]>
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> |
dff51cd1 | 26 | #include <linux/ratelimit.h> |
4b4e25f2 | 27 | #include "compat.h" |
74493f7a | 28 | #include "hash.h" |
fec577fb CM |
29 | #include "ctree.h" |
30 | #include "disk-io.h" | |
31 | #include "print-tree.h" | |
e089f05c | 32 | #include "transaction.h" |
0b86a832 | 33 | #include "volumes.h" |
925baedd | 34 | #include "locking.h" |
fa9c0d79 | 35 | #include "free-space-cache.h" |
3fed40cc | 36 | #include "math.h" |
fec577fb | 37 | |
709c0486 AJ |
38 | #undef SCRAMBLE_DELAYED_REFS |
39 | ||
9e622d6b MX |
40 | /* |
41 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
42 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
43 | * if we really need one. | |
44 | * | |
0e4f8f88 CM |
45 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
46 | * if we have very few chunks already allocated. This is | |
47 | * used as part of the clustering code to help make sure | |
48 | * we have a good pool of storage to cluster in, without | |
49 | * filling the FS with empty chunks | |
50 | * | |
9e622d6b MX |
51 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
52 | * | |
0e4f8f88 CM |
53 | */ |
54 | enum { | |
55 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
56 | CHUNK_ALLOC_LIMITED = 1, |
57 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
58 | }; |
59 | ||
fb25e914 JB |
60 | /* |
61 | * Control how reservations are dealt with. | |
62 | * | |
63 | * RESERVE_FREE - freeing a reservation. | |
64 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
65 | * ENOSPC accounting | |
66 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
67 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
68 | */ | |
69 | enum { | |
70 | RESERVE_FREE = 0, | |
71 | RESERVE_ALLOC = 1, | |
72 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
73 | }; | |
74 | ||
c53d613e | 75 | static int update_block_group(struct btrfs_root *root, |
f0486c68 | 76 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
77 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
78 | struct btrfs_root *root, | |
79 | u64 bytenr, u64 num_bytes, u64 parent, | |
80 | u64 root_objectid, u64 owner_objectid, | |
81 | u64 owner_offset, int refs_to_drop, | |
82 | struct btrfs_delayed_extent_op *extra_op); | |
83 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
84 | struct extent_buffer *leaf, | |
85 | struct btrfs_extent_item *ei); | |
86 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
87 | struct btrfs_root *root, | |
88 | u64 parent, u64 root_objectid, | |
89 | u64 flags, u64 owner, u64 offset, | |
90 | struct btrfs_key *ins, int ref_mod); | |
91 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
92 | struct btrfs_root *root, | |
93 | u64 parent, u64 root_objectid, | |
94 | u64 flags, struct btrfs_disk_key *key, | |
95 | int level, struct btrfs_key *ins); | |
6a63209f | 96 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 JB |
97 | struct btrfs_root *extent_root, u64 flags, |
98 | int force); | |
11833d66 YZ |
99 | static int find_next_key(struct btrfs_path *path, int level, |
100 | struct btrfs_key *key); | |
9ed74f2d JB |
101 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
102 | int dump_block_groups); | |
fb25e914 JB |
103 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
104 | u64 num_bytes, int reserve); | |
6a63209f | 105 | |
817d52f8 JB |
106 | static noinline int |
107 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
108 | { | |
109 | smp_mb(); | |
110 | return cache->cached == BTRFS_CACHE_FINISHED; | |
111 | } | |
112 | ||
0f9dd46c JB |
113 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
114 | { | |
115 | return (cache->flags & bits) == bits; | |
116 | } | |
117 | ||
62a45b60 | 118 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
119 | { |
120 | atomic_inc(&cache->count); | |
121 | } | |
122 | ||
123 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
124 | { | |
f0486c68 YZ |
125 | if (atomic_dec_and_test(&cache->count)) { |
126 | WARN_ON(cache->pinned > 0); | |
127 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 128 | kfree(cache->free_space_ctl); |
11dfe35a | 129 | kfree(cache); |
f0486c68 | 130 | } |
11dfe35a JB |
131 | } |
132 | ||
0f9dd46c JB |
133 | /* |
134 | * this adds the block group to the fs_info rb tree for the block group | |
135 | * cache | |
136 | */ | |
b2950863 | 137 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
138 | struct btrfs_block_group_cache *block_group) |
139 | { | |
140 | struct rb_node **p; | |
141 | struct rb_node *parent = NULL; | |
142 | struct btrfs_block_group_cache *cache; | |
143 | ||
144 | spin_lock(&info->block_group_cache_lock); | |
145 | p = &info->block_group_cache_tree.rb_node; | |
146 | ||
147 | while (*p) { | |
148 | parent = *p; | |
149 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
150 | cache_node); | |
151 | if (block_group->key.objectid < cache->key.objectid) { | |
152 | p = &(*p)->rb_left; | |
153 | } else if (block_group->key.objectid > cache->key.objectid) { | |
154 | p = &(*p)->rb_right; | |
155 | } else { | |
156 | spin_unlock(&info->block_group_cache_lock); | |
157 | return -EEXIST; | |
158 | } | |
159 | } | |
160 | ||
161 | rb_link_node(&block_group->cache_node, parent, p); | |
162 | rb_insert_color(&block_group->cache_node, | |
163 | &info->block_group_cache_tree); | |
a1897fdd LB |
164 | |
165 | if (info->first_logical_byte > block_group->key.objectid) | |
166 | info->first_logical_byte = block_group->key.objectid; | |
167 | ||
0f9dd46c JB |
168 | spin_unlock(&info->block_group_cache_lock); |
169 | ||
170 | return 0; | |
171 | } | |
172 | ||
173 | /* | |
174 | * This will return the block group at or after bytenr if contains is 0, else | |
175 | * it will return the block group that contains the bytenr | |
176 | */ | |
177 | static struct btrfs_block_group_cache * | |
178 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
179 | int contains) | |
180 | { | |
181 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
182 | struct rb_node *n; | |
183 | u64 end, start; | |
184 | ||
185 | spin_lock(&info->block_group_cache_lock); | |
186 | n = info->block_group_cache_tree.rb_node; | |
187 | ||
188 | while (n) { | |
189 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
190 | cache_node); | |
191 | end = cache->key.objectid + cache->key.offset - 1; | |
192 | start = cache->key.objectid; | |
193 | ||
194 | if (bytenr < start) { | |
195 | if (!contains && (!ret || start < ret->key.objectid)) | |
196 | ret = cache; | |
197 | n = n->rb_left; | |
198 | } else if (bytenr > start) { | |
199 | if (contains && bytenr <= end) { | |
200 | ret = cache; | |
201 | break; | |
202 | } | |
203 | n = n->rb_right; | |
204 | } else { | |
205 | ret = cache; | |
206 | break; | |
207 | } | |
208 | } | |
a1897fdd | 209 | if (ret) { |
11dfe35a | 210 | btrfs_get_block_group(ret); |
a1897fdd LB |
211 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
212 | info->first_logical_byte = ret->key.objectid; | |
213 | } | |
0f9dd46c JB |
214 | spin_unlock(&info->block_group_cache_lock); |
215 | ||
216 | return ret; | |
217 | } | |
218 | ||
11833d66 YZ |
219 | static int add_excluded_extent(struct btrfs_root *root, |
220 | u64 start, u64 num_bytes) | |
817d52f8 | 221 | { |
11833d66 YZ |
222 | u64 end = start + num_bytes - 1; |
223 | set_extent_bits(&root->fs_info->freed_extents[0], | |
224 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
225 | set_extent_bits(&root->fs_info->freed_extents[1], | |
226 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
227 | return 0; | |
228 | } | |
817d52f8 | 229 | |
11833d66 YZ |
230 | static void free_excluded_extents(struct btrfs_root *root, |
231 | struct btrfs_block_group_cache *cache) | |
232 | { | |
233 | u64 start, end; | |
817d52f8 | 234 | |
11833d66 YZ |
235 | start = cache->key.objectid; |
236 | end = start + cache->key.offset - 1; | |
237 | ||
238 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
239 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
240 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
241 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
242 | } |
243 | ||
11833d66 YZ |
244 | static int exclude_super_stripes(struct btrfs_root *root, |
245 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 246 | { |
817d52f8 JB |
247 | u64 bytenr; |
248 | u64 *logical; | |
249 | int stripe_len; | |
250 | int i, nr, ret; | |
251 | ||
06b2331f YZ |
252 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
253 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
254 | cache->bytes_super += stripe_len; | |
255 | ret = add_excluded_extent(root, cache->key.objectid, | |
256 | stripe_len); | |
79787eaa | 257 | BUG_ON(ret); /* -ENOMEM */ |
06b2331f YZ |
258 | } |
259 | ||
817d52f8 JB |
260 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
261 | bytenr = btrfs_sb_offset(i); | |
262 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
263 | cache->key.objectid, bytenr, | |
264 | 0, &logical, &nr, &stripe_len); | |
79787eaa | 265 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 | 266 | |
817d52f8 | 267 | while (nr--) { |
1b2da372 | 268 | cache->bytes_super += stripe_len; |
11833d66 YZ |
269 | ret = add_excluded_extent(root, logical[nr], |
270 | stripe_len); | |
79787eaa | 271 | BUG_ON(ret); /* -ENOMEM */ |
817d52f8 | 272 | } |
11833d66 | 273 | |
817d52f8 JB |
274 | kfree(logical); |
275 | } | |
817d52f8 JB |
276 | return 0; |
277 | } | |
278 | ||
11833d66 YZ |
279 | static struct btrfs_caching_control * |
280 | get_caching_control(struct btrfs_block_group_cache *cache) | |
281 | { | |
282 | struct btrfs_caching_control *ctl; | |
283 | ||
284 | spin_lock(&cache->lock); | |
285 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
286 | spin_unlock(&cache->lock); | |
287 | return NULL; | |
288 | } | |
289 | ||
dde5abee JB |
290 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
291 | if (!cache->caching_ctl) { | |
292 | spin_unlock(&cache->lock); | |
11833d66 YZ |
293 | return NULL; |
294 | } | |
295 | ||
296 | ctl = cache->caching_ctl; | |
297 | atomic_inc(&ctl->count); | |
298 | spin_unlock(&cache->lock); | |
299 | return ctl; | |
300 | } | |
301 | ||
302 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
303 | { | |
304 | if (atomic_dec_and_test(&ctl->count)) | |
305 | kfree(ctl); | |
306 | } | |
307 | ||
0f9dd46c JB |
308 | /* |
309 | * this is only called by cache_block_group, since we could have freed extents | |
310 | * we need to check the pinned_extents for any extents that can't be used yet | |
311 | * since their free space will be released as soon as the transaction commits. | |
312 | */ | |
817d52f8 | 313 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
314 | struct btrfs_fs_info *info, u64 start, u64 end) |
315 | { | |
817d52f8 | 316 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
317 | int ret; |
318 | ||
319 | while (start < end) { | |
11833d66 | 320 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 321 | &extent_start, &extent_end, |
e6138876 JB |
322 | EXTENT_DIRTY | EXTENT_UPTODATE, |
323 | NULL); | |
0f9dd46c JB |
324 | if (ret) |
325 | break; | |
326 | ||
06b2331f | 327 | if (extent_start <= start) { |
0f9dd46c JB |
328 | start = extent_end + 1; |
329 | } else if (extent_start > start && extent_start < end) { | |
330 | size = extent_start - start; | |
817d52f8 | 331 | total_added += size; |
ea6a478e JB |
332 | ret = btrfs_add_free_space(block_group, start, |
333 | size); | |
79787eaa | 334 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
335 | start = extent_end + 1; |
336 | } else { | |
337 | break; | |
338 | } | |
339 | } | |
340 | ||
341 | if (start < end) { | |
342 | size = end - start; | |
817d52f8 | 343 | total_added += size; |
ea6a478e | 344 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 345 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
346 | } |
347 | ||
817d52f8 | 348 | return total_added; |
0f9dd46c JB |
349 | } |
350 | ||
bab39bf9 | 351 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 352 | { |
bab39bf9 JB |
353 | struct btrfs_block_group_cache *block_group; |
354 | struct btrfs_fs_info *fs_info; | |
355 | struct btrfs_caching_control *caching_ctl; | |
356 | struct btrfs_root *extent_root; | |
e37c9e69 | 357 | struct btrfs_path *path; |
5f39d397 | 358 | struct extent_buffer *leaf; |
11833d66 | 359 | struct btrfs_key key; |
817d52f8 | 360 | u64 total_found = 0; |
11833d66 YZ |
361 | u64 last = 0; |
362 | u32 nritems; | |
363 | int ret = 0; | |
f510cfec | 364 | |
bab39bf9 JB |
365 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
366 | block_group = caching_ctl->block_group; | |
367 | fs_info = block_group->fs_info; | |
368 | extent_root = fs_info->extent_root; | |
369 | ||
e37c9e69 CM |
370 | path = btrfs_alloc_path(); |
371 | if (!path) | |
bab39bf9 | 372 | goto out; |
7d7d6068 | 373 | |
817d52f8 | 374 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 375 | |
5cd57b2c | 376 | /* |
817d52f8 JB |
377 | * We don't want to deadlock with somebody trying to allocate a new |
378 | * extent for the extent root while also trying to search the extent | |
379 | * root to add free space. So we skip locking and search the commit | |
380 | * root, since its read-only | |
5cd57b2c CM |
381 | */ |
382 | path->skip_locking = 1; | |
817d52f8 | 383 | path->search_commit_root = 1; |
026fd317 | 384 | path->reada = 1; |
817d52f8 | 385 | |
e4404d6e | 386 | key.objectid = last; |
e37c9e69 | 387 | key.offset = 0; |
11833d66 | 388 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 389 | again: |
11833d66 | 390 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
391 | /* need to make sure the commit_root doesn't disappear */ |
392 | down_read(&fs_info->extent_commit_sem); | |
393 | ||
11833d66 | 394 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 395 | if (ret < 0) |
ef8bbdfe | 396 | goto err; |
a512bbf8 | 397 | |
11833d66 YZ |
398 | leaf = path->nodes[0]; |
399 | nritems = btrfs_header_nritems(leaf); | |
400 | ||
d397712b | 401 | while (1) { |
7841cb28 | 402 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 403 | last = (u64)-1; |
817d52f8 | 404 | break; |
f25784b3 | 405 | } |
817d52f8 | 406 | |
11833d66 YZ |
407 | if (path->slots[0] < nritems) { |
408 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
409 | } else { | |
410 | ret = find_next_key(path, 0, &key); | |
411 | if (ret) | |
e37c9e69 | 412 | break; |
817d52f8 | 413 | |
589d8ade JB |
414 | if (need_resched() || |
415 | btrfs_next_leaf(extent_root, path)) { | |
416 | caching_ctl->progress = last; | |
ff5714cc | 417 | btrfs_release_path(path); |
589d8ade JB |
418 | up_read(&fs_info->extent_commit_sem); |
419 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 420 | cond_resched(); |
589d8ade JB |
421 | goto again; |
422 | } | |
423 | leaf = path->nodes[0]; | |
424 | nritems = btrfs_header_nritems(leaf); | |
425 | continue; | |
11833d66 | 426 | } |
817d52f8 | 427 | |
11833d66 YZ |
428 | if (key.objectid < block_group->key.objectid) { |
429 | path->slots[0]++; | |
817d52f8 | 430 | continue; |
e37c9e69 | 431 | } |
0f9dd46c | 432 | |
e37c9e69 | 433 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 434 | block_group->key.offset) |
e37c9e69 | 435 | break; |
7d7d6068 | 436 | |
11833d66 | 437 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
438 | total_found += add_new_free_space(block_group, |
439 | fs_info, last, | |
440 | key.objectid); | |
7d7d6068 | 441 | last = key.objectid + key.offset; |
817d52f8 | 442 | |
11833d66 YZ |
443 | if (total_found > (1024 * 1024 * 2)) { |
444 | total_found = 0; | |
445 | wake_up(&caching_ctl->wait); | |
446 | } | |
817d52f8 | 447 | } |
e37c9e69 CM |
448 | path->slots[0]++; |
449 | } | |
817d52f8 | 450 | ret = 0; |
e37c9e69 | 451 | |
817d52f8 JB |
452 | total_found += add_new_free_space(block_group, fs_info, last, |
453 | block_group->key.objectid + | |
454 | block_group->key.offset); | |
11833d66 | 455 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
456 | |
457 | spin_lock(&block_group->lock); | |
11833d66 | 458 | block_group->caching_ctl = NULL; |
817d52f8 JB |
459 | block_group->cached = BTRFS_CACHE_FINISHED; |
460 | spin_unlock(&block_group->lock); | |
0f9dd46c | 461 | |
54aa1f4d | 462 | err: |
e37c9e69 | 463 | btrfs_free_path(path); |
276e680d | 464 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 465 | |
11833d66 YZ |
466 | free_excluded_extents(extent_root, block_group); |
467 | ||
468 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 469 | out: |
11833d66 YZ |
470 | wake_up(&caching_ctl->wait); |
471 | ||
472 | put_caching_control(caching_ctl); | |
11dfe35a | 473 | btrfs_put_block_group(block_group); |
817d52f8 JB |
474 | } |
475 | ||
9d66e233 | 476 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 477 | int load_cache_only) |
817d52f8 | 478 | { |
291c7d2f | 479 | DEFINE_WAIT(wait); |
11833d66 YZ |
480 | struct btrfs_fs_info *fs_info = cache->fs_info; |
481 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
482 | int ret = 0; |
483 | ||
291c7d2f | 484 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
485 | if (!caching_ctl) |
486 | return -ENOMEM; | |
291c7d2f JB |
487 | |
488 | INIT_LIST_HEAD(&caching_ctl->list); | |
489 | mutex_init(&caching_ctl->mutex); | |
490 | init_waitqueue_head(&caching_ctl->wait); | |
491 | caching_ctl->block_group = cache; | |
492 | caching_ctl->progress = cache->key.objectid; | |
493 | atomic_set(&caching_ctl->count, 1); | |
494 | caching_ctl->work.func = caching_thread; | |
495 | ||
496 | spin_lock(&cache->lock); | |
497 | /* | |
498 | * This should be a rare occasion, but this could happen I think in the | |
499 | * case where one thread starts to load the space cache info, and then | |
500 | * some other thread starts a transaction commit which tries to do an | |
501 | * allocation while the other thread is still loading the space cache | |
502 | * info. The previous loop should have kept us from choosing this block | |
503 | * group, but if we've moved to the state where we will wait on caching | |
504 | * block groups we need to first check if we're doing a fast load here, | |
505 | * so we can wait for it to finish, otherwise we could end up allocating | |
506 | * from a block group who's cache gets evicted for one reason or | |
507 | * another. | |
508 | */ | |
509 | while (cache->cached == BTRFS_CACHE_FAST) { | |
510 | struct btrfs_caching_control *ctl; | |
511 | ||
512 | ctl = cache->caching_ctl; | |
513 | atomic_inc(&ctl->count); | |
514 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
515 | spin_unlock(&cache->lock); | |
516 | ||
517 | schedule(); | |
518 | ||
519 | finish_wait(&ctl->wait, &wait); | |
520 | put_caching_control(ctl); | |
521 | spin_lock(&cache->lock); | |
522 | } | |
523 | ||
524 | if (cache->cached != BTRFS_CACHE_NO) { | |
525 | spin_unlock(&cache->lock); | |
526 | kfree(caching_ctl); | |
11833d66 | 527 | return 0; |
291c7d2f JB |
528 | } |
529 | WARN_ON(cache->caching_ctl); | |
530 | cache->caching_ctl = caching_ctl; | |
531 | cache->cached = BTRFS_CACHE_FAST; | |
532 | spin_unlock(&cache->lock); | |
11833d66 | 533 | |
d53ba474 | 534 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
9d66e233 JB |
535 | ret = load_free_space_cache(fs_info, cache); |
536 | ||
537 | spin_lock(&cache->lock); | |
538 | if (ret == 1) { | |
291c7d2f | 539 | cache->caching_ctl = NULL; |
9d66e233 JB |
540 | cache->cached = BTRFS_CACHE_FINISHED; |
541 | cache->last_byte_to_unpin = (u64)-1; | |
542 | } else { | |
291c7d2f JB |
543 | if (load_cache_only) { |
544 | cache->caching_ctl = NULL; | |
545 | cache->cached = BTRFS_CACHE_NO; | |
546 | } else { | |
547 | cache->cached = BTRFS_CACHE_STARTED; | |
548 | } | |
9d66e233 JB |
549 | } |
550 | spin_unlock(&cache->lock); | |
291c7d2f | 551 | wake_up(&caching_ctl->wait); |
3c14874a | 552 | if (ret == 1) { |
291c7d2f | 553 | put_caching_control(caching_ctl); |
3c14874a | 554 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 555 | return 0; |
3c14874a | 556 | } |
291c7d2f JB |
557 | } else { |
558 | /* | |
559 | * We are not going to do the fast caching, set cached to the | |
560 | * appropriate value and wakeup any waiters. | |
561 | */ | |
562 | spin_lock(&cache->lock); | |
563 | if (load_cache_only) { | |
564 | cache->caching_ctl = NULL; | |
565 | cache->cached = BTRFS_CACHE_NO; | |
566 | } else { | |
567 | cache->cached = BTRFS_CACHE_STARTED; | |
568 | } | |
569 | spin_unlock(&cache->lock); | |
570 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
571 | } |
572 | ||
291c7d2f JB |
573 | if (load_cache_only) { |
574 | put_caching_control(caching_ctl); | |
11833d66 | 575 | return 0; |
817d52f8 | 576 | } |
817d52f8 | 577 | |
11833d66 | 578 | down_write(&fs_info->extent_commit_sem); |
291c7d2f | 579 | atomic_inc(&caching_ctl->count); |
11833d66 YZ |
580 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
581 | up_write(&fs_info->extent_commit_sem); | |
582 | ||
11dfe35a | 583 | btrfs_get_block_group(cache); |
11833d66 | 584 | |
bab39bf9 | 585 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 586 | |
ef8bbdfe | 587 | return ret; |
e37c9e69 CM |
588 | } |
589 | ||
0f9dd46c JB |
590 | /* |
591 | * return the block group that starts at or after bytenr | |
592 | */ | |
d397712b CM |
593 | static struct btrfs_block_group_cache * |
594 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 595 | { |
0f9dd46c | 596 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 597 | |
0f9dd46c | 598 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 599 | |
0f9dd46c | 600 | return cache; |
0ef3e66b CM |
601 | } |
602 | ||
0f9dd46c | 603 | /* |
9f55684c | 604 | * return the block group that contains the given bytenr |
0f9dd46c | 605 | */ |
d397712b CM |
606 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
607 | struct btrfs_fs_info *info, | |
608 | u64 bytenr) | |
be744175 | 609 | { |
0f9dd46c | 610 | struct btrfs_block_group_cache *cache; |
be744175 | 611 | |
0f9dd46c | 612 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 613 | |
0f9dd46c | 614 | return cache; |
be744175 | 615 | } |
0b86a832 | 616 | |
0f9dd46c JB |
617 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
618 | u64 flags) | |
6324fbf3 | 619 | { |
0f9dd46c | 620 | struct list_head *head = &info->space_info; |
0f9dd46c | 621 | struct btrfs_space_info *found; |
4184ea7f | 622 | |
52ba6929 | 623 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 624 | |
4184ea7f CM |
625 | rcu_read_lock(); |
626 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 627 | if (found->flags & flags) { |
4184ea7f | 628 | rcu_read_unlock(); |
0f9dd46c | 629 | return found; |
4184ea7f | 630 | } |
0f9dd46c | 631 | } |
4184ea7f | 632 | rcu_read_unlock(); |
0f9dd46c | 633 | return NULL; |
6324fbf3 CM |
634 | } |
635 | ||
4184ea7f CM |
636 | /* |
637 | * after adding space to the filesystem, we need to clear the full flags | |
638 | * on all the space infos. | |
639 | */ | |
640 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
641 | { | |
642 | struct list_head *head = &info->space_info; | |
643 | struct btrfs_space_info *found; | |
644 | ||
645 | rcu_read_lock(); | |
646 | list_for_each_entry_rcu(found, head, list) | |
647 | found->full = 0; | |
648 | rcu_read_unlock(); | |
649 | } | |
650 | ||
d2fb3437 YZ |
651 | u64 btrfs_find_block_group(struct btrfs_root *root, |
652 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 653 | { |
96b5179d | 654 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 655 | u64 used; |
d2fb3437 YZ |
656 | u64 last = max(search_hint, search_start); |
657 | u64 group_start = 0; | |
31f3c99b | 658 | int full_search = 0; |
d2fb3437 | 659 | int factor = 9; |
0ef3e66b | 660 | int wrapped = 0; |
31f3c99b | 661 | again: |
e8569813 ZY |
662 | while (1) { |
663 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
664 | if (!cache) |
665 | break; | |
96b5179d | 666 | |
c286ac48 | 667 | spin_lock(&cache->lock); |
96b5179d CM |
668 | last = cache->key.objectid + cache->key.offset; |
669 | used = btrfs_block_group_used(&cache->item); | |
670 | ||
d2fb3437 YZ |
671 | if ((full_search || !cache->ro) && |
672 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 673 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
674 | div_factor(cache->key.offset, factor)) { |
675 | group_start = cache->key.objectid; | |
c286ac48 | 676 | spin_unlock(&cache->lock); |
fa9c0d79 | 677 | btrfs_put_block_group(cache); |
8790d502 CM |
678 | goto found; |
679 | } | |
6324fbf3 | 680 | } |
c286ac48 | 681 | spin_unlock(&cache->lock); |
fa9c0d79 | 682 | btrfs_put_block_group(cache); |
de428b63 | 683 | cond_resched(); |
cd1bc465 | 684 | } |
0ef3e66b CM |
685 | if (!wrapped) { |
686 | last = search_start; | |
687 | wrapped = 1; | |
688 | goto again; | |
689 | } | |
690 | if (!full_search && factor < 10) { | |
be744175 | 691 | last = search_start; |
31f3c99b | 692 | full_search = 1; |
0ef3e66b | 693 | factor = 10; |
31f3c99b CM |
694 | goto again; |
695 | } | |
be744175 | 696 | found: |
d2fb3437 | 697 | return group_start; |
925baedd | 698 | } |
0f9dd46c | 699 | |
e02119d5 | 700 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 701 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
702 | { |
703 | int ret; | |
704 | struct btrfs_key key; | |
31840ae1 | 705 | struct btrfs_path *path; |
e02119d5 | 706 | |
31840ae1 | 707 | path = btrfs_alloc_path(); |
d8926bb3 MF |
708 | if (!path) |
709 | return -ENOMEM; | |
710 | ||
e02119d5 CM |
711 | key.objectid = start; |
712 | key.offset = len; | |
713 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
714 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
715 | 0, 0); | |
31840ae1 | 716 | btrfs_free_path(path); |
7bb86316 CM |
717 | return ret; |
718 | } | |
719 | ||
a22285a6 YZ |
720 | /* |
721 | * helper function to lookup reference count and flags of extent. | |
722 | * | |
723 | * the head node for delayed ref is used to store the sum of all the | |
724 | * reference count modifications queued up in the rbtree. the head | |
725 | * node may also store the extent flags to set. This way you can check | |
726 | * to see what the reference count and extent flags would be if all of | |
727 | * the delayed refs are not processed. | |
728 | */ | |
729 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
730 | struct btrfs_root *root, u64 bytenr, | |
731 | u64 num_bytes, u64 *refs, u64 *flags) | |
732 | { | |
733 | struct btrfs_delayed_ref_head *head; | |
734 | struct btrfs_delayed_ref_root *delayed_refs; | |
735 | struct btrfs_path *path; | |
736 | struct btrfs_extent_item *ei; | |
737 | struct extent_buffer *leaf; | |
738 | struct btrfs_key key; | |
739 | u32 item_size; | |
740 | u64 num_refs; | |
741 | u64 extent_flags; | |
742 | int ret; | |
743 | ||
744 | path = btrfs_alloc_path(); | |
745 | if (!path) | |
746 | return -ENOMEM; | |
747 | ||
748 | key.objectid = bytenr; | |
749 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
750 | key.offset = num_bytes; | |
751 | if (!trans) { | |
752 | path->skip_locking = 1; | |
753 | path->search_commit_root = 1; | |
754 | } | |
755 | again: | |
756 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
757 | &key, path, 0, 0); | |
758 | if (ret < 0) | |
759 | goto out_free; | |
760 | ||
761 | if (ret == 0) { | |
762 | leaf = path->nodes[0]; | |
763 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
764 | if (item_size >= sizeof(*ei)) { | |
765 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
766 | struct btrfs_extent_item); | |
767 | num_refs = btrfs_extent_refs(leaf, ei); | |
768 | extent_flags = btrfs_extent_flags(leaf, ei); | |
769 | } else { | |
770 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
771 | struct btrfs_extent_item_v0 *ei0; | |
772 | BUG_ON(item_size != sizeof(*ei0)); | |
773 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
774 | struct btrfs_extent_item_v0); | |
775 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
776 | /* FIXME: this isn't correct for data */ | |
777 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
778 | #else | |
779 | BUG(); | |
780 | #endif | |
781 | } | |
782 | BUG_ON(num_refs == 0); | |
783 | } else { | |
784 | num_refs = 0; | |
785 | extent_flags = 0; | |
786 | ret = 0; | |
787 | } | |
788 | ||
789 | if (!trans) | |
790 | goto out; | |
791 | ||
792 | delayed_refs = &trans->transaction->delayed_refs; | |
793 | spin_lock(&delayed_refs->lock); | |
794 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
795 | if (head) { | |
796 | if (!mutex_trylock(&head->mutex)) { | |
797 | atomic_inc(&head->node.refs); | |
798 | spin_unlock(&delayed_refs->lock); | |
799 | ||
b3b4aa74 | 800 | btrfs_release_path(path); |
a22285a6 | 801 | |
8cc33e5c DS |
802 | /* |
803 | * Mutex was contended, block until it's released and try | |
804 | * again | |
805 | */ | |
a22285a6 YZ |
806 | mutex_lock(&head->mutex); |
807 | mutex_unlock(&head->mutex); | |
808 | btrfs_put_delayed_ref(&head->node); | |
809 | goto again; | |
810 | } | |
811 | if (head->extent_op && head->extent_op->update_flags) | |
812 | extent_flags |= head->extent_op->flags_to_set; | |
813 | else | |
814 | BUG_ON(num_refs == 0); | |
815 | ||
816 | num_refs += head->node.ref_mod; | |
817 | mutex_unlock(&head->mutex); | |
818 | } | |
819 | spin_unlock(&delayed_refs->lock); | |
820 | out: | |
821 | WARN_ON(num_refs == 0); | |
822 | if (refs) | |
823 | *refs = num_refs; | |
824 | if (flags) | |
825 | *flags = extent_flags; | |
826 | out_free: | |
827 | btrfs_free_path(path); | |
828 | return ret; | |
829 | } | |
830 | ||
d8d5f3e1 CM |
831 | /* |
832 | * Back reference rules. Back refs have three main goals: | |
833 | * | |
834 | * 1) differentiate between all holders of references to an extent so that | |
835 | * when a reference is dropped we can make sure it was a valid reference | |
836 | * before freeing the extent. | |
837 | * | |
838 | * 2) Provide enough information to quickly find the holders of an extent | |
839 | * if we notice a given block is corrupted or bad. | |
840 | * | |
841 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
842 | * maintenance. This is actually the same as #2, but with a slightly | |
843 | * different use case. | |
844 | * | |
5d4f98a2 YZ |
845 | * There are two kinds of back refs. The implicit back refs is optimized |
846 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
847 | * back refs of this kind provide information about the block's owner tree | |
848 | * and the pointer's key. These information allow us to find the block by | |
849 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
850 | * referenced by their owner trees. The location of tree block is recorded | |
851 | * in the back refs. Actually the full back refs is generic, and can be | |
852 | * used in all cases the implicit back refs is used. The major shortcoming | |
853 | * of the full back refs is its overhead. Every time a tree block gets | |
854 | * COWed, we have to update back refs entry for all pointers in it. | |
855 | * | |
856 | * For a newly allocated tree block, we use implicit back refs for | |
857 | * pointers in it. This means most tree related operations only involve | |
858 | * implicit back refs. For a tree block created in old transaction, the | |
859 | * only way to drop a reference to it is COW it. So we can detect the | |
860 | * event that tree block loses its owner tree's reference and do the | |
861 | * back refs conversion. | |
862 | * | |
863 | * When a tree block is COW'd through a tree, there are four cases: | |
864 | * | |
865 | * The reference count of the block is one and the tree is the block's | |
866 | * owner tree. Nothing to do in this case. | |
867 | * | |
868 | * The reference count of the block is one and the tree is not the | |
869 | * block's owner tree. In this case, full back refs is used for pointers | |
870 | * in the block. Remove these full back refs, add implicit back refs for | |
871 | * every pointers in the new block. | |
872 | * | |
873 | * The reference count of the block is greater than one and the tree is | |
874 | * the block's owner tree. In this case, implicit back refs is used for | |
875 | * pointers in the block. Add full back refs for every pointers in the | |
876 | * block, increase lower level extents' reference counts. The original | |
877 | * implicit back refs are entailed to the new block. | |
878 | * | |
879 | * The reference count of the block is greater than one and the tree is | |
880 | * not the block's owner tree. Add implicit back refs for every pointer in | |
881 | * the new block, increase lower level extents' reference count. | |
882 | * | |
883 | * Back Reference Key composing: | |
884 | * | |
885 | * The key objectid corresponds to the first byte in the extent, | |
886 | * The key type is used to differentiate between types of back refs. | |
887 | * There are different meanings of the key offset for different types | |
888 | * of back refs. | |
889 | * | |
d8d5f3e1 CM |
890 | * File extents can be referenced by: |
891 | * | |
892 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 893 | * - different files inside a single subvolume |
d8d5f3e1 CM |
894 | * - different offsets inside a file (bookend extents in file.c) |
895 | * | |
5d4f98a2 | 896 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
897 | * |
898 | * - Objectid of the subvolume root | |
d8d5f3e1 | 899 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
900 | * - original offset in the file |
901 | * - how many bookend extents | |
d8d5f3e1 | 902 | * |
5d4f98a2 YZ |
903 | * The key offset for the implicit back refs is hash of the first |
904 | * three fields. | |
d8d5f3e1 | 905 | * |
5d4f98a2 | 906 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 907 | * |
5d4f98a2 | 908 | * - number of pointers in the tree leaf |
d8d5f3e1 | 909 | * |
5d4f98a2 YZ |
910 | * The key offset for the implicit back refs is the first byte of |
911 | * the tree leaf | |
d8d5f3e1 | 912 | * |
5d4f98a2 YZ |
913 | * When a file extent is allocated, The implicit back refs is used. |
914 | * the fields are filled in: | |
d8d5f3e1 | 915 | * |
5d4f98a2 | 916 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 917 | * |
5d4f98a2 YZ |
918 | * When a file extent is removed file truncation, we find the |
919 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 920 | * |
5d4f98a2 | 921 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 922 | * |
5d4f98a2 | 923 | * Btree extents can be referenced by: |
d8d5f3e1 | 924 | * |
5d4f98a2 | 925 | * - Different subvolumes |
d8d5f3e1 | 926 | * |
5d4f98a2 YZ |
927 | * Both the implicit back refs and the full back refs for tree blocks |
928 | * only consist of key. The key offset for the implicit back refs is | |
929 | * objectid of block's owner tree. The key offset for the full back refs | |
930 | * is the first byte of parent block. | |
d8d5f3e1 | 931 | * |
5d4f98a2 YZ |
932 | * When implicit back refs is used, information about the lowest key and |
933 | * level of the tree block are required. These information are stored in | |
934 | * tree block info structure. | |
d8d5f3e1 | 935 | */ |
31840ae1 | 936 | |
5d4f98a2 YZ |
937 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
938 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
939 | struct btrfs_root *root, | |
940 | struct btrfs_path *path, | |
941 | u64 owner, u32 extra_size) | |
7bb86316 | 942 | { |
5d4f98a2 YZ |
943 | struct btrfs_extent_item *item; |
944 | struct btrfs_extent_item_v0 *ei0; | |
945 | struct btrfs_extent_ref_v0 *ref0; | |
946 | struct btrfs_tree_block_info *bi; | |
947 | struct extent_buffer *leaf; | |
7bb86316 | 948 | struct btrfs_key key; |
5d4f98a2 YZ |
949 | struct btrfs_key found_key; |
950 | u32 new_size = sizeof(*item); | |
951 | u64 refs; | |
952 | int ret; | |
953 | ||
954 | leaf = path->nodes[0]; | |
955 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
956 | ||
957 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
958 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
959 | struct btrfs_extent_item_v0); | |
960 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
961 | ||
962 | if (owner == (u64)-1) { | |
963 | while (1) { | |
964 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
965 | ret = btrfs_next_leaf(root, path); | |
966 | if (ret < 0) | |
967 | return ret; | |
79787eaa | 968 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
969 | leaf = path->nodes[0]; |
970 | } | |
971 | btrfs_item_key_to_cpu(leaf, &found_key, | |
972 | path->slots[0]); | |
973 | BUG_ON(key.objectid != found_key.objectid); | |
974 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
975 | path->slots[0]++; | |
976 | continue; | |
977 | } | |
978 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
979 | struct btrfs_extent_ref_v0); | |
980 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
981 | break; | |
982 | } | |
983 | } | |
b3b4aa74 | 984 | btrfs_release_path(path); |
5d4f98a2 YZ |
985 | |
986 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
987 | new_size += sizeof(*bi); | |
988 | ||
989 | new_size -= sizeof(*ei0); | |
990 | ret = btrfs_search_slot(trans, root, &key, path, | |
991 | new_size + extra_size, 1); | |
992 | if (ret < 0) | |
993 | return ret; | |
79787eaa | 994 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 995 | |
143bede5 | 996 | btrfs_extend_item(trans, root, path, new_size); |
5d4f98a2 YZ |
997 | |
998 | leaf = path->nodes[0]; | |
999 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1000 | btrfs_set_extent_refs(leaf, item, refs); | |
1001 | /* FIXME: get real generation */ | |
1002 | btrfs_set_extent_generation(leaf, item, 0); | |
1003 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1004 | btrfs_set_extent_flags(leaf, item, | |
1005 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1006 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1007 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1008 | /* FIXME: get first key of the block */ | |
1009 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1010 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1011 | } else { | |
1012 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1013 | } | |
1014 | btrfs_mark_buffer_dirty(leaf); | |
1015 | return 0; | |
1016 | } | |
1017 | #endif | |
1018 | ||
1019 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1020 | { | |
1021 | u32 high_crc = ~(u32)0; | |
1022 | u32 low_crc = ~(u32)0; | |
1023 | __le64 lenum; | |
1024 | ||
1025 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 1026 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1027 | lenum = cpu_to_le64(owner); |
163e783e | 1028 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1029 | lenum = cpu_to_le64(offset); |
163e783e | 1030 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1031 | |
1032 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1033 | } | |
1034 | ||
1035 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1036 | struct btrfs_extent_data_ref *ref) | |
1037 | { | |
1038 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1039 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1040 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1041 | } | |
1042 | ||
1043 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1044 | struct btrfs_extent_data_ref *ref, | |
1045 | u64 root_objectid, u64 owner, u64 offset) | |
1046 | { | |
1047 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1048 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1049 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1050 | return 0; | |
1051 | return 1; | |
1052 | } | |
1053 | ||
1054 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1055 | struct btrfs_root *root, | |
1056 | struct btrfs_path *path, | |
1057 | u64 bytenr, u64 parent, | |
1058 | u64 root_objectid, | |
1059 | u64 owner, u64 offset) | |
1060 | { | |
1061 | struct btrfs_key key; | |
1062 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1063 | struct extent_buffer *leaf; |
5d4f98a2 | 1064 | u32 nritems; |
74493f7a | 1065 | int ret; |
5d4f98a2 YZ |
1066 | int recow; |
1067 | int err = -ENOENT; | |
74493f7a | 1068 | |
31840ae1 | 1069 | key.objectid = bytenr; |
5d4f98a2 YZ |
1070 | if (parent) { |
1071 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1072 | key.offset = parent; | |
1073 | } else { | |
1074 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1075 | key.offset = hash_extent_data_ref(root_objectid, | |
1076 | owner, offset); | |
1077 | } | |
1078 | again: | |
1079 | recow = 0; | |
1080 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1081 | if (ret < 0) { | |
1082 | err = ret; | |
1083 | goto fail; | |
1084 | } | |
31840ae1 | 1085 | |
5d4f98a2 YZ |
1086 | if (parent) { |
1087 | if (!ret) | |
1088 | return 0; | |
1089 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1090 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1091 | btrfs_release_path(path); |
5d4f98a2 YZ |
1092 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1093 | if (ret < 0) { | |
1094 | err = ret; | |
1095 | goto fail; | |
1096 | } | |
1097 | if (!ret) | |
1098 | return 0; | |
1099 | #endif | |
1100 | goto fail; | |
31840ae1 ZY |
1101 | } |
1102 | ||
1103 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1104 | nritems = btrfs_header_nritems(leaf); |
1105 | while (1) { | |
1106 | if (path->slots[0] >= nritems) { | |
1107 | ret = btrfs_next_leaf(root, path); | |
1108 | if (ret < 0) | |
1109 | err = ret; | |
1110 | if (ret) | |
1111 | goto fail; | |
1112 | ||
1113 | leaf = path->nodes[0]; | |
1114 | nritems = btrfs_header_nritems(leaf); | |
1115 | recow = 1; | |
1116 | } | |
1117 | ||
1118 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1119 | if (key.objectid != bytenr || | |
1120 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1121 | goto fail; | |
1122 | ||
1123 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1124 | struct btrfs_extent_data_ref); | |
1125 | ||
1126 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1127 | owner, offset)) { | |
1128 | if (recow) { | |
b3b4aa74 | 1129 | btrfs_release_path(path); |
5d4f98a2 YZ |
1130 | goto again; |
1131 | } | |
1132 | err = 0; | |
1133 | break; | |
1134 | } | |
1135 | path->slots[0]++; | |
31840ae1 | 1136 | } |
5d4f98a2 YZ |
1137 | fail: |
1138 | return err; | |
31840ae1 ZY |
1139 | } |
1140 | ||
5d4f98a2 YZ |
1141 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1142 | struct btrfs_root *root, | |
1143 | struct btrfs_path *path, | |
1144 | u64 bytenr, u64 parent, | |
1145 | u64 root_objectid, u64 owner, | |
1146 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1147 | { |
1148 | struct btrfs_key key; | |
1149 | struct extent_buffer *leaf; | |
5d4f98a2 | 1150 | u32 size; |
31840ae1 ZY |
1151 | u32 num_refs; |
1152 | int ret; | |
74493f7a | 1153 | |
74493f7a | 1154 | key.objectid = bytenr; |
5d4f98a2 YZ |
1155 | if (parent) { |
1156 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1157 | key.offset = parent; | |
1158 | size = sizeof(struct btrfs_shared_data_ref); | |
1159 | } else { | |
1160 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1161 | key.offset = hash_extent_data_ref(root_objectid, | |
1162 | owner, offset); | |
1163 | size = sizeof(struct btrfs_extent_data_ref); | |
1164 | } | |
74493f7a | 1165 | |
5d4f98a2 YZ |
1166 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1167 | if (ret && ret != -EEXIST) | |
1168 | goto fail; | |
1169 | ||
1170 | leaf = path->nodes[0]; | |
1171 | if (parent) { | |
1172 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1173 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1174 | struct btrfs_shared_data_ref); |
1175 | if (ret == 0) { | |
1176 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1177 | } else { | |
1178 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1179 | num_refs += refs_to_add; | |
1180 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1181 | } |
5d4f98a2 YZ |
1182 | } else { |
1183 | struct btrfs_extent_data_ref *ref; | |
1184 | while (ret == -EEXIST) { | |
1185 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1186 | struct btrfs_extent_data_ref); | |
1187 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1188 | owner, offset)) | |
1189 | break; | |
b3b4aa74 | 1190 | btrfs_release_path(path); |
5d4f98a2 YZ |
1191 | key.offset++; |
1192 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1193 | size); | |
1194 | if (ret && ret != -EEXIST) | |
1195 | goto fail; | |
31840ae1 | 1196 | |
5d4f98a2 YZ |
1197 | leaf = path->nodes[0]; |
1198 | } | |
1199 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1200 | struct btrfs_extent_data_ref); | |
1201 | if (ret == 0) { | |
1202 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1203 | root_objectid); | |
1204 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1205 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1206 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1207 | } else { | |
1208 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1209 | num_refs += refs_to_add; | |
1210 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1211 | } |
31840ae1 | 1212 | } |
5d4f98a2 YZ |
1213 | btrfs_mark_buffer_dirty(leaf); |
1214 | ret = 0; | |
1215 | fail: | |
b3b4aa74 | 1216 | btrfs_release_path(path); |
7bb86316 | 1217 | return ret; |
74493f7a CM |
1218 | } |
1219 | ||
5d4f98a2 YZ |
1220 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1221 | struct btrfs_root *root, | |
1222 | struct btrfs_path *path, | |
1223 | int refs_to_drop) | |
31840ae1 | 1224 | { |
5d4f98a2 YZ |
1225 | struct btrfs_key key; |
1226 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1227 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1228 | struct extent_buffer *leaf; |
5d4f98a2 | 1229 | u32 num_refs = 0; |
31840ae1 ZY |
1230 | int ret = 0; |
1231 | ||
1232 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1233 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1234 | ||
1235 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1236 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1237 | struct btrfs_extent_data_ref); | |
1238 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1239 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1240 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1241 | struct btrfs_shared_data_ref); | |
1242 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1243 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1244 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1245 | struct btrfs_extent_ref_v0 *ref0; | |
1246 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1247 | struct btrfs_extent_ref_v0); | |
1248 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1249 | #endif | |
1250 | } else { | |
1251 | BUG(); | |
1252 | } | |
1253 | ||
56bec294 CM |
1254 | BUG_ON(num_refs < refs_to_drop); |
1255 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1256 | |
31840ae1 ZY |
1257 | if (num_refs == 0) { |
1258 | ret = btrfs_del_item(trans, root, path); | |
1259 | } else { | |
5d4f98a2 YZ |
1260 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1261 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1262 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1263 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1264 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1265 | else { | |
1266 | struct btrfs_extent_ref_v0 *ref0; | |
1267 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1268 | struct btrfs_extent_ref_v0); | |
1269 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1270 | } | |
1271 | #endif | |
31840ae1 ZY |
1272 | btrfs_mark_buffer_dirty(leaf); |
1273 | } | |
31840ae1 ZY |
1274 | return ret; |
1275 | } | |
1276 | ||
5d4f98a2 YZ |
1277 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1278 | struct btrfs_path *path, | |
1279 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1280 | { |
5d4f98a2 YZ |
1281 | struct btrfs_key key; |
1282 | struct extent_buffer *leaf; | |
1283 | struct btrfs_extent_data_ref *ref1; | |
1284 | struct btrfs_shared_data_ref *ref2; | |
1285 | u32 num_refs = 0; | |
1286 | ||
1287 | leaf = path->nodes[0]; | |
1288 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1289 | if (iref) { | |
1290 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1291 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1292 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1293 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1294 | } else { | |
1295 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1296 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1297 | } | |
1298 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1299 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1300 | struct btrfs_extent_data_ref); | |
1301 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1302 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1303 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1304 | struct btrfs_shared_data_ref); | |
1305 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1306 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1307 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1308 | struct btrfs_extent_ref_v0 *ref0; | |
1309 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1310 | struct btrfs_extent_ref_v0); | |
1311 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1312 | #endif |
5d4f98a2 YZ |
1313 | } else { |
1314 | WARN_ON(1); | |
1315 | } | |
1316 | return num_refs; | |
1317 | } | |
15916de8 | 1318 | |
5d4f98a2 YZ |
1319 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1320 | struct btrfs_root *root, | |
1321 | struct btrfs_path *path, | |
1322 | u64 bytenr, u64 parent, | |
1323 | u64 root_objectid) | |
1f3c79a2 | 1324 | { |
5d4f98a2 | 1325 | struct btrfs_key key; |
1f3c79a2 | 1326 | int ret; |
1f3c79a2 | 1327 | |
5d4f98a2 YZ |
1328 | key.objectid = bytenr; |
1329 | if (parent) { | |
1330 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1331 | key.offset = parent; | |
1332 | } else { | |
1333 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1334 | key.offset = root_objectid; | |
1f3c79a2 LH |
1335 | } |
1336 | ||
5d4f98a2 YZ |
1337 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1338 | if (ret > 0) | |
1339 | ret = -ENOENT; | |
1340 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1341 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1342 | btrfs_release_path(path); |
5d4f98a2 YZ |
1343 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1344 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1345 | if (ret > 0) | |
1346 | ret = -ENOENT; | |
1347 | } | |
1f3c79a2 | 1348 | #endif |
5d4f98a2 | 1349 | return ret; |
1f3c79a2 LH |
1350 | } |
1351 | ||
5d4f98a2 YZ |
1352 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1353 | struct btrfs_root *root, | |
1354 | struct btrfs_path *path, | |
1355 | u64 bytenr, u64 parent, | |
1356 | u64 root_objectid) | |
31840ae1 | 1357 | { |
5d4f98a2 | 1358 | struct btrfs_key key; |
31840ae1 | 1359 | int ret; |
31840ae1 | 1360 | |
5d4f98a2 YZ |
1361 | key.objectid = bytenr; |
1362 | if (parent) { | |
1363 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1364 | key.offset = parent; | |
1365 | } else { | |
1366 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1367 | key.offset = root_objectid; | |
1368 | } | |
1369 | ||
1370 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1371 | btrfs_release_path(path); |
31840ae1 ZY |
1372 | return ret; |
1373 | } | |
1374 | ||
5d4f98a2 | 1375 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1376 | { |
5d4f98a2 YZ |
1377 | int type; |
1378 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1379 | if (parent > 0) | |
1380 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1381 | else | |
1382 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1383 | } else { | |
1384 | if (parent > 0) | |
1385 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1386 | else | |
1387 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1388 | } | |
1389 | return type; | |
31840ae1 | 1390 | } |
56bec294 | 1391 | |
2c47e605 YZ |
1392 | static int find_next_key(struct btrfs_path *path, int level, |
1393 | struct btrfs_key *key) | |
56bec294 | 1394 | |
02217ed2 | 1395 | { |
2c47e605 | 1396 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1397 | if (!path->nodes[level]) |
1398 | break; | |
5d4f98a2 YZ |
1399 | if (path->slots[level] + 1 >= |
1400 | btrfs_header_nritems(path->nodes[level])) | |
1401 | continue; | |
1402 | if (level == 0) | |
1403 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1404 | path->slots[level] + 1); | |
1405 | else | |
1406 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1407 | path->slots[level] + 1); | |
1408 | return 0; | |
1409 | } | |
1410 | return 1; | |
1411 | } | |
037e6390 | 1412 | |
5d4f98a2 YZ |
1413 | /* |
1414 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1415 | * to the address of inline back ref, and 0 is returned. | |
1416 | * | |
1417 | * if back ref isn't found, *ref_ret is set to the address where it | |
1418 | * should be inserted, and -ENOENT is returned. | |
1419 | * | |
1420 | * if insert is true and there are too many inline back refs, the path | |
1421 | * points to the extent item, and -EAGAIN is returned. | |
1422 | * | |
1423 | * NOTE: inline back refs are ordered in the same way that back ref | |
1424 | * items in the tree are ordered. | |
1425 | */ | |
1426 | static noinline_for_stack | |
1427 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1428 | struct btrfs_root *root, | |
1429 | struct btrfs_path *path, | |
1430 | struct btrfs_extent_inline_ref **ref_ret, | |
1431 | u64 bytenr, u64 num_bytes, | |
1432 | u64 parent, u64 root_objectid, | |
1433 | u64 owner, u64 offset, int insert) | |
1434 | { | |
1435 | struct btrfs_key key; | |
1436 | struct extent_buffer *leaf; | |
1437 | struct btrfs_extent_item *ei; | |
1438 | struct btrfs_extent_inline_ref *iref; | |
1439 | u64 flags; | |
1440 | u64 item_size; | |
1441 | unsigned long ptr; | |
1442 | unsigned long end; | |
1443 | int extra_size; | |
1444 | int type; | |
1445 | int want; | |
1446 | int ret; | |
1447 | int err = 0; | |
26b8003f | 1448 | |
db94535d | 1449 | key.objectid = bytenr; |
31840ae1 | 1450 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1451 | key.offset = num_bytes; |
31840ae1 | 1452 | |
5d4f98a2 YZ |
1453 | want = extent_ref_type(parent, owner); |
1454 | if (insert) { | |
1455 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1456 | path->keep_locks = 1; |
5d4f98a2 YZ |
1457 | } else |
1458 | extra_size = -1; | |
1459 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1460 | if (ret < 0) { |
5d4f98a2 YZ |
1461 | err = ret; |
1462 | goto out; | |
1463 | } | |
79787eaa JM |
1464 | if (ret && !insert) { |
1465 | err = -ENOENT; | |
1466 | goto out; | |
1467 | } | |
1468 | BUG_ON(ret); /* Corruption */ | |
5d4f98a2 YZ |
1469 | |
1470 | leaf = path->nodes[0]; | |
1471 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1472 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1473 | if (item_size < sizeof(*ei)) { | |
1474 | if (!insert) { | |
1475 | err = -ENOENT; | |
1476 | goto out; | |
1477 | } | |
1478 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1479 | extra_size); | |
1480 | if (ret < 0) { | |
1481 | err = ret; | |
1482 | goto out; | |
1483 | } | |
1484 | leaf = path->nodes[0]; | |
1485 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1486 | } | |
1487 | #endif | |
1488 | BUG_ON(item_size < sizeof(*ei)); | |
1489 | ||
5d4f98a2 YZ |
1490 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1491 | flags = btrfs_extent_flags(leaf, ei); | |
1492 | ||
1493 | ptr = (unsigned long)(ei + 1); | |
1494 | end = (unsigned long)ei + item_size; | |
1495 | ||
1496 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1497 | ptr += sizeof(struct btrfs_tree_block_info); | |
1498 | BUG_ON(ptr > end); | |
1499 | } else { | |
1500 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1501 | } | |
1502 | ||
1503 | err = -ENOENT; | |
1504 | while (1) { | |
1505 | if (ptr >= end) { | |
1506 | WARN_ON(ptr > end); | |
1507 | break; | |
1508 | } | |
1509 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1510 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1511 | if (want < type) | |
1512 | break; | |
1513 | if (want > type) { | |
1514 | ptr += btrfs_extent_inline_ref_size(type); | |
1515 | continue; | |
1516 | } | |
1517 | ||
1518 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1519 | struct btrfs_extent_data_ref *dref; | |
1520 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1521 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1522 | owner, offset)) { | |
1523 | err = 0; | |
1524 | break; | |
1525 | } | |
1526 | if (hash_extent_data_ref_item(leaf, dref) < | |
1527 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1528 | break; | |
1529 | } else { | |
1530 | u64 ref_offset; | |
1531 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1532 | if (parent > 0) { | |
1533 | if (parent == ref_offset) { | |
1534 | err = 0; | |
1535 | break; | |
1536 | } | |
1537 | if (ref_offset < parent) | |
1538 | break; | |
1539 | } else { | |
1540 | if (root_objectid == ref_offset) { | |
1541 | err = 0; | |
1542 | break; | |
1543 | } | |
1544 | if (ref_offset < root_objectid) | |
1545 | break; | |
1546 | } | |
1547 | } | |
1548 | ptr += btrfs_extent_inline_ref_size(type); | |
1549 | } | |
1550 | if (err == -ENOENT && insert) { | |
1551 | if (item_size + extra_size >= | |
1552 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1553 | err = -EAGAIN; | |
1554 | goto out; | |
1555 | } | |
1556 | /* | |
1557 | * To add new inline back ref, we have to make sure | |
1558 | * there is no corresponding back ref item. | |
1559 | * For simplicity, we just do not add new inline back | |
1560 | * ref if there is any kind of item for this block | |
1561 | */ | |
2c47e605 YZ |
1562 | if (find_next_key(path, 0, &key) == 0 && |
1563 | key.objectid == bytenr && | |
85d4198e | 1564 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1565 | err = -EAGAIN; |
1566 | goto out; | |
1567 | } | |
1568 | } | |
1569 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1570 | out: | |
85d4198e | 1571 | if (insert) { |
5d4f98a2 YZ |
1572 | path->keep_locks = 0; |
1573 | btrfs_unlock_up_safe(path, 1); | |
1574 | } | |
1575 | return err; | |
1576 | } | |
1577 | ||
1578 | /* | |
1579 | * helper to add new inline back ref | |
1580 | */ | |
1581 | static noinline_for_stack | |
143bede5 JM |
1582 | void setup_inline_extent_backref(struct btrfs_trans_handle *trans, |
1583 | struct btrfs_root *root, | |
1584 | struct btrfs_path *path, | |
1585 | struct btrfs_extent_inline_ref *iref, | |
1586 | u64 parent, u64 root_objectid, | |
1587 | u64 owner, u64 offset, int refs_to_add, | |
1588 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1589 | { |
1590 | struct extent_buffer *leaf; | |
1591 | struct btrfs_extent_item *ei; | |
1592 | unsigned long ptr; | |
1593 | unsigned long end; | |
1594 | unsigned long item_offset; | |
1595 | u64 refs; | |
1596 | int size; | |
1597 | int type; | |
5d4f98a2 YZ |
1598 | |
1599 | leaf = path->nodes[0]; | |
1600 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1601 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1602 | ||
1603 | type = extent_ref_type(parent, owner); | |
1604 | size = btrfs_extent_inline_ref_size(type); | |
1605 | ||
143bede5 | 1606 | btrfs_extend_item(trans, root, path, size); |
5d4f98a2 YZ |
1607 | |
1608 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1609 | refs = btrfs_extent_refs(leaf, ei); | |
1610 | refs += refs_to_add; | |
1611 | btrfs_set_extent_refs(leaf, ei, refs); | |
1612 | if (extent_op) | |
1613 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1614 | ||
1615 | ptr = (unsigned long)ei + item_offset; | |
1616 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1617 | if (ptr < end - size) | |
1618 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1619 | end - size - ptr); | |
1620 | ||
1621 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1622 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1623 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1624 | struct btrfs_extent_data_ref *dref; | |
1625 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1626 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1627 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1628 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1629 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1630 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1631 | struct btrfs_shared_data_ref *sref; | |
1632 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1633 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1634 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1635 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1636 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1637 | } else { | |
1638 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1639 | } | |
1640 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1641 | } |
1642 | ||
1643 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1644 | struct btrfs_root *root, | |
1645 | struct btrfs_path *path, | |
1646 | struct btrfs_extent_inline_ref **ref_ret, | |
1647 | u64 bytenr, u64 num_bytes, u64 parent, | |
1648 | u64 root_objectid, u64 owner, u64 offset) | |
1649 | { | |
1650 | int ret; | |
1651 | ||
1652 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1653 | bytenr, num_bytes, parent, | |
1654 | root_objectid, owner, offset, 0); | |
1655 | if (ret != -ENOENT) | |
54aa1f4d | 1656 | return ret; |
5d4f98a2 | 1657 | |
b3b4aa74 | 1658 | btrfs_release_path(path); |
5d4f98a2 YZ |
1659 | *ref_ret = NULL; |
1660 | ||
1661 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1662 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1663 | root_objectid); | |
1664 | } else { | |
1665 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1666 | root_objectid, owner, offset); | |
b9473439 | 1667 | } |
5d4f98a2 YZ |
1668 | return ret; |
1669 | } | |
31840ae1 | 1670 | |
5d4f98a2 YZ |
1671 | /* |
1672 | * helper to update/remove inline back ref | |
1673 | */ | |
1674 | static noinline_for_stack | |
143bede5 JM |
1675 | void update_inline_extent_backref(struct btrfs_trans_handle *trans, |
1676 | struct btrfs_root *root, | |
1677 | struct btrfs_path *path, | |
1678 | struct btrfs_extent_inline_ref *iref, | |
1679 | int refs_to_mod, | |
1680 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1681 | { |
1682 | struct extent_buffer *leaf; | |
1683 | struct btrfs_extent_item *ei; | |
1684 | struct btrfs_extent_data_ref *dref = NULL; | |
1685 | struct btrfs_shared_data_ref *sref = NULL; | |
1686 | unsigned long ptr; | |
1687 | unsigned long end; | |
1688 | u32 item_size; | |
1689 | int size; | |
1690 | int type; | |
5d4f98a2 YZ |
1691 | u64 refs; |
1692 | ||
1693 | leaf = path->nodes[0]; | |
1694 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1695 | refs = btrfs_extent_refs(leaf, ei); | |
1696 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1697 | refs += refs_to_mod; | |
1698 | btrfs_set_extent_refs(leaf, ei, refs); | |
1699 | if (extent_op) | |
1700 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1701 | ||
1702 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1703 | ||
1704 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1705 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1706 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1707 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1708 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1709 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1710 | } else { | |
1711 | refs = 1; | |
1712 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1713 | } |
31840ae1 | 1714 | |
5d4f98a2 YZ |
1715 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1716 | refs += refs_to_mod; | |
1717 | ||
1718 | if (refs > 0) { | |
1719 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1720 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1721 | else | |
1722 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1723 | } else { | |
1724 | size = btrfs_extent_inline_ref_size(type); | |
1725 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1726 | ptr = (unsigned long)iref; | |
1727 | end = (unsigned long)ei + item_size; | |
1728 | if (ptr + size < end) | |
1729 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1730 | end - ptr - size); | |
1731 | item_size -= size; | |
143bede5 | 1732 | btrfs_truncate_item(trans, root, path, item_size, 1); |
5d4f98a2 YZ |
1733 | } |
1734 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1735 | } |
1736 | ||
1737 | static noinline_for_stack | |
1738 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1739 | struct btrfs_root *root, | |
1740 | struct btrfs_path *path, | |
1741 | u64 bytenr, u64 num_bytes, u64 parent, | |
1742 | u64 root_objectid, u64 owner, | |
1743 | u64 offset, int refs_to_add, | |
1744 | struct btrfs_delayed_extent_op *extent_op) | |
1745 | { | |
1746 | struct btrfs_extent_inline_ref *iref; | |
1747 | int ret; | |
1748 | ||
1749 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1750 | bytenr, num_bytes, parent, | |
1751 | root_objectid, owner, offset, 1); | |
1752 | if (ret == 0) { | |
1753 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
143bede5 JM |
1754 | update_inline_extent_backref(trans, root, path, iref, |
1755 | refs_to_add, extent_op); | |
5d4f98a2 | 1756 | } else if (ret == -ENOENT) { |
143bede5 JM |
1757 | setup_inline_extent_backref(trans, root, path, iref, parent, |
1758 | root_objectid, owner, offset, | |
1759 | refs_to_add, extent_op); | |
1760 | ret = 0; | |
771ed689 | 1761 | } |
5d4f98a2 YZ |
1762 | return ret; |
1763 | } | |
31840ae1 | 1764 | |
5d4f98a2 YZ |
1765 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1766 | struct btrfs_root *root, | |
1767 | struct btrfs_path *path, | |
1768 | u64 bytenr, u64 parent, u64 root_objectid, | |
1769 | u64 owner, u64 offset, int refs_to_add) | |
1770 | { | |
1771 | int ret; | |
1772 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1773 | BUG_ON(refs_to_add != 1); | |
1774 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1775 | parent, root_objectid); | |
1776 | } else { | |
1777 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1778 | parent, root_objectid, | |
1779 | owner, offset, refs_to_add); | |
1780 | } | |
1781 | return ret; | |
1782 | } | |
56bec294 | 1783 | |
5d4f98a2 YZ |
1784 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1785 | struct btrfs_root *root, | |
1786 | struct btrfs_path *path, | |
1787 | struct btrfs_extent_inline_ref *iref, | |
1788 | int refs_to_drop, int is_data) | |
1789 | { | |
143bede5 | 1790 | int ret = 0; |
b9473439 | 1791 | |
5d4f98a2 YZ |
1792 | BUG_ON(!is_data && refs_to_drop != 1); |
1793 | if (iref) { | |
143bede5 JM |
1794 | update_inline_extent_backref(trans, root, path, iref, |
1795 | -refs_to_drop, NULL); | |
5d4f98a2 YZ |
1796 | } else if (is_data) { |
1797 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1798 | } else { | |
1799 | ret = btrfs_del_item(trans, root, path); | |
1800 | } | |
1801 | return ret; | |
1802 | } | |
1803 | ||
5378e607 | 1804 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1805 | u64 start, u64 len) |
1806 | { | |
5378e607 | 1807 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1808 | } |
5d4f98a2 YZ |
1809 | |
1810 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1811 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1812 | { |
5d4f98a2 | 1813 | int ret; |
5378e607 | 1814 | u64 discarded_bytes = 0; |
a1d3c478 | 1815 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1816 | |
e244a0ae | 1817 | |
5d4f98a2 | 1818 | /* Tell the block device(s) that the sectors can be discarded */ |
3ec706c8 | 1819 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
a1d3c478 | 1820 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 1821 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1822 | if (!ret) { |
a1d3c478 | 1823 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1824 | int i; |
1825 | ||
5d4f98a2 | 1826 | |
a1d3c478 | 1827 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d5e2003c JB |
1828 | if (!stripe->dev->can_discard) |
1829 | continue; | |
1830 | ||
5378e607 LD |
1831 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1832 | stripe->physical, | |
1833 | stripe->length); | |
1834 | if (!ret) | |
1835 | discarded_bytes += stripe->length; | |
1836 | else if (ret != -EOPNOTSUPP) | |
79787eaa | 1837 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
1838 | |
1839 | /* | |
1840 | * Just in case we get back EOPNOTSUPP for some reason, | |
1841 | * just ignore the return value so we don't screw up | |
1842 | * people calling discard_extent. | |
1843 | */ | |
1844 | ret = 0; | |
5d4f98a2 | 1845 | } |
a1d3c478 | 1846 | kfree(bbio); |
5d4f98a2 | 1847 | } |
5378e607 LD |
1848 | |
1849 | if (actual_bytes) | |
1850 | *actual_bytes = discarded_bytes; | |
1851 | ||
5d4f98a2 YZ |
1852 | |
1853 | return ret; | |
5d4f98a2 YZ |
1854 | } |
1855 | ||
79787eaa | 1856 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
1857 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
1858 | struct btrfs_root *root, | |
1859 | u64 bytenr, u64 num_bytes, u64 parent, | |
66d7e7f0 | 1860 | u64 root_objectid, u64 owner, u64 offset, int for_cow) |
5d4f98a2 YZ |
1861 | { |
1862 | int ret; | |
66d7e7f0 AJ |
1863 | struct btrfs_fs_info *fs_info = root->fs_info; |
1864 | ||
5d4f98a2 YZ |
1865 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
1866 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1867 | ||
1868 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
1869 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
1870 | num_bytes, | |
5d4f98a2 | 1871 | parent, root_objectid, (int)owner, |
66d7e7f0 | 1872 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 1873 | } else { |
66d7e7f0 AJ |
1874 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
1875 | num_bytes, | |
5d4f98a2 | 1876 | parent, root_objectid, owner, offset, |
66d7e7f0 | 1877 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 YZ |
1878 | } |
1879 | return ret; | |
1880 | } | |
1881 | ||
1882 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1883 | struct btrfs_root *root, | |
1884 | u64 bytenr, u64 num_bytes, | |
1885 | u64 parent, u64 root_objectid, | |
1886 | u64 owner, u64 offset, int refs_to_add, | |
1887 | struct btrfs_delayed_extent_op *extent_op) | |
1888 | { | |
1889 | struct btrfs_path *path; | |
1890 | struct extent_buffer *leaf; | |
1891 | struct btrfs_extent_item *item; | |
1892 | u64 refs; | |
1893 | int ret; | |
1894 | int err = 0; | |
1895 | ||
1896 | path = btrfs_alloc_path(); | |
1897 | if (!path) | |
1898 | return -ENOMEM; | |
1899 | ||
1900 | path->reada = 1; | |
1901 | path->leave_spinning = 1; | |
1902 | /* this will setup the path even if it fails to insert the back ref */ | |
1903 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1904 | path, bytenr, num_bytes, parent, | |
1905 | root_objectid, owner, offset, | |
1906 | refs_to_add, extent_op); | |
1907 | if (ret == 0) | |
1908 | goto out; | |
1909 | ||
1910 | if (ret != -EAGAIN) { | |
1911 | err = ret; | |
1912 | goto out; | |
1913 | } | |
1914 | ||
1915 | leaf = path->nodes[0]; | |
1916 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1917 | refs = btrfs_extent_refs(leaf, item); | |
1918 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1919 | if (extent_op) | |
1920 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1921 | |
5d4f98a2 | 1922 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1923 | btrfs_release_path(path); |
56bec294 CM |
1924 | |
1925 | path->reada = 1; | |
b9473439 CM |
1926 | path->leave_spinning = 1; |
1927 | ||
56bec294 CM |
1928 | /* now insert the actual backref */ |
1929 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1930 | path, bytenr, parent, root_objectid, |
1931 | owner, offset, refs_to_add); | |
79787eaa JM |
1932 | if (ret) |
1933 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 1934 | out: |
56bec294 | 1935 | btrfs_free_path(path); |
5d4f98a2 | 1936 | return err; |
56bec294 CM |
1937 | } |
1938 | ||
5d4f98a2 YZ |
1939 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1940 | struct btrfs_root *root, | |
1941 | struct btrfs_delayed_ref_node *node, | |
1942 | struct btrfs_delayed_extent_op *extent_op, | |
1943 | int insert_reserved) | |
56bec294 | 1944 | { |
5d4f98a2 YZ |
1945 | int ret = 0; |
1946 | struct btrfs_delayed_data_ref *ref; | |
1947 | struct btrfs_key ins; | |
1948 | u64 parent = 0; | |
1949 | u64 ref_root = 0; | |
1950 | u64 flags = 0; | |
1951 | ||
1952 | ins.objectid = node->bytenr; | |
1953 | ins.offset = node->num_bytes; | |
1954 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1955 | ||
1956 | ref = btrfs_delayed_node_to_data_ref(node); | |
1957 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1958 | parent = ref->parent; | |
1959 | else | |
1960 | ref_root = ref->root; | |
1961 | ||
1962 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1963 | if (extent_op) { | |
1964 | BUG_ON(extent_op->update_key); | |
1965 | flags |= extent_op->flags_to_set; | |
1966 | } | |
1967 | ret = alloc_reserved_file_extent(trans, root, | |
1968 | parent, ref_root, flags, | |
1969 | ref->objectid, ref->offset, | |
1970 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1971 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1972 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1973 | node->num_bytes, parent, | |
1974 | ref_root, ref->objectid, | |
1975 | ref->offset, node->ref_mod, | |
1976 | extent_op); | |
1977 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1978 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1979 | node->num_bytes, parent, | |
1980 | ref_root, ref->objectid, | |
1981 | ref->offset, node->ref_mod, | |
1982 | extent_op); | |
1983 | } else { | |
1984 | BUG(); | |
1985 | } | |
1986 | return ret; | |
1987 | } | |
1988 | ||
1989 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1990 | struct extent_buffer *leaf, | |
1991 | struct btrfs_extent_item *ei) | |
1992 | { | |
1993 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1994 | if (extent_op->update_flags) { | |
1995 | flags |= extent_op->flags_to_set; | |
1996 | btrfs_set_extent_flags(leaf, ei, flags); | |
1997 | } | |
1998 | ||
1999 | if (extent_op->update_key) { | |
2000 | struct btrfs_tree_block_info *bi; | |
2001 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2002 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2003 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2004 | } | |
2005 | } | |
2006 | ||
2007 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2008 | struct btrfs_root *root, | |
2009 | struct btrfs_delayed_ref_node *node, | |
2010 | struct btrfs_delayed_extent_op *extent_op) | |
2011 | { | |
2012 | struct btrfs_key key; | |
2013 | struct btrfs_path *path; | |
2014 | struct btrfs_extent_item *ei; | |
2015 | struct extent_buffer *leaf; | |
2016 | u32 item_size; | |
56bec294 | 2017 | int ret; |
5d4f98a2 YZ |
2018 | int err = 0; |
2019 | ||
79787eaa JM |
2020 | if (trans->aborted) |
2021 | return 0; | |
2022 | ||
5d4f98a2 YZ |
2023 | path = btrfs_alloc_path(); |
2024 | if (!path) | |
2025 | return -ENOMEM; | |
2026 | ||
2027 | key.objectid = node->bytenr; | |
2028 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2029 | key.offset = node->num_bytes; | |
2030 | ||
2031 | path->reada = 1; | |
2032 | path->leave_spinning = 1; | |
2033 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2034 | path, 0, 1); | |
2035 | if (ret < 0) { | |
2036 | err = ret; | |
2037 | goto out; | |
2038 | } | |
2039 | if (ret > 0) { | |
2040 | err = -EIO; | |
2041 | goto out; | |
2042 | } | |
2043 | ||
2044 | leaf = path->nodes[0]; | |
2045 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2046 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2047 | if (item_size < sizeof(*ei)) { | |
2048 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2049 | path, (u64)-1, 0); | |
2050 | if (ret < 0) { | |
2051 | err = ret; | |
2052 | goto out; | |
2053 | } | |
2054 | leaf = path->nodes[0]; | |
2055 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2056 | } | |
2057 | #endif | |
2058 | BUG_ON(item_size < sizeof(*ei)); | |
2059 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2060 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2061 | |
5d4f98a2 YZ |
2062 | btrfs_mark_buffer_dirty(leaf); |
2063 | out: | |
2064 | btrfs_free_path(path); | |
2065 | return err; | |
56bec294 CM |
2066 | } |
2067 | ||
5d4f98a2 YZ |
2068 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2069 | struct btrfs_root *root, | |
2070 | struct btrfs_delayed_ref_node *node, | |
2071 | struct btrfs_delayed_extent_op *extent_op, | |
2072 | int insert_reserved) | |
56bec294 CM |
2073 | { |
2074 | int ret = 0; | |
5d4f98a2 YZ |
2075 | struct btrfs_delayed_tree_ref *ref; |
2076 | struct btrfs_key ins; | |
2077 | u64 parent = 0; | |
2078 | u64 ref_root = 0; | |
56bec294 | 2079 | |
5d4f98a2 YZ |
2080 | ins.objectid = node->bytenr; |
2081 | ins.offset = node->num_bytes; | |
2082 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 2083 | |
5d4f98a2 YZ |
2084 | ref = btrfs_delayed_node_to_tree_ref(node); |
2085 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2086 | parent = ref->parent; | |
2087 | else | |
2088 | ref_root = ref->root; | |
2089 | ||
2090 | BUG_ON(node->ref_mod != 1); | |
2091 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2092 | BUG_ON(!extent_op || !extent_op->update_flags || | |
2093 | !extent_op->update_key); | |
2094 | ret = alloc_reserved_tree_block(trans, root, | |
2095 | parent, ref_root, | |
2096 | extent_op->flags_to_set, | |
2097 | &extent_op->key, | |
2098 | ref->level, &ins); | |
5d4f98a2 YZ |
2099 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2100 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2101 | node->num_bytes, parent, ref_root, | |
2102 | ref->level, 0, 1, extent_op); | |
2103 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2104 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2105 | node->num_bytes, parent, ref_root, | |
2106 | ref->level, 0, 1, extent_op); | |
2107 | } else { | |
2108 | BUG(); | |
2109 | } | |
56bec294 CM |
2110 | return ret; |
2111 | } | |
2112 | ||
2113 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2114 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2115 | struct btrfs_root *root, | |
2116 | struct btrfs_delayed_ref_node *node, | |
2117 | struct btrfs_delayed_extent_op *extent_op, | |
2118 | int insert_reserved) | |
56bec294 | 2119 | { |
79787eaa JM |
2120 | int ret = 0; |
2121 | ||
2122 | if (trans->aborted) | |
2123 | return 0; | |
2124 | ||
5d4f98a2 | 2125 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2126 | struct btrfs_delayed_ref_head *head; |
2127 | /* | |
2128 | * we've hit the end of the chain and we were supposed | |
2129 | * to insert this extent into the tree. But, it got | |
2130 | * deleted before we ever needed to insert it, so all | |
2131 | * we have to do is clean up the accounting | |
2132 | */ | |
5d4f98a2 YZ |
2133 | BUG_ON(extent_op); |
2134 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2135 | if (insert_reserved) { |
f0486c68 YZ |
2136 | btrfs_pin_extent(root, node->bytenr, |
2137 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2138 | if (head->is_data) { |
2139 | ret = btrfs_del_csums(trans, root, | |
2140 | node->bytenr, | |
2141 | node->num_bytes); | |
5d4f98a2 | 2142 | } |
56bec294 | 2143 | } |
79787eaa | 2144 | return ret; |
56bec294 CM |
2145 | } |
2146 | ||
5d4f98a2 YZ |
2147 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2148 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2149 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2150 | insert_reserved); | |
2151 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2152 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2153 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2154 | insert_reserved); | |
2155 | else | |
2156 | BUG(); | |
2157 | return ret; | |
56bec294 CM |
2158 | } |
2159 | ||
2160 | static noinline struct btrfs_delayed_ref_node * | |
2161 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2162 | { | |
2163 | struct rb_node *node; | |
2164 | struct btrfs_delayed_ref_node *ref; | |
2165 | int action = BTRFS_ADD_DELAYED_REF; | |
2166 | again: | |
2167 | /* | |
2168 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2169 | * this prevents ref count from going down to zero when | |
2170 | * there still are pending delayed ref. | |
2171 | */ | |
2172 | node = rb_prev(&head->node.rb_node); | |
2173 | while (1) { | |
2174 | if (!node) | |
2175 | break; | |
2176 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2177 | rb_node); | |
2178 | if (ref->bytenr != head->node.bytenr) | |
2179 | break; | |
5d4f98a2 | 2180 | if (ref->action == action) |
56bec294 CM |
2181 | return ref; |
2182 | node = rb_prev(node); | |
2183 | } | |
2184 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2185 | action = BTRFS_DROP_DELAYED_REF; | |
2186 | goto again; | |
2187 | } | |
2188 | return NULL; | |
2189 | } | |
2190 | ||
79787eaa JM |
2191 | /* |
2192 | * Returns 0 on success or if called with an already aborted transaction. | |
2193 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2194 | */ | |
c3e69d58 CM |
2195 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2196 | struct btrfs_root *root, | |
2197 | struct list_head *cluster) | |
56bec294 | 2198 | { |
56bec294 CM |
2199 | struct btrfs_delayed_ref_root *delayed_refs; |
2200 | struct btrfs_delayed_ref_node *ref; | |
2201 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2202 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2203 | struct btrfs_fs_info *fs_info = root->fs_info; |
56bec294 | 2204 | int ret; |
c3e69d58 | 2205 | int count = 0; |
56bec294 | 2206 | int must_insert_reserved = 0; |
56bec294 CM |
2207 | |
2208 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2209 | while (1) { |
2210 | if (!locked_ref) { | |
c3e69d58 CM |
2211 | /* pick a new head ref from the cluster list */ |
2212 | if (list_empty(cluster)) | |
56bec294 | 2213 | break; |
56bec294 | 2214 | |
c3e69d58 CM |
2215 | locked_ref = list_entry(cluster->next, |
2216 | struct btrfs_delayed_ref_head, cluster); | |
2217 | ||
2218 | /* grab the lock that says we are going to process | |
2219 | * all the refs for this head */ | |
2220 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2221 | ||
2222 | /* | |
2223 | * we may have dropped the spin lock to get the head | |
2224 | * mutex lock, and that might have given someone else | |
2225 | * time to free the head. If that's true, it has been | |
2226 | * removed from our list and we can move on. | |
2227 | */ | |
2228 | if (ret == -EAGAIN) { | |
2229 | locked_ref = NULL; | |
2230 | count++; | |
2231 | continue; | |
56bec294 CM |
2232 | } |
2233 | } | |
a28ec197 | 2234 | |
ae1e206b JB |
2235 | /* |
2236 | * We need to try and merge add/drops of the same ref since we | |
2237 | * can run into issues with relocate dropping the implicit ref | |
2238 | * and then it being added back again before the drop can | |
2239 | * finish. If we merged anything we need to re-loop so we can | |
2240 | * get a good ref. | |
2241 | */ | |
2242 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, | |
2243 | locked_ref); | |
2244 | ||
d1270cd9 AJ |
2245 | /* |
2246 | * locked_ref is the head node, so we have to go one | |
2247 | * node back for any delayed ref updates | |
2248 | */ | |
2249 | ref = select_delayed_ref(locked_ref); | |
2250 | ||
2251 | if (ref && ref->seq && | |
097b8a7c | 2252 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d1270cd9 AJ |
2253 | /* |
2254 | * there are still refs with lower seq numbers in the | |
2255 | * process of being added. Don't run this ref yet. | |
2256 | */ | |
2257 | list_del_init(&locked_ref->cluster); | |
093486c4 | 2258 | btrfs_delayed_ref_unlock(locked_ref); |
d1270cd9 AJ |
2259 | locked_ref = NULL; |
2260 | delayed_refs->num_heads_ready++; | |
2261 | spin_unlock(&delayed_refs->lock); | |
2262 | cond_resched(); | |
2263 | spin_lock(&delayed_refs->lock); | |
2264 | continue; | |
2265 | } | |
2266 | ||
56bec294 CM |
2267 | /* |
2268 | * record the must insert reserved flag before we | |
2269 | * drop the spin lock. | |
2270 | */ | |
2271 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2272 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2273 | |
5d4f98a2 YZ |
2274 | extent_op = locked_ref->extent_op; |
2275 | locked_ref->extent_op = NULL; | |
2276 | ||
56bec294 CM |
2277 | if (!ref) { |
2278 | /* All delayed refs have been processed, Go ahead | |
2279 | * and send the head node to run_one_delayed_ref, | |
2280 | * so that any accounting fixes can happen | |
2281 | */ | |
2282 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2283 | |
2284 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2285 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2286 | extent_op = NULL; |
2287 | } | |
2288 | ||
2289 | if (extent_op) { | |
2290 | spin_unlock(&delayed_refs->lock); | |
2291 | ||
2292 | ret = run_delayed_extent_op(trans, root, | |
2293 | ref, extent_op); | |
78a6184a | 2294 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2295 | |
79787eaa | 2296 | if (ret) { |
093486c4 MX |
2297 | printk(KERN_DEBUG |
2298 | "btrfs: run_delayed_extent_op " | |
2299 | "returned %d\n", ret); | |
253beebd | 2300 | spin_lock(&delayed_refs->lock); |
093486c4 | 2301 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2302 | return ret; |
2303 | } | |
2304 | ||
203bf287 | 2305 | goto next; |
5d4f98a2 | 2306 | } |
56bec294 | 2307 | } |
02217ed2 | 2308 | |
56bec294 CM |
2309 | ref->in_tree = 0; |
2310 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2311 | delayed_refs->num_entries--; | |
093486c4 | 2312 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2313 | /* |
2314 | * when we play the delayed ref, also correct the | |
2315 | * ref_mod on head | |
2316 | */ | |
2317 | switch (ref->action) { | |
2318 | case BTRFS_ADD_DELAYED_REF: | |
2319 | case BTRFS_ADD_DELAYED_EXTENT: | |
2320 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2321 | break; | |
2322 | case BTRFS_DROP_DELAYED_REF: | |
2323 | locked_ref->node.ref_mod += ref->ref_mod; | |
2324 | break; | |
2325 | default: | |
2326 | WARN_ON(1); | |
2327 | } | |
2328 | } | |
56bec294 | 2329 | spin_unlock(&delayed_refs->lock); |
925baedd | 2330 | |
5d4f98a2 | 2331 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2332 | must_insert_reserved); |
eb099670 | 2333 | |
78a6184a | 2334 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2335 | if (ret) { |
093486c4 MX |
2336 | btrfs_delayed_ref_unlock(locked_ref); |
2337 | btrfs_put_delayed_ref(ref); | |
2338 | printk(KERN_DEBUG | |
2339 | "btrfs: run_one_delayed_ref returned %d\n", ret); | |
253beebd | 2340 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2341 | return ret; |
2342 | } | |
2343 | ||
093486c4 MX |
2344 | /* |
2345 | * If this node is a head, that means all the refs in this head | |
2346 | * have been dealt with, and we will pick the next head to deal | |
2347 | * with, so we must unlock the head and drop it from the cluster | |
2348 | * list before we release it. | |
2349 | */ | |
2350 | if (btrfs_delayed_ref_is_head(ref)) { | |
2351 | list_del_init(&locked_ref->cluster); | |
2352 | btrfs_delayed_ref_unlock(locked_ref); | |
2353 | locked_ref = NULL; | |
2354 | } | |
2355 | btrfs_put_delayed_ref(ref); | |
2356 | count++; | |
203bf287 | 2357 | next: |
c3e69d58 CM |
2358 | cond_resched(); |
2359 | spin_lock(&delayed_refs->lock); | |
2360 | } | |
2361 | return count; | |
2362 | } | |
2363 | ||
709c0486 AJ |
2364 | #ifdef SCRAMBLE_DELAYED_REFS |
2365 | /* | |
2366 | * Normally delayed refs get processed in ascending bytenr order. This | |
2367 | * correlates in most cases to the order added. To expose dependencies on this | |
2368 | * order, we start to process the tree in the middle instead of the beginning | |
2369 | */ | |
2370 | static u64 find_middle(struct rb_root *root) | |
2371 | { | |
2372 | struct rb_node *n = root->rb_node; | |
2373 | struct btrfs_delayed_ref_node *entry; | |
2374 | int alt = 1; | |
2375 | u64 middle; | |
2376 | u64 first = 0, last = 0; | |
2377 | ||
2378 | n = rb_first(root); | |
2379 | if (n) { | |
2380 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2381 | first = entry->bytenr; | |
2382 | } | |
2383 | n = rb_last(root); | |
2384 | if (n) { | |
2385 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2386 | last = entry->bytenr; | |
2387 | } | |
2388 | n = root->rb_node; | |
2389 | ||
2390 | while (n) { | |
2391 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2392 | WARN_ON(!entry->in_tree); | |
2393 | ||
2394 | middle = entry->bytenr; | |
2395 | ||
2396 | if (alt) | |
2397 | n = n->rb_left; | |
2398 | else | |
2399 | n = n->rb_right; | |
2400 | ||
2401 | alt = 1 - alt; | |
2402 | } | |
2403 | return middle; | |
2404 | } | |
2405 | #endif | |
2406 | ||
bed92eae AJ |
2407 | int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans, |
2408 | struct btrfs_fs_info *fs_info) | |
2409 | { | |
2410 | struct qgroup_update *qgroup_update; | |
2411 | int ret = 0; | |
2412 | ||
2413 | if (list_empty(&trans->qgroup_ref_list) != | |
2414 | !trans->delayed_ref_elem.seq) { | |
2415 | /* list without seq or seq without list */ | |
2416 | printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n", | |
2417 | list_empty(&trans->qgroup_ref_list) ? "" : " not", | |
2418 | trans->delayed_ref_elem.seq); | |
2419 | BUG(); | |
2420 | } | |
2421 | ||
2422 | if (!trans->delayed_ref_elem.seq) | |
2423 | return 0; | |
2424 | ||
2425 | while (!list_empty(&trans->qgroup_ref_list)) { | |
2426 | qgroup_update = list_first_entry(&trans->qgroup_ref_list, | |
2427 | struct qgroup_update, list); | |
2428 | list_del(&qgroup_update->list); | |
2429 | if (!ret) | |
2430 | ret = btrfs_qgroup_account_ref( | |
2431 | trans, fs_info, qgroup_update->node, | |
2432 | qgroup_update->extent_op); | |
2433 | kfree(qgroup_update); | |
2434 | } | |
2435 | ||
2436 | btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
2437 | ||
2438 | return ret; | |
2439 | } | |
2440 | ||
c3e69d58 CM |
2441 | /* |
2442 | * this starts processing the delayed reference count updates and | |
2443 | * extent insertions we have queued up so far. count can be | |
2444 | * 0, which means to process everything in the tree at the start | |
2445 | * of the run (but not newly added entries), or it can be some target | |
2446 | * number you'd like to process. | |
79787eaa JM |
2447 | * |
2448 | * Returns 0 on success or if called with an aborted transaction | |
2449 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2450 | */ |
2451 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2452 | struct btrfs_root *root, unsigned long count) | |
2453 | { | |
2454 | struct rb_node *node; | |
2455 | struct btrfs_delayed_ref_root *delayed_refs; | |
2456 | struct btrfs_delayed_ref_node *ref; | |
2457 | struct list_head cluster; | |
2458 | int ret; | |
a168650c | 2459 | u64 delayed_start; |
c3e69d58 CM |
2460 | int run_all = count == (unsigned long)-1; |
2461 | int run_most = 0; | |
1fa11e26 | 2462 | int loops; |
c3e69d58 | 2463 | |
79787eaa JM |
2464 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2465 | if (trans->aborted) | |
2466 | return 0; | |
2467 | ||
c3e69d58 CM |
2468 | if (root == root->fs_info->extent_root) |
2469 | root = root->fs_info->tree_root; | |
2470 | ||
edf39272 JS |
2471 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); |
2472 | ||
c3e69d58 CM |
2473 | delayed_refs = &trans->transaction->delayed_refs; |
2474 | INIT_LIST_HEAD(&cluster); | |
2475 | again: | |
1fa11e26 | 2476 | loops = 0; |
c3e69d58 | 2477 | spin_lock(&delayed_refs->lock); |
097b8a7c | 2478 | |
709c0486 AJ |
2479 | #ifdef SCRAMBLE_DELAYED_REFS |
2480 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2481 | #endif | |
2482 | ||
c3e69d58 CM |
2483 | if (count == 0) { |
2484 | count = delayed_refs->num_entries * 2; | |
2485 | run_most = 1; | |
2486 | } | |
2487 | while (1) { | |
2488 | if (!(run_all || run_most) && | |
2489 | delayed_refs->num_heads_ready < 64) | |
2490 | break; | |
eb099670 | 2491 | |
56bec294 | 2492 | /* |
c3e69d58 CM |
2493 | * go find something we can process in the rbtree. We start at |
2494 | * the beginning of the tree, and then build a cluster | |
2495 | * of refs to process starting at the first one we are able to | |
2496 | * lock | |
56bec294 | 2497 | */ |
a168650c | 2498 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2499 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2500 | delayed_refs->run_delayed_start); | |
2501 | if (ret) | |
56bec294 CM |
2502 | break; |
2503 | ||
c3e69d58 | 2504 | ret = run_clustered_refs(trans, root, &cluster); |
79787eaa | 2505 | if (ret < 0) { |
093486c4 | 2506 | btrfs_release_ref_cluster(&cluster); |
79787eaa JM |
2507 | spin_unlock(&delayed_refs->lock); |
2508 | btrfs_abort_transaction(trans, root, ret); | |
2509 | return ret; | |
2510 | } | |
c3e69d58 CM |
2511 | |
2512 | count -= min_t(unsigned long, ret, count); | |
2513 | ||
2514 | if (count == 0) | |
2515 | break; | |
a168650c | 2516 | |
1fa11e26 AJ |
2517 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2518 | if (loops == 0) { | |
2519 | /* | |
2520 | * btrfs_find_ref_cluster looped. let's do one | |
2521 | * more cycle. if we don't run any delayed ref | |
2522 | * during that cycle (because we can't because | |
2523 | * all of them are blocked), bail out. | |
2524 | */ | |
2525 | loops = 1; | |
2526 | } else { | |
2527 | /* | |
2528 | * no runnable refs left, stop trying | |
2529 | */ | |
2530 | BUG_ON(run_all); | |
2531 | break; | |
2532 | } | |
2533 | } | |
2534 | if (ret) { | |
a168650c | 2535 | /* refs were run, let's reset staleness detection */ |
1fa11e26 | 2536 | loops = 0; |
a168650c | 2537 | } |
eb099670 | 2538 | } |
c3e69d58 | 2539 | |
56bec294 | 2540 | if (run_all) { |
ea658bad JB |
2541 | if (!list_empty(&trans->new_bgs)) { |
2542 | spin_unlock(&delayed_refs->lock); | |
2543 | btrfs_create_pending_block_groups(trans, root); | |
2544 | spin_lock(&delayed_refs->lock); | |
2545 | } | |
2546 | ||
56bec294 | 2547 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2548 | if (!node) |
56bec294 | 2549 | goto out; |
c3e69d58 | 2550 | count = (unsigned long)-1; |
e9d0b13b | 2551 | |
56bec294 CM |
2552 | while (node) { |
2553 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2554 | rb_node); | |
2555 | if (btrfs_delayed_ref_is_head(ref)) { | |
2556 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2557 | |
56bec294 CM |
2558 | head = btrfs_delayed_node_to_head(ref); |
2559 | atomic_inc(&ref->refs); | |
2560 | ||
2561 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2562 | /* |
2563 | * Mutex was contended, block until it's | |
2564 | * released and try again | |
2565 | */ | |
56bec294 CM |
2566 | mutex_lock(&head->mutex); |
2567 | mutex_unlock(&head->mutex); | |
2568 | ||
2569 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2570 | cond_resched(); |
56bec294 CM |
2571 | goto again; |
2572 | } | |
2573 | node = rb_next(node); | |
2574 | } | |
2575 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2576 | schedule_timeout(1); |
2577 | goto again; | |
5f39d397 | 2578 | } |
54aa1f4d | 2579 | out: |
c3e69d58 | 2580 | spin_unlock(&delayed_refs->lock); |
edf39272 | 2581 | assert_qgroups_uptodate(trans); |
a28ec197 CM |
2582 | return 0; |
2583 | } | |
2584 | ||
5d4f98a2 YZ |
2585 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2586 | struct btrfs_root *root, | |
2587 | u64 bytenr, u64 num_bytes, u64 flags, | |
2588 | int is_data) | |
2589 | { | |
2590 | struct btrfs_delayed_extent_op *extent_op; | |
2591 | int ret; | |
2592 | ||
78a6184a | 2593 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2594 | if (!extent_op) |
2595 | return -ENOMEM; | |
2596 | ||
2597 | extent_op->flags_to_set = flags; | |
2598 | extent_op->update_flags = 1; | |
2599 | extent_op->update_key = 0; | |
2600 | extent_op->is_data = is_data ? 1 : 0; | |
2601 | ||
66d7e7f0 AJ |
2602 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2603 | num_bytes, extent_op); | |
5d4f98a2 | 2604 | if (ret) |
78a6184a | 2605 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2606 | return ret; |
2607 | } | |
2608 | ||
2609 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2610 | struct btrfs_root *root, | |
2611 | struct btrfs_path *path, | |
2612 | u64 objectid, u64 offset, u64 bytenr) | |
2613 | { | |
2614 | struct btrfs_delayed_ref_head *head; | |
2615 | struct btrfs_delayed_ref_node *ref; | |
2616 | struct btrfs_delayed_data_ref *data_ref; | |
2617 | struct btrfs_delayed_ref_root *delayed_refs; | |
2618 | struct rb_node *node; | |
2619 | int ret = 0; | |
2620 | ||
2621 | ret = -ENOENT; | |
2622 | delayed_refs = &trans->transaction->delayed_refs; | |
2623 | spin_lock(&delayed_refs->lock); | |
2624 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2625 | if (!head) | |
2626 | goto out; | |
2627 | ||
2628 | if (!mutex_trylock(&head->mutex)) { | |
2629 | atomic_inc(&head->node.refs); | |
2630 | spin_unlock(&delayed_refs->lock); | |
2631 | ||
b3b4aa74 | 2632 | btrfs_release_path(path); |
5d4f98a2 | 2633 | |
8cc33e5c DS |
2634 | /* |
2635 | * Mutex was contended, block until it's released and let | |
2636 | * caller try again | |
2637 | */ | |
5d4f98a2 YZ |
2638 | mutex_lock(&head->mutex); |
2639 | mutex_unlock(&head->mutex); | |
2640 | btrfs_put_delayed_ref(&head->node); | |
2641 | return -EAGAIN; | |
2642 | } | |
2643 | ||
2644 | node = rb_prev(&head->node.rb_node); | |
2645 | if (!node) | |
2646 | goto out_unlock; | |
2647 | ||
2648 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2649 | ||
2650 | if (ref->bytenr != bytenr) | |
2651 | goto out_unlock; | |
2652 | ||
2653 | ret = 1; | |
2654 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2655 | goto out_unlock; | |
2656 | ||
2657 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2658 | ||
2659 | node = rb_prev(node); | |
2660 | if (node) { | |
df57dbe6 LB |
2661 | int seq = ref->seq; |
2662 | ||
5d4f98a2 | 2663 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); |
df57dbe6 | 2664 | if (ref->bytenr == bytenr && ref->seq == seq) |
5d4f98a2 YZ |
2665 | goto out_unlock; |
2666 | } | |
2667 | ||
2668 | if (data_ref->root != root->root_key.objectid || | |
2669 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2670 | goto out_unlock; | |
2671 | ||
2672 | ret = 0; | |
2673 | out_unlock: | |
2674 | mutex_unlock(&head->mutex); | |
2675 | out: | |
2676 | spin_unlock(&delayed_refs->lock); | |
2677 | return ret; | |
2678 | } | |
2679 | ||
2680 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2681 | struct btrfs_root *root, | |
2682 | struct btrfs_path *path, | |
2683 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2684 | { |
2685 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2686 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2687 | struct btrfs_extent_data_ref *ref; |
2688 | struct btrfs_extent_inline_ref *iref; | |
2689 | struct btrfs_extent_item *ei; | |
f321e491 | 2690 | struct btrfs_key key; |
5d4f98a2 | 2691 | u32 item_size; |
be20aa9d | 2692 | int ret; |
925baedd | 2693 | |
be20aa9d | 2694 | key.objectid = bytenr; |
31840ae1 | 2695 | key.offset = (u64)-1; |
f321e491 | 2696 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2697 | |
be20aa9d CM |
2698 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2699 | if (ret < 0) | |
2700 | goto out; | |
79787eaa | 2701 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2702 | |
2703 | ret = -ENOENT; | |
2704 | if (path->slots[0] == 0) | |
31840ae1 | 2705 | goto out; |
be20aa9d | 2706 | |
31840ae1 | 2707 | path->slots[0]--; |
f321e491 | 2708 | leaf = path->nodes[0]; |
5d4f98a2 | 2709 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2710 | |
5d4f98a2 | 2711 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2712 | goto out; |
f321e491 | 2713 | |
5d4f98a2 YZ |
2714 | ret = 1; |
2715 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2716 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2717 | if (item_size < sizeof(*ei)) { | |
2718 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2719 | goto out; | |
2720 | } | |
2721 | #endif | |
2722 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2723 | |
5d4f98a2 YZ |
2724 | if (item_size != sizeof(*ei) + |
2725 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2726 | goto out; | |
be20aa9d | 2727 | |
5d4f98a2 YZ |
2728 | if (btrfs_extent_generation(leaf, ei) <= |
2729 | btrfs_root_last_snapshot(&root->root_item)) | |
2730 | goto out; | |
2731 | ||
2732 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2733 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2734 | BTRFS_EXTENT_DATA_REF_KEY) | |
2735 | goto out; | |
2736 | ||
2737 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2738 | if (btrfs_extent_refs(leaf, ei) != | |
2739 | btrfs_extent_data_ref_count(leaf, ref) || | |
2740 | btrfs_extent_data_ref_root(leaf, ref) != | |
2741 | root->root_key.objectid || | |
2742 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2743 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2744 | goto out; | |
2745 | ||
2746 | ret = 0; | |
2747 | out: | |
2748 | return ret; | |
2749 | } | |
2750 | ||
2751 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2752 | struct btrfs_root *root, | |
2753 | u64 objectid, u64 offset, u64 bytenr) | |
2754 | { | |
2755 | struct btrfs_path *path; | |
2756 | int ret; | |
2757 | int ret2; | |
2758 | ||
2759 | path = btrfs_alloc_path(); | |
2760 | if (!path) | |
2761 | return -ENOENT; | |
2762 | ||
2763 | do { | |
2764 | ret = check_committed_ref(trans, root, path, objectid, | |
2765 | offset, bytenr); | |
2766 | if (ret && ret != -ENOENT) | |
f321e491 | 2767 | goto out; |
80ff3856 | 2768 | |
5d4f98a2 YZ |
2769 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2770 | offset, bytenr); | |
2771 | } while (ret2 == -EAGAIN); | |
2772 | ||
2773 | if (ret2 && ret2 != -ENOENT) { | |
2774 | ret = ret2; | |
2775 | goto out; | |
f321e491 | 2776 | } |
5d4f98a2 YZ |
2777 | |
2778 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2779 | ret = 0; | |
be20aa9d | 2780 | out: |
80ff3856 | 2781 | btrfs_free_path(path); |
f0486c68 YZ |
2782 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2783 | WARN_ON(ret > 0); | |
f321e491 | 2784 | return ret; |
be20aa9d | 2785 | } |
c5739bba | 2786 | |
5d4f98a2 | 2787 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2788 | struct btrfs_root *root, |
5d4f98a2 | 2789 | struct extent_buffer *buf, |
66d7e7f0 | 2790 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
2791 | { |
2792 | u64 bytenr; | |
5d4f98a2 YZ |
2793 | u64 num_bytes; |
2794 | u64 parent; | |
31840ae1 | 2795 | u64 ref_root; |
31840ae1 | 2796 | u32 nritems; |
31840ae1 ZY |
2797 | struct btrfs_key key; |
2798 | struct btrfs_file_extent_item *fi; | |
2799 | int i; | |
2800 | int level; | |
2801 | int ret = 0; | |
31840ae1 | 2802 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 2803 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
2804 | |
2805 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2806 | nritems = btrfs_header_nritems(buf); |
2807 | level = btrfs_header_level(buf); | |
2808 | ||
5d4f98a2 YZ |
2809 | if (!root->ref_cows && level == 0) |
2810 | return 0; | |
31840ae1 | 2811 | |
5d4f98a2 YZ |
2812 | if (inc) |
2813 | process_func = btrfs_inc_extent_ref; | |
2814 | else | |
2815 | process_func = btrfs_free_extent; | |
31840ae1 | 2816 | |
5d4f98a2 YZ |
2817 | if (full_backref) |
2818 | parent = buf->start; | |
2819 | else | |
2820 | parent = 0; | |
2821 | ||
2822 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2823 | if (level == 0) { |
5d4f98a2 | 2824 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2825 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2826 | continue; | |
5d4f98a2 | 2827 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2828 | struct btrfs_file_extent_item); |
2829 | if (btrfs_file_extent_type(buf, fi) == | |
2830 | BTRFS_FILE_EXTENT_INLINE) | |
2831 | continue; | |
2832 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2833 | if (bytenr == 0) | |
2834 | continue; | |
5d4f98a2 YZ |
2835 | |
2836 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2837 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2838 | ret = process_func(trans, root, bytenr, num_bytes, | |
2839 | parent, ref_root, key.objectid, | |
66d7e7f0 | 2840 | key.offset, for_cow); |
31840ae1 ZY |
2841 | if (ret) |
2842 | goto fail; | |
2843 | } else { | |
5d4f98a2 YZ |
2844 | bytenr = btrfs_node_blockptr(buf, i); |
2845 | num_bytes = btrfs_level_size(root, level - 1); | |
2846 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
2847 | parent, ref_root, level - 1, 0, |
2848 | for_cow); | |
31840ae1 ZY |
2849 | if (ret) |
2850 | goto fail; | |
2851 | } | |
2852 | } | |
2853 | return 0; | |
2854 | fail: | |
5d4f98a2 YZ |
2855 | return ret; |
2856 | } | |
2857 | ||
2858 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2859 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2860 | { |
66d7e7f0 | 2861 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
2862 | } |
2863 | ||
2864 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2865 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2866 | { |
66d7e7f0 | 2867 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
2868 | } |
2869 | ||
9078a3e1 CM |
2870 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2871 | struct btrfs_root *root, | |
2872 | struct btrfs_path *path, | |
2873 | struct btrfs_block_group_cache *cache) | |
2874 | { | |
2875 | int ret; | |
9078a3e1 | 2876 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2877 | unsigned long bi; |
2878 | struct extent_buffer *leaf; | |
9078a3e1 | 2879 | |
9078a3e1 | 2880 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2881 | if (ret < 0) |
2882 | goto fail; | |
79787eaa | 2883 | BUG_ON(ret); /* Corruption */ |
5f39d397 CM |
2884 | |
2885 | leaf = path->nodes[0]; | |
2886 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2887 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2888 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2889 | btrfs_release_path(path); |
54aa1f4d | 2890 | fail: |
79787eaa JM |
2891 | if (ret) { |
2892 | btrfs_abort_transaction(trans, root, ret); | |
9078a3e1 | 2893 | return ret; |
79787eaa | 2894 | } |
9078a3e1 CM |
2895 | return 0; |
2896 | ||
2897 | } | |
2898 | ||
4a8c9a62 YZ |
2899 | static struct btrfs_block_group_cache * |
2900 | next_block_group(struct btrfs_root *root, | |
2901 | struct btrfs_block_group_cache *cache) | |
2902 | { | |
2903 | struct rb_node *node; | |
2904 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2905 | node = rb_next(&cache->cache_node); | |
2906 | btrfs_put_block_group(cache); | |
2907 | if (node) { | |
2908 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2909 | cache_node); | |
11dfe35a | 2910 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2911 | } else |
2912 | cache = NULL; | |
2913 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2914 | return cache; | |
2915 | } | |
2916 | ||
0af3d00b JB |
2917 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2918 | struct btrfs_trans_handle *trans, | |
2919 | struct btrfs_path *path) | |
2920 | { | |
2921 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2922 | struct inode *inode = NULL; | |
2923 | u64 alloc_hint = 0; | |
2b20982e | 2924 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2925 | int num_pages = 0; |
2926 | int retries = 0; | |
2927 | int ret = 0; | |
2928 | ||
2929 | /* | |
2930 | * If this block group is smaller than 100 megs don't bother caching the | |
2931 | * block group. | |
2932 | */ | |
2933 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2934 | spin_lock(&block_group->lock); | |
2935 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2936 | spin_unlock(&block_group->lock); | |
2937 | return 0; | |
2938 | } | |
2939 | ||
2940 | again: | |
2941 | inode = lookup_free_space_inode(root, block_group, path); | |
2942 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2943 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2944 | btrfs_release_path(path); |
0af3d00b JB |
2945 | goto out; |
2946 | } | |
2947 | ||
2948 | if (IS_ERR(inode)) { | |
2949 | BUG_ON(retries); | |
2950 | retries++; | |
2951 | ||
2952 | if (block_group->ro) | |
2953 | goto out_free; | |
2954 | ||
2955 | ret = create_free_space_inode(root, trans, block_group, path); | |
2956 | if (ret) | |
2957 | goto out_free; | |
2958 | goto again; | |
2959 | } | |
2960 | ||
5b0e95bf JB |
2961 | /* We've already setup this transaction, go ahead and exit */ |
2962 | if (block_group->cache_generation == trans->transid && | |
2963 | i_size_read(inode)) { | |
2964 | dcs = BTRFS_DC_SETUP; | |
2965 | goto out_put; | |
2966 | } | |
2967 | ||
0af3d00b JB |
2968 | /* |
2969 | * We want to set the generation to 0, that way if anything goes wrong | |
2970 | * from here on out we know not to trust this cache when we load up next | |
2971 | * time. | |
2972 | */ | |
2973 | BTRFS_I(inode)->generation = 0; | |
2974 | ret = btrfs_update_inode(trans, root, inode); | |
2975 | WARN_ON(ret); | |
2976 | ||
2977 | if (i_size_read(inode) > 0) { | |
2978 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2979 | inode); | |
2980 | if (ret) | |
2981 | goto out_put; | |
2982 | } | |
2983 | ||
2984 | spin_lock(&block_group->lock); | |
cf7c1ef6 LB |
2985 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
2986 | !btrfs_test_opt(root, SPACE_CACHE)) { | |
2987 | /* | |
2988 | * don't bother trying to write stuff out _if_ | |
2989 | * a) we're not cached, | |
2990 | * b) we're with nospace_cache mount option. | |
2991 | */ | |
2b20982e | 2992 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
2993 | spin_unlock(&block_group->lock); |
2994 | goto out_put; | |
2995 | } | |
2996 | spin_unlock(&block_group->lock); | |
2997 | ||
6fc823b1 JB |
2998 | /* |
2999 | * Try to preallocate enough space based on how big the block group is. | |
3000 | * Keep in mind this has to include any pinned space which could end up | |
3001 | * taking up quite a bit since it's not folded into the other space | |
3002 | * cache. | |
3003 | */ | |
3004 | num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024); | |
0af3d00b JB |
3005 | if (!num_pages) |
3006 | num_pages = 1; | |
3007 | ||
0af3d00b JB |
3008 | num_pages *= 16; |
3009 | num_pages *= PAGE_CACHE_SIZE; | |
3010 | ||
3011 | ret = btrfs_check_data_free_space(inode, num_pages); | |
3012 | if (ret) | |
3013 | goto out_put; | |
3014 | ||
3015 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3016 | num_pages, num_pages, | |
3017 | &alloc_hint); | |
2b20982e JB |
3018 | if (!ret) |
3019 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 3020 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 3021 | |
0af3d00b JB |
3022 | out_put: |
3023 | iput(inode); | |
3024 | out_free: | |
b3b4aa74 | 3025 | btrfs_release_path(path); |
0af3d00b JB |
3026 | out: |
3027 | spin_lock(&block_group->lock); | |
e65cbb94 | 3028 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3029 | block_group->cache_generation = trans->transid; |
2b20982e | 3030 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3031 | spin_unlock(&block_group->lock); |
3032 | ||
3033 | return ret; | |
3034 | } | |
3035 | ||
96b5179d CM |
3036 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
3037 | struct btrfs_root *root) | |
9078a3e1 | 3038 | { |
4a8c9a62 | 3039 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 3040 | int err = 0; |
9078a3e1 | 3041 | struct btrfs_path *path; |
96b5179d | 3042 | u64 last = 0; |
9078a3e1 CM |
3043 | |
3044 | path = btrfs_alloc_path(); | |
3045 | if (!path) | |
3046 | return -ENOMEM; | |
3047 | ||
0af3d00b JB |
3048 | again: |
3049 | while (1) { | |
3050 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3051 | while (cache) { | |
3052 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3053 | break; | |
3054 | cache = next_block_group(root, cache); | |
3055 | } | |
3056 | if (!cache) { | |
3057 | if (last == 0) | |
3058 | break; | |
3059 | last = 0; | |
3060 | continue; | |
3061 | } | |
3062 | err = cache_save_setup(cache, trans, path); | |
3063 | last = cache->key.objectid + cache->key.offset; | |
3064 | btrfs_put_block_group(cache); | |
3065 | } | |
3066 | ||
d397712b | 3067 | while (1) { |
4a8c9a62 YZ |
3068 | if (last == 0) { |
3069 | err = btrfs_run_delayed_refs(trans, root, | |
3070 | (unsigned long)-1); | |
79787eaa JM |
3071 | if (err) /* File system offline */ |
3072 | goto out; | |
0f9dd46c | 3073 | } |
54aa1f4d | 3074 | |
4a8c9a62 YZ |
3075 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
3076 | while (cache) { | |
0af3d00b JB |
3077 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
3078 | btrfs_put_block_group(cache); | |
3079 | goto again; | |
3080 | } | |
3081 | ||
4a8c9a62 YZ |
3082 | if (cache->dirty) |
3083 | break; | |
3084 | cache = next_block_group(root, cache); | |
3085 | } | |
3086 | if (!cache) { | |
3087 | if (last == 0) | |
3088 | break; | |
3089 | last = 0; | |
3090 | continue; | |
3091 | } | |
0f9dd46c | 3092 | |
0cb59c99 JB |
3093 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
3094 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 3095 | cache->dirty = 0; |
4a8c9a62 | 3096 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 3097 | |
4a8c9a62 | 3098 | err = write_one_cache_group(trans, root, path, cache); |
79787eaa JM |
3099 | if (err) /* File system offline */ |
3100 | goto out; | |
3101 | ||
4a8c9a62 | 3102 | btrfs_put_block_group(cache); |
9078a3e1 | 3103 | } |
4a8c9a62 | 3104 | |
0cb59c99 JB |
3105 | while (1) { |
3106 | /* | |
3107 | * I don't think this is needed since we're just marking our | |
3108 | * preallocated extent as written, but just in case it can't | |
3109 | * hurt. | |
3110 | */ | |
3111 | if (last == 0) { | |
3112 | err = btrfs_run_delayed_refs(trans, root, | |
3113 | (unsigned long)-1); | |
79787eaa JM |
3114 | if (err) /* File system offline */ |
3115 | goto out; | |
0cb59c99 JB |
3116 | } |
3117 | ||
3118 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3119 | while (cache) { | |
3120 | /* | |
3121 | * Really this shouldn't happen, but it could if we | |
3122 | * couldn't write the entire preallocated extent and | |
3123 | * splitting the extent resulted in a new block. | |
3124 | */ | |
3125 | if (cache->dirty) { | |
3126 | btrfs_put_block_group(cache); | |
3127 | goto again; | |
3128 | } | |
3129 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3130 | break; | |
3131 | cache = next_block_group(root, cache); | |
3132 | } | |
3133 | if (!cache) { | |
3134 | if (last == 0) | |
3135 | break; | |
3136 | last = 0; | |
3137 | continue; | |
3138 | } | |
3139 | ||
79787eaa | 3140 | err = btrfs_write_out_cache(root, trans, cache, path); |
0cb59c99 JB |
3141 | |
3142 | /* | |
3143 | * If we didn't have an error then the cache state is still | |
3144 | * NEED_WRITE, so we can set it to WRITTEN. | |
3145 | */ | |
79787eaa | 3146 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
0cb59c99 JB |
3147 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
3148 | last = cache->key.objectid + cache->key.offset; | |
3149 | btrfs_put_block_group(cache); | |
3150 | } | |
79787eaa | 3151 | out: |
0cb59c99 | 3152 | |
9078a3e1 | 3153 | btrfs_free_path(path); |
79787eaa | 3154 | return err; |
9078a3e1 CM |
3155 | } |
3156 | ||
d2fb3437 YZ |
3157 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3158 | { | |
3159 | struct btrfs_block_group_cache *block_group; | |
3160 | int readonly = 0; | |
3161 | ||
3162 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3163 | if (!block_group || block_group->ro) | |
3164 | readonly = 1; | |
3165 | if (block_group) | |
fa9c0d79 | 3166 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3167 | return readonly; |
3168 | } | |
3169 | ||
593060d7 CM |
3170 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3171 | u64 total_bytes, u64 bytes_used, | |
3172 | struct btrfs_space_info **space_info) | |
3173 | { | |
3174 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3175 | int i; |
3176 | int factor; | |
3177 | ||
3178 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3179 | BTRFS_BLOCK_GROUP_RAID10)) | |
3180 | factor = 2; | |
3181 | else | |
3182 | factor = 1; | |
593060d7 CM |
3183 | |
3184 | found = __find_space_info(info, flags); | |
3185 | if (found) { | |
25179201 | 3186 | spin_lock(&found->lock); |
593060d7 | 3187 | found->total_bytes += total_bytes; |
89a55897 | 3188 | found->disk_total += total_bytes * factor; |
593060d7 | 3189 | found->bytes_used += bytes_used; |
b742bb82 | 3190 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3191 | found->full = 0; |
25179201 | 3192 | spin_unlock(&found->lock); |
593060d7 CM |
3193 | *space_info = found; |
3194 | return 0; | |
3195 | } | |
c146afad | 3196 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3197 | if (!found) |
3198 | return -ENOMEM; | |
3199 | ||
b742bb82 YZ |
3200 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3201 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3202 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3203 | spin_lock_init(&found->lock); |
52ba6929 | 3204 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3205 | found->total_bytes = total_bytes; |
89a55897 | 3206 | found->disk_total = total_bytes * factor; |
593060d7 | 3207 | found->bytes_used = bytes_used; |
b742bb82 | 3208 | found->disk_used = bytes_used * factor; |
593060d7 | 3209 | found->bytes_pinned = 0; |
e8569813 | 3210 | found->bytes_reserved = 0; |
c146afad | 3211 | found->bytes_readonly = 0; |
f0486c68 | 3212 | found->bytes_may_use = 0; |
593060d7 | 3213 | found->full = 0; |
0e4f8f88 | 3214 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3215 | found->chunk_alloc = 0; |
fdb5effd JB |
3216 | found->flush = 0; |
3217 | init_waitqueue_head(&found->wait); | |
593060d7 | 3218 | *space_info = found; |
4184ea7f | 3219 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3220 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3221 | info->data_sinfo = found; | |
593060d7 CM |
3222 | return 0; |
3223 | } | |
3224 | ||
8790d502 CM |
3225 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3226 | { | |
899c81ea ID |
3227 | u64 extra_flags = chunk_to_extended(flags) & |
3228 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3229 | |
de98ced9 | 3230 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3231 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3232 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3233 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3234 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3235 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3236 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3237 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3238 | } |
593060d7 | 3239 | |
fc67c450 ID |
3240 | /* |
3241 | * returns target flags in extended format or 0 if restripe for this | |
3242 | * chunk_type is not in progress | |
c6664b42 ID |
3243 | * |
3244 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3245 | */ |
3246 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3247 | { | |
3248 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3249 | u64 target = 0; | |
3250 | ||
fc67c450 ID |
3251 | if (!bctl) |
3252 | return 0; | |
3253 | ||
3254 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3255 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3256 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3257 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3258 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3259 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3260 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3261 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3262 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3263 | } | |
3264 | ||
3265 | return target; | |
3266 | } | |
3267 | ||
a46d11a8 ID |
3268 | /* |
3269 | * @flags: available profiles in extended format (see ctree.h) | |
3270 | * | |
e4d8ec0f ID |
3271 | * Returns reduced profile in chunk format. If profile changing is in |
3272 | * progress (either running or paused) picks the target profile (if it's | |
3273 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3274 | */ |
2b82032c | 3275 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3276 | { |
cd02dca5 CM |
3277 | /* |
3278 | * we add in the count of missing devices because we want | |
3279 | * to make sure that any RAID levels on a degraded FS | |
3280 | * continue to be honored. | |
3281 | */ | |
3282 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3283 | root->fs_info->fs_devices->missing_devices; | |
fc67c450 | 3284 | u64 target; |
a061fc8d | 3285 | |
fc67c450 ID |
3286 | /* |
3287 | * see if restripe for this chunk_type is in progress, if so | |
3288 | * try to reduce to the target profile | |
3289 | */ | |
e4d8ec0f | 3290 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3291 | target = get_restripe_target(root->fs_info, flags); |
3292 | if (target) { | |
3293 | /* pick target profile only if it's already available */ | |
3294 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3295 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3296 | return extended_to_chunk(target); |
e4d8ec0f ID |
3297 | } |
3298 | } | |
3299 | spin_unlock(&root->fs_info->balance_lock); | |
3300 | ||
a061fc8d CM |
3301 | if (num_devices == 1) |
3302 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
3303 | if (num_devices < 4) | |
3304 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3305 | ||
ec44a35c CM |
3306 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
3307 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 3308 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 3309 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 3310 | } |
ec44a35c CM |
3311 | |
3312 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 3313 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 3314 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 3315 | } |
ec44a35c CM |
3316 | |
3317 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
3318 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
3319 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
a46d11a8 | 3320 | (flags & BTRFS_BLOCK_GROUP_DUP))) { |
ec44a35c | 3321 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; |
a46d11a8 ID |
3322 | } |
3323 | ||
899c81ea | 3324 | return extended_to_chunk(flags); |
ec44a35c CM |
3325 | } |
3326 | ||
b742bb82 | 3327 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3328 | { |
de98ced9 MX |
3329 | unsigned seq; |
3330 | ||
3331 | do { | |
3332 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
3333 | ||
3334 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3335 | flags |= root->fs_info->avail_data_alloc_bits; | |
3336 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3337 | flags |= root->fs_info->avail_system_alloc_bits; | |
3338 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3339 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
3340 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
6fef8df1 | 3341 | |
b742bb82 | 3342 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3343 | } |
3344 | ||
6d07bcec | 3345 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3346 | { |
b742bb82 | 3347 | u64 flags; |
9ed74f2d | 3348 | |
b742bb82 YZ |
3349 | if (data) |
3350 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3351 | else if (root == root->fs_info->chunk_root) | |
3352 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3353 | else |
b742bb82 | 3354 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3355 | |
b742bb82 | 3356 | return get_alloc_profile(root, flags); |
6a63209f | 3357 | } |
9ed74f2d | 3358 | |
6a63209f | 3359 | /* |
6a63209f JB |
3360 | * This will check the space that the inode allocates from to make sure we have |
3361 | * enough space for bytes. | |
6a63209f | 3362 | */ |
0ca1f7ce | 3363 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3364 | { |
6a63209f | 3365 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3366 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 3367 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 3368 | u64 used; |
0af3d00b | 3369 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3370 | |
6a63209f JB |
3371 | /* make sure bytes are sectorsize aligned */ |
3372 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3373 | |
82d5902d LZ |
3374 | if (root == root->fs_info->tree_root || |
3375 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3376 | alloc_chunk = 0; |
3377 | committed = 1; | |
3378 | } | |
3379 | ||
b4d7c3c9 | 3380 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
3381 | if (!data_sinfo) |
3382 | goto alloc; | |
9ed74f2d | 3383 | |
6a63209f JB |
3384 | again: |
3385 | /* make sure we have enough space to handle the data first */ | |
3386 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3387 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3388 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3389 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3390 | |
3391 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3392 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3393 | |
6a63209f JB |
3394 | /* |
3395 | * if we don't have enough free bytes in this space then we need | |
3396 | * to alloc a new chunk. | |
3397 | */ | |
0af3d00b | 3398 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3399 | u64 alloc_target; |
9ed74f2d | 3400 | |
0e4f8f88 | 3401 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3402 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3403 | alloc: |
6a63209f | 3404 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3405 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3406 | if (IS_ERR(trans)) |
3407 | return PTR_ERR(trans); | |
9ed74f2d | 3408 | |
6a63209f | 3409 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
3410 | alloc_target, |
3411 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3412 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3413 | if (ret < 0) { |
3414 | if (ret != -ENOSPC) | |
3415 | return ret; | |
3416 | else | |
3417 | goto commit_trans; | |
3418 | } | |
9ed74f2d | 3419 | |
b4d7c3c9 LZ |
3420 | if (!data_sinfo) |
3421 | data_sinfo = fs_info->data_sinfo; | |
3422 | ||
6a63209f JB |
3423 | goto again; |
3424 | } | |
f2bb8f5c JB |
3425 | |
3426 | /* | |
3427 | * If we have less pinned bytes than we want to allocate then | |
3428 | * don't bother committing the transaction, it won't help us. | |
3429 | */ | |
3430 | if (data_sinfo->bytes_pinned < bytes) | |
3431 | committed = 1; | |
6a63209f | 3432 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3433 | |
4e06bdd6 | 3434 | /* commit the current transaction and try again */ |
d52a5b5f | 3435 | commit_trans: |
a4abeea4 JB |
3436 | if (!committed && |
3437 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3438 | committed = 1; |
7a7eaa40 | 3439 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3440 | if (IS_ERR(trans)) |
3441 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3442 | ret = btrfs_commit_transaction(trans, root); |
3443 | if (ret) | |
3444 | return ret; | |
3445 | goto again; | |
3446 | } | |
9ed74f2d | 3447 | |
6a63209f JB |
3448 | return -ENOSPC; |
3449 | } | |
3450 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3451 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3452 | data_sinfo->flags, bytes, 1); |
6a63209f | 3453 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3454 | |
9ed74f2d | 3455 | return 0; |
9ed74f2d | 3456 | } |
6a63209f | 3457 | |
6a63209f | 3458 | /* |
fb25e914 | 3459 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3460 | */ |
0ca1f7ce | 3461 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3462 | { |
0ca1f7ce | 3463 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3464 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3465 | |
6a63209f JB |
3466 | /* make sure bytes are sectorsize aligned */ |
3467 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3468 | |
b4d7c3c9 | 3469 | data_sinfo = root->fs_info->data_sinfo; |
6a63209f JB |
3470 | spin_lock(&data_sinfo->lock); |
3471 | data_sinfo->bytes_may_use -= bytes; | |
8c2a3ca2 | 3472 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3473 | data_sinfo->flags, bytes, 0); |
6a63209f | 3474 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3475 | } |
3476 | ||
97e728d4 | 3477 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3478 | { |
97e728d4 JB |
3479 | struct list_head *head = &info->space_info; |
3480 | struct btrfs_space_info *found; | |
e3ccfa98 | 3481 | |
97e728d4 JB |
3482 | rcu_read_lock(); |
3483 | list_for_each_entry_rcu(found, head, list) { | |
3484 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3485 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3486 | } |
97e728d4 | 3487 | rcu_read_unlock(); |
e3ccfa98 JB |
3488 | } |
3489 | ||
e5bc2458 | 3490 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 3491 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 3492 | { |
fb25e914 | 3493 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3494 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3495 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3496 | u64 thresh; |
e3ccfa98 | 3497 | |
0e4f8f88 CM |
3498 | if (force == CHUNK_ALLOC_FORCE) |
3499 | return 1; | |
3500 | ||
fb25e914 JB |
3501 | /* |
3502 | * We need to take into account the global rsv because for all intents | |
3503 | * and purposes it's used space. Don't worry about locking the | |
3504 | * global_rsv, it doesn't change except when the transaction commits. | |
3505 | */ | |
54338b5c JB |
3506 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3507 | num_allocated += global_rsv->size; | |
fb25e914 | 3508 | |
0e4f8f88 CM |
3509 | /* |
3510 | * in limited mode, we want to have some free space up to | |
3511 | * about 1% of the FS size. | |
3512 | */ | |
3513 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3514 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3515 | thresh = max_t(u64, 64 * 1024 * 1024, |
3516 | div_factor_fine(thresh, 1)); | |
3517 | ||
3518 | if (num_bytes - num_allocated < thresh) | |
3519 | return 1; | |
3520 | } | |
0e4f8f88 | 3521 | |
698d0082 | 3522 | if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8)) |
14ed0ca6 | 3523 | return 0; |
424499db | 3524 | return 1; |
32c00aff JB |
3525 | } |
3526 | ||
15d1ff81 LB |
3527 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
3528 | { | |
3529 | u64 num_dev; | |
3530 | ||
3531 | if (type & BTRFS_BLOCK_GROUP_RAID10 || | |
3532 | type & BTRFS_BLOCK_GROUP_RAID0) | |
3533 | num_dev = root->fs_info->fs_devices->rw_devices; | |
3534 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
3535 | num_dev = 2; | |
3536 | else | |
3537 | num_dev = 1; /* DUP or single */ | |
3538 | ||
3539 | /* metadata for updaing devices and chunk tree */ | |
3540 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); | |
3541 | } | |
3542 | ||
3543 | static void check_system_chunk(struct btrfs_trans_handle *trans, | |
3544 | struct btrfs_root *root, u64 type) | |
3545 | { | |
3546 | struct btrfs_space_info *info; | |
3547 | u64 left; | |
3548 | u64 thresh; | |
3549 | ||
3550 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3551 | spin_lock(&info->lock); | |
3552 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
3553 | info->bytes_reserved - info->bytes_readonly; | |
3554 | spin_unlock(&info->lock); | |
3555 | ||
3556 | thresh = get_system_chunk_thresh(root, type); | |
3557 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
3558 | printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n", | |
3559 | left, thresh, type); | |
3560 | dump_space_info(info, 0, 0); | |
3561 | } | |
3562 | ||
3563 | if (left < thresh) { | |
3564 | u64 flags; | |
3565 | ||
3566 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
3567 | btrfs_alloc_chunk(trans, root, flags); | |
3568 | } | |
3569 | } | |
3570 | ||
6324fbf3 | 3571 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 3572 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 3573 | { |
6324fbf3 | 3574 | struct btrfs_space_info *space_info; |
97e728d4 | 3575 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3576 | int wait_for_alloc = 0; |
9ed74f2d | 3577 | int ret = 0; |
9ed74f2d | 3578 | |
c6b305a8 JB |
3579 | /* Don't re-enter if we're already allocating a chunk */ |
3580 | if (trans->allocating_chunk) | |
3581 | return -ENOSPC; | |
3582 | ||
6324fbf3 | 3583 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3584 | if (!space_info) { |
3585 | ret = update_space_info(extent_root->fs_info, flags, | |
3586 | 0, 0, &space_info); | |
79787eaa | 3587 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 3588 | } |
79787eaa | 3589 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 3590 | |
6d74119f | 3591 | again: |
25179201 | 3592 | spin_lock(&space_info->lock); |
9e622d6b | 3593 | if (force < space_info->force_alloc) |
0e4f8f88 | 3594 | force = space_info->force_alloc; |
25179201 JB |
3595 | if (space_info->full) { |
3596 | spin_unlock(&space_info->lock); | |
6d74119f | 3597 | return 0; |
9ed74f2d JB |
3598 | } |
3599 | ||
698d0082 | 3600 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 3601 | spin_unlock(&space_info->lock); |
6d74119f JB |
3602 | return 0; |
3603 | } else if (space_info->chunk_alloc) { | |
3604 | wait_for_alloc = 1; | |
3605 | } else { | |
3606 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3607 | } |
0e4f8f88 | 3608 | |
25179201 | 3609 | spin_unlock(&space_info->lock); |
9ed74f2d | 3610 | |
6d74119f JB |
3611 | mutex_lock(&fs_info->chunk_mutex); |
3612 | ||
3613 | /* | |
3614 | * The chunk_mutex is held throughout the entirety of a chunk | |
3615 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3616 | * other guy is done and we need to recheck and see if we should | |
3617 | * allocate. | |
3618 | */ | |
3619 | if (wait_for_alloc) { | |
3620 | mutex_unlock(&fs_info->chunk_mutex); | |
3621 | wait_for_alloc = 0; | |
3622 | goto again; | |
3623 | } | |
3624 | ||
c6b305a8 JB |
3625 | trans->allocating_chunk = true; |
3626 | ||
67377734 JB |
3627 | /* |
3628 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3629 | * allocating mixed chunks instead of individual chunks. | |
3630 | */ | |
3631 | if (btrfs_mixed_space_info(space_info)) | |
3632 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3633 | ||
97e728d4 JB |
3634 | /* |
3635 | * if we're doing a data chunk, go ahead and make sure that | |
3636 | * we keep a reasonable number of metadata chunks allocated in the | |
3637 | * FS as well. | |
3638 | */ | |
9ed74f2d | 3639 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3640 | fs_info->data_chunk_allocations++; |
3641 | if (!(fs_info->data_chunk_allocations % | |
3642 | fs_info->metadata_ratio)) | |
3643 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3644 | } |
3645 | ||
15d1ff81 LB |
3646 | /* |
3647 | * Check if we have enough space in SYSTEM chunk because we may need | |
3648 | * to update devices. | |
3649 | */ | |
3650 | check_system_chunk(trans, extent_root, flags); | |
3651 | ||
2b82032c | 3652 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 3653 | trans->allocating_chunk = false; |
92b8e897 MF |
3654 | if (ret < 0 && ret != -ENOSPC) |
3655 | goto out; | |
3656 | ||
9ed74f2d | 3657 | spin_lock(&space_info->lock); |
9ed74f2d | 3658 | if (ret) |
6324fbf3 | 3659 | space_info->full = 1; |
424499db YZ |
3660 | else |
3661 | ret = 1; | |
6d74119f | 3662 | |
0e4f8f88 | 3663 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3664 | space_info->chunk_alloc = 0; |
9ed74f2d | 3665 | spin_unlock(&space_info->lock); |
92b8e897 | 3666 | out: |
a25c75d5 | 3667 | mutex_unlock(&fs_info->chunk_mutex); |
0f9dd46c | 3668 | return ret; |
6324fbf3 | 3669 | } |
9ed74f2d | 3670 | |
a80c8dcf JB |
3671 | static int can_overcommit(struct btrfs_root *root, |
3672 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 3673 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf JB |
3674 | { |
3675 | u64 profile = btrfs_get_alloc_profile(root, 0); | |
3676 | u64 avail; | |
3677 | u64 used; | |
3678 | ||
3679 | used = space_info->bytes_used + space_info->bytes_reserved + | |
3680 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3681 | space_info->bytes_may_use; | |
3682 | ||
3683 | spin_lock(&root->fs_info->free_chunk_lock); | |
3684 | avail = root->fs_info->free_chunk_space; | |
3685 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3686 | ||
3687 | /* | |
3688 | * If we have dup, raid1 or raid10 then only half of the free | |
3689 | * space is actually useable. | |
3690 | */ | |
3691 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3692 | BTRFS_BLOCK_GROUP_RAID1 | | |
3693 | BTRFS_BLOCK_GROUP_RAID10)) | |
3694 | avail >>= 1; | |
3695 | ||
3696 | /* | |
561c294d MX |
3697 | * If we aren't flushing all things, let us overcommit up to |
3698 | * 1/2th of the space. If we can flush, don't let us overcommit | |
3699 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 3700 | */ |
08e007d2 | 3701 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
a80c8dcf JB |
3702 | avail >>= 3; |
3703 | else | |
3704 | avail >>= 1; | |
3705 | ||
3706 | if (used + bytes < space_info->total_bytes + avail) | |
3707 | return 1; | |
3708 | return 0; | |
3709 | } | |
3710 | ||
da633a42 MX |
3711 | static inline int writeback_inodes_sb_nr_if_idle_safe(struct super_block *sb, |
3712 | unsigned long nr_pages, | |
3713 | enum wb_reason reason) | |
1135d6df | 3714 | { |
da633a42 MX |
3715 | /* the flusher is dealing with the dirty inodes now. */ |
3716 | if (writeback_in_progress(sb->s_bdi)) | |
3717 | return 1; | |
3718 | ||
3719 | if (down_read_trylock(&sb->s_umount)) { | |
1135d6df JB |
3720 | writeback_inodes_sb_nr(sb, nr_pages, reason); |
3721 | up_read(&sb->s_umount); | |
3722 | return 1; | |
3723 | } | |
3724 | ||
3725 | return 0; | |
3726 | } | |
3727 | ||
da633a42 MX |
3728 | void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
3729 | unsigned long nr_pages) | |
3730 | { | |
3731 | struct super_block *sb = root->fs_info->sb; | |
3732 | int started; | |
3733 | ||
3734 | /* If we can not start writeback, just sync all the delalloc file. */ | |
3735 | started = writeback_inodes_sb_nr_if_idle_safe(sb, nr_pages, | |
3736 | WB_REASON_FS_FREE_SPACE); | |
3737 | if (!started) { | |
3738 | /* | |
3739 | * We needn't worry the filesystem going from r/w to r/o though | |
3740 | * we don't acquire ->s_umount mutex, because the filesystem | |
3741 | * should guarantee the delalloc inodes list be empty after | |
3742 | * the filesystem is readonly(all dirty pages are written to | |
3743 | * the disk). | |
3744 | */ | |
3745 | btrfs_start_delalloc_inodes(root, 0); | |
3746 | btrfs_wait_ordered_extents(root, 0); | |
3747 | } | |
3748 | } | |
3749 | ||
9ed74f2d | 3750 | /* |
5da9d01b | 3751 | * shrink metadata reservation for delalloc |
9ed74f2d | 3752 | */ |
f4c738c2 JB |
3753 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
3754 | bool wait_ordered) | |
5da9d01b | 3755 | { |
0ca1f7ce | 3756 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3757 | struct btrfs_space_info *space_info; |
663350ac | 3758 | struct btrfs_trans_handle *trans; |
f4c738c2 | 3759 | u64 delalloc_bytes; |
5da9d01b | 3760 | u64 max_reclaim; |
b1953bce | 3761 | long time_left; |
877da174 | 3762 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3763 | int loops = 0; |
08e007d2 | 3764 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 3765 | |
663350ac | 3766 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 3767 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3768 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3769 | |
3770 | smp_mb(); | |
963d678b MX |
3771 | delalloc_bytes = percpu_counter_sum_positive( |
3772 | &root->fs_info->delalloc_bytes); | |
f4c738c2 | 3773 | if (delalloc_bytes == 0) { |
fdb5effd | 3774 | if (trans) |
f4c738c2 | 3775 | return; |
6bbe3a9c | 3776 | btrfs_wait_ordered_extents(root, 0); |
f4c738c2 | 3777 | return; |
fdb5effd JB |
3778 | } |
3779 | ||
f4c738c2 JB |
3780 | while (delalloc_bytes && loops < 3) { |
3781 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
3782 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; | |
da633a42 | 3783 | btrfs_writeback_inodes_sb_nr(root, nr_pages); |
dea31f52 JB |
3784 | /* |
3785 | * We need to wait for the async pages to actually start before | |
3786 | * we do anything. | |
3787 | */ | |
3788 | wait_event(root->fs_info->async_submit_wait, | |
3789 | !atomic_read(&root->fs_info->async_delalloc_pages)); | |
3790 | ||
08e007d2 MX |
3791 | if (!trans) |
3792 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
3793 | else | |
3794 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 3795 | spin_lock(&space_info->lock); |
08e007d2 | 3796 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
3797 | spin_unlock(&space_info->lock); |
3798 | break; | |
3799 | } | |
0019f10d | 3800 | spin_unlock(&space_info->lock); |
5da9d01b | 3801 | |
36e39c40 | 3802 | loops++; |
f104d044 | 3803 | if (wait_ordered && !trans) { |
6bbe3a9c | 3804 | btrfs_wait_ordered_extents(root, 0); |
f104d044 | 3805 | } else { |
f4c738c2 | 3806 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
3807 | if (time_left) |
3808 | break; | |
3809 | } | |
f4c738c2 | 3810 | smp_mb(); |
963d678b MX |
3811 | delalloc_bytes = percpu_counter_sum_positive( |
3812 | &root->fs_info->delalloc_bytes); | |
5da9d01b | 3813 | } |
5da9d01b YZ |
3814 | } |
3815 | ||
663350ac JB |
3816 | /** |
3817 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
3818 | * @root - the root we're allocating for | |
3819 | * @bytes - the number of bytes we want to reserve | |
3820 | * @force - force the commit | |
8bb8ab2e | 3821 | * |
663350ac JB |
3822 | * This will check to make sure that committing the transaction will actually |
3823 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
3824 | * will return -ENOSPC. | |
8bb8ab2e | 3825 | */ |
663350ac JB |
3826 | static int may_commit_transaction(struct btrfs_root *root, |
3827 | struct btrfs_space_info *space_info, | |
3828 | u64 bytes, int force) | |
3829 | { | |
3830 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
3831 | struct btrfs_trans_handle *trans; | |
3832 | ||
3833 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
3834 | if (trans) | |
3835 | return -EAGAIN; | |
3836 | ||
3837 | if (force) | |
3838 | goto commit; | |
3839 | ||
3840 | /* See if there is enough pinned space to make this reservation */ | |
3841 | spin_lock(&space_info->lock); | |
3842 | if (space_info->bytes_pinned >= bytes) { | |
3843 | spin_unlock(&space_info->lock); | |
3844 | goto commit; | |
3845 | } | |
3846 | spin_unlock(&space_info->lock); | |
3847 | ||
3848 | /* | |
3849 | * See if there is some space in the delayed insertion reservation for | |
3850 | * this reservation. | |
3851 | */ | |
3852 | if (space_info != delayed_rsv->space_info) | |
3853 | return -ENOSPC; | |
3854 | ||
d9b0218f | 3855 | spin_lock(&space_info->lock); |
663350ac | 3856 | spin_lock(&delayed_rsv->lock); |
d9b0218f | 3857 | if (space_info->bytes_pinned + delayed_rsv->size < bytes) { |
663350ac | 3858 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 3859 | spin_unlock(&space_info->lock); |
663350ac JB |
3860 | return -ENOSPC; |
3861 | } | |
3862 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 3863 | spin_unlock(&space_info->lock); |
663350ac JB |
3864 | |
3865 | commit: | |
3866 | trans = btrfs_join_transaction(root); | |
3867 | if (IS_ERR(trans)) | |
3868 | return -ENOSPC; | |
3869 | ||
3870 | return btrfs_commit_transaction(trans, root); | |
3871 | } | |
3872 | ||
96c3f433 | 3873 | enum flush_state { |
67b0fd63 JB |
3874 | FLUSH_DELAYED_ITEMS_NR = 1, |
3875 | FLUSH_DELAYED_ITEMS = 2, | |
3876 | FLUSH_DELALLOC = 3, | |
3877 | FLUSH_DELALLOC_WAIT = 4, | |
ea658bad JB |
3878 | ALLOC_CHUNK = 5, |
3879 | COMMIT_TRANS = 6, | |
96c3f433 JB |
3880 | }; |
3881 | ||
3882 | static int flush_space(struct btrfs_root *root, | |
3883 | struct btrfs_space_info *space_info, u64 num_bytes, | |
3884 | u64 orig_bytes, int state) | |
3885 | { | |
3886 | struct btrfs_trans_handle *trans; | |
3887 | int nr; | |
f4c738c2 | 3888 | int ret = 0; |
96c3f433 JB |
3889 | |
3890 | switch (state) { | |
96c3f433 JB |
3891 | case FLUSH_DELAYED_ITEMS_NR: |
3892 | case FLUSH_DELAYED_ITEMS: | |
3893 | if (state == FLUSH_DELAYED_ITEMS_NR) { | |
3894 | u64 bytes = btrfs_calc_trans_metadata_size(root, 1); | |
3895 | ||
3896 | nr = (int)div64_u64(num_bytes, bytes); | |
3897 | if (!nr) | |
3898 | nr = 1; | |
3899 | nr *= 2; | |
3900 | } else { | |
3901 | nr = -1; | |
3902 | } | |
3903 | trans = btrfs_join_transaction(root); | |
3904 | if (IS_ERR(trans)) { | |
3905 | ret = PTR_ERR(trans); | |
3906 | break; | |
3907 | } | |
3908 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
3909 | btrfs_end_transaction(trans, root); | |
3910 | break; | |
67b0fd63 JB |
3911 | case FLUSH_DELALLOC: |
3912 | case FLUSH_DELALLOC_WAIT: | |
3913 | shrink_delalloc(root, num_bytes, orig_bytes, | |
3914 | state == FLUSH_DELALLOC_WAIT); | |
3915 | break; | |
ea658bad JB |
3916 | case ALLOC_CHUNK: |
3917 | trans = btrfs_join_transaction(root); | |
3918 | if (IS_ERR(trans)) { | |
3919 | ret = PTR_ERR(trans); | |
3920 | break; | |
3921 | } | |
3922 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
3923 | btrfs_get_alloc_profile(root, 0), |
3924 | CHUNK_ALLOC_NO_FORCE); | |
3925 | btrfs_end_transaction(trans, root); | |
3926 | if (ret == -ENOSPC) | |
3927 | ret = 0; | |
3928 | break; | |
96c3f433 JB |
3929 | case COMMIT_TRANS: |
3930 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
3931 | break; | |
3932 | default: | |
3933 | ret = -ENOSPC; | |
3934 | break; | |
3935 | } | |
3936 | ||
3937 | return ret; | |
3938 | } | |
4a92b1b8 JB |
3939 | /** |
3940 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
3941 | * @root - the root we're allocating for | |
3942 | * @block_rsv - the block_rsv we're allocating for | |
3943 | * @orig_bytes - the number of bytes we want | |
3944 | * @flush - wether or not we can flush to make our reservation | |
8bb8ab2e | 3945 | * |
4a92b1b8 JB |
3946 | * This will reserve orgi_bytes number of bytes from the space info associated |
3947 | * with the block_rsv. If there is not enough space it will make an attempt to | |
3948 | * flush out space to make room. It will do this by flushing delalloc if | |
3949 | * possible or committing the transaction. If flush is 0 then no attempts to | |
3950 | * regain reservations will be made and this will fail if there is not enough | |
3951 | * space already. | |
8bb8ab2e | 3952 | */ |
4a92b1b8 | 3953 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e | 3954 | struct btrfs_block_rsv *block_rsv, |
08e007d2 MX |
3955 | u64 orig_bytes, |
3956 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 3957 | { |
f0486c68 | 3958 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 3959 | u64 used; |
8bb8ab2e | 3960 | u64 num_bytes = orig_bytes; |
67b0fd63 | 3961 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
8bb8ab2e | 3962 | int ret = 0; |
fdb5effd | 3963 | bool flushing = false; |
9ed74f2d | 3964 | |
8bb8ab2e | 3965 | again: |
fdb5effd | 3966 | ret = 0; |
8bb8ab2e | 3967 | spin_lock(&space_info->lock); |
fdb5effd | 3968 | /* |
08e007d2 MX |
3969 | * We only want to wait if somebody other than us is flushing and we |
3970 | * are actually allowed to flush all things. | |
fdb5effd | 3971 | */ |
08e007d2 MX |
3972 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && |
3973 | space_info->flush) { | |
fdb5effd JB |
3974 | spin_unlock(&space_info->lock); |
3975 | /* | |
3976 | * If we have a trans handle we can't wait because the flusher | |
3977 | * may have to commit the transaction, which would mean we would | |
3978 | * deadlock since we are waiting for the flusher to finish, but | |
3979 | * hold the current transaction open. | |
3980 | */ | |
663350ac | 3981 | if (current->journal_info) |
fdb5effd | 3982 | return -EAGAIN; |
b9688bb8 AJ |
3983 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
3984 | /* Must have been killed, return */ | |
3985 | if (ret) | |
fdb5effd JB |
3986 | return -EINTR; |
3987 | ||
3988 | spin_lock(&space_info->lock); | |
3989 | } | |
3990 | ||
3991 | ret = -ENOSPC; | |
2bf64758 JB |
3992 | used = space_info->bytes_used + space_info->bytes_reserved + |
3993 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3994 | space_info->bytes_may_use; | |
9ed74f2d | 3995 | |
8bb8ab2e JB |
3996 | /* |
3997 | * The idea here is that we've not already over-reserved the block group | |
3998 | * then we can go ahead and save our reservation first and then start | |
3999 | * flushing if we need to. Otherwise if we've already overcommitted | |
4000 | * lets start flushing stuff first and then come back and try to make | |
4001 | * our reservation. | |
4002 | */ | |
2bf64758 JB |
4003 | if (used <= space_info->total_bytes) { |
4004 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 4005 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 4006 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 4007 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
4008 | ret = 0; |
4009 | } else { | |
4010 | /* | |
4011 | * Ok set num_bytes to orig_bytes since we aren't | |
4012 | * overocmmitted, this way we only try and reclaim what | |
4013 | * we need. | |
4014 | */ | |
4015 | num_bytes = orig_bytes; | |
4016 | } | |
4017 | } else { | |
4018 | /* | |
4019 | * Ok we're over committed, set num_bytes to the overcommitted | |
4020 | * amount plus the amount of bytes that we need for this | |
4021 | * reservation. | |
4022 | */ | |
2bf64758 | 4023 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 4024 | (orig_bytes * 2); |
8bb8ab2e | 4025 | } |
9ed74f2d | 4026 | |
44734ed1 JB |
4027 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { |
4028 | space_info->bytes_may_use += orig_bytes; | |
4029 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4030 | space_info->flags, orig_bytes, | |
4031 | 1); | |
4032 | ret = 0; | |
2bf64758 JB |
4033 | } |
4034 | ||
8bb8ab2e JB |
4035 | /* |
4036 | * Couldn't make our reservation, save our place so while we're trying | |
4037 | * to reclaim space we can actually use it instead of somebody else | |
4038 | * stealing it from us. | |
08e007d2 MX |
4039 | * |
4040 | * We make the other tasks wait for the flush only when we can flush | |
4041 | * all things. | |
8bb8ab2e | 4042 | */ |
72bcd99d | 4043 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
fdb5effd JB |
4044 | flushing = true; |
4045 | space_info->flush = 1; | |
8bb8ab2e | 4046 | } |
9ed74f2d | 4047 | |
f0486c68 | 4048 | spin_unlock(&space_info->lock); |
9ed74f2d | 4049 | |
08e007d2 | 4050 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
8bb8ab2e | 4051 | goto out; |
f0486c68 | 4052 | |
96c3f433 JB |
4053 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
4054 | flush_state); | |
4055 | flush_state++; | |
08e007d2 MX |
4056 | |
4057 | /* | |
4058 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock | |
4059 | * would happen. So skip delalloc flush. | |
4060 | */ | |
4061 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
4062 | (flush_state == FLUSH_DELALLOC || | |
4063 | flush_state == FLUSH_DELALLOC_WAIT)) | |
4064 | flush_state = ALLOC_CHUNK; | |
4065 | ||
96c3f433 | 4066 | if (!ret) |
8bb8ab2e | 4067 | goto again; |
08e007d2 MX |
4068 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
4069 | flush_state < COMMIT_TRANS) | |
4070 | goto again; | |
4071 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
4072 | flush_state <= COMMIT_TRANS) | |
8bb8ab2e JB |
4073 | goto again; |
4074 | ||
4075 | out: | |
fdb5effd | 4076 | if (flushing) { |
8bb8ab2e | 4077 | spin_lock(&space_info->lock); |
fdb5effd JB |
4078 | space_info->flush = 0; |
4079 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 4080 | spin_unlock(&space_info->lock); |
f0486c68 | 4081 | } |
f0486c68 YZ |
4082 | return ret; |
4083 | } | |
4084 | ||
79787eaa JM |
4085 | static struct btrfs_block_rsv *get_block_rsv( |
4086 | const struct btrfs_trans_handle *trans, | |
4087 | const struct btrfs_root *root) | |
f0486c68 | 4088 | { |
4c13d758 JB |
4089 | struct btrfs_block_rsv *block_rsv = NULL; |
4090 | ||
0e721106 JB |
4091 | if (root->ref_cows) |
4092 | block_rsv = trans->block_rsv; | |
4093 | ||
4094 | if (root == root->fs_info->csum_root && trans->adding_csums) | |
f0486c68 | 4095 | block_rsv = trans->block_rsv; |
4c13d758 JB |
4096 | |
4097 | if (!block_rsv) | |
f0486c68 YZ |
4098 | block_rsv = root->block_rsv; |
4099 | ||
4100 | if (!block_rsv) | |
4101 | block_rsv = &root->fs_info->empty_block_rsv; | |
4102 | ||
4103 | return block_rsv; | |
4104 | } | |
4105 | ||
4106 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
4107 | u64 num_bytes) | |
4108 | { | |
4109 | int ret = -ENOSPC; | |
4110 | spin_lock(&block_rsv->lock); | |
4111 | if (block_rsv->reserved >= num_bytes) { | |
4112 | block_rsv->reserved -= num_bytes; | |
4113 | if (block_rsv->reserved < block_rsv->size) | |
4114 | block_rsv->full = 0; | |
4115 | ret = 0; | |
4116 | } | |
4117 | spin_unlock(&block_rsv->lock); | |
4118 | return ret; | |
4119 | } | |
4120 | ||
4121 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
4122 | u64 num_bytes, int update_size) | |
4123 | { | |
4124 | spin_lock(&block_rsv->lock); | |
4125 | block_rsv->reserved += num_bytes; | |
4126 | if (update_size) | |
4127 | block_rsv->size += num_bytes; | |
4128 | else if (block_rsv->reserved >= block_rsv->size) | |
4129 | block_rsv->full = 1; | |
4130 | spin_unlock(&block_rsv->lock); | |
4131 | } | |
4132 | ||
8c2a3ca2 JB |
4133 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
4134 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 4135 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
4136 | { |
4137 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
4138 | ||
4139 | spin_lock(&block_rsv->lock); | |
4140 | if (num_bytes == (u64)-1) | |
4141 | num_bytes = block_rsv->size; | |
4142 | block_rsv->size -= num_bytes; | |
4143 | if (block_rsv->reserved >= block_rsv->size) { | |
4144 | num_bytes = block_rsv->reserved - block_rsv->size; | |
4145 | block_rsv->reserved = block_rsv->size; | |
4146 | block_rsv->full = 1; | |
4147 | } else { | |
4148 | num_bytes = 0; | |
4149 | } | |
4150 | spin_unlock(&block_rsv->lock); | |
4151 | ||
4152 | if (num_bytes > 0) { | |
4153 | if (dest) { | |
e9e22899 JB |
4154 | spin_lock(&dest->lock); |
4155 | if (!dest->full) { | |
4156 | u64 bytes_to_add; | |
4157 | ||
4158 | bytes_to_add = dest->size - dest->reserved; | |
4159 | bytes_to_add = min(num_bytes, bytes_to_add); | |
4160 | dest->reserved += bytes_to_add; | |
4161 | if (dest->reserved >= dest->size) | |
4162 | dest->full = 1; | |
4163 | num_bytes -= bytes_to_add; | |
4164 | } | |
4165 | spin_unlock(&dest->lock); | |
4166 | } | |
4167 | if (num_bytes) { | |
f0486c68 | 4168 | spin_lock(&space_info->lock); |
fb25e914 | 4169 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4170 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4171 | space_info->flags, num_bytes, 0); |
36e39c40 | 4172 | space_info->reservation_progress++; |
f0486c68 | 4173 | spin_unlock(&space_info->lock); |
4e06bdd6 | 4174 | } |
9ed74f2d | 4175 | } |
f0486c68 | 4176 | } |
4e06bdd6 | 4177 | |
f0486c68 YZ |
4178 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
4179 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
4180 | { | |
4181 | int ret; | |
9ed74f2d | 4182 | |
f0486c68 YZ |
4183 | ret = block_rsv_use_bytes(src, num_bytes); |
4184 | if (ret) | |
4185 | return ret; | |
9ed74f2d | 4186 | |
f0486c68 | 4187 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
4188 | return 0; |
4189 | } | |
4190 | ||
66d8f3dd | 4191 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 4192 | { |
f0486c68 YZ |
4193 | memset(rsv, 0, sizeof(*rsv)); |
4194 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 4195 | rsv->type = type; |
f0486c68 YZ |
4196 | } |
4197 | ||
66d8f3dd MX |
4198 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
4199 | unsigned short type) | |
f0486c68 YZ |
4200 | { |
4201 | struct btrfs_block_rsv *block_rsv; | |
4202 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 4203 | |
f0486c68 YZ |
4204 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
4205 | if (!block_rsv) | |
4206 | return NULL; | |
9ed74f2d | 4207 | |
66d8f3dd | 4208 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
4209 | block_rsv->space_info = __find_space_info(fs_info, |
4210 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
4211 | return block_rsv; |
4212 | } | |
9ed74f2d | 4213 | |
f0486c68 YZ |
4214 | void btrfs_free_block_rsv(struct btrfs_root *root, |
4215 | struct btrfs_block_rsv *rsv) | |
4216 | { | |
2aaa6655 JB |
4217 | if (!rsv) |
4218 | return; | |
dabdb640 JB |
4219 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
4220 | kfree(rsv); | |
9ed74f2d JB |
4221 | } |
4222 | ||
08e007d2 MX |
4223 | int btrfs_block_rsv_add(struct btrfs_root *root, |
4224 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
4225 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4226 | { |
f0486c68 | 4227 | int ret; |
9ed74f2d | 4228 | |
f0486c68 YZ |
4229 | if (num_bytes == 0) |
4230 | return 0; | |
8bb8ab2e | 4231 | |
61b520a9 | 4232 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
4233 | if (!ret) { |
4234 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
4235 | return 0; | |
4236 | } | |
9ed74f2d | 4237 | |
f0486c68 | 4238 | return ret; |
f0486c68 | 4239 | } |
9ed74f2d | 4240 | |
4a92b1b8 | 4241 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4242 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4243 | { |
4244 | u64 num_bytes = 0; | |
f0486c68 | 4245 | int ret = -ENOSPC; |
9ed74f2d | 4246 | |
f0486c68 YZ |
4247 | if (!block_rsv) |
4248 | return 0; | |
9ed74f2d | 4249 | |
f0486c68 | 4250 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4251 | num_bytes = div_factor(block_rsv->size, min_factor); |
4252 | if (block_rsv->reserved >= num_bytes) | |
4253 | ret = 0; | |
4254 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4255 | |
36ba022a JB |
4256 | return ret; |
4257 | } | |
4258 | ||
08e007d2 MX |
4259 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4260 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
4261 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
4262 | { |
4263 | u64 num_bytes = 0; | |
4264 | int ret = -ENOSPC; | |
4265 | ||
4266 | if (!block_rsv) | |
4267 | return 0; | |
4268 | ||
4269 | spin_lock(&block_rsv->lock); | |
4270 | num_bytes = min_reserved; | |
13553e52 | 4271 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4272 | ret = 0; |
13553e52 | 4273 | else |
f0486c68 | 4274 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4275 | spin_unlock(&block_rsv->lock); |
13553e52 | 4276 | |
f0486c68 YZ |
4277 | if (!ret) |
4278 | return 0; | |
4279 | ||
aa38a711 | 4280 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4281 | if (!ret) { |
4282 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4283 | return 0; |
6a63209f | 4284 | } |
9ed74f2d | 4285 | |
13553e52 | 4286 | return ret; |
f0486c68 YZ |
4287 | } |
4288 | ||
4289 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
4290 | struct btrfs_block_rsv *dst_rsv, | |
4291 | u64 num_bytes) | |
4292 | { | |
4293 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4294 | } | |
4295 | ||
4296 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4297 | struct btrfs_block_rsv *block_rsv, | |
4298 | u64 num_bytes) | |
4299 | { | |
4300 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4301 | if (global_rsv->full || global_rsv == block_rsv || | |
4302 | block_rsv->space_info != global_rsv->space_info) | |
4303 | global_rsv = NULL; | |
8c2a3ca2 JB |
4304 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4305 | num_bytes); | |
6a63209f JB |
4306 | } |
4307 | ||
4308 | /* | |
8929ecfa YZ |
4309 | * helper to calculate size of global block reservation. |
4310 | * the desired value is sum of space used by extent tree, | |
4311 | * checksum tree and root tree | |
6a63209f | 4312 | */ |
8929ecfa | 4313 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4314 | { |
8929ecfa YZ |
4315 | struct btrfs_space_info *sinfo; |
4316 | u64 num_bytes; | |
4317 | u64 meta_used; | |
4318 | u64 data_used; | |
6c41761f | 4319 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4320 | |
8929ecfa YZ |
4321 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4322 | spin_lock(&sinfo->lock); | |
4323 | data_used = sinfo->bytes_used; | |
4324 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4325 | |
8929ecfa YZ |
4326 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4327 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4328 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4329 | data_used = 0; | |
8929ecfa YZ |
4330 | meta_used = sinfo->bytes_used; |
4331 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4332 | |
8929ecfa YZ |
4333 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4334 | csum_size * 2; | |
4335 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4336 | |
8929ecfa | 4337 | if (num_bytes * 3 > meta_used) |
8e62c2de | 4338 | num_bytes = div64_u64(meta_used, 3); |
ab6e2410 | 4339 | |
8929ecfa YZ |
4340 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4341 | } | |
6a63209f | 4342 | |
8929ecfa YZ |
4343 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4344 | { | |
4345 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4346 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4347 | u64 num_bytes; | |
6a63209f | 4348 | |
8929ecfa | 4349 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4350 | |
8929ecfa | 4351 | spin_lock(&sinfo->lock); |
1f699d38 | 4352 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 4353 | |
8929ecfa | 4354 | block_rsv->size = num_bytes; |
4e06bdd6 | 4355 | |
8929ecfa | 4356 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4357 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4358 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4359 | |
4360 | if (sinfo->total_bytes > num_bytes) { | |
4361 | num_bytes = sinfo->total_bytes - num_bytes; | |
4362 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4363 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4364 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4365 | sinfo->flags, num_bytes, 1); |
6a63209f | 4366 | } |
6a63209f | 4367 | |
8929ecfa YZ |
4368 | if (block_rsv->reserved >= block_rsv->size) { |
4369 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4370 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4371 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4372 | sinfo->flags, num_bytes, 0); |
36e39c40 | 4373 | sinfo->reservation_progress++; |
8929ecfa YZ |
4374 | block_rsv->reserved = block_rsv->size; |
4375 | block_rsv->full = 1; | |
4376 | } | |
182608c8 | 4377 | |
8929ecfa | 4378 | spin_unlock(&block_rsv->lock); |
1f699d38 | 4379 | spin_unlock(&sinfo->lock); |
6a63209f JB |
4380 | } |
4381 | ||
f0486c68 | 4382 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4383 | { |
f0486c68 | 4384 | struct btrfs_space_info *space_info; |
6a63209f | 4385 | |
f0486c68 YZ |
4386 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4387 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4388 | |
f0486c68 | 4389 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4390 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4391 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4392 | fs_info->trans_block_rsv.space_info = space_info; |
4393 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4394 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4395 | |
8929ecfa YZ |
4396 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4397 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4398 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4399 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4400 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4401 | |
8929ecfa | 4402 | update_global_block_rsv(fs_info); |
6a63209f JB |
4403 | } |
4404 | ||
8929ecfa | 4405 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4406 | { |
8c2a3ca2 JB |
4407 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4408 | (u64)-1); | |
8929ecfa YZ |
4409 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4410 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4411 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4412 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4413 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4414 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4415 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4416 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4417 | } |
4418 | ||
a22285a6 YZ |
4419 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4420 | struct btrfs_root *root) | |
6a63209f | 4421 | { |
0e721106 JB |
4422 | if (!trans->block_rsv) |
4423 | return; | |
4424 | ||
a22285a6 YZ |
4425 | if (!trans->bytes_reserved) |
4426 | return; | |
6a63209f | 4427 | |
e77266e4 | 4428 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 4429 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 4430 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4431 | trans->bytes_reserved = 0; |
4432 | } | |
6a63209f | 4433 | |
79787eaa | 4434 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
4435 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4436 | struct inode *inode) | |
4437 | { | |
4438 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4439 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4440 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4441 | ||
4442 | /* | |
fcb80c2a JB |
4443 | * We need to hold space in order to delete our orphan item once we've |
4444 | * added it, so this takes the reservation so we can release it later | |
4445 | * when we are truly done with the orphan item. | |
d68fc57b | 4446 | */ |
ff5714cc | 4447 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4448 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4449 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4450 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4451 | } |
4452 | ||
d68fc57b | 4453 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4454 | { |
d68fc57b | 4455 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4456 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4457 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4458 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4459 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4460 | } | |
97e728d4 | 4461 | |
a22285a6 YZ |
4462 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
4463 | struct btrfs_pending_snapshot *pending) | |
4464 | { | |
4465 | struct btrfs_root *root = pending->root; | |
4466 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4467 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
4468 | /* | |
48c03c4b MX |
4469 | * two for root back/forward refs, two for directory entries, |
4470 | * one for root of the snapshot and one for parent inode. | |
a22285a6 | 4471 | */ |
48c03c4b | 4472 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6); |
a22285a6 YZ |
4473 | dst_rsv->space_info = src_rsv->space_info; |
4474 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
4475 | } |
4476 | ||
7709cde3 JB |
4477 | /** |
4478 | * drop_outstanding_extent - drop an outstanding extent | |
4479 | * @inode: the inode we're dropping the extent for | |
4480 | * | |
4481 | * This is called when we are freeing up an outstanding extent, either called | |
4482 | * after an error or after an extent is written. This will return the number of | |
4483 | * reserved extents that need to be freed. This must be called with | |
4484 | * BTRFS_I(inode)->lock held. | |
4485 | */ | |
9e0baf60 JB |
4486 | static unsigned drop_outstanding_extent(struct inode *inode) |
4487 | { | |
7fd2ae21 | 4488 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4489 | unsigned dropped_extents = 0; |
4490 | ||
9e0baf60 JB |
4491 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4492 | BTRFS_I(inode)->outstanding_extents--; | |
4493 | ||
7fd2ae21 | 4494 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
4495 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4496 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 4497 | drop_inode_space = 1; |
7fd2ae21 | 4498 | |
9e0baf60 JB |
4499 | /* |
4500 | * If we have more or the same amount of outsanding extents than we have | |
4501 | * reserved then we need to leave the reserved extents count alone. | |
4502 | */ | |
4503 | if (BTRFS_I(inode)->outstanding_extents >= | |
4504 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4505 | return drop_inode_space; |
9e0baf60 JB |
4506 | |
4507 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4508 | BTRFS_I(inode)->outstanding_extents; | |
4509 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4510 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4511 | } |
4512 | ||
7709cde3 JB |
4513 | /** |
4514 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4515 | * reserved/free'd for the given bytes. | |
4516 | * @inode: the inode we're manipulating | |
4517 | * @num_bytes: the number of bytes in question | |
4518 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4519 | * | |
4520 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4521 | * correct amount of metadata that must either be reserved or freed. We | |
4522 | * calculate how many checksums we can fit into one leaf and then divide the | |
4523 | * number of bytes that will need to be checksumed by this value to figure out | |
4524 | * how many checksums will be required. If we are adding bytes then the number | |
4525 | * may go up and we will return the number of additional bytes that must be | |
4526 | * reserved. If it is going down we will return the number of bytes that must | |
4527 | * be freed. | |
4528 | * | |
4529 | * This must be called with BTRFS_I(inode)->lock held. | |
4530 | */ | |
4531 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4532 | int reserve) | |
6324fbf3 | 4533 | { |
7709cde3 JB |
4534 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4535 | u64 csum_size; | |
4536 | int num_csums_per_leaf; | |
4537 | int num_csums; | |
4538 | int old_csums; | |
4539 | ||
4540 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4541 | BTRFS_I(inode)->csum_bytes == 0) | |
4542 | return 0; | |
4543 | ||
4544 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4545 | if (reserve) | |
4546 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4547 | else | |
4548 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4549 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4550 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4551 | sizeof(struct btrfs_csum_item) + | |
4552 | sizeof(struct btrfs_disk_key)); | |
4553 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4554 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4555 | num_csums = num_csums / num_csums_per_leaf; | |
4556 | ||
4557 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4558 | old_csums = old_csums / num_csums_per_leaf; | |
4559 | ||
4560 | /* No change, no need to reserve more */ | |
4561 | if (old_csums == num_csums) | |
4562 | return 0; | |
4563 | ||
4564 | if (reserve) | |
4565 | return btrfs_calc_trans_metadata_size(root, | |
4566 | num_csums - old_csums); | |
4567 | ||
4568 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4569 | } |
c146afad | 4570 | |
0ca1f7ce YZ |
4571 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4572 | { | |
4573 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4574 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4575 | u64 to_reserve = 0; |
660d3f6c | 4576 | u64 csum_bytes; |
9e0baf60 | 4577 | unsigned nr_extents = 0; |
660d3f6c | 4578 | int extra_reserve = 0; |
08e007d2 | 4579 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 4580 | int ret = 0; |
c64c2bd8 | 4581 | bool delalloc_lock = true; |
6324fbf3 | 4582 | |
c64c2bd8 JB |
4583 | /* If we are a free space inode we need to not flush since we will be in |
4584 | * the middle of a transaction commit. We also don't need the delalloc | |
4585 | * mutex since we won't race with anybody. We need this mostly to make | |
4586 | * lockdep shut its filthy mouth. | |
4587 | */ | |
4588 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 4589 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 JB |
4590 | delalloc_lock = false; |
4591 | } | |
c09544e0 | 4592 | |
08e007d2 MX |
4593 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
4594 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4595 | schedule_timeout(1); |
ec44a35c | 4596 | |
c64c2bd8 JB |
4597 | if (delalloc_lock) |
4598 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
4599 | ||
0ca1f7ce | 4600 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4601 | |
9e0baf60 JB |
4602 | spin_lock(&BTRFS_I(inode)->lock); |
4603 | BTRFS_I(inode)->outstanding_extents++; | |
4604 | ||
4605 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4606 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4607 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4608 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4609 | |
7fd2ae21 JB |
4610 | /* |
4611 | * Add an item to reserve for updating the inode when we complete the | |
4612 | * delalloc io. | |
4613 | */ | |
72ac3c0d JB |
4614 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4615 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 4616 | nr_extents++; |
660d3f6c | 4617 | extra_reserve = 1; |
593060d7 | 4618 | } |
7fd2ae21 JB |
4619 | |
4620 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4621 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4622 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4623 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4624 | |
eb6b88d9 | 4625 | if (root->fs_info->quota_enabled) |
c5567237 AJ |
4626 | ret = btrfs_qgroup_reserve(root, num_bytes + |
4627 | nr_extents * root->leafsize); | |
c5567237 | 4628 | |
eb6b88d9 JS |
4629 | /* |
4630 | * ret != 0 here means the qgroup reservation failed, we go straight to | |
4631 | * the shared error handling then. | |
4632 | */ | |
4633 | if (ret == 0) | |
4634 | ret = reserve_metadata_bytes(root, block_rsv, | |
4635 | to_reserve, flush); | |
4636 | ||
9e0baf60 | 4637 | if (ret) { |
7ed49f18 | 4638 | u64 to_free = 0; |
9e0baf60 | 4639 | unsigned dropped; |
7ed49f18 | 4640 | |
7709cde3 | 4641 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4642 | dropped = drop_outstanding_extent(inode); |
9e0baf60 | 4643 | /* |
660d3f6c JB |
4644 | * If the inodes csum_bytes is the same as the original |
4645 | * csum_bytes then we know we haven't raced with any free()ers | |
4646 | * so we can just reduce our inodes csum bytes and carry on. | |
4647 | * Otherwise we have to do the normal free thing to account for | |
4648 | * the case that the free side didn't free up its reserve | |
4649 | * because of this outstanding reservation. | |
9e0baf60 | 4650 | */ |
660d3f6c JB |
4651 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) |
4652 | calc_csum_metadata_size(inode, num_bytes, 0); | |
4653 | else | |
4654 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
4655 | spin_unlock(&BTRFS_I(inode)->lock); | |
4656 | if (dropped) | |
4657 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
4658 | ||
8c2a3ca2 | 4659 | if (to_free) { |
7ed49f18 | 4660 | btrfs_block_rsv_release(root, block_rsv, to_free); |
8c2a3ca2 JB |
4661 | trace_btrfs_space_reservation(root->fs_info, |
4662 | "delalloc", | |
4663 | btrfs_ino(inode), | |
4664 | to_free, 0); | |
4665 | } | |
4b5829a8 MX |
4666 | if (root->fs_info->quota_enabled) { |
4667 | btrfs_qgroup_free(root, num_bytes + | |
4668 | nr_extents * root->leafsize); | |
4669 | } | |
c64c2bd8 JB |
4670 | if (delalloc_lock) |
4671 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
0ca1f7ce | 4672 | return ret; |
9e0baf60 | 4673 | } |
25179201 | 4674 | |
660d3f6c JB |
4675 | spin_lock(&BTRFS_I(inode)->lock); |
4676 | if (extra_reserve) { | |
72ac3c0d JB |
4677 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4678 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
4679 | nr_extents--; |
4680 | } | |
4681 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
4682 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
4683 | |
4684 | if (delalloc_lock) | |
4685 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 4686 | |
8c2a3ca2 JB |
4687 | if (to_reserve) |
4688 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
4689 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
4690 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4691 | ||
0ca1f7ce YZ |
4692 | return 0; |
4693 | } | |
4694 | ||
7709cde3 JB |
4695 | /** |
4696 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
4697 | * @inode: the inode to release the reservation for | |
4698 | * @num_bytes: the number of bytes we're releasing | |
4699 | * | |
4700 | * This will release the metadata reservation for an inode. This can be called | |
4701 | * once we complete IO for a given set of bytes to release their metadata | |
4702 | * reservations. | |
4703 | */ | |
0ca1f7ce YZ |
4704 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
4705 | { | |
4706 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4707 | u64 to_free = 0; |
4708 | unsigned dropped; | |
0ca1f7ce YZ |
4709 | |
4710 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 4711 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4712 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4713 | |
7709cde3 JB |
4714 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); |
4715 | spin_unlock(&BTRFS_I(inode)->lock); | |
9e0baf60 JB |
4716 | if (dropped > 0) |
4717 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 4718 | |
8c2a3ca2 JB |
4719 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
4720 | btrfs_ino(inode), to_free, 0); | |
c5567237 AJ |
4721 | if (root->fs_info->quota_enabled) { |
4722 | btrfs_qgroup_free(root, num_bytes + | |
4723 | dropped * root->leafsize); | |
4724 | } | |
4725 | ||
0ca1f7ce YZ |
4726 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
4727 | to_free); | |
4728 | } | |
4729 | ||
7709cde3 JB |
4730 | /** |
4731 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
4732 | * @inode: inode we're writing to | |
4733 | * @num_bytes: the number of bytes we want to allocate | |
4734 | * | |
4735 | * This will do the following things | |
4736 | * | |
4737 | * o reserve space in the data space info for num_bytes | |
4738 | * o reserve space in the metadata space info based on number of outstanding | |
4739 | * extents and how much csums will be needed | |
4740 | * o add to the inodes ->delalloc_bytes | |
4741 | * o add it to the fs_info's delalloc inodes list. | |
4742 | * | |
4743 | * This will return 0 for success and -ENOSPC if there is no space left. | |
4744 | */ | |
0ca1f7ce YZ |
4745 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
4746 | { | |
4747 | int ret; | |
4748 | ||
4749 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4750 | if (ret) |
0ca1f7ce YZ |
4751 | return ret; |
4752 | ||
4753 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4754 | if (ret) { | |
4755 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4756 | return ret; | |
4757 | } | |
4758 | ||
4759 | return 0; | |
4760 | } | |
4761 | ||
7709cde3 JB |
4762 | /** |
4763 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
4764 | * @inode: inode we're releasing space for | |
4765 | * @num_bytes: the number of bytes we want to free up | |
4766 | * | |
4767 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
4768 | * called in the case that we don't need the metadata AND data reservations | |
4769 | * anymore. So if there is an error or we insert an inline extent. | |
4770 | * | |
4771 | * This function will release the metadata space that was not used and will | |
4772 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
4773 | * list if there are no delalloc bytes left. | |
4774 | */ | |
0ca1f7ce YZ |
4775 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
4776 | { | |
4777 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4778 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4779 | } |
4780 | ||
c53d613e | 4781 | static int update_block_group(struct btrfs_root *root, |
f0486c68 | 4782 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4783 | { |
0af3d00b | 4784 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4785 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4786 | u64 total = num_bytes; |
9078a3e1 | 4787 | u64 old_val; |
db94535d | 4788 | u64 byte_in_group; |
0af3d00b | 4789 | int factor; |
3e1ad54f | 4790 | |
5d4f98a2 YZ |
4791 | /* block accounting for super block */ |
4792 | spin_lock(&info->delalloc_lock); | |
6c41761f | 4793 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
4794 | if (alloc) |
4795 | old_val += num_bytes; | |
4796 | else | |
4797 | old_val -= num_bytes; | |
6c41761f | 4798 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
5d4f98a2 YZ |
4799 | spin_unlock(&info->delalloc_lock); |
4800 | ||
d397712b | 4801 | while (total) { |
db94535d | 4802 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4803 | if (!cache) |
79787eaa | 4804 | return -ENOENT; |
b742bb82 YZ |
4805 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4806 | BTRFS_BLOCK_GROUP_RAID1 | | |
4807 | BTRFS_BLOCK_GROUP_RAID10)) | |
4808 | factor = 2; | |
4809 | else | |
4810 | factor = 1; | |
9d66e233 JB |
4811 | /* |
4812 | * If this block group has free space cache written out, we | |
4813 | * need to make sure to load it if we are removing space. This | |
4814 | * is because we need the unpinning stage to actually add the | |
4815 | * space back to the block group, otherwise we will leak space. | |
4816 | */ | |
4817 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 4818 | cache_block_group(cache, 1); |
0af3d00b | 4819 | |
db94535d CM |
4820 | byte_in_group = bytenr - cache->key.objectid; |
4821 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4822 | |
25179201 | 4823 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4824 | spin_lock(&cache->lock); |
0af3d00b | 4825 | |
73bc1876 | 4826 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
4827 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
4828 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4829 | ||
0f9dd46c | 4830 | cache->dirty = 1; |
9078a3e1 | 4831 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4832 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4833 | if (alloc) { |
db94535d | 4834 | old_val += num_bytes; |
11833d66 YZ |
4835 | btrfs_set_block_group_used(&cache->item, old_val); |
4836 | cache->reserved -= num_bytes; | |
11833d66 | 4837 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
4838 | cache->space_info->bytes_used += num_bytes; |
4839 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4840 | spin_unlock(&cache->lock); |
25179201 | 4841 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4842 | } else { |
db94535d | 4843 | old_val -= num_bytes; |
c286ac48 | 4844 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4845 | cache->pinned += num_bytes; |
4846 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4847 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4848 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4849 | spin_unlock(&cache->lock); |
25179201 | 4850 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4851 | |
f0486c68 YZ |
4852 | set_extent_dirty(info->pinned_extents, |
4853 | bytenr, bytenr + num_bytes - 1, | |
4854 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4855 | } |
fa9c0d79 | 4856 | btrfs_put_block_group(cache); |
db94535d CM |
4857 | total -= num_bytes; |
4858 | bytenr += num_bytes; | |
9078a3e1 CM |
4859 | } |
4860 | return 0; | |
4861 | } | |
6324fbf3 | 4862 | |
a061fc8d CM |
4863 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4864 | { | |
0f9dd46c | 4865 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4866 | u64 bytenr; |
0f9dd46c | 4867 | |
a1897fdd LB |
4868 | spin_lock(&root->fs_info->block_group_cache_lock); |
4869 | bytenr = root->fs_info->first_logical_byte; | |
4870 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
4871 | ||
4872 | if (bytenr < (u64)-1) | |
4873 | return bytenr; | |
4874 | ||
0f9dd46c JB |
4875 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
4876 | if (!cache) | |
a061fc8d | 4877 | return 0; |
0f9dd46c | 4878 | |
d2fb3437 | 4879 | bytenr = cache->key.objectid; |
fa9c0d79 | 4880 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4881 | |
4882 | return bytenr; | |
a061fc8d CM |
4883 | } |
4884 | ||
f0486c68 YZ |
4885 | static int pin_down_extent(struct btrfs_root *root, |
4886 | struct btrfs_block_group_cache *cache, | |
4887 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4888 | { |
11833d66 YZ |
4889 | spin_lock(&cache->space_info->lock); |
4890 | spin_lock(&cache->lock); | |
4891 | cache->pinned += num_bytes; | |
4892 | cache->space_info->bytes_pinned += num_bytes; | |
4893 | if (reserved) { | |
4894 | cache->reserved -= num_bytes; | |
4895 | cache->space_info->bytes_reserved -= num_bytes; | |
4896 | } | |
4897 | spin_unlock(&cache->lock); | |
4898 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4899 | |
f0486c68 YZ |
4900 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4901 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4902 | return 0; | |
4903 | } | |
68b38550 | 4904 | |
f0486c68 YZ |
4905 | /* |
4906 | * this function must be called within transaction | |
4907 | */ | |
4908 | int btrfs_pin_extent(struct btrfs_root *root, | |
4909 | u64 bytenr, u64 num_bytes, int reserved) | |
4910 | { | |
4911 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4912 | |
f0486c68 | 4913 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 4914 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
4915 | |
4916 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4917 | ||
4918 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4919 | return 0; |
4920 | } | |
4921 | ||
f0486c68 | 4922 | /* |
e688b725 CM |
4923 | * this function must be called within transaction |
4924 | */ | |
dcfac415 | 4925 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
e688b725 CM |
4926 | u64 bytenr, u64 num_bytes) |
4927 | { | |
4928 | struct btrfs_block_group_cache *cache; | |
4929 | ||
4930 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
79787eaa | 4931 | BUG_ON(!cache); /* Logic error */ |
e688b725 CM |
4932 | |
4933 | /* | |
4934 | * pull in the free space cache (if any) so that our pin | |
4935 | * removes the free space from the cache. We have load_only set | |
4936 | * to one because the slow code to read in the free extents does check | |
4937 | * the pinned extents. | |
4938 | */ | |
f6373bf3 | 4939 | cache_block_group(cache, 1); |
e688b725 CM |
4940 | |
4941 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
4942 | ||
4943 | /* remove us from the free space cache (if we're there at all) */ | |
4944 | btrfs_remove_free_space(cache, bytenr, num_bytes); | |
4945 | btrfs_put_block_group(cache); | |
4946 | return 0; | |
4947 | } | |
4948 | ||
fb25e914 JB |
4949 | /** |
4950 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
4951 | * @cache: The cache we are manipulating | |
4952 | * @num_bytes: The number of bytes in question | |
4953 | * @reserve: One of the reservation enums | |
4954 | * | |
4955 | * This is called by the allocator when it reserves space, or by somebody who is | |
4956 | * freeing space that was never actually used on disk. For example if you | |
4957 | * reserve some space for a new leaf in transaction A and before transaction A | |
4958 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
4959 | * clear the reservation. | |
4960 | * | |
4961 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
4962 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
4963 | * delalloc bits in the io_tree. We have to do this since we could end up | |
4964 | * allocating less disk space for the amount of data we have reserved in the | |
4965 | * case of compression. | |
4966 | * | |
4967 | * If this is a reservation and the block group has become read only we cannot | |
4968 | * make the reservation and return -EAGAIN, otherwise this function always | |
4969 | * succeeds. | |
f0486c68 | 4970 | */ |
fb25e914 JB |
4971 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4972 | u64 num_bytes, int reserve) | |
11833d66 | 4973 | { |
fb25e914 | 4974 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 4975 | int ret = 0; |
79787eaa | 4976 | |
fb25e914 JB |
4977 | spin_lock(&space_info->lock); |
4978 | spin_lock(&cache->lock); | |
4979 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
4980 | if (cache->ro) { |
4981 | ret = -EAGAIN; | |
4982 | } else { | |
fb25e914 JB |
4983 | cache->reserved += num_bytes; |
4984 | space_info->bytes_reserved += num_bytes; | |
4985 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 4986 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
4987 | "space_info", space_info->flags, |
4988 | num_bytes, 0); | |
fb25e914 JB |
4989 | space_info->bytes_may_use -= num_bytes; |
4990 | } | |
f0486c68 | 4991 | } |
fb25e914 JB |
4992 | } else { |
4993 | if (cache->ro) | |
4994 | space_info->bytes_readonly += num_bytes; | |
4995 | cache->reserved -= num_bytes; | |
4996 | space_info->bytes_reserved -= num_bytes; | |
4997 | space_info->reservation_progress++; | |
324ae4df | 4998 | } |
fb25e914 JB |
4999 | spin_unlock(&cache->lock); |
5000 | spin_unlock(&space_info->lock); | |
f0486c68 | 5001 | return ret; |
324ae4df | 5002 | } |
9078a3e1 | 5003 | |
143bede5 | 5004 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 5005 | struct btrfs_root *root) |
e8569813 | 5006 | { |
e8569813 | 5007 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
5008 | struct btrfs_caching_control *next; |
5009 | struct btrfs_caching_control *caching_ctl; | |
5010 | struct btrfs_block_group_cache *cache; | |
e8569813 | 5011 | |
11833d66 | 5012 | down_write(&fs_info->extent_commit_sem); |
25179201 | 5013 | |
11833d66 YZ |
5014 | list_for_each_entry_safe(caching_ctl, next, |
5015 | &fs_info->caching_block_groups, list) { | |
5016 | cache = caching_ctl->block_group; | |
5017 | if (block_group_cache_done(cache)) { | |
5018 | cache->last_byte_to_unpin = (u64)-1; | |
5019 | list_del_init(&caching_ctl->list); | |
5020 | put_caching_control(caching_ctl); | |
e8569813 | 5021 | } else { |
11833d66 | 5022 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 5023 | } |
e8569813 | 5024 | } |
11833d66 YZ |
5025 | |
5026 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
5027 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
5028 | else | |
5029 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
5030 | ||
5031 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
5032 | |
5033 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
5034 | } |
5035 | ||
11833d66 | 5036 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 5037 | { |
11833d66 YZ |
5038 | struct btrfs_fs_info *fs_info = root->fs_info; |
5039 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
5040 | struct btrfs_space_info *space_info; |
5041 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
11833d66 | 5042 | u64 len; |
7b398f8e | 5043 | bool readonly; |
ccd467d6 | 5044 | |
11833d66 | 5045 | while (start <= end) { |
7b398f8e | 5046 | readonly = false; |
11833d66 YZ |
5047 | if (!cache || |
5048 | start >= cache->key.objectid + cache->key.offset) { | |
5049 | if (cache) | |
5050 | btrfs_put_block_group(cache); | |
5051 | cache = btrfs_lookup_block_group(fs_info, start); | |
79787eaa | 5052 | BUG_ON(!cache); /* Logic error */ |
11833d66 YZ |
5053 | } |
5054 | ||
5055 | len = cache->key.objectid + cache->key.offset - start; | |
5056 | len = min(len, end + 1 - start); | |
5057 | ||
5058 | if (start < cache->last_byte_to_unpin) { | |
5059 | len = min(len, cache->last_byte_to_unpin - start); | |
5060 | btrfs_add_free_space(cache, start, len); | |
5061 | } | |
5062 | ||
f0486c68 | 5063 | start += len; |
7b398f8e | 5064 | space_info = cache->space_info; |
f0486c68 | 5065 | |
7b398f8e | 5066 | spin_lock(&space_info->lock); |
11833d66 YZ |
5067 | spin_lock(&cache->lock); |
5068 | cache->pinned -= len; | |
7b398f8e JB |
5069 | space_info->bytes_pinned -= len; |
5070 | if (cache->ro) { | |
5071 | space_info->bytes_readonly += len; | |
5072 | readonly = true; | |
5073 | } | |
11833d66 | 5074 | spin_unlock(&cache->lock); |
7b398f8e JB |
5075 | if (!readonly && global_rsv->space_info == space_info) { |
5076 | spin_lock(&global_rsv->lock); | |
5077 | if (!global_rsv->full) { | |
5078 | len = min(len, global_rsv->size - | |
5079 | global_rsv->reserved); | |
5080 | global_rsv->reserved += len; | |
5081 | space_info->bytes_may_use += len; | |
5082 | if (global_rsv->reserved >= global_rsv->size) | |
5083 | global_rsv->full = 1; | |
5084 | } | |
5085 | spin_unlock(&global_rsv->lock); | |
5086 | } | |
5087 | spin_unlock(&space_info->lock); | |
ccd467d6 | 5088 | } |
11833d66 YZ |
5089 | |
5090 | if (cache) | |
5091 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
5092 | return 0; |
5093 | } | |
5094 | ||
5095 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 5096 | struct btrfs_root *root) |
a28ec197 | 5097 | { |
11833d66 YZ |
5098 | struct btrfs_fs_info *fs_info = root->fs_info; |
5099 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
5100 | u64 start; |
5101 | u64 end; | |
a28ec197 | 5102 | int ret; |
a28ec197 | 5103 | |
79787eaa JM |
5104 | if (trans->aborted) |
5105 | return 0; | |
5106 | ||
11833d66 YZ |
5107 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
5108 | unpin = &fs_info->freed_extents[1]; | |
5109 | else | |
5110 | unpin = &fs_info->freed_extents[0]; | |
5111 | ||
d397712b | 5112 | while (1) { |
1a5bc167 | 5113 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 5114 | EXTENT_DIRTY, NULL); |
1a5bc167 | 5115 | if (ret) |
a28ec197 | 5116 | break; |
1f3c79a2 | 5117 | |
5378e607 LD |
5118 | if (btrfs_test_opt(root, DISCARD)) |
5119 | ret = btrfs_discard_extent(root, start, | |
5120 | end + 1 - start, NULL); | |
1f3c79a2 | 5121 | |
1a5bc167 | 5122 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 5123 | unpin_extent_range(root, start, end); |
b9473439 | 5124 | cond_resched(); |
a28ec197 | 5125 | } |
817d52f8 | 5126 | |
e20d96d6 CM |
5127 | return 0; |
5128 | } | |
5129 | ||
5d4f98a2 YZ |
5130 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
5131 | struct btrfs_root *root, | |
5132 | u64 bytenr, u64 num_bytes, u64 parent, | |
5133 | u64 root_objectid, u64 owner_objectid, | |
5134 | u64 owner_offset, int refs_to_drop, | |
5135 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 5136 | { |
e2fa7227 | 5137 | struct btrfs_key key; |
5d4f98a2 | 5138 | struct btrfs_path *path; |
1261ec42 CM |
5139 | struct btrfs_fs_info *info = root->fs_info; |
5140 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 5141 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
5142 | struct btrfs_extent_item *ei; |
5143 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 5144 | int ret; |
5d4f98a2 | 5145 | int is_data; |
952fccac CM |
5146 | int extent_slot = 0; |
5147 | int found_extent = 0; | |
5148 | int num_to_del = 1; | |
5d4f98a2 YZ |
5149 | u32 item_size; |
5150 | u64 refs; | |
037e6390 | 5151 | |
5caf2a00 | 5152 | path = btrfs_alloc_path(); |
54aa1f4d CM |
5153 | if (!path) |
5154 | return -ENOMEM; | |
5f26f772 | 5155 | |
3c12ac72 | 5156 | path->reada = 1; |
b9473439 | 5157 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5158 | |
5159 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
5160 | BUG_ON(!is_data && refs_to_drop != 1); | |
5161 | ||
5162 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
5163 | bytenr, num_bytes, parent, | |
5164 | root_objectid, owner_objectid, | |
5165 | owner_offset); | |
7bb86316 | 5166 | if (ret == 0) { |
952fccac | 5167 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
5168 | while (extent_slot >= 0) { |
5169 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 5170 | extent_slot); |
5d4f98a2 | 5171 | if (key.objectid != bytenr) |
952fccac | 5172 | break; |
5d4f98a2 YZ |
5173 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
5174 | key.offset == num_bytes) { | |
952fccac CM |
5175 | found_extent = 1; |
5176 | break; | |
5177 | } | |
5178 | if (path->slots[0] - extent_slot > 5) | |
5179 | break; | |
5d4f98a2 | 5180 | extent_slot--; |
952fccac | 5181 | } |
5d4f98a2 YZ |
5182 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
5183 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
5184 | if (found_extent && item_size < sizeof(*ei)) | |
5185 | found_extent = 0; | |
5186 | #endif | |
31840ae1 | 5187 | if (!found_extent) { |
5d4f98a2 | 5188 | BUG_ON(iref); |
56bec294 | 5189 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
5190 | NULL, refs_to_drop, |
5191 | is_data); | |
005d6427 DS |
5192 | if (ret) { |
5193 | btrfs_abort_transaction(trans, extent_root, ret); | |
5194 | goto out; | |
5195 | } | |
b3b4aa74 | 5196 | btrfs_release_path(path); |
b9473439 | 5197 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5198 | |
5199 | key.objectid = bytenr; | |
5200 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5201 | key.offset = num_bytes; | |
5202 | ||
31840ae1 ZY |
5203 | ret = btrfs_search_slot(trans, extent_root, |
5204 | &key, path, -1, 1); | |
f3465ca4 JB |
5205 | if (ret) { |
5206 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
5207 | ", was looking for %llu\n", ret, |
5208 | (unsigned long long)bytenr); | |
b783e62d JB |
5209 | if (ret > 0) |
5210 | btrfs_print_leaf(extent_root, | |
5211 | path->nodes[0]); | |
f3465ca4 | 5212 | } |
005d6427 DS |
5213 | if (ret < 0) { |
5214 | btrfs_abort_transaction(trans, extent_root, ret); | |
5215 | goto out; | |
5216 | } | |
31840ae1 ZY |
5217 | extent_slot = path->slots[0]; |
5218 | } | |
79787eaa | 5219 | } else if (ret == -ENOENT) { |
7bb86316 CM |
5220 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5221 | WARN_ON(1); | |
d397712b | 5222 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 5223 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 5224 | (unsigned long long)bytenr, |
56bec294 | 5225 | (unsigned long long)parent, |
d397712b | 5226 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
5227 | (unsigned long long)owner_objectid, |
5228 | (unsigned long long)owner_offset); | |
79787eaa | 5229 | } else { |
005d6427 DS |
5230 | btrfs_abort_transaction(trans, extent_root, ret); |
5231 | goto out; | |
7bb86316 | 5232 | } |
5f39d397 CM |
5233 | |
5234 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
5235 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
5236 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
5237 | if (item_size < sizeof(*ei)) { | |
5238 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
5239 | ret = convert_extent_item_v0(trans, extent_root, path, | |
5240 | owner_objectid, 0); | |
005d6427 DS |
5241 | if (ret < 0) { |
5242 | btrfs_abort_transaction(trans, extent_root, ret); | |
5243 | goto out; | |
5244 | } | |
5d4f98a2 | 5245 | |
b3b4aa74 | 5246 | btrfs_release_path(path); |
5d4f98a2 YZ |
5247 | path->leave_spinning = 1; |
5248 | ||
5249 | key.objectid = bytenr; | |
5250 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5251 | key.offset = num_bytes; | |
5252 | ||
5253 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
5254 | -1, 1); | |
5255 | if (ret) { | |
5256 | printk(KERN_ERR "umm, got %d back from search" | |
5257 | ", was looking for %llu\n", ret, | |
5258 | (unsigned long long)bytenr); | |
5259 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
5260 | } | |
005d6427 DS |
5261 | if (ret < 0) { |
5262 | btrfs_abort_transaction(trans, extent_root, ret); | |
5263 | goto out; | |
5264 | } | |
5265 | ||
5d4f98a2 YZ |
5266 | extent_slot = path->slots[0]; |
5267 | leaf = path->nodes[0]; | |
5268 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
5269 | } | |
5270 | #endif | |
5271 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 5272 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 5273 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5274 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
5275 | struct btrfs_tree_block_info *bi; | |
5276 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
5277 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
5278 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
5279 | } | |
56bec294 | 5280 | |
5d4f98a2 | 5281 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
5282 | BUG_ON(refs < refs_to_drop); |
5283 | refs -= refs_to_drop; | |
5f39d397 | 5284 | |
5d4f98a2 YZ |
5285 | if (refs > 0) { |
5286 | if (extent_op) | |
5287 | __run_delayed_extent_op(extent_op, leaf, ei); | |
5288 | /* | |
5289 | * In the case of inline back ref, reference count will | |
5290 | * be updated by remove_extent_backref | |
952fccac | 5291 | */ |
5d4f98a2 YZ |
5292 | if (iref) { |
5293 | BUG_ON(!found_extent); | |
5294 | } else { | |
5295 | btrfs_set_extent_refs(leaf, ei, refs); | |
5296 | btrfs_mark_buffer_dirty(leaf); | |
5297 | } | |
5298 | if (found_extent) { | |
5299 | ret = remove_extent_backref(trans, extent_root, path, | |
5300 | iref, refs_to_drop, | |
5301 | is_data); | |
005d6427 DS |
5302 | if (ret) { |
5303 | btrfs_abort_transaction(trans, extent_root, ret); | |
5304 | goto out; | |
5305 | } | |
952fccac | 5306 | } |
5d4f98a2 | 5307 | } else { |
5d4f98a2 YZ |
5308 | if (found_extent) { |
5309 | BUG_ON(is_data && refs_to_drop != | |
5310 | extent_data_ref_count(root, path, iref)); | |
5311 | if (iref) { | |
5312 | BUG_ON(path->slots[0] != extent_slot); | |
5313 | } else { | |
5314 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5315 | path->slots[0] = extent_slot; | |
5316 | num_to_del = 2; | |
5317 | } | |
78fae27e | 5318 | } |
b9473439 | 5319 | |
952fccac CM |
5320 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5321 | num_to_del); | |
005d6427 DS |
5322 | if (ret) { |
5323 | btrfs_abort_transaction(trans, extent_root, ret); | |
5324 | goto out; | |
5325 | } | |
b3b4aa74 | 5326 | btrfs_release_path(path); |
21af804c | 5327 | |
5d4f98a2 | 5328 | if (is_data) { |
459931ec | 5329 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 DS |
5330 | if (ret) { |
5331 | btrfs_abort_transaction(trans, extent_root, ret); | |
5332 | goto out; | |
5333 | } | |
459931ec CM |
5334 | } |
5335 | ||
c53d613e | 5336 | ret = update_block_group(root, bytenr, num_bytes, 0); |
005d6427 DS |
5337 | if (ret) { |
5338 | btrfs_abort_transaction(trans, extent_root, ret); | |
5339 | goto out; | |
5340 | } | |
a28ec197 | 5341 | } |
79787eaa | 5342 | out: |
5caf2a00 | 5343 | btrfs_free_path(path); |
a28ec197 CM |
5344 | return ret; |
5345 | } | |
5346 | ||
1887be66 | 5347 | /* |
f0486c68 | 5348 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5349 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5350 | * a given extent, and if there are no other delayed refs to be processed, it | |
5351 | * removes it from the tree. | |
5352 | */ | |
5353 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5354 | struct btrfs_root *root, u64 bytenr) | |
5355 | { | |
5356 | struct btrfs_delayed_ref_head *head; | |
5357 | struct btrfs_delayed_ref_root *delayed_refs; | |
5358 | struct btrfs_delayed_ref_node *ref; | |
5359 | struct rb_node *node; | |
f0486c68 | 5360 | int ret = 0; |
1887be66 CM |
5361 | |
5362 | delayed_refs = &trans->transaction->delayed_refs; | |
5363 | spin_lock(&delayed_refs->lock); | |
5364 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5365 | if (!head) | |
5366 | goto out; | |
5367 | ||
5368 | node = rb_prev(&head->node.rb_node); | |
5369 | if (!node) | |
5370 | goto out; | |
5371 | ||
5372 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5373 | ||
5374 | /* there are still entries for this ref, we can't drop it */ | |
5375 | if (ref->bytenr == bytenr) | |
5376 | goto out; | |
5377 | ||
5d4f98a2 YZ |
5378 | if (head->extent_op) { |
5379 | if (!head->must_insert_reserved) | |
5380 | goto out; | |
78a6184a | 5381 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
5382 | head->extent_op = NULL; |
5383 | } | |
5384 | ||
1887be66 CM |
5385 | /* |
5386 | * waiting for the lock here would deadlock. If someone else has it | |
5387 | * locked they are already in the process of dropping it anyway | |
5388 | */ | |
5389 | if (!mutex_trylock(&head->mutex)) | |
5390 | goto out; | |
5391 | ||
5392 | /* | |
5393 | * at this point we have a head with no other entries. Go | |
5394 | * ahead and process it. | |
5395 | */ | |
5396 | head->node.in_tree = 0; | |
5397 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5398 | |
1887be66 CM |
5399 | delayed_refs->num_entries--; |
5400 | ||
5401 | /* | |
5402 | * we don't take a ref on the node because we're removing it from the | |
5403 | * tree, so we just steal the ref the tree was holding. | |
5404 | */ | |
c3e69d58 CM |
5405 | delayed_refs->num_heads--; |
5406 | if (list_empty(&head->cluster)) | |
5407 | delayed_refs->num_heads_ready--; | |
5408 | ||
5409 | list_del_init(&head->cluster); | |
1887be66 CM |
5410 | spin_unlock(&delayed_refs->lock); |
5411 | ||
f0486c68 YZ |
5412 | BUG_ON(head->extent_op); |
5413 | if (head->must_insert_reserved) | |
5414 | ret = 1; | |
5415 | ||
5416 | mutex_unlock(&head->mutex); | |
1887be66 | 5417 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5418 | return ret; |
1887be66 CM |
5419 | out: |
5420 | spin_unlock(&delayed_refs->lock); | |
5421 | return 0; | |
5422 | } | |
5423 | ||
f0486c68 YZ |
5424 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5425 | struct btrfs_root *root, | |
5426 | struct extent_buffer *buf, | |
5581a51a | 5427 | u64 parent, int last_ref) |
f0486c68 | 5428 | { |
f0486c68 YZ |
5429 | struct btrfs_block_group_cache *cache = NULL; |
5430 | int ret; | |
5431 | ||
5432 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5433 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5434 | buf->start, buf->len, | |
5435 | parent, root->root_key.objectid, | |
5436 | btrfs_header_level(buf), | |
5581a51a | 5437 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
79787eaa | 5438 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
5439 | } |
5440 | ||
5441 | if (!last_ref) | |
5442 | return; | |
5443 | ||
f0486c68 | 5444 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5445 | |
5446 | if (btrfs_header_generation(buf) == trans->transid) { | |
5447 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5448 | ret = check_ref_cleanup(trans, root, buf->start); | |
5449 | if (!ret) | |
37be25bc | 5450 | goto out; |
f0486c68 YZ |
5451 | } |
5452 | ||
5453 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5454 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5455 | goto out; |
f0486c68 YZ |
5456 | } |
5457 | ||
5458 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5459 | ||
5460 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5461 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
f0486c68 YZ |
5462 | } |
5463 | out: | |
a826d6dc JB |
5464 | /* |
5465 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
5466 | * anymore. | |
5467 | */ | |
5468 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
5469 | btrfs_put_block_group(cache); |
5470 | } | |
5471 | ||
79787eaa | 5472 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
5473 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5474 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
5475 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
5476 | { |
5477 | int ret; | |
66d7e7f0 | 5478 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 5479 | |
56bec294 CM |
5480 | /* |
5481 | * tree log blocks never actually go into the extent allocation | |
5482 | * tree, just update pinning info and exit early. | |
56bec294 | 5483 | */ |
5d4f98a2 YZ |
5484 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
5485 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 5486 | /* unlocks the pinned mutex */ |
11833d66 | 5487 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 5488 | ret = 0; |
5d4f98a2 | 5489 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
5490 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
5491 | num_bytes, | |
5d4f98a2 | 5492 | parent, root_objectid, (int)owner, |
66d7e7f0 | 5493 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 5494 | } else { |
66d7e7f0 AJ |
5495 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5496 | num_bytes, | |
5497 | parent, root_objectid, owner, | |
5498 | offset, BTRFS_DROP_DELAYED_REF, | |
5499 | NULL, for_cow); | |
56bec294 | 5500 | } |
925baedd CM |
5501 | return ret; |
5502 | } | |
5503 | ||
87ee04eb CM |
5504 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
5505 | { | |
5506 | u64 mask = ((u64)root->stripesize - 1); | |
5507 | u64 ret = (val + mask) & ~mask; | |
5508 | return ret; | |
5509 | } | |
5510 | ||
817d52f8 JB |
5511 | /* |
5512 | * when we wait for progress in the block group caching, its because | |
5513 | * our allocation attempt failed at least once. So, we must sleep | |
5514 | * and let some progress happen before we try again. | |
5515 | * | |
5516 | * This function will sleep at least once waiting for new free space to | |
5517 | * show up, and then it will check the block group free space numbers | |
5518 | * for our min num_bytes. Another option is to have it go ahead | |
5519 | * and look in the rbtree for a free extent of a given size, but this | |
5520 | * is a good start. | |
5521 | */ | |
5522 | static noinline int | |
5523 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
5524 | u64 num_bytes) | |
5525 | { | |
11833d66 | 5526 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
5527 | DEFINE_WAIT(wait); |
5528 | ||
11833d66 YZ |
5529 | caching_ctl = get_caching_control(cache); |
5530 | if (!caching_ctl) | |
817d52f8 | 5531 | return 0; |
817d52f8 | 5532 | |
11833d66 | 5533 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5534 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5535 | |
5536 | put_caching_control(caching_ctl); | |
5537 | return 0; | |
5538 | } | |
5539 | ||
5540 | static noinline int | |
5541 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5542 | { | |
5543 | struct btrfs_caching_control *caching_ctl; | |
5544 | DEFINE_WAIT(wait); | |
5545 | ||
5546 | caching_ctl = get_caching_control(cache); | |
5547 | if (!caching_ctl) | |
5548 | return 0; | |
5549 | ||
5550 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
5551 | ||
5552 | put_caching_control(caching_ctl); | |
817d52f8 JB |
5553 | return 0; |
5554 | } | |
5555 | ||
31e50229 | 5556 | int __get_raid_index(u64 flags) |
b742bb82 | 5557 | { |
7738a53a | 5558 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 5559 | return BTRFS_RAID_RAID10; |
7738a53a | 5560 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 5561 | return BTRFS_RAID_RAID1; |
7738a53a | 5562 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 5563 | return BTRFS_RAID_DUP; |
7738a53a | 5564 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 5565 | return BTRFS_RAID_RAID0; |
b742bb82 | 5566 | else |
e6ec716f | 5567 | return BTRFS_RAID_SINGLE; |
b742bb82 YZ |
5568 | } |
5569 | ||
7738a53a ID |
5570 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
5571 | { | |
31e50229 | 5572 | return __get_raid_index(cache->flags); |
7738a53a ID |
5573 | } |
5574 | ||
817d52f8 | 5575 | enum btrfs_loop_type { |
285ff5af JB |
5576 | LOOP_CACHING_NOWAIT = 0, |
5577 | LOOP_CACHING_WAIT = 1, | |
5578 | LOOP_ALLOC_CHUNK = 2, | |
5579 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
5580 | }; |
5581 | ||
fec577fb CM |
5582 | /* |
5583 | * walks the btree of allocated extents and find a hole of a given size. | |
5584 | * The key ins is changed to record the hole: | |
5585 | * ins->objectid == block start | |
62e2749e | 5586 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
5587 | * ins->offset == number of blocks |
5588 | * Any available blocks before search_start are skipped. | |
5589 | */ | |
d397712b | 5590 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
5591 | struct btrfs_root *orig_root, |
5592 | u64 num_bytes, u64 empty_size, | |
98ed5174 | 5593 | u64 hint_byte, struct btrfs_key *ins, |
e0f54067 | 5594 | u64 data) |
fec577fb | 5595 | { |
80eb234a | 5596 | int ret = 0; |
d397712b | 5597 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 5598 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 5599 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 5600 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 5601 | u64 search_start = 0; |
239b14b3 | 5602 | int empty_cluster = 2 * 1024 * 1024; |
80eb234a | 5603 | struct btrfs_space_info *space_info; |
fa9c0d79 | 5604 | int loop = 0; |
ac5c9300 | 5605 | int index = __get_raid_index(data); |
fb25e914 JB |
5606 | int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ? |
5607 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
817d52f8 | 5608 | bool found_uncached_bg = false; |
0a24325e | 5609 | bool failed_cluster_refill = false; |
1cdda9b8 | 5610 | bool failed_alloc = false; |
67377734 | 5611 | bool use_cluster = true; |
60d2adbb | 5612 | bool have_caching_bg = false; |
fec577fb | 5613 | |
db94535d | 5614 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 5615 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
5616 | ins->objectid = 0; |
5617 | ins->offset = 0; | |
b1a4d965 | 5618 | |
3f7de037 JB |
5619 | trace_find_free_extent(orig_root, num_bytes, empty_size, data); |
5620 | ||
2552d17e | 5621 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 5622 | if (!space_info) { |
e0f54067 | 5623 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
5624 | return -ENOSPC; |
5625 | } | |
2552d17e | 5626 | |
67377734 JB |
5627 | /* |
5628 | * If the space info is for both data and metadata it means we have a | |
5629 | * small filesystem and we can't use the clustering stuff. | |
5630 | */ | |
5631 | if (btrfs_mixed_space_info(space_info)) | |
5632 | use_cluster = false; | |
5633 | ||
67377734 | 5634 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5635 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5636 | if (!btrfs_test_opt(root, SSD)) |
5637 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5638 | } |
5639 | ||
67377734 JB |
5640 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5641 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5642 | last_ptr = &root->fs_info->data_alloc_cluster; |
5643 | } | |
0f9dd46c | 5644 | |
239b14b3 | 5645 | if (last_ptr) { |
fa9c0d79 CM |
5646 | spin_lock(&last_ptr->lock); |
5647 | if (last_ptr->block_group) | |
5648 | hint_byte = last_ptr->window_start; | |
5649 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5650 | } |
fa9c0d79 | 5651 | |
a061fc8d | 5652 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5653 | search_start = max(search_start, hint_byte); |
0b86a832 | 5654 | |
817d52f8 | 5655 | if (!last_ptr) |
fa9c0d79 | 5656 | empty_cluster = 0; |
fa9c0d79 | 5657 | |
2552d17e | 5658 | if (search_start == hint_byte) { |
2552d17e JB |
5659 | block_group = btrfs_lookup_block_group(root->fs_info, |
5660 | search_start); | |
274bd4fb | 5661 | used_block_group = block_group; |
817d52f8 JB |
5662 | /* |
5663 | * we don't want to use the block group if it doesn't match our | |
5664 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5665 | * |
5666 | * However if we are re-searching with an ideal block group | |
5667 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5668 | */ |
5669 | if (block_group && block_group_bits(block_group, data) && | |
285ff5af | 5670 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 5671 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5672 | if (list_empty(&block_group->list) || |
5673 | block_group->ro) { | |
5674 | /* | |
5675 | * someone is removing this block group, | |
5676 | * we can't jump into the have_block_group | |
5677 | * target because our list pointers are not | |
5678 | * valid | |
5679 | */ | |
5680 | btrfs_put_block_group(block_group); | |
5681 | up_read(&space_info->groups_sem); | |
ccf0e725 | 5682 | } else { |
b742bb82 | 5683 | index = get_block_group_index(block_group); |
44fb5511 | 5684 | goto have_block_group; |
ccf0e725 | 5685 | } |
2552d17e | 5686 | } else if (block_group) { |
fa9c0d79 | 5687 | btrfs_put_block_group(block_group); |
2552d17e | 5688 | } |
42e70e7a | 5689 | } |
2552d17e | 5690 | search: |
60d2adbb | 5691 | have_caching_bg = false; |
80eb234a | 5692 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
5693 | list_for_each_entry(block_group, &space_info->block_groups[index], |
5694 | list) { | |
6226cb0a | 5695 | u64 offset; |
817d52f8 | 5696 | int cached; |
8a1413a2 | 5697 | |
274bd4fb | 5698 | used_block_group = block_group; |
11dfe35a | 5699 | btrfs_get_block_group(block_group); |
2552d17e | 5700 | search_start = block_group->key.objectid; |
42e70e7a | 5701 | |
83a50de9 CM |
5702 | /* |
5703 | * this can happen if we end up cycling through all the | |
5704 | * raid types, but we want to make sure we only allocate | |
5705 | * for the proper type. | |
5706 | */ | |
5707 | if (!block_group_bits(block_group, data)) { | |
5708 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
5709 | BTRFS_BLOCK_GROUP_RAID1 | | |
5710 | BTRFS_BLOCK_GROUP_RAID10; | |
5711 | ||
5712 | /* | |
5713 | * if they asked for extra copies and this block group | |
5714 | * doesn't provide them, bail. This does allow us to | |
5715 | * fill raid0 from raid1. | |
5716 | */ | |
5717 | if ((data & extra) && !(block_group->flags & extra)) | |
5718 | goto loop; | |
5719 | } | |
5720 | ||
2552d17e | 5721 | have_block_group: |
291c7d2f JB |
5722 | cached = block_group_cache_done(block_group); |
5723 | if (unlikely(!cached)) { | |
291c7d2f | 5724 | found_uncached_bg = true; |
f6373bf3 | 5725 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
5726 | BUG_ON(ret < 0); |
5727 | ret = 0; | |
817d52f8 JB |
5728 | } |
5729 | ||
ea6a478e | 5730 | if (unlikely(block_group->ro)) |
2552d17e | 5731 | goto loop; |
0f9dd46c | 5732 | |
0a24325e | 5733 | /* |
062c05c4 AO |
5734 | * Ok we want to try and use the cluster allocator, so |
5735 | * lets look there | |
0a24325e | 5736 | */ |
062c05c4 | 5737 | if (last_ptr) { |
fa9c0d79 CM |
5738 | /* |
5739 | * the refill lock keeps out other | |
5740 | * people trying to start a new cluster | |
5741 | */ | |
5742 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
5743 | used_block_group = last_ptr->block_group; |
5744 | if (used_block_group != block_group && | |
5745 | (!used_block_group || | |
5746 | used_block_group->ro || | |
5747 | !block_group_bits(used_block_group, data))) { | |
5748 | used_block_group = block_group; | |
44fb5511 | 5749 | goto refill_cluster; |
274bd4fb AO |
5750 | } |
5751 | ||
5752 | if (used_block_group != block_group) | |
5753 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 5754 | |
274bd4fb AO |
5755 | offset = btrfs_alloc_from_cluster(used_block_group, |
5756 | last_ptr, num_bytes, used_block_group->key.objectid); | |
fa9c0d79 CM |
5757 | if (offset) { |
5758 | /* we have a block, we're done */ | |
5759 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5760 | trace_btrfs_reserve_extent_cluster(root, |
5761 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
5762 | goto checks; |
5763 | } | |
5764 | ||
274bd4fb AO |
5765 | WARN_ON(last_ptr->block_group != used_block_group); |
5766 | if (used_block_group != block_group) { | |
5767 | btrfs_put_block_group(used_block_group); | |
5768 | used_block_group = block_group; | |
fa9c0d79 | 5769 | } |
44fb5511 | 5770 | refill_cluster: |
274bd4fb | 5771 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
5772 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
5773 | * set up a new clusters, so lets just skip it | |
5774 | * and let the allocator find whatever block | |
5775 | * it can find. If we reach this point, we | |
5776 | * will have tried the cluster allocator | |
5777 | * plenty of times and not have found | |
5778 | * anything, so we are likely way too | |
5779 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
5780 | * anything. |
5781 | * | |
5782 | * However, if the cluster is taken from the | |
5783 | * current block group, release the cluster | |
5784 | * first, so that we stand a better chance of | |
5785 | * succeeding in the unclustered | |
5786 | * allocation. */ | |
5787 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
5788 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
5789 | spin_unlock(&last_ptr->refill_lock); |
5790 | goto unclustered_alloc; | |
5791 | } | |
5792 | ||
fa9c0d79 CM |
5793 | /* |
5794 | * this cluster didn't work out, free it and | |
5795 | * start over | |
5796 | */ | |
5797 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5798 | ||
a5f6f719 AO |
5799 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
5800 | spin_unlock(&last_ptr->refill_lock); | |
5801 | goto unclustered_alloc; | |
5802 | } | |
5803 | ||
fa9c0d79 | 5804 | /* allocate a cluster in this block group */ |
451d7585 | 5805 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 5806 | block_group, last_ptr, |
1b22bad7 | 5807 | search_start, num_bytes, |
fa9c0d79 CM |
5808 | empty_cluster + empty_size); |
5809 | if (ret == 0) { | |
5810 | /* | |
5811 | * now pull our allocation out of this | |
5812 | * cluster | |
5813 | */ | |
5814 | offset = btrfs_alloc_from_cluster(block_group, | |
5815 | last_ptr, num_bytes, | |
5816 | search_start); | |
5817 | if (offset) { | |
5818 | /* we found one, proceed */ | |
5819 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5820 | trace_btrfs_reserve_extent_cluster(root, |
5821 | block_group, search_start, | |
5822 | num_bytes); | |
fa9c0d79 CM |
5823 | goto checks; |
5824 | } | |
0a24325e JB |
5825 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5826 | && !failed_cluster_refill) { | |
817d52f8 JB |
5827 | spin_unlock(&last_ptr->refill_lock); |
5828 | ||
0a24325e | 5829 | failed_cluster_refill = true; |
817d52f8 JB |
5830 | wait_block_group_cache_progress(block_group, |
5831 | num_bytes + empty_cluster + empty_size); | |
5832 | goto have_block_group; | |
fa9c0d79 | 5833 | } |
817d52f8 | 5834 | |
fa9c0d79 CM |
5835 | /* |
5836 | * at this point we either didn't find a cluster | |
5837 | * or we weren't able to allocate a block from our | |
5838 | * cluster. Free the cluster we've been trying | |
5839 | * to use, and go to the next block group | |
5840 | */ | |
0a24325e | 5841 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5842 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5843 | goto loop; |
fa9c0d79 CM |
5844 | } |
5845 | ||
062c05c4 | 5846 | unclustered_alloc: |
a5f6f719 AO |
5847 | spin_lock(&block_group->free_space_ctl->tree_lock); |
5848 | if (cached && | |
5849 | block_group->free_space_ctl->free_space < | |
5850 | num_bytes + empty_cluster + empty_size) { | |
5851 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5852 | goto loop; | |
5853 | } | |
5854 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5855 | ||
6226cb0a JB |
5856 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5857 | num_bytes, empty_size); | |
1cdda9b8 JB |
5858 | /* |
5859 | * If we didn't find a chunk, and we haven't failed on this | |
5860 | * block group before, and this block group is in the middle of | |
5861 | * caching and we are ok with waiting, then go ahead and wait | |
5862 | * for progress to be made, and set failed_alloc to true. | |
5863 | * | |
5864 | * If failed_alloc is true then we've already waited on this | |
5865 | * block group once and should move on to the next block group. | |
5866 | */ | |
5867 | if (!offset && !failed_alloc && !cached && | |
5868 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5869 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5870 | num_bytes + empty_size); |
5871 | failed_alloc = true; | |
817d52f8 | 5872 | goto have_block_group; |
1cdda9b8 | 5873 | } else if (!offset) { |
60d2adbb MX |
5874 | if (!cached) |
5875 | have_caching_bg = true; | |
1cdda9b8 | 5876 | goto loop; |
817d52f8 | 5877 | } |
fa9c0d79 | 5878 | checks: |
6226cb0a | 5879 | search_start = stripe_align(root, offset); |
25179201 | 5880 | |
2552d17e JB |
5881 | /* move on to the next group */ |
5882 | if (search_start + num_bytes > | |
274bd4fb AO |
5883 | used_block_group->key.objectid + used_block_group->key.offset) { |
5884 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 5885 | goto loop; |
6226cb0a | 5886 | } |
f5a31e16 | 5887 | |
f0486c68 | 5888 | if (offset < search_start) |
274bd4fb | 5889 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
5890 | search_start - offset); |
5891 | BUG_ON(offset > search_start); | |
2552d17e | 5892 | |
274bd4fb | 5893 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 5894 | alloc_type); |
f0486c68 | 5895 | if (ret == -EAGAIN) { |
274bd4fb | 5896 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 5897 | goto loop; |
0f9dd46c | 5898 | } |
0b86a832 | 5899 | |
f0486c68 | 5900 | /* we are all good, lets return */ |
2552d17e JB |
5901 | ins->objectid = search_start; |
5902 | ins->offset = num_bytes; | |
d2fb3437 | 5903 | |
3f7de037 JB |
5904 | trace_btrfs_reserve_extent(orig_root, block_group, |
5905 | search_start, num_bytes); | |
274bd4fb AO |
5906 | if (used_block_group != block_group) |
5907 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 5908 | btrfs_put_block_group(block_group); |
2552d17e JB |
5909 | break; |
5910 | loop: | |
0a24325e | 5911 | failed_cluster_refill = false; |
1cdda9b8 | 5912 | failed_alloc = false; |
b742bb82 | 5913 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
5914 | if (used_block_group != block_group) |
5915 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 5916 | btrfs_put_block_group(block_group); |
2552d17e JB |
5917 | } |
5918 | up_read(&space_info->groups_sem); | |
5919 | ||
60d2adbb MX |
5920 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
5921 | goto search; | |
5922 | ||
b742bb82 YZ |
5923 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5924 | goto search; | |
5925 | ||
285ff5af | 5926 | /* |
ccf0e725 JB |
5927 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
5928 | * caching kthreads as we move along | |
817d52f8 JB |
5929 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5930 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5931 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5932 | * again | |
fa9c0d79 | 5933 | */ |
723bda20 | 5934 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 5935 | index = 0; |
723bda20 | 5936 | loop++; |
817d52f8 | 5937 | if (loop == LOOP_ALLOC_CHUNK) { |
698d0082 | 5938 | ret = do_chunk_alloc(trans, root, data, |
ea658bad JB |
5939 | CHUNK_ALLOC_FORCE); |
5940 | /* | |
5941 | * Do not bail out on ENOSPC since we | |
5942 | * can do more things. | |
5943 | */ | |
5944 | if (ret < 0 && ret != -ENOSPC) { | |
5945 | btrfs_abort_transaction(trans, | |
5946 | root, ret); | |
5947 | goto out; | |
723bda20 | 5948 | } |
2552d17e JB |
5949 | } |
5950 | ||
723bda20 JB |
5951 | if (loop == LOOP_NO_EMPTY_SIZE) { |
5952 | empty_size = 0; | |
5953 | empty_cluster = 0; | |
fa9c0d79 | 5954 | } |
723bda20 JB |
5955 | |
5956 | goto search; | |
2552d17e JB |
5957 | } else if (!ins->objectid) { |
5958 | ret = -ENOSPC; | |
d82a6f1d | 5959 | } else if (ins->objectid) { |
80eb234a | 5960 | ret = 0; |
be744175 | 5961 | } |
79787eaa | 5962 | out: |
be744175 | 5963 | |
0f70abe2 | 5964 | return ret; |
fec577fb | 5965 | } |
ec44a35c | 5966 | |
9ed74f2d JB |
5967 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5968 | int dump_block_groups) | |
0f9dd46c JB |
5969 | { |
5970 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5971 | int index = 0; |
0f9dd46c | 5972 | |
9ed74f2d | 5973 | spin_lock(&info->lock); |
fb25e914 JB |
5974 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
5975 | (unsigned long long)info->flags, | |
d397712b | 5976 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 5977 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5978 | info->bytes_readonly), |
d397712b | 5979 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5980 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5981 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5982 | (unsigned long long)info->total_bytes, |
8929ecfa | 5983 | (unsigned long long)info->bytes_used, |
21380931 | 5984 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5985 | (unsigned long long)info->bytes_reserved, |
21380931 | 5986 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5987 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5988 | spin_unlock(&info->lock); |
5989 | ||
5990 | if (!dump_block_groups) | |
5991 | return; | |
0f9dd46c | 5992 | |
80eb234a | 5993 | down_read(&info->groups_sem); |
b742bb82 YZ |
5994 | again: |
5995 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5996 | spin_lock(&cache->lock); |
799ffc3c | 5997 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n", |
d397712b CM |
5998 | (unsigned long long)cache->key.objectid, |
5999 | (unsigned long long)cache->key.offset, | |
6000 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
6001 | (unsigned long long)cache->pinned, | |
799ffc3c LB |
6002 | (unsigned long long)cache->reserved, |
6003 | cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
6004 | btrfs_dump_free_space(cache, bytes); |
6005 | spin_unlock(&cache->lock); | |
6006 | } | |
b742bb82 YZ |
6007 | if (++index < BTRFS_NR_RAID_TYPES) |
6008 | goto again; | |
80eb234a | 6009 | up_read(&info->groups_sem); |
0f9dd46c | 6010 | } |
e8569813 | 6011 | |
11833d66 YZ |
6012 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
6013 | struct btrfs_root *root, | |
6014 | u64 num_bytes, u64 min_alloc_size, | |
6015 | u64 empty_size, u64 hint_byte, | |
81c9ad23 | 6016 | struct btrfs_key *ins, u64 data) |
fec577fb | 6017 | { |
9e622d6b | 6018 | bool final_tried = false; |
fec577fb | 6019 | int ret; |
925baedd | 6020 | |
6a63209f | 6021 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 6022 | again: |
db94535d CM |
6023 | WARN_ON(num_bytes < root->sectorsize); |
6024 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
81c9ad23 | 6025 | hint_byte, ins, data); |
3b951516 | 6026 | |
9e622d6b MX |
6027 | if (ret == -ENOSPC) { |
6028 | if (!final_tried) { | |
6029 | num_bytes = num_bytes >> 1; | |
6030 | num_bytes = num_bytes & ~(root->sectorsize - 1); | |
6031 | num_bytes = max(num_bytes, min_alloc_size); | |
9e622d6b MX |
6032 | if (num_bytes == min_alloc_size) |
6033 | final_tried = true; | |
6034 | goto again; | |
6035 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
6036 | struct btrfs_space_info *sinfo; | |
6037 | ||
6038 | sinfo = __find_space_info(root->fs_info, data); | |
6039 | printk(KERN_ERR "btrfs allocation failed flags %llu, " | |
6040 | "wanted %llu\n", (unsigned long long)data, | |
6041 | (unsigned long long)num_bytes); | |
53804280 JM |
6042 | if (sinfo) |
6043 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 6044 | } |
925baedd | 6045 | } |
0f9dd46c | 6046 | |
1abe9b8a | 6047 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
6048 | ||
0f9dd46c | 6049 | return ret; |
e6dcd2dc CM |
6050 | } |
6051 | ||
e688b725 CM |
6052 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
6053 | u64 start, u64 len, int pin) | |
65b51a00 | 6054 | { |
0f9dd46c | 6055 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 6056 | int ret = 0; |
0f9dd46c | 6057 | |
0f9dd46c JB |
6058 | cache = btrfs_lookup_block_group(root->fs_info, start); |
6059 | if (!cache) { | |
d397712b CM |
6060 | printk(KERN_ERR "Unable to find block group for %llu\n", |
6061 | (unsigned long long)start); | |
0f9dd46c JB |
6062 | return -ENOSPC; |
6063 | } | |
1f3c79a2 | 6064 | |
5378e607 LD |
6065 | if (btrfs_test_opt(root, DISCARD)) |
6066 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 6067 | |
e688b725 CM |
6068 | if (pin) |
6069 | pin_down_extent(root, cache, start, len, 1); | |
6070 | else { | |
6071 | btrfs_add_free_space(cache, start, len); | |
6072 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
6073 | } | |
fa9c0d79 | 6074 | btrfs_put_block_group(cache); |
817d52f8 | 6075 | |
1abe9b8a | 6076 | trace_btrfs_reserved_extent_free(root, start, len); |
6077 | ||
e6dcd2dc CM |
6078 | return ret; |
6079 | } | |
6080 | ||
e688b725 CM |
6081 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
6082 | u64 start, u64 len) | |
6083 | { | |
6084 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
6085 | } | |
6086 | ||
6087 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
6088 | u64 start, u64 len) | |
6089 | { | |
6090 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
6091 | } | |
6092 | ||
5d4f98a2 YZ |
6093 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
6094 | struct btrfs_root *root, | |
6095 | u64 parent, u64 root_objectid, | |
6096 | u64 flags, u64 owner, u64 offset, | |
6097 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
6098 | { |
6099 | int ret; | |
5d4f98a2 | 6100 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 6101 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 6102 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 6103 | struct btrfs_path *path; |
5d4f98a2 YZ |
6104 | struct extent_buffer *leaf; |
6105 | int type; | |
6106 | u32 size; | |
26b8003f | 6107 | |
5d4f98a2 YZ |
6108 | if (parent > 0) |
6109 | type = BTRFS_SHARED_DATA_REF_KEY; | |
6110 | else | |
6111 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 6112 | |
5d4f98a2 | 6113 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
6114 | |
6115 | path = btrfs_alloc_path(); | |
db5b493a TI |
6116 | if (!path) |
6117 | return -ENOMEM; | |
47e4bb98 | 6118 | |
b9473439 | 6119 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6120 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
6121 | ins, size); | |
79787eaa JM |
6122 | if (ret) { |
6123 | btrfs_free_path(path); | |
6124 | return ret; | |
6125 | } | |
0f9dd46c | 6126 | |
5d4f98a2 YZ |
6127 | leaf = path->nodes[0]; |
6128 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 6129 | struct btrfs_extent_item); |
5d4f98a2 YZ |
6130 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
6131 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6132 | btrfs_set_extent_flags(leaf, extent_item, | |
6133 | flags | BTRFS_EXTENT_FLAG_DATA); | |
6134 | ||
6135 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6136 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
6137 | if (parent > 0) { | |
6138 | struct btrfs_shared_data_ref *ref; | |
6139 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
6140 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6141 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
6142 | } else { | |
6143 | struct btrfs_extent_data_ref *ref; | |
6144 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
6145 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
6146 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
6147 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
6148 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
6149 | } | |
47e4bb98 CM |
6150 | |
6151 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 6152 | btrfs_free_path(path); |
f510cfec | 6153 | |
c53d613e | 6154 | ret = update_block_group(root, ins->objectid, ins->offset, 1); |
79787eaa | 6155 | if (ret) { /* -ENOENT, logic error */ |
d397712b CM |
6156 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6157 | "%llu\n", (unsigned long long)ins->objectid, | |
6158 | (unsigned long long)ins->offset); | |
f5947066 CM |
6159 | BUG(); |
6160 | } | |
e6dcd2dc CM |
6161 | return ret; |
6162 | } | |
6163 | ||
5d4f98a2 YZ |
6164 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
6165 | struct btrfs_root *root, | |
6166 | u64 parent, u64 root_objectid, | |
6167 | u64 flags, struct btrfs_disk_key *key, | |
6168 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
6169 | { |
6170 | int ret; | |
5d4f98a2 YZ |
6171 | struct btrfs_fs_info *fs_info = root->fs_info; |
6172 | struct btrfs_extent_item *extent_item; | |
6173 | struct btrfs_tree_block_info *block_info; | |
6174 | struct btrfs_extent_inline_ref *iref; | |
6175 | struct btrfs_path *path; | |
6176 | struct extent_buffer *leaf; | |
6177 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 6178 | |
5d4f98a2 | 6179 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6180 | if (!path) |
6181 | return -ENOMEM; | |
56bec294 | 6182 | |
5d4f98a2 YZ |
6183 | path->leave_spinning = 1; |
6184 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
6185 | ins, size); | |
79787eaa JM |
6186 | if (ret) { |
6187 | btrfs_free_path(path); | |
6188 | return ret; | |
6189 | } | |
5d4f98a2 YZ |
6190 | |
6191 | leaf = path->nodes[0]; | |
6192 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
6193 | struct btrfs_extent_item); | |
6194 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
6195 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6196 | btrfs_set_extent_flags(leaf, extent_item, | |
6197 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
6198 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
6199 | ||
6200 | btrfs_set_tree_block_key(leaf, block_info, key); | |
6201 | btrfs_set_tree_block_level(leaf, block_info, level); | |
6202 | ||
6203 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
6204 | if (parent > 0) { | |
6205 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
6206 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6207 | BTRFS_SHARED_BLOCK_REF_KEY); | |
6208 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6209 | } else { | |
6210 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6211 | BTRFS_TREE_BLOCK_REF_KEY); | |
6212 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
6213 | } | |
6214 | ||
6215 | btrfs_mark_buffer_dirty(leaf); | |
6216 | btrfs_free_path(path); | |
6217 | ||
c53d613e | 6218 | ret = update_block_group(root, ins->objectid, ins->offset, 1); |
79787eaa | 6219 | if (ret) { /* -ENOENT, logic error */ |
5d4f98a2 YZ |
6220 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6221 | "%llu\n", (unsigned long long)ins->objectid, | |
6222 | (unsigned long long)ins->offset); | |
6223 | BUG(); | |
6224 | } | |
6225 | return ret; | |
6226 | } | |
6227 | ||
6228 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
6229 | struct btrfs_root *root, | |
6230 | u64 root_objectid, u64 owner, | |
6231 | u64 offset, struct btrfs_key *ins) | |
6232 | { | |
6233 | int ret; | |
6234 | ||
6235 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
6236 | ||
66d7e7f0 AJ |
6237 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
6238 | ins->offset, 0, | |
6239 | root_objectid, owner, offset, | |
6240 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
6241 | return ret; |
6242 | } | |
e02119d5 CM |
6243 | |
6244 | /* | |
6245 | * this is used by the tree logging recovery code. It records that | |
6246 | * an extent has been allocated and makes sure to clear the free | |
6247 | * space cache bits as well | |
6248 | */ | |
5d4f98a2 YZ |
6249 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
6250 | struct btrfs_root *root, | |
6251 | u64 root_objectid, u64 owner, u64 offset, | |
6252 | struct btrfs_key *ins) | |
e02119d5 CM |
6253 | { |
6254 | int ret; | |
6255 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
6256 | struct btrfs_caching_control *caching_ctl; |
6257 | u64 start = ins->objectid; | |
6258 | u64 num_bytes = ins->offset; | |
e02119d5 | 6259 | |
e02119d5 | 6260 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
f6373bf3 | 6261 | cache_block_group(block_group, 0); |
11833d66 | 6262 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 6263 | |
11833d66 YZ |
6264 | if (!caching_ctl) { |
6265 | BUG_ON(!block_group_cache_done(block_group)); | |
6266 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
79787eaa | 6267 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6268 | } else { |
6269 | mutex_lock(&caching_ctl->mutex); | |
6270 | ||
6271 | if (start >= caching_ctl->progress) { | |
6272 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6273 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6274 | } else if (start + num_bytes <= caching_ctl->progress) { |
6275 | ret = btrfs_remove_free_space(block_group, | |
6276 | start, num_bytes); | |
79787eaa | 6277 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6278 | } else { |
6279 | num_bytes = caching_ctl->progress - start; | |
6280 | ret = btrfs_remove_free_space(block_group, | |
6281 | start, num_bytes); | |
79787eaa | 6282 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6283 | |
6284 | start = caching_ctl->progress; | |
6285 | num_bytes = ins->objectid + ins->offset - | |
6286 | caching_ctl->progress; | |
6287 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6288 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6289 | } |
6290 | ||
6291 | mutex_unlock(&caching_ctl->mutex); | |
6292 | put_caching_control(caching_ctl); | |
6293 | } | |
6294 | ||
fb25e914 JB |
6295 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
6296 | RESERVE_ALLOC_NO_ACCOUNT); | |
79787eaa | 6297 | BUG_ON(ret); /* logic error */ |
fa9c0d79 | 6298 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
6299 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
6300 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
6301 | return ret; |
6302 | } | |
6303 | ||
65b51a00 CM |
6304 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
6305 | struct btrfs_root *root, | |
4008c04a CM |
6306 | u64 bytenr, u32 blocksize, |
6307 | int level) | |
65b51a00 CM |
6308 | { |
6309 | struct extent_buffer *buf; | |
6310 | ||
6311 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6312 | if (!buf) | |
6313 | return ERR_PTR(-ENOMEM); | |
6314 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6315 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6316 | btrfs_tree_lock(buf); |
6317 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6318 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6319 | |
6320 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6321 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6322 | |
d0c803c4 | 6323 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6324 | /* |
6325 | * we allow two log transactions at a time, use different | |
6326 | * EXENT bit to differentiate dirty pages. | |
6327 | */ | |
6328 | if (root->log_transid % 2 == 0) | |
6329 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6330 | buf->start + buf->len - 1, GFP_NOFS); | |
6331 | else | |
6332 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6333 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6334 | } else { |
6335 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6336 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6337 | } |
65b51a00 | 6338 | trans->blocks_used++; |
b4ce94de | 6339 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6340 | return buf; |
6341 | } | |
6342 | ||
f0486c68 YZ |
6343 | static struct btrfs_block_rsv * |
6344 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6345 | struct btrfs_root *root, u32 blocksize) | |
6346 | { | |
6347 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6348 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
6349 | int ret; |
6350 | ||
6351 | block_rsv = get_block_rsv(trans, root); | |
6352 | ||
6353 | if (block_rsv->size == 0) { | |
08e007d2 MX |
6354 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, |
6355 | BTRFS_RESERVE_NO_FLUSH); | |
68a82277 JB |
6356 | /* |
6357 | * If we couldn't reserve metadata bytes try and use some from | |
6358 | * the global reserve. | |
6359 | */ | |
6360 | if (ret && block_rsv != global_rsv) { | |
6361 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6362 | if (!ret) | |
6363 | return global_rsv; | |
f0486c68 | 6364 | return ERR_PTR(ret); |
68a82277 | 6365 | } else if (ret) { |
f0486c68 | 6366 | return ERR_PTR(ret); |
68a82277 | 6367 | } |
f0486c68 YZ |
6368 | return block_rsv; |
6369 | } | |
6370 | ||
6371 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
6372 | if (!ret) | |
6373 | return block_rsv; | |
ca7e70f5 | 6374 | if (ret && !block_rsv->failfast) { |
dff51cd1 DS |
6375 | static DEFINE_RATELIMIT_STATE(_rs, |
6376 | DEFAULT_RATELIMIT_INTERVAL, | |
6377 | /*DEFAULT_RATELIMIT_BURST*/ 2); | |
31b1a2bd JL |
6378 | if (__ratelimit(&_rs)) |
6379 | WARN(1, KERN_DEBUG "btrfs: block rsv returned %d\n", | |
6380 | ret); | |
08e007d2 MX |
6381 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, |
6382 | BTRFS_RESERVE_NO_FLUSH); | |
68a82277 | 6383 | if (!ret) { |
68a82277 JB |
6384 | return block_rsv; |
6385 | } else if (ret && block_rsv != global_rsv) { | |
6386 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6387 | if (!ret) | |
6388 | return global_rsv; | |
6389 | } | |
6390 | } | |
f0486c68 | 6391 | |
f0486c68 YZ |
6392 | return ERR_PTR(-ENOSPC); |
6393 | } | |
6394 | ||
8c2a3ca2 JB |
6395 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6396 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6397 | { |
6398 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6399 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6400 | } |
6401 | ||
fec577fb | 6402 | /* |
f0486c68 YZ |
6403 | * finds a free extent and does all the dirty work required for allocation |
6404 | * returns the key for the extent through ins, and a tree buffer for | |
6405 | * the first block of the extent through buf. | |
6406 | * | |
fec577fb CM |
6407 | * returns the tree buffer or NULL. |
6408 | */ | |
5f39d397 | 6409 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6410 | struct btrfs_root *root, u32 blocksize, |
6411 | u64 parent, u64 root_objectid, | |
6412 | struct btrfs_disk_key *key, int level, | |
5581a51a | 6413 | u64 hint, u64 empty_size) |
fec577fb | 6414 | { |
e2fa7227 | 6415 | struct btrfs_key ins; |
f0486c68 | 6416 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6417 | struct extent_buffer *buf; |
f0486c68 YZ |
6418 | u64 flags = 0; |
6419 | int ret; | |
6420 | ||
fec577fb | 6421 | |
f0486c68 YZ |
6422 | block_rsv = use_block_rsv(trans, root, blocksize); |
6423 | if (IS_ERR(block_rsv)) | |
6424 | return ERR_CAST(block_rsv); | |
6425 | ||
6426 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
81c9ad23 | 6427 | empty_size, hint, &ins, 0); |
fec577fb | 6428 | if (ret) { |
8c2a3ca2 | 6429 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6430 | return ERR_PTR(ret); |
fec577fb | 6431 | } |
55c69072 | 6432 | |
4008c04a CM |
6433 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6434 | blocksize, level); | |
79787eaa | 6435 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
f0486c68 YZ |
6436 | |
6437 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6438 | if (parent == 0) | |
6439 | parent = ins.objectid; | |
6440 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6441 | } else | |
6442 | BUG_ON(parent > 0); | |
6443 | ||
6444 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6445 | struct btrfs_delayed_extent_op *extent_op; | |
78a6184a | 6446 | extent_op = btrfs_alloc_delayed_extent_op(); |
79787eaa | 6447 | BUG_ON(!extent_op); /* -ENOMEM */ |
f0486c68 YZ |
6448 | if (key) |
6449 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6450 | else | |
6451 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6452 | extent_op->flags_to_set = flags; | |
6453 | extent_op->update_key = 1; | |
6454 | extent_op->update_flags = 1; | |
6455 | extent_op->is_data = 0; | |
6456 | ||
66d7e7f0 AJ |
6457 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6458 | ins.objectid, | |
f0486c68 YZ |
6459 | ins.offset, parent, root_objectid, |
6460 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5581a51a | 6461 | extent_op, 0); |
79787eaa | 6462 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 | 6463 | } |
fec577fb CM |
6464 | return buf; |
6465 | } | |
a28ec197 | 6466 | |
2c47e605 YZ |
6467 | struct walk_control { |
6468 | u64 refs[BTRFS_MAX_LEVEL]; | |
6469 | u64 flags[BTRFS_MAX_LEVEL]; | |
6470 | struct btrfs_key update_progress; | |
6471 | int stage; | |
6472 | int level; | |
6473 | int shared_level; | |
6474 | int update_ref; | |
6475 | int keep_locks; | |
1c4850e2 YZ |
6476 | int reada_slot; |
6477 | int reada_count; | |
66d7e7f0 | 6478 | int for_reloc; |
2c47e605 YZ |
6479 | }; |
6480 | ||
6481 | #define DROP_REFERENCE 1 | |
6482 | #define UPDATE_BACKREF 2 | |
6483 | ||
1c4850e2 YZ |
6484 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6485 | struct btrfs_root *root, | |
6486 | struct walk_control *wc, | |
6487 | struct btrfs_path *path) | |
6407bf6d | 6488 | { |
1c4850e2 YZ |
6489 | u64 bytenr; |
6490 | u64 generation; | |
6491 | u64 refs; | |
94fcca9f | 6492 | u64 flags; |
5d4f98a2 | 6493 | u32 nritems; |
1c4850e2 YZ |
6494 | u32 blocksize; |
6495 | struct btrfs_key key; | |
6496 | struct extent_buffer *eb; | |
6407bf6d | 6497 | int ret; |
1c4850e2 YZ |
6498 | int slot; |
6499 | int nread = 0; | |
6407bf6d | 6500 | |
1c4850e2 YZ |
6501 | if (path->slots[wc->level] < wc->reada_slot) { |
6502 | wc->reada_count = wc->reada_count * 2 / 3; | |
6503 | wc->reada_count = max(wc->reada_count, 2); | |
6504 | } else { | |
6505 | wc->reada_count = wc->reada_count * 3 / 2; | |
6506 | wc->reada_count = min_t(int, wc->reada_count, | |
6507 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
6508 | } | |
7bb86316 | 6509 | |
1c4850e2 YZ |
6510 | eb = path->nodes[wc->level]; |
6511 | nritems = btrfs_header_nritems(eb); | |
6512 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 6513 | |
1c4850e2 YZ |
6514 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6515 | if (nread >= wc->reada_count) | |
6516 | break; | |
bd56b302 | 6517 | |
2dd3e67b | 6518 | cond_resched(); |
1c4850e2 YZ |
6519 | bytenr = btrfs_node_blockptr(eb, slot); |
6520 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6521 | |
1c4850e2 YZ |
6522 | if (slot == path->slots[wc->level]) |
6523 | goto reada; | |
5d4f98a2 | 6524 | |
1c4850e2 YZ |
6525 | if (wc->stage == UPDATE_BACKREF && |
6526 | generation <= root->root_key.offset) | |
bd56b302 CM |
6527 | continue; |
6528 | ||
94fcca9f YZ |
6529 | /* We don't lock the tree block, it's OK to be racy here */ |
6530 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
6531 | &refs, &flags); | |
79787eaa JM |
6532 | /* We don't care about errors in readahead. */ |
6533 | if (ret < 0) | |
6534 | continue; | |
94fcca9f YZ |
6535 | BUG_ON(refs == 0); |
6536 | ||
1c4850e2 | 6537 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6538 | if (refs == 1) |
6539 | goto reada; | |
bd56b302 | 6540 | |
94fcca9f YZ |
6541 | if (wc->level == 1 && |
6542 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6543 | continue; | |
1c4850e2 YZ |
6544 | if (!wc->update_ref || |
6545 | generation <= root->root_key.offset) | |
6546 | continue; | |
6547 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6548 | ret = btrfs_comp_cpu_keys(&key, | |
6549 | &wc->update_progress); | |
6550 | if (ret < 0) | |
6551 | continue; | |
94fcca9f YZ |
6552 | } else { |
6553 | if (wc->level == 1 && | |
6554 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6555 | continue; | |
6407bf6d | 6556 | } |
1c4850e2 YZ |
6557 | reada: |
6558 | ret = readahead_tree_block(root, bytenr, blocksize, | |
6559 | generation); | |
6560 | if (ret) | |
bd56b302 | 6561 | break; |
1c4850e2 | 6562 | nread++; |
20524f02 | 6563 | } |
1c4850e2 | 6564 | wc->reada_slot = slot; |
20524f02 | 6565 | } |
2c47e605 | 6566 | |
f82d02d9 | 6567 | /* |
2c47e605 YZ |
6568 | * hepler to process tree block while walking down the tree. |
6569 | * | |
2c47e605 YZ |
6570 | * when wc->stage == UPDATE_BACKREF, this function updates |
6571 | * back refs for pointers in the block. | |
6572 | * | |
6573 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 6574 | */ |
2c47e605 | 6575 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 6576 | struct btrfs_root *root, |
2c47e605 | 6577 | struct btrfs_path *path, |
94fcca9f | 6578 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 6579 | { |
2c47e605 YZ |
6580 | int level = wc->level; |
6581 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6582 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6583 | int ret; |
6584 | ||
2c47e605 YZ |
6585 | if (wc->stage == UPDATE_BACKREF && |
6586 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6587 | return 1; | |
f82d02d9 | 6588 | |
2c47e605 YZ |
6589 | /* |
6590 | * when reference count of tree block is 1, it won't increase | |
6591 | * again. once full backref flag is set, we never clear it. | |
6592 | */ | |
94fcca9f YZ |
6593 | if (lookup_info && |
6594 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6595 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6596 | BUG_ON(!path->locks[level]); |
6597 | ret = btrfs_lookup_extent_info(trans, root, | |
6598 | eb->start, eb->len, | |
6599 | &wc->refs[level], | |
6600 | &wc->flags[level]); | |
79787eaa JM |
6601 | BUG_ON(ret == -ENOMEM); |
6602 | if (ret) | |
6603 | return ret; | |
2c47e605 YZ |
6604 | BUG_ON(wc->refs[level] == 0); |
6605 | } | |
5d4f98a2 | 6606 | |
2c47e605 YZ |
6607 | if (wc->stage == DROP_REFERENCE) { |
6608 | if (wc->refs[level] > 1) | |
6609 | return 1; | |
f82d02d9 | 6610 | |
2c47e605 | 6611 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 6612 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6613 | path->locks[level] = 0; |
6614 | } | |
6615 | return 0; | |
6616 | } | |
f82d02d9 | 6617 | |
2c47e605 YZ |
6618 | /* wc->stage == UPDATE_BACKREF */ |
6619 | if (!(wc->flags[level] & flag)) { | |
6620 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 6621 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
79787eaa | 6622 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 6623 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
79787eaa | 6624 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6625 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
6626 | eb->len, flag, 0); | |
79787eaa | 6627 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6628 | wc->flags[level] |= flag; |
6629 | } | |
6630 | ||
6631 | /* | |
6632 | * the block is shared by multiple trees, so it's not good to | |
6633 | * keep the tree lock | |
6634 | */ | |
6635 | if (path->locks[level] && level > 0) { | |
bd681513 | 6636 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6637 | path->locks[level] = 0; |
6638 | } | |
6639 | return 0; | |
6640 | } | |
6641 | ||
1c4850e2 YZ |
6642 | /* |
6643 | * hepler to process tree block pointer. | |
6644 | * | |
6645 | * when wc->stage == DROP_REFERENCE, this function checks | |
6646 | * reference count of the block pointed to. if the block | |
6647 | * is shared and we need update back refs for the subtree | |
6648 | * rooted at the block, this function changes wc->stage to | |
6649 | * UPDATE_BACKREF. if the block is shared and there is no | |
6650 | * need to update back, this function drops the reference | |
6651 | * to the block. | |
6652 | * | |
6653 | * NOTE: return value 1 means we should stop walking down. | |
6654 | */ | |
6655 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
6656 | struct btrfs_root *root, | |
6657 | struct btrfs_path *path, | |
94fcca9f | 6658 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
6659 | { |
6660 | u64 bytenr; | |
6661 | u64 generation; | |
6662 | u64 parent; | |
6663 | u32 blocksize; | |
6664 | struct btrfs_key key; | |
6665 | struct extent_buffer *next; | |
6666 | int level = wc->level; | |
6667 | int reada = 0; | |
6668 | int ret = 0; | |
6669 | ||
6670 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
6671 | path->slots[level]); | |
6672 | /* | |
6673 | * if the lower level block was created before the snapshot | |
6674 | * was created, we know there is no need to update back refs | |
6675 | * for the subtree | |
6676 | */ | |
6677 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
6678 | generation <= root->root_key.offset) { |
6679 | *lookup_info = 1; | |
1c4850e2 | 6680 | return 1; |
94fcca9f | 6681 | } |
1c4850e2 YZ |
6682 | |
6683 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
6684 | blocksize = btrfs_level_size(root, level - 1); | |
6685 | ||
6686 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6687 | if (!next) { | |
6688 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6689 | if (!next) |
6690 | return -ENOMEM; | |
1c4850e2 YZ |
6691 | reada = 1; |
6692 | } | |
6693 | btrfs_tree_lock(next); | |
6694 | btrfs_set_lock_blocking(next); | |
6695 | ||
94fcca9f YZ |
6696 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6697 | &wc->refs[level - 1], | |
6698 | &wc->flags[level - 1]); | |
79787eaa JM |
6699 | if (ret < 0) { |
6700 | btrfs_tree_unlock(next); | |
6701 | return ret; | |
6702 | } | |
6703 | ||
94fcca9f YZ |
6704 | BUG_ON(wc->refs[level - 1] == 0); |
6705 | *lookup_info = 0; | |
1c4850e2 | 6706 | |
94fcca9f | 6707 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6708 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6709 | if (level == 1 && |
6710 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6711 | goto skip; | |
6712 | ||
1c4850e2 YZ |
6713 | if (!wc->update_ref || |
6714 | generation <= root->root_key.offset) | |
6715 | goto skip; | |
6716 | ||
6717 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6718 | path->slots[level]); | |
6719 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6720 | if (ret < 0) | |
6721 | goto skip; | |
6722 | ||
6723 | wc->stage = UPDATE_BACKREF; | |
6724 | wc->shared_level = level - 1; | |
6725 | } | |
94fcca9f YZ |
6726 | } else { |
6727 | if (level == 1 && | |
6728 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6729 | goto skip; | |
1c4850e2 YZ |
6730 | } |
6731 | ||
b9fab919 | 6732 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
6733 | btrfs_tree_unlock(next); |
6734 | free_extent_buffer(next); | |
6735 | next = NULL; | |
94fcca9f | 6736 | *lookup_info = 1; |
1c4850e2 YZ |
6737 | } |
6738 | ||
6739 | if (!next) { | |
6740 | if (reada && level == 1) | |
6741 | reada_walk_down(trans, root, wc, path); | |
6742 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6743 | if (!next) |
6744 | return -EIO; | |
1c4850e2 YZ |
6745 | btrfs_tree_lock(next); |
6746 | btrfs_set_lock_blocking(next); | |
6747 | } | |
6748 | ||
6749 | level--; | |
6750 | BUG_ON(level != btrfs_header_level(next)); | |
6751 | path->nodes[level] = next; | |
6752 | path->slots[level] = 0; | |
bd681513 | 6753 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
6754 | wc->level = level; |
6755 | if (wc->level == 1) | |
6756 | wc->reada_slot = 0; | |
6757 | return 0; | |
6758 | skip: | |
6759 | wc->refs[level - 1] = 0; | |
6760 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6761 | if (wc->stage == DROP_REFERENCE) { |
6762 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6763 | parent = path->nodes[level]->start; | |
6764 | } else { | |
6765 | BUG_ON(root->root_key.objectid != | |
6766 | btrfs_header_owner(path->nodes[level])); | |
6767 | parent = 0; | |
6768 | } | |
1c4850e2 | 6769 | |
94fcca9f | 6770 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 6771 | root->root_key.objectid, level - 1, 0, 0); |
79787eaa | 6772 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 6773 | } |
1c4850e2 YZ |
6774 | btrfs_tree_unlock(next); |
6775 | free_extent_buffer(next); | |
94fcca9f | 6776 | *lookup_info = 1; |
1c4850e2 YZ |
6777 | return 1; |
6778 | } | |
6779 | ||
2c47e605 YZ |
6780 | /* |
6781 | * hepler to process tree block while walking up the tree. | |
6782 | * | |
6783 | * when wc->stage == DROP_REFERENCE, this function drops | |
6784 | * reference count on the block. | |
6785 | * | |
6786 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6787 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6788 | * to UPDATE_BACKREF previously while processing the block. | |
6789 | * | |
6790 | * NOTE: return value 1 means we should stop walking up. | |
6791 | */ | |
6792 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6793 | struct btrfs_root *root, | |
6794 | struct btrfs_path *path, | |
6795 | struct walk_control *wc) | |
6796 | { | |
f0486c68 | 6797 | int ret; |
2c47e605 YZ |
6798 | int level = wc->level; |
6799 | struct extent_buffer *eb = path->nodes[level]; | |
6800 | u64 parent = 0; | |
6801 | ||
6802 | if (wc->stage == UPDATE_BACKREF) { | |
6803 | BUG_ON(wc->shared_level < level); | |
6804 | if (level < wc->shared_level) | |
6805 | goto out; | |
6806 | ||
2c47e605 YZ |
6807 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6808 | if (ret > 0) | |
6809 | wc->update_ref = 0; | |
6810 | ||
6811 | wc->stage = DROP_REFERENCE; | |
6812 | wc->shared_level = -1; | |
6813 | path->slots[level] = 0; | |
6814 | ||
6815 | /* | |
6816 | * check reference count again if the block isn't locked. | |
6817 | * we should start walking down the tree again if reference | |
6818 | * count is one. | |
6819 | */ | |
6820 | if (!path->locks[level]) { | |
6821 | BUG_ON(level == 0); | |
6822 | btrfs_tree_lock(eb); | |
6823 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6824 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6825 | |
6826 | ret = btrfs_lookup_extent_info(trans, root, | |
6827 | eb->start, eb->len, | |
6828 | &wc->refs[level], | |
6829 | &wc->flags[level]); | |
79787eaa JM |
6830 | if (ret < 0) { |
6831 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 6832 | path->locks[level] = 0; |
79787eaa JM |
6833 | return ret; |
6834 | } | |
2c47e605 YZ |
6835 | BUG_ON(wc->refs[level] == 0); |
6836 | if (wc->refs[level] == 1) { | |
bd681513 | 6837 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 6838 | path->locks[level] = 0; |
2c47e605 YZ |
6839 | return 1; |
6840 | } | |
f82d02d9 | 6841 | } |
2c47e605 | 6842 | } |
f82d02d9 | 6843 | |
2c47e605 YZ |
6844 | /* wc->stage == DROP_REFERENCE */ |
6845 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6846 | |
2c47e605 YZ |
6847 | if (wc->refs[level] == 1) { |
6848 | if (level == 0) { | |
6849 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
6850 | ret = btrfs_dec_ref(trans, root, eb, 1, |
6851 | wc->for_reloc); | |
2c47e605 | 6852 | else |
66d7e7f0 AJ |
6853 | ret = btrfs_dec_ref(trans, root, eb, 0, |
6854 | wc->for_reloc); | |
79787eaa | 6855 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6856 | } |
6857 | /* make block locked assertion in clean_tree_block happy */ | |
6858 | if (!path->locks[level] && | |
6859 | btrfs_header_generation(eb) == trans->transid) { | |
6860 | btrfs_tree_lock(eb); | |
6861 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6862 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6863 | } |
6864 | clean_tree_block(trans, root, eb); | |
6865 | } | |
6866 | ||
6867 | if (eb == root->node) { | |
6868 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6869 | parent = eb->start; | |
6870 | else | |
6871 | BUG_ON(root->root_key.objectid != | |
6872 | btrfs_header_owner(eb)); | |
6873 | } else { | |
6874 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6875 | parent = path->nodes[level + 1]->start; | |
6876 | else | |
6877 | BUG_ON(root->root_key.objectid != | |
6878 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6879 | } |
f82d02d9 | 6880 | |
5581a51a | 6881 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
6882 | out: |
6883 | wc->refs[level] = 0; | |
6884 | wc->flags[level] = 0; | |
f0486c68 | 6885 | return 0; |
2c47e605 YZ |
6886 | } |
6887 | ||
6888 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6889 | struct btrfs_root *root, | |
6890 | struct btrfs_path *path, | |
6891 | struct walk_control *wc) | |
6892 | { | |
2c47e605 | 6893 | int level = wc->level; |
94fcca9f | 6894 | int lookup_info = 1; |
2c47e605 YZ |
6895 | int ret; |
6896 | ||
6897 | while (level >= 0) { | |
94fcca9f | 6898 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6899 | if (ret > 0) |
6900 | break; | |
6901 | ||
6902 | if (level == 0) | |
6903 | break; | |
6904 | ||
7a7965f8 YZ |
6905 | if (path->slots[level] >= |
6906 | btrfs_header_nritems(path->nodes[level])) | |
6907 | break; | |
6908 | ||
94fcca9f | 6909 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6910 | if (ret > 0) { |
6911 | path->slots[level]++; | |
6912 | continue; | |
90d2c51d MX |
6913 | } else if (ret < 0) |
6914 | return ret; | |
1c4850e2 | 6915 | level = wc->level; |
f82d02d9 | 6916 | } |
f82d02d9 YZ |
6917 | return 0; |
6918 | } | |
6919 | ||
d397712b | 6920 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6921 | struct btrfs_root *root, |
f82d02d9 | 6922 | struct btrfs_path *path, |
2c47e605 | 6923 | struct walk_control *wc, int max_level) |
20524f02 | 6924 | { |
2c47e605 | 6925 | int level = wc->level; |
20524f02 | 6926 | int ret; |
9f3a7427 | 6927 | |
2c47e605 YZ |
6928 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6929 | while (level < max_level && path->nodes[level]) { | |
6930 | wc->level = level; | |
6931 | if (path->slots[level] + 1 < | |
6932 | btrfs_header_nritems(path->nodes[level])) { | |
6933 | path->slots[level]++; | |
20524f02 CM |
6934 | return 0; |
6935 | } else { | |
2c47e605 YZ |
6936 | ret = walk_up_proc(trans, root, path, wc); |
6937 | if (ret > 0) | |
6938 | return 0; | |
bd56b302 | 6939 | |
2c47e605 | 6940 | if (path->locks[level]) { |
bd681513 CM |
6941 | btrfs_tree_unlock_rw(path->nodes[level], |
6942 | path->locks[level]); | |
2c47e605 | 6943 | path->locks[level] = 0; |
f82d02d9 | 6944 | } |
2c47e605 YZ |
6945 | free_extent_buffer(path->nodes[level]); |
6946 | path->nodes[level] = NULL; | |
6947 | level++; | |
20524f02 CM |
6948 | } |
6949 | } | |
6950 | return 1; | |
6951 | } | |
6952 | ||
9aca1d51 | 6953 | /* |
2c47e605 YZ |
6954 | * drop a subvolume tree. |
6955 | * | |
6956 | * this function traverses the tree freeing any blocks that only | |
6957 | * referenced by the tree. | |
6958 | * | |
6959 | * when a shared tree block is found. this function decreases its | |
6960 | * reference count by one. if update_ref is true, this function | |
6961 | * also make sure backrefs for the shared block and all lower level | |
6962 | * blocks are properly updated. | |
9aca1d51 | 6963 | */ |
2c536799 | 6964 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
6965 | struct btrfs_block_rsv *block_rsv, int update_ref, |
6966 | int for_reloc) | |
20524f02 | 6967 | { |
5caf2a00 | 6968 | struct btrfs_path *path; |
2c47e605 YZ |
6969 | struct btrfs_trans_handle *trans; |
6970 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6971 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6972 | struct walk_control *wc; |
6973 | struct btrfs_key key; | |
6974 | int err = 0; | |
6975 | int ret; | |
6976 | int level; | |
20524f02 | 6977 | |
5caf2a00 | 6978 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
6979 | if (!path) { |
6980 | err = -ENOMEM; | |
6981 | goto out; | |
6982 | } | |
20524f02 | 6983 | |
2c47e605 | 6984 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
6985 | if (!wc) { |
6986 | btrfs_free_path(path); | |
cb1b69f4 TI |
6987 | err = -ENOMEM; |
6988 | goto out; | |
38a1a919 | 6989 | } |
2c47e605 | 6990 | |
a22285a6 | 6991 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
6992 | if (IS_ERR(trans)) { |
6993 | err = PTR_ERR(trans); | |
6994 | goto out_free; | |
6995 | } | |
98d5dc13 | 6996 | |
3fd0a558 YZ |
6997 | if (block_rsv) |
6998 | trans->block_rsv = block_rsv; | |
2c47e605 | 6999 | |
9f3a7427 | 7000 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 7001 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
7002 | path->nodes[level] = btrfs_lock_root_node(root); |
7003 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 7004 | path->slots[level] = 0; |
bd681513 | 7005 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7006 | memset(&wc->update_progress, 0, |
7007 | sizeof(wc->update_progress)); | |
9f3a7427 | 7008 | } else { |
9f3a7427 | 7009 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
7010 | memcpy(&wc->update_progress, &key, |
7011 | sizeof(wc->update_progress)); | |
7012 | ||
6702ed49 | 7013 | level = root_item->drop_level; |
2c47e605 | 7014 | BUG_ON(level == 0); |
6702ed49 | 7015 | path->lowest_level = level; |
2c47e605 YZ |
7016 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
7017 | path->lowest_level = 0; | |
7018 | if (ret < 0) { | |
7019 | err = ret; | |
79787eaa | 7020 | goto out_end_trans; |
9f3a7427 | 7021 | } |
1c4850e2 | 7022 | WARN_ON(ret > 0); |
2c47e605 | 7023 | |
7d9eb12c CM |
7024 | /* |
7025 | * unlock our path, this is safe because only this | |
7026 | * function is allowed to delete this snapshot | |
7027 | */ | |
5d4f98a2 | 7028 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
7029 | |
7030 | level = btrfs_header_level(root->node); | |
7031 | while (1) { | |
7032 | btrfs_tree_lock(path->nodes[level]); | |
7033 | btrfs_set_lock_blocking(path->nodes[level]); | |
7034 | ||
7035 | ret = btrfs_lookup_extent_info(trans, root, | |
7036 | path->nodes[level]->start, | |
7037 | path->nodes[level]->len, | |
7038 | &wc->refs[level], | |
7039 | &wc->flags[level]); | |
79787eaa JM |
7040 | if (ret < 0) { |
7041 | err = ret; | |
7042 | goto out_end_trans; | |
7043 | } | |
2c47e605 YZ |
7044 | BUG_ON(wc->refs[level] == 0); |
7045 | ||
7046 | if (level == root_item->drop_level) | |
7047 | break; | |
7048 | ||
7049 | btrfs_tree_unlock(path->nodes[level]); | |
7050 | WARN_ON(wc->refs[level] != 1); | |
7051 | level--; | |
7052 | } | |
9f3a7427 | 7053 | } |
2c47e605 YZ |
7054 | |
7055 | wc->level = level; | |
7056 | wc->shared_level = -1; | |
7057 | wc->stage = DROP_REFERENCE; | |
7058 | wc->update_ref = update_ref; | |
7059 | wc->keep_locks = 0; | |
66d7e7f0 | 7060 | wc->for_reloc = for_reloc; |
1c4850e2 | 7061 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 7062 | |
d397712b | 7063 | while (1) { |
2c47e605 YZ |
7064 | ret = walk_down_tree(trans, root, path, wc); |
7065 | if (ret < 0) { | |
7066 | err = ret; | |
20524f02 | 7067 | break; |
2c47e605 | 7068 | } |
9aca1d51 | 7069 | |
2c47e605 YZ |
7070 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
7071 | if (ret < 0) { | |
7072 | err = ret; | |
20524f02 | 7073 | break; |
2c47e605 YZ |
7074 | } |
7075 | ||
7076 | if (ret > 0) { | |
7077 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
7078 | break; |
7079 | } | |
2c47e605 YZ |
7080 | |
7081 | if (wc->stage == DROP_REFERENCE) { | |
7082 | level = wc->level; | |
7083 | btrfs_node_key(path->nodes[level], | |
7084 | &root_item->drop_progress, | |
7085 | path->slots[level]); | |
7086 | root_item->drop_level = level; | |
7087 | } | |
7088 | ||
7089 | BUG_ON(wc->level == 0); | |
3fd0a558 | 7090 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
7091 | ret = btrfs_update_root(trans, tree_root, |
7092 | &root->root_key, | |
7093 | root_item); | |
79787eaa JM |
7094 | if (ret) { |
7095 | btrfs_abort_transaction(trans, tree_root, ret); | |
7096 | err = ret; | |
7097 | goto out_end_trans; | |
7098 | } | |
2c47e605 | 7099 | |
3fd0a558 | 7100 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 7101 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7102 | if (IS_ERR(trans)) { |
7103 | err = PTR_ERR(trans); | |
7104 | goto out_free; | |
7105 | } | |
3fd0a558 YZ |
7106 | if (block_rsv) |
7107 | trans->block_rsv = block_rsv; | |
c3e69d58 | 7108 | } |
20524f02 | 7109 | } |
b3b4aa74 | 7110 | btrfs_release_path(path); |
79787eaa JM |
7111 | if (err) |
7112 | goto out_end_trans; | |
2c47e605 YZ |
7113 | |
7114 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
7115 | if (ret) { |
7116 | btrfs_abort_transaction(trans, tree_root, ret); | |
7117 | goto out_end_trans; | |
7118 | } | |
2c47e605 | 7119 | |
76dda93c YZ |
7120 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
7121 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
7122 | NULL, NULL); | |
79787eaa JM |
7123 | if (ret < 0) { |
7124 | btrfs_abort_transaction(trans, tree_root, ret); | |
7125 | err = ret; | |
7126 | goto out_end_trans; | |
7127 | } else if (ret > 0) { | |
84cd948c JB |
7128 | /* if we fail to delete the orphan item this time |
7129 | * around, it'll get picked up the next time. | |
7130 | * | |
7131 | * The most common failure here is just -ENOENT. | |
7132 | */ | |
7133 | btrfs_del_orphan_item(trans, tree_root, | |
7134 | root->root_key.objectid); | |
76dda93c YZ |
7135 | } |
7136 | } | |
7137 | ||
7138 | if (root->in_radix) { | |
7139 | btrfs_free_fs_root(tree_root->fs_info, root); | |
7140 | } else { | |
7141 | free_extent_buffer(root->node); | |
7142 | free_extent_buffer(root->commit_root); | |
7143 | kfree(root); | |
7144 | } | |
79787eaa | 7145 | out_end_trans: |
3fd0a558 | 7146 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 7147 | out_free: |
2c47e605 | 7148 | kfree(wc); |
5caf2a00 | 7149 | btrfs_free_path(path); |
cb1b69f4 TI |
7150 | out: |
7151 | if (err) | |
7152 | btrfs_std_error(root->fs_info, err); | |
2c536799 | 7153 | return err; |
20524f02 | 7154 | } |
9078a3e1 | 7155 | |
2c47e605 YZ |
7156 | /* |
7157 | * drop subtree rooted at tree block 'node'. | |
7158 | * | |
7159 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 7160 | * only used by relocation code |
2c47e605 | 7161 | */ |
f82d02d9 YZ |
7162 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
7163 | struct btrfs_root *root, | |
7164 | struct extent_buffer *node, | |
7165 | struct extent_buffer *parent) | |
7166 | { | |
7167 | struct btrfs_path *path; | |
2c47e605 | 7168 | struct walk_control *wc; |
f82d02d9 YZ |
7169 | int level; |
7170 | int parent_level; | |
7171 | int ret = 0; | |
7172 | int wret; | |
7173 | ||
2c47e605 YZ |
7174 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
7175 | ||
f82d02d9 | 7176 | path = btrfs_alloc_path(); |
db5b493a TI |
7177 | if (!path) |
7178 | return -ENOMEM; | |
f82d02d9 | 7179 | |
2c47e605 | 7180 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
7181 | if (!wc) { |
7182 | btrfs_free_path(path); | |
7183 | return -ENOMEM; | |
7184 | } | |
2c47e605 | 7185 | |
b9447ef8 | 7186 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
7187 | parent_level = btrfs_header_level(parent); |
7188 | extent_buffer_get(parent); | |
7189 | path->nodes[parent_level] = parent; | |
7190 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
7191 | ||
b9447ef8 | 7192 | btrfs_assert_tree_locked(node); |
f82d02d9 | 7193 | level = btrfs_header_level(node); |
f82d02d9 YZ |
7194 | path->nodes[level] = node; |
7195 | path->slots[level] = 0; | |
bd681513 | 7196 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7197 | |
7198 | wc->refs[parent_level] = 1; | |
7199 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7200 | wc->level = level; | |
7201 | wc->shared_level = -1; | |
7202 | wc->stage = DROP_REFERENCE; | |
7203 | wc->update_ref = 0; | |
7204 | wc->keep_locks = 1; | |
66d7e7f0 | 7205 | wc->for_reloc = 1; |
1c4850e2 | 7206 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
7207 | |
7208 | while (1) { | |
2c47e605 YZ |
7209 | wret = walk_down_tree(trans, root, path, wc); |
7210 | if (wret < 0) { | |
f82d02d9 | 7211 | ret = wret; |
f82d02d9 | 7212 | break; |
2c47e605 | 7213 | } |
f82d02d9 | 7214 | |
2c47e605 | 7215 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
7216 | if (wret < 0) |
7217 | ret = wret; | |
7218 | if (wret != 0) | |
7219 | break; | |
7220 | } | |
7221 | ||
2c47e605 | 7222 | kfree(wc); |
f82d02d9 YZ |
7223 | btrfs_free_path(path); |
7224 | return ret; | |
7225 | } | |
7226 | ||
ec44a35c CM |
7227 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7228 | { | |
7229 | u64 num_devices; | |
fc67c450 | 7230 | u64 stripped; |
e4d8ec0f | 7231 | |
fc67c450 ID |
7232 | /* |
7233 | * if restripe for this chunk_type is on pick target profile and | |
7234 | * return, otherwise do the usual balance | |
7235 | */ | |
7236 | stripped = get_restripe_target(root->fs_info, flags); | |
7237 | if (stripped) | |
7238 | return extended_to_chunk(stripped); | |
e4d8ec0f | 7239 | |
cd02dca5 CM |
7240 | /* |
7241 | * we add in the count of missing devices because we want | |
7242 | * to make sure that any RAID levels on a degraded FS | |
7243 | * continue to be honored. | |
7244 | */ | |
7245 | num_devices = root->fs_info->fs_devices->rw_devices + | |
7246 | root->fs_info->fs_devices->missing_devices; | |
7247 | ||
fc67c450 ID |
7248 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
7249 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
7250 | ||
ec44a35c CM |
7251 | if (num_devices == 1) { |
7252 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7253 | stripped = flags & ~stripped; | |
7254 | ||
7255 | /* turn raid0 into single device chunks */ | |
7256 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7257 | return stripped; | |
7258 | ||
7259 | /* turn mirroring into duplication */ | |
7260 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7261 | BTRFS_BLOCK_GROUP_RAID10)) | |
7262 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
7263 | } else { |
7264 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7265 | if (flags & stripped) |
7266 | return flags; | |
7267 | ||
7268 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7269 | stripped = flags & ~stripped; | |
7270 | ||
7271 | /* switch duplicated blocks with raid1 */ | |
7272 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7273 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7274 | ||
e3176ca2 | 7275 | /* this is drive concat, leave it alone */ |
ec44a35c | 7276 | } |
e3176ca2 | 7277 | |
ec44a35c CM |
7278 | return flags; |
7279 | } | |
7280 | ||
199c36ea | 7281 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 7282 | { |
f0486c68 YZ |
7283 | struct btrfs_space_info *sinfo = cache->space_info; |
7284 | u64 num_bytes; | |
199c36ea | 7285 | u64 min_allocable_bytes; |
f0486c68 | 7286 | int ret = -ENOSPC; |
0ef3e66b | 7287 | |
c286ac48 | 7288 | |
199c36ea MX |
7289 | /* |
7290 | * We need some metadata space and system metadata space for | |
7291 | * allocating chunks in some corner cases until we force to set | |
7292 | * it to be readonly. | |
7293 | */ | |
7294 | if ((sinfo->flags & | |
7295 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
7296 | !force) | |
7297 | min_allocable_bytes = 1 * 1024 * 1024; | |
7298 | else | |
7299 | min_allocable_bytes = 0; | |
7300 | ||
f0486c68 YZ |
7301 | spin_lock(&sinfo->lock); |
7302 | spin_lock(&cache->lock); | |
61cfea9b W |
7303 | |
7304 | if (cache->ro) { | |
7305 | ret = 0; | |
7306 | goto out; | |
7307 | } | |
7308 | ||
f0486c68 YZ |
7309 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
7310 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7311 | ||
7312 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
7313 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
7314 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 7315 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
7316 | cache->ro = 1; |
7317 | ret = 0; | |
7318 | } | |
61cfea9b | 7319 | out: |
f0486c68 YZ |
7320 | spin_unlock(&cache->lock); |
7321 | spin_unlock(&sinfo->lock); | |
7322 | return ret; | |
7323 | } | |
7d9eb12c | 7324 | |
f0486c68 YZ |
7325 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
7326 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 7327 | |
f0486c68 YZ |
7328 | { |
7329 | struct btrfs_trans_handle *trans; | |
7330 | u64 alloc_flags; | |
7331 | int ret; | |
7d9eb12c | 7332 | |
f0486c68 | 7333 | BUG_ON(cache->ro); |
0ef3e66b | 7334 | |
ff5714cc | 7335 | trans = btrfs_join_transaction(root); |
79787eaa JM |
7336 | if (IS_ERR(trans)) |
7337 | return PTR_ERR(trans); | |
5d4f98a2 | 7338 | |
f0486c68 | 7339 | alloc_flags = update_block_group_flags(root, cache->flags); |
79787eaa | 7340 | if (alloc_flags != cache->flags) { |
698d0082 | 7341 | ret = do_chunk_alloc(trans, root, alloc_flags, |
79787eaa JM |
7342 | CHUNK_ALLOC_FORCE); |
7343 | if (ret < 0) | |
7344 | goto out; | |
7345 | } | |
5d4f98a2 | 7346 | |
199c36ea | 7347 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7348 | if (!ret) |
7349 | goto out; | |
7350 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 7351 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7352 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
7353 | if (ret < 0) |
7354 | goto out; | |
199c36ea | 7355 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7356 | out: |
7357 | btrfs_end_transaction(trans, root); | |
7358 | return ret; | |
7359 | } | |
5d4f98a2 | 7360 | |
c87f08ca CM |
7361 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7362 | struct btrfs_root *root, u64 type) | |
7363 | { | |
7364 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 7365 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7366 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
7367 | } |
7368 | ||
6d07bcec MX |
7369 | /* |
7370 | * helper to account the unused space of all the readonly block group in the | |
7371 | * list. takes mirrors into account. | |
7372 | */ | |
7373 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7374 | { | |
7375 | struct btrfs_block_group_cache *block_group; | |
7376 | u64 free_bytes = 0; | |
7377 | int factor; | |
7378 | ||
7379 | list_for_each_entry(block_group, groups_list, list) { | |
7380 | spin_lock(&block_group->lock); | |
7381 | ||
7382 | if (!block_group->ro) { | |
7383 | spin_unlock(&block_group->lock); | |
7384 | continue; | |
7385 | } | |
7386 | ||
7387 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7388 | BTRFS_BLOCK_GROUP_RAID10 | | |
7389 | BTRFS_BLOCK_GROUP_DUP)) | |
7390 | factor = 2; | |
7391 | else | |
7392 | factor = 1; | |
7393 | ||
7394 | free_bytes += (block_group->key.offset - | |
7395 | btrfs_block_group_used(&block_group->item)) * | |
7396 | factor; | |
7397 | ||
7398 | spin_unlock(&block_group->lock); | |
7399 | } | |
7400 | ||
7401 | return free_bytes; | |
7402 | } | |
7403 | ||
7404 | /* | |
7405 | * helper to account the unused space of all the readonly block group in the | |
7406 | * space_info. takes mirrors into account. | |
7407 | */ | |
7408 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
7409 | { | |
7410 | int i; | |
7411 | u64 free_bytes = 0; | |
7412 | ||
7413 | spin_lock(&sinfo->lock); | |
7414 | ||
7415 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
7416 | if (!list_empty(&sinfo->block_groups[i])) | |
7417 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
7418 | &sinfo->block_groups[i]); | |
7419 | ||
7420 | spin_unlock(&sinfo->lock); | |
7421 | ||
7422 | return free_bytes; | |
7423 | } | |
7424 | ||
143bede5 | 7425 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
f0486c68 | 7426 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 7427 | { |
f0486c68 YZ |
7428 | struct btrfs_space_info *sinfo = cache->space_info; |
7429 | u64 num_bytes; | |
7430 | ||
7431 | BUG_ON(!cache->ro); | |
7432 | ||
7433 | spin_lock(&sinfo->lock); | |
7434 | spin_lock(&cache->lock); | |
7435 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7436 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7437 | sinfo->bytes_readonly -= num_bytes; | |
7438 | cache->ro = 0; | |
7439 | spin_unlock(&cache->lock); | |
7440 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
7441 | } |
7442 | ||
ba1bf481 JB |
7443 | /* |
7444 | * checks to see if its even possible to relocate this block group. | |
7445 | * | |
7446 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7447 | * ok to go ahead and try. | |
7448 | */ | |
7449 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7450 | { |
ba1bf481 JB |
7451 | struct btrfs_block_group_cache *block_group; |
7452 | struct btrfs_space_info *space_info; | |
7453 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7454 | struct btrfs_device *device; | |
cdcb725c | 7455 | u64 min_free; |
6719db6a JB |
7456 | u64 dev_min = 1; |
7457 | u64 dev_nr = 0; | |
4a5e98f5 | 7458 | u64 target; |
cdcb725c | 7459 | int index; |
ba1bf481 JB |
7460 | int full = 0; |
7461 | int ret = 0; | |
1a40e23b | 7462 | |
ba1bf481 | 7463 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7464 | |
ba1bf481 JB |
7465 | /* odd, couldn't find the block group, leave it alone */ |
7466 | if (!block_group) | |
7467 | return -1; | |
1a40e23b | 7468 | |
cdcb725c | 7469 | min_free = btrfs_block_group_used(&block_group->item); |
7470 | ||
ba1bf481 | 7471 | /* no bytes used, we're good */ |
cdcb725c | 7472 | if (!min_free) |
1a40e23b ZY |
7473 | goto out; |
7474 | ||
ba1bf481 JB |
7475 | space_info = block_group->space_info; |
7476 | spin_lock(&space_info->lock); | |
17d217fe | 7477 | |
ba1bf481 | 7478 | full = space_info->full; |
17d217fe | 7479 | |
ba1bf481 JB |
7480 | /* |
7481 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7482 | * relocate it unless we're able to allocate a new chunk below. |
7483 | * | |
7484 | * Otherwise, we need to make sure we have room in the space to handle | |
7485 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7486 | */ |
7ce618db | 7487 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 7488 | (space_info->bytes_used + space_info->bytes_reserved + |
7489 | space_info->bytes_pinned + space_info->bytes_readonly + | |
7490 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
7491 | spin_unlock(&space_info->lock); |
7492 | goto out; | |
17d217fe | 7493 | } |
ba1bf481 | 7494 | spin_unlock(&space_info->lock); |
ea8c2819 | 7495 | |
ba1bf481 JB |
7496 | /* |
7497 | * ok we don't have enough space, but maybe we have free space on our | |
7498 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
7499 | * alloc devices and guess if we have enough space. if this block |
7500 | * group is going to be restriped, run checks against the target | |
7501 | * profile instead of the current one. | |
ba1bf481 JB |
7502 | */ |
7503 | ret = -1; | |
ea8c2819 | 7504 | |
cdcb725c | 7505 | /* |
7506 | * index: | |
7507 | * 0: raid10 | |
7508 | * 1: raid1 | |
7509 | * 2: dup | |
7510 | * 3: raid0 | |
7511 | * 4: single | |
7512 | */ | |
4a5e98f5 ID |
7513 | target = get_restripe_target(root->fs_info, block_group->flags); |
7514 | if (target) { | |
31e50229 | 7515 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
7516 | } else { |
7517 | /* | |
7518 | * this is just a balance, so if we were marked as full | |
7519 | * we know there is no space for a new chunk | |
7520 | */ | |
7521 | if (full) | |
7522 | goto out; | |
7523 | ||
7524 | index = get_block_group_index(block_group); | |
7525 | } | |
7526 | ||
e6ec716f | 7527 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 7528 | dev_min = 4; |
6719db6a JB |
7529 | /* Divide by 2 */ |
7530 | min_free >>= 1; | |
e6ec716f | 7531 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 7532 | dev_min = 2; |
e6ec716f | 7533 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
7534 | /* Multiply by 2 */ |
7535 | min_free <<= 1; | |
e6ec716f | 7536 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 7537 | dev_min = fs_devices->rw_devices; |
6719db6a | 7538 | do_div(min_free, dev_min); |
cdcb725c | 7539 | } |
7540 | ||
ba1bf481 JB |
7541 | mutex_lock(&root->fs_info->chunk_mutex); |
7542 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 7543 | u64 dev_offset; |
56bec294 | 7544 | |
ba1bf481 JB |
7545 | /* |
7546 | * check to make sure we can actually find a chunk with enough | |
7547 | * space to fit our block group in. | |
7548 | */ | |
63a212ab SB |
7549 | if (device->total_bytes > device->bytes_used + min_free && |
7550 | !device->is_tgtdev_for_dev_replace) { | |
125ccb0a | 7551 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 7552 | &dev_offset, NULL); |
ba1bf481 | 7553 | if (!ret) |
cdcb725c | 7554 | dev_nr++; |
7555 | ||
7556 | if (dev_nr >= dev_min) | |
73e48b27 | 7557 | break; |
cdcb725c | 7558 | |
ba1bf481 | 7559 | ret = -1; |
725c8463 | 7560 | } |
edbd8d4e | 7561 | } |
ba1bf481 | 7562 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7563 | out: |
ba1bf481 | 7564 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7565 | return ret; |
7566 | } | |
7567 | ||
b2950863 CH |
7568 | static int find_first_block_group(struct btrfs_root *root, |
7569 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7570 | { |
925baedd | 7571 | int ret = 0; |
0b86a832 CM |
7572 | struct btrfs_key found_key; |
7573 | struct extent_buffer *leaf; | |
7574 | int slot; | |
edbd8d4e | 7575 | |
0b86a832 CM |
7576 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7577 | if (ret < 0) | |
925baedd CM |
7578 | goto out; |
7579 | ||
d397712b | 7580 | while (1) { |
0b86a832 | 7581 | slot = path->slots[0]; |
edbd8d4e | 7582 | leaf = path->nodes[0]; |
0b86a832 CM |
7583 | if (slot >= btrfs_header_nritems(leaf)) { |
7584 | ret = btrfs_next_leaf(root, path); | |
7585 | if (ret == 0) | |
7586 | continue; | |
7587 | if (ret < 0) | |
925baedd | 7588 | goto out; |
0b86a832 | 7589 | break; |
edbd8d4e | 7590 | } |
0b86a832 | 7591 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7592 | |
0b86a832 | 7593 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7594 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7595 | ret = 0; | |
7596 | goto out; | |
7597 | } | |
0b86a832 | 7598 | path->slots[0]++; |
edbd8d4e | 7599 | } |
925baedd | 7600 | out: |
0b86a832 | 7601 | return ret; |
edbd8d4e CM |
7602 | } |
7603 | ||
0af3d00b JB |
7604 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
7605 | { | |
7606 | struct btrfs_block_group_cache *block_group; | |
7607 | u64 last = 0; | |
7608 | ||
7609 | while (1) { | |
7610 | struct inode *inode; | |
7611 | ||
7612 | block_group = btrfs_lookup_first_block_group(info, last); | |
7613 | while (block_group) { | |
7614 | spin_lock(&block_group->lock); | |
7615 | if (block_group->iref) | |
7616 | break; | |
7617 | spin_unlock(&block_group->lock); | |
7618 | block_group = next_block_group(info->tree_root, | |
7619 | block_group); | |
7620 | } | |
7621 | if (!block_group) { | |
7622 | if (last == 0) | |
7623 | break; | |
7624 | last = 0; | |
7625 | continue; | |
7626 | } | |
7627 | ||
7628 | inode = block_group->inode; | |
7629 | block_group->iref = 0; | |
7630 | block_group->inode = NULL; | |
7631 | spin_unlock(&block_group->lock); | |
7632 | iput(inode); | |
7633 | last = block_group->key.objectid + block_group->key.offset; | |
7634 | btrfs_put_block_group(block_group); | |
7635 | } | |
7636 | } | |
7637 | ||
1a40e23b ZY |
7638 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7639 | { | |
7640 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7641 | struct btrfs_space_info *space_info; |
11833d66 | 7642 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7643 | struct rb_node *n; |
7644 | ||
11833d66 YZ |
7645 | down_write(&info->extent_commit_sem); |
7646 | while (!list_empty(&info->caching_block_groups)) { | |
7647 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7648 | struct btrfs_caching_control, list); | |
7649 | list_del(&caching_ctl->list); | |
7650 | put_caching_control(caching_ctl); | |
7651 | } | |
7652 | up_write(&info->extent_commit_sem); | |
7653 | ||
1a40e23b ZY |
7654 | spin_lock(&info->block_group_cache_lock); |
7655 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7656 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7657 | cache_node); | |
1a40e23b ZY |
7658 | rb_erase(&block_group->cache_node, |
7659 | &info->block_group_cache_tree); | |
d899e052 YZ |
7660 | spin_unlock(&info->block_group_cache_lock); |
7661 | ||
80eb234a | 7662 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7663 | list_del(&block_group->list); |
80eb234a | 7664 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7665 | |
817d52f8 | 7666 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7667 | wait_block_group_cache_done(block_group); |
817d52f8 | 7668 | |
3c14874a JB |
7669 | /* |
7670 | * We haven't cached this block group, which means we could | |
7671 | * possibly have excluded extents on this block group. | |
7672 | */ | |
7673 | if (block_group->cached == BTRFS_CACHE_NO) | |
7674 | free_excluded_extents(info->extent_root, block_group); | |
7675 | ||
817d52f8 | 7676 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 7677 | btrfs_put_block_group(block_group); |
d899e052 YZ |
7678 | |
7679 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7680 | } |
7681 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7682 | |
7683 | /* now that all the block groups are freed, go through and | |
7684 | * free all the space_info structs. This is only called during | |
7685 | * the final stages of unmount, and so we know nobody is | |
7686 | * using them. We call synchronize_rcu() once before we start, | |
7687 | * just to be on the safe side. | |
7688 | */ | |
7689 | synchronize_rcu(); | |
7690 | ||
8929ecfa YZ |
7691 | release_global_block_rsv(info); |
7692 | ||
4184ea7f CM |
7693 | while(!list_empty(&info->space_info)) { |
7694 | space_info = list_entry(info->space_info.next, | |
7695 | struct btrfs_space_info, | |
7696 | list); | |
f0486c68 | 7697 | if (space_info->bytes_pinned > 0 || |
fb25e914 JB |
7698 | space_info->bytes_reserved > 0 || |
7699 | space_info->bytes_may_use > 0) { | |
f0486c68 YZ |
7700 | WARN_ON(1); |
7701 | dump_space_info(space_info, 0, 0); | |
7702 | } | |
4184ea7f CM |
7703 | list_del(&space_info->list); |
7704 | kfree(space_info); | |
7705 | } | |
1a40e23b ZY |
7706 | return 0; |
7707 | } | |
7708 | ||
b742bb82 YZ |
7709 | static void __link_block_group(struct btrfs_space_info *space_info, |
7710 | struct btrfs_block_group_cache *cache) | |
7711 | { | |
7712 | int index = get_block_group_index(cache); | |
7713 | ||
7714 | down_write(&space_info->groups_sem); | |
7715 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
7716 | up_write(&space_info->groups_sem); | |
7717 | } | |
7718 | ||
9078a3e1 CM |
7719 | int btrfs_read_block_groups(struct btrfs_root *root) |
7720 | { | |
7721 | struct btrfs_path *path; | |
7722 | int ret; | |
9078a3e1 | 7723 | struct btrfs_block_group_cache *cache; |
be744175 | 7724 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7725 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7726 | struct btrfs_key key; |
7727 | struct btrfs_key found_key; | |
5f39d397 | 7728 | struct extent_buffer *leaf; |
0af3d00b JB |
7729 | int need_clear = 0; |
7730 | u64 cache_gen; | |
96b5179d | 7731 | |
be744175 | 7732 | root = info->extent_root; |
9078a3e1 | 7733 | key.objectid = 0; |
0b86a832 | 7734 | key.offset = 0; |
9078a3e1 | 7735 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7736 | path = btrfs_alloc_path(); |
7737 | if (!path) | |
7738 | return -ENOMEM; | |
026fd317 | 7739 | path->reada = 1; |
9078a3e1 | 7740 | |
6c41761f | 7741 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 7742 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 7743 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 7744 | need_clear = 1; |
88c2ba3b JB |
7745 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
7746 | need_clear = 1; | |
0af3d00b | 7747 | |
d397712b | 7748 | while (1) { |
0b86a832 | 7749 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
7750 | if (ret > 0) |
7751 | break; | |
0b86a832 CM |
7752 | if (ret != 0) |
7753 | goto error; | |
5f39d397 CM |
7754 | leaf = path->nodes[0]; |
7755 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7756 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7757 | if (!cache) { |
0b86a832 | 7758 | ret = -ENOMEM; |
f0486c68 | 7759 | goto error; |
9078a3e1 | 7760 | } |
34d52cb6 LZ |
7761 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7762 | GFP_NOFS); | |
7763 | if (!cache->free_space_ctl) { | |
7764 | kfree(cache); | |
7765 | ret = -ENOMEM; | |
7766 | goto error; | |
7767 | } | |
3e1ad54f | 7768 | |
d2fb3437 | 7769 | atomic_set(&cache->count, 1); |
c286ac48 | 7770 | spin_lock_init(&cache->lock); |
817d52f8 | 7771 | cache->fs_info = info; |
0f9dd46c | 7772 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7773 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 7774 | |
cf7c1ef6 LB |
7775 | if (need_clear) { |
7776 | /* | |
7777 | * When we mount with old space cache, we need to | |
7778 | * set BTRFS_DC_CLEAR and set dirty flag. | |
7779 | * | |
7780 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
7781 | * truncate the old free space cache inode and | |
7782 | * setup a new one. | |
7783 | * b) Setting 'dirty flag' makes sure that we flush | |
7784 | * the new space cache info onto disk. | |
7785 | */ | |
0af3d00b | 7786 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 LB |
7787 | if (btrfs_test_opt(root, SPACE_CACHE)) |
7788 | cache->dirty = 1; | |
7789 | } | |
0af3d00b | 7790 | |
5f39d397 CM |
7791 | read_extent_buffer(leaf, &cache->item, |
7792 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7793 | sizeof(cache->item)); | |
9078a3e1 | 7794 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7795 | |
9078a3e1 | 7796 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 7797 | btrfs_release_path(path); |
0b86a832 | 7798 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7799 | cache->sectorsize = root->sectorsize; |
7800 | ||
34d52cb6 LZ |
7801 | btrfs_init_free_space_ctl(cache); |
7802 | ||
3c14874a JB |
7803 | /* |
7804 | * We need to exclude the super stripes now so that the space | |
7805 | * info has super bytes accounted for, otherwise we'll think | |
7806 | * we have more space than we actually do. | |
7807 | */ | |
7808 | exclude_super_stripes(root, cache); | |
7809 | ||
817d52f8 JB |
7810 | /* |
7811 | * check for two cases, either we are full, and therefore | |
7812 | * don't need to bother with the caching work since we won't | |
7813 | * find any space, or we are empty, and we can just add all | |
7814 | * the space in and be done with it. This saves us _alot_ of | |
7815 | * time, particularly in the full case. | |
7816 | */ | |
7817 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 7818 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7819 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7820 | free_excluded_extents(root, cache); |
817d52f8 | 7821 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 7822 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
7823 | cache->cached = BTRFS_CACHE_FINISHED; |
7824 | add_new_free_space(cache, root->fs_info, | |
7825 | found_key.objectid, | |
7826 | found_key.objectid + | |
7827 | found_key.offset); | |
11833d66 | 7828 | free_excluded_extents(root, cache); |
817d52f8 | 7829 | } |
96b5179d | 7830 | |
6324fbf3 CM |
7831 | ret = update_space_info(info, cache->flags, found_key.offset, |
7832 | btrfs_block_group_used(&cache->item), | |
7833 | &space_info); | |
79787eaa | 7834 | BUG_ON(ret); /* -ENOMEM */ |
6324fbf3 | 7835 | cache->space_info = space_info; |
1b2da372 | 7836 | spin_lock(&cache->space_info->lock); |
f0486c68 | 7837 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7838 | spin_unlock(&cache->space_info->lock); |
7839 | ||
b742bb82 | 7840 | __link_block_group(space_info, cache); |
0f9dd46c JB |
7841 | |
7842 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
79787eaa | 7843 | BUG_ON(ret); /* Logic error */ |
75ccf47d CM |
7844 | |
7845 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 7846 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 7847 | set_block_group_ro(cache, 1); |
9078a3e1 | 7848 | } |
b742bb82 YZ |
7849 | |
7850 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
7851 | if (!(get_alloc_profile(root, space_info->flags) & | |
7852 | (BTRFS_BLOCK_GROUP_RAID10 | | |
7853 | BTRFS_BLOCK_GROUP_RAID1 | | |
7854 | BTRFS_BLOCK_GROUP_DUP))) | |
7855 | continue; | |
7856 | /* | |
7857 | * avoid allocating from un-mirrored block group if there are | |
7858 | * mirrored block groups. | |
7859 | */ | |
7860 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 7861 | set_block_group_ro(cache, 1); |
b742bb82 | 7862 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 7863 | set_block_group_ro(cache, 1); |
9078a3e1 | 7864 | } |
f0486c68 YZ |
7865 | |
7866 | init_global_block_rsv(info); | |
0b86a832 CM |
7867 | ret = 0; |
7868 | error: | |
9078a3e1 | 7869 | btrfs_free_path(path); |
0b86a832 | 7870 | return ret; |
9078a3e1 | 7871 | } |
6324fbf3 | 7872 | |
ea658bad JB |
7873 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
7874 | struct btrfs_root *root) | |
7875 | { | |
7876 | struct btrfs_block_group_cache *block_group, *tmp; | |
7877 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
7878 | struct btrfs_block_group_item item; | |
7879 | struct btrfs_key key; | |
7880 | int ret = 0; | |
7881 | ||
7882 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, | |
7883 | new_bg_list) { | |
7884 | list_del_init(&block_group->new_bg_list); | |
7885 | ||
7886 | if (ret) | |
7887 | continue; | |
7888 | ||
7889 | spin_lock(&block_group->lock); | |
7890 | memcpy(&item, &block_group->item, sizeof(item)); | |
7891 | memcpy(&key, &block_group->key, sizeof(key)); | |
7892 | spin_unlock(&block_group->lock); | |
7893 | ||
7894 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
7895 | sizeof(item)); | |
7896 | if (ret) | |
7897 | btrfs_abort_transaction(trans, extent_root, ret); | |
7898 | } | |
7899 | } | |
7900 | ||
6324fbf3 CM |
7901 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
7902 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7903 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7904 | u64 size) |
7905 | { | |
7906 | int ret; | |
6324fbf3 CM |
7907 | struct btrfs_root *extent_root; |
7908 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7909 | |
7910 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7911 | |
12fcfd22 | 7912 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7913 | |
8f18cf13 | 7914 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7915 | if (!cache) |
7916 | return -ENOMEM; | |
34d52cb6 LZ |
7917 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7918 | GFP_NOFS); | |
7919 | if (!cache->free_space_ctl) { | |
7920 | kfree(cache); | |
7921 | return -ENOMEM; | |
7922 | } | |
0f9dd46c | 7923 | |
e17cade2 | 7924 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7925 | cache->key.offset = size; |
d2fb3437 | 7926 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 7927 | cache->sectorsize = root->sectorsize; |
0af3d00b | 7928 | cache->fs_info = root->fs_info; |
96303081 | 7929 | |
d2fb3437 | 7930 | atomic_set(&cache->count, 1); |
c286ac48 | 7931 | spin_lock_init(&cache->lock); |
0f9dd46c | 7932 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7933 | INIT_LIST_HEAD(&cache->cluster_list); |
ea658bad | 7934 | INIT_LIST_HEAD(&cache->new_bg_list); |
0ef3e66b | 7935 | |
34d52cb6 LZ |
7936 | btrfs_init_free_space_ctl(cache); |
7937 | ||
6324fbf3 | 7938 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7939 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7940 | cache->flags = type; | |
7941 | btrfs_set_block_group_flags(&cache->item, type); | |
7942 | ||
11833d66 | 7943 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7944 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7945 | exclude_super_stripes(root, cache); |
96303081 | 7946 | |
817d52f8 JB |
7947 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7948 | chunk_offset + size); | |
7949 | ||
11833d66 YZ |
7950 | free_excluded_extents(root, cache); |
7951 | ||
6324fbf3 CM |
7952 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7953 | &cache->space_info); | |
79787eaa | 7954 | BUG_ON(ret); /* -ENOMEM */ |
c7c144db | 7955 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
7956 | |
7957 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 7958 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7959 | spin_unlock(&cache->space_info->lock); |
7960 | ||
b742bb82 | 7961 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 7962 | |
0f9dd46c | 7963 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
79787eaa | 7964 | BUG_ON(ret); /* Logic error */ |
c286ac48 | 7965 | |
ea658bad | 7966 | list_add_tail(&cache->new_bg_list, &trans->new_bgs); |
6324fbf3 | 7967 | |
d18a2c44 | 7968 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7969 | |
6324fbf3 CM |
7970 | return 0; |
7971 | } | |
1a40e23b | 7972 | |
10ea00f5 ID |
7973 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
7974 | { | |
899c81ea ID |
7975 | u64 extra_flags = chunk_to_extended(flags) & |
7976 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 7977 | |
de98ced9 | 7978 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
7979 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
7980 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
7981 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
7982 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
7983 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
7984 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 7985 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
7986 | } |
7987 | ||
1a40e23b ZY |
7988 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
7989 | struct btrfs_root *root, u64 group_start) | |
7990 | { | |
7991 | struct btrfs_path *path; | |
7992 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7993 | struct btrfs_free_cluster *cluster; |
0af3d00b | 7994 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 7995 | struct btrfs_key key; |
0af3d00b | 7996 | struct inode *inode; |
1a40e23b | 7997 | int ret; |
10ea00f5 | 7998 | int index; |
89a55897 | 7999 | int factor; |
1a40e23b | 8000 | |
1a40e23b ZY |
8001 | root = root->fs_info->extent_root; |
8002 | ||
8003 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
8004 | BUG_ON(!block_group); | |
c146afad | 8005 | BUG_ON(!block_group->ro); |
1a40e23b | 8006 | |
9f7c43c9 | 8007 | /* |
8008 | * Free the reserved super bytes from this block group before | |
8009 | * remove it. | |
8010 | */ | |
8011 | free_excluded_extents(root, block_group); | |
8012 | ||
1a40e23b | 8013 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 8014 | index = get_block_group_index(block_group); |
89a55897 JB |
8015 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
8016 | BTRFS_BLOCK_GROUP_RAID1 | | |
8017 | BTRFS_BLOCK_GROUP_RAID10)) | |
8018 | factor = 2; | |
8019 | else | |
8020 | factor = 1; | |
1a40e23b | 8021 | |
44fb5511 CM |
8022 | /* make sure this block group isn't part of an allocation cluster */ |
8023 | cluster = &root->fs_info->data_alloc_cluster; | |
8024 | spin_lock(&cluster->refill_lock); | |
8025 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8026 | spin_unlock(&cluster->refill_lock); | |
8027 | ||
8028 | /* | |
8029 | * make sure this block group isn't part of a metadata | |
8030 | * allocation cluster | |
8031 | */ | |
8032 | cluster = &root->fs_info->meta_alloc_cluster; | |
8033 | spin_lock(&cluster->refill_lock); | |
8034 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8035 | spin_unlock(&cluster->refill_lock); | |
8036 | ||
1a40e23b | 8037 | path = btrfs_alloc_path(); |
d8926bb3 MF |
8038 | if (!path) { |
8039 | ret = -ENOMEM; | |
8040 | goto out; | |
8041 | } | |
1a40e23b | 8042 | |
10b2f34d | 8043 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 8044 | if (!IS_ERR(inode)) { |
b532402e | 8045 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
8046 | if (ret) { |
8047 | btrfs_add_delayed_iput(inode); | |
8048 | goto out; | |
8049 | } | |
0af3d00b JB |
8050 | clear_nlink(inode); |
8051 | /* One for the block groups ref */ | |
8052 | spin_lock(&block_group->lock); | |
8053 | if (block_group->iref) { | |
8054 | block_group->iref = 0; | |
8055 | block_group->inode = NULL; | |
8056 | spin_unlock(&block_group->lock); | |
8057 | iput(inode); | |
8058 | } else { | |
8059 | spin_unlock(&block_group->lock); | |
8060 | } | |
8061 | /* One for our lookup ref */ | |
455757c3 | 8062 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
8063 | } |
8064 | ||
8065 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
8066 | key.offset = block_group->key.objectid; | |
8067 | key.type = 0; | |
8068 | ||
8069 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
8070 | if (ret < 0) | |
8071 | goto out; | |
8072 | if (ret > 0) | |
b3b4aa74 | 8073 | btrfs_release_path(path); |
0af3d00b JB |
8074 | if (ret == 0) { |
8075 | ret = btrfs_del_item(trans, tree_root, path); | |
8076 | if (ret) | |
8077 | goto out; | |
b3b4aa74 | 8078 | btrfs_release_path(path); |
0af3d00b JB |
8079 | } |
8080 | ||
3dfdb934 | 8081 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
8082 | rb_erase(&block_group->cache_node, |
8083 | &root->fs_info->block_group_cache_tree); | |
a1897fdd LB |
8084 | |
8085 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
8086 | root->fs_info->first_logical_byte = (u64)-1; | |
3dfdb934 | 8087 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 8088 | |
80eb234a | 8089 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
8090 | /* |
8091 | * we must use list_del_init so people can check to see if they | |
8092 | * are still on the list after taking the semaphore | |
8093 | */ | |
8094 | list_del_init(&block_group->list); | |
10ea00f5 ID |
8095 | if (list_empty(&block_group->space_info->block_groups[index])) |
8096 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 8097 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 8098 | |
817d52f8 | 8099 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8100 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
8101 | |
8102 | btrfs_remove_free_space_cache(block_group); | |
8103 | ||
c146afad YZ |
8104 | spin_lock(&block_group->space_info->lock); |
8105 | block_group->space_info->total_bytes -= block_group->key.offset; | |
8106 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 8107 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 8108 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 8109 | |
0af3d00b JB |
8110 | memcpy(&key, &block_group->key, sizeof(key)); |
8111 | ||
283bb197 | 8112 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 8113 | |
fa9c0d79 CM |
8114 | btrfs_put_block_group(block_group); |
8115 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
8116 | |
8117 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
8118 | if (ret > 0) | |
8119 | ret = -EIO; | |
8120 | if (ret < 0) | |
8121 | goto out; | |
8122 | ||
8123 | ret = btrfs_del_item(trans, root, path); | |
8124 | out: | |
8125 | btrfs_free_path(path); | |
8126 | return ret; | |
8127 | } | |
acce952b | 8128 | |
c59021f8 | 8129 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
8130 | { | |
8131 | struct btrfs_space_info *space_info; | |
1aba86d6 | 8132 | struct btrfs_super_block *disk_super; |
8133 | u64 features; | |
8134 | u64 flags; | |
8135 | int mixed = 0; | |
c59021f8 | 8136 | int ret; |
8137 | ||
6c41761f | 8138 | disk_super = fs_info->super_copy; |
1aba86d6 | 8139 | if (!btrfs_super_root(disk_super)) |
8140 | return 1; | |
c59021f8 | 8141 | |
1aba86d6 | 8142 | features = btrfs_super_incompat_flags(disk_super); |
8143 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
8144 | mixed = 1; | |
c59021f8 | 8145 | |
1aba86d6 | 8146 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
8147 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 8148 | if (ret) |
1aba86d6 | 8149 | goto out; |
c59021f8 | 8150 | |
1aba86d6 | 8151 | if (mixed) { |
8152 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
8153 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8154 | } else { | |
8155 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
8156 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8157 | if (ret) | |
8158 | goto out; | |
8159 | ||
8160 | flags = BTRFS_BLOCK_GROUP_DATA; | |
8161 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8162 | } | |
8163 | out: | |
c59021f8 | 8164 | return ret; |
8165 | } | |
8166 | ||
acce952b | 8167 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
8168 | { | |
8169 | return unpin_extent_range(root, start, end); | |
8170 | } | |
8171 | ||
8172 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 8173 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 8174 | { |
5378e607 | 8175 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 8176 | } |
f7039b1d LD |
8177 | |
8178 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
8179 | { | |
8180 | struct btrfs_fs_info *fs_info = root->fs_info; | |
8181 | struct btrfs_block_group_cache *cache = NULL; | |
8182 | u64 group_trimmed; | |
8183 | u64 start; | |
8184 | u64 end; | |
8185 | u64 trimmed = 0; | |
2cac13e4 | 8186 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
8187 | int ret = 0; |
8188 | ||
2cac13e4 LB |
8189 | /* |
8190 | * try to trim all FS space, our block group may start from non-zero. | |
8191 | */ | |
8192 | if (range->len == total_bytes) | |
8193 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
8194 | else | |
8195 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
8196 | |
8197 | while (cache) { | |
8198 | if (cache->key.objectid >= (range->start + range->len)) { | |
8199 | btrfs_put_block_group(cache); | |
8200 | break; | |
8201 | } | |
8202 | ||
8203 | start = max(range->start, cache->key.objectid); | |
8204 | end = min(range->start + range->len, | |
8205 | cache->key.objectid + cache->key.offset); | |
8206 | ||
8207 | if (end - start >= range->minlen) { | |
8208 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 8209 | ret = cache_block_group(cache, 0); |
f7039b1d LD |
8210 | if (!ret) |
8211 | wait_block_group_cache_done(cache); | |
8212 | } | |
8213 | ret = btrfs_trim_block_group(cache, | |
8214 | &group_trimmed, | |
8215 | start, | |
8216 | end, | |
8217 | range->minlen); | |
8218 | ||
8219 | trimmed += group_trimmed; | |
8220 | if (ret) { | |
8221 | btrfs_put_block_group(cache); | |
8222 | break; | |
8223 | } | |
8224 | } | |
8225 | ||
8226 | cache = next_block_group(fs_info->tree_root, cache); | |
8227 | } | |
8228 | ||
8229 | range->len = trimmed; | |
8230 | return ret; | |
8231 | } |