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" |
53b381b3 | 34 | #include "raid56.h" |
925baedd | 35 | #include "locking.h" |
fa9c0d79 | 36 | #include "free-space-cache.h" |
3fed40cc | 37 | #include "math.h" |
fec577fb | 38 | |
709c0486 AJ |
39 | #undef SCRAMBLE_DELAYED_REFS |
40 | ||
9e622d6b MX |
41 | /* |
42 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
43 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
44 | * if we really need one. | |
45 | * | |
0e4f8f88 CM |
46 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
47 | * if we have very few chunks already allocated. This is | |
48 | * used as part of the clustering code to help make sure | |
49 | * we have a good pool of storage to cluster in, without | |
50 | * filling the FS with empty chunks | |
51 | * | |
9e622d6b MX |
52 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
53 | * | |
0e4f8f88 CM |
54 | */ |
55 | enum { | |
56 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
57 | CHUNK_ALLOC_LIMITED = 1, |
58 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
59 | }; |
60 | ||
fb25e914 JB |
61 | /* |
62 | * Control how reservations are dealt with. | |
63 | * | |
64 | * RESERVE_FREE - freeing a reservation. | |
65 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
66 | * ENOSPC accounting | |
67 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
68 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
69 | */ | |
70 | enum { | |
71 | RESERVE_FREE = 0, | |
72 | RESERVE_ALLOC = 1, | |
73 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
74 | }; | |
75 | ||
c53d613e | 76 | static int update_block_group(struct btrfs_root *root, |
f0486c68 | 77 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
78 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
79 | struct btrfs_root *root, | |
80 | u64 bytenr, u64 num_bytes, u64 parent, | |
81 | u64 root_objectid, u64 owner_objectid, | |
82 | u64 owner_offset, int refs_to_drop, | |
83 | struct btrfs_delayed_extent_op *extra_op); | |
84 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
85 | struct extent_buffer *leaf, | |
86 | struct btrfs_extent_item *ei); | |
87 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
88 | struct btrfs_root *root, | |
89 | u64 parent, u64 root_objectid, | |
90 | u64 flags, u64 owner, u64 offset, | |
91 | struct btrfs_key *ins, int ref_mod); | |
92 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
93 | struct btrfs_root *root, | |
94 | u64 parent, u64 root_objectid, | |
95 | u64 flags, struct btrfs_disk_key *key, | |
96 | int level, struct btrfs_key *ins); | |
6a63209f | 97 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 JB |
98 | struct btrfs_root *extent_root, u64 flags, |
99 | int force); | |
11833d66 YZ |
100 | static int find_next_key(struct btrfs_path *path, int level, |
101 | struct btrfs_key *key); | |
9ed74f2d JB |
102 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
103 | int dump_block_groups); | |
fb25e914 JB |
104 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
105 | u64 num_bytes, int reserve); | |
5d80366e JB |
106 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
107 | u64 num_bytes); | |
6a63209f | 108 | |
817d52f8 JB |
109 | static noinline int |
110 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
111 | { | |
112 | smp_mb(); | |
113 | return cache->cached == BTRFS_CACHE_FINISHED; | |
114 | } | |
115 | ||
0f9dd46c JB |
116 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
117 | { | |
118 | return (cache->flags & bits) == bits; | |
119 | } | |
120 | ||
62a45b60 | 121 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
122 | { |
123 | atomic_inc(&cache->count); | |
124 | } | |
125 | ||
126 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
127 | { | |
f0486c68 YZ |
128 | if (atomic_dec_and_test(&cache->count)) { |
129 | WARN_ON(cache->pinned > 0); | |
130 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 131 | kfree(cache->free_space_ctl); |
11dfe35a | 132 | kfree(cache); |
f0486c68 | 133 | } |
11dfe35a JB |
134 | } |
135 | ||
0f9dd46c JB |
136 | /* |
137 | * this adds the block group to the fs_info rb tree for the block group | |
138 | * cache | |
139 | */ | |
b2950863 | 140 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
141 | struct btrfs_block_group_cache *block_group) |
142 | { | |
143 | struct rb_node **p; | |
144 | struct rb_node *parent = NULL; | |
145 | struct btrfs_block_group_cache *cache; | |
146 | ||
147 | spin_lock(&info->block_group_cache_lock); | |
148 | p = &info->block_group_cache_tree.rb_node; | |
149 | ||
150 | while (*p) { | |
151 | parent = *p; | |
152 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
153 | cache_node); | |
154 | if (block_group->key.objectid < cache->key.objectid) { | |
155 | p = &(*p)->rb_left; | |
156 | } else if (block_group->key.objectid > cache->key.objectid) { | |
157 | p = &(*p)->rb_right; | |
158 | } else { | |
159 | spin_unlock(&info->block_group_cache_lock); | |
160 | return -EEXIST; | |
161 | } | |
162 | } | |
163 | ||
164 | rb_link_node(&block_group->cache_node, parent, p); | |
165 | rb_insert_color(&block_group->cache_node, | |
166 | &info->block_group_cache_tree); | |
a1897fdd LB |
167 | |
168 | if (info->first_logical_byte > block_group->key.objectid) | |
169 | info->first_logical_byte = block_group->key.objectid; | |
170 | ||
0f9dd46c JB |
171 | spin_unlock(&info->block_group_cache_lock); |
172 | ||
173 | return 0; | |
174 | } | |
175 | ||
176 | /* | |
177 | * This will return the block group at or after bytenr if contains is 0, else | |
178 | * it will return the block group that contains the bytenr | |
179 | */ | |
180 | static struct btrfs_block_group_cache * | |
181 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
182 | int contains) | |
183 | { | |
184 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
185 | struct rb_node *n; | |
186 | u64 end, start; | |
187 | ||
188 | spin_lock(&info->block_group_cache_lock); | |
189 | n = info->block_group_cache_tree.rb_node; | |
190 | ||
191 | while (n) { | |
192 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
193 | cache_node); | |
194 | end = cache->key.objectid + cache->key.offset - 1; | |
195 | start = cache->key.objectid; | |
196 | ||
197 | if (bytenr < start) { | |
198 | if (!contains && (!ret || start < ret->key.objectid)) | |
199 | ret = cache; | |
200 | n = n->rb_left; | |
201 | } else if (bytenr > start) { | |
202 | if (contains && bytenr <= end) { | |
203 | ret = cache; | |
204 | break; | |
205 | } | |
206 | n = n->rb_right; | |
207 | } else { | |
208 | ret = cache; | |
209 | break; | |
210 | } | |
211 | } | |
a1897fdd | 212 | if (ret) { |
11dfe35a | 213 | btrfs_get_block_group(ret); |
a1897fdd LB |
214 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
215 | info->first_logical_byte = ret->key.objectid; | |
216 | } | |
0f9dd46c JB |
217 | spin_unlock(&info->block_group_cache_lock); |
218 | ||
219 | return ret; | |
220 | } | |
221 | ||
11833d66 YZ |
222 | static int add_excluded_extent(struct btrfs_root *root, |
223 | u64 start, u64 num_bytes) | |
817d52f8 | 224 | { |
11833d66 YZ |
225 | u64 end = start + num_bytes - 1; |
226 | set_extent_bits(&root->fs_info->freed_extents[0], | |
227 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
228 | set_extent_bits(&root->fs_info->freed_extents[1], | |
229 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
230 | return 0; | |
231 | } | |
817d52f8 | 232 | |
11833d66 YZ |
233 | static void free_excluded_extents(struct btrfs_root *root, |
234 | struct btrfs_block_group_cache *cache) | |
235 | { | |
236 | u64 start, end; | |
817d52f8 | 237 | |
11833d66 YZ |
238 | start = cache->key.objectid; |
239 | end = start + cache->key.offset - 1; | |
240 | ||
241 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
242 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
243 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
244 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
245 | } |
246 | ||
11833d66 YZ |
247 | static int exclude_super_stripes(struct btrfs_root *root, |
248 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 249 | { |
817d52f8 JB |
250 | u64 bytenr; |
251 | u64 *logical; | |
252 | int stripe_len; | |
253 | int i, nr, ret; | |
254 | ||
06b2331f YZ |
255 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
256 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
257 | cache->bytes_super += stripe_len; | |
258 | ret = add_excluded_extent(root, cache->key.objectid, | |
259 | stripe_len); | |
835d974f JB |
260 | if (ret) |
261 | return ret; | |
06b2331f YZ |
262 | } |
263 | ||
817d52f8 JB |
264 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
265 | bytenr = btrfs_sb_offset(i); | |
266 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
267 | cache->key.objectid, bytenr, | |
268 | 0, &logical, &nr, &stripe_len); | |
835d974f JB |
269 | if (ret) |
270 | return ret; | |
11833d66 | 271 | |
817d52f8 | 272 | while (nr--) { |
1b2da372 | 273 | cache->bytes_super += stripe_len; |
11833d66 YZ |
274 | ret = add_excluded_extent(root, logical[nr], |
275 | stripe_len); | |
835d974f JB |
276 | if (ret) { |
277 | kfree(logical); | |
278 | return ret; | |
279 | } | |
817d52f8 | 280 | } |
11833d66 | 281 | |
817d52f8 JB |
282 | kfree(logical); |
283 | } | |
817d52f8 JB |
284 | return 0; |
285 | } | |
286 | ||
11833d66 YZ |
287 | static struct btrfs_caching_control * |
288 | get_caching_control(struct btrfs_block_group_cache *cache) | |
289 | { | |
290 | struct btrfs_caching_control *ctl; | |
291 | ||
292 | spin_lock(&cache->lock); | |
293 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
294 | spin_unlock(&cache->lock); | |
295 | return NULL; | |
296 | } | |
297 | ||
dde5abee JB |
298 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
299 | if (!cache->caching_ctl) { | |
300 | spin_unlock(&cache->lock); | |
11833d66 YZ |
301 | return NULL; |
302 | } | |
303 | ||
304 | ctl = cache->caching_ctl; | |
305 | atomic_inc(&ctl->count); | |
306 | spin_unlock(&cache->lock); | |
307 | return ctl; | |
308 | } | |
309 | ||
310 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
311 | { | |
312 | if (atomic_dec_and_test(&ctl->count)) | |
313 | kfree(ctl); | |
314 | } | |
315 | ||
0f9dd46c JB |
316 | /* |
317 | * this is only called by cache_block_group, since we could have freed extents | |
318 | * we need to check the pinned_extents for any extents that can't be used yet | |
319 | * since their free space will be released as soon as the transaction commits. | |
320 | */ | |
817d52f8 | 321 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
322 | struct btrfs_fs_info *info, u64 start, u64 end) |
323 | { | |
817d52f8 | 324 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
325 | int ret; |
326 | ||
327 | while (start < end) { | |
11833d66 | 328 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 329 | &extent_start, &extent_end, |
e6138876 JB |
330 | EXTENT_DIRTY | EXTENT_UPTODATE, |
331 | NULL); | |
0f9dd46c JB |
332 | if (ret) |
333 | break; | |
334 | ||
06b2331f | 335 | if (extent_start <= start) { |
0f9dd46c JB |
336 | start = extent_end + 1; |
337 | } else if (extent_start > start && extent_start < end) { | |
338 | size = extent_start - start; | |
817d52f8 | 339 | total_added += size; |
ea6a478e JB |
340 | ret = btrfs_add_free_space(block_group, start, |
341 | size); | |
79787eaa | 342 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
343 | start = extent_end + 1; |
344 | } else { | |
345 | break; | |
346 | } | |
347 | } | |
348 | ||
349 | if (start < end) { | |
350 | size = end - start; | |
817d52f8 | 351 | total_added += size; |
ea6a478e | 352 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 353 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
354 | } |
355 | ||
817d52f8 | 356 | return total_added; |
0f9dd46c JB |
357 | } |
358 | ||
bab39bf9 | 359 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 360 | { |
bab39bf9 JB |
361 | struct btrfs_block_group_cache *block_group; |
362 | struct btrfs_fs_info *fs_info; | |
363 | struct btrfs_caching_control *caching_ctl; | |
364 | struct btrfs_root *extent_root; | |
e37c9e69 | 365 | struct btrfs_path *path; |
5f39d397 | 366 | struct extent_buffer *leaf; |
11833d66 | 367 | struct btrfs_key key; |
817d52f8 | 368 | u64 total_found = 0; |
11833d66 YZ |
369 | u64 last = 0; |
370 | u32 nritems; | |
371 | int ret = 0; | |
f510cfec | 372 | |
bab39bf9 JB |
373 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
374 | block_group = caching_ctl->block_group; | |
375 | fs_info = block_group->fs_info; | |
376 | extent_root = fs_info->extent_root; | |
377 | ||
e37c9e69 CM |
378 | path = btrfs_alloc_path(); |
379 | if (!path) | |
bab39bf9 | 380 | goto out; |
7d7d6068 | 381 | |
817d52f8 | 382 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 383 | |
5cd57b2c | 384 | /* |
817d52f8 JB |
385 | * We don't want to deadlock with somebody trying to allocate a new |
386 | * extent for the extent root while also trying to search the extent | |
387 | * root to add free space. So we skip locking and search the commit | |
388 | * root, since its read-only | |
5cd57b2c CM |
389 | */ |
390 | path->skip_locking = 1; | |
817d52f8 | 391 | path->search_commit_root = 1; |
026fd317 | 392 | path->reada = 1; |
817d52f8 | 393 | |
e4404d6e | 394 | key.objectid = last; |
e37c9e69 | 395 | key.offset = 0; |
11833d66 | 396 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 397 | again: |
11833d66 | 398 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
399 | /* need to make sure the commit_root doesn't disappear */ |
400 | down_read(&fs_info->extent_commit_sem); | |
401 | ||
11833d66 | 402 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 403 | if (ret < 0) |
ef8bbdfe | 404 | goto err; |
a512bbf8 | 405 | |
11833d66 YZ |
406 | leaf = path->nodes[0]; |
407 | nritems = btrfs_header_nritems(leaf); | |
408 | ||
d397712b | 409 | while (1) { |
7841cb28 | 410 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 411 | last = (u64)-1; |
817d52f8 | 412 | break; |
f25784b3 | 413 | } |
817d52f8 | 414 | |
11833d66 YZ |
415 | if (path->slots[0] < nritems) { |
416 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
417 | } else { | |
418 | ret = find_next_key(path, 0, &key); | |
419 | if (ret) | |
e37c9e69 | 420 | break; |
817d52f8 | 421 | |
589d8ade JB |
422 | if (need_resched() || |
423 | btrfs_next_leaf(extent_root, path)) { | |
424 | caching_ctl->progress = last; | |
ff5714cc | 425 | btrfs_release_path(path); |
589d8ade JB |
426 | up_read(&fs_info->extent_commit_sem); |
427 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 428 | cond_resched(); |
589d8ade JB |
429 | goto again; |
430 | } | |
431 | leaf = path->nodes[0]; | |
432 | nritems = btrfs_header_nritems(leaf); | |
433 | continue; | |
11833d66 | 434 | } |
817d52f8 | 435 | |
11833d66 YZ |
436 | if (key.objectid < block_group->key.objectid) { |
437 | path->slots[0]++; | |
817d52f8 | 438 | continue; |
e37c9e69 | 439 | } |
0f9dd46c | 440 | |
e37c9e69 | 441 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 442 | block_group->key.offset) |
e37c9e69 | 443 | break; |
7d7d6068 | 444 | |
3173a18f JB |
445 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
446 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
447 | total_found += add_new_free_space(block_group, |
448 | fs_info, last, | |
449 | key.objectid); | |
3173a18f JB |
450 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
451 | last = key.objectid + | |
452 | fs_info->tree_root->leafsize; | |
453 | else | |
454 | last = key.objectid + key.offset; | |
817d52f8 | 455 | |
11833d66 YZ |
456 | if (total_found > (1024 * 1024 * 2)) { |
457 | total_found = 0; | |
458 | wake_up(&caching_ctl->wait); | |
459 | } | |
817d52f8 | 460 | } |
e37c9e69 CM |
461 | path->slots[0]++; |
462 | } | |
817d52f8 | 463 | ret = 0; |
e37c9e69 | 464 | |
817d52f8 JB |
465 | total_found += add_new_free_space(block_group, fs_info, last, |
466 | block_group->key.objectid + | |
467 | block_group->key.offset); | |
11833d66 | 468 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
469 | |
470 | spin_lock(&block_group->lock); | |
11833d66 | 471 | block_group->caching_ctl = NULL; |
817d52f8 JB |
472 | block_group->cached = BTRFS_CACHE_FINISHED; |
473 | spin_unlock(&block_group->lock); | |
0f9dd46c | 474 | |
54aa1f4d | 475 | err: |
e37c9e69 | 476 | btrfs_free_path(path); |
276e680d | 477 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 478 | |
11833d66 YZ |
479 | free_excluded_extents(extent_root, block_group); |
480 | ||
481 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 482 | out: |
11833d66 YZ |
483 | wake_up(&caching_ctl->wait); |
484 | ||
485 | put_caching_control(caching_ctl); | |
11dfe35a | 486 | btrfs_put_block_group(block_group); |
817d52f8 JB |
487 | } |
488 | ||
9d66e233 | 489 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 490 | int load_cache_only) |
817d52f8 | 491 | { |
291c7d2f | 492 | DEFINE_WAIT(wait); |
11833d66 YZ |
493 | struct btrfs_fs_info *fs_info = cache->fs_info; |
494 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
495 | int ret = 0; |
496 | ||
291c7d2f | 497 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
498 | if (!caching_ctl) |
499 | return -ENOMEM; | |
291c7d2f JB |
500 | |
501 | INIT_LIST_HEAD(&caching_ctl->list); | |
502 | mutex_init(&caching_ctl->mutex); | |
503 | init_waitqueue_head(&caching_ctl->wait); | |
504 | caching_ctl->block_group = cache; | |
505 | caching_ctl->progress = cache->key.objectid; | |
506 | atomic_set(&caching_ctl->count, 1); | |
507 | caching_ctl->work.func = caching_thread; | |
508 | ||
509 | spin_lock(&cache->lock); | |
510 | /* | |
511 | * This should be a rare occasion, but this could happen I think in the | |
512 | * case where one thread starts to load the space cache info, and then | |
513 | * some other thread starts a transaction commit which tries to do an | |
514 | * allocation while the other thread is still loading the space cache | |
515 | * info. The previous loop should have kept us from choosing this block | |
516 | * group, but if we've moved to the state where we will wait on caching | |
517 | * block groups we need to first check if we're doing a fast load here, | |
518 | * so we can wait for it to finish, otherwise we could end up allocating | |
519 | * from a block group who's cache gets evicted for one reason or | |
520 | * another. | |
521 | */ | |
522 | while (cache->cached == BTRFS_CACHE_FAST) { | |
523 | struct btrfs_caching_control *ctl; | |
524 | ||
525 | ctl = cache->caching_ctl; | |
526 | atomic_inc(&ctl->count); | |
527 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
528 | spin_unlock(&cache->lock); | |
529 | ||
530 | schedule(); | |
531 | ||
532 | finish_wait(&ctl->wait, &wait); | |
533 | put_caching_control(ctl); | |
534 | spin_lock(&cache->lock); | |
535 | } | |
536 | ||
537 | if (cache->cached != BTRFS_CACHE_NO) { | |
538 | spin_unlock(&cache->lock); | |
539 | kfree(caching_ctl); | |
11833d66 | 540 | return 0; |
291c7d2f JB |
541 | } |
542 | WARN_ON(cache->caching_ctl); | |
543 | cache->caching_ctl = caching_ctl; | |
544 | cache->cached = BTRFS_CACHE_FAST; | |
545 | spin_unlock(&cache->lock); | |
11833d66 | 546 | |
d53ba474 | 547 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
9d66e233 JB |
548 | ret = load_free_space_cache(fs_info, cache); |
549 | ||
550 | spin_lock(&cache->lock); | |
551 | if (ret == 1) { | |
291c7d2f | 552 | cache->caching_ctl = NULL; |
9d66e233 JB |
553 | cache->cached = BTRFS_CACHE_FINISHED; |
554 | cache->last_byte_to_unpin = (u64)-1; | |
555 | } else { | |
291c7d2f JB |
556 | if (load_cache_only) { |
557 | cache->caching_ctl = NULL; | |
558 | cache->cached = BTRFS_CACHE_NO; | |
559 | } else { | |
560 | cache->cached = BTRFS_CACHE_STARTED; | |
561 | } | |
9d66e233 JB |
562 | } |
563 | spin_unlock(&cache->lock); | |
291c7d2f | 564 | wake_up(&caching_ctl->wait); |
3c14874a | 565 | if (ret == 1) { |
291c7d2f | 566 | put_caching_control(caching_ctl); |
3c14874a | 567 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 568 | return 0; |
3c14874a | 569 | } |
291c7d2f JB |
570 | } else { |
571 | /* | |
572 | * We are not going to do the fast caching, set cached to the | |
573 | * appropriate value and wakeup any waiters. | |
574 | */ | |
575 | spin_lock(&cache->lock); | |
576 | if (load_cache_only) { | |
577 | cache->caching_ctl = NULL; | |
578 | cache->cached = BTRFS_CACHE_NO; | |
579 | } else { | |
580 | cache->cached = BTRFS_CACHE_STARTED; | |
581 | } | |
582 | spin_unlock(&cache->lock); | |
583 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
584 | } |
585 | ||
291c7d2f JB |
586 | if (load_cache_only) { |
587 | put_caching_control(caching_ctl); | |
11833d66 | 588 | return 0; |
817d52f8 | 589 | } |
817d52f8 | 590 | |
11833d66 | 591 | down_write(&fs_info->extent_commit_sem); |
291c7d2f | 592 | atomic_inc(&caching_ctl->count); |
11833d66 YZ |
593 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
594 | up_write(&fs_info->extent_commit_sem); | |
595 | ||
11dfe35a | 596 | btrfs_get_block_group(cache); |
11833d66 | 597 | |
bab39bf9 | 598 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 599 | |
ef8bbdfe | 600 | return ret; |
e37c9e69 CM |
601 | } |
602 | ||
0f9dd46c JB |
603 | /* |
604 | * return the block group that starts at or after bytenr | |
605 | */ | |
d397712b CM |
606 | static struct btrfs_block_group_cache * |
607 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 608 | { |
0f9dd46c | 609 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 610 | |
0f9dd46c | 611 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 612 | |
0f9dd46c | 613 | return cache; |
0ef3e66b CM |
614 | } |
615 | ||
0f9dd46c | 616 | /* |
9f55684c | 617 | * return the block group that contains the given bytenr |
0f9dd46c | 618 | */ |
d397712b CM |
619 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
620 | struct btrfs_fs_info *info, | |
621 | u64 bytenr) | |
be744175 | 622 | { |
0f9dd46c | 623 | struct btrfs_block_group_cache *cache; |
be744175 | 624 | |
0f9dd46c | 625 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 626 | |
0f9dd46c | 627 | return cache; |
be744175 | 628 | } |
0b86a832 | 629 | |
0f9dd46c JB |
630 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
631 | u64 flags) | |
6324fbf3 | 632 | { |
0f9dd46c | 633 | struct list_head *head = &info->space_info; |
0f9dd46c | 634 | struct btrfs_space_info *found; |
4184ea7f | 635 | |
52ba6929 | 636 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 637 | |
4184ea7f CM |
638 | rcu_read_lock(); |
639 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 640 | if (found->flags & flags) { |
4184ea7f | 641 | rcu_read_unlock(); |
0f9dd46c | 642 | return found; |
4184ea7f | 643 | } |
0f9dd46c | 644 | } |
4184ea7f | 645 | rcu_read_unlock(); |
0f9dd46c | 646 | return NULL; |
6324fbf3 CM |
647 | } |
648 | ||
4184ea7f CM |
649 | /* |
650 | * after adding space to the filesystem, we need to clear the full flags | |
651 | * on all the space infos. | |
652 | */ | |
653 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
654 | { | |
655 | struct list_head *head = &info->space_info; | |
656 | struct btrfs_space_info *found; | |
657 | ||
658 | rcu_read_lock(); | |
659 | list_for_each_entry_rcu(found, head, list) | |
660 | found->full = 0; | |
661 | rcu_read_unlock(); | |
662 | } | |
663 | ||
d2fb3437 YZ |
664 | u64 btrfs_find_block_group(struct btrfs_root *root, |
665 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 666 | { |
96b5179d | 667 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 668 | u64 used; |
d2fb3437 YZ |
669 | u64 last = max(search_hint, search_start); |
670 | u64 group_start = 0; | |
31f3c99b | 671 | int full_search = 0; |
d2fb3437 | 672 | int factor = 9; |
0ef3e66b | 673 | int wrapped = 0; |
31f3c99b | 674 | again: |
e8569813 ZY |
675 | while (1) { |
676 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
677 | if (!cache) |
678 | break; | |
96b5179d | 679 | |
c286ac48 | 680 | spin_lock(&cache->lock); |
96b5179d CM |
681 | last = cache->key.objectid + cache->key.offset; |
682 | used = btrfs_block_group_used(&cache->item); | |
683 | ||
d2fb3437 YZ |
684 | if ((full_search || !cache->ro) && |
685 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 686 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
687 | div_factor(cache->key.offset, factor)) { |
688 | group_start = cache->key.objectid; | |
c286ac48 | 689 | spin_unlock(&cache->lock); |
fa9c0d79 | 690 | btrfs_put_block_group(cache); |
8790d502 CM |
691 | goto found; |
692 | } | |
6324fbf3 | 693 | } |
c286ac48 | 694 | spin_unlock(&cache->lock); |
fa9c0d79 | 695 | btrfs_put_block_group(cache); |
de428b63 | 696 | cond_resched(); |
cd1bc465 | 697 | } |
0ef3e66b CM |
698 | if (!wrapped) { |
699 | last = search_start; | |
700 | wrapped = 1; | |
701 | goto again; | |
702 | } | |
703 | if (!full_search && factor < 10) { | |
be744175 | 704 | last = search_start; |
31f3c99b | 705 | full_search = 1; |
0ef3e66b | 706 | factor = 10; |
31f3c99b CM |
707 | goto again; |
708 | } | |
be744175 | 709 | found: |
d2fb3437 | 710 | return group_start; |
925baedd | 711 | } |
0f9dd46c | 712 | |
e02119d5 | 713 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 714 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
715 | { |
716 | int ret; | |
717 | struct btrfs_key key; | |
31840ae1 | 718 | struct btrfs_path *path; |
e02119d5 | 719 | |
31840ae1 | 720 | path = btrfs_alloc_path(); |
d8926bb3 MF |
721 | if (!path) |
722 | return -ENOMEM; | |
723 | ||
e02119d5 CM |
724 | key.objectid = start; |
725 | key.offset = len; | |
3173a18f | 726 | key.type = BTRFS_EXTENT_ITEM_KEY; |
e02119d5 CM |
727 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, |
728 | 0, 0); | |
3173a18f JB |
729 | if (ret > 0) { |
730 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
731 | if (key.objectid == start && | |
732 | key.type == BTRFS_METADATA_ITEM_KEY) | |
733 | ret = 0; | |
734 | } | |
31840ae1 | 735 | btrfs_free_path(path); |
7bb86316 CM |
736 | return ret; |
737 | } | |
738 | ||
a22285a6 | 739 | /* |
3173a18f | 740 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
741 | * |
742 | * the head node for delayed ref is used to store the sum of all the | |
743 | * reference count modifications queued up in the rbtree. the head | |
744 | * node may also store the extent flags to set. This way you can check | |
745 | * to see what the reference count and extent flags would be if all of | |
746 | * the delayed refs are not processed. | |
747 | */ | |
748 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
749 | struct btrfs_root *root, u64 bytenr, | |
3173a18f | 750 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
751 | { |
752 | struct btrfs_delayed_ref_head *head; | |
753 | struct btrfs_delayed_ref_root *delayed_refs; | |
754 | struct btrfs_path *path; | |
755 | struct btrfs_extent_item *ei; | |
756 | struct extent_buffer *leaf; | |
757 | struct btrfs_key key; | |
758 | u32 item_size; | |
759 | u64 num_refs; | |
760 | u64 extent_flags; | |
761 | int ret; | |
762 | ||
3173a18f JB |
763 | /* |
764 | * If we don't have skinny metadata, don't bother doing anything | |
765 | * different | |
766 | */ | |
767 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { | |
768 | offset = root->leafsize; | |
769 | metadata = 0; | |
770 | } | |
771 | ||
a22285a6 YZ |
772 | path = btrfs_alloc_path(); |
773 | if (!path) | |
774 | return -ENOMEM; | |
775 | ||
3173a18f JB |
776 | if (metadata) { |
777 | key.objectid = bytenr; | |
778 | key.type = BTRFS_METADATA_ITEM_KEY; | |
779 | key.offset = offset; | |
780 | } else { | |
781 | key.objectid = bytenr; | |
782 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
783 | key.offset = offset; | |
784 | } | |
785 | ||
a22285a6 YZ |
786 | if (!trans) { |
787 | path->skip_locking = 1; | |
788 | path->search_commit_root = 1; | |
789 | } | |
790 | again: | |
791 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
792 | &key, path, 0, 0); | |
793 | if (ret < 0) | |
794 | goto out_free; | |
795 | ||
3173a18f JB |
796 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
797 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
798 | key.offset = root->leafsize; | |
799 | btrfs_release_path(path); | |
800 | goto again; | |
801 | } | |
802 | ||
a22285a6 YZ |
803 | if (ret == 0) { |
804 | leaf = path->nodes[0]; | |
805 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
806 | if (item_size >= sizeof(*ei)) { | |
807 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
808 | struct btrfs_extent_item); | |
809 | num_refs = btrfs_extent_refs(leaf, ei); | |
810 | extent_flags = btrfs_extent_flags(leaf, ei); | |
811 | } else { | |
812 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
813 | struct btrfs_extent_item_v0 *ei0; | |
814 | BUG_ON(item_size != sizeof(*ei0)); | |
815 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
816 | struct btrfs_extent_item_v0); | |
817 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
818 | /* FIXME: this isn't correct for data */ | |
819 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
820 | #else | |
821 | BUG(); | |
822 | #endif | |
823 | } | |
824 | BUG_ON(num_refs == 0); | |
825 | } else { | |
826 | num_refs = 0; | |
827 | extent_flags = 0; | |
828 | ret = 0; | |
829 | } | |
830 | ||
831 | if (!trans) | |
832 | goto out; | |
833 | ||
834 | delayed_refs = &trans->transaction->delayed_refs; | |
835 | spin_lock(&delayed_refs->lock); | |
836 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
837 | if (head) { | |
838 | if (!mutex_trylock(&head->mutex)) { | |
839 | atomic_inc(&head->node.refs); | |
840 | spin_unlock(&delayed_refs->lock); | |
841 | ||
b3b4aa74 | 842 | btrfs_release_path(path); |
a22285a6 | 843 | |
8cc33e5c DS |
844 | /* |
845 | * Mutex was contended, block until it's released and try | |
846 | * again | |
847 | */ | |
a22285a6 YZ |
848 | mutex_lock(&head->mutex); |
849 | mutex_unlock(&head->mutex); | |
850 | btrfs_put_delayed_ref(&head->node); | |
851 | goto again; | |
852 | } | |
853 | if (head->extent_op && head->extent_op->update_flags) | |
854 | extent_flags |= head->extent_op->flags_to_set; | |
855 | else | |
856 | BUG_ON(num_refs == 0); | |
857 | ||
858 | num_refs += head->node.ref_mod; | |
859 | mutex_unlock(&head->mutex); | |
860 | } | |
861 | spin_unlock(&delayed_refs->lock); | |
862 | out: | |
863 | WARN_ON(num_refs == 0); | |
864 | if (refs) | |
865 | *refs = num_refs; | |
866 | if (flags) | |
867 | *flags = extent_flags; | |
868 | out_free: | |
869 | btrfs_free_path(path); | |
870 | return ret; | |
871 | } | |
872 | ||
d8d5f3e1 CM |
873 | /* |
874 | * Back reference rules. Back refs have three main goals: | |
875 | * | |
876 | * 1) differentiate between all holders of references to an extent so that | |
877 | * when a reference is dropped we can make sure it was a valid reference | |
878 | * before freeing the extent. | |
879 | * | |
880 | * 2) Provide enough information to quickly find the holders of an extent | |
881 | * if we notice a given block is corrupted or bad. | |
882 | * | |
883 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
884 | * maintenance. This is actually the same as #2, but with a slightly | |
885 | * different use case. | |
886 | * | |
5d4f98a2 YZ |
887 | * There are two kinds of back refs. The implicit back refs is optimized |
888 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
889 | * back refs of this kind provide information about the block's owner tree | |
890 | * and the pointer's key. These information allow us to find the block by | |
891 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
892 | * referenced by their owner trees. The location of tree block is recorded | |
893 | * in the back refs. Actually the full back refs is generic, and can be | |
894 | * used in all cases the implicit back refs is used. The major shortcoming | |
895 | * of the full back refs is its overhead. Every time a tree block gets | |
896 | * COWed, we have to update back refs entry for all pointers in it. | |
897 | * | |
898 | * For a newly allocated tree block, we use implicit back refs for | |
899 | * pointers in it. This means most tree related operations only involve | |
900 | * implicit back refs. For a tree block created in old transaction, the | |
901 | * only way to drop a reference to it is COW it. So we can detect the | |
902 | * event that tree block loses its owner tree's reference and do the | |
903 | * back refs conversion. | |
904 | * | |
905 | * When a tree block is COW'd through a tree, there are four cases: | |
906 | * | |
907 | * The reference count of the block is one and the tree is the block's | |
908 | * owner tree. Nothing to do in this case. | |
909 | * | |
910 | * The reference count of the block is one and the tree is not the | |
911 | * block's owner tree. In this case, full back refs is used for pointers | |
912 | * in the block. Remove these full back refs, add implicit back refs for | |
913 | * every pointers in the new block. | |
914 | * | |
915 | * The reference count of the block is greater than one and the tree is | |
916 | * the block's owner tree. In this case, implicit back refs is used for | |
917 | * pointers in the block. Add full back refs for every pointers in the | |
918 | * block, increase lower level extents' reference counts. The original | |
919 | * implicit back refs are entailed to the new block. | |
920 | * | |
921 | * The reference count of the block is greater than one and the tree is | |
922 | * not the block's owner tree. Add implicit back refs for every pointer in | |
923 | * the new block, increase lower level extents' reference count. | |
924 | * | |
925 | * Back Reference Key composing: | |
926 | * | |
927 | * The key objectid corresponds to the first byte in the extent, | |
928 | * The key type is used to differentiate between types of back refs. | |
929 | * There are different meanings of the key offset for different types | |
930 | * of back refs. | |
931 | * | |
d8d5f3e1 CM |
932 | * File extents can be referenced by: |
933 | * | |
934 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 935 | * - different files inside a single subvolume |
d8d5f3e1 CM |
936 | * - different offsets inside a file (bookend extents in file.c) |
937 | * | |
5d4f98a2 | 938 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
939 | * |
940 | * - Objectid of the subvolume root | |
d8d5f3e1 | 941 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
942 | * - original offset in the file |
943 | * - how many bookend extents | |
d8d5f3e1 | 944 | * |
5d4f98a2 YZ |
945 | * The key offset for the implicit back refs is hash of the first |
946 | * three fields. | |
d8d5f3e1 | 947 | * |
5d4f98a2 | 948 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 949 | * |
5d4f98a2 | 950 | * - number of pointers in the tree leaf |
d8d5f3e1 | 951 | * |
5d4f98a2 YZ |
952 | * The key offset for the implicit back refs is the first byte of |
953 | * the tree leaf | |
d8d5f3e1 | 954 | * |
5d4f98a2 YZ |
955 | * When a file extent is allocated, The implicit back refs is used. |
956 | * the fields are filled in: | |
d8d5f3e1 | 957 | * |
5d4f98a2 | 958 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 959 | * |
5d4f98a2 YZ |
960 | * When a file extent is removed file truncation, we find the |
961 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 962 | * |
5d4f98a2 | 963 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 964 | * |
5d4f98a2 | 965 | * Btree extents can be referenced by: |
d8d5f3e1 | 966 | * |
5d4f98a2 | 967 | * - Different subvolumes |
d8d5f3e1 | 968 | * |
5d4f98a2 YZ |
969 | * Both the implicit back refs and the full back refs for tree blocks |
970 | * only consist of key. The key offset for the implicit back refs is | |
971 | * objectid of block's owner tree. The key offset for the full back refs | |
972 | * is the first byte of parent block. | |
d8d5f3e1 | 973 | * |
5d4f98a2 YZ |
974 | * When implicit back refs is used, information about the lowest key and |
975 | * level of the tree block are required. These information are stored in | |
976 | * tree block info structure. | |
d8d5f3e1 | 977 | */ |
31840ae1 | 978 | |
5d4f98a2 YZ |
979 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
980 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
981 | struct btrfs_root *root, | |
982 | struct btrfs_path *path, | |
983 | u64 owner, u32 extra_size) | |
7bb86316 | 984 | { |
5d4f98a2 YZ |
985 | struct btrfs_extent_item *item; |
986 | struct btrfs_extent_item_v0 *ei0; | |
987 | struct btrfs_extent_ref_v0 *ref0; | |
988 | struct btrfs_tree_block_info *bi; | |
989 | struct extent_buffer *leaf; | |
7bb86316 | 990 | struct btrfs_key key; |
5d4f98a2 YZ |
991 | struct btrfs_key found_key; |
992 | u32 new_size = sizeof(*item); | |
993 | u64 refs; | |
994 | int ret; | |
995 | ||
996 | leaf = path->nodes[0]; | |
997 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
998 | ||
999 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1000 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1001 | struct btrfs_extent_item_v0); | |
1002 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1003 | ||
1004 | if (owner == (u64)-1) { | |
1005 | while (1) { | |
1006 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1007 | ret = btrfs_next_leaf(root, path); | |
1008 | if (ret < 0) | |
1009 | return ret; | |
79787eaa | 1010 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1011 | leaf = path->nodes[0]; |
1012 | } | |
1013 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1014 | path->slots[0]); | |
1015 | BUG_ON(key.objectid != found_key.objectid); | |
1016 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1017 | path->slots[0]++; | |
1018 | continue; | |
1019 | } | |
1020 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1021 | struct btrfs_extent_ref_v0); | |
1022 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1023 | break; | |
1024 | } | |
1025 | } | |
b3b4aa74 | 1026 | btrfs_release_path(path); |
5d4f98a2 YZ |
1027 | |
1028 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1029 | new_size += sizeof(*bi); | |
1030 | ||
1031 | new_size -= sizeof(*ei0); | |
1032 | ret = btrfs_search_slot(trans, root, &key, path, | |
1033 | new_size + extra_size, 1); | |
1034 | if (ret < 0) | |
1035 | return ret; | |
79787eaa | 1036 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1037 | |
143bede5 | 1038 | btrfs_extend_item(trans, root, path, new_size); |
5d4f98a2 YZ |
1039 | |
1040 | leaf = path->nodes[0]; | |
1041 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1042 | btrfs_set_extent_refs(leaf, item, refs); | |
1043 | /* FIXME: get real generation */ | |
1044 | btrfs_set_extent_generation(leaf, item, 0); | |
1045 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1046 | btrfs_set_extent_flags(leaf, item, | |
1047 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1048 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1049 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1050 | /* FIXME: get first key of the block */ | |
1051 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1052 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1053 | } else { | |
1054 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1055 | } | |
1056 | btrfs_mark_buffer_dirty(leaf); | |
1057 | return 0; | |
1058 | } | |
1059 | #endif | |
1060 | ||
1061 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1062 | { | |
1063 | u32 high_crc = ~(u32)0; | |
1064 | u32 low_crc = ~(u32)0; | |
1065 | __le64 lenum; | |
1066 | ||
1067 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 1068 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1069 | lenum = cpu_to_le64(owner); |
163e783e | 1070 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1071 | lenum = cpu_to_le64(offset); |
163e783e | 1072 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1073 | |
1074 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1075 | } | |
1076 | ||
1077 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1078 | struct btrfs_extent_data_ref *ref) | |
1079 | { | |
1080 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1081 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1082 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1083 | } | |
1084 | ||
1085 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1086 | struct btrfs_extent_data_ref *ref, | |
1087 | u64 root_objectid, u64 owner, u64 offset) | |
1088 | { | |
1089 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1090 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1091 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1092 | return 0; | |
1093 | return 1; | |
1094 | } | |
1095 | ||
1096 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1097 | struct btrfs_root *root, | |
1098 | struct btrfs_path *path, | |
1099 | u64 bytenr, u64 parent, | |
1100 | u64 root_objectid, | |
1101 | u64 owner, u64 offset) | |
1102 | { | |
1103 | struct btrfs_key key; | |
1104 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1105 | struct extent_buffer *leaf; |
5d4f98a2 | 1106 | u32 nritems; |
74493f7a | 1107 | int ret; |
5d4f98a2 YZ |
1108 | int recow; |
1109 | int err = -ENOENT; | |
74493f7a | 1110 | |
31840ae1 | 1111 | key.objectid = bytenr; |
5d4f98a2 YZ |
1112 | if (parent) { |
1113 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1114 | key.offset = parent; | |
1115 | } else { | |
1116 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1117 | key.offset = hash_extent_data_ref(root_objectid, | |
1118 | owner, offset); | |
1119 | } | |
1120 | again: | |
1121 | recow = 0; | |
1122 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1123 | if (ret < 0) { | |
1124 | err = ret; | |
1125 | goto fail; | |
1126 | } | |
31840ae1 | 1127 | |
5d4f98a2 YZ |
1128 | if (parent) { |
1129 | if (!ret) | |
1130 | return 0; | |
1131 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1132 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1133 | btrfs_release_path(path); |
5d4f98a2 YZ |
1134 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1135 | if (ret < 0) { | |
1136 | err = ret; | |
1137 | goto fail; | |
1138 | } | |
1139 | if (!ret) | |
1140 | return 0; | |
1141 | #endif | |
1142 | goto fail; | |
31840ae1 ZY |
1143 | } |
1144 | ||
1145 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1146 | nritems = btrfs_header_nritems(leaf); |
1147 | while (1) { | |
1148 | if (path->slots[0] >= nritems) { | |
1149 | ret = btrfs_next_leaf(root, path); | |
1150 | if (ret < 0) | |
1151 | err = ret; | |
1152 | if (ret) | |
1153 | goto fail; | |
1154 | ||
1155 | leaf = path->nodes[0]; | |
1156 | nritems = btrfs_header_nritems(leaf); | |
1157 | recow = 1; | |
1158 | } | |
1159 | ||
1160 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1161 | if (key.objectid != bytenr || | |
1162 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1163 | goto fail; | |
1164 | ||
1165 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1166 | struct btrfs_extent_data_ref); | |
1167 | ||
1168 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1169 | owner, offset)) { | |
1170 | if (recow) { | |
b3b4aa74 | 1171 | btrfs_release_path(path); |
5d4f98a2 YZ |
1172 | goto again; |
1173 | } | |
1174 | err = 0; | |
1175 | break; | |
1176 | } | |
1177 | path->slots[0]++; | |
31840ae1 | 1178 | } |
5d4f98a2 YZ |
1179 | fail: |
1180 | return err; | |
31840ae1 ZY |
1181 | } |
1182 | ||
5d4f98a2 YZ |
1183 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1184 | struct btrfs_root *root, | |
1185 | struct btrfs_path *path, | |
1186 | u64 bytenr, u64 parent, | |
1187 | u64 root_objectid, u64 owner, | |
1188 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1189 | { |
1190 | struct btrfs_key key; | |
1191 | struct extent_buffer *leaf; | |
5d4f98a2 | 1192 | u32 size; |
31840ae1 ZY |
1193 | u32 num_refs; |
1194 | int ret; | |
74493f7a | 1195 | |
74493f7a | 1196 | key.objectid = bytenr; |
5d4f98a2 YZ |
1197 | if (parent) { |
1198 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1199 | key.offset = parent; | |
1200 | size = sizeof(struct btrfs_shared_data_ref); | |
1201 | } else { | |
1202 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1203 | key.offset = hash_extent_data_ref(root_objectid, | |
1204 | owner, offset); | |
1205 | size = sizeof(struct btrfs_extent_data_ref); | |
1206 | } | |
74493f7a | 1207 | |
5d4f98a2 YZ |
1208 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1209 | if (ret && ret != -EEXIST) | |
1210 | goto fail; | |
1211 | ||
1212 | leaf = path->nodes[0]; | |
1213 | if (parent) { | |
1214 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1215 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1216 | struct btrfs_shared_data_ref); |
1217 | if (ret == 0) { | |
1218 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1219 | } else { | |
1220 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1221 | num_refs += refs_to_add; | |
1222 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1223 | } |
5d4f98a2 YZ |
1224 | } else { |
1225 | struct btrfs_extent_data_ref *ref; | |
1226 | while (ret == -EEXIST) { | |
1227 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1228 | struct btrfs_extent_data_ref); | |
1229 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1230 | owner, offset)) | |
1231 | break; | |
b3b4aa74 | 1232 | btrfs_release_path(path); |
5d4f98a2 YZ |
1233 | key.offset++; |
1234 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1235 | size); | |
1236 | if (ret && ret != -EEXIST) | |
1237 | goto fail; | |
31840ae1 | 1238 | |
5d4f98a2 YZ |
1239 | leaf = path->nodes[0]; |
1240 | } | |
1241 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1242 | struct btrfs_extent_data_ref); | |
1243 | if (ret == 0) { | |
1244 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1245 | root_objectid); | |
1246 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1247 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1248 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1249 | } else { | |
1250 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1251 | num_refs += refs_to_add; | |
1252 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1253 | } |
31840ae1 | 1254 | } |
5d4f98a2 YZ |
1255 | btrfs_mark_buffer_dirty(leaf); |
1256 | ret = 0; | |
1257 | fail: | |
b3b4aa74 | 1258 | btrfs_release_path(path); |
7bb86316 | 1259 | return ret; |
74493f7a CM |
1260 | } |
1261 | ||
5d4f98a2 YZ |
1262 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1263 | struct btrfs_root *root, | |
1264 | struct btrfs_path *path, | |
1265 | int refs_to_drop) | |
31840ae1 | 1266 | { |
5d4f98a2 YZ |
1267 | struct btrfs_key key; |
1268 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1269 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1270 | struct extent_buffer *leaf; |
5d4f98a2 | 1271 | u32 num_refs = 0; |
31840ae1 ZY |
1272 | int ret = 0; |
1273 | ||
1274 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1275 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1276 | ||
1277 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1278 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1279 | struct btrfs_extent_data_ref); | |
1280 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1281 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1282 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1283 | struct btrfs_shared_data_ref); | |
1284 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1285 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1286 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1287 | struct btrfs_extent_ref_v0 *ref0; | |
1288 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1289 | struct btrfs_extent_ref_v0); | |
1290 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1291 | #endif | |
1292 | } else { | |
1293 | BUG(); | |
1294 | } | |
1295 | ||
56bec294 CM |
1296 | BUG_ON(num_refs < refs_to_drop); |
1297 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1298 | |
31840ae1 ZY |
1299 | if (num_refs == 0) { |
1300 | ret = btrfs_del_item(trans, root, path); | |
1301 | } else { | |
5d4f98a2 YZ |
1302 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1303 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1304 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1305 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1306 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1307 | else { | |
1308 | struct btrfs_extent_ref_v0 *ref0; | |
1309 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1310 | struct btrfs_extent_ref_v0); | |
1311 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1312 | } | |
1313 | #endif | |
31840ae1 ZY |
1314 | btrfs_mark_buffer_dirty(leaf); |
1315 | } | |
31840ae1 ZY |
1316 | return ret; |
1317 | } | |
1318 | ||
5d4f98a2 YZ |
1319 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1320 | struct btrfs_path *path, | |
1321 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1322 | { |
5d4f98a2 YZ |
1323 | struct btrfs_key key; |
1324 | struct extent_buffer *leaf; | |
1325 | struct btrfs_extent_data_ref *ref1; | |
1326 | struct btrfs_shared_data_ref *ref2; | |
1327 | u32 num_refs = 0; | |
1328 | ||
1329 | leaf = path->nodes[0]; | |
1330 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1331 | if (iref) { | |
1332 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1333 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1334 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1335 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1336 | } else { | |
1337 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1338 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1339 | } | |
1340 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1341 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1342 | struct btrfs_extent_data_ref); | |
1343 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1344 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1345 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1346 | struct btrfs_shared_data_ref); | |
1347 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1348 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1349 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1350 | struct btrfs_extent_ref_v0 *ref0; | |
1351 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1352 | struct btrfs_extent_ref_v0); | |
1353 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1354 | #endif |
5d4f98a2 YZ |
1355 | } else { |
1356 | WARN_ON(1); | |
1357 | } | |
1358 | return num_refs; | |
1359 | } | |
15916de8 | 1360 | |
5d4f98a2 YZ |
1361 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1362 | struct btrfs_root *root, | |
1363 | struct btrfs_path *path, | |
1364 | u64 bytenr, u64 parent, | |
1365 | u64 root_objectid) | |
1f3c79a2 | 1366 | { |
5d4f98a2 | 1367 | struct btrfs_key key; |
1f3c79a2 | 1368 | int ret; |
1f3c79a2 | 1369 | |
5d4f98a2 YZ |
1370 | key.objectid = bytenr; |
1371 | if (parent) { | |
1372 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1373 | key.offset = parent; | |
1374 | } else { | |
1375 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1376 | key.offset = root_objectid; | |
1f3c79a2 LH |
1377 | } |
1378 | ||
5d4f98a2 YZ |
1379 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1380 | if (ret > 0) | |
1381 | ret = -ENOENT; | |
1382 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1383 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1384 | btrfs_release_path(path); |
5d4f98a2 YZ |
1385 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1386 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1387 | if (ret > 0) | |
1388 | ret = -ENOENT; | |
1389 | } | |
1f3c79a2 | 1390 | #endif |
5d4f98a2 | 1391 | return ret; |
1f3c79a2 LH |
1392 | } |
1393 | ||
5d4f98a2 YZ |
1394 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1395 | struct btrfs_root *root, | |
1396 | struct btrfs_path *path, | |
1397 | u64 bytenr, u64 parent, | |
1398 | u64 root_objectid) | |
31840ae1 | 1399 | { |
5d4f98a2 | 1400 | struct btrfs_key key; |
31840ae1 | 1401 | int ret; |
31840ae1 | 1402 | |
5d4f98a2 YZ |
1403 | key.objectid = bytenr; |
1404 | if (parent) { | |
1405 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1406 | key.offset = parent; | |
1407 | } else { | |
1408 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1409 | key.offset = root_objectid; | |
1410 | } | |
1411 | ||
1412 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1413 | btrfs_release_path(path); |
31840ae1 ZY |
1414 | return ret; |
1415 | } | |
1416 | ||
5d4f98a2 | 1417 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1418 | { |
5d4f98a2 YZ |
1419 | int type; |
1420 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1421 | if (parent > 0) | |
1422 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1423 | else | |
1424 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1425 | } else { | |
1426 | if (parent > 0) | |
1427 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1428 | else | |
1429 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1430 | } | |
1431 | return type; | |
31840ae1 | 1432 | } |
56bec294 | 1433 | |
2c47e605 YZ |
1434 | static int find_next_key(struct btrfs_path *path, int level, |
1435 | struct btrfs_key *key) | |
56bec294 | 1436 | |
02217ed2 | 1437 | { |
2c47e605 | 1438 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1439 | if (!path->nodes[level]) |
1440 | break; | |
5d4f98a2 YZ |
1441 | if (path->slots[level] + 1 >= |
1442 | btrfs_header_nritems(path->nodes[level])) | |
1443 | continue; | |
1444 | if (level == 0) | |
1445 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1446 | path->slots[level] + 1); | |
1447 | else | |
1448 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1449 | path->slots[level] + 1); | |
1450 | return 0; | |
1451 | } | |
1452 | return 1; | |
1453 | } | |
037e6390 | 1454 | |
5d4f98a2 YZ |
1455 | /* |
1456 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1457 | * to the address of inline back ref, and 0 is returned. | |
1458 | * | |
1459 | * if back ref isn't found, *ref_ret is set to the address where it | |
1460 | * should be inserted, and -ENOENT is returned. | |
1461 | * | |
1462 | * if insert is true and there are too many inline back refs, the path | |
1463 | * points to the extent item, and -EAGAIN is returned. | |
1464 | * | |
1465 | * NOTE: inline back refs are ordered in the same way that back ref | |
1466 | * items in the tree are ordered. | |
1467 | */ | |
1468 | static noinline_for_stack | |
1469 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1470 | struct btrfs_root *root, | |
1471 | struct btrfs_path *path, | |
1472 | struct btrfs_extent_inline_ref **ref_ret, | |
1473 | u64 bytenr, u64 num_bytes, | |
1474 | u64 parent, u64 root_objectid, | |
1475 | u64 owner, u64 offset, int insert) | |
1476 | { | |
1477 | struct btrfs_key key; | |
1478 | struct extent_buffer *leaf; | |
1479 | struct btrfs_extent_item *ei; | |
1480 | struct btrfs_extent_inline_ref *iref; | |
1481 | u64 flags; | |
1482 | u64 item_size; | |
1483 | unsigned long ptr; | |
1484 | unsigned long end; | |
1485 | int extra_size; | |
1486 | int type; | |
1487 | int want; | |
1488 | int ret; | |
1489 | int err = 0; | |
3173a18f JB |
1490 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
1491 | SKINNY_METADATA); | |
26b8003f | 1492 | |
db94535d | 1493 | key.objectid = bytenr; |
31840ae1 | 1494 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1495 | key.offset = num_bytes; |
31840ae1 | 1496 | |
5d4f98a2 YZ |
1497 | want = extent_ref_type(parent, owner); |
1498 | if (insert) { | |
1499 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1500 | path->keep_locks = 1; |
5d4f98a2 YZ |
1501 | } else |
1502 | extra_size = -1; | |
3173a18f JB |
1503 | |
1504 | /* | |
1505 | * Owner is our parent level, so we can just add one to get the level | |
1506 | * for the block we are interested in. | |
1507 | */ | |
1508 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1509 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1510 | key.offset = owner; | |
1511 | } | |
1512 | ||
1513 | again: | |
5d4f98a2 | 1514 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1515 | if (ret < 0) { |
5d4f98a2 YZ |
1516 | err = ret; |
1517 | goto out; | |
1518 | } | |
3173a18f JB |
1519 | |
1520 | /* | |
1521 | * We may be a newly converted file system which still has the old fat | |
1522 | * extent entries for metadata, so try and see if we have one of those. | |
1523 | */ | |
1524 | if (ret > 0 && skinny_metadata) { | |
1525 | skinny_metadata = false; | |
1526 | if (path->slots[0]) { | |
1527 | path->slots[0]--; | |
1528 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1529 | path->slots[0]); | |
1530 | if (key.objectid == bytenr && | |
1531 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1532 | key.offset == num_bytes) | |
1533 | ret = 0; | |
1534 | } | |
1535 | if (ret) { | |
1536 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1537 | key.offset = num_bytes; | |
1538 | btrfs_release_path(path); | |
1539 | goto again; | |
1540 | } | |
1541 | } | |
1542 | ||
79787eaa JM |
1543 | if (ret && !insert) { |
1544 | err = -ENOENT; | |
1545 | goto out; | |
492104c8 JB |
1546 | } else if (ret) { |
1547 | err = -EIO; | |
1548 | WARN_ON(1); | |
1549 | goto out; | |
79787eaa | 1550 | } |
5d4f98a2 YZ |
1551 | |
1552 | leaf = path->nodes[0]; | |
1553 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1554 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1555 | if (item_size < sizeof(*ei)) { | |
1556 | if (!insert) { | |
1557 | err = -ENOENT; | |
1558 | goto out; | |
1559 | } | |
1560 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1561 | extra_size); | |
1562 | if (ret < 0) { | |
1563 | err = ret; | |
1564 | goto out; | |
1565 | } | |
1566 | leaf = path->nodes[0]; | |
1567 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1568 | } | |
1569 | #endif | |
1570 | BUG_ON(item_size < sizeof(*ei)); | |
1571 | ||
5d4f98a2 YZ |
1572 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1573 | flags = btrfs_extent_flags(leaf, ei); | |
1574 | ||
1575 | ptr = (unsigned long)(ei + 1); | |
1576 | end = (unsigned long)ei + item_size; | |
1577 | ||
3173a18f | 1578 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1579 | ptr += sizeof(struct btrfs_tree_block_info); |
1580 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1581 | } |
1582 | ||
1583 | err = -ENOENT; | |
1584 | while (1) { | |
1585 | if (ptr >= end) { | |
1586 | WARN_ON(ptr > end); | |
1587 | break; | |
1588 | } | |
1589 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1590 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1591 | if (want < type) | |
1592 | break; | |
1593 | if (want > type) { | |
1594 | ptr += btrfs_extent_inline_ref_size(type); | |
1595 | continue; | |
1596 | } | |
1597 | ||
1598 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1599 | struct btrfs_extent_data_ref *dref; | |
1600 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1601 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1602 | owner, offset)) { | |
1603 | err = 0; | |
1604 | break; | |
1605 | } | |
1606 | if (hash_extent_data_ref_item(leaf, dref) < | |
1607 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1608 | break; | |
1609 | } else { | |
1610 | u64 ref_offset; | |
1611 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1612 | if (parent > 0) { | |
1613 | if (parent == ref_offset) { | |
1614 | err = 0; | |
1615 | break; | |
1616 | } | |
1617 | if (ref_offset < parent) | |
1618 | break; | |
1619 | } else { | |
1620 | if (root_objectid == ref_offset) { | |
1621 | err = 0; | |
1622 | break; | |
1623 | } | |
1624 | if (ref_offset < root_objectid) | |
1625 | break; | |
1626 | } | |
1627 | } | |
1628 | ptr += btrfs_extent_inline_ref_size(type); | |
1629 | } | |
1630 | if (err == -ENOENT && insert) { | |
1631 | if (item_size + extra_size >= | |
1632 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1633 | err = -EAGAIN; | |
1634 | goto out; | |
1635 | } | |
1636 | /* | |
1637 | * To add new inline back ref, we have to make sure | |
1638 | * there is no corresponding back ref item. | |
1639 | * For simplicity, we just do not add new inline back | |
1640 | * ref if there is any kind of item for this block | |
1641 | */ | |
2c47e605 YZ |
1642 | if (find_next_key(path, 0, &key) == 0 && |
1643 | key.objectid == bytenr && | |
85d4198e | 1644 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1645 | err = -EAGAIN; |
1646 | goto out; | |
1647 | } | |
1648 | } | |
1649 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1650 | out: | |
85d4198e | 1651 | if (insert) { |
5d4f98a2 YZ |
1652 | path->keep_locks = 0; |
1653 | btrfs_unlock_up_safe(path, 1); | |
1654 | } | |
1655 | return err; | |
1656 | } | |
1657 | ||
1658 | /* | |
1659 | * helper to add new inline back ref | |
1660 | */ | |
1661 | static noinline_for_stack | |
143bede5 JM |
1662 | void setup_inline_extent_backref(struct btrfs_trans_handle *trans, |
1663 | struct btrfs_root *root, | |
1664 | struct btrfs_path *path, | |
1665 | struct btrfs_extent_inline_ref *iref, | |
1666 | u64 parent, u64 root_objectid, | |
1667 | u64 owner, u64 offset, int refs_to_add, | |
1668 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1669 | { |
1670 | struct extent_buffer *leaf; | |
1671 | struct btrfs_extent_item *ei; | |
1672 | unsigned long ptr; | |
1673 | unsigned long end; | |
1674 | unsigned long item_offset; | |
1675 | u64 refs; | |
1676 | int size; | |
1677 | int type; | |
5d4f98a2 YZ |
1678 | |
1679 | leaf = path->nodes[0]; | |
1680 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1681 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1682 | ||
1683 | type = extent_ref_type(parent, owner); | |
1684 | size = btrfs_extent_inline_ref_size(type); | |
1685 | ||
143bede5 | 1686 | btrfs_extend_item(trans, root, path, size); |
5d4f98a2 YZ |
1687 | |
1688 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1689 | refs = btrfs_extent_refs(leaf, ei); | |
1690 | refs += refs_to_add; | |
1691 | btrfs_set_extent_refs(leaf, ei, refs); | |
1692 | if (extent_op) | |
1693 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1694 | ||
1695 | ptr = (unsigned long)ei + item_offset; | |
1696 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1697 | if (ptr < end - size) | |
1698 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1699 | end - size - ptr); | |
1700 | ||
1701 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1702 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1703 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1704 | struct btrfs_extent_data_ref *dref; | |
1705 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1706 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1707 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1708 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1709 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1710 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1711 | struct btrfs_shared_data_ref *sref; | |
1712 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1713 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1714 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1715 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1716 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1717 | } else { | |
1718 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1719 | } | |
1720 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1721 | } |
1722 | ||
1723 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1724 | struct btrfs_root *root, | |
1725 | struct btrfs_path *path, | |
1726 | struct btrfs_extent_inline_ref **ref_ret, | |
1727 | u64 bytenr, u64 num_bytes, u64 parent, | |
1728 | u64 root_objectid, u64 owner, u64 offset) | |
1729 | { | |
1730 | int ret; | |
1731 | ||
1732 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1733 | bytenr, num_bytes, parent, | |
1734 | root_objectid, owner, offset, 0); | |
1735 | if (ret != -ENOENT) | |
54aa1f4d | 1736 | return ret; |
5d4f98a2 | 1737 | |
b3b4aa74 | 1738 | btrfs_release_path(path); |
5d4f98a2 YZ |
1739 | *ref_ret = NULL; |
1740 | ||
1741 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1742 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1743 | root_objectid); | |
1744 | } else { | |
1745 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1746 | root_objectid, owner, offset); | |
b9473439 | 1747 | } |
5d4f98a2 YZ |
1748 | return ret; |
1749 | } | |
31840ae1 | 1750 | |
5d4f98a2 YZ |
1751 | /* |
1752 | * helper to update/remove inline back ref | |
1753 | */ | |
1754 | static noinline_for_stack | |
143bede5 JM |
1755 | void update_inline_extent_backref(struct btrfs_trans_handle *trans, |
1756 | struct btrfs_root *root, | |
1757 | struct btrfs_path *path, | |
1758 | struct btrfs_extent_inline_ref *iref, | |
1759 | int refs_to_mod, | |
1760 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1761 | { |
1762 | struct extent_buffer *leaf; | |
1763 | struct btrfs_extent_item *ei; | |
1764 | struct btrfs_extent_data_ref *dref = NULL; | |
1765 | struct btrfs_shared_data_ref *sref = NULL; | |
1766 | unsigned long ptr; | |
1767 | unsigned long end; | |
1768 | u32 item_size; | |
1769 | int size; | |
1770 | int type; | |
5d4f98a2 YZ |
1771 | u64 refs; |
1772 | ||
1773 | leaf = path->nodes[0]; | |
1774 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1775 | refs = btrfs_extent_refs(leaf, ei); | |
1776 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1777 | refs += refs_to_mod; | |
1778 | btrfs_set_extent_refs(leaf, ei, refs); | |
1779 | if (extent_op) | |
1780 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1781 | ||
1782 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1783 | ||
1784 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1785 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1786 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1787 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1788 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1789 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1790 | } else { | |
1791 | refs = 1; | |
1792 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1793 | } |
31840ae1 | 1794 | |
5d4f98a2 YZ |
1795 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1796 | refs += refs_to_mod; | |
1797 | ||
1798 | if (refs > 0) { | |
1799 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1800 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1801 | else | |
1802 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1803 | } else { | |
1804 | size = btrfs_extent_inline_ref_size(type); | |
1805 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1806 | ptr = (unsigned long)iref; | |
1807 | end = (unsigned long)ei + item_size; | |
1808 | if (ptr + size < end) | |
1809 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1810 | end - ptr - size); | |
1811 | item_size -= size; | |
143bede5 | 1812 | btrfs_truncate_item(trans, root, path, item_size, 1); |
5d4f98a2 YZ |
1813 | } |
1814 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1815 | } |
1816 | ||
1817 | static noinline_for_stack | |
1818 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1819 | struct btrfs_root *root, | |
1820 | struct btrfs_path *path, | |
1821 | u64 bytenr, u64 num_bytes, u64 parent, | |
1822 | u64 root_objectid, u64 owner, | |
1823 | u64 offset, int refs_to_add, | |
1824 | struct btrfs_delayed_extent_op *extent_op) | |
1825 | { | |
1826 | struct btrfs_extent_inline_ref *iref; | |
1827 | int ret; | |
1828 | ||
1829 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1830 | bytenr, num_bytes, parent, | |
1831 | root_objectid, owner, offset, 1); | |
1832 | if (ret == 0) { | |
1833 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
143bede5 JM |
1834 | update_inline_extent_backref(trans, root, path, iref, |
1835 | refs_to_add, extent_op); | |
5d4f98a2 | 1836 | } else if (ret == -ENOENT) { |
143bede5 JM |
1837 | setup_inline_extent_backref(trans, root, path, iref, parent, |
1838 | root_objectid, owner, offset, | |
1839 | refs_to_add, extent_op); | |
1840 | ret = 0; | |
771ed689 | 1841 | } |
5d4f98a2 YZ |
1842 | return ret; |
1843 | } | |
31840ae1 | 1844 | |
5d4f98a2 YZ |
1845 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1846 | struct btrfs_root *root, | |
1847 | struct btrfs_path *path, | |
1848 | u64 bytenr, u64 parent, u64 root_objectid, | |
1849 | u64 owner, u64 offset, int refs_to_add) | |
1850 | { | |
1851 | int ret; | |
1852 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1853 | BUG_ON(refs_to_add != 1); | |
1854 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1855 | parent, root_objectid); | |
1856 | } else { | |
1857 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1858 | parent, root_objectid, | |
1859 | owner, offset, refs_to_add); | |
1860 | } | |
1861 | return ret; | |
1862 | } | |
56bec294 | 1863 | |
5d4f98a2 YZ |
1864 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1865 | struct btrfs_root *root, | |
1866 | struct btrfs_path *path, | |
1867 | struct btrfs_extent_inline_ref *iref, | |
1868 | int refs_to_drop, int is_data) | |
1869 | { | |
143bede5 | 1870 | int ret = 0; |
b9473439 | 1871 | |
5d4f98a2 YZ |
1872 | BUG_ON(!is_data && refs_to_drop != 1); |
1873 | if (iref) { | |
143bede5 JM |
1874 | update_inline_extent_backref(trans, root, path, iref, |
1875 | -refs_to_drop, NULL); | |
5d4f98a2 YZ |
1876 | } else if (is_data) { |
1877 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1878 | } else { | |
1879 | ret = btrfs_del_item(trans, root, path); | |
1880 | } | |
1881 | return ret; | |
1882 | } | |
1883 | ||
5378e607 | 1884 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1885 | u64 start, u64 len) |
1886 | { | |
5378e607 | 1887 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1888 | } |
5d4f98a2 YZ |
1889 | |
1890 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1891 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1892 | { |
5d4f98a2 | 1893 | int ret; |
5378e607 | 1894 | u64 discarded_bytes = 0; |
a1d3c478 | 1895 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1896 | |
e244a0ae | 1897 | |
5d4f98a2 | 1898 | /* Tell the block device(s) that the sectors can be discarded */ |
3ec706c8 | 1899 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
a1d3c478 | 1900 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 1901 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1902 | if (!ret) { |
a1d3c478 | 1903 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1904 | int i; |
1905 | ||
5d4f98a2 | 1906 | |
a1d3c478 | 1907 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d5e2003c JB |
1908 | if (!stripe->dev->can_discard) |
1909 | continue; | |
1910 | ||
5378e607 LD |
1911 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1912 | stripe->physical, | |
1913 | stripe->length); | |
1914 | if (!ret) | |
1915 | discarded_bytes += stripe->length; | |
1916 | else if (ret != -EOPNOTSUPP) | |
79787eaa | 1917 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
1918 | |
1919 | /* | |
1920 | * Just in case we get back EOPNOTSUPP for some reason, | |
1921 | * just ignore the return value so we don't screw up | |
1922 | * people calling discard_extent. | |
1923 | */ | |
1924 | ret = 0; | |
5d4f98a2 | 1925 | } |
a1d3c478 | 1926 | kfree(bbio); |
5d4f98a2 | 1927 | } |
5378e607 LD |
1928 | |
1929 | if (actual_bytes) | |
1930 | *actual_bytes = discarded_bytes; | |
1931 | ||
5d4f98a2 | 1932 | |
53b381b3 DW |
1933 | if (ret == -EOPNOTSUPP) |
1934 | ret = 0; | |
5d4f98a2 | 1935 | return ret; |
5d4f98a2 YZ |
1936 | } |
1937 | ||
79787eaa | 1938 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
1939 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
1940 | struct btrfs_root *root, | |
1941 | u64 bytenr, u64 num_bytes, u64 parent, | |
66d7e7f0 | 1942 | u64 root_objectid, u64 owner, u64 offset, int for_cow) |
5d4f98a2 YZ |
1943 | { |
1944 | int ret; | |
66d7e7f0 AJ |
1945 | struct btrfs_fs_info *fs_info = root->fs_info; |
1946 | ||
5d4f98a2 YZ |
1947 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
1948 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1949 | ||
1950 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
1951 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
1952 | num_bytes, | |
5d4f98a2 | 1953 | parent, root_objectid, (int)owner, |
66d7e7f0 | 1954 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 1955 | } else { |
66d7e7f0 AJ |
1956 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
1957 | num_bytes, | |
5d4f98a2 | 1958 | parent, root_objectid, owner, offset, |
66d7e7f0 | 1959 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 YZ |
1960 | } |
1961 | return ret; | |
1962 | } | |
1963 | ||
1964 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1965 | struct btrfs_root *root, | |
1966 | u64 bytenr, u64 num_bytes, | |
1967 | u64 parent, u64 root_objectid, | |
1968 | u64 owner, u64 offset, int refs_to_add, | |
1969 | struct btrfs_delayed_extent_op *extent_op) | |
1970 | { | |
1971 | struct btrfs_path *path; | |
1972 | struct extent_buffer *leaf; | |
1973 | struct btrfs_extent_item *item; | |
1974 | u64 refs; | |
1975 | int ret; | |
1976 | int err = 0; | |
1977 | ||
1978 | path = btrfs_alloc_path(); | |
1979 | if (!path) | |
1980 | return -ENOMEM; | |
1981 | ||
1982 | path->reada = 1; | |
1983 | path->leave_spinning = 1; | |
1984 | /* this will setup the path even if it fails to insert the back ref */ | |
1985 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1986 | path, bytenr, num_bytes, parent, | |
1987 | root_objectid, owner, offset, | |
1988 | refs_to_add, extent_op); | |
1989 | if (ret == 0) | |
1990 | goto out; | |
1991 | ||
1992 | if (ret != -EAGAIN) { | |
1993 | err = ret; | |
1994 | goto out; | |
1995 | } | |
1996 | ||
1997 | leaf = path->nodes[0]; | |
1998 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1999 | refs = btrfs_extent_refs(leaf, item); | |
2000 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2001 | if (extent_op) | |
2002 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2003 | |
5d4f98a2 | 2004 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2005 | btrfs_release_path(path); |
56bec294 CM |
2006 | |
2007 | path->reada = 1; | |
b9473439 CM |
2008 | path->leave_spinning = 1; |
2009 | ||
56bec294 CM |
2010 | /* now insert the actual backref */ |
2011 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
2012 | path, bytenr, parent, root_objectid, |
2013 | owner, offset, refs_to_add); | |
79787eaa JM |
2014 | if (ret) |
2015 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 2016 | out: |
56bec294 | 2017 | btrfs_free_path(path); |
5d4f98a2 | 2018 | return err; |
56bec294 CM |
2019 | } |
2020 | ||
5d4f98a2 YZ |
2021 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2022 | struct btrfs_root *root, | |
2023 | struct btrfs_delayed_ref_node *node, | |
2024 | struct btrfs_delayed_extent_op *extent_op, | |
2025 | int insert_reserved) | |
56bec294 | 2026 | { |
5d4f98a2 YZ |
2027 | int ret = 0; |
2028 | struct btrfs_delayed_data_ref *ref; | |
2029 | struct btrfs_key ins; | |
2030 | u64 parent = 0; | |
2031 | u64 ref_root = 0; | |
2032 | u64 flags = 0; | |
2033 | ||
2034 | ins.objectid = node->bytenr; | |
2035 | ins.offset = node->num_bytes; | |
2036 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2037 | ||
2038 | ref = btrfs_delayed_node_to_data_ref(node); | |
2039 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2040 | parent = ref->parent; | |
2041 | else | |
2042 | ref_root = ref->root; | |
2043 | ||
2044 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2045 | if (extent_op) |
5d4f98a2 | 2046 | flags |= extent_op->flags_to_set; |
5d4f98a2 YZ |
2047 | ret = alloc_reserved_file_extent(trans, root, |
2048 | parent, ref_root, flags, | |
2049 | ref->objectid, ref->offset, | |
2050 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
2051 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2052 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2053 | node->num_bytes, parent, | |
2054 | ref_root, ref->objectid, | |
2055 | ref->offset, node->ref_mod, | |
2056 | extent_op); | |
2057 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2058 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2059 | node->num_bytes, parent, | |
2060 | ref_root, ref->objectid, | |
2061 | ref->offset, node->ref_mod, | |
2062 | extent_op); | |
2063 | } else { | |
2064 | BUG(); | |
2065 | } | |
2066 | return ret; | |
2067 | } | |
2068 | ||
2069 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2070 | struct extent_buffer *leaf, | |
2071 | struct btrfs_extent_item *ei) | |
2072 | { | |
2073 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2074 | if (extent_op->update_flags) { | |
2075 | flags |= extent_op->flags_to_set; | |
2076 | btrfs_set_extent_flags(leaf, ei, flags); | |
2077 | } | |
2078 | ||
2079 | if (extent_op->update_key) { | |
2080 | struct btrfs_tree_block_info *bi; | |
2081 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2082 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2083 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2084 | } | |
2085 | } | |
2086 | ||
2087 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2088 | struct btrfs_root *root, | |
2089 | struct btrfs_delayed_ref_node *node, | |
2090 | struct btrfs_delayed_extent_op *extent_op) | |
2091 | { | |
2092 | struct btrfs_key key; | |
2093 | struct btrfs_path *path; | |
2094 | struct btrfs_extent_item *ei; | |
2095 | struct extent_buffer *leaf; | |
2096 | u32 item_size; | |
56bec294 | 2097 | int ret; |
5d4f98a2 | 2098 | int err = 0; |
3173a18f JB |
2099 | int metadata = (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2100 | node->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
5d4f98a2 | 2101 | |
79787eaa JM |
2102 | if (trans->aborted) |
2103 | return 0; | |
2104 | ||
3173a18f JB |
2105 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
2106 | metadata = 0; | |
2107 | ||
5d4f98a2 YZ |
2108 | path = btrfs_alloc_path(); |
2109 | if (!path) | |
2110 | return -ENOMEM; | |
2111 | ||
2112 | key.objectid = node->bytenr; | |
5d4f98a2 | 2113 | |
3173a18f JB |
2114 | if (metadata) { |
2115 | struct btrfs_delayed_tree_ref *tree_ref; | |
2116 | ||
2117 | tree_ref = btrfs_delayed_node_to_tree_ref(node); | |
2118 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2119 | key.offset = tree_ref->level; | |
2120 | } else { | |
2121 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2122 | key.offset = node->num_bytes; | |
2123 | } | |
2124 | ||
2125 | again: | |
5d4f98a2 YZ |
2126 | path->reada = 1; |
2127 | path->leave_spinning = 1; | |
2128 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2129 | path, 0, 1); | |
2130 | if (ret < 0) { | |
2131 | err = ret; | |
2132 | goto out; | |
2133 | } | |
2134 | if (ret > 0) { | |
3173a18f JB |
2135 | if (metadata) { |
2136 | btrfs_release_path(path); | |
2137 | metadata = 0; | |
2138 | ||
2139 | key.offset = node->num_bytes; | |
2140 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2141 | goto again; | |
2142 | } | |
5d4f98a2 YZ |
2143 | err = -EIO; |
2144 | goto out; | |
2145 | } | |
2146 | ||
2147 | leaf = path->nodes[0]; | |
2148 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2149 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2150 | if (item_size < sizeof(*ei)) { | |
2151 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2152 | path, (u64)-1, 0); | |
2153 | if (ret < 0) { | |
2154 | err = ret; | |
2155 | goto out; | |
2156 | } | |
2157 | leaf = path->nodes[0]; | |
2158 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2159 | } | |
2160 | #endif | |
2161 | BUG_ON(item_size < sizeof(*ei)); | |
2162 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2163 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2164 | |
5d4f98a2 YZ |
2165 | btrfs_mark_buffer_dirty(leaf); |
2166 | out: | |
2167 | btrfs_free_path(path); | |
2168 | return err; | |
56bec294 CM |
2169 | } |
2170 | ||
5d4f98a2 YZ |
2171 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2172 | struct btrfs_root *root, | |
2173 | struct btrfs_delayed_ref_node *node, | |
2174 | struct btrfs_delayed_extent_op *extent_op, | |
2175 | int insert_reserved) | |
56bec294 CM |
2176 | { |
2177 | int ret = 0; | |
5d4f98a2 YZ |
2178 | struct btrfs_delayed_tree_ref *ref; |
2179 | struct btrfs_key ins; | |
2180 | u64 parent = 0; | |
2181 | u64 ref_root = 0; | |
3173a18f JB |
2182 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
2183 | SKINNY_METADATA); | |
56bec294 | 2184 | |
5d4f98a2 YZ |
2185 | ref = btrfs_delayed_node_to_tree_ref(node); |
2186 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2187 | parent = ref->parent; | |
2188 | else | |
2189 | ref_root = ref->root; | |
2190 | ||
3173a18f JB |
2191 | ins.objectid = node->bytenr; |
2192 | if (skinny_metadata) { | |
2193 | ins.offset = ref->level; | |
2194 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2195 | } else { | |
2196 | ins.offset = node->num_bytes; | |
2197 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2198 | } | |
2199 | ||
5d4f98a2 YZ |
2200 | BUG_ON(node->ref_mod != 1); |
2201 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2202 | BUG_ON(!extent_op || !extent_op->update_flags); |
5d4f98a2 YZ |
2203 | ret = alloc_reserved_tree_block(trans, root, |
2204 | parent, ref_root, | |
2205 | extent_op->flags_to_set, | |
2206 | &extent_op->key, | |
2207 | ref->level, &ins); | |
5d4f98a2 YZ |
2208 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2209 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2210 | node->num_bytes, parent, ref_root, | |
2211 | ref->level, 0, 1, extent_op); | |
2212 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2213 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2214 | node->num_bytes, parent, ref_root, | |
2215 | ref->level, 0, 1, extent_op); | |
2216 | } else { | |
2217 | BUG(); | |
2218 | } | |
56bec294 CM |
2219 | return ret; |
2220 | } | |
2221 | ||
2222 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2223 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2224 | struct btrfs_root *root, | |
2225 | struct btrfs_delayed_ref_node *node, | |
2226 | struct btrfs_delayed_extent_op *extent_op, | |
2227 | int insert_reserved) | |
56bec294 | 2228 | { |
79787eaa JM |
2229 | int ret = 0; |
2230 | ||
2231 | if (trans->aborted) | |
2232 | return 0; | |
2233 | ||
5d4f98a2 | 2234 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2235 | struct btrfs_delayed_ref_head *head; |
2236 | /* | |
2237 | * we've hit the end of the chain and we were supposed | |
2238 | * to insert this extent into the tree. But, it got | |
2239 | * deleted before we ever needed to insert it, so all | |
2240 | * we have to do is clean up the accounting | |
2241 | */ | |
5d4f98a2 YZ |
2242 | BUG_ON(extent_op); |
2243 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2244 | if (insert_reserved) { |
f0486c68 YZ |
2245 | btrfs_pin_extent(root, node->bytenr, |
2246 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2247 | if (head->is_data) { |
2248 | ret = btrfs_del_csums(trans, root, | |
2249 | node->bytenr, | |
2250 | node->num_bytes); | |
5d4f98a2 | 2251 | } |
56bec294 | 2252 | } |
79787eaa | 2253 | return ret; |
56bec294 CM |
2254 | } |
2255 | ||
5d4f98a2 YZ |
2256 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2257 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2258 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2259 | insert_reserved); | |
2260 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2261 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2262 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2263 | insert_reserved); | |
2264 | else | |
2265 | BUG(); | |
2266 | return ret; | |
56bec294 CM |
2267 | } |
2268 | ||
2269 | static noinline struct btrfs_delayed_ref_node * | |
2270 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2271 | { | |
2272 | struct rb_node *node; | |
2273 | struct btrfs_delayed_ref_node *ref; | |
2274 | int action = BTRFS_ADD_DELAYED_REF; | |
2275 | again: | |
2276 | /* | |
2277 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2278 | * this prevents ref count from going down to zero when | |
2279 | * there still are pending delayed ref. | |
2280 | */ | |
2281 | node = rb_prev(&head->node.rb_node); | |
2282 | while (1) { | |
2283 | if (!node) | |
2284 | break; | |
2285 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2286 | rb_node); | |
2287 | if (ref->bytenr != head->node.bytenr) | |
2288 | break; | |
5d4f98a2 | 2289 | if (ref->action == action) |
56bec294 CM |
2290 | return ref; |
2291 | node = rb_prev(node); | |
2292 | } | |
2293 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2294 | action = BTRFS_DROP_DELAYED_REF; | |
2295 | goto again; | |
2296 | } | |
2297 | return NULL; | |
2298 | } | |
2299 | ||
79787eaa JM |
2300 | /* |
2301 | * Returns 0 on success or if called with an already aborted transaction. | |
2302 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2303 | */ | |
c3e69d58 CM |
2304 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2305 | struct btrfs_root *root, | |
2306 | struct list_head *cluster) | |
56bec294 | 2307 | { |
56bec294 CM |
2308 | struct btrfs_delayed_ref_root *delayed_refs; |
2309 | struct btrfs_delayed_ref_node *ref; | |
2310 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2311 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2312 | struct btrfs_fs_info *fs_info = root->fs_info; |
56bec294 | 2313 | int ret; |
c3e69d58 | 2314 | int count = 0; |
56bec294 | 2315 | int must_insert_reserved = 0; |
56bec294 CM |
2316 | |
2317 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2318 | while (1) { |
2319 | if (!locked_ref) { | |
c3e69d58 CM |
2320 | /* pick a new head ref from the cluster list */ |
2321 | if (list_empty(cluster)) | |
56bec294 | 2322 | break; |
56bec294 | 2323 | |
c3e69d58 CM |
2324 | locked_ref = list_entry(cluster->next, |
2325 | struct btrfs_delayed_ref_head, cluster); | |
2326 | ||
2327 | /* grab the lock that says we are going to process | |
2328 | * all the refs for this head */ | |
2329 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2330 | ||
2331 | /* | |
2332 | * we may have dropped the spin lock to get the head | |
2333 | * mutex lock, and that might have given someone else | |
2334 | * time to free the head. If that's true, it has been | |
2335 | * removed from our list and we can move on. | |
2336 | */ | |
2337 | if (ret == -EAGAIN) { | |
2338 | locked_ref = NULL; | |
2339 | count++; | |
2340 | continue; | |
56bec294 CM |
2341 | } |
2342 | } | |
a28ec197 | 2343 | |
ae1e206b JB |
2344 | /* |
2345 | * We need to try and merge add/drops of the same ref since we | |
2346 | * can run into issues with relocate dropping the implicit ref | |
2347 | * and then it being added back again before the drop can | |
2348 | * finish. If we merged anything we need to re-loop so we can | |
2349 | * get a good ref. | |
2350 | */ | |
2351 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, | |
2352 | locked_ref); | |
2353 | ||
d1270cd9 AJ |
2354 | /* |
2355 | * locked_ref is the head node, so we have to go one | |
2356 | * node back for any delayed ref updates | |
2357 | */ | |
2358 | ref = select_delayed_ref(locked_ref); | |
2359 | ||
2360 | if (ref && ref->seq && | |
097b8a7c | 2361 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d1270cd9 AJ |
2362 | /* |
2363 | * there are still refs with lower seq numbers in the | |
2364 | * process of being added. Don't run this ref yet. | |
2365 | */ | |
2366 | list_del_init(&locked_ref->cluster); | |
093486c4 | 2367 | btrfs_delayed_ref_unlock(locked_ref); |
d1270cd9 AJ |
2368 | locked_ref = NULL; |
2369 | delayed_refs->num_heads_ready++; | |
2370 | spin_unlock(&delayed_refs->lock); | |
2371 | cond_resched(); | |
2372 | spin_lock(&delayed_refs->lock); | |
2373 | continue; | |
2374 | } | |
2375 | ||
56bec294 CM |
2376 | /* |
2377 | * record the must insert reserved flag before we | |
2378 | * drop the spin lock. | |
2379 | */ | |
2380 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2381 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2382 | |
5d4f98a2 YZ |
2383 | extent_op = locked_ref->extent_op; |
2384 | locked_ref->extent_op = NULL; | |
2385 | ||
56bec294 CM |
2386 | if (!ref) { |
2387 | /* All delayed refs have been processed, Go ahead | |
2388 | * and send the head node to run_one_delayed_ref, | |
2389 | * so that any accounting fixes can happen | |
2390 | */ | |
2391 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2392 | |
2393 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2394 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2395 | extent_op = NULL; |
2396 | } | |
2397 | ||
2398 | if (extent_op) { | |
2399 | spin_unlock(&delayed_refs->lock); | |
2400 | ||
2401 | ret = run_delayed_extent_op(trans, root, | |
2402 | ref, extent_op); | |
78a6184a | 2403 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2404 | |
79787eaa | 2405 | if (ret) { |
093486c4 MX |
2406 | printk(KERN_DEBUG |
2407 | "btrfs: run_delayed_extent_op " | |
2408 | "returned %d\n", ret); | |
253beebd | 2409 | spin_lock(&delayed_refs->lock); |
093486c4 | 2410 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2411 | return ret; |
2412 | } | |
2413 | ||
203bf287 | 2414 | goto next; |
5d4f98a2 | 2415 | } |
56bec294 | 2416 | } |
02217ed2 | 2417 | |
56bec294 CM |
2418 | ref->in_tree = 0; |
2419 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2420 | delayed_refs->num_entries--; | |
093486c4 | 2421 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2422 | /* |
2423 | * when we play the delayed ref, also correct the | |
2424 | * ref_mod on head | |
2425 | */ | |
2426 | switch (ref->action) { | |
2427 | case BTRFS_ADD_DELAYED_REF: | |
2428 | case BTRFS_ADD_DELAYED_EXTENT: | |
2429 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2430 | break; | |
2431 | case BTRFS_DROP_DELAYED_REF: | |
2432 | locked_ref->node.ref_mod += ref->ref_mod; | |
2433 | break; | |
2434 | default: | |
2435 | WARN_ON(1); | |
2436 | } | |
2437 | } | |
56bec294 | 2438 | spin_unlock(&delayed_refs->lock); |
925baedd | 2439 | |
5d4f98a2 | 2440 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2441 | must_insert_reserved); |
eb099670 | 2442 | |
78a6184a | 2443 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2444 | if (ret) { |
093486c4 MX |
2445 | btrfs_delayed_ref_unlock(locked_ref); |
2446 | btrfs_put_delayed_ref(ref); | |
2447 | printk(KERN_DEBUG | |
2448 | "btrfs: run_one_delayed_ref returned %d\n", ret); | |
253beebd | 2449 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2450 | return ret; |
2451 | } | |
2452 | ||
093486c4 MX |
2453 | /* |
2454 | * If this node is a head, that means all the refs in this head | |
2455 | * have been dealt with, and we will pick the next head to deal | |
2456 | * with, so we must unlock the head and drop it from the cluster | |
2457 | * list before we release it. | |
2458 | */ | |
2459 | if (btrfs_delayed_ref_is_head(ref)) { | |
2460 | list_del_init(&locked_ref->cluster); | |
2461 | btrfs_delayed_ref_unlock(locked_ref); | |
2462 | locked_ref = NULL; | |
2463 | } | |
2464 | btrfs_put_delayed_ref(ref); | |
2465 | count++; | |
203bf287 | 2466 | next: |
c3e69d58 CM |
2467 | cond_resched(); |
2468 | spin_lock(&delayed_refs->lock); | |
2469 | } | |
2470 | return count; | |
2471 | } | |
2472 | ||
709c0486 AJ |
2473 | #ifdef SCRAMBLE_DELAYED_REFS |
2474 | /* | |
2475 | * Normally delayed refs get processed in ascending bytenr order. This | |
2476 | * correlates in most cases to the order added. To expose dependencies on this | |
2477 | * order, we start to process the tree in the middle instead of the beginning | |
2478 | */ | |
2479 | static u64 find_middle(struct rb_root *root) | |
2480 | { | |
2481 | struct rb_node *n = root->rb_node; | |
2482 | struct btrfs_delayed_ref_node *entry; | |
2483 | int alt = 1; | |
2484 | u64 middle; | |
2485 | u64 first = 0, last = 0; | |
2486 | ||
2487 | n = rb_first(root); | |
2488 | if (n) { | |
2489 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2490 | first = entry->bytenr; | |
2491 | } | |
2492 | n = rb_last(root); | |
2493 | if (n) { | |
2494 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2495 | last = entry->bytenr; | |
2496 | } | |
2497 | n = root->rb_node; | |
2498 | ||
2499 | while (n) { | |
2500 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2501 | WARN_ON(!entry->in_tree); | |
2502 | ||
2503 | middle = entry->bytenr; | |
2504 | ||
2505 | if (alt) | |
2506 | n = n->rb_left; | |
2507 | else | |
2508 | n = n->rb_right; | |
2509 | ||
2510 | alt = 1 - alt; | |
2511 | } | |
2512 | return middle; | |
2513 | } | |
2514 | #endif | |
2515 | ||
bed92eae AJ |
2516 | int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans, |
2517 | struct btrfs_fs_info *fs_info) | |
2518 | { | |
2519 | struct qgroup_update *qgroup_update; | |
2520 | int ret = 0; | |
2521 | ||
2522 | if (list_empty(&trans->qgroup_ref_list) != | |
2523 | !trans->delayed_ref_elem.seq) { | |
2524 | /* list without seq or seq without list */ | |
2525 | printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n", | |
2526 | list_empty(&trans->qgroup_ref_list) ? "" : " not", | |
2527 | trans->delayed_ref_elem.seq); | |
2528 | BUG(); | |
2529 | } | |
2530 | ||
2531 | if (!trans->delayed_ref_elem.seq) | |
2532 | return 0; | |
2533 | ||
2534 | while (!list_empty(&trans->qgroup_ref_list)) { | |
2535 | qgroup_update = list_first_entry(&trans->qgroup_ref_list, | |
2536 | struct qgroup_update, list); | |
2537 | list_del(&qgroup_update->list); | |
2538 | if (!ret) | |
2539 | ret = btrfs_qgroup_account_ref( | |
2540 | trans, fs_info, qgroup_update->node, | |
2541 | qgroup_update->extent_op); | |
2542 | kfree(qgroup_update); | |
2543 | } | |
2544 | ||
2545 | btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
2546 | ||
2547 | return ret; | |
2548 | } | |
2549 | ||
bb721703 CM |
2550 | static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq, |
2551 | int count) | |
2552 | { | |
2553 | int val = atomic_read(&delayed_refs->ref_seq); | |
2554 | ||
2555 | if (val < seq || val >= seq + count) | |
2556 | return 1; | |
2557 | return 0; | |
2558 | } | |
2559 | ||
c3e69d58 CM |
2560 | /* |
2561 | * this starts processing the delayed reference count updates and | |
2562 | * extent insertions we have queued up so far. count can be | |
2563 | * 0, which means to process everything in the tree at the start | |
2564 | * of the run (but not newly added entries), or it can be some target | |
2565 | * number you'd like to process. | |
79787eaa JM |
2566 | * |
2567 | * Returns 0 on success or if called with an aborted transaction | |
2568 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2569 | */ |
2570 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2571 | struct btrfs_root *root, unsigned long count) | |
2572 | { | |
2573 | struct rb_node *node; | |
2574 | struct btrfs_delayed_ref_root *delayed_refs; | |
2575 | struct btrfs_delayed_ref_node *ref; | |
2576 | struct list_head cluster; | |
2577 | int ret; | |
a168650c | 2578 | u64 delayed_start; |
c3e69d58 CM |
2579 | int run_all = count == (unsigned long)-1; |
2580 | int run_most = 0; | |
1fa11e26 | 2581 | int loops; |
c3e69d58 | 2582 | |
79787eaa JM |
2583 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2584 | if (trans->aborted) | |
2585 | return 0; | |
2586 | ||
c3e69d58 CM |
2587 | if (root == root->fs_info->extent_root) |
2588 | root = root->fs_info->tree_root; | |
2589 | ||
edf39272 JS |
2590 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); |
2591 | ||
c3e69d58 CM |
2592 | delayed_refs = &trans->transaction->delayed_refs; |
2593 | INIT_LIST_HEAD(&cluster); | |
bb721703 CM |
2594 | if (count == 0) { |
2595 | count = delayed_refs->num_entries * 2; | |
2596 | run_most = 1; | |
2597 | } | |
2598 | ||
2599 | if (!run_all && !run_most) { | |
2600 | int old; | |
2601 | int seq = atomic_read(&delayed_refs->ref_seq); | |
2602 | ||
2603 | progress: | |
2604 | old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1); | |
2605 | if (old) { | |
2606 | DEFINE_WAIT(__wait); | |
2607 | if (delayed_refs->num_entries < 16348) | |
2608 | return 0; | |
2609 | ||
2610 | prepare_to_wait(&delayed_refs->wait, &__wait, | |
2611 | TASK_UNINTERRUPTIBLE); | |
2612 | ||
2613 | old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1); | |
2614 | if (old) { | |
2615 | schedule(); | |
2616 | finish_wait(&delayed_refs->wait, &__wait); | |
2617 | ||
2618 | if (!refs_newer(delayed_refs, seq, 256)) | |
2619 | goto progress; | |
2620 | else | |
2621 | return 0; | |
2622 | } else { | |
2623 | finish_wait(&delayed_refs->wait, &__wait); | |
2624 | goto again; | |
2625 | } | |
2626 | } | |
2627 | ||
2628 | } else { | |
2629 | atomic_inc(&delayed_refs->procs_running_refs); | |
2630 | } | |
2631 | ||
c3e69d58 | 2632 | again: |
1fa11e26 | 2633 | loops = 0; |
c3e69d58 | 2634 | spin_lock(&delayed_refs->lock); |
097b8a7c | 2635 | |
709c0486 AJ |
2636 | #ifdef SCRAMBLE_DELAYED_REFS |
2637 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2638 | #endif | |
2639 | ||
c3e69d58 CM |
2640 | while (1) { |
2641 | if (!(run_all || run_most) && | |
2642 | delayed_refs->num_heads_ready < 64) | |
2643 | break; | |
eb099670 | 2644 | |
56bec294 | 2645 | /* |
c3e69d58 CM |
2646 | * go find something we can process in the rbtree. We start at |
2647 | * the beginning of the tree, and then build a cluster | |
2648 | * of refs to process starting at the first one we are able to | |
2649 | * lock | |
56bec294 | 2650 | */ |
a168650c | 2651 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2652 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2653 | delayed_refs->run_delayed_start); | |
2654 | if (ret) | |
56bec294 CM |
2655 | break; |
2656 | ||
c3e69d58 | 2657 | ret = run_clustered_refs(trans, root, &cluster); |
79787eaa | 2658 | if (ret < 0) { |
093486c4 | 2659 | btrfs_release_ref_cluster(&cluster); |
79787eaa JM |
2660 | spin_unlock(&delayed_refs->lock); |
2661 | btrfs_abort_transaction(trans, root, ret); | |
bb721703 | 2662 | atomic_dec(&delayed_refs->procs_running_refs); |
79787eaa JM |
2663 | return ret; |
2664 | } | |
c3e69d58 | 2665 | |
bb721703 CM |
2666 | atomic_add(ret, &delayed_refs->ref_seq); |
2667 | ||
c3e69d58 CM |
2668 | count -= min_t(unsigned long, ret, count); |
2669 | ||
2670 | if (count == 0) | |
2671 | break; | |
a168650c | 2672 | |
1fa11e26 AJ |
2673 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2674 | if (loops == 0) { | |
2675 | /* | |
2676 | * btrfs_find_ref_cluster looped. let's do one | |
2677 | * more cycle. if we don't run any delayed ref | |
2678 | * during that cycle (because we can't because | |
2679 | * all of them are blocked), bail out. | |
2680 | */ | |
2681 | loops = 1; | |
2682 | } else { | |
2683 | /* | |
2684 | * no runnable refs left, stop trying | |
2685 | */ | |
2686 | BUG_ON(run_all); | |
2687 | break; | |
2688 | } | |
2689 | } | |
2690 | if (ret) { | |
a168650c | 2691 | /* refs were run, let's reset staleness detection */ |
1fa11e26 | 2692 | loops = 0; |
a168650c | 2693 | } |
eb099670 | 2694 | } |
c3e69d58 | 2695 | |
56bec294 | 2696 | if (run_all) { |
ea658bad JB |
2697 | if (!list_empty(&trans->new_bgs)) { |
2698 | spin_unlock(&delayed_refs->lock); | |
2699 | btrfs_create_pending_block_groups(trans, root); | |
2700 | spin_lock(&delayed_refs->lock); | |
2701 | } | |
2702 | ||
56bec294 | 2703 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2704 | if (!node) |
56bec294 | 2705 | goto out; |
c3e69d58 | 2706 | count = (unsigned long)-1; |
e9d0b13b | 2707 | |
56bec294 CM |
2708 | while (node) { |
2709 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2710 | rb_node); | |
2711 | if (btrfs_delayed_ref_is_head(ref)) { | |
2712 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2713 | |
56bec294 CM |
2714 | head = btrfs_delayed_node_to_head(ref); |
2715 | atomic_inc(&ref->refs); | |
2716 | ||
2717 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2718 | /* |
2719 | * Mutex was contended, block until it's | |
2720 | * released and try again | |
2721 | */ | |
56bec294 CM |
2722 | mutex_lock(&head->mutex); |
2723 | mutex_unlock(&head->mutex); | |
2724 | ||
2725 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2726 | cond_resched(); |
56bec294 CM |
2727 | goto again; |
2728 | } | |
2729 | node = rb_next(node); | |
2730 | } | |
2731 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2732 | schedule_timeout(1); |
2733 | goto again; | |
5f39d397 | 2734 | } |
54aa1f4d | 2735 | out: |
bb721703 CM |
2736 | atomic_dec(&delayed_refs->procs_running_refs); |
2737 | smp_mb(); | |
2738 | if (waitqueue_active(&delayed_refs->wait)) | |
2739 | wake_up(&delayed_refs->wait); | |
2740 | ||
c3e69d58 | 2741 | spin_unlock(&delayed_refs->lock); |
edf39272 | 2742 | assert_qgroups_uptodate(trans); |
a28ec197 CM |
2743 | return 0; |
2744 | } | |
2745 | ||
5d4f98a2 YZ |
2746 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2747 | struct btrfs_root *root, | |
2748 | u64 bytenr, u64 num_bytes, u64 flags, | |
2749 | int is_data) | |
2750 | { | |
2751 | struct btrfs_delayed_extent_op *extent_op; | |
2752 | int ret; | |
2753 | ||
78a6184a | 2754 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2755 | if (!extent_op) |
2756 | return -ENOMEM; | |
2757 | ||
2758 | extent_op->flags_to_set = flags; | |
2759 | extent_op->update_flags = 1; | |
2760 | extent_op->update_key = 0; | |
2761 | extent_op->is_data = is_data ? 1 : 0; | |
2762 | ||
66d7e7f0 AJ |
2763 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2764 | num_bytes, extent_op); | |
5d4f98a2 | 2765 | if (ret) |
78a6184a | 2766 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2767 | return ret; |
2768 | } | |
2769 | ||
2770 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2771 | struct btrfs_root *root, | |
2772 | struct btrfs_path *path, | |
2773 | u64 objectid, u64 offset, u64 bytenr) | |
2774 | { | |
2775 | struct btrfs_delayed_ref_head *head; | |
2776 | struct btrfs_delayed_ref_node *ref; | |
2777 | struct btrfs_delayed_data_ref *data_ref; | |
2778 | struct btrfs_delayed_ref_root *delayed_refs; | |
2779 | struct rb_node *node; | |
2780 | int ret = 0; | |
2781 | ||
2782 | ret = -ENOENT; | |
2783 | delayed_refs = &trans->transaction->delayed_refs; | |
2784 | spin_lock(&delayed_refs->lock); | |
2785 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2786 | if (!head) | |
2787 | goto out; | |
2788 | ||
2789 | if (!mutex_trylock(&head->mutex)) { | |
2790 | atomic_inc(&head->node.refs); | |
2791 | spin_unlock(&delayed_refs->lock); | |
2792 | ||
b3b4aa74 | 2793 | btrfs_release_path(path); |
5d4f98a2 | 2794 | |
8cc33e5c DS |
2795 | /* |
2796 | * Mutex was contended, block until it's released and let | |
2797 | * caller try again | |
2798 | */ | |
5d4f98a2 YZ |
2799 | mutex_lock(&head->mutex); |
2800 | mutex_unlock(&head->mutex); | |
2801 | btrfs_put_delayed_ref(&head->node); | |
2802 | return -EAGAIN; | |
2803 | } | |
2804 | ||
2805 | node = rb_prev(&head->node.rb_node); | |
2806 | if (!node) | |
2807 | goto out_unlock; | |
2808 | ||
2809 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2810 | ||
2811 | if (ref->bytenr != bytenr) | |
2812 | goto out_unlock; | |
2813 | ||
2814 | ret = 1; | |
2815 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2816 | goto out_unlock; | |
2817 | ||
2818 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2819 | ||
2820 | node = rb_prev(node); | |
2821 | if (node) { | |
df57dbe6 LB |
2822 | int seq = ref->seq; |
2823 | ||
5d4f98a2 | 2824 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); |
df57dbe6 | 2825 | if (ref->bytenr == bytenr && ref->seq == seq) |
5d4f98a2 YZ |
2826 | goto out_unlock; |
2827 | } | |
2828 | ||
2829 | if (data_ref->root != root->root_key.objectid || | |
2830 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2831 | goto out_unlock; | |
2832 | ||
2833 | ret = 0; | |
2834 | out_unlock: | |
2835 | mutex_unlock(&head->mutex); | |
2836 | out: | |
2837 | spin_unlock(&delayed_refs->lock); | |
2838 | return ret; | |
2839 | } | |
2840 | ||
2841 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2842 | struct btrfs_root *root, | |
2843 | struct btrfs_path *path, | |
2844 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2845 | { |
2846 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2847 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2848 | struct btrfs_extent_data_ref *ref; |
2849 | struct btrfs_extent_inline_ref *iref; | |
2850 | struct btrfs_extent_item *ei; | |
f321e491 | 2851 | struct btrfs_key key; |
5d4f98a2 | 2852 | u32 item_size; |
be20aa9d | 2853 | int ret; |
925baedd | 2854 | |
be20aa9d | 2855 | key.objectid = bytenr; |
31840ae1 | 2856 | key.offset = (u64)-1; |
f321e491 | 2857 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2858 | |
be20aa9d CM |
2859 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2860 | if (ret < 0) | |
2861 | goto out; | |
79787eaa | 2862 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2863 | |
2864 | ret = -ENOENT; | |
2865 | if (path->slots[0] == 0) | |
31840ae1 | 2866 | goto out; |
be20aa9d | 2867 | |
31840ae1 | 2868 | path->slots[0]--; |
f321e491 | 2869 | leaf = path->nodes[0]; |
5d4f98a2 | 2870 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2871 | |
5d4f98a2 | 2872 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2873 | goto out; |
f321e491 | 2874 | |
5d4f98a2 YZ |
2875 | ret = 1; |
2876 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2877 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2878 | if (item_size < sizeof(*ei)) { | |
2879 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2880 | goto out; | |
2881 | } | |
2882 | #endif | |
2883 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2884 | |
5d4f98a2 YZ |
2885 | if (item_size != sizeof(*ei) + |
2886 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2887 | goto out; | |
be20aa9d | 2888 | |
5d4f98a2 YZ |
2889 | if (btrfs_extent_generation(leaf, ei) <= |
2890 | btrfs_root_last_snapshot(&root->root_item)) | |
2891 | goto out; | |
2892 | ||
2893 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2894 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2895 | BTRFS_EXTENT_DATA_REF_KEY) | |
2896 | goto out; | |
2897 | ||
2898 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2899 | if (btrfs_extent_refs(leaf, ei) != | |
2900 | btrfs_extent_data_ref_count(leaf, ref) || | |
2901 | btrfs_extent_data_ref_root(leaf, ref) != | |
2902 | root->root_key.objectid || | |
2903 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2904 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2905 | goto out; | |
2906 | ||
2907 | ret = 0; | |
2908 | out: | |
2909 | return ret; | |
2910 | } | |
2911 | ||
2912 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2913 | struct btrfs_root *root, | |
2914 | u64 objectid, u64 offset, u64 bytenr) | |
2915 | { | |
2916 | struct btrfs_path *path; | |
2917 | int ret; | |
2918 | int ret2; | |
2919 | ||
2920 | path = btrfs_alloc_path(); | |
2921 | if (!path) | |
2922 | return -ENOENT; | |
2923 | ||
2924 | do { | |
2925 | ret = check_committed_ref(trans, root, path, objectid, | |
2926 | offset, bytenr); | |
2927 | if (ret && ret != -ENOENT) | |
f321e491 | 2928 | goto out; |
80ff3856 | 2929 | |
5d4f98a2 YZ |
2930 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2931 | offset, bytenr); | |
2932 | } while (ret2 == -EAGAIN); | |
2933 | ||
2934 | if (ret2 && ret2 != -ENOENT) { | |
2935 | ret = ret2; | |
2936 | goto out; | |
f321e491 | 2937 | } |
5d4f98a2 YZ |
2938 | |
2939 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2940 | ret = 0; | |
be20aa9d | 2941 | out: |
80ff3856 | 2942 | btrfs_free_path(path); |
f0486c68 YZ |
2943 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2944 | WARN_ON(ret > 0); | |
f321e491 | 2945 | return ret; |
be20aa9d | 2946 | } |
c5739bba | 2947 | |
5d4f98a2 | 2948 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2949 | struct btrfs_root *root, |
5d4f98a2 | 2950 | struct extent_buffer *buf, |
66d7e7f0 | 2951 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
2952 | { |
2953 | u64 bytenr; | |
5d4f98a2 YZ |
2954 | u64 num_bytes; |
2955 | u64 parent; | |
31840ae1 | 2956 | u64 ref_root; |
31840ae1 | 2957 | u32 nritems; |
31840ae1 ZY |
2958 | struct btrfs_key key; |
2959 | struct btrfs_file_extent_item *fi; | |
2960 | int i; | |
2961 | int level; | |
2962 | int ret = 0; | |
31840ae1 | 2963 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 2964 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
2965 | |
2966 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2967 | nritems = btrfs_header_nritems(buf); |
2968 | level = btrfs_header_level(buf); | |
2969 | ||
5d4f98a2 YZ |
2970 | if (!root->ref_cows && level == 0) |
2971 | return 0; | |
31840ae1 | 2972 | |
5d4f98a2 YZ |
2973 | if (inc) |
2974 | process_func = btrfs_inc_extent_ref; | |
2975 | else | |
2976 | process_func = btrfs_free_extent; | |
31840ae1 | 2977 | |
5d4f98a2 YZ |
2978 | if (full_backref) |
2979 | parent = buf->start; | |
2980 | else | |
2981 | parent = 0; | |
2982 | ||
2983 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2984 | if (level == 0) { |
5d4f98a2 | 2985 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2986 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2987 | continue; | |
5d4f98a2 | 2988 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2989 | struct btrfs_file_extent_item); |
2990 | if (btrfs_file_extent_type(buf, fi) == | |
2991 | BTRFS_FILE_EXTENT_INLINE) | |
2992 | continue; | |
2993 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2994 | if (bytenr == 0) | |
2995 | continue; | |
5d4f98a2 YZ |
2996 | |
2997 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2998 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2999 | ret = process_func(trans, root, bytenr, num_bytes, | |
3000 | parent, ref_root, key.objectid, | |
66d7e7f0 | 3001 | key.offset, for_cow); |
31840ae1 ZY |
3002 | if (ret) |
3003 | goto fail; | |
3004 | } else { | |
5d4f98a2 YZ |
3005 | bytenr = btrfs_node_blockptr(buf, i); |
3006 | num_bytes = btrfs_level_size(root, level - 1); | |
3007 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
3008 | parent, ref_root, level - 1, 0, |
3009 | for_cow); | |
31840ae1 ZY |
3010 | if (ret) |
3011 | goto fail; | |
3012 | } | |
3013 | } | |
3014 | return 0; | |
3015 | fail: | |
5d4f98a2 YZ |
3016 | return ret; |
3017 | } | |
3018 | ||
3019 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 3020 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 3021 | { |
66d7e7f0 | 3022 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
3023 | } |
3024 | ||
3025 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 3026 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 3027 | { |
66d7e7f0 | 3028 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
3029 | } |
3030 | ||
9078a3e1 CM |
3031 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
3032 | struct btrfs_root *root, | |
3033 | struct btrfs_path *path, | |
3034 | struct btrfs_block_group_cache *cache) | |
3035 | { | |
3036 | int ret; | |
9078a3e1 | 3037 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
3038 | unsigned long bi; |
3039 | struct extent_buffer *leaf; | |
9078a3e1 | 3040 | |
9078a3e1 | 3041 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
3042 | if (ret < 0) |
3043 | goto fail; | |
79787eaa | 3044 | BUG_ON(ret); /* Corruption */ |
5f39d397 CM |
3045 | |
3046 | leaf = path->nodes[0]; | |
3047 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3048 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3049 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 3050 | btrfs_release_path(path); |
54aa1f4d | 3051 | fail: |
79787eaa JM |
3052 | if (ret) { |
3053 | btrfs_abort_transaction(trans, root, ret); | |
9078a3e1 | 3054 | return ret; |
79787eaa | 3055 | } |
9078a3e1 CM |
3056 | return 0; |
3057 | ||
3058 | } | |
3059 | ||
4a8c9a62 YZ |
3060 | static struct btrfs_block_group_cache * |
3061 | next_block_group(struct btrfs_root *root, | |
3062 | struct btrfs_block_group_cache *cache) | |
3063 | { | |
3064 | struct rb_node *node; | |
3065 | spin_lock(&root->fs_info->block_group_cache_lock); | |
3066 | node = rb_next(&cache->cache_node); | |
3067 | btrfs_put_block_group(cache); | |
3068 | if (node) { | |
3069 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3070 | cache_node); | |
11dfe35a | 3071 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3072 | } else |
3073 | cache = NULL; | |
3074 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3075 | return cache; | |
3076 | } | |
3077 | ||
0af3d00b JB |
3078 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3079 | struct btrfs_trans_handle *trans, | |
3080 | struct btrfs_path *path) | |
3081 | { | |
3082 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
3083 | struct inode *inode = NULL; | |
3084 | u64 alloc_hint = 0; | |
2b20982e | 3085 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
3086 | int num_pages = 0; |
3087 | int retries = 0; | |
3088 | int ret = 0; | |
3089 | ||
3090 | /* | |
3091 | * If this block group is smaller than 100 megs don't bother caching the | |
3092 | * block group. | |
3093 | */ | |
3094 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
3095 | spin_lock(&block_group->lock); | |
3096 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3097 | spin_unlock(&block_group->lock); | |
3098 | return 0; | |
3099 | } | |
3100 | ||
3101 | again: | |
3102 | inode = lookup_free_space_inode(root, block_group, path); | |
3103 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
3104 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3105 | btrfs_release_path(path); |
0af3d00b JB |
3106 | goto out; |
3107 | } | |
3108 | ||
3109 | if (IS_ERR(inode)) { | |
3110 | BUG_ON(retries); | |
3111 | retries++; | |
3112 | ||
3113 | if (block_group->ro) | |
3114 | goto out_free; | |
3115 | ||
3116 | ret = create_free_space_inode(root, trans, block_group, path); | |
3117 | if (ret) | |
3118 | goto out_free; | |
3119 | goto again; | |
3120 | } | |
3121 | ||
5b0e95bf JB |
3122 | /* We've already setup this transaction, go ahead and exit */ |
3123 | if (block_group->cache_generation == trans->transid && | |
3124 | i_size_read(inode)) { | |
3125 | dcs = BTRFS_DC_SETUP; | |
3126 | goto out_put; | |
3127 | } | |
3128 | ||
0af3d00b JB |
3129 | /* |
3130 | * We want to set the generation to 0, that way if anything goes wrong | |
3131 | * from here on out we know not to trust this cache when we load up next | |
3132 | * time. | |
3133 | */ | |
3134 | BTRFS_I(inode)->generation = 0; | |
3135 | ret = btrfs_update_inode(trans, root, inode); | |
3136 | WARN_ON(ret); | |
3137 | ||
3138 | if (i_size_read(inode) > 0) { | |
3139 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
3140 | inode); | |
3141 | if (ret) | |
3142 | goto out_put; | |
3143 | } | |
3144 | ||
3145 | spin_lock(&block_group->lock); | |
cf7c1ef6 LB |
3146 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
3147 | !btrfs_test_opt(root, SPACE_CACHE)) { | |
3148 | /* | |
3149 | * don't bother trying to write stuff out _if_ | |
3150 | * a) we're not cached, | |
3151 | * b) we're with nospace_cache mount option. | |
3152 | */ | |
2b20982e | 3153 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3154 | spin_unlock(&block_group->lock); |
3155 | goto out_put; | |
3156 | } | |
3157 | spin_unlock(&block_group->lock); | |
3158 | ||
6fc823b1 JB |
3159 | /* |
3160 | * Try to preallocate enough space based on how big the block group is. | |
3161 | * Keep in mind this has to include any pinned space which could end up | |
3162 | * taking up quite a bit since it's not folded into the other space | |
3163 | * cache. | |
3164 | */ | |
3165 | num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024); | |
0af3d00b JB |
3166 | if (!num_pages) |
3167 | num_pages = 1; | |
3168 | ||
0af3d00b JB |
3169 | num_pages *= 16; |
3170 | num_pages *= PAGE_CACHE_SIZE; | |
3171 | ||
3172 | ret = btrfs_check_data_free_space(inode, num_pages); | |
3173 | if (ret) | |
3174 | goto out_put; | |
3175 | ||
3176 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3177 | num_pages, num_pages, | |
3178 | &alloc_hint); | |
2b20982e JB |
3179 | if (!ret) |
3180 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 3181 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 3182 | |
0af3d00b JB |
3183 | out_put: |
3184 | iput(inode); | |
3185 | out_free: | |
b3b4aa74 | 3186 | btrfs_release_path(path); |
0af3d00b JB |
3187 | out: |
3188 | spin_lock(&block_group->lock); | |
e65cbb94 | 3189 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3190 | block_group->cache_generation = trans->transid; |
2b20982e | 3191 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3192 | spin_unlock(&block_group->lock); |
3193 | ||
3194 | return ret; | |
3195 | } | |
3196 | ||
96b5179d CM |
3197 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
3198 | struct btrfs_root *root) | |
9078a3e1 | 3199 | { |
4a8c9a62 | 3200 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 3201 | int err = 0; |
9078a3e1 | 3202 | struct btrfs_path *path; |
96b5179d | 3203 | u64 last = 0; |
9078a3e1 CM |
3204 | |
3205 | path = btrfs_alloc_path(); | |
3206 | if (!path) | |
3207 | return -ENOMEM; | |
3208 | ||
0af3d00b JB |
3209 | again: |
3210 | while (1) { | |
3211 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3212 | while (cache) { | |
3213 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3214 | break; | |
3215 | cache = next_block_group(root, cache); | |
3216 | } | |
3217 | if (!cache) { | |
3218 | if (last == 0) | |
3219 | break; | |
3220 | last = 0; | |
3221 | continue; | |
3222 | } | |
3223 | err = cache_save_setup(cache, trans, path); | |
3224 | last = cache->key.objectid + cache->key.offset; | |
3225 | btrfs_put_block_group(cache); | |
3226 | } | |
3227 | ||
d397712b | 3228 | while (1) { |
4a8c9a62 YZ |
3229 | if (last == 0) { |
3230 | err = btrfs_run_delayed_refs(trans, root, | |
3231 | (unsigned long)-1); | |
79787eaa JM |
3232 | if (err) /* File system offline */ |
3233 | goto out; | |
0f9dd46c | 3234 | } |
54aa1f4d | 3235 | |
4a8c9a62 YZ |
3236 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
3237 | while (cache) { | |
0af3d00b JB |
3238 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
3239 | btrfs_put_block_group(cache); | |
3240 | goto again; | |
3241 | } | |
3242 | ||
4a8c9a62 YZ |
3243 | if (cache->dirty) |
3244 | break; | |
3245 | cache = next_block_group(root, cache); | |
3246 | } | |
3247 | if (!cache) { | |
3248 | if (last == 0) | |
3249 | break; | |
3250 | last = 0; | |
3251 | continue; | |
3252 | } | |
0f9dd46c | 3253 | |
0cb59c99 JB |
3254 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
3255 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 3256 | cache->dirty = 0; |
4a8c9a62 | 3257 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 3258 | |
4a8c9a62 | 3259 | err = write_one_cache_group(trans, root, path, cache); |
79787eaa JM |
3260 | if (err) /* File system offline */ |
3261 | goto out; | |
3262 | ||
4a8c9a62 | 3263 | btrfs_put_block_group(cache); |
9078a3e1 | 3264 | } |
4a8c9a62 | 3265 | |
0cb59c99 JB |
3266 | while (1) { |
3267 | /* | |
3268 | * I don't think this is needed since we're just marking our | |
3269 | * preallocated extent as written, but just in case it can't | |
3270 | * hurt. | |
3271 | */ | |
3272 | if (last == 0) { | |
3273 | err = btrfs_run_delayed_refs(trans, root, | |
3274 | (unsigned long)-1); | |
79787eaa JM |
3275 | if (err) /* File system offline */ |
3276 | goto out; | |
0cb59c99 JB |
3277 | } |
3278 | ||
3279 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3280 | while (cache) { | |
3281 | /* | |
3282 | * Really this shouldn't happen, but it could if we | |
3283 | * couldn't write the entire preallocated extent and | |
3284 | * splitting the extent resulted in a new block. | |
3285 | */ | |
3286 | if (cache->dirty) { | |
3287 | btrfs_put_block_group(cache); | |
3288 | goto again; | |
3289 | } | |
3290 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3291 | break; | |
3292 | cache = next_block_group(root, cache); | |
3293 | } | |
3294 | if (!cache) { | |
3295 | if (last == 0) | |
3296 | break; | |
3297 | last = 0; | |
3298 | continue; | |
3299 | } | |
3300 | ||
79787eaa | 3301 | err = btrfs_write_out_cache(root, trans, cache, path); |
0cb59c99 JB |
3302 | |
3303 | /* | |
3304 | * If we didn't have an error then the cache state is still | |
3305 | * NEED_WRITE, so we can set it to WRITTEN. | |
3306 | */ | |
79787eaa | 3307 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
0cb59c99 JB |
3308 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
3309 | last = cache->key.objectid + cache->key.offset; | |
3310 | btrfs_put_block_group(cache); | |
3311 | } | |
79787eaa | 3312 | out: |
0cb59c99 | 3313 | |
9078a3e1 | 3314 | btrfs_free_path(path); |
79787eaa | 3315 | return err; |
9078a3e1 CM |
3316 | } |
3317 | ||
d2fb3437 YZ |
3318 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3319 | { | |
3320 | struct btrfs_block_group_cache *block_group; | |
3321 | int readonly = 0; | |
3322 | ||
3323 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3324 | if (!block_group || block_group->ro) | |
3325 | readonly = 1; | |
3326 | if (block_group) | |
fa9c0d79 | 3327 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3328 | return readonly; |
3329 | } | |
3330 | ||
593060d7 CM |
3331 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3332 | u64 total_bytes, u64 bytes_used, | |
3333 | struct btrfs_space_info **space_info) | |
3334 | { | |
3335 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3336 | int i; |
3337 | int factor; | |
3338 | ||
3339 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3340 | BTRFS_BLOCK_GROUP_RAID10)) | |
3341 | factor = 2; | |
3342 | else | |
3343 | factor = 1; | |
593060d7 CM |
3344 | |
3345 | found = __find_space_info(info, flags); | |
3346 | if (found) { | |
25179201 | 3347 | spin_lock(&found->lock); |
593060d7 | 3348 | found->total_bytes += total_bytes; |
89a55897 | 3349 | found->disk_total += total_bytes * factor; |
593060d7 | 3350 | found->bytes_used += bytes_used; |
b742bb82 | 3351 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3352 | found->full = 0; |
25179201 | 3353 | spin_unlock(&found->lock); |
593060d7 CM |
3354 | *space_info = found; |
3355 | return 0; | |
3356 | } | |
c146afad | 3357 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3358 | if (!found) |
3359 | return -ENOMEM; | |
3360 | ||
b742bb82 YZ |
3361 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3362 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3363 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3364 | spin_lock_init(&found->lock); |
52ba6929 | 3365 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3366 | found->total_bytes = total_bytes; |
89a55897 | 3367 | found->disk_total = total_bytes * factor; |
593060d7 | 3368 | found->bytes_used = bytes_used; |
b742bb82 | 3369 | found->disk_used = bytes_used * factor; |
593060d7 | 3370 | found->bytes_pinned = 0; |
e8569813 | 3371 | found->bytes_reserved = 0; |
c146afad | 3372 | found->bytes_readonly = 0; |
f0486c68 | 3373 | found->bytes_may_use = 0; |
593060d7 | 3374 | found->full = 0; |
0e4f8f88 | 3375 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3376 | found->chunk_alloc = 0; |
fdb5effd JB |
3377 | found->flush = 0; |
3378 | init_waitqueue_head(&found->wait); | |
593060d7 | 3379 | *space_info = found; |
4184ea7f | 3380 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3381 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3382 | info->data_sinfo = found; | |
593060d7 CM |
3383 | return 0; |
3384 | } | |
3385 | ||
8790d502 CM |
3386 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3387 | { | |
899c81ea ID |
3388 | u64 extra_flags = chunk_to_extended(flags) & |
3389 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3390 | |
de98ced9 | 3391 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3392 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3393 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3394 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3395 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3396 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3397 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3398 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3399 | } |
593060d7 | 3400 | |
fc67c450 ID |
3401 | /* |
3402 | * returns target flags in extended format or 0 if restripe for this | |
3403 | * chunk_type is not in progress | |
c6664b42 ID |
3404 | * |
3405 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3406 | */ |
3407 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3408 | { | |
3409 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3410 | u64 target = 0; | |
3411 | ||
fc67c450 ID |
3412 | if (!bctl) |
3413 | return 0; | |
3414 | ||
3415 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3416 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3417 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3418 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3419 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3420 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3421 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3422 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3423 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3424 | } | |
3425 | ||
3426 | return target; | |
3427 | } | |
3428 | ||
a46d11a8 ID |
3429 | /* |
3430 | * @flags: available profiles in extended format (see ctree.h) | |
3431 | * | |
e4d8ec0f ID |
3432 | * Returns reduced profile in chunk format. If profile changing is in |
3433 | * progress (either running or paused) picks the target profile (if it's | |
3434 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3435 | */ |
2b82032c | 3436 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3437 | { |
cd02dca5 CM |
3438 | /* |
3439 | * we add in the count of missing devices because we want | |
3440 | * to make sure that any RAID levels on a degraded FS | |
3441 | * continue to be honored. | |
3442 | */ | |
3443 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3444 | root->fs_info->fs_devices->missing_devices; | |
fc67c450 | 3445 | u64 target; |
53b381b3 | 3446 | u64 tmp; |
a061fc8d | 3447 | |
fc67c450 ID |
3448 | /* |
3449 | * see if restripe for this chunk_type is in progress, if so | |
3450 | * try to reduce to the target profile | |
3451 | */ | |
e4d8ec0f | 3452 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3453 | target = get_restripe_target(root->fs_info, flags); |
3454 | if (target) { | |
3455 | /* pick target profile only if it's already available */ | |
3456 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3457 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3458 | return extended_to_chunk(target); |
e4d8ec0f ID |
3459 | } |
3460 | } | |
3461 | spin_unlock(&root->fs_info->balance_lock); | |
3462 | ||
53b381b3 | 3463 | /* First, mask out the RAID levels which aren't possible */ |
a061fc8d | 3464 | if (num_devices == 1) |
53b381b3 DW |
3465 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 | |
3466 | BTRFS_BLOCK_GROUP_RAID5); | |
3467 | if (num_devices < 3) | |
3468 | flags &= ~BTRFS_BLOCK_GROUP_RAID6; | |
a061fc8d CM |
3469 | if (num_devices < 4) |
3470 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3471 | ||
53b381b3 DW |
3472 | tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | |
3473 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 | | |
3474 | BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10); | |
3475 | flags &= ~tmp; | |
ec44a35c | 3476 | |
53b381b3 DW |
3477 | if (tmp & BTRFS_BLOCK_GROUP_RAID6) |
3478 | tmp = BTRFS_BLOCK_GROUP_RAID6; | |
3479 | else if (tmp & BTRFS_BLOCK_GROUP_RAID5) | |
3480 | tmp = BTRFS_BLOCK_GROUP_RAID5; | |
3481 | else if (tmp & BTRFS_BLOCK_GROUP_RAID10) | |
3482 | tmp = BTRFS_BLOCK_GROUP_RAID10; | |
3483 | else if (tmp & BTRFS_BLOCK_GROUP_RAID1) | |
3484 | tmp = BTRFS_BLOCK_GROUP_RAID1; | |
3485 | else if (tmp & BTRFS_BLOCK_GROUP_RAID0) | |
3486 | tmp = BTRFS_BLOCK_GROUP_RAID0; | |
a46d11a8 | 3487 | |
53b381b3 | 3488 | return extended_to_chunk(flags | tmp); |
ec44a35c CM |
3489 | } |
3490 | ||
b742bb82 | 3491 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3492 | { |
de98ced9 MX |
3493 | unsigned seq; |
3494 | ||
3495 | do { | |
3496 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
3497 | ||
3498 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3499 | flags |= root->fs_info->avail_data_alloc_bits; | |
3500 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3501 | flags |= root->fs_info->avail_system_alloc_bits; | |
3502 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3503 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
3504 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
6fef8df1 | 3505 | |
b742bb82 | 3506 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3507 | } |
3508 | ||
6d07bcec | 3509 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3510 | { |
b742bb82 | 3511 | u64 flags; |
53b381b3 | 3512 | u64 ret; |
9ed74f2d | 3513 | |
b742bb82 YZ |
3514 | if (data) |
3515 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3516 | else if (root == root->fs_info->chunk_root) | |
3517 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3518 | else |
b742bb82 | 3519 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3520 | |
53b381b3 DW |
3521 | ret = get_alloc_profile(root, flags); |
3522 | return ret; | |
6a63209f | 3523 | } |
9ed74f2d | 3524 | |
6a63209f | 3525 | /* |
6a63209f JB |
3526 | * This will check the space that the inode allocates from to make sure we have |
3527 | * enough space for bytes. | |
6a63209f | 3528 | */ |
0ca1f7ce | 3529 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3530 | { |
6a63209f | 3531 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3532 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 3533 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 3534 | u64 used; |
0af3d00b | 3535 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3536 | |
6a63209f | 3537 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 3538 | bytes = ALIGN(bytes, root->sectorsize); |
6a63209f | 3539 | |
82d5902d LZ |
3540 | if (root == root->fs_info->tree_root || |
3541 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3542 | alloc_chunk = 0; |
3543 | committed = 1; | |
3544 | } | |
3545 | ||
b4d7c3c9 | 3546 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
3547 | if (!data_sinfo) |
3548 | goto alloc; | |
9ed74f2d | 3549 | |
6a63209f JB |
3550 | again: |
3551 | /* make sure we have enough space to handle the data first */ | |
3552 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3553 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3554 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3555 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3556 | |
3557 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3558 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3559 | |
6a63209f JB |
3560 | /* |
3561 | * if we don't have enough free bytes in this space then we need | |
3562 | * to alloc a new chunk. | |
3563 | */ | |
0af3d00b | 3564 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3565 | u64 alloc_target; |
9ed74f2d | 3566 | |
0e4f8f88 | 3567 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3568 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3569 | alloc: |
6a63209f | 3570 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3571 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3572 | if (IS_ERR(trans)) |
3573 | return PTR_ERR(trans); | |
9ed74f2d | 3574 | |
6a63209f | 3575 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
3576 | alloc_target, |
3577 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3578 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3579 | if (ret < 0) { |
3580 | if (ret != -ENOSPC) | |
3581 | return ret; | |
3582 | else | |
3583 | goto commit_trans; | |
3584 | } | |
9ed74f2d | 3585 | |
b4d7c3c9 LZ |
3586 | if (!data_sinfo) |
3587 | data_sinfo = fs_info->data_sinfo; | |
3588 | ||
6a63209f JB |
3589 | goto again; |
3590 | } | |
f2bb8f5c JB |
3591 | |
3592 | /* | |
3593 | * If we have less pinned bytes than we want to allocate then | |
3594 | * don't bother committing the transaction, it won't help us. | |
3595 | */ | |
3596 | if (data_sinfo->bytes_pinned < bytes) | |
3597 | committed = 1; | |
6a63209f | 3598 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3599 | |
4e06bdd6 | 3600 | /* commit the current transaction and try again */ |
d52a5b5f | 3601 | commit_trans: |
a4abeea4 JB |
3602 | if (!committed && |
3603 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3604 | committed = 1; |
7a7eaa40 | 3605 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3606 | if (IS_ERR(trans)) |
3607 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3608 | ret = btrfs_commit_transaction(trans, root); |
3609 | if (ret) | |
3610 | return ret; | |
3611 | goto again; | |
3612 | } | |
9ed74f2d | 3613 | |
6a63209f JB |
3614 | return -ENOSPC; |
3615 | } | |
3616 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3617 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3618 | data_sinfo->flags, bytes, 1); |
6a63209f | 3619 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3620 | |
9ed74f2d | 3621 | return 0; |
9ed74f2d | 3622 | } |
6a63209f | 3623 | |
6a63209f | 3624 | /* |
fb25e914 | 3625 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3626 | */ |
0ca1f7ce | 3627 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3628 | { |
0ca1f7ce | 3629 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3630 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3631 | |
6a63209f | 3632 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 3633 | bytes = ALIGN(bytes, root->sectorsize); |
e3ccfa98 | 3634 | |
b4d7c3c9 | 3635 | data_sinfo = root->fs_info->data_sinfo; |
6a63209f JB |
3636 | spin_lock(&data_sinfo->lock); |
3637 | data_sinfo->bytes_may_use -= bytes; | |
8c2a3ca2 | 3638 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3639 | data_sinfo->flags, bytes, 0); |
6a63209f | 3640 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3641 | } |
3642 | ||
97e728d4 | 3643 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3644 | { |
97e728d4 JB |
3645 | struct list_head *head = &info->space_info; |
3646 | struct btrfs_space_info *found; | |
e3ccfa98 | 3647 | |
97e728d4 JB |
3648 | rcu_read_lock(); |
3649 | list_for_each_entry_rcu(found, head, list) { | |
3650 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3651 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3652 | } |
97e728d4 | 3653 | rcu_read_unlock(); |
e3ccfa98 JB |
3654 | } |
3655 | ||
e5bc2458 | 3656 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 3657 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 3658 | { |
fb25e914 | 3659 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3660 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3661 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3662 | u64 thresh; |
e3ccfa98 | 3663 | |
0e4f8f88 CM |
3664 | if (force == CHUNK_ALLOC_FORCE) |
3665 | return 1; | |
3666 | ||
fb25e914 JB |
3667 | /* |
3668 | * We need to take into account the global rsv because for all intents | |
3669 | * and purposes it's used space. Don't worry about locking the | |
3670 | * global_rsv, it doesn't change except when the transaction commits. | |
3671 | */ | |
54338b5c JB |
3672 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3673 | num_allocated += global_rsv->size; | |
fb25e914 | 3674 | |
0e4f8f88 CM |
3675 | /* |
3676 | * in limited mode, we want to have some free space up to | |
3677 | * about 1% of the FS size. | |
3678 | */ | |
3679 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3680 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3681 | thresh = max_t(u64, 64 * 1024 * 1024, |
3682 | div_factor_fine(thresh, 1)); | |
3683 | ||
3684 | if (num_bytes - num_allocated < thresh) | |
3685 | return 1; | |
3686 | } | |
0e4f8f88 | 3687 | |
698d0082 | 3688 | if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8)) |
14ed0ca6 | 3689 | return 0; |
424499db | 3690 | return 1; |
32c00aff JB |
3691 | } |
3692 | ||
15d1ff81 LB |
3693 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
3694 | { | |
3695 | u64 num_dev; | |
3696 | ||
53b381b3 DW |
3697 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
3698 | BTRFS_BLOCK_GROUP_RAID0 | | |
3699 | BTRFS_BLOCK_GROUP_RAID5 | | |
3700 | BTRFS_BLOCK_GROUP_RAID6)) | |
15d1ff81 LB |
3701 | num_dev = root->fs_info->fs_devices->rw_devices; |
3702 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
3703 | num_dev = 2; | |
3704 | else | |
3705 | num_dev = 1; /* DUP or single */ | |
3706 | ||
3707 | /* metadata for updaing devices and chunk tree */ | |
3708 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); | |
3709 | } | |
3710 | ||
3711 | static void check_system_chunk(struct btrfs_trans_handle *trans, | |
3712 | struct btrfs_root *root, u64 type) | |
3713 | { | |
3714 | struct btrfs_space_info *info; | |
3715 | u64 left; | |
3716 | u64 thresh; | |
3717 | ||
3718 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3719 | spin_lock(&info->lock); | |
3720 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
3721 | info->bytes_reserved - info->bytes_readonly; | |
3722 | spin_unlock(&info->lock); | |
3723 | ||
3724 | thresh = get_system_chunk_thresh(root, type); | |
3725 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
3726 | printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n", | |
3727 | left, thresh, type); | |
3728 | dump_space_info(info, 0, 0); | |
3729 | } | |
3730 | ||
3731 | if (left < thresh) { | |
3732 | u64 flags; | |
3733 | ||
3734 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
3735 | btrfs_alloc_chunk(trans, root, flags); | |
3736 | } | |
3737 | } | |
3738 | ||
6324fbf3 | 3739 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 3740 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 3741 | { |
6324fbf3 | 3742 | struct btrfs_space_info *space_info; |
97e728d4 | 3743 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3744 | int wait_for_alloc = 0; |
9ed74f2d | 3745 | int ret = 0; |
9ed74f2d | 3746 | |
c6b305a8 JB |
3747 | /* Don't re-enter if we're already allocating a chunk */ |
3748 | if (trans->allocating_chunk) | |
3749 | return -ENOSPC; | |
3750 | ||
6324fbf3 | 3751 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3752 | if (!space_info) { |
3753 | ret = update_space_info(extent_root->fs_info, flags, | |
3754 | 0, 0, &space_info); | |
79787eaa | 3755 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 3756 | } |
79787eaa | 3757 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 3758 | |
6d74119f | 3759 | again: |
25179201 | 3760 | spin_lock(&space_info->lock); |
9e622d6b | 3761 | if (force < space_info->force_alloc) |
0e4f8f88 | 3762 | force = space_info->force_alloc; |
25179201 JB |
3763 | if (space_info->full) { |
3764 | spin_unlock(&space_info->lock); | |
6d74119f | 3765 | return 0; |
9ed74f2d JB |
3766 | } |
3767 | ||
698d0082 | 3768 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 3769 | spin_unlock(&space_info->lock); |
6d74119f JB |
3770 | return 0; |
3771 | } else if (space_info->chunk_alloc) { | |
3772 | wait_for_alloc = 1; | |
3773 | } else { | |
3774 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3775 | } |
0e4f8f88 | 3776 | |
25179201 | 3777 | spin_unlock(&space_info->lock); |
9ed74f2d | 3778 | |
6d74119f JB |
3779 | mutex_lock(&fs_info->chunk_mutex); |
3780 | ||
3781 | /* | |
3782 | * The chunk_mutex is held throughout the entirety of a chunk | |
3783 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3784 | * other guy is done and we need to recheck and see if we should | |
3785 | * allocate. | |
3786 | */ | |
3787 | if (wait_for_alloc) { | |
3788 | mutex_unlock(&fs_info->chunk_mutex); | |
3789 | wait_for_alloc = 0; | |
3790 | goto again; | |
3791 | } | |
3792 | ||
c6b305a8 JB |
3793 | trans->allocating_chunk = true; |
3794 | ||
67377734 JB |
3795 | /* |
3796 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3797 | * allocating mixed chunks instead of individual chunks. | |
3798 | */ | |
3799 | if (btrfs_mixed_space_info(space_info)) | |
3800 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3801 | ||
97e728d4 JB |
3802 | /* |
3803 | * if we're doing a data chunk, go ahead and make sure that | |
3804 | * we keep a reasonable number of metadata chunks allocated in the | |
3805 | * FS as well. | |
3806 | */ | |
9ed74f2d | 3807 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3808 | fs_info->data_chunk_allocations++; |
3809 | if (!(fs_info->data_chunk_allocations % | |
3810 | fs_info->metadata_ratio)) | |
3811 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3812 | } |
3813 | ||
15d1ff81 LB |
3814 | /* |
3815 | * Check if we have enough space in SYSTEM chunk because we may need | |
3816 | * to update devices. | |
3817 | */ | |
3818 | check_system_chunk(trans, extent_root, flags); | |
3819 | ||
2b82032c | 3820 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 3821 | trans->allocating_chunk = false; |
92b8e897 | 3822 | |
9ed74f2d | 3823 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
3824 | if (ret < 0 && ret != -ENOSPC) |
3825 | goto out; | |
9ed74f2d | 3826 | if (ret) |
6324fbf3 | 3827 | space_info->full = 1; |
424499db YZ |
3828 | else |
3829 | ret = 1; | |
6d74119f | 3830 | |
0e4f8f88 | 3831 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 3832 | out: |
6d74119f | 3833 | space_info->chunk_alloc = 0; |
9ed74f2d | 3834 | spin_unlock(&space_info->lock); |
a25c75d5 | 3835 | mutex_unlock(&fs_info->chunk_mutex); |
0f9dd46c | 3836 | return ret; |
6324fbf3 | 3837 | } |
9ed74f2d | 3838 | |
a80c8dcf JB |
3839 | static int can_overcommit(struct btrfs_root *root, |
3840 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 3841 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf | 3842 | { |
96f1bb57 | 3843 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
a80c8dcf | 3844 | u64 profile = btrfs_get_alloc_profile(root, 0); |
96f1bb57 | 3845 | u64 rsv_size = 0; |
a80c8dcf JB |
3846 | u64 avail; |
3847 | u64 used; | |
70afa399 | 3848 | u64 to_add; |
a80c8dcf JB |
3849 | |
3850 | used = space_info->bytes_used + space_info->bytes_reserved + | |
96f1bb57 JB |
3851 | space_info->bytes_pinned + space_info->bytes_readonly; |
3852 | ||
3853 | spin_lock(&global_rsv->lock); | |
3854 | rsv_size = global_rsv->size; | |
3855 | spin_unlock(&global_rsv->lock); | |
3856 | ||
3857 | /* | |
3858 | * We only want to allow over committing if we have lots of actual space | |
3859 | * free, but if we don't have enough space to handle the global reserve | |
3860 | * space then we could end up having a real enospc problem when trying | |
3861 | * to allocate a chunk or some other such important allocation. | |
3862 | */ | |
3863 | rsv_size <<= 1; | |
3864 | if (used + rsv_size >= space_info->total_bytes) | |
3865 | return 0; | |
3866 | ||
3867 | used += space_info->bytes_may_use; | |
a80c8dcf JB |
3868 | |
3869 | spin_lock(&root->fs_info->free_chunk_lock); | |
3870 | avail = root->fs_info->free_chunk_space; | |
3871 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3872 | ||
3873 | /* | |
3874 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
3875 | * space is actually useable. For raid56, the space info used |
3876 | * doesn't include the parity drive, so we don't have to | |
3877 | * change the math | |
a80c8dcf JB |
3878 | */ |
3879 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3880 | BTRFS_BLOCK_GROUP_RAID1 | | |
3881 | BTRFS_BLOCK_GROUP_RAID10)) | |
3882 | avail >>= 1; | |
3883 | ||
70afa399 JB |
3884 | to_add = space_info->total_bytes; |
3885 | ||
a80c8dcf | 3886 | /* |
561c294d MX |
3887 | * If we aren't flushing all things, let us overcommit up to |
3888 | * 1/2th of the space. If we can flush, don't let us overcommit | |
3889 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 3890 | */ |
08e007d2 | 3891 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
70afa399 | 3892 | to_add >>= 3; |
a80c8dcf | 3893 | else |
70afa399 | 3894 | to_add >>= 1; |
a80c8dcf | 3895 | |
70afa399 JB |
3896 | /* |
3897 | * Limit the overcommit to the amount of free space we could possibly | |
3898 | * allocate for chunks. | |
3899 | */ | |
3900 | to_add = min(avail, to_add); | |
a80c8dcf | 3901 | |
70afa399 | 3902 | if (used + bytes < space_info->total_bytes + to_add) |
a80c8dcf JB |
3903 | return 1; |
3904 | return 0; | |
3905 | } | |
3906 | ||
da633a42 MX |
3907 | void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
3908 | unsigned long nr_pages) | |
3909 | { | |
3910 | struct super_block *sb = root->fs_info->sb; | |
3911 | int started; | |
3912 | ||
3913 | /* If we can not start writeback, just sync all the delalloc file. */ | |
b695188d | 3914 | started = try_to_writeback_inodes_sb_nr(sb, nr_pages, |
da633a42 MX |
3915 | WB_REASON_FS_FREE_SPACE); |
3916 | if (!started) { | |
3917 | /* | |
3918 | * We needn't worry the filesystem going from r/w to r/o though | |
3919 | * we don't acquire ->s_umount mutex, because the filesystem | |
3920 | * should guarantee the delalloc inodes list be empty after | |
3921 | * the filesystem is readonly(all dirty pages are written to | |
3922 | * the disk). | |
3923 | */ | |
3924 | btrfs_start_delalloc_inodes(root, 0); | |
3925 | btrfs_wait_ordered_extents(root, 0); | |
3926 | } | |
3927 | } | |
3928 | ||
9ed74f2d | 3929 | /* |
5da9d01b | 3930 | * shrink metadata reservation for delalloc |
9ed74f2d | 3931 | */ |
f4c738c2 JB |
3932 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
3933 | bool wait_ordered) | |
5da9d01b | 3934 | { |
0ca1f7ce | 3935 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3936 | struct btrfs_space_info *space_info; |
663350ac | 3937 | struct btrfs_trans_handle *trans; |
f4c738c2 | 3938 | u64 delalloc_bytes; |
5da9d01b | 3939 | u64 max_reclaim; |
b1953bce | 3940 | long time_left; |
877da174 | 3941 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3942 | int loops = 0; |
08e007d2 | 3943 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 3944 | |
663350ac | 3945 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 3946 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3947 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3948 | |
3949 | smp_mb(); | |
963d678b MX |
3950 | delalloc_bytes = percpu_counter_sum_positive( |
3951 | &root->fs_info->delalloc_bytes); | |
f4c738c2 | 3952 | if (delalloc_bytes == 0) { |
fdb5effd | 3953 | if (trans) |
f4c738c2 | 3954 | return; |
6bbe3a9c | 3955 | btrfs_wait_ordered_extents(root, 0); |
f4c738c2 | 3956 | return; |
fdb5effd JB |
3957 | } |
3958 | ||
f4c738c2 JB |
3959 | while (delalloc_bytes && loops < 3) { |
3960 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
3961 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; | |
da633a42 | 3962 | btrfs_writeback_inodes_sb_nr(root, nr_pages); |
dea31f52 JB |
3963 | /* |
3964 | * We need to wait for the async pages to actually start before | |
3965 | * we do anything. | |
3966 | */ | |
3967 | wait_event(root->fs_info->async_submit_wait, | |
3968 | !atomic_read(&root->fs_info->async_delalloc_pages)); | |
3969 | ||
08e007d2 MX |
3970 | if (!trans) |
3971 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
3972 | else | |
3973 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 3974 | spin_lock(&space_info->lock); |
08e007d2 | 3975 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
3976 | spin_unlock(&space_info->lock); |
3977 | break; | |
3978 | } | |
0019f10d | 3979 | spin_unlock(&space_info->lock); |
5da9d01b | 3980 | |
36e39c40 | 3981 | loops++; |
f104d044 | 3982 | if (wait_ordered && !trans) { |
6bbe3a9c | 3983 | btrfs_wait_ordered_extents(root, 0); |
f104d044 | 3984 | } else { |
f4c738c2 | 3985 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
3986 | if (time_left) |
3987 | break; | |
3988 | } | |
f4c738c2 | 3989 | smp_mb(); |
963d678b MX |
3990 | delalloc_bytes = percpu_counter_sum_positive( |
3991 | &root->fs_info->delalloc_bytes); | |
5da9d01b | 3992 | } |
5da9d01b YZ |
3993 | } |
3994 | ||
663350ac JB |
3995 | /** |
3996 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
3997 | * @root - the root we're allocating for | |
3998 | * @bytes - the number of bytes we want to reserve | |
3999 | * @force - force the commit | |
8bb8ab2e | 4000 | * |
663350ac JB |
4001 | * This will check to make sure that committing the transaction will actually |
4002 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4003 | * will return -ENOSPC. | |
8bb8ab2e | 4004 | */ |
663350ac JB |
4005 | static int may_commit_transaction(struct btrfs_root *root, |
4006 | struct btrfs_space_info *space_info, | |
4007 | u64 bytes, int force) | |
4008 | { | |
4009 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
4010 | struct btrfs_trans_handle *trans; | |
4011 | ||
4012 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4013 | if (trans) | |
4014 | return -EAGAIN; | |
4015 | ||
4016 | if (force) | |
4017 | goto commit; | |
4018 | ||
4019 | /* See if there is enough pinned space to make this reservation */ | |
4020 | spin_lock(&space_info->lock); | |
4021 | if (space_info->bytes_pinned >= bytes) { | |
4022 | spin_unlock(&space_info->lock); | |
4023 | goto commit; | |
4024 | } | |
4025 | spin_unlock(&space_info->lock); | |
4026 | ||
4027 | /* | |
4028 | * See if there is some space in the delayed insertion reservation for | |
4029 | * this reservation. | |
4030 | */ | |
4031 | if (space_info != delayed_rsv->space_info) | |
4032 | return -ENOSPC; | |
4033 | ||
d9b0218f | 4034 | spin_lock(&space_info->lock); |
663350ac | 4035 | spin_lock(&delayed_rsv->lock); |
d9b0218f | 4036 | if (space_info->bytes_pinned + delayed_rsv->size < bytes) { |
663350ac | 4037 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 4038 | spin_unlock(&space_info->lock); |
663350ac JB |
4039 | return -ENOSPC; |
4040 | } | |
4041 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 4042 | spin_unlock(&space_info->lock); |
663350ac JB |
4043 | |
4044 | commit: | |
4045 | trans = btrfs_join_transaction(root); | |
4046 | if (IS_ERR(trans)) | |
4047 | return -ENOSPC; | |
4048 | ||
4049 | return btrfs_commit_transaction(trans, root); | |
4050 | } | |
4051 | ||
96c3f433 | 4052 | enum flush_state { |
67b0fd63 JB |
4053 | FLUSH_DELAYED_ITEMS_NR = 1, |
4054 | FLUSH_DELAYED_ITEMS = 2, | |
4055 | FLUSH_DELALLOC = 3, | |
4056 | FLUSH_DELALLOC_WAIT = 4, | |
ea658bad JB |
4057 | ALLOC_CHUNK = 5, |
4058 | COMMIT_TRANS = 6, | |
96c3f433 JB |
4059 | }; |
4060 | ||
4061 | static int flush_space(struct btrfs_root *root, | |
4062 | struct btrfs_space_info *space_info, u64 num_bytes, | |
4063 | u64 orig_bytes, int state) | |
4064 | { | |
4065 | struct btrfs_trans_handle *trans; | |
4066 | int nr; | |
f4c738c2 | 4067 | int ret = 0; |
96c3f433 JB |
4068 | |
4069 | switch (state) { | |
96c3f433 JB |
4070 | case FLUSH_DELAYED_ITEMS_NR: |
4071 | case FLUSH_DELAYED_ITEMS: | |
4072 | if (state == FLUSH_DELAYED_ITEMS_NR) { | |
4073 | u64 bytes = btrfs_calc_trans_metadata_size(root, 1); | |
4074 | ||
4075 | nr = (int)div64_u64(num_bytes, bytes); | |
4076 | if (!nr) | |
4077 | nr = 1; | |
4078 | nr *= 2; | |
4079 | } else { | |
4080 | nr = -1; | |
4081 | } | |
4082 | trans = btrfs_join_transaction(root); | |
4083 | if (IS_ERR(trans)) { | |
4084 | ret = PTR_ERR(trans); | |
4085 | break; | |
4086 | } | |
4087 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
4088 | btrfs_end_transaction(trans, root); | |
4089 | break; | |
67b0fd63 JB |
4090 | case FLUSH_DELALLOC: |
4091 | case FLUSH_DELALLOC_WAIT: | |
4092 | shrink_delalloc(root, num_bytes, orig_bytes, | |
4093 | state == FLUSH_DELALLOC_WAIT); | |
4094 | break; | |
ea658bad JB |
4095 | case ALLOC_CHUNK: |
4096 | trans = btrfs_join_transaction(root); | |
4097 | if (IS_ERR(trans)) { | |
4098 | ret = PTR_ERR(trans); | |
4099 | break; | |
4100 | } | |
4101 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
4102 | btrfs_get_alloc_profile(root, 0), |
4103 | CHUNK_ALLOC_NO_FORCE); | |
4104 | btrfs_end_transaction(trans, root); | |
4105 | if (ret == -ENOSPC) | |
4106 | ret = 0; | |
4107 | break; | |
96c3f433 JB |
4108 | case COMMIT_TRANS: |
4109 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
4110 | break; | |
4111 | default: | |
4112 | ret = -ENOSPC; | |
4113 | break; | |
4114 | } | |
4115 | ||
4116 | return ret; | |
4117 | } | |
4a92b1b8 JB |
4118 | /** |
4119 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
4120 | * @root - the root we're allocating for | |
4121 | * @block_rsv - the block_rsv we're allocating for | |
4122 | * @orig_bytes - the number of bytes we want | |
48fc7f7e | 4123 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 4124 | * |
4a92b1b8 JB |
4125 | * This will reserve orgi_bytes number of bytes from the space info associated |
4126 | * with the block_rsv. If there is not enough space it will make an attempt to | |
4127 | * flush out space to make room. It will do this by flushing delalloc if | |
4128 | * possible or committing the transaction. If flush is 0 then no attempts to | |
4129 | * regain reservations will be made and this will fail if there is not enough | |
4130 | * space already. | |
8bb8ab2e | 4131 | */ |
4a92b1b8 | 4132 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e | 4133 | struct btrfs_block_rsv *block_rsv, |
08e007d2 MX |
4134 | u64 orig_bytes, |
4135 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4136 | { |
f0486c68 | 4137 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 4138 | u64 used; |
8bb8ab2e | 4139 | u64 num_bytes = orig_bytes; |
67b0fd63 | 4140 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
8bb8ab2e | 4141 | int ret = 0; |
fdb5effd | 4142 | bool flushing = false; |
9ed74f2d | 4143 | |
8bb8ab2e | 4144 | again: |
fdb5effd | 4145 | ret = 0; |
8bb8ab2e | 4146 | spin_lock(&space_info->lock); |
fdb5effd | 4147 | /* |
08e007d2 MX |
4148 | * We only want to wait if somebody other than us is flushing and we |
4149 | * are actually allowed to flush all things. | |
fdb5effd | 4150 | */ |
08e007d2 MX |
4151 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && |
4152 | space_info->flush) { | |
fdb5effd JB |
4153 | spin_unlock(&space_info->lock); |
4154 | /* | |
4155 | * If we have a trans handle we can't wait because the flusher | |
4156 | * may have to commit the transaction, which would mean we would | |
4157 | * deadlock since we are waiting for the flusher to finish, but | |
4158 | * hold the current transaction open. | |
4159 | */ | |
663350ac | 4160 | if (current->journal_info) |
fdb5effd | 4161 | return -EAGAIN; |
b9688bb8 AJ |
4162 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
4163 | /* Must have been killed, return */ | |
4164 | if (ret) | |
fdb5effd JB |
4165 | return -EINTR; |
4166 | ||
4167 | spin_lock(&space_info->lock); | |
4168 | } | |
4169 | ||
4170 | ret = -ENOSPC; | |
2bf64758 JB |
4171 | used = space_info->bytes_used + space_info->bytes_reserved + |
4172 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4173 | space_info->bytes_may_use; | |
9ed74f2d | 4174 | |
8bb8ab2e JB |
4175 | /* |
4176 | * The idea here is that we've not already over-reserved the block group | |
4177 | * then we can go ahead and save our reservation first and then start | |
4178 | * flushing if we need to. Otherwise if we've already overcommitted | |
4179 | * lets start flushing stuff first and then come back and try to make | |
4180 | * our reservation. | |
4181 | */ | |
2bf64758 JB |
4182 | if (used <= space_info->total_bytes) { |
4183 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 4184 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 4185 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 4186 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
4187 | ret = 0; |
4188 | } else { | |
4189 | /* | |
4190 | * Ok set num_bytes to orig_bytes since we aren't | |
4191 | * overocmmitted, this way we only try and reclaim what | |
4192 | * we need. | |
4193 | */ | |
4194 | num_bytes = orig_bytes; | |
4195 | } | |
4196 | } else { | |
4197 | /* | |
4198 | * Ok we're over committed, set num_bytes to the overcommitted | |
4199 | * amount plus the amount of bytes that we need for this | |
4200 | * reservation. | |
4201 | */ | |
2bf64758 | 4202 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 4203 | (orig_bytes * 2); |
8bb8ab2e | 4204 | } |
9ed74f2d | 4205 | |
44734ed1 JB |
4206 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { |
4207 | space_info->bytes_may_use += orig_bytes; | |
4208 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4209 | space_info->flags, orig_bytes, | |
4210 | 1); | |
4211 | ret = 0; | |
2bf64758 JB |
4212 | } |
4213 | ||
8bb8ab2e JB |
4214 | /* |
4215 | * Couldn't make our reservation, save our place so while we're trying | |
4216 | * to reclaim space we can actually use it instead of somebody else | |
4217 | * stealing it from us. | |
08e007d2 MX |
4218 | * |
4219 | * We make the other tasks wait for the flush only when we can flush | |
4220 | * all things. | |
8bb8ab2e | 4221 | */ |
72bcd99d | 4222 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
fdb5effd JB |
4223 | flushing = true; |
4224 | space_info->flush = 1; | |
8bb8ab2e | 4225 | } |
9ed74f2d | 4226 | |
f0486c68 | 4227 | spin_unlock(&space_info->lock); |
9ed74f2d | 4228 | |
08e007d2 | 4229 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
8bb8ab2e | 4230 | goto out; |
f0486c68 | 4231 | |
96c3f433 JB |
4232 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
4233 | flush_state); | |
4234 | flush_state++; | |
08e007d2 MX |
4235 | |
4236 | /* | |
4237 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock | |
4238 | * would happen. So skip delalloc flush. | |
4239 | */ | |
4240 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
4241 | (flush_state == FLUSH_DELALLOC || | |
4242 | flush_state == FLUSH_DELALLOC_WAIT)) | |
4243 | flush_state = ALLOC_CHUNK; | |
4244 | ||
96c3f433 | 4245 | if (!ret) |
8bb8ab2e | 4246 | goto again; |
08e007d2 MX |
4247 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
4248 | flush_state < COMMIT_TRANS) | |
4249 | goto again; | |
4250 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
4251 | flush_state <= COMMIT_TRANS) | |
8bb8ab2e JB |
4252 | goto again; |
4253 | ||
4254 | out: | |
5d80366e JB |
4255 | if (ret == -ENOSPC && |
4256 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
4257 | struct btrfs_block_rsv *global_rsv = | |
4258 | &root->fs_info->global_block_rsv; | |
4259 | ||
4260 | if (block_rsv != global_rsv && | |
4261 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
4262 | ret = 0; | |
4263 | } | |
fdb5effd | 4264 | if (flushing) { |
8bb8ab2e | 4265 | spin_lock(&space_info->lock); |
fdb5effd JB |
4266 | space_info->flush = 0; |
4267 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 4268 | spin_unlock(&space_info->lock); |
f0486c68 | 4269 | } |
f0486c68 YZ |
4270 | return ret; |
4271 | } | |
4272 | ||
79787eaa JM |
4273 | static struct btrfs_block_rsv *get_block_rsv( |
4274 | const struct btrfs_trans_handle *trans, | |
4275 | const struct btrfs_root *root) | |
f0486c68 | 4276 | { |
4c13d758 JB |
4277 | struct btrfs_block_rsv *block_rsv = NULL; |
4278 | ||
0e721106 JB |
4279 | if (root->ref_cows) |
4280 | block_rsv = trans->block_rsv; | |
4281 | ||
4282 | if (root == root->fs_info->csum_root && trans->adding_csums) | |
f0486c68 | 4283 | block_rsv = trans->block_rsv; |
4c13d758 JB |
4284 | |
4285 | if (!block_rsv) | |
f0486c68 YZ |
4286 | block_rsv = root->block_rsv; |
4287 | ||
4288 | if (!block_rsv) | |
4289 | block_rsv = &root->fs_info->empty_block_rsv; | |
4290 | ||
4291 | return block_rsv; | |
4292 | } | |
4293 | ||
4294 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
4295 | u64 num_bytes) | |
4296 | { | |
4297 | int ret = -ENOSPC; | |
4298 | spin_lock(&block_rsv->lock); | |
4299 | if (block_rsv->reserved >= num_bytes) { | |
4300 | block_rsv->reserved -= num_bytes; | |
4301 | if (block_rsv->reserved < block_rsv->size) | |
4302 | block_rsv->full = 0; | |
4303 | ret = 0; | |
4304 | } | |
4305 | spin_unlock(&block_rsv->lock); | |
4306 | return ret; | |
4307 | } | |
4308 | ||
4309 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
4310 | u64 num_bytes, int update_size) | |
4311 | { | |
4312 | spin_lock(&block_rsv->lock); | |
4313 | block_rsv->reserved += num_bytes; | |
4314 | if (update_size) | |
4315 | block_rsv->size += num_bytes; | |
4316 | else if (block_rsv->reserved >= block_rsv->size) | |
4317 | block_rsv->full = 1; | |
4318 | spin_unlock(&block_rsv->lock); | |
4319 | } | |
4320 | ||
8c2a3ca2 JB |
4321 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
4322 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 4323 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
4324 | { |
4325 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
4326 | ||
4327 | spin_lock(&block_rsv->lock); | |
4328 | if (num_bytes == (u64)-1) | |
4329 | num_bytes = block_rsv->size; | |
4330 | block_rsv->size -= num_bytes; | |
4331 | if (block_rsv->reserved >= block_rsv->size) { | |
4332 | num_bytes = block_rsv->reserved - block_rsv->size; | |
4333 | block_rsv->reserved = block_rsv->size; | |
4334 | block_rsv->full = 1; | |
4335 | } else { | |
4336 | num_bytes = 0; | |
4337 | } | |
4338 | spin_unlock(&block_rsv->lock); | |
4339 | ||
4340 | if (num_bytes > 0) { | |
4341 | if (dest) { | |
e9e22899 JB |
4342 | spin_lock(&dest->lock); |
4343 | if (!dest->full) { | |
4344 | u64 bytes_to_add; | |
4345 | ||
4346 | bytes_to_add = dest->size - dest->reserved; | |
4347 | bytes_to_add = min(num_bytes, bytes_to_add); | |
4348 | dest->reserved += bytes_to_add; | |
4349 | if (dest->reserved >= dest->size) | |
4350 | dest->full = 1; | |
4351 | num_bytes -= bytes_to_add; | |
4352 | } | |
4353 | spin_unlock(&dest->lock); | |
4354 | } | |
4355 | if (num_bytes) { | |
f0486c68 | 4356 | spin_lock(&space_info->lock); |
fb25e914 | 4357 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4358 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4359 | space_info->flags, num_bytes, 0); |
36e39c40 | 4360 | space_info->reservation_progress++; |
f0486c68 | 4361 | spin_unlock(&space_info->lock); |
4e06bdd6 | 4362 | } |
9ed74f2d | 4363 | } |
f0486c68 | 4364 | } |
4e06bdd6 | 4365 | |
f0486c68 YZ |
4366 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
4367 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
4368 | { | |
4369 | int ret; | |
9ed74f2d | 4370 | |
f0486c68 YZ |
4371 | ret = block_rsv_use_bytes(src, num_bytes); |
4372 | if (ret) | |
4373 | return ret; | |
9ed74f2d | 4374 | |
f0486c68 | 4375 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
4376 | return 0; |
4377 | } | |
4378 | ||
66d8f3dd | 4379 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 4380 | { |
f0486c68 YZ |
4381 | memset(rsv, 0, sizeof(*rsv)); |
4382 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 4383 | rsv->type = type; |
f0486c68 YZ |
4384 | } |
4385 | ||
66d8f3dd MX |
4386 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
4387 | unsigned short type) | |
f0486c68 YZ |
4388 | { |
4389 | struct btrfs_block_rsv *block_rsv; | |
4390 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 4391 | |
f0486c68 YZ |
4392 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
4393 | if (!block_rsv) | |
4394 | return NULL; | |
9ed74f2d | 4395 | |
66d8f3dd | 4396 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
4397 | block_rsv->space_info = __find_space_info(fs_info, |
4398 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
4399 | return block_rsv; |
4400 | } | |
9ed74f2d | 4401 | |
f0486c68 YZ |
4402 | void btrfs_free_block_rsv(struct btrfs_root *root, |
4403 | struct btrfs_block_rsv *rsv) | |
4404 | { | |
2aaa6655 JB |
4405 | if (!rsv) |
4406 | return; | |
dabdb640 JB |
4407 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
4408 | kfree(rsv); | |
9ed74f2d JB |
4409 | } |
4410 | ||
08e007d2 MX |
4411 | int btrfs_block_rsv_add(struct btrfs_root *root, |
4412 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
4413 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4414 | { |
f0486c68 | 4415 | int ret; |
9ed74f2d | 4416 | |
f0486c68 YZ |
4417 | if (num_bytes == 0) |
4418 | return 0; | |
8bb8ab2e | 4419 | |
61b520a9 | 4420 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
4421 | if (!ret) { |
4422 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
4423 | return 0; | |
4424 | } | |
9ed74f2d | 4425 | |
f0486c68 | 4426 | return ret; |
f0486c68 | 4427 | } |
9ed74f2d | 4428 | |
4a92b1b8 | 4429 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4430 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4431 | { |
4432 | u64 num_bytes = 0; | |
f0486c68 | 4433 | int ret = -ENOSPC; |
9ed74f2d | 4434 | |
f0486c68 YZ |
4435 | if (!block_rsv) |
4436 | return 0; | |
9ed74f2d | 4437 | |
f0486c68 | 4438 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4439 | num_bytes = div_factor(block_rsv->size, min_factor); |
4440 | if (block_rsv->reserved >= num_bytes) | |
4441 | ret = 0; | |
4442 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4443 | |
36ba022a JB |
4444 | return ret; |
4445 | } | |
4446 | ||
08e007d2 MX |
4447 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4448 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
4449 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
4450 | { |
4451 | u64 num_bytes = 0; | |
4452 | int ret = -ENOSPC; | |
4453 | ||
4454 | if (!block_rsv) | |
4455 | return 0; | |
4456 | ||
4457 | spin_lock(&block_rsv->lock); | |
4458 | num_bytes = min_reserved; | |
13553e52 | 4459 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4460 | ret = 0; |
13553e52 | 4461 | else |
f0486c68 | 4462 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4463 | spin_unlock(&block_rsv->lock); |
13553e52 | 4464 | |
f0486c68 YZ |
4465 | if (!ret) |
4466 | return 0; | |
4467 | ||
aa38a711 | 4468 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4469 | if (!ret) { |
4470 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4471 | return 0; |
6a63209f | 4472 | } |
9ed74f2d | 4473 | |
13553e52 | 4474 | return ret; |
f0486c68 YZ |
4475 | } |
4476 | ||
4477 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
4478 | struct btrfs_block_rsv *dst_rsv, | |
4479 | u64 num_bytes) | |
4480 | { | |
4481 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4482 | } | |
4483 | ||
4484 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4485 | struct btrfs_block_rsv *block_rsv, | |
4486 | u64 num_bytes) | |
4487 | { | |
4488 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4489 | if (global_rsv->full || global_rsv == block_rsv || | |
4490 | block_rsv->space_info != global_rsv->space_info) | |
4491 | global_rsv = NULL; | |
8c2a3ca2 JB |
4492 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4493 | num_bytes); | |
6a63209f JB |
4494 | } |
4495 | ||
4496 | /* | |
8929ecfa YZ |
4497 | * helper to calculate size of global block reservation. |
4498 | * the desired value is sum of space used by extent tree, | |
4499 | * checksum tree and root tree | |
6a63209f | 4500 | */ |
8929ecfa | 4501 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4502 | { |
8929ecfa YZ |
4503 | struct btrfs_space_info *sinfo; |
4504 | u64 num_bytes; | |
4505 | u64 meta_used; | |
4506 | u64 data_used; | |
6c41761f | 4507 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4508 | |
8929ecfa YZ |
4509 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4510 | spin_lock(&sinfo->lock); | |
4511 | data_used = sinfo->bytes_used; | |
4512 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4513 | |
8929ecfa YZ |
4514 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4515 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4516 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4517 | data_used = 0; | |
8929ecfa YZ |
4518 | meta_used = sinfo->bytes_used; |
4519 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4520 | |
8929ecfa YZ |
4521 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4522 | csum_size * 2; | |
4523 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4524 | |
8929ecfa | 4525 | if (num_bytes * 3 > meta_used) |
8e62c2de | 4526 | num_bytes = div64_u64(meta_used, 3); |
ab6e2410 | 4527 | |
8929ecfa YZ |
4528 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4529 | } | |
6a63209f | 4530 | |
8929ecfa YZ |
4531 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4532 | { | |
4533 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4534 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4535 | u64 num_bytes; | |
6a63209f | 4536 | |
8929ecfa | 4537 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4538 | |
8929ecfa | 4539 | spin_lock(&sinfo->lock); |
1f699d38 | 4540 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 4541 | |
fdf30d1c | 4542 | block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024); |
4e06bdd6 | 4543 | |
8929ecfa | 4544 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4545 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4546 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4547 | |
4548 | if (sinfo->total_bytes > num_bytes) { | |
4549 | num_bytes = sinfo->total_bytes - num_bytes; | |
4550 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4551 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4552 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4553 | sinfo->flags, num_bytes, 1); |
6a63209f | 4554 | } |
6a63209f | 4555 | |
8929ecfa YZ |
4556 | if (block_rsv->reserved >= block_rsv->size) { |
4557 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4558 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4559 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4560 | sinfo->flags, num_bytes, 0); |
36e39c40 | 4561 | sinfo->reservation_progress++; |
8929ecfa YZ |
4562 | block_rsv->reserved = block_rsv->size; |
4563 | block_rsv->full = 1; | |
4564 | } | |
182608c8 | 4565 | |
8929ecfa | 4566 | spin_unlock(&block_rsv->lock); |
1f699d38 | 4567 | spin_unlock(&sinfo->lock); |
6a63209f JB |
4568 | } |
4569 | ||
f0486c68 | 4570 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4571 | { |
f0486c68 | 4572 | struct btrfs_space_info *space_info; |
6a63209f | 4573 | |
f0486c68 YZ |
4574 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4575 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4576 | |
f0486c68 | 4577 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4578 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4579 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4580 | fs_info->trans_block_rsv.space_info = space_info; |
4581 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4582 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4583 | |
8929ecfa YZ |
4584 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4585 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4586 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4587 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4588 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4589 | |
8929ecfa | 4590 | update_global_block_rsv(fs_info); |
6a63209f JB |
4591 | } |
4592 | ||
8929ecfa | 4593 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4594 | { |
8c2a3ca2 JB |
4595 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4596 | (u64)-1); | |
8929ecfa YZ |
4597 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4598 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4599 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4600 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4601 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4602 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4603 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4604 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4605 | } |
4606 | ||
a22285a6 YZ |
4607 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4608 | struct btrfs_root *root) | |
6a63209f | 4609 | { |
0e721106 JB |
4610 | if (!trans->block_rsv) |
4611 | return; | |
4612 | ||
a22285a6 YZ |
4613 | if (!trans->bytes_reserved) |
4614 | return; | |
6a63209f | 4615 | |
e77266e4 | 4616 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 4617 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 4618 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4619 | trans->bytes_reserved = 0; |
4620 | } | |
6a63209f | 4621 | |
79787eaa | 4622 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
4623 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4624 | struct inode *inode) | |
4625 | { | |
4626 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4627 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4628 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4629 | ||
4630 | /* | |
fcb80c2a JB |
4631 | * We need to hold space in order to delete our orphan item once we've |
4632 | * added it, so this takes the reservation so we can release it later | |
4633 | * when we are truly done with the orphan item. | |
d68fc57b | 4634 | */ |
ff5714cc | 4635 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4636 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4637 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4638 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4639 | } |
4640 | ||
d68fc57b | 4641 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4642 | { |
d68fc57b | 4643 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4644 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4645 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4646 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4647 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4648 | } | |
97e728d4 | 4649 | |
d5c12070 MX |
4650 | /* |
4651 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
4652 | * root: the root of the parent directory | |
4653 | * rsv: block reservation | |
4654 | * items: the number of items that we need do reservation | |
4655 | * qgroup_reserved: used to return the reserved size in qgroup | |
4656 | * | |
4657 | * This function is used to reserve the space for snapshot/subvolume | |
4658 | * creation and deletion. Those operations are different with the | |
4659 | * common file/directory operations, they change two fs/file trees | |
4660 | * and root tree, the number of items that the qgroup reserves is | |
4661 | * different with the free space reservation. So we can not use | |
4662 | * the space reseravtion mechanism in start_transaction(). | |
4663 | */ | |
4664 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
4665 | struct btrfs_block_rsv *rsv, | |
4666 | int items, | |
4667 | u64 *qgroup_reserved) | |
a22285a6 | 4668 | { |
d5c12070 MX |
4669 | u64 num_bytes; |
4670 | int ret; | |
4671 | ||
4672 | if (root->fs_info->quota_enabled) { | |
4673 | /* One for parent inode, two for dir entries */ | |
4674 | num_bytes = 3 * root->leafsize; | |
4675 | ret = btrfs_qgroup_reserve(root, num_bytes); | |
4676 | if (ret) | |
4677 | return ret; | |
4678 | } else { | |
4679 | num_bytes = 0; | |
4680 | } | |
4681 | ||
4682 | *qgroup_reserved = num_bytes; | |
4683 | ||
4684 | num_bytes = btrfs_calc_trans_metadata_size(root, items); | |
4685 | rsv->space_info = __find_space_info(root->fs_info, | |
4686 | BTRFS_BLOCK_GROUP_METADATA); | |
4687 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
4688 | BTRFS_RESERVE_FLUSH_ALL); | |
4689 | if (ret) { | |
4690 | if (*qgroup_reserved) | |
4691 | btrfs_qgroup_free(root, *qgroup_reserved); | |
4692 | } | |
4693 | ||
4694 | return ret; | |
4695 | } | |
4696 | ||
4697 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, | |
4698 | struct btrfs_block_rsv *rsv, | |
4699 | u64 qgroup_reserved) | |
4700 | { | |
4701 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
4702 | if (qgroup_reserved) | |
4703 | btrfs_qgroup_free(root, qgroup_reserved); | |
97e728d4 JB |
4704 | } |
4705 | ||
7709cde3 JB |
4706 | /** |
4707 | * drop_outstanding_extent - drop an outstanding extent | |
4708 | * @inode: the inode we're dropping the extent for | |
4709 | * | |
4710 | * This is called when we are freeing up an outstanding extent, either called | |
4711 | * after an error or after an extent is written. This will return the number of | |
4712 | * reserved extents that need to be freed. This must be called with | |
4713 | * BTRFS_I(inode)->lock held. | |
4714 | */ | |
9e0baf60 JB |
4715 | static unsigned drop_outstanding_extent(struct inode *inode) |
4716 | { | |
7fd2ae21 | 4717 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4718 | unsigned dropped_extents = 0; |
4719 | ||
9e0baf60 JB |
4720 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4721 | BTRFS_I(inode)->outstanding_extents--; | |
4722 | ||
7fd2ae21 | 4723 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
4724 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4725 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 4726 | drop_inode_space = 1; |
7fd2ae21 | 4727 | |
9e0baf60 JB |
4728 | /* |
4729 | * If we have more or the same amount of outsanding extents than we have | |
4730 | * reserved then we need to leave the reserved extents count alone. | |
4731 | */ | |
4732 | if (BTRFS_I(inode)->outstanding_extents >= | |
4733 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4734 | return drop_inode_space; |
9e0baf60 JB |
4735 | |
4736 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4737 | BTRFS_I(inode)->outstanding_extents; | |
4738 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4739 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4740 | } |
4741 | ||
7709cde3 JB |
4742 | /** |
4743 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4744 | * reserved/free'd for the given bytes. | |
4745 | * @inode: the inode we're manipulating | |
4746 | * @num_bytes: the number of bytes in question | |
4747 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4748 | * | |
4749 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4750 | * correct amount of metadata that must either be reserved or freed. We | |
4751 | * calculate how many checksums we can fit into one leaf and then divide the | |
4752 | * number of bytes that will need to be checksumed by this value to figure out | |
4753 | * how many checksums will be required. If we are adding bytes then the number | |
4754 | * may go up and we will return the number of additional bytes that must be | |
4755 | * reserved. If it is going down we will return the number of bytes that must | |
4756 | * be freed. | |
4757 | * | |
4758 | * This must be called with BTRFS_I(inode)->lock held. | |
4759 | */ | |
4760 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4761 | int reserve) | |
6324fbf3 | 4762 | { |
7709cde3 JB |
4763 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4764 | u64 csum_size; | |
4765 | int num_csums_per_leaf; | |
4766 | int num_csums; | |
4767 | int old_csums; | |
4768 | ||
4769 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4770 | BTRFS_I(inode)->csum_bytes == 0) | |
4771 | return 0; | |
4772 | ||
4773 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4774 | if (reserve) | |
4775 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4776 | else | |
4777 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4778 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4779 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4780 | sizeof(struct btrfs_csum_item) + | |
4781 | sizeof(struct btrfs_disk_key)); | |
4782 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4783 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4784 | num_csums = num_csums / num_csums_per_leaf; | |
4785 | ||
4786 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4787 | old_csums = old_csums / num_csums_per_leaf; | |
4788 | ||
4789 | /* No change, no need to reserve more */ | |
4790 | if (old_csums == num_csums) | |
4791 | return 0; | |
4792 | ||
4793 | if (reserve) | |
4794 | return btrfs_calc_trans_metadata_size(root, | |
4795 | num_csums - old_csums); | |
4796 | ||
4797 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4798 | } |
c146afad | 4799 | |
0ca1f7ce YZ |
4800 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4801 | { | |
4802 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4803 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4804 | u64 to_reserve = 0; |
660d3f6c | 4805 | u64 csum_bytes; |
9e0baf60 | 4806 | unsigned nr_extents = 0; |
660d3f6c | 4807 | int extra_reserve = 0; |
08e007d2 | 4808 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 4809 | int ret = 0; |
c64c2bd8 | 4810 | bool delalloc_lock = true; |
88e081bf WS |
4811 | u64 to_free = 0; |
4812 | unsigned dropped; | |
6324fbf3 | 4813 | |
c64c2bd8 JB |
4814 | /* If we are a free space inode we need to not flush since we will be in |
4815 | * the middle of a transaction commit. We also don't need the delalloc | |
4816 | * mutex since we won't race with anybody. We need this mostly to make | |
4817 | * lockdep shut its filthy mouth. | |
4818 | */ | |
4819 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 4820 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 JB |
4821 | delalloc_lock = false; |
4822 | } | |
c09544e0 | 4823 | |
08e007d2 MX |
4824 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
4825 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4826 | schedule_timeout(1); |
ec44a35c | 4827 | |
c64c2bd8 JB |
4828 | if (delalloc_lock) |
4829 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
4830 | ||
0ca1f7ce | 4831 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4832 | |
9e0baf60 JB |
4833 | spin_lock(&BTRFS_I(inode)->lock); |
4834 | BTRFS_I(inode)->outstanding_extents++; | |
4835 | ||
4836 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4837 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4838 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4839 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4840 | |
7fd2ae21 JB |
4841 | /* |
4842 | * Add an item to reserve for updating the inode when we complete the | |
4843 | * delalloc io. | |
4844 | */ | |
72ac3c0d JB |
4845 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4846 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 4847 | nr_extents++; |
660d3f6c | 4848 | extra_reserve = 1; |
593060d7 | 4849 | } |
7fd2ae21 JB |
4850 | |
4851 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4852 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4853 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4854 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4855 | |
88e081bf | 4856 | if (root->fs_info->quota_enabled) { |
c5567237 AJ |
4857 | ret = btrfs_qgroup_reserve(root, num_bytes + |
4858 | nr_extents * root->leafsize); | |
88e081bf WS |
4859 | if (ret) |
4860 | goto out_fail; | |
4861 | } | |
c5567237 | 4862 | |
88e081bf WS |
4863 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
4864 | if (unlikely(ret)) { | |
4865 | if (root->fs_info->quota_enabled) | |
4b5829a8 MX |
4866 | btrfs_qgroup_free(root, num_bytes + |
4867 | nr_extents * root->leafsize); | |
88e081bf | 4868 | goto out_fail; |
9e0baf60 | 4869 | } |
25179201 | 4870 | |
660d3f6c JB |
4871 | spin_lock(&BTRFS_I(inode)->lock); |
4872 | if (extra_reserve) { | |
72ac3c0d JB |
4873 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4874 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
4875 | nr_extents--; |
4876 | } | |
4877 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
4878 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
4879 | |
4880 | if (delalloc_lock) | |
4881 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 4882 | |
8c2a3ca2 JB |
4883 | if (to_reserve) |
4884 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
4885 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
4886 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4887 | ||
0ca1f7ce | 4888 | return 0; |
88e081bf WS |
4889 | |
4890 | out_fail: | |
4891 | spin_lock(&BTRFS_I(inode)->lock); | |
4892 | dropped = drop_outstanding_extent(inode); | |
4893 | /* | |
4894 | * If the inodes csum_bytes is the same as the original | |
4895 | * csum_bytes then we know we haven't raced with any free()ers | |
4896 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 4897 | */ |
f4881bc7 | 4898 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { |
88e081bf | 4899 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 JB |
4900 | } else { |
4901 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; | |
4902 | u64 bytes; | |
4903 | ||
4904 | /* | |
4905 | * This is tricky, but first we need to figure out how much we | |
4906 | * free'd from any free-ers that occured during this | |
4907 | * reservation, so we reset ->csum_bytes to the csum_bytes | |
4908 | * before we dropped our lock, and then call the free for the | |
4909 | * number of bytes that were freed while we were trying our | |
4910 | * reservation. | |
4911 | */ | |
4912 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; | |
4913 | BTRFS_I(inode)->csum_bytes = csum_bytes; | |
4914 | to_free = calc_csum_metadata_size(inode, bytes, 0); | |
4915 | ||
4916 | ||
4917 | /* | |
4918 | * Now we need to see how much we would have freed had we not | |
4919 | * been making this reservation and our ->csum_bytes were not | |
4920 | * artificially inflated. | |
4921 | */ | |
4922 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; | |
4923 | bytes = csum_bytes - orig_csum_bytes; | |
4924 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
4925 | ||
4926 | /* | |
4927 | * Now reset ->csum_bytes to what it should be. If bytes is | |
4928 | * more than to_free then we would have free'd more space had we | |
4929 | * not had an artificially high ->csum_bytes, so we need to free | |
4930 | * the remainder. If bytes is the same or less then we don't | |
4931 | * need to do anything, the other free-ers did the correct | |
4932 | * thing. | |
4933 | */ | |
4934 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; | |
4935 | if (bytes > to_free) | |
4936 | to_free = bytes - to_free; | |
4937 | else | |
4938 | to_free = 0; | |
4939 | } | |
88e081bf WS |
4940 | spin_unlock(&BTRFS_I(inode)->lock); |
4941 | if (dropped) | |
4942 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
4943 | ||
4944 | if (to_free) { | |
4945 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
4946 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
4947 | btrfs_ino(inode), to_free, 0); | |
4948 | } | |
4949 | if (delalloc_lock) | |
4950 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
4951 | return ret; | |
0ca1f7ce YZ |
4952 | } |
4953 | ||
7709cde3 JB |
4954 | /** |
4955 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
4956 | * @inode: the inode to release the reservation for | |
4957 | * @num_bytes: the number of bytes we're releasing | |
4958 | * | |
4959 | * This will release the metadata reservation for an inode. This can be called | |
4960 | * once we complete IO for a given set of bytes to release their metadata | |
4961 | * reservations. | |
4962 | */ | |
0ca1f7ce YZ |
4963 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
4964 | { | |
4965 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4966 | u64 to_free = 0; |
4967 | unsigned dropped; | |
0ca1f7ce YZ |
4968 | |
4969 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 4970 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4971 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4972 | |
0934856d MX |
4973 | if (num_bytes) |
4974 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
7709cde3 | 4975 | spin_unlock(&BTRFS_I(inode)->lock); |
9e0baf60 JB |
4976 | if (dropped > 0) |
4977 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 4978 | |
8c2a3ca2 JB |
4979 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
4980 | btrfs_ino(inode), to_free, 0); | |
c5567237 AJ |
4981 | if (root->fs_info->quota_enabled) { |
4982 | btrfs_qgroup_free(root, num_bytes + | |
4983 | dropped * root->leafsize); | |
4984 | } | |
4985 | ||
0ca1f7ce YZ |
4986 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
4987 | to_free); | |
4988 | } | |
4989 | ||
7709cde3 JB |
4990 | /** |
4991 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
4992 | * @inode: inode we're writing to | |
4993 | * @num_bytes: the number of bytes we want to allocate | |
4994 | * | |
4995 | * This will do the following things | |
4996 | * | |
4997 | * o reserve space in the data space info for num_bytes | |
4998 | * o reserve space in the metadata space info based on number of outstanding | |
4999 | * extents and how much csums will be needed | |
5000 | * o add to the inodes ->delalloc_bytes | |
5001 | * o add it to the fs_info's delalloc inodes list. | |
5002 | * | |
5003 | * This will return 0 for success and -ENOSPC if there is no space left. | |
5004 | */ | |
0ca1f7ce YZ |
5005 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
5006 | { | |
5007 | int ret; | |
5008 | ||
5009 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 5010 | if (ret) |
0ca1f7ce YZ |
5011 | return ret; |
5012 | ||
5013 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
5014 | if (ret) { | |
5015 | btrfs_free_reserved_data_space(inode, num_bytes); | |
5016 | return ret; | |
5017 | } | |
5018 | ||
5019 | return 0; | |
5020 | } | |
5021 | ||
7709cde3 JB |
5022 | /** |
5023 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
5024 | * @inode: inode we're releasing space for | |
5025 | * @num_bytes: the number of bytes we want to free up | |
5026 | * | |
5027 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
5028 | * called in the case that we don't need the metadata AND data reservations | |
5029 | * anymore. So if there is an error or we insert an inline extent. | |
5030 | * | |
5031 | * This function will release the metadata space that was not used and will | |
5032 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
5033 | * list if there are no delalloc bytes left. | |
5034 | */ | |
0ca1f7ce YZ |
5035 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
5036 | { | |
5037 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
5038 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
5039 | } |
5040 | ||
c53d613e | 5041 | static int update_block_group(struct btrfs_root *root, |
f0486c68 | 5042 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 5043 | { |
0af3d00b | 5044 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 5045 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 5046 | u64 total = num_bytes; |
9078a3e1 | 5047 | u64 old_val; |
db94535d | 5048 | u64 byte_in_group; |
0af3d00b | 5049 | int factor; |
3e1ad54f | 5050 | |
5d4f98a2 YZ |
5051 | /* block accounting for super block */ |
5052 | spin_lock(&info->delalloc_lock); | |
6c41761f | 5053 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
5054 | if (alloc) |
5055 | old_val += num_bytes; | |
5056 | else | |
5057 | old_val -= num_bytes; | |
6c41761f | 5058 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
5d4f98a2 YZ |
5059 | spin_unlock(&info->delalloc_lock); |
5060 | ||
d397712b | 5061 | while (total) { |
db94535d | 5062 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 5063 | if (!cache) |
79787eaa | 5064 | return -ENOENT; |
b742bb82 YZ |
5065 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
5066 | BTRFS_BLOCK_GROUP_RAID1 | | |
5067 | BTRFS_BLOCK_GROUP_RAID10)) | |
5068 | factor = 2; | |
5069 | else | |
5070 | factor = 1; | |
9d66e233 JB |
5071 | /* |
5072 | * If this block group has free space cache written out, we | |
5073 | * need to make sure to load it if we are removing space. This | |
5074 | * is because we need the unpinning stage to actually add the | |
5075 | * space back to the block group, otherwise we will leak space. | |
5076 | */ | |
5077 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 5078 | cache_block_group(cache, 1); |
0af3d00b | 5079 | |
db94535d CM |
5080 | byte_in_group = bytenr - cache->key.objectid; |
5081 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 5082 | |
25179201 | 5083 | spin_lock(&cache->space_info->lock); |
c286ac48 | 5084 | spin_lock(&cache->lock); |
0af3d00b | 5085 | |
73bc1876 | 5086 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
5087 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
5088 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
5089 | ||
0f9dd46c | 5090 | cache->dirty = 1; |
9078a3e1 | 5091 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 5092 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 5093 | if (alloc) { |
db94535d | 5094 | old_val += num_bytes; |
11833d66 YZ |
5095 | btrfs_set_block_group_used(&cache->item, old_val); |
5096 | cache->reserved -= num_bytes; | |
11833d66 | 5097 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
5098 | cache->space_info->bytes_used += num_bytes; |
5099 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 5100 | spin_unlock(&cache->lock); |
25179201 | 5101 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 5102 | } else { |
db94535d | 5103 | old_val -= num_bytes; |
c286ac48 | 5104 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
5105 | cache->pinned += num_bytes; |
5106 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 5107 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 5108 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 5109 | spin_unlock(&cache->lock); |
25179201 | 5110 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 5111 | |
f0486c68 YZ |
5112 | set_extent_dirty(info->pinned_extents, |
5113 | bytenr, bytenr + num_bytes - 1, | |
5114 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 5115 | } |
fa9c0d79 | 5116 | btrfs_put_block_group(cache); |
db94535d CM |
5117 | total -= num_bytes; |
5118 | bytenr += num_bytes; | |
9078a3e1 CM |
5119 | } |
5120 | return 0; | |
5121 | } | |
6324fbf3 | 5122 | |
a061fc8d CM |
5123 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
5124 | { | |
0f9dd46c | 5125 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 5126 | u64 bytenr; |
0f9dd46c | 5127 | |
a1897fdd LB |
5128 | spin_lock(&root->fs_info->block_group_cache_lock); |
5129 | bytenr = root->fs_info->first_logical_byte; | |
5130 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
5131 | ||
5132 | if (bytenr < (u64)-1) | |
5133 | return bytenr; | |
5134 | ||
0f9dd46c JB |
5135 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
5136 | if (!cache) | |
a061fc8d | 5137 | return 0; |
0f9dd46c | 5138 | |
d2fb3437 | 5139 | bytenr = cache->key.objectid; |
fa9c0d79 | 5140 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
5141 | |
5142 | return bytenr; | |
a061fc8d CM |
5143 | } |
5144 | ||
f0486c68 YZ |
5145 | static int pin_down_extent(struct btrfs_root *root, |
5146 | struct btrfs_block_group_cache *cache, | |
5147 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 5148 | { |
11833d66 YZ |
5149 | spin_lock(&cache->space_info->lock); |
5150 | spin_lock(&cache->lock); | |
5151 | cache->pinned += num_bytes; | |
5152 | cache->space_info->bytes_pinned += num_bytes; | |
5153 | if (reserved) { | |
5154 | cache->reserved -= num_bytes; | |
5155 | cache->space_info->bytes_reserved -= num_bytes; | |
5156 | } | |
5157 | spin_unlock(&cache->lock); | |
5158 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 5159 | |
f0486c68 YZ |
5160 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
5161 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
5162 | return 0; | |
5163 | } | |
68b38550 | 5164 | |
f0486c68 YZ |
5165 | /* |
5166 | * this function must be called within transaction | |
5167 | */ | |
5168 | int btrfs_pin_extent(struct btrfs_root *root, | |
5169 | u64 bytenr, u64 num_bytes, int reserved) | |
5170 | { | |
5171 | struct btrfs_block_group_cache *cache; | |
68b38550 | 5172 | |
f0486c68 | 5173 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 5174 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
5175 | |
5176 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
5177 | ||
5178 | btrfs_put_block_group(cache); | |
11833d66 YZ |
5179 | return 0; |
5180 | } | |
5181 | ||
f0486c68 | 5182 | /* |
e688b725 CM |
5183 | * this function must be called within transaction |
5184 | */ | |
dcfac415 | 5185 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
e688b725 CM |
5186 | u64 bytenr, u64 num_bytes) |
5187 | { | |
5188 | struct btrfs_block_group_cache *cache; | |
5189 | ||
5190 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
79787eaa | 5191 | BUG_ON(!cache); /* Logic error */ |
e688b725 CM |
5192 | |
5193 | /* | |
5194 | * pull in the free space cache (if any) so that our pin | |
5195 | * removes the free space from the cache. We have load_only set | |
5196 | * to one because the slow code to read in the free extents does check | |
5197 | * the pinned extents. | |
5198 | */ | |
f6373bf3 | 5199 | cache_block_group(cache, 1); |
e688b725 CM |
5200 | |
5201 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
5202 | ||
5203 | /* remove us from the free space cache (if we're there at all) */ | |
5204 | btrfs_remove_free_space(cache, bytenr, num_bytes); | |
5205 | btrfs_put_block_group(cache); | |
5206 | return 0; | |
5207 | } | |
5208 | ||
fb25e914 JB |
5209 | /** |
5210 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
5211 | * @cache: The cache we are manipulating | |
5212 | * @num_bytes: The number of bytes in question | |
5213 | * @reserve: One of the reservation enums | |
5214 | * | |
5215 | * This is called by the allocator when it reserves space, or by somebody who is | |
5216 | * freeing space that was never actually used on disk. For example if you | |
5217 | * reserve some space for a new leaf in transaction A and before transaction A | |
5218 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
5219 | * clear the reservation. | |
5220 | * | |
5221 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
5222 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
5223 | * delalloc bits in the io_tree. We have to do this since we could end up | |
5224 | * allocating less disk space for the amount of data we have reserved in the | |
5225 | * case of compression. | |
5226 | * | |
5227 | * If this is a reservation and the block group has become read only we cannot | |
5228 | * make the reservation and return -EAGAIN, otherwise this function always | |
5229 | * succeeds. | |
f0486c68 | 5230 | */ |
fb25e914 JB |
5231 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
5232 | u64 num_bytes, int reserve) | |
11833d66 | 5233 | { |
fb25e914 | 5234 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 5235 | int ret = 0; |
79787eaa | 5236 | |
fb25e914 JB |
5237 | spin_lock(&space_info->lock); |
5238 | spin_lock(&cache->lock); | |
5239 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
5240 | if (cache->ro) { |
5241 | ret = -EAGAIN; | |
5242 | } else { | |
fb25e914 JB |
5243 | cache->reserved += num_bytes; |
5244 | space_info->bytes_reserved += num_bytes; | |
5245 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 5246 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
5247 | "space_info", space_info->flags, |
5248 | num_bytes, 0); | |
fb25e914 JB |
5249 | space_info->bytes_may_use -= num_bytes; |
5250 | } | |
f0486c68 | 5251 | } |
fb25e914 JB |
5252 | } else { |
5253 | if (cache->ro) | |
5254 | space_info->bytes_readonly += num_bytes; | |
5255 | cache->reserved -= num_bytes; | |
5256 | space_info->bytes_reserved -= num_bytes; | |
5257 | space_info->reservation_progress++; | |
324ae4df | 5258 | } |
fb25e914 JB |
5259 | spin_unlock(&cache->lock); |
5260 | spin_unlock(&space_info->lock); | |
f0486c68 | 5261 | return ret; |
324ae4df | 5262 | } |
9078a3e1 | 5263 | |
143bede5 | 5264 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 5265 | struct btrfs_root *root) |
e8569813 | 5266 | { |
e8569813 | 5267 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
5268 | struct btrfs_caching_control *next; |
5269 | struct btrfs_caching_control *caching_ctl; | |
5270 | struct btrfs_block_group_cache *cache; | |
e8569813 | 5271 | |
11833d66 | 5272 | down_write(&fs_info->extent_commit_sem); |
25179201 | 5273 | |
11833d66 YZ |
5274 | list_for_each_entry_safe(caching_ctl, next, |
5275 | &fs_info->caching_block_groups, list) { | |
5276 | cache = caching_ctl->block_group; | |
5277 | if (block_group_cache_done(cache)) { | |
5278 | cache->last_byte_to_unpin = (u64)-1; | |
5279 | list_del_init(&caching_ctl->list); | |
5280 | put_caching_control(caching_ctl); | |
e8569813 | 5281 | } else { |
11833d66 | 5282 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 5283 | } |
e8569813 | 5284 | } |
11833d66 YZ |
5285 | |
5286 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
5287 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
5288 | else | |
5289 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
5290 | ||
5291 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
5292 | |
5293 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
5294 | } |
5295 | ||
11833d66 | 5296 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 5297 | { |
11833d66 YZ |
5298 | struct btrfs_fs_info *fs_info = root->fs_info; |
5299 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
5300 | struct btrfs_space_info *space_info; |
5301 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
11833d66 | 5302 | u64 len; |
7b398f8e | 5303 | bool readonly; |
ccd467d6 | 5304 | |
11833d66 | 5305 | while (start <= end) { |
7b398f8e | 5306 | readonly = false; |
11833d66 YZ |
5307 | if (!cache || |
5308 | start >= cache->key.objectid + cache->key.offset) { | |
5309 | if (cache) | |
5310 | btrfs_put_block_group(cache); | |
5311 | cache = btrfs_lookup_block_group(fs_info, start); | |
79787eaa | 5312 | BUG_ON(!cache); /* Logic error */ |
11833d66 YZ |
5313 | } |
5314 | ||
5315 | len = cache->key.objectid + cache->key.offset - start; | |
5316 | len = min(len, end + 1 - start); | |
5317 | ||
5318 | if (start < cache->last_byte_to_unpin) { | |
5319 | len = min(len, cache->last_byte_to_unpin - start); | |
5320 | btrfs_add_free_space(cache, start, len); | |
5321 | } | |
5322 | ||
f0486c68 | 5323 | start += len; |
7b398f8e | 5324 | space_info = cache->space_info; |
f0486c68 | 5325 | |
7b398f8e | 5326 | spin_lock(&space_info->lock); |
11833d66 YZ |
5327 | spin_lock(&cache->lock); |
5328 | cache->pinned -= len; | |
7b398f8e JB |
5329 | space_info->bytes_pinned -= len; |
5330 | if (cache->ro) { | |
5331 | space_info->bytes_readonly += len; | |
5332 | readonly = true; | |
5333 | } | |
11833d66 | 5334 | spin_unlock(&cache->lock); |
7b398f8e JB |
5335 | if (!readonly && global_rsv->space_info == space_info) { |
5336 | spin_lock(&global_rsv->lock); | |
5337 | if (!global_rsv->full) { | |
5338 | len = min(len, global_rsv->size - | |
5339 | global_rsv->reserved); | |
5340 | global_rsv->reserved += len; | |
5341 | space_info->bytes_may_use += len; | |
5342 | if (global_rsv->reserved >= global_rsv->size) | |
5343 | global_rsv->full = 1; | |
5344 | } | |
5345 | spin_unlock(&global_rsv->lock); | |
5346 | } | |
5347 | spin_unlock(&space_info->lock); | |
ccd467d6 | 5348 | } |
11833d66 YZ |
5349 | |
5350 | if (cache) | |
5351 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
5352 | return 0; |
5353 | } | |
5354 | ||
5355 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 5356 | struct btrfs_root *root) |
a28ec197 | 5357 | { |
11833d66 YZ |
5358 | struct btrfs_fs_info *fs_info = root->fs_info; |
5359 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
5360 | u64 start; |
5361 | u64 end; | |
a28ec197 | 5362 | int ret; |
a28ec197 | 5363 | |
79787eaa JM |
5364 | if (trans->aborted) |
5365 | return 0; | |
5366 | ||
11833d66 YZ |
5367 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
5368 | unpin = &fs_info->freed_extents[1]; | |
5369 | else | |
5370 | unpin = &fs_info->freed_extents[0]; | |
5371 | ||
d397712b | 5372 | while (1) { |
1a5bc167 | 5373 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 5374 | EXTENT_DIRTY, NULL); |
1a5bc167 | 5375 | if (ret) |
a28ec197 | 5376 | break; |
1f3c79a2 | 5377 | |
5378e607 LD |
5378 | if (btrfs_test_opt(root, DISCARD)) |
5379 | ret = btrfs_discard_extent(root, start, | |
5380 | end + 1 - start, NULL); | |
1f3c79a2 | 5381 | |
1a5bc167 | 5382 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 5383 | unpin_extent_range(root, start, end); |
b9473439 | 5384 | cond_resched(); |
a28ec197 | 5385 | } |
817d52f8 | 5386 | |
e20d96d6 CM |
5387 | return 0; |
5388 | } | |
5389 | ||
5d4f98a2 YZ |
5390 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
5391 | struct btrfs_root *root, | |
5392 | u64 bytenr, u64 num_bytes, u64 parent, | |
5393 | u64 root_objectid, u64 owner_objectid, | |
5394 | u64 owner_offset, int refs_to_drop, | |
5395 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 5396 | { |
e2fa7227 | 5397 | struct btrfs_key key; |
5d4f98a2 | 5398 | struct btrfs_path *path; |
1261ec42 CM |
5399 | struct btrfs_fs_info *info = root->fs_info; |
5400 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 5401 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
5402 | struct btrfs_extent_item *ei; |
5403 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 5404 | int ret; |
5d4f98a2 | 5405 | int is_data; |
952fccac CM |
5406 | int extent_slot = 0; |
5407 | int found_extent = 0; | |
5408 | int num_to_del = 1; | |
5d4f98a2 YZ |
5409 | u32 item_size; |
5410 | u64 refs; | |
3173a18f JB |
5411 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
5412 | SKINNY_METADATA); | |
037e6390 | 5413 | |
5caf2a00 | 5414 | path = btrfs_alloc_path(); |
54aa1f4d CM |
5415 | if (!path) |
5416 | return -ENOMEM; | |
5f26f772 | 5417 | |
3c12ac72 | 5418 | path->reada = 1; |
b9473439 | 5419 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5420 | |
5421 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
5422 | BUG_ON(!is_data && refs_to_drop != 1); | |
5423 | ||
3173a18f JB |
5424 | if (is_data) |
5425 | skinny_metadata = 0; | |
5426 | ||
5d4f98a2 YZ |
5427 | ret = lookup_extent_backref(trans, extent_root, path, &iref, |
5428 | bytenr, num_bytes, parent, | |
5429 | root_objectid, owner_objectid, | |
5430 | owner_offset); | |
7bb86316 | 5431 | if (ret == 0) { |
952fccac | 5432 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
5433 | while (extent_slot >= 0) { |
5434 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 5435 | extent_slot); |
5d4f98a2 | 5436 | if (key.objectid != bytenr) |
952fccac | 5437 | break; |
5d4f98a2 YZ |
5438 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
5439 | key.offset == num_bytes) { | |
952fccac CM |
5440 | found_extent = 1; |
5441 | break; | |
5442 | } | |
3173a18f JB |
5443 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
5444 | key.offset == owner_objectid) { | |
5445 | found_extent = 1; | |
5446 | break; | |
5447 | } | |
952fccac CM |
5448 | if (path->slots[0] - extent_slot > 5) |
5449 | break; | |
5d4f98a2 | 5450 | extent_slot--; |
952fccac | 5451 | } |
5d4f98a2 YZ |
5452 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
5453 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
5454 | if (found_extent && item_size < sizeof(*ei)) | |
5455 | found_extent = 0; | |
5456 | #endif | |
31840ae1 | 5457 | if (!found_extent) { |
5d4f98a2 | 5458 | BUG_ON(iref); |
56bec294 | 5459 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
5460 | NULL, refs_to_drop, |
5461 | is_data); | |
005d6427 DS |
5462 | if (ret) { |
5463 | btrfs_abort_transaction(trans, extent_root, ret); | |
5464 | goto out; | |
5465 | } | |
b3b4aa74 | 5466 | btrfs_release_path(path); |
b9473439 | 5467 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5468 | |
5469 | key.objectid = bytenr; | |
5470 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5471 | key.offset = num_bytes; | |
5472 | ||
3173a18f JB |
5473 | if (!is_data && skinny_metadata) { |
5474 | key.type = BTRFS_METADATA_ITEM_KEY; | |
5475 | key.offset = owner_objectid; | |
5476 | } | |
5477 | ||
31840ae1 ZY |
5478 | ret = btrfs_search_slot(trans, extent_root, |
5479 | &key, path, -1, 1); | |
3173a18f JB |
5480 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
5481 | /* | |
5482 | * Couldn't find our skinny metadata item, | |
5483 | * see if we have ye olde extent item. | |
5484 | */ | |
5485 | path->slots[0]--; | |
5486 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
5487 | path->slots[0]); | |
5488 | if (key.objectid == bytenr && | |
5489 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
5490 | key.offset == num_bytes) | |
5491 | ret = 0; | |
5492 | } | |
5493 | ||
5494 | if (ret > 0 && skinny_metadata) { | |
5495 | skinny_metadata = false; | |
5496 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5497 | key.offset = num_bytes; | |
5498 | btrfs_release_path(path); | |
5499 | ret = btrfs_search_slot(trans, extent_root, | |
5500 | &key, path, -1, 1); | |
5501 | } | |
5502 | ||
f3465ca4 JB |
5503 | if (ret) { |
5504 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
5505 | ", was looking for %llu\n", ret, |
5506 | (unsigned long long)bytenr); | |
b783e62d JB |
5507 | if (ret > 0) |
5508 | btrfs_print_leaf(extent_root, | |
5509 | path->nodes[0]); | |
f3465ca4 | 5510 | } |
005d6427 DS |
5511 | if (ret < 0) { |
5512 | btrfs_abort_transaction(trans, extent_root, ret); | |
5513 | goto out; | |
5514 | } | |
31840ae1 ZY |
5515 | extent_slot = path->slots[0]; |
5516 | } | |
79787eaa | 5517 | } else if (ret == -ENOENT) { |
7bb86316 CM |
5518 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5519 | WARN_ON(1); | |
d397712b | 5520 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 5521 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 5522 | (unsigned long long)bytenr, |
56bec294 | 5523 | (unsigned long long)parent, |
d397712b | 5524 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
5525 | (unsigned long long)owner_objectid, |
5526 | (unsigned long long)owner_offset); | |
79787eaa | 5527 | } else { |
005d6427 DS |
5528 | btrfs_abort_transaction(trans, extent_root, ret); |
5529 | goto out; | |
7bb86316 | 5530 | } |
5f39d397 CM |
5531 | |
5532 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
5533 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
5534 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
5535 | if (item_size < sizeof(*ei)) { | |
5536 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
5537 | ret = convert_extent_item_v0(trans, extent_root, path, | |
5538 | owner_objectid, 0); | |
005d6427 DS |
5539 | if (ret < 0) { |
5540 | btrfs_abort_transaction(trans, extent_root, ret); | |
5541 | goto out; | |
5542 | } | |
5d4f98a2 | 5543 | |
b3b4aa74 | 5544 | btrfs_release_path(path); |
5d4f98a2 YZ |
5545 | path->leave_spinning = 1; |
5546 | ||
5547 | key.objectid = bytenr; | |
5548 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5549 | key.offset = num_bytes; | |
5550 | ||
5551 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
5552 | -1, 1); | |
5553 | if (ret) { | |
5554 | printk(KERN_ERR "umm, got %d back from search" | |
5555 | ", was looking for %llu\n", ret, | |
5556 | (unsigned long long)bytenr); | |
5557 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
5558 | } | |
005d6427 DS |
5559 | if (ret < 0) { |
5560 | btrfs_abort_transaction(trans, extent_root, ret); | |
5561 | goto out; | |
5562 | } | |
5563 | ||
5d4f98a2 YZ |
5564 | extent_slot = path->slots[0]; |
5565 | leaf = path->nodes[0]; | |
5566 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
5567 | } | |
5568 | #endif | |
5569 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 5570 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 5571 | struct btrfs_extent_item); |
3173a18f JB |
5572 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
5573 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
5574 | struct btrfs_tree_block_info *bi; |
5575 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
5576 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
5577 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
5578 | } | |
56bec294 | 5579 | |
5d4f98a2 | 5580 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
5581 | BUG_ON(refs < refs_to_drop); |
5582 | refs -= refs_to_drop; | |
5f39d397 | 5583 | |
5d4f98a2 YZ |
5584 | if (refs > 0) { |
5585 | if (extent_op) | |
5586 | __run_delayed_extent_op(extent_op, leaf, ei); | |
5587 | /* | |
5588 | * In the case of inline back ref, reference count will | |
5589 | * be updated by remove_extent_backref | |
952fccac | 5590 | */ |
5d4f98a2 YZ |
5591 | if (iref) { |
5592 | BUG_ON(!found_extent); | |
5593 | } else { | |
5594 | btrfs_set_extent_refs(leaf, ei, refs); | |
5595 | btrfs_mark_buffer_dirty(leaf); | |
5596 | } | |
5597 | if (found_extent) { | |
5598 | ret = remove_extent_backref(trans, extent_root, path, | |
5599 | iref, refs_to_drop, | |
5600 | is_data); | |
005d6427 DS |
5601 | if (ret) { |
5602 | btrfs_abort_transaction(trans, extent_root, ret); | |
5603 | goto out; | |
5604 | } | |
952fccac | 5605 | } |
5d4f98a2 | 5606 | } else { |
5d4f98a2 YZ |
5607 | if (found_extent) { |
5608 | BUG_ON(is_data && refs_to_drop != | |
5609 | extent_data_ref_count(root, path, iref)); | |
5610 | if (iref) { | |
5611 | BUG_ON(path->slots[0] != extent_slot); | |
5612 | } else { | |
5613 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5614 | path->slots[0] = extent_slot; | |
5615 | num_to_del = 2; | |
5616 | } | |
78fae27e | 5617 | } |
b9473439 | 5618 | |
952fccac CM |
5619 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5620 | num_to_del); | |
005d6427 DS |
5621 | if (ret) { |
5622 | btrfs_abort_transaction(trans, extent_root, ret); | |
5623 | goto out; | |
5624 | } | |
b3b4aa74 | 5625 | btrfs_release_path(path); |
21af804c | 5626 | |
5d4f98a2 | 5627 | if (is_data) { |
459931ec | 5628 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 DS |
5629 | if (ret) { |
5630 | btrfs_abort_transaction(trans, extent_root, ret); | |
5631 | goto out; | |
5632 | } | |
459931ec CM |
5633 | } |
5634 | ||
c53d613e | 5635 | ret = update_block_group(root, bytenr, num_bytes, 0); |
005d6427 DS |
5636 | if (ret) { |
5637 | btrfs_abort_transaction(trans, extent_root, ret); | |
5638 | goto out; | |
5639 | } | |
a28ec197 | 5640 | } |
79787eaa | 5641 | out: |
5caf2a00 | 5642 | btrfs_free_path(path); |
a28ec197 CM |
5643 | return ret; |
5644 | } | |
5645 | ||
1887be66 | 5646 | /* |
f0486c68 | 5647 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5648 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5649 | * a given extent, and if there are no other delayed refs to be processed, it | |
5650 | * removes it from the tree. | |
5651 | */ | |
5652 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5653 | struct btrfs_root *root, u64 bytenr) | |
5654 | { | |
5655 | struct btrfs_delayed_ref_head *head; | |
5656 | struct btrfs_delayed_ref_root *delayed_refs; | |
5657 | struct btrfs_delayed_ref_node *ref; | |
5658 | struct rb_node *node; | |
f0486c68 | 5659 | int ret = 0; |
1887be66 CM |
5660 | |
5661 | delayed_refs = &trans->transaction->delayed_refs; | |
5662 | spin_lock(&delayed_refs->lock); | |
5663 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5664 | if (!head) | |
5665 | goto out; | |
5666 | ||
5667 | node = rb_prev(&head->node.rb_node); | |
5668 | if (!node) | |
5669 | goto out; | |
5670 | ||
5671 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5672 | ||
5673 | /* there are still entries for this ref, we can't drop it */ | |
5674 | if (ref->bytenr == bytenr) | |
5675 | goto out; | |
5676 | ||
5d4f98a2 YZ |
5677 | if (head->extent_op) { |
5678 | if (!head->must_insert_reserved) | |
5679 | goto out; | |
78a6184a | 5680 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
5681 | head->extent_op = NULL; |
5682 | } | |
5683 | ||
1887be66 CM |
5684 | /* |
5685 | * waiting for the lock here would deadlock. If someone else has it | |
5686 | * locked they are already in the process of dropping it anyway | |
5687 | */ | |
5688 | if (!mutex_trylock(&head->mutex)) | |
5689 | goto out; | |
5690 | ||
5691 | /* | |
5692 | * at this point we have a head with no other entries. Go | |
5693 | * ahead and process it. | |
5694 | */ | |
5695 | head->node.in_tree = 0; | |
5696 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5697 | |
1887be66 CM |
5698 | delayed_refs->num_entries--; |
5699 | ||
5700 | /* | |
5701 | * we don't take a ref on the node because we're removing it from the | |
5702 | * tree, so we just steal the ref the tree was holding. | |
5703 | */ | |
c3e69d58 CM |
5704 | delayed_refs->num_heads--; |
5705 | if (list_empty(&head->cluster)) | |
5706 | delayed_refs->num_heads_ready--; | |
5707 | ||
5708 | list_del_init(&head->cluster); | |
1887be66 CM |
5709 | spin_unlock(&delayed_refs->lock); |
5710 | ||
f0486c68 YZ |
5711 | BUG_ON(head->extent_op); |
5712 | if (head->must_insert_reserved) | |
5713 | ret = 1; | |
5714 | ||
5715 | mutex_unlock(&head->mutex); | |
1887be66 | 5716 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5717 | return ret; |
1887be66 CM |
5718 | out: |
5719 | spin_unlock(&delayed_refs->lock); | |
5720 | return 0; | |
5721 | } | |
5722 | ||
f0486c68 YZ |
5723 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5724 | struct btrfs_root *root, | |
5725 | struct extent_buffer *buf, | |
5581a51a | 5726 | u64 parent, int last_ref) |
f0486c68 | 5727 | { |
f0486c68 YZ |
5728 | struct btrfs_block_group_cache *cache = NULL; |
5729 | int ret; | |
5730 | ||
5731 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5732 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5733 | buf->start, buf->len, | |
5734 | parent, root->root_key.objectid, | |
5735 | btrfs_header_level(buf), | |
5581a51a | 5736 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
79787eaa | 5737 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
5738 | } |
5739 | ||
5740 | if (!last_ref) | |
5741 | return; | |
5742 | ||
f0486c68 | 5743 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5744 | |
5745 | if (btrfs_header_generation(buf) == trans->transid) { | |
5746 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5747 | ret = check_ref_cleanup(trans, root, buf->start); | |
5748 | if (!ret) | |
37be25bc | 5749 | goto out; |
f0486c68 YZ |
5750 | } |
5751 | ||
5752 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5753 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5754 | goto out; |
f0486c68 YZ |
5755 | } |
5756 | ||
5757 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5758 | ||
5759 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5760 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
f0486c68 YZ |
5761 | } |
5762 | out: | |
a826d6dc JB |
5763 | /* |
5764 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
5765 | * anymore. | |
5766 | */ | |
5767 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
5768 | btrfs_put_block_group(cache); |
5769 | } | |
5770 | ||
79787eaa | 5771 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
5772 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5773 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
5774 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
5775 | { |
5776 | int ret; | |
66d7e7f0 | 5777 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 5778 | |
56bec294 CM |
5779 | /* |
5780 | * tree log blocks never actually go into the extent allocation | |
5781 | * tree, just update pinning info and exit early. | |
56bec294 | 5782 | */ |
5d4f98a2 YZ |
5783 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
5784 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 5785 | /* unlocks the pinned mutex */ |
11833d66 | 5786 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 5787 | ret = 0; |
5d4f98a2 | 5788 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
5789 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
5790 | num_bytes, | |
5d4f98a2 | 5791 | parent, root_objectid, (int)owner, |
66d7e7f0 | 5792 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 5793 | } else { |
66d7e7f0 AJ |
5794 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5795 | num_bytes, | |
5796 | parent, root_objectid, owner, | |
5797 | offset, BTRFS_DROP_DELAYED_REF, | |
5798 | NULL, for_cow); | |
56bec294 | 5799 | } |
925baedd CM |
5800 | return ret; |
5801 | } | |
5802 | ||
53b381b3 DW |
5803 | static u64 stripe_align(struct btrfs_root *root, |
5804 | struct btrfs_block_group_cache *cache, | |
5805 | u64 val, u64 num_bytes) | |
87ee04eb | 5806 | { |
fda2832f | 5807 | u64 ret = ALIGN(val, root->stripesize); |
87ee04eb CM |
5808 | return ret; |
5809 | } | |
5810 | ||
817d52f8 JB |
5811 | /* |
5812 | * when we wait for progress in the block group caching, its because | |
5813 | * our allocation attempt failed at least once. So, we must sleep | |
5814 | * and let some progress happen before we try again. | |
5815 | * | |
5816 | * This function will sleep at least once waiting for new free space to | |
5817 | * show up, and then it will check the block group free space numbers | |
5818 | * for our min num_bytes. Another option is to have it go ahead | |
5819 | * and look in the rbtree for a free extent of a given size, but this | |
5820 | * is a good start. | |
5821 | */ | |
5822 | static noinline int | |
5823 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
5824 | u64 num_bytes) | |
5825 | { | |
11833d66 | 5826 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 5827 | |
11833d66 YZ |
5828 | caching_ctl = get_caching_control(cache); |
5829 | if (!caching_ctl) | |
817d52f8 | 5830 | return 0; |
817d52f8 | 5831 | |
11833d66 | 5832 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5833 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5834 | |
5835 | put_caching_control(caching_ctl); | |
5836 | return 0; | |
5837 | } | |
5838 | ||
5839 | static noinline int | |
5840 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5841 | { | |
5842 | struct btrfs_caching_control *caching_ctl; | |
11833d66 YZ |
5843 | |
5844 | caching_ctl = get_caching_control(cache); | |
5845 | if (!caching_ctl) | |
5846 | return 0; | |
5847 | ||
5848 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
5849 | ||
5850 | put_caching_control(caching_ctl); | |
817d52f8 JB |
5851 | return 0; |
5852 | } | |
5853 | ||
31e50229 | 5854 | int __get_raid_index(u64 flags) |
b742bb82 | 5855 | { |
7738a53a | 5856 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 5857 | return BTRFS_RAID_RAID10; |
7738a53a | 5858 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 5859 | return BTRFS_RAID_RAID1; |
7738a53a | 5860 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 5861 | return BTRFS_RAID_DUP; |
7738a53a | 5862 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 5863 | return BTRFS_RAID_RAID0; |
53b381b3 | 5864 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 5865 | return BTRFS_RAID_RAID5; |
53b381b3 | 5866 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 5867 | return BTRFS_RAID_RAID6; |
7738a53a | 5868 | |
e942f883 | 5869 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
5870 | } |
5871 | ||
7738a53a ID |
5872 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
5873 | { | |
31e50229 | 5874 | return __get_raid_index(cache->flags); |
7738a53a ID |
5875 | } |
5876 | ||
817d52f8 | 5877 | enum btrfs_loop_type { |
285ff5af JB |
5878 | LOOP_CACHING_NOWAIT = 0, |
5879 | LOOP_CACHING_WAIT = 1, | |
5880 | LOOP_ALLOC_CHUNK = 2, | |
5881 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
5882 | }; |
5883 | ||
fec577fb CM |
5884 | /* |
5885 | * walks the btree of allocated extents and find a hole of a given size. | |
5886 | * The key ins is changed to record the hole: | |
5887 | * ins->objectid == block start | |
62e2749e | 5888 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
5889 | * ins->offset == number of blocks |
5890 | * Any available blocks before search_start are skipped. | |
5891 | */ | |
d397712b | 5892 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
5893 | struct btrfs_root *orig_root, |
5894 | u64 num_bytes, u64 empty_size, | |
98ed5174 | 5895 | u64 hint_byte, struct btrfs_key *ins, |
e0f54067 | 5896 | u64 data) |
fec577fb | 5897 | { |
80eb234a | 5898 | int ret = 0; |
d397712b | 5899 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 5900 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 5901 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 5902 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 5903 | u64 search_start = 0; |
239b14b3 | 5904 | int empty_cluster = 2 * 1024 * 1024; |
80eb234a | 5905 | struct btrfs_space_info *space_info; |
fa9c0d79 | 5906 | int loop = 0; |
ac5c9300 | 5907 | int index = __get_raid_index(data); |
fb25e914 JB |
5908 | int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ? |
5909 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
817d52f8 | 5910 | bool found_uncached_bg = false; |
0a24325e | 5911 | bool failed_cluster_refill = false; |
1cdda9b8 | 5912 | bool failed_alloc = false; |
67377734 | 5913 | bool use_cluster = true; |
60d2adbb | 5914 | bool have_caching_bg = false; |
fec577fb | 5915 | |
db94535d | 5916 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 5917 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
5918 | ins->objectid = 0; |
5919 | ins->offset = 0; | |
b1a4d965 | 5920 | |
3f7de037 JB |
5921 | trace_find_free_extent(orig_root, num_bytes, empty_size, data); |
5922 | ||
2552d17e | 5923 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 5924 | if (!space_info) { |
e0f54067 | 5925 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
5926 | return -ENOSPC; |
5927 | } | |
2552d17e | 5928 | |
67377734 JB |
5929 | /* |
5930 | * If the space info is for both data and metadata it means we have a | |
5931 | * small filesystem and we can't use the clustering stuff. | |
5932 | */ | |
5933 | if (btrfs_mixed_space_info(space_info)) | |
5934 | use_cluster = false; | |
5935 | ||
67377734 | 5936 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5937 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5938 | if (!btrfs_test_opt(root, SSD)) |
5939 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5940 | } |
5941 | ||
67377734 JB |
5942 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5943 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5944 | last_ptr = &root->fs_info->data_alloc_cluster; |
5945 | } | |
0f9dd46c | 5946 | |
239b14b3 | 5947 | if (last_ptr) { |
fa9c0d79 CM |
5948 | spin_lock(&last_ptr->lock); |
5949 | if (last_ptr->block_group) | |
5950 | hint_byte = last_ptr->window_start; | |
5951 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5952 | } |
fa9c0d79 | 5953 | |
a061fc8d | 5954 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5955 | search_start = max(search_start, hint_byte); |
0b86a832 | 5956 | |
817d52f8 | 5957 | if (!last_ptr) |
fa9c0d79 | 5958 | empty_cluster = 0; |
fa9c0d79 | 5959 | |
2552d17e | 5960 | if (search_start == hint_byte) { |
2552d17e JB |
5961 | block_group = btrfs_lookup_block_group(root->fs_info, |
5962 | search_start); | |
274bd4fb | 5963 | used_block_group = block_group; |
817d52f8 JB |
5964 | /* |
5965 | * we don't want to use the block group if it doesn't match our | |
5966 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5967 | * |
5968 | * However if we are re-searching with an ideal block group | |
5969 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5970 | */ |
5971 | if (block_group && block_group_bits(block_group, data) && | |
285ff5af | 5972 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 5973 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5974 | if (list_empty(&block_group->list) || |
5975 | block_group->ro) { | |
5976 | /* | |
5977 | * someone is removing this block group, | |
5978 | * we can't jump into the have_block_group | |
5979 | * target because our list pointers are not | |
5980 | * valid | |
5981 | */ | |
5982 | btrfs_put_block_group(block_group); | |
5983 | up_read(&space_info->groups_sem); | |
ccf0e725 | 5984 | } else { |
b742bb82 | 5985 | index = get_block_group_index(block_group); |
44fb5511 | 5986 | goto have_block_group; |
ccf0e725 | 5987 | } |
2552d17e | 5988 | } else if (block_group) { |
fa9c0d79 | 5989 | btrfs_put_block_group(block_group); |
2552d17e | 5990 | } |
42e70e7a | 5991 | } |
2552d17e | 5992 | search: |
60d2adbb | 5993 | have_caching_bg = false; |
80eb234a | 5994 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
5995 | list_for_each_entry(block_group, &space_info->block_groups[index], |
5996 | list) { | |
6226cb0a | 5997 | u64 offset; |
817d52f8 | 5998 | int cached; |
8a1413a2 | 5999 | |
274bd4fb | 6000 | used_block_group = block_group; |
11dfe35a | 6001 | btrfs_get_block_group(block_group); |
2552d17e | 6002 | search_start = block_group->key.objectid; |
42e70e7a | 6003 | |
83a50de9 CM |
6004 | /* |
6005 | * this can happen if we end up cycling through all the | |
6006 | * raid types, but we want to make sure we only allocate | |
6007 | * for the proper type. | |
6008 | */ | |
6009 | if (!block_group_bits(block_group, data)) { | |
6010 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
6011 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
6012 | BTRFS_BLOCK_GROUP_RAID5 | |
6013 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
6014 | BTRFS_BLOCK_GROUP_RAID10; |
6015 | ||
6016 | /* | |
6017 | * if they asked for extra copies and this block group | |
6018 | * doesn't provide them, bail. This does allow us to | |
6019 | * fill raid0 from raid1. | |
6020 | */ | |
6021 | if ((data & extra) && !(block_group->flags & extra)) | |
6022 | goto loop; | |
6023 | } | |
6024 | ||
2552d17e | 6025 | have_block_group: |
291c7d2f JB |
6026 | cached = block_group_cache_done(block_group); |
6027 | if (unlikely(!cached)) { | |
291c7d2f | 6028 | found_uncached_bg = true; |
f6373bf3 | 6029 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
6030 | BUG_ON(ret < 0); |
6031 | ret = 0; | |
817d52f8 JB |
6032 | } |
6033 | ||
ea6a478e | 6034 | if (unlikely(block_group->ro)) |
2552d17e | 6035 | goto loop; |
0f9dd46c | 6036 | |
0a24325e | 6037 | /* |
062c05c4 AO |
6038 | * Ok we want to try and use the cluster allocator, so |
6039 | * lets look there | |
0a24325e | 6040 | */ |
062c05c4 | 6041 | if (last_ptr) { |
8de972b4 | 6042 | unsigned long aligned_cluster; |
fa9c0d79 CM |
6043 | /* |
6044 | * the refill lock keeps out other | |
6045 | * people trying to start a new cluster | |
6046 | */ | |
6047 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
6048 | used_block_group = last_ptr->block_group; |
6049 | if (used_block_group != block_group && | |
6050 | (!used_block_group || | |
6051 | used_block_group->ro || | |
6052 | !block_group_bits(used_block_group, data))) { | |
6053 | used_block_group = block_group; | |
44fb5511 | 6054 | goto refill_cluster; |
274bd4fb AO |
6055 | } |
6056 | ||
6057 | if (used_block_group != block_group) | |
6058 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 6059 | |
274bd4fb AO |
6060 | offset = btrfs_alloc_from_cluster(used_block_group, |
6061 | last_ptr, num_bytes, used_block_group->key.objectid); | |
fa9c0d79 CM |
6062 | if (offset) { |
6063 | /* we have a block, we're done */ | |
6064 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
6065 | trace_btrfs_reserve_extent_cluster(root, |
6066 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
6067 | goto checks; |
6068 | } | |
6069 | ||
274bd4fb AO |
6070 | WARN_ON(last_ptr->block_group != used_block_group); |
6071 | if (used_block_group != block_group) { | |
6072 | btrfs_put_block_group(used_block_group); | |
6073 | used_block_group = block_group; | |
fa9c0d79 | 6074 | } |
44fb5511 | 6075 | refill_cluster: |
274bd4fb | 6076 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
6077 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
6078 | * set up a new clusters, so lets just skip it | |
6079 | * and let the allocator find whatever block | |
6080 | * it can find. If we reach this point, we | |
6081 | * will have tried the cluster allocator | |
6082 | * plenty of times and not have found | |
6083 | * anything, so we are likely way too | |
6084 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
6085 | * anything. |
6086 | * | |
6087 | * However, if the cluster is taken from the | |
6088 | * current block group, release the cluster | |
6089 | * first, so that we stand a better chance of | |
6090 | * succeeding in the unclustered | |
6091 | * allocation. */ | |
6092 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
6093 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
6094 | spin_unlock(&last_ptr->refill_lock); |
6095 | goto unclustered_alloc; | |
6096 | } | |
6097 | ||
fa9c0d79 CM |
6098 | /* |
6099 | * this cluster didn't work out, free it and | |
6100 | * start over | |
6101 | */ | |
6102 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
6103 | ||
a5f6f719 AO |
6104 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
6105 | spin_unlock(&last_ptr->refill_lock); | |
6106 | goto unclustered_alloc; | |
6107 | } | |
6108 | ||
8de972b4 CM |
6109 | aligned_cluster = max_t(unsigned long, |
6110 | empty_cluster + empty_size, | |
6111 | block_group->full_stripe_len); | |
6112 | ||
fa9c0d79 | 6113 | /* allocate a cluster in this block group */ |
451d7585 | 6114 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 6115 | block_group, last_ptr, |
1b22bad7 | 6116 | search_start, num_bytes, |
8de972b4 | 6117 | aligned_cluster); |
fa9c0d79 CM |
6118 | if (ret == 0) { |
6119 | /* | |
6120 | * now pull our allocation out of this | |
6121 | * cluster | |
6122 | */ | |
6123 | offset = btrfs_alloc_from_cluster(block_group, | |
6124 | last_ptr, num_bytes, | |
6125 | search_start); | |
6126 | if (offset) { | |
6127 | /* we found one, proceed */ | |
6128 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
6129 | trace_btrfs_reserve_extent_cluster(root, |
6130 | block_group, search_start, | |
6131 | num_bytes); | |
fa9c0d79 CM |
6132 | goto checks; |
6133 | } | |
0a24325e JB |
6134 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
6135 | && !failed_cluster_refill) { | |
817d52f8 JB |
6136 | spin_unlock(&last_ptr->refill_lock); |
6137 | ||
0a24325e | 6138 | failed_cluster_refill = true; |
817d52f8 JB |
6139 | wait_block_group_cache_progress(block_group, |
6140 | num_bytes + empty_cluster + empty_size); | |
6141 | goto have_block_group; | |
fa9c0d79 | 6142 | } |
817d52f8 | 6143 | |
fa9c0d79 CM |
6144 | /* |
6145 | * at this point we either didn't find a cluster | |
6146 | * or we weren't able to allocate a block from our | |
6147 | * cluster. Free the cluster we've been trying | |
6148 | * to use, and go to the next block group | |
6149 | */ | |
0a24325e | 6150 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 6151 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 6152 | goto loop; |
fa9c0d79 CM |
6153 | } |
6154 | ||
062c05c4 | 6155 | unclustered_alloc: |
a5f6f719 AO |
6156 | spin_lock(&block_group->free_space_ctl->tree_lock); |
6157 | if (cached && | |
6158 | block_group->free_space_ctl->free_space < | |
6159 | num_bytes + empty_cluster + empty_size) { | |
6160 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
6161 | goto loop; | |
6162 | } | |
6163 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
6164 | ||
6226cb0a JB |
6165 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
6166 | num_bytes, empty_size); | |
1cdda9b8 JB |
6167 | /* |
6168 | * If we didn't find a chunk, and we haven't failed on this | |
6169 | * block group before, and this block group is in the middle of | |
6170 | * caching and we are ok with waiting, then go ahead and wait | |
6171 | * for progress to be made, and set failed_alloc to true. | |
6172 | * | |
6173 | * If failed_alloc is true then we've already waited on this | |
6174 | * block group once and should move on to the next block group. | |
6175 | */ | |
6176 | if (!offset && !failed_alloc && !cached && | |
6177 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 6178 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
6179 | num_bytes + empty_size); |
6180 | failed_alloc = true; | |
817d52f8 | 6181 | goto have_block_group; |
1cdda9b8 | 6182 | } else if (!offset) { |
60d2adbb MX |
6183 | if (!cached) |
6184 | have_caching_bg = true; | |
1cdda9b8 | 6185 | goto loop; |
817d52f8 | 6186 | } |
fa9c0d79 | 6187 | checks: |
53b381b3 DW |
6188 | search_start = stripe_align(root, used_block_group, |
6189 | offset, num_bytes); | |
25179201 | 6190 | |
2552d17e JB |
6191 | /* move on to the next group */ |
6192 | if (search_start + num_bytes > | |
274bd4fb AO |
6193 | used_block_group->key.objectid + used_block_group->key.offset) { |
6194 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 6195 | goto loop; |
6226cb0a | 6196 | } |
f5a31e16 | 6197 | |
f0486c68 | 6198 | if (offset < search_start) |
274bd4fb | 6199 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
6200 | search_start - offset); |
6201 | BUG_ON(offset > search_start); | |
2552d17e | 6202 | |
274bd4fb | 6203 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 6204 | alloc_type); |
f0486c68 | 6205 | if (ret == -EAGAIN) { |
274bd4fb | 6206 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 6207 | goto loop; |
0f9dd46c | 6208 | } |
0b86a832 | 6209 | |
f0486c68 | 6210 | /* we are all good, lets return */ |
2552d17e JB |
6211 | ins->objectid = search_start; |
6212 | ins->offset = num_bytes; | |
d2fb3437 | 6213 | |
3f7de037 JB |
6214 | trace_btrfs_reserve_extent(orig_root, block_group, |
6215 | search_start, num_bytes); | |
274bd4fb AO |
6216 | if (used_block_group != block_group) |
6217 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 6218 | btrfs_put_block_group(block_group); |
2552d17e JB |
6219 | break; |
6220 | loop: | |
0a24325e | 6221 | failed_cluster_refill = false; |
1cdda9b8 | 6222 | failed_alloc = false; |
b742bb82 | 6223 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
6224 | if (used_block_group != block_group) |
6225 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 6226 | btrfs_put_block_group(block_group); |
2552d17e JB |
6227 | } |
6228 | up_read(&space_info->groups_sem); | |
6229 | ||
60d2adbb MX |
6230 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
6231 | goto search; | |
6232 | ||
b742bb82 YZ |
6233 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
6234 | goto search; | |
6235 | ||
285ff5af | 6236 | /* |
ccf0e725 JB |
6237 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
6238 | * caching kthreads as we move along | |
817d52f8 JB |
6239 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
6240 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
6241 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
6242 | * again | |
fa9c0d79 | 6243 | */ |
723bda20 | 6244 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 6245 | index = 0; |
723bda20 | 6246 | loop++; |
817d52f8 | 6247 | if (loop == LOOP_ALLOC_CHUNK) { |
698d0082 | 6248 | ret = do_chunk_alloc(trans, root, data, |
ea658bad JB |
6249 | CHUNK_ALLOC_FORCE); |
6250 | /* | |
6251 | * Do not bail out on ENOSPC since we | |
6252 | * can do more things. | |
6253 | */ | |
6254 | if (ret < 0 && ret != -ENOSPC) { | |
6255 | btrfs_abort_transaction(trans, | |
6256 | root, ret); | |
6257 | goto out; | |
723bda20 | 6258 | } |
2552d17e JB |
6259 | } |
6260 | ||
723bda20 JB |
6261 | if (loop == LOOP_NO_EMPTY_SIZE) { |
6262 | empty_size = 0; | |
6263 | empty_cluster = 0; | |
fa9c0d79 | 6264 | } |
723bda20 JB |
6265 | |
6266 | goto search; | |
2552d17e JB |
6267 | } else if (!ins->objectid) { |
6268 | ret = -ENOSPC; | |
d82a6f1d | 6269 | } else if (ins->objectid) { |
80eb234a | 6270 | ret = 0; |
be744175 | 6271 | } |
79787eaa | 6272 | out: |
be744175 | 6273 | |
0f70abe2 | 6274 | return ret; |
fec577fb | 6275 | } |
ec44a35c | 6276 | |
9ed74f2d JB |
6277 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
6278 | int dump_block_groups) | |
0f9dd46c JB |
6279 | { |
6280 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 6281 | int index = 0; |
0f9dd46c | 6282 | |
9ed74f2d | 6283 | spin_lock(&info->lock); |
fb25e914 JB |
6284 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
6285 | (unsigned long long)info->flags, | |
d397712b | 6286 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 6287 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 6288 | info->bytes_readonly), |
d397712b | 6289 | (info->full) ? "" : "not "); |
8929ecfa YZ |
6290 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
6291 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 6292 | (unsigned long long)info->total_bytes, |
8929ecfa | 6293 | (unsigned long long)info->bytes_used, |
21380931 | 6294 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 6295 | (unsigned long long)info->bytes_reserved, |
21380931 | 6296 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 6297 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
6298 | spin_unlock(&info->lock); |
6299 | ||
6300 | if (!dump_block_groups) | |
6301 | return; | |
0f9dd46c | 6302 | |
80eb234a | 6303 | down_read(&info->groups_sem); |
b742bb82 YZ |
6304 | again: |
6305 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 6306 | spin_lock(&cache->lock); |
799ffc3c | 6307 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n", |
d397712b CM |
6308 | (unsigned long long)cache->key.objectid, |
6309 | (unsigned long long)cache->key.offset, | |
6310 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
6311 | (unsigned long long)cache->pinned, | |
799ffc3c LB |
6312 | (unsigned long long)cache->reserved, |
6313 | cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
6314 | btrfs_dump_free_space(cache, bytes); |
6315 | spin_unlock(&cache->lock); | |
6316 | } | |
b742bb82 YZ |
6317 | if (++index < BTRFS_NR_RAID_TYPES) |
6318 | goto again; | |
80eb234a | 6319 | up_read(&info->groups_sem); |
0f9dd46c | 6320 | } |
e8569813 | 6321 | |
11833d66 YZ |
6322 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
6323 | struct btrfs_root *root, | |
6324 | u64 num_bytes, u64 min_alloc_size, | |
6325 | u64 empty_size, u64 hint_byte, | |
81c9ad23 | 6326 | struct btrfs_key *ins, u64 data) |
fec577fb | 6327 | { |
9e622d6b | 6328 | bool final_tried = false; |
fec577fb | 6329 | int ret; |
925baedd | 6330 | |
6a63209f | 6331 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 6332 | again: |
db94535d CM |
6333 | WARN_ON(num_bytes < root->sectorsize); |
6334 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
81c9ad23 | 6335 | hint_byte, ins, data); |
3b951516 | 6336 | |
9e622d6b MX |
6337 | if (ret == -ENOSPC) { |
6338 | if (!final_tried) { | |
6339 | num_bytes = num_bytes >> 1; | |
24542bf7 | 6340 | num_bytes = round_down(num_bytes, root->sectorsize); |
9e622d6b | 6341 | num_bytes = max(num_bytes, min_alloc_size); |
9e622d6b MX |
6342 | if (num_bytes == min_alloc_size) |
6343 | final_tried = true; | |
6344 | goto again; | |
6345 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
6346 | struct btrfs_space_info *sinfo; | |
6347 | ||
6348 | sinfo = __find_space_info(root->fs_info, data); | |
6349 | printk(KERN_ERR "btrfs allocation failed flags %llu, " | |
6350 | "wanted %llu\n", (unsigned long long)data, | |
6351 | (unsigned long long)num_bytes); | |
53804280 JM |
6352 | if (sinfo) |
6353 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 6354 | } |
925baedd | 6355 | } |
0f9dd46c | 6356 | |
1abe9b8a | 6357 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
6358 | ||
0f9dd46c | 6359 | return ret; |
e6dcd2dc CM |
6360 | } |
6361 | ||
e688b725 CM |
6362 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
6363 | u64 start, u64 len, int pin) | |
65b51a00 | 6364 | { |
0f9dd46c | 6365 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 6366 | int ret = 0; |
0f9dd46c | 6367 | |
0f9dd46c JB |
6368 | cache = btrfs_lookup_block_group(root->fs_info, start); |
6369 | if (!cache) { | |
d397712b CM |
6370 | printk(KERN_ERR "Unable to find block group for %llu\n", |
6371 | (unsigned long long)start); | |
0f9dd46c JB |
6372 | return -ENOSPC; |
6373 | } | |
1f3c79a2 | 6374 | |
5378e607 LD |
6375 | if (btrfs_test_opt(root, DISCARD)) |
6376 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 6377 | |
e688b725 CM |
6378 | if (pin) |
6379 | pin_down_extent(root, cache, start, len, 1); | |
6380 | else { | |
6381 | btrfs_add_free_space(cache, start, len); | |
6382 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
6383 | } | |
fa9c0d79 | 6384 | btrfs_put_block_group(cache); |
817d52f8 | 6385 | |
1abe9b8a | 6386 | trace_btrfs_reserved_extent_free(root, start, len); |
6387 | ||
e6dcd2dc CM |
6388 | return ret; |
6389 | } | |
6390 | ||
e688b725 CM |
6391 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
6392 | u64 start, u64 len) | |
6393 | { | |
6394 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
6395 | } | |
6396 | ||
6397 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
6398 | u64 start, u64 len) | |
6399 | { | |
6400 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
6401 | } | |
6402 | ||
5d4f98a2 YZ |
6403 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
6404 | struct btrfs_root *root, | |
6405 | u64 parent, u64 root_objectid, | |
6406 | u64 flags, u64 owner, u64 offset, | |
6407 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
6408 | { |
6409 | int ret; | |
5d4f98a2 | 6410 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 6411 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 6412 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 6413 | struct btrfs_path *path; |
5d4f98a2 YZ |
6414 | struct extent_buffer *leaf; |
6415 | int type; | |
6416 | u32 size; | |
26b8003f | 6417 | |
5d4f98a2 YZ |
6418 | if (parent > 0) |
6419 | type = BTRFS_SHARED_DATA_REF_KEY; | |
6420 | else | |
6421 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 6422 | |
5d4f98a2 | 6423 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
6424 | |
6425 | path = btrfs_alloc_path(); | |
db5b493a TI |
6426 | if (!path) |
6427 | return -ENOMEM; | |
47e4bb98 | 6428 | |
b9473439 | 6429 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6430 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
6431 | ins, size); | |
79787eaa JM |
6432 | if (ret) { |
6433 | btrfs_free_path(path); | |
6434 | return ret; | |
6435 | } | |
0f9dd46c | 6436 | |
5d4f98a2 YZ |
6437 | leaf = path->nodes[0]; |
6438 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 6439 | struct btrfs_extent_item); |
5d4f98a2 YZ |
6440 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
6441 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6442 | btrfs_set_extent_flags(leaf, extent_item, | |
6443 | flags | BTRFS_EXTENT_FLAG_DATA); | |
6444 | ||
6445 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6446 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
6447 | if (parent > 0) { | |
6448 | struct btrfs_shared_data_ref *ref; | |
6449 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
6450 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6451 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
6452 | } else { | |
6453 | struct btrfs_extent_data_ref *ref; | |
6454 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
6455 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
6456 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
6457 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
6458 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
6459 | } | |
47e4bb98 CM |
6460 | |
6461 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 6462 | btrfs_free_path(path); |
f510cfec | 6463 | |
c53d613e | 6464 | ret = update_block_group(root, ins->objectid, ins->offset, 1); |
79787eaa | 6465 | if (ret) { /* -ENOENT, logic error */ |
d397712b CM |
6466 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6467 | "%llu\n", (unsigned long long)ins->objectid, | |
6468 | (unsigned long long)ins->offset); | |
f5947066 CM |
6469 | BUG(); |
6470 | } | |
e6dcd2dc CM |
6471 | return ret; |
6472 | } | |
6473 | ||
5d4f98a2 YZ |
6474 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
6475 | struct btrfs_root *root, | |
6476 | u64 parent, u64 root_objectid, | |
6477 | u64 flags, struct btrfs_disk_key *key, | |
6478 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
6479 | { |
6480 | int ret; | |
5d4f98a2 YZ |
6481 | struct btrfs_fs_info *fs_info = root->fs_info; |
6482 | struct btrfs_extent_item *extent_item; | |
6483 | struct btrfs_tree_block_info *block_info; | |
6484 | struct btrfs_extent_inline_ref *iref; | |
6485 | struct btrfs_path *path; | |
6486 | struct extent_buffer *leaf; | |
3173a18f JB |
6487 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
6488 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, | |
6489 | SKINNY_METADATA); | |
6490 | ||
6491 | if (!skinny_metadata) | |
6492 | size += sizeof(*block_info); | |
1c2308f8 | 6493 | |
5d4f98a2 | 6494 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6495 | if (!path) |
6496 | return -ENOMEM; | |
56bec294 | 6497 | |
5d4f98a2 YZ |
6498 | path->leave_spinning = 1; |
6499 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
6500 | ins, size); | |
79787eaa JM |
6501 | if (ret) { |
6502 | btrfs_free_path(path); | |
6503 | return ret; | |
6504 | } | |
5d4f98a2 YZ |
6505 | |
6506 | leaf = path->nodes[0]; | |
6507 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
6508 | struct btrfs_extent_item); | |
6509 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
6510 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6511 | btrfs_set_extent_flags(leaf, extent_item, | |
6512 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 6513 | |
3173a18f JB |
6514 | if (skinny_metadata) { |
6515 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6516 | } else { | |
6517 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
6518 | btrfs_set_tree_block_key(leaf, block_info, key); | |
6519 | btrfs_set_tree_block_level(leaf, block_info, level); | |
6520 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
6521 | } | |
5d4f98a2 | 6522 | |
5d4f98a2 YZ |
6523 | if (parent > 0) { |
6524 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
6525 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6526 | BTRFS_SHARED_BLOCK_REF_KEY); | |
6527 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6528 | } else { | |
6529 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6530 | BTRFS_TREE_BLOCK_REF_KEY); | |
6531 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
6532 | } | |
6533 | ||
6534 | btrfs_mark_buffer_dirty(leaf); | |
6535 | btrfs_free_path(path); | |
6536 | ||
3173a18f | 6537 | ret = update_block_group(root, ins->objectid, root->leafsize, 1); |
79787eaa | 6538 | if (ret) { /* -ENOENT, logic error */ |
5d4f98a2 YZ |
6539 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6540 | "%llu\n", (unsigned long long)ins->objectid, | |
6541 | (unsigned long long)ins->offset); | |
6542 | BUG(); | |
6543 | } | |
6544 | return ret; | |
6545 | } | |
6546 | ||
6547 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
6548 | struct btrfs_root *root, | |
6549 | u64 root_objectid, u64 owner, | |
6550 | u64 offset, struct btrfs_key *ins) | |
6551 | { | |
6552 | int ret; | |
6553 | ||
6554 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
6555 | ||
66d7e7f0 AJ |
6556 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
6557 | ins->offset, 0, | |
6558 | root_objectid, owner, offset, | |
6559 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
6560 | return ret; |
6561 | } | |
e02119d5 CM |
6562 | |
6563 | /* | |
6564 | * this is used by the tree logging recovery code. It records that | |
6565 | * an extent has been allocated and makes sure to clear the free | |
6566 | * space cache bits as well | |
6567 | */ | |
5d4f98a2 YZ |
6568 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
6569 | struct btrfs_root *root, | |
6570 | u64 root_objectid, u64 owner, u64 offset, | |
6571 | struct btrfs_key *ins) | |
e02119d5 CM |
6572 | { |
6573 | int ret; | |
6574 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
6575 | struct btrfs_caching_control *caching_ctl; |
6576 | u64 start = ins->objectid; | |
6577 | u64 num_bytes = ins->offset; | |
e02119d5 | 6578 | |
e02119d5 | 6579 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
f6373bf3 | 6580 | cache_block_group(block_group, 0); |
11833d66 | 6581 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 6582 | |
11833d66 YZ |
6583 | if (!caching_ctl) { |
6584 | BUG_ON(!block_group_cache_done(block_group)); | |
6585 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
79787eaa | 6586 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6587 | } else { |
6588 | mutex_lock(&caching_ctl->mutex); | |
6589 | ||
6590 | if (start >= caching_ctl->progress) { | |
6591 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6592 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6593 | } else if (start + num_bytes <= caching_ctl->progress) { |
6594 | ret = btrfs_remove_free_space(block_group, | |
6595 | start, num_bytes); | |
79787eaa | 6596 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6597 | } else { |
6598 | num_bytes = caching_ctl->progress - start; | |
6599 | ret = btrfs_remove_free_space(block_group, | |
6600 | start, num_bytes); | |
79787eaa | 6601 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6602 | |
6603 | start = caching_ctl->progress; | |
6604 | num_bytes = ins->objectid + ins->offset - | |
6605 | caching_ctl->progress; | |
6606 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6607 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6608 | } |
6609 | ||
6610 | mutex_unlock(&caching_ctl->mutex); | |
6611 | put_caching_control(caching_ctl); | |
6612 | } | |
6613 | ||
fb25e914 JB |
6614 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
6615 | RESERVE_ALLOC_NO_ACCOUNT); | |
79787eaa | 6616 | BUG_ON(ret); /* logic error */ |
fa9c0d79 | 6617 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
6618 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
6619 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
6620 | return ret; |
6621 | } | |
6622 | ||
65b51a00 CM |
6623 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
6624 | struct btrfs_root *root, | |
4008c04a CM |
6625 | u64 bytenr, u32 blocksize, |
6626 | int level) | |
65b51a00 CM |
6627 | { |
6628 | struct extent_buffer *buf; | |
6629 | ||
6630 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6631 | if (!buf) | |
6632 | return ERR_PTR(-ENOMEM); | |
6633 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6634 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6635 | btrfs_tree_lock(buf); |
6636 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6637 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6638 | |
6639 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6640 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6641 | |
d0c803c4 | 6642 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6643 | /* |
6644 | * we allow two log transactions at a time, use different | |
6645 | * EXENT bit to differentiate dirty pages. | |
6646 | */ | |
6647 | if (root->log_transid % 2 == 0) | |
6648 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6649 | buf->start + buf->len - 1, GFP_NOFS); | |
6650 | else | |
6651 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6652 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6653 | } else { |
6654 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6655 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6656 | } |
65b51a00 | 6657 | trans->blocks_used++; |
b4ce94de | 6658 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6659 | return buf; |
6660 | } | |
6661 | ||
f0486c68 YZ |
6662 | static struct btrfs_block_rsv * |
6663 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6664 | struct btrfs_root *root, u32 blocksize) | |
6665 | { | |
6666 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6667 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
6668 | int ret; |
6669 | ||
6670 | block_rsv = get_block_rsv(trans, root); | |
6671 | ||
6672 | if (block_rsv->size == 0) { | |
08e007d2 MX |
6673 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, |
6674 | BTRFS_RESERVE_NO_FLUSH); | |
68a82277 JB |
6675 | /* |
6676 | * If we couldn't reserve metadata bytes try and use some from | |
6677 | * the global reserve. | |
6678 | */ | |
6679 | if (ret && block_rsv != global_rsv) { | |
6680 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6681 | if (!ret) | |
6682 | return global_rsv; | |
f0486c68 | 6683 | return ERR_PTR(ret); |
68a82277 | 6684 | } else if (ret) { |
f0486c68 | 6685 | return ERR_PTR(ret); |
68a82277 | 6686 | } |
f0486c68 YZ |
6687 | return block_rsv; |
6688 | } | |
6689 | ||
6690 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
6691 | if (!ret) | |
6692 | return block_rsv; | |
ca7e70f5 | 6693 | if (ret && !block_rsv->failfast) { |
b069e0c3 DS |
6694 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { |
6695 | static DEFINE_RATELIMIT_STATE(_rs, | |
6696 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
6697 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
6698 | if (__ratelimit(&_rs)) | |
6699 | WARN(1, KERN_DEBUG | |
6700 | "btrfs: block rsv returned %d\n", ret); | |
6701 | } | |
08e007d2 MX |
6702 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, |
6703 | BTRFS_RESERVE_NO_FLUSH); | |
68a82277 | 6704 | if (!ret) { |
68a82277 JB |
6705 | return block_rsv; |
6706 | } else if (ret && block_rsv != global_rsv) { | |
6707 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6708 | if (!ret) | |
6709 | return global_rsv; | |
6710 | } | |
6711 | } | |
f0486c68 | 6712 | |
f0486c68 YZ |
6713 | return ERR_PTR(-ENOSPC); |
6714 | } | |
6715 | ||
8c2a3ca2 JB |
6716 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6717 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6718 | { |
6719 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6720 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6721 | } |
6722 | ||
fec577fb | 6723 | /* |
f0486c68 YZ |
6724 | * finds a free extent and does all the dirty work required for allocation |
6725 | * returns the key for the extent through ins, and a tree buffer for | |
6726 | * the first block of the extent through buf. | |
6727 | * | |
fec577fb CM |
6728 | * returns the tree buffer or NULL. |
6729 | */ | |
5f39d397 | 6730 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6731 | struct btrfs_root *root, u32 blocksize, |
6732 | u64 parent, u64 root_objectid, | |
6733 | struct btrfs_disk_key *key, int level, | |
5581a51a | 6734 | u64 hint, u64 empty_size) |
fec577fb | 6735 | { |
e2fa7227 | 6736 | struct btrfs_key ins; |
f0486c68 | 6737 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6738 | struct extent_buffer *buf; |
f0486c68 YZ |
6739 | u64 flags = 0; |
6740 | int ret; | |
3173a18f JB |
6741 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
6742 | SKINNY_METADATA); | |
fec577fb | 6743 | |
f0486c68 YZ |
6744 | block_rsv = use_block_rsv(trans, root, blocksize); |
6745 | if (IS_ERR(block_rsv)) | |
6746 | return ERR_CAST(block_rsv); | |
6747 | ||
6748 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
81c9ad23 | 6749 | empty_size, hint, &ins, 0); |
fec577fb | 6750 | if (ret) { |
8c2a3ca2 | 6751 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6752 | return ERR_PTR(ret); |
fec577fb | 6753 | } |
55c69072 | 6754 | |
4008c04a CM |
6755 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6756 | blocksize, level); | |
79787eaa | 6757 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
f0486c68 YZ |
6758 | |
6759 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6760 | if (parent == 0) | |
6761 | parent = ins.objectid; | |
6762 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6763 | } else | |
6764 | BUG_ON(parent > 0); | |
6765 | ||
6766 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6767 | struct btrfs_delayed_extent_op *extent_op; | |
78a6184a | 6768 | extent_op = btrfs_alloc_delayed_extent_op(); |
79787eaa | 6769 | BUG_ON(!extent_op); /* -ENOMEM */ |
f0486c68 YZ |
6770 | if (key) |
6771 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6772 | else | |
6773 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6774 | extent_op->flags_to_set = flags; | |
3173a18f JB |
6775 | if (skinny_metadata) |
6776 | extent_op->update_key = 0; | |
6777 | else | |
6778 | extent_op->update_key = 1; | |
f0486c68 YZ |
6779 | extent_op->update_flags = 1; |
6780 | extent_op->is_data = 0; | |
6781 | ||
66d7e7f0 AJ |
6782 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6783 | ins.objectid, | |
f0486c68 YZ |
6784 | ins.offset, parent, root_objectid, |
6785 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5581a51a | 6786 | extent_op, 0); |
79787eaa | 6787 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 | 6788 | } |
fec577fb CM |
6789 | return buf; |
6790 | } | |
a28ec197 | 6791 | |
2c47e605 YZ |
6792 | struct walk_control { |
6793 | u64 refs[BTRFS_MAX_LEVEL]; | |
6794 | u64 flags[BTRFS_MAX_LEVEL]; | |
6795 | struct btrfs_key update_progress; | |
6796 | int stage; | |
6797 | int level; | |
6798 | int shared_level; | |
6799 | int update_ref; | |
6800 | int keep_locks; | |
1c4850e2 YZ |
6801 | int reada_slot; |
6802 | int reada_count; | |
66d7e7f0 | 6803 | int for_reloc; |
2c47e605 YZ |
6804 | }; |
6805 | ||
6806 | #define DROP_REFERENCE 1 | |
6807 | #define UPDATE_BACKREF 2 | |
6808 | ||
1c4850e2 YZ |
6809 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6810 | struct btrfs_root *root, | |
6811 | struct walk_control *wc, | |
6812 | struct btrfs_path *path) | |
6407bf6d | 6813 | { |
1c4850e2 YZ |
6814 | u64 bytenr; |
6815 | u64 generation; | |
6816 | u64 refs; | |
94fcca9f | 6817 | u64 flags; |
5d4f98a2 | 6818 | u32 nritems; |
1c4850e2 YZ |
6819 | u32 blocksize; |
6820 | struct btrfs_key key; | |
6821 | struct extent_buffer *eb; | |
6407bf6d | 6822 | int ret; |
1c4850e2 YZ |
6823 | int slot; |
6824 | int nread = 0; | |
6407bf6d | 6825 | |
1c4850e2 YZ |
6826 | if (path->slots[wc->level] < wc->reada_slot) { |
6827 | wc->reada_count = wc->reada_count * 2 / 3; | |
6828 | wc->reada_count = max(wc->reada_count, 2); | |
6829 | } else { | |
6830 | wc->reada_count = wc->reada_count * 3 / 2; | |
6831 | wc->reada_count = min_t(int, wc->reada_count, | |
6832 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
6833 | } | |
7bb86316 | 6834 | |
1c4850e2 YZ |
6835 | eb = path->nodes[wc->level]; |
6836 | nritems = btrfs_header_nritems(eb); | |
6837 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 6838 | |
1c4850e2 YZ |
6839 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6840 | if (nread >= wc->reada_count) | |
6841 | break; | |
bd56b302 | 6842 | |
2dd3e67b | 6843 | cond_resched(); |
1c4850e2 YZ |
6844 | bytenr = btrfs_node_blockptr(eb, slot); |
6845 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6846 | |
1c4850e2 YZ |
6847 | if (slot == path->slots[wc->level]) |
6848 | goto reada; | |
5d4f98a2 | 6849 | |
1c4850e2 YZ |
6850 | if (wc->stage == UPDATE_BACKREF && |
6851 | generation <= root->root_key.offset) | |
bd56b302 CM |
6852 | continue; |
6853 | ||
94fcca9f | 6854 | /* We don't lock the tree block, it's OK to be racy here */ |
3173a18f JB |
6855 | ret = btrfs_lookup_extent_info(trans, root, bytenr, |
6856 | wc->level - 1, 1, &refs, | |
6857 | &flags); | |
79787eaa JM |
6858 | /* We don't care about errors in readahead. */ |
6859 | if (ret < 0) | |
6860 | continue; | |
94fcca9f YZ |
6861 | BUG_ON(refs == 0); |
6862 | ||
1c4850e2 | 6863 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6864 | if (refs == 1) |
6865 | goto reada; | |
bd56b302 | 6866 | |
94fcca9f YZ |
6867 | if (wc->level == 1 && |
6868 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6869 | continue; | |
1c4850e2 YZ |
6870 | if (!wc->update_ref || |
6871 | generation <= root->root_key.offset) | |
6872 | continue; | |
6873 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6874 | ret = btrfs_comp_cpu_keys(&key, | |
6875 | &wc->update_progress); | |
6876 | if (ret < 0) | |
6877 | continue; | |
94fcca9f YZ |
6878 | } else { |
6879 | if (wc->level == 1 && | |
6880 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6881 | continue; | |
6407bf6d | 6882 | } |
1c4850e2 YZ |
6883 | reada: |
6884 | ret = readahead_tree_block(root, bytenr, blocksize, | |
6885 | generation); | |
6886 | if (ret) | |
bd56b302 | 6887 | break; |
1c4850e2 | 6888 | nread++; |
20524f02 | 6889 | } |
1c4850e2 | 6890 | wc->reada_slot = slot; |
20524f02 | 6891 | } |
2c47e605 | 6892 | |
f82d02d9 | 6893 | /* |
2c016dc2 | 6894 | * helper to process tree block while walking down the tree. |
2c47e605 | 6895 | * |
2c47e605 YZ |
6896 | * when wc->stage == UPDATE_BACKREF, this function updates |
6897 | * back refs for pointers in the block. | |
6898 | * | |
6899 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 6900 | */ |
2c47e605 | 6901 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 6902 | struct btrfs_root *root, |
2c47e605 | 6903 | struct btrfs_path *path, |
94fcca9f | 6904 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 6905 | { |
2c47e605 YZ |
6906 | int level = wc->level; |
6907 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6908 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6909 | int ret; |
6910 | ||
2c47e605 YZ |
6911 | if (wc->stage == UPDATE_BACKREF && |
6912 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6913 | return 1; | |
f82d02d9 | 6914 | |
2c47e605 YZ |
6915 | /* |
6916 | * when reference count of tree block is 1, it won't increase | |
6917 | * again. once full backref flag is set, we never clear it. | |
6918 | */ | |
94fcca9f YZ |
6919 | if (lookup_info && |
6920 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6921 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6922 | BUG_ON(!path->locks[level]); |
6923 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 6924 | eb->start, level, 1, |
2c47e605 YZ |
6925 | &wc->refs[level], |
6926 | &wc->flags[level]); | |
79787eaa JM |
6927 | BUG_ON(ret == -ENOMEM); |
6928 | if (ret) | |
6929 | return ret; | |
2c47e605 YZ |
6930 | BUG_ON(wc->refs[level] == 0); |
6931 | } | |
5d4f98a2 | 6932 | |
2c47e605 YZ |
6933 | if (wc->stage == DROP_REFERENCE) { |
6934 | if (wc->refs[level] > 1) | |
6935 | return 1; | |
f82d02d9 | 6936 | |
2c47e605 | 6937 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 6938 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6939 | path->locks[level] = 0; |
6940 | } | |
6941 | return 0; | |
6942 | } | |
f82d02d9 | 6943 | |
2c47e605 YZ |
6944 | /* wc->stage == UPDATE_BACKREF */ |
6945 | if (!(wc->flags[level] & flag)) { | |
6946 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 6947 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
79787eaa | 6948 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 6949 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
79787eaa | 6950 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6951 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
6952 | eb->len, flag, 0); | |
79787eaa | 6953 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6954 | wc->flags[level] |= flag; |
6955 | } | |
6956 | ||
6957 | /* | |
6958 | * the block is shared by multiple trees, so it's not good to | |
6959 | * keep the tree lock | |
6960 | */ | |
6961 | if (path->locks[level] && level > 0) { | |
bd681513 | 6962 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6963 | path->locks[level] = 0; |
6964 | } | |
6965 | return 0; | |
6966 | } | |
6967 | ||
1c4850e2 | 6968 | /* |
2c016dc2 | 6969 | * helper to process tree block pointer. |
1c4850e2 YZ |
6970 | * |
6971 | * when wc->stage == DROP_REFERENCE, this function checks | |
6972 | * reference count of the block pointed to. if the block | |
6973 | * is shared and we need update back refs for the subtree | |
6974 | * rooted at the block, this function changes wc->stage to | |
6975 | * UPDATE_BACKREF. if the block is shared and there is no | |
6976 | * need to update back, this function drops the reference | |
6977 | * to the block. | |
6978 | * | |
6979 | * NOTE: return value 1 means we should stop walking down. | |
6980 | */ | |
6981 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
6982 | struct btrfs_root *root, | |
6983 | struct btrfs_path *path, | |
94fcca9f | 6984 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
6985 | { |
6986 | u64 bytenr; | |
6987 | u64 generation; | |
6988 | u64 parent; | |
6989 | u32 blocksize; | |
6990 | struct btrfs_key key; | |
6991 | struct extent_buffer *next; | |
6992 | int level = wc->level; | |
6993 | int reada = 0; | |
6994 | int ret = 0; | |
6995 | ||
6996 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
6997 | path->slots[level]); | |
6998 | /* | |
6999 | * if the lower level block was created before the snapshot | |
7000 | * was created, we know there is no need to update back refs | |
7001 | * for the subtree | |
7002 | */ | |
7003 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
7004 | generation <= root->root_key.offset) { |
7005 | *lookup_info = 1; | |
1c4850e2 | 7006 | return 1; |
94fcca9f | 7007 | } |
1c4850e2 YZ |
7008 | |
7009 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
7010 | blocksize = btrfs_level_size(root, level - 1); | |
7011 | ||
7012 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
7013 | if (!next) { | |
7014 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
7015 | if (!next) |
7016 | return -ENOMEM; | |
1c4850e2 YZ |
7017 | reada = 1; |
7018 | } | |
7019 | btrfs_tree_lock(next); | |
7020 | btrfs_set_lock_blocking(next); | |
7021 | ||
3173a18f | 7022 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, |
94fcca9f YZ |
7023 | &wc->refs[level - 1], |
7024 | &wc->flags[level - 1]); | |
79787eaa JM |
7025 | if (ret < 0) { |
7026 | btrfs_tree_unlock(next); | |
7027 | return ret; | |
7028 | } | |
7029 | ||
94fcca9f YZ |
7030 | BUG_ON(wc->refs[level - 1] == 0); |
7031 | *lookup_info = 0; | |
1c4850e2 | 7032 | |
94fcca9f | 7033 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 7034 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
7035 | if (level == 1 && |
7036 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7037 | goto skip; | |
7038 | ||
1c4850e2 YZ |
7039 | if (!wc->update_ref || |
7040 | generation <= root->root_key.offset) | |
7041 | goto skip; | |
7042 | ||
7043 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
7044 | path->slots[level]); | |
7045 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
7046 | if (ret < 0) | |
7047 | goto skip; | |
7048 | ||
7049 | wc->stage = UPDATE_BACKREF; | |
7050 | wc->shared_level = level - 1; | |
7051 | } | |
94fcca9f YZ |
7052 | } else { |
7053 | if (level == 1 && | |
7054 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7055 | goto skip; | |
1c4850e2 YZ |
7056 | } |
7057 | ||
b9fab919 | 7058 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
7059 | btrfs_tree_unlock(next); |
7060 | free_extent_buffer(next); | |
7061 | next = NULL; | |
94fcca9f | 7062 | *lookup_info = 1; |
1c4850e2 YZ |
7063 | } |
7064 | ||
7065 | if (!next) { | |
7066 | if (reada && level == 1) | |
7067 | reada_walk_down(trans, root, wc, path); | |
7068 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
7069 | if (!next) |
7070 | return -EIO; | |
1c4850e2 YZ |
7071 | btrfs_tree_lock(next); |
7072 | btrfs_set_lock_blocking(next); | |
7073 | } | |
7074 | ||
7075 | level--; | |
7076 | BUG_ON(level != btrfs_header_level(next)); | |
7077 | path->nodes[level] = next; | |
7078 | path->slots[level] = 0; | |
bd681513 | 7079 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
7080 | wc->level = level; |
7081 | if (wc->level == 1) | |
7082 | wc->reada_slot = 0; | |
7083 | return 0; | |
7084 | skip: | |
7085 | wc->refs[level - 1] = 0; | |
7086 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
7087 | if (wc->stage == DROP_REFERENCE) { |
7088 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
7089 | parent = path->nodes[level]->start; | |
7090 | } else { | |
7091 | BUG_ON(root->root_key.objectid != | |
7092 | btrfs_header_owner(path->nodes[level])); | |
7093 | parent = 0; | |
7094 | } | |
1c4850e2 | 7095 | |
94fcca9f | 7096 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 7097 | root->root_key.objectid, level - 1, 0, 0); |
79787eaa | 7098 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 7099 | } |
1c4850e2 YZ |
7100 | btrfs_tree_unlock(next); |
7101 | free_extent_buffer(next); | |
94fcca9f | 7102 | *lookup_info = 1; |
1c4850e2 YZ |
7103 | return 1; |
7104 | } | |
7105 | ||
2c47e605 | 7106 | /* |
2c016dc2 | 7107 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
7108 | * |
7109 | * when wc->stage == DROP_REFERENCE, this function drops | |
7110 | * reference count on the block. | |
7111 | * | |
7112 | * when wc->stage == UPDATE_BACKREF, this function changes | |
7113 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
7114 | * to UPDATE_BACKREF previously while processing the block. | |
7115 | * | |
7116 | * NOTE: return value 1 means we should stop walking up. | |
7117 | */ | |
7118 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
7119 | struct btrfs_root *root, | |
7120 | struct btrfs_path *path, | |
7121 | struct walk_control *wc) | |
7122 | { | |
f0486c68 | 7123 | int ret; |
2c47e605 YZ |
7124 | int level = wc->level; |
7125 | struct extent_buffer *eb = path->nodes[level]; | |
7126 | u64 parent = 0; | |
7127 | ||
7128 | if (wc->stage == UPDATE_BACKREF) { | |
7129 | BUG_ON(wc->shared_level < level); | |
7130 | if (level < wc->shared_level) | |
7131 | goto out; | |
7132 | ||
2c47e605 YZ |
7133 | ret = find_next_key(path, level + 1, &wc->update_progress); |
7134 | if (ret > 0) | |
7135 | wc->update_ref = 0; | |
7136 | ||
7137 | wc->stage = DROP_REFERENCE; | |
7138 | wc->shared_level = -1; | |
7139 | path->slots[level] = 0; | |
7140 | ||
7141 | /* | |
7142 | * check reference count again if the block isn't locked. | |
7143 | * we should start walking down the tree again if reference | |
7144 | * count is one. | |
7145 | */ | |
7146 | if (!path->locks[level]) { | |
7147 | BUG_ON(level == 0); | |
7148 | btrfs_tree_lock(eb); | |
7149 | btrfs_set_lock_blocking(eb); | |
bd681513 | 7150 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7151 | |
7152 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 7153 | eb->start, level, 1, |
2c47e605 YZ |
7154 | &wc->refs[level], |
7155 | &wc->flags[level]); | |
79787eaa JM |
7156 | if (ret < 0) { |
7157 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 7158 | path->locks[level] = 0; |
79787eaa JM |
7159 | return ret; |
7160 | } | |
2c47e605 YZ |
7161 | BUG_ON(wc->refs[level] == 0); |
7162 | if (wc->refs[level] == 1) { | |
bd681513 | 7163 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 7164 | path->locks[level] = 0; |
2c47e605 YZ |
7165 | return 1; |
7166 | } | |
f82d02d9 | 7167 | } |
2c47e605 | 7168 | } |
f82d02d9 | 7169 | |
2c47e605 YZ |
7170 | /* wc->stage == DROP_REFERENCE */ |
7171 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 7172 | |
2c47e605 YZ |
7173 | if (wc->refs[level] == 1) { |
7174 | if (level == 0) { | |
7175 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
7176 | ret = btrfs_dec_ref(trans, root, eb, 1, |
7177 | wc->for_reloc); | |
2c47e605 | 7178 | else |
66d7e7f0 AJ |
7179 | ret = btrfs_dec_ref(trans, root, eb, 0, |
7180 | wc->for_reloc); | |
79787eaa | 7181 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
7182 | } |
7183 | /* make block locked assertion in clean_tree_block happy */ | |
7184 | if (!path->locks[level] && | |
7185 | btrfs_header_generation(eb) == trans->transid) { | |
7186 | btrfs_tree_lock(eb); | |
7187 | btrfs_set_lock_blocking(eb); | |
bd681513 | 7188 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7189 | } |
7190 | clean_tree_block(trans, root, eb); | |
7191 | } | |
7192 | ||
7193 | if (eb == root->node) { | |
7194 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
7195 | parent = eb->start; | |
7196 | else | |
7197 | BUG_ON(root->root_key.objectid != | |
7198 | btrfs_header_owner(eb)); | |
7199 | } else { | |
7200 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
7201 | parent = path->nodes[level + 1]->start; | |
7202 | else | |
7203 | BUG_ON(root->root_key.objectid != | |
7204 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 7205 | } |
f82d02d9 | 7206 | |
5581a51a | 7207 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
7208 | out: |
7209 | wc->refs[level] = 0; | |
7210 | wc->flags[level] = 0; | |
f0486c68 | 7211 | return 0; |
2c47e605 YZ |
7212 | } |
7213 | ||
7214 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
7215 | struct btrfs_root *root, | |
7216 | struct btrfs_path *path, | |
7217 | struct walk_control *wc) | |
7218 | { | |
2c47e605 | 7219 | int level = wc->level; |
94fcca9f | 7220 | int lookup_info = 1; |
2c47e605 YZ |
7221 | int ret; |
7222 | ||
7223 | while (level >= 0) { | |
94fcca9f | 7224 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
7225 | if (ret > 0) |
7226 | break; | |
7227 | ||
7228 | if (level == 0) | |
7229 | break; | |
7230 | ||
7a7965f8 YZ |
7231 | if (path->slots[level] >= |
7232 | btrfs_header_nritems(path->nodes[level])) | |
7233 | break; | |
7234 | ||
94fcca9f | 7235 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
7236 | if (ret > 0) { |
7237 | path->slots[level]++; | |
7238 | continue; | |
90d2c51d MX |
7239 | } else if (ret < 0) |
7240 | return ret; | |
1c4850e2 | 7241 | level = wc->level; |
f82d02d9 | 7242 | } |
f82d02d9 YZ |
7243 | return 0; |
7244 | } | |
7245 | ||
d397712b | 7246 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 7247 | struct btrfs_root *root, |
f82d02d9 | 7248 | struct btrfs_path *path, |
2c47e605 | 7249 | struct walk_control *wc, int max_level) |
20524f02 | 7250 | { |
2c47e605 | 7251 | int level = wc->level; |
20524f02 | 7252 | int ret; |
9f3a7427 | 7253 | |
2c47e605 YZ |
7254 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
7255 | while (level < max_level && path->nodes[level]) { | |
7256 | wc->level = level; | |
7257 | if (path->slots[level] + 1 < | |
7258 | btrfs_header_nritems(path->nodes[level])) { | |
7259 | path->slots[level]++; | |
20524f02 CM |
7260 | return 0; |
7261 | } else { | |
2c47e605 YZ |
7262 | ret = walk_up_proc(trans, root, path, wc); |
7263 | if (ret > 0) | |
7264 | return 0; | |
bd56b302 | 7265 | |
2c47e605 | 7266 | if (path->locks[level]) { |
bd681513 CM |
7267 | btrfs_tree_unlock_rw(path->nodes[level], |
7268 | path->locks[level]); | |
2c47e605 | 7269 | path->locks[level] = 0; |
f82d02d9 | 7270 | } |
2c47e605 YZ |
7271 | free_extent_buffer(path->nodes[level]); |
7272 | path->nodes[level] = NULL; | |
7273 | level++; | |
20524f02 CM |
7274 | } |
7275 | } | |
7276 | return 1; | |
7277 | } | |
7278 | ||
9aca1d51 | 7279 | /* |
2c47e605 YZ |
7280 | * drop a subvolume tree. |
7281 | * | |
7282 | * this function traverses the tree freeing any blocks that only | |
7283 | * referenced by the tree. | |
7284 | * | |
7285 | * when a shared tree block is found. this function decreases its | |
7286 | * reference count by one. if update_ref is true, this function | |
7287 | * also make sure backrefs for the shared block and all lower level | |
7288 | * blocks are properly updated. | |
9d1a2a3a DS |
7289 | * |
7290 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 7291 | */ |
2c536799 | 7292 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
7293 | struct btrfs_block_rsv *block_rsv, int update_ref, |
7294 | int for_reloc) | |
20524f02 | 7295 | { |
5caf2a00 | 7296 | struct btrfs_path *path; |
2c47e605 YZ |
7297 | struct btrfs_trans_handle *trans; |
7298 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 7299 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
7300 | struct walk_control *wc; |
7301 | struct btrfs_key key; | |
7302 | int err = 0; | |
7303 | int ret; | |
7304 | int level; | |
20524f02 | 7305 | |
5caf2a00 | 7306 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
7307 | if (!path) { |
7308 | err = -ENOMEM; | |
7309 | goto out; | |
7310 | } | |
20524f02 | 7311 | |
2c47e605 | 7312 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
7313 | if (!wc) { |
7314 | btrfs_free_path(path); | |
cb1b69f4 TI |
7315 | err = -ENOMEM; |
7316 | goto out; | |
38a1a919 | 7317 | } |
2c47e605 | 7318 | |
a22285a6 | 7319 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7320 | if (IS_ERR(trans)) { |
7321 | err = PTR_ERR(trans); | |
7322 | goto out_free; | |
7323 | } | |
98d5dc13 | 7324 | |
3fd0a558 YZ |
7325 | if (block_rsv) |
7326 | trans->block_rsv = block_rsv; | |
2c47e605 | 7327 | |
9f3a7427 | 7328 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 7329 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
7330 | path->nodes[level] = btrfs_lock_root_node(root); |
7331 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 7332 | path->slots[level] = 0; |
bd681513 | 7333 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7334 | memset(&wc->update_progress, 0, |
7335 | sizeof(wc->update_progress)); | |
9f3a7427 | 7336 | } else { |
9f3a7427 | 7337 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
7338 | memcpy(&wc->update_progress, &key, |
7339 | sizeof(wc->update_progress)); | |
7340 | ||
6702ed49 | 7341 | level = root_item->drop_level; |
2c47e605 | 7342 | BUG_ON(level == 0); |
6702ed49 | 7343 | path->lowest_level = level; |
2c47e605 YZ |
7344 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
7345 | path->lowest_level = 0; | |
7346 | if (ret < 0) { | |
7347 | err = ret; | |
79787eaa | 7348 | goto out_end_trans; |
9f3a7427 | 7349 | } |
1c4850e2 | 7350 | WARN_ON(ret > 0); |
2c47e605 | 7351 | |
7d9eb12c CM |
7352 | /* |
7353 | * unlock our path, this is safe because only this | |
7354 | * function is allowed to delete this snapshot | |
7355 | */ | |
5d4f98a2 | 7356 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
7357 | |
7358 | level = btrfs_header_level(root->node); | |
7359 | while (1) { | |
7360 | btrfs_tree_lock(path->nodes[level]); | |
7361 | btrfs_set_lock_blocking(path->nodes[level]); | |
7362 | ||
7363 | ret = btrfs_lookup_extent_info(trans, root, | |
7364 | path->nodes[level]->start, | |
3173a18f | 7365 | level, 1, &wc->refs[level], |
2c47e605 | 7366 | &wc->flags[level]); |
79787eaa JM |
7367 | if (ret < 0) { |
7368 | err = ret; | |
7369 | goto out_end_trans; | |
7370 | } | |
2c47e605 YZ |
7371 | BUG_ON(wc->refs[level] == 0); |
7372 | ||
7373 | if (level == root_item->drop_level) | |
7374 | break; | |
7375 | ||
7376 | btrfs_tree_unlock(path->nodes[level]); | |
7377 | WARN_ON(wc->refs[level] != 1); | |
7378 | level--; | |
7379 | } | |
9f3a7427 | 7380 | } |
2c47e605 YZ |
7381 | |
7382 | wc->level = level; | |
7383 | wc->shared_level = -1; | |
7384 | wc->stage = DROP_REFERENCE; | |
7385 | wc->update_ref = update_ref; | |
7386 | wc->keep_locks = 0; | |
66d7e7f0 | 7387 | wc->for_reloc = for_reloc; |
1c4850e2 | 7388 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 7389 | |
d397712b | 7390 | while (1) { |
9d1a2a3a DS |
7391 | if (!for_reloc && btrfs_fs_closing(root->fs_info)) { |
7392 | pr_debug("btrfs: drop snapshot early exit\n"); | |
7393 | err = -EAGAIN; | |
7394 | goto out_end_trans; | |
7395 | } | |
7396 | ||
2c47e605 YZ |
7397 | ret = walk_down_tree(trans, root, path, wc); |
7398 | if (ret < 0) { | |
7399 | err = ret; | |
20524f02 | 7400 | break; |
2c47e605 | 7401 | } |
9aca1d51 | 7402 | |
2c47e605 YZ |
7403 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
7404 | if (ret < 0) { | |
7405 | err = ret; | |
20524f02 | 7406 | break; |
2c47e605 YZ |
7407 | } |
7408 | ||
7409 | if (ret > 0) { | |
7410 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
7411 | break; |
7412 | } | |
2c47e605 YZ |
7413 | |
7414 | if (wc->stage == DROP_REFERENCE) { | |
7415 | level = wc->level; | |
7416 | btrfs_node_key(path->nodes[level], | |
7417 | &root_item->drop_progress, | |
7418 | path->slots[level]); | |
7419 | root_item->drop_level = level; | |
7420 | } | |
7421 | ||
7422 | BUG_ON(wc->level == 0); | |
3fd0a558 | 7423 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
7424 | ret = btrfs_update_root(trans, tree_root, |
7425 | &root->root_key, | |
7426 | root_item); | |
79787eaa JM |
7427 | if (ret) { |
7428 | btrfs_abort_transaction(trans, tree_root, ret); | |
7429 | err = ret; | |
7430 | goto out_end_trans; | |
7431 | } | |
2c47e605 | 7432 | |
3fd0a558 | 7433 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 7434 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7435 | if (IS_ERR(trans)) { |
7436 | err = PTR_ERR(trans); | |
7437 | goto out_free; | |
7438 | } | |
3fd0a558 YZ |
7439 | if (block_rsv) |
7440 | trans->block_rsv = block_rsv; | |
c3e69d58 | 7441 | } |
20524f02 | 7442 | } |
b3b4aa74 | 7443 | btrfs_release_path(path); |
79787eaa JM |
7444 | if (err) |
7445 | goto out_end_trans; | |
2c47e605 YZ |
7446 | |
7447 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
7448 | if (ret) { |
7449 | btrfs_abort_transaction(trans, tree_root, ret); | |
7450 | goto out_end_trans; | |
7451 | } | |
2c47e605 | 7452 | |
76dda93c YZ |
7453 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
7454 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
7455 | NULL, NULL); | |
79787eaa JM |
7456 | if (ret < 0) { |
7457 | btrfs_abort_transaction(trans, tree_root, ret); | |
7458 | err = ret; | |
7459 | goto out_end_trans; | |
7460 | } else if (ret > 0) { | |
84cd948c JB |
7461 | /* if we fail to delete the orphan item this time |
7462 | * around, it'll get picked up the next time. | |
7463 | * | |
7464 | * The most common failure here is just -ENOENT. | |
7465 | */ | |
7466 | btrfs_del_orphan_item(trans, tree_root, | |
7467 | root->root_key.objectid); | |
76dda93c YZ |
7468 | } |
7469 | } | |
7470 | ||
7471 | if (root->in_radix) { | |
7472 | btrfs_free_fs_root(tree_root->fs_info, root); | |
7473 | } else { | |
7474 | free_extent_buffer(root->node); | |
7475 | free_extent_buffer(root->commit_root); | |
7476 | kfree(root); | |
7477 | } | |
79787eaa | 7478 | out_end_trans: |
3fd0a558 | 7479 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 7480 | out_free: |
2c47e605 | 7481 | kfree(wc); |
5caf2a00 | 7482 | btrfs_free_path(path); |
cb1b69f4 TI |
7483 | out: |
7484 | if (err) | |
7485 | btrfs_std_error(root->fs_info, err); | |
2c536799 | 7486 | return err; |
20524f02 | 7487 | } |
9078a3e1 | 7488 | |
2c47e605 YZ |
7489 | /* |
7490 | * drop subtree rooted at tree block 'node'. | |
7491 | * | |
7492 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 7493 | * only used by relocation code |
2c47e605 | 7494 | */ |
f82d02d9 YZ |
7495 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
7496 | struct btrfs_root *root, | |
7497 | struct extent_buffer *node, | |
7498 | struct extent_buffer *parent) | |
7499 | { | |
7500 | struct btrfs_path *path; | |
2c47e605 | 7501 | struct walk_control *wc; |
f82d02d9 YZ |
7502 | int level; |
7503 | int parent_level; | |
7504 | int ret = 0; | |
7505 | int wret; | |
7506 | ||
2c47e605 YZ |
7507 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
7508 | ||
f82d02d9 | 7509 | path = btrfs_alloc_path(); |
db5b493a TI |
7510 | if (!path) |
7511 | return -ENOMEM; | |
f82d02d9 | 7512 | |
2c47e605 | 7513 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
7514 | if (!wc) { |
7515 | btrfs_free_path(path); | |
7516 | return -ENOMEM; | |
7517 | } | |
2c47e605 | 7518 | |
b9447ef8 | 7519 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
7520 | parent_level = btrfs_header_level(parent); |
7521 | extent_buffer_get(parent); | |
7522 | path->nodes[parent_level] = parent; | |
7523 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
7524 | ||
b9447ef8 | 7525 | btrfs_assert_tree_locked(node); |
f82d02d9 | 7526 | level = btrfs_header_level(node); |
f82d02d9 YZ |
7527 | path->nodes[level] = node; |
7528 | path->slots[level] = 0; | |
bd681513 | 7529 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7530 | |
7531 | wc->refs[parent_level] = 1; | |
7532 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7533 | wc->level = level; | |
7534 | wc->shared_level = -1; | |
7535 | wc->stage = DROP_REFERENCE; | |
7536 | wc->update_ref = 0; | |
7537 | wc->keep_locks = 1; | |
66d7e7f0 | 7538 | wc->for_reloc = 1; |
1c4850e2 | 7539 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
7540 | |
7541 | while (1) { | |
2c47e605 YZ |
7542 | wret = walk_down_tree(trans, root, path, wc); |
7543 | if (wret < 0) { | |
f82d02d9 | 7544 | ret = wret; |
f82d02d9 | 7545 | break; |
2c47e605 | 7546 | } |
f82d02d9 | 7547 | |
2c47e605 | 7548 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
7549 | if (wret < 0) |
7550 | ret = wret; | |
7551 | if (wret != 0) | |
7552 | break; | |
7553 | } | |
7554 | ||
2c47e605 | 7555 | kfree(wc); |
f82d02d9 YZ |
7556 | btrfs_free_path(path); |
7557 | return ret; | |
7558 | } | |
7559 | ||
ec44a35c CM |
7560 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7561 | { | |
7562 | u64 num_devices; | |
fc67c450 | 7563 | u64 stripped; |
e4d8ec0f | 7564 | |
fc67c450 ID |
7565 | /* |
7566 | * if restripe for this chunk_type is on pick target profile and | |
7567 | * return, otherwise do the usual balance | |
7568 | */ | |
7569 | stripped = get_restripe_target(root->fs_info, flags); | |
7570 | if (stripped) | |
7571 | return extended_to_chunk(stripped); | |
e4d8ec0f | 7572 | |
cd02dca5 CM |
7573 | /* |
7574 | * we add in the count of missing devices because we want | |
7575 | * to make sure that any RAID levels on a degraded FS | |
7576 | * continue to be honored. | |
7577 | */ | |
7578 | num_devices = root->fs_info->fs_devices->rw_devices + | |
7579 | root->fs_info->fs_devices->missing_devices; | |
7580 | ||
fc67c450 | 7581 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 7582 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
7583 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
7584 | ||
ec44a35c CM |
7585 | if (num_devices == 1) { |
7586 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7587 | stripped = flags & ~stripped; | |
7588 | ||
7589 | /* turn raid0 into single device chunks */ | |
7590 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7591 | return stripped; | |
7592 | ||
7593 | /* turn mirroring into duplication */ | |
7594 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7595 | BTRFS_BLOCK_GROUP_RAID10)) | |
7596 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
7597 | } else { |
7598 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7599 | if (flags & stripped) |
7600 | return flags; | |
7601 | ||
7602 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7603 | stripped = flags & ~stripped; | |
7604 | ||
7605 | /* switch duplicated blocks with raid1 */ | |
7606 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7607 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7608 | ||
e3176ca2 | 7609 | /* this is drive concat, leave it alone */ |
ec44a35c | 7610 | } |
e3176ca2 | 7611 | |
ec44a35c CM |
7612 | return flags; |
7613 | } | |
7614 | ||
199c36ea | 7615 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 7616 | { |
f0486c68 YZ |
7617 | struct btrfs_space_info *sinfo = cache->space_info; |
7618 | u64 num_bytes; | |
199c36ea | 7619 | u64 min_allocable_bytes; |
f0486c68 | 7620 | int ret = -ENOSPC; |
0ef3e66b | 7621 | |
c286ac48 | 7622 | |
199c36ea MX |
7623 | /* |
7624 | * We need some metadata space and system metadata space for | |
7625 | * allocating chunks in some corner cases until we force to set | |
7626 | * it to be readonly. | |
7627 | */ | |
7628 | if ((sinfo->flags & | |
7629 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
7630 | !force) | |
7631 | min_allocable_bytes = 1 * 1024 * 1024; | |
7632 | else | |
7633 | min_allocable_bytes = 0; | |
7634 | ||
f0486c68 YZ |
7635 | spin_lock(&sinfo->lock); |
7636 | spin_lock(&cache->lock); | |
61cfea9b W |
7637 | |
7638 | if (cache->ro) { | |
7639 | ret = 0; | |
7640 | goto out; | |
7641 | } | |
7642 | ||
f0486c68 YZ |
7643 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
7644 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7645 | ||
7646 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
7647 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
7648 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 7649 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
7650 | cache->ro = 1; |
7651 | ret = 0; | |
7652 | } | |
61cfea9b | 7653 | out: |
f0486c68 YZ |
7654 | spin_unlock(&cache->lock); |
7655 | spin_unlock(&sinfo->lock); | |
7656 | return ret; | |
7657 | } | |
7d9eb12c | 7658 | |
f0486c68 YZ |
7659 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
7660 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 7661 | |
f0486c68 YZ |
7662 | { |
7663 | struct btrfs_trans_handle *trans; | |
7664 | u64 alloc_flags; | |
7665 | int ret; | |
7d9eb12c | 7666 | |
f0486c68 | 7667 | BUG_ON(cache->ro); |
0ef3e66b | 7668 | |
ff5714cc | 7669 | trans = btrfs_join_transaction(root); |
79787eaa JM |
7670 | if (IS_ERR(trans)) |
7671 | return PTR_ERR(trans); | |
5d4f98a2 | 7672 | |
f0486c68 | 7673 | alloc_flags = update_block_group_flags(root, cache->flags); |
79787eaa | 7674 | if (alloc_flags != cache->flags) { |
698d0082 | 7675 | ret = do_chunk_alloc(trans, root, alloc_flags, |
79787eaa JM |
7676 | CHUNK_ALLOC_FORCE); |
7677 | if (ret < 0) | |
7678 | goto out; | |
7679 | } | |
5d4f98a2 | 7680 | |
199c36ea | 7681 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7682 | if (!ret) |
7683 | goto out; | |
7684 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 7685 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7686 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
7687 | if (ret < 0) |
7688 | goto out; | |
199c36ea | 7689 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7690 | out: |
7691 | btrfs_end_transaction(trans, root); | |
7692 | return ret; | |
7693 | } | |
5d4f98a2 | 7694 | |
c87f08ca CM |
7695 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7696 | struct btrfs_root *root, u64 type) | |
7697 | { | |
7698 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 7699 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7700 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
7701 | } |
7702 | ||
6d07bcec MX |
7703 | /* |
7704 | * helper to account the unused space of all the readonly block group in the | |
7705 | * list. takes mirrors into account. | |
7706 | */ | |
7707 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7708 | { | |
7709 | struct btrfs_block_group_cache *block_group; | |
7710 | u64 free_bytes = 0; | |
7711 | int factor; | |
7712 | ||
7713 | list_for_each_entry(block_group, groups_list, list) { | |
7714 | spin_lock(&block_group->lock); | |
7715 | ||
7716 | if (!block_group->ro) { | |
7717 | spin_unlock(&block_group->lock); | |
7718 | continue; | |
7719 | } | |
7720 | ||
7721 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7722 | BTRFS_BLOCK_GROUP_RAID10 | | |
7723 | BTRFS_BLOCK_GROUP_DUP)) | |
7724 | factor = 2; | |
7725 | else | |
7726 | factor = 1; | |
7727 | ||
7728 | free_bytes += (block_group->key.offset - | |
7729 | btrfs_block_group_used(&block_group->item)) * | |
7730 | factor; | |
7731 | ||
7732 | spin_unlock(&block_group->lock); | |
7733 | } | |
7734 | ||
7735 | return free_bytes; | |
7736 | } | |
7737 | ||
7738 | /* | |
7739 | * helper to account the unused space of all the readonly block group in the | |
7740 | * space_info. takes mirrors into account. | |
7741 | */ | |
7742 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
7743 | { | |
7744 | int i; | |
7745 | u64 free_bytes = 0; | |
7746 | ||
7747 | spin_lock(&sinfo->lock); | |
7748 | ||
7749 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
7750 | if (!list_empty(&sinfo->block_groups[i])) | |
7751 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
7752 | &sinfo->block_groups[i]); | |
7753 | ||
7754 | spin_unlock(&sinfo->lock); | |
7755 | ||
7756 | return free_bytes; | |
7757 | } | |
7758 | ||
143bede5 | 7759 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
f0486c68 | 7760 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 7761 | { |
f0486c68 YZ |
7762 | struct btrfs_space_info *sinfo = cache->space_info; |
7763 | u64 num_bytes; | |
7764 | ||
7765 | BUG_ON(!cache->ro); | |
7766 | ||
7767 | spin_lock(&sinfo->lock); | |
7768 | spin_lock(&cache->lock); | |
7769 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7770 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7771 | sinfo->bytes_readonly -= num_bytes; | |
7772 | cache->ro = 0; | |
7773 | spin_unlock(&cache->lock); | |
7774 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
7775 | } |
7776 | ||
ba1bf481 JB |
7777 | /* |
7778 | * checks to see if its even possible to relocate this block group. | |
7779 | * | |
7780 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7781 | * ok to go ahead and try. | |
7782 | */ | |
7783 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7784 | { |
ba1bf481 JB |
7785 | struct btrfs_block_group_cache *block_group; |
7786 | struct btrfs_space_info *space_info; | |
7787 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7788 | struct btrfs_device *device; | |
cdcb725c | 7789 | u64 min_free; |
6719db6a JB |
7790 | u64 dev_min = 1; |
7791 | u64 dev_nr = 0; | |
4a5e98f5 | 7792 | u64 target; |
cdcb725c | 7793 | int index; |
ba1bf481 JB |
7794 | int full = 0; |
7795 | int ret = 0; | |
1a40e23b | 7796 | |
ba1bf481 | 7797 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7798 | |
ba1bf481 JB |
7799 | /* odd, couldn't find the block group, leave it alone */ |
7800 | if (!block_group) | |
7801 | return -1; | |
1a40e23b | 7802 | |
cdcb725c | 7803 | min_free = btrfs_block_group_used(&block_group->item); |
7804 | ||
ba1bf481 | 7805 | /* no bytes used, we're good */ |
cdcb725c | 7806 | if (!min_free) |
1a40e23b ZY |
7807 | goto out; |
7808 | ||
ba1bf481 JB |
7809 | space_info = block_group->space_info; |
7810 | spin_lock(&space_info->lock); | |
17d217fe | 7811 | |
ba1bf481 | 7812 | full = space_info->full; |
17d217fe | 7813 | |
ba1bf481 JB |
7814 | /* |
7815 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7816 | * relocate it unless we're able to allocate a new chunk below. |
7817 | * | |
7818 | * Otherwise, we need to make sure we have room in the space to handle | |
7819 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7820 | */ |
7ce618db | 7821 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 7822 | (space_info->bytes_used + space_info->bytes_reserved + |
7823 | space_info->bytes_pinned + space_info->bytes_readonly + | |
7824 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
7825 | spin_unlock(&space_info->lock); |
7826 | goto out; | |
17d217fe | 7827 | } |
ba1bf481 | 7828 | spin_unlock(&space_info->lock); |
ea8c2819 | 7829 | |
ba1bf481 JB |
7830 | /* |
7831 | * ok we don't have enough space, but maybe we have free space on our | |
7832 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
7833 | * alloc devices and guess if we have enough space. if this block |
7834 | * group is going to be restriped, run checks against the target | |
7835 | * profile instead of the current one. | |
ba1bf481 JB |
7836 | */ |
7837 | ret = -1; | |
ea8c2819 | 7838 | |
cdcb725c | 7839 | /* |
7840 | * index: | |
7841 | * 0: raid10 | |
7842 | * 1: raid1 | |
7843 | * 2: dup | |
7844 | * 3: raid0 | |
7845 | * 4: single | |
7846 | */ | |
4a5e98f5 ID |
7847 | target = get_restripe_target(root->fs_info, block_group->flags); |
7848 | if (target) { | |
31e50229 | 7849 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
7850 | } else { |
7851 | /* | |
7852 | * this is just a balance, so if we were marked as full | |
7853 | * we know there is no space for a new chunk | |
7854 | */ | |
7855 | if (full) | |
7856 | goto out; | |
7857 | ||
7858 | index = get_block_group_index(block_group); | |
7859 | } | |
7860 | ||
e6ec716f | 7861 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 7862 | dev_min = 4; |
6719db6a JB |
7863 | /* Divide by 2 */ |
7864 | min_free >>= 1; | |
e6ec716f | 7865 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 7866 | dev_min = 2; |
e6ec716f | 7867 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
7868 | /* Multiply by 2 */ |
7869 | min_free <<= 1; | |
e6ec716f | 7870 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 7871 | dev_min = fs_devices->rw_devices; |
6719db6a | 7872 | do_div(min_free, dev_min); |
cdcb725c | 7873 | } |
7874 | ||
ba1bf481 JB |
7875 | mutex_lock(&root->fs_info->chunk_mutex); |
7876 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 7877 | u64 dev_offset; |
56bec294 | 7878 | |
ba1bf481 JB |
7879 | /* |
7880 | * check to make sure we can actually find a chunk with enough | |
7881 | * space to fit our block group in. | |
7882 | */ | |
63a212ab SB |
7883 | if (device->total_bytes > device->bytes_used + min_free && |
7884 | !device->is_tgtdev_for_dev_replace) { | |
125ccb0a | 7885 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 7886 | &dev_offset, NULL); |
ba1bf481 | 7887 | if (!ret) |
cdcb725c | 7888 | dev_nr++; |
7889 | ||
7890 | if (dev_nr >= dev_min) | |
73e48b27 | 7891 | break; |
cdcb725c | 7892 | |
ba1bf481 | 7893 | ret = -1; |
725c8463 | 7894 | } |
edbd8d4e | 7895 | } |
ba1bf481 | 7896 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7897 | out: |
ba1bf481 | 7898 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7899 | return ret; |
7900 | } | |
7901 | ||
b2950863 CH |
7902 | static int find_first_block_group(struct btrfs_root *root, |
7903 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7904 | { |
925baedd | 7905 | int ret = 0; |
0b86a832 CM |
7906 | struct btrfs_key found_key; |
7907 | struct extent_buffer *leaf; | |
7908 | int slot; | |
edbd8d4e | 7909 | |
0b86a832 CM |
7910 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7911 | if (ret < 0) | |
925baedd CM |
7912 | goto out; |
7913 | ||
d397712b | 7914 | while (1) { |
0b86a832 | 7915 | slot = path->slots[0]; |
edbd8d4e | 7916 | leaf = path->nodes[0]; |
0b86a832 CM |
7917 | if (slot >= btrfs_header_nritems(leaf)) { |
7918 | ret = btrfs_next_leaf(root, path); | |
7919 | if (ret == 0) | |
7920 | continue; | |
7921 | if (ret < 0) | |
925baedd | 7922 | goto out; |
0b86a832 | 7923 | break; |
edbd8d4e | 7924 | } |
0b86a832 | 7925 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7926 | |
0b86a832 | 7927 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7928 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7929 | ret = 0; | |
7930 | goto out; | |
7931 | } | |
0b86a832 | 7932 | path->slots[0]++; |
edbd8d4e | 7933 | } |
925baedd | 7934 | out: |
0b86a832 | 7935 | return ret; |
edbd8d4e CM |
7936 | } |
7937 | ||
0af3d00b JB |
7938 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
7939 | { | |
7940 | struct btrfs_block_group_cache *block_group; | |
7941 | u64 last = 0; | |
7942 | ||
7943 | while (1) { | |
7944 | struct inode *inode; | |
7945 | ||
7946 | block_group = btrfs_lookup_first_block_group(info, last); | |
7947 | while (block_group) { | |
7948 | spin_lock(&block_group->lock); | |
7949 | if (block_group->iref) | |
7950 | break; | |
7951 | spin_unlock(&block_group->lock); | |
7952 | block_group = next_block_group(info->tree_root, | |
7953 | block_group); | |
7954 | } | |
7955 | if (!block_group) { | |
7956 | if (last == 0) | |
7957 | break; | |
7958 | last = 0; | |
7959 | continue; | |
7960 | } | |
7961 | ||
7962 | inode = block_group->inode; | |
7963 | block_group->iref = 0; | |
7964 | block_group->inode = NULL; | |
7965 | spin_unlock(&block_group->lock); | |
7966 | iput(inode); | |
7967 | last = block_group->key.objectid + block_group->key.offset; | |
7968 | btrfs_put_block_group(block_group); | |
7969 | } | |
7970 | } | |
7971 | ||
1a40e23b ZY |
7972 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7973 | { | |
7974 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7975 | struct btrfs_space_info *space_info; |
11833d66 | 7976 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7977 | struct rb_node *n; |
7978 | ||
11833d66 YZ |
7979 | down_write(&info->extent_commit_sem); |
7980 | while (!list_empty(&info->caching_block_groups)) { | |
7981 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7982 | struct btrfs_caching_control, list); | |
7983 | list_del(&caching_ctl->list); | |
7984 | put_caching_control(caching_ctl); | |
7985 | } | |
7986 | up_write(&info->extent_commit_sem); | |
7987 | ||
1a40e23b ZY |
7988 | spin_lock(&info->block_group_cache_lock); |
7989 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7990 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7991 | cache_node); | |
1a40e23b ZY |
7992 | rb_erase(&block_group->cache_node, |
7993 | &info->block_group_cache_tree); | |
d899e052 YZ |
7994 | spin_unlock(&info->block_group_cache_lock); |
7995 | ||
80eb234a | 7996 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7997 | list_del(&block_group->list); |
80eb234a | 7998 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7999 | |
817d52f8 | 8000 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8001 | wait_block_group_cache_done(block_group); |
817d52f8 | 8002 | |
3c14874a JB |
8003 | /* |
8004 | * We haven't cached this block group, which means we could | |
8005 | * possibly have excluded extents on this block group. | |
8006 | */ | |
8007 | if (block_group->cached == BTRFS_CACHE_NO) | |
8008 | free_excluded_extents(info->extent_root, block_group); | |
8009 | ||
817d52f8 | 8010 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 8011 | btrfs_put_block_group(block_group); |
d899e052 YZ |
8012 | |
8013 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
8014 | } |
8015 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
8016 | |
8017 | /* now that all the block groups are freed, go through and | |
8018 | * free all the space_info structs. This is only called during | |
8019 | * the final stages of unmount, and so we know nobody is | |
8020 | * using them. We call synchronize_rcu() once before we start, | |
8021 | * just to be on the safe side. | |
8022 | */ | |
8023 | synchronize_rcu(); | |
8024 | ||
8929ecfa YZ |
8025 | release_global_block_rsv(info); |
8026 | ||
4184ea7f CM |
8027 | while(!list_empty(&info->space_info)) { |
8028 | space_info = list_entry(info->space_info.next, | |
8029 | struct btrfs_space_info, | |
8030 | list); | |
b069e0c3 DS |
8031 | if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) { |
8032 | if (space_info->bytes_pinned > 0 || | |
8033 | space_info->bytes_reserved > 0 || | |
8034 | space_info->bytes_may_use > 0) { | |
8035 | WARN_ON(1); | |
8036 | dump_space_info(space_info, 0, 0); | |
8037 | } | |
f0486c68 | 8038 | } |
4184ea7f CM |
8039 | list_del(&space_info->list); |
8040 | kfree(space_info); | |
8041 | } | |
1a40e23b ZY |
8042 | return 0; |
8043 | } | |
8044 | ||
b742bb82 YZ |
8045 | static void __link_block_group(struct btrfs_space_info *space_info, |
8046 | struct btrfs_block_group_cache *cache) | |
8047 | { | |
8048 | int index = get_block_group_index(cache); | |
8049 | ||
8050 | down_write(&space_info->groups_sem); | |
8051 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
8052 | up_write(&space_info->groups_sem); | |
8053 | } | |
8054 | ||
9078a3e1 CM |
8055 | int btrfs_read_block_groups(struct btrfs_root *root) |
8056 | { | |
8057 | struct btrfs_path *path; | |
8058 | int ret; | |
9078a3e1 | 8059 | struct btrfs_block_group_cache *cache; |
be744175 | 8060 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 8061 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
8062 | struct btrfs_key key; |
8063 | struct btrfs_key found_key; | |
5f39d397 | 8064 | struct extent_buffer *leaf; |
0af3d00b JB |
8065 | int need_clear = 0; |
8066 | u64 cache_gen; | |
96b5179d | 8067 | |
be744175 | 8068 | root = info->extent_root; |
9078a3e1 | 8069 | key.objectid = 0; |
0b86a832 | 8070 | key.offset = 0; |
9078a3e1 | 8071 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
8072 | path = btrfs_alloc_path(); |
8073 | if (!path) | |
8074 | return -ENOMEM; | |
026fd317 | 8075 | path->reada = 1; |
9078a3e1 | 8076 | |
6c41761f | 8077 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 8078 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 8079 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 8080 | need_clear = 1; |
88c2ba3b JB |
8081 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
8082 | need_clear = 1; | |
0af3d00b | 8083 | |
d397712b | 8084 | while (1) { |
0b86a832 | 8085 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
8086 | if (ret > 0) |
8087 | break; | |
0b86a832 CM |
8088 | if (ret != 0) |
8089 | goto error; | |
5f39d397 CM |
8090 | leaf = path->nodes[0]; |
8091 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 8092 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 8093 | if (!cache) { |
0b86a832 | 8094 | ret = -ENOMEM; |
f0486c68 | 8095 | goto error; |
9078a3e1 | 8096 | } |
34d52cb6 LZ |
8097 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
8098 | GFP_NOFS); | |
8099 | if (!cache->free_space_ctl) { | |
8100 | kfree(cache); | |
8101 | ret = -ENOMEM; | |
8102 | goto error; | |
8103 | } | |
3e1ad54f | 8104 | |
d2fb3437 | 8105 | atomic_set(&cache->count, 1); |
c286ac48 | 8106 | spin_lock_init(&cache->lock); |
817d52f8 | 8107 | cache->fs_info = info; |
0f9dd46c | 8108 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8109 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 8110 | |
cf7c1ef6 LB |
8111 | if (need_clear) { |
8112 | /* | |
8113 | * When we mount with old space cache, we need to | |
8114 | * set BTRFS_DC_CLEAR and set dirty flag. | |
8115 | * | |
8116 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
8117 | * truncate the old free space cache inode and | |
8118 | * setup a new one. | |
8119 | * b) Setting 'dirty flag' makes sure that we flush | |
8120 | * the new space cache info onto disk. | |
8121 | */ | |
0af3d00b | 8122 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 LB |
8123 | if (btrfs_test_opt(root, SPACE_CACHE)) |
8124 | cache->dirty = 1; | |
8125 | } | |
0af3d00b | 8126 | |
5f39d397 CM |
8127 | read_extent_buffer(leaf, &cache->item, |
8128 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
8129 | sizeof(cache->item)); | |
9078a3e1 | 8130 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 8131 | |
9078a3e1 | 8132 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 8133 | btrfs_release_path(path); |
0b86a832 | 8134 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 | 8135 | cache->sectorsize = root->sectorsize; |
53b381b3 DW |
8136 | cache->full_stripe_len = btrfs_full_stripe_len(root, |
8137 | &root->fs_info->mapping_tree, | |
8138 | found_key.objectid); | |
34d52cb6 LZ |
8139 | btrfs_init_free_space_ctl(cache); |
8140 | ||
3c14874a JB |
8141 | /* |
8142 | * We need to exclude the super stripes now so that the space | |
8143 | * info has super bytes accounted for, otherwise we'll think | |
8144 | * we have more space than we actually do. | |
8145 | */ | |
835d974f JB |
8146 | ret = exclude_super_stripes(root, cache); |
8147 | if (ret) { | |
8148 | /* | |
8149 | * We may have excluded something, so call this just in | |
8150 | * case. | |
8151 | */ | |
8152 | free_excluded_extents(root, cache); | |
8153 | kfree(cache->free_space_ctl); | |
8154 | kfree(cache); | |
8155 | goto error; | |
8156 | } | |
3c14874a | 8157 | |
817d52f8 JB |
8158 | /* |
8159 | * check for two cases, either we are full, and therefore | |
8160 | * don't need to bother with the caching work since we won't | |
8161 | * find any space, or we are empty, and we can just add all | |
8162 | * the space in and be done with it. This saves us _alot_ of | |
8163 | * time, particularly in the full case. | |
8164 | */ | |
8165 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 8166 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8167 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 8168 | free_excluded_extents(root, cache); |
817d52f8 | 8169 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 8170 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
8171 | cache->cached = BTRFS_CACHE_FINISHED; |
8172 | add_new_free_space(cache, root->fs_info, | |
8173 | found_key.objectid, | |
8174 | found_key.objectid + | |
8175 | found_key.offset); | |
11833d66 | 8176 | free_excluded_extents(root, cache); |
817d52f8 | 8177 | } |
96b5179d | 8178 | |
6324fbf3 CM |
8179 | ret = update_space_info(info, cache->flags, found_key.offset, |
8180 | btrfs_block_group_used(&cache->item), | |
8181 | &space_info); | |
79787eaa | 8182 | BUG_ON(ret); /* -ENOMEM */ |
6324fbf3 | 8183 | cache->space_info = space_info; |
1b2da372 | 8184 | spin_lock(&cache->space_info->lock); |
f0486c68 | 8185 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8186 | spin_unlock(&cache->space_info->lock); |
8187 | ||
b742bb82 | 8188 | __link_block_group(space_info, cache); |
0f9dd46c JB |
8189 | |
8190 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
79787eaa | 8191 | BUG_ON(ret); /* Logic error */ |
75ccf47d CM |
8192 | |
8193 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 8194 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 8195 | set_block_group_ro(cache, 1); |
9078a3e1 | 8196 | } |
b742bb82 YZ |
8197 | |
8198 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
8199 | if (!(get_alloc_profile(root, space_info->flags) & | |
8200 | (BTRFS_BLOCK_GROUP_RAID10 | | |
8201 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
8202 | BTRFS_BLOCK_GROUP_RAID5 | |
8203 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
8204 | BTRFS_BLOCK_GROUP_DUP))) |
8205 | continue; | |
8206 | /* | |
8207 | * avoid allocating from un-mirrored block group if there are | |
8208 | * mirrored block groups. | |
8209 | */ | |
8210 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 8211 | set_block_group_ro(cache, 1); |
b742bb82 | 8212 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 8213 | set_block_group_ro(cache, 1); |
9078a3e1 | 8214 | } |
f0486c68 YZ |
8215 | |
8216 | init_global_block_rsv(info); | |
0b86a832 CM |
8217 | ret = 0; |
8218 | error: | |
9078a3e1 | 8219 | btrfs_free_path(path); |
0b86a832 | 8220 | return ret; |
9078a3e1 | 8221 | } |
6324fbf3 | 8222 | |
ea658bad JB |
8223 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
8224 | struct btrfs_root *root) | |
8225 | { | |
8226 | struct btrfs_block_group_cache *block_group, *tmp; | |
8227 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
8228 | struct btrfs_block_group_item item; | |
8229 | struct btrfs_key key; | |
8230 | int ret = 0; | |
8231 | ||
8232 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, | |
8233 | new_bg_list) { | |
8234 | list_del_init(&block_group->new_bg_list); | |
8235 | ||
8236 | if (ret) | |
8237 | continue; | |
8238 | ||
8239 | spin_lock(&block_group->lock); | |
8240 | memcpy(&item, &block_group->item, sizeof(item)); | |
8241 | memcpy(&key, &block_group->key, sizeof(key)); | |
8242 | spin_unlock(&block_group->lock); | |
8243 | ||
8244 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
8245 | sizeof(item)); | |
8246 | if (ret) | |
8247 | btrfs_abort_transaction(trans, extent_root, ret); | |
8248 | } | |
8249 | } | |
8250 | ||
6324fbf3 CM |
8251 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
8252 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 8253 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
8254 | u64 size) |
8255 | { | |
8256 | int ret; | |
6324fbf3 CM |
8257 | struct btrfs_root *extent_root; |
8258 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
8259 | |
8260 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 8261 | |
12fcfd22 | 8262 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 8263 | |
8f18cf13 | 8264 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
8265 | if (!cache) |
8266 | return -ENOMEM; | |
34d52cb6 LZ |
8267 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
8268 | GFP_NOFS); | |
8269 | if (!cache->free_space_ctl) { | |
8270 | kfree(cache); | |
8271 | return -ENOMEM; | |
8272 | } | |
0f9dd46c | 8273 | |
e17cade2 | 8274 | cache->key.objectid = chunk_offset; |
6324fbf3 | 8275 | cache->key.offset = size; |
d2fb3437 | 8276 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 8277 | cache->sectorsize = root->sectorsize; |
0af3d00b | 8278 | cache->fs_info = root->fs_info; |
53b381b3 DW |
8279 | cache->full_stripe_len = btrfs_full_stripe_len(root, |
8280 | &root->fs_info->mapping_tree, | |
8281 | chunk_offset); | |
96303081 | 8282 | |
d2fb3437 | 8283 | atomic_set(&cache->count, 1); |
c286ac48 | 8284 | spin_lock_init(&cache->lock); |
0f9dd46c | 8285 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8286 | INIT_LIST_HEAD(&cache->cluster_list); |
ea658bad | 8287 | INIT_LIST_HEAD(&cache->new_bg_list); |
0ef3e66b | 8288 | |
34d52cb6 LZ |
8289 | btrfs_init_free_space_ctl(cache); |
8290 | ||
6324fbf3 | 8291 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
8292 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
8293 | cache->flags = type; | |
8294 | btrfs_set_block_group_flags(&cache->item, type); | |
8295 | ||
11833d66 | 8296 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8297 | cache->cached = BTRFS_CACHE_FINISHED; |
835d974f JB |
8298 | ret = exclude_super_stripes(root, cache); |
8299 | if (ret) { | |
8300 | /* | |
8301 | * We may have excluded something, so call this just in | |
8302 | * case. | |
8303 | */ | |
8304 | free_excluded_extents(root, cache); | |
8305 | kfree(cache->free_space_ctl); | |
8306 | kfree(cache); | |
8307 | return ret; | |
8308 | } | |
96303081 | 8309 | |
817d52f8 JB |
8310 | add_new_free_space(cache, root->fs_info, chunk_offset, |
8311 | chunk_offset + size); | |
8312 | ||
11833d66 YZ |
8313 | free_excluded_extents(root, cache); |
8314 | ||
6324fbf3 CM |
8315 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
8316 | &cache->space_info); | |
79787eaa | 8317 | BUG_ON(ret); /* -ENOMEM */ |
c7c144db | 8318 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
8319 | |
8320 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 8321 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8322 | spin_unlock(&cache->space_info->lock); |
8323 | ||
b742bb82 | 8324 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 8325 | |
0f9dd46c | 8326 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
79787eaa | 8327 | BUG_ON(ret); /* Logic error */ |
c286ac48 | 8328 | |
ea658bad | 8329 | list_add_tail(&cache->new_bg_list, &trans->new_bgs); |
6324fbf3 | 8330 | |
d18a2c44 | 8331 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 8332 | |
6324fbf3 CM |
8333 | return 0; |
8334 | } | |
1a40e23b | 8335 | |
10ea00f5 ID |
8336 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
8337 | { | |
899c81ea ID |
8338 | u64 extra_flags = chunk_to_extended(flags) & |
8339 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 8340 | |
de98ced9 | 8341 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
8342 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
8343 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
8344 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
8345 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
8346 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
8347 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 8348 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
8349 | } |
8350 | ||
1a40e23b ZY |
8351 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
8352 | struct btrfs_root *root, u64 group_start) | |
8353 | { | |
8354 | struct btrfs_path *path; | |
8355 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 8356 | struct btrfs_free_cluster *cluster; |
0af3d00b | 8357 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 8358 | struct btrfs_key key; |
0af3d00b | 8359 | struct inode *inode; |
1a40e23b | 8360 | int ret; |
10ea00f5 | 8361 | int index; |
89a55897 | 8362 | int factor; |
1a40e23b | 8363 | |
1a40e23b ZY |
8364 | root = root->fs_info->extent_root; |
8365 | ||
8366 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
8367 | BUG_ON(!block_group); | |
c146afad | 8368 | BUG_ON(!block_group->ro); |
1a40e23b | 8369 | |
9f7c43c9 | 8370 | /* |
8371 | * Free the reserved super bytes from this block group before | |
8372 | * remove it. | |
8373 | */ | |
8374 | free_excluded_extents(root, block_group); | |
8375 | ||
1a40e23b | 8376 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 8377 | index = get_block_group_index(block_group); |
89a55897 JB |
8378 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
8379 | BTRFS_BLOCK_GROUP_RAID1 | | |
8380 | BTRFS_BLOCK_GROUP_RAID10)) | |
8381 | factor = 2; | |
8382 | else | |
8383 | factor = 1; | |
1a40e23b | 8384 | |
44fb5511 CM |
8385 | /* make sure this block group isn't part of an allocation cluster */ |
8386 | cluster = &root->fs_info->data_alloc_cluster; | |
8387 | spin_lock(&cluster->refill_lock); | |
8388 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8389 | spin_unlock(&cluster->refill_lock); | |
8390 | ||
8391 | /* | |
8392 | * make sure this block group isn't part of a metadata | |
8393 | * allocation cluster | |
8394 | */ | |
8395 | cluster = &root->fs_info->meta_alloc_cluster; | |
8396 | spin_lock(&cluster->refill_lock); | |
8397 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8398 | spin_unlock(&cluster->refill_lock); | |
8399 | ||
1a40e23b | 8400 | path = btrfs_alloc_path(); |
d8926bb3 MF |
8401 | if (!path) { |
8402 | ret = -ENOMEM; | |
8403 | goto out; | |
8404 | } | |
1a40e23b | 8405 | |
10b2f34d | 8406 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 8407 | if (!IS_ERR(inode)) { |
b532402e | 8408 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
8409 | if (ret) { |
8410 | btrfs_add_delayed_iput(inode); | |
8411 | goto out; | |
8412 | } | |
0af3d00b JB |
8413 | clear_nlink(inode); |
8414 | /* One for the block groups ref */ | |
8415 | spin_lock(&block_group->lock); | |
8416 | if (block_group->iref) { | |
8417 | block_group->iref = 0; | |
8418 | block_group->inode = NULL; | |
8419 | spin_unlock(&block_group->lock); | |
8420 | iput(inode); | |
8421 | } else { | |
8422 | spin_unlock(&block_group->lock); | |
8423 | } | |
8424 | /* One for our lookup ref */ | |
455757c3 | 8425 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
8426 | } |
8427 | ||
8428 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
8429 | key.offset = block_group->key.objectid; | |
8430 | key.type = 0; | |
8431 | ||
8432 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
8433 | if (ret < 0) | |
8434 | goto out; | |
8435 | if (ret > 0) | |
b3b4aa74 | 8436 | btrfs_release_path(path); |
0af3d00b JB |
8437 | if (ret == 0) { |
8438 | ret = btrfs_del_item(trans, tree_root, path); | |
8439 | if (ret) | |
8440 | goto out; | |
b3b4aa74 | 8441 | btrfs_release_path(path); |
0af3d00b JB |
8442 | } |
8443 | ||
3dfdb934 | 8444 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
8445 | rb_erase(&block_group->cache_node, |
8446 | &root->fs_info->block_group_cache_tree); | |
a1897fdd LB |
8447 | |
8448 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
8449 | root->fs_info->first_logical_byte = (u64)-1; | |
3dfdb934 | 8450 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 8451 | |
80eb234a | 8452 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
8453 | /* |
8454 | * we must use list_del_init so people can check to see if they | |
8455 | * are still on the list after taking the semaphore | |
8456 | */ | |
8457 | list_del_init(&block_group->list); | |
10ea00f5 ID |
8458 | if (list_empty(&block_group->space_info->block_groups[index])) |
8459 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 8460 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 8461 | |
817d52f8 | 8462 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8463 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
8464 | |
8465 | btrfs_remove_free_space_cache(block_group); | |
8466 | ||
c146afad YZ |
8467 | spin_lock(&block_group->space_info->lock); |
8468 | block_group->space_info->total_bytes -= block_group->key.offset; | |
8469 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 8470 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 8471 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 8472 | |
0af3d00b JB |
8473 | memcpy(&key, &block_group->key, sizeof(key)); |
8474 | ||
283bb197 | 8475 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 8476 | |
fa9c0d79 CM |
8477 | btrfs_put_block_group(block_group); |
8478 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
8479 | |
8480 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
8481 | if (ret > 0) | |
8482 | ret = -EIO; | |
8483 | if (ret < 0) | |
8484 | goto out; | |
8485 | ||
8486 | ret = btrfs_del_item(trans, root, path); | |
8487 | out: | |
8488 | btrfs_free_path(path); | |
8489 | return ret; | |
8490 | } | |
acce952b | 8491 | |
c59021f8 | 8492 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
8493 | { | |
8494 | struct btrfs_space_info *space_info; | |
1aba86d6 | 8495 | struct btrfs_super_block *disk_super; |
8496 | u64 features; | |
8497 | u64 flags; | |
8498 | int mixed = 0; | |
c59021f8 | 8499 | int ret; |
8500 | ||
6c41761f | 8501 | disk_super = fs_info->super_copy; |
1aba86d6 | 8502 | if (!btrfs_super_root(disk_super)) |
8503 | return 1; | |
c59021f8 | 8504 | |
1aba86d6 | 8505 | features = btrfs_super_incompat_flags(disk_super); |
8506 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
8507 | mixed = 1; | |
c59021f8 | 8508 | |
1aba86d6 | 8509 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
8510 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 8511 | if (ret) |
1aba86d6 | 8512 | goto out; |
c59021f8 | 8513 | |
1aba86d6 | 8514 | if (mixed) { |
8515 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
8516 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8517 | } else { | |
8518 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
8519 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8520 | if (ret) | |
8521 | goto out; | |
8522 | ||
8523 | flags = BTRFS_BLOCK_GROUP_DATA; | |
8524 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8525 | } | |
8526 | out: | |
c59021f8 | 8527 | return ret; |
8528 | } | |
8529 | ||
acce952b | 8530 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
8531 | { | |
8532 | return unpin_extent_range(root, start, end); | |
8533 | } | |
8534 | ||
8535 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 8536 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 8537 | { |
5378e607 | 8538 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 8539 | } |
f7039b1d LD |
8540 | |
8541 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
8542 | { | |
8543 | struct btrfs_fs_info *fs_info = root->fs_info; | |
8544 | struct btrfs_block_group_cache *cache = NULL; | |
8545 | u64 group_trimmed; | |
8546 | u64 start; | |
8547 | u64 end; | |
8548 | u64 trimmed = 0; | |
2cac13e4 | 8549 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
8550 | int ret = 0; |
8551 | ||
2cac13e4 LB |
8552 | /* |
8553 | * try to trim all FS space, our block group may start from non-zero. | |
8554 | */ | |
8555 | if (range->len == total_bytes) | |
8556 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
8557 | else | |
8558 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
8559 | |
8560 | while (cache) { | |
8561 | if (cache->key.objectid >= (range->start + range->len)) { | |
8562 | btrfs_put_block_group(cache); | |
8563 | break; | |
8564 | } | |
8565 | ||
8566 | start = max(range->start, cache->key.objectid); | |
8567 | end = min(range->start + range->len, | |
8568 | cache->key.objectid + cache->key.offset); | |
8569 | ||
8570 | if (end - start >= range->minlen) { | |
8571 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 8572 | ret = cache_block_group(cache, 0); |
f7039b1d LD |
8573 | if (!ret) |
8574 | wait_block_group_cache_done(cache); | |
8575 | } | |
8576 | ret = btrfs_trim_block_group(cache, | |
8577 | &group_trimmed, | |
8578 | start, | |
8579 | end, | |
8580 | range->minlen); | |
8581 | ||
8582 | trimmed += group_trimmed; | |
8583 | if (ret) { | |
8584 | btrfs_put_block_group(cache); | |
8585 | break; | |
8586 | } | |
8587 | } | |
8588 | ||
8589 | cache = next_block_group(fs_info->tree_root, cache); | |
8590 | } | |
8591 | ||
8592 | range->len = trimmed; | |
8593 | return ret; | |
8594 | } |