]>
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> |
f361bf4a | 19 | #include <linux/sched/signal.h> |
edbd8d4e | 20 | #include <linux/pagemap.h> |
ec44a35c | 21 | #include <linux/writeback.h> |
21af804c | 22 | #include <linux/blkdev.h> |
b7a9f29f | 23 | #include <linux/sort.h> |
4184ea7f | 24 | #include <linux/rcupdate.h> |
817d52f8 | 25 | #include <linux/kthread.h> |
5a0e3ad6 | 26 | #include <linux/slab.h> |
dff51cd1 | 27 | #include <linux/ratelimit.h> |
b150a4f1 | 28 | #include <linux/percpu_counter.h> |
74493f7a | 29 | #include "hash.h" |
995946dd | 30 | #include "tree-log.h" |
fec577fb CM |
31 | #include "disk-io.h" |
32 | #include "print-tree.h" | |
0b86a832 | 33 | #include "volumes.h" |
53b381b3 | 34 | #include "raid56.h" |
925baedd | 35 | #include "locking.h" |
fa9c0d79 | 36 | #include "free-space-cache.h" |
1e144fb8 | 37 | #include "free-space-tree.h" |
3fed40cc | 38 | #include "math.h" |
6ab0a202 | 39 | #include "sysfs.h" |
fcebe456 | 40 | #include "qgroup.h" |
fec577fb | 41 | |
709c0486 AJ |
42 | #undef SCRAMBLE_DELAYED_REFS |
43 | ||
9e622d6b MX |
44 | /* |
45 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
46 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
47 | * if we really need one. | |
48 | * | |
0e4f8f88 CM |
49 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
50 | * if we have very few chunks already allocated. This is | |
51 | * used as part of the clustering code to help make sure | |
52 | * we have a good pool of storage to cluster in, without | |
53 | * filling the FS with empty chunks | |
54 | * | |
9e622d6b MX |
55 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
56 | * | |
0e4f8f88 CM |
57 | */ |
58 | enum { | |
59 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
60 | CHUNK_ALLOC_LIMITED = 1, |
61 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
62 | }; |
63 | ||
ce93ec54 | 64 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 65 | struct btrfs_fs_info *fs_info, u64 bytenr, |
ce93ec54 | 66 | u64 num_bytes, int alloc); |
5d4f98a2 | 67 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 68 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 69 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
70 | u64 root_objectid, u64 owner_objectid, |
71 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 72 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
73 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
74 | struct extent_buffer *leaf, | |
75 | struct btrfs_extent_item *ei); | |
76 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
2ff7e61e | 77 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
78 | u64 parent, u64 root_objectid, |
79 | u64 flags, u64 owner, u64 offset, | |
80 | struct btrfs_key *ins, int ref_mod); | |
81 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
2ff7e61e | 82 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
83 | u64 parent, u64 root_objectid, |
84 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 85 | int level, struct btrfs_key *ins); |
6a63209f | 86 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 87 | struct btrfs_fs_info *fs_info, u64 flags, |
698d0082 | 88 | int force); |
11833d66 YZ |
89 | static int find_next_key(struct btrfs_path *path, int level, |
90 | struct btrfs_key *key); | |
ab8d0fc4 JM |
91 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
92 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 93 | int dump_block_groups); |
4824f1f4 | 94 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 95 | u64 ram_bytes, u64 num_bytes, int delalloc); |
4824f1f4 WX |
96 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
97 | u64 num_bytes, int delalloc); | |
5d80366e JB |
98 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
99 | u64 num_bytes); | |
957780eb JB |
100 | static int __reserve_metadata_bytes(struct btrfs_root *root, |
101 | struct btrfs_space_info *space_info, | |
102 | u64 orig_bytes, | |
103 | enum btrfs_reserve_flush_enum flush); | |
104 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
105 | struct btrfs_space_info *space_info, | |
106 | u64 num_bytes); | |
107 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
108 | struct btrfs_space_info *space_info, | |
109 | u64 num_bytes); | |
6a63209f | 110 | |
817d52f8 JB |
111 | static noinline int |
112 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
113 | { | |
114 | smp_mb(); | |
36cce922 JB |
115 | return cache->cached == BTRFS_CACHE_FINISHED || |
116 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
117 | } |
118 | ||
0f9dd46c JB |
119 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
120 | { | |
121 | return (cache->flags & bits) == bits; | |
122 | } | |
123 | ||
758f2dfc | 124 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
125 | { |
126 | atomic_inc(&cache->count); | |
127 | } | |
128 | ||
129 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
130 | { | |
f0486c68 YZ |
131 | if (atomic_dec_and_test(&cache->count)) { |
132 | WARN_ON(cache->pinned > 0); | |
133 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 134 | kfree(cache->free_space_ctl); |
11dfe35a | 135 | kfree(cache); |
f0486c68 | 136 | } |
11dfe35a JB |
137 | } |
138 | ||
0f9dd46c JB |
139 | /* |
140 | * this adds the block group to the fs_info rb tree for the block group | |
141 | * cache | |
142 | */ | |
b2950863 | 143 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
144 | struct btrfs_block_group_cache *block_group) |
145 | { | |
146 | struct rb_node **p; | |
147 | struct rb_node *parent = NULL; | |
148 | struct btrfs_block_group_cache *cache; | |
149 | ||
150 | spin_lock(&info->block_group_cache_lock); | |
151 | p = &info->block_group_cache_tree.rb_node; | |
152 | ||
153 | while (*p) { | |
154 | parent = *p; | |
155 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
156 | cache_node); | |
157 | if (block_group->key.objectid < cache->key.objectid) { | |
158 | p = &(*p)->rb_left; | |
159 | } else if (block_group->key.objectid > cache->key.objectid) { | |
160 | p = &(*p)->rb_right; | |
161 | } else { | |
162 | spin_unlock(&info->block_group_cache_lock); | |
163 | return -EEXIST; | |
164 | } | |
165 | } | |
166 | ||
167 | rb_link_node(&block_group->cache_node, parent, p); | |
168 | rb_insert_color(&block_group->cache_node, | |
169 | &info->block_group_cache_tree); | |
a1897fdd LB |
170 | |
171 | if (info->first_logical_byte > block_group->key.objectid) | |
172 | info->first_logical_byte = block_group->key.objectid; | |
173 | ||
0f9dd46c JB |
174 | spin_unlock(&info->block_group_cache_lock); |
175 | ||
176 | return 0; | |
177 | } | |
178 | ||
179 | /* | |
180 | * This will return the block group at or after bytenr if contains is 0, else | |
181 | * it will return the block group that contains the bytenr | |
182 | */ | |
183 | static struct btrfs_block_group_cache * | |
184 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
185 | int contains) | |
186 | { | |
187 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
188 | struct rb_node *n; | |
189 | u64 end, start; | |
190 | ||
191 | spin_lock(&info->block_group_cache_lock); | |
192 | n = info->block_group_cache_tree.rb_node; | |
193 | ||
194 | while (n) { | |
195 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
196 | cache_node); | |
197 | end = cache->key.objectid + cache->key.offset - 1; | |
198 | start = cache->key.objectid; | |
199 | ||
200 | if (bytenr < start) { | |
201 | if (!contains && (!ret || start < ret->key.objectid)) | |
202 | ret = cache; | |
203 | n = n->rb_left; | |
204 | } else if (bytenr > start) { | |
205 | if (contains && bytenr <= end) { | |
206 | ret = cache; | |
207 | break; | |
208 | } | |
209 | n = n->rb_right; | |
210 | } else { | |
211 | ret = cache; | |
212 | break; | |
213 | } | |
214 | } | |
a1897fdd | 215 | if (ret) { |
11dfe35a | 216 | btrfs_get_block_group(ret); |
a1897fdd LB |
217 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
218 | info->first_logical_byte = ret->key.objectid; | |
219 | } | |
0f9dd46c JB |
220 | spin_unlock(&info->block_group_cache_lock); |
221 | ||
222 | return ret; | |
223 | } | |
224 | ||
2ff7e61e | 225 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 226 | u64 start, u64 num_bytes) |
817d52f8 | 227 | { |
11833d66 | 228 | u64 end = start + num_bytes - 1; |
0b246afa | 229 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 230 | start, end, EXTENT_UPTODATE); |
0b246afa | 231 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 232 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
233 | return 0; |
234 | } | |
817d52f8 | 235 | |
2ff7e61e | 236 | static void free_excluded_extents(struct btrfs_fs_info *fs_info, |
11833d66 YZ |
237 | struct btrfs_block_group_cache *cache) |
238 | { | |
239 | u64 start, end; | |
817d52f8 | 240 | |
11833d66 YZ |
241 | start = cache->key.objectid; |
242 | end = start + cache->key.offset - 1; | |
243 | ||
0b246afa | 244 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 245 | start, end, EXTENT_UPTODATE); |
0b246afa | 246 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 247 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
248 | } |
249 | ||
2ff7e61e | 250 | static int exclude_super_stripes(struct btrfs_fs_info *fs_info, |
11833d66 | 251 | struct btrfs_block_group_cache *cache) |
817d52f8 | 252 | { |
817d52f8 JB |
253 | u64 bytenr; |
254 | u64 *logical; | |
255 | int stripe_len; | |
256 | int i, nr, ret; | |
257 | ||
06b2331f YZ |
258 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
259 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
260 | cache->bytes_super += stripe_len; | |
2ff7e61e | 261 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 262 | stripe_len); |
835d974f JB |
263 | if (ret) |
264 | return ret; | |
06b2331f YZ |
265 | } |
266 | ||
817d52f8 JB |
267 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
268 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 269 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
ab8d0fc4 | 270 | bytenr, 0, &logical, &nr, &stripe_len); |
835d974f JB |
271 | if (ret) |
272 | return ret; | |
11833d66 | 273 | |
817d52f8 | 274 | while (nr--) { |
51bf5f0b JB |
275 | u64 start, len; |
276 | ||
277 | if (logical[nr] > cache->key.objectid + | |
278 | cache->key.offset) | |
279 | continue; | |
280 | ||
281 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
282 | continue; | |
283 | ||
284 | start = logical[nr]; | |
285 | if (start < cache->key.objectid) { | |
286 | start = cache->key.objectid; | |
287 | len = (logical[nr] + stripe_len) - start; | |
288 | } else { | |
289 | len = min_t(u64, stripe_len, | |
290 | cache->key.objectid + | |
291 | cache->key.offset - start); | |
292 | } | |
293 | ||
294 | cache->bytes_super += len; | |
2ff7e61e | 295 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
296 | if (ret) { |
297 | kfree(logical); | |
298 | return ret; | |
299 | } | |
817d52f8 | 300 | } |
11833d66 | 301 | |
817d52f8 JB |
302 | kfree(logical); |
303 | } | |
817d52f8 JB |
304 | return 0; |
305 | } | |
306 | ||
11833d66 YZ |
307 | static struct btrfs_caching_control * |
308 | get_caching_control(struct btrfs_block_group_cache *cache) | |
309 | { | |
310 | struct btrfs_caching_control *ctl; | |
311 | ||
312 | spin_lock(&cache->lock); | |
dde5abee JB |
313 | if (!cache->caching_ctl) { |
314 | spin_unlock(&cache->lock); | |
11833d66 YZ |
315 | return NULL; |
316 | } | |
317 | ||
318 | ctl = cache->caching_ctl; | |
319 | atomic_inc(&ctl->count); | |
320 | spin_unlock(&cache->lock); | |
321 | return ctl; | |
322 | } | |
323 | ||
324 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
325 | { | |
326 | if (atomic_dec_and_test(&ctl->count)) | |
327 | kfree(ctl); | |
328 | } | |
329 | ||
d0bd4560 | 330 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 331 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 332 | { |
2ff7e61e | 333 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
334 | u64 start = block_group->key.objectid; |
335 | u64 len = block_group->key.offset; | |
336 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 337 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
338 | u64 step = chunk << 1; |
339 | ||
340 | while (len > chunk) { | |
341 | btrfs_remove_free_space(block_group, start, chunk); | |
342 | start += step; | |
343 | if (len < step) | |
344 | len = 0; | |
345 | else | |
346 | len -= step; | |
347 | } | |
348 | } | |
349 | #endif | |
350 | ||
0f9dd46c JB |
351 | /* |
352 | * this is only called by cache_block_group, since we could have freed extents | |
353 | * we need to check the pinned_extents for any extents that can't be used yet | |
354 | * since their free space will be released as soon as the transaction commits. | |
355 | */ | |
a5ed9182 OS |
356 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
357 | struct btrfs_fs_info *info, u64 start, u64 end) | |
0f9dd46c | 358 | { |
817d52f8 | 359 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
360 | int ret; |
361 | ||
362 | while (start < end) { | |
11833d66 | 363 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 364 | &extent_start, &extent_end, |
e6138876 JB |
365 | EXTENT_DIRTY | EXTENT_UPTODATE, |
366 | NULL); | |
0f9dd46c JB |
367 | if (ret) |
368 | break; | |
369 | ||
06b2331f | 370 | if (extent_start <= start) { |
0f9dd46c JB |
371 | start = extent_end + 1; |
372 | } else if (extent_start > start && extent_start < end) { | |
373 | size = extent_start - start; | |
817d52f8 | 374 | total_added += size; |
ea6a478e JB |
375 | ret = btrfs_add_free_space(block_group, start, |
376 | size); | |
79787eaa | 377 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
378 | start = extent_end + 1; |
379 | } else { | |
380 | break; | |
381 | } | |
382 | } | |
383 | ||
384 | if (start < end) { | |
385 | size = end - start; | |
817d52f8 | 386 | total_added += size; |
ea6a478e | 387 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 388 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
389 | } |
390 | ||
817d52f8 | 391 | return total_added; |
0f9dd46c JB |
392 | } |
393 | ||
73fa48b6 | 394 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 395 | { |
0b246afa JM |
396 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
397 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
398 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 399 | struct btrfs_path *path; |
5f39d397 | 400 | struct extent_buffer *leaf; |
11833d66 | 401 | struct btrfs_key key; |
817d52f8 | 402 | u64 total_found = 0; |
11833d66 YZ |
403 | u64 last = 0; |
404 | u32 nritems; | |
73fa48b6 | 405 | int ret; |
d0bd4560 | 406 | bool wakeup = true; |
f510cfec | 407 | |
e37c9e69 CM |
408 | path = btrfs_alloc_path(); |
409 | if (!path) | |
73fa48b6 | 410 | return -ENOMEM; |
7d7d6068 | 411 | |
817d52f8 | 412 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 413 | |
d0bd4560 JB |
414 | #ifdef CONFIG_BTRFS_DEBUG |
415 | /* | |
416 | * If we're fragmenting we don't want to make anybody think we can | |
417 | * allocate from this block group until we've had a chance to fragment | |
418 | * the free space. | |
419 | */ | |
2ff7e61e | 420 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
421 | wakeup = false; |
422 | #endif | |
5cd57b2c | 423 | /* |
817d52f8 JB |
424 | * We don't want to deadlock with somebody trying to allocate a new |
425 | * extent for the extent root while also trying to search the extent | |
426 | * root to add free space. So we skip locking and search the commit | |
427 | * root, since its read-only | |
5cd57b2c CM |
428 | */ |
429 | path->skip_locking = 1; | |
817d52f8 | 430 | path->search_commit_root = 1; |
e4058b54 | 431 | path->reada = READA_FORWARD; |
817d52f8 | 432 | |
e4404d6e | 433 | key.objectid = last; |
e37c9e69 | 434 | key.offset = 0; |
11833d66 | 435 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 436 | |
52ee28d2 | 437 | next: |
11833d66 | 438 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 439 | if (ret < 0) |
73fa48b6 | 440 | goto out; |
a512bbf8 | 441 | |
11833d66 YZ |
442 | leaf = path->nodes[0]; |
443 | nritems = btrfs_header_nritems(leaf); | |
444 | ||
d397712b | 445 | while (1) { |
7841cb28 | 446 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 447 | last = (u64)-1; |
817d52f8 | 448 | break; |
f25784b3 | 449 | } |
817d52f8 | 450 | |
11833d66 YZ |
451 | if (path->slots[0] < nritems) { |
452 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
453 | } else { | |
454 | ret = find_next_key(path, 0, &key); | |
455 | if (ret) | |
e37c9e69 | 456 | break; |
817d52f8 | 457 | |
c9ea7b24 | 458 | if (need_resched() || |
9e351cc8 | 459 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
460 | if (wakeup) |
461 | caching_ctl->progress = last; | |
ff5714cc | 462 | btrfs_release_path(path); |
9e351cc8 | 463 | up_read(&fs_info->commit_root_sem); |
589d8ade | 464 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 465 | cond_resched(); |
73fa48b6 OS |
466 | mutex_lock(&caching_ctl->mutex); |
467 | down_read(&fs_info->commit_root_sem); | |
468 | goto next; | |
589d8ade | 469 | } |
0a3896d0 JB |
470 | |
471 | ret = btrfs_next_leaf(extent_root, path); | |
472 | if (ret < 0) | |
73fa48b6 | 473 | goto out; |
0a3896d0 JB |
474 | if (ret) |
475 | break; | |
589d8ade JB |
476 | leaf = path->nodes[0]; |
477 | nritems = btrfs_header_nritems(leaf); | |
478 | continue; | |
11833d66 | 479 | } |
817d52f8 | 480 | |
52ee28d2 LB |
481 | if (key.objectid < last) { |
482 | key.objectid = last; | |
483 | key.offset = 0; | |
484 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
485 | ||
d0bd4560 JB |
486 | if (wakeup) |
487 | caching_ctl->progress = last; | |
52ee28d2 LB |
488 | btrfs_release_path(path); |
489 | goto next; | |
490 | } | |
491 | ||
11833d66 YZ |
492 | if (key.objectid < block_group->key.objectid) { |
493 | path->slots[0]++; | |
817d52f8 | 494 | continue; |
e37c9e69 | 495 | } |
0f9dd46c | 496 | |
e37c9e69 | 497 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 498 | block_group->key.offset) |
e37c9e69 | 499 | break; |
7d7d6068 | 500 | |
3173a18f JB |
501 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
502 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
503 | total_found += add_new_free_space(block_group, |
504 | fs_info, last, | |
505 | key.objectid); | |
3173a18f JB |
506 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
507 | last = key.objectid + | |
da17066c | 508 | fs_info->nodesize; |
3173a18f JB |
509 | else |
510 | last = key.objectid + key.offset; | |
817d52f8 | 511 | |
73fa48b6 | 512 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 513 | total_found = 0; |
d0bd4560 JB |
514 | if (wakeup) |
515 | wake_up(&caching_ctl->wait); | |
11833d66 | 516 | } |
817d52f8 | 517 | } |
e37c9e69 CM |
518 | path->slots[0]++; |
519 | } | |
817d52f8 | 520 | ret = 0; |
e37c9e69 | 521 | |
817d52f8 JB |
522 | total_found += add_new_free_space(block_group, fs_info, last, |
523 | block_group->key.objectid + | |
524 | block_group->key.offset); | |
11833d66 | 525 | caching_ctl->progress = (u64)-1; |
817d52f8 | 526 | |
73fa48b6 OS |
527 | out: |
528 | btrfs_free_path(path); | |
529 | return ret; | |
530 | } | |
531 | ||
532 | static noinline void caching_thread(struct btrfs_work *work) | |
533 | { | |
534 | struct btrfs_block_group_cache *block_group; | |
535 | struct btrfs_fs_info *fs_info; | |
536 | struct btrfs_caching_control *caching_ctl; | |
b4570aa9 | 537 | struct btrfs_root *extent_root; |
73fa48b6 OS |
538 | int ret; |
539 | ||
540 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
541 | block_group = caching_ctl->block_group; | |
542 | fs_info = block_group->fs_info; | |
b4570aa9 | 543 | extent_root = fs_info->extent_root; |
73fa48b6 OS |
544 | |
545 | mutex_lock(&caching_ctl->mutex); | |
546 | down_read(&fs_info->commit_root_sem); | |
547 | ||
1e144fb8 OS |
548 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
549 | ret = load_free_space_tree(caching_ctl); | |
550 | else | |
551 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 552 | |
817d52f8 | 553 | spin_lock(&block_group->lock); |
11833d66 | 554 | block_group->caching_ctl = NULL; |
73fa48b6 | 555 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 556 | spin_unlock(&block_group->lock); |
0f9dd46c | 557 | |
d0bd4560 | 558 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 559 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
560 | u64 bytes_used; |
561 | ||
562 | spin_lock(&block_group->space_info->lock); | |
563 | spin_lock(&block_group->lock); | |
564 | bytes_used = block_group->key.offset - | |
565 | btrfs_block_group_used(&block_group->item); | |
566 | block_group->space_info->bytes_used += bytes_used >> 1; | |
567 | spin_unlock(&block_group->lock); | |
568 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 569 | fragment_free_space(block_group); |
d0bd4560 JB |
570 | } |
571 | #endif | |
572 | ||
573 | caching_ctl->progress = (u64)-1; | |
11833d66 | 574 | |
9e351cc8 | 575 | up_read(&fs_info->commit_root_sem); |
2ff7e61e | 576 | free_excluded_extents(fs_info, block_group); |
11833d66 | 577 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 578 | |
11833d66 YZ |
579 | wake_up(&caching_ctl->wait); |
580 | ||
581 | put_caching_control(caching_ctl); | |
11dfe35a | 582 | btrfs_put_block_group(block_group); |
817d52f8 JB |
583 | } |
584 | ||
9d66e233 | 585 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 586 | int load_cache_only) |
817d52f8 | 587 | { |
291c7d2f | 588 | DEFINE_WAIT(wait); |
11833d66 YZ |
589 | struct btrfs_fs_info *fs_info = cache->fs_info; |
590 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
591 | int ret = 0; |
592 | ||
291c7d2f | 593 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
594 | if (!caching_ctl) |
595 | return -ENOMEM; | |
291c7d2f JB |
596 | |
597 | INIT_LIST_HEAD(&caching_ctl->list); | |
598 | mutex_init(&caching_ctl->mutex); | |
599 | init_waitqueue_head(&caching_ctl->wait); | |
600 | caching_ctl->block_group = cache; | |
601 | caching_ctl->progress = cache->key.objectid; | |
602 | atomic_set(&caching_ctl->count, 1); | |
9e0af237 LB |
603 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
604 | caching_thread, NULL, NULL); | |
291c7d2f JB |
605 | |
606 | spin_lock(&cache->lock); | |
607 | /* | |
608 | * This should be a rare occasion, but this could happen I think in the | |
609 | * case where one thread starts to load the space cache info, and then | |
610 | * some other thread starts a transaction commit which tries to do an | |
611 | * allocation while the other thread is still loading the space cache | |
612 | * info. The previous loop should have kept us from choosing this block | |
613 | * group, but if we've moved to the state where we will wait on caching | |
614 | * block groups we need to first check if we're doing a fast load here, | |
615 | * so we can wait for it to finish, otherwise we could end up allocating | |
616 | * from a block group who's cache gets evicted for one reason or | |
617 | * another. | |
618 | */ | |
619 | while (cache->cached == BTRFS_CACHE_FAST) { | |
620 | struct btrfs_caching_control *ctl; | |
621 | ||
622 | ctl = cache->caching_ctl; | |
623 | atomic_inc(&ctl->count); | |
624 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
625 | spin_unlock(&cache->lock); | |
626 | ||
627 | schedule(); | |
628 | ||
629 | finish_wait(&ctl->wait, &wait); | |
630 | put_caching_control(ctl); | |
631 | spin_lock(&cache->lock); | |
632 | } | |
633 | ||
634 | if (cache->cached != BTRFS_CACHE_NO) { | |
635 | spin_unlock(&cache->lock); | |
636 | kfree(caching_ctl); | |
11833d66 | 637 | return 0; |
291c7d2f JB |
638 | } |
639 | WARN_ON(cache->caching_ctl); | |
640 | cache->caching_ctl = caching_ctl; | |
641 | cache->cached = BTRFS_CACHE_FAST; | |
642 | spin_unlock(&cache->lock); | |
11833d66 | 643 | |
d53ba474 | 644 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
cb83b7b8 | 645 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
646 | ret = load_free_space_cache(fs_info, cache); |
647 | ||
648 | spin_lock(&cache->lock); | |
649 | if (ret == 1) { | |
291c7d2f | 650 | cache->caching_ctl = NULL; |
9d66e233 JB |
651 | cache->cached = BTRFS_CACHE_FINISHED; |
652 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 653 | caching_ctl->progress = (u64)-1; |
9d66e233 | 654 | } else { |
291c7d2f JB |
655 | if (load_cache_only) { |
656 | cache->caching_ctl = NULL; | |
657 | cache->cached = BTRFS_CACHE_NO; | |
658 | } else { | |
659 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 660 | cache->has_caching_ctl = 1; |
291c7d2f | 661 | } |
9d66e233 JB |
662 | } |
663 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
664 | #ifdef CONFIG_BTRFS_DEBUG |
665 | if (ret == 1 && | |
2ff7e61e | 666 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
667 | u64 bytes_used; |
668 | ||
669 | spin_lock(&cache->space_info->lock); | |
670 | spin_lock(&cache->lock); | |
671 | bytes_used = cache->key.offset - | |
672 | btrfs_block_group_used(&cache->item); | |
673 | cache->space_info->bytes_used += bytes_used >> 1; | |
674 | spin_unlock(&cache->lock); | |
675 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 676 | fragment_free_space(cache); |
d0bd4560 JB |
677 | } |
678 | #endif | |
cb83b7b8 JB |
679 | mutex_unlock(&caching_ctl->mutex); |
680 | ||
291c7d2f | 681 | wake_up(&caching_ctl->wait); |
3c14874a | 682 | if (ret == 1) { |
291c7d2f | 683 | put_caching_control(caching_ctl); |
2ff7e61e | 684 | free_excluded_extents(fs_info, cache); |
9d66e233 | 685 | return 0; |
3c14874a | 686 | } |
291c7d2f JB |
687 | } else { |
688 | /* | |
1e144fb8 OS |
689 | * We're either using the free space tree or no caching at all. |
690 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
691 | */ |
692 | spin_lock(&cache->lock); | |
693 | if (load_cache_only) { | |
694 | cache->caching_ctl = NULL; | |
695 | cache->cached = BTRFS_CACHE_NO; | |
696 | } else { | |
697 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 698 | cache->has_caching_ctl = 1; |
291c7d2f JB |
699 | } |
700 | spin_unlock(&cache->lock); | |
701 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
702 | } |
703 | ||
291c7d2f JB |
704 | if (load_cache_only) { |
705 | put_caching_control(caching_ctl); | |
11833d66 | 706 | return 0; |
817d52f8 | 707 | } |
817d52f8 | 708 | |
9e351cc8 | 709 | down_write(&fs_info->commit_root_sem); |
291c7d2f | 710 | atomic_inc(&caching_ctl->count); |
11833d66 | 711 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 712 | up_write(&fs_info->commit_root_sem); |
11833d66 | 713 | |
11dfe35a | 714 | btrfs_get_block_group(cache); |
11833d66 | 715 | |
e66f0bb1 | 716 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 717 | |
ef8bbdfe | 718 | return ret; |
e37c9e69 CM |
719 | } |
720 | ||
0f9dd46c JB |
721 | /* |
722 | * return the block group that starts at or after bytenr | |
723 | */ | |
d397712b CM |
724 | static struct btrfs_block_group_cache * |
725 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 726 | { |
e2c89907 | 727 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
728 | } |
729 | ||
0f9dd46c | 730 | /* |
9f55684c | 731 | * return the block group that contains the given bytenr |
0f9dd46c | 732 | */ |
d397712b CM |
733 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
734 | struct btrfs_fs_info *info, | |
735 | u64 bytenr) | |
be744175 | 736 | { |
e2c89907 | 737 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 738 | } |
0b86a832 | 739 | |
0f9dd46c JB |
740 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
741 | u64 flags) | |
6324fbf3 | 742 | { |
0f9dd46c | 743 | struct list_head *head = &info->space_info; |
0f9dd46c | 744 | struct btrfs_space_info *found; |
4184ea7f | 745 | |
52ba6929 | 746 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 747 | |
4184ea7f CM |
748 | rcu_read_lock(); |
749 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 750 | if (found->flags & flags) { |
4184ea7f | 751 | rcu_read_unlock(); |
0f9dd46c | 752 | return found; |
4184ea7f | 753 | } |
0f9dd46c | 754 | } |
4184ea7f | 755 | rcu_read_unlock(); |
0f9dd46c | 756 | return NULL; |
6324fbf3 CM |
757 | } |
758 | ||
4184ea7f CM |
759 | /* |
760 | * after adding space to the filesystem, we need to clear the full flags | |
761 | * on all the space infos. | |
762 | */ | |
763 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
764 | { | |
765 | struct list_head *head = &info->space_info; | |
766 | struct btrfs_space_info *found; | |
767 | ||
768 | rcu_read_lock(); | |
769 | list_for_each_entry_rcu(found, head, list) | |
770 | found->full = 0; | |
771 | rcu_read_unlock(); | |
772 | } | |
773 | ||
1a4ed8fd | 774 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 775 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
776 | { |
777 | int ret; | |
778 | struct btrfs_key key; | |
31840ae1 | 779 | struct btrfs_path *path; |
e02119d5 | 780 | |
31840ae1 | 781 | path = btrfs_alloc_path(); |
d8926bb3 MF |
782 | if (!path) |
783 | return -ENOMEM; | |
784 | ||
e02119d5 CM |
785 | key.objectid = start; |
786 | key.offset = len; | |
3173a18f | 787 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 788 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 789 | btrfs_free_path(path); |
7bb86316 CM |
790 | return ret; |
791 | } | |
792 | ||
a22285a6 | 793 | /* |
3173a18f | 794 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
795 | * |
796 | * the head node for delayed ref is used to store the sum of all the | |
797 | * reference count modifications queued up in the rbtree. the head | |
798 | * node may also store the extent flags to set. This way you can check | |
799 | * to see what the reference count and extent flags would be if all of | |
800 | * the delayed refs are not processed. | |
801 | */ | |
802 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 803 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 804 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
805 | { |
806 | struct btrfs_delayed_ref_head *head; | |
807 | struct btrfs_delayed_ref_root *delayed_refs; | |
808 | struct btrfs_path *path; | |
809 | struct btrfs_extent_item *ei; | |
810 | struct extent_buffer *leaf; | |
811 | struct btrfs_key key; | |
812 | u32 item_size; | |
813 | u64 num_refs; | |
814 | u64 extent_flags; | |
815 | int ret; | |
816 | ||
3173a18f JB |
817 | /* |
818 | * If we don't have skinny metadata, don't bother doing anything | |
819 | * different | |
820 | */ | |
0b246afa JM |
821 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
822 | offset = fs_info->nodesize; | |
3173a18f JB |
823 | metadata = 0; |
824 | } | |
825 | ||
a22285a6 YZ |
826 | path = btrfs_alloc_path(); |
827 | if (!path) | |
828 | return -ENOMEM; | |
829 | ||
a22285a6 YZ |
830 | if (!trans) { |
831 | path->skip_locking = 1; | |
832 | path->search_commit_root = 1; | |
833 | } | |
639eefc8 FDBM |
834 | |
835 | search_again: | |
836 | key.objectid = bytenr; | |
837 | key.offset = offset; | |
838 | if (metadata) | |
839 | key.type = BTRFS_METADATA_ITEM_KEY; | |
840 | else | |
841 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
842 | ||
0b246afa | 843 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
844 | if (ret < 0) |
845 | goto out_free; | |
846 | ||
3173a18f | 847 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
848 | if (path->slots[0]) { |
849 | path->slots[0]--; | |
850 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
851 | path->slots[0]); | |
852 | if (key.objectid == bytenr && | |
853 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 854 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
855 | ret = 0; |
856 | } | |
3173a18f JB |
857 | } |
858 | ||
a22285a6 YZ |
859 | if (ret == 0) { |
860 | leaf = path->nodes[0]; | |
861 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
862 | if (item_size >= sizeof(*ei)) { | |
863 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
864 | struct btrfs_extent_item); | |
865 | num_refs = btrfs_extent_refs(leaf, ei); | |
866 | extent_flags = btrfs_extent_flags(leaf, ei); | |
867 | } else { | |
868 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
869 | struct btrfs_extent_item_v0 *ei0; | |
870 | BUG_ON(item_size != sizeof(*ei0)); | |
871 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
872 | struct btrfs_extent_item_v0); | |
873 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
874 | /* FIXME: this isn't correct for data */ | |
875 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
876 | #else | |
877 | BUG(); | |
878 | #endif | |
879 | } | |
880 | BUG_ON(num_refs == 0); | |
881 | } else { | |
882 | num_refs = 0; | |
883 | extent_flags = 0; | |
884 | ret = 0; | |
885 | } | |
886 | ||
887 | if (!trans) | |
888 | goto out; | |
889 | ||
890 | delayed_refs = &trans->transaction->delayed_refs; | |
891 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 892 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
893 | if (head) { |
894 | if (!mutex_trylock(&head->mutex)) { | |
895 | atomic_inc(&head->node.refs); | |
896 | spin_unlock(&delayed_refs->lock); | |
897 | ||
b3b4aa74 | 898 | btrfs_release_path(path); |
a22285a6 | 899 | |
8cc33e5c DS |
900 | /* |
901 | * Mutex was contended, block until it's released and try | |
902 | * again | |
903 | */ | |
a22285a6 YZ |
904 | mutex_lock(&head->mutex); |
905 | mutex_unlock(&head->mutex); | |
906 | btrfs_put_delayed_ref(&head->node); | |
639eefc8 | 907 | goto search_again; |
a22285a6 | 908 | } |
d7df2c79 | 909 | spin_lock(&head->lock); |
a22285a6 YZ |
910 | if (head->extent_op && head->extent_op->update_flags) |
911 | extent_flags |= head->extent_op->flags_to_set; | |
912 | else | |
913 | BUG_ON(num_refs == 0); | |
914 | ||
915 | num_refs += head->node.ref_mod; | |
d7df2c79 | 916 | spin_unlock(&head->lock); |
a22285a6 YZ |
917 | mutex_unlock(&head->mutex); |
918 | } | |
919 | spin_unlock(&delayed_refs->lock); | |
920 | out: | |
921 | WARN_ON(num_refs == 0); | |
922 | if (refs) | |
923 | *refs = num_refs; | |
924 | if (flags) | |
925 | *flags = extent_flags; | |
926 | out_free: | |
927 | btrfs_free_path(path); | |
928 | return ret; | |
929 | } | |
930 | ||
d8d5f3e1 CM |
931 | /* |
932 | * Back reference rules. Back refs have three main goals: | |
933 | * | |
934 | * 1) differentiate between all holders of references to an extent so that | |
935 | * when a reference is dropped we can make sure it was a valid reference | |
936 | * before freeing the extent. | |
937 | * | |
938 | * 2) Provide enough information to quickly find the holders of an extent | |
939 | * if we notice a given block is corrupted or bad. | |
940 | * | |
941 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
942 | * maintenance. This is actually the same as #2, but with a slightly | |
943 | * different use case. | |
944 | * | |
5d4f98a2 YZ |
945 | * There are two kinds of back refs. The implicit back refs is optimized |
946 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
947 | * back refs of this kind provide information about the block's owner tree | |
948 | * and the pointer's key. These information allow us to find the block by | |
949 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
950 | * referenced by their owner trees. The location of tree block is recorded | |
951 | * in the back refs. Actually the full back refs is generic, and can be | |
952 | * used in all cases the implicit back refs is used. The major shortcoming | |
953 | * of the full back refs is its overhead. Every time a tree block gets | |
954 | * COWed, we have to update back refs entry for all pointers in it. | |
955 | * | |
956 | * For a newly allocated tree block, we use implicit back refs for | |
957 | * pointers in it. This means most tree related operations only involve | |
958 | * implicit back refs. For a tree block created in old transaction, the | |
959 | * only way to drop a reference to it is COW it. So we can detect the | |
960 | * event that tree block loses its owner tree's reference and do the | |
961 | * back refs conversion. | |
962 | * | |
01327610 | 963 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
964 | * |
965 | * The reference count of the block is one and the tree is the block's | |
966 | * owner tree. Nothing to do in this case. | |
967 | * | |
968 | * The reference count of the block is one and the tree is not the | |
969 | * block's owner tree. In this case, full back refs is used for pointers | |
970 | * in the block. Remove these full back refs, add implicit back refs for | |
971 | * every pointers in the new block. | |
972 | * | |
973 | * The reference count of the block is greater than one and the tree is | |
974 | * the block's owner tree. In this case, implicit back refs is used for | |
975 | * pointers in the block. Add full back refs for every pointers in the | |
976 | * block, increase lower level extents' reference counts. The original | |
977 | * implicit back refs are entailed to the new block. | |
978 | * | |
979 | * The reference count of the block is greater than one and the tree is | |
980 | * not the block's owner tree. Add implicit back refs for every pointer in | |
981 | * the new block, increase lower level extents' reference count. | |
982 | * | |
983 | * Back Reference Key composing: | |
984 | * | |
985 | * The key objectid corresponds to the first byte in the extent, | |
986 | * The key type is used to differentiate between types of back refs. | |
987 | * There are different meanings of the key offset for different types | |
988 | * of back refs. | |
989 | * | |
d8d5f3e1 CM |
990 | * File extents can be referenced by: |
991 | * | |
992 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 993 | * - different files inside a single subvolume |
d8d5f3e1 CM |
994 | * - different offsets inside a file (bookend extents in file.c) |
995 | * | |
5d4f98a2 | 996 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
997 | * |
998 | * - Objectid of the subvolume root | |
d8d5f3e1 | 999 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1000 | * - original offset in the file |
1001 | * - how many bookend extents | |
d8d5f3e1 | 1002 | * |
5d4f98a2 YZ |
1003 | * The key offset for the implicit back refs is hash of the first |
1004 | * three fields. | |
d8d5f3e1 | 1005 | * |
5d4f98a2 | 1006 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1007 | * |
5d4f98a2 | 1008 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1009 | * |
5d4f98a2 YZ |
1010 | * The key offset for the implicit back refs is the first byte of |
1011 | * the tree leaf | |
d8d5f3e1 | 1012 | * |
5d4f98a2 YZ |
1013 | * When a file extent is allocated, The implicit back refs is used. |
1014 | * the fields are filled in: | |
d8d5f3e1 | 1015 | * |
5d4f98a2 | 1016 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1017 | * |
5d4f98a2 YZ |
1018 | * When a file extent is removed file truncation, we find the |
1019 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1020 | * |
5d4f98a2 | 1021 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1022 | * |
5d4f98a2 | 1023 | * Btree extents can be referenced by: |
d8d5f3e1 | 1024 | * |
5d4f98a2 | 1025 | * - Different subvolumes |
d8d5f3e1 | 1026 | * |
5d4f98a2 YZ |
1027 | * Both the implicit back refs and the full back refs for tree blocks |
1028 | * only consist of key. The key offset for the implicit back refs is | |
1029 | * objectid of block's owner tree. The key offset for the full back refs | |
1030 | * is the first byte of parent block. | |
d8d5f3e1 | 1031 | * |
5d4f98a2 YZ |
1032 | * When implicit back refs is used, information about the lowest key and |
1033 | * level of the tree block are required. These information are stored in | |
1034 | * tree block info structure. | |
d8d5f3e1 | 1035 | */ |
31840ae1 | 1036 | |
5d4f98a2 YZ |
1037 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1038 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
87bde3cd | 1039 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1040 | struct btrfs_path *path, |
1041 | u64 owner, u32 extra_size) | |
7bb86316 | 1042 | { |
87bde3cd | 1043 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1044 | struct btrfs_extent_item *item; |
1045 | struct btrfs_extent_item_v0 *ei0; | |
1046 | struct btrfs_extent_ref_v0 *ref0; | |
1047 | struct btrfs_tree_block_info *bi; | |
1048 | struct extent_buffer *leaf; | |
7bb86316 | 1049 | struct btrfs_key key; |
5d4f98a2 YZ |
1050 | struct btrfs_key found_key; |
1051 | u32 new_size = sizeof(*item); | |
1052 | u64 refs; | |
1053 | int ret; | |
1054 | ||
1055 | leaf = path->nodes[0]; | |
1056 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1057 | ||
1058 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1059 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1060 | struct btrfs_extent_item_v0); | |
1061 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1062 | ||
1063 | if (owner == (u64)-1) { | |
1064 | while (1) { | |
1065 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1066 | ret = btrfs_next_leaf(root, path); | |
1067 | if (ret < 0) | |
1068 | return ret; | |
79787eaa | 1069 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1070 | leaf = path->nodes[0]; |
1071 | } | |
1072 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1073 | path->slots[0]); | |
1074 | BUG_ON(key.objectid != found_key.objectid); | |
1075 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1076 | path->slots[0]++; | |
1077 | continue; | |
1078 | } | |
1079 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1080 | struct btrfs_extent_ref_v0); | |
1081 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1082 | break; | |
1083 | } | |
1084 | } | |
b3b4aa74 | 1085 | btrfs_release_path(path); |
5d4f98a2 YZ |
1086 | |
1087 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1088 | new_size += sizeof(*bi); | |
1089 | ||
1090 | new_size -= sizeof(*ei0); | |
1091 | ret = btrfs_search_slot(trans, root, &key, path, | |
1092 | new_size + extra_size, 1); | |
1093 | if (ret < 0) | |
1094 | return ret; | |
79787eaa | 1095 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1096 | |
87bde3cd | 1097 | btrfs_extend_item(fs_info, path, new_size); |
5d4f98a2 YZ |
1098 | |
1099 | leaf = path->nodes[0]; | |
1100 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1101 | btrfs_set_extent_refs(leaf, item, refs); | |
1102 | /* FIXME: get real generation */ | |
1103 | btrfs_set_extent_generation(leaf, item, 0); | |
1104 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1105 | btrfs_set_extent_flags(leaf, item, | |
1106 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1107 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1108 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1109 | /* FIXME: get first key of the block */ | |
b159fa28 | 1110 | memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi)); |
5d4f98a2 YZ |
1111 | btrfs_set_tree_block_level(leaf, bi, (int)owner); |
1112 | } else { | |
1113 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1114 | } | |
1115 | btrfs_mark_buffer_dirty(leaf); | |
1116 | return 0; | |
1117 | } | |
1118 | #endif | |
1119 | ||
1120 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1121 | { | |
1122 | u32 high_crc = ~(u32)0; | |
1123 | u32 low_crc = ~(u32)0; | |
1124 | __le64 lenum; | |
1125 | ||
1126 | lenum = cpu_to_le64(root_objectid); | |
14a958e6 | 1127 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1128 | lenum = cpu_to_le64(owner); |
14a958e6 | 1129 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1130 | lenum = cpu_to_le64(offset); |
14a958e6 | 1131 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1132 | |
1133 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1134 | } | |
1135 | ||
1136 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1137 | struct btrfs_extent_data_ref *ref) | |
1138 | { | |
1139 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1140 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1141 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1142 | } | |
1143 | ||
1144 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1145 | struct btrfs_extent_data_ref *ref, | |
1146 | u64 root_objectid, u64 owner, u64 offset) | |
1147 | { | |
1148 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1149 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1150 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1151 | return 0; | |
1152 | return 1; | |
1153 | } | |
1154 | ||
1155 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1156 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1157 | struct btrfs_path *path, |
1158 | u64 bytenr, u64 parent, | |
1159 | u64 root_objectid, | |
1160 | u64 owner, u64 offset) | |
1161 | { | |
87bde3cd | 1162 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1163 | struct btrfs_key key; |
1164 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1165 | struct extent_buffer *leaf; |
5d4f98a2 | 1166 | u32 nritems; |
74493f7a | 1167 | int ret; |
5d4f98a2 YZ |
1168 | int recow; |
1169 | int err = -ENOENT; | |
74493f7a | 1170 | |
31840ae1 | 1171 | key.objectid = bytenr; |
5d4f98a2 YZ |
1172 | if (parent) { |
1173 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1174 | key.offset = parent; | |
1175 | } else { | |
1176 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1177 | key.offset = hash_extent_data_ref(root_objectid, | |
1178 | owner, offset); | |
1179 | } | |
1180 | again: | |
1181 | recow = 0; | |
1182 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1183 | if (ret < 0) { | |
1184 | err = ret; | |
1185 | goto fail; | |
1186 | } | |
31840ae1 | 1187 | |
5d4f98a2 YZ |
1188 | if (parent) { |
1189 | if (!ret) | |
1190 | return 0; | |
1191 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1192 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1193 | btrfs_release_path(path); |
5d4f98a2 YZ |
1194 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1195 | if (ret < 0) { | |
1196 | err = ret; | |
1197 | goto fail; | |
1198 | } | |
1199 | if (!ret) | |
1200 | return 0; | |
1201 | #endif | |
1202 | goto fail; | |
31840ae1 ZY |
1203 | } |
1204 | ||
1205 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1206 | nritems = btrfs_header_nritems(leaf); |
1207 | while (1) { | |
1208 | if (path->slots[0] >= nritems) { | |
1209 | ret = btrfs_next_leaf(root, path); | |
1210 | if (ret < 0) | |
1211 | err = ret; | |
1212 | if (ret) | |
1213 | goto fail; | |
1214 | ||
1215 | leaf = path->nodes[0]; | |
1216 | nritems = btrfs_header_nritems(leaf); | |
1217 | recow = 1; | |
1218 | } | |
1219 | ||
1220 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1221 | if (key.objectid != bytenr || | |
1222 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1223 | goto fail; | |
1224 | ||
1225 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1226 | struct btrfs_extent_data_ref); | |
1227 | ||
1228 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1229 | owner, offset)) { | |
1230 | if (recow) { | |
b3b4aa74 | 1231 | btrfs_release_path(path); |
5d4f98a2 YZ |
1232 | goto again; |
1233 | } | |
1234 | err = 0; | |
1235 | break; | |
1236 | } | |
1237 | path->slots[0]++; | |
31840ae1 | 1238 | } |
5d4f98a2 YZ |
1239 | fail: |
1240 | return err; | |
31840ae1 ZY |
1241 | } |
1242 | ||
5d4f98a2 | 1243 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1244 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1245 | struct btrfs_path *path, |
1246 | u64 bytenr, u64 parent, | |
1247 | u64 root_objectid, u64 owner, | |
1248 | u64 offset, int refs_to_add) | |
31840ae1 | 1249 | { |
87bde3cd | 1250 | struct btrfs_root *root = fs_info->extent_root; |
31840ae1 ZY |
1251 | struct btrfs_key key; |
1252 | struct extent_buffer *leaf; | |
5d4f98a2 | 1253 | u32 size; |
31840ae1 ZY |
1254 | u32 num_refs; |
1255 | int ret; | |
74493f7a | 1256 | |
74493f7a | 1257 | key.objectid = bytenr; |
5d4f98a2 YZ |
1258 | if (parent) { |
1259 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1260 | key.offset = parent; | |
1261 | size = sizeof(struct btrfs_shared_data_ref); | |
1262 | } else { | |
1263 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1264 | key.offset = hash_extent_data_ref(root_objectid, | |
1265 | owner, offset); | |
1266 | size = sizeof(struct btrfs_extent_data_ref); | |
1267 | } | |
74493f7a | 1268 | |
5d4f98a2 YZ |
1269 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1270 | if (ret && ret != -EEXIST) | |
1271 | goto fail; | |
1272 | ||
1273 | leaf = path->nodes[0]; | |
1274 | if (parent) { | |
1275 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1276 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1277 | struct btrfs_shared_data_ref); |
1278 | if (ret == 0) { | |
1279 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1280 | } else { | |
1281 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1282 | num_refs += refs_to_add; | |
1283 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1284 | } |
5d4f98a2 YZ |
1285 | } else { |
1286 | struct btrfs_extent_data_ref *ref; | |
1287 | while (ret == -EEXIST) { | |
1288 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1289 | struct btrfs_extent_data_ref); | |
1290 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1291 | owner, offset)) | |
1292 | break; | |
b3b4aa74 | 1293 | btrfs_release_path(path); |
5d4f98a2 YZ |
1294 | key.offset++; |
1295 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1296 | size); | |
1297 | if (ret && ret != -EEXIST) | |
1298 | goto fail; | |
31840ae1 | 1299 | |
5d4f98a2 YZ |
1300 | leaf = path->nodes[0]; |
1301 | } | |
1302 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1303 | struct btrfs_extent_data_ref); | |
1304 | if (ret == 0) { | |
1305 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1306 | root_objectid); | |
1307 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1308 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1309 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1310 | } else { | |
1311 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1312 | num_refs += refs_to_add; | |
1313 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1314 | } |
31840ae1 | 1315 | } |
5d4f98a2 YZ |
1316 | btrfs_mark_buffer_dirty(leaf); |
1317 | ret = 0; | |
1318 | fail: | |
b3b4aa74 | 1319 | btrfs_release_path(path); |
7bb86316 | 1320 | return ret; |
74493f7a CM |
1321 | } |
1322 | ||
5d4f98a2 | 1323 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1324 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 1325 | struct btrfs_path *path, |
fcebe456 | 1326 | int refs_to_drop, int *last_ref) |
31840ae1 | 1327 | { |
5d4f98a2 YZ |
1328 | struct btrfs_key key; |
1329 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1330 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1331 | struct extent_buffer *leaf; |
5d4f98a2 | 1332 | u32 num_refs = 0; |
31840ae1 ZY |
1333 | int ret = 0; |
1334 | ||
1335 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1336 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1337 | ||
1338 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1339 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1340 | struct btrfs_extent_data_ref); | |
1341 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1342 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1343 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1344 | struct btrfs_shared_data_ref); | |
1345 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1346 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1347 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1348 | struct btrfs_extent_ref_v0 *ref0; | |
1349 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1350 | struct btrfs_extent_ref_v0); | |
1351 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1352 | #endif | |
1353 | } else { | |
1354 | BUG(); | |
1355 | } | |
1356 | ||
56bec294 CM |
1357 | BUG_ON(num_refs < refs_to_drop); |
1358 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1359 | |
31840ae1 | 1360 | if (num_refs == 0) { |
87bde3cd | 1361 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
fcebe456 | 1362 | *last_ref = 1; |
31840ae1 | 1363 | } else { |
5d4f98a2 YZ |
1364 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1365 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1366 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1367 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1368 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1369 | else { | |
1370 | struct btrfs_extent_ref_v0 *ref0; | |
1371 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1372 | struct btrfs_extent_ref_v0); | |
1373 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1374 | } | |
1375 | #endif | |
31840ae1 ZY |
1376 | btrfs_mark_buffer_dirty(leaf); |
1377 | } | |
31840ae1 ZY |
1378 | return ret; |
1379 | } | |
1380 | ||
9ed0dea0 | 1381 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1382 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1383 | { |
5d4f98a2 YZ |
1384 | struct btrfs_key key; |
1385 | struct extent_buffer *leaf; | |
1386 | struct btrfs_extent_data_ref *ref1; | |
1387 | struct btrfs_shared_data_ref *ref2; | |
1388 | u32 num_refs = 0; | |
1389 | ||
1390 | leaf = path->nodes[0]; | |
1391 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1392 | if (iref) { | |
1393 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1394 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1395 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1396 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1397 | } else { | |
1398 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1399 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1400 | } | |
1401 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1402 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1403 | struct btrfs_extent_data_ref); | |
1404 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1405 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1406 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1407 | struct btrfs_shared_data_ref); | |
1408 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1409 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1410 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1411 | struct btrfs_extent_ref_v0 *ref0; | |
1412 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1413 | struct btrfs_extent_ref_v0); | |
1414 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1415 | #endif |
5d4f98a2 YZ |
1416 | } else { |
1417 | WARN_ON(1); | |
1418 | } | |
1419 | return num_refs; | |
1420 | } | |
15916de8 | 1421 | |
5d4f98a2 | 1422 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1423 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1424 | struct btrfs_path *path, |
1425 | u64 bytenr, u64 parent, | |
1426 | u64 root_objectid) | |
1f3c79a2 | 1427 | { |
87bde3cd | 1428 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 | 1429 | struct btrfs_key key; |
1f3c79a2 | 1430 | int ret; |
1f3c79a2 | 1431 | |
5d4f98a2 YZ |
1432 | key.objectid = bytenr; |
1433 | if (parent) { | |
1434 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1435 | key.offset = parent; | |
1436 | } else { | |
1437 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1438 | key.offset = root_objectid; | |
1f3c79a2 LH |
1439 | } |
1440 | ||
5d4f98a2 YZ |
1441 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1442 | if (ret > 0) | |
1443 | ret = -ENOENT; | |
1444 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1445 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1446 | btrfs_release_path(path); |
5d4f98a2 YZ |
1447 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1448 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1449 | if (ret > 0) | |
1450 | ret = -ENOENT; | |
1451 | } | |
1f3c79a2 | 1452 | #endif |
5d4f98a2 | 1453 | return ret; |
1f3c79a2 LH |
1454 | } |
1455 | ||
5d4f98a2 | 1456 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1457 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1458 | struct btrfs_path *path, |
1459 | u64 bytenr, u64 parent, | |
1460 | u64 root_objectid) | |
31840ae1 | 1461 | { |
5d4f98a2 | 1462 | struct btrfs_key key; |
31840ae1 | 1463 | int ret; |
31840ae1 | 1464 | |
5d4f98a2 YZ |
1465 | key.objectid = bytenr; |
1466 | if (parent) { | |
1467 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1468 | key.offset = parent; | |
1469 | } else { | |
1470 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1471 | key.offset = root_objectid; | |
1472 | } | |
1473 | ||
87bde3cd JM |
1474 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, |
1475 | path, &key, 0); | |
b3b4aa74 | 1476 | btrfs_release_path(path); |
31840ae1 ZY |
1477 | return ret; |
1478 | } | |
1479 | ||
5d4f98a2 | 1480 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1481 | { |
5d4f98a2 YZ |
1482 | int type; |
1483 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1484 | if (parent > 0) | |
1485 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1486 | else | |
1487 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1488 | } else { | |
1489 | if (parent > 0) | |
1490 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1491 | else | |
1492 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1493 | } | |
1494 | return type; | |
31840ae1 | 1495 | } |
56bec294 | 1496 | |
2c47e605 YZ |
1497 | static int find_next_key(struct btrfs_path *path, int level, |
1498 | struct btrfs_key *key) | |
56bec294 | 1499 | |
02217ed2 | 1500 | { |
2c47e605 | 1501 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1502 | if (!path->nodes[level]) |
1503 | break; | |
5d4f98a2 YZ |
1504 | if (path->slots[level] + 1 >= |
1505 | btrfs_header_nritems(path->nodes[level])) | |
1506 | continue; | |
1507 | if (level == 0) | |
1508 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1509 | path->slots[level] + 1); | |
1510 | else | |
1511 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1512 | path->slots[level] + 1); | |
1513 | return 0; | |
1514 | } | |
1515 | return 1; | |
1516 | } | |
037e6390 | 1517 | |
5d4f98a2 YZ |
1518 | /* |
1519 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1520 | * to the address of inline back ref, and 0 is returned. | |
1521 | * | |
1522 | * if back ref isn't found, *ref_ret is set to the address where it | |
1523 | * should be inserted, and -ENOENT is returned. | |
1524 | * | |
1525 | * if insert is true and there are too many inline back refs, the path | |
1526 | * points to the extent item, and -EAGAIN is returned. | |
1527 | * | |
1528 | * NOTE: inline back refs are ordered in the same way that back ref | |
1529 | * items in the tree are ordered. | |
1530 | */ | |
1531 | static noinline_for_stack | |
1532 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1533 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1534 | struct btrfs_path *path, |
1535 | struct btrfs_extent_inline_ref **ref_ret, | |
1536 | u64 bytenr, u64 num_bytes, | |
1537 | u64 parent, u64 root_objectid, | |
1538 | u64 owner, u64 offset, int insert) | |
1539 | { | |
87bde3cd | 1540 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1541 | struct btrfs_key key; |
1542 | struct extent_buffer *leaf; | |
1543 | struct btrfs_extent_item *ei; | |
1544 | struct btrfs_extent_inline_ref *iref; | |
1545 | u64 flags; | |
1546 | u64 item_size; | |
1547 | unsigned long ptr; | |
1548 | unsigned long end; | |
1549 | int extra_size; | |
1550 | int type; | |
1551 | int want; | |
1552 | int ret; | |
1553 | int err = 0; | |
0b246afa | 1554 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
26b8003f | 1555 | |
db94535d | 1556 | key.objectid = bytenr; |
31840ae1 | 1557 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1558 | key.offset = num_bytes; |
31840ae1 | 1559 | |
5d4f98a2 YZ |
1560 | want = extent_ref_type(parent, owner); |
1561 | if (insert) { | |
1562 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1563 | path->keep_locks = 1; |
5d4f98a2 YZ |
1564 | } else |
1565 | extra_size = -1; | |
3173a18f JB |
1566 | |
1567 | /* | |
1568 | * Owner is our parent level, so we can just add one to get the level | |
1569 | * for the block we are interested in. | |
1570 | */ | |
1571 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1572 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1573 | key.offset = owner; | |
1574 | } | |
1575 | ||
1576 | again: | |
5d4f98a2 | 1577 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1578 | if (ret < 0) { |
5d4f98a2 YZ |
1579 | err = ret; |
1580 | goto out; | |
1581 | } | |
3173a18f JB |
1582 | |
1583 | /* | |
1584 | * We may be a newly converted file system which still has the old fat | |
1585 | * extent entries for metadata, so try and see if we have one of those. | |
1586 | */ | |
1587 | if (ret > 0 && skinny_metadata) { | |
1588 | skinny_metadata = false; | |
1589 | if (path->slots[0]) { | |
1590 | path->slots[0]--; | |
1591 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1592 | path->slots[0]); | |
1593 | if (key.objectid == bytenr && | |
1594 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1595 | key.offset == num_bytes) | |
1596 | ret = 0; | |
1597 | } | |
1598 | if (ret) { | |
9ce49a0b | 1599 | key.objectid = bytenr; |
3173a18f JB |
1600 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1601 | key.offset = num_bytes; | |
1602 | btrfs_release_path(path); | |
1603 | goto again; | |
1604 | } | |
1605 | } | |
1606 | ||
79787eaa JM |
1607 | if (ret && !insert) { |
1608 | err = -ENOENT; | |
1609 | goto out; | |
fae7f21c | 1610 | } else if (WARN_ON(ret)) { |
492104c8 | 1611 | err = -EIO; |
492104c8 | 1612 | goto out; |
79787eaa | 1613 | } |
5d4f98a2 YZ |
1614 | |
1615 | leaf = path->nodes[0]; | |
1616 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1617 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1618 | if (item_size < sizeof(*ei)) { | |
1619 | if (!insert) { | |
1620 | err = -ENOENT; | |
1621 | goto out; | |
1622 | } | |
87bde3cd | 1623 | ret = convert_extent_item_v0(trans, fs_info, path, owner, |
5d4f98a2 YZ |
1624 | extra_size); |
1625 | if (ret < 0) { | |
1626 | err = ret; | |
1627 | goto out; | |
1628 | } | |
1629 | leaf = path->nodes[0]; | |
1630 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1631 | } | |
1632 | #endif | |
1633 | BUG_ON(item_size < sizeof(*ei)); | |
1634 | ||
5d4f98a2 YZ |
1635 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1636 | flags = btrfs_extent_flags(leaf, ei); | |
1637 | ||
1638 | ptr = (unsigned long)(ei + 1); | |
1639 | end = (unsigned long)ei + item_size; | |
1640 | ||
3173a18f | 1641 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1642 | ptr += sizeof(struct btrfs_tree_block_info); |
1643 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1644 | } |
1645 | ||
1646 | err = -ENOENT; | |
1647 | while (1) { | |
1648 | if (ptr >= end) { | |
1649 | WARN_ON(ptr > end); | |
1650 | break; | |
1651 | } | |
1652 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1653 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1654 | if (want < type) | |
1655 | break; | |
1656 | if (want > type) { | |
1657 | ptr += btrfs_extent_inline_ref_size(type); | |
1658 | continue; | |
1659 | } | |
1660 | ||
1661 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1662 | struct btrfs_extent_data_ref *dref; | |
1663 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1664 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1665 | owner, offset)) { | |
1666 | err = 0; | |
1667 | break; | |
1668 | } | |
1669 | if (hash_extent_data_ref_item(leaf, dref) < | |
1670 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1671 | break; | |
1672 | } else { | |
1673 | u64 ref_offset; | |
1674 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1675 | if (parent > 0) { | |
1676 | if (parent == ref_offset) { | |
1677 | err = 0; | |
1678 | break; | |
1679 | } | |
1680 | if (ref_offset < parent) | |
1681 | break; | |
1682 | } else { | |
1683 | if (root_objectid == ref_offset) { | |
1684 | err = 0; | |
1685 | break; | |
1686 | } | |
1687 | if (ref_offset < root_objectid) | |
1688 | break; | |
1689 | } | |
1690 | } | |
1691 | ptr += btrfs_extent_inline_ref_size(type); | |
1692 | } | |
1693 | if (err == -ENOENT && insert) { | |
1694 | if (item_size + extra_size >= | |
1695 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1696 | err = -EAGAIN; | |
1697 | goto out; | |
1698 | } | |
1699 | /* | |
1700 | * To add new inline back ref, we have to make sure | |
1701 | * there is no corresponding back ref item. | |
1702 | * For simplicity, we just do not add new inline back | |
1703 | * ref if there is any kind of item for this block | |
1704 | */ | |
2c47e605 YZ |
1705 | if (find_next_key(path, 0, &key) == 0 && |
1706 | key.objectid == bytenr && | |
85d4198e | 1707 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1708 | err = -EAGAIN; |
1709 | goto out; | |
1710 | } | |
1711 | } | |
1712 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1713 | out: | |
85d4198e | 1714 | if (insert) { |
5d4f98a2 YZ |
1715 | path->keep_locks = 0; |
1716 | btrfs_unlock_up_safe(path, 1); | |
1717 | } | |
1718 | return err; | |
1719 | } | |
1720 | ||
1721 | /* | |
1722 | * helper to add new inline back ref | |
1723 | */ | |
1724 | static noinline_for_stack | |
87bde3cd | 1725 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1726 | struct btrfs_path *path, |
1727 | struct btrfs_extent_inline_ref *iref, | |
1728 | u64 parent, u64 root_objectid, | |
1729 | u64 owner, u64 offset, int refs_to_add, | |
1730 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1731 | { |
1732 | struct extent_buffer *leaf; | |
1733 | struct btrfs_extent_item *ei; | |
1734 | unsigned long ptr; | |
1735 | unsigned long end; | |
1736 | unsigned long item_offset; | |
1737 | u64 refs; | |
1738 | int size; | |
1739 | int type; | |
5d4f98a2 YZ |
1740 | |
1741 | leaf = path->nodes[0]; | |
1742 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1743 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1744 | ||
1745 | type = extent_ref_type(parent, owner); | |
1746 | size = btrfs_extent_inline_ref_size(type); | |
1747 | ||
87bde3cd | 1748 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1749 | |
1750 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1751 | refs = btrfs_extent_refs(leaf, ei); | |
1752 | refs += refs_to_add; | |
1753 | btrfs_set_extent_refs(leaf, ei, refs); | |
1754 | if (extent_op) | |
1755 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1756 | ||
1757 | ptr = (unsigned long)ei + item_offset; | |
1758 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1759 | if (ptr < end - size) | |
1760 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1761 | end - size - ptr); | |
1762 | ||
1763 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1764 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1765 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1766 | struct btrfs_extent_data_ref *dref; | |
1767 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1768 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1769 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1770 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1771 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1772 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1773 | struct btrfs_shared_data_ref *sref; | |
1774 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1775 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1776 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1777 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1778 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1779 | } else { | |
1780 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1781 | } | |
1782 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1783 | } |
1784 | ||
1785 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1786 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1787 | struct btrfs_path *path, |
1788 | struct btrfs_extent_inline_ref **ref_ret, | |
1789 | u64 bytenr, u64 num_bytes, u64 parent, | |
1790 | u64 root_objectid, u64 owner, u64 offset) | |
1791 | { | |
1792 | int ret; | |
1793 | ||
87bde3cd | 1794 | ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret, |
5d4f98a2 YZ |
1795 | bytenr, num_bytes, parent, |
1796 | root_objectid, owner, offset, 0); | |
1797 | if (ret != -ENOENT) | |
54aa1f4d | 1798 | return ret; |
5d4f98a2 | 1799 | |
b3b4aa74 | 1800 | btrfs_release_path(path); |
5d4f98a2 YZ |
1801 | *ref_ret = NULL; |
1802 | ||
1803 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
87bde3cd JM |
1804 | ret = lookup_tree_block_ref(trans, fs_info, path, bytenr, |
1805 | parent, root_objectid); | |
5d4f98a2 | 1806 | } else { |
87bde3cd JM |
1807 | ret = lookup_extent_data_ref(trans, fs_info, path, bytenr, |
1808 | parent, root_objectid, owner, | |
1809 | offset); | |
b9473439 | 1810 | } |
5d4f98a2 YZ |
1811 | return ret; |
1812 | } | |
31840ae1 | 1813 | |
5d4f98a2 YZ |
1814 | /* |
1815 | * helper to update/remove inline back ref | |
1816 | */ | |
1817 | static noinline_for_stack | |
87bde3cd | 1818 | void update_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1819 | struct btrfs_path *path, |
1820 | struct btrfs_extent_inline_ref *iref, | |
1821 | int refs_to_mod, | |
fcebe456 JB |
1822 | struct btrfs_delayed_extent_op *extent_op, |
1823 | int *last_ref) | |
5d4f98a2 YZ |
1824 | { |
1825 | struct extent_buffer *leaf; | |
1826 | struct btrfs_extent_item *ei; | |
1827 | struct btrfs_extent_data_ref *dref = NULL; | |
1828 | struct btrfs_shared_data_ref *sref = NULL; | |
1829 | unsigned long ptr; | |
1830 | unsigned long end; | |
1831 | u32 item_size; | |
1832 | int size; | |
1833 | int type; | |
5d4f98a2 YZ |
1834 | u64 refs; |
1835 | ||
1836 | leaf = path->nodes[0]; | |
1837 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1838 | refs = btrfs_extent_refs(leaf, ei); | |
1839 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1840 | refs += refs_to_mod; | |
1841 | btrfs_set_extent_refs(leaf, ei, refs); | |
1842 | if (extent_op) | |
1843 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1844 | ||
1845 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1846 | ||
1847 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1848 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1849 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1850 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1851 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1852 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1853 | } else { | |
1854 | refs = 1; | |
1855 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1856 | } |
31840ae1 | 1857 | |
5d4f98a2 YZ |
1858 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1859 | refs += refs_to_mod; | |
1860 | ||
1861 | if (refs > 0) { | |
1862 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1863 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1864 | else | |
1865 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1866 | } else { | |
fcebe456 | 1867 | *last_ref = 1; |
5d4f98a2 YZ |
1868 | size = btrfs_extent_inline_ref_size(type); |
1869 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1870 | ptr = (unsigned long)iref; | |
1871 | end = (unsigned long)ei + item_size; | |
1872 | if (ptr + size < end) | |
1873 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1874 | end - ptr - size); | |
1875 | item_size -= size; | |
87bde3cd | 1876 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1877 | } |
1878 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1879 | } |
1880 | ||
1881 | static noinline_for_stack | |
1882 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1883 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1884 | struct btrfs_path *path, |
1885 | u64 bytenr, u64 num_bytes, u64 parent, | |
1886 | u64 root_objectid, u64 owner, | |
1887 | u64 offset, int refs_to_add, | |
1888 | struct btrfs_delayed_extent_op *extent_op) | |
1889 | { | |
1890 | struct btrfs_extent_inline_ref *iref; | |
1891 | int ret; | |
1892 | ||
87bde3cd | 1893 | ret = lookup_inline_extent_backref(trans, fs_info, path, &iref, |
5d4f98a2 YZ |
1894 | bytenr, num_bytes, parent, |
1895 | root_objectid, owner, offset, 1); | |
1896 | if (ret == 0) { | |
1897 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
87bde3cd | 1898 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 1899 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1900 | } else if (ret == -ENOENT) { |
87bde3cd | 1901 | setup_inline_extent_backref(fs_info, path, iref, parent, |
143bede5 JM |
1902 | root_objectid, owner, offset, |
1903 | refs_to_add, extent_op); | |
1904 | ret = 0; | |
771ed689 | 1905 | } |
5d4f98a2 YZ |
1906 | return ret; |
1907 | } | |
31840ae1 | 1908 | |
5d4f98a2 | 1909 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 1910 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1911 | struct btrfs_path *path, |
1912 | u64 bytenr, u64 parent, u64 root_objectid, | |
1913 | u64 owner, u64 offset, int refs_to_add) | |
1914 | { | |
1915 | int ret; | |
1916 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1917 | BUG_ON(refs_to_add != 1); | |
87bde3cd | 1918 | ret = insert_tree_block_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
1919 | parent, root_objectid); |
1920 | } else { | |
87bde3cd | 1921 | ret = insert_extent_data_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
1922 | parent, root_objectid, |
1923 | owner, offset, refs_to_add); | |
1924 | } | |
1925 | return ret; | |
1926 | } | |
56bec294 | 1927 | |
5d4f98a2 | 1928 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 1929 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1930 | struct btrfs_path *path, |
1931 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1932 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1933 | { |
143bede5 | 1934 | int ret = 0; |
b9473439 | 1935 | |
5d4f98a2 YZ |
1936 | BUG_ON(!is_data && refs_to_drop != 1); |
1937 | if (iref) { | |
87bde3cd | 1938 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 1939 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 1940 | } else if (is_data) { |
87bde3cd | 1941 | ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop, |
fcebe456 | 1942 | last_ref); |
5d4f98a2 | 1943 | } else { |
fcebe456 | 1944 | *last_ref = 1; |
87bde3cd | 1945 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
5d4f98a2 YZ |
1946 | } |
1947 | return ret; | |
1948 | } | |
1949 | ||
86557861 | 1950 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
1951 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1952 | u64 *discarded_bytes) | |
5d4f98a2 | 1953 | { |
86557861 JM |
1954 | int j, ret = 0; |
1955 | u64 bytes_left, end; | |
4d89d377 | 1956 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1957 | |
4d89d377 JM |
1958 | if (WARN_ON(start != aligned_start)) { |
1959 | len -= aligned_start - start; | |
1960 | len = round_down(len, 1 << 9); | |
1961 | start = aligned_start; | |
1962 | } | |
d04c6b88 | 1963 | |
4d89d377 | 1964 | *discarded_bytes = 0; |
86557861 JM |
1965 | |
1966 | if (!len) | |
1967 | return 0; | |
1968 | ||
1969 | end = start + len; | |
1970 | bytes_left = len; | |
1971 | ||
1972 | /* Skip any superblocks on this device. */ | |
1973 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1974 | u64 sb_start = btrfs_sb_offset(j); | |
1975 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1976 | u64 size = sb_start - start; | |
1977 | ||
1978 | if (!in_range(sb_start, start, bytes_left) && | |
1979 | !in_range(sb_end, start, bytes_left) && | |
1980 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1981 | continue; | |
1982 | ||
1983 | /* | |
1984 | * Superblock spans beginning of range. Adjust start and | |
1985 | * try again. | |
1986 | */ | |
1987 | if (sb_start <= start) { | |
1988 | start += sb_end - start; | |
1989 | if (start > end) { | |
1990 | bytes_left = 0; | |
1991 | break; | |
1992 | } | |
1993 | bytes_left = end - start; | |
1994 | continue; | |
1995 | } | |
1996 | ||
1997 | if (size) { | |
1998 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1999 | GFP_NOFS, 0); | |
2000 | if (!ret) | |
2001 | *discarded_bytes += size; | |
2002 | else if (ret != -EOPNOTSUPP) | |
2003 | return ret; | |
2004 | } | |
2005 | ||
2006 | start = sb_end; | |
2007 | if (start > end) { | |
2008 | bytes_left = 0; | |
2009 | break; | |
2010 | } | |
2011 | bytes_left = end - start; | |
2012 | } | |
2013 | ||
2014 | if (bytes_left) { | |
2015 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2016 | GFP_NOFS, 0); |
2017 | if (!ret) | |
86557861 | 2018 | *discarded_bytes += bytes_left; |
4d89d377 | 2019 | } |
d04c6b88 | 2020 | return ret; |
5d4f98a2 | 2021 | } |
5d4f98a2 | 2022 | |
2ff7e61e | 2023 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 2024 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 2025 | { |
5d4f98a2 | 2026 | int ret; |
5378e607 | 2027 | u64 discarded_bytes = 0; |
a1d3c478 | 2028 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2029 | |
e244a0ae | 2030 | |
2999241d FM |
2031 | /* |
2032 | * Avoid races with device replace and make sure our bbio has devices | |
2033 | * associated to its stripes that don't go away while we are discarding. | |
2034 | */ | |
0b246afa | 2035 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 2036 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
2037 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
2038 | &bbio, 0); | |
79787eaa | 2039 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2040 | if (!ret) { |
a1d3c478 | 2041 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2042 | int i; |
2043 | ||
5d4f98a2 | 2044 | |
a1d3c478 | 2045 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2046 | u64 bytes; |
d5e2003c JB |
2047 | if (!stripe->dev->can_discard) |
2048 | continue; | |
2049 | ||
5378e607 LD |
2050 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2051 | stripe->physical, | |
d04c6b88 JM |
2052 | stripe->length, |
2053 | &bytes); | |
5378e607 | 2054 | if (!ret) |
d04c6b88 | 2055 | discarded_bytes += bytes; |
5378e607 | 2056 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2057 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2058 | |
2059 | /* | |
2060 | * Just in case we get back EOPNOTSUPP for some reason, | |
2061 | * just ignore the return value so we don't screw up | |
2062 | * people calling discard_extent. | |
2063 | */ | |
2064 | ret = 0; | |
5d4f98a2 | 2065 | } |
6e9606d2 | 2066 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2067 | } |
0b246afa | 2068 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2069 | |
2070 | if (actual_bytes) | |
2071 | *actual_bytes = discarded_bytes; | |
2072 | ||
5d4f98a2 | 2073 | |
53b381b3 DW |
2074 | if (ret == -EOPNOTSUPP) |
2075 | ret = 0; | |
5d4f98a2 | 2076 | return ret; |
5d4f98a2 YZ |
2077 | } |
2078 | ||
79787eaa | 2079 | /* Can return -ENOMEM */ |
5d4f98a2 | 2080 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2081 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 2082 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2083 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
2084 | { |
2085 | int ret; | |
66d7e7f0 | 2086 | |
5d4f98a2 YZ |
2087 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2088 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2089 | ||
2090 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
2091 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
2092 | num_bytes, | |
5d4f98a2 | 2093 | parent, root_objectid, (int)owner, |
b06c4bf5 | 2094 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 | 2095 | } else { |
66d7e7f0 | 2096 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5846a3c2 QW |
2097 | num_bytes, parent, root_objectid, |
2098 | owner, offset, 0, | |
fef394f7 | 2099 | BTRFS_ADD_DELAYED_REF); |
5d4f98a2 YZ |
2100 | } |
2101 | return ret; | |
2102 | } | |
2103 | ||
2104 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2105 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 2106 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2107 | u64 parent, u64 root_objectid, |
2108 | u64 owner, u64 offset, int refs_to_add, | |
2109 | struct btrfs_delayed_extent_op *extent_op) | |
2110 | { | |
2111 | struct btrfs_path *path; | |
2112 | struct extent_buffer *leaf; | |
2113 | struct btrfs_extent_item *item; | |
fcebe456 | 2114 | struct btrfs_key key; |
c682f9b3 QW |
2115 | u64 bytenr = node->bytenr; |
2116 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2117 | u64 refs; |
2118 | int ret; | |
5d4f98a2 YZ |
2119 | |
2120 | path = btrfs_alloc_path(); | |
2121 | if (!path) | |
2122 | return -ENOMEM; | |
2123 | ||
e4058b54 | 2124 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2125 | path->leave_spinning = 1; |
2126 | /* this will setup the path even if it fails to insert the back ref */ | |
87bde3cd JM |
2127 | ret = insert_inline_extent_backref(trans, fs_info, path, bytenr, |
2128 | num_bytes, parent, root_objectid, | |
2129 | owner, offset, | |
5d4f98a2 | 2130 | refs_to_add, extent_op); |
0ed4792a | 2131 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2132 | goto out; |
fcebe456 JB |
2133 | |
2134 | /* | |
2135 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2136 | * inline extent ref, so just update the reference count and add a | |
2137 | * normal backref. | |
2138 | */ | |
5d4f98a2 | 2139 | leaf = path->nodes[0]; |
fcebe456 | 2140 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2141 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2142 | refs = btrfs_extent_refs(leaf, item); | |
2143 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2144 | if (extent_op) | |
2145 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2146 | |
5d4f98a2 | 2147 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2148 | btrfs_release_path(path); |
56bec294 | 2149 | |
e4058b54 | 2150 | path->reada = READA_FORWARD; |
b9473439 | 2151 | path->leave_spinning = 1; |
56bec294 | 2152 | /* now insert the actual backref */ |
87bde3cd JM |
2153 | ret = insert_extent_backref(trans, fs_info, path, bytenr, parent, |
2154 | root_objectid, owner, offset, refs_to_add); | |
79787eaa | 2155 | if (ret) |
66642832 | 2156 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2157 | out: |
56bec294 | 2158 | btrfs_free_path(path); |
30d133fc | 2159 | return ret; |
56bec294 CM |
2160 | } |
2161 | ||
5d4f98a2 | 2162 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2163 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2164 | struct btrfs_delayed_ref_node *node, |
2165 | struct btrfs_delayed_extent_op *extent_op, | |
2166 | int insert_reserved) | |
56bec294 | 2167 | { |
5d4f98a2 YZ |
2168 | int ret = 0; |
2169 | struct btrfs_delayed_data_ref *ref; | |
2170 | struct btrfs_key ins; | |
2171 | u64 parent = 0; | |
2172 | u64 ref_root = 0; | |
2173 | u64 flags = 0; | |
2174 | ||
2175 | ins.objectid = node->bytenr; | |
2176 | ins.offset = node->num_bytes; | |
2177 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2178 | ||
2179 | ref = btrfs_delayed_node_to_data_ref(node); | |
0b246afa | 2180 | trace_run_delayed_data_ref(fs_info, node, ref, node->action); |
599c75ec | 2181 | |
5d4f98a2 YZ |
2182 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2183 | parent = ref->parent; | |
fcebe456 | 2184 | ref_root = ref->root; |
5d4f98a2 YZ |
2185 | |
2186 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2187 | if (extent_op) |
5d4f98a2 | 2188 | flags |= extent_op->flags_to_set; |
2ff7e61e | 2189 | ret = alloc_reserved_file_extent(trans, fs_info, |
5d4f98a2 YZ |
2190 | parent, ref_root, flags, |
2191 | ref->objectid, ref->offset, | |
2192 | &ins, node->ref_mod); | |
5d4f98a2 | 2193 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2194 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2195 | ref_root, ref->objectid, |
2196 | ref->offset, node->ref_mod, | |
c682f9b3 | 2197 | extent_op); |
5d4f98a2 | 2198 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2199 | ret = __btrfs_free_extent(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2200 | ref_root, ref->objectid, |
2201 | ref->offset, node->ref_mod, | |
c682f9b3 | 2202 | extent_op); |
5d4f98a2 YZ |
2203 | } else { |
2204 | BUG(); | |
2205 | } | |
2206 | return ret; | |
2207 | } | |
2208 | ||
2209 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2210 | struct extent_buffer *leaf, | |
2211 | struct btrfs_extent_item *ei) | |
2212 | { | |
2213 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2214 | if (extent_op->update_flags) { | |
2215 | flags |= extent_op->flags_to_set; | |
2216 | btrfs_set_extent_flags(leaf, ei, flags); | |
2217 | } | |
2218 | ||
2219 | if (extent_op->update_key) { | |
2220 | struct btrfs_tree_block_info *bi; | |
2221 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2222 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2223 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2224 | } | |
2225 | } | |
2226 | ||
2227 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2228 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2229 | struct btrfs_delayed_ref_node *node, |
2230 | struct btrfs_delayed_extent_op *extent_op) | |
2231 | { | |
2232 | struct btrfs_key key; | |
2233 | struct btrfs_path *path; | |
2234 | struct btrfs_extent_item *ei; | |
2235 | struct extent_buffer *leaf; | |
2236 | u32 item_size; | |
56bec294 | 2237 | int ret; |
5d4f98a2 | 2238 | int err = 0; |
b1c79e09 | 2239 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2240 | |
79787eaa JM |
2241 | if (trans->aborted) |
2242 | return 0; | |
2243 | ||
0b246afa | 2244 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2245 | metadata = 0; |
2246 | ||
5d4f98a2 YZ |
2247 | path = btrfs_alloc_path(); |
2248 | if (!path) | |
2249 | return -ENOMEM; | |
2250 | ||
2251 | key.objectid = node->bytenr; | |
5d4f98a2 | 2252 | |
3173a18f | 2253 | if (metadata) { |
3173a18f | 2254 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2255 | key.offset = extent_op->level; |
3173a18f JB |
2256 | } else { |
2257 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2258 | key.offset = node->num_bytes; | |
2259 | } | |
2260 | ||
2261 | again: | |
e4058b54 | 2262 | path->reada = READA_FORWARD; |
5d4f98a2 | 2263 | path->leave_spinning = 1; |
0b246afa | 2264 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2265 | if (ret < 0) { |
2266 | err = ret; | |
2267 | goto out; | |
2268 | } | |
2269 | if (ret > 0) { | |
3173a18f | 2270 | if (metadata) { |
55994887 FDBM |
2271 | if (path->slots[0] > 0) { |
2272 | path->slots[0]--; | |
2273 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2274 | path->slots[0]); | |
2275 | if (key.objectid == node->bytenr && | |
2276 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2277 | key.offset == node->num_bytes) | |
2278 | ret = 0; | |
2279 | } | |
2280 | if (ret > 0) { | |
2281 | btrfs_release_path(path); | |
2282 | metadata = 0; | |
3173a18f | 2283 | |
55994887 FDBM |
2284 | key.objectid = node->bytenr; |
2285 | key.offset = node->num_bytes; | |
2286 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2287 | goto again; | |
2288 | } | |
2289 | } else { | |
2290 | err = -EIO; | |
2291 | goto out; | |
3173a18f | 2292 | } |
5d4f98a2 YZ |
2293 | } |
2294 | ||
2295 | leaf = path->nodes[0]; | |
2296 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2297 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2298 | if (item_size < sizeof(*ei)) { | |
87bde3cd | 2299 | ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0); |
5d4f98a2 YZ |
2300 | if (ret < 0) { |
2301 | err = ret; | |
2302 | goto out; | |
2303 | } | |
2304 | leaf = path->nodes[0]; | |
2305 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2306 | } | |
2307 | #endif | |
2308 | BUG_ON(item_size < sizeof(*ei)); | |
2309 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2310 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2311 | |
5d4f98a2 YZ |
2312 | btrfs_mark_buffer_dirty(leaf); |
2313 | out: | |
2314 | btrfs_free_path(path); | |
2315 | return err; | |
56bec294 CM |
2316 | } |
2317 | ||
5d4f98a2 | 2318 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2319 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2320 | struct btrfs_delayed_ref_node *node, |
2321 | struct btrfs_delayed_extent_op *extent_op, | |
2322 | int insert_reserved) | |
56bec294 CM |
2323 | { |
2324 | int ret = 0; | |
5d4f98a2 YZ |
2325 | struct btrfs_delayed_tree_ref *ref; |
2326 | struct btrfs_key ins; | |
2327 | u64 parent = 0; | |
2328 | u64 ref_root = 0; | |
0b246afa | 2329 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
56bec294 | 2330 | |
5d4f98a2 | 2331 | ref = btrfs_delayed_node_to_tree_ref(node); |
0b246afa | 2332 | trace_run_delayed_tree_ref(fs_info, node, ref, node->action); |
599c75ec | 2333 | |
5d4f98a2 YZ |
2334 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2335 | parent = ref->parent; | |
fcebe456 | 2336 | ref_root = ref->root; |
5d4f98a2 | 2337 | |
3173a18f JB |
2338 | ins.objectid = node->bytenr; |
2339 | if (skinny_metadata) { | |
2340 | ins.offset = ref->level; | |
2341 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2342 | } else { | |
2343 | ins.offset = node->num_bytes; | |
2344 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2345 | } | |
2346 | ||
02794222 | 2347 | if (node->ref_mod != 1) { |
2ff7e61e | 2348 | btrfs_err(fs_info, |
02794222 LB |
2349 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2350 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2351 | parent); | |
2352 | return -EIO; | |
2353 | } | |
5d4f98a2 | 2354 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2355 | BUG_ON(!extent_op || !extent_op->update_flags); |
2ff7e61e | 2356 | ret = alloc_reserved_tree_block(trans, fs_info, |
5d4f98a2 YZ |
2357 | parent, ref_root, |
2358 | extent_op->flags_to_set, | |
2359 | &extent_op->key, | |
b06c4bf5 | 2360 | ref->level, &ins); |
5d4f98a2 | 2361 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2362 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, |
c682f9b3 QW |
2363 | parent, ref_root, |
2364 | ref->level, 0, 1, | |
fcebe456 | 2365 | extent_op); |
5d4f98a2 | 2366 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2367 | ret = __btrfs_free_extent(trans, fs_info, node, |
c682f9b3 QW |
2368 | parent, ref_root, |
2369 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2370 | } else { |
2371 | BUG(); | |
2372 | } | |
56bec294 CM |
2373 | return ret; |
2374 | } | |
2375 | ||
2376 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2377 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2378 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2379 | struct btrfs_delayed_ref_node *node, |
2380 | struct btrfs_delayed_extent_op *extent_op, | |
2381 | int insert_reserved) | |
56bec294 | 2382 | { |
79787eaa JM |
2383 | int ret = 0; |
2384 | ||
857cc2fc JB |
2385 | if (trans->aborted) { |
2386 | if (insert_reserved) | |
2ff7e61e | 2387 | btrfs_pin_extent(fs_info, node->bytenr, |
857cc2fc | 2388 | node->num_bytes, 1); |
79787eaa | 2389 | return 0; |
857cc2fc | 2390 | } |
79787eaa | 2391 | |
5d4f98a2 | 2392 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2393 | struct btrfs_delayed_ref_head *head; |
2394 | /* | |
2395 | * we've hit the end of the chain and we were supposed | |
2396 | * to insert this extent into the tree. But, it got | |
2397 | * deleted before we ever needed to insert it, so all | |
2398 | * we have to do is clean up the accounting | |
2399 | */ | |
5d4f98a2 YZ |
2400 | BUG_ON(extent_op); |
2401 | head = btrfs_delayed_node_to_head(node); | |
0b246afa | 2402 | trace_run_delayed_ref_head(fs_info, node, head, node->action); |
599c75ec | 2403 | |
56bec294 | 2404 | if (insert_reserved) { |
2ff7e61e | 2405 | btrfs_pin_extent(fs_info, node->bytenr, |
f0486c68 | 2406 | node->num_bytes, 1); |
5d4f98a2 | 2407 | if (head->is_data) { |
0b246afa | 2408 | ret = btrfs_del_csums(trans, fs_info, |
5d4f98a2 YZ |
2409 | node->bytenr, |
2410 | node->num_bytes); | |
5d4f98a2 | 2411 | } |
56bec294 | 2412 | } |
297d750b QW |
2413 | |
2414 | /* Also free its reserved qgroup space */ | |
0b246afa | 2415 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, |
297d750b | 2416 | head->qgroup_reserved); |
79787eaa | 2417 | return ret; |
56bec294 CM |
2418 | } |
2419 | ||
5d4f98a2 YZ |
2420 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2421 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2ff7e61e | 2422 | ret = run_delayed_tree_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2423 | insert_reserved); |
2424 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2425 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2ff7e61e | 2426 | ret = run_delayed_data_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2427 | insert_reserved); |
2428 | else | |
2429 | BUG(); | |
2430 | return ret; | |
56bec294 CM |
2431 | } |
2432 | ||
c6fc2454 | 2433 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2434 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2435 | { | |
cffc3374 FM |
2436 | struct btrfs_delayed_ref_node *ref; |
2437 | ||
c6fc2454 QW |
2438 | if (list_empty(&head->ref_list)) |
2439 | return NULL; | |
d7df2c79 | 2440 | |
cffc3374 FM |
2441 | /* |
2442 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2443 | * This is to prevent a ref count from going down to zero, which deletes | |
2444 | * the extent item from the extent tree, when there still are references | |
2445 | * to add, which would fail because they would not find the extent item. | |
2446 | */ | |
1d57ee94 WX |
2447 | if (!list_empty(&head->ref_add_list)) |
2448 | return list_first_entry(&head->ref_add_list, | |
2449 | struct btrfs_delayed_ref_node, add_list); | |
2450 | ||
2451 | ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node, | |
2452 | list); | |
2453 | ASSERT(list_empty(&ref->add_list)); | |
2454 | return ref; | |
56bec294 CM |
2455 | } |
2456 | ||
79787eaa JM |
2457 | /* |
2458 | * Returns 0 on success or if called with an already aborted transaction. | |
2459 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2460 | */ | |
d7df2c79 | 2461 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2462 | struct btrfs_fs_info *fs_info, |
d7df2c79 | 2463 | unsigned long nr) |
56bec294 | 2464 | { |
56bec294 CM |
2465 | struct btrfs_delayed_ref_root *delayed_refs; |
2466 | struct btrfs_delayed_ref_node *ref; | |
2467 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2468 | struct btrfs_delayed_extent_op *extent_op; |
0a2b2a84 | 2469 | ktime_t start = ktime_get(); |
56bec294 | 2470 | int ret; |
d7df2c79 | 2471 | unsigned long count = 0; |
0a2b2a84 | 2472 | unsigned long actual_count = 0; |
56bec294 | 2473 | int must_insert_reserved = 0; |
56bec294 CM |
2474 | |
2475 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2476 | while (1) { |
2477 | if (!locked_ref) { | |
d7df2c79 | 2478 | if (count >= nr) |
56bec294 | 2479 | break; |
56bec294 | 2480 | |
d7df2c79 JB |
2481 | spin_lock(&delayed_refs->lock); |
2482 | locked_ref = btrfs_select_ref_head(trans); | |
2483 | if (!locked_ref) { | |
2484 | spin_unlock(&delayed_refs->lock); | |
2485 | break; | |
2486 | } | |
c3e69d58 CM |
2487 | |
2488 | /* grab the lock that says we are going to process | |
2489 | * all the refs for this head */ | |
2490 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2491 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2492 | /* |
2493 | * we may have dropped the spin lock to get the head | |
2494 | * mutex lock, and that might have given someone else | |
2495 | * time to free the head. If that's true, it has been | |
2496 | * removed from our list and we can move on. | |
2497 | */ | |
2498 | if (ret == -EAGAIN) { | |
2499 | locked_ref = NULL; | |
2500 | count++; | |
2501 | continue; | |
56bec294 CM |
2502 | } |
2503 | } | |
a28ec197 | 2504 | |
2c3cf7d5 FM |
2505 | /* |
2506 | * We need to try and merge add/drops of the same ref since we | |
2507 | * can run into issues with relocate dropping the implicit ref | |
2508 | * and then it being added back again before the drop can | |
2509 | * finish. If we merged anything we need to re-loop so we can | |
2510 | * get a good ref. | |
2511 | * Or we can get node references of the same type that weren't | |
2512 | * merged when created due to bumps in the tree mod seq, and | |
2513 | * we need to merge them to prevent adding an inline extent | |
2514 | * backref before dropping it (triggering a BUG_ON at | |
2515 | * insert_inline_extent_backref()). | |
2516 | */ | |
d7df2c79 | 2517 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2518 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2519 | locked_ref); | |
ae1e206b | 2520 | |
d1270cd9 AJ |
2521 | /* |
2522 | * locked_ref is the head node, so we have to go one | |
2523 | * node back for any delayed ref updates | |
2524 | */ | |
2525 | ref = select_delayed_ref(locked_ref); | |
2526 | ||
2527 | if (ref && ref->seq && | |
097b8a7c | 2528 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2529 | spin_unlock(&locked_ref->lock); |
d7df2c79 JB |
2530 | spin_lock(&delayed_refs->lock); |
2531 | locked_ref->processing = 0; | |
d1270cd9 AJ |
2532 | delayed_refs->num_heads_ready++; |
2533 | spin_unlock(&delayed_refs->lock); | |
d0280996 | 2534 | btrfs_delayed_ref_unlock(locked_ref); |
d7df2c79 | 2535 | locked_ref = NULL; |
d1270cd9 | 2536 | cond_resched(); |
27a377db | 2537 | count++; |
d1270cd9 AJ |
2538 | continue; |
2539 | } | |
2540 | ||
56bec294 CM |
2541 | /* |
2542 | * record the must insert reserved flag before we | |
2543 | * drop the spin lock. | |
2544 | */ | |
2545 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2546 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2547 | |
5d4f98a2 YZ |
2548 | extent_op = locked_ref->extent_op; |
2549 | locked_ref->extent_op = NULL; | |
2550 | ||
56bec294 | 2551 | if (!ref) { |
d7df2c79 JB |
2552 | |
2553 | ||
56bec294 CM |
2554 | /* All delayed refs have been processed, Go ahead |
2555 | * and send the head node to run_one_delayed_ref, | |
2556 | * so that any accounting fixes can happen | |
2557 | */ | |
2558 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2559 | |
2560 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2561 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2562 | extent_op = NULL; |
2563 | } | |
2564 | ||
2565 | if (extent_op) { | |
d7df2c79 | 2566 | spin_unlock(&locked_ref->lock); |
2ff7e61e | 2567 | ret = run_delayed_extent_op(trans, fs_info, |
5d4f98a2 | 2568 | ref, extent_op); |
78a6184a | 2569 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2570 | |
79787eaa | 2571 | if (ret) { |
857cc2fc JB |
2572 | /* |
2573 | * Need to reset must_insert_reserved if | |
2574 | * there was an error so the abort stuff | |
2575 | * can cleanup the reserved space | |
2576 | * properly. | |
2577 | */ | |
2578 | if (must_insert_reserved) | |
2579 | locked_ref->must_insert_reserved = 1; | |
aa7c8da3 | 2580 | spin_lock(&delayed_refs->lock); |
d7df2c79 | 2581 | locked_ref->processing = 0; |
aa7c8da3 JM |
2582 | delayed_refs->num_heads_ready++; |
2583 | spin_unlock(&delayed_refs->lock); | |
5d163e0e JM |
2584 | btrfs_debug(fs_info, |
2585 | "run_delayed_extent_op returned %d", | |
2586 | ret); | |
093486c4 | 2587 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2588 | return ret; |
2589 | } | |
d7df2c79 | 2590 | continue; |
5d4f98a2 | 2591 | } |
02217ed2 | 2592 | |
d7df2c79 | 2593 | /* |
01327610 | 2594 | * Need to drop our head ref lock and re-acquire the |
d7df2c79 JB |
2595 | * delayed ref lock and then re-check to make sure |
2596 | * nobody got added. | |
2597 | */ | |
2598 | spin_unlock(&locked_ref->lock); | |
2599 | spin_lock(&delayed_refs->lock); | |
2600 | spin_lock(&locked_ref->lock); | |
c6fc2454 | 2601 | if (!list_empty(&locked_ref->ref_list) || |
573a0755 | 2602 | locked_ref->extent_op) { |
d7df2c79 JB |
2603 | spin_unlock(&locked_ref->lock); |
2604 | spin_unlock(&delayed_refs->lock); | |
2605 | continue; | |
2606 | } | |
2607 | ref->in_tree = 0; | |
2608 | delayed_refs->num_heads--; | |
c46effa6 LB |
2609 | rb_erase(&locked_ref->href_node, |
2610 | &delayed_refs->href_root); | |
d7df2c79 JB |
2611 | spin_unlock(&delayed_refs->lock); |
2612 | } else { | |
0a2b2a84 | 2613 | actual_count++; |
d7df2c79 | 2614 | ref->in_tree = 0; |
c6fc2454 | 2615 | list_del(&ref->list); |
1d57ee94 WX |
2616 | if (!list_empty(&ref->add_list)) |
2617 | list_del(&ref->add_list); | |
c46effa6 | 2618 | } |
d7df2c79 JB |
2619 | atomic_dec(&delayed_refs->num_entries); |
2620 | ||
093486c4 | 2621 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2622 | /* |
2623 | * when we play the delayed ref, also correct the | |
2624 | * ref_mod on head | |
2625 | */ | |
2626 | switch (ref->action) { | |
2627 | case BTRFS_ADD_DELAYED_REF: | |
2628 | case BTRFS_ADD_DELAYED_EXTENT: | |
2629 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2630 | break; | |
2631 | case BTRFS_DROP_DELAYED_REF: | |
2632 | locked_ref->node.ref_mod += ref->ref_mod; | |
2633 | break; | |
2634 | default: | |
2635 | WARN_ON(1); | |
2636 | } | |
2637 | } | |
d7df2c79 | 2638 | spin_unlock(&locked_ref->lock); |
925baedd | 2639 | |
2ff7e61e | 2640 | ret = run_one_delayed_ref(trans, fs_info, ref, extent_op, |
56bec294 | 2641 | must_insert_reserved); |
eb099670 | 2642 | |
78a6184a | 2643 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2644 | if (ret) { |
9d1032cc | 2645 | spin_lock(&delayed_refs->lock); |
d7df2c79 | 2646 | locked_ref->processing = 0; |
9d1032cc WX |
2647 | delayed_refs->num_heads_ready++; |
2648 | spin_unlock(&delayed_refs->lock); | |
093486c4 MX |
2649 | btrfs_delayed_ref_unlock(locked_ref); |
2650 | btrfs_put_delayed_ref(ref); | |
5d163e0e JM |
2651 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
2652 | ret); | |
79787eaa JM |
2653 | return ret; |
2654 | } | |
2655 | ||
093486c4 MX |
2656 | /* |
2657 | * If this node is a head, that means all the refs in this head | |
2658 | * have been dealt with, and we will pick the next head to deal | |
2659 | * with, so we must unlock the head and drop it from the cluster | |
2660 | * list before we release it. | |
2661 | */ | |
2662 | if (btrfs_delayed_ref_is_head(ref)) { | |
1262133b JB |
2663 | if (locked_ref->is_data && |
2664 | locked_ref->total_ref_mod < 0) { | |
2665 | spin_lock(&delayed_refs->lock); | |
2666 | delayed_refs->pending_csums -= ref->num_bytes; | |
2667 | spin_unlock(&delayed_refs->lock); | |
2668 | } | |
093486c4 MX |
2669 | btrfs_delayed_ref_unlock(locked_ref); |
2670 | locked_ref = NULL; | |
2671 | } | |
2672 | btrfs_put_delayed_ref(ref); | |
2673 | count++; | |
c3e69d58 | 2674 | cond_resched(); |
c3e69d58 | 2675 | } |
0a2b2a84 JB |
2676 | |
2677 | /* | |
2678 | * We don't want to include ref heads since we can have empty ref heads | |
2679 | * and those will drastically skew our runtime down since we just do | |
2680 | * accounting, no actual extent tree updates. | |
2681 | */ | |
2682 | if (actual_count > 0) { | |
2683 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2684 | u64 avg; | |
2685 | ||
2686 | /* | |
2687 | * We weigh the current average higher than our current runtime | |
2688 | * to avoid large swings in the average. | |
2689 | */ | |
2690 | spin_lock(&delayed_refs->lock); | |
2691 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2692 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2693 | spin_unlock(&delayed_refs->lock); |
2694 | } | |
d7df2c79 | 2695 | return 0; |
c3e69d58 CM |
2696 | } |
2697 | ||
709c0486 AJ |
2698 | #ifdef SCRAMBLE_DELAYED_REFS |
2699 | /* | |
2700 | * Normally delayed refs get processed in ascending bytenr order. This | |
2701 | * correlates in most cases to the order added. To expose dependencies on this | |
2702 | * order, we start to process the tree in the middle instead of the beginning | |
2703 | */ | |
2704 | static u64 find_middle(struct rb_root *root) | |
2705 | { | |
2706 | struct rb_node *n = root->rb_node; | |
2707 | struct btrfs_delayed_ref_node *entry; | |
2708 | int alt = 1; | |
2709 | u64 middle; | |
2710 | u64 first = 0, last = 0; | |
2711 | ||
2712 | n = rb_first(root); | |
2713 | if (n) { | |
2714 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2715 | first = entry->bytenr; | |
2716 | } | |
2717 | n = rb_last(root); | |
2718 | if (n) { | |
2719 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2720 | last = entry->bytenr; | |
2721 | } | |
2722 | n = root->rb_node; | |
2723 | ||
2724 | while (n) { | |
2725 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2726 | WARN_ON(!entry->in_tree); | |
2727 | ||
2728 | middle = entry->bytenr; | |
2729 | ||
2730 | if (alt) | |
2731 | n = n->rb_left; | |
2732 | else | |
2733 | n = n->rb_right; | |
2734 | ||
2735 | alt = 1 - alt; | |
2736 | } | |
2737 | return middle; | |
2738 | } | |
2739 | #endif | |
2740 | ||
2ff7e61e | 2741 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2742 | { |
2743 | u64 num_bytes; | |
2744 | ||
2745 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2746 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2747 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2748 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2749 | ||
2750 | /* | |
2751 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2752 | * closer to what we're really going to want to use. |
1be41b78 | 2753 | */ |
0b246afa | 2754 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2755 | } |
2756 | ||
1262133b JB |
2757 | /* |
2758 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2759 | * would require to store the csums for that many bytes. | |
2760 | */ | |
2ff7e61e | 2761 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2762 | { |
2763 | u64 csum_size; | |
2764 | u64 num_csums_per_leaf; | |
2765 | u64 num_csums; | |
2766 | ||
0b246afa | 2767 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2768 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2769 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2770 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2771 | num_csums += num_csums_per_leaf - 1; |
2772 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2773 | return num_csums; | |
2774 | } | |
2775 | ||
0a2b2a84 | 2776 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2777 | struct btrfs_fs_info *fs_info) |
1be41b78 JB |
2778 | { |
2779 | struct btrfs_block_rsv *global_rsv; | |
2780 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2781 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
cb723e49 JB |
2782 | u64 num_dirty_bgs = trans->transaction->num_dirty_bgs; |
2783 | u64 num_bytes, num_dirty_bgs_bytes; | |
1be41b78 JB |
2784 | int ret = 0; |
2785 | ||
0b246afa | 2786 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2787 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2788 | if (num_heads > 1) |
0b246afa | 2789 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2790 | num_bytes <<= 1; |
2ff7e61e JM |
2791 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2792 | fs_info->nodesize; | |
0b246afa | 2793 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2794 | num_dirty_bgs); |
0b246afa | 2795 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2796 | |
2797 | /* | |
2798 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2799 | * wiggle room since running delayed refs can create more delayed refs. | |
2800 | */ | |
cb723e49 JB |
2801 | if (global_rsv->space_info->full) { |
2802 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2803 | num_bytes <<= 1; |
cb723e49 | 2804 | } |
1be41b78 JB |
2805 | |
2806 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2807 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2808 | ret = 1; |
2809 | spin_unlock(&global_rsv->lock); | |
2810 | return ret; | |
2811 | } | |
2812 | ||
0a2b2a84 | 2813 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2814 | struct btrfs_fs_info *fs_info) |
0a2b2a84 | 2815 | { |
0a2b2a84 JB |
2816 | u64 num_entries = |
2817 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2818 | u64 avg_runtime; | |
a79b7d4b | 2819 | u64 val; |
0a2b2a84 JB |
2820 | |
2821 | smp_mb(); | |
2822 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2823 | val = num_entries * avg_runtime; |
dc1a90c6 | 2824 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2825 | return 1; |
a79b7d4b CM |
2826 | if (val >= NSEC_PER_SEC / 2) |
2827 | return 2; | |
0a2b2a84 | 2828 | |
2ff7e61e | 2829 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
0a2b2a84 JB |
2830 | } |
2831 | ||
a79b7d4b CM |
2832 | struct async_delayed_refs { |
2833 | struct btrfs_root *root; | |
31b9655f | 2834 | u64 transid; |
a79b7d4b CM |
2835 | int count; |
2836 | int error; | |
2837 | int sync; | |
2838 | struct completion wait; | |
2839 | struct btrfs_work work; | |
2840 | }; | |
2841 | ||
2ff7e61e JM |
2842 | static inline struct async_delayed_refs * |
2843 | to_async_delayed_refs(struct btrfs_work *work) | |
2844 | { | |
2845 | return container_of(work, struct async_delayed_refs, work); | |
2846 | } | |
2847 | ||
a79b7d4b CM |
2848 | static void delayed_ref_async_start(struct btrfs_work *work) |
2849 | { | |
2ff7e61e | 2850 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2851 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2852 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2853 | int ret; |
2854 | ||
0f873eca | 2855 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2856 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2857 | goto done; |
31b9655f | 2858 | |
0f873eca CM |
2859 | trans = btrfs_join_transaction(async->root); |
2860 | if (IS_ERR(trans)) { | |
2861 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2862 | goto done; |
2863 | } | |
2864 | ||
2865 | /* | |
01327610 | 2866 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2867 | * wait on delayed refs |
2868 | */ | |
2869 | trans->sync = true; | |
0f873eca CM |
2870 | |
2871 | /* Don't bother flushing if we got into a different transaction */ | |
2872 | if (trans->transid > async->transid) | |
2873 | goto end; | |
2874 | ||
2ff7e61e | 2875 | ret = btrfs_run_delayed_refs(trans, fs_info, async->count); |
a79b7d4b CM |
2876 | if (ret) |
2877 | async->error = ret; | |
0f873eca | 2878 | end: |
3a45bb20 | 2879 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
2880 | if (ret && !async->error) |
2881 | async->error = ret; | |
2882 | done: | |
2883 | if (async->sync) | |
2884 | complete(&async->wait); | |
2885 | else | |
2886 | kfree(async); | |
2887 | } | |
2888 | ||
2ff7e61e | 2889 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 2890 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
2891 | { |
2892 | struct async_delayed_refs *async; | |
2893 | int ret; | |
2894 | ||
2895 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2896 | if (!async) | |
2897 | return -ENOMEM; | |
2898 | ||
0b246afa | 2899 | async->root = fs_info->tree_root; |
a79b7d4b CM |
2900 | async->count = count; |
2901 | async->error = 0; | |
31b9655f | 2902 | async->transid = transid; |
a79b7d4b CM |
2903 | if (wait) |
2904 | async->sync = 1; | |
2905 | else | |
2906 | async->sync = 0; | |
2907 | init_completion(&async->wait); | |
2908 | ||
9e0af237 LB |
2909 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
2910 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 2911 | |
0b246afa | 2912 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
2913 | |
2914 | if (wait) { | |
2915 | wait_for_completion(&async->wait); | |
2916 | ret = async->error; | |
2917 | kfree(async); | |
2918 | return ret; | |
2919 | } | |
2920 | return 0; | |
2921 | } | |
2922 | ||
c3e69d58 CM |
2923 | /* |
2924 | * this starts processing the delayed reference count updates and | |
2925 | * extent insertions we have queued up so far. count can be | |
2926 | * 0, which means to process everything in the tree at the start | |
2927 | * of the run (but not newly added entries), or it can be some target | |
2928 | * number you'd like to process. | |
79787eaa JM |
2929 | * |
2930 | * Returns 0 on success or if called with an aborted transaction | |
2931 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2932 | */ |
2933 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2934 | struct btrfs_fs_info *fs_info, unsigned long count) |
c3e69d58 CM |
2935 | { |
2936 | struct rb_node *node; | |
2937 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2938 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2939 | int ret; |
2940 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 2941 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 2942 | |
79787eaa JM |
2943 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2944 | if (trans->aborted) | |
2945 | return 0; | |
2946 | ||
0b246afa | 2947 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2948 | return 0; |
2949 | ||
c3e69d58 | 2950 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2951 | if (count == 0) |
d7df2c79 | 2952 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2953 | |
c3e69d58 | 2954 | again: |
709c0486 AJ |
2955 | #ifdef SCRAMBLE_DELAYED_REFS |
2956 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2957 | #endif | |
d9a0540a | 2958 | trans->can_flush_pending_bgs = false; |
2ff7e61e | 2959 | ret = __btrfs_run_delayed_refs(trans, fs_info, count); |
d7df2c79 | 2960 | if (ret < 0) { |
66642832 | 2961 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2962 | return ret; |
eb099670 | 2963 | } |
c3e69d58 | 2964 | |
56bec294 | 2965 | if (run_all) { |
d7df2c79 | 2966 | if (!list_empty(&trans->new_bgs)) |
2ff7e61e | 2967 | btrfs_create_pending_block_groups(trans, fs_info); |
ea658bad | 2968 | |
d7df2c79 | 2969 | spin_lock(&delayed_refs->lock); |
c46effa6 | 2970 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
2971 | if (!node) { |
2972 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2973 | goto out; |
d7df2c79 | 2974 | } |
e9d0b13b | 2975 | |
56bec294 | 2976 | while (node) { |
c46effa6 LB |
2977 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2978 | href_node); | |
2979 | if (btrfs_delayed_ref_is_head(&head->node)) { | |
2980 | struct btrfs_delayed_ref_node *ref; | |
5caf2a00 | 2981 | |
c46effa6 | 2982 | ref = &head->node; |
56bec294 CM |
2983 | atomic_inc(&ref->refs); |
2984 | ||
2985 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2986 | /* |
2987 | * Mutex was contended, block until it's | |
2988 | * released and try again | |
2989 | */ | |
56bec294 CM |
2990 | mutex_lock(&head->mutex); |
2991 | mutex_unlock(&head->mutex); | |
2992 | ||
2993 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2994 | cond_resched(); |
56bec294 | 2995 | goto again; |
c46effa6 LB |
2996 | } else { |
2997 | WARN_ON(1); | |
56bec294 CM |
2998 | } |
2999 | node = rb_next(node); | |
3000 | } | |
3001 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 3002 | cond_resched(); |
56bec294 | 3003 | goto again; |
5f39d397 | 3004 | } |
54aa1f4d | 3005 | out: |
edf39272 | 3006 | assert_qgroups_uptodate(trans); |
d9a0540a | 3007 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3008 | return 0; |
3009 | } | |
3010 | ||
5d4f98a2 | 3011 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 3012 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 3013 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 3014 | int level, int is_data) |
5d4f98a2 YZ |
3015 | { |
3016 | struct btrfs_delayed_extent_op *extent_op; | |
3017 | int ret; | |
3018 | ||
78a6184a | 3019 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3020 | if (!extent_op) |
3021 | return -ENOMEM; | |
3022 | ||
3023 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3024 | extent_op->update_flags = true; |
3025 | extent_op->update_key = false; | |
3026 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3027 | extent_op->level = level; |
5d4f98a2 | 3028 | |
0b246afa | 3029 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 3030 | num_bytes, extent_op); |
5d4f98a2 | 3031 | if (ret) |
78a6184a | 3032 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3033 | return ret; |
3034 | } | |
3035 | ||
e4c3b2dc | 3036 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3037 | struct btrfs_path *path, |
3038 | u64 objectid, u64 offset, u64 bytenr) | |
3039 | { | |
3040 | struct btrfs_delayed_ref_head *head; | |
3041 | struct btrfs_delayed_ref_node *ref; | |
3042 | struct btrfs_delayed_data_ref *data_ref; | |
3043 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 3044 | struct btrfs_transaction *cur_trans; |
5d4f98a2 YZ |
3045 | int ret = 0; |
3046 | ||
e4c3b2dc LB |
3047 | cur_trans = root->fs_info->running_transaction; |
3048 | if (!cur_trans) | |
3049 | return 0; | |
3050 | ||
3051 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3052 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3053 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3054 | if (!head) { |
3055 | spin_unlock(&delayed_refs->lock); | |
3056 | return 0; | |
3057 | } | |
5d4f98a2 YZ |
3058 | |
3059 | if (!mutex_trylock(&head->mutex)) { | |
3060 | atomic_inc(&head->node.refs); | |
3061 | spin_unlock(&delayed_refs->lock); | |
3062 | ||
b3b4aa74 | 3063 | btrfs_release_path(path); |
5d4f98a2 | 3064 | |
8cc33e5c DS |
3065 | /* |
3066 | * Mutex was contended, block until it's released and let | |
3067 | * caller try again | |
3068 | */ | |
5d4f98a2 YZ |
3069 | mutex_lock(&head->mutex); |
3070 | mutex_unlock(&head->mutex); | |
3071 | btrfs_put_delayed_ref(&head->node); | |
3072 | return -EAGAIN; | |
3073 | } | |
d7df2c79 | 3074 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3075 | |
d7df2c79 | 3076 | spin_lock(&head->lock); |
c6fc2454 | 3077 | list_for_each_entry(ref, &head->ref_list, list) { |
d7df2c79 JB |
3078 | /* If it's a shared ref we know a cross reference exists */ |
3079 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3080 | ret = 1; | |
3081 | break; | |
3082 | } | |
5d4f98a2 | 3083 | |
d7df2c79 | 3084 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3085 | |
d7df2c79 JB |
3086 | /* |
3087 | * If our ref doesn't match the one we're currently looking at | |
3088 | * then we have a cross reference. | |
3089 | */ | |
3090 | if (data_ref->root != root->root_key.objectid || | |
3091 | data_ref->objectid != objectid || | |
3092 | data_ref->offset != offset) { | |
3093 | ret = 1; | |
3094 | break; | |
3095 | } | |
5d4f98a2 | 3096 | } |
d7df2c79 | 3097 | spin_unlock(&head->lock); |
5d4f98a2 | 3098 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
3099 | return ret; |
3100 | } | |
3101 | ||
e4c3b2dc | 3102 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3103 | struct btrfs_path *path, |
3104 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3105 | { |
0b246afa JM |
3106 | struct btrfs_fs_info *fs_info = root->fs_info; |
3107 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3108 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3109 | struct btrfs_extent_data_ref *ref; |
3110 | struct btrfs_extent_inline_ref *iref; | |
3111 | struct btrfs_extent_item *ei; | |
f321e491 | 3112 | struct btrfs_key key; |
5d4f98a2 | 3113 | u32 item_size; |
be20aa9d | 3114 | int ret; |
925baedd | 3115 | |
be20aa9d | 3116 | key.objectid = bytenr; |
31840ae1 | 3117 | key.offset = (u64)-1; |
f321e491 | 3118 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3119 | |
be20aa9d CM |
3120 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3121 | if (ret < 0) | |
3122 | goto out; | |
79787eaa | 3123 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3124 | |
3125 | ret = -ENOENT; | |
3126 | if (path->slots[0] == 0) | |
31840ae1 | 3127 | goto out; |
be20aa9d | 3128 | |
31840ae1 | 3129 | path->slots[0]--; |
f321e491 | 3130 | leaf = path->nodes[0]; |
5d4f98a2 | 3131 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3132 | |
5d4f98a2 | 3133 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3134 | goto out; |
f321e491 | 3135 | |
5d4f98a2 YZ |
3136 | ret = 1; |
3137 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3138 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3139 | if (item_size < sizeof(*ei)) { | |
3140 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3141 | goto out; | |
3142 | } | |
3143 | #endif | |
3144 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3145 | |
5d4f98a2 YZ |
3146 | if (item_size != sizeof(*ei) + |
3147 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3148 | goto out; | |
be20aa9d | 3149 | |
5d4f98a2 YZ |
3150 | if (btrfs_extent_generation(leaf, ei) <= |
3151 | btrfs_root_last_snapshot(&root->root_item)) | |
3152 | goto out; | |
3153 | ||
3154 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3155 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
3156 | BTRFS_EXTENT_DATA_REF_KEY) | |
3157 | goto out; | |
3158 | ||
3159 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3160 | if (btrfs_extent_refs(leaf, ei) != | |
3161 | btrfs_extent_data_ref_count(leaf, ref) || | |
3162 | btrfs_extent_data_ref_root(leaf, ref) != | |
3163 | root->root_key.objectid || | |
3164 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3165 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3166 | goto out; | |
3167 | ||
3168 | ret = 0; | |
3169 | out: | |
3170 | return ret; | |
3171 | } | |
3172 | ||
e4c3b2dc LB |
3173 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3174 | u64 bytenr) | |
5d4f98a2 YZ |
3175 | { |
3176 | struct btrfs_path *path; | |
3177 | int ret; | |
3178 | int ret2; | |
3179 | ||
3180 | path = btrfs_alloc_path(); | |
3181 | if (!path) | |
3182 | return -ENOENT; | |
3183 | ||
3184 | do { | |
e4c3b2dc | 3185 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3186 | offset, bytenr); |
3187 | if (ret && ret != -ENOENT) | |
f321e491 | 3188 | goto out; |
80ff3856 | 3189 | |
e4c3b2dc | 3190 | ret2 = check_delayed_ref(root, path, objectid, |
5d4f98a2 YZ |
3191 | offset, bytenr); |
3192 | } while (ret2 == -EAGAIN); | |
3193 | ||
3194 | if (ret2 && ret2 != -ENOENT) { | |
3195 | ret = ret2; | |
3196 | goto out; | |
f321e491 | 3197 | } |
5d4f98a2 YZ |
3198 | |
3199 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3200 | ret = 0; | |
be20aa9d | 3201 | out: |
80ff3856 | 3202 | btrfs_free_path(path); |
f0486c68 YZ |
3203 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3204 | WARN_ON(ret > 0); | |
f321e491 | 3205 | return ret; |
be20aa9d | 3206 | } |
c5739bba | 3207 | |
5d4f98a2 | 3208 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3209 | struct btrfs_root *root, |
5d4f98a2 | 3210 | struct extent_buffer *buf, |
e339a6b0 | 3211 | int full_backref, int inc) |
31840ae1 | 3212 | { |
0b246afa | 3213 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3214 | u64 bytenr; |
5d4f98a2 YZ |
3215 | u64 num_bytes; |
3216 | u64 parent; | |
31840ae1 | 3217 | u64 ref_root; |
31840ae1 | 3218 | u32 nritems; |
31840ae1 ZY |
3219 | struct btrfs_key key; |
3220 | struct btrfs_file_extent_item *fi; | |
3221 | int i; | |
3222 | int level; | |
3223 | int ret = 0; | |
2ff7e61e JM |
3224 | int (*process_func)(struct btrfs_trans_handle *, |
3225 | struct btrfs_fs_info *, | |
b06c4bf5 | 3226 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3227 | |
fccb84c9 | 3228 | |
0b246afa | 3229 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3230 | return 0; |
fccb84c9 | 3231 | |
31840ae1 | 3232 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3233 | nritems = btrfs_header_nritems(buf); |
3234 | level = btrfs_header_level(buf); | |
3235 | ||
27cdeb70 | 3236 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3237 | return 0; |
31840ae1 | 3238 | |
5d4f98a2 YZ |
3239 | if (inc) |
3240 | process_func = btrfs_inc_extent_ref; | |
3241 | else | |
3242 | process_func = btrfs_free_extent; | |
31840ae1 | 3243 | |
5d4f98a2 YZ |
3244 | if (full_backref) |
3245 | parent = buf->start; | |
3246 | else | |
3247 | parent = 0; | |
3248 | ||
3249 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3250 | if (level == 0) { |
5d4f98a2 | 3251 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3252 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3253 | continue; |
5d4f98a2 | 3254 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3255 | struct btrfs_file_extent_item); |
3256 | if (btrfs_file_extent_type(buf, fi) == | |
3257 | BTRFS_FILE_EXTENT_INLINE) | |
3258 | continue; | |
3259 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3260 | if (bytenr == 0) | |
3261 | continue; | |
5d4f98a2 YZ |
3262 | |
3263 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3264 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2ff7e61e | 3265 | ret = process_func(trans, fs_info, bytenr, num_bytes, |
5d4f98a2 | 3266 | parent, ref_root, key.objectid, |
b06c4bf5 | 3267 | key.offset); |
31840ae1 ZY |
3268 | if (ret) |
3269 | goto fail; | |
3270 | } else { | |
5d4f98a2 | 3271 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3272 | num_bytes = fs_info->nodesize; |
2ff7e61e | 3273 | ret = process_func(trans, fs_info, bytenr, num_bytes, |
b06c4bf5 | 3274 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3275 | if (ret) |
3276 | goto fail; | |
3277 | } | |
3278 | } | |
3279 | return 0; | |
3280 | fail: | |
5d4f98a2 YZ |
3281 | return ret; |
3282 | } | |
3283 | ||
3284 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3285 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3286 | { |
e339a6b0 | 3287 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3288 | } |
3289 | ||
3290 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3291 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3292 | { |
e339a6b0 | 3293 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3294 | } |
3295 | ||
9078a3e1 | 3296 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3297 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3298 | struct btrfs_path *path, |
3299 | struct btrfs_block_group_cache *cache) | |
3300 | { | |
3301 | int ret; | |
0b246afa | 3302 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3303 | unsigned long bi; |
3304 | struct extent_buffer *leaf; | |
9078a3e1 | 3305 | |
9078a3e1 | 3306 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3307 | if (ret) { |
3308 | if (ret > 0) | |
3309 | ret = -ENOENT; | |
54aa1f4d | 3310 | goto fail; |
df95e7f0 | 3311 | } |
5f39d397 CM |
3312 | |
3313 | leaf = path->nodes[0]; | |
3314 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3315 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3316 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3317 | fail: |
24b89d08 | 3318 | btrfs_release_path(path); |
df95e7f0 | 3319 | return ret; |
9078a3e1 CM |
3320 | |
3321 | } | |
3322 | ||
4a8c9a62 | 3323 | static struct btrfs_block_group_cache * |
2ff7e61e | 3324 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3325 | struct btrfs_block_group_cache *cache) |
3326 | { | |
3327 | struct rb_node *node; | |
292cbd51 | 3328 | |
0b246afa | 3329 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3330 | |
3331 | /* If our block group was removed, we need a full search. */ | |
3332 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3333 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3334 | ||
0b246afa | 3335 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3336 | btrfs_put_block_group(cache); |
0b246afa | 3337 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3338 | } |
4a8c9a62 YZ |
3339 | node = rb_next(&cache->cache_node); |
3340 | btrfs_put_block_group(cache); | |
3341 | if (node) { | |
3342 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3343 | cache_node); | |
11dfe35a | 3344 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3345 | } else |
3346 | cache = NULL; | |
0b246afa | 3347 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3348 | return cache; |
3349 | } | |
3350 | ||
0af3d00b JB |
3351 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3352 | struct btrfs_trans_handle *trans, | |
3353 | struct btrfs_path *path) | |
3354 | { | |
0b246afa JM |
3355 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3356 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b JB |
3357 | struct inode *inode = NULL; |
3358 | u64 alloc_hint = 0; | |
2b20982e | 3359 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3360 | u64 num_pages = 0; |
0af3d00b JB |
3361 | int retries = 0; |
3362 | int ret = 0; | |
3363 | ||
3364 | /* | |
3365 | * If this block group is smaller than 100 megs don't bother caching the | |
3366 | * block group. | |
3367 | */ | |
ee22184b | 3368 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3369 | spin_lock(&block_group->lock); |
3370 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3371 | spin_unlock(&block_group->lock); | |
3372 | return 0; | |
3373 | } | |
3374 | ||
0c0ef4bc JB |
3375 | if (trans->aborted) |
3376 | return 0; | |
0af3d00b | 3377 | again: |
77ab86bf | 3378 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3379 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3380 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3381 | btrfs_release_path(path); |
0af3d00b JB |
3382 | goto out; |
3383 | } | |
3384 | ||
3385 | if (IS_ERR(inode)) { | |
3386 | BUG_ON(retries); | |
3387 | retries++; | |
3388 | ||
3389 | if (block_group->ro) | |
3390 | goto out_free; | |
3391 | ||
77ab86bf JM |
3392 | ret = create_free_space_inode(fs_info, trans, block_group, |
3393 | path); | |
0af3d00b JB |
3394 | if (ret) |
3395 | goto out_free; | |
3396 | goto again; | |
3397 | } | |
3398 | ||
5b0e95bf JB |
3399 | /* We've already setup this transaction, go ahead and exit */ |
3400 | if (block_group->cache_generation == trans->transid && | |
3401 | i_size_read(inode)) { | |
3402 | dcs = BTRFS_DC_SETUP; | |
3403 | goto out_put; | |
3404 | } | |
3405 | ||
0af3d00b JB |
3406 | /* |
3407 | * We want to set the generation to 0, that way if anything goes wrong | |
3408 | * from here on out we know not to trust this cache when we load up next | |
3409 | * time. | |
3410 | */ | |
3411 | BTRFS_I(inode)->generation = 0; | |
3412 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3413 | if (ret) { |
3414 | /* | |
3415 | * So theoretically we could recover from this, simply set the | |
3416 | * super cache generation to 0 so we know to invalidate the | |
3417 | * cache, but then we'd have to keep track of the block groups | |
3418 | * that fail this way so we know we _have_ to reset this cache | |
3419 | * before the next commit or risk reading stale cache. So to | |
3420 | * limit our exposure to horrible edge cases lets just abort the | |
3421 | * transaction, this only happens in really bad situations | |
3422 | * anyway. | |
3423 | */ | |
66642832 | 3424 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3425 | goto out_put; |
3426 | } | |
0af3d00b JB |
3427 | WARN_ON(ret); |
3428 | ||
3429 | if (i_size_read(inode) > 0) { | |
2ff7e61e | 3430 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3431 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3432 | if (ret) |
3433 | goto out_put; | |
3434 | ||
77ab86bf | 3435 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3436 | if (ret) |
3437 | goto out_put; | |
3438 | } | |
3439 | ||
3440 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3441 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3442 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3443 | /* |
3444 | * don't bother trying to write stuff out _if_ | |
3445 | * a) we're not cached, | |
3446 | * b) we're with nospace_cache mount option. | |
3447 | */ | |
2b20982e | 3448 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3449 | spin_unlock(&block_group->lock); |
3450 | goto out_put; | |
3451 | } | |
3452 | spin_unlock(&block_group->lock); | |
3453 | ||
2968b1f4 JB |
3454 | /* |
3455 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3456 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3457 | */ | |
3458 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3459 | ret = -ENOSPC; | |
3460 | goto out_put; | |
3461 | } | |
3462 | ||
6fc823b1 JB |
3463 | /* |
3464 | * Try to preallocate enough space based on how big the block group is. | |
3465 | * Keep in mind this has to include any pinned space which could end up | |
3466 | * taking up quite a bit since it's not folded into the other space | |
3467 | * cache. | |
3468 | */ | |
ee22184b | 3469 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3470 | if (!num_pages) |
3471 | num_pages = 1; | |
3472 | ||
0af3d00b | 3473 | num_pages *= 16; |
09cbfeaf | 3474 | num_pages *= PAGE_SIZE; |
0af3d00b | 3475 | |
7cf5b976 | 3476 | ret = btrfs_check_data_free_space(inode, 0, num_pages); |
0af3d00b JB |
3477 | if (ret) |
3478 | goto out_put; | |
3479 | ||
3480 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3481 | num_pages, num_pages, | |
3482 | &alloc_hint); | |
2968b1f4 JB |
3483 | /* |
3484 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3485 | * of metadata or split extents when writing the cache out, which means | |
3486 | * we can enospc if we are heavily fragmented in addition to just normal | |
3487 | * out of space conditions. So if we hit this just skip setting up any | |
3488 | * other block groups for this transaction, maybe we'll unpin enough | |
3489 | * space the next time around. | |
3490 | */ | |
2b20982e JB |
3491 | if (!ret) |
3492 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3493 | else if (ret == -ENOSPC) |
3494 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3495 | |
0af3d00b JB |
3496 | out_put: |
3497 | iput(inode); | |
3498 | out_free: | |
b3b4aa74 | 3499 | btrfs_release_path(path); |
0af3d00b JB |
3500 | out: |
3501 | spin_lock(&block_group->lock); | |
e65cbb94 | 3502 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3503 | block_group->cache_generation = trans->transid; |
2b20982e | 3504 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3505 | spin_unlock(&block_group->lock); |
3506 | ||
3507 | return ret; | |
3508 | } | |
3509 | ||
dcdf7f6d | 3510 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3511 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3512 | { |
3513 | struct btrfs_block_group_cache *cache, *tmp; | |
3514 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3515 | struct btrfs_path *path; | |
3516 | ||
3517 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3518 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3519 | return 0; |
3520 | ||
3521 | path = btrfs_alloc_path(); | |
3522 | if (!path) | |
3523 | return -ENOMEM; | |
3524 | ||
3525 | /* Could add new block groups, use _safe just in case */ | |
3526 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3527 | dirty_list) { | |
3528 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3529 | cache_save_setup(cache, trans, path); | |
3530 | } | |
3531 | ||
3532 | btrfs_free_path(path); | |
3533 | return 0; | |
3534 | } | |
3535 | ||
1bbc621e CM |
3536 | /* |
3537 | * transaction commit does final block group cache writeback during a | |
3538 | * critical section where nothing is allowed to change the FS. This is | |
3539 | * required in order for the cache to actually match the block group, | |
3540 | * but can introduce a lot of latency into the commit. | |
3541 | * | |
3542 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3543 | * cache IO. There's a chance we'll have to redo some of it if the | |
3544 | * block group changes again during the commit, but it greatly reduces | |
3545 | * the commit latency by getting rid of the easy block groups while | |
3546 | * we're still allowing others to join the commit. | |
3547 | */ | |
3548 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3549 | struct btrfs_fs_info *fs_info) |
9078a3e1 | 3550 | { |
4a8c9a62 | 3551 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3552 | struct btrfs_transaction *cur_trans = trans->transaction; |
3553 | int ret = 0; | |
c9dc4c65 | 3554 | int should_put; |
1bbc621e CM |
3555 | struct btrfs_path *path = NULL; |
3556 | LIST_HEAD(dirty); | |
3557 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3558 | int num_started = 0; |
1bbc621e CM |
3559 | int loops = 0; |
3560 | ||
3561 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3562 | if (list_empty(&cur_trans->dirty_bgs)) { |
3563 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3564 | return 0; | |
1bbc621e | 3565 | } |
b58d1a9e | 3566 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3567 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3568 | |
1bbc621e | 3569 | again: |
1bbc621e CM |
3570 | /* |
3571 | * make sure all the block groups on our dirty list actually | |
3572 | * exist | |
3573 | */ | |
2ff7e61e | 3574 | btrfs_create_pending_block_groups(trans, fs_info); |
1bbc621e CM |
3575 | |
3576 | if (!path) { | |
3577 | path = btrfs_alloc_path(); | |
3578 | if (!path) | |
3579 | return -ENOMEM; | |
3580 | } | |
3581 | ||
b58d1a9e FM |
3582 | /* |
3583 | * cache_write_mutex is here only to save us from balance or automatic | |
3584 | * removal of empty block groups deleting this block group while we are | |
3585 | * writing out the cache | |
3586 | */ | |
3587 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3588 | while (!list_empty(&dirty)) { |
3589 | cache = list_first_entry(&dirty, | |
3590 | struct btrfs_block_group_cache, | |
3591 | dirty_list); | |
1bbc621e CM |
3592 | /* |
3593 | * this can happen if something re-dirties a block | |
3594 | * group that is already under IO. Just wait for it to | |
3595 | * finish and then do it all again | |
3596 | */ | |
3597 | if (!list_empty(&cache->io_list)) { | |
3598 | list_del_init(&cache->io_list); | |
afdb5718 | 3599 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3600 | btrfs_put_block_group(cache); |
3601 | } | |
3602 | ||
3603 | ||
3604 | /* | |
3605 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3606 | * if it should update the cache_state. Don't delete | |
3607 | * until after we wait. | |
3608 | * | |
3609 | * Since we're not running in the commit critical section | |
3610 | * we need the dirty_bgs_lock to protect from update_block_group | |
3611 | */ | |
3612 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3613 | list_del_init(&cache->dirty_list); | |
3614 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3615 | ||
3616 | should_put = 1; | |
3617 | ||
3618 | cache_save_setup(cache, trans, path); | |
3619 | ||
3620 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3621 | cache->io_ctl.inode = NULL; | |
0b246afa | 3622 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3623 | cache, path); |
1bbc621e CM |
3624 | if (ret == 0 && cache->io_ctl.inode) { |
3625 | num_started++; | |
3626 | should_put = 0; | |
3627 | ||
3628 | /* | |
3629 | * the cache_write_mutex is protecting | |
3630 | * the io_list | |
3631 | */ | |
3632 | list_add_tail(&cache->io_list, io); | |
3633 | } else { | |
3634 | /* | |
3635 | * if we failed to write the cache, the | |
3636 | * generation will be bad and life goes on | |
3637 | */ | |
3638 | ret = 0; | |
3639 | } | |
3640 | } | |
ff1f8250 | 3641 | if (!ret) { |
2ff7e61e JM |
3642 | ret = write_one_cache_group(trans, fs_info, |
3643 | path, cache); | |
ff1f8250 FM |
3644 | /* |
3645 | * Our block group might still be attached to the list | |
3646 | * of new block groups in the transaction handle of some | |
3647 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3648 | * means its block group item isn't yet in the extent | |
3649 | * tree. If this happens ignore the error, as we will | |
3650 | * try again later in the critical section of the | |
3651 | * transaction commit. | |
3652 | */ | |
3653 | if (ret == -ENOENT) { | |
3654 | ret = 0; | |
3655 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3656 | if (list_empty(&cache->dirty_list)) { | |
3657 | list_add_tail(&cache->dirty_list, | |
3658 | &cur_trans->dirty_bgs); | |
3659 | btrfs_get_block_group(cache); | |
3660 | } | |
3661 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3662 | } else if (ret) { | |
66642832 | 3663 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3664 | } |
3665 | } | |
1bbc621e CM |
3666 | |
3667 | /* if its not on the io list, we need to put the block group */ | |
3668 | if (should_put) | |
3669 | btrfs_put_block_group(cache); | |
3670 | ||
3671 | if (ret) | |
3672 | break; | |
b58d1a9e FM |
3673 | |
3674 | /* | |
3675 | * Avoid blocking other tasks for too long. It might even save | |
3676 | * us from writing caches for block groups that are going to be | |
3677 | * removed. | |
3678 | */ | |
3679 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3680 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3681 | } |
b58d1a9e | 3682 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3683 | |
3684 | /* | |
3685 | * go through delayed refs for all the stuff we've just kicked off | |
3686 | * and then loop back (just once) | |
3687 | */ | |
2ff7e61e | 3688 | ret = btrfs_run_delayed_refs(trans, fs_info, 0); |
1bbc621e CM |
3689 | if (!ret && loops == 0) { |
3690 | loops++; | |
3691 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3692 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3693 | /* |
3694 | * dirty_bgs_lock protects us from concurrent block group | |
3695 | * deletes too (not just cache_write_mutex). | |
3696 | */ | |
3697 | if (!list_empty(&dirty)) { | |
3698 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3699 | goto again; | |
3700 | } | |
1bbc621e | 3701 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3702 | } else if (ret < 0) { |
2ff7e61e | 3703 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3704 | } |
3705 | ||
3706 | btrfs_free_path(path); | |
3707 | return ret; | |
3708 | } | |
3709 | ||
3710 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3711 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3712 | { |
3713 | struct btrfs_block_group_cache *cache; | |
3714 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3715 | int ret = 0; | |
3716 | int should_put; | |
3717 | struct btrfs_path *path; | |
3718 | struct list_head *io = &cur_trans->io_bgs; | |
3719 | int num_started = 0; | |
9078a3e1 CM |
3720 | |
3721 | path = btrfs_alloc_path(); | |
3722 | if (!path) | |
3723 | return -ENOMEM; | |
3724 | ||
ce93ec54 | 3725 | /* |
e44081ef FM |
3726 | * Even though we are in the critical section of the transaction commit, |
3727 | * we can still have concurrent tasks adding elements to this | |
3728 | * transaction's list of dirty block groups. These tasks correspond to | |
3729 | * endio free space workers started when writeback finishes for a | |
3730 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3731 | * allocate new block groups as a result of COWing nodes of the root | |
3732 | * tree when updating the free space inode. The writeback for the space | |
3733 | * caches is triggered by an earlier call to | |
3734 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3735 | * loop. | |
3736 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3737 | * delayed refs to make sure we have the best chance at doing this all |
3738 | * in one shot. | |
3739 | */ | |
e44081ef | 3740 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3741 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3742 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3743 | struct btrfs_block_group_cache, | |
3744 | dirty_list); | |
c9dc4c65 CM |
3745 | |
3746 | /* | |
3747 | * this can happen if cache_save_setup re-dirties a block | |
3748 | * group that is already under IO. Just wait for it to | |
3749 | * finish and then do it all again | |
3750 | */ | |
3751 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3752 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3753 | list_del_init(&cache->io_list); |
afdb5718 | 3754 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3755 | btrfs_put_block_group(cache); |
e44081ef | 3756 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3757 | } |
3758 | ||
1bbc621e CM |
3759 | /* |
3760 | * don't remove from the dirty list until after we've waited | |
3761 | * on any pending IO | |
3762 | */ | |
ce93ec54 | 3763 | list_del_init(&cache->dirty_list); |
e44081ef | 3764 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3765 | should_put = 1; |
3766 | ||
1bbc621e | 3767 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3768 | |
ce93ec54 | 3769 | if (!ret) |
2ff7e61e JM |
3770 | ret = btrfs_run_delayed_refs(trans, fs_info, |
3771 | (unsigned long) -1); | |
c9dc4c65 CM |
3772 | |
3773 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3774 | cache->io_ctl.inode = NULL; | |
0b246afa | 3775 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3776 | cache, path); |
c9dc4c65 CM |
3777 | if (ret == 0 && cache->io_ctl.inode) { |
3778 | num_started++; | |
3779 | should_put = 0; | |
1bbc621e | 3780 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3781 | } else { |
3782 | /* | |
3783 | * if we failed to write the cache, the | |
3784 | * generation will be bad and life goes on | |
3785 | */ | |
3786 | ret = 0; | |
3787 | } | |
3788 | } | |
ff1f8250 | 3789 | if (!ret) { |
2ff7e61e JM |
3790 | ret = write_one_cache_group(trans, fs_info, |
3791 | path, cache); | |
2bc0bb5f FM |
3792 | /* |
3793 | * One of the free space endio workers might have | |
3794 | * created a new block group while updating a free space | |
3795 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3796 | * and hasn't released its transaction handle yet, in | |
3797 | * which case the new block group is still attached to | |
3798 | * its transaction handle and its creation has not | |
3799 | * finished yet (no block group item in the extent tree | |
3800 | * yet, etc). If this is the case, wait for all free | |
3801 | * space endio workers to finish and retry. This is a | |
3802 | * a very rare case so no need for a more efficient and | |
3803 | * complex approach. | |
3804 | */ | |
3805 | if (ret == -ENOENT) { | |
3806 | wait_event(cur_trans->writer_wait, | |
3807 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3808 | ret = write_one_cache_group(trans, fs_info, |
3809 | path, cache); | |
2bc0bb5f | 3810 | } |
ff1f8250 | 3811 | if (ret) |
66642832 | 3812 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3813 | } |
c9dc4c65 CM |
3814 | |
3815 | /* if its not on the io list, we need to put the block group */ | |
3816 | if (should_put) | |
3817 | btrfs_put_block_group(cache); | |
e44081ef | 3818 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3819 | } |
e44081ef | 3820 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3821 | |
1bbc621e CM |
3822 | while (!list_empty(io)) { |
3823 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3824 | io_list); |
3825 | list_del_init(&cache->io_list); | |
afdb5718 | 3826 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3827 | btrfs_put_block_group(cache); |
3828 | } | |
3829 | ||
9078a3e1 | 3830 | btrfs_free_path(path); |
ce93ec54 | 3831 | return ret; |
9078a3e1 CM |
3832 | } |
3833 | ||
2ff7e61e | 3834 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3835 | { |
3836 | struct btrfs_block_group_cache *block_group; | |
3837 | int readonly = 0; | |
3838 | ||
0b246afa | 3839 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3840 | if (!block_group || block_group->ro) |
3841 | readonly = 1; | |
3842 | if (block_group) | |
fa9c0d79 | 3843 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3844 | return readonly; |
3845 | } | |
3846 | ||
f78c436c FM |
3847 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3848 | { | |
3849 | struct btrfs_block_group_cache *bg; | |
3850 | bool ret = true; | |
3851 | ||
3852 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3853 | if (!bg) | |
3854 | return false; | |
3855 | ||
3856 | spin_lock(&bg->lock); | |
3857 | if (bg->ro) | |
3858 | ret = false; | |
3859 | else | |
3860 | atomic_inc(&bg->nocow_writers); | |
3861 | spin_unlock(&bg->lock); | |
3862 | ||
3863 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3864 | if (!ret) | |
3865 | btrfs_put_block_group(bg); | |
3866 | ||
3867 | return ret; | |
3868 | ||
3869 | } | |
3870 | ||
3871 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3872 | { | |
3873 | struct btrfs_block_group_cache *bg; | |
3874 | ||
3875 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3876 | ASSERT(bg); | |
3877 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
3878 | wake_up_atomic_t(&bg->nocow_writers); | |
3879 | /* | |
3880 | * Once for our lookup and once for the lookup done by a previous call | |
3881 | * to btrfs_inc_nocow_writers() | |
3882 | */ | |
3883 | btrfs_put_block_group(bg); | |
3884 | btrfs_put_block_group(bg); | |
3885 | } | |
3886 | ||
3887 | static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a) | |
3888 | { | |
3889 | schedule(); | |
3890 | return 0; | |
3891 | } | |
3892 | ||
3893 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) | |
3894 | { | |
3895 | wait_on_atomic_t(&bg->nocow_writers, | |
3896 | btrfs_wait_nocow_writers_atomic_t, | |
3897 | TASK_UNINTERRUPTIBLE); | |
3898 | } | |
3899 | ||
6ab0a202 JM |
3900 | static const char *alloc_name(u64 flags) |
3901 | { | |
3902 | switch (flags) { | |
3903 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3904 | return "mixed"; | |
3905 | case BTRFS_BLOCK_GROUP_METADATA: | |
3906 | return "metadata"; | |
3907 | case BTRFS_BLOCK_GROUP_DATA: | |
3908 | return "data"; | |
3909 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3910 | return "system"; | |
3911 | default: | |
3912 | WARN_ON(1); | |
3913 | return "invalid-combination"; | |
3914 | }; | |
3915 | } | |
3916 | ||
593060d7 CM |
3917 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3918 | u64 total_bytes, u64 bytes_used, | |
e40edf2d | 3919 | u64 bytes_readonly, |
593060d7 CM |
3920 | struct btrfs_space_info **space_info) |
3921 | { | |
3922 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3923 | int i; |
3924 | int factor; | |
b150a4f1 | 3925 | int ret; |
b742bb82 YZ |
3926 | |
3927 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3928 | BTRFS_BLOCK_GROUP_RAID10)) | |
3929 | factor = 2; | |
3930 | else | |
3931 | factor = 1; | |
593060d7 CM |
3932 | |
3933 | found = __find_space_info(info, flags); | |
3934 | if (found) { | |
25179201 | 3935 | spin_lock(&found->lock); |
593060d7 | 3936 | found->total_bytes += total_bytes; |
89a55897 | 3937 | found->disk_total += total_bytes * factor; |
593060d7 | 3938 | found->bytes_used += bytes_used; |
b742bb82 | 3939 | found->disk_used += bytes_used * factor; |
e40edf2d | 3940 | found->bytes_readonly += bytes_readonly; |
2e6e5183 FM |
3941 | if (total_bytes > 0) |
3942 | found->full = 0; | |
957780eb JB |
3943 | space_info_add_new_bytes(info, found, total_bytes - |
3944 | bytes_used - bytes_readonly); | |
25179201 | 3945 | spin_unlock(&found->lock); |
593060d7 CM |
3946 | *space_info = found; |
3947 | return 0; | |
3948 | } | |
c146afad | 3949 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3950 | if (!found) |
3951 | return -ENOMEM; | |
3952 | ||
908c7f19 | 3953 | ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL); |
b150a4f1 JB |
3954 | if (ret) { |
3955 | kfree(found); | |
3956 | return ret; | |
3957 | } | |
3958 | ||
c1895442 | 3959 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
b742bb82 | 3960 | INIT_LIST_HEAD(&found->block_groups[i]); |
80eb234a | 3961 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3962 | spin_lock_init(&found->lock); |
52ba6929 | 3963 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3964 | found->total_bytes = total_bytes; |
89a55897 | 3965 | found->disk_total = total_bytes * factor; |
593060d7 | 3966 | found->bytes_used = bytes_used; |
b742bb82 | 3967 | found->disk_used = bytes_used * factor; |
593060d7 | 3968 | found->bytes_pinned = 0; |
e8569813 | 3969 | found->bytes_reserved = 0; |
e40edf2d | 3970 | found->bytes_readonly = bytes_readonly; |
f0486c68 | 3971 | found->bytes_may_use = 0; |
6af3e3ad | 3972 | found->full = 0; |
4f4db217 | 3973 | found->max_extent_size = 0; |
0e4f8f88 | 3974 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3975 | found->chunk_alloc = 0; |
fdb5effd JB |
3976 | found->flush = 0; |
3977 | init_waitqueue_head(&found->wait); | |
633c0aad | 3978 | INIT_LIST_HEAD(&found->ro_bgs); |
957780eb JB |
3979 | INIT_LIST_HEAD(&found->tickets); |
3980 | INIT_LIST_HEAD(&found->priority_tickets); | |
6ab0a202 JM |
3981 | |
3982 | ret = kobject_init_and_add(&found->kobj, &space_info_ktype, | |
3983 | info->space_info_kobj, "%s", | |
3984 | alloc_name(found->flags)); | |
3985 | if (ret) { | |
3986 | kfree(found); | |
3987 | return ret; | |
3988 | } | |
3989 | ||
593060d7 | 3990 | *space_info = found; |
4184ea7f | 3991 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3992 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3993 | info->data_sinfo = found; | |
6ab0a202 JM |
3994 | |
3995 | return ret; | |
593060d7 CM |
3996 | } |
3997 | ||
8790d502 CM |
3998 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3999 | { | |
899c81ea ID |
4000 | u64 extra_flags = chunk_to_extended(flags) & |
4001 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4002 | |
de98ced9 | 4003 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4004 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4005 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4006 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4007 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4008 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4009 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4010 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4011 | } |
593060d7 | 4012 | |
fc67c450 ID |
4013 | /* |
4014 | * returns target flags in extended format or 0 if restripe for this | |
4015 | * chunk_type is not in progress | |
c6664b42 ID |
4016 | * |
4017 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
4018 | */ |
4019 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4020 | { | |
4021 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4022 | u64 target = 0; | |
4023 | ||
fc67c450 ID |
4024 | if (!bctl) |
4025 | return 0; | |
4026 | ||
4027 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4028 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4029 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4030 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4031 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4032 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4033 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4034 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4035 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4036 | } | |
4037 | ||
4038 | return target; | |
4039 | } | |
4040 | ||
a46d11a8 ID |
4041 | /* |
4042 | * @flags: available profiles in extended format (see ctree.h) | |
4043 | * | |
e4d8ec0f ID |
4044 | * Returns reduced profile in chunk format. If profile changing is in |
4045 | * progress (either running or paused) picks the target profile (if it's | |
4046 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4047 | */ |
2ff7e61e | 4048 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4049 | { |
0b246afa | 4050 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4051 | u64 target; |
9c170b26 ZL |
4052 | u64 raid_type; |
4053 | u64 allowed = 0; | |
a061fc8d | 4054 | |
fc67c450 ID |
4055 | /* |
4056 | * see if restripe for this chunk_type is in progress, if so | |
4057 | * try to reduce to the target profile | |
4058 | */ | |
0b246afa JM |
4059 | spin_lock(&fs_info->balance_lock); |
4060 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4061 | if (target) { |
4062 | /* pick target profile only if it's already available */ | |
4063 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4064 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4065 | return extended_to_chunk(target); |
e4d8ec0f ID |
4066 | } |
4067 | } | |
0b246afa | 4068 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4069 | |
53b381b3 | 4070 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4071 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4072 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
4073 | allowed |= btrfs_raid_group[raid_type]; | |
4074 | } | |
4075 | allowed &= flags; | |
4076 | ||
4077 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4078 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4079 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4080 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4081 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4082 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4083 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4084 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4085 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4086 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4087 | ||
4088 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4089 | ||
4090 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4091 | } |
4092 | ||
2ff7e61e | 4093 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4094 | { |
de98ced9 | 4095 | unsigned seq; |
f8213bdc | 4096 | u64 flags; |
de98ced9 MX |
4097 | |
4098 | do { | |
f8213bdc | 4099 | flags = orig_flags; |
0b246afa | 4100 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4101 | |
4102 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4103 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4104 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4105 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4106 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4107 | flags |= fs_info->avail_metadata_alloc_bits; |
4108 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4109 | |
2ff7e61e | 4110 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4111 | } |
4112 | ||
6d07bcec | 4113 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 4114 | { |
0b246afa | 4115 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4116 | u64 flags; |
53b381b3 | 4117 | u64 ret; |
9ed74f2d | 4118 | |
b742bb82 YZ |
4119 | if (data) |
4120 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4121 | else if (root == fs_info->chunk_root) |
b742bb82 | 4122 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4123 | else |
b742bb82 | 4124 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4125 | |
2ff7e61e | 4126 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4127 | return ret; |
6a63209f | 4128 | } |
9ed74f2d | 4129 | |
4136135b LB |
4130 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4131 | bool may_use_included) | |
4132 | { | |
4133 | ASSERT(s_info); | |
4134 | return s_info->bytes_used + s_info->bytes_reserved + | |
4135 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4136 | (may_use_included ? s_info->bytes_may_use : 0); | |
4137 | } | |
4138 | ||
04f4f916 | 4139 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4140 | { |
6a63209f | 4141 | struct btrfs_space_info *data_sinfo; |
04f4f916 | 4142 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4143 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 4144 | u64 used; |
94b947b2 | 4145 | int ret = 0; |
c99f1b0c ZL |
4146 | int need_commit = 2; |
4147 | int have_pinned_space; | |
6a63209f | 4148 | |
6a63209f | 4149 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4150 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4151 | |
9dced186 | 4152 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4153 | need_commit = 0; |
9dced186 | 4154 | ASSERT(current->journal_info); |
0af3d00b JB |
4155 | } |
4156 | ||
b4d7c3c9 | 4157 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
4158 | if (!data_sinfo) |
4159 | goto alloc; | |
9ed74f2d | 4160 | |
6a63209f JB |
4161 | again: |
4162 | /* make sure we have enough space to handle the data first */ | |
4163 | spin_lock(&data_sinfo->lock); | |
4136135b | 4164 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4165 | |
4166 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4167 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4168 | |
6a63209f JB |
4169 | /* |
4170 | * if we don't have enough free bytes in this space then we need | |
4171 | * to alloc a new chunk. | |
4172 | */ | |
b9fd47cd | 4173 | if (!data_sinfo->full) { |
6a63209f | 4174 | u64 alloc_target; |
9ed74f2d | 4175 | |
0e4f8f88 | 4176 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4177 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 4178 | alloc: |
6a63209f | 4179 | alloc_target = btrfs_get_alloc_profile(root, 1); |
9dced186 MX |
4180 | /* |
4181 | * It is ugly that we don't call nolock join | |
4182 | * transaction for the free space inode case here. | |
4183 | * But it is safe because we only do the data space | |
4184 | * reservation for the free space cache in the | |
4185 | * transaction context, the common join transaction | |
4186 | * just increase the counter of the current transaction | |
4187 | * handler, doesn't try to acquire the trans_lock of | |
4188 | * the fs. | |
4189 | */ | |
7a7eaa40 | 4190 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4191 | if (IS_ERR(trans)) |
4192 | return PTR_ERR(trans); | |
9ed74f2d | 4193 | |
2ff7e61e | 4194 | ret = do_chunk_alloc(trans, fs_info, alloc_target, |
0e4f8f88 | 4195 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4196 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4197 | if (ret < 0) { |
4198 | if (ret != -ENOSPC) | |
4199 | return ret; | |
c99f1b0c ZL |
4200 | else { |
4201 | have_pinned_space = 1; | |
d52a5b5f | 4202 | goto commit_trans; |
c99f1b0c | 4203 | } |
d52a5b5f | 4204 | } |
9ed74f2d | 4205 | |
b4d7c3c9 LZ |
4206 | if (!data_sinfo) |
4207 | data_sinfo = fs_info->data_sinfo; | |
4208 | ||
6a63209f JB |
4209 | goto again; |
4210 | } | |
f2bb8f5c JB |
4211 | |
4212 | /* | |
b150a4f1 | 4213 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4214 | * allocation, and no removed chunk in current transaction, |
4215 | * don't bother committing the transaction. | |
f2bb8f5c | 4216 | */ |
c99f1b0c ZL |
4217 | have_pinned_space = percpu_counter_compare( |
4218 | &data_sinfo->total_bytes_pinned, | |
4219 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4220 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4221 | |
4e06bdd6 | 4222 | /* commit the current transaction and try again */ |
d52a5b5f | 4223 | commit_trans: |
c99f1b0c | 4224 | if (need_commit && |
0b246afa | 4225 | !atomic_read(&fs_info->open_ioctl_trans)) { |
c99f1b0c | 4226 | need_commit--; |
b150a4f1 | 4227 | |
e1746e83 ZL |
4228 | if (need_commit > 0) { |
4229 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
0b246afa JM |
4230 | btrfs_wait_ordered_roots(fs_info, -1, 0, |
4231 | (u64)-1); | |
e1746e83 | 4232 | } |
9a4e7276 | 4233 | |
7a7eaa40 | 4234 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4235 | if (IS_ERR(trans)) |
4236 | return PTR_ERR(trans); | |
c99f1b0c | 4237 | if (have_pinned_space >= 0 || |
3204d33c JB |
4238 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4239 | &trans->transaction->flags) || | |
c99f1b0c | 4240 | need_commit > 0) { |
3a45bb20 | 4241 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4242 | if (ret) |
4243 | return ret; | |
d7c15171 | 4244 | /* |
c2d6cb16 FM |
4245 | * The cleaner kthread might still be doing iput |
4246 | * operations. Wait for it to finish so that | |
4247 | * more space is released. | |
d7c15171 | 4248 | */ |
0b246afa JM |
4249 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4250 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4251 | goto again; |
4252 | } else { | |
3a45bb20 | 4253 | btrfs_end_transaction(trans); |
94b947b2 | 4254 | } |
4e06bdd6 | 4255 | } |
9ed74f2d | 4256 | |
0b246afa | 4257 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4258 | "space_info:enospc", |
4259 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4260 | return -ENOSPC; |
4261 | } | |
4262 | data_sinfo->bytes_may_use += bytes; | |
0b246afa | 4263 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4264 | data_sinfo->flags, bytes, 1); |
6a63209f | 4265 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4266 | |
237c0e9f | 4267 | return ret; |
9ed74f2d | 4268 | } |
6a63209f | 4269 | |
4ceff079 QW |
4270 | /* |
4271 | * New check_data_free_space() with ability for precious data reservation | |
4272 | * Will replace old btrfs_check_data_free_space(), but for patch split, | |
4273 | * add a new function first and then replace it. | |
4274 | */ | |
7cf5b976 | 4275 | int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len) |
4ceff079 | 4276 | { |
0b246afa | 4277 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4278 | int ret; |
4279 | ||
4280 | /* align the range */ | |
0b246afa JM |
4281 | len = round_up(start + len, fs_info->sectorsize) - |
4282 | round_down(start, fs_info->sectorsize); | |
4283 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4284 | |
04f4f916 | 4285 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4286 | if (ret < 0) |
4287 | return ret; | |
4288 | ||
1e5ec2e7 | 4289 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
4ceff079 | 4290 | ret = btrfs_qgroup_reserve_data(inode, start, len); |
1e5ec2e7 JB |
4291 | if (ret) |
4292 | btrfs_free_reserved_data_space_noquota(inode, start, len); | |
4ceff079 QW |
4293 | return ret; |
4294 | } | |
4295 | ||
4ceff079 QW |
4296 | /* |
4297 | * Called if we need to clear a data reservation for this inode | |
4298 | * Normally in a error case. | |
4299 | * | |
51773bec QW |
4300 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4301 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4302 | * Like clear_bit_hook(). | |
4ceff079 | 4303 | */ |
51773bec QW |
4304 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4305 | u64 len) | |
4ceff079 | 4306 | { |
0b246afa | 4307 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4308 | struct btrfs_space_info *data_sinfo; |
4309 | ||
4310 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4311 | len = round_up(start + len, fs_info->sectorsize) - |
4312 | round_down(start, fs_info->sectorsize); | |
4313 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4314 | |
0b246afa | 4315 | data_sinfo = fs_info->data_sinfo; |
4ceff079 QW |
4316 | spin_lock(&data_sinfo->lock); |
4317 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4318 | data_sinfo->bytes_may_use = 0; | |
4319 | else | |
4320 | data_sinfo->bytes_may_use -= len; | |
0b246afa | 4321 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4322 | data_sinfo->flags, len, 0); |
4323 | spin_unlock(&data_sinfo->lock); | |
4324 | } | |
4325 | ||
51773bec QW |
4326 | /* |
4327 | * Called if we need to clear a data reservation for this inode | |
4328 | * Normally in a error case. | |
4329 | * | |
01327610 | 4330 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4331 | * space framework. |
4332 | */ | |
4333 | void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len) | |
4334 | { | |
0c476a5d JM |
4335 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4336 | ||
4337 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4338 | len = round_up(start + len, root->fs_info->sectorsize) - |
4339 | round_down(start, root->fs_info->sectorsize); | |
4340 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4341 | |
51773bec QW |
4342 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
4343 | btrfs_qgroup_free_data(inode, start, len); | |
4344 | } | |
4345 | ||
97e728d4 | 4346 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4347 | { |
97e728d4 JB |
4348 | struct list_head *head = &info->space_info; |
4349 | struct btrfs_space_info *found; | |
e3ccfa98 | 4350 | |
97e728d4 JB |
4351 | rcu_read_lock(); |
4352 | list_for_each_entry_rcu(found, head, list) { | |
4353 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4354 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4355 | } |
97e728d4 | 4356 | rcu_read_unlock(); |
e3ccfa98 JB |
4357 | } |
4358 | ||
3c76cd84 MX |
4359 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4360 | { | |
4361 | return (global->size << 1); | |
4362 | } | |
4363 | ||
2ff7e61e | 4364 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4365 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4366 | { |
0b246afa | 4367 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
424499db | 4368 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 4369 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 4370 | u64 thresh; |
e3ccfa98 | 4371 | |
0e4f8f88 CM |
4372 | if (force == CHUNK_ALLOC_FORCE) |
4373 | return 1; | |
4374 | ||
fb25e914 JB |
4375 | /* |
4376 | * We need to take into account the global rsv because for all intents | |
4377 | * and purposes it's used space. Don't worry about locking the | |
4378 | * global_rsv, it doesn't change except when the transaction commits. | |
4379 | */ | |
54338b5c | 4380 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3c76cd84 | 4381 | num_allocated += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4382 | |
0e4f8f88 CM |
4383 | /* |
4384 | * in limited mode, we want to have some free space up to | |
4385 | * about 1% of the FS size. | |
4386 | */ | |
4387 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4388 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4389 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 CM |
4390 | |
4391 | if (num_bytes - num_allocated < thresh) | |
4392 | return 1; | |
4393 | } | |
0e4f8f88 | 4394 | |
ee22184b | 4395 | if (num_allocated + SZ_2M < div_factor(num_bytes, 8)) |
14ed0ca6 | 4396 | return 0; |
424499db | 4397 | return 1; |
32c00aff JB |
4398 | } |
4399 | ||
2ff7e61e | 4400 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4401 | { |
4402 | u64 num_dev; | |
4403 | ||
53b381b3 DW |
4404 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4405 | BTRFS_BLOCK_GROUP_RAID0 | | |
4406 | BTRFS_BLOCK_GROUP_RAID5 | | |
4407 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4408 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4409 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4410 | num_dev = 2; | |
4411 | else | |
4412 | num_dev = 1; /* DUP or single */ | |
4413 | ||
39c2d7fa | 4414 | return num_dev; |
15d1ff81 LB |
4415 | } |
4416 | ||
39c2d7fa FM |
4417 | /* |
4418 | * If @is_allocation is true, reserve space in the system space info necessary | |
4419 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4420 | * removing a chunk. | |
4421 | */ | |
4422 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 4423 | struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4424 | { |
4425 | struct btrfs_space_info *info; | |
4426 | u64 left; | |
4427 | u64 thresh; | |
4fbcdf66 | 4428 | int ret = 0; |
39c2d7fa | 4429 | u64 num_devs; |
4fbcdf66 FM |
4430 | |
4431 | /* | |
4432 | * Needed because we can end up allocating a system chunk and for an | |
4433 | * atomic and race free space reservation in the chunk block reserve. | |
4434 | */ | |
0b246afa | 4435 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
15d1ff81 | 4436 | |
0b246afa | 4437 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4438 | spin_lock(&info->lock); |
4136135b | 4439 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4440 | spin_unlock(&info->lock); |
4441 | ||
2ff7e61e | 4442 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4443 | |
4444 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4445 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4446 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4447 | |
0b246afa JM |
4448 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4449 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4450 | left, thresh, type); | |
4451 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4452 | } |
4453 | ||
4454 | if (left < thresh) { | |
4455 | u64 flags; | |
4456 | ||
0b246afa | 4457 | flags = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
4fbcdf66 FM |
4458 | /* |
4459 | * Ignore failure to create system chunk. We might end up not | |
4460 | * needing it, as we might not need to COW all nodes/leafs from | |
4461 | * the paths we visit in the chunk tree (they were already COWed | |
4462 | * or created in the current transaction for example). | |
4463 | */ | |
2ff7e61e | 4464 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
4fbcdf66 FM |
4465 | } |
4466 | ||
4467 | if (!ret) { | |
0b246afa JM |
4468 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4469 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4470 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4471 | if (!ret) | |
4472 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4473 | } |
4474 | } | |
4475 | ||
28b737f6 LB |
4476 | /* |
4477 | * If force is CHUNK_ALLOC_FORCE: | |
4478 | * - return 1 if it successfully allocates a chunk, | |
4479 | * - return errors including -ENOSPC otherwise. | |
4480 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4481 | * - return 0 if it doesn't need to allocate a new chunk, | |
4482 | * - return 1 if it successfully allocates a chunk, | |
4483 | * - return errors including -ENOSPC otherwise. | |
4484 | */ | |
6324fbf3 | 4485 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 4486 | struct btrfs_fs_info *fs_info, u64 flags, int force) |
9ed74f2d | 4487 | { |
6324fbf3 | 4488 | struct btrfs_space_info *space_info; |
6d74119f | 4489 | int wait_for_alloc = 0; |
9ed74f2d | 4490 | int ret = 0; |
9ed74f2d | 4491 | |
c6b305a8 JB |
4492 | /* Don't re-enter if we're already allocating a chunk */ |
4493 | if (trans->allocating_chunk) | |
4494 | return -ENOSPC; | |
4495 | ||
0b246afa | 4496 | space_info = __find_space_info(fs_info, flags); |
593060d7 | 4497 | if (!space_info) { |
0b246afa | 4498 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
79787eaa | 4499 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 4500 | } |
79787eaa | 4501 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 4502 | |
6d74119f | 4503 | again: |
25179201 | 4504 | spin_lock(&space_info->lock); |
9e622d6b | 4505 | if (force < space_info->force_alloc) |
0e4f8f88 | 4506 | force = space_info->force_alloc; |
25179201 | 4507 | if (space_info->full) { |
2ff7e61e | 4508 | if (should_alloc_chunk(fs_info, space_info, force)) |
09fb99a6 FDBM |
4509 | ret = -ENOSPC; |
4510 | else | |
4511 | ret = 0; | |
25179201 | 4512 | spin_unlock(&space_info->lock); |
09fb99a6 | 4513 | return ret; |
9ed74f2d JB |
4514 | } |
4515 | ||
2ff7e61e | 4516 | if (!should_alloc_chunk(fs_info, space_info, force)) { |
25179201 | 4517 | spin_unlock(&space_info->lock); |
6d74119f JB |
4518 | return 0; |
4519 | } else if (space_info->chunk_alloc) { | |
4520 | wait_for_alloc = 1; | |
4521 | } else { | |
4522 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4523 | } |
0e4f8f88 | 4524 | |
25179201 | 4525 | spin_unlock(&space_info->lock); |
9ed74f2d | 4526 | |
6d74119f JB |
4527 | mutex_lock(&fs_info->chunk_mutex); |
4528 | ||
4529 | /* | |
4530 | * The chunk_mutex is held throughout the entirety of a chunk | |
4531 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4532 | * other guy is done and we need to recheck and see if we should | |
4533 | * allocate. | |
4534 | */ | |
4535 | if (wait_for_alloc) { | |
4536 | mutex_unlock(&fs_info->chunk_mutex); | |
4537 | wait_for_alloc = 0; | |
4538 | goto again; | |
4539 | } | |
4540 | ||
c6b305a8 JB |
4541 | trans->allocating_chunk = true; |
4542 | ||
67377734 JB |
4543 | /* |
4544 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4545 | * allocating mixed chunks instead of individual chunks. | |
4546 | */ | |
4547 | if (btrfs_mixed_space_info(space_info)) | |
4548 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4549 | ||
97e728d4 JB |
4550 | /* |
4551 | * if we're doing a data chunk, go ahead and make sure that | |
4552 | * we keep a reasonable number of metadata chunks allocated in the | |
4553 | * FS as well. | |
4554 | */ | |
9ed74f2d | 4555 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4556 | fs_info->data_chunk_allocations++; |
4557 | if (!(fs_info->data_chunk_allocations % | |
4558 | fs_info->metadata_ratio)) | |
4559 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4560 | } |
4561 | ||
15d1ff81 LB |
4562 | /* |
4563 | * Check if we have enough space in SYSTEM chunk because we may need | |
4564 | * to update devices. | |
4565 | */ | |
2ff7e61e | 4566 | check_system_chunk(trans, fs_info, flags); |
15d1ff81 | 4567 | |
2ff7e61e | 4568 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
c6b305a8 | 4569 | trans->allocating_chunk = false; |
92b8e897 | 4570 | |
9ed74f2d | 4571 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4572 | if (ret < 0 && ret != -ENOSPC) |
4573 | goto out; | |
9ed74f2d | 4574 | if (ret) |
6324fbf3 | 4575 | space_info->full = 1; |
424499db YZ |
4576 | else |
4577 | ret = 1; | |
6d74119f | 4578 | |
0e4f8f88 | 4579 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4580 | out: |
6d74119f | 4581 | space_info->chunk_alloc = 0; |
9ed74f2d | 4582 | spin_unlock(&space_info->lock); |
a25c75d5 | 4583 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4584 | /* |
4585 | * When we allocate a new chunk we reserve space in the chunk block | |
4586 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4587 | * add new nodes/leafs to it if we end up needing to do it when | |
4588 | * inserting the chunk item and updating device items as part of the | |
4589 | * second phase of chunk allocation, performed by | |
4590 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4591 | * large number of new block groups to create in our transaction | |
4592 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4593 | * in extreme cases - like having a single transaction create many new | |
4594 | * block groups when starting to write out the free space caches of all | |
4595 | * the block groups that were made dirty during the lifetime of the | |
4596 | * transaction. | |
4597 | */ | |
d9a0540a | 4598 | if (trans->can_flush_pending_bgs && |
ee22184b | 4599 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
2ff7e61e | 4600 | btrfs_create_pending_block_groups(trans, fs_info); |
00d80e34 FM |
4601 | btrfs_trans_release_chunk_metadata(trans); |
4602 | } | |
0f9dd46c | 4603 | return ret; |
6324fbf3 | 4604 | } |
9ed74f2d | 4605 | |
a80c8dcf JB |
4606 | static int can_overcommit(struct btrfs_root *root, |
4607 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 4608 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf | 4609 | { |
0b246afa JM |
4610 | struct btrfs_fs_info *fs_info = root->fs_info; |
4611 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 4612 | u64 profile; |
3c76cd84 | 4613 | u64 space_size; |
a80c8dcf JB |
4614 | u64 avail; |
4615 | u64 used; | |
4616 | ||
957780eb JB |
4617 | /* Don't overcommit when in mixed mode. */ |
4618 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4619 | return 0; | |
4620 | ||
957780eb | 4621 | profile = btrfs_get_alloc_profile(root, 0); |
4136135b | 4622 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4623 | |
96f1bb57 JB |
4624 | /* |
4625 | * We only want to allow over committing if we have lots of actual space | |
4626 | * free, but if we don't have enough space to handle the global reserve | |
4627 | * space then we could end up having a real enospc problem when trying | |
4628 | * to allocate a chunk or some other such important allocation. | |
4629 | */ | |
3c76cd84 MX |
4630 | spin_lock(&global_rsv->lock); |
4631 | space_size = calc_global_rsv_need_space(global_rsv); | |
4632 | spin_unlock(&global_rsv->lock); | |
4633 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4634 | return 0; |
4635 | ||
4636 | used += space_info->bytes_may_use; | |
a80c8dcf | 4637 | |
0b246afa JM |
4638 | spin_lock(&fs_info->free_chunk_lock); |
4639 | avail = fs_info->free_chunk_space; | |
4640 | spin_unlock(&fs_info->free_chunk_lock); | |
a80c8dcf JB |
4641 | |
4642 | /* | |
4643 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4644 | * space is actually useable. For raid56, the space info used |
4645 | * doesn't include the parity drive, so we don't have to | |
4646 | * change the math | |
a80c8dcf JB |
4647 | */ |
4648 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4649 | BTRFS_BLOCK_GROUP_RAID1 | | |
4650 | BTRFS_BLOCK_GROUP_RAID10)) | |
4651 | avail >>= 1; | |
4652 | ||
4653 | /* | |
561c294d MX |
4654 | * If we aren't flushing all things, let us overcommit up to |
4655 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4656 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4657 | */ |
08e007d2 | 4658 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4659 | avail >>= 3; |
a80c8dcf | 4660 | else |
14575aef | 4661 | avail >>= 1; |
a80c8dcf | 4662 | |
14575aef | 4663 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4664 | return 1; |
4665 | return 0; | |
4666 | } | |
4667 | ||
2ff7e61e | 4668 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4669 | unsigned long nr_pages, int nr_items) |
da633a42 | 4670 | { |
0b246afa | 4671 | struct super_block *sb = fs_info->sb; |
da633a42 | 4672 | |
925a6efb JB |
4673 | if (down_read_trylock(&sb->s_umount)) { |
4674 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4675 | up_read(&sb->s_umount); | |
4676 | } else { | |
da633a42 MX |
4677 | /* |
4678 | * We needn't worry the filesystem going from r/w to r/o though | |
4679 | * we don't acquire ->s_umount mutex, because the filesystem | |
4680 | * should guarantee the delalloc inodes list be empty after | |
4681 | * the filesystem is readonly(all dirty pages are written to | |
4682 | * the disk). | |
4683 | */ | |
0b246afa | 4684 | btrfs_start_delalloc_roots(fs_info, 0, nr_items); |
98ad69cf | 4685 | if (!current->journal_info) |
0b246afa | 4686 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4687 | } |
4688 | } | |
4689 | ||
2ff7e61e JM |
4690 | static inline int calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
4691 | u64 to_reclaim) | |
18cd8ea6 MX |
4692 | { |
4693 | u64 bytes; | |
4694 | int nr; | |
4695 | ||
2ff7e61e | 4696 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
18cd8ea6 MX |
4697 | nr = (int)div64_u64(to_reclaim, bytes); |
4698 | if (!nr) | |
4699 | nr = 1; | |
4700 | return nr; | |
4701 | } | |
4702 | ||
ee22184b | 4703 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4704 | |
9ed74f2d | 4705 | /* |
5da9d01b | 4706 | * shrink metadata reservation for delalloc |
9ed74f2d | 4707 | */ |
f4c738c2 JB |
4708 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
4709 | bool wait_ordered) | |
5da9d01b | 4710 | { |
0b246afa | 4711 | struct btrfs_fs_info *fs_info = root->fs_info; |
0ca1f7ce | 4712 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4713 | struct btrfs_space_info *space_info; |
663350ac | 4714 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4715 | u64 delalloc_bytes; |
5da9d01b | 4716 | u64 max_reclaim; |
b1953bce | 4717 | long time_left; |
d3ee29e3 MX |
4718 | unsigned long nr_pages; |
4719 | int loops; | |
b0244199 | 4720 | int items; |
08e007d2 | 4721 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4722 | |
c61a16a7 | 4723 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4724 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
8eb0dfdb | 4725 | to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4726 | |
663350ac | 4727 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0b246afa | 4728 | block_rsv = &fs_info->delalloc_block_rsv; |
0019f10d | 4729 | space_info = block_rsv->space_info; |
bf9022e0 | 4730 | |
963d678b | 4731 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4732 | &fs_info->delalloc_bytes); |
f4c738c2 | 4733 | if (delalloc_bytes == 0) { |
fdb5effd | 4734 | if (trans) |
f4c738c2 | 4735 | return; |
38c135af | 4736 | if (wait_ordered) |
0b246afa | 4737 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4738 | return; |
fdb5effd JB |
4739 | } |
4740 | ||
d3ee29e3 | 4741 | loops = 0; |
f4c738c2 JB |
4742 | while (delalloc_bytes && loops < 3) { |
4743 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4744 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4745 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4746 | /* |
4747 | * We need to wait for the async pages to actually start before | |
4748 | * we do anything. | |
4749 | */ | |
0b246afa | 4750 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4751 | if (!max_reclaim) |
4752 | goto skip_async; | |
4753 | ||
4754 | if (max_reclaim <= nr_pages) | |
4755 | max_reclaim = 0; | |
4756 | else | |
4757 | max_reclaim -= nr_pages; | |
dea31f52 | 4758 | |
0b246afa JM |
4759 | wait_event(fs_info->async_submit_wait, |
4760 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4761 | (int)max_reclaim); |
4762 | skip_async: | |
08e007d2 MX |
4763 | if (!trans) |
4764 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4765 | else | |
4766 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4767 | spin_lock(&space_info->lock); |
08e007d2 | 4768 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
4769 | spin_unlock(&space_info->lock); |
4770 | break; | |
4771 | } | |
957780eb JB |
4772 | if (list_empty(&space_info->tickets) && |
4773 | list_empty(&space_info->priority_tickets)) { | |
4774 | spin_unlock(&space_info->lock); | |
4775 | break; | |
4776 | } | |
0019f10d | 4777 | spin_unlock(&space_info->lock); |
5da9d01b | 4778 | |
36e39c40 | 4779 | loops++; |
f104d044 | 4780 | if (wait_ordered && !trans) { |
0b246afa | 4781 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4782 | } else { |
f4c738c2 | 4783 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4784 | if (time_left) |
4785 | break; | |
4786 | } | |
963d678b | 4787 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4788 | &fs_info->delalloc_bytes); |
5da9d01b | 4789 | } |
5da9d01b YZ |
4790 | } |
4791 | ||
663350ac JB |
4792 | /** |
4793 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4794 | * @root - the root we're allocating for | |
4795 | * @bytes - the number of bytes we want to reserve | |
4796 | * @force - force the commit | |
8bb8ab2e | 4797 | * |
663350ac JB |
4798 | * This will check to make sure that committing the transaction will actually |
4799 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4800 | * will return -ENOSPC. | |
8bb8ab2e | 4801 | */ |
0c9ab349 | 4802 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
663350ac JB |
4803 | struct btrfs_space_info *space_info, |
4804 | u64 bytes, int force) | |
4805 | { | |
0b246afa | 4806 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac JB |
4807 | struct btrfs_trans_handle *trans; |
4808 | ||
4809 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4810 | if (trans) | |
4811 | return -EAGAIN; | |
4812 | ||
4813 | if (force) | |
4814 | goto commit; | |
4815 | ||
4816 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4817 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4818 | bytes) >= 0) |
663350ac | 4819 | goto commit; |
663350ac JB |
4820 | |
4821 | /* | |
4822 | * See if there is some space in the delayed insertion reservation for | |
4823 | * this reservation. | |
4824 | */ | |
4825 | if (space_info != delayed_rsv->space_info) | |
4826 | return -ENOSPC; | |
4827 | ||
4828 | spin_lock(&delayed_rsv->lock); | |
b150a4f1 JB |
4829 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
4830 | bytes - delayed_rsv->size) >= 0) { | |
663350ac JB |
4831 | spin_unlock(&delayed_rsv->lock); |
4832 | return -ENOSPC; | |
4833 | } | |
4834 | spin_unlock(&delayed_rsv->lock); | |
4835 | ||
4836 | commit: | |
0c9ab349 | 4837 | trans = btrfs_join_transaction(fs_info->fs_root); |
663350ac JB |
4838 | if (IS_ERR(trans)) |
4839 | return -ENOSPC; | |
4840 | ||
3a45bb20 | 4841 | return btrfs_commit_transaction(trans); |
663350ac JB |
4842 | } |
4843 | ||
957780eb JB |
4844 | struct reserve_ticket { |
4845 | u64 bytes; | |
4846 | int error; | |
4847 | struct list_head list; | |
4848 | wait_queue_head_t wait; | |
96c3f433 JB |
4849 | }; |
4850 | ||
0c9ab349 | 4851 | static int flush_space(struct btrfs_fs_info *fs_info, |
96c3f433 JB |
4852 | struct btrfs_space_info *space_info, u64 num_bytes, |
4853 | u64 orig_bytes, int state) | |
4854 | { | |
0c9ab349 | 4855 | struct btrfs_root *root = fs_info->fs_root; |
96c3f433 JB |
4856 | struct btrfs_trans_handle *trans; |
4857 | int nr; | |
f4c738c2 | 4858 | int ret = 0; |
96c3f433 JB |
4859 | |
4860 | switch (state) { | |
96c3f433 JB |
4861 | case FLUSH_DELAYED_ITEMS_NR: |
4862 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4863 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4864 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4865 | else |
96c3f433 | 4866 | nr = -1; |
18cd8ea6 | 4867 | |
96c3f433 JB |
4868 | trans = btrfs_join_transaction(root); |
4869 | if (IS_ERR(trans)) { | |
4870 | ret = PTR_ERR(trans); | |
4871 | break; | |
4872 | } | |
2ff7e61e | 4873 | ret = btrfs_run_delayed_items_nr(trans, fs_info, nr); |
3a45bb20 | 4874 | btrfs_end_transaction(trans); |
96c3f433 | 4875 | break; |
67b0fd63 JB |
4876 | case FLUSH_DELALLOC: |
4877 | case FLUSH_DELALLOC_WAIT: | |
24af7dd1 | 4878 | shrink_delalloc(root, num_bytes * 2, orig_bytes, |
67b0fd63 JB |
4879 | state == FLUSH_DELALLOC_WAIT); |
4880 | break; | |
ea658bad JB |
4881 | case ALLOC_CHUNK: |
4882 | trans = btrfs_join_transaction(root); | |
4883 | if (IS_ERR(trans)) { | |
4884 | ret = PTR_ERR(trans); | |
4885 | break; | |
4886 | } | |
2ff7e61e | 4887 | ret = do_chunk_alloc(trans, fs_info, |
ea658bad JB |
4888 | btrfs_get_alloc_profile(root, 0), |
4889 | CHUNK_ALLOC_NO_FORCE); | |
3a45bb20 | 4890 | btrfs_end_transaction(trans); |
eecba891 | 4891 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
4892 | ret = 0; |
4893 | break; | |
96c3f433 | 4894 | case COMMIT_TRANS: |
0c9ab349 JM |
4895 | ret = may_commit_transaction(fs_info, space_info, |
4896 | orig_bytes, 0); | |
96c3f433 JB |
4897 | break; |
4898 | default: | |
4899 | ret = -ENOSPC; | |
4900 | break; | |
4901 | } | |
4902 | ||
0b246afa | 4903 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, |
f376df2b | 4904 | orig_bytes, state, ret); |
96c3f433 JB |
4905 | return ret; |
4906 | } | |
21c7e756 MX |
4907 | |
4908 | static inline u64 | |
4909 | btrfs_calc_reclaim_metadata_size(struct btrfs_root *root, | |
4910 | struct btrfs_space_info *space_info) | |
4911 | { | |
957780eb | 4912 | struct reserve_ticket *ticket; |
21c7e756 MX |
4913 | u64 used; |
4914 | u64 expected; | |
957780eb | 4915 | u64 to_reclaim = 0; |
21c7e756 | 4916 | |
957780eb JB |
4917 | list_for_each_entry(ticket, &space_info->tickets, list) |
4918 | to_reclaim += ticket->bytes; | |
4919 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
4920 | to_reclaim += ticket->bytes; | |
4921 | if (to_reclaim) | |
4922 | return to_reclaim; | |
21c7e756 | 4923 | |
e0af2484 WX |
4924 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
4925 | if (can_overcommit(root, space_info, to_reclaim, | |
4926 | BTRFS_RESERVE_FLUSH_ALL)) | |
4927 | return 0; | |
4928 | ||
21c7e756 MX |
4929 | used = space_info->bytes_used + space_info->bytes_reserved + |
4930 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4931 | space_info->bytes_may_use; | |
ee22184b | 4932 | if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL)) |
21c7e756 MX |
4933 | expected = div_factor_fine(space_info->total_bytes, 95); |
4934 | else | |
4935 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4936 | ||
4937 | if (used > expected) | |
4938 | to_reclaim = used - expected; | |
4939 | else | |
4940 | to_reclaim = 0; | |
4941 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4942 | space_info->bytes_reserved); | |
21c7e756 MX |
4943 | return to_reclaim; |
4944 | } | |
4945 | ||
4946 | static inline int need_do_async_reclaim(struct btrfs_space_info *space_info, | |
87241c2e | 4947 | struct btrfs_root *root, u64 used) |
21c7e756 | 4948 | { |
0b246afa | 4949 | struct btrfs_fs_info *fs_info = root->fs_info; |
365c5313 JB |
4950 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4951 | ||
4952 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 4953 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
4954 | return 0; |
4955 | ||
87241c2e | 4956 | if (!btrfs_calc_reclaim_metadata_size(root, space_info)) |
d38b349c JB |
4957 | return 0; |
4958 | ||
0b246afa JM |
4959 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
4960 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
4961 | } |
4962 | ||
957780eb | 4963 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 4964 | { |
957780eb | 4965 | struct reserve_ticket *ticket; |
25ce459c | 4966 | |
957780eb JB |
4967 | while (!list_empty(head)) { |
4968 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
4969 | list_del_init(&ticket->list); | |
4970 | ticket->error = -ENOSPC; | |
4971 | wake_up(&ticket->wait); | |
21c7e756 | 4972 | } |
21c7e756 MX |
4973 | } |
4974 | ||
957780eb JB |
4975 | /* |
4976 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
4977 | * will loop and continuously try to flush as long as we are making progress. | |
4978 | * We count progress as clearing off tickets each time we have to loop. | |
4979 | */ | |
21c7e756 MX |
4980 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
4981 | { | |
4982 | struct btrfs_fs_info *fs_info; | |
4983 | struct btrfs_space_info *space_info; | |
4984 | u64 to_reclaim; | |
4985 | int flush_state; | |
957780eb | 4986 | int commit_cycles = 0; |
ce129655 | 4987 | u64 last_tickets_id; |
21c7e756 MX |
4988 | |
4989 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4990 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4991 | ||
957780eb | 4992 | spin_lock(&space_info->lock); |
21c7e756 MX |
4993 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, |
4994 | space_info); | |
957780eb JB |
4995 | if (!to_reclaim) { |
4996 | space_info->flush = 0; | |
4997 | spin_unlock(&space_info->lock); | |
21c7e756 | 4998 | return; |
957780eb | 4999 | } |
ce129655 | 5000 | last_tickets_id = space_info->tickets_id; |
957780eb | 5001 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5002 | |
5003 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb JB |
5004 | do { |
5005 | struct reserve_ticket *ticket; | |
5006 | int ret; | |
5007 | ||
0c9ab349 JM |
5008 | ret = flush_space(fs_info, space_info, to_reclaim, to_reclaim, |
5009 | flush_state); | |
957780eb JB |
5010 | spin_lock(&space_info->lock); |
5011 | if (list_empty(&space_info->tickets)) { | |
5012 | space_info->flush = 0; | |
5013 | spin_unlock(&space_info->lock); | |
5014 | return; | |
5015 | } | |
5016 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, | |
5017 | space_info); | |
5018 | ticket = list_first_entry(&space_info->tickets, | |
5019 | struct reserve_ticket, list); | |
ce129655 | 5020 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5021 | flush_state++; |
5022 | } else { | |
ce129655 | 5023 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5024 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5025 | if (commit_cycles) | |
5026 | commit_cycles--; | |
5027 | } | |
5028 | ||
5029 | if (flush_state > COMMIT_TRANS) { | |
5030 | commit_cycles++; | |
5031 | if (commit_cycles > 2) { | |
5032 | wake_all_tickets(&space_info->tickets); | |
5033 | space_info->flush = 0; | |
5034 | } else { | |
5035 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5036 | } | |
5037 | } | |
5038 | spin_unlock(&space_info->lock); | |
5039 | } while (flush_state <= COMMIT_TRANS); | |
5040 | } | |
5041 | ||
5042 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5043 | { | |
5044 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5045 | } | |
5046 | ||
5047 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5048 | struct btrfs_space_info *space_info, | |
5049 | struct reserve_ticket *ticket) | |
5050 | { | |
5051 | u64 to_reclaim; | |
5052 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5053 | ||
5054 | spin_lock(&space_info->lock); | |
5055 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, | |
5056 | space_info); | |
5057 | if (!to_reclaim) { | |
5058 | spin_unlock(&space_info->lock); | |
5059 | return; | |
5060 | } | |
5061 | spin_unlock(&space_info->lock); | |
5062 | ||
21c7e756 | 5063 | do { |
0c9ab349 JM |
5064 | flush_space(fs_info, space_info, to_reclaim, to_reclaim, |
5065 | flush_state); | |
21c7e756 | 5066 | flush_state++; |
957780eb JB |
5067 | spin_lock(&space_info->lock); |
5068 | if (ticket->bytes == 0) { | |
5069 | spin_unlock(&space_info->lock); | |
21c7e756 | 5070 | return; |
957780eb JB |
5071 | } |
5072 | spin_unlock(&space_info->lock); | |
5073 | ||
5074 | /* | |
5075 | * Priority flushers can't wait on delalloc without | |
5076 | * deadlocking. | |
5077 | */ | |
5078 | if (flush_state == FLUSH_DELALLOC || | |
5079 | flush_state == FLUSH_DELALLOC_WAIT) | |
5080 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5081 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5082 | } |
5083 | ||
957780eb JB |
5084 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5085 | struct btrfs_space_info *space_info, | |
5086 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5087 | ||
21c7e756 | 5088 | { |
957780eb JB |
5089 | DEFINE_WAIT(wait); |
5090 | int ret = 0; | |
5091 | ||
5092 | spin_lock(&space_info->lock); | |
5093 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5094 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5095 | if (ret) { | |
5096 | ret = -EINTR; | |
5097 | break; | |
5098 | } | |
5099 | spin_unlock(&space_info->lock); | |
5100 | ||
5101 | schedule(); | |
5102 | ||
5103 | finish_wait(&ticket->wait, &wait); | |
5104 | spin_lock(&space_info->lock); | |
5105 | } | |
5106 | if (!ret) | |
5107 | ret = ticket->error; | |
5108 | if (!list_empty(&ticket->list)) | |
5109 | list_del_init(&ticket->list); | |
5110 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5111 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5112 | space_info->bytes_may_use -= num_bytes; | |
5113 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5114 | space_info->flags, num_bytes, 0); | |
5115 | } | |
5116 | spin_unlock(&space_info->lock); | |
5117 | ||
5118 | return ret; | |
21c7e756 MX |
5119 | } |
5120 | ||
4a92b1b8 JB |
5121 | /** |
5122 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5123 | * @root - the root we're allocating for | |
957780eb | 5124 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5125 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5126 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5127 | * |
01327610 | 5128 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5129 | * with the block_rsv. If there is not enough space it will make an attempt to |
5130 | * flush out space to make room. It will do this by flushing delalloc if | |
5131 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5132 | * regain reservations will be made and this will fail if there is not enough | |
5133 | * space already. | |
8bb8ab2e | 5134 | */ |
957780eb JB |
5135 | static int __reserve_metadata_bytes(struct btrfs_root *root, |
5136 | struct btrfs_space_info *space_info, | |
5137 | u64 orig_bytes, | |
5138 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5139 | { |
0b246afa | 5140 | struct btrfs_fs_info *fs_info = root->fs_info; |
957780eb | 5141 | struct reserve_ticket ticket; |
2bf64758 | 5142 | u64 used; |
8bb8ab2e | 5143 | int ret = 0; |
9ed74f2d | 5144 | |
957780eb | 5145 | ASSERT(orig_bytes); |
8ca17f0f | 5146 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5147 | |
8bb8ab2e | 5148 | spin_lock(&space_info->lock); |
fdb5effd | 5149 | ret = -ENOSPC; |
4136135b | 5150 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5151 | |
8bb8ab2e | 5152 | /* |
957780eb JB |
5153 | * If we have enough space then hooray, make our reservation and carry |
5154 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5155 | * If not things get more complicated. | |
8bb8ab2e | 5156 | */ |
957780eb JB |
5157 | if (used + orig_bytes <= space_info->total_bytes) { |
5158 | space_info->bytes_may_use += orig_bytes; | |
0b246afa JM |
5159 | trace_btrfs_space_reservation(fs_info, "space_info", |
5160 | space_info->flags, orig_bytes, 1); | |
957780eb JB |
5161 | ret = 0; |
5162 | } else if (can_overcommit(root, space_info, orig_bytes, flush)) { | |
44734ed1 | 5163 | space_info->bytes_may_use += orig_bytes; |
0b246afa JM |
5164 | trace_btrfs_space_reservation(fs_info, "space_info", |
5165 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5166 | ret = 0; |
2bf64758 JB |
5167 | } |
5168 | ||
8bb8ab2e | 5169 | /* |
957780eb JB |
5170 | * If we couldn't make a reservation then setup our reservation ticket |
5171 | * and kick the async worker if it's not already running. | |
08e007d2 | 5172 | * |
957780eb JB |
5173 | * If we are a priority flusher then we just need to add our ticket to |
5174 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5175 | */ |
72bcd99d | 5176 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5177 | ticket.bytes = orig_bytes; |
5178 | ticket.error = 0; | |
5179 | init_waitqueue_head(&ticket.wait); | |
5180 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5181 | list_add_tail(&ticket.list, &space_info->tickets); | |
5182 | if (!space_info->flush) { | |
5183 | space_info->flush = 1; | |
0b246afa | 5184 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5185 | space_info->flags, |
5186 | orig_bytes, flush, | |
5187 | "enospc"); | |
957780eb JB |
5188 | queue_work(system_unbound_wq, |
5189 | &root->fs_info->async_reclaim_work); | |
5190 | } | |
5191 | } else { | |
5192 | list_add_tail(&ticket.list, | |
5193 | &space_info->priority_tickets); | |
5194 | } | |
21c7e756 MX |
5195 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5196 | used += orig_bytes; | |
f6acfd50 JB |
5197 | /* |
5198 | * We will do the space reservation dance during log replay, | |
5199 | * which means we won't have fs_info->fs_root set, so don't do | |
5200 | * the async reclaim as we will panic. | |
5201 | */ | |
0b246afa | 5202 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
87241c2e | 5203 | need_do_async_reclaim(space_info, root, used) && |
0b246afa JM |
5204 | !work_busy(&fs_info->async_reclaim_work)) { |
5205 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5206 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5207 | queue_work(system_unbound_wq, |
0b246afa | 5208 | &fs_info->async_reclaim_work); |
f376df2b | 5209 | } |
8bb8ab2e | 5210 | } |
f0486c68 | 5211 | spin_unlock(&space_info->lock); |
08e007d2 | 5212 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5213 | return ret; |
f0486c68 | 5214 | |
957780eb | 5215 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5216 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5217 | orig_bytes); |
08e007d2 | 5218 | |
957780eb | 5219 | ret = 0; |
0b246afa | 5220 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5221 | spin_lock(&space_info->lock); |
5222 | if (ticket.bytes) { | |
5223 | if (ticket.bytes < orig_bytes) { | |
5224 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5225 | space_info->bytes_may_use -= num_bytes; | |
0b246afa JM |
5226 | trace_btrfs_space_reservation(fs_info, "space_info", |
5227 | space_info->flags, | |
5228 | num_bytes, 0); | |
08e007d2 | 5229 | |
957780eb JB |
5230 | } |
5231 | list_del_init(&ticket.list); | |
5232 | ret = -ENOSPC; | |
5233 | } | |
5234 | spin_unlock(&space_info->lock); | |
5235 | ASSERT(list_empty(&ticket.list)); | |
5236 | return ret; | |
5237 | } | |
8bb8ab2e | 5238 | |
957780eb JB |
5239 | /** |
5240 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5241 | * @root - the root we're allocating for | |
5242 | * @block_rsv - the block_rsv we're allocating for | |
5243 | * @orig_bytes - the number of bytes we want | |
5244 | * @flush - whether or not we can flush to make our reservation | |
5245 | * | |
5246 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5247 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5248 | * flush out space to make room. It will do this by flushing delalloc if | |
5249 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5250 | * regain reservations will be made and this will fail if there is not enough | |
5251 | * space already. | |
5252 | */ | |
5253 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5254 | struct btrfs_block_rsv *block_rsv, | |
5255 | u64 orig_bytes, | |
5256 | enum btrfs_reserve_flush_enum flush) | |
5257 | { | |
0b246afa JM |
5258 | struct btrfs_fs_info *fs_info = root->fs_info; |
5259 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb JB |
5260 | int ret; |
5261 | ||
5262 | ret = __reserve_metadata_bytes(root, block_rsv->space_info, orig_bytes, | |
5263 | flush); | |
5d80366e JB |
5264 | if (ret == -ENOSPC && |
5265 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5266 | if (block_rsv != global_rsv && |
5267 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5268 | ret = 0; | |
5269 | } | |
cab45e22 | 5270 | if (ret == -ENOSPC) |
0b246afa | 5271 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5272 | block_rsv->space_info->flags, |
5273 | orig_bytes, 1); | |
f0486c68 YZ |
5274 | return ret; |
5275 | } | |
5276 | ||
79787eaa JM |
5277 | static struct btrfs_block_rsv *get_block_rsv( |
5278 | const struct btrfs_trans_handle *trans, | |
5279 | const struct btrfs_root *root) | |
f0486c68 | 5280 | { |
0b246afa | 5281 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5282 | struct btrfs_block_rsv *block_rsv = NULL; |
5283 | ||
e9cf439f | 5284 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5285 | (root == fs_info->csum_root && trans->adding_csums) || |
5286 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5287 | block_rsv = trans->block_rsv; |
5288 | ||
4c13d758 | 5289 | if (!block_rsv) |
f0486c68 YZ |
5290 | block_rsv = root->block_rsv; |
5291 | ||
5292 | if (!block_rsv) | |
0b246afa | 5293 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5294 | |
5295 | return block_rsv; | |
5296 | } | |
5297 | ||
5298 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5299 | u64 num_bytes) | |
5300 | { | |
5301 | int ret = -ENOSPC; | |
5302 | spin_lock(&block_rsv->lock); | |
5303 | if (block_rsv->reserved >= num_bytes) { | |
5304 | block_rsv->reserved -= num_bytes; | |
5305 | if (block_rsv->reserved < block_rsv->size) | |
5306 | block_rsv->full = 0; | |
5307 | ret = 0; | |
5308 | } | |
5309 | spin_unlock(&block_rsv->lock); | |
5310 | return ret; | |
5311 | } | |
5312 | ||
5313 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5314 | u64 num_bytes, int update_size) | |
5315 | { | |
5316 | spin_lock(&block_rsv->lock); | |
5317 | block_rsv->reserved += num_bytes; | |
5318 | if (update_size) | |
5319 | block_rsv->size += num_bytes; | |
5320 | else if (block_rsv->reserved >= block_rsv->size) | |
5321 | block_rsv->full = 1; | |
5322 | spin_unlock(&block_rsv->lock); | |
5323 | } | |
5324 | ||
d52be818 JB |
5325 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5326 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5327 | int min_factor) | |
5328 | { | |
5329 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5330 | u64 min_bytes; | |
5331 | ||
5332 | if (global_rsv->space_info != dest->space_info) | |
5333 | return -ENOSPC; | |
5334 | ||
5335 | spin_lock(&global_rsv->lock); | |
5336 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5337 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5338 | spin_unlock(&global_rsv->lock); | |
5339 | return -ENOSPC; | |
5340 | } | |
5341 | global_rsv->reserved -= num_bytes; | |
5342 | if (global_rsv->reserved < global_rsv->size) | |
5343 | global_rsv->full = 0; | |
5344 | spin_unlock(&global_rsv->lock); | |
5345 | ||
5346 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5347 | return 0; | |
5348 | } | |
5349 | ||
957780eb JB |
5350 | /* |
5351 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5352 | * basically when we're returning space from block_rsv's. | |
5353 | */ | |
5354 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5355 | struct btrfs_space_info *space_info, | |
5356 | u64 num_bytes) | |
5357 | { | |
5358 | struct reserve_ticket *ticket; | |
5359 | struct list_head *head; | |
5360 | u64 used; | |
5361 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5362 | bool check_overcommit = false; | |
5363 | ||
5364 | spin_lock(&space_info->lock); | |
5365 | head = &space_info->priority_tickets; | |
5366 | ||
5367 | /* | |
5368 | * If we are over our limit then we need to check and see if we can | |
5369 | * overcommit, and if we can't then we just need to free up our space | |
5370 | * and not satisfy any requests. | |
5371 | */ | |
5372 | used = space_info->bytes_used + space_info->bytes_reserved + | |
5373 | space_info->bytes_pinned + space_info->bytes_readonly + | |
5374 | space_info->bytes_may_use; | |
5375 | if (used - num_bytes >= space_info->total_bytes) | |
5376 | check_overcommit = true; | |
5377 | again: | |
5378 | while (!list_empty(head) && num_bytes) { | |
5379 | ticket = list_first_entry(head, struct reserve_ticket, | |
5380 | list); | |
5381 | /* | |
5382 | * We use 0 bytes because this space is already reserved, so | |
5383 | * adding the ticket space would be a double count. | |
5384 | */ | |
5385 | if (check_overcommit && | |
5386 | !can_overcommit(fs_info->extent_root, space_info, 0, | |
5387 | flush)) | |
5388 | break; | |
5389 | if (num_bytes >= ticket->bytes) { | |
5390 | list_del_init(&ticket->list); | |
5391 | num_bytes -= ticket->bytes; | |
5392 | ticket->bytes = 0; | |
ce129655 | 5393 | space_info->tickets_id++; |
957780eb JB |
5394 | wake_up(&ticket->wait); |
5395 | } else { | |
5396 | ticket->bytes -= num_bytes; | |
5397 | num_bytes = 0; | |
5398 | } | |
5399 | } | |
5400 | ||
5401 | if (num_bytes && head == &space_info->priority_tickets) { | |
5402 | head = &space_info->tickets; | |
5403 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5404 | goto again; | |
5405 | } | |
5406 | space_info->bytes_may_use -= num_bytes; | |
5407 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5408 | space_info->flags, num_bytes, 0); | |
5409 | spin_unlock(&space_info->lock); | |
5410 | } | |
5411 | ||
5412 | /* | |
5413 | * This is for newly allocated space that isn't accounted in | |
5414 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5415 | * we use this helper. | |
5416 | */ | |
5417 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5418 | struct btrfs_space_info *space_info, | |
5419 | u64 num_bytes) | |
5420 | { | |
5421 | struct reserve_ticket *ticket; | |
5422 | struct list_head *head = &space_info->priority_tickets; | |
5423 | ||
5424 | again: | |
5425 | while (!list_empty(head) && num_bytes) { | |
5426 | ticket = list_first_entry(head, struct reserve_ticket, | |
5427 | list); | |
5428 | if (num_bytes >= ticket->bytes) { | |
5429 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5430 | space_info->flags, | |
5431 | ticket->bytes, 1); | |
5432 | list_del_init(&ticket->list); | |
5433 | num_bytes -= ticket->bytes; | |
5434 | space_info->bytes_may_use += ticket->bytes; | |
5435 | ticket->bytes = 0; | |
ce129655 | 5436 | space_info->tickets_id++; |
957780eb JB |
5437 | wake_up(&ticket->wait); |
5438 | } else { | |
5439 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5440 | space_info->flags, | |
5441 | num_bytes, 1); | |
5442 | space_info->bytes_may_use += num_bytes; | |
5443 | ticket->bytes -= num_bytes; | |
5444 | num_bytes = 0; | |
5445 | } | |
5446 | } | |
5447 | ||
5448 | if (num_bytes && head == &space_info->priority_tickets) { | |
5449 | head = &space_info->tickets; | |
5450 | goto again; | |
5451 | } | |
5452 | } | |
5453 | ||
8c2a3ca2 JB |
5454 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
5455 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 5456 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
5457 | { |
5458 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
5459 | ||
5460 | spin_lock(&block_rsv->lock); | |
5461 | if (num_bytes == (u64)-1) | |
5462 | num_bytes = block_rsv->size; | |
5463 | block_rsv->size -= num_bytes; | |
5464 | if (block_rsv->reserved >= block_rsv->size) { | |
5465 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5466 | block_rsv->reserved = block_rsv->size; | |
5467 | block_rsv->full = 1; | |
5468 | } else { | |
5469 | num_bytes = 0; | |
5470 | } | |
5471 | spin_unlock(&block_rsv->lock); | |
5472 | ||
5473 | if (num_bytes > 0) { | |
5474 | if (dest) { | |
e9e22899 JB |
5475 | spin_lock(&dest->lock); |
5476 | if (!dest->full) { | |
5477 | u64 bytes_to_add; | |
5478 | ||
5479 | bytes_to_add = dest->size - dest->reserved; | |
5480 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5481 | dest->reserved += bytes_to_add; | |
5482 | if (dest->reserved >= dest->size) | |
5483 | dest->full = 1; | |
5484 | num_bytes -= bytes_to_add; | |
5485 | } | |
5486 | spin_unlock(&dest->lock); | |
5487 | } | |
957780eb JB |
5488 | if (num_bytes) |
5489 | space_info_add_old_bytes(fs_info, space_info, | |
5490 | num_bytes); | |
9ed74f2d | 5491 | } |
f0486c68 | 5492 | } |
4e06bdd6 | 5493 | |
25d609f8 JB |
5494 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5495 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
5496 | int update_size) | |
f0486c68 YZ |
5497 | { |
5498 | int ret; | |
9ed74f2d | 5499 | |
f0486c68 YZ |
5500 | ret = block_rsv_use_bytes(src, num_bytes); |
5501 | if (ret) | |
5502 | return ret; | |
9ed74f2d | 5503 | |
25d609f8 | 5504 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5505 | return 0; |
5506 | } | |
5507 | ||
66d8f3dd | 5508 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5509 | { |
f0486c68 YZ |
5510 | memset(rsv, 0, sizeof(*rsv)); |
5511 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5512 | rsv->type = type; |
f0486c68 YZ |
5513 | } |
5514 | ||
2ff7e61e | 5515 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5516 | unsigned short type) |
f0486c68 YZ |
5517 | { |
5518 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5519 | |
f0486c68 YZ |
5520 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5521 | if (!block_rsv) | |
5522 | return NULL; | |
9ed74f2d | 5523 | |
66d8f3dd | 5524 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
5525 | block_rsv->space_info = __find_space_info(fs_info, |
5526 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
5527 | return block_rsv; |
5528 | } | |
9ed74f2d | 5529 | |
2ff7e61e | 5530 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5531 | struct btrfs_block_rsv *rsv) |
5532 | { | |
2aaa6655 JB |
5533 | if (!rsv) |
5534 | return; | |
2ff7e61e | 5535 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5536 | kfree(rsv); |
9ed74f2d JB |
5537 | } |
5538 | ||
cdfb080e CM |
5539 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5540 | { | |
5541 | kfree(rsv); | |
5542 | } | |
5543 | ||
08e007d2 MX |
5544 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5545 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5546 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5547 | { |
f0486c68 | 5548 | int ret; |
9ed74f2d | 5549 | |
f0486c68 YZ |
5550 | if (num_bytes == 0) |
5551 | return 0; | |
8bb8ab2e | 5552 | |
61b520a9 | 5553 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5554 | if (!ret) { |
5555 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5556 | return 0; | |
5557 | } | |
9ed74f2d | 5558 | |
f0486c68 | 5559 | return ret; |
f0486c68 | 5560 | } |
9ed74f2d | 5561 | |
2ff7e61e | 5562 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5563 | { |
5564 | u64 num_bytes = 0; | |
f0486c68 | 5565 | int ret = -ENOSPC; |
9ed74f2d | 5566 | |
f0486c68 YZ |
5567 | if (!block_rsv) |
5568 | return 0; | |
9ed74f2d | 5569 | |
f0486c68 | 5570 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5571 | num_bytes = div_factor(block_rsv->size, min_factor); |
5572 | if (block_rsv->reserved >= num_bytes) | |
5573 | ret = 0; | |
5574 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5575 | |
36ba022a JB |
5576 | return ret; |
5577 | } | |
5578 | ||
08e007d2 MX |
5579 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5580 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5581 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5582 | { |
5583 | u64 num_bytes = 0; | |
5584 | int ret = -ENOSPC; | |
5585 | ||
5586 | if (!block_rsv) | |
5587 | return 0; | |
5588 | ||
5589 | spin_lock(&block_rsv->lock); | |
5590 | num_bytes = min_reserved; | |
13553e52 | 5591 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5592 | ret = 0; |
13553e52 | 5593 | else |
f0486c68 | 5594 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5595 | spin_unlock(&block_rsv->lock); |
13553e52 | 5596 | |
f0486c68 YZ |
5597 | if (!ret) |
5598 | return 0; | |
5599 | ||
aa38a711 | 5600 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5601 | if (!ret) { |
5602 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5603 | return 0; |
6a63209f | 5604 | } |
9ed74f2d | 5605 | |
13553e52 | 5606 | return ret; |
f0486c68 YZ |
5607 | } |
5608 | ||
2ff7e61e | 5609 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5610 | struct btrfs_block_rsv *block_rsv, |
5611 | u64 num_bytes) | |
5612 | { | |
0b246afa JM |
5613 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5614 | ||
17504584 | 5615 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5616 | block_rsv->space_info != global_rsv->space_info) |
5617 | global_rsv = NULL; | |
0b246afa | 5618 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes); |
6a63209f JB |
5619 | } |
5620 | ||
8929ecfa YZ |
5621 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5622 | { | |
5623 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5624 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5625 | u64 num_bytes; | |
6a63209f | 5626 | |
ae2e4728 JB |
5627 | /* |
5628 | * The global block rsv is based on the size of the extent tree, the | |
5629 | * checksum tree and the root tree. If the fs is empty we want to set | |
5630 | * it to a minimal amount for safety. | |
5631 | */ | |
5632 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5633 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5634 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5635 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5636 | |
8929ecfa | 5637 | spin_lock(&sinfo->lock); |
1f699d38 | 5638 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5639 | |
ee22184b | 5640 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5641 | |
fb4b10e5 | 5642 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5643 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5644 | if (sinfo->total_bytes > num_bytes) { |
5645 | num_bytes = sinfo->total_bytes - num_bytes; | |
5646 | num_bytes = min(num_bytes, | |
5647 | block_rsv->size - block_rsv->reserved); | |
5648 | block_rsv->reserved += num_bytes; | |
5649 | sinfo->bytes_may_use += num_bytes; | |
5650 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5651 | sinfo->flags, num_bytes, | |
5652 | 1); | |
5653 | } | |
5654 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5655 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5656 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5657 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5658 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5659 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5660 | } |
182608c8 | 5661 | |
fb4b10e5 JB |
5662 | if (block_rsv->reserved == block_rsv->size) |
5663 | block_rsv->full = 1; | |
5664 | else | |
5665 | block_rsv->full = 0; | |
5666 | ||
8929ecfa | 5667 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5668 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5669 | } |
5670 | ||
f0486c68 | 5671 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5672 | { |
f0486c68 | 5673 | struct btrfs_space_info *space_info; |
6a63209f | 5674 | |
f0486c68 YZ |
5675 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5676 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5677 | |
f0486c68 | 5678 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5679 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 5680 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
5681 | fs_info->trans_block_rsv.space_info = space_info; |
5682 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5683 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5684 | |
8929ecfa YZ |
5685 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5686 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5687 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5688 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5689 | if (fs_info->quota_root) |
5690 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5691 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5692 | |
8929ecfa | 5693 | update_global_block_rsv(fs_info); |
6a63209f JB |
5694 | } |
5695 | ||
8929ecfa | 5696 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5697 | { |
8c2a3ca2 JB |
5698 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5699 | (u64)-1); | |
8929ecfa YZ |
5700 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
5701 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
5702 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
5703 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5704 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5705 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5706 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5707 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5708 | } |
5709 | ||
a22285a6 | 5710 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
2ff7e61e | 5711 | struct btrfs_fs_info *fs_info) |
6a63209f | 5712 | { |
0e721106 JB |
5713 | if (!trans->block_rsv) |
5714 | return; | |
5715 | ||
a22285a6 YZ |
5716 | if (!trans->bytes_reserved) |
5717 | return; | |
6a63209f | 5718 | |
0b246afa | 5719 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 5720 | trans->transid, trans->bytes_reserved, 0); |
2ff7e61e JM |
5721 | btrfs_block_rsv_release(fs_info, trans->block_rsv, |
5722 | trans->bytes_reserved); | |
a22285a6 YZ |
5723 | trans->bytes_reserved = 0; |
5724 | } | |
6a63209f | 5725 | |
4fbcdf66 FM |
5726 | /* |
5727 | * To be called after all the new block groups attached to the transaction | |
5728 | * handle have been created (btrfs_create_pending_block_groups()). | |
5729 | */ | |
5730 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5731 | { | |
64b63580 | 5732 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5733 | |
5734 | if (!trans->chunk_bytes_reserved) | |
5735 | return; | |
5736 | ||
5737 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5738 | ||
5739 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5740 | trans->chunk_bytes_reserved); | |
5741 | trans->chunk_bytes_reserved = 0; | |
5742 | } | |
5743 | ||
79787eaa | 5744 | /* Can only return 0 or -ENOSPC */ |
d68fc57b | 5745 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
8ed7a2a0 | 5746 | struct btrfs_inode *inode) |
d68fc57b | 5747 | { |
8ed7a2a0 NB |
5748 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5749 | struct btrfs_root *root = inode->root; | |
40acc3ee JB |
5750 | /* |
5751 | * We always use trans->block_rsv here as we will have reserved space | |
5752 | * for our orphan when starting the transaction, using get_block_rsv() | |
5753 | * here will sometimes make us choose the wrong block rsv as we could be | |
5754 | * doing a reloc inode for a non refcounted root. | |
5755 | */ | |
5756 | struct btrfs_block_rsv *src_rsv = trans->block_rsv; | |
d68fc57b YZ |
5757 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; |
5758 | ||
5759 | /* | |
fcb80c2a JB |
5760 | * We need to hold space in order to delete our orphan item once we've |
5761 | * added it, so this takes the reservation so we can release it later | |
5762 | * when we are truly done with the orphan item. | |
d68fc57b | 5763 | */ |
0b246afa JM |
5764 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5765 | ||
8ed7a2a0 NB |
5766 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5767 | num_bytes, 1); | |
25d609f8 | 5768 | return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); |
6a63209f JB |
5769 | } |
5770 | ||
703b391a | 5771 | void btrfs_orphan_release_metadata(struct btrfs_inode *inode) |
97e728d4 | 5772 | { |
703b391a NB |
5773 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5774 | struct btrfs_root *root = inode->root; | |
0b246afa JM |
5775 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5776 | ||
703b391a NB |
5777 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5778 | num_bytes, 0); | |
2ff7e61e | 5779 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes); |
d68fc57b | 5780 | } |
97e728d4 | 5781 | |
d5c12070 MX |
5782 | /* |
5783 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5784 | * root: the root of the parent directory | |
5785 | * rsv: block reservation | |
5786 | * items: the number of items that we need do reservation | |
5787 | * qgroup_reserved: used to return the reserved size in qgroup | |
5788 | * | |
5789 | * This function is used to reserve the space for snapshot/subvolume | |
5790 | * creation and deletion. Those operations are different with the | |
5791 | * common file/directory operations, they change two fs/file trees | |
5792 | * and root tree, the number of items that the qgroup reserves is | |
5793 | * different with the free space reservation. So we can not use | |
01327610 | 5794 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5795 | */ |
5796 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5797 | struct btrfs_block_rsv *rsv, | |
5798 | int items, | |
ee3441b4 JM |
5799 | u64 *qgroup_reserved, |
5800 | bool use_global_rsv) | |
a22285a6 | 5801 | { |
d5c12070 MX |
5802 | u64 num_bytes; |
5803 | int ret; | |
0b246afa JM |
5804 | struct btrfs_fs_info *fs_info = root->fs_info; |
5805 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5806 | |
0b246afa | 5807 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5808 | /* One for parent inode, two for dir entries */ |
0b246afa | 5809 | num_bytes = 3 * fs_info->nodesize; |
003d7c59 | 5810 | ret = btrfs_qgroup_reserve_meta(root, num_bytes, true); |
d5c12070 MX |
5811 | if (ret) |
5812 | return ret; | |
5813 | } else { | |
5814 | num_bytes = 0; | |
5815 | } | |
5816 | ||
5817 | *qgroup_reserved = num_bytes; | |
5818 | ||
0b246afa JM |
5819 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
5820 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
5821 | BTRFS_BLOCK_GROUP_METADATA); |
5822 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5823 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5824 | |
5825 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 5826 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 5827 | |
7174109c QW |
5828 | if (ret && *qgroup_reserved) |
5829 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
d5c12070 MX |
5830 | |
5831 | return ret; | |
5832 | } | |
5833 | ||
2ff7e61e | 5834 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 5835 | struct btrfs_block_rsv *rsv) |
d5c12070 | 5836 | { |
2ff7e61e | 5837 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
5838 | } |
5839 | ||
7709cde3 JB |
5840 | /** |
5841 | * drop_outstanding_extent - drop an outstanding extent | |
5842 | * @inode: the inode we're dropping the extent for | |
01327610 | 5843 | * @num_bytes: the number of bytes we're releasing. |
7709cde3 JB |
5844 | * |
5845 | * This is called when we are freeing up an outstanding extent, either called | |
5846 | * after an error or after an extent is written. This will return the number of | |
5847 | * reserved extents that need to be freed. This must be called with | |
5848 | * BTRFS_I(inode)->lock held. | |
5849 | */ | |
baa3ba39 NB |
5850 | static unsigned drop_outstanding_extent(struct btrfs_inode *inode, |
5851 | u64 num_bytes) | |
9e0baf60 | 5852 | { |
7fd2ae21 | 5853 | unsigned drop_inode_space = 0; |
9e0baf60 | 5854 | unsigned dropped_extents = 0; |
823bb20a | 5855 | unsigned num_extents; |
9e0baf60 | 5856 | |
823bb20a | 5857 | num_extents = count_max_extents(num_bytes); |
dcab6a3b | 5858 | ASSERT(num_extents); |
baa3ba39 NB |
5859 | ASSERT(inode->outstanding_extents >= num_extents); |
5860 | inode->outstanding_extents -= num_extents; | |
9e0baf60 | 5861 | |
baa3ba39 | 5862 | if (inode->outstanding_extents == 0 && |
72ac3c0d | 5863 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
baa3ba39 | 5864 | &inode->runtime_flags)) |
7fd2ae21 | 5865 | drop_inode_space = 1; |
7fd2ae21 | 5866 | |
9e0baf60 | 5867 | /* |
01327610 | 5868 | * If we have more or the same amount of outstanding extents than we have |
9e0baf60 JB |
5869 | * reserved then we need to leave the reserved extents count alone. |
5870 | */ | |
baa3ba39 | 5871 | if (inode->outstanding_extents >= inode->reserved_extents) |
7fd2ae21 | 5872 | return drop_inode_space; |
9e0baf60 | 5873 | |
baa3ba39 NB |
5874 | dropped_extents = inode->reserved_extents - inode->outstanding_extents; |
5875 | inode->reserved_extents -= dropped_extents; | |
7fd2ae21 | 5876 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
5877 | } |
5878 | ||
7709cde3 | 5879 | /** |
01327610 NS |
5880 | * calc_csum_metadata_size - return the amount of metadata space that must be |
5881 | * reserved/freed for the given bytes. | |
7709cde3 JB |
5882 | * @inode: the inode we're manipulating |
5883 | * @num_bytes: the number of bytes in question | |
5884 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
5885 | * | |
5886 | * This adjusts the number of csum_bytes in the inode and then returns the | |
5887 | * correct amount of metadata that must either be reserved or freed. We | |
5888 | * calculate how many checksums we can fit into one leaf and then divide the | |
5889 | * number of bytes that will need to be checksumed by this value to figure out | |
5890 | * how many checksums will be required. If we are adding bytes then the number | |
5891 | * may go up and we will return the number of additional bytes that must be | |
5892 | * reserved. If it is going down we will return the number of bytes that must | |
5893 | * be freed. | |
5894 | * | |
5895 | * This must be called with BTRFS_I(inode)->lock held. | |
5896 | */ | |
0e6bf9b1 | 5897 | static u64 calc_csum_metadata_size(struct btrfs_inode *inode, u64 num_bytes, |
7709cde3 | 5898 | int reserve) |
6324fbf3 | 5899 | { |
0e6bf9b1 | 5900 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
1262133b | 5901 | u64 old_csums, num_csums; |
7709cde3 | 5902 | |
0e6bf9b1 | 5903 | if (inode->flags & BTRFS_INODE_NODATASUM && inode->csum_bytes == 0) |
7709cde3 JB |
5904 | return 0; |
5905 | ||
0e6bf9b1 | 5906 | old_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes); |
7709cde3 | 5907 | if (reserve) |
0e6bf9b1 | 5908 | inode->csum_bytes += num_bytes; |
7709cde3 | 5909 | else |
0e6bf9b1 NB |
5910 | inode->csum_bytes -= num_bytes; |
5911 | num_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes); | |
7709cde3 JB |
5912 | |
5913 | /* No change, no need to reserve more */ | |
5914 | if (old_csums == num_csums) | |
5915 | return 0; | |
5916 | ||
5917 | if (reserve) | |
0b246afa | 5918 | return btrfs_calc_trans_metadata_size(fs_info, |
7709cde3 JB |
5919 | num_csums - old_csums); |
5920 | ||
0b246afa | 5921 | return btrfs_calc_trans_metadata_size(fs_info, old_csums - num_csums); |
0ca1f7ce | 5922 | } |
c146afad | 5923 | |
9f3db423 | 5924 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 5925 | { |
9f3db423 NB |
5926 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5927 | struct btrfs_root *root = inode->root; | |
0b246afa | 5928 | struct btrfs_block_rsv *block_rsv = &fs_info->delalloc_block_rsv; |
9e0baf60 | 5929 | u64 to_reserve = 0; |
660d3f6c | 5930 | u64 csum_bytes; |
823bb20a | 5931 | unsigned nr_extents; |
08e007d2 | 5932 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5933 | int ret = 0; |
c64c2bd8 | 5934 | bool delalloc_lock = true; |
88e081bf WS |
5935 | u64 to_free = 0; |
5936 | unsigned dropped; | |
48c3d480 | 5937 | bool release_extra = false; |
6324fbf3 | 5938 | |
c64c2bd8 JB |
5939 | /* If we are a free space inode we need to not flush since we will be in |
5940 | * the middle of a transaction commit. We also don't need the delalloc | |
5941 | * mutex since we won't race with anybody. We need this mostly to make | |
5942 | * lockdep shut its filthy mouth. | |
bac357dc JB |
5943 | * |
5944 | * If we have a transaction open (can happen if we call truncate_block | |
5945 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
5946 | */ |
5947 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 5948 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 5949 | delalloc_lock = false; |
bac357dc JB |
5950 | } else if (current->journal_info) { |
5951 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c64c2bd8 | 5952 | } |
c09544e0 | 5953 | |
08e007d2 | 5954 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
0b246afa | 5955 | btrfs_transaction_in_commit(fs_info)) |
0ca1f7ce | 5956 | schedule_timeout(1); |
ec44a35c | 5957 | |
c64c2bd8 | 5958 | if (delalloc_lock) |
9f3db423 | 5959 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 5960 | |
0b246afa | 5961 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
8bb8ab2e | 5962 | |
9f3db423 | 5963 | spin_lock(&inode->lock); |
823bb20a | 5964 | nr_extents = count_max_extents(num_bytes); |
9f3db423 | 5965 | inode->outstanding_extents += nr_extents; |
9e0baf60 | 5966 | |
48c3d480 | 5967 | nr_extents = 0; |
9f3db423 NB |
5968 | if (inode->outstanding_extents > inode->reserved_extents) |
5969 | nr_extents += inode->outstanding_extents - | |
5970 | inode->reserved_extents; | |
57a45ced | 5971 | |
48c3d480 | 5972 | /* We always want to reserve a slot for updating the inode. */ |
0b246afa | 5973 | to_reserve = btrfs_calc_trans_metadata_size(fs_info, nr_extents + 1); |
7709cde3 | 5974 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
9f3db423 NB |
5975 | csum_bytes = inode->csum_bytes; |
5976 | spin_unlock(&inode->lock); | |
57a45ced | 5977 | |
0b246afa | 5978 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
7174109c | 5979 | ret = btrfs_qgroup_reserve_meta(root, |
003d7c59 | 5980 | nr_extents * fs_info->nodesize, true); |
88e081bf WS |
5981 | if (ret) |
5982 | goto out_fail; | |
5983 | } | |
c5567237 | 5984 | |
48c3d480 | 5985 | ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush); |
88e081bf | 5986 | if (unlikely(ret)) { |
da17066c | 5987 | btrfs_qgroup_free_meta(root, |
0b246afa | 5988 | nr_extents * fs_info->nodesize); |
88e081bf | 5989 | goto out_fail; |
9e0baf60 | 5990 | } |
25179201 | 5991 | |
9f3db423 | 5992 | spin_lock(&inode->lock); |
48c3d480 | 5993 | if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
9f3db423 | 5994 | &inode->runtime_flags)) { |
0b246afa | 5995 | to_reserve -= btrfs_calc_trans_metadata_size(fs_info, 1); |
48c3d480 | 5996 | release_extra = true; |
660d3f6c | 5997 | } |
9f3db423 NB |
5998 | inode->reserved_extents += nr_extents; |
5999 | spin_unlock(&inode->lock); | |
c64c2bd8 JB |
6000 | |
6001 | if (delalloc_lock) | |
9f3db423 | 6002 | mutex_unlock(&inode->delalloc_mutex); |
660d3f6c | 6003 | |
8c2a3ca2 | 6004 | if (to_reserve) |
0b246afa | 6005 | trace_btrfs_space_reservation(fs_info, "delalloc", |
9f3db423 | 6006 | btrfs_ino(inode), to_reserve, 1); |
48c3d480 | 6007 | if (release_extra) |
2ff7e61e | 6008 | btrfs_block_rsv_release(fs_info, block_rsv, |
0b246afa | 6009 | btrfs_calc_trans_metadata_size(fs_info, 1)); |
0ca1f7ce | 6010 | return 0; |
88e081bf WS |
6011 | |
6012 | out_fail: | |
9f3db423 | 6013 | spin_lock(&inode->lock); |
dcab6a3b | 6014 | dropped = drop_outstanding_extent(inode, num_bytes); |
88e081bf WS |
6015 | /* |
6016 | * If the inodes csum_bytes is the same as the original | |
6017 | * csum_bytes then we know we haven't raced with any free()ers | |
6018 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 6019 | */ |
9f3db423 | 6020 | if (inode->csum_bytes == csum_bytes) { |
88e081bf | 6021 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 | 6022 | } else { |
9f3db423 | 6023 | u64 orig_csum_bytes = inode->csum_bytes; |
f4881bc7 JB |
6024 | u64 bytes; |
6025 | ||
6026 | /* | |
6027 | * This is tricky, but first we need to figure out how much we | |
01327610 | 6028 | * freed from any free-ers that occurred during this |
f4881bc7 JB |
6029 | * reservation, so we reset ->csum_bytes to the csum_bytes |
6030 | * before we dropped our lock, and then call the free for the | |
6031 | * number of bytes that were freed while we were trying our | |
6032 | * reservation. | |
6033 | */ | |
9f3db423 NB |
6034 | bytes = csum_bytes - inode->csum_bytes; |
6035 | inode->csum_bytes = csum_bytes; | |
f4881bc7 JB |
6036 | to_free = calc_csum_metadata_size(inode, bytes, 0); |
6037 | ||
6038 | ||
6039 | /* | |
6040 | * Now we need to see how much we would have freed had we not | |
6041 | * been making this reservation and our ->csum_bytes were not | |
6042 | * artificially inflated. | |
6043 | */ | |
9f3db423 | 6044 | inode->csum_bytes = csum_bytes - num_bytes; |
f4881bc7 JB |
6045 | bytes = csum_bytes - orig_csum_bytes; |
6046 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
6047 | ||
6048 | /* | |
6049 | * Now reset ->csum_bytes to what it should be. If bytes is | |
01327610 | 6050 | * more than to_free then we would have freed more space had we |
f4881bc7 JB |
6051 | * not had an artificially high ->csum_bytes, so we need to free |
6052 | * the remainder. If bytes is the same or less then we don't | |
6053 | * need to do anything, the other free-ers did the correct | |
6054 | * thing. | |
6055 | */ | |
9f3db423 | 6056 | inode->csum_bytes = orig_csum_bytes - num_bytes; |
f4881bc7 JB |
6057 | if (bytes > to_free) |
6058 | to_free = bytes - to_free; | |
6059 | else | |
6060 | to_free = 0; | |
6061 | } | |
9f3db423 | 6062 | spin_unlock(&inode->lock); |
e2d1f923 | 6063 | if (dropped) |
0b246afa | 6064 | to_free += btrfs_calc_trans_metadata_size(fs_info, dropped); |
88e081bf WS |
6065 | |
6066 | if (to_free) { | |
2ff7e61e | 6067 | btrfs_block_rsv_release(fs_info, block_rsv, to_free); |
0b246afa | 6068 | trace_btrfs_space_reservation(fs_info, "delalloc", |
9f3db423 | 6069 | btrfs_ino(inode), to_free, 0); |
88e081bf WS |
6070 | } |
6071 | if (delalloc_lock) | |
9f3db423 | 6072 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6073 | return ret; |
0ca1f7ce YZ |
6074 | } |
6075 | ||
7709cde3 JB |
6076 | /** |
6077 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
6078 | * @inode: the inode to release the reservation for | |
6079 | * @num_bytes: the number of bytes we're releasing | |
6080 | * | |
6081 | * This will release the metadata reservation for an inode. This can be called | |
6082 | * once we complete IO for a given set of bytes to release their metadata | |
6083 | * reservations. | |
6084 | */ | |
691fa059 | 6085 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6086 | { |
691fa059 | 6087 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
9e0baf60 JB |
6088 | u64 to_free = 0; |
6089 | unsigned dropped; | |
0ca1f7ce | 6090 | |
0b246afa | 6091 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6092 | spin_lock(&inode->lock); |
dcab6a3b | 6093 | dropped = drop_outstanding_extent(inode, num_bytes); |
97e728d4 | 6094 | |
0934856d MX |
6095 | if (num_bytes) |
6096 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
691fa059 | 6097 | spin_unlock(&inode->lock); |
9e0baf60 | 6098 | if (dropped > 0) |
0b246afa | 6099 | to_free += btrfs_calc_trans_metadata_size(fs_info, dropped); |
0ca1f7ce | 6100 | |
0b246afa | 6101 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6102 | return; |
6103 | ||
691fa059 NB |
6104 | trace_btrfs_space_reservation(fs_info, "delalloc", btrfs_ino(inode), |
6105 | to_free, 0); | |
c5567237 | 6106 | |
2ff7e61e | 6107 | btrfs_block_rsv_release(fs_info, &fs_info->delalloc_block_rsv, to_free); |
0ca1f7ce YZ |
6108 | } |
6109 | ||
1ada3a62 | 6110 | /** |
7cf5b976 | 6111 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6112 | * delalloc |
6113 | * @inode: inode we're writing to | |
6114 | * @start: start range we are writing to | |
6115 | * @len: how long the range we are writing to | |
6116 | * | |
1ada3a62 QW |
6117 | * This will do the following things |
6118 | * | |
6119 | * o reserve space in data space info for num bytes | |
6120 | * and reserve precious corresponding qgroup space | |
6121 | * (Done in check_data_free_space) | |
6122 | * | |
6123 | * o reserve space for metadata space, based on the number of outstanding | |
6124 | * extents and how much csums will be needed | |
6125 | * also reserve metadata space in a per root over-reserve method. | |
6126 | * o add to the inodes->delalloc_bytes | |
6127 | * o add it to the fs_info's delalloc inodes list. | |
6128 | * (Above 3 all done in delalloc_reserve_metadata) | |
6129 | * | |
6130 | * Return 0 for success | |
6131 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6132 | */ | |
7cf5b976 | 6133 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
6134 | { |
6135 | int ret; | |
6136 | ||
7cf5b976 | 6137 | ret = btrfs_check_data_free_space(inode, start, len); |
1ada3a62 QW |
6138 | if (ret < 0) |
6139 | return ret; | |
9f3db423 | 6140 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6141 | if (ret < 0) |
7cf5b976 | 6142 | btrfs_free_reserved_data_space(inode, start, len); |
1ada3a62 QW |
6143 | return ret; |
6144 | } | |
6145 | ||
7709cde3 | 6146 | /** |
7cf5b976 | 6147 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6148 | * @inode: inode we're releasing space for |
6149 | * @start: start position of the space already reserved | |
6150 | * @len: the len of the space already reserved | |
6151 | * | |
6152 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
6153 | * called in the case that we don't need the metadata AND data reservations | |
6154 | * anymore. So if there is an error or we insert an inline extent. | |
6155 | * | |
6156 | * This function will release the metadata space that was not used and will | |
6157 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6158 | * list if there are no delalloc bytes left. | |
6159 | * Also it will handle the qgroup reserved space. | |
6160 | */ | |
7cf5b976 | 6161 | void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 | 6162 | { |
691fa059 | 6163 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len); |
7cf5b976 | 6164 | btrfs_free_reserved_data_space(inode, start, len); |
6324fbf3 CM |
6165 | } |
6166 | ||
ce93ec54 | 6167 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6168 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6169 | u64 num_bytes, int alloc) |
9078a3e1 | 6170 | { |
0af3d00b | 6171 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6172 | u64 total = num_bytes; |
9078a3e1 | 6173 | u64 old_val; |
db94535d | 6174 | u64 byte_in_group; |
0af3d00b | 6175 | int factor; |
3e1ad54f | 6176 | |
5d4f98a2 | 6177 | /* block accounting for super block */ |
eb73c1b7 | 6178 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6179 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6180 | if (alloc) |
6181 | old_val += num_bytes; | |
6182 | else | |
6183 | old_val -= num_bytes; | |
6c41761f | 6184 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6185 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6186 | |
d397712b | 6187 | while (total) { |
db94535d | 6188 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6189 | if (!cache) |
79787eaa | 6190 | return -ENOENT; |
b742bb82 YZ |
6191 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6192 | BTRFS_BLOCK_GROUP_RAID1 | | |
6193 | BTRFS_BLOCK_GROUP_RAID10)) | |
6194 | factor = 2; | |
6195 | else | |
6196 | factor = 1; | |
9d66e233 JB |
6197 | /* |
6198 | * If this block group has free space cache written out, we | |
6199 | * need to make sure to load it if we are removing space. This | |
6200 | * is because we need the unpinning stage to actually add the | |
6201 | * space back to the block group, otherwise we will leak space. | |
6202 | */ | |
6203 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6204 | cache_block_group(cache, 1); |
0af3d00b | 6205 | |
db94535d CM |
6206 | byte_in_group = bytenr - cache->key.objectid; |
6207 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6208 | |
25179201 | 6209 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6210 | spin_lock(&cache->lock); |
0af3d00b | 6211 | |
6202df69 | 6212 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6213 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6214 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6215 | ||
9078a3e1 | 6216 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6217 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6218 | if (alloc) { |
db94535d | 6219 | old_val += num_bytes; |
11833d66 YZ |
6220 | btrfs_set_block_group_used(&cache->item, old_val); |
6221 | cache->reserved -= num_bytes; | |
11833d66 | 6222 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6223 | cache->space_info->bytes_used += num_bytes; |
6224 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6225 | spin_unlock(&cache->lock); |
25179201 | 6226 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6227 | } else { |
db94535d | 6228 | old_val -= num_bytes; |
ae0ab003 FM |
6229 | btrfs_set_block_group_used(&cache->item, old_val); |
6230 | cache->pinned += num_bytes; | |
6231 | cache->space_info->bytes_pinned += num_bytes; | |
6232 | cache->space_info->bytes_used -= num_bytes; | |
6233 | cache->space_info->disk_used -= num_bytes * factor; | |
6234 | spin_unlock(&cache->lock); | |
6235 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6236 | |
0b246afa | 6237 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6238 | cache->space_info->flags, |
6239 | num_bytes, 1); | |
ae0ab003 FM |
6240 | set_extent_dirty(info->pinned_extents, |
6241 | bytenr, bytenr + num_bytes - 1, | |
6242 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6243 | } |
1bbc621e CM |
6244 | |
6245 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6246 | if (list_empty(&cache->dirty_list)) { | |
6247 | list_add_tail(&cache->dirty_list, | |
6248 | &trans->transaction->dirty_bgs); | |
6249 | trans->transaction->num_dirty_bgs++; | |
6250 | btrfs_get_block_group(cache); | |
6251 | } | |
6252 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6253 | ||
036a9348 FM |
6254 | /* |
6255 | * No longer have used bytes in this block group, queue it for | |
6256 | * deletion. We do this after adding the block group to the | |
6257 | * dirty list to avoid races between cleaner kthread and space | |
6258 | * cache writeout. | |
6259 | */ | |
6260 | if (!alloc && old_val == 0) { | |
6261 | spin_lock(&info->unused_bgs_lock); | |
6262 | if (list_empty(&cache->bg_list)) { | |
6263 | btrfs_get_block_group(cache); | |
6264 | list_add_tail(&cache->bg_list, | |
6265 | &info->unused_bgs); | |
6266 | } | |
6267 | spin_unlock(&info->unused_bgs_lock); | |
6268 | } | |
6269 | ||
fa9c0d79 | 6270 | btrfs_put_block_group(cache); |
db94535d CM |
6271 | total -= num_bytes; |
6272 | bytenr += num_bytes; | |
9078a3e1 CM |
6273 | } |
6274 | return 0; | |
6275 | } | |
6324fbf3 | 6276 | |
2ff7e61e | 6277 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6278 | { |
0f9dd46c | 6279 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6280 | u64 bytenr; |
0f9dd46c | 6281 | |
0b246afa JM |
6282 | spin_lock(&fs_info->block_group_cache_lock); |
6283 | bytenr = fs_info->first_logical_byte; | |
6284 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6285 | |
6286 | if (bytenr < (u64)-1) | |
6287 | return bytenr; | |
6288 | ||
0b246afa | 6289 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6290 | if (!cache) |
a061fc8d | 6291 | return 0; |
0f9dd46c | 6292 | |
d2fb3437 | 6293 | bytenr = cache->key.objectid; |
fa9c0d79 | 6294 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6295 | |
6296 | return bytenr; | |
a061fc8d CM |
6297 | } |
6298 | ||
2ff7e61e | 6299 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6300 | struct btrfs_block_group_cache *cache, |
6301 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6302 | { |
11833d66 YZ |
6303 | spin_lock(&cache->space_info->lock); |
6304 | spin_lock(&cache->lock); | |
6305 | cache->pinned += num_bytes; | |
6306 | cache->space_info->bytes_pinned += num_bytes; | |
6307 | if (reserved) { | |
6308 | cache->reserved -= num_bytes; | |
6309 | cache->space_info->bytes_reserved -= num_bytes; | |
6310 | } | |
6311 | spin_unlock(&cache->lock); | |
6312 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6313 | |
0b246afa | 6314 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6315 | cache->space_info->flags, num_bytes, 1); |
0b246afa | 6316 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6317 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6318 | return 0; | |
6319 | } | |
68b38550 | 6320 | |
f0486c68 YZ |
6321 | /* |
6322 | * this function must be called within transaction | |
6323 | */ | |
2ff7e61e | 6324 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6325 | u64 bytenr, u64 num_bytes, int reserved) |
6326 | { | |
6327 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6328 | |
0b246afa | 6329 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6330 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6331 | |
2ff7e61e | 6332 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6333 | |
6334 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6335 | return 0; |
6336 | } | |
6337 | ||
f0486c68 | 6338 | /* |
e688b725 CM |
6339 | * this function must be called within transaction |
6340 | */ | |
2ff7e61e | 6341 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6342 | u64 bytenr, u64 num_bytes) |
6343 | { | |
6344 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6345 | int ret; |
e688b725 | 6346 | |
0b246afa | 6347 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6348 | if (!cache) |
6349 | return -EINVAL; | |
e688b725 CM |
6350 | |
6351 | /* | |
6352 | * pull in the free space cache (if any) so that our pin | |
6353 | * removes the free space from the cache. We have load_only set | |
6354 | * to one because the slow code to read in the free extents does check | |
6355 | * the pinned extents. | |
6356 | */ | |
f6373bf3 | 6357 | cache_block_group(cache, 1); |
e688b725 | 6358 | |
2ff7e61e | 6359 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6360 | |
6361 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6362 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6363 | btrfs_put_block_group(cache); |
b50c6e25 | 6364 | return ret; |
e688b725 CM |
6365 | } |
6366 | ||
2ff7e61e JM |
6367 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6368 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6369 | { |
6370 | int ret; | |
6371 | struct btrfs_block_group_cache *block_group; | |
6372 | struct btrfs_caching_control *caching_ctl; | |
6373 | ||
0b246afa | 6374 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6375 | if (!block_group) |
6376 | return -EINVAL; | |
6377 | ||
6378 | cache_block_group(block_group, 0); | |
6379 | caching_ctl = get_caching_control(block_group); | |
6380 | ||
6381 | if (!caching_ctl) { | |
6382 | /* Logic error */ | |
6383 | BUG_ON(!block_group_cache_done(block_group)); | |
6384 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6385 | } else { | |
6386 | mutex_lock(&caching_ctl->mutex); | |
6387 | ||
6388 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6389 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6390 | } else if (start + num_bytes <= caching_ctl->progress) { |
6391 | ret = btrfs_remove_free_space(block_group, | |
6392 | start, num_bytes); | |
6393 | } else { | |
6394 | num_bytes = caching_ctl->progress - start; | |
6395 | ret = btrfs_remove_free_space(block_group, | |
6396 | start, num_bytes); | |
6397 | if (ret) | |
6398 | goto out_lock; | |
6399 | ||
6400 | num_bytes = (start + num_bytes) - | |
6401 | caching_ctl->progress; | |
6402 | start = caching_ctl->progress; | |
2ff7e61e | 6403 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6404 | } |
6405 | out_lock: | |
6406 | mutex_unlock(&caching_ctl->mutex); | |
6407 | put_caching_control(caching_ctl); | |
6408 | } | |
6409 | btrfs_put_block_group(block_group); | |
6410 | return ret; | |
6411 | } | |
6412 | ||
2ff7e61e | 6413 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6414 | struct extent_buffer *eb) |
6415 | { | |
6416 | struct btrfs_file_extent_item *item; | |
6417 | struct btrfs_key key; | |
6418 | int found_type; | |
6419 | int i; | |
6420 | ||
2ff7e61e | 6421 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6422 | return 0; |
6423 | ||
6424 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6425 | btrfs_item_key_to_cpu(eb, &key, i); | |
6426 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6427 | continue; | |
6428 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6429 | found_type = btrfs_file_extent_type(eb, item); | |
6430 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6431 | continue; | |
6432 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6433 | continue; | |
6434 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6435 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
2ff7e61e | 6436 | __exclude_logged_extent(fs_info, key.objectid, key.offset); |
8c2a1a30 JB |
6437 | } |
6438 | ||
6439 | return 0; | |
6440 | } | |
6441 | ||
9cfa3e34 FM |
6442 | static void |
6443 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6444 | { | |
6445 | atomic_inc(&bg->reservations); | |
6446 | } | |
6447 | ||
6448 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6449 | const u64 start) | |
6450 | { | |
6451 | struct btrfs_block_group_cache *bg; | |
6452 | ||
6453 | bg = btrfs_lookup_block_group(fs_info, start); | |
6454 | ASSERT(bg); | |
6455 | if (atomic_dec_and_test(&bg->reservations)) | |
6456 | wake_up_atomic_t(&bg->reservations); | |
6457 | btrfs_put_block_group(bg); | |
6458 | } | |
6459 | ||
6460 | static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a) | |
6461 | { | |
6462 | schedule(); | |
6463 | return 0; | |
6464 | } | |
6465 | ||
6466 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6467 | { | |
6468 | struct btrfs_space_info *space_info = bg->space_info; | |
6469 | ||
6470 | ASSERT(bg->ro); | |
6471 | ||
6472 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6473 | return; | |
6474 | ||
6475 | /* | |
6476 | * Our block group is read only but before we set it to read only, | |
6477 | * some task might have had allocated an extent from it already, but it | |
6478 | * has not yet created a respective ordered extent (and added it to a | |
6479 | * root's list of ordered extents). | |
6480 | * Therefore wait for any task currently allocating extents, since the | |
6481 | * block group's reservations counter is incremented while a read lock | |
6482 | * on the groups' semaphore is held and decremented after releasing | |
6483 | * the read access on that semaphore and creating the ordered extent. | |
6484 | */ | |
6485 | down_write(&space_info->groups_sem); | |
6486 | up_write(&space_info->groups_sem); | |
6487 | ||
6488 | wait_on_atomic_t(&bg->reservations, | |
6489 | btrfs_wait_bg_reservations_atomic_t, | |
6490 | TASK_UNINTERRUPTIBLE); | |
6491 | } | |
6492 | ||
fb25e914 | 6493 | /** |
4824f1f4 | 6494 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6495 | * @cache: The cache we are manipulating |
18513091 WX |
6496 | * @ram_bytes: The number of bytes of file content, and will be same to |
6497 | * @num_bytes except for the compress path. | |
fb25e914 | 6498 | * @num_bytes: The number of bytes in question |
e570fd27 | 6499 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6500 | * |
745699ef XW |
6501 | * This is called by the allocator when it reserves space. If this is a |
6502 | * reservation and the block group has become read only we cannot make the | |
6503 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6504 | */ |
4824f1f4 | 6505 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6506 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6507 | { |
fb25e914 | 6508 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6509 | int ret = 0; |
79787eaa | 6510 | |
fb25e914 JB |
6511 | spin_lock(&space_info->lock); |
6512 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6513 | if (cache->ro) { |
6514 | ret = -EAGAIN; | |
fb25e914 | 6515 | } else { |
4824f1f4 WX |
6516 | cache->reserved += num_bytes; |
6517 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6518 | |
18513091 WX |
6519 | trace_btrfs_space_reservation(cache->fs_info, |
6520 | "space_info", space_info->flags, | |
6521 | ram_bytes, 0); | |
6522 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6523 | if (delalloc) |
4824f1f4 | 6524 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6525 | } |
fb25e914 JB |
6526 | spin_unlock(&cache->lock); |
6527 | spin_unlock(&space_info->lock); | |
f0486c68 | 6528 | return ret; |
324ae4df | 6529 | } |
9078a3e1 | 6530 | |
4824f1f4 WX |
6531 | /** |
6532 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6533 | * @cache: The cache we are manipulating | |
6534 | * @num_bytes: The number of bytes in question | |
6535 | * @delalloc: The blocks are allocated for the delalloc write | |
6536 | * | |
6537 | * This is called by somebody who is freeing space that was never actually used | |
6538 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6539 | * A and before transaction A commits you free that leaf, you call this with | |
6540 | * reserve set to 0 in order to clear the reservation. | |
6541 | */ | |
6542 | ||
6543 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6544 | u64 num_bytes, int delalloc) | |
6545 | { | |
6546 | struct btrfs_space_info *space_info = cache->space_info; | |
6547 | int ret = 0; | |
6548 | ||
6549 | spin_lock(&space_info->lock); | |
6550 | spin_lock(&cache->lock); | |
6551 | if (cache->ro) | |
6552 | space_info->bytes_readonly += num_bytes; | |
6553 | cache->reserved -= num_bytes; | |
6554 | space_info->bytes_reserved -= num_bytes; | |
6555 | ||
6556 | if (delalloc) | |
6557 | cache->delalloc_bytes -= num_bytes; | |
6558 | spin_unlock(&cache->lock); | |
6559 | spin_unlock(&space_info->lock); | |
6560 | return ret; | |
6561 | } | |
8b74c03e | 6562 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6563 | { |
11833d66 YZ |
6564 | struct btrfs_caching_control *next; |
6565 | struct btrfs_caching_control *caching_ctl; | |
6566 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6567 | |
9e351cc8 | 6568 | down_write(&fs_info->commit_root_sem); |
25179201 | 6569 | |
11833d66 YZ |
6570 | list_for_each_entry_safe(caching_ctl, next, |
6571 | &fs_info->caching_block_groups, list) { | |
6572 | cache = caching_ctl->block_group; | |
6573 | if (block_group_cache_done(cache)) { | |
6574 | cache->last_byte_to_unpin = (u64)-1; | |
6575 | list_del_init(&caching_ctl->list); | |
6576 | put_caching_control(caching_ctl); | |
e8569813 | 6577 | } else { |
11833d66 | 6578 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6579 | } |
e8569813 | 6580 | } |
11833d66 YZ |
6581 | |
6582 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6583 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6584 | else | |
6585 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6586 | ||
9e351cc8 | 6587 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6588 | |
6589 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6590 | } |
6591 | ||
c759c4e1 JB |
6592 | /* |
6593 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6594 | * what it should be based on the mount options. | |
6595 | */ | |
6596 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6597 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6598 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6599 | { |
6600 | struct btrfs_free_cluster *ret = NULL; | |
0b246afa | 6601 | bool ssd = btrfs_test_opt(fs_info, SSD); |
c759c4e1 JB |
6602 | |
6603 | *empty_cluster = 0; | |
6604 | if (btrfs_mixed_space_info(space_info)) | |
6605 | return ret; | |
6606 | ||
6607 | if (ssd) | |
ee22184b | 6608 | *empty_cluster = SZ_2M; |
c759c4e1 | 6609 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6610 | ret = &fs_info->meta_alloc_cluster; |
c759c4e1 | 6611 | if (!ssd) |
ee22184b | 6612 | *empty_cluster = SZ_64K; |
c759c4e1 | 6613 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) { |
0b246afa | 6614 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6615 | } |
6616 | ||
6617 | return ret; | |
6618 | } | |
6619 | ||
2ff7e61e JM |
6620 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6621 | u64 start, u64 end, | |
678886bd | 6622 | const bool return_free_space) |
ccd467d6 | 6623 | { |
11833d66 | 6624 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6625 | struct btrfs_space_info *space_info; |
6626 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6627 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6628 | u64 len; |
c759c4e1 JB |
6629 | u64 total_unpinned = 0; |
6630 | u64 empty_cluster = 0; | |
7b398f8e | 6631 | bool readonly; |
ccd467d6 | 6632 | |
11833d66 | 6633 | while (start <= end) { |
7b398f8e | 6634 | readonly = false; |
11833d66 YZ |
6635 | if (!cache || |
6636 | start >= cache->key.objectid + cache->key.offset) { | |
6637 | if (cache) | |
6638 | btrfs_put_block_group(cache); | |
c759c4e1 | 6639 | total_unpinned = 0; |
11833d66 | 6640 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6641 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6642 | |
2ff7e61e | 6643 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6644 | cache->space_info, |
6645 | &empty_cluster); | |
6646 | empty_cluster <<= 1; | |
11833d66 YZ |
6647 | } |
6648 | ||
6649 | len = cache->key.objectid + cache->key.offset - start; | |
6650 | len = min(len, end + 1 - start); | |
6651 | ||
6652 | if (start < cache->last_byte_to_unpin) { | |
6653 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6654 | if (return_free_space) |
6655 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6656 | } |
6657 | ||
f0486c68 | 6658 | start += len; |
c759c4e1 | 6659 | total_unpinned += len; |
7b398f8e | 6660 | space_info = cache->space_info; |
f0486c68 | 6661 | |
c759c4e1 JB |
6662 | /* |
6663 | * If this space cluster has been marked as fragmented and we've | |
6664 | * unpinned enough in this block group to potentially allow a | |
6665 | * cluster to be created inside of it go ahead and clear the | |
6666 | * fragmented check. | |
6667 | */ | |
6668 | if (cluster && cluster->fragmented && | |
6669 | total_unpinned > empty_cluster) { | |
6670 | spin_lock(&cluster->lock); | |
6671 | cluster->fragmented = 0; | |
6672 | spin_unlock(&cluster->lock); | |
6673 | } | |
6674 | ||
7b398f8e | 6675 | spin_lock(&space_info->lock); |
11833d66 YZ |
6676 | spin_lock(&cache->lock); |
6677 | cache->pinned -= len; | |
7b398f8e | 6678 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6679 | |
6680 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6681 | space_info->flags, len, 0); | |
4f4db217 | 6682 | space_info->max_extent_size = 0; |
d288db5d | 6683 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6684 | if (cache->ro) { |
6685 | space_info->bytes_readonly += len; | |
6686 | readonly = true; | |
6687 | } | |
11833d66 | 6688 | spin_unlock(&cache->lock); |
957780eb JB |
6689 | if (!readonly && return_free_space && |
6690 | global_rsv->space_info == space_info) { | |
6691 | u64 to_add = len; | |
6692 | WARN_ON(!return_free_space); | |
7b398f8e JB |
6693 | spin_lock(&global_rsv->lock); |
6694 | if (!global_rsv->full) { | |
957780eb JB |
6695 | to_add = min(len, global_rsv->size - |
6696 | global_rsv->reserved); | |
6697 | global_rsv->reserved += to_add; | |
6698 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6699 | if (global_rsv->reserved >= global_rsv->size) |
6700 | global_rsv->full = 1; | |
957780eb JB |
6701 | trace_btrfs_space_reservation(fs_info, |
6702 | "space_info", | |
6703 | space_info->flags, | |
6704 | to_add, 1); | |
6705 | len -= to_add; | |
7b398f8e JB |
6706 | } |
6707 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6708 | /* Add to any tickets we may have */ |
6709 | if (len) | |
6710 | space_info_add_new_bytes(fs_info, space_info, | |
6711 | len); | |
7b398f8e JB |
6712 | } |
6713 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6714 | } |
11833d66 YZ |
6715 | |
6716 | if (cache) | |
6717 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6718 | return 0; |
6719 | } | |
6720 | ||
6721 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
2ff7e61e | 6722 | struct btrfs_fs_info *fs_info) |
a28ec197 | 6723 | { |
e33e17ee JM |
6724 | struct btrfs_block_group_cache *block_group, *tmp; |
6725 | struct list_head *deleted_bgs; | |
11833d66 | 6726 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6727 | u64 start; |
6728 | u64 end; | |
a28ec197 | 6729 | int ret; |
a28ec197 | 6730 | |
11833d66 YZ |
6731 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6732 | unpin = &fs_info->freed_extents[1]; | |
6733 | else | |
6734 | unpin = &fs_info->freed_extents[0]; | |
6735 | ||
e33e17ee | 6736 | while (!trans->aborted) { |
d4b450cd | 6737 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6738 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6739 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6740 | if (ret) { |
6741 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6742 | break; |
d4b450cd | 6743 | } |
1f3c79a2 | 6744 | |
0b246afa | 6745 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6746 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6747 | end + 1 - start, NULL); |
1f3c79a2 | 6748 | |
af6f8f60 | 6749 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6750 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6751 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6752 | cond_resched(); |
a28ec197 | 6753 | } |
817d52f8 | 6754 | |
e33e17ee JM |
6755 | /* |
6756 | * Transaction is finished. We don't need the lock anymore. We | |
6757 | * do need to clean up the block groups in case of a transaction | |
6758 | * abort. | |
6759 | */ | |
6760 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6761 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6762 | u64 trimmed = 0; | |
6763 | ||
6764 | ret = -EROFS; | |
6765 | if (!trans->aborted) | |
2ff7e61e | 6766 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6767 | block_group->key.objectid, |
6768 | block_group->key.offset, | |
6769 | &trimmed); | |
6770 | ||
6771 | list_del_init(&block_group->bg_list); | |
6772 | btrfs_put_block_group_trimming(block_group); | |
6773 | btrfs_put_block_group(block_group); | |
6774 | ||
6775 | if (ret) { | |
6776 | const char *errstr = btrfs_decode_error(ret); | |
6777 | btrfs_warn(fs_info, | |
6778 | "Discard failed while removing blockgroup: errno=%d %s\n", | |
6779 | ret, errstr); | |
6780 | } | |
6781 | } | |
6782 | ||
e20d96d6 CM |
6783 | return 0; |
6784 | } | |
6785 | ||
b150a4f1 JB |
6786 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, |
6787 | u64 owner, u64 root_objectid) | |
6788 | { | |
6789 | struct btrfs_space_info *space_info; | |
6790 | u64 flags; | |
6791 | ||
6792 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
6793 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
6794 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
6795 | else | |
6796 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
6797 | } else { | |
6798 | flags = BTRFS_BLOCK_GROUP_DATA; | |
6799 | } | |
6800 | ||
6801 | space_info = __find_space_info(fs_info, flags); | |
6802 | BUG_ON(!space_info); /* Logic bug */ | |
6803 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); | |
6804 | } | |
6805 | ||
6806 | ||
5d4f98a2 | 6807 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 6808 | struct btrfs_fs_info *info, |
c682f9b3 | 6809 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6810 | u64 root_objectid, u64 owner_objectid, |
6811 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6812 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6813 | { |
e2fa7227 | 6814 | struct btrfs_key key; |
5d4f98a2 | 6815 | struct btrfs_path *path; |
1261ec42 | 6816 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6817 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6818 | struct btrfs_extent_item *ei; |
6819 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6820 | int ret; |
5d4f98a2 | 6821 | int is_data; |
952fccac CM |
6822 | int extent_slot = 0; |
6823 | int found_extent = 0; | |
6824 | int num_to_del = 1; | |
5d4f98a2 YZ |
6825 | u32 item_size; |
6826 | u64 refs; | |
c682f9b3 QW |
6827 | u64 bytenr = node->bytenr; |
6828 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6829 | int last_ref = 0; |
0b246afa | 6830 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6831 | |
5caf2a00 | 6832 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6833 | if (!path) |
6834 | return -ENOMEM; | |
5f26f772 | 6835 | |
e4058b54 | 6836 | path->reada = READA_FORWARD; |
b9473439 | 6837 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6838 | |
6839 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6840 | BUG_ON(!is_data && refs_to_drop != 1); | |
6841 | ||
3173a18f JB |
6842 | if (is_data) |
6843 | skinny_metadata = 0; | |
6844 | ||
87bde3cd | 6845 | ret = lookup_extent_backref(trans, info, path, &iref, |
5d4f98a2 YZ |
6846 | bytenr, num_bytes, parent, |
6847 | root_objectid, owner_objectid, | |
6848 | owner_offset); | |
7bb86316 | 6849 | if (ret == 0) { |
952fccac | 6850 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6851 | while (extent_slot >= 0) { |
6852 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6853 | extent_slot); |
5d4f98a2 | 6854 | if (key.objectid != bytenr) |
952fccac | 6855 | break; |
5d4f98a2 YZ |
6856 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6857 | key.offset == num_bytes) { | |
952fccac CM |
6858 | found_extent = 1; |
6859 | break; | |
6860 | } | |
3173a18f JB |
6861 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6862 | key.offset == owner_objectid) { | |
6863 | found_extent = 1; | |
6864 | break; | |
6865 | } | |
952fccac CM |
6866 | if (path->slots[0] - extent_slot > 5) |
6867 | break; | |
5d4f98a2 | 6868 | extent_slot--; |
952fccac | 6869 | } |
5d4f98a2 YZ |
6870 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6871 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6872 | if (found_extent && item_size < sizeof(*ei)) | |
6873 | found_extent = 0; | |
6874 | #endif | |
31840ae1 | 6875 | if (!found_extent) { |
5d4f98a2 | 6876 | BUG_ON(iref); |
87bde3cd JM |
6877 | ret = remove_extent_backref(trans, info, path, NULL, |
6878 | refs_to_drop, | |
fcebe456 | 6879 | is_data, &last_ref); |
005d6427 | 6880 | if (ret) { |
66642832 | 6881 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6882 | goto out; |
6883 | } | |
b3b4aa74 | 6884 | btrfs_release_path(path); |
b9473439 | 6885 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6886 | |
6887 | key.objectid = bytenr; | |
6888 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6889 | key.offset = num_bytes; | |
6890 | ||
3173a18f JB |
6891 | if (!is_data && skinny_metadata) { |
6892 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6893 | key.offset = owner_objectid; | |
6894 | } | |
6895 | ||
31840ae1 ZY |
6896 | ret = btrfs_search_slot(trans, extent_root, |
6897 | &key, path, -1, 1); | |
3173a18f JB |
6898 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6899 | /* | |
6900 | * Couldn't find our skinny metadata item, | |
6901 | * see if we have ye olde extent item. | |
6902 | */ | |
6903 | path->slots[0]--; | |
6904 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6905 | path->slots[0]); | |
6906 | if (key.objectid == bytenr && | |
6907 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6908 | key.offset == num_bytes) | |
6909 | ret = 0; | |
6910 | } | |
6911 | ||
6912 | if (ret > 0 && skinny_metadata) { | |
6913 | skinny_metadata = false; | |
9ce49a0b | 6914 | key.objectid = bytenr; |
3173a18f JB |
6915 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6916 | key.offset = num_bytes; | |
6917 | btrfs_release_path(path); | |
6918 | ret = btrfs_search_slot(trans, extent_root, | |
6919 | &key, path, -1, 1); | |
6920 | } | |
6921 | ||
f3465ca4 | 6922 | if (ret) { |
5d163e0e JM |
6923 | btrfs_err(info, |
6924 | "umm, got %d back from search, was looking for %llu", | |
6925 | ret, bytenr); | |
b783e62d | 6926 | if (ret > 0) |
2ff7e61e | 6927 | btrfs_print_leaf(info, path->nodes[0]); |
f3465ca4 | 6928 | } |
005d6427 | 6929 | if (ret < 0) { |
66642832 | 6930 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6931 | goto out; |
6932 | } | |
31840ae1 ZY |
6933 | extent_slot = path->slots[0]; |
6934 | } | |
fae7f21c | 6935 | } else if (WARN_ON(ret == -ENOENT)) { |
2ff7e61e | 6936 | btrfs_print_leaf(info, path->nodes[0]); |
c2cf52eb SK |
6937 | btrfs_err(info, |
6938 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6939 | bytenr, parent, root_objectid, owner_objectid, |
6940 | owner_offset); | |
66642832 | 6941 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6942 | goto out; |
79787eaa | 6943 | } else { |
66642832 | 6944 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6945 | goto out; |
7bb86316 | 6946 | } |
5f39d397 CM |
6947 | |
6948 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6949 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6950 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6951 | if (item_size < sizeof(*ei)) { | |
6952 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
87bde3cd JM |
6953 | ret = convert_extent_item_v0(trans, info, path, owner_objectid, |
6954 | 0); | |
005d6427 | 6955 | if (ret < 0) { |
66642832 | 6956 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6957 | goto out; |
6958 | } | |
5d4f98a2 | 6959 | |
b3b4aa74 | 6960 | btrfs_release_path(path); |
5d4f98a2 YZ |
6961 | path->leave_spinning = 1; |
6962 | ||
6963 | key.objectid = bytenr; | |
6964 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6965 | key.offset = num_bytes; | |
6966 | ||
6967 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
6968 | -1, 1); | |
6969 | if (ret) { | |
5d163e0e JM |
6970 | btrfs_err(info, |
6971 | "umm, got %d back from search, was looking for %llu", | |
c1c9ff7c | 6972 | ret, bytenr); |
2ff7e61e | 6973 | btrfs_print_leaf(info, path->nodes[0]); |
5d4f98a2 | 6974 | } |
005d6427 | 6975 | if (ret < 0) { |
66642832 | 6976 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6977 | goto out; |
6978 | } | |
6979 | ||
5d4f98a2 YZ |
6980 | extent_slot = path->slots[0]; |
6981 | leaf = path->nodes[0]; | |
6982 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
6983 | } | |
6984 | #endif | |
6985 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 6986 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 6987 | struct btrfs_extent_item); |
3173a18f JB |
6988 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
6989 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
6990 | struct btrfs_tree_block_info *bi; |
6991 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
6992 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
6993 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
6994 | } | |
56bec294 | 6995 | |
5d4f98a2 | 6996 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 6997 | if (refs < refs_to_drop) { |
5d163e0e JM |
6998 | btrfs_err(info, |
6999 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7000 | refs_to_drop, refs, bytenr); | |
32b02538 | 7001 | ret = -EINVAL; |
66642832 | 7002 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7003 | goto out; |
7004 | } | |
56bec294 | 7005 | refs -= refs_to_drop; |
5f39d397 | 7006 | |
5d4f98a2 YZ |
7007 | if (refs > 0) { |
7008 | if (extent_op) | |
7009 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7010 | /* | |
7011 | * In the case of inline back ref, reference count will | |
7012 | * be updated by remove_extent_backref | |
952fccac | 7013 | */ |
5d4f98a2 YZ |
7014 | if (iref) { |
7015 | BUG_ON(!found_extent); | |
7016 | } else { | |
7017 | btrfs_set_extent_refs(leaf, ei, refs); | |
7018 | btrfs_mark_buffer_dirty(leaf); | |
7019 | } | |
7020 | if (found_extent) { | |
87bde3cd | 7021 | ret = remove_extent_backref(trans, info, path, |
5d4f98a2 | 7022 | iref, refs_to_drop, |
fcebe456 | 7023 | is_data, &last_ref); |
005d6427 | 7024 | if (ret) { |
66642832 | 7025 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7026 | goto out; |
7027 | } | |
952fccac | 7028 | } |
0b246afa | 7029 | add_pinned_bytes(info, -num_bytes, owner_objectid, |
b150a4f1 | 7030 | root_objectid); |
5d4f98a2 | 7031 | } else { |
5d4f98a2 YZ |
7032 | if (found_extent) { |
7033 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7034 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7035 | if (iref) { |
7036 | BUG_ON(path->slots[0] != extent_slot); | |
7037 | } else { | |
7038 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7039 | path->slots[0] = extent_slot; | |
7040 | num_to_del = 2; | |
7041 | } | |
78fae27e | 7042 | } |
b9473439 | 7043 | |
fcebe456 | 7044 | last_ref = 1; |
952fccac CM |
7045 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7046 | num_to_del); | |
005d6427 | 7047 | if (ret) { |
66642832 | 7048 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7049 | goto out; |
7050 | } | |
b3b4aa74 | 7051 | btrfs_release_path(path); |
21af804c | 7052 | |
5d4f98a2 | 7053 | if (is_data) { |
5b4aacef | 7054 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7055 | if (ret) { |
66642832 | 7056 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7057 | goto out; |
7058 | } | |
459931ec CM |
7059 | } |
7060 | ||
0b246afa | 7061 | ret = add_to_free_space_tree(trans, info, bytenr, num_bytes); |
1e144fb8 | 7062 | if (ret) { |
66642832 | 7063 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7064 | goto out; |
7065 | } | |
7066 | ||
0b246afa | 7067 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 7068 | if (ret) { |
66642832 | 7069 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7070 | goto out; |
7071 | } | |
a28ec197 | 7072 | } |
fcebe456 JB |
7073 | btrfs_release_path(path); |
7074 | ||
79787eaa | 7075 | out: |
5caf2a00 | 7076 | btrfs_free_path(path); |
a28ec197 CM |
7077 | return ret; |
7078 | } | |
7079 | ||
1887be66 | 7080 | /* |
f0486c68 | 7081 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7082 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7083 | * a given extent, and if there are no other delayed refs to be processed, it | |
7084 | * removes it from the tree. | |
7085 | */ | |
7086 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7087 | u64 bytenr) |
1887be66 CM |
7088 | { |
7089 | struct btrfs_delayed_ref_head *head; | |
7090 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7091 | int ret = 0; |
1887be66 CM |
7092 | |
7093 | delayed_refs = &trans->transaction->delayed_refs; | |
7094 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7095 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7096 | if (!head) |
cf93da7b | 7097 | goto out_delayed_unlock; |
1887be66 | 7098 | |
d7df2c79 | 7099 | spin_lock(&head->lock); |
c6fc2454 | 7100 | if (!list_empty(&head->ref_list)) |
1887be66 CM |
7101 | goto out; |
7102 | ||
5d4f98a2 YZ |
7103 | if (head->extent_op) { |
7104 | if (!head->must_insert_reserved) | |
7105 | goto out; | |
78a6184a | 7106 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7107 | head->extent_op = NULL; |
7108 | } | |
7109 | ||
1887be66 CM |
7110 | /* |
7111 | * waiting for the lock here would deadlock. If someone else has it | |
7112 | * locked they are already in the process of dropping it anyway | |
7113 | */ | |
7114 | if (!mutex_trylock(&head->mutex)) | |
7115 | goto out; | |
7116 | ||
7117 | /* | |
7118 | * at this point we have a head with no other entries. Go | |
7119 | * ahead and process it. | |
7120 | */ | |
7121 | head->node.in_tree = 0; | |
c46effa6 | 7122 | rb_erase(&head->href_node, &delayed_refs->href_root); |
c3e69d58 | 7123 | |
d7df2c79 | 7124 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7125 | |
7126 | /* | |
7127 | * we don't take a ref on the node because we're removing it from the | |
7128 | * tree, so we just steal the ref the tree was holding. | |
7129 | */ | |
c3e69d58 | 7130 | delayed_refs->num_heads--; |
d7df2c79 | 7131 | if (head->processing == 0) |
c3e69d58 | 7132 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7133 | head->processing = 0; |
7134 | spin_unlock(&head->lock); | |
1887be66 CM |
7135 | spin_unlock(&delayed_refs->lock); |
7136 | ||
f0486c68 YZ |
7137 | BUG_ON(head->extent_op); |
7138 | if (head->must_insert_reserved) | |
7139 | ret = 1; | |
7140 | ||
7141 | mutex_unlock(&head->mutex); | |
1887be66 | 7142 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 7143 | return ret; |
1887be66 | 7144 | out: |
d7df2c79 | 7145 | spin_unlock(&head->lock); |
cf93da7b CM |
7146 | |
7147 | out_delayed_unlock: | |
1887be66 CM |
7148 | spin_unlock(&delayed_refs->lock); |
7149 | return 0; | |
7150 | } | |
7151 | ||
f0486c68 YZ |
7152 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7153 | struct btrfs_root *root, | |
7154 | struct extent_buffer *buf, | |
5581a51a | 7155 | u64 parent, int last_ref) |
f0486c68 | 7156 | { |
0b246afa | 7157 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7158 | int pin = 1; |
f0486c68 YZ |
7159 | int ret; |
7160 | ||
7161 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
0b246afa JM |
7162 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, |
7163 | buf->start, buf->len, | |
7164 | parent, | |
7165 | root->root_key.objectid, | |
7166 | btrfs_header_level(buf), | |
7167 | BTRFS_DROP_DELAYED_REF, NULL); | |
79787eaa | 7168 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
7169 | } |
7170 | ||
7171 | if (!last_ref) | |
7172 | return; | |
7173 | ||
f0486c68 | 7174 | if (btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7175 | struct btrfs_block_group_cache *cache; |
7176 | ||
f0486c68 | 7177 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7178 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7179 | if (!ret) |
37be25bc | 7180 | goto out; |
f0486c68 YZ |
7181 | } |
7182 | ||
0b246afa | 7183 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7184 | |
f0486c68 | 7185 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7186 | pin_down_extent(fs_info, cache, buf->start, |
7187 | buf->len, 1); | |
6219872d | 7188 | btrfs_put_block_group(cache); |
37be25bc | 7189 | goto out; |
f0486c68 YZ |
7190 | } |
7191 | ||
7192 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7193 | ||
7194 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7195 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7196 | btrfs_put_block_group(cache); |
71ff6437 | 7197 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
b150a4f1 | 7198 | pin = 0; |
f0486c68 YZ |
7199 | } |
7200 | out: | |
b150a4f1 | 7201 | if (pin) |
0b246afa | 7202 | add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf), |
b150a4f1 JB |
7203 | root->root_key.objectid); |
7204 | ||
a826d6dc JB |
7205 | /* |
7206 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
7207 | * anymore. | |
7208 | */ | |
7209 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
7210 | } |
7211 | ||
79787eaa | 7212 | /* Can return -ENOMEM */ |
2ff7e61e JM |
7213 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
7214 | struct btrfs_fs_info *fs_info, | |
66d7e7f0 | 7215 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7216 | u64 owner, u64 offset) |
925baedd CM |
7217 | { |
7218 | int ret; | |
7219 | ||
f5ee5c9a | 7220 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7221 | return 0; |
fccb84c9 | 7222 | |
0b246afa | 7223 | add_pinned_bytes(fs_info, num_bytes, owner, root_objectid); |
b150a4f1 | 7224 | |
56bec294 CM |
7225 | /* |
7226 | * tree log blocks never actually go into the extent allocation | |
7227 | * tree, just update pinning info and exit early. | |
56bec294 | 7228 | */ |
5d4f98a2 YZ |
7229 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7230 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7231 | /* unlocks the pinned mutex */ |
2ff7e61e | 7232 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
56bec294 | 7233 | ret = 0; |
5d4f98a2 | 7234 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
7235 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7236 | num_bytes, | |
5d4f98a2 | 7237 | parent, root_objectid, (int)owner, |
b06c4bf5 | 7238 | BTRFS_DROP_DELAYED_REF, NULL); |
5d4f98a2 | 7239 | } else { |
66d7e7f0 AJ |
7240 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7241 | num_bytes, | |
7242 | parent, root_objectid, owner, | |
5846a3c2 | 7243 | offset, 0, |
fef394f7 | 7244 | BTRFS_DROP_DELAYED_REF); |
56bec294 | 7245 | } |
925baedd CM |
7246 | return ret; |
7247 | } | |
7248 | ||
817d52f8 JB |
7249 | /* |
7250 | * when we wait for progress in the block group caching, its because | |
7251 | * our allocation attempt failed at least once. So, we must sleep | |
7252 | * and let some progress happen before we try again. | |
7253 | * | |
7254 | * This function will sleep at least once waiting for new free space to | |
7255 | * show up, and then it will check the block group free space numbers | |
7256 | * for our min num_bytes. Another option is to have it go ahead | |
7257 | * and look in the rbtree for a free extent of a given size, but this | |
7258 | * is a good start. | |
36cce922 JB |
7259 | * |
7260 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7261 | * any of the information in this block group. | |
817d52f8 | 7262 | */ |
36cce922 | 7263 | static noinline void |
817d52f8 JB |
7264 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7265 | u64 num_bytes) | |
7266 | { | |
11833d66 | 7267 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7268 | |
11833d66 YZ |
7269 | caching_ctl = get_caching_control(cache); |
7270 | if (!caching_ctl) | |
36cce922 | 7271 | return; |
817d52f8 | 7272 | |
11833d66 | 7273 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7274 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7275 | |
7276 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7277 | } |
7278 | ||
7279 | static noinline int | |
7280 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7281 | { | |
7282 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7283 | int ret = 0; |
11833d66 YZ |
7284 | |
7285 | caching_ctl = get_caching_control(cache); | |
7286 | if (!caching_ctl) | |
36cce922 | 7287 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7288 | |
7289 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7290 | if (cache->cached == BTRFS_CACHE_ERROR) |
7291 | ret = -EIO; | |
11833d66 | 7292 | put_caching_control(caching_ctl); |
36cce922 | 7293 | return ret; |
817d52f8 JB |
7294 | } |
7295 | ||
31e50229 | 7296 | int __get_raid_index(u64 flags) |
b742bb82 | 7297 | { |
7738a53a | 7298 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 7299 | return BTRFS_RAID_RAID10; |
7738a53a | 7300 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 7301 | return BTRFS_RAID_RAID1; |
7738a53a | 7302 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 7303 | return BTRFS_RAID_DUP; |
7738a53a | 7304 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 7305 | return BTRFS_RAID_RAID0; |
53b381b3 | 7306 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 7307 | return BTRFS_RAID_RAID5; |
53b381b3 | 7308 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 7309 | return BTRFS_RAID_RAID6; |
7738a53a | 7310 | |
e942f883 | 7311 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
7312 | } |
7313 | ||
6ab0a202 | 7314 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
7738a53a | 7315 | { |
31e50229 | 7316 | return __get_raid_index(cache->flags); |
7738a53a ID |
7317 | } |
7318 | ||
6ab0a202 JM |
7319 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
7320 | [BTRFS_RAID_RAID10] = "raid10", | |
7321 | [BTRFS_RAID_RAID1] = "raid1", | |
7322 | [BTRFS_RAID_DUP] = "dup", | |
7323 | [BTRFS_RAID_RAID0] = "raid0", | |
7324 | [BTRFS_RAID_SINGLE] = "single", | |
7325 | [BTRFS_RAID_RAID5] = "raid5", | |
7326 | [BTRFS_RAID_RAID6] = "raid6", | |
7327 | }; | |
7328 | ||
1b8e5df6 | 7329 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
7330 | { |
7331 | if (type >= BTRFS_NR_RAID_TYPES) | |
7332 | return NULL; | |
7333 | ||
7334 | return btrfs_raid_type_names[type]; | |
7335 | } | |
7336 | ||
817d52f8 | 7337 | enum btrfs_loop_type { |
285ff5af JB |
7338 | LOOP_CACHING_NOWAIT = 0, |
7339 | LOOP_CACHING_WAIT = 1, | |
7340 | LOOP_ALLOC_CHUNK = 2, | |
7341 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7342 | }; |
7343 | ||
e570fd27 MX |
7344 | static inline void |
7345 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7346 | int delalloc) | |
7347 | { | |
7348 | if (delalloc) | |
7349 | down_read(&cache->data_rwsem); | |
7350 | } | |
7351 | ||
7352 | static inline void | |
7353 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7354 | int delalloc) | |
7355 | { | |
7356 | btrfs_get_block_group(cache); | |
7357 | if (delalloc) | |
7358 | down_read(&cache->data_rwsem); | |
7359 | } | |
7360 | ||
7361 | static struct btrfs_block_group_cache * | |
7362 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7363 | struct btrfs_free_cluster *cluster, | |
7364 | int delalloc) | |
7365 | { | |
89771cc9 | 7366 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7367 | |
e570fd27 | 7368 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7369 | while (1) { |
7370 | used_bg = cluster->block_group; | |
7371 | if (!used_bg) | |
7372 | return NULL; | |
7373 | ||
7374 | if (used_bg == block_group) | |
e570fd27 MX |
7375 | return used_bg; |
7376 | ||
6719afdc | 7377 | btrfs_get_block_group(used_bg); |
e570fd27 | 7378 | |
6719afdc GU |
7379 | if (!delalloc) |
7380 | return used_bg; | |
e570fd27 | 7381 | |
6719afdc GU |
7382 | if (down_read_trylock(&used_bg->data_rwsem)) |
7383 | return used_bg; | |
e570fd27 | 7384 | |
6719afdc | 7385 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7386 | |
e321f8a8 LB |
7387 | /* We should only have one-level nested. */ |
7388 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7389 | |
6719afdc GU |
7390 | spin_lock(&cluster->refill_lock); |
7391 | if (used_bg == cluster->block_group) | |
7392 | return used_bg; | |
e570fd27 | 7393 | |
6719afdc GU |
7394 | up_read(&used_bg->data_rwsem); |
7395 | btrfs_put_block_group(used_bg); | |
7396 | } | |
e570fd27 MX |
7397 | } |
7398 | ||
7399 | static inline void | |
7400 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7401 | int delalloc) | |
7402 | { | |
7403 | if (delalloc) | |
7404 | up_read(&cache->data_rwsem); | |
7405 | btrfs_put_block_group(cache); | |
7406 | } | |
7407 | ||
fec577fb CM |
7408 | /* |
7409 | * walks the btree of allocated extents and find a hole of a given size. | |
7410 | * The key ins is changed to record the hole: | |
a4820398 | 7411 | * ins->objectid == start position |
62e2749e | 7412 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7413 | * ins->offset == the size of the hole. |
fec577fb | 7414 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7415 | * |
7416 | * If there is no suitable free space, we will record the max size of | |
7417 | * the free space extent currently. | |
fec577fb | 7418 | */ |
87bde3cd | 7419 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7420 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7421 | u64 hint_byte, struct btrfs_key *ins, | |
7422 | u64 flags, int delalloc) | |
fec577fb | 7423 | { |
80eb234a | 7424 | int ret = 0; |
0b246afa | 7425 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7426 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7427 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7428 | u64 search_start = 0; |
a4820398 | 7429 | u64 max_extent_size = 0; |
c759c4e1 | 7430 | u64 empty_cluster = 0; |
80eb234a | 7431 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7432 | int loop = 0; |
b6919a58 | 7433 | int index = __get_raid_index(flags); |
0a24325e | 7434 | bool failed_cluster_refill = false; |
1cdda9b8 | 7435 | bool failed_alloc = false; |
67377734 | 7436 | bool use_cluster = true; |
60d2adbb | 7437 | bool have_caching_bg = false; |
13a0db5a | 7438 | bool orig_have_caching_bg = false; |
a5e681d9 | 7439 | bool full_search = false; |
fec577fb | 7440 | |
0b246afa | 7441 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7442 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7443 | ins->objectid = 0; |
7444 | ins->offset = 0; | |
b1a4d965 | 7445 | |
71ff6437 | 7446 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7447 | |
0b246afa | 7448 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7449 | if (!space_info) { |
0b246afa | 7450 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7451 | return -ENOSPC; |
7452 | } | |
2552d17e | 7453 | |
67377734 | 7454 | /* |
4f4db217 JB |
7455 | * If our free space is heavily fragmented we may not be able to make |
7456 | * big contiguous allocations, so instead of doing the expensive search | |
7457 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7458 | * can go ahead and search for a more manageable chunk. | |
7459 | * | |
7460 | * If our max_extent_size is large enough for our allocation simply | |
7461 | * disable clustering since we will likely not be able to find enough | |
7462 | * space to create a cluster and induce latency trying. | |
67377734 | 7463 | */ |
4f4db217 JB |
7464 | if (unlikely(space_info->max_extent_size)) { |
7465 | spin_lock(&space_info->lock); | |
7466 | if (space_info->max_extent_size && | |
7467 | num_bytes > space_info->max_extent_size) { | |
7468 | ins->offset = space_info->max_extent_size; | |
7469 | spin_unlock(&space_info->lock); | |
7470 | return -ENOSPC; | |
7471 | } else if (space_info->max_extent_size) { | |
7472 | use_cluster = false; | |
7473 | } | |
7474 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7475 | } |
0f9dd46c | 7476 | |
2ff7e61e | 7477 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7478 | if (last_ptr) { |
fa9c0d79 CM |
7479 | spin_lock(&last_ptr->lock); |
7480 | if (last_ptr->block_group) | |
7481 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7482 | if (last_ptr->fragmented) { |
7483 | /* | |
7484 | * We still set window_start so we can keep track of the | |
7485 | * last place we found an allocation to try and save | |
7486 | * some time. | |
7487 | */ | |
7488 | hint_byte = last_ptr->window_start; | |
7489 | use_cluster = false; | |
7490 | } | |
fa9c0d79 | 7491 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7492 | } |
fa9c0d79 | 7493 | |
2ff7e61e | 7494 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7495 | search_start = max(search_start, hint_byte); |
2552d17e | 7496 | if (search_start == hint_byte) { |
0b246afa | 7497 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7498 | /* |
7499 | * we don't want to use the block group if it doesn't match our | |
7500 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7501 | * |
7502 | * However if we are re-searching with an ideal block group | |
7503 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7504 | */ |
b6919a58 | 7505 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7506 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7507 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7508 | if (list_empty(&block_group->list) || |
7509 | block_group->ro) { | |
7510 | /* | |
7511 | * someone is removing this block group, | |
7512 | * we can't jump into the have_block_group | |
7513 | * target because our list pointers are not | |
7514 | * valid | |
7515 | */ | |
7516 | btrfs_put_block_group(block_group); | |
7517 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7518 | } else { |
b742bb82 | 7519 | index = get_block_group_index(block_group); |
e570fd27 | 7520 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7521 | goto have_block_group; |
ccf0e725 | 7522 | } |
2552d17e | 7523 | } else if (block_group) { |
fa9c0d79 | 7524 | btrfs_put_block_group(block_group); |
2552d17e | 7525 | } |
42e70e7a | 7526 | } |
2552d17e | 7527 | search: |
60d2adbb | 7528 | have_caching_bg = false; |
a5e681d9 JB |
7529 | if (index == 0 || index == __get_raid_index(flags)) |
7530 | full_search = true; | |
80eb234a | 7531 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7532 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7533 | list) { | |
6226cb0a | 7534 | u64 offset; |
817d52f8 | 7535 | int cached; |
8a1413a2 | 7536 | |
e570fd27 | 7537 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7538 | search_start = block_group->key.objectid; |
42e70e7a | 7539 | |
83a50de9 CM |
7540 | /* |
7541 | * this can happen if we end up cycling through all the | |
7542 | * raid types, but we want to make sure we only allocate | |
7543 | * for the proper type. | |
7544 | */ | |
b6919a58 | 7545 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7546 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7547 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7548 | BTRFS_BLOCK_GROUP_RAID5 | |
7549 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7550 | BTRFS_BLOCK_GROUP_RAID10; |
7551 | ||
7552 | /* | |
7553 | * if they asked for extra copies and this block group | |
7554 | * doesn't provide them, bail. This does allow us to | |
7555 | * fill raid0 from raid1. | |
7556 | */ | |
b6919a58 | 7557 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7558 | goto loop; |
7559 | } | |
7560 | ||
2552d17e | 7561 | have_block_group: |
291c7d2f JB |
7562 | cached = block_group_cache_done(block_group); |
7563 | if (unlikely(!cached)) { | |
a5e681d9 | 7564 | have_caching_bg = true; |
f6373bf3 | 7565 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7566 | BUG_ON(ret < 0); |
7567 | ret = 0; | |
817d52f8 JB |
7568 | } |
7569 | ||
36cce922 JB |
7570 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7571 | goto loop; | |
ea6a478e | 7572 | if (unlikely(block_group->ro)) |
2552d17e | 7573 | goto loop; |
0f9dd46c | 7574 | |
0a24325e | 7575 | /* |
062c05c4 AO |
7576 | * Ok we want to try and use the cluster allocator, so |
7577 | * lets look there | |
0a24325e | 7578 | */ |
c759c4e1 | 7579 | if (last_ptr && use_cluster) { |
215a63d1 | 7580 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7581 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7582 | /* |
7583 | * the refill lock keeps out other | |
7584 | * people trying to start a new cluster | |
7585 | */ | |
e570fd27 MX |
7586 | used_block_group = btrfs_lock_cluster(block_group, |
7587 | last_ptr, | |
7588 | delalloc); | |
7589 | if (!used_block_group) | |
44fb5511 | 7590 | goto refill_cluster; |
274bd4fb | 7591 | |
e570fd27 MX |
7592 | if (used_block_group != block_group && |
7593 | (used_block_group->ro || | |
7594 | !block_group_bits(used_block_group, flags))) | |
7595 | goto release_cluster; | |
44fb5511 | 7596 | |
274bd4fb | 7597 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7598 | last_ptr, |
7599 | num_bytes, | |
7600 | used_block_group->key.objectid, | |
7601 | &max_extent_size); | |
fa9c0d79 CM |
7602 | if (offset) { |
7603 | /* we have a block, we're done */ | |
7604 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7605 | trace_btrfs_reserve_extent_cluster(fs_info, |
89d4346a MX |
7606 | used_block_group, |
7607 | search_start, num_bytes); | |
215a63d1 | 7608 | if (used_block_group != block_group) { |
e570fd27 MX |
7609 | btrfs_release_block_group(block_group, |
7610 | delalloc); | |
215a63d1 MX |
7611 | block_group = used_block_group; |
7612 | } | |
fa9c0d79 CM |
7613 | goto checks; |
7614 | } | |
7615 | ||
274bd4fb | 7616 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7617 | release_cluster: |
062c05c4 AO |
7618 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7619 | * set up a new clusters, so lets just skip it | |
7620 | * and let the allocator find whatever block | |
7621 | * it can find. If we reach this point, we | |
7622 | * will have tried the cluster allocator | |
7623 | * plenty of times and not have found | |
7624 | * anything, so we are likely way too | |
7625 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7626 | * anything. |
7627 | * | |
7628 | * However, if the cluster is taken from the | |
7629 | * current block group, release the cluster | |
7630 | * first, so that we stand a better chance of | |
7631 | * succeeding in the unclustered | |
7632 | * allocation. */ | |
7633 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7634 | used_block_group != block_group) { |
062c05c4 | 7635 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7636 | btrfs_release_block_group(used_block_group, |
7637 | delalloc); | |
062c05c4 AO |
7638 | goto unclustered_alloc; |
7639 | } | |
7640 | ||
fa9c0d79 CM |
7641 | /* |
7642 | * this cluster didn't work out, free it and | |
7643 | * start over | |
7644 | */ | |
7645 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7646 | ||
e570fd27 MX |
7647 | if (used_block_group != block_group) |
7648 | btrfs_release_block_group(used_block_group, | |
7649 | delalloc); | |
7650 | refill_cluster: | |
a5f6f719 AO |
7651 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7652 | spin_unlock(&last_ptr->refill_lock); | |
7653 | goto unclustered_alloc; | |
7654 | } | |
7655 | ||
8de972b4 CM |
7656 | aligned_cluster = max_t(unsigned long, |
7657 | empty_cluster + empty_size, | |
7658 | block_group->full_stripe_len); | |
7659 | ||
fa9c0d79 | 7660 | /* allocate a cluster in this block group */ |
2ff7e61e | 7661 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7662 | last_ptr, search_start, |
7663 | num_bytes, | |
7664 | aligned_cluster); | |
fa9c0d79 CM |
7665 | if (ret == 0) { |
7666 | /* | |
7667 | * now pull our allocation out of this | |
7668 | * cluster | |
7669 | */ | |
7670 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7671 | last_ptr, |
7672 | num_bytes, | |
7673 | search_start, | |
7674 | &max_extent_size); | |
fa9c0d79 CM |
7675 | if (offset) { |
7676 | /* we found one, proceed */ | |
7677 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7678 | trace_btrfs_reserve_extent_cluster(fs_info, |
3f7de037 JB |
7679 | block_group, search_start, |
7680 | num_bytes); | |
fa9c0d79 CM |
7681 | goto checks; |
7682 | } | |
0a24325e JB |
7683 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7684 | && !failed_cluster_refill) { | |
817d52f8 JB |
7685 | spin_unlock(&last_ptr->refill_lock); |
7686 | ||
0a24325e | 7687 | failed_cluster_refill = true; |
817d52f8 JB |
7688 | wait_block_group_cache_progress(block_group, |
7689 | num_bytes + empty_cluster + empty_size); | |
7690 | goto have_block_group; | |
fa9c0d79 | 7691 | } |
817d52f8 | 7692 | |
fa9c0d79 CM |
7693 | /* |
7694 | * at this point we either didn't find a cluster | |
7695 | * or we weren't able to allocate a block from our | |
7696 | * cluster. Free the cluster we've been trying | |
7697 | * to use, and go to the next block group | |
7698 | */ | |
0a24325e | 7699 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7700 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7701 | goto loop; |
fa9c0d79 CM |
7702 | } |
7703 | ||
062c05c4 | 7704 | unclustered_alloc: |
c759c4e1 JB |
7705 | /* |
7706 | * We are doing an unclustered alloc, set the fragmented flag so | |
7707 | * we don't bother trying to setup a cluster again until we get | |
7708 | * more space. | |
7709 | */ | |
7710 | if (unlikely(last_ptr)) { | |
7711 | spin_lock(&last_ptr->lock); | |
7712 | last_ptr->fragmented = 1; | |
7713 | spin_unlock(&last_ptr->lock); | |
7714 | } | |
0c9b36e0 LB |
7715 | if (cached) { |
7716 | struct btrfs_free_space_ctl *ctl = | |
7717 | block_group->free_space_ctl; | |
7718 | ||
7719 | spin_lock(&ctl->tree_lock); | |
7720 | if (ctl->free_space < | |
7721 | num_bytes + empty_cluster + empty_size) { | |
7722 | if (ctl->free_space > max_extent_size) | |
7723 | max_extent_size = ctl->free_space; | |
7724 | spin_unlock(&ctl->tree_lock); | |
7725 | goto loop; | |
7726 | } | |
7727 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7728 | } |
a5f6f719 | 7729 | |
6226cb0a | 7730 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7731 | num_bytes, empty_size, |
7732 | &max_extent_size); | |
1cdda9b8 JB |
7733 | /* |
7734 | * If we didn't find a chunk, and we haven't failed on this | |
7735 | * block group before, and this block group is in the middle of | |
7736 | * caching and we are ok with waiting, then go ahead and wait | |
7737 | * for progress to be made, and set failed_alloc to true. | |
7738 | * | |
7739 | * If failed_alloc is true then we've already waited on this | |
7740 | * block group once and should move on to the next block group. | |
7741 | */ | |
7742 | if (!offset && !failed_alloc && !cached && | |
7743 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7744 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7745 | num_bytes + empty_size); |
7746 | failed_alloc = true; | |
817d52f8 | 7747 | goto have_block_group; |
1cdda9b8 JB |
7748 | } else if (!offset) { |
7749 | goto loop; | |
817d52f8 | 7750 | } |
fa9c0d79 | 7751 | checks: |
0b246afa | 7752 | search_start = ALIGN(offset, fs_info->stripesize); |
25179201 | 7753 | |
2552d17e JB |
7754 | /* move on to the next group */ |
7755 | if (search_start + num_bytes > | |
215a63d1 MX |
7756 | block_group->key.objectid + block_group->key.offset) { |
7757 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7758 | goto loop; |
6226cb0a | 7759 | } |
f5a31e16 | 7760 | |
f0486c68 | 7761 | if (offset < search_start) |
215a63d1 | 7762 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7763 | search_start - offset); |
7764 | BUG_ON(offset > search_start); | |
2552d17e | 7765 | |
18513091 WX |
7766 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7767 | num_bytes, delalloc); | |
f0486c68 | 7768 | if (ret == -EAGAIN) { |
215a63d1 | 7769 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7770 | goto loop; |
0f9dd46c | 7771 | } |
9cfa3e34 | 7772 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7773 | |
f0486c68 | 7774 | /* we are all good, lets return */ |
2552d17e JB |
7775 | ins->objectid = search_start; |
7776 | ins->offset = num_bytes; | |
d2fb3437 | 7777 | |
71ff6437 | 7778 | trace_btrfs_reserve_extent(fs_info, block_group, |
3f7de037 | 7779 | search_start, num_bytes); |
e570fd27 | 7780 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7781 | break; |
7782 | loop: | |
0a24325e | 7783 | failed_cluster_refill = false; |
1cdda9b8 | 7784 | failed_alloc = false; |
b742bb82 | 7785 | BUG_ON(index != get_block_group_index(block_group)); |
e570fd27 | 7786 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7787 | } |
7788 | up_read(&space_info->groups_sem); | |
7789 | ||
13a0db5a | 7790 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7791 | && !orig_have_caching_bg) | |
7792 | orig_have_caching_bg = true; | |
7793 | ||
60d2adbb MX |
7794 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7795 | goto search; | |
7796 | ||
b742bb82 YZ |
7797 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7798 | goto search; | |
7799 | ||
285ff5af | 7800 | /* |
ccf0e725 JB |
7801 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7802 | * caching kthreads as we move along | |
817d52f8 JB |
7803 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7804 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7805 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7806 | * again | |
fa9c0d79 | 7807 | */ |
723bda20 | 7808 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7809 | index = 0; |
a5e681d9 JB |
7810 | if (loop == LOOP_CACHING_NOWAIT) { |
7811 | /* | |
7812 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7813 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7814 | * full search through. |
7815 | */ | |
13a0db5a | 7816 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7817 | loop = LOOP_CACHING_WAIT; |
7818 | else | |
7819 | loop = LOOP_ALLOC_CHUNK; | |
7820 | } else { | |
7821 | loop++; | |
7822 | } | |
7823 | ||
817d52f8 | 7824 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7825 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7826 | int exist = 0; |
7827 | ||
7828 | trans = current->journal_info; | |
7829 | if (trans) | |
7830 | exist = 1; | |
7831 | else | |
7832 | trans = btrfs_join_transaction(root); | |
00361589 | 7833 | |
00361589 JB |
7834 | if (IS_ERR(trans)) { |
7835 | ret = PTR_ERR(trans); | |
7836 | goto out; | |
7837 | } | |
7838 | ||
2ff7e61e | 7839 | ret = do_chunk_alloc(trans, fs_info, flags, |
ea658bad | 7840 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7841 | |
7842 | /* | |
7843 | * If we can't allocate a new chunk we've already looped | |
7844 | * through at least once, move on to the NO_EMPTY_SIZE | |
7845 | * case. | |
7846 | */ | |
7847 | if (ret == -ENOSPC) | |
7848 | loop = LOOP_NO_EMPTY_SIZE; | |
7849 | ||
ea658bad JB |
7850 | /* |
7851 | * Do not bail out on ENOSPC since we | |
7852 | * can do more things. | |
7853 | */ | |
00361589 | 7854 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7855 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7856 | else |
7857 | ret = 0; | |
f017f15f | 7858 | if (!exist) |
3a45bb20 | 7859 | btrfs_end_transaction(trans); |
00361589 | 7860 | if (ret) |
ea658bad | 7861 | goto out; |
2552d17e JB |
7862 | } |
7863 | ||
723bda20 | 7864 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7865 | /* |
7866 | * Don't loop again if we already have no empty_size and | |
7867 | * no empty_cluster. | |
7868 | */ | |
7869 | if (empty_size == 0 && | |
7870 | empty_cluster == 0) { | |
7871 | ret = -ENOSPC; | |
7872 | goto out; | |
7873 | } | |
723bda20 JB |
7874 | empty_size = 0; |
7875 | empty_cluster = 0; | |
fa9c0d79 | 7876 | } |
723bda20 JB |
7877 | |
7878 | goto search; | |
2552d17e JB |
7879 | } else if (!ins->objectid) { |
7880 | ret = -ENOSPC; | |
d82a6f1d | 7881 | } else if (ins->objectid) { |
c759c4e1 JB |
7882 | if (!use_cluster && last_ptr) { |
7883 | spin_lock(&last_ptr->lock); | |
7884 | last_ptr->window_start = ins->objectid; | |
7885 | spin_unlock(&last_ptr->lock); | |
7886 | } | |
80eb234a | 7887 | ret = 0; |
be744175 | 7888 | } |
79787eaa | 7889 | out: |
4f4db217 JB |
7890 | if (ret == -ENOSPC) { |
7891 | spin_lock(&space_info->lock); | |
7892 | space_info->max_extent_size = max_extent_size; | |
7893 | spin_unlock(&space_info->lock); | |
a4820398 | 7894 | ins->offset = max_extent_size; |
4f4db217 | 7895 | } |
0f70abe2 | 7896 | return ret; |
fec577fb | 7897 | } |
ec44a35c | 7898 | |
ab8d0fc4 JM |
7899 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7900 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7901 | int dump_block_groups) |
0f9dd46c JB |
7902 | { |
7903 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7904 | int index = 0; |
0f9dd46c | 7905 | |
9ed74f2d | 7906 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7907 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7908 | info->flags, | |
4136135b LB |
7909 | info->total_bytes - btrfs_space_info_used(info, true), |
7910 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7911 | btrfs_info(fs_info, |
7912 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7913 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7914 | info->bytes_reserved, info->bytes_may_use, | |
7915 | info->bytes_readonly); | |
9ed74f2d JB |
7916 | spin_unlock(&info->lock); |
7917 | ||
7918 | if (!dump_block_groups) | |
7919 | return; | |
0f9dd46c | 7920 | |
80eb234a | 7921 | down_read(&info->groups_sem); |
b742bb82 YZ |
7922 | again: |
7923 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7924 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7925 | btrfs_info(fs_info, |
7926 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7927 | cache->key.objectid, cache->key.offset, | |
7928 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7929 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7930 | btrfs_dump_free_space(cache, bytes); |
7931 | spin_unlock(&cache->lock); | |
7932 | } | |
b742bb82 YZ |
7933 | if (++index < BTRFS_NR_RAID_TYPES) |
7934 | goto again; | |
80eb234a | 7935 | up_read(&info->groups_sem); |
0f9dd46c | 7936 | } |
e8569813 | 7937 | |
18513091 | 7938 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
7939 | u64 num_bytes, u64 min_alloc_size, |
7940 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7941 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7942 | { |
ab8d0fc4 | 7943 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 7944 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7945 | u64 flags; |
fec577fb | 7946 | int ret; |
925baedd | 7947 | |
b6919a58 | 7948 | flags = btrfs_get_alloc_profile(root, is_data); |
98d20f67 | 7949 | again: |
0b246afa | 7950 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 7951 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 7952 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 7953 | if (!ret && !is_data) { |
ab8d0fc4 | 7954 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 7955 | } else if (ret == -ENOSPC) { |
a4820398 MX |
7956 | if (!final_tried && ins->offset) { |
7957 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 7958 | num_bytes = round_down(num_bytes, |
0b246afa | 7959 | fs_info->sectorsize); |
9e622d6b | 7960 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 7961 | ram_bytes = num_bytes; |
9e622d6b MX |
7962 | if (num_bytes == min_alloc_size) |
7963 | final_tried = true; | |
7964 | goto again; | |
ab8d0fc4 | 7965 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
7966 | struct btrfs_space_info *sinfo; |
7967 | ||
ab8d0fc4 | 7968 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 7969 | btrfs_err(fs_info, |
5d163e0e JM |
7970 | "allocation failed flags %llu, wanted %llu", |
7971 | flags, num_bytes); | |
53804280 | 7972 | if (sinfo) |
ab8d0fc4 | 7973 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 7974 | } |
925baedd | 7975 | } |
0f9dd46c JB |
7976 | |
7977 | return ret; | |
e6dcd2dc CM |
7978 | } |
7979 | ||
2ff7e61e | 7980 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
7981 | u64 start, u64 len, |
7982 | int pin, int delalloc) | |
65b51a00 | 7983 | { |
0f9dd46c | 7984 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 7985 | int ret = 0; |
0f9dd46c | 7986 | |
0b246afa | 7987 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 7988 | if (!cache) { |
0b246afa JM |
7989 | btrfs_err(fs_info, "Unable to find block group for %llu", |
7990 | start); | |
0f9dd46c JB |
7991 | return -ENOSPC; |
7992 | } | |
1f3c79a2 | 7993 | |
e688b725 | 7994 | if (pin) |
2ff7e61e | 7995 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 7996 | else { |
0b246afa | 7997 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 7998 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 7999 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8000 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8001 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8002 | } |
31193213 | 8003 | |
fa9c0d79 | 8004 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8005 | return ret; |
8006 | } | |
8007 | ||
2ff7e61e | 8008 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8009 | u64 start, u64 len, int delalloc) |
e688b725 | 8010 | { |
2ff7e61e | 8011 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8012 | } |
8013 | ||
2ff7e61e | 8014 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8015 | u64 start, u64 len) |
8016 | { | |
2ff7e61e | 8017 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8018 | } |
8019 | ||
5d4f98a2 | 8020 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8021 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8022 | u64 parent, u64 root_objectid, |
8023 | u64 flags, u64 owner, u64 offset, | |
8024 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
8025 | { |
8026 | int ret; | |
e6dcd2dc | 8027 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8028 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8029 | struct btrfs_path *path; |
5d4f98a2 YZ |
8030 | struct extent_buffer *leaf; |
8031 | int type; | |
8032 | u32 size; | |
26b8003f | 8033 | |
5d4f98a2 YZ |
8034 | if (parent > 0) |
8035 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8036 | else | |
8037 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8038 | |
5d4f98a2 | 8039 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8040 | |
8041 | path = btrfs_alloc_path(); | |
db5b493a TI |
8042 | if (!path) |
8043 | return -ENOMEM; | |
47e4bb98 | 8044 | |
b9473439 | 8045 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8046 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8047 | ins, size); | |
79787eaa JM |
8048 | if (ret) { |
8049 | btrfs_free_path(path); | |
8050 | return ret; | |
8051 | } | |
0f9dd46c | 8052 | |
5d4f98a2 YZ |
8053 | leaf = path->nodes[0]; |
8054 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8055 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8056 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8057 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8058 | btrfs_set_extent_flags(leaf, extent_item, | |
8059 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8060 | ||
8061 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8062 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8063 | if (parent > 0) { | |
8064 | struct btrfs_shared_data_ref *ref; | |
8065 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8066 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8067 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8068 | } else { | |
8069 | struct btrfs_extent_data_ref *ref; | |
8070 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8071 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8072 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8073 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8074 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8075 | } | |
47e4bb98 CM |
8076 | |
8077 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8078 | btrfs_free_path(path); |
f510cfec | 8079 | |
1e144fb8 OS |
8080 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8081 | ins->offset); | |
8082 | if (ret) | |
8083 | return ret; | |
8084 | ||
6202df69 | 8085 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 8086 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8087 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8088 | ins->objectid, ins->offset); |
f5947066 CM |
8089 | BUG(); |
8090 | } | |
71ff6437 | 8091 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8092 | return ret; |
8093 | } | |
8094 | ||
5d4f98a2 | 8095 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
2ff7e61e | 8096 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8097 | u64 parent, u64 root_objectid, |
8098 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 8099 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
8100 | { |
8101 | int ret; | |
5d4f98a2 YZ |
8102 | struct btrfs_extent_item *extent_item; |
8103 | struct btrfs_tree_block_info *block_info; | |
8104 | struct btrfs_extent_inline_ref *iref; | |
8105 | struct btrfs_path *path; | |
8106 | struct extent_buffer *leaf; | |
3173a18f | 8107 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 8108 | u64 num_bytes = ins->offset; |
0b246afa | 8109 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f JB |
8110 | |
8111 | if (!skinny_metadata) | |
8112 | size += sizeof(*block_info); | |
1c2308f8 | 8113 | |
5d4f98a2 | 8114 | path = btrfs_alloc_path(); |
857cc2fc | 8115 | if (!path) { |
2ff7e61e | 8116 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8117 | fs_info->nodesize); |
d8926bb3 | 8118 | return -ENOMEM; |
857cc2fc | 8119 | } |
56bec294 | 8120 | |
5d4f98a2 YZ |
8121 | path->leave_spinning = 1; |
8122 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
8123 | ins, size); | |
79787eaa | 8124 | if (ret) { |
dd825259 | 8125 | btrfs_free_path(path); |
2ff7e61e | 8126 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8127 | fs_info->nodesize); |
79787eaa JM |
8128 | return ret; |
8129 | } | |
5d4f98a2 YZ |
8130 | |
8131 | leaf = path->nodes[0]; | |
8132 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8133 | struct btrfs_extent_item); | |
8134 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8135 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8136 | btrfs_set_extent_flags(leaf, extent_item, | |
8137 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8138 | |
3173a18f JB |
8139 | if (skinny_metadata) { |
8140 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
0b246afa | 8141 | num_bytes = fs_info->nodesize; |
3173a18f JB |
8142 | } else { |
8143 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
8144 | btrfs_set_tree_block_key(leaf, block_info, key); | |
8145 | btrfs_set_tree_block_level(leaf, block_info, level); | |
8146 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
8147 | } | |
5d4f98a2 | 8148 | |
5d4f98a2 YZ |
8149 | if (parent > 0) { |
8150 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
8151 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8152 | BTRFS_SHARED_BLOCK_REF_KEY); | |
8153 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8154 | } else { | |
8155 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8156 | BTRFS_TREE_BLOCK_REF_KEY); | |
8157 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
8158 | } | |
8159 | ||
8160 | btrfs_mark_buffer_dirty(leaf); | |
8161 | btrfs_free_path(path); | |
8162 | ||
1e144fb8 OS |
8163 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8164 | num_bytes); | |
8165 | if (ret) | |
8166 | return ret; | |
8167 | ||
6202df69 JM |
8168 | ret = update_block_group(trans, fs_info, ins->objectid, |
8169 | fs_info->nodesize, 1); | |
79787eaa | 8170 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8171 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8172 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
8173 | BUG(); |
8174 | } | |
0be5dc67 | 8175 | |
71ff6437 | 8176 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, |
0b246afa | 8177 | fs_info->nodesize); |
5d4f98a2 YZ |
8178 | return ret; |
8179 | } | |
8180 | ||
8181 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 | 8182 | u64 root_objectid, u64 owner, |
5846a3c2 QW |
8183 | u64 offset, u64 ram_bytes, |
8184 | struct btrfs_key *ins) | |
5d4f98a2 | 8185 | { |
2ff7e61e | 8186 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
8187 | int ret; |
8188 | ||
8189 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
8190 | ||
0b246afa | 8191 | ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid, |
66d7e7f0 AJ |
8192 | ins->offset, 0, |
8193 | root_objectid, owner, offset, | |
fef394f7 | 8194 | ram_bytes, BTRFS_ADD_DELAYED_EXTENT); |
e6dcd2dc CM |
8195 | return ret; |
8196 | } | |
e02119d5 CM |
8197 | |
8198 | /* | |
8199 | * this is used by the tree logging recovery code. It records that | |
8200 | * an extent has been allocated and makes sure to clear the free | |
8201 | * space cache bits as well | |
8202 | */ | |
5d4f98a2 | 8203 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8204 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8205 | u64 root_objectid, u64 owner, u64 offset, |
8206 | struct btrfs_key *ins) | |
e02119d5 CM |
8207 | { |
8208 | int ret; | |
8209 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8210 | struct btrfs_space_info *space_info; |
11833d66 | 8211 | |
8c2a1a30 JB |
8212 | /* |
8213 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8214 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8215 | */ |
0b246afa | 8216 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8217 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8218 | ins->offset); | |
b50c6e25 | 8219 | if (ret) |
8c2a1a30 | 8220 | return ret; |
11833d66 YZ |
8221 | } |
8222 | ||
0b246afa | 8223 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8224 | if (!block_group) |
8225 | return -EINVAL; | |
8226 | ||
ed7a6948 WX |
8227 | space_info = block_group->space_info; |
8228 | spin_lock(&space_info->lock); | |
8229 | spin_lock(&block_group->lock); | |
8230 | space_info->bytes_reserved += ins->offset; | |
8231 | block_group->reserved += ins->offset; | |
8232 | spin_unlock(&block_group->lock); | |
8233 | spin_unlock(&space_info->lock); | |
8234 | ||
2ff7e61e | 8235 | ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid, |
5d4f98a2 | 8236 | 0, owner, offset, ins, 1); |
b50c6e25 | 8237 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8238 | return ret; |
8239 | } | |
8240 | ||
48a3b636 ES |
8241 | static struct extent_buffer * |
8242 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8243 | u64 bytenr, int level) |
65b51a00 | 8244 | { |
0b246afa | 8245 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8246 | struct extent_buffer *buf; |
8247 | ||
2ff7e61e | 8248 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8249 | if (IS_ERR(buf)) |
8250 | return buf; | |
8251 | ||
65b51a00 | 8252 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8253 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8254 | btrfs_tree_lock(buf); |
7c302b49 | 8255 | clean_tree_block(fs_info, buf); |
3083ee2e | 8256 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8257 | |
8258 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8259 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8260 | |
d0c803c4 | 8261 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8262 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8263 | /* |
8264 | * we allow two log transactions at a time, use different | |
8265 | * EXENT bit to differentiate dirty pages. | |
8266 | */ | |
656f30db | 8267 | if (buf->log_index == 0) |
8cef4e16 YZ |
8268 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8269 | buf->start + buf->len - 1, GFP_NOFS); | |
8270 | else | |
8271 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8272 | buf->start + buf->len - 1); |
d0c803c4 | 8273 | } else { |
656f30db | 8274 | buf->log_index = -1; |
d0c803c4 | 8275 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8276 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8277 | } |
64c12921 | 8278 | trans->dirty = true; |
b4ce94de | 8279 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8280 | return buf; |
8281 | } | |
8282 | ||
f0486c68 YZ |
8283 | static struct btrfs_block_rsv * |
8284 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8285 | struct btrfs_root *root, u32 blocksize) | |
8286 | { | |
0b246afa | 8287 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8288 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8289 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8290 | int ret; |
d88033db | 8291 | bool global_updated = false; |
f0486c68 YZ |
8292 | |
8293 | block_rsv = get_block_rsv(trans, root); | |
8294 | ||
b586b323 MX |
8295 | if (unlikely(block_rsv->size == 0)) |
8296 | goto try_reserve; | |
d88033db | 8297 | again: |
f0486c68 YZ |
8298 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8299 | if (!ret) | |
8300 | return block_rsv; | |
8301 | ||
b586b323 MX |
8302 | if (block_rsv->failfast) |
8303 | return ERR_PTR(ret); | |
8304 | ||
d88033db MX |
8305 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8306 | global_updated = true; | |
0b246afa | 8307 | update_global_block_rsv(fs_info); |
d88033db MX |
8308 | goto again; |
8309 | } | |
8310 | ||
0b246afa | 8311 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8312 | static DEFINE_RATELIMIT_STATE(_rs, |
8313 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8314 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8315 | if (__ratelimit(&_rs)) | |
8316 | WARN(1, KERN_DEBUG | |
efe120a0 | 8317 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8318 | } |
8319 | try_reserve: | |
8320 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8321 | BTRFS_RESERVE_NO_FLUSH); | |
8322 | if (!ret) | |
8323 | return block_rsv; | |
8324 | /* | |
8325 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8326 | * the global reserve if its space type is the same as the global |
8327 | * reservation. | |
b586b323 | 8328 | */ |
5881cfc9 MX |
8329 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8330 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8331 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8332 | if (!ret) | |
8333 | return global_rsv; | |
8334 | } | |
8335 | return ERR_PTR(ret); | |
f0486c68 YZ |
8336 | } |
8337 | ||
8c2a3ca2 JB |
8338 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8339 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8340 | { |
8341 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 8342 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
8343 | } |
8344 | ||
fec577fb | 8345 | /* |
f0486c68 | 8346 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8347 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8348 | */ |
4d75f8a9 | 8349 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8350 | struct btrfs_root *root, |
8351 | u64 parent, u64 root_objectid, | |
8352 | const struct btrfs_disk_key *key, | |
8353 | int level, u64 hint, | |
8354 | u64 empty_size) | |
fec577fb | 8355 | { |
0b246afa | 8356 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8357 | struct btrfs_key ins; |
f0486c68 | 8358 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8359 | struct extent_buffer *buf; |
67b7859e | 8360 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8361 | u64 flags = 0; |
8362 | int ret; | |
0b246afa JM |
8363 | u32 blocksize = fs_info->nodesize; |
8364 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8365 | |
05653ef3 | 8366 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8367 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8368 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8369 | level); |
faa2dbf0 JB |
8370 | if (!IS_ERR(buf)) |
8371 | root->alloc_bytenr += blocksize; | |
8372 | return buf; | |
8373 | } | |
05653ef3 | 8374 | #endif |
fccb84c9 | 8375 | |
f0486c68 YZ |
8376 | block_rsv = use_block_rsv(trans, root, blocksize); |
8377 | if (IS_ERR(block_rsv)) | |
8378 | return ERR_CAST(block_rsv); | |
8379 | ||
18513091 | 8380 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8381 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8382 | if (ret) |
8383 | goto out_unuse; | |
55c69072 | 8384 | |
fe864576 | 8385 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8386 | if (IS_ERR(buf)) { |
8387 | ret = PTR_ERR(buf); | |
8388 | goto out_free_reserved; | |
8389 | } | |
f0486c68 YZ |
8390 | |
8391 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8392 | if (parent == 0) | |
8393 | parent = ins.objectid; | |
8394 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8395 | } else | |
8396 | BUG_ON(parent > 0); | |
8397 | ||
8398 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8399 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8400 | if (!extent_op) { |
8401 | ret = -ENOMEM; | |
8402 | goto out_free_buf; | |
8403 | } | |
f0486c68 YZ |
8404 | if (key) |
8405 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8406 | else | |
8407 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8408 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8409 | extent_op->update_key = skinny_metadata ? false : true; |
8410 | extent_op->update_flags = true; | |
8411 | extent_op->is_data = false; | |
b1c79e09 | 8412 | extent_op->level = level; |
f0486c68 | 8413 | |
0b246afa | 8414 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, |
67b7859e OS |
8415 | ins.objectid, ins.offset, |
8416 | parent, root_objectid, level, | |
8417 | BTRFS_ADD_DELAYED_EXTENT, | |
b06c4bf5 | 8418 | extent_op); |
67b7859e OS |
8419 | if (ret) |
8420 | goto out_free_delayed; | |
f0486c68 | 8421 | } |
fec577fb | 8422 | return buf; |
67b7859e OS |
8423 | |
8424 | out_free_delayed: | |
8425 | btrfs_free_delayed_extent_op(extent_op); | |
8426 | out_free_buf: | |
8427 | free_extent_buffer(buf); | |
8428 | out_free_reserved: | |
2ff7e61e | 8429 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8430 | out_unuse: |
0b246afa | 8431 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8432 | return ERR_PTR(ret); |
fec577fb | 8433 | } |
a28ec197 | 8434 | |
2c47e605 YZ |
8435 | struct walk_control { |
8436 | u64 refs[BTRFS_MAX_LEVEL]; | |
8437 | u64 flags[BTRFS_MAX_LEVEL]; | |
8438 | struct btrfs_key update_progress; | |
8439 | int stage; | |
8440 | int level; | |
8441 | int shared_level; | |
8442 | int update_ref; | |
8443 | int keep_locks; | |
1c4850e2 YZ |
8444 | int reada_slot; |
8445 | int reada_count; | |
66d7e7f0 | 8446 | int for_reloc; |
2c47e605 YZ |
8447 | }; |
8448 | ||
8449 | #define DROP_REFERENCE 1 | |
8450 | #define UPDATE_BACKREF 2 | |
8451 | ||
1c4850e2 YZ |
8452 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8453 | struct btrfs_root *root, | |
8454 | struct walk_control *wc, | |
8455 | struct btrfs_path *path) | |
6407bf6d | 8456 | { |
0b246afa | 8457 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8458 | u64 bytenr; |
8459 | u64 generation; | |
8460 | u64 refs; | |
94fcca9f | 8461 | u64 flags; |
5d4f98a2 | 8462 | u32 nritems; |
1c4850e2 YZ |
8463 | struct btrfs_key key; |
8464 | struct extent_buffer *eb; | |
6407bf6d | 8465 | int ret; |
1c4850e2 YZ |
8466 | int slot; |
8467 | int nread = 0; | |
6407bf6d | 8468 | |
1c4850e2 YZ |
8469 | if (path->slots[wc->level] < wc->reada_slot) { |
8470 | wc->reada_count = wc->reada_count * 2 / 3; | |
8471 | wc->reada_count = max(wc->reada_count, 2); | |
8472 | } else { | |
8473 | wc->reada_count = wc->reada_count * 3 / 2; | |
8474 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8475 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8476 | } |
7bb86316 | 8477 | |
1c4850e2 YZ |
8478 | eb = path->nodes[wc->level]; |
8479 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8480 | |
1c4850e2 YZ |
8481 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8482 | if (nread >= wc->reada_count) | |
8483 | break; | |
bd56b302 | 8484 | |
2dd3e67b | 8485 | cond_resched(); |
1c4850e2 YZ |
8486 | bytenr = btrfs_node_blockptr(eb, slot); |
8487 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8488 | |
1c4850e2 YZ |
8489 | if (slot == path->slots[wc->level]) |
8490 | goto reada; | |
5d4f98a2 | 8491 | |
1c4850e2 YZ |
8492 | if (wc->stage == UPDATE_BACKREF && |
8493 | generation <= root->root_key.offset) | |
bd56b302 CM |
8494 | continue; |
8495 | ||
94fcca9f | 8496 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8497 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8498 | wc->level - 1, 1, &refs, |
8499 | &flags); | |
79787eaa JM |
8500 | /* We don't care about errors in readahead. */ |
8501 | if (ret < 0) | |
8502 | continue; | |
94fcca9f YZ |
8503 | BUG_ON(refs == 0); |
8504 | ||
1c4850e2 | 8505 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8506 | if (refs == 1) |
8507 | goto reada; | |
bd56b302 | 8508 | |
94fcca9f YZ |
8509 | if (wc->level == 1 && |
8510 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8511 | continue; | |
1c4850e2 YZ |
8512 | if (!wc->update_ref || |
8513 | generation <= root->root_key.offset) | |
8514 | continue; | |
8515 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8516 | ret = btrfs_comp_cpu_keys(&key, | |
8517 | &wc->update_progress); | |
8518 | if (ret < 0) | |
8519 | continue; | |
94fcca9f YZ |
8520 | } else { |
8521 | if (wc->level == 1 && | |
8522 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8523 | continue; | |
6407bf6d | 8524 | } |
1c4850e2 | 8525 | reada: |
2ff7e61e | 8526 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8527 | nread++; |
20524f02 | 8528 | } |
1c4850e2 | 8529 | wc->reada_slot = slot; |
20524f02 | 8530 | } |
2c47e605 | 8531 | |
f82d02d9 | 8532 | /* |
2c016dc2 | 8533 | * helper to process tree block while walking down the tree. |
2c47e605 | 8534 | * |
2c47e605 YZ |
8535 | * when wc->stage == UPDATE_BACKREF, this function updates |
8536 | * back refs for pointers in the block. | |
8537 | * | |
8538 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8539 | */ |
2c47e605 | 8540 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8541 | struct btrfs_root *root, |
2c47e605 | 8542 | struct btrfs_path *path, |
94fcca9f | 8543 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8544 | { |
2ff7e61e | 8545 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8546 | int level = wc->level; |
8547 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8548 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8549 | int ret; |
8550 | ||
2c47e605 YZ |
8551 | if (wc->stage == UPDATE_BACKREF && |
8552 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8553 | return 1; | |
f82d02d9 | 8554 | |
2c47e605 YZ |
8555 | /* |
8556 | * when reference count of tree block is 1, it won't increase | |
8557 | * again. once full backref flag is set, we never clear it. | |
8558 | */ | |
94fcca9f YZ |
8559 | if (lookup_info && |
8560 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8561 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8562 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8563 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8564 | eb->start, level, 1, |
2c47e605 YZ |
8565 | &wc->refs[level], |
8566 | &wc->flags[level]); | |
79787eaa JM |
8567 | BUG_ON(ret == -ENOMEM); |
8568 | if (ret) | |
8569 | return ret; | |
2c47e605 YZ |
8570 | BUG_ON(wc->refs[level] == 0); |
8571 | } | |
5d4f98a2 | 8572 | |
2c47e605 YZ |
8573 | if (wc->stage == DROP_REFERENCE) { |
8574 | if (wc->refs[level] > 1) | |
8575 | return 1; | |
f82d02d9 | 8576 | |
2c47e605 | 8577 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8578 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8579 | path->locks[level] = 0; |
8580 | } | |
8581 | return 0; | |
8582 | } | |
f82d02d9 | 8583 | |
2c47e605 YZ |
8584 | /* wc->stage == UPDATE_BACKREF */ |
8585 | if (!(wc->flags[level] & flag)) { | |
8586 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8587 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8588 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8589 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8590 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8591 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8592 | eb->len, flag, |
8593 | btrfs_header_level(eb), 0); | |
79787eaa | 8594 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8595 | wc->flags[level] |= flag; |
8596 | } | |
8597 | ||
8598 | /* | |
8599 | * the block is shared by multiple trees, so it's not good to | |
8600 | * keep the tree lock | |
8601 | */ | |
8602 | if (path->locks[level] && level > 0) { | |
bd681513 | 8603 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8604 | path->locks[level] = 0; |
8605 | } | |
8606 | return 0; | |
8607 | } | |
8608 | ||
1c4850e2 | 8609 | /* |
2c016dc2 | 8610 | * helper to process tree block pointer. |
1c4850e2 YZ |
8611 | * |
8612 | * when wc->stage == DROP_REFERENCE, this function checks | |
8613 | * reference count of the block pointed to. if the block | |
8614 | * is shared and we need update back refs for the subtree | |
8615 | * rooted at the block, this function changes wc->stage to | |
8616 | * UPDATE_BACKREF. if the block is shared and there is no | |
8617 | * need to update back, this function drops the reference | |
8618 | * to the block. | |
8619 | * | |
8620 | * NOTE: return value 1 means we should stop walking down. | |
8621 | */ | |
8622 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8623 | struct btrfs_root *root, | |
8624 | struct btrfs_path *path, | |
94fcca9f | 8625 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8626 | { |
0b246afa | 8627 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8628 | u64 bytenr; |
8629 | u64 generation; | |
8630 | u64 parent; | |
8631 | u32 blocksize; | |
8632 | struct btrfs_key key; | |
8633 | struct extent_buffer *next; | |
8634 | int level = wc->level; | |
8635 | int reada = 0; | |
8636 | int ret = 0; | |
1152651a | 8637 | bool need_account = false; |
1c4850e2 YZ |
8638 | |
8639 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8640 | path->slots[level]); | |
8641 | /* | |
8642 | * if the lower level block was created before the snapshot | |
8643 | * was created, we know there is no need to update back refs | |
8644 | * for the subtree | |
8645 | */ | |
8646 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8647 | generation <= root->root_key.offset) { |
8648 | *lookup_info = 1; | |
1c4850e2 | 8649 | return 1; |
94fcca9f | 8650 | } |
1c4850e2 YZ |
8651 | |
8652 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
0b246afa | 8653 | blocksize = fs_info->nodesize; |
1c4850e2 | 8654 | |
0b246afa | 8655 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8656 | if (!next) { |
2ff7e61e | 8657 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8658 | if (IS_ERR(next)) |
8659 | return PTR_ERR(next); | |
8660 | ||
b2aaaa3b JB |
8661 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8662 | level - 1); | |
1c4850e2 YZ |
8663 | reada = 1; |
8664 | } | |
8665 | btrfs_tree_lock(next); | |
8666 | btrfs_set_lock_blocking(next); | |
8667 | ||
2ff7e61e | 8668 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8669 | &wc->refs[level - 1], |
8670 | &wc->flags[level - 1]); | |
4867268c JB |
8671 | if (ret < 0) |
8672 | goto out_unlock; | |
79787eaa | 8673 | |
c2cf52eb | 8674 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8675 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8676 | ret = -EIO; |
8677 | goto out_unlock; | |
c2cf52eb | 8678 | } |
94fcca9f | 8679 | *lookup_info = 0; |
1c4850e2 | 8680 | |
94fcca9f | 8681 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8682 | if (wc->refs[level - 1] > 1) { |
1152651a | 8683 | need_account = true; |
94fcca9f YZ |
8684 | if (level == 1 && |
8685 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8686 | goto skip; | |
8687 | ||
1c4850e2 YZ |
8688 | if (!wc->update_ref || |
8689 | generation <= root->root_key.offset) | |
8690 | goto skip; | |
8691 | ||
8692 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8693 | path->slots[level]); | |
8694 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8695 | if (ret < 0) | |
8696 | goto skip; | |
8697 | ||
8698 | wc->stage = UPDATE_BACKREF; | |
8699 | wc->shared_level = level - 1; | |
8700 | } | |
94fcca9f YZ |
8701 | } else { |
8702 | if (level == 1 && | |
8703 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8704 | goto skip; | |
1c4850e2 YZ |
8705 | } |
8706 | ||
b9fab919 | 8707 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8708 | btrfs_tree_unlock(next); |
8709 | free_extent_buffer(next); | |
8710 | next = NULL; | |
94fcca9f | 8711 | *lookup_info = 1; |
1c4850e2 YZ |
8712 | } |
8713 | ||
8714 | if (!next) { | |
8715 | if (reada && level == 1) | |
8716 | reada_walk_down(trans, root, wc, path); | |
2ff7e61e | 8717 | next = read_tree_block(fs_info, bytenr, generation); |
64c043de LB |
8718 | if (IS_ERR(next)) { |
8719 | return PTR_ERR(next); | |
8720 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8721 | free_extent_buffer(next); |
97d9a8a4 | 8722 | return -EIO; |
416bc658 | 8723 | } |
1c4850e2 YZ |
8724 | btrfs_tree_lock(next); |
8725 | btrfs_set_lock_blocking(next); | |
8726 | } | |
8727 | ||
8728 | level--; | |
4867268c JB |
8729 | ASSERT(level == btrfs_header_level(next)); |
8730 | if (level != btrfs_header_level(next)) { | |
8731 | btrfs_err(root->fs_info, "mismatched level"); | |
8732 | ret = -EIO; | |
8733 | goto out_unlock; | |
8734 | } | |
1c4850e2 YZ |
8735 | path->nodes[level] = next; |
8736 | path->slots[level] = 0; | |
bd681513 | 8737 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8738 | wc->level = level; |
8739 | if (wc->level == 1) | |
8740 | wc->reada_slot = 0; | |
8741 | return 0; | |
8742 | skip: | |
8743 | wc->refs[level - 1] = 0; | |
8744 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8745 | if (wc->stage == DROP_REFERENCE) { |
8746 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8747 | parent = path->nodes[level]->start; | |
8748 | } else { | |
4867268c | 8749 | ASSERT(root->root_key.objectid == |
94fcca9f | 8750 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8751 | if (root->root_key.objectid != |
8752 | btrfs_header_owner(path->nodes[level])) { | |
8753 | btrfs_err(root->fs_info, | |
8754 | "mismatched block owner"); | |
8755 | ret = -EIO; | |
8756 | goto out_unlock; | |
8757 | } | |
94fcca9f YZ |
8758 | parent = 0; |
8759 | } | |
1c4850e2 | 8760 | |
1152651a | 8761 | if (need_account) { |
33d1f05c QW |
8762 | ret = btrfs_qgroup_trace_subtree(trans, root, next, |
8763 | generation, level - 1); | |
1152651a | 8764 | if (ret) { |
0b246afa | 8765 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8766 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8767 | ret); | |
1152651a MF |
8768 | } |
8769 | } | |
2ff7e61e JM |
8770 | ret = btrfs_free_extent(trans, fs_info, bytenr, blocksize, |
8771 | parent, root->root_key.objectid, | |
8772 | level - 1, 0); | |
4867268c JB |
8773 | if (ret) |
8774 | goto out_unlock; | |
1c4850e2 | 8775 | } |
4867268c JB |
8776 | |
8777 | *lookup_info = 1; | |
8778 | ret = 1; | |
8779 | ||
8780 | out_unlock: | |
1c4850e2 YZ |
8781 | btrfs_tree_unlock(next); |
8782 | free_extent_buffer(next); | |
4867268c JB |
8783 | |
8784 | return ret; | |
1c4850e2 YZ |
8785 | } |
8786 | ||
2c47e605 | 8787 | /* |
2c016dc2 | 8788 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8789 | * |
8790 | * when wc->stage == DROP_REFERENCE, this function drops | |
8791 | * reference count on the block. | |
8792 | * | |
8793 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8794 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8795 | * to UPDATE_BACKREF previously while processing the block. | |
8796 | * | |
8797 | * NOTE: return value 1 means we should stop walking up. | |
8798 | */ | |
8799 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8800 | struct btrfs_root *root, | |
8801 | struct btrfs_path *path, | |
8802 | struct walk_control *wc) | |
8803 | { | |
0b246afa | 8804 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8805 | int ret; |
2c47e605 YZ |
8806 | int level = wc->level; |
8807 | struct extent_buffer *eb = path->nodes[level]; | |
8808 | u64 parent = 0; | |
8809 | ||
8810 | if (wc->stage == UPDATE_BACKREF) { | |
8811 | BUG_ON(wc->shared_level < level); | |
8812 | if (level < wc->shared_level) | |
8813 | goto out; | |
8814 | ||
2c47e605 YZ |
8815 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8816 | if (ret > 0) | |
8817 | wc->update_ref = 0; | |
8818 | ||
8819 | wc->stage = DROP_REFERENCE; | |
8820 | wc->shared_level = -1; | |
8821 | path->slots[level] = 0; | |
8822 | ||
8823 | /* | |
8824 | * check reference count again if the block isn't locked. | |
8825 | * we should start walking down the tree again if reference | |
8826 | * count is one. | |
8827 | */ | |
8828 | if (!path->locks[level]) { | |
8829 | BUG_ON(level == 0); | |
8830 | btrfs_tree_lock(eb); | |
8831 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8832 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8833 | |
2ff7e61e | 8834 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8835 | eb->start, level, 1, |
2c47e605 YZ |
8836 | &wc->refs[level], |
8837 | &wc->flags[level]); | |
79787eaa JM |
8838 | if (ret < 0) { |
8839 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8840 | path->locks[level] = 0; |
79787eaa JM |
8841 | return ret; |
8842 | } | |
2c47e605 YZ |
8843 | BUG_ON(wc->refs[level] == 0); |
8844 | if (wc->refs[level] == 1) { | |
bd681513 | 8845 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8846 | path->locks[level] = 0; |
2c47e605 YZ |
8847 | return 1; |
8848 | } | |
f82d02d9 | 8849 | } |
2c47e605 | 8850 | } |
f82d02d9 | 8851 | |
2c47e605 YZ |
8852 | /* wc->stage == DROP_REFERENCE */ |
8853 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8854 | |
2c47e605 YZ |
8855 | if (wc->refs[level] == 1) { |
8856 | if (level == 0) { | |
8857 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8858 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8859 | else |
e339a6b0 | 8860 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8861 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8862 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb); |
1152651a | 8863 | if (ret) { |
0b246afa | 8864 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8865 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
8866 | ret); | |
1152651a | 8867 | } |
2c47e605 YZ |
8868 | } |
8869 | /* make block locked assertion in clean_tree_block happy */ | |
8870 | if (!path->locks[level] && | |
8871 | btrfs_header_generation(eb) == trans->transid) { | |
8872 | btrfs_tree_lock(eb); | |
8873 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8874 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8875 | } |
7c302b49 | 8876 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
8877 | } |
8878 | ||
8879 | if (eb == root->node) { | |
8880 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8881 | parent = eb->start; | |
8882 | else | |
8883 | BUG_ON(root->root_key.objectid != | |
8884 | btrfs_header_owner(eb)); | |
8885 | } else { | |
8886 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8887 | parent = path->nodes[level + 1]->start; | |
8888 | else | |
8889 | BUG_ON(root->root_key.objectid != | |
8890 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8891 | } |
f82d02d9 | 8892 | |
5581a51a | 8893 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8894 | out: |
8895 | wc->refs[level] = 0; | |
8896 | wc->flags[level] = 0; | |
f0486c68 | 8897 | return 0; |
2c47e605 YZ |
8898 | } |
8899 | ||
8900 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8901 | struct btrfs_root *root, | |
8902 | struct btrfs_path *path, | |
8903 | struct walk_control *wc) | |
8904 | { | |
2c47e605 | 8905 | int level = wc->level; |
94fcca9f | 8906 | int lookup_info = 1; |
2c47e605 YZ |
8907 | int ret; |
8908 | ||
8909 | while (level >= 0) { | |
94fcca9f | 8910 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8911 | if (ret > 0) |
8912 | break; | |
8913 | ||
8914 | if (level == 0) | |
8915 | break; | |
8916 | ||
7a7965f8 YZ |
8917 | if (path->slots[level] >= |
8918 | btrfs_header_nritems(path->nodes[level])) | |
8919 | break; | |
8920 | ||
94fcca9f | 8921 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8922 | if (ret > 0) { |
8923 | path->slots[level]++; | |
8924 | continue; | |
90d2c51d MX |
8925 | } else if (ret < 0) |
8926 | return ret; | |
1c4850e2 | 8927 | level = wc->level; |
f82d02d9 | 8928 | } |
f82d02d9 YZ |
8929 | return 0; |
8930 | } | |
8931 | ||
d397712b | 8932 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 8933 | struct btrfs_root *root, |
f82d02d9 | 8934 | struct btrfs_path *path, |
2c47e605 | 8935 | struct walk_control *wc, int max_level) |
20524f02 | 8936 | { |
2c47e605 | 8937 | int level = wc->level; |
20524f02 | 8938 | int ret; |
9f3a7427 | 8939 | |
2c47e605 YZ |
8940 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
8941 | while (level < max_level && path->nodes[level]) { | |
8942 | wc->level = level; | |
8943 | if (path->slots[level] + 1 < | |
8944 | btrfs_header_nritems(path->nodes[level])) { | |
8945 | path->slots[level]++; | |
20524f02 CM |
8946 | return 0; |
8947 | } else { | |
2c47e605 YZ |
8948 | ret = walk_up_proc(trans, root, path, wc); |
8949 | if (ret > 0) | |
8950 | return 0; | |
bd56b302 | 8951 | |
2c47e605 | 8952 | if (path->locks[level]) { |
bd681513 CM |
8953 | btrfs_tree_unlock_rw(path->nodes[level], |
8954 | path->locks[level]); | |
2c47e605 | 8955 | path->locks[level] = 0; |
f82d02d9 | 8956 | } |
2c47e605 YZ |
8957 | free_extent_buffer(path->nodes[level]); |
8958 | path->nodes[level] = NULL; | |
8959 | level++; | |
20524f02 CM |
8960 | } |
8961 | } | |
8962 | return 1; | |
8963 | } | |
8964 | ||
9aca1d51 | 8965 | /* |
2c47e605 YZ |
8966 | * drop a subvolume tree. |
8967 | * | |
8968 | * this function traverses the tree freeing any blocks that only | |
8969 | * referenced by the tree. | |
8970 | * | |
8971 | * when a shared tree block is found. this function decreases its | |
8972 | * reference count by one. if update_ref is true, this function | |
8973 | * also make sure backrefs for the shared block and all lower level | |
8974 | * blocks are properly updated. | |
9d1a2a3a DS |
8975 | * |
8976 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 8977 | */ |
2c536799 | 8978 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
8979 | struct btrfs_block_rsv *block_rsv, int update_ref, |
8980 | int for_reloc) | |
20524f02 | 8981 | { |
ab8d0fc4 | 8982 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 8983 | struct btrfs_path *path; |
2c47e605 | 8984 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 8985 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 8986 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
8987 | struct walk_control *wc; |
8988 | struct btrfs_key key; | |
8989 | int err = 0; | |
8990 | int ret; | |
8991 | int level; | |
d29a9f62 | 8992 | bool root_dropped = false; |
20524f02 | 8993 | |
ab8d0fc4 | 8994 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
1152651a | 8995 | |
5caf2a00 | 8996 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
8997 | if (!path) { |
8998 | err = -ENOMEM; | |
8999 | goto out; | |
9000 | } | |
20524f02 | 9001 | |
2c47e605 | 9002 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9003 | if (!wc) { |
9004 | btrfs_free_path(path); | |
cb1b69f4 TI |
9005 | err = -ENOMEM; |
9006 | goto out; | |
38a1a919 | 9007 | } |
2c47e605 | 9008 | |
a22285a6 | 9009 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9010 | if (IS_ERR(trans)) { |
9011 | err = PTR_ERR(trans); | |
9012 | goto out_free; | |
9013 | } | |
98d5dc13 | 9014 | |
3fd0a558 YZ |
9015 | if (block_rsv) |
9016 | trans->block_rsv = block_rsv; | |
2c47e605 | 9017 | |
9f3a7427 | 9018 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9019 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9020 | path->nodes[level] = btrfs_lock_root_node(root); |
9021 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9022 | path->slots[level] = 0; |
bd681513 | 9023 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9024 | memset(&wc->update_progress, 0, |
9025 | sizeof(wc->update_progress)); | |
9f3a7427 | 9026 | } else { |
9f3a7427 | 9027 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9028 | memcpy(&wc->update_progress, &key, |
9029 | sizeof(wc->update_progress)); | |
9030 | ||
6702ed49 | 9031 | level = root_item->drop_level; |
2c47e605 | 9032 | BUG_ON(level == 0); |
6702ed49 | 9033 | path->lowest_level = level; |
2c47e605 YZ |
9034 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9035 | path->lowest_level = 0; | |
9036 | if (ret < 0) { | |
9037 | err = ret; | |
79787eaa | 9038 | goto out_end_trans; |
9f3a7427 | 9039 | } |
1c4850e2 | 9040 | WARN_ON(ret > 0); |
2c47e605 | 9041 | |
7d9eb12c CM |
9042 | /* |
9043 | * unlock our path, this is safe because only this | |
9044 | * function is allowed to delete this snapshot | |
9045 | */ | |
5d4f98a2 | 9046 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9047 | |
9048 | level = btrfs_header_level(root->node); | |
9049 | while (1) { | |
9050 | btrfs_tree_lock(path->nodes[level]); | |
9051 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9052 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9053 | |
2ff7e61e | 9054 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9055 | path->nodes[level]->start, |
3173a18f | 9056 | level, 1, &wc->refs[level], |
2c47e605 | 9057 | &wc->flags[level]); |
79787eaa JM |
9058 | if (ret < 0) { |
9059 | err = ret; | |
9060 | goto out_end_trans; | |
9061 | } | |
2c47e605 YZ |
9062 | BUG_ON(wc->refs[level] == 0); |
9063 | ||
9064 | if (level == root_item->drop_level) | |
9065 | break; | |
9066 | ||
9067 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9068 | path->locks[level] = 0; |
2c47e605 YZ |
9069 | WARN_ON(wc->refs[level] != 1); |
9070 | level--; | |
9071 | } | |
9f3a7427 | 9072 | } |
2c47e605 YZ |
9073 | |
9074 | wc->level = level; | |
9075 | wc->shared_level = -1; | |
9076 | wc->stage = DROP_REFERENCE; | |
9077 | wc->update_ref = update_ref; | |
9078 | wc->keep_locks = 0; | |
66d7e7f0 | 9079 | wc->for_reloc = for_reloc; |
0b246afa | 9080 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9081 | |
d397712b | 9082 | while (1) { |
9d1a2a3a | 9083 | |
2c47e605 YZ |
9084 | ret = walk_down_tree(trans, root, path, wc); |
9085 | if (ret < 0) { | |
9086 | err = ret; | |
20524f02 | 9087 | break; |
2c47e605 | 9088 | } |
9aca1d51 | 9089 | |
2c47e605 YZ |
9090 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9091 | if (ret < 0) { | |
9092 | err = ret; | |
20524f02 | 9093 | break; |
2c47e605 YZ |
9094 | } |
9095 | ||
9096 | if (ret > 0) { | |
9097 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9098 | break; |
9099 | } | |
2c47e605 YZ |
9100 | |
9101 | if (wc->stage == DROP_REFERENCE) { | |
9102 | level = wc->level; | |
9103 | btrfs_node_key(path->nodes[level], | |
9104 | &root_item->drop_progress, | |
9105 | path->slots[level]); | |
9106 | root_item->drop_level = level; | |
9107 | } | |
9108 | ||
9109 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9110 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9111 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9112 | ret = btrfs_update_root(trans, tree_root, |
9113 | &root->root_key, | |
9114 | root_item); | |
79787eaa | 9115 | if (ret) { |
66642832 | 9116 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9117 | err = ret; |
9118 | goto out_end_trans; | |
9119 | } | |
2c47e605 | 9120 | |
3a45bb20 | 9121 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9122 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9123 | btrfs_debug(fs_info, |
9124 | "drop snapshot early exit"); | |
3c8f2422 JB |
9125 | err = -EAGAIN; |
9126 | goto out_free; | |
9127 | } | |
9128 | ||
a22285a6 | 9129 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9130 | if (IS_ERR(trans)) { |
9131 | err = PTR_ERR(trans); | |
9132 | goto out_free; | |
9133 | } | |
3fd0a558 YZ |
9134 | if (block_rsv) |
9135 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9136 | } |
20524f02 | 9137 | } |
b3b4aa74 | 9138 | btrfs_release_path(path); |
79787eaa JM |
9139 | if (err) |
9140 | goto out_end_trans; | |
2c47e605 YZ |
9141 | |
9142 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa | 9143 | if (ret) { |
66642832 | 9144 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9145 | goto out_end_trans; |
9146 | } | |
2c47e605 | 9147 | |
76dda93c | 9148 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9149 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9150 | NULL, NULL); | |
79787eaa | 9151 | if (ret < 0) { |
66642832 | 9152 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9153 | err = ret; |
9154 | goto out_end_trans; | |
9155 | } else if (ret > 0) { | |
84cd948c JB |
9156 | /* if we fail to delete the orphan item this time |
9157 | * around, it'll get picked up the next time. | |
9158 | * | |
9159 | * The most common failure here is just -ENOENT. | |
9160 | */ | |
9161 | btrfs_del_orphan_item(trans, tree_root, | |
9162 | root->root_key.objectid); | |
76dda93c YZ |
9163 | } |
9164 | } | |
9165 | ||
27cdeb70 | 9166 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9167 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9168 | } else { |
9169 | free_extent_buffer(root->node); | |
9170 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9171 | btrfs_put_fs_root(root); |
76dda93c | 9172 | } |
d29a9f62 | 9173 | root_dropped = true; |
79787eaa | 9174 | out_end_trans: |
3a45bb20 | 9175 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9176 | out_free: |
2c47e605 | 9177 | kfree(wc); |
5caf2a00 | 9178 | btrfs_free_path(path); |
cb1b69f4 | 9179 | out: |
d29a9f62 JB |
9180 | /* |
9181 | * So if we need to stop dropping the snapshot for whatever reason we | |
9182 | * need to make sure to add it back to the dead root list so that we | |
9183 | * keep trying to do the work later. This also cleans up roots if we | |
9184 | * don't have it in the radix (like when we recover after a power fail | |
9185 | * or unmount) so we don't leak memory. | |
9186 | */ | |
b37b39cd | 9187 | if (!for_reloc && root_dropped == false) |
d29a9f62 | 9188 | btrfs_add_dead_root(root); |
90515e7f | 9189 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9190 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9191 | return err; |
20524f02 | 9192 | } |
9078a3e1 | 9193 | |
2c47e605 YZ |
9194 | /* |
9195 | * drop subtree rooted at tree block 'node'. | |
9196 | * | |
9197 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9198 | * only used by relocation code |
2c47e605 | 9199 | */ |
f82d02d9 YZ |
9200 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9201 | struct btrfs_root *root, | |
9202 | struct extent_buffer *node, | |
9203 | struct extent_buffer *parent) | |
9204 | { | |
0b246afa | 9205 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9206 | struct btrfs_path *path; |
2c47e605 | 9207 | struct walk_control *wc; |
f82d02d9 YZ |
9208 | int level; |
9209 | int parent_level; | |
9210 | int ret = 0; | |
9211 | int wret; | |
9212 | ||
2c47e605 YZ |
9213 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9214 | ||
f82d02d9 | 9215 | path = btrfs_alloc_path(); |
db5b493a TI |
9216 | if (!path) |
9217 | return -ENOMEM; | |
f82d02d9 | 9218 | |
2c47e605 | 9219 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9220 | if (!wc) { |
9221 | btrfs_free_path(path); | |
9222 | return -ENOMEM; | |
9223 | } | |
2c47e605 | 9224 | |
b9447ef8 | 9225 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9226 | parent_level = btrfs_header_level(parent); |
9227 | extent_buffer_get(parent); | |
9228 | path->nodes[parent_level] = parent; | |
9229 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9230 | ||
b9447ef8 | 9231 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9232 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9233 | path->nodes[level] = node; |
9234 | path->slots[level] = 0; | |
bd681513 | 9235 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9236 | |
9237 | wc->refs[parent_level] = 1; | |
9238 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9239 | wc->level = level; | |
9240 | wc->shared_level = -1; | |
9241 | wc->stage = DROP_REFERENCE; | |
9242 | wc->update_ref = 0; | |
9243 | wc->keep_locks = 1; | |
66d7e7f0 | 9244 | wc->for_reloc = 1; |
0b246afa | 9245 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9246 | |
9247 | while (1) { | |
2c47e605 YZ |
9248 | wret = walk_down_tree(trans, root, path, wc); |
9249 | if (wret < 0) { | |
f82d02d9 | 9250 | ret = wret; |
f82d02d9 | 9251 | break; |
2c47e605 | 9252 | } |
f82d02d9 | 9253 | |
2c47e605 | 9254 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9255 | if (wret < 0) |
9256 | ret = wret; | |
9257 | if (wret != 0) | |
9258 | break; | |
9259 | } | |
9260 | ||
2c47e605 | 9261 | kfree(wc); |
f82d02d9 YZ |
9262 | btrfs_free_path(path); |
9263 | return ret; | |
9264 | } | |
9265 | ||
6202df69 | 9266 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9267 | { |
9268 | u64 num_devices; | |
fc67c450 | 9269 | u64 stripped; |
e4d8ec0f | 9270 | |
fc67c450 ID |
9271 | /* |
9272 | * if restripe for this chunk_type is on pick target profile and | |
9273 | * return, otherwise do the usual balance | |
9274 | */ | |
6202df69 | 9275 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9276 | if (stripped) |
9277 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9278 | |
6202df69 | 9279 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9280 | |
fc67c450 | 9281 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9282 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9283 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9284 | ||
ec44a35c CM |
9285 | if (num_devices == 1) { |
9286 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9287 | stripped = flags & ~stripped; | |
9288 | ||
9289 | /* turn raid0 into single device chunks */ | |
9290 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9291 | return stripped; | |
9292 | ||
9293 | /* turn mirroring into duplication */ | |
9294 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9295 | BTRFS_BLOCK_GROUP_RAID10)) | |
9296 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9297 | } else { |
9298 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9299 | if (flags & stripped) |
9300 | return flags; | |
9301 | ||
9302 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9303 | stripped = flags & ~stripped; | |
9304 | ||
9305 | /* switch duplicated blocks with raid1 */ | |
9306 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9307 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9308 | ||
e3176ca2 | 9309 | /* this is drive concat, leave it alone */ |
ec44a35c | 9310 | } |
e3176ca2 | 9311 | |
ec44a35c CM |
9312 | return flags; |
9313 | } | |
9314 | ||
868f401a | 9315 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9316 | { |
f0486c68 YZ |
9317 | struct btrfs_space_info *sinfo = cache->space_info; |
9318 | u64 num_bytes; | |
199c36ea | 9319 | u64 min_allocable_bytes; |
f0486c68 | 9320 | int ret = -ENOSPC; |
0ef3e66b | 9321 | |
199c36ea MX |
9322 | /* |
9323 | * We need some metadata space and system metadata space for | |
9324 | * allocating chunks in some corner cases until we force to set | |
9325 | * it to be readonly. | |
9326 | */ | |
9327 | if ((sinfo->flags & | |
9328 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9329 | !force) | |
ee22184b | 9330 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9331 | else |
9332 | min_allocable_bytes = 0; | |
9333 | ||
f0486c68 YZ |
9334 | spin_lock(&sinfo->lock); |
9335 | spin_lock(&cache->lock); | |
61cfea9b W |
9336 | |
9337 | if (cache->ro) { | |
868f401a | 9338 | cache->ro++; |
61cfea9b W |
9339 | ret = 0; |
9340 | goto out; | |
9341 | } | |
9342 | ||
f0486c68 YZ |
9343 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9344 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9345 | ||
4136135b | 9346 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9347 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9348 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9349 | cache->ro++; |
633c0aad | 9350 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9351 | ret = 0; |
9352 | } | |
61cfea9b | 9353 | out: |
f0486c68 YZ |
9354 | spin_unlock(&cache->lock); |
9355 | spin_unlock(&sinfo->lock); | |
9356 | return ret; | |
9357 | } | |
7d9eb12c | 9358 | |
5e00f193 | 9359 | int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info, |
f0486c68 | 9360 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9361 | |
f0486c68 YZ |
9362 | { |
9363 | struct btrfs_trans_handle *trans; | |
9364 | u64 alloc_flags; | |
9365 | int ret; | |
7d9eb12c | 9366 | |
1bbc621e | 9367 | again: |
5e00f193 | 9368 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9369 | if (IS_ERR(trans)) |
9370 | return PTR_ERR(trans); | |
5d4f98a2 | 9371 | |
1bbc621e CM |
9372 | /* |
9373 | * we're not allowed to set block groups readonly after the dirty | |
9374 | * block groups cache has started writing. If it already started, | |
9375 | * back off and let this transaction commit | |
9376 | */ | |
0b246afa | 9377 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9378 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9379 | u64 transid = trans->transid; |
9380 | ||
0b246afa | 9381 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9382 | btrfs_end_transaction(trans); |
1bbc621e | 9383 | |
2ff7e61e | 9384 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9385 | if (ret) |
9386 | return ret; | |
9387 | goto again; | |
9388 | } | |
9389 | ||
153c35b6 CM |
9390 | /* |
9391 | * if we are changing raid levels, try to allocate a corresponding | |
9392 | * block group with the new raid level. | |
9393 | */ | |
0b246afa | 9394 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9395 | if (alloc_flags != cache->flags) { |
2ff7e61e | 9396 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, |
153c35b6 CM |
9397 | CHUNK_ALLOC_FORCE); |
9398 | /* | |
9399 | * ENOSPC is allowed here, we may have enough space | |
9400 | * already allocated at the new raid level to | |
9401 | * carry on | |
9402 | */ | |
9403 | if (ret == -ENOSPC) | |
9404 | ret = 0; | |
9405 | if (ret < 0) | |
9406 | goto out; | |
9407 | } | |
1bbc621e | 9408 | |
868f401a | 9409 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9410 | if (!ret) |
9411 | goto out; | |
2ff7e61e JM |
9412 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
9413 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, | |
0e4f8f88 | 9414 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9415 | if (ret < 0) |
9416 | goto out; | |
868f401a | 9417 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9418 | out: |
2f081088 | 9419 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9420 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9421 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 9422 | check_system_chunk(trans, fs_info, alloc_flags); |
34441361 | 9423 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9424 | } |
0b246afa | 9425 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9426 | |
3a45bb20 | 9427 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9428 | return ret; |
9429 | } | |
5d4f98a2 | 9430 | |
c87f08ca | 9431 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 9432 | struct btrfs_fs_info *fs_info, u64 type) |
c87f08ca | 9433 | { |
2ff7e61e JM |
9434 | u64 alloc_flags = get_alloc_profile(fs_info, type); |
9435 | ||
9436 | return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
9437 | } |
9438 | ||
6d07bcec MX |
9439 | /* |
9440 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9441 | * space_info. takes mirrors into account. |
6d07bcec | 9442 | */ |
633c0aad | 9443 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9444 | { |
9445 | struct btrfs_block_group_cache *block_group; | |
9446 | u64 free_bytes = 0; | |
9447 | int factor; | |
9448 | ||
01327610 | 9449 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9450 | if (list_empty(&sinfo->ro_bgs)) |
9451 | return 0; | |
9452 | ||
9453 | spin_lock(&sinfo->lock); | |
9454 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9455 | spin_lock(&block_group->lock); |
9456 | ||
9457 | if (!block_group->ro) { | |
9458 | spin_unlock(&block_group->lock); | |
9459 | continue; | |
9460 | } | |
9461 | ||
9462 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9463 | BTRFS_BLOCK_GROUP_RAID10 | | |
9464 | BTRFS_BLOCK_GROUP_DUP)) | |
9465 | factor = 2; | |
9466 | else | |
9467 | factor = 1; | |
9468 | ||
9469 | free_bytes += (block_group->key.offset - | |
9470 | btrfs_block_group_used(&block_group->item)) * | |
9471 | factor; | |
9472 | ||
9473 | spin_unlock(&block_group->lock); | |
9474 | } | |
6d07bcec MX |
9475 | spin_unlock(&sinfo->lock); |
9476 | ||
9477 | return free_bytes; | |
9478 | } | |
9479 | ||
2ff7e61e | 9480 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9481 | { |
f0486c68 YZ |
9482 | struct btrfs_space_info *sinfo = cache->space_info; |
9483 | u64 num_bytes; | |
9484 | ||
9485 | BUG_ON(!cache->ro); | |
9486 | ||
9487 | spin_lock(&sinfo->lock); | |
9488 | spin_lock(&cache->lock); | |
868f401a Z |
9489 | if (!--cache->ro) { |
9490 | num_bytes = cache->key.offset - cache->reserved - | |
9491 | cache->pinned - cache->bytes_super - | |
9492 | btrfs_block_group_used(&cache->item); | |
9493 | sinfo->bytes_readonly -= num_bytes; | |
9494 | list_del_init(&cache->ro_list); | |
9495 | } | |
f0486c68 YZ |
9496 | spin_unlock(&cache->lock); |
9497 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9498 | } |
9499 | ||
ba1bf481 JB |
9500 | /* |
9501 | * checks to see if its even possible to relocate this block group. | |
9502 | * | |
9503 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9504 | * ok to go ahead and try. | |
9505 | */ | |
6bccf3ab | 9506 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9507 | { |
6bccf3ab | 9508 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9509 | struct btrfs_block_group_cache *block_group; |
9510 | struct btrfs_space_info *space_info; | |
0b246afa | 9511 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9512 | struct btrfs_device *device; |
6df9a95e | 9513 | struct btrfs_trans_handle *trans; |
cdcb725c | 9514 | u64 min_free; |
6719db6a JB |
9515 | u64 dev_min = 1; |
9516 | u64 dev_nr = 0; | |
4a5e98f5 | 9517 | u64 target; |
0305bc27 | 9518 | int debug; |
cdcb725c | 9519 | int index; |
ba1bf481 JB |
9520 | int full = 0; |
9521 | int ret = 0; | |
1a40e23b | 9522 | |
0b246afa | 9523 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9524 | |
0b246afa | 9525 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9526 | |
ba1bf481 | 9527 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9528 | if (!block_group) { |
9529 | if (debug) | |
0b246afa | 9530 | btrfs_warn(fs_info, |
0305bc27 QW |
9531 | "can't find block group for bytenr %llu", |
9532 | bytenr); | |
ba1bf481 | 9533 | return -1; |
0305bc27 | 9534 | } |
1a40e23b | 9535 | |
cdcb725c | 9536 | min_free = btrfs_block_group_used(&block_group->item); |
9537 | ||
ba1bf481 | 9538 | /* no bytes used, we're good */ |
cdcb725c | 9539 | if (!min_free) |
1a40e23b ZY |
9540 | goto out; |
9541 | ||
ba1bf481 JB |
9542 | space_info = block_group->space_info; |
9543 | spin_lock(&space_info->lock); | |
17d217fe | 9544 | |
ba1bf481 | 9545 | full = space_info->full; |
17d217fe | 9546 | |
ba1bf481 JB |
9547 | /* |
9548 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9549 | * relocate it unless we're able to allocate a new chunk below. |
9550 | * | |
9551 | * Otherwise, we need to make sure we have room in the space to handle | |
9552 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9553 | */ |
7ce618db | 9554 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9555 | (btrfs_space_info_used(space_info, false) + min_free < |
9556 | space_info->total_bytes)) { | |
ba1bf481 JB |
9557 | spin_unlock(&space_info->lock); |
9558 | goto out; | |
17d217fe | 9559 | } |
ba1bf481 | 9560 | spin_unlock(&space_info->lock); |
ea8c2819 | 9561 | |
ba1bf481 JB |
9562 | /* |
9563 | * ok we don't have enough space, but maybe we have free space on our | |
9564 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9565 | * alloc devices and guess if we have enough space. if this block |
9566 | * group is going to be restriped, run checks against the target | |
9567 | * profile instead of the current one. | |
ba1bf481 JB |
9568 | */ |
9569 | ret = -1; | |
ea8c2819 | 9570 | |
cdcb725c | 9571 | /* |
9572 | * index: | |
9573 | * 0: raid10 | |
9574 | * 1: raid1 | |
9575 | * 2: dup | |
9576 | * 3: raid0 | |
9577 | * 4: single | |
9578 | */ | |
0b246afa | 9579 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9580 | if (target) { |
31e50229 | 9581 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9582 | } else { |
9583 | /* | |
9584 | * this is just a balance, so if we were marked as full | |
9585 | * we know there is no space for a new chunk | |
9586 | */ | |
0305bc27 QW |
9587 | if (full) { |
9588 | if (debug) | |
0b246afa JM |
9589 | btrfs_warn(fs_info, |
9590 | "no space to alloc new chunk for block group %llu", | |
9591 | block_group->key.objectid); | |
4a5e98f5 | 9592 | goto out; |
0305bc27 | 9593 | } |
4a5e98f5 ID |
9594 | |
9595 | index = get_block_group_index(block_group); | |
9596 | } | |
9597 | ||
e6ec716f | 9598 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9599 | dev_min = 4; |
6719db6a JB |
9600 | /* Divide by 2 */ |
9601 | min_free >>= 1; | |
e6ec716f | 9602 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9603 | dev_min = 2; |
e6ec716f | 9604 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9605 | /* Multiply by 2 */ |
9606 | min_free <<= 1; | |
e6ec716f | 9607 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9608 | dev_min = fs_devices->rw_devices; |
47c5713f | 9609 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9610 | } |
9611 | ||
6df9a95e JB |
9612 | /* We need to do this so that we can look at pending chunks */ |
9613 | trans = btrfs_join_transaction(root); | |
9614 | if (IS_ERR(trans)) { | |
9615 | ret = PTR_ERR(trans); | |
9616 | goto out; | |
9617 | } | |
9618 | ||
0b246afa | 9619 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9620 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9621 | u64 dev_offset; |
56bec294 | 9622 | |
ba1bf481 JB |
9623 | /* |
9624 | * check to make sure we can actually find a chunk with enough | |
9625 | * space to fit our block group in. | |
9626 | */ | |
63a212ab SB |
9627 | if (device->total_bytes > device->bytes_used + min_free && |
9628 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 9629 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9630 | &dev_offset, NULL); |
ba1bf481 | 9631 | if (!ret) |
cdcb725c | 9632 | dev_nr++; |
9633 | ||
9634 | if (dev_nr >= dev_min) | |
73e48b27 | 9635 | break; |
cdcb725c | 9636 | |
ba1bf481 | 9637 | ret = -1; |
725c8463 | 9638 | } |
edbd8d4e | 9639 | } |
0305bc27 | 9640 | if (debug && ret == -1) |
0b246afa JM |
9641 | btrfs_warn(fs_info, |
9642 | "no space to allocate a new chunk for block group %llu", | |
9643 | block_group->key.objectid); | |
9644 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9645 | btrfs_end_transaction(trans); |
edbd8d4e | 9646 | out: |
ba1bf481 | 9647 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9648 | return ret; |
9649 | } | |
9650 | ||
6bccf3ab JM |
9651 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9652 | struct btrfs_path *path, | |
9653 | struct btrfs_key *key) | |
0b86a832 | 9654 | { |
6bccf3ab | 9655 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9656 | int ret = 0; |
0b86a832 CM |
9657 | struct btrfs_key found_key; |
9658 | struct extent_buffer *leaf; | |
9659 | int slot; | |
edbd8d4e | 9660 | |
0b86a832 CM |
9661 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9662 | if (ret < 0) | |
925baedd CM |
9663 | goto out; |
9664 | ||
d397712b | 9665 | while (1) { |
0b86a832 | 9666 | slot = path->slots[0]; |
edbd8d4e | 9667 | leaf = path->nodes[0]; |
0b86a832 CM |
9668 | if (slot >= btrfs_header_nritems(leaf)) { |
9669 | ret = btrfs_next_leaf(root, path); | |
9670 | if (ret == 0) | |
9671 | continue; | |
9672 | if (ret < 0) | |
925baedd | 9673 | goto out; |
0b86a832 | 9674 | break; |
edbd8d4e | 9675 | } |
0b86a832 | 9676 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9677 | |
0b86a832 | 9678 | if (found_key.objectid >= key->objectid && |
925baedd | 9679 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9680 | struct extent_map_tree *em_tree; |
9681 | struct extent_map *em; | |
9682 | ||
9683 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9684 | read_lock(&em_tree->lock); | |
9685 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9686 | found_key.offset); | |
9687 | read_unlock(&em_tree->lock); | |
9688 | if (!em) { | |
0b246afa | 9689 | btrfs_err(fs_info, |
6fb37b75 LB |
9690 | "logical %llu len %llu found bg but no related chunk", |
9691 | found_key.objectid, found_key.offset); | |
9692 | ret = -ENOENT; | |
9693 | } else { | |
9694 | ret = 0; | |
9695 | } | |
187ee58c | 9696 | free_extent_map(em); |
925baedd CM |
9697 | goto out; |
9698 | } | |
0b86a832 | 9699 | path->slots[0]++; |
edbd8d4e | 9700 | } |
925baedd | 9701 | out: |
0b86a832 | 9702 | return ret; |
edbd8d4e CM |
9703 | } |
9704 | ||
0af3d00b JB |
9705 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9706 | { | |
9707 | struct btrfs_block_group_cache *block_group; | |
9708 | u64 last = 0; | |
9709 | ||
9710 | while (1) { | |
9711 | struct inode *inode; | |
9712 | ||
9713 | block_group = btrfs_lookup_first_block_group(info, last); | |
9714 | while (block_group) { | |
9715 | spin_lock(&block_group->lock); | |
9716 | if (block_group->iref) | |
9717 | break; | |
9718 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9719 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9720 | } |
9721 | if (!block_group) { | |
9722 | if (last == 0) | |
9723 | break; | |
9724 | last = 0; | |
9725 | continue; | |
9726 | } | |
9727 | ||
9728 | inode = block_group->inode; | |
9729 | block_group->iref = 0; | |
9730 | block_group->inode = NULL; | |
9731 | spin_unlock(&block_group->lock); | |
f3bca802 | 9732 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9733 | iput(inode); |
9734 | last = block_group->key.objectid + block_group->key.offset; | |
9735 | btrfs_put_block_group(block_group); | |
9736 | } | |
9737 | } | |
9738 | ||
5cdd7db6 FM |
9739 | /* |
9740 | * Must be called only after stopping all workers, since we could have block | |
9741 | * group caching kthreads running, and therefore they could race with us if we | |
9742 | * freed the block groups before stopping them. | |
9743 | */ | |
1a40e23b ZY |
9744 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9745 | { | |
9746 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9747 | struct btrfs_space_info *space_info; |
11833d66 | 9748 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9749 | struct rb_node *n; |
9750 | ||
9e351cc8 | 9751 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9752 | while (!list_empty(&info->caching_block_groups)) { |
9753 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9754 | struct btrfs_caching_control, list); | |
9755 | list_del(&caching_ctl->list); | |
9756 | put_caching_control(caching_ctl); | |
9757 | } | |
9e351cc8 | 9758 | up_write(&info->commit_root_sem); |
11833d66 | 9759 | |
47ab2a6c JB |
9760 | spin_lock(&info->unused_bgs_lock); |
9761 | while (!list_empty(&info->unused_bgs)) { | |
9762 | block_group = list_first_entry(&info->unused_bgs, | |
9763 | struct btrfs_block_group_cache, | |
9764 | bg_list); | |
9765 | list_del_init(&block_group->bg_list); | |
9766 | btrfs_put_block_group(block_group); | |
9767 | } | |
9768 | spin_unlock(&info->unused_bgs_lock); | |
9769 | ||
1a40e23b ZY |
9770 | spin_lock(&info->block_group_cache_lock); |
9771 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9772 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9773 | cache_node); | |
1a40e23b ZY |
9774 | rb_erase(&block_group->cache_node, |
9775 | &info->block_group_cache_tree); | |
01eacb27 | 9776 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9777 | spin_unlock(&info->block_group_cache_lock); |
9778 | ||
80eb234a | 9779 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9780 | list_del(&block_group->list); |
80eb234a | 9781 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9782 | |
3c14874a JB |
9783 | /* |
9784 | * We haven't cached this block group, which means we could | |
9785 | * possibly have excluded extents on this block group. | |
9786 | */ | |
36cce922 JB |
9787 | if (block_group->cached == BTRFS_CACHE_NO || |
9788 | block_group->cached == BTRFS_CACHE_ERROR) | |
2ff7e61e | 9789 | free_excluded_extents(info, block_group); |
3c14874a | 9790 | |
817d52f8 | 9791 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9792 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9793 | ASSERT(list_empty(&block_group->dirty_list)); |
9794 | ASSERT(list_empty(&block_group->io_list)); | |
9795 | ASSERT(list_empty(&block_group->bg_list)); | |
9796 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9797 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9798 | |
9799 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9800 | } |
9801 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9802 | |
9803 | /* now that all the block groups are freed, go through and | |
9804 | * free all the space_info structs. This is only called during | |
9805 | * the final stages of unmount, and so we know nobody is | |
9806 | * using them. We call synchronize_rcu() once before we start, | |
9807 | * just to be on the safe side. | |
9808 | */ | |
9809 | synchronize_rcu(); | |
9810 | ||
8929ecfa YZ |
9811 | release_global_block_rsv(info); |
9812 | ||
67871254 | 9813 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9814 | int i; |
9815 | ||
4184ea7f CM |
9816 | space_info = list_entry(info->space_info.next, |
9817 | struct btrfs_space_info, | |
9818 | list); | |
d555b6c3 JB |
9819 | |
9820 | /* | |
9821 | * Do not hide this behind enospc_debug, this is actually | |
9822 | * important and indicates a real bug if this happens. | |
9823 | */ | |
9824 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9825 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9826 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9827 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9828 | list_del(&space_info->list); |
6ab0a202 JM |
9829 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9830 | struct kobject *kobj; | |
c1895442 JM |
9831 | kobj = space_info->block_group_kobjs[i]; |
9832 | space_info->block_group_kobjs[i] = NULL; | |
9833 | if (kobj) { | |
6ab0a202 JM |
9834 | kobject_del(kobj); |
9835 | kobject_put(kobj); | |
9836 | } | |
9837 | } | |
9838 | kobject_del(&space_info->kobj); | |
9839 | kobject_put(&space_info->kobj); | |
4184ea7f | 9840 | } |
1a40e23b ZY |
9841 | return 0; |
9842 | } | |
9843 | ||
b742bb82 YZ |
9844 | static void __link_block_group(struct btrfs_space_info *space_info, |
9845 | struct btrfs_block_group_cache *cache) | |
9846 | { | |
9847 | int index = get_block_group_index(cache); | |
ed55b6ac | 9848 | bool first = false; |
b742bb82 YZ |
9849 | |
9850 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9851 | if (list_empty(&space_info->block_groups[index])) |
9852 | first = true; | |
9853 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9854 | up_write(&space_info->groups_sem); | |
9855 | ||
9856 | if (first) { | |
c1895442 | 9857 | struct raid_kobject *rkobj; |
6ab0a202 JM |
9858 | int ret; |
9859 | ||
c1895442 JM |
9860 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9861 | if (!rkobj) | |
9862 | goto out_err; | |
9863 | rkobj->raid_type = index; | |
9864 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9865 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9866 | "%s", get_raid_name(index)); | |
6ab0a202 | 9867 | if (ret) { |
c1895442 JM |
9868 | kobject_put(&rkobj->kobj); |
9869 | goto out_err; | |
6ab0a202 | 9870 | } |
c1895442 | 9871 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9872 | } |
c1895442 JM |
9873 | |
9874 | return; | |
9875 | out_err: | |
ab8d0fc4 JM |
9876 | btrfs_warn(cache->fs_info, |
9877 | "failed to add kobject for block cache, ignoring"); | |
b742bb82 YZ |
9878 | } |
9879 | ||
920e4a58 | 9880 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
9881 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
9882 | u64 start, u64 size) | |
920e4a58 MX |
9883 | { |
9884 | struct btrfs_block_group_cache *cache; | |
9885 | ||
9886 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9887 | if (!cache) | |
9888 | return NULL; | |
9889 | ||
9890 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9891 | GFP_NOFS); | |
9892 | if (!cache->free_space_ctl) { | |
9893 | kfree(cache); | |
9894 | return NULL; | |
9895 | } | |
9896 | ||
9897 | cache->key.objectid = start; | |
9898 | cache->key.offset = size; | |
9899 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9900 | ||
0b246afa JM |
9901 | cache->sectorsize = fs_info->sectorsize; |
9902 | cache->fs_info = fs_info; | |
2ff7e61e JM |
9903 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, |
9904 | &fs_info->mapping_tree, | |
9905 | start); | |
1e144fb8 OS |
9906 | set_free_space_tree_thresholds(cache); |
9907 | ||
920e4a58 MX |
9908 | atomic_set(&cache->count, 1); |
9909 | spin_lock_init(&cache->lock); | |
e570fd27 | 9910 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
9911 | INIT_LIST_HEAD(&cache->list); |
9912 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 9913 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 9914 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 9915 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 9916 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 9917 | btrfs_init_free_space_ctl(cache); |
04216820 | 9918 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 9919 | mutex_init(&cache->free_space_lock); |
920e4a58 MX |
9920 | |
9921 | return cache; | |
9922 | } | |
9923 | ||
5b4aacef | 9924 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
9925 | { |
9926 | struct btrfs_path *path; | |
9927 | int ret; | |
9078a3e1 | 9928 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 9929 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
9930 | struct btrfs_key key; |
9931 | struct btrfs_key found_key; | |
5f39d397 | 9932 | struct extent_buffer *leaf; |
0af3d00b JB |
9933 | int need_clear = 0; |
9934 | u64 cache_gen; | |
49303381 LB |
9935 | u64 feature; |
9936 | int mixed; | |
9937 | ||
9938 | feature = btrfs_super_incompat_flags(info->super_copy); | |
9939 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 9940 | |
9078a3e1 | 9941 | key.objectid = 0; |
0b86a832 | 9942 | key.offset = 0; |
962a298f | 9943 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
9944 | path = btrfs_alloc_path(); |
9945 | if (!path) | |
9946 | return -ENOMEM; | |
e4058b54 | 9947 | path->reada = READA_FORWARD; |
9078a3e1 | 9948 | |
0b246afa JM |
9949 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
9950 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
9951 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 9952 | need_clear = 1; |
0b246afa | 9953 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 9954 | need_clear = 1; |
0af3d00b | 9955 | |
d397712b | 9956 | while (1) { |
6bccf3ab | 9957 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
9958 | if (ret > 0) |
9959 | break; | |
0b86a832 CM |
9960 | if (ret != 0) |
9961 | goto error; | |
920e4a58 | 9962 | |
5f39d397 CM |
9963 | leaf = path->nodes[0]; |
9964 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 9965 | |
2ff7e61e | 9966 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 9967 | found_key.offset); |
9078a3e1 | 9968 | if (!cache) { |
0b86a832 | 9969 | ret = -ENOMEM; |
f0486c68 | 9970 | goto error; |
9078a3e1 | 9971 | } |
96303081 | 9972 | |
cf7c1ef6 LB |
9973 | if (need_clear) { |
9974 | /* | |
9975 | * When we mount with old space cache, we need to | |
9976 | * set BTRFS_DC_CLEAR and set dirty flag. | |
9977 | * | |
9978 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
9979 | * truncate the old free space cache inode and | |
9980 | * setup a new one. | |
9981 | * b) Setting 'dirty flag' makes sure that we flush | |
9982 | * the new space cache info onto disk. | |
9983 | */ | |
0b246afa | 9984 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 9985 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 9986 | } |
0af3d00b | 9987 | |
5f39d397 CM |
9988 | read_extent_buffer(leaf, &cache->item, |
9989 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
9990 | sizeof(cache->item)); | |
920e4a58 | 9991 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
9992 | if (!mixed && |
9993 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
9994 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
9995 | btrfs_err(info, | |
9996 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
9997 | cache->key.objectid); | |
9998 | ret = -EINVAL; | |
9999 | goto error; | |
10000 | } | |
0b86a832 | 10001 | |
9078a3e1 | 10002 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10003 | btrfs_release_path(path); |
34d52cb6 | 10004 | |
3c14874a JB |
10005 | /* |
10006 | * We need to exclude the super stripes now so that the space | |
10007 | * info has super bytes accounted for, otherwise we'll think | |
10008 | * we have more space than we actually do. | |
10009 | */ | |
2ff7e61e | 10010 | ret = exclude_super_stripes(info, cache); |
835d974f JB |
10011 | if (ret) { |
10012 | /* | |
10013 | * We may have excluded something, so call this just in | |
10014 | * case. | |
10015 | */ | |
2ff7e61e | 10016 | free_excluded_extents(info, cache); |
920e4a58 | 10017 | btrfs_put_block_group(cache); |
835d974f JB |
10018 | goto error; |
10019 | } | |
3c14874a | 10020 | |
817d52f8 JB |
10021 | /* |
10022 | * check for two cases, either we are full, and therefore | |
10023 | * don't need to bother with the caching work since we won't | |
10024 | * find any space, or we are empty, and we can just add all | |
10025 | * the space in and be done with it. This saves us _alot_ of | |
10026 | * time, particularly in the full case. | |
10027 | */ | |
10028 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10029 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10030 | cache->cached = BTRFS_CACHE_FINISHED; |
2ff7e61e | 10031 | free_excluded_extents(info, cache); |
817d52f8 | 10032 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10033 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10034 | cache->cached = BTRFS_CACHE_FINISHED; |
0b246afa | 10035 | add_new_free_space(cache, info, |
817d52f8 JB |
10036 | found_key.objectid, |
10037 | found_key.objectid + | |
10038 | found_key.offset); | |
2ff7e61e | 10039 | free_excluded_extents(info, cache); |
817d52f8 | 10040 | } |
96b5179d | 10041 | |
0b246afa | 10042 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10043 | if (ret) { |
10044 | btrfs_remove_free_space_cache(cache); | |
10045 | btrfs_put_block_group(cache); | |
10046 | goto error; | |
10047 | } | |
10048 | ||
0b246afa | 10049 | trace_btrfs_add_block_group(info, cache, 0); |
6324fbf3 CM |
10050 | ret = update_space_info(info, cache->flags, found_key.offset, |
10051 | btrfs_block_group_used(&cache->item), | |
e40edf2d | 10052 | cache->bytes_super, &space_info); |
8c579fe7 JB |
10053 | if (ret) { |
10054 | btrfs_remove_free_space_cache(cache); | |
10055 | spin_lock(&info->block_group_cache_lock); | |
10056 | rb_erase(&cache->cache_node, | |
10057 | &info->block_group_cache_tree); | |
01eacb27 | 10058 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
10059 | spin_unlock(&info->block_group_cache_lock); |
10060 | btrfs_put_block_group(cache); | |
10061 | goto error; | |
10062 | } | |
10063 | ||
6324fbf3 | 10064 | cache->space_info = space_info; |
1b2da372 | 10065 | |
b742bb82 | 10066 | __link_block_group(space_info, cache); |
0f9dd46c | 10067 | |
0b246afa | 10068 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10069 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10070 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10071 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10072 | spin_lock(&info->unused_bgs_lock); | |
10073 | /* Should always be true but just in case. */ | |
10074 | if (list_empty(&cache->bg_list)) { | |
10075 | btrfs_get_block_group(cache); | |
10076 | list_add_tail(&cache->bg_list, | |
10077 | &info->unused_bgs); | |
10078 | } | |
10079 | spin_unlock(&info->unused_bgs_lock); | |
10080 | } | |
9078a3e1 | 10081 | } |
b742bb82 | 10082 | |
0b246afa | 10083 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10084 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10085 | (BTRFS_BLOCK_GROUP_RAID10 | |
10086 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10087 | BTRFS_BLOCK_GROUP_RAID5 | |
10088 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10089 | BTRFS_BLOCK_GROUP_DUP))) |
10090 | continue; | |
10091 | /* | |
10092 | * avoid allocating from un-mirrored block group if there are | |
10093 | * mirrored block groups. | |
10094 | */ | |
1095cc0d | 10095 | list_for_each_entry(cache, |
10096 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10097 | list) | |
868f401a | 10098 | inc_block_group_ro(cache, 1); |
1095cc0d | 10099 | list_for_each_entry(cache, |
10100 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10101 | list) | |
868f401a | 10102 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10103 | } |
f0486c68 YZ |
10104 | |
10105 | init_global_block_rsv(info); | |
0b86a832 CM |
10106 | ret = 0; |
10107 | error: | |
9078a3e1 | 10108 | btrfs_free_path(path); |
0b86a832 | 10109 | return ret; |
9078a3e1 | 10110 | } |
6324fbf3 | 10111 | |
ea658bad | 10112 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
2ff7e61e | 10113 | struct btrfs_fs_info *fs_info) |
ea658bad JB |
10114 | { |
10115 | struct btrfs_block_group_cache *block_group, *tmp; | |
0b246afa | 10116 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10117 | struct btrfs_block_group_item item; |
10118 | struct btrfs_key key; | |
10119 | int ret = 0; | |
d9a0540a | 10120 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10121 | |
d9a0540a | 10122 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10123 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10124 | if (ret) |
c92f6be3 | 10125 | goto next; |
ea658bad JB |
10126 | |
10127 | spin_lock(&block_group->lock); | |
10128 | memcpy(&item, &block_group->item, sizeof(item)); | |
10129 | memcpy(&key, &block_group->key, sizeof(key)); | |
10130 | spin_unlock(&block_group->lock); | |
10131 | ||
10132 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10133 | sizeof(item)); | |
10134 | if (ret) | |
66642832 | 10135 | btrfs_abort_transaction(trans, ret); |
0b246afa JM |
10136 | ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid, |
10137 | key.offset); | |
6df9a95e | 10138 | if (ret) |
66642832 | 10139 | btrfs_abort_transaction(trans, ret); |
0b246afa | 10140 | add_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 | 10141 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10142 | next: |
10143 | list_del_init(&block_group->bg_list); | |
ea658bad | 10144 | } |
d9a0540a | 10145 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10146 | } |
10147 | ||
6324fbf3 | 10148 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 10149 | struct btrfs_fs_info *fs_info, u64 bytes_used, |
e17cade2 | 10150 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
10151 | u64 size) |
10152 | { | |
6324fbf3 | 10153 | struct btrfs_block_group_cache *cache; |
0b246afa | 10154 | int ret; |
6324fbf3 | 10155 | |
0b246afa | 10156 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10157 | |
2ff7e61e | 10158 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10159 | if (!cache) |
10160 | return -ENOMEM; | |
34d52cb6 | 10161 | |
6324fbf3 | 10162 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 | 10163 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
6324fbf3 CM |
10164 | btrfs_set_block_group_flags(&cache->item, type); |
10165 | ||
920e4a58 | 10166 | cache->flags = type; |
11833d66 | 10167 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10168 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10169 | cache->needs_free_space = 1; |
2ff7e61e | 10170 | ret = exclude_super_stripes(fs_info, cache); |
835d974f JB |
10171 | if (ret) { |
10172 | /* | |
10173 | * We may have excluded something, so call this just in | |
10174 | * case. | |
10175 | */ | |
2ff7e61e | 10176 | free_excluded_extents(fs_info, cache); |
920e4a58 | 10177 | btrfs_put_block_group(cache); |
835d974f JB |
10178 | return ret; |
10179 | } | |
96303081 | 10180 | |
0b246afa | 10181 | add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size); |
817d52f8 | 10182 | |
2ff7e61e | 10183 | free_excluded_extents(fs_info, cache); |
11833d66 | 10184 | |
d0bd4560 | 10185 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10186 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10187 | u64 new_bytes_used = size - bytes_used; |
10188 | ||
10189 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10190 | fragment_free_space(cache); |
d0bd4560 JB |
10191 | } |
10192 | #endif | |
2e6e5183 FM |
10193 | /* |
10194 | * Call to ensure the corresponding space_info object is created and | |
10195 | * assigned to our block group, but don't update its counters just yet. | |
10196 | * We want our bg to be added to the rbtree with its ->space_info set. | |
10197 | */ | |
0b246afa | 10198 | ret = update_space_info(fs_info, cache->flags, 0, 0, 0, |
2e6e5183 FM |
10199 | &cache->space_info); |
10200 | if (ret) { | |
10201 | btrfs_remove_free_space_cache(cache); | |
10202 | btrfs_put_block_group(cache); | |
10203 | return ret; | |
10204 | } | |
10205 | ||
0b246afa | 10206 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10207 | if (ret) { |
10208 | btrfs_remove_free_space_cache(cache); | |
10209 | btrfs_put_block_group(cache); | |
10210 | return ret; | |
10211 | } | |
10212 | ||
2e6e5183 FM |
10213 | /* |
10214 | * Now that our block group has its ->space_info set and is inserted in | |
10215 | * the rbtree, update the space info's counters. | |
10216 | */ | |
0b246afa JM |
10217 | trace_btrfs_add_block_group(fs_info, cache, 1); |
10218 | ret = update_space_info(fs_info, cache->flags, size, bytes_used, | |
e40edf2d | 10219 | cache->bytes_super, &cache->space_info); |
8c579fe7 JB |
10220 | if (ret) { |
10221 | btrfs_remove_free_space_cache(cache); | |
0b246afa | 10222 | spin_lock(&fs_info->block_group_cache_lock); |
8c579fe7 | 10223 | rb_erase(&cache->cache_node, |
0b246afa | 10224 | &fs_info->block_group_cache_tree); |
01eacb27 | 10225 | RB_CLEAR_NODE(&cache->cache_node); |
0b246afa | 10226 | spin_unlock(&fs_info->block_group_cache_lock); |
8c579fe7 JB |
10227 | btrfs_put_block_group(cache); |
10228 | return ret; | |
10229 | } | |
0b246afa | 10230 | update_global_block_rsv(fs_info); |
1b2da372 | 10231 | |
b742bb82 | 10232 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 10233 | |
47ab2a6c | 10234 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10235 | |
0b246afa | 10236 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10237 | return 0; |
10238 | } | |
1a40e23b | 10239 | |
10ea00f5 ID |
10240 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10241 | { | |
899c81ea ID |
10242 | u64 extra_flags = chunk_to_extended(flags) & |
10243 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10244 | |
de98ced9 | 10245 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10246 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10247 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10248 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10249 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10250 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10251 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10252 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10253 | } |
10254 | ||
1a40e23b | 10255 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
6bccf3ab | 10256 | struct btrfs_fs_info *fs_info, u64 group_start, |
04216820 | 10257 | struct extent_map *em) |
1a40e23b | 10258 | { |
6bccf3ab | 10259 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10260 | struct btrfs_path *path; |
10261 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10262 | struct btrfs_free_cluster *cluster; |
0b246afa | 10263 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10264 | struct btrfs_key key; |
0af3d00b | 10265 | struct inode *inode; |
c1895442 | 10266 | struct kobject *kobj = NULL; |
1a40e23b | 10267 | int ret; |
10ea00f5 | 10268 | int index; |
89a55897 | 10269 | int factor; |
4f69cb98 | 10270 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10271 | bool remove_em; |
1a40e23b | 10272 | |
6bccf3ab | 10273 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10274 | BUG_ON(!block_group); |
c146afad | 10275 | BUG_ON(!block_group->ro); |
1a40e23b | 10276 | |
9f7c43c9 | 10277 | /* |
10278 | * Free the reserved super bytes from this block group before | |
10279 | * remove it. | |
10280 | */ | |
2ff7e61e | 10281 | free_excluded_extents(fs_info, block_group); |
9f7c43c9 | 10282 | |
1a40e23b | 10283 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 10284 | index = get_block_group_index(block_group); |
89a55897 JB |
10285 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10286 | BTRFS_BLOCK_GROUP_RAID1 | | |
10287 | BTRFS_BLOCK_GROUP_RAID10)) | |
10288 | factor = 2; | |
10289 | else | |
10290 | factor = 1; | |
1a40e23b | 10291 | |
44fb5511 | 10292 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10293 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10294 | spin_lock(&cluster->refill_lock); |
10295 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10296 | spin_unlock(&cluster->refill_lock); | |
10297 | ||
10298 | /* | |
10299 | * make sure this block group isn't part of a metadata | |
10300 | * allocation cluster | |
10301 | */ | |
0b246afa | 10302 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10303 | spin_lock(&cluster->refill_lock); |
10304 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10305 | spin_unlock(&cluster->refill_lock); | |
10306 | ||
1a40e23b | 10307 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10308 | if (!path) { |
10309 | ret = -ENOMEM; | |
10310 | goto out; | |
10311 | } | |
1a40e23b | 10312 | |
1bbc621e CM |
10313 | /* |
10314 | * get the inode first so any iput calls done for the io_list | |
10315 | * aren't the final iput (no unlinks allowed now) | |
10316 | */ | |
77ab86bf | 10317 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10318 | |
10319 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10320 | /* | |
10321 | * make sure our free spache cache IO is done before remove the | |
10322 | * free space inode | |
10323 | */ | |
10324 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10325 | if (!list_empty(&block_group->io_list)) { | |
10326 | list_del_init(&block_group->io_list); | |
10327 | ||
10328 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10329 | ||
10330 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10331 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10332 | btrfs_put_block_group(block_group); |
10333 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10334 | } | |
10335 | ||
10336 | if (!list_empty(&block_group->dirty_list)) { | |
10337 | list_del_init(&block_group->dirty_list); | |
10338 | btrfs_put_block_group(block_group); | |
10339 | } | |
10340 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10341 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10342 | ||
0af3d00b | 10343 | if (!IS_ERR(inode)) { |
73f2e545 | 10344 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10345 | if (ret) { |
10346 | btrfs_add_delayed_iput(inode); | |
10347 | goto out; | |
10348 | } | |
0af3d00b JB |
10349 | clear_nlink(inode); |
10350 | /* One for the block groups ref */ | |
10351 | spin_lock(&block_group->lock); | |
10352 | if (block_group->iref) { | |
10353 | block_group->iref = 0; | |
10354 | block_group->inode = NULL; | |
10355 | spin_unlock(&block_group->lock); | |
10356 | iput(inode); | |
10357 | } else { | |
10358 | spin_unlock(&block_group->lock); | |
10359 | } | |
10360 | /* One for our lookup ref */ | |
455757c3 | 10361 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10362 | } |
10363 | ||
10364 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10365 | key.offset = block_group->key.objectid; | |
10366 | key.type = 0; | |
10367 | ||
10368 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10369 | if (ret < 0) | |
10370 | goto out; | |
10371 | if (ret > 0) | |
b3b4aa74 | 10372 | btrfs_release_path(path); |
0af3d00b JB |
10373 | if (ret == 0) { |
10374 | ret = btrfs_del_item(trans, tree_root, path); | |
10375 | if (ret) | |
10376 | goto out; | |
b3b4aa74 | 10377 | btrfs_release_path(path); |
0af3d00b JB |
10378 | } |
10379 | ||
0b246afa | 10380 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10381 | rb_erase(&block_group->cache_node, |
0b246afa | 10382 | &fs_info->block_group_cache_tree); |
292cbd51 | 10383 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10384 | |
0b246afa JM |
10385 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10386 | fs_info->first_logical_byte = (u64)-1; | |
10387 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10388 | |
80eb234a | 10389 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10390 | /* |
10391 | * we must use list_del_init so people can check to see if they | |
10392 | * are still on the list after taking the semaphore | |
10393 | */ | |
10394 | list_del_init(&block_group->list); | |
6ab0a202 | 10395 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10396 | kobj = block_group->space_info->block_group_kobjs[index]; |
10397 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10398 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10399 | } |
80eb234a | 10400 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10401 | if (kobj) { |
10402 | kobject_del(kobj); | |
10403 | kobject_put(kobj); | |
10404 | } | |
1a40e23b | 10405 | |
4f69cb98 FM |
10406 | if (block_group->has_caching_ctl) |
10407 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10408 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10409 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10410 | if (block_group->has_caching_ctl) { |
0b246afa | 10411 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10412 | if (!caching_ctl) { |
10413 | struct btrfs_caching_control *ctl; | |
10414 | ||
10415 | list_for_each_entry(ctl, | |
0b246afa | 10416 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10417 | if (ctl->block_group == block_group) { |
10418 | caching_ctl = ctl; | |
10419 | atomic_inc(&caching_ctl->count); | |
10420 | break; | |
10421 | } | |
10422 | } | |
10423 | if (caching_ctl) | |
10424 | list_del_init(&caching_ctl->list); | |
0b246afa | 10425 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10426 | if (caching_ctl) { |
10427 | /* Once for the caching bgs list and once for us. */ | |
10428 | put_caching_control(caching_ctl); | |
10429 | put_caching_control(caching_ctl); | |
10430 | } | |
10431 | } | |
817d52f8 | 10432 | |
ce93ec54 JB |
10433 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10434 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10435 | WARN_ON(1); |
10436 | } | |
10437 | if (!list_empty(&block_group->io_list)) { | |
10438 | WARN_ON(1); | |
ce93ec54 JB |
10439 | } |
10440 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10441 | btrfs_remove_free_space_cache(block_group); |
10442 | ||
c146afad | 10443 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10444 | list_del_init(&block_group->ro_list); |
18d018ad | 10445 | |
0b246afa | 10446 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10447 | WARN_ON(block_group->space_info->total_bytes |
10448 | < block_group->key.offset); | |
10449 | WARN_ON(block_group->space_info->bytes_readonly | |
10450 | < block_group->key.offset); | |
10451 | WARN_ON(block_group->space_info->disk_total | |
10452 | < block_group->key.offset * factor); | |
10453 | } | |
c146afad YZ |
10454 | block_group->space_info->total_bytes -= block_group->key.offset; |
10455 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10456 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10457 | |
c146afad | 10458 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10459 | |
0af3d00b JB |
10460 | memcpy(&key, &block_group->key, sizeof(key)); |
10461 | ||
34441361 | 10462 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10463 | if (!list_empty(&em->list)) { |
10464 | /* We're in the transaction->pending_chunks list. */ | |
10465 | free_extent_map(em); | |
10466 | } | |
04216820 FM |
10467 | spin_lock(&block_group->lock); |
10468 | block_group->removed = 1; | |
10469 | /* | |
10470 | * At this point trimming can't start on this block group, because we | |
10471 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10472 | * so no one can't find it anymore and even if someone already got this | |
10473 | * block group before we removed it from the rbtree, they have already | |
10474 | * incremented block_group->trimming - if they didn't, they won't find | |
10475 | * any free space entries because we already removed them all when we | |
10476 | * called btrfs_remove_free_space_cache(). | |
10477 | * | |
10478 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10479 | * to prevent the same logical address range and physical device space | |
10480 | * ranges from being reused for a new block group. This is because our | |
10481 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10482 | * completely transactionless, so while it is trimming a range the | |
10483 | * currently running transaction might finish and a new one start, | |
10484 | * allowing for new block groups to be created that can reuse the same | |
10485 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10486 | * |
10487 | * There may also be an implicit trim operation if the file system | |
10488 | * is mounted with -odiscard. The same protections must remain | |
10489 | * in place until the extents have been discarded completely when | |
10490 | * the transaction commit has completed. | |
04216820 FM |
10491 | */ |
10492 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10493 | /* | |
10494 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10495 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10496 | * before checking block_group->removed). | |
10497 | */ | |
10498 | if (!remove_em) { | |
10499 | /* | |
10500 | * Our em might be in trans->transaction->pending_chunks which | |
10501 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10502 | * and so is the fs_info->pinned_chunks list. | |
10503 | * | |
10504 | * So at this point we must be holding the chunk_mutex to avoid | |
10505 | * any races with chunk allocation (more specifically at | |
10506 | * volumes.c:contains_pending_extent()), to ensure it always | |
10507 | * sees the em, either in the pending_chunks list or in the | |
10508 | * pinned_chunks list. | |
10509 | */ | |
0b246afa | 10510 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10511 | } |
10512 | spin_unlock(&block_group->lock); | |
04216820 FM |
10513 | |
10514 | if (remove_em) { | |
10515 | struct extent_map_tree *em_tree; | |
10516 | ||
0b246afa | 10517 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10518 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10519 | /* |
10520 | * The em might be in the pending_chunks list, so make sure the | |
10521 | * chunk mutex is locked, since remove_extent_mapping() will | |
10522 | * delete us from that list. | |
10523 | */ | |
04216820 FM |
10524 | remove_extent_mapping(em_tree, em); |
10525 | write_unlock(&em_tree->lock); | |
10526 | /* once for the tree */ | |
10527 | free_extent_map(em); | |
10528 | } | |
10529 | ||
34441361 | 10530 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10531 | |
0b246afa | 10532 | ret = remove_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 OS |
10533 | if (ret) |
10534 | goto out; | |
10535 | ||
fa9c0d79 CM |
10536 | btrfs_put_block_group(block_group); |
10537 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10538 | |
10539 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10540 | if (ret > 0) | |
10541 | ret = -EIO; | |
10542 | if (ret < 0) | |
10543 | goto out; | |
10544 | ||
10545 | ret = btrfs_del_item(trans, root, path); | |
10546 | out: | |
10547 | btrfs_free_path(path); | |
10548 | return ret; | |
10549 | } | |
acce952b | 10550 | |
8eab77ff | 10551 | struct btrfs_trans_handle * |
7fd01182 FM |
10552 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10553 | const u64 chunk_offset) | |
8eab77ff | 10554 | { |
7fd01182 FM |
10555 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10556 | struct extent_map *em; | |
10557 | struct map_lookup *map; | |
10558 | unsigned int num_items; | |
10559 | ||
10560 | read_lock(&em_tree->lock); | |
10561 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10562 | read_unlock(&em_tree->lock); | |
10563 | ASSERT(em && em->start == chunk_offset); | |
10564 | ||
8eab77ff | 10565 | /* |
7fd01182 FM |
10566 | * We need to reserve 3 + N units from the metadata space info in order |
10567 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10568 | * btrfs_remove_block_group()), which are used for: | |
10569 | * | |
8eab77ff FM |
10570 | * 1 unit for adding the free space inode's orphan (located in the tree |
10571 | * of tree roots). | |
7fd01182 FM |
10572 | * 1 unit for deleting the block group item (located in the extent |
10573 | * tree). | |
10574 | * 1 unit for deleting the free space item (located in tree of tree | |
10575 | * roots). | |
10576 | * N units for deleting N device extent items corresponding to each | |
10577 | * stripe (located in the device tree). | |
10578 | * | |
10579 | * In order to remove a block group we also need to reserve units in the | |
10580 | * system space info in order to update the chunk tree (update one or | |
10581 | * more device items and remove one chunk item), but this is done at | |
10582 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10583 | */ |
95617d69 | 10584 | map = em->map_lookup; |
7fd01182 FM |
10585 | num_items = 3 + map->num_stripes; |
10586 | free_extent_map(em); | |
10587 | ||
8eab77ff | 10588 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10589 | num_items, 1); |
8eab77ff FM |
10590 | } |
10591 | ||
47ab2a6c JB |
10592 | /* |
10593 | * Process the unused_bgs list and remove any that don't have any allocated | |
10594 | * space inside of them. | |
10595 | */ | |
10596 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10597 | { | |
10598 | struct btrfs_block_group_cache *block_group; | |
10599 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10600 | struct btrfs_trans_handle *trans; |
10601 | int ret = 0; | |
10602 | ||
afcdd129 | 10603 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10604 | return; |
10605 | ||
10606 | spin_lock(&fs_info->unused_bgs_lock); | |
10607 | while (!list_empty(&fs_info->unused_bgs)) { | |
10608 | u64 start, end; | |
e33e17ee | 10609 | int trimming; |
47ab2a6c JB |
10610 | |
10611 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10612 | struct btrfs_block_group_cache, | |
10613 | bg_list); | |
47ab2a6c | 10614 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10615 | |
10616 | space_info = block_group->space_info; | |
10617 | ||
47ab2a6c JB |
10618 | if (ret || btrfs_mixed_space_info(space_info)) { |
10619 | btrfs_put_block_group(block_group); | |
10620 | continue; | |
10621 | } | |
10622 | spin_unlock(&fs_info->unused_bgs_lock); | |
10623 | ||
d5f2e33b | 10624 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10625 | |
47ab2a6c JB |
10626 | /* Don't want to race with allocators so take the groups_sem */ |
10627 | down_write(&space_info->groups_sem); | |
10628 | spin_lock(&block_group->lock); | |
10629 | if (block_group->reserved || | |
10630 | btrfs_block_group_used(&block_group->item) || | |
19c4d2f9 | 10631 | block_group->ro || |
aefbe9a6 | 10632 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10633 | /* |
10634 | * We want to bail if we made new allocations or have | |
10635 | * outstanding allocations in this block group. We do | |
10636 | * the ro check in case balance is currently acting on | |
10637 | * this block group. | |
10638 | */ | |
10639 | spin_unlock(&block_group->lock); | |
10640 | up_write(&space_info->groups_sem); | |
10641 | goto next; | |
10642 | } | |
10643 | spin_unlock(&block_group->lock); | |
10644 | ||
10645 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10646 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10647 | up_write(&space_info->groups_sem); |
10648 | if (ret < 0) { | |
10649 | ret = 0; | |
10650 | goto next; | |
10651 | } | |
10652 | ||
10653 | /* | |
10654 | * Want to do this before we do anything else so we can recover | |
10655 | * properly if we fail to join the transaction. | |
10656 | */ | |
7fd01182 FM |
10657 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10658 | block_group->key.objectid); | |
47ab2a6c | 10659 | if (IS_ERR(trans)) { |
2ff7e61e | 10660 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10661 | ret = PTR_ERR(trans); |
10662 | goto next; | |
10663 | } | |
10664 | ||
10665 | /* | |
10666 | * We could have pending pinned extents for this block group, | |
10667 | * just delete them, we don't care about them anymore. | |
10668 | */ | |
10669 | start = block_group->key.objectid; | |
10670 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10671 | /* |
10672 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10673 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10674 | * another task might be running finish_extent_commit() for the | |
10675 | * previous transaction N - 1, and have seen a range belonging | |
10676 | * to the block group in freed_extents[] before we were able to | |
10677 | * clear the whole block group range from freed_extents[]. This | |
10678 | * means that task can lookup for the block group after we | |
10679 | * unpinned it from freed_extents[] and removed it, leading to | |
10680 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10681 | */ | |
10682 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10683 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10684 | EXTENT_DIRTY); |
758eb51e | 10685 | if (ret) { |
d4b450cd | 10686 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10687 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10688 | goto end_trans; |
10689 | } | |
10690 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10691 | EXTENT_DIRTY); |
758eb51e | 10692 | if (ret) { |
d4b450cd | 10693 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10694 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10695 | goto end_trans; |
10696 | } | |
d4b450cd | 10697 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10698 | |
10699 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10700 | spin_lock(&space_info->lock); |
10701 | spin_lock(&block_group->lock); | |
10702 | ||
10703 | space_info->bytes_pinned -= block_group->pinned; | |
10704 | space_info->bytes_readonly += block_group->pinned; | |
10705 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10706 | -block_group->pinned); | |
47ab2a6c JB |
10707 | block_group->pinned = 0; |
10708 | ||
c30666d4 ZL |
10709 | spin_unlock(&block_group->lock); |
10710 | spin_unlock(&space_info->lock); | |
10711 | ||
e33e17ee | 10712 | /* DISCARD can flip during remount */ |
0b246afa | 10713 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10714 | |
10715 | /* Implicit trim during transaction commit. */ | |
10716 | if (trimming) | |
10717 | btrfs_get_block_group_trimming(block_group); | |
10718 | ||
47ab2a6c JB |
10719 | /* |
10720 | * Btrfs_remove_chunk will abort the transaction if things go | |
10721 | * horribly wrong. | |
10722 | */ | |
5b4aacef | 10723 | ret = btrfs_remove_chunk(trans, fs_info, |
47ab2a6c | 10724 | block_group->key.objectid); |
e33e17ee JM |
10725 | |
10726 | if (ret) { | |
10727 | if (trimming) | |
10728 | btrfs_put_block_group_trimming(block_group); | |
10729 | goto end_trans; | |
10730 | } | |
10731 | ||
10732 | /* | |
10733 | * If we're not mounted with -odiscard, we can just forget | |
10734 | * about this block group. Otherwise we'll need to wait | |
10735 | * until transaction commit to do the actual discard. | |
10736 | */ | |
10737 | if (trimming) { | |
348a0013 FM |
10738 | spin_lock(&fs_info->unused_bgs_lock); |
10739 | /* | |
10740 | * A concurrent scrub might have added us to the list | |
10741 | * fs_info->unused_bgs, so use a list_move operation | |
10742 | * to add the block group to the deleted_bgs list. | |
10743 | */ | |
e33e17ee JM |
10744 | list_move(&block_group->bg_list, |
10745 | &trans->transaction->deleted_bgs); | |
348a0013 | 10746 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10747 | btrfs_get_block_group(block_group); |
10748 | } | |
758eb51e | 10749 | end_trans: |
3a45bb20 | 10750 | btrfs_end_transaction(trans); |
47ab2a6c | 10751 | next: |
d5f2e33b | 10752 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10753 | btrfs_put_block_group(block_group); |
10754 | spin_lock(&fs_info->unused_bgs_lock); | |
10755 | } | |
10756 | spin_unlock(&fs_info->unused_bgs_lock); | |
10757 | } | |
10758 | ||
c59021f8 | 10759 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10760 | { | |
10761 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10762 | struct btrfs_super_block *disk_super; |
10763 | u64 features; | |
10764 | u64 flags; | |
10765 | int mixed = 0; | |
c59021f8 | 10766 | int ret; |
10767 | ||
6c41761f | 10768 | disk_super = fs_info->super_copy; |
1aba86d6 | 10769 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10770 | return -EINVAL; |
c59021f8 | 10771 | |
1aba86d6 | 10772 | features = btrfs_super_incompat_flags(disk_super); |
10773 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10774 | mixed = 1; | |
c59021f8 | 10775 | |
1aba86d6 | 10776 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
e40edf2d | 10777 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
c59021f8 | 10778 | if (ret) |
1aba86d6 | 10779 | goto out; |
c59021f8 | 10780 | |
1aba86d6 | 10781 | if (mixed) { |
10782 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
e40edf2d | 10783 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10784 | } else { |
10785 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
e40edf2d | 10786 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10787 | if (ret) |
10788 | goto out; | |
10789 | ||
10790 | flags = BTRFS_BLOCK_GROUP_DATA; | |
e40edf2d | 10791 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10792 | } |
10793 | out: | |
c59021f8 | 10794 | return ret; |
10795 | } | |
10796 | ||
2ff7e61e JM |
10797 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10798 | u64 start, u64 end) | |
acce952b | 10799 | { |
2ff7e61e | 10800 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10801 | } |
10802 | ||
499f377f JM |
10803 | /* |
10804 | * It used to be that old block groups would be left around forever. | |
10805 | * Iterating over them would be enough to trim unused space. Since we | |
10806 | * now automatically remove them, we also need to iterate over unallocated | |
10807 | * space. | |
10808 | * | |
10809 | * We don't want a transaction for this since the discard may take a | |
10810 | * substantial amount of time. We don't require that a transaction be | |
10811 | * running, but we do need to take a running transaction into account | |
10812 | * to ensure that we're not discarding chunks that were released in | |
10813 | * the current transaction. | |
10814 | * | |
10815 | * Holding the chunks lock will prevent other threads from allocating | |
10816 | * or releasing chunks, but it won't prevent a running transaction | |
10817 | * from committing and releasing the memory that the pending chunks | |
10818 | * list head uses. For that, we need to take a reference to the | |
10819 | * transaction. | |
10820 | */ | |
10821 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10822 | u64 minlen, u64 *trimmed) | |
10823 | { | |
10824 | u64 start = 0, len = 0; | |
10825 | int ret; | |
10826 | ||
10827 | *trimmed = 0; | |
10828 | ||
10829 | /* Not writeable = nothing to do. */ | |
10830 | if (!device->writeable) | |
10831 | return 0; | |
10832 | ||
10833 | /* No free space = nothing to do. */ | |
10834 | if (device->total_bytes <= device->bytes_used) | |
10835 | return 0; | |
10836 | ||
10837 | ret = 0; | |
10838 | ||
10839 | while (1) { | |
fb456252 | 10840 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10841 | struct btrfs_transaction *trans; |
10842 | u64 bytes; | |
10843 | ||
10844 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10845 | if (ret) | |
10846 | return ret; | |
10847 | ||
10848 | down_read(&fs_info->commit_root_sem); | |
10849 | ||
10850 | spin_lock(&fs_info->trans_lock); | |
10851 | trans = fs_info->running_transaction; | |
10852 | if (trans) | |
10853 | atomic_inc(&trans->use_count); | |
10854 | spin_unlock(&fs_info->trans_lock); | |
10855 | ||
10856 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10857 | &start, &len); | |
10858 | if (trans) | |
10859 | btrfs_put_transaction(trans); | |
10860 | ||
10861 | if (ret) { | |
10862 | up_read(&fs_info->commit_root_sem); | |
10863 | mutex_unlock(&fs_info->chunk_mutex); | |
10864 | if (ret == -ENOSPC) | |
10865 | ret = 0; | |
10866 | break; | |
10867 | } | |
10868 | ||
10869 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10870 | up_read(&fs_info->commit_root_sem); | |
10871 | mutex_unlock(&fs_info->chunk_mutex); | |
10872 | ||
10873 | if (ret) | |
10874 | break; | |
10875 | ||
10876 | start += len; | |
10877 | *trimmed += bytes; | |
10878 | ||
10879 | if (fatal_signal_pending(current)) { | |
10880 | ret = -ERESTARTSYS; | |
10881 | break; | |
10882 | } | |
10883 | ||
10884 | cond_resched(); | |
10885 | } | |
10886 | ||
10887 | return ret; | |
10888 | } | |
10889 | ||
2ff7e61e | 10890 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 10891 | { |
f7039b1d | 10892 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
10893 | struct btrfs_device *device; |
10894 | struct list_head *devices; | |
f7039b1d LD |
10895 | u64 group_trimmed; |
10896 | u64 start; | |
10897 | u64 end; | |
10898 | u64 trimmed = 0; | |
2cac13e4 | 10899 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10900 | int ret = 0; |
10901 | ||
2cac13e4 LB |
10902 | /* |
10903 | * try to trim all FS space, our block group may start from non-zero. | |
10904 | */ | |
10905 | if (range->len == total_bytes) | |
10906 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10907 | else | |
10908 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10909 | |
10910 | while (cache) { | |
10911 | if (cache->key.objectid >= (range->start + range->len)) { | |
10912 | btrfs_put_block_group(cache); | |
10913 | break; | |
10914 | } | |
10915 | ||
10916 | start = max(range->start, cache->key.objectid); | |
10917 | end = min(range->start + range->len, | |
10918 | cache->key.objectid + cache->key.offset); | |
10919 | ||
10920 | if (end - start >= range->minlen) { | |
10921 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10922 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
10923 | if (ret) { |
10924 | btrfs_put_block_group(cache); | |
10925 | break; | |
10926 | } | |
10927 | ret = wait_block_group_cache_done(cache); | |
10928 | if (ret) { | |
10929 | btrfs_put_block_group(cache); | |
10930 | break; | |
10931 | } | |
f7039b1d LD |
10932 | } |
10933 | ret = btrfs_trim_block_group(cache, | |
10934 | &group_trimmed, | |
10935 | start, | |
10936 | end, | |
10937 | range->minlen); | |
10938 | ||
10939 | trimmed += group_trimmed; | |
10940 | if (ret) { | |
10941 | btrfs_put_block_group(cache); | |
10942 | break; | |
10943 | } | |
10944 | } | |
10945 | ||
2ff7e61e | 10946 | cache = next_block_group(fs_info, cache); |
f7039b1d LD |
10947 | } |
10948 | ||
0b246afa JM |
10949 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
10950 | devices = &fs_info->fs_devices->alloc_list; | |
499f377f JM |
10951 | list_for_each_entry(device, devices, dev_alloc_list) { |
10952 | ret = btrfs_trim_free_extents(device, range->minlen, | |
10953 | &group_trimmed); | |
10954 | if (ret) | |
10955 | break; | |
10956 | ||
10957 | trimmed += group_trimmed; | |
10958 | } | |
0b246afa | 10959 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 10960 | |
f7039b1d LD |
10961 | range->len = trimmed; |
10962 | return ret; | |
10963 | } | |
8257b2dc MX |
10964 | |
10965 | /* | |
9ea24bbe FM |
10966 | * btrfs_{start,end}_write_no_snapshoting() are similar to |
10967 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing | |
10968 | * data into the page cache through nocow before the subvolume is snapshoted, | |
10969 | * but flush the data into disk after the snapshot creation, or to prevent | |
10970 | * operations while snapshoting is ongoing and that cause the snapshot to be | |
10971 | * inconsistent (writes followed by expanding truncates for example). | |
8257b2dc | 10972 | */ |
9ea24bbe | 10973 | void btrfs_end_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc MX |
10974 | { |
10975 | percpu_counter_dec(&root->subv_writers->counter); | |
10976 | /* | |
a83342aa | 10977 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
10978 | */ |
10979 | smp_mb(); | |
10980 | if (waitqueue_active(&root->subv_writers->wait)) | |
10981 | wake_up(&root->subv_writers->wait); | |
10982 | } | |
10983 | ||
9ea24bbe | 10984 | int btrfs_start_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc | 10985 | { |
ee39b432 | 10986 | if (atomic_read(&root->will_be_snapshoted)) |
8257b2dc MX |
10987 | return 0; |
10988 | ||
10989 | percpu_counter_inc(&root->subv_writers->counter); | |
10990 | /* | |
10991 | * Make sure counter is updated before we check for snapshot creation. | |
10992 | */ | |
10993 | smp_mb(); | |
ee39b432 | 10994 | if (atomic_read(&root->will_be_snapshoted)) { |
9ea24bbe | 10995 | btrfs_end_write_no_snapshoting(root); |
8257b2dc MX |
10996 | return 0; |
10997 | } | |
10998 | return 1; | |
10999 | } | |
0bc19f90 ZL |
11000 | |
11001 | static int wait_snapshoting_atomic_t(atomic_t *a) | |
11002 | { | |
11003 | schedule(); | |
11004 | return 0; | |
11005 | } | |
11006 | ||
11007 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) | |
11008 | { | |
11009 | while (true) { | |
11010 | int ret; | |
11011 | ||
11012 | ret = btrfs_start_write_no_snapshoting(root); | |
11013 | if (ret) | |
11014 | break; | |
11015 | wait_on_atomic_t(&root->will_be_snapshoted, | |
11016 | wait_snapshoting_atomic_t, | |
11017 | TASK_UNINTERRUPTIBLE); | |
11018 | } | |
11019 | } |