diff options
Diffstat (limited to 'fs/btrfs/free-space-cache.c')
-rw-r--r-- | fs/btrfs/free-space-cache.c | 751 |
1 files changed, 751 insertions, 0 deletions
diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c index f488fac04d99..22ee0dc2e6b8 100644 --- a/fs/btrfs/free-space-cache.c +++ b/fs/btrfs/free-space-cache.c | |||
@@ -23,10 +23,761 @@ | |||
23 | #include "ctree.h" | 23 | #include "ctree.h" |
24 | #include "free-space-cache.h" | 24 | #include "free-space-cache.h" |
25 | #include "transaction.h" | 25 | #include "transaction.h" |
26 | #include "disk-io.h" | ||
26 | 27 | ||
27 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) | 28 | #define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8) |
28 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) | 29 | #define MAX_CACHE_BYTES_PER_GIG (32 * 1024) |
29 | 30 | ||
31 | static void recalculate_thresholds(struct btrfs_block_group_cache | ||
32 | *block_group); | ||
33 | static int link_free_space(struct btrfs_block_group_cache *block_group, | ||
34 | struct btrfs_free_space *info); | ||
35 | |||
36 | struct inode *lookup_free_space_inode(struct btrfs_root *root, | ||
37 | struct btrfs_block_group_cache | ||
38 | *block_group, struct btrfs_path *path) | ||
39 | { | ||
40 | struct btrfs_key key; | ||
41 | struct btrfs_key location; | ||
42 | struct btrfs_disk_key disk_key; | ||
43 | struct btrfs_free_space_header *header; | ||
44 | struct extent_buffer *leaf; | ||
45 | struct inode *inode = NULL; | ||
46 | int ret; | ||
47 | |||
48 | spin_lock(&block_group->lock); | ||
49 | if (block_group->inode) | ||
50 | inode = igrab(block_group->inode); | ||
51 | spin_unlock(&block_group->lock); | ||
52 | if (inode) | ||
53 | return inode; | ||
54 | |||
55 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | ||
56 | key.offset = block_group->key.objectid; | ||
57 | key.type = 0; | ||
58 | |||
59 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | ||
60 | if (ret < 0) | ||
61 | return ERR_PTR(ret); | ||
62 | if (ret > 0) { | ||
63 | btrfs_release_path(root, path); | ||
64 | return ERR_PTR(-ENOENT); | ||
65 | } | ||
66 | |||
67 | leaf = path->nodes[0]; | ||
68 | header = btrfs_item_ptr(leaf, path->slots[0], | ||
69 | struct btrfs_free_space_header); | ||
70 | btrfs_free_space_key(leaf, header, &disk_key); | ||
71 | btrfs_disk_key_to_cpu(&location, &disk_key); | ||
72 | btrfs_release_path(root, path); | ||
73 | |||
74 | inode = btrfs_iget(root->fs_info->sb, &location, root, NULL); | ||
75 | if (!inode) | ||
76 | return ERR_PTR(-ENOENT); | ||
77 | if (IS_ERR(inode)) | ||
78 | return inode; | ||
79 | if (is_bad_inode(inode)) { | ||
80 | iput(inode); | ||
81 | return ERR_PTR(-ENOENT); | ||
82 | } | ||
83 | |||
84 | spin_lock(&block_group->lock); | ||
85 | if (!root->fs_info->closing) { | ||
86 | block_group->inode = igrab(inode); | ||
87 | block_group->iref = 1; | ||
88 | } | ||
89 | spin_unlock(&block_group->lock); | ||
90 | |||
91 | return inode; | ||
92 | } | ||
93 | |||
94 | int create_free_space_inode(struct btrfs_root *root, | ||
95 | struct btrfs_trans_handle *trans, | ||
96 | struct btrfs_block_group_cache *block_group, | ||
97 | struct btrfs_path *path) | ||
98 | { | ||
99 | struct btrfs_key key; | ||
100 | struct btrfs_disk_key disk_key; | ||
101 | struct btrfs_free_space_header *header; | ||
102 | struct btrfs_inode_item *inode_item; | ||
103 | struct extent_buffer *leaf; | ||
104 | u64 objectid; | ||
105 | int ret; | ||
106 | |||
107 | ret = btrfs_find_free_objectid(trans, root, 0, &objectid); | ||
108 | if (ret < 0) | ||
109 | return ret; | ||
110 | |||
111 | ret = btrfs_insert_empty_inode(trans, root, path, objectid); | ||
112 | if (ret) | ||
113 | return ret; | ||
114 | |||
115 | leaf = path->nodes[0]; | ||
116 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | ||
117 | struct btrfs_inode_item); | ||
118 | btrfs_item_key(leaf, &disk_key, path->slots[0]); | ||
119 | memset_extent_buffer(leaf, 0, (unsigned long)inode_item, | ||
120 | sizeof(*inode_item)); | ||
121 | btrfs_set_inode_generation(leaf, inode_item, trans->transid); | ||
122 | btrfs_set_inode_size(leaf, inode_item, 0); | ||
123 | btrfs_set_inode_nbytes(leaf, inode_item, 0); | ||
124 | btrfs_set_inode_uid(leaf, inode_item, 0); | ||
125 | btrfs_set_inode_gid(leaf, inode_item, 0); | ||
126 | btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600); | ||
127 | btrfs_set_inode_flags(leaf, inode_item, BTRFS_INODE_NOCOMPRESS | | ||
128 | BTRFS_INODE_PREALLOC | BTRFS_INODE_NODATASUM); | ||
129 | btrfs_set_inode_nlink(leaf, inode_item, 1); | ||
130 | btrfs_set_inode_transid(leaf, inode_item, trans->transid); | ||
131 | btrfs_set_inode_block_group(leaf, inode_item, | ||
132 | block_group->key.objectid); | ||
133 | btrfs_mark_buffer_dirty(leaf); | ||
134 | btrfs_release_path(root, path); | ||
135 | |||
136 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | ||
137 | key.offset = block_group->key.objectid; | ||
138 | key.type = 0; | ||
139 | |||
140 | ret = btrfs_insert_empty_item(trans, root, path, &key, | ||
141 | sizeof(struct btrfs_free_space_header)); | ||
142 | if (ret < 0) { | ||
143 | btrfs_release_path(root, path); | ||
144 | return ret; | ||
145 | } | ||
146 | leaf = path->nodes[0]; | ||
147 | header = btrfs_item_ptr(leaf, path->slots[0], | ||
148 | struct btrfs_free_space_header); | ||
149 | memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header)); | ||
150 | btrfs_set_free_space_key(leaf, header, &disk_key); | ||
151 | btrfs_mark_buffer_dirty(leaf); | ||
152 | btrfs_release_path(root, path); | ||
153 | |||
154 | return 0; | ||
155 | } | ||
156 | |||
157 | int btrfs_truncate_free_space_cache(struct btrfs_root *root, | ||
158 | struct btrfs_trans_handle *trans, | ||
159 | struct btrfs_path *path, | ||
160 | struct inode *inode) | ||
161 | { | ||
162 | loff_t oldsize; | ||
163 | int ret = 0; | ||
164 | |||
165 | trans->block_rsv = root->orphan_block_rsv; | ||
166 | ret = btrfs_block_rsv_check(trans, root, | ||
167 | root->orphan_block_rsv, | ||
168 | 0, 5); | ||
169 | if (ret) | ||
170 | return ret; | ||
171 | |||
172 | oldsize = i_size_read(inode); | ||
173 | btrfs_i_size_write(inode, 0); | ||
174 | truncate_pagecache(inode, oldsize, 0); | ||
175 | |||
176 | /* | ||
177 | * We don't need an orphan item because truncating the free space cache | ||
178 | * will never be split across transactions. | ||
179 | */ | ||
180 | ret = btrfs_truncate_inode_items(trans, root, inode, | ||
181 | 0, BTRFS_EXTENT_DATA_KEY); | ||
182 | if (ret) { | ||
183 | WARN_ON(1); | ||
184 | return ret; | ||
185 | } | ||
186 | |||
187 | return btrfs_update_inode(trans, root, inode); | ||
188 | } | ||
189 | |||
190 | static int readahead_cache(struct inode *inode) | ||
191 | { | ||
192 | struct file_ra_state *ra; | ||
193 | unsigned long last_index; | ||
194 | |||
195 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | ||
196 | if (!ra) | ||
197 | return -ENOMEM; | ||
198 | |||
199 | file_ra_state_init(ra, inode->i_mapping); | ||
200 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | ||
201 | |||
202 | page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index); | ||
203 | |||
204 | kfree(ra); | ||
205 | |||
206 | return 0; | ||
207 | } | ||
208 | |||
209 | int load_free_space_cache(struct btrfs_fs_info *fs_info, | ||
210 | struct btrfs_block_group_cache *block_group) | ||
211 | { | ||
212 | struct btrfs_root *root = fs_info->tree_root; | ||
213 | struct inode *inode; | ||
214 | struct btrfs_free_space_header *header; | ||
215 | struct extent_buffer *leaf; | ||
216 | struct page *page; | ||
217 | struct btrfs_path *path; | ||
218 | u32 *checksums = NULL, *crc; | ||
219 | char *disk_crcs = NULL; | ||
220 | struct btrfs_key key; | ||
221 | struct list_head bitmaps; | ||
222 | u64 num_entries; | ||
223 | u64 num_bitmaps; | ||
224 | u64 generation; | ||
225 | u32 cur_crc = ~(u32)0; | ||
226 | pgoff_t index = 0; | ||
227 | unsigned long first_page_offset; | ||
228 | int num_checksums; | ||
229 | int ret = 0; | ||
230 | |||
231 | /* | ||
232 | * If we're unmounting then just return, since this does a search on the | ||
233 | * normal root and not the commit root and we could deadlock. | ||
234 | */ | ||
235 | smp_mb(); | ||
236 | if (fs_info->closing) | ||
237 | return 0; | ||
238 | |||
239 | /* | ||
240 | * If this block group has been marked to be cleared for one reason or | ||
241 | * another then we can't trust the on disk cache, so just return. | ||
242 | */ | ||
243 | spin_lock(&block_group->lock); | ||
244 | if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) { | ||
245 | spin_unlock(&block_group->lock); | ||
246 | return 0; | ||
247 | } | ||
248 | spin_unlock(&block_group->lock); | ||
249 | |||
250 | INIT_LIST_HEAD(&bitmaps); | ||
251 | |||
252 | path = btrfs_alloc_path(); | ||
253 | if (!path) | ||
254 | return 0; | ||
255 | |||
256 | inode = lookup_free_space_inode(root, block_group, path); | ||
257 | if (IS_ERR(inode)) { | ||
258 | btrfs_free_path(path); | ||
259 | return 0; | ||
260 | } | ||
261 | |||
262 | /* Nothing in the space cache, goodbye */ | ||
263 | if (!i_size_read(inode)) { | ||
264 | btrfs_free_path(path); | ||
265 | goto out; | ||
266 | } | ||
267 | |||
268 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | ||
269 | key.offset = block_group->key.objectid; | ||
270 | key.type = 0; | ||
271 | |||
272 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | ||
273 | if (ret) { | ||
274 | btrfs_free_path(path); | ||
275 | goto out; | ||
276 | } | ||
277 | |||
278 | leaf = path->nodes[0]; | ||
279 | header = btrfs_item_ptr(leaf, path->slots[0], | ||
280 | struct btrfs_free_space_header); | ||
281 | num_entries = btrfs_free_space_entries(leaf, header); | ||
282 | num_bitmaps = btrfs_free_space_bitmaps(leaf, header); | ||
283 | generation = btrfs_free_space_generation(leaf, header); | ||
284 | btrfs_free_path(path); | ||
285 | |||
286 | if (BTRFS_I(inode)->generation != generation) { | ||
287 | printk(KERN_ERR "btrfs: free space inode generation (%llu) did" | ||
288 | " not match free space cache generation (%llu) for " | ||
289 | "block group %llu\n", | ||
290 | (unsigned long long)BTRFS_I(inode)->generation, | ||
291 | (unsigned long long)generation, | ||
292 | (unsigned long long)block_group->key.objectid); | ||
293 | goto out; | ||
294 | } | ||
295 | |||
296 | if (!num_entries) | ||
297 | goto out; | ||
298 | |||
299 | /* Setup everything for doing checksumming */ | ||
300 | num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE; | ||
301 | checksums = crc = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS); | ||
302 | if (!checksums) | ||
303 | goto out; | ||
304 | first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64); | ||
305 | disk_crcs = kzalloc(first_page_offset, GFP_NOFS); | ||
306 | if (!disk_crcs) | ||
307 | goto out; | ||
308 | |||
309 | ret = readahead_cache(inode); | ||
310 | if (ret) { | ||
311 | ret = 0; | ||
312 | goto out; | ||
313 | } | ||
314 | |||
315 | while (1) { | ||
316 | struct btrfs_free_space_entry *entry; | ||
317 | struct btrfs_free_space *e; | ||
318 | void *addr; | ||
319 | unsigned long offset = 0; | ||
320 | unsigned long start_offset = 0; | ||
321 | int need_loop = 0; | ||
322 | |||
323 | if (!num_entries && !num_bitmaps) | ||
324 | break; | ||
325 | |||
326 | if (index == 0) { | ||
327 | start_offset = first_page_offset; | ||
328 | offset = start_offset; | ||
329 | } | ||
330 | |||
331 | page = grab_cache_page(inode->i_mapping, index); | ||
332 | if (!page) { | ||
333 | ret = 0; | ||
334 | goto free_cache; | ||
335 | } | ||
336 | |||
337 | if (!PageUptodate(page)) { | ||
338 | btrfs_readpage(NULL, page); | ||
339 | lock_page(page); | ||
340 | if (!PageUptodate(page)) { | ||
341 | unlock_page(page); | ||
342 | page_cache_release(page); | ||
343 | printk(KERN_ERR "btrfs: error reading free " | ||
344 | "space cache: %llu\n", | ||
345 | (unsigned long long) | ||
346 | block_group->key.objectid); | ||
347 | goto free_cache; | ||
348 | } | ||
349 | } | ||
350 | addr = kmap(page); | ||
351 | |||
352 | if (index == 0) { | ||
353 | u64 *gen; | ||
354 | |||
355 | memcpy(disk_crcs, addr, first_page_offset); | ||
356 | gen = addr + (sizeof(u32) * num_checksums); | ||
357 | if (*gen != BTRFS_I(inode)->generation) { | ||
358 | printk(KERN_ERR "btrfs: space cache generation" | ||
359 | " (%llu) does not match inode (%llu) " | ||
360 | "for block group %llu\n", | ||
361 | (unsigned long long)*gen, | ||
362 | (unsigned long long) | ||
363 | BTRFS_I(inode)->generation, | ||
364 | (unsigned long long) | ||
365 | block_group->key.objectid); | ||
366 | kunmap(page); | ||
367 | unlock_page(page); | ||
368 | page_cache_release(page); | ||
369 | goto free_cache; | ||
370 | } | ||
371 | crc = (u32 *)disk_crcs; | ||
372 | } | ||
373 | entry = addr + start_offset; | ||
374 | |||
375 | /* First lets check our crc before we do anything fun */ | ||
376 | cur_crc = ~(u32)0; | ||
377 | cur_crc = btrfs_csum_data(root, addr + start_offset, cur_crc, | ||
378 | PAGE_CACHE_SIZE - start_offset); | ||
379 | btrfs_csum_final(cur_crc, (char *)&cur_crc); | ||
380 | if (cur_crc != *crc) { | ||
381 | printk(KERN_ERR "btrfs: crc mismatch for page %lu in " | ||
382 | "block group %llu\n", index, | ||
383 | (unsigned long long)block_group->key.objectid); | ||
384 | kunmap(page); | ||
385 | unlock_page(page); | ||
386 | page_cache_release(page); | ||
387 | goto free_cache; | ||
388 | } | ||
389 | crc++; | ||
390 | |||
391 | while (1) { | ||
392 | if (!num_entries) | ||
393 | break; | ||
394 | |||
395 | need_loop = 1; | ||
396 | e = kzalloc(sizeof(struct btrfs_free_space), GFP_NOFS); | ||
397 | if (!e) { | ||
398 | kunmap(page); | ||
399 | unlock_page(page); | ||
400 | page_cache_release(page); | ||
401 | goto free_cache; | ||
402 | } | ||
403 | |||
404 | e->offset = le64_to_cpu(entry->offset); | ||
405 | e->bytes = le64_to_cpu(entry->bytes); | ||
406 | if (!e->bytes) { | ||
407 | kunmap(page); | ||
408 | kfree(e); | ||
409 | unlock_page(page); | ||
410 | page_cache_release(page); | ||
411 | goto free_cache; | ||
412 | } | ||
413 | |||
414 | if (entry->type == BTRFS_FREE_SPACE_EXTENT) { | ||
415 | spin_lock(&block_group->tree_lock); | ||
416 | ret = link_free_space(block_group, e); | ||
417 | spin_unlock(&block_group->tree_lock); | ||
418 | BUG_ON(ret); | ||
419 | } else { | ||
420 | e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS); | ||
421 | if (!e->bitmap) { | ||
422 | kunmap(page); | ||
423 | kfree(e); | ||
424 | unlock_page(page); | ||
425 | page_cache_release(page); | ||
426 | goto free_cache; | ||
427 | } | ||
428 | spin_lock(&block_group->tree_lock); | ||
429 | ret = link_free_space(block_group, e); | ||
430 | block_group->total_bitmaps++; | ||
431 | recalculate_thresholds(block_group); | ||
432 | spin_unlock(&block_group->tree_lock); | ||
433 | list_add_tail(&e->list, &bitmaps); | ||
434 | } | ||
435 | |||
436 | num_entries--; | ||
437 | offset += sizeof(struct btrfs_free_space_entry); | ||
438 | if (offset + sizeof(struct btrfs_free_space_entry) >= | ||
439 | PAGE_CACHE_SIZE) | ||
440 | break; | ||
441 | entry++; | ||
442 | } | ||
443 | |||
444 | /* | ||
445 | * We read an entry out of this page, we need to move on to the | ||
446 | * next page. | ||
447 | */ | ||
448 | if (need_loop) { | ||
449 | kunmap(page); | ||
450 | goto next; | ||
451 | } | ||
452 | |||
453 | /* | ||
454 | * We add the bitmaps at the end of the entries in order that | ||
455 | * the bitmap entries are added to the cache. | ||
456 | */ | ||
457 | e = list_entry(bitmaps.next, struct btrfs_free_space, list); | ||
458 | list_del_init(&e->list); | ||
459 | memcpy(e->bitmap, addr, PAGE_CACHE_SIZE); | ||
460 | kunmap(page); | ||
461 | num_bitmaps--; | ||
462 | next: | ||
463 | unlock_page(page); | ||
464 | page_cache_release(page); | ||
465 | index++; | ||
466 | } | ||
467 | |||
468 | ret = 1; | ||
469 | out: | ||
470 | kfree(checksums); | ||
471 | kfree(disk_crcs); | ||
472 | iput(inode); | ||
473 | return ret; | ||
474 | |||
475 | free_cache: | ||
476 | /* This cache is bogus, make sure it gets cleared */ | ||
477 | spin_lock(&block_group->lock); | ||
478 | block_group->disk_cache_state = BTRFS_DC_CLEAR; | ||
479 | spin_unlock(&block_group->lock); | ||
480 | btrfs_remove_free_space_cache(block_group); | ||
481 | goto out; | ||
482 | } | ||
483 | |||
484 | int btrfs_write_out_cache(struct btrfs_root *root, | ||
485 | struct btrfs_trans_handle *trans, | ||
486 | struct btrfs_block_group_cache *block_group, | ||
487 | struct btrfs_path *path) | ||
488 | { | ||
489 | struct btrfs_free_space_header *header; | ||
490 | struct extent_buffer *leaf; | ||
491 | struct inode *inode; | ||
492 | struct rb_node *node; | ||
493 | struct list_head *pos, *n; | ||
494 | struct page *page; | ||
495 | struct extent_state *cached_state = NULL; | ||
496 | struct list_head bitmap_list; | ||
497 | struct btrfs_key key; | ||
498 | u64 bytes = 0; | ||
499 | u32 *crc, *checksums; | ||
500 | pgoff_t index = 0, last_index = 0; | ||
501 | unsigned long first_page_offset; | ||
502 | int num_checksums; | ||
503 | int entries = 0; | ||
504 | int bitmaps = 0; | ||
505 | int ret = 0; | ||
506 | |||
507 | root = root->fs_info->tree_root; | ||
508 | |||
509 | INIT_LIST_HEAD(&bitmap_list); | ||
510 | |||
511 | spin_lock(&block_group->lock); | ||
512 | if (block_group->disk_cache_state < BTRFS_DC_SETUP) { | ||
513 | spin_unlock(&block_group->lock); | ||
514 | return 0; | ||
515 | } | ||
516 | spin_unlock(&block_group->lock); | ||
517 | |||
518 | inode = lookup_free_space_inode(root, block_group, path); | ||
519 | if (IS_ERR(inode)) | ||
520 | return 0; | ||
521 | |||
522 | if (!i_size_read(inode)) { | ||
523 | iput(inode); | ||
524 | return 0; | ||
525 | } | ||
526 | |||
527 | last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | ||
528 | filemap_write_and_wait(inode->i_mapping); | ||
529 | btrfs_wait_ordered_range(inode, inode->i_size & | ||
530 | ~(root->sectorsize - 1), (u64)-1); | ||
531 | |||
532 | /* We need a checksum per page. */ | ||
533 | num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE; | ||
534 | crc = checksums = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS); | ||
535 | if (!crc) { | ||
536 | iput(inode); | ||
537 | return 0; | ||
538 | } | ||
539 | |||
540 | /* Since the first page has all of our checksums and our generation we | ||
541 | * need to calculate the offset into the page that we can start writing | ||
542 | * our entries. | ||
543 | */ | ||
544 | first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64); | ||
545 | |||
546 | node = rb_first(&block_group->free_space_offset); | ||
547 | if (!node) | ||
548 | goto out_free; | ||
549 | |||
550 | /* | ||
551 | * Lock all pages first so we can lock the extent safely. | ||
552 | * | ||
553 | * NOTE: Because we hold the ref the entire time we're going to write to | ||
554 | * the page find_get_page should never fail, so we don't do a check | ||
555 | * after find_get_page at this point. Just putting this here so people | ||
556 | * know and don't freak out. | ||
557 | */ | ||
558 | while (index <= last_index) { | ||
559 | page = grab_cache_page(inode->i_mapping, index); | ||
560 | if (!page) { | ||
561 | pgoff_t i = 0; | ||
562 | |||
563 | while (i < index) { | ||
564 | page = find_get_page(inode->i_mapping, i); | ||
565 | unlock_page(page); | ||
566 | page_cache_release(page); | ||
567 | page_cache_release(page); | ||
568 | i++; | ||
569 | } | ||
570 | goto out_free; | ||
571 | } | ||
572 | index++; | ||
573 | } | ||
574 | |||
575 | index = 0; | ||
576 | lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1, | ||
577 | 0, &cached_state, GFP_NOFS); | ||
578 | |||
579 | /* Write out the extent entries */ | ||
580 | do { | ||
581 | struct btrfs_free_space_entry *entry; | ||
582 | void *addr; | ||
583 | unsigned long offset = 0; | ||
584 | unsigned long start_offset = 0; | ||
585 | |||
586 | if (index == 0) { | ||
587 | start_offset = first_page_offset; | ||
588 | offset = start_offset; | ||
589 | } | ||
590 | |||
591 | page = find_get_page(inode->i_mapping, index); | ||
592 | |||
593 | addr = kmap(page); | ||
594 | entry = addr + start_offset; | ||
595 | |||
596 | memset(addr, 0, PAGE_CACHE_SIZE); | ||
597 | while (1) { | ||
598 | struct btrfs_free_space *e; | ||
599 | |||
600 | e = rb_entry(node, struct btrfs_free_space, offset_index); | ||
601 | entries++; | ||
602 | |||
603 | entry->offset = cpu_to_le64(e->offset); | ||
604 | entry->bytes = cpu_to_le64(e->bytes); | ||
605 | if (e->bitmap) { | ||
606 | entry->type = BTRFS_FREE_SPACE_BITMAP; | ||
607 | list_add_tail(&e->list, &bitmap_list); | ||
608 | bitmaps++; | ||
609 | } else { | ||
610 | entry->type = BTRFS_FREE_SPACE_EXTENT; | ||
611 | } | ||
612 | node = rb_next(node); | ||
613 | if (!node) | ||
614 | break; | ||
615 | offset += sizeof(struct btrfs_free_space_entry); | ||
616 | if (offset + sizeof(struct btrfs_free_space_entry) >= | ||
617 | PAGE_CACHE_SIZE) | ||
618 | break; | ||
619 | entry++; | ||
620 | } | ||
621 | *crc = ~(u32)0; | ||
622 | *crc = btrfs_csum_data(root, addr + start_offset, *crc, | ||
623 | PAGE_CACHE_SIZE - start_offset); | ||
624 | kunmap(page); | ||
625 | |||
626 | btrfs_csum_final(*crc, (char *)crc); | ||
627 | crc++; | ||
628 | |||
629 | bytes += PAGE_CACHE_SIZE; | ||
630 | |||
631 | ClearPageChecked(page); | ||
632 | set_page_extent_mapped(page); | ||
633 | SetPageUptodate(page); | ||
634 | set_page_dirty(page); | ||
635 | |||
636 | /* | ||
637 | * We need to release our reference we got for grab_cache_page, | ||
638 | * except for the first page which will hold our checksums, we | ||
639 | * do that below. | ||
640 | */ | ||
641 | if (index != 0) { | ||
642 | unlock_page(page); | ||
643 | page_cache_release(page); | ||
644 | } | ||
645 | |||
646 | page_cache_release(page); | ||
647 | |||
648 | index++; | ||
649 | } while (node); | ||
650 | |||
651 | /* Write out the bitmaps */ | ||
652 | list_for_each_safe(pos, n, &bitmap_list) { | ||
653 | void *addr; | ||
654 | struct btrfs_free_space *entry = | ||
655 | list_entry(pos, struct btrfs_free_space, list); | ||
656 | |||
657 | page = find_get_page(inode->i_mapping, index); | ||
658 | |||
659 | addr = kmap(page); | ||
660 | memcpy(addr, entry->bitmap, PAGE_CACHE_SIZE); | ||
661 | *crc = ~(u32)0; | ||
662 | *crc = btrfs_csum_data(root, addr, *crc, PAGE_CACHE_SIZE); | ||
663 | kunmap(page); | ||
664 | btrfs_csum_final(*crc, (char *)crc); | ||
665 | crc++; | ||
666 | bytes += PAGE_CACHE_SIZE; | ||
667 | |||
668 | ClearPageChecked(page); | ||
669 | set_page_extent_mapped(page); | ||
670 | SetPageUptodate(page); | ||
671 | set_page_dirty(page); | ||
672 | unlock_page(page); | ||
673 | page_cache_release(page); | ||
674 | page_cache_release(page); | ||
675 | list_del_init(&entry->list); | ||
676 | index++; | ||
677 | } | ||
678 | |||
679 | /* Zero out the rest of the pages just to make sure */ | ||
680 | while (index <= last_index) { | ||
681 | void *addr; | ||
682 | |||
683 | page = find_get_page(inode->i_mapping, index); | ||
684 | |||
685 | addr = kmap(page); | ||
686 | memset(addr, 0, PAGE_CACHE_SIZE); | ||
687 | kunmap(page); | ||
688 | ClearPageChecked(page); | ||
689 | set_page_extent_mapped(page); | ||
690 | SetPageUptodate(page); | ||
691 | set_page_dirty(page); | ||
692 | unlock_page(page); | ||
693 | page_cache_release(page); | ||
694 | page_cache_release(page); | ||
695 | bytes += PAGE_CACHE_SIZE; | ||
696 | index++; | ||
697 | } | ||
698 | |||
699 | btrfs_set_extent_delalloc(inode, 0, bytes - 1, &cached_state); | ||
700 | |||
701 | /* Write the checksums and trans id to the first page */ | ||
702 | { | ||
703 | void *addr; | ||
704 | u64 *gen; | ||
705 | |||
706 | page = find_get_page(inode->i_mapping, 0); | ||
707 | |||
708 | addr = kmap(page); | ||
709 | memcpy(addr, checksums, sizeof(u32) * num_checksums); | ||
710 | gen = addr + (sizeof(u32) * num_checksums); | ||
711 | *gen = trans->transid; | ||
712 | kunmap(page); | ||
713 | ClearPageChecked(page); | ||
714 | set_page_extent_mapped(page); | ||
715 | SetPageUptodate(page); | ||
716 | set_page_dirty(page); | ||
717 | unlock_page(page); | ||
718 | page_cache_release(page); | ||
719 | page_cache_release(page); | ||
720 | } | ||
721 | BTRFS_I(inode)->generation = trans->transid; | ||
722 | |||
723 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0, | ||
724 | i_size_read(inode) - 1, &cached_state, GFP_NOFS); | ||
725 | |||
726 | filemap_write_and_wait(inode->i_mapping); | ||
727 | |||
728 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | ||
729 | key.offset = block_group->key.objectid; | ||
730 | key.type = 0; | ||
731 | |||
732 | ret = btrfs_search_slot(trans, root, &key, path, 1, 1); | ||
733 | if (ret < 0) { | ||
734 | ret = 0; | ||
735 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1, | ||
736 | EXTENT_DIRTY | EXTENT_DELALLOC | | ||
737 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS); | ||
738 | goto out_free; | ||
739 | } | ||
740 | leaf = path->nodes[0]; | ||
741 | if (ret > 0) { | ||
742 | struct btrfs_key found_key; | ||
743 | BUG_ON(!path->slots[0]); | ||
744 | path->slots[0]--; | ||
745 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | ||
746 | if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID || | ||
747 | found_key.offset != block_group->key.objectid) { | ||
748 | ret = 0; | ||
749 | clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1, | ||
750 | EXTENT_DIRTY | EXTENT_DELALLOC | | ||
751 | EXTENT_DO_ACCOUNTING, 0, 0, NULL, | ||
752 | GFP_NOFS); | ||
753 | btrfs_release_path(root, path); | ||
754 | goto out_free; | ||
755 | } | ||
756 | } | ||
757 | header = btrfs_item_ptr(leaf, path->slots[0], | ||
758 | struct btrfs_free_space_header); | ||
759 | btrfs_set_free_space_entries(leaf, header, entries); | ||
760 | btrfs_set_free_space_bitmaps(leaf, header, bitmaps); | ||
761 | btrfs_set_free_space_generation(leaf, header, trans->transid); | ||
762 | btrfs_mark_buffer_dirty(leaf); | ||
763 | btrfs_release_path(root, path); | ||
764 | |||
765 | ret = 1; | ||
766 | |||
767 | out_free: | ||
768 | if (ret == 0) { | ||
769 | invalidate_inode_pages2_range(inode->i_mapping, 0, index); | ||
770 | spin_lock(&block_group->lock); | ||
771 | block_group->disk_cache_state = BTRFS_DC_ERROR; | ||
772 | spin_unlock(&block_group->lock); | ||
773 | BTRFS_I(inode)->generation = 0; | ||
774 | } | ||
775 | kfree(checksums); | ||
776 | btrfs_update_inode(trans, root, inode); | ||
777 | iput(inode); | ||
778 | return ret; | ||
779 | } | ||
780 | |||
30 | static inline unsigned long offset_to_bit(u64 bitmap_start, u64 sectorsize, | 781 | static inline unsigned long offset_to_bit(u64 bitmap_start, u64 sectorsize, |
31 | u64 offset) | 782 | u64 offset) |
32 | { | 783 | { |