
1512 * Be _very_ careful from here on. Bad things can happen if
1513 * two buffer heads end IO at almost the same time and both
1514 * decide that the page is now completely done.
1515 */
1516 first = page_buffers(page);
1517 local_irq_save(flags);
1518 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
1519 clear_buffer_async_read(bh);
1520 unlock_buffer(bh);
1521 tmp = bh;
1522 do {
1523 if (!buffer_uptodate(tmp))
1524 page_uptodate = 0;
1525 if (buffer_async_read(tmp)) {
1526 BUG_ON(!buffer_locked(tmp));
1527 goto still_busy;
1528 }
1529 tmp = tmp->b_this_page;
1530 } while (tmp != bh);
1531 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
1532 local_irq_restore(flags);
1533
1534 /*
1535 * If none of the buffers had errors and they are all
1536 * uptodate then we can set the page uptodate.
1537 */
1538 if (page_uptodate && !PageError(page))
1539 SetPageUptodate(page);
1540 unlock_page(page);
1541 return;
1542
1543still_busy:
1544 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
1545 local_irq_restore(flags);
1546 return;
1547} 1653}
1548 1654
1549/* 1655static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1550 * Generic "read page" function for block devices that have the normal
1551 * get_block functionality. This is most of the block device filesystems.
1552 * Reads the page asynchronously --- the unlock_buffer() and
1553 * set/clear_buffer_uptodate() functions propagate buffer state into the
1554 * page struct once IO has completed.
1555 */
1556int btrfs_readpage(struct file *file, struct page *page)
1557{ 1656{
1558 struct inode *inode = page->mapping->host; 1657 struct extent_map_tree *tree;
1559 sector_t iblock, lblock; 1658 int ret;
1560 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
1561 unsigned int blocksize;
1562 int nr, i;
1563 int fully_mapped = 1;
1564
1565 BUG_ON(!PageLocked(page));
1566 blocksize = 1 << inode->i_blkbits;
1567 if (!page_has_buffers(page))
1568 create_empty_buffers(page, blocksize, 0);
1569 head = page_buffers(page);
1570
1571 iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1572 lblock = (i_size_read(inode)+blocksize-1) >> inode->i_blkbits;
1573 bh = head;
1574 nr = 0;
1575 i = 0;
1576
1577 do {
1578 if (buffer_uptodate(bh))
1579 continue;
1580
1581 if (!buffer_mapped(bh)) {
1582 int err = 0;
1583
1584 fully_mapped = 0;
1585 if (iblock < lblock) {
1586 WARN_ON(bh->b_size != blocksize);
1587 err = btrfs_get_block_csum(inode, iblock,
1588 bh, 0);
1589 if (err)
1590 SetPageError(page);
1591 }
1592 if (!buffer_mapped(bh)) {
1593 void *kaddr = kmap_atomic(page, KM_USER0);
1594 memset(kaddr + i * blocksize, 0, blocksize);
1595 flush_dcache_page(page);
1596 kunmap_atomic(kaddr, KM_USER0);
1597 if (!err)
1598 set_buffer_uptodate(bh);
1599 continue;
1600 }
1601 /*
1602 * get_block() might have updated the buffer
1603 * synchronously
1604 */
1605 if (buffer_uptodate(bh))
1606 continue;
1607 }
1608 arr[nr++] = bh;
1609 } while (i++, iblock++, (bh = bh->b_this_page) != head);
1610
1611 if (fully_mapped)
1612 SetPageMappedToDisk(page);
1613
1614 if (!nr) {
1615 /*
1616 * All buffers are uptodate - we can set the page uptodate
1617 * as well. But not if get_block() returned an error.
1618 */
1619 if (!PageError(page))
1620 SetPageUptodate(page);
1621 unlock_page(page);
1622 return 0;
1623 }
1624
1625 /* Stage two: lock the buffers */
1626 for (i = 0; i < nr; i++) {
1627 bh = arr[i];
1628 lock_buffer(bh);
1629 bh->b_end_io = btrfs_end_buffer_async_read;
1630 set_buffer_async_read(bh);
1631 }
1632
1633 /*
1634 * Stage 3: start the IO. Check for uptodateness
1635 * inside the buffer lock in case another process reading
1636 * the underlying blockdev brought it uptodate (the sct fix).
1637 */
1638 for (i = 0; i < nr; i++) {
1639 bh = arr[i];
1640 if (buffer_uptodate(bh))
1641 btrfs_end_buffer_async_read(bh, 1);
1642 else
1643 submit_bh(READ, bh);
1644 }
1645 return 0;
1646}
1647
1648/*
1649 * Aside from a tiny bit of packed file data handling, this is the
1650 * same as the generic code.
1651 *
1652 * While block_write_full_page is writing back the dirty buffers under
1653 * the page lock, whoever dirtied the buffers may decide to clean them
1654 * again at any time. We handle that by only looking at the buffer
1655 * state inside lock_buffer().
1656 *
1657 * If block_write_full_page() is called for regular writeback
1658 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1659 * locked buffer. This only can happen if someone has written the buffer
1660 * directly, with submit_bh(). At the address_space level PageWriteback
1661 * prevents this contention from occurring.
1662 */
1663static int __btrfs_write_full_page(struct inode *inode, struct page *page,
1664 struct writeback_control *wbc)
1665{
1666 int err;
1667 sector_t block;
1668 sector_t last_block;
1669 struct buffer_head *bh, *head;
1670 const unsigned blocksize = 1 << inode->i_blkbits;
1671 int nr_underway = 0;
1672 struct btrfs_root *root = BTRFS_I(inode)->root;
1673
1674 BUG_ON(!PageLocked(page));
1675
1676 last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
1677
1678 /* no csumming allowed when from PF_MEMALLOC */
1679 if (current->flags & PF_MEMALLOC) {
1680 redirty_page_for_writepage(wbc, page);
1681 unlock_page(page);
1682 return 0;
1683 }
1684 1659
1685 if (!page_has_buffers(page)) { 1660 if (page->private != 1) {
1686 create_empty_buffers(page, blocksize, 1661 WARN_ON(1);
1687 (1 << BH_Dirty)|(1 << BH_Uptodate)); 1662 return try_to_free_buffers(page);
1688 } 1663 }
1689 1664 tree = &BTRFS_I(page->mapping->host)->extent_tree;
1690 /* 1665 ret = try_release_extent_mapping(tree, page);
1691 * Be very careful. We have no exclusion from __set_page_dirty_buffers 1666 if (ret == 1) {
1692 * here, and the (potentially unmapped) buffers may become dirty at 1667 ClearPagePrivate(page);
1693 * any time. If a buffer becomes dirty here after we've inspected it 1668 set_page_private(page, 0);
1694 * then we just miss that fact, and the page stays dirty. 1669 page_cache_release(page);
1695 *
1696 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1697 * handle that here by just cleaning them.
1698 */
1699
1700 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1701 head = page_buffers(page);
1702 bh = head;
1703
1704 /*
1705 * Get all the dirty buffers mapped to disk addresses and
1706 * handle any aliases from the underlying blockdev's mapping.
1707 */
1708 do {
1709 if (block > last_block) {
1710 /*
1711 * mapped buffers outside i_size will occur, because
1712 * this page can be outside i_size when there is a
1713 * truncate in progress.
1714 */
1715 /*
1716 * The buffer was zeroed by block_write_full_page()
1717 */
1718 clear_buffer_dirty(bh);
1719 set_buffer_uptodate(bh);
1720 } else if (!buffer_mapped(bh) && buffer_dirty(bh)) {
1721 WARN_ON(bh->b_size != blocksize);
1722 err = btrfs_get_block(inode, block, bh, 0);
1723 if (err) {
1724 goto recover;
1725 }
1726 if (buffer_new(bh)) {
1727 /* blockdev mappings never come here */
1728 clear_buffer_new(bh);
1729 }
1730 }
1731 bh = bh->b_this_page;
1732 block++;
1733 } while (bh != head);
1734
1735 do {
1736 if (!buffer_mapped(bh))
1737 continue;
1738 /*
1739 * If it's a fully non-blocking write attempt and we cannot
1740 * lock the buffer then redirty the page. Note that this can
1741 * potentially cause a busy-wait loop from pdflush and kswapd
1742 * activity, but those code paths have their own higher-level
1743 * throttling.
1744 */
1745 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
1746 lock_buffer(bh);
1747 } else if (test_set_buffer_locked(bh)) {
1748 redirty_page_for_writepage(wbc, page);
1749 continue;
1750 }
1751 if (test_clear_buffer_dirty(bh) && bh->b_blocknr != 0) {
1752 struct btrfs_trans_handle *trans;
1753 int ret;
1754 u64 off = page->index << PAGE_CACHE_SHIFT;
1755 char *kaddr;
1756
1757 off += bh_offset(bh);
1758 mutex_lock(&root->fs_info->fs_mutex);
1759 trans = btrfs_start_transaction(root, 1);
1760 btrfs_set_trans_block_group(trans, inode);
1761 kaddr = kmap(page);
1762 btrfs_csum_file_block(trans, root, inode->i_ino,
1763 off, kaddr + bh_offset(bh),
1764 bh->b_size);
1765 kunmap(page);
1766 ret = btrfs_end_transaction(trans, root);
1767 BUG_ON(ret);
1768 mutex_unlock(&root->fs_info->fs_mutex);
1769 mark_buffer_async_write(bh);
1770 } else {
1771 unlock_buffer(bh);
1772 }
1773 } while ((bh = bh->b_this_page) != head);
1774
1775 /*
1776 * The page and its buffers are protected by PageWriteback(), so we can
1777 * drop the bh refcounts early.
1778 */
1779 BUG_ON(PageWriteback(page));
1780 set_page_writeback(page);
1781
1782 do {
1783 struct buffer_head *next = bh->b_this_page;
1784 if (buffer_async_write(bh)) {
1785 submit_bh(WRITE, bh);
1786 nr_underway++;
1787 }
1788 bh = next;
1789 } while (bh != head);
1790 unlock_page(page);
1791
1792 err = 0;
1793done:
1794 if (nr_underway == 0) {
1795 /*
1796 * The page was marked dirty, but the buffers were
1797 * clean. Someone wrote them back by hand with
1798 * ll_rw_block/submit_bh. A rare case.
1799 */
1800 int uptodate = 1;
1801 do {
1802 if (!buffer_uptodate(bh)) {
1803 uptodate = 0;
1804 break;
1805 }
1806 bh = bh->b_this_page;
1807 } while (bh != head);
1808 if (uptodate)
1809 SetPageUptodate(page);
1810 end_page_writeback(page);
1811 } 1670 }
1812 return err; 1671 return ret;
1813
1814recover:
1815 /*
1816 * ENOSPC, or some other error. We may already have added some
1817 * blocks to the file, so we need to write these out to avoid
1818 * exposing stale data.
1819 * The page is currently locked and not marked for writeback
1820 */
1821 bh = head;
1822 /* Recovery: lock and submit the mapped buffers */
1823 do {
1824 if (buffer_mapped(bh) && buffer_dirty(bh)) {
1825 lock_buffer(bh);
1826 mark_buffer_async_write(bh);
1827 } else {
1828 /*
1829 * The buffer may have been set dirty during
1830 * attachment to a dirty page.
1831 */
1832 clear_buffer_dirty(bh);
1833 }
1834 } while ((bh = bh->b_this_page) != head);
1835 SetPageError(page);
1836 BUG_ON(PageWriteback(page));
1837 set_page_writeback(page);
1838 do {
1839 struct buffer_head *next = bh->b_this_page;
1840 if (buffer_async_write(bh)) {
1841 clear_buffer_dirty(bh);
1842 submit_bh(WRITE, bh);
1843 nr_underway++;
1844 }
1845 bh = next;
1846 } while (bh != head);
1847 unlock_page(page);
1848 goto done;
1849} 1672}
1850 1673
1851static int btrfs_writepage(struct page *page, struct writeback_control *wbc) 1674static void btrfs_invalidatepage(struct page *page, unsigned long offset)
1852{ 1675{
1853 struct inode * const inode = page->mapping->host; 1676 struct extent_map_tree *tree;
1854 loff_t i_size = i_size_read(inode);
1855 const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
1856 unsigned offset;
1857 void *kaddr;
1858
1859 /* Is the page fully inside i_size? */
1860 if (page->index < end_index)
1861 return __btrfs_write_full_page(inode, page, wbc);
1862
1863 /* Is the page fully outside i_size? (truncate in progress) */
1864 offset = i_size & (PAGE_CACHE_SIZE-1);
1865 if (page->index >= end_index+1 || !offset) {
1866 /*
1867 * The page may have dirty, unmapped buffers. For example,
1868 * they may have been added in ext3_writepage(). Make them
1869 * freeable here, so the page does not leak.
1870 */
1871 block_invalidatepage(page, 0);
1872 unlock_page(page);
1873 return 0; /* don't care */
1874 }
1875 1677
1876 /* 1678 tree = &BTRFS_I(page->mapping->host)->extent_tree;
1877 * The page straddles i_size. It must be zeroed out on each and every 1679 extent_invalidatepage(tree, page, offset);
1878 * writepage invokation because it may be mmapped. "A file is mapped 1680 btrfs_releasepage(page, GFP_NOFS);
1879 * in multiples of the page size. For a file that is not a multiple of
1880 * the page size, the remaining memory is zeroed when mapped, and
1881 * writes to that region are not written out to the file."
1882 */
1883 kaddr = kmap_atomic(page, KM_USER0);
1884 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1885 flush_dcache_page(page);
1886 kunmap_atomic(kaddr, KM_USER0);
1887 return __btrfs_write_full_page(inode, page, wbc);
1888} 1681}
1889 1682
1890/* 1683/*
@@ -1905,28 +1698,39 @@ static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1905int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page) 1698int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
1906{ 1699{
1907 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1700 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1701 struct btrfs_root *root = BTRFS_I(inode)->root;
1702 struct btrfs_trans_handle *trans;
1908 unsigned long end; 1703 unsigned long end;
1909 loff_t size; 1704 loff_t size;
1910 int ret = -EINVAL; 1705 int ret = -EINVAL;
1706 u64 page_start;
1911 1707
1912 lock_page(page); 1708 lock_page(page);
1913 wait_on_page_writeback(page); 1709 wait_on_page_writeback(page);
1914 size = i_size_read(inode); 1710 size = i_size_read(inode);
1711 page_start = page->index << PAGE_CACHE_SHIFT;
1712
1915 if ((page->mapping != inode->i_mapping) || 1713 if ((page->mapping != inode->i_mapping) ||
1916 ((page->index << PAGE_CACHE_SHIFT) > size)) { 1714 (page_start > size)) {
1917 /* page got truncated out from underneath us */ 1715 /* page got truncated out from underneath us */
1918 goto out_unlock; 1716 goto out_unlock;
1919 } 1717 }
1920 1718
1921 /* page is wholly or partially inside EOF */ 1719 /* page is wholly or partially inside EOF */
1922 if (((page->index + 1) << PAGE_CACHE_SHIFT) > size) 1720 if (page_start + PAGE_CACHE_SIZE > size)
1923 end = size & ~PAGE_CACHE_MASK; 1721 end = size & ~PAGE_CACHE_MASK;
1924 else 1722 else
1925 end = PAGE_CACHE_SIZE; 1723 end = PAGE_CACHE_SIZE;
1926 1724
1927 ret = btrfs_prepare_write(NULL, page, 0, end); 1725 mutex_lock(&root->fs_info->fs_mutex);
1928 if (!ret) 1726 trans = btrfs_start_transaction(root, 1);
1929 ret = btrfs_commit_write(NULL, page, 0, end); 1727 ret = btrfs_cow_one_page(trans, inode, page, end);
1728 btrfs_end_transaction(trans, root);
1729 mutex_unlock(&root->fs_info->fs_mutex);
1730 set_extent_dirty(&BTRFS_I(inode)->extent_tree,
1731 page_start, page_start + PAGE_CACHE_SIZE - 1,
1732 GFP_NOFS);
1733 set_page_dirty(page);
1930 1734
1931out_unlock: 1735out_unlock:
1932 unlock_page(page); 1736 unlock_page(page);
@@ -1962,21 +1766,8 @@ static void btrfs_truncate(struct inode *inode)
1962int btrfs_commit_write(struct file *file, struct page *page, 1766int btrfs_commit_write(struct file *file, struct page *page,
1963 unsigned from, unsigned to) 1767 unsigned from, unsigned to)
1964{ 1768{
1965 struct inode *inode = page->mapping->host; 1769 return extent_commit_write(&BTRFS_I(page->mapping->host)->extent_tree,
1966 struct buffer_head *bh; 1770 page->mapping->host, page, from, to);
1967 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1968
1969 SetPageUptodate(page);
1970 bh = page_buffers(page);
1971 set_buffer_uptodate(bh);
1972 if (buffer_mapped(bh) && bh->b_blocknr != 0) {
1973 set_page_dirty(page);
1974 }
1975 if (pos > inode->i_size) {
1976 i_size_write(inode, pos);
1977 mark_inode_dirty(inode);
1978 }
1979 return 0;
1980} 1771}
1981 1772
1982static int create_subvol(struct btrfs_root *root, char *name, int namelen) 1773static int create_subvol(struct btrfs_root *root, char *name, int namelen)
@@ -2471,6 +2262,8 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2471 inode->i_mapping->a_ops = &btrfs_aops; 2262 inode->i_mapping->a_ops = &btrfs_aops;
2472 inode->i_fop = &btrfs_file_operations; 2263 inode->i_fop = &btrfs_file_operations;
2473 inode->i_op = &btrfs_file_inode_operations; 2264 inode->i_op = &btrfs_file_inode_operations;
2265 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2266 inode->i_mapping, GFP_NOFS);
2474 } 2267 }
2475 dir->i_sb->s_dirt = 1; 2268 dir->i_sb->s_dirt = 1;
2476 btrfs_update_inode_block_group(trans, inode); 2269 btrfs_update_inode_block_group(trans, inode);
@@ -2553,6 +2346,9 @@ static struct address_space_operations btrfs_aops = {
2553 .prepare_write = btrfs_prepare_write, 2346 .prepare_write = btrfs_prepare_write,
2554 .commit_write = btrfs_commit_write, 2347 .commit_write = btrfs_commit_write,
2555 .bmap = btrfs_bmap, 2348 .bmap = btrfs_bmap,
2349 .invalidatepage = btrfs_invalidatepage,
2350 .releasepage = btrfs_releasepage,
2351 .set_page_dirty = __set_page_dirty_nobuffers,
2556}; 2352};
2557 2353
2558static struct address_space_operations btrfs_symlink_aops = { 2354static struct address_space_operations btrfs_symlink_aops = {
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 74f3de47423c..7b7c32c73a2c 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -185,6 +185,7 @@ static int __init init_btrfs_fs(void)
185 err = btrfs_init_cachep(); 185 err = btrfs_init_cachep();
186 if (err) 186 if (err)
187 return err; 187 return err;
188 extent_map_init();
188 return register_filesystem(&btrfs_fs_type); 189 return register_filesystem(&btrfs_fs_type);
189} 190}
190 191
@@ -192,6 +193,7 @@ static void __exit exit_btrfs_fs(void)
192{ 193{
193 btrfs_exit_transaction_sys(); 194 btrfs_exit_transaction_sys();
194 btrfs_destroy_cachep(); 195 btrfs_destroy_cachep();
196 extent_map_exit();
195 unregister_filesystem(&btrfs_fs_type); 197 unregister_filesystem(&btrfs_fs_type);
196} 198}
197 199
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-25 20:19:50 -0500