Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
edbd8d4e | 19 | #include <linux/pagemap.h> |
ec44a35c | 20 | #include <linux/writeback.h> |
21af804c | 21 | #include <linux/blkdev.h> |
b7a9f29f | 22 | #include <linux/sort.h> |
4184ea7f | 23 | #include <linux/rcupdate.h> |
817d52f8 | 24 | #include <linux/kthread.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
dff51cd1 | 26 | #include <linux/ratelimit.h> |
4b4e25f2 | 27 | #include "compat.h" |
74493f7a | 28 | #include "hash.h" |
fec577fb CM |
29 | #include "ctree.h" |
30 | #include "disk-io.h" | |
31 | #include "print-tree.h" | |
e089f05c | 32 | #include "transaction.h" |
0b86a832 | 33 | #include "volumes.h" |
925baedd | 34 | #include "locking.h" |
fa9c0d79 | 35 | #include "free-space-cache.h" |
fec577fb | 36 | |
9e622d6b MX |
37 | /* |
38 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
39 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
40 | * if we really need one. | |
41 | * | |
0e4f8f88 CM |
42 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
43 | * if we have very few chunks already allocated. This is | |
44 | * used as part of the clustering code to help make sure | |
45 | * we have a good pool of storage to cluster in, without | |
46 | * filling the FS with empty chunks | |
47 | * | |
9e622d6b MX |
48 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
49 | * | |
0e4f8f88 CM |
50 | */ |
51 | enum { | |
52 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
53 | CHUNK_ALLOC_LIMITED = 1, |
54 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
55 | }; |
56 | ||
fb25e914 JB |
57 | /* |
58 | * Control how reservations are dealt with. | |
59 | * | |
60 | * RESERVE_FREE - freeing a reservation. | |
61 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
62 | * ENOSPC accounting | |
63 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
64 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
65 | */ | |
66 | enum { | |
67 | RESERVE_FREE = 0, | |
68 | RESERVE_ALLOC = 1, | |
69 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
70 | }; | |
71 | ||
f3465ca4 JB |
72 | static int update_block_group(struct btrfs_trans_handle *trans, |
73 | struct btrfs_root *root, | |
f0486c68 | 74 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
75 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
76 | struct btrfs_root *root, | |
77 | u64 bytenr, u64 num_bytes, u64 parent, | |
78 | u64 root_objectid, u64 owner_objectid, | |
79 | u64 owner_offset, int refs_to_drop, | |
80 | struct btrfs_delayed_extent_op *extra_op); | |
81 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
82 | struct extent_buffer *leaf, | |
83 | struct btrfs_extent_item *ei); | |
84 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
85 | struct btrfs_root *root, | |
86 | u64 parent, u64 root_objectid, | |
87 | u64 flags, u64 owner, u64 offset, | |
88 | struct btrfs_key *ins, int ref_mod); | |
89 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
90 | struct btrfs_root *root, | |
91 | u64 parent, u64 root_objectid, | |
92 | u64 flags, struct btrfs_disk_key *key, | |
93 | int level, struct btrfs_key *ins); | |
6a63209f JB |
94 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
95 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
96 | u64 flags, int force); | |
11833d66 YZ |
97 | static int find_next_key(struct btrfs_path *path, int level, |
98 | struct btrfs_key *key); | |
9ed74f2d JB |
99 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
100 | int dump_block_groups); | |
fb25e914 JB |
101 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
102 | u64 num_bytes, int reserve); | |
6a63209f | 103 | |
817d52f8 JB |
104 | static noinline int |
105 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
106 | { | |
107 | smp_mb(); | |
108 | return cache->cached == BTRFS_CACHE_FINISHED; | |
109 | } | |
110 | ||
0f9dd46c JB |
111 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
112 | { | |
113 | return (cache->flags & bits) == bits; | |
114 | } | |
115 | ||
62a45b60 | 116 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
117 | { |
118 | atomic_inc(&cache->count); | |
119 | } | |
120 | ||
121 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
122 | { | |
f0486c68 YZ |
123 | if (atomic_dec_and_test(&cache->count)) { |
124 | WARN_ON(cache->pinned > 0); | |
125 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 126 | kfree(cache->free_space_ctl); |
11dfe35a | 127 | kfree(cache); |
f0486c68 | 128 | } |
11dfe35a JB |
129 | } |
130 | ||
0f9dd46c JB |
131 | /* |
132 | * this adds the block group to the fs_info rb tree for the block group | |
133 | * cache | |
134 | */ | |
b2950863 | 135 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
136 | struct btrfs_block_group_cache *block_group) |
137 | { | |
138 | struct rb_node **p; | |
139 | struct rb_node *parent = NULL; | |
140 | struct btrfs_block_group_cache *cache; | |
141 | ||
142 | spin_lock(&info->block_group_cache_lock); | |
143 | p = &info->block_group_cache_tree.rb_node; | |
144 | ||
145 | while (*p) { | |
146 | parent = *p; | |
147 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
148 | cache_node); | |
149 | if (block_group->key.objectid < cache->key.objectid) { | |
150 | p = &(*p)->rb_left; | |
151 | } else if (block_group->key.objectid > cache->key.objectid) { | |
152 | p = &(*p)->rb_right; | |
153 | } else { | |
154 | spin_unlock(&info->block_group_cache_lock); | |
155 | return -EEXIST; | |
156 | } | |
157 | } | |
158 | ||
159 | rb_link_node(&block_group->cache_node, parent, p); | |
160 | rb_insert_color(&block_group->cache_node, | |
161 | &info->block_group_cache_tree); | |
162 | spin_unlock(&info->block_group_cache_lock); | |
163 | ||
164 | return 0; | |
165 | } | |
166 | ||
167 | /* | |
168 | * This will return the block group at or after bytenr if contains is 0, else | |
169 | * it will return the block group that contains the bytenr | |
170 | */ | |
171 | static struct btrfs_block_group_cache * | |
172 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
173 | int contains) | |
174 | { | |
175 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
176 | struct rb_node *n; | |
177 | u64 end, start; | |
178 | ||
179 | spin_lock(&info->block_group_cache_lock); | |
180 | n = info->block_group_cache_tree.rb_node; | |
181 | ||
182 | while (n) { | |
183 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
184 | cache_node); | |
185 | end = cache->key.objectid + cache->key.offset - 1; | |
186 | start = cache->key.objectid; | |
187 | ||
188 | if (bytenr < start) { | |
189 | if (!contains && (!ret || start < ret->key.objectid)) | |
190 | ret = cache; | |
191 | n = n->rb_left; | |
192 | } else if (bytenr > start) { | |
193 | if (contains && bytenr <= end) { | |
194 | ret = cache; | |
195 | break; | |
196 | } | |
197 | n = n->rb_right; | |
198 | } else { | |
199 | ret = cache; | |
200 | break; | |
201 | } | |
202 | } | |
d2fb3437 | 203 | if (ret) |
11dfe35a | 204 | btrfs_get_block_group(ret); |
0f9dd46c JB |
205 | spin_unlock(&info->block_group_cache_lock); |
206 | ||
207 | return ret; | |
208 | } | |
209 | ||
11833d66 YZ |
210 | static int add_excluded_extent(struct btrfs_root *root, |
211 | u64 start, u64 num_bytes) | |
817d52f8 | 212 | { |
11833d66 YZ |
213 | u64 end = start + num_bytes - 1; |
214 | set_extent_bits(&root->fs_info->freed_extents[0], | |
215 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
216 | set_extent_bits(&root->fs_info->freed_extents[1], | |
217 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
218 | return 0; | |
219 | } | |
817d52f8 | 220 | |
11833d66 YZ |
221 | static void free_excluded_extents(struct btrfs_root *root, |
222 | struct btrfs_block_group_cache *cache) | |
223 | { | |
224 | u64 start, end; | |
817d52f8 | 225 | |
11833d66 YZ |
226 | start = cache->key.objectid; |
227 | end = start + cache->key.offset - 1; | |
228 | ||
229 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
230 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
231 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
232 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
233 | } |
234 | ||
11833d66 YZ |
235 | static int exclude_super_stripes(struct btrfs_root *root, |
236 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 237 | { |
817d52f8 JB |
238 | u64 bytenr; |
239 | u64 *logical; | |
240 | int stripe_len; | |
241 | int i, nr, ret; | |
242 | ||
06b2331f YZ |
243 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
244 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
245 | cache->bytes_super += stripe_len; | |
246 | ret = add_excluded_extent(root, cache->key.objectid, | |
247 | stripe_len); | |
248 | BUG_ON(ret); | |
249 | } | |
250 | ||
817d52f8 JB |
251 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
252 | bytenr = btrfs_sb_offset(i); | |
253 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
254 | cache->key.objectid, bytenr, | |
255 | 0, &logical, &nr, &stripe_len); | |
256 | BUG_ON(ret); | |
11833d66 | 257 | |
817d52f8 | 258 | while (nr--) { |
1b2da372 | 259 | cache->bytes_super += stripe_len; |
11833d66 YZ |
260 | ret = add_excluded_extent(root, logical[nr], |
261 | stripe_len); | |
262 | BUG_ON(ret); | |
817d52f8 | 263 | } |
11833d66 | 264 | |
817d52f8 JB |
265 | kfree(logical); |
266 | } | |
817d52f8 JB |
267 | return 0; |
268 | } | |
269 | ||
11833d66 YZ |
270 | static struct btrfs_caching_control * |
271 | get_caching_control(struct btrfs_block_group_cache *cache) | |
272 | { | |
273 | struct btrfs_caching_control *ctl; | |
274 | ||
275 | spin_lock(&cache->lock); | |
276 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
277 | spin_unlock(&cache->lock); | |
278 | return NULL; | |
279 | } | |
280 | ||
dde5abee JB |
281 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
282 | if (!cache->caching_ctl) { | |
283 | spin_unlock(&cache->lock); | |
11833d66 YZ |
284 | return NULL; |
285 | } | |
286 | ||
287 | ctl = cache->caching_ctl; | |
288 | atomic_inc(&ctl->count); | |
289 | spin_unlock(&cache->lock); | |
290 | return ctl; | |
291 | } | |
292 | ||
293 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
294 | { | |
295 | if (atomic_dec_and_test(&ctl->count)) | |
296 | kfree(ctl); | |
297 | } | |
298 | ||
0f9dd46c JB |
299 | /* |
300 | * this is only called by cache_block_group, since we could have freed extents | |
301 | * we need to check the pinned_extents for any extents that can't be used yet | |
302 | * since their free space will be released as soon as the transaction commits. | |
303 | */ | |
817d52f8 | 304 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
305 | struct btrfs_fs_info *info, u64 start, u64 end) |
306 | { | |
817d52f8 | 307 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
308 | int ret; |
309 | ||
310 | while (start < end) { | |
11833d66 | 311 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 312 | &extent_start, &extent_end, |
11833d66 | 313 | EXTENT_DIRTY | EXTENT_UPTODATE); |
0f9dd46c JB |
314 | if (ret) |
315 | break; | |
316 | ||
06b2331f | 317 | if (extent_start <= start) { |
0f9dd46c JB |
318 | start = extent_end + 1; |
319 | } else if (extent_start > start && extent_start < end) { | |
320 | size = extent_start - start; | |
817d52f8 | 321 | total_added += size; |
ea6a478e JB |
322 | ret = btrfs_add_free_space(block_group, start, |
323 | size); | |
0f9dd46c JB |
324 | BUG_ON(ret); |
325 | start = extent_end + 1; | |
326 | } else { | |
327 | break; | |
328 | } | |
329 | } | |
330 | ||
331 | if (start < end) { | |
332 | size = end - start; | |
817d52f8 | 333 | total_added += size; |
ea6a478e | 334 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
335 | BUG_ON(ret); |
336 | } | |
337 | ||
817d52f8 | 338 | return total_added; |
0f9dd46c JB |
339 | } |
340 | ||
bab39bf9 | 341 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 342 | { |
bab39bf9 JB |
343 | struct btrfs_block_group_cache *block_group; |
344 | struct btrfs_fs_info *fs_info; | |
345 | struct btrfs_caching_control *caching_ctl; | |
346 | struct btrfs_root *extent_root; | |
e37c9e69 | 347 | struct btrfs_path *path; |
5f39d397 | 348 | struct extent_buffer *leaf; |
11833d66 | 349 | struct btrfs_key key; |
817d52f8 | 350 | u64 total_found = 0; |
11833d66 YZ |
351 | u64 last = 0; |
352 | u32 nritems; | |
353 | int ret = 0; | |
f510cfec | 354 | |
bab39bf9 JB |
355 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
356 | block_group = caching_ctl->block_group; | |
357 | fs_info = block_group->fs_info; | |
358 | extent_root = fs_info->extent_root; | |
359 | ||
e37c9e69 CM |
360 | path = btrfs_alloc_path(); |
361 | if (!path) | |
bab39bf9 | 362 | goto out; |
7d7d6068 | 363 | |
817d52f8 | 364 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 365 | |
5cd57b2c | 366 | /* |
817d52f8 JB |
367 | * We don't want to deadlock with somebody trying to allocate a new |
368 | * extent for the extent root while also trying to search the extent | |
369 | * root to add free space. So we skip locking and search the commit | |
370 | * root, since its read-only | |
5cd57b2c CM |
371 | */ |
372 | path->skip_locking = 1; | |
817d52f8 | 373 | path->search_commit_root = 1; |
026fd317 | 374 | path->reada = 1; |
817d52f8 | 375 | |
e4404d6e | 376 | key.objectid = last; |
e37c9e69 | 377 | key.offset = 0; |
11833d66 | 378 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 379 | again: |
11833d66 | 380 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
381 | /* need to make sure the commit_root doesn't disappear */ |
382 | down_read(&fs_info->extent_commit_sem); | |
383 | ||
11833d66 | 384 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 385 | if (ret < 0) |
ef8bbdfe | 386 | goto err; |
a512bbf8 | 387 | |
11833d66 YZ |
388 | leaf = path->nodes[0]; |
389 | nritems = btrfs_header_nritems(leaf); | |
390 | ||
d397712b | 391 | while (1) { |
7841cb28 | 392 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 393 | last = (u64)-1; |
817d52f8 | 394 | break; |
f25784b3 | 395 | } |
817d52f8 | 396 | |
11833d66 YZ |
397 | if (path->slots[0] < nritems) { |
398 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
399 | } else { | |
400 | ret = find_next_key(path, 0, &key); | |
401 | if (ret) | |
e37c9e69 | 402 | break; |
817d52f8 | 403 | |
589d8ade JB |
404 | if (need_resched() || |
405 | btrfs_next_leaf(extent_root, path)) { | |
406 | caching_ctl->progress = last; | |
ff5714cc | 407 | btrfs_release_path(path); |
589d8ade JB |
408 | up_read(&fs_info->extent_commit_sem); |
409 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 410 | cond_resched(); |
589d8ade JB |
411 | goto again; |
412 | } | |
413 | leaf = path->nodes[0]; | |
414 | nritems = btrfs_header_nritems(leaf); | |
415 | continue; | |
11833d66 | 416 | } |
817d52f8 | 417 | |
11833d66 YZ |
418 | if (key.objectid < block_group->key.objectid) { |
419 | path->slots[0]++; | |
817d52f8 | 420 | continue; |
e37c9e69 | 421 | } |
0f9dd46c | 422 | |
e37c9e69 | 423 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 424 | block_group->key.offset) |
e37c9e69 | 425 | break; |
7d7d6068 | 426 | |
11833d66 | 427 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
428 | total_found += add_new_free_space(block_group, |
429 | fs_info, last, | |
430 | key.objectid); | |
7d7d6068 | 431 | last = key.objectid + key.offset; |
817d52f8 | 432 | |
11833d66 YZ |
433 | if (total_found > (1024 * 1024 * 2)) { |
434 | total_found = 0; | |
435 | wake_up(&caching_ctl->wait); | |
436 | } | |
817d52f8 | 437 | } |
e37c9e69 CM |
438 | path->slots[0]++; |
439 | } | |
817d52f8 | 440 | ret = 0; |
e37c9e69 | 441 | |
817d52f8 JB |
442 | total_found += add_new_free_space(block_group, fs_info, last, |
443 | block_group->key.objectid + | |
444 | block_group->key.offset); | |
11833d66 | 445 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
446 | |
447 | spin_lock(&block_group->lock); | |
11833d66 | 448 | block_group->caching_ctl = NULL; |
817d52f8 JB |
449 | block_group->cached = BTRFS_CACHE_FINISHED; |
450 | spin_unlock(&block_group->lock); | |
0f9dd46c | 451 | |
54aa1f4d | 452 | err: |
e37c9e69 | 453 | btrfs_free_path(path); |
276e680d | 454 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 455 | |
11833d66 YZ |
456 | free_excluded_extents(extent_root, block_group); |
457 | ||
458 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 459 | out: |
11833d66 YZ |
460 | wake_up(&caching_ctl->wait); |
461 | ||
462 | put_caching_control(caching_ctl); | |
11dfe35a | 463 | btrfs_put_block_group(block_group); |
817d52f8 JB |
464 | } |
465 | ||
9d66e233 JB |
466 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
467 | struct btrfs_trans_handle *trans, | |
b8399dee | 468 | struct btrfs_root *root, |
9d66e233 | 469 | int load_cache_only) |
817d52f8 | 470 | { |
291c7d2f | 471 | DEFINE_WAIT(wait); |
11833d66 YZ |
472 | struct btrfs_fs_info *fs_info = cache->fs_info; |
473 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
474 | int ret = 0; |
475 | ||
291c7d2f JB |
476 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
477 | BUG_ON(!caching_ctl); | |
478 | ||
479 | INIT_LIST_HEAD(&caching_ctl->list); | |
480 | mutex_init(&caching_ctl->mutex); | |
481 | init_waitqueue_head(&caching_ctl->wait); | |
482 | caching_ctl->block_group = cache; | |
483 | caching_ctl->progress = cache->key.objectid; | |
484 | atomic_set(&caching_ctl->count, 1); | |
485 | caching_ctl->work.func = caching_thread; | |
486 | ||
487 | spin_lock(&cache->lock); | |
488 | /* | |
489 | * This should be a rare occasion, but this could happen I think in the | |
490 | * case where one thread starts to load the space cache info, and then | |
491 | * some other thread starts a transaction commit which tries to do an | |
492 | * allocation while the other thread is still loading the space cache | |
493 | * info. The previous loop should have kept us from choosing this block | |
494 | * group, but if we've moved to the state where we will wait on caching | |
495 | * block groups we need to first check if we're doing a fast load here, | |
496 | * so we can wait for it to finish, otherwise we could end up allocating | |
497 | * from a block group who's cache gets evicted for one reason or | |
498 | * another. | |
499 | */ | |
500 | while (cache->cached == BTRFS_CACHE_FAST) { | |
501 | struct btrfs_caching_control *ctl; | |
502 | ||
503 | ctl = cache->caching_ctl; | |
504 | atomic_inc(&ctl->count); | |
505 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
506 | spin_unlock(&cache->lock); | |
507 | ||
508 | schedule(); | |
509 | ||
510 | finish_wait(&ctl->wait, &wait); | |
511 | put_caching_control(ctl); | |
512 | spin_lock(&cache->lock); | |
513 | } | |
514 | ||
515 | if (cache->cached != BTRFS_CACHE_NO) { | |
516 | spin_unlock(&cache->lock); | |
517 | kfree(caching_ctl); | |
11833d66 | 518 | return 0; |
291c7d2f JB |
519 | } |
520 | WARN_ON(cache->caching_ctl); | |
521 | cache->caching_ctl = caching_ctl; | |
522 | cache->cached = BTRFS_CACHE_FAST; | |
523 | spin_unlock(&cache->lock); | |
11833d66 | 524 | |
9d66e233 JB |
525 | /* |
526 | * We can't do the read from on-disk cache during a commit since we need | |
b8399dee JB |
527 | * to have the normal tree locking. Also if we are currently trying to |
528 | * allocate blocks for the tree root we can't do the fast caching since | |
529 | * we likely hold important locks. | |
9d66e233 | 530 | */ |
f7039b1d | 531 | if (trans && (!trans->transaction->in_commit) && |
73bc1876 JB |
532 | (root && root != root->fs_info->tree_root) && |
533 | btrfs_test_opt(root, SPACE_CACHE)) { | |
9d66e233 JB |
534 | ret = load_free_space_cache(fs_info, cache); |
535 | ||
536 | spin_lock(&cache->lock); | |
537 | if (ret == 1) { | |
291c7d2f | 538 | cache->caching_ctl = NULL; |
9d66e233 JB |
539 | cache->cached = BTRFS_CACHE_FINISHED; |
540 | cache->last_byte_to_unpin = (u64)-1; | |
541 | } else { | |
291c7d2f JB |
542 | if (load_cache_only) { |
543 | cache->caching_ctl = NULL; | |
544 | cache->cached = BTRFS_CACHE_NO; | |
545 | } else { | |
546 | cache->cached = BTRFS_CACHE_STARTED; | |
547 | } | |
9d66e233 JB |
548 | } |
549 | spin_unlock(&cache->lock); | |
291c7d2f | 550 | wake_up(&caching_ctl->wait); |
3c14874a | 551 | if (ret == 1) { |
291c7d2f | 552 | put_caching_control(caching_ctl); |
3c14874a | 553 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 554 | return 0; |
3c14874a | 555 | } |
291c7d2f JB |
556 | } else { |
557 | /* | |
558 | * We are not going to do the fast caching, set cached to the | |
559 | * appropriate value and wakeup any waiters. | |
560 | */ | |
561 | spin_lock(&cache->lock); | |
562 | if (load_cache_only) { | |
563 | cache->caching_ctl = NULL; | |
564 | cache->cached = BTRFS_CACHE_NO; | |
565 | } else { | |
566 | cache->cached = BTRFS_CACHE_STARTED; | |
567 | } | |
568 | spin_unlock(&cache->lock); | |
569 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
570 | } |
571 | ||
291c7d2f JB |
572 | if (load_cache_only) { |
573 | put_caching_control(caching_ctl); | |
11833d66 | 574 | return 0; |
817d52f8 | 575 | } |
817d52f8 | 576 | |
11833d66 | 577 | down_write(&fs_info->extent_commit_sem); |
291c7d2f | 578 | atomic_inc(&caching_ctl->count); |
11833d66 YZ |
579 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
580 | up_write(&fs_info->extent_commit_sem); | |
581 | ||
11dfe35a | 582 | btrfs_get_block_group(cache); |
11833d66 | 583 | |
bab39bf9 | 584 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 585 | |
ef8bbdfe | 586 | return ret; |
e37c9e69 CM |
587 | } |
588 | ||
0f9dd46c JB |
589 | /* |
590 | * return the block group that starts at or after bytenr | |
591 | */ | |
d397712b CM |
592 | static struct btrfs_block_group_cache * |
593 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 594 | { |
0f9dd46c | 595 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 596 | |
0f9dd46c | 597 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 598 | |
0f9dd46c | 599 | return cache; |
0ef3e66b CM |
600 | } |
601 | ||
0f9dd46c | 602 | /* |
9f55684c | 603 | * return the block group that contains the given bytenr |
0f9dd46c | 604 | */ |
d397712b CM |
605 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
606 | struct btrfs_fs_info *info, | |
607 | u64 bytenr) | |
be744175 | 608 | { |
0f9dd46c | 609 | struct btrfs_block_group_cache *cache; |
be744175 | 610 | |
0f9dd46c | 611 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 612 | |
0f9dd46c | 613 | return cache; |
be744175 | 614 | } |
0b86a832 | 615 | |
0f9dd46c JB |
616 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
617 | u64 flags) | |
6324fbf3 | 618 | { |
0f9dd46c | 619 | struct list_head *head = &info->space_info; |
0f9dd46c | 620 | struct btrfs_space_info *found; |
4184ea7f | 621 | |
52ba6929 | 622 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 623 | |
4184ea7f CM |
624 | rcu_read_lock(); |
625 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 626 | if (found->flags & flags) { |
4184ea7f | 627 | rcu_read_unlock(); |
0f9dd46c | 628 | return found; |
4184ea7f | 629 | } |
0f9dd46c | 630 | } |
4184ea7f | 631 | rcu_read_unlock(); |
0f9dd46c | 632 | return NULL; |
6324fbf3 CM |
633 | } |
634 | ||
4184ea7f CM |
635 | /* |
636 | * after adding space to the filesystem, we need to clear the full flags | |
637 | * on all the space infos. | |
638 | */ | |
639 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
640 | { | |
641 | struct list_head *head = &info->space_info; | |
642 | struct btrfs_space_info *found; | |
643 | ||
644 | rcu_read_lock(); | |
645 | list_for_each_entry_rcu(found, head, list) | |
646 | found->full = 0; | |
647 | rcu_read_unlock(); | |
648 | } | |
649 | ||
80eb234a JB |
650 | static u64 div_factor(u64 num, int factor) |
651 | { | |
652 | if (factor == 10) | |
653 | return num; | |
654 | num *= factor; | |
655 | do_div(num, 10); | |
656 | return num; | |
657 | } | |
658 | ||
e5bc2458 CM |
659 | static u64 div_factor_fine(u64 num, int factor) |
660 | { | |
661 | if (factor == 100) | |
662 | return num; | |
663 | num *= factor; | |
664 | do_div(num, 100); | |
665 | return num; | |
666 | } | |
667 | ||
d2fb3437 YZ |
668 | u64 btrfs_find_block_group(struct btrfs_root *root, |
669 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 670 | { |
96b5179d | 671 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 672 | u64 used; |
d2fb3437 YZ |
673 | u64 last = max(search_hint, search_start); |
674 | u64 group_start = 0; | |
31f3c99b | 675 | int full_search = 0; |
d2fb3437 | 676 | int factor = 9; |
0ef3e66b | 677 | int wrapped = 0; |
31f3c99b | 678 | again: |
e8569813 ZY |
679 | while (1) { |
680 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
681 | if (!cache) |
682 | break; | |
96b5179d | 683 | |
c286ac48 | 684 | spin_lock(&cache->lock); |
96b5179d CM |
685 | last = cache->key.objectid + cache->key.offset; |
686 | used = btrfs_block_group_used(&cache->item); | |
687 | ||
d2fb3437 YZ |
688 | if ((full_search || !cache->ro) && |
689 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 690 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
691 | div_factor(cache->key.offset, factor)) { |
692 | group_start = cache->key.objectid; | |
c286ac48 | 693 | spin_unlock(&cache->lock); |
fa9c0d79 | 694 | btrfs_put_block_group(cache); |
8790d502 CM |
695 | goto found; |
696 | } | |
6324fbf3 | 697 | } |
c286ac48 | 698 | spin_unlock(&cache->lock); |
fa9c0d79 | 699 | btrfs_put_block_group(cache); |
de428b63 | 700 | cond_resched(); |
cd1bc465 | 701 | } |
0ef3e66b CM |
702 | if (!wrapped) { |
703 | last = search_start; | |
704 | wrapped = 1; | |
705 | goto again; | |
706 | } | |
707 | if (!full_search && factor < 10) { | |
be744175 | 708 | last = search_start; |
31f3c99b | 709 | full_search = 1; |
0ef3e66b | 710 | factor = 10; |
31f3c99b CM |
711 | goto again; |
712 | } | |
be744175 | 713 | found: |
d2fb3437 | 714 | return group_start; |
925baedd | 715 | } |
0f9dd46c | 716 | |
e02119d5 | 717 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 718 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
719 | { |
720 | int ret; | |
721 | struct btrfs_key key; | |
31840ae1 | 722 | struct btrfs_path *path; |
e02119d5 | 723 | |
31840ae1 | 724 | path = btrfs_alloc_path(); |
d8926bb3 MF |
725 | if (!path) |
726 | return -ENOMEM; | |
727 | ||
e02119d5 CM |
728 | key.objectid = start; |
729 | key.offset = len; | |
730 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
731 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
732 | 0, 0); | |
31840ae1 | 733 | btrfs_free_path(path); |
7bb86316 CM |
734 | return ret; |
735 | } | |
736 | ||
a22285a6 YZ |
737 | /* |
738 | * helper function to lookup reference count and flags of extent. | |
739 | * | |
740 | * the head node for delayed ref is used to store the sum of all the | |
741 | * reference count modifications queued up in the rbtree. the head | |
742 | * node may also store the extent flags to set. This way you can check | |
743 | * to see what the reference count and extent flags would be if all of | |
744 | * the delayed refs are not processed. | |
745 | */ | |
746 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
747 | struct btrfs_root *root, u64 bytenr, | |
748 | u64 num_bytes, u64 *refs, u64 *flags) | |
749 | { | |
750 | struct btrfs_delayed_ref_head *head; | |
751 | struct btrfs_delayed_ref_root *delayed_refs; | |
752 | struct btrfs_path *path; | |
753 | struct btrfs_extent_item *ei; | |
754 | struct extent_buffer *leaf; | |
755 | struct btrfs_key key; | |
756 | u32 item_size; | |
757 | u64 num_refs; | |
758 | u64 extent_flags; | |
759 | int ret; | |
760 | ||
761 | path = btrfs_alloc_path(); | |
762 | if (!path) | |
763 | return -ENOMEM; | |
764 | ||
765 | key.objectid = bytenr; | |
766 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
767 | key.offset = num_bytes; | |
768 | if (!trans) { | |
769 | path->skip_locking = 1; | |
770 | path->search_commit_root = 1; | |
771 | } | |
772 | again: | |
773 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
774 | &key, path, 0, 0); | |
775 | if (ret < 0) | |
776 | goto out_free; | |
777 | ||
778 | if (ret == 0) { | |
779 | leaf = path->nodes[0]; | |
780 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
781 | if (item_size >= sizeof(*ei)) { | |
782 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
783 | struct btrfs_extent_item); | |
784 | num_refs = btrfs_extent_refs(leaf, ei); | |
785 | extent_flags = btrfs_extent_flags(leaf, ei); | |
786 | } else { | |
787 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
788 | struct btrfs_extent_item_v0 *ei0; | |
789 | BUG_ON(item_size != sizeof(*ei0)); | |
790 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
791 | struct btrfs_extent_item_v0); | |
792 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
793 | /* FIXME: this isn't correct for data */ | |
794 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
795 | #else | |
796 | BUG(); | |
797 | #endif | |
798 | } | |
799 | BUG_ON(num_refs == 0); | |
800 | } else { | |
801 | num_refs = 0; | |
802 | extent_flags = 0; | |
803 | ret = 0; | |
804 | } | |
805 | ||
806 | if (!trans) | |
807 | goto out; | |
808 | ||
809 | delayed_refs = &trans->transaction->delayed_refs; | |
810 | spin_lock(&delayed_refs->lock); | |
811 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
812 | if (head) { | |
813 | if (!mutex_trylock(&head->mutex)) { | |
814 | atomic_inc(&head->node.refs); | |
815 | spin_unlock(&delayed_refs->lock); | |
816 | ||
b3b4aa74 | 817 | btrfs_release_path(path); |
a22285a6 | 818 | |
8cc33e5c DS |
819 | /* |
820 | * Mutex was contended, block until it's released and try | |
821 | * again | |
822 | */ | |
a22285a6 YZ |
823 | mutex_lock(&head->mutex); |
824 | mutex_unlock(&head->mutex); | |
825 | btrfs_put_delayed_ref(&head->node); | |
826 | goto again; | |
827 | } | |
828 | if (head->extent_op && head->extent_op->update_flags) | |
829 | extent_flags |= head->extent_op->flags_to_set; | |
830 | else | |
831 | BUG_ON(num_refs == 0); | |
832 | ||
833 | num_refs += head->node.ref_mod; | |
834 | mutex_unlock(&head->mutex); | |
835 | } | |
836 | spin_unlock(&delayed_refs->lock); | |
837 | out: | |
838 | WARN_ON(num_refs == 0); | |
839 | if (refs) | |
840 | *refs = num_refs; | |
841 | if (flags) | |
842 | *flags = extent_flags; | |
843 | out_free: | |
844 | btrfs_free_path(path); | |
845 | return ret; | |
846 | } | |
847 | ||
d8d5f3e1 CM |
848 | /* |
849 | * Back reference rules. Back refs have three main goals: | |
850 | * | |
851 | * 1) differentiate between all holders of references to an extent so that | |
852 | * when a reference is dropped we can make sure it was a valid reference | |
853 | * before freeing the extent. | |
854 | * | |
855 | * 2) Provide enough information to quickly find the holders of an extent | |
856 | * if we notice a given block is corrupted or bad. | |
857 | * | |
858 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
859 | * maintenance. This is actually the same as #2, but with a slightly | |
860 | * different use case. | |
861 | * | |
5d4f98a2 YZ |
862 | * There are two kinds of back refs. The implicit back refs is optimized |
863 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
864 | * back refs of this kind provide information about the block's owner tree | |
865 | * and the pointer's key. These information allow us to find the block by | |
866 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
867 | * referenced by their owner trees. The location of tree block is recorded | |
868 | * in the back refs. Actually the full back refs is generic, and can be | |
869 | * used in all cases the implicit back refs is used. The major shortcoming | |
870 | * of the full back refs is its overhead. Every time a tree block gets | |
871 | * COWed, we have to update back refs entry for all pointers in it. | |
872 | * | |
873 | * For a newly allocated tree block, we use implicit back refs for | |
874 | * pointers in it. This means most tree related operations only involve | |
875 | * implicit back refs. For a tree block created in old transaction, the | |
876 | * only way to drop a reference to it is COW it. So we can detect the | |
877 | * event that tree block loses its owner tree's reference and do the | |
878 | * back refs conversion. | |
879 | * | |
880 | * When a tree block is COW'd through a tree, there are four cases: | |
881 | * | |
882 | * The reference count of the block is one and the tree is the block's | |
883 | * owner tree. Nothing to do in this case. | |
884 | * | |
885 | * The reference count of the block is one and the tree is not the | |
886 | * block's owner tree. In this case, full back refs is used for pointers | |
887 | * in the block. Remove these full back refs, add implicit back refs for | |
888 | * every pointers in the new block. | |
889 | * | |
890 | * The reference count of the block is greater than one and the tree is | |
891 | * the block's owner tree. In this case, implicit back refs is used for | |
892 | * pointers in the block. Add full back refs for every pointers in the | |
893 | * block, increase lower level extents' reference counts. The original | |
894 | * implicit back refs are entailed to the new block. | |
895 | * | |
896 | * The reference count of the block is greater than one and the tree is | |
897 | * not the block's owner tree. Add implicit back refs for every pointer in | |
898 | * the new block, increase lower level extents' reference count. | |
899 | * | |
900 | * Back Reference Key composing: | |
901 | * | |
902 | * The key objectid corresponds to the first byte in the extent, | |
903 | * The key type is used to differentiate between types of back refs. | |
904 | * There are different meanings of the key offset for different types | |
905 | * of back refs. | |
906 | * | |
d8d5f3e1 CM |
907 | * File extents can be referenced by: |
908 | * | |
909 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 910 | * - different files inside a single subvolume |
d8d5f3e1 CM |
911 | * - different offsets inside a file (bookend extents in file.c) |
912 | * | |
5d4f98a2 | 913 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
914 | * |
915 | * - Objectid of the subvolume root | |
d8d5f3e1 | 916 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
917 | * - original offset in the file |
918 | * - how many bookend extents | |
d8d5f3e1 | 919 | * |
5d4f98a2 YZ |
920 | * The key offset for the implicit back refs is hash of the first |
921 | * three fields. | |
d8d5f3e1 | 922 | * |
5d4f98a2 | 923 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 924 | * |
5d4f98a2 | 925 | * - number of pointers in the tree leaf |
d8d5f3e1 | 926 | * |
5d4f98a2 YZ |
927 | * The key offset for the implicit back refs is the first byte of |
928 | * the tree leaf | |
d8d5f3e1 | 929 | * |
5d4f98a2 YZ |
930 | * When a file extent is allocated, The implicit back refs is used. |
931 | * the fields are filled in: | |
d8d5f3e1 | 932 | * |
5d4f98a2 | 933 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 934 | * |
5d4f98a2 YZ |
935 | * When a file extent is removed file truncation, we find the |
936 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 937 | * |
5d4f98a2 | 938 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 939 | * |
5d4f98a2 | 940 | * Btree extents can be referenced by: |
d8d5f3e1 | 941 | * |
5d4f98a2 | 942 | * - Different subvolumes |
d8d5f3e1 | 943 | * |
5d4f98a2 YZ |
944 | * Both the implicit back refs and the full back refs for tree blocks |
945 | * only consist of key. The key offset for the implicit back refs is | |
946 | * objectid of block's owner tree. The key offset for the full back refs | |
947 | * is the first byte of parent block. | |
d8d5f3e1 | 948 | * |
5d4f98a2 YZ |
949 | * When implicit back refs is used, information about the lowest key and |
950 | * level of the tree block are required. These information are stored in | |
951 | * tree block info structure. | |
d8d5f3e1 | 952 | */ |
31840ae1 | 953 | |
5d4f98a2 YZ |
954 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
955 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
956 | struct btrfs_root *root, | |
957 | struct btrfs_path *path, | |
958 | u64 owner, u32 extra_size) | |
7bb86316 | 959 | { |
5d4f98a2 YZ |
960 | struct btrfs_extent_item *item; |
961 | struct btrfs_extent_item_v0 *ei0; | |
962 | struct btrfs_extent_ref_v0 *ref0; | |
963 | struct btrfs_tree_block_info *bi; | |
964 | struct extent_buffer *leaf; | |
7bb86316 | 965 | struct btrfs_key key; |
5d4f98a2 YZ |
966 | struct btrfs_key found_key; |
967 | u32 new_size = sizeof(*item); | |
968 | u64 refs; | |
969 | int ret; | |
970 | ||
971 | leaf = path->nodes[0]; | |
972 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
973 | ||
974 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
975 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
976 | struct btrfs_extent_item_v0); | |
977 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
978 | ||
979 | if (owner == (u64)-1) { | |
980 | while (1) { | |
981 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
982 | ret = btrfs_next_leaf(root, path); | |
983 | if (ret < 0) | |
984 | return ret; | |
985 | BUG_ON(ret > 0); | |
986 | leaf = path->nodes[0]; | |
987 | } | |
988 | btrfs_item_key_to_cpu(leaf, &found_key, | |
989 | path->slots[0]); | |
990 | BUG_ON(key.objectid != found_key.objectid); | |
991 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
992 | path->slots[0]++; | |
993 | continue; | |
994 | } | |
995 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
996 | struct btrfs_extent_ref_v0); | |
997 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
998 | break; | |
999 | } | |
1000 | } | |
b3b4aa74 | 1001 | btrfs_release_path(path); |
5d4f98a2 YZ |
1002 | |
1003 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1004 | new_size += sizeof(*bi); | |
1005 | ||
1006 | new_size -= sizeof(*ei0); | |
1007 | ret = btrfs_search_slot(trans, root, &key, path, | |
1008 | new_size + extra_size, 1); | |
1009 | if (ret < 0) | |
1010 | return ret; | |
1011 | BUG_ON(ret); | |
1012 | ||
1013 | ret = btrfs_extend_item(trans, root, path, new_size); | |
5d4f98a2 YZ |
1014 | |
1015 | leaf = path->nodes[0]; | |
1016 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1017 | btrfs_set_extent_refs(leaf, item, refs); | |
1018 | /* FIXME: get real generation */ | |
1019 | btrfs_set_extent_generation(leaf, item, 0); | |
1020 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1021 | btrfs_set_extent_flags(leaf, item, | |
1022 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1023 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1024 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1025 | /* FIXME: get first key of the block */ | |
1026 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1027 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1028 | } else { | |
1029 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1030 | } | |
1031 | btrfs_mark_buffer_dirty(leaf); | |
1032 | return 0; | |
1033 | } | |
1034 | #endif | |
1035 | ||
1036 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1037 | { | |
1038 | u32 high_crc = ~(u32)0; | |
1039 | u32 low_crc = ~(u32)0; | |
1040 | __le64 lenum; | |
1041 | ||
1042 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 1043 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1044 | lenum = cpu_to_le64(owner); |
163e783e | 1045 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1046 | lenum = cpu_to_le64(offset); |
163e783e | 1047 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1048 | |
1049 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1050 | } | |
1051 | ||
1052 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1053 | struct btrfs_extent_data_ref *ref) | |
1054 | { | |
1055 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1056 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1057 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1058 | } | |
1059 | ||
1060 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1061 | struct btrfs_extent_data_ref *ref, | |
1062 | u64 root_objectid, u64 owner, u64 offset) | |
1063 | { | |
1064 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1065 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1066 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1067 | return 0; | |
1068 | return 1; | |
1069 | } | |
1070 | ||
1071 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1072 | struct btrfs_root *root, | |
1073 | struct btrfs_path *path, | |
1074 | u64 bytenr, u64 parent, | |
1075 | u64 root_objectid, | |
1076 | u64 owner, u64 offset) | |
1077 | { | |
1078 | struct btrfs_key key; | |
1079 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1080 | struct extent_buffer *leaf; |
5d4f98a2 | 1081 | u32 nritems; |
74493f7a | 1082 | int ret; |
5d4f98a2 YZ |
1083 | int recow; |
1084 | int err = -ENOENT; | |
74493f7a | 1085 | |
31840ae1 | 1086 | key.objectid = bytenr; |
5d4f98a2 YZ |
1087 | if (parent) { |
1088 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1089 | key.offset = parent; | |
1090 | } else { | |
1091 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1092 | key.offset = hash_extent_data_ref(root_objectid, | |
1093 | owner, offset); | |
1094 | } | |
1095 | again: | |
1096 | recow = 0; | |
1097 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1098 | if (ret < 0) { | |
1099 | err = ret; | |
1100 | goto fail; | |
1101 | } | |
31840ae1 | 1102 | |
5d4f98a2 YZ |
1103 | if (parent) { |
1104 | if (!ret) | |
1105 | return 0; | |
1106 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1107 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1108 | btrfs_release_path(path); |
5d4f98a2 YZ |
1109 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1110 | if (ret < 0) { | |
1111 | err = ret; | |
1112 | goto fail; | |
1113 | } | |
1114 | if (!ret) | |
1115 | return 0; | |
1116 | #endif | |
1117 | goto fail; | |
31840ae1 ZY |
1118 | } |
1119 | ||
1120 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1121 | nritems = btrfs_header_nritems(leaf); |
1122 | while (1) { | |
1123 | if (path->slots[0] >= nritems) { | |
1124 | ret = btrfs_next_leaf(root, path); | |
1125 | if (ret < 0) | |
1126 | err = ret; | |
1127 | if (ret) | |
1128 | goto fail; | |
1129 | ||
1130 | leaf = path->nodes[0]; | |
1131 | nritems = btrfs_header_nritems(leaf); | |
1132 | recow = 1; | |
1133 | } | |
1134 | ||
1135 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1136 | if (key.objectid != bytenr || | |
1137 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1138 | goto fail; | |
1139 | ||
1140 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1141 | struct btrfs_extent_data_ref); | |
1142 | ||
1143 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1144 | owner, offset)) { | |
1145 | if (recow) { | |
b3b4aa74 | 1146 | btrfs_release_path(path); |
5d4f98a2 YZ |
1147 | goto again; |
1148 | } | |
1149 | err = 0; | |
1150 | break; | |
1151 | } | |
1152 | path->slots[0]++; | |
31840ae1 | 1153 | } |
5d4f98a2 YZ |
1154 | fail: |
1155 | return err; | |
31840ae1 ZY |
1156 | } |
1157 | ||
5d4f98a2 YZ |
1158 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1159 | struct btrfs_root *root, | |
1160 | struct btrfs_path *path, | |
1161 | u64 bytenr, u64 parent, | |
1162 | u64 root_objectid, u64 owner, | |
1163 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1164 | { |
1165 | struct btrfs_key key; | |
1166 | struct extent_buffer *leaf; | |
5d4f98a2 | 1167 | u32 size; |
31840ae1 ZY |
1168 | u32 num_refs; |
1169 | int ret; | |
74493f7a | 1170 | |
74493f7a | 1171 | key.objectid = bytenr; |
5d4f98a2 YZ |
1172 | if (parent) { |
1173 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1174 | key.offset = parent; | |
1175 | size = sizeof(struct btrfs_shared_data_ref); | |
1176 | } else { | |
1177 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1178 | key.offset = hash_extent_data_ref(root_objectid, | |
1179 | owner, offset); | |
1180 | size = sizeof(struct btrfs_extent_data_ref); | |
1181 | } | |
74493f7a | 1182 | |
5d4f98a2 YZ |
1183 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1184 | if (ret && ret != -EEXIST) | |
1185 | goto fail; | |
1186 | ||
1187 | leaf = path->nodes[0]; | |
1188 | if (parent) { | |
1189 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1190 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1191 | struct btrfs_shared_data_ref); |
1192 | if (ret == 0) { | |
1193 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1194 | } else { | |
1195 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1196 | num_refs += refs_to_add; | |
1197 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1198 | } |
5d4f98a2 YZ |
1199 | } else { |
1200 | struct btrfs_extent_data_ref *ref; | |
1201 | while (ret == -EEXIST) { | |
1202 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1203 | struct btrfs_extent_data_ref); | |
1204 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1205 | owner, offset)) | |
1206 | break; | |
b3b4aa74 | 1207 | btrfs_release_path(path); |
5d4f98a2 YZ |
1208 | key.offset++; |
1209 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1210 | size); | |
1211 | if (ret && ret != -EEXIST) | |
1212 | goto fail; | |
31840ae1 | 1213 | |
5d4f98a2 YZ |
1214 | leaf = path->nodes[0]; |
1215 | } | |
1216 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1217 | struct btrfs_extent_data_ref); | |
1218 | if (ret == 0) { | |
1219 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1220 | root_objectid); | |
1221 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1222 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1223 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1224 | } else { | |
1225 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1226 | num_refs += refs_to_add; | |
1227 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1228 | } |
31840ae1 | 1229 | } |
5d4f98a2 YZ |
1230 | btrfs_mark_buffer_dirty(leaf); |
1231 | ret = 0; | |
1232 | fail: | |
b3b4aa74 | 1233 | btrfs_release_path(path); |
7bb86316 | 1234 | return ret; |
74493f7a CM |
1235 | } |
1236 | ||
5d4f98a2 YZ |
1237 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1238 | struct btrfs_root *root, | |
1239 | struct btrfs_path *path, | |
1240 | int refs_to_drop) | |
31840ae1 | 1241 | { |
5d4f98a2 YZ |
1242 | struct btrfs_key key; |
1243 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1244 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1245 | struct extent_buffer *leaf; |
5d4f98a2 | 1246 | u32 num_refs = 0; |
31840ae1 ZY |
1247 | int ret = 0; |
1248 | ||
1249 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1250 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1251 | ||
1252 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1253 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1254 | struct btrfs_extent_data_ref); | |
1255 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1256 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1257 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1258 | struct btrfs_shared_data_ref); | |
1259 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1260 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1261 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1262 | struct btrfs_extent_ref_v0 *ref0; | |
1263 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1264 | struct btrfs_extent_ref_v0); | |
1265 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1266 | #endif | |
1267 | } else { | |
1268 | BUG(); | |
1269 | } | |
1270 | ||
56bec294 CM |
1271 | BUG_ON(num_refs < refs_to_drop); |
1272 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1273 | |
31840ae1 ZY |
1274 | if (num_refs == 0) { |
1275 | ret = btrfs_del_item(trans, root, path); | |
1276 | } else { | |
5d4f98a2 YZ |
1277 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1278 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1279 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1280 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1281 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1282 | else { | |
1283 | struct btrfs_extent_ref_v0 *ref0; | |
1284 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1285 | struct btrfs_extent_ref_v0); | |
1286 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1287 | } | |
1288 | #endif | |
31840ae1 ZY |
1289 | btrfs_mark_buffer_dirty(leaf); |
1290 | } | |
31840ae1 ZY |
1291 | return ret; |
1292 | } | |
1293 | ||
5d4f98a2 YZ |
1294 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1295 | struct btrfs_path *path, | |
1296 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1297 | { |
5d4f98a2 YZ |
1298 | struct btrfs_key key; |
1299 | struct extent_buffer *leaf; | |
1300 | struct btrfs_extent_data_ref *ref1; | |
1301 | struct btrfs_shared_data_ref *ref2; | |
1302 | u32 num_refs = 0; | |
1303 | ||
1304 | leaf = path->nodes[0]; | |
1305 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1306 | if (iref) { | |
1307 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1308 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1309 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1310 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1311 | } else { | |
1312 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1313 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1314 | } | |
1315 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1316 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1317 | struct btrfs_extent_data_ref); | |
1318 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1319 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1320 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1321 | struct btrfs_shared_data_ref); | |
1322 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1323 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1324 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1325 | struct btrfs_extent_ref_v0 *ref0; | |
1326 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1327 | struct btrfs_extent_ref_v0); | |
1328 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1329 | #endif |
5d4f98a2 YZ |
1330 | } else { |
1331 | WARN_ON(1); | |
1332 | } | |
1333 | return num_refs; | |
1334 | } | |
15916de8 | 1335 | |
5d4f98a2 YZ |
1336 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1337 | struct btrfs_root *root, | |
1338 | struct btrfs_path *path, | |
1339 | u64 bytenr, u64 parent, | |
1340 | u64 root_objectid) | |
1f3c79a2 | 1341 | { |
5d4f98a2 | 1342 | struct btrfs_key key; |
1f3c79a2 | 1343 | int ret; |
1f3c79a2 | 1344 | |
5d4f98a2 YZ |
1345 | key.objectid = bytenr; |
1346 | if (parent) { | |
1347 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1348 | key.offset = parent; | |
1349 | } else { | |
1350 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1351 | key.offset = root_objectid; | |
1f3c79a2 LH |
1352 | } |
1353 | ||
5d4f98a2 YZ |
1354 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1355 | if (ret > 0) | |
1356 | ret = -ENOENT; | |
1357 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1358 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1359 | btrfs_release_path(path); |
5d4f98a2 YZ |
1360 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1361 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1362 | if (ret > 0) | |
1363 | ret = -ENOENT; | |
1364 | } | |
1f3c79a2 | 1365 | #endif |
5d4f98a2 | 1366 | return ret; |
1f3c79a2 LH |
1367 | } |
1368 | ||
5d4f98a2 YZ |
1369 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1370 | struct btrfs_root *root, | |
1371 | struct btrfs_path *path, | |
1372 | u64 bytenr, u64 parent, | |
1373 | u64 root_objectid) | |
31840ae1 | 1374 | { |
5d4f98a2 | 1375 | struct btrfs_key key; |
31840ae1 | 1376 | int ret; |
31840ae1 | 1377 | |
5d4f98a2 YZ |
1378 | key.objectid = bytenr; |
1379 | if (parent) { | |
1380 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1381 | key.offset = parent; | |
1382 | } else { | |
1383 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1384 | key.offset = root_objectid; | |
1385 | } | |
1386 | ||
1387 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1388 | btrfs_release_path(path); |
31840ae1 ZY |
1389 | return ret; |
1390 | } | |
1391 | ||
5d4f98a2 | 1392 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1393 | { |
5d4f98a2 YZ |
1394 | int type; |
1395 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1396 | if (parent > 0) | |
1397 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1398 | else | |
1399 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1400 | } else { | |
1401 | if (parent > 0) | |
1402 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1403 | else | |
1404 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1405 | } | |
1406 | return type; | |
31840ae1 | 1407 | } |
56bec294 | 1408 | |
2c47e605 YZ |
1409 | static int find_next_key(struct btrfs_path *path, int level, |
1410 | struct btrfs_key *key) | |
56bec294 | 1411 | |
02217ed2 | 1412 | { |
2c47e605 | 1413 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1414 | if (!path->nodes[level]) |
1415 | break; | |
5d4f98a2 YZ |
1416 | if (path->slots[level] + 1 >= |
1417 | btrfs_header_nritems(path->nodes[level])) | |
1418 | continue; | |
1419 | if (level == 0) | |
1420 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1421 | path->slots[level] + 1); | |
1422 | else | |
1423 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1424 | path->slots[level] + 1); | |
1425 | return 0; | |
1426 | } | |
1427 | return 1; | |
1428 | } | |
037e6390 | 1429 | |
5d4f98a2 YZ |
1430 | /* |
1431 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1432 | * to the address of inline back ref, and 0 is returned. | |
1433 | * | |
1434 | * if back ref isn't found, *ref_ret is set to the address where it | |
1435 | * should be inserted, and -ENOENT is returned. | |
1436 | * | |
1437 | * if insert is true and there are too many inline back refs, the path | |
1438 | * points to the extent item, and -EAGAIN is returned. | |
1439 | * | |
1440 | * NOTE: inline back refs are ordered in the same way that back ref | |
1441 | * items in the tree are ordered. | |
1442 | */ | |
1443 | static noinline_for_stack | |
1444 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1445 | struct btrfs_root *root, | |
1446 | struct btrfs_path *path, | |
1447 | struct btrfs_extent_inline_ref **ref_ret, | |
1448 | u64 bytenr, u64 num_bytes, | |
1449 | u64 parent, u64 root_objectid, | |
1450 | u64 owner, u64 offset, int insert) | |
1451 | { | |
1452 | struct btrfs_key key; | |
1453 | struct extent_buffer *leaf; | |
1454 | struct btrfs_extent_item *ei; | |
1455 | struct btrfs_extent_inline_ref *iref; | |
1456 | u64 flags; | |
1457 | u64 item_size; | |
1458 | unsigned long ptr; | |
1459 | unsigned long end; | |
1460 | int extra_size; | |
1461 | int type; | |
1462 | int want; | |
1463 | int ret; | |
1464 | int err = 0; | |
26b8003f | 1465 | |
db94535d | 1466 | key.objectid = bytenr; |
31840ae1 | 1467 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1468 | key.offset = num_bytes; |
31840ae1 | 1469 | |
5d4f98a2 YZ |
1470 | want = extent_ref_type(parent, owner); |
1471 | if (insert) { | |
1472 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1473 | path->keep_locks = 1; |
5d4f98a2 YZ |
1474 | } else |
1475 | extra_size = -1; | |
1476 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1477 | if (ret < 0) { |
5d4f98a2 YZ |
1478 | err = ret; |
1479 | goto out; | |
1480 | } | |
1481 | BUG_ON(ret); | |
1482 | ||
1483 | leaf = path->nodes[0]; | |
1484 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1485 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1486 | if (item_size < sizeof(*ei)) { | |
1487 | if (!insert) { | |
1488 | err = -ENOENT; | |
1489 | goto out; | |
1490 | } | |
1491 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1492 | extra_size); | |
1493 | if (ret < 0) { | |
1494 | err = ret; | |
1495 | goto out; | |
1496 | } | |
1497 | leaf = path->nodes[0]; | |
1498 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1499 | } | |
1500 | #endif | |
1501 | BUG_ON(item_size < sizeof(*ei)); | |
1502 | ||
5d4f98a2 YZ |
1503 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1504 | flags = btrfs_extent_flags(leaf, ei); | |
1505 | ||
1506 | ptr = (unsigned long)(ei + 1); | |
1507 | end = (unsigned long)ei + item_size; | |
1508 | ||
1509 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1510 | ptr += sizeof(struct btrfs_tree_block_info); | |
1511 | BUG_ON(ptr > end); | |
1512 | } else { | |
1513 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1514 | } | |
1515 | ||
1516 | err = -ENOENT; | |
1517 | while (1) { | |
1518 | if (ptr >= end) { | |
1519 | WARN_ON(ptr > end); | |
1520 | break; | |
1521 | } | |
1522 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1523 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1524 | if (want < type) | |
1525 | break; | |
1526 | if (want > type) { | |
1527 | ptr += btrfs_extent_inline_ref_size(type); | |
1528 | continue; | |
1529 | } | |
1530 | ||
1531 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1532 | struct btrfs_extent_data_ref *dref; | |
1533 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1534 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1535 | owner, offset)) { | |
1536 | err = 0; | |
1537 | break; | |
1538 | } | |
1539 | if (hash_extent_data_ref_item(leaf, dref) < | |
1540 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1541 | break; | |
1542 | } else { | |
1543 | u64 ref_offset; | |
1544 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1545 | if (parent > 0) { | |
1546 | if (parent == ref_offset) { | |
1547 | err = 0; | |
1548 | break; | |
1549 | } | |
1550 | if (ref_offset < parent) | |
1551 | break; | |
1552 | } else { | |
1553 | if (root_objectid == ref_offset) { | |
1554 | err = 0; | |
1555 | break; | |
1556 | } | |
1557 | if (ref_offset < root_objectid) | |
1558 | break; | |
1559 | } | |
1560 | } | |
1561 | ptr += btrfs_extent_inline_ref_size(type); | |
1562 | } | |
1563 | if (err == -ENOENT && insert) { | |
1564 | if (item_size + extra_size >= | |
1565 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1566 | err = -EAGAIN; | |
1567 | goto out; | |
1568 | } | |
1569 | /* | |
1570 | * To add new inline back ref, we have to make sure | |
1571 | * there is no corresponding back ref item. | |
1572 | * For simplicity, we just do not add new inline back | |
1573 | * ref if there is any kind of item for this block | |
1574 | */ | |
2c47e605 YZ |
1575 | if (find_next_key(path, 0, &key) == 0 && |
1576 | key.objectid == bytenr && | |
85d4198e | 1577 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1578 | err = -EAGAIN; |
1579 | goto out; | |
1580 | } | |
1581 | } | |
1582 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1583 | out: | |
85d4198e | 1584 | if (insert) { |
5d4f98a2 YZ |
1585 | path->keep_locks = 0; |
1586 | btrfs_unlock_up_safe(path, 1); | |
1587 | } | |
1588 | return err; | |
1589 | } | |
1590 | ||
1591 | /* | |
1592 | * helper to add new inline back ref | |
1593 | */ | |
1594 | static noinline_for_stack | |
1595 | int setup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1596 | struct btrfs_root *root, | |
1597 | struct btrfs_path *path, | |
1598 | struct btrfs_extent_inline_ref *iref, | |
1599 | u64 parent, u64 root_objectid, | |
1600 | u64 owner, u64 offset, int refs_to_add, | |
1601 | struct btrfs_delayed_extent_op *extent_op) | |
1602 | { | |
1603 | struct extent_buffer *leaf; | |
1604 | struct btrfs_extent_item *ei; | |
1605 | unsigned long ptr; | |
1606 | unsigned long end; | |
1607 | unsigned long item_offset; | |
1608 | u64 refs; | |
1609 | int size; | |
1610 | int type; | |
1611 | int ret; | |
1612 | ||
1613 | leaf = path->nodes[0]; | |
1614 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1615 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1616 | ||
1617 | type = extent_ref_type(parent, owner); | |
1618 | size = btrfs_extent_inline_ref_size(type); | |
1619 | ||
1620 | ret = btrfs_extend_item(trans, root, path, size); | |
5d4f98a2 YZ |
1621 | |
1622 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1623 | refs = btrfs_extent_refs(leaf, ei); | |
1624 | refs += refs_to_add; | |
1625 | btrfs_set_extent_refs(leaf, ei, refs); | |
1626 | if (extent_op) | |
1627 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1628 | ||
1629 | ptr = (unsigned long)ei + item_offset; | |
1630 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1631 | if (ptr < end - size) | |
1632 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1633 | end - size - ptr); | |
1634 | ||
1635 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1636 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1637 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1638 | struct btrfs_extent_data_ref *dref; | |
1639 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1640 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1641 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1642 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1643 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1644 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1645 | struct btrfs_shared_data_ref *sref; | |
1646 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1647 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1648 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1649 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1650 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1651 | } else { | |
1652 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1653 | } | |
1654 | btrfs_mark_buffer_dirty(leaf); | |
1655 | return 0; | |
1656 | } | |
1657 | ||
1658 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1659 | struct btrfs_root *root, | |
1660 | struct btrfs_path *path, | |
1661 | struct btrfs_extent_inline_ref **ref_ret, | |
1662 | u64 bytenr, u64 num_bytes, u64 parent, | |
1663 | u64 root_objectid, u64 owner, u64 offset) | |
1664 | { | |
1665 | int ret; | |
1666 | ||
1667 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1668 | bytenr, num_bytes, parent, | |
1669 | root_objectid, owner, offset, 0); | |
1670 | if (ret != -ENOENT) | |
54aa1f4d | 1671 | return ret; |
5d4f98a2 | 1672 | |
b3b4aa74 | 1673 | btrfs_release_path(path); |
5d4f98a2 YZ |
1674 | *ref_ret = NULL; |
1675 | ||
1676 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1677 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1678 | root_objectid); | |
1679 | } else { | |
1680 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1681 | root_objectid, owner, offset); | |
b9473439 | 1682 | } |
5d4f98a2 YZ |
1683 | return ret; |
1684 | } | |
31840ae1 | 1685 | |
5d4f98a2 YZ |
1686 | /* |
1687 | * helper to update/remove inline back ref | |
1688 | */ | |
1689 | static noinline_for_stack | |
1690 | int update_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1691 | struct btrfs_root *root, | |
1692 | struct btrfs_path *path, | |
1693 | struct btrfs_extent_inline_ref *iref, | |
1694 | int refs_to_mod, | |
1695 | struct btrfs_delayed_extent_op *extent_op) | |
1696 | { | |
1697 | struct extent_buffer *leaf; | |
1698 | struct btrfs_extent_item *ei; | |
1699 | struct btrfs_extent_data_ref *dref = NULL; | |
1700 | struct btrfs_shared_data_ref *sref = NULL; | |
1701 | unsigned long ptr; | |
1702 | unsigned long end; | |
1703 | u32 item_size; | |
1704 | int size; | |
1705 | int type; | |
1706 | int ret; | |
1707 | u64 refs; | |
1708 | ||
1709 | leaf = path->nodes[0]; | |
1710 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1711 | refs = btrfs_extent_refs(leaf, ei); | |
1712 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1713 | refs += refs_to_mod; | |
1714 | btrfs_set_extent_refs(leaf, ei, refs); | |
1715 | if (extent_op) | |
1716 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1717 | ||
1718 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1719 | ||
1720 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1721 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1722 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1723 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1724 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1725 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1726 | } else { | |
1727 | refs = 1; | |
1728 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1729 | } |
31840ae1 | 1730 | |
5d4f98a2 YZ |
1731 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1732 | refs += refs_to_mod; | |
1733 | ||
1734 | if (refs > 0) { | |
1735 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1736 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1737 | else | |
1738 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1739 | } else { | |
1740 | size = btrfs_extent_inline_ref_size(type); | |
1741 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1742 | ptr = (unsigned long)iref; | |
1743 | end = (unsigned long)ei + item_size; | |
1744 | if (ptr + size < end) | |
1745 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1746 | end - ptr - size); | |
1747 | item_size -= size; | |
1748 | ret = btrfs_truncate_item(trans, root, path, item_size, 1); | |
5d4f98a2 YZ |
1749 | } |
1750 | btrfs_mark_buffer_dirty(leaf); | |
1751 | return 0; | |
1752 | } | |
1753 | ||
1754 | static noinline_for_stack | |
1755 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1756 | struct btrfs_root *root, | |
1757 | struct btrfs_path *path, | |
1758 | u64 bytenr, u64 num_bytes, u64 parent, | |
1759 | u64 root_objectid, u64 owner, | |
1760 | u64 offset, int refs_to_add, | |
1761 | struct btrfs_delayed_extent_op *extent_op) | |
1762 | { | |
1763 | struct btrfs_extent_inline_ref *iref; | |
1764 | int ret; | |
1765 | ||
1766 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1767 | bytenr, num_bytes, parent, | |
1768 | root_objectid, owner, offset, 1); | |
1769 | if (ret == 0) { | |
1770 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
1771 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1772 | refs_to_add, extent_op); | |
1773 | } else if (ret == -ENOENT) { | |
1774 | ret = setup_inline_extent_backref(trans, root, path, iref, | |
1775 | parent, root_objectid, | |
1776 | owner, offset, refs_to_add, | |
1777 | extent_op); | |
771ed689 | 1778 | } |
5d4f98a2 YZ |
1779 | return ret; |
1780 | } | |
31840ae1 | 1781 | |
5d4f98a2 YZ |
1782 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1783 | struct btrfs_root *root, | |
1784 | struct btrfs_path *path, | |
1785 | u64 bytenr, u64 parent, u64 root_objectid, | |
1786 | u64 owner, u64 offset, int refs_to_add) | |
1787 | { | |
1788 | int ret; | |
1789 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1790 | BUG_ON(refs_to_add != 1); | |
1791 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1792 | parent, root_objectid); | |
1793 | } else { | |
1794 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1795 | parent, root_objectid, | |
1796 | owner, offset, refs_to_add); | |
1797 | } | |
1798 | return ret; | |
1799 | } | |
56bec294 | 1800 | |
5d4f98a2 YZ |
1801 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1802 | struct btrfs_root *root, | |
1803 | struct btrfs_path *path, | |
1804 | struct btrfs_extent_inline_ref *iref, | |
1805 | int refs_to_drop, int is_data) | |
1806 | { | |
1807 | int ret; | |
b9473439 | 1808 | |
5d4f98a2 YZ |
1809 | BUG_ON(!is_data && refs_to_drop != 1); |
1810 | if (iref) { | |
1811 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1812 | -refs_to_drop, NULL); | |
1813 | } else if (is_data) { | |
1814 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1815 | } else { | |
1816 | ret = btrfs_del_item(trans, root, path); | |
1817 | } | |
1818 | return ret; | |
1819 | } | |
1820 | ||
5378e607 | 1821 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1822 | u64 start, u64 len) |
1823 | { | |
5378e607 | 1824 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1825 | } |
5d4f98a2 YZ |
1826 | |
1827 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1828 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1829 | { |
5d4f98a2 | 1830 | int ret; |
5378e607 | 1831 | u64 discarded_bytes = 0; |
a1d3c478 | 1832 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1833 | |
e244a0ae | 1834 | |
5d4f98a2 | 1835 | /* Tell the block device(s) that the sectors can be discarded */ |
5378e607 | 1836 | ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD, |
a1d3c478 | 1837 | bytenr, &num_bytes, &bbio, 0); |
5d4f98a2 | 1838 | if (!ret) { |
a1d3c478 | 1839 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1840 | int i; |
1841 | ||
5d4f98a2 | 1842 | |
a1d3c478 | 1843 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d5e2003c JB |
1844 | if (!stripe->dev->can_discard) |
1845 | continue; | |
1846 | ||
5378e607 LD |
1847 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1848 | stripe->physical, | |
1849 | stripe->length); | |
1850 | if (!ret) | |
1851 | discarded_bytes += stripe->length; | |
1852 | else if (ret != -EOPNOTSUPP) | |
1853 | break; | |
d5e2003c JB |
1854 | |
1855 | /* | |
1856 | * Just in case we get back EOPNOTSUPP for some reason, | |
1857 | * just ignore the return value so we don't screw up | |
1858 | * people calling discard_extent. | |
1859 | */ | |
1860 | ret = 0; | |
5d4f98a2 | 1861 | } |
a1d3c478 | 1862 | kfree(bbio); |
5d4f98a2 | 1863 | } |
5378e607 LD |
1864 | |
1865 | if (actual_bytes) | |
1866 | *actual_bytes = discarded_bytes; | |
1867 | ||
5d4f98a2 YZ |
1868 | |
1869 | return ret; | |
5d4f98a2 YZ |
1870 | } |
1871 | ||
1872 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1873 | struct btrfs_root *root, | |
1874 | u64 bytenr, u64 num_bytes, u64 parent, | |
66d7e7f0 | 1875 | u64 root_objectid, u64 owner, u64 offset, int for_cow) |
5d4f98a2 YZ |
1876 | { |
1877 | int ret; | |
66d7e7f0 AJ |
1878 | struct btrfs_fs_info *fs_info = root->fs_info; |
1879 | ||
5d4f98a2 YZ |
1880 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
1881 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1882 | ||
1883 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
1884 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
1885 | num_bytes, | |
5d4f98a2 | 1886 | parent, root_objectid, (int)owner, |
66d7e7f0 | 1887 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 1888 | } else { |
66d7e7f0 AJ |
1889 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
1890 | num_bytes, | |
5d4f98a2 | 1891 | parent, root_objectid, owner, offset, |
66d7e7f0 | 1892 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 YZ |
1893 | } |
1894 | return ret; | |
1895 | } | |
1896 | ||
1897 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1898 | struct btrfs_root *root, | |
1899 | u64 bytenr, u64 num_bytes, | |
1900 | u64 parent, u64 root_objectid, | |
1901 | u64 owner, u64 offset, int refs_to_add, | |
1902 | struct btrfs_delayed_extent_op *extent_op) | |
1903 | { | |
1904 | struct btrfs_path *path; | |
1905 | struct extent_buffer *leaf; | |
1906 | struct btrfs_extent_item *item; | |
1907 | u64 refs; | |
1908 | int ret; | |
1909 | int err = 0; | |
1910 | ||
1911 | path = btrfs_alloc_path(); | |
1912 | if (!path) | |
1913 | return -ENOMEM; | |
1914 | ||
1915 | path->reada = 1; | |
1916 | path->leave_spinning = 1; | |
1917 | /* this will setup the path even if it fails to insert the back ref */ | |
1918 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1919 | path, bytenr, num_bytes, parent, | |
1920 | root_objectid, owner, offset, | |
1921 | refs_to_add, extent_op); | |
1922 | if (ret == 0) | |
1923 | goto out; | |
1924 | ||
1925 | if (ret != -EAGAIN) { | |
1926 | err = ret; | |
1927 | goto out; | |
1928 | } | |
1929 | ||
1930 | leaf = path->nodes[0]; | |
1931 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1932 | refs = btrfs_extent_refs(leaf, item); | |
1933 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1934 | if (extent_op) | |
1935 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1936 | |
5d4f98a2 | 1937 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1938 | btrfs_release_path(path); |
56bec294 CM |
1939 | |
1940 | path->reada = 1; | |
b9473439 CM |
1941 | path->leave_spinning = 1; |
1942 | ||
56bec294 CM |
1943 | /* now insert the actual backref */ |
1944 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1945 | path, bytenr, parent, root_objectid, |
1946 | owner, offset, refs_to_add); | |
56bec294 | 1947 | BUG_ON(ret); |
5d4f98a2 | 1948 | out: |
56bec294 | 1949 | btrfs_free_path(path); |
5d4f98a2 | 1950 | return err; |
56bec294 CM |
1951 | } |
1952 | ||
5d4f98a2 YZ |
1953 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1954 | struct btrfs_root *root, | |
1955 | struct btrfs_delayed_ref_node *node, | |
1956 | struct btrfs_delayed_extent_op *extent_op, | |
1957 | int insert_reserved) | |
56bec294 | 1958 | { |
5d4f98a2 YZ |
1959 | int ret = 0; |
1960 | struct btrfs_delayed_data_ref *ref; | |
1961 | struct btrfs_key ins; | |
1962 | u64 parent = 0; | |
1963 | u64 ref_root = 0; | |
1964 | u64 flags = 0; | |
1965 | ||
1966 | ins.objectid = node->bytenr; | |
1967 | ins.offset = node->num_bytes; | |
1968 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1969 | ||
1970 | ref = btrfs_delayed_node_to_data_ref(node); | |
1971 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1972 | parent = ref->parent; | |
1973 | else | |
1974 | ref_root = ref->root; | |
1975 | ||
1976 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1977 | if (extent_op) { | |
1978 | BUG_ON(extent_op->update_key); | |
1979 | flags |= extent_op->flags_to_set; | |
1980 | } | |
1981 | ret = alloc_reserved_file_extent(trans, root, | |
1982 | parent, ref_root, flags, | |
1983 | ref->objectid, ref->offset, | |
1984 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1985 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1986 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1987 | node->num_bytes, parent, | |
1988 | ref_root, ref->objectid, | |
1989 | ref->offset, node->ref_mod, | |
1990 | extent_op); | |
1991 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1992 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1993 | node->num_bytes, parent, | |
1994 | ref_root, ref->objectid, | |
1995 | ref->offset, node->ref_mod, | |
1996 | extent_op); | |
1997 | } else { | |
1998 | BUG(); | |
1999 | } | |
2000 | return ret; | |
2001 | } | |
2002 | ||
2003 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2004 | struct extent_buffer *leaf, | |
2005 | struct btrfs_extent_item *ei) | |
2006 | { | |
2007 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2008 | if (extent_op->update_flags) { | |
2009 | flags |= extent_op->flags_to_set; | |
2010 | btrfs_set_extent_flags(leaf, ei, flags); | |
2011 | } | |
2012 | ||
2013 | if (extent_op->update_key) { | |
2014 | struct btrfs_tree_block_info *bi; | |
2015 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2016 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2017 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2018 | } | |
2019 | } | |
2020 | ||
2021 | static int run_delayed_extent_op(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 | { | |
2026 | struct btrfs_key key; | |
2027 | struct btrfs_path *path; | |
2028 | struct btrfs_extent_item *ei; | |
2029 | struct extent_buffer *leaf; | |
2030 | u32 item_size; | |
56bec294 | 2031 | int ret; |
5d4f98a2 YZ |
2032 | int err = 0; |
2033 | ||
2034 | path = btrfs_alloc_path(); | |
2035 | if (!path) | |
2036 | return -ENOMEM; | |
2037 | ||
2038 | key.objectid = node->bytenr; | |
2039 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2040 | key.offset = node->num_bytes; | |
2041 | ||
2042 | path->reada = 1; | |
2043 | path->leave_spinning = 1; | |
2044 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2045 | path, 0, 1); | |
2046 | if (ret < 0) { | |
2047 | err = ret; | |
2048 | goto out; | |
2049 | } | |
2050 | if (ret > 0) { | |
2051 | err = -EIO; | |
2052 | goto out; | |
2053 | } | |
2054 | ||
2055 | leaf = path->nodes[0]; | |
2056 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2057 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2058 | if (item_size < sizeof(*ei)) { | |
2059 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2060 | path, (u64)-1, 0); | |
2061 | if (ret < 0) { | |
2062 | err = ret; | |
2063 | goto out; | |
2064 | } | |
2065 | leaf = path->nodes[0]; | |
2066 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2067 | } | |
2068 | #endif | |
2069 | BUG_ON(item_size < sizeof(*ei)); | |
2070 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2071 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2072 | |
5d4f98a2 YZ |
2073 | btrfs_mark_buffer_dirty(leaf); |
2074 | out: | |
2075 | btrfs_free_path(path); | |
2076 | return err; | |
56bec294 CM |
2077 | } |
2078 | ||
5d4f98a2 YZ |
2079 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2080 | struct btrfs_root *root, | |
2081 | struct btrfs_delayed_ref_node *node, | |
2082 | struct btrfs_delayed_extent_op *extent_op, | |
2083 | int insert_reserved) | |
56bec294 CM |
2084 | { |
2085 | int ret = 0; | |
5d4f98a2 YZ |
2086 | struct btrfs_delayed_tree_ref *ref; |
2087 | struct btrfs_key ins; | |
2088 | u64 parent = 0; | |
2089 | u64 ref_root = 0; | |
56bec294 | 2090 | |
5d4f98a2 YZ |
2091 | ins.objectid = node->bytenr; |
2092 | ins.offset = node->num_bytes; | |
2093 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 2094 | |
5d4f98a2 YZ |
2095 | ref = btrfs_delayed_node_to_tree_ref(node); |
2096 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2097 | parent = ref->parent; | |
2098 | else | |
2099 | ref_root = ref->root; | |
2100 | ||
2101 | BUG_ON(node->ref_mod != 1); | |
2102 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2103 | BUG_ON(!extent_op || !extent_op->update_flags || | |
2104 | !extent_op->update_key); | |
2105 | ret = alloc_reserved_tree_block(trans, root, | |
2106 | parent, ref_root, | |
2107 | extent_op->flags_to_set, | |
2108 | &extent_op->key, | |
2109 | ref->level, &ins); | |
5d4f98a2 YZ |
2110 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2111 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2112 | node->num_bytes, parent, ref_root, | |
2113 | ref->level, 0, 1, extent_op); | |
2114 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2115 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2116 | node->num_bytes, parent, ref_root, | |
2117 | ref->level, 0, 1, extent_op); | |
2118 | } else { | |
2119 | BUG(); | |
2120 | } | |
56bec294 CM |
2121 | return ret; |
2122 | } | |
2123 | ||
2124 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2125 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2126 | struct btrfs_root *root, | |
2127 | struct btrfs_delayed_ref_node *node, | |
2128 | struct btrfs_delayed_extent_op *extent_op, | |
2129 | int insert_reserved) | |
56bec294 CM |
2130 | { |
2131 | int ret; | |
5d4f98a2 | 2132 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2133 | struct btrfs_delayed_ref_head *head; |
2134 | /* | |
2135 | * we've hit the end of the chain and we were supposed | |
2136 | * to insert this extent into the tree. But, it got | |
2137 | * deleted before we ever needed to insert it, so all | |
2138 | * we have to do is clean up the accounting | |
2139 | */ | |
5d4f98a2 YZ |
2140 | BUG_ON(extent_op); |
2141 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2142 | if (insert_reserved) { |
f0486c68 YZ |
2143 | btrfs_pin_extent(root, node->bytenr, |
2144 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2145 | if (head->is_data) { |
2146 | ret = btrfs_del_csums(trans, root, | |
2147 | node->bytenr, | |
2148 | node->num_bytes); | |
2149 | BUG_ON(ret); | |
2150 | } | |
56bec294 | 2151 | } |
56bec294 CM |
2152 | mutex_unlock(&head->mutex); |
2153 | return 0; | |
2154 | } | |
2155 | ||
5d4f98a2 YZ |
2156 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2157 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2158 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2159 | insert_reserved); | |
2160 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2161 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2162 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2163 | insert_reserved); | |
2164 | else | |
2165 | BUG(); | |
2166 | return ret; | |
56bec294 CM |
2167 | } |
2168 | ||
2169 | static noinline struct btrfs_delayed_ref_node * | |
2170 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2171 | { | |
2172 | struct rb_node *node; | |
2173 | struct btrfs_delayed_ref_node *ref; | |
2174 | int action = BTRFS_ADD_DELAYED_REF; | |
2175 | again: | |
2176 | /* | |
2177 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2178 | * this prevents ref count from going down to zero when | |
2179 | * there still are pending delayed ref. | |
2180 | */ | |
2181 | node = rb_prev(&head->node.rb_node); | |
2182 | while (1) { | |
2183 | if (!node) | |
2184 | break; | |
2185 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2186 | rb_node); | |
2187 | if (ref->bytenr != head->node.bytenr) | |
2188 | break; | |
5d4f98a2 | 2189 | if (ref->action == action) |
56bec294 CM |
2190 | return ref; |
2191 | node = rb_prev(node); | |
2192 | } | |
2193 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2194 | action = BTRFS_DROP_DELAYED_REF; | |
2195 | goto again; | |
2196 | } | |
2197 | return NULL; | |
2198 | } | |
2199 | ||
c3e69d58 CM |
2200 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2201 | struct btrfs_root *root, | |
2202 | struct list_head *cluster) | |
56bec294 | 2203 | { |
56bec294 CM |
2204 | struct btrfs_delayed_ref_root *delayed_refs; |
2205 | struct btrfs_delayed_ref_node *ref; | |
2206 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2207 | struct btrfs_delayed_extent_op *extent_op; |
56bec294 | 2208 | int ret; |
c3e69d58 | 2209 | int count = 0; |
56bec294 | 2210 | int must_insert_reserved = 0; |
56bec294 CM |
2211 | |
2212 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2213 | while (1) { |
2214 | if (!locked_ref) { | |
c3e69d58 CM |
2215 | /* pick a new head ref from the cluster list */ |
2216 | if (list_empty(cluster)) | |
56bec294 | 2217 | break; |
56bec294 | 2218 | |
c3e69d58 CM |
2219 | locked_ref = list_entry(cluster->next, |
2220 | struct btrfs_delayed_ref_head, cluster); | |
2221 | ||
2222 | /* grab the lock that says we are going to process | |
2223 | * all the refs for this head */ | |
2224 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2225 | ||
2226 | /* | |
2227 | * we may have dropped the spin lock to get the head | |
2228 | * mutex lock, and that might have given someone else | |
2229 | * time to free the head. If that's true, it has been | |
2230 | * removed from our list and we can move on. | |
2231 | */ | |
2232 | if (ret == -EAGAIN) { | |
2233 | locked_ref = NULL; | |
2234 | count++; | |
2235 | continue; | |
56bec294 CM |
2236 | } |
2237 | } | |
a28ec197 | 2238 | |
d1270cd9 AJ |
2239 | /* |
2240 | * locked_ref is the head node, so we have to go one | |
2241 | * node back for any delayed ref updates | |
2242 | */ | |
2243 | ref = select_delayed_ref(locked_ref); | |
2244 | ||
2245 | if (ref && ref->seq && | |
2246 | btrfs_check_delayed_seq(delayed_refs, ref->seq)) { | |
2247 | /* | |
2248 | * there are still refs with lower seq numbers in the | |
2249 | * process of being added. Don't run this ref yet. | |
2250 | */ | |
2251 | list_del_init(&locked_ref->cluster); | |
2252 | mutex_unlock(&locked_ref->mutex); | |
2253 | locked_ref = NULL; | |
2254 | delayed_refs->num_heads_ready++; | |
2255 | spin_unlock(&delayed_refs->lock); | |
2256 | cond_resched(); | |
2257 | spin_lock(&delayed_refs->lock); | |
2258 | continue; | |
2259 | } | |
2260 | ||
56bec294 CM |
2261 | /* |
2262 | * record the must insert reserved flag before we | |
2263 | * drop the spin lock. | |
2264 | */ | |
2265 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2266 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2267 | |
5d4f98a2 YZ |
2268 | extent_op = locked_ref->extent_op; |
2269 | locked_ref->extent_op = NULL; | |
2270 | ||
56bec294 CM |
2271 | if (!ref) { |
2272 | /* All delayed refs have been processed, Go ahead | |
2273 | * and send the head node to run_one_delayed_ref, | |
2274 | * so that any accounting fixes can happen | |
2275 | */ | |
2276 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2277 | |
2278 | if (extent_op && must_insert_reserved) { | |
2279 | kfree(extent_op); | |
2280 | extent_op = NULL; | |
2281 | } | |
2282 | ||
2283 | if (extent_op) { | |
2284 | spin_unlock(&delayed_refs->lock); | |
2285 | ||
2286 | ret = run_delayed_extent_op(trans, root, | |
2287 | ref, extent_op); | |
2288 | BUG_ON(ret); | |
2289 | kfree(extent_op); | |
2290 | ||
203bf287 | 2291 | goto next; |
5d4f98a2 YZ |
2292 | } |
2293 | ||
c3e69d58 | 2294 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2295 | locked_ref = NULL; |
2296 | } | |
02217ed2 | 2297 | |
56bec294 CM |
2298 | ref->in_tree = 0; |
2299 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2300 | delayed_refs->num_entries--; | |
a168650c JS |
2301 | /* |
2302 | * we modified num_entries, but as we're currently running | |
2303 | * delayed refs, skip | |
2304 | * wake_up(&delayed_refs->seq_wait); | |
2305 | * here. | |
2306 | */ | |
56bec294 | 2307 | spin_unlock(&delayed_refs->lock); |
925baedd | 2308 | |
5d4f98a2 | 2309 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 CM |
2310 | must_insert_reserved); |
2311 | BUG_ON(ret); | |
eb099670 | 2312 | |
5d4f98a2 YZ |
2313 | btrfs_put_delayed_ref(ref); |
2314 | kfree(extent_op); | |
c3e69d58 | 2315 | count++; |
203bf287 CM |
2316 | next: |
2317 | do_chunk_alloc(trans, root->fs_info->extent_root, | |
2318 | 2 * 1024 * 1024, | |
2319 | btrfs_get_alloc_profile(root, 0), | |
2320 | CHUNK_ALLOC_NO_FORCE); | |
c3e69d58 CM |
2321 | cond_resched(); |
2322 | spin_lock(&delayed_refs->lock); | |
2323 | } | |
2324 | return count; | |
2325 | } | |
2326 | ||
a168650c JS |
2327 | |
2328 | static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs, | |
2329 | unsigned long num_refs) | |
2330 | { | |
2331 | struct list_head *first_seq = delayed_refs->seq_head.next; | |
2332 | ||
2333 | spin_unlock(&delayed_refs->lock); | |
2334 | pr_debug("waiting for more refs (num %ld, first %p)\n", | |
2335 | num_refs, first_seq); | |
2336 | wait_event(delayed_refs->seq_wait, | |
2337 | num_refs != delayed_refs->num_entries || | |
2338 | delayed_refs->seq_head.next != first_seq); | |
2339 | pr_debug("done waiting for more refs (num %ld, first %p)\n", | |
2340 | delayed_refs->num_entries, delayed_refs->seq_head.next); | |
2341 | spin_lock(&delayed_refs->lock); | |
2342 | } | |
2343 | ||
c3e69d58 CM |
2344 | /* |
2345 | * this starts processing the delayed reference count updates and | |
2346 | * extent insertions we have queued up so far. count can be | |
2347 | * 0, which means to process everything in the tree at the start | |
2348 | * of the run (but not newly added entries), or it can be some target | |
2349 | * number you'd like to process. | |
2350 | */ | |
2351 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2352 | struct btrfs_root *root, unsigned long count) | |
2353 | { | |
2354 | struct rb_node *node; | |
2355 | struct btrfs_delayed_ref_root *delayed_refs; | |
2356 | struct btrfs_delayed_ref_node *ref; | |
2357 | struct list_head cluster; | |
2358 | int ret; | |
a168650c | 2359 | u64 delayed_start; |
c3e69d58 CM |
2360 | int run_all = count == (unsigned long)-1; |
2361 | int run_most = 0; | |
a168650c JS |
2362 | unsigned long num_refs = 0; |
2363 | int consider_waiting; | |
c3e69d58 CM |
2364 | |
2365 | if (root == root->fs_info->extent_root) | |
2366 | root = root->fs_info->tree_root; | |
2367 | ||
203bf287 CM |
2368 | do_chunk_alloc(trans, root->fs_info->extent_root, |
2369 | 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0), | |
2370 | CHUNK_ALLOC_NO_FORCE); | |
2371 | ||
c3e69d58 CM |
2372 | delayed_refs = &trans->transaction->delayed_refs; |
2373 | INIT_LIST_HEAD(&cluster); | |
2374 | again: | |
a168650c | 2375 | consider_waiting = 0; |
c3e69d58 CM |
2376 | spin_lock(&delayed_refs->lock); |
2377 | if (count == 0) { | |
2378 | count = delayed_refs->num_entries * 2; | |
2379 | run_most = 1; | |
2380 | } | |
2381 | while (1) { | |
2382 | if (!(run_all || run_most) && | |
2383 | delayed_refs->num_heads_ready < 64) | |
2384 | break; | |
eb099670 | 2385 | |
56bec294 | 2386 | /* |
c3e69d58 CM |
2387 | * go find something we can process in the rbtree. We start at |
2388 | * the beginning of the tree, and then build a cluster | |
2389 | * of refs to process starting at the first one we are able to | |
2390 | * lock | |
56bec294 | 2391 | */ |
a168650c | 2392 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2393 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2394 | delayed_refs->run_delayed_start); | |
2395 | if (ret) | |
56bec294 CM |
2396 | break; |
2397 | ||
a168650c JS |
2398 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2399 | if (consider_waiting == 0) { | |
2400 | /* | |
2401 | * btrfs_find_ref_cluster looped. let's do one | |
2402 | * more cycle. if we don't run any delayed ref | |
2403 | * during that cycle (because we can't because | |
2404 | * all of them are blocked) and if the number of | |
2405 | * refs doesn't change, we avoid busy waiting. | |
2406 | */ | |
2407 | consider_waiting = 1; | |
2408 | num_refs = delayed_refs->num_entries; | |
2409 | } else { | |
2410 | wait_for_more_refs(delayed_refs, num_refs); | |
2411 | /* | |
2412 | * after waiting, things have changed. we | |
2413 | * dropped the lock and someone else might have | |
2414 | * run some refs, built new clusters and so on. | |
2415 | * therefore, we restart staleness detection. | |
2416 | */ | |
2417 | consider_waiting = 0; | |
2418 | } | |
2419 | } | |
2420 | ||
c3e69d58 CM |
2421 | ret = run_clustered_refs(trans, root, &cluster); |
2422 | BUG_ON(ret < 0); | |
2423 | ||
2424 | count -= min_t(unsigned long, ret, count); | |
2425 | ||
2426 | if (count == 0) | |
2427 | break; | |
a168650c JS |
2428 | |
2429 | if (ret || delayed_refs->run_delayed_start == 0) { | |
2430 | /* refs were run, let's reset staleness detection */ | |
2431 | consider_waiting = 0; | |
2432 | } | |
eb099670 | 2433 | } |
c3e69d58 | 2434 | |
56bec294 | 2435 | if (run_all) { |
56bec294 | 2436 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2437 | if (!node) |
56bec294 | 2438 | goto out; |
c3e69d58 | 2439 | count = (unsigned long)-1; |
e9d0b13b | 2440 | |
56bec294 CM |
2441 | while (node) { |
2442 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2443 | rb_node); | |
2444 | if (btrfs_delayed_ref_is_head(ref)) { | |
2445 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2446 | |
56bec294 CM |
2447 | head = btrfs_delayed_node_to_head(ref); |
2448 | atomic_inc(&ref->refs); | |
2449 | ||
2450 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2451 | /* |
2452 | * Mutex was contended, block until it's | |
2453 | * released and try again | |
2454 | */ | |
56bec294 CM |
2455 | mutex_lock(&head->mutex); |
2456 | mutex_unlock(&head->mutex); | |
2457 | ||
2458 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2459 | cond_resched(); |
56bec294 CM |
2460 | goto again; |
2461 | } | |
2462 | node = rb_next(node); | |
2463 | } | |
2464 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2465 | schedule_timeout(1); |
2466 | goto again; | |
5f39d397 | 2467 | } |
54aa1f4d | 2468 | out: |
c3e69d58 | 2469 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2470 | return 0; |
2471 | } | |
2472 | ||
5d4f98a2 YZ |
2473 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2474 | struct btrfs_root *root, | |
2475 | u64 bytenr, u64 num_bytes, u64 flags, | |
2476 | int is_data) | |
2477 | { | |
2478 | struct btrfs_delayed_extent_op *extent_op; | |
2479 | int ret; | |
2480 | ||
2481 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2482 | if (!extent_op) | |
2483 | return -ENOMEM; | |
2484 | ||
2485 | extent_op->flags_to_set = flags; | |
2486 | extent_op->update_flags = 1; | |
2487 | extent_op->update_key = 0; | |
2488 | extent_op->is_data = is_data ? 1 : 0; | |
2489 | ||
66d7e7f0 AJ |
2490 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2491 | num_bytes, extent_op); | |
5d4f98a2 YZ |
2492 | if (ret) |
2493 | kfree(extent_op); | |
2494 | return ret; | |
2495 | } | |
2496 | ||
2497 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2498 | struct btrfs_root *root, | |
2499 | struct btrfs_path *path, | |
2500 | u64 objectid, u64 offset, u64 bytenr) | |
2501 | { | |
2502 | struct btrfs_delayed_ref_head *head; | |
2503 | struct btrfs_delayed_ref_node *ref; | |
2504 | struct btrfs_delayed_data_ref *data_ref; | |
2505 | struct btrfs_delayed_ref_root *delayed_refs; | |
2506 | struct rb_node *node; | |
2507 | int ret = 0; | |
2508 | ||
2509 | ret = -ENOENT; | |
2510 | delayed_refs = &trans->transaction->delayed_refs; | |
2511 | spin_lock(&delayed_refs->lock); | |
2512 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2513 | if (!head) | |
2514 | goto out; | |
2515 | ||
2516 | if (!mutex_trylock(&head->mutex)) { | |
2517 | atomic_inc(&head->node.refs); | |
2518 | spin_unlock(&delayed_refs->lock); | |
2519 | ||
b3b4aa74 | 2520 | btrfs_release_path(path); |
5d4f98a2 | 2521 | |
8cc33e5c DS |
2522 | /* |
2523 | * Mutex was contended, block until it's released and let | |
2524 | * caller try again | |
2525 | */ | |
5d4f98a2 YZ |
2526 | mutex_lock(&head->mutex); |
2527 | mutex_unlock(&head->mutex); | |
2528 | btrfs_put_delayed_ref(&head->node); | |
2529 | return -EAGAIN; | |
2530 | } | |
2531 | ||
2532 | node = rb_prev(&head->node.rb_node); | |
2533 | if (!node) | |
2534 | goto out_unlock; | |
2535 | ||
2536 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2537 | ||
2538 | if (ref->bytenr != bytenr) | |
2539 | goto out_unlock; | |
2540 | ||
2541 | ret = 1; | |
2542 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2543 | goto out_unlock; | |
2544 | ||
2545 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2546 | ||
2547 | node = rb_prev(node); | |
2548 | if (node) { | |
2549 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2550 | if (ref->bytenr == bytenr) | |
2551 | goto out_unlock; | |
2552 | } | |
2553 | ||
2554 | if (data_ref->root != root->root_key.objectid || | |
2555 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2556 | goto out_unlock; | |
2557 | ||
2558 | ret = 0; | |
2559 | out_unlock: | |
2560 | mutex_unlock(&head->mutex); | |
2561 | out: | |
2562 | spin_unlock(&delayed_refs->lock); | |
2563 | return ret; | |
2564 | } | |
2565 | ||
2566 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2567 | struct btrfs_root *root, | |
2568 | struct btrfs_path *path, | |
2569 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2570 | { |
2571 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2572 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2573 | struct btrfs_extent_data_ref *ref; |
2574 | struct btrfs_extent_inline_ref *iref; | |
2575 | struct btrfs_extent_item *ei; | |
f321e491 | 2576 | struct btrfs_key key; |
5d4f98a2 | 2577 | u32 item_size; |
be20aa9d | 2578 | int ret; |
925baedd | 2579 | |
be20aa9d | 2580 | key.objectid = bytenr; |
31840ae1 | 2581 | key.offset = (u64)-1; |
f321e491 | 2582 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2583 | |
be20aa9d CM |
2584 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2585 | if (ret < 0) | |
2586 | goto out; | |
2587 | BUG_ON(ret == 0); | |
80ff3856 YZ |
2588 | |
2589 | ret = -ENOENT; | |
2590 | if (path->slots[0] == 0) | |
31840ae1 | 2591 | goto out; |
be20aa9d | 2592 | |
31840ae1 | 2593 | path->slots[0]--; |
f321e491 | 2594 | leaf = path->nodes[0]; |
5d4f98a2 | 2595 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2596 | |
5d4f98a2 | 2597 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2598 | goto out; |
f321e491 | 2599 | |
5d4f98a2 YZ |
2600 | ret = 1; |
2601 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2602 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2603 | if (item_size < sizeof(*ei)) { | |
2604 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2605 | goto out; | |
2606 | } | |
2607 | #endif | |
2608 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2609 | |
5d4f98a2 YZ |
2610 | if (item_size != sizeof(*ei) + |
2611 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2612 | goto out; | |
be20aa9d | 2613 | |
5d4f98a2 YZ |
2614 | if (btrfs_extent_generation(leaf, ei) <= |
2615 | btrfs_root_last_snapshot(&root->root_item)) | |
2616 | goto out; | |
2617 | ||
2618 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2619 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2620 | BTRFS_EXTENT_DATA_REF_KEY) | |
2621 | goto out; | |
2622 | ||
2623 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2624 | if (btrfs_extent_refs(leaf, ei) != | |
2625 | btrfs_extent_data_ref_count(leaf, ref) || | |
2626 | btrfs_extent_data_ref_root(leaf, ref) != | |
2627 | root->root_key.objectid || | |
2628 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2629 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2630 | goto out; | |
2631 | ||
2632 | ret = 0; | |
2633 | out: | |
2634 | return ret; | |
2635 | } | |
2636 | ||
2637 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2638 | struct btrfs_root *root, | |
2639 | u64 objectid, u64 offset, u64 bytenr) | |
2640 | { | |
2641 | struct btrfs_path *path; | |
2642 | int ret; | |
2643 | int ret2; | |
2644 | ||
2645 | path = btrfs_alloc_path(); | |
2646 | if (!path) | |
2647 | return -ENOENT; | |
2648 | ||
2649 | do { | |
2650 | ret = check_committed_ref(trans, root, path, objectid, | |
2651 | offset, bytenr); | |
2652 | if (ret && ret != -ENOENT) | |
f321e491 | 2653 | goto out; |
80ff3856 | 2654 | |
5d4f98a2 YZ |
2655 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2656 | offset, bytenr); | |
2657 | } while (ret2 == -EAGAIN); | |
2658 | ||
2659 | if (ret2 && ret2 != -ENOENT) { | |
2660 | ret = ret2; | |
2661 | goto out; | |
f321e491 | 2662 | } |
5d4f98a2 YZ |
2663 | |
2664 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2665 | ret = 0; | |
be20aa9d | 2666 | out: |
80ff3856 | 2667 | btrfs_free_path(path); |
f0486c68 YZ |
2668 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2669 | WARN_ON(ret > 0); | |
f321e491 | 2670 | return ret; |
be20aa9d | 2671 | } |
c5739bba | 2672 | |
5d4f98a2 | 2673 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2674 | struct btrfs_root *root, |
5d4f98a2 | 2675 | struct extent_buffer *buf, |
66d7e7f0 | 2676 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
2677 | { |
2678 | u64 bytenr; | |
5d4f98a2 YZ |
2679 | u64 num_bytes; |
2680 | u64 parent; | |
31840ae1 | 2681 | u64 ref_root; |
31840ae1 | 2682 | u32 nritems; |
31840ae1 ZY |
2683 | struct btrfs_key key; |
2684 | struct btrfs_file_extent_item *fi; | |
2685 | int i; | |
2686 | int level; | |
2687 | int ret = 0; | |
31840ae1 | 2688 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 2689 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
2690 | |
2691 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2692 | nritems = btrfs_header_nritems(buf); |
2693 | level = btrfs_header_level(buf); | |
2694 | ||
5d4f98a2 YZ |
2695 | if (!root->ref_cows && level == 0) |
2696 | return 0; | |
31840ae1 | 2697 | |
5d4f98a2 YZ |
2698 | if (inc) |
2699 | process_func = btrfs_inc_extent_ref; | |
2700 | else | |
2701 | process_func = btrfs_free_extent; | |
31840ae1 | 2702 | |
5d4f98a2 YZ |
2703 | if (full_backref) |
2704 | parent = buf->start; | |
2705 | else | |
2706 | parent = 0; | |
2707 | ||
2708 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2709 | if (level == 0) { |
5d4f98a2 | 2710 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2711 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2712 | continue; | |
5d4f98a2 | 2713 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2714 | struct btrfs_file_extent_item); |
2715 | if (btrfs_file_extent_type(buf, fi) == | |
2716 | BTRFS_FILE_EXTENT_INLINE) | |
2717 | continue; | |
2718 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2719 | if (bytenr == 0) | |
2720 | continue; | |
5d4f98a2 YZ |
2721 | |
2722 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2723 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2724 | ret = process_func(trans, root, bytenr, num_bytes, | |
2725 | parent, ref_root, key.objectid, | |
66d7e7f0 | 2726 | key.offset, for_cow); |
31840ae1 ZY |
2727 | if (ret) |
2728 | goto fail; | |
2729 | } else { | |
5d4f98a2 YZ |
2730 | bytenr = btrfs_node_blockptr(buf, i); |
2731 | num_bytes = btrfs_level_size(root, level - 1); | |
2732 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
2733 | parent, ref_root, level - 1, 0, |
2734 | for_cow); | |
31840ae1 ZY |
2735 | if (ret) |
2736 | goto fail; | |
2737 | } | |
2738 | } | |
2739 | return 0; | |
2740 | fail: | |
5d4f98a2 YZ |
2741 | BUG(); |
2742 | return ret; | |
2743 | } | |
2744 | ||
2745 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2746 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2747 | { |
66d7e7f0 | 2748 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
2749 | } |
2750 | ||
2751 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2752 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2753 | { |
66d7e7f0 | 2754 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
2755 | } |
2756 | ||
9078a3e1 CM |
2757 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2758 | struct btrfs_root *root, | |
2759 | struct btrfs_path *path, | |
2760 | struct btrfs_block_group_cache *cache) | |
2761 | { | |
2762 | int ret; | |
9078a3e1 | 2763 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2764 | unsigned long bi; |
2765 | struct extent_buffer *leaf; | |
9078a3e1 | 2766 | |
9078a3e1 | 2767 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2768 | if (ret < 0) |
2769 | goto fail; | |
9078a3e1 | 2770 | BUG_ON(ret); |
5f39d397 CM |
2771 | |
2772 | leaf = path->nodes[0]; | |
2773 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2774 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2775 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2776 | btrfs_release_path(path); |
54aa1f4d | 2777 | fail: |
9078a3e1 CM |
2778 | if (ret) |
2779 | return ret; | |
9078a3e1 CM |
2780 | return 0; |
2781 | ||
2782 | } | |
2783 | ||
4a8c9a62 YZ |
2784 | static struct btrfs_block_group_cache * |
2785 | next_block_group(struct btrfs_root *root, | |
2786 | struct btrfs_block_group_cache *cache) | |
2787 | { | |
2788 | struct rb_node *node; | |
2789 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2790 | node = rb_next(&cache->cache_node); | |
2791 | btrfs_put_block_group(cache); | |
2792 | if (node) { | |
2793 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2794 | cache_node); | |
11dfe35a | 2795 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2796 | } else |
2797 | cache = NULL; | |
2798 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2799 | return cache; | |
2800 | } | |
2801 | ||
0af3d00b JB |
2802 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2803 | struct btrfs_trans_handle *trans, | |
2804 | struct btrfs_path *path) | |
2805 | { | |
2806 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2807 | struct inode *inode = NULL; | |
2808 | u64 alloc_hint = 0; | |
2b20982e | 2809 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2810 | int num_pages = 0; |
2811 | int retries = 0; | |
2812 | int ret = 0; | |
2813 | ||
2814 | /* | |
2815 | * If this block group is smaller than 100 megs don't bother caching the | |
2816 | * block group. | |
2817 | */ | |
2818 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2819 | spin_lock(&block_group->lock); | |
2820 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2821 | spin_unlock(&block_group->lock); | |
2822 | return 0; | |
2823 | } | |
2824 | ||
2825 | again: | |
2826 | inode = lookup_free_space_inode(root, block_group, path); | |
2827 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2828 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2829 | btrfs_release_path(path); |
0af3d00b JB |
2830 | goto out; |
2831 | } | |
2832 | ||
2833 | if (IS_ERR(inode)) { | |
2834 | BUG_ON(retries); | |
2835 | retries++; | |
2836 | ||
2837 | if (block_group->ro) | |
2838 | goto out_free; | |
2839 | ||
2840 | ret = create_free_space_inode(root, trans, block_group, path); | |
2841 | if (ret) | |
2842 | goto out_free; | |
2843 | goto again; | |
2844 | } | |
2845 | ||
5b0e95bf JB |
2846 | /* We've already setup this transaction, go ahead and exit */ |
2847 | if (block_group->cache_generation == trans->transid && | |
2848 | i_size_read(inode)) { | |
2849 | dcs = BTRFS_DC_SETUP; | |
2850 | goto out_put; | |
2851 | } | |
2852 | ||
0af3d00b JB |
2853 | /* |
2854 | * We want to set the generation to 0, that way if anything goes wrong | |
2855 | * from here on out we know not to trust this cache when we load up next | |
2856 | * time. | |
2857 | */ | |
2858 | BTRFS_I(inode)->generation = 0; | |
2859 | ret = btrfs_update_inode(trans, root, inode); | |
2860 | WARN_ON(ret); | |
2861 | ||
2862 | if (i_size_read(inode) > 0) { | |
2863 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2864 | inode); | |
2865 | if (ret) | |
2866 | goto out_put; | |
2867 | } | |
2868 | ||
2869 | spin_lock(&block_group->lock); | |
2870 | if (block_group->cached != BTRFS_CACHE_FINISHED) { | |
2b20982e JB |
2871 | /* We're not cached, don't bother trying to write stuff out */ |
2872 | dcs = BTRFS_DC_WRITTEN; | |
0af3d00b JB |
2873 | spin_unlock(&block_group->lock); |
2874 | goto out_put; | |
2875 | } | |
2876 | spin_unlock(&block_group->lock); | |
2877 | ||
2878 | num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024); | |
2879 | if (!num_pages) | |
2880 | num_pages = 1; | |
2881 | ||
2882 | /* | |
2883 | * Just to make absolutely sure we have enough space, we're going to | |
2884 | * preallocate 12 pages worth of space for each block group. In | |
2885 | * practice we ought to use at most 8, but we need extra space so we can | |
2886 | * add our header and have a terminator between the extents and the | |
2887 | * bitmaps. | |
2888 | */ | |
2889 | num_pages *= 16; | |
2890 | num_pages *= PAGE_CACHE_SIZE; | |
2891 | ||
2892 | ret = btrfs_check_data_free_space(inode, num_pages); | |
2893 | if (ret) | |
2894 | goto out_put; | |
2895 | ||
2896 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2897 | num_pages, num_pages, | |
2898 | &alloc_hint); | |
2b20982e JB |
2899 | if (!ret) |
2900 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 2901 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 2902 | |
0af3d00b JB |
2903 | out_put: |
2904 | iput(inode); | |
2905 | out_free: | |
b3b4aa74 | 2906 | btrfs_release_path(path); |
0af3d00b JB |
2907 | out: |
2908 | spin_lock(&block_group->lock); | |
e65cbb94 | 2909 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 2910 | block_group->cache_generation = trans->transid; |
2b20982e | 2911 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
2912 | spin_unlock(&block_group->lock); |
2913 | ||
2914 | return ret; | |
2915 | } | |
2916 | ||
96b5179d CM |
2917 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2918 | struct btrfs_root *root) | |
9078a3e1 | 2919 | { |
4a8c9a62 | 2920 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2921 | int err = 0; |
9078a3e1 | 2922 | struct btrfs_path *path; |
96b5179d | 2923 | u64 last = 0; |
9078a3e1 CM |
2924 | |
2925 | path = btrfs_alloc_path(); | |
2926 | if (!path) | |
2927 | return -ENOMEM; | |
2928 | ||
0af3d00b JB |
2929 | again: |
2930 | while (1) { | |
2931 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2932 | while (cache) { | |
2933 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2934 | break; | |
2935 | cache = next_block_group(root, cache); | |
2936 | } | |
2937 | if (!cache) { | |
2938 | if (last == 0) | |
2939 | break; | |
2940 | last = 0; | |
2941 | continue; | |
2942 | } | |
2943 | err = cache_save_setup(cache, trans, path); | |
2944 | last = cache->key.objectid + cache->key.offset; | |
2945 | btrfs_put_block_group(cache); | |
2946 | } | |
2947 | ||
d397712b | 2948 | while (1) { |
4a8c9a62 YZ |
2949 | if (last == 0) { |
2950 | err = btrfs_run_delayed_refs(trans, root, | |
2951 | (unsigned long)-1); | |
2952 | BUG_ON(err); | |
0f9dd46c | 2953 | } |
54aa1f4d | 2954 | |
4a8c9a62 YZ |
2955 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2956 | while (cache) { | |
0af3d00b JB |
2957 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
2958 | btrfs_put_block_group(cache); | |
2959 | goto again; | |
2960 | } | |
2961 | ||
4a8c9a62 YZ |
2962 | if (cache->dirty) |
2963 | break; | |
2964 | cache = next_block_group(root, cache); | |
2965 | } | |
2966 | if (!cache) { | |
2967 | if (last == 0) | |
2968 | break; | |
2969 | last = 0; | |
2970 | continue; | |
2971 | } | |
0f9dd46c | 2972 | |
0cb59c99 JB |
2973 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
2974 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 2975 | cache->dirty = 0; |
4a8c9a62 | 2976 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 2977 | |
4a8c9a62 YZ |
2978 | err = write_one_cache_group(trans, root, path, cache); |
2979 | BUG_ON(err); | |
2980 | btrfs_put_block_group(cache); | |
9078a3e1 | 2981 | } |
4a8c9a62 | 2982 | |
0cb59c99 JB |
2983 | while (1) { |
2984 | /* | |
2985 | * I don't think this is needed since we're just marking our | |
2986 | * preallocated extent as written, but just in case it can't | |
2987 | * hurt. | |
2988 | */ | |
2989 | if (last == 0) { | |
2990 | err = btrfs_run_delayed_refs(trans, root, | |
2991 | (unsigned long)-1); | |
2992 | BUG_ON(err); | |
2993 | } | |
2994 | ||
2995 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2996 | while (cache) { | |
2997 | /* | |
2998 | * Really this shouldn't happen, but it could if we | |
2999 | * couldn't write the entire preallocated extent and | |
3000 | * splitting the extent resulted in a new block. | |
3001 | */ | |
3002 | if (cache->dirty) { | |
3003 | btrfs_put_block_group(cache); | |
3004 | goto again; | |
3005 | } | |
3006 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3007 | break; | |
3008 | cache = next_block_group(root, cache); | |
3009 | } | |
3010 | if (!cache) { | |
3011 | if (last == 0) | |
3012 | break; | |
3013 | last = 0; | |
3014 | continue; | |
3015 | } | |
3016 | ||
3017 | btrfs_write_out_cache(root, trans, cache, path); | |
3018 | ||
3019 | /* | |
3020 | * If we didn't have an error then the cache state is still | |
3021 | * NEED_WRITE, so we can set it to WRITTEN. | |
3022 | */ | |
3023 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3024 | cache->disk_cache_state = BTRFS_DC_WRITTEN; | |
3025 | last = cache->key.objectid + cache->key.offset; | |
3026 | btrfs_put_block_group(cache); | |
3027 | } | |
3028 | ||
9078a3e1 | 3029 | btrfs_free_path(path); |
4a8c9a62 | 3030 | return 0; |
9078a3e1 CM |
3031 | } |
3032 | ||
d2fb3437 YZ |
3033 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3034 | { | |
3035 | struct btrfs_block_group_cache *block_group; | |
3036 | int readonly = 0; | |
3037 | ||
3038 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3039 | if (!block_group || block_group->ro) | |
3040 | readonly = 1; | |
3041 | if (block_group) | |
fa9c0d79 | 3042 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3043 | return readonly; |
3044 | } | |
3045 | ||
593060d7 CM |
3046 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3047 | u64 total_bytes, u64 bytes_used, | |
3048 | struct btrfs_space_info **space_info) | |
3049 | { | |
3050 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3051 | int i; |
3052 | int factor; | |
3053 | ||
3054 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3055 | BTRFS_BLOCK_GROUP_RAID10)) | |
3056 | factor = 2; | |
3057 | else | |
3058 | factor = 1; | |
593060d7 CM |
3059 | |
3060 | found = __find_space_info(info, flags); | |
3061 | if (found) { | |
25179201 | 3062 | spin_lock(&found->lock); |
593060d7 | 3063 | found->total_bytes += total_bytes; |
89a55897 | 3064 | found->disk_total += total_bytes * factor; |
593060d7 | 3065 | found->bytes_used += bytes_used; |
b742bb82 | 3066 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3067 | found->full = 0; |
25179201 | 3068 | spin_unlock(&found->lock); |
593060d7 CM |
3069 | *space_info = found; |
3070 | return 0; | |
3071 | } | |
c146afad | 3072 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3073 | if (!found) |
3074 | return -ENOMEM; | |
3075 | ||
b742bb82 YZ |
3076 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3077 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3078 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3079 | spin_lock_init(&found->lock); |
52ba6929 | 3080 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3081 | found->total_bytes = total_bytes; |
89a55897 | 3082 | found->disk_total = total_bytes * factor; |
593060d7 | 3083 | found->bytes_used = bytes_used; |
b742bb82 | 3084 | found->disk_used = bytes_used * factor; |
593060d7 | 3085 | found->bytes_pinned = 0; |
e8569813 | 3086 | found->bytes_reserved = 0; |
c146afad | 3087 | found->bytes_readonly = 0; |
f0486c68 | 3088 | found->bytes_may_use = 0; |
593060d7 | 3089 | found->full = 0; |
0e4f8f88 | 3090 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3091 | found->chunk_alloc = 0; |
fdb5effd JB |
3092 | found->flush = 0; |
3093 | init_waitqueue_head(&found->wait); | |
593060d7 | 3094 | *space_info = found; |
4184ea7f | 3095 | list_add_rcu(&found->list, &info->space_info); |
593060d7 CM |
3096 | return 0; |
3097 | } | |
3098 | ||
8790d502 CM |
3099 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3100 | { | |
52ba6929 | 3101 | u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK; |
a46d11a8 ID |
3102 | |
3103 | /* chunk -> extended profile */ | |
3104 | if (extra_flags == 0) | |
3105 | extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
3106 | ||
3107 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3108 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3109 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3110 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3111 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3112 | fs_info->avail_system_alloc_bits |= extra_flags; | |
8790d502 | 3113 | } |
593060d7 | 3114 | |
a46d11a8 ID |
3115 | /* |
3116 | * @flags: available profiles in extended format (see ctree.h) | |
3117 | * | |
e4d8ec0f ID |
3118 | * Returns reduced profile in chunk format. If profile changing is in |
3119 | * progress (either running or paused) picks the target profile (if it's | |
3120 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3121 | */ |
2b82032c | 3122 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3123 | { |
cd02dca5 CM |
3124 | /* |
3125 | * we add in the count of missing devices because we want | |
3126 | * to make sure that any RAID levels on a degraded FS | |
3127 | * continue to be honored. | |
3128 | */ | |
3129 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3130 | root->fs_info->fs_devices->missing_devices; | |
a061fc8d | 3131 | |
e4d8ec0f ID |
3132 | /* pick restriper's target profile if it's available */ |
3133 | spin_lock(&root->fs_info->balance_lock); | |
3134 | if (root->fs_info->balance_ctl) { | |
3135 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
3136 | u64 tgt = 0; | |
3137 | ||
3138 | if ((flags & BTRFS_BLOCK_GROUP_DATA) && | |
3139 | (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3140 | (flags & bctl->data.target)) { | |
3141 | tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3142 | } else if ((flags & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3143 | (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3144 | (flags & bctl->sys.target)) { | |
3145 | tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3146 | } else if ((flags & BTRFS_BLOCK_GROUP_METADATA) && | |
3147 | (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3148 | (flags & bctl->meta.target)) { | |
3149 | tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3150 | } | |
3151 | ||
3152 | if (tgt) { | |
3153 | spin_unlock(&root->fs_info->balance_lock); | |
3154 | flags = tgt; | |
3155 | goto out; | |
3156 | } | |
3157 | } | |
3158 | spin_unlock(&root->fs_info->balance_lock); | |
3159 | ||
a061fc8d CM |
3160 | if (num_devices == 1) |
3161 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
3162 | if (num_devices < 4) | |
3163 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3164 | ||
ec44a35c CM |
3165 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
3166 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 3167 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 3168 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 3169 | } |
ec44a35c CM |
3170 | |
3171 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 3172 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 3173 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 3174 | } |
ec44a35c CM |
3175 | |
3176 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
3177 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
3178 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
a46d11a8 | 3179 | (flags & BTRFS_BLOCK_GROUP_DUP))) { |
ec44a35c | 3180 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; |
a46d11a8 ID |
3181 | } |
3182 | ||
e4d8ec0f | 3183 | out: |
a46d11a8 ID |
3184 | /* extended -> chunk profile */ |
3185 | flags &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
ec44a35c CM |
3186 | return flags; |
3187 | } | |
3188 | ||
b742bb82 | 3189 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3190 | { |
b742bb82 | 3191 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
6fef8df1 | 3192 | flags |= root->fs_info->avail_data_alloc_bits; |
b742bb82 | 3193 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
6fef8df1 | 3194 | flags |= root->fs_info->avail_system_alloc_bits; |
b742bb82 | 3195 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
6fef8df1 ID |
3196 | flags |= root->fs_info->avail_metadata_alloc_bits; |
3197 | ||
b742bb82 | 3198 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3199 | } |
3200 | ||
6d07bcec | 3201 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3202 | { |
b742bb82 | 3203 | u64 flags; |
9ed74f2d | 3204 | |
b742bb82 YZ |
3205 | if (data) |
3206 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3207 | else if (root == root->fs_info->chunk_root) | |
3208 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3209 | else |
b742bb82 | 3210 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3211 | |
b742bb82 | 3212 | return get_alloc_profile(root, flags); |
6a63209f | 3213 | } |
9ed74f2d | 3214 | |
6a63209f JB |
3215 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) |
3216 | { | |
6a63209f | 3217 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, |
f0486c68 | 3218 | BTRFS_BLOCK_GROUP_DATA); |
9ed74f2d JB |
3219 | } |
3220 | ||
6a63209f | 3221 | /* |
6a63209f JB |
3222 | * This will check the space that the inode allocates from to make sure we have |
3223 | * enough space for bytes. | |
6a63209f | 3224 | */ |
0ca1f7ce | 3225 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3226 | { |
6a63209f | 3227 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3228 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ab6e2410 | 3229 | u64 used; |
0af3d00b | 3230 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3231 | |
6a63209f JB |
3232 | /* make sure bytes are sectorsize aligned */ |
3233 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3234 | |
82d5902d LZ |
3235 | if (root == root->fs_info->tree_root || |
3236 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3237 | alloc_chunk = 0; |
3238 | committed = 1; | |
3239 | } | |
3240 | ||
6a63209f | 3241 | data_sinfo = BTRFS_I(inode)->space_info; |
33b4d47f CM |
3242 | if (!data_sinfo) |
3243 | goto alloc; | |
9ed74f2d | 3244 | |
6a63209f JB |
3245 | again: |
3246 | /* make sure we have enough space to handle the data first */ | |
3247 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3248 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3249 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3250 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3251 | |
3252 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3253 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3254 | |
6a63209f JB |
3255 | /* |
3256 | * if we don't have enough free bytes in this space then we need | |
3257 | * to alloc a new chunk. | |
3258 | */ | |
0af3d00b | 3259 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3260 | u64 alloc_target; |
9ed74f2d | 3261 | |
0e4f8f88 | 3262 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3263 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3264 | alloc: |
6a63209f | 3265 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3266 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3267 | if (IS_ERR(trans)) |
3268 | return PTR_ERR(trans); | |
9ed74f2d | 3269 | |
6a63209f JB |
3270 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
3271 | bytes + 2 * 1024 * 1024, | |
0e4f8f88 CM |
3272 | alloc_target, |
3273 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3274 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3275 | if (ret < 0) { |
3276 | if (ret != -ENOSPC) | |
3277 | return ret; | |
3278 | else | |
3279 | goto commit_trans; | |
3280 | } | |
9ed74f2d | 3281 | |
33b4d47f CM |
3282 | if (!data_sinfo) { |
3283 | btrfs_set_inode_space_info(root, inode); | |
3284 | data_sinfo = BTRFS_I(inode)->space_info; | |
3285 | } | |
6a63209f JB |
3286 | goto again; |
3287 | } | |
f2bb8f5c JB |
3288 | |
3289 | /* | |
3290 | * If we have less pinned bytes than we want to allocate then | |
3291 | * don't bother committing the transaction, it won't help us. | |
3292 | */ | |
3293 | if (data_sinfo->bytes_pinned < bytes) | |
3294 | committed = 1; | |
6a63209f | 3295 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3296 | |
4e06bdd6 | 3297 | /* commit the current transaction and try again */ |
d52a5b5f | 3298 | commit_trans: |
a4abeea4 JB |
3299 | if (!committed && |
3300 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3301 | committed = 1; |
7a7eaa40 | 3302 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3303 | if (IS_ERR(trans)) |
3304 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3305 | ret = btrfs_commit_transaction(trans, root); |
3306 | if (ret) | |
3307 | return ret; | |
3308 | goto again; | |
3309 | } | |
9ed74f2d | 3310 | |
6a63209f JB |
3311 | return -ENOSPC; |
3312 | } | |
3313 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3314 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
e77266e4 CM |
3315 | (u64)(unsigned long)data_sinfo, |
3316 | bytes, 1); | |
6a63209f | 3317 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3318 | |
9ed74f2d | 3319 | return 0; |
9ed74f2d | 3320 | } |
6a63209f | 3321 | |
6a63209f | 3322 | /* |
fb25e914 | 3323 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3324 | */ |
0ca1f7ce | 3325 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3326 | { |
0ca1f7ce | 3327 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3328 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3329 | |
6a63209f JB |
3330 | /* make sure bytes are sectorsize aligned */ |
3331 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3332 | |
6a63209f JB |
3333 | data_sinfo = BTRFS_I(inode)->space_info; |
3334 | spin_lock(&data_sinfo->lock); | |
3335 | data_sinfo->bytes_may_use -= bytes; | |
8c2a3ca2 | 3336 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
e77266e4 CM |
3337 | (u64)(unsigned long)data_sinfo, |
3338 | bytes, 0); | |
6a63209f | 3339 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3340 | } |
3341 | ||
97e728d4 | 3342 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3343 | { |
97e728d4 JB |
3344 | struct list_head *head = &info->space_info; |
3345 | struct btrfs_space_info *found; | |
e3ccfa98 | 3346 | |
97e728d4 JB |
3347 | rcu_read_lock(); |
3348 | list_for_each_entry_rcu(found, head, list) { | |
3349 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3350 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3351 | } |
97e728d4 | 3352 | rcu_read_unlock(); |
e3ccfa98 JB |
3353 | } |
3354 | ||
e5bc2458 | 3355 | static int should_alloc_chunk(struct btrfs_root *root, |
0e4f8f88 CM |
3356 | struct btrfs_space_info *sinfo, u64 alloc_bytes, |
3357 | int force) | |
32c00aff | 3358 | { |
fb25e914 | 3359 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3360 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3361 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3362 | u64 thresh; |
e3ccfa98 | 3363 | |
0e4f8f88 CM |
3364 | if (force == CHUNK_ALLOC_FORCE) |
3365 | return 1; | |
3366 | ||
fb25e914 JB |
3367 | /* |
3368 | * We need to take into account the global rsv because for all intents | |
3369 | * and purposes it's used space. Don't worry about locking the | |
3370 | * global_rsv, it doesn't change except when the transaction commits. | |
3371 | */ | |
3372 | num_allocated += global_rsv->size; | |
3373 | ||
0e4f8f88 CM |
3374 | /* |
3375 | * in limited mode, we want to have some free space up to | |
3376 | * about 1% of the FS size. | |
3377 | */ | |
3378 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3379 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3380 | thresh = max_t(u64, 64 * 1024 * 1024, |
3381 | div_factor_fine(thresh, 1)); | |
3382 | ||
3383 | if (num_bytes - num_allocated < thresh) | |
3384 | return 1; | |
3385 | } | |
6c41761f | 3386 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 | 3387 | |
cf1d72c9 CM |
3388 | /* 256MB or 2% of the FS */ |
3389 | thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2)); | |
96bdc7dc CM |
3390 | /* system chunks need a much small threshold */ |
3391 | if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3392 | thresh = 32 * 1024 * 1024; | |
e5bc2458 | 3393 | |
cf1d72c9 | 3394 | if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8)) |
14ed0ca6 | 3395 | return 0; |
424499db | 3396 | return 1; |
32c00aff JB |
3397 | } |
3398 | ||
6324fbf3 CM |
3399 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3400 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3401 | u64 flags, int force) |
9ed74f2d | 3402 | { |
6324fbf3 | 3403 | struct btrfs_space_info *space_info; |
97e728d4 | 3404 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3405 | int wait_for_alloc = 0; |
9ed74f2d | 3406 | int ret = 0; |
9ed74f2d | 3407 | |
70922617 | 3408 | BUG_ON(!profile_is_valid(flags, 0)); |
ec44a35c | 3409 | |
6324fbf3 | 3410 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3411 | if (!space_info) { |
3412 | ret = update_space_info(extent_root->fs_info, flags, | |
3413 | 0, 0, &space_info); | |
3414 | BUG_ON(ret); | |
9ed74f2d | 3415 | } |
6324fbf3 | 3416 | BUG_ON(!space_info); |
9ed74f2d | 3417 | |
6d74119f | 3418 | again: |
25179201 | 3419 | spin_lock(&space_info->lock); |
9e622d6b | 3420 | if (force < space_info->force_alloc) |
0e4f8f88 | 3421 | force = space_info->force_alloc; |
25179201 JB |
3422 | if (space_info->full) { |
3423 | spin_unlock(&space_info->lock); | |
6d74119f | 3424 | return 0; |
9ed74f2d JB |
3425 | } |
3426 | ||
0e4f8f88 | 3427 | if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) { |
25179201 | 3428 | spin_unlock(&space_info->lock); |
6d74119f JB |
3429 | return 0; |
3430 | } else if (space_info->chunk_alloc) { | |
3431 | wait_for_alloc = 1; | |
3432 | } else { | |
3433 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3434 | } |
0e4f8f88 | 3435 | |
25179201 | 3436 | spin_unlock(&space_info->lock); |
9ed74f2d | 3437 | |
6d74119f JB |
3438 | mutex_lock(&fs_info->chunk_mutex); |
3439 | ||
3440 | /* | |
3441 | * The chunk_mutex is held throughout the entirety of a chunk | |
3442 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3443 | * other guy is done and we need to recheck and see if we should | |
3444 | * allocate. | |
3445 | */ | |
3446 | if (wait_for_alloc) { | |
3447 | mutex_unlock(&fs_info->chunk_mutex); | |
3448 | wait_for_alloc = 0; | |
3449 | goto again; | |
3450 | } | |
3451 | ||
67377734 JB |
3452 | /* |
3453 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3454 | * allocating mixed chunks instead of individual chunks. | |
3455 | */ | |
3456 | if (btrfs_mixed_space_info(space_info)) | |
3457 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3458 | ||
97e728d4 JB |
3459 | /* |
3460 | * if we're doing a data chunk, go ahead and make sure that | |
3461 | * we keep a reasonable number of metadata chunks allocated in the | |
3462 | * FS as well. | |
3463 | */ | |
9ed74f2d | 3464 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3465 | fs_info->data_chunk_allocations++; |
3466 | if (!(fs_info->data_chunk_allocations % | |
3467 | fs_info->metadata_ratio)) | |
3468 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3469 | } |
3470 | ||
2b82032c | 3471 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
92b8e897 MF |
3472 | if (ret < 0 && ret != -ENOSPC) |
3473 | goto out; | |
3474 | ||
9ed74f2d | 3475 | spin_lock(&space_info->lock); |
9ed74f2d | 3476 | if (ret) |
6324fbf3 | 3477 | space_info->full = 1; |
424499db YZ |
3478 | else |
3479 | ret = 1; | |
6d74119f | 3480 | |
0e4f8f88 | 3481 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3482 | space_info->chunk_alloc = 0; |
9ed74f2d | 3483 | spin_unlock(&space_info->lock); |
92b8e897 | 3484 | out: |
c146afad | 3485 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 3486 | return ret; |
6324fbf3 | 3487 | } |
9ed74f2d | 3488 | |
9ed74f2d | 3489 | /* |
5da9d01b | 3490 | * shrink metadata reservation for delalloc |
9ed74f2d | 3491 | */ |
663350ac | 3492 | static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, |
f104d044 | 3493 | bool wait_ordered) |
5da9d01b | 3494 | { |
0ca1f7ce | 3495 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3496 | struct btrfs_space_info *space_info; |
663350ac | 3497 | struct btrfs_trans_handle *trans; |
5da9d01b YZ |
3498 | u64 reserved; |
3499 | u64 max_reclaim; | |
3500 | u64 reclaimed = 0; | |
b1953bce | 3501 | long time_left; |
877da174 | 3502 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3503 | int loops = 0; |
36e39c40 | 3504 | unsigned long progress; |
5da9d01b | 3505 | |
663350ac | 3506 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 3507 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3508 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3509 | |
3510 | smp_mb(); | |
fb25e914 | 3511 | reserved = space_info->bytes_may_use; |
36e39c40 | 3512 | progress = space_info->reservation_progress; |
5da9d01b YZ |
3513 | |
3514 | if (reserved == 0) | |
3515 | return 0; | |
c4f675cd | 3516 | |
fdb5effd JB |
3517 | smp_mb(); |
3518 | if (root->fs_info->delalloc_bytes == 0) { | |
3519 | if (trans) | |
3520 | return 0; | |
3521 | btrfs_wait_ordered_extents(root, 0, 0); | |
3522 | return 0; | |
3523 | } | |
3524 | ||
5da9d01b | 3525 | max_reclaim = min(reserved, to_reclaim); |
877da174 JB |
3526 | nr_pages = max_t(unsigned long, nr_pages, |
3527 | max_reclaim >> PAGE_CACHE_SHIFT); | |
b1953bce | 3528 | while (loops < 1024) { |
bf9022e0 CM |
3529 | /* have the flusher threads jump in and do some IO */ |
3530 | smp_mb(); | |
3531 | nr_pages = min_t(unsigned long, nr_pages, | |
3532 | root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT); | |
0e175a18 CW |
3533 | writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages, |
3534 | WB_REASON_FS_FREE_SPACE); | |
5da9d01b | 3535 | |
0019f10d | 3536 | spin_lock(&space_info->lock); |
fb25e914 JB |
3537 | if (reserved > space_info->bytes_may_use) |
3538 | reclaimed += reserved - space_info->bytes_may_use; | |
3539 | reserved = space_info->bytes_may_use; | |
0019f10d | 3540 | spin_unlock(&space_info->lock); |
5da9d01b | 3541 | |
36e39c40 CM |
3542 | loops++; |
3543 | ||
5da9d01b YZ |
3544 | if (reserved == 0 || reclaimed >= max_reclaim) |
3545 | break; | |
3546 | ||
3547 | if (trans && trans->transaction->blocked) | |
3548 | return -EAGAIN; | |
bf9022e0 | 3549 | |
f104d044 JB |
3550 | if (wait_ordered && !trans) { |
3551 | btrfs_wait_ordered_extents(root, 0, 0); | |
3552 | } else { | |
3553 | time_left = schedule_timeout_interruptible(1); | |
b1953bce | 3554 | |
f104d044 JB |
3555 | /* We were interrupted, exit */ |
3556 | if (time_left) | |
3557 | break; | |
3558 | } | |
b1953bce | 3559 | |
36e39c40 CM |
3560 | /* we've kicked the IO a few times, if anything has been freed, |
3561 | * exit. There is no sense in looping here for a long time | |
3562 | * when we really need to commit the transaction, or there are | |
3563 | * just too many writers without enough free space | |
3564 | */ | |
3565 | ||
3566 | if (loops > 3) { | |
3567 | smp_mb(); | |
3568 | if (progress != space_info->reservation_progress) | |
3569 | break; | |
3570 | } | |
bf9022e0 | 3571 | |
5da9d01b | 3572 | } |
f104d044 | 3573 | |
5da9d01b YZ |
3574 | return reclaimed >= to_reclaim; |
3575 | } | |
3576 | ||
663350ac JB |
3577 | /** |
3578 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
3579 | * @root - the root we're allocating for | |
3580 | * @bytes - the number of bytes we want to reserve | |
3581 | * @force - force the commit | |
8bb8ab2e | 3582 | * |
663350ac JB |
3583 | * This will check to make sure that committing the transaction will actually |
3584 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
3585 | * will return -ENOSPC. | |
8bb8ab2e | 3586 | */ |
663350ac JB |
3587 | static int may_commit_transaction(struct btrfs_root *root, |
3588 | struct btrfs_space_info *space_info, | |
3589 | u64 bytes, int force) | |
3590 | { | |
3591 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
3592 | struct btrfs_trans_handle *trans; | |
3593 | ||
3594 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
3595 | if (trans) | |
3596 | return -EAGAIN; | |
3597 | ||
3598 | if (force) | |
3599 | goto commit; | |
3600 | ||
3601 | /* See if there is enough pinned space to make this reservation */ | |
3602 | spin_lock(&space_info->lock); | |
3603 | if (space_info->bytes_pinned >= bytes) { | |
3604 | spin_unlock(&space_info->lock); | |
3605 | goto commit; | |
3606 | } | |
3607 | spin_unlock(&space_info->lock); | |
3608 | ||
3609 | /* | |
3610 | * See if there is some space in the delayed insertion reservation for | |
3611 | * this reservation. | |
3612 | */ | |
3613 | if (space_info != delayed_rsv->space_info) | |
3614 | return -ENOSPC; | |
3615 | ||
d9b0218f | 3616 | spin_lock(&space_info->lock); |
663350ac | 3617 | spin_lock(&delayed_rsv->lock); |
d9b0218f | 3618 | if (space_info->bytes_pinned + delayed_rsv->size < bytes) { |
663350ac | 3619 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 3620 | spin_unlock(&space_info->lock); |
663350ac JB |
3621 | return -ENOSPC; |
3622 | } | |
3623 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 3624 | spin_unlock(&space_info->lock); |
663350ac JB |
3625 | |
3626 | commit: | |
3627 | trans = btrfs_join_transaction(root); | |
3628 | if (IS_ERR(trans)) | |
3629 | return -ENOSPC; | |
3630 | ||
3631 | return btrfs_commit_transaction(trans, root); | |
3632 | } | |
3633 | ||
4a92b1b8 JB |
3634 | /** |
3635 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
3636 | * @root - the root we're allocating for | |
3637 | * @block_rsv - the block_rsv we're allocating for | |
3638 | * @orig_bytes - the number of bytes we want | |
3639 | * @flush - wether or not we can flush to make our reservation | |
8bb8ab2e | 3640 | * |
4a92b1b8 JB |
3641 | * This will reserve orgi_bytes number of bytes from the space info associated |
3642 | * with the block_rsv. If there is not enough space it will make an attempt to | |
3643 | * flush out space to make room. It will do this by flushing delalloc if | |
3644 | * possible or committing the transaction. If flush is 0 then no attempts to | |
3645 | * regain reservations will be made and this will fail if there is not enough | |
3646 | * space already. | |
8bb8ab2e | 3647 | */ |
4a92b1b8 | 3648 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e JB |
3649 | struct btrfs_block_rsv *block_rsv, |
3650 | u64 orig_bytes, int flush) | |
9ed74f2d | 3651 | { |
f0486c68 | 3652 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 3653 | u64 used; |
8bb8ab2e JB |
3654 | u64 num_bytes = orig_bytes; |
3655 | int retries = 0; | |
3656 | int ret = 0; | |
38227933 | 3657 | bool committed = false; |
fdb5effd | 3658 | bool flushing = false; |
f104d044 | 3659 | bool wait_ordered = false; |
9ed74f2d | 3660 | |
8bb8ab2e | 3661 | again: |
fdb5effd | 3662 | ret = 0; |
8bb8ab2e | 3663 | spin_lock(&space_info->lock); |
fdb5effd JB |
3664 | /* |
3665 | * We only want to wait if somebody other than us is flushing and we are | |
3666 | * actually alloed to flush. | |
3667 | */ | |
3668 | while (flush && !flushing && space_info->flush) { | |
3669 | spin_unlock(&space_info->lock); | |
3670 | /* | |
3671 | * If we have a trans handle we can't wait because the flusher | |
3672 | * may have to commit the transaction, which would mean we would | |
3673 | * deadlock since we are waiting for the flusher to finish, but | |
3674 | * hold the current transaction open. | |
3675 | */ | |
663350ac | 3676 | if (current->journal_info) |
fdb5effd JB |
3677 | return -EAGAIN; |
3678 | ret = wait_event_interruptible(space_info->wait, | |
3679 | !space_info->flush); | |
3680 | /* Must have been interrupted, return */ | |
3681 | if (ret) | |
3682 | return -EINTR; | |
3683 | ||
3684 | spin_lock(&space_info->lock); | |
3685 | } | |
3686 | ||
3687 | ret = -ENOSPC; | |
2bf64758 JB |
3688 | used = space_info->bytes_used + space_info->bytes_reserved + |
3689 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3690 | space_info->bytes_may_use; | |
9ed74f2d | 3691 | |
8bb8ab2e JB |
3692 | /* |
3693 | * The idea here is that we've not already over-reserved the block group | |
3694 | * then we can go ahead and save our reservation first and then start | |
3695 | * flushing if we need to. Otherwise if we've already overcommitted | |
3696 | * lets start flushing stuff first and then come back and try to make | |
3697 | * our reservation. | |
3698 | */ | |
2bf64758 JB |
3699 | if (used <= space_info->total_bytes) { |
3700 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 3701 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3702 | trace_btrfs_space_reservation(root->fs_info, |
e77266e4 CM |
3703 | "space_info", |
3704 | (u64)(unsigned long)space_info, | |
3705 | orig_bytes, 1); | |
8bb8ab2e JB |
3706 | ret = 0; |
3707 | } else { | |
3708 | /* | |
3709 | * Ok set num_bytes to orig_bytes since we aren't | |
3710 | * overocmmitted, this way we only try and reclaim what | |
3711 | * we need. | |
3712 | */ | |
3713 | num_bytes = orig_bytes; | |
3714 | } | |
3715 | } else { | |
3716 | /* | |
3717 | * Ok we're over committed, set num_bytes to the overcommitted | |
3718 | * amount plus the amount of bytes that we need for this | |
3719 | * reservation. | |
3720 | */ | |
f104d044 | 3721 | wait_ordered = true; |
2bf64758 | 3722 | num_bytes = used - space_info->total_bytes + |
8bb8ab2e JB |
3723 | (orig_bytes * (retries + 1)); |
3724 | } | |
9ed74f2d | 3725 | |
36ba022a | 3726 | if (ret) { |
2bf64758 JB |
3727 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3728 | u64 avail; | |
3729 | ||
7e355b83 JB |
3730 | /* |
3731 | * If we have a lot of space that's pinned, don't bother doing | |
3732 | * the overcommit dance yet and just commit the transaction. | |
3733 | */ | |
3734 | avail = (space_info->total_bytes - space_info->bytes_used) * 8; | |
3735 | do_div(avail, 10); | |
663350ac | 3736 | if (space_info->bytes_pinned >= avail && flush && !committed) { |
7e355b83 JB |
3737 | space_info->flush = 1; |
3738 | flushing = true; | |
3739 | spin_unlock(&space_info->lock); | |
663350ac JB |
3740 | ret = may_commit_transaction(root, space_info, |
3741 | orig_bytes, 1); | |
3742 | if (ret) | |
3743 | goto out; | |
3744 | committed = true; | |
3745 | goto again; | |
7e355b83 JB |
3746 | } |
3747 | ||
2bf64758 JB |
3748 | spin_lock(&root->fs_info->free_chunk_lock); |
3749 | avail = root->fs_info->free_chunk_space; | |
3750 | ||
3751 | /* | |
3752 | * If we have dup, raid1 or raid10 then only half of the free | |
3753 | * space is actually useable. | |
3754 | */ | |
3755 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3756 | BTRFS_BLOCK_GROUP_RAID1 | | |
3757 | BTRFS_BLOCK_GROUP_RAID10)) | |
3758 | avail >>= 1; | |
3759 | ||
3760 | /* | |
3761 | * If we aren't flushing don't let us overcommit too much, say | |
3762 | * 1/8th of the space. If we can flush, let it overcommit up to | |
3763 | * 1/2 of the space. | |
3764 | */ | |
3765 | if (flush) | |
3766 | avail >>= 3; | |
3767 | else | |
3768 | avail >>= 1; | |
3769 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3770 | ||
9a82ca65 | 3771 | if (used + num_bytes < space_info->total_bytes + avail) { |
2bf64758 | 3772 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3773 | trace_btrfs_space_reservation(root->fs_info, |
e77266e4 CM |
3774 | "space_info", |
3775 | (u64)(unsigned long)space_info, | |
3776 | orig_bytes, 1); | |
2bf64758 | 3777 | ret = 0; |
f104d044 JB |
3778 | } else { |
3779 | wait_ordered = true; | |
2bf64758 JB |
3780 | } |
3781 | } | |
3782 | ||
8bb8ab2e JB |
3783 | /* |
3784 | * Couldn't make our reservation, save our place so while we're trying | |
3785 | * to reclaim space we can actually use it instead of somebody else | |
3786 | * stealing it from us. | |
3787 | */ | |
fdb5effd JB |
3788 | if (ret && flush) { |
3789 | flushing = true; | |
3790 | space_info->flush = 1; | |
8bb8ab2e | 3791 | } |
9ed74f2d | 3792 | |
f0486c68 | 3793 | spin_unlock(&space_info->lock); |
9ed74f2d | 3794 | |
fdb5effd | 3795 | if (!ret || !flush) |
8bb8ab2e | 3796 | goto out; |
f0486c68 | 3797 | |
8bb8ab2e JB |
3798 | /* |
3799 | * We do synchronous shrinking since we don't actually unreserve | |
3800 | * metadata until after the IO is completed. | |
3801 | */ | |
663350ac | 3802 | ret = shrink_delalloc(root, num_bytes, wait_ordered); |
fdb5effd | 3803 | if (ret < 0) |
8bb8ab2e | 3804 | goto out; |
f0486c68 | 3805 | |
75c195a2 CM |
3806 | ret = 0; |
3807 | ||
8bb8ab2e JB |
3808 | /* |
3809 | * So if we were overcommitted it's possible that somebody else flushed | |
3810 | * out enough space and we simply didn't have enough space to reclaim, | |
3811 | * so go back around and try again. | |
3812 | */ | |
3813 | if (retries < 2) { | |
f104d044 | 3814 | wait_ordered = true; |
8bb8ab2e JB |
3815 | retries++; |
3816 | goto again; | |
3817 | } | |
f0486c68 | 3818 | |
8bb8ab2e | 3819 | ret = -ENOSPC; |
75c195a2 CM |
3820 | if (committed) |
3821 | goto out; | |
3822 | ||
663350ac | 3823 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); |
38227933 | 3824 | if (!ret) { |
38227933 | 3825 | committed = true; |
8bb8ab2e | 3826 | goto again; |
38227933 | 3827 | } |
8bb8ab2e JB |
3828 | |
3829 | out: | |
fdb5effd | 3830 | if (flushing) { |
8bb8ab2e | 3831 | spin_lock(&space_info->lock); |
fdb5effd JB |
3832 | space_info->flush = 0; |
3833 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 3834 | spin_unlock(&space_info->lock); |
f0486c68 | 3835 | } |
f0486c68 YZ |
3836 | return ret; |
3837 | } | |
3838 | ||
3839 | static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans, | |
3840 | struct btrfs_root *root) | |
3841 | { | |
4c13d758 JB |
3842 | struct btrfs_block_rsv *block_rsv = NULL; |
3843 | ||
3844 | if (root->ref_cows || root == root->fs_info->csum_root) | |
f0486c68 | 3845 | block_rsv = trans->block_rsv; |
4c13d758 JB |
3846 | |
3847 | if (!block_rsv) | |
f0486c68 YZ |
3848 | block_rsv = root->block_rsv; |
3849 | ||
3850 | if (!block_rsv) | |
3851 | block_rsv = &root->fs_info->empty_block_rsv; | |
3852 | ||
3853 | return block_rsv; | |
3854 | } | |
3855 | ||
3856 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
3857 | u64 num_bytes) | |
3858 | { | |
3859 | int ret = -ENOSPC; | |
3860 | spin_lock(&block_rsv->lock); | |
3861 | if (block_rsv->reserved >= num_bytes) { | |
3862 | block_rsv->reserved -= num_bytes; | |
3863 | if (block_rsv->reserved < block_rsv->size) | |
3864 | block_rsv->full = 0; | |
3865 | ret = 0; | |
3866 | } | |
3867 | spin_unlock(&block_rsv->lock); | |
3868 | return ret; | |
3869 | } | |
3870 | ||
3871 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3872 | u64 num_bytes, int update_size) | |
3873 | { | |
3874 | spin_lock(&block_rsv->lock); | |
3875 | block_rsv->reserved += num_bytes; | |
3876 | if (update_size) | |
3877 | block_rsv->size += num_bytes; | |
3878 | else if (block_rsv->reserved >= block_rsv->size) | |
3879 | block_rsv->full = 1; | |
3880 | spin_unlock(&block_rsv->lock); | |
3881 | } | |
3882 | ||
8c2a3ca2 JB |
3883 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
3884 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 3885 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
3886 | { |
3887 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
3888 | ||
3889 | spin_lock(&block_rsv->lock); | |
3890 | if (num_bytes == (u64)-1) | |
3891 | num_bytes = block_rsv->size; | |
3892 | block_rsv->size -= num_bytes; | |
3893 | if (block_rsv->reserved >= block_rsv->size) { | |
3894 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3895 | block_rsv->reserved = block_rsv->size; | |
3896 | block_rsv->full = 1; | |
3897 | } else { | |
3898 | num_bytes = 0; | |
3899 | } | |
3900 | spin_unlock(&block_rsv->lock); | |
3901 | ||
3902 | if (num_bytes > 0) { | |
3903 | if (dest) { | |
e9e22899 JB |
3904 | spin_lock(&dest->lock); |
3905 | if (!dest->full) { | |
3906 | u64 bytes_to_add; | |
3907 | ||
3908 | bytes_to_add = dest->size - dest->reserved; | |
3909 | bytes_to_add = min(num_bytes, bytes_to_add); | |
3910 | dest->reserved += bytes_to_add; | |
3911 | if (dest->reserved >= dest->size) | |
3912 | dest->full = 1; | |
3913 | num_bytes -= bytes_to_add; | |
3914 | } | |
3915 | spin_unlock(&dest->lock); | |
3916 | } | |
3917 | if (num_bytes) { | |
f0486c68 | 3918 | spin_lock(&space_info->lock); |
fb25e914 | 3919 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 3920 | trace_btrfs_space_reservation(fs_info, "space_info", |
e77266e4 CM |
3921 | (u64)(unsigned long)space_info, |
3922 | num_bytes, 0); | |
36e39c40 | 3923 | space_info->reservation_progress++; |
f0486c68 | 3924 | spin_unlock(&space_info->lock); |
4e06bdd6 | 3925 | } |
9ed74f2d | 3926 | } |
f0486c68 | 3927 | } |
4e06bdd6 | 3928 | |
f0486c68 YZ |
3929 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
3930 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
3931 | { | |
3932 | int ret; | |
9ed74f2d | 3933 | |
f0486c68 YZ |
3934 | ret = block_rsv_use_bytes(src, num_bytes); |
3935 | if (ret) | |
3936 | return ret; | |
9ed74f2d | 3937 | |
f0486c68 | 3938 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
3939 | return 0; |
3940 | } | |
3941 | ||
f0486c68 | 3942 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv) |
9ed74f2d | 3943 | { |
f0486c68 YZ |
3944 | memset(rsv, 0, sizeof(*rsv)); |
3945 | spin_lock_init(&rsv->lock); | |
f0486c68 YZ |
3946 | } |
3947 | ||
3948 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root) | |
3949 | { | |
3950 | struct btrfs_block_rsv *block_rsv; | |
3951 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 3952 | |
f0486c68 YZ |
3953 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
3954 | if (!block_rsv) | |
3955 | return NULL; | |
9ed74f2d | 3956 | |
f0486c68 | 3957 | btrfs_init_block_rsv(block_rsv); |
f0486c68 YZ |
3958 | block_rsv->space_info = __find_space_info(fs_info, |
3959 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
3960 | return block_rsv; |
3961 | } | |
9ed74f2d | 3962 | |
f0486c68 YZ |
3963 | void btrfs_free_block_rsv(struct btrfs_root *root, |
3964 | struct btrfs_block_rsv *rsv) | |
3965 | { | |
dabdb640 JB |
3966 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
3967 | kfree(rsv); | |
9ed74f2d JB |
3968 | } |
3969 | ||
61b520a9 MX |
3970 | static inline int __block_rsv_add(struct btrfs_root *root, |
3971 | struct btrfs_block_rsv *block_rsv, | |
3972 | u64 num_bytes, int flush) | |
9ed74f2d | 3973 | { |
f0486c68 | 3974 | int ret; |
9ed74f2d | 3975 | |
f0486c68 YZ |
3976 | if (num_bytes == 0) |
3977 | return 0; | |
8bb8ab2e | 3978 | |
61b520a9 | 3979 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
3980 | if (!ret) { |
3981 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
3982 | return 0; | |
3983 | } | |
9ed74f2d | 3984 | |
f0486c68 | 3985 | return ret; |
f0486c68 | 3986 | } |
9ed74f2d | 3987 | |
61b520a9 MX |
3988 | int btrfs_block_rsv_add(struct btrfs_root *root, |
3989 | struct btrfs_block_rsv *block_rsv, | |
3990 | u64 num_bytes) | |
3991 | { | |
3992 | return __block_rsv_add(root, block_rsv, num_bytes, 1); | |
3993 | } | |
3994 | ||
c06a0e12 JB |
3995 | int btrfs_block_rsv_add_noflush(struct btrfs_root *root, |
3996 | struct btrfs_block_rsv *block_rsv, | |
3997 | u64 num_bytes) | |
f0486c68 | 3998 | { |
61b520a9 | 3999 | return __block_rsv_add(root, block_rsv, num_bytes, 0); |
f0486c68 | 4000 | } |
9ed74f2d | 4001 | |
4a92b1b8 | 4002 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4003 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4004 | { |
4005 | u64 num_bytes = 0; | |
f0486c68 | 4006 | int ret = -ENOSPC; |
9ed74f2d | 4007 | |
f0486c68 YZ |
4008 | if (!block_rsv) |
4009 | return 0; | |
9ed74f2d | 4010 | |
f0486c68 | 4011 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4012 | num_bytes = div_factor(block_rsv->size, min_factor); |
4013 | if (block_rsv->reserved >= num_bytes) | |
4014 | ret = 0; | |
4015 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4016 | |
36ba022a JB |
4017 | return ret; |
4018 | } | |
4019 | ||
aa38a711 MX |
4020 | static inline int __btrfs_block_rsv_refill(struct btrfs_root *root, |
4021 | struct btrfs_block_rsv *block_rsv, | |
4022 | u64 min_reserved, int flush) | |
36ba022a JB |
4023 | { |
4024 | u64 num_bytes = 0; | |
4025 | int ret = -ENOSPC; | |
4026 | ||
4027 | if (!block_rsv) | |
4028 | return 0; | |
4029 | ||
4030 | spin_lock(&block_rsv->lock); | |
4031 | num_bytes = min_reserved; | |
13553e52 | 4032 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4033 | ret = 0; |
13553e52 | 4034 | else |
f0486c68 | 4035 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4036 | spin_unlock(&block_rsv->lock); |
13553e52 | 4037 | |
f0486c68 YZ |
4038 | if (!ret) |
4039 | return 0; | |
4040 | ||
aa38a711 | 4041 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4042 | if (!ret) { |
4043 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4044 | return 0; |
6a63209f | 4045 | } |
9ed74f2d | 4046 | |
13553e52 | 4047 | return ret; |
f0486c68 YZ |
4048 | } |
4049 | ||
aa38a711 MX |
4050 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4051 | struct btrfs_block_rsv *block_rsv, | |
4052 | u64 min_reserved) | |
4053 | { | |
4054 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1); | |
4055 | } | |
4056 | ||
4057 | int btrfs_block_rsv_refill_noflush(struct btrfs_root *root, | |
4058 | struct btrfs_block_rsv *block_rsv, | |
4059 | u64 min_reserved) | |
4060 | { | |
4061 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0); | |
4062 | } | |
4063 | ||
f0486c68 YZ |
4064 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, |
4065 | struct btrfs_block_rsv *dst_rsv, | |
4066 | u64 num_bytes) | |
4067 | { | |
4068 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4069 | } | |
4070 | ||
4071 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4072 | struct btrfs_block_rsv *block_rsv, | |
4073 | u64 num_bytes) | |
4074 | { | |
4075 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4076 | if (global_rsv->full || global_rsv == block_rsv || | |
4077 | block_rsv->space_info != global_rsv->space_info) | |
4078 | global_rsv = NULL; | |
8c2a3ca2 JB |
4079 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4080 | num_bytes); | |
6a63209f JB |
4081 | } |
4082 | ||
4083 | /* | |
8929ecfa YZ |
4084 | * helper to calculate size of global block reservation. |
4085 | * the desired value is sum of space used by extent tree, | |
4086 | * checksum tree and root tree | |
6a63209f | 4087 | */ |
8929ecfa | 4088 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4089 | { |
8929ecfa YZ |
4090 | struct btrfs_space_info *sinfo; |
4091 | u64 num_bytes; | |
4092 | u64 meta_used; | |
4093 | u64 data_used; | |
6c41761f | 4094 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4095 | |
8929ecfa YZ |
4096 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4097 | spin_lock(&sinfo->lock); | |
4098 | data_used = sinfo->bytes_used; | |
4099 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4100 | |
8929ecfa YZ |
4101 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4102 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4103 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4104 | data_used = 0; | |
8929ecfa YZ |
4105 | meta_used = sinfo->bytes_used; |
4106 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4107 | |
8929ecfa YZ |
4108 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4109 | csum_size * 2; | |
4110 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4111 | |
8929ecfa | 4112 | if (num_bytes * 3 > meta_used) |
5500cdbe | 4113 | num_bytes = div64_u64(meta_used, 3) * 2; |
ab6e2410 | 4114 | |
8929ecfa YZ |
4115 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4116 | } | |
6a63209f | 4117 | |
8929ecfa YZ |
4118 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4119 | { | |
4120 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4121 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4122 | u64 num_bytes; | |
6a63209f | 4123 | |
8929ecfa | 4124 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4125 | |
8929ecfa YZ |
4126 | spin_lock(&block_rsv->lock); |
4127 | spin_lock(&sinfo->lock); | |
4e06bdd6 | 4128 | |
8929ecfa | 4129 | block_rsv->size = num_bytes; |
4e06bdd6 | 4130 | |
8929ecfa | 4131 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4132 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4133 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4134 | |
4135 | if (sinfo->total_bytes > num_bytes) { | |
4136 | num_bytes = sinfo->total_bytes - num_bytes; | |
4137 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4138 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4139 | trace_btrfs_space_reservation(fs_info, "space_info", |
e77266e4 | 4140 | (u64)(unsigned long)sinfo, num_bytes, 1); |
6a63209f | 4141 | } |
6a63209f | 4142 | |
8929ecfa YZ |
4143 | if (block_rsv->reserved >= block_rsv->size) { |
4144 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4145 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4146 | trace_btrfs_space_reservation(fs_info, "space_info", |
e77266e4 | 4147 | (u64)(unsigned long)sinfo, num_bytes, 0); |
36e39c40 | 4148 | sinfo->reservation_progress++; |
8929ecfa YZ |
4149 | block_rsv->reserved = block_rsv->size; |
4150 | block_rsv->full = 1; | |
4151 | } | |
182608c8 | 4152 | |
8929ecfa YZ |
4153 | spin_unlock(&sinfo->lock); |
4154 | spin_unlock(&block_rsv->lock); | |
6a63209f JB |
4155 | } |
4156 | ||
f0486c68 | 4157 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4158 | { |
f0486c68 | 4159 | struct btrfs_space_info *space_info; |
6a63209f | 4160 | |
f0486c68 YZ |
4161 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4162 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4163 | |
f0486c68 | 4164 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4165 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4166 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4167 | fs_info->trans_block_rsv.space_info = space_info; |
4168 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4169 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4170 | |
8929ecfa YZ |
4171 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4172 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4173 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4174 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4175 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4176 | |
8929ecfa | 4177 | update_global_block_rsv(fs_info); |
6a63209f JB |
4178 | } |
4179 | ||
8929ecfa | 4180 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4181 | { |
8c2a3ca2 JB |
4182 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4183 | (u64)-1); | |
8929ecfa YZ |
4184 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4185 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4186 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4187 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4188 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4189 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4190 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4191 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4192 | } |
4193 | ||
a22285a6 YZ |
4194 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4195 | struct btrfs_root *root) | |
6a63209f | 4196 | { |
a22285a6 YZ |
4197 | if (!trans->bytes_reserved) |
4198 | return; | |
6a63209f | 4199 | |
e77266e4 CM |
4200 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
4201 | (u64)(unsigned long)trans, | |
8c2a3ca2 | 4202 | trans->bytes_reserved, 0); |
b24e03db | 4203 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4204 | trans->bytes_reserved = 0; |
4205 | } | |
6a63209f | 4206 | |
d68fc57b YZ |
4207 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4208 | struct inode *inode) | |
4209 | { | |
4210 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4211 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4212 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4213 | ||
4214 | /* | |
fcb80c2a JB |
4215 | * We need to hold space in order to delete our orphan item once we've |
4216 | * added it, so this takes the reservation so we can release it later | |
4217 | * when we are truly done with the orphan item. | |
d68fc57b | 4218 | */ |
ff5714cc | 4219 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4220 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4221 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4222 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4223 | } |
4224 | ||
d68fc57b | 4225 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4226 | { |
d68fc57b | 4227 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4228 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4229 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4230 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4231 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4232 | } | |
97e728d4 | 4233 | |
a22285a6 YZ |
4234 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
4235 | struct btrfs_pending_snapshot *pending) | |
4236 | { | |
4237 | struct btrfs_root *root = pending->root; | |
4238 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4239 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
4240 | /* | |
4241 | * two for root back/forward refs, two for directory entries | |
4242 | * and one for root of the snapshot. | |
4243 | */ | |
16cdcec7 | 4244 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); |
a22285a6 YZ |
4245 | dst_rsv->space_info = src_rsv->space_info; |
4246 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
4247 | } |
4248 | ||
7709cde3 JB |
4249 | /** |
4250 | * drop_outstanding_extent - drop an outstanding extent | |
4251 | * @inode: the inode we're dropping the extent for | |
4252 | * | |
4253 | * This is called when we are freeing up an outstanding extent, either called | |
4254 | * after an error or after an extent is written. This will return the number of | |
4255 | * reserved extents that need to be freed. This must be called with | |
4256 | * BTRFS_I(inode)->lock held. | |
4257 | */ | |
9e0baf60 JB |
4258 | static unsigned drop_outstanding_extent(struct inode *inode) |
4259 | { | |
7fd2ae21 | 4260 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4261 | unsigned dropped_extents = 0; |
4262 | ||
9e0baf60 JB |
4263 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4264 | BTRFS_I(inode)->outstanding_extents--; | |
4265 | ||
7fd2ae21 JB |
4266 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
4267 | BTRFS_I(inode)->delalloc_meta_reserved) { | |
4268 | drop_inode_space = 1; | |
4269 | BTRFS_I(inode)->delalloc_meta_reserved = 0; | |
4270 | } | |
4271 | ||
9e0baf60 JB |
4272 | /* |
4273 | * If we have more or the same amount of outsanding extents than we have | |
4274 | * reserved then we need to leave the reserved extents count alone. | |
4275 | */ | |
4276 | if (BTRFS_I(inode)->outstanding_extents >= | |
4277 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4278 | return drop_inode_space; |
9e0baf60 JB |
4279 | |
4280 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4281 | BTRFS_I(inode)->outstanding_extents; | |
4282 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4283 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4284 | } |
4285 | ||
7709cde3 JB |
4286 | /** |
4287 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4288 | * reserved/free'd for the given bytes. | |
4289 | * @inode: the inode we're manipulating | |
4290 | * @num_bytes: the number of bytes in question | |
4291 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4292 | * | |
4293 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4294 | * correct amount of metadata that must either be reserved or freed. We | |
4295 | * calculate how many checksums we can fit into one leaf and then divide the | |
4296 | * number of bytes that will need to be checksumed by this value to figure out | |
4297 | * how many checksums will be required. If we are adding bytes then the number | |
4298 | * may go up and we will return the number of additional bytes that must be | |
4299 | * reserved. If it is going down we will return the number of bytes that must | |
4300 | * be freed. | |
4301 | * | |
4302 | * This must be called with BTRFS_I(inode)->lock held. | |
4303 | */ | |
4304 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4305 | int reserve) | |
6324fbf3 | 4306 | { |
7709cde3 JB |
4307 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4308 | u64 csum_size; | |
4309 | int num_csums_per_leaf; | |
4310 | int num_csums; | |
4311 | int old_csums; | |
4312 | ||
4313 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4314 | BTRFS_I(inode)->csum_bytes == 0) | |
4315 | return 0; | |
4316 | ||
4317 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4318 | if (reserve) | |
4319 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4320 | else | |
4321 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4322 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4323 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4324 | sizeof(struct btrfs_csum_item) + | |
4325 | sizeof(struct btrfs_disk_key)); | |
4326 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4327 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4328 | num_csums = num_csums / num_csums_per_leaf; | |
4329 | ||
4330 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4331 | old_csums = old_csums / num_csums_per_leaf; | |
4332 | ||
4333 | /* No change, no need to reserve more */ | |
4334 | if (old_csums == num_csums) | |
4335 | return 0; | |
4336 | ||
4337 | if (reserve) | |
4338 | return btrfs_calc_trans_metadata_size(root, | |
4339 | num_csums - old_csums); | |
4340 | ||
4341 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4342 | } |
c146afad | 4343 | |
0ca1f7ce YZ |
4344 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4345 | { | |
4346 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4347 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4348 | u64 to_reserve = 0; |
660d3f6c | 4349 | u64 csum_bytes; |
9e0baf60 | 4350 | unsigned nr_extents = 0; |
660d3f6c | 4351 | int extra_reserve = 0; |
c09544e0 | 4352 | int flush = 1; |
0ca1f7ce | 4353 | int ret; |
6324fbf3 | 4354 | |
660d3f6c | 4355 | /* Need to be holding the i_mutex here if we aren't free space cache */ |
c09544e0 JB |
4356 | if (btrfs_is_free_space_inode(root, inode)) |
4357 | flush = 0; | |
4358 | ||
4359 | if (flush && btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4360 | schedule_timeout(1); |
ec44a35c | 4361 | |
f248679e | 4362 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4363 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4364 | |
9e0baf60 JB |
4365 | spin_lock(&BTRFS_I(inode)->lock); |
4366 | BTRFS_I(inode)->outstanding_extents++; | |
4367 | ||
4368 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4369 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4370 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4371 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4372 | |
7fd2ae21 JB |
4373 | /* |
4374 | * Add an item to reserve for updating the inode when we complete the | |
4375 | * delalloc io. | |
4376 | */ | |
4377 | if (!BTRFS_I(inode)->delalloc_meta_reserved) { | |
4378 | nr_extents++; | |
660d3f6c | 4379 | extra_reserve = 1; |
593060d7 | 4380 | } |
7fd2ae21 JB |
4381 | |
4382 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4383 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4384 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4385 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4386 | |
36ba022a | 4387 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
9e0baf60 | 4388 | if (ret) { |
7ed49f18 | 4389 | u64 to_free = 0; |
9e0baf60 | 4390 | unsigned dropped; |
7ed49f18 | 4391 | |
7709cde3 | 4392 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4393 | dropped = drop_outstanding_extent(inode); |
9e0baf60 | 4394 | /* |
660d3f6c JB |
4395 | * If the inodes csum_bytes is the same as the original |
4396 | * csum_bytes then we know we haven't raced with any free()ers | |
4397 | * so we can just reduce our inodes csum bytes and carry on. | |
4398 | * Otherwise we have to do the normal free thing to account for | |
4399 | * the case that the free side didn't free up its reserve | |
4400 | * because of this outstanding reservation. | |
9e0baf60 | 4401 | */ |
660d3f6c JB |
4402 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) |
4403 | calc_csum_metadata_size(inode, num_bytes, 0); | |
4404 | else | |
4405 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
4406 | spin_unlock(&BTRFS_I(inode)->lock); | |
4407 | if (dropped) | |
4408 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
4409 | ||
8c2a3ca2 | 4410 | if (to_free) { |
7ed49f18 | 4411 | btrfs_block_rsv_release(root, block_rsv, to_free); |
8c2a3ca2 JB |
4412 | trace_btrfs_space_reservation(root->fs_info, |
4413 | "delalloc", | |
4414 | btrfs_ino(inode), | |
4415 | to_free, 0); | |
4416 | } | |
f248679e | 4417 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4418 | return ret; |
9e0baf60 | 4419 | } |
25179201 | 4420 | |
660d3f6c JB |
4421 | spin_lock(&BTRFS_I(inode)->lock); |
4422 | if (extra_reserve) { | |
4423 | BTRFS_I(inode)->delalloc_meta_reserved = 1; | |
4424 | nr_extents--; | |
4425 | } | |
4426 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
4427 | spin_unlock(&BTRFS_I(inode)->lock); | |
f248679e | 4428 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
660d3f6c | 4429 | |
8c2a3ca2 JB |
4430 | if (to_reserve) |
4431 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
4432 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
4433 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4434 | ||
0ca1f7ce YZ |
4435 | return 0; |
4436 | } | |
4437 | ||
7709cde3 JB |
4438 | /** |
4439 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
4440 | * @inode: the inode to release the reservation for | |
4441 | * @num_bytes: the number of bytes we're releasing | |
4442 | * | |
4443 | * This will release the metadata reservation for an inode. This can be called | |
4444 | * once we complete IO for a given set of bytes to release their metadata | |
4445 | * reservations. | |
4446 | */ | |
0ca1f7ce YZ |
4447 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
4448 | { | |
4449 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4450 | u64 to_free = 0; |
4451 | unsigned dropped; | |
0ca1f7ce YZ |
4452 | |
4453 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 4454 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4455 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4456 | |
7709cde3 JB |
4457 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); |
4458 | spin_unlock(&BTRFS_I(inode)->lock); | |
9e0baf60 JB |
4459 | if (dropped > 0) |
4460 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 4461 | |
8c2a3ca2 JB |
4462 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
4463 | btrfs_ino(inode), to_free, 0); | |
0ca1f7ce YZ |
4464 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
4465 | to_free); | |
4466 | } | |
4467 | ||
7709cde3 JB |
4468 | /** |
4469 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
4470 | * @inode: inode we're writing to | |
4471 | * @num_bytes: the number of bytes we want to allocate | |
4472 | * | |
4473 | * This will do the following things | |
4474 | * | |
4475 | * o reserve space in the data space info for num_bytes | |
4476 | * o reserve space in the metadata space info based on number of outstanding | |
4477 | * extents and how much csums will be needed | |
4478 | * o add to the inodes ->delalloc_bytes | |
4479 | * o add it to the fs_info's delalloc inodes list. | |
4480 | * | |
4481 | * This will return 0 for success and -ENOSPC if there is no space left. | |
4482 | */ | |
0ca1f7ce YZ |
4483 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
4484 | { | |
4485 | int ret; | |
4486 | ||
4487 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4488 | if (ret) |
0ca1f7ce YZ |
4489 | return ret; |
4490 | ||
4491 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4492 | if (ret) { | |
4493 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4494 | return ret; | |
4495 | } | |
4496 | ||
4497 | return 0; | |
4498 | } | |
4499 | ||
7709cde3 JB |
4500 | /** |
4501 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
4502 | * @inode: inode we're releasing space for | |
4503 | * @num_bytes: the number of bytes we want to free up | |
4504 | * | |
4505 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
4506 | * called in the case that we don't need the metadata AND data reservations | |
4507 | * anymore. So if there is an error or we insert an inline extent. | |
4508 | * | |
4509 | * This function will release the metadata space that was not used and will | |
4510 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
4511 | * list if there are no delalloc bytes left. | |
4512 | */ | |
0ca1f7ce YZ |
4513 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
4514 | { | |
4515 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4516 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4517 | } |
4518 | ||
9078a3e1 CM |
4519 | static int update_block_group(struct btrfs_trans_handle *trans, |
4520 | struct btrfs_root *root, | |
f0486c68 | 4521 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4522 | { |
0af3d00b | 4523 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4524 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4525 | u64 total = num_bytes; |
9078a3e1 | 4526 | u64 old_val; |
db94535d | 4527 | u64 byte_in_group; |
0af3d00b | 4528 | int factor; |
3e1ad54f | 4529 | |
5d4f98a2 YZ |
4530 | /* block accounting for super block */ |
4531 | spin_lock(&info->delalloc_lock); | |
6c41761f | 4532 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
4533 | if (alloc) |
4534 | old_val += num_bytes; | |
4535 | else | |
4536 | old_val -= num_bytes; | |
6c41761f | 4537 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
5d4f98a2 YZ |
4538 | spin_unlock(&info->delalloc_lock); |
4539 | ||
d397712b | 4540 | while (total) { |
db94535d | 4541 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4542 | if (!cache) |
9078a3e1 | 4543 | return -1; |
b742bb82 YZ |
4544 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4545 | BTRFS_BLOCK_GROUP_RAID1 | | |
4546 | BTRFS_BLOCK_GROUP_RAID10)) | |
4547 | factor = 2; | |
4548 | else | |
4549 | factor = 1; | |
9d66e233 JB |
4550 | /* |
4551 | * If this block group has free space cache written out, we | |
4552 | * need to make sure to load it if we are removing space. This | |
4553 | * is because we need the unpinning stage to actually add the | |
4554 | * space back to the block group, otherwise we will leak space. | |
4555 | */ | |
4556 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4557 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4558 | |
db94535d CM |
4559 | byte_in_group = bytenr - cache->key.objectid; |
4560 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4561 | |
25179201 | 4562 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4563 | spin_lock(&cache->lock); |
0af3d00b | 4564 | |
73bc1876 | 4565 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
4566 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
4567 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4568 | ||
0f9dd46c | 4569 | cache->dirty = 1; |
9078a3e1 | 4570 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4571 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4572 | if (alloc) { |
db94535d | 4573 | old_val += num_bytes; |
11833d66 YZ |
4574 | btrfs_set_block_group_used(&cache->item, old_val); |
4575 | cache->reserved -= num_bytes; | |
11833d66 | 4576 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
4577 | cache->space_info->bytes_used += num_bytes; |
4578 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4579 | spin_unlock(&cache->lock); |
25179201 | 4580 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4581 | } else { |
db94535d | 4582 | old_val -= num_bytes; |
c286ac48 | 4583 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4584 | cache->pinned += num_bytes; |
4585 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4586 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4587 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4588 | spin_unlock(&cache->lock); |
25179201 | 4589 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4590 | |
f0486c68 YZ |
4591 | set_extent_dirty(info->pinned_extents, |
4592 | bytenr, bytenr + num_bytes - 1, | |
4593 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4594 | } |
fa9c0d79 | 4595 | btrfs_put_block_group(cache); |
db94535d CM |
4596 | total -= num_bytes; |
4597 | bytenr += num_bytes; | |
9078a3e1 CM |
4598 | } |
4599 | return 0; | |
4600 | } | |
6324fbf3 | 4601 | |
a061fc8d CM |
4602 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4603 | { | |
0f9dd46c | 4604 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4605 | u64 bytenr; |
0f9dd46c JB |
4606 | |
4607 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4608 | if (!cache) | |
a061fc8d | 4609 | return 0; |
0f9dd46c | 4610 | |
d2fb3437 | 4611 | bytenr = cache->key.objectid; |
fa9c0d79 | 4612 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4613 | |
4614 | return bytenr; | |
a061fc8d CM |
4615 | } |
4616 | ||
f0486c68 YZ |
4617 | static int pin_down_extent(struct btrfs_root *root, |
4618 | struct btrfs_block_group_cache *cache, | |
4619 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4620 | { |
11833d66 YZ |
4621 | spin_lock(&cache->space_info->lock); |
4622 | spin_lock(&cache->lock); | |
4623 | cache->pinned += num_bytes; | |
4624 | cache->space_info->bytes_pinned += num_bytes; | |
4625 | if (reserved) { | |
4626 | cache->reserved -= num_bytes; | |
4627 | cache->space_info->bytes_reserved -= num_bytes; | |
4628 | } | |
4629 | spin_unlock(&cache->lock); | |
4630 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4631 | |
f0486c68 YZ |
4632 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4633 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4634 | return 0; | |
4635 | } | |
68b38550 | 4636 | |
f0486c68 YZ |
4637 | /* |
4638 | * this function must be called within transaction | |
4639 | */ | |
4640 | int btrfs_pin_extent(struct btrfs_root *root, | |
4641 | u64 bytenr, u64 num_bytes, int reserved) | |
4642 | { | |
4643 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4644 | |
f0486c68 YZ |
4645 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
4646 | BUG_ON(!cache); | |
4647 | ||
4648 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4649 | ||
4650 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4651 | return 0; |
4652 | } | |
4653 | ||
f0486c68 | 4654 | /* |
e688b725 CM |
4655 | * this function must be called within transaction |
4656 | */ | |
4657 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, | |
4658 | struct btrfs_root *root, | |
4659 | u64 bytenr, u64 num_bytes) | |
4660 | { | |
4661 | struct btrfs_block_group_cache *cache; | |
4662 | ||
4663 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
4664 | BUG_ON(!cache); | |
4665 | ||
4666 | /* | |
4667 | * pull in the free space cache (if any) so that our pin | |
4668 | * removes the free space from the cache. We have load_only set | |
4669 | * to one because the slow code to read in the free extents does check | |
4670 | * the pinned extents. | |
4671 | */ | |
4672 | cache_block_group(cache, trans, root, 1); | |
4673 | ||
4674 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
4675 | ||
4676 | /* remove us from the free space cache (if we're there at all) */ | |
4677 | btrfs_remove_free_space(cache, bytenr, num_bytes); | |
4678 | btrfs_put_block_group(cache); | |
4679 | return 0; | |
4680 | } | |
4681 | ||
fb25e914 JB |
4682 | /** |
4683 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
4684 | * @cache: The cache we are manipulating | |
4685 | * @num_bytes: The number of bytes in question | |
4686 | * @reserve: One of the reservation enums | |
4687 | * | |
4688 | * This is called by the allocator when it reserves space, or by somebody who is | |
4689 | * freeing space that was never actually used on disk. For example if you | |
4690 | * reserve some space for a new leaf in transaction A and before transaction A | |
4691 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
4692 | * clear the reservation. | |
4693 | * | |
4694 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
4695 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
4696 | * delalloc bits in the io_tree. We have to do this since we could end up | |
4697 | * allocating less disk space for the amount of data we have reserved in the | |
4698 | * case of compression. | |
4699 | * | |
4700 | * If this is a reservation and the block group has become read only we cannot | |
4701 | * make the reservation and return -EAGAIN, otherwise this function always | |
4702 | * succeeds. | |
f0486c68 | 4703 | */ |
fb25e914 JB |
4704 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4705 | u64 num_bytes, int reserve) | |
11833d66 | 4706 | { |
fb25e914 | 4707 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 4708 | int ret = 0; |
fb25e914 JB |
4709 | spin_lock(&space_info->lock); |
4710 | spin_lock(&cache->lock); | |
4711 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
4712 | if (cache->ro) { |
4713 | ret = -EAGAIN; | |
4714 | } else { | |
fb25e914 JB |
4715 | cache->reserved += num_bytes; |
4716 | space_info->bytes_reserved += num_bytes; | |
4717 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 4718 | trace_btrfs_space_reservation(cache->fs_info, |
e77266e4 CM |
4719 | "space_info", |
4720 | (u64)(unsigned long)space_info, | |
4721 | num_bytes, 0); | |
fb25e914 JB |
4722 | space_info->bytes_may_use -= num_bytes; |
4723 | } | |
f0486c68 | 4724 | } |
fb25e914 JB |
4725 | } else { |
4726 | if (cache->ro) | |
4727 | space_info->bytes_readonly += num_bytes; | |
4728 | cache->reserved -= num_bytes; | |
4729 | space_info->bytes_reserved -= num_bytes; | |
4730 | space_info->reservation_progress++; | |
324ae4df | 4731 | } |
fb25e914 JB |
4732 | spin_unlock(&cache->lock); |
4733 | spin_unlock(&space_info->lock); | |
f0486c68 | 4734 | return ret; |
324ae4df | 4735 | } |
9078a3e1 | 4736 | |
11833d66 YZ |
4737 | int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
4738 | struct btrfs_root *root) | |
e8569813 | 4739 | { |
e8569813 | 4740 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4741 | struct btrfs_caching_control *next; |
4742 | struct btrfs_caching_control *caching_ctl; | |
4743 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4744 | |
11833d66 | 4745 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4746 | |
11833d66 YZ |
4747 | list_for_each_entry_safe(caching_ctl, next, |
4748 | &fs_info->caching_block_groups, list) { | |
4749 | cache = caching_ctl->block_group; | |
4750 | if (block_group_cache_done(cache)) { | |
4751 | cache->last_byte_to_unpin = (u64)-1; | |
4752 | list_del_init(&caching_ctl->list); | |
4753 | put_caching_control(caching_ctl); | |
e8569813 | 4754 | } else { |
11833d66 | 4755 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 4756 | } |
e8569813 | 4757 | } |
11833d66 YZ |
4758 | |
4759 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
4760 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
4761 | else | |
4762 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
4763 | ||
4764 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
4765 | |
4766 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
4767 | return 0; |
4768 | } | |
4769 | ||
11833d66 | 4770 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 4771 | { |
11833d66 YZ |
4772 | struct btrfs_fs_info *fs_info = root->fs_info; |
4773 | struct btrfs_block_group_cache *cache = NULL; | |
4774 | u64 len; | |
ccd467d6 | 4775 | |
11833d66 YZ |
4776 | while (start <= end) { |
4777 | if (!cache || | |
4778 | start >= cache->key.objectid + cache->key.offset) { | |
4779 | if (cache) | |
4780 | btrfs_put_block_group(cache); | |
4781 | cache = btrfs_lookup_block_group(fs_info, start); | |
4782 | BUG_ON(!cache); | |
4783 | } | |
4784 | ||
4785 | len = cache->key.objectid + cache->key.offset - start; | |
4786 | len = min(len, end + 1 - start); | |
4787 | ||
4788 | if (start < cache->last_byte_to_unpin) { | |
4789 | len = min(len, cache->last_byte_to_unpin - start); | |
4790 | btrfs_add_free_space(cache, start, len); | |
4791 | } | |
4792 | ||
f0486c68 YZ |
4793 | start += len; |
4794 | ||
11833d66 YZ |
4795 | spin_lock(&cache->space_info->lock); |
4796 | spin_lock(&cache->lock); | |
4797 | cache->pinned -= len; | |
4798 | cache->space_info->bytes_pinned -= len; | |
37be25bc | 4799 | if (cache->ro) |
f0486c68 | 4800 | cache->space_info->bytes_readonly += len; |
11833d66 YZ |
4801 | spin_unlock(&cache->lock); |
4802 | spin_unlock(&cache->space_info->lock); | |
ccd467d6 | 4803 | } |
11833d66 YZ |
4804 | |
4805 | if (cache) | |
4806 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
4807 | return 0; |
4808 | } | |
4809 | ||
4810 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 4811 | struct btrfs_root *root) |
a28ec197 | 4812 | { |
11833d66 YZ |
4813 | struct btrfs_fs_info *fs_info = root->fs_info; |
4814 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
4815 | u64 start; |
4816 | u64 end; | |
a28ec197 | 4817 | int ret; |
a28ec197 | 4818 | |
11833d66 YZ |
4819 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
4820 | unpin = &fs_info->freed_extents[1]; | |
4821 | else | |
4822 | unpin = &fs_info->freed_extents[0]; | |
4823 | ||
d397712b | 4824 | while (1) { |
1a5bc167 CM |
4825 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
4826 | EXTENT_DIRTY); | |
4827 | if (ret) | |
a28ec197 | 4828 | break; |
1f3c79a2 | 4829 | |
5378e607 LD |
4830 | if (btrfs_test_opt(root, DISCARD)) |
4831 | ret = btrfs_discard_extent(root, start, | |
4832 | end + 1 - start, NULL); | |
1f3c79a2 | 4833 | |
1a5bc167 | 4834 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 4835 | unpin_extent_range(root, start, end); |
b9473439 | 4836 | cond_resched(); |
a28ec197 | 4837 | } |
817d52f8 | 4838 | |
e20d96d6 CM |
4839 | return 0; |
4840 | } | |
4841 | ||
5d4f98a2 YZ |
4842 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
4843 | struct btrfs_root *root, | |
4844 | u64 bytenr, u64 num_bytes, u64 parent, | |
4845 | u64 root_objectid, u64 owner_objectid, | |
4846 | u64 owner_offset, int refs_to_drop, | |
4847 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 4848 | { |
e2fa7227 | 4849 | struct btrfs_key key; |
5d4f98a2 | 4850 | struct btrfs_path *path; |
1261ec42 CM |
4851 | struct btrfs_fs_info *info = root->fs_info; |
4852 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 4853 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
4854 | struct btrfs_extent_item *ei; |
4855 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 4856 | int ret; |
5d4f98a2 | 4857 | int is_data; |
952fccac CM |
4858 | int extent_slot = 0; |
4859 | int found_extent = 0; | |
4860 | int num_to_del = 1; | |
5d4f98a2 YZ |
4861 | u32 item_size; |
4862 | u64 refs; | |
037e6390 | 4863 | |
5caf2a00 | 4864 | path = btrfs_alloc_path(); |
54aa1f4d CM |
4865 | if (!path) |
4866 | return -ENOMEM; | |
5f26f772 | 4867 | |
3c12ac72 | 4868 | path->reada = 1; |
b9473439 | 4869 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4870 | |
4871 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
4872 | BUG_ON(!is_data && refs_to_drop != 1); | |
4873 | ||
4874 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
4875 | bytenr, num_bytes, parent, | |
4876 | root_objectid, owner_objectid, | |
4877 | owner_offset); | |
7bb86316 | 4878 | if (ret == 0) { |
952fccac | 4879 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
4880 | while (extent_slot >= 0) { |
4881 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 4882 | extent_slot); |
5d4f98a2 | 4883 | if (key.objectid != bytenr) |
952fccac | 4884 | break; |
5d4f98a2 YZ |
4885 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
4886 | key.offset == num_bytes) { | |
952fccac CM |
4887 | found_extent = 1; |
4888 | break; | |
4889 | } | |
4890 | if (path->slots[0] - extent_slot > 5) | |
4891 | break; | |
5d4f98a2 | 4892 | extent_slot--; |
952fccac | 4893 | } |
5d4f98a2 YZ |
4894 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
4895 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
4896 | if (found_extent && item_size < sizeof(*ei)) | |
4897 | found_extent = 0; | |
4898 | #endif | |
31840ae1 | 4899 | if (!found_extent) { |
5d4f98a2 | 4900 | BUG_ON(iref); |
56bec294 | 4901 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
4902 | NULL, refs_to_drop, |
4903 | is_data); | |
31840ae1 | 4904 | BUG_ON(ret); |
b3b4aa74 | 4905 | btrfs_release_path(path); |
b9473439 | 4906 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4907 | |
4908 | key.objectid = bytenr; | |
4909 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4910 | key.offset = num_bytes; | |
4911 | ||
31840ae1 ZY |
4912 | ret = btrfs_search_slot(trans, extent_root, |
4913 | &key, path, -1, 1); | |
f3465ca4 JB |
4914 | if (ret) { |
4915 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
4916 | ", was looking for %llu\n", ret, |
4917 | (unsigned long long)bytenr); | |
b783e62d JB |
4918 | if (ret > 0) |
4919 | btrfs_print_leaf(extent_root, | |
4920 | path->nodes[0]); | |
f3465ca4 | 4921 | } |
31840ae1 ZY |
4922 | BUG_ON(ret); |
4923 | extent_slot = path->slots[0]; | |
4924 | } | |
7bb86316 CM |
4925 | } else { |
4926 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4927 | WARN_ON(1); | |
d397712b | 4928 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 4929 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 4930 | (unsigned long long)bytenr, |
56bec294 | 4931 | (unsigned long long)parent, |
d397712b | 4932 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
4933 | (unsigned long long)owner_objectid, |
4934 | (unsigned long long)owner_offset); | |
7bb86316 | 4935 | } |
5f39d397 CM |
4936 | |
4937 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
4938 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
4939 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
4940 | if (item_size < sizeof(*ei)) { | |
4941 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
4942 | ret = convert_extent_item_v0(trans, extent_root, path, | |
4943 | owner_objectid, 0); | |
4944 | BUG_ON(ret < 0); | |
4945 | ||
b3b4aa74 | 4946 | btrfs_release_path(path); |
5d4f98a2 YZ |
4947 | path->leave_spinning = 1; |
4948 | ||
4949 | key.objectid = bytenr; | |
4950 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4951 | key.offset = num_bytes; | |
4952 | ||
4953 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
4954 | -1, 1); | |
4955 | if (ret) { | |
4956 | printk(KERN_ERR "umm, got %d back from search" | |
4957 | ", was looking for %llu\n", ret, | |
4958 | (unsigned long long)bytenr); | |
4959 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4960 | } | |
4961 | BUG_ON(ret); | |
4962 | extent_slot = path->slots[0]; | |
4963 | leaf = path->nodes[0]; | |
4964 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
4965 | } | |
4966 | #endif | |
4967 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 4968 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 4969 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4970 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
4971 | struct btrfs_tree_block_info *bi; | |
4972 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
4973 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
4974 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
4975 | } | |
56bec294 | 4976 | |
5d4f98a2 | 4977 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
4978 | BUG_ON(refs < refs_to_drop); |
4979 | refs -= refs_to_drop; | |
5f39d397 | 4980 | |
5d4f98a2 YZ |
4981 | if (refs > 0) { |
4982 | if (extent_op) | |
4983 | __run_delayed_extent_op(extent_op, leaf, ei); | |
4984 | /* | |
4985 | * In the case of inline back ref, reference count will | |
4986 | * be updated by remove_extent_backref | |
952fccac | 4987 | */ |
5d4f98a2 YZ |
4988 | if (iref) { |
4989 | BUG_ON(!found_extent); | |
4990 | } else { | |
4991 | btrfs_set_extent_refs(leaf, ei, refs); | |
4992 | btrfs_mark_buffer_dirty(leaf); | |
4993 | } | |
4994 | if (found_extent) { | |
4995 | ret = remove_extent_backref(trans, extent_root, path, | |
4996 | iref, refs_to_drop, | |
4997 | is_data); | |
952fccac CM |
4998 | BUG_ON(ret); |
4999 | } | |
5d4f98a2 | 5000 | } else { |
5d4f98a2 YZ |
5001 | if (found_extent) { |
5002 | BUG_ON(is_data && refs_to_drop != | |
5003 | extent_data_ref_count(root, path, iref)); | |
5004 | if (iref) { | |
5005 | BUG_ON(path->slots[0] != extent_slot); | |
5006 | } else { | |
5007 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5008 | path->slots[0] = extent_slot; | |
5009 | num_to_del = 2; | |
5010 | } | |
78fae27e | 5011 | } |
b9473439 | 5012 | |
952fccac CM |
5013 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5014 | num_to_del); | |
31840ae1 | 5015 | BUG_ON(ret); |
b3b4aa74 | 5016 | btrfs_release_path(path); |
21af804c | 5017 | |
5d4f98a2 | 5018 | if (is_data) { |
459931ec CM |
5019 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
5020 | BUG_ON(ret); | |
5021 | } | |
5022 | ||
f0486c68 | 5023 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
dcbdd4dc | 5024 | BUG_ON(ret); |
a28ec197 | 5025 | } |
5caf2a00 | 5026 | btrfs_free_path(path); |
a28ec197 CM |
5027 | return ret; |
5028 | } | |
5029 | ||
1887be66 | 5030 | /* |
f0486c68 | 5031 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5032 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5033 | * a given extent, and if there are no other delayed refs to be processed, it | |
5034 | * removes it from the tree. | |
5035 | */ | |
5036 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5037 | struct btrfs_root *root, u64 bytenr) | |
5038 | { | |
5039 | struct btrfs_delayed_ref_head *head; | |
5040 | struct btrfs_delayed_ref_root *delayed_refs; | |
5041 | struct btrfs_delayed_ref_node *ref; | |
5042 | struct rb_node *node; | |
f0486c68 | 5043 | int ret = 0; |
1887be66 CM |
5044 | |
5045 | delayed_refs = &trans->transaction->delayed_refs; | |
5046 | spin_lock(&delayed_refs->lock); | |
5047 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5048 | if (!head) | |
5049 | goto out; | |
5050 | ||
5051 | node = rb_prev(&head->node.rb_node); | |
5052 | if (!node) | |
5053 | goto out; | |
5054 | ||
5055 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5056 | ||
5057 | /* there are still entries for this ref, we can't drop it */ | |
5058 | if (ref->bytenr == bytenr) | |
5059 | goto out; | |
5060 | ||
5d4f98a2 YZ |
5061 | if (head->extent_op) { |
5062 | if (!head->must_insert_reserved) | |
5063 | goto out; | |
5064 | kfree(head->extent_op); | |
5065 | head->extent_op = NULL; | |
5066 | } | |
5067 | ||
1887be66 CM |
5068 | /* |
5069 | * waiting for the lock here would deadlock. If someone else has it | |
5070 | * locked they are already in the process of dropping it anyway | |
5071 | */ | |
5072 | if (!mutex_trylock(&head->mutex)) | |
5073 | goto out; | |
5074 | ||
5075 | /* | |
5076 | * at this point we have a head with no other entries. Go | |
5077 | * ahead and process it. | |
5078 | */ | |
5079 | head->node.in_tree = 0; | |
5080 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5081 | |
1887be66 | 5082 | delayed_refs->num_entries--; |
a168650c JS |
5083 | if (waitqueue_active(&delayed_refs->seq_wait)) |
5084 | wake_up(&delayed_refs->seq_wait); | |
1887be66 CM |
5085 | |
5086 | /* | |
5087 | * we don't take a ref on the node because we're removing it from the | |
5088 | * tree, so we just steal the ref the tree was holding. | |
5089 | */ | |
c3e69d58 CM |
5090 | delayed_refs->num_heads--; |
5091 | if (list_empty(&head->cluster)) | |
5092 | delayed_refs->num_heads_ready--; | |
5093 | ||
5094 | list_del_init(&head->cluster); | |
1887be66 CM |
5095 | spin_unlock(&delayed_refs->lock); |
5096 | ||
f0486c68 YZ |
5097 | BUG_ON(head->extent_op); |
5098 | if (head->must_insert_reserved) | |
5099 | ret = 1; | |
5100 | ||
5101 | mutex_unlock(&head->mutex); | |
1887be66 | 5102 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5103 | return ret; |
1887be66 CM |
5104 | out: |
5105 | spin_unlock(&delayed_refs->lock); | |
5106 | return 0; | |
5107 | } | |
5108 | ||
f0486c68 YZ |
5109 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5110 | struct btrfs_root *root, | |
5111 | struct extent_buffer *buf, | |
66d7e7f0 | 5112 | u64 parent, int last_ref, int for_cow) |
f0486c68 | 5113 | { |
f0486c68 YZ |
5114 | struct btrfs_block_group_cache *cache = NULL; |
5115 | int ret; | |
5116 | ||
5117 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5118 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5119 | buf->start, buf->len, | |
5120 | parent, root->root_key.objectid, | |
5121 | btrfs_header_level(buf), | |
5122 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); | |
f0486c68 YZ |
5123 | BUG_ON(ret); |
5124 | } | |
5125 | ||
5126 | if (!last_ref) | |
5127 | return; | |
5128 | ||
f0486c68 | 5129 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5130 | |
5131 | if (btrfs_header_generation(buf) == trans->transid) { | |
5132 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5133 | ret = check_ref_cleanup(trans, root, buf->start); | |
5134 | if (!ret) | |
37be25bc | 5135 | goto out; |
f0486c68 YZ |
5136 | } |
5137 | ||
5138 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5139 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5140 | goto out; |
f0486c68 YZ |
5141 | } |
5142 | ||
5143 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5144 | ||
5145 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5146 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
f0486c68 YZ |
5147 | } |
5148 | out: | |
a826d6dc JB |
5149 | /* |
5150 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
5151 | * anymore. | |
5152 | */ | |
5153 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
5154 | btrfs_put_block_group(cache); |
5155 | } | |
5156 | ||
66d7e7f0 AJ |
5157 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5158 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
5159 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
5160 | { |
5161 | int ret; | |
66d7e7f0 | 5162 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 5163 | |
56bec294 CM |
5164 | /* |
5165 | * tree log blocks never actually go into the extent allocation | |
5166 | * tree, just update pinning info and exit early. | |
56bec294 | 5167 | */ |
5d4f98a2 YZ |
5168 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
5169 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 5170 | /* unlocks the pinned mutex */ |
11833d66 | 5171 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 5172 | ret = 0; |
5d4f98a2 | 5173 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
5174 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
5175 | num_bytes, | |
5d4f98a2 | 5176 | parent, root_objectid, (int)owner, |
66d7e7f0 | 5177 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
1887be66 | 5178 | BUG_ON(ret); |
5d4f98a2 | 5179 | } else { |
66d7e7f0 AJ |
5180 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5181 | num_bytes, | |
5182 | parent, root_objectid, owner, | |
5183 | offset, BTRFS_DROP_DELAYED_REF, | |
5184 | NULL, for_cow); | |
5d4f98a2 | 5185 | BUG_ON(ret); |
56bec294 | 5186 | } |
925baedd CM |
5187 | return ret; |
5188 | } | |
5189 | ||
87ee04eb CM |
5190 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
5191 | { | |
5192 | u64 mask = ((u64)root->stripesize - 1); | |
5193 | u64 ret = (val + mask) & ~mask; | |
5194 | return ret; | |
5195 | } | |
5196 | ||
817d52f8 JB |
5197 | /* |
5198 | * when we wait for progress in the block group caching, its because | |
5199 | * our allocation attempt failed at least once. So, we must sleep | |
5200 | * and let some progress happen before we try again. | |
5201 | * | |
5202 | * This function will sleep at least once waiting for new free space to | |
5203 | * show up, and then it will check the block group free space numbers | |
5204 | * for our min num_bytes. Another option is to have it go ahead | |
5205 | * and look in the rbtree for a free extent of a given size, but this | |
5206 | * is a good start. | |
5207 | */ | |
5208 | static noinline int | |
5209 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
5210 | u64 num_bytes) | |
5211 | { | |
11833d66 | 5212 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
5213 | DEFINE_WAIT(wait); |
5214 | ||
11833d66 YZ |
5215 | caching_ctl = get_caching_control(cache); |
5216 | if (!caching_ctl) | |
817d52f8 | 5217 | return 0; |
817d52f8 | 5218 | |
11833d66 | 5219 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5220 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5221 | |
5222 | put_caching_control(caching_ctl); | |
5223 | return 0; | |
5224 | } | |
5225 | ||
5226 | static noinline int | |
5227 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5228 | { | |
5229 | struct btrfs_caching_control *caching_ctl; | |
5230 | DEFINE_WAIT(wait); | |
5231 | ||
5232 | caching_ctl = get_caching_control(cache); | |
5233 | if (!caching_ctl) | |
5234 | return 0; | |
5235 | ||
5236 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
5237 | ||
5238 | put_caching_control(caching_ctl); | |
817d52f8 JB |
5239 | return 0; |
5240 | } | |
5241 | ||
b742bb82 YZ |
5242 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
5243 | { | |
5244 | int index; | |
5245 | if (cache->flags & BTRFS_BLOCK_GROUP_RAID10) | |
5246 | index = 0; | |
5247 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1) | |
5248 | index = 1; | |
5249 | else if (cache->flags & BTRFS_BLOCK_GROUP_DUP) | |
5250 | index = 2; | |
5251 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0) | |
5252 | index = 3; | |
5253 | else | |
5254 | index = 4; | |
5255 | return index; | |
5256 | } | |
5257 | ||
817d52f8 | 5258 | enum btrfs_loop_type { |
285ff5af JB |
5259 | LOOP_CACHING_NOWAIT = 0, |
5260 | LOOP_CACHING_WAIT = 1, | |
5261 | LOOP_ALLOC_CHUNK = 2, | |
5262 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
5263 | }; |
5264 | ||
fec577fb CM |
5265 | /* |
5266 | * walks the btree of allocated extents and find a hole of a given size. | |
5267 | * The key ins is changed to record the hole: | |
5268 | * ins->objectid == block start | |
62e2749e | 5269 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
5270 | * ins->offset == number of blocks |
5271 | * Any available blocks before search_start are skipped. | |
5272 | */ | |
d397712b | 5273 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
5274 | struct btrfs_root *orig_root, |
5275 | u64 num_bytes, u64 empty_size, | |
98ed5174 | 5276 | u64 hint_byte, struct btrfs_key *ins, |
e0f54067 | 5277 | u64 data) |
fec577fb | 5278 | { |
80eb234a | 5279 | int ret = 0; |
d397712b | 5280 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 5281 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 5282 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 5283 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 5284 | u64 search_start = 0; |
239b14b3 | 5285 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 5286 | int allowed_chunk_alloc = 0; |
ccf0e725 | 5287 | int done_chunk_alloc = 0; |
80eb234a | 5288 | struct btrfs_space_info *space_info; |
fa9c0d79 | 5289 | int loop = 0; |
f0486c68 | 5290 | int index = 0; |
fb25e914 JB |
5291 | int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ? |
5292 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
817d52f8 | 5293 | bool found_uncached_bg = false; |
0a24325e | 5294 | bool failed_cluster_refill = false; |
1cdda9b8 | 5295 | bool failed_alloc = false; |
67377734 | 5296 | bool use_cluster = true; |
60d2adbb | 5297 | bool have_caching_bg = false; |
fec577fb | 5298 | |
db94535d | 5299 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 5300 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
5301 | ins->objectid = 0; |
5302 | ins->offset = 0; | |
b1a4d965 | 5303 | |
3f7de037 JB |
5304 | trace_find_free_extent(orig_root, num_bytes, empty_size, data); |
5305 | ||
2552d17e | 5306 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 5307 | if (!space_info) { |
e0f54067 | 5308 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
5309 | return -ENOSPC; |
5310 | } | |
2552d17e | 5311 | |
67377734 JB |
5312 | /* |
5313 | * If the space info is for both data and metadata it means we have a | |
5314 | * small filesystem and we can't use the clustering stuff. | |
5315 | */ | |
5316 | if (btrfs_mixed_space_info(space_info)) | |
5317 | use_cluster = false; | |
5318 | ||
0ef3e66b CM |
5319 | if (orig_root->ref_cows || empty_size) |
5320 | allowed_chunk_alloc = 1; | |
5321 | ||
67377734 | 5322 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5323 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5324 | if (!btrfs_test_opt(root, SSD)) |
5325 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5326 | } |
5327 | ||
67377734 JB |
5328 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5329 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5330 | last_ptr = &root->fs_info->data_alloc_cluster; |
5331 | } | |
0f9dd46c | 5332 | |
239b14b3 | 5333 | if (last_ptr) { |
fa9c0d79 CM |
5334 | spin_lock(&last_ptr->lock); |
5335 | if (last_ptr->block_group) | |
5336 | hint_byte = last_ptr->window_start; | |
5337 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5338 | } |
fa9c0d79 | 5339 | |
a061fc8d | 5340 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5341 | search_start = max(search_start, hint_byte); |
0b86a832 | 5342 | |
817d52f8 | 5343 | if (!last_ptr) |
fa9c0d79 | 5344 | empty_cluster = 0; |
fa9c0d79 | 5345 | |
2552d17e | 5346 | if (search_start == hint_byte) { |
2552d17e JB |
5347 | block_group = btrfs_lookup_block_group(root->fs_info, |
5348 | search_start); | |
274bd4fb | 5349 | used_block_group = block_group; |
817d52f8 JB |
5350 | /* |
5351 | * we don't want to use the block group if it doesn't match our | |
5352 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5353 | * |
5354 | * However if we are re-searching with an ideal block group | |
5355 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5356 | */ |
5357 | if (block_group && block_group_bits(block_group, data) && | |
285ff5af | 5358 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 5359 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5360 | if (list_empty(&block_group->list) || |
5361 | block_group->ro) { | |
5362 | /* | |
5363 | * someone is removing this block group, | |
5364 | * we can't jump into the have_block_group | |
5365 | * target because our list pointers are not | |
5366 | * valid | |
5367 | */ | |
5368 | btrfs_put_block_group(block_group); | |
5369 | up_read(&space_info->groups_sem); | |
ccf0e725 | 5370 | } else { |
b742bb82 | 5371 | index = get_block_group_index(block_group); |
44fb5511 | 5372 | goto have_block_group; |
ccf0e725 | 5373 | } |
2552d17e | 5374 | } else if (block_group) { |
fa9c0d79 | 5375 | btrfs_put_block_group(block_group); |
2552d17e | 5376 | } |
42e70e7a | 5377 | } |
2552d17e | 5378 | search: |
60d2adbb | 5379 | have_caching_bg = false; |
80eb234a | 5380 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
5381 | list_for_each_entry(block_group, &space_info->block_groups[index], |
5382 | list) { | |
6226cb0a | 5383 | u64 offset; |
817d52f8 | 5384 | int cached; |
8a1413a2 | 5385 | |
274bd4fb | 5386 | used_block_group = block_group; |
11dfe35a | 5387 | btrfs_get_block_group(block_group); |
2552d17e | 5388 | search_start = block_group->key.objectid; |
42e70e7a | 5389 | |
83a50de9 CM |
5390 | /* |
5391 | * this can happen if we end up cycling through all the | |
5392 | * raid types, but we want to make sure we only allocate | |
5393 | * for the proper type. | |
5394 | */ | |
5395 | if (!block_group_bits(block_group, data)) { | |
5396 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
5397 | BTRFS_BLOCK_GROUP_RAID1 | | |
5398 | BTRFS_BLOCK_GROUP_RAID10; | |
5399 | ||
5400 | /* | |
5401 | * if they asked for extra copies and this block group | |
5402 | * doesn't provide them, bail. This does allow us to | |
5403 | * fill raid0 from raid1. | |
5404 | */ | |
5405 | if ((data & extra) && !(block_group->flags & extra)) | |
5406 | goto loop; | |
5407 | } | |
5408 | ||
2552d17e | 5409 | have_block_group: |
291c7d2f JB |
5410 | cached = block_group_cache_done(block_group); |
5411 | if (unlikely(!cached)) { | |
291c7d2f | 5412 | found_uncached_bg = true; |
b8399dee | 5413 | ret = cache_block_group(block_group, trans, |
285ff5af JB |
5414 | orig_root, 0); |
5415 | BUG_ON(ret); | |
817d52f8 JB |
5416 | } |
5417 | ||
ea6a478e | 5418 | if (unlikely(block_group->ro)) |
2552d17e | 5419 | goto loop; |
0f9dd46c | 5420 | |
0a24325e | 5421 | /* |
062c05c4 AO |
5422 | * Ok we want to try and use the cluster allocator, so |
5423 | * lets look there | |
0a24325e | 5424 | */ |
062c05c4 | 5425 | if (last_ptr) { |
fa9c0d79 CM |
5426 | /* |
5427 | * the refill lock keeps out other | |
5428 | * people trying to start a new cluster | |
5429 | */ | |
5430 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
5431 | used_block_group = last_ptr->block_group; |
5432 | if (used_block_group != block_group && | |
5433 | (!used_block_group || | |
5434 | used_block_group->ro || | |
5435 | !block_group_bits(used_block_group, data))) { | |
5436 | used_block_group = block_group; | |
44fb5511 | 5437 | goto refill_cluster; |
274bd4fb AO |
5438 | } |
5439 | ||
5440 | if (used_block_group != block_group) | |
5441 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 5442 | |
274bd4fb AO |
5443 | offset = btrfs_alloc_from_cluster(used_block_group, |
5444 | last_ptr, num_bytes, used_block_group->key.objectid); | |
fa9c0d79 CM |
5445 | if (offset) { |
5446 | /* we have a block, we're done */ | |
5447 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5448 | trace_btrfs_reserve_extent_cluster(root, |
5449 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
5450 | goto checks; |
5451 | } | |
5452 | ||
274bd4fb AO |
5453 | WARN_ON(last_ptr->block_group != used_block_group); |
5454 | if (used_block_group != block_group) { | |
5455 | btrfs_put_block_group(used_block_group); | |
5456 | used_block_group = block_group; | |
fa9c0d79 | 5457 | } |
44fb5511 | 5458 | refill_cluster: |
274bd4fb | 5459 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
5460 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
5461 | * set up a new clusters, so lets just skip it | |
5462 | * and let the allocator find whatever block | |
5463 | * it can find. If we reach this point, we | |
5464 | * will have tried the cluster allocator | |
5465 | * plenty of times and not have found | |
5466 | * anything, so we are likely way too | |
5467 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
5468 | * anything. |
5469 | * | |
5470 | * However, if the cluster is taken from the | |
5471 | * current block group, release the cluster | |
5472 | * first, so that we stand a better chance of | |
5473 | * succeeding in the unclustered | |
5474 | * allocation. */ | |
5475 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
5476 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
5477 | spin_unlock(&last_ptr->refill_lock); |
5478 | goto unclustered_alloc; | |
5479 | } | |
5480 | ||
fa9c0d79 CM |
5481 | /* |
5482 | * this cluster didn't work out, free it and | |
5483 | * start over | |
5484 | */ | |
5485 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5486 | ||
a5f6f719 AO |
5487 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
5488 | spin_unlock(&last_ptr->refill_lock); | |
5489 | goto unclustered_alloc; | |
5490 | } | |
5491 | ||
fa9c0d79 | 5492 | /* allocate a cluster in this block group */ |
451d7585 | 5493 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 5494 | block_group, last_ptr, |
1b22bad7 | 5495 | search_start, num_bytes, |
fa9c0d79 CM |
5496 | empty_cluster + empty_size); |
5497 | if (ret == 0) { | |
5498 | /* | |
5499 | * now pull our allocation out of this | |
5500 | * cluster | |
5501 | */ | |
5502 | offset = btrfs_alloc_from_cluster(block_group, | |
5503 | last_ptr, num_bytes, | |
5504 | search_start); | |
5505 | if (offset) { | |
5506 | /* we found one, proceed */ | |
5507 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5508 | trace_btrfs_reserve_extent_cluster(root, |
5509 | block_group, search_start, | |
5510 | num_bytes); | |
fa9c0d79 CM |
5511 | goto checks; |
5512 | } | |
0a24325e JB |
5513 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5514 | && !failed_cluster_refill) { | |
817d52f8 JB |
5515 | spin_unlock(&last_ptr->refill_lock); |
5516 | ||
0a24325e | 5517 | failed_cluster_refill = true; |
817d52f8 JB |
5518 | wait_block_group_cache_progress(block_group, |
5519 | num_bytes + empty_cluster + empty_size); | |
5520 | goto have_block_group; | |
fa9c0d79 | 5521 | } |
817d52f8 | 5522 | |
fa9c0d79 CM |
5523 | /* |
5524 | * at this point we either didn't find a cluster | |
5525 | * or we weren't able to allocate a block from our | |
5526 | * cluster. Free the cluster we've been trying | |
5527 | * to use, and go to the next block group | |
5528 | */ | |
0a24325e | 5529 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5530 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5531 | goto loop; |
fa9c0d79 CM |
5532 | } |
5533 | ||
062c05c4 | 5534 | unclustered_alloc: |
a5f6f719 AO |
5535 | spin_lock(&block_group->free_space_ctl->tree_lock); |
5536 | if (cached && | |
5537 | block_group->free_space_ctl->free_space < | |
5538 | num_bytes + empty_cluster + empty_size) { | |
5539 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5540 | goto loop; | |
5541 | } | |
5542 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5543 | ||
6226cb0a JB |
5544 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5545 | num_bytes, empty_size); | |
1cdda9b8 JB |
5546 | /* |
5547 | * If we didn't find a chunk, and we haven't failed on this | |
5548 | * block group before, and this block group is in the middle of | |
5549 | * caching and we are ok with waiting, then go ahead and wait | |
5550 | * for progress to be made, and set failed_alloc to true. | |
5551 | * | |
5552 | * If failed_alloc is true then we've already waited on this | |
5553 | * block group once and should move on to the next block group. | |
5554 | */ | |
5555 | if (!offset && !failed_alloc && !cached && | |
5556 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5557 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5558 | num_bytes + empty_size); |
5559 | failed_alloc = true; | |
817d52f8 | 5560 | goto have_block_group; |
1cdda9b8 | 5561 | } else if (!offset) { |
60d2adbb MX |
5562 | if (!cached) |
5563 | have_caching_bg = true; | |
1cdda9b8 | 5564 | goto loop; |
817d52f8 | 5565 | } |
fa9c0d79 | 5566 | checks: |
6226cb0a | 5567 | search_start = stripe_align(root, offset); |
25179201 | 5568 | |
2552d17e JB |
5569 | /* move on to the next group */ |
5570 | if (search_start + num_bytes > | |
274bd4fb AO |
5571 | used_block_group->key.objectid + used_block_group->key.offset) { |
5572 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 5573 | goto loop; |
6226cb0a | 5574 | } |
f5a31e16 | 5575 | |
f0486c68 | 5576 | if (offset < search_start) |
274bd4fb | 5577 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
5578 | search_start - offset); |
5579 | BUG_ON(offset > search_start); | |
2552d17e | 5580 | |
274bd4fb | 5581 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 5582 | alloc_type); |
f0486c68 | 5583 | if (ret == -EAGAIN) { |
274bd4fb | 5584 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 5585 | goto loop; |
0f9dd46c | 5586 | } |
0b86a832 | 5587 | |
f0486c68 | 5588 | /* we are all good, lets return */ |
2552d17e JB |
5589 | ins->objectid = search_start; |
5590 | ins->offset = num_bytes; | |
d2fb3437 | 5591 | |
3f7de037 JB |
5592 | trace_btrfs_reserve_extent(orig_root, block_group, |
5593 | search_start, num_bytes); | |
6226cb0a | 5594 | if (offset < search_start) |
274bd4fb | 5595 | btrfs_add_free_space(used_block_group, offset, |
6226cb0a JB |
5596 | search_start - offset); |
5597 | BUG_ON(offset > search_start); | |
274bd4fb AO |
5598 | if (used_block_group != block_group) |
5599 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 5600 | btrfs_put_block_group(block_group); |
2552d17e JB |
5601 | break; |
5602 | loop: | |
0a24325e | 5603 | failed_cluster_refill = false; |
1cdda9b8 | 5604 | failed_alloc = false; |
b742bb82 | 5605 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
5606 | if (used_block_group != block_group) |
5607 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 5608 | btrfs_put_block_group(block_group); |
2552d17e JB |
5609 | } |
5610 | up_read(&space_info->groups_sem); | |
5611 | ||
60d2adbb MX |
5612 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
5613 | goto search; | |
5614 | ||
b742bb82 YZ |
5615 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5616 | goto search; | |
5617 | ||
285ff5af | 5618 | /* |
ccf0e725 JB |
5619 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
5620 | * caching kthreads as we move along | |
817d52f8 JB |
5621 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5622 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5623 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5624 | * again | |
fa9c0d79 | 5625 | */ |
723bda20 | 5626 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 5627 | index = 0; |
723bda20 | 5628 | loop++; |
817d52f8 | 5629 | if (loop == LOOP_ALLOC_CHUNK) { |
723bda20 JB |
5630 | if (allowed_chunk_alloc) { |
5631 | ret = do_chunk_alloc(trans, root, num_bytes + | |
5632 | 2 * 1024 * 1024, data, | |
5633 | CHUNK_ALLOC_LIMITED); | |
5634 | allowed_chunk_alloc = 0; | |
5635 | if (ret == 1) | |
5636 | done_chunk_alloc = 1; | |
5637 | } else if (!done_chunk_alloc && | |
5638 | space_info->force_alloc == | |
5639 | CHUNK_ALLOC_NO_FORCE) { | |
5640 | space_info->force_alloc = CHUNK_ALLOC_LIMITED; | |
5641 | } | |
2552d17e | 5642 | |
723bda20 JB |
5643 | /* |
5644 | * We didn't allocate a chunk, go ahead and drop the | |
5645 | * empty size and loop again. | |
5646 | */ | |
5647 | if (!done_chunk_alloc) | |
5648 | loop = LOOP_NO_EMPTY_SIZE; | |
2552d17e JB |
5649 | } |
5650 | ||
723bda20 JB |
5651 | if (loop == LOOP_NO_EMPTY_SIZE) { |
5652 | empty_size = 0; | |
5653 | empty_cluster = 0; | |
fa9c0d79 | 5654 | } |
723bda20 JB |
5655 | |
5656 | goto search; | |
2552d17e JB |
5657 | } else if (!ins->objectid) { |
5658 | ret = -ENOSPC; | |
d82a6f1d | 5659 | } else if (ins->objectid) { |
80eb234a | 5660 | ret = 0; |
be744175 | 5661 | } |
be744175 | 5662 | |
0f70abe2 | 5663 | return ret; |
fec577fb | 5664 | } |
ec44a35c | 5665 | |
9ed74f2d JB |
5666 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5667 | int dump_block_groups) | |
0f9dd46c JB |
5668 | { |
5669 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5670 | int index = 0; |
0f9dd46c | 5671 | |
9ed74f2d | 5672 | spin_lock(&info->lock); |
fb25e914 JB |
5673 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
5674 | (unsigned long long)info->flags, | |
d397712b | 5675 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 5676 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5677 | info->bytes_readonly), |
d397712b | 5678 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5679 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5680 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5681 | (unsigned long long)info->total_bytes, |
8929ecfa | 5682 | (unsigned long long)info->bytes_used, |
21380931 | 5683 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5684 | (unsigned long long)info->bytes_reserved, |
21380931 | 5685 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5686 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5687 | spin_unlock(&info->lock); |
5688 | ||
5689 | if (!dump_block_groups) | |
5690 | return; | |
0f9dd46c | 5691 | |
80eb234a | 5692 | down_read(&info->groups_sem); |
b742bb82 YZ |
5693 | again: |
5694 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5695 | spin_lock(&cache->lock); |
d397712b CM |
5696 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
5697 | "%llu pinned %llu reserved\n", | |
5698 | (unsigned long long)cache->key.objectid, | |
5699 | (unsigned long long)cache->key.offset, | |
5700 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5701 | (unsigned long long)cache->pinned, | |
5702 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
5703 | btrfs_dump_free_space(cache, bytes); |
5704 | spin_unlock(&cache->lock); | |
5705 | } | |
b742bb82 YZ |
5706 | if (++index < BTRFS_NR_RAID_TYPES) |
5707 | goto again; | |
80eb234a | 5708 | up_read(&info->groups_sem); |
0f9dd46c | 5709 | } |
e8569813 | 5710 | |
11833d66 YZ |
5711 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5712 | struct btrfs_root *root, | |
5713 | u64 num_bytes, u64 min_alloc_size, | |
5714 | u64 empty_size, u64 hint_byte, | |
81c9ad23 | 5715 | struct btrfs_key *ins, u64 data) |
fec577fb | 5716 | { |
9e622d6b | 5717 | bool final_tried = false; |
fec577fb | 5718 | int ret; |
925baedd | 5719 | |
6a63209f | 5720 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 5721 | again: |
0ef3e66b CM |
5722 | /* |
5723 | * the only place that sets empty_size is btrfs_realloc_node, which | |
5724 | * is not called recursively on allocations | |
5725 | */ | |
83d3c969 | 5726 | if (empty_size || root->ref_cows) |
6324fbf3 | 5727 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
5728 | num_bytes + 2 * 1024 * 1024, data, |
5729 | CHUNK_ALLOC_NO_FORCE); | |
0b86a832 | 5730 | |
db94535d CM |
5731 | WARN_ON(num_bytes < root->sectorsize); |
5732 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
81c9ad23 | 5733 | hint_byte, ins, data); |
3b951516 | 5734 | |
9e622d6b MX |
5735 | if (ret == -ENOSPC) { |
5736 | if (!final_tried) { | |
5737 | num_bytes = num_bytes >> 1; | |
5738 | num_bytes = num_bytes & ~(root->sectorsize - 1); | |
5739 | num_bytes = max(num_bytes, min_alloc_size); | |
5740 | do_chunk_alloc(trans, root->fs_info->extent_root, | |
5741 | num_bytes, data, CHUNK_ALLOC_FORCE); | |
5742 | if (num_bytes == min_alloc_size) | |
5743 | final_tried = true; | |
5744 | goto again; | |
5745 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
5746 | struct btrfs_space_info *sinfo; | |
5747 | ||
5748 | sinfo = __find_space_info(root->fs_info, data); | |
5749 | printk(KERN_ERR "btrfs allocation failed flags %llu, " | |
5750 | "wanted %llu\n", (unsigned long long)data, | |
5751 | (unsigned long long)num_bytes); | |
5752 | dump_space_info(sinfo, num_bytes, 1); | |
5753 | } | |
925baedd | 5754 | } |
0f9dd46c | 5755 | |
1abe9b8a | 5756 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
5757 | ||
0f9dd46c | 5758 | return ret; |
e6dcd2dc CM |
5759 | } |
5760 | ||
e688b725 CM |
5761 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
5762 | u64 start, u64 len, int pin) | |
65b51a00 | 5763 | { |
0f9dd46c | 5764 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 5765 | int ret = 0; |
0f9dd46c | 5766 | |
0f9dd46c JB |
5767 | cache = btrfs_lookup_block_group(root->fs_info, start); |
5768 | if (!cache) { | |
d397712b CM |
5769 | printk(KERN_ERR "Unable to find block group for %llu\n", |
5770 | (unsigned long long)start); | |
0f9dd46c JB |
5771 | return -ENOSPC; |
5772 | } | |
1f3c79a2 | 5773 | |
5378e607 LD |
5774 | if (btrfs_test_opt(root, DISCARD)) |
5775 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 5776 | |
e688b725 CM |
5777 | if (pin) |
5778 | pin_down_extent(root, cache, start, len, 1); | |
5779 | else { | |
5780 | btrfs_add_free_space(cache, start, len); | |
5781 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
5782 | } | |
fa9c0d79 | 5783 | btrfs_put_block_group(cache); |
817d52f8 | 5784 | |
1abe9b8a | 5785 | trace_btrfs_reserved_extent_free(root, start, len); |
5786 | ||
e6dcd2dc CM |
5787 | return ret; |
5788 | } | |
5789 | ||
e688b725 CM |
5790 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
5791 | u64 start, u64 len) | |
5792 | { | |
5793 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
5794 | } | |
5795 | ||
5796 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
5797 | u64 start, u64 len) | |
5798 | { | |
5799 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
5800 | } | |
5801 | ||
5d4f98a2 YZ |
5802 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5803 | struct btrfs_root *root, | |
5804 | u64 parent, u64 root_objectid, | |
5805 | u64 flags, u64 owner, u64 offset, | |
5806 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
5807 | { |
5808 | int ret; | |
5d4f98a2 | 5809 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 5810 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 5811 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 5812 | struct btrfs_path *path; |
5d4f98a2 YZ |
5813 | struct extent_buffer *leaf; |
5814 | int type; | |
5815 | u32 size; | |
26b8003f | 5816 | |
5d4f98a2 YZ |
5817 | if (parent > 0) |
5818 | type = BTRFS_SHARED_DATA_REF_KEY; | |
5819 | else | |
5820 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 5821 | |
5d4f98a2 | 5822 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
5823 | |
5824 | path = btrfs_alloc_path(); | |
db5b493a TI |
5825 | if (!path) |
5826 | return -ENOMEM; | |
47e4bb98 | 5827 | |
b9473439 | 5828 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5829 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
5830 | ins, size); | |
ccd467d6 | 5831 | BUG_ON(ret); |
0f9dd46c | 5832 | |
5d4f98a2 YZ |
5833 | leaf = path->nodes[0]; |
5834 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 5835 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5836 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
5837 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5838 | btrfs_set_extent_flags(leaf, extent_item, | |
5839 | flags | BTRFS_EXTENT_FLAG_DATA); | |
5840 | ||
5841 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
5842 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
5843 | if (parent > 0) { | |
5844 | struct btrfs_shared_data_ref *ref; | |
5845 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
5846 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5847 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
5848 | } else { | |
5849 | struct btrfs_extent_data_ref *ref; | |
5850 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
5851 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
5852 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
5853 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
5854 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
5855 | } | |
47e4bb98 CM |
5856 | |
5857 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 5858 | btrfs_free_path(path); |
f510cfec | 5859 | |
f0486c68 | 5860 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
f5947066 | 5861 | if (ret) { |
d397712b CM |
5862 | printk(KERN_ERR "btrfs update block group failed for %llu " |
5863 | "%llu\n", (unsigned long long)ins->objectid, | |
5864 | (unsigned long long)ins->offset); | |
f5947066 CM |
5865 | BUG(); |
5866 | } | |
e6dcd2dc CM |
5867 | return ret; |
5868 | } | |
5869 | ||
5d4f98a2 YZ |
5870 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
5871 | struct btrfs_root *root, | |
5872 | u64 parent, u64 root_objectid, | |
5873 | u64 flags, struct btrfs_disk_key *key, | |
5874 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
5875 | { |
5876 | int ret; | |
5d4f98a2 YZ |
5877 | struct btrfs_fs_info *fs_info = root->fs_info; |
5878 | struct btrfs_extent_item *extent_item; | |
5879 | struct btrfs_tree_block_info *block_info; | |
5880 | struct btrfs_extent_inline_ref *iref; | |
5881 | struct btrfs_path *path; | |
5882 | struct extent_buffer *leaf; | |
5883 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 5884 | |
5d4f98a2 | 5885 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5886 | if (!path) |
5887 | return -ENOMEM; | |
56bec294 | 5888 | |
5d4f98a2 YZ |
5889 | path->leave_spinning = 1; |
5890 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
5891 | ins, size); | |
56bec294 | 5892 | BUG_ON(ret); |
5d4f98a2 YZ |
5893 | |
5894 | leaf = path->nodes[0]; | |
5895 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
5896 | struct btrfs_extent_item); | |
5897 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
5898 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5899 | btrfs_set_extent_flags(leaf, extent_item, | |
5900 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5901 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
5902 | ||
5903 | btrfs_set_tree_block_key(leaf, block_info, key); | |
5904 | btrfs_set_tree_block_level(leaf, block_info, level); | |
5905 | ||
5906 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
5907 | if (parent > 0) { | |
5908 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
5909 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5910 | BTRFS_SHARED_BLOCK_REF_KEY); | |
5911 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5912 | } else { | |
5913 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5914 | BTRFS_TREE_BLOCK_REF_KEY); | |
5915 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
5916 | } | |
5917 | ||
5918 | btrfs_mark_buffer_dirty(leaf); | |
5919 | btrfs_free_path(path); | |
5920 | ||
f0486c68 | 5921 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
5d4f98a2 YZ |
5922 | if (ret) { |
5923 | printk(KERN_ERR "btrfs update block group failed for %llu " | |
5924 | "%llu\n", (unsigned long long)ins->objectid, | |
5925 | (unsigned long long)ins->offset); | |
5926 | BUG(); | |
5927 | } | |
5928 | return ret; | |
5929 | } | |
5930 | ||
5931 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5932 | struct btrfs_root *root, | |
5933 | u64 root_objectid, u64 owner, | |
5934 | u64 offset, struct btrfs_key *ins) | |
5935 | { | |
5936 | int ret; | |
5937 | ||
5938 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
5939 | ||
66d7e7f0 AJ |
5940 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
5941 | ins->offset, 0, | |
5942 | root_objectid, owner, offset, | |
5943 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
5944 | return ret; |
5945 | } | |
e02119d5 CM |
5946 | |
5947 | /* | |
5948 | * this is used by the tree logging recovery code. It records that | |
5949 | * an extent has been allocated and makes sure to clear the free | |
5950 | * space cache bits as well | |
5951 | */ | |
5d4f98a2 YZ |
5952 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5953 | struct btrfs_root *root, | |
5954 | u64 root_objectid, u64 owner, u64 offset, | |
5955 | struct btrfs_key *ins) | |
e02119d5 CM |
5956 | { |
5957 | int ret; | |
5958 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
5959 | struct btrfs_caching_control *caching_ctl; |
5960 | u64 start = ins->objectid; | |
5961 | u64 num_bytes = ins->offset; | |
e02119d5 | 5962 | |
e02119d5 | 5963 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 5964 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 5965 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 5966 | |
11833d66 YZ |
5967 | if (!caching_ctl) { |
5968 | BUG_ON(!block_group_cache_done(block_group)); | |
5969 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
5970 | BUG_ON(ret); | |
5971 | } else { | |
5972 | mutex_lock(&caching_ctl->mutex); | |
5973 | ||
5974 | if (start >= caching_ctl->progress) { | |
5975 | ret = add_excluded_extent(root, start, num_bytes); | |
5976 | BUG_ON(ret); | |
5977 | } else if (start + num_bytes <= caching_ctl->progress) { | |
5978 | ret = btrfs_remove_free_space(block_group, | |
5979 | start, num_bytes); | |
5980 | BUG_ON(ret); | |
5981 | } else { | |
5982 | num_bytes = caching_ctl->progress - start; | |
5983 | ret = btrfs_remove_free_space(block_group, | |
5984 | start, num_bytes); | |
5985 | BUG_ON(ret); | |
5986 | ||
5987 | start = caching_ctl->progress; | |
5988 | num_bytes = ins->objectid + ins->offset - | |
5989 | caching_ctl->progress; | |
5990 | ret = add_excluded_extent(root, start, num_bytes); | |
5991 | BUG_ON(ret); | |
5992 | } | |
5993 | ||
5994 | mutex_unlock(&caching_ctl->mutex); | |
5995 | put_caching_control(caching_ctl); | |
5996 | } | |
5997 | ||
fb25e914 JB |
5998 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
5999 | RESERVE_ALLOC_NO_ACCOUNT); | |
f0486c68 | 6000 | BUG_ON(ret); |
fa9c0d79 | 6001 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
6002 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
6003 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
6004 | return ret; |
6005 | } | |
6006 | ||
65b51a00 CM |
6007 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
6008 | struct btrfs_root *root, | |
4008c04a CM |
6009 | u64 bytenr, u32 blocksize, |
6010 | int level) | |
65b51a00 CM |
6011 | { |
6012 | struct extent_buffer *buf; | |
6013 | ||
6014 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6015 | if (!buf) | |
6016 | return ERR_PTR(-ENOMEM); | |
6017 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6018 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6019 | btrfs_tree_lock(buf); |
6020 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6021 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6022 | |
6023 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6024 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6025 | |
d0c803c4 | 6026 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6027 | /* |
6028 | * we allow two log transactions at a time, use different | |
6029 | * EXENT bit to differentiate dirty pages. | |
6030 | */ | |
6031 | if (root->log_transid % 2 == 0) | |
6032 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6033 | buf->start + buf->len - 1, GFP_NOFS); | |
6034 | else | |
6035 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6036 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6037 | } else { |
6038 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6039 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6040 | } |
65b51a00 | 6041 | trans->blocks_used++; |
b4ce94de | 6042 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6043 | return buf; |
6044 | } | |
6045 | ||
f0486c68 YZ |
6046 | static struct btrfs_block_rsv * |
6047 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6048 | struct btrfs_root *root, u32 blocksize) | |
6049 | { | |
6050 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6051 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
6052 | int ret; |
6053 | ||
6054 | block_rsv = get_block_rsv(trans, root); | |
6055 | ||
6056 | if (block_rsv->size == 0) { | |
36ba022a | 6057 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 JB |
6058 | /* |
6059 | * If we couldn't reserve metadata bytes try and use some from | |
6060 | * the global reserve. | |
6061 | */ | |
6062 | if (ret && block_rsv != global_rsv) { | |
6063 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6064 | if (!ret) | |
6065 | return global_rsv; | |
f0486c68 | 6066 | return ERR_PTR(ret); |
68a82277 | 6067 | } else if (ret) { |
f0486c68 | 6068 | return ERR_PTR(ret); |
68a82277 | 6069 | } |
f0486c68 YZ |
6070 | return block_rsv; |
6071 | } | |
6072 | ||
6073 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
6074 | if (!ret) | |
6075 | return block_rsv; | |
68a82277 | 6076 | if (ret) { |
dff51cd1 DS |
6077 | static DEFINE_RATELIMIT_STATE(_rs, |
6078 | DEFAULT_RATELIMIT_INTERVAL, | |
6079 | /*DEFAULT_RATELIMIT_BURST*/ 2); | |
6080 | if (__ratelimit(&_rs)) { | |
6081 | printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret); | |
6082 | WARN_ON(1); | |
6083 | } | |
36ba022a | 6084 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 | 6085 | if (!ret) { |
68a82277 JB |
6086 | return block_rsv; |
6087 | } else if (ret && block_rsv != global_rsv) { | |
6088 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6089 | if (!ret) | |
6090 | return global_rsv; | |
6091 | } | |
6092 | } | |
f0486c68 | 6093 | |
f0486c68 YZ |
6094 | return ERR_PTR(-ENOSPC); |
6095 | } | |
6096 | ||
8c2a3ca2 JB |
6097 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6098 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6099 | { |
6100 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6101 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6102 | } |
6103 | ||
fec577fb | 6104 | /* |
f0486c68 YZ |
6105 | * finds a free extent and does all the dirty work required for allocation |
6106 | * returns the key for the extent through ins, and a tree buffer for | |
6107 | * the first block of the extent through buf. | |
6108 | * | |
fec577fb CM |
6109 | * returns the tree buffer or NULL. |
6110 | */ | |
5f39d397 | 6111 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6112 | struct btrfs_root *root, u32 blocksize, |
6113 | u64 parent, u64 root_objectid, | |
6114 | struct btrfs_disk_key *key, int level, | |
66d7e7f0 | 6115 | u64 hint, u64 empty_size, int for_cow) |
fec577fb | 6116 | { |
e2fa7227 | 6117 | struct btrfs_key ins; |
f0486c68 | 6118 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6119 | struct extent_buffer *buf; |
f0486c68 YZ |
6120 | u64 flags = 0; |
6121 | int ret; | |
6122 | ||
fec577fb | 6123 | |
f0486c68 YZ |
6124 | block_rsv = use_block_rsv(trans, root, blocksize); |
6125 | if (IS_ERR(block_rsv)) | |
6126 | return ERR_CAST(block_rsv); | |
6127 | ||
6128 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
81c9ad23 | 6129 | empty_size, hint, &ins, 0); |
fec577fb | 6130 | if (ret) { |
8c2a3ca2 | 6131 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6132 | return ERR_PTR(ret); |
fec577fb | 6133 | } |
55c69072 | 6134 | |
4008c04a CM |
6135 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6136 | blocksize, level); | |
f0486c68 YZ |
6137 | BUG_ON(IS_ERR(buf)); |
6138 | ||
6139 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6140 | if (parent == 0) | |
6141 | parent = ins.objectid; | |
6142 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6143 | } else | |
6144 | BUG_ON(parent > 0); | |
6145 | ||
6146 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6147 | struct btrfs_delayed_extent_op *extent_op; | |
6148 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
6149 | BUG_ON(!extent_op); | |
6150 | if (key) | |
6151 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6152 | else | |
6153 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6154 | extent_op->flags_to_set = flags; | |
6155 | extent_op->update_key = 1; | |
6156 | extent_op->update_flags = 1; | |
6157 | extent_op->is_data = 0; | |
6158 | ||
66d7e7f0 AJ |
6159 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6160 | ins.objectid, | |
f0486c68 YZ |
6161 | ins.offset, parent, root_objectid, |
6162 | level, BTRFS_ADD_DELAYED_EXTENT, | |
66d7e7f0 | 6163 | extent_op, for_cow); |
f0486c68 YZ |
6164 | BUG_ON(ret); |
6165 | } | |
fec577fb CM |
6166 | return buf; |
6167 | } | |
a28ec197 | 6168 | |
2c47e605 YZ |
6169 | struct walk_control { |
6170 | u64 refs[BTRFS_MAX_LEVEL]; | |
6171 | u64 flags[BTRFS_MAX_LEVEL]; | |
6172 | struct btrfs_key update_progress; | |
6173 | int stage; | |
6174 | int level; | |
6175 | int shared_level; | |
6176 | int update_ref; | |
6177 | int keep_locks; | |
1c4850e2 YZ |
6178 | int reada_slot; |
6179 | int reada_count; | |
66d7e7f0 | 6180 | int for_reloc; |
2c47e605 YZ |
6181 | }; |
6182 | ||
6183 | #define DROP_REFERENCE 1 | |
6184 | #define UPDATE_BACKREF 2 | |
6185 | ||
1c4850e2 YZ |
6186 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6187 | struct btrfs_root *root, | |
6188 | struct walk_control *wc, | |
6189 | struct btrfs_path *path) | |
6407bf6d | 6190 | { |
1c4850e2 YZ |
6191 | u64 bytenr; |
6192 | u64 generation; | |
6193 | u64 refs; | |
94fcca9f | 6194 | u64 flags; |
5d4f98a2 | 6195 | u32 nritems; |
1c4850e2 YZ |
6196 | u32 blocksize; |
6197 | struct btrfs_key key; | |
6198 | struct extent_buffer *eb; | |
6407bf6d | 6199 | int ret; |
1c4850e2 YZ |
6200 | int slot; |
6201 | int nread = 0; | |
6407bf6d | 6202 | |
1c4850e2 YZ |
6203 | if (path->slots[wc->level] < wc->reada_slot) { |
6204 | wc->reada_count = wc->reada_count * 2 / 3; | |
6205 | wc->reada_count = max(wc->reada_count, 2); | |
6206 | } else { | |
6207 | wc->reada_count = wc->reada_count * 3 / 2; | |
6208 | wc->reada_count = min_t(int, wc->reada_count, | |
6209 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
6210 | } | |
7bb86316 | 6211 | |
1c4850e2 YZ |
6212 | eb = path->nodes[wc->level]; |
6213 | nritems = btrfs_header_nritems(eb); | |
6214 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 6215 | |
1c4850e2 YZ |
6216 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6217 | if (nread >= wc->reada_count) | |
6218 | break; | |
bd56b302 | 6219 | |
2dd3e67b | 6220 | cond_resched(); |
1c4850e2 YZ |
6221 | bytenr = btrfs_node_blockptr(eb, slot); |
6222 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6223 | |
1c4850e2 YZ |
6224 | if (slot == path->slots[wc->level]) |
6225 | goto reada; | |
5d4f98a2 | 6226 | |
1c4850e2 YZ |
6227 | if (wc->stage == UPDATE_BACKREF && |
6228 | generation <= root->root_key.offset) | |
bd56b302 CM |
6229 | continue; |
6230 | ||
94fcca9f YZ |
6231 | /* We don't lock the tree block, it's OK to be racy here */ |
6232 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
6233 | &refs, &flags); | |
6234 | BUG_ON(ret); | |
6235 | BUG_ON(refs == 0); | |
6236 | ||
1c4850e2 | 6237 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6238 | if (refs == 1) |
6239 | goto reada; | |
bd56b302 | 6240 | |
94fcca9f YZ |
6241 | if (wc->level == 1 && |
6242 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6243 | continue; | |
1c4850e2 YZ |
6244 | if (!wc->update_ref || |
6245 | generation <= root->root_key.offset) | |
6246 | continue; | |
6247 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6248 | ret = btrfs_comp_cpu_keys(&key, | |
6249 | &wc->update_progress); | |
6250 | if (ret < 0) | |
6251 | continue; | |
94fcca9f YZ |
6252 | } else { |
6253 | if (wc->level == 1 && | |
6254 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6255 | continue; | |
6407bf6d | 6256 | } |
1c4850e2 YZ |
6257 | reada: |
6258 | ret = readahead_tree_block(root, bytenr, blocksize, | |
6259 | generation); | |
6260 | if (ret) | |
bd56b302 | 6261 | break; |
1c4850e2 | 6262 | nread++; |
20524f02 | 6263 | } |
1c4850e2 | 6264 | wc->reada_slot = slot; |
20524f02 | 6265 | } |
2c47e605 | 6266 | |
f82d02d9 | 6267 | /* |
2c47e605 YZ |
6268 | * hepler to process tree block while walking down the tree. |
6269 | * | |
2c47e605 YZ |
6270 | * when wc->stage == UPDATE_BACKREF, this function updates |
6271 | * back refs for pointers in the block. | |
6272 | * | |
6273 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 6274 | */ |
2c47e605 | 6275 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 6276 | struct btrfs_root *root, |
2c47e605 | 6277 | struct btrfs_path *path, |
94fcca9f | 6278 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 6279 | { |
2c47e605 YZ |
6280 | int level = wc->level; |
6281 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6282 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6283 | int ret; |
6284 | ||
2c47e605 YZ |
6285 | if (wc->stage == UPDATE_BACKREF && |
6286 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6287 | return 1; | |
f82d02d9 | 6288 | |
2c47e605 YZ |
6289 | /* |
6290 | * when reference count of tree block is 1, it won't increase | |
6291 | * again. once full backref flag is set, we never clear it. | |
6292 | */ | |
94fcca9f YZ |
6293 | if (lookup_info && |
6294 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6295 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6296 | BUG_ON(!path->locks[level]); |
6297 | ret = btrfs_lookup_extent_info(trans, root, | |
6298 | eb->start, eb->len, | |
6299 | &wc->refs[level], | |
6300 | &wc->flags[level]); | |
6301 | BUG_ON(ret); | |
6302 | BUG_ON(wc->refs[level] == 0); | |
6303 | } | |
5d4f98a2 | 6304 | |
2c47e605 YZ |
6305 | if (wc->stage == DROP_REFERENCE) { |
6306 | if (wc->refs[level] > 1) | |
6307 | return 1; | |
f82d02d9 | 6308 | |
2c47e605 | 6309 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 6310 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6311 | path->locks[level] = 0; |
6312 | } | |
6313 | return 0; | |
6314 | } | |
f82d02d9 | 6315 | |
2c47e605 YZ |
6316 | /* wc->stage == UPDATE_BACKREF */ |
6317 | if (!(wc->flags[level] & flag)) { | |
6318 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 6319 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
f82d02d9 | 6320 | BUG_ON(ret); |
66d7e7f0 | 6321 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
2c47e605 YZ |
6322 | BUG_ON(ret); |
6323 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, | |
6324 | eb->len, flag, 0); | |
6325 | BUG_ON(ret); | |
6326 | wc->flags[level] |= flag; | |
6327 | } | |
6328 | ||
6329 | /* | |
6330 | * the block is shared by multiple trees, so it's not good to | |
6331 | * keep the tree lock | |
6332 | */ | |
6333 | if (path->locks[level] && level > 0) { | |
bd681513 | 6334 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6335 | path->locks[level] = 0; |
6336 | } | |
6337 | return 0; | |
6338 | } | |
6339 | ||
1c4850e2 YZ |
6340 | /* |
6341 | * hepler to process tree block pointer. | |
6342 | * | |
6343 | * when wc->stage == DROP_REFERENCE, this function checks | |
6344 | * reference count of the block pointed to. if the block | |
6345 | * is shared and we need update back refs for the subtree | |
6346 | * rooted at the block, this function changes wc->stage to | |
6347 | * UPDATE_BACKREF. if the block is shared and there is no | |
6348 | * need to update back, this function drops the reference | |
6349 | * to the block. | |
6350 | * | |
6351 | * NOTE: return value 1 means we should stop walking down. | |
6352 | */ | |
6353 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
6354 | struct btrfs_root *root, | |
6355 | struct btrfs_path *path, | |
94fcca9f | 6356 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
6357 | { |
6358 | u64 bytenr; | |
6359 | u64 generation; | |
6360 | u64 parent; | |
6361 | u32 blocksize; | |
6362 | struct btrfs_key key; | |
6363 | struct extent_buffer *next; | |
6364 | int level = wc->level; | |
6365 | int reada = 0; | |
6366 | int ret = 0; | |
6367 | ||
6368 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
6369 | path->slots[level]); | |
6370 | /* | |
6371 | * if the lower level block was created before the snapshot | |
6372 | * was created, we know there is no need to update back refs | |
6373 | * for the subtree | |
6374 | */ | |
6375 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
6376 | generation <= root->root_key.offset) { |
6377 | *lookup_info = 1; | |
1c4850e2 | 6378 | return 1; |
94fcca9f | 6379 | } |
1c4850e2 YZ |
6380 | |
6381 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
6382 | blocksize = btrfs_level_size(root, level - 1); | |
6383 | ||
6384 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6385 | if (!next) { | |
6386 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6387 | if (!next) |
6388 | return -ENOMEM; | |
1c4850e2 YZ |
6389 | reada = 1; |
6390 | } | |
6391 | btrfs_tree_lock(next); | |
6392 | btrfs_set_lock_blocking(next); | |
6393 | ||
94fcca9f YZ |
6394 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6395 | &wc->refs[level - 1], | |
6396 | &wc->flags[level - 1]); | |
6397 | BUG_ON(ret); | |
6398 | BUG_ON(wc->refs[level - 1] == 0); | |
6399 | *lookup_info = 0; | |
1c4850e2 | 6400 | |
94fcca9f | 6401 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6402 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6403 | if (level == 1 && |
6404 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6405 | goto skip; | |
6406 | ||
1c4850e2 YZ |
6407 | if (!wc->update_ref || |
6408 | generation <= root->root_key.offset) | |
6409 | goto skip; | |
6410 | ||
6411 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6412 | path->slots[level]); | |
6413 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6414 | if (ret < 0) | |
6415 | goto skip; | |
6416 | ||
6417 | wc->stage = UPDATE_BACKREF; | |
6418 | wc->shared_level = level - 1; | |
6419 | } | |
94fcca9f YZ |
6420 | } else { |
6421 | if (level == 1 && | |
6422 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6423 | goto skip; | |
1c4850e2 YZ |
6424 | } |
6425 | ||
6426 | if (!btrfs_buffer_uptodate(next, generation)) { | |
6427 | btrfs_tree_unlock(next); | |
6428 | free_extent_buffer(next); | |
6429 | next = NULL; | |
94fcca9f | 6430 | *lookup_info = 1; |
1c4850e2 YZ |
6431 | } |
6432 | ||
6433 | if (!next) { | |
6434 | if (reada && level == 1) | |
6435 | reada_walk_down(trans, root, wc, path); | |
6436 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6437 | if (!next) |
6438 | return -EIO; | |
1c4850e2 YZ |
6439 | btrfs_tree_lock(next); |
6440 | btrfs_set_lock_blocking(next); | |
6441 | } | |
6442 | ||
6443 | level--; | |
6444 | BUG_ON(level != btrfs_header_level(next)); | |
6445 | path->nodes[level] = next; | |
6446 | path->slots[level] = 0; | |
bd681513 | 6447 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
6448 | wc->level = level; |
6449 | if (wc->level == 1) | |
6450 | wc->reada_slot = 0; | |
6451 | return 0; | |
6452 | skip: | |
6453 | wc->refs[level - 1] = 0; | |
6454 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6455 | if (wc->stage == DROP_REFERENCE) { |
6456 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6457 | parent = path->nodes[level]->start; | |
6458 | } else { | |
6459 | BUG_ON(root->root_key.objectid != | |
6460 | btrfs_header_owner(path->nodes[level])); | |
6461 | parent = 0; | |
6462 | } | |
1c4850e2 | 6463 | |
94fcca9f | 6464 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 6465 | root->root_key.objectid, level - 1, 0, 0); |
94fcca9f | 6466 | BUG_ON(ret); |
1c4850e2 | 6467 | } |
1c4850e2 YZ |
6468 | btrfs_tree_unlock(next); |
6469 | free_extent_buffer(next); | |
94fcca9f | 6470 | *lookup_info = 1; |
1c4850e2 YZ |
6471 | return 1; |
6472 | } | |
6473 | ||
2c47e605 YZ |
6474 | /* |
6475 | * hepler to process tree block while walking up the tree. | |
6476 | * | |
6477 | * when wc->stage == DROP_REFERENCE, this function drops | |
6478 | * reference count on the block. | |
6479 | * | |
6480 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6481 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6482 | * to UPDATE_BACKREF previously while processing the block. | |
6483 | * | |
6484 | * NOTE: return value 1 means we should stop walking up. | |
6485 | */ | |
6486 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6487 | struct btrfs_root *root, | |
6488 | struct btrfs_path *path, | |
6489 | struct walk_control *wc) | |
6490 | { | |
f0486c68 | 6491 | int ret; |
2c47e605 YZ |
6492 | int level = wc->level; |
6493 | struct extent_buffer *eb = path->nodes[level]; | |
6494 | u64 parent = 0; | |
6495 | ||
6496 | if (wc->stage == UPDATE_BACKREF) { | |
6497 | BUG_ON(wc->shared_level < level); | |
6498 | if (level < wc->shared_level) | |
6499 | goto out; | |
6500 | ||
2c47e605 YZ |
6501 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6502 | if (ret > 0) | |
6503 | wc->update_ref = 0; | |
6504 | ||
6505 | wc->stage = DROP_REFERENCE; | |
6506 | wc->shared_level = -1; | |
6507 | path->slots[level] = 0; | |
6508 | ||
6509 | /* | |
6510 | * check reference count again if the block isn't locked. | |
6511 | * we should start walking down the tree again if reference | |
6512 | * count is one. | |
6513 | */ | |
6514 | if (!path->locks[level]) { | |
6515 | BUG_ON(level == 0); | |
6516 | btrfs_tree_lock(eb); | |
6517 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6518 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6519 | |
6520 | ret = btrfs_lookup_extent_info(trans, root, | |
6521 | eb->start, eb->len, | |
6522 | &wc->refs[level], | |
6523 | &wc->flags[level]); | |
f82d02d9 | 6524 | BUG_ON(ret); |
2c47e605 YZ |
6525 | BUG_ON(wc->refs[level] == 0); |
6526 | if (wc->refs[level] == 1) { | |
bd681513 | 6527 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6528 | return 1; |
6529 | } | |
f82d02d9 | 6530 | } |
2c47e605 | 6531 | } |
f82d02d9 | 6532 | |
2c47e605 YZ |
6533 | /* wc->stage == DROP_REFERENCE */ |
6534 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6535 | |
2c47e605 YZ |
6536 | if (wc->refs[level] == 1) { |
6537 | if (level == 0) { | |
6538 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
6539 | ret = btrfs_dec_ref(trans, root, eb, 1, |
6540 | wc->for_reloc); | |
2c47e605 | 6541 | else |
66d7e7f0 AJ |
6542 | ret = btrfs_dec_ref(trans, root, eb, 0, |
6543 | wc->for_reloc); | |
2c47e605 YZ |
6544 | BUG_ON(ret); |
6545 | } | |
6546 | /* make block locked assertion in clean_tree_block happy */ | |
6547 | if (!path->locks[level] && | |
6548 | btrfs_header_generation(eb) == trans->transid) { | |
6549 | btrfs_tree_lock(eb); | |
6550 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6551 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6552 | } |
6553 | clean_tree_block(trans, root, eb); | |
6554 | } | |
6555 | ||
6556 | if (eb == root->node) { | |
6557 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6558 | parent = eb->start; | |
6559 | else | |
6560 | BUG_ON(root->root_key.objectid != | |
6561 | btrfs_header_owner(eb)); | |
6562 | } else { | |
6563 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6564 | parent = path->nodes[level + 1]->start; | |
6565 | else | |
6566 | BUG_ON(root->root_key.objectid != | |
6567 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6568 | } |
f82d02d9 | 6569 | |
66d7e7f0 | 6570 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0); |
2c47e605 YZ |
6571 | out: |
6572 | wc->refs[level] = 0; | |
6573 | wc->flags[level] = 0; | |
f0486c68 | 6574 | return 0; |
2c47e605 YZ |
6575 | } |
6576 | ||
6577 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6578 | struct btrfs_root *root, | |
6579 | struct btrfs_path *path, | |
6580 | struct walk_control *wc) | |
6581 | { | |
2c47e605 | 6582 | int level = wc->level; |
94fcca9f | 6583 | int lookup_info = 1; |
2c47e605 YZ |
6584 | int ret; |
6585 | ||
6586 | while (level >= 0) { | |
94fcca9f | 6587 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6588 | if (ret > 0) |
6589 | break; | |
6590 | ||
6591 | if (level == 0) | |
6592 | break; | |
6593 | ||
7a7965f8 YZ |
6594 | if (path->slots[level] >= |
6595 | btrfs_header_nritems(path->nodes[level])) | |
6596 | break; | |
6597 | ||
94fcca9f | 6598 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6599 | if (ret > 0) { |
6600 | path->slots[level]++; | |
6601 | continue; | |
90d2c51d MX |
6602 | } else if (ret < 0) |
6603 | return ret; | |
1c4850e2 | 6604 | level = wc->level; |
f82d02d9 | 6605 | } |
f82d02d9 YZ |
6606 | return 0; |
6607 | } | |
6608 | ||
d397712b | 6609 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6610 | struct btrfs_root *root, |
f82d02d9 | 6611 | struct btrfs_path *path, |
2c47e605 | 6612 | struct walk_control *wc, int max_level) |
20524f02 | 6613 | { |
2c47e605 | 6614 | int level = wc->level; |
20524f02 | 6615 | int ret; |
9f3a7427 | 6616 | |
2c47e605 YZ |
6617 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6618 | while (level < max_level && path->nodes[level]) { | |
6619 | wc->level = level; | |
6620 | if (path->slots[level] + 1 < | |
6621 | btrfs_header_nritems(path->nodes[level])) { | |
6622 | path->slots[level]++; | |
20524f02 CM |
6623 | return 0; |
6624 | } else { | |
2c47e605 YZ |
6625 | ret = walk_up_proc(trans, root, path, wc); |
6626 | if (ret > 0) | |
6627 | return 0; | |
bd56b302 | 6628 | |
2c47e605 | 6629 | if (path->locks[level]) { |
bd681513 CM |
6630 | btrfs_tree_unlock_rw(path->nodes[level], |
6631 | path->locks[level]); | |
2c47e605 | 6632 | path->locks[level] = 0; |
f82d02d9 | 6633 | } |
2c47e605 YZ |
6634 | free_extent_buffer(path->nodes[level]); |
6635 | path->nodes[level] = NULL; | |
6636 | level++; | |
20524f02 CM |
6637 | } |
6638 | } | |
6639 | return 1; | |
6640 | } | |
6641 | ||
9aca1d51 | 6642 | /* |
2c47e605 YZ |
6643 | * drop a subvolume tree. |
6644 | * | |
6645 | * this function traverses the tree freeing any blocks that only | |
6646 | * referenced by the tree. | |
6647 | * | |
6648 | * when a shared tree block is found. this function decreases its | |
6649 | * reference count by one. if update_ref is true, this function | |
6650 | * also make sure backrefs for the shared block and all lower level | |
6651 | * blocks are properly updated. | |
9aca1d51 | 6652 | */ |
cb1b69f4 | 6653 | void btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
6654 | struct btrfs_block_rsv *block_rsv, int update_ref, |
6655 | int for_reloc) | |
20524f02 | 6656 | { |
5caf2a00 | 6657 | struct btrfs_path *path; |
2c47e605 YZ |
6658 | struct btrfs_trans_handle *trans; |
6659 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6660 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6661 | struct walk_control *wc; |
6662 | struct btrfs_key key; | |
6663 | int err = 0; | |
6664 | int ret; | |
6665 | int level; | |
20524f02 | 6666 | |
5caf2a00 | 6667 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
6668 | if (!path) { |
6669 | err = -ENOMEM; | |
6670 | goto out; | |
6671 | } | |
20524f02 | 6672 | |
2c47e605 | 6673 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
6674 | if (!wc) { |
6675 | btrfs_free_path(path); | |
cb1b69f4 TI |
6676 | err = -ENOMEM; |
6677 | goto out; | |
38a1a919 | 6678 | } |
2c47e605 | 6679 | |
a22285a6 | 6680 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 TI |
6681 | BUG_ON(IS_ERR(trans)); |
6682 | ||
3fd0a558 YZ |
6683 | if (block_rsv) |
6684 | trans->block_rsv = block_rsv; | |
2c47e605 | 6685 | |
9f3a7427 | 6686 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6687 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6688 | path->nodes[level] = btrfs_lock_root_node(root); |
6689 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6690 | path->slots[level] = 0; |
bd681513 | 6691 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6692 | memset(&wc->update_progress, 0, |
6693 | sizeof(wc->update_progress)); | |
9f3a7427 | 6694 | } else { |
9f3a7427 | 6695 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6696 | memcpy(&wc->update_progress, &key, |
6697 | sizeof(wc->update_progress)); | |
6698 | ||
6702ed49 | 6699 | level = root_item->drop_level; |
2c47e605 | 6700 | BUG_ON(level == 0); |
6702ed49 | 6701 | path->lowest_level = level; |
2c47e605 YZ |
6702 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6703 | path->lowest_level = 0; | |
6704 | if (ret < 0) { | |
6705 | err = ret; | |
cb1b69f4 | 6706 | goto out_free; |
9f3a7427 | 6707 | } |
1c4850e2 | 6708 | WARN_ON(ret > 0); |
2c47e605 | 6709 | |
7d9eb12c CM |
6710 | /* |
6711 | * unlock our path, this is safe because only this | |
6712 | * function is allowed to delete this snapshot | |
6713 | */ | |
5d4f98a2 | 6714 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
6715 | |
6716 | level = btrfs_header_level(root->node); | |
6717 | while (1) { | |
6718 | btrfs_tree_lock(path->nodes[level]); | |
6719 | btrfs_set_lock_blocking(path->nodes[level]); | |
6720 | ||
6721 | ret = btrfs_lookup_extent_info(trans, root, | |
6722 | path->nodes[level]->start, | |
6723 | path->nodes[level]->len, | |
6724 | &wc->refs[level], | |
6725 | &wc->flags[level]); | |
6726 | BUG_ON(ret); | |
6727 | BUG_ON(wc->refs[level] == 0); | |
6728 | ||
6729 | if (level == root_item->drop_level) | |
6730 | break; | |
6731 | ||
6732 | btrfs_tree_unlock(path->nodes[level]); | |
6733 | WARN_ON(wc->refs[level] != 1); | |
6734 | level--; | |
6735 | } | |
9f3a7427 | 6736 | } |
2c47e605 YZ |
6737 | |
6738 | wc->level = level; | |
6739 | wc->shared_level = -1; | |
6740 | wc->stage = DROP_REFERENCE; | |
6741 | wc->update_ref = update_ref; | |
6742 | wc->keep_locks = 0; | |
66d7e7f0 | 6743 | wc->for_reloc = for_reloc; |
1c4850e2 | 6744 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 6745 | |
d397712b | 6746 | while (1) { |
2c47e605 YZ |
6747 | ret = walk_down_tree(trans, root, path, wc); |
6748 | if (ret < 0) { | |
6749 | err = ret; | |
20524f02 | 6750 | break; |
2c47e605 | 6751 | } |
9aca1d51 | 6752 | |
2c47e605 YZ |
6753 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
6754 | if (ret < 0) { | |
6755 | err = ret; | |
20524f02 | 6756 | break; |
2c47e605 YZ |
6757 | } |
6758 | ||
6759 | if (ret > 0) { | |
6760 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
6761 | break; |
6762 | } | |
2c47e605 YZ |
6763 | |
6764 | if (wc->stage == DROP_REFERENCE) { | |
6765 | level = wc->level; | |
6766 | btrfs_node_key(path->nodes[level], | |
6767 | &root_item->drop_progress, | |
6768 | path->slots[level]); | |
6769 | root_item->drop_level = level; | |
6770 | } | |
6771 | ||
6772 | BUG_ON(wc->level == 0); | |
3fd0a558 | 6773 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
6774 | ret = btrfs_update_root(trans, tree_root, |
6775 | &root->root_key, | |
6776 | root_item); | |
6777 | BUG_ON(ret); | |
6778 | ||
3fd0a558 | 6779 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 6780 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 | 6781 | BUG_ON(IS_ERR(trans)); |
3fd0a558 YZ |
6782 | if (block_rsv) |
6783 | trans->block_rsv = block_rsv; | |
c3e69d58 | 6784 | } |
20524f02 | 6785 | } |
b3b4aa74 | 6786 | btrfs_release_path(path); |
2c47e605 YZ |
6787 | BUG_ON(err); |
6788 | ||
6789 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
6790 | BUG_ON(ret); | |
6791 | ||
76dda93c YZ |
6792 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
6793 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
6794 | NULL, NULL); | |
6795 | BUG_ON(ret < 0); | |
6796 | if (ret > 0) { | |
84cd948c JB |
6797 | /* if we fail to delete the orphan item this time |
6798 | * around, it'll get picked up the next time. | |
6799 | * | |
6800 | * The most common failure here is just -ENOENT. | |
6801 | */ | |
6802 | btrfs_del_orphan_item(trans, tree_root, | |
6803 | root->root_key.objectid); | |
76dda93c YZ |
6804 | } |
6805 | } | |
6806 | ||
6807 | if (root->in_radix) { | |
6808 | btrfs_free_fs_root(tree_root->fs_info, root); | |
6809 | } else { | |
6810 | free_extent_buffer(root->node); | |
6811 | free_extent_buffer(root->commit_root); | |
6812 | kfree(root); | |
6813 | } | |
cb1b69f4 | 6814 | out_free: |
3fd0a558 | 6815 | btrfs_end_transaction_throttle(trans, tree_root); |
2c47e605 | 6816 | kfree(wc); |
5caf2a00 | 6817 | btrfs_free_path(path); |
cb1b69f4 TI |
6818 | out: |
6819 | if (err) | |
6820 | btrfs_std_error(root->fs_info, err); | |
6821 | return; | |
20524f02 | 6822 | } |
9078a3e1 | 6823 | |
2c47e605 YZ |
6824 | /* |
6825 | * drop subtree rooted at tree block 'node'. | |
6826 | * | |
6827 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 6828 | * only used by relocation code |
2c47e605 | 6829 | */ |
f82d02d9 YZ |
6830 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
6831 | struct btrfs_root *root, | |
6832 | struct extent_buffer *node, | |
6833 | struct extent_buffer *parent) | |
6834 | { | |
6835 | struct btrfs_path *path; | |
2c47e605 | 6836 | struct walk_control *wc; |
f82d02d9 YZ |
6837 | int level; |
6838 | int parent_level; | |
6839 | int ret = 0; | |
6840 | int wret; | |
6841 | ||
2c47e605 YZ |
6842 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
6843 | ||
f82d02d9 | 6844 | path = btrfs_alloc_path(); |
db5b493a TI |
6845 | if (!path) |
6846 | return -ENOMEM; | |
f82d02d9 | 6847 | |
2c47e605 | 6848 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
6849 | if (!wc) { |
6850 | btrfs_free_path(path); | |
6851 | return -ENOMEM; | |
6852 | } | |
2c47e605 | 6853 | |
b9447ef8 | 6854 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
6855 | parent_level = btrfs_header_level(parent); |
6856 | extent_buffer_get(parent); | |
6857 | path->nodes[parent_level] = parent; | |
6858 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
6859 | ||
b9447ef8 | 6860 | btrfs_assert_tree_locked(node); |
f82d02d9 | 6861 | level = btrfs_header_level(node); |
f82d02d9 YZ |
6862 | path->nodes[level] = node; |
6863 | path->slots[level] = 0; | |
bd681513 | 6864 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6865 | |
6866 | wc->refs[parent_level] = 1; | |
6867 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6868 | wc->level = level; | |
6869 | wc->shared_level = -1; | |
6870 | wc->stage = DROP_REFERENCE; | |
6871 | wc->update_ref = 0; | |
6872 | wc->keep_locks = 1; | |
66d7e7f0 | 6873 | wc->for_reloc = 1; |
1c4850e2 | 6874 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
6875 | |
6876 | while (1) { | |
2c47e605 YZ |
6877 | wret = walk_down_tree(trans, root, path, wc); |
6878 | if (wret < 0) { | |
f82d02d9 | 6879 | ret = wret; |
f82d02d9 | 6880 | break; |
2c47e605 | 6881 | } |
f82d02d9 | 6882 | |
2c47e605 | 6883 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
6884 | if (wret < 0) |
6885 | ret = wret; | |
6886 | if (wret != 0) | |
6887 | break; | |
6888 | } | |
6889 | ||
2c47e605 | 6890 | kfree(wc); |
f82d02d9 YZ |
6891 | btrfs_free_path(path); |
6892 | return ret; | |
6893 | } | |
6894 | ||
ec44a35c CM |
6895 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
6896 | { | |
6897 | u64 num_devices; | |
6898 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
6899 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
6900 | ||
e4d8ec0f ID |
6901 | if (root->fs_info->balance_ctl) { |
6902 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
6903 | u64 tgt = 0; | |
6904 | ||
6905 | /* pick restriper's target profile and return */ | |
6906 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
6907 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
6908 | tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
6909 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
6910 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
6911 | tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
6912 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
6913 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
6914 | tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
6915 | } | |
6916 | ||
6917 | if (tgt) { | |
6918 | /* extended -> chunk profile */ | |
6919 | tgt &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
6920 | return tgt; | |
6921 | } | |
6922 | } | |
6923 | ||
cd02dca5 CM |
6924 | /* |
6925 | * we add in the count of missing devices because we want | |
6926 | * to make sure that any RAID levels on a degraded FS | |
6927 | * continue to be honored. | |
6928 | */ | |
6929 | num_devices = root->fs_info->fs_devices->rw_devices + | |
6930 | root->fs_info->fs_devices->missing_devices; | |
6931 | ||
ec44a35c CM |
6932 | if (num_devices == 1) { |
6933 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6934 | stripped = flags & ~stripped; | |
6935 | ||
6936 | /* turn raid0 into single device chunks */ | |
6937 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
6938 | return stripped; | |
6939 | ||
6940 | /* turn mirroring into duplication */ | |
6941 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
6942 | BTRFS_BLOCK_GROUP_RAID10)) | |
6943 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
6944 | return flags; | |
6945 | } else { | |
6946 | /* they already had raid on here, just return */ | |
ec44a35c CM |
6947 | if (flags & stripped) |
6948 | return flags; | |
6949 | ||
6950 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6951 | stripped = flags & ~stripped; | |
6952 | ||
6953 | /* switch duplicated blocks with raid1 */ | |
6954 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
6955 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
6956 | ||
6957 | /* turn single device chunks into raid0 */ | |
6958 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
6959 | } | |
6960 | return flags; | |
6961 | } | |
6962 | ||
199c36ea | 6963 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 6964 | { |
f0486c68 YZ |
6965 | struct btrfs_space_info *sinfo = cache->space_info; |
6966 | u64 num_bytes; | |
199c36ea | 6967 | u64 min_allocable_bytes; |
f0486c68 | 6968 | int ret = -ENOSPC; |
0ef3e66b | 6969 | |
c286ac48 | 6970 | |
199c36ea MX |
6971 | /* |
6972 | * We need some metadata space and system metadata space for | |
6973 | * allocating chunks in some corner cases until we force to set | |
6974 | * it to be readonly. | |
6975 | */ | |
6976 | if ((sinfo->flags & | |
6977 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
6978 | !force) | |
6979 | min_allocable_bytes = 1 * 1024 * 1024; | |
6980 | else | |
6981 | min_allocable_bytes = 0; | |
6982 | ||
f0486c68 YZ |
6983 | spin_lock(&sinfo->lock); |
6984 | spin_lock(&cache->lock); | |
61cfea9b W |
6985 | |
6986 | if (cache->ro) { | |
6987 | ret = 0; | |
6988 | goto out; | |
6989 | } | |
6990 | ||
f0486c68 YZ |
6991 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
6992 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
6993 | ||
6994 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
6995 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
6996 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 6997 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
6998 | cache->ro = 1; |
6999 | ret = 0; | |
7000 | } | |
61cfea9b | 7001 | out: |
f0486c68 YZ |
7002 | spin_unlock(&cache->lock); |
7003 | spin_unlock(&sinfo->lock); | |
7004 | return ret; | |
7005 | } | |
7d9eb12c | 7006 | |
f0486c68 YZ |
7007 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
7008 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 7009 | |
f0486c68 YZ |
7010 | { |
7011 | struct btrfs_trans_handle *trans; | |
7012 | u64 alloc_flags; | |
7013 | int ret; | |
7d9eb12c | 7014 | |
f0486c68 | 7015 | BUG_ON(cache->ro); |
0ef3e66b | 7016 | |
ff5714cc | 7017 | trans = btrfs_join_transaction(root); |
f0486c68 | 7018 | BUG_ON(IS_ERR(trans)); |
5d4f98a2 | 7019 | |
f0486c68 YZ |
7020 | alloc_flags = update_block_group_flags(root, cache->flags); |
7021 | if (alloc_flags != cache->flags) | |
0e4f8f88 CM |
7022 | do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7023 | CHUNK_ALLOC_FORCE); | |
5d4f98a2 | 7024 | |
199c36ea | 7025 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7026 | if (!ret) |
7027 | goto out; | |
7028 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
0e4f8f88 CM |
7029 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7030 | CHUNK_ALLOC_FORCE); | |
f0486c68 YZ |
7031 | if (ret < 0) |
7032 | goto out; | |
199c36ea | 7033 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7034 | out: |
7035 | btrfs_end_transaction(trans, root); | |
7036 | return ret; | |
7037 | } | |
5d4f98a2 | 7038 | |
c87f08ca CM |
7039 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7040 | struct btrfs_root *root, u64 type) | |
7041 | { | |
7042 | u64 alloc_flags = get_alloc_profile(root, type); | |
0e4f8f88 CM |
7043 | return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7044 | CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
7045 | } |
7046 | ||
6d07bcec MX |
7047 | /* |
7048 | * helper to account the unused space of all the readonly block group in the | |
7049 | * list. takes mirrors into account. | |
7050 | */ | |
7051 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7052 | { | |
7053 | struct btrfs_block_group_cache *block_group; | |
7054 | u64 free_bytes = 0; | |
7055 | int factor; | |
7056 | ||
7057 | list_for_each_entry(block_group, groups_list, list) { | |
7058 | spin_lock(&block_group->lock); | |
7059 | ||
7060 | if (!block_group->ro) { | |
7061 | spin_unlock(&block_group->lock); | |
7062 | continue; | |
7063 | } | |
7064 | ||
7065 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7066 | BTRFS_BLOCK_GROUP_RAID10 | | |
7067 | BTRFS_BLOCK_GROUP_DUP)) | |
7068 | factor = 2; | |
7069 | else | |
7070 | factor = 1; | |
7071 | ||
7072 | free_bytes += (block_group->key.offset - | |
7073 | btrfs_block_group_used(&block_group->item)) * | |
7074 | factor; | |
7075 | ||
7076 | spin_unlock(&block_group->lock); | |
7077 | } | |
7078 | ||
7079 | return free_bytes; | |
7080 | } | |
7081 | ||
7082 | /* | |
7083 | * helper to account the unused space of all the readonly block group in the | |
7084 | * space_info. takes mirrors into account. | |
7085 | */ | |
7086 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
7087 | { | |
7088 | int i; | |
7089 | u64 free_bytes = 0; | |
7090 | ||
7091 | spin_lock(&sinfo->lock); | |
7092 | ||
7093 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
7094 | if (!list_empty(&sinfo->block_groups[i])) | |
7095 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
7096 | &sinfo->block_groups[i]); | |
7097 | ||
7098 | spin_unlock(&sinfo->lock); | |
7099 | ||
7100 | return free_bytes; | |
7101 | } | |
7102 | ||
f0486c68 YZ |
7103 | int btrfs_set_block_group_rw(struct btrfs_root *root, |
7104 | struct btrfs_block_group_cache *cache) | |
5d4f98a2 | 7105 | { |
f0486c68 YZ |
7106 | struct btrfs_space_info *sinfo = cache->space_info; |
7107 | u64 num_bytes; | |
7108 | ||
7109 | BUG_ON(!cache->ro); | |
7110 | ||
7111 | spin_lock(&sinfo->lock); | |
7112 | spin_lock(&cache->lock); | |
7113 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7114 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7115 | sinfo->bytes_readonly -= num_bytes; | |
7116 | cache->ro = 0; | |
7117 | spin_unlock(&cache->lock); | |
7118 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
7119 | return 0; |
7120 | } | |
7121 | ||
ba1bf481 JB |
7122 | /* |
7123 | * checks to see if its even possible to relocate this block group. | |
7124 | * | |
7125 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7126 | * ok to go ahead and try. | |
7127 | */ | |
7128 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7129 | { |
ba1bf481 JB |
7130 | struct btrfs_block_group_cache *block_group; |
7131 | struct btrfs_space_info *space_info; | |
7132 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7133 | struct btrfs_device *device; | |
cdcb725c | 7134 | u64 min_free; |
6719db6a JB |
7135 | u64 dev_min = 1; |
7136 | u64 dev_nr = 0; | |
cdcb725c | 7137 | int index; |
ba1bf481 JB |
7138 | int full = 0; |
7139 | int ret = 0; | |
1a40e23b | 7140 | |
ba1bf481 | 7141 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7142 | |
ba1bf481 JB |
7143 | /* odd, couldn't find the block group, leave it alone */ |
7144 | if (!block_group) | |
7145 | return -1; | |
1a40e23b | 7146 | |
cdcb725c | 7147 | min_free = btrfs_block_group_used(&block_group->item); |
7148 | ||
ba1bf481 | 7149 | /* no bytes used, we're good */ |
cdcb725c | 7150 | if (!min_free) |
1a40e23b ZY |
7151 | goto out; |
7152 | ||
ba1bf481 JB |
7153 | space_info = block_group->space_info; |
7154 | spin_lock(&space_info->lock); | |
17d217fe | 7155 | |
ba1bf481 | 7156 | full = space_info->full; |
17d217fe | 7157 | |
ba1bf481 JB |
7158 | /* |
7159 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7160 | * relocate it unless we're able to allocate a new chunk below. |
7161 | * | |
7162 | * Otherwise, we need to make sure we have room in the space to handle | |
7163 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7164 | */ |
7ce618db | 7165 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 7166 | (space_info->bytes_used + space_info->bytes_reserved + |
7167 | space_info->bytes_pinned + space_info->bytes_readonly + | |
7168 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
7169 | spin_unlock(&space_info->lock); |
7170 | goto out; | |
17d217fe | 7171 | } |
ba1bf481 | 7172 | spin_unlock(&space_info->lock); |
ea8c2819 | 7173 | |
ba1bf481 JB |
7174 | /* |
7175 | * ok we don't have enough space, but maybe we have free space on our | |
7176 | * devices to allocate new chunks for relocation, so loop through our | |
7177 | * alloc devices and guess if we have enough space. However, if we | |
7178 | * were marked as full, then we know there aren't enough chunks, and we | |
7179 | * can just return. | |
7180 | */ | |
7181 | ret = -1; | |
7182 | if (full) | |
7183 | goto out; | |
ea8c2819 | 7184 | |
cdcb725c | 7185 | /* |
7186 | * index: | |
7187 | * 0: raid10 | |
7188 | * 1: raid1 | |
7189 | * 2: dup | |
7190 | * 3: raid0 | |
7191 | * 4: single | |
7192 | */ | |
7193 | index = get_block_group_index(block_group); | |
7194 | if (index == 0) { | |
7195 | dev_min = 4; | |
6719db6a JB |
7196 | /* Divide by 2 */ |
7197 | min_free >>= 1; | |
cdcb725c | 7198 | } else if (index == 1) { |
7199 | dev_min = 2; | |
7200 | } else if (index == 2) { | |
6719db6a JB |
7201 | /* Multiply by 2 */ |
7202 | min_free <<= 1; | |
cdcb725c | 7203 | } else if (index == 3) { |
7204 | dev_min = fs_devices->rw_devices; | |
6719db6a | 7205 | do_div(min_free, dev_min); |
cdcb725c | 7206 | } |
7207 | ||
ba1bf481 JB |
7208 | mutex_lock(&root->fs_info->chunk_mutex); |
7209 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 7210 | u64 dev_offset; |
56bec294 | 7211 | |
ba1bf481 JB |
7212 | /* |
7213 | * check to make sure we can actually find a chunk with enough | |
7214 | * space to fit our block group in. | |
7215 | */ | |
7216 | if (device->total_bytes > device->bytes_used + min_free) { | |
125ccb0a | 7217 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 7218 | &dev_offset, NULL); |
ba1bf481 | 7219 | if (!ret) |
cdcb725c | 7220 | dev_nr++; |
7221 | ||
7222 | if (dev_nr >= dev_min) | |
73e48b27 | 7223 | break; |
cdcb725c | 7224 | |
ba1bf481 | 7225 | ret = -1; |
725c8463 | 7226 | } |
edbd8d4e | 7227 | } |
ba1bf481 | 7228 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7229 | out: |
ba1bf481 | 7230 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7231 | return ret; |
7232 | } | |
7233 | ||
b2950863 CH |
7234 | static int find_first_block_group(struct btrfs_root *root, |
7235 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7236 | { |
925baedd | 7237 | int ret = 0; |
0b86a832 CM |
7238 | struct btrfs_key found_key; |
7239 | struct extent_buffer *leaf; | |
7240 | int slot; | |
edbd8d4e | 7241 | |
0b86a832 CM |
7242 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7243 | if (ret < 0) | |
925baedd CM |
7244 | goto out; |
7245 | ||
d397712b | 7246 | while (1) { |
0b86a832 | 7247 | slot = path->slots[0]; |
edbd8d4e | 7248 | leaf = path->nodes[0]; |
0b86a832 CM |
7249 | if (slot >= btrfs_header_nritems(leaf)) { |
7250 | ret = btrfs_next_leaf(root, path); | |
7251 | if (ret == 0) | |
7252 | continue; | |
7253 | if (ret < 0) | |
925baedd | 7254 | goto out; |
0b86a832 | 7255 | break; |
edbd8d4e | 7256 | } |
0b86a832 | 7257 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7258 | |
0b86a832 | 7259 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7260 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7261 | ret = 0; | |
7262 | goto out; | |
7263 | } | |
0b86a832 | 7264 | path->slots[0]++; |
edbd8d4e | 7265 | } |
925baedd | 7266 | out: |
0b86a832 | 7267 | return ret; |
edbd8d4e CM |
7268 | } |
7269 | ||
0af3d00b JB |
7270 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
7271 | { | |
7272 | struct btrfs_block_group_cache *block_group; | |
7273 | u64 last = 0; | |
7274 | ||
7275 | while (1) { | |
7276 | struct inode *inode; | |
7277 | ||
7278 | block_group = btrfs_lookup_first_block_group(info, last); | |
7279 | while (block_group) { | |
7280 | spin_lock(&block_group->lock); | |
7281 | if (block_group->iref) | |
7282 | break; | |
7283 | spin_unlock(&block_group->lock); | |
7284 | block_group = next_block_group(info->tree_root, | |
7285 | block_group); | |
7286 | } | |
7287 | if (!block_group) { | |
7288 | if (last == 0) | |
7289 | break; | |
7290 | last = 0; | |
7291 | continue; | |
7292 | } | |
7293 | ||
7294 | inode = block_group->inode; | |
7295 | block_group->iref = 0; | |
7296 | block_group->inode = NULL; | |
7297 | spin_unlock(&block_group->lock); | |
7298 | iput(inode); | |
7299 | last = block_group->key.objectid + block_group->key.offset; | |
7300 | btrfs_put_block_group(block_group); | |
7301 | } | |
7302 | } | |
7303 | ||
1a40e23b ZY |
7304 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7305 | { | |
7306 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7307 | struct btrfs_space_info *space_info; |
11833d66 | 7308 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7309 | struct rb_node *n; |
7310 | ||
11833d66 YZ |
7311 | down_write(&info->extent_commit_sem); |
7312 | while (!list_empty(&info->caching_block_groups)) { | |
7313 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7314 | struct btrfs_caching_control, list); | |
7315 | list_del(&caching_ctl->list); | |
7316 | put_caching_control(caching_ctl); | |
7317 | } | |
7318 | up_write(&info->extent_commit_sem); | |
7319 | ||
1a40e23b ZY |
7320 | spin_lock(&info->block_group_cache_lock); |
7321 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7322 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7323 | cache_node); | |
1a40e23b ZY |
7324 | rb_erase(&block_group->cache_node, |
7325 | &info->block_group_cache_tree); | |
d899e052 YZ |
7326 | spin_unlock(&info->block_group_cache_lock); |
7327 | ||
80eb234a | 7328 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7329 | list_del(&block_group->list); |
80eb234a | 7330 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7331 | |
817d52f8 | 7332 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7333 | wait_block_group_cache_done(block_group); |
817d52f8 | 7334 | |
3c14874a JB |
7335 | /* |
7336 | * We haven't cached this block group, which means we could | |
7337 | * possibly have excluded extents on this block group. | |
7338 | */ | |
7339 | if (block_group->cached == BTRFS_CACHE_NO) | |
7340 | free_excluded_extents(info->extent_root, block_group); | |
7341 | ||
817d52f8 | 7342 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 7343 | btrfs_put_block_group(block_group); |
d899e052 YZ |
7344 | |
7345 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7346 | } |
7347 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7348 | |
7349 | /* now that all the block groups are freed, go through and | |
7350 | * free all the space_info structs. This is only called during | |
7351 | * the final stages of unmount, and so we know nobody is | |
7352 | * using them. We call synchronize_rcu() once before we start, | |
7353 | * just to be on the safe side. | |
7354 | */ | |
7355 | synchronize_rcu(); | |
7356 | ||
8929ecfa YZ |
7357 | release_global_block_rsv(info); |
7358 | ||
4184ea7f CM |
7359 | while(!list_empty(&info->space_info)) { |
7360 | space_info = list_entry(info->space_info.next, | |
7361 | struct btrfs_space_info, | |
7362 | list); | |
f0486c68 | 7363 | if (space_info->bytes_pinned > 0 || |
fb25e914 JB |
7364 | space_info->bytes_reserved > 0 || |
7365 | space_info->bytes_may_use > 0) { | |
f0486c68 YZ |
7366 | WARN_ON(1); |
7367 | dump_space_info(space_info, 0, 0); | |
7368 | } | |
4184ea7f CM |
7369 | list_del(&space_info->list); |
7370 | kfree(space_info); | |
7371 | } | |
1a40e23b ZY |
7372 | return 0; |
7373 | } | |
7374 | ||
b742bb82 YZ |
7375 | static void __link_block_group(struct btrfs_space_info *space_info, |
7376 | struct btrfs_block_group_cache *cache) | |
7377 | { | |
7378 | int index = get_block_group_index(cache); | |
7379 | ||
7380 | down_write(&space_info->groups_sem); | |
7381 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
7382 | up_write(&space_info->groups_sem); | |
7383 | } | |
7384 | ||
9078a3e1 CM |
7385 | int btrfs_read_block_groups(struct btrfs_root *root) |
7386 | { | |
7387 | struct btrfs_path *path; | |
7388 | int ret; | |
9078a3e1 | 7389 | struct btrfs_block_group_cache *cache; |
be744175 | 7390 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7391 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7392 | struct btrfs_key key; |
7393 | struct btrfs_key found_key; | |
5f39d397 | 7394 | struct extent_buffer *leaf; |
0af3d00b JB |
7395 | int need_clear = 0; |
7396 | u64 cache_gen; | |
96b5179d | 7397 | |
be744175 | 7398 | root = info->extent_root; |
9078a3e1 | 7399 | key.objectid = 0; |
0b86a832 | 7400 | key.offset = 0; |
9078a3e1 | 7401 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7402 | path = btrfs_alloc_path(); |
7403 | if (!path) | |
7404 | return -ENOMEM; | |
026fd317 | 7405 | path->reada = 1; |
9078a3e1 | 7406 | |
6c41761f | 7407 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 7408 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 7409 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 7410 | need_clear = 1; |
88c2ba3b JB |
7411 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
7412 | need_clear = 1; | |
0af3d00b | 7413 | |
d397712b | 7414 | while (1) { |
0b86a832 | 7415 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
7416 | if (ret > 0) |
7417 | break; | |
0b86a832 CM |
7418 | if (ret != 0) |
7419 | goto error; | |
5f39d397 CM |
7420 | leaf = path->nodes[0]; |
7421 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7422 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7423 | if (!cache) { |
0b86a832 | 7424 | ret = -ENOMEM; |
f0486c68 | 7425 | goto error; |
9078a3e1 | 7426 | } |
34d52cb6 LZ |
7427 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7428 | GFP_NOFS); | |
7429 | if (!cache->free_space_ctl) { | |
7430 | kfree(cache); | |
7431 | ret = -ENOMEM; | |
7432 | goto error; | |
7433 | } | |
3e1ad54f | 7434 | |
d2fb3437 | 7435 | atomic_set(&cache->count, 1); |
c286ac48 | 7436 | spin_lock_init(&cache->lock); |
817d52f8 | 7437 | cache->fs_info = info; |
0f9dd46c | 7438 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7439 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 7440 | |
0af3d00b JB |
7441 | if (need_clear) |
7442 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
7443 | ||
5f39d397 CM |
7444 | read_extent_buffer(leaf, &cache->item, |
7445 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7446 | sizeof(cache->item)); | |
9078a3e1 | 7447 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7448 | |
9078a3e1 | 7449 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 7450 | btrfs_release_path(path); |
0b86a832 | 7451 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7452 | cache->sectorsize = root->sectorsize; |
7453 | ||
34d52cb6 LZ |
7454 | btrfs_init_free_space_ctl(cache); |
7455 | ||
3c14874a JB |
7456 | /* |
7457 | * We need to exclude the super stripes now so that the space | |
7458 | * info has super bytes accounted for, otherwise we'll think | |
7459 | * we have more space than we actually do. | |
7460 | */ | |
7461 | exclude_super_stripes(root, cache); | |
7462 | ||
817d52f8 JB |
7463 | /* |
7464 | * check for two cases, either we are full, and therefore | |
7465 | * don't need to bother with the caching work since we won't | |
7466 | * find any space, or we are empty, and we can just add all | |
7467 | * the space in and be done with it. This saves us _alot_ of | |
7468 | * time, particularly in the full case. | |
7469 | */ | |
7470 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 7471 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7472 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7473 | free_excluded_extents(root, cache); |
817d52f8 | 7474 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 7475 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
7476 | cache->cached = BTRFS_CACHE_FINISHED; |
7477 | add_new_free_space(cache, root->fs_info, | |
7478 | found_key.objectid, | |
7479 | found_key.objectid + | |
7480 | found_key.offset); | |
11833d66 | 7481 | free_excluded_extents(root, cache); |
817d52f8 | 7482 | } |
96b5179d | 7483 | |
6324fbf3 CM |
7484 | ret = update_space_info(info, cache->flags, found_key.offset, |
7485 | btrfs_block_group_used(&cache->item), | |
7486 | &space_info); | |
7487 | BUG_ON(ret); | |
7488 | cache->space_info = space_info; | |
1b2da372 | 7489 | spin_lock(&cache->space_info->lock); |
f0486c68 | 7490 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7491 | spin_unlock(&cache->space_info->lock); |
7492 | ||
b742bb82 | 7493 | __link_block_group(space_info, cache); |
0f9dd46c JB |
7494 | |
7495 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
7496 | BUG_ON(ret); | |
75ccf47d CM |
7497 | |
7498 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 7499 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 7500 | set_block_group_ro(cache, 1); |
9078a3e1 | 7501 | } |
b742bb82 YZ |
7502 | |
7503 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
7504 | if (!(get_alloc_profile(root, space_info->flags) & | |
7505 | (BTRFS_BLOCK_GROUP_RAID10 | | |
7506 | BTRFS_BLOCK_GROUP_RAID1 | | |
7507 | BTRFS_BLOCK_GROUP_DUP))) | |
7508 | continue; | |
7509 | /* | |
7510 | * avoid allocating from un-mirrored block group if there are | |
7511 | * mirrored block groups. | |
7512 | */ | |
7513 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 7514 | set_block_group_ro(cache, 1); |
b742bb82 | 7515 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 7516 | set_block_group_ro(cache, 1); |
9078a3e1 | 7517 | } |
f0486c68 YZ |
7518 | |
7519 | init_global_block_rsv(info); | |
0b86a832 CM |
7520 | ret = 0; |
7521 | error: | |
9078a3e1 | 7522 | btrfs_free_path(path); |
0b86a832 | 7523 | return ret; |
9078a3e1 | 7524 | } |
6324fbf3 CM |
7525 | |
7526 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7527 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7528 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7529 | u64 size) |
7530 | { | |
7531 | int ret; | |
6324fbf3 CM |
7532 | struct btrfs_root *extent_root; |
7533 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7534 | |
7535 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7536 | |
12fcfd22 | 7537 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7538 | |
8f18cf13 | 7539 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7540 | if (!cache) |
7541 | return -ENOMEM; | |
34d52cb6 LZ |
7542 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7543 | GFP_NOFS); | |
7544 | if (!cache->free_space_ctl) { | |
7545 | kfree(cache); | |
7546 | return -ENOMEM; | |
7547 | } | |
0f9dd46c | 7548 | |
e17cade2 | 7549 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7550 | cache->key.offset = size; |
d2fb3437 | 7551 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 7552 | cache->sectorsize = root->sectorsize; |
0af3d00b | 7553 | cache->fs_info = root->fs_info; |
96303081 | 7554 | |
d2fb3437 | 7555 | atomic_set(&cache->count, 1); |
c286ac48 | 7556 | spin_lock_init(&cache->lock); |
0f9dd46c | 7557 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7558 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7559 | |
34d52cb6 LZ |
7560 | btrfs_init_free_space_ctl(cache); |
7561 | ||
6324fbf3 | 7562 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7563 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7564 | cache->flags = type; | |
7565 | btrfs_set_block_group_flags(&cache->item, type); | |
7566 | ||
11833d66 | 7567 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7568 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7569 | exclude_super_stripes(root, cache); |
96303081 | 7570 | |
817d52f8 JB |
7571 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7572 | chunk_offset + size); | |
7573 | ||
11833d66 YZ |
7574 | free_excluded_extents(root, cache); |
7575 | ||
6324fbf3 CM |
7576 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7577 | &cache->space_info); | |
7578 | BUG_ON(ret); | |
c7c144db | 7579 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
7580 | |
7581 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 7582 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7583 | spin_unlock(&cache->space_info->lock); |
7584 | ||
b742bb82 | 7585 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 7586 | |
0f9dd46c JB |
7587 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
7588 | BUG_ON(ret); | |
c286ac48 | 7589 | |
6324fbf3 CM |
7590 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7591 | sizeof(cache->item)); | |
7592 | BUG_ON(ret); | |
7593 | ||
d18a2c44 | 7594 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7595 | |
6324fbf3 CM |
7596 | return 0; |
7597 | } | |
1a40e23b | 7598 | |
10ea00f5 ID |
7599 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
7600 | { | |
7601 | u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
7602 | ||
7603 | /* chunk -> extended profile */ | |
7604 | if (extra_flags == 0) | |
7605 | extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
7606 | ||
7607 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
7608 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
7609 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
7610 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
7611 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
7612 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
7613 | } | |
7614 | ||
1a40e23b ZY |
7615 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
7616 | struct btrfs_root *root, u64 group_start) | |
7617 | { | |
7618 | struct btrfs_path *path; | |
7619 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7620 | struct btrfs_free_cluster *cluster; |
0af3d00b | 7621 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 7622 | struct btrfs_key key; |
0af3d00b | 7623 | struct inode *inode; |
1a40e23b | 7624 | int ret; |
10ea00f5 | 7625 | int index; |
89a55897 | 7626 | int factor; |
1a40e23b | 7627 | |
1a40e23b ZY |
7628 | root = root->fs_info->extent_root; |
7629 | ||
7630 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7631 | BUG_ON(!block_group); | |
c146afad | 7632 | BUG_ON(!block_group->ro); |
1a40e23b | 7633 | |
9f7c43c9 | 7634 | /* |
7635 | * Free the reserved super bytes from this block group before | |
7636 | * remove it. | |
7637 | */ | |
7638 | free_excluded_extents(root, block_group); | |
7639 | ||
1a40e23b | 7640 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 7641 | index = get_block_group_index(block_group); |
89a55897 JB |
7642 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
7643 | BTRFS_BLOCK_GROUP_RAID1 | | |
7644 | BTRFS_BLOCK_GROUP_RAID10)) | |
7645 | factor = 2; | |
7646 | else | |
7647 | factor = 1; | |
1a40e23b | 7648 | |
44fb5511 CM |
7649 | /* make sure this block group isn't part of an allocation cluster */ |
7650 | cluster = &root->fs_info->data_alloc_cluster; | |
7651 | spin_lock(&cluster->refill_lock); | |
7652 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7653 | spin_unlock(&cluster->refill_lock); | |
7654 | ||
7655 | /* | |
7656 | * make sure this block group isn't part of a metadata | |
7657 | * allocation cluster | |
7658 | */ | |
7659 | cluster = &root->fs_info->meta_alloc_cluster; | |
7660 | spin_lock(&cluster->refill_lock); | |
7661 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7662 | spin_unlock(&cluster->refill_lock); | |
7663 | ||
1a40e23b | 7664 | path = btrfs_alloc_path(); |
d8926bb3 MF |
7665 | if (!path) { |
7666 | ret = -ENOMEM; | |
7667 | goto out; | |
7668 | } | |
1a40e23b | 7669 | |
10b2f34d | 7670 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 7671 | if (!IS_ERR(inode)) { |
b532402e TI |
7672 | ret = btrfs_orphan_add(trans, inode); |
7673 | BUG_ON(ret); | |
0af3d00b JB |
7674 | clear_nlink(inode); |
7675 | /* One for the block groups ref */ | |
7676 | spin_lock(&block_group->lock); | |
7677 | if (block_group->iref) { | |
7678 | block_group->iref = 0; | |
7679 | block_group->inode = NULL; | |
7680 | spin_unlock(&block_group->lock); | |
7681 | iput(inode); | |
7682 | } else { | |
7683 | spin_unlock(&block_group->lock); | |
7684 | } | |
7685 | /* One for our lookup ref */ | |
455757c3 | 7686 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
7687 | } |
7688 | ||
7689 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
7690 | key.offset = block_group->key.objectid; | |
7691 | key.type = 0; | |
7692 | ||
7693 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
7694 | if (ret < 0) | |
7695 | goto out; | |
7696 | if (ret > 0) | |
b3b4aa74 | 7697 | btrfs_release_path(path); |
0af3d00b JB |
7698 | if (ret == 0) { |
7699 | ret = btrfs_del_item(trans, tree_root, path); | |
7700 | if (ret) | |
7701 | goto out; | |
b3b4aa74 | 7702 | btrfs_release_path(path); |
0af3d00b JB |
7703 | } |
7704 | ||
3dfdb934 | 7705 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
7706 | rb_erase(&block_group->cache_node, |
7707 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 7708 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 7709 | |
80eb234a | 7710 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
7711 | /* |
7712 | * we must use list_del_init so people can check to see if they | |
7713 | * are still on the list after taking the semaphore | |
7714 | */ | |
7715 | list_del_init(&block_group->list); | |
10ea00f5 ID |
7716 | if (list_empty(&block_group->space_info->block_groups[index])) |
7717 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 7718 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 7719 | |
817d52f8 | 7720 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7721 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
7722 | |
7723 | btrfs_remove_free_space_cache(block_group); | |
7724 | ||
c146afad YZ |
7725 | spin_lock(&block_group->space_info->lock); |
7726 | block_group->space_info->total_bytes -= block_group->key.offset; | |
7727 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 7728 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 7729 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 7730 | |
0af3d00b JB |
7731 | memcpy(&key, &block_group->key, sizeof(key)); |
7732 | ||
283bb197 | 7733 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 7734 | |
fa9c0d79 CM |
7735 | btrfs_put_block_group(block_group); |
7736 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
7737 | |
7738 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
7739 | if (ret > 0) | |
7740 | ret = -EIO; | |
7741 | if (ret < 0) | |
7742 | goto out; | |
7743 | ||
7744 | ret = btrfs_del_item(trans, root, path); | |
7745 | out: | |
7746 | btrfs_free_path(path); | |
7747 | return ret; | |
7748 | } | |
acce952b | 7749 | |
c59021f8 | 7750 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
7751 | { | |
7752 | struct btrfs_space_info *space_info; | |
1aba86d6 | 7753 | struct btrfs_super_block *disk_super; |
7754 | u64 features; | |
7755 | u64 flags; | |
7756 | int mixed = 0; | |
c59021f8 | 7757 | int ret; |
7758 | ||
6c41761f | 7759 | disk_super = fs_info->super_copy; |
1aba86d6 | 7760 | if (!btrfs_super_root(disk_super)) |
7761 | return 1; | |
c59021f8 | 7762 | |
1aba86d6 | 7763 | features = btrfs_super_incompat_flags(disk_super); |
7764 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
7765 | mixed = 1; | |
c59021f8 | 7766 | |
1aba86d6 | 7767 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
7768 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 7769 | if (ret) |
1aba86d6 | 7770 | goto out; |
c59021f8 | 7771 | |
1aba86d6 | 7772 | if (mixed) { |
7773 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
7774 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7775 | } else { | |
7776 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
7777 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7778 | if (ret) | |
7779 | goto out; | |
7780 | ||
7781 | flags = BTRFS_BLOCK_GROUP_DATA; | |
7782 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7783 | } | |
7784 | out: | |
c59021f8 | 7785 | return ret; |
7786 | } | |
7787 | ||
acce952b | 7788 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
7789 | { | |
7790 | return unpin_extent_range(root, start, end); | |
7791 | } | |
7792 | ||
7793 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 7794 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 7795 | { |
5378e607 | 7796 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 7797 | } |
f7039b1d LD |
7798 | |
7799 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
7800 | { | |
7801 | struct btrfs_fs_info *fs_info = root->fs_info; | |
7802 | struct btrfs_block_group_cache *cache = NULL; | |
7803 | u64 group_trimmed; | |
7804 | u64 start; | |
7805 | u64 end; | |
7806 | u64 trimmed = 0; | |
2cac13e4 | 7807 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
7808 | int ret = 0; |
7809 | ||
2cac13e4 LB |
7810 | /* |
7811 | * try to trim all FS space, our block group may start from non-zero. | |
7812 | */ | |
7813 | if (range->len == total_bytes) | |
7814 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
7815 | else | |
7816 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
7817 | |
7818 | while (cache) { | |
7819 | if (cache->key.objectid >= (range->start + range->len)) { | |
7820 | btrfs_put_block_group(cache); | |
7821 | break; | |
7822 | } | |
7823 | ||
7824 | start = max(range->start, cache->key.objectid); | |
7825 | end = min(range->start + range->len, | |
7826 | cache->key.objectid + cache->key.offset); | |
7827 | ||
7828 | if (end - start >= range->minlen) { | |
7829 | if (!block_group_cache_done(cache)) { | |
7830 | ret = cache_block_group(cache, NULL, root, 0); | |
7831 | if (!ret) | |
7832 | wait_block_group_cache_done(cache); | |
7833 | } | |
7834 | ret = btrfs_trim_block_group(cache, | |
7835 | &group_trimmed, | |
7836 | start, | |
7837 | end, | |
7838 | range->minlen); | |
7839 | ||
7840 | trimmed += group_trimmed; | |
7841 | if (ret) { | |
7842 | btrfs_put_block_group(cache); | |
7843 | break; | |
7844 | } | |
7845 | } | |
7846 | ||
7847 | cache = next_block_group(fs_info->tree_root, cache); | |
7848 | } | |
7849 | ||
7850 | range->len = trimmed; | |
7851 | return ret; | |
7852 | } |