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 | |
f3465ca4 JB |
36 | static int update_block_group(struct btrfs_trans_handle *trans, |
37 | struct btrfs_root *root, | |
f0486c68 YZ |
38 | u64 bytenr, u64 num_bytes, int alloc); |
39 | static int update_reserved_bytes(struct btrfs_block_group_cache *cache, | |
40 | u64 num_bytes, int reserve, int sinfo); | |
5d4f98a2 YZ |
41 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
42 | struct btrfs_root *root, | |
43 | u64 bytenr, u64 num_bytes, u64 parent, | |
44 | u64 root_objectid, u64 owner_objectid, | |
45 | u64 owner_offset, int refs_to_drop, | |
46 | struct btrfs_delayed_extent_op *extra_op); | |
47 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
48 | struct extent_buffer *leaf, | |
49 | struct btrfs_extent_item *ei); | |
50 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
51 | struct btrfs_root *root, | |
52 | u64 parent, u64 root_objectid, | |
53 | u64 flags, u64 owner, u64 offset, | |
54 | struct btrfs_key *ins, int ref_mod); | |
55 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
56 | struct btrfs_root *root, | |
57 | u64 parent, u64 root_objectid, | |
58 | u64 flags, struct btrfs_disk_key *key, | |
59 | int level, struct btrfs_key *ins); | |
6a63209f JB |
60 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
61 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
62 | u64 flags, int force); | |
11833d66 YZ |
63 | static int find_next_key(struct btrfs_path *path, int level, |
64 | struct btrfs_key *key); | |
9ed74f2d JB |
65 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
66 | int dump_block_groups); | |
6a63209f | 67 | |
817d52f8 JB |
68 | static noinline int |
69 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
70 | { | |
71 | smp_mb(); | |
72 | return cache->cached == BTRFS_CACHE_FINISHED; | |
73 | } | |
74 | ||
0f9dd46c JB |
75 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
76 | { | |
77 | return (cache->flags & bits) == bits; | |
78 | } | |
79 | ||
11dfe35a JB |
80 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
81 | { | |
82 | atomic_inc(&cache->count); | |
83 | } | |
84 | ||
85 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
86 | { | |
f0486c68 YZ |
87 | if (atomic_dec_and_test(&cache->count)) { |
88 | WARN_ON(cache->pinned > 0); | |
89 | WARN_ON(cache->reserved > 0); | |
90 | WARN_ON(cache->reserved_pinned > 0); | |
11dfe35a | 91 | kfree(cache); |
f0486c68 | 92 | } |
11dfe35a JB |
93 | } |
94 | ||
0f9dd46c JB |
95 | /* |
96 | * this adds the block group to the fs_info rb tree for the block group | |
97 | * cache | |
98 | */ | |
b2950863 | 99 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
100 | struct btrfs_block_group_cache *block_group) |
101 | { | |
102 | struct rb_node **p; | |
103 | struct rb_node *parent = NULL; | |
104 | struct btrfs_block_group_cache *cache; | |
105 | ||
106 | spin_lock(&info->block_group_cache_lock); | |
107 | p = &info->block_group_cache_tree.rb_node; | |
108 | ||
109 | while (*p) { | |
110 | parent = *p; | |
111 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
112 | cache_node); | |
113 | if (block_group->key.objectid < cache->key.objectid) { | |
114 | p = &(*p)->rb_left; | |
115 | } else if (block_group->key.objectid > cache->key.objectid) { | |
116 | p = &(*p)->rb_right; | |
117 | } else { | |
118 | spin_unlock(&info->block_group_cache_lock); | |
119 | return -EEXIST; | |
120 | } | |
121 | } | |
122 | ||
123 | rb_link_node(&block_group->cache_node, parent, p); | |
124 | rb_insert_color(&block_group->cache_node, | |
125 | &info->block_group_cache_tree); | |
126 | spin_unlock(&info->block_group_cache_lock); | |
127 | ||
128 | return 0; | |
129 | } | |
130 | ||
131 | /* | |
132 | * This will return the block group at or after bytenr if contains is 0, else | |
133 | * it will return the block group that contains the bytenr | |
134 | */ | |
135 | static struct btrfs_block_group_cache * | |
136 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
137 | int contains) | |
138 | { | |
139 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
140 | struct rb_node *n; | |
141 | u64 end, start; | |
142 | ||
143 | spin_lock(&info->block_group_cache_lock); | |
144 | n = info->block_group_cache_tree.rb_node; | |
145 | ||
146 | while (n) { | |
147 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
148 | cache_node); | |
149 | end = cache->key.objectid + cache->key.offset - 1; | |
150 | start = cache->key.objectid; | |
151 | ||
152 | if (bytenr < start) { | |
153 | if (!contains && (!ret || start < ret->key.objectid)) | |
154 | ret = cache; | |
155 | n = n->rb_left; | |
156 | } else if (bytenr > start) { | |
157 | if (contains && bytenr <= end) { | |
158 | ret = cache; | |
159 | break; | |
160 | } | |
161 | n = n->rb_right; | |
162 | } else { | |
163 | ret = cache; | |
164 | break; | |
165 | } | |
166 | } | |
d2fb3437 | 167 | if (ret) |
11dfe35a | 168 | btrfs_get_block_group(ret); |
0f9dd46c JB |
169 | spin_unlock(&info->block_group_cache_lock); |
170 | ||
171 | return ret; | |
172 | } | |
173 | ||
11833d66 YZ |
174 | static int add_excluded_extent(struct btrfs_root *root, |
175 | u64 start, u64 num_bytes) | |
817d52f8 | 176 | { |
11833d66 YZ |
177 | u64 end = start + num_bytes - 1; |
178 | set_extent_bits(&root->fs_info->freed_extents[0], | |
179 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
180 | set_extent_bits(&root->fs_info->freed_extents[1], | |
181 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
182 | return 0; | |
183 | } | |
817d52f8 | 184 | |
11833d66 YZ |
185 | static void free_excluded_extents(struct btrfs_root *root, |
186 | struct btrfs_block_group_cache *cache) | |
187 | { | |
188 | u64 start, end; | |
817d52f8 | 189 | |
11833d66 YZ |
190 | start = cache->key.objectid; |
191 | end = start + cache->key.offset - 1; | |
192 | ||
193 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
194 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
195 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
196 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
197 | } |
198 | ||
11833d66 YZ |
199 | static int exclude_super_stripes(struct btrfs_root *root, |
200 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 201 | { |
817d52f8 JB |
202 | u64 bytenr; |
203 | u64 *logical; | |
204 | int stripe_len; | |
205 | int i, nr, ret; | |
206 | ||
06b2331f YZ |
207 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
208 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
209 | cache->bytes_super += stripe_len; | |
210 | ret = add_excluded_extent(root, cache->key.objectid, | |
211 | stripe_len); | |
212 | BUG_ON(ret); | |
213 | } | |
214 | ||
817d52f8 JB |
215 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
216 | bytenr = btrfs_sb_offset(i); | |
217 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
218 | cache->key.objectid, bytenr, | |
219 | 0, &logical, &nr, &stripe_len); | |
220 | BUG_ON(ret); | |
11833d66 | 221 | |
817d52f8 | 222 | while (nr--) { |
1b2da372 | 223 | cache->bytes_super += stripe_len; |
11833d66 YZ |
224 | ret = add_excluded_extent(root, logical[nr], |
225 | stripe_len); | |
226 | BUG_ON(ret); | |
817d52f8 | 227 | } |
11833d66 | 228 | |
817d52f8 JB |
229 | kfree(logical); |
230 | } | |
817d52f8 JB |
231 | return 0; |
232 | } | |
233 | ||
11833d66 YZ |
234 | static struct btrfs_caching_control * |
235 | get_caching_control(struct btrfs_block_group_cache *cache) | |
236 | { | |
237 | struct btrfs_caching_control *ctl; | |
238 | ||
239 | spin_lock(&cache->lock); | |
240 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
241 | spin_unlock(&cache->lock); | |
dde5abee JB |
242 | return NULL; |
243 | } | |
244 | ||
245 | /* We're loading it the fast way, so we don't have a caching_ctl. */ | |
246 | if (!cache->caching_ctl) { | |
247 | spin_unlock(&cache->lock); | |
11833d66 YZ |
248 | return NULL; |
249 | } | |
250 | ||
251 | ctl = cache->caching_ctl; | |
252 | atomic_inc(&ctl->count); | |
253 | spin_unlock(&cache->lock); | |
254 | return ctl; | |
255 | } | |
256 | ||
257 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
258 | { | |
259 | if (atomic_dec_and_test(&ctl->count)) | |
260 | kfree(ctl); | |
261 | } | |
262 | ||
0f9dd46c JB |
263 | /* |
264 | * this is only called by cache_block_group, since we could have freed extents | |
265 | * we need to check the pinned_extents for any extents that can't be used yet | |
266 | * since their free space will be released as soon as the transaction commits. | |
267 | */ | |
817d52f8 | 268 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
269 | struct btrfs_fs_info *info, u64 start, u64 end) |
270 | { | |
817d52f8 | 271 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
272 | int ret; |
273 | ||
274 | while (start < end) { | |
11833d66 | 275 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 276 | &extent_start, &extent_end, |
11833d66 | 277 | EXTENT_DIRTY | EXTENT_UPTODATE); |
0f9dd46c JB |
278 | if (ret) |
279 | break; | |
280 | ||
06b2331f | 281 | if (extent_start <= start) { |
0f9dd46c JB |
282 | start = extent_end + 1; |
283 | } else if (extent_start > start && extent_start < end) { | |
284 | size = extent_start - start; | |
817d52f8 | 285 | total_added += size; |
ea6a478e JB |
286 | ret = btrfs_add_free_space(block_group, start, |
287 | size); | |
0f9dd46c JB |
288 | BUG_ON(ret); |
289 | start = extent_end + 1; | |
290 | } else { | |
291 | break; | |
292 | } | |
293 | } | |
294 | ||
295 | if (start < end) { | |
296 | size = end - start; | |
817d52f8 | 297 | total_added += size; |
ea6a478e | 298 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
299 | BUG_ON(ret); |
300 | } | |
301 | ||
817d52f8 | 302 | return total_added; |
0f9dd46c JB |
303 | } |
304 | ||
817d52f8 | 305 | static int caching_kthread(void *data) |
e37c9e69 | 306 | { |
817d52f8 JB |
307 | struct btrfs_block_group_cache *block_group = data; |
308 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
11833d66 YZ |
309 | struct btrfs_caching_control *caching_ctl = block_group->caching_ctl; |
310 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 311 | struct btrfs_path *path; |
5f39d397 | 312 | struct extent_buffer *leaf; |
11833d66 | 313 | struct btrfs_key key; |
817d52f8 | 314 | u64 total_found = 0; |
11833d66 YZ |
315 | u64 last = 0; |
316 | u32 nritems; | |
317 | int ret = 0; | |
f510cfec | 318 | |
e37c9e69 CM |
319 | path = btrfs_alloc_path(); |
320 | if (!path) | |
321 | return -ENOMEM; | |
7d7d6068 | 322 | |
11833d66 | 323 | exclude_super_stripes(extent_root, block_group); |
1b2da372 | 324 | spin_lock(&block_group->space_info->lock); |
f0486c68 | 325 | block_group->space_info->bytes_readonly += block_group->bytes_super; |
1b2da372 | 326 | spin_unlock(&block_group->space_info->lock); |
11833d66 | 327 | |
817d52f8 | 328 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 329 | |
5cd57b2c | 330 | /* |
817d52f8 JB |
331 | * We don't want to deadlock with somebody trying to allocate a new |
332 | * extent for the extent root while also trying to search the extent | |
333 | * root to add free space. So we skip locking and search the commit | |
334 | * root, since its read-only | |
5cd57b2c CM |
335 | */ |
336 | path->skip_locking = 1; | |
817d52f8 JB |
337 | path->search_commit_root = 1; |
338 | path->reada = 2; | |
339 | ||
e4404d6e | 340 | key.objectid = last; |
e37c9e69 | 341 | key.offset = 0; |
11833d66 | 342 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 343 | again: |
11833d66 | 344 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
345 | /* need to make sure the commit_root doesn't disappear */ |
346 | down_read(&fs_info->extent_commit_sem); | |
347 | ||
11833d66 | 348 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 349 | if (ret < 0) |
ef8bbdfe | 350 | goto err; |
a512bbf8 | 351 | |
11833d66 YZ |
352 | leaf = path->nodes[0]; |
353 | nritems = btrfs_header_nritems(leaf); | |
354 | ||
d397712b | 355 | while (1) { |
817d52f8 | 356 | smp_mb(); |
11833d66 | 357 | if (fs_info->closing > 1) { |
f25784b3 | 358 | last = (u64)-1; |
817d52f8 | 359 | break; |
f25784b3 | 360 | } |
817d52f8 | 361 | |
11833d66 YZ |
362 | if (path->slots[0] < nritems) { |
363 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
364 | } else { | |
365 | ret = find_next_key(path, 0, &key); | |
366 | if (ret) | |
e37c9e69 | 367 | break; |
817d52f8 | 368 | |
11833d66 YZ |
369 | caching_ctl->progress = last; |
370 | btrfs_release_path(extent_root, path); | |
371 | up_read(&fs_info->extent_commit_sem); | |
372 | mutex_unlock(&caching_ctl->mutex); | |
373 | if (btrfs_transaction_in_commit(fs_info)) | |
f36f3042 | 374 | schedule_timeout(1); |
11833d66 YZ |
375 | else |
376 | cond_resched(); | |
377 | goto again; | |
378 | } | |
817d52f8 | 379 | |
11833d66 YZ |
380 | if (key.objectid < block_group->key.objectid) { |
381 | path->slots[0]++; | |
817d52f8 | 382 | continue; |
e37c9e69 | 383 | } |
0f9dd46c | 384 | |
e37c9e69 | 385 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 386 | block_group->key.offset) |
e37c9e69 | 387 | break; |
7d7d6068 | 388 | |
11833d66 | 389 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
390 | total_found += add_new_free_space(block_group, |
391 | fs_info, last, | |
392 | key.objectid); | |
7d7d6068 | 393 | last = key.objectid + key.offset; |
817d52f8 | 394 | |
11833d66 YZ |
395 | if (total_found > (1024 * 1024 * 2)) { |
396 | total_found = 0; | |
397 | wake_up(&caching_ctl->wait); | |
398 | } | |
817d52f8 | 399 | } |
e37c9e69 CM |
400 | path->slots[0]++; |
401 | } | |
817d52f8 | 402 | ret = 0; |
e37c9e69 | 403 | |
817d52f8 JB |
404 | total_found += add_new_free_space(block_group, fs_info, last, |
405 | block_group->key.objectid + | |
406 | block_group->key.offset); | |
11833d66 | 407 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
408 | |
409 | spin_lock(&block_group->lock); | |
11833d66 | 410 | block_group->caching_ctl = NULL; |
817d52f8 JB |
411 | block_group->cached = BTRFS_CACHE_FINISHED; |
412 | spin_unlock(&block_group->lock); | |
0f9dd46c | 413 | |
54aa1f4d | 414 | err: |
e37c9e69 | 415 | btrfs_free_path(path); |
276e680d | 416 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 417 | |
11833d66 YZ |
418 | free_excluded_extents(extent_root, block_group); |
419 | ||
420 | mutex_unlock(&caching_ctl->mutex); | |
421 | wake_up(&caching_ctl->wait); | |
422 | ||
423 | put_caching_control(caching_ctl); | |
424 | atomic_dec(&block_group->space_info->caching_threads); | |
11dfe35a JB |
425 | btrfs_put_block_group(block_group); |
426 | ||
817d52f8 JB |
427 | return 0; |
428 | } | |
429 | ||
9d66e233 JB |
430 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
431 | struct btrfs_trans_handle *trans, | |
b8399dee | 432 | struct btrfs_root *root, |
9d66e233 | 433 | int load_cache_only) |
817d52f8 | 434 | { |
11833d66 YZ |
435 | struct btrfs_fs_info *fs_info = cache->fs_info; |
436 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
437 | struct task_struct *tsk; |
438 | int ret = 0; | |
439 | ||
11833d66 YZ |
440 | smp_mb(); |
441 | if (cache->cached != BTRFS_CACHE_NO) | |
442 | return 0; | |
443 | ||
9d66e233 JB |
444 | /* |
445 | * We can't do the read from on-disk cache during a commit since we need | |
b8399dee JB |
446 | * to have the normal tree locking. Also if we are currently trying to |
447 | * allocate blocks for the tree root we can't do the fast caching since | |
448 | * we likely hold important locks. | |
9d66e233 | 449 | */ |
b8399dee JB |
450 | if (!trans->transaction->in_commit && |
451 | (root && root != root->fs_info->tree_root)) { | |
9d66e233 JB |
452 | spin_lock(&cache->lock); |
453 | if (cache->cached != BTRFS_CACHE_NO) { | |
454 | spin_unlock(&cache->lock); | |
455 | return 0; | |
456 | } | |
457 | cache->cached = BTRFS_CACHE_STARTED; | |
458 | spin_unlock(&cache->lock); | |
459 | ||
460 | ret = load_free_space_cache(fs_info, cache); | |
461 | ||
462 | spin_lock(&cache->lock); | |
463 | if (ret == 1) { | |
464 | cache->cached = BTRFS_CACHE_FINISHED; | |
465 | cache->last_byte_to_unpin = (u64)-1; | |
466 | } else { | |
467 | cache->cached = BTRFS_CACHE_NO; | |
468 | } | |
469 | spin_unlock(&cache->lock); | |
470 | if (ret == 1) | |
471 | return 0; | |
472 | } | |
473 | ||
474 | if (load_cache_only) | |
475 | return 0; | |
476 | ||
11833d66 YZ |
477 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL); |
478 | BUG_ON(!caching_ctl); | |
479 | ||
480 | INIT_LIST_HEAD(&caching_ctl->list); | |
481 | mutex_init(&caching_ctl->mutex); | |
482 | init_waitqueue_head(&caching_ctl->wait); | |
483 | caching_ctl->block_group = cache; | |
484 | caching_ctl->progress = cache->key.objectid; | |
485 | /* one for caching kthread, one for caching block group list */ | |
486 | atomic_set(&caching_ctl->count, 2); | |
487 | ||
817d52f8 JB |
488 | spin_lock(&cache->lock); |
489 | if (cache->cached != BTRFS_CACHE_NO) { | |
490 | spin_unlock(&cache->lock); | |
11833d66 YZ |
491 | kfree(caching_ctl); |
492 | return 0; | |
817d52f8 | 493 | } |
11833d66 | 494 | cache->caching_ctl = caching_ctl; |
817d52f8 JB |
495 | cache->cached = BTRFS_CACHE_STARTED; |
496 | spin_unlock(&cache->lock); | |
497 | ||
11833d66 YZ |
498 | down_write(&fs_info->extent_commit_sem); |
499 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
500 | up_write(&fs_info->extent_commit_sem); | |
501 | ||
502 | atomic_inc(&cache->space_info->caching_threads); | |
11dfe35a | 503 | btrfs_get_block_group(cache); |
11833d66 | 504 | |
817d52f8 JB |
505 | tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n", |
506 | cache->key.objectid); | |
507 | if (IS_ERR(tsk)) { | |
508 | ret = PTR_ERR(tsk); | |
509 | printk(KERN_ERR "error running thread %d\n", ret); | |
510 | BUG(); | |
511 | } | |
512 | ||
ef8bbdfe | 513 | return ret; |
e37c9e69 CM |
514 | } |
515 | ||
0f9dd46c JB |
516 | /* |
517 | * return the block group that starts at or after bytenr | |
518 | */ | |
d397712b CM |
519 | static struct btrfs_block_group_cache * |
520 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 521 | { |
0f9dd46c | 522 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 523 | |
0f9dd46c | 524 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 525 | |
0f9dd46c | 526 | return cache; |
0ef3e66b CM |
527 | } |
528 | ||
0f9dd46c | 529 | /* |
9f55684c | 530 | * return the block group that contains the given bytenr |
0f9dd46c | 531 | */ |
d397712b CM |
532 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
533 | struct btrfs_fs_info *info, | |
534 | u64 bytenr) | |
be744175 | 535 | { |
0f9dd46c | 536 | struct btrfs_block_group_cache *cache; |
be744175 | 537 | |
0f9dd46c | 538 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 539 | |
0f9dd46c | 540 | return cache; |
be744175 | 541 | } |
0b86a832 | 542 | |
0f9dd46c JB |
543 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
544 | u64 flags) | |
6324fbf3 | 545 | { |
0f9dd46c | 546 | struct list_head *head = &info->space_info; |
0f9dd46c | 547 | struct btrfs_space_info *found; |
4184ea7f | 548 | |
b742bb82 YZ |
549 | flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM | |
550 | BTRFS_BLOCK_GROUP_METADATA; | |
551 | ||
4184ea7f CM |
552 | rcu_read_lock(); |
553 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 554 | if (found->flags & flags) { |
4184ea7f | 555 | rcu_read_unlock(); |
0f9dd46c | 556 | return found; |
4184ea7f | 557 | } |
0f9dd46c | 558 | } |
4184ea7f | 559 | rcu_read_unlock(); |
0f9dd46c | 560 | return NULL; |
6324fbf3 CM |
561 | } |
562 | ||
4184ea7f CM |
563 | /* |
564 | * after adding space to the filesystem, we need to clear the full flags | |
565 | * on all the space infos. | |
566 | */ | |
567 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
568 | { | |
569 | struct list_head *head = &info->space_info; | |
570 | struct btrfs_space_info *found; | |
571 | ||
572 | rcu_read_lock(); | |
573 | list_for_each_entry_rcu(found, head, list) | |
574 | found->full = 0; | |
575 | rcu_read_unlock(); | |
576 | } | |
577 | ||
80eb234a JB |
578 | static u64 div_factor(u64 num, int factor) |
579 | { | |
580 | if (factor == 10) | |
581 | return num; | |
582 | num *= factor; | |
583 | do_div(num, 10); | |
584 | return num; | |
585 | } | |
586 | ||
e5bc2458 CM |
587 | static u64 div_factor_fine(u64 num, int factor) |
588 | { | |
589 | if (factor == 100) | |
590 | return num; | |
591 | num *= factor; | |
592 | do_div(num, 100); | |
593 | return num; | |
594 | } | |
595 | ||
d2fb3437 YZ |
596 | u64 btrfs_find_block_group(struct btrfs_root *root, |
597 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 598 | { |
96b5179d | 599 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 600 | u64 used; |
d2fb3437 YZ |
601 | u64 last = max(search_hint, search_start); |
602 | u64 group_start = 0; | |
31f3c99b | 603 | int full_search = 0; |
d2fb3437 | 604 | int factor = 9; |
0ef3e66b | 605 | int wrapped = 0; |
31f3c99b | 606 | again: |
e8569813 ZY |
607 | while (1) { |
608 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
609 | if (!cache) |
610 | break; | |
96b5179d | 611 | |
c286ac48 | 612 | spin_lock(&cache->lock); |
96b5179d CM |
613 | last = cache->key.objectid + cache->key.offset; |
614 | used = btrfs_block_group_used(&cache->item); | |
615 | ||
d2fb3437 YZ |
616 | if ((full_search || !cache->ro) && |
617 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 618 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
619 | div_factor(cache->key.offset, factor)) { |
620 | group_start = cache->key.objectid; | |
c286ac48 | 621 | spin_unlock(&cache->lock); |
fa9c0d79 | 622 | btrfs_put_block_group(cache); |
8790d502 CM |
623 | goto found; |
624 | } | |
6324fbf3 | 625 | } |
c286ac48 | 626 | spin_unlock(&cache->lock); |
fa9c0d79 | 627 | btrfs_put_block_group(cache); |
de428b63 | 628 | cond_resched(); |
cd1bc465 | 629 | } |
0ef3e66b CM |
630 | if (!wrapped) { |
631 | last = search_start; | |
632 | wrapped = 1; | |
633 | goto again; | |
634 | } | |
635 | if (!full_search && factor < 10) { | |
be744175 | 636 | last = search_start; |
31f3c99b | 637 | full_search = 1; |
0ef3e66b | 638 | factor = 10; |
31f3c99b CM |
639 | goto again; |
640 | } | |
be744175 | 641 | found: |
d2fb3437 | 642 | return group_start; |
925baedd | 643 | } |
0f9dd46c | 644 | |
e02119d5 | 645 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 646 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
647 | { |
648 | int ret; | |
649 | struct btrfs_key key; | |
31840ae1 | 650 | struct btrfs_path *path; |
e02119d5 | 651 | |
31840ae1 ZY |
652 | path = btrfs_alloc_path(); |
653 | BUG_ON(!path); | |
e02119d5 CM |
654 | key.objectid = start; |
655 | key.offset = len; | |
656 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
657 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
658 | 0, 0); | |
31840ae1 | 659 | btrfs_free_path(path); |
7bb86316 CM |
660 | return ret; |
661 | } | |
662 | ||
a22285a6 YZ |
663 | /* |
664 | * helper function to lookup reference count and flags of extent. | |
665 | * | |
666 | * the head node for delayed ref is used to store the sum of all the | |
667 | * reference count modifications queued up in the rbtree. the head | |
668 | * node may also store the extent flags to set. This way you can check | |
669 | * to see what the reference count and extent flags would be if all of | |
670 | * the delayed refs are not processed. | |
671 | */ | |
672 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
673 | struct btrfs_root *root, u64 bytenr, | |
674 | u64 num_bytes, u64 *refs, u64 *flags) | |
675 | { | |
676 | struct btrfs_delayed_ref_head *head; | |
677 | struct btrfs_delayed_ref_root *delayed_refs; | |
678 | struct btrfs_path *path; | |
679 | struct btrfs_extent_item *ei; | |
680 | struct extent_buffer *leaf; | |
681 | struct btrfs_key key; | |
682 | u32 item_size; | |
683 | u64 num_refs; | |
684 | u64 extent_flags; | |
685 | int ret; | |
686 | ||
687 | path = btrfs_alloc_path(); | |
688 | if (!path) | |
689 | return -ENOMEM; | |
690 | ||
691 | key.objectid = bytenr; | |
692 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
693 | key.offset = num_bytes; | |
694 | if (!trans) { | |
695 | path->skip_locking = 1; | |
696 | path->search_commit_root = 1; | |
697 | } | |
698 | again: | |
699 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
700 | &key, path, 0, 0); | |
701 | if (ret < 0) | |
702 | goto out_free; | |
703 | ||
704 | if (ret == 0) { | |
705 | leaf = path->nodes[0]; | |
706 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
707 | if (item_size >= sizeof(*ei)) { | |
708 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
709 | struct btrfs_extent_item); | |
710 | num_refs = btrfs_extent_refs(leaf, ei); | |
711 | extent_flags = btrfs_extent_flags(leaf, ei); | |
712 | } else { | |
713 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
714 | struct btrfs_extent_item_v0 *ei0; | |
715 | BUG_ON(item_size != sizeof(*ei0)); | |
716 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
717 | struct btrfs_extent_item_v0); | |
718 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
719 | /* FIXME: this isn't correct for data */ | |
720 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
721 | #else | |
722 | BUG(); | |
723 | #endif | |
724 | } | |
725 | BUG_ON(num_refs == 0); | |
726 | } else { | |
727 | num_refs = 0; | |
728 | extent_flags = 0; | |
729 | ret = 0; | |
730 | } | |
731 | ||
732 | if (!trans) | |
733 | goto out; | |
734 | ||
735 | delayed_refs = &trans->transaction->delayed_refs; | |
736 | spin_lock(&delayed_refs->lock); | |
737 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
738 | if (head) { | |
739 | if (!mutex_trylock(&head->mutex)) { | |
740 | atomic_inc(&head->node.refs); | |
741 | spin_unlock(&delayed_refs->lock); | |
742 | ||
743 | btrfs_release_path(root->fs_info->extent_root, path); | |
744 | ||
745 | mutex_lock(&head->mutex); | |
746 | mutex_unlock(&head->mutex); | |
747 | btrfs_put_delayed_ref(&head->node); | |
748 | goto again; | |
749 | } | |
750 | if (head->extent_op && head->extent_op->update_flags) | |
751 | extent_flags |= head->extent_op->flags_to_set; | |
752 | else | |
753 | BUG_ON(num_refs == 0); | |
754 | ||
755 | num_refs += head->node.ref_mod; | |
756 | mutex_unlock(&head->mutex); | |
757 | } | |
758 | spin_unlock(&delayed_refs->lock); | |
759 | out: | |
760 | WARN_ON(num_refs == 0); | |
761 | if (refs) | |
762 | *refs = num_refs; | |
763 | if (flags) | |
764 | *flags = extent_flags; | |
765 | out_free: | |
766 | btrfs_free_path(path); | |
767 | return ret; | |
768 | } | |
769 | ||
d8d5f3e1 CM |
770 | /* |
771 | * Back reference rules. Back refs have three main goals: | |
772 | * | |
773 | * 1) differentiate between all holders of references to an extent so that | |
774 | * when a reference is dropped we can make sure it was a valid reference | |
775 | * before freeing the extent. | |
776 | * | |
777 | * 2) Provide enough information to quickly find the holders of an extent | |
778 | * if we notice a given block is corrupted or bad. | |
779 | * | |
780 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
781 | * maintenance. This is actually the same as #2, but with a slightly | |
782 | * different use case. | |
783 | * | |
5d4f98a2 YZ |
784 | * There are two kinds of back refs. The implicit back refs is optimized |
785 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
786 | * back refs of this kind provide information about the block's owner tree | |
787 | * and the pointer's key. These information allow us to find the block by | |
788 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
789 | * referenced by their owner trees. The location of tree block is recorded | |
790 | * in the back refs. Actually the full back refs is generic, and can be | |
791 | * used in all cases the implicit back refs is used. The major shortcoming | |
792 | * of the full back refs is its overhead. Every time a tree block gets | |
793 | * COWed, we have to update back refs entry for all pointers in it. | |
794 | * | |
795 | * For a newly allocated tree block, we use implicit back refs for | |
796 | * pointers in it. This means most tree related operations only involve | |
797 | * implicit back refs. For a tree block created in old transaction, the | |
798 | * only way to drop a reference to it is COW it. So we can detect the | |
799 | * event that tree block loses its owner tree's reference and do the | |
800 | * back refs conversion. | |
801 | * | |
802 | * When a tree block is COW'd through a tree, there are four cases: | |
803 | * | |
804 | * The reference count of the block is one and the tree is the block's | |
805 | * owner tree. Nothing to do in this case. | |
806 | * | |
807 | * The reference count of the block is one and the tree is not the | |
808 | * block's owner tree. In this case, full back refs is used for pointers | |
809 | * in the block. Remove these full back refs, add implicit back refs for | |
810 | * every pointers in the new block. | |
811 | * | |
812 | * The reference count of the block is greater than one and the tree is | |
813 | * the block's owner tree. In this case, implicit back refs is used for | |
814 | * pointers in the block. Add full back refs for every pointers in the | |
815 | * block, increase lower level extents' reference counts. The original | |
816 | * implicit back refs are entailed to the new block. | |
817 | * | |
818 | * The reference count of the block is greater than one and the tree is | |
819 | * not the block's owner tree. Add implicit back refs for every pointer in | |
820 | * the new block, increase lower level extents' reference count. | |
821 | * | |
822 | * Back Reference Key composing: | |
823 | * | |
824 | * The key objectid corresponds to the first byte in the extent, | |
825 | * The key type is used to differentiate between types of back refs. | |
826 | * There are different meanings of the key offset for different types | |
827 | * of back refs. | |
828 | * | |
d8d5f3e1 CM |
829 | * File extents can be referenced by: |
830 | * | |
831 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 832 | * - different files inside a single subvolume |
d8d5f3e1 CM |
833 | * - different offsets inside a file (bookend extents in file.c) |
834 | * | |
5d4f98a2 | 835 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
836 | * |
837 | * - Objectid of the subvolume root | |
d8d5f3e1 | 838 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
839 | * - original offset in the file |
840 | * - how many bookend extents | |
d8d5f3e1 | 841 | * |
5d4f98a2 YZ |
842 | * The key offset for the implicit back refs is hash of the first |
843 | * three fields. | |
d8d5f3e1 | 844 | * |
5d4f98a2 | 845 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 846 | * |
5d4f98a2 | 847 | * - number of pointers in the tree leaf |
d8d5f3e1 | 848 | * |
5d4f98a2 YZ |
849 | * The key offset for the implicit back refs is the first byte of |
850 | * the tree leaf | |
d8d5f3e1 | 851 | * |
5d4f98a2 YZ |
852 | * When a file extent is allocated, The implicit back refs is used. |
853 | * the fields are filled in: | |
d8d5f3e1 | 854 | * |
5d4f98a2 | 855 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 856 | * |
5d4f98a2 YZ |
857 | * When a file extent is removed file truncation, we find the |
858 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 859 | * |
5d4f98a2 | 860 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 861 | * |
5d4f98a2 | 862 | * Btree extents can be referenced by: |
d8d5f3e1 | 863 | * |
5d4f98a2 | 864 | * - Different subvolumes |
d8d5f3e1 | 865 | * |
5d4f98a2 YZ |
866 | * Both the implicit back refs and the full back refs for tree blocks |
867 | * only consist of key. The key offset for the implicit back refs is | |
868 | * objectid of block's owner tree. The key offset for the full back refs | |
869 | * is the first byte of parent block. | |
d8d5f3e1 | 870 | * |
5d4f98a2 YZ |
871 | * When implicit back refs is used, information about the lowest key and |
872 | * level of the tree block are required. These information are stored in | |
873 | * tree block info structure. | |
d8d5f3e1 | 874 | */ |
31840ae1 | 875 | |
5d4f98a2 YZ |
876 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
877 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
878 | struct btrfs_root *root, | |
879 | struct btrfs_path *path, | |
880 | u64 owner, u32 extra_size) | |
7bb86316 | 881 | { |
5d4f98a2 YZ |
882 | struct btrfs_extent_item *item; |
883 | struct btrfs_extent_item_v0 *ei0; | |
884 | struct btrfs_extent_ref_v0 *ref0; | |
885 | struct btrfs_tree_block_info *bi; | |
886 | struct extent_buffer *leaf; | |
7bb86316 | 887 | struct btrfs_key key; |
5d4f98a2 YZ |
888 | struct btrfs_key found_key; |
889 | u32 new_size = sizeof(*item); | |
890 | u64 refs; | |
891 | int ret; | |
892 | ||
893 | leaf = path->nodes[0]; | |
894 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
895 | ||
896 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
897 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
898 | struct btrfs_extent_item_v0); | |
899 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
900 | ||
901 | if (owner == (u64)-1) { | |
902 | while (1) { | |
903 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
904 | ret = btrfs_next_leaf(root, path); | |
905 | if (ret < 0) | |
906 | return ret; | |
907 | BUG_ON(ret > 0); | |
908 | leaf = path->nodes[0]; | |
909 | } | |
910 | btrfs_item_key_to_cpu(leaf, &found_key, | |
911 | path->slots[0]); | |
912 | BUG_ON(key.objectid != found_key.objectid); | |
913 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
914 | path->slots[0]++; | |
915 | continue; | |
916 | } | |
917 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
918 | struct btrfs_extent_ref_v0); | |
919 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
920 | break; | |
921 | } | |
922 | } | |
923 | btrfs_release_path(root, path); | |
924 | ||
925 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
926 | new_size += sizeof(*bi); | |
927 | ||
928 | new_size -= sizeof(*ei0); | |
929 | ret = btrfs_search_slot(trans, root, &key, path, | |
930 | new_size + extra_size, 1); | |
931 | if (ret < 0) | |
932 | return ret; | |
933 | BUG_ON(ret); | |
934 | ||
935 | ret = btrfs_extend_item(trans, root, path, new_size); | |
936 | BUG_ON(ret); | |
937 | ||
938 | leaf = path->nodes[0]; | |
939 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
940 | btrfs_set_extent_refs(leaf, item, refs); | |
941 | /* FIXME: get real generation */ | |
942 | btrfs_set_extent_generation(leaf, item, 0); | |
943 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
944 | btrfs_set_extent_flags(leaf, item, | |
945 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
946 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
947 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
948 | /* FIXME: get first key of the block */ | |
949 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
950 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
951 | } else { | |
952 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
953 | } | |
954 | btrfs_mark_buffer_dirty(leaf); | |
955 | return 0; | |
956 | } | |
957 | #endif | |
958 | ||
959 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
960 | { | |
961 | u32 high_crc = ~(u32)0; | |
962 | u32 low_crc = ~(u32)0; | |
963 | __le64 lenum; | |
964 | ||
965 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 966 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 967 | lenum = cpu_to_le64(owner); |
163e783e | 968 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 969 | lenum = cpu_to_le64(offset); |
163e783e | 970 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
971 | |
972 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
973 | } | |
974 | ||
975 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
976 | struct btrfs_extent_data_ref *ref) | |
977 | { | |
978 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
979 | btrfs_extent_data_ref_objectid(leaf, ref), | |
980 | btrfs_extent_data_ref_offset(leaf, ref)); | |
981 | } | |
982 | ||
983 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
984 | struct btrfs_extent_data_ref *ref, | |
985 | u64 root_objectid, u64 owner, u64 offset) | |
986 | { | |
987 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
988 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
989 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
990 | return 0; | |
991 | return 1; | |
992 | } | |
993 | ||
994 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
995 | struct btrfs_root *root, | |
996 | struct btrfs_path *path, | |
997 | u64 bytenr, u64 parent, | |
998 | u64 root_objectid, | |
999 | u64 owner, u64 offset) | |
1000 | { | |
1001 | struct btrfs_key key; | |
1002 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1003 | struct extent_buffer *leaf; |
5d4f98a2 | 1004 | u32 nritems; |
74493f7a | 1005 | int ret; |
5d4f98a2 YZ |
1006 | int recow; |
1007 | int err = -ENOENT; | |
74493f7a | 1008 | |
31840ae1 | 1009 | key.objectid = bytenr; |
5d4f98a2 YZ |
1010 | if (parent) { |
1011 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1012 | key.offset = parent; | |
1013 | } else { | |
1014 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1015 | key.offset = hash_extent_data_ref(root_objectid, | |
1016 | owner, offset); | |
1017 | } | |
1018 | again: | |
1019 | recow = 0; | |
1020 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1021 | if (ret < 0) { | |
1022 | err = ret; | |
1023 | goto fail; | |
1024 | } | |
31840ae1 | 1025 | |
5d4f98a2 YZ |
1026 | if (parent) { |
1027 | if (!ret) | |
1028 | return 0; | |
1029 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1030 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
1031 | btrfs_release_path(root, path); | |
1032 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1033 | if (ret < 0) { | |
1034 | err = ret; | |
1035 | goto fail; | |
1036 | } | |
1037 | if (!ret) | |
1038 | return 0; | |
1039 | #endif | |
1040 | goto fail; | |
31840ae1 ZY |
1041 | } |
1042 | ||
1043 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1044 | nritems = btrfs_header_nritems(leaf); |
1045 | while (1) { | |
1046 | if (path->slots[0] >= nritems) { | |
1047 | ret = btrfs_next_leaf(root, path); | |
1048 | if (ret < 0) | |
1049 | err = ret; | |
1050 | if (ret) | |
1051 | goto fail; | |
1052 | ||
1053 | leaf = path->nodes[0]; | |
1054 | nritems = btrfs_header_nritems(leaf); | |
1055 | recow = 1; | |
1056 | } | |
1057 | ||
1058 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1059 | if (key.objectid != bytenr || | |
1060 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1061 | goto fail; | |
1062 | ||
1063 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1064 | struct btrfs_extent_data_ref); | |
1065 | ||
1066 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1067 | owner, offset)) { | |
1068 | if (recow) { | |
1069 | btrfs_release_path(root, path); | |
1070 | goto again; | |
1071 | } | |
1072 | err = 0; | |
1073 | break; | |
1074 | } | |
1075 | path->slots[0]++; | |
31840ae1 | 1076 | } |
5d4f98a2 YZ |
1077 | fail: |
1078 | return err; | |
31840ae1 ZY |
1079 | } |
1080 | ||
5d4f98a2 YZ |
1081 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1082 | struct btrfs_root *root, | |
1083 | struct btrfs_path *path, | |
1084 | u64 bytenr, u64 parent, | |
1085 | u64 root_objectid, u64 owner, | |
1086 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1087 | { |
1088 | struct btrfs_key key; | |
1089 | struct extent_buffer *leaf; | |
5d4f98a2 | 1090 | u32 size; |
31840ae1 ZY |
1091 | u32 num_refs; |
1092 | int ret; | |
74493f7a | 1093 | |
74493f7a | 1094 | key.objectid = bytenr; |
5d4f98a2 YZ |
1095 | if (parent) { |
1096 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1097 | key.offset = parent; | |
1098 | size = sizeof(struct btrfs_shared_data_ref); | |
1099 | } else { | |
1100 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1101 | key.offset = hash_extent_data_ref(root_objectid, | |
1102 | owner, offset); | |
1103 | size = sizeof(struct btrfs_extent_data_ref); | |
1104 | } | |
74493f7a | 1105 | |
5d4f98a2 YZ |
1106 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1107 | if (ret && ret != -EEXIST) | |
1108 | goto fail; | |
1109 | ||
1110 | leaf = path->nodes[0]; | |
1111 | if (parent) { | |
1112 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1113 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1114 | struct btrfs_shared_data_ref); |
1115 | if (ret == 0) { | |
1116 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1117 | } else { | |
1118 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1119 | num_refs += refs_to_add; | |
1120 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1121 | } |
5d4f98a2 YZ |
1122 | } else { |
1123 | struct btrfs_extent_data_ref *ref; | |
1124 | while (ret == -EEXIST) { | |
1125 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1126 | struct btrfs_extent_data_ref); | |
1127 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1128 | owner, offset)) | |
1129 | break; | |
1130 | btrfs_release_path(root, path); | |
1131 | key.offset++; | |
1132 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1133 | size); | |
1134 | if (ret && ret != -EEXIST) | |
1135 | goto fail; | |
31840ae1 | 1136 | |
5d4f98a2 YZ |
1137 | leaf = path->nodes[0]; |
1138 | } | |
1139 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1140 | struct btrfs_extent_data_ref); | |
1141 | if (ret == 0) { | |
1142 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1143 | root_objectid); | |
1144 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1145 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1146 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1147 | } else { | |
1148 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1149 | num_refs += refs_to_add; | |
1150 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1151 | } |
31840ae1 | 1152 | } |
5d4f98a2 YZ |
1153 | btrfs_mark_buffer_dirty(leaf); |
1154 | ret = 0; | |
1155 | fail: | |
7bb86316 CM |
1156 | btrfs_release_path(root, path); |
1157 | return ret; | |
74493f7a CM |
1158 | } |
1159 | ||
5d4f98a2 YZ |
1160 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1161 | struct btrfs_root *root, | |
1162 | struct btrfs_path *path, | |
1163 | int refs_to_drop) | |
31840ae1 | 1164 | { |
5d4f98a2 YZ |
1165 | struct btrfs_key key; |
1166 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1167 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1168 | struct extent_buffer *leaf; |
5d4f98a2 | 1169 | u32 num_refs = 0; |
31840ae1 ZY |
1170 | int ret = 0; |
1171 | ||
1172 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1173 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1174 | ||
1175 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1176 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1177 | struct btrfs_extent_data_ref); | |
1178 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1179 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1180 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1181 | struct btrfs_shared_data_ref); | |
1182 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1183 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1184 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1185 | struct btrfs_extent_ref_v0 *ref0; | |
1186 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1187 | struct btrfs_extent_ref_v0); | |
1188 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1189 | #endif | |
1190 | } else { | |
1191 | BUG(); | |
1192 | } | |
1193 | ||
56bec294 CM |
1194 | BUG_ON(num_refs < refs_to_drop); |
1195 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1196 | |
31840ae1 ZY |
1197 | if (num_refs == 0) { |
1198 | ret = btrfs_del_item(trans, root, path); | |
1199 | } else { | |
5d4f98a2 YZ |
1200 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1201 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1202 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1203 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1204 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1205 | else { | |
1206 | struct btrfs_extent_ref_v0 *ref0; | |
1207 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1208 | struct btrfs_extent_ref_v0); | |
1209 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1210 | } | |
1211 | #endif | |
31840ae1 ZY |
1212 | btrfs_mark_buffer_dirty(leaf); |
1213 | } | |
31840ae1 ZY |
1214 | return ret; |
1215 | } | |
1216 | ||
5d4f98a2 YZ |
1217 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1218 | struct btrfs_path *path, | |
1219 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1220 | { |
5d4f98a2 YZ |
1221 | struct btrfs_key key; |
1222 | struct extent_buffer *leaf; | |
1223 | struct btrfs_extent_data_ref *ref1; | |
1224 | struct btrfs_shared_data_ref *ref2; | |
1225 | u32 num_refs = 0; | |
1226 | ||
1227 | leaf = path->nodes[0]; | |
1228 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1229 | if (iref) { | |
1230 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1231 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1232 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1233 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1234 | } else { | |
1235 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1236 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1237 | } | |
1238 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1239 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1240 | struct btrfs_extent_data_ref); | |
1241 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1242 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1243 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1244 | struct btrfs_shared_data_ref); | |
1245 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1246 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1247 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1248 | struct btrfs_extent_ref_v0 *ref0; | |
1249 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1250 | struct btrfs_extent_ref_v0); | |
1251 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1252 | #endif |
5d4f98a2 YZ |
1253 | } else { |
1254 | WARN_ON(1); | |
1255 | } | |
1256 | return num_refs; | |
1257 | } | |
15916de8 | 1258 | |
5d4f98a2 YZ |
1259 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1260 | struct btrfs_root *root, | |
1261 | struct btrfs_path *path, | |
1262 | u64 bytenr, u64 parent, | |
1263 | u64 root_objectid) | |
1f3c79a2 | 1264 | { |
5d4f98a2 | 1265 | struct btrfs_key key; |
1f3c79a2 | 1266 | int ret; |
1f3c79a2 | 1267 | |
5d4f98a2 YZ |
1268 | key.objectid = bytenr; |
1269 | if (parent) { | |
1270 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1271 | key.offset = parent; | |
1272 | } else { | |
1273 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1274 | key.offset = root_objectid; | |
1f3c79a2 LH |
1275 | } |
1276 | ||
5d4f98a2 YZ |
1277 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1278 | if (ret > 0) | |
1279 | ret = -ENOENT; | |
1280 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1281 | if (ret == -ENOENT && parent) { | |
1282 | btrfs_release_path(root, path); | |
1283 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
1284 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1285 | if (ret > 0) | |
1286 | ret = -ENOENT; | |
1287 | } | |
1f3c79a2 | 1288 | #endif |
5d4f98a2 | 1289 | return ret; |
1f3c79a2 LH |
1290 | } |
1291 | ||
5d4f98a2 YZ |
1292 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1293 | struct btrfs_root *root, | |
1294 | struct btrfs_path *path, | |
1295 | u64 bytenr, u64 parent, | |
1296 | u64 root_objectid) | |
31840ae1 | 1297 | { |
5d4f98a2 | 1298 | struct btrfs_key key; |
31840ae1 | 1299 | int ret; |
31840ae1 | 1300 | |
5d4f98a2 YZ |
1301 | key.objectid = bytenr; |
1302 | if (parent) { | |
1303 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1304 | key.offset = parent; | |
1305 | } else { | |
1306 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1307 | key.offset = root_objectid; | |
1308 | } | |
1309 | ||
1310 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1311 | btrfs_release_path(root, path); | |
31840ae1 ZY |
1312 | return ret; |
1313 | } | |
1314 | ||
5d4f98a2 | 1315 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1316 | { |
5d4f98a2 YZ |
1317 | int type; |
1318 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1319 | if (parent > 0) | |
1320 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1321 | else | |
1322 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1323 | } else { | |
1324 | if (parent > 0) | |
1325 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1326 | else | |
1327 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1328 | } | |
1329 | return type; | |
31840ae1 | 1330 | } |
56bec294 | 1331 | |
2c47e605 YZ |
1332 | static int find_next_key(struct btrfs_path *path, int level, |
1333 | struct btrfs_key *key) | |
56bec294 | 1334 | |
02217ed2 | 1335 | { |
2c47e605 | 1336 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1337 | if (!path->nodes[level]) |
1338 | break; | |
5d4f98a2 YZ |
1339 | if (path->slots[level] + 1 >= |
1340 | btrfs_header_nritems(path->nodes[level])) | |
1341 | continue; | |
1342 | if (level == 0) | |
1343 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1344 | path->slots[level] + 1); | |
1345 | else | |
1346 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1347 | path->slots[level] + 1); | |
1348 | return 0; | |
1349 | } | |
1350 | return 1; | |
1351 | } | |
037e6390 | 1352 | |
5d4f98a2 YZ |
1353 | /* |
1354 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1355 | * to the address of inline back ref, and 0 is returned. | |
1356 | * | |
1357 | * if back ref isn't found, *ref_ret is set to the address where it | |
1358 | * should be inserted, and -ENOENT is returned. | |
1359 | * | |
1360 | * if insert is true and there are too many inline back refs, the path | |
1361 | * points to the extent item, and -EAGAIN is returned. | |
1362 | * | |
1363 | * NOTE: inline back refs are ordered in the same way that back ref | |
1364 | * items in the tree are ordered. | |
1365 | */ | |
1366 | static noinline_for_stack | |
1367 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1368 | struct btrfs_root *root, | |
1369 | struct btrfs_path *path, | |
1370 | struct btrfs_extent_inline_ref **ref_ret, | |
1371 | u64 bytenr, u64 num_bytes, | |
1372 | u64 parent, u64 root_objectid, | |
1373 | u64 owner, u64 offset, int insert) | |
1374 | { | |
1375 | struct btrfs_key key; | |
1376 | struct extent_buffer *leaf; | |
1377 | struct btrfs_extent_item *ei; | |
1378 | struct btrfs_extent_inline_ref *iref; | |
1379 | u64 flags; | |
1380 | u64 item_size; | |
1381 | unsigned long ptr; | |
1382 | unsigned long end; | |
1383 | int extra_size; | |
1384 | int type; | |
1385 | int want; | |
1386 | int ret; | |
1387 | int err = 0; | |
26b8003f | 1388 | |
db94535d | 1389 | key.objectid = bytenr; |
31840ae1 | 1390 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1391 | key.offset = num_bytes; |
31840ae1 | 1392 | |
5d4f98a2 YZ |
1393 | want = extent_ref_type(parent, owner); |
1394 | if (insert) { | |
1395 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1396 | path->keep_locks = 1; |
5d4f98a2 YZ |
1397 | } else |
1398 | extra_size = -1; | |
1399 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1400 | if (ret < 0) { |
5d4f98a2 YZ |
1401 | err = ret; |
1402 | goto out; | |
1403 | } | |
1404 | BUG_ON(ret); | |
1405 | ||
1406 | leaf = path->nodes[0]; | |
1407 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1408 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1409 | if (item_size < sizeof(*ei)) { | |
1410 | if (!insert) { | |
1411 | err = -ENOENT; | |
1412 | goto out; | |
1413 | } | |
1414 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1415 | extra_size); | |
1416 | if (ret < 0) { | |
1417 | err = ret; | |
1418 | goto out; | |
1419 | } | |
1420 | leaf = path->nodes[0]; | |
1421 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1422 | } | |
1423 | #endif | |
1424 | BUG_ON(item_size < sizeof(*ei)); | |
1425 | ||
5d4f98a2 YZ |
1426 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1427 | flags = btrfs_extent_flags(leaf, ei); | |
1428 | ||
1429 | ptr = (unsigned long)(ei + 1); | |
1430 | end = (unsigned long)ei + item_size; | |
1431 | ||
1432 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1433 | ptr += sizeof(struct btrfs_tree_block_info); | |
1434 | BUG_ON(ptr > end); | |
1435 | } else { | |
1436 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1437 | } | |
1438 | ||
1439 | err = -ENOENT; | |
1440 | while (1) { | |
1441 | if (ptr >= end) { | |
1442 | WARN_ON(ptr > end); | |
1443 | break; | |
1444 | } | |
1445 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1446 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1447 | if (want < type) | |
1448 | break; | |
1449 | if (want > type) { | |
1450 | ptr += btrfs_extent_inline_ref_size(type); | |
1451 | continue; | |
1452 | } | |
1453 | ||
1454 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1455 | struct btrfs_extent_data_ref *dref; | |
1456 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1457 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1458 | owner, offset)) { | |
1459 | err = 0; | |
1460 | break; | |
1461 | } | |
1462 | if (hash_extent_data_ref_item(leaf, dref) < | |
1463 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1464 | break; | |
1465 | } else { | |
1466 | u64 ref_offset; | |
1467 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1468 | if (parent > 0) { | |
1469 | if (parent == ref_offset) { | |
1470 | err = 0; | |
1471 | break; | |
1472 | } | |
1473 | if (ref_offset < parent) | |
1474 | break; | |
1475 | } else { | |
1476 | if (root_objectid == ref_offset) { | |
1477 | err = 0; | |
1478 | break; | |
1479 | } | |
1480 | if (ref_offset < root_objectid) | |
1481 | break; | |
1482 | } | |
1483 | } | |
1484 | ptr += btrfs_extent_inline_ref_size(type); | |
1485 | } | |
1486 | if (err == -ENOENT && insert) { | |
1487 | if (item_size + extra_size >= | |
1488 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1489 | err = -EAGAIN; | |
1490 | goto out; | |
1491 | } | |
1492 | /* | |
1493 | * To add new inline back ref, we have to make sure | |
1494 | * there is no corresponding back ref item. | |
1495 | * For simplicity, we just do not add new inline back | |
1496 | * ref if there is any kind of item for this block | |
1497 | */ | |
2c47e605 YZ |
1498 | if (find_next_key(path, 0, &key) == 0 && |
1499 | key.objectid == bytenr && | |
85d4198e | 1500 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1501 | err = -EAGAIN; |
1502 | goto out; | |
1503 | } | |
1504 | } | |
1505 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1506 | out: | |
85d4198e | 1507 | if (insert) { |
5d4f98a2 YZ |
1508 | path->keep_locks = 0; |
1509 | btrfs_unlock_up_safe(path, 1); | |
1510 | } | |
1511 | return err; | |
1512 | } | |
1513 | ||
1514 | /* | |
1515 | * helper to add new inline back ref | |
1516 | */ | |
1517 | static noinline_for_stack | |
1518 | int setup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1519 | struct btrfs_root *root, | |
1520 | struct btrfs_path *path, | |
1521 | struct btrfs_extent_inline_ref *iref, | |
1522 | u64 parent, u64 root_objectid, | |
1523 | u64 owner, u64 offset, int refs_to_add, | |
1524 | struct btrfs_delayed_extent_op *extent_op) | |
1525 | { | |
1526 | struct extent_buffer *leaf; | |
1527 | struct btrfs_extent_item *ei; | |
1528 | unsigned long ptr; | |
1529 | unsigned long end; | |
1530 | unsigned long item_offset; | |
1531 | u64 refs; | |
1532 | int size; | |
1533 | int type; | |
1534 | int ret; | |
1535 | ||
1536 | leaf = path->nodes[0]; | |
1537 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1538 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1539 | ||
1540 | type = extent_ref_type(parent, owner); | |
1541 | size = btrfs_extent_inline_ref_size(type); | |
1542 | ||
1543 | ret = btrfs_extend_item(trans, root, path, size); | |
1544 | BUG_ON(ret); | |
1545 | ||
1546 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1547 | refs = btrfs_extent_refs(leaf, ei); | |
1548 | refs += refs_to_add; | |
1549 | btrfs_set_extent_refs(leaf, ei, refs); | |
1550 | if (extent_op) | |
1551 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1552 | ||
1553 | ptr = (unsigned long)ei + item_offset; | |
1554 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1555 | if (ptr < end - size) | |
1556 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1557 | end - size - ptr); | |
1558 | ||
1559 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1560 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1561 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1562 | struct btrfs_extent_data_ref *dref; | |
1563 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1564 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1565 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1566 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1567 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1568 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1569 | struct btrfs_shared_data_ref *sref; | |
1570 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1571 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1572 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1573 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1574 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1575 | } else { | |
1576 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1577 | } | |
1578 | btrfs_mark_buffer_dirty(leaf); | |
1579 | return 0; | |
1580 | } | |
1581 | ||
1582 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1583 | struct btrfs_root *root, | |
1584 | struct btrfs_path *path, | |
1585 | struct btrfs_extent_inline_ref **ref_ret, | |
1586 | u64 bytenr, u64 num_bytes, u64 parent, | |
1587 | u64 root_objectid, u64 owner, u64 offset) | |
1588 | { | |
1589 | int ret; | |
1590 | ||
1591 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1592 | bytenr, num_bytes, parent, | |
1593 | root_objectid, owner, offset, 0); | |
1594 | if (ret != -ENOENT) | |
54aa1f4d | 1595 | return ret; |
5d4f98a2 YZ |
1596 | |
1597 | btrfs_release_path(root, path); | |
1598 | *ref_ret = NULL; | |
1599 | ||
1600 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1601 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1602 | root_objectid); | |
1603 | } else { | |
1604 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1605 | root_objectid, owner, offset); | |
b9473439 | 1606 | } |
5d4f98a2 YZ |
1607 | return ret; |
1608 | } | |
31840ae1 | 1609 | |
5d4f98a2 YZ |
1610 | /* |
1611 | * helper to update/remove inline back ref | |
1612 | */ | |
1613 | static noinline_for_stack | |
1614 | int update_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1615 | struct btrfs_root *root, | |
1616 | struct btrfs_path *path, | |
1617 | struct btrfs_extent_inline_ref *iref, | |
1618 | int refs_to_mod, | |
1619 | struct btrfs_delayed_extent_op *extent_op) | |
1620 | { | |
1621 | struct extent_buffer *leaf; | |
1622 | struct btrfs_extent_item *ei; | |
1623 | struct btrfs_extent_data_ref *dref = NULL; | |
1624 | struct btrfs_shared_data_ref *sref = NULL; | |
1625 | unsigned long ptr; | |
1626 | unsigned long end; | |
1627 | u32 item_size; | |
1628 | int size; | |
1629 | int type; | |
1630 | int ret; | |
1631 | u64 refs; | |
1632 | ||
1633 | leaf = path->nodes[0]; | |
1634 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1635 | refs = btrfs_extent_refs(leaf, ei); | |
1636 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1637 | refs += refs_to_mod; | |
1638 | btrfs_set_extent_refs(leaf, ei, refs); | |
1639 | if (extent_op) | |
1640 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1641 | ||
1642 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1643 | ||
1644 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1645 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1646 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1647 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1648 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1649 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1650 | } else { | |
1651 | refs = 1; | |
1652 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1653 | } |
31840ae1 | 1654 | |
5d4f98a2 YZ |
1655 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1656 | refs += refs_to_mod; | |
1657 | ||
1658 | if (refs > 0) { | |
1659 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1660 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1661 | else | |
1662 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1663 | } else { | |
1664 | size = btrfs_extent_inline_ref_size(type); | |
1665 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1666 | ptr = (unsigned long)iref; | |
1667 | end = (unsigned long)ei + item_size; | |
1668 | if (ptr + size < end) | |
1669 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1670 | end - ptr - size); | |
1671 | item_size -= size; | |
1672 | ret = btrfs_truncate_item(trans, root, path, item_size, 1); | |
1673 | BUG_ON(ret); | |
1674 | } | |
1675 | btrfs_mark_buffer_dirty(leaf); | |
1676 | return 0; | |
1677 | } | |
1678 | ||
1679 | static noinline_for_stack | |
1680 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1681 | struct btrfs_root *root, | |
1682 | struct btrfs_path *path, | |
1683 | u64 bytenr, u64 num_bytes, u64 parent, | |
1684 | u64 root_objectid, u64 owner, | |
1685 | u64 offset, int refs_to_add, | |
1686 | struct btrfs_delayed_extent_op *extent_op) | |
1687 | { | |
1688 | struct btrfs_extent_inline_ref *iref; | |
1689 | int ret; | |
1690 | ||
1691 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1692 | bytenr, num_bytes, parent, | |
1693 | root_objectid, owner, offset, 1); | |
1694 | if (ret == 0) { | |
1695 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
1696 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1697 | refs_to_add, extent_op); | |
1698 | } else if (ret == -ENOENT) { | |
1699 | ret = setup_inline_extent_backref(trans, root, path, iref, | |
1700 | parent, root_objectid, | |
1701 | owner, offset, refs_to_add, | |
1702 | extent_op); | |
771ed689 | 1703 | } |
5d4f98a2 YZ |
1704 | return ret; |
1705 | } | |
31840ae1 | 1706 | |
5d4f98a2 YZ |
1707 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1708 | struct btrfs_root *root, | |
1709 | struct btrfs_path *path, | |
1710 | u64 bytenr, u64 parent, u64 root_objectid, | |
1711 | u64 owner, u64 offset, int refs_to_add) | |
1712 | { | |
1713 | int ret; | |
1714 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1715 | BUG_ON(refs_to_add != 1); | |
1716 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1717 | parent, root_objectid); | |
1718 | } else { | |
1719 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1720 | parent, root_objectid, | |
1721 | owner, offset, refs_to_add); | |
1722 | } | |
1723 | return ret; | |
1724 | } | |
56bec294 | 1725 | |
5d4f98a2 YZ |
1726 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1727 | struct btrfs_root *root, | |
1728 | struct btrfs_path *path, | |
1729 | struct btrfs_extent_inline_ref *iref, | |
1730 | int refs_to_drop, int is_data) | |
1731 | { | |
1732 | int ret; | |
b9473439 | 1733 | |
5d4f98a2 YZ |
1734 | BUG_ON(!is_data && refs_to_drop != 1); |
1735 | if (iref) { | |
1736 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1737 | -refs_to_drop, NULL); | |
1738 | } else if (is_data) { | |
1739 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1740 | } else { | |
1741 | ret = btrfs_del_item(trans, root, path); | |
1742 | } | |
1743 | return ret; | |
1744 | } | |
1745 | ||
5d4f98a2 YZ |
1746 | static void btrfs_issue_discard(struct block_device *bdev, |
1747 | u64 start, u64 len) | |
1748 | { | |
746cd1e7 | 1749 | blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL, |
fbd9b09a | 1750 | BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER); |
5d4f98a2 | 1751 | } |
5d4f98a2 YZ |
1752 | |
1753 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
1754 | u64 num_bytes) | |
1755 | { | |
5d4f98a2 YZ |
1756 | int ret; |
1757 | u64 map_length = num_bytes; | |
1758 | struct btrfs_multi_bio *multi = NULL; | |
1759 | ||
e244a0ae CH |
1760 | if (!btrfs_test_opt(root, DISCARD)) |
1761 | return 0; | |
1762 | ||
5d4f98a2 YZ |
1763 | /* Tell the block device(s) that the sectors can be discarded */ |
1764 | ret = btrfs_map_block(&root->fs_info->mapping_tree, READ, | |
1765 | bytenr, &map_length, &multi, 0); | |
1766 | if (!ret) { | |
1767 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
1768 | int i; | |
1769 | ||
1770 | if (map_length > num_bytes) | |
1771 | map_length = num_bytes; | |
1772 | ||
1773 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
1774 | btrfs_issue_discard(stripe->dev->bdev, | |
1775 | stripe->physical, | |
1776 | map_length); | |
1777 | } | |
1778 | kfree(multi); | |
1779 | } | |
1780 | ||
1781 | return ret; | |
5d4f98a2 YZ |
1782 | } |
1783 | ||
1784 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1785 | struct btrfs_root *root, | |
1786 | u64 bytenr, u64 num_bytes, u64 parent, | |
1787 | u64 root_objectid, u64 owner, u64 offset) | |
1788 | { | |
1789 | int ret; | |
1790 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && | |
1791 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1792 | ||
1793 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1794 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
1795 | parent, root_objectid, (int)owner, | |
1796 | BTRFS_ADD_DELAYED_REF, NULL); | |
1797 | } else { | |
1798 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
1799 | parent, root_objectid, owner, offset, | |
1800 | BTRFS_ADD_DELAYED_REF, NULL); | |
1801 | } | |
1802 | return ret; | |
1803 | } | |
1804 | ||
1805 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1806 | struct btrfs_root *root, | |
1807 | u64 bytenr, u64 num_bytes, | |
1808 | u64 parent, u64 root_objectid, | |
1809 | u64 owner, u64 offset, int refs_to_add, | |
1810 | struct btrfs_delayed_extent_op *extent_op) | |
1811 | { | |
1812 | struct btrfs_path *path; | |
1813 | struct extent_buffer *leaf; | |
1814 | struct btrfs_extent_item *item; | |
1815 | u64 refs; | |
1816 | int ret; | |
1817 | int err = 0; | |
1818 | ||
1819 | path = btrfs_alloc_path(); | |
1820 | if (!path) | |
1821 | return -ENOMEM; | |
1822 | ||
1823 | path->reada = 1; | |
1824 | path->leave_spinning = 1; | |
1825 | /* this will setup the path even if it fails to insert the back ref */ | |
1826 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1827 | path, bytenr, num_bytes, parent, | |
1828 | root_objectid, owner, offset, | |
1829 | refs_to_add, extent_op); | |
1830 | if (ret == 0) | |
1831 | goto out; | |
1832 | ||
1833 | if (ret != -EAGAIN) { | |
1834 | err = ret; | |
1835 | goto out; | |
1836 | } | |
1837 | ||
1838 | leaf = path->nodes[0]; | |
1839 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1840 | refs = btrfs_extent_refs(leaf, item); | |
1841 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1842 | if (extent_op) | |
1843 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1844 | |
5d4f98a2 | 1845 | btrfs_mark_buffer_dirty(leaf); |
56bec294 CM |
1846 | btrfs_release_path(root->fs_info->extent_root, path); |
1847 | ||
1848 | path->reada = 1; | |
b9473439 CM |
1849 | path->leave_spinning = 1; |
1850 | ||
56bec294 CM |
1851 | /* now insert the actual backref */ |
1852 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1853 | path, bytenr, parent, root_objectid, |
1854 | owner, offset, refs_to_add); | |
56bec294 | 1855 | BUG_ON(ret); |
5d4f98a2 | 1856 | out: |
56bec294 | 1857 | btrfs_free_path(path); |
5d4f98a2 | 1858 | return err; |
56bec294 CM |
1859 | } |
1860 | ||
5d4f98a2 YZ |
1861 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1862 | struct btrfs_root *root, | |
1863 | struct btrfs_delayed_ref_node *node, | |
1864 | struct btrfs_delayed_extent_op *extent_op, | |
1865 | int insert_reserved) | |
56bec294 | 1866 | { |
5d4f98a2 YZ |
1867 | int ret = 0; |
1868 | struct btrfs_delayed_data_ref *ref; | |
1869 | struct btrfs_key ins; | |
1870 | u64 parent = 0; | |
1871 | u64 ref_root = 0; | |
1872 | u64 flags = 0; | |
1873 | ||
1874 | ins.objectid = node->bytenr; | |
1875 | ins.offset = node->num_bytes; | |
1876 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1877 | ||
1878 | ref = btrfs_delayed_node_to_data_ref(node); | |
1879 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1880 | parent = ref->parent; | |
1881 | else | |
1882 | ref_root = ref->root; | |
1883 | ||
1884 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1885 | if (extent_op) { | |
1886 | BUG_ON(extent_op->update_key); | |
1887 | flags |= extent_op->flags_to_set; | |
1888 | } | |
1889 | ret = alloc_reserved_file_extent(trans, root, | |
1890 | parent, ref_root, flags, | |
1891 | ref->objectid, ref->offset, | |
1892 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1893 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1894 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1895 | node->num_bytes, parent, | |
1896 | ref_root, ref->objectid, | |
1897 | ref->offset, node->ref_mod, | |
1898 | extent_op); | |
1899 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1900 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1901 | node->num_bytes, parent, | |
1902 | ref_root, ref->objectid, | |
1903 | ref->offset, node->ref_mod, | |
1904 | extent_op); | |
1905 | } else { | |
1906 | BUG(); | |
1907 | } | |
1908 | return ret; | |
1909 | } | |
1910 | ||
1911 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1912 | struct extent_buffer *leaf, | |
1913 | struct btrfs_extent_item *ei) | |
1914 | { | |
1915 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1916 | if (extent_op->update_flags) { | |
1917 | flags |= extent_op->flags_to_set; | |
1918 | btrfs_set_extent_flags(leaf, ei, flags); | |
1919 | } | |
1920 | ||
1921 | if (extent_op->update_key) { | |
1922 | struct btrfs_tree_block_info *bi; | |
1923 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1924 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1925 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1926 | } | |
1927 | } | |
1928 | ||
1929 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
1930 | struct btrfs_root *root, | |
1931 | struct btrfs_delayed_ref_node *node, | |
1932 | struct btrfs_delayed_extent_op *extent_op) | |
1933 | { | |
1934 | struct btrfs_key key; | |
1935 | struct btrfs_path *path; | |
1936 | struct btrfs_extent_item *ei; | |
1937 | struct extent_buffer *leaf; | |
1938 | u32 item_size; | |
56bec294 | 1939 | int ret; |
5d4f98a2 YZ |
1940 | int err = 0; |
1941 | ||
1942 | path = btrfs_alloc_path(); | |
1943 | if (!path) | |
1944 | return -ENOMEM; | |
1945 | ||
1946 | key.objectid = node->bytenr; | |
1947 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1948 | key.offset = node->num_bytes; | |
1949 | ||
1950 | path->reada = 1; | |
1951 | path->leave_spinning = 1; | |
1952 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
1953 | path, 0, 1); | |
1954 | if (ret < 0) { | |
1955 | err = ret; | |
1956 | goto out; | |
1957 | } | |
1958 | if (ret > 0) { | |
1959 | err = -EIO; | |
1960 | goto out; | |
1961 | } | |
1962 | ||
1963 | leaf = path->nodes[0]; | |
1964 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1965 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1966 | if (item_size < sizeof(*ei)) { | |
1967 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
1968 | path, (u64)-1, 0); | |
1969 | if (ret < 0) { | |
1970 | err = ret; | |
1971 | goto out; | |
1972 | } | |
1973 | leaf = path->nodes[0]; | |
1974 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1975 | } | |
1976 | #endif | |
1977 | BUG_ON(item_size < sizeof(*ei)); | |
1978 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1979 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1980 | |
5d4f98a2 YZ |
1981 | btrfs_mark_buffer_dirty(leaf); |
1982 | out: | |
1983 | btrfs_free_path(path); | |
1984 | return err; | |
56bec294 CM |
1985 | } |
1986 | ||
5d4f98a2 YZ |
1987 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
1988 | struct btrfs_root *root, | |
1989 | struct btrfs_delayed_ref_node *node, | |
1990 | struct btrfs_delayed_extent_op *extent_op, | |
1991 | int insert_reserved) | |
56bec294 CM |
1992 | { |
1993 | int ret = 0; | |
5d4f98a2 YZ |
1994 | struct btrfs_delayed_tree_ref *ref; |
1995 | struct btrfs_key ins; | |
1996 | u64 parent = 0; | |
1997 | u64 ref_root = 0; | |
56bec294 | 1998 | |
5d4f98a2 YZ |
1999 | ins.objectid = node->bytenr; |
2000 | ins.offset = node->num_bytes; | |
2001 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 2002 | |
5d4f98a2 YZ |
2003 | ref = btrfs_delayed_node_to_tree_ref(node); |
2004 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2005 | parent = ref->parent; | |
2006 | else | |
2007 | ref_root = ref->root; | |
2008 | ||
2009 | BUG_ON(node->ref_mod != 1); | |
2010 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2011 | BUG_ON(!extent_op || !extent_op->update_flags || | |
2012 | !extent_op->update_key); | |
2013 | ret = alloc_reserved_tree_block(trans, root, | |
2014 | parent, ref_root, | |
2015 | extent_op->flags_to_set, | |
2016 | &extent_op->key, | |
2017 | ref->level, &ins); | |
5d4f98a2 YZ |
2018 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2019 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2020 | node->num_bytes, parent, ref_root, | |
2021 | ref->level, 0, 1, extent_op); | |
2022 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2023 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2024 | node->num_bytes, parent, ref_root, | |
2025 | ref->level, 0, 1, extent_op); | |
2026 | } else { | |
2027 | BUG(); | |
2028 | } | |
56bec294 CM |
2029 | return ret; |
2030 | } | |
2031 | ||
2032 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2033 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2034 | struct btrfs_root *root, | |
2035 | struct btrfs_delayed_ref_node *node, | |
2036 | struct btrfs_delayed_extent_op *extent_op, | |
2037 | int insert_reserved) | |
56bec294 CM |
2038 | { |
2039 | int ret; | |
5d4f98a2 | 2040 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2041 | struct btrfs_delayed_ref_head *head; |
2042 | /* | |
2043 | * we've hit the end of the chain and we were supposed | |
2044 | * to insert this extent into the tree. But, it got | |
2045 | * deleted before we ever needed to insert it, so all | |
2046 | * we have to do is clean up the accounting | |
2047 | */ | |
5d4f98a2 YZ |
2048 | BUG_ON(extent_op); |
2049 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2050 | if (insert_reserved) { |
f0486c68 YZ |
2051 | btrfs_pin_extent(root, node->bytenr, |
2052 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2053 | if (head->is_data) { |
2054 | ret = btrfs_del_csums(trans, root, | |
2055 | node->bytenr, | |
2056 | node->num_bytes); | |
2057 | BUG_ON(ret); | |
2058 | } | |
56bec294 | 2059 | } |
56bec294 CM |
2060 | mutex_unlock(&head->mutex); |
2061 | return 0; | |
2062 | } | |
2063 | ||
5d4f98a2 YZ |
2064 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2065 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2066 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2067 | insert_reserved); | |
2068 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2069 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2070 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2071 | insert_reserved); | |
2072 | else | |
2073 | BUG(); | |
2074 | return ret; | |
56bec294 CM |
2075 | } |
2076 | ||
2077 | static noinline struct btrfs_delayed_ref_node * | |
2078 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2079 | { | |
2080 | struct rb_node *node; | |
2081 | struct btrfs_delayed_ref_node *ref; | |
2082 | int action = BTRFS_ADD_DELAYED_REF; | |
2083 | again: | |
2084 | /* | |
2085 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2086 | * this prevents ref count from going down to zero when | |
2087 | * there still are pending delayed ref. | |
2088 | */ | |
2089 | node = rb_prev(&head->node.rb_node); | |
2090 | while (1) { | |
2091 | if (!node) | |
2092 | break; | |
2093 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2094 | rb_node); | |
2095 | if (ref->bytenr != head->node.bytenr) | |
2096 | break; | |
5d4f98a2 | 2097 | if (ref->action == action) |
56bec294 CM |
2098 | return ref; |
2099 | node = rb_prev(node); | |
2100 | } | |
2101 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2102 | action = BTRFS_DROP_DELAYED_REF; | |
2103 | goto again; | |
2104 | } | |
2105 | return NULL; | |
2106 | } | |
2107 | ||
c3e69d58 CM |
2108 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2109 | struct btrfs_root *root, | |
2110 | struct list_head *cluster) | |
56bec294 | 2111 | { |
56bec294 CM |
2112 | struct btrfs_delayed_ref_root *delayed_refs; |
2113 | struct btrfs_delayed_ref_node *ref; | |
2114 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2115 | struct btrfs_delayed_extent_op *extent_op; |
56bec294 | 2116 | int ret; |
c3e69d58 | 2117 | int count = 0; |
56bec294 | 2118 | int must_insert_reserved = 0; |
56bec294 CM |
2119 | |
2120 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2121 | while (1) { |
2122 | if (!locked_ref) { | |
c3e69d58 CM |
2123 | /* pick a new head ref from the cluster list */ |
2124 | if (list_empty(cluster)) | |
56bec294 | 2125 | break; |
56bec294 | 2126 | |
c3e69d58 CM |
2127 | locked_ref = list_entry(cluster->next, |
2128 | struct btrfs_delayed_ref_head, cluster); | |
2129 | ||
2130 | /* grab the lock that says we are going to process | |
2131 | * all the refs for this head */ | |
2132 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2133 | ||
2134 | /* | |
2135 | * we may have dropped the spin lock to get the head | |
2136 | * mutex lock, and that might have given someone else | |
2137 | * time to free the head. If that's true, it has been | |
2138 | * removed from our list and we can move on. | |
2139 | */ | |
2140 | if (ret == -EAGAIN) { | |
2141 | locked_ref = NULL; | |
2142 | count++; | |
2143 | continue; | |
56bec294 CM |
2144 | } |
2145 | } | |
a28ec197 | 2146 | |
56bec294 CM |
2147 | /* |
2148 | * record the must insert reserved flag before we | |
2149 | * drop the spin lock. | |
2150 | */ | |
2151 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2152 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2153 | |
5d4f98a2 YZ |
2154 | extent_op = locked_ref->extent_op; |
2155 | locked_ref->extent_op = NULL; | |
2156 | ||
56bec294 CM |
2157 | /* |
2158 | * locked_ref is the head node, so we have to go one | |
2159 | * node back for any delayed ref updates | |
2160 | */ | |
56bec294 CM |
2161 | ref = select_delayed_ref(locked_ref); |
2162 | if (!ref) { | |
2163 | /* All delayed refs have been processed, Go ahead | |
2164 | * and send the head node to run_one_delayed_ref, | |
2165 | * so that any accounting fixes can happen | |
2166 | */ | |
2167 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2168 | |
2169 | if (extent_op && must_insert_reserved) { | |
2170 | kfree(extent_op); | |
2171 | extent_op = NULL; | |
2172 | } | |
2173 | ||
2174 | if (extent_op) { | |
2175 | spin_unlock(&delayed_refs->lock); | |
2176 | ||
2177 | ret = run_delayed_extent_op(trans, root, | |
2178 | ref, extent_op); | |
2179 | BUG_ON(ret); | |
2180 | kfree(extent_op); | |
2181 | ||
2182 | cond_resched(); | |
2183 | spin_lock(&delayed_refs->lock); | |
2184 | continue; | |
2185 | } | |
2186 | ||
c3e69d58 | 2187 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2188 | locked_ref = NULL; |
2189 | } | |
02217ed2 | 2190 | |
56bec294 CM |
2191 | ref->in_tree = 0; |
2192 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2193 | delayed_refs->num_entries--; | |
5d4f98a2 | 2194 | |
56bec294 | 2195 | spin_unlock(&delayed_refs->lock); |
925baedd | 2196 | |
5d4f98a2 | 2197 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 CM |
2198 | must_insert_reserved); |
2199 | BUG_ON(ret); | |
eb099670 | 2200 | |
5d4f98a2 YZ |
2201 | btrfs_put_delayed_ref(ref); |
2202 | kfree(extent_op); | |
c3e69d58 | 2203 | count++; |
5d4f98a2 | 2204 | |
c3e69d58 CM |
2205 | cond_resched(); |
2206 | spin_lock(&delayed_refs->lock); | |
2207 | } | |
2208 | return count; | |
2209 | } | |
2210 | ||
2211 | /* | |
2212 | * this starts processing the delayed reference count updates and | |
2213 | * extent insertions we have queued up so far. count can be | |
2214 | * 0, which means to process everything in the tree at the start | |
2215 | * of the run (but not newly added entries), or it can be some target | |
2216 | * number you'd like to process. | |
2217 | */ | |
2218 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2219 | struct btrfs_root *root, unsigned long count) | |
2220 | { | |
2221 | struct rb_node *node; | |
2222 | struct btrfs_delayed_ref_root *delayed_refs; | |
2223 | struct btrfs_delayed_ref_node *ref; | |
2224 | struct list_head cluster; | |
2225 | int ret; | |
2226 | int run_all = count == (unsigned long)-1; | |
2227 | int run_most = 0; | |
2228 | ||
2229 | if (root == root->fs_info->extent_root) | |
2230 | root = root->fs_info->tree_root; | |
2231 | ||
2232 | delayed_refs = &trans->transaction->delayed_refs; | |
2233 | INIT_LIST_HEAD(&cluster); | |
2234 | again: | |
2235 | spin_lock(&delayed_refs->lock); | |
2236 | if (count == 0) { | |
2237 | count = delayed_refs->num_entries * 2; | |
2238 | run_most = 1; | |
2239 | } | |
2240 | while (1) { | |
2241 | if (!(run_all || run_most) && | |
2242 | delayed_refs->num_heads_ready < 64) | |
2243 | break; | |
eb099670 | 2244 | |
56bec294 | 2245 | /* |
c3e69d58 CM |
2246 | * go find something we can process in the rbtree. We start at |
2247 | * the beginning of the tree, and then build a cluster | |
2248 | * of refs to process starting at the first one we are able to | |
2249 | * lock | |
56bec294 | 2250 | */ |
c3e69d58 CM |
2251 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2252 | delayed_refs->run_delayed_start); | |
2253 | if (ret) | |
56bec294 CM |
2254 | break; |
2255 | ||
c3e69d58 CM |
2256 | ret = run_clustered_refs(trans, root, &cluster); |
2257 | BUG_ON(ret < 0); | |
2258 | ||
2259 | count -= min_t(unsigned long, ret, count); | |
2260 | ||
2261 | if (count == 0) | |
2262 | break; | |
eb099670 | 2263 | } |
c3e69d58 | 2264 | |
56bec294 | 2265 | if (run_all) { |
56bec294 | 2266 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2267 | if (!node) |
56bec294 | 2268 | goto out; |
c3e69d58 | 2269 | count = (unsigned long)-1; |
e9d0b13b | 2270 | |
56bec294 CM |
2271 | while (node) { |
2272 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2273 | rb_node); | |
2274 | if (btrfs_delayed_ref_is_head(ref)) { | |
2275 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2276 | |
56bec294 CM |
2277 | head = btrfs_delayed_node_to_head(ref); |
2278 | atomic_inc(&ref->refs); | |
2279 | ||
2280 | spin_unlock(&delayed_refs->lock); | |
2281 | mutex_lock(&head->mutex); | |
2282 | mutex_unlock(&head->mutex); | |
2283 | ||
2284 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2285 | cond_resched(); |
56bec294 CM |
2286 | goto again; |
2287 | } | |
2288 | node = rb_next(node); | |
2289 | } | |
2290 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2291 | schedule_timeout(1); |
2292 | goto again; | |
5f39d397 | 2293 | } |
54aa1f4d | 2294 | out: |
c3e69d58 | 2295 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2296 | return 0; |
2297 | } | |
2298 | ||
5d4f98a2 YZ |
2299 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2300 | struct btrfs_root *root, | |
2301 | u64 bytenr, u64 num_bytes, u64 flags, | |
2302 | int is_data) | |
2303 | { | |
2304 | struct btrfs_delayed_extent_op *extent_op; | |
2305 | int ret; | |
2306 | ||
2307 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2308 | if (!extent_op) | |
2309 | return -ENOMEM; | |
2310 | ||
2311 | extent_op->flags_to_set = flags; | |
2312 | extent_op->update_flags = 1; | |
2313 | extent_op->update_key = 0; | |
2314 | extent_op->is_data = is_data ? 1 : 0; | |
2315 | ||
2316 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); | |
2317 | if (ret) | |
2318 | kfree(extent_op); | |
2319 | return ret; | |
2320 | } | |
2321 | ||
2322 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2323 | struct btrfs_root *root, | |
2324 | struct btrfs_path *path, | |
2325 | u64 objectid, u64 offset, u64 bytenr) | |
2326 | { | |
2327 | struct btrfs_delayed_ref_head *head; | |
2328 | struct btrfs_delayed_ref_node *ref; | |
2329 | struct btrfs_delayed_data_ref *data_ref; | |
2330 | struct btrfs_delayed_ref_root *delayed_refs; | |
2331 | struct rb_node *node; | |
2332 | int ret = 0; | |
2333 | ||
2334 | ret = -ENOENT; | |
2335 | delayed_refs = &trans->transaction->delayed_refs; | |
2336 | spin_lock(&delayed_refs->lock); | |
2337 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2338 | if (!head) | |
2339 | goto out; | |
2340 | ||
2341 | if (!mutex_trylock(&head->mutex)) { | |
2342 | atomic_inc(&head->node.refs); | |
2343 | spin_unlock(&delayed_refs->lock); | |
2344 | ||
2345 | btrfs_release_path(root->fs_info->extent_root, path); | |
2346 | ||
2347 | mutex_lock(&head->mutex); | |
2348 | mutex_unlock(&head->mutex); | |
2349 | btrfs_put_delayed_ref(&head->node); | |
2350 | return -EAGAIN; | |
2351 | } | |
2352 | ||
2353 | node = rb_prev(&head->node.rb_node); | |
2354 | if (!node) | |
2355 | goto out_unlock; | |
2356 | ||
2357 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2358 | ||
2359 | if (ref->bytenr != bytenr) | |
2360 | goto out_unlock; | |
2361 | ||
2362 | ret = 1; | |
2363 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2364 | goto out_unlock; | |
2365 | ||
2366 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2367 | ||
2368 | node = rb_prev(node); | |
2369 | if (node) { | |
2370 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2371 | if (ref->bytenr == bytenr) | |
2372 | goto out_unlock; | |
2373 | } | |
2374 | ||
2375 | if (data_ref->root != root->root_key.objectid || | |
2376 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2377 | goto out_unlock; | |
2378 | ||
2379 | ret = 0; | |
2380 | out_unlock: | |
2381 | mutex_unlock(&head->mutex); | |
2382 | out: | |
2383 | spin_unlock(&delayed_refs->lock); | |
2384 | return ret; | |
2385 | } | |
2386 | ||
2387 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2388 | struct btrfs_root *root, | |
2389 | struct btrfs_path *path, | |
2390 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2391 | { |
2392 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2393 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2394 | struct btrfs_extent_data_ref *ref; |
2395 | struct btrfs_extent_inline_ref *iref; | |
2396 | struct btrfs_extent_item *ei; | |
f321e491 | 2397 | struct btrfs_key key; |
5d4f98a2 | 2398 | u32 item_size; |
be20aa9d | 2399 | int ret; |
925baedd | 2400 | |
be20aa9d | 2401 | key.objectid = bytenr; |
31840ae1 | 2402 | key.offset = (u64)-1; |
f321e491 | 2403 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2404 | |
be20aa9d CM |
2405 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2406 | if (ret < 0) | |
2407 | goto out; | |
2408 | BUG_ON(ret == 0); | |
80ff3856 YZ |
2409 | |
2410 | ret = -ENOENT; | |
2411 | if (path->slots[0] == 0) | |
31840ae1 | 2412 | goto out; |
be20aa9d | 2413 | |
31840ae1 | 2414 | path->slots[0]--; |
f321e491 | 2415 | leaf = path->nodes[0]; |
5d4f98a2 | 2416 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2417 | |
5d4f98a2 | 2418 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2419 | goto out; |
f321e491 | 2420 | |
5d4f98a2 YZ |
2421 | ret = 1; |
2422 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2423 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2424 | if (item_size < sizeof(*ei)) { | |
2425 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2426 | goto out; | |
2427 | } | |
2428 | #endif | |
2429 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2430 | |
5d4f98a2 YZ |
2431 | if (item_size != sizeof(*ei) + |
2432 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2433 | goto out; | |
be20aa9d | 2434 | |
5d4f98a2 YZ |
2435 | if (btrfs_extent_generation(leaf, ei) <= |
2436 | btrfs_root_last_snapshot(&root->root_item)) | |
2437 | goto out; | |
2438 | ||
2439 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2440 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2441 | BTRFS_EXTENT_DATA_REF_KEY) | |
2442 | goto out; | |
2443 | ||
2444 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2445 | if (btrfs_extent_refs(leaf, ei) != | |
2446 | btrfs_extent_data_ref_count(leaf, ref) || | |
2447 | btrfs_extent_data_ref_root(leaf, ref) != | |
2448 | root->root_key.objectid || | |
2449 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2450 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2451 | goto out; | |
2452 | ||
2453 | ret = 0; | |
2454 | out: | |
2455 | return ret; | |
2456 | } | |
2457 | ||
2458 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2459 | struct btrfs_root *root, | |
2460 | u64 objectid, u64 offset, u64 bytenr) | |
2461 | { | |
2462 | struct btrfs_path *path; | |
2463 | int ret; | |
2464 | int ret2; | |
2465 | ||
2466 | path = btrfs_alloc_path(); | |
2467 | if (!path) | |
2468 | return -ENOENT; | |
2469 | ||
2470 | do { | |
2471 | ret = check_committed_ref(trans, root, path, objectid, | |
2472 | offset, bytenr); | |
2473 | if (ret && ret != -ENOENT) | |
f321e491 | 2474 | goto out; |
80ff3856 | 2475 | |
5d4f98a2 YZ |
2476 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2477 | offset, bytenr); | |
2478 | } while (ret2 == -EAGAIN); | |
2479 | ||
2480 | if (ret2 && ret2 != -ENOENT) { | |
2481 | ret = ret2; | |
2482 | goto out; | |
f321e491 | 2483 | } |
5d4f98a2 YZ |
2484 | |
2485 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2486 | ret = 0; | |
be20aa9d | 2487 | out: |
80ff3856 | 2488 | btrfs_free_path(path); |
f0486c68 YZ |
2489 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2490 | WARN_ON(ret > 0); | |
f321e491 | 2491 | return ret; |
be20aa9d | 2492 | } |
c5739bba | 2493 | |
5d4f98a2 | 2494 | #if 0 |
31840ae1 ZY |
2495 | int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2496 | struct extent_buffer *buf, u32 nr_extents) | |
02217ed2 | 2497 | { |
5f39d397 | 2498 | struct btrfs_key key; |
6407bf6d | 2499 | struct btrfs_file_extent_item *fi; |
e4657689 ZY |
2500 | u64 root_gen; |
2501 | u32 nritems; | |
02217ed2 | 2502 | int i; |
db94535d | 2503 | int level; |
31840ae1 | 2504 | int ret = 0; |
e4657689 | 2505 | int shared = 0; |
a28ec197 | 2506 | |
3768f368 | 2507 | if (!root->ref_cows) |
a28ec197 | 2508 | return 0; |
5f39d397 | 2509 | |
e4657689 ZY |
2510 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
2511 | shared = 0; | |
2512 | root_gen = root->root_key.offset; | |
2513 | } else { | |
2514 | shared = 1; | |
2515 | root_gen = trans->transid - 1; | |
2516 | } | |
2517 | ||
db94535d | 2518 | level = btrfs_header_level(buf); |
5f39d397 | 2519 | nritems = btrfs_header_nritems(buf); |
4a096752 | 2520 | |
31840ae1 | 2521 | if (level == 0) { |
31153d81 YZ |
2522 | struct btrfs_leaf_ref *ref; |
2523 | struct btrfs_extent_info *info; | |
2524 | ||
31840ae1 | 2525 | ref = btrfs_alloc_leaf_ref(root, nr_extents); |
31153d81 | 2526 | if (!ref) { |
31840ae1 | 2527 | ret = -ENOMEM; |
31153d81 YZ |
2528 | goto out; |
2529 | } | |
2530 | ||
e4657689 | 2531 | ref->root_gen = root_gen; |
31153d81 YZ |
2532 | ref->bytenr = buf->start; |
2533 | ref->owner = btrfs_header_owner(buf); | |
2534 | ref->generation = btrfs_header_generation(buf); | |
31840ae1 | 2535 | ref->nritems = nr_extents; |
31153d81 | 2536 | info = ref->extents; |
bcc63abb | 2537 | |
31840ae1 | 2538 | for (i = 0; nr_extents > 0 && i < nritems; i++) { |
31153d81 YZ |
2539 | u64 disk_bytenr; |
2540 | btrfs_item_key_to_cpu(buf, &key, i); | |
2541 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
2542 | continue; | |
2543 | fi = btrfs_item_ptr(buf, i, | |
2544 | struct btrfs_file_extent_item); | |
2545 | if (btrfs_file_extent_type(buf, fi) == | |
2546 | BTRFS_FILE_EXTENT_INLINE) | |
2547 | continue; | |
2548 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2549 | if (disk_bytenr == 0) | |
2550 | continue; | |
2551 | ||
2552 | info->bytenr = disk_bytenr; | |
2553 | info->num_bytes = | |
2554 | btrfs_file_extent_disk_num_bytes(buf, fi); | |
2555 | info->objectid = key.objectid; | |
2556 | info->offset = key.offset; | |
2557 | info++; | |
2558 | } | |
2559 | ||
e4657689 | 2560 | ret = btrfs_add_leaf_ref(root, ref, shared); |
5b84e8d6 YZ |
2561 | if (ret == -EEXIST && shared) { |
2562 | struct btrfs_leaf_ref *old; | |
2563 | old = btrfs_lookup_leaf_ref(root, ref->bytenr); | |
2564 | BUG_ON(!old); | |
2565 | btrfs_remove_leaf_ref(root, old); | |
2566 | btrfs_free_leaf_ref(root, old); | |
2567 | ret = btrfs_add_leaf_ref(root, ref, shared); | |
2568 | } | |
31153d81 | 2569 | WARN_ON(ret); |
bcc63abb | 2570 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
2571 | } |
2572 | out: | |
31840ae1 ZY |
2573 | return ret; |
2574 | } | |
2575 | ||
b7a9f29f CM |
2576 | /* when a block goes through cow, we update the reference counts of |
2577 | * everything that block points to. The internal pointers of the block | |
2578 | * can be in just about any order, and it is likely to have clusters of | |
2579 | * things that are close together and clusters of things that are not. | |
2580 | * | |
2581 | * To help reduce the seeks that come with updating all of these reference | |
2582 | * counts, sort them by byte number before actual updates are done. | |
2583 | * | |
2584 | * struct refsort is used to match byte number to slot in the btree block. | |
2585 | * we sort based on the byte number and then use the slot to actually | |
2586 | * find the item. | |
bd56b302 CM |
2587 | * |
2588 | * struct refsort is smaller than strcut btrfs_item and smaller than | |
2589 | * struct btrfs_key_ptr. Since we're currently limited to the page size | |
2590 | * for a btree block, there's no way for a kmalloc of refsorts for a | |
2591 | * single node to be bigger than a page. | |
b7a9f29f CM |
2592 | */ |
2593 | struct refsort { | |
2594 | u64 bytenr; | |
2595 | u32 slot; | |
2596 | }; | |
2597 | ||
2598 | /* | |
2599 | * for passing into sort() | |
2600 | */ | |
2601 | static int refsort_cmp(const void *a_void, const void *b_void) | |
2602 | { | |
2603 | const struct refsort *a = a_void; | |
2604 | const struct refsort *b = b_void; | |
2605 | ||
2606 | if (a->bytenr < b->bytenr) | |
2607 | return -1; | |
2608 | if (a->bytenr > b->bytenr) | |
2609 | return 1; | |
2610 | return 0; | |
2611 | } | |
5d4f98a2 | 2612 | #endif |
b7a9f29f | 2613 | |
5d4f98a2 | 2614 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2615 | struct btrfs_root *root, |
5d4f98a2 YZ |
2616 | struct extent_buffer *buf, |
2617 | int full_backref, int inc) | |
31840ae1 ZY |
2618 | { |
2619 | u64 bytenr; | |
5d4f98a2 YZ |
2620 | u64 num_bytes; |
2621 | u64 parent; | |
31840ae1 | 2622 | u64 ref_root; |
31840ae1 | 2623 | u32 nritems; |
31840ae1 ZY |
2624 | struct btrfs_key key; |
2625 | struct btrfs_file_extent_item *fi; | |
2626 | int i; | |
2627 | int level; | |
2628 | int ret = 0; | |
31840ae1 | 2629 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
5d4f98a2 | 2630 | u64, u64, u64, u64, u64, u64); |
31840ae1 ZY |
2631 | |
2632 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2633 | nritems = btrfs_header_nritems(buf); |
2634 | level = btrfs_header_level(buf); | |
2635 | ||
5d4f98a2 YZ |
2636 | if (!root->ref_cows && level == 0) |
2637 | return 0; | |
31840ae1 | 2638 | |
5d4f98a2 YZ |
2639 | if (inc) |
2640 | process_func = btrfs_inc_extent_ref; | |
2641 | else | |
2642 | process_func = btrfs_free_extent; | |
31840ae1 | 2643 | |
5d4f98a2 YZ |
2644 | if (full_backref) |
2645 | parent = buf->start; | |
2646 | else | |
2647 | parent = 0; | |
2648 | ||
2649 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2650 | if (level == 0) { |
5d4f98a2 | 2651 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2652 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2653 | continue; | |
5d4f98a2 | 2654 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2655 | struct btrfs_file_extent_item); |
2656 | if (btrfs_file_extent_type(buf, fi) == | |
2657 | BTRFS_FILE_EXTENT_INLINE) | |
2658 | continue; | |
2659 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2660 | if (bytenr == 0) | |
2661 | continue; | |
5d4f98a2 YZ |
2662 | |
2663 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2664 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2665 | ret = process_func(trans, root, bytenr, num_bytes, | |
2666 | parent, ref_root, key.objectid, | |
2667 | key.offset); | |
31840ae1 ZY |
2668 | if (ret) |
2669 | goto fail; | |
2670 | } else { | |
5d4f98a2 YZ |
2671 | bytenr = btrfs_node_blockptr(buf, i); |
2672 | num_bytes = btrfs_level_size(root, level - 1); | |
2673 | ret = process_func(trans, root, bytenr, num_bytes, | |
2674 | parent, ref_root, level - 1, 0); | |
31840ae1 ZY |
2675 | if (ret) |
2676 | goto fail; | |
2677 | } | |
2678 | } | |
2679 | return 0; | |
2680 | fail: | |
5d4f98a2 YZ |
2681 | BUG(); |
2682 | return ret; | |
2683 | } | |
2684 | ||
2685 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2686 | struct extent_buffer *buf, int full_backref) | |
2687 | { | |
2688 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); | |
2689 | } | |
2690 | ||
2691 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2692 | struct extent_buffer *buf, int full_backref) | |
2693 | { | |
2694 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); | |
31840ae1 ZY |
2695 | } |
2696 | ||
9078a3e1 CM |
2697 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2698 | struct btrfs_root *root, | |
2699 | struct btrfs_path *path, | |
2700 | struct btrfs_block_group_cache *cache) | |
2701 | { | |
2702 | int ret; | |
9078a3e1 | 2703 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2704 | unsigned long bi; |
2705 | struct extent_buffer *leaf; | |
9078a3e1 | 2706 | |
9078a3e1 | 2707 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2708 | if (ret < 0) |
2709 | goto fail; | |
9078a3e1 | 2710 | BUG_ON(ret); |
5f39d397 CM |
2711 | |
2712 | leaf = path->nodes[0]; | |
2713 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2714 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2715 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 2716 | btrfs_release_path(extent_root, path); |
54aa1f4d | 2717 | fail: |
9078a3e1 CM |
2718 | if (ret) |
2719 | return ret; | |
9078a3e1 CM |
2720 | return 0; |
2721 | ||
2722 | } | |
2723 | ||
4a8c9a62 YZ |
2724 | static struct btrfs_block_group_cache * |
2725 | next_block_group(struct btrfs_root *root, | |
2726 | struct btrfs_block_group_cache *cache) | |
2727 | { | |
2728 | struct rb_node *node; | |
2729 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2730 | node = rb_next(&cache->cache_node); | |
2731 | btrfs_put_block_group(cache); | |
2732 | if (node) { | |
2733 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2734 | cache_node); | |
11dfe35a | 2735 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2736 | } else |
2737 | cache = NULL; | |
2738 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2739 | return cache; | |
2740 | } | |
2741 | ||
0af3d00b JB |
2742 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2743 | struct btrfs_trans_handle *trans, | |
2744 | struct btrfs_path *path) | |
2745 | { | |
2746 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2747 | struct inode *inode = NULL; | |
2748 | u64 alloc_hint = 0; | |
2b20982e | 2749 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2750 | int num_pages = 0; |
2751 | int retries = 0; | |
2752 | int ret = 0; | |
2753 | ||
2754 | /* | |
2755 | * If this block group is smaller than 100 megs don't bother caching the | |
2756 | * block group. | |
2757 | */ | |
2758 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2759 | spin_lock(&block_group->lock); | |
2760 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2761 | spin_unlock(&block_group->lock); | |
2762 | return 0; | |
2763 | } | |
2764 | ||
2765 | again: | |
2766 | inode = lookup_free_space_inode(root, block_group, path); | |
2767 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2768 | ret = PTR_ERR(inode); | |
2769 | btrfs_release_path(root, path); | |
2770 | goto out; | |
2771 | } | |
2772 | ||
2773 | if (IS_ERR(inode)) { | |
2774 | BUG_ON(retries); | |
2775 | retries++; | |
2776 | ||
2777 | if (block_group->ro) | |
2778 | goto out_free; | |
2779 | ||
2780 | ret = create_free_space_inode(root, trans, block_group, path); | |
2781 | if (ret) | |
2782 | goto out_free; | |
2783 | goto again; | |
2784 | } | |
2785 | ||
2786 | /* | |
2787 | * We want to set the generation to 0, that way if anything goes wrong | |
2788 | * from here on out we know not to trust this cache when we load up next | |
2789 | * time. | |
2790 | */ | |
2791 | BTRFS_I(inode)->generation = 0; | |
2792 | ret = btrfs_update_inode(trans, root, inode); | |
2793 | WARN_ON(ret); | |
2794 | ||
2795 | if (i_size_read(inode) > 0) { | |
2796 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2797 | inode); | |
2798 | if (ret) | |
2799 | goto out_put; | |
2800 | } | |
2801 | ||
2802 | spin_lock(&block_group->lock); | |
2803 | if (block_group->cached != BTRFS_CACHE_FINISHED) { | |
2b20982e JB |
2804 | /* We're not cached, don't bother trying to write stuff out */ |
2805 | dcs = BTRFS_DC_WRITTEN; | |
0af3d00b JB |
2806 | spin_unlock(&block_group->lock); |
2807 | goto out_put; | |
2808 | } | |
2809 | spin_unlock(&block_group->lock); | |
2810 | ||
2811 | num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024); | |
2812 | if (!num_pages) | |
2813 | num_pages = 1; | |
2814 | ||
2815 | /* | |
2816 | * Just to make absolutely sure we have enough space, we're going to | |
2817 | * preallocate 12 pages worth of space for each block group. In | |
2818 | * practice we ought to use at most 8, but we need extra space so we can | |
2819 | * add our header and have a terminator between the extents and the | |
2820 | * bitmaps. | |
2821 | */ | |
2822 | num_pages *= 16; | |
2823 | num_pages *= PAGE_CACHE_SIZE; | |
2824 | ||
2825 | ret = btrfs_check_data_free_space(inode, num_pages); | |
2826 | if (ret) | |
2827 | goto out_put; | |
2828 | ||
2829 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2830 | num_pages, num_pages, | |
2831 | &alloc_hint); | |
2b20982e JB |
2832 | if (!ret) |
2833 | dcs = BTRFS_DC_SETUP; | |
0af3d00b JB |
2834 | btrfs_free_reserved_data_space(inode, num_pages); |
2835 | out_put: | |
2836 | iput(inode); | |
2837 | out_free: | |
2838 | btrfs_release_path(root, path); | |
2839 | out: | |
2840 | spin_lock(&block_group->lock); | |
2b20982e | 2841 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
2842 | spin_unlock(&block_group->lock); |
2843 | ||
2844 | return ret; | |
2845 | } | |
2846 | ||
96b5179d CM |
2847 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2848 | struct btrfs_root *root) | |
9078a3e1 | 2849 | { |
4a8c9a62 | 2850 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2851 | int err = 0; |
9078a3e1 | 2852 | struct btrfs_path *path; |
96b5179d | 2853 | u64 last = 0; |
9078a3e1 CM |
2854 | |
2855 | path = btrfs_alloc_path(); | |
2856 | if (!path) | |
2857 | return -ENOMEM; | |
2858 | ||
0af3d00b JB |
2859 | again: |
2860 | while (1) { | |
2861 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2862 | while (cache) { | |
2863 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2864 | break; | |
2865 | cache = next_block_group(root, cache); | |
2866 | } | |
2867 | if (!cache) { | |
2868 | if (last == 0) | |
2869 | break; | |
2870 | last = 0; | |
2871 | continue; | |
2872 | } | |
2873 | err = cache_save_setup(cache, trans, path); | |
2874 | last = cache->key.objectid + cache->key.offset; | |
2875 | btrfs_put_block_group(cache); | |
2876 | } | |
2877 | ||
d397712b | 2878 | while (1) { |
4a8c9a62 YZ |
2879 | if (last == 0) { |
2880 | err = btrfs_run_delayed_refs(trans, root, | |
2881 | (unsigned long)-1); | |
2882 | BUG_ON(err); | |
0f9dd46c | 2883 | } |
54aa1f4d | 2884 | |
4a8c9a62 YZ |
2885 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2886 | while (cache) { | |
0af3d00b JB |
2887 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
2888 | btrfs_put_block_group(cache); | |
2889 | goto again; | |
2890 | } | |
2891 | ||
4a8c9a62 YZ |
2892 | if (cache->dirty) |
2893 | break; | |
2894 | cache = next_block_group(root, cache); | |
2895 | } | |
2896 | if (!cache) { | |
2897 | if (last == 0) | |
2898 | break; | |
2899 | last = 0; | |
2900 | continue; | |
2901 | } | |
0f9dd46c | 2902 | |
0cb59c99 JB |
2903 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
2904 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 2905 | cache->dirty = 0; |
4a8c9a62 | 2906 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 2907 | |
4a8c9a62 YZ |
2908 | err = write_one_cache_group(trans, root, path, cache); |
2909 | BUG_ON(err); | |
2910 | btrfs_put_block_group(cache); | |
9078a3e1 | 2911 | } |
4a8c9a62 | 2912 | |
0cb59c99 JB |
2913 | while (1) { |
2914 | /* | |
2915 | * I don't think this is needed since we're just marking our | |
2916 | * preallocated extent as written, but just in case it can't | |
2917 | * hurt. | |
2918 | */ | |
2919 | if (last == 0) { | |
2920 | err = btrfs_run_delayed_refs(trans, root, | |
2921 | (unsigned long)-1); | |
2922 | BUG_ON(err); | |
2923 | } | |
2924 | ||
2925 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2926 | while (cache) { | |
2927 | /* | |
2928 | * Really this shouldn't happen, but it could if we | |
2929 | * couldn't write the entire preallocated extent and | |
2930 | * splitting the extent resulted in a new block. | |
2931 | */ | |
2932 | if (cache->dirty) { | |
2933 | btrfs_put_block_group(cache); | |
2934 | goto again; | |
2935 | } | |
2936 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
2937 | break; | |
2938 | cache = next_block_group(root, cache); | |
2939 | } | |
2940 | if (!cache) { | |
2941 | if (last == 0) | |
2942 | break; | |
2943 | last = 0; | |
2944 | continue; | |
2945 | } | |
2946 | ||
2947 | btrfs_write_out_cache(root, trans, cache, path); | |
2948 | ||
2949 | /* | |
2950 | * If we didn't have an error then the cache state is still | |
2951 | * NEED_WRITE, so we can set it to WRITTEN. | |
2952 | */ | |
2953 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
2954 | cache->disk_cache_state = BTRFS_DC_WRITTEN; | |
2955 | last = cache->key.objectid + cache->key.offset; | |
2956 | btrfs_put_block_group(cache); | |
2957 | } | |
2958 | ||
9078a3e1 | 2959 | btrfs_free_path(path); |
4a8c9a62 | 2960 | return 0; |
9078a3e1 CM |
2961 | } |
2962 | ||
d2fb3437 YZ |
2963 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
2964 | { | |
2965 | struct btrfs_block_group_cache *block_group; | |
2966 | int readonly = 0; | |
2967 | ||
2968 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
2969 | if (!block_group || block_group->ro) | |
2970 | readonly = 1; | |
2971 | if (block_group) | |
fa9c0d79 | 2972 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2973 | return readonly; |
2974 | } | |
2975 | ||
593060d7 CM |
2976 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
2977 | u64 total_bytes, u64 bytes_used, | |
2978 | struct btrfs_space_info **space_info) | |
2979 | { | |
2980 | struct btrfs_space_info *found; | |
b742bb82 YZ |
2981 | int i; |
2982 | int factor; | |
2983 | ||
2984 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
2985 | BTRFS_BLOCK_GROUP_RAID10)) | |
2986 | factor = 2; | |
2987 | else | |
2988 | factor = 1; | |
593060d7 CM |
2989 | |
2990 | found = __find_space_info(info, flags); | |
2991 | if (found) { | |
25179201 | 2992 | spin_lock(&found->lock); |
593060d7 | 2993 | found->total_bytes += total_bytes; |
89a55897 | 2994 | found->disk_total += total_bytes * factor; |
593060d7 | 2995 | found->bytes_used += bytes_used; |
b742bb82 | 2996 | found->disk_used += bytes_used * factor; |
8f18cf13 | 2997 | found->full = 0; |
25179201 | 2998 | spin_unlock(&found->lock); |
593060d7 CM |
2999 | *space_info = found; |
3000 | return 0; | |
3001 | } | |
c146afad | 3002 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3003 | if (!found) |
3004 | return -ENOMEM; | |
3005 | ||
b742bb82 YZ |
3006 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3007 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3008 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3009 | spin_lock_init(&found->lock); |
b742bb82 YZ |
3010 | found->flags = flags & (BTRFS_BLOCK_GROUP_DATA | |
3011 | BTRFS_BLOCK_GROUP_SYSTEM | | |
3012 | BTRFS_BLOCK_GROUP_METADATA); | |
593060d7 | 3013 | found->total_bytes = total_bytes; |
89a55897 | 3014 | found->disk_total = total_bytes * factor; |
593060d7 | 3015 | found->bytes_used = bytes_used; |
b742bb82 | 3016 | found->disk_used = bytes_used * factor; |
593060d7 | 3017 | found->bytes_pinned = 0; |
e8569813 | 3018 | found->bytes_reserved = 0; |
c146afad | 3019 | found->bytes_readonly = 0; |
f0486c68 | 3020 | found->bytes_may_use = 0; |
593060d7 | 3021 | found->full = 0; |
0ef3e66b | 3022 | found->force_alloc = 0; |
593060d7 | 3023 | *space_info = found; |
4184ea7f | 3024 | list_add_rcu(&found->list, &info->space_info); |
817d52f8 | 3025 | atomic_set(&found->caching_threads, 0); |
593060d7 CM |
3026 | return 0; |
3027 | } | |
3028 | ||
8790d502 CM |
3029 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3030 | { | |
3031 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 3032 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 3033 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 3034 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
3035 | if (extra_flags) { |
3036 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3037 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3038 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3039 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3040 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3041 | fs_info->avail_system_alloc_bits |= extra_flags; | |
3042 | } | |
3043 | } | |
593060d7 | 3044 | |
2b82032c | 3045 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3046 | { |
cd02dca5 CM |
3047 | /* |
3048 | * we add in the count of missing devices because we want | |
3049 | * to make sure that any RAID levels on a degraded FS | |
3050 | * continue to be honored. | |
3051 | */ | |
3052 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3053 | root->fs_info->fs_devices->missing_devices; | |
a061fc8d CM |
3054 | |
3055 | if (num_devices == 1) | |
3056 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
3057 | if (num_devices < 4) | |
3058 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3059 | ||
ec44a35c CM |
3060 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
3061 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 3062 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 3063 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 3064 | } |
ec44a35c CM |
3065 | |
3066 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 3067 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 3068 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 3069 | } |
ec44a35c CM |
3070 | |
3071 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
3072 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
3073 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
3074 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
3075 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
3076 | return flags; | |
3077 | } | |
3078 | ||
b742bb82 | 3079 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3080 | { |
b742bb82 YZ |
3081 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3082 | flags |= root->fs_info->avail_data_alloc_bits & | |
3083 | root->fs_info->data_alloc_profile; | |
3084 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3085 | flags |= root->fs_info->avail_system_alloc_bits & | |
3086 | root->fs_info->system_alloc_profile; | |
3087 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3088 | flags |= root->fs_info->avail_metadata_alloc_bits & | |
3089 | root->fs_info->metadata_alloc_profile; | |
3090 | return btrfs_reduce_alloc_profile(root, flags); | |
6a63209f JB |
3091 | } |
3092 | ||
6d07bcec | 3093 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3094 | { |
b742bb82 | 3095 | u64 flags; |
9ed74f2d | 3096 | |
b742bb82 YZ |
3097 | if (data) |
3098 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3099 | else if (root == root->fs_info->chunk_root) | |
3100 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3101 | else |
b742bb82 | 3102 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3103 | |
b742bb82 | 3104 | return get_alloc_profile(root, flags); |
6a63209f | 3105 | } |
9ed74f2d | 3106 | |
6a63209f JB |
3107 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) |
3108 | { | |
6a63209f | 3109 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, |
f0486c68 | 3110 | BTRFS_BLOCK_GROUP_DATA); |
9ed74f2d JB |
3111 | } |
3112 | ||
6a63209f | 3113 | /* |
6a63209f JB |
3114 | * This will check the space that the inode allocates from to make sure we have |
3115 | * enough space for bytes. | |
6a63209f | 3116 | */ |
0ca1f7ce | 3117 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3118 | { |
6a63209f | 3119 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3120 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ab6e2410 | 3121 | u64 used; |
0af3d00b | 3122 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3123 | |
6a63209f JB |
3124 | /* make sure bytes are sectorsize aligned */ |
3125 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3126 | |
0af3d00b JB |
3127 | if (root == root->fs_info->tree_root) { |
3128 | alloc_chunk = 0; | |
3129 | committed = 1; | |
3130 | } | |
3131 | ||
6a63209f | 3132 | data_sinfo = BTRFS_I(inode)->space_info; |
33b4d47f CM |
3133 | if (!data_sinfo) |
3134 | goto alloc; | |
9ed74f2d | 3135 | |
6a63209f JB |
3136 | again: |
3137 | /* make sure we have enough space to handle the data first */ | |
3138 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3139 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3140 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3141 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3142 | |
3143 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3144 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3145 | |
6a63209f JB |
3146 | /* |
3147 | * if we don't have enough free bytes in this space then we need | |
3148 | * to alloc a new chunk. | |
3149 | */ | |
0af3d00b | 3150 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3151 | u64 alloc_target; |
9ed74f2d | 3152 | |
6a63209f JB |
3153 | data_sinfo->force_alloc = 1; |
3154 | spin_unlock(&data_sinfo->lock); | |
33b4d47f | 3155 | alloc: |
6a63209f | 3156 | alloc_target = btrfs_get_alloc_profile(root, 1); |
a22285a6 YZ |
3157 | trans = btrfs_join_transaction(root, 1); |
3158 | if (IS_ERR(trans)) | |
3159 | return PTR_ERR(trans); | |
9ed74f2d | 3160 | |
6a63209f JB |
3161 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
3162 | bytes + 2 * 1024 * 1024, | |
3163 | alloc_target, 0); | |
3164 | btrfs_end_transaction(trans, root); | |
d52a5b5f MX |
3165 | if (ret < 0) { |
3166 | if (ret != -ENOSPC) | |
3167 | return ret; | |
3168 | else | |
3169 | goto commit_trans; | |
3170 | } | |
9ed74f2d | 3171 | |
33b4d47f CM |
3172 | if (!data_sinfo) { |
3173 | btrfs_set_inode_space_info(root, inode); | |
3174 | data_sinfo = BTRFS_I(inode)->space_info; | |
3175 | } | |
6a63209f JB |
3176 | goto again; |
3177 | } | |
3178 | spin_unlock(&data_sinfo->lock); | |
6a63209f | 3179 | |
4e06bdd6 | 3180 | /* commit the current transaction and try again */ |
d52a5b5f | 3181 | commit_trans: |
dd7e0b7b | 3182 | if (!committed && !root->fs_info->open_ioctl_trans) { |
4e06bdd6 JB |
3183 | committed = 1; |
3184 | trans = btrfs_join_transaction(root, 1); | |
a22285a6 YZ |
3185 | if (IS_ERR(trans)) |
3186 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3187 | ret = btrfs_commit_transaction(trans, root); |
3188 | if (ret) | |
3189 | return ret; | |
3190 | goto again; | |
3191 | } | |
9ed74f2d | 3192 | |
933b585f | 3193 | #if 0 /* I hope we never need this code again, just in case */ |
8929ecfa YZ |
3194 | printk(KERN_ERR "no space left, need %llu, %llu bytes_used, " |
3195 | "%llu bytes_reserved, " "%llu bytes_pinned, " | |
3196 | "%llu bytes_readonly, %llu may use %llu total\n", | |
3197 | (unsigned long long)bytes, | |
21380931 JB |
3198 | (unsigned long long)data_sinfo->bytes_used, |
3199 | (unsigned long long)data_sinfo->bytes_reserved, | |
3200 | (unsigned long long)data_sinfo->bytes_pinned, | |
3201 | (unsigned long long)data_sinfo->bytes_readonly, | |
3202 | (unsigned long long)data_sinfo->bytes_may_use, | |
3203 | (unsigned long long)data_sinfo->total_bytes); | |
933b585f | 3204 | #endif |
6a63209f JB |
3205 | return -ENOSPC; |
3206 | } | |
3207 | data_sinfo->bytes_may_use += bytes; | |
3208 | BTRFS_I(inode)->reserved_bytes += bytes; | |
3209 | spin_unlock(&data_sinfo->lock); | |
6a63209f | 3210 | |
9ed74f2d | 3211 | return 0; |
9ed74f2d | 3212 | } |
6a63209f | 3213 | |
6a63209f | 3214 | /* |
0ca1f7ce YZ |
3215 | * called when we are clearing an delalloc extent from the |
3216 | * inode's io_tree or there was an error for whatever reason | |
3217 | * after calling btrfs_check_data_free_space | |
6a63209f | 3218 | */ |
0ca1f7ce | 3219 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3220 | { |
0ca1f7ce | 3221 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3222 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3223 | |
6a63209f JB |
3224 | /* make sure bytes are sectorsize aligned */ |
3225 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3226 | |
6a63209f JB |
3227 | data_sinfo = BTRFS_I(inode)->space_info; |
3228 | spin_lock(&data_sinfo->lock); | |
3229 | data_sinfo->bytes_may_use -= bytes; | |
3230 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
3231 | spin_unlock(&data_sinfo->lock); | |
e3ccfa98 JB |
3232 | } |
3233 | ||
97e728d4 | 3234 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3235 | { |
97e728d4 JB |
3236 | struct list_head *head = &info->space_info; |
3237 | struct btrfs_space_info *found; | |
e3ccfa98 | 3238 | |
97e728d4 JB |
3239 | rcu_read_lock(); |
3240 | list_for_each_entry_rcu(found, head, list) { | |
3241 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
3242 | found->force_alloc = 1; | |
e3ccfa98 | 3243 | } |
97e728d4 | 3244 | rcu_read_unlock(); |
e3ccfa98 JB |
3245 | } |
3246 | ||
e5bc2458 CM |
3247 | static int should_alloc_chunk(struct btrfs_root *root, |
3248 | struct btrfs_space_info *sinfo, u64 alloc_bytes) | |
32c00aff | 3249 | { |
424499db | 3250 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
e5bc2458 | 3251 | u64 thresh; |
e3ccfa98 | 3252 | |
424499db YZ |
3253 | if (sinfo->bytes_used + sinfo->bytes_reserved + |
3254 | alloc_bytes + 256 * 1024 * 1024 < num_bytes) | |
3255 | return 0; | |
e3ccfa98 | 3256 | |
424499db YZ |
3257 | if (sinfo->bytes_used + sinfo->bytes_reserved + |
3258 | alloc_bytes < div_factor(num_bytes, 8)) | |
3259 | return 0; | |
32c00aff | 3260 | |
e5bc2458 CM |
3261 | thresh = btrfs_super_total_bytes(&root->fs_info->super_copy); |
3262 | thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5)); | |
3263 | ||
3264 | if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3)) | |
14ed0ca6 JB |
3265 | return 0; |
3266 | ||
424499db | 3267 | return 1; |
32c00aff JB |
3268 | } |
3269 | ||
6324fbf3 CM |
3270 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3271 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3272 | u64 flags, int force) |
9ed74f2d | 3273 | { |
6324fbf3 | 3274 | struct btrfs_space_info *space_info; |
97e728d4 | 3275 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
9ed74f2d | 3276 | int ret = 0; |
9ed74f2d | 3277 | |
97e728d4 | 3278 | mutex_lock(&fs_info->chunk_mutex); |
9ed74f2d | 3279 | |
2b82032c | 3280 | flags = btrfs_reduce_alloc_profile(extent_root, flags); |
ec44a35c | 3281 | |
6324fbf3 | 3282 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3283 | if (!space_info) { |
3284 | ret = update_space_info(extent_root->fs_info, flags, | |
3285 | 0, 0, &space_info); | |
3286 | BUG_ON(ret); | |
9ed74f2d | 3287 | } |
6324fbf3 | 3288 | BUG_ON(!space_info); |
9ed74f2d | 3289 | |
25179201 | 3290 | spin_lock(&space_info->lock); |
9ed74f2d | 3291 | if (space_info->force_alloc) |
0ef3e66b | 3292 | force = 1; |
25179201 JB |
3293 | if (space_info->full) { |
3294 | spin_unlock(&space_info->lock); | |
925baedd | 3295 | goto out; |
9ed74f2d JB |
3296 | } |
3297 | ||
e5bc2458 CM |
3298 | if (!force && !should_alloc_chunk(extent_root, space_info, |
3299 | alloc_bytes)) { | |
25179201 | 3300 | spin_unlock(&space_info->lock); |
925baedd | 3301 | goto out; |
9ed74f2d | 3302 | } |
25179201 | 3303 | spin_unlock(&space_info->lock); |
9ed74f2d | 3304 | |
67377734 JB |
3305 | /* |
3306 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3307 | * allocating mixed chunks instead of individual chunks. | |
3308 | */ | |
3309 | if (btrfs_mixed_space_info(space_info)) | |
3310 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3311 | ||
97e728d4 JB |
3312 | /* |
3313 | * if we're doing a data chunk, go ahead and make sure that | |
3314 | * we keep a reasonable number of metadata chunks allocated in the | |
3315 | * FS as well. | |
3316 | */ | |
9ed74f2d | 3317 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3318 | fs_info->data_chunk_allocations++; |
3319 | if (!(fs_info->data_chunk_allocations % | |
3320 | fs_info->metadata_ratio)) | |
3321 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3322 | } |
3323 | ||
2b82032c | 3324 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
9ed74f2d | 3325 | spin_lock(&space_info->lock); |
9ed74f2d | 3326 | if (ret) |
6324fbf3 | 3327 | space_info->full = 1; |
424499db YZ |
3328 | else |
3329 | ret = 1; | |
9ed74f2d JB |
3330 | space_info->force_alloc = 0; |
3331 | spin_unlock(&space_info->lock); | |
9ed74f2d | 3332 | out: |
c146afad | 3333 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 3334 | return ret; |
6324fbf3 | 3335 | } |
9ed74f2d | 3336 | |
9ed74f2d | 3337 | /* |
5da9d01b | 3338 | * shrink metadata reservation for delalloc |
9ed74f2d | 3339 | */ |
5da9d01b | 3340 | static int shrink_delalloc(struct btrfs_trans_handle *trans, |
0019f10d | 3341 | struct btrfs_root *root, u64 to_reclaim, int sync) |
5da9d01b | 3342 | { |
0ca1f7ce | 3343 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3344 | struct btrfs_space_info *space_info; |
5da9d01b YZ |
3345 | u64 reserved; |
3346 | u64 max_reclaim; | |
3347 | u64 reclaimed = 0; | |
3348 | int pause = 1; | |
bf9022e0 | 3349 | int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
5da9d01b | 3350 | |
0ca1f7ce | 3351 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3352 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3353 | |
3354 | smp_mb(); | |
0019f10d | 3355 | reserved = space_info->bytes_reserved; |
5da9d01b YZ |
3356 | |
3357 | if (reserved == 0) | |
3358 | return 0; | |
3359 | ||
3360 | max_reclaim = min(reserved, to_reclaim); | |
3361 | ||
3362 | while (1) { | |
bf9022e0 CM |
3363 | /* have the flusher threads jump in and do some IO */ |
3364 | smp_mb(); | |
3365 | nr_pages = min_t(unsigned long, nr_pages, | |
3366 | root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT); | |
3367 | writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages); | |
5da9d01b | 3368 | |
0019f10d JB |
3369 | spin_lock(&space_info->lock); |
3370 | if (reserved > space_info->bytes_reserved) | |
3371 | reclaimed += reserved - space_info->bytes_reserved; | |
3372 | reserved = space_info->bytes_reserved; | |
3373 | spin_unlock(&space_info->lock); | |
5da9d01b YZ |
3374 | |
3375 | if (reserved == 0 || reclaimed >= max_reclaim) | |
3376 | break; | |
3377 | ||
3378 | if (trans && trans->transaction->blocked) | |
3379 | return -EAGAIN; | |
bf9022e0 CM |
3380 | |
3381 | __set_current_state(TASK_INTERRUPTIBLE); | |
3382 | schedule_timeout(pause); | |
3383 | pause <<= 1; | |
3384 | if (pause > HZ / 10) | |
3385 | pause = HZ / 10; | |
3386 | ||
5da9d01b YZ |
3387 | } |
3388 | return reclaimed >= to_reclaim; | |
3389 | } | |
3390 | ||
8bb8ab2e JB |
3391 | /* |
3392 | * Retries tells us how many times we've called reserve_metadata_bytes. The | |
3393 | * idea is if this is the first call (retries == 0) then we will add to our | |
3394 | * reserved count if we can't make the allocation in order to hold our place | |
3395 | * while we go and try and free up space. That way for retries > 1 we don't try | |
3396 | * and add space, we just check to see if the amount of unused space is >= the | |
3397 | * total space, meaning that our reservation is valid. | |
3398 | * | |
3399 | * However if we don't intend to retry this reservation, pass -1 as retries so | |
3400 | * that it short circuits this logic. | |
3401 | */ | |
3402 | static int reserve_metadata_bytes(struct btrfs_trans_handle *trans, | |
3403 | struct btrfs_root *root, | |
3404 | struct btrfs_block_rsv *block_rsv, | |
3405 | u64 orig_bytes, int flush) | |
9ed74f2d | 3406 | { |
f0486c68 | 3407 | struct btrfs_space_info *space_info = block_rsv->space_info; |
8bb8ab2e JB |
3408 | u64 unused; |
3409 | u64 num_bytes = orig_bytes; | |
3410 | int retries = 0; | |
3411 | int ret = 0; | |
3412 | bool reserved = false; | |
38227933 | 3413 | bool committed = false; |
9ed74f2d | 3414 | |
8bb8ab2e JB |
3415 | again: |
3416 | ret = -ENOSPC; | |
3417 | if (reserved) | |
3418 | num_bytes = 0; | |
9ed74f2d | 3419 | |
8bb8ab2e JB |
3420 | spin_lock(&space_info->lock); |
3421 | unused = space_info->bytes_used + space_info->bytes_reserved + | |
3422 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3423 | space_info->bytes_may_use; | |
9ed74f2d | 3424 | |
8bb8ab2e JB |
3425 | /* |
3426 | * The idea here is that we've not already over-reserved the block group | |
3427 | * then we can go ahead and save our reservation first and then start | |
3428 | * flushing if we need to. Otherwise if we've already overcommitted | |
3429 | * lets start flushing stuff first and then come back and try to make | |
3430 | * our reservation. | |
3431 | */ | |
3432 | if (unused <= space_info->total_bytes) { | |
6f334348 | 3433 | unused = space_info->total_bytes - unused; |
8bb8ab2e JB |
3434 | if (unused >= num_bytes) { |
3435 | if (!reserved) | |
3436 | space_info->bytes_reserved += orig_bytes; | |
3437 | ret = 0; | |
3438 | } else { | |
3439 | /* | |
3440 | * Ok set num_bytes to orig_bytes since we aren't | |
3441 | * overocmmitted, this way we only try and reclaim what | |
3442 | * we need. | |
3443 | */ | |
3444 | num_bytes = orig_bytes; | |
3445 | } | |
3446 | } else { | |
3447 | /* | |
3448 | * Ok we're over committed, set num_bytes to the overcommitted | |
3449 | * amount plus the amount of bytes that we need for this | |
3450 | * reservation. | |
3451 | */ | |
3452 | num_bytes = unused - space_info->total_bytes + | |
3453 | (orig_bytes * (retries + 1)); | |
3454 | } | |
9ed74f2d | 3455 | |
8bb8ab2e JB |
3456 | /* |
3457 | * Couldn't make our reservation, save our place so while we're trying | |
3458 | * to reclaim space we can actually use it instead of somebody else | |
3459 | * stealing it from us. | |
3460 | */ | |
3461 | if (ret && !reserved) { | |
3462 | space_info->bytes_reserved += orig_bytes; | |
3463 | reserved = true; | |
3464 | } | |
9ed74f2d | 3465 | |
f0486c68 | 3466 | spin_unlock(&space_info->lock); |
9ed74f2d | 3467 | |
8bb8ab2e JB |
3468 | if (!ret) |
3469 | return 0; | |
9ed74f2d | 3470 | |
8bb8ab2e JB |
3471 | if (!flush) |
3472 | goto out; | |
f0486c68 | 3473 | |
8bb8ab2e JB |
3474 | /* |
3475 | * We do synchronous shrinking since we don't actually unreserve | |
3476 | * metadata until after the IO is completed. | |
3477 | */ | |
3478 | ret = shrink_delalloc(trans, root, num_bytes, 1); | |
3479 | if (ret > 0) | |
3480 | return 0; | |
3481 | else if (ret < 0) | |
3482 | goto out; | |
f0486c68 | 3483 | |
8bb8ab2e JB |
3484 | /* |
3485 | * So if we were overcommitted it's possible that somebody else flushed | |
3486 | * out enough space and we simply didn't have enough space to reclaim, | |
3487 | * so go back around and try again. | |
3488 | */ | |
3489 | if (retries < 2) { | |
3490 | retries++; | |
3491 | goto again; | |
3492 | } | |
f0486c68 YZ |
3493 | |
3494 | spin_lock(&space_info->lock); | |
8bb8ab2e JB |
3495 | /* |
3496 | * Not enough space to be reclaimed, don't bother committing the | |
3497 | * transaction. | |
3498 | */ | |
3499 | if (space_info->bytes_pinned < orig_bytes) | |
3500 | ret = -ENOSPC; | |
3501 | spin_unlock(&space_info->lock); | |
3502 | if (ret) | |
3503 | goto out; | |
f0486c68 | 3504 | |
8bb8ab2e | 3505 | ret = -EAGAIN; |
38227933 | 3506 | if (trans || committed) |
8bb8ab2e | 3507 | goto out; |
f0486c68 | 3508 | |
8bb8ab2e JB |
3509 | ret = -ENOSPC; |
3510 | trans = btrfs_join_transaction(root, 1); | |
3511 | if (IS_ERR(trans)) | |
3512 | goto out; | |
3513 | ret = btrfs_commit_transaction(trans, root); | |
38227933 JB |
3514 | if (!ret) { |
3515 | trans = NULL; | |
3516 | committed = true; | |
8bb8ab2e | 3517 | goto again; |
38227933 | 3518 | } |
8bb8ab2e JB |
3519 | |
3520 | out: | |
3521 | if (reserved) { | |
3522 | spin_lock(&space_info->lock); | |
3523 | space_info->bytes_reserved -= orig_bytes; | |
3524 | spin_unlock(&space_info->lock); | |
f0486c68 | 3525 | } |
4e06bdd6 | 3526 | |
f0486c68 YZ |
3527 | return ret; |
3528 | } | |
3529 | ||
3530 | static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans, | |
3531 | struct btrfs_root *root) | |
3532 | { | |
3533 | struct btrfs_block_rsv *block_rsv; | |
3534 | if (root->ref_cows) | |
3535 | block_rsv = trans->block_rsv; | |
3536 | else | |
3537 | block_rsv = root->block_rsv; | |
3538 | ||
3539 | if (!block_rsv) | |
3540 | block_rsv = &root->fs_info->empty_block_rsv; | |
3541 | ||
3542 | return block_rsv; | |
3543 | } | |
3544 | ||
3545 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
3546 | u64 num_bytes) | |
3547 | { | |
3548 | int ret = -ENOSPC; | |
3549 | spin_lock(&block_rsv->lock); | |
3550 | if (block_rsv->reserved >= num_bytes) { | |
3551 | block_rsv->reserved -= num_bytes; | |
3552 | if (block_rsv->reserved < block_rsv->size) | |
3553 | block_rsv->full = 0; | |
3554 | ret = 0; | |
3555 | } | |
3556 | spin_unlock(&block_rsv->lock); | |
3557 | return ret; | |
3558 | } | |
3559 | ||
3560 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3561 | u64 num_bytes, int update_size) | |
3562 | { | |
3563 | spin_lock(&block_rsv->lock); | |
3564 | block_rsv->reserved += num_bytes; | |
3565 | if (update_size) | |
3566 | block_rsv->size += num_bytes; | |
3567 | else if (block_rsv->reserved >= block_rsv->size) | |
3568 | block_rsv->full = 1; | |
3569 | spin_unlock(&block_rsv->lock); | |
3570 | } | |
3571 | ||
3572 | void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv, | |
3573 | struct btrfs_block_rsv *dest, u64 num_bytes) | |
3574 | { | |
3575 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
3576 | ||
3577 | spin_lock(&block_rsv->lock); | |
3578 | if (num_bytes == (u64)-1) | |
3579 | num_bytes = block_rsv->size; | |
3580 | block_rsv->size -= num_bytes; | |
3581 | if (block_rsv->reserved >= block_rsv->size) { | |
3582 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3583 | block_rsv->reserved = block_rsv->size; | |
3584 | block_rsv->full = 1; | |
3585 | } else { | |
3586 | num_bytes = 0; | |
3587 | } | |
3588 | spin_unlock(&block_rsv->lock); | |
3589 | ||
3590 | if (num_bytes > 0) { | |
3591 | if (dest) { | |
e9e22899 JB |
3592 | spin_lock(&dest->lock); |
3593 | if (!dest->full) { | |
3594 | u64 bytes_to_add; | |
3595 | ||
3596 | bytes_to_add = dest->size - dest->reserved; | |
3597 | bytes_to_add = min(num_bytes, bytes_to_add); | |
3598 | dest->reserved += bytes_to_add; | |
3599 | if (dest->reserved >= dest->size) | |
3600 | dest->full = 1; | |
3601 | num_bytes -= bytes_to_add; | |
3602 | } | |
3603 | spin_unlock(&dest->lock); | |
3604 | } | |
3605 | if (num_bytes) { | |
f0486c68 YZ |
3606 | spin_lock(&space_info->lock); |
3607 | space_info->bytes_reserved -= num_bytes; | |
3608 | spin_unlock(&space_info->lock); | |
4e06bdd6 | 3609 | } |
9ed74f2d | 3610 | } |
f0486c68 | 3611 | } |
4e06bdd6 | 3612 | |
f0486c68 YZ |
3613 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
3614 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
3615 | { | |
3616 | int ret; | |
9ed74f2d | 3617 | |
f0486c68 YZ |
3618 | ret = block_rsv_use_bytes(src, num_bytes); |
3619 | if (ret) | |
3620 | return ret; | |
9ed74f2d | 3621 | |
f0486c68 | 3622 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
3623 | return 0; |
3624 | } | |
3625 | ||
f0486c68 | 3626 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv) |
9ed74f2d | 3627 | { |
f0486c68 YZ |
3628 | memset(rsv, 0, sizeof(*rsv)); |
3629 | spin_lock_init(&rsv->lock); | |
3630 | atomic_set(&rsv->usage, 1); | |
3631 | rsv->priority = 6; | |
3632 | INIT_LIST_HEAD(&rsv->list); | |
3633 | } | |
3634 | ||
3635 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root) | |
3636 | { | |
3637 | struct btrfs_block_rsv *block_rsv; | |
3638 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 3639 | |
f0486c68 YZ |
3640 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
3641 | if (!block_rsv) | |
3642 | return NULL; | |
9ed74f2d | 3643 | |
f0486c68 | 3644 | btrfs_init_block_rsv(block_rsv); |
f0486c68 YZ |
3645 | block_rsv->space_info = __find_space_info(fs_info, |
3646 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
3647 | return block_rsv; |
3648 | } | |
9ed74f2d | 3649 | |
f0486c68 YZ |
3650 | void btrfs_free_block_rsv(struct btrfs_root *root, |
3651 | struct btrfs_block_rsv *rsv) | |
3652 | { | |
3653 | if (rsv && atomic_dec_and_test(&rsv->usage)) { | |
3654 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
3655 | if (!rsv->durable) | |
3656 | kfree(rsv); | |
3657 | } | |
9ed74f2d JB |
3658 | } |
3659 | ||
3660 | /* | |
f0486c68 YZ |
3661 | * make the block_rsv struct be able to capture freed space. |
3662 | * the captured space will re-add to the the block_rsv struct | |
3663 | * after transaction commit | |
9ed74f2d | 3664 | */ |
f0486c68 YZ |
3665 | void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info, |
3666 | struct btrfs_block_rsv *block_rsv) | |
9ed74f2d | 3667 | { |
f0486c68 YZ |
3668 | block_rsv->durable = 1; |
3669 | mutex_lock(&fs_info->durable_block_rsv_mutex); | |
3670 | list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list); | |
3671 | mutex_unlock(&fs_info->durable_block_rsv_mutex); | |
3672 | } | |
9ed74f2d | 3673 | |
f0486c68 YZ |
3674 | int btrfs_block_rsv_add(struct btrfs_trans_handle *trans, |
3675 | struct btrfs_root *root, | |
3676 | struct btrfs_block_rsv *block_rsv, | |
8bb8ab2e | 3677 | u64 num_bytes) |
f0486c68 YZ |
3678 | { |
3679 | int ret; | |
9ed74f2d | 3680 | |
f0486c68 YZ |
3681 | if (num_bytes == 0) |
3682 | return 0; | |
8bb8ab2e JB |
3683 | |
3684 | ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1); | |
f0486c68 YZ |
3685 | if (!ret) { |
3686 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
3687 | return 0; | |
3688 | } | |
9ed74f2d | 3689 | |
f0486c68 YZ |
3690 | return ret; |
3691 | } | |
9ed74f2d | 3692 | |
f0486c68 YZ |
3693 | int btrfs_block_rsv_check(struct btrfs_trans_handle *trans, |
3694 | struct btrfs_root *root, | |
3695 | struct btrfs_block_rsv *block_rsv, | |
3696 | u64 min_reserved, int min_factor) | |
3697 | { | |
3698 | u64 num_bytes = 0; | |
3699 | int commit_trans = 0; | |
3700 | int ret = -ENOSPC; | |
9ed74f2d | 3701 | |
f0486c68 YZ |
3702 | if (!block_rsv) |
3703 | return 0; | |
9ed74f2d | 3704 | |
f0486c68 YZ |
3705 | spin_lock(&block_rsv->lock); |
3706 | if (min_factor > 0) | |
3707 | num_bytes = div_factor(block_rsv->size, min_factor); | |
3708 | if (min_reserved > num_bytes) | |
3709 | num_bytes = min_reserved; | |
9ed74f2d | 3710 | |
f0486c68 YZ |
3711 | if (block_rsv->reserved >= num_bytes) { |
3712 | ret = 0; | |
3713 | } else { | |
3714 | num_bytes -= block_rsv->reserved; | |
3715 | if (block_rsv->durable && | |
3716 | block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes) | |
3717 | commit_trans = 1; | |
3718 | } | |
3719 | spin_unlock(&block_rsv->lock); | |
3720 | if (!ret) | |
3721 | return 0; | |
3722 | ||
3723 | if (block_rsv->refill_used) { | |
8bb8ab2e JB |
3724 | ret = reserve_metadata_bytes(trans, root, block_rsv, |
3725 | num_bytes, 0); | |
f0486c68 YZ |
3726 | if (!ret) { |
3727 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
3728 | return 0; | |
4e06bdd6 | 3729 | } |
f0486c68 | 3730 | } |
9ed74f2d | 3731 | |
f0486c68 YZ |
3732 | if (commit_trans) { |
3733 | if (trans) | |
3734 | return -EAGAIN; | |
3735 | ||
3736 | trans = btrfs_join_transaction(root, 1); | |
3737 | BUG_ON(IS_ERR(trans)); | |
3738 | ret = btrfs_commit_transaction(trans, root); | |
3739 | return 0; | |
6a63209f | 3740 | } |
9ed74f2d | 3741 | |
f0486c68 YZ |
3742 | return -ENOSPC; |
3743 | } | |
3744 | ||
3745 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
3746 | struct btrfs_block_rsv *dst_rsv, | |
3747 | u64 num_bytes) | |
3748 | { | |
3749 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
3750 | } | |
3751 | ||
3752 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
3753 | struct btrfs_block_rsv *block_rsv, | |
3754 | u64 num_bytes) | |
3755 | { | |
3756 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
3757 | if (global_rsv->full || global_rsv == block_rsv || | |
3758 | block_rsv->space_info != global_rsv->space_info) | |
3759 | global_rsv = NULL; | |
3760 | block_rsv_release_bytes(block_rsv, global_rsv, num_bytes); | |
6a63209f JB |
3761 | } |
3762 | ||
3763 | /* | |
8929ecfa YZ |
3764 | * helper to calculate size of global block reservation. |
3765 | * the desired value is sum of space used by extent tree, | |
3766 | * checksum tree and root tree | |
6a63209f | 3767 | */ |
8929ecfa | 3768 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 3769 | { |
8929ecfa YZ |
3770 | struct btrfs_space_info *sinfo; |
3771 | u64 num_bytes; | |
3772 | u64 meta_used; | |
3773 | u64 data_used; | |
3774 | int csum_size = btrfs_super_csum_size(&fs_info->super_copy); | |
3775 | #if 0 | |
3776 | /* | |
3777 | * per tree used space accounting can be inaccuracy, so we | |
3778 | * can't rely on it. | |
3779 | */ | |
3780 | spin_lock(&fs_info->extent_root->accounting_lock); | |
3781 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item); | |
3782 | spin_unlock(&fs_info->extent_root->accounting_lock); | |
6a63209f | 3783 | |
8929ecfa YZ |
3784 | spin_lock(&fs_info->csum_root->accounting_lock); |
3785 | num_bytes += btrfs_root_used(&fs_info->csum_root->root_item); | |
3786 | spin_unlock(&fs_info->csum_root->accounting_lock); | |
6a63209f | 3787 | |
8929ecfa YZ |
3788 | spin_lock(&fs_info->tree_root->accounting_lock); |
3789 | num_bytes += btrfs_root_used(&fs_info->tree_root->root_item); | |
3790 | spin_unlock(&fs_info->tree_root->accounting_lock); | |
3791 | #endif | |
3792 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
3793 | spin_lock(&sinfo->lock); | |
3794 | data_used = sinfo->bytes_used; | |
3795 | spin_unlock(&sinfo->lock); | |
33b4d47f | 3796 | |
8929ecfa YZ |
3797 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
3798 | spin_lock(&sinfo->lock); | |
6d48755d JB |
3799 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
3800 | data_used = 0; | |
8929ecfa YZ |
3801 | meta_used = sinfo->bytes_used; |
3802 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 3803 | |
8929ecfa YZ |
3804 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
3805 | csum_size * 2; | |
3806 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 3807 | |
8929ecfa YZ |
3808 | if (num_bytes * 3 > meta_used) |
3809 | num_bytes = div64_u64(meta_used, 3); | |
ab6e2410 | 3810 | |
8929ecfa YZ |
3811 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
3812 | } | |
6a63209f | 3813 | |
8929ecfa YZ |
3814 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
3815 | { | |
3816 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
3817 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
3818 | u64 num_bytes; | |
6a63209f | 3819 | |
8929ecfa | 3820 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 3821 | |
8929ecfa YZ |
3822 | spin_lock(&block_rsv->lock); |
3823 | spin_lock(&sinfo->lock); | |
4e06bdd6 | 3824 | |
8929ecfa | 3825 | block_rsv->size = num_bytes; |
4e06bdd6 | 3826 | |
8929ecfa | 3827 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
3828 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
3829 | sinfo->bytes_may_use; | |
8929ecfa YZ |
3830 | |
3831 | if (sinfo->total_bytes > num_bytes) { | |
3832 | num_bytes = sinfo->total_bytes - num_bytes; | |
3833 | block_rsv->reserved += num_bytes; | |
3834 | sinfo->bytes_reserved += num_bytes; | |
6a63209f | 3835 | } |
6a63209f | 3836 | |
8929ecfa YZ |
3837 | if (block_rsv->reserved >= block_rsv->size) { |
3838 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3839 | sinfo->bytes_reserved -= num_bytes; | |
3840 | block_rsv->reserved = block_rsv->size; | |
3841 | block_rsv->full = 1; | |
3842 | } | |
3843 | #if 0 | |
3844 | printk(KERN_INFO"global block rsv size %llu reserved %llu\n", | |
3845 | block_rsv->size, block_rsv->reserved); | |
3846 | #endif | |
3847 | spin_unlock(&sinfo->lock); | |
3848 | spin_unlock(&block_rsv->lock); | |
6a63209f JB |
3849 | } |
3850 | ||
f0486c68 | 3851 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 3852 | { |
f0486c68 | 3853 | struct btrfs_space_info *space_info; |
6a63209f | 3854 | |
f0486c68 YZ |
3855 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
3856 | fs_info->chunk_block_rsv.space_info = space_info; | |
3857 | fs_info->chunk_block_rsv.priority = 10; | |
6a63209f | 3858 | |
f0486c68 | 3859 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa YZ |
3860 | fs_info->global_block_rsv.space_info = space_info; |
3861 | fs_info->global_block_rsv.priority = 10; | |
3862 | fs_info->global_block_rsv.refill_used = 1; | |
3863 | fs_info->delalloc_block_rsv.space_info = space_info; | |
f0486c68 YZ |
3864 | fs_info->trans_block_rsv.space_info = space_info; |
3865 | fs_info->empty_block_rsv.space_info = space_info; | |
3866 | fs_info->empty_block_rsv.priority = 10; | |
3867 | ||
8929ecfa YZ |
3868 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
3869 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
3870 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
3871 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 3872 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa YZ |
3873 | |
3874 | btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv); | |
3875 | ||
3876 | btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv); | |
3877 | ||
3878 | update_global_block_rsv(fs_info); | |
6a63209f JB |
3879 | } |
3880 | ||
8929ecfa | 3881 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 3882 | { |
8929ecfa YZ |
3883 | block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1); |
3884 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); | |
3885 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
3886 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
3887 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
3888 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
3889 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
f0486c68 | 3890 | } |
6a63209f | 3891 | |
a22285a6 YZ |
3892 | static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items) |
3893 | { | |
3894 | return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) * | |
3895 | 3 * num_items; | |
3896 | } | |
6a63209f | 3897 | |
a22285a6 YZ |
3898 | int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans, |
3899 | struct btrfs_root *root, | |
8bb8ab2e | 3900 | int num_items) |
a22285a6 YZ |
3901 | { |
3902 | u64 num_bytes; | |
3903 | int ret; | |
6a63209f | 3904 | |
a22285a6 YZ |
3905 | if (num_items == 0 || root->fs_info->chunk_root == root) |
3906 | return 0; | |
6a63209f | 3907 | |
a22285a6 YZ |
3908 | num_bytes = calc_trans_metadata_size(root, num_items); |
3909 | ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv, | |
8bb8ab2e | 3910 | num_bytes); |
a22285a6 YZ |
3911 | if (!ret) { |
3912 | trans->bytes_reserved += num_bytes; | |
3913 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
3914 | } | |
3915 | return ret; | |
6a63209f JB |
3916 | } |
3917 | ||
a22285a6 YZ |
3918 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
3919 | struct btrfs_root *root) | |
6a63209f | 3920 | { |
a22285a6 YZ |
3921 | if (!trans->bytes_reserved) |
3922 | return; | |
6a63209f | 3923 | |
a22285a6 YZ |
3924 | BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv); |
3925 | btrfs_block_rsv_release(root, trans->block_rsv, | |
3926 | trans->bytes_reserved); | |
3927 | trans->bytes_reserved = 0; | |
3928 | } | |
6a63209f | 3929 | |
d68fc57b YZ |
3930 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
3931 | struct inode *inode) | |
3932 | { | |
3933 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3934 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
3935 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
3936 | ||
3937 | /* | |
3938 | * one for deleting orphan item, one for updating inode and | |
3939 | * two for calling btrfs_truncate_inode_items. | |
3940 | * | |
3941 | * btrfs_truncate_inode_items is a delete operation, it frees | |
3942 | * more space than it uses in most cases. So two units of | |
3943 | * metadata space should be enough for calling it many times. | |
3944 | * If all of the metadata space is used, we can commit | |
3945 | * transaction and use space it freed. | |
3946 | */ | |
3947 | u64 num_bytes = calc_trans_metadata_size(root, 4); | |
3948 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
6a63209f JB |
3949 | } |
3950 | ||
d68fc57b | 3951 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 3952 | { |
d68fc57b YZ |
3953 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3954 | u64 num_bytes = calc_trans_metadata_size(root, 4); | |
3955 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); | |
3956 | } | |
97e728d4 | 3957 | |
a22285a6 YZ |
3958 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
3959 | struct btrfs_pending_snapshot *pending) | |
3960 | { | |
3961 | struct btrfs_root *root = pending->root; | |
3962 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
3963 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
3964 | /* | |
3965 | * two for root back/forward refs, two for directory entries | |
3966 | * and one for root of the snapshot. | |
3967 | */ | |
3968 | u64 num_bytes = calc_trans_metadata_size(root, 5); | |
3969 | dst_rsv->space_info = src_rsv->space_info; | |
3970 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
3971 | } |
3972 | ||
0ca1f7ce | 3973 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes) |
6324fbf3 | 3974 | { |
0ca1f7ce YZ |
3975 | return num_bytes >>= 3; |
3976 | } | |
c146afad | 3977 | |
0ca1f7ce YZ |
3978 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
3979 | { | |
3980 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3981 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
3982 | u64 to_reserve; | |
3983 | int nr_extents; | |
0ca1f7ce | 3984 | int ret; |
6324fbf3 | 3985 | |
0ca1f7ce YZ |
3986 | if (btrfs_transaction_in_commit(root->fs_info)) |
3987 | schedule_timeout(1); | |
ec44a35c | 3988 | |
0ca1f7ce | 3989 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 3990 | |
0ca1f7ce YZ |
3991 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
3992 | nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1; | |
3993 | if (nr_extents > BTRFS_I(inode)->reserved_extents) { | |
3994 | nr_extents -= BTRFS_I(inode)->reserved_extents; | |
3995 | to_reserve = calc_trans_metadata_size(root, nr_extents); | |
3996 | } else { | |
3997 | nr_extents = 0; | |
3998 | to_reserve = 0; | |
593060d7 | 3999 | } |
8bb8ab2e | 4000 | spin_unlock(&BTRFS_I(inode)->accounting_lock); |
6324fbf3 | 4001 | |
0ca1f7ce | 4002 | to_reserve += calc_csum_metadata_size(inode, num_bytes); |
8bb8ab2e JB |
4003 | ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1); |
4004 | if (ret) | |
0ca1f7ce | 4005 | return ret; |
6324fbf3 | 4006 | |
8bb8ab2e | 4007 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
0ca1f7ce YZ |
4008 | BTRFS_I(inode)->reserved_extents += nr_extents; |
4009 | atomic_inc(&BTRFS_I(inode)->outstanding_extents); | |
4010 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
25179201 | 4011 | |
0ca1f7ce YZ |
4012 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4013 | ||
4014 | if (block_rsv->size > 512 * 1024 * 1024) | |
0019f10d | 4015 | shrink_delalloc(NULL, root, to_reserve, 0); |
0ca1f7ce YZ |
4016 | |
4017 | return 0; | |
4018 | } | |
4019 | ||
4020 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) | |
4021 | { | |
4022 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4023 | u64 to_free; | |
4024 | int nr_extents; | |
4025 | ||
4026 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
4027 | atomic_dec(&BTRFS_I(inode)->outstanding_extents); | |
4028 | ||
4029 | spin_lock(&BTRFS_I(inode)->accounting_lock); | |
4030 | nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents); | |
4031 | if (nr_extents < BTRFS_I(inode)->reserved_extents) { | |
4032 | nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents; | |
4033 | BTRFS_I(inode)->reserved_extents -= nr_extents; | |
4034 | } else { | |
4035 | nr_extents = 0; | |
97e728d4 | 4036 | } |
0ca1f7ce | 4037 | spin_unlock(&BTRFS_I(inode)->accounting_lock); |
97e728d4 | 4038 | |
0ca1f7ce YZ |
4039 | to_free = calc_csum_metadata_size(inode, num_bytes); |
4040 | if (nr_extents > 0) | |
4041 | to_free += calc_trans_metadata_size(root, nr_extents); | |
4042 | ||
4043 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, | |
4044 | to_free); | |
4045 | } | |
4046 | ||
4047 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) | |
4048 | { | |
4049 | int ret; | |
4050 | ||
4051 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4052 | if (ret) |
0ca1f7ce YZ |
4053 | return ret; |
4054 | ||
4055 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4056 | if (ret) { | |
4057 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4058 | return ret; | |
4059 | } | |
4060 | ||
4061 | return 0; | |
4062 | } | |
4063 | ||
4064 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) | |
4065 | { | |
4066 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4067 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4068 | } |
4069 | ||
9078a3e1 CM |
4070 | static int update_block_group(struct btrfs_trans_handle *trans, |
4071 | struct btrfs_root *root, | |
f0486c68 | 4072 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4073 | { |
0af3d00b | 4074 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4075 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4076 | u64 total = num_bytes; |
9078a3e1 | 4077 | u64 old_val; |
db94535d | 4078 | u64 byte_in_group; |
0af3d00b | 4079 | int factor; |
3e1ad54f | 4080 | |
5d4f98a2 YZ |
4081 | /* block accounting for super block */ |
4082 | spin_lock(&info->delalloc_lock); | |
4083 | old_val = btrfs_super_bytes_used(&info->super_copy); | |
4084 | if (alloc) | |
4085 | old_val += num_bytes; | |
4086 | else | |
4087 | old_val -= num_bytes; | |
4088 | btrfs_set_super_bytes_used(&info->super_copy, old_val); | |
5d4f98a2 YZ |
4089 | spin_unlock(&info->delalloc_lock); |
4090 | ||
d397712b | 4091 | while (total) { |
db94535d | 4092 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4093 | if (!cache) |
9078a3e1 | 4094 | return -1; |
b742bb82 YZ |
4095 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4096 | BTRFS_BLOCK_GROUP_RAID1 | | |
4097 | BTRFS_BLOCK_GROUP_RAID10)) | |
4098 | factor = 2; | |
4099 | else | |
4100 | factor = 1; | |
9d66e233 JB |
4101 | /* |
4102 | * If this block group has free space cache written out, we | |
4103 | * need to make sure to load it if we are removing space. This | |
4104 | * is because we need the unpinning stage to actually add the | |
4105 | * space back to the block group, otherwise we will leak space. | |
4106 | */ | |
4107 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4108 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4109 | |
db94535d CM |
4110 | byte_in_group = bytenr - cache->key.objectid; |
4111 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4112 | |
25179201 | 4113 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4114 | spin_lock(&cache->lock); |
0af3d00b JB |
4115 | |
4116 | if (btrfs_super_cache_generation(&info->super_copy) != 0 && | |
4117 | cache->disk_cache_state < BTRFS_DC_CLEAR) | |
4118 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4119 | ||
0f9dd46c | 4120 | cache->dirty = 1; |
9078a3e1 | 4121 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4122 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4123 | if (alloc) { |
db94535d | 4124 | old_val += num_bytes; |
11833d66 YZ |
4125 | btrfs_set_block_group_used(&cache->item, old_val); |
4126 | cache->reserved -= num_bytes; | |
11833d66 | 4127 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
4128 | cache->space_info->bytes_used += num_bytes; |
4129 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4130 | spin_unlock(&cache->lock); |
25179201 | 4131 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4132 | } else { |
db94535d | 4133 | old_val -= num_bytes; |
c286ac48 | 4134 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4135 | cache->pinned += num_bytes; |
4136 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4137 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4138 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4139 | spin_unlock(&cache->lock); |
25179201 | 4140 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4141 | |
f0486c68 YZ |
4142 | set_extent_dirty(info->pinned_extents, |
4143 | bytenr, bytenr + num_bytes - 1, | |
4144 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4145 | } |
fa9c0d79 | 4146 | btrfs_put_block_group(cache); |
db94535d CM |
4147 | total -= num_bytes; |
4148 | bytenr += num_bytes; | |
9078a3e1 CM |
4149 | } |
4150 | return 0; | |
4151 | } | |
6324fbf3 | 4152 | |
a061fc8d CM |
4153 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4154 | { | |
0f9dd46c | 4155 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4156 | u64 bytenr; |
0f9dd46c JB |
4157 | |
4158 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4159 | if (!cache) | |
a061fc8d | 4160 | return 0; |
0f9dd46c | 4161 | |
d2fb3437 | 4162 | bytenr = cache->key.objectid; |
fa9c0d79 | 4163 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4164 | |
4165 | return bytenr; | |
a061fc8d CM |
4166 | } |
4167 | ||
f0486c68 YZ |
4168 | static int pin_down_extent(struct btrfs_root *root, |
4169 | struct btrfs_block_group_cache *cache, | |
4170 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4171 | { |
11833d66 YZ |
4172 | spin_lock(&cache->space_info->lock); |
4173 | spin_lock(&cache->lock); | |
4174 | cache->pinned += num_bytes; | |
4175 | cache->space_info->bytes_pinned += num_bytes; | |
4176 | if (reserved) { | |
4177 | cache->reserved -= num_bytes; | |
4178 | cache->space_info->bytes_reserved -= num_bytes; | |
4179 | } | |
4180 | spin_unlock(&cache->lock); | |
4181 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4182 | |
f0486c68 YZ |
4183 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4184 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4185 | return 0; | |
4186 | } | |
68b38550 | 4187 | |
f0486c68 YZ |
4188 | /* |
4189 | * this function must be called within transaction | |
4190 | */ | |
4191 | int btrfs_pin_extent(struct btrfs_root *root, | |
4192 | u64 bytenr, u64 num_bytes, int reserved) | |
4193 | { | |
4194 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4195 | |
f0486c68 YZ |
4196 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
4197 | BUG_ON(!cache); | |
4198 | ||
4199 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4200 | ||
4201 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4202 | return 0; |
4203 | } | |
4204 | ||
f0486c68 YZ |
4205 | /* |
4206 | * update size of reserved extents. this function may return -EAGAIN | |
4207 | * if 'reserve' is true or 'sinfo' is false. | |
4208 | */ | |
4209 | static int update_reserved_bytes(struct btrfs_block_group_cache *cache, | |
4210 | u64 num_bytes, int reserve, int sinfo) | |
11833d66 | 4211 | { |
f0486c68 YZ |
4212 | int ret = 0; |
4213 | if (sinfo) { | |
4214 | struct btrfs_space_info *space_info = cache->space_info; | |
4215 | spin_lock(&space_info->lock); | |
4216 | spin_lock(&cache->lock); | |
4217 | if (reserve) { | |
4218 | if (cache->ro) { | |
4219 | ret = -EAGAIN; | |
4220 | } else { | |
4221 | cache->reserved += num_bytes; | |
4222 | space_info->bytes_reserved += num_bytes; | |
4223 | } | |
4224 | } else { | |
4225 | if (cache->ro) | |
4226 | space_info->bytes_readonly += num_bytes; | |
4227 | cache->reserved -= num_bytes; | |
4228 | space_info->bytes_reserved -= num_bytes; | |
4229 | } | |
4230 | spin_unlock(&cache->lock); | |
4231 | spin_unlock(&space_info->lock); | |
11833d66 | 4232 | } else { |
f0486c68 YZ |
4233 | spin_lock(&cache->lock); |
4234 | if (cache->ro) { | |
4235 | ret = -EAGAIN; | |
4236 | } else { | |
4237 | if (reserve) | |
4238 | cache->reserved += num_bytes; | |
4239 | else | |
4240 | cache->reserved -= num_bytes; | |
4241 | } | |
4242 | spin_unlock(&cache->lock); | |
324ae4df | 4243 | } |
f0486c68 | 4244 | return ret; |
324ae4df | 4245 | } |
9078a3e1 | 4246 | |
11833d66 YZ |
4247 | int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
4248 | struct btrfs_root *root) | |
e8569813 | 4249 | { |
e8569813 | 4250 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4251 | struct btrfs_caching_control *next; |
4252 | struct btrfs_caching_control *caching_ctl; | |
4253 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4254 | |
11833d66 | 4255 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4256 | |
11833d66 YZ |
4257 | list_for_each_entry_safe(caching_ctl, next, |
4258 | &fs_info->caching_block_groups, list) { | |
4259 | cache = caching_ctl->block_group; | |
4260 | if (block_group_cache_done(cache)) { | |
4261 | cache->last_byte_to_unpin = (u64)-1; | |
4262 | list_del_init(&caching_ctl->list); | |
4263 | put_caching_control(caching_ctl); | |
e8569813 | 4264 | } else { |
11833d66 | 4265 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 4266 | } |
e8569813 | 4267 | } |
11833d66 YZ |
4268 | |
4269 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
4270 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
4271 | else | |
4272 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
4273 | ||
4274 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
4275 | |
4276 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
4277 | return 0; |
4278 | } | |
4279 | ||
11833d66 | 4280 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 4281 | { |
11833d66 YZ |
4282 | struct btrfs_fs_info *fs_info = root->fs_info; |
4283 | struct btrfs_block_group_cache *cache = NULL; | |
4284 | u64 len; | |
ccd467d6 | 4285 | |
11833d66 YZ |
4286 | while (start <= end) { |
4287 | if (!cache || | |
4288 | start >= cache->key.objectid + cache->key.offset) { | |
4289 | if (cache) | |
4290 | btrfs_put_block_group(cache); | |
4291 | cache = btrfs_lookup_block_group(fs_info, start); | |
4292 | BUG_ON(!cache); | |
4293 | } | |
4294 | ||
4295 | len = cache->key.objectid + cache->key.offset - start; | |
4296 | len = min(len, end + 1 - start); | |
4297 | ||
4298 | if (start < cache->last_byte_to_unpin) { | |
4299 | len = min(len, cache->last_byte_to_unpin - start); | |
4300 | btrfs_add_free_space(cache, start, len); | |
4301 | } | |
4302 | ||
f0486c68 YZ |
4303 | start += len; |
4304 | ||
11833d66 YZ |
4305 | spin_lock(&cache->space_info->lock); |
4306 | spin_lock(&cache->lock); | |
4307 | cache->pinned -= len; | |
4308 | cache->space_info->bytes_pinned -= len; | |
f0486c68 YZ |
4309 | if (cache->ro) { |
4310 | cache->space_info->bytes_readonly += len; | |
4311 | } else if (cache->reserved_pinned > 0) { | |
4312 | len = min(len, cache->reserved_pinned); | |
4313 | cache->reserved_pinned -= len; | |
4314 | cache->space_info->bytes_reserved += len; | |
4315 | } | |
11833d66 YZ |
4316 | spin_unlock(&cache->lock); |
4317 | spin_unlock(&cache->space_info->lock); | |
ccd467d6 | 4318 | } |
11833d66 YZ |
4319 | |
4320 | if (cache) | |
4321 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
4322 | return 0; |
4323 | } | |
4324 | ||
4325 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 4326 | struct btrfs_root *root) |
a28ec197 | 4327 | { |
11833d66 YZ |
4328 | struct btrfs_fs_info *fs_info = root->fs_info; |
4329 | struct extent_io_tree *unpin; | |
f0486c68 YZ |
4330 | struct btrfs_block_rsv *block_rsv; |
4331 | struct btrfs_block_rsv *next_rsv; | |
1a5bc167 CM |
4332 | u64 start; |
4333 | u64 end; | |
f0486c68 | 4334 | int idx; |
a28ec197 | 4335 | int ret; |
a28ec197 | 4336 | |
11833d66 YZ |
4337 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
4338 | unpin = &fs_info->freed_extents[1]; | |
4339 | else | |
4340 | unpin = &fs_info->freed_extents[0]; | |
4341 | ||
d397712b | 4342 | while (1) { |
1a5bc167 CM |
4343 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
4344 | EXTENT_DIRTY); | |
4345 | if (ret) | |
a28ec197 | 4346 | break; |
1f3c79a2 LH |
4347 | |
4348 | ret = btrfs_discard_extent(root, start, end + 1 - start); | |
4349 | ||
1a5bc167 | 4350 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 4351 | unpin_extent_range(root, start, end); |
b9473439 | 4352 | cond_resched(); |
a28ec197 | 4353 | } |
817d52f8 | 4354 | |
f0486c68 YZ |
4355 | mutex_lock(&fs_info->durable_block_rsv_mutex); |
4356 | list_for_each_entry_safe(block_rsv, next_rsv, | |
4357 | &fs_info->durable_block_rsv_list, list) { | |
444528b3 | 4358 | |
f0486c68 YZ |
4359 | idx = trans->transid & 0x1; |
4360 | if (block_rsv->freed[idx] > 0) { | |
4361 | block_rsv_add_bytes(block_rsv, | |
4362 | block_rsv->freed[idx], 0); | |
4363 | block_rsv->freed[idx] = 0; | |
4364 | } | |
4365 | if (atomic_read(&block_rsv->usage) == 0) { | |
4366 | btrfs_block_rsv_release(root, block_rsv, (u64)-1); | |
31840ae1 | 4367 | |
f0486c68 YZ |
4368 | if (block_rsv->freed[0] == 0 && |
4369 | block_rsv->freed[1] == 0) { | |
4370 | list_del_init(&block_rsv->list); | |
4371 | kfree(block_rsv); | |
4372 | } | |
4373 | } else { | |
4374 | btrfs_block_rsv_release(root, block_rsv, 0); | |
8ef97622 | 4375 | } |
f4b9aa8d | 4376 | } |
f0486c68 | 4377 | mutex_unlock(&fs_info->durable_block_rsv_mutex); |
31840ae1 | 4378 | |
e20d96d6 CM |
4379 | return 0; |
4380 | } | |
4381 | ||
5d4f98a2 YZ |
4382 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
4383 | struct btrfs_root *root, | |
4384 | u64 bytenr, u64 num_bytes, u64 parent, | |
4385 | u64 root_objectid, u64 owner_objectid, | |
4386 | u64 owner_offset, int refs_to_drop, | |
4387 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 4388 | { |
e2fa7227 | 4389 | struct btrfs_key key; |
5d4f98a2 | 4390 | struct btrfs_path *path; |
1261ec42 CM |
4391 | struct btrfs_fs_info *info = root->fs_info; |
4392 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 4393 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
4394 | struct btrfs_extent_item *ei; |
4395 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 4396 | int ret; |
5d4f98a2 | 4397 | int is_data; |
952fccac CM |
4398 | int extent_slot = 0; |
4399 | int found_extent = 0; | |
4400 | int num_to_del = 1; | |
5d4f98a2 YZ |
4401 | u32 item_size; |
4402 | u64 refs; | |
037e6390 | 4403 | |
5caf2a00 | 4404 | path = btrfs_alloc_path(); |
54aa1f4d CM |
4405 | if (!path) |
4406 | return -ENOMEM; | |
5f26f772 | 4407 | |
3c12ac72 | 4408 | path->reada = 1; |
b9473439 | 4409 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4410 | |
4411 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
4412 | BUG_ON(!is_data && refs_to_drop != 1); | |
4413 | ||
4414 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
4415 | bytenr, num_bytes, parent, | |
4416 | root_objectid, owner_objectid, | |
4417 | owner_offset); | |
7bb86316 | 4418 | if (ret == 0) { |
952fccac | 4419 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
4420 | while (extent_slot >= 0) { |
4421 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 4422 | extent_slot); |
5d4f98a2 | 4423 | if (key.objectid != bytenr) |
952fccac | 4424 | break; |
5d4f98a2 YZ |
4425 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
4426 | key.offset == num_bytes) { | |
952fccac CM |
4427 | found_extent = 1; |
4428 | break; | |
4429 | } | |
4430 | if (path->slots[0] - extent_slot > 5) | |
4431 | break; | |
5d4f98a2 | 4432 | extent_slot--; |
952fccac | 4433 | } |
5d4f98a2 YZ |
4434 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
4435 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
4436 | if (found_extent && item_size < sizeof(*ei)) | |
4437 | found_extent = 0; | |
4438 | #endif | |
31840ae1 | 4439 | if (!found_extent) { |
5d4f98a2 | 4440 | BUG_ON(iref); |
56bec294 | 4441 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
4442 | NULL, refs_to_drop, |
4443 | is_data); | |
31840ae1 ZY |
4444 | BUG_ON(ret); |
4445 | btrfs_release_path(extent_root, path); | |
b9473439 | 4446 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4447 | |
4448 | key.objectid = bytenr; | |
4449 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4450 | key.offset = num_bytes; | |
4451 | ||
31840ae1 ZY |
4452 | ret = btrfs_search_slot(trans, extent_root, |
4453 | &key, path, -1, 1); | |
f3465ca4 JB |
4454 | if (ret) { |
4455 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
4456 | ", was looking for %llu\n", ret, |
4457 | (unsigned long long)bytenr); | |
f3465ca4 JB |
4458 | btrfs_print_leaf(extent_root, path->nodes[0]); |
4459 | } | |
31840ae1 ZY |
4460 | BUG_ON(ret); |
4461 | extent_slot = path->slots[0]; | |
4462 | } | |
7bb86316 CM |
4463 | } else { |
4464 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4465 | WARN_ON(1); | |
d397712b | 4466 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 4467 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 4468 | (unsigned long long)bytenr, |
56bec294 | 4469 | (unsigned long long)parent, |
d397712b | 4470 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
4471 | (unsigned long long)owner_objectid, |
4472 | (unsigned long long)owner_offset); | |
7bb86316 | 4473 | } |
5f39d397 CM |
4474 | |
4475 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
4476 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
4477 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
4478 | if (item_size < sizeof(*ei)) { | |
4479 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
4480 | ret = convert_extent_item_v0(trans, extent_root, path, | |
4481 | owner_objectid, 0); | |
4482 | BUG_ON(ret < 0); | |
4483 | ||
4484 | btrfs_release_path(extent_root, path); | |
4485 | path->leave_spinning = 1; | |
4486 | ||
4487 | key.objectid = bytenr; | |
4488 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4489 | key.offset = num_bytes; | |
4490 | ||
4491 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
4492 | -1, 1); | |
4493 | if (ret) { | |
4494 | printk(KERN_ERR "umm, got %d back from search" | |
4495 | ", was looking for %llu\n", ret, | |
4496 | (unsigned long long)bytenr); | |
4497 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4498 | } | |
4499 | BUG_ON(ret); | |
4500 | extent_slot = path->slots[0]; | |
4501 | leaf = path->nodes[0]; | |
4502 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
4503 | } | |
4504 | #endif | |
4505 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 4506 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 4507 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4508 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
4509 | struct btrfs_tree_block_info *bi; | |
4510 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
4511 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
4512 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
4513 | } | |
56bec294 | 4514 | |
5d4f98a2 | 4515 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
4516 | BUG_ON(refs < refs_to_drop); |
4517 | refs -= refs_to_drop; | |
5f39d397 | 4518 | |
5d4f98a2 YZ |
4519 | if (refs > 0) { |
4520 | if (extent_op) | |
4521 | __run_delayed_extent_op(extent_op, leaf, ei); | |
4522 | /* | |
4523 | * In the case of inline back ref, reference count will | |
4524 | * be updated by remove_extent_backref | |
952fccac | 4525 | */ |
5d4f98a2 YZ |
4526 | if (iref) { |
4527 | BUG_ON(!found_extent); | |
4528 | } else { | |
4529 | btrfs_set_extent_refs(leaf, ei, refs); | |
4530 | btrfs_mark_buffer_dirty(leaf); | |
4531 | } | |
4532 | if (found_extent) { | |
4533 | ret = remove_extent_backref(trans, extent_root, path, | |
4534 | iref, refs_to_drop, | |
4535 | is_data); | |
952fccac CM |
4536 | BUG_ON(ret); |
4537 | } | |
5d4f98a2 | 4538 | } else { |
5d4f98a2 YZ |
4539 | if (found_extent) { |
4540 | BUG_ON(is_data && refs_to_drop != | |
4541 | extent_data_ref_count(root, path, iref)); | |
4542 | if (iref) { | |
4543 | BUG_ON(path->slots[0] != extent_slot); | |
4544 | } else { | |
4545 | BUG_ON(path->slots[0] != extent_slot + 1); | |
4546 | path->slots[0] = extent_slot; | |
4547 | num_to_del = 2; | |
4548 | } | |
78fae27e | 4549 | } |
b9473439 | 4550 | |
952fccac CM |
4551 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
4552 | num_to_del); | |
31840ae1 | 4553 | BUG_ON(ret); |
25179201 | 4554 | btrfs_release_path(extent_root, path); |
21af804c | 4555 | |
5d4f98a2 | 4556 | if (is_data) { |
459931ec CM |
4557 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
4558 | BUG_ON(ret); | |
d57e62b8 CM |
4559 | } else { |
4560 | invalidate_mapping_pages(info->btree_inode->i_mapping, | |
4561 | bytenr >> PAGE_CACHE_SHIFT, | |
4562 | (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT); | |
459931ec CM |
4563 | } |
4564 | ||
f0486c68 | 4565 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
dcbdd4dc | 4566 | BUG_ON(ret); |
a28ec197 | 4567 | } |
5caf2a00 | 4568 | btrfs_free_path(path); |
a28ec197 CM |
4569 | return ret; |
4570 | } | |
4571 | ||
1887be66 | 4572 | /* |
f0486c68 | 4573 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
4574 | * delayed ref for that extent as well. This searches the delayed ref tree for |
4575 | * a given extent, and if there are no other delayed refs to be processed, it | |
4576 | * removes it from the tree. | |
4577 | */ | |
4578 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
4579 | struct btrfs_root *root, u64 bytenr) | |
4580 | { | |
4581 | struct btrfs_delayed_ref_head *head; | |
4582 | struct btrfs_delayed_ref_root *delayed_refs; | |
4583 | struct btrfs_delayed_ref_node *ref; | |
4584 | struct rb_node *node; | |
f0486c68 | 4585 | int ret = 0; |
1887be66 CM |
4586 | |
4587 | delayed_refs = &trans->transaction->delayed_refs; | |
4588 | spin_lock(&delayed_refs->lock); | |
4589 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
4590 | if (!head) | |
4591 | goto out; | |
4592 | ||
4593 | node = rb_prev(&head->node.rb_node); | |
4594 | if (!node) | |
4595 | goto out; | |
4596 | ||
4597 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
4598 | ||
4599 | /* there are still entries for this ref, we can't drop it */ | |
4600 | if (ref->bytenr == bytenr) | |
4601 | goto out; | |
4602 | ||
5d4f98a2 YZ |
4603 | if (head->extent_op) { |
4604 | if (!head->must_insert_reserved) | |
4605 | goto out; | |
4606 | kfree(head->extent_op); | |
4607 | head->extent_op = NULL; | |
4608 | } | |
4609 | ||
1887be66 CM |
4610 | /* |
4611 | * waiting for the lock here would deadlock. If someone else has it | |
4612 | * locked they are already in the process of dropping it anyway | |
4613 | */ | |
4614 | if (!mutex_trylock(&head->mutex)) | |
4615 | goto out; | |
4616 | ||
4617 | /* | |
4618 | * at this point we have a head with no other entries. Go | |
4619 | * ahead and process it. | |
4620 | */ | |
4621 | head->node.in_tree = 0; | |
4622 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 4623 | |
1887be66 CM |
4624 | delayed_refs->num_entries--; |
4625 | ||
4626 | /* | |
4627 | * we don't take a ref on the node because we're removing it from the | |
4628 | * tree, so we just steal the ref the tree was holding. | |
4629 | */ | |
c3e69d58 CM |
4630 | delayed_refs->num_heads--; |
4631 | if (list_empty(&head->cluster)) | |
4632 | delayed_refs->num_heads_ready--; | |
4633 | ||
4634 | list_del_init(&head->cluster); | |
1887be66 CM |
4635 | spin_unlock(&delayed_refs->lock); |
4636 | ||
f0486c68 YZ |
4637 | BUG_ON(head->extent_op); |
4638 | if (head->must_insert_reserved) | |
4639 | ret = 1; | |
4640 | ||
4641 | mutex_unlock(&head->mutex); | |
1887be66 | 4642 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 4643 | return ret; |
1887be66 CM |
4644 | out: |
4645 | spin_unlock(&delayed_refs->lock); | |
4646 | return 0; | |
4647 | } | |
4648 | ||
f0486c68 YZ |
4649 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
4650 | struct btrfs_root *root, | |
4651 | struct extent_buffer *buf, | |
4652 | u64 parent, int last_ref) | |
4653 | { | |
4654 | struct btrfs_block_rsv *block_rsv; | |
4655 | struct btrfs_block_group_cache *cache = NULL; | |
4656 | int ret; | |
4657 | ||
4658 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
4659 | ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len, | |
4660 | parent, root->root_key.objectid, | |
4661 | btrfs_header_level(buf), | |
4662 | BTRFS_DROP_DELAYED_REF, NULL); | |
4663 | BUG_ON(ret); | |
4664 | } | |
4665 | ||
4666 | if (!last_ref) | |
4667 | return; | |
4668 | ||
4669 | block_rsv = get_block_rsv(trans, root); | |
4670 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); | |
3bf84a5a YZ |
4671 | if (block_rsv->space_info != cache->space_info) |
4672 | goto out; | |
f0486c68 YZ |
4673 | |
4674 | if (btrfs_header_generation(buf) == trans->transid) { | |
4675 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
4676 | ret = check_ref_cleanup(trans, root, buf->start); | |
4677 | if (!ret) | |
4678 | goto pin; | |
4679 | } | |
4680 | ||
4681 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
4682 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
4683 | goto pin; | |
4684 | } | |
4685 | ||
4686 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
4687 | ||
4688 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4689 | ret = update_reserved_bytes(cache, buf->len, 0, 0); | |
4690 | if (ret == -EAGAIN) { | |
4691 | /* block group became read-only */ | |
4692 | update_reserved_bytes(cache, buf->len, 0, 1); | |
4693 | goto out; | |
4694 | } | |
4695 | ||
4696 | ret = 1; | |
4697 | spin_lock(&block_rsv->lock); | |
4698 | if (block_rsv->reserved < block_rsv->size) { | |
4699 | block_rsv->reserved += buf->len; | |
4700 | ret = 0; | |
4701 | } | |
4702 | spin_unlock(&block_rsv->lock); | |
4703 | ||
4704 | if (ret) { | |
4705 | spin_lock(&cache->space_info->lock); | |
4706 | cache->space_info->bytes_reserved -= buf->len; | |
4707 | spin_unlock(&cache->space_info->lock); | |
4708 | } | |
4709 | goto out; | |
4710 | } | |
4711 | pin: | |
4712 | if (block_rsv->durable && !cache->ro) { | |
4713 | ret = 0; | |
4714 | spin_lock(&cache->lock); | |
4715 | if (!cache->ro) { | |
4716 | cache->reserved_pinned += buf->len; | |
4717 | ret = 1; | |
4718 | } | |
4719 | spin_unlock(&cache->lock); | |
4720 | ||
4721 | if (ret) { | |
4722 | spin_lock(&block_rsv->lock); | |
4723 | block_rsv->freed[trans->transid & 0x1] += buf->len; | |
4724 | spin_unlock(&block_rsv->lock); | |
4725 | } | |
4726 | } | |
4727 | out: | |
4728 | btrfs_put_block_group(cache); | |
4729 | } | |
4730 | ||
925baedd | 4731 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
4732 | struct btrfs_root *root, |
4733 | u64 bytenr, u64 num_bytes, u64 parent, | |
5d4f98a2 | 4734 | u64 root_objectid, u64 owner, u64 offset) |
925baedd CM |
4735 | { |
4736 | int ret; | |
4737 | ||
56bec294 CM |
4738 | /* |
4739 | * tree log blocks never actually go into the extent allocation | |
4740 | * tree, just update pinning info and exit early. | |
56bec294 | 4741 | */ |
5d4f98a2 YZ |
4742 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
4743 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 4744 | /* unlocks the pinned mutex */ |
11833d66 | 4745 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 4746 | ret = 0; |
5d4f98a2 YZ |
4747 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
4748 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
4749 | parent, root_objectid, (int)owner, | |
4750 | BTRFS_DROP_DELAYED_REF, NULL); | |
1887be66 | 4751 | BUG_ON(ret); |
5d4f98a2 YZ |
4752 | } else { |
4753 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
4754 | parent, root_objectid, owner, | |
4755 | offset, BTRFS_DROP_DELAYED_REF, NULL); | |
4756 | BUG_ON(ret); | |
56bec294 | 4757 | } |
925baedd CM |
4758 | return ret; |
4759 | } | |
4760 | ||
87ee04eb CM |
4761 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
4762 | { | |
4763 | u64 mask = ((u64)root->stripesize - 1); | |
4764 | u64 ret = (val + mask) & ~mask; | |
4765 | return ret; | |
4766 | } | |
4767 | ||
817d52f8 JB |
4768 | /* |
4769 | * when we wait for progress in the block group caching, its because | |
4770 | * our allocation attempt failed at least once. So, we must sleep | |
4771 | * and let some progress happen before we try again. | |
4772 | * | |
4773 | * This function will sleep at least once waiting for new free space to | |
4774 | * show up, and then it will check the block group free space numbers | |
4775 | * for our min num_bytes. Another option is to have it go ahead | |
4776 | * and look in the rbtree for a free extent of a given size, but this | |
4777 | * is a good start. | |
4778 | */ | |
4779 | static noinline int | |
4780 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
4781 | u64 num_bytes) | |
4782 | { | |
11833d66 | 4783 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
4784 | DEFINE_WAIT(wait); |
4785 | ||
11833d66 YZ |
4786 | caching_ctl = get_caching_control(cache); |
4787 | if (!caching_ctl) | |
817d52f8 | 4788 | return 0; |
817d52f8 | 4789 | |
11833d66 | 4790 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
817d52f8 | 4791 | (cache->free_space >= num_bytes)); |
11833d66 YZ |
4792 | |
4793 | put_caching_control(caching_ctl); | |
4794 | return 0; | |
4795 | } | |
4796 | ||
4797 | static noinline int | |
4798 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
4799 | { | |
4800 | struct btrfs_caching_control *caching_ctl; | |
4801 | DEFINE_WAIT(wait); | |
4802 | ||
4803 | caching_ctl = get_caching_control(cache); | |
4804 | if (!caching_ctl) | |
4805 | return 0; | |
4806 | ||
4807 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
4808 | ||
4809 | put_caching_control(caching_ctl); | |
817d52f8 JB |
4810 | return 0; |
4811 | } | |
4812 | ||
b742bb82 YZ |
4813 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
4814 | { | |
4815 | int index; | |
4816 | if (cache->flags & BTRFS_BLOCK_GROUP_RAID10) | |
4817 | index = 0; | |
4818 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1) | |
4819 | index = 1; | |
4820 | else if (cache->flags & BTRFS_BLOCK_GROUP_DUP) | |
4821 | index = 2; | |
4822 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0) | |
4823 | index = 3; | |
4824 | else | |
4825 | index = 4; | |
4826 | return index; | |
4827 | } | |
4828 | ||
817d52f8 | 4829 | enum btrfs_loop_type { |
ccf0e725 | 4830 | LOOP_FIND_IDEAL = 0, |
817d52f8 JB |
4831 | LOOP_CACHING_NOWAIT = 1, |
4832 | LOOP_CACHING_WAIT = 2, | |
4833 | LOOP_ALLOC_CHUNK = 3, | |
4834 | LOOP_NO_EMPTY_SIZE = 4, | |
4835 | }; | |
4836 | ||
fec577fb CM |
4837 | /* |
4838 | * walks the btree of allocated extents and find a hole of a given size. | |
4839 | * The key ins is changed to record the hole: | |
4840 | * ins->objectid == block start | |
62e2749e | 4841 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
4842 | * ins->offset == number of blocks |
4843 | * Any available blocks before search_start are skipped. | |
4844 | */ | |
d397712b | 4845 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
4846 | struct btrfs_root *orig_root, |
4847 | u64 num_bytes, u64 empty_size, | |
4848 | u64 search_start, u64 search_end, | |
4849 | u64 hint_byte, struct btrfs_key *ins, | |
98ed5174 | 4850 | int data) |
fec577fb | 4851 | { |
80eb234a | 4852 | int ret = 0; |
d397712b | 4853 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 4854 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 4855 | struct btrfs_block_group_cache *block_group = NULL; |
239b14b3 | 4856 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 4857 | int allowed_chunk_alloc = 0; |
ccf0e725 | 4858 | int done_chunk_alloc = 0; |
80eb234a | 4859 | struct btrfs_space_info *space_info; |
fa9c0d79 CM |
4860 | int last_ptr_loop = 0; |
4861 | int loop = 0; | |
f0486c68 | 4862 | int index = 0; |
817d52f8 | 4863 | bool found_uncached_bg = false; |
0a24325e | 4864 | bool failed_cluster_refill = false; |
1cdda9b8 | 4865 | bool failed_alloc = false; |
67377734 | 4866 | bool use_cluster = true; |
ccf0e725 JB |
4867 | u64 ideal_cache_percent = 0; |
4868 | u64 ideal_cache_offset = 0; | |
fec577fb | 4869 | |
db94535d | 4870 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 4871 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
4872 | ins->objectid = 0; |
4873 | ins->offset = 0; | |
b1a4d965 | 4874 | |
2552d17e | 4875 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 JB |
4876 | if (!space_info) { |
4877 | printk(KERN_ERR "No space info for %d\n", data); | |
4878 | return -ENOSPC; | |
4879 | } | |
2552d17e | 4880 | |
67377734 JB |
4881 | /* |
4882 | * If the space info is for both data and metadata it means we have a | |
4883 | * small filesystem and we can't use the clustering stuff. | |
4884 | */ | |
4885 | if (btrfs_mixed_space_info(space_info)) | |
4886 | use_cluster = false; | |
4887 | ||
0ef3e66b CM |
4888 | if (orig_root->ref_cows || empty_size) |
4889 | allowed_chunk_alloc = 1; | |
4890 | ||
67377734 | 4891 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 4892 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
4893 | if (!btrfs_test_opt(root, SSD)) |
4894 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
4895 | } |
4896 | ||
67377734 JB |
4897 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
4898 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
4899 | last_ptr = &root->fs_info->data_alloc_cluster; |
4900 | } | |
0f9dd46c | 4901 | |
239b14b3 | 4902 | if (last_ptr) { |
fa9c0d79 CM |
4903 | spin_lock(&last_ptr->lock); |
4904 | if (last_ptr->block_group) | |
4905 | hint_byte = last_ptr->window_start; | |
4906 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 4907 | } |
fa9c0d79 | 4908 | |
a061fc8d | 4909 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 4910 | search_start = max(search_start, hint_byte); |
0b86a832 | 4911 | |
817d52f8 | 4912 | if (!last_ptr) |
fa9c0d79 | 4913 | empty_cluster = 0; |
fa9c0d79 | 4914 | |
2552d17e | 4915 | if (search_start == hint_byte) { |
ccf0e725 | 4916 | ideal_cache: |
2552d17e JB |
4917 | block_group = btrfs_lookup_block_group(root->fs_info, |
4918 | search_start); | |
817d52f8 JB |
4919 | /* |
4920 | * we don't want to use the block group if it doesn't match our | |
4921 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4922 | * |
4923 | * However if we are re-searching with an ideal block group | |
4924 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
4925 | */ |
4926 | if (block_group && block_group_bits(block_group, data) && | |
ccf0e725 JB |
4927 | (block_group->cached != BTRFS_CACHE_NO || |
4928 | search_start == ideal_cache_offset)) { | |
2552d17e | 4929 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4930 | if (list_empty(&block_group->list) || |
4931 | block_group->ro) { | |
4932 | /* | |
4933 | * someone is removing this block group, | |
4934 | * we can't jump into the have_block_group | |
4935 | * target because our list pointers are not | |
4936 | * valid | |
4937 | */ | |
4938 | btrfs_put_block_group(block_group); | |
4939 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4940 | } else { |
b742bb82 | 4941 | index = get_block_group_index(block_group); |
44fb5511 | 4942 | goto have_block_group; |
ccf0e725 | 4943 | } |
2552d17e | 4944 | } else if (block_group) { |
fa9c0d79 | 4945 | btrfs_put_block_group(block_group); |
2552d17e | 4946 | } |
42e70e7a | 4947 | } |
2552d17e | 4948 | search: |
80eb234a | 4949 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
4950 | list_for_each_entry(block_group, &space_info->block_groups[index], |
4951 | list) { | |
6226cb0a | 4952 | u64 offset; |
817d52f8 | 4953 | int cached; |
8a1413a2 | 4954 | |
11dfe35a | 4955 | btrfs_get_block_group(block_group); |
2552d17e | 4956 | search_start = block_group->key.objectid; |
42e70e7a | 4957 | |
83a50de9 CM |
4958 | /* |
4959 | * this can happen if we end up cycling through all the | |
4960 | * raid types, but we want to make sure we only allocate | |
4961 | * for the proper type. | |
4962 | */ | |
4963 | if (!block_group_bits(block_group, data)) { | |
4964 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
4965 | BTRFS_BLOCK_GROUP_RAID1 | | |
4966 | BTRFS_BLOCK_GROUP_RAID10; | |
4967 | ||
4968 | /* | |
4969 | * if they asked for extra copies and this block group | |
4970 | * doesn't provide them, bail. This does allow us to | |
4971 | * fill raid0 from raid1. | |
4972 | */ | |
4973 | if ((data & extra) && !(block_group->flags & extra)) | |
4974 | goto loop; | |
4975 | } | |
4976 | ||
2552d17e | 4977 | have_block_group: |
817d52f8 | 4978 | if (unlikely(block_group->cached == BTRFS_CACHE_NO)) { |
ccf0e725 JB |
4979 | u64 free_percent; |
4980 | ||
b8399dee JB |
4981 | ret = cache_block_group(block_group, trans, |
4982 | orig_root, 1); | |
9d66e233 JB |
4983 | if (block_group->cached == BTRFS_CACHE_FINISHED) |
4984 | goto have_block_group; | |
4985 | ||
ccf0e725 JB |
4986 | free_percent = btrfs_block_group_used(&block_group->item); |
4987 | free_percent *= 100; | |
4988 | free_percent = div64_u64(free_percent, | |
4989 | block_group->key.offset); | |
4990 | free_percent = 100 - free_percent; | |
4991 | if (free_percent > ideal_cache_percent && | |
4992 | likely(!block_group->ro)) { | |
4993 | ideal_cache_offset = block_group->key.objectid; | |
4994 | ideal_cache_percent = free_percent; | |
4995 | } | |
4996 | ||
817d52f8 | 4997 | /* |
ccf0e725 JB |
4998 | * We only want to start kthread caching if we are at |
4999 | * the point where we will wait for caching to make | |
5000 | * progress, or if our ideal search is over and we've | |
5001 | * found somebody to start caching. | |
817d52f8 JB |
5002 | */ |
5003 | if (loop > LOOP_CACHING_NOWAIT || | |
ccf0e725 JB |
5004 | (loop > LOOP_FIND_IDEAL && |
5005 | atomic_read(&space_info->caching_threads) < 2)) { | |
b8399dee JB |
5006 | ret = cache_block_group(block_group, trans, |
5007 | orig_root, 0); | |
817d52f8 | 5008 | BUG_ON(ret); |
2552d17e | 5009 | } |
817d52f8 JB |
5010 | found_uncached_bg = true; |
5011 | ||
ccf0e725 JB |
5012 | /* |
5013 | * If loop is set for cached only, try the next block | |
5014 | * group. | |
5015 | */ | |
5016 | if (loop == LOOP_FIND_IDEAL) | |
817d52f8 JB |
5017 | goto loop; |
5018 | } | |
5019 | ||
ccf0e725 JB |
5020 | cached = block_group_cache_done(block_group); |
5021 | if (unlikely(!cached)) | |
5022 | found_uncached_bg = true; | |
5023 | ||
ea6a478e | 5024 | if (unlikely(block_group->ro)) |
2552d17e | 5025 | goto loop; |
0f9dd46c | 5026 | |
0a24325e JB |
5027 | /* |
5028 | * Ok we want to try and use the cluster allocator, so lets look | |
5029 | * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will | |
5030 | * have tried the cluster allocator plenty of times at this | |
5031 | * point and not have found anything, so we are likely way too | |
5032 | * fragmented for the clustering stuff to find anything, so lets | |
5033 | * just skip it and let the allocator find whatever block it can | |
5034 | * find | |
5035 | */ | |
5036 | if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) { | |
fa9c0d79 CM |
5037 | /* |
5038 | * the refill lock keeps out other | |
5039 | * people trying to start a new cluster | |
5040 | */ | |
5041 | spin_lock(&last_ptr->refill_lock); | |
44fb5511 CM |
5042 | if (last_ptr->block_group && |
5043 | (last_ptr->block_group->ro || | |
5044 | !block_group_bits(last_ptr->block_group, data))) { | |
5045 | offset = 0; | |
5046 | goto refill_cluster; | |
5047 | } | |
5048 | ||
fa9c0d79 CM |
5049 | offset = btrfs_alloc_from_cluster(block_group, last_ptr, |
5050 | num_bytes, search_start); | |
5051 | if (offset) { | |
5052 | /* we have a block, we're done */ | |
5053 | spin_unlock(&last_ptr->refill_lock); | |
5054 | goto checks; | |
5055 | } | |
5056 | ||
5057 | spin_lock(&last_ptr->lock); | |
5058 | /* | |
5059 | * whoops, this cluster doesn't actually point to | |
5060 | * this block group. Get a ref on the block | |
5061 | * group is does point to and try again | |
5062 | */ | |
5063 | if (!last_ptr_loop && last_ptr->block_group && | |
5064 | last_ptr->block_group != block_group) { | |
5065 | ||
5066 | btrfs_put_block_group(block_group); | |
5067 | block_group = last_ptr->block_group; | |
11dfe35a | 5068 | btrfs_get_block_group(block_group); |
fa9c0d79 CM |
5069 | spin_unlock(&last_ptr->lock); |
5070 | spin_unlock(&last_ptr->refill_lock); | |
5071 | ||
5072 | last_ptr_loop = 1; | |
5073 | search_start = block_group->key.objectid; | |
44fb5511 CM |
5074 | /* |
5075 | * we know this block group is properly | |
5076 | * in the list because | |
5077 | * btrfs_remove_block_group, drops the | |
5078 | * cluster before it removes the block | |
5079 | * group from the list | |
5080 | */ | |
fa9c0d79 CM |
5081 | goto have_block_group; |
5082 | } | |
5083 | spin_unlock(&last_ptr->lock); | |
44fb5511 | 5084 | refill_cluster: |
fa9c0d79 CM |
5085 | /* |
5086 | * this cluster didn't work out, free it and | |
5087 | * start over | |
5088 | */ | |
5089 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5090 | ||
5091 | last_ptr_loop = 0; | |
5092 | ||
5093 | /* allocate a cluster in this block group */ | |
451d7585 | 5094 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 CM |
5095 | block_group, last_ptr, |
5096 | offset, num_bytes, | |
5097 | empty_cluster + empty_size); | |
5098 | if (ret == 0) { | |
5099 | /* | |
5100 | * now pull our allocation out of this | |
5101 | * cluster | |
5102 | */ | |
5103 | offset = btrfs_alloc_from_cluster(block_group, | |
5104 | last_ptr, num_bytes, | |
5105 | search_start); | |
5106 | if (offset) { | |
5107 | /* we found one, proceed */ | |
5108 | spin_unlock(&last_ptr->refill_lock); | |
5109 | goto checks; | |
5110 | } | |
0a24325e JB |
5111 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5112 | && !failed_cluster_refill) { | |
817d52f8 JB |
5113 | spin_unlock(&last_ptr->refill_lock); |
5114 | ||
0a24325e | 5115 | failed_cluster_refill = true; |
817d52f8 JB |
5116 | wait_block_group_cache_progress(block_group, |
5117 | num_bytes + empty_cluster + empty_size); | |
5118 | goto have_block_group; | |
fa9c0d79 | 5119 | } |
817d52f8 | 5120 | |
fa9c0d79 CM |
5121 | /* |
5122 | * at this point we either didn't find a cluster | |
5123 | * or we weren't able to allocate a block from our | |
5124 | * cluster. Free the cluster we've been trying | |
5125 | * to use, and go to the next block group | |
5126 | */ | |
0a24325e | 5127 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5128 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5129 | goto loop; |
fa9c0d79 CM |
5130 | } |
5131 | ||
6226cb0a JB |
5132 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5133 | num_bytes, empty_size); | |
1cdda9b8 JB |
5134 | /* |
5135 | * If we didn't find a chunk, and we haven't failed on this | |
5136 | * block group before, and this block group is in the middle of | |
5137 | * caching and we are ok with waiting, then go ahead and wait | |
5138 | * for progress to be made, and set failed_alloc to true. | |
5139 | * | |
5140 | * If failed_alloc is true then we've already waited on this | |
5141 | * block group once and should move on to the next block group. | |
5142 | */ | |
5143 | if (!offset && !failed_alloc && !cached && | |
5144 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5145 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5146 | num_bytes + empty_size); |
5147 | failed_alloc = true; | |
817d52f8 | 5148 | goto have_block_group; |
1cdda9b8 JB |
5149 | } else if (!offset) { |
5150 | goto loop; | |
817d52f8 | 5151 | } |
fa9c0d79 | 5152 | checks: |
6226cb0a | 5153 | search_start = stripe_align(root, offset); |
2552d17e | 5154 | /* move on to the next group */ |
6226cb0a JB |
5155 | if (search_start + num_bytes >= search_end) { |
5156 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 5157 | goto loop; |
6226cb0a | 5158 | } |
25179201 | 5159 | |
2552d17e JB |
5160 | /* move on to the next group */ |
5161 | if (search_start + num_bytes > | |
6226cb0a JB |
5162 | block_group->key.objectid + block_group->key.offset) { |
5163 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 5164 | goto loop; |
6226cb0a | 5165 | } |
f5a31e16 | 5166 | |
f0486c68 YZ |
5167 | ins->objectid = search_start; |
5168 | ins->offset = num_bytes; | |
2552d17e | 5169 | |
f0486c68 YZ |
5170 | if (offset < search_start) |
5171 | btrfs_add_free_space(block_group, offset, | |
5172 | search_start - offset); | |
5173 | BUG_ON(offset > search_start); | |
2552d17e | 5174 | |
f0486c68 YZ |
5175 | ret = update_reserved_bytes(block_group, num_bytes, 1, |
5176 | (data & BTRFS_BLOCK_GROUP_DATA)); | |
5177 | if (ret == -EAGAIN) { | |
6226cb0a | 5178 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 5179 | goto loop; |
0f9dd46c | 5180 | } |
0b86a832 | 5181 | |
f0486c68 | 5182 | /* we are all good, lets return */ |
2552d17e JB |
5183 | ins->objectid = search_start; |
5184 | ins->offset = num_bytes; | |
d2fb3437 | 5185 | |
6226cb0a JB |
5186 | if (offset < search_start) |
5187 | btrfs_add_free_space(block_group, offset, | |
5188 | search_start - offset); | |
5189 | BUG_ON(offset > search_start); | |
2552d17e JB |
5190 | break; |
5191 | loop: | |
0a24325e | 5192 | failed_cluster_refill = false; |
1cdda9b8 | 5193 | failed_alloc = false; |
b742bb82 | 5194 | BUG_ON(index != get_block_group_index(block_group)); |
fa9c0d79 | 5195 | btrfs_put_block_group(block_group); |
2552d17e JB |
5196 | } |
5197 | up_read(&space_info->groups_sem); | |
5198 | ||
b742bb82 YZ |
5199 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5200 | goto search; | |
5201 | ||
ccf0e725 JB |
5202 | /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for |
5203 | * for them to make caching progress. Also | |
5204 | * determine the best possible bg to cache | |
5205 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
5206 | * caching kthreads as we move along | |
817d52f8 JB |
5207 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5208 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5209 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5210 | * again | |
fa9c0d79 | 5211 | */ |
817d52f8 JB |
5212 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE && |
5213 | (found_uncached_bg || empty_size || empty_cluster || | |
5214 | allowed_chunk_alloc)) { | |
b742bb82 | 5215 | index = 0; |
ccf0e725 | 5216 | if (loop == LOOP_FIND_IDEAL && found_uncached_bg) { |
817d52f8 | 5217 | found_uncached_bg = false; |
ccf0e725 JB |
5218 | loop++; |
5219 | if (!ideal_cache_percent && | |
5220 | atomic_read(&space_info->caching_threads)) | |
817d52f8 | 5221 | goto search; |
ccf0e725 JB |
5222 | |
5223 | /* | |
5224 | * 1 of the following 2 things have happened so far | |
5225 | * | |
5226 | * 1) We found an ideal block group for caching that | |
5227 | * is mostly full and will cache quickly, so we might | |
5228 | * as well wait for it. | |
5229 | * | |
5230 | * 2) We searched for cached only and we didn't find | |
5231 | * anything, and we didn't start any caching kthreads | |
5232 | * either, so chances are we will loop through and | |
5233 | * start a couple caching kthreads, and then come back | |
5234 | * around and just wait for them. This will be slower | |
5235 | * because we will have 2 caching kthreads reading at | |
5236 | * the same time when we could have just started one | |
5237 | * and waited for it to get far enough to give us an | |
5238 | * allocation, so go ahead and go to the wait caching | |
5239 | * loop. | |
5240 | */ | |
5241 | loop = LOOP_CACHING_WAIT; | |
5242 | search_start = ideal_cache_offset; | |
5243 | ideal_cache_percent = 0; | |
5244 | goto ideal_cache; | |
5245 | } else if (loop == LOOP_FIND_IDEAL) { | |
5246 | /* | |
5247 | * Didn't find a uncached bg, wait on anything we find | |
5248 | * next. | |
5249 | */ | |
5250 | loop = LOOP_CACHING_WAIT; | |
5251 | goto search; | |
5252 | } | |
5253 | ||
5254 | if (loop < LOOP_CACHING_WAIT) { | |
5255 | loop++; | |
5256 | goto search; | |
817d52f8 JB |
5257 | } |
5258 | ||
5259 | if (loop == LOOP_ALLOC_CHUNK) { | |
fa9c0d79 CM |
5260 | empty_size = 0; |
5261 | empty_cluster = 0; | |
5262 | } | |
2552d17e JB |
5263 | |
5264 | if (allowed_chunk_alloc) { | |
5265 | ret = do_chunk_alloc(trans, root, num_bytes + | |
5266 | 2 * 1024 * 1024, data, 1); | |
2552d17e | 5267 | allowed_chunk_alloc = 0; |
ccf0e725 JB |
5268 | done_chunk_alloc = 1; |
5269 | } else if (!done_chunk_alloc) { | |
2552d17e JB |
5270 | space_info->force_alloc = 1; |
5271 | } | |
5272 | ||
817d52f8 | 5273 | if (loop < LOOP_NO_EMPTY_SIZE) { |
fa9c0d79 | 5274 | loop++; |
2552d17e | 5275 | goto search; |
fa9c0d79 | 5276 | } |
2552d17e JB |
5277 | ret = -ENOSPC; |
5278 | } else if (!ins->objectid) { | |
5279 | ret = -ENOSPC; | |
f2654de4 | 5280 | } |
0b86a832 | 5281 | |
80eb234a JB |
5282 | /* we found what we needed */ |
5283 | if (ins->objectid) { | |
5284 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) | |
d2fb3437 | 5285 | trans->block_group = block_group->key.objectid; |
0f9dd46c | 5286 | |
fa9c0d79 | 5287 | btrfs_put_block_group(block_group); |
80eb234a | 5288 | ret = 0; |
be744175 | 5289 | } |
be744175 | 5290 | |
0f70abe2 | 5291 | return ret; |
fec577fb | 5292 | } |
ec44a35c | 5293 | |
9ed74f2d JB |
5294 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5295 | int dump_block_groups) | |
0f9dd46c JB |
5296 | { |
5297 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5298 | int index = 0; |
0f9dd46c | 5299 | |
9ed74f2d | 5300 | spin_lock(&info->lock); |
d397712b CM |
5301 | printk(KERN_INFO "space_info has %llu free, is %sfull\n", |
5302 | (unsigned long long)(info->total_bytes - info->bytes_used - | |
9ed74f2d | 5303 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5304 | info->bytes_readonly), |
d397712b | 5305 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5306 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5307 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5308 | (unsigned long long)info->total_bytes, |
8929ecfa | 5309 | (unsigned long long)info->bytes_used, |
21380931 | 5310 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5311 | (unsigned long long)info->bytes_reserved, |
21380931 | 5312 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5313 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5314 | spin_unlock(&info->lock); |
5315 | ||
5316 | if (!dump_block_groups) | |
5317 | return; | |
0f9dd46c | 5318 | |
80eb234a | 5319 | down_read(&info->groups_sem); |
b742bb82 YZ |
5320 | again: |
5321 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5322 | spin_lock(&cache->lock); |
d397712b CM |
5323 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
5324 | "%llu pinned %llu reserved\n", | |
5325 | (unsigned long long)cache->key.objectid, | |
5326 | (unsigned long long)cache->key.offset, | |
5327 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5328 | (unsigned long long)cache->pinned, | |
5329 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
5330 | btrfs_dump_free_space(cache, bytes); |
5331 | spin_unlock(&cache->lock); | |
5332 | } | |
b742bb82 YZ |
5333 | if (++index < BTRFS_NR_RAID_TYPES) |
5334 | goto again; | |
80eb234a | 5335 | up_read(&info->groups_sem); |
0f9dd46c | 5336 | } |
e8569813 | 5337 | |
11833d66 YZ |
5338 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5339 | struct btrfs_root *root, | |
5340 | u64 num_bytes, u64 min_alloc_size, | |
5341 | u64 empty_size, u64 hint_byte, | |
5342 | u64 search_end, struct btrfs_key *ins, | |
5343 | u64 data) | |
fec577fb CM |
5344 | { |
5345 | int ret; | |
fbdc762b | 5346 | u64 search_start = 0; |
925baedd | 5347 | |
6a63209f | 5348 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 5349 | again: |
0ef3e66b CM |
5350 | /* |
5351 | * the only place that sets empty_size is btrfs_realloc_node, which | |
5352 | * is not called recursively on allocations | |
5353 | */ | |
83d3c969 | 5354 | if (empty_size || root->ref_cows) |
6324fbf3 | 5355 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b | 5356 | num_bytes + 2 * 1024 * 1024, data, 0); |
0b86a832 | 5357 | |
db94535d CM |
5358 | WARN_ON(num_bytes < root->sectorsize); |
5359 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
f0486c68 YZ |
5360 | search_start, search_end, hint_byte, |
5361 | ins, data); | |
3b951516 | 5362 | |
98d20f67 CM |
5363 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
5364 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 5365 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 5366 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b CM |
5367 | do_chunk_alloc(trans, root->fs_info->extent_root, |
5368 | num_bytes, data, 1); | |
98d20f67 CM |
5369 | goto again; |
5370 | } | |
817d52f8 | 5371 | if (ret == -ENOSPC) { |
0f9dd46c JB |
5372 | struct btrfs_space_info *sinfo; |
5373 | ||
5374 | sinfo = __find_space_info(root->fs_info, data); | |
d397712b CM |
5375 | printk(KERN_ERR "btrfs allocation failed flags %llu, " |
5376 | "wanted %llu\n", (unsigned long long)data, | |
5377 | (unsigned long long)num_bytes); | |
9ed74f2d | 5378 | dump_space_info(sinfo, num_bytes, 1); |
925baedd | 5379 | } |
0f9dd46c JB |
5380 | |
5381 | return ret; | |
e6dcd2dc CM |
5382 | } |
5383 | ||
65b51a00 CM |
5384 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
5385 | { | |
0f9dd46c | 5386 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 5387 | int ret = 0; |
0f9dd46c | 5388 | |
0f9dd46c JB |
5389 | cache = btrfs_lookup_block_group(root->fs_info, start); |
5390 | if (!cache) { | |
d397712b CM |
5391 | printk(KERN_ERR "Unable to find block group for %llu\n", |
5392 | (unsigned long long)start); | |
0f9dd46c JB |
5393 | return -ENOSPC; |
5394 | } | |
1f3c79a2 LH |
5395 | |
5396 | ret = btrfs_discard_extent(root, start, len); | |
5397 | ||
0f9dd46c | 5398 | btrfs_add_free_space(cache, start, len); |
f0486c68 | 5399 | update_reserved_bytes(cache, len, 0, 1); |
fa9c0d79 | 5400 | btrfs_put_block_group(cache); |
817d52f8 | 5401 | |
e6dcd2dc CM |
5402 | return ret; |
5403 | } | |
5404 | ||
5d4f98a2 YZ |
5405 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5406 | struct btrfs_root *root, | |
5407 | u64 parent, u64 root_objectid, | |
5408 | u64 flags, u64 owner, u64 offset, | |
5409 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
5410 | { |
5411 | int ret; | |
5d4f98a2 | 5412 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 5413 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 5414 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 5415 | struct btrfs_path *path; |
5d4f98a2 YZ |
5416 | struct extent_buffer *leaf; |
5417 | int type; | |
5418 | u32 size; | |
26b8003f | 5419 | |
5d4f98a2 YZ |
5420 | if (parent > 0) |
5421 | type = BTRFS_SHARED_DATA_REF_KEY; | |
5422 | else | |
5423 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 5424 | |
5d4f98a2 | 5425 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
5426 | |
5427 | path = btrfs_alloc_path(); | |
5428 | BUG_ON(!path); | |
47e4bb98 | 5429 | |
b9473439 | 5430 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5431 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
5432 | ins, size); | |
ccd467d6 | 5433 | BUG_ON(ret); |
0f9dd46c | 5434 | |
5d4f98a2 YZ |
5435 | leaf = path->nodes[0]; |
5436 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 5437 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5438 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
5439 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5440 | btrfs_set_extent_flags(leaf, extent_item, | |
5441 | flags | BTRFS_EXTENT_FLAG_DATA); | |
5442 | ||
5443 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
5444 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
5445 | if (parent > 0) { | |
5446 | struct btrfs_shared_data_ref *ref; | |
5447 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
5448 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5449 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
5450 | } else { | |
5451 | struct btrfs_extent_data_ref *ref; | |
5452 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
5453 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
5454 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
5455 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
5456 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
5457 | } | |
47e4bb98 CM |
5458 | |
5459 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 5460 | btrfs_free_path(path); |
f510cfec | 5461 | |
f0486c68 | 5462 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
f5947066 | 5463 | if (ret) { |
d397712b CM |
5464 | printk(KERN_ERR "btrfs update block group failed for %llu " |
5465 | "%llu\n", (unsigned long long)ins->objectid, | |
5466 | (unsigned long long)ins->offset); | |
f5947066 CM |
5467 | BUG(); |
5468 | } | |
e6dcd2dc CM |
5469 | return ret; |
5470 | } | |
5471 | ||
5d4f98a2 YZ |
5472 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
5473 | struct btrfs_root *root, | |
5474 | u64 parent, u64 root_objectid, | |
5475 | u64 flags, struct btrfs_disk_key *key, | |
5476 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
5477 | { |
5478 | int ret; | |
5d4f98a2 YZ |
5479 | struct btrfs_fs_info *fs_info = root->fs_info; |
5480 | struct btrfs_extent_item *extent_item; | |
5481 | struct btrfs_tree_block_info *block_info; | |
5482 | struct btrfs_extent_inline_ref *iref; | |
5483 | struct btrfs_path *path; | |
5484 | struct extent_buffer *leaf; | |
5485 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 5486 | |
5d4f98a2 YZ |
5487 | path = btrfs_alloc_path(); |
5488 | BUG_ON(!path); | |
56bec294 | 5489 | |
5d4f98a2 YZ |
5490 | path->leave_spinning = 1; |
5491 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
5492 | ins, size); | |
56bec294 | 5493 | BUG_ON(ret); |
5d4f98a2 YZ |
5494 | |
5495 | leaf = path->nodes[0]; | |
5496 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
5497 | struct btrfs_extent_item); | |
5498 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
5499 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5500 | btrfs_set_extent_flags(leaf, extent_item, | |
5501 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5502 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
5503 | ||
5504 | btrfs_set_tree_block_key(leaf, block_info, key); | |
5505 | btrfs_set_tree_block_level(leaf, block_info, level); | |
5506 | ||
5507 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
5508 | if (parent > 0) { | |
5509 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
5510 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5511 | BTRFS_SHARED_BLOCK_REF_KEY); | |
5512 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5513 | } else { | |
5514 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5515 | BTRFS_TREE_BLOCK_REF_KEY); | |
5516 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
5517 | } | |
5518 | ||
5519 | btrfs_mark_buffer_dirty(leaf); | |
5520 | btrfs_free_path(path); | |
5521 | ||
f0486c68 | 5522 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
5d4f98a2 YZ |
5523 | if (ret) { |
5524 | printk(KERN_ERR "btrfs update block group failed for %llu " | |
5525 | "%llu\n", (unsigned long long)ins->objectid, | |
5526 | (unsigned long long)ins->offset); | |
5527 | BUG(); | |
5528 | } | |
5529 | return ret; | |
5530 | } | |
5531 | ||
5532 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5533 | struct btrfs_root *root, | |
5534 | u64 root_objectid, u64 owner, | |
5535 | u64 offset, struct btrfs_key *ins) | |
5536 | { | |
5537 | int ret; | |
5538 | ||
5539 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
5540 | ||
5541 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset, | |
5542 | 0, root_objectid, owner, offset, | |
5543 | BTRFS_ADD_DELAYED_EXTENT, NULL); | |
e6dcd2dc CM |
5544 | return ret; |
5545 | } | |
e02119d5 CM |
5546 | |
5547 | /* | |
5548 | * this is used by the tree logging recovery code. It records that | |
5549 | * an extent has been allocated and makes sure to clear the free | |
5550 | * space cache bits as well | |
5551 | */ | |
5d4f98a2 YZ |
5552 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5553 | struct btrfs_root *root, | |
5554 | u64 root_objectid, u64 owner, u64 offset, | |
5555 | struct btrfs_key *ins) | |
e02119d5 CM |
5556 | { |
5557 | int ret; | |
5558 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
5559 | struct btrfs_caching_control *caching_ctl; |
5560 | u64 start = ins->objectid; | |
5561 | u64 num_bytes = ins->offset; | |
e02119d5 | 5562 | |
e02119d5 | 5563 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 5564 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 5565 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 5566 | |
11833d66 YZ |
5567 | if (!caching_ctl) { |
5568 | BUG_ON(!block_group_cache_done(block_group)); | |
5569 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
5570 | BUG_ON(ret); | |
5571 | } else { | |
5572 | mutex_lock(&caching_ctl->mutex); | |
5573 | ||
5574 | if (start >= caching_ctl->progress) { | |
5575 | ret = add_excluded_extent(root, start, num_bytes); | |
5576 | BUG_ON(ret); | |
5577 | } else if (start + num_bytes <= caching_ctl->progress) { | |
5578 | ret = btrfs_remove_free_space(block_group, | |
5579 | start, num_bytes); | |
5580 | BUG_ON(ret); | |
5581 | } else { | |
5582 | num_bytes = caching_ctl->progress - start; | |
5583 | ret = btrfs_remove_free_space(block_group, | |
5584 | start, num_bytes); | |
5585 | BUG_ON(ret); | |
5586 | ||
5587 | start = caching_ctl->progress; | |
5588 | num_bytes = ins->objectid + ins->offset - | |
5589 | caching_ctl->progress; | |
5590 | ret = add_excluded_extent(root, start, num_bytes); | |
5591 | BUG_ON(ret); | |
5592 | } | |
5593 | ||
5594 | mutex_unlock(&caching_ctl->mutex); | |
5595 | put_caching_control(caching_ctl); | |
5596 | } | |
5597 | ||
f0486c68 YZ |
5598 | ret = update_reserved_bytes(block_group, ins->offset, 1, 1); |
5599 | BUG_ON(ret); | |
fa9c0d79 | 5600 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
5601 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
5602 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
5603 | return ret; |
5604 | } | |
5605 | ||
65b51a00 CM |
5606 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
5607 | struct btrfs_root *root, | |
4008c04a CM |
5608 | u64 bytenr, u32 blocksize, |
5609 | int level) | |
65b51a00 CM |
5610 | { |
5611 | struct extent_buffer *buf; | |
5612 | ||
5613 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
5614 | if (!buf) | |
5615 | return ERR_PTR(-ENOMEM); | |
5616 | btrfs_set_header_generation(buf, trans->transid); | |
4008c04a | 5617 | btrfs_set_buffer_lockdep_class(buf, level); |
65b51a00 CM |
5618 | btrfs_tree_lock(buf); |
5619 | clean_tree_block(trans, root, buf); | |
b4ce94de CM |
5620 | |
5621 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 5622 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 5623 | |
d0c803c4 | 5624 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
5625 | /* |
5626 | * we allow two log transactions at a time, use different | |
5627 | * EXENT bit to differentiate dirty pages. | |
5628 | */ | |
5629 | if (root->log_transid % 2 == 0) | |
5630 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
5631 | buf->start + buf->len - 1, GFP_NOFS); | |
5632 | else | |
5633 | set_extent_new(&root->dirty_log_pages, buf->start, | |
5634 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
5635 | } else { |
5636 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 5637 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 5638 | } |
65b51a00 | 5639 | trans->blocks_used++; |
b4ce94de | 5640 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
5641 | return buf; |
5642 | } | |
5643 | ||
f0486c68 YZ |
5644 | static struct btrfs_block_rsv * |
5645 | use_block_rsv(struct btrfs_trans_handle *trans, | |
5646 | struct btrfs_root *root, u32 blocksize) | |
5647 | { | |
5648 | struct btrfs_block_rsv *block_rsv; | |
5649 | int ret; | |
5650 | ||
5651 | block_rsv = get_block_rsv(trans, root); | |
5652 | ||
5653 | if (block_rsv->size == 0) { | |
8bb8ab2e JB |
5654 | ret = reserve_metadata_bytes(trans, root, block_rsv, |
5655 | blocksize, 0); | |
f0486c68 YZ |
5656 | if (ret) |
5657 | return ERR_PTR(ret); | |
5658 | return block_rsv; | |
5659 | } | |
5660 | ||
5661 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
5662 | if (!ret) | |
5663 | return block_rsv; | |
5664 | ||
f0486c68 YZ |
5665 | return ERR_PTR(-ENOSPC); |
5666 | } | |
5667 | ||
5668 | static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
5669 | { | |
5670 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
5671 | block_rsv_release_bytes(block_rsv, NULL, 0); | |
5672 | } | |
5673 | ||
fec577fb | 5674 | /* |
f0486c68 YZ |
5675 | * finds a free extent and does all the dirty work required for allocation |
5676 | * returns the key for the extent through ins, and a tree buffer for | |
5677 | * the first block of the extent through buf. | |
5678 | * | |
fec577fb CM |
5679 | * returns the tree buffer or NULL. |
5680 | */ | |
5f39d397 | 5681 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
5682 | struct btrfs_root *root, u32 blocksize, |
5683 | u64 parent, u64 root_objectid, | |
5684 | struct btrfs_disk_key *key, int level, | |
5685 | u64 hint, u64 empty_size) | |
fec577fb | 5686 | { |
e2fa7227 | 5687 | struct btrfs_key ins; |
f0486c68 | 5688 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 5689 | struct extent_buffer *buf; |
f0486c68 YZ |
5690 | u64 flags = 0; |
5691 | int ret; | |
5692 | ||
fec577fb | 5693 | |
f0486c68 YZ |
5694 | block_rsv = use_block_rsv(trans, root, blocksize); |
5695 | if (IS_ERR(block_rsv)) | |
5696 | return ERR_CAST(block_rsv); | |
5697 | ||
5698 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
5699 | empty_size, hint, (u64)-1, &ins, 0); | |
fec577fb | 5700 | if (ret) { |
f0486c68 | 5701 | unuse_block_rsv(block_rsv, blocksize); |
54aa1f4d | 5702 | return ERR_PTR(ret); |
fec577fb | 5703 | } |
55c69072 | 5704 | |
4008c04a CM |
5705 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
5706 | blocksize, level); | |
f0486c68 YZ |
5707 | BUG_ON(IS_ERR(buf)); |
5708 | ||
5709 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
5710 | if (parent == 0) | |
5711 | parent = ins.objectid; | |
5712 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5713 | } else | |
5714 | BUG_ON(parent > 0); | |
5715 | ||
5716 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5717 | struct btrfs_delayed_extent_op *extent_op; | |
5718 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
5719 | BUG_ON(!extent_op); | |
5720 | if (key) | |
5721 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
5722 | else | |
5723 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
5724 | extent_op->flags_to_set = flags; | |
5725 | extent_op->update_key = 1; | |
5726 | extent_op->update_flags = 1; | |
5727 | extent_op->is_data = 0; | |
5728 | ||
5729 | ret = btrfs_add_delayed_tree_ref(trans, ins.objectid, | |
5730 | ins.offset, parent, root_objectid, | |
5731 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5732 | extent_op); | |
5733 | BUG_ON(ret); | |
5734 | } | |
fec577fb CM |
5735 | return buf; |
5736 | } | |
a28ec197 | 5737 | |
2c47e605 YZ |
5738 | struct walk_control { |
5739 | u64 refs[BTRFS_MAX_LEVEL]; | |
5740 | u64 flags[BTRFS_MAX_LEVEL]; | |
5741 | struct btrfs_key update_progress; | |
5742 | int stage; | |
5743 | int level; | |
5744 | int shared_level; | |
5745 | int update_ref; | |
5746 | int keep_locks; | |
1c4850e2 YZ |
5747 | int reada_slot; |
5748 | int reada_count; | |
2c47e605 YZ |
5749 | }; |
5750 | ||
5751 | #define DROP_REFERENCE 1 | |
5752 | #define UPDATE_BACKREF 2 | |
5753 | ||
1c4850e2 YZ |
5754 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
5755 | struct btrfs_root *root, | |
5756 | struct walk_control *wc, | |
5757 | struct btrfs_path *path) | |
6407bf6d | 5758 | { |
1c4850e2 YZ |
5759 | u64 bytenr; |
5760 | u64 generation; | |
5761 | u64 refs; | |
94fcca9f | 5762 | u64 flags; |
5d4f98a2 | 5763 | u32 nritems; |
1c4850e2 YZ |
5764 | u32 blocksize; |
5765 | struct btrfs_key key; | |
5766 | struct extent_buffer *eb; | |
6407bf6d | 5767 | int ret; |
1c4850e2 YZ |
5768 | int slot; |
5769 | int nread = 0; | |
6407bf6d | 5770 | |
1c4850e2 YZ |
5771 | if (path->slots[wc->level] < wc->reada_slot) { |
5772 | wc->reada_count = wc->reada_count * 2 / 3; | |
5773 | wc->reada_count = max(wc->reada_count, 2); | |
5774 | } else { | |
5775 | wc->reada_count = wc->reada_count * 3 / 2; | |
5776 | wc->reada_count = min_t(int, wc->reada_count, | |
5777 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
5778 | } | |
7bb86316 | 5779 | |
1c4850e2 YZ |
5780 | eb = path->nodes[wc->level]; |
5781 | nritems = btrfs_header_nritems(eb); | |
5782 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 5783 | |
1c4850e2 YZ |
5784 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
5785 | if (nread >= wc->reada_count) | |
5786 | break; | |
bd56b302 | 5787 | |
2dd3e67b | 5788 | cond_resched(); |
1c4850e2 YZ |
5789 | bytenr = btrfs_node_blockptr(eb, slot); |
5790 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 5791 | |
1c4850e2 YZ |
5792 | if (slot == path->slots[wc->level]) |
5793 | goto reada; | |
5d4f98a2 | 5794 | |
1c4850e2 YZ |
5795 | if (wc->stage == UPDATE_BACKREF && |
5796 | generation <= root->root_key.offset) | |
bd56b302 CM |
5797 | continue; |
5798 | ||
94fcca9f YZ |
5799 | /* We don't lock the tree block, it's OK to be racy here */ |
5800 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
5801 | &refs, &flags); | |
5802 | BUG_ON(ret); | |
5803 | BUG_ON(refs == 0); | |
5804 | ||
1c4850e2 | 5805 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
5806 | if (refs == 1) |
5807 | goto reada; | |
bd56b302 | 5808 | |
94fcca9f YZ |
5809 | if (wc->level == 1 && |
5810 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5811 | continue; | |
1c4850e2 YZ |
5812 | if (!wc->update_ref || |
5813 | generation <= root->root_key.offset) | |
5814 | continue; | |
5815 | btrfs_node_key_to_cpu(eb, &key, slot); | |
5816 | ret = btrfs_comp_cpu_keys(&key, | |
5817 | &wc->update_progress); | |
5818 | if (ret < 0) | |
5819 | continue; | |
94fcca9f YZ |
5820 | } else { |
5821 | if (wc->level == 1 && | |
5822 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5823 | continue; | |
6407bf6d | 5824 | } |
1c4850e2 YZ |
5825 | reada: |
5826 | ret = readahead_tree_block(root, bytenr, blocksize, | |
5827 | generation); | |
5828 | if (ret) | |
bd56b302 | 5829 | break; |
1c4850e2 | 5830 | nread++; |
20524f02 | 5831 | } |
1c4850e2 | 5832 | wc->reada_slot = slot; |
20524f02 | 5833 | } |
2c47e605 | 5834 | |
f82d02d9 | 5835 | /* |
2c47e605 YZ |
5836 | * hepler to process tree block while walking down the tree. |
5837 | * | |
2c47e605 YZ |
5838 | * when wc->stage == UPDATE_BACKREF, this function updates |
5839 | * back refs for pointers in the block. | |
5840 | * | |
5841 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 5842 | */ |
2c47e605 | 5843 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 5844 | struct btrfs_root *root, |
2c47e605 | 5845 | struct btrfs_path *path, |
94fcca9f | 5846 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 5847 | { |
2c47e605 YZ |
5848 | int level = wc->level; |
5849 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 5850 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
5851 | int ret; |
5852 | ||
2c47e605 YZ |
5853 | if (wc->stage == UPDATE_BACKREF && |
5854 | btrfs_header_owner(eb) != root->root_key.objectid) | |
5855 | return 1; | |
f82d02d9 | 5856 | |
2c47e605 YZ |
5857 | /* |
5858 | * when reference count of tree block is 1, it won't increase | |
5859 | * again. once full backref flag is set, we never clear it. | |
5860 | */ | |
94fcca9f YZ |
5861 | if (lookup_info && |
5862 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
5863 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
5864 | BUG_ON(!path->locks[level]); |
5865 | ret = btrfs_lookup_extent_info(trans, root, | |
5866 | eb->start, eb->len, | |
5867 | &wc->refs[level], | |
5868 | &wc->flags[level]); | |
5869 | BUG_ON(ret); | |
5870 | BUG_ON(wc->refs[level] == 0); | |
5871 | } | |
5d4f98a2 | 5872 | |
2c47e605 YZ |
5873 | if (wc->stage == DROP_REFERENCE) { |
5874 | if (wc->refs[level] > 1) | |
5875 | return 1; | |
f82d02d9 | 5876 | |
2c47e605 YZ |
5877 | if (path->locks[level] && !wc->keep_locks) { |
5878 | btrfs_tree_unlock(eb); | |
5879 | path->locks[level] = 0; | |
5880 | } | |
5881 | return 0; | |
5882 | } | |
f82d02d9 | 5883 | |
2c47e605 YZ |
5884 | /* wc->stage == UPDATE_BACKREF */ |
5885 | if (!(wc->flags[level] & flag)) { | |
5886 | BUG_ON(!path->locks[level]); | |
5887 | ret = btrfs_inc_ref(trans, root, eb, 1); | |
f82d02d9 | 5888 | BUG_ON(ret); |
2c47e605 YZ |
5889 | ret = btrfs_dec_ref(trans, root, eb, 0); |
5890 | BUG_ON(ret); | |
5891 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, | |
5892 | eb->len, flag, 0); | |
5893 | BUG_ON(ret); | |
5894 | wc->flags[level] |= flag; | |
5895 | } | |
5896 | ||
5897 | /* | |
5898 | * the block is shared by multiple trees, so it's not good to | |
5899 | * keep the tree lock | |
5900 | */ | |
5901 | if (path->locks[level] && level > 0) { | |
5902 | btrfs_tree_unlock(eb); | |
5903 | path->locks[level] = 0; | |
5904 | } | |
5905 | return 0; | |
5906 | } | |
5907 | ||
1c4850e2 YZ |
5908 | /* |
5909 | * hepler to process tree block pointer. | |
5910 | * | |
5911 | * when wc->stage == DROP_REFERENCE, this function checks | |
5912 | * reference count of the block pointed to. if the block | |
5913 | * is shared and we need update back refs for the subtree | |
5914 | * rooted at the block, this function changes wc->stage to | |
5915 | * UPDATE_BACKREF. if the block is shared and there is no | |
5916 | * need to update back, this function drops the reference | |
5917 | * to the block. | |
5918 | * | |
5919 | * NOTE: return value 1 means we should stop walking down. | |
5920 | */ | |
5921 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
5922 | struct btrfs_root *root, | |
5923 | struct btrfs_path *path, | |
94fcca9f | 5924 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
5925 | { |
5926 | u64 bytenr; | |
5927 | u64 generation; | |
5928 | u64 parent; | |
5929 | u32 blocksize; | |
5930 | struct btrfs_key key; | |
5931 | struct extent_buffer *next; | |
5932 | int level = wc->level; | |
5933 | int reada = 0; | |
5934 | int ret = 0; | |
5935 | ||
5936 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
5937 | path->slots[level]); | |
5938 | /* | |
5939 | * if the lower level block was created before the snapshot | |
5940 | * was created, we know there is no need to update back refs | |
5941 | * for the subtree | |
5942 | */ | |
5943 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5944 | generation <= root->root_key.offset) { |
5945 | *lookup_info = 1; | |
1c4850e2 | 5946 | return 1; |
94fcca9f | 5947 | } |
1c4850e2 YZ |
5948 | |
5949 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
5950 | blocksize = btrfs_level_size(root, level - 1); | |
5951 | ||
5952 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
5953 | if (!next) { | |
5954 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
5955 | if (!next) |
5956 | return -ENOMEM; | |
1c4850e2 YZ |
5957 | reada = 1; |
5958 | } | |
5959 | btrfs_tree_lock(next); | |
5960 | btrfs_set_lock_blocking(next); | |
5961 | ||
94fcca9f YZ |
5962 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
5963 | &wc->refs[level - 1], | |
5964 | &wc->flags[level - 1]); | |
5965 | BUG_ON(ret); | |
5966 | BUG_ON(wc->refs[level - 1] == 0); | |
5967 | *lookup_info = 0; | |
1c4850e2 | 5968 | |
94fcca9f | 5969 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 5970 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
5971 | if (level == 1 && |
5972 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5973 | goto skip; | |
5974 | ||
1c4850e2 YZ |
5975 | if (!wc->update_ref || |
5976 | generation <= root->root_key.offset) | |
5977 | goto skip; | |
5978 | ||
5979 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
5980 | path->slots[level]); | |
5981 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
5982 | if (ret < 0) | |
5983 | goto skip; | |
5984 | ||
5985 | wc->stage = UPDATE_BACKREF; | |
5986 | wc->shared_level = level - 1; | |
5987 | } | |
94fcca9f YZ |
5988 | } else { |
5989 | if (level == 1 && | |
5990 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5991 | goto skip; | |
1c4850e2 YZ |
5992 | } |
5993 | ||
5994 | if (!btrfs_buffer_uptodate(next, generation)) { | |
5995 | btrfs_tree_unlock(next); | |
5996 | free_extent_buffer(next); | |
5997 | next = NULL; | |
94fcca9f | 5998 | *lookup_info = 1; |
1c4850e2 YZ |
5999 | } |
6000 | ||
6001 | if (!next) { | |
6002 | if (reada && level == 1) | |
6003 | reada_walk_down(trans, root, wc, path); | |
6004 | next = read_tree_block(root, bytenr, blocksize, generation); | |
6005 | btrfs_tree_lock(next); | |
6006 | btrfs_set_lock_blocking(next); | |
6007 | } | |
6008 | ||
6009 | level--; | |
6010 | BUG_ON(level != btrfs_header_level(next)); | |
6011 | path->nodes[level] = next; | |
6012 | path->slots[level] = 0; | |
6013 | path->locks[level] = 1; | |
6014 | wc->level = level; | |
6015 | if (wc->level == 1) | |
6016 | wc->reada_slot = 0; | |
6017 | return 0; | |
6018 | skip: | |
6019 | wc->refs[level - 1] = 0; | |
6020 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6021 | if (wc->stage == DROP_REFERENCE) { |
6022 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6023 | parent = path->nodes[level]->start; | |
6024 | } else { | |
6025 | BUG_ON(root->root_key.objectid != | |
6026 | btrfs_header_owner(path->nodes[level])); | |
6027 | parent = 0; | |
6028 | } | |
1c4850e2 | 6029 | |
94fcca9f YZ |
6030 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
6031 | root->root_key.objectid, level - 1, 0); | |
6032 | BUG_ON(ret); | |
1c4850e2 | 6033 | } |
1c4850e2 YZ |
6034 | btrfs_tree_unlock(next); |
6035 | free_extent_buffer(next); | |
94fcca9f | 6036 | *lookup_info = 1; |
1c4850e2 YZ |
6037 | return 1; |
6038 | } | |
6039 | ||
2c47e605 YZ |
6040 | /* |
6041 | * hepler to process tree block while walking up the tree. | |
6042 | * | |
6043 | * when wc->stage == DROP_REFERENCE, this function drops | |
6044 | * reference count on the block. | |
6045 | * | |
6046 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6047 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6048 | * to UPDATE_BACKREF previously while processing the block. | |
6049 | * | |
6050 | * NOTE: return value 1 means we should stop walking up. | |
6051 | */ | |
6052 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6053 | struct btrfs_root *root, | |
6054 | struct btrfs_path *path, | |
6055 | struct walk_control *wc) | |
6056 | { | |
f0486c68 | 6057 | int ret; |
2c47e605 YZ |
6058 | int level = wc->level; |
6059 | struct extent_buffer *eb = path->nodes[level]; | |
6060 | u64 parent = 0; | |
6061 | ||
6062 | if (wc->stage == UPDATE_BACKREF) { | |
6063 | BUG_ON(wc->shared_level < level); | |
6064 | if (level < wc->shared_level) | |
6065 | goto out; | |
6066 | ||
2c47e605 YZ |
6067 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6068 | if (ret > 0) | |
6069 | wc->update_ref = 0; | |
6070 | ||
6071 | wc->stage = DROP_REFERENCE; | |
6072 | wc->shared_level = -1; | |
6073 | path->slots[level] = 0; | |
6074 | ||
6075 | /* | |
6076 | * check reference count again if the block isn't locked. | |
6077 | * we should start walking down the tree again if reference | |
6078 | * count is one. | |
6079 | */ | |
6080 | if (!path->locks[level]) { | |
6081 | BUG_ON(level == 0); | |
6082 | btrfs_tree_lock(eb); | |
6083 | btrfs_set_lock_blocking(eb); | |
6084 | path->locks[level] = 1; | |
6085 | ||
6086 | ret = btrfs_lookup_extent_info(trans, root, | |
6087 | eb->start, eb->len, | |
6088 | &wc->refs[level], | |
6089 | &wc->flags[level]); | |
f82d02d9 | 6090 | BUG_ON(ret); |
2c47e605 YZ |
6091 | BUG_ON(wc->refs[level] == 0); |
6092 | if (wc->refs[level] == 1) { | |
6093 | btrfs_tree_unlock(eb); | |
6094 | path->locks[level] = 0; | |
6095 | return 1; | |
6096 | } | |
f82d02d9 | 6097 | } |
2c47e605 | 6098 | } |
f82d02d9 | 6099 | |
2c47e605 YZ |
6100 | /* wc->stage == DROP_REFERENCE */ |
6101 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6102 | |
2c47e605 YZ |
6103 | if (wc->refs[level] == 1) { |
6104 | if (level == 0) { | |
6105 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6106 | ret = btrfs_dec_ref(trans, root, eb, 1); | |
6107 | else | |
6108 | ret = btrfs_dec_ref(trans, root, eb, 0); | |
6109 | BUG_ON(ret); | |
6110 | } | |
6111 | /* make block locked assertion in clean_tree_block happy */ | |
6112 | if (!path->locks[level] && | |
6113 | btrfs_header_generation(eb) == trans->transid) { | |
6114 | btrfs_tree_lock(eb); | |
6115 | btrfs_set_lock_blocking(eb); | |
6116 | path->locks[level] = 1; | |
6117 | } | |
6118 | clean_tree_block(trans, root, eb); | |
6119 | } | |
6120 | ||
6121 | if (eb == root->node) { | |
6122 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6123 | parent = eb->start; | |
6124 | else | |
6125 | BUG_ON(root->root_key.objectid != | |
6126 | btrfs_header_owner(eb)); | |
6127 | } else { | |
6128 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6129 | parent = path->nodes[level + 1]->start; | |
6130 | else | |
6131 | BUG_ON(root->root_key.objectid != | |
6132 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6133 | } |
f82d02d9 | 6134 | |
f0486c68 | 6135 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
6136 | out: |
6137 | wc->refs[level] = 0; | |
6138 | wc->flags[level] = 0; | |
f0486c68 | 6139 | return 0; |
2c47e605 YZ |
6140 | } |
6141 | ||
6142 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6143 | struct btrfs_root *root, | |
6144 | struct btrfs_path *path, | |
6145 | struct walk_control *wc) | |
6146 | { | |
2c47e605 | 6147 | int level = wc->level; |
94fcca9f | 6148 | int lookup_info = 1; |
2c47e605 YZ |
6149 | int ret; |
6150 | ||
6151 | while (level >= 0) { | |
94fcca9f | 6152 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6153 | if (ret > 0) |
6154 | break; | |
6155 | ||
6156 | if (level == 0) | |
6157 | break; | |
6158 | ||
7a7965f8 YZ |
6159 | if (path->slots[level] >= |
6160 | btrfs_header_nritems(path->nodes[level])) | |
6161 | break; | |
6162 | ||
94fcca9f | 6163 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6164 | if (ret > 0) { |
6165 | path->slots[level]++; | |
6166 | continue; | |
90d2c51d MX |
6167 | } else if (ret < 0) |
6168 | return ret; | |
1c4850e2 | 6169 | level = wc->level; |
f82d02d9 | 6170 | } |
f82d02d9 YZ |
6171 | return 0; |
6172 | } | |
6173 | ||
d397712b | 6174 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6175 | struct btrfs_root *root, |
f82d02d9 | 6176 | struct btrfs_path *path, |
2c47e605 | 6177 | struct walk_control *wc, int max_level) |
20524f02 | 6178 | { |
2c47e605 | 6179 | int level = wc->level; |
20524f02 | 6180 | int ret; |
9f3a7427 | 6181 | |
2c47e605 YZ |
6182 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6183 | while (level < max_level && path->nodes[level]) { | |
6184 | wc->level = level; | |
6185 | if (path->slots[level] + 1 < | |
6186 | btrfs_header_nritems(path->nodes[level])) { | |
6187 | path->slots[level]++; | |
20524f02 CM |
6188 | return 0; |
6189 | } else { | |
2c47e605 YZ |
6190 | ret = walk_up_proc(trans, root, path, wc); |
6191 | if (ret > 0) | |
6192 | return 0; | |
bd56b302 | 6193 | |
2c47e605 YZ |
6194 | if (path->locks[level]) { |
6195 | btrfs_tree_unlock(path->nodes[level]); | |
6196 | path->locks[level] = 0; | |
f82d02d9 | 6197 | } |
2c47e605 YZ |
6198 | free_extent_buffer(path->nodes[level]); |
6199 | path->nodes[level] = NULL; | |
6200 | level++; | |
20524f02 CM |
6201 | } |
6202 | } | |
6203 | return 1; | |
6204 | } | |
6205 | ||
9aca1d51 | 6206 | /* |
2c47e605 YZ |
6207 | * drop a subvolume tree. |
6208 | * | |
6209 | * this function traverses the tree freeing any blocks that only | |
6210 | * referenced by the tree. | |
6211 | * | |
6212 | * when a shared tree block is found. this function decreases its | |
6213 | * reference count by one. if update_ref is true, this function | |
6214 | * also make sure backrefs for the shared block and all lower level | |
6215 | * blocks are properly updated. | |
9aca1d51 | 6216 | */ |
3fd0a558 YZ |
6217 | int btrfs_drop_snapshot(struct btrfs_root *root, |
6218 | struct btrfs_block_rsv *block_rsv, int update_ref) | |
20524f02 | 6219 | { |
5caf2a00 | 6220 | struct btrfs_path *path; |
2c47e605 YZ |
6221 | struct btrfs_trans_handle *trans; |
6222 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6223 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6224 | struct walk_control *wc; |
6225 | struct btrfs_key key; | |
6226 | int err = 0; | |
6227 | int ret; | |
6228 | int level; | |
20524f02 | 6229 | |
5caf2a00 CM |
6230 | path = btrfs_alloc_path(); |
6231 | BUG_ON(!path); | |
20524f02 | 6232 | |
2c47e605 YZ |
6233 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
6234 | BUG_ON(!wc); | |
6235 | ||
a22285a6 | 6236 | trans = btrfs_start_transaction(tree_root, 0); |
3fd0a558 YZ |
6237 | if (block_rsv) |
6238 | trans->block_rsv = block_rsv; | |
2c47e605 | 6239 | |
9f3a7427 | 6240 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6241 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6242 | path->nodes[level] = btrfs_lock_root_node(root); |
6243 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6244 | path->slots[level] = 0; |
5d4f98a2 | 6245 | path->locks[level] = 1; |
2c47e605 YZ |
6246 | memset(&wc->update_progress, 0, |
6247 | sizeof(wc->update_progress)); | |
9f3a7427 | 6248 | } else { |
9f3a7427 | 6249 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6250 | memcpy(&wc->update_progress, &key, |
6251 | sizeof(wc->update_progress)); | |
6252 | ||
6702ed49 | 6253 | level = root_item->drop_level; |
2c47e605 | 6254 | BUG_ON(level == 0); |
6702ed49 | 6255 | path->lowest_level = level; |
2c47e605 YZ |
6256 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6257 | path->lowest_level = 0; | |
6258 | if (ret < 0) { | |
6259 | err = ret; | |
9f3a7427 CM |
6260 | goto out; |
6261 | } | |
1c4850e2 | 6262 | WARN_ON(ret > 0); |
2c47e605 | 6263 | |
7d9eb12c CM |
6264 | /* |
6265 | * unlock our path, this is safe because only this | |
6266 | * function is allowed to delete this snapshot | |
6267 | */ | |
5d4f98a2 | 6268 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
6269 | |
6270 | level = btrfs_header_level(root->node); | |
6271 | while (1) { | |
6272 | btrfs_tree_lock(path->nodes[level]); | |
6273 | btrfs_set_lock_blocking(path->nodes[level]); | |
6274 | ||
6275 | ret = btrfs_lookup_extent_info(trans, root, | |
6276 | path->nodes[level]->start, | |
6277 | path->nodes[level]->len, | |
6278 | &wc->refs[level], | |
6279 | &wc->flags[level]); | |
6280 | BUG_ON(ret); | |
6281 | BUG_ON(wc->refs[level] == 0); | |
6282 | ||
6283 | if (level == root_item->drop_level) | |
6284 | break; | |
6285 | ||
6286 | btrfs_tree_unlock(path->nodes[level]); | |
6287 | WARN_ON(wc->refs[level] != 1); | |
6288 | level--; | |
6289 | } | |
9f3a7427 | 6290 | } |
2c47e605 YZ |
6291 | |
6292 | wc->level = level; | |
6293 | wc->shared_level = -1; | |
6294 | wc->stage = DROP_REFERENCE; | |
6295 | wc->update_ref = update_ref; | |
6296 | wc->keep_locks = 0; | |
1c4850e2 | 6297 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 6298 | |
d397712b | 6299 | while (1) { |
2c47e605 YZ |
6300 | ret = walk_down_tree(trans, root, path, wc); |
6301 | if (ret < 0) { | |
6302 | err = ret; | |
20524f02 | 6303 | break; |
2c47e605 | 6304 | } |
9aca1d51 | 6305 | |
2c47e605 YZ |
6306 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
6307 | if (ret < 0) { | |
6308 | err = ret; | |
20524f02 | 6309 | break; |
2c47e605 YZ |
6310 | } |
6311 | ||
6312 | if (ret > 0) { | |
6313 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
6314 | break; |
6315 | } | |
2c47e605 YZ |
6316 | |
6317 | if (wc->stage == DROP_REFERENCE) { | |
6318 | level = wc->level; | |
6319 | btrfs_node_key(path->nodes[level], | |
6320 | &root_item->drop_progress, | |
6321 | path->slots[level]); | |
6322 | root_item->drop_level = level; | |
6323 | } | |
6324 | ||
6325 | BUG_ON(wc->level == 0); | |
3fd0a558 | 6326 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
6327 | ret = btrfs_update_root(trans, tree_root, |
6328 | &root->root_key, | |
6329 | root_item); | |
6330 | BUG_ON(ret); | |
6331 | ||
3fd0a558 | 6332 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 6333 | trans = btrfs_start_transaction(tree_root, 0); |
3fd0a558 YZ |
6334 | if (block_rsv) |
6335 | trans->block_rsv = block_rsv; | |
c3e69d58 | 6336 | } |
20524f02 | 6337 | } |
2c47e605 YZ |
6338 | btrfs_release_path(root, path); |
6339 | BUG_ON(err); | |
6340 | ||
6341 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
6342 | BUG_ON(ret); | |
6343 | ||
76dda93c YZ |
6344 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
6345 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
6346 | NULL, NULL); | |
6347 | BUG_ON(ret < 0); | |
6348 | if (ret > 0) { | |
84cd948c JB |
6349 | /* if we fail to delete the orphan item this time |
6350 | * around, it'll get picked up the next time. | |
6351 | * | |
6352 | * The most common failure here is just -ENOENT. | |
6353 | */ | |
6354 | btrfs_del_orphan_item(trans, tree_root, | |
6355 | root->root_key.objectid); | |
76dda93c YZ |
6356 | } |
6357 | } | |
6358 | ||
6359 | if (root->in_radix) { | |
6360 | btrfs_free_fs_root(tree_root->fs_info, root); | |
6361 | } else { | |
6362 | free_extent_buffer(root->node); | |
6363 | free_extent_buffer(root->commit_root); | |
6364 | kfree(root); | |
6365 | } | |
9f3a7427 | 6366 | out: |
3fd0a558 | 6367 | btrfs_end_transaction_throttle(trans, tree_root); |
2c47e605 | 6368 | kfree(wc); |
5caf2a00 | 6369 | btrfs_free_path(path); |
2c47e605 | 6370 | return err; |
20524f02 | 6371 | } |
9078a3e1 | 6372 | |
2c47e605 YZ |
6373 | /* |
6374 | * drop subtree rooted at tree block 'node'. | |
6375 | * | |
6376 | * NOTE: this function will unlock and release tree block 'node' | |
6377 | */ | |
f82d02d9 YZ |
6378 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
6379 | struct btrfs_root *root, | |
6380 | struct extent_buffer *node, | |
6381 | struct extent_buffer *parent) | |
6382 | { | |
6383 | struct btrfs_path *path; | |
2c47e605 | 6384 | struct walk_control *wc; |
f82d02d9 YZ |
6385 | int level; |
6386 | int parent_level; | |
6387 | int ret = 0; | |
6388 | int wret; | |
6389 | ||
2c47e605 YZ |
6390 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
6391 | ||
f82d02d9 YZ |
6392 | path = btrfs_alloc_path(); |
6393 | BUG_ON(!path); | |
6394 | ||
2c47e605 YZ |
6395 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
6396 | BUG_ON(!wc); | |
6397 | ||
b9447ef8 | 6398 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
6399 | parent_level = btrfs_header_level(parent); |
6400 | extent_buffer_get(parent); | |
6401 | path->nodes[parent_level] = parent; | |
6402 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
6403 | ||
b9447ef8 | 6404 | btrfs_assert_tree_locked(node); |
f82d02d9 | 6405 | level = btrfs_header_level(node); |
f82d02d9 YZ |
6406 | path->nodes[level] = node; |
6407 | path->slots[level] = 0; | |
2c47e605 YZ |
6408 | path->locks[level] = 1; |
6409 | ||
6410 | wc->refs[parent_level] = 1; | |
6411 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6412 | wc->level = level; | |
6413 | wc->shared_level = -1; | |
6414 | wc->stage = DROP_REFERENCE; | |
6415 | wc->update_ref = 0; | |
6416 | wc->keep_locks = 1; | |
1c4850e2 | 6417 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
6418 | |
6419 | while (1) { | |
2c47e605 YZ |
6420 | wret = walk_down_tree(trans, root, path, wc); |
6421 | if (wret < 0) { | |
f82d02d9 | 6422 | ret = wret; |
f82d02d9 | 6423 | break; |
2c47e605 | 6424 | } |
f82d02d9 | 6425 | |
2c47e605 | 6426 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
6427 | if (wret < 0) |
6428 | ret = wret; | |
6429 | if (wret != 0) | |
6430 | break; | |
6431 | } | |
6432 | ||
2c47e605 | 6433 | kfree(wc); |
f82d02d9 YZ |
6434 | btrfs_free_path(path); |
6435 | return ret; | |
6436 | } | |
6437 | ||
5d4f98a2 | 6438 | #if 0 |
8e7bf94f CM |
6439 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
6440 | unsigned long nr) | |
6441 | { | |
6442 | return min(last, start + nr - 1); | |
6443 | } | |
6444 | ||
d397712b | 6445 | static noinline int relocate_inode_pages(struct inode *inode, u64 start, |
98ed5174 | 6446 | u64 len) |
edbd8d4e CM |
6447 | { |
6448 | u64 page_start; | |
6449 | u64 page_end; | |
1a40e23b | 6450 | unsigned long first_index; |
edbd8d4e | 6451 | unsigned long last_index; |
edbd8d4e CM |
6452 | unsigned long i; |
6453 | struct page *page; | |
d1310b2e | 6454 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 6455 | struct file_ra_state *ra; |
3eaa2885 | 6456 | struct btrfs_ordered_extent *ordered; |
1a40e23b ZY |
6457 | unsigned int total_read = 0; |
6458 | unsigned int total_dirty = 0; | |
6459 | int ret = 0; | |
4313b399 CM |
6460 | |
6461 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
6462 | |
6463 | mutex_lock(&inode->i_mutex); | |
1a40e23b | 6464 | first_index = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
6465 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
6466 | ||
1a40e23b ZY |
6467 | /* make sure the dirty trick played by the caller work */ |
6468 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
6469 | first_index, last_index); | |
6470 | if (ret) | |
6471 | goto out_unlock; | |
8e7bf94f | 6472 | |
4313b399 | 6473 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 6474 | |
1a40e23b ZY |
6475 | for (i = first_index ; i <= last_index; i++) { |
6476 | if (total_read % ra->ra_pages == 0) { | |
8e7bf94f | 6477 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, |
1a40e23b | 6478 | calc_ra(i, last_index, ra->ra_pages)); |
8e7bf94f CM |
6479 | } |
6480 | total_read++; | |
3eaa2885 CM |
6481 | again: |
6482 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
1a40e23b | 6483 | BUG_ON(1); |
edbd8d4e | 6484 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 6485 | if (!page) { |
1a40e23b | 6486 | ret = -ENOMEM; |
edbd8d4e | 6487 | goto out_unlock; |
a061fc8d | 6488 | } |
edbd8d4e CM |
6489 | if (!PageUptodate(page)) { |
6490 | btrfs_readpage(NULL, page); | |
6491 | lock_page(page); | |
6492 | if (!PageUptodate(page)) { | |
6493 | unlock_page(page); | |
6494 | page_cache_release(page); | |
1a40e23b | 6495 | ret = -EIO; |
edbd8d4e CM |
6496 | goto out_unlock; |
6497 | } | |
6498 | } | |
ec44a35c | 6499 | wait_on_page_writeback(page); |
3eaa2885 | 6500 | |
edbd8d4e CM |
6501 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
6502 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
d1310b2e | 6503 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 6504 | |
3eaa2885 CM |
6505 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6506 | if (ordered) { | |
6507 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
6508 | unlock_page(page); | |
6509 | page_cache_release(page); | |
6510 | btrfs_start_ordered_extent(inode, ordered, 1); | |
6511 | btrfs_put_ordered_extent(ordered); | |
6512 | goto again; | |
6513 | } | |
6514 | set_page_extent_mapped(page); | |
6515 | ||
1a40e23b ZY |
6516 | if (i == first_index) |
6517 | set_extent_bits(io_tree, page_start, page_end, | |
6518 | EXTENT_BOUNDARY, GFP_NOFS); | |
1f80e4db | 6519 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
1a40e23b | 6520 | |
a061fc8d | 6521 | set_page_dirty(page); |
1a40e23b | 6522 | total_dirty++; |
edbd8d4e | 6523 | |
d1310b2e | 6524 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
6525 | unlock_page(page); |
6526 | page_cache_release(page); | |
6527 | } | |
6528 | ||
6529 | out_unlock: | |
ec44a35c | 6530 | kfree(ra); |
edbd8d4e | 6531 | mutex_unlock(&inode->i_mutex); |
1a40e23b ZY |
6532 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty); |
6533 | return ret; | |
edbd8d4e CM |
6534 | } |
6535 | ||
d397712b | 6536 | static noinline int relocate_data_extent(struct inode *reloc_inode, |
1a40e23b ZY |
6537 | struct btrfs_key *extent_key, |
6538 | u64 offset) | |
6539 | { | |
6540 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
6541 | struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree; | |
6542 | struct extent_map *em; | |
6643558d YZ |
6543 | u64 start = extent_key->objectid - offset; |
6544 | u64 end = start + extent_key->offset - 1; | |
bf4ef679 | 6545 | |
1a40e23b ZY |
6546 | em = alloc_extent_map(GFP_NOFS); |
6547 | BUG_ON(!em || IS_ERR(em)); | |
bf4ef679 | 6548 | |
6643558d | 6549 | em->start = start; |
1a40e23b | 6550 | em->len = extent_key->offset; |
c8b97818 | 6551 | em->block_len = extent_key->offset; |
1a40e23b ZY |
6552 | em->block_start = extent_key->objectid; |
6553 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
6554 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
6555 | ||
6556 | /* setup extent map to cheat btrfs_readpage */ | |
6643558d | 6557 | lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
1a40e23b ZY |
6558 | while (1) { |
6559 | int ret; | |
890871be | 6560 | write_lock(&em_tree->lock); |
1a40e23b | 6561 | ret = add_extent_mapping(em_tree, em); |
890871be | 6562 | write_unlock(&em_tree->lock); |
1a40e23b ZY |
6563 | if (ret != -EEXIST) { |
6564 | free_extent_map(em); | |
bf4ef679 CM |
6565 | break; |
6566 | } | |
6643558d | 6567 | btrfs_drop_extent_cache(reloc_inode, start, end, 0); |
bf4ef679 | 6568 | } |
6643558d | 6569 | unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
bf4ef679 | 6570 | |
6643558d | 6571 | return relocate_inode_pages(reloc_inode, start, extent_key->offset); |
1a40e23b | 6572 | } |
edbd8d4e | 6573 | |
1a40e23b ZY |
6574 | struct btrfs_ref_path { |
6575 | u64 extent_start; | |
6576 | u64 nodes[BTRFS_MAX_LEVEL]; | |
6577 | u64 root_objectid; | |
6578 | u64 root_generation; | |
6579 | u64 owner_objectid; | |
1a40e23b ZY |
6580 | u32 num_refs; |
6581 | int lowest_level; | |
6582 | int current_level; | |
f82d02d9 YZ |
6583 | int shared_level; |
6584 | ||
6585 | struct btrfs_key node_keys[BTRFS_MAX_LEVEL]; | |
6586 | u64 new_nodes[BTRFS_MAX_LEVEL]; | |
1a40e23b | 6587 | }; |
7d9eb12c | 6588 | |
1a40e23b | 6589 | struct disk_extent { |
c8b97818 | 6590 | u64 ram_bytes; |
1a40e23b ZY |
6591 | u64 disk_bytenr; |
6592 | u64 disk_num_bytes; | |
6593 | u64 offset; | |
6594 | u64 num_bytes; | |
c8b97818 CM |
6595 | u8 compression; |
6596 | u8 encryption; | |
6597 | u16 other_encoding; | |
1a40e23b | 6598 | }; |
4313b399 | 6599 | |
1a40e23b ZY |
6600 | static int is_cowonly_root(u64 root_objectid) |
6601 | { | |
6602 | if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || | |
6603 | root_objectid == BTRFS_EXTENT_TREE_OBJECTID || | |
6604 | root_objectid == BTRFS_CHUNK_TREE_OBJECTID || | |
6605 | root_objectid == BTRFS_DEV_TREE_OBJECTID || | |
0403e47e YZ |
6606 | root_objectid == BTRFS_TREE_LOG_OBJECTID || |
6607 | root_objectid == BTRFS_CSUM_TREE_OBJECTID) | |
1a40e23b ZY |
6608 | return 1; |
6609 | return 0; | |
6610 | } | |
edbd8d4e | 6611 | |
d397712b | 6612 | static noinline int __next_ref_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6613 | struct btrfs_root *extent_root, |
6614 | struct btrfs_ref_path *ref_path, | |
6615 | int first_time) | |
6616 | { | |
6617 | struct extent_buffer *leaf; | |
6618 | struct btrfs_path *path; | |
6619 | struct btrfs_extent_ref *ref; | |
6620 | struct btrfs_key key; | |
6621 | struct btrfs_key found_key; | |
6622 | u64 bytenr; | |
6623 | u32 nritems; | |
6624 | int level; | |
6625 | int ret = 1; | |
edbd8d4e | 6626 | |
1a40e23b ZY |
6627 | path = btrfs_alloc_path(); |
6628 | if (!path) | |
6629 | return -ENOMEM; | |
bf4ef679 | 6630 | |
1a40e23b ZY |
6631 | if (first_time) { |
6632 | ref_path->lowest_level = -1; | |
6633 | ref_path->current_level = -1; | |
f82d02d9 | 6634 | ref_path->shared_level = -1; |
1a40e23b ZY |
6635 | goto walk_up; |
6636 | } | |
6637 | walk_down: | |
6638 | level = ref_path->current_level - 1; | |
6639 | while (level >= -1) { | |
6640 | u64 parent; | |
6641 | if (level < ref_path->lowest_level) | |
6642 | break; | |
bf4ef679 | 6643 | |
d397712b | 6644 | if (level >= 0) |
1a40e23b | 6645 | bytenr = ref_path->nodes[level]; |
d397712b | 6646 | else |
1a40e23b | 6647 | bytenr = ref_path->extent_start; |
1a40e23b | 6648 | BUG_ON(bytenr == 0); |
bf4ef679 | 6649 | |
1a40e23b ZY |
6650 | parent = ref_path->nodes[level + 1]; |
6651 | ref_path->nodes[level + 1] = 0; | |
6652 | ref_path->current_level = level; | |
6653 | BUG_ON(parent == 0); | |
0ef3e66b | 6654 | |
1a40e23b ZY |
6655 | key.objectid = bytenr; |
6656 | key.offset = parent + 1; | |
6657 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 6658 | |
1a40e23b ZY |
6659 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
6660 | if (ret < 0) | |
edbd8d4e | 6661 | goto out; |
1a40e23b | 6662 | BUG_ON(ret == 0); |
7d9eb12c | 6663 | |
1a40e23b ZY |
6664 | leaf = path->nodes[0]; |
6665 | nritems = btrfs_header_nritems(leaf); | |
6666 | if (path->slots[0] >= nritems) { | |
6667 | ret = btrfs_next_leaf(extent_root, path); | |
6668 | if (ret < 0) | |
6669 | goto out; | |
6670 | if (ret > 0) | |
6671 | goto next; | |
6672 | leaf = path->nodes[0]; | |
6673 | } | |
0ef3e66b | 6674 | |
1a40e23b ZY |
6675 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6676 | if (found_key.objectid == bytenr && | |
f82d02d9 YZ |
6677 | found_key.type == BTRFS_EXTENT_REF_KEY) { |
6678 | if (level < ref_path->shared_level) | |
6679 | ref_path->shared_level = level; | |
1a40e23b | 6680 | goto found; |
f82d02d9 | 6681 | } |
1a40e23b ZY |
6682 | next: |
6683 | level--; | |
6684 | btrfs_release_path(extent_root, path); | |
d899e052 | 6685 | cond_resched(); |
1a40e23b ZY |
6686 | } |
6687 | /* reached lowest level */ | |
6688 | ret = 1; | |
6689 | goto out; | |
6690 | walk_up: | |
6691 | level = ref_path->current_level; | |
6692 | while (level < BTRFS_MAX_LEVEL - 1) { | |
6693 | u64 ref_objectid; | |
d397712b CM |
6694 | |
6695 | if (level >= 0) | |
1a40e23b | 6696 | bytenr = ref_path->nodes[level]; |
d397712b | 6697 | else |
1a40e23b | 6698 | bytenr = ref_path->extent_start; |
d397712b | 6699 | |
1a40e23b | 6700 | BUG_ON(bytenr == 0); |
edbd8d4e | 6701 | |
1a40e23b ZY |
6702 | key.objectid = bytenr; |
6703 | key.offset = 0; | |
6704 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 6705 | |
1a40e23b ZY |
6706 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
6707 | if (ret < 0) | |
6708 | goto out; | |
edbd8d4e | 6709 | |
1a40e23b ZY |
6710 | leaf = path->nodes[0]; |
6711 | nritems = btrfs_header_nritems(leaf); | |
6712 | if (path->slots[0] >= nritems) { | |
6713 | ret = btrfs_next_leaf(extent_root, path); | |
6714 | if (ret < 0) | |
6715 | goto out; | |
6716 | if (ret > 0) { | |
6717 | /* the extent was freed by someone */ | |
6718 | if (ref_path->lowest_level == level) | |
6719 | goto out; | |
6720 | btrfs_release_path(extent_root, path); | |
6721 | goto walk_down; | |
6722 | } | |
6723 | leaf = path->nodes[0]; | |
6724 | } | |
edbd8d4e | 6725 | |
1a40e23b ZY |
6726 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6727 | if (found_key.objectid != bytenr || | |
6728 | found_key.type != BTRFS_EXTENT_REF_KEY) { | |
6729 | /* the extent was freed by someone */ | |
6730 | if (ref_path->lowest_level == level) { | |
6731 | ret = 1; | |
6732 | goto out; | |
6733 | } | |
6734 | btrfs_release_path(extent_root, path); | |
6735 | goto walk_down; | |
6736 | } | |
6737 | found: | |
6738 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
6739 | struct btrfs_extent_ref); | |
6740 | ref_objectid = btrfs_ref_objectid(leaf, ref); | |
6741 | if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
6742 | if (first_time) { | |
6743 | level = (int)ref_objectid; | |
6744 | BUG_ON(level >= BTRFS_MAX_LEVEL); | |
6745 | ref_path->lowest_level = level; | |
6746 | ref_path->current_level = level; | |
6747 | ref_path->nodes[level] = bytenr; | |
6748 | } else { | |
6749 | WARN_ON(ref_objectid != level); | |
6750 | } | |
6751 | } else { | |
6752 | WARN_ON(level != -1); | |
6753 | } | |
6754 | first_time = 0; | |
bf4ef679 | 6755 | |
1a40e23b ZY |
6756 | if (ref_path->lowest_level == level) { |
6757 | ref_path->owner_objectid = ref_objectid; | |
1a40e23b ZY |
6758 | ref_path->num_refs = btrfs_ref_num_refs(leaf, ref); |
6759 | } | |
bf4ef679 | 6760 | |
7d9eb12c | 6761 | /* |
1a40e23b ZY |
6762 | * the block is tree root or the block isn't in reference |
6763 | * counted tree. | |
7d9eb12c | 6764 | */ |
1a40e23b ZY |
6765 | if (found_key.objectid == found_key.offset || |
6766 | is_cowonly_root(btrfs_ref_root(leaf, ref))) { | |
6767 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
6768 | ref_path->root_generation = | |
6769 | btrfs_ref_generation(leaf, ref); | |
6770 | if (level < 0) { | |
6771 | /* special reference from the tree log */ | |
6772 | ref_path->nodes[0] = found_key.offset; | |
6773 | ref_path->current_level = 0; | |
6774 | } | |
6775 | ret = 0; | |
6776 | goto out; | |
6777 | } | |
7d9eb12c | 6778 | |
1a40e23b ZY |
6779 | level++; |
6780 | BUG_ON(ref_path->nodes[level] != 0); | |
6781 | ref_path->nodes[level] = found_key.offset; | |
6782 | ref_path->current_level = level; | |
bf4ef679 | 6783 | |
1a40e23b ZY |
6784 | /* |
6785 | * the reference was created in the running transaction, | |
6786 | * no need to continue walking up. | |
6787 | */ | |
6788 | if (btrfs_ref_generation(leaf, ref) == trans->transid) { | |
6789 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
6790 | ref_path->root_generation = | |
6791 | btrfs_ref_generation(leaf, ref); | |
6792 | ret = 0; | |
6793 | goto out; | |
7d9eb12c CM |
6794 | } |
6795 | ||
1a40e23b | 6796 | btrfs_release_path(extent_root, path); |
d899e052 | 6797 | cond_resched(); |
7d9eb12c | 6798 | } |
1a40e23b ZY |
6799 | /* reached max tree level, but no tree root found. */ |
6800 | BUG(); | |
edbd8d4e | 6801 | out: |
1a40e23b ZY |
6802 | btrfs_free_path(path); |
6803 | return ret; | |
edbd8d4e CM |
6804 | } |
6805 | ||
1a40e23b ZY |
6806 | static int btrfs_first_ref_path(struct btrfs_trans_handle *trans, |
6807 | struct btrfs_root *extent_root, | |
6808 | struct btrfs_ref_path *ref_path, | |
6809 | u64 extent_start) | |
a061fc8d | 6810 | { |
1a40e23b ZY |
6811 | memset(ref_path, 0, sizeof(*ref_path)); |
6812 | ref_path->extent_start = extent_start; | |
a061fc8d | 6813 | |
1a40e23b | 6814 | return __next_ref_path(trans, extent_root, ref_path, 1); |
a061fc8d CM |
6815 | } |
6816 | ||
1a40e23b ZY |
6817 | static int btrfs_next_ref_path(struct btrfs_trans_handle *trans, |
6818 | struct btrfs_root *extent_root, | |
6819 | struct btrfs_ref_path *ref_path) | |
edbd8d4e | 6820 | { |
1a40e23b ZY |
6821 | return __next_ref_path(trans, extent_root, ref_path, 0); |
6822 | } | |
6823 | ||
d397712b | 6824 | static noinline int get_new_locations(struct inode *reloc_inode, |
1a40e23b ZY |
6825 | struct btrfs_key *extent_key, |
6826 | u64 offset, int no_fragment, | |
6827 | struct disk_extent **extents, | |
6828 | int *nr_extents) | |
6829 | { | |
6830 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
6831 | struct btrfs_path *path; | |
6832 | struct btrfs_file_extent_item *fi; | |
edbd8d4e | 6833 | struct extent_buffer *leaf; |
1a40e23b ZY |
6834 | struct disk_extent *exts = *extents; |
6835 | struct btrfs_key found_key; | |
6836 | u64 cur_pos; | |
6837 | u64 last_byte; | |
edbd8d4e | 6838 | u32 nritems; |
1a40e23b ZY |
6839 | int nr = 0; |
6840 | int max = *nr_extents; | |
6841 | int ret; | |
edbd8d4e | 6842 | |
1a40e23b ZY |
6843 | WARN_ON(!no_fragment && *extents); |
6844 | if (!exts) { | |
6845 | max = 1; | |
6846 | exts = kmalloc(sizeof(*exts) * max, GFP_NOFS); | |
6847 | if (!exts) | |
6848 | return -ENOMEM; | |
a061fc8d | 6849 | } |
edbd8d4e | 6850 | |
1a40e23b ZY |
6851 | path = btrfs_alloc_path(); |
6852 | BUG_ON(!path); | |
edbd8d4e | 6853 | |
1a40e23b ZY |
6854 | cur_pos = extent_key->objectid - offset; |
6855 | last_byte = extent_key->objectid + extent_key->offset; | |
6856 | ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino, | |
6857 | cur_pos, 0); | |
6858 | if (ret < 0) | |
6859 | goto out; | |
6860 | if (ret > 0) { | |
6861 | ret = -ENOENT; | |
6862 | goto out; | |
6863 | } | |
edbd8d4e | 6864 | |
1a40e23b | 6865 | while (1) { |
edbd8d4e CM |
6866 | leaf = path->nodes[0]; |
6867 | nritems = btrfs_header_nritems(leaf); | |
1a40e23b ZY |
6868 | if (path->slots[0] >= nritems) { |
6869 | ret = btrfs_next_leaf(root, path); | |
a061fc8d CM |
6870 | if (ret < 0) |
6871 | goto out; | |
1a40e23b ZY |
6872 | if (ret > 0) |
6873 | break; | |
bf4ef679 | 6874 | leaf = path->nodes[0]; |
a061fc8d | 6875 | } |
edbd8d4e CM |
6876 | |
6877 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b ZY |
6878 | if (found_key.offset != cur_pos || |
6879 | found_key.type != BTRFS_EXTENT_DATA_KEY || | |
6880 | found_key.objectid != reloc_inode->i_ino) | |
edbd8d4e CM |
6881 | break; |
6882 | ||
1a40e23b ZY |
6883 | fi = btrfs_item_ptr(leaf, path->slots[0], |
6884 | struct btrfs_file_extent_item); | |
6885 | if (btrfs_file_extent_type(leaf, fi) != | |
6886 | BTRFS_FILE_EXTENT_REG || | |
6887 | btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
edbd8d4e | 6888 | break; |
1a40e23b ZY |
6889 | |
6890 | if (nr == max) { | |
6891 | struct disk_extent *old = exts; | |
6892 | max *= 2; | |
6893 | exts = kzalloc(sizeof(*exts) * max, GFP_NOFS); | |
6894 | memcpy(exts, old, sizeof(*exts) * nr); | |
6895 | if (old != *extents) | |
6896 | kfree(old); | |
a061fc8d | 6897 | } |
edbd8d4e | 6898 | |
1a40e23b ZY |
6899 | exts[nr].disk_bytenr = |
6900 | btrfs_file_extent_disk_bytenr(leaf, fi); | |
6901 | exts[nr].disk_num_bytes = | |
6902 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6903 | exts[nr].offset = btrfs_file_extent_offset(leaf, fi); | |
6904 | exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
c8b97818 CM |
6905 | exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
6906 | exts[nr].compression = btrfs_file_extent_compression(leaf, fi); | |
6907 | exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi); | |
6908 | exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf, | |
6909 | fi); | |
d899e052 YZ |
6910 | BUG_ON(exts[nr].offset > 0); |
6911 | BUG_ON(exts[nr].compression || exts[nr].encryption); | |
6912 | BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes); | |
edbd8d4e | 6913 | |
1a40e23b ZY |
6914 | cur_pos += exts[nr].num_bytes; |
6915 | nr++; | |
6916 | ||
6917 | if (cur_pos + offset >= last_byte) | |
6918 | break; | |
6919 | ||
6920 | if (no_fragment) { | |
6921 | ret = 1; | |
edbd8d4e | 6922 | goto out; |
1a40e23b ZY |
6923 | } |
6924 | path->slots[0]++; | |
6925 | } | |
6926 | ||
1f80e4db | 6927 | BUG_ON(cur_pos + offset > last_byte); |
1a40e23b ZY |
6928 | if (cur_pos + offset < last_byte) { |
6929 | ret = -ENOENT; | |
6930 | goto out; | |
edbd8d4e CM |
6931 | } |
6932 | ret = 0; | |
6933 | out: | |
1a40e23b ZY |
6934 | btrfs_free_path(path); |
6935 | if (ret) { | |
6936 | if (exts != *extents) | |
6937 | kfree(exts); | |
6938 | } else { | |
6939 | *extents = exts; | |
6940 | *nr_extents = nr; | |
6941 | } | |
6942 | return ret; | |
6943 | } | |
6944 | ||
d397712b | 6945 | static noinline int replace_one_extent(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6946 | struct btrfs_root *root, |
6947 | struct btrfs_path *path, | |
6948 | struct btrfs_key *extent_key, | |
6949 | struct btrfs_key *leaf_key, | |
6950 | struct btrfs_ref_path *ref_path, | |
6951 | struct disk_extent *new_extents, | |
6952 | int nr_extents) | |
6953 | { | |
6954 | struct extent_buffer *leaf; | |
6955 | struct btrfs_file_extent_item *fi; | |
6956 | struct inode *inode = NULL; | |
6957 | struct btrfs_key key; | |
6958 | u64 lock_start = 0; | |
6959 | u64 lock_end = 0; | |
6960 | u64 num_bytes; | |
6961 | u64 ext_offset; | |
86288a19 | 6962 | u64 search_end = (u64)-1; |
1a40e23b | 6963 | u32 nritems; |
3bb1a1bc | 6964 | int nr_scaned = 0; |
1a40e23b | 6965 | int extent_locked = 0; |
d899e052 | 6966 | int extent_type; |
1a40e23b ZY |
6967 | int ret; |
6968 | ||
3bb1a1bc | 6969 | memcpy(&key, leaf_key, sizeof(key)); |
1a40e23b | 6970 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { |
3bb1a1bc YZ |
6971 | if (key.objectid < ref_path->owner_objectid || |
6972 | (key.objectid == ref_path->owner_objectid && | |
6973 | key.type < BTRFS_EXTENT_DATA_KEY)) { | |
6974 | key.objectid = ref_path->owner_objectid; | |
6975 | key.type = BTRFS_EXTENT_DATA_KEY; | |
6976 | key.offset = 0; | |
6977 | } | |
1a40e23b ZY |
6978 | } |
6979 | ||
6980 | while (1) { | |
6981 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
6982 | if (ret < 0) | |
6983 | goto out; | |
6984 | ||
6985 | leaf = path->nodes[0]; | |
6986 | nritems = btrfs_header_nritems(leaf); | |
6987 | next: | |
6988 | if (extent_locked && ret > 0) { | |
6989 | /* | |
6990 | * the file extent item was modified by someone | |
6991 | * before the extent got locked. | |
6992 | */ | |
1a40e23b ZY |
6993 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
6994 | lock_end, GFP_NOFS); | |
6995 | extent_locked = 0; | |
6996 | } | |
6997 | ||
6998 | if (path->slots[0] >= nritems) { | |
3bb1a1bc | 6999 | if (++nr_scaned > 2) |
1a40e23b ZY |
7000 | break; |
7001 | ||
7002 | BUG_ON(extent_locked); | |
7003 | ret = btrfs_next_leaf(root, path); | |
7004 | if (ret < 0) | |
7005 | goto out; | |
7006 | if (ret > 0) | |
7007 | break; | |
7008 | leaf = path->nodes[0]; | |
7009 | nritems = btrfs_header_nritems(leaf); | |
7010 | } | |
7011 | ||
7012 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
7013 | ||
7014 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { | |
7015 | if ((key.objectid > ref_path->owner_objectid) || | |
7016 | (key.objectid == ref_path->owner_objectid && | |
7017 | key.type > BTRFS_EXTENT_DATA_KEY) || | |
86288a19 | 7018 | key.offset >= search_end) |
1a40e23b ZY |
7019 | break; |
7020 | } | |
7021 | ||
7022 | if (inode && key.objectid != inode->i_ino) { | |
7023 | BUG_ON(extent_locked); | |
7024 | btrfs_release_path(root, path); | |
7025 | mutex_unlock(&inode->i_mutex); | |
7026 | iput(inode); | |
7027 | inode = NULL; | |
7028 | continue; | |
7029 | } | |
7030 | ||
7031 | if (key.type != BTRFS_EXTENT_DATA_KEY) { | |
7032 | path->slots[0]++; | |
7033 | ret = 1; | |
7034 | goto next; | |
7035 | } | |
7036 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
7037 | struct btrfs_file_extent_item); | |
d899e052 YZ |
7038 | extent_type = btrfs_file_extent_type(leaf, fi); |
7039 | if ((extent_type != BTRFS_FILE_EXTENT_REG && | |
7040 | extent_type != BTRFS_FILE_EXTENT_PREALLOC) || | |
1a40e23b ZY |
7041 | (btrfs_file_extent_disk_bytenr(leaf, fi) != |
7042 | extent_key->objectid)) { | |
7043 | path->slots[0]++; | |
7044 | ret = 1; | |
7045 | goto next; | |
7046 | } | |
7047 | ||
7048 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
7049 | ext_offset = btrfs_file_extent_offset(leaf, fi); | |
7050 | ||
86288a19 YZ |
7051 | if (search_end == (u64)-1) { |
7052 | search_end = key.offset - ext_offset + | |
7053 | btrfs_file_extent_ram_bytes(leaf, fi); | |
7054 | } | |
1a40e23b ZY |
7055 | |
7056 | if (!extent_locked) { | |
7057 | lock_start = key.offset; | |
7058 | lock_end = lock_start + num_bytes - 1; | |
7059 | } else { | |
6643558d YZ |
7060 | if (lock_start > key.offset || |
7061 | lock_end + 1 < key.offset + num_bytes) { | |
7062 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
7063 | lock_start, lock_end, GFP_NOFS); | |
7064 | extent_locked = 0; | |
7065 | } | |
1a40e23b ZY |
7066 | } |
7067 | ||
7068 | if (!inode) { | |
7069 | btrfs_release_path(root, path); | |
7070 | ||
7071 | inode = btrfs_iget_locked(root->fs_info->sb, | |
7072 | key.objectid, root); | |
7073 | if (inode->i_state & I_NEW) { | |
7074 | BTRFS_I(inode)->root = root; | |
7075 | BTRFS_I(inode)->location.objectid = | |
7076 | key.objectid; | |
7077 | BTRFS_I(inode)->location.type = | |
7078 | BTRFS_INODE_ITEM_KEY; | |
7079 | BTRFS_I(inode)->location.offset = 0; | |
7080 | btrfs_read_locked_inode(inode); | |
7081 | unlock_new_inode(inode); | |
7082 | } | |
7083 | /* | |
7084 | * some code call btrfs_commit_transaction while | |
7085 | * holding the i_mutex, so we can't use mutex_lock | |
7086 | * here. | |
7087 | */ | |
7088 | if (is_bad_inode(inode) || | |
7089 | !mutex_trylock(&inode->i_mutex)) { | |
7090 | iput(inode); | |
7091 | inode = NULL; | |
7092 | key.offset = (u64)-1; | |
7093 | goto skip; | |
7094 | } | |
7095 | } | |
7096 | ||
7097 | if (!extent_locked) { | |
7098 | struct btrfs_ordered_extent *ordered; | |
7099 | ||
7100 | btrfs_release_path(root, path); | |
7101 | ||
7102 | lock_extent(&BTRFS_I(inode)->io_tree, lock_start, | |
7103 | lock_end, GFP_NOFS); | |
7104 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
7105 | lock_end); | |
7106 | if (ordered && | |
7107 | ordered->file_offset <= lock_end && | |
7108 | ordered->file_offset + ordered->len > lock_start) { | |
7109 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
7110 | lock_start, lock_end, GFP_NOFS); | |
7111 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7112 | btrfs_put_ordered_extent(ordered); | |
7113 | key.offset += num_bytes; | |
7114 | goto skip; | |
7115 | } | |
7116 | if (ordered) | |
7117 | btrfs_put_ordered_extent(ordered); | |
7118 | ||
1a40e23b ZY |
7119 | extent_locked = 1; |
7120 | continue; | |
7121 | } | |
7122 | ||
7123 | if (nr_extents == 1) { | |
7124 | /* update extent pointer in place */ | |
1a40e23b ZY |
7125 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
7126 | new_extents[0].disk_bytenr); | |
7127 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
7128 | new_extents[0].disk_num_bytes); | |
1a40e23b ZY |
7129 | btrfs_mark_buffer_dirty(leaf); |
7130 | ||
7131 | btrfs_drop_extent_cache(inode, key.offset, | |
7132 | key.offset + num_bytes - 1, 0); | |
7133 | ||
7134 | ret = btrfs_inc_extent_ref(trans, root, | |
7135 | new_extents[0].disk_bytenr, | |
7136 | new_extents[0].disk_num_bytes, | |
7137 | leaf->start, | |
7138 | root->root_key.objectid, | |
7139 | trans->transid, | |
3bb1a1bc | 7140 | key.objectid); |
1a40e23b ZY |
7141 | BUG_ON(ret); |
7142 | ||
7143 | ret = btrfs_free_extent(trans, root, | |
7144 | extent_key->objectid, | |
7145 | extent_key->offset, | |
7146 | leaf->start, | |
7147 | btrfs_header_owner(leaf), | |
7148 | btrfs_header_generation(leaf), | |
3bb1a1bc | 7149 | key.objectid, 0); |
1a40e23b ZY |
7150 | BUG_ON(ret); |
7151 | ||
7152 | btrfs_release_path(root, path); | |
7153 | key.offset += num_bytes; | |
7154 | } else { | |
d899e052 YZ |
7155 | BUG_ON(1); |
7156 | #if 0 | |
1a40e23b ZY |
7157 | u64 alloc_hint; |
7158 | u64 extent_len; | |
7159 | int i; | |
7160 | /* | |
7161 | * drop old extent pointer at first, then insert the | |
7162 | * new pointers one bye one | |
7163 | */ | |
7164 | btrfs_release_path(root, path); | |
7165 | ret = btrfs_drop_extents(trans, root, inode, key.offset, | |
7166 | key.offset + num_bytes, | |
7167 | key.offset, &alloc_hint); | |
7168 | BUG_ON(ret); | |
7169 | ||
7170 | for (i = 0; i < nr_extents; i++) { | |
7171 | if (ext_offset >= new_extents[i].num_bytes) { | |
7172 | ext_offset -= new_extents[i].num_bytes; | |
7173 | continue; | |
7174 | } | |
7175 | extent_len = min(new_extents[i].num_bytes - | |
7176 | ext_offset, num_bytes); | |
7177 | ||
7178 | ret = btrfs_insert_empty_item(trans, root, | |
7179 | path, &key, | |
7180 | sizeof(*fi)); | |
7181 | BUG_ON(ret); | |
7182 | ||
7183 | leaf = path->nodes[0]; | |
7184 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
7185 | struct btrfs_file_extent_item); | |
7186 | btrfs_set_file_extent_generation(leaf, fi, | |
7187 | trans->transid); | |
7188 | btrfs_set_file_extent_type(leaf, fi, | |
7189 | BTRFS_FILE_EXTENT_REG); | |
7190 | btrfs_set_file_extent_disk_bytenr(leaf, fi, | |
7191 | new_extents[i].disk_bytenr); | |
7192 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
7193 | new_extents[i].disk_num_bytes); | |
c8b97818 CM |
7194 | btrfs_set_file_extent_ram_bytes(leaf, fi, |
7195 | new_extents[i].ram_bytes); | |
7196 | ||
7197 | btrfs_set_file_extent_compression(leaf, fi, | |
7198 | new_extents[i].compression); | |
7199 | btrfs_set_file_extent_encryption(leaf, fi, | |
7200 | new_extents[i].encryption); | |
7201 | btrfs_set_file_extent_other_encoding(leaf, fi, | |
7202 | new_extents[i].other_encoding); | |
7203 | ||
1a40e23b ZY |
7204 | btrfs_set_file_extent_num_bytes(leaf, fi, |
7205 | extent_len); | |
7206 | ext_offset += new_extents[i].offset; | |
7207 | btrfs_set_file_extent_offset(leaf, fi, | |
7208 | ext_offset); | |
7209 | btrfs_mark_buffer_dirty(leaf); | |
7210 | ||
7211 | btrfs_drop_extent_cache(inode, key.offset, | |
7212 | key.offset + extent_len - 1, 0); | |
7213 | ||
7214 | ret = btrfs_inc_extent_ref(trans, root, | |
7215 | new_extents[i].disk_bytenr, | |
7216 | new_extents[i].disk_num_bytes, | |
7217 | leaf->start, | |
7218 | root->root_key.objectid, | |
3bb1a1bc | 7219 | trans->transid, key.objectid); |
1a40e23b ZY |
7220 | BUG_ON(ret); |
7221 | btrfs_release_path(root, path); | |
7222 | ||
a76a3cd4 | 7223 | inode_add_bytes(inode, extent_len); |
1a40e23b ZY |
7224 | |
7225 | ext_offset = 0; | |
7226 | num_bytes -= extent_len; | |
7227 | key.offset += extent_len; | |
7228 | ||
7229 | if (num_bytes == 0) | |
7230 | break; | |
7231 | } | |
7232 | BUG_ON(i >= nr_extents); | |
d899e052 | 7233 | #endif |
1a40e23b ZY |
7234 | } |
7235 | ||
7236 | if (extent_locked) { | |
1a40e23b ZY |
7237 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
7238 | lock_end, GFP_NOFS); | |
7239 | extent_locked = 0; | |
7240 | } | |
7241 | skip: | |
7242 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS && | |
86288a19 | 7243 | key.offset >= search_end) |
1a40e23b ZY |
7244 | break; |
7245 | ||
7246 | cond_resched(); | |
7247 | } | |
7248 | ret = 0; | |
7249 | out: | |
7250 | btrfs_release_path(root, path); | |
7251 | if (inode) { | |
7252 | mutex_unlock(&inode->i_mutex); | |
7253 | if (extent_locked) { | |
1a40e23b ZY |
7254 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
7255 | lock_end, GFP_NOFS); | |
7256 | } | |
7257 | iput(inode); | |
7258 | } | |
7259 | return ret; | |
7260 | } | |
7261 | ||
1a40e23b ZY |
7262 | int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, |
7263 | struct btrfs_root *root, | |
7264 | struct extent_buffer *buf, u64 orig_start) | |
7265 | { | |
7266 | int level; | |
7267 | int ret; | |
7268 | ||
7269 | BUG_ON(btrfs_header_generation(buf) != trans->transid); | |
7270 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
7271 | ||
7272 | level = btrfs_header_level(buf); | |
7273 | if (level == 0) { | |
7274 | struct btrfs_leaf_ref *ref; | |
7275 | struct btrfs_leaf_ref *orig_ref; | |
7276 | ||
7277 | orig_ref = btrfs_lookup_leaf_ref(root, orig_start); | |
7278 | if (!orig_ref) | |
7279 | return -ENOENT; | |
7280 | ||
7281 | ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems); | |
7282 | if (!ref) { | |
7283 | btrfs_free_leaf_ref(root, orig_ref); | |
7284 | return -ENOMEM; | |
7285 | } | |
7286 | ||
7287 | ref->nritems = orig_ref->nritems; | |
7288 | memcpy(ref->extents, orig_ref->extents, | |
7289 | sizeof(ref->extents[0]) * ref->nritems); | |
7290 | ||
7291 | btrfs_free_leaf_ref(root, orig_ref); | |
7292 | ||
7293 | ref->root_gen = trans->transid; | |
7294 | ref->bytenr = buf->start; | |
7295 | ref->owner = btrfs_header_owner(buf); | |
7296 | ref->generation = btrfs_header_generation(buf); | |
bd56b302 | 7297 | |
1a40e23b ZY |
7298 | ret = btrfs_add_leaf_ref(root, ref, 0); |
7299 | WARN_ON(ret); | |
7300 | btrfs_free_leaf_ref(root, ref); | |
7301 | } | |
7302 | return 0; | |
7303 | } | |
7304 | ||
d397712b | 7305 | static noinline int invalidate_extent_cache(struct btrfs_root *root, |
1a40e23b ZY |
7306 | struct extent_buffer *leaf, |
7307 | struct btrfs_block_group_cache *group, | |
7308 | struct btrfs_root *target_root) | |
7309 | { | |
7310 | struct btrfs_key key; | |
7311 | struct inode *inode = NULL; | |
7312 | struct btrfs_file_extent_item *fi; | |
2ac55d41 | 7313 | struct extent_state *cached_state = NULL; |
1a40e23b ZY |
7314 | u64 num_bytes; |
7315 | u64 skip_objectid = 0; | |
7316 | u32 nritems; | |
7317 | u32 i; | |
7318 | ||
7319 | nritems = btrfs_header_nritems(leaf); | |
7320 | for (i = 0; i < nritems; i++) { | |
7321 | btrfs_item_key_to_cpu(leaf, &key, i); | |
7322 | if (key.objectid == skip_objectid || | |
7323 | key.type != BTRFS_EXTENT_DATA_KEY) | |
7324 | continue; | |
7325 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
7326 | if (btrfs_file_extent_type(leaf, fi) == | |
7327 | BTRFS_FILE_EXTENT_INLINE) | |
7328 | continue; | |
7329 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
7330 | continue; | |
7331 | if (!inode || inode->i_ino != key.objectid) { | |
7332 | iput(inode); | |
7333 | inode = btrfs_ilookup(target_root->fs_info->sb, | |
7334 | key.objectid, target_root, 1); | |
7335 | } | |
7336 | if (!inode) { | |
7337 | skip_objectid = key.objectid; | |
7338 | continue; | |
7339 | } | |
7340 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
7341 | ||
2ac55d41 JB |
7342 | lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset, |
7343 | key.offset + num_bytes - 1, 0, &cached_state, | |
7344 | GFP_NOFS); | |
1a40e23b ZY |
7345 | btrfs_drop_extent_cache(inode, key.offset, |
7346 | key.offset + num_bytes - 1, 1); | |
2ac55d41 JB |
7347 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset, |
7348 | key.offset + num_bytes - 1, &cached_state, | |
7349 | GFP_NOFS); | |
1a40e23b ZY |
7350 | cond_resched(); |
7351 | } | |
7352 | iput(inode); | |
7353 | return 0; | |
7354 | } | |
7355 | ||
d397712b | 7356 | static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
7357 | struct btrfs_root *root, |
7358 | struct extent_buffer *leaf, | |
7359 | struct btrfs_block_group_cache *group, | |
7360 | struct inode *reloc_inode) | |
7361 | { | |
7362 | struct btrfs_key key; | |
7363 | struct btrfs_key extent_key; | |
7364 | struct btrfs_file_extent_item *fi; | |
7365 | struct btrfs_leaf_ref *ref; | |
7366 | struct disk_extent *new_extent; | |
7367 | u64 bytenr; | |
7368 | u64 num_bytes; | |
7369 | u32 nritems; | |
7370 | u32 i; | |
7371 | int ext_index; | |
7372 | int nr_extent; | |
7373 | int ret; | |
7374 | ||
7375 | new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS); | |
7376 | BUG_ON(!new_extent); | |
7377 | ||
7378 | ref = btrfs_lookup_leaf_ref(root, leaf->start); | |
7379 | BUG_ON(!ref); | |
7380 | ||
7381 | ext_index = -1; | |
7382 | nritems = btrfs_header_nritems(leaf); | |
7383 | for (i = 0; i < nritems; i++) { | |
7384 | btrfs_item_key_to_cpu(leaf, &key, i); | |
7385 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
7386 | continue; | |
7387 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
7388 | if (btrfs_file_extent_type(leaf, fi) == | |
7389 | BTRFS_FILE_EXTENT_INLINE) | |
7390 | continue; | |
7391 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
7392 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7393 | if (bytenr == 0) | |
7394 | continue; | |
7395 | ||
7396 | ext_index++; | |
7397 | if (bytenr >= group->key.objectid + group->key.offset || | |
7398 | bytenr + num_bytes <= group->key.objectid) | |
7399 | continue; | |
7400 | ||
7401 | extent_key.objectid = bytenr; | |
7402 | extent_key.offset = num_bytes; | |
7403 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
7404 | nr_extent = 1; | |
7405 | ret = get_new_locations(reloc_inode, &extent_key, | |
7406 | group->key.objectid, 1, | |
7407 | &new_extent, &nr_extent); | |
7408 | if (ret > 0) | |
7409 | continue; | |
7410 | BUG_ON(ret < 0); | |
7411 | ||
7412 | BUG_ON(ref->extents[ext_index].bytenr != bytenr); | |
7413 | BUG_ON(ref->extents[ext_index].num_bytes != num_bytes); | |
7414 | ref->extents[ext_index].bytenr = new_extent->disk_bytenr; | |
7415 | ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes; | |
7416 | ||
1a40e23b ZY |
7417 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
7418 | new_extent->disk_bytenr); | |
7419 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
7420 | new_extent->disk_num_bytes); | |
1a40e23b ZY |
7421 | btrfs_mark_buffer_dirty(leaf); |
7422 | ||
7423 | ret = btrfs_inc_extent_ref(trans, root, | |
7424 | new_extent->disk_bytenr, | |
7425 | new_extent->disk_num_bytes, | |
7426 | leaf->start, | |
7427 | root->root_key.objectid, | |
3bb1a1bc | 7428 | trans->transid, key.objectid); |
1a40e23b | 7429 | BUG_ON(ret); |
56bec294 | 7430 | |
1a40e23b ZY |
7431 | ret = btrfs_free_extent(trans, root, |
7432 | bytenr, num_bytes, leaf->start, | |
7433 | btrfs_header_owner(leaf), | |
7434 | btrfs_header_generation(leaf), | |
3bb1a1bc | 7435 | key.objectid, 0); |
1a40e23b ZY |
7436 | BUG_ON(ret); |
7437 | cond_resched(); | |
7438 | } | |
7439 | kfree(new_extent); | |
7440 | BUG_ON(ext_index + 1 != ref->nritems); | |
7441 | btrfs_free_leaf_ref(root, ref); | |
7442 | return 0; | |
7443 | } | |
7444 | ||
f82d02d9 YZ |
7445 | int btrfs_free_reloc_root(struct btrfs_trans_handle *trans, |
7446 | struct btrfs_root *root) | |
1a40e23b ZY |
7447 | { |
7448 | struct btrfs_root *reloc_root; | |
f82d02d9 | 7449 | int ret; |
1a40e23b ZY |
7450 | |
7451 | if (root->reloc_root) { | |
7452 | reloc_root = root->reloc_root; | |
7453 | root->reloc_root = NULL; | |
7454 | list_add(&reloc_root->dead_list, | |
7455 | &root->fs_info->dead_reloc_roots); | |
f82d02d9 YZ |
7456 | |
7457 | btrfs_set_root_bytenr(&reloc_root->root_item, | |
7458 | reloc_root->node->start); | |
7459 | btrfs_set_root_level(&root->root_item, | |
7460 | btrfs_header_level(reloc_root->node)); | |
7461 | memset(&reloc_root->root_item.drop_progress, 0, | |
7462 | sizeof(struct btrfs_disk_key)); | |
7463 | reloc_root->root_item.drop_level = 0; | |
7464 | ||
7465 | ret = btrfs_update_root(trans, root->fs_info->tree_root, | |
7466 | &reloc_root->root_key, | |
7467 | &reloc_root->root_item); | |
7468 | BUG_ON(ret); | |
1a40e23b ZY |
7469 | } |
7470 | return 0; | |
7471 | } | |
7472 | ||
7473 | int btrfs_drop_dead_reloc_roots(struct btrfs_root *root) | |
7474 | { | |
7475 | struct btrfs_trans_handle *trans; | |
7476 | struct btrfs_root *reloc_root; | |
7477 | struct btrfs_root *prev_root = NULL; | |
7478 | struct list_head dead_roots; | |
7479 | int ret; | |
7480 | unsigned long nr; | |
7481 | ||
7482 | INIT_LIST_HEAD(&dead_roots); | |
7483 | list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots); | |
7484 | ||
7485 | while (!list_empty(&dead_roots)) { | |
7486 | reloc_root = list_entry(dead_roots.prev, | |
7487 | struct btrfs_root, dead_list); | |
7488 | list_del_init(&reloc_root->dead_list); | |
7489 | ||
7490 | BUG_ON(reloc_root->commit_root != NULL); | |
7491 | while (1) { | |
7492 | trans = btrfs_join_transaction(root, 1); | |
3612b495 | 7493 | BUG_ON(IS_ERR(trans)); |
1a40e23b ZY |
7494 | |
7495 | mutex_lock(&root->fs_info->drop_mutex); | |
7496 | ret = btrfs_drop_snapshot(trans, reloc_root); | |
7497 | if (ret != -EAGAIN) | |
7498 | break; | |
7499 | mutex_unlock(&root->fs_info->drop_mutex); | |
7500 | ||
7501 | nr = trans->blocks_used; | |
7502 | ret = btrfs_end_transaction(trans, root); | |
7503 | BUG_ON(ret); | |
7504 | btrfs_btree_balance_dirty(root, nr); | |
7505 | } | |
7506 | ||
7507 | free_extent_buffer(reloc_root->node); | |
7508 | ||
7509 | ret = btrfs_del_root(trans, root->fs_info->tree_root, | |
7510 | &reloc_root->root_key); | |
7511 | BUG_ON(ret); | |
7512 | mutex_unlock(&root->fs_info->drop_mutex); | |
7513 | ||
7514 | nr = trans->blocks_used; | |
7515 | ret = btrfs_end_transaction(trans, root); | |
7516 | BUG_ON(ret); | |
7517 | btrfs_btree_balance_dirty(root, nr); | |
7518 | ||
7519 | kfree(prev_root); | |
7520 | prev_root = reloc_root; | |
7521 | } | |
7522 | if (prev_root) { | |
7523 | btrfs_remove_leaf_refs(prev_root, (u64)-1, 0); | |
7524 | kfree(prev_root); | |
7525 | } | |
7526 | return 0; | |
7527 | } | |
7528 | ||
7529 | int btrfs_add_dead_reloc_root(struct btrfs_root *root) | |
7530 | { | |
7531 | list_add(&root->dead_list, &root->fs_info->dead_reloc_roots); | |
7532 | return 0; | |
7533 | } | |
7534 | ||
7535 | int btrfs_cleanup_reloc_trees(struct btrfs_root *root) | |
7536 | { | |
7537 | struct btrfs_root *reloc_root; | |
7538 | struct btrfs_trans_handle *trans; | |
7539 | struct btrfs_key location; | |
7540 | int found; | |
7541 | int ret; | |
7542 | ||
7543 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
7544 | ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL); | |
7545 | BUG_ON(ret); | |
7546 | found = !list_empty(&root->fs_info->dead_reloc_roots); | |
7547 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
7548 | ||
7549 | if (found) { | |
7550 | trans = btrfs_start_transaction(root, 1); | |
7551 | BUG_ON(!trans); | |
7552 | ret = btrfs_commit_transaction(trans, root); | |
7553 | BUG_ON(ret); | |
7554 | } | |
7555 | ||
7556 | location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
7557 | location.offset = (u64)-1; | |
7558 | location.type = BTRFS_ROOT_ITEM_KEY; | |
7559 | ||
7560 | reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location); | |
7561 | BUG_ON(!reloc_root); | |
7562 | btrfs_orphan_cleanup(reloc_root); | |
7563 | return 0; | |
7564 | } | |
7565 | ||
d397712b | 7566 | static noinline int init_reloc_tree(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
7567 | struct btrfs_root *root) |
7568 | { | |
7569 | struct btrfs_root *reloc_root; | |
7570 | struct extent_buffer *eb; | |
7571 | struct btrfs_root_item *root_item; | |
7572 | struct btrfs_key root_key; | |
7573 | int ret; | |
7574 | ||
7575 | BUG_ON(!root->ref_cows); | |
7576 | if (root->reloc_root) | |
7577 | return 0; | |
7578 | ||
7579 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); | |
7580 | BUG_ON(!root_item); | |
7581 | ||
7582 | ret = btrfs_copy_root(trans, root, root->commit_root, | |
7583 | &eb, BTRFS_TREE_RELOC_OBJECTID); | |
7584 | BUG_ON(ret); | |
7585 | ||
7586 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
7587 | root_key.offset = root->root_key.objectid; | |
7588 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
7589 | ||
7590 | memcpy(root_item, &root->root_item, sizeof(root_item)); | |
7591 | btrfs_set_root_refs(root_item, 0); | |
7592 | btrfs_set_root_bytenr(root_item, eb->start); | |
7593 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
84234f3a | 7594 | btrfs_set_root_generation(root_item, trans->transid); |
1a40e23b ZY |
7595 | |
7596 | btrfs_tree_unlock(eb); | |
7597 | free_extent_buffer(eb); | |
7598 | ||
7599 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, | |
7600 | &root_key, root_item); | |
7601 | BUG_ON(ret); | |
7602 | kfree(root_item); | |
7603 | ||
7604 | reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root, | |
7605 | &root_key); | |
7606 | BUG_ON(!reloc_root); | |
7607 | reloc_root->last_trans = trans->transid; | |
7608 | reloc_root->commit_root = NULL; | |
7609 | reloc_root->ref_tree = &root->fs_info->reloc_ref_tree; | |
7610 | ||
7611 | root->reloc_root = reloc_root; | |
7612 | return 0; | |
7613 | } | |
7614 | ||
7615 | /* | |
7616 | * Core function of space balance. | |
7617 | * | |
7618 | * The idea is using reloc trees to relocate tree blocks in reference | |
f82d02d9 YZ |
7619 | * counted roots. There is one reloc tree for each subvol, and all |
7620 | * reloc trees share same root key objectid. Reloc trees are snapshots | |
7621 | * of the latest committed roots of subvols (root->commit_root). | |
7622 | * | |
7623 | * To relocate a tree block referenced by a subvol, there are two steps. | |
7624 | * COW the block through subvol's reloc tree, then update block pointer | |
7625 | * in the subvol to point to the new block. Since all reloc trees share | |
7626 | * same root key objectid, doing special handing for tree blocks owned | |
7627 | * by them is easy. Once a tree block has been COWed in one reloc tree, | |
7628 | * we can use the resulting new block directly when the same block is | |
7629 | * required to COW again through other reloc trees. By this way, relocated | |
7630 | * tree blocks are shared between reloc trees, so they are also shared | |
7631 | * between subvols. | |
1a40e23b | 7632 | */ |
d397712b | 7633 | static noinline int relocate_one_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
7634 | struct btrfs_root *root, |
7635 | struct btrfs_path *path, | |
7636 | struct btrfs_key *first_key, | |
7637 | struct btrfs_ref_path *ref_path, | |
7638 | struct btrfs_block_group_cache *group, | |
7639 | struct inode *reloc_inode) | |
7640 | { | |
7641 | struct btrfs_root *reloc_root; | |
7642 | struct extent_buffer *eb = NULL; | |
7643 | struct btrfs_key *keys; | |
7644 | u64 *nodes; | |
7645 | int level; | |
f82d02d9 | 7646 | int shared_level; |
1a40e23b | 7647 | int lowest_level = 0; |
1a40e23b ZY |
7648 | int ret; |
7649 | ||
7650 | if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
7651 | lowest_level = ref_path->owner_objectid; | |
7652 | ||
f82d02d9 | 7653 | if (!root->ref_cows) { |
1a40e23b ZY |
7654 | path->lowest_level = lowest_level; |
7655 | ret = btrfs_search_slot(trans, root, first_key, path, 0, 1); | |
7656 | BUG_ON(ret < 0); | |
7657 | path->lowest_level = 0; | |
7658 | btrfs_release_path(root, path); | |
7659 | return 0; | |
7660 | } | |
7661 | ||
1a40e23b ZY |
7662 | mutex_lock(&root->fs_info->tree_reloc_mutex); |
7663 | ret = init_reloc_tree(trans, root); | |
7664 | BUG_ON(ret); | |
7665 | reloc_root = root->reloc_root; | |
7666 | ||
f82d02d9 YZ |
7667 | shared_level = ref_path->shared_level; |
7668 | ref_path->shared_level = BTRFS_MAX_LEVEL - 1; | |
1a40e23b | 7669 | |
f82d02d9 YZ |
7670 | keys = ref_path->node_keys; |
7671 | nodes = ref_path->new_nodes; | |
7672 | memset(&keys[shared_level + 1], 0, | |
7673 | sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
7674 | memset(&nodes[shared_level + 1], 0, | |
7675 | sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
1a40e23b | 7676 | |
f82d02d9 YZ |
7677 | if (nodes[lowest_level] == 0) { |
7678 | path->lowest_level = lowest_level; | |
7679 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, | |
7680 | 0, 1); | |
7681 | BUG_ON(ret); | |
7682 | for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) { | |
7683 | eb = path->nodes[level]; | |
7684 | if (!eb || eb == reloc_root->node) | |
7685 | break; | |
7686 | nodes[level] = eb->start; | |
7687 | if (level == 0) | |
7688 | btrfs_item_key_to_cpu(eb, &keys[level], 0); | |
7689 | else | |
7690 | btrfs_node_key_to_cpu(eb, &keys[level], 0); | |
7691 | } | |
2b82032c YZ |
7692 | if (nodes[0] && |
7693 | ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
7694 | eb = path->nodes[0]; |
7695 | ret = replace_extents_in_leaf(trans, reloc_root, eb, | |
7696 | group, reloc_inode); | |
7697 | BUG_ON(ret); | |
7698 | } | |
7699 | btrfs_release_path(reloc_root, path); | |
7700 | } else { | |
1a40e23b | 7701 | ret = btrfs_merge_path(trans, reloc_root, keys, nodes, |
f82d02d9 | 7702 | lowest_level); |
1a40e23b ZY |
7703 | BUG_ON(ret); |
7704 | } | |
7705 | ||
1a40e23b ZY |
7706 | /* |
7707 | * replace tree blocks in the fs tree with tree blocks in | |
7708 | * the reloc tree. | |
7709 | */ | |
7710 | ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level); | |
7711 | BUG_ON(ret < 0); | |
7712 | ||
7713 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
7714 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, |
7715 | 0, 0); | |
7716 | BUG_ON(ret); | |
7717 | extent_buffer_get(path->nodes[0]); | |
7718 | eb = path->nodes[0]; | |
7719 | btrfs_release_path(reloc_root, path); | |
1a40e23b ZY |
7720 | ret = invalidate_extent_cache(reloc_root, eb, group, root); |
7721 | BUG_ON(ret); | |
7722 | free_extent_buffer(eb); | |
7723 | } | |
1a40e23b | 7724 | |
f82d02d9 | 7725 | mutex_unlock(&root->fs_info->tree_reloc_mutex); |
1a40e23b | 7726 | path->lowest_level = 0; |
1a40e23b ZY |
7727 | return 0; |
7728 | } | |
7729 | ||
d397712b | 7730 | static noinline int relocate_tree_block(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
7731 | struct btrfs_root *root, |
7732 | struct btrfs_path *path, | |
7733 | struct btrfs_key *first_key, | |
7734 | struct btrfs_ref_path *ref_path) | |
7735 | { | |
7736 | int ret; | |
1a40e23b ZY |
7737 | |
7738 | ret = relocate_one_path(trans, root, path, first_key, | |
7739 | ref_path, NULL, NULL); | |
7740 | BUG_ON(ret); | |
7741 | ||
1a40e23b ZY |
7742 | return 0; |
7743 | } | |
7744 | ||
d397712b | 7745 | static noinline int del_extent_zero(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
7746 | struct btrfs_root *extent_root, |
7747 | struct btrfs_path *path, | |
7748 | struct btrfs_key *extent_key) | |
7749 | { | |
7750 | int ret; | |
7751 | ||
1a40e23b ZY |
7752 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); |
7753 | if (ret) | |
7754 | goto out; | |
7755 | ret = btrfs_del_item(trans, extent_root, path); | |
7756 | out: | |
7757 | btrfs_release_path(extent_root, path); | |
1a40e23b ZY |
7758 | return ret; |
7759 | } | |
7760 | ||
d397712b | 7761 | static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info, |
1a40e23b ZY |
7762 | struct btrfs_ref_path *ref_path) |
7763 | { | |
7764 | struct btrfs_key root_key; | |
7765 | ||
7766 | root_key.objectid = ref_path->root_objectid; | |
7767 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
7768 | if (is_cowonly_root(ref_path->root_objectid)) | |
7769 | root_key.offset = 0; | |
7770 | else | |
7771 | root_key.offset = (u64)-1; | |
7772 | ||
7773 | return btrfs_read_fs_root_no_name(fs_info, &root_key); | |
7774 | } | |
7775 | ||
d397712b | 7776 | static noinline int relocate_one_extent(struct btrfs_root *extent_root, |
1a40e23b ZY |
7777 | struct btrfs_path *path, |
7778 | struct btrfs_key *extent_key, | |
7779 | struct btrfs_block_group_cache *group, | |
7780 | struct inode *reloc_inode, int pass) | |
7781 | { | |
7782 | struct btrfs_trans_handle *trans; | |
7783 | struct btrfs_root *found_root; | |
7784 | struct btrfs_ref_path *ref_path = NULL; | |
7785 | struct disk_extent *new_extents = NULL; | |
7786 | int nr_extents = 0; | |
7787 | int loops; | |
7788 | int ret; | |
7789 | int level; | |
7790 | struct btrfs_key first_key; | |
7791 | u64 prev_block = 0; | |
7792 | ||
1a40e23b ZY |
7793 | |
7794 | trans = btrfs_start_transaction(extent_root, 1); | |
7795 | BUG_ON(!trans); | |
7796 | ||
7797 | if (extent_key->objectid == 0) { | |
7798 | ret = del_extent_zero(trans, extent_root, path, extent_key); | |
7799 | goto out; | |
7800 | } | |
7801 | ||
7802 | ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS); | |
7803 | if (!ref_path) { | |
d397712b CM |
7804 | ret = -ENOMEM; |
7805 | goto out; | |
1a40e23b ZY |
7806 | } |
7807 | ||
7808 | for (loops = 0; ; loops++) { | |
7809 | if (loops == 0) { | |
7810 | ret = btrfs_first_ref_path(trans, extent_root, ref_path, | |
7811 | extent_key->objectid); | |
7812 | } else { | |
7813 | ret = btrfs_next_ref_path(trans, extent_root, ref_path); | |
7814 | } | |
7815 | if (ret < 0) | |
7816 | goto out; | |
7817 | if (ret > 0) | |
7818 | break; | |
7819 | ||
7820 | if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID || | |
7821 | ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
7822 | continue; | |
7823 | ||
7824 | found_root = read_ref_root(extent_root->fs_info, ref_path); | |
7825 | BUG_ON(!found_root); | |
7826 | /* | |
7827 | * for reference counted tree, only process reference paths | |
7828 | * rooted at the latest committed root. | |
7829 | */ | |
7830 | if (found_root->ref_cows && | |
7831 | ref_path->root_generation != found_root->root_key.offset) | |
7832 | continue; | |
7833 | ||
7834 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
7835 | if (pass == 0) { | |
7836 | /* | |
7837 | * copy data extents to new locations | |
7838 | */ | |
7839 | u64 group_start = group->key.objectid; | |
7840 | ret = relocate_data_extent(reloc_inode, | |
7841 | extent_key, | |
7842 | group_start); | |
7843 | if (ret < 0) | |
7844 | goto out; | |
7845 | break; | |
7846 | } | |
7847 | level = 0; | |
7848 | } else { | |
7849 | level = ref_path->owner_objectid; | |
7850 | } | |
7851 | ||
7852 | if (prev_block != ref_path->nodes[level]) { | |
7853 | struct extent_buffer *eb; | |
7854 | u64 block_start = ref_path->nodes[level]; | |
7855 | u64 block_size = btrfs_level_size(found_root, level); | |
7856 | ||
7857 | eb = read_tree_block(found_root, block_start, | |
7858 | block_size, 0); | |
7859 | btrfs_tree_lock(eb); | |
7860 | BUG_ON(level != btrfs_header_level(eb)); | |
7861 | ||
7862 | if (level == 0) | |
7863 | btrfs_item_key_to_cpu(eb, &first_key, 0); | |
7864 | else | |
7865 | btrfs_node_key_to_cpu(eb, &first_key, 0); | |
7866 | ||
7867 | btrfs_tree_unlock(eb); | |
7868 | free_extent_buffer(eb); | |
7869 | prev_block = block_start; | |
7870 | } | |
7871 | ||
24562425 | 7872 | mutex_lock(&extent_root->fs_info->trans_mutex); |
e4404d6e | 7873 | btrfs_record_root_in_trans(found_root); |
24562425 | 7874 | mutex_unlock(&extent_root->fs_info->trans_mutex); |
e4404d6e YZ |
7875 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { |
7876 | /* | |
7877 | * try to update data extent references while | |
7878 | * keeping metadata shared between snapshots. | |
7879 | */ | |
7880 | if (pass == 1) { | |
7881 | ret = relocate_one_path(trans, found_root, | |
7882 | path, &first_key, ref_path, | |
7883 | group, reloc_inode); | |
7884 | if (ret < 0) | |
7885 | goto out; | |
7886 | continue; | |
7887 | } | |
1a40e23b ZY |
7888 | /* |
7889 | * use fallback method to process the remaining | |
7890 | * references. | |
7891 | */ | |
7892 | if (!new_extents) { | |
7893 | u64 group_start = group->key.objectid; | |
d899e052 YZ |
7894 | new_extents = kmalloc(sizeof(*new_extents), |
7895 | GFP_NOFS); | |
7896 | nr_extents = 1; | |
1a40e23b ZY |
7897 | ret = get_new_locations(reloc_inode, |
7898 | extent_key, | |
d899e052 | 7899 | group_start, 1, |
1a40e23b ZY |
7900 | &new_extents, |
7901 | &nr_extents); | |
d899e052 | 7902 | if (ret) |
1a40e23b ZY |
7903 | goto out; |
7904 | } | |
1a40e23b ZY |
7905 | ret = replace_one_extent(trans, found_root, |
7906 | path, extent_key, | |
7907 | &first_key, ref_path, | |
7908 | new_extents, nr_extents); | |
e4404d6e | 7909 | } else { |
1a40e23b ZY |
7910 | ret = relocate_tree_block(trans, found_root, path, |
7911 | &first_key, ref_path); | |
1a40e23b ZY |
7912 | } |
7913 | if (ret < 0) | |
7914 | goto out; | |
7915 | } | |
7916 | ret = 0; | |
7917 | out: | |
7918 | btrfs_end_transaction(trans, extent_root); | |
7919 | kfree(new_extents); | |
7920 | kfree(ref_path); | |
1a40e23b ZY |
7921 | return ret; |
7922 | } | |
5d4f98a2 | 7923 | #endif |
1a40e23b | 7924 | |
ec44a35c CM |
7925 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7926 | { | |
7927 | u64 num_devices; | |
7928 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
7929 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
7930 | ||
cd02dca5 CM |
7931 | /* |
7932 | * we add in the count of missing devices because we want | |
7933 | * to make sure that any RAID levels on a degraded FS | |
7934 | * continue to be honored. | |
7935 | */ | |
7936 | num_devices = root->fs_info->fs_devices->rw_devices + | |
7937 | root->fs_info->fs_devices->missing_devices; | |
7938 | ||
ec44a35c CM |
7939 | if (num_devices == 1) { |
7940 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7941 | stripped = flags & ~stripped; | |
7942 | ||
7943 | /* turn raid0 into single device chunks */ | |
7944 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7945 | return stripped; | |
7946 | ||
7947 | /* turn mirroring into duplication */ | |
7948 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7949 | BTRFS_BLOCK_GROUP_RAID10)) | |
7950 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
7951 | return flags; | |
7952 | } else { | |
7953 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7954 | if (flags & stripped) |
7955 | return flags; | |
7956 | ||
7957 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7958 | stripped = flags & ~stripped; | |
7959 | ||
7960 | /* switch duplicated blocks with raid1 */ | |
7961 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7962 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7963 | ||
7964 | /* turn single device chunks into raid0 */ | |
7965 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
7966 | } | |
7967 | return flags; | |
7968 | } | |
7969 | ||
f0486c68 | 7970 | static int set_block_group_ro(struct btrfs_block_group_cache *cache) |
0ef3e66b | 7971 | { |
f0486c68 YZ |
7972 | struct btrfs_space_info *sinfo = cache->space_info; |
7973 | u64 num_bytes; | |
7974 | int ret = -ENOSPC; | |
0ef3e66b | 7975 | |
f0486c68 YZ |
7976 | if (cache->ro) |
7977 | return 0; | |
c286ac48 | 7978 | |
f0486c68 YZ |
7979 | spin_lock(&sinfo->lock); |
7980 | spin_lock(&cache->lock); | |
7981 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7982 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7983 | ||
7984 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
7985 | sinfo->bytes_may_use + sinfo->bytes_readonly + | |
65e5341b | 7986 | cache->reserved_pinned + num_bytes <= sinfo->total_bytes) { |
f0486c68 YZ |
7987 | sinfo->bytes_readonly += num_bytes; |
7988 | sinfo->bytes_reserved += cache->reserved_pinned; | |
7989 | cache->reserved_pinned = 0; | |
7990 | cache->ro = 1; | |
7991 | ret = 0; | |
7992 | } | |
65e5341b | 7993 | |
f0486c68 YZ |
7994 | spin_unlock(&cache->lock); |
7995 | spin_unlock(&sinfo->lock); | |
7996 | return ret; | |
7997 | } | |
7d9eb12c | 7998 | |
f0486c68 YZ |
7999 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
8000 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 8001 | |
f0486c68 YZ |
8002 | { |
8003 | struct btrfs_trans_handle *trans; | |
8004 | u64 alloc_flags; | |
8005 | int ret; | |
7d9eb12c | 8006 | |
f0486c68 | 8007 | BUG_ON(cache->ro); |
0ef3e66b | 8008 | |
f0486c68 YZ |
8009 | trans = btrfs_join_transaction(root, 1); |
8010 | BUG_ON(IS_ERR(trans)); | |
5d4f98a2 | 8011 | |
f0486c68 YZ |
8012 | alloc_flags = update_block_group_flags(root, cache->flags); |
8013 | if (alloc_flags != cache->flags) | |
8014 | do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1); | |
5d4f98a2 | 8015 | |
f0486c68 YZ |
8016 | ret = set_block_group_ro(cache); |
8017 | if (!ret) | |
8018 | goto out; | |
8019 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
8020 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1); | |
8021 | if (ret < 0) | |
8022 | goto out; | |
8023 | ret = set_block_group_ro(cache); | |
8024 | out: | |
8025 | btrfs_end_transaction(trans, root); | |
8026 | return ret; | |
8027 | } | |
5d4f98a2 | 8028 | |
6d07bcec MX |
8029 | /* |
8030 | * helper to account the unused space of all the readonly block group in the | |
8031 | * list. takes mirrors into account. | |
8032 | */ | |
8033 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
8034 | { | |
8035 | struct btrfs_block_group_cache *block_group; | |
8036 | u64 free_bytes = 0; | |
8037 | int factor; | |
8038 | ||
8039 | list_for_each_entry(block_group, groups_list, list) { | |
8040 | spin_lock(&block_group->lock); | |
8041 | ||
8042 | if (!block_group->ro) { | |
8043 | spin_unlock(&block_group->lock); | |
8044 | continue; | |
8045 | } | |
8046 | ||
8047 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
8048 | BTRFS_BLOCK_GROUP_RAID10 | | |
8049 | BTRFS_BLOCK_GROUP_DUP)) | |
8050 | factor = 2; | |
8051 | else | |
8052 | factor = 1; | |
8053 | ||
8054 | free_bytes += (block_group->key.offset - | |
8055 | btrfs_block_group_used(&block_group->item)) * | |
8056 | factor; | |
8057 | ||
8058 | spin_unlock(&block_group->lock); | |
8059 | } | |
8060 | ||
8061 | return free_bytes; | |
8062 | } | |
8063 | ||
8064 | /* | |
8065 | * helper to account the unused space of all the readonly block group in the | |
8066 | * space_info. takes mirrors into account. | |
8067 | */ | |
8068 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
8069 | { | |
8070 | int i; | |
8071 | u64 free_bytes = 0; | |
8072 | ||
8073 | spin_lock(&sinfo->lock); | |
8074 | ||
8075 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
8076 | if (!list_empty(&sinfo->block_groups[i])) | |
8077 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
8078 | &sinfo->block_groups[i]); | |
8079 | ||
8080 | spin_unlock(&sinfo->lock); | |
8081 | ||
8082 | return free_bytes; | |
8083 | } | |
8084 | ||
f0486c68 YZ |
8085 | int btrfs_set_block_group_rw(struct btrfs_root *root, |
8086 | struct btrfs_block_group_cache *cache) | |
5d4f98a2 | 8087 | { |
f0486c68 YZ |
8088 | struct btrfs_space_info *sinfo = cache->space_info; |
8089 | u64 num_bytes; | |
8090 | ||
8091 | BUG_ON(!cache->ro); | |
8092 | ||
8093 | spin_lock(&sinfo->lock); | |
8094 | spin_lock(&cache->lock); | |
8095 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
8096 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
8097 | sinfo->bytes_readonly -= num_bytes; | |
8098 | cache->ro = 0; | |
8099 | spin_unlock(&cache->lock); | |
8100 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
8101 | return 0; |
8102 | } | |
8103 | ||
ba1bf481 JB |
8104 | /* |
8105 | * checks to see if its even possible to relocate this block group. | |
8106 | * | |
8107 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
8108 | * ok to go ahead and try. | |
8109 | */ | |
8110 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 8111 | { |
ba1bf481 JB |
8112 | struct btrfs_block_group_cache *block_group; |
8113 | struct btrfs_space_info *space_info; | |
8114 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
8115 | struct btrfs_device *device; | |
8116 | int full = 0; | |
8117 | int ret = 0; | |
1a40e23b | 8118 | |
ba1bf481 | 8119 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 8120 | |
ba1bf481 JB |
8121 | /* odd, couldn't find the block group, leave it alone */ |
8122 | if (!block_group) | |
8123 | return -1; | |
1a40e23b | 8124 | |
ba1bf481 JB |
8125 | /* no bytes used, we're good */ |
8126 | if (!btrfs_block_group_used(&block_group->item)) | |
1a40e23b ZY |
8127 | goto out; |
8128 | ||
ba1bf481 JB |
8129 | space_info = block_group->space_info; |
8130 | spin_lock(&space_info->lock); | |
17d217fe | 8131 | |
ba1bf481 | 8132 | full = space_info->full; |
17d217fe | 8133 | |
ba1bf481 JB |
8134 | /* |
8135 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
8136 | * relocate it unless we're able to allocate a new chunk below. |
8137 | * | |
8138 | * Otherwise, we need to make sure we have room in the space to handle | |
8139 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 8140 | */ |
7ce618db CM |
8141 | if ((space_info->total_bytes != block_group->key.offset) && |
8142 | (space_info->bytes_used + space_info->bytes_reserved + | |
ba1bf481 JB |
8143 | space_info->bytes_pinned + space_info->bytes_readonly + |
8144 | btrfs_block_group_used(&block_group->item) < | |
7ce618db | 8145 | space_info->total_bytes)) { |
ba1bf481 JB |
8146 | spin_unlock(&space_info->lock); |
8147 | goto out; | |
17d217fe | 8148 | } |
ba1bf481 | 8149 | spin_unlock(&space_info->lock); |
ea8c2819 | 8150 | |
ba1bf481 JB |
8151 | /* |
8152 | * ok we don't have enough space, but maybe we have free space on our | |
8153 | * devices to allocate new chunks for relocation, so loop through our | |
8154 | * alloc devices and guess if we have enough space. However, if we | |
8155 | * were marked as full, then we know there aren't enough chunks, and we | |
8156 | * can just return. | |
8157 | */ | |
8158 | ret = -1; | |
8159 | if (full) | |
8160 | goto out; | |
ea8c2819 | 8161 | |
ba1bf481 JB |
8162 | mutex_lock(&root->fs_info->chunk_mutex); |
8163 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
8164 | u64 min_free = btrfs_block_group_used(&block_group->item); | |
7bfc837d | 8165 | u64 dev_offset; |
56bec294 | 8166 | |
ba1bf481 JB |
8167 | /* |
8168 | * check to make sure we can actually find a chunk with enough | |
8169 | * space to fit our block group in. | |
8170 | */ | |
8171 | if (device->total_bytes > device->bytes_used + min_free) { | |
8172 | ret = find_free_dev_extent(NULL, device, min_free, | |
7bfc837d | 8173 | &dev_offset, NULL); |
ba1bf481 | 8174 | if (!ret) |
73e48b27 | 8175 | break; |
ba1bf481 | 8176 | ret = -1; |
725c8463 | 8177 | } |
edbd8d4e | 8178 | } |
ba1bf481 | 8179 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 8180 | out: |
ba1bf481 | 8181 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
8182 | return ret; |
8183 | } | |
8184 | ||
b2950863 CH |
8185 | static int find_first_block_group(struct btrfs_root *root, |
8186 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 8187 | { |
925baedd | 8188 | int ret = 0; |
0b86a832 CM |
8189 | struct btrfs_key found_key; |
8190 | struct extent_buffer *leaf; | |
8191 | int slot; | |
edbd8d4e | 8192 | |
0b86a832 CM |
8193 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
8194 | if (ret < 0) | |
925baedd CM |
8195 | goto out; |
8196 | ||
d397712b | 8197 | while (1) { |
0b86a832 | 8198 | slot = path->slots[0]; |
edbd8d4e | 8199 | leaf = path->nodes[0]; |
0b86a832 CM |
8200 | if (slot >= btrfs_header_nritems(leaf)) { |
8201 | ret = btrfs_next_leaf(root, path); | |
8202 | if (ret == 0) | |
8203 | continue; | |
8204 | if (ret < 0) | |
925baedd | 8205 | goto out; |
0b86a832 | 8206 | break; |
edbd8d4e | 8207 | } |
0b86a832 | 8208 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 8209 | |
0b86a832 | 8210 | if (found_key.objectid >= key->objectid && |
925baedd CM |
8211 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
8212 | ret = 0; | |
8213 | goto out; | |
8214 | } | |
0b86a832 | 8215 | path->slots[0]++; |
edbd8d4e | 8216 | } |
925baedd | 8217 | out: |
0b86a832 | 8218 | return ret; |
edbd8d4e CM |
8219 | } |
8220 | ||
0af3d00b JB |
8221 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
8222 | { | |
8223 | struct btrfs_block_group_cache *block_group; | |
8224 | u64 last = 0; | |
8225 | ||
8226 | while (1) { | |
8227 | struct inode *inode; | |
8228 | ||
8229 | block_group = btrfs_lookup_first_block_group(info, last); | |
8230 | while (block_group) { | |
8231 | spin_lock(&block_group->lock); | |
8232 | if (block_group->iref) | |
8233 | break; | |
8234 | spin_unlock(&block_group->lock); | |
8235 | block_group = next_block_group(info->tree_root, | |
8236 | block_group); | |
8237 | } | |
8238 | if (!block_group) { | |
8239 | if (last == 0) | |
8240 | break; | |
8241 | last = 0; | |
8242 | continue; | |
8243 | } | |
8244 | ||
8245 | inode = block_group->inode; | |
8246 | block_group->iref = 0; | |
8247 | block_group->inode = NULL; | |
8248 | spin_unlock(&block_group->lock); | |
8249 | iput(inode); | |
8250 | last = block_group->key.objectid + block_group->key.offset; | |
8251 | btrfs_put_block_group(block_group); | |
8252 | } | |
8253 | } | |
8254 | ||
1a40e23b ZY |
8255 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
8256 | { | |
8257 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 8258 | struct btrfs_space_info *space_info; |
11833d66 | 8259 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
8260 | struct rb_node *n; |
8261 | ||
11833d66 YZ |
8262 | down_write(&info->extent_commit_sem); |
8263 | while (!list_empty(&info->caching_block_groups)) { | |
8264 | caching_ctl = list_entry(info->caching_block_groups.next, | |
8265 | struct btrfs_caching_control, list); | |
8266 | list_del(&caching_ctl->list); | |
8267 | put_caching_control(caching_ctl); | |
8268 | } | |
8269 | up_write(&info->extent_commit_sem); | |
8270 | ||
1a40e23b ZY |
8271 | spin_lock(&info->block_group_cache_lock); |
8272 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
8273 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
8274 | cache_node); | |
1a40e23b ZY |
8275 | rb_erase(&block_group->cache_node, |
8276 | &info->block_group_cache_tree); | |
d899e052 YZ |
8277 | spin_unlock(&info->block_group_cache_lock); |
8278 | ||
80eb234a | 8279 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 8280 | list_del(&block_group->list); |
80eb234a | 8281 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 8282 | |
817d52f8 | 8283 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8284 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
8285 | |
8286 | btrfs_remove_free_space_cache(block_group); | |
11dfe35a | 8287 | btrfs_put_block_group(block_group); |
d899e052 YZ |
8288 | |
8289 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
8290 | } |
8291 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
8292 | |
8293 | /* now that all the block groups are freed, go through and | |
8294 | * free all the space_info structs. This is only called during | |
8295 | * the final stages of unmount, and so we know nobody is | |
8296 | * using them. We call synchronize_rcu() once before we start, | |
8297 | * just to be on the safe side. | |
8298 | */ | |
8299 | synchronize_rcu(); | |
8300 | ||
8929ecfa YZ |
8301 | release_global_block_rsv(info); |
8302 | ||
4184ea7f CM |
8303 | while(!list_empty(&info->space_info)) { |
8304 | space_info = list_entry(info->space_info.next, | |
8305 | struct btrfs_space_info, | |
8306 | list); | |
f0486c68 YZ |
8307 | if (space_info->bytes_pinned > 0 || |
8308 | space_info->bytes_reserved > 0) { | |
8309 | WARN_ON(1); | |
8310 | dump_space_info(space_info, 0, 0); | |
8311 | } | |
4184ea7f CM |
8312 | list_del(&space_info->list); |
8313 | kfree(space_info); | |
8314 | } | |
1a40e23b ZY |
8315 | return 0; |
8316 | } | |
8317 | ||
b742bb82 YZ |
8318 | static void __link_block_group(struct btrfs_space_info *space_info, |
8319 | struct btrfs_block_group_cache *cache) | |
8320 | { | |
8321 | int index = get_block_group_index(cache); | |
8322 | ||
8323 | down_write(&space_info->groups_sem); | |
8324 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
8325 | up_write(&space_info->groups_sem); | |
8326 | } | |
8327 | ||
9078a3e1 CM |
8328 | int btrfs_read_block_groups(struct btrfs_root *root) |
8329 | { | |
8330 | struct btrfs_path *path; | |
8331 | int ret; | |
9078a3e1 | 8332 | struct btrfs_block_group_cache *cache; |
be744175 | 8333 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 8334 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
8335 | struct btrfs_key key; |
8336 | struct btrfs_key found_key; | |
5f39d397 | 8337 | struct extent_buffer *leaf; |
0af3d00b JB |
8338 | int need_clear = 0; |
8339 | u64 cache_gen; | |
96b5179d | 8340 | |
be744175 | 8341 | root = info->extent_root; |
9078a3e1 | 8342 | key.objectid = 0; |
0b86a832 | 8343 | key.offset = 0; |
9078a3e1 | 8344 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
8345 | path = btrfs_alloc_path(); |
8346 | if (!path) | |
8347 | return -ENOMEM; | |
8348 | ||
0af3d00b JB |
8349 | cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy); |
8350 | if (cache_gen != 0 && | |
8351 | btrfs_super_generation(&root->fs_info->super_copy) != cache_gen) | |
8352 | need_clear = 1; | |
88c2ba3b JB |
8353 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
8354 | need_clear = 1; | |
8216ef86 JB |
8355 | if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen) |
8356 | printk(KERN_INFO "btrfs: disk space caching is enabled\n"); | |
0af3d00b | 8357 | |
d397712b | 8358 | while (1) { |
0b86a832 | 8359 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
8360 | if (ret > 0) |
8361 | break; | |
0b86a832 CM |
8362 | if (ret != 0) |
8363 | goto error; | |
5f39d397 CM |
8364 | leaf = path->nodes[0]; |
8365 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 8366 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 8367 | if (!cache) { |
0b86a832 | 8368 | ret = -ENOMEM; |
f0486c68 | 8369 | goto error; |
9078a3e1 | 8370 | } |
3e1ad54f | 8371 | |
d2fb3437 | 8372 | atomic_set(&cache->count, 1); |
c286ac48 | 8373 | spin_lock_init(&cache->lock); |
6226cb0a | 8374 | spin_lock_init(&cache->tree_lock); |
817d52f8 | 8375 | cache->fs_info = info; |
0f9dd46c | 8376 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8377 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 8378 | |
0af3d00b JB |
8379 | if (need_clear) |
8380 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
8381 | ||
96303081 JB |
8382 | /* |
8383 | * we only want to have 32k of ram per block group for keeping | |
8384 | * track of free space, and if we pass 1/2 of that we want to | |
8385 | * start converting things over to using bitmaps | |
8386 | */ | |
8387 | cache->extents_thresh = ((1024 * 32) / 2) / | |
8388 | sizeof(struct btrfs_free_space); | |
8389 | ||
5f39d397 CM |
8390 | read_extent_buffer(leaf, &cache->item, |
8391 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
8392 | sizeof(cache->item)); | |
9078a3e1 | 8393 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 8394 | |
9078a3e1 CM |
8395 | key.objectid = found_key.objectid + found_key.offset; |
8396 | btrfs_release_path(root, path); | |
0b86a832 | 8397 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
8398 | cache->sectorsize = root->sectorsize; |
8399 | ||
817d52f8 JB |
8400 | /* |
8401 | * check for two cases, either we are full, and therefore | |
8402 | * don't need to bother with the caching work since we won't | |
8403 | * find any space, or we are empty, and we can just add all | |
8404 | * the space in and be done with it. This saves us _alot_ of | |
8405 | * time, particularly in the full case. | |
8406 | */ | |
8407 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
1b2da372 | 8408 | exclude_super_stripes(root, cache); |
11833d66 | 8409 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8410 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 8411 | free_excluded_extents(root, cache); |
817d52f8 | 8412 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 YZ |
8413 | exclude_super_stripes(root, cache); |
8414 | cache->last_byte_to_unpin = (u64)-1; | |
817d52f8 JB |
8415 | cache->cached = BTRFS_CACHE_FINISHED; |
8416 | add_new_free_space(cache, root->fs_info, | |
8417 | found_key.objectid, | |
8418 | found_key.objectid + | |
8419 | found_key.offset); | |
11833d66 | 8420 | free_excluded_extents(root, cache); |
817d52f8 | 8421 | } |
96b5179d | 8422 | |
6324fbf3 CM |
8423 | ret = update_space_info(info, cache->flags, found_key.offset, |
8424 | btrfs_block_group_used(&cache->item), | |
8425 | &space_info); | |
8426 | BUG_ON(ret); | |
8427 | cache->space_info = space_info; | |
1b2da372 | 8428 | spin_lock(&cache->space_info->lock); |
f0486c68 | 8429 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8430 | spin_unlock(&cache->space_info->lock); |
8431 | ||
b742bb82 | 8432 | __link_block_group(space_info, cache); |
0f9dd46c JB |
8433 | |
8434 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
8435 | BUG_ON(ret); | |
75ccf47d CM |
8436 | |
8437 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 8438 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
f0486c68 | 8439 | set_block_group_ro(cache); |
9078a3e1 | 8440 | } |
b742bb82 YZ |
8441 | |
8442 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
8443 | if (!(get_alloc_profile(root, space_info->flags) & | |
8444 | (BTRFS_BLOCK_GROUP_RAID10 | | |
8445 | BTRFS_BLOCK_GROUP_RAID1 | | |
8446 | BTRFS_BLOCK_GROUP_DUP))) | |
8447 | continue; | |
8448 | /* | |
8449 | * avoid allocating from un-mirrored block group if there are | |
8450 | * mirrored block groups. | |
8451 | */ | |
8452 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
f0486c68 | 8453 | set_block_group_ro(cache); |
b742bb82 | 8454 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
f0486c68 | 8455 | set_block_group_ro(cache); |
9078a3e1 | 8456 | } |
f0486c68 YZ |
8457 | |
8458 | init_global_block_rsv(info); | |
0b86a832 CM |
8459 | ret = 0; |
8460 | error: | |
9078a3e1 | 8461 | btrfs_free_path(path); |
0b86a832 | 8462 | return ret; |
9078a3e1 | 8463 | } |
6324fbf3 CM |
8464 | |
8465 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
8466 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 8467 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
8468 | u64 size) |
8469 | { | |
8470 | int ret; | |
6324fbf3 CM |
8471 | struct btrfs_root *extent_root; |
8472 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
8473 | |
8474 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 8475 | |
12fcfd22 | 8476 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 8477 | |
8f18cf13 | 8478 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
8479 | if (!cache) |
8480 | return -ENOMEM; | |
8481 | ||
e17cade2 | 8482 | cache->key.objectid = chunk_offset; |
6324fbf3 | 8483 | cache->key.offset = size; |
d2fb3437 | 8484 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 8485 | cache->sectorsize = root->sectorsize; |
0af3d00b | 8486 | cache->fs_info = root->fs_info; |
96303081 JB |
8487 | |
8488 | /* | |
8489 | * we only want to have 32k of ram per block group for keeping track | |
8490 | * of free space, and if we pass 1/2 of that we want to start | |
8491 | * converting things over to using bitmaps | |
8492 | */ | |
8493 | cache->extents_thresh = ((1024 * 32) / 2) / | |
8494 | sizeof(struct btrfs_free_space); | |
d2fb3437 | 8495 | atomic_set(&cache->count, 1); |
c286ac48 | 8496 | spin_lock_init(&cache->lock); |
6226cb0a | 8497 | spin_lock_init(&cache->tree_lock); |
0f9dd46c | 8498 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8499 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 8500 | |
6324fbf3 | 8501 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
8502 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
8503 | cache->flags = type; | |
8504 | btrfs_set_block_group_flags(&cache->item, type); | |
8505 | ||
11833d66 | 8506 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8507 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 8508 | exclude_super_stripes(root, cache); |
96303081 | 8509 | |
817d52f8 JB |
8510 | add_new_free_space(cache, root->fs_info, chunk_offset, |
8511 | chunk_offset + size); | |
8512 | ||
11833d66 YZ |
8513 | free_excluded_extents(root, cache); |
8514 | ||
6324fbf3 CM |
8515 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
8516 | &cache->space_info); | |
8517 | BUG_ON(ret); | |
1b2da372 JB |
8518 | |
8519 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 8520 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8521 | spin_unlock(&cache->space_info->lock); |
8522 | ||
b742bb82 | 8523 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 8524 | |
0f9dd46c JB |
8525 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
8526 | BUG_ON(ret); | |
c286ac48 | 8527 | |
6324fbf3 CM |
8528 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
8529 | sizeof(cache->item)); | |
8530 | BUG_ON(ret); | |
8531 | ||
d18a2c44 | 8532 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 8533 | |
6324fbf3 CM |
8534 | return 0; |
8535 | } | |
1a40e23b ZY |
8536 | |
8537 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
8538 | struct btrfs_root *root, u64 group_start) | |
8539 | { | |
8540 | struct btrfs_path *path; | |
8541 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 8542 | struct btrfs_free_cluster *cluster; |
0af3d00b | 8543 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 8544 | struct btrfs_key key; |
0af3d00b | 8545 | struct inode *inode; |
1a40e23b | 8546 | int ret; |
89a55897 | 8547 | int factor; |
1a40e23b | 8548 | |
1a40e23b ZY |
8549 | root = root->fs_info->extent_root; |
8550 | ||
8551 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
8552 | BUG_ON(!block_group); | |
c146afad | 8553 | BUG_ON(!block_group->ro); |
1a40e23b ZY |
8554 | |
8555 | memcpy(&key, &block_group->key, sizeof(key)); | |
89a55897 JB |
8556 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
8557 | BTRFS_BLOCK_GROUP_RAID1 | | |
8558 | BTRFS_BLOCK_GROUP_RAID10)) | |
8559 | factor = 2; | |
8560 | else | |
8561 | factor = 1; | |
1a40e23b | 8562 | |
44fb5511 CM |
8563 | /* make sure this block group isn't part of an allocation cluster */ |
8564 | cluster = &root->fs_info->data_alloc_cluster; | |
8565 | spin_lock(&cluster->refill_lock); | |
8566 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8567 | spin_unlock(&cluster->refill_lock); | |
8568 | ||
8569 | /* | |
8570 | * make sure this block group isn't part of a metadata | |
8571 | * allocation cluster | |
8572 | */ | |
8573 | cluster = &root->fs_info->meta_alloc_cluster; | |
8574 | spin_lock(&cluster->refill_lock); | |
8575 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8576 | spin_unlock(&cluster->refill_lock); | |
8577 | ||
1a40e23b ZY |
8578 | path = btrfs_alloc_path(); |
8579 | BUG_ON(!path); | |
8580 | ||
0af3d00b JB |
8581 | inode = lookup_free_space_inode(root, block_group, path); |
8582 | if (!IS_ERR(inode)) { | |
8583 | btrfs_orphan_add(trans, inode); | |
8584 | clear_nlink(inode); | |
8585 | /* One for the block groups ref */ | |
8586 | spin_lock(&block_group->lock); | |
8587 | if (block_group->iref) { | |
8588 | block_group->iref = 0; | |
8589 | block_group->inode = NULL; | |
8590 | spin_unlock(&block_group->lock); | |
8591 | iput(inode); | |
8592 | } else { | |
8593 | spin_unlock(&block_group->lock); | |
8594 | } | |
8595 | /* One for our lookup ref */ | |
8596 | iput(inode); | |
8597 | } | |
8598 | ||
8599 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
8600 | key.offset = block_group->key.objectid; | |
8601 | key.type = 0; | |
8602 | ||
8603 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
8604 | if (ret < 0) | |
8605 | goto out; | |
8606 | if (ret > 0) | |
8607 | btrfs_release_path(tree_root, path); | |
8608 | if (ret == 0) { | |
8609 | ret = btrfs_del_item(trans, tree_root, path); | |
8610 | if (ret) | |
8611 | goto out; | |
8612 | btrfs_release_path(tree_root, path); | |
8613 | } | |
8614 | ||
3dfdb934 | 8615 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
8616 | rb_erase(&block_group->cache_node, |
8617 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 8618 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 8619 | |
80eb234a | 8620 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
8621 | /* |
8622 | * we must use list_del_init so people can check to see if they | |
8623 | * are still on the list after taking the semaphore | |
8624 | */ | |
8625 | list_del_init(&block_group->list); | |
80eb234a | 8626 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 8627 | |
817d52f8 | 8628 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8629 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
8630 | |
8631 | btrfs_remove_free_space_cache(block_group); | |
8632 | ||
c146afad YZ |
8633 | spin_lock(&block_group->space_info->lock); |
8634 | block_group->space_info->total_bytes -= block_group->key.offset; | |
8635 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 8636 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 8637 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 8638 | |
0af3d00b JB |
8639 | memcpy(&key, &block_group->key, sizeof(key)); |
8640 | ||
283bb197 | 8641 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 8642 | |
fa9c0d79 CM |
8643 | btrfs_put_block_group(block_group); |
8644 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
8645 | |
8646 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
8647 | if (ret > 0) | |
8648 | ret = -EIO; | |
8649 | if (ret < 0) | |
8650 | goto out; | |
8651 | ||
8652 | ret = btrfs_del_item(trans, root, path); | |
8653 | out: | |
8654 | btrfs_free_path(path); | |
8655 | return ret; | |
8656 | } | |
acce952b | 8657 | |
8658 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) | |
8659 | { | |
8660 | return unpin_extent_range(root, start, end); | |
8661 | } | |
8662 | ||
8663 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
8664 | u64 num_bytes) | |
8665 | { | |
8666 | return btrfs_discard_extent(root, bytenr, num_bytes); | |
8667 | } |