Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
edbd8d4e | 19 | #include <linux/pagemap.h> |
ec44a35c | 20 | #include <linux/writeback.h> |
21af804c | 21 | #include <linux/blkdev.h> |
b7a9f29f | 22 | #include <linux/sort.h> |
4184ea7f | 23 | #include <linux/rcupdate.h> |
4b4e25f2 | 24 | #include "compat.h" |
74493f7a | 25 | #include "hash.h" |
a5eb62e3 | 26 | #include "crc32c.h" |
fec577fb CM |
27 | #include "ctree.h" |
28 | #include "disk-io.h" | |
29 | #include "print-tree.h" | |
e089f05c | 30 | #include "transaction.h" |
0b86a832 | 31 | #include "volumes.h" |
925baedd | 32 | #include "locking.h" |
31153d81 | 33 | #include "ref-cache.h" |
fec577fb | 34 | |
31840ae1 ZY |
35 | #define PENDING_EXTENT_INSERT 0 |
36 | #define PENDING_EXTENT_DELETE 1 | |
37 | #define PENDING_BACKREF_UPDATE 2 | |
38 | ||
39 | struct pending_extent_op { | |
40 | int type; | |
41 | u64 bytenr; | |
42 | u64 num_bytes; | |
43 | u64 parent; | |
44 | u64 orig_parent; | |
45 | u64 generation; | |
46 | u64 orig_generation; | |
47 | int level; | |
f3465ca4 JB |
48 | struct list_head list; |
49 | int del; | |
31840ae1 ZY |
50 | }; |
51 | ||
56bec294 CM |
52 | static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, |
53 | struct btrfs_root *root, u64 parent, | |
54 | u64 root_objectid, u64 ref_generation, | |
55 | u64 owner, struct btrfs_key *ins, | |
56 | int ref_mod); | |
57 | static int update_reserved_extents(struct btrfs_root *root, | |
58 | u64 bytenr, u64 num, int reserve); | |
f3465ca4 JB |
59 | static int update_block_group(struct btrfs_trans_handle *trans, |
60 | struct btrfs_root *root, | |
61 | u64 bytenr, u64 num_bytes, int alloc, | |
62 | int mark_free); | |
56bec294 CM |
63 | static noinline int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
64 | struct btrfs_root *root, | |
65 | u64 bytenr, u64 num_bytes, u64 parent, | |
66 | u64 root_objectid, u64 ref_generation, | |
67 | u64 owner_objectid, int pin, | |
68 | int ref_to_drop); | |
d548ee51 | 69 | |
6a63209f JB |
70 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
71 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
72 | u64 flags, int force); | |
73 | ||
0f9dd46c JB |
74 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
75 | { | |
76 | return (cache->flags & bits) == bits; | |
77 | } | |
78 | ||
79 | /* | |
80 | * this adds the block group to the fs_info rb tree for the block group | |
81 | * cache | |
82 | */ | |
b2950863 | 83 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
84 | struct btrfs_block_group_cache *block_group) |
85 | { | |
86 | struct rb_node **p; | |
87 | struct rb_node *parent = NULL; | |
88 | struct btrfs_block_group_cache *cache; | |
89 | ||
90 | spin_lock(&info->block_group_cache_lock); | |
91 | p = &info->block_group_cache_tree.rb_node; | |
92 | ||
93 | while (*p) { | |
94 | parent = *p; | |
95 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
96 | cache_node); | |
97 | if (block_group->key.objectid < cache->key.objectid) { | |
98 | p = &(*p)->rb_left; | |
99 | } else if (block_group->key.objectid > cache->key.objectid) { | |
100 | p = &(*p)->rb_right; | |
101 | } else { | |
102 | spin_unlock(&info->block_group_cache_lock); | |
103 | return -EEXIST; | |
104 | } | |
105 | } | |
106 | ||
107 | rb_link_node(&block_group->cache_node, parent, p); | |
108 | rb_insert_color(&block_group->cache_node, | |
109 | &info->block_group_cache_tree); | |
110 | spin_unlock(&info->block_group_cache_lock); | |
111 | ||
112 | return 0; | |
113 | } | |
114 | ||
115 | /* | |
116 | * This will return the block group at or after bytenr if contains is 0, else | |
117 | * it will return the block group that contains the bytenr | |
118 | */ | |
119 | static struct btrfs_block_group_cache * | |
120 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
121 | int contains) | |
122 | { | |
123 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
124 | struct rb_node *n; | |
125 | u64 end, start; | |
126 | ||
127 | spin_lock(&info->block_group_cache_lock); | |
128 | n = info->block_group_cache_tree.rb_node; | |
129 | ||
130 | while (n) { | |
131 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
132 | cache_node); | |
133 | end = cache->key.objectid + cache->key.offset - 1; | |
134 | start = cache->key.objectid; | |
135 | ||
136 | if (bytenr < start) { | |
137 | if (!contains && (!ret || start < ret->key.objectid)) | |
138 | ret = cache; | |
139 | n = n->rb_left; | |
140 | } else if (bytenr > start) { | |
141 | if (contains && bytenr <= end) { | |
142 | ret = cache; | |
143 | break; | |
144 | } | |
145 | n = n->rb_right; | |
146 | } else { | |
147 | ret = cache; | |
148 | break; | |
149 | } | |
150 | } | |
d2fb3437 YZ |
151 | if (ret) |
152 | atomic_inc(&ret->count); | |
0f9dd46c JB |
153 | spin_unlock(&info->block_group_cache_lock); |
154 | ||
155 | return ret; | |
156 | } | |
157 | ||
158 | /* | |
159 | * this is only called by cache_block_group, since we could have freed extents | |
160 | * we need to check the pinned_extents for any extents that can't be used yet | |
161 | * since their free space will be released as soon as the transaction commits. | |
162 | */ | |
163 | static int add_new_free_space(struct btrfs_block_group_cache *block_group, | |
164 | struct btrfs_fs_info *info, u64 start, u64 end) | |
165 | { | |
166 | u64 extent_start, extent_end, size; | |
167 | int ret; | |
168 | ||
169 | while (start < end) { | |
170 | ret = find_first_extent_bit(&info->pinned_extents, start, | |
171 | &extent_start, &extent_end, | |
172 | EXTENT_DIRTY); | |
173 | if (ret) | |
174 | break; | |
175 | ||
176 | if (extent_start == start) { | |
177 | start = extent_end + 1; | |
178 | } else if (extent_start > start && extent_start < end) { | |
179 | size = extent_start - start; | |
ea6a478e JB |
180 | ret = btrfs_add_free_space(block_group, start, |
181 | size); | |
0f9dd46c JB |
182 | BUG_ON(ret); |
183 | start = extent_end + 1; | |
184 | } else { | |
185 | break; | |
186 | } | |
187 | } | |
188 | ||
189 | if (start < end) { | |
190 | size = end - start; | |
ea6a478e | 191 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
192 | BUG_ON(ret); |
193 | } | |
194 | ||
195 | return 0; | |
196 | } | |
197 | ||
a512bbf8 YZ |
198 | static int remove_sb_from_cache(struct btrfs_root *root, |
199 | struct btrfs_block_group_cache *cache) | |
200 | { | |
201 | u64 bytenr; | |
202 | u64 *logical; | |
203 | int stripe_len; | |
204 | int i, nr, ret; | |
205 | ||
206 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
207 | bytenr = btrfs_sb_offset(i); | |
208 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
209 | cache->key.objectid, bytenr, 0, | |
210 | &logical, &nr, &stripe_len); | |
211 | BUG_ON(ret); | |
212 | while (nr--) { | |
213 | btrfs_remove_free_space(cache, logical[nr], | |
214 | stripe_len); | |
215 | } | |
216 | kfree(logical); | |
217 | } | |
218 | return 0; | |
219 | } | |
220 | ||
e37c9e69 CM |
221 | static int cache_block_group(struct btrfs_root *root, |
222 | struct btrfs_block_group_cache *block_group) | |
223 | { | |
224 | struct btrfs_path *path; | |
ef8bbdfe | 225 | int ret = 0; |
e37c9e69 | 226 | struct btrfs_key key; |
5f39d397 | 227 | struct extent_buffer *leaf; |
e37c9e69 | 228 | int slot; |
e4404d6e | 229 | u64 last; |
e37c9e69 | 230 | |
00f5c795 CM |
231 | if (!block_group) |
232 | return 0; | |
233 | ||
e37c9e69 | 234 | root = root->fs_info->extent_root; |
e37c9e69 CM |
235 | |
236 | if (block_group->cached) | |
237 | return 0; | |
f510cfec | 238 | |
e37c9e69 CM |
239 | path = btrfs_alloc_path(); |
240 | if (!path) | |
241 | return -ENOMEM; | |
7d7d6068 | 242 | |
2cc58cf2 | 243 | path->reada = 2; |
5cd57b2c CM |
244 | /* |
245 | * we get into deadlocks with paths held by callers of this function. | |
246 | * since the alloc_mutex is protecting things right now, just | |
247 | * skip the locking here | |
248 | */ | |
249 | path->skip_locking = 1; | |
e4404d6e YZ |
250 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
251 | key.objectid = last; | |
e37c9e69 CM |
252 | key.offset = 0; |
253 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
254 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
255 | if (ret < 0) | |
ef8bbdfe | 256 | goto err; |
a512bbf8 | 257 | |
d397712b | 258 | while (1) { |
5f39d397 | 259 | leaf = path->nodes[0]; |
e37c9e69 | 260 | slot = path->slots[0]; |
5f39d397 | 261 | if (slot >= btrfs_header_nritems(leaf)) { |
e37c9e69 | 262 | ret = btrfs_next_leaf(root, path); |
54aa1f4d CM |
263 | if (ret < 0) |
264 | goto err; | |
0f9dd46c | 265 | if (ret == 0) |
e37c9e69 | 266 | continue; |
0f9dd46c | 267 | else |
e37c9e69 | 268 | break; |
e37c9e69 | 269 | } |
5f39d397 | 270 | btrfs_item_key_to_cpu(leaf, &key, slot); |
0f9dd46c | 271 | if (key.objectid < block_group->key.objectid) |
7d7d6068 | 272 | goto next; |
0f9dd46c | 273 | |
e37c9e69 | 274 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 275 | block_group->key.offset) |
e37c9e69 | 276 | break; |
7d7d6068 | 277 | |
e37c9e69 | 278 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
0f9dd46c JB |
279 | add_new_free_space(block_group, root->fs_info, last, |
280 | key.objectid); | |
281 | ||
7d7d6068 | 282 | last = key.objectid + key.offset; |
e37c9e69 | 283 | } |
7d7d6068 | 284 | next: |
e37c9e69 CM |
285 | path->slots[0]++; |
286 | } | |
287 | ||
0f9dd46c JB |
288 | add_new_free_space(block_group, root->fs_info, last, |
289 | block_group->key.objectid + | |
290 | block_group->key.offset); | |
291 | ||
e37c9e69 | 292 | block_group->cached = 1; |
70cb0743 | 293 | remove_sb_from_cache(root, block_group); |
ef8bbdfe | 294 | ret = 0; |
54aa1f4d | 295 | err: |
e37c9e69 | 296 | btrfs_free_path(path); |
ef8bbdfe | 297 | return ret; |
e37c9e69 CM |
298 | } |
299 | ||
0f9dd46c JB |
300 | /* |
301 | * return the block group that starts at or after bytenr | |
302 | */ | |
d397712b CM |
303 | static struct btrfs_block_group_cache * |
304 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 305 | { |
0f9dd46c | 306 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 307 | |
0f9dd46c | 308 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 309 | |
0f9dd46c | 310 | return cache; |
0ef3e66b CM |
311 | } |
312 | ||
0f9dd46c JB |
313 | /* |
314 | * return the block group that contains teh given bytenr | |
315 | */ | |
d397712b CM |
316 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
317 | struct btrfs_fs_info *info, | |
318 | u64 bytenr) | |
be744175 | 319 | { |
0f9dd46c | 320 | struct btrfs_block_group_cache *cache; |
be744175 | 321 | |
0f9dd46c | 322 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 323 | |
0f9dd46c | 324 | return cache; |
be744175 | 325 | } |
0b86a832 | 326 | |
d2fb3437 YZ |
327 | static inline void put_block_group(struct btrfs_block_group_cache *cache) |
328 | { | |
329 | if (atomic_dec_and_test(&cache->count)) | |
330 | kfree(cache); | |
331 | } | |
332 | ||
0f9dd46c JB |
333 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
334 | u64 flags) | |
6324fbf3 | 335 | { |
0f9dd46c | 336 | struct list_head *head = &info->space_info; |
0f9dd46c | 337 | struct btrfs_space_info *found; |
4184ea7f CM |
338 | |
339 | rcu_read_lock(); | |
340 | list_for_each_entry_rcu(found, head, list) { | |
341 | if (found->flags == flags) { | |
342 | rcu_read_unlock(); | |
0f9dd46c | 343 | return found; |
4184ea7f | 344 | } |
0f9dd46c | 345 | } |
4184ea7f | 346 | rcu_read_unlock(); |
0f9dd46c | 347 | return NULL; |
6324fbf3 CM |
348 | } |
349 | ||
4184ea7f CM |
350 | /* |
351 | * after adding space to the filesystem, we need to clear the full flags | |
352 | * on all the space infos. | |
353 | */ | |
354 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
355 | { | |
356 | struct list_head *head = &info->space_info; | |
357 | struct btrfs_space_info *found; | |
358 | ||
359 | rcu_read_lock(); | |
360 | list_for_each_entry_rcu(found, head, list) | |
361 | found->full = 0; | |
362 | rcu_read_unlock(); | |
363 | } | |
364 | ||
80eb234a JB |
365 | static u64 div_factor(u64 num, int factor) |
366 | { | |
367 | if (factor == 10) | |
368 | return num; | |
369 | num *= factor; | |
370 | do_div(num, 10); | |
371 | return num; | |
372 | } | |
373 | ||
d2fb3437 YZ |
374 | u64 btrfs_find_block_group(struct btrfs_root *root, |
375 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 376 | { |
96b5179d | 377 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 378 | u64 used; |
d2fb3437 YZ |
379 | u64 last = max(search_hint, search_start); |
380 | u64 group_start = 0; | |
31f3c99b | 381 | int full_search = 0; |
d2fb3437 | 382 | int factor = 9; |
0ef3e66b | 383 | int wrapped = 0; |
31f3c99b | 384 | again: |
e8569813 ZY |
385 | while (1) { |
386 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
387 | if (!cache) |
388 | break; | |
96b5179d | 389 | |
c286ac48 | 390 | spin_lock(&cache->lock); |
96b5179d CM |
391 | last = cache->key.objectid + cache->key.offset; |
392 | used = btrfs_block_group_used(&cache->item); | |
393 | ||
d2fb3437 YZ |
394 | if ((full_search || !cache->ro) && |
395 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 396 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
397 | div_factor(cache->key.offset, factor)) { |
398 | group_start = cache->key.objectid; | |
c286ac48 | 399 | spin_unlock(&cache->lock); |
d2fb3437 | 400 | put_block_group(cache); |
8790d502 CM |
401 | goto found; |
402 | } | |
6324fbf3 | 403 | } |
c286ac48 | 404 | spin_unlock(&cache->lock); |
d2fb3437 | 405 | put_block_group(cache); |
de428b63 | 406 | cond_resched(); |
cd1bc465 | 407 | } |
0ef3e66b CM |
408 | if (!wrapped) { |
409 | last = search_start; | |
410 | wrapped = 1; | |
411 | goto again; | |
412 | } | |
413 | if (!full_search && factor < 10) { | |
be744175 | 414 | last = search_start; |
31f3c99b | 415 | full_search = 1; |
0ef3e66b | 416 | factor = 10; |
31f3c99b CM |
417 | goto again; |
418 | } | |
be744175 | 419 | found: |
d2fb3437 | 420 | return group_start; |
925baedd | 421 | } |
0f9dd46c | 422 | |
e02119d5 | 423 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 424 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
425 | { |
426 | int ret; | |
427 | struct btrfs_key key; | |
31840ae1 | 428 | struct btrfs_path *path; |
e02119d5 | 429 | |
31840ae1 ZY |
430 | path = btrfs_alloc_path(); |
431 | BUG_ON(!path); | |
e02119d5 CM |
432 | key.objectid = start; |
433 | key.offset = len; | |
434 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
435 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
436 | 0, 0); | |
31840ae1 | 437 | btrfs_free_path(path); |
7bb86316 CM |
438 | return ret; |
439 | } | |
440 | ||
d8d5f3e1 CM |
441 | /* |
442 | * Back reference rules. Back refs have three main goals: | |
443 | * | |
444 | * 1) differentiate between all holders of references to an extent so that | |
445 | * when a reference is dropped we can make sure it was a valid reference | |
446 | * before freeing the extent. | |
447 | * | |
448 | * 2) Provide enough information to quickly find the holders of an extent | |
449 | * if we notice a given block is corrupted or bad. | |
450 | * | |
451 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
452 | * maintenance. This is actually the same as #2, but with a slightly | |
453 | * different use case. | |
454 | * | |
455 | * File extents can be referenced by: | |
456 | * | |
457 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 458 | * - different files inside a single subvolume |
d8d5f3e1 CM |
459 | * - different offsets inside a file (bookend extents in file.c) |
460 | * | |
461 | * The extent ref structure has fields for: | |
462 | * | |
463 | * - Objectid of the subvolume root | |
464 | * - Generation number of the tree holding the reference | |
465 | * - objectid of the file holding the reference | |
31840ae1 ZY |
466 | * - number of references holding by parent node (alway 1 for tree blocks) |
467 | * | |
468 | * Btree leaf may hold multiple references to a file extent. In most cases, | |
469 | * these references are from same file and the corresponding offsets inside | |
3bb1a1bc | 470 | * the file are close together. |
d8d5f3e1 CM |
471 | * |
472 | * When a file extent is allocated the fields are filled in: | |
3bb1a1bc | 473 | * (root_key.objectid, trans->transid, inode objectid, 1) |
d8d5f3e1 CM |
474 | * |
475 | * When a leaf is cow'd new references are added for every file extent found | |
31840ae1 ZY |
476 | * in the leaf. It looks similar to the create case, but trans->transid will |
477 | * be different when the block is cow'd. | |
d8d5f3e1 | 478 | * |
3bb1a1bc | 479 | * (root_key.objectid, trans->transid, inode objectid, |
31840ae1 | 480 | * number of references in the leaf) |
d8d5f3e1 | 481 | * |
3bb1a1bc YZ |
482 | * When a file extent is removed either during snapshot deletion or |
483 | * file truncation, we find the corresponding back reference and check | |
484 | * the following fields: | |
d8d5f3e1 | 485 | * |
3bb1a1bc YZ |
486 | * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), |
487 | * inode objectid) | |
d8d5f3e1 CM |
488 | * |
489 | * Btree extents can be referenced by: | |
490 | * | |
491 | * - Different subvolumes | |
492 | * - Different generations of the same subvolume | |
493 | * | |
d8d5f3e1 CM |
494 | * When a tree block is created, back references are inserted: |
495 | * | |
3bb1a1bc | 496 | * (root->root_key.objectid, trans->transid, level, 1) |
d8d5f3e1 | 497 | * |
31840ae1 ZY |
498 | * When a tree block is cow'd, new back references are added for all the |
499 | * blocks it points to. If the tree block isn't in reference counted root, | |
500 | * the old back references are removed. These new back references are of | |
501 | * the form (trans->transid will have increased since creation): | |
d8d5f3e1 | 502 | * |
3bb1a1bc | 503 | * (root->root_key.objectid, trans->transid, level, 1) |
d8d5f3e1 | 504 | * |
31840ae1 | 505 | * When a backref is in deleting, the following fields are checked: |
d8d5f3e1 CM |
506 | * |
507 | * if backref was for a tree root: | |
3bb1a1bc | 508 | * (btrfs_header_owner(itself), btrfs_header_generation(itself), level) |
d8d5f3e1 | 509 | * else |
3bb1a1bc | 510 | * (btrfs_header_owner(parent), btrfs_header_generation(parent), level) |
d8d5f3e1 | 511 | * |
31840ae1 | 512 | * Back Reference Key composing: |
d8d5f3e1 | 513 | * |
31840ae1 ZY |
514 | * The key objectid corresponds to the first byte in the extent, the key |
515 | * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first | |
516 | * byte of parent extent. If a extent is tree root, the key offset is set | |
517 | * to the key objectid. | |
d8d5f3e1 | 518 | */ |
31840ae1 | 519 | |
d397712b | 520 | static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans, |
31840ae1 | 521 | struct btrfs_root *root, |
3bb1a1bc YZ |
522 | struct btrfs_path *path, |
523 | u64 bytenr, u64 parent, | |
524 | u64 ref_root, u64 ref_generation, | |
525 | u64 owner_objectid, int del) | |
7bb86316 | 526 | { |
7bb86316 | 527 | struct btrfs_key key; |
31840ae1 ZY |
528 | struct btrfs_extent_ref *ref; |
529 | struct extent_buffer *leaf; | |
3bb1a1bc | 530 | u64 ref_objectid; |
74493f7a CM |
531 | int ret; |
532 | ||
31840ae1 ZY |
533 | key.objectid = bytenr; |
534 | key.type = BTRFS_EXTENT_REF_KEY; | |
535 | key.offset = parent; | |
536 | ||
537 | ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1); | |
538 | if (ret < 0) | |
539 | goto out; | |
540 | if (ret > 0) { | |
541 | ret = -ENOENT; | |
542 | goto out; | |
543 | } | |
544 | ||
545 | leaf = path->nodes[0]; | |
546 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); | |
3bb1a1bc | 547 | ref_objectid = btrfs_ref_objectid(leaf, ref); |
31840ae1 | 548 | if (btrfs_ref_root(leaf, ref) != ref_root || |
3bb1a1bc YZ |
549 | btrfs_ref_generation(leaf, ref) != ref_generation || |
550 | (ref_objectid != owner_objectid && | |
551 | ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) { | |
31840ae1 ZY |
552 | ret = -EIO; |
553 | WARN_ON(1); | |
554 | goto out; | |
555 | } | |
556 | ret = 0; | |
557 | out: | |
558 | return ret; | |
559 | } | |
560 | ||
d397712b | 561 | static noinline int insert_extent_backref(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
562 | struct btrfs_root *root, |
563 | struct btrfs_path *path, | |
564 | u64 bytenr, u64 parent, | |
565 | u64 ref_root, u64 ref_generation, | |
56bec294 CM |
566 | u64 owner_objectid, |
567 | int refs_to_add) | |
31840ae1 ZY |
568 | { |
569 | struct btrfs_key key; | |
570 | struct extent_buffer *leaf; | |
571 | struct btrfs_extent_ref *ref; | |
572 | u32 num_refs; | |
573 | int ret; | |
74493f7a | 574 | |
74493f7a CM |
575 | key.objectid = bytenr; |
576 | key.type = BTRFS_EXTENT_REF_KEY; | |
31840ae1 | 577 | key.offset = parent; |
74493f7a | 578 | |
31840ae1 ZY |
579 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref)); |
580 | if (ret == 0) { | |
581 | leaf = path->nodes[0]; | |
582 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
583 | struct btrfs_extent_ref); | |
584 | btrfs_set_ref_root(leaf, ref, ref_root); | |
585 | btrfs_set_ref_generation(leaf, ref, ref_generation); | |
586 | btrfs_set_ref_objectid(leaf, ref, owner_objectid); | |
56bec294 | 587 | btrfs_set_ref_num_refs(leaf, ref, refs_to_add); |
31840ae1 ZY |
588 | } else if (ret == -EEXIST) { |
589 | u64 existing_owner; | |
56bec294 | 590 | |
31840ae1 ZY |
591 | BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID); |
592 | leaf = path->nodes[0]; | |
593 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
594 | struct btrfs_extent_ref); | |
595 | if (btrfs_ref_root(leaf, ref) != ref_root || | |
596 | btrfs_ref_generation(leaf, ref) != ref_generation) { | |
597 | ret = -EIO; | |
598 | WARN_ON(1); | |
7bb86316 | 599 | goto out; |
31840ae1 ZY |
600 | } |
601 | ||
602 | num_refs = btrfs_ref_num_refs(leaf, ref); | |
603 | BUG_ON(num_refs == 0); | |
56bec294 | 604 | btrfs_set_ref_num_refs(leaf, ref, num_refs + refs_to_add); |
31840ae1 ZY |
605 | |
606 | existing_owner = btrfs_ref_objectid(leaf, ref); | |
3bb1a1bc YZ |
607 | if (existing_owner != owner_objectid && |
608 | existing_owner != BTRFS_MULTIPLE_OBJECTIDS) { | |
31840ae1 ZY |
609 | btrfs_set_ref_objectid(leaf, ref, |
610 | BTRFS_MULTIPLE_OBJECTIDS); | |
31840ae1 ZY |
611 | } |
612 | ret = 0; | |
613 | } else { | |
7bb86316 | 614 | goto out; |
31840ae1 | 615 | } |
b9473439 | 616 | btrfs_unlock_up_safe(path, 1); |
7bb86316 CM |
617 | btrfs_mark_buffer_dirty(path->nodes[0]); |
618 | out: | |
619 | btrfs_release_path(root, path); | |
620 | return ret; | |
74493f7a CM |
621 | } |
622 | ||
d397712b | 623 | static noinline int remove_extent_backref(struct btrfs_trans_handle *trans, |
31840ae1 | 624 | struct btrfs_root *root, |
56bec294 CM |
625 | struct btrfs_path *path, |
626 | int refs_to_drop) | |
31840ae1 ZY |
627 | { |
628 | struct extent_buffer *leaf; | |
629 | struct btrfs_extent_ref *ref; | |
630 | u32 num_refs; | |
631 | int ret = 0; | |
632 | ||
633 | leaf = path->nodes[0]; | |
634 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); | |
635 | num_refs = btrfs_ref_num_refs(leaf, ref); | |
56bec294 CM |
636 | BUG_ON(num_refs < refs_to_drop); |
637 | num_refs -= refs_to_drop; | |
31840ae1 ZY |
638 | if (num_refs == 0) { |
639 | ret = btrfs_del_item(trans, root, path); | |
640 | } else { | |
641 | btrfs_set_ref_num_refs(leaf, ref, num_refs); | |
642 | btrfs_mark_buffer_dirty(leaf); | |
643 | } | |
644 | btrfs_release_path(root, path); | |
645 | return ret; | |
646 | } | |
647 | ||
4b4e25f2 | 648 | #ifdef BIO_RW_DISCARD |
15916de8 CM |
649 | static void btrfs_issue_discard(struct block_device *bdev, |
650 | u64 start, u64 len) | |
651 | { | |
15916de8 | 652 | blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL); |
15916de8 | 653 | } |
4b4e25f2 | 654 | #endif |
15916de8 | 655 | |
1f3c79a2 LH |
656 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, |
657 | u64 num_bytes) | |
658 | { | |
659 | #ifdef BIO_RW_DISCARD | |
660 | int ret; | |
661 | u64 map_length = num_bytes; | |
662 | struct btrfs_multi_bio *multi = NULL; | |
663 | ||
664 | /* Tell the block device(s) that the sectors can be discarded */ | |
665 | ret = btrfs_map_block(&root->fs_info->mapping_tree, READ, | |
666 | bytenr, &map_length, &multi, 0); | |
667 | if (!ret) { | |
668 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
669 | int i; | |
670 | ||
671 | if (map_length > num_bytes) | |
672 | map_length = num_bytes; | |
673 | ||
674 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
675 | btrfs_issue_discard(stripe->dev->bdev, | |
676 | stripe->physical, | |
677 | map_length); | |
678 | } | |
679 | kfree(multi); | |
680 | } | |
681 | ||
682 | return ret; | |
683 | #else | |
684 | return 0; | |
685 | #endif | |
686 | } | |
687 | ||
31840ae1 ZY |
688 | static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, |
689 | struct btrfs_root *root, u64 bytenr, | |
56bec294 | 690 | u64 num_bytes, |
31840ae1 ZY |
691 | u64 orig_parent, u64 parent, |
692 | u64 orig_root, u64 ref_root, | |
693 | u64 orig_generation, u64 ref_generation, | |
3bb1a1bc | 694 | u64 owner_objectid) |
31840ae1 ZY |
695 | { |
696 | int ret; | |
56bec294 | 697 | int pin = owner_objectid < BTRFS_FIRST_FREE_OBJECTID; |
31840ae1 | 698 | |
56bec294 CM |
699 | ret = btrfs_update_delayed_ref(trans, bytenr, num_bytes, |
700 | orig_parent, parent, orig_root, | |
701 | ref_root, orig_generation, | |
702 | ref_generation, owner_objectid, pin); | |
31840ae1 | 703 | BUG_ON(ret); |
31840ae1 ZY |
704 | return ret; |
705 | } | |
706 | ||
707 | int btrfs_update_extent_ref(struct btrfs_trans_handle *trans, | |
708 | struct btrfs_root *root, u64 bytenr, | |
56bec294 | 709 | u64 num_bytes, u64 orig_parent, u64 parent, |
31840ae1 | 710 | u64 ref_root, u64 ref_generation, |
3bb1a1bc | 711 | u64 owner_objectid) |
31840ae1 ZY |
712 | { |
713 | int ret; | |
714 | if (ref_root == BTRFS_TREE_LOG_OBJECTID && | |
715 | owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
716 | return 0; | |
56bec294 CM |
717 | |
718 | ret = __btrfs_update_extent_ref(trans, root, bytenr, num_bytes, | |
719 | orig_parent, parent, ref_root, | |
720 | ref_root, ref_generation, | |
721 | ref_generation, owner_objectid); | |
31840ae1 ZY |
722 | return ret; |
723 | } | |
925baedd | 724 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
31840ae1 | 725 | struct btrfs_root *root, u64 bytenr, |
56bec294 | 726 | u64 num_bytes, |
31840ae1 ZY |
727 | u64 orig_parent, u64 parent, |
728 | u64 orig_root, u64 ref_root, | |
729 | u64 orig_generation, u64 ref_generation, | |
3bb1a1bc | 730 | u64 owner_objectid) |
56bec294 CM |
731 | { |
732 | int ret; | |
733 | ||
734 | ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent, ref_root, | |
735 | ref_generation, owner_objectid, | |
736 | BTRFS_ADD_DELAYED_REF, 0); | |
737 | BUG_ON(ret); | |
738 | return ret; | |
739 | } | |
740 | ||
741 | static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans, | |
742 | struct btrfs_root *root, u64 bytenr, | |
743 | u64 num_bytes, u64 parent, u64 ref_root, | |
744 | u64 ref_generation, u64 owner_objectid, | |
745 | int refs_to_add) | |
02217ed2 | 746 | { |
5caf2a00 | 747 | struct btrfs_path *path; |
02217ed2 | 748 | int ret; |
e2fa7227 | 749 | struct btrfs_key key; |
5f39d397 | 750 | struct extent_buffer *l; |
234b63a0 | 751 | struct btrfs_extent_item *item; |
cf27e1ee | 752 | u32 refs; |
037e6390 | 753 | |
5caf2a00 | 754 | path = btrfs_alloc_path(); |
54aa1f4d CM |
755 | if (!path) |
756 | return -ENOMEM; | |
26b8003f | 757 | |
3c12ac72 | 758 | path->reada = 1; |
b9473439 | 759 | path->leave_spinning = 1; |
db94535d | 760 | key.objectid = bytenr; |
31840ae1 | 761 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 762 | key.offset = num_bytes; |
31840ae1 | 763 | |
56bec294 CM |
764 | /* first find the extent item and update its reference count */ |
765 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
766 | path, 0, 1); | |
b9473439 CM |
767 | if (ret < 0) { |
768 | btrfs_set_path_blocking(path); | |
54aa1f4d | 769 | return ret; |
b9473439 | 770 | } |
31840ae1 | 771 | |
56bec294 CM |
772 | if (ret > 0) { |
773 | WARN_ON(1); | |
774 | btrfs_free_path(path); | |
775 | return -EIO; | |
776 | } | |
5f39d397 | 777 | l = path->nodes[0]; |
31840ae1 ZY |
778 | |
779 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
771ed689 CM |
780 | if (key.objectid != bytenr) { |
781 | btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]); | |
d397712b CM |
782 | printk(KERN_ERR "btrfs wanted %llu found %llu\n", |
783 | (unsigned long long)bytenr, | |
784 | (unsigned long long)key.objectid); | |
771ed689 CM |
785 | BUG(); |
786 | } | |
31840ae1 ZY |
787 | BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY); |
788 | ||
56bec294 CM |
789 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
790 | ||
791 | refs = btrfs_extent_refs(l, item); | |
792 | btrfs_set_extent_refs(l, item, refs + refs_to_add); | |
b9473439 CM |
793 | btrfs_unlock_up_safe(path, 1); |
794 | ||
56bec294 CM |
795 | btrfs_mark_buffer_dirty(path->nodes[0]); |
796 | ||
797 | btrfs_release_path(root->fs_info->extent_root, path); | |
798 | ||
799 | path->reada = 1; | |
b9473439 CM |
800 | path->leave_spinning = 1; |
801 | ||
56bec294 CM |
802 | /* now insert the actual backref */ |
803 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
804 | path, bytenr, parent, | |
805 | ref_root, ref_generation, | |
806 | owner_objectid, refs_to_add); | |
807 | BUG_ON(ret); | |
808 | btrfs_free_path(path); | |
809 | return 0; | |
810 | } | |
811 | ||
812 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
813 | struct btrfs_root *root, | |
814 | u64 bytenr, u64 num_bytes, u64 parent, | |
815 | u64 ref_root, u64 ref_generation, | |
816 | u64 owner_objectid) | |
817 | { | |
818 | int ret; | |
819 | if (ref_root == BTRFS_TREE_LOG_OBJECTID && | |
820 | owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
821 | return 0; | |
822 | ||
823 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, parent, | |
824 | 0, ref_root, 0, ref_generation, | |
825 | owner_objectid); | |
826 | return ret; | |
827 | } | |
828 | ||
829 | static int drop_delayed_ref(struct btrfs_trans_handle *trans, | |
830 | struct btrfs_root *root, | |
831 | struct btrfs_delayed_ref_node *node) | |
832 | { | |
833 | int ret = 0; | |
834 | struct btrfs_delayed_ref *ref = btrfs_delayed_node_to_ref(node); | |
835 | ||
836 | BUG_ON(node->ref_mod == 0); | |
837 | ret = __btrfs_free_extent(trans, root, node->bytenr, node->num_bytes, | |
838 | node->parent, ref->root, ref->generation, | |
839 | ref->owner_objectid, ref->pin, node->ref_mod); | |
840 | ||
841 | return ret; | |
842 | } | |
843 | ||
844 | /* helper function to actually process a single delayed ref entry */ | |
845 | static noinline int run_one_delayed_ref(struct btrfs_trans_handle *trans, | |
846 | struct btrfs_root *root, | |
847 | struct btrfs_delayed_ref_node *node, | |
848 | int insert_reserved) | |
849 | { | |
850 | int ret; | |
851 | struct btrfs_delayed_ref *ref; | |
852 | ||
853 | if (node->parent == (u64)-1) { | |
854 | struct btrfs_delayed_ref_head *head; | |
855 | /* | |
856 | * we've hit the end of the chain and we were supposed | |
857 | * to insert this extent into the tree. But, it got | |
858 | * deleted before we ever needed to insert it, so all | |
859 | * we have to do is clean up the accounting | |
860 | */ | |
861 | if (insert_reserved) { | |
862 | update_reserved_extents(root, node->bytenr, | |
863 | node->num_bytes, 0); | |
864 | } | |
865 | head = btrfs_delayed_node_to_head(node); | |
866 | mutex_unlock(&head->mutex); | |
867 | return 0; | |
868 | } | |
869 | ||
870 | ref = btrfs_delayed_node_to_ref(node); | |
871 | if (ref->action == BTRFS_ADD_DELAYED_REF) { | |
872 | if (insert_reserved) { | |
873 | struct btrfs_key ins; | |
874 | ||
875 | ins.objectid = node->bytenr; | |
876 | ins.offset = node->num_bytes; | |
877 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
878 | ||
879 | /* record the full extent allocation */ | |
880 | ret = __btrfs_alloc_reserved_extent(trans, root, | |
881 | node->parent, ref->root, | |
882 | ref->generation, ref->owner_objectid, | |
883 | &ins, node->ref_mod); | |
884 | update_reserved_extents(root, node->bytenr, | |
885 | node->num_bytes, 0); | |
886 | } else { | |
887 | /* just add one backref */ | |
888 | ret = add_extent_ref(trans, root, node->bytenr, | |
889 | node->num_bytes, | |
890 | node->parent, ref->root, ref->generation, | |
891 | ref->owner_objectid, node->ref_mod); | |
892 | } | |
893 | BUG_ON(ret); | |
894 | } else if (ref->action == BTRFS_DROP_DELAYED_REF) { | |
895 | WARN_ON(insert_reserved); | |
896 | ret = drop_delayed_ref(trans, root, node); | |
897 | } | |
898 | return 0; | |
899 | } | |
900 | ||
901 | static noinline struct btrfs_delayed_ref_node * | |
902 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
903 | { | |
904 | struct rb_node *node; | |
905 | struct btrfs_delayed_ref_node *ref; | |
906 | int action = BTRFS_ADD_DELAYED_REF; | |
907 | again: | |
908 | /* | |
909 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
910 | * this prevents ref count from going down to zero when | |
911 | * there still are pending delayed ref. | |
912 | */ | |
913 | node = rb_prev(&head->node.rb_node); | |
914 | while (1) { | |
915 | if (!node) | |
916 | break; | |
917 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
918 | rb_node); | |
919 | if (ref->bytenr != head->node.bytenr) | |
920 | break; | |
921 | if (btrfs_delayed_node_to_ref(ref)->action == action) | |
922 | return ref; | |
923 | node = rb_prev(node); | |
924 | } | |
925 | if (action == BTRFS_ADD_DELAYED_REF) { | |
926 | action = BTRFS_DROP_DELAYED_REF; | |
927 | goto again; | |
928 | } | |
929 | return NULL; | |
930 | } | |
931 | ||
c3e69d58 CM |
932 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
933 | struct btrfs_root *root, | |
934 | struct list_head *cluster) | |
56bec294 | 935 | { |
56bec294 CM |
936 | struct btrfs_delayed_ref_root *delayed_refs; |
937 | struct btrfs_delayed_ref_node *ref; | |
938 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
939 | int ret; | |
c3e69d58 | 940 | int count = 0; |
56bec294 | 941 | int must_insert_reserved = 0; |
56bec294 CM |
942 | |
943 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
944 | while (1) { |
945 | if (!locked_ref) { | |
c3e69d58 CM |
946 | /* pick a new head ref from the cluster list */ |
947 | if (list_empty(cluster)) | |
56bec294 | 948 | break; |
56bec294 | 949 | |
c3e69d58 CM |
950 | locked_ref = list_entry(cluster->next, |
951 | struct btrfs_delayed_ref_head, cluster); | |
952 | ||
953 | /* grab the lock that says we are going to process | |
954 | * all the refs for this head */ | |
955 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
956 | ||
957 | /* | |
958 | * we may have dropped the spin lock to get the head | |
959 | * mutex lock, and that might have given someone else | |
960 | * time to free the head. If that's true, it has been | |
961 | * removed from our list and we can move on. | |
962 | */ | |
963 | if (ret == -EAGAIN) { | |
964 | locked_ref = NULL; | |
965 | count++; | |
966 | continue; | |
56bec294 CM |
967 | } |
968 | } | |
a28ec197 | 969 | |
56bec294 CM |
970 | /* |
971 | * record the must insert reserved flag before we | |
972 | * drop the spin lock. | |
973 | */ | |
974 | must_insert_reserved = locked_ref->must_insert_reserved; | |
975 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 976 | |
56bec294 CM |
977 | /* |
978 | * locked_ref is the head node, so we have to go one | |
979 | * node back for any delayed ref updates | |
980 | */ | |
56bec294 CM |
981 | ref = select_delayed_ref(locked_ref); |
982 | if (!ref) { | |
983 | /* All delayed refs have been processed, Go ahead | |
984 | * and send the head node to run_one_delayed_ref, | |
985 | * so that any accounting fixes can happen | |
986 | */ | |
987 | ref = &locked_ref->node; | |
c3e69d58 | 988 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
989 | locked_ref = NULL; |
990 | } | |
02217ed2 | 991 | |
56bec294 CM |
992 | ref->in_tree = 0; |
993 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
994 | delayed_refs->num_entries--; | |
995 | spin_unlock(&delayed_refs->lock); | |
925baedd | 996 | |
56bec294 CM |
997 | ret = run_one_delayed_ref(trans, root, ref, |
998 | must_insert_reserved); | |
999 | BUG_ON(ret); | |
1000 | btrfs_put_delayed_ref(ref); | |
eb099670 | 1001 | |
c3e69d58 CM |
1002 | count++; |
1003 | cond_resched(); | |
1004 | spin_lock(&delayed_refs->lock); | |
1005 | } | |
1006 | return count; | |
1007 | } | |
1008 | ||
1009 | /* | |
1010 | * this starts processing the delayed reference count updates and | |
1011 | * extent insertions we have queued up so far. count can be | |
1012 | * 0, which means to process everything in the tree at the start | |
1013 | * of the run (but not newly added entries), or it can be some target | |
1014 | * number you'd like to process. | |
1015 | */ | |
1016 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
1017 | struct btrfs_root *root, unsigned long count) | |
1018 | { | |
1019 | struct rb_node *node; | |
1020 | struct btrfs_delayed_ref_root *delayed_refs; | |
1021 | struct btrfs_delayed_ref_node *ref; | |
1022 | struct list_head cluster; | |
1023 | int ret; | |
1024 | int run_all = count == (unsigned long)-1; | |
1025 | int run_most = 0; | |
1026 | ||
1027 | if (root == root->fs_info->extent_root) | |
1028 | root = root->fs_info->tree_root; | |
1029 | ||
1030 | delayed_refs = &trans->transaction->delayed_refs; | |
1031 | INIT_LIST_HEAD(&cluster); | |
1032 | again: | |
1033 | spin_lock(&delayed_refs->lock); | |
1034 | if (count == 0) { | |
1035 | count = delayed_refs->num_entries * 2; | |
1036 | run_most = 1; | |
1037 | } | |
1038 | while (1) { | |
1039 | if (!(run_all || run_most) && | |
1040 | delayed_refs->num_heads_ready < 64) | |
1041 | break; | |
eb099670 | 1042 | |
56bec294 | 1043 | /* |
c3e69d58 CM |
1044 | * go find something we can process in the rbtree. We start at |
1045 | * the beginning of the tree, and then build a cluster | |
1046 | * of refs to process starting at the first one we are able to | |
1047 | * lock | |
56bec294 | 1048 | */ |
c3e69d58 CM |
1049 | ret = btrfs_find_ref_cluster(trans, &cluster, |
1050 | delayed_refs->run_delayed_start); | |
1051 | if (ret) | |
56bec294 CM |
1052 | break; |
1053 | ||
c3e69d58 CM |
1054 | ret = run_clustered_refs(trans, root, &cluster); |
1055 | BUG_ON(ret < 0); | |
1056 | ||
1057 | count -= min_t(unsigned long, ret, count); | |
1058 | ||
1059 | if (count == 0) | |
1060 | break; | |
eb099670 | 1061 | } |
c3e69d58 | 1062 | |
56bec294 | 1063 | if (run_all) { |
56bec294 | 1064 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 1065 | if (!node) |
56bec294 | 1066 | goto out; |
c3e69d58 | 1067 | count = (unsigned long)-1; |
e9d0b13b | 1068 | |
56bec294 CM |
1069 | while (node) { |
1070 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
1071 | rb_node); | |
1072 | if (btrfs_delayed_ref_is_head(ref)) { | |
1073 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 1074 | |
56bec294 CM |
1075 | head = btrfs_delayed_node_to_head(ref); |
1076 | atomic_inc(&ref->refs); | |
1077 | ||
1078 | spin_unlock(&delayed_refs->lock); | |
1079 | mutex_lock(&head->mutex); | |
1080 | mutex_unlock(&head->mutex); | |
1081 | ||
1082 | btrfs_put_delayed_ref(ref); | |
1887be66 | 1083 | cond_resched(); |
56bec294 CM |
1084 | goto again; |
1085 | } | |
1086 | node = rb_next(node); | |
1087 | } | |
1088 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
1089 | schedule_timeout(1); |
1090 | goto again; | |
5f39d397 | 1091 | } |
54aa1f4d | 1092 | out: |
c3e69d58 | 1093 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
1094 | return 0; |
1095 | } | |
1096 | ||
80ff3856 | 1097 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, |
17d217fe | 1098 | struct btrfs_root *root, u64 objectid, u64 bytenr) |
be20aa9d CM |
1099 | { |
1100 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
1101 | struct btrfs_path *path; | |
f321e491 YZ |
1102 | struct extent_buffer *leaf; |
1103 | struct btrfs_extent_ref *ref_item; | |
1104 | struct btrfs_key key; | |
1105 | struct btrfs_key found_key; | |
80ff3856 YZ |
1106 | u64 ref_root; |
1107 | u64 last_snapshot; | |
be20aa9d CM |
1108 | u32 nritems; |
1109 | int ret; | |
925baedd | 1110 | |
be20aa9d | 1111 | key.objectid = bytenr; |
31840ae1 | 1112 | key.offset = (u64)-1; |
f321e491 | 1113 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 1114 | |
f321e491 | 1115 | path = btrfs_alloc_path(); |
be20aa9d CM |
1116 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
1117 | if (ret < 0) | |
1118 | goto out; | |
1119 | BUG_ON(ret == 0); | |
80ff3856 YZ |
1120 | |
1121 | ret = -ENOENT; | |
1122 | if (path->slots[0] == 0) | |
31840ae1 | 1123 | goto out; |
be20aa9d | 1124 | |
31840ae1 | 1125 | path->slots[0]--; |
f321e491 YZ |
1126 | leaf = path->nodes[0]; |
1127 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
be20aa9d CM |
1128 | |
1129 | if (found_key.objectid != bytenr || | |
80ff3856 | 1130 | found_key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 1131 | goto out; |
f321e491 | 1132 | |
80ff3856 | 1133 | last_snapshot = btrfs_root_last_snapshot(&root->root_item); |
be20aa9d | 1134 | while (1) { |
f321e491 YZ |
1135 | leaf = path->nodes[0]; |
1136 | nritems = btrfs_header_nritems(leaf); | |
be20aa9d CM |
1137 | if (path->slots[0] >= nritems) { |
1138 | ret = btrfs_next_leaf(extent_root, path); | |
f321e491 YZ |
1139 | if (ret < 0) |
1140 | goto out; | |
be20aa9d CM |
1141 | if (ret == 0) |
1142 | continue; | |
1143 | break; | |
1144 | } | |
f321e491 | 1145 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
be20aa9d CM |
1146 | if (found_key.objectid != bytenr) |
1147 | break; | |
bd09835d | 1148 | |
be20aa9d CM |
1149 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
1150 | path->slots[0]++; | |
1151 | continue; | |
1152 | } | |
1153 | ||
f321e491 | 1154 | ref_item = btrfs_item_ptr(leaf, path->slots[0], |
be20aa9d | 1155 | struct btrfs_extent_ref); |
80ff3856 | 1156 | ref_root = btrfs_ref_root(leaf, ref_item); |
17d217fe YZ |
1157 | if ((ref_root != root->root_key.objectid && |
1158 | ref_root != BTRFS_TREE_LOG_OBJECTID) || | |
1159 | objectid != btrfs_ref_objectid(leaf, ref_item)) { | |
80ff3856 | 1160 | ret = 1; |
f321e491 | 1161 | goto out; |
80ff3856 YZ |
1162 | } |
1163 | if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) { | |
f321e491 YZ |
1164 | ret = 1; |
1165 | goto out; | |
1166 | } | |
80ff3856 YZ |
1167 | |
1168 | path->slots[0]++; | |
f321e491 YZ |
1169 | } |
1170 | ret = 0; | |
be20aa9d | 1171 | out: |
80ff3856 | 1172 | btrfs_free_path(path); |
f321e491 | 1173 | return ret; |
be20aa9d | 1174 | } |
c5739bba | 1175 | |
31840ae1 ZY |
1176 | int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1177 | struct extent_buffer *buf, u32 nr_extents) | |
02217ed2 | 1178 | { |
5f39d397 | 1179 | struct btrfs_key key; |
6407bf6d | 1180 | struct btrfs_file_extent_item *fi; |
e4657689 ZY |
1181 | u64 root_gen; |
1182 | u32 nritems; | |
02217ed2 | 1183 | int i; |
db94535d | 1184 | int level; |
31840ae1 | 1185 | int ret = 0; |
e4657689 | 1186 | int shared = 0; |
a28ec197 | 1187 | |
3768f368 | 1188 | if (!root->ref_cows) |
a28ec197 | 1189 | return 0; |
5f39d397 | 1190 | |
e4657689 ZY |
1191 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
1192 | shared = 0; | |
1193 | root_gen = root->root_key.offset; | |
1194 | } else { | |
1195 | shared = 1; | |
1196 | root_gen = trans->transid - 1; | |
1197 | } | |
1198 | ||
db94535d | 1199 | level = btrfs_header_level(buf); |
5f39d397 | 1200 | nritems = btrfs_header_nritems(buf); |
4a096752 | 1201 | |
31840ae1 | 1202 | if (level == 0) { |
31153d81 YZ |
1203 | struct btrfs_leaf_ref *ref; |
1204 | struct btrfs_extent_info *info; | |
1205 | ||
31840ae1 | 1206 | ref = btrfs_alloc_leaf_ref(root, nr_extents); |
31153d81 | 1207 | if (!ref) { |
31840ae1 | 1208 | ret = -ENOMEM; |
31153d81 YZ |
1209 | goto out; |
1210 | } | |
1211 | ||
e4657689 | 1212 | ref->root_gen = root_gen; |
31153d81 YZ |
1213 | ref->bytenr = buf->start; |
1214 | ref->owner = btrfs_header_owner(buf); | |
1215 | ref->generation = btrfs_header_generation(buf); | |
31840ae1 | 1216 | ref->nritems = nr_extents; |
31153d81 | 1217 | info = ref->extents; |
bcc63abb | 1218 | |
31840ae1 | 1219 | for (i = 0; nr_extents > 0 && i < nritems; i++) { |
31153d81 YZ |
1220 | u64 disk_bytenr; |
1221 | btrfs_item_key_to_cpu(buf, &key, i); | |
1222 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
1223 | continue; | |
1224 | fi = btrfs_item_ptr(buf, i, | |
1225 | struct btrfs_file_extent_item); | |
1226 | if (btrfs_file_extent_type(buf, fi) == | |
1227 | BTRFS_FILE_EXTENT_INLINE) | |
1228 | continue; | |
1229 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
1230 | if (disk_bytenr == 0) | |
1231 | continue; | |
1232 | ||
1233 | info->bytenr = disk_bytenr; | |
1234 | info->num_bytes = | |
1235 | btrfs_file_extent_disk_num_bytes(buf, fi); | |
1236 | info->objectid = key.objectid; | |
1237 | info->offset = key.offset; | |
1238 | info++; | |
1239 | } | |
1240 | ||
e4657689 | 1241 | ret = btrfs_add_leaf_ref(root, ref, shared); |
5b84e8d6 YZ |
1242 | if (ret == -EEXIST && shared) { |
1243 | struct btrfs_leaf_ref *old; | |
1244 | old = btrfs_lookup_leaf_ref(root, ref->bytenr); | |
1245 | BUG_ON(!old); | |
1246 | btrfs_remove_leaf_ref(root, old); | |
1247 | btrfs_free_leaf_ref(root, old); | |
1248 | ret = btrfs_add_leaf_ref(root, ref, shared); | |
1249 | } | |
31153d81 | 1250 | WARN_ON(ret); |
bcc63abb | 1251 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
1252 | } |
1253 | out: | |
31840ae1 ZY |
1254 | return ret; |
1255 | } | |
1256 | ||
b7a9f29f CM |
1257 | /* when a block goes through cow, we update the reference counts of |
1258 | * everything that block points to. The internal pointers of the block | |
1259 | * can be in just about any order, and it is likely to have clusters of | |
1260 | * things that are close together and clusters of things that are not. | |
1261 | * | |
1262 | * To help reduce the seeks that come with updating all of these reference | |
1263 | * counts, sort them by byte number before actual updates are done. | |
1264 | * | |
1265 | * struct refsort is used to match byte number to slot in the btree block. | |
1266 | * we sort based on the byte number and then use the slot to actually | |
1267 | * find the item. | |
bd56b302 CM |
1268 | * |
1269 | * struct refsort is smaller than strcut btrfs_item and smaller than | |
1270 | * struct btrfs_key_ptr. Since we're currently limited to the page size | |
1271 | * for a btree block, there's no way for a kmalloc of refsorts for a | |
1272 | * single node to be bigger than a page. | |
b7a9f29f CM |
1273 | */ |
1274 | struct refsort { | |
1275 | u64 bytenr; | |
1276 | u32 slot; | |
1277 | }; | |
1278 | ||
1279 | /* | |
1280 | * for passing into sort() | |
1281 | */ | |
1282 | static int refsort_cmp(const void *a_void, const void *b_void) | |
1283 | { | |
1284 | const struct refsort *a = a_void; | |
1285 | const struct refsort *b = b_void; | |
1286 | ||
1287 | if (a->bytenr < b->bytenr) | |
1288 | return -1; | |
1289 | if (a->bytenr > b->bytenr) | |
1290 | return 1; | |
1291 | return 0; | |
1292 | } | |
1293 | ||
1294 | ||
1295 | noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans, | |
1296 | struct btrfs_root *root, | |
1297 | struct extent_buffer *orig_buf, | |
1298 | struct extent_buffer *buf, u32 *nr_extents) | |
31840ae1 ZY |
1299 | { |
1300 | u64 bytenr; | |
1301 | u64 ref_root; | |
1302 | u64 orig_root; | |
1303 | u64 ref_generation; | |
1304 | u64 orig_generation; | |
b7a9f29f | 1305 | struct refsort *sorted; |
31840ae1 ZY |
1306 | u32 nritems; |
1307 | u32 nr_file_extents = 0; | |
1308 | struct btrfs_key key; | |
1309 | struct btrfs_file_extent_item *fi; | |
1310 | int i; | |
1311 | int level; | |
1312 | int ret = 0; | |
1313 | int faili = 0; | |
b7a9f29f CM |
1314 | int refi = 0; |
1315 | int slot; | |
31840ae1 | 1316 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
56bec294 | 1317 | u64, u64, u64, u64, u64, u64, u64, u64, u64); |
31840ae1 ZY |
1318 | |
1319 | ref_root = btrfs_header_owner(buf); | |
1320 | ref_generation = btrfs_header_generation(buf); | |
1321 | orig_root = btrfs_header_owner(orig_buf); | |
1322 | orig_generation = btrfs_header_generation(orig_buf); | |
1323 | ||
1324 | nritems = btrfs_header_nritems(buf); | |
1325 | level = btrfs_header_level(buf); | |
1326 | ||
b7a9f29f CM |
1327 | sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS); |
1328 | BUG_ON(!sorted); | |
1329 | ||
31840ae1 ZY |
1330 | if (root->ref_cows) { |
1331 | process_func = __btrfs_inc_extent_ref; | |
1332 | } else { | |
1333 | if (level == 0 && | |
1334 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
1335 | goto out; | |
1336 | if (level != 0 && | |
1337 | root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) | |
1338 | goto out; | |
1339 | process_func = __btrfs_update_extent_ref; | |
1340 | } | |
1341 | ||
b7a9f29f CM |
1342 | /* |
1343 | * we make two passes through the items. In the first pass we | |
1344 | * only record the byte number and slot. Then we sort based on | |
1345 | * byte number and do the actual work based on the sorted results | |
1346 | */ | |
31840ae1 ZY |
1347 | for (i = 0; i < nritems; i++) { |
1348 | cond_resched(); | |
db94535d | 1349 | if (level == 0) { |
5f39d397 CM |
1350 | btrfs_item_key_to_cpu(buf, &key, i); |
1351 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
54aa1f4d | 1352 | continue; |
5f39d397 | 1353 | fi = btrfs_item_ptr(buf, i, |
54aa1f4d | 1354 | struct btrfs_file_extent_item); |
5f39d397 | 1355 | if (btrfs_file_extent_type(buf, fi) == |
54aa1f4d CM |
1356 | BTRFS_FILE_EXTENT_INLINE) |
1357 | continue; | |
31840ae1 ZY |
1358 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
1359 | if (bytenr == 0) | |
54aa1f4d | 1360 | continue; |
31840ae1 ZY |
1361 | |
1362 | nr_file_extents++; | |
b7a9f29f CM |
1363 | sorted[refi].bytenr = bytenr; |
1364 | sorted[refi].slot = i; | |
1365 | refi++; | |
1366 | } else { | |
1367 | bytenr = btrfs_node_blockptr(buf, i); | |
1368 | sorted[refi].bytenr = bytenr; | |
1369 | sorted[refi].slot = i; | |
1370 | refi++; | |
1371 | } | |
1372 | } | |
1373 | /* | |
1374 | * if refi == 0, we didn't actually put anything into the sorted | |
1375 | * array and we're done | |
1376 | */ | |
1377 | if (refi == 0) | |
1378 | goto out; | |
1379 | ||
1380 | sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); | |
1381 | ||
1382 | for (i = 0; i < refi; i++) { | |
1383 | cond_resched(); | |
1384 | slot = sorted[i].slot; | |
1385 | bytenr = sorted[i].bytenr; | |
1386 | ||
1387 | if (level == 0) { | |
1388 | btrfs_item_key_to_cpu(buf, &key, slot); | |
56bec294 CM |
1389 | fi = btrfs_item_ptr(buf, slot, |
1390 | struct btrfs_file_extent_item); | |
1391 | ||
1392 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
1393 | if (bytenr == 0) | |
1394 | continue; | |
31840ae1 | 1395 | |
31840ae1 | 1396 | ret = process_func(trans, root, bytenr, |
56bec294 CM |
1397 | btrfs_file_extent_disk_num_bytes(buf, fi), |
1398 | orig_buf->start, buf->start, | |
1399 | orig_root, ref_root, | |
1400 | orig_generation, ref_generation, | |
1401 | key.objectid); | |
31840ae1 ZY |
1402 | |
1403 | if (ret) { | |
b7a9f29f | 1404 | faili = slot; |
31840ae1 ZY |
1405 | WARN_ON(1); |
1406 | goto fail; | |
1407 | } | |
54aa1f4d | 1408 | } else { |
56bec294 | 1409 | ret = process_func(trans, root, bytenr, buf->len, |
31840ae1 ZY |
1410 | orig_buf->start, buf->start, |
1411 | orig_root, ref_root, | |
1412 | orig_generation, ref_generation, | |
3bb1a1bc | 1413 | level - 1); |
31840ae1 | 1414 | if (ret) { |
b7a9f29f | 1415 | faili = slot; |
31840ae1 ZY |
1416 | WARN_ON(1); |
1417 | goto fail; | |
1418 | } | |
54aa1f4d CM |
1419 | } |
1420 | } | |
31840ae1 | 1421 | out: |
b7a9f29f | 1422 | kfree(sorted); |
31840ae1 ZY |
1423 | if (nr_extents) { |
1424 | if (level == 0) | |
1425 | *nr_extents = nr_file_extents; | |
1426 | else | |
1427 | *nr_extents = nritems; | |
1428 | } | |
1429 | return 0; | |
1430 | fail: | |
b7a9f29f | 1431 | kfree(sorted); |
31840ae1 | 1432 | WARN_ON(1); |
54aa1f4d | 1433 | return ret; |
02217ed2 CM |
1434 | } |
1435 | ||
31840ae1 ZY |
1436 | int btrfs_update_ref(struct btrfs_trans_handle *trans, |
1437 | struct btrfs_root *root, struct extent_buffer *orig_buf, | |
1438 | struct extent_buffer *buf, int start_slot, int nr) | |
1439 | ||
1440 | { | |
1441 | u64 bytenr; | |
1442 | u64 ref_root; | |
1443 | u64 orig_root; | |
1444 | u64 ref_generation; | |
1445 | u64 orig_generation; | |
1446 | struct btrfs_key key; | |
1447 | struct btrfs_file_extent_item *fi; | |
1448 | int i; | |
1449 | int ret; | |
1450 | int slot; | |
1451 | int level; | |
1452 | ||
1453 | BUG_ON(start_slot < 0); | |
1454 | BUG_ON(start_slot + nr > btrfs_header_nritems(buf)); | |
1455 | ||
1456 | ref_root = btrfs_header_owner(buf); | |
1457 | ref_generation = btrfs_header_generation(buf); | |
1458 | orig_root = btrfs_header_owner(orig_buf); | |
1459 | orig_generation = btrfs_header_generation(orig_buf); | |
1460 | level = btrfs_header_level(buf); | |
1461 | ||
1462 | if (!root->ref_cows) { | |
1463 | if (level == 0 && | |
1464 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
1465 | return 0; | |
1466 | if (level != 0 && | |
1467 | root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) | |
1468 | return 0; | |
1469 | } | |
1470 | ||
1471 | for (i = 0, slot = start_slot; i < nr; i++, slot++) { | |
1472 | cond_resched(); | |
1473 | if (level == 0) { | |
1474 | btrfs_item_key_to_cpu(buf, &key, slot); | |
1475 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
1476 | continue; | |
1477 | fi = btrfs_item_ptr(buf, slot, | |
1478 | struct btrfs_file_extent_item); | |
1479 | if (btrfs_file_extent_type(buf, fi) == | |
1480 | BTRFS_FILE_EXTENT_INLINE) | |
1481 | continue; | |
1482 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
1483 | if (bytenr == 0) | |
1484 | continue; | |
31840ae1 | 1485 | ret = __btrfs_update_extent_ref(trans, root, bytenr, |
56bec294 CM |
1486 | btrfs_file_extent_disk_num_bytes(buf, fi), |
1487 | orig_buf->start, buf->start, | |
1488 | orig_root, ref_root, orig_generation, | |
1489 | ref_generation, key.objectid); | |
31840ae1 ZY |
1490 | if (ret) |
1491 | goto fail; | |
1492 | } else { | |
1493 | bytenr = btrfs_node_blockptr(buf, slot); | |
31840ae1 | 1494 | ret = __btrfs_update_extent_ref(trans, root, bytenr, |
56bec294 CM |
1495 | buf->len, orig_buf->start, |
1496 | buf->start, orig_root, ref_root, | |
31840ae1 | 1497 | orig_generation, ref_generation, |
3bb1a1bc | 1498 | level - 1); |
31840ae1 ZY |
1499 | if (ret) |
1500 | goto fail; | |
1501 | } | |
1502 | } | |
1503 | return 0; | |
1504 | fail: | |
1505 | WARN_ON(1); | |
1506 | return -1; | |
1507 | } | |
1508 | ||
9078a3e1 CM |
1509 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
1510 | struct btrfs_root *root, | |
1511 | struct btrfs_path *path, | |
1512 | struct btrfs_block_group_cache *cache) | |
1513 | { | |
1514 | int ret; | |
9078a3e1 | 1515 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
1516 | unsigned long bi; |
1517 | struct extent_buffer *leaf; | |
9078a3e1 | 1518 | |
9078a3e1 | 1519 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
1520 | if (ret < 0) |
1521 | goto fail; | |
9078a3e1 | 1522 | BUG_ON(ret); |
5f39d397 CM |
1523 | |
1524 | leaf = path->nodes[0]; | |
1525 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1526 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
1527 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 1528 | btrfs_release_path(extent_root, path); |
54aa1f4d | 1529 | fail: |
9078a3e1 CM |
1530 | if (ret) |
1531 | return ret; | |
9078a3e1 CM |
1532 | return 0; |
1533 | ||
1534 | } | |
1535 | ||
96b5179d CM |
1536 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
1537 | struct btrfs_root *root) | |
9078a3e1 | 1538 | { |
0f9dd46c JB |
1539 | struct btrfs_block_group_cache *cache, *entry; |
1540 | struct rb_node *n; | |
9078a3e1 CM |
1541 | int err = 0; |
1542 | int werr = 0; | |
9078a3e1 | 1543 | struct btrfs_path *path; |
96b5179d | 1544 | u64 last = 0; |
9078a3e1 CM |
1545 | |
1546 | path = btrfs_alloc_path(); | |
1547 | if (!path) | |
1548 | return -ENOMEM; | |
1549 | ||
d397712b | 1550 | while (1) { |
0f9dd46c JB |
1551 | cache = NULL; |
1552 | spin_lock(&root->fs_info->block_group_cache_lock); | |
1553 | for (n = rb_first(&root->fs_info->block_group_cache_tree); | |
1554 | n; n = rb_next(n)) { | |
1555 | entry = rb_entry(n, struct btrfs_block_group_cache, | |
1556 | cache_node); | |
1557 | if (entry->dirty) { | |
1558 | cache = entry; | |
1559 | break; | |
1560 | } | |
1561 | } | |
1562 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
54aa1f4d | 1563 | |
0f9dd46c | 1564 | if (!cache) |
96b5179d | 1565 | break; |
0f9dd46c | 1566 | |
e8569813 | 1567 | cache->dirty = 0; |
0f9dd46c JB |
1568 | last += cache->key.offset; |
1569 | ||
96b5179d CM |
1570 | err = write_one_cache_group(trans, root, |
1571 | path, cache); | |
1572 | /* | |
1573 | * if we fail to write the cache group, we want | |
1574 | * to keep it marked dirty in hopes that a later | |
1575 | * write will work | |
1576 | */ | |
1577 | if (err) { | |
1578 | werr = err; | |
1579 | continue; | |
9078a3e1 CM |
1580 | } |
1581 | } | |
1582 | btrfs_free_path(path); | |
1583 | return werr; | |
1584 | } | |
1585 | ||
d2fb3437 YZ |
1586 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
1587 | { | |
1588 | struct btrfs_block_group_cache *block_group; | |
1589 | int readonly = 0; | |
1590 | ||
1591 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
1592 | if (!block_group || block_group->ro) | |
1593 | readonly = 1; | |
1594 | if (block_group) | |
1595 | put_block_group(block_group); | |
1596 | return readonly; | |
1597 | } | |
1598 | ||
593060d7 CM |
1599 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
1600 | u64 total_bytes, u64 bytes_used, | |
1601 | struct btrfs_space_info **space_info) | |
1602 | { | |
1603 | struct btrfs_space_info *found; | |
1604 | ||
1605 | found = __find_space_info(info, flags); | |
1606 | if (found) { | |
25179201 | 1607 | spin_lock(&found->lock); |
593060d7 CM |
1608 | found->total_bytes += total_bytes; |
1609 | found->bytes_used += bytes_used; | |
8f18cf13 | 1610 | found->full = 0; |
25179201 | 1611 | spin_unlock(&found->lock); |
593060d7 CM |
1612 | *space_info = found; |
1613 | return 0; | |
1614 | } | |
c146afad | 1615 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
1616 | if (!found) |
1617 | return -ENOMEM; | |
1618 | ||
0f9dd46c | 1619 | INIT_LIST_HEAD(&found->block_groups); |
80eb234a | 1620 | init_rwsem(&found->groups_sem); |
0f9dd46c | 1621 | spin_lock_init(&found->lock); |
593060d7 CM |
1622 | found->flags = flags; |
1623 | found->total_bytes = total_bytes; | |
1624 | found->bytes_used = bytes_used; | |
1625 | found->bytes_pinned = 0; | |
e8569813 | 1626 | found->bytes_reserved = 0; |
c146afad | 1627 | found->bytes_readonly = 0; |
6a63209f | 1628 | found->bytes_delalloc = 0; |
593060d7 | 1629 | found->full = 0; |
0ef3e66b | 1630 | found->force_alloc = 0; |
593060d7 | 1631 | *space_info = found; |
4184ea7f | 1632 | list_add_rcu(&found->list, &info->space_info); |
593060d7 CM |
1633 | return 0; |
1634 | } | |
1635 | ||
8790d502 CM |
1636 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
1637 | { | |
1638 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 1639 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 1640 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 1641 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
1642 | if (extra_flags) { |
1643 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1644 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1645 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1646 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1647 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1648 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1649 | } | |
1650 | } | |
593060d7 | 1651 | |
c146afad YZ |
1652 | static void set_block_group_readonly(struct btrfs_block_group_cache *cache) |
1653 | { | |
1654 | spin_lock(&cache->space_info->lock); | |
1655 | spin_lock(&cache->lock); | |
1656 | if (!cache->ro) { | |
1657 | cache->space_info->bytes_readonly += cache->key.offset - | |
1658 | btrfs_block_group_used(&cache->item); | |
1659 | cache->ro = 1; | |
1660 | } | |
1661 | spin_unlock(&cache->lock); | |
1662 | spin_unlock(&cache->space_info->lock); | |
1663 | } | |
1664 | ||
2b82032c | 1665 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 1666 | { |
2b82032c | 1667 | u64 num_devices = root->fs_info->fs_devices->rw_devices; |
a061fc8d CM |
1668 | |
1669 | if (num_devices == 1) | |
1670 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
1671 | if (num_devices < 4) | |
1672 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
1673 | ||
ec44a35c CM |
1674 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
1675 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 1676 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 1677 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 1678 | } |
ec44a35c CM |
1679 | |
1680 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 1681 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 1682 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 1683 | } |
ec44a35c CM |
1684 | |
1685 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
1686 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
1687 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
1688 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
1689 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
1690 | return flags; | |
1691 | } | |
1692 | ||
6a63209f JB |
1693 | static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data) |
1694 | { | |
1695 | struct btrfs_fs_info *info = root->fs_info; | |
1696 | u64 alloc_profile; | |
1697 | ||
1698 | if (data) { | |
1699 | alloc_profile = info->avail_data_alloc_bits & | |
1700 | info->data_alloc_profile; | |
1701 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
1702 | } else if (root == root->fs_info->chunk_root) { | |
1703 | alloc_profile = info->avail_system_alloc_bits & | |
1704 | info->system_alloc_profile; | |
1705 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
1706 | } else { | |
1707 | alloc_profile = info->avail_metadata_alloc_bits & | |
1708 | info->metadata_alloc_profile; | |
1709 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
1710 | } | |
1711 | ||
1712 | return btrfs_reduce_alloc_profile(root, data); | |
1713 | } | |
1714 | ||
1715 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) | |
1716 | { | |
1717 | u64 alloc_target; | |
1718 | ||
1719 | alloc_target = btrfs_get_alloc_profile(root, 1); | |
1720 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, | |
1721 | alloc_target); | |
1722 | } | |
1723 | ||
1724 | /* | |
1725 | * for now this just makes sure we have at least 5% of our metadata space free | |
1726 | * for use. | |
1727 | */ | |
1728 | int btrfs_check_metadata_free_space(struct btrfs_root *root) | |
1729 | { | |
1730 | struct btrfs_fs_info *info = root->fs_info; | |
1731 | struct btrfs_space_info *meta_sinfo; | |
1732 | u64 alloc_target, thresh; | |
4e06bdd6 | 1733 | int committed = 0, ret; |
6a63209f JB |
1734 | |
1735 | /* get the space info for where the metadata will live */ | |
1736 | alloc_target = btrfs_get_alloc_profile(root, 0); | |
1737 | meta_sinfo = __find_space_info(info, alloc_target); | |
1738 | ||
4e06bdd6 | 1739 | again: |
6a63209f | 1740 | spin_lock(&meta_sinfo->lock); |
4e06bdd6 JB |
1741 | if (!meta_sinfo->full) |
1742 | thresh = meta_sinfo->total_bytes * 80; | |
1743 | else | |
1744 | thresh = meta_sinfo->total_bytes * 95; | |
6a63209f JB |
1745 | |
1746 | do_div(thresh, 100); | |
1747 | ||
1748 | if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + | |
1749 | meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) { | |
4e06bdd6 JB |
1750 | struct btrfs_trans_handle *trans; |
1751 | if (!meta_sinfo->full) { | |
1752 | meta_sinfo->force_alloc = 1; | |
1753 | spin_unlock(&meta_sinfo->lock); | |
1754 | ||
1755 | trans = btrfs_start_transaction(root, 1); | |
1756 | if (!trans) | |
1757 | return -ENOMEM; | |
1758 | ||
1759 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
1760 | 2 * 1024 * 1024, alloc_target, 0); | |
1761 | btrfs_end_transaction(trans, root); | |
1762 | goto again; | |
1763 | } | |
6a63209f | 1764 | spin_unlock(&meta_sinfo->lock); |
4e06bdd6 JB |
1765 | |
1766 | if (!committed) { | |
1767 | committed = 1; | |
1768 | trans = btrfs_join_transaction(root, 1); | |
1769 | if (!trans) | |
1770 | return -ENOMEM; | |
1771 | ret = btrfs_commit_transaction(trans, root); | |
1772 | if (ret) | |
1773 | return ret; | |
1774 | goto again; | |
1775 | } | |
6a63209f JB |
1776 | return -ENOSPC; |
1777 | } | |
1778 | spin_unlock(&meta_sinfo->lock); | |
1779 | ||
1780 | return 0; | |
1781 | } | |
1782 | ||
1783 | /* | |
1784 | * This will check the space that the inode allocates from to make sure we have | |
1785 | * enough space for bytes. | |
1786 | */ | |
1787 | int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode, | |
1788 | u64 bytes) | |
1789 | { | |
1790 | struct btrfs_space_info *data_sinfo; | |
4e06bdd6 | 1791 | int ret = 0, committed = 0; |
6a63209f JB |
1792 | |
1793 | /* make sure bytes are sectorsize aligned */ | |
1794 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
1795 | ||
1796 | data_sinfo = BTRFS_I(inode)->space_info; | |
1797 | again: | |
1798 | /* make sure we have enough space to handle the data first */ | |
1799 | spin_lock(&data_sinfo->lock); | |
1800 | if (data_sinfo->total_bytes - data_sinfo->bytes_used - | |
1801 | data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved - | |
1802 | data_sinfo->bytes_pinned - data_sinfo->bytes_readonly - | |
1803 | data_sinfo->bytes_may_use < bytes) { | |
4e06bdd6 JB |
1804 | struct btrfs_trans_handle *trans; |
1805 | ||
6a63209f JB |
1806 | /* |
1807 | * if we don't have enough free bytes in this space then we need | |
1808 | * to alloc a new chunk. | |
1809 | */ | |
1810 | if (!data_sinfo->full) { | |
1811 | u64 alloc_target; | |
6a63209f JB |
1812 | |
1813 | data_sinfo->force_alloc = 1; | |
1814 | spin_unlock(&data_sinfo->lock); | |
1815 | ||
1816 | alloc_target = btrfs_get_alloc_profile(root, 1); | |
1817 | trans = btrfs_start_transaction(root, 1); | |
1818 | if (!trans) | |
1819 | return -ENOMEM; | |
1820 | ||
1821 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
1822 | bytes + 2 * 1024 * 1024, | |
1823 | alloc_target, 0); | |
1824 | btrfs_end_transaction(trans, root); | |
1825 | if (ret) | |
1826 | return ret; | |
1827 | goto again; | |
1828 | } | |
1829 | spin_unlock(&data_sinfo->lock); | |
4e06bdd6 JB |
1830 | |
1831 | /* commit the current transaction and try again */ | |
1832 | if (!committed) { | |
1833 | committed = 1; | |
1834 | trans = btrfs_join_transaction(root, 1); | |
1835 | if (!trans) | |
1836 | return -ENOMEM; | |
1837 | ret = btrfs_commit_transaction(trans, root); | |
1838 | if (ret) | |
1839 | return ret; | |
1840 | goto again; | |
1841 | } | |
1842 | ||
6a63209f JB |
1843 | printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes" |
1844 | ", %llu bytes_used, %llu bytes_reserved, " | |
1845 | "%llu bytes_pinned, %llu bytes_readonly, %llu may use" | |
1846 | "%llu total\n", bytes, data_sinfo->bytes_delalloc, | |
1847 | data_sinfo->bytes_used, data_sinfo->bytes_reserved, | |
1848 | data_sinfo->bytes_pinned, data_sinfo->bytes_readonly, | |
1849 | data_sinfo->bytes_may_use, data_sinfo->total_bytes); | |
1850 | return -ENOSPC; | |
1851 | } | |
1852 | data_sinfo->bytes_may_use += bytes; | |
1853 | BTRFS_I(inode)->reserved_bytes += bytes; | |
1854 | spin_unlock(&data_sinfo->lock); | |
1855 | ||
1856 | return btrfs_check_metadata_free_space(root); | |
1857 | } | |
1858 | ||
1859 | /* | |
1860 | * if there was an error for whatever reason after calling | |
1861 | * btrfs_check_data_free_space, call this so we can cleanup the counters. | |
1862 | */ | |
1863 | void btrfs_free_reserved_data_space(struct btrfs_root *root, | |
1864 | struct inode *inode, u64 bytes) | |
1865 | { | |
1866 | struct btrfs_space_info *data_sinfo; | |
1867 | ||
1868 | /* make sure bytes are sectorsize aligned */ | |
1869 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
1870 | ||
1871 | data_sinfo = BTRFS_I(inode)->space_info; | |
1872 | spin_lock(&data_sinfo->lock); | |
1873 | data_sinfo->bytes_may_use -= bytes; | |
1874 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
1875 | spin_unlock(&data_sinfo->lock); | |
1876 | } | |
1877 | ||
1878 | /* called when we are adding a delalloc extent to the inode's io_tree */ | |
1879 | void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode, | |
1880 | u64 bytes) | |
1881 | { | |
1882 | struct btrfs_space_info *data_sinfo; | |
1883 | ||
1884 | /* get the space info for where this inode will be storing its data */ | |
1885 | data_sinfo = BTRFS_I(inode)->space_info; | |
1886 | ||
1887 | /* make sure we have enough space to handle the data first */ | |
1888 | spin_lock(&data_sinfo->lock); | |
1889 | data_sinfo->bytes_delalloc += bytes; | |
1890 | ||
1891 | /* | |
1892 | * we are adding a delalloc extent without calling | |
1893 | * btrfs_check_data_free_space first. This happens on a weird | |
1894 | * writepage condition, but shouldn't hurt our accounting | |
1895 | */ | |
1896 | if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) { | |
1897 | data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes; | |
1898 | BTRFS_I(inode)->reserved_bytes = 0; | |
1899 | } else { | |
1900 | data_sinfo->bytes_may_use -= bytes; | |
1901 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
1902 | } | |
1903 | ||
1904 | spin_unlock(&data_sinfo->lock); | |
1905 | } | |
1906 | ||
1907 | /* called when we are clearing an delalloc extent from the inode's io_tree */ | |
1908 | void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode, | |
1909 | u64 bytes) | |
1910 | { | |
1911 | struct btrfs_space_info *info; | |
1912 | ||
1913 | info = BTRFS_I(inode)->space_info; | |
1914 | ||
1915 | spin_lock(&info->lock); | |
1916 | info->bytes_delalloc -= bytes; | |
1917 | spin_unlock(&info->lock); | |
1918 | } | |
1919 | ||
6324fbf3 CM |
1920 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
1921 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 1922 | u64 flags, int force) |
6324fbf3 CM |
1923 | { |
1924 | struct btrfs_space_info *space_info; | |
1925 | u64 thresh; | |
c146afad YZ |
1926 | int ret = 0; |
1927 | ||
1928 | mutex_lock(&extent_root->fs_info->chunk_mutex); | |
6324fbf3 | 1929 | |
2b82032c | 1930 | flags = btrfs_reduce_alloc_profile(extent_root, flags); |
ec44a35c | 1931 | |
6324fbf3 | 1932 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
1933 | if (!space_info) { |
1934 | ret = update_space_info(extent_root->fs_info, flags, | |
1935 | 0, 0, &space_info); | |
1936 | BUG_ON(ret); | |
1937 | } | |
6324fbf3 CM |
1938 | BUG_ON(!space_info); |
1939 | ||
25179201 | 1940 | spin_lock(&space_info->lock); |
0ef3e66b CM |
1941 | if (space_info->force_alloc) { |
1942 | force = 1; | |
1943 | space_info->force_alloc = 0; | |
1944 | } | |
25179201 JB |
1945 | if (space_info->full) { |
1946 | spin_unlock(&space_info->lock); | |
925baedd | 1947 | goto out; |
25179201 | 1948 | } |
6324fbf3 | 1949 | |
c146afad YZ |
1950 | thresh = space_info->total_bytes - space_info->bytes_readonly; |
1951 | thresh = div_factor(thresh, 6); | |
0ef3e66b | 1952 | if (!force && |
e8569813 | 1953 | (space_info->bytes_used + space_info->bytes_pinned + |
25179201 JB |
1954 | space_info->bytes_reserved + alloc_bytes) < thresh) { |
1955 | spin_unlock(&space_info->lock); | |
925baedd | 1956 | goto out; |
25179201 | 1957 | } |
25179201 JB |
1958 | spin_unlock(&space_info->lock); |
1959 | ||
2b82032c | 1960 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
d397712b | 1961 | if (ret) |
6324fbf3 | 1962 | space_info->full = 1; |
a74a4b97 | 1963 | out: |
c146afad | 1964 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 1965 | return ret; |
6324fbf3 CM |
1966 | } |
1967 | ||
9078a3e1 CM |
1968 | static int update_block_group(struct btrfs_trans_handle *trans, |
1969 | struct btrfs_root *root, | |
db94535d | 1970 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 1971 | int mark_free) |
9078a3e1 CM |
1972 | { |
1973 | struct btrfs_block_group_cache *cache; | |
1974 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 1975 | u64 total = num_bytes; |
9078a3e1 | 1976 | u64 old_val; |
db94535d | 1977 | u64 byte_in_group; |
3e1ad54f | 1978 | |
d397712b | 1979 | while (total) { |
db94535d | 1980 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 1981 | if (!cache) |
9078a3e1 | 1982 | return -1; |
db94535d CM |
1983 | byte_in_group = bytenr - cache->key.objectid; |
1984 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 1985 | |
25179201 | 1986 | spin_lock(&cache->space_info->lock); |
c286ac48 | 1987 | spin_lock(&cache->lock); |
0f9dd46c | 1988 | cache->dirty = 1; |
9078a3e1 | 1989 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 1990 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 1991 | if (alloc) { |
db94535d | 1992 | old_val += num_bytes; |
6324fbf3 | 1993 | cache->space_info->bytes_used += num_bytes; |
a512bbf8 | 1994 | if (cache->ro) |
c146afad | 1995 | cache->space_info->bytes_readonly -= num_bytes; |
c286ac48 CM |
1996 | btrfs_set_block_group_used(&cache->item, old_val); |
1997 | spin_unlock(&cache->lock); | |
25179201 | 1998 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 1999 | } else { |
db94535d | 2000 | old_val -= num_bytes; |
6324fbf3 | 2001 | cache->space_info->bytes_used -= num_bytes; |
c146afad YZ |
2002 | if (cache->ro) |
2003 | cache->space_info->bytes_readonly += num_bytes; | |
c286ac48 CM |
2004 | btrfs_set_block_group_used(&cache->item, old_val); |
2005 | spin_unlock(&cache->lock); | |
25179201 | 2006 | spin_unlock(&cache->space_info->lock); |
f510cfec | 2007 | if (mark_free) { |
0f9dd46c | 2008 | int ret; |
1f3c79a2 LH |
2009 | |
2010 | ret = btrfs_discard_extent(root, bytenr, | |
2011 | num_bytes); | |
2012 | WARN_ON(ret); | |
2013 | ||
0f9dd46c JB |
2014 | ret = btrfs_add_free_space(cache, bytenr, |
2015 | num_bytes); | |
d2fb3437 | 2016 | WARN_ON(ret); |
e37c9e69 | 2017 | } |
cd1bc465 | 2018 | } |
d2fb3437 | 2019 | put_block_group(cache); |
db94535d CM |
2020 | total -= num_bytes; |
2021 | bytenr += num_bytes; | |
9078a3e1 CM |
2022 | } |
2023 | return 0; | |
2024 | } | |
6324fbf3 | 2025 | |
a061fc8d CM |
2026 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
2027 | { | |
0f9dd46c | 2028 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 2029 | u64 bytenr; |
0f9dd46c JB |
2030 | |
2031 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
2032 | if (!cache) | |
a061fc8d | 2033 | return 0; |
0f9dd46c | 2034 | |
d2fb3437 YZ |
2035 | bytenr = cache->key.objectid; |
2036 | put_block_group(cache); | |
2037 | ||
2038 | return bytenr; | |
a061fc8d CM |
2039 | } |
2040 | ||
e02119d5 | 2041 | int btrfs_update_pinned_extents(struct btrfs_root *root, |
324ae4df Y |
2042 | u64 bytenr, u64 num, int pin) |
2043 | { | |
2044 | u64 len; | |
2045 | struct btrfs_block_group_cache *cache; | |
2046 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2047 | ||
2048 | if (pin) { | |
2049 | set_extent_dirty(&fs_info->pinned_extents, | |
2050 | bytenr, bytenr + num - 1, GFP_NOFS); | |
2051 | } else { | |
2052 | clear_extent_dirty(&fs_info->pinned_extents, | |
2053 | bytenr, bytenr + num - 1, GFP_NOFS); | |
2054 | } | |
b9473439 | 2055 | |
324ae4df Y |
2056 | while (num > 0) { |
2057 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
e8569813 ZY |
2058 | BUG_ON(!cache); |
2059 | len = min(num, cache->key.offset - | |
2060 | (bytenr - cache->key.objectid)); | |
324ae4df | 2061 | if (pin) { |
25179201 | 2062 | spin_lock(&cache->space_info->lock); |
e8569813 ZY |
2063 | spin_lock(&cache->lock); |
2064 | cache->pinned += len; | |
2065 | cache->space_info->bytes_pinned += len; | |
2066 | spin_unlock(&cache->lock); | |
25179201 | 2067 | spin_unlock(&cache->space_info->lock); |
324ae4df Y |
2068 | fs_info->total_pinned += len; |
2069 | } else { | |
25179201 | 2070 | spin_lock(&cache->space_info->lock); |
e8569813 ZY |
2071 | spin_lock(&cache->lock); |
2072 | cache->pinned -= len; | |
2073 | cache->space_info->bytes_pinned -= len; | |
2074 | spin_unlock(&cache->lock); | |
25179201 | 2075 | spin_unlock(&cache->space_info->lock); |
324ae4df | 2076 | fs_info->total_pinned -= len; |
07103a3c JB |
2077 | if (cache->cached) |
2078 | btrfs_add_free_space(cache, bytenr, len); | |
324ae4df | 2079 | } |
d2fb3437 | 2080 | put_block_group(cache); |
324ae4df Y |
2081 | bytenr += len; |
2082 | num -= len; | |
2083 | } | |
2084 | return 0; | |
2085 | } | |
9078a3e1 | 2086 | |
e8569813 ZY |
2087 | static int update_reserved_extents(struct btrfs_root *root, |
2088 | u64 bytenr, u64 num, int reserve) | |
2089 | { | |
2090 | u64 len; | |
2091 | struct btrfs_block_group_cache *cache; | |
2092 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2093 | ||
e8569813 ZY |
2094 | while (num > 0) { |
2095 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
2096 | BUG_ON(!cache); | |
2097 | len = min(num, cache->key.offset - | |
2098 | (bytenr - cache->key.objectid)); | |
25179201 JB |
2099 | |
2100 | spin_lock(&cache->space_info->lock); | |
2101 | spin_lock(&cache->lock); | |
e8569813 | 2102 | if (reserve) { |
e8569813 ZY |
2103 | cache->reserved += len; |
2104 | cache->space_info->bytes_reserved += len; | |
e8569813 | 2105 | } else { |
e8569813 ZY |
2106 | cache->reserved -= len; |
2107 | cache->space_info->bytes_reserved -= len; | |
e8569813 | 2108 | } |
25179201 JB |
2109 | spin_unlock(&cache->lock); |
2110 | spin_unlock(&cache->space_info->lock); | |
d2fb3437 | 2111 | put_block_group(cache); |
e8569813 ZY |
2112 | bytenr += len; |
2113 | num -= len; | |
2114 | } | |
2115 | return 0; | |
2116 | } | |
2117 | ||
d1310b2e | 2118 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) |
ccd467d6 | 2119 | { |
ccd467d6 | 2120 | u64 last = 0; |
1a5bc167 CM |
2121 | u64 start; |
2122 | u64 end; | |
d1310b2e | 2123 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; |
ccd467d6 | 2124 | int ret; |
ccd467d6 | 2125 | |
d397712b | 2126 | while (1) { |
1a5bc167 CM |
2127 | ret = find_first_extent_bit(pinned_extents, last, |
2128 | &start, &end, EXTENT_DIRTY); | |
2129 | if (ret) | |
ccd467d6 | 2130 | break; |
1a5bc167 CM |
2131 | set_extent_dirty(copy, start, end, GFP_NOFS); |
2132 | last = end + 1; | |
ccd467d6 CM |
2133 | } |
2134 | return 0; | |
2135 | } | |
2136 | ||
2137 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
2138 | struct btrfs_root *root, | |
d1310b2e | 2139 | struct extent_io_tree *unpin) |
a28ec197 | 2140 | { |
1a5bc167 CM |
2141 | u64 start; |
2142 | u64 end; | |
a28ec197 | 2143 | int ret; |
a28ec197 | 2144 | |
d397712b | 2145 | while (1) { |
1a5bc167 CM |
2146 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
2147 | EXTENT_DIRTY); | |
2148 | if (ret) | |
a28ec197 | 2149 | break; |
1f3c79a2 LH |
2150 | |
2151 | ret = btrfs_discard_extent(root, start, end + 1 - start); | |
2152 | ||
b9473439 | 2153 | /* unlocks the pinned mutex */ |
e02119d5 | 2154 | btrfs_update_pinned_extents(root, start, end + 1 - start, 0); |
1a5bc167 | 2155 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
1f3c79a2 | 2156 | |
b9473439 | 2157 | cond_resched(); |
a28ec197 | 2158 | } |
1f3c79a2 | 2159 | return ret; |
a28ec197 CM |
2160 | } |
2161 | ||
31840ae1 ZY |
2162 | static int pin_down_bytes(struct btrfs_trans_handle *trans, |
2163 | struct btrfs_root *root, | |
b9473439 CM |
2164 | struct btrfs_path *path, |
2165 | u64 bytenr, u64 num_bytes, int is_data, | |
2166 | struct extent_buffer **must_clean) | |
e20d96d6 | 2167 | { |
1a5bc167 | 2168 | int err = 0; |
31840ae1 | 2169 | struct extent_buffer *buf; |
8ef97622 | 2170 | |
31840ae1 ZY |
2171 | if (is_data) |
2172 | goto pinit; | |
2173 | ||
2174 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); | |
2175 | if (!buf) | |
2176 | goto pinit; | |
2177 | ||
2178 | /* we can reuse a block if it hasn't been written | |
2179 | * and it is from this transaction. We can't | |
2180 | * reuse anything from the tree log root because | |
2181 | * it has tiny sub-transactions. | |
2182 | */ | |
2183 | if (btrfs_buffer_uptodate(buf, 0) && | |
2184 | btrfs_try_tree_lock(buf)) { | |
2185 | u64 header_owner = btrfs_header_owner(buf); | |
2186 | u64 header_transid = btrfs_header_generation(buf); | |
2187 | if (header_owner != BTRFS_TREE_LOG_OBJECTID && | |
1a40e23b | 2188 | header_owner != BTRFS_TREE_RELOC_OBJECTID && |
56bec294 | 2189 | header_owner != BTRFS_DATA_RELOC_TREE_OBJECTID && |
31840ae1 ZY |
2190 | header_transid == trans->transid && |
2191 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
b9473439 | 2192 | *must_clean = buf; |
31840ae1 | 2193 | return 1; |
8ef97622 | 2194 | } |
31840ae1 | 2195 | btrfs_tree_unlock(buf); |
f4b9aa8d | 2196 | } |
31840ae1 ZY |
2197 | free_extent_buffer(buf); |
2198 | pinit: | |
b9473439 | 2199 | btrfs_set_path_blocking(path); |
b9473439 | 2200 | /* unlocks the pinned mutex */ |
31840ae1 ZY |
2201 | btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); |
2202 | ||
be744175 | 2203 | BUG_ON(err < 0); |
e20d96d6 CM |
2204 | return 0; |
2205 | } | |
2206 | ||
fec577fb | 2207 | /* |
a28ec197 | 2208 | * remove an extent from the root, returns 0 on success |
fec577fb | 2209 | */ |
31840ae1 ZY |
2210 | static int __free_extent(struct btrfs_trans_handle *trans, |
2211 | struct btrfs_root *root, | |
2212 | u64 bytenr, u64 num_bytes, u64 parent, | |
7bb86316 | 2213 | u64 root_objectid, u64 ref_generation, |
56bec294 CM |
2214 | u64 owner_objectid, int pin, int mark_free, |
2215 | int refs_to_drop) | |
a28ec197 | 2216 | { |
5caf2a00 | 2217 | struct btrfs_path *path; |
e2fa7227 | 2218 | struct btrfs_key key; |
1261ec42 CM |
2219 | struct btrfs_fs_info *info = root->fs_info; |
2220 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 2221 | struct extent_buffer *leaf; |
a28ec197 | 2222 | int ret; |
952fccac CM |
2223 | int extent_slot = 0; |
2224 | int found_extent = 0; | |
2225 | int num_to_del = 1; | |
234b63a0 | 2226 | struct btrfs_extent_item *ei; |
cf27e1ee | 2227 | u32 refs; |
037e6390 | 2228 | |
db94535d | 2229 | key.objectid = bytenr; |
62e2749e | 2230 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 2231 | key.offset = num_bytes; |
5caf2a00 | 2232 | path = btrfs_alloc_path(); |
54aa1f4d CM |
2233 | if (!path) |
2234 | return -ENOMEM; | |
5f26f772 | 2235 | |
3c12ac72 | 2236 | path->reada = 1; |
b9473439 | 2237 | path->leave_spinning = 1; |
3bb1a1bc YZ |
2238 | ret = lookup_extent_backref(trans, extent_root, path, |
2239 | bytenr, parent, root_objectid, | |
2240 | ref_generation, owner_objectid, 1); | |
7bb86316 | 2241 | if (ret == 0) { |
952fccac CM |
2242 | struct btrfs_key found_key; |
2243 | extent_slot = path->slots[0]; | |
d397712b | 2244 | while (extent_slot > 0) { |
952fccac CM |
2245 | extent_slot--; |
2246 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2247 | extent_slot); | |
2248 | if (found_key.objectid != bytenr) | |
2249 | break; | |
2250 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && | |
2251 | found_key.offset == num_bytes) { | |
2252 | found_extent = 1; | |
2253 | break; | |
2254 | } | |
2255 | if (path->slots[0] - extent_slot > 5) | |
2256 | break; | |
2257 | } | |
31840ae1 | 2258 | if (!found_extent) { |
56bec294 CM |
2259 | ret = remove_extent_backref(trans, extent_root, path, |
2260 | refs_to_drop); | |
31840ae1 ZY |
2261 | BUG_ON(ret); |
2262 | btrfs_release_path(extent_root, path); | |
b9473439 | 2263 | path->leave_spinning = 1; |
31840ae1 ZY |
2264 | ret = btrfs_search_slot(trans, extent_root, |
2265 | &key, path, -1, 1); | |
f3465ca4 JB |
2266 | if (ret) { |
2267 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
2268 | ", was looking for %llu\n", ret, |
2269 | (unsigned long long)bytenr); | |
f3465ca4 JB |
2270 | btrfs_print_leaf(extent_root, path->nodes[0]); |
2271 | } | |
31840ae1 ZY |
2272 | BUG_ON(ret); |
2273 | extent_slot = path->slots[0]; | |
2274 | } | |
7bb86316 CM |
2275 | } else { |
2276 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
2277 | WARN_ON(1); | |
d397712b | 2278 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
56bec294 | 2279 | "parent %llu root %llu gen %llu owner %llu\n", |
d397712b | 2280 | (unsigned long long)bytenr, |
56bec294 | 2281 | (unsigned long long)parent, |
d397712b CM |
2282 | (unsigned long long)root_objectid, |
2283 | (unsigned long long)ref_generation, | |
2284 | (unsigned long long)owner_objectid); | |
7bb86316 | 2285 | } |
5f39d397 CM |
2286 | |
2287 | leaf = path->nodes[0]; | |
952fccac | 2288 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 2289 | struct btrfs_extent_item); |
5f39d397 | 2290 | refs = btrfs_extent_refs(leaf, ei); |
952fccac | 2291 | |
56bec294 CM |
2292 | /* |
2293 | * we're not allowed to delete the extent item if there | |
2294 | * are other delayed ref updates pending | |
2295 | */ | |
2296 | ||
2297 | BUG_ON(refs < refs_to_drop); | |
2298 | refs -= refs_to_drop; | |
2299 | btrfs_set_extent_refs(leaf, ei, refs); | |
5f39d397 CM |
2300 | btrfs_mark_buffer_dirty(leaf); |
2301 | ||
56bec294 CM |
2302 | if (refs == 0 && found_extent && |
2303 | path->slots[0] == extent_slot + 1) { | |
31840ae1 ZY |
2304 | struct btrfs_extent_ref *ref; |
2305 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
2306 | struct btrfs_extent_ref); | |
56bec294 | 2307 | BUG_ON(btrfs_ref_num_refs(leaf, ref) != refs_to_drop); |
952fccac CM |
2308 | /* if the back ref and the extent are next to each other |
2309 | * they get deleted below in one shot | |
2310 | */ | |
2311 | path->slots[0] = extent_slot; | |
2312 | num_to_del = 2; | |
2313 | } else if (found_extent) { | |
2314 | /* otherwise delete the extent back ref */ | |
56bec294 CM |
2315 | ret = remove_extent_backref(trans, extent_root, path, |
2316 | refs_to_drop); | |
952fccac CM |
2317 | BUG_ON(ret); |
2318 | /* if refs are 0, we need to setup the path for deletion */ | |
2319 | if (refs == 0) { | |
2320 | btrfs_release_path(extent_root, path); | |
b9473439 | 2321 | path->leave_spinning = 1; |
952fccac CM |
2322 | ret = btrfs_search_slot(trans, extent_root, &key, path, |
2323 | -1, 1); | |
952fccac CM |
2324 | BUG_ON(ret); |
2325 | } | |
2326 | } | |
2327 | ||
cf27e1ee | 2328 | if (refs == 0) { |
db94535d CM |
2329 | u64 super_used; |
2330 | u64 root_used; | |
b9473439 | 2331 | struct extent_buffer *must_clean = NULL; |
78fae27e CM |
2332 | |
2333 | if (pin) { | |
b9473439 CM |
2334 | ret = pin_down_bytes(trans, root, path, |
2335 | bytenr, num_bytes, | |
2336 | owner_objectid >= BTRFS_FIRST_FREE_OBJECTID, | |
2337 | &must_clean); | |
c549228f Y |
2338 | if (ret > 0) |
2339 | mark_free = 1; | |
2340 | BUG_ON(ret < 0); | |
78fae27e | 2341 | } |
b9473439 | 2342 | |
58176a96 | 2343 | /* block accounting for super block */ |
75eff68e | 2344 | spin_lock(&info->delalloc_lock); |
db94535d CM |
2345 | super_used = btrfs_super_bytes_used(&info->super_copy); |
2346 | btrfs_set_super_bytes_used(&info->super_copy, | |
2347 | super_used - num_bytes); | |
58176a96 JB |
2348 | |
2349 | /* block accounting for root item */ | |
db94535d | 2350 | root_used = btrfs_root_used(&root->root_item); |
5f39d397 | 2351 | btrfs_set_root_used(&root->root_item, |
db94535d | 2352 | root_used - num_bytes); |
34bf63c4 | 2353 | spin_unlock(&info->delalloc_lock); |
b9473439 CM |
2354 | |
2355 | /* | |
2356 | * it is going to be very rare for someone to be waiting | |
2357 | * on the block we're freeing. del_items might need to | |
2358 | * schedule, so rather than get fancy, just force it | |
2359 | * to blocking here | |
2360 | */ | |
2361 | if (must_clean) | |
2362 | btrfs_set_lock_blocking(must_clean); | |
2363 | ||
952fccac CM |
2364 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
2365 | num_to_del); | |
31840ae1 | 2366 | BUG_ON(ret); |
25179201 | 2367 | btrfs_release_path(extent_root, path); |
21af804c | 2368 | |
b9473439 CM |
2369 | if (must_clean) { |
2370 | clean_tree_block(NULL, root, must_clean); | |
2371 | btrfs_tree_unlock(must_clean); | |
2372 | free_extent_buffer(must_clean); | |
2373 | } | |
2374 | ||
459931ec CM |
2375 | if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { |
2376 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); | |
2377 | BUG_ON(ret); | |
d57e62b8 CM |
2378 | } else { |
2379 | invalidate_mapping_pages(info->btree_inode->i_mapping, | |
2380 | bytenr >> PAGE_CACHE_SHIFT, | |
2381 | (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT); | |
459931ec CM |
2382 | } |
2383 | ||
dcbdd4dc CM |
2384 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
2385 | mark_free); | |
2386 | BUG_ON(ret); | |
a28ec197 | 2387 | } |
5caf2a00 | 2388 | btrfs_free_path(path); |
a28ec197 CM |
2389 | return ret; |
2390 | } | |
2391 | ||
fec577fb CM |
2392 | /* |
2393 | * remove an extent from the root, returns 0 on success | |
2394 | */ | |
925baedd | 2395 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
56bec294 CM |
2396 | struct btrfs_root *root, |
2397 | u64 bytenr, u64 num_bytes, u64 parent, | |
2398 | u64 root_objectid, u64 ref_generation, | |
2399 | u64 owner_objectid, int pin, | |
2400 | int refs_to_drop) | |
fec577fb | 2401 | { |
db94535d | 2402 | WARN_ON(num_bytes < root->sectorsize); |
31840ae1 | 2403 | |
56bec294 CM |
2404 | /* |
2405 | * if metadata always pin | |
2406 | * if data pin when any transaction has committed this | |
2407 | */ | |
2408 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID || | |
2409 | ref_generation != trans->transid) | |
4bef0848 CM |
2410 | pin = 1; |
2411 | ||
4bef0848 CM |
2412 | if (ref_generation != trans->transid) |
2413 | pin = 1; | |
2414 | ||
56bec294 | 2415 | return __free_extent(trans, root, bytenr, num_bytes, parent, |
3bb1a1bc | 2416 | root_objectid, ref_generation, |
56bec294 | 2417 | owner_objectid, pin, pin == 0, refs_to_drop); |
fec577fb CM |
2418 | } |
2419 | ||
1887be66 CM |
2420 | /* |
2421 | * when we free an extent, it is possible (and likely) that we free the last | |
2422 | * delayed ref for that extent as well. This searches the delayed ref tree for | |
2423 | * a given extent, and if there are no other delayed refs to be processed, it | |
2424 | * removes it from the tree. | |
2425 | */ | |
2426 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2427 | struct btrfs_root *root, u64 bytenr) | |
2428 | { | |
2429 | struct btrfs_delayed_ref_head *head; | |
2430 | struct btrfs_delayed_ref_root *delayed_refs; | |
2431 | struct btrfs_delayed_ref_node *ref; | |
2432 | struct rb_node *node; | |
2433 | int ret; | |
2434 | ||
2435 | delayed_refs = &trans->transaction->delayed_refs; | |
2436 | spin_lock(&delayed_refs->lock); | |
2437 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2438 | if (!head) | |
2439 | goto out; | |
2440 | ||
2441 | node = rb_prev(&head->node.rb_node); | |
2442 | if (!node) | |
2443 | goto out; | |
2444 | ||
2445 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2446 | ||
2447 | /* there are still entries for this ref, we can't drop it */ | |
2448 | if (ref->bytenr == bytenr) | |
2449 | goto out; | |
2450 | ||
2451 | /* | |
2452 | * waiting for the lock here would deadlock. If someone else has it | |
2453 | * locked they are already in the process of dropping it anyway | |
2454 | */ | |
2455 | if (!mutex_trylock(&head->mutex)) | |
2456 | goto out; | |
2457 | ||
2458 | /* | |
2459 | * at this point we have a head with no other entries. Go | |
2460 | * ahead and process it. | |
2461 | */ | |
2462 | head->node.in_tree = 0; | |
2463 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 2464 | |
1887be66 CM |
2465 | delayed_refs->num_entries--; |
2466 | ||
2467 | /* | |
2468 | * we don't take a ref on the node because we're removing it from the | |
2469 | * tree, so we just steal the ref the tree was holding. | |
2470 | */ | |
c3e69d58 CM |
2471 | delayed_refs->num_heads--; |
2472 | if (list_empty(&head->cluster)) | |
2473 | delayed_refs->num_heads_ready--; | |
2474 | ||
2475 | list_del_init(&head->cluster); | |
1887be66 CM |
2476 | spin_unlock(&delayed_refs->lock); |
2477 | ||
2478 | ret = run_one_delayed_ref(trans, root->fs_info->tree_root, | |
2479 | &head->node, head->must_insert_reserved); | |
2480 | BUG_ON(ret); | |
2481 | btrfs_put_delayed_ref(&head->node); | |
2482 | return 0; | |
2483 | out: | |
2484 | spin_unlock(&delayed_refs->lock); | |
2485 | return 0; | |
2486 | } | |
2487 | ||
925baedd | 2488 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
2489 | struct btrfs_root *root, |
2490 | u64 bytenr, u64 num_bytes, u64 parent, | |
2491 | u64 root_objectid, u64 ref_generation, | |
3bb1a1bc | 2492 | u64 owner_objectid, int pin) |
925baedd CM |
2493 | { |
2494 | int ret; | |
2495 | ||
56bec294 CM |
2496 | /* |
2497 | * tree log blocks never actually go into the extent allocation | |
2498 | * tree, just update pinning info and exit early. | |
2499 | * | |
2500 | * data extents referenced by the tree log do need to have | |
2501 | * their reference counts bumped. | |
2502 | */ | |
2503 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID && | |
2504 | owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
b9473439 | 2505 | /* unlocks the pinned mutex */ |
56bec294 | 2506 | btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); |
56bec294 CM |
2507 | update_reserved_extents(root, bytenr, num_bytes, 0); |
2508 | ret = 0; | |
2509 | } else { | |
2510 | ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent, | |
2511 | root_objectid, ref_generation, | |
2512 | owner_objectid, | |
2513 | BTRFS_DROP_DELAYED_REF, 1); | |
1887be66 CM |
2514 | BUG_ON(ret); |
2515 | ret = check_ref_cleanup(trans, root, bytenr); | |
2516 | BUG_ON(ret); | |
56bec294 | 2517 | } |
925baedd CM |
2518 | return ret; |
2519 | } | |
2520 | ||
87ee04eb CM |
2521 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
2522 | { | |
2523 | u64 mask = ((u64)root->stripesize - 1); | |
2524 | u64 ret = (val + mask) & ~mask; | |
2525 | return ret; | |
2526 | } | |
2527 | ||
fec577fb CM |
2528 | /* |
2529 | * walks the btree of allocated extents and find a hole of a given size. | |
2530 | * The key ins is changed to record the hole: | |
2531 | * ins->objectid == block start | |
62e2749e | 2532 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
2533 | * ins->offset == number of blocks |
2534 | * Any available blocks before search_start are skipped. | |
2535 | */ | |
d397712b | 2536 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2537 | struct btrfs_root *orig_root, |
2538 | u64 num_bytes, u64 empty_size, | |
2539 | u64 search_start, u64 search_end, | |
2540 | u64 hint_byte, struct btrfs_key *ins, | |
2541 | u64 exclude_start, u64 exclude_nr, | |
2542 | int data) | |
fec577fb | 2543 | { |
80eb234a | 2544 | int ret = 0; |
d397712b | 2545 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
239b14b3 | 2546 | u64 *last_ptr = NULL; |
80eb234a | 2547 | struct btrfs_block_group_cache *block_group = NULL; |
239b14b3 | 2548 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 2549 | int allowed_chunk_alloc = 0; |
2552d17e | 2550 | int using_hint = 0; |
80eb234a | 2551 | struct btrfs_space_info *space_info; |
fec577fb | 2552 | |
db94535d | 2553 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 2554 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
2555 | ins->objectid = 0; |
2556 | ins->offset = 0; | |
b1a4d965 | 2557 | |
2552d17e JB |
2558 | space_info = __find_space_info(root->fs_info, data); |
2559 | ||
0ef3e66b CM |
2560 | if (orig_root->ref_cows || empty_size) |
2561 | allowed_chunk_alloc = 1; | |
2562 | ||
239b14b3 CM |
2563 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
2564 | last_ptr = &root->fs_info->last_alloc; | |
536ac8ae CM |
2565 | if (!btrfs_test_opt(root, SSD)) |
2566 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
2567 | } |
2568 | ||
0f9dd46c | 2569 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) |
239b14b3 | 2570 | last_ptr = &root->fs_info->last_data_alloc; |
0f9dd46c | 2571 | |
239b14b3 | 2572 | if (last_ptr) { |
2552d17e | 2573 | if (*last_ptr) |
239b14b3 | 2574 | hint_byte = *last_ptr; |
2552d17e | 2575 | else |
239b14b3 | 2576 | empty_size += empty_cluster; |
4366211c CM |
2577 | } else { |
2578 | empty_cluster = 0; | |
239b14b3 | 2579 | } |
a061fc8d | 2580 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 2581 | search_start = max(search_start, hint_byte); |
0b86a832 | 2582 | |
2552d17e JB |
2583 | if (search_start == hint_byte) { |
2584 | using_hint = 1; | |
2585 | block_group = btrfs_lookup_block_group(root->fs_info, | |
2586 | search_start); | |
2587 | if (block_group && block_group_bits(block_group, data)) { | |
2552d17e JB |
2588 | down_read(&space_info->groups_sem); |
2589 | goto have_block_group; | |
2590 | } else if (block_group) { | |
2591 | put_block_group(block_group); | |
2592 | } | |
2593 | ||
42e70e7a | 2594 | empty_size += empty_cluster; |
2552d17e | 2595 | using_hint = 0; |
42e70e7a | 2596 | } |
4366211c | 2597 | |
2552d17e | 2598 | search: |
80eb234a | 2599 | down_read(&space_info->groups_sem); |
2552d17e | 2600 | list_for_each_entry(block_group, &space_info->block_groups, list) { |
6226cb0a | 2601 | u64 offset; |
8a1413a2 | 2602 | |
2552d17e JB |
2603 | atomic_inc(&block_group->count); |
2604 | search_start = block_group->key.objectid; | |
42e70e7a | 2605 | |
2552d17e | 2606 | have_block_group: |
ea6a478e JB |
2607 | if (unlikely(!block_group->cached)) { |
2608 | mutex_lock(&block_group->cache_mutex); | |
2609 | ret = cache_block_group(root, block_group); | |
2610 | mutex_unlock(&block_group->cache_mutex); | |
2552d17e JB |
2611 | if (ret) { |
2612 | put_block_group(block_group); | |
ea6a478e | 2613 | break; |
2552d17e | 2614 | } |
ea6a478e JB |
2615 | } |
2616 | ||
ea6a478e | 2617 | if (unlikely(block_group->ro)) |
2552d17e | 2618 | goto loop; |
0f9dd46c | 2619 | |
6226cb0a JB |
2620 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
2621 | num_bytes, empty_size); | |
2622 | if (!offset) | |
2552d17e | 2623 | goto loop; |
3b7885bf | 2624 | |
6226cb0a | 2625 | search_start = stripe_align(root, offset); |
80eb234a | 2626 | |
2552d17e | 2627 | /* move on to the next group */ |
6226cb0a JB |
2628 | if (search_start + num_bytes >= search_end) { |
2629 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 2630 | goto loop; |
6226cb0a | 2631 | } |
25179201 | 2632 | |
2552d17e JB |
2633 | /* move on to the next group */ |
2634 | if (search_start + num_bytes > | |
6226cb0a JB |
2635 | block_group->key.objectid + block_group->key.offset) { |
2636 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 2637 | goto loop; |
6226cb0a | 2638 | } |
f5a31e16 | 2639 | |
6226cb0a JB |
2640 | if (using_hint && search_start > hint_byte) { |
2641 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 2642 | goto loop; |
6226cb0a | 2643 | } |
2552d17e JB |
2644 | |
2645 | if (exclude_nr > 0 && | |
2646 | (search_start + num_bytes > exclude_start && | |
2647 | search_start < exclude_start + exclude_nr)) { | |
2648 | search_start = exclude_start + exclude_nr; | |
2649 | ||
6226cb0a | 2650 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e JB |
2651 | /* |
2652 | * if search_start is still in this block group | |
2653 | * then we just re-search this block group | |
f5a31e16 | 2654 | */ |
2552d17e JB |
2655 | if (search_start >= block_group->key.objectid && |
2656 | search_start < (block_group->key.objectid + | |
6226cb0a | 2657 | block_group->key.offset)) |
2552d17e | 2658 | goto have_block_group; |
2552d17e | 2659 | goto loop; |
0f9dd46c | 2660 | } |
0b86a832 | 2661 | |
2552d17e JB |
2662 | ins->objectid = search_start; |
2663 | ins->offset = num_bytes; | |
d2fb3437 | 2664 | |
6226cb0a JB |
2665 | if (offset < search_start) |
2666 | btrfs_add_free_space(block_group, offset, | |
2667 | search_start - offset); | |
2668 | BUG_ON(offset > search_start); | |
2669 | ||
2552d17e | 2670 | /* we are all good, lets return */ |
2552d17e JB |
2671 | break; |
2672 | loop: | |
2552d17e JB |
2673 | put_block_group(block_group); |
2674 | if (using_hint) { | |
2675 | empty_size += empty_cluster; | |
2552d17e JB |
2676 | using_hint = 0; |
2677 | up_read(&space_info->groups_sem); | |
2678 | goto search; | |
2679 | } | |
2680 | } | |
2681 | up_read(&space_info->groups_sem); | |
2682 | ||
2683 | if (!ins->objectid && (empty_size || allowed_chunk_alloc)) { | |
2684 | int try_again = empty_size; | |
2685 | ||
2552d17e JB |
2686 | empty_size = 0; |
2687 | ||
2688 | if (allowed_chunk_alloc) { | |
2689 | ret = do_chunk_alloc(trans, root, num_bytes + | |
2690 | 2 * 1024 * 1024, data, 1); | |
2691 | if (!ret) | |
2692 | try_again = 1; | |
2693 | allowed_chunk_alloc = 0; | |
2694 | } else { | |
2695 | space_info->force_alloc = 1; | |
2696 | } | |
2697 | ||
2698 | if (try_again) | |
2699 | goto search; | |
2700 | ret = -ENOSPC; | |
2701 | } else if (!ins->objectid) { | |
2702 | ret = -ENOSPC; | |
f2654de4 | 2703 | } |
0b86a832 | 2704 | |
80eb234a JB |
2705 | /* we found what we needed */ |
2706 | if (ins->objectid) { | |
2707 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) | |
d2fb3437 | 2708 | trans->block_group = block_group->key.objectid; |
0f9dd46c | 2709 | |
80eb234a JB |
2710 | if (last_ptr) |
2711 | *last_ptr = ins->objectid + ins->offset; | |
2552d17e | 2712 | put_block_group(block_group); |
80eb234a | 2713 | ret = 0; |
be744175 | 2714 | } |
be744175 | 2715 | |
0f70abe2 | 2716 | return ret; |
fec577fb | 2717 | } |
ec44a35c | 2718 | |
0f9dd46c JB |
2719 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes) |
2720 | { | |
2721 | struct btrfs_block_group_cache *cache; | |
0f9dd46c | 2722 | |
d397712b CM |
2723 | printk(KERN_INFO "space_info has %llu free, is %sfull\n", |
2724 | (unsigned long long)(info->total_bytes - info->bytes_used - | |
2725 | info->bytes_pinned - info->bytes_reserved), | |
2726 | (info->full) ? "" : "not "); | |
6a63209f JB |
2727 | printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu," |
2728 | " may_use=%llu, used=%llu\n", info->total_bytes, | |
2729 | info->bytes_pinned, info->bytes_delalloc, info->bytes_may_use, | |
2730 | info->bytes_used); | |
0f9dd46c | 2731 | |
80eb234a | 2732 | down_read(&info->groups_sem); |
c6e30871 | 2733 | list_for_each_entry(cache, &info->block_groups, list) { |
0f9dd46c | 2734 | spin_lock(&cache->lock); |
d397712b CM |
2735 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
2736 | "%llu pinned %llu reserved\n", | |
2737 | (unsigned long long)cache->key.objectid, | |
2738 | (unsigned long long)cache->key.offset, | |
2739 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
2740 | (unsigned long long)cache->pinned, | |
2741 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
2742 | btrfs_dump_free_space(cache, bytes); |
2743 | spin_unlock(&cache->lock); | |
2744 | } | |
80eb234a | 2745 | up_read(&info->groups_sem); |
0f9dd46c | 2746 | } |
e8569813 | 2747 | |
e6dcd2dc CM |
2748 | static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
2749 | struct btrfs_root *root, | |
2750 | u64 num_bytes, u64 min_alloc_size, | |
2751 | u64 empty_size, u64 hint_byte, | |
2752 | u64 search_end, struct btrfs_key *ins, | |
2753 | u64 data) | |
fec577fb CM |
2754 | { |
2755 | int ret; | |
fbdc762b | 2756 | u64 search_start = 0; |
1261ec42 | 2757 | struct btrfs_fs_info *info = root->fs_info; |
925baedd | 2758 | |
6a63209f | 2759 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 2760 | again: |
0ef3e66b CM |
2761 | /* |
2762 | * the only place that sets empty_size is btrfs_realloc_node, which | |
2763 | * is not called recursively on allocations | |
2764 | */ | |
2765 | if (empty_size || root->ref_cows) { | |
593060d7 | 2766 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 2767 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b CM |
2768 | 2 * 1024 * 1024, |
2769 | BTRFS_BLOCK_GROUP_METADATA | | |
2770 | (info->metadata_alloc_profile & | |
2771 | info->avail_metadata_alloc_bits), 0); | |
6324fbf3 CM |
2772 | } |
2773 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
0ef3e66b | 2774 | num_bytes + 2 * 1024 * 1024, data, 0); |
6324fbf3 | 2775 | } |
0b86a832 | 2776 | |
db94535d CM |
2777 | WARN_ON(num_bytes < root->sectorsize); |
2778 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
2779 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
2780 | trans->alloc_exclude_start, |
2781 | trans->alloc_exclude_nr, data); | |
3b951516 | 2782 | |
98d20f67 CM |
2783 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
2784 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 2785 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 2786 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b CM |
2787 | do_chunk_alloc(trans, root->fs_info->extent_root, |
2788 | num_bytes, data, 1); | |
98d20f67 CM |
2789 | goto again; |
2790 | } | |
ec44a35c | 2791 | if (ret) { |
0f9dd46c JB |
2792 | struct btrfs_space_info *sinfo; |
2793 | ||
2794 | sinfo = __find_space_info(root->fs_info, data); | |
d397712b CM |
2795 | printk(KERN_ERR "btrfs allocation failed flags %llu, " |
2796 | "wanted %llu\n", (unsigned long long)data, | |
2797 | (unsigned long long)num_bytes); | |
0f9dd46c | 2798 | dump_space_info(sinfo, num_bytes); |
925baedd | 2799 | BUG(); |
925baedd | 2800 | } |
0f9dd46c JB |
2801 | |
2802 | return ret; | |
e6dcd2dc CM |
2803 | } |
2804 | ||
65b51a00 CM |
2805 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
2806 | { | |
0f9dd46c | 2807 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 2808 | int ret = 0; |
0f9dd46c | 2809 | |
0f9dd46c JB |
2810 | cache = btrfs_lookup_block_group(root->fs_info, start); |
2811 | if (!cache) { | |
d397712b CM |
2812 | printk(KERN_ERR "Unable to find block group for %llu\n", |
2813 | (unsigned long long)start); | |
0f9dd46c JB |
2814 | return -ENOSPC; |
2815 | } | |
1f3c79a2 LH |
2816 | |
2817 | ret = btrfs_discard_extent(root, start, len); | |
2818 | ||
0f9dd46c | 2819 | btrfs_add_free_space(cache, start, len); |
d2fb3437 | 2820 | put_block_group(cache); |
1a40e23b | 2821 | update_reserved_extents(root, start, len, 0); |
1f3c79a2 LH |
2822 | |
2823 | return ret; | |
65b51a00 CM |
2824 | } |
2825 | ||
e6dcd2dc CM |
2826 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
2827 | struct btrfs_root *root, | |
2828 | u64 num_bytes, u64 min_alloc_size, | |
2829 | u64 empty_size, u64 hint_byte, | |
2830 | u64 search_end, struct btrfs_key *ins, | |
2831 | u64 data) | |
2832 | { | |
2833 | int ret; | |
e6dcd2dc CM |
2834 | ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, |
2835 | empty_size, hint_byte, search_end, ins, | |
2836 | data); | |
e8569813 | 2837 | update_reserved_extents(root, ins->objectid, ins->offset, 1); |
e6dcd2dc CM |
2838 | return ret; |
2839 | } | |
2840 | ||
2841 | static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
31840ae1 | 2842 | struct btrfs_root *root, u64 parent, |
e6dcd2dc | 2843 | u64 root_objectid, u64 ref_generation, |
56bec294 CM |
2844 | u64 owner, struct btrfs_key *ins, |
2845 | int ref_mod) | |
e6dcd2dc CM |
2846 | { |
2847 | int ret; | |
e6dcd2dc CM |
2848 | u64 super_used; |
2849 | u64 root_used; | |
2850 | u64 num_bytes = ins->offset; | |
2851 | u32 sizes[2]; | |
2852 | struct btrfs_fs_info *info = root->fs_info; | |
2853 | struct btrfs_root *extent_root = info->extent_root; | |
2854 | struct btrfs_extent_item *extent_item; | |
2855 | struct btrfs_extent_ref *ref; | |
2856 | struct btrfs_path *path; | |
2857 | struct btrfs_key keys[2]; | |
fec577fb | 2858 | |
31840ae1 ZY |
2859 | if (parent == 0) |
2860 | parent = ins->objectid; | |
2861 | ||
58176a96 | 2862 | /* block accounting for super block */ |
75eff68e | 2863 | spin_lock(&info->delalloc_lock); |
db94535d CM |
2864 | super_used = btrfs_super_bytes_used(&info->super_copy); |
2865 | btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); | |
26b8003f | 2866 | |
58176a96 | 2867 | /* block accounting for root item */ |
db94535d CM |
2868 | root_used = btrfs_root_used(&root->root_item); |
2869 | btrfs_set_root_used(&root->root_item, root_used + num_bytes); | |
34bf63c4 | 2870 | spin_unlock(&info->delalloc_lock); |
58176a96 | 2871 | |
47e4bb98 | 2872 | memcpy(&keys[0], ins, sizeof(*ins)); |
47e4bb98 CM |
2873 | keys[1].objectid = ins->objectid; |
2874 | keys[1].type = BTRFS_EXTENT_REF_KEY; | |
31840ae1 | 2875 | keys[1].offset = parent; |
47e4bb98 CM |
2876 | sizes[0] = sizeof(*extent_item); |
2877 | sizes[1] = sizeof(*ref); | |
7bb86316 CM |
2878 | |
2879 | path = btrfs_alloc_path(); | |
2880 | BUG_ON(!path); | |
47e4bb98 | 2881 | |
b9473439 | 2882 | path->leave_spinning = 1; |
47e4bb98 CM |
2883 | ret = btrfs_insert_empty_items(trans, extent_root, path, keys, |
2884 | sizes, 2); | |
ccd467d6 | 2885 | BUG_ON(ret); |
0f9dd46c | 2886 | |
47e4bb98 CM |
2887 | extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2888 | struct btrfs_extent_item); | |
56bec294 | 2889 | btrfs_set_extent_refs(path->nodes[0], extent_item, ref_mod); |
47e4bb98 CM |
2890 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, |
2891 | struct btrfs_extent_ref); | |
2892 | ||
2893 | btrfs_set_ref_root(path->nodes[0], ref, root_objectid); | |
2894 | btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); | |
2895 | btrfs_set_ref_objectid(path->nodes[0], ref, owner); | |
56bec294 | 2896 | btrfs_set_ref_num_refs(path->nodes[0], ref, ref_mod); |
47e4bb98 CM |
2897 | |
2898 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
2899 | ||
2900 | trans->alloc_exclude_start = 0; | |
2901 | trans->alloc_exclude_nr = 0; | |
7bb86316 | 2902 | btrfs_free_path(path); |
f510cfec | 2903 | |
925baedd CM |
2904 | if (ret) |
2905 | goto out; | |
26b8003f | 2906 | |
d397712b CM |
2907 | ret = update_block_group(trans, root, ins->objectid, |
2908 | ins->offset, 1, 0); | |
f5947066 | 2909 | if (ret) { |
d397712b CM |
2910 | printk(KERN_ERR "btrfs update block group failed for %llu " |
2911 | "%llu\n", (unsigned long long)ins->objectid, | |
2912 | (unsigned long long)ins->offset); | |
f5947066 CM |
2913 | BUG(); |
2914 | } | |
925baedd | 2915 | out: |
e6dcd2dc CM |
2916 | return ret; |
2917 | } | |
2918 | ||
2919 | int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
31840ae1 | 2920 | struct btrfs_root *root, u64 parent, |
e6dcd2dc | 2921 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 2922 | u64 owner, struct btrfs_key *ins) |
e6dcd2dc CM |
2923 | { |
2924 | int ret; | |
1c2308f8 CM |
2925 | |
2926 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) | |
2927 | return 0; | |
56bec294 CM |
2928 | |
2929 | ret = btrfs_add_delayed_ref(trans, ins->objectid, | |
2930 | ins->offset, parent, root_objectid, | |
2931 | ref_generation, owner, | |
2932 | BTRFS_ADD_DELAYED_EXTENT, 0); | |
2933 | BUG_ON(ret); | |
e6dcd2dc CM |
2934 | return ret; |
2935 | } | |
e02119d5 CM |
2936 | |
2937 | /* | |
2938 | * this is used by the tree logging recovery code. It records that | |
2939 | * an extent has been allocated and makes sure to clear the free | |
2940 | * space cache bits as well | |
2941 | */ | |
2942 | int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans, | |
31840ae1 | 2943 | struct btrfs_root *root, u64 parent, |
e02119d5 | 2944 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 2945 | u64 owner, struct btrfs_key *ins) |
e02119d5 CM |
2946 | { |
2947 | int ret; | |
2948 | struct btrfs_block_group_cache *block_group; | |
2949 | ||
e02119d5 | 2950 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
ea6a478e | 2951 | mutex_lock(&block_group->cache_mutex); |
e02119d5 | 2952 | cache_block_group(root, block_group); |
ea6a478e | 2953 | mutex_unlock(&block_group->cache_mutex); |
e02119d5 | 2954 | |
ea6a478e JB |
2955 | ret = btrfs_remove_free_space(block_group, ins->objectid, |
2956 | ins->offset); | |
0f9dd46c | 2957 | BUG_ON(ret); |
d2fb3437 | 2958 | put_block_group(block_group); |
3bb1a1bc | 2959 | ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, |
56bec294 | 2960 | ref_generation, owner, ins, 1); |
e02119d5 CM |
2961 | return ret; |
2962 | } | |
2963 | ||
e6dcd2dc CM |
2964 | /* |
2965 | * finds a free extent and does all the dirty work required for allocation | |
2966 | * returns the key for the extent through ins, and a tree buffer for | |
2967 | * the first block of the extent through buf. | |
2968 | * | |
2969 | * returns 0 if everything worked, non-zero otherwise. | |
2970 | */ | |
2971 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, | |
2972 | struct btrfs_root *root, | |
31840ae1 | 2973 | u64 num_bytes, u64 parent, u64 min_alloc_size, |
e6dcd2dc | 2974 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 2975 | u64 owner_objectid, u64 empty_size, u64 hint_byte, |
e6dcd2dc CM |
2976 | u64 search_end, struct btrfs_key *ins, u64 data) |
2977 | { | |
2978 | int ret; | |
e6dcd2dc CM |
2979 | ret = __btrfs_reserve_extent(trans, root, num_bytes, |
2980 | min_alloc_size, empty_size, hint_byte, | |
2981 | search_end, ins, data); | |
2982 | BUG_ON(ret); | |
d00aff00 | 2983 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { |
56bec294 CM |
2984 | ret = btrfs_add_delayed_ref(trans, ins->objectid, |
2985 | ins->offset, parent, root_objectid, | |
2986 | ref_generation, owner_objectid, | |
2987 | BTRFS_ADD_DELAYED_EXTENT, 0); | |
d00aff00 | 2988 | BUG_ON(ret); |
d00aff00 | 2989 | } |
56bec294 | 2990 | update_reserved_extents(root, ins->objectid, ins->offset, 1); |
925baedd | 2991 | return ret; |
fec577fb | 2992 | } |
65b51a00 CM |
2993 | |
2994 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, | |
2995 | struct btrfs_root *root, | |
4008c04a CM |
2996 | u64 bytenr, u32 blocksize, |
2997 | int level) | |
65b51a00 CM |
2998 | { |
2999 | struct extent_buffer *buf; | |
3000 | ||
3001 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
3002 | if (!buf) | |
3003 | return ERR_PTR(-ENOMEM); | |
3004 | btrfs_set_header_generation(buf, trans->transid); | |
4008c04a | 3005 | btrfs_set_buffer_lockdep_class(buf, level); |
65b51a00 CM |
3006 | btrfs_tree_lock(buf); |
3007 | clean_tree_block(trans, root, buf); | |
b4ce94de CM |
3008 | |
3009 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 3010 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 3011 | |
d0c803c4 CM |
3012 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
3013 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
3014 | buf->start + buf->len - 1, GFP_NOFS); | |
3015 | } else { | |
3016 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 3017 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 3018 | } |
65b51a00 | 3019 | trans->blocks_used++; |
b4ce94de | 3020 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
3021 | return buf; |
3022 | } | |
3023 | ||
fec577fb CM |
3024 | /* |
3025 | * helper function to allocate a block for a given tree | |
3026 | * returns the tree buffer or NULL. | |
3027 | */ | |
5f39d397 | 3028 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
7bb86316 | 3029 | struct btrfs_root *root, |
31840ae1 | 3030 | u32 blocksize, u64 parent, |
7bb86316 CM |
3031 | u64 root_objectid, |
3032 | u64 ref_generation, | |
7bb86316 CM |
3033 | int level, |
3034 | u64 hint, | |
5f39d397 | 3035 | u64 empty_size) |
fec577fb | 3036 | { |
e2fa7227 | 3037 | struct btrfs_key ins; |
fec577fb | 3038 | int ret; |
5f39d397 | 3039 | struct extent_buffer *buf; |
fec577fb | 3040 | |
31840ae1 | 3041 | ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize, |
3bb1a1bc | 3042 | root_objectid, ref_generation, level, |
31840ae1 | 3043 | empty_size, hint, (u64)-1, &ins, 0); |
fec577fb | 3044 | if (ret) { |
54aa1f4d CM |
3045 | BUG_ON(ret > 0); |
3046 | return ERR_PTR(ret); | |
fec577fb | 3047 | } |
55c69072 | 3048 | |
4008c04a CM |
3049 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
3050 | blocksize, level); | |
fec577fb CM |
3051 | return buf; |
3052 | } | |
a28ec197 | 3053 | |
e02119d5 CM |
3054 | int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, |
3055 | struct btrfs_root *root, struct extent_buffer *leaf) | |
6407bf6d | 3056 | { |
7bb86316 CM |
3057 | u64 leaf_owner; |
3058 | u64 leaf_generation; | |
bd56b302 | 3059 | struct refsort *sorted; |
5f39d397 | 3060 | struct btrfs_key key; |
6407bf6d CM |
3061 | struct btrfs_file_extent_item *fi; |
3062 | int i; | |
3063 | int nritems; | |
3064 | int ret; | |
bd56b302 CM |
3065 | int refi = 0; |
3066 | int slot; | |
6407bf6d | 3067 | |
5f39d397 CM |
3068 | BUG_ON(!btrfs_is_leaf(leaf)); |
3069 | nritems = btrfs_header_nritems(leaf); | |
7bb86316 CM |
3070 | leaf_owner = btrfs_header_owner(leaf); |
3071 | leaf_generation = btrfs_header_generation(leaf); | |
3072 | ||
bd56b302 CM |
3073 | sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS); |
3074 | /* we do this loop twice. The first time we build a list | |
3075 | * of the extents we have a reference on, then we sort the list | |
3076 | * by bytenr. The second time around we actually do the | |
3077 | * extent freeing. | |
3078 | */ | |
6407bf6d | 3079 | for (i = 0; i < nritems; i++) { |
db94535d | 3080 | u64 disk_bytenr; |
e34a5b4f | 3081 | cond_resched(); |
5f39d397 CM |
3082 | |
3083 | btrfs_item_key_to_cpu(leaf, &key, i); | |
bd56b302 CM |
3084 | |
3085 | /* only extents have references, skip everything else */ | |
5f39d397 | 3086 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
6407bf6d | 3087 | continue; |
bd56b302 | 3088 | |
6407bf6d | 3089 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); |
bd56b302 CM |
3090 | |
3091 | /* inline extents live in the btree, they don't have refs */ | |
5f39d397 CM |
3092 | if (btrfs_file_extent_type(leaf, fi) == |
3093 | BTRFS_FILE_EXTENT_INLINE) | |
236454df | 3094 | continue; |
bd56b302 | 3095 | |
db94535d | 3096 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
bd56b302 CM |
3097 | |
3098 | /* holes don't have refs */ | |
db94535d | 3099 | if (disk_bytenr == 0) |
3a686375 | 3100 | continue; |
4a096752 | 3101 | |
bd56b302 CM |
3102 | sorted[refi].bytenr = disk_bytenr; |
3103 | sorted[refi].slot = i; | |
3104 | refi++; | |
3105 | } | |
3106 | ||
3107 | if (refi == 0) | |
3108 | goto out; | |
3109 | ||
3110 | sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); | |
3111 | ||
3112 | for (i = 0; i < refi; i++) { | |
3113 | u64 disk_bytenr; | |
3114 | ||
3115 | disk_bytenr = sorted[i].bytenr; | |
3116 | slot = sorted[i].slot; | |
3117 | ||
3118 | cond_resched(); | |
3119 | ||
3120 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
3121 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
3122 | continue; | |
3123 | ||
3124 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
3125 | ||
56bec294 | 3126 | ret = btrfs_free_extent(trans, root, disk_bytenr, |
7bb86316 | 3127 | btrfs_file_extent_disk_num_bytes(leaf, fi), |
31840ae1 | 3128 | leaf->start, leaf_owner, leaf_generation, |
3bb1a1bc | 3129 | key.objectid, 0); |
31840ae1 | 3130 | BUG_ON(ret); |
2dd3e67b CM |
3131 | |
3132 | atomic_inc(&root->fs_info->throttle_gen); | |
3133 | wake_up(&root->fs_info->transaction_throttle); | |
3134 | cond_resched(); | |
6407bf6d | 3135 | } |
bd56b302 CM |
3136 | out: |
3137 | kfree(sorted); | |
6407bf6d CM |
3138 | return 0; |
3139 | } | |
3140 | ||
d397712b | 3141 | static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans, |
e02119d5 CM |
3142 | struct btrfs_root *root, |
3143 | struct btrfs_leaf_ref *ref) | |
31153d81 YZ |
3144 | { |
3145 | int i; | |
3146 | int ret; | |
bd56b302 CM |
3147 | struct btrfs_extent_info *info; |
3148 | struct refsort *sorted; | |
3149 | ||
3150 | if (ref->nritems == 0) | |
3151 | return 0; | |
31153d81 | 3152 | |
bd56b302 CM |
3153 | sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS); |
3154 | for (i = 0; i < ref->nritems; i++) { | |
3155 | sorted[i].bytenr = ref->extents[i].bytenr; | |
3156 | sorted[i].slot = i; | |
3157 | } | |
3158 | sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL); | |
3159 | ||
3160 | /* | |
3161 | * the items in the ref were sorted when the ref was inserted | |
3162 | * into the ref cache, so this is already in order | |
3163 | */ | |
31153d81 | 3164 | for (i = 0; i < ref->nritems; i++) { |
bd56b302 | 3165 | info = ref->extents + sorted[i].slot; |
56bec294 | 3166 | ret = btrfs_free_extent(trans, root, info->bytenr, |
31840ae1 ZY |
3167 | info->num_bytes, ref->bytenr, |
3168 | ref->owner, ref->generation, | |
3bb1a1bc | 3169 | info->objectid, 0); |
2dd3e67b CM |
3170 | |
3171 | atomic_inc(&root->fs_info->throttle_gen); | |
3172 | wake_up(&root->fs_info->transaction_throttle); | |
3173 | cond_resched(); | |
3174 | ||
31153d81 YZ |
3175 | BUG_ON(ret); |
3176 | info++; | |
3177 | } | |
31153d81 | 3178 | |
806638bc | 3179 | kfree(sorted); |
31153d81 YZ |
3180 | return 0; |
3181 | } | |
3182 | ||
56bec294 CM |
3183 | static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans, |
3184 | struct btrfs_root *root, u64 start, | |
d397712b | 3185 | u64 len, u32 *refs) |
333db94c | 3186 | { |
017e5369 | 3187 | int ret; |
f87f057b | 3188 | |
56bec294 | 3189 | ret = btrfs_lookup_extent_ref(trans, root, start, len, refs); |
f87f057b CM |
3190 | BUG_ON(ret); |
3191 | ||
d397712b | 3192 | #if 0 /* some debugging code in case we see problems here */ |
f87f057b CM |
3193 | /* if the refs count is one, it won't get increased again. But |
3194 | * if the ref count is > 1, someone may be decreasing it at | |
3195 | * the same time we are. | |
3196 | */ | |
3197 | if (*refs != 1) { | |
3198 | struct extent_buffer *eb = NULL; | |
3199 | eb = btrfs_find_create_tree_block(root, start, len); | |
3200 | if (eb) | |
3201 | btrfs_tree_lock(eb); | |
3202 | ||
3203 | mutex_lock(&root->fs_info->alloc_mutex); | |
3204 | ret = lookup_extent_ref(NULL, root, start, len, refs); | |
3205 | BUG_ON(ret); | |
3206 | mutex_unlock(&root->fs_info->alloc_mutex); | |
3207 | ||
3208 | if (eb) { | |
3209 | btrfs_tree_unlock(eb); | |
3210 | free_extent_buffer(eb); | |
3211 | } | |
3212 | if (*refs == 1) { | |
d397712b CM |
3213 | printk(KERN_ERR "btrfs block %llu went down to one " |
3214 | "during drop_snap\n", (unsigned long long)start); | |
f87f057b CM |
3215 | } |
3216 | ||
3217 | } | |
3218 | #endif | |
3219 | ||
e7a84565 | 3220 | cond_resched(); |
017e5369 | 3221 | return ret; |
333db94c CM |
3222 | } |
3223 | ||
bd56b302 CM |
3224 | /* |
3225 | * this is used while deleting old snapshots, and it drops the refs | |
3226 | * on a whole subtree starting from a level 1 node. | |
3227 | * | |
3228 | * The idea is to sort all the leaf pointers, and then drop the | |
3229 | * ref on all the leaves in order. Most of the time the leaves | |
3230 | * will have ref cache entries, so no leaf IOs will be required to | |
3231 | * find the extents they have references on. | |
3232 | * | |
3233 | * For each leaf, any references it has are also dropped in order | |
3234 | * | |
3235 | * This ends up dropping the references in something close to optimal | |
3236 | * order for reading and modifying the extent allocation tree. | |
3237 | */ | |
3238 | static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans, | |
3239 | struct btrfs_root *root, | |
3240 | struct btrfs_path *path) | |
3241 | { | |
3242 | u64 bytenr; | |
3243 | u64 root_owner; | |
3244 | u64 root_gen; | |
3245 | struct extent_buffer *eb = path->nodes[1]; | |
3246 | struct extent_buffer *leaf; | |
3247 | struct btrfs_leaf_ref *ref; | |
3248 | struct refsort *sorted = NULL; | |
3249 | int nritems = btrfs_header_nritems(eb); | |
3250 | int ret; | |
3251 | int i; | |
3252 | int refi = 0; | |
3253 | int slot = path->slots[1]; | |
3254 | u32 blocksize = btrfs_level_size(root, 0); | |
3255 | u32 refs; | |
3256 | ||
3257 | if (nritems == 0) | |
3258 | goto out; | |
3259 | ||
3260 | root_owner = btrfs_header_owner(eb); | |
3261 | root_gen = btrfs_header_generation(eb); | |
3262 | sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS); | |
3263 | ||
3264 | /* | |
3265 | * step one, sort all the leaf pointers so we don't scribble | |
3266 | * randomly into the extent allocation tree | |
3267 | */ | |
3268 | for (i = slot; i < nritems; i++) { | |
3269 | sorted[refi].bytenr = btrfs_node_blockptr(eb, i); | |
3270 | sorted[refi].slot = i; | |
3271 | refi++; | |
3272 | } | |
3273 | ||
3274 | /* | |
3275 | * nritems won't be zero, but if we're picking up drop_snapshot | |
3276 | * after a crash, slot might be > 0, so double check things | |
3277 | * just in case. | |
3278 | */ | |
3279 | if (refi == 0) | |
3280 | goto out; | |
3281 | ||
3282 | sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); | |
3283 | ||
3284 | /* | |
3285 | * the first loop frees everything the leaves point to | |
3286 | */ | |
3287 | for (i = 0; i < refi; i++) { | |
3288 | u64 ptr_gen; | |
3289 | ||
3290 | bytenr = sorted[i].bytenr; | |
3291 | ||
3292 | /* | |
3293 | * check the reference count on this leaf. If it is > 1 | |
3294 | * we just decrement it below and don't update any | |
3295 | * of the refs the leaf points to. | |
3296 | */ | |
56bec294 CM |
3297 | ret = drop_snap_lookup_refcount(trans, root, bytenr, |
3298 | blocksize, &refs); | |
bd56b302 CM |
3299 | BUG_ON(ret); |
3300 | if (refs != 1) | |
3301 | continue; | |
3302 | ||
3303 | ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot); | |
3304 | ||
3305 | /* | |
3306 | * the leaf only had one reference, which means the | |
3307 | * only thing pointing to this leaf is the snapshot | |
3308 | * we're deleting. It isn't possible for the reference | |
3309 | * count to increase again later | |
3310 | * | |
3311 | * The reference cache is checked for the leaf, | |
3312 | * and if found we'll be able to drop any refs held by | |
3313 | * the leaf without needing to read it in. | |
3314 | */ | |
3315 | ref = btrfs_lookup_leaf_ref(root, bytenr); | |
3316 | if (ref && ref->generation != ptr_gen) { | |
3317 | btrfs_free_leaf_ref(root, ref); | |
3318 | ref = NULL; | |
3319 | } | |
3320 | if (ref) { | |
3321 | ret = cache_drop_leaf_ref(trans, root, ref); | |
3322 | BUG_ON(ret); | |
3323 | btrfs_remove_leaf_ref(root, ref); | |
3324 | btrfs_free_leaf_ref(root, ref); | |
3325 | } else { | |
3326 | /* | |
3327 | * the leaf wasn't in the reference cache, so | |
3328 | * we have to read it. | |
3329 | */ | |
3330 | leaf = read_tree_block(root, bytenr, blocksize, | |
3331 | ptr_gen); | |
3332 | ret = btrfs_drop_leaf_ref(trans, root, leaf); | |
3333 | BUG_ON(ret); | |
3334 | free_extent_buffer(leaf); | |
3335 | } | |
3336 | atomic_inc(&root->fs_info->throttle_gen); | |
3337 | wake_up(&root->fs_info->transaction_throttle); | |
3338 | cond_resched(); | |
3339 | } | |
3340 | ||
3341 | /* | |
3342 | * run through the loop again to free the refs on the leaves. | |
3343 | * This is faster than doing it in the loop above because | |
3344 | * the leaves are likely to be clustered together. We end up | |
3345 | * working in nice chunks on the extent allocation tree. | |
3346 | */ | |
3347 | for (i = 0; i < refi; i++) { | |
3348 | bytenr = sorted[i].bytenr; | |
56bec294 | 3349 | ret = btrfs_free_extent(trans, root, bytenr, |
bd56b302 CM |
3350 | blocksize, eb->start, |
3351 | root_owner, root_gen, 0, 1); | |
3352 | BUG_ON(ret); | |
3353 | ||
3354 | atomic_inc(&root->fs_info->throttle_gen); | |
3355 | wake_up(&root->fs_info->transaction_throttle); | |
3356 | cond_resched(); | |
3357 | } | |
3358 | out: | |
3359 | kfree(sorted); | |
3360 | ||
3361 | /* | |
3362 | * update the path to show we've processed the entire level 1 | |
3363 | * node. This will get saved into the root's drop_snapshot_progress | |
3364 | * field so these drops are not repeated again if this transaction | |
3365 | * commits. | |
3366 | */ | |
3367 | path->slots[1] = nritems; | |
3368 | return 0; | |
3369 | } | |
3370 | ||
9aca1d51 CM |
3371 | /* |
3372 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
3373 | * counts as it goes. | |
3374 | */ | |
d397712b | 3375 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, |
98ed5174 CM |
3376 | struct btrfs_root *root, |
3377 | struct btrfs_path *path, int *level) | |
20524f02 | 3378 | { |
7bb86316 CM |
3379 | u64 root_owner; |
3380 | u64 root_gen; | |
3381 | u64 bytenr; | |
ca7a79ad | 3382 | u64 ptr_gen; |
5f39d397 CM |
3383 | struct extent_buffer *next; |
3384 | struct extent_buffer *cur; | |
7bb86316 | 3385 | struct extent_buffer *parent; |
db94535d | 3386 | u32 blocksize; |
20524f02 CM |
3387 | int ret; |
3388 | u32 refs; | |
3389 | ||
5caf2a00 CM |
3390 | WARN_ON(*level < 0); |
3391 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
56bec294 | 3392 | ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start, |
db94535d | 3393 | path->nodes[*level]->len, &refs); |
20524f02 CM |
3394 | BUG_ON(ret); |
3395 | if (refs > 1) | |
3396 | goto out; | |
e011599b | 3397 | |
9aca1d51 CM |
3398 | /* |
3399 | * walk down to the last node level and free all the leaves | |
3400 | */ | |
d397712b | 3401 | while (*level >= 0) { |
5caf2a00 CM |
3402 | WARN_ON(*level < 0); |
3403 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
20524f02 | 3404 | cur = path->nodes[*level]; |
e011599b | 3405 | |
5f39d397 | 3406 | if (btrfs_header_level(cur) != *level) |
2c90e5d6 | 3407 | WARN_ON(1); |
e011599b | 3408 | |
7518a238 | 3409 | if (path->slots[*level] >= |
5f39d397 | 3410 | btrfs_header_nritems(cur)) |
20524f02 | 3411 | break; |
bd56b302 CM |
3412 | |
3413 | /* the new code goes down to level 1 and does all the | |
3414 | * leaves pointed to that node in bulk. So, this check | |
3415 | * for level 0 will always be false. | |
3416 | * | |
3417 | * But, the disk format allows the drop_snapshot_progress | |
3418 | * field in the root to leave things in a state where | |
3419 | * a leaf will need cleaning up here. If someone crashes | |
3420 | * with the old code and then boots with the new code, | |
3421 | * we might find a leaf here. | |
3422 | */ | |
6407bf6d | 3423 | if (*level == 0) { |
e02119d5 | 3424 | ret = btrfs_drop_leaf_ref(trans, root, cur); |
6407bf6d CM |
3425 | BUG_ON(ret); |
3426 | break; | |
3427 | } | |
bd56b302 CM |
3428 | |
3429 | /* | |
3430 | * once we get to level one, process the whole node | |
3431 | * at once, including everything below it. | |
3432 | */ | |
3433 | if (*level == 1) { | |
3434 | ret = drop_level_one_refs(trans, root, path); | |
3435 | BUG_ON(ret); | |
3436 | break; | |
3437 | } | |
3438 | ||
db94535d | 3439 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
ca7a79ad | 3440 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); |
db94535d | 3441 | blocksize = btrfs_level_size(root, *level - 1); |
925baedd | 3442 | |
56bec294 CM |
3443 | ret = drop_snap_lookup_refcount(trans, root, bytenr, |
3444 | blocksize, &refs); | |
6407bf6d | 3445 | BUG_ON(ret); |
bd56b302 CM |
3446 | |
3447 | /* | |
3448 | * if there is more than one reference, we don't need | |
3449 | * to read that node to drop any references it has. We | |
3450 | * just drop the ref we hold on that node and move on to the | |
3451 | * next slot in this level. | |
3452 | */ | |
6407bf6d | 3453 | if (refs != 1) { |
7bb86316 CM |
3454 | parent = path->nodes[*level]; |
3455 | root_owner = btrfs_header_owner(parent); | |
3456 | root_gen = btrfs_header_generation(parent); | |
20524f02 | 3457 | path->slots[*level]++; |
f87f057b | 3458 | |
56bec294 | 3459 | ret = btrfs_free_extent(trans, root, bytenr, |
31840ae1 | 3460 | blocksize, parent->start, |
3bb1a1bc YZ |
3461 | root_owner, root_gen, |
3462 | *level - 1, 1); | |
20524f02 | 3463 | BUG_ON(ret); |
18e35e0a CM |
3464 | |
3465 | atomic_inc(&root->fs_info->throttle_gen); | |
3466 | wake_up(&root->fs_info->transaction_throttle); | |
2dd3e67b | 3467 | cond_resched(); |
18e35e0a | 3468 | |
20524f02 CM |
3469 | continue; |
3470 | } | |
bd56b302 | 3471 | |
f87f057b | 3472 | /* |
bd56b302 CM |
3473 | * we need to keep freeing things in the next level down. |
3474 | * read the block and loop around to process it | |
f87f057b | 3475 | */ |
bd56b302 | 3476 | next = read_tree_block(root, bytenr, blocksize, ptr_gen); |
5caf2a00 | 3477 | WARN_ON(*level <= 0); |
83e15a28 | 3478 | if (path->nodes[*level-1]) |
5f39d397 | 3479 | free_extent_buffer(path->nodes[*level-1]); |
20524f02 | 3480 | path->nodes[*level-1] = next; |
5f39d397 | 3481 | *level = btrfs_header_level(next); |
20524f02 | 3482 | path->slots[*level] = 0; |
2dd3e67b | 3483 | cond_resched(); |
20524f02 CM |
3484 | } |
3485 | out: | |
5caf2a00 CM |
3486 | WARN_ON(*level < 0); |
3487 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
7bb86316 CM |
3488 | |
3489 | if (path->nodes[*level] == root->node) { | |
7bb86316 | 3490 | parent = path->nodes[*level]; |
31153d81 | 3491 | bytenr = path->nodes[*level]->start; |
7bb86316 CM |
3492 | } else { |
3493 | parent = path->nodes[*level + 1]; | |
31153d81 | 3494 | bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]); |
7bb86316 CM |
3495 | } |
3496 | ||
31153d81 YZ |
3497 | blocksize = btrfs_level_size(root, *level); |
3498 | root_owner = btrfs_header_owner(parent); | |
7bb86316 | 3499 | root_gen = btrfs_header_generation(parent); |
31153d81 | 3500 | |
bd56b302 CM |
3501 | /* |
3502 | * cleanup and free the reference on the last node | |
3503 | * we processed | |
3504 | */ | |
56bec294 | 3505 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
31840ae1 | 3506 | parent->start, root_owner, root_gen, |
3bb1a1bc | 3507 | *level, 1); |
5f39d397 | 3508 | free_extent_buffer(path->nodes[*level]); |
20524f02 | 3509 | path->nodes[*level] = NULL; |
bd56b302 | 3510 | |
20524f02 CM |
3511 | *level += 1; |
3512 | BUG_ON(ret); | |
f87f057b | 3513 | |
e7a84565 | 3514 | cond_resched(); |
20524f02 CM |
3515 | return 0; |
3516 | } | |
3517 | ||
f82d02d9 YZ |
3518 | /* |
3519 | * helper function for drop_subtree, this function is similar to | |
3520 | * walk_down_tree. The main difference is that it checks reference | |
3521 | * counts while tree blocks are locked. | |
3522 | */ | |
d397712b | 3523 | static noinline int walk_down_subtree(struct btrfs_trans_handle *trans, |
f82d02d9 YZ |
3524 | struct btrfs_root *root, |
3525 | struct btrfs_path *path, int *level) | |
3526 | { | |
3527 | struct extent_buffer *next; | |
3528 | struct extent_buffer *cur; | |
3529 | struct extent_buffer *parent; | |
3530 | u64 bytenr; | |
3531 | u64 ptr_gen; | |
3532 | u32 blocksize; | |
3533 | u32 refs; | |
3534 | int ret; | |
3535 | ||
3536 | cur = path->nodes[*level]; | |
3537 | ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len, | |
3538 | &refs); | |
3539 | BUG_ON(ret); | |
3540 | if (refs > 1) | |
3541 | goto out; | |
3542 | ||
3543 | while (*level >= 0) { | |
3544 | cur = path->nodes[*level]; | |
3545 | if (*level == 0) { | |
3546 | ret = btrfs_drop_leaf_ref(trans, root, cur); | |
3547 | BUG_ON(ret); | |
3548 | clean_tree_block(trans, root, cur); | |
3549 | break; | |
3550 | } | |
3551 | if (path->slots[*level] >= btrfs_header_nritems(cur)) { | |
3552 | clean_tree_block(trans, root, cur); | |
3553 | break; | |
3554 | } | |
3555 | ||
3556 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
3557 | blocksize = btrfs_level_size(root, *level - 1); | |
3558 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
3559 | ||
3560 | next = read_tree_block(root, bytenr, blocksize, ptr_gen); | |
3561 | btrfs_tree_lock(next); | |
b4ce94de | 3562 | btrfs_set_lock_blocking(next); |
f82d02d9 YZ |
3563 | |
3564 | ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize, | |
3565 | &refs); | |
3566 | BUG_ON(ret); | |
3567 | if (refs > 1) { | |
3568 | parent = path->nodes[*level]; | |
3569 | ret = btrfs_free_extent(trans, root, bytenr, | |
3570 | blocksize, parent->start, | |
3571 | btrfs_header_owner(parent), | |
3572 | btrfs_header_generation(parent), | |
3573 | *level - 1, 1); | |
3574 | BUG_ON(ret); | |
3575 | path->slots[*level]++; | |
3576 | btrfs_tree_unlock(next); | |
3577 | free_extent_buffer(next); | |
3578 | continue; | |
3579 | } | |
3580 | ||
3581 | *level = btrfs_header_level(next); | |
3582 | path->nodes[*level] = next; | |
3583 | path->slots[*level] = 0; | |
3584 | path->locks[*level] = 1; | |
3585 | cond_resched(); | |
3586 | } | |
3587 | out: | |
3588 | parent = path->nodes[*level + 1]; | |
3589 | bytenr = path->nodes[*level]->start; | |
3590 | blocksize = path->nodes[*level]->len; | |
3591 | ||
3592 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, | |
3593 | parent->start, btrfs_header_owner(parent), | |
3594 | btrfs_header_generation(parent), *level, 1); | |
3595 | BUG_ON(ret); | |
3596 | ||
3597 | if (path->locks[*level]) { | |
3598 | btrfs_tree_unlock(path->nodes[*level]); | |
3599 | path->locks[*level] = 0; | |
3600 | } | |
3601 | free_extent_buffer(path->nodes[*level]); | |
3602 | path->nodes[*level] = NULL; | |
3603 | *level += 1; | |
3604 | cond_resched(); | |
3605 | return 0; | |
3606 | } | |
3607 | ||
9aca1d51 CM |
3608 | /* |
3609 | * helper for dropping snapshots. This walks back up the tree in the path | |
3610 | * to find the first node higher up where we haven't yet gone through | |
3611 | * all the slots | |
3612 | */ | |
d397712b | 3613 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 3614 | struct btrfs_root *root, |
f82d02d9 YZ |
3615 | struct btrfs_path *path, |
3616 | int *level, int max_level) | |
20524f02 | 3617 | { |
7bb86316 CM |
3618 | u64 root_owner; |
3619 | u64 root_gen; | |
3620 | struct btrfs_root_item *root_item = &root->root_item; | |
20524f02 CM |
3621 | int i; |
3622 | int slot; | |
3623 | int ret; | |
9f3a7427 | 3624 | |
f82d02d9 | 3625 | for (i = *level; i < max_level && path->nodes[i]; i++) { |
20524f02 | 3626 | slot = path->slots[i]; |
5f39d397 CM |
3627 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { |
3628 | struct extent_buffer *node; | |
3629 | struct btrfs_disk_key disk_key; | |
bd56b302 CM |
3630 | |
3631 | /* | |
3632 | * there is more work to do in this level. | |
3633 | * Update the drop_progress marker to reflect | |
3634 | * the work we've done so far, and then bump | |
3635 | * the slot number | |
3636 | */ | |
5f39d397 | 3637 | node = path->nodes[i]; |
20524f02 CM |
3638 | path->slots[i]++; |
3639 | *level = i; | |
9f3a7427 | 3640 | WARN_ON(*level == 0); |
5f39d397 | 3641 | btrfs_node_key(node, &disk_key, path->slots[i]); |
9f3a7427 | 3642 | memcpy(&root_item->drop_progress, |
5f39d397 | 3643 | &disk_key, sizeof(disk_key)); |
9f3a7427 | 3644 | root_item->drop_level = i; |
20524f02 CM |
3645 | return 0; |
3646 | } else { | |
31840ae1 | 3647 | struct extent_buffer *parent; |
bd56b302 CM |
3648 | |
3649 | /* | |
3650 | * this whole node is done, free our reference | |
3651 | * on it and go up one level | |
3652 | */ | |
31840ae1 ZY |
3653 | if (path->nodes[*level] == root->node) |
3654 | parent = path->nodes[*level]; | |
3655 | else | |
3656 | parent = path->nodes[*level + 1]; | |
3657 | ||
3658 | root_owner = btrfs_header_owner(parent); | |
3659 | root_gen = btrfs_header_generation(parent); | |
f82d02d9 YZ |
3660 | |
3661 | clean_tree_block(trans, root, path->nodes[*level]); | |
e089f05c | 3662 | ret = btrfs_free_extent(trans, root, |
db94535d | 3663 | path->nodes[*level]->start, |
7bb86316 | 3664 | path->nodes[*level]->len, |
3bb1a1bc YZ |
3665 | parent->start, root_owner, |
3666 | root_gen, *level, 1); | |
6407bf6d | 3667 | BUG_ON(ret); |
f82d02d9 YZ |
3668 | if (path->locks[*level]) { |
3669 | btrfs_tree_unlock(path->nodes[*level]); | |
3670 | path->locks[*level] = 0; | |
3671 | } | |
5f39d397 | 3672 | free_extent_buffer(path->nodes[*level]); |
83e15a28 | 3673 | path->nodes[*level] = NULL; |
20524f02 | 3674 | *level = i + 1; |
20524f02 CM |
3675 | } |
3676 | } | |
3677 | return 1; | |
3678 | } | |
3679 | ||
9aca1d51 CM |
3680 | /* |
3681 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
3682 | * the tree freeing any blocks that have a ref count of zero after being | |
3683 | * decremented. | |
3684 | */ | |
e089f05c | 3685 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
9f3a7427 | 3686 | *root) |
20524f02 | 3687 | { |
3768f368 | 3688 | int ret = 0; |
9aca1d51 | 3689 | int wret; |
20524f02 | 3690 | int level; |
5caf2a00 | 3691 | struct btrfs_path *path; |
20524f02 CM |
3692 | int i; |
3693 | int orig_level; | |
c3e69d58 | 3694 | int update_count; |
9f3a7427 | 3695 | struct btrfs_root_item *root_item = &root->root_item; |
20524f02 | 3696 | |
a2135011 | 3697 | WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); |
5caf2a00 CM |
3698 | path = btrfs_alloc_path(); |
3699 | BUG_ON(!path); | |
20524f02 | 3700 | |
5f39d397 | 3701 | level = btrfs_header_level(root->node); |
20524f02 | 3702 | orig_level = level; |
9f3a7427 CM |
3703 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
3704 | path->nodes[level] = root->node; | |
f510cfec | 3705 | extent_buffer_get(root->node); |
9f3a7427 CM |
3706 | path->slots[level] = 0; |
3707 | } else { | |
3708 | struct btrfs_key key; | |
5f39d397 CM |
3709 | struct btrfs_disk_key found_key; |
3710 | struct extent_buffer *node; | |
6702ed49 | 3711 | |
9f3a7427 | 3712 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
6702ed49 CM |
3713 | level = root_item->drop_level; |
3714 | path->lowest_level = level; | |
9f3a7427 | 3715 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6702ed49 | 3716 | if (wret < 0) { |
9f3a7427 CM |
3717 | ret = wret; |
3718 | goto out; | |
3719 | } | |
5f39d397 CM |
3720 | node = path->nodes[level]; |
3721 | btrfs_node_key(node, &found_key, path->slots[level]); | |
3722 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, | |
3723 | sizeof(found_key))); | |
7d9eb12c CM |
3724 | /* |
3725 | * unlock our path, this is safe because only this | |
3726 | * function is allowed to delete this snapshot | |
3727 | */ | |
925baedd CM |
3728 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3729 | if (path->nodes[i] && path->locks[i]) { | |
3730 | path->locks[i] = 0; | |
3731 | btrfs_tree_unlock(path->nodes[i]); | |
3732 | } | |
3733 | } | |
9f3a7427 | 3734 | } |
d397712b | 3735 | while (1) { |
c3e69d58 | 3736 | unsigned long update; |
5caf2a00 | 3737 | wret = walk_down_tree(trans, root, path, &level); |
9aca1d51 | 3738 | if (wret > 0) |
20524f02 | 3739 | break; |
9aca1d51 CM |
3740 | if (wret < 0) |
3741 | ret = wret; | |
3742 | ||
f82d02d9 YZ |
3743 | wret = walk_up_tree(trans, root, path, &level, |
3744 | BTRFS_MAX_LEVEL); | |
9aca1d51 | 3745 | if (wret > 0) |
20524f02 | 3746 | break; |
9aca1d51 CM |
3747 | if (wret < 0) |
3748 | ret = wret; | |
b7ec40d7 CM |
3749 | if (trans->transaction->in_commit || |
3750 | trans->transaction->delayed_refs.flushing) { | |
e7a84565 CM |
3751 | ret = -EAGAIN; |
3752 | break; | |
3753 | } | |
18e35e0a | 3754 | atomic_inc(&root->fs_info->throttle_gen); |
017e5369 | 3755 | wake_up(&root->fs_info->transaction_throttle); |
c3e69d58 CM |
3756 | for (update_count = 0; update_count < 16; update_count++) { |
3757 | update = trans->delayed_ref_updates; | |
3758 | trans->delayed_ref_updates = 0; | |
3759 | if (update) | |
3760 | btrfs_run_delayed_refs(trans, root, update); | |
3761 | else | |
3762 | break; | |
3763 | } | |
20524f02 | 3764 | } |
83e15a28 | 3765 | for (i = 0; i <= orig_level; i++) { |
5caf2a00 | 3766 | if (path->nodes[i]) { |
5f39d397 | 3767 | free_extent_buffer(path->nodes[i]); |
0f82731f | 3768 | path->nodes[i] = NULL; |
83e15a28 | 3769 | } |
20524f02 | 3770 | } |
9f3a7427 | 3771 | out: |
5caf2a00 | 3772 | btrfs_free_path(path); |
9aca1d51 | 3773 | return ret; |
20524f02 | 3774 | } |
9078a3e1 | 3775 | |
f82d02d9 YZ |
3776 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
3777 | struct btrfs_root *root, | |
3778 | struct extent_buffer *node, | |
3779 | struct extent_buffer *parent) | |
3780 | { | |
3781 | struct btrfs_path *path; | |
3782 | int level; | |
3783 | int parent_level; | |
3784 | int ret = 0; | |
3785 | int wret; | |
3786 | ||
3787 | path = btrfs_alloc_path(); | |
3788 | BUG_ON(!path); | |
3789 | ||
b9447ef8 | 3790 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
3791 | parent_level = btrfs_header_level(parent); |
3792 | extent_buffer_get(parent); | |
3793 | path->nodes[parent_level] = parent; | |
3794 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
3795 | ||
b9447ef8 | 3796 | btrfs_assert_tree_locked(node); |
f82d02d9 YZ |
3797 | level = btrfs_header_level(node); |
3798 | extent_buffer_get(node); | |
3799 | path->nodes[level] = node; | |
3800 | path->slots[level] = 0; | |
3801 | ||
3802 | while (1) { | |
3803 | wret = walk_down_subtree(trans, root, path, &level); | |
3804 | if (wret < 0) | |
3805 | ret = wret; | |
3806 | if (wret != 0) | |
3807 | break; | |
3808 | ||
3809 | wret = walk_up_tree(trans, root, path, &level, parent_level); | |
3810 | if (wret < 0) | |
3811 | ret = wret; | |
3812 | if (wret != 0) | |
3813 | break; | |
3814 | } | |
3815 | ||
3816 | btrfs_free_path(path); | |
3817 | return ret; | |
3818 | } | |
3819 | ||
8e7bf94f CM |
3820 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
3821 | unsigned long nr) | |
3822 | { | |
3823 | return min(last, start + nr - 1); | |
3824 | } | |
3825 | ||
d397712b | 3826 | static noinline int relocate_inode_pages(struct inode *inode, u64 start, |
98ed5174 | 3827 | u64 len) |
edbd8d4e CM |
3828 | { |
3829 | u64 page_start; | |
3830 | u64 page_end; | |
1a40e23b | 3831 | unsigned long first_index; |
edbd8d4e | 3832 | unsigned long last_index; |
edbd8d4e CM |
3833 | unsigned long i; |
3834 | struct page *page; | |
d1310b2e | 3835 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 3836 | struct file_ra_state *ra; |
3eaa2885 | 3837 | struct btrfs_ordered_extent *ordered; |
1a40e23b ZY |
3838 | unsigned int total_read = 0; |
3839 | unsigned int total_dirty = 0; | |
3840 | int ret = 0; | |
4313b399 CM |
3841 | |
3842 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
3843 | |
3844 | mutex_lock(&inode->i_mutex); | |
1a40e23b | 3845 | first_index = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
3846 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
3847 | ||
1a40e23b ZY |
3848 | /* make sure the dirty trick played by the caller work */ |
3849 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
3850 | first_index, last_index); | |
3851 | if (ret) | |
3852 | goto out_unlock; | |
8e7bf94f | 3853 | |
4313b399 | 3854 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 3855 | |
1a40e23b ZY |
3856 | for (i = first_index ; i <= last_index; i++) { |
3857 | if (total_read % ra->ra_pages == 0) { | |
8e7bf94f | 3858 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, |
1a40e23b | 3859 | calc_ra(i, last_index, ra->ra_pages)); |
8e7bf94f CM |
3860 | } |
3861 | total_read++; | |
3eaa2885 CM |
3862 | again: |
3863 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
1a40e23b | 3864 | BUG_ON(1); |
edbd8d4e | 3865 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 3866 | if (!page) { |
1a40e23b | 3867 | ret = -ENOMEM; |
edbd8d4e | 3868 | goto out_unlock; |
a061fc8d | 3869 | } |
edbd8d4e CM |
3870 | if (!PageUptodate(page)) { |
3871 | btrfs_readpage(NULL, page); | |
3872 | lock_page(page); | |
3873 | if (!PageUptodate(page)) { | |
3874 | unlock_page(page); | |
3875 | page_cache_release(page); | |
1a40e23b | 3876 | ret = -EIO; |
edbd8d4e CM |
3877 | goto out_unlock; |
3878 | } | |
3879 | } | |
ec44a35c | 3880 | wait_on_page_writeback(page); |
3eaa2885 | 3881 | |
edbd8d4e CM |
3882 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
3883 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
d1310b2e | 3884 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 3885 | |
3eaa2885 CM |
3886 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
3887 | if (ordered) { | |
3888 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3889 | unlock_page(page); | |
3890 | page_cache_release(page); | |
3891 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3892 | btrfs_put_ordered_extent(ordered); | |
3893 | goto again; | |
3894 | } | |
3895 | set_page_extent_mapped(page); | |
3896 | ||
1a40e23b ZY |
3897 | if (i == first_index) |
3898 | set_extent_bits(io_tree, page_start, page_end, | |
3899 | EXTENT_BOUNDARY, GFP_NOFS); | |
1f80e4db | 3900 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
1a40e23b | 3901 | |
a061fc8d | 3902 | set_page_dirty(page); |
1a40e23b | 3903 | total_dirty++; |
edbd8d4e | 3904 | |
d1310b2e | 3905 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
3906 | unlock_page(page); |
3907 | page_cache_release(page); | |
3908 | } | |
3909 | ||
3910 | out_unlock: | |
ec44a35c | 3911 | kfree(ra); |
edbd8d4e | 3912 | mutex_unlock(&inode->i_mutex); |
1a40e23b ZY |
3913 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty); |
3914 | return ret; | |
edbd8d4e CM |
3915 | } |
3916 | ||
d397712b | 3917 | static noinline int relocate_data_extent(struct inode *reloc_inode, |
1a40e23b ZY |
3918 | struct btrfs_key *extent_key, |
3919 | u64 offset) | |
3920 | { | |
3921 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
3922 | struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree; | |
3923 | struct extent_map *em; | |
6643558d YZ |
3924 | u64 start = extent_key->objectid - offset; |
3925 | u64 end = start + extent_key->offset - 1; | |
bf4ef679 | 3926 | |
1a40e23b ZY |
3927 | em = alloc_extent_map(GFP_NOFS); |
3928 | BUG_ON(!em || IS_ERR(em)); | |
bf4ef679 | 3929 | |
6643558d | 3930 | em->start = start; |
1a40e23b | 3931 | em->len = extent_key->offset; |
c8b97818 | 3932 | em->block_len = extent_key->offset; |
1a40e23b ZY |
3933 | em->block_start = extent_key->objectid; |
3934 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
3935 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
3936 | ||
3937 | /* setup extent map to cheat btrfs_readpage */ | |
6643558d | 3938 | lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
1a40e23b ZY |
3939 | while (1) { |
3940 | int ret; | |
3941 | spin_lock(&em_tree->lock); | |
3942 | ret = add_extent_mapping(em_tree, em); | |
3943 | spin_unlock(&em_tree->lock); | |
3944 | if (ret != -EEXIST) { | |
3945 | free_extent_map(em); | |
bf4ef679 CM |
3946 | break; |
3947 | } | |
6643558d | 3948 | btrfs_drop_extent_cache(reloc_inode, start, end, 0); |
bf4ef679 | 3949 | } |
6643558d | 3950 | unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
bf4ef679 | 3951 | |
6643558d | 3952 | return relocate_inode_pages(reloc_inode, start, extent_key->offset); |
1a40e23b | 3953 | } |
edbd8d4e | 3954 | |
1a40e23b ZY |
3955 | struct btrfs_ref_path { |
3956 | u64 extent_start; | |
3957 | u64 nodes[BTRFS_MAX_LEVEL]; | |
3958 | u64 root_objectid; | |
3959 | u64 root_generation; | |
3960 | u64 owner_objectid; | |
1a40e23b ZY |
3961 | u32 num_refs; |
3962 | int lowest_level; | |
3963 | int current_level; | |
f82d02d9 YZ |
3964 | int shared_level; |
3965 | ||
3966 | struct btrfs_key node_keys[BTRFS_MAX_LEVEL]; | |
3967 | u64 new_nodes[BTRFS_MAX_LEVEL]; | |
1a40e23b | 3968 | }; |
7d9eb12c | 3969 | |
1a40e23b | 3970 | struct disk_extent { |
c8b97818 | 3971 | u64 ram_bytes; |
1a40e23b ZY |
3972 | u64 disk_bytenr; |
3973 | u64 disk_num_bytes; | |
3974 | u64 offset; | |
3975 | u64 num_bytes; | |
c8b97818 CM |
3976 | u8 compression; |
3977 | u8 encryption; | |
3978 | u16 other_encoding; | |
1a40e23b | 3979 | }; |
4313b399 | 3980 | |
1a40e23b ZY |
3981 | static int is_cowonly_root(u64 root_objectid) |
3982 | { | |
3983 | if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || | |
3984 | root_objectid == BTRFS_EXTENT_TREE_OBJECTID || | |
3985 | root_objectid == BTRFS_CHUNK_TREE_OBJECTID || | |
3986 | root_objectid == BTRFS_DEV_TREE_OBJECTID || | |
0403e47e YZ |
3987 | root_objectid == BTRFS_TREE_LOG_OBJECTID || |
3988 | root_objectid == BTRFS_CSUM_TREE_OBJECTID) | |
1a40e23b ZY |
3989 | return 1; |
3990 | return 0; | |
3991 | } | |
edbd8d4e | 3992 | |
d397712b | 3993 | static noinline int __next_ref_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
3994 | struct btrfs_root *extent_root, |
3995 | struct btrfs_ref_path *ref_path, | |
3996 | int first_time) | |
3997 | { | |
3998 | struct extent_buffer *leaf; | |
3999 | struct btrfs_path *path; | |
4000 | struct btrfs_extent_ref *ref; | |
4001 | struct btrfs_key key; | |
4002 | struct btrfs_key found_key; | |
4003 | u64 bytenr; | |
4004 | u32 nritems; | |
4005 | int level; | |
4006 | int ret = 1; | |
edbd8d4e | 4007 | |
1a40e23b ZY |
4008 | path = btrfs_alloc_path(); |
4009 | if (!path) | |
4010 | return -ENOMEM; | |
bf4ef679 | 4011 | |
1a40e23b ZY |
4012 | if (first_time) { |
4013 | ref_path->lowest_level = -1; | |
4014 | ref_path->current_level = -1; | |
f82d02d9 | 4015 | ref_path->shared_level = -1; |
1a40e23b ZY |
4016 | goto walk_up; |
4017 | } | |
4018 | walk_down: | |
4019 | level = ref_path->current_level - 1; | |
4020 | while (level >= -1) { | |
4021 | u64 parent; | |
4022 | if (level < ref_path->lowest_level) | |
4023 | break; | |
bf4ef679 | 4024 | |
d397712b | 4025 | if (level >= 0) |
1a40e23b | 4026 | bytenr = ref_path->nodes[level]; |
d397712b | 4027 | else |
1a40e23b | 4028 | bytenr = ref_path->extent_start; |
1a40e23b | 4029 | BUG_ON(bytenr == 0); |
bf4ef679 | 4030 | |
1a40e23b ZY |
4031 | parent = ref_path->nodes[level + 1]; |
4032 | ref_path->nodes[level + 1] = 0; | |
4033 | ref_path->current_level = level; | |
4034 | BUG_ON(parent == 0); | |
0ef3e66b | 4035 | |
1a40e23b ZY |
4036 | key.objectid = bytenr; |
4037 | key.offset = parent + 1; | |
4038 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 4039 | |
1a40e23b ZY |
4040 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
4041 | if (ret < 0) | |
edbd8d4e | 4042 | goto out; |
1a40e23b | 4043 | BUG_ON(ret == 0); |
7d9eb12c | 4044 | |
1a40e23b ZY |
4045 | leaf = path->nodes[0]; |
4046 | nritems = btrfs_header_nritems(leaf); | |
4047 | if (path->slots[0] >= nritems) { | |
4048 | ret = btrfs_next_leaf(extent_root, path); | |
4049 | if (ret < 0) | |
4050 | goto out; | |
4051 | if (ret > 0) | |
4052 | goto next; | |
4053 | leaf = path->nodes[0]; | |
4054 | } | |
0ef3e66b | 4055 | |
1a40e23b ZY |
4056 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4057 | if (found_key.objectid == bytenr && | |
f82d02d9 YZ |
4058 | found_key.type == BTRFS_EXTENT_REF_KEY) { |
4059 | if (level < ref_path->shared_level) | |
4060 | ref_path->shared_level = level; | |
1a40e23b | 4061 | goto found; |
f82d02d9 | 4062 | } |
1a40e23b ZY |
4063 | next: |
4064 | level--; | |
4065 | btrfs_release_path(extent_root, path); | |
d899e052 | 4066 | cond_resched(); |
1a40e23b ZY |
4067 | } |
4068 | /* reached lowest level */ | |
4069 | ret = 1; | |
4070 | goto out; | |
4071 | walk_up: | |
4072 | level = ref_path->current_level; | |
4073 | while (level < BTRFS_MAX_LEVEL - 1) { | |
4074 | u64 ref_objectid; | |
d397712b CM |
4075 | |
4076 | if (level >= 0) | |
1a40e23b | 4077 | bytenr = ref_path->nodes[level]; |
d397712b | 4078 | else |
1a40e23b | 4079 | bytenr = ref_path->extent_start; |
d397712b | 4080 | |
1a40e23b | 4081 | BUG_ON(bytenr == 0); |
edbd8d4e | 4082 | |
1a40e23b ZY |
4083 | key.objectid = bytenr; |
4084 | key.offset = 0; | |
4085 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 4086 | |
1a40e23b ZY |
4087 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
4088 | if (ret < 0) | |
4089 | goto out; | |
edbd8d4e | 4090 | |
1a40e23b ZY |
4091 | leaf = path->nodes[0]; |
4092 | nritems = btrfs_header_nritems(leaf); | |
4093 | if (path->slots[0] >= nritems) { | |
4094 | ret = btrfs_next_leaf(extent_root, path); | |
4095 | if (ret < 0) | |
4096 | goto out; | |
4097 | if (ret > 0) { | |
4098 | /* the extent was freed by someone */ | |
4099 | if (ref_path->lowest_level == level) | |
4100 | goto out; | |
4101 | btrfs_release_path(extent_root, path); | |
4102 | goto walk_down; | |
4103 | } | |
4104 | leaf = path->nodes[0]; | |
4105 | } | |
edbd8d4e | 4106 | |
1a40e23b ZY |
4107 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4108 | if (found_key.objectid != bytenr || | |
4109 | found_key.type != BTRFS_EXTENT_REF_KEY) { | |
4110 | /* the extent was freed by someone */ | |
4111 | if (ref_path->lowest_level == level) { | |
4112 | ret = 1; | |
4113 | goto out; | |
4114 | } | |
4115 | btrfs_release_path(extent_root, path); | |
4116 | goto walk_down; | |
4117 | } | |
4118 | found: | |
4119 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
4120 | struct btrfs_extent_ref); | |
4121 | ref_objectid = btrfs_ref_objectid(leaf, ref); | |
4122 | if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
4123 | if (first_time) { | |
4124 | level = (int)ref_objectid; | |
4125 | BUG_ON(level >= BTRFS_MAX_LEVEL); | |
4126 | ref_path->lowest_level = level; | |
4127 | ref_path->current_level = level; | |
4128 | ref_path->nodes[level] = bytenr; | |
4129 | } else { | |
4130 | WARN_ON(ref_objectid != level); | |
4131 | } | |
4132 | } else { | |
4133 | WARN_ON(level != -1); | |
4134 | } | |
4135 | first_time = 0; | |
bf4ef679 | 4136 | |
1a40e23b ZY |
4137 | if (ref_path->lowest_level == level) { |
4138 | ref_path->owner_objectid = ref_objectid; | |
1a40e23b ZY |
4139 | ref_path->num_refs = btrfs_ref_num_refs(leaf, ref); |
4140 | } | |
bf4ef679 | 4141 | |
7d9eb12c | 4142 | /* |
1a40e23b ZY |
4143 | * the block is tree root or the block isn't in reference |
4144 | * counted tree. | |
7d9eb12c | 4145 | */ |
1a40e23b ZY |
4146 | if (found_key.objectid == found_key.offset || |
4147 | is_cowonly_root(btrfs_ref_root(leaf, ref))) { | |
4148 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
4149 | ref_path->root_generation = | |
4150 | btrfs_ref_generation(leaf, ref); | |
4151 | if (level < 0) { | |
4152 | /* special reference from the tree log */ | |
4153 | ref_path->nodes[0] = found_key.offset; | |
4154 | ref_path->current_level = 0; | |
4155 | } | |
4156 | ret = 0; | |
4157 | goto out; | |
4158 | } | |
7d9eb12c | 4159 | |
1a40e23b ZY |
4160 | level++; |
4161 | BUG_ON(ref_path->nodes[level] != 0); | |
4162 | ref_path->nodes[level] = found_key.offset; | |
4163 | ref_path->current_level = level; | |
bf4ef679 | 4164 | |
1a40e23b ZY |
4165 | /* |
4166 | * the reference was created in the running transaction, | |
4167 | * no need to continue walking up. | |
4168 | */ | |
4169 | if (btrfs_ref_generation(leaf, ref) == trans->transid) { | |
4170 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
4171 | ref_path->root_generation = | |
4172 | btrfs_ref_generation(leaf, ref); | |
4173 | ret = 0; | |
4174 | goto out; | |
7d9eb12c CM |
4175 | } |
4176 | ||
1a40e23b | 4177 | btrfs_release_path(extent_root, path); |
d899e052 | 4178 | cond_resched(); |
7d9eb12c | 4179 | } |
1a40e23b ZY |
4180 | /* reached max tree level, but no tree root found. */ |
4181 | BUG(); | |
edbd8d4e | 4182 | out: |
1a40e23b ZY |
4183 | btrfs_free_path(path); |
4184 | return ret; | |
edbd8d4e CM |
4185 | } |
4186 | ||
1a40e23b ZY |
4187 | static int btrfs_first_ref_path(struct btrfs_trans_handle *trans, |
4188 | struct btrfs_root *extent_root, | |
4189 | struct btrfs_ref_path *ref_path, | |
4190 | u64 extent_start) | |
a061fc8d | 4191 | { |
1a40e23b ZY |
4192 | memset(ref_path, 0, sizeof(*ref_path)); |
4193 | ref_path->extent_start = extent_start; | |
a061fc8d | 4194 | |
1a40e23b | 4195 | return __next_ref_path(trans, extent_root, ref_path, 1); |
a061fc8d CM |
4196 | } |
4197 | ||
1a40e23b ZY |
4198 | static int btrfs_next_ref_path(struct btrfs_trans_handle *trans, |
4199 | struct btrfs_root *extent_root, | |
4200 | struct btrfs_ref_path *ref_path) | |
edbd8d4e | 4201 | { |
1a40e23b ZY |
4202 | return __next_ref_path(trans, extent_root, ref_path, 0); |
4203 | } | |
4204 | ||
d397712b | 4205 | static noinline int get_new_locations(struct inode *reloc_inode, |
1a40e23b ZY |
4206 | struct btrfs_key *extent_key, |
4207 | u64 offset, int no_fragment, | |
4208 | struct disk_extent **extents, | |
4209 | int *nr_extents) | |
4210 | { | |
4211 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
4212 | struct btrfs_path *path; | |
4213 | struct btrfs_file_extent_item *fi; | |
edbd8d4e | 4214 | struct extent_buffer *leaf; |
1a40e23b ZY |
4215 | struct disk_extent *exts = *extents; |
4216 | struct btrfs_key found_key; | |
4217 | u64 cur_pos; | |
4218 | u64 last_byte; | |
edbd8d4e | 4219 | u32 nritems; |
1a40e23b ZY |
4220 | int nr = 0; |
4221 | int max = *nr_extents; | |
4222 | int ret; | |
edbd8d4e | 4223 | |
1a40e23b ZY |
4224 | WARN_ON(!no_fragment && *extents); |
4225 | if (!exts) { | |
4226 | max = 1; | |
4227 | exts = kmalloc(sizeof(*exts) * max, GFP_NOFS); | |
4228 | if (!exts) | |
4229 | return -ENOMEM; | |
a061fc8d | 4230 | } |
edbd8d4e | 4231 | |
1a40e23b ZY |
4232 | path = btrfs_alloc_path(); |
4233 | BUG_ON(!path); | |
edbd8d4e | 4234 | |
1a40e23b ZY |
4235 | cur_pos = extent_key->objectid - offset; |
4236 | last_byte = extent_key->objectid + extent_key->offset; | |
4237 | ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino, | |
4238 | cur_pos, 0); | |
4239 | if (ret < 0) | |
4240 | goto out; | |
4241 | if (ret > 0) { | |
4242 | ret = -ENOENT; | |
4243 | goto out; | |
4244 | } | |
edbd8d4e | 4245 | |
1a40e23b | 4246 | while (1) { |
edbd8d4e CM |
4247 | leaf = path->nodes[0]; |
4248 | nritems = btrfs_header_nritems(leaf); | |
1a40e23b ZY |
4249 | if (path->slots[0] >= nritems) { |
4250 | ret = btrfs_next_leaf(root, path); | |
a061fc8d CM |
4251 | if (ret < 0) |
4252 | goto out; | |
1a40e23b ZY |
4253 | if (ret > 0) |
4254 | break; | |
bf4ef679 | 4255 | leaf = path->nodes[0]; |
a061fc8d | 4256 | } |
edbd8d4e CM |
4257 | |
4258 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b ZY |
4259 | if (found_key.offset != cur_pos || |
4260 | found_key.type != BTRFS_EXTENT_DATA_KEY || | |
4261 | found_key.objectid != reloc_inode->i_ino) | |
edbd8d4e CM |
4262 | break; |
4263 | ||
1a40e23b ZY |
4264 | fi = btrfs_item_ptr(leaf, path->slots[0], |
4265 | struct btrfs_file_extent_item); | |
4266 | if (btrfs_file_extent_type(leaf, fi) != | |
4267 | BTRFS_FILE_EXTENT_REG || | |
4268 | btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
edbd8d4e | 4269 | break; |
1a40e23b ZY |
4270 | |
4271 | if (nr == max) { | |
4272 | struct disk_extent *old = exts; | |
4273 | max *= 2; | |
4274 | exts = kzalloc(sizeof(*exts) * max, GFP_NOFS); | |
4275 | memcpy(exts, old, sizeof(*exts) * nr); | |
4276 | if (old != *extents) | |
4277 | kfree(old); | |
a061fc8d | 4278 | } |
edbd8d4e | 4279 | |
1a40e23b ZY |
4280 | exts[nr].disk_bytenr = |
4281 | btrfs_file_extent_disk_bytenr(leaf, fi); | |
4282 | exts[nr].disk_num_bytes = | |
4283 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
4284 | exts[nr].offset = btrfs_file_extent_offset(leaf, fi); | |
4285 | exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
c8b97818 CM |
4286 | exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
4287 | exts[nr].compression = btrfs_file_extent_compression(leaf, fi); | |
4288 | exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi); | |
4289 | exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf, | |
4290 | fi); | |
d899e052 YZ |
4291 | BUG_ON(exts[nr].offset > 0); |
4292 | BUG_ON(exts[nr].compression || exts[nr].encryption); | |
4293 | BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes); | |
edbd8d4e | 4294 | |
1a40e23b ZY |
4295 | cur_pos += exts[nr].num_bytes; |
4296 | nr++; | |
4297 | ||
4298 | if (cur_pos + offset >= last_byte) | |
4299 | break; | |
4300 | ||
4301 | if (no_fragment) { | |
4302 | ret = 1; | |
edbd8d4e | 4303 | goto out; |
1a40e23b ZY |
4304 | } |
4305 | path->slots[0]++; | |
4306 | } | |
4307 | ||
1f80e4db | 4308 | BUG_ON(cur_pos + offset > last_byte); |
1a40e23b ZY |
4309 | if (cur_pos + offset < last_byte) { |
4310 | ret = -ENOENT; | |
4311 | goto out; | |
edbd8d4e CM |
4312 | } |
4313 | ret = 0; | |
4314 | out: | |
1a40e23b ZY |
4315 | btrfs_free_path(path); |
4316 | if (ret) { | |
4317 | if (exts != *extents) | |
4318 | kfree(exts); | |
4319 | } else { | |
4320 | *extents = exts; | |
4321 | *nr_extents = nr; | |
4322 | } | |
4323 | return ret; | |
4324 | } | |
4325 | ||
d397712b | 4326 | static noinline int replace_one_extent(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
4327 | struct btrfs_root *root, |
4328 | struct btrfs_path *path, | |
4329 | struct btrfs_key *extent_key, | |
4330 | struct btrfs_key *leaf_key, | |
4331 | struct btrfs_ref_path *ref_path, | |
4332 | struct disk_extent *new_extents, | |
4333 | int nr_extents) | |
4334 | { | |
4335 | struct extent_buffer *leaf; | |
4336 | struct btrfs_file_extent_item *fi; | |
4337 | struct inode *inode = NULL; | |
4338 | struct btrfs_key key; | |
4339 | u64 lock_start = 0; | |
4340 | u64 lock_end = 0; | |
4341 | u64 num_bytes; | |
4342 | u64 ext_offset; | |
86288a19 | 4343 | u64 search_end = (u64)-1; |
1a40e23b | 4344 | u32 nritems; |
3bb1a1bc | 4345 | int nr_scaned = 0; |
1a40e23b | 4346 | int extent_locked = 0; |
d899e052 | 4347 | int extent_type; |
1a40e23b ZY |
4348 | int ret; |
4349 | ||
3bb1a1bc | 4350 | memcpy(&key, leaf_key, sizeof(key)); |
1a40e23b | 4351 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { |
3bb1a1bc YZ |
4352 | if (key.objectid < ref_path->owner_objectid || |
4353 | (key.objectid == ref_path->owner_objectid && | |
4354 | key.type < BTRFS_EXTENT_DATA_KEY)) { | |
4355 | key.objectid = ref_path->owner_objectid; | |
4356 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4357 | key.offset = 0; | |
4358 | } | |
1a40e23b ZY |
4359 | } |
4360 | ||
4361 | while (1) { | |
4362 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
4363 | if (ret < 0) | |
4364 | goto out; | |
4365 | ||
4366 | leaf = path->nodes[0]; | |
4367 | nritems = btrfs_header_nritems(leaf); | |
4368 | next: | |
4369 | if (extent_locked && ret > 0) { | |
4370 | /* | |
4371 | * the file extent item was modified by someone | |
4372 | * before the extent got locked. | |
4373 | */ | |
1a40e23b ZY |
4374 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
4375 | lock_end, GFP_NOFS); | |
4376 | extent_locked = 0; | |
4377 | } | |
4378 | ||
4379 | if (path->slots[0] >= nritems) { | |
3bb1a1bc | 4380 | if (++nr_scaned > 2) |
1a40e23b ZY |
4381 | break; |
4382 | ||
4383 | BUG_ON(extent_locked); | |
4384 | ret = btrfs_next_leaf(root, path); | |
4385 | if (ret < 0) | |
4386 | goto out; | |
4387 | if (ret > 0) | |
4388 | break; | |
4389 | leaf = path->nodes[0]; | |
4390 | nritems = btrfs_header_nritems(leaf); | |
4391 | } | |
4392 | ||
4393 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4394 | ||
4395 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { | |
4396 | if ((key.objectid > ref_path->owner_objectid) || | |
4397 | (key.objectid == ref_path->owner_objectid && | |
4398 | key.type > BTRFS_EXTENT_DATA_KEY) || | |
86288a19 | 4399 | key.offset >= search_end) |
1a40e23b ZY |
4400 | break; |
4401 | } | |
4402 | ||
4403 | if (inode && key.objectid != inode->i_ino) { | |
4404 | BUG_ON(extent_locked); | |
4405 | btrfs_release_path(root, path); | |
4406 | mutex_unlock(&inode->i_mutex); | |
4407 | iput(inode); | |
4408 | inode = NULL; | |
4409 | continue; | |
4410 | } | |
4411 | ||
4412 | if (key.type != BTRFS_EXTENT_DATA_KEY) { | |
4413 | path->slots[0]++; | |
4414 | ret = 1; | |
4415 | goto next; | |
4416 | } | |
4417 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4418 | struct btrfs_file_extent_item); | |
d899e052 YZ |
4419 | extent_type = btrfs_file_extent_type(leaf, fi); |
4420 | if ((extent_type != BTRFS_FILE_EXTENT_REG && | |
4421 | extent_type != BTRFS_FILE_EXTENT_PREALLOC) || | |
1a40e23b ZY |
4422 | (btrfs_file_extent_disk_bytenr(leaf, fi) != |
4423 | extent_key->objectid)) { | |
4424 | path->slots[0]++; | |
4425 | ret = 1; | |
4426 | goto next; | |
4427 | } | |
4428 | ||
4429 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
4430 | ext_offset = btrfs_file_extent_offset(leaf, fi); | |
4431 | ||
86288a19 YZ |
4432 | if (search_end == (u64)-1) { |
4433 | search_end = key.offset - ext_offset + | |
4434 | btrfs_file_extent_ram_bytes(leaf, fi); | |
4435 | } | |
1a40e23b ZY |
4436 | |
4437 | if (!extent_locked) { | |
4438 | lock_start = key.offset; | |
4439 | lock_end = lock_start + num_bytes - 1; | |
4440 | } else { | |
6643558d YZ |
4441 | if (lock_start > key.offset || |
4442 | lock_end + 1 < key.offset + num_bytes) { | |
4443 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
4444 | lock_start, lock_end, GFP_NOFS); | |
4445 | extent_locked = 0; | |
4446 | } | |
1a40e23b ZY |
4447 | } |
4448 | ||
4449 | if (!inode) { | |
4450 | btrfs_release_path(root, path); | |
4451 | ||
4452 | inode = btrfs_iget_locked(root->fs_info->sb, | |
4453 | key.objectid, root); | |
4454 | if (inode->i_state & I_NEW) { | |
4455 | BTRFS_I(inode)->root = root; | |
4456 | BTRFS_I(inode)->location.objectid = | |
4457 | key.objectid; | |
4458 | BTRFS_I(inode)->location.type = | |
4459 | BTRFS_INODE_ITEM_KEY; | |
4460 | BTRFS_I(inode)->location.offset = 0; | |
4461 | btrfs_read_locked_inode(inode); | |
4462 | unlock_new_inode(inode); | |
4463 | } | |
4464 | /* | |
4465 | * some code call btrfs_commit_transaction while | |
4466 | * holding the i_mutex, so we can't use mutex_lock | |
4467 | * here. | |
4468 | */ | |
4469 | if (is_bad_inode(inode) || | |
4470 | !mutex_trylock(&inode->i_mutex)) { | |
4471 | iput(inode); | |
4472 | inode = NULL; | |
4473 | key.offset = (u64)-1; | |
4474 | goto skip; | |
4475 | } | |
4476 | } | |
4477 | ||
4478 | if (!extent_locked) { | |
4479 | struct btrfs_ordered_extent *ordered; | |
4480 | ||
4481 | btrfs_release_path(root, path); | |
4482 | ||
4483 | lock_extent(&BTRFS_I(inode)->io_tree, lock_start, | |
4484 | lock_end, GFP_NOFS); | |
4485 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
4486 | lock_end); | |
4487 | if (ordered && | |
4488 | ordered->file_offset <= lock_end && | |
4489 | ordered->file_offset + ordered->len > lock_start) { | |
4490 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
4491 | lock_start, lock_end, GFP_NOFS); | |
4492 | btrfs_start_ordered_extent(inode, ordered, 1); | |
4493 | btrfs_put_ordered_extent(ordered); | |
4494 | key.offset += num_bytes; | |
4495 | goto skip; | |
4496 | } | |
4497 | if (ordered) | |
4498 | btrfs_put_ordered_extent(ordered); | |
4499 | ||
1a40e23b ZY |
4500 | extent_locked = 1; |
4501 | continue; | |
4502 | } | |
4503 | ||
4504 | if (nr_extents == 1) { | |
4505 | /* update extent pointer in place */ | |
1a40e23b ZY |
4506 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
4507 | new_extents[0].disk_bytenr); | |
4508 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
4509 | new_extents[0].disk_num_bytes); | |
1a40e23b ZY |
4510 | btrfs_mark_buffer_dirty(leaf); |
4511 | ||
4512 | btrfs_drop_extent_cache(inode, key.offset, | |
4513 | key.offset + num_bytes - 1, 0); | |
4514 | ||
4515 | ret = btrfs_inc_extent_ref(trans, root, | |
4516 | new_extents[0].disk_bytenr, | |
4517 | new_extents[0].disk_num_bytes, | |
4518 | leaf->start, | |
4519 | root->root_key.objectid, | |
4520 | trans->transid, | |
3bb1a1bc | 4521 | key.objectid); |
1a40e23b ZY |
4522 | BUG_ON(ret); |
4523 | ||
4524 | ret = btrfs_free_extent(trans, root, | |
4525 | extent_key->objectid, | |
4526 | extent_key->offset, | |
4527 | leaf->start, | |
4528 | btrfs_header_owner(leaf), | |
4529 | btrfs_header_generation(leaf), | |
3bb1a1bc | 4530 | key.objectid, 0); |
1a40e23b ZY |
4531 | BUG_ON(ret); |
4532 | ||
4533 | btrfs_release_path(root, path); | |
4534 | key.offset += num_bytes; | |
4535 | } else { | |
d899e052 YZ |
4536 | BUG_ON(1); |
4537 | #if 0 | |
1a40e23b ZY |
4538 | u64 alloc_hint; |
4539 | u64 extent_len; | |
4540 | int i; | |
4541 | /* | |
4542 | * drop old extent pointer at first, then insert the | |
4543 | * new pointers one bye one | |
4544 | */ | |
4545 | btrfs_release_path(root, path); | |
4546 | ret = btrfs_drop_extents(trans, root, inode, key.offset, | |
4547 | key.offset + num_bytes, | |
4548 | key.offset, &alloc_hint); | |
4549 | BUG_ON(ret); | |
4550 | ||
4551 | for (i = 0; i < nr_extents; i++) { | |
4552 | if (ext_offset >= new_extents[i].num_bytes) { | |
4553 | ext_offset -= new_extents[i].num_bytes; | |
4554 | continue; | |
4555 | } | |
4556 | extent_len = min(new_extents[i].num_bytes - | |
4557 | ext_offset, num_bytes); | |
4558 | ||
4559 | ret = btrfs_insert_empty_item(trans, root, | |
4560 | path, &key, | |
4561 | sizeof(*fi)); | |
4562 | BUG_ON(ret); | |
4563 | ||
4564 | leaf = path->nodes[0]; | |
4565 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4566 | struct btrfs_file_extent_item); | |
4567 | btrfs_set_file_extent_generation(leaf, fi, | |
4568 | trans->transid); | |
4569 | btrfs_set_file_extent_type(leaf, fi, | |
4570 | BTRFS_FILE_EXTENT_REG); | |
4571 | btrfs_set_file_extent_disk_bytenr(leaf, fi, | |
4572 | new_extents[i].disk_bytenr); | |
4573 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
4574 | new_extents[i].disk_num_bytes); | |
c8b97818 CM |
4575 | btrfs_set_file_extent_ram_bytes(leaf, fi, |
4576 | new_extents[i].ram_bytes); | |
4577 | ||
4578 | btrfs_set_file_extent_compression(leaf, fi, | |
4579 | new_extents[i].compression); | |
4580 | btrfs_set_file_extent_encryption(leaf, fi, | |
4581 | new_extents[i].encryption); | |
4582 | btrfs_set_file_extent_other_encoding(leaf, fi, | |
4583 | new_extents[i].other_encoding); | |
4584 | ||
1a40e23b ZY |
4585 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4586 | extent_len); | |
4587 | ext_offset += new_extents[i].offset; | |
4588 | btrfs_set_file_extent_offset(leaf, fi, | |
4589 | ext_offset); | |
4590 | btrfs_mark_buffer_dirty(leaf); | |
4591 | ||
4592 | btrfs_drop_extent_cache(inode, key.offset, | |
4593 | key.offset + extent_len - 1, 0); | |
4594 | ||
4595 | ret = btrfs_inc_extent_ref(trans, root, | |
4596 | new_extents[i].disk_bytenr, | |
4597 | new_extents[i].disk_num_bytes, | |
4598 | leaf->start, | |
4599 | root->root_key.objectid, | |
3bb1a1bc | 4600 | trans->transid, key.objectid); |
1a40e23b ZY |
4601 | BUG_ON(ret); |
4602 | btrfs_release_path(root, path); | |
4603 | ||
a76a3cd4 | 4604 | inode_add_bytes(inode, extent_len); |
1a40e23b ZY |
4605 | |
4606 | ext_offset = 0; | |
4607 | num_bytes -= extent_len; | |
4608 | key.offset += extent_len; | |
4609 | ||
4610 | if (num_bytes == 0) | |
4611 | break; | |
4612 | } | |
4613 | BUG_ON(i >= nr_extents); | |
d899e052 | 4614 | #endif |
1a40e23b ZY |
4615 | } |
4616 | ||
4617 | if (extent_locked) { | |
1a40e23b ZY |
4618 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
4619 | lock_end, GFP_NOFS); | |
4620 | extent_locked = 0; | |
4621 | } | |
4622 | skip: | |
4623 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS && | |
86288a19 | 4624 | key.offset >= search_end) |
1a40e23b ZY |
4625 | break; |
4626 | ||
4627 | cond_resched(); | |
4628 | } | |
4629 | ret = 0; | |
4630 | out: | |
4631 | btrfs_release_path(root, path); | |
4632 | if (inode) { | |
4633 | mutex_unlock(&inode->i_mutex); | |
4634 | if (extent_locked) { | |
1a40e23b ZY |
4635 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
4636 | lock_end, GFP_NOFS); | |
4637 | } | |
4638 | iput(inode); | |
4639 | } | |
4640 | return ret; | |
4641 | } | |
4642 | ||
1a40e23b ZY |
4643 | int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, |
4644 | struct btrfs_root *root, | |
4645 | struct extent_buffer *buf, u64 orig_start) | |
4646 | { | |
4647 | int level; | |
4648 | int ret; | |
4649 | ||
4650 | BUG_ON(btrfs_header_generation(buf) != trans->transid); | |
4651 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
4652 | ||
4653 | level = btrfs_header_level(buf); | |
4654 | if (level == 0) { | |
4655 | struct btrfs_leaf_ref *ref; | |
4656 | struct btrfs_leaf_ref *orig_ref; | |
4657 | ||
4658 | orig_ref = btrfs_lookup_leaf_ref(root, orig_start); | |
4659 | if (!orig_ref) | |
4660 | return -ENOENT; | |
4661 | ||
4662 | ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems); | |
4663 | if (!ref) { | |
4664 | btrfs_free_leaf_ref(root, orig_ref); | |
4665 | return -ENOMEM; | |
4666 | } | |
4667 | ||
4668 | ref->nritems = orig_ref->nritems; | |
4669 | memcpy(ref->extents, orig_ref->extents, | |
4670 | sizeof(ref->extents[0]) * ref->nritems); | |
4671 | ||
4672 | btrfs_free_leaf_ref(root, orig_ref); | |
4673 | ||
4674 | ref->root_gen = trans->transid; | |
4675 | ref->bytenr = buf->start; | |
4676 | ref->owner = btrfs_header_owner(buf); | |
4677 | ref->generation = btrfs_header_generation(buf); | |
bd56b302 | 4678 | |
1a40e23b ZY |
4679 | ret = btrfs_add_leaf_ref(root, ref, 0); |
4680 | WARN_ON(ret); | |
4681 | btrfs_free_leaf_ref(root, ref); | |
4682 | } | |
4683 | return 0; | |
4684 | } | |
4685 | ||
d397712b | 4686 | static noinline int invalidate_extent_cache(struct btrfs_root *root, |
1a40e23b ZY |
4687 | struct extent_buffer *leaf, |
4688 | struct btrfs_block_group_cache *group, | |
4689 | struct btrfs_root *target_root) | |
4690 | { | |
4691 | struct btrfs_key key; | |
4692 | struct inode *inode = NULL; | |
4693 | struct btrfs_file_extent_item *fi; | |
4694 | u64 num_bytes; | |
4695 | u64 skip_objectid = 0; | |
4696 | u32 nritems; | |
4697 | u32 i; | |
4698 | ||
4699 | nritems = btrfs_header_nritems(leaf); | |
4700 | for (i = 0; i < nritems; i++) { | |
4701 | btrfs_item_key_to_cpu(leaf, &key, i); | |
4702 | if (key.objectid == skip_objectid || | |
4703 | key.type != BTRFS_EXTENT_DATA_KEY) | |
4704 | continue; | |
4705 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
4706 | if (btrfs_file_extent_type(leaf, fi) == | |
4707 | BTRFS_FILE_EXTENT_INLINE) | |
4708 | continue; | |
4709 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
4710 | continue; | |
4711 | if (!inode || inode->i_ino != key.objectid) { | |
4712 | iput(inode); | |
4713 | inode = btrfs_ilookup(target_root->fs_info->sb, | |
4714 | key.objectid, target_root, 1); | |
4715 | } | |
4716 | if (!inode) { | |
4717 | skip_objectid = key.objectid; | |
4718 | continue; | |
4719 | } | |
4720 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
4721 | ||
4722 | lock_extent(&BTRFS_I(inode)->io_tree, key.offset, | |
4723 | key.offset + num_bytes - 1, GFP_NOFS); | |
1a40e23b ZY |
4724 | btrfs_drop_extent_cache(inode, key.offset, |
4725 | key.offset + num_bytes - 1, 1); | |
1a40e23b ZY |
4726 | unlock_extent(&BTRFS_I(inode)->io_tree, key.offset, |
4727 | key.offset + num_bytes - 1, GFP_NOFS); | |
4728 | cond_resched(); | |
4729 | } | |
4730 | iput(inode); | |
4731 | return 0; | |
4732 | } | |
4733 | ||
d397712b | 4734 | static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
4735 | struct btrfs_root *root, |
4736 | struct extent_buffer *leaf, | |
4737 | struct btrfs_block_group_cache *group, | |
4738 | struct inode *reloc_inode) | |
4739 | { | |
4740 | struct btrfs_key key; | |
4741 | struct btrfs_key extent_key; | |
4742 | struct btrfs_file_extent_item *fi; | |
4743 | struct btrfs_leaf_ref *ref; | |
4744 | struct disk_extent *new_extent; | |
4745 | u64 bytenr; | |
4746 | u64 num_bytes; | |
4747 | u32 nritems; | |
4748 | u32 i; | |
4749 | int ext_index; | |
4750 | int nr_extent; | |
4751 | int ret; | |
4752 | ||
4753 | new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS); | |
4754 | BUG_ON(!new_extent); | |
4755 | ||
4756 | ref = btrfs_lookup_leaf_ref(root, leaf->start); | |
4757 | BUG_ON(!ref); | |
4758 | ||
4759 | ext_index = -1; | |
4760 | nritems = btrfs_header_nritems(leaf); | |
4761 | for (i = 0; i < nritems; i++) { | |
4762 | btrfs_item_key_to_cpu(leaf, &key, i); | |
4763 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
4764 | continue; | |
4765 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
4766 | if (btrfs_file_extent_type(leaf, fi) == | |
4767 | BTRFS_FILE_EXTENT_INLINE) | |
4768 | continue; | |
4769 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
4770 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
4771 | if (bytenr == 0) | |
4772 | continue; | |
4773 | ||
4774 | ext_index++; | |
4775 | if (bytenr >= group->key.objectid + group->key.offset || | |
4776 | bytenr + num_bytes <= group->key.objectid) | |
4777 | continue; | |
4778 | ||
4779 | extent_key.objectid = bytenr; | |
4780 | extent_key.offset = num_bytes; | |
4781 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
4782 | nr_extent = 1; | |
4783 | ret = get_new_locations(reloc_inode, &extent_key, | |
4784 | group->key.objectid, 1, | |
4785 | &new_extent, &nr_extent); | |
4786 | if (ret > 0) | |
4787 | continue; | |
4788 | BUG_ON(ret < 0); | |
4789 | ||
4790 | BUG_ON(ref->extents[ext_index].bytenr != bytenr); | |
4791 | BUG_ON(ref->extents[ext_index].num_bytes != num_bytes); | |
4792 | ref->extents[ext_index].bytenr = new_extent->disk_bytenr; | |
4793 | ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes; | |
4794 | ||
1a40e23b ZY |
4795 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
4796 | new_extent->disk_bytenr); | |
4797 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
4798 | new_extent->disk_num_bytes); | |
1a40e23b ZY |
4799 | btrfs_mark_buffer_dirty(leaf); |
4800 | ||
4801 | ret = btrfs_inc_extent_ref(trans, root, | |
4802 | new_extent->disk_bytenr, | |
4803 | new_extent->disk_num_bytes, | |
4804 | leaf->start, | |
4805 | root->root_key.objectid, | |
3bb1a1bc | 4806 | trans->transid, key.objectid); |
1a40e23b | 4807 | BUG_ON(ret); |
56bec294 | 4808 | |
1a40e23b ZY |
4809 | ret = btrfs_free_extent(trans, root, |
4810 | bytenr, num_bytes, leaf->start, | |
4811 | btrfs_header_owner(leaf), | |
4812 | btrfs_header_generation(leaf), | |
3bb1a1bc | 4813 | key.objectid, 0); |
1a40e23b ZY |
4814 | BUG_ON(ret); |
4815 | cond_resched(); | |
4816 | } | |
4817 | kfree(new_extent); | |
4818 | BUG_ON(ext_index + 1 != ref->nritems); | |
4819 | btrfs_free_leaf_ref(root, ref); | |
4820 | return 0; | |
4821 | } | |
4822 | ||
f82d02d9 YZ |
4823 | int btrfs_free_reloc_root(struct btrfs_trans_handle *trans, |
4824 | struct btrfs_root *root) | |
1a40e23b ZY |
4825 | { |
4826 | struct btrfs_root *reloc_root; | |
f82d02d9 | 4827 | int ret; |
1a40e23b ZY |
4828 | |
4829 | if (root->reloc_root) { | |
4830 | reloc_root = root->reloc_root; | |
4831 | root->reloc_root = NULL; | |
4832 | list_add(&reloc_root->dead_list, | |
4833 | &root->fs_info->dead_reloc_roots); | |
f82d02d9 YZ |
4834 | |
4835 | btrfs_set_root_bytenr(&reloc_root->root_item, | |
4836 | reloc_root->node->start); | |
4837 | btrfs_set_root_level(&root->root_item, | |
4838 | btrfs_header_level(reloc_root->node)); | |
4839 | memset(&reloc_root->root_item.drop_progress, 0, | |
4840 | sizeof(struct btrfs_disk_key)); | |
4841 | reloc_root->root_item.drop_level = 0; | |
4842 | ||
4843 | ret = btrfs_update_root(trans, root->fs_info->tree_root, | |
4844 | &reloc_root->root_key, | |
4845 | &reloc_root->root_item); | |
4846 | BUG_ON(ret); | |
1a40e23b ZY |
4847 | } |
4848 | return 0; | |
4849 | } | |
4850 | ||
4851 | int btrfs_drop_dead_reloc_roots(struct btrfs_root *root) | |
4852 | { | |
4853 | struct btrfs_trans_handle *trans; | |
4854 | struct btrfs_root *reloc_root; | |
4855 | struct btrfs_root *prev_root = NULL; | |
4856 | struct list_head dead_roots; | |
4857 | int ret; | |
4858 | unsigned long nr; | |
4859 | ||
4860 | INIT_LIST_HEAD(&dead_roots); | |
4861 | list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots); | |
4862 | ||
4863 | while (!list_empty(&dead_roots)) { | |
4864 | reloc_root = list_entry(dead_roots.prev, | |
4865 | struct btrfs_root, dead_list); | |
4866 | list_del_init(&reloc_root->dead_list); | |
4867 | ||
4868 | BUG_ON(reloc_root->commit_root != NULL); | |
4869 | while (1) { | |
4870 | trans = btrfs_join_transaction(root, 1); | |
4871 | BUG_ON(!trans); | |
4872 | ||
4873 | mutex_lock(&root->fs_info->drop_mutex); | |
4874 | ret = btrfs_drop_snapshot(trans, reloc_root); | |
4875 | if (ret != -EAGAIN) | |
4876 | break; | |
4877 | mutex_unlock(&root->fs_info->drop_mutex); | |
4878 | ||
4879 | nr = trans->blocks_used; | |
4880 | ret = btrfs_end_transaction(trans, root); | |
4881 | BUG_ON(ret); | |
4882 | btrfs_btree_balance_dirty(root, nr); | |
4883 | } | |
4884 | ||
4885 | free_extent_buffer(reloc_root->node); | |
4886 | ||
4887 | ret = btrfs_del_root(trans, root->fs_info->tree_root, | |
4888 | &reloc_root->root_key); | |
4889 | BUG_ON(ret); | |
4890 | mutex_unlock(&root->fs_info->drop_mutex); | |
4891 | ||
4892 | nr = trans->blocks_used; | |
4893 | ret = btrfs_end_transaction(trans, root); | |
4894 | BUG_ON(ret); | |
4895 | btrfs_btree_balance_dirty(root, nr); | |
4896 | ||
4897 | kfree(prev_root); | |
4898 | prev_root = reloc_root; | |
4899 | } | |
4900 | if (prev_root) { | |
4901 | btrfs_remove_leaf_refs(prev_root, (u64)-1, 0); | |
4902 | kfree(prev_root); | |
4903 | } | |
4904 | return 0; | |
4905 | } | |
4906 | ||
4907 | int btrfs_add_dead_reloc_root(struct btrfs_root *root) | |
4908 | { | |
4909 | list_add(&root->dead_list, &root->fs_info->dead_reloc_roots); | |
4910 | return 0; | |
4911 | } | |
4912 | ||
4913 | int btrfs_cleanup_reloc_trees(struct btrfs_root *root) | |
4914 | { | |
4915 | struct btrfs_root *reloc_root; | |
4916 | struct btrfs_trans_handle *trans; | |
4917 | struct btrfs_key location; | |
4918 | int found; | |
4919 | int ret; | |
4920 | ||
4921 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
4922 | ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL); | |
4923 | BUG_ON(ret); | |
4924 | found = !list_empty(&root->fs_info->dead_reloc_roots); | |
4925 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
4926 | ||
4927 | if (found) { | |
4928 | trans = btrfs_start_transaction(root, 1); | |
4929 | BUG_ON(!trans); | |
4930 | ret = btrfs_commit_transaction(trans, root); | |
4931 | BUG_ON(ret); | |
4932 | } | |
4933 | ||
4934 | location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
4935 | location.offset = (u64)-1; | |
4936 | location.type = BTRFS_ROOT_ITEM_KEY; | |
4937 | ||
4938 | reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location); | |
4939 | BUG_ON(!reloc_root); | |
4940 | btrfs_orphan_cleanup(reloc_root); | |
4941 | return 0; | |
4942 | } | |
4943 | ||
d397712b | 4944 | static noinline int init_reloc_tree(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
4945 | struct btrfs_root *root) |
4946 | { | |
4947 | struct btrfs_root *reloc_root; | |
4948 | struct extent_buffer *eb; | |
4949 | struct btrfs_root_item *root_item; | |
4950 | struct btrfs_key root_key; | |
4951 | int ret; | |
4952 | ||
4953 | BUG_ON(!root->ref_cows); | |
4954 | if (root->reloc_root) | |
4955 | return 0; | |
4956 | ||
4957 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); | |
4958 | BUG_ON(!root_item); | |
4959 | ||
4960 | ret = btrfs_copy_root(trans, root, root->commit_root, | |
4961 | &eb, BTRFS_TREE_RELOC_OBJECTID); | |
4962 | BUG_ON(ret); | |
4963 | ||
4964 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
4965 | root_key.offset = root->root_key.objectid; | |
4966 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
4967 | ||
4968 | memcpy(root_item, &root->root_item, sizeof(root_item)); | |
4969 | btrfs_set_root_refs(root_item, 0); | |
4970 | btrfs_set_root_bytenr(root_item, eb->start); | |
4971 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
84234f3a | 4972 | btrfs_set_root_generation(root_item, trans->transid); |
1a40e23b ZY |
4973 | |
4974 | btrfs_tree_unlock(eb); | |
4975 | free_extent_buffer(eb); | |
4976 | ||
4977 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, | |
4978 | &root_key, root_item); | |
4979 | BUG_ON(ret); | |
4980 | kfree(root_item); | |
4981 | ||
4982 | reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root, | |
4983 | &root_key); | |
4984 | BUG_ON(!reloc_root); | |
4985 | reloc_root->last_trans = trans->transid; | |
4986 | reloc_root->commit_root = NULL; | |
4987 | reloc_root->ref_tree = &root->fs_info->reloc_ref_tree; | |
4988 | ||
4989 | root->reloc_root = reloc_root; | |
4990 | return 0; | |
4991 | } | |
4992 | ||
4993 | /* | |
4994 | * Core function of space balance. | |
4995 | * | |
4996 | * The idea is using reloc trees to relocate tree blocks in reference | |
f82d02d9 YZ |
4997 | * counted roots. There is one reloc tree for each subvol, and all |
4998 | * reloc trees share same root key objectid. Reloc trees are snapshots | |
4999 | * of the latest committed roots of subvols (root->commit_root). | |
5000 | * | |
5001 | * To relocate a tree block referenced by a subvol, there are two steps. | |
5002 | * COW the block through subvol's reloc tree, then update block pointer | |
5003 | * in the subvol to point to the new block. Since all reloc trees share | |
5004 | * same root key objectid, doing special handing for tree blocks owned | |
5005 | * by them is easy. Once a tree block has been COWed in one reloc tree, | |
5006 | * we can use the resulting new block directly when the same block is | |
5007 | * required to COW again through other reloc trees. By this way, relocated | |
5008 | * tree blocks are shared between reloc trees, so they are also shared | |
5009 | * between subvols. | |
1a40e23b | 5010 | */ |
d397712b | 5011 | static noinline int relocate_one_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
5012 | struct btrfs_root *root, |
5013 | struct btrfs_path *path, | |
5014 | struct btrfs_key *first_key, | |
5015 | struct btrfs_ref_path *ref_path, | |
5016 | struct btrfs_block_group_cache *group, | |
5017 | struct inode *reloc_inode) | |
5018 | { | |
5019 | struct btrfs_root *reloc_root; | |
5020 | struct extent_buffer *eb = NULL; | |
5021 | struct btrfs_key *keys; | |
5022 | u64 *nodes; | |
5023 | int level; | |
f82d02d9 | 5024 | int shared_level; |
1a40e23b | 5025 | int lowest_level = 0; |
1a40e23b ZY |
5026 | int ret; |
5027 | ||
5028 | if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
5029 | lowest_level = ref_path->owner_objectid; | |
5030 | ||
f82d02d9 | 5031 | if (!root->ref_cows) { |
1a40e23b ZY |
5032 | path->lowest_level = lowest_level; |
5033 | ret = btrfs_search_slot(trans, root, first_key, path, 0, 1); | |
5034 | BUG_ON(ret < 0); | |
5035 | path->lowest_level = 0; | |
5036 | btrfs_release_path(root, path); | |
5037 | return 0; | |
5038 | } | |
5039 | ||
1a40e23b ZY |
5040 | mutex_lock(&root->fs_info->tree_reloc_mutex); |
5041 | ret = init_reloc_tree(trans, root); | |
5042 | BUG_ON(ret); | |
5043 | reloc_root = root->reloc_root; | |
5044 | ||
f82d02d9 YZ |
5045 | shared_level = ref_path->shared_level; |
5046 | ref_path->shared_level = BTRFS_MAX_LEVEL - 1; | |
1a40e23b | 5047 | |
f82d02d9 YZ |
5048 | keys = ref_path->node_keys; |
5049 | nodes = ref_path->new_nodes; | |
5050 | memset(&keys[shared_level + 1], 0, | |
5051 | sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
5052 | memset(&nodes[shared_level + 1], 0, | |
5053 | sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
1a40e23b | 5054 | |
f82d02d9 YZ |
5055 | if (nodes[lowest_level] == 0) { |
5056 | path->lowest_level = lowest_level; | |
5057 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, | |
5058 | 0, 1); | |
5059 | BUG_ON(ret); | |
5060 | for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) { | |
5061 | eb = path->nodes[level]; | |
5062 | if (!eb || eb == reloc_root->node) | |
5063 | break; | |
5064 | nodes[level] = eb->start; | |
5065 | if (level == 0) | |
5066 | btrfs_item_key_to_cpu(eb, &keys[level], 0); | |
5067 | else | |
5068 | btrfs_node_key_to_cpu(eb, &keys[level], 0); | |
5069 | } | |
2b82032c YZ |
5070 | if (nodes[0] && |
5071 | ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
5072 | eb = path->nodes[0]; |
5073 | ret = replace_extents_in_leaf(trans, reloc_root, eb, | |
5074 | group, reloc_inode); | |
5075 | BUG_ON(ret); | |
5076 | } | |
5077 | btrfs_release_path(reloc_root, path); | |
5078 | } else { | |
1a40e23b | 5079 | ret = btrfs_merge_path(trans, reloc_root, keys, nodes, |
f82d02d9 | 5080 | lowest_level); |
1a40e23b ZY |
5081 | BUG_ON(ret); |
5082 | } | |
5083 | ||
1a40e23b ZY |
5084 | /* |
5085 | * replace tree blocks in the fs tree with tree blocks in | |
5086 | * the reloc tree. | |
5087 | */ | |
5088 | ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level); | |
5089 | BUG_ON(ret < 0); | |
5090 | ||
5091 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
5092 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, |
5093 | 0, 0); | |
5094 | BUG_ON(ret); | |
5095 | extent_buffer_get(path->nodes[0]); | |
5096 | eb = path->nodes[0]; | |
5097 | btrfs_release_path(reloc_root, path); | |
1a40e23b ZY |
5098 | ret = invalidate_extent_cache(reloc_root, eb, group, root); |
5099 | BUG_ON(ret); | |
5100 | free_extent_buffer(eb); | |
5101 | } | |
1a40e23b | 5102 | |
f82d02d9 | 5103 | mutex_unlock(&root->fs_info->tree_reloc_mutex); |
1a40e23b | 5104 | path->lowest_level = 0; |
1a40e23b ZY |
5105 | return 0; |
5106 | } | |
5107 | ||
d397712b | 5108 | static noinline int relocate_tree_block(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
5109 | struct btrfs_root *root, |
5110 | struct btrfs_path *path, | |
5111 | struct btrfs_key *first_key, | |
5112 | struct btrfs_ref_path *ref_path) | |
5113 | { | |
5114 | int ret; | |
1a40e23b ZY |
5115 | |
5116 | ret = relocate_one_path(trans, root, path, first_key, | |
5117 | ref_path, NULL, NULL); | |
5118 | BUG_ON(ret); | |
5119 | ||
1a40e23b ZY |
5120 | return 0; |
5121 | } | |
5122 | ||
d397712b | 5123 | static noinline int del_extent_zero(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
5124 | struct btrfs_root *extent_root, |
5125 | struct btrfs_path *path, | |
5126 | struct btrfs_key *extent_key) | |
5127 | { | |
5128 | int ret; | |
5129 | ||
1a40e23b ZY |
5130 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); |
5131 | if (ret) | |
5132 | goto out; | |
5133 | ret = btrfs_del_item(trans, extent_root, path); | |
5134 | out: | |
5135 | btrfs_release_path(extent_root, path); | |
1a40e23b ZY |
5136 | return ret; |
5137 | } | |
5138 | ||
d397712b | 5139 | static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info, |
1a40e23b ZY |
5140 | struct btrfs_ref_path *ref_path) |
5141 | { | |
5142 | struct btrfs_key root_key; | |
5143 | ||
5144 | root_key.objectid = ref_path->root_objectid; | |
5145 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
5146 | if (is_cowonly_root(ref_path->root_objectid)) | |
5147 | root_key.offset = 0; | |
5148 | else | |
5149 | root_key.offset = (u64)-1; | |
5150 | ||
5151 | return btrfs_read_fs_root_no_name(fs_info, &root_key); | |
5152 | } | |
5153 | ||
d397712b | 5154 | static noinline int relocate_one_extent(struct btrfs_root *extent_root, |
1a40e23b ZY |
5155 | struct btrfs_path *path, |
5156 | struct btrfs_key *extent_key, | |
5157 | struct btrfs_block_group_cache *group, | |
5158 | struct inode *reloc_inode, int pass) | |
5159 | { | |
5160 | struct btrfs_trans_handle *trans; | |
5161 | struct btrfs_root *found_root; | |
5162 | struct btrfs_ref_path *ref_path = NULL; | |
5163 | struct disk_extent *new_extents = NULL; | |
5164 | int nr_extents = 0; | |
5165 | int loops; | |
5166 | int ret; | |
5167 | int level; | |
5168 | struct btrfs_key first_key; | |
5169 | u64 prev_block = 0; | |
5170 | ||
1a40e23b ZY |
5171 | |
5172 | trans = btrfs_start_transaction(extent_root, 1); | |
5173 | BUG_ON(!trans); | |
5174 | ||
5175 | if (extent_key->objectid == 0) { | |
5176 | ret = del_extent_zero(trans, extent_root, path, extent_key); | |
5177 | goto out; | |
5178 | } | |
5179 | ||
5180 | ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS); | |
5181 | if (!ref_path) { | |
d397712b CM |
5182 | ret = -ENOMEM; |
5183 | goto out; | |
1a40e23b ZY |
5184 | } |
5185 | ||
5186 | for (loops = 0; ; loops++) { | |
5187 | if (loops == 0) { | |
5188 | ret = btrfs_first_ref_path(trans, extent_root, ref_path, | |
5189 | extent_key->objectid); | |
5190 | } else { | |
5191 | ret = btrfs_next_ref_path(trans, extent_root, ref_path); | |
5192 | } | |
5193 | if (ret < 0) | |
5194 | goto out; | |
5195 | if (ret > 0) | |
5196 | break; | |
5197 | ||
5198 | if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID || | |
5199 | ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
5200 | continue; | |
5201 | ||
5202 | found_root = read_ref_root(extent_root->fs_info, ref_path); | |
5203 | BUG_ON(!found_root); | |
5204 | /* | |
5205 | * for reference counted tree, only process reference paths | |
5206 | * rooted at the latest committed root. | |
5207 | */ | |
5208 | if (found_root->ref_cows && | |
5209 | ref_path->root_generation != found_root->root_key.offset) | |
5210 | continue; | |
5211 | ||
5212 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
5213 | if (pass == 0) { | |
5214 | /* | |
5215 | * copy data extents to new locations | |
5216 | */ | |
5217 | u64 group_start = group->key.objectid; | |
5218 | ret = relocate_data_extent(reloc_inode, | |
5219 | extent_key, | |
5220 | group_start); | |
5221 | if (ret < 0) | |
5222 | goto out; | |
5223 | break; | |
5224 | } | |
5225 | level = 0; | |
5226 | } else { | |
5227 | level = ref_path->owner_objectid; | |
5228 | } | |
5229 | ||
5230 | if (prev_block != ref_path->nodes[level]) { | |
5231 | struct extent_buffer *eb; | |
5232 | u64 block_start = ref_path->nodes[level]; | |
5233 | u64 block_size = btrfs_level_size(found_root, level); | |
5234 | ||
5235 | eb = read_tree_block(found_root, block_start, | |
5236 | block_size, 0); | |
5237 | btrfs_tree_lock(eb); | |
5238 | BUG_ON(level != btrfs_header_level(eb)); | |
5239 | ||
5240 | if (level == 0) | |
5241 | btrfs_item_key_to_cpu(eb, &first_key, 0); | |
5242 | else | |
5243 | btrfs_node_key_to_cpu(eb, &first_key, 0); | |
5244 | ||
5245 | btrfs_tree_unlock(eb); | |
5246 | free_extent_buffer(eb); | |
5247 | prev_block = block_start; | |
5248 | } | |
5249 | ||
24562425 | 5250 | mutex_lock(&extent_root->fs_info->trans_mutex); |
e4404d6e | 5251 | btrfs_record_root_in_trans(found_root); |
24562425 | 5252 | mutex_unlock(&extent_root->fs_info->trans_mutex); |
e4404d6e YZ |
5253 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { |
5254 | /* | |
5255 | * try to update data extent references while | |
5256 | * keeping metadata shared between snapshots. | |
5257 | */ | |
5258 | if (pass == 1) { | |
5259 | ret = relocate_one_path(trans, found_root, | |
5260 | path, &first_key, ref_path, | |
5261 | group, reloc_inode); | |
5262 | if (ret < 0) | |
5263 | goto out; | |
5264 | continue; | |
5265 | } | |
1a40e23b ZY |
5266 | /* |
5267 | * use fallback method to process the remaining | |
5268 | * references. | |
5269 | */ | |
5270 | if (!new_extents) { | |
5271 | u64 group_start = group->key.objectid; | |
d899e052 YZ |
5272 | new_extents = kmalloc(sizeof(*new_extents), |
5273 | GFP_NOFS); | |
5274 | nr_extents = 1; | |
1a40e23b ZY |
5275 | ret = get_new_locations(reloc_inode, |
5276 | extent_key, | |
d899e052 | 5277 | group_start, 1, |
1a40e23b ZY |
5278 | &new_extents, |
5279 | &nr_extents); | |
d899e052 | 5280 | if (ret) |
1a40e23b ZY |
5281 | goto out; |
5282 | } | |
1a40e23b ZY |
5283 | ret = replace_one_extent(trans, found_root, |
5284 | path, extent_key, | |
5285 | &first_key, ref_path, | |
5286 | new_extents, nr_extents); | |
e4404d6e | 5287 | } else { |
1a40e23b ZY |
5288 | ret = relocate_tree_block(trans, found_root, path, |
5289 | &first_key, ref_path); | |
1a40e23b ZY |
5290 | } |
5291 | if (ret < 0) | |
5292 | goto out; | |
5293 | } | |
5294 | ret = 0; | |
5295 | out: | |
5296 | btrfs_end_transaction(trans, extent_root); | |
5297 | kfree(new_extents); | |
5298 | kfree(ref_path); | |
1a40e23b ZY |
5299 | return ret; |
5300 | } | |
5301 | ||
ec44a35c CM |
5302 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
5303 | { | |
5304 | u64 num_devices; | |
5305 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
5306 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
5307 | ||
2b82032c | 5308 | num_devices = root->fs_info->fs_devices->rw_devices; |
ec44a35c CM |
5309 | if (num_devices == 1) { |
5310 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
5311 | stripped = flags & ~stripped; | |
5312 | ||
5313 | /* turn raid0 into single device chunks */ | |
5314 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
5315 | return stripped; | |
5316 | ||
5317 | /* turn mirroring into duplication */ | |
5318 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
5319 | BTRFS_BLOCK_GROUP_RAID10)) | |
5320 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
5321 | return flags; | |
5322 | } else { | |
5323 | /* they already had raid on here, just return */ | |
ec44a35c CM |
5324 | if (flags & stripped) |
5325 | return flags; | |
5326 | ||
5327 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
5328 | stripped = flags & ~stripped; | |
5329 | ||
5330 | /* switch duplicated blocks with raid1 */ | |
5331 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
5332 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
5333 | ||
5334 | /* turn single device chunks into raid0 */ | |
5335 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
5336 | } | |
5337 | return flags; | |
5338 | } | |
5339 | ||
b2950863 | 5340 | static int __alloc_chunk_for_shrink(struct btrfs_root *root, |
0ef3e66b CM |
5341 | struct btrfs_block_group_cache *shrink_block_group, |
5342 | int force) | |
5343 | { | |
5344 | struct btrfs_trans_handle *trans; | |
5345 | u64 new_alloc_flags; | |
5346 | u64 calc; | |
5347 | ||
c286ac48 | 5348 | spin_lock(&shrink_block_group->lock); |
0ef3e66b | 5349 | if (btrfs_block_group_used(&shrink_block_group->item) > 0) { |
c286ac48 | 5350 | spin_unlock(&shrink_block_group->lock); |
c286ac48 | 5351 | |
0ef3e66b | 5352 | trans = btrfs_start_transaction(root, 1); |
c286ac48 | 5353 | spin_lock(&shrink_block_group->lock); |
7d9eb12c | 5354 | |
0ef3e66b CM |
5355 | new_alloc_flags = update_block_group_flags(root, |
5356 | shrink_block_group->flags); | |
5357 | if (new_alloc_flags != shrink_block_group->flags) { | |
5358 | calc = | |
5359 | btrfs_block_group_used(&shrink_block_group->item); | |
5360 | } else { | |
5361 | calc = shrink_block_group->key.offset; | |
5362 | } | |
c286ac48 CM |
5363 | spin_unlock(&shrink_block_group->lock); |
5364 | ||
0ef3e66b CM |
5365 | do_chunk_alloc(trans, root->fs_info->extent_root, |
5366 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
7d9eb12c | 5367 | |
0ef3e66b | 5368 | btrfs_end_transaction(trans, root); |
c286ac48 CM |
5369 | } else |
5370 | spin_unlock(&shrink_block_group->lock); | |
0ef3e66b CM |
5371 | return 0; |
5372 | } | |
5373 | ||
1a40e23b ZY |
5374 | static int __insert_orphan_inode(struct btrfs_trans_handle *trans, |
5375 | struct btrfs_root *root, | |
5376 | u64 objectid, u64 size) | |
5377 | { | |
5378 | struct btrfs_path *path; | |
5379 | struct btrfs_inode_item *item; | |
5380 | struct extent_buffer *leaf; | |
5381 | int ret; | |
5382 | ||
5383 | path = btrfs_alloc_path(); | |
5384 | if (!path) | |
5385 | return -ENOMEM; | |
5386 | ||
b9473439 | 5387 | path->leave_spinning = 1; |
1a40e23b ZY |
5388 | ret = btrfs_insert_empty_inode(trans, root, path, objectid); |
5389 | if (ret) | |
5390 | goto out; | |
5391 | ||
5392 | leaf = path->nodes[0]; | |
5393 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); | |
5394 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
5395 | btrfs_set_inode_generation(leaf, item, 1); | |
5396 | btrfs_set_inode_size(leaf, item, size); | |
5397 | btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); | |
0403e47e | 5398 | btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS); |
1a40e23b ZY |
5399 | btrfs_mark_buffer_dirty(leaf); |
5400 | btrfs_release_path(root, path); | |
5401 | out: | |
5402 | btrfs_free_path(path); | |
5403 | return ret; | |
5404 | } | |
5405 | ||
d397712b | 5406 | static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, |
1a40e23b ZY |
5407 | struct btrfs_block_group_cache *group) |
5408 | { | |
5409 | struct inode *inode = NULL; | |
5410 | struct btrfs_trans_handle *trans; | |
5411 | struct btrfs_root *root; | |
5412 | struct btrfs_key root_key; | |
5413 | u64 objectid = BTRFS_FIRST_FREE_OBJECTID; | |
5414 | int err = 0; | |
5415 | ||
5416 | root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
5417 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
5418 | root_key.offset = (u64)-1; | |
5419 | root = btrfs_read_fs_root_no_name(fs_info, &root_key); | |
5420 | if (IS_ERR(root)) | |
5421 | return ERR_CAST(root); | |
5422 | ||
5423 | trans = btrfs_start_transaction(root, 1); | |
5424 | BUG_ON(!trans); | |
5425 | ||
5426 | err = btrfs_find_free_objectid(trans, root, objectid, &objectid); | |
5427 | if (err) | |
5428 | goto out; | |
5429 | ||
5430 | err = __insert_orphan_inode(trans, root, objectid, group->key.offset); | |
5431 | BUG_ON(err); | |
5432 | ||
5433 | err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0, | |
c8b97818 CM |
5434 | group->key.offset, 0, group->key.offset, |
5435 | 0, 0, 0); | |
1a40e23b ZY |
5436 | BUG_ON(err); |
5437 | ||
5438 | inode = btrfs_iget_locked(root->fs_info->sb, objectid, root); | |
5439 | if (inode->i_state & I_NEW) { | |
5440 | BTRFS_I(inode)->root = root; | |
5441 | BTRFS_I(inode)->location.objectid = objectid; | |
5442 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
5443 | BTRFS_I(inode)->location.offset = 0; | |
5444 | btrfs_read_locked_inode(inode); | |
5445 | unlock_new_inode(inode); | |
5446 | BUG_ON(is_bad_inode(inode)); | |
5447 | } else { | |
5448 | BUG_ON(1); | |
5449 | } | |
17d217fe | 5450 | BTRFS_I(inode)->index_cnt = group->key.objectid; |
1a40e23b ZY |
5451 | |
5452 | err = btrfs_orphan_add(trans, inode); | |
5453 | out: | |
5454 | btrfs_end_transaction(trans, root); | |
5455 | if (err) { | |
5456 | if (inode) | |
5457 | iput(inode); | |
5458 | inode = ERR_PTR(err); | |
5459 | } | |
5460 | return inode; | |
5461 | } | |
5462 | ||
17d217fe YZ |
5463 | int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) |
5464 | { | |
5465 | ||
5466 | struct btrfs_ordered_sum *sums; | |
5467 | struct btrfs_sector_sum *sector_sum; | |
5468 | struct btrfs_ordered_extent *ordered; | |
5469 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5470 | struct list_head list; | |
5471 | size_t offset; | |
5472 | int ret; | |
5473 | u64 disk_bytenr; | |
5474 | ||
5475 | INIT_LIST_HEAD(&list); | |
5476 | ||
5477 | ordered = btrfs_lookup_ordered_extent(inode, file_pos); | |
5478 | BUG_ON(ordered->file_offset != file_pos || ordered->len != len); | |
5479 | ||
5480 | disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; | |
07d400a6 | 5481 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr, |
17d217fe YZ |
5482 | disk_bytenr + len - 1, &list); |
5483 | ||
5484 | while (!list_empty(&list)) { | |
5485 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
5486 | list_del_init(&sums->list); | |
5487 | ||
5488 | sector_sum = sums->sums; | |
5489 | sums->bytenr = ordered->start; | |
5490 | ||
5491 | offset = 0; | |
5492 | while (offset < sums->len) { | |
5493 | sector_sum->bytenr += ordered->start - disk_bytenr; | |
5494 | sector_sum++; | |
5495 | offset += root->sectorsize; | |
5496 | } | |
5497 | ||
5498 | btrfs_add_ordered_sum(inode, ordered, sums); | |
5499 | } | |
5500 | btrfs_put_ordered_extent(ordered); | |
5501 | return 0; | |
5502 | } | |
5503 | ||
1a40e23b | 5504 | int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) |
edbd8d4e CM |
5505 | { |
5506 | struct btrfs_trans_handle *trans; | |
edbd8d4e | 5507 | struct btrfs_path *path; |
1a40e23b ZY |
5508 | struct btrfs_fs_info *info = root->fs_info; |
5509 | struct extent_buffer *leaf; | |
5510 | struct inode *reloc_inode; | |
5511 | struct btrfs_block_group_cache *block_group; | |
5512 | struct btrfs_key key; | |
d899e052 | 5513 | u64 skipped; |
edbd8d4e CM |
5514 | u64 cur_byte; |
5515 | u64 total_found; | |
edbd8d4e CM |
5516 | u32 nritems; |
5517 | int ret; | |
a061fc8d | 5518 | int progress; |
1a40e23b | 5519 | int pass = 0; |
edbd8d4e | 5520 | |
1a40e23b ZY |
5521 | root = root->fs_info->extent_root; |
5522 | ||
5523 | block_group = btrfs_lookup_block_group(info, group_start); | |
5524 | BUG_ON(!block_group); | |
8f18cf13 | 5525 | |
d397712b | 5526 | printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n", |
1a40e23b ZY |
5527 | (unsigned long long)block_group->key.objectid, |
5528 | (unsigned long long)block_group->flags); | |
8f18cf13 | 5529 | |
edbd8d4e | 5530 | path = btrfs_alloc_path(); |
1a40e23b | 5531 | BUG_ON(!path); |
edbd8d4e | 5532 | |
1a40e23b ZY |
5533 | reloc_inode = create_reloc_inode(info, block_group); |
5534 | BUG_ON(IS_ERR(reloc_inode)); | |
323da79c | 5535 | |
1a40e23b | 5536 | __alloc_chunk_for_shrink(root, block_group, 1); |
c146afad | 5537 | set_block_group_readonly(block_group); |
323da79c | 5538 | |
1a40e23b ZY |
5539 | btrfs_start_delalloc_inodes(info->tree_root); |
5540 | btrfs_wait_ordered_extents(info->tree_root, 0); | |
5541 | again: | |
d899e052 | 5542 | skipped = 0; |
edbd8d4e | 5543 | total_found = 0; |
a061fc8d | 5544 | progress = 0; |
1a40e23b | 5545 | key.objectid = block_group->key.objectid; |
edbd8d4e CM |
5546 | key.offset = 0; |
5547 | key.type = 0; | |
73e48b27 | 5548 | cur_byte = key.objectid; |
4313b399 | 5549 | |
1a40e23b ZY |
5550 | trans = btrfs_start_transaction(info->tree_root, 1); |
5551 | btrfs_commit_transaction(trans, info->tree_root); | |
ea8c2819 | 5552 | |
1a40e23b ZY |
5553 | mutex_lock(&root->fs_info->cleaner_mutex); |
5554 | btrfs_clean_old_snapshots(info->tree_root); | |
5555 | btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); | |
5556 | mutex_unlock(&root->fs_info->cleaner_mutex); | |
ea8c2819 | 5557 | |
56bec294 CM |
5558 | trans = btrfs_start_transaction(info->tree_root, 1); |
5559 | btrfs_commit_transaction(trans, info->tree_root); | |
5560 | ||
d397712b | 5561 | while (1) { |
edbd8d4e CM |
5562 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5563 | if (ret < 0) | |
5564 | goto out; | |
7d9eb12c | 5565 | next: |
edbd8d4e | 5566 | leaf = path->nodes[0]; |
73e48b27 | 5567 | nritems = btrfs_header_nritems(leaf); |
73e48b27 Y |
5568 | if (path->slots[0] >= nritems) { |
5569 | ret = btrfs_next_leaf(root, path); | |
5570 | if (ret < 0) | |
5571 | goto out; | |
5572 | if (ret == 1) { | |
5573 | ret = 0; | |
5574 | break; | |
edbd8d4e | 5575 | } |
73e48b27 Y |
5576 | leaf = path->nodes[0]; |
5577 | nritems = btrfs_header_nritems(leaf); | |
edbd8d4e | 5578 | } |
73e48b27 | 5579 | |
1a40e23b | 5580 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
725c8463 | 5581 | |
1a40e23b ZY |
5582 | if (key.objectid >= block_group->key.objectid + |
5583 | block_group->key.offset) | |
8f18cf13 CM |
5584 | break; |
5585 | ||
725c8463 | 5586 | if (progress && need_resched()) { |
725c8463 | 5587 | btrfs_release_path(root, path); |
1a40e23b | 5588 | cond_resched(); |
725c8463 | 5589 | progress = 0; |
1a40e23b | 5590 | continue; |
725c8463 CM |
5591 | } |
5592 | progress = 1; | |
5593 | ||
1a40e23b ZY |
5594 | if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY || |
5595 | key.objectid + key.offset <= cur_byte) { | |
edbd8d4e | 5596 | path->slots[0]++; |
edbd8d4e CM |
5597 | goto next; |
5598 | } | |
73e48b27 | 5599 | |
edbd8d4e | 5600 | total_found++; |
1a40e23b | 5601 | cur_byte = key.objectid + key.offset; |
edbd8d4e | 5602 | btrfs_release_path(root, path); |
edbd8d4e | 5603 | |
1a40e23b ZY |
5604 | __alloc_chunk_for_shrink(root, block_group, 0); |
5605 | ret = relocate_one_extent(root, path, &key, block_group, | |
5606 | reloc_inode, pass); | |
5607 | BUG_ON(ret < 0); | |
d899e052 YZ |
5608 | if (ret > 0) |
5609 | skipped++; | |
edbd8d4e | 5610 | |
1a40e23b ZY |
5611 | key.objectid = cur_byte; |
5612 | key.type = 0; | |
5613 | key.offset = 0; | |
edbd8d4e CM |
5614 | } |
5615 | ||
1a40e23b | 5616 | btrfs_release_path(root, path); |
7d9eb12c | 5617 | |
1a40e23b ZY |
5618 | if (pass == 0) { |
5619 | btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1); | |
5620 | invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1); | |
8e8a1e31 | 5621 | } |
73e48b27 | 5622 | |
1a40e23b | 5623 | if (total_found > 0) { |
d397712b | 5624 | printk(KERN_INFO "btrfs found %llu extents in pass %d\n", |
1a40e23b ZY |
5625 | (unsigned long long)total_found, pass); |
5626 | pass++; | |
d899e052 YZ |
5627 | if (total_found == skipped && pass > 2) { |
5628 | iput(reloc_inode); | |
5629 | reloc_inode = create_reloc_inode(info, block_group); | |
5630 | pass = 0; | |
5631 | } | |
1a40e23b | 5632 | goto again; |
0f9dd46c | 5633 | } |
0ef3e66b | 5634 | |
1a40e23b ZY |
5635 | /* delete reloc_inode */ |
5636 | iput(reloc_inode); | |
5637 | ||
5638 | /* unpin extents in this range */ | |
5639 | trans = btrfs_start_transaction(info->tree_root, 1); | |
5640 | btrfs_commit_transaction(trans, info->tree_root); | |
0ef3e66b | 5641 | |
1a40e23b ZY |
5642 | spin_lock(&block_group->lock); |
5643 | WARN_ON(block_group->pinned > 0); | |
5644 | WARN_ON(block_group->reserved > 0); | |
5645 | WARN_ON(btrfs_block_group_used(&block_group->item) > 0); | |
5646 | spin_unlock(&block_group->lock); | |
d2fb3437 | 5647 | put_block_group(block_group); |
1a40e23b | 5648 | ret = 0; |
edbd8d4e | 5649 | out: |
1a40e23b | 5650 | btrfs_free_path(path); |
edbd8d4e CM |
5651 | return ret; |
5652 | } | |
5653 | ||
b2950863 CH |
5654 | static int find_first_block_group(struct btrfs_root *root, |
5655 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 5656 | { |
925baedd | 5657 | int ret = 0; |
0b86a832 CM |
5658 | struct btrfs_key found_key; |
5659 | struct extent_buffer *leaf; | |
5660 | int slot; | |
edbd8d4e | 5661 | |
0b86a832 CM |
5662 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
5663 | if (ret < 0) | |
925baedd CM |
5664 | goto out; |
5665 | ||
d397712b | 5666 | while (1) { |
0b86a832 | 5667 | slot = path->slots[0]; |
edbd8d4e | 5668 | leaf = path->nodes[0]; |
0b86a832 CM |
5669 | if (slot >= btrfs_header_nritems(leaf)) { |
5670 | ret = btrfs_next_leaf(root, path); | |
5671 | if (ret == 0) | |
5672 | continue; | |
5673 | if (ret < 0) | |
925baedd | 5674 | goto out; |
0b86a832 | 5675 | break; |
edbd8d4e | 5676 | } |
0b86a832 | 5677 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 5678 | |
0b86a832 | 5679 | if (found_key.objectid >= key->objectid && |
925baedd CM |
5680 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5681 | ret = 0; | |
5682 | goto out; | |
5683 | } | |
0b86a832 | 5684 | path->slots[0]++; |
edbd8d4e | 5685 | } |
0b86a832 | 5686 | ret = -ENOENT; |
925baedd | 5687 | out: |
0b86a832 | 5688 | return ret; |
edbd8d4e CM |
5689 | } |
5690 | ||
1a40e23b ZY |
5691 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
5692 | { | |
5693 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 5694 | struct btrfs_space_info *space_info; |
1a40e23b ZY |
5695 | struct rb_node *n; |
5696 | ||
1a40e23b ZY |
5697 | spin_lock(&info->block_group_cache_lock); |
5698 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
5699 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
5700 | cache_node); | |
1a40e23b ZY |
5701 | rb_erase(&block_group->cache_node, |
5702 | &info->block_group_cache_tree); | |
d899e052 YZ |
5703 | spin_unlock(&info->block_group_cache_lock); |
5704 | ||
5705 | btrfs_remove_free_space_cache(block_group); | |
80eb234a | 5706 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 5707 | list_del(&block_group->list); |
80eb234a | 5708 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 YZ |
5709 | |
5710 | WARN_ON(atomic_read(&block_group->count) != 1); | |
1a40e23b | 5711 | kfree(block_group); |
d899e052 YZ |
5712 | |
5713 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
5714 | } |
5715 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
5716 | |
5717 | /* now that all the block groups are freed, go through and | |
5718 | * free all the space_info structs. This is only called during | |
5719 | * the final stages of unmount, and so we know nobody is | |
5720 | * using them. We call synchronize_rcu() once before we start, | |
5721 | * just to be on the safe side. | |
5722 | */ | |
5723 | synchronize_rcu(); | |
5724 | ||
5725 | while(!list_empty(&info->space_info)) { | |
5726 | space_info = list_entry(info->space_info.next, | |
5727 | struct btrfs_space_info, | |
5728 | list); | |
5729 | ||
5730 | list_del(&space_info->list); | |
5731 | kfree(space_info); | |
5732 | } | |
1a40e23b ZY |
5733 | return 0; |
5734 | } | |
5735 | ||
9078a3e1 CM |
5736 | int btrfs_read_block_groups(struct btrfs_root *root) |
5737 | { | |
5738 | struct btrfs_path *path; | |
5739 | int ret; | |
9078a3e1 | 5740 | struct btrfs_block_group_cache *cache; |
be744175 | 5741 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 5742 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
5743 | struct btrfs_key key; |
5744 | struct btrfs_key found_key; | |
5f39d397 | 5745 | struct extent_buffer *leaf; |
96b5179d | 5746 | |
be744175 | 5747 | root = info->extent_root; |
9078a3e1 | 5748 | key.objectid = 0; |
0b86a832 | 5749 | key.offset = 0; |
9078a3e1 | 5750 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
5751 | path = btrfs_alloc_path(); |
5752 | if (!path) | |
5753 | return -ENOMEM; | |
5754 | ||
d397712b | 5755 | while (1) { |
0b86a832 CM |
5756 | ret = find_first_block_group(root, path, &key); |
5757 | if (ret > 0) { | |
5758 | ret = 0; | |
5759 | goto error; | |
9078a3e1 | 5760 | } |
0b86a832 CM |
5761 | if (ret != 0) |
5762 | goto error; | |
5763 | ||
5f39d397 CM |
5764 | leaf = path->nodes[0]; |
5765 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 5766 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 5767 | if (!cache) { |
0b86a832 | 5768 | ret = -ENOMEM; |
9078a3e1 CM |
5769 | break; |
5770 | } | |
3e1ad54f | 5771 | |
d2fb3437 | 5772 | atomic_set(&cache->count, 1); |
c286ac48 | 5773 | spin_lock_init(&cache->lock); |
6226cb0a | 5774 | spin_lock_init(&cache->tree_lock); |
ea6a478e | 5775 | mutex_init(&cache->cache_mutex); |
0f9dd46c | 5776 | INIT_LIST_HEAD(&cache->list); |
5f39d397 CM |
5777 | read_extent_buffer(leaf, &cache->item, |
5778 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
5779 | sizeof(cache->item)); | |
9078a3e1 | 5780 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 5781 | |
9078a3e1 CM |
5782 | key.objectid = found_key.objectid + found_key.offset; |
5783 | btrfs_release_path(root, path); | |
0b86a832 | 5784 | cache->flags = btrfs_block_group_flags(&cache->item); |
96b5179d | 5785 | |
6324fbf3 CM |
5786 | ret = update_space_info(info, cache->flags, found_key.offset, |
5787 | btrfs_block_group_used(&cache->item), | |
5788 | &space_info); | |
5789 | BUG_ON(ret); | |
5790 | cache->space_info = space_info; | |
80eb234a JB |
5791 | down_write(&space_info->groups_sem); |
5792 | list_add_tail(&cache->list, &space_info->block_groups); | |
5793 | up_write(&space_info->groups_sem); | |
0f9dd46c JB |
5794 | |
5795 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
5796 | BUG_ON(ret); | |
75ccf47d CM |
5797 | |
5798 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c YZ |
5799 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
5800 | set_block_group_readonly(cache); | |
9078a3e1 | 5801 | } |
0b86a832 CM |
5802 | ret = 0; |
5803 | error: | |
9078a3e1 | 5804 | btrfs_free_path(path); |
0b86a832 | 5805 | return ret; |
9078a3e1 | 5806 | } |
6324fbf3 CM |
5807 | |
5808 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
5809 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 5810 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
5811 | u64 size) |
5812 | { | |
5813 | int ret; | |
6324fbf3 CM |
5814 | struct btrfs_root *extent_root; |
5815 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
5816 | |
5817 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 5818 | |
12fcfd22 | 5819 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 5820 | |
8f18cf13 | 5821 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
5822 | if (!cache) |
5823 | return -ENOMEM; | |
5824 | ||
e17cade2 | 5825 | cache->key.objectid = chunk_offset; |
6324fbf3 | 5826 | cache->key.offset = size; |
d2fb3437 YZ |
5827 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
5828 | atomic_set(&cache->count, 1); | |
c286ac48 | 5829 | spin_lock_init(&cache->lock); |
6226cb0a | 5830 | spin_lock_init(&cache->tree_lock); |
ea6a478e | 5831 | mutex_init(&cache->cache_mutex); |
0f9dd46c | 5832 | INIT_LIST_HEAD(&cache->list); |
0ef3e66b | 5833 | |
6324fbf3 | 5834 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
5835 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
5836 | cache->flags = type; | |
5837 | btrfs_set_block_group_flags(&cache->item, type); | |
5838 | ||
5839 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, | |
5840 | &cache->space_info); | |
5841 | BUG_ON(ret); | |
80eb234a JB |
5842 | down_write(&cache->space_info->groups_sem); |
5843 | list_add_tail(&cache->list, &cache->space_info->block_groups); | |
5844 | up_write(&cache->space_info->groups_sem); | |
6324fbf3 | 5845 | |
0f9dd46c JB |
5846 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
5847 | BUG_ON(ret); | |
c286ac48 | 5848 | |
6324fbf3 CM |
5849 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
5850 | sizeof(cache->item)); | |
5851 | BUG_ON(ret); | |
5852 | ||
d18a2c44 | 5853 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 5854 | |
6324fbf3 CM |
5855 | return 0; |
5856 | } | |
1a40e23b ZY |
5857 | |
5858 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
5859 | struct btrfs_root *root, u64 group_start) | |
5860 | { | |
5861 | struct btrfs_path *path; | |
5862 | struct btrfs_block_group_cache *block_group; | |
5863 | struct btrfs_key key; | |
5864 | int ret; | |
5865 | ||
1a40e23b ZY |
5866 | root = root->fs_info->extent_root; |
5867 | ||
5868 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
5869 | BUG_ON(!block_group); | |
c146afad | 5870 | BUG_ON(!block_group->ro); |
1a40e23b ZY |
5871 | |
5872 | memcpy(&key, &block_group->key, sizeof(key)); | |
5873 | ||
5874 | path = btrfs_alloc_path(); | |
5875 | BUG_ON(!path); | |
5876 | ||
3dfdb934 | 5877 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
5878 | rb_erase(&block_group->cache_node, |
5879 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 YZ |
5880 | spin_unlock(&root->fs_info->block_group_cache_lock); |
5881 | btrfs_remove_free_space_cache(block_group); | |
80eb234a | 5882 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 5883 | list_del(&block_group->list); |
80eb234a | 5884 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 5885 | |
c146afad YZ |
5886 | spin_lock(&block_group->space_info->lock); |
5887 | block_group->space_info->total_bytes -= block_group->key.offset; | |
5888 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
5889 | spin_unlock(&block_group->space_info->lock); | |
2b82032c | 5890 | block_group->space_info->full = 0; |
c146afad | 5891 | |
d2fb3437 YZ |
5892 | put_block_group(block_group); |
5893 | put_block_group(block_group); | |
1a40e23b ZY |
5894 | |
5895 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5896 | if (ret > 0) | |
5897 | ret = -EIO; | |
5898 | if (ret < 0) | |
5899 | goto out; | |
5900 | ||
5901 | ret = btrfs_del_item(trans, root, path); | |
5902 | out: | |
5903 | btrfs_free_path(path); | |
5904 | return ret; | |
5905 | } |