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> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
4b4e25f2 | 26 | #include "compat.h" |
74493f7a | 27 | #include "hash.h" |
fec577fb CM |
28 | #include "ctree.h" |
29 | #include "disk-io.h" | |
30 | #include "print-tree.h" | |
e089f05c | 31 | #include "transaction.h" |
0b86a832 | 32 | #include "volumes.h" |
925baedd | 33 | #include "locking.h" |
fa9c0d79 | 34 | #include "free-space-cache.h" |
fec577fb | 35 | |
0e4f8f88 CM |
36 | /* control flags for do_chunk_alloc's force field |
37 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk | |
38 | * if we really need one. | |
39 | * | |
40 | * CHUNK_ALLOC_FORCE means it must try to allocate one | |
41 | * | |
42 | * CHUNK_ALLOC_LIMITED means to only try and allocate one | |
43 | * if we have very few chunks already allocated. This is | |
44 | * used as part of the clustering code to help make sure | |
45 | * we have a good pool of storage to cluster in, without | |
46 | * filling the FS with empty chunks | |
47 | * | |
48 | */ | |
49 | enum { | |
50 | CHUNK_ALLOC_NO_FORCE = 0, | |
51 | CHUNK_ALLOC_FORCE = 1, | |
52 | CHUNK_ALLOC_LIMITED = 2, | |
53 | }; | |
54 | ||
f3465ca4 JB |
55 | static int update_block_group(struct btrfs_trans_handle *trans, |
56 | struct btrfs_root *root, | |
f0486c68 | 57 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
58 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
59 | struct btrfs_root *root, | |
60 | u64 bytenr, u64 num_bytes, u64 parent, | |
61 | u64 root_objectid, u64 owner_objectid, | |
62 | u64 owner_offset, int refs_to_drop, | |
63 | struct btrfs_delayed_extent_op *extra_op); | |
64 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
65 | struct extent_buffer *leaf, | |
66 | struct btrfs_extent_item *ei); | |
67 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
68 | struct btrfs_root *root, | |
69 | u64 parent, u64 root_objectid, | |
70 | u64 flags, u64 owner, u64 offset, | |
71 | struct btrfs_key *ins, int ref_mod); | |
72 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
73 | struct btrfs_root *root, | |
74 | u64 parent, u64 root_objectid, | |
75 | u64 flags, struct btrfs_disk_key *key, | |
76 | int level, struct btrfs_key *ins); | |
6a63209f JB |
77 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
78 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
79 | u64 flags, int force); | |
11833d66 YZ |
80 | static int find_next_key(struct btrfs_path *path, int level, |
81 | struct btrfs_key *key); | |
9ed74f2d JB |
82 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
83 | int dump_block_groups); | |
6a63209f | 84 | |
817d52f8 JB |
85 | static noinline int |
86 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
87 | { | |
88 | smp_mb(); | |
89 | return cache->cached == BTRFS_CACHE_FINISHED; | |
90 | } | |
91 | ||
0f9dd46c JB |
92 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
93 | { | |
94 | return (cache->flags & bits) == bits; | |
95 | } | |
96 | ||
62a45b60 | 97 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
98 | { |
99 | atomic_inc(&cache->count); | |
100 | } | |
101 | ||
102 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
103 | { | |
f0486c68 YZ |
104 | if (atomic_dec_and_test(&cache->count)) { |
105 | WARN_ON(cache->pinned > 0); | |
106 | WARN_ON(cache->reserved > 0); | |
107 | WARN_ON(cache->reserved_pinned > 0); | |
11dfe35a | 108 | kfree(cache); |
f0486c68 | 109 | } |
11dfe35a JB |
110 | } |
111 | ||
0f9dd46c JB |
112 | /* |
113 | * this adds the block group to the fs_info rb tree for the block group | |
114 | * cache | |
115 | */ | |
b2950863 | 116 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
117 | struct btrfs_block_group_cache *block_group) |
118 | { | |
119 | struct rb_node **p; | |
120 | struct rb_node *parent = NULL; | |
121 | struct btrfs_block_group_cache *cache; | |
122 | ||
123 | spin_lock(&info->block_group_cache_lock); | |
124 | p = &info->block_group_cache_tree.rb_node; | |
125 | ||
126 | while (*p) { | |
127 | parent = *p; | |
128 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
129 | cache_node); | |
130 | if (block_group->key.objectid < cache->key.objectid) { | |
131 | p = &(*p)->rb_left; | |
132 | } else if (block_group->key.objectid > cache->key.objectid) { | |
133 | p = &(*p)->rb_right; | |
134 | } else { | |
135 | spin_unlock(&info->block_group_cache_lock); | |
136 | return -EEXIST; | |
137 | } | |
138 | } | |
139 | ||
140 | rb_link_node(&block_group->cache_node, parent, p); | |
141 | rb_insert_color(&block_group->cache_node, | |
142 | &info->block_group_cache_tree); | |
143 | spin_unlock(&info->block_group_cache_lock); | |
144 | ||
145 | return 0; | |
146 | } | |
147 | ||
148 | /* | |
149 | * This will return the block group at or after bytenr if contains is 0, else | |
150 | * it will return the block group that contains the bytenr | |
151 | */ | |
152 | static struct btrfs_block_group_cache * | |
153 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
154 | int contains) | |
155 | { | |
156 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
157 | struct rb_node *n; | |
158 | u64 end, start; | |
159 | ||
160 | spin_lock(&info->block_group_cache_lock); | |
161 | n = info->block_group_cache_tree.rb_node; | |
162 | ||
163 | while (n) { | |
164 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
165 | cache_node); | |
166 | end = cache->key.objectid + cache->key.offset - 1; | |
167 | start = cache->key.objectid; | |
168 | ||
169 | if (bytenr < start) { | |
170 | if (!contains && (!ret || start < ret->key.objectid)) | |
171 | ret = cache; | |
172 | n = n->rb_left; | |
173 | } else if (bytenr > start) { | |
174 | if (contains && bytenr <= end) { | |
175 | ret = cache; | |
176 | break; | |
177 | } | |
178 | n = n->rb_right; | |
179 | } else { | |
180 | ret = cache; | |
181 | break; | |
182 | } | |
183 | } | |
d2fb3437 | 184 | if (ret) |
11dfe35a | 185 | btrfs_get_block_group(ret); |
0f9dd46c JB |
186 | spin_unlock(&info->block_group_cache_lock); |
187 | ||
188 | return ret; | |
189 | } | |
190 | ||
11833d66 YZ |
191 | static int add_excluded_extent(struct btrfs_root *root, |
192 | u64 start, u64 num_bytes) | |
817d52f8 | 193 | { |
11833d66 YZ |
194 | u64 end = start + num_bytes - 1; |
195 | set_extent_bits(&root->fs_info->freed_extents[0], | |
196 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
197 | set_extent_bits(&root->fs_info->freed_extents[1], | |
198 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
199 | return 0; | |
200 | } | |
817d52f8 | 201 | |
11833d66 YZ |
202 | static void free_excluded_extents(struct btrfs_root *root, |
203 | struct btrfs_block_group_cache *cache) | |
204 | { | |
205 | u64 start, end; | |
817d52f8 | 206 | |
11833d66 YZ |
207 | start = cache->key.objectid; |
208 | end = start + cache->key.offset - 1; | |
209 | ||
210 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
211 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
212 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
213 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
214 | } |
215 | ||
11833d66 YZ |
216 | static int exclude_super_stripes(struct btrfs_root *root, |
217 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 218 | { |
817d52f8 JB |
219 | u64 bytenr; |
220 | u64 *logical; | |
221 | int stripe_len; | |
222 | int i, nr, ret; | |
223 | ||
06b2331f YZ |
224 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
225 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
226 | cache->bytes_super += stripe_len; | |
227 | ret = add_excluded_extent(root, cache->key.objectid, | |
228 | stripe_len); | |
229 | BUG_ON(ret); | |
230 | } | |
231 | ||
817d52f8 JB |
232 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
233 | bytenr = btrfs_sb_offset(i); | |
234 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
235 | cache->key.objectid, bytenr, | |
236 | 0, &logical, &nr, &stripe_len); | |
237 | BUG_ON(ret); | |
11833d66 | 238 | |
817d52f8 | 239 | while (nr--) { |
1b2da372 | 240 | cache->bytes_super += stripe_len; |
11833d66 YZ |
241 | ret = add_excluded_extent(root, logical[nr], |
242 | stripe_len); | |
243 | BUG_ON(ret); | |
817d52f8 | 244 | } |
11833d66 | 245 | |
817d52f8 JB |
246 | kfree(logical); |
247 | } | |
817d52f8 JB |
248 | return 0; |
249 | } | |
250 | ||
11833d66 YZ |
251 | static struct btrfs_caching_control * |
252 | get_caching_control(struct btrfs_block_group_cache *cache) | |
253 | { | |
254 | struct btrfs_caching_control *ctl; | |
255 | ||
256 | spin_lock(&cache->lock); | |
257 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
258 | spin_unlock(&cache->lock); | |
259 | return NULL; | |
260 | } | |
261 | ||
dde5abee JB |
262 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
263 | if (!cache->caching_ctl) { | |
264 | spin_unlock(&cache->lock); | |
11833d66 YZ |
265 | return NULL; |
266 | } | |
267 | ||
268 | ctl = cache->caching_ctl; | |
269 | atomic_inc(&ctl->count); | |
270 | spin_unlock(&cache->lock); | |
271 | return ctl; | |
272 | } | |
273 | ||
274 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
275 | { | |
276 | if (atomic_dec_and_test(&ctl->count)) | |
277 | kfree(ctl); | |
278 | } | |
279 | ||
0f9dd46c JB |
280 | /* |
281 | * this is only called by cache_block_group, since we could have freed extents | |
282 | * we need to check the pinned_extents for any extents that can't be used yet | |
283 | * since their free space will be released as soon as the transaction commits. | |
284 | */ | |
817d52f8 | 285 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
286 | struct btrfs_fs_info *info, u64 start, u64 end) |
287 | { | |
817d52f8 | 288 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
289 | int ret; |
290 | ||
291 | while (start < end) { | |
11833d66 | 292 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 293 | &extent_start, &extent_end, |
11833d66 | 294 | EXTENT_DIRTY | EXTENT_UPTODATE); |
0f9dd46c JB |
295 | if (ret) |
296 | break; | |
297 | ||
06b2331f | 298 | if (extent_start <= start) { |
0f9dd46c JB |
299 | start = extent_end + 1; |
300 | } else if (extent_start > start && extent_start < end) { | |
301 | size = extent_start - start; | |
817d52f8 | 302 | total_added += size; |
ea6a478e JB |
303 | ret = btrfs_add_free_space(block_group, start, |
304 | size); | |
0f9dd46c JB |
305 | BUG_ON(ret); |
306 | start = extent_end + 1; | |
307 | } else { | |
308 | break; | |
309 | } | |
310 | } | |
311 | ||
312 | if (start < end) { | |
313 | size = end - start; | |
817d52f8 | 314 | total_added += size; |
ea6a478e | 315 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
316 | BUG_ON(ret); |
317 | } | |
318 | ||
817d52f8 | 319 | return total_added; |
0f9dd46c JB |
320 | } |
321 | ||
817d52f8 | 322 | static int caching_kthread(void *data) |
e37c9e69 | 323 | { |
817d52f8 JB |
324 | struct btrfs_block_group_cache *block_group = data; |
325 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
11833d66 YZ |
326 | struct btrfs_caching_control *caching_ctl = block_group->caching_ctl; |
327 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 328 | struct btrfs_path *path; |
5f39d397 | 329 | struct extent_buffer *leaf; |
11833d66 | 330 | struct btrfs_key key; |
817d52f8 | 331 | u64 total_found = 0; |
11833d66 YZ |
332 | u64 last = 0; |
333 | u32 nritems; | |
334 | int ret = 0; | |
f510cfec | 335 | |
e37c9e69 CM |
336 | path = btrfs_alloc_path(); |
337 | if (!path) | |
338 | return -ENOMEM; | |
7d7d6068 | 339 | |
817d52f8 | 340 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 341 | |
5cd57b2c | 342 | /* |
817d52f8 JB |
343 | * We don't want to deadlock with somebody trying to allocate a new |
344 | * extent for the extent root while also trying to search the extent | |
345 | * root to add free space. So we skip locking and search the commit | |
346 | * root, since its read-only | |
5cd57b2c CM |
347 | */ |
348 | path->skip_locking = 1; | |
817d52f8 JB |
349 | path->search_commit_root = 1; |
350 | path->reada = 2; | |
351 | ||
e4404d6e | 352 | key.objectid = last; |
e37c9e69 | 353 | key.offset = 0; |
11833d66 | 354 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 355 | again: |
11833d66 | 356 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
357 | /* need to make sure the commit_root doesn't disappear */ |
358 | down_read(&fs_info->extent_commit_sem); | |
359 | ||
11833d66 | 360 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 361 | if (ret < 0) |
ef8bbdfe | 362 | goto err; |
a512bbf8 | 363 | |
11833d66 YZ |
364 | leaf = path->nodes[0]; |
365 | nritems = btrfs_header_nritems(leaf); | |
366 | ||
d397712b | 367 | while (1) { |
817d52f8 | 368 | smp_mb(); |
11833d66 | 369 | if (fs_info->closing > 1) { |
f25784b3 | 370 | last = (u64)-1; |
817d52f8 | 371 | break; |
f25784b3 | 372 | } |
817d52f8 | 373 | |
11833d66 YZ |
374 | if (path->slots[0] < nritems) { |
375 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
376 | } else { | |
377 | ret = find_next_key(path, 0, &key); | |
378 | if (ret) | |
e37c9e69 | 379 | break; |
817d52f8 | 380 | |
11833d66 | 381 | caching_ctl->progress = last; |
b3b4aa74 | 382 | btrfs_release_path(path); |
11833d66 YZ |
383 | up_read(&fs_info->extent_commit_sem); |
384 | mutex_unlock(&caching_ctl->mutex); | |
385 | if (btrfs_transaction_in_commit(fs_info)) | |
f36f3042 | 386 | schedule_timeout(1); |
11833d66 YZ |
387 | else |
388 | cond_resched(); | |
389 | goto again; | |
390 | } | |
817d52f8 | 391 | |
11833d66 YZ |
392 | if (key.objectid < block_group->key.objectid) { |
393 | path->slots[0]++; | |
817d52f8 | 394 | continue; |
e37c9e69 | 395 | } |
0f9dd46c | 396 | |
e37c9e69 | 397 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 398 | block_group->key.offset) |
e37c9e69 | 399 | break; |
7d7d6068 | 400 | |
11833d66 | 401 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
402 | total_found += add_new_free_space(block_group, |
403 | fs_info, last, | |
404 | key.objectid); | |
7d7d6068 | 405 | last = key.objectid + key.offset; |
817d52f8 | 406 | |
11833d66 YZ |
407 | if (total_found > (1024 * 1024 * 2)) { |
408 | total_found = 0; | |
409 | wake_up(&caching_ctl->wait); | |
410 | } | |
817d52f8 | 411 | } |
e37c9e69 CM |
412 | path->slots[0]++; |
413 | } | |
817d52f8 | 414 | ret = 0; |
e37c9e69 | 415 | |
817d52f8 JB |
416 | total_found += add_new_free_space(block_group, fs_info, last, |
417 | block_group->key.objectid + | |
418 | block_group->key.offset); | |
11833d66 | 419 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
420 | |
421 | spin_lock(&block_group->lock); | |
11833d66 | 422 | block_group->caching_ctl = NULL; |
817d52f8 JB |
423 | block_group->cached = BTRFS_CACHE_FINISHED; |
424 | spin_unlock(&block_group->lock); | |
0f9dd46c | 425 | |
54aa1f4d | 426 | err: |
e37c9e69 | 427 | btrfs_free_path(path); |
276e680d | 428 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 429 | |
11833d66 YZ |
430 | free_excluded_extents(extent_root, block_group); |
431 | ||
432 | mutex_unlock(&caching_ctl->mutex); | |
433 | wake_up(&caching_ctl->wait); | |
434 | ||
435 | put_caching_control(caching_ctl); | |
436 | atomic_dec(&block_group->space_info->caching_threads); | |
11dfe35a JB |
437 | btrfs_put_block_group(block_group); |
438 | ||
817d52f8 JB |
439 | return 0; |
440 | } | |
441 | ||
9d66e233 JB |
442 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
443 | struct btrfs_trans_handle *trans, | |
b8399dee | 444 | struct btrfs_root *root, |
9d66e233 | 445 | int load_cache_only) |
817d52f8 | 446 | { |
11833d66 YZ |
447 | struct btrfs_fs_info *fs_info = cache->fs_info; |
448 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
449 | struct task_struct *tsk; |
450 | int ret = 0; | |
451 | ||
11833d66 YZ |
452 | smp_mb(); |
453 | if (cache->cached != BTRFS_CACHE_NO) | |
454 | return 0; | |
455 | ||
9d66e233 JB |
456 | /* |
457 | * We can't do the read from on-disk cache during a commit since we need | |
b8399dee JB |
458 | * to have the normal tree locking. Also if we are currently trying to |
459 | * allocate blocks for the tree root we can't do the fast caching since | |
460 | * we likely hold important locks. | |
9d66e233 | 461 | */ |
f7039b1d | 462 | if (trans && (!trans->transaction->in_commit) && |
b8399dee | 463 | (root && root != root->fs_info->tree_root)) { |
9d66e233 JB |
464 | spin_lock(&cache->lock); |
465 | if (cache->cached != BTRFS_CACHE_NO) { | |
466 | spin_unlock(&cache->lock); | |
467 | return 0; | |
468 | } | |
469 | cache->cached = BTRFS_CACHE_STARTED; | |
470 | spin_unlock(&cache->lock); | |
471 | ||
472 | ret = load_free_space_cache(fs_info, cache); | |
473 | ||
474 | spin_lock(&cache->lock); | |
475 | if (ret == 1) { | |
476 | cache->cached = BTRFS_CACHE_FINISHED; | |
477 | cache->last_byte_to_unpin = (u64)-1; | |
478 | } else { | |
479 | cache->cached = BTRFS_CACHE_NO; | |
480 | } | |
481 | spin_unlock(&cache->lock); | |
3c14874a JB |
482 | if (ret == 1) { |
483 | free_excluded_extents(fs_info->extent_root, cache); | |
9d66e233 | 484 | return 0; |
3c14874a | 485 | } |
9d66e233 JB |
486 | } |
487 | ||
488 | if (load_cache_only) | |
489 | return 0; | |
490 | ||
fc0e4a31 | 491 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
11833d66 YZ |
492 | BUG_ON(!caching_ctl); |
493 | ||
494 | INIT_LIST_HEAD(&caching_ctl->list); | |
495 | mutex_init(&caching_ctl->mutex); | |
496 | init_waitqueue_head(&caching_ctl->wait); | |
497 | caching_ctl->block_group = cache; | |
498 | caching_ctl->progress = cache->key.objectid; | |
499 | /* one for caching kthread, one for caching block group list */ | |
500 | atomic_set(&caching_ctl->count, 2); | |
501 | ||
817d52f8 JB |
502 | spin_lock(&cache->lock); |
503 | if (cache->cached != BTRFS_CACHE_NO) { | |
504 | spin_unlock(&cache->lock); | |
11833d66 YZ |
505 | kfree(caching_ctl); |
506 | return 0; | |
817d52f8 | 507 | } |
11833d66 | 508 | cache->caching_ctl = caching_ctl; |
817d52f8 JB |
509 | cache->cached = BTRFS_CACHE_STARTED; |
510 | spin_unlock(&cache->lock); | |
511 | ||
11833d66 YZ |
512 | down_write(&fs_info->extent_commit_sem); |
513 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
514 | up_write(&fs_info->extent_commit_sem); | |
515 | ||
516 | atomic_inc(&cache->space_info->caching_threads); | |
11dfe35a | 517 | btrfs_get_block_group(cache); |
11833d66 | 518 | |
817d52f8 JB |
519 | tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n", |
520 | cache->key.objectid); | |
521 | if (IS_ERR(tsk)) { | |
522 | ret = PTR_ERR(tsk); | |
523 | printk(KERN_ERR "error running thread %d\n", ret); | |
524 | BUG(); | |
525 | } | |
526 | ||
ef8bbdfe | 527 | return ret; |
e37c9e69 CM |
528 | } |
529 | ||
0f9dd46c JB |
530 | /* |
531 | * return the block group that starts at or after bytenr | |
532 | */ | |
d397712b CM |
533 | static struct btrfs_block_group_cache * |
534 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 535 | { |
0f9dd46c | 536 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 537 | |
0f9dd46c | 538 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 539 | |
0f9dd46c | 540 | return cache; |
0ef3e66b CM |
541 | } |
542 | ||
0f9dd46c | 543 | /* |
9f55684c | 544 | * return the block group that contains the given bytenr |
0f9dd46c | 545 | */ |
d397712b CM |
546 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
547 | struct btrfs_fs_info *info, | |
548 | u64 bytenr) | |
be744175 | 549 | { |
0f9dd46c | 550 | struct btrfs_block_group_cache *cache; |
be744175 | 551 | |
0f9dd46c | 552 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 553 | |
0f9dd46c | 554 | return cache; |
be744175 | 555 | } |
0b86a832 | 556 | |
0f9dd46c JB |
557 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
558 | u64 flags) | |
6324fbf3 | 559 | { |
0f9dd46c | 560 | struct list_head *head = &info->space_info; |
0f9dd46c | 561 | struct btrfs_space_info *found; |
4184ea7f | 562 | |
b742bb82 YZ |
563 | flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM | |
564 | BTRFS_BLOCK_GROUP_METADATA; | |
565 | ||
4184ea7f CM |
566 | rcu_read_lock(); |
567 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 568 | if (found->flags & flags) { |
4184ea7f | 569 | rcu_read_unlock(); |
0f9dd46c | 570 | return found; |
4184ea7f | 571 | } |
0f9dd46c | 572 | } |
4184ea7f | 573 | rcu_read_unlock(); |
0f9dd46c | 574 | return NULL; |
6324fbf3 CM |
575 | } |
576 | ||
4184ea7f CM |
577 | /* |
578 | * after adding space to the filesystem, we need to clear the full flags | |
579 | * on all the space infos. | |
580 | */ | |
581 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
582 | { | |
583 | struct list_head *head = &info->space_info; | |
584 | struct btrfs_space_info *found; | |
585 | ||
586 | rcu_read_lock(); | |
587 | list_for_each_entry_rcu(found, head, list) | |
588 | found->full = 0; | |
589 | rcu_read_unlock(); | |
590 | } | |
591 | ||
80eb234a JB |
592 | static u64 div_factor(u64 num, int factor) |
593 | { | |
594 | if (factor == 10) | |
595 | return num; | |
596 | num *= factor; | |
597 | do_div(num, 10); | |
598 | return num; | |
599 | } | |
600 | ||
e5bc2458 CM |
601 | static u64 div_factor_fine(u64 num, int factor) |
602 | { | |
603 | if (factor == 100) | |
604 | return num; | |
605 | num *= factor; | |
606 | do_div(num, 100); | |
607 | return num; | |
608 | } | |
609 | ||
d2fb3437 YZ |
610 | u64 btrfs_find_block_group(struct btrfs_root *root, |
611 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 612 | { |
96b5179d | 613 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 614 | u64 used; |
d2fb3437 YZ |
615 | u64 last = max(search_hint, search_start); |
616 | u64 group_start = 0; | |
31f3c99b | 617 | int full_search = 0; |
d2fb3437 | 618 | int factor = 9; |
0ef3e66b | 619 | int wrapped = 0; |
31f3c99b | 620 | again: |
e8569813 ZY |
621 | while (1) { |
622 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
623 | if (!cache) |
624 | break; | |
96b5179d | 625 | |
c286ac48 | 626 | spin_lock(&cache->lock); |
96b5179d CM |
627 | last = cache->key.objectid + cache->key.offset; |
628 | used = btrfs_block_group_used(&cache->item); | |
629 | ||
d2fb3437 YZ |
630 | if ((full_search || !cache->ro) && |
631 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 632 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
633 | div_factor(cache->key.offset, factor)) { |
634 | group_start = cache->key.objectid; | |
c286ac48 | 635 | spin_unlock(&cache->lock); |
fa9c0d79 | 636 | btrfs_put_block_group(cache); |
8790d502 CM |
637 | goto found; |
638 | } | |
6324fbf3 | 639 | } |
c286ac48 | 640 | spin_unlock(&cache->lock); |
fa9c0d79 | 641 | btrfs_put_block_group(cache); |
de428b63 | 642 | cond_resched(); |
cd1bc465 | 643 | } |
0ef3e66b CM |
644 | if (!wrapped) { |
645 | last = search_start; | |
646 | wrapped = 1; | |
647 | goto again; | |
648 | } | |
649 | if (!full_search && factor < 10) { | |
be744175 | 650 | last = search_start; |
31f3c99b | 651 | full_search = 1; |
0ef3e66b | 652 | factor = 10; |
31f3c99b CM |
653 | goto again; |
654 | } | |
be744175 | 655 | found: |
d2fb3437 | 656 | return group_start; |
925baedd | 657 | } |
0f9dd46c | 658 | |
e02119d5 | 659 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 660 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
661 | { |
662 | int ret; | |
663 | struct btrfs_key key; | |
31840ae1 | 664 | struct btrfs_path *path; |
e02119d5 | 665 | |
31840ae1 ZY |
666 | path = btrfs_alloc_path(); |
667 | BUG_ON(!path); | |
e02119d5 CM |
668 | key.objectid = start; |
669 | key.offset = len; | |
670 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
671 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
672 | 0, 0); | |
31840ae1 | 673 | btrfs_free_path(path); |
7bb86316 CM |
674 | return ret; |
675 | } | |
676 | ||
a22285a6 YZ |
677 | /* |
678 | * helper function to lookup reference count and flags of extent. | |
679 | * | |
680 | * the head node for delayed ref is used to store the sum of all the | |
681 | * reference count modifications queued up in the rbtree. the head | |
682 | * node may also store the extent flags to set. This way you can check | |
683 | * to see what the reference count and extent flags would be if all of | |
684 | * the delayed refs are not processed. | |
685 | */ | |
686 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
687 | struct btrfs_root *root, u64 bytenr, | |
688 | u64 num_bytes, u64 *refs, u64 *flags) | |
689 | { | |
690 | struct btrfs_delayed_ref_head *head; | |
691 | struct btrfs_delayed_ref_root *delayed_refs; | |
692 | struct btrfs_path *path; | |
693 | struct btrfs_extent_item *ei; | |
694 | struct extent_buffer *leaf; | |
695 | struct btrfs_key key; | |
696 | u32 item_size; | |
697 | u64 num_refs; | |
698 | u64 extent_flags; | |
699 | int ret; | |
700 | ||
701 | path = btrfs_alloc_path(); | |
702 | if (!path) | |
703 | return -ENOMEM; | |
704 | ||
705 | key.objectid = bytenr; | |
706 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
707 | key.offset = num_bytes; | |
708 | if (!trans) { | |
709 | path->skip_locking = 1; | |
710 | path->search_commit_root = 1; | |
711 | } | |
712 | again: | |
713 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
714 | &key, path, 0, 0); | |
715 | if (ret < 0) | |
716 | goto out_free; | |
717 | ||
718 | if (ret == 0) { | |
719 | leaf = path->nodes[0]; | |
720 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
721 | if (item_size >= sizeof(*ei)) { | |
722 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
723 | struct btrfs_extent_item); | |
724 | num_refs = btrfs_extent_refs(leaf, ei); | |
725 | extent_flags = btrfs_extent_flags(leaf, ei); | |
726 | } else { | |
727 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
728 | struct btrfs_extent_item_v0 *ei0; | |
729 | BUG_ON(item_size != sizeof(*ei0)); | |
730 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
731 | struct btrfs_extent_item_v0); | |
732 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
733 | /* FIXME: this isn't correct for data */ | |
734 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
735 | #else | |
736 | BUG(); | |
737 | #endif | |
738 | } | |
739 | BUG_ON(num_refs == 0); | |
740 | } else { | |
741 | num_refs = 0; | |
742 | extent_flags = 0; | |
743 | ret = 0; | |
744 | } | |
745 | ||
746 | if (!trans) | |
747 | goto out; | |
748 | ||
749 | delayed_refs = &trans->transaction->delayed_refs; | |
750 | spin_lock(&delayed_refs->lock); | |
751 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
752 | if (head) { | |
753 | if (!mutex_trylock(&head->mutex)) { | |
754 | atomic_inc(&head->node.refs); | |
755 | spin_unlock(&delayed_refs->lock); | |
756 | ||
b3b4aa74 | 757 | btrfs_release_path(path); |
a22285a6 | 758 | |
8cc33e5c DS |
759 | /* |
760 | * Mutex was contended, block until it's released and try | |
761 | * again | |
762 | */ | |
a22285a6 YZ |
763 | mutex_lock(&head->mutex); |
764 | mutex_unlock(&head->mutex); | |
765 | btrfs_put_delayed_ref(&head->node); | |
766 | goto again; | |
767 | } | |
768 | if (head->extent_op && head->extent_op->update_flags) | |
769 | extent_flags |= head->extent_op->flags_to_set; | |
770 | else | |
771 | BUG_ON(num_refs == 0); | |
772 | ||
773 | num_refs += head->node.ref_mod; | |
774 | mutex_unlock(&head->mutex); | |
775 | } | |
776 | spin_unlock(&delayed_refs->lock); | |
777 | out: | |
778 | WARN_ON(num_refs == 0); | |
779 | if (refs) | |
780 | *refs = num_refs; | |
781 | if (flags) | |
782 | *flags = extent_flags; | |
783 | out_free: | |
784 | btrfs_free_path(path); | |
785 | return ret; | |
786 | } | |
787 | ||
d8d5f3e1 CM |
788 | /* |
789 | * Back reference rules. Back refs have three main goals: | |
790 | * | |
791 | * 1) differentiate between all holders of references to an extent so that | |
792 | * when a reference is dropped we can make sure it was a valid reference | |
793 | * before freeing the extent. | |
794 | * | |
795 | * 2) Provide enough information to quickly find the holders of an extent | |
796 | * if we notice a given block is corrupted or bad. | |
797 | * | |
798 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
799 | * maintenance. This is actually the same as #2, but with a slightly | |
800 | * different use case. | |
801 | * | |
5d4f98a2 YZ |
802 | * There are two kinds of back refs. The implicit back refs is optimized |
803 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
804 | * back refs of this kind provide information about the block's owner tree | |
805 | * and the pointer's key. These information allow us to find the block by | |
806 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
807 | * referenced by their owner trees. The location of tree block is recorded | |
808 | * in the back refs. Actually the full back refs is generic, and can be | |
809 | * used in all cases the implicit back refs is used. The major shortcoming | |
810 | * of the full back refs is its overhead. Every time a tree block gets | |
811 | * COWed, we have to update back refs entry for all pointers in it. | |
812 | * | |
813 | * For a newly allocated tree block, we use implicit back refs for | |
814 | * pointers in it. This means most tree related operations only involve | |
815 | * implicit back refs. For a tree block created in old transaction, the | |
816 | * only way to drop a reference to it is COW it. So we can detect the | |
817 | * event that tree block loses its owner tree's reference and do the | |
818 | * back refs conversion. | |
819 | * | |
820 | * When a tree block is COW'd through a tree, there are four cases: | |
821 | * | |
822 | * The reference count of the block is one and the tree is the block's | |
823 | * owner tree. Nothing to do in this case. | |
824 | * | |
825 | * The reference count of the block is one and the tree is not the | |
826 | * block's owner tree. In this case, full back refs is used for pointers | |
827 | * in the block. Remove these full back refs, add implicit back refs for | |
828 | * every pointers in the new block. | |
829 | * | |
830 | * The reference count of the block is greater than one and the tree is | |
831 | * the block's owner tree. In this case, implicit back refs is used for | |
832 | * pointers in the block. Add full back refs for every pointers in the | |
833 | * block, increase lower level extents' reference counts. The original | |
834 | * implicit back refs are entailed to the new block. | |
835 | * | |
836 | * The reference count of the block is greater than one and the tree is | |
837 | * not the block's owner tree. Add implicit back refs for every pointer in | |
838 | * the new block, increase lower level extents' reference count. | |
839 | * | |
840 | * Back Reference Key composing: | |
841 | * | |
842 | * The key objectid corresponds to the first byte in the extent, | |
843 | * The key type is used to differentiate between types of back refs. | |
844 | * There are different meanings of the key offset for different types | |
845 | * of back refs. | |
846 | * | |
d8d5f3e1 CM |
847 | * File extents can be referenced by: |
848 | * | |
849 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 850 | * - different files inside a single subvolume |
d8d5f3e1 CM |
851 | * - different offsets inside a file (bookend extents in file.c) |
852 | * | |
5d4f98a2 | 853 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
854 | * |
855 | * - Objectid of the subvolume root | |
d8d5f3e1 | 856 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
857 | * - original offset in the file |
858 | * - how many bookend extents | |
d8d5f3e1 | 859 | * |
5d4f98a2 YZ |
860 | * The key offset for the implicit back refs is hash of the first |
861 | * three fields. | |
d8d5f3e1 | 862 | * |
5d4f98a2 | 863 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 864 | * |
5d4f98a2 | 865 | * - number of pointers in the tree leaf |
d8d5f3e1 | 866 | * |
5d4f98a2 YZ |
867 | * The key offset for the implicit back refs is the first byte of |
868 | * the tree leaf | |
d8d5f3e1 | 869 | * |
5d4f98a2 YZ |
870 | * When a file extent is allocated, The implicit back refs is used. |
871 | * the fields are filled in: | |
d8d5f3e1 | 872 | * |
5d4f98a2 | 873 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 874 | * |
5d4f98a2 YZ |
875 | * When a file extent is removed file truncation, we find the |
876 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 877 | * |
5d4f98a2 | 878 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 879 | * |
5d4f98a2 | 880 | * Btree extents can be referenced by: |
d8d5f3e1 | 881 | * |
5d4f98a2 | 882 | * - Different subvolumes |
d8d5f3e1 | 883 | * |
5d4f98a2 YZ |
884 | * Both the implicit back refs and the full back refs for tree blocks |
885 | * only consist of key. The key offset for the implicit back refs is | |
886 | * objectid of block's owner tree. The key offset for the full back refs | |
887 | * is the first byte of parent block. | |
d8d5f3e1 | 888 | * |
5d4f98a2 YZ |
889 | * When implicit back refs is used, information about the lowest key and |
890 | * level of the tree block are required. These information are stored in | |
891 | * tree block info structure. | |
d8d5f3e1 | 892 | */ |
31840ae1 | 893 | |
5d4f98a2 YZ |
894 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
895 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
896 | struct btrfs_root *root, | |
897 | struct btrfs_path *path, | |
898 | u64 owner, u32 extra_size) | |
7bb86316 | 899 | { |
5d4f98a2 YZ |
900 | struct btrfs_extent_item *item; |
901 | struct btrfs_extent_item_v0 *ei0; | |
902 | struct btrfs_extent_ref_v0 *ref0; | |
903 | struct btrfs_tree_block_info *bi; | |
904 | struct extent_buffer *leaf; | |
7bb86316 | 905 | struct btrfs_key key; |
5d4f98a2 YZ |
906 | struct btrfs_key found_key; |
907 | u32 new_size = sizeof(*item); | |
908 | u64 refs; | |
909 | int ret; | |
910 | ||
911 | leaf = path->nodes[0]; | |
912 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
913 | ||
914 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
915 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
916 | struct btrfs_extent_item_v0); | |
917 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
918 | ||
919 | if (owner == (u64)-1) { | |
920 | while (1) { | |
921 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
922 | ret = btrfs_next_leaf(root, path); | |
923 | if (ret < 0) | |
924 | return ret; | |
925 | BUG_ON(ret > 0); | |
926 | leaf = path->nodes[0]; | |
927 | } | |
928 | btrfs_item_key_to_cpu(leaf, &found_key, | |
929 | path->slots[0]); | |
930 | BUG_ON(key.objectid != found_key.objectid); | |
931 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
932 | path->slots[0]++; | |
933 | continue; | |
934 | } | |
935 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
936 | struct btrfs_extent_ref_v0); | |
937 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
938 | break; | |
939 | } | |
940 | } | |
b3b4aa74 | 941 | btrfs_release_path(path); |
5d4f98a2 YZ |
942 | |
943 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
944 | new_size += sizeof(*bi); | |
945 | ||
946 | new_size -= sizeof(*ei0); | |
947 | ret = btrfs_search_slot(trans, root, &key, path, | |
948 | new_size + extra_size, 1); | |
949 | if (ret < 0) | |
950 | return ret; | |
951 | BUG_ON(ret); | |
952 | ||
953 | ret = btrfs_extend_item(trans, root, path, new_size); | |
954 | BUG_ON(ret); | |
955 | ||
956 | leaf = path->nodes[0]; | |
957 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
958 | btrfs_set_extent_refs(leaf, item, refs); | |
959 | /* FIXME: get real generation */ | |
960 | btrfs_set_extent_generation(leaf, item, 0); | |
961 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
962 | btrfs_set_extent_flags(leaf, item, | |
963 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
964 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
965 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
966 | /* FIXME: get first key of the block */ | |
967 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
968 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
969 | } else { | |
970 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
971 | } | |
972 | btrfs_mark_buffer_dirty(leaf); | |
973 | return 0; | |
974 | } | |
975 | #endif | |
976 | ||
977 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
978 | { | |
979 | u32 high_crc = ~(u32)0; | |
980 | u32 low_crc = ~(u32)0; | |
981 | __le64 lenum; | |
982 | ||
983 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 984 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 985 | lenum = cpu_to_le64(owner); |
163e783e | 986 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 987 | lenum = cpu_to_le64(offset); |
163e783e | 988 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
989 | |
990 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
991 | } | |
992 | ||
993 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
994 | struct btrfs_extent_data_ref *ref) | |
995 | { | |
996 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
997 | btrfs_extent_data_ref_objectid(leaf, ref), | |
998 | btrfs_extent_data_ref_offset(leaf, ref)); | |
999 | } | |
1000 | ||
1001 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1002 | struct btrfs_extent_data_ref *ref, | |
1003 | u64 root_objectid, u64 owner, u64 offset) | |
1004 | { | |
1005 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1006 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1007 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1008 | return 0; | |
1009 | return 1; | |
1010 | } | |
1011 | ||
1012 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1013 | struct btrfs_root *root, | |
1014 | struct btrfs_path *path, | |
1015 | u64 bytenr, u64 parent, | |
1016 | u64 root_objectid, | |
1017 | u64 owner, u64 offset) | |
1018 | { | |
1019 | struct btrfs_key key; | |
1020 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1021 | struct extent_buffer *leaf; |
5d4f98a2 | 1022 | u32 nritems; |
74493f7a | 1023 | int ret; |
5d4f98a2 YZ |
1024 | int recow; |
1025 | int err = -ENOENT; | |
74493f7a | 1026 | |
31840ae1 | 1027 | key.objectid = bytenr; |
5d4f98a2 YZ |
1028 | if (parent) { |
1029 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1030 | key.offset = parent; | |
1031 | } else { | |
1032 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1033 | key.offset = hash_extent_data_ref(root_objectid, | |
1034 | owner, offset); | |
1035 | } | |
1036 | again: | |
1037 | recow = 0; | |
1038 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1039 | if (ret < 0) { | |
1040 | err = ret; | |
1041 | goto fail; | |
1042 | } | |
31840ae1 | 1043 | |
5d4f98a2 YZ |
1044 | if (parent) { |
1045 | if (!ret) | |
1046 | return 0; | |
1047 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1048 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1049 | btrfs_release_path(path); |
5d4f98a2 YZ |
1050 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1051 | if (ret < 0) { | |
1052 | err = ret; | |
1053 | goto fail; | |
1054 | } | |
1055 | if (!ret) | |
1056 | return 0; | |
1057 | #endif | |
1058 | goto fail; | |
31840ae1 ZY |
1059 | } |
1060 | ||
1061 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1062 | nritems = btrfs_header_nritems(leaf); |
1063 | while (1) { | |
1064 | if (path->slots[0] >= nritems) { | |
1065 | ret = btrfs_next_leaf(root, path); | |
1066 | if (ret < 0) | |
1067 | err = ret; | |
1068 | if (ret) | |
1069 | goto fail; | |
1070 | ||
1071 | leaf = path->nodes[0]; | |
1072 | nritems = btrfs_header_nritems(leaf); | |
1073 | recow = 1; | |
1074 | } | |
1075 | ||
1076 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1077 | if (key.objectid != bytenr || | |
1078 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1079 | goto fail; | |
1080 | ||
1081 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1082 | struct btrfs_extent_data_ref); | |
1083 | ||
1084 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1085 | owner, offset)) { | |
1086 | if (recow) { | |
b3b4aa74 | 1087 | btrfs_release_path(path); |
5d4f98a2 YZ |
1088 | goto again; |
1089 | } | |
1090 | err = 0; | |
1091 | break; | |
1092 | } | |
1093 | path->slots[0]++; | |
31840ae1 | 1094 | } |
5d4f98a2 YZ |
1095 | fail: |
1096 | return err; | |
31840ae1 ZY |
1097 | } |
1098 | ||
5d4f98a2 YZ |
1099 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1100 | struct btrfs_root *root, | |
1101 | struct btrfs_path *path, | |
1102 | u64 bytenr, u64 parent, | |
1103 | u64 root_objectid, u64 owner, | |
1104 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1105 | { |
1106 | struct btrfs_key key; | |
1107 | struct extent_buffer *leaf; | |
5d4f98a2 | 1108 | u32 size; |
31840ae1 ZY |
1109 | u32 num_refs; |
1110 | int ret; | |
74493f7a | 1111 | |
74493f7a | 1112 | key.objectid = bytenr; |
5d4f98a2 YZ |
1113 | if (parent) { |
1114 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1115 | key.offset = parent; | |
1116 | size = sizeof(struct btrfs_shared_data_ref); | |
1117 | } else { | |
1118 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1119 | key.offset = hash_extent_data_ref(root_objectid, | |
1120 | owner, offset); | |
1121 | size = sizeof(struct btrfs_extent_data_ref); | |
1122 | } | |
74493f7a | 1123 | |
5d4f98a2 YZ |
1124 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1125 | if (ret && ret != -EEXIST) | |
1126 | goto fail; | |
1127 | ||
1128 | leaf = path->nodes[0]; | |
1129 | if (parent) { | |
1130 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1131 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1132 | struct btrfs_shared_data_ref); |
1133 | if (ret == 0) { | |
1134 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1135 | } else { | |
1136 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1137 | num_refs += refs_to_add; | |
1138 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1139 | } |
5d4f98a2 YZ |
1140 | } else { |
1141 | struct btrfs_extent_data_ref *ref; | |
1142 | while (ret == -EEXIST) { | |
1143 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1144 | struct btrfs_extent_data_ref); | |
1145 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1146 | owner, offset)) | |
1147 | break; | |
b3b4aa74 | 1148 | btrfs_release_path(path); |
5d4f98a2 YZ |
1149 | key.offset++; |
1150 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1151 | size); | |
1152 | if (ret && ret != -EEXIST) | |
1153 | goto fail; | |
31840ae1 | 1154 | |
5d4f98a2 YZ |
1155 | leaf = path->nodes[0]; |
1156 | } | |
1157 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1158 | struct btrfs_extent_data_ref); | |
1159 | if (ret == 0) { | |
1160 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1161 | root_objectid); | |
1162 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1163 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1164 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1165 | } else { | |
1166 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1167 | num_refs += refs_to_add; | |
1168 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1169 | } |
31840ae1 | 1170 | } |
5d4f98a2 YZ |
1171 | btrfs_mark_buffer_dirty(leaf); |
1172 | ret = 0; | |
1173 | fail: | |
b3b4aa74 | 1174 | btrfs_release_path(path); |
7bb86316 | 1175 | return ret; |
74493f7a CM |
1176 | } |
1177 | ||
5d4f98a2 YZ |
1178 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1179 | struct btrfs_root *root, | |
1180 | struct btrfs_path *path, | |
1181 | int refs_to_drop) | |
31840ae1 | 1182 | { |
5d4f98a2 YZ |
1183 | struct btrfs_key key; |
1184 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1185 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1186 | struct extent_buffer *leaf; |
5d4f98a2 | 1187 | u32 num_refs = 0; |
31840ae1 ZY |
1188 | int ret = 0; |
1189 | ||
1190 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1191 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1192 | ||
1193 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1194 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1195 | struct btrfs_extent_data_ref); | |
1196 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1197 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1198 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1199 | struct btrfs_shared_data_ref); | |
1200 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1201 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1202 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1203 | struct btrfs_extent_ref_v0 *ref0; | |
1204 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1205 | struct btrfs_extent_ref_v0); | |
1206 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1207 | #endif | |
1208 | } else { | |
1209 | BUG(); | |
1210 | } | |
1211 | ||
56bec294 CM |
1212 | BUG_ON(num_refs < refs_to_drop); |
1213 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1214 | |
31840ae1 ZY |
1215 | if (num_refs == 0) { |
1216 | ret = btrfs_del_item(trans, root, path); | |
1217 | } else { | |
5d4f98a2 YZ |
1218 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1219 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1220 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1221 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1222 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1223 | else { | |
1224 | struct btrfs_extent_ref_v0 *ref0; | |
1225 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1226 | struct btrfs_extent_ref_v0); | |
1227 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1228 | } | |
1229 | #endif | |
31840ae1 ZY |
1230 | btrfs_mark_buffer_dirty(leaf); |
1231 | } | |
31840ae1 ZY |
1232 | return ret; |
1233 | } | |
1234 | ||
5d4f98a2 YZ |
1235 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1236 | struct btrfs_path *path, | |
1237 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1238 | { |
5d4f98a2 YZ |
1239 | struct btrfs_key key; |
1240 | struct extent_buffer *leaf; | |
1241 | struct btrfs_extent_data_ref *ref1; | |
1242 | struct btrfs_shared_data_ref *ref2; | |
1243 | u32 num_refs = 0; | |
1244 | ||
1245 | leaf = path->nodes[0]; | |
1246 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1247 | if (iref) { | |
1248 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1249 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1250 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1251 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1252 | } else { | |
1253 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1254 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1255 | } | |
1256 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1257 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1258 | struct btrfs_extent_data_ref); | |
1259 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1260 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1261 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1262 | struct btrfs_shared_data_ref); | |
1263 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1264 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1265 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1266 | struct btrfs_extent_ref_v0 *ref0; | |
1267 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1268 | struct btrfs_extent_ref_v0); | |
1269 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1270 | #endif |
5d4f98a2 YZ |
1271 | } else { |
1272 | WARN_ON(1); | |
1273 | } | |
1274 | return num_refs; | |
1275 | } | |
15916de8 | 1276 | |
5d4f98a2 YZ |
1277 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1278 | struct btrfs_root *root, | |
1279 | struct btrfs_path *path, | |
1280 | u64 bytenr, u64 parent, | |
1281 | u64 root_objectid) | |
1f3c79a2 | 1282 | { |
5d4f98a2 | 1283 | struct btrfs_key key; |
1f3c79a2 | 1284 | int ret; |
1f3c79a2 | 1285 | |
5d4f98a2 YZ |
1286 | key.objectid = bytenr; |
1287 | if (parent) { | |
1288 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1289 | key.offset = parent; | |
1290 | } else { | |
1291 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1292 | key.offset = root_objectid; | |
1f3c79a2 LH |
1293 | } |
1294 | ||
5d4f98a2 YZ |
1295 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1296 | if (ret > 0) | |
1297 | ret = -ENOENT; | |
1298 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1299 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1300 | btrfs_release_path(path); |
5d4f98a2 YZ |
1301 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1302 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1303 | if (ret > 0) | |
1304 | ret = -ENOENT; | |
1305 | } | |
1f3c79a2 | 1306 | #endif |
5d4f98a2 | 1307 | return ret; |
1f3c79a2 LH |
1308 | } |
1309 | ||
5d4f98a2 YZ |
1310 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1311 | struct btrfs_root *root, | |
1312 | struct btrfs_path *path, | |
1313 | u64 bytenr, u64 parent, | |
1314 | u64 root_objectid) | |
31840ae1 | 1315 | { |
5d4f98a2 | 1316 | struct btrfs_key key; |
31840ae1 | 1317 | int ret; |
31840ae1 | 1318 | |
5d4f98a2 YZ |
1319 | key.objectid = bytenr; |
1320 | if (parent) { | |
1321 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1322 | key.offset = parent; | |
1323 | } else { | |
1324 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1325 | key.offset = root_objectid; | |
1326 | } | |
1327 | ||
1328 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1329 | btrfs_release_path(path); |
31840ae1 ZY |
1330 | return ret; |
1331 | } | |
1332 | ||
5d4f98a2 | 1333 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1334 | { |
5d4f98a2 YZ |
1335 | int type; |
1336 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1337 | if (parent > 0) | |
1338 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1339 | else | |
1340 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1341 | } else { | |
1342 | if (parent > 0) | |
1343 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1344 | else | |
1345 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1346 | } | |
1347 | return type; | |
31840ae1 | 1348 | } |
56bec294 | 1349 | |
2c47e605 YZ |
1350 | static int find_next_key(struct btrfs_path *path, int level, |
1351 | struct btrfs_key *key) | |
56bec294 | 1352 | |
02217ed2 | 1353 | { |
2c47e605 | 1354 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1355 | if (!path->nodes[level]) |
1356 | break; | |
5d4f98a2 YZ |
1357 | if (path->slots[level] + 1 >= |
1358 | btrfs_header_nritems(path->nodes[level])) | |
1359 | continue; | |
1360 | if (level == 0) | |
1361 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1362 | path->slots[level] + 1); | |
1363 | else | |
1364 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1365 | path->slots[level] + 1); | |
1366 | return 0; | |
1367 | } | |
1368 | return 1; | |
1369 | } | |
037e6390 | 1370 | |
5d4f98a2 YZ |
1371 | /* |
1372 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1373 | * to the address of inline back ref, and 0 is returned. | |
1374 | * | |
1375 | * if back ref isn't found, *ref_ret is set to the address where it | |
1376 | * should be inserted, and -ENOENT is returned. | |
1377 | * | |
1378 | * if insert is true and there are too many inline back refs, the path | |
1379 | * points to the extent item, and -EAGAIN is returned. | |
1380 | * | |
1381 | * NOTE: inline back refs are ordered in the same way that back ref | |
1382 | * items in the tree are ordered. | |
1383 | */ | |
1384 | static noinline_for_stack | |
1385 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1386 | struct btrfs_root *root, | |
1387 | struct btrfs_path *path, | |
1388 | struct btrfs_extent_inline_ref **ref_ret, | |
1389 | u64 bytenr, u64 num_bytes, | |
1390 | u64 parent, u64 root_objectid, | |
1391 | u64 owner, u64 offset, int insert) | |
1392 | { | |
1393 | struct btrfs_key key; | |
1394 | struct extent_buffer *leaf; | |
1395 | struct btrfs_extent_item *ei; | |
1396 | struct btrfs_extent_inline_ref *iref; | |
1397 | u64 flags; | |
1398 | u64 item_size; | |
1399 | unsigned long ptr; | |
1400 | unsigned long end; | |
1401 | int extra_size; | |
1402 | int type; | |
1403 | int want; | |
1404 | int ret; | |
1405 | int err = 0; | |
26b8003f | 1406 | |
db94535d | 1407 | key.objectid = bytenr; |
31840ae1 | 1408 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1409 | key.offset = num_bytes; |
31840ae1 | 1410 | |
5d4f98a2 YZ |
1411 | want = extent_ref_type(parent, owner); |
1412 | if (insert) { | |
1413 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1414 | path->keep_locks = 1; |
5d4f98a2 YZ |
1415 | } else |
1416 | extra_size = -1; | |
1417 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1418 | if (ret < 0) { |
5d4f98a2 YZ |
1419 | err = ret; |
1420 | goto out; | |
1421 | } | |
1422 | BUG_ON(ret); | |
1423 | ||
1424 | leaf = path->nodes[0]; | |
1425 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1426 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1427 | if (item_size < sizeof(*ei)) { | |
1428 | if (!insert) { | |
1429 | err = -ENOENT; | |
1430 | goto out; | |
1431 | } | |
1432 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1433 | extra_size); | |
1434 | if (ret < 0) { | |
1435 | err = ret; | |
1436 | goto out; | |
1437 | } | |
1438 | leaf = path->nodes[0]; | |
1439 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1440 | } | |
1441 | #endif | |
1442 | BUG_ON(item_size < sizeof(*ei)); | |
1443 | ||
5d4f98a2 YZ |
1444 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1445 | flags = btrfs_extent_flags(leaf, ei); | |
1446 | ||
1447 | ptr = (unsigned long)(ei + 1); | |
1448 | end = (unsigned long)ei + item_size; | |
1449 | ||
1450 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1451 | ptr += sizeof(struct btrfs_tree_block_info); | |
1452 | BUG_ON(ptr > end); | |
1453 | } else { | |
1454 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1455 | } | |
1456 | ||
1457 | err = -ENOENT; | |
1458 | while (1) { | |
1459 | if (ptr >= end) { | |
1460 | WARN_ON(ptr > end); | |
1461 | break; | |
1462 | } | |
1463 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1464 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1465 | if (want < type) | |
1466 | break; | |
1467 | if (want > type) { | |
1468 | ptr += btrfs_extent_inline_ref_size(type); | |
1469 | continue; | |
1470 | } | |
1471 | ||
1472 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1473 | struct btrfs_extent_data_ref *dref; | |
1474 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1475 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1476 | owner, offset)) { | |
1477 | err = 0; | |
1478 | break; | |
1479 | } | |
1480 | if (hash_extent_data_ref_item(leaf, dref) < | |
1481 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1482 | break; | |
1483 | } else { | |
1484 | u64 ref_offset; | |
1485 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1486 | if (parent > 0) { | |
1487 | if (parent == ref_offset) { | |
1488 | err = 0; | |
1489 | break; | |
1490 | } | |
1491 | if (ref_offset < parent) | |
1492 | break; | |
1493 | } else { | |
1494 | if (root_objectid == ref_offset) { | |
1495 | err = 0; | |
1496 | break; | |
1497 | } | |
1498 | if (ref_offset < root_objectid) | |
1499 | break; | |
1500 | } | |
1501 | } | |
1502 | ptr += btrfs_extent_inline_ref_size(type); | |
1503 | } | |
1504 | if (err == -ENOENT && insert) { | |
1505 | if (item_size + extra_size >= | |
1506 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1507 | err = -EAGAIN; | |
1508 | goto out; | |
1509 | } | |
1510 | /* | |
1511 | * To add new inline back ref, we have to make sure | |
1512 | * there is no corresponding back ref item. | |
1513 | * For simplicity, we just do not add new inline back | |
1514 | * ref if there is any kind of item for this block | |
1515 | */ | |
2c47e605 YZ |
1516 | if (find_next_key(path, 0, &key) == 0 && |
1517 | key.objectid == bytenr && | |
85d4198e | 1518 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1519 | err = -EAGAIN; |
1520 | goto out; | |
1521 | } | |
1522 | } | |
1523 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1524 | out: | |
85d4198e | 1525 | if (insert) { |
5d4f98a2 YZ |
1526 | path->keep_locks = 0; |
1527 | btrfs_unlock_up_safe(path, 1); | |
1528 | } | |
1529 | return err; | |
1530 | } | |
1531 | ||
1532 | /* | |
1533 | * helper to add new inline back ref | |
1534 | */ | |
1535 | static noinline_for_stack | |
1536 | int setup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1537 | struct btrfs_root *root, | |
1538 | struct btrfs_path *path, | |
1539 | struct btrfs_extent_inline_ref *iref, | |
1540 | u64 parent, u64 root_objectid, | |
1541 | u64 owner, u64 offset, int refs_to_add, | |
1542 | struct btrfs_delayed_extent_op *extent_op) | |
1543 | { | |
1544 | struct extent_buffer *leaf; | |
1545 | struct btrfs_extent_item *ei; | |
1546 | unsigned long ptr; | |
1547 | unsigned long end; | |
1548 | unsigned long item_offset; | |
1549 | u64 refs; | |
1550 | int size; | |
1551 | int type; | |
1552 | int ret; | |
1553 | ||
1554 | leaf = path->nodes[0]; | |
1555 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1556 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1557 | ||
1558 | type = extent_ref_type(parent, owner); | |
1559 | size = btrfs_extent_inline_ref_size(type); | |
1560 | ||
1561 | ret = btrfs_extend_item(trans, root, path, size); | |
1562 | BUG_ON(ret); | |
1563 | ||
1564 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1565 | refs = btrfs_extent_refs(leaf, ei); | |
1566 | refs += refs_to_add; | |
1567 | btrfs_set_extent_refs(leaf, ei, refs); | |
1568 | if (extent_op) | |
1569 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1570 | ||
1571 | ptr = (unsigned long)ei + item_offset; | |
1572 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1573 | if (ptr < end - size) | |
1574 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1575 | end - size - ptr); | |
1576 | ||
1577 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1578 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1579 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1580 | struct btrfs_extent_data_ref *dref; | |
1581 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1582 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1583 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1584 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1585 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1586 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1587 | struct btrfs_shared_data_ref *sref; | |
1588 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1589 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1590 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1591 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1592 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1593 | } else { | |
1594 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1595 | } | |
1596 | btrfs_mark_buffer_dirty(leaf); | |
1597 | return 0; | |
1598 | } | |
1599 | ||
1600 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1601 | struct btrfs_root *root, | |
1602 | struct btrfs_path *path, | |
1603 | struct btrfs_extent_inline_ref **ref_ret, | |
1604 | u64 bytenr, u64 num_bytes, u64 parent, | |
1605 | u64 root_objectid, u64 owner, u64 offset) | |
1606 | { | |
1607 | int ret; | |
1608 | ||
1609 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1610 | bytenr, num_bytes, parent, | |
1611 | root_objectid, owner, offset, 0); | |
1612 | if (ret != -ENOENT) | |
54aa1f4d | 1613 | return ret; |
5d4f98a2 | 1614 | |
b3b4aa74 | 1615 | btrfs_release_path(path); |
5d4f98a2 YZ |
1616 | *ref_ret = NULL; |
1617 | ||
1618 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1619 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1620 | root_objectid); | |
1621 | } else { | |
1622 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1623 | root_objectid, owner, offset); | |
b9473439 | 1624 | } |
5d4f98a2 YZ |
1625 | return ret; |
1626 | } | |
31840ae1 | 1627 | |
5d4f98a2 YZ |
1628 | /* |
1629 | * helper to update/remove inline back ref | |
1630 | */ | |
1631 | static noinline_for_stack | |
1632 | int update_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1633 | struct btrfs_root *root, | |
1634 | struct btrfs_path *path, | |
1635 | struct btrfs_extent_inline_ref *iref, | |
1636 | int refs_to_mod, | |
1637 | struct btrfs_delayed_extent_op *extent_op) | |
1638 | { | |
1639 | struct extent_buffer *leaf; | |
1640 | struct btrfs_extent_item *ei; | |
1641 | struct btrfs_extent_data_ref *dref = NULL; | |
1642 | struct btrfs_shared_data_ref *sref = NULL; | |
1643 | unsigned long ptr; | |
1644 | unsigned long end; | |
1645 | u32 item_size; | |
1646 | int size; | |
1647 | int type; | |
1648 | int ret; | |
1649 | u64 refs; | |
1650 | ||
1651 | leaf = path->nodes[0]; | |
1652 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1653 | refs = btrfs_extent_refs(leaf, ei); | |
1654 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1655 | refs += refs_to_mod; | |
1656 | btrfs_set_extent_refs(leaf, ei, refs); | |
1657 | if (extent_op) | |
1658 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1659 | ||
1660 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1661 | ||
1662 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1663 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1664 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1665 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1666 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1667 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1668 | } else { | |
1669 | refs = 1; | |
1670 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1671 | } |
31840ae1 | 1672 | |
5d4f98a2 YZ |
1673 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1674 | refs += refs_to_mod; | |
1675 | ||
1676 | if (refs > 0) { | |
1677 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1678 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1679 | else | |
1680 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1681 | } else { | |
1682 | size = btrfs_extent_inline_ref_size(type); | |
1683 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1684 | ptr = (unsigned long)iref; | |
1685 | end = (unsigned long)ei + item_size; | |
1686 | if (ptr + size < end) | |
1687 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1688 | end - ptr - size); | |
1689 | item_size -= size; | |
1690 | ret = btrfs_truncate_item(trans, root, path, item_size, 1); | |
1691 | BUG_ON(ret); | |
1692 | } | |
1693 | btrfs_mark_buffer_dirty(leaf); | |
1694 | return 0; | |
1695 | } | |
1696 | ||
1697 | static noinline_for_stack | |
1698 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1699 | struct btrfs_root *root, | |
1700 | struct btrfs_path *path, | |
1701 | u64 bytenr, u64 num_bytes, u64 parent, | |
1702 | u64 root_objectid, u64 owner, | |
1703 | u64 offset, int refs_to_add, | |
1704 | struct btrfs_delayed_extent_op *extent_op) | |
1705 | { | |
1706 | struct btrfs_extent_inline_ref *iref; | |
1707 | int ret; | |
1708 | ||
1709 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1710 | bytenr, num_bytes, parent, | |
1711 | root_objectid, owner, offset, 1); | |
1712 | if (ret == 0) { | |
1713 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
1714 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1715 | refs_to_add, extent_op); | |
1716 | } else if (ret == -ENOENT) { | |
1717 | ret = setup_inline_extent_backref(trans, root, path, iref, | |
1718 | parent, root_objectid, | |
1719 | owner, offset, refs_to_add, | |
1720 | extent_op); | |
771ed689 | 1721 | } |
5d4f98a2 YZ |
1722 | return ret; |
1723 | } | |
31840ae1 | 1724 | |
5d4f98a2 YZ |
1725 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1726 | struct btrfs_root *root, | |
1727 | struct btrfs_path *path, | |
1728 | u64 bytenr, u64 parent, u64 root_objectid, | |
1729 | u64 owner, u64 offset, int refs_to_add) | |
1730 | { | |
1731 | int ret; | |
1732 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1733 | BUG_ON(refs_to_add != 1); | |
1734 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1735 | parent, root_objectid); | |
1736 | } else { | |
1737 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1738 | parent, root_objectid, | |
1739 | owner, offset, refs_to_add); | |
1740 | } | |
1741 | return ret; | |
1742 | } | |
56bec294 | 1743 | |
5d4f98a2 YZ |
1744 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1745 | struct btrfs_root *root, | |
1746 | struct btrfs_path *path, | |
1747 | struct btrfs_extent_inline_ref *iref, | |
1748 | int refs_to_drop, int is_data) | |
1749 | { | |
1750 | int ret; | |
b9473439 | 1751 | |
5d4f98a2 YZ |
1752 | BUG_ON(!is_data && refs_to_drop != 1); |
1753 | if (iref) { | |
1754 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1755 | -refs_to_drop, NULL); | |
1756 | } else if (is_data) { | |
1757 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1758 | } else { | |
1759 | ret = btrfs_del_item(trans, root, path); | |
1760 | } | |
1761 | return ret; | |
1762 | } | |
1763 | ||
5378e607 | 1764 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1765 | u64 start, u64 len) |
1766 | { | |
5378e607 | 1767 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1768 | } |
5d4f98a2 YZ |
1769 | |
1770 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1771 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1772 | { |
5d4f98a2 | 1773 | int ret; |
5378e607 | 1774 | u64 discarded_bytes = 0; |
5d4f98a2 YZ |
1775 | struct btrfs_multi_bio *multi = NULL; |
1776 | ||
e244a0ae | 1777 | |
5d4f98a2 | 1778 | /* Tell the block device(s) that the sectors can be discarded */ |
5378e607 LD |
1779 | ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD, |
1780 | bytenr, &num_bytes, &multi, 0); | |
5d4f98a2 YZ |
1781 | if (!ret) { |
1782 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
1783 | int i; | |
1784 | ||
5d4f98a2 YZ |
1785 | |
1786 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
5378e607 LD |
1787 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1788 | stripe->physical, | |
1789 | stripe->length); | |
1790 | if (!ret) | |
1791 | discarded_bytes += stripe->length; | |
1792 | else if (ret != -EOPNOTSUPP) | |
1793 | break; | |
5d4f98a2 YZ |
1794 | } |
1795 | kfree(multi); | |
1796 | } | |
5378e607 LD |
1797 | if (discarded_bytes && ret == -EOPNOTSUPP) |
1798 | ret = 0; | |
1799 | ||
1800 | if (actual_bytes) | |
1801 | *actual_bytes = discarded_bytes; | |
1802 | ||
5d4f98a2 YZ |
1803 | |
1804 | return ret; | |
5d4f98a2 YZ |
1805 | } |
1806 | ||
1807 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1808 | struct btrfs_root *root, | |
1809 | u64 bytenr, u64 num_bytes, u64 parent, | |
1810 | u64 root_objectid, u64 owner, u64 offset) | |
1811 | { | |
1812 | int ret; | |
1813 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && | |
1814 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1815 | ||
1816 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1817 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
1818 | parent, root_objectid, (int)owner, | |
1819 | BTRFS_ADD_DELAYED_REF, NULL); | |
1820 | } else { | |
1821 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
1822 | parent, root_objectid, owner, offset, | |
1823 | BTRFS_ADD_DELAYED_REF, NULL); | |
1824 | } | |
1825 | return ret; | |
1826 | } | |
1827 | ||
1828 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1829 | struct btrfs_root *root, | |
1830 | u64 bytenr, u64 num_bytes, | |
1831 | u64 parent, u64 root_objectid, | |
1832 | u64 owner, u64 offset, int refs_to_add, | |
1833 | struct btrfs_delayed_extent_op *extent_op) | |
1834 | { | |
1835 | struct btrfs_path *path; | |
1836 | struct extent_buffer *leaf; | |
1837 | struct btrfs_extent_item *item; | |
1838 | u64 refs; | |
1839 | int ret; | |
1840 | int err = 0; | |
1841 | ||
1842 | path = btrfs_alloc_path(); | |
1843 | if (!path) | |
1844 | return -ENOMEM; | |
1845 | ||
1846 | path->reada = 1; | |
1847 | path->leave_spinning = 1; | |
1848 | /* this will setup the path even if it fails to insert the back ref */ | |
1849 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1850 | path, bytenr, num_bytes, parent, | |
1851 | root_objectid, owner, offset, | |
1852 | refs_to_add, extent_op); | |
1853 | if (ret == 0) | |
1854 | goto out; | |
1855 | ||
1856 | if (ret != -EAGAIN) { | |
1857 | err = ret; | |
1858 | goto out; | |
1859 | } | |
1860 | ||
1861 | leaf = path->nodes[0]; | |
1862 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1863 | refs = btrfs_extent_refs(leaf, item); | |
1864 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1865 | if (extent_op) | |
1866 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1867 | |
5d4f98a2 | 1868 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1869 | btrfs_release_path(path); |
56bec294 CM |
1870 | |
1871 | path->reada = 1; | |
b9473439 CM |
1872 | path->leave_spinning = 1; |
1873 | ||
56bec294 CM |
1874 | /* now insert the actual backref */ |
1875 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1876 | path, bytenr, parent, root_objectid, |
1877 | owner, offset, refs_to_add); | |
56bec294 | 1878 | BUG_ON(ret); |
5d4f98a2 | 1879 | out: |
56bec294 | 1880 | btrfs_free_path(path); |
5d4f98a2 | 1881 | return err; |
56bec294 CM |
1882 | } |
1883 | ||
5d4f98a2 YZ |
1884 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1885 | struct btrfs_root *root, | |
1886 | struct btrfs_delayed_ref_node *node, | |
1887 | struct btrfs_delayed_extent_op *extent_op, | |
1888 | int insert_reserved) | |
56bec294 | 1889 | { |
5d4f98a2 YZ |
1890 | int ret = 0; |
1891 | struct btrfs_delayed_data_ref *ref; | |
1892 | struct btrfs_key ins; | |
1893 | u64 parent = 0; | |
1894 | u64 ref_root = 0; | |
1895 | u64 flags = 0; | |
1896 | ||
1897 | ins.objectid = node->bytenr; | |
1898 | ins.offset = node->num_bytes; | |
1899 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1900 | ||
1901 | ref = btrfs_delayed_node_to_data_ref(node); | |
1902 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1903 | parent = ref->parent; | |
1904 | else | |
1905 | ref_root = ref->root; | |
1906 | ||
1907 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1908 | if (extent_op) { | |
1909 | BUG_ON(extent_op->update_key); | |
1910 | flags |= extent_op->flags_to_set; | |
1911 | } | |
1912 | ret = alloc_reserved_file_extent(trans, root, | |
1913 | parent, ref_root, flags, | |
1914 | ref->objectid, ref->offset, | |
1915 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1916 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1917 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1918 | node->num_bytes, parent, | |
1919 | ref_root, ref->objectid, | |
1920 | ref->offset, node->ref_mod, | |
1921 | extent_op); | |
1922 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1923 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1924 | node->num_bytes, parent, | |
1925 | ref_root, ref->objectid, | |
1926 | ref->offset, node->ref_mod, | |
1927 | extent_op); | |
1928 | } else { | |
1929 | BUG(); | |
1930 | } | |
1931 | return ret; | |
1932 | } | |
1933 | ||
1934 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1935 | struct extent_buffer *leaf, | |
1936 | struct btrfs_extent_item *ei) | |
1937 | { | |
1938 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1939 | if (extent_op->update_flags) { | |
1940 | flags |= extent_op->flags_to_set; | |
1941 | btrfs_set_extent_flags(leaf, ei, flags); | |
1942 | } | |
1943 | ||
1944 | if (extent_op->update_key) { | |
1945 | struct btrfs_tree_block_info *bi; | |
1946 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1947 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1948 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1949 | } | |
1950 | } | |
1951 | ||
1952 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
1953 | struct btrfs_root *root, | |
1954 | struct btrfs_delayed_ref_node *node, | |
1955 | struct btrfs_delayed_extent_op *extent_op) | |
1956 | { | |
1957 | struct btrfs_key key; | |
1958 | struct btrfs_path *path; | |
1959 | struct btrfs_extent_item *ei; | |
1960 | struct extent_buffer *leaf; | |
1961 | u32 item_size; | |
56bec294 | 1962 | int ret; |
5d4f98a2 YZ |
1963 | int err = 0; |
1964 | ||
1965 | path = btrfs_alloc_path(); | |
1966 | if (!path) | |
1967 | return -ENOMEM; | |
1968 | ||
1969 | key.objectid = node->bytenr; | |
1970 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1971 | key.offset = node->num_bytes; | |
1972 | ||
1973 | path->reada = 1; | |
1974 | path->leave_spinning = 1; | |
1975 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
1976 | path, 0, 1); | |
1977 | if (ret < 0) { | |
1978 | err = ret; | |
1979 | goto out; | |
1980 | } | |
1981 | if (ret > 0) { | |
1982 | err = -EIO; | |
1983 | goto out; | |
1984 | } | |
1985 | ||
1986 | leaf = path->nodes[0]; | |
1987 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1988 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1989 | if (item_size < sizeof(*ei)) { | |
1990 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
1991 | path, (u64)-1, 0); | |
1992 | if (ret < 0) { | |
1993 | err = ret; | |
1994 | goto out; | |
1995 | } | |
1996 | leaf = path->nodes[0]; | |
1997 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1998 | } | |
1999 | #endif | |
2000 | BUG_ON(item_size < sizeof(*ei)); | |
2001 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2002 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2003 | |
5d4f98a2 YZ |
2004 | btrfs_mark_buffer_dirty(leaf); |
2005 | out: | |
2006 | btrfs_free_path(path); | |
2007 | return err; | |
56bec294 CM |
2008 | } |
2009 | ||
5d4f98a2 YZ |
2010 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2011 | struct btrfs_root *root, | |
2012 | struct btrfs_delayed_ref_node *node, | |
2013 | struct btrfs_delayed_extent_op *extent_op, | |
2014 | int insert_reserved) | |
56bec294 CM |
2015 | { |
2016 | int ret = 0; | |
5d4f98a2 YZ |
2017 | struct btrfs_delayed_tree_ref *ref; |
2018 | struct btrfs_key ins; | |
2019 | u64 parent = 0; | |
2020 | u64 ref_root = 0; | |
56bec294 | 2021 | |
5d4f98a2 YZ |
2022 | ins.objectid = node->bytenr; |
2023 | ins.offset = node->num_bytes; | |
2024 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 2025 | |
5d4f98a2 YZ |
2026 | ref = btrfs_delayed_node_to_tree_ref(node); |
2027 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2028 | parent = ref->parent; | |
2029 | else | |
2030 | ref_root = ref->root; | |
2031 | ||
2032 | BUG_ON(node->ref_mod != 1); | |
2033 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2034 | BUG_ON(!extent_op || !extent_op->update_flags || | |
2035 | !extent_op->update_key); | |
2036 | ret = alloc_reserved_tree_block(trans, root, | |
2037 | parent, ref_root, | |
2038 | extent_op->flags_to_set, | |
2039 | &extent_op->key, | |
2040 | ref->level, &ins); | |
5d4f98a2 YZ |
2041 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2042 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2043 | node->num_bytes, parent, ref_root, | |
2044 | ref->level, 0, 1, extent_op); | |
2045 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2046 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2047 | node->num_bytes, parent, ref_root, | |
2048 | ref->level, 0, 1, extent_op); | |
2049 | } else { | |
2050 | BUG(); | |
2051 | } | |
56bec294 CM |
2052 | return ret; |
2053 | } | |
2054 | ||
2055 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2056 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2057 | struct btrfs_root *root, | |
2058 | struct btrfs_delayed_ref_node *node, | |
2059 | struct btrfs_delayed_extent_op *extent_op, | |
2060 | int insert_reserved) | |
56bec294 CM |
2061 | { |
2062 | int ret; | |
5d4f98a2 | 2063 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2064 | struct btrfs_delayed_ref_head *head; |
2065 | /* | |
2066 | * we've hit the end of the chain and we were supposed | |
2067 | * to insert this extent into the tree. But, it got | |
2068 | * deleted before we ever needed to insert it, so all | |
2069 | * we have to do is clean up the accounting | |
2070 | */ | |
5d4f98a2 YZ |
2071 | BUG_ON(extent_op); |
2072 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2073 | if (insert_reserved) { |
f0486c68 YZ |
2074 | btrfs_pin_extent(root, node->bytenr, |
2075 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2076 | if (head->is_data) { |
2077 | ret = btrfs_del_csums(trans, root, | |
2078 | node->bytenr, | |
2079 | node->num_bytes); | |
2080 | BUG_ON(ret); | |
2081 | } | |
56bec294 | 2082 | } |
56bec294 CM |
2083 | mutex_unlock(&head->mutex); |
2084 | return 0; | |
2085 | } | |
2086 | ||
5d4f98a2 YZ |
2087 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2088 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2089 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2090 | insert_reserved); | |
2091 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2092 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2093 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2094 | insert_reserved); | |
2095 | else | |
2096 | BUG(); | |
2097 | return ret; | |
56bec294 CM |
2098 | } |
2099 | ||
2100 | static noinline struct btrfs_delayed_ref_node * | |
2101 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2102 | { | |
2103 | struct rb_node *node; | |
2104 | struct btrfs_delayed_ref_node *ref; | |
2105 | int action = BTRFS_ADD_DELAYED_REF; | |
2106 | again: | |
2107 | /* | |
2108 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2109 | * this prevents ref count from going down to zero when | |
2110 | * there still are pending delayed ref. | |
2111 | */ | |
2112 | node = rb_prev(&head->node.rb_node); | |
2113 | while (1) { | |
2114 | if (!node) | |
2115 | break; | |
2116 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2117 | rb_node); | |
2118 | if (ref->bytenr != head->node.bytenr) | |
2119 | break; | |
5d4f98a2 | 2120 | if (ref->action == action) |
56bec294 CM |
2121 | return ref; |
2122 | node = rb_prev(node); | |
2123 | } | |
2124 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2125 | action = BTRFS_DROP_DELAYED_REF; | |
2126 | goto again; | |
2127 | } | |
2128 | return NULL; | |
2129 | } | |
2130 | ||
c3e69d58 CM |
2131 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2132 | struct btrfs_root *root, | |
2133 | struct list_head *cluster) | |
56bec294 | 2134 | { |
56bec294 CM |
2135 | struct btrfs_delayed_ref_root *delayed_refs; |
2136 | struct btrfs_delayed_ref_node *ref; | |
2137 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2138 | struct btrfs_delayed_extent_op *extent_op; |
56bec294 | 2139 | int ret; |
c3e69d58 | 2140 | int count = 0; |
56bec294 | 2141 | int must_insert_reserved = 0; |
56bec294 CM |
2142 | |
2143 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2144 | while (1) { |
2145 | if (!locked_ref) { | |
c3e69d58 CM |
2146 | /* pick a new head ref from the cluster list */ |
2147 | if (list_empty(cluster)) | |
56bec294 | 2148 | break; |
56bec294 | 2149 | |
c3e69d58 CM |
2150 | locked_ref = list_entry(cluster->next, |
2151 | struct btrfs_delayed_ref_head, cluster); | |
2152 | ||
2153 | /* grab the lock that says we are going to process | |
2154 | * all the refs for this head */ | |
2155 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2156 | ||
2157 | /* | |
2158 | * we may have dropped the spin lock to get the head | |
2159 | * mutex lock, and that might have given someone else | |
2160 | * time to free the head. If that's true, it has been | |
2161 | * removed from our list and we can move on. | |
2162 | */ | |
2163 | if (ret == -EAGAIN) { | |
2164 | locked_ref = NULL; | |
2165 | count++; | |
2166 | continue; | |
56bec294 CM |
2167 | } |
2168 | } | |
a28ec197 | 2169 | |
56bec294 CM |
2170 | /* |
2171 | * record the must insert reserved flag before we | |
2172 | * drop the spin lock. | |
2173 | */ | |
2174 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2175 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2176 | |
5d4f98a2 YZ |
2177 | extent_op = locked_ref->extent_op; |
2178 | locked_ref->extent_op = NULL; | |
2179 | ||
56bec294 CM |
2180 | /* |
2181 | * locked_ref is the head node, so we have to go one | |
2182 | * node back for any delayed ref updates | |
2183 | */ | |
56bec294 CM |
2184 | ref = select_delayed_ref(locked_ref); |
2185 | if (!ref) { | |
2186 | /* All delayed refs have been processed, Go ahead | |
2187 | * and send the head node to run_one_delayed_ref, | |
2188 | * so that any accounting fixes can happen | |
2189 | */ | |
2190 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2191 | |
2192 | if (extent_op && must_insert_reserved) { | |
2193 | kfree(extent_op); | |
2194 | extent_op = NULL; | |
2195 | } | |
2196 | ||
2197 | if (extent_op) { | |
2198 | spin_unlock(&delayed_refs->lock); | |
2199 | ||
2200 | ret = run_delayed_extent_op(trans, root, | |
2201 | ref, extent_op); | |
2202 | BUG_ON(ret); | |
2203 | kfree(extent_op); | |
2204 | ||
2205 | cond_resched(); | |
2206 | spin_lock(&delayed_refs->lock); | |
2207 | continue; | |
2208 | } | |
2209 | ||
c3e69d58 | 2210 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2211 | locked_ref = NULL; |
2212 | } | |
02217ed2 | 2213 | |
56bec294 CM |
2214 | ref->in_tree = 0; |
2215 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2216 | delayed_refs->num_entries--; | |
5d4f98a2 | 2217 | |
56bec294 | 2218 | spin_unlock(&delayed_refs->lock); |
925baedd | 2219 | |
5d4f98a2 | 2220 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 CM |
2221 | must_insert_reserved); |
2222 | BUG_ON(ret); | |
eb099670 | 2223 | |
5d4f98a2 YZ |
2224 | btrfs_put_delayed_ref(ref); |
2225 | kfree(extent_op); | |
c3e69d58 | 2226 | count++; |
5d4f98a2 | 2227 | |
c3e69d58 CM |
2228 | cond_resched(); |
2229 | spin_lock(&delayed_refs->lock); | |
2230 | } | |
2231 | return count; | |
2232 | } | |
2233 | ||
2234 | /* | |
2235 | * this starts processing the delayed reference count updates and | |
2236 | * extent insertions we have queued up so far. count can be | |
2237 | * 0, which means to process everything in the tree at the start | |
2238 | * of the run (but not newly added entries), or it can be some target | |
2239 | * number you'd like to process. | |
2240 | */ | |
2241 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2242 | struct btrfs_root *root, unsigned long count) | |
2243 | { | |
2244 | struct rb_node *node; | |
2245 | struct btrfs_delayed_ref_root *delayed_refs; | |
2246 | struct btrfs_delayed_ref_node *ref; | |
2247 | struct list_head cluster; | |
2248 | int ret; | |
2249 | int run_all = count == (unsigned long)-1; | |
2250 | int run_most = 0; | |
2251 | ||
2252 | if (root == root->fs_info->extent_root) | |
2253 | root = root->fs_info->tree_root; | |
2254 | ||
2255 | delayed_refs = &trans->transaction->delayed_refs; | |
2256 | INIT_LIST_HEAD(&cluster); | |
2257 | again: | |
2258 | spin_lock(&delayed_refs->lock); | |
2259 | if (count == 0) { | |
2260 | count = delayed_refs->num_entries * 2; | |
2261 | run_most = 1; | |
2262 | } | |
2263 | while (1) { | |
2264 | if (!(run_all || run_most) && | |
2265 | delayed_refs->num_heads_ready < 64) | |
2266 | break; | |
eb099670 | 2267 | |
56bec294 | 2268 | /* |
c3e69d58 CM |
2269 | * go find something we can process in the rbtree. We start at |
2270 | * the beginning of the tree, and then build a cluster | |
2271 | * of refs to process starting at the first one we are able to | |
2272 | * lock | |
56bec294 | 2273 | */ |
c3e69d58 CM |
2274 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2275 | delayed_refs->run_delayed_start); | |
2276 | if (ret) | |
56bec294 CM |
2277 | break; |
2278 | ||
c3e69d58 CM |
2279 | ret = run_clustered_refs(trans, root, &cluster); |
2280 | BUG_ON(ret < 0); | |
2281 | ||
2282 | count -= min_t(unsigned long, ret, count); | |
2283 | ||
2284 | if (count == 0) | |
2285 | break; | |
eb099670 | 2286 | } |
c3e69d58 | 2287 | |
56bec294 | 2288 | if (run_all) { |
56bec294 | 2289 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2290 | if (!node) |
56bec294 | 2291 | goto out; |
c3e69d58 | 2292 | count = (unsigned long)-1; |
e9d0b13b | 2293 | |
56bec294 CM |
2294 | while (node) { |
2295 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2296 | rb_node); | |
2297 | if (btrfs_delayed_ref_is_head(ref)) { | |
2298 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2299 | |
56bec294 CM |
2300 | head = btrfs_delayed_node_to_head(ref); |
2301 | atomic_inc(&ref->refs); | |
2302 | ||
2303 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2304 | /* |
2305 | * Mutex was contended, block until it's | |
2306 | * released and try again | |
2307 | */ | |
56bec294 CM |
2308 | mutex_lock(&head->mutex); |
2309 | mutex_unlock(&head->mutex); | |
2310 | ||
2311 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2312 | cond_resched(); |
56bec294 CM |
2313 | goto again; |
2314 | } | |
2315 | node = rb_next(node); | |
2316 | } | |
2317 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2318 | schedule_timeout(1); |
2319 | goto again; | |
5f39d397 | 2320 | } |
54aa1f4d | 2321 | out: |
c3e69d58 | 2322 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2323 | return 0; |
2324 | } | |
2325 | ||
5d4f98a2 YZ |
2326 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2327 | struct btrfs_root *root, | |
2328 | u64 bytenr, u64 num_bytes, u64 flags, | |
2329 | int is_data) | |
2330 | { | |
2331 | struct btrfs_delayed_extent_op *extent_op; | |
2332 | int ret; | |
2333 | ||
2334 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2335 | if (!extent_op) | |
2336 | return -ENOMEM; | |
2337 | ||
2338 | extent_op->flags_to_set = flags; | |
2339 | extent_op->update_flags = 1; | |
2340 | extent_op->update_key = 0; | |
2341 | extent_op->is_data = is_data ? 1 : 0; | |
2342 | ||
2343 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); | |
2344 | if (ret) | |
2345 | kfree(extent_op); | |
2346 | return ret; | |
2347 | } | |
2348 | ||
2349 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2350 | struct btrfs_root *root, | |
2351 | struct btrfs_path *path, | |
2352 | u64 objectid, u64 offset, u64 bytenr) | |
2353 | { | |
2354 | struct btrfs_delayed_ref_head *head; | |
2355 | struct btrfs_delayed_ref_node *ref; | |
2356 | struct btrfs_delayed_data_ref *data_ref; | |
2357 | struct btrfs_delayed_ref_root *delayed_refs; | |
2358 | struct rb_node *node; | |
2359 | int ret = 0; | |
2360 | ||
2361 | ret = -ENOENT; | |
2362 | delayed_refs = &trans->transaction->delayed_refs; | |
2363 | spin_lock(&delayed_refs->lock); | |
2364 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2365 | if (!head) | |
2366 | goto out; | |
2367 | ||
2368 | if (!mutex_trylock(&head->mutex)) { | |
2369 | atomic_inc(&head->node.refs); | |
2370 | spin_unlock(&delayed_refs->lock); | |
2371 | ||
b3b4aa74 | 2372 | btrfs_release_path(path); |
5d4f98a2 | 2373 | |
8cc33e5c DS |
2374 | /* |
2375 | * Mutex was contended, block until it's released and let | |
2376 | * caller try again | |
2377 | */ | |
5d4f98a2 YZ |
2378 | mutex_lock(&head->mutex); |
2379 | mutex_unlock(&head->mutex); | |
2380 | btrfs_put_delayed_ref(&head->node); | |
2381 | return -EAGAIN; | |
2382 | } | |
2383 | ||
2384 | node = rb_prev(&head->node.rb_node); | |
2385 | if (!node) | |
2386 | goto out_unlock; | |
2387 | ||
2388 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2389 | ||
2390 | if (ref->bytenr != bytenr) | |
2391 | goto out_unlock; | |
2392 | ||
2393 | ret = 1; | |
2394 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2395 | goto out_unlock; | |
2396 | ||
2397 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2398 | ||
2399 | node = rb_prev(node); | |
2400 | if (node) { | |
2401 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2402 | if (ref->bytenr == bytenr) | |
2403 | goto out_unlock; | |
2404 | } | |
2405 | ||
2406 | if (data_ref->root != root->root_key.objectid || | |
2407 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2408 | goto out_unlock; | |
2409 | ||
2410 | ret = 0; | |
2411 | out_unlock: | |
2412 | mutex_unlock(&head->mutex); | |
2413 | out: | |
2414 | spin_unlock(&delayed_refs->lock); | |
2415 | return ret; | |
2416 | } | |
2417 | ||
2418 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2419 | struct btrfs_root *root, | |
2420 | struct btrfs_path *path, | |
2421 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2422 | { |
2423 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2424 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2425 | struct btrfs_extent_data_ref *ref; |
2426 | struct btrfs_extent_inline_ref *iref; | |
2427 | struct btrfs_extent_item *ei; | |
f321e491 | 2428 | struct btrfs_key key; |
5d4f98a2 | 2429 | u32 item_size; |
be20aa9d | 2430 | int ret; |
925baedd | 2431 | |
be20aa9d | 2432 | key.objectid = bytenr; |
31840ae1 | 2433 | key.offset = (u64)-1; |
f321e491 | 2434 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2435 | |
be20aa9d CM |
2436 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2437 | if (ret < 0) | |
2438 | goto out; | |
2439 | BUG_ON(ret == 0); | |
80ff3856 YZ |
2440 | |
2441 | ret = -ENOENT; | |
2442 | if (path->slots[0] == 0) | |
31840ae1 | 2443 | goto out; |
be20aa9d | 2444 | |
31840ae1 | 2445 | path->slots[0]--; |
f321e491 | 2446 | leaf = path->nodes[0]; |
5d4f98a2 | 2447 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2448 | |
5d4f98a2 | 2449 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2450 | goto out; |
f321e491 | 2451 | |
5d4f98a2 YZ |
2452 | ret = 1; |
2453 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2454 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2455 | if (item_size < sizeof(*ei)) { | |
2456 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2457 | goto out; | |
2458 | } | |
2459 | #endif | |
2460 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2461 | |
5d4f98a2 YZ |
2462 | if (item_size != sizeof(*ei) + |
2463 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2464 | goto out; | |
be20aa9d | 2465 | |
5d4f98a2 YZ |
2466 | if (btrfs_extent_generation(leaf, ei) <= |
2467 | btrfs_root_last_snapshot(&root->root_item)) | |
2468 | goto out; | |
2469 | ||
2470 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2471 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2472 | BTRFS_EXTENT_DATA_REF_KEY) | |
2473 | goto out; | |
2474 | ||
2475 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2476 | if (btrfs_extent_refs(leaf, ei) != | |
2477 | btrfs_extent_data_ref_count(leaf, ref) || | |
2478 | btrfs_extent_data_ref_root(leaf, ref) != | |
2479 | root->root_key.objectid || | |
2480 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2481 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2482 | goto out; | |
2483 | ||
2484 | ret = 0; | |
2485 | out: | |
2486 | return ret; | |
2487 | } | |
2488 | ||
2489 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2490 | struct btrfs_root *root, | |
2491 | u64 objectid, u64 offset, u64 bytenr) | |
2492 | { | |
2493 | struct btrfs_path *path; | |
2494 | int ret; | |
2495 | int ret2; | |
2496 | ||
2497 | path = btrfs_alloc_path(); | |
2498 | if (!path) | |
2499 | return -ENOENT; | |
2500 | ||
2501 | do { | |
2502 | ret = check_committed_ref(trans, root, path, objectid, | |
2503 | offset, bytenr); | |
2504 | if (ret && ret != -ENOENT) | |
f321e491 | 2505 | goto out; |
80ff3856 | 2506 | |
5d4f98a2 YZ |
2507 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2508 | offset, bytenr); | |
2509 | } while (ret2 == -EAGAIN); | |
2510 | ||
2511 | if (ret2 && ret2 != -ENOENT) { | |
2512 | ret = ret2; | |
2513 | goto out; | |
f321e491 | 2514 | } |
5d4f98a2 YZ |
2515 | |
2516 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2517 | ret = 0; | |
be20aa9d | 2518 | out: |
80ff3856 | 2519 | btrfs_free_path(path); |
f0486c68 YZ |
2520 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2521 | WARN_ON(ret > 0); | |
f321e491 | 2522 | return ret; |
be20aa9d | 2523 | } |
c5739bba | 2524 | |
5d4f98a2 | 2525 | #if 0 |
31840ae1 ZY |
2526 | int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2527 | struct extent_buffer *buf, u32 nr_extents) | |
02217ed2 | 2528 | { |
5f39d397 | 2529 | struct btrfs_key key; |
6407bf6d | 2530 | struct btrfs_file_extent_item *fi; |
e4657689 ZY |
2531 | u64 root_gen; |
2532 | u32 nritems; | |
02217ed2 | 2533 | int i; |
db94535d | 2534 | int level; |
31840ae1 | 2535 | int ret = 0; |
e4657689 | 2536 | int shared = 0; |
a28ec197 | 2537 | |
3768f368 | 2538 | if (!root->ref_cows) |
a28ec197 | 2539 | return 0; |
5f39d397 | 2540 | |
e4657689 ZY |
2541 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
2542 | shared = 0; | |
2543 | root_gen = root->root_key.offset; | |
2544 | } else { | |
2545 | shared = 1; | |
2546 | root_gen = trans->transid - 1; | |
2547 | } | |
2548 | ||
db94535d | 2549 | level = btrfs_header_level(buf); |
5f39d397 | 2550 | nritems = btrfs_header_nritems(buf); |
4a096752 | 2551 | |
31840ae1 | 2552 | if (level == 0) { |
31153d81 YZ |
2553 | struct btrfs_leaf_ref *ref; |
2554 | struct btrfs_extent_info *info; | |
2555 | ||
31840ae1 | 2556 | ref = btrfs_alloc_leaf_ref(root, nr_extents); |
31153d81 | 2557 | if (!ref) { |
31840ae1 | 2558 | ret = -ENOMEM; |
31153d81 YZ |
2559 | goto out; |
2560 | } | |
2561 | ||
e4657689 | 2562 | ref->root_gen = root_gen; |
31153d81 YZ |
2563 | ref->bytenr = buf->start; |
2564 | ref->owner = btrfs_header_owner(buf); | |
2565 | ref->generation = btrfs_header_generation(buf); | |
31840ae1 | 2566 | ref->nritems = nr_extents; |
31153d81 | 2567 | info = ref->extents; |
bcc63abb | 2568 | |
31840ae1 | 2569 | for (i = 0; nr_extents > 0 && i < nritems; i++) { |
31153d81 YZ |
2570 | u64 disk_bytenr; |
2571 | btrfs_item_key_to_cpu(buf, &key, i); | |
2572 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
2573 | continue; | |
2574 | fi = btrfs_item_ptr(buf, i, | |
2575 | struct btrfs_file_extent_item); | |
2576 | if (btrfs_file_extent_type(buf, fi) == | |
2577 | BTRFS_FILE_EXTENT_INLINE) | |
2578 | continue; | |
2579 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2580 | if (disk_bytenr == 0) | |
2581 | continue; | |
2582 | ||
2583 | info->bytenr = disk_bytenr; | |
2584 | info->num_bytes = | |
2585 | btrfs_file_extent_disk_num_bytes(buf, fi); | |
2586 | info->objectid = key.objectid; | |
2587 | info->offset = key.offset; | |
2588 | info++; | |
2589 | } | |
2590 | ||
e4657689 | 2591 | ret = btrfs_add_leaf_ref(root, ref, shared); |
5b84e8d6 YZ |
2592 | if (ret == -EEXIST && shared) { |
2593 | struct btrfs_leaf_ref *old; | |
2594 | old = btrfs_lookup_leaf_ref(root, ref->bytenr); | |
2595 | BUG_ON(!old); | |
2596 | btrfs_remove_leaf_ref(root, old); | |
2597 | btrfs_free_leaf_ref(root, old); | |
2598 | ret = btrfs_add_leaf_ref(root, ref, shared); | |
2599 | } | |
31153d81 | 2600 | WARN_ON(ret); |
bcc63abb | 2601 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
2602 | } |
2603 | out: | |
31840ae1 ZY |
2604 | return ret; |
2605 | } | |
2606 | ||
b7a9f29f CM |
2607 | /* when a block goes through cow, we update the reference counts of |
2608 | * everything that block points to. The internal pointers of the block | |
2609 | * can be in just about any order, and it is likely to have clusters of | |
2610 | * things that are close together and clusters of things that are not. | |
2611 | * | |
2612 | * To help reduce the seeks that come with updating all of these reference | |
2613 | * counts, sort them by byte number before actual updates are done. | |
2614 | * | |
2615 | * struct refsort is used to match byte number to slot in the btree block. | |
2616 | * we sort based on the byte number and then use the slot to actually | |
2617 | * find the item. | |
bd56b302 CM |
2618 | * |
2619 | * struct refsort is smaller than strcut btrfs_item and smaller than | |
2620 | * struct btrfs_key_ptr. Since we're currently limited to the page size | |
2621 | * for a btree block, there's no way for a kmalloc of refsorts for a | |
2622 | * single node to be bigger than a page. | |
b7a9f29f CM |
2623 | */ |
2624 | struct refsort { | |
2625 | u64 bytenr; | |
2626 | u32 slot; | |
2627 | }; | |
2628 | ||
2629 | /* | |
2630 | * for passing into sort() | |
2631 | */ | |
2632 | static int refsort_cmp(const void *a_void, const void *b_void) | |
2633 | { | |
2634 | const struct refsort *a = a_void; | |
2635 | const struct refsort *b = b_void; | |
2636 | ||
2637 | if (a->bytenr < b->bytenr) | |
2638 | return -1; | |
2639 | if (a->bytenr > b->bytenr) | |
2640 | return 1; | |
2641 | return 0; | |
2642 | } | |
5d4f98a2 | 2643 | #endif |
b7a9f29f | 2644 | |
5d4f98a2 | 2645 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2646 | struct btrfs_root *root, |
5d4f98a2 YZ |
2647 | struct extent_buffer *buf, |
2648 | int full_backref, int inc) | |
31840ae1 ZY |
2649 | { |
2650 | u64 bytenr; | |
5d4f98a2 YZ |
2651 | u64 num_bytes; |
2652 | u64 parent; | |
31840ae1 | 2653 | u64 ref_root; |
31840ae1 | 2654 | u32 nritems; |
31840ae1 ZY |
2655 | struct btrfs_key key; |
2656 | struct btrfs_file_extent_item *fi; | |
2657 | int i; | |
2658 | int level; | |
2659 | int ret = 0; | |
31840ae1 | 2660 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
5d4f98a2 | 2661 | u64, u64, u64, u64, u64, u64); |
31840ae1 ZY |
2662 | |
2663 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2664 | nritems = btrfs_header_nritems(buf); |
2665 | level = btrfs_header_level(buf); | |
2666 | ||
5d4f98a2 YZ |
2667 | if (!root->ref_cows && level == 0) |
2668 | return 0; | |
31840ae1 | 2669 | |
5d4f98a2 YZ |
2670 | if (inc) |
2671 | process_func = btrfs_inc_extent_ref; | |
2672 | else | |
2673 | process_func = btrfs_free_extent; | |
31840ae1 | 2674 | |
5d4f98a2 YZ |
2675 | if (full_backref) |
2676 | parent = buf->start; | |
2677 | else | |
2678 | parent = 0; | |
2679 | ||
2680 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2681 | if (level == 0) { |
5d4f98a2 | 2682 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2683 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2684 | continue; | |
5d4f98a2 | 2685 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2686 | struct btrfs_file_extent_item); |
2687 | if (btrfs_file_extent_type(buf, fi) == | |
2688 | BTRFS_FILE_EXTENT_INLINE) | |
2689 | continue; | |
2690 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2691 | if (bytenr == 0) | |
2692 | continue; | |
5d4f98a2 YZ |
2693 | |
2694 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2695 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2696 | ret = process_func(trans, root, bytenr, num_bytes, | |
2697 | parent, ref_root, key.objectid, | |
2698 | key.offset); | |
31840ae1 ZY |
2699 | if (ret) |
2700 | goto fail; | |
2701 | } else { | |
5d4f98a2 YZ |
2702 | bytenr = btrfs_node_blockptr(buf, i); |
2703 | num_bytes = btrfs_level_size(root, level - 1); | |
2704 | ret = process_func(trans, root, bytenr, num_bytes, | |
2705 | parent, ref_root, level - 1, 0); | |
31840ae1 ZY |
2706 | if (ret) |
2707 | goto fail; | |
2708 | } | |
2709 | } | |
2710 | return 0; | |
2711 | fail: | |
5d4f98a2 YZ |
2712 | BUG(); |
2713 | return ret; | |
2714 | } | |
2715 | ||
2716 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2717 | struct extent_buffer *buf, int full_backref) | |
2718 | { | |
2719 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); | |
2720 | } | |
2721 | ||
2722 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2723 | struct extent_buffer *buf, int full_backref) | |
2724 | { | |
2725 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); | |
31840ae1 ZY |
2726 | } |
2727 | ||
9078a3e1 CM |
2728 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2729 | struct btrfs_root *root, | |
2730 | struct btrfs_path *path, | |
2731 | struct btrfs_block_group_cache *cache) | |
2732 | { | |
2733 | int ret; | |
9078a3e1 | 2734 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2735 | unsigned long bi; |
2736 | struct extent_buffer *leaf; | |
9078a3e1 | 2737 | |
9078a3e1 | 2738 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2739 | if (ret < 0) |
2740 | goto fail; | |
9078a3e1 | 2741 | BUG_ON(ret); |
5f39d397 CM |
2742 | |
2743 | leaf = path->nodes[0]; | |
2744 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2745 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2746 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2747 | btrfs_release_path(path); |
54aa1f4d | 2748 | fail: |
9078a3e1 CM |
2749 | if (ret) |
2750 | return ret; | |
9078a3e1 CM |
2751 | return 0; |
2752 | ||
2753 | } | |
2754 | ||
4a8c9a62 YZ |
2755 | static struct btrfs_block_group_cache * |
2756 | next_block_group(struct btrfs_root *root, | |
2757 | struct btrfs_block_group_cache *cache) | |
2758 | { | |
2759 | struct rb_node *node; | |
2760 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2761 | node = rb_next(&cache->cache_node); | |
2762 | btrfs_put_block_group(cache); | |
2763 | if (node) { | |
2764 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2765 | cache_node); | |
11dfe35a | 2766 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2767 | } else |
2768 | cache = NULL; | |
2769 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2770 | return cache; | |
2771 | } | |
2772 | ||
0af3d00b JB |
2773 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2774 | struct btrfs_trans_handle *trans, | |
2775 | struct btrfs_path *path) | |
2776 | { | |
2777 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2778 | struct inode *inode = NULL; | |
2779 | u64 alloc_hint = 0; | |
2b20982e | 2780 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2781 | int num_pages = 0; |
2782 | int retries = 0; | |
2783 | int ret = 0; | |
2784 | ||
2785 | /* | |
2786 | * If this block group is smaller than 100 megs don't bother caching the | |
2787 | * block group. | |
2788 | */ | |
2789 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2790 | spin_lock(&block_group->lock); | |
2791 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2792 | spin_unlock(&block_group->lock); | |
2793 | return 0; | |
2794 | } | |
2795 | ||
2796 | again: | |
2797 | inode = lookup_free_space_inode(root, block_group, path); | |
2798 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2799 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2800 | btrfs_release_path(path); |
0af3d00b JB |
2801 | goto out; |
2802 | } | |
2803 | ||
2804 | if (IS_ERR(inode)) { | |
2805 | BUG_ON(retries); | |
2806 | retries++; | |
2807 | ||
2808 | if (block_group->ro) | |
2809 | goto out_free; | |
2810 | ||
2811 | ret = create_free_space_inode(root, trans, block_group, path); | |
2812 | if (ret) | |
2813 | goto out_free; | |
2814 | goto again; | |
2815 | } | |
2816 | ||
2817 | /* | |
2818 | * We want to set the generation to 0, that way if anything goes wrong | |
2819 | * from here on out we know not to trust this cache when we load up next | |
2820 | * time. | |
2821 | */ | |
2822 | BTRFS_I(inode)->generation = 0; | |
2823 | ret = btrfs_update_inode(trans, root, inode); | |
2824 | WARN_ON(ret); | |
2825 | ||
2826 | if (i_size_read(inode) > 0) { | |
2827 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2828 | inode); | |
2829 | if (ret) | |
2830 | goto out_put; | |
2831 | } | |
2832 | ||
2833 | spin_lock(&block_group->lock); | |
2834 | if (block_group->cached != BTRFS_CACHE_FINISHED) { | |
2b20982e JB |
2835 | /* We're not cached, don't bother trying to write stuff out */ |
2836 | dcs = BTRFS_DC_WRITTEN; | |
0af3d00b JB |
2837 | spin_unlock(&block_group->lock); |
2838 | goto out_put; | |
2839 | } | |
2840 | spin_unlock(&block_group->lock); | |
2841 | ||
2842 | num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024); | |
2843 | if (!num_pages) | |
2844 | num_pages = 1; | |
2845 | ||
2846 | /* | |
2847 | * Just to make absolutely sure we have enough space, we're going to | |
2848 | * preallocate 12 pages worth of space for each block group. In | |
2849 | * practice we ought to use at most 8, but we need extra space so we can | |
2850 | * add our header and have a terminator between the extents and the | |
2851 | * bitmaps. | |
2852 | */ | |
2853 | num_pages *= 16; | |
2854 | num_pages *= PAGE_CACHE_SIZE; | |
2855 | ||
2856 | ret = btrfs_check_data_free_space(inode, num_pages); | |
2857 | if (ret) | |
2858 | goto out_put; | |
2859 | ||
2860 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2861 | num_pages, num_pages, | |
2862 | &alloc_hint); | |
2b20982e JB |
2863 | if (!ret) |
2864 | dcs = BTRFS_DC_SETUP; | |
0af3d00b JB |
2865 | btrfs_free_reserved_data_space(inode, num_pages); |
2866 | out_put: | |
2867 | iput(inode); | |
2868 | out_free: | |
b3b4aa74 | 2869 | btrfs_release_path(path); |
0af3d00b JB |
2870 | out: |
2871 | spin_lock(&block_group->lock); | |
2b20982e | 2872 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
2873 | spin_unlock(&block_group->lock); |
2874 | ||
2875 | return ret; | |
2876 | } | |
2877 | ||
96b5179d CM |
2878 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2879 | struct btrfs_root *root) | |
9078a3e1 | 2880 | { |
4a8c9a62 | 2881 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2882 | int err = 0; |
9078a3e1 | 2883 | struct btrfs_path *path; |
96b5179d | 2884 | u64 last = 0; |
9078a3e1 CM |
2885 | |
2886 | path = btrfs_alloc_path(); | |
2887 | if (!path) | |
2888 | return -ENOMEM; | |
2889 | ||
0af3d00b JB |
2890 | again: |
2891 | while (1) { | |
2892 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2893 | while (cache) { | |
2894 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2895 | break; | |
2896 | cache = next_block_group(root, cache); | |
2897 | } | |
2898 | if (!cache) { | |
2899 | if (last == 0) | |
2900 | break; | |
2901 | last = 0; | |
2902 | continue; | |
2903 | } | |
2904 | err = cache_save_setup(cache, trans, path); | |
2905 | last = cache->key.objectid + cache->key.offset; | |
2906 | btrfs_put_block_group(cache); | |
2907 | } | |
2908 | ||
d397712b | 2909 | while (1) { |
4a8c9a62 YZ |
2910 | if (last == 0) { |
2911 | err = btrfs_run_delayed_refs(trans, root, | |
2912 | (unsigned long)-1); | |
2913 | BUG_ON(err); | |
0f9dd46c | 2914 | } |
54aa1f4d | 2915 | |
4a8c9a62 YZ |
2916 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2917 | while (cache) { | |
0af3d00b JB |
2918 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
2919 | btrfs_put_block_group(cache); | |
2920 | goto again; | |
2921 | } | |
2922 | ||
4a8c9a62 YZ |
2923 | if (cache->dirty) |
2924 | break; | |
2925 | cache = next_block_group(root, cache); | |
2926 | } | |
2927 | if (!cache) { | |
2928 | if (last == 0) | |
2929 | break; | |
2930 | last = 0; | |
2931 | continue; | |
2932 | } | |
0f9dd46c | 2933 | |
0cb59c99 JB |
2934 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
2935 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 2936 | cache->dirty = 0; |
4a8c9a62 | 2937 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 2938 | |
4a8c9a62 YZ |
2939 | err = write_one_cache_group(trans, root, path, cache); |
2940 | BUG_ON(err); | |
2941 | btrfs_put_block_group(cache); | |
9078a3e1 | 2942 | } |
4a8c9a62 | 2943 | |
0cb59c99 JB |
2944 | while (1) { |
2945 | /* | |
2946 | * I don't think this is needed since we're just marking our | |
2947 | * preallocated extent as written, but just in case it can't | |
2948 | * hurt. | |
2949 | */ | |
2950 | if (last == 0) { | |
2951 | err = btrfs_run_delayed_refs(trans, root, | |
2952 | (unsigned long)-1); | |
2953 | BUG_ON(err); | |
2954 | } | |
2955 | ||
2956 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2957 | while (cache) { | |
2958 | /* | |
2959 | * Really this shouldn't happen, but it could if we | |
2960 | * couldn't write the entire preallocated extent and | |
2961 | * splitting the extent resulted in a new block. | |
2962 | */ | |
2963 | if (cache->dirty) { | |
2964 | btrfs_put_block_group(cache); | |
2965 | goto again; | |
2966 | } | |
2967 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
2968 | break; | |
2969 | cache = next_block_group(root, cache); | |
2970 | } | |
2971 | if (!cache) { | |
2972 | if (last == 0) | |
2973 | break; | |
2974 | last = 0; | |
2975 | continue; | |
2976 | } | |
2977 | ||
2978 | btrfs_write_out_cache(root, trans, cache, path); | |
2979 | ||
2980 | /* | |
2981 | * If we didn't have an error then the cache state is still | |
2982 | * NEED_WRITE, so we can set it to WRITTEN. | |
2983 | */ | |
2984 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
2985 | cache->disk_cache_state = BTRFS_DC_WRITTEN; | |
2986 | last = cache->key.objectid + cache->key.offset; | |
2987 | btrfs_put_block_group(cache); | |
2988 | } | |
2989 | ||
9078a3e1 | 2990 | btrfs_free_path(path); |
4a8c9a62 | 2991 | return 0; |
9078a3e1 CM |
2992 | } |
2993 | ||
d2fb3437 YZ |
2994 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
2995 | { | |
2996 | struct btrfs_block_group_cache *block_group; | |
2997 | int readonly = 0; | |
2998 | ||
2999 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3000 | if (!block_group || block_group->ro) | |
3001 | readonly = 1; | |
3002 | if (block_group) | |
fa9c0d79 | 3003 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3004 | return readonly; |
3005 | } | |
3006 | ||
593060d7 CM |
3007 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3008 | u64 total_bytes, u64 bytes_used, | |
3009 | struct btrfs_space_info **space_info) | |
3010 | { | |
3011 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3012 | int i; |
3013 | int factor; | |
3014 | ||
3015 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3016 | BTRFS_BLOCK_GROUP_RAID10)) | |
3017 | factor = 2; | |
3018 | else | |
3019 | factor = 1; | |
593060d7 CM |
3020 | |
3021 | found = __find_space_info(info, flags); | |
3022 | if (found) { | |
25179201 | 3023 | spin_lock(&found->lock); |
593060d7 | 3024 | found->total_bytes += total_bytes; |
89a55897 | 3025 | found->disk_total += total_bytes * factor; |
593060d7 | 3026 | found->bytes_used += bytes_used; |
b742bb82 | 3027 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3028 | found->full = 0; |
25179201 | 3029 | spin_unlock(&found->lock); |
593060d7 CM |
3030 | *space_info = found; |
3031 | return 0; | |
3032 | } | |
c146afad | 3033 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3034 | if (!found) |
3035 | return -ENOMEM; | |
3036 | ||
b742bb82 YZ |
3037 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3038 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3039 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3040 | spin_lock_init(&found->lock); |
b742bb82 YZ |
3041 | found->flags = flags & (BTRFS_BLOCK_GROUP_DATA | |
3042 | BTRFS_BLOCK_GROUP_SYSTEM | | |
3043 | BTRFS_BLOCK_GROUP_METADATA); | |
593060d7 | 3044 | found->total_bytes = total_bytes; |
89a55897 | 3045 | found->disk_total = total_bytes * factor; |
593060d7 | 3046 | found->bytes_used = bytes_used; |
b742bb82 | 3047 | found->disk_used = bytes_used * factor; |
593060d7 | 3048 | found->bytes_pinned = 0; |
e8569813 | 3049 | found->bytes_reserved = 0; |
c146afad | 3050 | found->bytes_readonly = 0; |
f0486c68 | 3051 | found->bytes_may_use = 0; |
593060d7 | 3052 | found->full = 0; |
0e4f8f88 | 3053 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3054 | found->chunk_alloc = 0; |
593060d7 | 3055 | *space_info = found; |
4184ea7f | 3056 | list_add_rcu(&found->list, &info->space_info); |
817d52f8 | 3057 | atomic_set(&found->caching_threads, 0); |
593060d7 CM |
3058 | return 0; |
3059 | } | |
3060 | ||
8790d502 CM |
3061 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3062 | { | |
3063 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 3064 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 3065 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 3066 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
3067 | if (extra_flags) { |
3068 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3069 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3070 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3071 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3072 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3073 | fs_info->avail_system_alloc_bits |= extra_flags; | |
3074 | } | |
3075 | } | |
593060d7 | 3076 | |
2b82032c | 3077 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3078 | { |
cd02dca5 CM |
3079 | /* |
3080 | * we add in the count of missing devices because we want | |
3081 | * to make sure that any RAID levels on a degraded FS | |
3082 | * continue to be honored. | |
3083 | */ | |
3084 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3085 | root->fs_info->fs_devices->missing_devices; | |
a061fc8d CM |
3086 | |
3087 | if (num_devices == 1) | |
3088 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
3089 | if (num_devices < 4) | |
3090 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3091 | ||
ec44a35c CM |
3092 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
3093 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 3094 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 3095 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 3096 | } |
ec44a35c CM |
3097 | |
3098 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 3099 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 3100 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 3101 | } |
ec44a35c CM |
3102 | |
3103 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
3104 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
3105 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
3106 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
3107 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
3108 | return flags; | |
3109 | } | |
3110 | ||
b742bb82 | 3111 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3112 | { |
b742bb82 YZ |
3113 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3114 | flags |= root->fs_info->avail_data_alloc_bits & | |
3115 | root->fs_info->data_alloc_profile; | |
3116 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3117 | flags |= root->fs_info->avail_system_alloc_bits & | |
3118 | root->fs_info->system_alloc_profile; | |
3119 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3120 | flags |= root->fs_info->avail_metadata_alloc_bits & | |
3121 | root->fs_info->metadata_alloc_profile; | |
3122 | return btrfs_reduce_alloc_profile(root, flags); | |
6a63209f JB |
3123 | } |
3124 | ||
6d07bcec | 3125 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3126 | { |
b742bb82 | 3127 | u64 flags; |
9ed74f2d | 3128 | |
b742bb82 YZ |
3129 | if (data) |
3130 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3131 | else if (root == root->fs_info->chunk_root) | |
3132 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3133 | else |
b742bb82 | 3134 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3135 | |
b742bb82 | 3136 | return get_alloc_profile(root, flags); |
6a63209f | 3137 | } |
9ed74f2d | 3138 | |
6a63209f JB |
3139 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) |
3140 | { | |
6a63209f | 3141 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, |
f0486c68 | 3142 | BTRFS_BLOCK_GROUP_DATA); |
9ed74f2d JB |
3143 | } |
3144 | ||
6a63209f | 3145 | /* |
6a63209f JB |
3146 | * This will check the space that the inode allocates from to make sure we have |
3147 | * enough space for bytes. | |
6a63209f | 3148 | */ |
0ca1f7ce | 3149 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3150 | { |
6a63209f | 3151 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3152 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ab6e2410 | 3153 | u64 used; |
0af3d00b | 3154 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3155 | |
6a63209f JB |
3156 | /* make sure bytes are sectorsize aligned */ |
3157 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3158 | |
0af3d00b JB |
3159 | if (root == root->fs_info->tree_root) { |
3160 | alloc_chunk = 0; | |
3161 | committed = 1; | |
3162 | } | |
3163 | ||
6a63209f | 3164 | data_sinfo = BTRFS_I(inode)->space_info; |
33b4d47f CM |
3165 | if (!data_sinfo) |
3166 | goto alloc; | |
9ed74f2d | 3167 | |
6a63209f JB |
3168 | again: |
3169 | /* make sure we have enough space to handle the data first */ | |
3170 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3171 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3172 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3173 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3174 | |
3175 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3176 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3177 | |
6a63209f JB |
3178 | /* |
3179 | * if we don't have enough free bytes in this space then we need | |
3180 | * to alloc a new chunk. | |
3181 | */ | |
0af3d00b | 3182 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3183 | u64 alloc_target; |
9ed74f2d | 3184 | |
0e4f8f88 | 3185 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3186 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3187 | alloc: |
6a63209f | 3188 | alloc_target = btrfs_get_alloc_profile(root, 1); |
a22285a6 YZ |
3189 | trans = btrfs_join_transaction(root, 1); |
3190 | if (IS_ERR(trans)) | |
3191 | return PTR_ERR(trans); | |
9ed74f2d | 3192 | |
6a63209f JB |
3193 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
3194 | bytes + 2 * 1024 * 1024, | |
0e4f8f88 CM |
3195 | alloc_target, |
3196 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3197 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3198 | if (ret < 0) { |
3199 | if (ret != -ENOSPC) | |
3200 | return ret; | |
3201 | else | |
3202 | goto commit_trans; | |
3203 | } | |
9ed74f2d | 3204 | |
33b4d47f CM |
3205 | if (!data_sinfo) { |
3206 | btrfs_set_inode_space_info(root, inode); | |
3207 | data_sinfo = BTRFS_I(inode)->space_info; | |
3208 | } | |
6a63209f JB |
3209 | goto again; |
3210 | } | |
3211 | spin_unlock(&data_sinfo->lock); | |
6a63209f | 3212 | |
4e06bdd6 | 3213 | /* commit the current transaction and try again */ |
d52a5b5f | 3214 | commit_trans: |
dd7e0b7b | 3215 | if (!committed && !root->fs_info->open_ioctl_trans) { |
4e06bdd6 JB |
3216 | committed = 1; |
3217 | trans = btrfs_join_transaction(root, 1); | |
a22285a6 YZ |
3218 | if (IS_ERR(trans)) |
3219 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3220 | ret = btrfs_commit_transaction(trans, root); |
3221 | if (ret) | |
3222 | return ret; | |
3223 | goto again; | |
3224 | } | |
9ed74f2d | 3225 | |
933b585f | 3226 | #if 0 /* I hope we never need this code again, just in case */ |
8929ecfa YZ |
3227 | printk(KERN_ERR "no space left, need %llu, %llu bytes_used, " |
3228 | "%llu bytes_reserved, " "%llu bytes_pinned, " | |
3229 | "%llu bytes_readonly, %llu may use %llu total\n", | |
3230 | (unsigned long long)bytes, | |
21380931 JB |
3231 | (unsigned long long)data_sinfo->bytes_used, |
3232 | (unsigned long long)data_sinfo->bytes_reserved, | |
3233 | (unsigned long long)data_sinfo->bytes_pinned, | |
3234 | (unsigned long long)data_sinfo->bytes_readonly, | |
3235 | (unsigned long long)data_sinfo->bytes_may_use, | |
3236 | (unsigned long long)data_sinfo->total_bytes); | |
933b585f | 3237 | #endif |
6a63209f JB |
3238 | return -ENOSPC; |
3239 | } | |
3240 | data_sinfo->bytes_may_use += bytes; | |
3241 | BTRFS_I(inode)->reserved_bytes += bytes; | |
3242 | spin_unlock(&data_sinfo->lock); | |
6a63209f | 3243 | |
9ed74f2d | 3244 | return 0; |
9ed74f2d | 3245 | } |
6a63209f | 3246 | |
6a63209f | 3247 | /* |
0ca1f7ce YZ |
3248 | * called when we are clearing an delalloc extent from the |
3249 | * inode's io_tree or there was an error for whatever reason | |
3250 | * after calling btrfs_check_data_free_space | |
6a63209f | 3251 | */ |
0ca1f7ce | 3252 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3253 | { |
0ca1f7ce | 3254 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3255 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3256 | |
6a63209f JB |
3257 | /* make sure bytes are sectorsize aligned */ |
3258 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3259 | |
6a63209f JB |
3260 | data_sinfo = BTRFS_I(inode)->space_info; |
3261 | spin_lock(&data_sinfo->lock); | |
3262 | data_sinfo->bytes_may_use -= bytes; | |
3263 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
3264 | spin_unlock(&data_sinfo->lock); | |
e3ccfa98 JB |
3265 | } |
3266 | ||
97e728d4 | 3267 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3268 | { |
97e728d4 JB |
3269 | struct list_head *head = &info->space_info; |
3270 | struct btrfs_space_info *found; | |
e3ccfa98 | 3271 | |
97e728d4 JB |
3272 | rcu_read_lock(); |
3273 | list_for_each_entry_rcu(found, head, list) { | |
3274 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3275 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3276 | } |
97e728d4 | 3277 | rcu_read_unlock(); |
e3ccfa98 JB |
3278 | } |
3279 | ||
e5bc2458 | 3280 | static int should_alloc_chunk(struct btrfs_root *root, |
0e4f8f88 CM |
3281 | struct btrfs_space_info *sinfo, u64 alloc_bytes, |
3282 | int force) | |
32c00aff | 3283 | { |
424499db | 3284 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3285 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3286 | u64 thresh; |
e3ccfa98 | 3287 | |
0e4f8f88 CM |
3288 | if (force == CHUNK_ALLOC_FORCE) |
3289 | return 1; | |
3290 | ||
3291 | /* | |
3292 | * in limited mode, we want to have some free space up to | |
3293 | * about 1% of the FS size. | |
3294 | */ | |
3295 | if (force == CHUNK_ALLOC_LIMITED) { | |
3296 | thresh = btrfs_super_total_bytes(&root->fs_info->super_copy); | |
3297 | thresh = max_t(u64, 64 * 1024 * 1024, | |
3298 | div_factor_fine(thresh, 1)); | |
3299 | ||
3300 | if (num_bytes - num_allocated < thresh) | |
3301 | return 1; | |
3302 | } | |
3303 | ||
3304 | /* | |
3305 | * we have two similar checks here, one based on percentage | |
3306 | * and once based on a hard number of 256MB. The idea | |
3307 | * is that if we have a good amount of free | |
3308 | * room, don't allocate a chunk. A good mount is | |
3309 | * less than 80% utilized of the chunks we have allocated, | |
3310 | * or more than 256MB free | |
3311 | */ | |
3312 | if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes) | |
424499db | 3313 | return 0; |
e3ccfa98 | 3314 | |
0e4f8f88 | 3315 | if (num_allocated + alloc_bytes < div_factor(num_bytes, 8)) |
424499db | 3316 | return 0; |
32c00aff | 3317 | |
e5bc2458 | 3318 | thresh = btrfs_super_total_bytes(&root->fs_info->super_copy); |
0e4f8f88 CM |
3319 | |
3320 | /* 256MB or 5% of the FS */ | |
e5bc2458 CM |
3321 | thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5)); |
3322 | ||
3323 | if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3)) | |
14ed0ca6 | 3324 | return 0; |
424499db | 3325 | return 1; |
32c00aff JB |
3326 | } |
3327 | ||
6324fbf3 CM |
3328 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3329 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3330 | u64 flags, int force) |
9ed74f2d | 3331 | { |
6324fbf3 | 3332 | struct btrfs_space_info *space_info; |
97e728d4 | 3333 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3334 | int wait_for_alloc = 0; |
9ed74f2d | 3335 | int ret = 0; |
9ed74f2d | 3336 | |
2b82032c | 3337 | flags = btrfs_reduce_alloc_profile(extent_root, flags); |
ec44a35c | 3338 | |
6324fbf3 | 3339 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3340 | if (!space_info) { |
3341 | ret = update_space_info(extent_root->fs_info, flags, | |
3342 | 0, 0, &space_info); | |
3343 | BUG_ON(ret); | |
9ed74f2d | 3344 | } |
6324fbf3 | 3345 | BUG_ON(!space_info); |
9ed74f2d | 3346 | |
6d74119f | 3347 | again: |
25179201 | 3348 | spin_lock(&space_info->lock); |
9ed74f2d | 3349 | if (space_info->force_alloc) |
0e4f8f88 | 3350 | force = space_info->force_alloc; |
25179201 JB |
3351 | if (space_info->full) { |
3352 | spin_unlock(&space_info->lock); | |
6d74119f | 3353 | return 0; |
9ed74f2d JB |
3354 | } |
3355 | ||
0e4f8f88 | 3356 | if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) { |
25179201 | 3357 | spin_unlock(&space_info->lock); |
6d74119f JB |
3358 | return 0; |
3359 | } else if (space_info->chunk_alloc) { | |
3360 | wait_for_alloc = 1; | |
3361 | } else { | |
3362 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3363 | } |
0e4f8f88 | 3364 | |
25179201 | 3365 | spin_unlock(&space_info->lock); |
9ed74f2d | 3366 | |
6d74119f JB |
3367 | mutex_lock(&fs_info->chunk_mutex); |
3368 | ||
3369 | /* | |
3370 | * The chunk_mutex is held throughout the entirety of a chunk | |
3371 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3372 | * other guy is done and we need to recheck and see if we should | |
3373 | * allocate. | |
3374 | */ | |
3375 | if (wait_for_alloc) { | |
3376 | mutex_unlock(&fs_info->chunk_mutex); | |
3377 | wait_for_alloc = 0; | |
3378 | goto again; | |
3379 | } | |
3380 | ||
67377734 JB |
3381 | /* |
3382 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3383 | * allocating mixed chunks instead of individual chunks. | |
3384 | */ | |
3385 | if (btrfs_mixed_space_info(space_info)) | |
3386 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3387 | ||
97e728d4 JB |
3388 | /* |
3389 | * if we're doing a data chunk, go ahead and make sure that | |
3390 | * we keep a reasonable number of metadata chunks allocated in the | |
3391 | * FS as well. | |
3392 | */ | |
9ed74f2d | 3393 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3394 | fs_info->data_chunk_allocations++; |
3395 | if (!(fs_info->data_chunk_allocations % | |
3396 | fs_info->metadata_ratio)) | |
3397 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3398 | } |
3399 | ||
2b82032c | 3400 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
9ed74f2d | 3401 | spin_lock(&space_info->lock); |
9ed74f2d | 3402 | if (ret) |
6324fbf3 | 3403 | space_info->full = 1; |
424499db YZ |
3404 | else |
3405 | ret = 1; | |
6d74119f | 3406 | |
0e4f8f88 | 3407 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3408 | space_info->chunk_alloc = 0; |
9ed74f2d | 3409 | spin_unlock(&space_info->lock); |
c146afad | 3410 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 3411 | return ret; |
6324fbf3 | 3412 | } |
9ed74f2d | 3413 | |
9ed74f2d | 3414 | /* |
5da9d01b | 3415 | * shrink metadata reservation for delalloc |
9ed74f2d | 3416 | */ |
5da9d01b | 3417 | static int shrink_delalloc(struct btrfs_trans_handle *trans, |
0019f10d | 3418 | struct btrfs_root *root, u64 to_reclaim, int sync) |
5da9d01b | 3419 | { |
0ca1f7ce | 3420 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3421 | struct btrfs_space_info *space_info; |
5da9d01b YZ |
3422 | u64 reserved; |
3423 | u64 max_reclaim; | |
3424 | u64 reclaimed = 0; | |
b1953bce | 3425 | long time_left; |
bf9022e0 | 3426 | int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3427 | int loops = 0; |
36e39c40 | 3428 | unsigned long progress; |
5da9d01b | 3429 | |
0ca1f7ce | 3430 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3431 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3432 | |
3433 | smp_mb(); | |
0019f10d | 3434 | reserved = space_info->bytes_reserved; |
36e39c40 | 3435 | progress = space_info->reservation_progress; |
5da9d01b YZ |
3436 | |
3437 | if (reserved == 0) | |
3438 | return 0; | |
3439 | ||
3440 | max_reclaim = min(reserved, to_reclaim); | |
3441 | ||
b1953bce | 3442 | while (loops < 1024) { |
bf9022e0 CM |
3443 | /* have the flusher threads jump in and do some IO */ |
3444 | smp_mb(); | |
3445 | nr_pages = min_t(unsigned long, nr_pages, | |
3446 | root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT); | |
3447 | writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages); | |
5da9d01b | 3448 | |
0019f10d | 3449 | spin_lock(&space_info->lock); |
36e39c40 | 3450 | if (reserved > space_info->bytes_reserved) |
0019f10d JB |
3451 | reclaimed += reserved - space_info->bytes_reserved; |
3452 | reserved = space_info->bytes_reserved; | |
3453 | spin_unlock(&space_info->lock); | |
5da9d01b | 3454 | |
36e39c40 CM |
3455 | loops++; |
3456 | ||
5da9d01b YZ |
3457 | if (reserved == 0 || reclaimed >= max_reclaim) |
3458 | break; | |
3459 | ||
3460 | if (trans && trans->transaction->blocked) | |
3461 | return -EAGAIN; | |
bf9022e0 | 3462 | |
36e39c40 | 3463 | time_left = schedule_timeout_interruptible(1); |
b1953bce JB |
3464 | |
3465 | /* We were interrupted, exit */ | |
3466 | if (time_left) | |
3467 | break; | |
3468 | ||
36e39c40 CM |
3469 | /* we've kicked the IO a few times, if anything has been freed, |
3470 | * exit. There is no sense in looping here for a long time | |
3471 | * when we really need to commit the transaction, or there are | |
3472 | * just too many writers without enough free space | |
3473 | */ | |
3474 | ||
3475 | if (loops > 3) { | |
3476 | smp_mb(); | |
3477 | if (progress != space_info->reservation_progress) | |
3478 | break; | |
3479 | } | |
bf9022e0 | 3480 | |
5da9d01b YZ |
3481 | } |
3482 | return reclaimed >= to_reclaim; | |
3483 | } | |
3484 | ||
8bb8ab2e JB |
3485 | /* |
3486 | * Retries tells us how many times we've called reserve_metadata_bytes. The | |
3487 | * idea is if this is the first call (retries == 0) then we will add to our | |
3488 | * reserved count if we can't make the allocation in order to hold our place | |
3489 | * while we go and try and free up space. That way for retries > 1 we don't try | |
3490 | * and add space, we just check to see if the amount of unused space is >= the | |
3491 | * total space, meaning that our reservation is valid. | |
3492 | * | |
3493 | * However if we don't intend to retry this reservation, pass -1 as retries so | |
3494 | * that it short circuits this logic. | |
3495 | */ | |
3496 | static int reserve_metadata_bytes(struct btrfs_trans_handle *trans, | |
3497 | struct btrfs_root *root, | |
3498 | struct btrfs_block_rsv *block_rsv, | |
3499 | u64 orig_bytes, int flush) | |
9ed74f2d | 3500 | { |
f0486c68 | 3501 | struct btrfs_space_info *space_info = block_rsv->space_info; |
8bb8ab2e JB |
3502 | u64 unused; |
3503 | u64 num_bytes = orig_bytes; | |
3504 | int retries = 0; | |
3505 | int ret = 0; | |
3506 | bool reserved = false; | |
38227933 | 3507 | bool committed = false; |
9ed74f2d | 3508 | |
8bb8ab2e JB |
3509 | again: |
3510 | ret = -ENOSPC; | |
3511 | if (reserved) | |
3512 | num_bytes = 0; | |
9ed74f2d | 3513 | |
8bb8ab2e JB |
3514 | spin_lock(&space_info->lock); |
3515 | unused = space_info->bytes_used + space_info->bytes_reserved + | |
3516 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3517 | space_info->bytes_may_use; | |
9ed74f2d | 3518 | |
8bb8ab2e JB |
3519 | /* |
3520 | * The idea here is that we've not already over-reserved the block group | |
3521 | * then we can go ahead and save our reservation first and then start | |
3522 | * flushing if we need to. Otherwise if we've already overcommitted | |
3523 | * lets start flushing stuff first and then come back and try to make | |
3524 | * our reservation. | |
3525 | */ | |
3526 | if (unused <= space_info->total_bytes) { | |
6f334348 | 3527 | unused = space_info->total_bytes - unused; |
8bb8ab2e JB |
3528 | if (unused >= num_bytes) { |
3529 | if (!reserved) | |
3530 | space_info->bytes_reserved += orig_bytes; | |
3531 | ret = 0; | |
3532 | } else { | |
3533 | /* | |
3534 | * Ok set num_bytes to orig_bytes since we aren't | |
3535 | * overocmmitted, this way we only try and reclaim what | |
3536 | * we need. | |
3537 | */ | |
3538 | num_bytes = orig_bytes; | |
3539 | } | |
3540 | } else { | |
3541 | /* | |
3542 | * Ok we're over committed, set num_bytes to the overcommitted | |
3543 | * amount plus the amount of bytes that we need for this | |
3544 | * reservation. | |
3545 | */ | |
3546 | num_bytes = unused - space_info->total_bytes + | |
3547 | (orig_bytes * (retries + 1)); | |
3548 | } | |
9ed74f2d | 3549 | |
8bb8ab2e JB |
3550 | /* |
3551 | * Couldn't make our reservation, save our place so while we're trying | |
3552 | * to reclaim space we can actually use it instead of somebody else | |
3553 | * stealing it from us. | |
3554 | */ | |
3555 | if (ret && !reserved) { | |
3556 | space_info->bytes_reserved += orig_bytes; | |
3557 | reserved = true; | |
3558 | } | |
9ed74f2d | 3559 | |
f0486c68 | 3560 | spin_unlock(&space_info->lock); |
9ed74f2d | 3561 | |
8bb8ab2e JB |
3562 | if (!ret) |
3563 | return 0; | |
9ed74f2d | 3564 | |
8bb8ab2e JB |
3565 | if (!flush) |
3566 | goto out; | |
f0486c68 | 3567 | |
8bb8ab2e JB |
3568 | /* |
3569 | * We do synchronous shrinking since we don't actually unreserve | |
3570 | * metadata until after the IO is completed. | |
3571 | */ | |
3572 | ret = shrink_delalloc(trans, root, num_bytes, 1); | |
3573 | if (ret > 0) | |
3574 | return 0; | |
3575 | else if (ret < 0) | |
3576 | goto out; | |
f0486c68 | 3577 | |
8bb8ab2e JB |
3578 | /* |
3579 | * So if we were overcommitted it's possible that somebody else flushed | |
3580 | * out enough space and we simply didn't have enough space to reclaim, | |
3581 | * so go back around and try again. | |
3582 | */ | |
3583 | if (retries < 2) { | |
3584 | retries++; | |
3585 | goto again; | |
3586 | } | |
f0486c68 YZ |
3587 | |
3588 | spin_lock(&space_info->lock); | |
8bb8ab2e JB |
3589 | /* |
3590 | * Not enough space to be reclaimed, don't bother committing the | |
3591 | * transaction. | |
3592 | */ | |
3593 | if (space_info->bytes_pinned < orig_bytes) | |
3594 | ret = -ENOSPC; | |
3595 | spin_unlock(&space_info->lock); | |
3596 | if (ret) | |
3597 | goto out; | |
f0486c68 | 3598 | |
8bb8ab2e | 3599 | ret = -EAGAIN; |
38227933 | 3600 | if (trans || committed) |
8bb8ab2e | 3601 | goto out; |
f0486c68 | 3602 | |
8bb8ab2e JB |
3603 | ret = -ENOSPC; |
3604 | trans = btrfs_join_transaction(root, 1); | |
3605 | if (IS_ERR(trans)) | |
3606 | goto out; | |
3607 | ret = btrfs_commit_transaction(trans, root); | |
38227933 JB |
3608 | if (!ret) { |
3609 | trans = NULL; | |
3610 | committed = true; | |
8bb8ab2e | 3611 | goto again; |
38227933 | 3612 | } |
8bb8ab2e JB |
3613 | |
3614 | out: | |
3615 | if (reserved) { | |
3616 | spin_lock(&space_info->lock); | |
3617 | space_info->bytes_reserved -= orig_bytes; | |
3618 | spin_unlock(&space_info->lock); | |
f0486c68 | 3619 | } |
4e06bdd6 | 3620 | |
f0486c68 YZ |
3621 | return ret; |
3622 | } | |
3623 | ||
3624 | static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans, | |
3625 | struct btrfs_root *root) | |
3626 | { | |
3627 | struct btrfs_block_rsv *block_rsv; | |
3628 | if (root->ref_cows) | |
3629 | block_rsv = trans->block_rsv; | |
3630 | else | |
3631 | block_rsv = root->block_rsv; | |
3632 | ||
3633 | if (!block_rsv) | |
3634 | block_rsv = &root->fs_info->empty_block_rsv; | |
3635 | ||
3636 | return block_rsv; | |
3637 | } | |
3638 | ||
3639 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
3640 | u64 num_bytes) | |
3641 | { | |
3642 | int ret = -ENOSPC; | |
3643 | spin_lock(&block_rsv->lock); | |
3644 | if (block_rsv->reserved >= num_bytes) { | |
3645 | block_rsv->reserved -= num_bytes; | |
3646 | if (block_rsv->reserved < block_rsv->size) | |
3647 | block_rsv->full = 0; | |
3648 | ret = 0; | |
3649 | } | |
3650 | spin_unlock(&block_rsv->lock); | |
3651 | return ret; | |
3652 | } | |
3653 | ||
3654 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3655 | u64 num_bytes, int update_size) | |
3656 | { | |
3657 | spin_lock(&block_rsv->lock); | |
3658 | block_rsv->reserved += num_bytes; | |
3659 | if (update_size) | |
3660 | block_rsv->size += num_bytes; | |
3661 | else if (block_rsv->reserved >= block_rsv->size) | |
3662 | block_rsv->full = 1; | |
3663 | spin_unlock(&block_rsv->lock); | |
3664 | } | |
3665 | ||
62a45b60 DS |
3666 | static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv, |
3667 | struct btrfs_block_rsv *dest, u64 num_bytes) | |
f0486c68 YZ |
3668 | { |
3669 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
3670 | ||
3671 | spin_lock(&block_rsv->lock); | |
3672 | if (num_bytes == (u64)-1) | |
3673 | num_bytes = block_rsv->size; | |
3674 | block_rsv->size -= num_bytes; | |
3675 | if (block_rsv->reserved >= block_rsv->size) { | |
3676 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3677 | block_rsv->reserved = block_rsv->size; | |
3678 | block_rsv->full = 1; | |
3679 | } else { | |
3680 | num_bytes = 0; | |
3681 | } | |
3682 | spin_unlock(&block_rsv->lock); | |
3683 | ||
3684 | if (num_bytes > 0) { | |
3685 | if (dest) { | |
e9e22899 JB |
3686 | spin_lock(&dest->lock); |
3687 | if (!dest->full) { | |
3688 | u64 bytes_to_add; | |
3689 | ||
3690 | bytes_to_add = dest->size - dest->reserved; | |
3691 | bytes_to_add = min(num_bytes, bytes_to_add); | |
3692 | dest->reserved += bytes_to_add; | |
3693 | if (dest->reserved >= dest->size) | |
3694 | dest->full = 1; | |
3695 | num_bytes -= bytes_to_add; | |
3696 | } | |
3697 | spin_unlock(&dest->lock); | |
3698 | } | |
3699 | if (num_bytes) { | |
f0486c68 YZ |
3700 | spin_lock(&space_info->lock); |
3701 | space_info->bytes_reserved -= num_bytes; | |
36e39c40 | 3702 | space_info->reservation_progress++; |
f0486c68 | 3703 | spin_unlock(&space_info->lock); |
4e06bdd6 | 3704 | } |
9ed74f2d | 3705 | } |
f0486c68 | 3706 | } |
4e06bdd6 | 3707 | |
f0486c68 YZ |
3708 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
3709 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
3710 | { | |
3711 | int ret; | |
9ed74f2d | 3712 | |
f0486c68 YZ |
3713 | ret = block_rsv_use_bytes(src, num_bytes); |
3714 | if (ret) | |
3715 | return ret; | |
9ed74f2d | 3716 | |
f0486c68 | 3717 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
3718 | return 0; |
3719 | } | |
3720 | ||
f0486c68 | 3721 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv) |
9ed74f2d | 3722 | { |
f0486c68 YZ |
3723 | memset(rsv, 0, sizeof(*rsv)); |
3724 | spin_lock_init(&rsv->lock); | |
3725 | atomic_set(&rsv->usage, 1); | |
3726 | rsv->priority = 6; | |
3727 | INIT_LIST_HEAD(&rsv->list); | |
3728 | } | |
3729 | ||
3730 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root) | |
3731 | { | |
3732 | struct btrfs_block_rsv *block_rsv; | |
3733 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 3734 | |
f0486c68 YZ |
3735 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
3736 | if (!block_rsv) | |
3737 | return NULL; | |
9ed74f2d | 3738 | |
f0486c68 | 3739 | btrfs_init_block_rsv(block_rsv); |
f0486c68 YZ |
3740 | block_rsv->space_info = __find_space_info(fs_info, |
3741 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
3742 | return block_rsv; |
3743 | } | |
9ed74f2d | 3744 | |
f0486c68 YZ |
3745 | void btrfs_free_block_rsv(struct btrfs_root *root, |
3746 | struct btrfs_block_rsv *rsv) | |
3747 | { | |
3748 | if (rsv && atomic_dec_and_test(&rsv->usage)) { | |
3749 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
3750 | if (!rsv->durable) | |
3751 | kfree(rsv); | |
3752 | } | |
9ed74f2d JB |
3753 | } |
3754 | ||
3755 | /* | |
f0486c68 YZ |
3756 | * make the block_rsv struct be able to capture freed space. |
3757 | * the captured space will re-add to the the block_rsv struct | |
3758 | * after transaction commit | |
9ed74f2d | 3759 | */ |
f0486c68 YZ |
3760 | void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info, |
3761 | struct btrfs_block_rsv *block_rsv) | |
9ed74f2d | 3762 | { |
f0486c68 YZ |
3763 | block_rsv->durable = 1; |
3764 | mutex_lock(&fs_info->durable_block_rsv_mutex); | |
3765 | list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list); | |
3766 | mutex_unlock(&fs_info->durable_block_rsv_mutex); | |
3767 | } | |
9ed74f2d | 3768 | |
f0486c68 YZ |
3769 | int btrfs_block_rsv_add(struct btrfs_trans_handle *trans, |
3770 | struct btrfs_root *root, | |
3771 | struct btrfs_block_rsv *block_rsv, | |
8bb8ab2e | 3772 | u64 num_bytes) |
f0486c68 YZ |
3773 | { |
3774 | int ret; | |
9ed74f2d | 3775 | |
f0486c68 YZ |
3776 | if (num_bytes == 0) |
3777 | return 0; | |
8bb8ab2e JB |
3778 | |
3779 | ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1); | |
f0486c68 YZ |
3780 | if (!ret) { |
3781 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
3782 | return 0; | |
3783 | } | |
9ed74f2d | 3784 | |
f0486c68 YZ |
3785 | return ret; |
3786 | } | |
9ed74f2d | 3787 | |
f0486c68 YZ |
3788 | int btrfs_block_rsv_check(struct btrfs_trans_handle *trans, |
3789 | struct btrfs_root *root, | |
3790 | struct btrfs_block_rsv *block_rsv, | |
3791 | u64 min_reserved, int min_factor) | |
3792 | { | |
3793 | u64 num_bytes = 0; | |
3794 | int commit_trans = 0; | |
3795 | int ret = -ENOSPC; | |
9ed74f2d | 3796 | |
f0486c68 YZ |
3797 | if (!block_rsv) |
3798 | return 0; | |
9ed74f2d | 3799 | |
f0486c68 YZ |
3800 | spin_lock(&block_rsv->lock); |
3801 | if (min_factor > 0) | |
3802 | num_bytes = div_factor(block_rsv->size, min_factor); | |
3803 | if (min_reserved > num_bytes) | |
3804 | num_bytes = min_reserved; | |
9ed74f2d | 3805 | |
f0486c68 YZ |
3806 | if (block_rsv->reserved >= num_bytes) { |
3807 | ret = 0; | |
3808 | } else { | |
3809 | num_bytes -= block_rsv->reserved; | |
3810 | if (block_rsv->durable && | |
3811 | block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes) | |
3812 | commit_trans = 1; | |
3813 | } | |
3814 | spin_unlock(&block_rsv->lock); | |
3815 | if (!ret) | |
3816 | return 0; | |
3817 | ||
3818 | if (block_rsv->refill_used) { | |
8bb8ab2e JB |
3819 | ret = reserve_metadata_bytes(trans, root, block_rsv, |
3820 | num_bytes, 0); | |
f0486c68 YZ |
3821 | if (!ret) { |
3822 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
3823 | return 0; | |
4e06bdd6 | 3824 | } |
f0486c68 | 3825 | } |
9ed74f2d | 3826 | |
f0486c68 YZ |
3827 | if (commit_trans) { |
3828 | if (trans) | |
3829 | return -EAGAIN; | |
3830 | ||
3831 | trans = btrfs_join_transaction(root, 1); | |
3832 | BUG_ON(IS_ERR(trans)); | |
3833 | ret = btrfs_commit_transaction(trans, root); | |
3834 | return 0; | |
6a63209f | 3835 | } |
9ed74f2d | 3836 | |
f0486c68 YZ |
3837 | return -ENOSPC; |
3838 | } | |
3839 | ||
3840 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
3841 | struct btrfs_block_rsv *dst_rsv, | |
3842 | u64 num_bytes) | |
3843 | { | |
3844 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
3845 | } | |
3846 | ||
3847 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
3848 | struct btrfs_block_rsv *block_rsv, | |
3849 | u64 num_bytes) | |
3850 | { | |
3851 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
3852 | if (global_rsv->full || global_rsv == block_rsv || | |
3853 | block_rsv->space_info != global_rsv->space_info) | |
3854 | global_rsv = NULL; | |
3855 | block_rsv_release_bytes(block_rsv, global_rsv, num_bytes); | |
6a63209f JB |
3856 | } |
3857 | ||
3858 | /* | |
8929ecfa YZ |
3859 | * helper to calculate size of global block reservation. |
3860 | * the desired value is sum of space used by extent tree, | |
3861 | * checksum tree and root tree | |
6a63209f | 3862 | */ |
8929ecfa | 3863 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 3864 | { |
8929ecfa YZ |
3865 | struct btrfs_space_info *sinfo; |
3866 | u64 num_bytes; | |
3867 | u64 meta_used; | |
3868 | u64 data_used; | |
3869 | int csum_size = btrfs_super_csum_size(&fs_info->super_copy); | |
3870 | #if 0 | |
3871 | /* | |
3872 | * per tree used space accounting can be inaccuracy, so we | |
3873 | * can't rely on it. | |
3874 | */ | |
3875 | spin_lock(&fs_info->extent_root->accounting_lock); | |
3876 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item); | |
3877 | spin_unlock(&fs_info->extent_root->accounting_lock); | |
6a63209f | 3878 | |
8929ecfa YZ |
3879 | spin_lock(&fs_info->csum_root->accounting_lock); |
3880 | num_bytes += btrfs_root_used(&fs_info->csum_root->root_item); | |
3881 | spin_unlock(&fs_info->csum_root->accounting_lock); | |
6a63209f | 3882 | |
8929ecfa YZ |
3883 | spin_lock(&fs_info->tree_root->accounting_lock); |
3884 | num_bytes += btrfs_root_used(&fs_info->tree_root->root_item); | |
3885 | spin_unlock(&fs_info->tree_root->accounting_lock); | |
3886 | #endif | |
3887 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
3888 | spin_lock(&sinfo->lock); | |
3889 | data_used = sinfo->bytes_used; | |
3890 | spin_unlock(&sinfo->lock); | |
33b4d47f | 3891 | |
8929ecfa YZ |
3892 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
3893 | spin_lock(&sinfo->lock); | |
6d48755d JB |
3894 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
3895 | data_used = 0; | |
8929ecfa YZ |
3896 | meta_used = sinfo->bytes_used; |
3897 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 3898 | |
8929ecfa YZ |
3899 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
3900 | csum_size * 2; | |
3901 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 3902 | |
8929ecfa YZ |
3903 | if (num_bytes * 3 > meta_used) |
3904 | num_bytes = div64_u64(meta_used, 3); | |
ab6e2410 | 3905 | |
8929ecfa YZ |
3906 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
3907 | } | |
6a63209f | 3908 | |
8929ecfa YZ |
3909 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
3910 | { | |
3911 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
3912 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
3913 | u64 num_bytes; | |
6a63209f | 3914 | |
8929ecfa | 3915 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 3916 | |
8929ecfa YZ |
3917 | spin_lock(&block_rsv->lock); |
3918 | spin_lock(&sinfo->lock); | |
4e06bdd6 | 3919 | |
8929ecfa | 3920 | block_rsv->size = num_bytes; |
4e06bdd6 | 3921 | |
8929ecfa | 3922 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
3923 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
3924 | sinfo->bytes_may_use; | |
8929ecfa YZ |
3925 | |
3926 | if (sinfo->total_bytes > num_bytes) { | |
3927 | num_bytes = sinfo->total_bytes - num_bytes; | |
3928 | block_rsv->reserved += num_bytes; | |
3929 | sinfo->bytes_reserved += num_bytes; | |
6a63209f | 3930 | } |
6a63209f | 3931 | |
8929ecfa YZ |
3932 | if (block_rsv->reserved >= block_rsv->size) { |
3933 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3934 | sinfo->bytes_reserved -= num_bytes; | |
36e39c40 | 3935 | sinfo->reservation_progress++; |
8929ecfa YZ |
3936 | block_rsv->reserved = block_rsv->size; |
3937 | block_rsv->full = 1; | |
3938 | } | |
3939 | #if 0 | |
3940 | printk(KERN_INFO"global block rsv size %llu reserved %llu\n", | |
3941 | block_rsv->size, block_rsv->reserved); | |
3942 | #endif | |
3943 | spin_unlock(&sinfo->lock); | |
3944 | spin_unlock(&block_rsv->lock); | |
6a63209f JB |
3945 | } |
3946 | ||
f0486c68 | 3947 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 3948 | { |
f0486c68 | 3949 | struct btrfs_space_info *space_info; |
6a63209f | 3950 | |
f0486c68 YZ |
3951 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
3952 | fs_info->chunk_block_rsv.space_info = space_info; | |
3953 | fs_info->chunk_block_rsv.priority = 10; | |
6a63209f | 3954 | |
f0486c68 | 3955 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa YZ |
3956 | fs_info->global_block_rsv.space_info = space_info; |
3957 | fs_info->global_block_rsv.priority = 10; | |
3958 | fs_info->global_block_rsv.refill_used = 1; | |
3959 | fs_info->delalloc_block_rsv.space_info = space_info; | |
f0486c68 YZ |
3960 | fs_info->trans_block_rsv.space_info = space_info; |
3961 | fs_info->empty_block_rsv.space_info = space_info; | |
3962 | fs_info->empty_block_rsv.priority = 10; | |
3963 | ||
8929ecfa YZ |
3964 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
3965 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
3966 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
3967 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 3968 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa YZ |
3969 | |
3970 | btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv); | |
3971 | ||
3972 | btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv); | |
3973 | ||
3974 | update_global_block_rsv(fs_info); | |
6a63209f JB |
3975 | } |
3976 | ||
8929ecfa | 3977 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 3978 | { |
8929ecfa YZ |
3979 | block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1); |
3980 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); | |
3981 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
3982 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
3983 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
3984 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
3985 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
f0486c68 | 3986 | } |
6a63209f | 3987 | |
a22285a6 YZ |
3988 | static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items) |
3989 | { | |
3990 | return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) * | |
3991 | 3 * num_items; | |
3992 | } | |
6a63209f | 3993 | |
a22285a6 YZ |
3994 | int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans, |
3995 | struct btrfs_root *root, | |
8bb8ab2e | 3996 | int num_items) |
a22285a6 YZ |
3997 | { |
3998 | u64 num_bytes; | |
3999 | int ret; | |
6a63209f | 4000 | |
a22285a6 YZ |
4001 | if (num_items == 0 || root->fs_info->chunk_root == root) |
4002 | return 0; | |
6a63209f | 4003 | |
a22285a6 YZ |
4004 | num_bytes = calc_trans_metadata_size(root, num_items); |
4005 | ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv, | |
8bb8ab2e | 4006 | num_bytes); |
a22285a6 YZ |
4007 | if (!ret) { |
4008 | trans->bytes_reserved += num_bytes; | |
4009 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
4010 | } | |
4011 | return ret; | |
6a63209f JB |
4012 | } |
4013 | ||
a22285a6 YZ |
4014 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4015 | struct btrfs_root *root) | |
6a63209f | 4016 | { |
a22285a6 YZ |
4017 | if (!trans->bytes_reserved) |
4018 | return; | |
6a63209f | 4019 | |
a22285a6 YZ |
4020 | BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv); |
4021 | btrfs_block_rsv_release(root, trans->block_rsv, | |
4022 | trans->bytes_reserved); | |
4023 | trans->bytes_reserved = 0; | |
4024 | } | |
6a63209f | 4025 | |
d68fc57b YZ |
4026 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4027 | struct inode *inode) | |
4028 | { | |
4029 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4030 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4031 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4032 | ||
4033 | /* | |
4034 | * one for deleting orphan item, one for updating inode and | |
4035 | * two for calling btrfs_truncate_inode_items. | |
4036 | * | |
4037 | * btrfs_truncate_inode_items is a delete operation, it frees | |
4038 | * more space than it uses in most cases. So two units of | |
4039 | * metadata space should be enough for calling it many times. | |
4040 | * If all of the metadata space is used, we can commit | |
4041 | * transaction and use space it freed. | |
4042 | */ | |
4043 | u64 num_bytes = calc_trans_metadata_size(root, 4); | |
4044 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
6a63209f JB |
4045 | } |
4046 | ||
d68fc57b | 4047 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4048 | { |
d68fc57b YZ |
4049 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4050 | u64 num_bytes = calc_trans_metadata_size(root, 4); | |
4051 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); | |
4052 | } | |
97e728d4 | 4053 | |
a22285a6 YZ |
4054 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
4055 | struct btrfs_pending_snapshot *pending) | |
4056 | { | |
4057 | struct btrfs_root *root = pending->root; | |
4058 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4059 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
4060 | /* | |
4061 | * two for root back/forward refs, two for directory entries | |
4062 | * and one for root of the snapshot. | |
4063 | */ | |
4064 | u64 num_bytes = calc_trans_metadata_size(root, 5); | |
4065 | dst_rsv->space_info = src_rsv->space_info; | |
4066 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
4067 | } |
4068 | ||
0ca1f7ce | 4069 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes) |
6324fbf3 | 4070 | { |
0ca1f7ce YZ |
4071 | return num_bytes >>= 3; |
4072 | } | |
c146afad | 4073 | |
0ca1f7ce YZ |
4074 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4075 | { | |
4076 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4077 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
4078 | u64 to_reserve; | |
4079 | int nr_extents; | |
57a45ced | 4080 | int reserved_extents; |
0ca1f7ce | 4081 | int ret; |
6324fbf3 | 4082 | |
0ca1f7ce YZ |
4083 | if (btrfs_transaction_in_commit(root->fs_info)) |
4084 | schedule_timeout(1); | |
ec44a35c | 4085 | |
0ca1f7ce | 4086 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4087 | |
0ca1f7ce | 4088 | nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1; |
57a45ced JB |
4089 | reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents); |
4090 | ||
4091 | if (nr_extents > reserved_extents) { | |
4092 | nr_extents -= reserved_extents; | |
0ca1f7ce YZ |
4093 | to_reserve = calc_trans_metadata_size(root, nr_extents); |
4094 | } else { | |
4095 | nr_extents = 0; | |
4096 | to_reserve = 0; | |
593060d7 | 4097 | } |
57a45ced | 4098 | |
0ca1f7ce | 4099 | to_reserve += calc_csum_metadata_size(inode, num_bytes); |
8bb8ab2e JB |
4100 | ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1); |
4101 | if (ret) | |
0ca1f7ce | 4102 | return ret; |
6324fbf3 | 4103 | |
57a45ced | 4104 | atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents); |
0ca1f7ce | 4105 | atomic_inc(&BTRFS_I(inode)->outstanding_extents); |
25179201 | 4106 | |
0ca1f7ce YZ |
4107 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4108 | ||
4109 | if (block_rsv->size > 512 * 1024 * 1024) | |
0019f10d | 4110 | shrink_delalloc(NULL, root, to_reserve, 0); |
0ca1f7ce YZ |
4111 | |
4112 | return 0; | |
4113 | } | |
4114 | ||
4115 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) | |
4116 | { | |
4117 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4118 | u64 to_free; | |
4119 | int nr_extents; | |
57a45ced | 4120 | int reserved_extents; |
0ca1f7ce YZ |
4121 | |
4122 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
4123 | atomic_dec(&BTRFS_I(inode)->outstanding_extents); | |
3c14874a | 4124 | WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0); |
0ca1f7ce | 4125 | |
57a45ced JB |
4126 | reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents); |
4127 | do { | |
4128 | int old, new; | |
4129 | ||
4130 | nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents); | |
4131 | if (nr_extents >= reserved_extents) { | |
4132 | nr_extents = 0; | |
4133 | break; | |
4134 | } | |
4135 | old = reserved_extents; | |
4136 | nr_extents = reserved_extents - nr_extents; | |
4137 | new = reserved_extents - nr_extents; | |
4138 | old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents, | |
4139 | reserved_extents, new); | |
4140 | if (likely(old == reserved_extents)) | |
4141 | break; | |
4142 | reserved_extents = old; | |
4143 | } while (1); | |
97e728d4 | 4144 | |
0ca1f7ce YZ |
4145 | to_free = calc_csum_metadata_size(inode, num_bytes); |
4146 | if (nr_extents > 0) | |
4147 | to_free += calc_trans_metadata_size(root, nr_extents); | |
4148 | ||
4149 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, | |
4150 | to_free); | |
4151 | } | |
4152 | ||
4153 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) | |
4154 | { | |
4155 | int ret; | |
4156 | ||
4157 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4158 | if (ret) |
0ca1f7ce YZ |
4159 | return ret; |
4160 | ||
4161 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4162 | if (ret) { | |
4163 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4164 | return ret; | |
4165 | } | |
4166 | ||
4167 | return 0; | |
4168 | } | |
4169 | ||
4170 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) | |
4171 | { | |
4172 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4173 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4174 | } |
4175 | ||
9078a3e1 CM |
4176 | static int update_block_group(struct btrfs_trans_handle *trans, |
4177 | struct btrfs_root *root, | |
f0486c68 | 4178 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4179 | { |
0af3d00b | 4180 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4181 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4182 | u64 total = num_bytes; |
9078a3e1 | 4183 | u64 old_val; |
db94535d | 4184 | u64 byte_in_group; |
0af3d00b | 4185 | int factor; |
3e1ad54f | 4186 | |
5d4f98a2 YZ |
4187 | /* block accounting for super block */ |
4188 | spin_lock(&info->delalloc_lock); | |
4189 | old_val = btrfs_super_bytes_used(&info->super_copy); | |
4190 | if (alloc) | |
4191 | old_val += num_bytes; | |
4192 | else | |
4193 | old_val -= num_bytes; | |
4194 | btrfs_set_super_bytes_used(&info->super_copy, old_val); | |
5d4f98a2 YZ |
4195 | spin_unlock(&info->delalloc_lock); |
4196 | ||
d397712b | 4197 | while (total) { |
db94535d | 4198 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4199 | if (!cache) |
9078a3e1 | 4200 | return -1; |
b742bb82 YZ |
4201 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4202 | BTRFS_BLOCK_GROUP_RAID1 | | |
4203 | BTRFS_BLOCK_GROUP_RAID10)) | |
4204 | factor = 2; | |
4205 | else | |
4206 | factor = 1; | |
9d66e233 JB |
4207 | /* |
4208 | * If this block group has free space cache written out, we | |
4209 | * need to make sure to load it if we are removing space. This | |
4210 | * is because we need the unpinning stage to actually add the | |
4211 | * space back to the block group, otherwise we will leak space. | |
4212 | */ | |
4213 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4214 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4215 | |
db94535d CM |
4216 | byte_in_group = bytenr - cache->key.objectid; |
4217 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4218 | |
25179201 | 4219 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4220 | spin_lock(&cache->lock); |
0af3d00b JB |
4221 | |
4222 | if (btrfs_super_cache_generation(&info->super_copy) != 0 && | |
4223 | cache->disk_cache_state < BTRFS_DC_CLEAR) | |
4224 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4225 | ||
0f9dd46c | 4226 | cache->dirty = 1; |
9078a3e1 | 4227 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4228 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4229 | if (alloc) { |
db94535d | 4230 | old_val += num_bytes; |
11833d66 YZ |
4231 | btrfs_set_block_group_used(&cache->item, old_val); |
4232 | cache->reserved -= num_bytes; | |
11833d66 | 4233 | cache->space_info->bytes_reserved -= num_bytes; |
36e39c40 | 4234 | cache->space_info->reservation_progress++; |
b742bb82 YZ |
4235 | cache->space_info->bytes_used += num_bytes; |
4236 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4237 | spin_unlock(&cache->lock); |
25179201 | 4238 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4239 | } else { |
db94535d | 4240 | old_val -= num_bytes; |
c286ac48 | 4241 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4242 | cache->pinned += num_bytes; |
4243 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4244 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4245 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4246 | spin_unlock(&cache->lock); |
25179201 | 4247 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4248 | |
f0486c68 YZ |
4249 | set_extent_dirty(info->pinned_extents, |
4250 | bytenr, bytenr + num_bytes - 1, | |
4251 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4252 | } |
fa9c0d79 | 4253 | btrfs_put_block_group(cache); |
db94535d CM |
4254 | total -= num_bytes; |
4255 | bytenr += num_bytes; | |
9078a3e1 CM |
4256 | } |
4257 | return 0; | |
4258 | } | |
6324fbf3 | 4259 | |
a061fc8d CM |
4260 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4261 | { | |
0f9dd46c | 4262 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4263 | u64 bytenr; |
0f9dd46c JB |
4264 | |
4265 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4266 | if (!cache) | |
a061fc8d | 4267 | return 0; |
0f9dd46c | 4268 | |
d2fb3437 | 4269 | bytenr = cache->key.objectid; |
fa9c0d79 | 4270 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4271 | |
4272 | return bytenr; | |
a061fc8d CM |
4273 | } |
4274 | ||
f0486c68 YZ |
4275 | static int pin_down_extent(struct btrfs_root *root, |
4276 | struct btrfs_block_group_cache *cache, | |
4277 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4278 | { |
11833d66 YZ |
4279 | spin_lock(&cache->space_info->lock); |
4280 | spin_lock(&cache->lock); | |
4281 | cache->pinned += num_bytes; | |
4282 | cache->space_info->bytes_pinned += num_bytes; | |
4283 | if (reserved) { | |
4284 | cache->reserved -= num_bytes; | |
4285 | cache->space_info->bytes_reserved -= num_bytes; | |
36e39c40 | 4286 | cache->space_info->reservation_progress++; |
11833d66 YZ |
4287 | } |
4288 | spin_unlock(&cache->lock); | |
4289 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4290 | |
f0486c68 YZ |
4291 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4292 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4293 | return 0; | |
4294 | } | |
68b38550 | 4295 | |
f0486c68 YZ |
4296 | /* |
4297 | * this function must be called within transaction | |
4298 | */ | |
4299 | int btrfs_pin_extent(struct btrfs_root *root, | |
4300 | u64 bytenr, u64 num_bytes, int reserved) | |
4301 | { | |
4302 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4303 | |
f0486c68 YZ |
4304 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
4305 | BUG_ON(!cache); | |
4306 | ||
4307 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4308 | ||
4309 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4310 | return 0; |
4311 | } | |
4312 | ||
f0486c68 YZ |
4313 | /* |
4314 | * update size of reserved extents. this function may return -EAGAIN | |
4315 | * if 'reserve' is true or 'sinfo' is false. | |
4316 | */ | |
b4d00d56 LD |
4317 | int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4318 | u64 num_bytes, int reserve, int sinfo) | |
11833d66 | 4319 | { |
f0486c68 YZ |
4320 | int ret = 0; |
4321 | if (sinfo) { | |
4322 | struct btrfs_space_info *space_info = cache->space_info; | |
4323 | spin_lock(&space_info->lock); | |
4324 | spin_lock(&cache->lock); | |
4325 | if (reserve) { | |
4326 | if (cache->ro) { | |
4327 | ret = -EAGAIN; | |
4328 | } else { | |
4329 | cache->reserved += num_bytes; | |
4330 | space_info->bytes_reserved += num_bytes; | |
4331 | } | |
4332 | } else { | |
4333 | if (cache->ro) | |
4334 | space_info->bytes_readonly += num_bytes; | |
4335 | cache->reserved -= num_bytes; | |
4336 | space_info->bytes_reserved -= num_bytes; | |
36e39c40 | 4337 | space_info->reservation_progress++; |
f0486c68 YZ |
4338 | } |
4339 | spin_unlock(&cache->lock); | |
4340 | spin_unlock(&space_info->lock); | |
11833d66 | 4341 | } else { |
f0486c68 YZ |
4342 | spin_lock(&cache->lock); |
4343 | if (cache->ro) { | |
4344 | ret = -EAGAIN; | |
4345 | } else { | |
4346 | if (reserve) | |
4347 | cache->reserved += num_bytes; | |
4348 | else | |
4349 | cache->reserved -= num_bytes; | |
4350 | } | |
4351 | spin_unlock(&cache->lock); | |
324ae4df | 4352 | } |
f0486c68 | 4353 | return ret; |
324ae4df | 4354 | } |
9078a3e1 | 4355 | |
11833d66 YZ |
4356 | int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
4357 | struct btrfs_root *root) | |
e8569813 | 4358 | { |
e8569813 | 4359 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4360 | struct btrfs_caching_control *next; |
4361 | struct btrfs_caching_control *caching_ctl; | |
4362 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4363 | |
11833d66 | 4364 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4365 | |
11833d66 YZ |
4366 | list_for_each_entry_safe(caching_ctl, next, |
4367 | &fs_info->caching_block_groups, list) { | |
4368 | cache = caching_ctl->block_group; | |
4369 | if (block_group_cache_done(cache)) { | |
4370 | cache->last_byte_to_unpin = (u64)-1; | |
4371 | list_del_init(&caching_ctl->list); | |
4372 | put_caching_control(caching_ctl); | |
e8569813 | 4373 | } else { |
11833d66 | 4374 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 4375 | } |
e8569813 | 4376 | } |
11833d66 YZ |
4377 | |
4378 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
4379 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
4380 | else | |
4381 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
4382 | ||
4383 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
4384 | |
4385 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
4386 | return 0; |
4387 | } | |
4388 | ||
11833d66 | 4389 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 4390 | { |
11833d66 YZ |
4391 | struct btrfs_fs_info *fs_info = root->fs_info; |
4392 | struct btrfs_block_group_cache *cache = NULL; | |
4393 | u64 len; | |
ccd467d6 | 4394 | |
11833d66 YZ |
4395 | while (start <= end) { |
4396 | if (!cache || | |
4397 | start >= cache->key.objectid + cache->key.offset) { | |
4398 | if (cache) | |
4399 | btrfs_put_block_group(cache); | |
4400 | cache = btrfs_lookup_block_group(fs_info, start); | |
4401 | BUG_ON(!cache); | |
4402 | } | |
4403 | ||
4404 | len = cache->key.objectid + cache->key.offset - start; | |
4405 | len = min(len, end + 1 - start); | |
4406 | ||
4407 | if (start < cache->last_byte_to_unpin) { | |
4408 | len = min(len, cache->last_byte_to_unpin - start); | |
4409 | btrfs_add_free_space(cache, start, len); | |
4410 | } | |
4411 | ||
f0486c68 YZ |
4412 | start += len; |
4413 | ||
11833d66 YZ |
4414 | spin_lock(&cache->space_info->lock); |
4415 | spin_lock(&cache->lock); | |
4416 | cache->pinned -= len; | |
4417 | cache->space_info->bytes_pinned -= len; | |
f0486c68 YZ |
4418 | if (cache->ro) { |
4419 | cache->space_info->bytes_readonly += len; | |
4420 | } else if (cache->reserved_pinned > 0) { | |
4421 | len = min(len, cache->reserved_pinned); | |
4422 | cache->reserved_pinned -= len; | |
4423 | cache->space_info->bytes_reserved += len; | |
4424 | } | |
11833d66 YZ |
4425 | spin_unlock(&cache->lock); |
4426 | spin_unlock(&cache->space_info->lock); | |
ccd467d6 | 4427 | } |
11833d66 YZ |
4428 | |
4429 | if (cache) | |
4430 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
4431 | return 0; |
4432 | } | |
4433 | ||
4434 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 4435 | struct btrfs_root *root) |
a28ec197 | 4436 | { |
11833d66 YZ |
4437 | struct btrfs_fs_info *fs_info = root->fs_info; |
4438 | struct extent_io_tree *unpin; | |
f0486c68 YZ |
4439 | struct btrfs_block_rsv *block_rsv; |
4440 | struct btrfs_block_rsv *next_rsv; | |
1a5bc167 CM |
4441 | u64 start; |
4442 | u64 end; | |
f0486c68 | 4443 | int idx; |
a28ec197 | 4444 | int ret; |
a28ec197 | 4445 | |
11833d66 YZ |
4446 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
4447 | unpin = &fs_info->freed_extents[1]; | |
4448 | else | |
4449 | unpin = &fs_info->freed_extents[0]; | |
4450 | ||
d397712b | 4451 | while (1) { |
1a5bc167 CM |
4452 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
4453 | EXTENT_DIRTY); | |
4454 | if (ret) | |
a28ec197 | 4455 | break; |
1f3c79a2 | 4456 | |
5378e607 LD |
4457 | if (btrfs_test_opt(root, DISCARD)) |
4458 | ret = btrfs_discard_extent(root, start, | |
4459 | end + 1 - start, NULL); | |
1f3c79a2 | 4460 | |
1a5bc167 | 4461 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 4462 | unpin_extent_range(root, start, end); |
b9473439 | 4463 | cond_resched(); |
a28ec197 | 4464 | } |
817d52f8 | 4465 | |
f0486c68 YZ |
4466 | mutex_lock(&fs_info->durable_block_rsv_mutex); |
4467 | list_for_each_entry_safe(block_rsv, next_rsv, | |
4468 | &fs_info->durable_block_rsv_list, list) { | |
444528b3 | 4469 | |
f0486c68 YZ |
4470 | idx = trans->transid & 0x1; |
4471 | if (block_rsv->freed[idx] > 0) { | |
4472 | block_rsv_add_bytes(block_rsv, | |
4473 | block_rsv->freed[idx], 0); | |
4474 | block_rsv->freed[idx] = 0; | |
4475 | } | |
4476 | if (atomic_read(&block_rsv->usage) == 0) { | |
4477 | btrfs_block_rsv_release(root, block_rsv, (u64)-1); | |
31840ae1 | 4478 | |
f0486c68 YZ |
4479 | if (block_rsv->freed[0] == 0 && |
4480 | block_rsv->freed[1] == 0) { | |
4481 | list_del_init(&block_rsv->list); | |
4482 | kfree(block_rsv); | |
4483 | } | |
4484 | } else { | |
4485 | btrfs_block_rsv_release(root, block_rsv, 0); | |
8ef97622 | 4486 | } |
f4b9aa8d | 4487 | } |
f0486c68 | 4488 | mutex_unlock(&fs_info->durable_block_rsv_mutex); |
31840ae1 | 4489 | |
e20d96d6 CM |
4490 | return 0; |
4491 | } | |
4492 | ||
5d4f98a2 YZ |
4493 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
4494 | struct btrfs_root *root, | |
4495 | u64 bytenr, u64 num_bytes, u64 parent, | |
4496 | u64 root_objectid, u64 owner_objectid, | |
4497 | u64 owner_offset, int refs_to_drop, | |
4498 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 4499 | { |
e2fa7227 | 4500 | struct btrfs_key key; |
5d4f98a2 | 4501 | struct btrfs_path *path; |
1261ec42 CM |
4502 | struct btrfs_fs_info *info = root->fs_info; |
4503 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 4504 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
4505 | struct btrfs_extent_item *ei; |
4506 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 4507 | int ret; |
5d4f98a2 | 4508 | int is_data; |
952fccac CM |
4509 | int extent_slot = 0; |
4510 | int found_extent = 0; | |
4511 | int num_to_del = 1; | |
5d4f98a2 YZ |
4512 | u32 item_size; |
4513 | u64 refs; | |
037e6390 | 4514 | |
5caf2a00 | 4515 | path = btrfs_alloc_path(); |
54aa1f4d CM |
4516 | if (!path) |
4517 | return -ENOMEM; | |
5f26f772 | 4518 | |
3c12ac72 | 4519 | path->reada = 1; |
b9473439 | 4520 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4521 | |
4522 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
4523 | BUG_ON(!is_data && refs_to_drop != 1); | |
4524 | ||
4525 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
4526 | bytenr, num_bytes, parent, | |
4527 | root_objectid, owner_objectid, | |
4528 | owner_offset); | |
7bb86316 | 4529 | if (ret == 0) { |
952fccac | 4530 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
4531 | while (extent_slot >= 0) { |
4532 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 4533 | extent_slot); |
5d4f98a2 | 4534 | if (key.objectid != bytenr) |
952fccac | 4535 | break; |
5d4f98a2 YZ |
4536 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
4537 | key.offset == num_bytes) { | |
952fccac CM |
4538 | found_extent = 1; |
4539 | break; | |
4540 | } | |
4541 | if (path->slots[0] - extent_slot > 5) | |
4542 | break; | |
5d4f98a2 | 4543 | extent_slot--; |
952fccac | 4544 | } |
5d4f98a2 YZ |
4545 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
4546 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
4547 | if (found_extent && item_size < sizeof(*ei)) | |
4548 | found_extent = 0; | |
4549 | #endif | |
31840ae1 | 4550 | if (!found_extent) { |
5d4f98a2 | 4551 | BUG_ON(iref); |
56bec294 | 4552 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
4553 | NULL, refs_to_drop, |
4554 | is_data); | |
31840ae1 | 4555 | BUG_ON(ret); |
b3b4aa74 | 4556 | btrfs_release_path(path); |
b9473439 | 4557 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4558 | |
4559 | key.objectid = bytenr; | |
4560 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4561 | key.offset = num_bytes; | |
4562 | ||
31840ae1 ZY |
4563 | ret = btrfs_search_slot(trans, extent_root, |
4564 | &key, path, -1, 1); | |
f3465ca4 JB |
4565 | if (ret) { |
4566 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
4567 | ", was looking for %llu\n", ret, |
4568 | (unsigned long long)bytenr); | |
f3465ca4 JB |
4569 | btrfs_print_leaf(extent_root, path->nodes[0]); |
4570 | } | |
31840ae1 ZY |
4571 | BUG_ON(ret); |
4572 | extent_slot = path->slots[0]; | |
4573 | } | |
7bb86316 CM |
4574 | } else { |
4575 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4576 | WARN_ON(1); | |
d397712b | 4577 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 4578 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 4579 | (unsigned long long)bytenr, |
56bec294 | 4580 | (unsigned long long)parent, |
d397712b | 4581 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
4582 | (unsigned long long)owner_objectid, |
4583 | (unsigned long long)owner_offset); | |
7bb86316 | 4584 | } |
5f39d397 CM |
4585 | |
4586 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
4587 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
4588 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
4589 | if (item_size < sizeof(*ei)) { | |
4590 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
4591 | ret = convert_extent_item_v0(trans, extent_root, path, | |
4592 | owner_objectid, 0); | |
4593 | BUG_ON(ret < 0); | |
4594 | ||
b3b4aa74 | 4595 | btrfs_release_path(path); |
5d4f98a2 YZ |
4596 | path->leave_spinning = 1; |
4597 | ||
4598 | key.objectid = bytenr; | |
4599 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4600 | key.offset = num_bytes; | |
4601 | ||
4602 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
4603 | -1, 1); | |
4604 | if (ret) { | |
4605 | printk(KERN_ERR "umm, got %d back from search" | |
4606 | ", was looking for %llu\n", ret, | |
4607 | (unsigned long long)bytenr); | |
4608 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4609 | } | |
4610 | BUG_ON(ret); | |
4611 | extent_slot = path->slots[0]; | |
4612 | leaf = path->nodes[0]; | |
4613 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
4614 | } | |
4615 | #endif | |
4616 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 4617 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 4618 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4619 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
4620 | struct btrfs_tree_block_info *bi; | |
4621 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
4622 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
4623 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
4624 | } | |
56bec294 | 4625 | |
5d4f98a2 | 4626 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
4627 | BUG_ON(refs < refs_to_drop); |
4628 | refs -= refs_to_drop; | |
5f39d397 | 4629 | |
5d4f98a2 YZ |
4630 | if (refs > 0) { |
4631 | if (extent_op) | |
4632 | __run_delayed_extent_op(extent_op, leaf, ei); | |
4633 | /* | |
4634 | * In the case of inline back ref, reference count will | |
4635 | * be updated by remove_extent_backref | |
952fccac | 4636 | */ |
5d4f98a2 YZ |
4637 | if (iref) { |
4638 | BUG_ON(!found_extent); | |
4639 | } else { | |
4640 | btrfs_set_extent_refs(leaf, ei, refs); | |
4641 | btrfs_mark_buffer_dirty(leaf); | |
4642 | } | |
4643 | if (found_extent) { | |
4644 | ret = remove_extent_backref(trans, extent_root, path, | |
4645 | iref, refs_to_drop, | |
4646 | is_data); | |
952fccac CM |
4647 | BUG_ON(ret); |
4648 | } | |
5d4f98a2 | 4649 | } else { |
5d4f98a2 YZ |
4650 | if (found_extent) { |
4651 | BUG_ON(is_data && refs_to_drop != | |
4652 | extent_data_ref_count(root, path, iref)); | |
4653 | if (iref) { | |
4654 | BUG_ON(path->slots[0] != extent_slot); | |
4655 | } else { | |
4656 | BUG_ON(path->slots[0] != extent_slot + 1); | |
4657 | path->slots[0] = extent_slot; | |
4658 | num_to_del = 2; | |
4659 | } | |
78fae27e | 4660 | } |
b9473439 | 4661 | |
952fccac CM |
4662 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
4663 | num_to_del); | |
31840ae1 | 4664 | BUG_ON(ret); |
b3b4aa74 | 4665 | btrfs_release_path(path); |
21af804c | 4666 | |
5d4f98a2 | 4667 | if (is_data) { |
459931ec CM |
4668 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
4669 | BUG_ON(ret); | |
d57e62b8 CM |
4670 | } else { |
4671 | invalidate_mapping_pages(info->btree_inode->i_mapping, | |
4672 | bytenr >> PAGE_CACHE_SHIFT, | |
4673 | (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT); | |
459931ec CM |
4674 | } |
4675 | ||
f0486c68 | 4676 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
dcbdd4dc | 4677 | BUG_ON(ret); |
a28ec197 | 4678 | } |
5caf2a00 | 4679 | btrfs_free_path(path); |
a28ec197 CM |
4680 | return ret; |
4681 | } | |
4682 | ||
1887be66 | 4683 | /* |
f0486c68 | 4684 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
4685 | * delayed ref for that extent as well. This searches the delayed ref tree for |
4686 | * a given extent, and if there are no other delayed refs to be processed, it | |
4687 | * removes it from the tree. | |
4688 | */ | |
4689 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
4690 | struct btrfs_root *root, u64 bytenr) | |
4691 | { | |
4692 | struct btrfs_delayed_ref_head *head; | |
4693 | struct btrfs_delayed_ref_root *delayed_refs; | |
4694 | struct btrfs_delayed_ref_node *ref; | |
4695 | struct rb_node *node; | |
f0486c68 | 4696 | int ret = 0; |
1887be66 CM |
4697 | |
4698 | delayed_refs = &trans->transaction->delayed_refs; | |
4699 | spin_lock(&delayed_refs->lock); | |
4700 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
4701 | if (!head) | |
4702 | goto out; | |
4703 | ||
4704 | node = rb_prev(&head->node.rb_node); | |
4705 | if (!node) | |
4706 | goto out; | |
4707 | ||
4708 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
4709 | ||
4710 | /* there are still entries for this ref, we can't drop it */ | |
4711 | if (ref->bytenr == bytenr) | |
4712 | goto out; | |
4713 | ||
5d4f98a2 YZ |
4714 | if (head->extent_op) { |
4715 | if (!head->must_insert_reserved) | |
4716 | goto out; | |
4717 | kfree(head->extent_op); | |
4718 | head->extent_op = NULL; | |
4719 | } | |
4720 | ||
1887be66 CM |
4721 | /* |
4722 | * waiting for the lock here would deadlock. If someone else has it | |
4723 | * locked they are already in the process of dropping it anyway | |
4724 | */ | |
4725 | if (!mutex_trylock(&head->mutex)) | |
4726 | goto out; | |
4727 | ||
4728 | /* | |
4729 | * at this point we have a head with no other entries. Go | |
4730 | * ahead and process it. | |
4731 | */ | |
4732 | head->node.in_tree = 0; | |
4733 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 4734 | |
1887be66 CM |
4735 | delayed_refs->num_entries--; |
4736 | ||
4737 | /* | |
4738 | * we don't take a ref on the node because we're removing it from the | |
4739 | * tree, so we just steal the ref the tree was holding. | |
4740 | */ | |
c3e69d58 CM |
4741 | delayed_refs->num_heads--; |
4742 | if (list_empty(&head->cluster)) | |
4743 | delayed_refs->num_heads_ready--; | |
4744 | ||
4745 | list_del_init(&head->cluster); | |
1887be66 CM |
4746 | spin_unlock(&delayed_refs->lock); |
4747 | ||
f0486c68 YZ |
4748 | BUG_ON(head->extent_op); |
4749 | if (head->must_insert_reserved) | |
4750 | ret = 1; | |
4751 | ||
4752 | mutex_unlock(&head->mutex); | |
1887be66 | 4753 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 4754 | return ret; |
1887be66 CM |
4755 | out: |
4756 | spin_unlock(&delayed_refs->lock); | |
4757 | return 0; | |
4758 | } | |
4759 | ||
f0486c68 YZ |
4760 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
4761 | struct btrfs_root *root, | |
4762 | struct extent_buffer *buf, | |
4763 | u64 parent, int last_ref) | |
4764 | { | |
4765 | struct btrfs_block_rsv *block_rsv; | |
4766 | struct btrfs_block_group_cache *cache = NULL; | |
4767 | int ret; | |
4768 | ||
4769 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
4770 | ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len, | |
4771 | parent, root->root_key.objectid, | |
4772 | btrfs_header_level(buf), | |
4773 | BTRFS_DROP_DELAYED_REF, NULL); | |
4774 | BUG_ON(ret); | |
4775 | } | |
4776 | ||
4777 | if (!last_ref) | |
4778 | return; | |
4779 | ||
4780 | block_rsv = get_block_rsv(trans, root); | |
4781 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); | |
3bf84a5a YZ |
4782 | if (block_rsv->space_info != cache->space_info) |
4783 | goto out; | |
f0486c68 YZ |
4784 | |
4785 | if (btrfs_header_generation(buf) == trans->transid) { | |
4786 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
4787 | ret = check_ref_cleanup(trans, root, buf->start); | |
4788 | if (!ret) | |
4789 | goto pin; | |
4790 | } | |
4791 | ||
4792 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
4793 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
4794 | goto pin; | |
4795 | } | |
4796 | ||
4797 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
4798 | ||
4799 | btrfs_add_free_space(cache, buf->start, buf->len); | |
b4d00d56 | 4800 | ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0); |
f0486c68 YZ |
4801 | if (ret == -EAGAIN) { |
4802 | /* block group became read-only */ | |
b4d00d56 | 4803 | btrfs_update_reserved_bytes(cache, buf->len, 0, 1); |
f0486c68 YZ |
4804 | goto out; |
4805 | } | |
4806 | ||
4807 | ret = 1; | |
4808 | spin_lock(&block_rsv->lock); | |
4809 | if (block_rsv->reserved < block_rsv->size) { | |
4810 | block_rsv->reserved += buf->len; | |
4811 | ret = 0; | |
4812 | } | |
4813 | spin_unlock(&block_rsv->lock); | |
4814 | ||
4815 | if (ret) { | |
4816 | spin_lock(&cache->space_info->lock); | |
4817 | cache->space_info->bytes_reserved -= buf->len; | |
36e39c40 | 4818 | cache->space_info->reservation_progress++; |
f0486c68 YZ |
4819 | spin_unlock(&cache->space_info->lock); |
4820 | } | |
4821 | goto out; | |
4822 | } | |
4823 | pin: | |
4824 | if (block_rsv->durable && !cache->ro) { | |
4825 | ret = 0; | |
4826 | spin_lock(&cache->lock); | |
4827 | if (!cache->ro) { | |
4828 | cache->reserved_pinned += buf->len; | |
4829 | ret = 1; | |
4830 | } | |
4831 | spin_unlock(&cache->lock); | |
4832 | ||
4833 | if (ret) { | |
4834 | spin_lock(&block_rsv->lock); | |
4835 | block_rsv->freed[trans->transid & 0x1] += buf->len; | |
4836 | spin_unlock(&block_rsv->lock); | |
4837 | } | |
4838 | } | |
4839 | out: | |
a826d6dc JB |
4840 | /* |
4841 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
4842 | * anymore. | |
4843 | */ | |
4844 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
4845 | btrfs_put_block_group(cache); |
4846 | } | |
4847 | ||
925baedd | 4848 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
4849 | struct btrfs_root *root, |
4850 | u64 bytenr, u64 num_bytes, u64 parent, | |
5d4f98a2 | 4851 | u64 root_objectid, u64 owner, u64 offset) |
925baedd CM |
4852 | { |
4853 | int ret; | |
4854 | ||
56bec294 CM |
4855 | /* |
4856 | * tree log blocks never actually go into the extent allocation | |
4857 | * tree, just update pinning info and exit early. | |
56bec294 | 4858 | */ |
5d4f98a2 YZ |
4859 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
4860 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 4861 | /* unlocks the pinned mutex */ |
11833d66 | 4862 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 4863 | ret = 0; |
5d4f98a2 YZ |
4864 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
4865 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
4866 | parent, root_objectid, (int)owner, | |
4867 | BTRFS_DROP_DELAYED_REF, NULL); | |
1887be66 | 4868 | BUG_ON(ret); |
5d4f98a2 YZ |
4869 | } else { |
4870 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
4871 | parent, root_objectid, owner, | |
4872 | offset, BTRFS_DROP_DELAYED_REF, NULL); | |
4873 | BUG_ON(ret); | |
56bec294 | 4874 | } |
925baedd CM |
4875 | return ret; |
4876 | } | |
4877 | ||
87ee04eb CM |
4878 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
4879 | { | |
4880 | u64 mask = ((u64)root->stripesize - 1); | |
4881 | u64 ret = (val + mask) & ~mask; | |
4882 | return ret; | |
4883 | } | |
4884 | ||
817d52f8 JB |
4885 | /* |
4886 | * when we wait for progress in the block group caching, its because | |
4887 | * our allocation attempt failed at least once. So, we must sleep | |
4888 | * and let some progress happen before we try again. | |
4889 | * | |
4890 | * This function will sleep at least once waiting for new free space to | |
4891 | * show up, and then it will check the block group free space numbers | |
4892 | * for our min num_bytes. Another option is to have it go ahead | |
4893 | * and look in the rbtree for a free extent of a given size, but this | |
4894 | * is a good start. | |
4895 | */ | |
4896 | static noinline int | |
4897 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
4898 | u64 num_bytes) | |
4899 | { | |
11833d66 | 4900 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
4901 | DEFINE_WAIT(wait); |
4902 | ||
11833d66 YZ |
4903 | caching_ctl = get_caching_control(cache); |
4904 | if (!caching_ctl) | |
817d52f8 | 4905 | return 0; |
817d52f8 | 4906 | |
11833d66 | 4907 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
817d52f8 | 4908 | (cache->free_space >= num_bytes)); |
11833d66 YZ |
4909 | |
4910 | put_caching_control(caching_ctl); | |
4911 | return 0; | |
4912 | } | |
4913 | ||
4914 | static noinline int | |
4915 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
4916 | { | |
4917 | struct btrfs_caching_control *caching_ctl; | |
4918 | DEFINE_WAIT(wait); | |
4919 | ||
4920 | caching_ctl = get_caching_control(cache); | |
4921 | if (!caching_ctl) | |
4922 | return 0; | |
4923 | ||
4924 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
4925 | ||
4926 | put_caching_control(caching_ctl); | |
817d52f8 JB |
4927 | return 0; |
4928 | } | |
4929 | ||
b742bb82 YZ |
4930 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
4931 | { | |
4932 | int index; | |
4933 | if (cache->flags & BTRFS_BLOCK_GROUP_RAID10) | |
4934 | index = 0; | |
4935 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1) | |
4936 | index = 1; | |
4937 | else if (cache->flags & BTRFS_BLOCK_GROUP_DUP) | |
4938 | index = 2; | |
4939 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0) | |
4940 | index = 3; | |
4941 | else | |
4942 | index = 4; | |
4943 | return index; | |
4944 | } | |
4945 | ||
817d52f8 | 4946 | enum btrfs_loop_type { |
ccf0e725 | 4947 | LOOP_FIND_IDEAL = 0, |
817d52f8 JB |
4948 | LOOP_CACHING_NOWAIT = 1, |
4949 | LOOP_CACHING_WAIT = 2, | |
4950 | LOOP_ALLOC_CHUNK = 3, | |
4951 | LOOP_NO_EMPTY_SIZE = 4, | |
4952 | }; | |
4953 | ||
fec577fb CM |
4954 | /* |
4955 | * walks the btree of allocated extents and find a hole of a given size. | |
4956 | * The key ins is changed to record the hole: | |
4957 | * ins->objectid == block start | |
62e2749e | 4958 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
4959 | * ins->offset == number of blocks |
4960 | * Any available blocks before search_start are skipped. | |
4961 | */ | |
d397712b | 4962 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
4963 | struct btrfs_root *orig_root, |
4964 | u64 num_bytes, u64 empty_size, | |
4965 | u64 search_start, u64 search_end, | |
4966 | u64 hint_byte, struct btrfs_key *ins, | |
98ed5174 | 4967 | int data) |
fec577fb | 4968 | { |
80eb234a | 4969 | int ret = 0; |
d397712b | 4970 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 4971 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 4972 | struct btrfs_block_group_cache *block_group = NULL; |
239b14b3 | 4973 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 4974 | int allowed_chunk_alloc = 0; |
ccf0e725 | 4975 | int done_chunk_alloc = 0; |
80eb234a | 4976 | struct btrfs_space_info *space_info; |
fa9c0d79 CM |
4977 | int last_ptr_loop = 0; |
4978 | int loop = 0; | |
f0486c68 | 4979 | int index = 0; |
817d52f8 | 4980 | bool found_uncached_bg = false; |
0a24325e | 4981 | bool failed_cluster_refill = false; |
1cdda9b8 | 4982 | bool failed_alloc = false; |
67377734 | 4983 | bool use_cluster = true; |
ccf0e725 JB |
4984 | u64 ideal_cache_percent = 0; |
4985 | u64 ideal_cache_offset = 0; | |
fec577fb | 4986 | |
db94535d | 4987 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 4988 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
4989 | ins->objectid = 0; |
4990 | ins->offset = 0; | |
b1a4d965 | 4991 | |
2552d17e | 4992 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 JB |
4993 | if (!space_info) { |
4994 | printk(KERN_ERR "No space info for %d\n", data); | |
4995 | return -ENOSPC; | |
4996 | } | |
2552d17e | 4997 | |
67377734 JB |
4998 | /* |
4999 | * If the space info is for both data and metadata it means we have a | |
5000 | * small filesystem and we can't use the clustering stuff. | |
5001 | */ | |
5002 | if (btrfs_mixed_space_info(space_info)) | |
5003 | use_cluster = false; | |
5004 | ||
0ef3e66b CM |
5005 | if (orig_root->ref_cows || empty_size) |
5006 | allowed_chunk_alloc = 1; | |
5007 | ||
67377734 | 5008 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5009 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5010 | if (!btrfs_test_opt(root, SSD)) |
5011 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5012 | } |
5013 | ||
67377734 JB |
5014 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5015 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5016 | last_ptr = &root->fs_info->data_alloc_cluster; |
5017 | } | |
0f9dd46c | 5018 | |
239b14b3 | 5019 | if (last_ptr) { |
fa9c0d79 CM |
5020 | spin_lock(&last_ptr->lock); |
5021 | if (last_ptr->block_group) | |
5022 | hint_byte = last_ptr->window_start; | |
5023 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5024 | } |
fa9c0d79 | 5025 | |
a061fc8d | 5026 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5027 | search_start = max(search_start, hint_byte); |
0b86a832 | 5028 | |
817d52f8 | 5029 | if (!last_ptr) |
fa9c0d79 | 5030 | empty_cluster = 0; |
fa9c0d79 | 5031 | |
2552d17e | 5032 | if (search_start == hint_byte) { |
ccf0e725 | 5033 | ideal_cache: |
2552d17e JB |
5034 | block_group = btrfs_lookup_block_group(root->fs_info, |
5035 | search_start); | |
817d52f8 JB |
5036 | /* |
5037 | * we don't want to use the block group if it doesn't match our | |
5038 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5039 | * |
5040 | * However if we are re-searching with an ideal block group | |
5041 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5042 | */ |
5043 | if (block_group && block_group_bits(block_group, data) && | |
ccf0e725 JB |
5044 | (block_group->cached != BTRFS_CACHE_NO || |
5045 | search_start == ideal_cache_offset)) { | |
2552d17e | 5046 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5047 | if (list_empty(&block_group->list) || |
5048 | block_group->ro) { | |
5049 | /* | |
5050 | * someone is removing this block group, | |
5051 | * we can't jump into the have_block_group | |
5052 | * target because our list pointers are not | |
5053 | * valid | |
5054 | */ | |
5055 | btrfs_put_block_group(block_group); | |
5056 | up_read(&space_info->groups_sem); | |
ccf0e725 | 5057 | } else { |
b742bb82 | 5058 | index = get_block_group_index(block_group); |
44fb5511 | 5059 | goto have_block_group; |
ccf0e725 | 5060 | } |
2552d17e | 5061 | } else if (block_group) { |
fa9c0d79 | 5062 | btrfs_put_block_group(block_group); |
2552d17e | 5063 | } |
42e70e7a | 5064 | } |
2552d17e | 5065 | search: |
80eb234a | 5066 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
5067 | list_for_each_entry(block_group, &space_info->block_groups[index], |
5068 | list) { | |
6226cb0a | 5069 | u64 offset; |
817d52f8 | 5070 | int cached; |
8a1413a2 | 5071 | |
11dfe35a | 5072 | btrfs_get_block_group(block_group); |
2552d17e | 5073 | search_start = block_group->key.objectid; |
42e70e7a | 5074 | |
83a50de9 CM |
5075 | /* |
5076 | * this can happen if we end up cycling through all the | |
5077 | * raid types, but we want to make sure we only allocate | |
5078 | * for the proper type. | |
5079 | */ | |
5080 | if (!block_group_bits(block_group, data)) { | |
5081 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
5082 | BTRFS_BLOCK_GROUP_RAID1 | | |
5083 | BTRFS_BLOCK_GROUP_RAID10; | |
5084 | ||
5085 | /* | |
5086 | * if they asked for extra copies and this block group | |
5087 | * doesn't provide them, bail. This does allow us to | |
5088 | * fill raid0 from raid1. | |
5089 | */ | |
5090 | if ((data & extra) && !(block_group->flags & extra)) | |
5091 | goto loop; | |
5092 | } | |
5093 | ||
2552d17e | 5094 | have_block_group: |
817d52f8 | 5095 | if (unlikely(block_group->cached == BTRFS_CACHE_NO)) { |
ccf0e725 JB |
5096 | u64 free_percent; |
5097 | ||
b8399dee JB |
5098 | ret = cache_block_group(block_group, trans, |
5099 | orig_root, 1); | |
9d66e233 JB |
5100 | if (block_group->cached == BTRFS_CACHE_FINISHED) |
5101 | goto have_block_group; | |
5102 | ||
ccf0e725 JB |
5103 | free_percent = btrfs_block_group_used(&block_group->item); |
5104 | free_percent *= 100; | |
5105 | free_percent = div64_u64(free_percent, | |
5106 | block_group->key.offset); | |
5107 | free_percent = 100 - free_percent; | |
5108 | if (free_percent > ideal_cache_percent && | |
5109 | likely(!block_group->ro)) { | |
5110 | ideal_cache_offset = block_group->key.objectid; | |
5111 | ideal_cache_percent = free_percent; | |
5112 | } | |
5113 | ||
817d52f8 | 5114 | /* |
ccf0e725 JB |
5115 | * We only want to start kthread caching if we are at |
5116 | * the point where we will wait for caching to make | |
5117 | * progress, or if our ideal search is over and we've | |
5118 | * found somebody to start caching. | |
817d52f8 JB |
5119 | */ |
5120 | if (loop > LOOP_CACHING_NOWAIT || | |
ccf0e725 JB |
5121 | (loop > LOOP_FIND_IDEAL && |
5122 | atomic_read(&space_info->caching_threads) < 2)) { | |
b8399dee JB |
5123 | ret = cache_block_group(block_group, trans, |
5124 | orig_root, 0); | |
817d52f8 | 5125 | BUG_ON(ret); |
2552d17e | 5126 | } |
817d52f8 JB |
5127 | found_uncached_bg = true; |
5128 | ||
ccf0e725 JB |
5129 | /* |
5130 | * If loop is set for cached only, try the next block | |
5131 | * group. | |
5132 | */ | |
5133 | if (loop == LOOP_FIND_IDEAL) | |
817d52f8 JB |
5134 | goto loop; |
5135 | } | |
5136 | ||
ccf0e725 JB |
5137 | cached = block_group_cache_done(block_group); |
5138 | if (unlikely(!cached)) | |
5139 | found_uncached_bg = true; | |
5140 | ||
ea6a478e | 5141 | if (unlikely(block_group->ro)) |
2552d17e | 5142 | goto loop; |
0f9dd46c | 5143 | |
0a24325e JB |
5144 | /* |
5145 | * Ok we want to try and use the cluster allocator, so lets look | |
5146 | * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will | |
5147 | * have tried the cluster allocator plenty of times at this | |
5148 | * point and not have found anything, so we are likely way too | |
5149 | * fragmented for the clustering stuff to find anything, so lets | |
5150 | * just skip it and let the allocator find whatever block it can | |
5151 | * find | |
5152 | */ | |
5153 | if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) { | |
fa9c0d79 CM |
5154 | /* |
5155 | * the refill lock keeps out other | |
5156 | * people trying to start a new cluster | |
5157 | */ | |
5158 | spin_lock(&last_ptr->refill_lock); | |
44fb5511 CM |
5159 | if (last_ptr->block_group && |
5160 | (last_ptr->block_group->ro || | |
5161 | !block_group_bits(last_ptr->block_group, data))) { | |
5162 | offset = 0; | |
5163 | goto refill_cluster; | |
5164 | } | |
5165 | ||
fa9c0d79 CM |
5166 | offset = btrfs_alloc_from_cluster(block_group, last_ptr, |
5167 | num_bytes, search_start); | |
5168 | if (offset) { | |
5169 | /* we have a block, we're done */ | |
5170 | spin_unlock(&last_ptr->refill_lock); | |
5171 | goto checks; | |
5172 | } | |
5173 | ||
5174 | spin_lock(&last_ptr->lock); | |
5175 | /* | |
5176 | * whoops, this cluster doesn't actually point to | |
5177 | * this block group. Get a ref on the block | |
5178 | * group is does point to and try again | |
5179 | */ | |
5180 | if (!last_ptr_loop && last_ptr->block_group && | |
5181 | last_ptr->block_group != block_group) { | |
5182 | ||
5183 | btrfs_put_block_group(block_group); | |
5184 | block_group = last_ptr->block_group; | |
11dfe35a | 5185 | btrfs_get_block_group(block_group); |
fa9c0d79 CM |
5186 | spin_unlock(&last_ptr->lock); |
5187 | spin_unlock(&last_ptr->refill_lock); | |
5188 | ||
5189 | last_ptr_loop = 1; | |
5190 | search_start = block_group->key.objectid; | |
44fb5511 CM |
5191 | /* |
5192 | * we know this block group is properly | |
5193 | * in the list because | |
5194 | * btrfs_remove_block_group, drops the | |
5195 | * cluster before it removes the block | |
5196 | * group from the list | |
5197 | */ | |
fa9c0d79 CM |
5198 | goto have_block_group; |
5199 | } | |
5200 | spin_unlock(&last_ptr->lock); | |
44fb5511 | 5201 | refill_cluster: |
fa9c0d79 CM |
5202 | /* |
5203 | * this cluster didn't work out, free it and | |
5204 | * start over | |
5205 | */ | |
5206 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5207 | ||
5208 | last_ptr_loop = 0; | |
5209 | ||
5210 | /* allocate a cluster in this block group */ | |
451d7585 | 5211 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 CM |
5212 | block_group, last_ptr, |
5213 | offset, num_bytes, | |
5214 | empty_cluster + empty_size); | |
5215 | if (ret == 0) { | |
5216 | /* | |
5217 | * now pull our allocation out of this | |
5218 | * cluster | |
5219 | */ | |
5220 | offset = btrfs_alloc_from_cluster(block_group, | |
5221 | last_ptr, num_bytes, | |
5222 | search_start); | |
5223 | if (offset) { | |
5224 | /* we found one, proceed */ | |
5225 | spin_unlock(&last_ptr->refill_lock); | |
5226 | goto checks; | |
5227 | } | |
0a24325e JB |
5228 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5229 | && !failed_cluster_refill) { | |
817d52f8 JB |
5230 | spin_unlock(&last_ptr->refill_lock); |
5231 | ||
0a24325e | 5232 | failed_cluster_refill = true; |
817d52f8 JB |
5233 | wait_block_group_cache_progress(block_group, |
5234 | num_bytes + empty_cluster + empty_size); | |
5235 | goto have_block_group; | |
fa9c0d79 | 5236 | } |
817d52f8 | 5237 | |
fa9c0d79 CM |
5238 | /* |
5239 | * at this point we either didn't find a cluster | |
5240 | * or we weren't able to allocate a block from our | |
5241 | * cluster. Free the cluster we've been trying | |
5242 | * to use, and go to the next block group | |
5243 | */ | |
0a24325e | 5244 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5245 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5246 | goto loop; |
fa9c0d79 CM |
5247 | } |
5248 | ||
6226cb0a JB |
5249 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5250 | num_bytes, empty_size); | |
1cdda9b8 JB |
5251 | /* |
5252 | * If we didn't find a chunk, and we haven't failed on this | |
5253 | * block group before, and this block group is in the middle of | |
5254 | * caching and we are ok with waiting, then go ahead and wait | |
5255 | * for progress to be made, and set failed_alloc to true. | |
5256 | * | |
5257 | * If failed_alloc is true then we've already waited on this | |
5258 | * block group once and should move on to the next block group. | |
5259 | */ | |
5260 | if (!offset && !failed_alloc && !cached && | |
5261 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5262 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5263 | num_bytes + empty_size); |
5264 | failed_alloc = true; | |
817d52f8 | 5265 | goto have_block_group; |
1cdda9b8 JB |
5266 | } else if (!offset) { |
5267 | goto loop; | |
817d52f8 | 5268 | } |
fa9c0d79 | 5269 | checks: |
6226cb0a | 5270 | search_start = stripe_align(root, offset); |
2552d17e | 5271 | /* move on to the next group */ |
6226cb0a JB |
5272 | if (search_start + num_bytes >= search_end) { |
5273 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 5274 | goto loop; |
6226cb0a | 5275 | } |
25179201 | 5276 | |
2552d17e JB |
5277 | /* move on to the next group */ |
5278 | if (search_start + num_bytes > | |
6226cb0a JB |
5279 | block_group->key.objectid + block_group->key.offset) { |
5280 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 5281 | goto loop; |
6226cb0a | 5282 | } |
f5a31e16 | 5283 | |
f0486c68 YZ |
5284 | ins->objectid = search_start; |
5285 | ins->offset = num_bytes; | |
2552d17e | 5286 | |
f0486c68 YZ |
5287 | if (offset < search_start) |
5288 | btrfs_add_free_space(block_group, offset, | |
5289 | search_start - offset); | |
5290 | BUG_ON(offset > search_start); | |
2552d17e | 5291 | |
b4d00d56 | 5292 | ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1, |
f0486c68 YZ |
5293 | (data & BTRFS_BLOCK_GROUP_DATA)); |
5294 | if (ret == -EAGAIN) { | |
6226cb0a | 5295 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 5296 | goto loop; |
0f9dd46c | 5297 | } |
0b86a832 | 5298 | |
f0486c68 | 5299 | /* we are all good, lets return */ |
2552d17e JB |
5300 | ins->objectid = search_start; |
5301 | ins->offset = num_bytes; | |
d2fb3437 | 5302 | |
6226cb0a JB |
5303 | if (offset < search_start) |
5304 | btrfs_add_free_space(block_group, offset, | |
5305 | search_start - offset); | |
5306 | BUG_ON(offset > search_start); | |
2552d17e JB |
5307 | break; |
5308 | loop: | |
0a24325e | 5309 | failed_cluster_refill = false; |
1cdda9b8 | 5310 | failed_alloc = false; |
b742bb82 | 5311 | BUG_ON(index != get_block_group_index(block_group)); |
fa9c0d79 | 5312 | btrfs_put_block_group(block_group); |
2552d17e JB |
5313 | } |
5314 | up_read(&space_info->groups_sem); | |
5315 | ||
b742bb82 YZ |
5316 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5317 | goto search; | |
5318 | ||
ccf0e725 JB |
5319 | /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for |
5320 | * for them to make caching progress. Also | |
5321 | * determine the best possible bg to cache | |
5322 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
5323 | * caching kthreads as we move along | |
817d52f8 JB |
5324 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5325 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5326 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5327 | * again | |
fa9c0d79 | 5328 | */ |
817d52f8 JB |
5329 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE && |
5330 | (found_uncached_bg || empty_size || empty_cluster || | |
5331 | allowed_chunk_alloc)) { | |
b742bb82 | 5332 | index = 0; |
ccf0e725 | 5333 | if (loop == LOOP_FIND_IDEAL && found_uncached_bg) { |
817d52f8 | 5334 | found_uncached_bg = false; |
ccf0e725 JB |
5335 | loop++; |
5336 | if (!ideal_cache_percent && | |
5337 | atomic_read(&space_info->caching_threads)) | |
817d52f8 | 5338 | goto search; |
ccf0e725 JB |
5339 | |
5340 | /* | |
5341 | * 1 of the following 2 things have happened so far | |
5342 | * | |
5343 | * 1) We found an ideal block group for caching that | |
5344 | * is mostly full and will cache quickly, so we might | |
5345 | * as well wait for it. | |
5346 | * | |
5347 | * 2) We searched for cached only and we didn't find | |
5348 | * anything, and we didn't start any caching kthreads | |
5349 | * either, so chances are we will loop through and | |
5350 | * start a couple caching kthreads, and then come back | |
5351 | * around and just wait for them. This will be slower | |
5352 | * because we will have 2 caching kthreads reading at | |
5353 | * the same time when we could have just started one | |
5354 | * and waited for it to get far enough to give us an | |
5355 | * allocation, so go ahead and go to the wait caching | |
5356 | * loop. | |
5357 | */ | |
5358 | loop = LOOP_CACHING_WAIT; | |
5359 | search_start = ideal_cache_offset; | |
5360 | ideal_cache_percent = 0; | |
5361 | goto ideal_cache; | |
5362 | } else if (loop == LOOP_FIND_IDEAL) { | |
5363 | /* | |
5364 | * Didn't find a uncached bg, wait on anything we find | |
5365 | * next. | |
5366 | */ | |
5367 | loop = LOOP_CACHING_WAIT; | |
5368 | goto search; | |
5369 | } | |
5370 | ||
5371 | if (loop < LOOP_CACHING_WAIT) { | |
5372 | loop++; | |
5373 | goto search; | |
817d52f8 JB |
5374 | } |
5375 | ||
5376 | if (loop == LOOP_ALLOC_CHUNK) { | |
fa9c0d79 CM |
5377 | empty_size = 0; |
5378 | empty_cluster = 0; | |
5379 | } | |
2552d17e JB |
5380 | |
5381 | if (allowed_chunk_alloc) { | |
5382 | ret = do_chunk_alloc(trans, root, num_bytes + | |
0e4f8f88 CM |
5383 | 2 * 1024 * 1024, data, |
5384 | CHUNK_ALLOC_LIMITED); | |
2552d17e | 5385 | allowed_chunk_alloc = 0; |
ccf0e725 | 5386 | done_chunk_alloc = 1; |
0e4f8f88 CM |
5387 | } else if (!done_chunk_alloc && |
5388 | space_info->force_alloc == CHUNK_ALLOC_NO_FORCE) { | |
5389 | space_info->force_alloc = CHUNK_ALLOC_LIMITED; | |
2552d17e JB |
5390 | } |
5391 | ||
817d52f8 | 5392 | if (loop < LOOP_NO_EMPTY_SIZE) { |
fa9c0d79 | 5393 | loop++; |
2552d17e | 5394 | goto search; |
fa9c0d79 | 5395 | } |
2552d17e JB |
5396 | ret = -ENOSPC; |
5397 | } else if (!ins->objectid) { | |
5398 | ret = -ENOSPC; | |
f2654de4 | 5399 | } |
0b86a832 | 5400 | |
80eb234a JB |
5401 | /* we found what we needed */ |
5402 | if (ins->objectid) { | |
5403 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) | |
d2fb3437 | 5404 | trans->block_group = block_group->key.objectid; |
0f9dd46c | 5405 | |
fa9c0d79 | 5406 | btrfs_put_block_group(block_group); |
80eb234a | 5407 | ret = 0; |
be744175 | 5408 | } |
be744175 | 5409 | |
0f70abe2 | 5410 | return ret; |
fec577fb | 5411 | } |
ec44a35c | 5412 | |
9ed74f2d JB |
5413 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5414 | int dump_block_groups) | |
0f9dd46c JB |
5415 | { |
5416 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5417 | int index = 0; |
0f9dd46c | 5418 | |
9ed74f2d | 5419 | spin_lock(&info->lock); |
d397712b CM |
5420 | printk(KERN_INFO "space_info has %llu free, is %sfull\n", |
5421 | (unsigned long long)(info->total_bytes - info->bytes_used - | |
9ed74f2d | 5422 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5423 | info->bytes_readonly), |
d397712b | 5424 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5425 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5426 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5427 | (unsigned long long)info->total_bytes, |
8929ecfa | 5428 | (unsigned long long)info->bytes_used, |
21380931 | 5429 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5430 | (unsigned long long)info->bytes_reserved, |
21380931 | 5431 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5432 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5433 | spin_unlock(&info->lock); |
5434 | ||
5435 | if (!dump_block_groups) | |
5436 | return; | |
0f9dd46c | 5437 | |
80eb234a | 5438 | down_read(&info->groups_sem); |
b742bb82 YZ |
5439 | again: |
5440 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5441 | spin_lock(&cache->lock); |
d397712b CM |
5442 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
5443 | "%llu pinned %llu reserved\n", | |
5444 | (unsigned long long)cache->key.objectid, | |
5445 | (unsigned long long)cache->key.offset, | |
5446 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5447 | (unsigned long long)cache->pinned, | |
5448 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
5449 | btrfs_dump_free_space(cache, bytes); |
5450 | spin_unlock(&cache->lock); | |
5451 | } | |
b742bb82 YZ |
5452 | if (++index < BTRFS_NR_RAID_TYPES) |
5453 | goto again; | |
80eb234a | 5454 | up_read(&info->groups_sem); |
0f9dd46c | 5455 | } |
e8569813 | 5456 | |
11833d66 YZ |
5457 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5458 | struct btrfs_root *root, | |
5459 | u64 num_bytes, u64 min_alloc_size, | |
5460 | u64 empty_size, u64 hint_byte, | |
5461 | u64 search_end, struct btrfs_key *ins, | |
5462 | u64 data) | |
fec577fb CM |
5463 | { |
5464 | int ret; | |
fbdc762b | 5465 | u64 search_start = 0; |
925baedd | 5466 | |
6a63209f | 5467 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 5468 | again: |
0ef3e66b CM |
5469 | /* |
5470 | * the only place that sets empty_size is btrfs_realloc_node, which | |
5471 | * is not called recursively on allocations | |
5472 | */ | |
83d3c969 | 5473 | if (empty_size || root->ref_cows) |
6324fbf3 | 5474 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
5475 | num_bytes + 2 * 1024 * 1024, data, |
5476 | CHUNK_ALLOC_NO_FORCE); | |
0b86a832 | 5477 | |
db94535d CM |
5478 | WARN_ON(num_bytes < root->sectorsize); |
5479 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
f0486c68 YZ |
5480 | search_start, search_end, hint_byte, |
5481 | ins, data); | |
3b951516 | 5482 | |
98d20f67 CM |
5483 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
5484 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 5485 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 5486 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b | 5487 | do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 | 5488 | num_bytes, data, CHUNK_ALLOC_FORCE); |
98d20f67 CM |
5489 | goto again; |
5490 | } | |
91435650 | 5491 | if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) { |
0f9dd46c JB |
5492 | struct btrfs_space_info *sinfo; |
5493 | ||
5494 | sinfo = __find_space_info(root->fs_info, data); | |
d397712b CM |
5495 | printk(KERN_ERR "btrfs allocation failed flags %llu, " |
5496 | "wanted %llu\n", (unsigned long long)data, | |
5497 | (unsigned long long)num_bytes); | |
9ed74f2d | 5498 | dump_space_info(sinfo, num_bytes, 1); |
925baedd | 5499 | } |
0f9dd46c | 5500 | |
1abe9b8a | 5501 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
5502 | ||
0f9dd46c | 5503 | return ret; |
e6dcd2dc CM |
5504 | } |
5505 | ||
65b51a00 CM |
5506 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
5507 | { | |
0f9dd46c | 5508 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 5509 | int ret = 0; |
0f9dd46c | 5510 | |
0f9dd46c JB |
5511 | cache = btrfs_lookup_block_group(root->fs_info, start); |
5512 | if (!cache) { | |
d397712b CM |
5513 | printk(KERN_ERR "Unable to find block group for %llu\n", |
5514 | (unsigned long long)start); | |
0f9dd46c JB |
5515 | return -ENOSPC; |
5516 | } | |
1f3c79a2 | 5517 | |
5378e607 LD |
5518 | if (btrfs_test_opt(root, DISCARD)) |
5519 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 5520 | |
0f9dd46c | 5521 | btrfs_add_free_space(cache, start, len); |
b4d00d56 | 5522 | btrfs_update_reserved_bytes(cache, len, 0, 1); |
fa9c0d79 | 5523 | btrfs_put_block_group(cache); |
817d52f8 | 5524 | |
1abe9b8a | 5525 | trace_btrfs_reserved_extent_free(root, start, len); |
5526 | ||
e6dcd2dc CM |
5527 | return ret; |
5528 | } | |
5529 | ||
5d4f98a2 YZ |
5530 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5531 | struct btrfs_root *root, | |
5532 | u64 parent, u64 root_objectid, | |
5533 | u64 flags, u64 owner, u64 offset, | |
5534 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
5535 | { |
5536 | int ret; | |
5d4f98a2 | 5537 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 5538 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 5539 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 5540 | struct btrfs_path *path; |
5d4f98a2 YZ |
5541 | struct extent_buffer *leaf; |
5542 | int type; | |
5543 | u32 size; | |
26b8003f | 5544 | |
5d4f98a2 YZ |
5545 | if (parent > 0) |
5546 | type = BTRFS_SHARED_DATA_REF_KEY; | |
5547 | else | |
5548 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 5549 | |
5d4f98a2 | 5550 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
5551 | |
5552 | path = btrfs_alloc_path(); | |
db5b493a TI |
5553 | if (!path) |
5554 | return -ENOMEM; | |
47e4bb98 | 5555 | |
b9473439 | 5556 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5557 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
5558 | ins, size); | |
ccd467d6 | 5559 | BUG_ON(ret); |
0f9dd46c | 5560 | |
5d4f98a2 YZ |
5561 | leaf = path->nodes[0]; |
5562 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 5563 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5564 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
5565 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5566 | btrfs_set_extent_flags(leaf, extent_item, | |
5567 | flags | BTRFS_EXTENT_FLAG_DATA); | |
5568 | ||
5569 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
5570 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
5571 | if (parent > 0) { | |
5572 | struct btrfs_shared_data_ref *ref; | |
5573 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
5574 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5575 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
5576 | } else { | |
5577 | struct btrfs_extent_data_ref *ref; | |
5578 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
5579 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
5580 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
5581 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
5582 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
5583 | } | |
47e4bb98 CM |
5584 | |
5585 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 5586 | btrfs_free_path(path); |
f510cfec | 5587 | |
f0486c68 | 5588 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
f5947066 | 5589 | if (ret) { |
d397712b CM |
5590 | printk(KERN_ERR "btrfs update block group failed for %llu " |
5591 | "%llu\n", (unsigned long long)ins->objectid, | |
5592 | (unsigned long long)ins->offset); | |
f5947066 CM |
5593 | BUG(); |
5594 | } | |
e6dcd2dc CM |
5595 | return ret; |
5596 | } | |
5597 | ||
5d4f98a2 YZ |
5598 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
5599 | struct btrfs_root *root, | |
5600 | u64 parent, u64 root_objectid, | |
5601 | u64 flags, struct btrfs_disk_key *key, | |
5602 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
5603 | { |
5604 | int ret; | |
5d4f98a2 YZ |
5605 | struct btrfs_fs_info *fs_info = root->fs_info; |
5606 | struct btrfs_extent_item *extent_item; | |
5607 | struct btrfs_tree_block_info *block_info; | |
5608 | struct btrfs_extent_inline_ref *iref; | |
5609 | struct btrfs_path *path; | |
5610 | struct extent_buffer *leaf; | |
5611 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 5612 | |
5d4f98a2 YZ |
5613 | path = btrfs_alloc_path(); |
5614 | BUG_ON(!path); | |
56bec294 | 5615 | |
5d4f98a2 YZ |
5616 | path->leave_spinning = 1; |
5617 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
5618 | ins, size); | |
56bec294 | 5619 | BUG_ON(ret); |
5d4f98a2 YZ |
5620 | |
5621 | leaf = path->nodes[0]; | |
5622 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
5623 | struct btrfs_extent_item); | |
5624 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
5625 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5626 | btrfs_set_extent_flags(leaf, extent_item, | |
5627 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5628 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
5629 | ||
5630 | btrfs_set_tree_block_key(leaf, block_info, key); | |
5631 | btrfs_set_tree_block_level(leaf, block_info, level); | |
5632 | ||
5633 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
5634 | if (parent > 0) { | |
5635 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
5636 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5637 | BTRFS_SHARED_BLOCK_REF_KEY); | |
5638 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5639 | } else { | |
5640 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5641 | BTRFS_TREE_BLOCK_REF_KEY); | |
5642 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
5643 | } | |
5644 | ||
5645 | btrfs_mark_buffer_dirty(leaf); | |
5646 | btrfs_free_path(path); | |
5647 | ||
f0486c68 | 5648 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
5d4f98a2 YZ |
5649 | if (ret) { |
5650 | printk(KERN_ERR "btrfs update block group failed for %llu " | |
5651 | "%llu\n", (unsigned long long)ins->objectid, | |
5652 | (unsigned long long)ins->offset); | |
5653 | BUG(); | |
5654 | } | |
5655 | return ret; | |
5656 | } | |
5657 | ||
5658 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5659 | struct btrfs_root *root, | |
5660 | u64 root_objectid, u64 owner, | |
5661 | u64 offset, struct btrfs_key *ins) | |
5662 | { | |
5663 | int ret; | |
5664 | ||
5665 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
5666 | ||
5667 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset, | |
5668 | 0, root_objectid, owner, offset, | |
5669 | BTRFS_ADD_DELAYED_EXTENT, NULL); | |
e6dcd2dc CM |
5670 | return ret; |
5671 | } | |
e02119d5 CM |
5672 | |
5673 | /* | |
5674 | * this is used by the tree logging recovery code. It records that | |
5675 | * an extent has been allocated and makes sure to clear the free | |
5676 | * space cache bits as well | |
5677 | */ | |
5d4f98a2 YZ |
5678 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5679 | struct btrfs_root *root, | |
5680 | u64 root_objectid, u64 owner, u64 offset, | |
5681 | struct btrfs_key *ins) | |
e02119d5 CM |
5682 | { |
5683 | int ret; | |
5684 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
5685 | struct btrfs_caching_control *caching_ctl; |
5686 | u64 start = ins->objectid; | |
5687 | u64 num_bytes = ins->offset; | |
e02119d5 | 5688 | |
e02119d5 | 5689 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 5690 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 5691 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 5692 | |
11833d66 YZ |
5693 | if (!caching_ctl) { |
5694 | BUG_ON(!block_group_cache_done(block_group)); | |
5695 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
5696 | BUG_ON(ret); | |
5697 | } else { | |
5698 | mutex_lock(&caching_ctl->mutex); | |
5699 | ||
5700 | if (start >= caching_ctl->progress) { | |
5701 | ret = add_excluded_extent(root, start, num_bytes); | |
5702 | BUG_ON(ret); | |
5703 | } else if (start + num_bytes <= caching_ctl->progress) { | |
5704 | ret = btrfs_remove_free_space(block_group, | |
5705 | start, num_bytes); | |
5706 | BUG_ON(ret); | |
5707 | } else { | |
5708 | num_bytes = caching_ctl->progress - start; | |
5709 | ret = btrfs_remove_free_space(block_group, | |
5710 | start, num_bytes); | |
5711 | BUG_ON(ret); | |
5712 | ||
5713 | start = caching_ctl->progress; | |
5714 | num_bytes = ins->objectid + ins->offset - | |
5715 | caching_ctl->progress; | |
5716 | ret = add_excluded_extent(root, start, num_bytes); | |
5717 | BUG_ON(ret); | |
5718 | } | |
5719 | ||
5720 | mutex_unlock(&caching_ctl->mutex); | |
5721 | put_caching_control(caching_ctl); | |
5722 | } | |
5723 | ||
b4d00d56 | 5724 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1); |
f0486c68 | 5725 | BUG_ON(ret); |
fa9c0d79 | 5726 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
5727 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
5728 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
5729 | return ret; |
5730 | } | |
5731 | ||
65b51a00 CM |
5732 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
5733 | struct btrfs_root *root, | |
4008c04a CM |
5734 | u64 bytenr, u32 blocksize, |
5735 | int level) | |
65b51a00 CM |
5736 | { |
5737 | struct extent_buffer *buf; | |
5738 | ||
5739 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
5740 | if (!buf) | |
5741 | return ERR_PTR(-ENOMEM); | |
5742 | btrfs_set_header_generation(buf, trans->transid); | |
4008c04a | 5743 | btrfs_set_buffer_lockdep_class(buf, level); |
65b51a00 CM |
5744 | btrfs_tree_lock(buf); |
5745 | clean_tree_block(trans, root, buf); | |
b4ce94de CM |
5746 | |
5747 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 5748 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 5749 | |
d0c803c4 | 5750 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
5751 | /* |
5752 | * we allow two log transactions at a time, use different | |
5753 | * EXENT bit to differentiate dirty pages. | |
5754 | */ | |
5755 | if (root->log_transid % 2 == 0) | |
5756 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
5757 | buf->start + buf->len - 1, GFP_NOFS); | |
5758 | else | |
5759 | set_extent_new(&root->dirty_log_pages, buf->start, | |
5760 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
5761 | } else { |
5762 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 5763 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 5764 | } |
65b51a00 | 5765 | trans->blocks_used++; |
b4ce94de | 5766 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
5767 | return buf; |
5768 | } | |
5769 | ||
f0486c68 YZ |
5770 | static struct btrfs_block_rsv * |
5771 | use_block_rsv(struct btrfs_trans_handle *trans, | |
5772 | struct btrfs_root *root, u32 blocksize) | |
5773 | { | |
5774 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 5775 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
5776 | int ret; |
5777 | ||
5778 | block_rsv = get_block_rsv(trans, root); | |
5779 | ||
5780 | if (block_rsv->size == 0) { | |
8bb8ab2e JB |
5781 | ret = reserve_metadata_bytes(trans, root, block_rsv, |
5782 | blocksize, 0); | |
68a82277 JB |
5783 | /* |
5784 | * If we couldn't reserve metadata bytes try and use some from | |
5785 | * the global reserve. | |
5786 | */ | |
5787 | if (ret && block_rsv != global_rsv) { | |
5788 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
5789 | if (!ret) | |
5790 | return global_rsv; | |
f0486c68 | 5791 | return ERR_PTR(ret); |
68a82277 | 5792 | } else if (ret) { |
f0486c68 | 5793 | return ERR_PTR(ret); |
68a82277 | 5794 | } |
f0486c68 YZ |
5795 | return block_rsv; |
5796 | } | |
5797 | ||
5798 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
5799 | if (!ret) | |
5800 | return block_rsv; | |
68a82277 JB |
5801 | if (ret) { |
5802 | WARN_ON(1); | |
5803 | ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize, | |
5804 | 0); | |
5805 | if (!ret) { | |
5806 | spin_lock(&block_rsv->lock); | |
5807 | block_rsv->size += blocksize; | |
5808 | spin_unlock(&block_rsv->lock); | |
5809 | return block_rsv; | |
5810 | } else if (ret && block_rsv != global_rsv) { | |
5811 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
5812 | if (!ret) | |
5813 | return global_rsv; | |
5814 | } | |
5815 | } | |
f0486c68 | 5816 | |
f0486c68 YZ |
5817 | return ERR_PTR(-ENOSPC); |
5818 | } | |
5819 | ||
5820 | static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
5821 | { | |
5822 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
5823 | block_rsv_release_bytes(block_rsv, NULL, 0); | |
5824 | } | |
5825 | ||
fec577fb | 5826 | /* |
f0486c68 YZ |
5827 | * finds a free extent and does all the dirty work required for allocation |
5828 | * returns the key for the extent through ins, and a tree buffer for | |
5829 | * the first block of the extent through buf. | |
5830 | * | |
fec577fb CM |
5831 | * returns the tree buffer or NULL. |
5832 | */ | |
5f39d397 | 5833 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
5834 | struct btrfs_root *root, u32 blocksize, |
5835 | u64 parent, u64 root_objectid, | |
5836 | struct btrfs_disk_key *key, int level, | |
5837 | u64 hint, u64 empty_size) | |
fec577fb | 5838 | { |
e2fa7227 | 5839 | struct btrfs_key ins; |
f0486c68 | 5840 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 5841 | struct extent_buffer *buf; |
f0486c68 YZ |
5842 | u64 flags = 0; |
5843 | int ret; | |
5844 | ||
fec577fb | 5845 | |
f0486c68 YZ |
5846 | block_rsv = use_block_rsv(trans, root, blocksize); |
5847 | if (IS_ERR(block_rsv)) | |
5848 | return ERR_CAST(block_rsv); | |
5849 | ||
5850 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
5851 | empty_size, hint, (u64)-1, &ins, 0); | |
fec577fb | 5852 | if (ret) { |
f0486c68 | 5853 | unuse_block_rsv(block_rsv, blocksize); |
54aa1f4d | 5854 | return ERR_PTR(ret); |
fec577fb | 5855 | } |
55c69072 | 5856 | |
4008c04a CM |
5857 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
5858 | blocksize, level); | |
f0486c68 YZ |
5859 | BUG_ON(IS_ERR(buf)); |
5860 | ||
5861 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
5862 | if (parent == 0) | |
5863 | parent = ins.objectid; | |
5864 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5865 | } else | |
5866 | BUG_ON(parent > 0); | |
5867 | ||
5868 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5869 | struct btrfs_delayed_extent_op *extent_op; | |
5870 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
5871 | BUG_ON(!extent_op); | |
5872 | if (key) | |
5873 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
5874 | else | |
5875 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
5876 | extent_op->flags_to_set = flags; | |
5877 | extent_op->update_key = 1; | |
5878 | extent_op->update_flags = 1; | |
5879 | extent_op->is_data = 0; | |
5880 | ||
5881 | ret = btrfs_add_delayed_tree_ref(trans, ins.objectid, | |
5882 | ins.offset, parent, root_objectid, | |
5883 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5884 | extent_op); | |
5885 | BUG_ON(ret); | |
5886 | } | |
fec577fb CM |
5887 | return buf; |
5888 | } | |
a28ec197 | 5889 | |
2c47e605 YZ |
5890 | struct walk_control { |
5891 | u64 refs[BTRFS_MAX_LEVEL]; | |
5892 | u64 flags[BTRFS_MAX_LEVEL]; | |
5893 | struct btrfs_key update_progress; | |
5894 | int stage; | |
5895 | int level; | |
5896 | int shared_level; | |
5897 | int update_ref; | |
5898 | int keep_locks; | |
1c4850e2 YZ |
5899 | int reada_slot; |
5900 | int reada_count; | |
2c47e605 YZ |
5901 | }; |
5902 | ||
5903 | #define DROP_REFERENCE 1 | |
5904 | #define UPDATE_BACKREF 2 | |
5905 | ||
1c4850e2 YZ |
5906 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
5907 | struct btrfs_root *root, | |
5908 | struct walk_control *wc, | |
5909 | struct btrfs_path *path) | |
6407bf6d | 5910 | { |
1c4850e2 YZ |
5911 | u64 bytenr; |
5912 | u64 generation; | |
5913 | u64 refs; | |
94fcca9f | 5914 | u64 flags; |
5d4f98a2 | 5915 | u32 nritems; |
1c4850e2 YZ |
5916 | u32 blocksize; |
5917 | struct btrfs_key key; | |
5918 | struct extent_buffer *eb; | |
6407bf6d | 5919 | int ret; |
1c4850e2 YZ |
5920 | int slot; |
5921 | int nread = 0; | |
6407bf6d | 5922 | |
1c4850e2 YZ |
5923 | if (path->slots[wc->level] < wc->reada_slot) { |
5924 | wc->reada_count = wc->reada_count * 2 / 3; | |
5925 | wc->reada_count = max(wc->reada_count, 2); | |
5926 | } else { | |
5927 | wc->reada_count = wc->reada_count * 3 / 2; | |
5928 | wc->reada_count = min_t(int, wc->reada_count, | |
5929 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
5930 | } | |
7bb86316 | 5931 | |
1c4850e2 YZ |
5932 | eb = path->nodes[wc->level]; |
5933 | nritems = btrfs_header_nritems(eb); | |
5934 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 5935 | |
1c4850e2 YZ |
5936 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
5937 | if (nread >= wc->reada_count) | |
5938 | break; | |
bd56b302 | 5939 | |
2dd3e67b | 5940 | cond_resched(); |
1c4850e2 YZ |
5941 | bytenr = btrfs_node_blockptr(eb, slot); |
5942 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 5943 | |
1c4850e2 YZ |
5944 | if (slot == path->slots[wc->level]) |
5945 | goto reada; | |
5d4f98a2 | 5946 | |
1c4850e2 YZ |
5947 | if (wc->stage == UPDATE_BACKREF && |
5948 | generation <= root->root_key.offset) | |
bd56b302 CM |
5949 | continue; |
5950 | ||
94fcca9f YZ |
5951 | /* We don't lock the tree block, it's OK to be racy here */ |
5952 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
5953 | &refs, &flags); | |
5954 | BUG_ON(ret); | |
5955 | BUG_ON(refs == 0); | |
5956 | ||
1c4850e2 | 5957 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
5958 | if (refs == 1) |
5959 | goto reada; | |
bd56b302 | 5960 | |
94fcca9f YZ |
5961 | if (wc->level == 1 && |
5962 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5963 | continue; | |
1c4850e2 YZ |
5964 | if (!wc->update_ref || |
5965 | generation <= root->root_key.offset) | |
5966 | continue; | |
5967 | btrfs_node_key_to_cpu(eb, &key, slot); | |
5968 | ret = btrfs_comp_cpu_keys(&key, | |
5969 | &wc->update_progress); | |
5970 | if (ret < 0) | |
5971 | continue; | |
94fcca9f YZ |
5972 | } else { |
5973 | if (wc->level == 1 && | |
5974 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5975 | continue; | |
6407bf6d | 5976 | } |
1c4850e2 YZ |
5977 | reada: |
5978 | ret = readahead_tree_block(root, bytenr, blocksize, | |
5979 | generation); | |
5980 | if (ret) | |
bd56b302 | 5981 | break; |
1c4850e2 | 5982 | nread++; |
20524f02 | 5983 | } |
1c4850e2 | 5984 | wc->reada_slot = slot; |
20524f02 | 5985 | } |
2c47e605 | 5986 | |
f82d02d9 | 5987 | /* |
2c47e605 YZ |
5988 | * hepler to process tree block while walking down the tree. |
5989 | * | |
2c47e605 YZ |
5990 | * when wc->stage == UPDATE_BACKREF, this function updates |
5991 | * back refs for pointers in the block. | |
5992 | * | |
5993 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 5994 | */ |
2c47e605 | 5995 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 5996 | struct btrfs_root *root, |
2c47e605 | 5997 | struct btrfs_path *path, |
94fcca9f | 5998 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 5999 | { |
2c47e605 YZ |
6000 | int level = wc->level; |
6001 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6002 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6003 | int ret; |
6004 | ||
2c47e605 YZ |
6005 | if (wc->stage == UPDATE_BACKREF && |
6006 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6007 | return 1; | |
f82d02d9 | 6008 | |
2c47e605 YZ |
6009 | /* |
6010 | * when reference count of tree block is 1, it won't increase | |
6011 | * again. once full backref flag is set, we never clear it. | |
6012 | */ | |
94fcca9f YZ |
6013 | if (lookup_info && |
6014 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6015 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6016 | BUG_ON(!path->locks[level]); |
6017 | ret = btrfs_lookup_extent_info(trans, root, | |
6018 | eb->start, eb->len, | |
6019 | &wc->refs[level], | |
6020 | &wc->flags[level]); | |
6021 | BUG_ON(ret); | |
6022 | BUG_ON(wc->refs[level] == 0); | |
6023 | } | |
5d4f98a2 | 6024 | |
2c47e605 YZ |
6025 | if (wc->stage == DROP_REFERENCE) { |
6026 | if (wc->refs[level] > 1) | |
6027 | return 1; | |
f82d02d9 | 6028 | |
2c47e605 YZ |
6029 | if (path->locks[level] && !wc->keep_locks) { |
6030 | btrfs_tree_unlock(eb); | |
6031 | path->locks[level] = 0; | |
6032 | } | |
6033 | return 0; | |
6034 | } | |
f82d02d9 | 6035 | |
2c47e605 YZ |
6036 | /* wc->stage == UPDATE_BACKREF */ |
6037 | if (!(wc->flags[level] & flag)) { | |
6038 | BUG_ON(!path->locks[level]); | |
6039 | ret = btrfs_inc_ref(trans, root, eb, 1); | |
f82d02d9 | 6040 | BUG_ON(ret); |
2c47e605 YZ |
6041 | ret = btrfs_dec_ref(trans, root, eb, 0); |
6042 | BUG_ON(ret); | |
6043 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, | |
6044 | eb->len, flag, 0); | |
6045 | BUG_ON(ret); | |
6046 | wc->flags[level] |= flag; | |
6047 | } | |
6048 | ||
6049 | /* | |
6050 | * the block is shared by multiple trees, so it's not good to | |
6051 | * keep the tree lock | |
6052 | */ | |
6053 | if (path->locks[level] && level > 0) { | |
6054 | btrfs_tree_unlock(eb); | |
6055 | path->locks[level] = 0; | |
6056 | } | |
6057 | return 0; | |
6058 | } | |
6059 | ||
1c4850e2 YZ |
6060 | /* |
6061 | * hepler to process tree block pointer. | |
6062 | * | |
6063 | * when wc->stage == DROP_REFERENCE, this function checks | |
6064 | * reference count of the block pointed to. if the block | |
6065 | * is shared and we need update back refs for the subtree | |
6066 | * rooted at the block, this function changes wc->stage to | |
6067 | * UPDATE_BACKREF. if the block is shared and there is no | |
6068 | * need to update back, this function drops the reference | |
6069 | * to the block. | |
6070 | * | |
6071 | * NOTE: return value 1 means we should stop walking down. | |
6072 | */ | |
6073 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
6074 | struct btrfs_root *root, | |
6075 | struct btrfs_path *path, | |
94fcca9f | 6076 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
6077 | { |
6078 | u64 bytenr; | |
6079 | u64 generation; | |
6080 | u64 parent; | |
6081 | u32 blocksize; | |
6082 | struct btrfs_key key; | |
6083 | struct extent_buffer *next; | |
6084 | int level = wc->level; | |
6085 | int reada = 0; | |
6086 | int ret = 0; | |
6087 | ||
6088 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
6089 | path->slots[level]); | |
6090 | /* | |
6091 | * if the lower level block was created before the snapshot | |
6092 | * was created, we know there is no need to update back refs | |
6093 | * for the subtree | |
6094 | */ | |
6095 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
6096 | generation <= root->root_key.offset) { |
6097 | *lookup_info = 1; | |
1c4850e2 | 6098 | return 1; |
94fcca9f | 6099 | } |
1c4850e2 YZ |
6100 | |
6101 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
6102 | blocksize = btrfs_level_size(root, level - 1); | |
6103 | ||
6104 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6105 | if (!next) { | |
6106 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6107 | if (!next) |
6108 | return -ENOMEM; | |
1c4850e2 YZ |
6109 | reada = 1; |
6110 | } | |
6111 | btrfs_tree_lock(next); | |
6112 | btrfs_set_lock_blocking(next); | |
6113 | ||
94fcca9f YZ |
6114 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6115 | &wc->refs[level - 1], | |
6116 | &wc->flags[level - 1]); | |
6117 | BUG_ON(ret); | |
6118 | BUG_ON(wc->refs[level - 1] == 0); | |
6119 | *lookup_info = 0; | |
1c4850e2 | 6120 | |
94fcca9f | 6121 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6122 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6123 | if (level == 1 && |
6124 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6125 | goto skip; | |
6126 | ||
1c4850e2 YZ |
6127 | if (!wc->update_ref || |
6128 | generation <= root->root_key.offset) | |
6129 | goto skip; | |
6130 | ||
6131 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6132 | path->slots[level]); | |
6133 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6134 | if (ret < 0) | |
6135 | goto skip; | |
6136 | ||
6137 | wc->stage = UPDATE_BACKREF; | |
6138 | wc->shared_level = level - 1; | |
6139 | } | |
94fcca9f YZ |
6140 | } else { |
6141 | if (level == 1 && | |
6142 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6143 | goto skip; | |
1c4850e2 YZ |
6144 | } |
6145 | ||
6146 | if (!btrfs_buffer_uptodate(next, generation)) { | |
6147 | btrfs_tree_unlock(next); | |
6148 | free_extent_buffer(next); | |
6149 | next = NULL; | |
94fcca9f | 6150 | *lookup_info = 1; |
1c4850e2 YZ |
6151 | } |
6152 | ||
6153 | if (!next) { | |
6154 | if (reada && level == 1) | |
6155 | reada_walk_down(trans, root, wc, path); | |
6156 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6157 | if (!next) |
6158 | return -EIO; | |
1c4850e2 YZ |
6159 | btrfs_tree_lock(next); |
6160 | btrfs_set_lock_blocking(next); | |
6161 | } | |
6162 | ||
6163 | level--; | |
6164 | BUG_ON(level != btrfs_header_level(next)); | |
6165 | path->nodes[level] = next; | |
6166 | path->slots[level] = 0; | |
6167 | path->locks[level] = 1; | |
6168 | wc->level = level; | |
6169 | if (wc->level == 1) | |
6170 | wc->reada_slot = 0; | |
6171 | return 0; | |
6172 | skip: | |
6173 | wc->refs[level - 1] = 0; | |
6174 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6175 | if (wc->stage == DROP_REFERENCE) { |
6176 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6177 | parent = path->nodes[level]->start; | |
6178 | } else { | |
6179 | BUG_ON(root->root_key.objectid != | |
6180 | btrfs_header_owner(path->nodes[level])); | |
6181 | parent = 0; | |
6182 | } | |
1c4850e2 | 6183 | |
94fcca9f YZ |
6184 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
6185 | root->root_key.objectid, level - 1, 0); | |
6186 | BUG_ON(ret); | |
1c4850e2 | 6187 | } |
1c4850e2 YZ |
6188 | btrfs_tree_unlock(next); |
6189 | free_extent_buffer(next); | |
94fcca9f | 6190 | *lookup_info = 1; |
1c4850e2 YZ |
6191 | return 1; |
6192 | } | |
6193 | ||
2c47e605 YZ |
6194 | /* |
6195 | * hepler to process tree block while walking up the tree. | |
6196 | * | |
6197 | * when wc->stage == DROP_REFERENCE, this function drops | |
6198 | * reference count on the block. | |
6199 | * | |
6200 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6201 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6202 | * to UPDATE_BACKREF previously while processing the block. | |
6203 | * | |
6204 | * NOTE: return value 1 means we should stop walking up. | |
6205 | */ | |
6206 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6207 | struct btrfs_root *root, | |
6208 | struct btrfs_path *path, | |
6209 | struct walk_control *wc) | |
6210 | { | |
f0486c68 | 6211 | int ret; |
2c47e605 YZ |
6212 | int level = wc->level; |
6213 | struct extent_buffer *eb = path->nodes[level]; | |
6214 | u64 parent = 0; | |
6215 | ||
6216 | if (wc->stage == UPDATE_BACKREF) { | |
6217 | BUG_ON(wc->shared_level < level); | |
6218 | if (level < wc->shared_level) | |
6219 | goto out; | |
6220 | ||
2c47e605 YZ |
6221 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6222 | if (ret > 0) | |
6223 | wc->update_ref = 0; | |
6224 | ||
6225 | wc->stage = DROP_REFERENCE; | |
6226 | wc->shared_level = -1; | |
6227 | path->slots[level] = 0; | |
6228 | ||
6229 | /* | |
6230 | * check reference count again if the block isn't locked. | |
6231 | * we should start walking down the tree again if reference | |
6232 | * count is one. | |
6233 | */ | |
6234 | if (!path->locks[level]) { | |
6235 | BUG_ON(level == 0); | |
6236 | btrfs_tree_lock(eb); | |
6237 | btrfs_set_lock_blocking(eb); | |
6238 | path->locks[level] = 1; | |
6239 | ||
6240 | ret = btrfs_lookup_extent_info(trans, root, | |
6241 | eb->start, eb->len, | |
6242 | &wc->refs[level], | |
6243 | &wc->flags[level]); | |
f82d02d9 | 6244 | BUG_ON(ret); |
2c47e605 YZ |
6245 | BUG_ON(wc->refs[level] == 0); |
6246 | if (wc->refs[level] == 1) { | |
6247 | btrfs_tree_unlock(eb); | |
6248 | path->locks[level] = 0; | |
6249 | return 1; | |
6250 | } | |
f82d02d9 | 6251 | } |
2c47e605 | 6252 | } |
f82d02d9 | 6253 | |
2c47e605 YZ |
6254 | /* wc->stage == DROP_REFERENCE */ |
6255 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6256 | |
2c47e605 YZ |
6257 | if (wc->refs[level] == 1) { |
6258 | if (level == 0) { | |
6259 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6260 | ret = btrfs_dec_ref(trans, root, eb, 1); | |
6261 | else | |
6262 | ret = btrfs_dec_ref(trans, root, eb, 0); | |
6263 | BUG_ON(ret); | |
6264 | } | |
6265 | /* make block locked assertion in clean_tree_block happy */ | |
6266 | if (!path->locks[level] && | |
6267 | btrfs_header_generation(eb) == trans->transid) { | |
6268 | btrfs_tree_lock(eb); | |
6269 | btrfs_set_lock_blocking(eb); | |
6270 | path->locks[level] = 1; | |
6271 | } | |
6272 | clean_tree_block(trans, root, eb); | |
6273 | } | |
6274 | ||
6275 | if (eb == root->node) { | |
6276 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6277 | parent = eb->start; | |
6278 | else | |
6279 | BUG_ON(root->root_key.objectid != | |
6280 | btrfs_header_owner(eb)); | |
6281 | } else { | |
6282 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6283 | parent = path->nodes[level + 1]->start; | |
6284 | else | |
6285 | BUG_ON(root->root_key.objectid != | |
6286 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6287 | } |
f82d02d9 | 6288 | |
f0486c68 | 6289 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
6290 | out: |
6291 | wc->refs[level] = 0; | |
6292 | wc->flags[level] = 0; | |
f0486c68 | 6293 | return 0; |
2c47e605 YZ |
6294 | } |
6295 | ||
6296 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6297 | struct btrfs_root *root, | |
6298 | struct btrfs_path *path, | |
6299 | struct walk_control *wc) | |
6300 | { | |
2c47e605 | 6301 | int level = wc->level; |
94fcca9f | 6302 | int lookup_info = 1; |
2c47e605 YZ |
6303 | int ret; |
6304 | ||
6305 | while (level >= 0) { | |
94fcca9f | 6306 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6307 | if (ret > 0) |
6308 | break; | |
6309 | ||
6310 | if (level == 0) | |
6311 | break; | |
6312 | ||
7a7965f8 YZ |
6313 | if (path->slots[level] >= |
6314 | btrfs_header_nritems(path->nodes[level])) | |
6315 | break; | |
6316 | ||
94fcca9f | 6317 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6318 | if (ret > 0) { |
6319 | path->slots[level]++; | |
6320 | continue; | |
90d2c51d MX |
6321 | } else if (ret < 0) |
6322 | return ret; | |
1c4850e2 | 6323 | level = wc->level; |
f82d02d9 | 6324 | } |
f82d02d9 YZ |
6325 | return 0; |
6326 | } | |
6327 | ||
d397712b | 6328 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6329 | struct btrfs_root *root, |
f82d02d9 | 6330 | struct btrfs_path *path, |
2c47e605 | 6331 | struct walk_control *wc, int max_level) |
20524f02 | 6332 | { |
2c47e605 | 6333 | int level = wc->level; |
20524f02 | 6334 | int ret; |
9f3a7427 | 6335 | |
2c47e605 YZ |
6336 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6337 | while (level < max_level && path->nodes[level]) { | |
6338 | wc->level = level; | |
6339 | if (path->slots[level] + 1 < | |
6340 | btrfs_header_nritems(path->nodes[level])) { | |
6341 | path->slots[level]++; | |
20524f02 CM |
6342 | return 0; |
6343 | } else { | |
2c47e605 YZ |
6344 | ret = walk_up_proc(trans, root, path, wc); |
6345 | if (ret > 0) | |
6346 | return 0; | |
bd56b302 | 6347 | |
2c47e605 YZ |
6348 | if (path->locks[level]) { |
6349 | btrfs_tree_unlock(path->nodes[level]); | |
6350 | path->locks[level] = 0; | |
f82d02d9 | 6351 | } |
2c47e605 YZ |
6352 | free_extent_buffer(path->nodes[level]); |
6353 | path->nodes[level] = NULL; | |
6354 | level++; | |
20524f02 CM |
6355 | } |
6356 | } | |
6357 | return 1; | |
6358 | } | |
6359 | ||
9aca1d51 | 6360 | /* |
2c47e605 YZ |
6361 | * drop a subvolume tree. |
6362 | * | |
6363 | * this function traverses the tree freeing any blocks that only | |
6364 | * referenced by the tree. | |
6365 | * | |
6366 | * when a shared tree block is found. this function decreases its | |
6367 | * reference count by one. if update_ref is true, this function | |
6368 | * also make sure backrefs for the shared block and all lower level | |
6369 | * blocks are properly updated. | |
9aca1d51 | 6370 | */ |
3fd0a558 YZ |
6371 | int btrfs_drop_snapshot(struct btrfs_root *root, |
6372 | struct btrfs_block_rsv *block_rsv, int update_ref) | |
20524f02 | 6373 | { |
5caf2a00 | 6374 | struct btrfs_path *path; |
2c47e605 YZ |
6375 | struct btrfs_trans_handle *trans; |
6376 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6377 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6378 | struct walk_control *wc; |
6379 | struct btrfs_key key; | |
6380 | int err = 0; | |
6381 | int ret; | |
6382 | int level; | |
20524f02 | 6383 | |
5caf2a00 CM |
6384 | path = btrfs_alloc_path(); |
6385 | BUG_ON(!path); | |
20524f02 | 6386 | |
2c47e605 YZ |
6387 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
6388 | BUG_ON(!wc); | |
6389 | ||
a22285a6 | 6390 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 TI |
6391 | BUG_ON(IS_ERR(trans)); |
6392 | ||
3fd0a558 YZ |
6393 | if (block_rsv) |
6394 | trans->block_rsv = block_rsv; | |
2c47e605 | 6395 | |
9f3a7427 | 6396 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6397 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6398 | path->nodes[level] = btrfs_lock_root_node(root); |
6399 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6400 | path->slots[level] = 0; |
5d4f98a2 | 6401 | path->locks[level] = 1; |
2c47e605 YZ |
6402 | memset(&wc->update_progress, 0, |
6403 | sizeof(wc->update_progress)); | |
9f3a7427 | 6404 | } else { |
9f3a7427 | 6405 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6406 | memcpy(&wc->update_progress, &key, |
6407 | sizeof(wc->update_progress)); | |
6408 | ||
6702ed49 | 6409 | level = root_item->drop_level; |
2c47e605 | 6410 | BUG_ON(level == 0); |
6702ed49 | 6411 | path->lowest_level = level; |
2c47e605 YZ |
6412 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6413 | path->lowest_level = 0; | |
6414 | if (ret < 0) { | |
6415 | err = ret; | |
9f3a7427 CM |
6416 | goto out; |
6417 | } | |
1c4850e2 | 6418 | WARN_ON(ret > 0); |
2c47e605 | 6419 | |
7d9eb12c CM |
6420 | /* |
6421 | * unlock our path, this is safe because only this | |
6422 | * function is allowed to delete this snapshot | |
6423 | */ | |
5d4f98a2 | 6424 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
6425 | |
6426 | level = btrfs_header_level(root->node); | |
6427 | while (1) { | |
6428 | btrfs_tree_lock(path->nodes[level]); | |
6429 | btrfs_set_lock_blocking(path->nodes[level]); | |
6430 | ||
6431 | ret = btrfs_lookup_extent_info(trans, root, | |
6432 | path->nodes[level]->start, | |
6433 | path->nodes[level]->len, | |
6434 | &wc->refs[level], | |
6435 | &wc->flags[level]); | |
6436 | BUG_ON(ret); | |
6437 | BUG_ON(wc->refs[level] == 0); | |
6438 | ||
6439 | if (level == root_item->drop_level) | |
6440 | break; | |
6441 | ||
6442 | btrfs_tree_unlock(path->nodes[level]); | |
6443 | WARN_ON(wc->refs[level] != 1); | |
6444 | level--; | |
6445 | } | |
9f3a7427 | 6446 | } |
2c47e605 YZ |
6447 | |
6448 | wc->level = level; | |
6449 | wc->shared_level = -1; | |
6450 | wc->stage = DROP_REFERENCE; | |
6451 | wc->update_ref = update_ref; | |
6452 | wc->keep_locks = 0; | |
1c4850e2 | 6453 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 6454 | |
d397712b | 6455 | while (1) { |
2c47e605 YZ |
6456 | ret = walk_down_tree(trans, root, path, wc); |
6457 | if (ret < 0) { | |
6458 | err = ret; | |
20524f02 | 6459 | break; |
2c47e605 | 6460 | } |
9aca1d51 | 6461 | |
2c47e605 YZ |
6462 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
6463 | if (ret < 0) { | |
6464 | err = ret; | |
20524f02 | 6465 | break; |
2c47e605 YZ |
6466 | } |
6467 | ||
6468 | if (ret > 0) { | |
6469 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
6470 | break; |
6471 | } | |
2c47e605 YZ |
6472 | |
6473 | if (wc->stage == DROP_REFERENCE) { | |
6474 | level = wc->level; | |
6475 | btrfs_node_key(path->nodes[level], | |
6476 | &root_item->drop_progress, | |
6477 | path->slots[level]); | |
6478 | root_item->drop_level = level; | |
6479 | } | |
6480 | ||
6481 | BUG_ON(wc->level == 0); | |
3fd0a558 | 6482 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
6483 | ret = btrfs_update_root(trans, tree_root, |
6484 | &root->root_key, | |
6485 | root_item); | |
6486 | BUG_ON(ret); | |
6487 | ||
3fd0a558 | 6488 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 6489 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 | 6490 | BUG_ON(IS_ERR(trans)); |
3fd0a558 YZ |
6491 | if (block_rsv) |
6492 | trans->block_rsv = block_rsv; | |
c3e69d58 | 6493 | } |
20524f02 | 6494 | } |
b3b4aa74 | 6495 | btrfs_release_path(path); |
2c47e605 YZ |
6496 | BUG_ON(err); |
6497 | ||
6498 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
6499 | BUG_ON(ret); | |
6500 | ||
76dda93c YZ |
6501 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
6502 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
6503 | NULL, NULL); | |
6504 | BUG_ON(ret < 0); | |
6505 | if (ret > 0) { | |
84cd948c JB |
6506 | /* if we fail to delete the orphan item this time |
6507 | * around, it'll get picked up the next time. | |
6508 | * | |
6509 | * The most common failure here is just -ENOENT. | |
6510 | */ | |
6511 | btrfs_del_orphan_item(trans, tree_root, | |
6512 | root->root_key.objectid); | |
76dda93c YZ |
6513 | } |
6514 | } | |
6515 | ||
6516 | if (root->in_radix) { | |
6517 | btrfs_free_fs_root(tree_root->fs_info, root); | |
6518 | } else { | |
6519 | free_extent_buffer(root->node); | |
6520 | free_extent_buffer(root->commit_root); | |
6521 | kfree(root); | |
6522 | } | |
9f3a7427 | 6523 | out: |
3fd0a558 | 6524 | btrfs_end_transaction_throttle(trans, tree_root); |
2c47e605 | 6525 | kfree(wc); |
5caf2a00 | 6526 | btrfs_free_path(path); |
2c47e605 | 6527 | return err; |
20524f02 | 6528 | } |
9078a3e1 | 6529 | |
2c47e605 YZ |
6530 | /* |
6531 | * drop subtree rooted at tree block 'node'. | |
6532 | * | |
6533 | * NOTE: this function will unlock and release tree block 'node' | |
6534 | */ | |
f82d02d9 YZ |
6535 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
6536 | struct btrfs_root *root, | |
6537 | struct extent_buffer *node, | |
6538 | struct extent_buffer *parent) | |
6539 | { | |
6540 | struct btrfs_path *path; | |
2c47e605 | 6541 | struct walk_control *wc; |
f82d02d9 YZ |
6542 | int level; |
6543 | int parent_level; | |
6544 | int ret = 0; | |
6545 | int wret; | |
6546 | ||
2c47e605 YZ |
6547 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
6548 | ||
f82d02d9 | 6549 | path = btrfs_alloc_path(); |
db5b493a TI |
6550 | if (!path) |
6551 | return -ENOMEM; | |
f82d02d9 | 6552 | |
2c47e605 | 6553 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
6554 | if (!wc) { |
6555 | btrfs_free_path(path); | |
6556 | return -ENOMEM; | |
6557 | } | |
2c47e605 | 6558 | |
b9447ef8 | 6559 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
6560 | parent_level = btrfs_header_level(parent); |
6561 | extent_buffer_get(parent); | |
6562 | path->nodes[parent_level] = parent; | |
6563 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
6564 | ||
b9447ef8 | 6565 | btrfs_assert_tree_locked(node); |
f82d02d9 | 6566 | level = btrfs_header_level(node); |
f82d02d9 YZ |
6567 | path->nodes[level] = node; |
6568 | path->slots[level] = 0; | |
2c47e605 YZ |
6569 | path->locks[level] = 1; |
6570 | ||
6571 | wc->refs[parent_level] = 1; | |
6572 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6573 | wc->level = level; | |
6574 | wc->shared_level = -1; | |
6575 | wc->stage = DROP_REFERENCE; | |
6576 | wc->update_ref = 0; | |
6577 | wc->keep_locks = 1; | |
1c4850e2 | 6578 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
6579 | |
6580 | while (1) { | |
2c47e605 YZ |
6581 | wret = walk_down_tree(trans, root, path, wc); |
6582 | if (wret < 0) { | |
f82d02d9 | 6583 | ret = wret; |
f82d02d9 | 6584 | break; |
2c47e605 | 6585 | } |
f82d02d9 | 6586 | |
2c47e605 | 6587 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
6588 | if (wret < 0) |
6589 | ret = wret; | |
6590 | if (wret != 0) | |
6591 | break; | |
6592 | } | |
6593 | ||
2c47e605 | 6594 | kfree(wc); |
f82d02d9 YZ |
6595 | btrfs_free_path(path); |
6596 | return ret; | |
6597 | } | |
6598 | ||
5d4f98a2 | 6599 | #if 0 |
8e7bf94f CM |
6600 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
6601 | unsigned long nr) | |
6602 | { | |
6603 | return min(last, start + nr - 1); | |
6604 | } | |
6605 | ||
d397712b | 6606 | static noinline int relocate_inode_pages(struct inode *inode, u64 start, |
98ed5174 | 6607 | u64 len) |
edbd8d4e CM |
6608 | { |
6609 | u64 page_start; | |
6610 | u64 page_end; | |
1a40e23b | 6611 | unsigned long first_index; |
edbd8d4e | 6612 | unsigned long last_index; |
edbd8d4e CM |
6613 | unsigned long i; |
6614 | struct page *page; | |
d1310b2e | 6615 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 6616 | struct file_ra_state *ra; |
3eaa2885 | 6617 | struct btrfs_ordered_extent *ordered; |
1a40e23b ZY |
6618 | unsigned int total_read = 0; |
6619 | unsigned int total_dirty = 0; | |
6620 | int ret = 0; | |
4313b399 CM |
6621 | |
6622 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
5df67083 TI |
6623 | if (!ra) |
6624 | return -ENOMEM; | |
edbd8d4e CM |
6625 | |
6626 | mutex_lock(&inode->i_mutex); | |
1a40e23b | 6627 | first_index = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
6628 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
6629 | ||
1a40e23b ZY |
6630 | /* make sure the dirty trick played by the caller work */ |
6631 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
6632 | first_index, last_index); | |
6633 | if (ret) | |
6634 | goto out_unlock; | |
8e7bf94f | 6635 | |
4313b399 | 6636 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 6637 | |
1a40e23b ZY |
6638 | for (i = first_index ; i <= last_index; i++) { |
6639 | if (total_read % ra->ra_pages == 0) { | |
8e7bf94f | 6640 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, |
1a40e23b | 6641 | calc_ra(i, last_index, ra->ra_pages)); |
8e7bf94f CM |
6642 | } |
6643 | total_read++; | |
3eaa2885 CM |
6644 | again: |
6645 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
1a40e23b | 6646 | BUG_ON(1); |
edbd8d4e | 6647 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 6648 | if (!page) { |
1a40e23b | 6649 | ret = -ENOMEM; |
edbd8d4e | 6650 | goto out_unlock; |
a061fc8d | 6651 | } |
edbd8d4e CM |
6652 | if (!PageUptodate(page)) { |
6653 | btrfs_readpage(NULL, page); | |
6654 | lock_page(page); | |
6655 | if (!PageUptodate(page)) { | |
6656 | unlock_page(page); | |
6657 | page_cache_release(page); | |
1a40e23b | 6658 | ret = -EIO; |
edbd8d4e CM |
6659 | goto out_unlock; |
6660 | } | |
6661 | } | |
ec44a35c | 6662 | wait_on_page_writeback(page); |
3eaa2885 | 6663 | |
edbd8d4e CM |
6664 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
6665 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
d1310b2e | 6666 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 6667 | |
3eaa2885 CM |
6668 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6669 | if (ordered) { | |
6670 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
6671 | unlock_page(page); | |
6672 | page_cache_release(page); | |
6673 | btrfs_start_ordered_extent(inode, ordered, 1); | |
6674 | btrfs_put_ordered_extent(ordered); | |
6675 | goto again; | |
6676 | } | |
6677 | set_page_extent_mapped(page); | |
6678 | ||
1a40e23b ZY |
6679 | if (i == first_index) |
6680 | set_extent_bits(io_tree, page_start, page_end, | |
6681 | EXTENT_BOUNDARY, GFP_NOFS); | |
1f80e4db | 6682 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
1a40e23b | 6683 | |
a061fc8d | 6684 | set_page_dirty(page); |
1a40e23b | 6685 | total_dirty++; |
edbd8d4e | 6686 | |
d1310b2e | 6687 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
6688 | unlock_page(page); |
6689 | page_cache_release(page); | |
6690 | } | |
6691 | ||
6692 | out_unlock: | |
ec44a35c | 6693 | kfree(ra); |
edbd8d4e | 6694 | mutex_unlock(&inode->i_mutex); |
1a40e23b ZY |
6695 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty); |
6696 | return ret; | |
edbd8d4e CM |
6697 | } |
6698 | ||
d397712b | 6699 | static noinline int relocate_data_extent(struct inode *reloc_inode, |
1a40e23b ZY |
6700 | struct btrfs_key *extent_key, |
6701 | u64 offset) | |
6702 | { | |
6703 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
6704 | struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree; | |
6705 | struct extent_map *em; | |
6643558d YZ |
6706 | u64 start = extent_key->objectid - offset; |
6707 | u64 end = start + extent_key->offset - 1; | |
bf4ef679 | 6708 | |
172ddd60 | 6709 | em = alloc_extent_map(); |
c26a9203 | 6710 | BUG_ON(!em); |
bf4ef679 | 6711 | |
6643558d | 6712 | em->start = start; |
1a40e23b | 6713 | em->len = extent_key->offset; |
c8b97818 | 6714 | em->block_len = extent_key->offset; |
1a40e23b ZY |
6715 | em->block_start = extent_key->objectid; |
6716 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
6717 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
6718 | ||
6719 | /* setup extent map to cheat btrfs_readpage */ | |
6643558d | 6720 | lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
1a40e23b ZY |
6721 | while (1) { |
6722 | int ret; | |
890871be | 6723 | write_lock(&em_tree->lock); |
1a40e23b | 6724 | ret = add_extent_mapping(em_tree, em); |
890871be | 6725 | write_unlock(&em_tree->lock); |
1a40e23b ZY |
6726 | if (ret != -EEXIST) { |
6727 | free_extent_map(em); | |
bf4ef679 CM |
6728 | break; |
6729 | } | |
6643558d | 6730 | btrfs_drop_extent_cache(reloc_inode, start, end, 0); |
bf4ef679 | 6731 | } |
6643558d | 6732 | unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
bf4ef679 | 6733 | |
6643558d | 6734 | return relocate_inode_pages(reloc_inode, start, extent_key->offset); |
1a40e23b | 6735 | } |
edbd8d4e | 6736 | |
1a40e23b ZY |
6737 | struct btrfs_ref_path { |
6738 | u64 extent_start; | |
6739 | u64 nodes[BTRFS_MAX_LEVEL]; | |
6740 | u64 root_objectid; | |
6741 | u64 root_generation; | |
6742 | u64 owner_objectid; | |
1a40e23b ZY |
6743 | u32 num_refs; |
6744 | int lowest_level; | |
6745 | int current_level; | |
f82d02d9 YZ |
6746 | int shared_level; |
6747 | ||
6748 | struct btrfs_key node_keys[BTRFS_MAX_LEVEL]; | |
6749 | u64 new_nodes[BTRFS_MAX_LEVEL]; | |
1a40e23b | 6750 | }; |
7d9eb12c | 6751 | |
1a40e23b | 6752 | struct disk_extent { |
c8b97818 | 6753 | u64 ram_bytes; |
1a40e23b ZY |
6754 | u64 disk_bytenr; |
6755 | u64 disk_num_bytes; | |
6756 | u64 offset; | |
6757 | u64 num_bytes; | |
c8b97818 CM |
6758 | u8 compression; |
6759 | u8 encryption; | |
6760 | u16 other_encoding; | |
1a40e23b | 6761 | }; |
4313b399 | 6762 | |
1a40e23b ZY |
6763 | static int is_cowonly_root(u64 root_objectid) |
6764 | { | |
6765 | if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || | |
6766 | root_objectid == BTRFS_EXTENT_TREE_OBJECTID || | |
6767 | root_objectid == BTRFS_CHUNK_TREE_OBJECTID || | |
6768 | root_objectid == BTRFS_DEV_TREE_OBJECTID || | |
0403e47e YZ |
6769 | root_objectid == BTRFS_TREE_LOG_OBJECTID || |
6770 | root_objectid == BTRFS_CSUM_TREE_OBJECTID) | |
1a40e23b ZY |
6771 | return 1; |
6772 | return 0; | |
6773 | } | |
edbd8d4e | 6774 | |
d397712b | 6775 | static noinline int __next_ref_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6776 | struct btrfs_root *extent_root, |
6777 | struct btrfs_ref_path *ref_path, | |
6778 | int first_time) | |
6779 | { | |
6780 | struct extent_buffer *leaf; | |
6781 | struct btrfs_path *path; | |
6782 | struct btrfs_extent_ref *ref; | |
6783 | struct btrfs_key key; | |
6784 | struct btrfs_key found_key; | |
6785 | u64 bytenr; | |
6786 | u32 nritems; | |
6787 | int level; | |
6788 | int ret = 1; | |
edbd8d4e | 6789 | |
1a40e23b ZY |
6790 | path = btrfs_alloc_path(); |
6791 | if (!path) | |
6792 | return -ENOMEM; | |
bf4ef679 | 6793 | |
1a40e23b ZY |
6794 | if (first_time) { |
6795 | ref_path->lowest_level = -1; | |
6796 | ref_path->current_level = -1; | |
f82d02d9 | 6797 | ref_path->shared_level = -1; |
1a40e23b ZY |
6798 | goto walk_up; |
6799 | } | |
6800 | walk_down: | |
6801 | level = ref_path->current_level - 1; | |
6802 | while (level >= -1) { | |
6803 | u64 parent; | |
6804 | if (level < ref_path->lowest_level) | |
6805 | break; | |
bf4ef679 | 6806 | |
d397712b | 6807 | if (level >= 0) |
1a40e23b | 6808 | bytenr = ref_path->nodes[level]; |
d397712b | 6809 | else |
1a40e23b | 6810 | bytenr = ref_path->extent_start; |
1a40e23b | 6811 | BUG_ON(bytenr == 0); |
bf4ef679 | 6812 | |
1a40e23b ZY |
6813 | parent = ref_path->nodes[level + 1]; |
6814 | ref_path->nodes[level + 1] = 0; | |
6815 | ref_path->current_level = level; | |
6816 | BUG_ON(parent == 0); | |
0ef3e66b | 6817 | |
1a40e23b ZY |
6818 | key.objectid = bytenr; |
6819 | key.offset = parent + 1; | |
6820 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 6821 | |
1a40e23b ZY |
6822 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
6823 | if (ret < 0) | |
edbd8d4e | 6824 | goto out; |
1a40e23b | 6825 | BUG_ON(ret == 0); |
7d9eb12c | 6826 | |
1a40e23b ZY |
6827 | leaf = path->nodes[0]; |
6828 | nritems = btrfs_header_nritems(leaf); | |
6829 | if (path->slots[0] >= nritems) { | |
6830 | ret = btrfs_next_leaf(extent_root, path); | |
6831 | if (ret < 0) | |
6832 | goto out; | |
6833 | if (ret > 0) | |
6834 | goto next; | |
6835 | leaf = path->nodes[0]; | |
6836 | } | |
0ef3e66b | 6837 | |
1a40e23b ZY |
6838 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6839 | if (found_key.objectid == bytenr && | |
f82d02d9 YZ |
6840 | found_key.type == BTRFS_EXTENT_REF_KEY) { |
6841 | if (level < ref_path->shared_level) | |
6842 | ref_path->shared_level = level; | |
1a40e23b | 6843 | goto found; |
f82d02d9 | 6844 | } |
1a40e23b ZY |
6845 | next: |
6846 | level--; | |
6847 | btrfs_release_path(extent_root, path); | |
d899e052 | 6848 | cond_resched(); |
1a40e23b ZY |
6849 | } |
6850 | /* reached lowest level */ | |
6851 | ret = 1; | |
6852 | goto out; | |
6853 | walk_up: | |
6854 | level = ref_path->current_level; | |
6855 | while (level < BTRFS_MAX_LEVEL - 1) { | |
6856 | u64 ref_objectid; | |
d397712b CM |
6857 | |
6858 | if (level >= 0) | |
1a40e23b | 6859 | bytenr = ref_path->nodes[level]; |
d397712b | 6860 | else |
1a40e23b | 6861 | bytenr = ref_path->extent_start; |
d397712b | 6862 | |
1a40e23b | 6863 | BUG_ON(bytenr == 0); |
edbd8d4e | 6864 | |
1a40e23b ZY |
6865 | key.objectid = bytenr; |
6866 | key.offset = 0; | |
6867 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 6868 | |
1a40e23b ZY |
6869 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
6870 | if (ret < 0) | |
6871 | goto out; | |
edbd8d4e | 6872 | |
1a40e23b ZY |
6873 | leaf = path->nodes[0]; |
6874 | nritems = btrfs_header_nritems(leaf); | |
6875 | if (path->slots[0] >= nritems) { | |
6876 | ret = btrfs_next_leaf(extent_root, path); | |
6877 | if (ret < 0) | |
6878 | goto out; | |
6879 | if (ret > 0) { | |
6880 | /* the extent was freed by someone */ | |
6881 | if (ref_path->lowest_level == level) | |
6882 | goto out; | |
6883 | btrfs_release_path(extent_root, path); | |
6884 | goto walk_down; | |
6885 | } | |
6886 | leaf = path->nodes[0]; | |
6887 | } | |
edbd8d4e | 6888 | |
1a40e23b ZY |
6889 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6890 | if (found_key.objectid != bytenr || | |
6891 | found_key.type != BTRFS_EXTENT_REF_KEY) { | |
6892 | /* the extent was freed by someone */ | |
6893 | if (ref_path->lowest_level == level) { | |
6894 | ret = 1; | |
6895 | goto out; | |
6896 | } | |
6897 | btrfs_release_path(extent_root, path); | |
6898 | goto walk_down; | |
6899 | } | |
6900 | found: | |
6901 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
6902 | struct btrfs_extent_ref); | |
6903 | ref_objectid = btrfs_ref_objectid(leaf, ref); | |
6904 | if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
6905 | if (first_time) { | |
6906 | level = (int)ref_objectid; | |
6907 | BUG_ON(level >= BTRFS_MAX_LEVEL); | |
6908 | ref_path->lowest_level = level; | |
6909 | ref_path->current_level = level; | |
6910 | ref_path->nodes[level] = bytenr; | |
6911 | } else { | |
6912 | WARN_ON(ref_objectid != level); | |
6913 | } | |
6914 | } else { | |
6915 | WARN_ON(level != -1); | |
6916 | } | |
6917 | first_time = 0; | |
bf4ef679 | 6918 | |
1a40e23b ZY |
6919 | if (ref_path->lowest_level == level) { |
6920 | ref_path->owner_objectid = ref_objectid; | |
1a40e23b ZY |
6921 | ref_path->num_refs = btrfs_ref_num_refs(leaf, ref); |
6922 | } | |
bf4ef679 | 6923 | |
7d9eb12c | 6924 | /* |
1a40e23b ZY |
6925 | * the block is tree root or the block isn't in reference |
6926 | * counted tree. | |
7d9eb12c | 6927 | */ |
1a40e23b ZY |
6928 | if (found_key.objectid == found_key.offset || |
6929 | is_cowonly_root(btrfs_ref_root(leaf, ref))) { | |
6930 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
6931 | ref_path->root_generation = | |
6932 | btrfs_ref_generation(leaf, ref); | |
6933 | if (level < 0) { | |
6934 | /* special reference from the tree log */ | |
6935 | ref_path->nodes[0] = found_key.offset; | |
6936 | ref_path->current_level = 0; | |
6937 | } | |
6938 | ret = 0; | |
6939 | goto out; | |
6940 | } | |
7d9eb12c | 6941 | |
1a40e23b ZY |
6942 | level++; |
6943 | BUG_ON(ref_path->nodes[level] != 0); | |
6944 | ref_path->nodes[level] = found_key.offset; | |
6945 | ref_path->current_level = level; | |
bf4ef679 | 6946 | |
1a40e23b ZY |
6947 | /* |
6948 | * the reference was created in the running transaction, | |
6949 | * no need to continue walking up. | |
6950 | */ | |
6951 | if (btrfs_ref_generation(leaf, ref) == trans->transid) { | |
6952 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
6953 | ref_path->root_generation = | |
6954 | btrfs_ref_generation(leaf, ref); | |
6955 | ret = 0; | |
6956 | goto out; | |
7d9eb12c CM |
6957 | } |
6958 | ||
1a40e23b | 6959 | btrfs_release_path(extent_root, path); |
d899e052 | 6960 | cond_resched(); |
7d9eb12c | 6961 | } |
1a40e23b ZY |
6962 | /* reached max tree level, but no tree root found. */ |
6963 | BUG(); | |
edbd8d4e | 6964 | out: |
1a40e23b ZY |
6965 | btrfs_free_path(path); |
6966 | return ret; | |
edbd8d4e CM |
6967 | } |
6968 | ||
1a40e23b ZY |
6969 | static int btrfs_first_ref_path(struct btrfs_trans_handle *trans, |
6970 | struct btrfs_root *extent_root, | |
6971 | struct btrfs_ref_path *ref_path, | |
6972 | u64 extent_start) | |
a061fc8d | 6973 | { |
1a40e23b ZY |
6974 | memset(ref_path, 0, sizeof(*ref_path)); |
6975 | ref_path->extent_start = extent_start; | |
a061fc8d | 6976 | |
1a40e23b | 6977 | return __next_ref_path(trans, extent_root, ref_path, 1); |
a061fc8d CM |
6978 | } |
6979 | ||
1a40e23b ZY |
6980 | static int btrfs_next_ref_path(struct btrfs_trans_handle *trans, |
6981 | struct btrfs_root *extent_root, | |
6982 | struct btrfs_ref_path *ref_path) | |
edbd8d4e | 6983 | { |
1a40e23b ZY |
6984 | return __next_ref_path(trans, extent_root, ref_path, 0); |
6985 | } | |
6986 | ||
d397712b | 6987 | static noinline int get_new_locations(struct inode *reloc_inode, |
1a40e23b ZY |
6988 | struct btrfs_key *extent_key, |
6989 | u64 offset, int no_fragment, | |
6990 | struct disk_extent **extents, | |
6991 | int *nr_extents) | |
6992 | { | |
6993 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
6994 | struct btrfs_path *path; | |
6995 | struct btrfs_file_extent_item *fi; | |
edbd8d4e | 6996 | struct extent_buffer *leaf; |
1a40e23b ZY |
6997 | struct disk_extent *exts = *extents; |
6998 | struct btrfs_key found_key; | |
6999 | u64 cur_pos; | |
7000 | u64 last_byte; | |
edbd8d4e | 7001 | u32 nritems; |
1a40e23b ZY |
7002 | int nr = 0; |
7003 | int max = *nr_extents; | |
7004 | int ret; | |
edbd8d4e | 7005 | |
1a40e23b ZY |
7006 | WARN_ON(!no_fragment && *extents); |
7007 | if (!exts) { | |
7008 | max = 1; | |
7009 | exts = kmalloc(sizeof(*exts) * max, GFP_NOFS); | |
7010 | if (!exts) | |
7011 | return -ENOMEM; | |
a061fc8d | 7012 | } |
edbd8d4e | 7013 | |
1a40e23b | 7014 | path = btrfs_alloc_path(); |
db5b493a TI |
7015 | if (!path) { |
7016 | if (exts != *extents) | |
7017 | kfree(exts); | |
7018 | return -ENOMEM; | |
7019 | } | |
edbd8d4e | 7020 | |
1a40e23b ZY |
7021 | cur_pos = extent_key->objectid - offset; |
7022 | last_byte = extent_key->objectid + extent_key->offset; | |
7023 | ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino, | |
7024 | cur_pos, 0); | |
7025 | if (ret < 0) | |
7026 | goto out; | |
7027 | if (ret > 0) { | |
7028 | ret = -ENOENT; | |
7029 | goto out; | |
7030 | } | |
edbd8d4e | 7031 | |
1a40e23b | 7032 | while (1) { |
edbd8d4e CM |
7033 | leaf = path->nodes[0]; |
7034 | nritems = btrfs_header_nritems(leaf); | |
1a40e23b ZY |
7035 | if (path->slots[0] >= nritems) { |
7036 | ret = btrfs_next_leaf(root, path); | |
a061fc8d CM |
7037 | if (ret < 0) |
7038 | goto out; | |
1a40e23b ZY |
7039 | if (ret > 0) |
7040 | break; | |
bf4ef679 | 7041 | leaf = path->nodes[0]; |
a061fc8d | 7042 | } |
edbd8d4e CM |
7043 | |
7044 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b ZY |
7045 | if (found_key.offset != cur_pos || |
7046 | found_key.type != BTRFS_EXTENT_DATA_KEY || | |
7047 | found_key.objectid != reloc_inode->i_ino) | |
edbd8d4e CM |
7048 | break; |
7049 | ||
1a40e23b ZY |
7050 | fi = btrfs_item_ptr(leaf, path->slots[0], |
7051 | struct btrfs_file_extent_item); | |
7052 | if (btrfs_file_extent_type(leaf, fi) != | |
7053 | BTRFS_FILE_EXTENT_REG || | |
7054 | btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
edbd8d4e | 7055 | break; |
1a40e23b ZY |
7056 | |
7057 | if (nr == max) { | |
7058 | struct disk_extent *old = exts; | |
7059 | max *= 2; | |
7060 | exts = kzalloc(sizeof(*exts) * max, GFP_NOFS); | |
dac97e51 YS |
7061 | if (!exts) { |
7062 | ret = -ENOMEM; | |
7063 | goto out; | |
7064 | } | |
1a40e23b ZY |
7065 | memcpy(exts, old, sizeof(*exts) * nr); |
7066 | if (old != *extents) | |
7067 | kfree(old); | |
a061fc8d | 7068 | } |
edbd8d4e | 7069 | |
1a40e23b ZY |
7070 | exts[nr].disk_bytenr = |
7071 | btrfs_file_extent_disk_bytenr(leaf, fi); | |
7072 | exts[nr].disk_num_bytes = | |
7073 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7074 | exts[nr].offset = btrfs_file_extent_offset(leaf, fi); | |
7075 | exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
c8b97818 CM |
7076 | exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
7077 | exts[nr].compression = btrfs_file_extent_compression(leaf, fi); | |
7078 | exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi); | |
7079 | exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf, | |
7080 | fi); | |
d899e052 YZ |
7081 | BUG_ON(exts[nr].offset > 0); |
7082 | BUG_ON(exts[nr].compression || exts[nr].encryption); | |
7083 | BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes); | |
edbd8d4e | 7084 | |
1a40e23b ZY |
7085 | cur_pos += exts[nr].num_bytes; |
7086 | nr++; | |
7087 | ||
7088 | if (cur_pos + offset >= last_byte) | |
7089 | break; | |
7090 | ||
7091 | if (no_fragment) { | |
7092 | ret = 1; | |
edbd8d4e | 7093 | goto out; |
1a40e23b ZY |
7094 | } |
7095 | path->slots[0]++; | |
7096 | } | |
7097 | ||
1f80e4db | 7098 | BUG_ON(cur_pos + offset > last_byte); |
1a40e23b ZY |
7099 | if (cur_pos + offset < last_byte) { |
7100 | ret = -ENOENT; | |
7101 | goto out; | |
edbd8d4e CM |
7102 | } |
7103 | ret = 0; | |
7104 | out: | |
1a40e23b ZY |
7105 | btrfs_free_path(path); |
7106 | if (ret) { | |
7107 | if (exts != *extents) | |
7108 | kfree(exts); | |
7109 | } else { | |
7110 | *extents = exts; | |
7111 | *nr_extents = nr; | |
7112 | } | |
7113 | return ret; | |
7114 | } | |
7115 | ||
d397712b | 7116 | static noinline int replace_one_extent(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
7117 | struct btrfs_root *root, |
7118 | struct btrfs_path *path, | |
7119 | struct btrfs_key *extent_key, | |
7120 | struct btrfs_key *leaf_key, | |
7121 | struct btrfs_ref_path *ref_path, | |
7122 | struct disk_extent *new_extents, | |
7123 | int nr_extents) | |
7124 | { | |
7125 | struct extent_buffer *leaf; | |
7126 | struct btrfs_file_extent_item *fi; | |
7127 | struct inode *inode = NULL; | |
7128 | struct btrfs_key key; | |
7129 | u64 lock_start = 0; | |
7130 | u64 lock_end = 0; | |
7131 | u64 num_bytes; | |
7132 | u64 ext_offset; | |
86288a19 | 7133 | u64 search_end = (u64)-1; |
1a40e23b | 7134 | u32 nritems; |
3bb1a1bc | 7135 | int nr_scaned = 0; |
1a40e23b | 7136 | int extent_locked = 0; |
d899e052 | 7137 | int extent_type; |
1a40e23b ZY |
7138 | int ret; |
7139 | ||
3bb1a1bc | 7140 | memcpy(&key, leaf_key, sizeof(key)); |
1a40e23b | 7141 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { |
3bb1a1bc YZ |
7142 | if (key.objectid < ref_path->owner_objectid || |
7143 | (key.objectid == ref_path->owner_objectid && | |
7144 | key.type < BTRFS_EXTENT_DATA_KEY)) { | |
7145 | key.objectid = ref_path->owner_objectid; | |
7146 | key.type = BTRFS_EXTENT_DATA_KEY; | |
7147 | key.offset = 0; | |
7148 | } | |
1a40e23b ZY |
7149 | } |
7150 | ||
7151 | while (1) { | |
7152 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
7153 | if (ret < 0) | |
7154 | goto out; | |
7155 | ||
7156 | leaf = path->nodes[0]; | |
7157 | nritems = btrfs_header_nritems(leaf); | |
7158 | next: | |
7159 | if (extent_locked && ret > 0) { | |
7160 | /* | |
7161 | * the file extent item was modified by someone | |
7162 | * before the extent got locked. | |
7163 | */ | |
1a40e23b ZY |
7164 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
7165 | lock_end, GFP_NOFS); | |
7166 | extent_locked = 0; | |
7167 | } | |
7168 | ||
7169 | if (path->slots[0] >= nritems) { | |
3bb1a1bc | 7170 | if (++nr_scaned > 2) |
1a40e23b ZY |
7171 | break; |
7172 | ||
7173 | BUG_ON(extent_locked); | |
7174 | ret = btrfs_next_leaf(root, path); | |
7175 | if (ret < 0) | |
7176 | goto out; | |
7177 | if (ret > 0) | |
7178 | break; | |
7179 | leaf = path->nodes[0]; | |
7180 | nritems = btrfs_header_nritems(leaf); | |
7181 | } | |
7182 | ||
7183 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
7184 | ||
7185 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { | |
7186 | if ((key.objectid > ref_path->owner_objectid) || | |
7187 | (key.objectid == ref_path->owner_objectid && | |
7188 | key.type > BTRFS_EXTENT_DATA_KEY) || | |
86288a19 | 7189 | key.offset >= search_end) |
1a40e23b ZY |
7190 | break; |
7191 | } | |
7192 | ||
7193 | if (inode && key.objectid != inode->i_ino) { | |
7194 | BUG_ON(extent_locked); | |
7195 | btrfs_release_path(root, path); | |
7196 | mutex_unlock(&inode->i_mutex); | |
7197 | iput(inode); | |
7198 | inode = NULL; | |
7199 | continue; | |
7200 | } | |
7201 | ||
7202 | if (key.type != BTRFS_EXTENT_DATA_KEY) { | |
7203 | path->slots[0]++; | |
7204 | ret = 1; | |
7205 | goto next; | |
7206 | } | |
7207 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
7208 | struct btrfs_file_extent_item); | |
d899e052 YZ |
7209 | extent_type = btrfs_file_extent_type(leaf, fi); |
7210 | if ((extent_type != BTRFS_FILE_EXTENT_REG && | |
7211 | extent_type != BTRFS_FILE_EXTENT_PREALLOC) || | |
1a40e23b ZY |
7212 | (btrfs_file_extent_disk_bytenr(leaf, fi) != |
7213 | extent_key->objectid)) { | |
7214 | path->slots[0]++; | |
7215 | ret = 1; | |
7216 | goto next; | |
7217 | } | |
7218 | ||
7219 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
7220 | ext_offset = btrfs_file_extent_offset(leaf, fi); | |
7221 | ||
86288a19 YZ |
7222 | if (search_end == (u64)-1) { |
7223 | search_end = key.offset - ext_offset + | |
7224 | btrfs_file_extent_ram_bytes(leaf, fi); | |
7225 | } | |
1a40e23b ZY |
7226 | |
7227 | if (!extent_locked) { | |
7228 | lock_start = key.offset; | |
7229 | lock_end = lock_start + num_bytes - 1; | |
7230 | } else { | |
6643558d YZ |
7231 | if (lock_start > key.offset || |
7232 | lock_end + 1 < key.offset + num_bytes) { | |
7233 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
7234 | lock_start, lock_end, GFP_NOFS); | |
7235 | extent_locked = 0; | |
7236 | } | |
1a40e23b ZY |
7237 | } |
7238 | ||
7239 | if (!inode) { | |
7240 | btrfs_release_path(root, path); | |
7241 | ||
7242 | inode = btrfs_iget_locked(root->fs_info->sb, | |
7243 | key.objectid, root); | |
7244 | if (inode->i_state & I_NEW) { | |
7245 | BTRFS_I(inode)->root = root; | |
7246 | BTRFS_I(inode)->location.objectid = | |
7247 | key.objectid; | |
7248 | BTRFS_I(inode)->location.type = | |
7249 | BTRFS_INODE_ITEM_KEY; | |
7250 | BTRFS_I(inode)->location.offset = 0; | |
7251 | btrfs_read_locked_inode(inode); | |
7252 | unlock_new_inode(inode); | |
7253 | } | |
7254 | /* | |
7255 | * some code call btrfs_commit_transaction while | |
7256 | * holding the i_mutex, so we can't use mutex_lock | |
7257 | * here. | |
7258 | */ | |
7259 | if (is_bad_inode(inode) || | |
7260 | !mutex_trylock(&inode->i_mutex)) { | |
7261 | iput(inode); | |
7262 | inode = NULL; | |
7263 | key.offset = (u64)-1; | |
7264 | goto skip; | |
7265 | } | |
7266 | } | |
7267 | ||
7268 | if (!extent_locked) { | |
7269 | struct btrfs_ordered_extent *ordered; | |
7270 | ||
7271 | btrfs_release_path(root, path); | |
7272 | ||
7273 | lock_extent(&BTRFS_I(inode)->io_tree, lock_start, | |
7274 | lock_end, GFP_NOFS); | |
7275 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
7276 | lock_end); | |
7277 | if (ordered && | |
7278 | ordered->file_offset <= lock_end && | |
7279 | ordered->file_offset + ordered->len > lock_start) { | |
7280 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
7281 | lock_start, lock_end, GFP_NOFS); | |
7282 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7283 | btrfs_put_ordered_extent(ordered); | |
7284 | key.offset += num_bytes; | |
7285 | goto skip; | |
7286 | } | |
7287 | if (ordered) | |
7288 | btrfs_put_ordered_extent(ordered); | |
7289 | ||
1a40e23b ZY |
7290 | extent_locked = 1; |
7291 | continue; | |
7292 | } | |
7293 | ||
7294 | if (nr_extents == 1) { | |
7295 | /* update extent pointer in place */ | |
1a40e23b ZY |
7296 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
7297 | new_extents[0].disk_bytenr); | |
7298 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
7299 | new_extents[0].disk_num_bytes); | |
1a40e23b ZY |
7300 | btrfs_mark_buffer_dirty(leaf); |
7301 | ||
7302 | btrfs_drop_extent_cache(inode, key.offset, | |
7303 | key.offset + num_bytes - 1, 0); | |
7304 | ||
7305 | ret = btrfs_inc_extent_ref(trans, root, | |
7306 | new_extents[0].disk_bytenr, | |
7307 | new_extents[0].disk_num_bytes, | |
7308 | leaf->start, | |
7309 | root->root_key.objectid, | |
7310 | trans->transid, | |
3bb1a1bc | 7311 | key.objectid); |
1a40e23b ZY |
7312 | BUG_ON(ret); |
7313 | ||
7314 | ret = btrfs_free_extent(trans, root, | |
7315 | extent_key->objectid, | |
7316 | extent_key->offset, | |
7317 | leaf->start, | |
7318 | btrfs_header_owner(leaf), | |
7319 | btrfs_header_generation(leaf), | |
3bb1a1bc | 7320 | key.objectid, 0); |
1a40e23b ZY |
7321 | BUG_ON(ret); |
7322 | ||
7323 | btrfs_release_path(root, path); | |
7324 | key.offset += num_bytes; | |
7325 | } else { | |
d899e052 YZ |
7326 | BUG_ON(1); |
7327 | #if 0 | |
1a40e23b ZY |
7328 | u64 alloc_hint; |
7329 | u64 extent_len; | |
7330 | int i; | |
7331 | /* | |
7332 | * drop old extent pointer at first, then insert the | |
7333 | * new pointers one bye one | |
7334 | */ | |
7335 | btrfs_release_path(root, path); | |
7336 | ret = btrfs_drop_extents(trans, root, inode, key.offset, | |
7337 | key.offset + num_bytes, | |
7338 | key.offset, &alloc_hint); | |
7339 | BUG_ON(ret); | |
7340 | ||
7341 | for (i = 0; i < nr_extents; i++) { | |
7342 | if (ext_offset >= new_extents[i].num_bytes) { | |
7343 | ext_offset -= new_extents[i].num_bytes; | |
7344 | continue; | |
7345 | } | |
7346 | extent_len = min(new_extents[i].num_bytes - | |
7347 | ext_offset, num_bytes); | |
7348 | ||
7349 | ret = btrfs_insert_empty_item(trans, root, | |
7350 | path, &key, | |
7351 | sizeof(*fi)); | |
7352 | BUG_ON(ret); | |
7353 | ||
7354 | leaf = path->nodes[0]; | |
7355 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
7356 | struct btrfs_file_extent_item); | |
7357 | btrfs_set_file_extent_generation(leaf, fi, | |
7358 | trans->transid); | |
7359 | btrfs_set_file_extent_type(leaf, fi, | |
7360 | BTRFS_FILE_EXTENT_REG); | |
7361 | btrfs_set_file_extent_disk_bytenr(leaf, fi, | |
7362 | new_extents[i].disk_bytenr); | |
7363 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
7364 | new_extents[i].disk_num_bytes); | |
c8b97818 CM |
7365 | btrfs_set_file_extent_ram_bytes(leaf, fi, |
7366 | new_extents[i].ram_bytes); | |
7367 | ||
7368 | btrfs_set_file_extent_compression(leaf, fi, | |
7369 | new_extents[i].compression); | |
7370 | btrfs_set_file_extent_encryption(leaf, fi, | |
7371 | new_extents[i].encryption); | |
7372 | btrfs_set_file_extent_other_encoding(leaf, fi, | |
7373 | new_extents[i].other_encoding); | |
7374 | ||
1a40e23b ZY |
7375 | btrfs_set_file_extent_num_bytes(leaf, fi, |
7376 | extent_len); | |
7377 | ext_offset += new_extents[i].offset; | |
7378 | btrfs_set_file_extent_offset(leaf, fi, | |
7379 | ext_offset); | |
7380 | btrfs_mark_buffer_dirty(leaf); | |
7381 | ||
7382 | btrfs_drop_extent_cache(inode, key.offset, | |
7383 | key.offset + extent_len - 1, 0); | |
7384 | ||
7385 | ret = btrfs_inc_extent_ref(trans, root, | |
7386 | new_extents[i].disk_bytenr, | |
7387 | new_extents[i].disk_num_bytes, | |
7388 | leaf->start, | |
7389 | root->root_key.objectid, | |
3bb1a1bc | 7390 | trans->transid, key.objectid); |
1a40e23b ZY |
7391 | BUG_ON(ret); |
7392 | btrfs_release_path(root, path); | |
7393 | ||
a76a3cd4 | 7394 | inode_add_bytes(inode, extent_len); |
1a40e23b ZY |
7395 | |
7396 | ext_offset = 0; | |
7397 | num_bytes -= extent_len; | |
7398 | key.offset += extent_len; | |
7399 | ||
7400 | if (num_bytes == 0) | |
7401 | break; | |
7402 | } | |
7403 | BUG_ON(i >= nr_extents); | |
d899e052 | 7404 | #endif |
1a40e23b ZY |
7405 | } |
7406 | ||
7407 | if (extent_locked) { | |
1a40e23b ZY |
7408 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
7409 | lock_end, GFP_NOFS); | |
7410 | extent_locked = 0; | |
7411 | } | |
7412 | skip: | |
7413 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS && | |
86288a19 | 7414 | key.offset >= search_end) |
1a40e23b ZY |
7415 | break; |
7416 | ||
7417 | cond_resched(); | |
7418 | } | |
7419 | ret = 0; | |
7420 | out: | |
7421 | btrfs_release_path(root, path); | |
7422 | if (inode) { | |
7423 | mutex_unlock(&inode->i_mutex); | |
7424 | if (extent_locked) { | |
1a40e23b ZY |
7425 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
7426 | lock_end, GFP_NOFS); | |
7427 | } | |
7428 | iput(inode); | |
7429 | } | |
7430 | return ret; | |
7431 | } | |
7432 | ||
1a40e23b ZY |
7433 | int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, |
7434 | struct btrfs_root *root, | |
7435 | struct extent_buffer *buf, u64 orig_start) | |
7436 | { | |
7437 | int level; | |
7438 | int ret; | |
7439 | ||
7440 | BUG_ON(btrfs_header_generation(buf) != trans->transid); | |
7441 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
7442 | ||
7443 | level = btrfs_header_level(buf); | |
7444 | if (level == 0) { | |
7445 | struct btrfs_leaf_ref *ref; | |
7446 | struct btrfs_leaf_ref *orig_ref; | |
7447 | ||
7448 | orig_ref = btrfs_lookup_leaf_ref(root, orig_start); | |
7449 | if (!orig_ref) | |
7450 | return -ENOENT; | |
7451 | ||
7452 | ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems); | |
7453 | if (!ref) { | |
7454 | btrfs_free_leaf_ref(root, orig_ref); | |
7455 | return -ENOMEM; | |
7456 | } | |
7457 | ||
7458 | ref->nritems = orig_ref->nritems; | |
7459 | memcpy(ref->extents, orig_ref->extents, | |
7460 | sizeof(ref->extents[0]) * ref->nritems); | |
7461 | ||
7462 | btrfs_free_leaf_ref(root, orig_ref); | |
7463 | ||
7464 | ref->root_gen = trans->transid; | |
7465 | ref->bytenr = buf->start; | |
7466 | ref->owner = btrfs_header_owner(buf); | |
7467 | ref->generation = btrfs_header_generation(buf); | |
bd56b302 | 7468 | |
1a40e23b ZY |
7469 | ret = btrfs_add_leaf_ref(root, ref, 0); |
7470 | WARN_ON(ret); | |
7471 | btrfs_free_leaf_ref(root, ref); | |
7472 | } | |
7473 | return 0; | |
7474 | } | |
7475 | ||
d397712b | 7476 | static noinline int invalidate_extent_cache(struct btrfs_root *root, |
1a40e23b ZY |
7477 | struct extent_buffer *leaf, |
7478 | struct btrfs_block_group_cache *group, | |
7479 | struct btrfs_root *target_root) | |
7480 | { | |
7481 | struct btrfs_key key; | |
7482 | struct inode *inode = NULL; | |
7483 | struct btrfs_file_extent_item *fi; | |
2ac55d41 | 7484 | struct extent_state *cached_state = NULL; |
1a40e23b ZY |
7485 | u64 num_bytes; |
7486 | u64 skip_objectid = 0; | |
7487 | u32 nritems; | |
7488 | u32 i; | |
7489 | ||
7490 | nritems = btrfs_header_nritems(leaf); | |
7491 | for (i = 0; i < nritems; i++) { | |
7492 | btrfs_item_key_to_cpu(leaf, &key, i); | |
7493 | if (key.objectid == skip_objectid || | |
7494 | key.type != BTRFS_EXTENT_DATA_KEY) | |
7495 | continue; | |
7496 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
7497 | if (btrfs_file_extent_type(leaf, fi) == | |
7498 | BTRFS_FILE_EXTENT_INLINE) | |
7499 | continue; | |
7500 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
7501 | continue; | |
7502 | if (!inode || inode->i_ino != key.objectid) { | |
7503 | iput(inode); | |
7504 | inode = btrfs_ilookup(target_root->fs_info->sb, | |
7505 | key.objectid, target_root, 1); | |
7506 | } | |
7507 | if (!inode) { | |
7508 | skip_objectid = key.objectid; | |
7509 | continue; | |
7510 | } | |
7511 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
7512 | ||
2ac55d41 JB |
7513 | lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset, |
7514 | key.offset + num_bytes - 1, 0, &cached_state, | |
7515 | GFP_NOFS); | |
1a40e23b ZY |
7516 | btrfs_drop_extent_cache(inode, key.offset, |
7517 | key.offset + num_bytes - 1, 1); | |
2ac55d41 JB |
7518 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset, |
7519 | key.offset + num_bytes - 1, &cached_state, | |
7520 | GFP_NOFS); | |
1a40e23b ZY |
7521 | cond_resched(); |
7522 | } | |
7523 | iput(inode); | |
7524 | return 0; | |
7525 | } | |
7526 | ||
d397712b | 7527 | static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
7528 | struct btrfs_root *root, |
7529 | struct extent_buffer *leaf, | |
7530 | struct btrfs_block_group_cache *group, | |
7531 | struct inode *reloc_inode) | |
7532 | { | |
7533 | struct btrfs_key key; | |
7534 | struct btrfs_key extent_key; | |
7535 | struct btrfs_file_extent_item *fi; | |
7536 | struct btrfs_leaf_ref *ref; | |
7537 | struct disk_extent *new_extent; | |
7538 | u64 bytenr; | |
7539 | u64 num_bytes; | |
7540 | u32 nritems; | |
7541 | u32 i; | |
7542 | int ext_index; | |
7543 | int nr_extent; | |
7544 | int ret; | |
7545 | ||
7546 | new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS); | |
db5b493a TI |
7547 | if (!new_extent) |
7548 | return -ENOMEM; | |
1a40e23b ZY |
7549 | |
7550 | ref = btrfs_lookup_leaf_ref(root, leaf->start); | |
7551 | BUG_ON(!ref); | |
7552 | ||
7553 | ext_index = -1; | |
7554 | nritems = btrfs_header_nritems(leaf); | |
7555 | for (i = 0; i < nritems; i++) { | |
7556 | btrfs_item_key_to_cpu(leaf, &key, i); | |
7557 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
7558 | continue; | |
7559 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
7560 | if (btrfs_file_extent_type(leaf, fi) == | |
7561 | BTRFS_FILE_EXTENT_INLINE) | |
7562 | continue; | |
7563 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
7564 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7565 | if (bytenr == 0) | |
7566 | continue; | |
7567 | ||
7568 | ext_index++; | |
7569 | if (bytenr >= group->key.objectid + group->key.offset || | |
7570 | bytenr + num_bytes <= group->key.objectid) | |
7571 | continue; | |
7572 | ||
7573 | extent_key.objectid = bytenr; | |
7574 | extent_key.offset = num_bytes; | |
7575 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
7576 | nr_extent = 1; | |
7577 | ret = get_new_locations(reloc_inode, &extent_key, | |
7578 | group->key.objectid, 1, | |
7579 | &new_extent, &nr_extent); | |
7580 | if (ret > 0) | |
7581 | continue; | |
7582 | BUG_ON(ret < 0); | |
7583 | ||
7584 | BUG_ON(ref->extents[ext_index].bytenr != bytenr); | |
7585 | BUG_ON(ref->extents[ext_index].num_bytes != num_bytes); | |
7586 | ref->extents[ext_index].bytenr = new_extent->disk_bytenr; | |
7587 | ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes; | |
7588 | ||
1a40e23b ZY |
7589 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
7590 | new_extent->disk_bytenr); | |
7591 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
7592 | new_extent->disk_num_bytes); | |
1a40e23b ZY |
7593 | btrfs_mark_buffer_dirty(leaf); |
7594 | ||
7595 | ret = btrfs_inc_extent_ref(trans, root, | |
7596 | new_extent->disk_bytenr, | |
7597 | new_extent->disk_num_bytes, | |
7598 | leaf->start, | |
7599 | root->root_key.objectid, | |
3bb1a1bc | 7600 | trans->transid, key.objectid); |
1a40e23b | 7601 | BUG_ON(ret); |
56bec294 | 7602 | |
1a40e23b ZY |
7603 | ret = btrfs_free_extent(trans, root, |
7604 | bytenr, num_bytes, leaf->start, | |
7605 | btrfs_header_owner(leaf), | |
7606 | btrfs_header_generation(leaf), | |
3bb1a1bc | 7607 | key.objectid, 0); |
1a40e23b ZY |
7608 | BUG_ON(ret); |
7609 | cond_resched(); | |
7610 | } | |
7611 | kfree(new_extent); | |
7612 | BUG_ON(ext_index + 1 != ref->nritems); | |
7613 | btrfs_free_leaf_ref(root, ref); | |
7614 | return 0; | |
7615 | } | |
7616 | ||
f82d02d9 YZ |
7617 | int btrfs_free_reloc_root(struct btrfs_trans_handle *trans, |
7618 | struct btrfs_root *root) | |
1a40e23b ZY |
7619 | { |
7620 | struct btrfs_root *reloc_root; | |
f82d02d9 | 7621 | int ret; |
1a40e23b ZY |
7622 | |
7623 | if (root->reloc_root) { | |
7624 | reloc_root = root->reloc_root; | |
7625 | root->reloc_root = NULL; | |
7626 | list_add(&reloc_root->dead_list, | |
7627 | &root->fs_info->dead_reloc_roots); | |
f82d02d9 YZ |
7628 | |
7629 | btrfs_set_root_bytenr(&reloc_root->root_item, | |
7630 | reloc_root->node->start); | |
7631 | btrfs_set_root_level(&root->root_item, | |
7632 | btrfs_header_level(reloc_root->node)); | |
7633 | memset(&reloc_root->root_item.drop_progress, 0, | |
7634 | sizeof(struct btrfs_disk_key)); | |
7635 | reloc_root->root_item.drop_level = 0; | |
7636 | ||
7637 | ret = btrfs_update_root(trans, root->fs_info->tree_root, | |
7638 | &reloc_root->root_key, | |
7639 | &reloc_root->root_item); | |
7640 | BUG_ON(ret); | |
1a40e23b ZY |
7641 | } |
7642 | return 0; | |
7643 | } | |
7644 | ||
7645 | int btrfs_drop_dead_reloc_roots(struct btrfs_root *root) | |
7646 | { | |
7647 | struct btrfs_trans_handle *trans; | |
7648 | struct btrfs_root *reloc_root; | |
7649 | struct btrfs_root *prev_root = NULL; | |
7650 | struct list_head dead_roots; | |
7651 | int ret; | |
7652 | unsigned long nr; | |
7653 | ||
7654 | INIT_LIST_HEAD(&dead_roots); | |
7655 | list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots); | |
7656 | ||
7657 | while (!list_empty(&dead_roots)) { | |
7658 | reloc_root = list_entry(dead_roots.prev, | |
7659 | struct btrfs_root, dead_list); | |
7660 | list_del_init(&reloc_root->dead_list); | |
7661 | ||
7662 | BUG_ON(reloc_root->commit_root != NULL); | |
7663 | while (1) { | |
7664 | trans = btrfs_join_transaction(root, 1); | |
3612b495 | 7665 | BUG_ON(IS_ERR(trans)); |
1a40e23b ZY |
7666 | |
7667 | mutex_lock(&root->fs_info->drop_mutex); | |
7668 | ret = btrfs_drop_snapshot(trans, reloc_root); | |
7669 | if (ret != -EAGAIN) | |
7670 | break; | |
7671 | mutex_unlock(&root->fs_info->drop_mutex); | |
7672 | ||
7673 | nr = trans->blocks_used; | |
7674 | ret = btrfs_end_transaction(trans, root); | |
7675 | BUG_ON(ret); | |
7676 | btrfs_btree_balance_dirty(root, nr); | |
7677 | } | |
7678 | ||
7679 | free_extent_buffer(reloc_root->node); | |
7680 | ||
7681 | ret = btrfs_del_root(trans, root->fs_info->tree_root, | |
7682 | &reloc_root->root_key); | |
7683 | BUG_ON(ret); | |
7684 | mutex_unlock(&root->fs_info->drop_mutex); | |
7685 | ||
7686 | nr = trans->blocks_used; | |
7687 | ret = btrfs_end_transaction(trans, root); | |
7688 | BUG_ON(ret); | |
7689 | btrfs_btree_balance_dirty(root, nr); | |
7690 | ||
7691 | kfree(prev_root); | |
7692 | prev_root = reloc_root; | |
7693 | } | |
7694 | if (prev_root) { | |
7695 | btrfs_remove_leaf_refs(prev_root, (u64)-1, 0); | |
7696 | kfree(prev_root); | |
7697 | } | |
7698 | return 0; | |
7699 | } | |
7700 | ||
7701 | int btrfs_add_dead_reloc_root(struct btrfs_root *root) | |
7702 | { | |
7703 | list_add(&root->dead_list, &root->fs_info->dead_reloc_roots); | |
7704 | return 0; | |
7705 | } | |
7706 | ||
7707 | int btrfs_cleanup_reloc_trees(struct btrfs_root *root) | |
7708 | { | |
7709 | struct btrfs_root *reloc_root; | |
7710 | struct btrfs_trans_handle *trans; | |
7711 | struct btrfs_key location; | |
7712 | int found; | |
7713 | int ret; | |
7714 | ||
7715 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
7716 | ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL); | |
7717 | BUG_ON(ret); | |
7718 | found = !list_empty(&root->fs_info->dead_reloc_roots); | |
7719 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
7720 | ||
7721 | if (found) { | |
7722 | trans = btrfs_start_transaction(root, 1); | |
98d5dc13 | 7723 | BUG_ON(IS_ERR(trans)); |
1a40e23b ZY |
7724 | ret = btrfs_commit_transaction(trans, root); |
7725 | BUG_ON(ret); | |
7726 | } | |
7727 | ||
7728 | location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
7729 | location.offset = (u64)-1; | |
7730 | location.type = BTRFS_ROOT_ITEM_KEY; | |
7731 | ||
7732 | reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location); | |
7733 | BUG_ON(!reloc_root); | |
66b4ffd1 JB |
7734 | ret = btrfs_orphan_cleanup(reloc_root); |
7735 | BUG_ON(ret); | |
1a40e23b ZY |
7736 | return 0; |
7737 | } | |
7738 | ||
d397712b | 7739 | static noinline int init_reloc_tree(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
7740 | struct btrfs_root *root) |
7741 | { | |
7742 | struct btrfs_root *reloc_root; | |
7743 | struct extent_buffer *eb; | |
7744 | struct btrfs_root_item *root_item; | |
7745 | struct btrfs_key root_key; | |
7746 | int ret; | |
7747 | ||
7748 | BUG_ON(!root->ref_cows); | |
7749 | if (root->reloc_root) | |
7750 | return 0; | |
7751 | ||
7752 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); | |
db5b493a TI |
7753 | if (!root_item) |
7754 | return -ENOMEM; | |
1a40e23b ZY |
7755 | |
7756 | ret = btrfs_copy_root(trans, root, root->commit_root, | |
7757 | &eb, BTRFS_TREE_RELOC_OBJECTID); | |
7758 | BUG_ON(ret); | |
7759 | ||
7760 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
7761 | root_key.offset = root->root_key.objectid; | |
7762 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
7763 | ||
7764 | memcpy(root_item, &root->root_item, sizeof(root_item)); | |
7765 | btrfs_set_root_refs(root_item, 0); | |
7766 | btrfs_set_root_bytenr(root_item, eb->start); | |
7767 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
84234f3a | 7768 | btrfs_set_root_generation(root_item, trans->transid); |
1a40e23b ZY |
7769 | |
7770 | btrfs_tree_unlock(eb); | |
7771 | free_extent_buffer(eb); | |
7772 | ||
7773 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, | |
7774 | &root_key, root_item); | |
7775 | BUG_ON(ret); | |
7776 | kfree(root_item); | |
7777 | ||
7778 | reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root, | |
7779 | &root_key); | |
db5b493a | 7780 | BUG_ON(IS_ERR(reloc_root)); |
1a40e23b ZY |
7781 | reloc_root->last_trans = trans->transid; |
7782 | reloc_root->commit_root = NULL; | |
7783 | reloc_root->ref_tree = &root->fs_info->reloc_ref_tree; | |
7784 | ||
7785 | root->reloc_root = reloc_root; | |
7786 | return 0; | |
7787 | } | |
7788 | ||
7789 | /* | |
7790 | * Core function of space balance. | |
7791 | * | |
7792 | * The idea is using reloc trees to relocate tree blocks in reference | |
f82d02d9 YZ |
7793 | * counted roots. There is one reloc tree for each subvol, and all |
7794 | * reloc trees share same root key objectid. Reloc trees are snapshots | |
7795 | * of the latest committed roots of subvols (root->commit_root). | |
7796 | * | |
7797 | * To relocate a tree block referenced by a subvol, there are two steps. | |
7798 | * COW the block through subvol's reloc tree, then update block pointer | |
7799 | * in the subvol to point to the new block. Since all reloc trees share | |
7800 | * same root key objectid, doing special handing for tree blocks owned | |
7801 | * by them is easy. Once a tree block has been COWed in one reloc tree, | |
7802 | * we can use the resulting new block directly when the same block is | |
7803 | * required to COW again through other reloc trees. By this way, relocated | |
7804 | * tree blocks are shared between reloc trees, so they are also shared | |
7805 | * between subvols. | |
1a40e23b | 7806 | */ |
d397712b | 7807 | static noinline int relocate_one_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
7808 | struct btrfs_root *root, |
7809 | struct btrfs_path *path, | |
7810 | struct btrfs_key *first_key, | |
7811 | struct btrfs_ref_path *ref_path, | |
7812 | struct btrfs_block_group_cache *group, | |
7813 | struct inode *reloc_inode) | |
7814 | { | |
7815 | struct btrfs_root *reloc_root; | |
7816 | struct extent_buffer *eb = NULL; | |
7817 | struct btrfs_key *keys; | |
7818 | u64 *nodes; | |
7819 | int level; | |
f82d02d9 | 7820 | int shared_level; |
1a40e23b | 7821 | int lowest_level = 0; |
1a40e23b ZY |
7822 | int ret; |
7823 | ||
7824 | if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
7825 | lowest_level = ref_path->owner_objectid; | |
7826 | ||
f82d02d9 | 7827 | if (!root->ref_cows) { |
1a40e23b ZY |
7828 | path->lowest_level = lowest_level; |
7829 | ret = btrfs_search_slot(trans, root, first_key, path, 0, 1); | |
7830 | BUG_ON(ret < 0); | |
7831 | path->lowest_level = 0; | |
7832 | btrfs_release_path(root, path); | |
7833 | return 0; | |
7834 | } | |
7835 | ||
1a40e23b ZY |
7836 | mutex_lock(&root->fs_info->tree_reloc_mutex); |
7837 | ret = init_reloc_tree(trans, root); | |
7838 | BUG_ON(ret); | |
7839 | reloc_root = root->reloc_root; | |
7840 | ||
f82d02d9 YZ |
7841 | shared_level = ref_path->shared_level; |
7842 | ref_path->shared_level = BTRFS_MAX_LEVEL - 1; | |
1a40e23b | 7843 | |
f82d02d9 YZ |
7844 | keys = ref_path->node_keys; |
7845 | nodes = ref_path->new_nodes; | |
7846 | memset(&keys[shared_level + 1], 0, | |
7847 | sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
7848 | memset(&nodes[shared_level + 1], 0, | |
7849 | sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
1a40e23b | 7850 | |
f82d02d9 YZ |
7851 | if (nodes[lowest_level] == 0) { |
7852 | path->lowest_level = lowest_level; | |
7853 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, | |
7854 | 0, 1); | |
7855 | BUG_ON(ret); | |
7856 | for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) { | |
7857 | eb = path->nodes[level]; | |
7858 | if (!eb || eb == reloc_root->node) | |
7859 | break; | |
7860 | nodes[level] = eb->start; | |
7861 | if (level == 0) | |
7862 | btrfs_item_key_to_cpu(eb, &keys[level], 0); | |
7863 | else | |
7864 | btrfs_node_key_to_cpu(eb, &keys[level], 0); | |
7865 | } | |
2b82032c YZ |
7866 | if (nodes[0] && |
7867 | ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
7868 | eb = path->nodes[0]; |
7869 | ret = replace_extents_in_leaf(trans, reloc_root, eb, | |
7870 | group, reloc_inode); | |
7871 | BUG_ON(ret); | |
7872 | } | |
7873 | btrfs_release_path(reloc_root, path); | |
7874 | } else { | |
1a40e23b | 7875 | ret = btrfs_merge_path(trans, reloc_root, keys, nodes, |
f82d02d9 | 7876 | lowest_level); |
1a40e23b ZY |
7877 | BUG_ON(ret); |
7878 | } | |
7879 | ||
1a40e23b ZY |
7880 | /* |
7881 | * replace tree blocks in the fs tree with tree blocks in | |
7882 | * the reloc tree. | |
7883 | */ | |
7884 | ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level); | |
7885 | BUG_ON(ret < 0); | |
7886 | ||
7887 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
7888 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, |
7889 | 0, 0); | |
7890 | BUG_ON(ret); | |
7891 | extent_buffer_get(path->nodes[0]); | |
7892 | eb = path->nodes[0]; | |
7893 | btrfs_release_path(reloc_root, path); | |
1a40e23b ZY |
7894 | ret = invalidate_extent_cache(reloc_root, eb, group, root); |
7895 | BUG_ON(ret); | |
7896 | free_extent_buffer(eb); | |
7897 | } | |
1a40e23b | 7898 | |
f82d02d9 | 7899 | mutex_unlock(&root->fs_info->tree_reloc_mutex); |
1a40e23b | 7900 | path->lowest_level = 0; |
1a40e23b ZY |
7901 | return 0; |
7902 | } | |
7903 | ||
d397712b | 7904 | static noinline int relocate_tree_block(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
7905 | struct btrfs_root *root, |
7906 | struct btrfs_path *path, | |
7907 | struct btrfs_key *first_key, | |
7908 | struct btrfs_ref_path *ref_path) | |
7909 | { | |
7910 | int ret; | |
1a40e23b ZY |
7911 | |
7912 | ret = relocate_one_path(trans, root, path, first_key, | |
7913 | ref_path, NULL, NULL); | |
7914 | BUG_ON(ret); | |
7915 | ||
1a40e23b ZY |
7916 | return 0; |
7917 | } | |
7918 | ||
d397712b | 7919 | static noinline int del_extent_zero(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
7920 | struct btrfs_root *extent_root, |
7921 | struct btrfs_path *path, | |
7922 | struct btrfs_key *extent_key) | |
7923 | { | |
7924 | int ret; | |
7925 | ||
1a40e23b ZY |
7926 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); |
7927 | if (ret) | |
7928 | goto out; | |
7929 | ret = btrfs_del_item(trans, extent_root, path); | |
7930 | out: | |
7931 | btrfs_release_path(extent_root, path); | |
1a40e23b ZY |
7932 | return ret; |
7933 | } | |
7934 | ||
d397712b | 7935 | static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info, |
1a40e23b ZY |
7936 | struct btrfs_ref_path *ref_path) |
7937 | { | |
7938 | struct btrfs_key root_key; | |
7939 | ||
7940 | root_key.objectid = ref_path->root_objectid; | |
7941 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
7942 | if (is_cowonly_root(ref_path->root_objectid)) | |
7943 | root_key.offset = 0; | |
7944 | else | |
7945 | root_key.offset = (u64)-1; | |
7946 | ||
7947 | return btrfs_read_fs_root_no_name(fs_info, &root_key); | |
7948 | } | |
7949 | ||
d397712b | 7950 | static noinline int relocate_one_extent(struct btrfs_root *extent_root, |
1a40e23b ZY |
7951 | struct btrfs_path *path, |
7952 | struct btrfs_key *extent_key, | |
7953 | struct btrfs_block_group_cache *group, | |
7954 | struct inode *reloc_inode, int pass) | |
7955 | { | |
7956 | struct btrfs_trans_handle *trans; | |
7957 | struct btrfs_root *found_root; | |
7958 | struct btrfs_ref_path *ref_path = NULL; | |
7959 | struct disk_extent *new_extents = NULL; | |
7960 | int nr_extents = 0; | |
7961 | int loops; | |
7962 | int ret; | |
7963 | int level; | |
7964 | struct btrfs_key first_key; | |
7965 | u64 prev_block = 0; | |
7966 | ||
1a40e23b ZY |
7967 | |
7968 | trans = btrfs_start_transaction(extent_root, 1); | |
98d5dc13 | 7969 | BUG_ON(IS_ERR(trans)); |
1a40e23b ZY |
7970 | |
7971 | if (extent_key->objectid == 0) { | |
7972 | ret = del_extent_zero(trans, extent_root, path, extent_key); | |
7973 | goto out; | |
7974 | } | |
7975 | ||
7976 | ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS); | |
7977 | if (!ref_path) { | |
d397712b CM |
7978 | ret = -ENOMEM; |
7979 | goto out; | |
1a40e23b ZY |
7980 | } |
7981 | ||
7982 | for (loops = 0; ; loops++) { | |
7983 | if (loops == 0) { | |
7984 | ret = btrfs_first_ref_path(trans, extent_root, ref_path, | |
7985 | extent_key->objectid); | |
7986 | } else { | |
7987 | ret = btrfs_next_ref_path(trans, extent_root, ref_path); | |
7988 | } | |
7989 | if (ret < 0) | |
7990 | goto out; | |
7991 | if (ret > 0) | |
7992 | break; | |
7993 | ||
7994 | if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID || | |
7995 | ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
7996 | continue; | |
7997 | ||
7998 | found_root = read_ref_root(extent_root->fs_info, ref_path); | |
7999 | BUG_ON(!found_root); | |
8000 | /* | |
8001 | * for reference counted tree, only process reference paths | |
8002 | * rooted at the latest committed root. | |
8003 | */ | |
8004 | if (found_root->ref_cows && | |
8005 | ref_path->root_generation != found_root->root_key.offset) | |
8006 | continue; | |
8007 | ||
8008 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
8009 | if (pass == 0) { | |
8010 | /* | |
8011 | * copy data extents to new locations | |
8012 | */ | |
8013 | u64 group_start = group->key.objectid; | |
8014 | ret = relocate_data_extent(reloc_inode, | |
8015 | extent_key, | |
8016 | group_start); | |
8017 | if (ret < 0) | |
8018 | goto out; | |
8019 | break; | |
8020 | } | |
8021 | level = 0; | |
8022 | } else { | |
8023 | level = ref_path->owner_objectid; | |
8024 | } | |
8025 | ||
8026 | if (prev_block != ref_path->nodes[level]) { | |
8027 | struct extent_buffer *eb; | |
8028 | u64 block_start = ref_path->nodes[level]; | |
8029 | u64 block_size = btrfs_level_size(found_root, level); | |
8030 | ||
8031 | eb = read_tree_block(found_root, block_start, | |
8032 | block_size, 0); | |
97d9a8a4 TI |
8033 | if (!eb) { |
8034 | ret = -EIO; | |
8035 | goto out; | |
8036 | } | |
1a40e23b ZY |
8037 | btrfs_tree_lock(eb); |
8038 | BUG_ON(level != btrfs_header_level(eb)); | |
8039 | ||
8040 | if (level == 0) | |
8041 | btrfs_item_key_to_cpu(eb, &first_key, 0); | |
8042 | else | |
8043 | btrfs_node_key_to_cpu(eb, &first_key, 0); | |
8044 | ||
8045 | btrfs_tree_unlock(eb); | |
8046 | free_extent_buffer(eb); | |
8047 | prev_block = block_start; | |
8048 | } | |
8049 | ||
24562425 | 8050 | mutex_lock(&extent_root->fs_info->trans_mutex); |
e4404d6e | 8051 | btrfs_record_root_in_trans(found_root); |
24562425 | 8052 | mutex_unlock(&extent_root->fs_info->trans_mutex); |
e4404d6e YZ |
8053 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { |
8054 | /* | |
8055 | * try to update data extent references while | |
8056 | * keeping metadata shared between snapshots. | |
8057 | */ | |
8058 | if (pass == 1) { | |
8059 | ret = relocate_one_path(trans, found_root, | |
8060 | path, &first_key, ref_path, | |
8061 | group, reloc_inode); | |
8062 | if (ret < 0) | |
8063 | goto out; | |
8064 | continue; | |
8065 | } | |
1a40e23b ZY |
8066 | /* |
8067 | * use fallback method to process the remaining | |
8068 | * references. | |
8069 | */ | |
8070 | if (!new_extents) { | |
8071 | u64 group_start = group->key.objectid; | |
d899e052 YZ |
8072 | new_extents = kmalloc(sizeof(*new_extents), |
8073 | GFP_NOFS); | |
8d413713 TI |
8074 | if (!new_extents) { |
8075 | ret = -ENOMEM; | |
8076 | goto out; | |
8077 | } | |
d899e052 | 8078 | nr_extents = 1; |
1a40e23b ZY |
8079 | ret = get_new_locations(reloc_inode, |
8080 | extent_key, | |
d899e052 | 8081 | group_start, 1, |
1a40e23b ZY |
8082 | &new_extents, |
8083 | &nr_extents); | |
d899e052 | 8084 | if (ret) |
1a40e23b ZY |
8085 | goto out; |
8086 | } | |
1a40e23b ZY |
8087 | ret = replace_one_extent(trans, found_root, |
8088 | path, extent_key, | |
8089 | &first_key, ref_path, | |
8090 | new_extents, nr_extents); | |
e4404d6e | 8091 | } else { |
1a40e23b ZY |
8092 | ret = relocate_tree_block(trans, found_root, path, |
8093 | &first_key, ref_path); | |
1a40e23b ZY |
8094 | } |
8095 | if (ret < 0) | |
8096 | goto out; | |
8097 | } | |
8098 | ret = 0; | |
8099 | out: | |
8100 | btrfs_end_transaction(trans, extent_root); | |
8101 | kfree(new_extents); | |
8102 | kfree(ref_path); | |
1a40e23b ZY |
8103 | return ret; |
8104 | } | |
5d4f98a2 | 8105 | #endif |
1a40e23b | 8106 | |
ec44a35c CM |
8107 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
8108 | { | |
8109 | u64 num_devices; | |
8110 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
8111 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
8112 | ||
cd02dca5 CM |
8113 | /* |
8114 | * we add in the count of missing devices because we want | |
8115 | * to make sure that any RAID levels on a degraded FS | |
8116 | * continue to be honored. | |
8117 | */ | |
8118 | num_devices = root->fs_info->fs_devices->rw_devices + | |
8119 | root->fs_info->fs_devices->missing_devices; | |
8120 | ||
ec44a35c CM |
8121 | if (num_devices == 1) { |
8122 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
8123 | stripped = flags & ~stripped; | |
8124 | ||
8125 | /* turn raid0 into single device chunks */ | |
8126 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
8127 | return stripped; | |
8128 | ||
8129 | /* turn mirroring into duplication */ | |
8130 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
8131 | BTRFS_BLOCK_GROUP_RAID10)) | |
8132 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
8133 | return flags; | |
8134 | } else { | |
8135 | /* they already had raid on here, just return */ | |
ec44a35c CM |
8136 | if (flags & stripped) |
8137 | return flags; | |
8138 | ||
8139 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
8140 | stripped = flags & ~stripped; | |
8141 | ||
8142 | /* switch duplicated blocks with raid1 */ | |
8143 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
8144 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
8145 | ||
8146 | /* turn single device chunks into raid0 */ | |
8147 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
8148 | } | |
8149 | return flags; | |
8150 | } | |
8151 | ||
f0486c68 | 8152 | static int set_block_group_ro(struct btrfs_block_group_cache *cache) |
0ef3e66b | 8153 | { |
f0486c68 YZ |
8154 | struct btrfs_space_info *sinfo = cache->space_info; |
8155 | u64 num_bytes; | |
8156 | int ret = -ENOSPC; | |
0ef3e66b | 8157 | |
f0486c68 YZ |
8158 | if (cache->ro) |
8159 | return 0; | |
c286ac48 | 8160 | |
f0486c68 YZ |
8161 | spin_lock(&sinfo->lock); |
8162 | spin_lock(&cache->lock); | |
8163 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
8164 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
8165 | ||
8166 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
8167 | sinfo->bytes_may_use + sinfo->bytes_readonly + | |
65e5341b | 8168 | cache->reserved_pinned + num_bytes <= sinfo->total_bytes) { |
f0486c68 YZ |
8169 | sinfo->bytes_readonly += num_bytes; |
8170 | sinfo->bytes_reserved += cache->reserved_pinned; | |
8171 | cache->reserved_pinned = 0; | |
8172 | cache->ro = 1; | |
8173 | ret = 0; | |
8174 | } | |
65e5341b | 8175 | |
f0486c68 YZ |
8176 | spin_unlock(&cache->lock); |
8177 | spin_unlock(&sinfo->lock); | |
8178 | return ret; | |
8179 | } | |
7d9eb12c | 8180 | |
f0486c68 YZ |
8181 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
8182 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 8183 | |
f0486c68 YZ |
8184 | { |
8185 | struct btrfs_trans_handle *trans; | |
8186 | u64 alloc_flags; | |
8187 | int ret; | |
7d9eb12c | 8188 | |
f0486c68 | 8189 | BUG_ON(cache->ro); |
0ef3e66b | 8190 | |
f0486c68 YZ |
8191 | trans = btrfs_join_transaction(root, 1); |
8192 | BUG_ON(IS_ERR(trans)); | |
5d4f98a2 | 8193 | |
f0486c68 YZ |
8194 | alloc_flags = update_block_group_flags(root, cache->flags); |
8195 | if (alloc_flags != cache->flags) | |
0e4f8f88 CM |
8196 | do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
8197 | CHUNK_ALLOC_FORCE); | |
5d4f98a2 | 8198 | |
f0486c68 YZ |
8199 | ret = set_block_group_ro(cache); |
8200 | if (!ret) | |
8201 | goto out; | |
8202 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
0e4f8f88 CM |
8203 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
8204 | CHUNK_ALLOC_FORCE); | |
f0486c68 YZ |
8205 | if (ret < 0) |
8206 | goto out; | |
8207 | ret = set_block_group_ro(cache); | |
8208 | out: | |
8209 | btrfs_end_transaction(trans, root); | |
8210 | return ret; | |
8211 | } | |
5d4f98a2 | 8212 | |
c87f08ca CM |
8213 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
8214 | struct btrfs_root *root, u64 type) | |
8215 | { | |
8216 | u64 alloc_flags = get_alloc_profile(root, type); | |
0e4f8f88 CM |
8217 | return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
8218 | CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
8219 | } |
8220 | ||
6d07bcec MX |
8221 | /* |
8222 | * helper to account the unused space of all the readonly block group in the | |
8223 | * list. takes mirrors into account. | |
8224 | */ | |
8225 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
8226 | { | |
8227 | struct btrfs_block_group_cache *block_group; | |
8228 | u64 free_bytes = 0; | |
8229 | int factor; | |
8230 | ||
8231 | list_for_each_entry(block_group, groups_list, list) { | |
8232 | spin_lock(&block_group->lock); | |
8233 | ||
8234 | if (!block_group->ro) { | |
8235 | spin_unlock(&block_group->lock); | |
8236 | continue; | |
8237 | } | |
8238 | ||
8239 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
8240 | BTRFS_BLOCK_GROUP_RAID10 | | |
8241 | BTRFS_BLOCK_GROUP_DUP)) | |
8242 | factor = 2; | |
8243 | else | |
8244 | factor = 1; | |
8245 | ||
8246 | free_bytes += (block_group->key.offset - | |
8247 | btrfs_block_group_used(&block_group->item)) * | |
8248 | factor; | |
8249 | ||
8250 | spin_unlock(&block_group->lock); | |
8251 | } | |
8252 | ||
8253 | return free_bytes; | |
8254 | } | |
8255 | ||
8256 | /* | |
8257 | * helper to account the unused space of all the readonly block group in the | |
8258 | * space_info. takes mirrors into account. | |
8259 | */ | |
8260 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
8261 | { | |
8262 | int i; | |
8263 | u64 free_bytes = 0; | |
8264 | ||
8265 | spin_lock(&sinfo->lock); | |
8266 | ||
8267 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
8268 | if (!list_empty(&sinfo->block_groups[i])) | |
8269 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
8270 | &sinfo->block_groups[i]); | |
8271 | ||
8272 | spin_unlock(&sinfo->lock); | |
8273 | ||
8274 | return free_bytes; | |
8275 | } | |
8276 | ||
f0486c68 YZ |
8277 | int btrfs_set_block_group_rw(struct btrfs_root *root, |
8278 | struct btrfs_block_group_cache *cache) | |
5d4f98a2 | 8279 | { |
f0486c68 YZ |
8280 | struct btrfs_space_info *sinfo = cache->space_info; |
8281 | u64 num_bytes; | |
8282 | ||
8283 | BUG_ON(!cache->ro); | |
8284 | ||
8285 | spin_lock(&sinfo->lock); | |
8286 | spin_lock(&cache->lock); | |
8287 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
8288 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
8289 | sinfo->bytes_readonly -= num_bytes; | |
8290 | cache->ro = 0; | |
8291 | spin_unlock(&cache->lock); | |
8292 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
8293 | return 0; |
8294 | } | |
8295 | ||
ba1bf481 JB |
8296 | /* |
8297 | * checks to see if its even possible to relocate this block group. | |
8298 | * | |
8299 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
8300 | * ok to go ahead and try. | |
8301 | */ | |
8302 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 8303 | { |
ba1bf481 JB |
8304 | struct btrfs_block_group_cache *block_group; |
8305 | struct btrfs_space_info *space_info; | |
8306 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
8307 | struct btrfs_device *device; | |
8308 | int full = 0; | |
8309 | int ret = 0; | |
1a40e23b | 8310 | |
ba1bf481 | 8311 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 8312 | |
ba1bf481 JB |
8313 | /* odd, couldn't find the block group, leave it alone */ |
8314 | if (!block_group) | |
8315 | return -1; | |
1a40e23b | 8316 | |
ba1bf481 JB |
8317 | /* no bytes used, we're good */ |
8318 | if (!btrfs_block_group_used(&block_group->item)) | |
1a40e23b ZY |
8319 | goto out; |
8320 | ||
ba1bf481 JB |
8321 | space_info = block_group->space_info; |
8322 | spin_lock(&space_info->lock); | |
17d217fe | 8323 | |
ba1bf481 | 8324 | full = space_info->full; |
17d217fe | 8325 | |
ba1bf481 JB |
8326 | /* |
8327 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
8328 | * relocate it unless we're able to allocate a new chunk below. |
8329 | * | |
8330 | * Otherwise, we need to make sure we have room in the space to handle | |
8331 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 8332 | */ |
7ce618db CM |
8333 | if ((space_info->total_bytes != block_group->key.offset) && |
8334 | (space_info->bytes_used + space_info->bytes_reserved + | |
ba1bf481 JB |
8335 | space_info->bytes_pinned + space_info->bytes_readonly + |
8336 | btrfs_block_group_used(&block_group->item) < | |
7ce618db | 8337 | space_info->total_bytes)) { |
ba1bf481 JB |
8338 | spin_unlock(&space_info->lock); |
8339 | goto out; | |
17d217fe | 8340 | } |
ba1bf481 | 8341 | spin_unlock(&space_info->lock); |
ea8c2819 | 8342 | |
ba1bf481 JB |
8343 | /* |
8344 | * ok we don't have enough space, but maybe we have free space on our | |
8345 | * devices to allocate new chunks for relocation, so loop through our | |
8346 | * alloc devices and guess if we have enough space. However, if we | |
8347 | * were marked as full, then we know there aren't enough chunks, and we | |
8348 | * can just return. | |
8349 | */ | |
8350 | ret = -1; | |
8351 | if (full) | |
8352 | goto out; | |
ea8c2819 | 8353 | |
ba1bf481 JB |
8354 | mutex_lock(&root->fs_info->chunk_mutex); |
8355 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
8356 | u64 min_free = btrfs_block_group_used(&block_group->item); | |
7bfc837d | 8357 | u64 dev_offset; |
56bec294 | 8358 | |
ba1bf481 JB |
8359 | /* |
8360 | * check to make sure we can actually find a chunk with enough | |
8361 | * space to fit our block group in. | |
8362 | */ | |
8363 | if (device->total_bytes > device->bytes_used + min_free) { | |
8364 | ret = find_free_dev_extent(NULL, device, min_free, | |
7bfc837d | 8365 | &dev_offset, NULL); |
ba1bf481 | 8366 | if (!ret) |
73e48b27 | 8367 | break; |
ba1bf481 | 8368 | ret = -1; |
725c8463 | 8369 | } |
edbd8d4e | 8370 | } |
ba1bf481 | 8371 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 8372 | out: |
ba1bf481 | 8373 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
8374 | return ret; |
8375 | } | |
8376 | ||
b2950863 CH |
8377 | static int find_first_block_group(struct btrfs_root *root, |
8378 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 8379 | { |
925baedd | 8380 | int ret = 0; |
0b86a832 CM |
8381 | struct btrfs_key found_key; |
8382 | struct extent_buffer *leaf; | |
8383 | int slot; | |
edbd8d4e | 8384 | |
0b86a832 CM |
8385 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
8386 | if (ret < 0) | |
925baedd CM |
8387 | goto out; |
8388 | ||
d397712b | 8389 | while (1) { |
0b86a832 | 8390 | slot = path->slots[0]; |
edbd8d4e | 8391 | leaf = path->nodes[0]; |
0b86a832 CM |
8392 | if (slot >= btrfs_header_nritems(leaf)) { |
8393 | ret = btrfs_next_leaf(root, path); | |
8394 | if (ret == 0) | |
8395 | continue; | |
8396 | if (ret < 0) | |
925baedd | 8397 | goto out; |
0b86a832 | 8398 | break; |
edbd8d4e | 8399 | } |
0b86a832 | 8400 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 8401 | |
0b86a832 | 8402 | if (found_key.objectid >= key->objectid && |
925baedd CM |
8403 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
8404 | ret = 0; | |
8405 | goto out; | |
8406 | } | |
0b86a832 | 8407 | path->slots[0]++; |
edbd8d4e | 8408 | } |
925baedd | 8409 | out: |
0b86a832 | 8410 | return ret; |
edbd8d4e CM |
8411 | } |
8412 | ||
0af3d00b JB |
8413 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
8414 | { | |
8415 | struct btrfs_block_group_cache *block_group; | |
8416 | u64 last = 0; | |
8417 | ||
8418 | while (1) { | |
8419 | struct inode *inode; | |
8420 | ||
8421 | block_group = btrfs_lookup_first_block_group(info, last); | |
8422 | while (block_group) { | |
8423 | spin_lock(&block_group->lock); | |
8424 | if (block_group->iref) | |
8425 | break; | |
8426 | spin_unlock(&block_group->lock); | |
8427 | block_group = next_block_group(info->tree_root, | |
8428 | block_group); | |
8429 | } | |
8430 | if (!block_group) { | |
8431 | if (last == 0) | |
8432 | break; | |
8433 | last = 0; | |
8434 | continue; | |
8435 | } | |
8436 | ||
8437 | inode = block_group->inode; | |
8438 | block_group->iref = 0; | |
8439 | block_group->inode = NULL; | |
8440 | spin_unlock(&block_group->lock); | |
8441 | iput(inode); | |
8442 | last = block_group->key.objectid + block_group->key.offset; | |
8443 | btrfs_put_block_group(block_group); | |
8444 | } | |
8445 | } | |
8446 | ||
1a40e23b ZY |
8447 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
8448 | { | |
8449 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 8450 | struct btrfs_space_info *space_info; |
11833d66 | 8451 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
8452 | struct rb_node *n; |
8453 | ||
11833d66 YZ |
8454 | down_write(&info->extent_commit_sem); |
8455 | while (!list_empty(&info->caching_block_groups)) { | |
8456 | caching_ctl = list_entry(info->caching_block_groups.next, | |
8457 | struct btrfs_caching_control, list); | |
8458 | list_del(&caching_ctl->list); | |
8459 | put_caching_control(caching_ctl); | |
8460 | } | |
8461 | up_write(&info->extent_commit_sem); | |
8462 | ||
1a40e23b ZY |
8463 | spin_lock(&info->block_group_cache_lock); |
8464 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
8465 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
8466 | cache_node); | |
1a40e23b ZY |
8467 | rb_erase(&block_group->cache_node, |
8468 | &info->block_group_cache_tree); | |
d899e052 YZ |
8469 | spin_unlock(&info->block_group_cache_lock); |
8470 | ||
80eb234a | 8471 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 8472 | list_del(&block_group->list); |
80eb234a | 8473 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 8474 | |
817d52f8 | 8475 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8476 | wait_block_group_cache_done(block_group); |
817d52f8 | 8477 | |
3c14874a JB |
8478 | /* |
8479 | * We haven't cached this block group, which means we could | |
8480 | * possibly have excluded extents on this block group. | |
8481 | */ | |
8482 | if (block_group->cached == BTRFS_CACHE_NO) | |
8483 | free_excluded_extents(info->extent_root, block_group); | |
8484 | ||
817d52f8 | 8485 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 8486 | btrfs_put_block_group(block_group); |
d899e052 YZ |
8487 | |
8488 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
8489 | } |
8490 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
8491 | |
8492 | /* now that all the block groups are freed, go through and | |
8493 | * free all the space_info structs. This is only called during | |
8494 | * the final stages of unmount, and so we know nobody is | |
8495 | * using them. We call synchronize_rcu() once before we start, | |
8496 | * just to be on the safe side. | |
8497 | */ | |
8498 | synchronize_rcu(); | |
8499 | ||
8929ecfa YZ |
8500 | release_global_block_rsv(info); |
8501 | ||
4184ea7f CM |
8502 | while(!list_empty(&info->space_info)) { |
8503 | space_info = list_entry(info->space_info.next, | |
8504 | struct btrfs_space_info, | |
8505 | list); | |
f0486c68 YZ |
8506 | if (space_info->bytes_pinned > 0 || |
8507 | space_info->bytes_reserved > 0) { | |
8508 | WARN_ON(1); | |
8509 | dump_space_info(space_info, 0, 0); | |
8510 | } | |
4184ea7f CM |
8511 | list_del(&space_info->list); |
8512 | kfree(space_info); | |
8513 | } | |
1a40e23b ZY |
8514 | return 0; |
8515 | } | |
8516 | ||
b742bb82 YZ |
8517 | static void __link_block_group(struct btrfs_space_info *space_info, |
8518 | struct btrfs_block_group_cache *cache) | |
8519 | { | |
8520 | int index = get_block_group_index(cache); | |
8521 | ||
8522 | down_write(&space_info->groups_sem); | |
8523 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
8524 | up_write(&space_info->groups_sem); | |
8525 | } | |
8526 | ||
9078a3e1 CM |
8527 | int btrfs_read_block_groups(struct btrfs_root *root) |
8528 | { | |
8529 | struct btrfs_path *path; | |
8530 | int ret; | |
9078a3e1 | 8531 | struct btrfs_block_group_cache *cache; |
be744175 | 8532 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 8533 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
8534 | struct btrfs_key key; |
8535 | struct btrfs_key found_key; | |
5f39d397 | 8536 | struct extent_buffer *leaf; |
0af3d00b JB |
8537 | int need_clear = 0; |
8538 | u64 cache_gen; | |
96b5179d | 8539 | |
be744175 | 8540 | root = info->extent_root; |
9078a3e1 | 8541 | key.objectid = 0; |
0b86a832 | 8542 | key.offset = 0; |
9078a3e1 | 8543 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
8544 | path = btrfs_alloc_path(); |
8545 | if (!path) | |
8546 | return -ENOMEM; | |
8547 | ||
0af3d00b JB |
8548 | cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy); |
8549 | if (cache_gen != 0 && | |
8550 | btrfs_super_generation(&root->fs_info->super_copy) != cache_gen) | |
8551 | need_clear = 1; | |
88c2ba3b JB |
8552 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
8553 | need_clear = 1; | |
8216ef86 JB |
8554 | if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen) |
8555 | printk(KERN_INFO "btrfs: disk space caching is enabled\n"); | |
0af3d00b | 8556 | |
d397712b | 8557 | while (1) { |
0b86a832 | 8558 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
8559 | if (ret > 0) |
8560 | break; | |
0b86a832 CM |
8561 | if (ret != 0) |
8562 | goto error; | |
5f39d397 CM |
8563 | leaf = path->nodes[0]; |
8564 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 8565 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 8566 | if (!cache) { |
0b86a832 | 8567 | ret = -ENOMEM; |
f0486c68 | 8568 | goto error; |
9078a3e1 | 8569 | } |
3e1ad54f | 8570 | |
d2fb3437 | 8571 | atomic_set(&cache->count, 1); |
c286ac48 | 8572 | spin_lock_init(&cache->lock); |
6226cb0a | 8573 | spin_lock_init(&cache->tree_lock); |
817d52f8 | 8574 | cache->fs_info = info; |
0f9dd46c | 8575 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8576 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 8577 | |
0af3d00b JB |
8578 | if (need_clear) |
8579 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
8580 | ||
96303081 JB |
8581 | /* |
8582 | * we only want to have 32k of ram per block group for keeping | |
8583 | * track of free space, and if we pass 1/2 of that we want to | |
8584 | * start converting things over to using bitmaps | |
8585 | */ | |
8586 | cache->extents_thresh = ((1024 * 32) / 2) / | |
8587 | sizeof(struct btrfs_free_space); | |
8588 | ||
5f39d397 CM |
8589 | read_extent_buffer(leaf, &cache->item, |
8590 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
8591 | sizeof(cache->item)); | |
9078a3e1 | 8592 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 8593 | |
9078a3e1 | 8594 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 8595 | btrfs_release_path(path); |
0b86a832 | 8596 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
8597 | cache->sectorsize = root->sectorsize; |
8598 | ||
3c14874a JB |
8599 | /* |
8600 | * We need to exclude the super stripes now so that the space | |
8601 | * info has super bytes accounted for, otherwise we'll think | |
8602 | * we have more space than we actually do. | |
8603 | */ | |
8604 | exclude_super_stripes(root, cache); | |
8605 | ||
817d52f8 JB |
8606 | /* |
8607 | * check for two cases, either we are full, and therefore | |
8608 | * don't need to bother with the caching work since we won't | |
8609 | * find any space, or we are empty, and we can just add all | |
8610 | * the space in and be done with it. This saves us _alot_ of | |
8611 | * time, particularly in the full case. | |
8612 | */ | |
8613 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 8614 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8615 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 8616 | free_excluded_extents(root, cache); |
817d52f8 | 8617 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 8618 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
8619 | cache->cached = BTRFS_CACHE_FINISHED; |
8620 | add_new_free_space(cache, root->fs_info, | |
8621 | found_key.objectid, | |
8622 | found_key.objectid + | |
8623 | found_key.offset); | |
11833d66 | 8624 | free_excluded_extents(root, cache); |
817d52f8 | 8625 | } |
96b5179d | 8626 | |
6324fbf3 CM |
8627 | ret = update_space_info(info, cache->flags, found_key.offset, |
8628 | btrfs_block_group_used(&cache->item), | |
8629 | &space_info); | |
8630 | BUG_ON(ret); | |
8631 | cache->space_info = space_info; | |
1b2da372 | 8632 | spin_lock(&cache->space_info->lock); |
f0486c68 | 8633 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8634 | spin_unlock(&cache->space_info->lock); |
8635 | ||
b742bb82 | 8636 | __link_block_group(space_info, cache); |
0f9dd46c JB |
8637 | |
8638 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
8639 | BUG_ON(ret); | |
75ccf47d CM |
8640 | |
8641 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 8642 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
f0486c68 | 8643 | set_block_group_ro(cache); |
9078a3e1 | 8644 | } |
b742bb82 YZ |
8645 | |
8646 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
8647 | if (!(get_alloc_profile(root, space_info->flags) & | |
8648 | (BTRFS_BLOCK_GROUP_RAID10 | | |
8649 | BTRFS_BLOCK_GROUP_RAID1 | | |
8650 | BTRFS_BLOCK_GROUP_DUP))) | |
8651 | continue; | |
8652 | /* | |
8653 | * avoid allocating from un-mirrored block group if there are | |
8654 | * mirrored block groups. | |
8655 | */ | |
8656 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
f0486c68 | 8657 | set_block_group_ro(cache); |
b742bb82 | 8658 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
f0486c68 | 8659 | set_block_group_ro(cache); |
9078a3e1 | 8660 | } |
f0486c68 YZ |
8661 | |
8662 | init_global_block_rsv(info); | |
0b86a832 CM |
8663 | ret = 0; |
8664 | error: | |
9078a3e1 | 8665 | btrfs_free_path(path); |
0b86a832 | 8666 | return ret; |
9078a3e1 | 8667 | } |
6324fbf3 CM |
8668 | |
8669 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
8670 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 8671 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
8672 | u64 size) |
8673 | { | |
8674 | int ret; | |
6324fbf3 CM |
8675 | struct btrfs_root *extent_root; |
8676 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
8677 | |
8678 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 8679 | |
12fcfd22 | 8680 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 8681 | |
8f18cf13 | 8682 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
8683 | if (!cache) |
8684 | return -ENOMEM; | |
8685 | ||
e17cade2 | 8686 | cache->key.objectid = chunk_offset; |
6324fbf3 | 8687 | cache->key.offset = size; |
d2fb3437 | 8688 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 8689 | cache->sectorsize = root->sectorsize; |
0af3d00b | 8690 | cache->fs_info = root->fs_info; |
96303081 JB |
8691 | |
8692 | /* | |
8693 | * we only want to have 32k of ram per block group for keeping track | |
8694 | * of free space, and if we pass 1/2 of that we want to start | |
8695 | * converting things over to using bitmaps | |
8696 | */ | |
8697 | cache->extents_thresh = ((1024 * 32) / 2) / | |
8698 | sizeof(struct btrfs_free_space); | |
d2fb3437 | 8699 | atomic_set(&cache->count, 1); |
c286ac48 | 8700 | spin_lock_init(&cache->lock); |
6226cb0a | 8701 | spin_lock_init(&cache->tree_lock); |
0f9dd46c | 8702 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8703 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 8704 | |
6324fbf3 | 8705 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
8706 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
8707 | cache->flags = type; | |
8708 | btrfs_set_block_group_flags(&cache->item, type); | |
8709 | ||
11833d66 | 8710 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8711 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 8712 | exclude_super_stripes(root, cache); |
96303081 | 8713 | |
817d52f8 JB |
8714 | add_new_free_space(cache, root->fs_info, chunk_offset, |
8715 | chunk_offset + size); | |
8716 | ||
11833d66 YZ |
8717 | free_excluded_extents(root, cache); |
8718 | ||
6324fbf3 CM |
8719 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
8720 | &cache->space_info); | |
8721 | BUG_ON(ret); | |
1b2da372 JB |
8722 | |
8723 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 8724 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8725 | spin_unlock(&cache->space_info->lock); |
8726 | ||
b742bb82 | 8727 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 8728 | |
0f9dd46c JB |
8729 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
8730 | BUG_ON(ret); | |
c286ac48 | 8731 | |
6324fbf3 CM |
8732 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
8733 | sizeof(cache->item)); | |
8734 | BUG_ON(ret); | |
8735 | ||
d18a2c44 | 8736 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 8737 | |
6324fbf3 CM |
8738 | return 0; |
8739 | } | |
1a40e23b ZY |
8740 | |
8741 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
8742 | struct btrfs_root *root, u64 group_start) | |
8743 | { | |
8744 | struct btrfs_path *path; | |
8745 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 8746 | struct btrfs_free_cluster *cluster; |
0af3d00b | 8747 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 8748 | struct btrfs_key key; |
0af3d00b | 8749 | struct inode *inode; |
1a40e23b | 8750 | int ret; |
89a55897 | 8751 | int factor; |
1a40e23b | 8752 | |
1a40e23b ZY |
8753 | root = root->fs_info->extent_root; |
8754 | ||
8755 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
8756 | BUG_ON(!block_group); | |
c146afad | 8757 | BUG_ON(!block_group->ro); |
1a40e23b | 8758 | |
9f7c43c9 | 8759 | /* |
8760 | * Free the reserved super bytes from this block group before | |
8761 | * remove it. | |
8762 | */ | |
8763 | free_excluded_extents(root, block_group); | |
8764 | ||
1a40e23b | 8765 | memcpy(&key, &block_group->key, sizeof(key)); |
89a55897 JB |
8766 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
8767 | BTRFS_BLOCK_GROUP_RAID1 | | |
8768 | BTRFS_BLOCK_GROUP_RAID10)) | |
8769 | factor = 2; | |
8770 | else | |
8771 | factor = 1; | |
1a40e23b | 8772 | |
44fb5511 CM |
8773 | /* make sure this block group isn't part of an allocation cluster */ |
8774 | cluster = &root->fs_info->data_alloc_cluster; | |
8775 | spin_lock(&cluster->refill_lock); | |
8776 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8777 | spin_unlock(&cluster->refill_lock); | |
8778 | ||
8779 | /* | |
8780 | * make sure this block group isn't part of a metadata | |
8781 | * allocation cluster | |
8782 | */ | |
8783 | cluster = &root->fs_info->meta_alloc_cluster; | |
8784 | spin_lock(&cluster->refill_lock); | |
8785 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8786 | spin_unlock(&cluster->refill_lock); | |
8787 | ||
1a40e23b ZY |
8788 | path = btrfs_alloc_path(); |
8789 | BUG_ON(!path); | |
8790 | ||
0af3d00b JB |
8791 | inode = lookup_free_space_inode(root, block_group, path); |
8792 | if (!IS_ERR(inode)) { | |
8793 | btrfs_orphan_add(trans, inode); | |
8794 | clear_nlink(inode); | |
8795 | /* One for the block groups ref */ | |
8796 | spin_lock(&block_group->lock); | |
8797 | if (block_group->iref) { | |
8798 | block_group->iref = 0; | |
8799 | block_group->inode = NULL; | |
8800 | spin_unlock(&block_group->lock); | |
8801 | iput(inode); | |
8802 | } else { | |
8803 | spin_unlock(&block_group->lock); | |
8804 | } | |
8805 | /* One for our lookup ref */ | |
8806 | iput(inode); | |
8807 | } | |
8808 | ||
8809 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
8810 | key.offset = block_group->key.objectid; | |
8811 | key.type = 0; | |
8812 | ||
8813 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
8814 | if (ret < 0) | |
8815 | goto out; | |
8816 | if (ret > 0) | |
b3b4aa74 | 8817 | btrfs_release_path(path); |
0af3d00b JB |
8818 | if (ret == 0) { |
8819 | ret = btrfs_del_item(trans, tree_root, path); | |
8820 | if (ret) | |
8821 | goto out; | |
b3b4aa74 | 8822 | btrfs_release_path(path); |
0af3d00b JB |
8823 | } |
8824 | ||
3dfdb934 | 8825 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
8826 | rb_erase(&block_group->cache_node, |
8827 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 8828 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 8829 | |
80eb234a | 8830 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
8831 | /* |
8832 | * we must use list_del_init so people can check to see if they | |
8833 | * are still on the list after taking the semaphore | |
8834 | */ | |
8835 | list_del_init(&block_group->list); | |
80eb234a | 8836 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 8837 | |
817d52f8 | 8838 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8839 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
8840 | |
8841 | btrfs_remove_free_space_cache(block_group); | |
8842 | ||
c146afad YZ |
8843 | spin_lock(&block_group->space_info->lock); |
8844 | block_group->space_info->total_bytes -= block_group->key.offset; | |
8845 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 8846 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 8847 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 8848 | |
0af3d00b JB |
8849 | memcpy(&key, &block_group->key, sizeof(key)); |
8850 | ||
283bb197 | 8851 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 8852 | |
fa9c0d79 CM |
8853 | btrfs_put_block_group(block_group); |
8854 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
8855 | |
8856 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
8857 | if (ret > 0) | |
8858 | ret = -EIO; | |
8859 | if (ret < 0) | |
8860 | goto out; | |
8861 | ||
8862 | ret = btrfs_del_item(trans, root, path); | |
8863 | out: | |
8864 | btrfs_free_path(path); | |
8865 | return ret; | |
8866 | } | |
acce952b | 8867 | |
c59021f8 | 8868 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
8869 | { | |
8870 | struct btrfs_space_info *space_info; | |
8871 | int ret; | |
8872 | ||
8873 | ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM, 0, 0, | |
8874 | &space_info); | |
8875 | if (ret) | |
8876 | return ret; | |
8877 | ||
8878 | ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA, 0, 0, | |
8879 | &space_info); | |
8880 | if (ret) | |
8881 | return ret; | |
8882 | ||
8883 | ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA, 0, 0, | |
8884 | &space_info); | |
8885 | if (ret) | |
8886 | return ret; | |
8887 | ||
8888 | return ret; | |
8889 | } | |
8890 | ||
acce952b | 8891 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
8892 | { | |
8893 | return unpin_extent_range(root, start, end); | |
8894 | } | |
8895 | ||
8896 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 8897 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 8898 | { |
5378e607 | 8899 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 8900 | } |
f7039b1d LD |
8901 | |
8902 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
8903 | { | |
8904 | struct btrfs_fs_info *fs_info = root->fs_info; | |
8905 | struct btrfs_block_group_cache *cache = NULL; | |
8906 | u64 group_trimmed; | |
8907 | u64 start; | |
8908 | u64 end; | |
8909 | u64 trimmed = 0; | |
8910 | int ret = 0; | |
8911 | ||
8912 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
8913 | ||
8914 | while (cache) { | |
8915 | if (cache->key.objectid >= (range->start + range->len)) { | |
8916 | btrfs_put_block_group(cache); | |
8917 | break; | |
8918 | } | |
8919 | ||
8920 | start = max(range->start, cache->key.objectid); | |
8921 | end = min(range->start + range->len, | |
8922 | cache->key.objectid + cache->key.offset); | |
8923 | ||
8924 | if (end - start >= range->minlen) { | |
8925 | if (!block_group_cache_done(cache)) { | |
8926 | ret = cache_block_group(cache, NULL, root, 0); | |
8927 | if (!ret) | |
8928 | wait_block_group_cache_done(cache); | |
8929 | } | |
8930 | ret = btrfs_trim_block_group(cache, | |
8931 | &group_trimmed, | |
8932 | start, | |
8933 | end, | |
8934 | range->minlen); | |
8935 | ||
8936 | trimmed += group_trimmed; | |
8937 | if (ret) { | |
8938 | btrfs_put_block_group(cache); | |
8939 | break; | |
8940 | } | |
8941 | } | |
8942 | ||
8943 | cache = next_block_group(fs_info->tree_root, cache); | |
8944 | } | |
8945 | ||
8946 | range->len = trimmed; | |
8947 | return ret; | |
8948 | } |