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> |
817d52f8 | 24 | #include <linux/kthread.h> |
4b4e25f2 | 25 | #include "compat.h" |
74493f7a | 26 | #include "hash.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" |
fa9c0d79 | 33 | #include "free-space-cache.h" |
fec577fb | 34 | |
56bec294 CM |
35 | static int update_reserved_extents(struct btrfs_root *root, |
36 | u64 bytenr, u64 num, int reserve); | |
f3465ca4 JB |
37 | static int update_block_group(struct btrfs_trans_handle *trans, |
38 | struct btrfs_root *root, | |
39 | u64 bytenr, u64 num_bytes, int alloc, | |
40 | int mark_free); | |
5d4f98a2 YZ |
41 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
42 | struct btrfs_root *root, | |
43 | u64 bytenr, u64 num_bytes, u64 parent, | |
44 | u64 root_objectid, u64 owner_objectid, | |
45 | u64 owner_offset, int refs_to_drop, | |
46 | struct btrfs_delayed_extent_op *extra_op); | |
47 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
48 | struct extent_buffer *leaf, | |
49 | struct btrfs_extent_item *ei); | |
50 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
51 | struct btrfs_root *root, | |
52 | u64 parent, u64 root_objectid, | |
53 | u64 flags, u64 owner, u64 offset, | |
54 | struct btrfs_key *ins, int ref_mod); | |
55 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
56 | struct btrfs_root *root, | |
57 | u64 parent, u64 root_objectid, | |
58 | u64 flags, struct btrfs_disk_key *key, | |
59 | int level, struct btrfs_key *ins); | |
d548ee51 | 60 | |
6a63209f JB |
61 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
62 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
63 | u64 flags, int force); | |
64 | ||
817d52f8 JB |
65 | static noinline int |
66 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
67 | { | |
68 | smp_mb(); | |
69 | return cache->cached == BTRFS_CACHE_FINISHED; | |
70 | } | |
71 | ||
0f9dd46c JB |
72 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
73 | { | |
74 | return (cache->flags & bits) == bits; | |
75 | } | |
76 | ||
77 | /* | |
78 | * this adds the block group to the fs_info rb tree for the block group | |
79 | * cache | |
80 | */ | |
b2950863 | 81 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
82 | struct btrfs_block_group_cache *block_group) |
83 | { | |
84 | struct rb_node **p; | |
85 | struct rb_node *parent = NULL; | |
86 | struct btrfs_block_group_cache *cache; | |
87 | ||
88 | spin_lock(&info->block_group_cache_lock); | |
89 | p = &info->block_group_cache_tree.rb_node; | |
90 | ||
91 | while (*p) { | |
92 | parent = *p; | |
93 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
94 | cache_node); | |
95 | if (block_group->key.objectid < cache->key.objectid) { | |
96 | p = &(*p)->rb_left; | |
97 | } else if (block_group->key.objectid > cache->key.objectid) { | |
98 | p = &(*p)->rb_right; | |
99 | } else { | |
100 | spin_unlock(&info->block_group_cache_lock); | |
101 | return -EEXIST; | |
102 | } | |
103 | } | |
104 | ||
105 | rb_link_node(&block_group->cache_node, parent, p); | |
106 | rb_insert_color(&block_group->cache_node, | |
107 | &info->block_group_cache_tree); | |
108 | spin_unlock(&info->block_group_cache_lock); | |
109 | ||
110 | return 0; | |
111 | } | |
112 | ||
113 | /* | |
114 | * This will return the block group at or after bytenr if contains is 0, else | |
115 | * it will return the block group that contains the bytenr | |
116 | */ | |
117 | static struct btrfs_block_group_cache * | |
118 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
119 | int contains) | |
120 | { | |
121 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
122 | struct rb_node *n; | |
123 | u64 end, start; | |
124 | ||
125 | spin_lock(&info->block_group_cache_lock); | |
126 | n = info->block_group_cache_tree.rb_node; | |
127 | ||
128 | while (n) { | |
129 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
130 | cache_node); | |
131 | end = cache->key.objectid + cache->key.offset - 1; | |
132 | start = cache->key.objectid; | |
133 | ||
134 | if (bytenr < start) { | |
135 | if (!contains && (!ret || start < ret->key.objectid)) | |
136 | ret = cache; | |
137 | n = n->rb_left; | |
138 | } else if (bytenr > start) { | |
139 | if (contains && bytenr <= end) { | |
140 | ret = cache; | |
141 | break; | |
142 | } | |
143 | n = n->rb_right; | |
144 | } else { | |
145 | ret = cache; | |
146 | break; | |
147 | } | |
148 | } | |
d2fb3437 YZ |
149 | if (ret) |
150 | atomic_inc(&ret->count); | |
0f9dd46c JB |
151 | spin_unlock(&info->block_group_cache_lock); |
152 | ||
153 | return ret; | |
154 | } | |
155 | ||
68b38550 JB |
156 | /* |
157 | * We always set EXTENT_LOCKED for the super mirror extents so we don't | |
158 | * overwrite them, so those bits need to be unset. Also, if we are unmounting | |
159 | * with pinned extents still sitting there because we had a block group caching, | |
160 | * we need to clear those now, since we are done. | |
161 | */ | |
162 | void btrfs_free_pinned_extents(struct btrfs_fs_info *info) | |
817d52f8 JB |
163 | { |
164 | u64 start, end, last = 0; | |
165 | int ret; | |
166 | ||
167 | while (1) { | |
168 | ret = find_first_extent_bit(&info->pinned_extents, last, | |
68b38550 JB |
169 | &start, &end, |
170 | EXTENT_LOCKED|EXTENT_DIRTY); | |
817d52f8 JB |
171 | if (ret) |
172 | break; | |
173 | ||
68b38550 JB |
174 | clear_extent_bits(&info->pinned_extents, start, end, |
175 | EXTENT_LOCKED|EXTENT_DIRTY, GFP_NOFS); | |
817d52f8 JB |
176 | last = end+1; |
177 | } | |
178 | } | |
179 | ||
180 | static int remove_sb_from_cache(struct btrfs_root *root, | |
181 | struct btrfs_block_group_cache *cache) | |
182 | { | |
183 | struct btrfs_fs_info *fs_info = root->fs_info; | |
184 | u64 bytenr; | |
185 | u64 *logical; | |
186 | int stripe_len; | |
187 | int i, nr, ret; | |
188 | ||
189 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
190 | bytenr = btrfs_sb_offset(i); | |
191 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
192 | cache->key.objectid, bytenr, | |
193 | 0, &logical, &nr, &stripe_len); | |
194 | BUG_ON(ret); | |
195 | while (nr--) { | |
196 | try_lock_extent(&fs_info->pinned_extents, | |
197 | logical[nr], | |
198 | logical[nr] + stripe_len - 1, GFP_NOFS); | |
199 | } | |
200 | kfree(logical); | |
201 | } | |
202 | ||
203 | return 0; | |
204 | } | |
205 | ||
0f9dd46c JB |
206 | /* |
207 | * this is only called by cache_block_group, since we could have freed extents | |
208 | * we need to check the pinned_extents for any extents that can't be used yet | |
209 | * since their free space will be released as soon as the transaction commits. | |
210 | */ | |
817d52f8 | 211 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
212 | struct btrfs_fs_info *info, u64 start, u64 end) |
213 | { | |
817d52f8 | 214 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
215 | int ret; |
216 | ||
217 | while (start < end) { | |
218 | ret = find_first_extent_bit(&info->pinned_extents, start, | |
219 | &extent_start, &extent_end, | |
68b38550 | 220 | EXTENT_DIRTY|EXTENT_LOCKED); |
0f9dd46c JB |
221 | if (ret) |
222 | break; | |
223 | ||
224 | if (extent_start == start) { | |
225 | start = extent_end + 1; | |
226 | } else if (extent_start > start && extent_start < end) { | |
227 | size = extent_start - start; | |
817d52f8 | 228 | total_added += size; |
ea6a478e JB |
229 | ret = btrfs_add_free_space(block_group, start, |
230 | size); | |
0f9dd46c JB |
231 | BUG_ON(ret); |
232 | start = extent_end + 1; | |
233 | } else { | |
234 | break; | |
235 | } | |
236 | } | |
237 | ||
238 | if (start < end) { | |
239 | size = end - start; | |
817d52f8 | 240 | total_added += size; |
ea6a478e | 241 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
242 | BUG_ON(ret); |
243 | } | |
244 | ||
817d52f8 | 245 | return total_added; |
0f9dd46c JB |
246 | } |
247 | ||
817d52f8 | 248 | static int caching_kthread(void *data) |
e37c9e69 | 249 | { |
817d52f8 JB |
250 | struct btrfs_block_group_cache *block_group = data; |
251 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
252 | u64 last = 0; | |
e37c9e69 | 253 | struct btrfs_path *path; |
ef8bbdfe | 254 | int ret = 0; |
e37c9e69 | 255 | struct btrfs_key key; |
5f39d397 | 256 | struct extent_buffer *leaf; |
e37c9e69 | 257 | int slot; |
817d52f8 | 258 | u64 total_found = 0; |
00f5c795 | 259 | |
817d52f8 | 260 | BUG_ON(!fs_info); |
f510cfec | 261 | |
e37c9e69 CM |
262 | path = btrfs_alloc_path(); |
263 | if (!path) | |
264 | return -ENOMEM; | |
7d7d6068 | 265 | |
817d52f8 JB |
266 | atomic_inc(&block_group->space_info->caching_threads); |
267 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); | |
268 | again: | |
269 | /* need to make sure the commit_root doesn't disappear */ | |
270 | down_read(&fs_info->extent_root->commit_root_sem); | |
271 | ||
5cd57b2c | 272 | /* |
817d52f8 JB |
273 | * We don't want to deadlock with somebody trying to allocate a new |
274 | * extent for the extent root while also trying to search the extent | |
275 | * root to add free space. So we skip locking and search the commit | |
276 | * root, since its read-only | |
5cd57b2c CM |
277 | */ |
278 | path->skip_locking = 1; | |
817d52f8 JB |
279 | path->search_commit_root = 1; |
280 | path->reada = 2; | |
281 | ||
e4404d6e | 282 | key.objectid = last; |
e37c9e69 CM |
283 | key.offset = 0; |
284 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
817d52f8 | 285 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
e37c9e69 | 286 | if (ret < 0) |
ef8bbdfe | 287 | goto err; |
a512bbf8 | 288 | |
d397712b | 289 | while (1) { |
817d52f8 JB |
290 | smp_mb(); |
291 | if (block_group->fs_info->closing) | |
292 | break; | |
293 | ||
5f39d397 | 294 | leaf = path->nodes[0]; |
e37c9e69 | 295 | slot = path->slots[0]; |
5f39d397 | 296 | if (slot >= btrfs_header_nritems(leaf)) { |
817d52f8 | 297 | ret = btrfs_next_leaf(fs_info->extent_root, path); |
54aa1f4d CM |
298 | if (ret < 0) |
299 | goto err; | |
817d52f8 | 300 | else if (ret) |
e37c9e69 | 301 | break; |
817d52f8 JB |
302 | |
303 | if (need_resched()) { | |
304 | btrfs_release_path(fs_info->extent_root, path); | |
305 | up_read(&fs_info->extent_root->commit_root_sem); | |
306 | cond_resched(); | |
307 | goto again; | |
308 | } | |
309 | ||
310 | continue; | |
e37c9e69 | 311 | } |
5f39d397 | 312 | btrfs_item_key_to_cpu(leaf, &key, slot); |
0f9dd46c | 313 | if (key.objectid < block_group->key.objectid) |
7d7d6068 | 314 | goto next; |
0f9dd46c | 315 | |
e37c9e69 | 316 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 317 | block_group->key.offset) |
e37c9e69 | 318 | break; |
7d7d6068 | 319 | |
e37c9e69 | 320 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
321 | total_found += add_new_free_space(block_group, |
322 | fs_info, last, | |
323 | key.objectid); | |
7d7d6068 | 324 | last = key.objectid + key.offset; |
e37c9e69 | 325 | } |
817d52f8 JB |
326 | |
327 | if (total_found > (1024 * 1024 * 2)) { | |
328 | total_found = 0; | |
329 | wake_up(&block_group->caching_q); | |
330 | } | |
7d7d6068 | 331 | next: |
e37c9e69 CM |
332 | path->slots[0]++; |
333 | } | |
817d52f8 | 334 | ret = 0; |
e37c9e69 | 335 | |
817d52f8 JB |
336 | total_found += add_new_free_space(block_group, fs_info, last, |
337 | block_group->key.objectid + | |
338 | block_group->key.offset); | |
339 | ||
340 | spin_lock(&block_group->lock); | |
341 | block_group->cached = BTRFS_CACHE_FINISHED; | |
342 | spin_unlock(&block_group->lock); | |
0f9dd46c | 343 | |
54aa1f4d | 344 | err: |
e37c9e69 | 345 | btrfs_free_path(path); |
817d52f8 | 346 | up_read(&fs_info->extent_root->commit_root_sem); |
817d52f8 JB |
347 | atomic_dec(&block_group->space_info->caching_threads); |
348 | wake_up(&block_group->caching_q); | |
349 | ||
817d52f8 JB |
350 | return 0; |
351 | } | |
352 | ||
353 | static int cache_block_group(struct btrfs_block_group_cache *cache) | |
354 | { | |
355 | struct task_struct *tsk; | |
356 | int ret = 0; | |
357 | ||
358 | spin_lock(&cache->lock); | |
359 | if (cache->cached != BTRFS_CACHE_NO) { | |
360 | spin_unlock(&cache->lock); | |
361 | return ret; | |
362 | } | |
363 | cache->cached = BTRFS_CACHE_STARTED; | |
364 | spin_unlock(&cache->lock); | |
365 | ||
366 | tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n", | |
367 | cache->key.objectid); | |
368 | if (IS_ERR(tsk)) { | |
369 | ret = PTR_ERR(tsk); | |
370 | printk(KERN_ERR "error running thread %d\n", ret); | |
371 | BUG(); | |
372 | } | |
373 | ||
ef8bbdfe | 374 | return ret; |
e37c9e69 CM |
375 | } |
376 | ||
0f9dd46c JB |
377 | /* |
378 | * return the block group that starts at or after bytenr | |
379 | */ | |
d397712b CM |
380 | static struct btrfs_block_group_cache * |
381 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 382 | { |
0f9dd46c | 383 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 384 | |
0f9dd46c | 385 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 386 | |
0f9dd46c | 387 | return cache; |
0ef3e66b CM |
388 | } |
389 | ||
0f9dd46c | 390 | /* |
9f55684c | 391 | * return the block group that contains the given bytenr |
0f9dd46c | 392 | */ |
d397712b CM |
393 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
394 | struct btrfs_fs_info *info, | |
395 | u64 bytenr) | |
be744175 | 396 | { |
0f9dd46c | 397 | struct btrfs_block_group_cache *cache; |
be744175 | 398 | |
0f9dd46c | 399 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 400 | |
0f9dd46c | 401 | return cache; |
be744175 | 402 | } |
0b86a832 | 403 | |
fa9c0d79 | 404 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) |
d2fb3437 YZ |
405 | { |
406 | if (atomic_dec_and_test(&cache->count)) | |
407 | kfree(cache); | |
408 | } | |
409 | ||
0f9dd46c JB |
410 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
411 | u64 flags) | |
6324fbf3 | 412 | { |
0f9dd46c | 413 | struct list_head *head = &info->space_info; |
0f9dd46c | 414 | struct btrfs_space_info *found; |
4184ea7f CM |
415 | |
416 | rcu_read_lock(); | |
417 | list_for_each_entry_rcu(found, head, list) { | |
418 | if (found->flags == flags) { | |
419 | rcu_read_unlock(); | |
0f9dd46c | 420 | return found; |
4184ea7f | 421 | } |
0f9dd46c | 422 | } |
4184ea7f | 423 | rcu_read_unlock(); |
0f9dd46c | 424 | return NULL; |
6324fbf3 CM |
425 | } |
426 | ||
4184ea7f CM |
427 | /* |
428 | * after adding space to the filesystem, we need to clear the full flags | |
429 | * on all the space infos. | |
430 | */ | |
431 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
432 | { | |
433 | struct list_head *head = &info->space_info; | |
434 | struct btrfs_space_info *found; | |
435 | ||
436 | rcu_read_lock(); | |
437 | list_for_each_entry_rcu(found, head, list) | |
438 | found->full = 0; | |
439 | rcu_read_unlock(); | |
440 | } | |
441 | ||
80eb234a JB |
442 | static u64 div_factor(u64 num, int factor) |
443 | { | |
444 | if (factor == 10) | |
445 | return num; | |
446 | num *= factor; | |
447 | do_div(num, 10); | |
448 | return num; | |
449 | } | |
450 | ||
d2fb3437 YZ |
451 | u64 btrfs_find_block_group(struct btrfs_root *root, |
452 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 453 | { |
96b5179d | 454 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 455 | u64 used; |
d2fb3437 YZ |
456 | u64 last = max(search_hint, search_start); |
457 | u64 group_start = 0; | |
31f3c99b | 458 | int full_search = 0; |
d2fb3437 | 459 | int factor = 9; |
0ef3e66b | 460 | int wrapped = 0; |
31f3c99b | 461 | again: |
e8569813 ZY |
462 | while (1) { |
463 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
464 | if (!cache) |
465 | break; | |
96b5179d | 466 | |
c286ac48 | 467 | spin_lock(&cache->lock); |
96b5179d CM |
468 | last = cache->key.objectid + cache->key.offset; |
469 | used = btrfs_block_group_used(&cache->item); | |
470 | ||
d2fb3437 YZ |
471 | if ((full_search || !cache->ro) && |
472 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 473 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
474 | div_factor(cache->key.offset, factor)) { |
475 | group_start = cache->key.objectid; | |
c286ac48 | 476 | spin_unlock(&cache->lock); |
fa9c0d79 | 477 | btrfs_put_block_group(cache); |
8790d502 CM |
478 | goto found; |
479 | } | |
6324fbf3 | 480 | } |
c286ac48 | 481 | spin_unlock(&cache->lock); |
fa9c0d79 | 482 | btrfs_put_block_group(cache); |
de428b63 | 483 | cond_resched(); |
cd1bc465 | 484 | } |
0ef3e66b CM |
485 | if (!wrapped) { |
486 | last = search_start; | |
487 | wrapped = 1; | |
488 | goto again; | |
489 | } | |
490 | if (!full_search && factor < 10) { | |
be744175 | 491 | last = search_start; |
31f3c99b | 492 | full_search = 1; |
0ef3e66b | 493 | factor = 10; |
31f3c99b CM |
494 | goto again; |
495 | } | |
be744175 | 496 | found: |
d2fb3437 | 497 | return group_start; |
925baedd | 498 | } |
0f9dd46c | 499 | |
e02119d5 | 500 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 501 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
502 | { |
503 | int ret; | |
504 | struct btrfs_key key; | |
31840ae1 | 505 | struct btrfs_path *path; |
e02119d5 | 506 | |
31840ae1 ZY |
507 | path = btrfs_alloc_path(); |
508 | BUG_ON(!path); | |
e02119d5 CM |
509 | key.objectid = start; |
510 | key.offset = len; | |
511 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
512 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
513 | 0, 0); | |
31840ae1 | 514 | btrfs_free_path(path); |
7bb86316 CM |
515 | return ret; |
516 | } | |
517 | ||
d8d5f3e1 CM |
518 | /* |
519 | * Back reference rules. Back refs have three main goals: | |
520 | * | |
521 | * 1) differentiate between all holders of references to an extent so that | |
522 | * when a reference is dropped we can make sure it was a valid reference | |
523 | * before freeing the extent. | |
524 | * | |
525 | * 2) Provide enough information to quickly find the holders of an extent | |
526 | * if we notice a given block is corrupted or bad. | |
527 | * | |
528 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
529 | * maintenance. This is actually the same as #2, but with a slightly | |
530 | * different use case. | |
531 | * | |
5d4f98a2 YZ |
532 | * There are two kinds of back refs. The implicit back refs is optimized |
533 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
534 | * back refs of this kind provide information about the block's owner tree | |
535 | * and the pointer's key. These information allow us to find the block by | |
536 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
537 | * referenced by their owner trees. The location of tree block is recorded | |
538 | * in the back refs. Actually the full back refs is generic, and can be | |
539 | * used in all cases the implicit back refs is used. The major shortcoming | |
540 | * of the full back refs is its overhead. Every time a tree block gets | |
541 | * COWed, we have to update back refs entry for all pointers in it. | |
542 | * | |
543 | * For a newly allocated tree block, we use implicit back refs for | |
544 | * pointers in it. This means most tree related operations only involve | |
545 | * implicit back refs. For a tree block created in old transaction, the | |
546 | * only way to drop a reference to it is COW it. So we can detect the | |
547 | * event that tree block loses its owner tree's reference and do the | |
548 | * back refs conversion. | |
549 | * | |
550 | * When a tree block is COW'd through a tree, there are four cases: | |
551 | * | |
552 | * The reference count of the block is one and the tree is the block's | |
553 | * owner tree. Nothing to do in this case. | |
554 | * | |
555 | * The reference count of the block is one and the tree is not the | |
556 | * block's owner tree. In this case, full back refs is used for pointers | |
557 | * in the block. Remove these full back refs, add implicit back refs for | |
558 | * every pointers in the new block. | |
559 | * | |
560 | * The reference count of the block is greater than one and the tree is | |
561 | * the block's owner tree. In this case, implicit back refs is used for | |
562 | * pointers in the block. Add full back refs for every pointers in the | |
563 | * block, increase lower level extents' reference counts. The original | |
564 | * implicit back refs are entailed to the new block. | |
565 | * | |
566 | * The reference count of the block is greater than one and the tree is | |
567 | * not the block's owner tree. Add implicit back refs for every pointer in | |
568 | * the new block, increase lower level extents' reference count. | |
569 | * | |
570 | * Back Reference Key composing: | |
571 | * | |
572 | * The key objectid corresponds to the first byte in the extent, | |
573 | * The key type is used to differentiate between types of back refs. | |
574 | * There are different meanings of the key offset for different types | |
575 | * of back refs. | |
576 | * | |
d8d5f3e1 CM |
577 | * File extents can be referenced by: |
578 | * | |
579 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 580 | * - different files inside a single subvolume |
d8d5f3e1 CM |
581 | * - different offsets inside a file (bookend extents in file.c) |
582 | * | |
5d4f98a2 | 583 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
584 | * |
585 | * - Objectid of the subvolume root | |
d8d5f3e1 | 586 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
587 | * - original offset in the file |
588 | * - how many bookend extents | |
d8d5f3e1 | 589 | * |
5d4f98a2 YZ |
590 | * The key offset for the implicit back refs is hash of the first |
591 | * three fields. | |
d8d5f3e1 | 592 | * |
5d4f98a2 | 593 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 594 | * |
5d4f98a2 | 595 | * - number of pointers in the tree leaf |
d8d5f3e1 | 596 | * |
5d4f98a2 YZ |
597 | * The key offset for the implicit back refs is the first byte of |
598 | * the tree leaf | |
d8d5f3e1 | 599 | * |
5d4f98a2 YZ |
600 | * When a file extent is allocated, The implicit back refs is used. |
601 | * the fields are filled in: | |
d8d5f3e1 | 602 | * |
5d4f98a2 | 603 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 604 | * |
5d4f98a2 YZ |
605 | * When a file extent is removed file truncation, we find the |
606 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 607 | * |
5d4f98a2 | 608 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 609 | * |
5d4f98a2 | 610 | * Btree extents can be referenced by: |
d8d5f3e1 | 611 | * |
5d4f98a2 | 612 | * - Different subvolumes |
d8d5f3e1 | 613 | * |
5d4f98a2 YZ |
614 | * Both the implicit back refs and the full back refs for tree blocks |
615 | * only consist of key. The key offset for the implicit back refs is | |
616 | * objectid of block's owner tree. The key offset for the full back refs | |
617 | * is the first byte of parent block. | |
d8d5f3e1 | 618 | * |
5d4f98a2 YZ |
619 | * When implicit back refs is used, information about the lowest key and |
620 | * level of the tree block are required. These information are stored in | |
621 | * tree block info structure. | |
d8d5f3e1 | 622 | */ |
31840ae1 | 623 | |
5d4f98a2 YZ |
624 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
625 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
626 | struct btrfs_root *root, | |
627 | struct btrfs_path *path, | |
628 | u64 owner, u32 extra_size) | |
7bb86316 | 629 | { |
5d4f98a2 YZ |
630 | struct btrfs_extent_item *item; |
631 | struct btrfs_extent_item_v0 *ei0; | |
632 | struct btrfs_extent_ref_v0 *ref0; | |
633 | struct btrfs_tree_block_info *bi; | |
634 | struct extent_buffer *leaf; | |
7bb86316 | 635 | struct btrfs_key key; |
5d4f98a2 YZ |
636 | struct btrfs_key found_key; |
637 | u32 new_size = sizeof(*item); | |
638 | u64 refs; | |
639 | int ret; | |
640 | ||
641 | leaf = path->nodes[0]; | |
642 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
643 | ||
644 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
645 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
646 | struct btrfs_extent_item_v0); | |
647 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
648 | ||
649 | if (owner == (u64)-1) { | |
650 | while (1) { | |
651 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
652 | ret = btrfs_next_leaf(root, path); | |
653 | if (ret < 0) | |
654 | return ret; | |
655 | BUG_ON(ret > 0); | |
656 | leaf = path->nodes[0]; | |
657 | } | |
658 | btrfs_item_key_to_cpu(leaf, &found_key, | |
659 | path->slots[0]); | |
660 | BUG_ON(key.objectid != found_key.objectid); | |
661 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
662 | path->slots[0]++; | |
663 | continue; | |
664 | } | |
665 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
666 | struct btrfs_extent_ref_v0); | |
667 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
668 | break; | |
669 | } | |
670 | } | |
671 | btrfs_release_path(root, path); | |
672 | ||
673 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
674 | new_size += sizeof(*bi); | |
675 | ||
676 | new_size -= sizeof(*ei0); | |
677 | ret = btrfs_search_slot(trans, root, &key, path, | |
678 | new_size + extra_size, 1); | |
679 | if (ret < 0) | |
680 | return ret; | |
681 | BUG_ON(ret); | |
682 | ||
683 | ret = btrfs_extend_item(trans, root, path, new_size); | |
684 | BUG_ON(ret); | |
685 | ||
686 | leaf = path->nodes[0]; | |
687 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
688 | btrfs_set_extent_refs(leaf, item, refs); | |
689 | /* FIXME: get real generation */ | |
690 | btrfs_set_extent_generation(leaf, item, 0); | |
691 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
692 | btrfs_set_extent_flags(leaf, item, | |
693 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
694 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
695 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
696 | /* FIXME: get first key of the block */ | |
697 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
698 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
699 | } else { | |
700 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
701 | } | |
702 | btrfs_mark_buffer_dirty(leaf); | |
703 | return 0; | |
704 | } | |
705 | #endif | |
706 | ||
707 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
708 | { | |
709 | u32 high_crc = ~(u32)0; | |
710 | u32 low_crc = ~(u32)0; | |
711 | __le64 lenum; | |
712 | ||
713 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 714 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 715 | lenum = cpu_to_le64(owner); |
163e783e | 716 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 717 | lenum = cpu_to_le64(offset); |
163e783e | 718 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
719 | |
720 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
721 | } | |
722 | ||
723 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
724 | struct btrfs_extent_data_ref *ref) | |
725 | { | |
726 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
727 | btrfs_extent_data_ref_objectid(leaf, ref), | |
728 | btrfs_extent_data_ref_offset(leaf, ref)); | |
729 | } | |
730 | ||
731 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
732 | struct btrfs_extent_data_ref *ref, | |
733 | u64 root_objectid, u64 owner, u64 offset) | |
734 | { | |
735 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
736 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
737 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
738 | return 0; | |
739 | return 1; | |
740 | } | |
741 | ||
742 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
743 | struct btrfs_root *root, | |
744 | struct btrfs_path *path, | |
745 | u64 bytenr, u64 parent, | |
746 | u64 root_objectid, | |
747 | u64 owner, u64 offset) | |
748 | { | |
749 | struct btrfs_key key; | |
750 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 751 | struct extent_buffer *leaf; |
5d4f98a2 | 752 | u32 nritems; |
74493f7a | 753 | int ret; |
5d4f98a2 YZ |
754 | int recow; |
755 | int err = -ENOENT; | |
74493f7a | 756 | |
31840ae1 | 757 | key.objectid = bytenr; |
5d4f98a2 YZ |
758 | if (parent) { |
759 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
760 | key.offset = parent; | |
761 | } else { | |
762 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
763 | key.offset = hash_extent_data_ref(root_objectid, | |
764 | owner, offset); | |
765 | } | |
766 | again: | |
767 | recow = 0; | |
768 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
769 | if (ret < 0) { | |
770 | err = ret; | |
771 | goto fail; | |
772 | } | |
31840ae1 | 773 | |
5d4f98a2 YZ |
774 | if (parent) { |
775 | if (!ret) | |
776 | return 0; | |
777 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
778 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
779 | btrfs_release_path(root, path); | |
780 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
781 | if (ret < 0) { | |
782 | err = ret; | |
783 | goto fail; | |
784 | } | |
785 | if (!ret) | |
786 | return 0; | |
787 | #endif | |
788 | goto fail; | |
31840ae1 ZY |
789 | } |
790 | ||
791 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
792 | nritems = btrfs_header_nritems(leaf); |
793 | while (1) { | |
794 | if (path->slots[0] >= nritems) { | |
795 | ret = btrfs_next_leaf(root, path); | |
796 | if (ret < 0) | |
797 | err = ret; | |
798 | if (ret) | |
799 | goto fail; | |
800 | ||
801 | leaf = path->nodes[0]; | |
802 | nritems = btrfs_header_nritems(leaf); | |
803 | recow = 1; | |
804 | } | |
805 | ||
806 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
807 | if (key.objectid != bytenr || | |
808 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
809 | goto fail; | |
810 | ||
811 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
812 | struct btrfs_extent_data_ref); | |
813 | ||
814 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
815 | owner, offset)) { | |
816 | if (recow) { | |
817 | btrfs_release_path(root, path); | |
818 | goto again; | |
819 | } | |
820 | err = 0; | |
821 | break; | |
822 | } | |
823 | path->slots[0]++; | |
31840ae1 | 824 | } |
5d4f98a2 YZ |
825 | fail: |
826 | return err; | |
31840ae1 ZY |
827 | } |
828 | ||
5d4f98a2 YZ |
829 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
830 | struct btrfs_root *root, | |
831 | struct btrfs_path *path, | |
832 | u64 bytenr, u64 parent, | |
833 | u64 root_objectid, u64 owner, | |
834 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
835 | { |
836 | struct btrfs_key key; | |
837 | struct extent_buffer *leaf; | |
5d4f98a2 | 838 | u32 size; |
31840ae1 ZY |
839 | u32 num_refs; |
840 | int ret; | |
74493f7a | 841 | |
74493f7a | 842 | key.objectid = bytenr; |
5d4f98a2 YZ |
843 | if (parent) { |
844 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
845 | key.offset = parent; | |
846 | size = sizeof(struct btrfs_shared_data_ref); | |
847 | } else { | |
848 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
849 | key.offset = hash_extent_data_ref(root_objectid, | |
850 | owner, offset); | |
851 | size = sizeof(struct btrfs_extent_data_ref); | |
852 | } | |
74493f7a | 853 | |
5d4f98a2 YZ |
854 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
855 | if (ret && ret != -EEXIST) | |
856 | goto fail; | |
857 | ||
858 | leaf = path->nodes[0]; | |
859 | if (parent) { | |
860 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 861 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
862 | struct btrfs_shared_data_ref); |
863 | if (ret == 0) { | |
864 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
865 | } else { | |
866 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
867 | num_refs += refs_to_add; | |
868 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 869 | } |
5d4f98a2 YZ |
870 | } else { |
871 | struct btrfs_extent_data_ref *ref; | |
872 | while (ret == -EEXIST) { | |
873 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
874 | struct btrfs_extent_data_ref); | |
875 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
876 | owner, offset)) | |
877 | break; | |
878 | btrfs_release_path(root, path); | |
879 | key.offset++; | |
880 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
881 | size); | |
882 | if (ret && ret != -EEXIST) | |
883 | goto fail; | |
31840ae1 | 884 | |
5d4f98a2 YZ |
885 | leaf = path->nodes[0]; |
886 | } | |
887 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
888 | struct btrfs_extent_data_ref); | |
889 | if (ret == 0) { | |
890 | btrfs_set_extent_data_ref_root(leaf, ref, | |
891 | root_objectid); | |
892 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
893 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
894 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
895 | } else { | |
896 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
897 | num_refs += refs_to_add; | |
898 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 899 | } |
31840ae1 | 900 | } |
5d4f98a2 YZ |
901 | btrfs_mark_buffer_dirty(leaf); |
902 | ret = 0; | |
903 | fail: | |
7bb86316 CM |
904 | btrfs_release_path(root, path); |
905 | return ret; | |
74493f7a CM |
906 | } |
907 | ||
5d4f98a2 YZ |
908 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
909 | struct btrfs_root *root, | |
910 | struct btrfs_path *path, | |
911 | int refs_to_drop) | |
31840ae1 | 912 | { |
5d4f98a2 YZ |
913 | struct btrfs_key key; |
914 | struct btrfs_extent_data_ref *ref1 = NULL; | |
915 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 916 | struct extent_buffer *leaf; |
5d4f98a2 | 917 | u32 num_refs = 0; |
31840ae1 ZY |
918 | int ret = 0; |
919 | ||
920 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
921 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
922 | ||
923 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
924 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
925 | struct btrfs_extent_data_ref); | |
926 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
927 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
928 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
929 | struct btrfs_shared_data_ref); | |
930 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
931 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
932 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
933 | struct btrfs_extent_ref_v0 *ref0; | |
934 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
935 | struct btrfs_extent_ref_v0); | |
936 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
937 | #endif | |
938 | } else { | |
939 | BUG(); | |
940 | } | |
941 | ||
56bec294 CM |
942 | BUG_ON(num_refs < refs_to_drop); |
943 | num_refs -= refs_to_drop; | |
5d4f98a2 | 944 | |
31840ae1 ZY |
945 | if (num_refs == 0) { |
946 | ret = btrfs_del_item(trans, root, path); | |
947 | } else { | |
5d4f98a2 YZ |
948 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
949 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
950 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
951 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
952 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
953 | else { | |
954 | struct btrfs_extent_ref_v0 *ref0; | |
955 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
956 | struct btrfs_extent_ref_v0); | |
957 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
958 | } | |
959 | #endif | |
31840ae1 ZY |
960 | btrfs_mark_buffer_dirty(leaf); |
961 | } | |
31840ae1 ZY |
962 | return ret; |
963 | } | |
964 | ||
5d4f98a2 YZ |
965 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
966 | struct btrfs_path *path, | |
967 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 968 | { |
5d4f98a2 YZ |
969 | struct btrfs_key key; |
970 | struct extent_buffer *leaf; | |
971 | struct btrfs_extent_data_ref *ref1; | |
972 | struct btrfs_shared_data_ref *ref2; | |
973 | u32 num_refs = 0; | |
974 | ||
975 | leaf = path->nodes[0]; | |
976 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
977 | if (iref) { | |
978 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
979 | BTRFS_EXTENT_DATA_REF_KEY) { | |
980 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
981 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
982 | } else { | |
983 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
984 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
985 | } | |
986 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
987 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
988 | struct btrfs_extent_data_ref); | |
989 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
990 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
991 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
992 | struct btrfs_shared_data_ref); | |
993 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
994 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
995 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
996 | struct btrfs_extent_ref_v0 *ref0; | |
997 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
998 | struct btrfs_extent_ref_v0); | |
999 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1000 | #endif |
5d4f98a2 YZ |
1001 | } else { |
1002 | WARN_ON(1); | |
1003 | } | |
1004 | return num_refs; | |
1005 | } | |
15916de8 | 1006 | |
5d4f98a2 YZ |
1007 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1008 | struct btrfs_root *root, | |
1009 | struct btrfs_path *path, | |
1010 | u64 bytenr, u64 parent, | |
1011 | u64 root_objectid) | |
1f3c79a2 | 1012 | { |
5d4f98a2 | 1013 | struct btrfs_key key; |
1f3c79a2 | 1014 | int ret; |
1f3c79a2 | 1015 | |
5d4f98a2 YZ |
1016 | key.objectid = bytenr; |
1017 | if (parent) { | |
1018 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1019 | key.offset = parent; | |
1020 | } else { | |
1021 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1022 | key.offset = root_objectid; | |
1f3c79a2 LH |
1023 | } |
1024 | ||
5d4f98a2 YZ |
1025 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1026 | if (ret > 0) | |
1027 | ret = -ENOENT; | |
1028 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1029 | if (ret == -ENOENT && parent) { | |
1030 | btrfs_release_path(root, path); | |
1031 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
1032 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1033 | if (ret > 0) | |
1034 | ret = -ENOENT; | |
1035 | } | |
1f3c79a2 | 1036 | #endif |
5d4f98a2 | 1037 | return ret; |
1f3c79a2 LH |
1038 | } |
1039 | ||
5d4f98a2 YZ |
1040 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1041 | struct btrfs_root *root, | |
1042 | struct btrfs_path *path, | |
1043 | u64 bytenr, u64 parent, | |
1044 | u64 root_objectid) | |
31840ae1 | 1045 | { |
5d4f98a2 | 1046 | struct btrfs_key key; |
31840ae1 | 1047 | int ret; |
31840ae1 | 1048 | |
5d4f98a2 YZ |
1049 | key.objectid = bytenr; |
1050 | if (parent) { | |
1051 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1052 | key.offset = parent; | |
1053 | } else { | |
1054 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1055 | key.offset = root_objectid; | |
1056 | } | |
1057 | ||
1058 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1059 | btrfs_release_path(root, path); | |
31840ae1 ZY |
1060 | return ret; |
1061 | } | |
1062 | ||
5d4f98a2 | 1063 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1064 | { |
5d4f98a2 YZ |
1065 | int type; |
1066 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1067 | if (parent > 0) | |
1068 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1069 | else | |
1070 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1071 | } else { | |
1072 | if (parent > 0) | |
1073 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1074 | else | |
1075 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1076 | } | |
1077 | return type; | |
31840ae1 | 1078 | } |
56bec294 | 1079 | |
2c47e605 YZ |
1080 | static int find_next_key(struct btrfs_path *path, int level, |
1081 | struct btrfs_key *key) | |
56bec294 | 1082 | |
02217ed2 | 1083 | { |
2c47e605 | 1084 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1085 | if (!path->nodes[level]) |
1086 | break; | |
5d4f98a2 YZ |
1087 | if (path->slots[level] + 1 >= |
1088 | btrfs_header_nritems(path->nodes[level])) | |
1089 | continue; | |
1090 | if (level == 0) | |
1091 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1092 | path->slots[level] + 1); | |
1093 | else | |
1094 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1095 | path->slots[level] + 1); | |
1096 | return 0; | |
1097 | } | |
1098 | return 1; | |
1099 | } | |
037e6390 | 1100 | |
5d4f98a2 YZ |
1101 | /* |
1102 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1103 | * to the address of inline back ref, and 0 is returned. | |
1104 | * | |
1105 | * if back ref isn't found, *ref_ret is set to the address where it | |
1106 | * should be inserted, and -ENOENT is returned. | |
1107 | * | |
1108 | * if insert is true and there are too many inline back refs, the path | |
1109 | * points to the extent item, and -EAGAIN is returned. | |
1110 | * | |
1111 | * NOTE: inline back refs are ordered in the same way that back ref | |
1112 | * items in the tree are ordered. | |
1113 | */ | |
1114 | static noinline_for_stack | |
1115 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1116 | struct btrfs_root *root, | |
1117 | struct btrfs_path *path, | |
1118 | struct btrfs_extent_inline_ref **ref_ret, | |
1119 | u64 bytenr, u64 num_bytes, | |
1120 | u64 parent, u64 root_objectid, | |
1121 | u64 owner, u64 offset, int insert) | |
1122 | { | |
1123 | struct btrfs_key key; | |
1124 | struct extent_buffer *leaf; | |
1125 | struct btrfs_extent_item *ei; | |
1126 | struct btrfs_extent_inline_ref *iref; | |
1127 | u64 flags; | |
1128 | u64 item_size; | |
1129 | unsigned long ptr; | |
1130 | unsigned long end; | |
1131 | int extra_size; | |
1132 | int type; | |
1133 | int want; | |
1134 | int ret; | |
1135 | int err = 0; | |
26b8003f | 1136 | |
db94535d | 1137 | key.objectid = bytenr; |
31840ae1 | 1138 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1139 | key.offset = num_bytes; |
31840ae1 | 1140 | |
5d4f98a2 YZ |
1141 | want = extent_ref_type(parent, owner); |
1142 | if (insert) { | |
1143 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1144 | path->keep_locks = 1; |
5d4f98a2 YZ |
1145 | } else |
1146 | extra_size = -1; | |
1147 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1148 | if (ret < 0) { |
5d4f98a2 YZ |
1149 | err = ret; |
1150 | goto out; | |
1151 | } | |
1152 | BUG_ON(ret); | |
1153 | ||
1154 | leaf = path->nodes[0]; | |
1155 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1156 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1157 | if (item_size < sizeof(*ei)) { | |
1158 | if (!insert) { | |
1159 | err = -ENOENT; | |
1160 | goto out; | |
1161 | } | |
1162 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1163 | extra_size); | |
1164 | if (ret < 0) { | |
1165 | err = ret; | |
1166 | goto out; | |
1167 | } | |
1168 | leaf = path->nodes[0]; | |
1169 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1170 | } | |
1171 | #endif | |
1172 | BUG_ON(item_size < sizeof(*ei)); | |
1173 | ||
5d4f98a2 YZ |
1174 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1175 | flags = btrfs_extent_flags(leaf, ei); | |
1176 | ||
1177 | ptr = (unsigned long)(ei + 1); | |
1178 | end = (unsigned long)ei + item_size; | |
1179 | ||
1180 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1181 | ptr += sizeof(struct btrfs_tree_block_info); | |
1182 | BUG_ON(ptr > end); | |
1183 | } else { | |
1184 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1185 | } | |
1186 | ||
1187 | err = -ENOENT; | |
1188 | while (1) { | |
1189 | if (ptr >= end) { | |
1190 | WARN_ON(ptr > end); | |
1191 | break; | |
1192 | } | |
1193 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1194 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1195 | if (want < type) | |
1196 | break; | |
1197 | if (want > type) { | |
1198 | ptr += btrfs_extent_inline_ref_size(type); | |
1199 | continue; | |
1200 | } | |
1201 | ||
1202 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1203 | struct btrfs_extent_data_ref *dref; | |
1204 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1205 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1206 | owner, offset)) { | |
1207 | err = 0; | |
1208 | break; | |
1209 | } | |
1210 | if (hash_extent_data_ref_item(leaf, dref) < | |
1211 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1212 | break; | |
1213 | } else { | |
1214 | u64 ref_offset; | |
1215 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1216 | if (parent > 0) { | |
1217 | if (parent == ref_offset) { | |
1218 | err = 0; | |
1219 | break; | |
1220 | } | |
1221 | if (ref_offset < parent) | |
1222 | break; | |
1223 | } else { | |
1224 | if (root_objectid == ref_offset) { | |
1225 | err = 0; | |
1226 | break; | |
1227 | } | |
1228 | if (ref_offset < root_objectid) | |
1229 | break; | |
1230 | } | |
1231 | } | |
1232 | ptr += btrfs_extent_inline_ref_size(type); | |
1233 | } | |
1234 | if (err == -ENOENT && insert) { | |
1235 | if (item_size + extra_size >= | |
1236 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1237 | err = -EAGAIN; | |
1238 | goto out; | |
1239 | } | |
1240 | /* | |
1241 | * To add new inline back ref, we have to make sure | |
1242 | * there is no corresponding back ref item. | |
1243 | * For simplicity, we just do not add new inline back | |
1244 | * ref if there is any kind of item for this block | |
1245 | */ | |
2c47e605 YZ |
1246 | if (find_next_key(path, 0, &key) == 0 && |
1247 | key.objectid == bytenr && | |
85d4198e | 1248 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1249 | err = -EAGAIN; |
1250 | goto out; | |
1251 | } | |
1252 | } | |
1253 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1254 | out: | |
85d4198e | 1255 | if (insert) { |
5d4f98a2 YZ |
1256 | path->keep_locks = 0; |
1257 | btrfs_unlock_up_safe(path, 1); | |
1258 | } | |
1259 | return err; | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * helper to add new inline back ref | |
1264 | */ | |
1265 | static noinline_for_stack | |
1266 | int setup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1267 | struct btrfs_root *root, | |
1268 | struct btrfs_path *path, | |
1269 | struct btrfs_extent_inline_ref *iref, | |
1270 | u64 parent, u64 root_objectid, | |
1271 | u64 owner, u64 offset, int refs_to_add, | |
1272 | struct btrfs_delayed_extent_op *extent_op) | |
1273 | { | |
1274 | struct extent_buffer *leaf; | |
1275 | struct btrfs_extent_item *ei; | |
1276 | unsigned long ptr; | |
1277 | unsigned long end; | |
1278 | unsigned long item_offset; | |
1279 | u64 refs; | |
1280 | int size; | |
1281 | int type; | |
1282 | int ret; | |
1283 | ||
1284 | leaf = path->nodes[0]; | |
1285 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1286 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1287 | ||
1288 | type = extent_ref_type(parent, owner); | |
1289 | size = btrfs_extent_inline_ref_size(type); | |
1290 | ||
1291 | ret = btrfs_extend_item(trans, root, path, size); | |
1292 | BUG_ON(ret); | |
1293 | ||
1294 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1295 | refs = btrfs_extent_refs(leaf, ei); | |
1296 | refs += refs_to_add; | |
1297 | btrfs_set_extent_refs(leaf, ei, refs); | |
1298 | if (extent_op) | |
1299 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1300 | ||
1301 | ptr = (unsigned long)ei + item_offset; | |
1302 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1303 | if (ptr < end - size) | |
1304 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1305 | end - size - ptr); | |
1306 | ||
1307 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1308 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1309 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1310 | struct btrfs_extent_data_ref *dref; | |
1311 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1312 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1313 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1314 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1315 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1316 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1317 | struct btrfs_shared_data_ref *sref; | |
1318 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1319 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1320 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1321 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1322 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1323 | } else { | |
1324 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1325 | } | |
1326 | btrfs_mark_buffer_dirty(leaf); | |
1327 | return 0; | |
1328 | } | |
1329 | ||
1330 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1331 | struct btrfs_root *root, | |
1332 | struct btrfs_path *path, | |
1333 | struct btrfs_extent_inline_ref **ref_ret, | |
1334 | u64 bytenr, u64 num_bytes, u64 parent, | |
1335 | u64 root_objectid, u64 owner, u64 offset) | |
1336 | { | |
1337 | int ret; | |
1338 | ||
1339 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1340 | bytenr, num_bytes, parent, | |
1341 | root_objectid, owner, offset, 0); | |
1342 | if (ret != -ENOENT) | |
54aa1f4d | 1343 | return ret; |
5d4f98a2 YZ |
1344 | |
1345 | btrfs_release_path(root, path); | |
1346 | *ref_ret = NULL; | |
1347 | ||
1348 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1349 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1350 | root_objectid); | |
1351 | } else { | |
1352 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1353 | root_objectid, owner, offset); | |
b9473439 | 1354 | } |
5d4f98a2 YZ |
1355 | return ret; |
1356 | } | |
31840ae1 | 1357 | |
5d4f98a2 YZ |
1358 | /* |
1359 | * helper to update/remove inline back ref | |
1360 | */ | |
1361 | static noinline_for_stack | |
1362 | int update_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1363 | struct btrfs_root *root, | |
1364 | struct btrfs_path *path, | |
1365 | struct btrfs_extent_inline_ref *iref, | |
1366 | int refs_to_mod, | |
1367 | struct btrfs_delayed_extent_op *extent_op) | |
1368 | { | |
1369 | struct extent_buffer *leaf; | |
1370 | struct btrfs_extent_item *ei; | |
1371 | struct btrfs_extent_data_ref *dref = NULL; | |
1372 | struct btrfs_shared_data_ref *sref = NULL; | |
1373 | unsigned long ptr; | |
1374 | unsigned long end; | |
1375 | u32 item_size; | |
1376 | int size; | |
1377 | int type; | |
1378 | int ret; | |
1379 | u64 refs; | |
1380 | ||
1381 | leaf = path->nodes[0]; | |
1382 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1383 | refs = btrfs_extent_refs(leaf, ei); | |
1384 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1385 | refs += refs_to_mod; | |
1386 | btrfs_set_extent_refs(leaf, ei, refs); | |
1387 | if (extent_op) | |
1388 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1389 | ||
1390 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1391 | ||
1392 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1393 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1394 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1395 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1396 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1397 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1398 | } else { | |
1399 | refs = 1; | |
1400 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1401 | } |
31840ae1 | 1402 | |
5d4f98a2 YZ |
1403 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1404 | refs += refs_to_mod; | |
1405 | ||
1406 | if (refs > 0) { | |
1407 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1408 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1409 | else | |
1410 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1411 | } else { | |
1412 | size = btrfs_extent_inline_ref_size(type); | |
1413 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1414 | ptr = (unsigned long)iref; | |
1415 | end = (unsigned long)ei + item_size; | |
1416 | if (ptr + size < end) | |
1417 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1418 | end - ptr - size); | |
1419 | item_size -= size; | |
1420 | ret = btrfs_truncate_item(trans, root, path, item_size, 1); | |
1421 | BUG_ON(ret); | |
1422 | } | |
1423 | btrfs_mark_buffer_dirty(leaf); | |
1424 | return 0; | |
1425 | } | |
1426 | ||
1427 | static noinline_for_stack | |
1428 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1429 | struct btrfs_root *root, | |
1430 | struct btrfs_path *path, | |
1431 | u64 bytenr, u64 num_bytes, u64 parent, | |
1432 | u64 root_objectid, u64 owner, | |
1433 | u64 offset, int refs_to_add, | |
1434 | struct btrfs_delayed_extent_op *extent_op) | |
1435 | { | |
1436 | struct btrfs_extent_inline_ref *iref; | |
1437 | int ret; | |
1438 | ||
1439 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1440 | bytenr, num_bytes, parent, | |
1441 | root_objectid, owner, offset, 1); | |
1442 | if (ret == 0) { | |
1443 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
1444 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1445 | refs_to_add, extent_op); | |
1446 | } else if (ret == -ENOENT) { | |
1447 | ret = setup_inline_extent_backref(trans, root, path, iref, | |
1448 | parent, root_objectid, | |
1449 | owner, offset, refs_to_add, | |
1450 | extent_op); | |
771ed689 | 1451 | } |
5d4f98a2 YZ |
1452 | return ret; |
1453 | } | |
31840ae1 | 1454 | |
5d4f98a2 YZ |
1455 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1456 | struct btrfs_root *root, | |
1457 | struct btrfs_path *path, | |
1458 | u64 bytenr, u64 parent, u64 root_objectid, | |
1459 | u64 owner, u64 offset, int refs_to_add) | |
1460 | { | |
1461 | int ret; | |
1462 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1463 | BUG_ON(refs_to_add != 1); | |
1464 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1465 | parent, root_objectid); | |
1466 | } else { | |
1467 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1468 | parent, root_objectid, | |
1469 | owner, offset, refs_to_add); | |
1470 | } | |
1471 | return ret; | |
1472 | } | |
56bec294 | 1473 | |
5d4f98a2 YZ |
1474 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1475 | struct btrfs_root *root, | |
1476 | struct btrfs_path *path, | |
1477 | struct btrfs_extent_inline_ref *iref, | |
1478 | int refs_to_drop, int is_data) | |
1479 | { | |
1480 | int ret; | |
b9473439 | 1481 | |
5d4f98a2 YZ |
1482 | BUG_ON(!is_data && refs_to_drop != 1); |
1483 | if (iref) { | |
1484 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1485 | -refs_to_drop, NULL); | |
1486 | } else if (is_data) { | |
1487 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1488 | } else { | |
1489 | ret = btrfs_del_item(trans, root, path); | |
1490 | } | |
1491 | return ret; | |
1492 | } | |
1493 | ||
1494 | #ifdef BIO_RW_DISCARD | |
1495 | static void btrfs_issue_discard(struct block_device *bdev, | |
1496 | u64 start, u64 len) | |
1497 | { | |
1498 | blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL); | |
1499 | } | |
1500 | #endif | |
1501 | ||
1502 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
1503 | u64 num_bytes) | |
1504 | { | |
1505 | #ifdef BIO_RW_DISCARD | |
1506 | int ret; | |
1507 | u64 map_length = num_bytes; | |
1508 | struct btrfs_multi_bio *multi = NULL; | |
1509 | ||
1510 | /* Tell the block device(s) that the sectors can be discarded */ | |
1511 | ret = btrfs_map_block(&root->fs_info->mapping_tree, READ, | |
1512 | bytenr, &map_length, &multi, 0); | |
1513 | if (!ret) { | |
1514 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
1515 | int i; | |
1516 | ||
1517 | if (map_length > num_bytes) | |
1518 | map_length = num_bytes; | |
1519 | ||
1520 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
1521 | btrfs_issue_discard(stripe->dev->bdev, | |
1522 | stripe->physical, | |
1523 | map_length); | |
1524 | } | |
1525 | kfree(multi); | |
1526 | } | |
1527 | ||
1528 | return ret; | |
1529 | #else | |
1530 | return 0; | |
1531 | #endif | |
1532 | } | |
1533 | ||
1534 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1535 | struct btrfs_root *root, | |
1536 | u64 bytenr, u64 num_bytes, u64 parent, | |
1537 | u64 root_objectid, u64 owner, u64 offset) | |
1538 | { | |
1539 | int ret; | |
1540 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && | |
1541 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1542 | ||
1543 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1544 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
1545 | parent, root_objectid, (int)owner, | |
1546 | BTRFS_ADD_DELAYED_REF, NULL); | |
1547 | } else { | |
1548 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
1549 | parent, root_objectid, owner, offset, | |
1550 | BTRFS_ADD_DELAYED_REF, NULL); | |
1551 | } | |
1552 | return ret; | |
1553 | } | |
1554 | ||
1555 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1556 | struct btrfs_root *root, | |
1557 | u64 bytenr, u64 num_bytes, | |
1558 | u64 parent, u64 root_objectid, | |
1559 | u64 owner, u64 offset, int refs_to_add, | |
1560 | struct btrfs_delayed_extent_op *extent_op) | |
1561 | { | |
1562 | struct btrfs_path *path; | |
1563 | struct extent_buffer *leaf; | |
1564 | struct btrfs_extent_item *item; | |
1565 | u64 refs; | |
1566 | int ret; | |
1567 | int err = 0; | |
1568 | ||
1569 | path = btrfs_alloc_path(); | |
1570 | if (!path) | |
1571 | return -ENOMEM; | |
1572 | ||
1573 | path->reada = 1; | |
1574 | path->leave_spinning = 1; | |
1575 | /* this will setup the path even if it fails to insert the back ref */ | |
1576 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1577 | path, bytenr, num_bytes, parent, | |
1578 | root_objectid, owner, offset, | |
1579 | refs_to_add, extent_op); | |
1580 | if (ret == 0) | |
1581 | goto out; | |
1582 | ||
1583 | if (ret != -EAGAIN) { | |
1584 | err = ret; | |
1585 | goto out; | |
1586 | } | |
1587 | ||
1588 | leaf = path->nodes[0]; | |
1589 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1590 | refs = btrfs_extent_refs(leaf, item); | |
1591 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1592 | if (extent_op) | |
1593 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1594 | |
5d4f98a2 | 1595 | btrfs_mark_buffer_dirty(leaf); |
56bec294 CM |
1596 | btrfs_release_path(root->fs_info->extent_root, path); |
1597 | ||
1598 | path->reada = 1; | |
b9473439 CM |
1599 | path->leave_spinning = 1; |
1600 | ||
56bec294 CM |
1601 | /* now insert the actual backref */ |
1602 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1603 | path, bytenr, parent, root_objectid, |
1604 | owner, offset, refs_to_add); | |
56bec294 | 1605 | BUG_ON(ret); |
5d4f98a2 | 1606 | out: |
56bec294 | 1607 | btrfs_free_path(path); |
5d4f98a2 | 1608 | return err; |
56bec294 CM |
1609 | } |
1610 | ||
5d4f98a2 YZ |
1611 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1612 | struct btrfs_root *root, | |
1613 | struct btrfs_delayed_ref_node *node, | |
1614 | struct btrfs_delayed_extent_op *extent_op, | |
1615 | int insert_reserved) | |
56bec294 | 1616 | { |
5d4f98a2 YZ |
1617 | int ret = 0; |
1618 | struct btrfs_delayed_data_ref *ref; | |
1619 | struct btrfs_key ins; | |
1620 | u64 parent = 0; | |
1621 | u64 ref_root = 0; | |
1622 | u64 flags = 0; | |
1623 | ||
1624 | ins.objectid = node->bytenr; | |
1625 | ins.offset = node->num_bytes; | |
1626 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1627 | ||
1628 | ref = btrfs_delayed_node_to_data_ref(node); | |
1629 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1630 | parent = ref->parent; | |
1631 | else | |
1632 | ref_root = ref->root; | |
1633 | ||
1634 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1635 | if (extent_op) { | |
1636 | BUG_ON(extent_op->update_key); | |
1637 | flags |= extent_op->flags_to_set; | |
1638 | } | |
1639 | ret = alloc_reserved_file_extent(trans, root, | |
1640 | parent, ref_root, flags, | |
1641 | ref->objectid, ref->offset, | |
1642 | &ins, node->ref_mod); | |
1643 | update_reserved_extents(root, ins.objectid, ins.offset, 0); | |
1644 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { | |
1645 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1646 | node->num_bytes, parent, | |
1647 | ref_root, ref->objectid, | |
1648 | ref->offset, node->ref_mod, | |
1649 | extent_op); | |
1650 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1651 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1652 | node->num_bytes, parent, | |
1653 | ref_root, ref->objectid, | |
1654 | ref->offset, node->ref_mod, | |
1655 | extent_op); | |
1656 | } else { | |
1657 | BUG(); | |
1658 | } | |
1659 | return ret; | |
1660 | } | |
1661 | ||
1662 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1663 | struct extent_buffer *leaf, | |
1664 | struct btrfs_extent_item *ei) | |
1665 | { | |
1666 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1667 | if (extent_op->update_flags) { | |
1668 | flags |= extent_op->flags_to_set; | |
1669 | btrfs_set_extent_flags(leaf, ei, flags); | |
1670 | } | |
1671 | ||
1672 | if (extent_op->update_key) { | |
1673 | struct btrfs_tree_block_info *bi; | |
1674 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1675 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1676 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1677 | } | |
1678 | } | |
1679 | ||
1680 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
1681 | struct btrfs_root *root, | |
1682 | struct btrfs_delayed_ref_node *node, | |
1683 | struct btrfs_delayed_extent_op *extent_op) | |
1684 | { | |
1685 | struct btrfs_key key; | |
1686 | struct btrfs_path *path; | |
1687 | struct btrfs_extent_item *ei; | |
1688 | struct extent_buffer *leaf; | |
1689 | u32 item_size; | |
56bec294 | 1690 | int ret; |
5d4f98a2 YZ |
1691 | int err = 0; |
1692 | ||
1693 | path = btrfs_alloc_path(); | |
1694 | if (!path) | |
1695 | return -ENOMEM; | |
1696 | ||
1697 | key.objectid = node->bytenr; | |
1698 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1699 | key.offset = node->num_bytes; | |
1700 | ||
1701 | path->reada = 1; | |
1702 | path->leave_spinning = 1; | |
1703 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
1704 | path, 0, 1); | |
1705 | if (ret < 0) { | |
1706 | err = ret; | |
1707 | goto out; | |
1708 | } | |
1709 | if (ret > 0) { | |
1710 | err = -EIO; | |
1711 | goto out; | |
1712 | } | |
1713 | ||
1714 | leaf = path->nodes[0]; | |
1715 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1716 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1717 | if (item_size < sizeof(*ei)) { | |
1718 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
1719 | path, (u64)-1, 0); | |
1720 | if (ret < 0) { | |
1721 | err = ret; | |
1722 | goto out; | |
1723 | } | |
1724 | leaf = path->nodes[0]; | |
1725 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1726 | } | |
1727 | #endif | |
1728 | BUG_ON(item_size < sizeof(*ei)); | |
1729 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1730 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1731 | |
5d4f98a2 YZ |
1732 | btrfs_mark_buffer_dirty(leaf); |
1733 | out: | |
1734 | btrfs_free_path(path); | |
1735 | return err; | |
56bec294 CM |
1736 | } |
1737 | ||
5d4f98a2 YZ |
1738 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
1739 | struct btrfs_root *root, | |
1740 | struct btrfs_delayed_ref_node *node, | |
1741 | struct btrfs_delayed_extent_op *extent_op, | |
1742 | int insert_reserved) | |
56bec294 CM |
1743 | { |
1744 | int ret = 0; | |
5d4f98a2 YZ |
1745 | struct btrfs_delayed_tree_ref *ref; |
1746 | struct btrfs_key ins; | |
1747 | u64 parent = 0; | |
1748 | u64 ref_root = 0; | |
56bec294 | 1749 | |
5d4f98a2 YZ |
1750 | ins.objectid = node->bytenr; |
1751 | ins.offset = node->num_bytes; | |
1752 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 1753 | |
5d4f98a2 YZ |
1754 | ref = btrfs_delayed_node_to_tree_ref(node); |
1755 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
1756 | parent = ref->parent; | |
1757 | else | |
1758 | ref_root = ref->root; | |
1759 | ||
1760 | BUG_ON(node->ref_mod != 1); | |
1761 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1762 | BUG_ON(!extent_op || !extent_op->update_flags || | |
1763 | !extent_op->update_key); | |
1764 | ret = alloc_reserved_tree_block(trans, root, | |
1765 | parent, ref_root, | |
1766 | extent_op->flags_to_set, | |
1767 | &extent_op->key, | |
1768 | ref->level, &ins); | |
1769 | update_reserved_extents(root, ins.objectid, ins.offset, 0); | |
1770 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { | |
1771 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1772 | node->num_bytes, parent, ref_root, | |
1773 | ref->level, 0, 1, extent_op); | |
1774 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1775 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1776 | node->num_bytes, parent, ref_root, | |
1777 | ref->level, 0, 1, extent_op); | |
1778 | } else { | |
1779 | BUG(); | |
1780 | } | |
56bec294 CM |
1781 | return ret; |
1782 | } | |
1783 | ||
5d4f98a2 | 1784 | |
56bec294 | 1785 | /* helper function to actually process a single delayed ref entry */ |
5d4f98a2 YZ |
1786 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
1787 | struct btrfs_root *root, | |
1788 | struct btrfs_delayed_ref_node *node, | |
1789 | struct btrfs_delayed_extent_op *extent_op, | |
1790 | int insert_reserved) | |
56bec294 CM |
1791 | { |
1792 | int ret; | |
5d4f98a2 | 1793 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
1794 | struct btrfs_delayed_ref_head *head; |
1795 | /* | |
1796 | * we've hit the end of the chain and we were supposed | |
1797 | * to insert this extent into the tree. But, it got | |
1798 | * deleted before we ever needed to insert it, so all | |
1799 | * we have to do is clean up the accounting | |
1800 | */ | |
5d4f98a2 YZ |
1801 | BUG_ON(extent_op); |
1802 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 1803 | if (insert_reserved) { |
5d4f98a2 YZ |
1804 | if (head->is_data) { |
1805 | ret = btrfs_del_csums(trans, root, | |
1806 | node->bytenr, | |
1807 | node->num_bytes); | |
1808 | BUG_ON(ret); | |
1809 | } | |
1810 | btrfs_update_pinned_extents(root, node->bytenr, | |
68b38550 | 1811 | node->num_bytes, 1); |
56bec294 CM |
1812 | update_reserved_extents(root, node->bytenr, |
1813 | node->num_bytes, 0); | |
1814 | } | |
56bec294 CM |
1815 | mutex_unlock(&head->mutex); |
1816 | return 0; | |
1817 | } | |
1818 | ||
5d4f98a2 YZ |
1819 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1820 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
1821 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
1822 | insert_reserved); | |
1823 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1824 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1825 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
1826 | insert_reserved); | |
1827 | else | |
1828 | BUG(); | |
1829 | return ret; | |
56bec294 CM |
1830 | } |
1831 | ||
1832 | static noinline struct btrfs_delayed_ref_node * | |
1833 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
1834 | { | |
1835 | struct rb_node *node; | |
1836 | struct btrfs_delayed_ref_node *ref; | |
1837 | int action = BTRFS_ADD_DELAYED_REF; | |
1838 | again: | |
1839 | /* | |
1840 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1841 | * this prevents ref count from going down to zero when | |
1842 | * there still are pending delayed ref. | |
1843 | */ | |
1844 | node = rb_prev(&head->node.rb_node); | |
1845 | while (1) { | |
1846 | if (!node) | |
1847 | break; | |
1848 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
1849 | rb_node); | |
1850 | if (ref->bytenr != head->node.bytenr) | |
1851 | break; | |
5d4f98a2 | 1852 | if (ref->action == action) |
56bec294 CM |
1853 | return ref; |
1854 | node = rb_prev(node); | |
1855 | } | |
1856 | if (action == BTRFS_ADD_DELAYED_REF) { | |
1857 | action = BTRFS_DROP_DELAYED_REF; | |
1858 | goto again; | |
1859 | } | |
1860 | return NULL; | |
1861 | } | |
1862 | ||
c3e69d58 CM |
1863 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
1864 | struct btrfs_root *root, | |
1865 | struct list_head *cluster) | |
56bec294 | 1866 | { |
56bec294 CM |
1867 | struct btrfs_delayed_ref_root *delayed_refs; |
1868 | struct btrfs_delayed_ref_node *ref; | |
1869 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 1870 | struct btrfs_delayed_extent_op *extent_op; |
56bec294 | 1871 | int ret; |
c3e69d58 | 1872 | int count = 0; |
56bec294 | 1873 | int must_insert_reserved = 0; |
56bec294 CM |
1874 | |
1875 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
1876 | while (1) { |
1877 | if (!locked_ref) { | |
c3e69d58 CM |
1878 | /* pick a new head ref from the cluster list */ |
1879 | if (list_empty(cluster)) | |
56bec294 | 1880 | break; |
56bec294 | 1881 | |
c3e69d58 CM |
1882 | locked_ref = list_entry(cluster->next, |
1883 | struct btrfs_delayed_ref_head, cluster); | |
1884 | ||
1885 | /* grab the lock that says we are going to process | |
1886 | * all the refs for this head */ | |
1887 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
1888 | ||
1889 | /* | |
1890 | * we may have dropped the spin lock to get the head | |
1891 | * mutex lock, and that might have given someone else | |
1892 | * time to free the head. If that's true, it has been | |
1893 | * removed from our list and we can move on. | |
1894 | */ | |
1895 | if (ret == -EAGAIN) { | |
1896 | locked_ref = NULL; | |
1897 | count++; | |
1898 | continue; | |
56bec294 CM |
1899 | } |
1900 | } | |
a28ec197 | 1901 | |
56bec294 CM |
1902 | /* |
1903 | * record the must insert reserved flag before we | |
1904 | * drop the spin lock. | |
1905 | */ | |
1906 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1907 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 1908 | |
5d4f98a2 YZ |
1909 | extent_op = locked_ref->extent_op; |
1910 | locked_ref->extent_op = NULL; | |
1911 | ||
56bec294 CM |
1912 | /* |
1913 | * locked_ref is the head node, so we have to go one | |
1914 | * node back for any delayed ref updates | |
1915 | */ | |
56bec294 CM |
1916 | ref = select_delayed_ref(locked_ref); |
1917 | if (!ref) { | |
1918 | /* All delayed refs have been processed, Go ahead | |
1919 | * and send the head node to run_one_delayed_ref, | |
1920 | * so that any accounting fixes can happen | |
1921 | */ | |
1922 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
1923 | |
1924 | if (extent_op && must_insert_reserved) { | |
1925 | kfree(extent_op); | |
1926 | extent_op = NULL; | |
1927 | } | |
1928 | ||
1929 | if (extent_op) { | |
1930 | spin_unlock(&delayed_refs->lock); | |
1931 | ||
1932 | ret = run_delayed_extent_op(trans, root, | |
1933 | ref, extent_op); | |
1934 | BUG_ON(ret); | |
1935 | kfree(extent_op); | |
1936 | ||
1937 | cond_resched(); | |
1938 | spin_lock(&delayed_refs->lock); | |
1939 | continue; | |
1940 | } | |
1941 | ||
c3e69d58 | 1942 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
1943 | locked_ref = NULL; |
1944 | } | |
02217ed2 | 1945 | |
56bec294 CM |
1946 | ref->in_tree = 0; |
1947 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
1948 | delayed_refs->num_entries--; | |
5d4f98a2 | 1949 | |
56bec294 | 1950 | spin_unlock(&delayed_refs->lock); |
925baedd | 1951 | |
5d4f98a2 | 1952 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 CM |
1953 | must_insert_reserved); |
1954 | BUG_ON(ret); | |
eb099670 | 1955 | |
5d4f98a2 YZ |
1956 | btrfs_put_delayed_ref(ref); |
1957 | kfree(extent_op); | |
c3e69d58 | 1958 | count++; |
5d4f98a2 | 1959 | |
c3e69d58 CM |
1960 | cond_resched(); |
1961 | spin_lock(&delayed_refs->lock); | |
1962 | } | |
1963 | return count; | |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * this starts processing the delayed reference count updates and | |
1968 | * extent insertions we have queued up so far. count can be | |
1969 | * 0, which means to process everything in the tree at the start | |
1970 | * of the run (but not newly added entries), or it can be some target | |
1971 | * number you'd like to process. | |
1972 | */ | |
1973 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
1974 | struct btrfs_root *root, unsigned long count) | |
1975 | { | |
1976 | struct rb_node *node; | |
1977 | struct btrfs_delayed_ref_root *delayed_refs; | |
1978 | struct btrfs_delayed_ref_node *ref; | |
1979 | struct list_head cluster; | |
1980 | int ret; | |
1981 | int run_all = count == (unsigned long)-1; | |
1982 | int run_most = 0; | |
1983 | ||
1984 | if (root == root->fs_info->extent_root) | |
1985 | root = root->fs_info->tree_root; | |
1986 | ||
1987 | delayed_refs = &trans->transaction->delayed_refs; | |
1988 | INIT_LIST_HEAD(&cluster); | |
1989 | again: | |
1990 | spin_lock(&delayed_refs->lock); | |
1991 | if (count == 0) { | |
1992 | count = delayed_refs->num_entries * 2; | |
1993 | run_most = 1; | |
1994 | } | |
1995 | while (1) { | |
1996 | if (!(run_all || run_most) && | |
1997 | delayed_refs->num_heads_ready < 64) | |
1998 | break; | |
eb099670 | 1999 | |
56bec294 | 2000 | /* |
c3e69d58 CM |
2001 | * go find something we can process in the rbtree. We start at |
2002 | * the beginning of the tree, and then build a cluster | |
2003 | * of refs to process starting at the first one we are able to | |
2004 | * lock | |
56bec294 | 2005 | */ |
c3e69d58 CM |
2006 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2007 | delayed_refs->run_delayed_start); | |
2008 | if (ret) | |
56bec294 CM |
2009 | break; |
2010 | ||
c3e69d58 CM |
2011 | ret = run_clustered_refs(trans, root, &cluster); |
2012 | BUG_ON(ret < 0); | |
2013 | ||
2014 | count -= min_t(unsigned long, ret, count); | |
2015 | ||
2016 | if (count == 0) | |
2017 | break; | |
eb099670 | 2018 | } |
c3e69d58 | 2019 | |
56bec294 | 2020 | if (run_all) { |
56bec294 | 2021 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2022 | if (!node) |
56bec294 | 2023 | goto out; |
c3e69d58 | 2024 | count = (unsigned long)-1; |
e9d0b13b | 2025 | |
56bec294 CM |
2026 | while (node) { |
2027 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2028 | rb_node); | |
2029 | if (btrfs_delayed_ref_is_head(ref)) { | |
2030 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2031 | |
56bec294 CM |
2032 | head = btrfs_delayed_node_to_head(ref); |
2033 | atomic_inc(&ref->refs); | |
2034 | ||
2035 | spin_unlock(&delayed_refs->lock); | |
2036 | mutex_lock(&head->mutex); | |
2037 | mutex_unlock(&head->mutex); | |
2038 | ||
2039 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2040 | cond_resched(); |
56bec294 CM |
2041 | goto again; |
2042 | } | |
2043 | node = rb_next(node); | |
2044 | } | |
2045 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2046 | schedule_timeout(1); |
2047 | goto again; | |
5f39d397 | 2048 | } |
54aa1f4d | 2049 | out: |
c3e69d58 | 2050 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2051 | return 0; |
2052 | } | |
2053 | ||
5d4f98a2 YZ |
2054 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2055 | struct btrfs_root *root, | |
2056 | u64 bytenr, u64 num_bytes, u64 flags, | |
2057 | int is_data) | |
2058 | { | |
2059 | struct btrfs_delayed_extent_op *extent_op; | |
2060 | int ret; | |
2061 | ||
2062 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2063 | if (!extent_op) | |
2064 | return -ENOMEM; | |
2065 | ||
2066 | extent_op->flags_to_set = flags; | |
2067 | extent_op->update_flags = 1; | |
2068 | extent_op->update_key = 0; | |
2069 | extent_op->is_data = is_data ? 1 : 0; | |
2070 | ||
2071 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); | |
2072 | if (ret) | |
2073 | kfree(extent_op); | |
2074 | return ret; | |
2075 | } | |
2076 | ||
2077 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2078 | struct btrfs_root *root, | |
2079 | struct btrfs_path *path, | |
2080 | u64 objectid, u64 offset, u64 bytenr) | |
2081 | { | |
2082 | struct btrfs_delayed_ref_head *head; | |
2083 | struct btrfs_delayed_ref_node *ref; | |
2084 | struct btrfs_delayed_data_ref *data_ref; | |
2085 | struct btrfs_delayed_ref_root *delayed_refs; | |
2086 | struct rb_node *node; | |
2087 | int ret = 0; | |
2088 | ||
2089 | ret = -ENOENT; | |
2090 | delayed_refs = &trans->transaction->delayed_refs; | |
2091 | spin_lock(&delayed_refs->lock); | |
2092 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2093 | if (!head) | |
2094 | goto out; | |
2095 | ||
2096 | if (!mutex_trylock(&head->mutex)) { | |
2097 | atomic_inc(&head->node.refs); | |
2098 | spin_unlock(&delayed_refs->lock); | |
2099 | ||
2100 | btrfs_release_path(root->fs_info->extent_root, path); | |
2101 | ||
2102 | mutex_lock(&head->mutex); | |
2103 | mutex_unlock(&head->mutex); | |
2104 | btrfs_put_delayed_ref(&head->node); | |
2105 | return -EAGAIN; | |
2106 | } | |
2107 | ||
2108 | node = rb_prev(&head->node.rb_node); | |
2109 | if (!node) | |
2110 | goto out_unlock; | |
2111 | ||
2112 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2113 | ||
2114 | if (ref->bytenr != bytenr) | |
2115 | goto out_unlock; | |
2116 | ||
2117 | ret = 1; | |
2118 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2119 | goto out_unlock; | |
2120 | ||
2121 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2122 | ||
2123 | node = rb_prev(node); | |
2124 | if (node) { | |
2125 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2126 | if (ref->bytenr == bytenr) | |
2127 | goto out_unlock; | |
2128 | } | |
2129 | ||
2130 | if (data_ref->root != root->root_key.objectid || | |
2131 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2132 | goto out_unlock; | |
2133 | ||
2134 | ret = 0; | |
2135 | out_unlock: | |
2136 | mutex_unlock(&head->mutex); | |
2137 | out: | |
2138 | spin_unlock(&delayed_refs->lock); | |
2139 | return ret; | |
2140 | } | |
2141 | ||
2142 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2143 | struct btrfs_root *root, | |
2144 | struct btrfs_path *path, | |
2145 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2146 | { |
2147 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2148 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2149 | struct btrfs_extent_data_ref *ref; |
2150 | struct btrfs_extent_inline_ref *iref; | |
2151 | struct btrfs_extent_item *ei; | |
f321e491 | 2152 | struct btrfs_key key; |
5d4f98a2 | 2153 | u32 item_size; |
be20aa9d | 2154 | int ret; |
925baedd | 2155 | |
be20aa9d | 2156 | key.objectid = bytenr; |
31840ae1 | 2157 | key.offset = (u64)-1; |
f321e491 | 2158 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2159 | |
be20aa9d CM |
2160 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2161 | if (ret < 0) | |
2162 | goto out; | |
2163 | BUG_ON(ret == 0); | |
80ff3856 YZ |
2164 | |
2165 | ret = -ENOENT; | |
2166 | if (path->slots[0] == 0) | |
31840ae1 | 2167 | goto out; |
be20aa9d | 2168 | |
31840ae1 | 2169 | path->slots[0]--; |
f321e491 | 2170 | leaf = path->nodes[0]; |
5d4f98a2 | 2171 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2172 | |
5d4f98a2 | 2173 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2174 | goto out; |
f321e491 | 2175 | |
5d4f98a2 YZ |
2176 | ret = 1; |
2177 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2178 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2179 | if (item_size < sizeof(*ei)) { | |
2180 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2181 | goto out; | |
2182 | } | |
2183 | #endif | |
2184 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2185 | |
5d4f98a2 YZ |
2186 | if (item_size != sizeof(*ei) + |
2187 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2188 | goto out; | |
be20aa9d | 2189 | |
5d4f98a2 YZ |
2190 | if (btrfs_extent_generation(leaf, ei) <= |
2191 | btrfs_root_last_snapshot(&root->root_item)) | |
2192 | goto out; | |
2193 | ||
2194 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2195 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2196 | BTRFS_EXTENT_DATA_REF_KEY) | |
2197 | goto out; | |
2198 | ||
2199 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2200 | if (btrfs_extent_refs(leaf, ei) != | |
2201 | btrfs_extent_data_ref_count(leaf, ref) || | |
2202 | btrfs_extent_data_ref_root(leaf, ref) != | |
2203 | root->root_key.objectid || | |
2204 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2205 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2206 | goto out; | |
2207 | ||
2208 | ret = 0; | |
2209 | out: | |
2210 | return ret; | |
2211 | } | |
2212 | ||
2213 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2214 | struct btrfs_root *root, | |
2215 | u64 objectid, u64 offset, u64 bytenr) | |
2216 | { | |
2217 | struct btrfs_path *path; | |
2218 | int ret; | |
2219 | int ret2; | |
2220 | ||
2221 | path = btrfs_alloc_path(); | |
2222 | if (!path) | |
2223 | return -ENOENT; | |
2224 | ||
2225 | do { | |
2226 | ret = check_committed_ref(trans, root, path, objectid, | |
2227 | offset, bytenr); | |
2228 | if (ret && ret != -ENOENT) | |
f321e491 | 2229 | goto out; |
80ff3856 | 2230 | |
5d4f98a2 YZ |
2231 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2232 | offset, bytenr); | |
2233 | } while (ret2 == -EAGAIN); | |
2234 | ||
2235 | if (ret2 && ret2 != -ENOENT) { | |
2236 | ret = ret2; | |
2237 | goto out; | |
f321e491 | 2238 | } |
5d4f98a2 YZ |
2239 | |
2240 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2241 | ret = 0; | |
be20aa9d | 2242 | out: |
80ff3856 | 2243 | btrfs_free_path(path); |
f321e491 | 2244 | return ret; |
be20aa9d | 2245 | } |
c5739bba | 2246 | |
5d4f98a2 | 2247 | #if 0 |
31840ae1 ZY |
2248 | int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2249 | struct extent_buffer *buf, u32 nr_extents) | |
02217ed2 | 2250 | { |
5f39d397 | 2251 | struct btrfs_key key; |
6407bf6d | 2252 | struct btrfs_file_extent_item *fi; |
e4657689 ZY |
2253 | u64 root_gen; |
2254 | u32 nritems; | |
02217ed2 | 2255 | int i; |
db94535d | 2256 | int level; |
31840ae1 | 2257 | int ret = 0; |
e4657689 | 2258 | int shared = 0; |
a28ec197 | 2259 | |
3768f368 | 2260 | if (!root->ref_cows) |
a28ec197 | 2261 | return 0; |
5f39d397 | 2262 | |
e4657689 ZY |
2263 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
2264 | shared = 0; | |
2265 | root_gen = root->root_key.offset; | |
2266 | } else { | |
2267 | shared = 1; | |
2268 | root_gen = trans->transid - 1; | |
2269 | } | |
2270 | ||
db94535d | 2271 | level = btrfs_header_level(buf); |
5f39d397 | 2272 | nritems = btrfs_header_nritems(buf); |
4a096752 | 2273 | |
31840ae1 | 2274 | if (level == 0) { |
31153d81 YZ |
2275 | struct btrfs_leaf_ref *ref; |
2276 | struct btrfs_extent_info *info; | |
2277 | ||
31840ae1 | 2278 | ref = btrfs_alloc_leaf_ref(root, nr_extents); |
31153d81 | 2279 | if (!ref) { |
31840ae1 | 2280 | ret = -ENOMEM; |
31153d81 YZ |
2281 | goto out; |
2282 | } | |
2283 | ||
e4657689 | 2284 | ref->root_gen = root_gen; |
31153d81 YZ |
2285 | ref->bytenr = buf->start; |
2286 | ref->owner = btrfs_header_owner(buf); | |
2287 | ref->generation = btrfs_header_generation(buf); | |
31840ae1 | 2288 | ref->nritems = nr_extents; |
31153d81 | 2289 | info = ref->extents; |
bcc63abb | 2290 | |
31840ae1 | 2291 | for (i = 0; nr_extents > 0 && i < nritems; i++) { |
31153d81 YZ |
2292 | u64 disk_bytenr; |
2293 | btrfs_item_key_to_cpu(buf, &key, i); | |
2294 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
2295 | continue; | |
2296 | fi = btrfs_item_ptr(buf, i, | |
2297 | struct btrfs_file_extent_item); | |
2298 | if (btrfs_file_extent_type(buf, fi) == | |
2299 | BTRFS_FILE_EXTENT_INLINE) | |
2300 | continue; | |
2301 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2302 | if (disk_bytenr == 0) | |
2303 | continue; | |
2304 | ||
2305 | info->bytenr = disk_bytenr; | |
2306 | info->num_bytes = | |
2307 | btrfs_file_extent_disk_num_bytes(buf, fi); | |
2308 | info->objectid = key.objectid; | |
2309 | info->offset = key.offset; | |
2310 | info++; | |
2311 | } | |
2312 | ||
e4657689 | 2313 | ret = btrfs_add_leaf_ref(root, ref, shared); |
5b84e8d6 YZ |
2314 | if (ret == -EEXIST && shared) { |
2315 | struct btrfs_leaf_ref *old; | |
2316 | old = btrfs_lookup_leaf_ref(root, ref->bytenr); | |
2317 | BUG_ON(!old); | |
2318 | btrfs_remove_leaf_ref(root, old); | |
2319 | btrfs_free_leaf_ref(root, old); | |
2320 | ret = btrfs_add_leaf_ref(root, ref, shared); | |
2321 | } | |
31153d81 | 2322 | WARN_ON(ret); |
bcc63abb | 2323 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
2324 | } |
2325 | out: | |
31840ae1 ZY |
2326 | return ret; |
2327 | } | |
2328 | ||
b7a9f29f CM |
2329 | /* when a block goes through cow, we update the reference counts of |
2330 | * everything that block points to. The internal pointers of the block | |
2331 | * can be in just about any order, and it is likely to have clusters of | |
2332 | * things that are close together and clusters of things that are not. | |
2333 | * | |
2334 | * To help reduce the seeks that come with updating all of these reference | |
2335 | * counts, sort them by byte number before actual updates are done. | |
2336 | * | |
2337 | * struct refsort is used to match byte number to slot in the btree block. | |
2338 | * we sort based on the byte number and then use the slot to actually | |
2339 | * find the item. | |
bd56b302 CM |
2340 | * |
2341 | * struct refsort is smaller than strcut btrfs_item and smaller than | |
2342 | * struct btrfs_key_ptr. Since we're currently limited to the page size | |
2343 | * for a btree block, there's no way for a kmalloc of refsorts for a | |
2344 | * single node to be bigger than a page. | |
b7a9f29f CM |
2345 | */ |
2346 | struct refsort { | |
2347 | u64 bytenr; | |
2348 | u32 slot; | |
2349 | }; | |
2350 | ||
2351 | /* | |
2352 | * for passing into sort() | |
2353 | */ | |
2354 | static int refsort_cmp(const void *a_void, const void *b_void) | |
2355 | { | |
2356 | const struct refsort *a = a_void; | |
2357 | const struct refsort *b = b_void; | |
2358 | ||
2359 | if (a->bytenr < b->bytenr) | |
2360 | return -1; | |
2361 | if (a->bytenr > b->bytenr) | |
2362 | return 1; | |
2363 | return 0; | |
2364 | } | |
5d4f98a2 | 2365 | #endif |
b7a9f29f | 2366 | |
5d4f98a2 | 2367 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2368 | struct btrfs_root *root, |
5d4f98a2 YZ |
2369 | struct extent_buffer *buf, |
2370 | int full_backref, int inc) | |
31840ae1 ZY |
2371 | { |
2372 | u64 bytenr; | |
5d4f98a2 YZ |
2373 | u64 num_bytes; |
2374 | u64 parent; | |
31840ae1 | 2375 | u64 ref_root; |
31840ae1 | 2376 | u32 nritems; |
31840ae1 ZY |
2377 | struct btrfs_key key; |
2378 | struct btrfs_file_extent_item *fi; | |
2379 | int i; | |
2380 | int level; | |
2381 | int ret = 0; | |
31840ae1 | 2382 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
5d4f98a2 | 2383 | u64, u64, u64, u64, u64, u64); |
31840ae1 ZY |
2384 | |
2385 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2386 | nritems = btrfs_header_nritems(buf); |
2387 | level = btrfs_header_level(buf); | |
2388 | ||
5d4f98a2 YZ |
2389 | if (!root->ref_cows && level == 0) |
2390 | return 0; | |
31840ae1 | 2391 | |
5d4f98a2 YZ |
2392 | if (inc) |
2393 | process_func = btrfs_inc_extent_ref; | |
2394 | else | |
2395 | process_func = btrfs_free_extent; | |
31840ae1 | 2396 | |
5d4f98a2 YZ |
2397 | if (full_backref) |
2398 | parent = buf->start; | |
2399 | else | |
2400 | parent = 0; | |
2401 | ||
2402 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2403 | if (level == 0) { |
5d4f98a2 | 2404 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2405 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2406 | continue; | |
5d4f98a2 | 2407 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2408 | struct btrfs_file_extent_item); |
2409 | if (btrfs_file_extent_type(buf, fi) == | |
2410 | BTRFS_FILE_EXTENT_INLINE) | |
2411 | continue; | |
2412 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2413 | if (bytenr == 0) | |
2414 | continue; | |
5d4f98a2 YZ |
2415 | |
2416 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2417 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2418 | ret = process_func(trans, root, bytenr, num_bytes, | |
2419 | parent, ref_root, key.objectid, | |
2420 | key.offset); | |
31840ae1 ZY |
2421 | if (ret) |
2422 | goto fail; | |
2423 | } else { | |
5d4f98a2 YZ |
2424 | bytenr = btrfs_node_blockptr(buf, i); |
2425 | num_bytes = btrfs_level_size(root, level - 1); | |
2426 | ret = process_func(trans, root, bytenr, num_bytes, | |
2427 | parent, ref_root, level - 1, 0); | |
31840ae1 ZY |
2428 | if (ret) |
2429 | goto fail; | |
2430 | } | |
2431 | } | |
2432 | return 0; | |
2433 | fail: | |
5d4f98a2 YZ |
2434 | BUG(); |
2435 | return ret; | |
2436 | } | |
2437 | ||
2438 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2439 | struct extent_buffer *buf, int full_backref) | |
2440 | { | |
2441 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); | |
2442 | } | |
2443 | ||
2444 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2445 | struct extent_buffer *buf, int full_backref) | |
2446 | { | |
2447 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); | |
31840ae1 ZY |
2448 | } |
2449 | ||
9078a3e1 CM |
2450 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2451 | struct btrfs_root *root, | |
2452 | struct btrfs_path *path, | |
2453 | struct btrfs_block_group_cache *cache) | |
2454 | { | |
2455 | int ret; | |
9078a3e1 | 2456 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2457 | unsigned long bi; |
2458 | struct extent_buffer *leaf; | |
9078a3e1 | 2459 | |
9078a3e1 | 2460 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2461 | if (ret < 0) |
2462 | goto fail; | |
9078a3e1 | 2463 | BUG_ON(ret); |
5f39d397 CM |
2464 | |
2465 | leaf = path->nodes[0]; | |
2466 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2467 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2468 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 2469 | btrfs_release_path(extent_root, path); |
54aa1f4d | 2470 | fail: |
9078a3e1 CM |
2471 | if (ret) |
2472 | return ret; | |
9078a3e1 CM |
2473 | return 0; |
2474 | ||
2475 | } | |
2476 | ||
4a8c9a62 YZ |
2477 | static struct btrfs_block_group_cache * |
2478 | next_block_group(struct btrfs_root *root, | |
2479 | struct btrfs_block_group_cache *cache) | |
2480 | { | |
2481 | struct rb_node *node; | |
2482 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2483 | node = rb_next(&cache->cache_node); | |
2484 | btrfs_put_block_group(cache); | |
2485 | if (node) { | |
2486 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2487 | cache_node); | |
2488 | atomic_inc(&cache->count); | |
2489 | } else | |
2490 | cache = NULL; | |
2491 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2492 | return cache; | |
2493 | } | |
2494 | ||
96b5179d CM |
2495 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2496 | struct btrfs_root *root) | |
9078a3e1 | 2497 | { |
4a8c9a62 | 2498 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2499 | int err = 0; |
9078a3e1 | 2500 | struct btrfs_path *path; |
96b5179d | 2501 | u64 last = 0; |
9078a3e1 CM |
2502 | |
2503 | path = btrfs_alloc_path(); | |
2504 | if (!path) | |
2505 | return -ENOMEM; | |
2506 | ||
d397712b | 2507 | while (1) { |
4a8c9a62 YZ |
2508 | if (last == 0) { |
2509 | err = btrfs_run_delayed_refs(trans, root, | |
2510 | (unsigned long)-1); | |
2511 | BUG_ON(err); | |
0f9dd46c | 2512 | } |
54aa1f4d | 2513 | |
4a8c9a62 YZ |
2514 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2515 | while (cache) { | |
2516 | if (cache->dirty) | |
2517 | break; | |
2518 | cache = next_block_group(root, cache); | |
2519 | } | |
2520 | if (!cache) { | |
2521 | if (last == 0) | |
2522 | break; | |
2523 | last = 0; | |
2524 | continue; | |
2525 | } | |
0f9dd46c | 2526 | |
e8569813 | 2527 | cache->dirty = 0; |
4a8c9a62 | 2528 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 2529 | |
4a8c9a62 YZ |
2530 | err = write_one_cache_group(trans, root, path, cache); |
2531 | BUG_ON(err); | |
2532 | btrfs_put_block_group(cache); | |
9078a3e1 | 2533 | } |
4a8c9a62 | 2534 | |
9078a3e1 | 2535 | btrfs_free_path(path); |
4a8c9a62 | 2536 | return 0; |
9078a3e1 CM |
2537 | } |
2538 | ||
d2fb3437 YZ |
2539 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
2540 | { | |
2541 | struct btrfs_block_group_cache *block_group; | |
2542 | int readonly = 0; | |
2543 | ||
2544 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
2545 | if (!block_group || block_group->ro) | |
2546 | readonly = 1; | |
2547 | if (block_group) | |
fa9c0d79 | 2548 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2549 | return readonly; |
2550 | } | |
2551 | ||
593060d7 CM |
2552 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
2553 | u64 total_bytes, u64 bytes_used, | |
2554 | struct btrfs_space_info **space_info) | |
2555 | { | |
2556 | struct btrfs_space_info *found; | |
2557 | ||
2558 | found = __find_space_info(info, flags); | |
2559 | if (found) { | |
25179201 | 2560 | spin_lock(&found->lock); |
593060d7 CM |
2561 | found->total_bytes += total_bytes; |
2562 | found->bytes_used += bytes_used; | |
8f18cf13 | 2563 | found->full = 0; |
25179201 | 2564 | spin_unlock(&found->lock); |
593060d7 CM |
2565 | *space_info = found; |
2566 | return 0; | |
2567 | } | |
c146afad | 2568 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
2569 | if (!found) |
2570 | return -ENOMEM; | |
2571 | ||
0f9dd46c | 2572 | INIT_LIST_HEAD(&found->block_groups); |
80eb234a | 2573 | init_rwsem(&found->groups_sem); |
0f9dd46c | 2574 | spin_lock_init(&found->lock); |
593060d7 CM |
2575 | found->flags = flags; |
2576 | found->total_bytes = total_bytes; | |
2577 | found->bytes_used = bytes_used; | |
2578 | found->bytes_pinned = 0; | |
e8569813 | 2579 | found->bytes_reserved = 0; |
c146afad | 2580 | found->bytes_readonly = 0; |
6a63209f | 2581 | found->bytes_delalloc = 0; |
593060d7 | 2582 | found->full = 0; |
0ef3e66b | 2583 | found->force_alloc = 0; |
593060d7 | 2584 | *space_info = found; |
4184ea7f | 2585 | list_add_rcu(&found->list, &info->space_info); |
817d52f8 | 2586 | atomic_set(&found->caching_threads, 0); |
593060d7 CM |
2587 | return 0; |
2588 | } | |
2589 | ||
8790d502 CM |
2590 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
2591 | { | |
2592 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 2593 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 2594 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 2595 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
2596 | if (extra_flags) { |
2597 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
2598 | fs_info->avail_data_alloc_bits |= extra_flags; | |
2599 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
2600 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
2601 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
2602 | fs_info->avail_system_alloc_bits |= extra_flags; | |
2603 | } | |
2604 | } | |
593060d7 | 2605 | |
c146afad YZ |
2606 | static void set_block_group_readonly(struct btrfs_block_group_cache *cache) |
2607 | { | |
2608 | spin_lock(&cache->space_info->lock); | |
2609 | spin_lock(&cache->lock); | |
2610 | if (!cache->ro) { | |
2611 | cache->space_info->bytes_readonly += cache->key.offset - | |
2612 | btrfs_block_group_used(&cache->item); | |
2613 | cache->ro = 1; | |
2614 | } | |
2615 | spin_unlock(&cache->lock); | |
2616 | spin_unlock(&cache->space_info->lock); | |
2617 | } | |
2618 | ||
2b82032c | 2619 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 2620 | { |
2b82032c | 2621 | u64 num_devices = root->fs_info->fs_devices->rw_devices; |
a061fc8d CM |
2622 | |
2623 | if (num_devices == 1) | |
2624 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
2625 | if (num_devices < 4) | |
2626 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
2627 | ||
ec44a35c CM |
2628 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
2629 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 2630 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 2631 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 2632 | } |
ec44a35c CM |
2633 | |
2634 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 2635 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 2636 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 2637 | } |
ec44a35c CM |
2638 | |
2639 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
2640 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
2641 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
2642 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
2643 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
2644 | return flags; | |
2645 | } | |
2646 | ||
6a63209f JB |
2647 | static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data) |
2648 | { | |
2649 | struct btrfs_fs_info *info = root->fs_info; | |
2650 | u64 alloc_profile; | |
2651 | ||
2652 | if (data) { | |
2653 | alloc_profile = info->avail_data_alloc_bits & | |
2654 | info->data_alloc_profile; | |
2655 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
2656 | } else if (root == root->fs_info->chunk_root) { | |
2657 | alloc_profile = info->avail_system_alloc_bits & | |
2658 | info->system_alloc_profile; | |
2659 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
2660 | } else { | |
2661 | alloc_profile = info->avail_metadata_alloc_bits & | |
2662 | info->metadata_alloc_profile; | |
2663 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
2664 | } | |
2665 | ||
2666 | return btrfs_reduce_alloc_profile(root, data); | |
2667 | } | |
2668 | ||
2669 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) | |
2670 | { | |
2671 | u64 alloc_target; | |
2672 | ||
2673 | alloc_target = btrfs_get_alloc_profile(root, 1); | |
2674 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, | |
2675 | alloc_target); | |
2676 | } | |
2677 | ||
2678 | /* | |
2679 | * for now this just makes sure we have at least 5% of our metadata space free | |
2680 | * for use. | |
2681 | */ | |
2682 | int btrfs_check_metadata_free_space(struct btrfs_root *root) | |
2683 | { | |
2684 | struct btrfs_fs_info *info = root->fs_info; | |
2685 | struct btrfs_space_info *meta_sinfo; | |
2686 | u64 alloc_target, thresh; | |
4e06bdd6 | 2687 | int committed = 0, ret; |
6a63209f JB |
2688 | |
2689 | /* get the space info for where the metadata will live */ | |
2690 | alloc_target = btrfs_get_alloc_profile(root, 0); | |
2691 | meta_sinfo = __find_space_info(info, alloc_target); | |
2692 | ||
4e06bdd6 | 2693 | again: |
6a63209f | 2694 | spin_lock(&meta_sinfo->lock); |
4e06bdd6 JB |
2695 | if (!meta_sinfo->full) |
2696 | thresh = meta_sinfo->total_bytes * 80; | |
2697 | else | |
2698 | thresh = meta_sinfo->total_bytes * 95; | |
6a63209f JB |
2699 | |
2700 | do_div(thresh, 100); | |
2701 | ||
2702 | if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + | |
2703 | meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) { | |
4e06bdd6 JB |
2704 | struct btrfs_trans_handle *trans; |
2705 | if (!meta_sinfo->full) { | |
2706 | meta_sinfo->force_alloc = 1; | |
2707 | spin_unlock(&meta_sinfo->lock); | |
2708 | ||
2709 | trans = btrfs_start_transaction(root, 1); | |
2710 | if (!trans) | |
2711 | return -ENOMEM; | |
2712 | ||
2713 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
2714 | 2 * 1024 * 1024, alloc_target, 0); | |
2715 | btrfs_end_transaction(trans, root); | |
2716 | goto again; | |
2717 | } | |
6a63209f | 2718 | spin_unlock(&meta_sinfo->lock); |
4e06bdd6 JB |
2719 | |
2720 | if (!committed) { | |
2721 | committed = 1; | |
2722 | trans = btrfs_join_transaction(root, 1); | |
2723 | if (!trans) | |
2724 | return -ENOMEM; | |
2725 | ret = btrfs_commit_transaction(trans, root); | |
2726 | if (ret) | |
2727 | return ret; | |
2728 | goto again; | |
2729 | } | |
6a63209f JB |
2730 | return -ENOSPC; |
2731 | } | |
2732 | spin_unlock(&meta_sinfo->lock); | |
2733 | ||
2734 | return 0; | |
2735 | } | |
2736 | ||
2737 | /* | |
2738 | * This will check the space that the inode allocates from to make sure we have | |
2739 | * enough space for bytes. | |
2740 | */ | |
2741 | int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode, | |
2742 | u64 bytes) | |
2743 | { | |
2744 | struct btrfs_space_info *data_sinfo; | |
4e06bdd6 | 2745 | int ret = 0, committed = 0; |
6a63209f JB |
2746 | |
2747 | /* make sure bytes are sectorsize aligned */ | |
2748 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
2749 | ||
2750 | data_sinfo = BTRFS_I(inode)->space_info; | |
2751 | again: | |
2752 | /* make sure we have enough space to handle the data first */ | |
2753 | spin_lock(&data_sinfo->lock); | |
2754 | if (data_sinfo->total_bytes - data_sinfo->bytes_used - | |
2755 | data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved - | |
2756 | data_sinfo->bytes_pinned - data_sinfo->bytes_readonly - | |
2757 | data_sinfo->bytes_may_use < bytes) { | |
4e06bdd6 JB |
2758 | struct btrfs_trans_handle *trans; |
2759 | ||
6a63209f JB |
2760 | /* |
2761 | * if we don't have enough free bytes in this space then we need | |
2762 | * to alloc a new chunk. | |
2763 | */ | |
2764 | if (!data_sinfo->full) { | |
2765 | u64 alloc_target; | |
6a63209f JB |
2766 | |
2767 | data_sinfo->force_alloc = 1; | |
2768 | spin_unlock(&data_sinfo->lock); | |
2769 | ||
2770 | alloc_target = btrfs_get_alloc_profile(root, 1); | |
2771 | trans = btrfs_start_transaction(root, 1); | |
2772 | if (!trans) | |
2773 | return -ENOMEM; | |
2774 | ||
2775 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
2776 | bytes + 2 * 1024 * 1024, | |
2777 | alloc_target, 0); | |
2778 | btrfs_end_transaction(trans, root); | |
2779 | if (ret) | |
2780 | return ret; | |
2781 | goto again; | |
2782 | } | |
2783 | spin_unlock(&data_sinfo->lock); | |
4e06bdd6 JB |
2784 | |
2785 | /* commit the current transaction and try again */ | |
2786 | if (!committed) { | |
2787 | committed = 1; | |
2788 | trans = btrfs_join_transaction(root, 1); | |
2789 | if (!trans) | |
2790 | return -ENOMEM; | |
2791 | ret = btrfs_commit_transaction(trans, root); | |
2792 | if (ret) | |
2793 | return ret; | |
2794 | goto again; | |
2795 | } | |
2796 | ||
6a63209f JB |
2797 | printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes" |
2798 | ", %llu bytes_used, %llu bytes_reserved, " | |
68f5a38c | 2799 | "%llu bytes_pinned, %llu bytes_readonly, %llu may use " |
21380931 JB |
2800 | "%llu total\n", (unsigned long long)bytes, |
2801 | (unsigned long long)data_sinfo->bytes_delalloc, | |
2802 | (unsigned long long)data_sinfo->bytes_used, | |
2803 | (unsigned long long)data_sinfo->bytes_reserved, | |
2804 | (unsigned long long)data_sinfo->bytes_pinned, | |
2805 | (unsigned long long)data_sinfo->bytes_readonly, | |
2806 | (unsigned long long)data_sinfo->bytes_may_use, | |
2807 | (unsigned long long)data_sinfo->total_bytes); | |
6a63209f JB |
2808 | return -ENOSPC; |
2809 | } | |
2810 | data_sinfo->bytes_may_use += bytes; | |
2811 | BTRFS_I(inode)->reserved_bytes += bytes; | |
2812 | spin_unlock(&data_sinfo->lock); | |
2813 | ||
2814 | return btrfs_check_metadata_free_space(root); | |
2815 | } | |
2816 | ||
2817 | /* | |
2818 | * if there was an error for whatever reason after calling | |
2819 | * btrfs_check_data_free_space, call this so we can cleanup the counters. | |
2820 | */ | |
2821 | void btrfs_free_reserved_data_space(struct btrfs_root *root, | |
2822 | struct inode *inode, u64 bytes) | |
2823 | { | |
2824 | struct btrfs_space_info *data_sinfo; | |
2825 | ||
2826 | /* make sure bytes are sectorsize aligned */ | |
2827 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
2828 | ||
2829 | data_sinfo = BTRFS_I(inode)->space_info; | |
2830 | spin_lock(&data_sinfo->lock); | |
2831 | data_sinfo->bytes_may_use -= bytes; | |
2832 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
2833 | spin_unlock(&data_sinfo->lock); | |
2834 | } | |
2835 | ||
2836 | /* called when we are adding a delalloc extent to the inode's io_tree */ | |
2837 | void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode, | |
2838 | u64 bytes) | |
2839 | { | |
2840 | struct btrfs_space_info *data_sinfo; | |
2841 | ||
2842 | /* get the space info for where this inode will be storing its data */ | |
2843 | data_sinfo = BTRFS_I(inode)->space_info; | |
2844 | ||
2845 | /* make sure we have enough space to handle the data first */ | |
2846 | spin_lock(&data_sinfo->lock); | |
2847 | data_sinfo->bytes_delalloc += bytes; | |
2848 | ||
2849 | /* | |
2850 | * we are adding a delalloc extent without calling | |
2851 | * btrfs_check_data_free_space first. This happens on a weird | |
2852 | * writepage condition, but shouldn't hurt our accounting | |
2853 | */ | |
2854 | if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) { | |
2855 | data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes; | |
2856 | BTRFS_I(inode)->reserved_bytes = 0; | |
2857 | } else { | |
2858 | data_sinfo->bytes_may_use -= bytes; | |
2859 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
2860 | } | |
2861 | ||
2862 | spin_unlock(&data_sinfo->lock); | |
2863 | } | |
2864 | ||
2865 | /* called when we are clearing an delalloc extent from the inode's io_tree */ | |
2866 | void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode, | |
2867 | u64 bytes) | |
2868 | { | |
2869 | struct btrfs_space_info *info; | |
2870 | ||
2871 | info = BTRFS_I(inode)->space_info; | |
2872 | ||
2873 | spin_lock(&info->lock); | |
2874 | info->bytes_delalloc -= bytes; | |
2875 | spin_unlock(&info->lock); | |
2876 | } | |
2877 | ||
97e728d4 JB |
2878 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
2879 | { | |
2880 | struct list_head *head = &info->space_info; | |
2881 | struct btrfs_space_info *found; | |
2882 | ||
2883 | rcu_read_lock(); | |
2884 | list_for_each_entry_rcu(found, head, list) { | |
2885 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
2886 | found->force_alloc = 1; | |
2887 | } | |
2888 | rcu_read_unlock(); | |
2889 | } | |
2890 | ||
6324fbf3 CM |
2891 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2892 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 2893 | u64 flags, int force) |
6324fbf3 CM |
2894 | { |
2895 | struct btrfs_space_info *space_info; | |
97e728d4 | 2896 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6324fbf3 | 2897 | u64 thresh; |
c146afad YZ |
2898 | int ret = 0; |
2899 | ||
97e728d4 | 2900 | mutex_lock(&fs_info->chunk_mutex); |
6324fbf3 | 2901 | |
2b82032c | 2902 | flags = btrfs_reduce_alloc_profile(extent_root, flags); |
ec44a35c | 2903 | |
6324fbf3 | 2904 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
2905 | if (!space_info) { |
2906 | ret = update_space_info(extent_root->fs_info, flags, | |
2907 | 0, 0, &space_info); | |
2908 | BUG_ON(ret); | |
2909 | } | |
6324fbf3 CM |
2910 | BUG_ON(!space_info); |
2911 | ||
25179201 | 2912 | spin_lock(&space_info->lock); |
0ef3e66b CM |
2913 | if (space_info->force_alloc) { |
2914 | force = 1; | |
2915 | space_info->force_alloc = 0; | |
2916 | } | |
25179201 JB |
2917 | if (space_info->full) { |
2918 | spin_unlock(&space_info->lock); | |
925baedd | 2919 | goto out; |
25179201 | 2920 | } |
6324fbf3 | 2921 | |
c146afad YZ |
2922 | thresh = space_info->total_bytes - space_info->bytes_readonly; |
2923 | thresh = div_factor(thresh, 6); | |
0ef3e66b | 2924 | if (!force && |
e8569813 | 2925 | (space_info->bytes_used + space_info->bytes_pinned + |
25179201 JB |
2926 | space_info->bytes_reserved + alloc_bytes) < thresh) { |
2927 | spin_unlock(&space_info->lock); | |
925baedd | 2928 | goto out; |
25179201 | 2929 | } |
25179201 JB |
2930 | spin_unlock(&space_info->lock); |
2931 | ||
97e728d4 JB |
2932 | /* |
2933 | * if we're doing a data chunk, go ahead and make sure that | |
2934 | * we keep a reasonable number of metadata chunks allocated in the | |
2935 | * FS as well. | |
2936 | */ | |
2937 | if (flags & BTRFS_BLOCK_GROUP_DATA) { | |
2938 | fs_info->data_chunk_allocations++; | |
2939 | if (!(fs_info->data_chunk_allocations % | |
2940 | fs_info->metadata_ratio)) | |
2941 | force_metadata_allocation(fs_info); | |
2942 | } | |
2943 | ||
2b82032c | 2944 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
d397712b | 2945 | if (ret) |
6324fbf3 | 2946 | space_info->full = 1; |
a74a4b97 | 2947 | out: |
c146afad | 2948 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 2949 | return ret; |
6324fbf3 CM |
2950 | } |
2951 | ||
9078a3e1 CM |
2952 | static int update_block_group(struct btrfs_trans_handle *trans, |
2953 | struct btrfs_root *root, | |
db94535d | 2954 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 2955 | int mark_free) |
9078a3e1 CM |
2956 | { |
2957 | struct btrfs_block_group_cache *cache; | |
2958 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 2959 | u64 total = num_bytes; |
9078a3e1 | 2960 | u64 old_val; |
db94535d | 2961 | u64 byte_in_group; |
3e1ad54f | 2962 | |
5d4f98a2 YZ |
2963 | /* block accounting for super block */ |
2964 | spin_lock(&info->delalloc_lock); | |
2965 | old_val = btrfs_super_bytes_used(&info->super_copy); | |
2966 | if (alloc) | |
2967 | old_val += num_bytes; | |
2968 | else | |
2969 | old_val -= num_bytes; | |
2970 | btrfs_set_super_bytes_used(&info->super_copy, old_val); | |
2971 | ||
2972 | /* block accounting for root item */ | |
2973 | old_val = btrfs_root_used(&root->root_item); | |
2974 | if (alloc) | |
2975 | old_val += num_bytes; | |
2976 | else | |
2977 | old_val -= num_bytes; | |
2978 | btrfs_set_root_used(&root->root_item, old_val); | |
2979 | spin_unlock(&info->delalloc_lock); | |
2980 | ||
d397712b | 2981 | while (total) { |
db94535d | 2982 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 2983 | if (!cache) |
9078a3e1 | 2984 | return -1; |
db94535d CM |
2985 | byte_in_group = bytenr - cache->key.objectid; |
2986 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 2987 | |
25179201 | 2988 | spin_lock(&cache->space_info->lock); |
c286ac48 | 2989 | spin_lock(&cache->lock); |
0f9dd46c | 2990 | cache->dirty = 1; |
9078a3e1 | 2991 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 2992 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 2993 | if (alloc) { |
db94535d | 2994 | old_val += num_bytes; |
6324fbf3 | 2995 | cache->space_info->bytes_used += num_bytes; |
a512bbf8 | 2996 | if (cache->ro) |
c146afad | 2997 | cache->space_info->bytes_readonly -= num_bytes; |
c286ac48 CM |
2998 | btrfs_set_block_group_used(&cache->item, old_val); |
2999 | spin_unlock(&cache->lock); | |
25179201 | 3000 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 3001 | } else { |
db94535d | 3002 | old_val -= num_bytes; |
6324fbf3 | 3003 | cache->space_info->bytes_used -= num_bytes; |
c146afad YZ |
3004 | if (cache->ro) |
3005 | cache->space_info->bytes_readonly += num_bytes; | |
c286ac48 CM |
3006 | btrfs_set_block_group_used(&cache->item, old_val); |
3007 | spin_unlock(&cache->lock); | |
25179201 | 3008 | spin_unlock(&cache->space_info->lock); |
f510cfec | 3009 | if (mark_free) { |
0f9dd46c | 3010 | int ret; |
1f3c79a2 LH |
3011 | |
3012 | ret = btrfs_discard_extent(root, bytenr, | |
3013 | num_bytes); | |
3014 | WARN_ON(ret); | |
3015 | ||
0f9dd46c JB |
3016 | ret = btrfs_add_free_space(cache, bytenr, |
3017 | num_bytes); | |
d2fb3437 | 3018 | WARN_ON(ret); |
e37c9e69 | 3019 | } |
cd1bc465 | 3020 | } |
fa9c0d79 | 3021 | btrfs_put_block_group(cache); |
db94535d CM |
3022 | total -= num_bytes; |
3023 | bytenr += num_bytes; | |
9078a3e1 CM |
3024 | } |
3025 | return 0; | |
3026 | } | |
6324fbf3 | 3027 | |
a061fc8d CM |
3028 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
3029 | { | |
0f9dd46c | 3030 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 3031 | u64 bytenr; |
0f9dd46c JB |
3032 | |
3033 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
3034 | if (!cache) | |
a061fc8d | 3035 | return 0; |
0f9dd46c | 3036 | |
d2fb3437 | 3037 | bytenr = cache->key.objectid; |
fa9c0d79 | 3038 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
3039 | |
3040 | return bytenr; | |
a061fc8d CM |
3041 | } |
3042 | ||
e02119d5 | 3043 | int btrfs_update_pinned_extents(struct btrfs_root *root, |
68b38550 | 3044 | u64 bytenr, u64 num, int pin) |
324ae4df Y |
3045 | { |
3046 | u64 len; | |
3047 | struct btrfs_block_group_cache *cache; | |
3048 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3049 | ||
68b38550 | 3050 | if (pin) |
324ae4df Y |
3051 | set_extent_dirty(&fs_info->pinned_extents, |
3052 | bytenr, bytenr + num - 1, GFP_NOFS); | |
b9473439 | 3053 | |
324ae4df Y |
3054 | while (num > 0) { |
3055 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
e8569813 ZY |
3056 | BUG_ON(!cache); |
3057 | len = min(num, cache->key.offset - | |
3058 | (bytenr - cache->key.objectid)); | |
324ae4df | 3059 | if (pin) { |
25179201 | 3060 | spin_lock(&cache->space_info->lock); |
e8569813 ZY |
3061 | spin_lock(&cache->lock); |
3062 | cache->pinned += len; | |
3063 | cache->space_info->bytes_pinned += len; | |
3064 | spin_unlock(&cache->lock); | |
25179201 | 3065 | spin_unlock(&cache->space_info->lock); |
324ae4df Y |
3066 | fs_info->total_pinned += len; |
3067 | } else { | |
68b38550 JB |
3068 | int unpin = 0; |
3069 | ||
3070 | /* | |
3071 | * in order to not race with the block group caching, we | |
3072 | * only want to unpin the extent if we are cached. If | |
3073 | * we aren't cached, we want to start async caching this | |
3074 | * block group so we can free the extent the next time | |
3075 | * around. | |
3076 | */ | |
25179201 | 3077 | spin_lock(&cache->space_info->lock); |
e8569813 | 3078 | spin_lock(&cache->lock); |
68b38550 JB |
3079 | unpin = (cache->cached == BTRFS_CACHE_FINISHED); |
3080 | if (likely(unpin)) { | |
3081 | cache->pinned -= len; | |
3082 | cache->space_info->bytes_pinned -= len; | |
3083 | fs_info->total_pinned -= len; | |
3084 | } | |
e8569813 | 3085 | spin_unlock(&cache->lock); |
25179201 | 3086 | spin_unlock(&cache->space_info->lock); |
68b38550 JB |
3087 | |
3088 | if (likely(unpin)) | |
3089 | clear_extent_dirty(&fs_info->pinned_extents, | |
3090 | bytenr, bytenr + len -1, | |
3091 | GFP_NOFS); | |
3092 | else | |
3093 | cache_block_group(cache); | |
3094 | ||
3095 | if (unpin) | |
07103a3c | 3096 | btrfs_add_free_space(cache, bytenr, len); |
324ae4df | 3097 | } |
fa9c0d79 | 3098 | btrfs_put_block_group(cache); |
324ae4df Y |
3099 | bytenr += len; |
3100 | num -= len; | |
3101 | } | |
3102 | return 0; | |
3103 | } | |
9078a3e1 | 3104 | |
e8569813 ZY |
3105 | static int update_reserved_extents(struct btrfs_root *root, |
3106 | u64 bytenr, u64 num, int reserve) | |
3107 | { | |
3108 | u64 len; | |
3109 | struct btrfs_block_group_cache *cache; | |
3110 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3111 | ||
e8569813 ZY |
3112 | while (num > 0) { |
3113 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
3114 | BUG_ON(!cache); | |
3115 | len = min(num, cache->key.offset - | |
3116 | (bytenr - cache->key.objectid)); | |
25179201 JB |
3117 | |
3118 | spin_lock(&cache->space_info->lock); | |
3119 | spin_lock(&cache->lock); | |
e8569813 | 3120 | if (reserve) { |
e8569813 ZY |
3121 | cache->reserved += len; |
3122 | cache->space_info->bytes_reserved += len; | |
e8569813 | 3123 | } else { |
e8569813 ZY |
3124 | cache->reserved -= len; |
3125 | cache->space_info->bytes_reserved -= len; | |
e8569813 | 3126 | } |
25179201 JB |
3127 | spin_unlock(&cache->lock); |
3128 | spin_unlock(&cache->space_info->lock); | |
fa9c0d79 | 3129 | btrfs_put_block_group(cache); |
e8569813 ZY |
3130 | bytenr += len; |
3131 | num -= len; | |
3132 | } | |
3133 | return 0; | |
3134 | } | |
3135 | ||
d1310b2e | 3136 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) |
ccd467d6 | 3137 | { |
ccd467d6 | 3138 | u64 last = 0; |
1a5bc167 CM |
3139 | u64 start; |
3140 | u64 end; | |
d1310b2e | 3141 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; |
ccd467d6 | 3142 | int ret; |
ccd467d6 | 3143 | |
d397712b | 3144 | while (1) { |
1a5bc167 CM |
3145 | ret = find_first_extent_bit(pinned_extents, last, |
3146 | &start, &end, EXTENT_DIRTY); | |
3147 | if (ret) | |
ccd467d6 | 3148 | break; |
817d52f8 | 3149 | |
1a5bc167 CM |
3150 | set_extent_dirty(copy, start, end, GFP_NOFS); |
3151 | last = end + 1; | |
ccd467d6 CM |
3152 | } |
3153 | return 0; | |
3154 | } | |
3155 | ||
3156 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
3157 | struct btrfs_root *root, | |
d1310b2e | 3158 | struct extent_io_tree *unpin) |
a28ec197 | 3159 | { |
1a5bc167 CM |
3160 | u64 start; |
3161 | u64 end; | |
a28ec197 | 3162 | int ret; |
a28ec197 | 3163 | |
d397712b | 3164 | while (1) { |
1a5bc167 CM |
3165 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
3166 | EXTENT_DIRTY); | |
3167 | if (ret) | |
a28ec197 | 3168 | break; |
1f3c79a2 LH |
3169 | |
3170 | ret = btrfs_discard_extent(root, start, end + 1 - start); | |
3171 | ||
b9473439 | 3172 | /* unlocks the pinned mutex */ |
68b38550 | 3173 | btrfs_update_pinned_extents(root, start, end + 1 - start, 0); |
1a5bc167 | 3174 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
1f3c79a2 | 3175 | |
b9473439 | 3176 | cond_resched(); |
a28ec197 | 3177 | } |
817d52f8 | 3178 | |
1f3c79a2 | 3179 | return ret; |
a28ec197 CM |
3180 | } |
3181 | ||
31840ae1 ZY |
3182 | static int pin_down_bytes(struct btrfs_trans_handle *trans, |
3183 | struct btrfs_root *root, | |
b9473439 CM |
3184 | struct btrfs_path *path, |
3185 | u64 bytenr, u64 num_bytes, int is_data, | |
3186 | struct extent_buffer **must_clean) | |
e20d96d6 | 3187 | { |
1a5bc167 | 3188 | int err = 0; |
31840ae1 | 3189 | struct extent_buffer *buf; |
8ef97622 | 3190 | |
31840ae1 ZY |
3191 | if (is_data) |
3192 | goto pinit; | |
3193 | ||
3194 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); | |
3195 | if (!buf) | |
3196 | goto pinit; | |
3197 | ||
3198 | /* we can reuse a block if it hasn't been written | |
3199 | * and it is from this transaction. We can't | |
3200 | * reuse anything from the tree log root because | |
3201 | * it has tiny sub-transactions. | |
3202 | */ | |
3203 | if (btrfs_buffer_uptodate(buf, 0) && | |
3204 | btrfs_try_tree_lock(buf)) { | |
3205 | u64 header_owner = btrfs_header_owner(buf); | |
3206 | u64 header_transid = btrfs_header_generation(buf); | |
3207 | if (header_owner != BTRFS_TREE_LOG_OBJECTID && | |
3208 | header_transid == trans->transid && | |
3209 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
b9473439 | 3210 | *must_clean = buf; |
31840ae1 | 3211 | return 1; |
8ef97622 | 3212 | } |
31840ae1 | 3213 | btrfs_tree_unlock(buf); |
f4b9aa8d | 3214 | } |
31840ae1 ZY |
3215 | free_extent_buffer(buf); |
3216 | pinit: | |
b9473439 | 3217 | btrfs_set_path_blocking(path); |
b9473439 | 3218 | /* unlocks the pinned mutex */ |
68b38550 | 3219 | btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); |
31840ae1 | 3220 | |
be744175 | 3221 | BUG_ON(err < 0); |
e20d96d6 CM |
3222 | return 0; |
3223 | } | |
3224 | ||
5d4f98a2 YZ |
3225 | |
3226 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, | |
3227 | struct btrfs_root *root, | |
3228 | u64 bytenr, u64 num_bytes, u64 parent, | |
3229 | u64 root_objectid, u64 owner_objectid, | |
3230 | u64 owner_offset, int refs_to_drop, | |
3231 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 3232 | { |
e2fa7227 | 3233 | struct btrfs_key key; |
5d4f98a2 | 3234 | struct btrfs_path *path; |
1261ec42 CM |
3235 | struct btrfs_fs_info *info = root->fs_info; |
3236 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 3237 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3238 | struct btrfs_extent_item *ei; |
3239 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 3240 | int ret; |
5d4f98a2 | 3241 | int is_data; |
952fccac CM |
3242 | int extent_slot = 0; |
3243 | int found_extent = 0; | |
3244 | int num_to_del = 1; | |
5d4f98a2 YZ |
3245 | u32 item_size; |
3246 | u64 refs; | |
037e6390 | 3247 | |
5caf2a00 | 3248 | path = btrfs_alloc_path(); |
54aa1f4d CM |
3249 | if (!path) |
3250 | return -ENOMEM; | |
5f26f772 | 3251 | |
3c12ac72 | 3252 | path->reada = 1; |
b9473439 | 3253 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3254 | |
3255 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
3256 | BUG_ON(!is_data && refs_to_drop != 1); | |
3257 | ||
3258 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
3259 | bytenr, num_bytes, parent, | |
3260 | root_objectid, owner_objectid, | |
3261 | owner_offset); | |
7bb86316 | 3262 | if (ret == 0) { |
952fccac | 3263 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
3264 | while (extent_slot >= 0) { |
3265 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 3266 | extent_slot); |
5d4f98a2 | 3267 | if (key.objectid != bytenr) |
952fccac | 3268 | break; |
5d4f98a2 YZ |
3269 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
3270 | key.offset == num_bytes) { | |
952fccac CM |
3271 | found_extent = 1; |
3272 | break; | |
3273 | } | |
3274 | if (path->slots[0] - extent_slot > 5) | |
3275 | break; | |
5d4f98a2 | 3276 | extent_slot--; |
952fccac | 3277 | } |
5d4f98a2 YZ |
3278 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
3279 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
3280 | if (found_extent && item_size < sizeof(*ei)) | |
3281 | found_extent = 0; | |
3282 | #endif | |
31840ae1 | 3283 | if (!found_extent) { |
5d4f98a2 | 3284 | BUG_ON(iref); |
56bec294 | 3285 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
3286 | NULL, refs_to_drop, |
3287 | is_data); | |
31840ae1 ZY |
3288 | BUG_ON(ret); |
3289 | btrfs_release_path(extent_root, path); | |
b9473439 | 3290 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3291 | |
3292 | key.objectid = bytenr; | |
3293 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3294 | key.offset = num_bytes; | |
3295 | ||
31840ae1 ZY |
3296 | ret = btrfs_search_slot(trans, extent_root, |
3297 | &key, path, -1, 1); | |
f3465ca4 JB |
3298 | if (ret) { |
3299 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
3300 | ", was looking for %llu\n", ret, |
3301 | (unsigned long long)bytenr); | |
f3465ca4 JB |
3302 | btrfs_print_leaf(extent_root, path->nodes[0]); |
3303 | } | |
31840ae1 ZY |
3304 | BUG_ON(ret); |
3305 | extent_slot = path->slots[0]; | |
3306 | } | |
7bb86316 CM |
3307 | } else { |
3308 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
3309 | WARN_ON(1); | |
d397712b | 3310 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 3311 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 3312 | (unsigned long long)bytenr, |
56bec294 | 3313 | (unsigned long long)parent, |
d397712b | 3314 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
3315 | (unsigned long long)owner_objectid, |
3316 | (unsigned long long)owner_offset); | |
7bb86316 | 3317 | } |
5f39d397 CM |
3318 | |
3319 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
3320 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
3321 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3322 | if (item_size < sizeof(*ei)) { | |
3323 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
3324 | ret = convert_extent_item_v0(trans, extent_root, path, | |
3325 | owner_objectid, 0); | |
3326 | BUG_ON(ret < 0); | |
3327 | ||
3328 | btrfs_release_path(extent_root, path); | |
3329 | path->leave_spinning = 1; | |
3330 | ||
3331 | key.objectid = bytenr; | |
3332 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3333 | key.offset = num_bytes; | |
3334 | ||
3335 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
3336 | -1, 1); | |
3337 | if (ret) { | |
3338 | printk(KERN_ERR "umm, got %d back from search" | |
3339 | ", was looking for %llu\n", ret, | |
3340 | (unsigned long long)bytenr); | |
3341 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
3342 | } | |
3343 | BUG_ON(ret); | |
3344 | extent_slot = path->slots[0]; | |
3345 | leaf = path->nodes[0]; | |
3346 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
3347 | } | |
3348 | #endif | |
3349 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 3350 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3351 | struct btrfs_extent_item); |
5d4f98a2 YZ |
3352 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
3353 | struct btrfs_tree_block_info *bi; | |
3354 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
3355 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
3356 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3357 | } | |
56bec294 | 3358 | |
5d4f98a2 | 3359 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
3360 | BUG_ON(refs < refs_to_drop); |
3361 | refs -= refs_to_drop; | |
5f39d397 | 3362 | |
5d4f98a2 YZ |
3363 | if (refs > 0) { |
3364 | if (extent_op) | |
3365 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3366 | /* | |
3367 | * In the case of inline back ref, reference count will | |
3368 | * be updated by remove_extent_backref | |
952fccac | 3369 | */ |
5d4f98a2 YZ |
3370 | if (iref) { |
3371 | BUG_ON(!found_extent); | |
3372 | } else { | |
3373 | btrfs_set_extent_refs(leaf, ei, refs); | |
3374 | btrfs_mark_buffer_dirty(leaf); | |
3375 | } | |
3376 | if (found_extent) { | |
3377 | ret = remove_extent_backref(trans, extent_root, path, | |
3378 | iref, refs_to_drop, | |
3379 | is_data); | |
952fccac CM |
3380 | BUG_ON(ret); |
3381 | } | |
5d4f98a2 YZ |
3382 | } else { |
3383 | int mark_free = 0; | |
b9473439 | 3384 | struct extent_buffer *must_clean = NULL; |
78fae27e | 3385 | |
5d4f98a2 YZ |
3386 | if (found_extent) { |
3387 | BUG_ON(is_data && refs_to_drop != | |
3388 | extent_data_ref_count(root, path, iref)); | |
3389 | if (iref) { | |
3390 | BUG_ON(path->slots[0] != extent_slot); | |
3391 | } else { | |
3392 | BUG_ON(path->slots[0] != extent_slot + 1); | |
3393 | path->slots[0] = extent_slot; | |
3394 | num_to_del = 2; | |
3395 | } | |
78fae27e | 3396 | } |
b9473439 | 3397 | |
5d4f98a2 YZ |
3398 | ret = pin_down_bytes(trans, root, path, bytenr, |
3399 | num_bytes, is_data, &must_clean); | |
3400 | if (ret > 0) | |
3401 | mark_free = 1; | |
3402 | BUG_ON(ret < 0); | |
b9473439 CM |
3403 | /* |
3404 | * it is going to be very rare for someone to be waiting | |
3405 | * on the block we're freeing. del_items might need to | |
3406 | * schedule, so rather than get fancy, just force it | |
3407 | * to blocking here | |
3408 | */ | |
3409 | if (must_clean) | |
3410 | btrfs_set_lock_blocking(must_clean); | |
3411 | ||
952fccac CM |
3412 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3413 | num_to_del); | |
31840ae1 | 3414 | BUG_ON(ret); |
25179201 | 3415 | btrfs_release_path(extent_root, path); |
21af804c | 3416 | |
b9473439 CM |
3417 | if (must_clean) { |
3418 | clean_tree_block(NULL, root, must_clean); | |
3419 | btrfs_tree_unlock(must_clean); | |
3420 | free_extent_buffer(must_clean); | |
3421 | } | |
3422 | ||
5d4f98a2 | 3423 | if (is_data) { |
459931ec CM |
3424 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
3425 | BUG_ON(ret); | |
d57e62b8 CM |
3426 | } else { |
3427 | invalidate_mapping_pages(info->btree_inode->i_mapping, | |
3428 | bytenr >> PAGE_CACHE_SHIFT, | |
3429 | (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT); | |
459931ec CM |
3430 | } |
3431 | ||
dcbdd4dc CM |
3432 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
3433 | mark_free); | |
3434 | BUG_ON(ret); | |
a28ec197 | 3435 | } |
5caf2a00 | 3436 | btrfs_free_path(path); |
a28ec197 CM |
3437 | return ret; |
3438 | } | |
3439 | ||
1887be66 CM |
3440 | /* |
3441 | * when we free an extent, it is possible (and likely) that we free the last | |
3442 | * delayed ref for that extent as well. This searches the delayed ref tree for | |
3443 | * a given extent, and if there are no other delayed refs to be processed, it | |
3444 | * removes it from the tree. | |
3445 | */ | |
3446 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
3447 | struct btrfs_root *root, u64 bytenr) | |
3448 | { | |
3449 | struct btrfs_delayed_ref_head *head; | |
3450 | struct btrfs_delayed_ref_root *delayed_refs; | |
3451 | struct btrfs_delayed_ref_node *ref; | |
3452 | struct rb_node *node; | |
3453 | int ret; | |
3454 | ||
3455 | delayed_refs = &trans->transaction->delayed_refs; | |
3456 | spin_lock(&delayed_refs->lock); | |
3457 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
3458 | if (!head) | |
3459 | goto out; | |
3460 | ||
3461 | node = rb_prev(&head->node.rb_node); | |
3462 | if (!node) | |
3463 | goto out; | |
3464 | ||
3465 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
3466 | ||
3467 | /* there are still entries for this ref, we can't drop it */ | |
3468 | if (ref->bytenr == bytenr) | |
3469 | goto out; | |
3470 | ||
5d4f98a2 YZ |
3471 | if (head->extent_op) { |
3472 | if (!head->must_insert_reserved) | |
3473 | goto out; | |
3474 | kfree(head->extent_op); | |
3475 | head->extent_op = NULL; | |
3476 | } | |
3477 | ||
1887be66 CM |
3478 | /* |
3479 | * waiting for the lock here would deadlock. If someone else has it | |
3480 | * locked they are already in the process of dropping it anyway | |
3481 | */ | |
3482 | if (!mutex_trylock(&head->mutex)) | |
3483 | goto out; | |
3484 | ||
3485 | /* | |
3486 | * at this point we have a head with no other entries. Go | |
3487 | * ahead and process it. | |
3488 | */ | |
3489 | head->node.in_tree = 0; | |
3490 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 3491 | |
1887be66 CM |
3492 | delayed_refs->num_entries--; |
3493 | ||
3494 | /* | |
3495 | * we don't take a ref on the node because we're removing it from the | |
3496 | * tree, so we just steal the ref the tree was holding. | |
3497 | */ | |
c3e69d58 CM |
3498 | delayed_refs->num_heads--; |
3499 | if (list_empty(&head->cluster)) | |
3500 | delayed_refs->num_heads_ready--; | |
3501 | ||
3502 | list_del_init(&head->cluster); | |
1887be66 CM |
3503 | spin_unlock(&delayed_refs->lock); |
3504 | ||
3505 | ret = run_one_delayed_ref(trans, root->fs_info->tree_root, | |
5d4f98a2 YZ |
3506 | &head->node, head->extent_op, |
3507 | head->must_insert_reserved); | |
1887be66 CM |
3508 | BUG_ON(ret); |
3509 | btrfs_put_delayed_ref(&head->node); | |
3510 | return 0; | |
3511 | out: | |
3512 | spin_unlock(&delayed_refs->lock); | |
3513 | return 0; | |
3514 | } | |
3515 | ||
925baedd | 3516 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
3517 | struct btrfs_root *root, |
3518 | u64 bytenr, u64 num_bytes, u64 parent, | |
5d4f98a2 | 3519 | u64 root_objectid, u64 owner, u64 offset) |
925baedd CM |
3520 | { |
3521 | int ret; | |
3522 | ||
56bec294 CM |
3523 | /* |
3524 | * tree log blocks never actually go into the extent allocation | |
3525 | * tree, just update pinning info and exit early. | |
56bec294 | 3526 | */ |
5d4f98a2 YZ |
3527 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
3528 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 3529 | /* unlocks the pinned mutex */ |
68b38550 | 3530 | btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); |
56bec294 CM |
3531 | update_reserved_extents(root, bytenr, num_bytes, 0); |
3532 | ret = 0; | |
5d4f98a2 YZ |
3533 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
3534 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
3535 | parent, root_objectid, (int)owner, | |
3536 | BTRFS_DROP_DELAYED_REF, NULL); | |
1887be66 CM |
3537 | BUG_ON(ret); |
3538 | ret = check_ref_cleanup(trans, root, bytenr); | |
3539 | BUG_ON(ret); | |
5d4f98a2 YZ |
3540 | } else { |
3541 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
3542 | parent, root_objectid, owner, | |
3543 | offset, BTRFS_DROP_DELAYED_REF, NULL); | |
3544 | BUG_ON(ret); | |
56bec294 | 3545 | } |
925baedd CM |
3546 | return ret; |
3547 | } | |
3548 | ||
87ee04eb CM |
3549 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
3550 | { | |
3551 | u64 mask = ((u64)root->stripesize - 1); | |
3552 | u64 ret = (val + mask) & ~mask; | |
3553 | return ret; | |
3554 | } | |
3555 | ||
817d52f8 JB |
3556 | /* |
3557 | * when we wait for progress in the block group caching, its because | |
3558 | * our allocation attempt failed at least once. So, we must sleep | |
3559 | * and let some progress happen before we try again. | |
3560 | * | |
3561 | * This function will sleep at least once waiting for new free space to | |
3562 | * show up, and then it will check the block group free space numbers | |
3563 | * for our min num_bytes. Another option is to have it go ahead | |
3564 | * and look in the rbtree for a free extent of a given size, but this | |
3565 | * is a good start. | |
3566 | */ | |
3567 | static noinline int | |
3568 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
3569 | u64 num_bytes) | |
3570 | { | |
3571 | DEFINE_WAIT(wait); | |
3572 | ||
3573 | prepare_to_wait(&cache->caching_q, &wait, TASK_UNINTERRUPTIBLE); | |
3574 | ||
3575 | if (block_group_cache_done(cache)) { | |
3576 | finish_wait(&cache->caching_q, &wait); | |
3577 | return 0; | |
3578 | } | |
3579 | schedule(); | |
3580 | finish_wait(&cache->caching_q, &wait); | |
3581 | ||
3582 | wait_event(cache->caching_q, block_group_cache_done(cache) || | |
3583 | (cache->free_space >= num_bytes)); | |
3584 | return 0; | |
3585 | } | |
3586 | ||
3587 | enum btrfs_loop_type { | |
3588 | LOOP_CACHED_ONLY = 0, | |
3589 | LOOP_CACHING_NOWAIT = 1, | |
3590 | LOOP_CACHING_WAIT = 2, | |
3591 | LOOP_ALLOC_CHUNK = 3, | |
3592 | LOOP_NO_EMPTY_SIZE = 4, | |
3593 | }; | |
3594 | ||
fec577fb CM |
3595 | /* |
3596 | * walks the btree of allocated extents and find a hole of a given size. | |
3597 | * The key ins is changed to record the hole: | |
3598 | * ins->objectid == block start | |
62e2749e | 3599 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
3600 | * ins->offset == number of blocks |
3601 | * Any available blocks before search_start are skipped. | |
3602 | */ | |
d397712b | 3603 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
3604 | struct btrfs_root *orig_root, |
3605 | u64 num_bytes, u64 empty_size, | |
3606 | u64 search_start, u64 search_end, | |
3607 | u64 hint_byte, struct btrfs_key *ins, | |
3608 | u64 exclude_start, u64 exclude_nr, | |
3609 | int data) | |
fec577fb | 3610 | { |
80eb234a | 3611 | int ret = 0; |
d397712b | 3612 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 3613 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 3614 | struct btrfs_block_group_cache *block_group = NULL; |
239b14b3 | 3615 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 3616 | int allowed_chunk_alloc = 0; |
80eb234a | 3617 | struct btrfs_space_info *space_info; |
fa9c0d79 CM |
3618 | int last_ptr_loop = 0; |
3619 | int loop = 0; | |
817d52f8 | 3620 | bool found_uncached_bg = false; |
fec577fb | 3621 | |
db94535d | 3622 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 3623 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
3624 | ins->objectid = 0; |
3625 | ins->offset = 0; | |
b1a4d965 | 3626 | |
2552d17e JB |
3627 | space_info = __find_space_info(root->fs_info, data); |
3628 | ||
0ef3e66b CM |
3629 | if (orig_root->ref_cows || empty_size) |
3630 | allowed_chunk_alloc = 1; | |
3631 | ||
239b14b3 | 3632 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
fa9c0d79 | 3633 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
3634 | if (!btrfs_test_opt(root, SSD)) |
3635 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
3636 | } |
3637 | ||
fa9c0d79 CM |
3638 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { |
3639 | last_ptr = &root->fs_info->data_alloc_cluster; | |
3640 | } | |
0f9dd46c | 3641 | |
239b14b3 | 3642 | if (last_ptr) { |
fa9c0d79 CM |
3643 | spin_lock(&last_ptr->lock); |
3644 | if (last_ptr->block_group) | |
3645 | hint_byte = last_ptr->window_start; | |
3646 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 3647 | } |
fa9c0d79 | 3648 | |
a061fc8d | 3649 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 3650 | search_start = max(search_start, hint_byte); |
0b86a832 | 3651 | |
817d52f8 | 3652 | if (!last_ptr) |
fa9c0d79 | 3653 | empty_cluster = 0; |
fa9c0d79 | 3654 | |
2552d17e | 3655 | if (search_start == hint_byte) { |
2552d17e JB |
3656 | block_group = btrfs_lookup_block_group(root->fs_info, |
3657 | search_start); | |
817d52f8 JB |
3658 | /* |
3659 | * we don't want to use the block group if it doesn't match our | |
3660 | * allocation bits, or if its not cached. | |
3661 | */ | |
3662 | if (block_group && block_group_bits(block_group, data) && | |
3663 | block_group_cache_done(block_group)) { | |
2552d17e | 3664 | down_read(&space_info->groups_sem); |
44fb5511 CM |
3665 | if (list_empty(&block_group->list) || |
3666 | block_group->ro) { | |
3667 | /* | |
3668 | * someone is removing this block group, | |
3669 | * we can't jump into the have_block_group | |
3670 | * target because our list pointers are not | |
3671 | * valid | |
3672 | */ | |
3673 | btrfs_put_block_group(block_group); | |
3674 | up_read(&space_info->groups_sem); | |
3675 | } else | |
3676 | goto have_block_group; | |
2552d17e | 3677 | } else if (block_group) { |
fa9c0d79 | 3678 | btrfs_put_block_group(block_group); |
2552d17e | 3679 | } |
42e70e7a | 3680 | } |
4366211c | 3681 | |
2552d17e | 3682 | search: |
80eb234a | 3683 | down_read(&space_info->groups_sem); |
2552d17e | 3684 | list_for_each_entry(block_group, &space_info->block_groups, list) { |
6226cb0a | 3685 | u64 offset; |
817d52f8 | 3686 | int cached; |
8a1413a2 | 3687 | |
2552d17e JB |
3688 | atomic_inc(&block_group->count); |
3689 | search_start = block_group->key.objectid; | |
42e70e7a | 3690 | |
2552d17e | 3691 | have_block_group: |
817d52f8 JB |
3692 | if (unlikely(block_group->cached == BTRFS_CACHE_NO)) { |
3693 | /* | |
3694 | * we want to start caching kthreads, but not too many | |
3695 | * right off the bat so we don't overwhelm the system, | |
3696 | * so only start them if there are less than 2 and we're | |
3697 | * in the initial allocation phase. | |
3698 | */ | |
3699 | if (loop > LOOP_CACHING_NOWAIT || | |
3700 | atomic_read(&space_info->caching_threads) < 2) { | |
3701 | ret = cache_block_group(block_group); | |
3702 | BUG_ON(ret); | |
2552d17e | 3703 | } |
ea6a478e JB |
3704 | } |
3705 | ||
817d52f8 JB |
3706 | cached = block_group_cache_done(block_group); |
3707 | if (unlikely(!cached)) { | |
3708 | found_uncached_bg = true; | |
3709 | ||
3710 | /* if we only want cached bgs, loop */ | |
3711 | if (loop == LOOP_CACHED_ONLY) | |
3712 | goto loop; | |
3713 | } | |
3714 | ||
ea6a478e | 3715 | if (unlikely(block_group->ro)) |
2552d17e | 3716 | goto loop; |
0f9dd46c | 3717 | |
fa9c0d79 CM |
3718 | if (last_ptr) { |
3719 | /* | |
3720 | * the refill lock keeps out other | |
3721 | * people trying to start a new cluster | |
3722 | */ | |
3723 | spin_lock(&last_ptr->refill_lock); | |
44fb5511 CM |
3724 | if (last_ptr->block_group && |
3725 | (last_ptr->block_group->ro || | |
3726 | !block_group_bits(last_ptr->block_group, data))) { | |
3727 | offset = 0; | |
3728 | goto refill_cluster; | |
3729 | } | |
3730 | ||
fa9c0d79 CM |
3731 | offset = btrfs_alloc_from_cluster(block_group, last_ptr, |
3732 | num_bytes, search_start); | |
3733 | if (offset) { | |
3734 | /* we have a block, we're done */ | |
3735 | spin_unlock(&last_ptr->refill_lock); | |
3736 | goto checks; | |
3737 | } | |
3738 | ||
3739 | spin_lock(&last_ptr->lock); | |
3740 | /* | |
3741 | * whoops, this cluster doesn't actually point to | |
3742 | * this block group. Get a ref on the block | |
3743 | * group is does point to and try again | |
3744 | */ | |
3745 | if (!last_ptr_loop && last_ptr->block_group && | |
3746 | last_ptr->block_group != block_group) { | |
3747 | ||
3748 | btrfs_put_block_group(block_group); | |
3749 | block_group = last_ptr->block_group; | |
3750 | atomic_inc(&block_group->count); | |
3751 | spin_unlock(&last_ptr->lock); | |
3752 | spin_unlock(&last_ptr->refill_lock); | |
3753 | ||
3754 | last_ptr_loop = 1; | |
3755 | search_start = block_group->key.objectid; | |
44fb5511 CM |
3756 | /* |
3757 | * we know this block group is properly | |
3758 | * in the list because | |
3759 | * btrfs_remove_block_group, drops the | |
3760 | * cluster before it removes the block | |
3761 | * group from the list | |
3762 | */ | |
fa9c0d79 CM |
3763 | goto have_block_group; |
3764 | } | |
3765 | spin_unlock(&last_ptr->lock); | |
44fb5511 | 3766 | refill_cluster: |
fa9c0d79 CM |
3767 | /* |
3768 | * this cluster didn't work out, free it and | |
3769 | * start over | |
3770 | */ | |
3771 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3772 | ||
3773 | last_ptr_loop = 0; | |
3774 | ||
3775 | /* allocate a cluster in this block group */ | |
451d7585 | 3776 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 CM |
3777 | block_group, last_ptr, |
3778 | offset, num_bytes, | |
3779 | empty_cluster + empty_size); | |
3780 | if (ret == 0) { | |
3781 | /* | |
3782 | * now pull our allocation out of this | |
3783 | * cluster | |
3784 | */ | |
3785 | offset = btrfs_alloc_from_cluster(block_group, | |
3786 | last_ptr, num_bytes, | |
3787 | search_start); | |
3788 | if (offset) { | |
3789 | /* we found one, proceed */ | |
3790 | spin_unlock(&last_ptr->refill_lock); | |
3791 | goto checks; | |
3792 | } | |
817d52f8 JB |
3793 | } else if (!cached && loop > LOOP_CACHING_NOWAIT) { |
3794 | spin_unlock(&last_ptr->refill_lock); | |
3795 | ||
3796 | wait_block_group_cache_progress(block_group, | |
3797 | num_bytes + empty_cluster + empty_size); | |
3798 | goto have_block_group; | |
fa9c0d79 | 3799 | } |
817d52f8 | 3800 | |
fa9c0d79 CM |
3801 | /* |
3802 | * at this point we either didn't find a cluster | |
3803 | * or we weren't able to allocate a block from our | |
3804 | * cluster. Free the cluster we've been trying | |
3805 | * to use, and go to the next block group | |
3806 | */ | |
817d52f8 | 3807 | if (loop < LOOP_NO_EMPTY_SIZE) { |
fa9c0d79 CM |
3808 | btrfs_return_cluster_to_free_space(NULL, |
3809 | last_ptr); | |
3810 | spin_unlock(&last_ptr->refill_lock); | |
3811 | goto loop; | |
3812 | } | |
3813 | spin_unlock(&last_ptr->refill_lock); | |
3814 | } | |
3815 | ||
6226cb0a JB |
3816 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
3817 | num_bytes, empty_size); | |
817d52f8 JB |
3818 | if (!offset && (cached || (!cached && |
3819 | loop == LOOP_CACHING_NOWAIT))) { | |
2552d17e | 3820 | goto loop; |
817d52f8 JB |
3821 | } else if (!offset && (!cached && |
3822 | loop > LOOP_CACHING_NOWAIT)) { | |
3823 | wait_block_group_cache_progress(block_group, | |
3824 | num_bytes + empty_size); | |
3825 | goto have_block_group; | |
3826 | } | |
fa9c0d79 | 3827 | checks: |
6226cb0a | 3828 | search_start = stripe_align(root, offset); |
2552d17e | 3829 | /* move on to the next group */ |
6226cb0a JB |
3830 | if (search_start + num_bytes >= search_end) { |
3831 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 3832 | goto loop; |
6226cb0a | 3833 | } |
25179201 | 3834 | |
2552d17e JB |
3835 | /* move on to the next group */ |
3836 | if (search_start + num_bytes > | |
6226cb0a JB |
3837 | block_group->key.objectid + block_group->key.offset) { |
3838 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 3839 | goto loop; |
6226cb0a | 3840 | } |
f5a31e16 | 3841 | |
2552d17e JB |
3842 | if (exclude_nr > 0 && |
3843 | (search_start + num_bytes > exclude_start && | |
3844 | search_start < exclude_start + exclude_nr)) { | |
3845 | search_start = exclude_start + exclude_nr; | |
3846 | ||
6226cb0a | 3847 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e JB |
3848 | /* |
3849 | * if search_start is still in this block group | |
3850 | * then we just re-search this block group | |
f5a31e16 | 3851 | */ |
2552d17e JB |
3852 | if (search_start >= block_group->key.objectid && |
3853 | search_start < (block_group->key.objectid + | |
6226cb0a | 3854 | block_group->key.offset)) |
2552d17e | 3855 | goto have_block_group; |
2552d17e | 3856 | goto loop; |
0f9dd46c | 3857 | } |
0b86a832 | 3858 | |
2552d17e JB |
3859 | ins->objectid = search_start; |
3860 | ins->offset = num_bytes; | |
d2fb3437 | 3861 | |
6226cb0a JB |
3862 | if (offset < search_start) |
3863 | btrfs_add_free_space(block_group, offset, | |
3864 | search_start - offset); | |
3865 | BUG_ON(offset > search_start); | |
3866 | ||
2552d17e | 3867 | /* we are all good, lets return */ |
2552d17e JB |
3868 | break; |
3869 | loop: | |
fa9c0d79 | 3870 | btrfs_put_block_group(block_group); |
2552d17e JB |
3871 | } |
3872 | up_read(&space_info->groups_sem); | |
3873 | ||
817d52f8 JB |
3874 | /* LOOP_CACHED_ONLY, only search fully cached block groups |
3875 | * LOOP_CACHING_NOWAIT, search partially cached block groups, but | |
3876 | * dont wait foR them to finish caching | |
3877 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
3878 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
3879 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
3880 | * again | |
fa9c0d79 | 3881 | */ |
817d52f8 JB |
3882 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE && |
3883 | (found_uncached_bg || empty_size || empty_cluster || | |
3884 | allowed_chunk_alloc)) { | |
3885 | if (found_uncached_bg) { | |
3886 | found_uncached_bg = false; | |
3887 | if (loop < LOOP_CACHING_WAIT) { | |
3888 | loop++; | |
3889 | goto search; | |
3890 | } | |
3891 | } | |
3892 | ||
3893 | if (loop == LOOP_ALLOC_CHUNK) { | |
fa9c0d79 CM |
3894 | empty_size = 0; |
3895 | empty_cluster = 0; | |
3896 | } | |
2552d17e JB |
3897 | |
3898 | if (allowed_chunk_alloc) { | |
3899 | ret = do_chunk_alloc(trans, root, num_bytes + | |
3900 | 2 * 1024 * 1024, data, 1); | |
2552d17e JB |
3901 | allowed_chunk_alloc = 0; |
3902 | } else { | |
3903 | space_info->force_alloc = 1; | |
3904 | } | |
3905 | ||
817d52f8 | 3906 | if (loop < LOOP_NO_EMPTY_SIZE) { |
fa9c0d79 | 3907 | loop++; |
2552d17e | 3908 | goto search; |
fa9c0d79 | 3909 | } |
2552d17e JB |
3910 | ret = -ENOSPC; |
3911 | } else if (!ins->objectid) { | |
3912 | ret = -ENOSPC; | |
f2654de4 | 3913 | } |
0b86a832 | 3914 | |
80eb234a JB |
3915 | /* we found what we needed */ |
3916 | if (ins->objectid) { | |
3917 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) | |
d2fb3437 | 3918 | trans->block_group = block_group->key.objectid; |
0f9dd46c | 3919 | |
fa9c0d79 | 3920 | btrfs_put_block_group(block_group); |
80eb234a | 3921 | ret = 0; |
be744175 | 3922 | } |
be744175 | 3923 | |
0f70abe2 | 3924 | return ret; |
fec577fb | 3925 | } |
ec44a35c | 3926 | |
0f9dd46c JB |
3927 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes) |
3928 | { | |
3929 | struct btrfs_block_group_cache *cache; | |
0f9dd46c | 3930 | |
d397712b CM |
3931 | printk(KERN_INFO "space_info has %llu free, is %sfull\n", |
3932 | (unsigned long long)(info->total_bytes - info->bytes_used - | |
3933 | info->bytes_pinned - info->bytes_reserved), | |
3934 | (info->full) ? "" : "not "); | |
6a63209f | 3935 | printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu," |
21380931 JB |
3936 | " may_use=%llu, used=%llu\n", |
3937 | (unsigned long long)info->total_bytes, | |
3938 | (unsigned long long)info->bytes_pinned, | |
3939 | (unsigned long long)info->bytes_delalloc, | |
3940 | (unsigned long long)info->bytes_may_use, | |
3941 | (unsigned long long)info->bytes_used); | |
0f9dd46c | 3942 | |
80eb234a | 3943 | down_read(&info->groups_sem); |
c6e30871 | 3944 | list_for_each_entry(cache, &info->block_groups, list) { |
0f9dd46c | 3945 | spin_lock(&cache->lock); |
d397712b CM |
3946 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
3947 | "%llu pinned %llu reserved\n", | |
3948 | (unsigned long long)cache->key.objectid, | |
3949 | (unsigned long long)cache->key.offset, | |
3950 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
3951 | (unsigned long long)cache->pinned, | |
3952 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
3953 | btrfs_dump_free_space(cache, bytes); |
3954 | spin_unlock(&cache->lock); | |
3955 | } | |
80eb234a | 3956 | up_read(&info->groups_sem); |
0f9dd46c | 3957 | } |
e8569813 | 3958 | |
e6dcd2dc CM |
3959 | static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
3960 | struct btrfs_root *root, | |
3961 | u64 num_bytes, u64 min_alloc_size, | |
3962 | u64 empty_size, u64 hint_byte, | |
3963 | u64 search_end, struct btrfs_key *ins, | |
3964 | u64 data) | |
fec577fb CM |
3965 | { |
3966 | int ret; | |
fbdc762b | 3967 | u64 search_start = 0; |
1261ec42 | 3968 | struct btrfs_fs_info *info = root->fs_info; |
925baedd | 3969 | |
6a63209f | 3970 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 3971 | again: |
0ef3e66b CM |
3972 | /* |
3973 | * the only place that sets empty_size is btrfs_realloc_node, which | |
3974 | * is not called recursively on allocations | |
3975 | */ | |
3976 | if (empty_size || root->ref_cows) { | |
593060d7 | 3977 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 3978 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b CM |
3979 | 2 * 1024 * 1024, |
3980 | BTRFS_BLOCK_GROUP_METADATA | | |
3981 | (info->metadata_alloc_profile & | |
3982 | info->avail_metadata_alloc_bits), 0); | |
6324fbf3 CM |
3983 | } |
3984 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
0ef3e66b | 3985 | num_bytes + 2 * 1024 * 1024, data, 0); |
6324fbf3 | 3986 | } |
0b86a832 | 3987 | |
db94535d CM |
3988 | WARN_ON(num_bytes < root->sectorsize); |
3989 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
3990 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
3991 | trans->alloc_exclude_start, |
3992 | trans->alloc_exclude_nr, data); | |
3b951516 | 3993 | |
98d20f67 CM |
3994 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
3995 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 3996 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 3997 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b CM |
3998 | do_chunk_alloc(trans, root->fs_info->extent_root, |
3999 | num_bytes, data, 1); | |
98d20f67 CM |
4000 | goto again; |
4001 | } | |
817d52f8 | 4002 | if (ret == -ENOSPC) { |
0f9dd46c JB |
4003 | struct btrfs_space_info *sinfo; |
4004 | ||
4005 | sinfo = __find_space_info(root->fs_info, data); | |
d397712b CM |
4006 | printk(KERN_ERR "btrfs allocation failed flags %llu, " |
4007 | "wanted %llu\n", (unsigned long long)data, | |
4008 | (unsigned long long)num_bytes); | |
0f9dd46c | 4009 | dump_space_info(sinfo, num_bytes); |
925baedd | 4010 | } |
0f9dd46c JB |
4011 | |
4012 | return ret; | |
e6dcd2dc CM |
4013 | } |
4014 | ||
65b51a00 CM |
4015 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
4016 | { | |
0f9dd46c | 4017 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 4018 | int ret = 0; |
0f9dd46c | 4019 | |
0f9dd46c JB |
4020 | cache = btrfs_lookup_block_group(root->fs_info, start); |
4021 | if (!cache) { | |
d397712b CM |
4022 | printk(KERN_ERR "Unable to find block group for %llu\n", |
4023 | (unsigned long long)start); | |
0f9dd46c JB |
4024 | return -ENOSPC; |
4025 | } | |
1f3c79a2 LH |
4026 | |
4027 | ret = btrfs_discard_extent(root, start, len); | |
4028 | ||
0f9dd46c | 4029 | btrfs_add_free_space(cache, start, len); |
fa9c0d79 | 4030 | btrfs_put_block_group(cache); |
1a40e23b | 4031 | update_reserved_extents(root, start, len, 0); |
1f3c79a2 LH |
4032 | |
4033 | return ret; | |
65b51a00 CM |
4034 | } |
4035 | ||
e6dcd2dc CM |
4036 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
4037 | struct btrfs_root *root, | |
4038 | u64 num_bytes, u64 min_alloc_size, | |
4039 | u64 empty_size, u64 hint_byte, | |
4040 | u64 search_end, struct btrfs_key *ins, | |
4041 | u64 data) | |
4042 | { | |
4043 | int ret; | |
e6dcd2dc CM |
4044 | ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, |
4045 | empty_size, hint_byte, search_end, ins, | |
4046 | data); | |
817d52f8 JB |
4047 | if (!ret) |
4048 | update_reserved_extents(root, ins->objectid, ins->offset, 1); | |
4049 | ||
e6dcd2dc CM |
4050 | return ret; |
4051 | } | |
4052 | ||
5d4f98a2 YZ |
4053 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
4054 | struct btrfs_root *root, | |
4055 | u64 parent, u64 root_objectid, | |
4056 | u64 flags, u64 owner, u64 offset, | |
4057 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
4058 | { |
4059 | int ret; | |
5d4f98a2 | 4060 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 4061 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4062 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4063 | struct btrfs_path *path; |
5d4f98a2 YZ |
4064 | struct extent_buffer *leaf; |
4065 | int type; | |
4066 | u32 size; | |
26b8003f | 4067 | |
5d4f98a2 YZ |
4068 | if (parent > 0) |
4069 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4070 | else | |
4071 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4072 | |
5d4f98a2 | 4073 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4074 | |
4075 | path = btrfs_alloc_path(); | |
4076 | BUG_ON(!path); | |
47e4bb98 | 4077 | |
b9473439 | 4078 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4079 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4080 | ins, size); | |
ccd467d6 | 4081 | BUG_ON(ret); |
0f9dd46c | 4082 | |
5d4f98a2 YZ |
4083 | leaf = path->nodes[0]; |
4084 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4085 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4086 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4087 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4088 | btrfs_set_extent_flags(leaf, extent_item, | |
4089 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4090 | ||
4091 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4092 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4093 | if (parent > 0) { | |
4094 | struct btrfs_shared_data_ref *ref; | |
4095 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4096 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4097 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4098 | } else { | |
4099 | struct btrfs_extent_data_ref *ref; | |
4100 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4101 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4102 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4103 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4104 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4105 | } | |
47e4bb98 CM |
4106 | |
4107 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4108 | btrfs_free_path(path); |
f510cfec | 4109 | |
5d4f98a2 YZ |
4110 | ret = update_block_group(trans, root, ins->objectid, ins->offset, |
4111 | 1, 0); | |
f5947066 | 4112 | if (ret) { |
d397712b CM |
4113 | printk(KERN_ERR "btrfs update block group failed for %llu " |
4114 | "%llu\n", (unsigned long long)ins->objectid, | |
4115 | (unsigned long long)ins->offset); | |
f5947066 CM |
4116 | BUG(); |
4117 | } | |
e6dcd2dc CM |
4118 | return ret; |
4119 | } | |
4120 | ||
5d4f98a2 YZ |
4121 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4122 | struct btrfs_root *root, | |
4123 | u64 parent, u64 root_objectid, | |
4124 | u64 flags, struct btrfs_disk_key *key, | |
4125 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
4126 | { |
4127 | int ret; | |
5d4f98a2 YZ |
4128 | struct btrfs_fs_info *fs_info = root->fs_info; |
4129 | struct btrfs_extent_item *extent_item; | |
4130 | struct btrfs_tree_block_info *block_info; | |
4131 | struct btrfs_extent_inline_ref *iref; | |
4132 | struct btrfs_path *path; | |
4133 | struct extent_buffer *leaf; | |
4134 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 4135 | |
5d4f98a2 YZ |
4136 | path = btrfs_alloc_path(); |
4137 | BUG_ON(!path); | |
56bec294 | 4138 | |
5d4f98a2 YZ |
4139 | path->leave_spinning = 1; |
4140 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4141 | ins, size); | |
56bec294 | 4142 | BUG_ON(ret); |
5d4f98a2 YZ |
4143 | |
4144 | leaf = path->nodes[0]; | |
4145 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4146 | struct btrfs_extent_item); | |
4147 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4148 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4149 | btrfs_set_extent_flags(leaf, extent_item, | |
4150 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
4151 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
4152 | ||
4153 | btrfs_set_tree_block_key(leaf, block_info, key); | |
4154 | btrfs_set_tree_block_level(leaf, block_info, level); | |
4155 | ||
4156 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
4157 | if (parent > 0) { | |
4158 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
4159 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4160 | BTRFS_SHARED_BLOCK_REF_KEY); | |
4161 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4162 | } else { | |
4163 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4164 | BTRFS_TREE_BLOCK_REF_KEY); | |
4165 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
4166 | } | |
4167 | ||
4168 | btrfs_mark_buffer_dirty(leaf); | |
4169 | btrfs_free_path(path); | |
4170 | ||
4171 | ret = update_block_group(trans, root, ins->objectid, ins->offset, | |
4172 | 1, 0); | |
4173 | if (ret) { | |
4174 | printk(KERN_ERR "btrfs update block group failed for %llu " | |
4175 | "%llu\n", (unsigned long long)ins->objectid, | |
4176 | (unsigned long long)ins->offset); | |
4177 | BUG(); | |
4178 | } | |
4179 | return ret; | |
4180 | } | |
4181 | ||
4182 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
4183 | struct btrfs_root *root, | |
4184 | u64 root_objectid, u64 owner, | |
4185 | u64 offset, struct btrfs_key *ins) | |
4186 | { | |
4187 | int ret; | |
4188 | ||
4189 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
4190 | ||
4191 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset, | |
4192 | 0, root_objectid, owner, offset, | |
4193 | BTRFS_ADD_DELAYED_EXTENT, NULL); | |
e6dcd2dc CM |
4194 | return ret; |
4195 | } | |
e02119d5 CM |
4196 | |
4197 | /* | |
4198 | * this is used by the tree logging recovery code. It records that | |
4199 | * an extent has been allocated and makes sure to clear the free | |
4200 | * space cache bits as well | |
4201 | */ | |
5d4f98a2 YZ |
4202 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
4203 | struct btrfs_root *root, | |
4204 | u64 root_objectid, u64 owner, u64 offset, | |
4205 | struct btrfs_key *ins) | |
e02119d5 CM |
4206 | { |
4207 | int ret; | |
4208 | struct btrfs_block_group_cache *block_group; | |
4209 | ||
e02119d5 | 4210 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
817d52f8 JB |
4211 | cache_block_group(block_group); |
4212 | wait_event(block_group->caching_q, | |
4213 | block_group_cache_done(block_group)); | |
e02119d5 | 4214 | |
ea6a478e JB |
4215 | ret = btrfs_remove_free_space(block_group, ins->objectid, |
4216 | ins->offset); | |
0f9dd46c | 4217 | BUG_ON(ret); |
fa9c0d79 | 4218 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
4219 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
4220 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
4221 | return ret; |
4222 | } | |
4223 | ||
e6dcd2dc CM |
4224 | /* |
4225 | * finds a free extent and does all the dirty work required for allocation | |
4226 | * returns the key for the extent through ins, and a tree buffer for | |
4227 | * the first block of the extent through buf. | |
4228 | * | |
4229 | * returns 0 if everything worked, non-zero otherwise. | |
4230 | */ | |
5d4f98a2 YZ |
4231 | static int alloc_tree_block(struct btrfs_trans_handle *trans, |
4232 | struct btrfs_root *root, | |
4233 | u64 num_bytes, u64 parent, u64 root_objectid, | |
4234 | struct btrfs_disk_key *key, int level, | |
4235 | u64 empty_size, u64 hint_byte, u64 search_end, | |
4236 | struct btrfs_key *ins) | |
e6dcd2dc CM |
4237 | { |
4238 | int ret; | |
5d4f98a2 YZ |
4239 | u64 flags = 0; |
4240 | ||
4241 | ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes, | |
4242 | empty_size, hint_byte, search_end, | |
4243 | ins, 0); | |
817d52f8 JB |
4244 | if (ret) |
4245 | return ret; | |
5d4f98a2 YZ |
4246 | |
4247 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4248 | if (parent == 0) | |
4249 | parent = ins->objectid; | |
4250 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4251 | } else | |
4252 | BUG_ON(parent > 0); | |
4253 | ||
4254 | update_reserved_extents(root, ins->objectid, ins->offset, 1); | |
d00aff00 | 4255 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { |
5d4f98a2 YZ |
4256 | struct btrfs_delayed_extent_op *extent_op; |
4257 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
4258 | BUG_ON(!extent_op); | |
4259 | if (key) | |
4260 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4261 | else | |
4262 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4263 | extent_op->flags_to_set = flags; | |
4264 | extent_op->update_key = 1; | |
4265 | extent_op->update_flags = 1; | |
4266 | extent_op->is_data = 0; | |
4267 | ||
4268 | ret = btrfs_add_delayed_tree_ref(trans, ins->objectid, | |
4269 | ins->offset, parent, root_objectid, | |
4270 | level, BTRFS_ADD_DELAYED_EXTENT, | |
4271 | extent_op); | |
d00aff00 | 4272 | BUG_ON(ret); |
d00aff00 | 4273 | } |
925baedd | 4274 | return ret; |
fec577fb | 4275 | } |
65b51a00 CM |
4276 | |
4277 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, | |
4278 | struct btrfs_root *root, | |
4008c04a CM |
4279 | u64 bytenr, u32 blocksize, |
4280 | int level) | |
65b51a00 CM |
4281 | { |
4282 | struct extent_buffer *buf; | |
4283 | ||
4284 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
4285 | if (!buf) | |
4286 | return ERR_PTR(-ENOMEM); | |
4287 | btrfs_set_header_generation(buf, trans->transid); | |
4008c04a | 4288 | btrfs_set_buffer_lockdep_class(buf, level); |
65b51a00 CM |
4289 | btrfs_tree_lock(buf); |
4290 | clean_tree_block(trans, root, buf); | |
b4ce94de CM |
4291 | |
4292 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 4293 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 4294 | |
d0c803c4 CM |
4295 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
4296 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
4297 | buf->start + buf->len - 1, GFP_NOFS); | |
4298 | } else { | |
4299 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 4300 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4301 | } |
65b51a00 | 4302 | trans->blocks_used++; |
b4ce94de | 4303 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4304 | return buf; |
4305 | } | |
4306 | ||
fec577fb CM |
4307 | /* |
4308 | * helper function to allocate a block for a given tree | |
4309 | * returns the tree buffer or NULL. | |
4310 | */ | |
5f39d397 | 4311 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4312 | struct btrfs_root *root, u32 blocksize, |
4313 | u64 parent, u64 root_objectid, | |
4314 | struct btrfs_disk_key *key, int level, | |
4315 | u64 hint, u64 empty_size) | |
fec577fb | 4316 | { |
e2fa7227 | 4317 | struct btrfs_key ins; |
fec577fb | 4318 | int ret; |
5f39d397 | 4319 | struct extent_buffer *buf; |
fec577fb | 4320 | |
5d4f98a2 YZ |
4321 | ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid, |
4322 | key, level, empty_size, hint, (u64)-1, &ins); | |
fec577fb | 4323 | if (ret) { |
54aa1f4d CM |
4324 | BUG_ON(ret > 0); |
4325 | return ERR_PTR(ret); | |
fec577fb | 4326 | } |
55c69072 | 4327 | |
4008c04a CM |
4328 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
4329 | blocksize, level); | |
fec577fb CM |
4330 | return buf; |
4331 | } | |
a28ec197 | 4332 | |
2c47e605 | 4333 | #if 0 |
e02119d5 CM |
4334 | int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, |
4335 | struct btrfs_root *root, struct extent_buffer *leaf) | |
6407bf6d | 4336 | { |
5d4f98a2 YZ |
4337 | u64 disk_bytenr; |
4338 | u64 num_bytes; | |
5f39d397 | 4339 | struct btrfs_key key; |
6407bf6d | 4340 | struct btrfs_file_extent_item *fi; |
5d4f98a2 | 4341 | u32 nritems; |
6407bf6d | 4342 | int i; |
6407bf6d CM |
4343 | int ret; |
4344 | ||
5f39d397 CM |
4345 | BUG_ON(!btrfs_is_leaf(leaf)); |
4346 | nritems = btrfs_header_nritems(leaf); | |
7bb86316 | 4347 | |
6407bf6d | 4348 | for (i = 0; i < nritems; i++) { |
e34a5b4f | 4349 | cond_resched(); |
5f39d397 | 4350 | btrfs_item_key_to_cpu(leaf, &key, i); |
bd56b302 CM |
4351 | |
4352 | /* only extents have references, skip everything else */ | |
5f39d397 | 4353 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
6407bf6d | 4354 | continue; |
bd56b302 | 4355 | |
6407bf6d | 4356 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); |
bd56b302 CM |
4357 | |
4358 | /* inline extents live in the btree, they don't have refs */ | |
5f39d397 CM |
4359 | if (btrfs_file_extent_type(leaf, fi) == |
4360 | BTRFS_FILE_EXTENT_INLINE) | |
236454df | 4361 | continue; |
bd56b302 | 4362 | |
db94535d | 4363 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
bd56b302 CM |
4364 | |
4365 | /* holes don't have refs */ | |
db94535d | 4366 | if (disk_bytenr == 0) |
3a686375 | 4367 | continue; |
4a096752 | 4368 | |
5d4f98a2 YZ |
4369 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); |
4370 | ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes, | |
4371 | leaf->start, 0, key.objectid, 0); | |
31840ae1 | 4372 | BUG_ON(ret); |
6407bf6d CM |
4373 | } |
4374 | return 0; | |
4375 | } | |
4376 | ||
d397712b | 4377 | static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans, |
e02119d5 CM |
4378 | struct btrfs_root *root, |
4379 | struct btrfs_leaf_ref *ref) | |
31153d81 YZ |
4380 | { |
4381 | int i; | |
4382 | int ret; | |
bd56b302 CM |
4383 | struct btrfs_extent_info *info; |
4384 | struct refsort *sorted; | |
4385 | ||
4386 | if (ref->nritems == 0) | |
4387 | return 0; | |
31153d81 | 4388 | |
bd56b302 CM |
4389 | sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS); |
4390 | for (i = 0; i < ref->nritems; i++) { | |
4391 | sorted[i].bytenr = ref->extents[i].bytenr; | |
4392 | sorted[i].slot = i; | |
4393 | } | |
4394 | sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL); | |
4395 | ||
4396 | /* | |
4397 | * the items in the ref were sorted when the ref was inserted | |
4398 | * into the ref cache, so this is already in order | |
4399 | */ | |
31153d81 | 4400 | for (i = 0; i < ref->nritems; i++) { |
bd56b302 | 4401 | info = ref->extents + sorted[i].slot; |
56bec294 | 4402 | ret = btrfs_free_extent(trans, root, info->bytenr, |
31840ae1 ZY |
4403 | info->num_bytes, ref->bytenr, |
4404 | ref->owner, ref->generation, | |
3bb1a1bc | 4405 | info->objectid, 0); |
2dd3e67b CM |
4406 | |
4407 | atomic_inc(&root->fs_info->throttle_gen); | |
4408 | wake_up(&root->fs_info->transaction_throttle); | |
4409 | cond_resched(); | |
4410 | ||
31153d81 YZ |
4411 | BUG_ON(ret); |
4412 | info++; | |
4413 | } | |
31153d81 | 4414 | |
806638bc | 4415 | kfree(sorted); |
31153d81 YZ |
4416 | return 0; |
4417 | } | |
4418 | ||
5d4f98a2 | 4419 | |
56bec294 CM |
4420 | static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans, |
4421 | struct btrfs_root *root, u64 start, | |
d397712b | 4422 | u64 len, u32 *refs) |
333db94c | 4423 | { |
017e5369 | 4424 | int ret; |
f87f057b | 4425 | |
5d4f98a2 | 4426 | ret = btrfs_lookup_extent_refs(trans, root, start, len, refs); |
f87f057b CM |
4427 | BUG_ON(ret); |
4428 | ||
d397712b | 4429 | #if 0 /* some debugging code in case we see problems here */ |
f87f057b CM |
4430 | /* if the refs count is one, it won't get increased again. But |
4431 | * if the ref count is > 1, someone may be decreasing it at | |
4432 | * the same time we are. | |
4433 | */ | |
4434 | if (*refs != 1) { | |
4435 | struct extent_buffer *eb = NULL; | |
4436 | eb = btrfs_find_create_tree_block(root, start, len); | |
4437 | if (eb) | |
4438 | btrfs_tree_lock(eb); | |
4439 | ||
4440 | mutex_lock(&root->fs_info->alloc_mutex); | |
4441 | ret = lookup_extent_ref(NULL, root, start, len, refs); | |
4442 | BUG_ON(ret); | |
4443 | mutex_unlock(&root->fs_info->alloc_mutex); | |
4444 | ||
4445 | if (eb) { | |
4446 | btrfs_tree_unlock(eb); | |
4447 | free_extent_buffer(eb); | |
4448 | } | |
4449 | if (*refs == 1) { | |
d397712b CM |
4450 | printk(KERN_ERR "btrfs block %llu went down to one " |
4451 | "during drop_snap\n", (unsigned long long)start); | |
f87f057b CM |
4452 | } |
4453 | ||
4454 | } | |
4455 | #endif | |
4456 | ||
e7a84565 | 4457 | cond_resched(); |
017e5369 | 4458 | return ret; |
333db94c CM |
4459 | } |
4460 | ||
5d4f98a2 | 4461 | |
bd56b302 CM |
4462 | /* |
4463 | * this is used while deleting old snapshots, and it drops the refs | |
4464 | * on a whole subtree starting from a level 1 node. | |
4465 | * | |
4466 | * The idea is to sort all the leaf pointers, and then drop the | |
4467 | * ref on all the leaves in order. Most of the time the leaves | |
4468 | * will have ref cache entries, so no leaf IOs will be required to | |
4469 | * find the extents they have references on. | |
4470 | * | |
4471 | * For each leaf, any references it has are also dropped in order | |
4472 | * | |
4473 | * This ends up dropping the references in something close to optimal | |
4474 | * order for reading and modifying the extent allocation tree. | |
4475 | */ | |
4476 | static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans, | |
4477 | struct btrfs_root *root, | |
4478 | struct btrfs_path *path) | |
4479 | { | |
4480 | u64 bytenr; | |
4481 | u64 root_owner; | |
4482 | u64 root_gen; | |
4483 | struct extent_buffer *eb = path->nodes[1]; | |
4484 | struct extent_buffer *leaf; | |
4485 | struct btrfs_leaf_ref *ref; | |
4486 | struct refsort *sorted = NULL; | |
4487 | int nritems = btrfs_header_nritems(eb); | |
4488 | int ret; | |
4489 | int i; | |
4490 | int refi = 0; | |
4491 | int slot = path->slots[1]; | |
4492 | u32 blocksize = btrfs_level_size(root, 0); | |
4493 | u32 refs; | |
4494 | ||
4495 | if (nritems == 0) | |
4496 | goto out; | |
4497 | ||
4498 | root_owner = btrfs_header_owner(eb); | |
4499 | root_gen = btrfs_header_generation(eb); | |
4500 | sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS); | |
4501 | ||
4502 | /* | |
4503 | * step one, sort all the leaf pointers so we don't scribble | |
4504 | * randomly into the extent allocation tree | |
4505 | */ | |
4506 | for (i = slot; i < nritems; i++) { | |
4507 | sorted[refi].bytenr = btrfs_node_blockptr(eb, i); | |
4508 | sorted[refi].slot = i; | |
4509 | refi++; | |
4510 | } | |
4511 | ||
4512 | /* | |
4513 | * nritems won't be zero, but if we're picking up drop_snapshot | |
4514 | * after a crash, slot might be > 0, so double check things | |
4515 | * just in case. | |
4516 | */ | |
4517 | if (refi == 0) | |
4518 | goto out; | |
4519 | ||
4520 | sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL); | |
4521 | ||
4522 | /* | |
4523 | * the first loop frees everything the leaves point to | |
4524 | */ | |
4525 | for (i = 0; i < refi; i++) { | |
4526 | u64 ptr_gen; | |
4527 | ||
4528 | bytenr = sorted[i].bytenr; | |
4529 | ||
4530 | /* | |
4531 | * check the reference count on this leaf. If it is > 1 | |
4532 | * we just decrement it below and don't update any | |
4533 | * of the refs the leaf points to. | |
4534 | */ | |
56bec294 CM |
4535 | ret = drop_snap_lookup_refcount(trans, root, bytenr, |
4536 | blocksize, &refs); | |
bd56b302 CM |
4537 | BUG_ON(ret); |
4538 | if (refs != 1) | |
4539 | continue; | |
4540 | ||
4541 | ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot); | |
4542 | ||
4543 | /* | |
4544 | * the leaf only had one reference, which means the | |
4545 | * only thing pointing to this leaf is the snapshot | |
4546 | * we're deleting. It isn't possible for the reference | |
4547 | * count to increase again later | |
4548 | * | |
4549 | * The reference cache is checked for the leaf, | |
4550 | * and if found we'll be able to drop any refs held by | |
4551 | * the leaf without needing to read it in. | |
4552 | */ | |
4553 | ref = btrfs_lookup_leaf_ref(root, bytenr); | |
4554 | if (ref && ref->generation != ptr_gen) { | |
4555 | btrfs_free_leaf_ref(root, ref); | |
4556 | ref = NULL; | |
4557 | } | |
4558 | if (ref) { | |
4559 | ret = cache_drop_leaf_ref(trans, root, ref); | |
4560 | BUG_ON(ret); | |
4561 | btrfs_remove_leaf_ref(root, ref); | |
4562 | btrfs_free_leaf_ref(root, ref); | |
4563 | } else { | |
4564 | /* | |
4565 | * the leaf wasn't in the reference cache, so | |
4566 | * we have to read it. | |
4567 | */ | |
4568 | leaf = read_tree_block(root, bytenr, blocksize, | |
4569 | ptr_gen); | |
4570 | ret = btrfs_drop_leaf_ref(trans, root, leaf); | |
4571 | BUG_ON(ret); | |
4572 | free_extent_buffer(leaf); | |
4573 | } | |
4574 | atomic_inc(&root->fs_info->throttle_gen); | |
4575 | wake_up(&root->fs_info->transaction_throttle); | |
4576 | cond_resched(); | |
4577 | } | |
4578 | ||
4579 | /* | |
4580 | * run through the loop again to free the refs on the leaves. | |
4581 | * This is faster than doing it in the loop above because | |
4582 | * the leaves are likely to be clustered together. We end up | |
4583 | * working in nice chunks on the extent allocation tree. | |
4584 | */ | |
4585 | for (i = 0; i < refi; i++) { | |
4586 | bytenr = sorted[i].bytenr; | |
56bec294 | 4587 | ret = btrfs_free_extent(trans, root, bytenr, |
bd56b302 CM |
4588 | blocksize, eb->start, |
4589 | root_owner, root_gen, 0, 1); | |
4590 | BUG_ON(ret); | |
4591 | ||
4592 | atomic_inc(&root->fs_info->throttle_gen); | |
4593 | wake_up(&root->fs_info->transaction_throttle); | |
4594 | cond_resched(); | |
4595 | } | |
4596 | out: | |
4597 | kfree(sorted); | |
4598 | ||
4599 | /* | |
4600 | * update the path to show we've processed the entire level 1 | |
4601 | * node. This will get saved into the root's drop_snapshot_progress | |
4602 | * field so these drops are not repeated again if this transaction | |
4603 | * commits. | |
4604 | */ | |
4605 | path->slots[1] = nritems; | |
4606 | return 0; | |
4607 | } | |
4608 | ||
9aca1d51 CM |
4609 | /* |
4610 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
4611 | * counts as it goes. | |
4612 | */ | |
d397712b | 4613 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, |
98ed5174 CM |
4614 | struct btrfs_root *root, |
4615 | struct btrfs_path *path, int *level) | |
20524f02 | 4616 | { |
7bb86316 CM |
4617 | u64 root_owner; |
4618 | u64 root_gen; | |
4619 | u64 bytenr; | |
ca7a79ad | 4620 | u64 ptr_gen; |
5f39d397 CM |
4621 | struct extent_buffer *next; |
4622 | struct extent_buffer *cur; | |
7bb86316 | 4623 | struct extent_buffer *parent; |
db94535d | 4624 | u32 blocksize; |
20524f02 CM |
4625 | int ret; |
4626 | u32 refs; | |
4627 | ||
5caf2a00 CM |
4628 | WARN_ON(*level < 0); |
4629 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
56bec294 | 4630 | ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start, |
db94535d | 4631 | path->nodes[*level]->len, &refs); |
20524f02 CM |
4632 | BUG_ON(ret); |
4633 | if (refs > 1) | |
4634 | goto out; | |
e011599b | 4635 | |
9aca1d51 CM |
4636 | /* |
4637 | * walk down to the last node level and free all the leaves | |
4638 | */ | |
d397712b | 4639 | while (*level >= 0) { |
5caf2a00 CM |
4640 | WARN_ON(*level < 0); |
4641 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
20524f02 | 4642 | cur = path->nodes[*level]; |
e011599b | 4643 | |
5f39d397 | 4644 | if (btrfs_header_level(cur) != *level) |
2c90e5d6 | 4645 | WARN_ON(1); |
e011599b | 4646 | |
7518a238 | 4647 | if (path->slots[*level] >= |
5f39d397 | 4648 | btrfs_header_nritems(cur)) |
20524f02 | 4649 | break; |
bd56b302 CM |
4650 | |
4651 | /* the new code goes down to level 1 and does all the | |
4652 | * leaves pointed to that node in bulk. So, this check | |
4653 | * for level 0 will always be false. | |
4654 | * | |
4655 | * But, the disk format allows the drop_snapshot_progress | |
4656 | * field in the root to leave things in a state where | |
4657 | * a leaf will need cleaning up here. If someone crashes | |
4658 | * with the old code and then boots with the new code, | |
4659 | * we might find a leaf here. | |
4660 | */ | |
6407bf6d | 4661 | if (*level == 0) { |
e02119d5 | 4662 | ret = btrfs_drop_leaf_ref(trans, root, cur); |
6407bf6d CM |
4663 | BUG_ON(ret); |
4664 | break; | |
4665 | } | |
bd56b302 CM |
4666 | |
4667 | /* | |
4668 | * once we get to level one, process the whole node | |
4669 | * at once, including everything below it. | |
4670 | */ | |
4671 | if (*level == 1) { | |
4672 | ret = drop_level_one_refs(trans, root, path); | |
4673 | BUG_ON(ret); | |
4674 | break; | |
4675 | } | |
4676 | ||
db94535d | 4677 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
ca7a79ad | 4678 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); |
db94535d | 4679 | blocksize = btrfs_level_size(root, *level - 1); |
925baedd | 4680 | |
56bec294 CM |
4681 | ret = drop_snap_lookup_refcount(trans, root, bytenr, |
4682 | blocksize, &refs); | |
6407bf6d | 4683 | BUG_ON(ret); |
bd56b302 CM |
4684 | |
4685 | /* | |
4686 | * if there is more than one reference, we don't need | |
4687 | * to read that node to drop any references it has. We | |
4688 | * just drop the ref we hold on that node and move on to the | |
4689 | * next slot in this level. | |
4690 | */ | |
6407bf6d | 4691 | if (refs != 1) { |
7bb86316 CM |
4692 | parent = path->nodes[*level]; |
4693 | root_owner = btrfs_header_owner(parent); | |
4694 | root_gen = btrfs_header_generation(parent); | |
20524f02 | 4695 | path->slots[*level]++; |
f87f057b | 4696 | |
56bec294 | 4697 | ret = btrfs_free_extent(trans, root, bytenr, |
31840ae1 | 4698 | blocksize, parent->start, |
3bb1a1bc YZ |
4699 | root_owner, root_gen, |
4700 | *level - 1, 1); | |
20524f02 | 4701 | BUG_ON(ret); |
18e35e0a CM |
4702 | |
4703 | atomic_inc(&root->fs_info->throttle_gen); | |
4704 | wake_up(&root->fs_info->transaction_throttle); | |
2dd3e67b | 4705 | cond_resched(); |
18e35e0a | 4706 | |
20524f02 CM |
4707 | continue; |
4708 | } | |
bd56b302 | 4709 | |
f87f057b | 4710 | /* |
bd56b302 CM |
4711 | * we need to keep freeing things in the next level down. |
4712 | * read the block and loop around to process it | |
f87f057b | 4713 | */ |
bd56b302 | 4714 | next = read_tree_block(root, bytenr, blocksize, ptr_gen); |
5caf2a00 | 4715 | WARN_ON(*level <= 0); |
83e15a28 | 4716 | if (path->nodes[*level-1]) |
5f39d397 | 4717 | free_extent_buffer(path->nodes[*level-1]); |
20524f02 | 4718 | path->nodes[*level-1] = next; |
5f39d397 | 4719 | *level = btrfs_header_level(next); |
20524f02 | 4720 | path->slots[*level] = 0; |
2dd3e67b | 4721 | cond_resched(); |
20524f02 CM |
4722 | } |
4723 | out: | |
5caf2a00 CM |
4724 | WARN_ON(*level < 0); |
4725 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
7bb86316 CM |
4726 | |
4727 | if (path->nodes[*level] == root->node) { | |
7bb86316 | 4728 | parent = path->nodes[*level]; |
31153d81 | 4729 | bytenr = path->nodes[*level]->start; |
7bb86316 CM |
4730 | } else { |
4731 | parent = path->nodes[*level + 1]; | |
31153d81 | 4732 | bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]); |
7bb86316 CM |
4733 | } |
4734 | ||
31153d81 YZ |
4735 | blocksize = btrfs_level_size(root, *level); |
4736 | root_owner = btrfs_header_owner(parent); | |
7bb86316 | 4737 | root_gen = btrfs_header_generation(parent); |
31153d81 | 4738 | |
bd56b302 CM |
4739 | /* |
4740 | * cleanup and free the reference on the last node | |
4741 | * we processed | |
4742 | */ | |
56bec294 | 4743 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
31840ae1 | 4744 | parent->start, root_owner, root_gen, |
3bb1a1bc | 4745 | *level, 1); |
5f39d397 | 4746 | free_extent_buffer(path->nodes[*level]); |
20524f02 | 4747 | path->nodes[*level] = NULL; |
bd56b302 | 4748 | |
20524f02 CM |
4749 | *level += 1; |
4750 | BUG_ON(ret); | |
f87f057b | 4751 | |
e7a84565 | 4752 | cond_resched(); |
20524f02 CM |
4753 | return 0; |
4754 | } | |
5d4f98a2 | 4755 | #endif |
20524f02 | 4756 | |
2c47e605 YZ |
4757 | struct walk_control { |
4758 | u64 refs[BTRFS_MAX_LEVEL]; | |
4759 | u64 flags[BTRFS_MAX_LEVEL]; | |
4760 | struct btrfs_key update_progress; | |
4761 | int stage; | |
4762 | int level; | |
4763 | int shared_level; | |
4764 | int update_ref; | |
4765 | int keep_locks; | |
4766 | }; | |
4767 | ||
4768 | #define DROP_REFERENCE 1 | |
4769 | #define UPDATE_BACKREF 2 | |
4770 | ||
f82d02d9 | 4771 | /* |
2c47e605 YZ |
4772 | * hepler to process tree block while walking down the tree. |
4773 | * | |
4774 | * when wc->stage == DROP_REFERENCE, this function checks | |
4775 | * reference count of the block. if the block is shared and | |
4776 | * we need update back refs for the subtree rooted at the | |
4777 | * block, this function changes wc->stage to UPDATE_BACKREF | |
4778 | * | |
4779 | * when wc->stage == UPDATE_BACKREF, this function updates | |
4780 | * back refs for pointers in the block. | |
4781 | * | |
4782 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 4783 | */ |
2c47e605 | 4784 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 4785 | struct btrfs_root *root, |
2c47e605 YZ |
4786 | struct btrfs_path *path, |
4787 | struct walk_control *wc) | |
f82d02d9 | 4788 | { |
2c47e605 YZ |
4789 | int level = wc->level; |
4790 | struct extent_buffer *eb = path->nodes[level]; | |
4791 | struct btrfs_key key; | |
4792 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
f82d02d9 YZ |
4793 | int ret; |
4794 | ||
2c47e605 YZ |
4795 | if (wc->stage == UPDATE_BACKREF && |
4796 | btrfs_header_owner(eb) != root->root_key.objectid) | |
4797 | return 1; | |
f82d02d9 | 4798 | |
2c47e605 YZ |
4799 | /* |
4800 | * when reference count of tree block is 1, it won't increase | |
4801 | * again. once full backref flag is set, we never clear it. | |
4802 | */ | |
4803 | if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
4804 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) { | |
4805 | BUG_ON(!path->locks[level]); | |
4806 | ret = btrfs_lookup_extent_info(trans, root, | |
4807 | eb->start, eb->len, | |
4808 | &wc->refs[level], | |
4809 | &wc->flags[level]); | |
4810 | BUG_ON(ret); | |
4811 | BUG_ON(wc->refs[level] == 0); | |
4812 | } | |
5d4f98a2 | 4813 | |
2c47e605 YZ |
4814 | if (wc->stage == DROP_REFERENCE && |
4815 | wc->update_ref && wc->refs[level] > 1) { | |
4816 | BUG_ON(eb == root->node); | |
4817 | BUG_ON(path->slots[level] > 0); | |
4818 | if (level == 0) | |
4819 | btrfs_item_key_to_cpu(eb, &key, path->slots[level]); | |
4820 | else | |
4821 | btrfs_node_key_to_cpu(eb, &key, path->slots[level]); | |
4822 | if (btrfs_header_owner(eb) == root->root_key.objectid && | |
4823 | btrfs_comp_cpu_keys(&key, &wc->update_progress) >= 0) { | |
4824 | wc->stage = UPDATE_BACKREF; | |
4825 | wc->shared_level = level; | |
f82d02d9 | 4826 | } |
2c47e605 | 4827 | } |
f82d02d9 | 4828 | |
2c47e605 YZ |
4829 | if (wc->stage == DROP_REFERENCE) { |
4830 | if (wc->refs[level] > 1) | |
4831 | return 1; | |
f82d02d9 | 4832 | |
2c47e605 YZ |
4833 | if (path->locks[level] && !wc->keep_locks) { |
4834 | btrfs_tree_unlock(eb); | |
4835 | path->locks[level] = 0; | |
4836 | } | |
4837 | return 0; | |
4838 | } | |
f82d02d9 | 4839 | |
2c47e605 YZ |
4840 | /* wc->stage == UPDATE_BACKREF */ |
4841 | if (!(wc->flags[level] & flag)) { | |
4842 | BUG_ON(!path->locks[level]); | |
4843 | ret = btrfs_inc_ref(trans, root, eb, 1); | |
f82d02d9 | 4844 | BUG_ON(ret); |
2c47e605 YZ |
4845 | ret = btrfs_dec_ref(trans, root, eb, 0); |
4846 | BUG_ON(ret); | |
4847 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, | |
4848 | eb->len, flag, 0); | |
4849 | BUG_ON(ret); | |
4850 | wc->flags[level] |= flag; | |
4851 | } | |
4852 | ||
4853 | /* | |
4854 | * the block is shared by multiple trees, so it's not good to | |
4855 | * keep the tree lock | |
4856 | */ | |
4857 | if (path->locks[level] && level > 0) { | |
4858 | btrfs_tree_unlock(eb); | |
4859 | path->locks[level] = 0; | |
4860 | } | |
4861 | return 0; | |
4862 | } | |
4863 | ||
4864 | /* | |
4865 | * hepler to process tree block while walking up the tree. | |
4866 | * | |
4867 | * when wc->stage == DROP_REFERENCE, this function drops | |
4868 | * reference count on the block. | |
4869 | * | |
4870 | * when wc->stage == UPDATE_BACKREF, this function changes | |
4871 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
4872 | * to UPDATE_BACKREF previously while processing the block. | |
4873 | * | |
4874 | * NOTE: return value 1 means we should stop walking up. | |
4875 | */ | |
4876 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
4877 | struct btrfs_root *root, | |
4878 | struct btrfs_path *path, | |
4879 | struct walk_control *wc) | |
4880 | { | |
4881 | int ret = 0; | |
4882 | int level = wc->level; | |
4883 | struct extent_buffer *eb = path->nodes[level]; | |
4884 | u64 parent = 0; | |
4885 | ||
4886 | if (wc->stage == UPDATE_BACKREF) { | |
4887 | BUG_ON(wc->shared_level < level); | |
4888 | if (level < wc->shared_level) | |
4889 | goto out; | |
4890 | ||
4891 | BUG_ON(wc->refs[level] <= 1); | |
4892 | ret = find_next_key(path, level + 1, &wc->update_progress); | |
4893 | if (ret > 0) | |
4894 | wc->update_ref = 0; | |
4895 | ||
4896 | wc->stage = DROP_REFERENCE; | |
4897 | wc->shared_level = -1; | |
4898 | path->slots[level] = 0; | |
4899 | ||
4900 | /* | |
4901 | * check reference count again if the block isn't locked. | |
4902 | * we should start walking down the tree again if reference | |
4903 | * count is one. | |
4904 | */ | |
4905 | if (!path->locks[level]) { | |
4906 | BUG_ON(level == 0); | |
4907 | btrfs_tree_lock(eb); | |
4908 | btrfs_set_lock_blocking(eb); | |
4909 | path->locks[level] = 1; | |
4910 | ||
4911 | ret = btrfs_lookup_extent_info(trans, root, | |
4912 | eb->start, eb->len, | |
4913 | &wc->refs[level], | |
4914 | &wc->flags[level]); | |
f82d02d9 | 4915 | BUG_ON(ret); |
2c47e605 YZ |
4916 | BUG_ON(wc->refs[level] == 0); |
4917 | if (wc->refs[level] == 1) { | |
4918 | btrfs_tree_unlock(eb); | |
4919 | path->locks[level] = 0; | |
4920 | return 1; | |
4921 | } | |
4922 | } else { | |
4923 | BUG_ON(level != 0); | |
f82d02d9 | 4924 | } |
2c47e605 | 4925 | } |
f82d02d9 | 4926 | |
2c47e605 YZ |
4927 | /* wc->stage == DROP_REFERENCE */ |
4928 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 4929 | |
2c47e605 YZ |
4930 | if (wc->refs[level] == 1) { |
4931 | if (level == 0) { | |
4932 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
4933 | ret = btrfs_dec_ref(trans, root, eb, 1); | |
4934 | else | |
4935 | ret = btrfs_dec_ref(trans, root, eb, 0); | |
4936 | BUG_ON(ret); | |
4937 | } | |
4938 | /* make block locked assertion in clean_tree_block happy */ | |
4939 | if (!path->locks[level] && | |
4940 | btrfs_header_generation(eb) == trans->transid) { | |
4941 | btrfs_tree_lock(eb); | |
4942 | btrfs_set_lock_blocking(eb); | |
4943 | path->locks[level] = 1; | |
4944 | } | |
4945 | clean_tree_block(trans, root, eb); | |
4946 | } | |
4947 | ||
4948 | if (eb == root->node) { | |
4949 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
4950 | parent = eb->start; | |
4951 | else | |
4952 | BUG_ON(root->root_key.objectid != | |
4953 | btrfs_header_owner(eb)); | |
4954 | } else { | |
4955 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
4956 | parent = path->nodes[level + 1]->start; | |
4957 | else | |
4958 | BUG_ON(root->root_key.objectid != | |
4959 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 4960 | } |
f82d02d9 | 4961 | |
2c47e605 YZ |
4962 | ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent, |
4963 | root->root_key.objectid, level, 0); | |
f82d02d9 | 4964 | BUG_ON(ret); |
2c47e605 YZ |
4965 | out: |
4966 | wc->refs[level] = 0; | |
4967 | wc->flags[level] = 0; | |
4968 | return ret; | |
4969 | } | |
4970 | ||
4971 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
4972 | struct btrfs_root *root, | |
4973 | struct btrfs_path *path, | |
4974 | struct walk_control *wc) | |
4975 | { | |
4976 | struct extent_buffer *next; | |
4977 | struct extent_buffer *cur; | |
4978 | u64 bytenr; | |
4979 | u64 ptr_gen; | |
4980 | u32 blocksize; | |
4981 | int level = wc->level; | |
4982 | int ret; | |
4983 | ||
4984 | while (level >= 0) { | |
4985 | cur = path->nodes[level]; | |
4986 | BUG_ON(path->slots[level] >= btrfs_header_nritems(cur)); | |
f82d02d9 | 4987 | |
2c47e605 YZ |
4988 | ret = walk_down_proc(trans, root, path, wc); |
4989 | if (ret > 0) | |
4990 | break; | |
4991 | ||
4992 | if (level == 0) | |
4993 | break; | |
4994 | ||
4995 | bytenr = btrfs_node_blockptr(cur, path->slots[level]); | |
4996 | blocksize = btrfs_level_size(root, level - 1); | |
4997 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[level]); | |
4998 | ||
4999 | next = read_tree_block(root, bytenr, blocksize, ptr_gen); | |
5000 | btrfs_tree_lock(next); | |
5001 | btrfs_set_lock_blocking(next); | |
5002 | ||
5003 | level--; | |
5004 | BUG_ON(level != btrfs_header_level(next)); | |
5005 | path->nodes[level] = next; | |
5006 | path->slots[level] = 0; | |
5007 | path->locks[level] = 1; | |
5008 | wc->level = level; | |
f82d02d9 | 5009 | } |
f82d02d9 YZ |
5010 | return 0; |
5011 | } | |
5012 | ||
d397712b | 5013 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5014 | struct btrfs_root *root, |
f82d02d9 | 5015 | struct btrfs_path *path, |
2c47e605 | 5016 | struct walk_control *wc, int max_level) |
20524f02 | 5017 | { |
2c47e605 | 5018 | int level = wc->level; |
20524f02 | 5019 | int ret; |
9f3a7427 | 5020 | |
2c47e605 YZ |
5021 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5022 | while (level < max_level && path->nodes[level]) { | |
5023 | wc->level = level; | |
5024 | if (path->slots[level] + 1 < | |
5025 | btrfs_header_nritems(path->nodes[level])) { | |
5026 | path->slots[level]++; | |
20524f02 CM |
5027 | return 0; |
5028 | } else { | |
2c47e605 YZ |
5029 | ret = walk_up_proc(trans, root, path, wc); |
5030 | if (ret > 0) | |
5031 | return 0; | |
bd56b302 | 5032 | |
2c47e605 YZ |
5033 | if (path->locks[level]) { |
5034 | btrfs_tree_unlock(path->nodes[level]); | |
5035 | path->locks[level] = 0; | |
f82d02d9 | 5036 | } |
2c47e605 YZ |
5037 | free_extent_buffer(path->nodes[level]); |
5038 | path->nodes[level] = NULL; | |
5039 | level++; | |
20524f02 CM |
5040 | } |
5041 | } | |
5042 | return 1; | |
5043 | } | |
5044 | ||
9aca1d51 | 5045 | /* |
2c47e605 YZ |
5046 | * drop a subvolume tree. |
5047 | * | |
5048 | * this function traverses the tree freeing any blocks that only | |
5049 | * referenced by the tree. | |
5050 | * | |
5051 | * when a shared tree block is found. this function decreases its | |
5052 | * reference count by one. if update_ref is true, this function | |
5053 | * also make sure backrefs for the shared block and all lower level | |
5054 | * blocks are properly updated. | |
9aca1d51 | 5055 | */ |
2c47e605 | 5056 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref) |
20524f02 | 5057 | { |
5caf2a00 | 5058 | struct btrfs_path *path; |
2c47e605 YZ |
5059 | struct btrfs_trans_handle *trans; |
5060 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 5061 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5062 | struct walk_control *wc; |
5063 | struct btrfs_key key; | |
5064 | int err = 0; | |
5065 | int ret; | |
5066 | int level; | |
20524f02 | 5067 | |
5caf2a00 CM |
5068 | path = btrfs_alloc_path(); |
5069 | BUG_ON(!path); | |
20524f02 | 5070 | |
2c47e605 YZ |
5071 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
5072 | BUG_ON(!wc); | |
5073 | ||
5074 | trans = btrfs_start_transaction(tree_root, 1); | |
5075 | ||
9f3a7427 | 5076 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5077 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
5078 | path->nodes[level] = btrfs_lock_root_node(root); |
5079 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 5080 | path->slots[level] = 0; |
5d4f98a2 | 5081 | path->locks[level] = 1; |
2c47e605 YZ |
5082 | memset(&wc->update_progress, 0, |
5083 | sizeof(wc->update_progress)); | |
9f3a7427 | 5084 | } else { |
9f3a7427 | 5085 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5086 | memcpy(&wc->update_progress, &key, |
5087 | sizeof(wc->update_progress)); | |
5088 | ||
6702ed49 | 5089 | level = root_item->drop_level; |
2c47e605 | 5090 | BUG_ON(level == 0); |
6702ed49 | 5091 | path->lowest_level = level; |
2c47e605 YZ |
5092 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5093 | path->lowest_level = 0; | |
5094 | if (ret < 0) { | |
5095 | err = ret; | |
9f3a7427 CM |
5096 | goto out; |
5097 | } | |
2c47e605 YZ |
5098 | btrfs_node_key_to_cpu(path->nodes[level], &key, |
5099 | path->slots[level]); | |
5100 | WARN_ON(memcmp(&key, &wc->update_progress, sizeof(key))); | |
5101 | ||
7d9eb12c CM |
5102 | /* |
5103 | * unlock our path, this is safe because only this | |
5104 | * function is allowed to delete this snapshot | |
5105 | */ | |
5d4f98a2 | 5106 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5107 | |
5108 | level = btrfs_header_level(root->node); | |
5109 | while (1) { | |
5110 | btrfs_tree_lock(path->nodes[level]); | |
5111 | btrfs_set_lock_blocking(path->nodes[level]); | |
5112 | ||
5113 | ret = btrfs_lookup_extent_info(trans, root, | |
5114 | path->nodes[level]->start, | |
5115 | path->nodes[level]->len, | |
5116 | &wc->refs[level], | |
5117 | &wc->flags[level]); | |
5118 | BUG_ON(ret); | |
5119 | BUG_ON(wc->refs[level] == 0); | |
5120 | ||
5121 | if (level == root_item->drop_level) | |
5122 | break; | |
5123 | ||
5124 | btrfs_tree_unlock(path->nodes[level]); | |
5125 | WARN_ON(wc->refs[level] != 1); | |
5126 | level--; | |
5127 | } | |
9f3a7427 | 5128 | } |
2c47e605 YZ |
5129 | |
5130 | wc->level = level; | |
5131 | wc->shared_level = -1; | |
5132 | wc->stage = DROP_REFERENCE; | |
5133 | wc->update_ref = update_ref; | |
5134 | wc->keep_locks = 0; | |
5135 | ||
d397712b | 5136 | while (1) { |
2c47e605 YZ |
5137 | ret = walk_down_tree(trans, root, path, wc); |
5138 | if (ret < 0) { | |
5139 | err = ret; | |
20524f02 | 5140 | break; |
2c47e605 | 5141 | } |
9aca1d51 | 5142 | |
2c47e605 YZ |
5143 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5144 | if (ret < 0) { | |
5145 | err = ret; | |
20524f02 | 5146 | break; |
2c47e605 YZ |
5147 | } |
5148 | ||
5149 | if (ret > 0) { | |
5150 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5151 | break; |
5152 | } | |
2c47e605 YZ |
5153 | |
5154 | if (wc->stage == DROP_REFERENCE) { | |
5155 | level = wc->level; | |
5156 | btrfs_node_key(path->nodes[level], | |
5157 | &root_item->drop_progress, | |
5158 | path->slots[level]); | |
5159 | root_item->drop_level = level; | |
5160 | } | |
5161 | ||
5162 | BUG_ON(wc->level == 0); | |
5163 | if (trans->transaction->in_commit || | |
5164 | trans->transaction->delayed_refs.flushing) { | |
5165 | ret = btrfs_update_root(trans, tree_root, | |
5166 | &root->root_key, | |
5167 | root_item); | |
5168 | BUG_ON(ret); | |
5169 | ||
5170 | btrfs_end_transaction(trans, tree_root); | |
5171 | trans = btrfs_start_transaction(tree_root, 1); | |
5172 | } else { | |
5173 | unsigned long update; | |
c3e69d58 CM |
5174 | update = trans->delayed_ref_updates; |
5175 | trans->delayed_ref_updates = 0; | |
5176 | if (update) | |
2c47e605 YZ |
5177 | btrfs_run_delayed_refs(trans, tree_root, |
5178 | update); | |
c3e69d58 | 5179 | } |
20524f02 | 5180 | } |
2c47e605 YZ |
5181 | btrfs_release_path(root, path); |
5182 | BUG_ON(err); | |
5183 | ||
5184 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
5185 | BUG_ON(ret); | |
5186 | ||
5187 | free_extent_buffer(root->node); | |
5188 | free_extent_buffer(root->commit_root); | |
5189 | kfree(root); | |
9f3a7427 | 5190 | out: |
2c47e605 YZ |
5191 | btrfs_end_transaction(trans, tree_root); |
5192 | kfree(wc); | |
5caf2a00 | 5193 | btrfs_free_path(path); |
2c47e605 | 5194 | return err; |
20524f02 | 5195 | } |
9078a3e1 | 5196 | |
2c47e605 YZ |
5197 | /* |
5198 | * drop subtree rooted at tree block 'node'. | |
5199 | * | |
5200 | * NOTE: this function will unlock and release tree block 'node' | |
5201 | */ | |
f82d02d9 YZ |
5202 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5203 | struct btrfs_root *root, | |
5204 | struct extent_buffer *node, | |
5205 | struct extent_buffer *parent) | |
5206 | { | |
5207 | struct btrfs_path *path; | |
2c47e605 | 5208 | struct walk_control *wc; |
f82d02d9 YZ |
5209 | int level; |
5210 | int parent_level; | |
5211 | int ret = 0; | |
5212 | int wret; | |
5213 | ||
2c47e605 YZ |
5214 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5215 | ||
f82d02d9 YZ |
5216 | path = btrfs_alloc_path(); |
5217 | BUG_ON(!path); | |
5218 | ||
2c47e605 YZ |
5219 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
5220 | BUG_ON(!wc); | |
5221 | ||
b9447ef8 | 5222 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
5223 | parent_level = btrfs_header_level(parent); |
5224 | extent_buffer_get(parent); | |
5225 | path->nodes[parent_level] = parent; | |
5226 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5227 | ||
b9447ef8 | 5228 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5229 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5230 | path->nodes[level] = node; |
5231 | path->slots[level] = 0; | |
2c47e605 YZ |
5232 | path->locks[level] = 1; |
5233 | ||
5234 | wc->refs[parent_level] = 1; | |
5235 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5236 | wc->level = level; | |
5237 | wc->shared_level = -1; | |
5238 | wc->stage = DROP_REFERENCE; | |
5239 | wc->update_ref = 0; | |
5240 | wc->keep_locks = 1; | |
f82d02d9 YZ |
5241 | |
5242 | while (1) { | |
2c47e605 YZ |
5243 | wret = walk_down_tree(trans, root, path, wc); |
5244 | if (wret < 0) { | |
f82d02d9 | 5245 | ret = wret; |
f82d02d9 | 5246 | break; |
2c47e605 | 5247 | } |
f82d02d9 | 5248 | |
2c47e605 | 5249 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5250 | if (wret < 0) |
5251 | ret = wret; | |
5252 | if (wret != 0) | |
5253 | break; | |
5254 | } | |
5255 | ||
2c47e605 | 5256 | kfree(wc); |
f82d02d9 YZ |
5257 | btrfs_free_path(path); |
5258 | return ret; | |
5259 | } | |
5260 | ||
5d4f98a2 | 5261 | #if 0 |
8e7bf94f CM |
5262 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
5263 | unsigned long nr) | |
5264 | { | |
5265 | return min(last, start + nr - 1); | |
5266 | } | |
5267 | ||
d397712b | 5268 | static noinline int relocate_inode_pages(struct inode *inode, u64 start, |
98ed5174 | 5269 | u64 len) |
edbd8d4e CM |
5270 | { |
5271 | u64 page_start; | |
5272 | u64 page_end; | |
1a40e23b | 5273 | unsigned long first_index; |
edbd8d4e | 5274 | unsigned long last_index; |
edbd8d4e CM |
5275 | unsigned long i; |
5276 | struct page *page; | |
d1310b2e | 5277 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 5278 | struct file_ra_state *ra; |
3eaa2885 | 5279 | struct btrfs_ordered_extent *ordered; |
1a40e23b ZY |
5280 | unsigned int total_read = 0; |
5281 | unsigned int total_dirty = 0; | |
5282 | int ret = 0; | |
4313b399 CM |
5283 | |
5284 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
5285 | |
5286 | mutex_lock(&inode->i_mutex); | |
1a40e23b | 5287 | first_index = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
5288 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
5289 | ||
1a40e23b ZY |
5290 | /* make sure the dirty trick played by the caller work */ |
5291 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
5292 | first_index, last_index); | |
5293 | if (ret) | |
5294 | goto out_unlock; | |
8e7bf94f | 5295 | |
4313b399 | 5296 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 5297 | |
1a40e23b ZY |
5298 | for (i = first_index ; i <= last_index; i++) { |
5299 | if (total_read % ra->ra_pages == 0) { | |
8e7bf94f | 5300 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, |
1a40e23b | 5301 | calc_ra(i, last_index, ra->ra_pages)); |
8e7bf94f CM |
5302 | } |
5303 | total_read++; | |
3eaa2885 CM |
5304 | again: |
5305 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
1a40e23b | 5306 | BUG_ON(1); |
edbd8d4e | 5307 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 5308 | if (!page) { |
1a40e23b | 5309 | ret = -ENOMEM; |
edbd8d4e | 5310 | goto out_unlock; |
a061fc8d | 5311 | } |
edbd8d4e CM |
5312 | if (!PageUptodate(page)) { |
5313 | btrfs_readpage(NULL, page); | |
5314 | lock_page(page); | |
5315 | if (!PageUptodate(page)) { | |
5316 | unlock_page(page); | |
5317 | page_cache_release(page); | |
1a40e23b | 5318 | ret = -EIO; |
edbd8d4e CM |
5319 | goto out_unlock; |
5320 | } | |
5321 | } | |
ec44a35c | 5322 | wait_on_page_writeback(page); |
3eaa2885 | 5323 | |
edbd8d4e CM |
5324 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
5325 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
d1310b2e | 5326 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 5327 | |
3eaa2885 CM |
5328 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
5329 | if (ordered) { | |
5330 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5331 | unlock_page(page); | |
5332 | page_cache_release(page); | |
5333 | btrfs_start_ordered_extent(inode, ordered, 1); | |
5334 | btrfs_put_ordered_extent(ordered); | |
5335 | goto again; | |
5336 | } | |
5337 | set_page_extent_mapped(page); | |
5338 | ||
1a40e23b ZY |
5339 | if (i == first_index) |
5340 | set_extent_bits(io_tree, page_start, page_end, | |
5341 | EXTENT_BOUNDARY, GFP_NOFS); | |
1f80e4db | 5342 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
1a40e23b | 5343 | |
a061fc8d | 5344 | set_page_dirty(page); |
1a40e23b | 5345 | total_dirty++; |
edbd8d4e | 5346 | |
d1310b2e | 5347 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
5348 | unlock_page(page); |
5349 | page_cache_release(page); | |
5350 | } | |
5351 | ||
5352 | out_unlock: | |
ec44a35c | 5353 | kfree(ra); |
edbd8d4e | 5354 | mutex_unlock(&inode->i_mutex); |
1a40e23b ZY |
5355 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty); |
5356 | return ret; | |
edbd8d4e CM |
5357 | } |
5358 | ||
d397712b | 5359 | static noinline int relocate_data_extent(struct inode *reloc_inode, |
1a40e23b ZY |
5360 | struct btrfs_key *extent_key, |
5361 | u64 offset) | |
5362 | { | |
5363 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
5364 | struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree; | |
5365 | struct extent_map *em; | |
6643558d YZ |
5366 | u64 start = extent_key->objectid - offset; |
5367 | u64 end = start + extent_key->offset - 1; | |
bf4ef679 | 5368 | |
1a40e23b ZY |
5369 | em = alloc_extent_map(GFP_NOFS); |
5370 | BUG_ON(!em || IS_ERR(em)); | |
bf4ef679 | 5371 | |
6643558d | 5372 | em->start = start; |
1a40e23b | 5373 | em->len = extent_key->offset; |
c8b97818 | 5374 | em->block_len = extent_key->offset; |
1a40e23b ZY |
5375 | em->block_start = extent_key->objectid; |
5376 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
5377 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
5378 | ||
5379 | /* setup extent map to cheat btrfs_readpage */ | |
6643558d | 5380 | lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
1a40e23b ZY |
5381 | while (1) { |
5382 | int ret; | |
5383 | spin_lock(&em_tree->lock); | |
5384 | ret = add_extent_mapping(em_tree, em); | |
5385 | spin_unlock(&em_tree->lock); | |
5386 | if (ret != -EEXIST) { | |
5387 | free_extent_map(em); | |
bf4ef679 CM |
5388 | break; |
5389 | } | |
6643558d | 5390 | btrfs_drop_extent_cache(reloc_inode, start, end, 0); |
bf4ef679 | 5391 | } |
6643558d | 5392 | unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
bf4ef679 | 5393 | |
6643558d | 5394 | return relocate_inode_pages(reloc_inode, start, extent_key->offset); |
1a40e23b | 5395 | } |
edbd8d4e | 5396 | |
1a40e23b ZY |
5397 | struct btrfs_ref_path { |
5398 | u64 extent_start; | |
5399 | u64 nodes[BTRFS_MAX_LEVEL]; | |
5400 | u64 root_objectid; | |
5401 | u64 root_generation; | |
5402 | u64 owner_objectid; | |
1a40e23b ZY |
5403 | u32 num_refs; |
5404 | int lowest_level; | |
5405 | int current_level; | |
f82d02d9 YZ |
5406 | int shared_level; |
5407 | ||
5408 | struct btrfs_key node_keys[BTRFS_MAX_LEVEL]; | |
5409 | u64 new_nodes[BTRFS_MAX_LEVEL]; | |
1a40e23b | 5410 | }; |
7d9eb12c | 5411 | |
1a40e23b | 5412 | struct disk_extent { |
c8b97818 | 5413 | u64 ram_bytes; |
1a40e23b ZY |
5414 | u64 disk_bytenr; |
5415 | u64 disk_num_bytes; | |
5416 | u64 offset; | |
5417 | u64 num_bytes; | |
c8b97818 CM |
5418 | u8 compression; |
5419 | u8 encryption; | |
5420 | u16 other_encoding; | |
1a40e23b | 5421 | }; |
4313b399 | 5422 | |
1a40e23b ZY |
5423 | static int is_cowonly_root(u64 root_objectid) |
5424 | { | |
5425 | if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || | |
5426 | root_objectid == BTRFS_EXTENT_TREE_OBJECTID || | |
5427 | root_objectid == BTRFS_CHUNK_TREE_OBJECTID || | |
5428 | root_objectid == BTRFS_DEV_TREE_OBJECTID || | |
0403e47e YZ |
5429 | root_objectid == BTRFS_TREE_LOG_OBJECTID || |
5430 | root_objectid == BTRFS_CSUM_TREE_OBJECTID) | |
1a40e23b ZY |
5431 | return 1; |
5432 | return 0; | |
5433 | } | |
edbd8d4e | 5434 | |
d397712b | 5435 | static noinline int __next_ref_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
5436 | struct btrfs_root *extent_root, |
5437 | struct btrfs_ref_path *ref_path, | |
5438 | int first_time) | |
5439 | { | |
5440 | struct extent_buffer *leaf; | |
5441 | struct btrfs_path *path; | |
5442 | struct btrfs_extent_ref *ref; | |
5443 | struct btrfs_key key; | |
5444 | struct btrfs_key found_key; | |
5445 | u64 bytenr; | |
5446 | u32 nritems; | |
5447 | int level; | |
5448 | int ret = 1; | |
edbd8d4e | 5449 | |
1a40e23b ZY |
5450 | path = btrfs_alloc_path(); |
5451 | if (!path) | |
5452 | return -ENOMEM; | |
bf4ef679 | 5453 | |
1a40e23b ZY |
5454 | if (first_time) { |
5455 | ref_path->lowest_level = -1; | |
5456 | ref_path->current_level = -1; | |
f82d02d9 | 5457 | ref_path->shared_level = -1; |
1a40e23b ZY |
5458 | goto walk_up; |
5459 | } | |
5460 | walk_down: | |
5461 | level = ref_path->current_level - 1; | |
5462 | while (level >= -1) { | |
5463 | u64 parent; | |
5464 | if (level < ref_path->lowest_level) | |
5465 | break; | |
bf4ef679 | 5466 | |
d397712b | 5467 | if (level >= 0) |
1a40e23b | 5468 | bytenr = ref_path->nodes[level]; |
d397712b | 5469 | else |
1a40e23b | 5470 | bytenr = ref_path->extent_start; |
1a40e23b | 5471 | BUG_ON(bytenr == 0); |
bf4ef679 | 5472 | |
1a40e23b ZY |
5473 | parent = ref_path->nodes[level + 1]; |
5474 | ref_path->nodes[level + 1] = 0; | |
5475 | ref_path->current_level = level; | |
5476 | BUG_ON(parent == 0); | |
0ef3e66b | 5477 | |
1a40e23b ZY |
5478 | key.objectid = bytenr; |
5479 | key.offset = parent + 1; | |
5480 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 5481 | |
1a40e23b ZY |
5482 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
5483 | if (ret < 0) | |
edbd8d4e | 5484 | goto out; |
1a40e23b | 5485 | BUG_ON(ret == 0); |
7d9eb12c | 5486 | |
1a40e23b ZY |
5487 | leaf = path->nodes[0]; |
5488 | nritems = btrfs_header_nritems(leaf); | |
5489 | if (path->slots[0] >= nritems) { | |
5490 | ret = btrfs_next_leaf(extent_root, path); | |
5491 | if (ret < 0) | |
5492 | goto out; | |
5493 | if (ret > 0) | |
5494 | goto next; | |
5495 | leaf = path->nodes[0]; | |
5496 | } | |
0ef3e66b | 5497 | |
1a40e23b ZY |
5498 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5499 | if (found_key.objectid == bytenr && | |
f82d02d9 YZ |
5500 | found_key.type == BTRFS_EXTENT_REF_KEY) { |
5501 | if (level < ref_path->shared_level) | |
5502 | ref_path->shared_level = level; | |
1a40e23b | 5503 | goto found; |
f82d02d9 | 5504 | } |
1a40e23b ZY |
5505 | next: |
5506 | level--; | |
5507 | btrfs_release_path(extent_root, path); | |
d899e052 | 5508 | cond_resched(); |
1a40e23b ZY |
5509 | } |
5510 | /* reached lowest level */ | |
5511 | ret = 1; | |
5512 | goto out; | |
5513 | walk_up: | |
5514 | level = ref_path->current_level; | |
5515 | while (level < BTRFS_MAX_LEVEL - 1) { | |
5516 | u64 ref_objectid; | |
d397712b CM |
5517 | |
5518 | if (level >= 0) | |
1a40e23b | 5519 | bytenr = ref_path->nodes[level]; |
d397712b | 5520 | else |
1a40e23b | 5521 | bytenr = ref_path->extent_start; |
d397712b | 5522 | |
1a40e23b | 5523 | BUG_ON(bytenr == 0); |
edbd8d4e | 5524 | |
1a40e23b ZY |
5525 | key.objectid = bytenr; |
5526 | key.offset = 0; | |
5527 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 5528 | |
1a40e23b ZY |
5529 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
5530 | if (ret < 0) | |
5531 | goto out; | |
edbd8d4e | 5532 | |
1a40e23b ZY |
5533 | leaf = path->nodes[0]; |
5534 | nritems = btrfs_header_nritems(leaf); | |
5535 | if (path->slots[0] >= nritems) { | |
5536 | ret = btrfs_next_leaf(extent_root, path); | |
5537 | if (ret < 0) | |
5538 | goto out; | |
5539 | if (ret > 0) { | |
5540 | /* the extent was freed by someone */ | |
5541 | if (ref_path->lowest_level == level) | |
5542 | goto out; | |
5543 | btrfs_release_path(extent_root, path); | |
5544 | goto walk_down; | |
5545 | } | |
5546 | leaf = path->nodes[0]; | |
5547 | } | |
edbd8d4e | 5548 | |
1a40e23b ZY |
5549 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5550 | if (found_key.objectid != bytenr || | |
5551 | found_key.type != BTRFS_EXTENT_REF_KEY) { | |
5552 | /* the extent was freed by someone */ | |
5553 | if (ref_path->lowest_level == level) { | |
5554 | ret = 1; | |
5555 | goto out; | |
5556 | } | |
5557 | btrfs_release_path(extent_root, path); | |
5558 | goto walk_down; | |
5559 | } | |
5560 | found: | |
5561 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
5562 | struct btrfs_extent_ref); | |
5563 | ref_objectid = btrfs_ref_objectid(leaf, ref); | |
5564 | if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
5565 | if (first_time) { | |
5566 | level = (int)ref_objectid; | |
5567 | BUG_ON(level >= BTRFS_MAX_LEVEL); | |
5568 | ref_path->lowest_level = level; | |
5569 | ref_path->current_level = level; | |
5570 | ref_path->nodes[level] = bytenr; | |
5571 | } else { | |
5572 | WARN_ON(ref_objectid != level); | |
5573 | } | |
5574 | } else { | |
5575 | WARN_ON(level != -1); | |
5576 | } | |
5577 | first_time = 0; | |
bf4ef679 | 5578 | |
1a40e23b ZY |
5579 | if (ref_path->lowest_level == level) { |
5580 | ref_path->owner_objectid = ref_objectid; | |
1a40e23b ZY |
5581 | ref_path->num_refs = btrfs_ref_num_refs(leaf, ref); |
5582 | } | |
bf4ef679 | 5583 | |
7d9eb12c | 5584 | /* |
1a40e23b ZY |
5585 | * the block is tree root or the block isn't in reference |
5586 | * counted tree. | |
7d9eb12c | 5587 | */ |
1a40e23b ZY |
5588 | if (found_key.objectid == found_key.offset || |
5589 | is_cowonly_root(btrfs_ref_root(leaf, ref))) { | |
5590 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
5591 | ref_path->root_generation = | |
5592 | btrfs_ref_generation(leaf, ref); | |
5593 | if (level < 0) { | |
5594 | /* special reference from the tree log */ | |
5595 | ref_path->nodes[0] = found_key.offset; | |
5596 | ref_path->current_level = 0; | |
5597 | } | |
5598 | ret = 0; | |
5599 | goto out; | |
5600 | } | |
7d9eb12c | 5601 | |
1a40e23b ZY |
5602 | level++; |
5603 | BUG_ON(ref_path->nodes[level] != 0); | |
5604 | ref_path->nodes[level] = found_key.offset; | |
5605 | ref_path->current_level = level; | |
bf4ef679 | 5606 | |
1a40e23b ZY |
5607 | /* |
5608 | * the reference was created in the running transaction, | |
5609 | * no need to continue walking up. | |
5610 | */ | |
5611 | if (btrfs_ref_generation(leaf, ref) == trans->transid) { | |
5612 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
5613 | ref_path->root_generation = | |
5614 | btrfs_ref_generation(leaf, ref); | |
5615 | ret = 0; | |
5616 | goto out; | |
7d9eb12c CM |
5617 | } |
5618 | ||
1a40e23b | 5619 | btrfs_release_path(extent_root, path); |
d899e052 | 5620 | cond_resched(); |
7d9eb12c | 5621 | } |
1a40e23b ZY |
5622 | /* reached max tree level, but no tree root found. */ |
5623 | BUG(); | |
edbd8d4e | 5624 | out: |
1a40e23b ZY |
5625 | btrfs_free_path(path); |
5626 | return ret; | |
edbd8d4e CM |
5627 | } |
5628 | ||
1a40e23b ZY |
5629 | static int btrfs_first_ref_path(struct btrfs_trans_handle *trans, |
5630 | struct btrfs_root *extent_root, | |
5631 | struct btrfs_ref_path *ref_path, | |
5632 | u64 extent_start) | |
a061fc8d | 5633 | { |
1a40e23b ZY |
5634 | memset(ref_path, 0, sizeof(*ref_path)); |
5635 | ref_path->extent_start = extent_start; | |
a061fc8d | 5636 | |
1a40e23b | 5637 | return __next_ref_path(trans, extent_root, ref_path, 1); |
a061fc8d CM |
5638 | } |
5639 | ||
1a40e23b ZY |
5640 | static int btrfs_next_ref_path(struct btrfs_trans_handle *trans, |
5641 | struct btrfs_root *extent_root, | |
5642 | struct btrfs_ref_path *ref_path) | |
edbd8d4e | 5643 | { |
1a40e23b ZY |
5644 | return __next_ref_path(trans, extent_root, ref_path, 0); |
5645 | } | |
5646 | ||
d397712b | 5647 | static noinline int get_new_locations(struct inode *reloc_inode, |
1a40e23b ZY |
5648 | struct btrfs_key *extent_key, |
5649 | u64 offset, int no_fragment, | |
5650 | struct disk_extent **extents, | |
5651 | int *nr_extents) | |
5652 | { | |
5653 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
5654 | struct btrfs_path *path; | |
5655 | struct btrfs_file_extent_item *fi; | |
edbd8d4e | 5656 | struct extent_buffer *leaf; |
1a40e23b ZY |
5657 | struct disk_extent *exts = *extents; |
5658 | struct btrfs_key found_key; | |
5659 | u64 cur_pos; | |
5660 | u64 last_byte; | |
edbd8d4e | 5661 | u32 nritems; |
1a40e23b ZY |
5662 | int nr = 0; |
5663 | int max = *nr_extents; | |
5664 | int ret; | |
edbd8d4e | 5665 | |
1a40e23b ZY |
5666 | WARN_ON(!no_fragment && *extents); |
5667 | if (!exts) { | |
5668 | max = 1; | |
5669 | exts = kmalloc(sizeof(*exts) * max, GFP_NOFS); | |
5670 | if (!exts) | |
5671 | return -ENOMEM; | |
a061fc8d | 5672 | } |
edbd8d4e | 5673 | |
1a40e23b ZY |
5674 | path = btrfs_alloc_path(); |
5675 | BUG_ON(!path); | |
edbd8d4e | 5676 | |
1a40e23b ZY |
5677 | cur_pos = extent_key->objectid - offset; |
5678 | last_byte = extent_key->objectid + extent_key->offset; | |
5679 | ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino, | |
5680 | cur_pos, 0); | |
5681 | if (ret < 0) | |
5682 | goto out; | |
5683 | if (ret > 0) { | |
5684 | ret = -ENOENT; | |
5685 | goto out; | |
5686 | } | |
edbd8d4e | 5687 | |
1a40e23b | 5688 | while (1) { |
edbd8d4e CM |
5689 | leaf = path->nodes[0]; |
5690 | nritems = btrfs_header_nritems(leaf); | |
1a40e23b ZY |
5691 | if (path->slots[0] >= nritems) { |
5692 | ret = btrfs_next_leaf(root, path); | |
a061fc8d CM |
5693 | if (ret < 0) |
5694 | goto out; | |
1a40e23b ZY |
5695 | if (ret > 0) |
5696 | break; | |
bf4ef679 | 5697 | leaf = path->nodes[0]; |
a061fc8d | 5698 | } |
edbd8d4e CM |
5699 | |
5700 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b ZY |
5701 | if (found_key.offset != cur_pos || |
5702 | found_key.type != BTRFS_EXTENT_DATA_KEY || | |
5703 | found_key.objectid != reloc_inode->i_ino) | |
edbd8d4e CM |
5704 | break; |
5705 | ||
1a40e23b ZY |
5706 | fi = btrfs_item_ptr(leaf, path->slots[0], |
5707 | struct btrfs_file_extent_item); | |
5708 | if (btrfs_file_extent_type(leaf, fi) != | |
5709 | BTRFS_FILE_EXTENT_REG || | |
5710 | btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
edbd8d4e | 5711 | break; |
1a40e23b ZY |
5712 | |
5713 | if (nr == max) { | |
5714 | struct disk_extent *old = exts; | |
5715 | max *= 2; | |
5716 | exts = kzalloc(sizeof(*exts) * max, GFP_NOFS); | |
5717 | memcpy(exts, old, sizeof(*exts) * nr); | |
5718 | if (old != *extents) | |
5719 | kfree(old); | |
a061fc8d | 5720 | } |
edbd8d4e | 5721 | |
1a40e23b ZY |
5722 | exts[nr].disk_bytenr = |
5723 | btrfs_file_extent_disk_bytenr(leaf, fi); | |
5724 | exts[nr].disk_num_bytes = | |
5725 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
5726 | exts[nr].offset = btrfs_file_extent_offset(leaf, fi); | |
5727 | exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
c8b97818 CM |
5728 | exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
5729 | exts[nr].compression = btrfs_file_extent_compression(leaf, fi); | |
5730 | exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi); | |
5731 | exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf, | |
5732 | fi); | |
d899e052 YZ |
5733 | BUG_ON(exts[nr].offset > 0); |
5734 | BUG_ON(exts[nr].compression || exts[nr].encryption); | |
5735 | BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes); | |
edbd8d4e | 5736 | |
1a40e23b ZY |
5737 | cur_pos += exts[nr].num_bytes; |
5738 | nr++; | |
5739 | ||
5740 | if (cur_pos + offset >= last_byte) | |
5741 | break; | |
5742 | ||
5743 | if (no_fragment) { | |
5744 | ret = 1; | |
edbd8d4e | 5745 | goto out; |
1a40e23b ZY |
5746 | } |
5747 | path->slots[0]++; | |
5748 | } | |
5749 | ||
1f80e4db | 5750 | BUG_ON(cur_pos + offset > last_byte); |
1a40e23b ZY |
5751 | if (cur_pos + offset < last_byte) { |
5752 | ret = -ENOENT; | |
5753 | goto out; | |
edbd8d4e CM |
5754 | } |
5755 | ret = 0; | |
5756 | out: | |
1a40e23b ZY |
5757 | btrfs_free_path(path); |
5758 | if (ret) { | |
5759 | if (exts != *extents) | |
5760 | kfree(exts); | |
5761 | } else { | |
5762 | *extents = exts; | |
5763 | *nr_extents = nr; | |
5764 | } | |
5765 | return ret; | |
5766 | } | |
5767 | ||
d397712b | 5768 | static noinline int replace_one_extent(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
5769 | struct btrfs_root *root, |
5770 | struct btrfs_path *path, | |
5771 | struct btrfs_key *extent_key, | |
5772 | struct btrfs_key *leaf_key, | |
5773 | struct btrfs_ref_path *ref_path, | |
5774 | struct disk_extent *new_extents, | |
5775 | int nr_extents) | |
5776 | { | |
5777 | struct extent_buffer *leaf; | |
5778 | struct btrfs_file_extent_item *fi; | |
5779 | struct inode *inode = NULL; | |
5780 | struct btrfs_key key; | |
5781 | u64 lock_start = 0; | |
5782 | u64 lock_end = 0; | |
5783 | u64 num_bytes; | |
5784 | u64 ext_offset; | |
86288a19 | 5785 | u64 search_end = (u64)-1; |
1a40e23b | 5786 | u32 nritems; |
3bb1a1bc | 5787 | int nr_scaned = 0; |
1a40e23b | 5788 | int extent_locked = 0; |
d899e052 | 5789 | int extent_type; |
1a40e23b ZY |
5790 | int ret; |
5791 | ||
3bb1a1bc | 5792 | memcpy(&key, leaf_key, sizeof(key)); |
1a40e23b | 5793 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { |
3bb1a1bc YZ |
5794 | if (key.objectid < ref_path->owner_objectid || |
5795 | (key.objectid == ref_path->owner_objectid && | |
5796 | key.type < BTRFS_EXTENT_DATA_KEY)) { | |
5797 | key.objectid = ref_path->owner_objectid; | |
5798 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5799 | key.offset = 0; | |
5800 | } | |
1a40e23b ZY |
5801 | } |
5802 | ||
5803 | while (1) { | |
5804 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
5805 | if (ret < 0) | |
5806 | goto out; | |
5807 | ||
5808 | leaf = path->nodes[0]; | |
5809 | nritems = btrfs_header_nritems(leaf); | |
5810 | next: | |
5811 | if (extent_locked && ret > 0) { | |
5812 | /* | |
5813 | * the file extent item was modified by someone | |
5814 | * before the extent got locked. | |
5815 | */ | |
1a40e23b ZY |
5816 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
5817 | lock_end, GFP_NOFS); | |
5818 | extent_locked = 0; | |
5819 | } | |
5820 | ||
5821 | if (path->slots[0] >= nritems) { | |
3bb1a1bc | 5822 | if (++nr_scaned > 2) |
1a40e23b ZY |
5823 | break; |
5824 | ||
5825 | BUG_ON(extent_locked); | |
5826 | ret = btrfs_next_leaf(root, path); | |
5827 | if (ret < 0) | |
5828 | goto out; | |
5829 | if (ret > 0) | |
5830 | break; | |
5831 | leaf = path->nodes[0]; | |
5832 | nritems = btrfs_header_nritems(leaf); | |
5833 | } | |
5834 | ||
5835 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
5836 | ||
5837 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { | |
5838 | if ((key.objectid > ref_path->owner_objectid) || | |
5839 | (key.objectid == ref_path->owner_objectid && | |
5840 | key.type > BTRFS_EXTENT_DATA_KEY) || | |
86288a19 | 5841 | key.offset >= search_end) |
1a40e23b ZY |
5842 | break; |
5843 | } | |
5844 | ||
5845 | if (inode && key.objectid != inode->i_ino) { | |
5846 | BUG_ON(extent_locked); | |
5847 | btrfs_release_path(root, path); | |
5848 | mutex_unlock(&inode->i_mutex); | |
5849 | iput(inode); | |
5850 | inode = NULL; | |
5851 | continue; | |
5852 | } | |
5853 | ||
5854 | if (key.type != BTRFS_EXTENT_DATA_KEY) { | |
5855 | path->slots[0]++; | |
5856 | ret = 1; | |
5857 | goto next; | |
5858 | } | |
5859 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
5860 | struct btrfs_file_extent_item); | |
d899e052 YZ |
5861 | extent_type = btrfs_file_extent_type(leaf, fi); |
5862 | if ((extent_type != BTRFS_FILE_EXTENT_REG && | |
5863 | extent_type != BTRFS_FILE_EXTENT_PREALLOC) || | |
1a40e23b ZY |
5864 | (btrfs_file_extent_disk_bytenr(leaf, fi) != |
5865 | extent_key->objectid)) { | |
5866 | path->slots[0]++; | |
5867 | ret = 1; | |
5868 | goto next; | |
5869 | } | |
5870 | ||
5871 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
5872 | ext_offset = btrfs_file_extent_offset(leaf, fi); | |
5873 | ||
86288a19 YZ |
5874 | if (search_end == (u64)-1) { |
5875 | search_end = key.offset - ext_offset + | |
5876 | btrfs_file_extent_ram_bytes(leaf, fi); | |
5877 | } | |
1a40e23b ZY |
5878 | |
5879 | if (!extent_locked) { | |
5880 | lock_start = key.offset; | |
5881 | lock_end = lock_start + num_bytes - 1; | |
5882 | } else { | |
6643558d YZ |
5883 | if (lock_start > key.offset || |
5884 | lock_end + 1 < key.offset + num_bytes) { | |
5885 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
5886 | lock_start, lock_end, GFP_NOFS); | |
5887 | extent_locked = 0; | |
5888 | } | |
1a40e23b ZY |
5889 | } |
5890 | ||
5891 | if (!inode) { | |
5892 | btrfs_release_path(root, path); | |
5893 | ||
5894 | inode = btrfs_iget_locked(root->fs_info->sb, | |
5895 | key.objectid, root); | |
5896 | if (inode->i_state & I_NEW) { | |
5897 | BTRFS_I(inode)->root = root; | |
5898 | BTRFS_I(inode)->location.objectid = | |
5899 | key.objectid; | |
5900 | BTRFS_I(inode)->location.type = | |
5901 | BTRFS_INODE_ITEM_KEY; | |
5902 | BTRFS_I(inode)->location.offset = 0; | |
5903 | btrfs_read_locked_inode(inode); | |
5904 | unlock_new_inode(inode); | |
5905 | } | |
5906 | /* | |
5907 | * some code call btrfs_commit_transaction while | |
5908 | * holding the i_mutex, so we can't use mutex_lock | |
5909 | * here. | |
5910 | */ | |
5911 | if (is_bad_inode(inode) || | |
5912 | !mutex_trylock(&inode->i_mutex)) { | |
5913 | iput(inode); | |
5914 | inode = NULL; | |
5915 | key.offset = (u64)-1; | |
5916 | goto skip; | |
5917 | } | |
5918 | } | |
5919 | ||
5920 | if (!extent_locked) { | |
5921 | struct btrfs_ordered_extent *ordered; | |
5922 | ||
5923 | btrfs_release_path(root, path); | |
5924 | ||
5925 | lock_extent(&BTRFS_I(inode)->io_tree, lock_start, | |
5926 | lock_end, GFP_NOFS); | |
5927 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
5928 | lock_end); | |
5929 | if (ordered && | |
5930 | ordered->file_offset <= lock_end && | |
5931 | ordered->file_offset + ordered->len > lock_start) { | |
5932 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
5933 | lock_start, lock_end, GFP_NOFS); | |
5934 | btrfs_start_ordered_extent(inode, ordered, 1); | |
5935 | btrfs_put_ordered_extent(ordered); | |
5936 | key.offset += num_bytes; | |
5937 | goto skip; | |
5938 | } | |
5939 | if (ordered) | |
5940 | btrfs_put_ordered_extent(ordered); | |
5941 | ||
1a40e23b ZY |
5942 | extent_locked = 1; |
5943 | continue; | |
5944 | } | |
5945 | ||
5946 | if (nr_extents == 1) { | |
5947 | /* update extent pointer in place */ | |
1a40e23b ZY |
5948 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
5949 | new_extents[0].disk_bytenr); | |
5950 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
5951 | new_extents[0].disk_num_bytes); | |
1a40e23b ZY |
5952 | btrfs_mark_buffer_dirty(leaf); |
5953 | ||
5954 | btrfs_drop_extent_cache(inode, key.offset, | |
5955 | key.offset + num_bytes - 1, 0); | |
5956 | ||
5957 | ret = btrfs_inc_extent_ref(trans, root, | |
5958 | new_extents[0].disk_bytenr, | |
5959 | new_extents[0].disk_num_bytes, | |
5960 | leaf->start, | |
5961 | root->root_key.objectid, | |
5962 | trans->transid, | |
3bb1a1bc | 5963 | key.objectid); |
1a40e23b ZY |
5964 | BUG_ON(ret); |
5965 | ||
5966 | ret = btrfs_free_extent(trans, root, | |
5967 | extent_key->objectid, | |
5968 | extent_key->offset, | |
5969 | leaf->start, | |
5970 | btrfs_header_owner(leaf), | |
5971 | btrfs_header_generation(leaf), | |
3bb1a1bc | 5972 | key.objectid, 0); |
1a40e23b ZY |
5973 | BUG_ON(ret); |
5974 | ||
5975 | btrfs_release_path(root, path); | |
5976 | key.offset += num_bytes; | |
5977 | } else { | |
d899e052 YZ |
5978 | BUG_ON(1); |
5979 | #if 0 | |
1a40e23b ZY |
5980 | u64 alloc_hint; |
5981 | u64 extent_len; | |
5982 | int i; | |
5983 | /* | |
5984 | * drop old extent pointer at first, then insert the | |
5985 | * new pointers one bye one | |
5986 | */ | |
5987 | btrfs_release_path(root, path); | |
5988 | ret = btrfs_drop_extents(trans, root, inode, key.offset, | |
5989 | key.offset + num_bytes, | |
5990 | key.offset, &alloc_hint); | |
5991 | BUG_ON(ret); | |
5992 | ||
5993 | for (i = 0; i < nr_extents; i++) { | |
5994 | if (ext_offset >= new_extents[i].num_bytes) { | |
5995 | ext_offset -= new_extents[i].num_bytes; | |
5996 | continue; | |
5997 | } | |
5998 | extent_len = min(new_extents[i].num_bytes - | |
5999 | ext_offset, num_bytes); | |
6000 | ||
6001 | ret = btrfs_insert_empty_item(trans, root, | |
6002 | path, &key, | |
6003 | sizeof(*fi)); | |
6004 | BUG_ON(ret); | |
6005 | ||
6006 | leaf = path->nodes[0]; | |
6007 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
6008 | struct btrfs_file_extent_item); | |
6009 | btrfs_set_file_extent_generation(leaf, fi, | |
6010 | trans->transid); | |
6011 | btrfs_set_file_extent_type(leaf, fi, | |
6012 | BTRFS_FILE_EXTENT_REG); | |
6013 | btrfs_set_file_extent_disk_bytenr(leaf, fi, | |
6014 | new_extents[i].disk_bytenr); | |
6015 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
6016 | new_extents[i].disk_num_bytes); | |
c8b97818 CM |
6017 | btrfs_set_file_extent_ram_bytes(leaf, fi, |
6018 | new_extents[i].ram_bytes); | |
6019 | ||
6020 | btrfs_set_file_extent_compression(leaf, fi, | |
6021 | new_extents[i].compression); | |
6022 | btrfs_set_file_extent_encryption(leaf, fi, | |
6023 | new_extents[i].encryption); | |
6024 | btrfs_set_file_extent_other_encoding(leaf, fi, | |
6025 | new_extents[i].other_encoding); | |
6026 | ||
1a40e23b ZY |
6027 | btrfs_set_file_extent_num_bytes(leaf, fi, |
6028 | extent_len); | |
6029 | ext_offset += new_extents[i].offset; | |
6030 | btrfs_set_file_extent_offset(leaf, fi, | |
6031 | ext_offset); | |
6032 | btrfs_mark_buffer_dirty(leaf); | |
6033 | ||
6034 | btrfs_drop_extent_cache(inode, key.offset, | |
6035 | key.offset + extent_len - 1, 0); | |
6036 | ||
6037 | ret = btrfs_inc_extent_ref(trans, root, | |
6038 | new_extents[i].disk_bytenr, | |
6039 | new_extents[i].disk_num_bytes, | |
6040 | leaf->start, | |
6041 | root->root_key.objectid, | |
3bb1a1bc | 6042 | trans->transid, key.objectid); |
1a40e23b ZY |
6043 | BUG_ON(ret); |
6044 | btrfs_release_path(root, path); | |
6045 | ||
a76a3cd4 | 6046 | inode_add_bytes(inode, extent_len); |
1a40e23b ZY |
6047 | |
6048 | ext_offset = 0; | |
6049 | num_bytes -= extent_len; | |
6050 | key.offset += extent_len; | |
6051 | ||
6052 | if (num_bytes == 0) | |
6053 | break; | |
6054 | } | |
6055 | BUG_ON(i >= nr_extents); | |
d899e052 | 6056 | #endif |
1a40e23b ZY |
6057 | } |
6058 | ||
6059 | if (extent_locked) { | |
1a40e23b ZY |
6060 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
6061 | lock_end, GFP_NOFS); | |
6062 | extent_locked = 0; | |
6063 | } | |
6064 | skip: | |
6065 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS && | |
86288a19 | 6066 | key.offset >= search_end) |
1a40e23b ZY |
6067 | break; |
6068 | ||
6069 | cond_resched(); | |
6070 | } | |
6071 | ret = 0; | |
6072 | out: | |
6073 | btrfs_release_path(root, path); | |
6074 | if (inode) { | |
6075 | mutex_unlock(&inode->i_mutex); | |
6076 | if (extent_locked) { | |
1a40e23b ZY |
6077 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
6078 | lock_end, GFP_NOFS); | |
6079 | } | |
6080 | iput(inode); | |
6081 | } | |
6082 | return ret; | |
6083 | } | |
6084 | ||
1a40e23b ZY |
6085 | int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, |
6086 | struct btrfs_root *root, | |
6087 | struct extent_buffer *buf, u64 orig_start) | |
6088 | { | |
6089 | int level; | |
6090 | int ret; | |
6091 | ||
6092 | BUG_ON(btrfs_header_generation(buf) != trans->transid); | |
6093 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
6094 | ||
6095 | level = btrfs_header_level(buf); | |
6096 | if (level == 0) { | |
6097 | struct btrfs_leaf_ref *ref; | |
6098 | struct btrfs_leaf_ref *orig_ref; | |
6099 | ||
6100 | orig_ref = btrfs_lookup_leaf_ref(root, orig_start); | |
6101 | if (!orig_ref) | |
6102 | return -ENOENT; | |
6103 | ||
6104 | ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems); | |
6105 | if (!ref) { | |
6106 | btrfs_free_leaf_ref(root, orig_ref); | |
6107 | return -ENOMEM; | |
6108 | } | |
6109 | ||
6110 | ref->nritems = orig_ref->nritems; | |
6111 | memcpy(ref->extents, orig_ref->extents, | |
6112 | sizeof(ref->extents[0]) * ref->nritems); | |
6113 | ||
6114 | btrfs_free_leaf_ref(root, orig_ref); | |
6115 | ||
6116 | ref->root_gen = trans->transid; | |
6117 | ref->bytenr = buf->start; | |
6118 | ref->owner = btrfs_header_owner(buf); | |
6119 | ref->generation = btrfs_header_generation(buf); | |
bd56b302 | 6120 | |
1a40e23b ZY |
6121 | ret = btrfs_add_leaf_ref(root, ref, 0); |
6122 | WARN_ON(ret); | |
6123 | btrfs_free_leaf_ref(root, ref); | |
6124 | } | |
6125 | return 0; | |
6126 | } | |
6127 | ||
d397712b | 6128 | static noinline int invalidate_extent_cache(struct btrfs_root *root, |
1a40e23b ZY |
6129 | struct extent_buffer *leaf, |
6130 | struct btrfs_block_group_cache *group, | |
6131 | struct btrfs_root *target_root) | |
6132 | { | |
6133 | struct btrfs_key key; | |
6134 | struct inode *inode = NULL; | |
6135 | struct btrfs_file_extent_item *fi; | |
6136 | u64 num_bytes; | |
6137 | u64 skip_objectid = 0; | |
6138 | u32 nritems; | |
6139 | u32 i; | |
6140 | ||
6141 | nritems = btrfs_header_nritems(leaf); | |
6142 | for (i = 0; i < nritems; i++) { | |
6143 | btrfs_item_key_to_cpu(leaf, &key, i); | |
6144 | if (key.objectid == skip_objectid || | |
6145 | key.type != BTRFS_EXTENT_DATA_KEY) | |
6146 | continue; | |
6147 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
6148 | if (btrfs_file_extent_type(leaf, fi) == | |
6149 | BTRFS_FILE_EXTENT_INLINE) | |
6150 | continue; | |
6151 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
6152 | continue; | |
6153 | if (!inode || inode->i_ino != key.objectid) { | |
6154 | iput(inode); | |
6155 | inode = btrfs_ilookup(target_root->fs_info->sb, | |
6156 | key.objectid, target_root, 1); | |
6157 | } | |
6158 | if (!inode) { | |
6159 | skip_objectid = key.objectid; | |
6160 | continue; | |
6161 | } | |
6162 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
6163 | ||
6164 | lock_extent(&BTRFS_I(inode)->io_tree, key.offset, | |
6165 | key.offset + num_bytes - 1, GFP_NOFS); | |
1a40e23b ZY |
6166 | btrfs_drop_extent_cache(inode, key.offset, |
6167 | key.offset + num_bytes - 1, 1); | |
1a40e23b ZY |
6168 | unlock_extent(&BTRFS_I(inode)->io_tree, key.offset, |
6169 | key.offset + num_bytes - 1, GFP_NOFS); | |
6170 | cond_resched(); | |
6171 | } | |
6172 | iput(inode); | |
6173 | return 0; | |
6174 | } | |
6175 | ||
d397712b | 6176 | static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6177 | struct btrfs_root *root, |
6178 | struct extent_buffer *leaf, | |
6179 | struct btrfs_block_group_cache *group, | |
6180 | struct inode *reloc_inode) | |
6181 | { | |
6182 | struct btrfs_key key; | |
6183 | struct btrfs_key extent_key; | |
6184 | struct btrfs_file_extent_item *fi; | |
6185 | struct btrfs_leaf_ref *ref; | |
6186 | struct disk_extent *new_extent; | |
6187 | u64 bytenr; | |
6188 | u64 num_bytes; | |
6189 | u32 nritems; | |
6190 | u32 i; | |
6191 | int ext_index; | |
6192 | int nr_extent; | |
6193 | int ret; | |
6194 | ||
6195 | new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS); | |
6196 | BUG_ON(!new_extent); | |
6197 | ||
6198 | ref = btrfs_lookup_leaf_ref(root, leaf->start); | |
6199 | BUG_ON(!ref); | |
6200 | ||
6201 | ext_index = -1; | |
6202 | nritems = btrfs_header_nritems(leaf); | |
6203 | for (i = 0; i < nritems; i++) { | |
6204 | btrfs_item_key_to_cpu(leaf, &key, i); | |
6205 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6206 | continue; | |
6207 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
6208 | if (btrfs_file_extent_type(leaf, fi) == | |
6209 | BTRFS_FILE_EXTENT_INLINE) | |
6210 | continue; | |
6211 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
6212 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6213 | if (bytenr == 0) | |
6214 | continue; | |
6215 | ||
6216 | ext_index++; | |
6217 | if (bytenr >= group->key.objectid + group->key.offset || | |
6218 | bytenr + num_bytes <= group->key.objectid) | |
6219 | continue; | |
6220 | ||
6221 | extent_key.objectid = bytenr; | |
6222 | extent_key.offset = num_bytes; | |
6223 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
6224 | nr_extent = 1; | |
6225 | ret = get_new_locations(reloc_inode, &extent_key, | |
6226 | group->key.objectid, 1, | |
6227 | &new_extent, &nr_extent); | |
6228 | if (ret > 0) | |
6229 | continue; | |
6230 | BUG_ON(ret < 0); | |
6231 | ||
6232 | BUG_ON(ref->extents[ext_index].bytenr != bytenr); | |
6233 | BUG_ON(ref->extents[ext_index].num_bytes != num_bytes); | |
6234 | ref->extents[ext_index].bytenr = new_extent->disk_bytenr; | |
6235 | ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes; | |
6236 | ||
1a40e23b ZY |
6237 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
6238 | new_extent->disk_bytenr); | |
6239 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
6240 | new_extent->disk_num_bytes); | |
1a40e23b ZY |
6241 | btrfs_mark_buffer_dirty(leaf); |
6242 | ||
6243 | ret = btrfs_inc_extent_ref(trans, root, | |
6244 | new_extent->disk_bytenr, | |
6245 | new_extent->disk_num_bytes, | |
6246 | leaf->start, | |
6247 | root->root_key.objectid, | |
3bb1a1bc | 6248 | trans->transid, key.objectid); |
1a40e23b | 6249 | BUG_ON(ret); |
56bec294 | 6250 | |
1a40e23b ZY |
6251 | ret = btrfs_free_extent(trans, root, |
6252 | bytenr, num_bytes, leaf->start, | |
6253 | btrfs_header_owner(leaf), | |
6254 | btrfs_header_generation(leaf), | |
3bb1a1bc | 6255 | key.objectid, 0); |
1a40e23b ZY |
6256 | BUG_ON(ret); |
6257 | cond_resched(); | |
6258 | } | |
6259 | kfree(new_extent); | |
6260 | BUG_ON(ext_index + 1 != ref->nritems); | |
6261 | btrfs_free_leaf_ref(root, ref); | |
6262 | return 0; | |
6263 | } | |
6264 | ||
f82d02d9 YZ |
6265 | int btrfs_free_reloc_root(struct btrfs_trans_handle *trans, |
6266 | struct btrfs_root *root) | |
1a40e23b ZY |
6267 | { |
6268 | struct btrfs_root *reloc_root; | |
f82d02d9 | 6269 | int ret; |
1a40e23b ZY |
6270 | |
6271 | if (root->reloc_root) { | |
6272 | reloc_root = root->reloc_root; | |
6273 | root->reloc_root = NULL; | |
6274 | list_add(&reloc_root->dead_list, | |
6275 | &root->fs_info->dead_reloc_roots); | |
f82d02d9 YZ |
6276 | |
6277 | btrfs_set_root_bytenr(&reloc_root->root_item, | |
6278 | reloc_root->node->start); | |
6279 | btrfs_set_root_level(&root->root_item, | |
6280 | btrfs_header_level(reloc_root->node)); | |
6281 | memset(&reloc_root->root_item.drop_progress, 0, | |
6282 | sizeof(struct btrfs_disk_key)); | |
6283 | reloc_root->root_item.drop_level = 0; | |
6284 | ||
6285 | ret = btrfs_update_root(trans, root->fs_info->tree_root, | |
6286 | &reloc_root->root_key, | |
6287 | &reloc_root->root_item); | |
6288 | BUG_ON(ret); | |
1a40e23b ZY |
6289 | } |
6290 | return 0; | |
6291 | } | |
6292 | ||
6293 | int btrfs_drop_dead_reloc_roots(struct btrfs_root *root) | |
6294 | { | |
6295 | struct btrfs_trans_handle *trans; | |
6296 | struct btrfs_root *reloc_root; | |
6297 | struct btrfs_root *prev_root = NULL; | |
6298 | struct list_head dead_roots; | |
6299 | int ret; | |
6300 | unsigned long nr; | |
6301 | ||
6302 | INIT_LIST_HEAD(&dead_roots); | |
6303 | list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots); | |
6304 | ||
6305 | while (!list_empty(&dead_roots)) { | |
6306 | reloc_root = list_entry(dead_roots.prev, | |
6307 | struct btrfs_root, dead_list); | |
6308 | list_del_init(&reloc_root->dead_list); | |
6309 | ||
6310 | BUG_ON(reloc_root->commit_root != NULL); | |
6311 | while (1) { | |
6312 | trans = btrfs_join_transaction(root, 1); | |
6313 | BUG_ON(!trans); | |
6314 | ||
6315 | mutex_lock(&root->fs_info->drop_mutex); | |
6316 | ret = btrfs_drop_snapshot(trans, reloc_root); | |
6317 | if (ret != -EAGAIN) | |
6318 | break; | |
6319 | mutex_unlock(&root->fs_info->drop_mutex); | |
6320 | ||
6321 | nr = trans->blocks_used; | |
6322 | ret = btrfs_end_transaction(trans, root); | |
6323 | BUG_ON(ret); | |
6324 | btrfs_btree_balance_dirty(root, nr); | |
6325 | } | |
6326 | ||
6327 | free_extent_buffer(reloc_root->node); | |
6328 | ||
6329 | ret = btrfs_del_root(trans, root->fs_info->tree_root, | |
6330 | &reloc_root->root_key); | |
6331 | BUG_ON(ret); | |
6332 | mutex_unlock(&root->fs_info->drop_mutex); | |
6333 | ||
6334 | nr = trans->blocks_used; | |
6335 | ret = btrfs_end_transaction(trans, root); | |
6336 | BUG_ON(ret); | |
6337 | btrfs_btree_balance_dirty(root, nr); | |
6338 | ||
6339 | kfree(prev_root); | |
6340 | prev_root = reloc_root; | |
6341 | } | |
6342 | if (prev_root) { | |
6343 | btrfs_remove_leaf_refs(prev_root, (u64)-1, 0); | |
6344 | kfree(prev_root); | |
6345 | } | |
6346 | return 0; | |
6347 | } | |
6348 | ||
6349 | int btrfs_add_dead_reloc_root(struct btrfs_root *root) | |
6350 | { | |
6351 | list_add(&root->dead_list, &root->fs_info->dead_reloc_roots); | |
6352 | return 0; | |
6353 | } | |
6354 | ||
6355 | int btrfs_cleanup_reloc_trees(struct btrfs_root *root) | |
6356 | { | |
6357 | struct btrfs_root *reloc_root; | |
6358 | struct btrfs_trans_handle *trans; | |
6359 | struct btrfs_key location; | |
6360 | int found; | |
6361 | int ret; | |
6362 | ||
6363 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
6364 | ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL); | |
6365 | BUG_ON(ret); | |
6366 | found = !list_empty(&root->fs_info->dead_reloc_roots); | |
6367 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
6368 | ||
6369 | if (found) { | |
6370 | trans = btrfs_start_transaction(root, 1); | |
6371 | BUG_ON(!trans); | |
6372 | ret = btrfs_commit_transaction(trans, root); | |
6373 | BUG_ON(ret); | |
6374 | } | |
6375 | ||
6376 | location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
6377 | location.offset = (u64)-1; | |
6378 | location.type = BTRFS_ROOT_ITEM_KEY; | |
6379 | ||
6380 | reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location); | |
6381 | BUG_ON(!reloc_root); | |
6382 | btrfs_orphan_cleanup(reloc_root); | |
6383 | return 0; | |
6384 | } | |
6385 | ||
d397712b | 6386 | static noinline int init_reloc_tree(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6387 | struct btrfs_root *root) |
6388 | { | |
6389 | struct btrfs_root *reloc_root; | |
6390 | struct extent_buffer *eb; | |
6391 | struct btrfs_root_item *root_item; | |
6392 | struct btrfs_key root_key; | |
6393 | int ret; | |
6394 | ||
6395 | BUG_ON(!root->ref_cows); | |
6396 | if (root->reloc_root) | |
6397 | return 0; | |
6398 | ||
6399 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); | |
6400 | BUG_ON(!root_item); | |
6401 | ||
6402 | ret = btrfs_copy_root(trans, root, root->commit_root, | |
6403 | &eb, BTRFS_TREE_RELOC_OBJECTID); | |
6404 | BUG_ON(ret); | |
6405 | ||
6406 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
6407 | root_key.offset = root->root_key.objectid; | |
6408 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
6409 | ||
6410 | memcpy(root_item, &root->root_item, sizeof(root_item)); | |
6411 | btrfs_set_root_refs(root_item, 0); | |
6412 | btrfs_set_root_bytenr(root_item, eb->start); | |
6413 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
84234f3a | 6414 | btrfs_set_root_generation(root_item, trans->transid); |
1a40e23b ZY |
6415 | |
6416 | btrfs_tree_unlock(eb); | |
6417 | free_extent_buffer(eb); | |
6418 | ||
6419 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, | |
6420 | &root_key, root_item); | |
6421 | BUG_ON(ret); | |
6422 | kfree(root_item); | |
6423 | ||
6424 | reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root, | |
6425 | &root_key); | |
6426 | BUG_ON(!reloc_root); | |
6427 | reloc_root->last_trans = trans->transid; | |
6428 | reloc_root->commit_root = NULL; | |
6429 | reloc_root->ref_tree = &root->fs_info->reloc_ref_tree; | |
6430 | ||
6431 | root->reloc_root = reloc_root; | |
6432 | return 0; | |
6433 | } | |
6434 | ||
6435 | /* | |
6436 | * Core function of space balance. | |
6437 | * | |
6438 | * The idea is using reloc trees to relocate tree blocks in reference | |
f82d02d9 YZ |
6439 | * counted roots. There is one reloc tree for each subvol, and all |
6440 | * reloc trees share same root key objectid. Reloc trees are snapshots | |
6441 | * of the latest committed roots of subvols (root->commit_root). | |
6442 | * | |
6443 | * To relocate a tree block referenced by a subvol, there are two steps. | |
6444 | * COW the block through subvol's reloc tree, then update block pointer | |
6445 | * in the subvol to point to the new block. Since all reloc trees share | |
6446 | * same root key objectid, doing special handing for tree blocks owned | |
6447 | * by them is easy. Once a tree block has been COWed in one reloc tree, | |
6448 | * we can use the resulting new block directly when the same block is | |
6449 | * required to COW again through other reloc trees. By this way, relocated | |
6450 | * tree blocks are shared between reloc trees, so they are also shared | |
6451 | * between subvols. | |
1a40e23b | 6452 | */ |
d397712b | 6453 | static noinline int relocate_one_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6454 | struct btrfs_root *root, |
6455 | struct btrfs_path *path, | |
6456 | struct btrfs_key *first_key, | |
6457 | struct btrfs_ref_path *ref_path, | |
6458 | struct btrfs_block_group_cache *group, | |
6459 | struct inode *reloc_inode) | |
6460 | { | |
6461 | struct btrfs_root *reloc_root; | |
6462 | struct extent_buffer *eb = NULL; | |
6463 | struct btrfs_key *keys; | |
6464 | u64 *nodes; | |
6465 | int level; | |
f82d02d9 | 6466 | int shared_level; |
1a40e23b | 6467 | int lowest_level = 0; |
1a40e23b ZY |
6468 | int ret; |
6469 | ||
6470 | if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
6471 | lowest_level = ref_path->owner_objectid; | |
6472 | ||
f82d02d9 | 6473 | if (!root->ref_cows) { |
1a40e23b ZY |
6474 | path->lowest_level = lowest_level; |
6475 | ret = btrfs_search_slot(trans, root, first_key, path, 0, 1); | |
6476 | BUG_ON(ret < 0); | |
6477 | path->lowest_level = 0; | |
6478 | btrfs_release_path(root, path); | |
6479 | return 0; | |
6480 | } | |
6481 | ||
1a40e23b ZY |
6482 | mutex_lock(&root->fs_info->tree_reloc_mutex); |
6483 | ret = init_reloc_tree(trans, root); | |
6484 | BUG_ON(ret); | |
6485 | reloc_root = root->reloc_root; | |
6486 | ||
f82d02d9 YZ |
6487 | shared_level = ref_path->shared_level; |
6488 | ref_path->shared_level = BTRFS_MAX_LEVEL - 1; | |
1a40e23b | 6489 | |
f82d02d9 YZ |
6490 | keys = ref_path->node_keys; |
6491 | nodes = ref_path->new_nodes; | |
6492 | memset(&keys[shared_level + 1], 0, | |
6493 | sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
6494 | memset(&nodes[shared_level + 1], 0, | |
6495 | sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
1a40e23b | 6496 | |
f82d02d9 YZ |
6497 | if (nodes[lowest_level] == 0) { |
6498 | path->lowest_level = lowest_level; | |
6499 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, | |
6500 | 0, 1); | |
6501 | BUG_ON(ret); | |
6502 | for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) { | |
6503 | eb = path->nodes[level]; | |
6504 | if (!eb || eb == reloc_root->node) | |
6505 | break; | |
6506 | nodes[level] = eb->start; | |
6507 | if (level == 0) | |
6508 | btrfs_item_key_to_cpu(eb, &keys[level], 0); | |
6509 | else | |
6510 | btrfs_node_key_to_cpu(eb, &keys[level], 0); | |
6511 | } | |
2b82032c YZ |
6512 | if (nodes[0] && |
6513 | ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
6514 | eb = path->nodes[0]; |
6515 | ret = replace_extents_in_leaf(trans, reloc_root, eb, | |
6516 | group, reloc_inode); | |
6517 | BUG_ON(ret); | |
6518 | } | |
6519 | btrfs_release_path(reloc_root, path); | |
6520 | } else { | |
1a40e23b | 6521 | ret = btrfs_merge_path(trans, reloc_root, keys, nodes, |
f82d02d9 | 6522 | lowest_level); |
1a40e23b ZY |
6523 | BUG_ON(ret); |
6524 | } | |
6525 | ||
1a40e23b ZY |
6526 | /* |
6527 | * replace tree blocks in the fs tree with tree blocks in | |
6528 | * the reloc tree. | |
6529 | */ | |
6530 | ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level); | |
6531 | BUG_ON(ret < 0); | |
6532 | ||
6533 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
6534 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, |
6535 | 0, 0); | |
6536 | BUG_ON(ret); | |
6537 | extent_buffer_get(path->nodes[0]); | |
6538 | eb = path->nodes[0]; | |
6539 | btrfs_release_path(reloc_root, path); | |
1a40e23b ZY |
6540 | ret = invalidate_extent_cache(reloc_root, eb, group, root); |
6541 | BUG_ON(ret); | |
6542 | free_extent_buffer(eb); | |
6543 | } | |
1a40e23b | 6544 | |
f82d02d9 | 6545 | mutex_unlock(&root->fs_info->tree_reloc_mutex); |
1a40e23b | 6546 | path->lowest_level = 0; |
1a40e23b ZY |
6547 | return 0; |
6548 | } | |
6549 | ||
d397712b | 6550 | static noinline int relocate_tree_block(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6551 | struct btrfs_root *root, |
6552 | struct btrfs_path *path, | |
6553 | struct btrfs_key *first_key, | |
6554 | struct btrfs_ref_path *ref_path) | |
6555 | { | |
6556 | int ret; | |
1a40e23b ZY |
6557 | |
6558 | ret = relocate_one_path(trans, root, path, first_key, | |
6559 | ref_path, NULL, NULL); | |
6560 | BUG_ON(ret); | |
6561 | ||
1a40e23b ZY |
6562 | return 0; |
6563 | } | |
6564 | ||
d397712b | 6565 | static noinline int del_extent_zero(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6566 | struct btrfs_root *extent_root, |
6567 | struct btrfs_path *path, | |
6568 | struct btrfs_key *extent_key) | |
6569 | { | |
6570 | int ret; | |
6571 | ||
1a40e23b ZY |
6572 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); |
6573 | if (ret) | |
6574 | goto out; | |
6575 | ret = btrfs_del_item(trans, extent_root, path); | |
6576 | out: | |
6577 | btrfs_release_path(extent_root, path); | |
1a40e23b ZY |
6578 | return ret; |
6579 | } | |
6580 | ||
d397712b | 6581 | static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info, |
1a40e23b ZY |
6582 | struct btrfs_ref_path *ref_path) |
6583 | { | |
6584 | struct btrfs_key root_key; | |
6585 | ||
6586 | root_key.objectid = ref_path->root_objectid; | |
6587 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
6588 | if (is_cowonly_root(ref_path->root_objectid)) | |
6589 | root_key.offset = 0; | |
6590 | else | |
6591 | root_key.offset = (u64)-1; | |
6592 | ||
6593 | return btrfs_read_fs_root_no_name(fs_info, &root_key); | |
6594 | } | |
6595 | ||
d397712b | 6596 | static noinline int relocate_one_extent(struct btrfs_root *extent_root, |
1a40e23b ZY |
6597 | struct btrfs_path *path, |
6598 | struct btrfs_key *extent_key, | |
6599 | struct btrfs_block_group_cache *group, | |
6600 | struct inode *reloc_inode, int pass) | |
6601 | { | |
6602 | struct btrfs_trans_handle *trans; | |
6603 | struct btrfs_root *found_root; | |
6604 | struct btrfs_ref_path *ref_path = NULL; | |
6605 | struct disk_extent *new_extents = NULL; | |
6606 | int nr_extents = 0; | |
6607 | int loops; | |
6608 | int ret; | |
6609 | int level; | |
6610 | struct btrfs_key first_key; | |
6611 | u64 prev_block = 0; | |
6612 | ||
1a40e23b ZY |
6613 | |
6614 | trans = btrfs_start_transaction(extent_root, 1); | |
6615 | BUG_ON(!trans); | |
6616 | ||
6617 | if (extent_key->objectid == 0) { | |
6618 | ret = del_extent_zero(trans, extent_root, path, extent_key); | |
6619 | goto out; | |
6620 | } | |
6621 | ||
6622 | ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS); | |
6623 | if (!ref_path) { | |
d397712b CM |
6624 | ret = -ENOMEM; |
6625 | goto out; | |
1a40e23b ZY |
6626 | } |
6627 | ||
6628 | for (loops = 0; ; loops++) { | |
6629 | if (loops == 0) { | |
6630 | ret = btrfs_first_ref_path(trans, extent_root, ref_path, | |
6631 | extent_key->objectid); | |
6632 | } else { | |
6633 | ret = btrfs_next_ref_path(trans, extent_root, ref_path); | |
6634 | } | |
6635 | if (ret < 0) | |
6636 | goto out; | |
6637 | if (ret > 0) | |
6638 | break; | |
6639 | ||
6640 | if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID || | |
6641 | ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
6642 | continue; | |
6643 | ||
6644 | found_root = read_ref_root(extent_root->fs_info, ref_path); | |
6645 | BUG_ON(!found_root); | |
6646 | /* | |
6647 | * for reference counted tree, only process reference paths | |
6648 | * rooted at the latest committed root. | |
6649 | */ | |
6650 | if (found_root->ref_cows && | |
6651 | ref_path->root_generation != found_root->root_key.offset) | |
6652 | continue; | |
6653 | ||
6654 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
6655 | if (pass == 0) { | |
6656 | /* | |
6657 | * copy data extents to new locations | |
6658 | */ | |
6659 | u64 group_start = group->key.objectid; | |
6660 | ret = relocate_data_extent(reloc_inode, | |
6661 | extent_key, | |
6662 | group_start); | |
6663 | if (ret < 0) | |
6664 | goto out; | |
6665 | break; | |
6666 | } | |
6667 | level = 0; | |
6668 | } else { | |
6669 | level = ref_path->owner_objectid; | |
6670 | } | |
6671 | ||
6672 | if (prev_block != ref_path->nodes[level]) { | |
6673 | struct extent_buffer *eb; | |
6674 | u64 block_start = ref_path->nodes[level]; | |
6675 | u64 block_size = btrfs_level_size(found_root, level); | |
6676 | ||
6677 | eb = read_tree_block(found_root, block_start, | |
6678 | block_size, 0); | |
6679 | btrfs_tree_lock(eb); | |
6680 | BUG_ON(level != btrfs_header_level(eb)); | |
6681 | ||
6682 | if (level == 0) | |
6683 | btrfs_item_key_to_cpu(eb, &first_key, 0); | |
6684 | else | |
6685 | btrfs_node_key_to_cpu(eb, &first_key, 0); | |
6686 | ||
6687 | btrfs_tree_unlock(eb); | |
6688 | free_extent_buffer(eb); | |
6689 | prev_block = block_start; | |
6690 | } | |
6691 | ||
24562425 | 6692 | mutex_lock(&extent_root->fs_info->trans_mutex); |
e4404d6e | 6693 | btrfs_record_root_in_trans(found_root); |
24562425 | 6694 | mutex_unlock(&extent_root->fs_info->trans_mutex); |
e4404d6e YZ |
6695 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { |
6696 | /* | |
6697 | * try to update data extent references while | |
6698 | * keeping metadata shared between snapshots. | |
6699 | */ | |
6700 | if (pass == 1) { | |
6701 | ret = relocate_one_path(trans, found_root, | |
6702 | path, &first_key, ref_path, | |
6703 | group, reloc_inode); | |
6704 | if (ret < 0) | |
6705 | goto out; | |
6706 | continue; | |
6707 | } | |
1a40e23b ZY |
6708 | /* |
6709 | * use fallback method to process the remaining | |
6710 | * references. | |
6711 | */ | |
6712 | if (!new_extents) { | |
6713 | u64 group_start = group->key.objectid; | |
d899e052 YZ |
6714 | new_extents = kmalloc(sizeof(*new_extents), |
6715 | GFP_NOFS); | |
6716 | nr_extents = 1; | |
1a40e23b ZY |
6717 | ret = get_new_locations(reloc_inode, |
6718 | extent_key, | |
d899e052 | 6719 | group_start, 1, |
1a40e23b ZY |
6720 | &new_extents, |
6721 | &nr_extents); | |
d899e052 | 6722 | if (ret) |
1a40e23b ZY |
6723 | goto out; |
6724 | } | |
1a40e23b ZY |
6725 | ret = replace_one_extent(trans, found_root, |
6726 | path, extent_key, | |
6727 | &first_key, ref_path, | |
6728 | new_extents, nr_extents); | |
e4404d6e | 6729 | } else { |
1a40e23b ZY |
6730 | ret = relocate_tree_block(trans, found_root, path, |
6731 | &first_key, ref_path); | |
1a40e23b ZY |
6732 | } |
6733 | if (ret < 0) | |
6734 | goto out; | |
6735 | } | |
6736 | ret = 0; | |
6737 | out: | |
6738 | btrfs_end_transaction(trans, extent_root); | |
6739 | kfree(new_extents); | |
6740 | kfree(ref_path); | |
1a40e23b ZY |
6741 | return ret; |
6742 | } | |
5d4f98a2 | 6743 | #endif |
1a40e23b | 6744 | |
ec44a35c CM |
6745 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
6746 | { | |
6747 | u64 num_devices; | |
6748 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
6749 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
6750 | ||
2b82032c | 6751 | num_devices = root->fs_info->fs_devices->rw_devices; |
ec44a35c CM |
6752 | if (num_devices == 1) { |
6753 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6754 | stripped = flags & ~stripped; | |
6755 | ||
6756 | /* turn raid0 into single device chunks */ | |
6757 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
6758 | return stripped; | |
6759 | ||
6760 | /* turn mirroring into duplication */ | |
6761 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
6762 | BTRFS_BLOCK_GROUP_RAID10)) | |
6763 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
6764 | return flags; | |
6765 | } else { | |
6766 | /* they already had raid on here, just return */ | |
ec44a35c CM |
6767 | if (flags & stripped) |
6768 | return flags; | |
6769 | ||
6770 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6771 | stripped = flags & ~stripped; | |
6772 | ||
6773 | /* switch duplicated blocks with raid1 */ | |
6774 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
6775 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
6776 | ||
6777 | /* turn single device chunks into raid0 */ | |
6778 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
6779 | } | |
6780 | return flags; | |
6781 | } | |
6782 | ||
b2950863 | 6783 | static int __alloc_chunk_for_shrink(struct btrfs_root *root, |
0ef3e66b CM |
6784 | struct btrfs_block_group_cache *shrink_block_group, |
6785 | int force) | |
6786 | { | |
6787 | struct btrfs_trans_handle *trans; | |
6788 | u64 new_alloc_flags; | |
6789 | u64 calc; | |
6790 | ||
c286ac48 | 6791 | spin_lock(&shrink_block_group->lock); |
5d4f98a2 YZ |
6792 | if (btrfs_block_group_used(&shrink_block_group->item) + |
6793 | shrink_block_group->reserved > 0) { | |
c286ac48 | 6794 | spin_unlock(&shrink_block_group->lock); |
c286ac48 | 6795 | |
0ef3e66b | 6796 | trans = btrfs_start_transaction(root, 1); |
c286ac48 | 6797 | spin_lock(&shrink_block_group->lock); |
7d9eb12c | 6798 | |
0ef3e66b CM |
6799 | new_alloc_flags = update_block_group_flags(root, |
6800 | shrink_block_group->flags); | |
6801 | if (new_alloc_flags != shrink_block_group->flags) { | |
6802 | calc = | |
6803 | btrfs_block_group_used(&shrink_block_group->item); | |
6804 | } else { | |
6805 | calc = shrink_block_group->key.offset; | |
6806 | } | |
c286ac48 CM |
6807 | spin_unlock(&shrink_block_group->lock); |
6808 | ||
0ef3e66b CM |
6809 | do_chunk_alloc(trans, root->fs_info->extent_root, |
6810 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
7d9eb12c | 6811 | |
0ef3e66b | 6812 | btrfs_end_transaction(trans, root); |
c286ac48 CM |
6813 | } else |
6814 | spin_unlock(&shrink_block_group->lock); | |
0ef3e66b CM |
6815 | return 0; |
6816 | } | |
6817 | ||
5d4f98a2 YZ |
6818 | |
6819 | int btrfs_prepare_block_group_relocation(struct btrfs_root *root, | |
6820 | struct btrfs_block_group_cache *group) | |
6821 | ||
6822 | { | |
6823 | __alloc_chunk_for_shrink(root, group, 1); | |
6824 | set_block_group_readonly(group); | |
6825 | return 0; | |
6826 | } | |
6827 | ||
6828 | #if 0 | |
1a40e23b ZY |
6829 | static int __insert_orphan_inode(struct btrfs_trans_handle *trans, |
6830 | struct btrfs_root *root, | |
6831 | u64 objectid, u64 size) | |
6832 | { | |
6833 | struct btrfs_path *path; | |
6834 | struct btrfs_inode_item *item; | |
6835 | struct extent_buffer *leaf; | |
6836 | int ret; | |
6837 | ||
6838 | path = btrfs_alloc_path(); | |
6839 | if (!path) | |
6840 | return -ENOMEM; | |
6841 | ||
b9473439 | 6842 | path->leave_spinning = 1; |
1a40e23b ZY |
6843 | ret = btrfs_insert_empty_inode(trans, root, path, objectid); |
6844 | if (ret) | |
6845 | goto out; | |
6846 | ||
6847 | leaf = path->nodes[0]; | |
6848 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); | |
6849 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
6850 | btrfs_set_inode_generation(leaf, item, 1); | |
6851 | btrfs_set_inode_size(leaf, item, size); | |
6852 | btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); | |
0403e47e | 6853 | btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS); |
1a40e23b ZY |
6854 | btrfs_mark_buffer_dirty(leaf); |
6855 | btrfs_release_path(root, path); | |
6856 | out: | |
6857 | btrfs_free_path(path); | |
6858 | return ret; | |
6859 | } | |
6860 | ||
d397712b | 6861 | static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, |
1a40e23b ZY |
6862 | struct btrfs_block_group_cache *group) |
6863 | { | |
6864 | struct inode *inode = NULL; | |
6865 | struct btrfs_trans_handle *trans; | |
6866 | struct btrfs_root *root; | |
6867 | struct btrfs_key root_key; | |
6868 | u64 objectid = BTRFS_FIRST_FREE_OBJECTID; | |
6869 | int err = 0; | |
6870 | ||
6871 | root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
6872 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
6873 | root_key.offset = (u64)-1; | |
6874 | root = btrfs_read_fs_root_no_name(fs_info, &root_key); | |
6875 | if (IS_ERR(root)) | |
6876 | return ERR_CAST(root); | |
6877 | ||
6878 | trans = btrfs_start_transaction(root, 1); | |
6879 | BUG_ON(!trans); | |
6880 | ||
6881 | err = btrfs_find_free_objectid(trans, root, objectid, &objectid); | |
6882 | if (err) | |
6883 | goto out; | |
6884 | ||
6885 | err = __insert_orphan_inode(trans, root, objectid, group->key.offset); | |
6886 | BUG_ON(err); | |
6887 | ||
6888 | err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0, | |
c8b97818 CM |
6889 | group->key.offset, 0, group->key.offset, |
6890 | 0, 0, 0); | |
1a40e23b ZY |
6891 | BUG_ON(err); |
6892 | ||
6893 | inode = btrfs_iget_locked(root->fs_info->sb, objectid, root); | |
6894 | if (inode->i_state & I_NEW) { | |
6895 | BTRFS_I(inode)->root = root; | |
6896 | BTRFS_I(inode)->location.objectid = objectid; | |
6897 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
6898 | BTRFS_I(inode)->location.offset = 0; | |
6899 | btrfs_read_locked_inode(inode); | |
6900 | unlock_new_inode(inode); | |
6901 | BUG_ON(is_bad_inode(inode)); | |
6902 | } else { | |
6903 | BUG_ON(1); | |
6904 | } | |
17d217fe | 6905 | BTRFS_I(inode)->index_cnt = group->key.objectid; |
1a40e23b ZY |
6906 | |
6907 | err = btrfs_orphan_add(trans, inode); | |
6908 | out: | |
6909 | btrfs_end_transaction(trans, root); | |
6910 | if (err) { | |
6911 | if (inode) | |
6912 | iput(inode); | |
6913 | inode = ERR_PTR(err); | |
6914 | } | |
6915 | return inode; | |
6916 | } | |
6917 | ||
17d217fe YZ |
6918 | int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) |
6919 | { | |
6920 | ||
6921 | struct btrfs_ordered_sum *sums; | |
6922 | struct btrfs_sector_sum *sector_sum; | |
6923 | struct btrfs_ordered_extent *ordered; | |
6924 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6925 | struct list_head list; | |
6926 | size_t offset; | |
6927 | int ret; | |
6928 | u64 disk_bytenr; | |
6929 | ||
6930 | INIT_LIST_HEAD(&list); | |
6931 | ||
6932 | ordered = btrfs_lookup_ordered_extent(inode, file_pos); | |
6933 | BUG_ON(ordered->file_offset != file_pos || ordered->len != len); | |
6934 | ||
6935 | disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; | |
07d400a6 | 6936 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr, |
17d217fe YZ |
6937 | disk_bytenr + len - 1, &list); |
6938 | ||
6939 | while (!list_empty(&list)) { | |
6940 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
6941 | list_del_init(&sums->list); | |
6942 | ||
6943 | sector_sum = sums->sums; | |
6944 | sums->bytenr = ordered->start; | |
6945 | ||
6946 | offset = 0; | |
6947 | while (offset < sums->len) { | |
6948 | sector_sum->bytenr += ordered->start - disk_bytenr; | |
6949 | sector_sum++; | |
6950 | offset += root->sectorsize; | |
6951 | } | |
6952 | ||
6953 | btrfs_add_ordered_sum(inode, ordered, sums); | |
6954 | } | |
6955 | btrfs_put_ordered_extent(ordered); | |
6956 | return 0; | |
6957 | } | |
6958 | ||
1a40e23b | 6959 | int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) |
edbd8d4e CM |
6960 | { |
6961 | struct btrfs_trans_handle *trans; | |
edbd8d4e | 6962 | struct btrfs_path *path; |
1a40e23b ZY |
6963 | struct btrfs_fs_info *info = root->fs_info; |
6964 | struct extent_buffer *leaf; | |
6965 | struct inode *reloc_inode; | |
6966 | struct btrfs_block_group_cache *block_group; | |
6967 | struct btrfs_key key; | |
d899e052 | 6968 | u64 skipped; |
edbd8d4e CM |
6969 | u64 cur_byte; |
6970 | u64 total_found; | |
edbd8d4e CM |
6971 | u32 nritems; |
6972 | int ret; | |
a061fc8d | 6973 | int progress; |
1a40e23b | 6974 | int pass = 0; |
edbd8d4e | 6975 | |
1a40e23b ZY |
6976 | root = root->fs_info->extent_root; |
6977 | ||
6978 | block_group = btrfs_lookup_block_group(info, group_start); | |
6979 | BUG_ON(!block_group); | |
8f18cf13 | 6980 | |
d397712b | 6981 | printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n", |
1a40e23b ZY |
6982 | (unsigned long long)block_group->key.objectid, |
6983 | (unsigned long long)block_group->flags); | |
8f18cf13 | 6984 | |
edbd8d4e | 6985 | path = btrfs_alloc_path(); |
1a40e23b | 6986 | BUG_ON(!path); |
edbd8d4e | 6987 | |
1a40e23b ZY |
6988 | reloc_inode = create_reloc_inode(info, block_group); |
6989 | BUG_ON(IS_ERR(reloc_inode)); | |
323da79c | 6990 | |
1a40e23b | 6991 | __alloc_chunk_for_shrink(root, block_group, 1); |
c146afad | 6992 | set_block_group_readonly(block_group); |
323da79c | 6993 | |
1a40e23b ZY |
6994 | btrfs_start_delalloc_inodes(info->tree_root); |
6995 | btrfs_wait_ordered_extents(info->tree_root, 0); | |
6996 | again: | |
d899e052 | 6997 | skipped = 0; |
edbd8d4e | 6998 | total_found = 0; |
a061fc8d | 6999 | progress = 0; |
1a40e23b | 7000 | key.objectid = block_group->key.objectid; |
edbd8d4e CM |
7001 | key.offset = 0; |
7002 | key.type = 0; | |
73e48b27 | 7003 | cur_byte = key.objectid; |
4313b399 | 7004 | |
1a40e23b ZY |
7005 | trans = btrfs_start_transaction(info->tree_root, 1); |
7006 | btrfs_commit_transaction(trans, info->tree_root); | |
ea8c2819 | 7007 | |
1a40e23b ZY |
7008 | mutex_lock(&root->fs_info->cleaner_mutex); |
7009 | btrfs_clean_old_snapshots(info->tree_root); | |
7010 | btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); | |
7011 | mutex_unlock(&root->fs_info->cleaner_mutex); | |
ea8c2819 | 7012 | |
56bec294 CM |
7013 | trans = btrfs_start_transaction(info->tree_root, 1); |
7014 | btrfs_commit_transaction(trans, info->tree_root); | |
7015 | ||
d397712b | 7016 | while (1) { |
edbd8d4e CM |
7017 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
7018 | if (ret < 0) | |
7019 | goto out; | |
7d9eb12c | 7020 | next: |
edbd8d4e | 7021 | leaf = path->nodes[0]; |
73e48b27 | 7022 | nritems = btrfs_header_nritems(leaf); |
73e48b27 Y |
7023 | if (path->slots[0] >= nritems) { |
7024 | ret = btrfs_next_leaf(root, path); | |
7025 | if (ret < 0) | |
7026 | goto out; | |
7027 | if (ret == 1) { | |
7028 | ret = 0; | |
7029 | break; | |
edbd8d4e | 7030 | } |
73e48b27 Y |
7031 | leaf = path->nodes[0]; |
7032 | nritems = btrfs_header_nritems(leaf); | |
edbd8d4e | 7033 | } |
73e48b27 | 7034 | |
1a40e23b | 7035 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
725c8463 | 7036 | |
1a40e23b ZY |
7037 | if (key.objectid >= block_group->key.objectid + |
7038 | block_group->key.offset) | |
8f18cf13 CM |
7039 | break; |
7040 | ||
725c8463 | 7041 | if (progress && need_resched()) { |
725c8463 | 7042 | btrfs_release_path(root, path); |
1a40e23b | 7043 | cond_resched(); |
725c8463 | 7044 | progress = 0; |
1a40e23b | 7045 | continue; |
725c8463 CM |
7046 | } |
7047 | progress = 1; | |
7048 | ||
1a40e23b ZY |
7049 | if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY || |
7050 | key.objectid + key.offset <= cur_byte) { | |
edbd8d4e | 7051 | path->slots[0]++; |
edbd8d4e CM |
7052 | goto next; |
7053 | } | |
73e48b27 | 7054 | |
edbd8d4e | 7055 | total_found++; |
1a40e23b | 7056 | cur_byte = key.objectid + key.offset; |
edbd8d4e | 7057 | btrfs_release_path(root, path); |
edbd8d4e | 7058 | |
1a40e23b ZY |
7059 | __alloc_chunk_for_shrink(root, block_group, 0); |
7060 | ret = relocate_one_extent(root, path, &key, block_group, | |
7061 | reloc_inode, pass); | |
7062 | BUG_ON(ret < 0); | |
d899e052 YZ |
7063 | if (ret > 0) |
7064 | skipped++; | |
edbd8d4e | 7065 | |
1a40e23b ZY |
7066 | key.objectid = cur_byte; |
7067 | key.type = 0; | |
7068 | key.offset = 0; | |
edbd8d4e CM |
7069 | } |
7070 | ||
1a40e23b | 7071 | btrfs_release_path(root, path); |
7d9eb12c | 7072 | |
1a40e23b ZY |
7073 | if (pass == 0) { |
7074 | btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1); | |
7075 | invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1); | |
8e8a1e31 | 7076 | } |
73e48b27 | 7077 | |
1a40e23b | 7078 | if (total_found > 0) { |
d397712b | 7079 | printk(KERN_INFO "btrfs found %llu extents in pass %d\n", |
1a40e23b ZY |
7080 | (unsigned long long)total_found, pass); |
7081 | pass++; | |
d899e052 YZ |
7082 | if (total_found == skipped && pass > 2) { |
7083 | iput(reloc_inode); | |
7084 | reloc_inode = create_reloc_inode(info, block_group); | |
7085 | pass = 0; | |
7086 | } | |
1a40e23b | 7087 | goto again; |
0f9dd46c | 7088 | } |
0ef3e66b | 7089 | |
1a40e23b ZY |
7090 | /* delete reloc_inode */ |
7091 | iput(reloc_inode); | |
7092 | ||
7093 | /* unpin extents in this range */ | |
7094 | trans = btrfs_start_transaction(info->tree_root, 1); | |
7095 | btrfs_commit_transaction(trans, info->tree_root); | |
0ef3e66b | 7096 | |
1a40e23b ZY |
7097 | spin_lock(&block_group->lock); |
7098 | WARN_ON(block_group->pinned > 0); | |
7099 | WARN_ON(block_group->reserved > 0); | |
7100 | WARN_ON(btrfs_block_group_used(&block_group->item) > 0); | |
7101 | spin_unlock(&block_group->lock); | |
fa9c0d79 | 7102 | btrfs_put_block_group(block_group); |
1a40e23b | 7103 | ret = 0; |
edbd8d4e | 7104 | out: |
1a40e23b | 7105 | btrfs_free_path(path); |
edbd8d4e CM |
7106 | return ret; |
7107 | } | |
5d4f98a2 | 7108 | #endif |
edbd8d4e | 7109 | |
b2950863 CH |
7110 | static int find_first_block_group(struct btrfs_root *root, |
7111 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7112 | { |
925baedd | 7113 | int ret = 0; |
0b86a832 CM |
7114 | struct btrfs_key found_key; |
7115 | struct extent_buffer *leaf; | |
7116 | int slot; | |
edbd8d4e | 7117 | |
0b86a832 CM |
7118 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7119 | if (ret < 0) | |
925baedd CM |
7120 | goto out; |
7121 | ||
d397712b | 7122 | while (1) { |
0b86a832 | 7123 | slot = path->slots[0]; |
edbd8d4e | 7124 | leaf = path->nodes[0]; |
0b86a832 CM |
7125 | if (slot >= btrfs_header_nritems(leaf)) { |
7126 | ret = btrfs_next_leaf(root, path); | |
7127 | if (ret == 0) | |
7128 | continue; | |
7129 | if (ret < 0) | |
925baedd | 7130 | goto out; |
0b86a832 | 7131 | break; |
edbd8d4e | 7132 | } |
0b86a832 | 7133 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7134 | |
0b86a832 | 7135 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7136 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7137 | ret = 0; | |
7138 | goto out; | |
7139 | } | |
0b86a832 | 7140 | path->slots[0]++; |
edbd8d4e | 7141 | } |
0b86a832 | 7142 | ret = -ENOENT; |
925baedd | 7143 | out: |
0b86a832 | 7144 | return ret; |
edbd8d4e CM |
7145 | } |
7146 | ||
1a40e23b ZY |
7147 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7148 | { | |
7149 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7150 | struct btrfs_space_info *space_info; |
1a40e23b ZY |
7151 | struct rb_node *n; |
7152 | ||
1a40e23b ZY |
7153 | spin_lock(&info->block_group_cache_lock); |
7154 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7155 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7156 | cache_node); | |
1a40e23b ZY |
7157 | rb_erase(&block_group->cache_node, |
7158 | &info->block_group_cache_tree); | |
d899e052 YZ |
7159 | spin_unlock(&info->block_group_cache_lock); |
7160 | ||
80eb234a | 7161 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7162 | list_del(&block_group->list); |
80eb234a | 7163 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7164 | |
817d52f8 JB |
7165 | if (block_group->cached == BTRFS_CACHE_STARTED) |
7166 | wait_event(block_group->caching_q, | |
7167 | block_group_cache_done(block_group)); | |
7168 | ||
7169 | btrfs_remove_free_space_cache(block_group); | |
7170 | ||
d2fb3437 | 7171 | WARN_ON(atomic_read(&block_group->count) != 1); |
1a40e23b | 7172 | kfree(block_group); |
d899e052 YZ |
7173 | |
7174 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7175 | } |
7176 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7177 | |
7178 | /* now that all the block groups are freed, go through and | |
7179 | * free all the space_info structs. This is only called during | |
7180 | * the final stages of unmount, and so we know nobody is | |
7181 | * using them. We call synchronize_rcu() once before we start, | |
7182 | * just to be on the safe side. | |
7183 | */ | |
7184 | synchronize_rcu(); | |
7185 | ||
7186 | while(!list_empty(&info->space_info)) { | |
7187 | space_info = list_entry(info->space_info.next, | |
7188 | struct btrfs_space_info, | |
7189 | list); | |
7190 | ||
7191 | list_del(&space_info->list); | |
7192 | kfree(space_info); | |
7193 | } | |
1a40e23b ZY |
7194 | return 0; |
7195 | } | |
7196 | ||
9078a3e1 CM |
7197 | int btrfs_read_block_groups(struct btrfs_root *root) |
7198 | { | |
7199 | struct btrfs_path *path; | |
7200 | int ret; | |
9078a3e1 | 7201 | struct btrfs_block_group_cache *cache; |
be744175 | 7202 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7203 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7204 | struct btrfs_key key; |
7205 | struct btrfs_key found_key; | |
5f39d397 | 7206 | struct extent_buffer *leaf; |
96b5179d | 7207 | |
be744175 | 7208 | root = info->extent_root; |
9078a3e1 | 7209 | key.objectid = 0; |
0b86a832 | 7210 | key.offset = 0; |
9078a3e1 | 7211 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7212 | path = btrfs_alloc_path(); |
7213 | if (!path) | |
7214 | return -ENOMEM; | |
7215 | ||
d397712b | 7216 | while (1) { |
0b86a832 CM |
7217 | ret = find_first_block_group(root, path, &key); |
7218 | if (ret > 0) { | |
7219 | ret = 0; | |
7220 | goto error; | |
9078a3e1 | 7221 | } |
0b86a832 CM |
7222 | if (ret != 0) |
7223 | goto error; | |
7224 | ||
5f39d397 CM |
7225 | leaf = path->nodes[0]; |
7226 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7227 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7228 | if (!cache) { |
0b86a832 | 7229 | ret = -ENOMEM; |
9078a3e1 CM |
7230 | break; |
7231 | } | |
3e1ad54f | 7232 | |
d2fb3437 | 7233 | atomic_set(&cache->count, 1); |
c286ac48 | 7234 | spin_lock_init(&cache->lock); |
6226cb0a | 7235 | spin_lock_init(&cache->tree_lock); |
817d52f8 JB |
7236 | cache->fs_info = info; |
7237 | init_waitqueue_head(&cache->caching_q); | |
0f9dd46c | 7238 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7239 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 JB |
7240 | |
7241 | /* | |
7242 | * we only want to have 32k of ram per block group for keeping | |
7243 | * track of free space, and if we pass 1/2 of that we want to | |
7244 | * start converting things over to using bitmaps | |
7245 | */ | |
7246 | cache->extents_thresh = ((1024 * 32) / 2) / | |
7247 | sizeof(struct btrfs_free_space); | |
7248 | ||
5f39d397 CM |
7249 | read_extent_buffer(leaf, &cache->item, |
7250 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7251 | sizeof(cache->item)); | |
9078a3e1 | 7252 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7253 | |
9078a3e1 CM |
7254 | key.objectid = found_key.objectid + found_key.offset; |
7255 | btrfs_release_path(root, path); | |
0b86a832 | 7256 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7257 | cache->sectorsize = root->sectorsize; |
7258 | ||
7259 | remove_sb_from_cache(root, cache); | |
7260 | ||
7261 | /* | |
7262 | * check for two cases, either we are full, and therefore | |
7263 | * don't need to bother with the caching work since we won't | |
7264 | * find any space, or we are empty, and we can just add all | |
7265 | * the space in and be done with it. This saves us _alot_ of | |
7266 | * time, particularly in the full case. | |
7267 | */ | |
7268 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
7269 | cache->cached = BTRFS_CACHE_FINISHED; | |
7270 | } else if (btrfs_block_group_used(&cache->item) == 0) { | |
7271 | cache->cached = BTRFS_CACHE_FINISHED; | |
7272 | add_new_free_space(cache, root->fs_info, | |
7273 | found_key.objectid, | |
7274 | found_key.objectid + | |
7275 | found_key.offset); | |
7276 | } | |
96b5179d | 7277 | |
6324fbf3 CM |
7278 | ret = update_space_info(info, cache->flags, found_key.offset, |
7279 | btrfs_block_group_used(&cache->item), | |
7280 | &space_info); | |
7281 | BUG_ON(ret); | |
7282 | cache->space_info = space_info; | |
80eb234a JB |
7283 | down_write(&space_info->groups_sem); |
7284 | list_add_tail(&cache->list, &space_info->block_groups); | |
7285 | up_write(&space_info->groups_sem); | |
0f9dd46c JB |
7286 | |
7287 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
7288 | BUG_ON(ret); | |
75ccf47d CM |
7289 | |
7290 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c YZ |
7291 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
7292 | set_block_group_readonly(cache); | |
9078a3e1 | 7293 | } |
0b86a832 CM |
7294 | ret = 0; |
7295 | error: | |
9078a3e1 | 7296 | btrfs_free_path(path); |
0b86a832 | 7297 | return ret; |
9078a3e1 | 7298 | } |
6324fbf3 CM |
7299 | |
7300 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7301 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7302 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7303 | u64 size) |
7304 | { | |
7305 | int ret; | |
6324fbf3 CM |
7306 | struct btrfs_root *extent_root; |
7307 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7308 | |
7309 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7310 | |
12fcfd22 | 7311 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7312 | |
8f18cf13 | 7313 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7314 | if (!cache) |
7315 | return -ENOMEM; | |
7316 | ||
e17cade2 | 7317 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7318 | cache->key.offset = size; |
d2fb3437 | 7319 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 JB |
7320 | cache->sectorsize = root->sectorsize; |
7321 | ||
7322 | /* | |
7323 | * we only want to have 32k of ram per block group for keeping track | |
7324 | * of free space, and if we pass 1/2 of that we want to start | |
7325 | * converting things over to using bitmaps | |
7326 | */ | |
7327 | cache->extents_thresh = ((1024 * 32) / 2) / | |
7328 | sizeof(struct btrfs_free_space); | |
d2fb3437 | 7329 | atomic_set(&cache->count, 1); |
c286ac48 | 7330 | spin_lock_init(&cache->lock); |
6226cb0a | 7331 | spin_lock_init(&cache->tree_lock); |
817d52f8 | 7332 | init_waitqueue_head(&cache->caching_q); |
0f9dd46c | 7333 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7334 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7335 | |
6324fbf3 | 7336 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7337 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7338 | cache->flags = type; | |
7339 | btrfs_set_block_group_flags(&cache->item, type); | |
7340 | ||
817d52f8 | 7341 | cache->cached = BTRFS_CACHE_FINISHED; |
96303081 JB |
7342 | remove_sb_from_cache(root, cache); |
7343 | ||
817d52f8 JB |
7344 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7345 | chunk_offset + size); | |
7346 | ||
6324fbf3 CM |
7347 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7348 | &cache->space_info); | |
7349 | BUG_ON(ret); | |
80eb234a JB |
7350 | down_write(&cache->space_info->groups_sem); |
7351 | list_add_tail(&cache->list, &cache->space_info->block_groups); | |
7352 | up_write(&cache->space_info->groups_sem); | |
6324fbf3 | 7353 | |
0f9dd46c JB |
7354 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
7355 | BUG_ON(ret); | |
c286ac48 | 7356 | |
6324fbf3 CM |
7357 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7358 | sizeof(cache->item)); | |
7359 | BUG_ON(ret); | |
7360 | ||
d18a2c44 | 7361 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7362 | |
6324fbf3 CM |
7363 | return 0; |
7364 | } | |
1a40e23b ZY |
7365 | |
7366 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
7367 | struct btrfs_root *root, u64 group_start) | |
7368 | { | |
7369 | struct btrfs_path *path; | |
7370 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7371 | struct btrfs_free_cluster *cluster; |
1a40e23b ZY |
7372 | struct btrfs_key key; |
7373 | int ret; | |
7374 | ||
1a40e23b ZY |
7375 | root = root->fs_info->extent_root; |
7376 | ||
7377 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7378 | BUG_ON(!block_group); | |
c146afad | 7379 | BUG_ON(!block_group->ro); |
1a40e23b ZY |
7380 | |
7381 | memcpy(&key, &block_group->key, sizeof(key)); | |
7382 | ||
44fb5511 CM |
7383 | /* make sure this block group isn't part of an allocation cluster */ |
7384 | cluster = &root->fs_info->data_alloc_cluster; | |
7385 | spin_lock(&cluster->refill_lock); | |
7386 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7387 | spin_unlock(&cluster->refill_lock); | |
7388 | ||
7389 | /* | |
7390 | * make sure this block group isn't part of a metadata | |
7391 | * allocation cluster | |
7392 | */ | |
7393 | cluster = &root->fs_info->meta_alloc_cluster; | |
7394 | spin_lock(&cluster->refill_lock); | |
7395 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7396 | spin_unlock(&cluster->refill_lock); | |
7397 | ||
1a40e23b ZY |
7398 | path = btrfs_alloc_path(); |
7399 | BUG_ON(!path); | |
7400 | ||
3dfdb934 | 7401 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
7402 | rb_erase(&block_group->cache_node, |
7403 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 7404 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 7405 | |
80eb234a | 7406 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
7407 | /* |
7408 | * we must use list_del_init so people can check to see if they | |
7409 | * are still on the list after taking the semaphore | |
7410 | */ | |
7411 | list_del_init(&block_group->list); | |
80eb234a | 7412 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 7413 | |
817d52f8 JB |
7414 | if (block_group->cached == BTRFS_CACHE_STARTED) |
7415 | wait_event(block_group->caching_q, | |
7416 | block_group_cache_done(block_group)); | |
7417 | ||
7418 | btrfs_remove_free_space_cache(block_group); | |
7419 | ||
c146afad YZ |
7420 | spin_lock(&block_group->space_info->lock); |
7421 | block_group->space_info->total_bytes -= block_group->key.offset; | |
7422 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
7423 | spin_unlock(&block_group->space_info->lock); | |
283bb197 CM |
7424 | |
7425 | btrfs_clear_space_info_full(root->fs_info); | |
c146afad | 7426 | |
fa9c0d79 CM |
7427 | btrfs_put_block_group(block_group); |
7428 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
7429 | |
7430 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
7431 | if (ret > 0) | |
7432 | ret = -EIO; | |
7433 | if (ret < 0) | |
7434 | goto out; | |
7435 | ||
7436 | ret = btrfs_del_item(trans, root, path); | |
7437 | out: | |
7438 | btrfs_free_path(path); | |
7439 | return ret; | |
7440 | } |