]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
edbd8d4e | 19 | #include <linux/pagemap.h> |
ec44a35c | 20 | #include <linux/writeback.h> |
21af804c | 21 | #include <linux/blkdev.h> |
b7a9f29f | 22 | #include <linux/sort.h> |
4184ea7f | 23 | #include <linux/rcupdate.h> |
817d52f8 | 24 | #include <linux/kthread.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
dff51cd1 | 26 | #include <linux/ratelimit.h> |
b150a4f1 | 27 | #include <linux/percpu_counter.h> |
74493f7a | 28 | #include "hash.h" |
995946dd | 29 | #include "tree-log.h" |
fec577fb CM |
30 | #include "disk-io.h" |
31 | #include "print-tree.h" | |
0b86a832 | 32 | #include "volumes.h" |
53b381b3 | 33 | #include "raid56.h" |
925baedd | 34 | #include "locking.h" |
fa9c0d79 | 35 | #include "free-space-cache.h" |
1e144fb8 | 36 | #include "free-space-tree.h" |
3fed40cc | 37 | #include "math.h" |
6ab0a202 | 38 | #include "sysfs.h" |
fcebe456 | 39 | #include "qgroup.h" |
fec577fb | 40 | |
709c0486 AJ |
41 | #undef SCRAMBLE_DELAYED_REFS |
42 | ||
9e622d6b MX |
43 | /* |
44 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
45 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
46 | * if we really need one. | |
47 | * | |
0e4f8f88 CM |
48 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
49 | * if we have very few chunks already allocated. This is | |
50 | * used as part of the clustering code to help make sure | |
51 | * we have a good pool of storage to cluster in, without | |
52 | * filling the FS with empty chunks | |
53 | * | |
9e622d6b MX |
54 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
55 | * | |
0e4f8f88 CM |
56 | */ |
57 | enum { | |
58 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
59 | CHUNK_ALLOC_LIMITED = 1, |
60 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
61 | }; |
62 | ||
ce93ec54 JB |
63 | static int update_block_group(struct btrfs_trans_handle *trans, |
64 | struct btrfs_root *root, u64 bytenr, | |
65 | u64 num_bytes, int alloc); | |
5d4f98a2 YZ |
66 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
67 | struct btrfs_root *root, | |
c682f9b3 | 68 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
69 | u64 root_objectid, u64 owner_objectid, |
70 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 71 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
72 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
73 | struct extent_buffer *leaf, | |
74 | struct btrfs_extent_item *ei); | |
75 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
76 | struct btrfs_root *root, | |
77 | u64 parent, u64 root_objectid, | |
78 | u64 flags, u64 owner, u64 offset, | |
79 | struct btrfs_key *ins, int ref_mod); | |
80 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
81 | struct btrfs_root *root, | |
82 | u64 parent, u64 root_objectid, | |
83 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 84 | int level, struct btrfs_key *ins); |
6a63209f | 85 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 JB |
86 | struct btrfs_root *extent_root, u64 flags, |
87 | int force); | |
11833d66 YZ |
88 | static int find_next_key(struct btrfs_path *path, int level, |
89 | struct btrfs_key *key); | |
ab8d0fc4 JM |
90 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
91 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 92 | int dump_block_groups); |
4824f1f4 | 93 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 94 | u64 ram_bytes, u64 num_bytes, int delalloc); |
4824f1f4 WX |
95 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
96 | u64 num_bytes, int delalloc); | |
5d80366e JB |
97 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
98 | u64 num_bytes); | |
48a3b636 ES |
99 | int btrfs_pin_extent(struct btrfs_root *root, |
100 | u64 bytenr, u64 num_bytes, int reserved); | |
957780eb JB |
101 | static int __reserve_metadata_bytes(struct btrfs_root *root, |
102 | struct btrfs_space_info *space_info, | |
103 | u64 orig_bytes, | |
104 | enum btrfs_reserve_flush_enum flush); | |
105 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
106 | struct btrfs_space_info *space_info, | |
107 | u64 num_bytes); | |
108 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
109 | struct btrfs_space_info *space_info, | |
110 | u64 num_bytes); | |
6a63209f | 111 | |
817d52f8 JB |
112 | static noinline int |
113 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
114 | { | |
115 | smp_mb(); | |
36cce922 JB |
116 | return cache->cached == BTRFS_CACHE_FINISHED || |
117 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
118 | } |
119 | ||
0f9dd46c JB |
120 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
121 | { | |
122 | return (cache->flags & bits) == bits; | |
123 | } | |
124 | ||
758f2dfc | 125 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
126 | { |
127 | atomic_inc(&cache->count); | |
128 | } | |
129 | ||
130 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
131 | { | |
f0486c68 YZ |
132 | if (atomic_dec_and_test(&cache->count)) { |
133 | WARN_ON(cache->pinned > 0); | |
134 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 135 | kfree(cache->free_space_ctl); |
11dfe35a | 136 | kfree(cache); |
f0486c68 | 137 | } |
11dfe35a JB |
138 | } |
139 | ||
0f9dd46c JB |
140 | /* |
141 | * this adds the block group to the fs_info rb tree for the block group | |
142 | * cache | |
143 | */ | |
b2950863 | 144 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
145 | struct btrfs_block_group_cache *block_group) |
146 | { | |
147 | struct rb_node **p; | |
148 | struct rb_node *parent = NULL; | |
149 | struct btrfs_block_group_cache *cache; | |
150 | ||
151 | spin_lock(&info->block_group_cache_lock); | |
152 | p = &info->block_group_cache_tree.rb_node; | |
153 | ||
154 | while (*p) { | |
155 | parent = *p; | |
156 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
157 | cache_node); | |
158 | if (block_group->key.objectid < cache->key.objectid) { | |
159 | p = &(*p)->rb_left; | |
160 | } else if (block_group->key.objectid > cache->key.objectid) { | |
161 | p = &(*p)->rb_right; | |
162 | } else { | |
163 | spin_unlock(&info->block_group_cache_lock); | |
164 | return -EEXIST; | |
165 | } | |
166 | } | |
167 | ||
168 | rb_link_node(&block_group->cache_node, parent, p); | |
169 | rb_insert_color(&block_group->cache_node, | |
170 | &info->block_group_cache_tree); | |
a1897fdd LB |
171 | |
172 | if (info->first_logical_byte > block_group->key.objectid) | |
173 | info->first_logical_byte = block_group->key.objectid; | |
174 | ||
0f9dd46c JB |
175 | spin_unlock(&info->block_group_cache_lock); |
176 | ||
177 | return 0; | |
178 | } | |
179 | ||
180 | /* | |
181 | * This will return the block group at or after bytenr if contains is 0, else | |
182 | * it will return the block group that contains the bytenr | |
183 | */ | |
184 | static struct btrfs_block_group_cache * | |
185 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
186 | int contains) | |
187 | { | |
188 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
189 | struct rb_node *n; | |
190 | u64 end, start; | |
191 | ||
192 | spin_lock(&info->block_group_cache_lock); | |
193 | n = info->block_group_cache_tree.rb_node; | |
194 | ||
195 | while (n) { | |
196 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
197 | cache_node); | |
198 | end = cache->key.objectid + cache->key.offset - 1; | |
199 | start = cache->key.objectid; | |
200 | ||
201 | if (bytenr < start) { | |
202 | if (!contains && (!ret || start < ret->key.objectid)) | |
203 | ret = cache; | |
204 | n = n->rb_left; | |
205 | } else if (bytenr > start) { | |
206 | if (contains && bytenr <= end) { | |
207 | ret = cache; | |
208 | break; | |
209 | } | |
210 | n = n->rb_right; | |
211 | } else { | |
212 | ret = cache; | |
213 | break; | |
214 | } | |
215 | } | |
a1897fdd | 216 | if (ret) { |
11dfe35a | 217 | btrfs_get_block_group(ret); |
a1897fdd LB |
218 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
219 | info->first_logical_byte = ret->key.objectid; | |
220 | } | |
0f9dd46c JB |
221 | spin_unlock(&info->block_group_cache_lock); |
222 | ||
223 | return ret; | |
224 | } | |
225 | ||
11833d66 YZ |
226 | static int add_excluded_extent(struct btrfs_root *root, |
227 | u64 start, u64 num_bytes) | |
817d52f8 | 228 | { |
11833d66 YZ |
229 | u64 end = start + num_bytes - 1; |
230 | set_extent_bits(&root->fs_info->freed_extents[0], | |
ceeb0ae7 | 231 | start, end, EXTENT_UPTODATE); |
11833d66 | 232 | set_extent_bits(&root->fs_info->freed_extents[1], |
ceeb0ae7 | 233 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
234 | return 0; |
235 | } | |
817d52f8 | 236 | |
11833d66 YZ |
237 | static void free_excluded_extents(struct btrfs_root *root, |
238 | struct btrfs_block_group_cache *cache) | |
239 | { | |
240 | u64 start, end; | |
817d52f8 | 241 | |
11833d66 YZ |
242 | start = cache->key.objectid; |
243 | end = start + cache->key.offset - 1; | |
244 | ||
245 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
91166212 | 246 | start, end, EXTENT_UPTODATE); |
11833d66 | 247 | clear_extent_bits(&root->fs_info->freed_extents[1], |
91166212 | 248 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
249 | } |
250 | ||
11833d66 YZ |
251 | static int exclude_super_stripes(struct btrfs_root *root, |
252 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 253 | { |
817d52f8 JB |
254 | u64 bytenr; |
255 | u64 *logical; | |
256 | int stripe_len; | |
257 | int i, nr, ret; | |
258 | ||
06b2331f YZ |
259 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
260 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
261 | cache->bytes_super += stripe_len; | |
262 | ret = add_excluded_extent(root, cache->key.objectid, | |
263 | stripe_len); | |
835d974f JB |
264 | if (ret) |
265 | return ret; | |
06b2331f YZ |
266 | } |
267 | ||
817d52f8 JB |
268 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
269 | bytenr = btrfs_sb_offset(i); | |
ab8d0fc4 JM |
270 | ret = btrfs_rmap_block(root->fs_info, cache->key.objectid, |
271 | bytenr, 0, &logical, &nr, &stripe_len); | |
835d974f JB |
272 | if (ret) |
273 | return ret; | |
11833d66 | 274 | |
817d52f8 | 275 | while (nr--) { |
51bf5f0b JB |
276 | u64 start, len; |
277 | ||
278 | if (logical[nr] > cache->key.objectid + | |
279 | cache->key.offset) | |
280 | continue; | |
281 | ||
282 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
283 | continue; | |
284 | ||
285 | start = logical[nr]; | |
286 | if (start < cache->key.objectid) { | |
287 | start = cache->key.objectid; | |
288 | len = (logical[nr] + stripe_len) - start; | |
289 | } else { | |
290 | len = min_t(u64, stripe_len, | |
291 | cache->key.objectid + | |
292 | cache->key.offset - start); | |
293 | } | |
294 | ||
295 | cache->bytes_super += len; | |
296 | ret = add_excluded_extent(root, start, len); | |
835d974f JB |
297 | if (ret) { |
298 | kfree(logical); | |
299 | return ret; | |
300 | } | |
817d52f8 | 301 | } |
11833d66 | 302 | |
817d52f8 JB |
303 | kfree(logical); |
304 | } | |
817d52f8 JB |
305 | return 0; |
306 | } | |
307 | ||
11833d66 YZ |
308 | static struct btrfs_caching_control * |
309 | get_caching_control(struct btrfs_block_group_cache *cache) | |
310 | { | |
311 | struct btrfs_caching_control *ctl; | |
312 | ||
313 | spin_lock(&cache->lock); | |
dde5abee JB |
314 | if (!cache->caching_ctl) { |
315 | spin_unlock(&cache->lock); | |
11833d66 YZ |
316 | return NULL; |
317 | } | |
318 | ||
319 | ctl = cache->caching_ctl; | |
320 | atomic_inc(&ctl->count); | |
321 | spin_unlock(&cache->lock); | |
322 | return ctl; | |
323 | } | |
324 | ||
325 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
326 | { | |
327 | if (atomic_dec_and_test(&ctl->count)) | |
328 | kfree(ctl); | |
329 | } | |
330 | ||
d0bd4560 JB |
331 | #ifdef CONFIG_BTRFS_DEBUG |
332 | static void fragment_free_space(struct btrfs_root *root, | |
333 | struct btrfs_block_group_cache *block_group) | |
334 | { | |
335 | u64 start = block_group->key.objectid; | |
336 | u64 len = block_group->key.offset; | |
337 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
338 | root->nodesize : root->sectorsize; | |
339 | u64 step = chunk << 1; | |
340 | ||
341 | while (len > chunk) { | |
342 | btrfs_remove_free_space(block_group, start, chunk); | |
343 | start += step; | |
344 | if (len < step) | |
345 | len = 0; | |
346 | else | |
347 | len -= step; | |
348 | } | |
349 | } | |
350 | #endif | |
351 | ||
0f9dd46c JB |
352 | /* |
353 | * this is only called by cache_block_group, since we could have freed extents | |
354 | * we need to check the pinned_extents for any extents that can't be used yet | |
355 | * since their free space will be released as soon as the transaction commits. | |
356 | */ | |
a5ed9182 OS |
357 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
358 | struct btrfs_fs_info *info, u64 start, u64 end) | |
0f9dd46c | 359 | { |
817d52f8 | 360 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
361 | int ret; |
362 | ||
363 | while (start < end) { | |
11833d66 | 364 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 365 | &extent_start, &extent_end, |
e6138876 JB |
366 | EXTENT_DIRTY | EXTENT_UPTODATE, |
367 | NULL); | |
0f9dd46c JB |
368 | if (ret) |
369 | break; | |
370 | ||
06b2331f | 371 | if (extent_start <= start) { |
0f9dd46c JB |
372 | start = extent_end + 1; |
373 | } else if (extent_start > start && extent_start < end) { | |
374 | size = extent_start - start; | |
817d52f8 | 375 | total_added += size; |
ea6a478e JB |
376 | ret = btrfs_add_free_space(block_group, start, |
377 | size); | |
79787eaa | 378 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
379 | start = extent_end + 1; |
380 | } else { | |
381 | break; | |
382 | } | |
383 | } | |
384 | ||
385 | if (start < end) { | |
386 | size = end - start; | |
817d52f8 | 387 | total_added += size; |
ea6a478e | 388 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 389 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
390 | } |
391 | ||
817d52f8 | 392 | return total_added; |
0f9dd46c JB |
393 | } |
394 | ||
73fa48b6 | 395 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 396 | { |
bab39bf9 JB |
397 | struct btrfs_block_group_cache *block_group; |
398 | struct btrfs_fs_info *fs_info; | |
bab39bf9 | 399 | struct btrfs_root *extent_root; |
e37c9e69 | 400 | struct btrfs_path *path; |
5f39d397 | 401 | struct extent_buffer *leaf; |
11833d66 | 402 | struct btrfs_key key; |
817d52f8 | 403 | u64 total_found = 0; |
11833d66 YZ |
404 | u64 last = 0; |
405 | u32 nritems; | |
73fa48b6 | 406 | int ret; |
d0bd4560 | 407 | bool wakeup = true; |
f510cfec | 408 | |
bab39bf9 JB |
409 | block_group = caching_ctl->block_group; |
410 | fs_info = block_group->fs_info; | |
411 | extent_root = fs_info->extent_root; | |
412 | ||
e37c9e69 CM |
413 | path = btrfs_alloc_path(); |
414 | if (!path) | |
73fa48b6 | 415 | return -ENOMEM; |
7d7d6068 | 416 | |
817d52f8 | 417 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 418 | |
d0bd4560 JB |
419 | #ifdef CONFIG_BTRFS_DEBUG |
420 | /* | |
421 | * If we're fragmenting we don't want to make anybody think we can | |
422 | * allocate from this block group until we've had a chance to fragment | |
423 | * the free space. | |
424 | */ | |
425 | if (btrfs_should_fragment_free_space(extent_root, block_group)) | |
426 | wakeup = false; | |
427 | #endif | |
5cd57b2c | 428 | /* |
817d52f8 JB |
429 | * We don't want to deadlock with somebody trying to allocate a new |
430 | * extent for the extent root while also trying to search the extent | |
431 | * root to add free space. So we skip locking and search the commit | |
432 | * root, since its read-only | |
5cd57b2c CM |
433 | */ |
434 | path->skip_locking = 1; | |
817d52f8 | 435 | path->search_commit_root = 1; |
e4058b54 | 436 | path->reada = READA_FORWARD; |
817d52f8 | 437 | |
e4404d6e | 438 | key.objectid = last; |
e37c9e69 | 439 | key.offset = 0; |
11833d66 | 440 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 441 | |
52ee28d2 | 442 | next: |
11833d66 | 443 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 444 | if (ret < 0) |
73fa48b6 | 445 | goto out; |
a512bbf8 | 446 | |
11833d66 YZ |
447 | leaf = path->nodes[0]; |
448 | nritems = btrfs_header_nritems(leaf); | |
449 | ||
d397712b | 450 | while (1) { |
7841cb28 | 451 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 452 | last = (u64)-1; |
817d52f8 | 453 | break; |
f25784b3 | 454 | } |
817d52f8 | 455 | |
11833d66 YZ |
456 | if (path->slots[0] < nritems) { |
457 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
458 | } else { | |
459 | ret = find_next_key(path, 0, &key); | |
460 | if (ret) | |
e37c9e69 | 461 | break; |
817d52f8 | 462 | |
c9ea7b24 | 463 | if (need_resched() || |
9e351cc8 | 464 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
465 | if (wakeup) |
466 | caching_ctl->progress = last; | |
ff5714cc | 467 | btrfs_release_path(path); |
9e351cc8 | 468 | up_read(&fs_info->commit_root_sem); |
589d8ade | 469 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 470 | cond_resched(); |
73fa48b6 OS |
471 | mutex_lock(&caching_ctl->mutex); |
472 | down_read(&fs_info->commit_root_sem); | |
473 | goto next; | |
589d8ade | 474 | } |
0a3896d0 JB |
475 | |
476 | ret = btrfs_next_leaf(extent_root, path); | |
477 | if (ret < 0) | |
73fa48b6 | 478 | goto out; |
0a3896d0 JB |
479 | if (ret) |
480 | break; | |
589d8ade JB |
481 | leaf = path->nodes[0]; |
482 | nritems = btrfs_header_nritems(leaf); | |
483 | continue; | |
11833d66 | 484 | } |
817d52f8 | 485 | |
52ee28d2 LB |
486 | if (key.objectid < last) { |
487 | key.objectid = last; | |
488 | key.offset = 0; | |
489 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
490 | ||
d0bd4560 JB |
491 | if (wakeup) |
492 | caching_ctl->progress = last; | |
52ee28d2 LB |
493 | btrfs_release_path(path); |
494 | goto next; | |
495 | } | |
496 | ||
11833d66 YZ |
497 | if (key.objectid < block_group->key.objectid) { |
498 | path->slots[0]++; | |
817d52f8 | 499 | continue; |
e37c9e69 | 500 | } |
0f9dd46c | 501 | |
e37c9e69 | 502 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 503 | block_group->key.offset) |
e37c9e69 | 504 | break; |
7d7d6068 | 505 | |
3173a18f JB |
506 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
507 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
508 | total_found += add_new_free_space(block_group, |
509 | fs_info, last, | |
510 | key.objectid); | |
3173a18f JB |
511 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
512 | last = key.objectid + | |
707e8a07 | 513 | fs_info->tree_root->nodesize; |
3173a18f JB |
514 | else |
515 | last = key.objectid + key.offset; | |
817d52f8 | 516 | |
73fa48b6 | 517 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 518 | total_found = 0; |
d0bd4560 JB |
519 | if (wakeup) |
520 | wake_up(&caching_ctl->wait); | |
11833d66 | 521 | } |
817d52f8 | 522 | } |
e37c9e69 CM |
523 | path->slots[0]++; |
524 | } | |
817d52f8 | 525 | ret = 0; |
e37c9e69 | 526 | |
817d52f8 JB |
527 | total_found += add_new_free_space(block_group, fs_info, last, |
528 | block_group->key.objectid + | |
529 | block_group->key.offset); | |
11833d66 | 530 | caching_ctl->progress = (u64)-1; |
817d52f8 | 531 | |
73fa48b6 OS |
532 | out: |
533 | btrfs_free_path(path); | |
534 | return ret; | |
535 | } | |
536 | ||
537 | static noinline void caching_thread(struct btrfs_work *work) | |
538 | { | |
539 | struct btrfs_block_group_cache *block_group; | |
540 | struct btrfs_fs_info *fs_info; | |
541 | struct btrfs_caching_control *caching_ctl; | |
b4570aa9 | 542 | struct btrfs_root *extent_root; |
73fa48b6 OS |
543 | int ret; |
544 | ||
545 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
546 | block_group = caching_ctl->block_group; | |
547 | fs_info = block_group->fs_info; | |
b4570aa9 | 548 | extent_root = fs_info->extent_root; |
73fa48b6 OS |
549 | |
550 | mutex_lock(&caching_ctl->mutex); | |
551 | down_read(&fs_info->commit_root_sem); | |
552 | ||
1e144fb8 OS |
553 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
554 | ret = load_free_space_tree(caching_ctl); | |
555 | else | |
556 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 557 | |
817d52f8 | 558 | spin_lock(&block_group->lock); |
11833d66 | 559 | block_group->caching_ctl = NULL; |
73fa48b6 | 560 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 561 | spin_unlock(&block_group->lock); |
0f9dd46c | 562 | |
d0bd4560 JB |
563 | #ifdef CONFIG_BTRFS_DEBUG |
564 | if (btrfs_should_fragment_free_space(extent_root, block_group)) { | |
565 | u64 bytes_used; | |
566 | ||
567 | spin_lock(&block_group->space_info->lock); | |
568 | spin_lock(&block_group->lock); | |
569 | bytes_used = block_group->key.offset - | |
570 | btrfs_block_group_used(&block_group->item); | |
571 | block_group->space_info->bytes_used += bytes_used >> 1; | |
572 | spin_unlock(&block_group->lock); | |
573 | spin_unlock(&block_group->space_info->lock); | |
574 | fragment_free_space(extent_root, block_group); | |
575 | } | |
576 | #endif | |
577 | ||
578 | caching_ctl->progress = (u64)-1; | |
11833d66 | 579 | |
9e351cc8 | 580 | up_read(&fs_info->commit_root_sem); |
73fa48b6 | 581 | free_excluded_extents(fs_info->extent_root, block_group); |
11833d66 | 582 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 583 | |
11833d66 YZ |
584 | wake_up(&caching_ctl->wait); |
585 | ||
586 | put_caching_control(caching_ctl); | |
11dfe35a | 587 | btrfs_put_block_group(block_group); |
817d52f8 JB |
588 | } |
589 | ||
9d66e233 | 590 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 591 | int load_cache_only) |
817d52f8 | 592 | { |
291c7d2f | 593 | DEFINE_WAIT(wait); |
11833d66 YZ |
594 | struct btrfs_fs_info *fs_info = cache->fs_info; |
595 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
596 | int ret = 0; |
597 | ||
291c7d2f | 598 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
599 | if (!caching_ctl) |
600 | return -ENOMEM; | |
291c7d2f JB |
601 | |
602 | INIT_LIST_HEAD(&caching_ctl->list); | |
603 | mutex_init(&caching_ctl->mutex); | |
604 | init_waitqueue_head(&caching_ctl->wait); | |
605 | caching_ctl->block_group = cache; | |
606 | caching_ctl->progress = cache->key.objectid; | |
607 | atomic_set(&caching_ctl->count, 1); | |
9e0af237 LB |
608 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
609 | caching_thread, NULL, NULL); | |
291c7d2f JB |
610 | |
611 | spin_lock(&cache->lock); | |
612 | /* | |
613 | * This should be a rare occasion, but this could happen I think in the | |
614 | * case where one thread starts to load the space cache info, and then | |
615 | * some other thread starts a transaction commit which tries to do an | |
616 | * allocation while the other thread is still loading the space cache | |
617 | * info. The previous loop should have kept us from choosing this block | |
618 | * group, but if we've moved to the state where we will wait on caching | |
619 | * block groups we need to first check if we're doing a fast load here, | |
620 | * so we can wait for it to finish, otherwise we could end up allocating | |
621 | * from a block group who's cache gets evicted for one reason or | |
622 | * another. | |
623 | */ | |
624 | while (cache->cached == BTRFS_CACHE_FAST) { | |
625 | struct btrfs_caching_control *ctl; | |
626 | ||
627 | ctl = cache->caching_ctl; | |
628 | atomic_inc(&ctl->count); | |
629 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
630 | spin_unlock(&cache->lock); | |
631 | ||
632 | schedule(); | |
633 | ||
634 | finish_wait(&ctl->wait, &wait); | |
635 | put_caching_control(ctl); | |
636 | spin_lock(&cache->lock); | |
637 | } | |
638 | ||
639 | if (cache->cached != BTRFS_CACHE_NO) { | |
640 | spin_unlock(&cache->lock); | |
641 | kfree(caching_ctl); | |
11833d66 | 642 | return 0; |
291c7d2f JB |
643 | } |
644 | WARN_ON(cache->caching_ctl); | |
645 | cache->caching_ctl = caching_ctl; | |
646 | cache->cached = BTRFS_CACHE_FAST; | |
647 | spin_unlock(&cache->lock); | |
11833d66 | 648 | |
d53ba474 | 649 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
cb83b7b8 | 650 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
651 | ret = load_free_space_cache(fs_info, cache); |
652 | ||
653 | spin_lock(&cache->lock); | |
654 | if (ret == 1) { | |
291c7d2f | 655 | cache->caching_ctl = NULL; |
9d66e233 JB |
656 | cache->cached = BTRFS_CACHE_FINISHED; |
657 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 658 | caching_ctl->progress = (u64)-1; |
9d66e233 | 659 | } else { |
291c7d2f JB |
660 | if (load_cache_only) { |
661 | cache->caching_ctl = NULL; | |
662 | cache->cached = BTRFS_CACHE_NO; | |
663 | } else { | |
664 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 665 | cache->has_caching_ctl = 1; |
291c7d2f | 666 | } |
9d66e233 JB |
667 | } |
668 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
669 | #ifdef CONFIG_BTRFS_DEBUG |
670 | if (ret == 1 && | |
671 | btrfs_should_fragment_free_space(fs_info->extent_root, | |
672 | cache)) { | |
673 | u64 bytes_used; | |
674 | ||
675 | spin_lock(&cache->space_info->lock); | |
676 | spin_lock(&cache->lock); | |
677 | bytes_used = cache->key.offset - | |
678 | btrfs_block_group_used(&cache->item); | |
679 | cache->space_info->bytes_used += bytes_used >> 1; | |
680 | spin_unlock(&cache->lock); | |
681 | spin_unlock(&cache->space_info->lock); | |
682 | fragment_free_space(fs_info->extent_root, cache); | |
683 | } | |
684 | #endif | |
cb83b7b8 JB |
685 | mutex_unlock(&caching_ctl->mutex); |
686 | ||
291c7d2f | 687 | wake_up(&caching_ctl->wait); |
3c14874a | 688 | if (ret == 1) { |
291c7d2f | 689 | put_caching_control(caching_ctl); |
3c14874a | 690 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 691 | return 0; |
3c14874a | 692 | } |
291c7d2f JB |
693 | } else { |
694 | /* | |
1e144fb8 OS |
695 | * We're either using the free space tree or no caching at all. |
696 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
697 | */ |
698 | spin_lock(&cache->lock); | |
699 | if (load_cache_only) { | |
700 | cache->caching_ctl = NULL; | |
701 | cache->cached = BTRFS_CACHE_NO; | |
702 | } else { | |
703 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 704 | cache->has_caching_ctl = 1; |
291c7d2f JB |
705 | } |
706 | spin_unlock(&cache->lock); | |
707 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
708 | } |
709 | ||
291c7d2f JB |
710 | if (load_cache_only) { |
711 | put_caching_control(caching_ctl); | |
11833d66 | 712 | return 0; |
817d52f8 | 713 | } |
817d52f8 | 714 | |
9e351cc8 | 715 | down_write(&fs_info->commit_root_sem); |
291c7d2f | 716 | atomic_inc(&caching_ctl->count); |
11833d66 | 717 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 718 | up_write(&fs_info->commit_root_sem); |
11833d66 | 719 | |
11dfe35a | 720 | btrfs_get_block_group(cache); |
11833d66 | 721 | |
e66f0bb1 | 722 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 723 | |
ef8bbdfe | 724 | return ret; |
e37c9e69 CM |
725 | } |
726 | ||
0f9dd46c JB |
727 | /* |
728 | * return the block group that starts at or after bytenr | |
729 | */ | |
d397712b CM |
730 | static struct btrfs_block_group_cache * |
731 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 732 | { |
e2c89907 | 733 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
734 | } |
735 | ||
0f9dd46c | 736 | /* |
9f55684c | 737 | * return the block group that contains the given bytenr |
0f9dd46c | 738 | */ |
d397712b CM |
739 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
740 | struct btrfs_fs_info *info, | |
741 | u64 bytenr) | |
be744175 | 742 | { |
e2c89907 | 743 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 744 | } |
0b86a832 | 745 | |
0f9dd46c JB |
746 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
747 | u64 flags) | |
6324fbf3 | 748 | { |
0f9dd46c | 749 | struct list_head *head = &info->space_info; |
0f9dd46c | 750 | struct btrfs_space_info *found; |
4184ea7f | 751 | |
52ba6929 | 752 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 753 | |
4184ea7f CM |
754 | rcu_read_lock(); |
755 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 756 | if (found->flags & flags) { |
4184ea7f | 757 | rcu_read_unlock(); |
0f9dd46c | 758 | return found; |
4184ea7f | 759 | } |
0f9dd46c | 760 | } |
4184ea7f | 761 | rcu_read_unlock(); |
0f9dd46c | 762 | return NULL; |
6324fbf3 CM |
763 | } |
764 | ||
4184ea7f CM |
765 | /* |
766 | * after adding space to the filesystem, we need to clear the full flags | |
767 | * on all the space infos. | |
768 | */ | |
769 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
770 | { | |
771 | struct list_head *head = &info->space_info; | |
772 | struct btrfs_space_info *found; | |
773 | ||
774 | rcu_read_lock(); | |
775 | list_for_each_entry_rcu(found, head, list) | |
776 | found->full = 0; | |
777 | rcu_read_unlock(); | |
778 | } | |
779 | ||
1a4ed8fd FM |
780 | /* simple helper to search for an existing data extent at a given offset */ |
781 | int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len) | |
e02119d5 CM |
782 | { |
783 | int ret; | |
784 | struct btrfs_key key; | |
31840ae1 | 785 | struct btrfs_path *path; |
e02119d5 | 786 | |
31840ae1 | 787 | path = btrfs_alloc_path(); |
d8926bb3 MF |
788 | if (!path) |
789 | return -ENOMEM; | |
790 | ||
e02119d5 CM |
791 | key.objectid = start; |
792 | key.offset = len; | |
3173a18f | 793 | key.type = BTRFS_EXTENT_ITEM_KEY; |
e02119d5 CM |
794 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, |
795 | 0, 0); | |
31840ae1 | 796 | btrfs_free_path(path); |
7bb86316 CM |
797 | return ret; |
798 | } | |
799 | ||
a22285a6 | 800 | /* |
3173a18f | 801 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
802 | * |
803 | * the head node for delayed ref is used to store the sum of all the | |
804 | * reference count modifications queued up in the rbtree. the head | |
805 | * node may also store the extent flags to set. This way you can check | |
806 | * to see what the reference count and extent flags would be if all of | |
807 | * the delayed refs are not processed. | |
808 | */ | |
809 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
810 | struct btrfs_root *root, u64 bytenr, | |
3173a18f | 811 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
812 | { |
813 | struct btrfs_delayed_ref_head *head; | |
814 | struct btrfs_delayed_ref_root *delayed_refs; | |
815 | struct btrfs_path *path; | |
816 | struct btrfs_extent_item *ei; | |
817 | struct extent_buffer *leaf; | |
818 | struct btrfs_key key; | |
819 | u32 item_size; | |
820 | u64 num_refs; | |
821 | u64 extent_flags; | |
822 | int ret; | |
823 | ||
3173a18f JB |
824 | /* |
825 | * If we don't have skinny metadata, don't bother doing anything | |
826 | * different | |
827 | */ | |
828 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { | |
707e8a07 | 829 | offset = root->nodesize; |
3173a18f JB |
830 | metadata = 0; |
831 | } | |
832 | ||
a22285a6 YZ |
833 | path = btrfs_alloc_path(); |
834 | if (!path) | |
835 | return -ENOMEM; | |
836 | ||
a22285a6 YZ |
837 | if (!trans) { |
838 | path->skip_locking = 1; | |
839 | path->search_commit_root = 1; | |
840 | } | |
639eefc8 FDBM |
841 | |
842 | search_again: | |
843 | key.objectid = bytenr; | |
844 | key.offset = offset; | |
845 | if (metadata) | |
846 | key.type = BTRFS_METADATA_ITEM_KEY; | |
847 | else | |
848 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
849 | ||
a22285a6 YZ |
850 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, |
851 | &key, path, 0, 0); | |
852 | if (ret < 0) | |
853 | goto out_free; | |
854 | ||
3173a18f | 855 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
856 | if (path->slots[0]) { |
857 | path->slots[0]--; | |
858 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
859 | path->slots[0]); | |
860 | if (key.objectid == bytenr && | |
861 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
707e8a07 | 862 | key.offset == root->nodesize) |
74be9510 FDBM |
863 | ret = 0; |
864 | } | |
3173a18f JB |
865 | } |
866 | ||
a22285a6 YZ |
867 | if (ret == 0) { |
868 | leaf = path->nodes[0]; | |
869 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
870 | if (item_size >= sizeof(*ei)) { | |
871 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
872 | struct btrfs_extent_item); | |
873 | num_refs = btrfs_extent_refs(leaf, ei); | |
874 | extent_flags = btrfs_extent_flags(leaf, ei); | |
875 | } else { | |
876 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
877 | struct btrfs_extent_item_v0 *ei0; | |
878 | BUG_ON(item_size != sizeof(*ei0)); | |
879 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
880 | struct btrfs_extent_item_v0); | |
881 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
882 | /* FIXME: this isn't correct for data */ | |
883 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
884 | #else | |
885 | BUG(); | |
886 | #endif | |
887 | } | |
888 | BUG_ON(num_refs == 0); | |
889 | } else { | |
890 | num_refs = 0; | |
891 | extent_flags = 0; | |
892 | ret = 0; | |
893 | } | |
894 | ||
895 | if (!trans) | |
896 | goto out; | |
897 | ||
898 | delayed_refs = &trans->transaction->delayed_refs; | |
899 | spin_lock(&delayed_refs->lock); | |
900 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
901 | if (head) { | |
902 | if (!mutex_trylock(&head->mutex)) { | |
903 | atomic_inc(&head->node.refs); | |
904 | spin_unlock(&delayed_refs->lock); | |
905 | ||
b3b4aa74 | 906 | btrfs_release_path(path); |
a22285a6 | 907 | |
8cc33e5c DS |
908 | /* |
909 | * Mutex was contended, block until it's released and try | |
910 | * again | |
911 | */ | |
a22285a6 YZ |
912 | mutex_lock(&head->mutex); |
913 | mutex_unlock(&head->mutex); | |
914 | btrfs_put_delayed_ref(&head->node); | |
639eefc8 | 915 | goto search_again; |
a22285a6 | 916 | } |
d7df2c79 | 917 | spin_lock(&head->lock); |
a22285a6 YZ |
918 | if (head->extent_op && head->extent_op->update_flags) |
919 | extent_flags |= head->extent_op->flags_to_set; | |
920 | else | |
921 | BUG_ON(num_refs == 0); | |
922 | ||
923 | num_refs += head->node.ref_mod; | |
d7df2c79 | 924 | spin_unlock(&head->lock); |
a22285a6 YZ |
925 | mutex_unlock(&head->mutex); |
926 | } | |
927 | spin_unlock(&delayed_refs->lock); | |
928 | out: | |
929 | WARN_ON(num_refs == 0); | |
930 | if (refs) | |
931 | *refs = num_refs; | |
932 | if (flags) | |
933 | *flags = extent_flags; | |
934 | out_free: | |
935 | btrfs_free_path(path); | |
936 | return ret; | |
937 | } | |
938 | ||
d8d5f3e1 CM |
939 | /* |
940 | * Back reference rules. Back refs have three main goals: | |
941 | * | |
942 | * 1) differentiate between all holders of references to an extent so that | |
943 | * when a reference is dropped we can make sure it was a valid reference | |
944 | * before freeing the extent. | |
945 | * | |
946 | * 2) Provide enough information to quickly find the holders of an extent | |
947 | * if we notice a given block is corrupted or bad. | |
948 | * | |
949 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
950 | * maintenance. This is actually the same as #2, but with a slightly | |
951 | * different use case. | |
952 | * | |
5d4f98a2 YZ |
953 | * There are two kinds of back refs. The implicit back refs is optimized |
954 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
955 | * back refs of this kind provide information about the block's owner tree | |
956 | * and the pointer's key. These information allow us to find the block by | |
957 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
958 | * referenced by their owner trees. The location of tree block is recorded | |
959 | * in the back refs. Actually the full back refs is generic, and can be | |
960 | * used in all cases the implicit back refs is used. The major shortcoming | |
961 | * of the full back refs is its overhead. Every time a tree block gets | |
962 | * COWed, we have to update back refs entry for all pointers in it. | |
963 | * | |
964 | * For a newly allocated tree block, we use implicit back refs for | |
965 | * pointers in it. This means most tree related operations only involve | |
966 | * implicit back refs. For a tree block created in old transaction, the | |
967 | * only way to drop a reference to it is COW it. So we can detect the | |
968 | * event that tree block loses its owner tree's reference and do the | |
969 | * back refs conversion. | |
970 | * | |
01327610 | 971 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
972 | * |
973 | * The reference count of the block is one and the tree is the block's | |
974 | * owner tree. Nothing to do in this case. | |
975 | * | |
976 | * The reference count of the block is one and the tree is not the | |
977 | * block's owner tree. In this case, full back refs is used for pointers | |
978 | * in the block. Remove these full back refs, add implicit back refs for | |
979 | * every pointers in the new block. | |
980 | * | |
981 | * The reference count of the block is greater than one and the tree is | |
982 | * the block's owner tree. In this case, implicit back refs is used for | |
983 | * pointers in the block. Add full back refs for every pointers in the | |
984 | * block, increase lower level extents' reference counts. The original | |
985 | * implicit back refs are entailed to the new block. | |
986 | * | |
987 | * The reference count of the block is greater than one and the tree is | |
988 | * not the block's owner tree. Add implicit back refs for every pointer in | |
989 | * the new block, increase lower level extents' reference count. | |
990 | * | |
991 | * Back Reference Key composing: | |
992 | * | |
993 | * The key objectid corresponds to the first byte in the extent, | |
994 | * The key type is used to differentiate between types of back refs. | |
995 | * There are different meanings of the key offset for different types | |
996 | * of back refs. | |
997 | * | |
d8d5f3e1 CM |
998 | * File extents can be referenced by: |
999 | * | |
1000 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1001 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1002 | * - different offsets inside a file (bookend extents in file.c) |
1003 | * | |
5d4f98a2 | 1004 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1005 | * |
1006 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1007 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1008 | * - original offset in the file |
1009 | * - how many bookend extents | |
d8d5f3e1 | 1010 | * |
5d4f98a2 YZ |
1011 | * The key offset for the implicit back refs is hash of the first |
1012 | * three fields. | |
d8d5f3e1 | 1013 | * |
5d4f98a2 | 1014 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1015 | * |
5d4f98a2 | 1016 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1017 | * |
5d4f98a2 YZ |
1018 | * The key offset for the implicit back refs is the first byte of |
1019 | * the tree leaf | |
d8d5f3e1 | 1020 | * |
5d4f98a2 YZ |
1021 | * When a file extent is allocated, The implicit back refs is used. |
1022 | * the fields are filled in: | |
d8d5f3e1 | 1023 | * |
5d4f98a2 | 1024 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1025 | * |
5d4f98a2 YZ |
1026 | * When a file extent is removed file truncation, we find the |
1027 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1028 | * |
5d4f98a2 | 1029 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1030 | * |
5d4f98a2 | 1031 | * Btree extents can be referenced by: |
d8d5f3e1 | 1032 | * |
5d4f98a2 | 1033 | * - Different subvolumes |
d8d5f3e1 | 1034 | * |
5d4f98a2 YZ |
1035 | * Both the implicit back refs and the full back refs for tree blocks |
1036 | * only consist of key. The key offset for the implicit back refs is | |
1037 | * objectid of block's owner tree. The key offset for the full back refs | |
1038 | * is the first byte of parent block. | |
d8d5f3e1 | 1039 | * |
5d4f98a2 YZ |
1040 | * When implicit back refs is used, information about the lowest key and |
1041 | * level of the tree block are required. These information are stored in | |
1042 | * tree block info structure. | |
d8d5f3e1 | 1043 | */ |
31840ae1 | 1044 | |
5d4f98a2 YZ |
1045 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1046 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
1047 | struct btrfs_root *root, | |
1048 | struct btrfs_path *path, | |
1049 | u64 owner, u32 extra_size) | |
7bb86316 | 1050 | { |
5d4f98a2 YZ |
1051 | struct btrfs_extent_item *item; |
1052 | struct btrfs_extent_item_v0 *ei0; | |
1053 | struct btrfs_extent_ref_v0 *ref0; | |
1054 | struct btrfs_tree_block_info *bi; | |
1055 | struct extent_buffer *leaf; | |
7bb86316 | 1056 | struct btrfs_key key; |
5d4f98a2 YZ |
1057 | struct btrfs_key found_key; |
1058 | u32 new_size = sizeof(*item); | |
1059 | u64 refs; | |
1060 | int ret; | |
1061 | ||
1062 | leaf = path->nodes[0]; | |
1063 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1064 | ||
1065 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1066 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1067 | struct btrfs_extent_item_v0); | |
1068 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1069 | ||
1070 | if (owner == (u64)-1) { | |
1071 | while (1) { | |
1072 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1073 | ret = btrfs_next_leaf(root, path); | |
1074 | if (ret < 0) | |
1075 | return ret; | |
79787eaa | 1076 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1077 | leaf = path->nodes[0]; |
1078 | } | |
1079 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1080 | path->slots[0]); | |
1081 | BUG_ON(key.objectid != found_key.objectid); | |
1082 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1083 | path->slots[0]++; | |
1084 | continue; | |
1085 | } | |
1086 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1087 | struct btrfs_extent_ref_v0); | |
1088 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1089 | break; | |
1090 | } | |
1091 | } | |
b3b4aa74 | 1092 | btrfs_release_path(path); |
5d4f98a2 YZ |
1093 | |
1094 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1095 | new_size += sizeof(*bi); | |
1096 | ||
1097 | new_size -= sizeof(*ei0); | |
1098 | ret = btrfs_search_slot(trans, root, &key, path, | |
1099 | new_size + extra_size, 1); | |
1100 | if (ret < 0) | |
1101 | return ret; | |
79787eaa | 1102 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1103 | |
4b90c680 | 1104 | btrfs_extend_item(root, path, new_size); |
5d4f98a2 YZ |
1105 | |
1106 | leaf = path->nodes[0]; | |
1107 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1108 | btrfs_set_extent_refs(leaf, item, refs); | |
1109 | /* FIXME: get real generation */ | |
1110 | btrfs_set_extent_generation(leaf, item, 0); | |
1111 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1112 | btrfs_set_extent_flags(leaf, item, | |
1113 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1114 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1115 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1116 | /* FIXME: get first key of the block */ | |
1117 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1118 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1119 | } else { | |
1120 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1121 | } | |
1122 | btrfs_mark_buffer_dirty(leaf); | |
1123 | return 0; | |
1124 | } | |
1125 | #endif | |
1126 | ||
1127 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1128 | { | |
1129 | u32 high_crc = ~(u32)0; | |
1130 | u32 low_crc = ~(u32)0; | |
1131 | __le64 lenum; | |
1132 | ||
1133 | lenum = cpu_to_le64(root_objectid); | |
14a958e6 | 1134 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1135 | lenum = cpu_to_le64(owner); |
14a958e6 | 1136 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1137 | lenum = cpu_to_le64(offset); |
14a958e6 | 1138 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1139 | |
1140 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1141 | } | |
1142 | ||
1143 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1144 | struct btrfs_extent_data_ref *ref) | |
1145 | { | |
1146 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1147 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1148 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1149 | } | |
1150 | ||
1151 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1152 | struct btrfs_extent_data_ref *ref, | |
1153 | u64 root_objectid, u64 owner, u64 offset) | |
1154 | { | |
1155 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1156 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1157 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1158 | return 0; | |
1159 | return 1; | |
1160 | } | |
1161 | ||
1162 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1163 | struct btrfs_root *root, | |
1164 | struct btrfs_path *path, | |
1165 | u64 bytenr, u64 parent, | |
1166 | u64 root_objectid, | |
1167 | u64 owner, u64 offset) | |
1168 | { | |
1169 | struct btrfs_key key; | |
1170 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1171 | struct extent_buffer *leaf; |
5d4f98a2 | 1172 | u32 nritems; |
74493f7a | 1173 | int ret; |
5d4f98a2 YZ |
1174 | int recow; |
1175 | int err = -ENOENT; | |
74493f7a | 1176 | |
31840ae1 | 1177 | key.objectid = bytenr; |
5d4f98a2 YZ |
1178 | if (parent) { |
1179 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1180 | key.offset = parent; | |
1181 | } else { | |
1182 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1183 | key.offset = hash_extent_data_ref(root_objectid, | |
1184 | owner, offset); | |
1185 | } | |
1186 | again: | |
1187 | recow = 0; | |
1188 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1189 | if (ret < 0) { | |
1190 | err = ret; | |
1191 | goto fail; | |
1192 | } | |
31840ae1 | 1193 | |
5d4f98a2 YZ |
1194 | if (parent) { |
1195 | if (!ret) | |
1196 | return 0; | |
1197 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1198 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1199 | btrfs_release_path(path); |
5d4f98a2 YZ |
1200 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1201 | if (ret < 0) { | |
1202 | err = ret; | |
1203 | goto fail; | |
1204 | } | |
1205 | if (!ret) | |
1206 | return 0; | |
1207 | #endif | |
1208 | goto fail; | |
31840ae1 ZY |
1209 | } |
1210 | ||
1211 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1212 | nritems = btrfs_header_nritems(leaf); |
1213 | while (1) { | |
1214 | if (path->slots[0] >= nritems) { | |
1215 | ret = btrfs_next_leaf(root, path); | |
1216 | if (ret < 0) | |
1217 | err = ret; | |
1218 | if (ret) | |
1219 | goto fail; | |
1220 | ||
1221 | leaf = path->nodes[0]; | |
1222 | nritems = btrfs_header_nritems(leaf); | |
1223 | recow = 1; | |
1224 | } | |
1225 | ||
1226 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1227 | if (key.objectid != bytenr || | |
1228 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1229 | goto fail; | |
1230 | ||
1231 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1232 | struct btrfs_extent_data_ref); | |
1233 | ||
1234 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1235 | owner, offset)) { | |
1236 | if (recow) { | |
b3b4aa74 | 1237 | btrfs_release_path(path); |
5d4f98a2 YZ |
1238 | goto again; |
1239 | } | |
1240 | err = 0; | |
1241 | break; | |
1242 | } | |
1243 | path->slots[0]++; | |
31840ae1 | 1244 | } |
5d4f98a2 YZ |
1245 | fail: |
1246 | return err; | |
31840ae1 ZY |
1247 | } |
1248 | ||
5d4f98a2 YZ |
1249 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1250 | struct btrfs_root *root, | |
1251 | struct btrfs_path *path, | |
1252 | u64 bytenr, u64 parent, | |
1253 | u64 root_objectid, u64 owner, | |
1254 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1255 | { |
1256 | struct btrfs_key key; | |
1257 | struct extent_buffer *leaf; | |
5d4f98a2 | 1258 | u32 size; |
31840ae1 ZY |
1259 | u32 num_refs; |
1260 | int ret; | |
74493f7a | 1261 | |
74493f7a | 1262 | key.objectid = bytenr; |
5d4f98a2 YZ |
1263 | if (parent) { |
1264 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1265 | key.offset = parent; | |
1266 | size = sizeof(struct btrfs_shared_data_ref); | |
1267 | } else { | |
1268 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1269 | key.offset = hash_extent_data_ref(root_objectid, | |
1270 | owner, offset); | |
1271 | size = sizeof(struct btrfs_extent_data_ref); | |
1272 | } | |
74493f7a | 1273 | |
5d4f98a2 YZ |
1274 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1275 | if (ret && ret != -EEXIST) | |
1276 | goto fail; | |
1277 | ||
1278 | leaf = path->nodes[0]; | |
1279 | if (parent) { | |
1280 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1281 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1282 | struct btrfs_shared_data_ref); |
1283 | if (ret == 0) { | |
1284 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1285 | } else { | |
1286 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1287 | num_refs += refs_to_add; | |
1288 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1289 | } |
5d4f98a2 YZ |
1290 | } else { |
1291 | struct btrfs_extent_data_ref *ref; | |
1292 | while (ret == -EEXIST) { | |
1293 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1294 | struct btrfs_extent_data_ref); | |
1295 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1296 | owner, offset)) | |
1297 | break; | |
b3b4aa74 | 1298 | btrfs_release_path(path); |
5d4f98a2 YZ |
1299 | key.offset++; |
1300 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1301 | size); | |
1302 | if (ret && ret != -EEXIST) | |
1303 | goto fail; | |
31840ae1 | 1304 | |
5d4f98a2 YZ |
1305 | leaf = path->nodes[0]; |
1306 | } | |
1307 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1308 | struct btrfs_extent_data_ref); | |
1309 | if (ret == 0) { | |
1310 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1311 | root_objectid); | |
1312 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1313 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1314 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1315 | } else { | |
1316 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1317 | num_refs += refs_to_add; | |
1318 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1319 | } |
31840ae1 | 1320 | } |
5d4f98a2 YZ |
1321 | btrfs_mark_buffer_dirty(leaf); |
1322 | ret = 0; | |
1323 | fail: | |
b3b4aa74 | 1324 | btrfs_release_path(path); |
7bb86316 | 1325 | return ret; |
74493f7a CM |
1326 | } |
1327 | ||
5d4f98a2 YZ |
1328 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1329 | struct btrfs_root *root, | |
1330 | struct btrfs_path *path, | |
fcebe456 | 1331 | int refs_to_drop, int *last_ref) |
31840ae1 | 1332 | { |
5d4f98a2 YZ |
1333 | struct btrfs_key key; |
1334 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1335 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1336 | struct extent_buffer *leaf; |
5d4f98a2 | 1337 | u32 num_refs = 0; |
31840ae1 ZY |
1338 | int ret = 0; |
1339 | ||
1340 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1341 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1342 | ||
1343 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1344 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1345 | struct btrfs_extent_data_ref); | |
1346 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1347 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1348 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1349 | struct btrfs_shared_data_ref); | |
1350 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1351 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1352 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1353 | struct btrfs_extent_ref_v0 *ref0; | |
1354 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1355 | struct btrfs_extent_ref_v0); | |
1356 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1357 | #endif | |
1358 | } else { | |
1359 | BUG(); | |
1360 | } | |
1361 | ||
56bec294 CM |
1362 | BUG_ON(num_refs < refs_to_drop); |
1363 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1364 | |
31840ae1 ZY |
1365 | if (num_refs == 0) { |
1366 | ret = btrfs_del_item(trans, root, path); | |
fcebe456 | 1367 | *last_ref = 1; |
31840ae1 | 1368 | } else { |
5d4f98a2 YZ |
1369 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1370 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1371 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1372 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1373 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1374 | else { | |
1375 | struct btrfs_extent_ref_v0 *ref0; | |
1376 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1377 | struct btrfs_extent_ref_v0); | |
1378 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1379 | } | |
1380 | #endif | |
31840ae1 ZY |
1381 | btrfs_mark_buffer_dirty(leaf); |
1382 | } | |
31840ae1 ZY |
1383 | return ret; |
1384 | } | |
1385 | ||
9ed0dea0 | 1386 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1387 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1388 | { |
5d4f98a2 YZ |
1389 | struct btrfs_key key; |
1390 | struct extent_buffer *leaf; | |
1391 | struct btrfs_extent_data_ref *ref1; | |
1392 | struct btrfs_shared_data_ref *ref2; | |
1393 | u32 num_refs = 0; | |
1394 | ||
1395 | leaf = path->nodes[0]; | |
1396 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1397 | if (iref) { | |
1398 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1399 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1400 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1401 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1402 | } else { | |
1403 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1404 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1405 | } | |
1406 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1407 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1408 | struct btrfs_extent_data_ref); | |
1409 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1410 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1411 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1412 | struct btrfs_shared_data_ref); | |
1413 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1414 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1415 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1416 | struct btrfs_extent_ref_v0 *ref0; | |
1417 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1418 | struct btrfs_extent_ref_v0); | |
1419 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1420 | #endif |
5d4f98a2 YZ |
1421 | } else { |
1422 | WARN_ON(1); | |
1423 | } | |
1424 | return num_refs; | |
1425 | } | |
15916de8 | 1426 | |
5d4f98a2 YZ |
1427 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1428 | struct btrfs_root *root, | |
1429 | struct btrfs_path *path, | |
1430 | u64 bytenr, u64 parent, | |
1431 | u64 root_objectid) | |
1f3c79a2 | 1432 | { |
5d4f98a2 | 1433 | struct btrfs_key key; |
1f3c79a2 | 1434 | int ret; |
1f3c79a2 | 1435 | |
5d4f98a2 YZ |
1436 | key.objectid = bytenr; |
1437 | if (parent) { | |
1438 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1439 | key.offset = parent; | |
1440 | } else { | |
1441 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1442 | key.offset = root_objectid; | |
1f3c79a2 LH |
1443 | } |
1444 | ||
5d4f98a2 YZ |
1445 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1446 | if (ret > 0) | |
1447 | ret = -ENOENT; | |
1448 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1449 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1450 | btrfs_release_path(path); |
5d4f98a2 YZ |
1451 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1452 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1453 | if (ret > 0) | |
1454 | ret = -ENOENT; | |
1455 | } | |
1f3c79a2 | 1456 | #endif |
5d4f98a2 | 1457 | return ret; |
1f3c79a2 LH |
1458 | } |
1459 | ||
5d4f98a2 YZ |
1460 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1461 | struct btrfs_root *root, | |
1462 | struct btrfs_path *path, | |
1463 | u64 bytenr, u64 parent, | |
1464 | u64 root_objectid) | |
31840ae1 | 1465 | { |
5d4f98a2 | 1466 | struct btrfs_key key; |
31840ae1 | 1467 | int ret; |
31840ae1 | 1468 | |
5d4f98a2 YZ |
1469 | key.objectid = bytenr; |
1470 | if (parent) { | |
1471 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1472 | key.offset = parent; | |
1473 | } else { | |
1474 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1475 | key.offset = root_objectid; | |
1476 | } | |
1477 | ||
1478 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1479 | btrfs_release_path(path); |
31840ae1 ZY |
1480 | return ret; |
1481 | } | |
1482 | ||
5d4f98a2 | 1483 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1484 | { |
5d4f98a2 YZ |
1485 | int type; |
1486 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1487 | if (parent > 0) | |
1488 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1489 | else | |
1490 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1491 | } else { | |
1492 | if (parent > 0) | |
1493 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1494 | else | |
1495 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1496 | } | |
1497 | return type; | |
31840ae1 | 1498 | } |
56bec294 | 1499 | |
2c47e605 YZ |
1500 | static int find_next_key(struct btrfs_path *path, int level, |
1501 | struct btrfs_key *key) | |
56bec294 | 1502 | |
02217ed2 | 1503 | { |
2c47e605 | 1504 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1505 | if (!path->nodes[level]) |
1506 | break; | |
5d4f98a2 YZ |
1507 | if (path->slots[level] + 1 >= |
1508 | btrfs_header_nritems(path->nodes[level])) | |
1509 | continue; | |
1510 | if (level == 0) | |
1511 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1512 | path->slots[level] + 1); | |
1513 | else | |
1514 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1515 | path->slots[level] + 1); | |
1516 | return 0; | |
1517 | } | |
1518 | return 1; | |
1519 | } | |
037e6390 | 1520 | |
5d4f98a2 YZ |
1521 | /* |
1522 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1523 | * to the address of inline back ref, and 0 is returned. | |
1524 | * | |
1525 | * if back ref isn't found, *ref_ret is set to the address where it | |
1526 | * should be inserted, and -ENOENT is returned. | |
1527 | * | |
1528 | * if insert is true and there are too many inline back refs, the path | |
1529 | * points to the extent item, and -EAGAIN is returned. | |
1530 | * | |
1531 | * NOTE: inline back refs are ordered in the same way that back ref | |
1532 | * items in the tree are ordered. | |
1533 | */ | |
1534 | static noinline_for_stack | |
1535 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1536 | struct btrfs_root *root, | |
1537 | struct btrfs_path *path, | |
1538 | struct btrfs_extent_inline_ref **ref_ret, | |
1539 | u64 bytenr, u64 num_bytes, | |
1540 | u64 parent, u64 root_objectid, | |
1541 | u64 owner, u64 offset, int insert) | |
1542 | { | |
1543 | struct btrfs_key key; | |
1544 | struct extent_buffer *leaf; | |
1545 | struct btrfs_extent_item *ei; | |
1546 | struct btrfs_extent_inline_ref *iref; | |
1547 | u64 flags; | |
1548 | u64 item_size; | |
1549 | unsigned long ptr; | |
1550 | unsigned long end; | |
1551 | int extra_size; | |
1552 | int type; | |
1553 | int want; | |
1554 | int ret; | |
1555 | int err = 0; | |
3173a18f JB |
1556 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
1557 | SKINNY_METADATA); | |
26b8003f | 1558 | |
db94535d | 1559 | key.objectid = bytenr; |
31840ae1 | 1560 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1561 | key.offset = num_bytes; |
31840ae1 | 1562 | |
5d4f98a2 YZ |
1563 | want = extent_ref_type(parent, owner); |
1564 | if (insert) { | |
1565 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1566 | path->keep_locks = 1; |
5d4f98a2 YZ |
1567 | } else |
1568 | extra_size = -1; | |
3173a18f JB |
1569 | |
1570 | /* | |
1571 | * Owner is our parent level, so we can just add one to get the level | |
1572 | * for the block we are interested in. | |
1573 | */ | |
1574 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1575 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1576 | key.offset = owner; | |
1577 | } | |
1578 | ||
1579 | again: | |
5d4f98a2 | 1580 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1581 | if (ret < 0) { |
5d4f98a2 YZ |
1582 | err = ret; |
1583 | goto out; | |
1584 | } | |
3173a18f JB |
1585 | |
1586 | /* | |
1587 | * We may be a newly converted file system which still has the old fat | |
1588 | * extent entries for metadata, so try and see if we have one of those. | |
1589 | */ | |
1590 | if (ret > 0 && skinny_metadata) { | |
1591 | skinny_metadata = false; | |
1592 | if (path->slots[0]) { | |
1593 | path->slots[0]--; | |
1594 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1595 | path->slots[0]); | |
1596 | if (key.objectid == bytenr && | |
1597 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1598 | key.offset == num_bytes) | |
1599 | ret = 0; | |
1600 | } | |
1601 | if (ret) { | |
9ce49a0b | 1602 | key.objectid = bytenr; |
3173a18f JB |
1603 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1604 | key.offset = num_bytes; | |
1605 | btrfs_release_path(path); | |
1606 | goto again; | |
1607 | } | |
1608 | } | |
1609 | ||
79787eaa JM |
1610 | if (ret && !insert) { |
1611 | err = -ENOENT; | |
1612 | goto out; | |
fae7f21c | 1613 | } else if (WARN_ON(ret)) { |
492104c8 | 1614 | err = -EIO; |
492104c8 | 1615 | goto out; |
79787eaa | 1616 | } |
5d4f98a2 YZ |
1617 | |
1618 | leaf = path->nodes[0]; | |
1619 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1620 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1621 | if (item_size < sizeof(*ei)) { | |
1622 | if (!insert) { | |
1623 | err = -ENOENT; | |
1624 | goto out; | |
1625 | } | |
1626 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1627 | extra_size); | |
1628 | if (ret < 0) { | |
1629 | err = ret; | |
1630 | goto out; | |
1631 | } | |
1632 | leaf = path->nodes[0]; | |
1633 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1634 | } | |
1635 | #endif | |
1636 | BUG_ON(item_size < sizeof(*ei)); | |
1637 | ||
5d4f98a2 YZ |
1638 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1639 | flags = btrfs_extent_flags(leaf, ei); | |
1640 | ||
1641 | ptr = (unsigned long)(ei + 1); | |
1642 | end = (unsigned long)ei + item_size; | |
1643 | ||
3173a18f | 1644 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1645 | ptr += sizeof(struct btrfs_tree_block_info); |
1646 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1647 | } |
1648 | ||
1649 | err = -ENOENT; | |
1650 | while (1) { | |
1651 | if (ptr >= end) { | |
1652 | WARN_ON(ptr > end); | |
1653 | break; | |
1654 | } | |
1655 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1656 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1657 | if (want < type) | |
1658 | break; | |
1659 | if (want > type) { | |
1660 | ptr += btrfs_extent_inline_ref_size(type); | |
1661 | continue; | |
1662 | } | |
1663 | ||
1664 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1665 | struct btrfs_extent_data_ref *dref; | |
1666 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1667 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1668 | owner, offset)) { | |
1669 | err = 0; | |
1670 | break; | |
1671 | } | |
1672 | if (hash_extent_data_ref_item(leaf, dref) < | |
1673 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1674 | break; | |
1675 | } else { | |
1676 | u64 ref_offset; | |
1677 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1678 | if (parent > 0) { | |
1679 | if (parent == ref_offset) { | |
1680 | err = 0; | |
1681 | break; | |
1682 | } | |
1683 | if (ref_offset < parent) | |
1684 | break; | |
1685 | } else { | |
1686 | if (root_objectid == ref_offset) { | |
1687 | err = 0; | |
1688 | break; | |
1689 | } | |
1690 | if (ref_offset < root_objectid) | |
1691 | break; | |
1692 | } | |
1693 | } | |
1694 | ptr += btrfs_extent_inline_ref_size(type); | |
1695 | } | |
1696 | if (err == -ENOENT && insert) { | |
1697 | if (item_size + extra_size >= | |
1698 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1699 | err = -EAGAIN; | |
1700 | goto out; | |
1701 | } | |
1702 | /* | |
1703 | * To add new inline back ref, we have to make sure | |
1704 | * there is no corresponding back ref item. | |
1705 | * For simplicity, we just do not add new inline back | |
1706 | * ref if there is any kind of item for this block | |
1707 | */ | |
2c47e605 YZ |
1708 | if (find_next_key(path, 0, &key) == 0 && |
1709 | key.objectid == bytenr && | |
85d4198e | 1710 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1711 | err = -EAGAIN; |
1712 | goto out; | |
1713 | } | |
1714 | } | |
1715 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1716 | out: | |
85d4198e | 1717 | if (insert) { |
5d4f98a2 YZ |
1718 | path->keep_locks = 0; |
1719 | btrfs_unlock_up_safe(path, 1); | |
1720 | } | |
1721 | return err; | |
1722 | } | |
1723 | ||
1724 | /* | |
1725 | * helper to add new inline back ref | |
1726 | */ | |
1727 | static noinline_for_stack | |
fd279fae | 1728 | void setup_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1729 | struct btrfs_path *path, |
1730 | struct btrfs_extent_inline_ref *iref, | |
1731 | u64 parent, u64 root_objectid, | |
1732 | u64 owner, u64 offset, int refs_to_add, | |
1733 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1734 | { |
1735 | struct extent_buffer *leaf; | |
1736 | struct btrfs_extent_item *ei; | |
1737 | unsigned long ptr; | |
1738 | unsigned long end; | |
1739 | unsigned long item_offset; | |
1740 | u64 refs; | |
1741 | int size; | |
1742 | int type; | |
5d4f98a2 YZ |
1743 | |
1744 | leaf = path->nodes[0]; | |
1745 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1746 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1747 | ||
1748 | type = extent_ref_type(parent, owner); | |
1749 | size = btrfs_extent_inline_ref_size(type); | |
1750 | ||
4b90c680 | 1751 | btrfs_extend_item(root, path, size); |
5d4f98a2 YZ |
1752 | |
1753 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1754 | refs = btrfs_extent_refs(leaf, ei); | |
1755 | refs += refs_to_add; | |
1756 | btrfs_set_extent_refs(leaf, ei, refs); | |
1757 | if (extent_op) | |
1758 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1759 | ||
1760 | ptr = (unsigned long)ei + item_offset; | |
1761 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1762 | if (ptr < end - size) | |
1763 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1764 | end - size - ptr); | |
1765 | ||
1766 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1767 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1768 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1769 | struct btrfs_extent_data_ref *dref; | |
1770 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1771 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1772 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1773 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1774 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1775 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1776 | struct btrfs_shared_data_ref *sref; | |
1777 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1778 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1779 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1780 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1781 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1782 | } else { | |
1783 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1784 | } | |
1785 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1786 | } |
1787 | ||
1788 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1789 | struct btrfs_root *root, | |
1790 | struct btrfs_path *path, | |
1791 | struct btrfs_extent_inline_ref **ref_ret, | |
1792 | u64 bytenr, u64 num_bytes, u64 parent, | |
1793 | u64 root_objectid, u64 owner, u64 offset) | |
1794 | { | |
1795 | int ret; | |
1796 | ||
1797 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1798 | bytenr, num_bytes, parent, | |
1799 | root_objectid, owner, offset, 0); | |
1800 | if (ret != -ENOENT) | |
54aa1f4d | 1801 | return ret; |
5d4f98a2 | 1802 | |
b3b4aa74 | 1803 | btrfs_release_path(path); |
5d4f98a2 YZ |
1804 | *ref_ret = NULL; |
1805 | ||
1806 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1807 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1808 | root_objectid); | |
1809 | } else { | |
1810 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1811 | root_objectid, owner, offset); | |
b9473439 | 1812 | } |
5d4f98a2 YZ |
1813 | return ret; |
1814 | } | |
31840ae1 | 1815 | |
5d4f98a2 YZ |
1816 | /* |
1817 | * helper to update/remove inline back ref | |
1818 | */ | |
1819 | static noinline_for_stack | |
afe5fea7 | 1820 | void update_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1821 | struct btrfs_path *path, |
1822 | struct btrfs_extent_inline_ref *iref, | |
1823 | int refs_to_mod, | |
fcebe456 JB |
1824 | struct btrfs_delayed_extent_op *extent_op, |
1825 | int *last_ref) | |
5d4f98a2 YZ |
1826 | { |
1827 | struct extent_buffer *leaf; | |
1828 | struct btrfs_extent_item *ei; | |
1829 | struct btrfs_extent_data_ref *dref = NULL; | |
1830 | struct btrfs_shared_data_ref *sref = NULL; | |
1831 | unsigned long ptr; | |
1832 | unsigned long end; | |
1833 | u32 item_size; | |
1834 | int size; | |
1835 | int type; | |
5d4f98a2 YZ |
1836 | u64 refs; |
1837 | ||
1838 | leaf = path->nodes[0]; | |
1839 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1840 | refs = btrfs_extent_refs(leaf, ei); | |
1841 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1842 | refs += refs_to_mod; | |
1843 | btrfs_set_extent_refs(leaf, ei, refs); | |
1844 | if (extent_op) | |
1845 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1846 | ||
1847 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1848 | ||
1849 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1850 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1851 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1852 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1853 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1854 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1855 | } else { | |
1856 | refs = 1; | |
1857 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1858 | } |
31840ae1 | 1859 | |
5d4f98a2 YZ |
1860 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1861 | refs += refs_to_mod; | |
1862 | ||
1863 | if (refs > 0) { | |
1864 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1865 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1866 | else | |
1867 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1868 | } else { | |
fcebe456 | 1869 | *last_ref = 1; |
5d4f98a2 YZ |
1870 | size = btrfs_extent_inline_ref_size(type); |
1871 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1872 | ptr = (unsigned long)iref; | |
1873 | end = (unsigned long)ei + item_size; | |
1874 | if (ptr + size < end) | |
1875 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1876 | end - ptr - size); | |
1877 | item_size -= size; | |
afe5fea7 | 1878 | btrfs_truncate_item(root, path, item_size, 1); |
5d4f98a2 YZ |
1879 | } |
1880 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1881 | } |
1882 | ||
1883 | static noinline_for_stack | |
1884 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1885 | struct btrfs_root *root, | |
1886 | struct btrfs_path *path, | |
1887 | u64 bytenr, u64 num_bytes, u64 parent, | |
1888 | u64 root_objectid, u64 owner, | |
1889 | u64 offset, int refs_to_add, | |
1890 | struct btrfs_delayed_extent_op *extent_op) | |
1891 | { | |
1892 | struct btrfs_extent_inline_ref *iref; | |
1893 | int ret; | |
1894 | ||
1895 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1896 | bytenr, num_bytes, parent, | |
1897 | root_objectid, owner, offset, 1); | |
1898 | if (ret == 0) { | |
1899 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
afe5fea7 | 1900 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1901 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1902 | } else if (ret == -ENOENT) { |
fd279fae | 1903 | setup_inline_extent_backref(root, path, iref, parent, |
143bede5 JM |
1904 | root_objectid, owner, offset, |
1905 | refs_to_add, extent_op); | |
1906 | ret = 0; | |
771ed689 | 1907 | } |
5d4f98a2 YZ |
1908 | return ret; |
1909 | } | |
31840ae1 | 1910 | |
5d4f98a2 YZ |
1911 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1912 | struct btrfs_root *root, | |
1913 | struct btrfs_path *path, | |
1914 | u64 bytenr, u64 parent, u64 root_objectid, | |
1915 | u64 owner, u64 offset, int refs_to_add) | |
1916 | { | |
1917 | int ret; | |
1918 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1919 | BUG_ON(refs_to_add != 1); | |
1920 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1921 | parent, root_objectid); | |
1922 | } else { | |
1923 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1924 | parent, root_objectid, | |
1925 | owner, offset, refs_to_add); | |
1926 | } | |
1927 | return ret; | |
1928 | } | |
56bec294 | 1929 | |
5d4f98a2 YZ |
1930 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1931 | struct btrfs_root *root, | |
1932 | struct btrfs_path *path, | |
1933 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1934 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1935 | { |
143bede5 | 1936 | int ret = 0; |
b9473439 | 1937 | |
5d4f98a2 YZ |
1938 | BUG_ON(!is_data && refs_to_drop != 1); |
1939 | if (iref) { | |
afe5fea7 | 1940 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1941 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 1942 | } else if (is_data) { |
fcebe456 JB |
1943 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop, |
1944 | last_ref); | |
5d4f98a2 | 1945 | } else { |
fcebe456 | 1946 | *last_ref = 1; |
5d4f98a2 YZ |
1947 | ret = btrfs_del_item(trans, root, path); |
1948 | } | |
1949 | return ret; | |
1950 | } | |
1951 | ||
86557861 | 1952 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
1953 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1954 | u64 *discarded_bytes) | |
5d4f98a2 | 1955 | { |
86557861 JM |
1956 | int j, ret = 0; |
1957 | u64 bytes_left, end; | |
4d89d377 | 1958 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1959 | |
4d89d377 JM |
1960 | if (WARN_ON(start != aligned_start)) { |
1961 | len -= aligned_start - start; | |
1962 | len = round_down(len, 1 << 9); | |
1963 | start = aligned_start; | |
1964 | } | |
d04c6b88 | 1965 | |
4d89d377 | 1966 | *discarded_bytes = 0; |
86557861 JM |
1967 | |
1968 | if (!len) | |
1969 | return 0; | |
1970 | ||
1971 | end = start + len; | |
1972 | bytes_left = len; | |
1973 | ||
1974 | /* Skip any superblocks on this device. */ | |
1975 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1976 | u64 sb_start = btrfs_sb_offset(j); | |
1977 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1978 | u64 size = sb_start - start; | |
1979 | ||
1980 | if (!in_range(sb_start, start, bytes_left) && | |
1981 | !in_range(sb_end, start, bytes_left) && | |
1982 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1983 | continue; | |
1984 | ||
1985 | /* | |
1986 | * Superblock spans beginning of range. Adjust start and | |
1987 | * try again. | |
1988 | */ | |
1989 | if (sb_start <= start) { | |
1990 | start += sb_end - start; | |
1991 | if (start > end) { | |
1992 | bytes_left = 0; | |
1993 | break; | |
1994 | } | |
1995 | bytes_left = end - start; | |
1996 | continue; | |
1997 | } | |
1998 | ||
1999 | if (size) { | |
2000 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2001 | GFP_NOFS, 0); | |
2002 | if (!ret) | |
2003 | *discarded_bytes += size; | |
2004 | else if (ret != -EOPNOTSUPP) | |
2005 | return ret; | |
2006 | } | |
2007 | ||
2008 | start = sb_end; | |
2009 | if (start > end) { | |
2010 | bytes_left = 0; | |
2011 | break; | |
2012 | } | |
2013 | bytes_left = end - start; | |
2014 | } | |
2015 | ||
2016 | if (bytes_left) { | |
2017 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2018 | GFP_NOFS, 0); |
2019 | if (!ret) | |
86557861 | 2020 | *discarded_bytes += bytes_left; |
4d89d377 | 2021 | } |
d04c6b88 | 2022 | return ret; |
5d4f98a2 | 2023 | } |
5d4f98a2 | 2024 | |
1edb647b FM |
2025 | int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, |
2026 | u64 num_bytes, u64 *actual_bytes) | |
5d4f98a2 | 2027 | { |
5d4f98a2 | 2028 | int ret; |
5378e607 | 2029 | u64 discarded_bytes = 0; |
a1d3c478 | 2030 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2031 | |
e244a0ae | 2032 | |
2999241d FM |
2033 | /* |
2034 | * Avoid races with device replace and make sure our bbio has devices | |
2035 | * associated to its stripes that don't go away while we are discarding. | |
2036 | */ | |
2037 | btrfs_bio_counter_inc_blocked(root->fs_info); | |
5d4f98a2 | 2038 | /* Tell the block device(s) that the sectors can be discarded */ |
b3d3fa51 | 2039 | ret = btrfs_map_block(root->fs_info, REQ_OP_DISCARD, |
a1d3c478 | 2040 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 2041 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2042 | if (!ret) { |
a1d3c478 | 2043 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2044 | int i; |
2045 | ||
5d4f98a2 | 2046 | |
a1d3c478 | 2047 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2048 | u64 bytes; |
d5e2003c JB |
2049 | if (!stripe->dev->can_discard) |
2050 | continue; | |
2051 | ||
5378e607 LD |
2052 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2053 | stripe->physical, | |
d04c6b88 JM |
2054 | stripe->length, |
2055 | &bytes); | |
5378e607 | 2056 | if (!ret) |
d04c6b88 | 2057 | discarded_bytes += bytes; |
5378e607 | 2058 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2059 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2060 | |
2061 | /* | |
2062 | * Just in case we get back EOPNOTSUPP for some reason, | |
2063 | * just ignore the return value so we don't screw up | |
2064 | * people calling discard_extent. | |
2065 | */ | |
2066 | ret = 0; | |
5d4f98a2 | 2067 | } |
6e9606d2 | 2068 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2069 | } |
2999241d | 2070 | btrfs_bio_counter_dec(root->fs_info); |
5378e607 LD |
2071 | |
2072 | if (actual_bytes) | |
2073 | *actual_bytes = discarded_bytes; | |
2074 | ||
5d4f98a2 | 2075 | |
53b381b3 DW |
2076 | if (ret == -EOPNOTSUPP) |
2077 | ret = 0; | |
5d4f98a2 | 2078 | return ret; |
5d4f98a2 YZ |
2079 | } |
2080 | ||
79787eaa | 2081 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
2082 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
2083 | struct btrfs_root *root, | |
2084 | u64 bytenr, u64 num_bytes, u64 parent, | |
b06c4bf5 | 2085 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
2086 | { |
2087 | int ret; | |
66d7e7f0 AJ |
2088 | struct btrfs_fs_info *fs_info = root->fs_info; |
2089 | ||
5d4f98a2 YZ |
2090 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2091 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2092 | ||
2093 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
2094 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
2095 | num_bytes, | |
5d4f98a2 | 2096 | parent, root_objectid, (int)owner, |
b06c4bf5 | 2097 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 | 2098 | } else { |
66d7e7f0 | 2099 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5846a3c2 QW |
2100 | num_bytes, parent, root_objectid, |
2101 | owner, offset, 0, | |
b06c4bf5 | 2102 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 YZ |
2103 | } |
2104 | return ret; | |
2105 | } | |
2106 | ||
2107 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2108 | struct btrfs_root *root, | |
c682f9b3 | 2109 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2110 | u64 parent, u64 root_objectid, |
2111 | u64 owner, u64 offset, int refs_to_add, | |
2112 | struct btrfs_delayed_extent_op *extent_op) | |
2113 | { | |
fcebe456 | 2114 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
2115 | struct btrfs_path *path; |
2116 | struct extent_buffer *leaf; | |
2117 | struct btrfs_extent_item *item; | |
fcebe456 | 2118 | struct btrfs_key key; |
c682f9b3 QW |
2119 | u64 bytenr = node->bytenr; |
2120 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2121 | u64 refs; |
2122 | int ret; | |
5d4f98a2 YZ |
2123 | |
2124 | path = btrfs_alloc_path(); | |
2125 | if (!path) | |
2126 | return -ENOMEM; | |
2127 | ||
e4058b54 | 2128 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2129 | path->leave_spinning = 1; |
2130 | /* this will setup the path even if it fails to insert the back ref */ | |
fcebe456 JB |
2131 | ret = insert_inline_extent_backref(trans, fs_info->extent_root, path, |
2132 | bytenr, num_bytes, parent, | |
5d4f98a2 YZ |
2133 | root_objectid, owner, offset, |
2134 | refs_to_add, extent_op); | |
0ed4792a | 2135 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2136 | goto out; |
fcebe456 JB |
2137 | |
2138 | /* | |
2139 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2140 | * inline extent ref, so just update the reference count and add a | |
2141 | * normal backref. | |
2142 | */ | |
5d4f98a2 | 2143 | leaf = path->nodes[0]; |
fcebe456 | 2144 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2145 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2146 | refs = btrfs_extent_refs(leaf, item); | |
2147 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2148 | if (extent_op) | |
2149 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2150 | |
5d4f98a2 | 2151 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2152 | btrfs_release_path(path); |
56bec294 | 2153 | |
e4058b54 | 2154 | path->reada = READA_FORWARD; |
b9473439 | 2155 | path->leave_spinning = 1; |
56bec294 CM |
2156 | /* now insert the actual backref */ |
2157 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
2158 | path, bytenr, parent, root_objectid, |
2159 | owner, offset, refs_to_add); | |
79787eaa | 2160 | if (ret) |
66642832 | 2161 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2162 | out: |
56bec294 | 2163 | btrfs_free_path(path); |
30d133fc | 2164 | return ret; |
56bec294 CM |
2165 | } |
2166 | ||
5d4f98a2 YZ |
2167 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2168 | struct btrfs_root *root, | |
2169 | struct btrfs_delayed_ref_node *node, | |
2170 | struct btrfs_delayed_extent_op *extent_op, | |
2171 | int insert_reserved) | |
56bec294 | 2172 | { |
5d4f98a2 YZ |
2173 | int ret = 0; |
2174 | struct btrfs_delayed_data_ref *ref; | |
2175 | struct btrfs_key ins; | |
2176 | u64 parent = 0; | |
2177 | u64 ref_root = 0; | |
2178 | u64 flags = 0; | |
2179 | ||
2180 | ins.objectid = node->bytenr; | |
2181 | ins.offset = node->num_bytes; | |
2182 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2183 | ||
2184 | ref = btrfs_delayed_node_to_data_ref(node); | |
bc074524 | 2185 | trace_run_delayed_data_ref(root->fs_info, node, ref, node->action); |
599c75ec | 2186 | |
5d4f98a2 YZ |
2187 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2188 | parent = ref->parent; | |
fcebe456 | 2189 | ref_root = ref->root; |
5d4f98a2 YZ |
2190 | |
2191 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2192 | if (extent_op) |
5d4f98a2 | 2193 | flags |= extent_op->flags_to_set; |
5d4f98a2 YZ |
2194 | ret = alloc_reserved_file_extent(trans, root, |
2195 | parent, ref_root, flags, | |
2196 | ref->objectid, ref->offset, | |
2197 | &ins, node->ref_mod); | |
5d4f98a2 | 2198 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
c682f9b3 | 2199 | ret = __btrfs_inc_extent_ref(trans, root, node, parent, |
5d4f98a2 YZ |
2200 | ref_root, ref->objectid, |
2201 | ref->offset, node->ref_mod, | |
c682f9b3 | 2202 | extent_op); |
5d4f98a2 | 2203 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
c682f9b3 | 2204 | ret = __btrfs_free_extent(trans, root, node, parent, |
5d4f98a2 YZ |
2205 | ref_root, ref->objectid, |
2206 | ref->offset, node->ref_mod, | |
c682f9b3 | 2207 | extent_op); |
5d4f98a2 YZ |
2208 | } else { |
2209 | BUG(); | |
2210 | } | |
2211 | return ret; | |
2212 | } | |
2213 | ||
2214 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2215 | struct extent_buffer *leaf, | |
2216 | struct btrfs_extent_item *ei) | |
2217 | { | |
2218 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2219 | if (extent_op->update_flags) { | |
2220 | flags |= extent_op->flags_to_set; | |
2221 | btrfs_set_extent_flags(leaf, ei, flags); | |
2222 | } | |
2223 | ||
2224 | if (extent_op->update_key) { | |
2225 | struct btrfs_tree_block_info *bi; | |
2226 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2227 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2228 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2229 | } | |
2230 | } | |
2231 | ||
2232 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2233 | struct btrfs_root *root, | |
2234 | struct btrfs_delayed_ref_node *node, | |
2235 | struct btrfs_delayed_extent_op *extent_op) | |
2236 | { | |
2237 | struct btrfs_key key; | |
2238 | struct btrfs_path *path; | |
2239 | struct btrfs_extent_item *ei; | |
2240 | struct extent_buffer *leaf; | |
2241 | u32 item_size; | |
56bec294 | 2242 | int ret; |
5d4f98a2 | 2243 | int err = 0; |
b1c79e09 | 2244 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2245 | |
79787eaa JM |
2246 | if (trans->aborted) |
2247 | return 0; | |
2248 | ||
3173a18f JB |
2249 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
2250 | metadata = 0; | |
2251 | ||
5d4f98a2 YZ |
2252 | path = btrfs_alloc_path(); |
2253 | if (!path) | |
2254 | return -ENOMEM; | |
2255 | ||
2256 | key.objectid = node->bytenr; | |
5d4f98a2 | 2257 | |
3173a18f | 2258 | if (metadata) { |
3173a18f | 2259 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2260 | key.offset = extent_op->level; |
3173a18f JB |
2261 | } else { |
2262 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2263 | key.offset = node->num_bytes; | |
2264 | } | |
2265 | ||
2266 | again: | |
e4058b54 | 2267 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2268 | path->leave_spinning = 1; |
2269 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2270 | path, 0, 1); | |
2271 | if (ret < 0) { | |
2272 | err = ret; | |
2273 | goto out; | |
2274 | } | |
2275 | if (ret > 0) { | |
3173a18f | 2276 | if (metadata) { |
55994887 FDBM |
2277 | if (path->slots[0] > 0) { |
2278 | path->slots[0]--; | |
2279 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2280 | path->slots[0]); | |
2281 | if (key.objectid == node->bytenr && | |
2282 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2283 | key.offset == node->num_bytes) | |
2284 | ret = 0; | |
2285 | } | |
2286 | if (ret > 0) { | |
2287 | btrfs_release_path(path); | |
2288 | metadata = 0; | |
3173a18f | 2289 | |
55994887 FDBM |
2290 | key.objectid = node->bytenr; |
2291 | key.offset = node->num_bytes; | |
2292 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2293 | goto again; | |
2294 | } | |
2295 | } else { | |
2296 | err = -EIO; | |
2297 | goto out; | |
3173a18f | 2298 | } |
5d4f98a2 YZ |
2299 | } |
2300 | ||
2301 | leaf = path->nodes[0]; | |
2302 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2303 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2304 | if (item_size < sizeof(*ei)) { | |
2305 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2306 | path, (u64)-1, 0); | |
2307 | if (ret < 0) { | |
2308 | err = ret; | |
2309 | goto out; | |
2310 | } | |
2311 | leaf = path->nodes[0]; | |
2312 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2313 | } | |
2314 | #endif | |
2315 | BUG_ON(item_size < sizeof(*ei)); | |
2316 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2317 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2318 | |
5d4f98a2 YZ |
2319 | btrfs_mark_buffer_dirty(leaf); |
2320 | out: | |
2321 | btrfs_free_path(path); | |
2322 | return err; | |
56bec294 CM |
2323 | } |
2324 | ||
5d4f98a2 YZ |
2325 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2326 | struct btrfs_root *root, | |
2327 | struct btrfs_delayed_ref_node *node, | |
2328 | struct btrfs_delayed_extent_op *extent_op, | |
2329 | int insert_reserved) | |
56bec294 CM |
2330 | { |
2331 | int ret = 0; | |
5d4f98a2 YZ |
2332 | struct btrfs_delayed_tree_ref *ref; |
2333 | struct btrfs_key ins; | |
2334 | u64 parent = 0; | |
2335 | u64 ref_root = 0; | |
3173a18f JB |
2336 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
2337 | SKINNY_METADATA); | |
56bec294 | 2338 | |
5d4f98a2 | 2339 | ref = btrfs_delayed_node_to_tree_ref(node); |
bc074524 | 2340 | trace_run_delayed_tree_ref(root->fs_info, node, ref, node->action); |
599c75ec | 2341 | |
5d4f98a2 YZ |
2342 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2343 | parent = ref->parent; | |
fcebe456 | 2344 | ref_root = ref->root; |
5d4f98a2 | 2345 | |
3173a18f JB |
2346 | ins.objectid = node->bytenr; |
2347 | if (skinny_metadata) { | |
2348 | ins.offset = ref->level; | |
2349 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2350 | } else { | |
2351 | ins.offset = node->num_bytes; | |
2352 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2353 | } | |
2354 | ||
02794222 LB |
2355 | if (node->ref_mod != 1) { |
2356 | btrfs_err(root->fs_info, | |
2357 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", | |
2358 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2359 | parent); | |
2360 | return -EIO; | |
2361 | } | |
5d4f98a2 | 2362 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2363 | BUG_ON(!extent_op || !extent_op->update_flags); |
5d4f98a2 YZ |
2364 | ret = alloc_reserved_tree_block(trans, root, |
2365 | parent, ref_root, | |
2366 | extent_op->flags_to_set, | |
2367 | &extent_op->key, | |
b06c4bf5 | 2368 | ref->level, &ins); |
5d4f98a2 | 2369 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
c682f9b3 QW |
2370 | ret = __btrfs_inc_extent_ref(trans, root, node, |
2371 | parent, ref_root, | |
2372 | ref->level, 0, 1, | |
fcebe456 | 2373 | extent_op); |
5d4f98a2 | 2374 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
c682f9b3 QW |
2375 | ret = __btrfs_free_extent(trans, root, node, |
2376 | parent, ref_root, | |
2377 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2378 | } else { |
2379 | BUG(); | |
2380 | } | |
56bec294 CM |
2381 | return ret; |
2382 | } | |
2383 | ||
2384 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2385 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2386 | struct btrfs_root *root, | |
2387 | struct btrfs_delayed_ref_node *node, | |
2388 | struct btrfs_delayed_extent_op *extent_op, | |
2389 | int insert_reserved) | |
56bec294 | 2390 | { |
79787eaa JM |
2391 | int ret = 0; |
2392 | ||
857cc2fc JB |
2393 | if (trans->aborted) { |
2394 | if (insert_reserved) | |
2395 | btrfs_pin_extent(root, node->bytenr, | |
2396 | node->num_bytes, 1); | |
79787eaa | 2397 | return 0; |
857cc2fc | 2398 | } |
79787eaa | 2399 | |
5d4f98a2 | 2400 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2401 | struct btrfs_delayed_ref_head *head; |
2402 | /* | |
2403 | * we've hit the end of the chain and we were supposed | |
2404 | * to insert this extent into the tree. But, it got | |
2405 | * deleted before we ever needed to insert it, so all | |
2406 | * we have to do is clean up the accounting | |
2407 | */ | |
5d4f98a2 YZ |
2408 | BUG_ON(extent_op); |
2409 | head = btrfs_delayed_node_to_head(node); | |
bc074524 JM |
2410 | trace_run_delayed_ref_head(root->fs_info, node, head, |
2411 | node->action); | |
599c75ec | 2412 | |
56bec294 | 2413 | if (insert_reserved) { |
f0486c68 YZ |
2414 | btrfs_pin_extent(root, node->bytenr, |
2415 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2416 | if (head->is_data) { |
2417 | ret = btrfs_del_csums(trans, root, | |
2418 | node->bytenr, | |
2419 | node->num_bytes); | |
5d4f98a2 | 2420 | } |
56bec294 | 2421 | } |
297d750b QW |
2422 | |
2423 | /* Also free its reserved qgroup space */ | |
2424 | btrfs_qgroup_free_delayed_ref(root->fs_info, | |
2425 | head->qgroup_ref_root, | |
2426 | head->qgroup_reserved); | |
79787eaa | 2427 | return ret; |
56bec294 CM |
2428 | } |
2429 | ||
5d4f98a2 YZ |
2430 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2431 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2432 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2433 | insert_reserved); | |
2434 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2435 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2436 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2437 | insert_reserved); | |
2438 | else | |
2439 | BUG(); | |
2440 | return ret; | |
56bec294 CM |
2441 | } |
2442 | ||
c6fc2454 | 2443 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2444 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2445 | { | |
cffc3374 FM |
2446 | struct btrfs_delayed_ref_node *ref; |
2447 | ||
c6fc2454 QW |
2448 | if (list_empty(&head->ref_list)) |
2449 | return NULL; | |
d7df2c79 | 2450 | |
cffc3374 FM |
2451 | /* |
2452 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2453 | * This is to prevent a ref count from going down to zero, which deletes | |
2454 | * the extent item from the extent tree, when there still are references | |
2455 | * to add, which would fail because they would not find the extent item. | |
2456 | */ | |
2457 | list_for_each_entry(ref, &head->ref_list, list) { | |
2458 | if (ref->action == BTRFS_ADD_DELAYED_REF) | |
2459 | return ref; | |
2460 | } | |
2461 | ||
c6fc2454 QW |
2462 | return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node, |
2463 | list); | |
56bec294 CM |
2464 | } |
2465 | ||
79787eaa JM |
2466 | /* |
2467 | * Returns 0 on success or if called with an already aborted transaction. | |
2468 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2469 | */ | |
d7df2c79 JB |
2470 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2471 | struct btrfs_root *root, | |
2472 | unsigned long nr) | |
56bec294 | 2473 | { |
56bec294 CM |
2474 | struct btrfs_delayed_ref_root *delayed_refs; |
2475 | struct btrfs_delayed_ref_node *ref; | |
2476 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2477 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2478 | struct btrfs_fs_info *fs_info = root->fs_info; |
0a2b2a84 | 2479 | ktime_t start = ktime_get(); |
56bec294 | 2480 | int ret; |
d7df2c79 | 2481 | unsigned long count = 0; |
0a2b2a84 | 2482 | unsigned long actual_count = 0; |
56bec294 | 2483 | int must_insert_reserved = 0; |
56bec294 CM |
2484 | |
2485 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2486 | while (1) { |
2487 | if (!locked_ref) { | |
d7df2c79 | 2488 | if (count >= nr) |
56bec294 | 2489 | break; |
56bec294 | 2490 | |
d7df2c79 JB |
2491 | spin_lock(&delayed_refs->lock); |
2492 | locked_ref = btrfs_select_ref_head(trans); | |
2493 | if (!locked_ref) { | |
2494 | spin_unlock(&delayed_refs->lock); | |
2495 | break; | |
2496 | } | |
c3e69d58 CM |
2497 | |
2498 | /* grab the lock that says we are going to process | |
2499 | * all the refs for this head */ | |
2500 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2501 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2502 | /* |
2503 | * we may have dropped the spin lock to get the head | |
2504 | * mutex lock, and that might have given someone else | |
2505 | * time to free the head. If that's true, it has been | |
2506 | * removed from our list and we can move on. | |
2507 | */ | |
2508 | if (ret == -EAGAIN) { | |
2509 | locked_ref = NULL; | |
2510 | count++; | |
2511 | continue; | |
56bec294 CM |
2512 | } |
2513 | } | |
a28ec197 | 2514 | |
2c3cf7d5 FM |
2515 | /* |
2516 | * We need to try and merge add/drops of the same ref since we | |
2517 | * can run into issues with relocate dropping the implicit ref | |
2518 | * and then it being added back again before the drop can | |
2519 | * finish. If we merged anything we need to re-loop so we can | |
2520 | * get a good ref. | |
2521 | * Or we can get node references of the same type that weren't | |
2522 | * merged when created due to bumps in the tree mod seq, and | |
2523 | * we need to merge them to prevent adding an inline extent | |
2524 | * backref before dropping it (triggering a BUG_ON at | |
2525 | * insert_inline_extent_backref()). | |
2526 | */ | |
d7df2c79 | 2527 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2528 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2529 | locked_ref); | |
ae1e206b | 2530 | |
d1270cd9 AJ |
2531 | /* |
2532 | * locked_ref is the head node, so we have to go one | |
2533 | * node back for any delayed ref updates | |
2534 | */ | |
2535 | ref = select_delayed_ref(locked_ref); | |
2536 | ||
2537 | if (ref && ref->seq && | |
097b8a7c | 2538 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2539 | spin_unlock(&locked_ref->lock); |
093486c4 | 2540 | btrfs_delayed_ref_unlock(locked_ref); |
d7df2c79 JB |
2541 | spin_lock(&delayed_refs->lock); |
2542 | locked_ref->processing = 0; | |
d1270cd9 AJ |
2543 | delayed_refs->num_heads_ready++; |
2544 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 2545 | locked_ref = NULL; |
d1270cd9 | 2546 | cond_resched(); |
27a377db | 2547 | count++; |
d1270cd9 AJ |
2548 | continue; |
2549 | } | |
2550 | ||
56bec294 CM |
2551 | /* |
2552 | * record the must insert reserved flag before we | |
2553 | * drop the spin lock. | |
2554 | */ | |
2555 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2556 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2557 | |
5d4f98a2 YZ |
2558 | extent_op = locked_ref->extent_op; |
2559 | locked_ref->extent_op = NULL; | |
2560 | ||
56bec294 | 2561 | if (!ref) { |
d7df2c79 JB |
2562 | |
2563 | ||
56bec294 CM |
2564 | /* All delayed refs have been processed, Go ahead |
2565 | * and send the head node to run_one_delayed_ref, | |
2566 | * so that any accounting fixes can happen | |
2567 | */ | |
2568 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2569 | |
2570 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2571 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2572 | extent_op = NULL; |
2573 | } | |
2574 | ||
2575 | if (extent_op) { | |
d7df2c79 | 2576 | spin_unlock(&locked_ref->lock); |
5d4f98a2 YZ |
2577 | ret = run_delayed_extent_op(trans, root, |
2578 | ref, extent_op); | |
78a6184a | 2579 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2580 | |
79787eaa | 2581 | if (ret) { |
857cc2fc JB |
2582 | /* |
2583 | * Need to reset must_insert_reserved if | |
2584 | * there was an error so the abort stuff | |
2585 | * can cleanup the reserved space | |
2586 | * properly. | |
2587 | */ | |
2588 | if (must_insert_reserved) | |
2589 | locked_ref->must_insert_reserved = 1; | |
d7df2c79 | 2590 | locked_ref->processing = 0; |
5d163e0e JM |
2591 | btrfs_debug(fs_info, |
2592 | "run_delayed_extent_op returned %d", | |
2593 | ret); | |
093486c4 | 2594 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2595 | return ret; |
2596 | } | |
d7df2c79 | 2597 | continue; |
5d4f98a2 | 2598 | } |
02217ed2 | 2599 | |
d7df2c79 | 2600 | /* |
01327610 | 2601 | * Need to drop our head ref lock and re-acquire the |
d7df2c79 JB |
2602 | * delayed ref lock and then re-check to make sure |
2603 | * nobody got added. | |
2604 | */ | |
2605 | spin_unlock(&locked_ref->lock); | |
2606 | spin_lock(&delayed_refs->lock); | |
2607 | spin_lock(&locked_ref->lock); | |
c6fc2454 | 2608 | if (!list_empty(&locked_ref->ref_list) || |
573a0755 | 2609 | locked_ref->extent_op) { |
d7df2c79 JB |
2610 | spin_unlock(&locked_ref->lock); |
2611 | spin_unlock(&delayed_refs->lock); | |
2612 | continue; | |
2613 | } | |
2614 | ref->in_tree = 0; | |
2615 | delayed_refs->num_heads--; | |
c46effa6 LB |
2616 | rb_erase(&locked_ref->href_node, |
2617 | &delayed_refs->href_root); | |
d7df2c79 JB |
2618 | spin_unlock(&delayed_refs->lock); |
2619 | } else { | |
0a2b2a84 | 2620 | actual_count++; |
d7df2c79 | 2621 | ref->in_tree = 0; |
c6fc2454 | 2622 | list_del(&ref->list); |
c46effa6 | 2623 | } |
d7df2c79 JB |
2624 | atomic_dec(&delayed_refs->num_entries); |
2625 | ||
093486c4 | 2626 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2627 | /* |
2628 | * when we play the delayed ref, also correct the | |
2629 | * ref_mod on head | |
2630 | */ | |
2631 | switch (ref->action) { | |
2632 | case BTRFS_ADD_DELAYED_REF: | |
2633 | case BTRFS_ADD_DELAYED_EXTENT: | |
2634 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2635 | break; | |
2636 | case BTRFS_DROP_DELAYED_REF: | |
2637 | locked_ref->node.ref_mod += ref->ref_mod; | |
2638 | break; | |
2639 | default: | |
2640 | WARN_ON(1); | |
2641 | } | |
2642 | } | |
d7df2c79 | 2643 | spin_unlock(&locked_ref->lock); |
925baedd | 2644 | |
5d4f98a2 | 2645 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2646 | must_insert_reserved); |
eb099670 | 2647 | |
78a6184a | 2648 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2649 | if (ret) { |
d7df2c79 | 2650 | locked_ref->processing = 0; |
093486c4 MX |
2651 | btrfs_delayed_ref_unlock(locked_ref); |
2652 | btrfs_put_delayed_ref(ref); | |
5d163e0e JM |
2653 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
2654 | ret); | |
79787eaa JM |
2655 | return ret; |
2656 | } | |
2657 | ||
093486c4 MX |
2658 | /* |
2659 | * If this node is a head, that means all the refs in this head | |
2660 | * have been dealt with, and we will pick the next head to deal | |
2661 | * with, so we must unlock the head and drop it from the cluster | |
2662 | * list before we release it. | |
2663 | */ | |
2664 | if (btrfs_delayed_ref_is_head(ref)) { | |
1262133b JB |
2665 | if (locked_ref->is_data && |
2666 | locked_ref->total_ref_mod < 0) { | |
2667 | spin_lock(&delayed_refs->lock); | |
2668 | delayed_refs->pending_csums -= ref->num_bytes; | |
2669 | spin_unlock(&delayed_refs->lock); | |
2670 | } | |
093486c4 MX |
2671 | btrfs_delayed_ref_unlock(locked_ref); |
2672 | locked_ref = NULL; | |
2673 | } | |
2674 | btrfs_put_delayed_ref(ref); | |
2675 | count++; | |
c3e69d58 | 2676 | cond_resched(); |
c3e69d58 | 2677 | } |
0a2b2a84 JB |
2678 | |
2679 | /* | |
2680 | * We don't want to include ref heads since we can have empty ref heads | |
2681 | * and those will drastically skew our runtime down since we just do | |
2682 | * accounting, no actual extent tree updates. | |
2683 | */ | |
2684 | if (actual_count > 0) { | |
2685 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2686 | u64 avg; | |
2687 | ||
2688 | /* | |
2689 | * We weigh the current average higher than our current runtime | |
2690 | * to avoid large swings in the average. | |
2691 | */ | |
2692 | spin_lock(&delayed_refs->lock); | |
2693 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2694 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2695 | spin_unlock(&delayed_refs->lock); |
2696 | } | |
d7df2c79 | 2697 | return 0; |
c3e69d58 CM |
2698 | } |
2699 | ||
709c0486 AJ |
2700 | #ifdef SCRAMBLE_DELAYED_REFS |
2701 | /* | |
2702 | * Normally delayed refs get processed in ascending bytenr order. This | |
2703 | * correlates in most cases to the order added. To expose dependencies on this | |
2704 | * order, we start to process the tree in the middle instead of the beginning | |
2705 | */ | |
2706 | static u64 find_middle(struct rb_root *root) | |
2707 | { | |
2708 | struct rb_node *n = root->rb_node; | |
2709 | struct btrfs_delayed_ref_node *entry; | |
2710 | int alt = 1; | |
2711 | u64 middle; | |
2712 | u64 first = 0, last = 0; | |
2713 | ||
2714 | n = rb_first(root); | |
2715 | if (n) { | |
2716 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2717 | first = entry->bytenr; | |
2718 | } | |
2719 | n = rb_last(root); | |
2720 | if (n) { | |
2721 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2722 | last = entry->bytenr; | |
2723 | } | |
2724 | n = root->rb_node; | |
2725 | ||
2726 | while (n) { | |
2727 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2728 | WARN_ON(!entry->in_tree); | |
2729 | ||
2730 | middle = entry->bytenr; | |
2731 | ||
2732 | if (alt) | |
2733 | n = n->rb_left; | |
2734 | else | |
2735 | n = n->rb_right; | |
2736 | ||
2737 | alt = 1 - alt; | |
2738 | } | |
2739 | return middle; | |
2740 | } | |
2741 | #endif | |
2742 | ||
1be41b78 JB |
2743 | static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads) |
2744 | { | |
2745 | u64 num_bytes; | |
2746 | ||
2747 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2748 | sizeof(struct btrfs_extent_inline_ref)); | |
2749 | if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) | |
2750 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); | |
2751 | ||
2752 | /* | |
2753 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2754 | * closer to what we're really going to want to use. |
1be41b78 | 2755 | */ |
f8c269d7 | 2756 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root)); |
1be41b78 JB |
2757 | } |
2758 | ||
1262133b JB |
2759 | /* |
2760 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2761 | * would require to store the csums for that many bytes. | |
2762 | */ | |
28f75a0e | 2763 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes) |
1262133b JB |
2764 | { |
2765 | u64 csum_size; | |
2766 | u64 num_csums_per_leaf; | |
2767 | u64 num_csums; | |
2768 | ||
14a1e067 | 2769 | csum_size = BTRFS_MAX_ITEM_SIZE(root); |
1262133b JB |
2770 | num_csums_per_leaf = div64_u64(csum_size, |
2771 | (u64)btrfs_super_csum_size(root->fs_info->super_copy)); | |
2772 | num_csums = div64_u64(csum_bytes, root->sectorsize); | |
2773 | num_csums += num_csums_per_leaf - 1; | |
2774 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2775 | return num_csums; | |
2776 | } | |
2777 | ||
0a2b2a84 | 2778 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
1be41b78 JB |
2779 | struct btrfs_root *root) |
2780 | { | |
2781 | struct btrfs_block_rsv *global_rsv; | |
2782 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2783 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
cb723e49 JB |
2784 | u64 num_dirty_bgs = trans->transaction->num_dirty_bgs; |
2785 | u64 num_bytes, num_dirty_bgs_bytes; | |
1be41b78 JB |
2786 | int ret = 0; |
2787 | ||
2788 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
2789 | num_heads = heads_to_leaves(root, num_heads); | |
2790 | if (num_heads > 1) | |
707e8a07 | 2791 | num_bytes += (num_heads - 1) * root->nodesize; |
1be41b78 | 2792 | num_bytes <<= 1; |
28f75a0e | 2793 | num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize; |
cb723e49 JB |
2794 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root, |
2795 | num_dirty_bgs); | |
1be41b78 JB |
2796 | global_rsv = &root->fs_info->global_block_rsv; |
2797 | ||
2798 | /* | |
2799 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2800 | * wiggle room since running delayed refs can create more delayed refs. | |
2801 | */ | |
cb723e49 JB |
2802 | if (global_rsv->space_info->full) { |
2803 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2804 | num_bytes <<= 1; |
cb723e49 | 2805 | } |
1be41b78 JB |
2806 | |
2807 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2808 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2809 | ret = 1; |
2810 | spin_unlock(&global_rsv->lock); | |
2811 | return ret; | |
2812 | } | |
2813 | ||
0a2b2a84 JB |
2814 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2815 | struct btrfs_root *root) | |
2816 | { | |
2817 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2818 | u64 num_entries = | |
2819 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2820 | u64 avg_runtime; | |
a79b7d4b | 2821 | u64 val; |
0a2b2a84 JB |
2822 | |
2823 | smp_mb(); | |
2824 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2825 | val = num_entries * avg_runtime; |
0a2b2a84 JB |
2826 | if (num_entries * avg_runtime >= NSEC_PER_SEC) |
2827 | return 1; | |
a79b7d4b CM |
2828 | if (val >= NSEC_PER_SEC / 2) |
2829 | return 2; | |
0a2b2a84 JB |
2830 | |
2831 | return btrfs_check_space_for_delayed_refs(trans, root); | |
2832 | } | |
2833 | ||
a79b7d4b CM |
2834 | struct async_delayed_refs { |
2835 | struct btrfs_root *root; | |
31b9655f | 2836 | u64 transid; |
a79b7d4b CM |
2837 | int count; |
2838 | int error; | |
2839 | int sync; | |
2840 | struct completion wait; | |
2841 | struct btrfs_work work; | |
2842 | }; | |
2843 | ||
2844 | static void delayed_ref_async_start(struct btrfs_work *work) | |
2845 | { | |
2846 | struct async_delayed_refs *async; | |
2847 | struct btrfs_trans_handle *trans; | |
2848 | int ret; | |
2849 | ||
2850 | async = container_of(work, struct async_delayed_refs, work); | |
2851 | ||
0f873eca CM |
2852 | /* if the commit is already started, we don't need to wait here */ |
2853 | if (btrfs_transaction_blocked(async->root->fs_info)) | |
31b9655f | 2854 | goto done; |
31b9655f | 2855 | |
0f873eca CM |
2856 | trans = btrfs_join_transaction(async->root); |
2857 | if (IS_ERR(trans)) { | |
2858 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2859 | goto done; |
2860 | } | |
2861 | ||
2862 | /* | |
01327610 | 2863 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2864 | * wait on delayed refs |
2865 | */ | |
2866 | trans->sync = true; | |
0f873eca CM |
2867 | |
2868 | /* Don't bother flushing if we got into a different transaction */ | |
2869 | if (trans->transid > async->transid) | |
2870 | goto end; | |
2871 | ||
a79b7d4b CM |
2872 | ret = btrfs_run_delayed_refs(trans, async->root, async->count); |
2873 | if (ret) | |
2874 | async->error = ret; | |
0f873eca | 2875 | end: |
a79b7d4b CM |
2876 | ret = btrfs_end_transaction(trans, async->root); |
2877 | if (ret && !async->error) | |
2878 | async->error = ret; | |
2879 | done: | |
2880 | if (async->sync) | |
2881 | complete(&async->wait); | |
2882 | else | |
2883 | kfree(async); | |
2884 | } | |
2885 | ||
2886 | int btrfs_async_run_delayed_refs(struct btrfs_root *root, | |
31b9655f | 2887 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
2888 | { |
2889 | struct async_delayed_refs *async; | |
2890 | int ret; | |
2891 | ||
2892 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2893 | if (!async) | |
2894 | return -ENOMEM; | |
2895 | ||
2896 | async->root = root->fs_info->tree_root; | |
2897 | async->count = count; | |
2898 | async->error = 0; | |
31b9655f | 2899 | async->transid = transid; |
a79b7d4b CM |
2900 | if (wait) |
2901 | async->sync = 1; | |
2902 | else | |
2903 | async->sync = 0; | |
2904 | init_completion(&async->wait); | |
2905 | ||
9e0af237 LB |
2906 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
2907 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b CM |
2908 | |
2909 | btrfs_queue_work(root->fs_info->extent_workers, &async->work); | |
2910 | ||
2911 | if (wait) { | |
2912 | wait_for_completion(&async->wait); | |
2913 | ret = async->error; | |
2914 | kfree(async); | |
2915 | return ret; | |
2916 | } | |
2917 | return 0; | |
2918 | } | |
2919 | ||
c3e69d58 CM |
2920 | /* |
2921 | * this starts processing the delayed reference count updates and | |
2922 | * extent insertions we have queued up so far. count can be | |
2923 | * 0, which means to process everything in the tree at the start | |
2924 | * of the run (but not newly added entries), or it can be some target | |
2925 | * number you'd like to process. | |
79787eaa JM |
2926 | * |
2927 | * Returns 0 on success or if called with an aborted transaction | |
2928 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2929 | */ |
2930 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2931 | struct btrfs_root *root, unsigned long count) | |
2932 | { | |
2933 | struct rb_node *node; | |
2934 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2935 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2936 | int ret; |
2937 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 2938 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 2939 | |
79787eaa JM |
2940 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2941 | if (trans->aborted) | |
2942 | return 0; | |
2943 | ||
afcdd129 | 2944 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &root->fs_info->flags)) |
511711af CM |
2945 | return 0; |
2946 | ||
c3e69d58 CM |
2947 | if (root == root->fs_info->extent_root) |
2948 | root = root->fs_info->tree_root; | |
2949 | ||
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; |
d7df2c79 JB |
2959 | ret = __btrfs_run_delayed_refs(trans, root, count); |
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)) |
ea658bad | 2967 | btrfs_create_pending_block_groups(trans, root); |
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 YZ |
3011 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
3012 | struct btrfs_root *root, | |
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 | |
66d7e7f0 AJ |
3029 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
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 | ||
3036 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
3037 | struct btrfs_root *root, | |
3038 | struct btrfs_path *path, | |
3039 | u64 objectid, u64 offset, u64 bytenr) | |
3040 | { | |
3041 | struct btrfs_delayed_ref_head *head; | |
3042 | struct btrfs_delayed_ref_node *ref; | |
3043 | struct btrfs_delayed_data_ref *data_ref; | |
3044 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 YZ |
3045 | int ret = 0; |
3046 | ||
5d4f98a2 YZ |
3047 | delayed_refs = &trans->transaction->delayed_refs; |
3048 | spin_lock(&delayed_refs->lock); | |
3049 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
d7df2c79 JB |
3050 | if (!head) { |
3051 | spin_unlock(&delayed_refs->lock); | |
3052 | return 0; | |
3053 | } | |
5d4f98a2 YZ |
3054 | |
3055 | if (!mutex_trylock(&head->mutex)) { | |
3056 | atomic_inc(&head->node.refs); | |
3057 | spin_unlock(&delayed_refs->lock); | |
3058 | ||
b3b4aa74 | 3059 | btrfs_release_path(path); |
5d4f98a2 | 3060 | |
8cc33e5c DS |
3061 | /* |
3062 | * Mutex was contended, block until it's released and let | |
3063 | * caller try again | |
3064 | */ | |
5d4f98a2 YZ |
3065 | mutex_lock(&head->mutex); |
3066 | mutex_unlock(&head->mutex); | |
3067 | btrfs_put_delayed_ref(&head->node); | |
3068 | return -EAGAIN; | |
3069 | } | |
d7df2c79 | 3070 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3071 | |
d7df2c79 | 3072 | spin_lock(&head->lock); |
c6fc2454 | 3073 | list_for_each_entry(ref, &head->ref_list, list) { |
d7df2c79 JB |
3074 | /* If it's a shared ref we know a cross reference exists */ |
3075 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3076 | ret = 1; | |
3077 | break; | |
3078 | } | |
5d4f98a2 | 3079 | |
d7df2c79 | 3080 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3081 | |
d7df2c79 JB |
3082 | /* |
3083 | * If our ref doesn't match the one we're currently looking at | |
3084 | * then we have a cross reference. | |
3085 | */ | |
3086 | if (data_ref->root != root->root_key.objectid || | |
3087 | data_ref->objectid != objectid || | |
3088 | data_ref->offset != offset) { | |
3089 | ret = 1; | |
3090 | break; | |
3091 | } | |
5d4f98a2 | 3092 | } |
d7df2c79 | 3093 | spin_unlock(&head->lock); |
5d4f98a2 | 3094 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
3095 | return ret; |
3096 | } | |
3097 | ||
3098 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
3099 | struct btrfs_root *root, | |
3100 | struct btrfs_path *path, | |
3101 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
3102 | { |
3103 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 3104 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3105 | struct btrfs_extent_data_ref *ref; |
3106 | struct btrfs_extent_inline_ref *iref; | |
3107 | struct btrfs_extent_item *ei; | |
f321e491 | 3108 | struct btrfs_key key; |
5d4f98a2 | 3109 | u32 item_size; |
be20aa9d | 3110 | int ret; |
925baedd | 3111 | |
be20aa9d | 3112 | key.objectid = bytenr; |
31840ae1 | 3113 | key.offset = (u64)-1; |
f321e491 | 3114 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3115 | |
be20aa9d CM |
3116 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3117 | if (ret < 0) | |
3118 | goto out; | |
79787eaa | 3119 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3120 | |
3121 | ret = -ENOENT; | |
3122 | if (path->slots[0] == 0) | |
31840ae1 | 3123 | goto out; |
be20aa9d | 3124 | |
31840ae1 | 3125 | path->slots[0]--; |
f321e491 | 3126 | leaf = path->nodes[0]; |
5d4f98a2 | 3127 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3128 | |
5d4f98a2 | 3129 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3130 | goto out; |
f321e491 | 3131 | |
5d4f98a2 YZ |
3132 | ret = 1; |
3133 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3134 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3135 | if (item_size < sizeof(*ei)) { | |
3136 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3137 | goto out; | |
3138 | } | |
3139 | #endif | |
3140 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3141 | |
5d4f98a2 YZ |
3142 | if (item_size != sizeof(*ei) + |
3143 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3144 | goto out; | |
be20aa9d | 3145 | |
5d4f98a2 YZ |
3146 | if (btrfs_extent_generation(leaf, ei) <= |
3147 | btrfs_root_last_snapshot(&root->root_item)) | |
3148 | goto out; | |
3149 | ||
3150 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3151 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
3152 | BTRFS_EXTENT_DATA_REF_KEY) | |
3153 | goto out; | |
3154 | ||
3155 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3156 | if (btrfs_extent_refs(leaf, ei) != | |
3157 | btrfs_extent_data_ref_count(leaf, ref) || | |
3158 | btrfs_extent_data_ref_root(leaf, ref) != | |
3159 | root->root_key.objectid || | |
3160 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3161 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3162 | goto out; | |
3163 | ||
3164 | ret = 0; | |
3165 | out: | |
3166 | return ret; | |
3167 | } | |
3168 | ||
3169 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
3170 | struct btrfs_root *root, | |
3171 | u64 objectid, u64 offset, u64 bytenr) | |
3172 | { | |
3173 | struct btrfs_path *path; | |
3174 | int ret; | |
3175 | int ret2; | |
3176 | ||
3177 | path = btrfs_alloc_path(); | |
3178 | if (!path) | |
3179 | return -ENOENT; | |
3180 | ||
3181 | do { | |
3182 | ret = check_committed_ref(trans, root, path, objectid, | |
3183 | offset, bytenr); | |
3184 | if (ret && ret != -ENOENT) | |
f321e491 | 3185 | goto out; |
80ff3856 | 3186 | |
5d4f98a2 YZ |
3187 | ret2 = check_delayed_ref(trans, root, path, objectid, |
3188 | offset, bytenr); | |
3189 | } while (ret2 == -EAGAIN); | |
3190 | ||
3191 | if (ret2 && ret2 != -ENOENT) { | |
3192 | ret = ret2; | |
3193 | goto out; | |
f321e491 | 3194 | } |
5d4f98a2 YZ |
3195 | |
3196 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3197 | ret = 0; | |
be20aa9d | 3198 | out: |
80ff3856 | 3199 | btrfs_free_path(path); |
f0486c68 YZ |
3200 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3201 | WARN_ON(ret > 0); | |
f321e491 | 3202 | return ret; |
be20aa9d | 3203 | } |
c5739bba | 3204 | |
5d4f98a2 | 3205 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3206 | struct btrfs_root *root, |
5d4f98a2 | 3207 | struct extent_buffer *buf, |
e339a6b0 | 3208 | int full_backref, int inc) |
31840ae1 ZY |
3209 | { |
3210 | u64 bytenr; | |
5d4f98a2 YZ |
3211 | u64 num_bytes; |
3212 | u64 parent; | |
31840ae1 | 3213 | u64 ref_root; |
31840ae1 | 3214 | u32 nritems; |
31840ae1 ZY |
3215 | struct btrfs_key key; |
3216 | struct btrfs_file_extent_item *fi; | |
3217 | int i; | |
3218 | int level; | |
3219 | int ret = 0; | |
31840ae1 | 3220 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
b06c4bf5 | 3221 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3222 | |
fccb84c9 | 3223 | |
f5ee5c9a | 3224 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 3225 | return 0; |
fccb84c9 | 3226 | |
31840ae1 | 3227 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3228 | nritems = btrfs_header_nritems(buf); |
3229 | level = btrfs_header_level(buf); | |
3230 | ||
27cdeb70 | 3231 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3232 | return 0; |
31840ae1 | 3233 | |
5d4f98a2 YZ |
3234 | if (inc) |
3235 | process_func = btrfs_inc_extent_ref; | |
3236 | else | |
3237 | process_func = btrfs_free_extent; | |
31840ae1 | 3238 | |
5d4f98a2 YZ |
3239 | if (full_backref) |
3240 | parent = buf->start; | |
3241 | else | |
3242 | parent = 0; | |
3243 | ||
3244 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3245 | if (level == 0) { |
5d4f98a2 | 3246 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3247 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3248 | continue; |
5d4f98a2 | 3249 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3250 | struct btrfs_file_extent_item); |
3251 | if (btrfs_file_extent_type(buf, fi) == | |
3252 | BTRFS_FILE_EXTENT_INLINE) | |
3253 | continue; | |
3254 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3255 | if (bytenr == 0) | |
3256 | continue; | |
5d4f98a2 YZ |
3257 | |
3258 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3259 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
3260 | ret = process_func(trans, root, bytenr, num_bytes, | |
3261 | parent, ref_root, key.objectid, | |
b06c4bf5 | 3262 | key.offset); |
31840ae1 ZY |
3263 | if (ret) |
3264 | goto fail; | |
3265 | } else { | |
5d4f98a2 | 3266 | bytenr = btrfs_node_blockptr(buf, i); |
707e8a07 | 3267 | num_bytes = root->nodesize; |
5d4f98a2 | 3268 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3269 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3270 | if (ret) |
3271 | goto fail; | |
3272 | } | |
3273 | } | |
3274 | return 0; | |
3275 | fail: | |
5d4f98a2 YZ |
3276 | return ret; |
3277 | } | |
3278 | ||
3279 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3280 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3281 | { |
e339a6b0 | 3282 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3283 | } |
3284 | ||
3285 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3286 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3287 | { |
e339a6b0 | 3288 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3289 | } |
3290 | ||
9078a3e1 CM |
3291 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
3292 | struct btrfs_root *root, | |
3293 | struct btrfs_path *path, | |
3294 | struct btrfs_block_group_cache *cache) | |
3295 | { | |
3296 | int ret; | |
9078a3e1 | 3297 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
3298 | unsigned long bi; |
3299 | struct extent_buffer *leaf; | |
9078a3e1 | 3300 | |
9078a3e1 | 3301 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3302 | if (ret) { |
3303 | if (ret > 0) | |
3304 | ret = -ENOENT; | |
54aa1f4d | 3305 | goto fail; |
df95e7f0 | 3306 | } |
5f39d397 CM |
3307 | |
3308 | leaf = path->nodes[0]; | |
3309 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3310 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3311 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3312 | fail: |
24b89d08 | 3313 | btrfs_release_path(path); |
df95e7f0 | 3314 | return ret; |
9078a3e1 CM |
3315 | |
3316 | } | |
3317 | ||
4a8c9a62 YZ |
3318 | static struct btrfs_block_group_cache * |
3319 | next_block_group(struct btrfs_root *root, | |
3320 | struct btrfs_block_group_cache *cache) | |
3321 | { | |
3322 | struct rb_node *node; | |
292cbd51 | 3323 | |
4a8c9a62 | 3324 | spin_lock(&root->fs_info->block_group_cache_lock); |
292cbd51 FM |
3325 | |
3326 | /* If our block group was removed, we need a full search. */ | |
3327 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3328 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3329 | ||
3330 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3331 | btrfs_put_block_group(cache); | |
3332 | cache = btrfs_lookup_first_block_group(root->fs_info, | |
3333 | next_bytenr); | |
3334 | return cache; | |
3335 | } | |
4a8c9a62 YZ |
3336 | node = rb_next(&cache->cache_node); |
3337 | btrfs_put_block_group(cache); | |
3338 | if (node) { | |
3339 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3340 | cache_node); | |
11dfe35a | 3341 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3342 | } else |
3343 | cache = NULL; | |
3344 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3345 | return cache; | |
3346 | } | |
3347 | ||
0af3d00b JB |
3348 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3349 | struct btrfs_trans_handle *trans, | |
3350 | struct btrfs_path *path) | |
3351 | { | |
3352 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
3353 | struct inode *inode = NULL; | |
3354 | u64 alloc_hint = 0; | |
2b20982e | 3355 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3356 | u64 num_pages = 0; |
0af3d00b JB |
3357 | int retries = 0; |
3358 | int ret = 0; | |
3359 | ||
3360 | /* | |
3361 | * If this block group is smaller than 100 megs don't bother caching the | |
3362 | * block group. | |
3363 | */ | |
ee22184b | 3364 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3365 | spin_lock(&block_group->lock); |
3366 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3367 | spin_unlock(&block_group->lock); | |
3368 | return 0; | |
3369 | } | |
3370 | ||
0c0ef4bc JB |
3371 | if (trans->aborted) |
3372 | return 0; | |
0af3d00b JB |
3373 | again: |
3374 | inode = lookup_free_space_inode(root, block_group, path); | |
3375 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
3376 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3377 | btrfs_release_path(path); |
0af3d00b JB |
3378 | goto out; |
3379 | } | |
3380 | ||
3381 | if (IS_ERR(inode)) { | |
3382 | BUG_ON(retries); | |
3383 | retries++; | |
3384 | ||
3385 | if (block_group->ro) | |
3386 | goto out_free; | |
3387 | ||
3388 | ret = create_free_space_inode(root, trans, block_group, path); | |
3389 | if (ret) | |
3390 | goto out_free; | |
3391 | goto again; | |
3392 | } | |
3393 | ||
5b0e95bf JB |
3394 | /* We've already setup this transaction, go ahead and exit */ |
3395 | if (block_group->cache_generation == trans->transid && | |
3396 | i_size_read(inode)) { | |
3397 | dcs = BTRFS_DC_SETUP; | |
3398 | goto out_put; | |
3399 | } | |
3400 | ||
0af3d00b JB |
3401 | /* |
3402 | * We want to set the generation to 0, that way if anything goes wrong | |
3403 | * from here on out we know not to trust this cache when we load up next | |
3404 | * time. | |
3405 | */ | |
3406 | BTRFS_I(inode)->generation = 0; | |
3407 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3408 | if (ret) { |
3409 | /* | |
3410 | * So theoretically we could recover from this, simply set the | |
3411 | * super cache generation to 0 so we know to invalidate the | |
3412 | * cache, but then we'd have to keep track of the block groups | |
3413 | * that fail this way so we know we _have_ to reset this cache | |
3414 | * before the next commit or risk reading stale cache. So to | |
3415 | * limit our exposure to horrible edge cases lets just abort the | |
3416 | * transaction, this only happens in really bad situations | |
3417 | * anyway. | |
3418 | */ | |
66642832 | 3419 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3420 | goto out_put; |
3421 | } | |
0af3d00b JB |
3422 | WARN_ON(ret); |
3423 | ||
3424 | if (i_size_read(inode) > 0) { | |
7b61cd92 MX |
3425 | ret = btrfs_check_trunc_cache_free_space(root, |
3426 | &root->fs_info->global_block_rsv); | |
3427 | if (ret) | |
3428 | goto out_put; | |
3429 | ||
1bbc621e | 3430 | ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode); |
0af3d00b JB |
3431 | if (ret) |
3432 | goto out_put; | |
3433 | } | |
3434 | ||
3435 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3436 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
3cdde224 | 3437 | !btrfs_test_opt(root->fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3438 | /* |
3439 | * don't bother trying to write stuff out _if_ | |
3440 | * a) we're not cached, | |
3441 | * b) we're with nospace_cache mount option. | |
3442 | */ | |
2b20982e | 3443 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3444 | spin_unlock(&block_group->lock); |
3445 | goto out_put; | |
3446 | } | |
3447 | spin_unlock(&block_group->lock); | |
3448 | ||
2968b1f4 JB |
3449 | /* |
3450 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3451 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3452 | */ | |
3453 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3454 | ret = -ENOSPC; | |
3455 | goto out_put; | |
3456 | } | |
3457 | ||
6fc823b1 JB |
3458 | /* |
3459 | * Try to preallocate enough space based on how big the block group is. | |
3460 | * Keep in mind this has to include any pinned space which could end up | |
3461 | * taking up quite a bit since it's not folded into the other space | |
3462 | * cache. | |
3463 | */ | |
ee22184b | 3464 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3465 | if (!num_pages) |
3466 | num_pages = 1; | |
3467 | ||
0af3d00b | 3468 | num_pages *= 16; |
09cbfeaf | 3469 | num_pages *= PAGE_SIZE; |
0af3d00b | 3470 | |
7cf5b976 | 3471 | ret = btrfs_check_data_free_space(inode, 0, num_pages); |
0af3d00b JB |
3472 | if (ret) |
3473 | goto out_put; | |
3474 | ||
3475 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3476 | num_pages, num_pages, | |
3477 | &alloc_hint); | |
2968b1f4 JB |
3478 | /* |
3479 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3480 | * of metadata or split extents when writing the cache out, which means | |
3481 | * we can enospc if we are heavily fragmented in addition to just normal | |
3482 | * out of space conditions. So if we hit this just skip setting up any | |
3483 | * other block groups for this transaction, maybe we'll unpin enough | |
3484 | * space the next time around. | |
3485 | */ | |
2b20982e JB |
3486 | if (!ret) |
3487 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3488 | else if (ret == -ENOSPC) |
3489 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3490 | |
0af3d00b JB |
3491 | out_put: |
3492 | iput(inode); | |
3493 | out_free: | |
b3b4aa74 | 3494 | btrfs_release_path(path); |
0af3d00b JB |
3495 | out: |
3496 | spin_lock(&block_group->lock); | |
e65cbb94 | 3497 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3498 | block_group->cache_generation = trans->transid; |
2b20982e | 3499 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3500 | spin_unlock(&block_group->lock); |
3501 | ||
3502 | return ret; | |
3503 | } | |
3504 | ||
dcdf7f6d JB |
3505 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
3506 | struct btrfs_root *root) | |
3507 | { | |
3508 | struct btrfs_block_group_cache *cache, *tmp; | |
3509 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3510 | struct btrfs_path *path; | |
3511 | ||
3512 | if (list_empty(&cur_trans->dirty_bgs) || | |
3cdde224 | 3513 | !btrfs_test_opt(root->fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3514 | return 0; |
3515 | ||
3516 | path = btrfs_alloc_path(); | |
3517 | if (!path) | |
3518 | return -ENOMEM; | |
3519 | ||
3520 | /* Could add new block groups, use _safe just in case */ | |
3521 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3522 | dirty_list) { | |
3523 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3524 | cache_save_setup(cache, trans, path); | |
3525 | } | |
3526 | ||
3527 | btrfs_free_path(path); | |
3528 | return 0; | |
3529 | } | |
3530 | ||
1bbc621e CM |
3531 | /* |
3532 | * transaction commit does final block group cache writeback during a | |
3533 | * critical section where nothing is allowed to change the FS. This is | |
3534 | * required in order for the cache to actually match the block group, | |
3535 | * but can introduce a lot of latency into the commit. | |
3536 | * | |
3537 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3538 | * cache IO. There's a chance we'll have to redo some of it if the | |
3539 | * block group changes again during the commit, but it greatly reduces | |
3540 | * the commit latency by getting rid of the easy block groups while | |
3541 | * we're still allowing others to join the commit. | |
3542 | */ | |
3543 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans, | |
96b5179d | 3544 | struct btrfs_root *root) |
9078a3e1 | 3545 | { |
4a8c9a62 | 3546 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3547 | struct btrfs_transaction *cur_trans = trans->transaction; |
3548 | int ret = 0; | |
c9dc4c65 | 3549 | int should_put; |
1bbc621e CM |
3550 | struct btrfs_path *path = NULL; |
3551 | LIST_HEAD(dirty); | |
3552 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3553 | int num_started = 0; |
1bbc621e CM |
3554 | int loops = 0; |
3555 | ||
3556 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3557 | if (list_empty(&cur_trans->dirty_bgs)) { |
3558 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3559 | return 0; | |
1bbc621e | 3560 | } |
b58d1a9e | 3561 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3562 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3563 | |
1bbc621e | 3564 | again: |
1bbc621e CM |
3565 | /* |
3566 | * make sure all the block groups on our dirty list actually | |
3567 | * exist | |
3568 | */ | |
3569 | btrfs_create_pending_block_groups(trans, root); | |
3570 | ||
3571 | if (!path) { | |
3572 | path = btrfs_alloc_path(); | |
3573 | if (!path) | |
3574 | return -ENOMEM; | |
3575 | } | |
3576 | ||
b58d1a9e FM |
3577 | /* |
3578 | * cache_write_mutex is here only to save us from balance or automatic | |
3579 | * removal of empty block groups deleting this block group while we are | |
3580 | * writing out the cache | |
3581 | */ | |
3582 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3583 | while (!list_empty(&dirty)) { |
3584 | cache = list_first_entry(&dirty, | |
3585 | struct btrfs_block_group_cache, | |
3586 | dirty_list); | |
1bbc621e CM |
3587 | /* |
3588 | * this can happen if something re-dirties a block | |
3589 | * group that is already under IO. Just wait for it to | |
3590 | * finish and then do it all again | |
3591 | */ | |
3592 | if (!list_empty(&cache->io_list)) { | |
3593 | list_del_init(&cache->io_list); | |
3594 | btrfs_wait_cache_io(root, trans, cache, | |
3595 | &cache->io_ctl, path, | |
3596 | cache->key.objectid); | |
3597 | btrfs_put_block_group(cache); | |
3598 | } | |
3599 | ||
3600 | ||
3601 | /* | |
3602 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3603 | * if it should update the cache_state. Don't delete | |
3604 | * until after we wait. | |
3605 | * | |
3606 | * Since we're not running in the commit critical section | |
3607 | * we need the dirty_bgs_lock to protect from update_block_group | |
3608 | */ | |
3609 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3610 | list_del_init(&cache->dirty_list); | |
3611 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3612 | ||
3613 | should_put = 1; | |
3614 | ||
3615 | cache_save_setup(cache, trans, path); | |
3616 | ||
3617 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3618 | cache->io_ctl.inode = NULL; | |
3619 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3620 | if (ret == 0 && cache->io_ctl.inode) { | |
3621 | num_started++; | |
3622 | should_put = 0; | |
3623 | ||
3624 | /* | |
3625 | * the cache_write_mutex is protecting | |
3626 | * the io_list | |
3627 | */ | |
3628 | list_add_tail(&cache->io_list, io); | |
3629 | } else { | |
3630 | /* | |
3631 | * if we failed to write the cache, the | |
3632 | * generation will be bad and life goes on | |
3633 | */ | |
3634 | ret = 0; | |
3635 | } | |
3636 | } | |
ff1f8250 | 3637 | if (!ret) { |
1bbc621e | 3638 | ret = write_one_cache_group(trans, root, path, cache); |
ff1f8250 FM |
3639 | /* |
3640 | * Our block group might still be attached to the list | |
3641 | * of new block groups in the transaction handle of some | |
3642 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3643 | * means its block group item isn't yet in the extent | |
3644 | * tree. If this happens ignore the error, as we will | |
3645 | * try again later in the critical section of the | |
3646 | * transaction commit. | |
3647 | */ | |
3648 | if (ret == -ENOENT) { | |
3649 | ret = 0; | |
3650 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3651 | if (list_empty(&cache->dirty_list)) { | |
3652 | list_add_tail(&cache->dirty_list, | |
3653 | &cur_trans->dirty_bgs); | |
3654 | btrfs_get_block_group(cache); | |
3655 | } | |
3656 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3657 | } else if (ret) { | |
66642832 | 3658 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3659 | } |
3660 | } | |
1bbc621e CM |
3661 | |
3662 | /* if its not on the io list, we need to put the block group */ | |
3663 | if (should_put) | |
3664 | btrfs_put_block_group(cache); | |
3665 | ||
3666 | if (ret) | |
3667 | break; | |
b58d1a9e FM |
3668 | |
3669 | /* | |
3670 | * Avoid blocking other tasks for too long. It might even save | |
3671 | * us from writing caches for block groups that are going to be | |
3672 | * removed. | |
3673 | */ | |
3674 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3675 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3676 | } |
b58d1a9e | 3677 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3678 | |
3679 | /* | |
3680 | * go through delayed refs for all the stuff we've just kicked off | |
3681 | * and then loop back (just once) | |
3682 | */ | |
3683 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
3684 | if (!ret && loops == 0) { | |
3685 | loops++; | |
3686 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3687 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3688 | /* |
3689 | * dirty_bgs_lock protects us from concurrent block group | |
3690 | * deletes too (not just cache_write_mutex). | |
3691 | */ | |
3692 | if (!list_empty(&dirty)) { | |
3693 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3694 | goto again; | |
3695 | } | |
1bbc621e | 3696 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 LB |
3697 | } else if (ret < 0) { |
3698 | btrfs_cleanup_dirty_bgs(cur_trans, root); | |
1bbc621e CM |
3699 | } |
3700 | ||
3701 | btrfs_free_path(path); | |
3702 | return ret; | |
3703 | } | |
3704 | ||
3705 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
3706 | struct btrfs_root *root) | |
3707 | { | |
3708 | struct btrfs_block_group_cache *cache; | |
3709 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3710 | int ret = 0; | |
3711 | int should_put; | |
3712 | struct btrfs_path *path; | |
3713 | struct list_head *io = &cur_trans->io_bgs; | |
3714 | int num_started = 0; | |
9078a3e1 CM |
3715 | |
3716 | path = btrfs_alloc_path(); | |
3717 | if (!path) | |
3718 | return -ENOMEM; | |
3719 | ||
ce93ec54 | 3720 | /* |
e44081ef FM |
3721 | * Even though we are in the critical section of the transaction commit, |
3722 | * we can still have concurrent tasks adding elements to this | |
3723 | * transaction's list of dirty block groups. These tasks correspond to | |
3724 | * endio free space workers started when writeback finishes for a | |
3725 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3726 | * allocate new block groups as a result of COWing nodes of the root | |
3727 | * tree when updating the free space inode. The writeback for the space | |
3728 | * caches is triggered by an earlier call to | |
3729 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3730 | * loop. | |
3731 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3732 | * delayed refs to make sure we have the best chance at doing this all |
3733 | * in one shot. | |
3734 | */ | |
e44081ef | 3735 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3736 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3737 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3738 | struct btrfs_block_group_cache, | |
3739 | dirty_list); | |
c9dc4c65 CM |
3740 | |
3741 | /* | |
3742 | * this can happen if cache_save_setup re-dirties a block | |
3743 | * group that is already under IO. Just wait for it to | |
3744 | * finish and then do it all again | |
3745 | */ | |
3746 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3747 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3748 | list_del_init(&cache->io_list); |
3749 | btrfs_wait_cache_io(root, trans, cache, | |
3750 | &cache->io_ctl, path, | |
3751 | cache->key.objectid); | |
3752 | btrfs_put_block_group(cache); | |
e44081ef | 3753 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3754 | } |
3755 | ||
1bbc621e CM |
3756 | /* |
3757 | * don't remove from the dirty list until after we've waited | |
3758 | * on any pending IO | |
3759 | */ | |
ce93ec54 | 3760 | list_del_init(&cache->dirty_list); |
e44081ef | 3761 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3762 | should_put = 1; |
3763 | ||
1bbc621e | 3764 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3765 | |
ce93ec54 | 3766 | if (!ret) |
c9dc4c65 CM |
3767 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1); |
3768 | ||
3769 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3770 | cache->io_ctl.inode = NULL; | |
3771 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3772 | if (ret == 0 && cache->io_ctl.inode) { | |
3773 | num_started++; | |
3774 | should_put = 0; | |
1bbc621e | 3775 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3776 | } else { |
3777 | /* | |
3778 | * if we failed to write the cache, the | |
3779 | * generation will be bad and life goes on | |
3780 | */ | |
3781 | ret = 0; | |
3782 | } | |
3783 | } | |
ff1f8250 | 3784 | if (!ret) { |
ce93ec54 | 3785 | ret = write_one_cache_group(trans, root, path, cache); |
2bc0bb5f FM |
3786 | /* |
3787 | * One of the free space endio workers might have | |
3788 | * created a new block group while updating a free space | |
3789 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3790 | * and hasn't released its transaction handle yet, in | |
3791 | * which case the new block group is still attached to | |
3792 | * its transaction handle and its creation has not | |
3793 | * finished yet (no block group item in the extent tree | |
3794 | * yet, etc). If this is the case, wait for all free | |
3795 | * space endio workers to finish and retry. This is a | |
3796 | * a very rare case so no need for a more efficient and | |
3797 | * complex approach. | |
3798 | */ | |
3799 | if (ret == -ENOENT) { | |
3800 | wait_event(cur_trans->writer_wait, | |
3801 | atomic_read(&cur_trans->num_writers) == 1); | |
3802 | ret = write_one_cache_group(trans, root, path, | |
3803 | cache); | |
3804 | } | |
ff1f8250 | 3805 | if (ret) |
66642832 | 3806 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3807 | } |
c9dc4c65 CM |
3808 | |
3809 | /* if its not on the io list, we need to put the block group */ | |
3810 | if (should_put) | |
3811 | btrfs_put_block_group(cache); | |
e44081ef | 3812 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3813 | } |
e44081ef | 3814 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3815 | |
1bbc621e CM |
3816 | while (!list_empty(io)) { |
3817 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3818 | io_list); |
3819 | list_del_init(&cache->io_list); | |
c9dc4c65 CM |
3820 | btrfs_wait_cache_io(root, trans, cache, |
3821 | &cache->io_ctl, path, cache->key.objectid); | |
0cb59c99 JB |
3822 | btrfs_put_block_group(cache); |
3823 | } | |
3824 | ||
9078a3e1 | 3825 | btrfs_free_path(path); |
ce93ec54 | 3826 | return ret; |
9078a3e1 CM |
3827 | } |
3828 | ||
d2fb3437 YZ |
3829 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3830 | { | |
3831 | struct btrfs_block_group_cache *block_group; | |
3832 | int readonly = 0; | |
3833 | ||
3834 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3835 | if (!block_group || block_group->ro) | |
3836 | readonly = 1; | |
3837 | if (block_group) | |
fa9c0d79 | 3838 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3839 | return readonly; |
3840 | } | |
3841 | ||
f78c436c FM |
3842 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3843 | { | |
3844 | struct btrfs_block_group_cache *bg; | |
3845 | bool ret = true; | |
3846 | ||
3847 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3848 | if (!bg) | |
3849 | return false; | |
3850 | ||
3851 | spin_lock(&bg->lock); | |
3852 | if (bg->ro) | |
3853 | ret = false; | |
3854 | else | |
3855 | atomic_inc(&bg->nocow_writers); | |
3856 | spin_unlock(&bg->lock); | |
3857 | ||
3858 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3859 | if (!ret) | |
3860 | btrfs_put_block_group(bg); | |
3861 | ||
3862 | return ret; | |
3863 | ||
3864 | } | |
3865 | ||
3866 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3867 | { | |
3868 | struct btrfs_block_group_cache *bg; | |
3869 | ||
3870 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3871 | ASSERT(bg); | |
3872 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
3873 | wake_up_atomic_t(&bg->nocow_writers); | |
3874 | /* | |
3875 | * Once for our lookup and once for the lookup done by a previous call | |
3876 | * to btrfs_inc_nocow_writers() | |
3877 | */ | |
3878 | btrfs_put_block_group(bg); | |
3879 | btrfs_put_block_group(bg); | |
3880 | } | |
3881 | ||
3882 | static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a) | |
3883 | { | |
3884 | schedule(); | |
3885 | return 0; | |
3886 | } | |
3887 | ||
3888 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) | |
3889 | { | |
3890 | wait_on_atomic_t(&bg->nocow_writers, | |
3891 | btrfs_wait_nocow_writers_atomic_t, | |
3892 | TASK_UNINTERRUPTIBLE); | |
3893 | } | |
3894 | ||
6ab0a202 JM |
3895 | static const char *alloc_name(u64 flags) |
3896 | { | |
3897 | switch (flags) { | |
3898 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3899 | return "mixed"; | |
3900 | case BTRFS_BLOCK_GROUP_METADATA: | |
3901 | return "metadata"; | |
3902 | case BTRFS_BLOCK_GROUP_DATA: | |
3903 | return "data"; | |
3904 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3905 | return "system"; | |
3906 | default: | |
3907 | WARN_ON(1); | |
3908 | return "invalid-combination"; | |
3909 | }; | |
3910 | } | |
3911 | ||
593060d7 CM |
3912 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3913 | u64 total_bytes, u64 bytes_used, | |
e40edf2d | 3914 | u64 bytes_readonly, |
593060d7 CM |
3915 | struct btrfs_space_info **space_info) |
3916 | { | |
3917 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3918 | int i; |
3919 | int factor; | |
b150a4f1 | 3920 | int ret; |
b742bb82 YZ |
3921 | |
3922 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3923 | BTRFS_BLOCK_GROUP_RAID10)) | |
3924 | factor = 2; | |
3925 | else | |
3926 | factor = 1; | |
593060d7 CM |
3927 | |
3928 | found = __find_space_info(info, flags); | |
3929 | if (found) { | |
25179201 | 3930 | spin_lock(&found->lock); |
593060d7 | 3931 | found->total_bytes += total_bytes; |
89a55897 | 3932 | found->disk_total += total_bytes * factor; |
593060d7 | 3933 | found->bytes_used += bytes_used; |
b742bb82 | 3934 | found->disk_used += bytes_used * factor; |
e40edf2d | 3935 | found->bytes_readonly += bytes_readonly; |
2e6e5183 FM |
3936 | if (total_bytes > 0) |
3937 | found->full = 0; | |
957780eb JB |
3938 | space_info_add_new_bytes(info, found, total_bytes - |
3939 | bytes_used - bytes_readonly); | |
25179201 | 3940 | spin_unlock(&found->lock); |
593060d7 CM |
3941 | *space_info = found; |
3942 | return 0; | |
3943 | } | |
c146afad | 3944 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3945 | if (!found) |
3946 | return -ENOMEM; | |
3947 | ||
908c7f19 | 3948 | ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL); |
b150a4f1 JB |
3949 | if (ret) { |
3950 | kfree(found); | |
3951 | return ret; | |
3952 | } | |
3953 | ||
c1895442 | 3954 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
b742bb82 | 3955 | INIT_LIST_HEAD(&found->block_groups[i]); |
80eb234a | 3956 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3957 | spin_lock_init(&found->lock); |
52ba6929 | 3958 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3959 | found->total_bytes = total_bytes; |
89a55897 | 3960 | found->disk_total = total_bytes * factor; |
593060d7 | 3961 | found->bytes_used = bytes_used; |
b742bb82 | 3962 | found->disk_used = bytes_used * factor; |
593060d7 | 3963 | found->bytes_pinned = 0; |
e8569813 | 3964 | found->bytes_reserved = 0; |
e40edf2d | 3965 | found->bytes_readonly = bytes_readonly; |
f0486c68 | 3966 | found->bytes_may_use = 0; |
6af3e3ad | 3967 | found->full = 0; |
4f4db217 | 3968 | found->max_extent_size = 0; |
0e4f8f88 | 3969 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3970 | found->chunk_alloc = 0; |
fdb5effd JB |
3971 | found->flush = 0; |
3972 | init_waitqueue_head(&found->wait); | |
633c0aad | 3973 | INIT_LIST_HEAD(&found->ro_bgs); |
957780eb JB |
3974 | INIT_LIST_HEAD(&found->tickets); |
3975 | INIT_LIST_HEAD(&found->priority_tickets); | |
6ab0a202 JM |
3976 | |
3977 | ret = kobject_init_and_add(&found->kobj, &space_info_ktype, | |
3978 | info->space_info_kobj, "%s", | |
3979 | alloc_name(found->flags)); | |
3980 | if (ret) { | |
3981 | kfree(found); | |
3982 | return ret; | |
3983 | } | |
3984 | ||
593060d7 | 3985 | *space_info = found; |
4184ea7f | 3986 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3987 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3988 | info->data_sinfo = found; | |
6ab0a202 JM |
3989 | |
3990 | return ret; | |
593060d7 CM |
3991 | } |
3992 | ||
8790d502 CM |
3993 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3994 | { | |
899c81ea ID |
3995 | u64 extra_flags = chunk_to_extended(flags) & |
3996 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3997 | |
de98ced9 | 3998 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3999 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4000 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4001 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4002 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4003 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4004 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4005 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4006 | } |
593060d7 | 4007 | |
fc67c450 ID |
4008 | /* |
4009 | * returns target flags in extended format or 0 if restripe for this | |
4010 | * chunk_type is not in progress | |
c6664b42 ID |
4011 | * |
4012 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
4013 | */ |
4014 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4015 | { | |
4016 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4017 | u64 target = 0; | |
4018 | ||
fc67c450 ID |
4019 | if (!bctl) |
4020 | return 0; | |
4021 | ||
4022 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4023 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4024 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4025 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4026 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4027 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4028 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4029 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4030 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4031 | } | |
4032 | ||
4033 | return target; | |
4034 | } | |
4035 | ||
a46d11a8 ID |
4036 | /* |
4037 | * @flags: available profiles in extended format (see ctree.h) | |
4038 | * | |
e4d8ec0f ID |
4039 | * Returns reduced profile in chunk format. If profile changing is in |
4040 | * progress (either running or paused) picks the target profile (if it's | |
4041 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4042 | */ |
48a3b636 | 4043 | static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 4044 | { |
95669976 | 4045 | u64 num_devices = root->fs_info->fs_devices->rw_devices; |
fc67c450 | 4046 | u64 target; |
9c170b26 ZL |
4047 | u64 raid_type; |
4048 | u64 allowed = 0; | |
a061fc8d | 4049 | |
fc67c450 ID |
4050 | /* |
4051 | * see if restripe for this chunk_type is in progress, if so | |
4052 | * try to reduce to the target profile | |
4053 | */ | |
e4d8ec0f | 4054 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
4055 | target = get_restripe_target(root->fs_info, flags); |
4056 | if (target) { | |
4057 | /* pick target profile only if it's already available */ | |
4058 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 4059 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 4060 | return extended_to_chunk(target); |
e4d8ec0f ID |
4061 | } |
4062 | } | |
4063 | spin_unlock(&root->fs_info->balance_lock); | |
4064 | ||
53b381b3 | 4065 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4066 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4067 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
4068 | allowed |= btrfs_raid_group[raid_type]; | |
4069 | } | |
4070 | allowed &= flags; | |
4071 | ||
4072 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4073 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4074 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4075 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4076 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4077 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4078 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4079 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4080 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4081 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4082 | ||
4083 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4084 | ||
4085 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4086 | } |
4087 | ||
f8213bdc | 4088 | static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags) |
6a63209f | 4089 | { |
de98ced9 | 4090 | unsigned seq; |
f8213bdc | 4091 | u64 flags; |
de98ced9 MX |
4092 | |
4093 | do { | |
f8213bdc | 4094 | flags = orig_flags; |
de98ced9 MX |
4095 | seq = read_seqbegin(&root->fs_info->profiles_lock); |
4096 | ||
4097 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4098 | flags |= root->fs_info->avail_data_alloc_bits; | |
4099 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4100 | flags |= root->fs_info->avail_system_alloc_bits; | |
4101 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4102 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
4103 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
6fef8df1 | 4104 | |
b742bb82 | 4105 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
4106 | } |
4107 | ||
6d07bcec | 4108 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 4109 | { |
b742bb82 | 4110 | u64 flags; |
53b381b3 | 4111 | u64 ret; |
9ed74f2d | 4112 | |
b742bb82 YZ |
4113 | if (data) |
4114 | flags = BTRFS_BLOCK_GROUP_DATA; | |
4115 | else if (root == root->fs_info->chunk_root) | |
4116 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 4117 | else |
b742bb82 | 4118 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4119 | |
53b381b3 DW |
4120 | ret = get_alloc_profile(root, flags); |
4121 | return ret; | |
6a63209f | 4122 | } |
9ed74f2d | 4123 | |
4ceff079 | 4124 | int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes) |
6a63209f | 4125 | { |
6a63209f | 4126 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 4127 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 4128 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 4129 | u64 used; |
94b947b2 | 4130 | int ret = 0; |
c99f1b0c ZL |
4131 | int need_commit = 2; |
4132 | int have_pinned_space; | |
6a63209f | 4133 | |
6a63209f | 4134 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 4135 | bytes = ALIGN(bytes, root->sectorsize); |
6a63209f | 4136 | |
9dced186 | 4137 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4138 | need_commit = 0; |
9dced186 | 4139 | ASSERT(current->journal_info); |
0af3d00b JB |
4140 | } |
4141 | ||
b4d7c3c9 | 4142 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
4143 | if (!data_sinfo) |
4144 | goto alloc; | |
9ed74f2d | 4145 | |
6a63209f JB |
4146 | again: |
4147 | /* make sure we have enough space to handle the data first */ | |
4148 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
4149 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
4150 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
4151 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
4152 | |
4153 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4154 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4155 | |
6a63209f JB |
4156 | /* |
4157 | * if we don't have enough free bytes in this space then we need | |
4158 | * to alloc a new chunk. | |
4159 | */ | |
b9fd47cd | 4160 | if (!data_sinfo->full) { |
6a63209f | 4161 | u64 alloc_target; |
9ed74f2d | 4162 | |
0e4f8f88 | 4163 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4164 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 4165 | alloc: |
6a63209f | 4166 | alloc_target = btrfs_get_alloc_profile(root, 1); |
9dced186 MX |
4167 | /* |
4168 | * It is ugly that we don't call nolock join | |
4169 | * transaction for the free space inode case here. | |
4170 | * But it is safe because we only do the data space | |
4171 | * reservation for the free space cache in the | |
4172 | * transaction context, the common join transaction | |
4173 | * just increase the counter of the current transaction | |
4174 | * handler, doesn't try to acquire the trans_lock of | |
4175 | * the fs. | |
4176 | */ | |
7a7eaa40 | 4177 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4178 | if (IS_ERR(trans)) |
4179 | return PTR_ERR(trans); | |
9ed74f2d | 4180 | |
6a63209f | 4181 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
4182 | alloc_target, |
4183 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 4184 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
4185 | if (ret < 0) { |
4186 | if (ret != -ENOSPC) | |
4187 | return ret; | |
c99f1b0c ZL |
4188 | else { |
4189 | have_pinned_space = 1; | |
d52a5b5f | 4190 | goto commit_trans; |
c99f1b0c | 4191 | } |
d52a5b5f | 4192 | } |
9ed74f2d | 4193 | |
b4d7c3c9 LZ |
4194 | if (!data_sinfo) |
4195 | data_sinfo = fs_info->data_sinfo; | |
4196 | ||
6a63209f JB |
4197 | goto again; |
4198 | } | |
f2bb8f5c JB |
4199 | |
4200 | /* | |
b150a4f1 | 4201 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4202 | * allocation, and no removed chunk in current transaction, |
4203 | * don't bother committing the transaction. | |
f2bb8f5c | 4204 | */ |
c99f1b0c ZL |
4205 | have_pinned_space = percpu_counter_compare( |
4206 | &data_sinfo->total_bytes_pinned, | |
4207 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4208 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4209 | |
4e06bdd6 | 4210 | /* commit the current transaction and try again */ |
d52a5b5f | 4211 | commit_trans: |
c99f1b0c | 4212 | if (need_commit && |
a4abeea4 | 4213 | !atomic_read(&root->fs_info->open_ioctl_trans)) { |
c99f1b0c | 4214 | need_commit--; |
b150a4f1 | 4215 | |
e1746e83 ZL |
4216 | if (need_commit > 0) { |
4217 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
578def7c | 4218 | btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1); |
e1746e83 | 4219 | } |
9a4e7276 | 4220 | |
7a7eaa40 | 4221 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4222 | if (IS_ERR(trans)) |
4223 | return PTR_ERR(trans); | |
c99f1b0c | 4224 | if (have_pinned_space >= 0 || |
3204d33c JB |
4225 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4226 | &trans->transaction->flags) || | |
c99f1b0c | 4227 | need_commit > 0) { |
94b947b2 ZL |
4228 | ret = btrfs_commit_transaction(trans, root); |
4229 | if (ret) | |
4230 | return ret; | |
d7c15171 | 4231 | /* |
c2d6cb16 FM |
4232 | * The cleaner kthread might still be doing iput |
4233 | * operations. Wait for it to finish so that | |
4234 | * more space is released. | |
d7c15171 | 4235 | */ |
c2d6cb16 FM |
4236 | mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex); |
4237 | mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4238 | goto again; |
4239 | } else { | |
4240 | btrfs_end_transaction(trans, root); | |
4241 | } | |
4e06bdd6 | 4242 | } |
9ed74f2d | 4243 | |
cab45e22 JM |
4244 | trace_btrfs_space_reservation(root->fs_info, |
4245 | "space_info:enospc", | |
4246 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4247 | return -ENOSPC; |
4248 | } | |
4249 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 4250 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 4251 | data_sinfo->flags, bytes, 1); |
6a63209f | 4252 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4253 | |
237c0e9f | 4254 | return ret; |
9ed74f2d | 4255 | } |
6a63209f | 4256 | |
4ceff079 QW |
4257 | /* |
4258 | * New check_data_free_space() with ability for precious data reservation | |
4259 | * Will replace old btrfs_check_data_free_space(), but for patch split, | |
4260 | * add a new function first and then replace it. | |
4261 | */ | |
7cf5b976 | 4262 | int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len) |
4ceff079 QW |
4263 | { |
4264 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4265 | int ret; | |
4266 | ||
4267 | /* align the range */ | |
4268 | len = round_up(start + len, root->sectorsize) - | |
4269 | round_down(start, root->sectorsize); | |
4270 | start = round_down(start, root->sectorsize); | |
4271 | ||
4272 | ret = btrfs_alloc_data_chunk_ondemand(inode, len); | |
4273 | if (ret < 0) | |
4274 | return ret; | |
4275 | ||
1e5ec2e7 | 4276 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
4ceff079 | 4277 | ret = btrfs_qgroup_reserve_data(inode, start, len); |
1e5ec2e7 JB |
4278 | if (ret) |
4279 | btrfs_free_reserved_data_space_noquota(inode, start, len); | |
4ceff079 QW |
4280 | return ret; |
4281 | } | |
4282 | ||
4ceff079 QW |
4283 | /* |
4284 | * Called if we need to clear a data reservation for this inode | |
4285 | * Normally in a error case. | |
4286 | * | |
51773bec QW |
4287 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4288 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4289 | * Like clear_bit_hook(). | |
4ceff079 | 4290 | */ |
51773bec QW |
4291 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4292 | u64 len) | |
4ceff079 QW |
4293 | { |
4294 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4295 | struct btrfs_space_info *data_sinfo; | |
4296 | ||
4297 | /* Make sure the range is aligned to sectorsize */ | |
4298 | len = round_up(start + len, root->sectorsize) - | |
4299 | round_down(start, root->sectorsize); | |
4300 | start = round_down(start, root->sectorsize); | |
4301 | ||
4ceff079 QW |
4302 | data_sinfo = root->fs_info->data_sinfo; |
4303 | spin_lock(&data_sinfo->lock); | |
4304 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4305 | data_sinfo->bytes_may_use = 0; | |
4306 | else | |
4307 | data_sinfo->bytes_may_use -= len; | |
4308 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4309 | data_sinfo->flags, len, 0); | |
4310 | spin_unlock(&data_sinfo->lock); | |
4311 | } | |
4312 | ||
51773bec QW |
4313 | /* |
4314 | * Called if we need to clear a data reservation for this inode | |
4315 | * Normally in a error case. | |
4316 | * | |
01327610 | 4317 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4318 | * space framework. |
4319 | */ | |
4320 | void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len) | |
4321 | { | |
4322 | btrfs_free_reserved_data_space_noquota(inode, start, len); | |
4323 | btrfs_qgroup_free_data(inode, start, len); | |
4324 | } | |
4325 | ||
97e728d4 | 4326 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4327 | { |
97e728d4 JB |
4328 | struct list_head *head = &info->space_info; |
4329 | struct btrfs_space_info *found; | |
e3ccfa98 | 4330 | |
97e728d4 JB |
4331 | rcu_read_lock(); |
4332 | list_for_each_entry_rcu(found, head, list) { | |
4333 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4334 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4335 | } |
97e728d4 | 4336 | rcu_read_unlock(); |
e3ccfa98 JB |
4337 | } |
4338 | ||
3c76cd84 MX |
4339 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4340 | { | |
4341 | return (global->size << 1); | |
4342 | } | |
4343 | ||
e5bc2458 | 4344 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 4345 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4346 | { |
fb25e914 | 4347 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 4348 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 4349 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 4350 | u64 thresh; |
e3ccfa98 | 4351 | |
0e4f8f88 CM |
4352 | if (force == CHUNK_ALLOC_FORCE) |
4353 | return 1; | |
4354 | ||
fb25e914 JB |
4355 | /* |
4356 | * We need to take into account the global rsv because for all intents | |
4357 | * and purposes it's used space. Don't worry about locking the | |
4358 | * global_rsv, it doesn't change except when the transaction commits. | |
4359 | */ | |
54338b5c | 4360 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3c76cd84 | 4361 | num_allocated += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4362 | |
0e4f8f88 CM |
4363 | /* |
4364 | * in limited mode, we want to have some free space up to | |
4365 | * about 1% of the FS size. | |
4366 | */ | |
4367 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 4368 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
ee22184b | 4369 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 CM |
4370 | |
4371 | if (num_bytes - num_allocated < thresh) | |
4372 | return 1; | |
4373 | } | |
0e4f8f88 | 4374 | |
ee22184b | 4375 | if (num_allocated + SZ_2M < div_factor(num_bytes, 8)) |
14ed0ca6 | 4376 | return 0; |
424499db | 4377 | return 1; |
32c00aff JB |
4378 | } |
4379 | ||
39c2d7fa | 4380 | static u64 get_profile_num_devs(struct btrfs_root *root, u64 type) |
15d1ff81 LB |
4381 | { |
4382 | u64 num_dev; | |
4383 | ||
53b381b3 DW |
4384 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4385 | BTRFS_BLOCK_GROUP_RAID0 | | |
4386 | BTRFS_BLOCK_GROUP_RAID5 | | |
4387 | BTRFS_BLOCK_GROUP_RAID6)) | |
15d1ff81 LB |
4388 | num_dev = root->fs_info->fs_devices->rw_devices; |
4389 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
4390 | num_dev = 2; | |
4391 | else | |
4392 | num_dev = 1; /* DUP or single */ | |
4393 | ||
39c2d7fa | 4394 | return num_dev; |
15d1ff81 LB |
4395 | } |
4396 | ||
39c2d7fa FM |
4397 | /* |
4398 | * If @is_allocation is true, reserve space in the system space info necessary | |
4399 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4400 | * removing a chunk. | |
4401 | */ | |
4402 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
4403 | struct btrfs_root *root, | |
4617ea3a | 4404 | u64 type) |
15d1ff81 LB |
4405 | { |
4406 | struct btrfs_space_info *info; | |
4407 | u64 left; | |
4408 | u64 thresh; | |
4fbcdf66 | 4409 | int ret = 0; |
39c2d7fa | 4410 | u64 num_devs; |
4fbcdf66 FM |
4411 | |
4412 | /* | |
4413 | * Needed because we can end up allocating a system chunk and for an | |
4414 | * atomic and race free space reservation in the chunk block reserve. | |
4415 | */ | |
4416 | ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex)); | |
15d1ff81 LB |
4417 | |
4418 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4419 | spin_lock(&info->lock); | |
4420 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
4fbcdf66 FM |
4421 | info->bytes_reserved - info->bytes_readonly - |
4422 | info->bytes_may_use; | |
15d1ff81 LB |
4423 | spin_unlock(&info->lock); |
4424 | ||
39c2d7fa FM |
4425 | num_devs = get_profile_num_devs(root, type); |
4426 | ||
4427 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
4617ea3a FM |
4428 | thresh = btrfs_calc_trunc_metadata_size(root, num_devs) + |
4429 | btrfs_calc_trans_metadata_size(root, 1); | |
39c2d7fa | 4430 | |
3cdde224 | 4431 | if (left < thresh && btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) { |
c2cf52eb SK |
4432 | btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu", |
4433 | left, thresh, type); | |
ab8d0fc4 | 4434 | dump_space_info(root->fs_info, info, 0, 0); |
15d1ff81 LB |
4435 | } |
4436 | ||
4437 | if (left < thresh) { | |
4438 | u64 flags; | |
4439 | ||
4440 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
4fbcdf66 FM |
4441 | /* |
4442 | * Ignore failure to create system chunk. We might end up not | |
4443 | * needing it, as we might not need to COW all nodes/leafs from | |
4444 | * the paths we visit in the chunk tree (they were already COWed | |
4445 | * or created in the current transaction for example). | |
4446 | */ | |
4447 | ret = btrfs_alloc_chunk(trans, root, flags); | |
4448 | } | |
4449 | ||
4450 | if (!ret) { | |
4451 | ret = btrfs_block_rsv_add(root->fs_info->chunk_root, | |
4452 | &root->fs_info->chunk_block_rsv, | |
4453 | thresh, BTRFS_RESERVE_NO_FLUSH); | |
4454 | if (!ret) | |
4455 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4456 | } |
4457 | } | |
4458 | ||
28b737f6 LB |
4459 | /* |
4460 | * If force is CHUNK_ALLOC_FORCE: | |
4461 | * - return 1 if it successfully allocates a chunk, | |
4462 | * - return errors including -ENOSPC otherwise. | |
4463 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4464 | * - return 0 if it doesn't need to allocate a new chunk, | |
4465 | * - return 1 if it successfully allocates a chunk, | |
4466 | * - return errors including -ENOSPC otherwise. | |
4467 | */ | |
6324fbf3 | 4468 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 4469 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 4470 | { |
6324fbf3 | 4471 | struct btrfs_space_info *space_info; |
97e728d4 | 4472 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 4473 | int wait_for_alloc = 0; |
9ed74f2d | 4474 | int ret = 0; |
9ed74f2d | 4475 | |
c6b305a8 JB |
4476 | /* Don't re-enter if we're already allocating a chunk */ |
4477 | if (trans->allocating_chunk) | |
4478 | return -ENOSPC; | |
4479 | ||
6324fbf3 | 4480 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
4481 | if (!space_info) { |
4482 | ret = update_space_info(extent_root->fs_info, flags, | |
e40edf2d | 4483 | 0, 0, 0, &space_info); |
79787eaa | 4484 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 4485 | } |
79787eaa | 4486 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 4487 | |
6d74119f | 4488 | again: |
25179201 | 4489 | spin_lock(&space_info->lock); |
9e622d6b | 4490 | if (force < space_info->force_alloc) |
0e4f8f88 | 4491 | force = space_info->force_alloc; |
25179201 | 4492 | if (space_info->full) { |
09fb99a6 FDBM |
4493 | if (should_alloc_chunk(extent_root, space_info, force)) |
4494 | ret = -ENOSPC; | |
4495 | else | |
4496 | ret = 0; | |
25179201 | 4497 | spin_unlock(&space_info->lock); |
09fb99a6 | 4498 | return ret; |
9ed74f2d JB |
4499 | } |
4500 | ||
698d0082 | 4501 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 4502 | spin_unlock(&space_info->lock); |
6d74119f JB |
4503 | return 0; |
4504 | } else if (space_info->chunk_alloc) { | |
4505 | wait_for_alloc = 1; | |
4506 | } else { | |
4507 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4508 | } |
0e4f8f88 | 4509 | |
25179201 | 4510 | spin_unlock(&space_info->lock); |
9ed74f2d | 4511 | |
6d74119f JB |
4512 | mutex_lock(&fs_info->chunk_mutex); |
4513 | ||
4514 | /* | |
4515 | * The chunk_mutex is held throughout the entirety of a chunk | |
4516 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4517 | * other guy is done and we need to recheck and see if we should | |
4518 | * allocate. | |
4519 | */ | |
4520 | if (wait_for_alloc) { | |
4521 | mutex_unlock(&fs_info->chunk_mutex); | |
4522 | wait_for_alloc = 0; | |
4523 | goto again; | |
4524 | } | |
4525 | ||
c6b305a8 JB |
4526 | trans->allocating_chunk = true; |
4527 | ||
67377734 JB |
4528 | /* |
4529 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4530 | * allocating mixed chunks instead of individual chunks. | |
4531 | */ | |
4532 | if (btrfs_mixed_space_info(space_info)) | |
4533 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4534 | ||
97e728d4 JB |
4535 | /* |
4536 | * if we're doing a data chunk, go ahead and make sure that | |
4537 | * we keep a reasonable number of metadata chunks allocated in the | |
4538 | * FS as well. | |
4539 | */ | |
9ed74f2d | 4540 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4541 | fs_info->data_chunk_allocations++; |
4542 | if (!(fs_info->data_chunk_allocations % | |
4543 | fs_info->metadata_ratio)) | |
4544 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4545 | } |
4546 | ||
15d1ff81 LB |
4547 | /* |
4548 | * Check if we have enough space in SYSTEM chunk because we may need | |
4549 | * to update devices. | |
4550 | */ | |
4617ea3a | 4551 | check_system_chunk(trans, extent_root, flags); |
15d1ff81 | 4552 | |
2b82032c | 4553 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 4554 | trans->allocating_chunk = false; |
92b8e897 | 4555 | |
9ed74f2d | 4556 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4557 | if (ret < 0 && ret != -ENOSPC) |
4558 | goto out; | |
9ed74f2d | 4559 | if (ret) |
6324fbf3 | 4560 | space_info->full = 1; |
424499db YZ |
4561 | else |
4562 | ret = 1; | |
6d74119f | 4563 | |
0e4f8f88 | 4564 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4565 | out: |
6d74119f | 4566 | space_info->chunk_alloc = 0; |
9ed74f2d | 4567 | spin_unlock(&space_info->lock); |
a25c75d5 | 4568 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4569 | /* |
4570 | * When we allocate a new chunk we reserve space in the chunk block | |
4571 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4572 | * add new nodes/leafs to it if we end up needing to do it when | |
4573 | * inserting the chunk item and updating device items as part of the | |
4574 | * second phase of chunk allocation, performed by | |
4575 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4576 | * large number of new block groups to create in our transaction | |
4577 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4578 | * in extreme cases - like having a single transaction create many new | |
4579 | * block groups when starting to write out the free space caches of all | |
4580 | * the block groups that were made dirty during the lifetime of the | |
4581 | * transaction. | |
4582 | */ | |
d9a0540a | 4583 | if (trans->can_flush_pending_bgs && |
ee22184b | 4584 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
64b63580 | 4585 | btrfs_create_pending_block_groups(trans, extent_root); |
00d80e34 FM |
4586 | btrfs_trans_release_chunk_metadata(trans); |
4587 | } | |
0f9dd46c | 4588 | return ret; |
6324fbf3 | 4589 | } |
9ed74f2d | 4590 | |
a80c8dcf JB |
4591 | static int can_overcommit(struct btrfs_root *root, |
4592 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 4593 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf | 4594 | { |
957780eb JB |
4595 | struct btrfs_block_rsv *global_rsv; |
4596 | u64 profile; | |
3c76cd84 | 4597 | u64 space_size; |
a80c8dcf JB |
4598 | u64 avail; |
4599 | u64 used; | |
4600 | ||
957780eb JB |
4601 | /* Don't overcommit when in mixed mode. */ |
4602 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4603 | return 0; | |
4604 | ||
4605 | BUG_ON(root->fs_info == NULL); | |
4606 | global_rsv = &root->fs_info->global_block_rsv; | |
4607 | profile = btrfs_get_alloc_profile(root, 0); | |
a80c8dcf | 4608 | used = space_info->bytes_used + space_info->bytes_reserved + |
96f1bb57 JB |
4609 | space_info->bytes_pinned + space_info->bytes_readonly; |
4610 | ||
96f1bb57 JB |
4611 | /* |
4612 | * We only want to allow over committing if we have lots of actual space | |
4613 | * free, but if we don't have enough space to handle the global reserve | |
4614 | * space then we could end up having a real enospc problem when trying | |
4615 | * to allocate a chunk or some other such important allocation. | |
4616 | */ | |
3c76cd84 MX |
4617 | spin_lock(&global_rsv->lock); |
4618 | space_size = calc_global_rsv_need_space(global_rsv); | |
4619 | spin_unlock(&global_rsv->lock); | |
4620 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4621 | return 0; |
4622 | ||
4623 | used += space_info->bytes_may_use; | |
a80c8dcf JB |
4624 | |
4625 | spin_lock(&root->fs_info->free_chunk_lock); | |
4626 | avail = root->fs_info->free_chunk_space; | |
4627 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4628 | ||
4629 | /* | |
4630 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4631 | * space is actually useable. For raid56, the space info used |
4632 | * doesn't include the parity drive, so we don't have to | |
4633 | * change the math | |
a80c8dcf JB |
4634 | */ |
4635 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4636 | BTRFS_BLOCK_GROUP_RAID1 | | |
4637 | BTRFS_BLOCK_GROUP_RAID10)) | |
4638 | avail >>= 1; | |
4639 | ||
4640 | /* | |
561c294d MX |
4641 | * If we aren't flushing all things, let us overcommit up to |
4642 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4643 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4644 | */ |
08e007d2 | 4645 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4646 | avail >>= 3; |
a80c8dcf | 4647 | else |
14575aef | 4648 | avail >>= 1; |
a80c8dcf | 4649 | |
14575aef | 4650 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4651 | return 1; |
4652 | return 0; | |
4653 | } | |
4654 | ||
48a3b636 | 4655 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
6c255e67 | 4656 | unsigned long nr_pages, int nr_items) |
da633a42 MX |
4657 | { |
4658 | struct super_block *sb = root->fs_info->sb; | |
da633a42 | 4659 | |
925a6efb JB |
4660 | if (down_read_trylock(&sb->s_umount)) { |
4661 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4662 | up_read(&sb->s_umount); | |
4663 | } else { | |
da633a42 MX |
4664 | /* |
4665 | * We needn't worry the filesystem going from r/w to r/o though | |
4666 | * we don't acquire ->s_umount mutex, because the filesystem | |
4667 | * should guarantee the delalloc inodes list be empty after | |
4668 | * the filesystem is readonly(all dirty pages are written to | |
4669 | * the disk). | |
4670 | */ | |
6c255e67 | 4671 | btrfs_start_delalloc_roots(root->fs_info, 0, nr_items); |
98ad69cf | 4672 | if (!current->journal_info) |
578def7c FM |
4673 | btrfs_wait_ordered_roots(root->fs_info, nr_items, |
4674 | 0, (u64)-1); | |
da633a42 MX |
4675 | } |
4676 | } | |
4677 | ||
18cd8ea6 MX |
4678 | static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim) |
4679 | { | |
4680 | u64 bytes; | |
4681 | int nr; | |
4682 | ||
4683 | bytes = btrfs_calc_trans_metadata_size(root, 1); | |
4684 | nr = (int)div64_u64(to_reclaim, bytes); | |
4685 | if (!nr) | |
4686 | nr = 1; | |
4687 | return nr; | |
4688 | } | |
4689 | ||
ee22184b | 4690 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4691 | |
9ed74f2d | 4692 | /* |
5da9d01b | 4693 | * shrink metadata reservation for delalloc |
9ed74f2d | 4694 | */ |
f4c738c2 JB |
4695 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
4696 | bool wait_ordered) | |
5da9d01b | 4697 | { |
0ca1f7ce | 4698 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4699 | struct btrfs_space_info *space_info; |
663350ac | 4700 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4701 | u64 delalloc_bytes; |
5da9d01b | 4702 | u64 max_reclaim; |
b1953bce | 4703 | long time_left; |
d3ee29e3 MX |
4704 | unsigned long nr_pages; |
4705 | int loops; | |
b0244199 | 4706 | int items; |
08e007d2 | 4707 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4708 | |
c61a16a7 | 4709 | /* Calc the number of the pages we need flush for space reservation */ |
b0244199 | 4710 | items = calc_reclaim_items_nr(root, to_reclaim); |
8eb0dfdb | 4711 | to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4712 | |
663350ac | 4713 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 4714 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 4715 | space_info = block_rsv->space_info; |
bf9022e0 | 4716 | |
963d678b MX |
4717 | delalloc_bytes = percpu_counter_sum_positive( |
4718 | &root->fs_info->delalloc_bytes); | |
f4c738c2 | 4719 | if (delalloc_bytes == 0) { |
fdb5effd | 4720 | if (trans) |
f4c738c2 | 4721 | return; |
38c135af | 4722 | if (wait_ordered) |
578def7c FM |
4723 | btrfs_wait_ordered_roots(root->fs_info, items, |
4724 | 0, (u64)-1); | |
f4c738c2 | 4725 | return; |
fdb5effd JB |
4726 | } |
4727 | ||
d3ee29e3 | 4728 | loops = 0; |
f4c738c2 JB |
4729 | while (delalloc_bytes && loops < 3) { |
4730 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4731 | nr_pages = max_reclaim >> PAGE_SHIFT; |
6c255e67 | 4732 | btrfs_writeback_inodes_sb_nr(root, nr_pages, items); |
dea31f52 JB |
4733 | /* |
4734 | * We need to wait for the async pages to actually start before | |
4735 | * we do anything. | |
4736 | */ | |
9f3a074d MX |
4737 | max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages); |
4738 | if (!max_reclaim) | |
4739 | goto skip_async; | |
4740 | ||
4741 | if (max_reclaim <= nr_pages) | |
4742 | max_reclaim = 0; | |
4743 | else | |
4744 | max_reclaim -= nr_pages; | |
dea31f52 | 4745 | |
9f3a074d MX |
4746 | wait_event(root->fs_info->async_submit_wait, |
4747 | atomic_read(&root->fs_info->async_delalloc_pages) <= | |
4748 | (int)max_reclaim); | |
4749 | skip_async: | |
08e007d2 MX |
4750 | if (!trans) |
4751 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4752 | else | |
4753 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4754 | spin_lock(&space_info->lock); |
08e007d2 | 4755 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
4756 | spin_unlock(&space_info->lock); |
4757 | break; | |
4758 | } | |
957780eb JB |
4759 | if (list_empty(&space_info->tickets) && |
4760 | list_empty(&space_info->priority_tickets)) { | |
4761 | spin_unlock(&space_info->lock); | |
4762 | break; | |
4763 | } | |
0019f10d | 4764 | spin_unlock(&space_info->lock); |
5da9d01b | 4765 | |
36e39c40 | 4766 | loops++; |
f104d044 | 4767 | if (wait_ordered && !trans) { |
578def7c FM |
4768 | btrfs_wait_ordered_roots(root->fs_info, items, |
4769 | 0, (u64)-1); | |
f104d044 | 4770 | } else { |
f4c738c2 | 4771 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4772 | if (time_left) |
4773 | break; | |
4774 | } | |
963d678b MX |
4775 | delalloc_bytes = percpu_counter_sum_positive( |
4776 | &root->fs_info->delalloc_bytes); | |
5da9d01b | 4777 | } |
5da9d01b YZ |
4778 | } |
4779 | ||
663350ac JB |
4780 | /** |
4781 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4782 | * @root - the root we're allocating for | |
4783 | * @bytes - the number of bytes we want to reserve | |
4784 | * @force - force the commit | |
8bb8ab2e | 4785 | * |
663350ac JB |
4786 | * This will check to make sure that committing the transaction will actually |
4787 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4788 | * will return -ENOSPC. | |
8bb8ab2e | 4789 | */ |
663350ac JB |
4790 | static int may_commit_transaction(struct btrfs_root *root, |
4791 | struct btrfs_space_info *space_info, | |
4792 | u64 bytes, int force) | |
4793 | { | |
4794 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
4795 | struct btrfs_trans_handle *trans; | |
4796 | ||
4797 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4798 | if (trans) | |
4799 | return -EAGAIN; | |
4800 | ||
4801 | if (force) | |
4802 | goto commit; | |
4803 | ||
4804 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4805 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4806 | bytes) >= 0) |
663350ac | 4807 | goto commit; |
663350ac JB |
4808 | |
4809 | /* | |
4810 | * See if there is some space in the delayed insertion reservation for | |
4811 | * this reservation. | |
4812 | */ | |
4813 | if (space_info != delayed_rsv->space_info) | |
4814 | return -ENOSPC; | |
4815 | ||
4816 | spin_lock(&delayed_rsv->lock); | |
b150a4f1 JB |
4817 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
4818 | bytes - delayed_rsv->size) >= 0) { | |
663350ac JB |
4819 | spin_unlock(&delayed_rsv->lock); |
4820 | return -ENOSPC; | |
4821 | } | |
4822 | spin_unlock(&delayed_rsv->lock); | |
4823 | ||
4824 | commit: | |
4825 | trans = btrfs_join_transaction(root); | |
4826 | if (IS_ERR(trans)) | |
4827 | return -ENOSPC; | |
4828 | ||
4829 | return btrfs_commit_transaction(trans, root); | |
4830 | } | |
4831 | ||
957780eb JB |
4832 | struct reserve_ticket { |
4833 | u64 bytes; | |
4834 | int error; | |
4835 | struct list_head list; | |
4836 | wait_queue_head_t wait; | |
96c3f433 JB |
4837 | }; |
4838 | ||
4839 | static int flush_space(struct btrfs_root *root, | |
4840 | struct btrfs_space_info *space_info, u64 num_bytes, | |
4841 | u64 orig_bytes, int state) | |
4842 | { | |
4843 | struct btrfs_trans_handle *trans; | |
4844 | int nr; | |
f4c738c2 | 4845 | int ret = 0; |
96c3f433 JB |
4846 | |
4847 | switch (state) { | |
96c3f433 JB |
4848 | case FLUSH_DELAYED_ITEMS_NR: |
4849 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 MX |
4850 | if (state == FLUSH_DELAYED_ITEMS_NR) |
4851 | nr = calc_reclaim_items_nr(root, num_bytes) * 2; | |
4852 | else | |
96c3f433 | 4853 | nr = -1; |
18cd8ea6 | 4854 | |
96c3f433 JB |
4855 | trans = btrfs_join_transaction(root); |
4856 | if (IS_ERR(trans)) { | |
4857 | ret = PTR_ERR(trans); | |
4858 | break; | |
4859 | } | |
4860 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
4861 | btrfs_end_transaction(trans, root); | |
4862 | break; | |
67b0fd63 JB |
4863 | case FLUSH_DELALLOC: |
4864 | case FLUSH_DELALLOC_WAIT: | |
24af7dd1 | 4865 | shrink_delalloc(root, num_bytes * 2, orig_bytes, |
67b0fd63 JB |
4866 | state == FLUSH_DELALLOC_WAIT); |
4867 | break; | |
ea658bad JB |
4868 | case ALLOC_CHUNK: |
4869 | trans = btrfs_join_transaction(root); | |
4870 | if (IS_ERR(trans)) { | |
4871 | ret = PTR_ERR(trans); | |
4872 | break; | |
4873 | } | |
4874 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
4875 | btrfs_get_alloc_profile(root, 0), |
4876 | CHUNK_ALLOC_NO_FORCE); | |
4877 | btrfs_end_transaction(trans, root); | |
eecba891 | 4878 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
4879 | ret = 0; |
4880 | break; | |
96c3f433 JB |
4881 | case COMMIT_TRANS: |
4882 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
4883 | break; | |
4884 | default: | |
4885 | ret = -ENOSPC; | |
4886 | break; | |
4887 | } | |
4888 | ||
f376df2b JB |
4889 | trace_btrfs_flush_space(root->fs_info, space_info->flags, num_bytes, |
4890 | orig_bytes, state, ret); | |
96c3f433 JB |
4891 | return ret; |
4892 | } | |
21c7e756 MX |
4893 | |
4894 | static inline u64 | |
4895 | btrfs_calc_reclaim_metadata_size(struct btrfs_root *root, | |
4896 | struct btrfs_space_info *space_info) | |
4897 | { | |
957780eb | 4898 | struct reserve_ticket *ticket; |
21c7e756 MX |
4899 | u64 used; |
4900 | u64 expected; | |
957780eb | 4901 | u64 to_reclaim = 0; |
21c7e756 | 4902 | |
957780eb JB |
4903 | list_for_each_entry(ticket, &space_info->tickets, list) |
4904 | to_reclaim += ticket->bytes; | |
4905 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
4906 | to_reclaim += ticket->bytes; | |
4907 | if (to_reclaim) | |
4908 | return to_reclaim; | |
21c7e756 | 4909 | |
e0af2484 WX |
4910 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
4911 | if (can_overcommit(root, space_info, to_reclaim, | |
4912 | BTRFS_RESERVE_FLUSH_ALL)) | |
4913 | return 0; | |
4914 | ||
21c7e756 MX |
4915 | used = space_info->bytes_used + space_info->bytes_reserved + |
4916 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4917 | space_info->bytes_may_use; | |
ee22184b | 4918 | if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL)) |
21c7e756 MX |
4919 | expected = div_factor_fine(space_info->total_bytes, 95); |
4920 | else | |
4921 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4922 | ||
4923 | if (used > expected) | |
4924 | to_reclaim = used - expected; | |
4925 | else | |
4926 | to_reclaim = 0; | |
4927 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4928 | space_info->bytes_reserved); | |
21c7e756 MX |
4929 | return to_reclaim; |
4930 | } | |
4931 | ||
4932 | static inline int need_do_async_reclaim(struct btrfs_space_info *space_info, | |
87241c2e | 4933 | struct btrfs_root *root, u64 used) |
21c7e756 | 4934 | { |
365c5313 JB |
4935 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4936 | ||
4937 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 4938 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
4939 | return 0; |
4940 | ||
87241c2e | 4941 | if (!btrfs_calc_reclaim_metadata_size(root, space_info)) |
d38b349c JB |
4942 | return 0; |
4943 | ||
87241c2e JB |
4944 | return (used >= thresh && !btrfs_fs_closing(root->fs_info) && |
4945 | !test_bit(BTRFS_FS_STATE_REMOUNTING, | |
4946 | &root->fs_info->fs_state)); | |
21c7e756 MX |
4947 | } |
4948 | ||
957780eb | 4949 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 4950 | { |
957780eb | 4951 | struct reserve_ticket *ticket; |
25ce459c | 4952 | |
957780eb JB |
4953 | while (!list_empty(head)) { |
4954 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
4955 | list_del_init(&ticket->list); | |
4956 | ticket->error = -ENOSPC; | |
4957 | wake_up(&ticket->wait); | |
21c7e756 | 4958 | } |
21c7e756 MX |
4959 | } |
4960 | ||
957780eb JB |
4961 | /* |
4962 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
4963 | * will loop and continuously try to flush as long as we are making progress. | |
4964 | * We count progress as clearing off tickets each time we have to loop. | |
4965 | */ | |
21c7e756 MX |
4966 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
4967 | { | |
4968 | struct btrfs_fs_info *fs_info; | |
4969 | struct btrfs_space_info *space_info; | |
4970 | u64 to_reclaim; | |
4971 | int flush_state; | |
957780eb | 4972 | int commit_cycles = 0; |
ce129655 | 4973 | u64 last_tickets_id; |
21c7e756 MX |
4974 | |
4975 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4976 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4977 | ||
957780eb | 4978 | spin_lock(&space_info->lock); |
21c7e756 MX |
4979 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, |
4980 | space_info); | |
957780eb JB |
4981 | if (!to_reclaim) { |
4982 | space_info->flush = 0; | |
4983 | spin_unlock(&space_info->lock); | |
21c7e756 | 4984 | return; |
957780eb | 4985 | } |
ce129655 | 4986 | last_tickets_id = space_info->tickets_id; |
957780eb | 4987 | spin_unlock(&space_info->lock); |
21c7e756 MX |
4988 | |
4989 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb JB |
4990 | do { |
4991 | struct reserve_ticket *ticket; | |
4992 | int ret; | |
4993 | ||
4994 | ret = flush_space(fs_info->fs_root, space_info, to_reclaim, | |
4995 | to_reclaim, flush_state); | |
4996 | spin_lock(&space_info->lock); | |
4997 | if (list_empty(&space_info->tickets)) { | |
4998 | space_info->flush = 0; | |
4999 | spin_unlock(&space_info->lock); | |
5000 | return; | |
5001 | } | |
5002 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, | |
5003 | space_info); | |
5004 | ticket = list_first_entry(&space_info->tickets, | |
5005 | struct reserve_ticket, list); | |
ce129655 | 5006 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5007 | flush_state++; |
5008 | } else { | |
ce129655 | 5009 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5010 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5011 | if (commit_cycles) | |
5012 | commit_cycles--; | |
5013 | } | |
5014 | ||
5015 | if (flush_state > COMMIT_TRANS) { | |
5016 | commit_cycles++; | |
5017 | if (commit_cycles > 2) { | |
5018 | wake_all_tickets(&space_info->tickets); | |
5019 | space_info->flush = 0; | |
5020 | } else { | |
5021 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5022 | } | |
5023 | } | |
5024 | spin_unlock(&space_info->lock); | |
5025 | } while (flush_state <= COMMIT_TRANS); | |
5026 | } | |
5027 | ||
5028 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5029 | { | |
5030 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5031 | } | |
5032 | ||
5033 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5034 | struct btrfs_space_info *space_info, | |
5035 | struct reserve_ticket *ticket) | |
5036 | { | |
5037 | u64 to_reclaim; | |
5038 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5039 | ||
5040 | spin_lock(&space_info->lock); | |
5041 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, | |
5042 | space_info); | |
5043 | if (!to_reclaim) { | |
5044 | spin_unlock(&space_info->lock); | |
5045 | return; | |
5046 | } | |
5047 | spin_unlock(&space_info->lock); | |
5048 | ||
21c7e756 MX |
5049 | do { |
5050 | flush_space(fs_info->fs_root, space_info, to_reclaim, | |
5051 | to_reclaim, flush_state); | |
5052 | flush_state++; | |
957780eb JB |
5053 | spin_lock(&space_info->lock); |
5054 | if (ticket->bytes == 0) { | |
5055 | spin_unlock(&space_info->lock); | |
21c7e756 | 5056 | return; |
957780eb JB |
5057 | } |
5058 | spin_unlock(&space_info->lock); | |
5059 | ||
5060 | /* | |
5061 | * Priority flushers can't wait on delalloc without | |
5062 | * deadlocking. | |
5063 | */ | |
5064 | if (flush_state == FLUSH_DELALLOC || | |
5065 | flush_state == FLUSH_DELALLOC_WAIT) | |
5066 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5067 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5068 | } |
5069 | ||
957780eb JB |
5070 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5071 | struct btrfs_space_info *space_info, | |
5072 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5073 | ||
21c7e756 | 5074 | { |
957780eb JB |
5075 | DEFINE_WAIT(wait); |
5076 | int ret = 0; | |
5077 | ||
5078 | spin_lock(&space_info->lock); | |
5079 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5080 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5081 | if (ret) { | |
5082 | ret = -EINTR; | |
5083 | break; | |
5084 | } | |
5085 | spin_unlock(&space_info->lock); | |
5086 | ||
5087 | schedule(); | |
5088 | ||
5089 | finish_wait(&ticket->wait, &wait); | |
5090 | spin_lock(&space_info->lock); | |
5091 | } | |
5092 | if (!ret) | |
5093 | ret = ticket->error; | |
5094 | if (!list_empty(&ticket->list)) | |
5095 | list_del_init(&ticket->list); | |
5096 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5097 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5098 | space_info->bytes_may_use -= num_bytes; | |
5099 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5100 | space_info->flags, num_bytes, 0); | |
5101 | } | |
5102 | spin_unlock(&space_info->lock); | |
5103 | ||
5104 | return ret; | |
21c7e756 MX |
5105 | } |
5106 | ||
4a92b1b8 JB |
5107 | /** |
5108 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5109 | * @root - the root we're allocating for | |
957780eb | 5110 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5111 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5112 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5113 | * |
01327610 | 5114 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5115 | * with the block_rsv. If there is not enough space it will make an attempt to |
5116 | * flush out space to make room. It will do this by flushing delalloc if | |
5117 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5118 | * regain reservations will be made and this will fail if there is not enough | |
5119 | * space already. | |
8bb8ab2e | 5120 | */ |
957780eb JB |
5121 | static int __reserve_metadata_bytes(struct btrfs_root *root, |
5122 | struct btrfs_space_info *space_info, | |
5123 | u64 orig_bytes, | |
5124 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5125 | { |
957780eb | 5126 | struct reserve_ticket ticket; |
2bf64758 | 5127 | u64 used; |
8bb8ab2e | 5128 | int ret = 0; |
9ed74f2d | 5129 | |
957780eb | 5130 | ASSERT(orig_bytes); |
8ca17f0f | 5131 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5132 | |
8bb8ab2e | 5133 | spin_lock(&space_info->lock); |
fdb5effd | 5134 | ret = -ENOSPC; |
2bf64758 JB |
5135 | used = space_info->bytes_used + space_info->bytes_reserved + |
5136 | space_info->bytes_pinned + space_info->bytes_readonly + | |
5137 | space_info->bytes_may_use; | |
9ed74f2d | 5138 | |
8bb8ab2e | 5139 | /* |
957780eb JB |
5140 | * If we have enough space then hooray, make our reservation and carry |
5141 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5142 | * If not things get more complicated. | |
8bb8ab2e | 5143 | */ |
957780eb JB |
5144 | if (used + orig_bytes <= space_info->total_bytes) { |
5145 | space_info->bytes_may_use += orig_bytes; | |
5146 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
5147 | space_info->flags, orig_bytes, | |
5148 | 1); | |
5149 | ret = 0; | |
5150 | } else if (can_overcommit(root, space_info, orig_bytes, flush)) { | |
44734ed1 JB |
5151 | space_info->bytes_may_use += orig_bytes; |
5152 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
5153 | space_info->flags, orig_bytes, | |
5154 | 1); | |
5155 | ret = 0; | |
2bf64758 JB |
5156 | } |
5157 | ||
8bb8ab2e | 5158 | /* |
957780eb JB |
5159 | * If we couldn't make a reservation then setup our reservation ticket |
5160 | * and kick the async worker if it's not already running. | |
08e007d2 | 5161 | * |
957780eb JB |
5162 | * If we are a priority flusher then we just need to add our ticket to |
5163 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5164 | */ |
72bcd99d | 5165 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5166 | ticket.bytes = orig_bytes; |
5167 | ticket.error = 0; | |
5168 | init_waitqueue_head(&ticket.wait); | |
5169 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5170 | list_add_tail(&ticket.list, &space_info->tickets); | |
5171 | if (!space_info->flush) { | |
5172 | space_info->flush = 1; | |
f376df2b JB |
5173 | trace_btrfs_trigger_flush(root->fs_info, |
5174 | space_info->flags, | |
5175 | orig_bytes, flush, | |
5176 | "enospc"); | |
957780eb JB |
5177 | queue_work(system_unbound_wq, |
5178 | &root->fs_info->async_reclaim_work); | |
5179 | } | |
5180 | } else { | |
5181 | list_add_tail(&ticket.list, | |
5182 | &space_info->priority_tickets); | |
5183 | } | |
21c7e756 MX |
5184 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5185 | used += orig_bytes; | |
f6acfd50 JB |
5186 | /* |
5187 | * We will do the space reservation dance during log replay, | |
5188 | * which means we won't have fs_info->fs_root set, so don't do | |
5189 | * the async reclaim as we will panic. | |
5190 | */ | |
afcdd129 | 5191 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags) && |
87241c2e | 5192 | need_do_async_reclaim(space_info, root, used) && |
f376df2b JB |
5193 | !work_busy(&root->fs_info->async_reclaim_work)) { |
5194 | trace_btrfs_trigger_flush(root->fs_info, | |
5195 | space_info->flags, | |
5196 | orig_bytes, flush, | |
5197 | "preempt"); | |
21c7e756 MX |
5198 | queue_work(system_unbound_wq, |
5199 | &root->fs_info->async_reclaim_work); | |
f376df2b | 5200 | } |
8bb8ab2e | 5201 | } |
f0486c68 | 5202 | spin_unlock(&space_info->lock); |
08e007d2 | 5203 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5204 | return ret; |
f0486c68 | 5205 | |
957780eb JB |
5206 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
5207 | return wait_reserve_ticket(root->fs_info, space_info, &ticket, | |
5208 | orig_bytes); | |
08e007d2 | 5209 | |
957780eb JB |
5210 | ret = 0; |
5211 | priority_reclaim_metadata_space(root->fs_info, space_info, &ticket); | |
5212 | spin_lock(&space_info->lock); | |
5213 | if (ticket.bytes) { | |
5214 | if (ticket.bytes < orig_bytes) { | |
5215 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5216 | space_info->bytes_may_use -= num_bytes; | |
5217 | trace_btrfs_space_reservation(root->fs_info, | |
5218 | "space_info", space_info->flags, | |
5219 | num_bytes, 0); | |
08e007d2 | 5220 | |
957780eb JB |
5221 | } |
5222 | list_del_init(&ticket.list); | |
5223 | ret = -ENOSPC; | |
5224 | } | |
5225 | spin_unlock(&space_info->lock); | |
5226 | ASSERT(list_empty(&ticket.list)); | |
5227 | return ret; | |
5228 | } | |
8bb8ab2e | 5229 | |
957780eb JB |
5230 | /** |
5231 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5232 | * @root - the root we're allocating for | |
5233 | * @block_rsv - the block_rsv we're allocating for | |
5234 | * @orig_bytes - the number of bytes we want | |
5235 | * @flush - whether or not we can flush to make our reservation | |
5236 | * | |
5237 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5238 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5239 | * flush out space to make room. It will do this by flushing delalloc if | |
5240 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5241 | * regain reservations will be made and this will fail if there is not enough | |
5242 | * space already. | |
5243 | */ | |
5244 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5245 | struct btrfs_block_rsv *block_rsv, | |
5246 | u64 orig_bytes, | |
5247 | enum btrfs_reserve_flush_enum flush) | |
5248 | { | |
5249 | int ret; | |
5250 | ||
5251 | ret = __reserve_metadata_bytes(root, block_rsv->space_info, orig_bytes, | |
5252 | flush); | |
5d80366e JB |
5253 | if (ret == -ENOSPC && |
5254 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5255 | struct btrfs_block_rsv *global_rsv = | |
5256 | &root->fs_info->global_block_rsv; | |
5257 | ||
5258 | if (block_rsv != global_rsv && | |
5259 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5260 | ret = 0; | |
5261 | } | |
cab45e22 JM |
5262 | if (ret == -ENOSPC) |
5263 | trace_btrfs_space_reservation(root->fs_info, | |
5264 | "space_info:enospc", | |
957780eb JB |
5265 | block_rsv->space_info->flags, |
5266 | orig_bytes, 1); | |
f0486c68 YZ |
5267 | return ret; |
5268 | } | |
5269 | ||
79787eaa JM |
5270 | static struct btrfs_block_rsv *get_block_rsv( |
5271 | const struct btrfs_trans_handle *trans, | |
5272 | const struct btrfs_root *root) | |
f0486c68 | 5273 | { |
4c13d758 JB |
5274 | struct btrfs_block_rsv *block_rsv = NULL; |
5275 | ||
e9cf439f AM |
5276 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
5277 | (root == root->fs_info->csum_root && trans->adding_csums) || | |
5278 | (root == root->fs_info->uuid_root)) | |
f7a81ea4 SB |
5279 | block_rsv = trans->block_rsv; |
5280 | ||
4c13d758 | 5281 | if (!block_rsv) |
f0486c68 YZ |
5282 | block_rsv = root->block_rsv; |
5283 | ||
5284 | if (!block_rsv) | |
5285 | block_rsv = &root->fs_info->empty_block_rsv; | |
5286 | ||
5287 | return block_rsv; | |
5288 | } | |
5289 | ||
5290 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5291 | u64 num_bytes) | |
5292 | { | |
5293 | int ret = -ENOSPC; | |
5294 | spin_lock(&block_rsv->lock); | |
5295 | if (block_rsv->reserved >= num_bytes) { | |
5296 | block_rsv->reserved -= num_bytes; | |
5297 | if (block_rsv->reserved < block_rsv->size) | |
5298 | block_rsv->full = 0; | |
5299 | ret = 0; | |
5300 | } | |
5301 | spin_unlock(&block_rsv->lock); | |
5302 | return ret; | |
5303 | } | |
5304 | ||
5305 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5306 | u64 num_bytes, int update_size) | |
5307 | { | |
5308 | spin_lock(&block_rsv->lock); | |
5309 | block_rsv->reserved += num_bytes; | |
5310 | if (update_size) | |
5311 | block_rsv->size += num_bytes; | |
5312 | else if (block_rsv->reserved >= block_rsv->size) | |
5313 | block_rsv->full = 1; | |
5314 | spin_unlock(&block_rsv->lock); | |
5315 | } | |
5316 | ||
d52be818 JB |
5317 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5318 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5319 | int min_factor) | |
5320 | { | |
5321 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5322 | u64 min_bytes; | |
5323 | ||
5324 | if (global_rsv->space_info != dest->space_info) | |
5325 | return -ENOSPC; | |
5326 | ||
5327 | spin_lock(&global_rsv->lock); | |
5328 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5329 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5330 | spin_unlock(&global_rsv->lock); | |
5331 | return -ENOSPC; | |
5332 | } | |
5333 | global_rsv->reserved -= num_bytes; | |
5334 | if (global_rsv->reserved < global_rsv->size) | |
5335 | global_rsv->full = 0; | |
5336 | spin_unlock(&global_rsv->lock); | |
5337 | ||
5338 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5339 | return 0; | |
5340 | } | |
5341 | ||
957780eb JB |
5342 | /* |
5343 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5344 | * basically when we're returning space from block_rsv's. | |
5345 | */ | |
5346 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5347 | struct btrfs_space_info *space_info, | |
5348 | u64 num_bytes) | |
5349 | { | |
5350 | struct reserve_ticket *ticket; | |
5351 | struct list_head *head; | |
5352 | u64 used; | |
5353 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5354 | bool check_overcommit = false; | |
5355 | ||
5356 | spin_lock(&space_info->lock); | |
5357 | head = &space_info->priority_tickets; | |
5358 | ||
5359 | /* | |
5360 | * If we are over our limit then we need to check and see if we can | |
5361 | * overcommit, and if we can't then we just need to free up our space | |
5362 | * and not satisfy any requests. | |
5363 | */ | |
5364 | used = space_info->bytes_used + space_info->bytes_reserved + | |
5365 | space_info->bytes_pinned + space_info->bytes_readonly + | |
5366 | space_info->bytes_may_use; | |
5367 | if (used - num_bytes >= space_info->total_bytes) | |
5368 | check_overcommit = true; | |
5369 | again: | |
5370 | while (!list_empty(head) && num_bytes) { | |
5371 | ticket = list_first_entry(head, struct reserve_ticket, | |
5372 | list); | |
5373 | /* | |
5374 | * We use 0 bytes because this space is already reserved, so | |
5375 | * adding the ticket space would be a double count. | |
5376 | */ | |
5377 | if (check_overcommit && | |
5378 | !can_overcommit(fs_info->extent_root, space_info, 0, | |
5379 | flush)) | |
5380 | break; | |
5381 | if (num_bytes >= ticket->bytes) { | |
5382 | list_del_init(&ticket->list); | |
5383 | num_bytes -= ticket->bytes; | |
5384 | ticket->bytes = 0; | |
ce129655 | 5385 | space_info->tickets_id++; |
957780eb JB |
5386 | wake_up(&ticket->wait); |
5387 | } else { | |
5388 | ticket->bytes -= num_bytes; | |
5389 | num_bytes = 0; | |
5390 | } | |
5391 | } | |
5392 | ||
5393 | if (num_bytes && head == &space_info->priority_tickets) { | |
5394 | head = &space_info->tickets; | |
5395 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5396 | goto again; | |
5397 | } | |
5398 | space_info->bytes_may_use -= num_bytes; | |
5399 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5400 | space_info->flags, num_bytes, 0); | |
5401 | spin_unlock(&space_info->lock); | |
5402 | } | |
5403 | ||
5404 | /* | |
5405 | * This is for newly allocated space that isn't accounted in | |
5406 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5407 | * we use this helper. | |
5408 | */ | |
5409 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5410 | struct btrfs_space_info *space_info, | |
5411 | u64 num_bytes) | |
5412 | { | |
5413 | struct reserve_ticket *ticket; | |
5414 | struct list_head *head = &space_info->priority_tickets; | |
5415 | ||
5416 | again: | |
5417 | while (!list_empty(head) && num_bytes) { | |
5418 | ticket = list_first_entry(head, struct reserve_ticket, | |
5419 | list); | |
5420 | if (num_bytes >= ticket->bytes) { | |
5421 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5422 | space_info->flags, | |
5423 | ticket->bytes, 1); | |
5424 | list_del_init(&ticket->list); | |
5425 | num_bytes -= ticket->bytes; | |
5426 | space_info->bytes_may_use += ticket->bytes; | |
5427 | ticket->bytes = 0; | |
ce129655 | 5428 | space_info->tickets_id++; |
957780eb JB |
5429 | wake_up(&ticket->wait); |
5430 | } else { | |
5431 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5432 | space_info->flags, | |
5433 | num_bytes, 1); | |
5434 | space_info->bytes_may_use += num_bytes; | |
5435 | ticket->bytes -= num_bytes; | |
5436 | num_bytes = 0; | |
5437 | } | |
5438 | } | |
5439 | ||
5440 | if (num_bytes && head == &space_info->priority_tickets) { | |
5441 | head = &space_info->tickets; | |
5442 | goto again; | |
5443 | } | |
5444 | } | |
5445 | ||
8c2a3ca2 JB |
5446 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
5447 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 5448 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
5449 | { |
5450 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
5451 | ||
5452 | spin_lock(&block_rsv->lock); | |
5453 | if (num_bytes == (u64)-1) | |
5454 | num_bytes = block_rsv->size; | |
5455 | block_rsv->size -= num_bytes; | |
5456 | if (block_rsv->reserved >= block_rsv->size) { | |
5457 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5458 | block_rsv->reserved = block_rsv->size; | |
5459 | block_rsv->full = 1; | |
5460 | } else { | |
5461 | num_bytes = 0; | |
5462 | } | |
5463 | spin_unlock(&block_rsv->lock); | |
5464 | ||
5465 | if (num_bytes > 0) { | |
5466 | if (dest) { | |
e9e22899 JB |
5467 | spin_lock(&dest->lock); |
5468 | if (!dest->full) { | |
5469 | u64 bytes_to_add; | |
5470 | ||
5471 | bytes_to_add = dest->size - dest->reserved; | |
5472 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5473 | dest->reserved += bytes_to_add; | |
5474 | if (dest->reserved >= dest->size) | |
5475 | dest->full = 1; | |
5476 | num_bytes -= bytes_to_add; | |
5477 | } | |
5478 | spin_unlock(&dest->lock); | |
5479 | } | |
957780eb JB |
5480 | if (num_bytes) |
5481 | space_info_add_old_bytes(fs_info, space_info, | |
5482 | num_bytes); | |
9ed74f2d | 5483 | } |
f0486c68 | 5484 | } |
4e06bdd6 | 5485 | |
25d609f8 JB |
5486 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5487 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
5488 | int update_size) | |
f0486c68 YZ |
5489 | { |
5490 | int ret; | |
9ed74f2d | 5491 | |
f0486c68 YZ |
5492 | ret = block_rsv_use_bytes(src, num_bytes); |
5493 | if (ret) | |
5494 | return ret; | |
9ed74f2d | 5495 | |
25d609f8 | 5496 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5497 | return 0; |
5498 | } | |
5499 | ||
66d8f3dd | 5500 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5501 | { |
f0486c68 YZ |
5502 | memset(rsv, 0, sizeof(*rsv)); |
5503 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5504 | rsv->type = type; |
f0486c68 YZ |
5505 | } |
5506 | ||
66d8f3dd MX |
5507 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
5508 | unsigned short type) | |
f0486c68 YZ |
5509 | { |
5510 | struct btrfs_block_rsv *block_rsv; | |
5511 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 5512 | |
f0486c68 YZ |
5513 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5514 | if (!block_rsv) | |
5515 | return NULL; | |
9ed74f2d | 5516 | |
66d8f3dd | 5517 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
5518 | block_rsv->space_info = __find_space_info(fs_info, |
5519 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
5520 | return block_rsv; |
5521 | } | |
9ed74f2d | 5522 | |
f0486c68 YZ |
5523 | void btrfs_free_block_rsv(struct btrfs_root *root, |
5524 | struct btrfs_block_rsv *rsv) | |
5525 | { | |
2aaa6655 JB |
5526 | if (!rsv) |
5527 | return; | |
dabdb640 JB |
5528 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
5529 | kfree(rsv); | |
9ed74f2d JB |
5530 | } |
5531 | ||
cdfb080e CM |
5532 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5533 | { | |
5534 | kfree(rsv); | |
5535 | } | |
5536 | ||
08e007d2 MX |
5537 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5538 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5539 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5540 | { |
f0486c68 | 5541 | int ret; |
9ed74f2d | 5542 | |
f0486c68 YZ |
5543 | if (num_bytes == 0) |
5544 | return 0; | |
8bb8ab2e | 5545 | |
61b520a9 | 5546 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5547 | if (!ret) { |
5548 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5549 | return 0; | |
5550 | } | |
9ed74f2d | 5551 | |
f0486c68 | 5552 | return ret; |
f0486c68 | 5553 | } |
9ed74f2d | 5554 | |
4a92b1b8 | 5555 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 5556 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5557 | { |
5558 | u64 num_bytes = 0; | |
f0486c68 | 5559 | int ret = -ENOSPC; |
9ed74f2d | 5560 | |
f0486c68 YZ |
5561 | if (!block_rsv) |
5562 | return 0; | |
9ed74f2d | 5563 | |
f0486c68 | 5564 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5565 | num_bytes = div_factor(block_rsv->size, min_factor); |
5566 | if (block_rsv->reserved >= num_bytes) | |
5567 | ret = 0; | |
5568 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5569 | |
36ba022a JB |
5570 | return ret; |
5571 | } | |
5572 | ||
08e007d2 MX |
5573 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5574 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5575 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5576 | { |
5577 | u64 num_bytes = 0; | |
5578 | int ret = -ENOSPC; | |
5579 | ||
5580 | if (!block_rsv) | |
5581 | return 0; | |
5582 | ||
5583 | spin_lock(&block_rsv->lock); | |
5584 | num_bytes = min_reserved; | |
13553e52 | 5585 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5586 | ret = 0; |
13553e52 | 5587 | else |
f0486c68 | 5588 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5589 | spin_unlock(&block_rsv->lock); |
13553e52 | 5590 | |
f0486c68 YZ |
5591 | if (!ret) |
5592 | return 0; | |
5593 | ||
aa38a711 | 5594 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5595 | if (!ret) { |
5596 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5597 | return 0; |
6a63209f | 5598 | } |
9ed74f2d | 5599 | |
13553e52 | 5600 | return ret; |
f0486c68 YZ |
5601 | } |
5602 | ||
f0486c68 YZ |
5603 | void btrfs_block_rsv_release(struct btrfs_root *root, |
5604 | struct btrfs_block_rsv *block_rsv, | |
5605 | u64 num_bytes) | |
5606 | { | |
5607 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
17504584 | 5608 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5609 | block_rsv->space_info != global_rsv->space_info) |
5610 | global_rsv = NULL; | |
8c2a3ca2 JB |
5611 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
5612 | num_bytes); | |
6a63209f JB |
5613 | } |
5614 | ||
8929ecfa YZ |
5615 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5616 | { | |
5617 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5618 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5619 | u64 num_bytes; | |
6a63209f | 5620 | |
ae2e4728 JB |
5621 | /* |
5622 | * The global block rsv is based on the size of the extent tree, the | |
5623 | * checksum tree and the root tree. If the fs is empty we want to set | |
5624 | * it to a minimal amount for safety. | |
5625 | */ | |
5626 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5627 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5628 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5629 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5630 | |
8929ecfa | 5631 | spin_lock(&sinfo->lock); |
1f699d38 | 5632 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5633 | |
ee22184b | 5634 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5635 | |
fb4b10e5 JB |
5636 | if (block_rsv->reserved < block_rsv->size) { |
5637 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + | |
5638 | sinfo->bytes_reserved + sinfo->bytes_readonly + | |
5639 | sinfo->bytes_may_use; | |
5640 | if (sinfo->total_bytes > num_bytes) { | |
5641 | num_bytes = sinfo->total_bytes - num_bytes; | |
5642 | num_bytes = min(num_bytes, | |
5643 | block_rsv->size - block_rsv->reserved); | |
5644 | block_rsv->reserved += num_bytes; | |
5645 | sinfo->bytes_may_use += num_bytes; | |
5646 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5647 | sinfo->flags, num_bytes, | |
5648 | 1); | |
5649 | } | |
5650 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5651 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5652 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5653 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5654 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5655 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5656 | } |
182608c8 | 5657 | |
fb4b10e5 JB |
5658 | if (block_rsv->reserved == block_rsv->size) |
5659 | block_rsv->full = 1; | |
5660 | else | |
5661 | block_rsv->full = 0; | |
5662 | ||
8929ecfa | 5663 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5664 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5665 | } |
5666 | ||
f0486c68 | 5667 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5668 | { |
f0486c68 | 5669 | struct btrfs_space_info *space_info; |
6a63209f | 5670 | |
f0486c68 YZ |
5671 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5672 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5673 | |
f0486c68 | 5674 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5675 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 5676 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
5677 | fs_info->trans_block_rsv.space_info = space_info; |
5678 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5679 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5680 | |
8929ecfa YZ |
5681 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5682 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5683 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5684 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5685 | if (fs_info->quota_root) |
5686 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5687 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5688 | |
8929ecfa | 5689 | update_global_block_rsv(fs_info); |
6a63209f JB |
5690 | } |
5691 | ||
8929ecfa | 5692 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5693 | { |
8c2a3ca2 JB |
5694 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5695 | (u64)-1); | |
8929ecfa YZ |
5696 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
5697 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
5698 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
5699 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5700 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5701 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5702 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5703 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5704 | } |
5705 | ||
a22285a6 YZ |
5706 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
5707 | struct btrfs_root *root) | |
6a63209f | 5708 | { |
0e721106 JB |
5709 | if (!trans->block_rsv) |
5710 | return; | |
5711 | ||
a22285a6 YZ |
5712 | if (!trans->bytes_reserved) |
5713 | return; | |
6a63209f | 5714 | |
e77266e4 | 5715 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 5716 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 5717 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
5718 | trans->bytes_reserved = 0; |
5719 | } | |
6a63209f | 5720 | |
4fbcdf66 FM |
5721 | /* |
5722 | * To be called after all the new block groups attached to the transaction | |
5723 | * handle have been created (btrfs_create_pending_block_groups()). | |
5724 | */ | |
5725 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5726 | { | |
64b63580 | 5727 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5728 | |
5729 | if (!trans->chunk_bytes_reserved) | |
5730 | return; | |
5731 | ||
5732 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5733 | ||
5734 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5735 | trans->chunk_bytes_reserved); | |
5736 | trans->chunk_bytes_reserved = 0; | |
5737 | } | |
5738 | ||
79787eaa | 5739 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
5740 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
5741 | struct inode *inode) | |
5742 | { | |
5743 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
40acc3ee JB |
5744 | /* |
5745 | * We always use trans->block_rsv here as we will have reserved space | |
5746 | * for our orphan when starting the transaction, using get_block_rsv() | |
5747 | * here will sometimes make us choose the wrong block rsv as we could be | |
5748 | * doing a reloc inode for a non refcounted root. | |
5749 | */ | |
5750 | struct btrfs_block_rsv *src_rsv = trans->block_rsv; | |
d68fc57b YZ |
5751 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; |
5752 | ||
5753 | /* | |
fcb80c2a JB |
5754 | * We need to hold space in order to delete our orphan item once we've |
5755 | * added it, so this takes the reservation so we can release it later | |
5756 | * when we are truly done with the orphan item. | |
d68fc57b | 5757 | */ |
ff5714cc | 5758 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
5759 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
5760 | btrfs_ino(inode), num_bytes, 1); | |
25d609f8 | 5761 | return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); |
6a63209f JB |
5762 | } |
5763 | ||
d68fc57b | 5764 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 5765 | { |
d68fc57b | 5766 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 5767 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
5768 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
5769 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
5770 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
5771 | } | |
97e728d4 | 5772 | |
d5c12070 MX |
5773 | /* |
5774 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5775 | * root: the root of the parent directory | |
5776 | * rsv: block reservation | |
5777 | * items: the number of items that we need do reservation | |
5778 | * qgroup_reserved: used to return the reserved size in qgroup | |
5779 | * | |
5780 | * This function is used to reserve the space for snapshot/subvolume | |
5781 | * creation and deletion. Those operations are different with the | |
5782 | * common file/directory operations, they change two fs/file trees | |
5783 | * and root tree, the number of items that the qgroup reserves is | |
5784 | * different with the free space reservation. So we can not use | |
01327610 | 5785 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5786 | */ |
5787 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5788 | struct btrfs_block_rsv *rsv, | |
5789 | int items, | |
ee3441b4 JM |
5790 | u64 *qgroup_reserved, |
5791 | bool use_global_rsv) | |
a22285a6 | 5792 | { |
d5c12070 MX |
5793 | u64 num_bytes; |
5794 | int ret; | |
ee3441b4 | 5795 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
d5c12070 | 5796 | |
afcdd129 | 5797 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags)) { |
d5c12070 | 5798 | /* One for parent inode, two for dir entries */ |
707e8a07 | 5799 | num_bytes = 3 * root->nodesize; |
7174109c | 5800 | ret = btrfs_qgroup_reserve_meta(root, num_bytes); |
d5c12070 MX |
5801 | if (ret) |
5802 | return ret; | |
5803 | } else { | |
5804 | num_bytes = 0; | |
5805 | } | |
5806 | ||
5807 | *qgroup_reserved = num_bytes; | |
5808 | ||
5809 | num_bytes = btrfs_calc_trans_metadata_size(root, items); | |
5810 | rsv->space_info = __find_space_info(root->fs_info, | |
5811 | BTRFS_BLOCK_GROUP_METADATA); | |
5812 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5813 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5814 | |
5815 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 5816 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 5817 | |
7174109c QW |
5818 | if (ret && *qgroup_reserved) |
5819 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
d5c12070 MX |
5820 | |
5821 | return ret; | |
5822 | } | |
5823 | ||
5824 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, | |
5825 | struct btrfs_block_rsv *rsv, | |
5826 | u64 qgroup_reserved) | |
5827 | { | |
5828 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
97e728d4 JB |
5829 | } |
5830 | ||
7709cde3 JB |
5831 | /** |
5832 | * drop_outstanding_extent - drop an outstanding extent | |
5833 | * @inode: the inode we're dropping the extent for | |
01327610 | 5834 | * @num_bytes: the number of bytes we're releasing. |
7709cde3 JB |
5835 | * |
5836 | * This is called when we are freeing up an outstanding extent, either called | |
5837 | * after an error or after an extent is written. This will return the number of | |
5838 | * reserved extents that need to be freed. This must be called with | |
5839 | * BTRFS_I(inode)->lock held. | |
5840 | */ | |
dcab6a3b | 5841 | static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes) |
9e0baf60 | 5842 | { |
7fd2ae21 | 5843 | unsigned drop_inode_space = 0; |
9e0baf60 | 5844 | unsigned dropped_extents = 0; |
dcab6a3b | 5845 | unsigned num_extents = 0; |
9e0baf60 | 5846 | |
dcab6a3b JB |
5847 | num_extents = (unsigned)div64_u64(num_bytes + |
5848 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5849 | BTRFS_MAX_EXTENT_SIZE); | |
5850 | ASSERT(num_extents); | |
5851 | ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents); | |
5852 | BTRFS_I(inode)->outstanding_extents -= num_extents; | |
9e0baf60 | 5853 | |
7fd2ae21 | 5854 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
5855 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5856 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 5857 | drop_inode_space = 1; |
7fd2ae21 | 5858 | |
9e0baf60 | 5859 | /* |
01327610 | 5860 | * If we have more or the same amount of outstanding extents than we have |
9e0baf60 JB |
5861 | * reserved then we need to leave the reserved extents count alone. |
5862 | */ | |
5863 | if (BTRFS_I(inode)->outstanding_extents >= | |
5864 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 5865 | return drop_inode_space; |
9e0baf60 JB |
5866 | |
5867 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
5868 | BTRFS_I(inode)->outstanding_extents; | |
5869 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 5870 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
5871 | } |
5872 | ||
7709cde3 | 5873 | /** |
01327610 NS |
5874 | * calc_csum_metadata_size - return the amount of metadata space that must be |
5875 | * reserved/freed for the given bytes. | |
7709cde3 JB |
5876 | * @inode: the inode we're manipulating |
5877 | * @num_bytes: the number of bytes in question | |
5878 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
5879 | * | |
5880 | * This adjusts the number of csum_bytes in the inode and then returns the | |
5881 | * correct amount of metadata that must either be reserved or freed. We | |
5882 | * calculate how many checksums we can fit into one leaf and then divide the | |
5883 | * number of bytes that will need to be checksumed by this value to figure out | |
5884 | * how many checksums will be required. If we are adding bytes then the number | |
5885 | * may go up and we will return the number of additional bytes that must be | |
5886 | * reserved. If it is going down we will return the number of bytes that must | |
5887 | * be freed. | |
5888 | * | |
5889 | * This must be called with BTRFS_I(inode)->lock held. | |
5890 | */ | |
5891 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
5892 | int reserve) | |
6324fbf3 | 5893 | { |
7709cde3 | 5894 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1262133b | 5895 | u64 old_csums, num_csums; |
7709cde3 JB |
5896 | |
5897 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
5898 | BTRFS_I(inode)->csum_bytes == 0) | |
5899 | return 0; | |
5900 | ||
28f75a0e | 5901 | old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); |
7709cde3 JB |
5902 | if (reserve) |
5903 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
5904 | else | |
5905 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
28f75a0e | 5906 | num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); |
7709cde3 JB |
5907 | |
5908 | /* No change, no need to reserve more */ | |
5909 | if (old_csums == num_csums) | |
5910 | return 0; | |
5911 | ||
5912 | if (reserve) | |
5913 | return btrfs_calc_trans_metadata_size(root, | |
5914 | num_csums - old_csums); | |
5915 | ||
5916 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 5917 | } |
c146afad | 5918 | |
0ca1f7ce YZ |
5919 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
5920 | { | |
5921 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5922 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 5923 | u64 to_reserve = 0; |
660d3f6c | 5924 | u64 csum_bytes; |
9e0baf60 | 5925 | unsigned nr_extents = 0; |
08e007d2 | 5926 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5927 | int ret = 0; |
c64c2bd8 | 5928 | bool delalloc_lock = true; |
88e081bf WS |
5929 | u64 to_free = 0; |
5930 | unsigned dropped; | |
48c3d480 | 5931 | bool release_extra = false; |
6324fbf3 | 5932 | |
c64c2bd8 JB |
5933 | /* If we are a free space inode we need to not flush since we will be in |
5934 | * the middle of a transaction commit. We also don't need the delalloc | |
5935 | * mutex since we won't race with anybody. We need this mostly to make | |
5936 | * lockdep shut its filthy mouth. | |
bac357dc JB |
5937 | * |
5938 | * If we have a transaction open (can happen if we call truncate_block | |
5939 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
5940 | */ |
5941 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 5942 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 5943 | delalloc_lock = false; |
bac357dc JB |
5944 | } else if (current->journal_info) { |
5945 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c64c2bd8 | 5946 | } |
c09544e0 | 5947 | |
08e007d2 MX |
5948 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
5949 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 5950 | schedule_timeout(1); |
ec44a35c | 5951 | |
c64c2bd8 JB |
5952 | if (delalloc_lock) |
5953 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
5954 | ||
0ca1f7ce | 5955 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 5956 | |
9e0baf60 | 5957 | spin_lock(&BTRFS_I(inode)->lock); |
6a41dd09 JB |
5958 | nr_extents = (unsigned)div64_u64(num_bytes + |
5959 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5960 | BTRFS_MAX_EXTENT_SIZE); | |
5961 | BTRFS_I(inode)->outstanding_extents += nr_extents; | |
9e0baf60 | 5962 | |
48c3d480 | 5963 | nr_extents = 0; |
9e0baf60 | 5964 | if (BTRFS_I(inode)->outstanding_extents > |
660d3f6c | 5965 | BTRFS_I(inode)->reserved_extents) |
48c3d480 | 5966 | nr_extents += BTRFS_I(inode)->outstanding_extents - |
9e0baf60 | 5967 | BTRFS_I(inode)->reserved_extents; |
57a45ced | 5968 | |
48c3d480 JB |
5969 | /* We always want to reserve a slot for updating the inode. */ |
5970 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents + 1); | |
7709cde3 | 5971 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 5972 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 5973 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 5974 | |
afcdd129 | 5975 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags)) { |
7174109c QW |
5976 | ret = btrfs_qgroup_reserve_meta(root, |
5977 | nr_extents * root->nodesize); | |
88e081bf WS |
5978 | if (ret) |
5979 | goto out_fail; | |
5980 | } | |
c5567237 | 5981 | |
48c3d480 | 5982 | ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush); |
88e081bf | 5983 | if (unlikely(ret)) { |
7174109c | 5984 | btrfs_qgroup_free_meta(root, nr_extents * root->nodesize); |
88e081bf | 5985 | goto out_fail; |
9e0baf60 | 5986 | } |
25179201 | 5987 | |
660d3f6c | 5988 | spin_lock(&BTRFS_I(inode)->lock); |
48c3d480 | 5989 | if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
f485c9ee JB |
5990 | &BTRFS_I(inode)->runtime_flags)) { |
5991 | to_reserve -= btrfs_calc_trans_metadata_size(root, 1); | |
48c3d480 | 5992 | release_extra = true; |
660d3f6c JB |
5993 | } |
5994 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
5995 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
5996 | |
5997 | if (delalloc_lock) | |
5998 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 5999 | |
8c2a3ca2 | 6000 | if (to_reserve) |
67871254 | 6001 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
8c2a3ca2 | 6002 | btrfs_ino(inode), to_reserve, 1); |
48c3d480 JB |
6003 | if (release_extra) |
6004 | btrfs_block_rsv_release(root, block_rsv, | |
6005 | btrfs_calc_trans_metadata_size(root, | |
6006 | 1)); | |
0ca1f7ce | 6007 | return 0; |
88e081bf WS |
6008 | |
6009 | out_fail: | |
6010 | spin_lock(&BTRFS_I(inode)->lock); | |
dcab6a3b | 6011 | dropped = drop_outstanding_extent(inode, num_bytes); |
88e081bf WS |
6012 | /* |
6013 | * If the inodes csum_bytes is the same as the original | |
6014 | * csum_bytes then we know we haven't raced with any free()ers | |
6015 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 6016 | */ |
f4881bc7 | 6017 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { |
88e081bf | 6018 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 JB |
6019 | } else { |
6020 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; | |
6021 | u64 bytes; | |
6022 | ||
6023 | /* | |
6024 | * This is tricky, but first we need to figure out how much we | |
01327610 | 6025 | * freed from any free-ers that occurred during this |
f4881bc7 JB |
6026 | * reservation, so we reset ->csum_bytes to the csum_bytes |
6027 | * before we dropped our lock, and then call the free for the | |
6028 | * number of bytes that were freed while we were trying our | |
6029 | * reservation. | |
6030 | */ | |
6031 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; | |
6032 | BTRFS_I(inode)->csum_bytes = csum_bytes; | |
6033 | to_free = calc_csum_metadata_size(inode, bytes, 0); | |
6034 | ||
6035 | ||
6036 | /* | |
6037 | * Now we need to see how much we would have freed had we not | |
6038 | * been making this reservation and our ->csum_bytes were not | |
6039 | * artificially inflated. | |
6040 | */ | |
6041 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; | |
6042 | bytes = csum_bytes - orig_csum_bytes; | |
6043 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
6044 | ||
6045 | /* | |
6046 | * Now reset ->csum_bytes to what it should be. If bytes is | |
01327610 | 6047 | * more than to_free then we would have freed more space had we |
f4881bc7 JB |
6048 | * not had an artificially high ->csum_bytes, so we need to free |
6049 | * the remainder. If bytes is the same or less then we don't | |
6050 | * need to do anything, the other free-ers did the correct | |
6051 | * thing. | |
6052 | */ | |
6053 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; | |
6054 | if (bytes > to_free) | |
6055 | to_free = bytes - to_free; | |
6056 | else | |
6057 | to_free = 0; | |
6058 | } | |
88e081bf | 6059 | spin_unlock(&BTRFS_I(inode)->lock); |
e2d1f923 | 6060 | if (dropped) |
88e081bf WS |
6061 | to_free += btrfs_calc_trans_metadata_size(root, dropped); |
6062 | ||
6063 | if (to_free) { | |
6064 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
6065 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
6066 | btrfs_ino(inode), to_free, 0); | |
6067 | } | |
6068 | if (delalloc_lock) | |
6069 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
6070 | return ret; | |
0ca1f7ce YZ |
6071 | } |
6072 | ||
7709cde3 JB |
6073 | /** |
6074 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
6075 | * @inode: the inode to release the reservation for | |
6076 | * @num_bytes: the number of bytes we're releasing | |
6077 | * | |
6078 | * This will release the metadata reservation for an inode. This can be called | |
6079 | * once we complete IO for a given set of bytes to release their metadata | |
6080 | * reservations. | |
6081 | */ | |
0ca1f7ce YZ |
6082 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
6083 | { | |
6084 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
6085 | u64 to_free = 0; |
6086 | unsigned dropped; | |
0ca1f7ce YZ |
6087 | |
6088 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 6089 | spin_lock(&BTRFS_I(inode)->lock); |
dcab6a3b | 6090 | dropped = drop_outstanding_extent(inode, num_bytes); |
97e728d4 | 6091 | |
0934856d MX |
6092 | if (num_bytes) |
6093 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
7709cde3 | 6094 | spin_unlock(&BTRFS_I(inode)->lock); |
9e0baf60 JB |
6095 | if (dropped > 0) |
6096 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 6097 | |
f5ee5c9a | 6098 | if (btrfs_is_testing(root->fs_info)) |
6a3891c5 JB |
6099 | return; |
6100 | ||
8c2a3ca2 JB |
6101 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
6102 | btrfs_ino(inode), to_free, 0); | |
c5567237 | 6103 | |
0ca1f7ce YZ |
6104 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
6105 | to_free); | |
6106 | } | |
6107 | ||
1ada3a62 | 6108 | /** |
7cf5b976 | 6109 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6110 | * delalloc |
6111 | * @inode: inode we're writing to | |
6112 | * @start: start range we are writing to | |
6113 | * @len: how long the range we are writing to | |
6114 | * | |
1ada3a62 QW |
6115 | * This will do the following things |
6116 | * | |
6117 | * o reserve space in data space info for num bytes | |
6118 | * and reserve precious corresponding qgroup space | |
6119 | * (Done in check_data_free_space) | |
6120 | * | |
6121 | * o reserve space for metadata space, based on the number of outstanding | |
6122 | * extents and how much csums will be needed | |
6123 | * also reserve metadata space in a per root over-reserve method. | |
6124 | * o add to the inodes->delalloc_bytes | |
6125 | * o add it to the fs_info's delalloc inodes list. | |
6126 | * (Above 3 all done in delalloc_reserve_metadata) | |
6127 | * | |
6128 | * Return 0 for success | |
6129 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6130 | */ | |
7cf5b976 | 6131 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
6132 | { |
6133 | int ret; | |
6134 | ||
7cf5b976 | 6135 | ret = btrfs_check_data_free_space(inode, start, len); |
1ada3a62 QW |
6136 | if (ret < 0) |
6137 | return ret; | |
6138 | ret = btrfs_delalloc_reserve_metadata(inode, len); | |
6139 | if (ret < 0) | |
7cf5b976 | 6140 | btrfs_free_reserved_data_space(inode, start, len); |
1ada3a62 QW |
6141 | return ret; |
6142 | } | |
6143 | ||
7709cde3 | 6144 | /** |
7cf5b976 | 6145 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6146 | * @inode: inode we're releasing space for |
6147 | * @start: start position of the space already reserved | |
6148 | * @len: the len of the space already reserved | |
6149 | * | |
6150 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
6151 | * called in the case that we don't need the metadata AND data reservations | |
6152 | * anymore. So if there is an error or we insert an inline extent. | |
6153 | * | |
6154 | * This function will release the metadata space that was not used and will | |
6155 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6156 | * list if there are no delalloc bytes left. | |
6157 | * Also it will handle the qgroup reserved space. | |
6158 | */ | |
7cf5b976 | 6159 | void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
6160 | { |
6161 | btrfs_delalloc_release_metadata(inode, len); | |
7cf5b976 | 6162 | btrfs_free_reserved_data_space(inode, start, len); |
6324fbf3 CM |
6163 | } |
6164 | ||
ce93ec54 JB |
6165 | static int update_block_group(struct btrfs_trans_handle *trans, |
6166 | struct btrfs_root *root, u64 bytenr, | |
6167 | u64 num_bytes, int alloc) | |
9078a3e1 | 6168 | { |
0af3d00b | 6169 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 6170 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 6171 | u64 total = num_bytes; |
9078a3e1 | 6172 | u64 old_val; |
db94535d | 6173 | u64 byte_in_group; |
0af3d00b | 6174 | int factor; |
3e1ad54f | 6175 | |
5d4f98a2 | 6176 | /* block accounting for super block */ |
eb73c1b7 | 6177 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6178 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6179 | if (alloc) |
6180 | old_val += num_bytes; | |
6181 | else | |
6182 | old_val -= num_bytes; | |
6c41761f | 6183 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6184 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6185 | |
d397712b | 6186 | while (total) { |
db94535d | 6187 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6188 | if (!cache) |
79787eaa | 6189 | return -ENOENT; |
b742bb82 YZ |
6190 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6191 | BTRFS_BLOCK_GROUP_RAID1 | | |
6192 | BTRFS_BLOCK_GROUP_RAID10)) | |
6193 | factor = 2; | |
6194 | else | |
6195 | factor = 1; | |
9d66e233 JB |
6196 | /* |
6197 | * If this block group has free space cache written out, we | |
6198 | * need to make sure to load it if we are removing space. This | |
6199 | * is because we need the unpinning stage to actually add the | |
6200 | * space back to the block group, otherwise we will leak space. | |
6201 | */ | |
6202 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6203 | cache_block_group(cache, 1); |
0af3d00b | 6204 | |
db94535d CM |
6205 | byte_in_group = bytenr - cache->key.objectid; |
6206 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6207 | |
25179201 | 6208 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6209 | spin_lock(&cache->lock); |
0af3d00b | 6210 | |
3cdde224 | 6211 | if (btrfs_test_opt(root->fs_info, SPACE_CACHE) && |
0af3d00b JB |
6212 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6213 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6214 | ||
9078a3e1 | 6215 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6216 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6217 | if (alloc) { |
db94535d | 6218 | old_val += num_bytes; |
11833d66 YZ |
6219 | btrfs_set_block_group_used(&cache->item, old_val); |
6220 | cache->reserved -= num_bytes; | |
11833d66 | 6221 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6222 | cache->space_info->bytes_used += num_bytes; |
6223 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6224 | spin_unlock(&cache->lock); |
25179201 | 6225 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6226 | } else { |
db94535d | 6227 | old_val -= num_bytes; |
ae0ab003 FM |
6228 | btrfs_set_block_group_used(&cache->item, old_val); |
6229 | cache->pinned += num_bytes; | |
6230 | cache->space_info->bytes_pinned += num_bytes; | |
6231 | cache->space_info->bytes_used -= num_bytes; | |
6232 | cache->space_info->disk_used -= num_bytes * factor; | |
6233 | spin_unlock(&cache->lock); | |
6234 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6235 | |
c51e7bb1 JB |
6236 | trace_btrfs_space_reservation(root->fs_info, "pinned", |
6237 | cache->space_info->flags, | |
6238 | num_bytes, 1); | |
ae0ab003 FM |
6239 | set_extent_dirty(info->pinned_extents, |
6240 | bytenr, bytenr + num_bytes - 1, | |
6241 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6242 | } |
1bbc621e CM |
6243 | |
6244 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6245 | if (list_empty(&cache->dirty_list)) { | |
6246 | list_add_tail(&cache->dirty_list, | |
6247 | &trans->transaction->dirty_bgs); | |
6248 | trans->transaction->num_dirty_bgs++; | |
6249 | btrfs_get_block_group(cache); | |
6250 | } | |
6251 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6252 | ||
036a9348 FM |
6253 | /* |
6254 | * No longer have used bytes in this block group, queue it for | |
6255 | * deletion. We do this after adding the block group to the | |
6256 | * dirty list to avoid races between cleaner kthread and space | |
6257 | * cache writeout. | |
6258 | */ | |
6259 | if (!alloc && old_val == 0) { | |
6260 | spin_lock(&info->unused_bgs_lock); | |
6261 | if (list_empty(&cache->bg_list)) { | |
6262 | btrfs_get_block_group(cache); | |
6263 | list_add_tail(&cache->bg_list, | |
6264 | &info->unused_bgs); | |
6265 | } | |
6266 | spin_unlock(&info->unused_bgs_lock); | |
6267 | } | |
6268 | ||
fa9c0d79 | 6269 | btrfs_put_block_group(cache); |
db94535d CM |
6270 | total -= num_bytes; |
6271 | bytenr += num_bytes; | |
9078a3e1 CM |
6272 | } |
6273 | return 0; | |
6274 | } | |
6324fbf3 | 6275 | |
a061fc8d CM |
6276 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
6277 | { | |
0f9dd46c | 6278 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6279 | u64 bytenr; |
0f9dd46c | 6280 | |
a1897fdd LB |
6281 | spin_lock(&root->fs_info->block_group_cache_lock); |
6282 | bytenr = root->fs_info->first_logical_byte; | |
6283 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
6284 | ||
6285 | if (bytenr < (u64)-1) | |
6286 | return bytenr; | |
6287 | ||
0f9dd46c JB |
6288 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
6289 | if (!cache) | |
a061fc8d | 6290 | return 0; |
0f9dd46c | 6291 | |
d2fb3437 | 6292 | bytenr = cache->key.objectid; |
fa9c0d79 | 6293 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6294 | |
6295 | return bytenr; | |
a061fc8d CM |
6296 | } |
6297 | ||
f0486c68 YZ |
6298 | static int pin_down_extent(struct btrfs_root *root, |
6299 | struct btrfs_block_group_cache *cache, | |
6300 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6301 | { |
11833d66 YZ |
6302 | spin_lock(&cache->space_info->lock); |
6303 | spin_lock(&cache->lock); | |
6304 | cache->pinned += num_bytes; | |
6305 | cache->space_info->bytes_pinned += num_bytes; | |
6306 | if (reserved) { | |
6307 | cache->reserved -= num_bytes; | |
6308 | cache->space_info->bytes_reserved -= num_bytes; | |
6309 | } | |
6310 | spin_unlock(&cache->lock); | |
6311 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6312 | |
c51e7bb1 JB |
6313 | trace_btrfs_space_reservation(root->fs_info, "pinned", |
6314 | cache->space_info->flags, num_bytes, 1); | |
f0486c68 YZ |
6315 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
6316 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
6317 | return 0; | |
6318 | } | |
68b38550 | 6319 | |
f0486c68 YZ |
6320 | /* |
6321 | * this function must be called within transaction | |
6322 | */ | |
6323 | int btrfs_pin_extent(struct btrfs_root *root, | |
6324 | u64 bytenr, u64 num_bytes, int reserved) | |
6325 | { | |
6326 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6327 | |
f0486c68 | 6328 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 6329 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
6330 | |
6331 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
6332 | ||
6333 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6334 | return 0; |
6335 | } | |
6336 | ||
f0486c68 | 6337 | /* |
e688b725 CM |
6338 | * this function must be called within transaction |
6339 | */ | |
dcfac415 | 6340 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
e688b725 CM |
6341 | u64 bytenr, u64 num_bytes) |
6342 | { | |
6343 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6344 | int ret; |
e688b725 CM |
6345 | |
6346 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
b50c6e25 JB |
6347 | if (!cache) |
6348 | return -EINVAL; | |
e688b725 CM |
6349 | |
6350 | /* | |
6351 | * pull in the free space cache (if any) so that our pin | |
6352 | * removes the free space from the cache. We have load_only set | |
6353 | * to one because the slow code to read in the free extents does check | |
6354 | * the pinned extents. | |
6355 | */ | |
f6373bf3 | 6356 | cache_block_group(cache, 1); |
e688b725 CM |
6357 | |
6358 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
6359 | ||
6360 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6361 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6362 | btrfs_put_block_group(cache); |
b50c6e25 | 6363 | return ret; |
e688b725 CM |
6364 | } |
6365 | ||
8c2a1a30 JB |
6366 | static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes) |
6367 | { | |
6368 | int ret; | |
6369 | struct btrfs_block_group_cache *block_group; | |
6370 | struct btrfs_caching_control *caching_ctl; | |
6371 | ||
6372 | block_group = btrfs_lookup_block_group(root->fs_info, start); | |
6373 | if (!block_group) | |
6374 | return -EINVAL; | |
6375 | ||
6376 | cache_block_group(block_group, 0); | |
6377 | caching_ctl = get_caching_control(block_group); | |
6378 | ||
6379 | if (!caching_ctl) { | |
6380 | /* Logic error */ | |
6381 | BUG_ON(!block_group_cache_done(block_group)); | |
6382 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6383 | } else { | |
6384 | mutex_lock(&caching_ctl->mutex); | |
6385 | ||
6386 | if (start >= caching_ctl->progress) { | |
6387 | ret = add_excluded_extent(root, start, num_bytes); | |
6388 | } else if (start + num_bytes <= caching_ctl->progress) { | |
6389 | ret = btrfs_remove_free_space(block_group, | |
6390 | start, num_bytes); | |
6391 | } else { | |
6392 | num_bytes = caching_ctl->progress - start; | |
6393 | ret = btrfs_remove_free_space(block_group, | |
6394 | start, num_bytes); | |
6395 | if (ret) | |
6396 | goto out_lock; | |
6397 | ||
6398 | num_bytes = (start + num_bytes) - | |
6399 | caching_ctl->progress; | |
6400 | start = caching_ctl->progress; | |
6401 | ret = add_excluded_extent(root, start, num_bytes); | |
6402 | } | |
6403 | out_lock: | |
6404 | mutex_unlock(&caching_ctl->mutex); | |
6405 | put_caching_control(caching_ctl); | |
6406 | } | |
6407 | btrfs_put_block_group(block_group); | |
6408 | return ret; | |
6409 | } | |
6410 | ||
6411 | int btrfs_exclude_logged_extents(struct btrfs_root *log, | |
6412 | struct extent_buffer *eb) | |
6413 | { | |
6414 | struct btrfs_file_extent_item *item; | |
6415 | struct btrfs_key key; | |
6416 | int found_type; | |
6417 | int i; | |
6418 | ||
6419 | if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) | |
6420 | return 0; | |
6421 | ||
6422 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6423 | btrfs_item_key_to_cpu(eb, &key, i); | |
6424 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6425 | continue; | |
6426 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6427 | found_type = btrfs_file_extent_type(eb, item); | |
6428 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6429 | continue; | |
6430 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6431 | continue; | |
6432 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6433 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
6434 | __exclude_logged_extent(log, key.objectid, key.offset); | |
6435 | } | |
6436 | ||
6437 | return 0; | |
6438 | } | |
6439 | ||
9cfa3e34 FM |
6440 | static void |
6441 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6442 | { | |
6443 | atomic_inc(&bg->reservations); | |
6444 | } | |
6445 | ||
6446 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6447 | const u64 start) | |
6448 | { | |
6449 | struct btrfs_block_group_cache *bg; | |
6450 | ||
6451 | bg = btrfs_lookup_block_group(fs_info, start); | |
6452 | ASSERT(bg); | |
6453 | if (atomic_dec_and_test(&bg->reservations)) | |
6454 | wake_up_atomic_t(&bg->reservations); | |
6455 | btrfs_put_block_group(bg); | |
6456 | } | |
6457 | ||
6458 | static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a) | |
6459 | { | |
6460 | schedule(); | |
6461 | return 0; | |
6462 | } | |
6463 | ||
6464 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6465 | { | |
6466 | struct btrfs_space_info *space_info = bg->space_info; | |
6467 | ||
6468 | ASSERT(bg->ro); | |
6469 | ||
6470 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6471 | return; | |
6472 | ||
6473 | /* | |
6474 | * Our block group is read only but before we set it to read only, | |
6475 | * some task might have had allocated an extent from it already, but it | |
6476 | * has not yet created a respective ordered extent (and added it to a | |
6477 | * root's list of ordered extents). | |
6478 | * Therefore wait for any task currently allocating extents, since the | |
6479 | * block group's reservations counter is incremented while a read lock | |
6480 | * on the groups' semaphore is held and decremented after releasing | |
6481 | * the read access on that semaphore and creating the ordered extent. | |
6482 | */ | |
6483 | down_write(&space_info->groups_sem); | |
6484 | up_write(&space_info->groups_sem); | |
6485 | ||
6486 | wait_on_atomic_t(&bg->reservations, | |
6487 | btrfs_wait_bg_reservations_atomic_t, | |
6488 | TASK_UNINTERRUPTIBLE); | |
6489 | } | |
6490 | ||
fb25e914 | 6491 | /** |
4824f1f4 | 6492 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6493 | * @cache: The cache we are manipulating |
18513091 WX |
6494 | * @ram_bytes: The number of bytes of file content, and will be same to |
6495 | * @num_bytes except for the compress path. | |
fb25e914 | 6496 | * @num_bytes: The number of bytes in question |
e570fd27 | 6497 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6498 | * |
4824f1f4 WX |
6499 | * This is called by the allocator when it reserves space. Metadata |
6500 | * reservations should be called with RESERVE_ALLOC so we do the proper | |
fb25e914 JB |
6501 | * ENOSPC accounting. For data we handle the reservation through clearing the |
6502 | * delalloc bits in the io_tree. We have to do this since we could end up | |
6503 | * allocating less disk space for the amount of data we have reserved in the | |
6504 | * case of compression. | |
6505 | * | |
6506 | * If this is a reservation and the block group has become read only we cannot | |
6507 | * make the reservation and return -EAGAIN, otherwise this function always | |
6508 | * succeeds. | |
f0486c68 | 6509 | */ |
4824f1f4 | 6510 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6511 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6512 | { |
fb25e914 | 6513 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6514 | int ret = 0; |
79787eaa | 6515 | |
fb25e914 JB |
6516 | spin_lock(&space_info->lock); |
6517 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6518 | if (cache->ro) { |
6519 | ret = -EAGAIN; | |
fb25e914 | 6520 | } else { |
4824f1f4 WX |
6521 | cache->reserved += num_bytes; |
6522 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6523 | |
18513091 WX |
6524 | trace_btrfs_space_reservation(cache->fs_info, |
6525 | "space_info", space_info->flags, | |
6526 | ram_bytes, 0); | |
6527 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6528 | if (delalloc) |
4824f1f4 | 6529 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6530 | } |
fb25e914 JB |
6531 | spin_unlock(&cache->lock); |
6532 | spin_unlock(&space_info->lock); | |
f0486c68 | 6533 | return ret; |
324ae4df | 6534 | } |
9078a3e1 | 6535 | |
4824f1f4 WX |
6536 | /** |
6537 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6538 | * @cache: The cache we are manipulating | |
6539 | * @num_bytes: The number of bytes in question | |
6540 | * @delalloc: The blocks are allocated for the delalloc write | |
6541 | * | |
6542 | * This is called by somebody who is freeing space that was never actually used | |
6543 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6544 | * A and before transaction A commits you free that leaf, you call this with | |
6545 | * reserve set to 0 in order to clear the reservation. | |
6546 | */ | |
6547 | ||
6548 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6549 | u64 num_bytes, int delalloc) | |
6550 | { | |
6551 | struct btrfs_space_info *space_info = cache->space_info; | |
6552 | int ret = 0; | |
6553 | ||
6554 | spin_lock(&space_info->lock); | |
6555 | spin_lock(&cache->lock); | |
6556 | if (cache->ro) | |
6557 | space_info->bytes_readonly += num_bytes; | |
6558 | cache->reserved -= num_bytes; | |
6559 | space_info->bytes_reserved -= num_bytes; | |
6560 | ||
6561 | if (delalloc) | |
6562 | cache->delalloc_bytes -= num_bytes; | |
6563 | spin_unlock(&cache->lock); | |
6564 | spin_unlock(&space_info->lock); | |
6565 | return ret; | |
6566 | } | |
143bede5 | 6567 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 6568 | struct btrfs_root *root) |
e8569813 | 6569 | { |
e8569813 | 6570 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
6571 | struct btrfs_caching_control *next; |
6572 | struct btrfs_caching_control *caching_ctl; | |
6573 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6574 | |
9e351cc8 | 6575 | down_write(&fs_info->commit_root_sem); |
25179201 | 6576 | |
11833d66 YZ |
6577 | list_for_each_entry_safe(caching_ctl, next, |
6578 | &fs_info->caching_block_groups, list) { | |
6579 | cache = caching_ctl->block_group; | |
6580 | if (block_group_cache_done(cache)) { | |
6581 | cache->last_byte_to_unpin = (u64)-1; | |
6582 | list_del_init(&caching_ctl->list); | |
6583 | put_caching_control(caching_ctl); | |
e8569813 | 6584 | } else { |
11833d66 | 6585 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6586 | } |
e8569813 | 6587 | } |
11833d66 YZ |
6588 | |
6589 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6590 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6591 | else | |
6592 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6593 | ||
9e351cc8 | 6594 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6595 | |
6596 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6597 | } |
6598 | ||
c759c4e1 JB |
6599 | /* |
6600 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6601 | * what it should be based on the mount options. | |
6602 | */ | |
6603 | static struct btrfs_free_cluster * | |
6604 | fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info, | |
6605 | u64 *empty_cluster) | |
6606 | { | |
6607 | struct btrfs_free_cluster *ret = NULL; | |
3cdde224 | 6608 | bool ssd = btrfs_test_opt(root->fs_info, SSD); |
c759c4e1 JB |
6609 | |
6610 | *empty_cluster = 0; | |
6611 | if (btrfs_mixed_space_info(space_info)) | |
6612 | return ret; | |
6613 | ||
6614 | if (ssd) | |
ee22184b | 6615 | *empty_cluster = SZ_2M; |
c759c4e1 JB |
6616 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
6617 | ret = &root->fs_info->meta_alloc_cluster; | |
6618 | if (!ssd) | |
ee22184b | 6619 | *empty_cluster = SZ_64K; |
c759c4e1 JB |
6620 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) { |
6621 | ret = &root->fs_info->data_alloc_cluster; | |
6622 | } | |
6623 | ||
6624 | return ret; | |
6625 | } | |
6626 | ||
678886bd FM |
6627 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end, |
6628 | const bool return_free_space) | |
ccd467d6 | 6629 | { |
11833d66 YZ |
6630 | struct btrfs_fs_info *fs_info = root->fs_info; |
6631 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
6632 | struct btrfs_space_info *space_info; |
6633 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6634 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6635 | u64 len; |
c759c4e1 JB |
6636 | u64 total_unpinned = 0; |
6637 | u64 empty_cluster = 0; | |
7b398f8e | 6638 | bool readonly; |
ccd467d6 | 6639 | |
11833d66 | 6640 | while (start <= end) { |
7b398f8e | 6641 | readonly = false; |
11833d66 YZ |
6642 | if (!cache || |
6643 | start >= cache->key.objectid + cache->key.offset) { | |
6644 | if (cache) | |
6645 | btrfs_put_block_group(cache); | |
c759c4e1 | 6646 | total_unpinned = 0; |
11833d66 | 6647 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6648 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 JB |
6649 | |
6650 | cluster = fetch_cluster_info(root, | |
6651 | cache->space_info, | |
6652 | &empty_cluster); | |
6653 | empty_cluster <<= 1; | |
11833d66 YZ |
6654 | } |
6655 | ||
6656 | len = cache->key.objectid + cache->key.offset - start; | |
6657 | len = min(len, end + 1 - start); | |
6658 | ||
6659 | if (start < cache->last_byte_to_unpin) { | |
6660 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6661 | if (return_free_space) |
6662 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6663 | } |
6664 | ||
f0486c68 | 6665 | start += len; |
c759c4e1 | 6666 | total_unpinned += len; |
7b398f8e | 6667 | space_info = cache->space_info; |
f0486c68 | 6668 | |
c759c4e1 JB |
6669 | /* |
6670 | * If this space cluster has been marked as fragmented and we've | |
6671 | * unpinned enough in this block group to potentially allow a | |
6672 | * cluster to be created inside of it go ahead and clear the | |
6673 | * fragmented check. | |
6674 | */ | |
6675 | if (cluster && cluster->fragmented && | |
6676 | total_unpinned > empty_cluster) { | |
6677 | spin_lock(&cluster->lock); | |
6678 | cluster->fragmented = 0; | |
6679 | spin_unlock(&cluster->lock); | |
6680 | } | |
6681 | ||
7b398f8e | 6682 | spin_lock(&space_info->lock); |
11833d66 YZ |
6683 | spin_lock(&cache->lock); |
6684 | cache->pinned -= len; | |
7b398f8e | 6685 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6686 | |
6687 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6688 | space_info->flags, len, 0); | |
4f4db217 | 6689 | space_info->max_extent_size = 0; |
d288db5d | 6690 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6691 | if (cache->ro) { |
6692 | space_info->bytes_readonly += len; | |
6693 | readonly = true; | |
6694 | } | |
11833d66 | 6695 | spin_unlock(&cache->lock); |
957780eb JB |
6696 | if (!readonly && return_free_space && |
6697 | global_rsv->space_info == space_info) { | |
6698 | u64 to_add = len; | |
6699 | WARN_ON(!return_free_space); | |
7b398f8e JB |
6700 | spin_lock(&global_rsv->lock); |
6701 | if (!global_rsv->full) { | |
957780eb JB |
6702 | to_add = min(len, global_rsv->size - |
6703 | global_rsv->reserved); | |
6704 | global_rsv->reserved += to_add; | |
6705 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6706 | if (global_rsv->reserved >= global_rsv->size) |
6707 | global_rsv->full = 1; | |
957780eb JB |
6708 | trace_btrfs_space_reservation(fs_info, |
6709 | "space_info", | |
6710 | space_info->flags, | |
6711 | to_add, 1); | |
6712 | len -= to_add; | |
7b398f8e JB |
6713 | } |
6714 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6715 | /* Add to any tickets we may have */ |
6716 | if (len) | |
6717 | space_info_add_new_bytes(fs_info, space_info, | |
6718 | len); | |
7b398f8e JB |
6719 | } |
6720 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6721 | } |
11833d66 YZ |
6722 | |
6723 | if (cache) | |
6724 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6725 | return 0; |
6726 | } | |
6727 | ||
6728 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 6729 | struct btrfs_root *root) |
a28ec197 | 6730 | { |
11833d66 | 6731 | struct btrfs_fs_info *fs_info = root->fs_info; |
e33e17ee JM |
6732 | struct btrfs_block_group_cache *block_group, *tmp; |
6733 | struct list_head *deleted_bgs; | |
11833d66 | 6734 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6735 | u64 start; |
6736 | u64 end; | |
a28ec197 | 6737 | int ret; |
a28ec197 | 6738 | |
11833d66 YZ |
6739 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6740 | unpin = &fs_info->freed_extents[1]; | |
6741 | else | |
6742 | unpin = &fs_info->freed_extents[0]; | |
6743 | ||
e33e17ee | 6744 | while (!trans->aborted) { |
d4b450cd | 6745 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6746 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6747 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6748 | if (ret) { |
6749 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6750 | break; |
d4b450cd | 6751 | } |
1f3c79a2 | 6752 | |
3cdde224 | 6753 | if (btrfs_test_opt(root->fs_info, DISCARD)) |
5378e607 LD |
6754 | ret = btrfs_discard_extent(root, start, |
6755 | end + 1 - start, NULL); | |
1f3c79a2 | 6756 | |
af6f8f60 | 6757 | clear_extent_dirty(unpin, start, end); |
678886bd | 6758 | unpin_extent_range(root, start, end, true); |
d4b450cd | 6759 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6760 | cond_resched(); |
a28ec197 | 6761 | } |
817d52f8 | 6762 | |
e33e17ee JM |
6763 | /* |
6764 | * Transaction is finished. We don't need the lock anymore. We | |
6765 | * do need to clean up the block groups in case of a transaction | |
6766 | * abort. | |
6767 | */ | |
6768 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6769 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6770 | u64 trimmed = 0; | |
6771 | ||
6772 | ret = -EROFS; | |
6773 | if (!trans->aborted) | |
6774 | ret = btrfs_discard_extent(root, | |
6775 | block_group->key.objectid, | |
6776 | block_group->key.offset, | |
6777 | &trimmed); | |
6778 | ||
6779 | list_del_init(&block_group->bg_list); | |
6780 | btrfs_put_block_group_trimming(block_group); | |
6781 | btrfs_put_block_group(block_group); | |
6782 | ||
6783 | if (ret) { | |
6784 | const char *errstr = btrfs_decode_error(ret); | |
6785 | btrfs_warn(fs_info, | |
6786 | "Discard failed while removing blockgroup: errno=%d %s\n", | |
6787 | ret, errstr); | |
6788 | } | |
6789 | } | |
6790 | ||
e20d96d6 CM |
6791 | return 0; |
6792 | } | |
6793 | ||
b150a4f1 JB |
6794 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, |
6795 | u64 owner, u64 root_objectid) | |
6796 | { | |
6797 | struct btrfs_space_info *space_info; | |
6798 | u64 flags; | |
6799 | ||
6800 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
6801 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
6802 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
6803 | else | |
6804 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
6805 | } else { | |
6806 | flags = BTRFS_BLOCK_GROUP_DATA; | |
6807 | } | |
6808 | ||
6809 | space_info = __find_space_info(fs_info, flags); | |
6810 | BUG_ON(!space_info); /* Logic bug */ | |
6811 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); | |
6812 | } | |
6813 | ||
6814 | ||
5d4f98a2 YZ |
6815 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
6816 | struct btrfs_root *root, | |
c682f9b3 | 6817 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6818 | u64 root_objectid, u64 owner_objectid, |
6819 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6820 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6821 | { |
e2fa7227 | 6822 | struct btrfs_key key; |
5d4f98a2 | 6823 | struct btrfs_path *path; |
1261ec42 CM |
6824 | struct btrfs_fs_info *info = root->fs_info; |
6825 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 6826 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6827 | struct btrfs_extent_item *ei; |
6828 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6829 | int ret; |
5d4f98a2 | 6830 | int is_data; |
952fccac CM |
6831 | int extent_slot = 0; |
6832 | int found_extent = 0; | |
6833 | int num_to_del = 1; | |
5d4f98a2 YZ |
6834 | u32 item_size; |
6835 | u64 refs; | |
c682f9b3 QW |
6836 | u64 bytenr = node->bytenr; |
6837 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6838 | int last_ref = 0; |
3173a18f JB |
6839 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
6840 | SKINNY_METADATA); | |
037e6390 | 6841 | |
5caf2a00 | 6842 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6843 | if (!path) |
6844 | return -ENOMEM; | |
5f26f772 | 6845 | |
e4058b54 | 6846 | path->reada = READA_FORWARD; |
b9473439 | 6847 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6848 | |
6849 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6850 | BUG_ON(!is_data && refs_to_drop != 1); | |
6851 | ||
3173a18f JB |
6852 | if (is_data) |
6853 | skinny_metadata = 0; | |
6854 | ||
5d4f98a2 YZ |
6855 | ret = lookup_extent_backref(trans, extent_root, path, &iref, |
6856 | bytenr, num_bytes, parent, | |
6857 | root_objectid, owner_objectid, | |
6858 | owner_offset); | |
7bb86316 | 6859 | if (ret == 0) { |
952fccac | 6860 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6861 | while (extent_slot >= 0) { |
6862 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6863 | extent_slot); |
5d4f98a2 | 6864 | if (key.objectid != bytenr) |
952fccac | 6865 | break; |
5d4f98a2 YZ |
6866 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6867 | key.offset == num_bytes) { | |
952fccac CM |
6868 | found_extent = 1; |
6869 | break; | |
6870 | } | |
3173a18f JB |
6871 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6872 | key.offset == owner_objectid) { | |
6873 | found_extent = 1; | |
6874 | break; | |
6875 | } | |
952fccac CM |
6876 | if (path->slots[0] - extent_slot > 5) |
6877 | break; | |
5d4f98a2 | 6878 | extent_slot--; |
952fccac | 6879 | } |
5d4f98a2 YZ |
6880 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6881 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6882 | if (found_extent && item_size < sizeof(*ei)) | |
6883 | found_extent = 0; | |
6884 | #endif | |
31840ae1 | 6885 | if (!found_extent) { |
5d4f98a2 | 6886 | BUG_ON(iref); |
56bec294 | 6887 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 | 6888 | NULL, refs_to_drop, |
fcebe456 | 6889 | is_data, &last_ref); |
005d6427 | 6890 | if (ret) { |
66642832 | 6891 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6892 | goto out; |
6893 | } | |
b3b4aa74 | 6894 | btrfs_release_path(path); |
b9473439 | 6895 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6896 | |
6897 | key.objectid = bytenr; | |
6898 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6899 | key.offset = num_bytes; | |
6900 | ||
3173a18f JB |
6901 | if (!is_data && skinny_metadata) { |
6902 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6903 | key.offset = owner_objectid; | |
6904 | } | |
6905 | ||
31840ae1 ZY |
6906 | ret = btrfs_search_slot(trans, extent_root, |
6907 | &key, path, -1, 1); | |
3173a18f JB |
6908 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6909 | /* | |
6910 | * Couldn't find our skinny metadata item, | |
6911 | * see if we have ye olde extent item. | |
6912 | */ | |
6913 | path->slots[0]--; | |
6914 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6915 | path->slots[0]); | |
6916 | if (key.objectid == bytenr && | |
6917 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6918 | key.offset == num_bytes) | |
6919 | ret = 0; | |
6920 | } | |
6921 | ||
6922 | if (ret > 0 && skinny_metadata) { | |
6923 | skinny_metadata = false; | |
9ce49a0b | 6924 | key.objectid = bytenr; |
3173a18f JB |
6925 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6926 | key.offset = num_bytes; | |
6927 | btrfs_release_path(path); | |
6928 | ret = btrfs_search_slot(trans, extent_root, | |
6929 | &key, path, -1, 1); | |
6930 | } | |
6931 | ||
f3465ca4 | 6932 | if (ret) { |
5d163e0e JM |
6933 | btrfs_err(info, |
6934 | "umm, got %d back from search, was looking for %llu", | |
6935 | ret, bytenr); | |
b783e62d JB |
6936 | if (ret > 0) |
6937 | btrfs_print_leaf(extent_root, | |
6938 | path->nodes[0]); | |
f3465ca4 | 6939 | } |
005d6427 | 6940 | if (ret < 0) { |
66642832 | 6941 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6942 | goto out; |
6943 | } | |
31840ae1 ZY |
6944 | extent_slot = path->slots[0]; |
6945 | } | |
fae7f21c | 6946 | } else if (WARN_ON(ret == -ENOENT)) { |
7bb86316 | 6947 | btrfs_print_leaf(extent_root, path->nodes[0]); |
c2cf52eb SK |
6948 | btrfs_err(info, |
6949 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6950 | bytenr, parent, root_objectid, owner_objectid, |
6951 | owner_offset); | |
66642832 | 6952 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6953 | goto out; |
79787eaa | 6954 | } else { |
66642832 | 6955 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6956 | goto out; |
7bb86316 | 6957 | } |
5f39d397 CM |
6958 | |
6959 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6960 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6961 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6962 | if (item_size < sizeof(*ei)) { | |
6963 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
6964 | ret = convert_extent_item_v0(trans, extent_root, path, | |
6965 | owner_objectid, 0); | |
005d6427 | 6966 | if (ret < 0) { |
66642832 | 6967 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6968 | goto out; |
6969 | } | |
5d4f98a2 | 6970 | |
b3b4aa74 | 6971 | btrfs_release_path(path); |
5d4f98a2 YZ |
6972 | path->leave_spinning = 1; |
6973 | ||
6974 | key.objectid = bytenr; | |
6975 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6976 | key.offset = num_bytes; | |
6977 | ||
6978 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
6979 | -1, 1); | |
6980 | if (ret) { | |
5d163e0e JM |
6981 | btrfs_err(info, |
6982 | "umm, got %d back from search, was looking for %llu", | |
c1c9ff7c | 6983 | ret, bytenr); |
5d4f98a2 YZ |
6984 | btrfs_print_leaf(extent_root, path->nodes[0]); |
6985 | } | |
005d6427 | 6986 | if (ret < 0) { |
66642832 | 6987 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6988 | goto out; |
6989 | } | |
6990 | ||
5d4f98a2 YZ |
6991 | extent_slot = path->slots[0]; |
6992 | leaf = path->nodes[0]; | |
6993 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
6994 | } | |
6995 | #endif | |
6996 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 6997 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 6998 | struct btrfs_extent_item); |
3173a18f JB |
6999 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7000 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7001 | struct btrfs_tree_block_info *bi; |
7002 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7003 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7004 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7005 | } | |
56bec294 | 7006 | |
5d4f98a2 | 7007 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7008 | if (refs < refs_to_drop) { |
5d163e0e JM |
7009 | btrfs_err(info, |
7010 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7011 | refs_to_drop, refs, bytenr); | |
32b02538 | 7012 | ret = -EINVAL; |
66642832 | 7013 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7014 | goto out; |
7015 | } | |
56bec294 | 7016 | refs -= refs_to_drop; |
5f39d397 | 7017 | |
5d4f98a2 YZ |
7018 | if (refs > 0) { |
7019 | if (extent_op) | |
7020 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7021 | /* | |
7022 | * In the case of inline back ref, reference count will | |
7023 | * be updated by remove_extent_backref | |
952fccac | 7024 | */ |
5d4f98a2 YZ |
7025 | if (iref) { |
7026 | BUG_ON(!found_extent); | |
7027 | } else { | |
7028 | btrfs_set_extent_refs(leaf, ei, refs); | |
7029 | btrfs_mark_buffer_dirty(leaf); | |
7030 | } | |
7031 | if (found_extent) { | |
7032 | ret = remove_extent_backref(trans, extent_root, path, | |
7033 | iref, refs_to_drop, | |
fcebe456 | 7034 | is_data, &last_ref); |
005d6427 | 7035 | if (ret) { |
66642832 | 7036 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7037 | goto out; |
7038 | } | |
952fccac | 7039 | } |
b150a4f1 JB |
7040 | add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid, |
7041 | root_objectid); | |
5d4f98a2 | 7042 | } else { |
5d4f98a2 YZ |
7043 | if (found_extent) { |
7044 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7045 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7046 | if (iref) { |
7047 | BUG_ON(path->slots[0] != extent_slot); | |
7048 | } else { | |
7049 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7050 | path->slots[0] = extent_slot; | |
7051 | num_to_del = 2; | |
7052 | } | |
78fae27e | 7053 | } |
b9473439 | 7054 | |
fcebe456 | 7055 | last_ref = 1; |
952fccac CM |
7056 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7057 | num_to_del); | |
005d6427 | 7058 | if (ret) { |
66642832 | 7059 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7060 | goto out; |
7061 | } | |
b3b4aa74 | 7062 | btrfs_release_path(path); |
21af804c | 7063 | |
5d4f98a2 | 7064 | if (is_data) { |
459931ec | 7065 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 | 7066 | if (ret) { |
66642832 | 7067 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7068 | goto out; |
7069 | } | |
459931ec CM |
7070 | } |
7071 | ||
1e144fb8 OS |
7072 | ret = add_to_free_space_tree(trans, root->fs_info, bytenr, |
7073 | num_bytes); | |
7074 | if (ret) { | |
66642832 | 7075 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7076 | goto out; |
7077 | } | |
7078 | ||
ce93ec54 | 7079 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
005d6427 | 7080 | if (ret) { |
66642832 | 7081 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7082 | goto out; |
7083 | } | |
a28ec197 | 7084 | } |
fcebe456 JB |
7085 | btrfs_release_path(path); |
7086 | ||
79787eaa | 7087 | out: |
5caf2a00 | 7088 | btrfs_free_path(path); |
a28ec197 CM |
7089 | return ret; |
7090 | } | |
7091 | ||
1887be66 | 7092 | /* |
f0486c68 | 7093 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7094 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7095 | * a given extent, and if there are no other delayed refs to be processed, it | |
7096 | * removes it from the tree. | |
7097 | */ | |
7098 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
7099 | struct btrfs_root *root, u64 bytenr) | |
7100 | { | |
7101 | struct btrfs_delayed_ref_head *head; | |
7102 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7103 | int ret = 0; |
1887be66 CM |
7104 | |
7105 | delayed_refs = &trans->transaction->delayed_refs; | |
7106 | spin_lock(&delayed_refs->lock); | |
7107 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
7108 | if (!head) | |
cf93da7b | 7109 | goto out_delayed_unlock; |
1887be66 | 7110 | |
d7df2c79 | 7111 | spin_lock(&head->lock); |
c6fc2454 | 7112 | if (!list_empty(&head->ref_list)) |
1887be66 CM |
7113 | goto out; |
7114 | ||
5d4f98a2 YZ |
7115 | if (head->extent_op) { |
7116 | if (!head->must_insert_reserved) | |
7117 | goto out; | |
78a6184a | 7118 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7119 | head->extent_op = NULL; |
7120 | } | |
7121 | ||
1887be66 CM |
7122 | /* |
7123 | * waiting for the lock here would deadlock. If someone else has it | |
7124 | * locked they are already in the process of dropping it anyway | |
7125 | */ | |
7126 | if (!mutex_trylock(&head->mutex)) | |
7127 | goto out; | |
7128 | ||
7129 | /* | |
7130 | * at this point we have a head with no other entries. Go | |
7131 | * ahead and process it. | |
7132 | */ | |
7133 | head->node.in_tree = 0; | |
c46effa6 | 7134 | rb_erase(&head->href_node, &delayed_refs->href_root); |
c3e69d58 | 7135 | |
d7df2c79 | 7136 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7137 | |
7138 | /* | |
7139 | * we don't take a ref on the node because we're removing it from the | |
7140 | * tree, so we just steal the ref the tree was holding. | |
7141 | */ | |
c3e69d58 | 7142 | delayed_refs->num_heads--; |
d7df2c79 | 7143 | if (head->processing == 0) |
c3e69d58 | 7144 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7145 | head->processing = 0; |
7146 | spin_unlock(&head->lock); | |
1887be66 CM |
7147 | spin_unlock(&delayed_refs->lock); |
7148 | ||
f0486c68 YZ |
7149 | BUG_ON(head->extent_op); |
7150 | if (head->must_insert_reserved) | |
7151 | ret = 1; | |
7152 | ||
7153 | mutex_unlock(&head->mutex); | |
1887be66 | 7154 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 7155 | return ret; |
1887be66 | 7156 | out: |
d7df2c79 | 7157 | spin_unlock(&head->lock); |
cf93da7b CM |
7158 | |
7159 | out_delayed_unlock: | |
1887be66 CM |
7160 | spin_unlock(&delayed_refs->lock); |
7161 | return 0; | |
7162 | } | |
7163 | ||
f0486c68 YZ |
7164 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7165 | struct btrfs_root *root, | |
7166 | struct extent_buffer *buf, | |
5581a51a | 7167 | u64 parent, int last_ref) |
f0486c68 | 7168 | { |
b150a4f1 | 7169 | int pin = 1; |
f0486c68 YZ |
7170 | int ret; |
7171 | ||
7172 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
7173 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
7174 | buf->start, buf->len, | |
7175 | parent, root->root_key.objectid, | |
7176 | btrfs_header_level(buf), | |
b06c4bf5 | 7177 | BTRFS_DROP_DELAYED_REF, NULL); |
79787eaa | 7178 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
7179 | } |
7180 | ||
7181 | if (!last_ref) | |
7182 | return; | |
7183 | ||
f0486c68 | 7184 | if (btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7185 | struct btrfs_block_group_cache *cache; |
7186 | ||
f0486c68 YZ |
7187 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
7188 | ret = check_ref_cleanup(trans, root, buf->start); | |
7189 | if (!ret) | |
37be25bc | 7190 | goto out; |
f0486c68 YZ |
7191 | } |
7192 | ||
6219872d FM |
7193 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
7194 | ||
f0486c68 YZ |
7195 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
7196 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
6219872d | 7197 | btrfs_put_block_group(cache); |
37be25bc | 7198 | goto out; |
f0486c68 YZ |
7199 | } |
7200 | ||
7201 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7202 | ||
7203 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7204 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7205 | btrfs_put_block_group(cache); |
0be5dc67 | 7206 | trace_btrfs_reserved_extent_free(root, buf->start, buf->len); |
b150a4f1 | 7207 | pin = 0; |
f0486c68 YZ |
7208 | } |
7209 | out: | |
b150a4f1 JB |
7210 | if (pin) |
7211 | add_pinned_bytes(root->fs_info, buf->len, | |
7212 | btrfs_header_level(buf), | |
7213 | root->root_key.objectid); | |
7214 | ||
a826d6dc JB |
7215 | /* |
7216 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
7217 | * anymore. | |
7218 | */ | |
7219 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
7220 | } |
7221 | ||
79787eaa | 7222 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
7223 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
7224 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
b06c4bf5 | 7225 | u64 owner, u64 offset) |
925baedd CM |
7226 | { |
7227 | int ret; | |
66d7e7f0 | 7228 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 7229 | |
f5ee5c9a | 7230 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7231 | return 0; |
fccb84c9 | 7232 | |
b150a4f1 JB |
7233 | add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid); |
7234 | ||
56bec294 CM |
7235 | /* |
7236 | * tree log blocks never actually go into the extent allocation | |
7237 | * tree, just update pinning info and exit early. | |
56bec294 | 7238 | */ |
5d4f98a2 YZ |
7239 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7240 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7241 | /* unlocks the pinned mutex */ |
11833d66 | 7242 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 7243 | ret = 0; |
5d4f98a2 | 7244 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
7245 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7246 | num_bytes, | |
5d4f98a2 | 7247 | parent, root_objectid, (int)owner, |
b06c4bf5 | 7248 | BTRFS_DROP_DELAYED_REF, NULL); |
5d4f98a2 | 7249 | } else { |
66d7e7f0 AJ |
7250 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7251 | num_bytes, | |
7252 | parent, root_objectid, owner, | |
5846a3c2 QW |
7253 | offset, 0, |
7254 | BTRFS_DROP_DELAYED_REF, NULL); | |
56bec294 | 7255 | } |
925baedd CM |
7256 | return ret; |
7257 | } | |
7258 | ||
817d52f8 JB |
7259 | /* |
7260 | * when we wait for progress in the block group caching, its because | |
7261 | * our allocation attempt failed at least once. So, we must sleep | |
7262 | * and let some progress happen before we try again. | |
7263 | * | |
7264 | * This function will sleep at least once waiting for new free space to | |
7265 | * show up, and then it will check the block group free space numbers | |
7266 | * for our min num_bytes. Another option is to have it go ahead | |
7267 | * and look in the rbtree for a free extent of a given size, but this | |
7268 | * is a good start. | |
36cce922 JB |
7269 | * |
7270 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7271 | * any of the information in this block group. | |
817d52f8 | 7272 | */ |
36cce922 | 7273 | static noinline void |
817d52f8 JB |
7274 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7275 | u64 num_bytes) | |
7276 | { | |
11833d66 | 7277 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7278 | |
11833d66 YZ |
7279 | caching_ctl = get_caching_control(cache); |
7280 | if (!caching_ctl) | |
36cce922 | 7281 | return; |
817d52f8 | 7282 | |
11833d66 | 7283 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7284 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7285 | |
7286 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7287 | } |
7288 | ||
7289 | static noinline int | |
7290 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7291 | { | |
7292 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7293 | int ret = 0; |
11833d66 YZ |
7294 | |
7295 | caching_ctl = get_caching_control(cache); | |
7296 | if (!caching_ctl) | |
36cce922 | 7297 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7298 | |
7299 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7300 | if (cache->cached == BTRFS_CACHE_ERROR) |
7301 | ret = -EIO; | |
11833d66 | 7302 | put_caching_control(caching_ctl); |
36cce922 | 7303 | return ret; |
817d52f8 JB |
7304 | } |
7305 | ||
31e50229 | 7306 | int __get_raid_index(u64 flags) |
b742bb82 | 7307 | { |
7738a53a | 7308 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 7309 | return BTRFS_RAID_RAID10; |
7738a53a | 7310 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 7311 | return BTRFS_RAID_RAID1; |
7738a53a | 7312 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 7313 | return BTRFS_RAID_DUP; |
7738a53a | 7314 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 7315 | return BTRFS_RAID_RAID0; |
53b381b3 | 7316 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 7317 | return BTRFS_RAID_RAID5; |
53b381b3 | 7318 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 7319 | return BTRFS_RAID_RAID6; |
7738a53a | 7320 | |
e942f883 | 7321 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
7322 | } |
7323 | ||
6ab0a202 | 7324 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
7738a53a | 7325 | { |
31e50229 | 7326 | return __get_raid_index(cache->flags); |
7738a53a ID |
7327 | } |
7328 | ||
6ab0a202 JM |
7329 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
7330 | [BTRFS_RAID_RAID10] = "raid10", | |
7331 | [BTRFS_RAID_RAID1] = "raid1", | |
7332 | [BTRFS_RAID_DUP] = "dup", | |
7333 | [BTRFS_RAID_RAID0] = "raid0", | |
7334 | [BTRFS_RAID_SINGLE] = "single", | |
7335 | [BTRFS_RAID_RAID5] = "raid5", | |
7336 | [BTRFS_RAID_RAID6] = "raid6", | |
7337 | }; | |
7338 | ||
1b8e5df6 | 7339 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
7340 | { |
7341 | if (type >= BTRFS_NR_RAID_TYPES) | |
7342 | return NULL; | |
7343 | ||
7344 | return btrfs_raid_type_names[type]; | |
7345 | } | |
7346 | ||
817d52f8 | 7347 | enum btrfs_loop_type { |
285ff5af JB |
7348 | LOOP_CACHING_NOWAIT = 0, |
7349 | LOOP_CACHING_WAIT = 1, | |
7350 | LOOP_ALLOC_CHUNK = 2, | |
7351 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7352 | }; |
7353 | ||
e570fd27 MX |
7354 | static inline void |
7355 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7356 | int delalloc) | |
7357 | { | |
7358 | if (delalloc) | |
7359 | down_read(&cache->data_rwsem); | |
7360 | } | |
7361 | ||
7362 | static inline void | |
7363 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7364 | int delalloc) | |
7365 | { | |
7366 | btrfs_get_block_group(cache); | |
7367 | if (delalloc) | |
7368 | down_read(&cache->data_rwsem); | |
7369 | } | |
7370 | ||
7371 | static struct btrfs_block_group_cache * | |
7372 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7373 | struct btrfs_free_cluster *cluster, | |
7374 | int delalloc) | |
7375 | { | |
89771cc9 | 7376 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7377 | |
e570fd27 | 7378 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7379 | while (1) { |
7380 | used_bg = cluster->block_group; | |
7381 | if (!used_bg) | |
7382 | return NULL; | |
7383 | ||
7384 | if (used_bg == block_group) | |
e570fd27 MX |
7385 | return used_bg; |
7386 | ||
6719afdc | 7387 | btrfs_get_block_group(used_bg); |
e570fd27 | 7388 | |
6719afdc GU |
7389 | if (!delalloc) |
7390 | return used_bg; | |
e570fd27 | 7391 | |
6719afdc GU |
7392 | if (down_read_trylock(&used_bg->data_rwsem)) |
7393 | return used_bg; | |
e570fd27 | 7394 | |
6719afdc | 7395 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7396 | |
6719afdc | 7397 | down_read(&used_bg->data_rwsem); |
e570fd27 | 7398 | |
6719afdc GU |
7399 | spin_lock(&cluster->refill_lock); |
7400 | if (used_bg == cluster->block_group) | |
7401 | return used_bg; | |
e570fd27 | 7402 | |
6719afdc GU |
7403 | up_read(&used_bg->data_rwsem); |
7404 | btrfs_put_block_group(used_bg); | |
7405 | } | |
e570fd27 MX |
7406 | } |
7407 | ||
7408 | static inline void | |
7409 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7410 | int delalloc) | |
7411 | { | |
7412 | if (delalloc) | |
7413 | up_read(&cache->data_rwsem); | |
7414 | btrfs_put_block_group(cache); | |
7415 | } | |
7416 | ||
fec577fb CM |
7417 | /* |
7418 | * walks the btree of allocated extents and find a hole of a given size. | |
7419 | * The key ins is changed to record the hole: | |
a4820398 | 7420 | * ins->objectid == start position |
62e2749e | 7421 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7422 | * ins->offset == the size of the hole. |
fec577fb | 7423 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7424 | * |
7425 | * If there is no suitable free space, we will record the max size of | |
7426 | * the free space extent currently. | |
fec577fb | 7427 | */ |
00361589 | 7428 | static noinline int find_free_extent(struct btrfs_root *orig_root, |
18513091 WX |
7429 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7430 | u64 hint_byte, struct btrfs_key *ins, | |
7431 | u64 flags, int delalloc) | |
fec577fb | 7432 | { |
80eb234a | 7433 | int ret = 0; |
d397712b | 7434 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 7435 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7436 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7437 | u64 search_start = 0; |
a4820398 | 7438 | u64 max_extent_size = 0; |
c759c4e1 | 7439 | u64 empty_cluster = 0; |
80eb234a | 7440 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7441 | int loop = 0; |
b6919a58 | 7442 | int index = __get_raid_index(flags); |
0a24325e | 7443 | bool failed_cluster_refill = false; |
1cdda9b8 | 7444 | bool failed_alloc = false; |
67377734 | 7445 | bool use_cluster = true; |
60d2adbb | 7446 | bool have_caching_bg = false; |
13a0db5a | 7447 | bool orig_have_caching_bg = false; |
a5e681d9 | 7448 | bool full_search = false; |
fec577fb | 7449 | |
db94535d | 7450 | WARN_ON(num_bytes < root->sectorsize); |
962a298f | 7451 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7452 | ins->objectid = 0; |
7453 | ins->offset = 0; | |
b1a4d965 | 7454 | |
b6919a58 | 7455 | trace_find_free_extent(orig_root, num_bytes, empty_size, flags); |
3f7de037 | 7456 | |
b6919a58 | 7457 | space_info = __find_space_info(root->fs_info, flags); |
1b1d1f66 | 7458 | if (!space_info) { |
b6919a58 | 7459 | btrfs_err(root->fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7460 | return -ENOSPC; |
7461 | } | |
2552d17e | 7462 | |
67377734 | 7463 | /* |
4f4db217 JB |
7464 | * If our free space is heavily fragmented we may not be able to make |
7465 | * big contiguous allocations, so instead of doing the expensive search | |
7466 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7467 | * can go ahead and search for a more manageable chunk. | |
7468 | * | |
7469 | * If our max_extent_size is large enough for our allocation simply | |
7470 | * disable clustering since we will likely not be able to find enough | |
7471 | * space to create a cluster and induce latency trying. | |
67377734 | 7472 | */ |
4f4db217 JB |
7473 | if (unlikely(space_info->max_extent_size)) { |
7474 | spin_lock(&space_info->lock); | |
7475 | if (space_info->max_extent_size && | |
7476 | num_bytes > space_info->max_extent_size) { | |
7477 | ins->offset = space_info->max_extent_size; | |
7478 | spin_unlock(&space_info->lock); | |
7479 | return -ENOSPC; | |
7480 | } else if (space_info->max_extent_size) { | |
7481 | use_cluster = false; | |
7482 | } | |
7483 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7484 | } |
0f9dd46c | 7485 | |
c759c4e1 | 7486 | last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster); |
239b14b3 | 7487 | if (last_ptr) { |
fa9c0d79 CM |
7488 | spin_lock(&last_ptr->lock); |
7489 | if (last_ptr->block_group) | |
7490 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7491 | if (last_ptr->fragmented) { |
7492 | /* | |
7493 | * We still set window_start so we can keep track of the | |
7494 | * last place we found an allocation to try and save | |
7495 | * some time. | |
7496 | */ | |
7497 | hint_byte = last_ptr->window_start; | |
7498 | use_cluster = false; | |
7499 | } | |
fa9c0d79 | 7500 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7501 | } |
fa9c0d79 | 7502 | |
a061fc8d | 7503 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 7504 | search_start = max(search_start, hint_byte); |
2552d17e | 7505 | if (search_start == hint_byte) { |
2552d17e JB |
7506 | block_group = btrfs_lookup_block_group(root->fs_info, |
7507 | search_start); | |
817d52f8 JB |
7508 | /* |
7509 | * we don't want to use the block group if it doesn't match our | |
7510 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7511 | * |
7512 | * However if we are re-searching with an ideal block group | |
7513 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7514 | */ |
b6919a58 | 7515 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7516 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7517 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7518 | if (list_empty(&block_group->list) || |
7519 | block_group->ro) { | |
7520 | /* | |
7521 | * someone is removing this block group, | |
7522 | * we can't jump into the have_block_group | |
7523 | * target because our list pointers are not | |
7524 | * valid | |
7525 | */ | |
7526 | btrfs_put_block_group(block_group); | |
7527 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7528 | } else { |
b742bb82 | 7529 | index = get_block_group_index(block_group); |
e570fd27 | 7530 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7531 | goto have_block_group; |
ccf0e725 | 7532 | } |
2552d17e | 7533 | } else if (block_group) { |
fa9c0d79 | 7534 | btrfs_put_block_group(block_group); |
2552d17e | 7535 | } |
42e70e7a | 7536 | } |
2552d17e | 7537 | search: |
60d2adbb | 7538 | have_caching_bg = false; |
a5e681d9 JB |
7539 | if (index == 0 || index == __get_raid_index(flags)) |
7540 | full_search = true; | |
80eb234a | 7541 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7542 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7543 | list) { | |
6226cb0a | 7544 | u64 offset; |
817d52f8 | 7545 | int cached; |
8a1413a2 | 7546 | |
e570fd27 | 7547 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7548 | search_start = block_group->key.objectid; |
42e70e7a | 7549 | |
83a50de9 CM |
7550 | /* |
7551 | * this can happen if we end up cycling through all the | |
7552 | * raid types, but we want to make sure we only allocate | |
7553 | * for the proper type. | |
7554 | */ | |
b6919a58 | 7555 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7556 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7557 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7558 | BTRFS_BLOCK_GROUP_RAID5 | |
7559 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7560 | BTRFS_BLOCK_GROUP_RAID10; |
7561 | ||
7562 | /* | |
7563 | * if they asked for extra copies and this block group | |
7564 | * doesn't provide them, bail. This does allow us to | |
7565 | * fill raid0 from raid1. | |
7566 | */ | |
b6919a58 | 7567 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7568 | goto loop; |
7569 | } | |
7570 | ||
2552d17e | 7571 | have_block_group: |
291c7d2f JB |
7572 | cached = block_group_cache_done(block_group); |
7573 | if (unlikely(!cached)) { | |
a5e681d9 | 7574 | have_caching_bg = true; |
f6373bf3 | 7575 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7576 | BUG_ON(ret < 0); |
7577 | ret = 0; | |
817d52f8 JB |
7578 | } |
7579 | ||
36cce922 JB |
7580 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7581 | goto loop; | |
ea6a478e | 7582 | if (unlikely(block_group->ro)) |
2552d17e | 7583 | goto loop; |
0f9dd46c | 7584 | |
0a24325e | 7585 | /* |
062c05c4 AO |
7586 | * Ok we want to try and use the cluster allocator, so |
7587 | * lets look there | |
0a24325e | 7588 | */ |
c759c4e1 | 7589 | if (last_ptr && use_cluster) { |
215a63d1 | 7590 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7591 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7592 | /* |
7593 | * the refill lock keeps out other | |
7594 | * people trying to start a new cluster | |
7595 | */ | |
e570fd27 MX |
7596 | used_block_group = btrfs_lock_cluster(block_group, |
7597 | last_ptr, | |
7598 | delalloc); | |
7599 | if (!used_block_group) | |
44fb5511 | 7600 | goto refill_cluster; |
274bd4fb | 7601 | |
e570fd27 MX |
7602 | if (used_block_group != block_group && |
7603 | (used_block_group->ro || | |
7604 | !block_group_bits(used_block_group, flags))) | |
7605 | goto release_cluster; | |
44fb5511 | 7606 | |
274bd4fb | 7607 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7608 | last_ptr, |
7609 | num_bytes, | |
7610 | used_block_group->key.objectid, | |
7611 | &max_extent_size); | |
fa9c0d79 CM |
7612 | if (offset) { |
7613 | /* we have a block, we're done */ | |
7614 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 | 7615 | trace_btrfs_reserve_extent_cluster(root, |
89d4346a MX |
7616 | used_block_group, |
7617 | search_start, num_bytes); | |
215a63d1 | 7618 | if (used_block_group != block_group) { |
e570fd27 MX |
7619 | btrfs_release_block_group(block_group, |
7620 | delalloc); | |
215a63d1 MX |
7621 | block_group = used_block_group; |
7622 | } | |
fa9c0d79 CM |
7623 | goto checks; |
7624 | } | |
7625 | ||
274bd4fb | 7626 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7627 | release_cluster: |
062c05c4 AO |
7628 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7629 | * set up a new clusters, so lets just skip it | |
7630 | * and let the allocator find whatever block | |
7631 | * it can find. If we reach this point, we | |
7632 | * will have tried the cluster allocator | |
7633 | * plenty of times and not have found | |
7634 | * anything, so we are likely way too | |
7635 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7636 | * anything. |
7637 | * | |
7638 | * However, if the cluster is taken from the | |
7639 | * current block group, release the cluster | |
7640 | * first, so that we stand a better chance of | |
7641 | * succeeding in the unclustered | |
7642 | * allocation. */ | |
7643 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7644 | used_block_group != block_group) { |
062c05c4 | 7645 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7646 | btrfs_release_block_group(used_block_group, |
7647 | delalloc); | |
062c05c4 AO |
7648 | goto unclustered_alloc; |
7649 | } | |
7650 | ||
fa9c0d79 CM |
7651 | /* |
7652 | * this cluster didn't work out, free it and | |
7653 | * start over | |
7654 | */ | |
7655 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7656 | ||
e570fd27 MX |
7657 | if (used_block_group != block_group) |
7658 | btrfs_release_block_group(used_block_group, | |
7659 | delalloc); | |
7660 | refill_cluster: | |
a5f6f719 AO |
7661 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7662 | spin_unlock(&last_ptr->refill_lock); | |
7663 | goto unclustered_alloc; | |
7664 | } | |
7665 | ||
8de972b4 CM |
7666 | aligned_cluster = max_t(unsigned long, |
7667 | empty_cluster + empty_size, | |
7668 | block_group->full_stripe_len); | |
7669 | ||
fa9c0d79 | 7670 | /* allocate a cluster in this block group */ |
00361589 JB |
7671 | ret = btrfs_find_space_cluster(root, block_group, |
7672 | last_ptr, search_start, | |
7673 | num_bytes, | |
7674 | aligned_cluster); | |
fa9c0d79 CM |
7675 | if (ret == 0) { |
7676 | /* | |
7677 | * now pull our allocation out of this | |
7678 | * cluster | |
7679 | */ | |
7680 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7681 | last_ptr, |
7682 | num_bytes, | |
7683 | search_start, | |
7684 | &max_extent_size); | |
fa9c0d79 CM |
7685 | if (offset) { |
7686 | /* we found one, proceed */ | |
7687 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
7688 | trace_btrfs_reserve_extent_cluster(root, |
7689 | block_group, search_start, | |
7690 | num_bytes); | |
fa9c0d79 CM |
7691 | goto checks; |
7692 | } | |
0a24325e JB |
7693 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7694 | && !failed_cluster_refill) { | |
817d52f8 JB |
7695 | spin_unlock(&last_ptr->refill_lock); |
7696 | ||
0a24325e | 7697 | failed_cluster_refill = true; |
817d52f8 JB |
7698 | wait_block_group_cache_progress(block_group, |
7699 | num_bytes + empty_cluster + empty_size); | |
7700 | goto have_block_group; | |
fa9c0d79 | 7701 | } |
817d52f8 | 7702 | |
fa9c0d79 CM |
7703 | /* |
7704 | * at this point we either didn't find a cluster | |
7705 | * or we weren't able to allocate a block from our | |
7706 | * cluster. Free the cluster we've been trying | |
7707 | * to use, and go to the next block group | |
7708 | */ | |
0a24325e | 7709 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7710 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7711 | goto loop; |
fa9c0d79 CM |
7712 | } |
7713 | ||
062c05c4 | 7714 | unclustered_alloc: |
c759c4e1 JB |
7715 | /* |
7716 | * We are doing an unclustered alloc, set the fragmented flag so | |
7717 | * we don't bother trying to setup a cluster again until we get | |
7718 | * more space. | |
7719 | */ | |
7720 | if (unlikely(last_ptr)) { | |
7721 | spin_lock(&last_ptr->lock); | |
7722 | last_ptr->fragmented = 1; | |
7723 | spin_unlock(&last_ptr->lock); | |
7724 | } | |
a5f6f719 AO |
7725 | spin_lock(&block_group->free_space_ctl->tree_lock); |
7726 | if (cached && | |
7727 | block_group->free_space_ctl->free_space < | |
7728 | num_bytes + empty_cluster + empty_size) { | |
a4820398 MX |
7729 | if (block_group->free_space_ctl->free_space > |
7730 | max_extent_size) | |
7731 | max_extent_size = | |
7732 | block_group->free_space_ctl->free_space; | |
a5f6f719 AO |
7733 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
7734 | goto loop; | |
7735 | } | |
7736 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
7737 | ||
6226cb0a | 7738 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7739 | num_bytes, empty_size, |
7740 | &max_extent_size); | |
1cdda9b8 JB |
7741 | /* |
7742 | * If we didn't find a chunk, and we haven't failed on this | |
7743 | * block group before, and this block group is in the middle of | |
7744 | * caching and we are ok with waiting, then go ahead and wait | |
7745 | * for progress to be made, and set failed_alloc to true. | |
7746 | * | |
7747 | * If failed_alloc is true then we've already waited on this | |
7748 | * block group once and should move on to the next block group. | |
7749 | */ | |
7750 | if (!offset && !failed_alloc && !cached && | |
7751 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7752 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7753 | num_bytes + empty_size); |
7754 | failed_alloc = true; | |
817d52f8 | 7755 | goto have_block_group; |
1cdda9b8 JB |
7756 | } else if (!offset) { |
7757 | goto loop; | |
817d52f8 | 7758 | } |
fa9c0d79 | 7759 | checks: |
4e54b17a | 7760 | search_start = ALIGN(offset, root->stripesize); |
25179201 | 7761 | |
2552d17e JB |
7762 | /* move on to the next group */ |
7763 | if (search_start + num_bytes > | |
215a63d1 MX |
7764 | block_group->key.objectid + block_group->key.offset) { |
7765 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7766 | goto loop; |
6226cb0a | 7767 | } |
f5a31e16 | 7768 | |
f0486c68 | 7769 | if (offset < search_start) |
215a63d1 | 7770 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7771 | search_start - offset); |
7772 | BUG_ON(offset > search_start); | |
2552d17e | 7773 | |
18513091 WX |
7774 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7775 | num_bytes, delalloc); | |
f0486c68 | 7776 | if (ret == -EAGAIN) { |
215a63d1 | 7777 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7778 | goto loop; |
0f9dd46c | 7779 | } |
9cfa3e34 | 7780 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7781 | |
f0486c68 | 7782 | /* we are all good, lets return */ |
2552d17e JB |
7783 | ins->objectid = search_start; |
7784 | ins->offset = num_bytes; | |
d2fb3437 | 7785 | |
3f7de037 JB |
7786 | trace_btrfs_reserve_extent(orig_root, block_group, |
7787 | search_start, num_bytes); | |
e570fd27 | 7788 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7789 | break; |
7790 | loop: | |
0a24325e | 7791 | failed_cluster_refill = false; |
1cdda9b8 | 7792 | failed_alloc = false; |
b742bb82 | 7793 | BUG_ON(index != get_block_group_index(block_group)); |
e570fd27 | 7794 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7795 | } |
7796 | up_read(&space_info->groups_sem); | |
7797 | ||
13a0db5a | 7798 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7799 | && !orig_have_caching_bg) | |
7800 | orig_have_caching_bg = true; | |
7801 | ||
60d2adbb MX |
7802 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7803 | goto search; | |
7804 | ||
b742bb82 YZ |
7805 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7806 | goto search; | |
7807 | ||
285ff5af | 7808 | /* |
ccf0e725 JB |
7809 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7810 | * caching kthreads as we move along | |
817d52f8 JB |
7811 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7812 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7813 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7814 | * again | |
fa9c0d79 | 7815 | */ |
723bda20 | 7816 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7817 | index = 0; |
a5e681d9 JB |
7818 | if (loop == LOOP_CACHING_NOWAIT) { |
7819 | /* | |
7820 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7821 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7822 | * full search through. |
7823 | */ | |
13a0db5a | 7824 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7825 | loop = LOOP_CACHING_WAIT; |
7826 | else | |
7827 | loop = LOOP_ALLOC_CHUNK; | |
7828 | } else { | |
7829 | loop++; | |
7830 | } | |
7831 | ||
817d52f8 | 7832 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7833 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7834 | int exist = 0; |
7835 | ||
7836 | trans = current->journal_info; | |
7837 | if (trans) | |
7838 | exist = 1; | |
7839 | else | |
7840 | trans = btrfs_join_transaction(root); | |
00361589 | 7841 | |
00361589 JB |
7842 | if (IS_ERR(trans)) { |
7843 | ret = PTR_ERR(trans); | |
7844 | goto out; | |
7845 | } | |
7846 | ||
b6919a58 | 7847 | ret = do_chunk_alloc(trans, root, flags, |
ea658bad | 7848 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7849 | |
7850 | /* | |
7851 | * If we can't allocate a new chunk we've already looped | |
7852 | * through at least once, move on to the NO_EMPTY_SIZE | |
7853 | * case. | |
7854 | */ | |
7855 | if (ret == -ENOSPC) | |
7856 | loop = LOOP_NO_EMPTY_SIZE; | |
7857 | ||
ea658bad JB |
7858 | /* |
7859 | * Do not bail out on ENOSPC since we | |
7860 | * can do more things. | |
7861 | */ | |
00361589 | 7862 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7863 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7864 | else |
7865 | ret = 0; | |
f017f15f WS |
7866 | if (!exist) |
7867 | btrfs_end_transaction(trans, root); | |
00361589 | 7868 | if (ret) |
ea658bad | 7869 | goto out; |
2552d17e JB |
7870 | } |
7871 | ||
723bda20 | 7872 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7873 | /* |
7874 | * Don't loop again if we already have no empty_size and | |
7875 | * no empty_cluster. | |
7876 | */ | |
7877 | if (empty_size == 0 && | |
7878 | empty_cluster == 0) { | |
7879 | ret = -ENOSPC; | |
7880 | goto out; | |
7881 | } | |
723bda20 JB |
7882 | empty_size = 0; |
7883 | empty_cluster = 0; | |
fa9c0d79 | 7884 | } |
723bda20 JB |
7885 | |
7886 | goto search; | |
2552d17e JB |
7887 | } else if (!ins->objectid) { |
7888 | ret = -ENOSPC; | |
d82a6f1d | 7889 | } else if (ins->objectid) { |
c759c4e1 JB |
7890 | if (!use_cluster && last_ptr) { |
7891 | spin_lock(&last_ptr->lock); | |
7892 | last_ptr->window_start = ins->objectid; | |
7893 | spin_unlock(&last_ptr->lock); | |
7894 | } | |
80eb234a | 7895 | ret = 0; |
be744175 | 7896 | } |
79787eaa | 7897 | out: |
4f4db217 JB |
7898 | if (ret == -ENOSPC) { |
7899 | spin_lock(&space_info->lock); | |
7900 | space_info->max_extent_size = max_extent_size; | |
7901 | spin_unlock(&space_info->lock); | |
a4820398 | 7902 | ins->offset = max_extent_size; |
4f4db217 | 7903 | } |
0f70abe2 | 7904 | return ret; |
fec577fb | 7905 | } |
ec44a35c | 7906 | |
ab8d0fc4 JM |
7907 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7908 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7909 | int dump_block_groups) |
0f9dd46c JB |
7910 | { |
7911 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7912 | int index = 0; |
0f9dd46c | 7913 | |
9ed74f2d | 7914 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7915 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7916 | info->flags, | |
7917 | info->total_bytes - info->bytes_used - info->bytes_pinned - | |
7918 | info->bytes_reserved - info->bytes_readonly - | |
7919 | info->bytes_may_use, (info->full) ? "" : "not "); | |
7920 | btrfs_info(fs_info, | |
7921 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7922 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7923 | info->bytes_reserved, info->bytes_may_use, | |
7924 | info->bytes_readonly); | |
9ed74f2d JB |
7925 | spin_unlock(&info->lock); |
7926 | ||
7927 | if (!dump_block_groups) | |
7928 | return; | |
0f9dd46c | 7929 | |
80eb234a | 7930 | down_read(&info->groups_sem); |
b742bb82 YZ |
7931 | again: |
7932 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7933 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7934 | btrfs_info(fs_info, |
7935 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7936 | cache->key.objectid, cache->key.offset, | |
7937 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7938 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7939 | btrfs_dump_free_space(cache, bytes); |
7940 | spin_unlock(&cache->lock); | |
7941 | } | |
b742bb82 YZ |
7942 | if (++index < BTRFS_NR_RAID_TYPES) |
7943 | goto again; | |
80eb234a | 7944 | up_read(&info->groups_sem); |
0f9dd46c | 7945 | } |
e8569813 | 7946 | |
18513091 | 7947 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
7948 | u64 num_bytes, u64 min_alloc_size, |
7949 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7950 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7951 | { |
ab8d0fc4 | 7952 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 7953 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7954 | u64 flags; |
fec577fb | 7955 | int ret; |
925baedd | 7956 | |
b6919a58 | 7957 | flags = btrfs_get_alloc_profile(root, is_data); |
98d20f67 | 7958 | again: |
db94535d | 7959 | WARN_ON(num_bytes < root->sectorsize); |
18513091 WX |
7960 | ret = find_free_extent(root, ram_bytes, num_bytes, empty_size, |
7961 | hint_byte, ins, flags, delalloc); | |
9cfa3e34 | 7962 | if (!ret && !is_data) { |
ab8d0fc4 | 7963 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 7964 | } else if (ret == -ENOSPC) { |
a4820398 MX |
7965 | if (!final_tried && ins->offset) { |
7966 | num_bytes = min(num_bytes >> 1, ins->offset); | |
24542bf7 | 7967 | num_bytes = round_down(num_bytes, root->sectorsize); |
9e622d6b | 7968 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 7969 | ram_bytes = num_bytes; |
9e622d6b MX |
7970 | if (num_bytes == min_alloc_size) |
7971 | final_tried = true; | |
7972 | goto again; | |
ab8d0fc4 | 7973 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
7974 | struct btrfs_space_info *sinfo; |
7975 | ||
ab8d0fc4 | 7976 | sinfo = __find_space_info(fs_info, flags); |
5d163e0e JM |
7977 | btrfs_err(root->fs_info, |
7978 | "allocation failed flags %llu, wanted %llu", | |
7979 | flags, num_bytes); | |
53804280 | 7980 | if (sinfo) |
ab8d0fc4 | 7981 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 7982 | } |
925baedd | 7983 | } |
0f9dd46c JB |
7984 | |
7985 | return ret; | |
e6dcd2dc CM |
7986 | } |
7987 | ||
e688b725 | 7988 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 MX |
7989 | u64 start, u64 len, |
7990 | int pin, int delalloc) | |
65b51a00 | 7991 | { |
0f9dd46c | 7992 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 7993 | int ret = 0; |
0f9dd46c | 7994 | |
0f9dd46c JB |
7995 | cache = btrfs_lookup_block_group(root->fs_info, start); |
7996 | if (!cache) { | |
c2cf52eb | 7997 | btrfs_err(root->fs_info, "Unable to find block group for %llu", |
c1c9ff7c | 7998 | start); |
0f9dd46c JB |
7999 | return -ENOSPC; |
8000 | } | |
1f3c79a2 | 8001 | |
e688b725 CM |
8002 | if (pin) |
8003 | pin_down_extent(root, cache, start, len, 1); | |
8004 | else { | |
3cdde224 | 8005 | if (btrfs_test_opt(root->fs_info, DISCARD)) |
dcc82f47 | 8006 | ret = btrfs_discard_extent(root, start, len, NULL); |
e688b725 | 8007 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8008 | btrfs_free_reserved_bytes(cache, len, delalloc); |
31bada7c | 8009 | trace_btrfs_reserved_extent_free(root, start, len); |
e688b725 | 8010 | } |
31193213 | 8011 | |
fa9c0d79 | 8012 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8013 | return ret; |
8014 | } | |
8015 | ||
e688b725 | 8016 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 | 8017 | u64 start, u64 len, int delalloc) |
e688b725 | 8018 | { |
e570fd27 | 8019 | return __btrfs_free_reserved_extent(root, start, len, 0, delalloc); |
e688b725 CM |
8020 | } |
8021 | ||
8022 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
8023 | u64 start, u64 len) | |
8024 | { | |
e570fd27 | 8025 | return __btrfs_free_reserved_extent(root, start, len, 1, 0); |
e688b725 CM |
8026 | } |
8027 | ||
5d4f98a2 YZ |
8028 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
8029 | struct btrfs_root *root, | |
8030 | u64 parent, u64 root_objectid, | |
8031 | u64 flags, u64 owner, u64 offset, | |
8032 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
8033 | { |
8034 | int ret; | |
5d4f98a2 | 8035 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 8036 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8037 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8038 | struct btrfs_path *path; |
5d4f98a2 YZ |
8039 | struct extent_buffer *leaf; |
8040 | int type; | |
8041 | u32 size; | |
26b8003f | 8042 | |
5d4f98a2 YZ |
8043 | if (parent > 0) |
8044 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8045 | else | |
8046 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8047 | |
5d4f98a2 | 8048 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8049 | |
8050 | path = btrfs_alloc_path(); | |
db5b493a TI |
8051 | if (!path) |
8052 | return -ENOMEM; | |
47e4bb98 | 8053 | |
b9473439 | 8054 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8055 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8056 | ins, size); | |
79787eaa JM |
8057 | if (ret) { |
8058 | btrfs_free_path(path); | |
8059 | return ret; | |
8060 | } | |
0f9dd46c | 8061 | |
5d4f98a2 YZ |
8062 | leaf = path->nodes[0]; |
8063 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8064 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8065 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8066 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8067 | btrfs_set_extent_flags(leaf, extent_item, | |
8068 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8069 | ||
8070 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8071 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8072 | if (parent > 0) { | |
8073 | struct btrfs_shared_data_ref *ref; | |
8074 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8075 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8076 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8077 | } else { | |
8078 | struct btrfs_extent_data_ref *ref; | |
8079 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8080 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8081 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8082 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8083 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8084 | } | |
47e4bb98 CM |
8085 | |
8086 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8087 | btrfs_free_path(path); |
f510cfec | 8088 | |
1e144fb8 OS |
8089 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8090 | ins->offset); | |
8091 | if (ret) | |
8092 | return ret; | |
8093 | ||
ce93ec54 | 8094 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 8095 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8096 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8097 | ins->objectid, ins->offset); |
f5947066 CM |
8098 | BUG(); |
8099 | } | |
0be5dc67 | 8100 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
e6dcd2dc CM |
8101 | return ret; |
8102 | } | |
8103 | ||
5d4f98a2 YZ |
8104 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
8105 | struct btrfs_root *root, | |
8106 | u64 parent, u64 root_objectid, | |
8107 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 8108 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
8109 | { |
8110 | int ret; | |
5d4f98a2 YZ |
8111 | struct btrfs_fs_info *fs_info = root->fs_info; |
8112 | struct btrfs_extent_item *extent_item; | |
8113 | struct btrfs_tree_block_info *block_info; | |
8114 | struct btrfs_extent_inline_ref *iref; | |
8115 | struct btrfs_path *path; | |
8116 | struct extent_buffer *leaf; | |
3173a18f | 8117 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 8118 | u64 num_bytes = ins->offset; |
3173a18f JB |
8119 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
8120 | SKINNY_METADATA); | |
8121 | ||
8122 | if (!skinny_metadata) | |
8123 | size += sizeof(*block_info); | |
1c2308f8 | 8124 | |
5d4f98a2 | 8125 | path = btrfs_alloc_path(); |
857cc2fc JB |
8126 | if (!path) { |
8127 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, | |
707e8a07 | 8128 | root->nodesize); |
d8926bb3 | 8129 | return -ENOMEM; |
857cc2fc | 8130 | } |
56bec294 | 8131 | |
5d4f98a2 YZ |
8132 | path->leave_spinning = 1; |
8133 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
8134 | ins, size); | |
79787eaa | 8135 | if (ret) { |
dd825259 | 8136 | btrfs_free_path(path); |
857cc2fc | 8137 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, |
707e8a07 | 8138 | root->nodesize); |
79787eaa JM |
8139 | return ret; |
8140 | } | |
5d4f98a2 YZ |
8141 | |
8142 | leaf = path->nodes[0]; | |
8143 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8144 | struct btrfs_extent_item); | |
8145 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8146 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8147 | btrfs_set_extent_flags(leaf, extent_item, | |
8148 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8149 | |
3173a18f JB |
8150 | if (skinny_metadata) { |
8151 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
707e8a07 | 8152 | num_bytes = root->nodesize; |
3173a18f JB |
8153 | } else { |
8154 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
8155 | btrfs_set_tree_block_key(leaf, block_info, key); | |
8156 | btrfs_set_tree_block_level(leaf, block_info, level); | |
8157 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
8158 | } | |
5d4f98a2 | 8159 | |
5d4f98a2 YZ |
8160 | if (parent > 0) { |
8161 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
8162 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8163 | BTRFS_SHARED_BLOCK_REF_KEY); | |
8164 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8165 | } else { | |
8166 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8167 | BTRFS_TREE_BLOCK_REF_KEY); | |
8168 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
8169 | } | |
8170 | ||
8171 | btrfs_mark_buffer_dirty(leaf); | |
8172 | btrfs_free_path(path); | |
8173 | ||
1e144fb8 OS |
8174 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8175 | num_bytes); | |
8176 | if (ret) | |
8177 | return ret; | |
8178 | ||
ce93ec54 JB |
8179 | ret = update_block_group(trans, root, ins->objectid, root->nodesize, |
8180 | 1); | |
79787eaa | 8181 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8182 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8183 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
8184 | BUG(); |
8185 | } | |
0be5dc67 | 8186 | |
707e8a07 | 8187 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize); |
5d4f98a2 YZ |
8188 | return ret; |
8189 | } | |
8190 | ||
8191 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
8192 | struct btrfs_root *root, | |
8193 | u64 root_objectid, u64 owner, | |
5846a3c2 QW |
8194 | u64 offset, u64 ram_bytes, |
8195 | struct btrfs_key *ins) | |
5d4f98a2 YZ |
8196 | { |
8197 | int ret; | |
8198 | ||
8199 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
8200 | ||
66d7e7f0 AJ |
8201 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
8202 | ins->offset, 0, | |
8203 | root_objectid, owner, offset, | |
5846a3c2 QW |
8204 | ram_bytes, BTRFS_ADD_DELAYED_EXTENT, |
8205 | NULL); | |
e6dcd2dc CM |
8206 | return ret; |
8207 | } | |
e02119d5 CM |
8208 | |
8209 | /* | |
8210 | * this is used by the tree logging recovery code. It records that | |
8211 | * an extent has been allocated and makes sure to clear the free | |
8212 | * space cache bits as well | |
8213 | */ | |
5d4f98a2 YZ |
8214 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
8215 | struct btrfs_root *root, | |
8216 | u64 root_objectid, u64 owner, u64 offset, | |
8217 | struct btrfs_key *ins) | |
e02119d5 CM |
8218 | { |
8219 | int ret; | |
8220 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8221 | struct btrfs_space_info *space_info; |
11833d66 | 8222 | |
8c2a1a30 JB |
8223 | /* |
8224 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8225 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 JB |
8226 | */ |
8227 | if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) { | |
8228 | ret = __exclude_logged_extent(root, ins->objectid, ins->offset); | |
b50c6e25 | 8229 | if (ret) |
8c2a1a30 | 8230 | return ret; |
11833d66 YZ |
8231 | } |
8232 | ||
8c2a1a30 JB |
8233 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
8234 | if (!block_group) | |
8235 | return -EINVAL; | |
8236 | ||
ed7a6948 WX |
8237 | space_info = block_group->space_info; |
8238 | spin_lock(&space_info->lock); | |
8239 | spin_lock(&block_group->lock); | |
8240 | space_info->bytes_reserved += ins->offset; | |
8241 | block_group->reserved += ins->offset; | |
8242 | spin_unlock(&block_group->lock); | |
8243 | spin_unlock(&space_info->lock); | |
8244 | ||
5d4f98a2 YZ |
8245 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
8246 | 0, owner, offset, ins, 1); | |
b50c6e25 | 8247 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8248 | return ret; |
8249 | } | |
8250 | ||
48a3b636 ES |
8251 | static struct extent_buffer * |
8252 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8253 | u64 bytenr, int level) |
65b51a00 CM |
8254 | { |
8255 | struct extent_buffer *buf; | |
8256 | ||
a83fffb7 | 8257 | buf = btrfs_find_create_tree_block(root, bytenr); |
c871b0f2 LB |
8258 | if (IS_ERR(buf)) |
8259 | return buf; | |
8260 | ||
65b51a00 | 8261 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8262 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8263 | btrfs_tree_lock(buf); |
01d58472 | 8264 | clean_tree_block(trans, root->fs_info, buf); |
3083ee2e | 8265 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8266 | |
8267 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8268 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8269 | |
d0c803c4 | 8270 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8271 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8272 | /* |
8273 | * we allow two log transactions at a time, use different | |
8274 | * EXENT bit to differentiate dirty pages. | |
8275 | */ | |
656f30db | 8276 | if (buf->log_index == 0) |
8cef4e16 YZ |
8277 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8278 | buf->start + buf->len - 1, GFP_NOFS); | |
8279 | else | |
8280 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8281 | buf->start + buf->len - 1); |
d0c803c4 | 8282 | } else { |
656f30db | 8283 | buf->log_index = -1; |
d0c803c4 | 8284 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8285 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8286 | } |
64c12921 | 8287 | trans->dirty = true; |
b4ce94de | 8288 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8289 | return buf; |
8290 | } | |
8291 | ||
f0486c68 YZ |
8292 | static struct btrfs_block_rsv * |
8293 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8294 | struct btrfs_root *root, u32 blocksize) | |
8295 | { | |
8296 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 8297 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 | 8298 | int ret; |
d88033db | 8299 | bool global_updated = false; |
f0486c68 YZ |
8300 | |
8301 | block_rsv = get_block_rsv(trans, root); | |
8302 | ||
b586b323 MX |
8303 | if (unlikely(block_rsv->size == 0)) |
8304 | goto try_reserve; | |
d88033db | 8305 | again: |
f0486c68 YZ |
8306 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8307 | if (!ret) | |
8308 | return block_rsv; | |
8309 | ||
b586b323 MX |
8310 | if (block_rsv->failfast) |
8311 | return ERR_PTR(ret); | |
8312 | ||
d88033db MX |
8313 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8314 | global_updated = true; | |
8315 | update_global_block_rsv(root->fs_info); | |
8316 | goto again; | |
8317 | } | |
8318 | ||
3cdde224 | 8319 | if (btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8320 | static DEFINE_RATELIMIT_STATE(_rs, |
8321 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8322 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8323 | if (__ratelimit(&_rs)) | |
8324 | WARN(1, KERN_DEBUG | |
efe120a0 | 8325 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8326 | } |
8327 | try_reserve: | |
8328 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8329 | BTRFS_RESERVE_NO_FLUSH); | |
8330 | if (!ret) | |
8331 | return block_rsv; | |
8332 | /* | |
8333 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8334 | * the global reserve if its space type is the same as the global |
8335 | * reservation. | |
b586b323 | 8336 | */ |
5881cfc9 MX |
8337 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8338 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8339 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8340 | if (!ret) | |
8341 | return global_rsv; | |
8342 | } | |
8343 | return ERR_PTR(ret); | |
f0486c68 YZ |
8344 | } |
8345 | ||
8c2a3ca2 JB |
8346 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8347 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8348 | { |
8349 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 8350 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
8351 | } |
8352 | ||
fec577fb | 8353 | /* |
f0486c68 | 8354 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8355 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8356 | */ |
4d75f8a9 DS |
8357 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
8358 | struct btrfs_root *root, | |
5d4f98a2 YZ |
8359 | u64 parent, u64 root_objectid, |
8360 | struct btrfs_disk_key *key, int level, | |
5581a51a | 8361 | u64 hint, u64 empty_size) |
fec577fb | 8362 | { |
e2fa7227 | 8363 | struct btrfs_key ins; |
f0486c68 | 8364 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8365 | struct extent_buffer *buf; |
67b7859e | 8366 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8367 | u64 flags = 0; |
8368 | int ret; | |
4d75f8a9 | 8369 | u32 blocksize = root->nodesize; |
3173a18f JB |
8370 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
8371 | SKINNY_METADATA); | |
fec577fb | 8372 | |
05653ef3 | 8373 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
f5ee5c9a | 8374 | if (btrfs_is_testing(root->fs_info)) { |
faa2dbf0 | 8375 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8376 | level); |
faa2dbf0 JB |
8377 | if (!IS_ERR(buf)) |
8378 | root->alloc_bytenr += blocksize; | |
8379 | return buf; | |
8380 | } | |
05653ef3 | 8381 | #endif |
fccb84c9 | 8382 | |
f0486c68 YZ |
8383 | block_rsv = use_block_rsv(trans, root, blocksize); |
8384 | if (IS_ERR(block_rsv)) | |
8385 | return ERR_CAST(block_rsv); | |
8386 | ||
18513091 | 8387 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8388 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8389 | if (ret) |
8390 | goto out_unuse; | |
55c69072 | 8391 | |
fe864576 | 8392 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8393 | if (IS_ERR(buf)) { |
8394 | ret = PTR_ERR(buf); | |
8395 | goto out_free_reserved; | |
8396 | } | |
f0486c68 YZ |
8397 | |
8398 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8399 | if (parent == 0) | |
8400 | parent = ins.objectid; | |
8401 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8402 | } else | |
8403 | BUG_ON(parent > 0); | |
8404 | ||
8405 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8406 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8407 | if (!extent_op) { |
8408 | ret = -ENOMEM; | |
8409 | goto out_free_buf; | |
8410 | } | |
f0486c68 YZ |
8411 | if (key) |
8412 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8413 | else | |
8414 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8415 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8416 | extent_op->update_key = skinny_metadata ? false : true; |
8417 | extent_op->update_flags = true; | |
8418 | extent_op->is_data = false; | |
b1c79e09 | 8419 | extent_op->level = level; |
f0486c68 | 8420 | |
66d7e7f0 | 8421 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
67b7859e OS |
8422 | ins.objectid, ins.offset, |
8423 | parent, root_objectid, level, | |
8424 | BTRFS_ADD_DELAYED_EXTENT, | |
b06c4bf5 | 8425 | extent_op); |
67b7859e OS |
8426 | if (ret) |
8427 | goto out_free_delayed; | |
f0486c68 | 8428 | } |
fec577fb | 8429 | return buf; |
67b7859e OS |
8430 | |
8431 | out_free_delayed: | |
8432 | btrfs_free_delayed_extent_op(extent_op); | |
8433 | out_free_buf: | |
8434 | free_extent_buffer(buf); | |
8435 | out_free_reserved: | |
8436 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0); | |
8437 | out_unuse: | |
8438 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); | |
8439 | return ERR_PTR(ret); | |
fec577fb | 8440 | } |
a28ec197 | 8441 | |
2c47e605 YZ |
8442 | struct walk_control { |
8443 | u64 refs[BTRFS_MAX_LEVEL]; | |
8444 | u64 flags[BTRFS_MAX_LEVEL]; | |
8445 | struct btrfs_key update_progress; | |
8446 | int stage; | |
8447 | int level; | |
8448 | int shared_level; | |
8449 | int update_ref; | |
8450 | int keep_locks; | |
1c4850e2 YZ |
8451 | int reada_slot; |
8452 | int reada_count; | |
66d7e7f0 | 8453 | int for_reloc; |
2c47e605 YZ |
8454 | }; |
8455 | ||
8456 | #define DROP_REFERENCE 1 | |
8457 | #define UPDATE_BACKREF 2 | |
8458 | ||
1c4850e2 YZ |
8459 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8460 | struct btrfs_root *root, | |
8461 | struct walk_control *wc, | |
8462 | struct btrfs_path *path) | |
6407bf6d | 8463 | { |
1c4850e2 YZ |
8464 | u64 bytenr; |
8465 | u64 generation; | |
8466 | u64 refs; | |
94fcca9f | 8467 | u64 flags; |
5d4f98a2 | 8468 | u32 nritems; |
1c4850e2 YZ |
8469 | struct btrfs_key key; |
8470 | struct extent_buffer *eb; | |
6407bf6d | 8471 | int ret; |
1c4850e2 YZ |
8472 | int slot; |
8473 | int nread = 0; | |
6407bf6d | 8474 | |
1c4850e2 YZ |
8475 | if (path->slots[wc->level] < wc->reada_slot) { |
8476 | wc->reada_count = wc->reada_count * 2 / 3; | |
8477 | wc->reada_count = max(wc->reada_count, 2); | |
8478 | } else { | |
8479 | wc->reada_count = wc->reada_count * 3 / 2; | |
8480 | wc->reada_count = min_t(int, wc->reada_count, | |
8481 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
8482 | } | |
7bb86316 | 8483 | |
1c4850e2 YZ |
8484 | eb = path->nodes[wc->level]; |
8485 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8486 | |
1c4850e2 YZ |
8487 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8488 | if (nread >= wc->reada_count) | |
8489 | break; | |
bd56b302 | 8490 | |
2dd3e67b | 8491 | cond_resched(); |
1c4850e2 YZ |
8492 | bytenr = btrfs_node_blockptr(eb, slot); |
8493 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8494 | |
1c4850e2 YZ |
8495 | if (slot == path->slots[wc->level]) |
8496 | goto reada; | |
5d4f98a2 | 8497 | |
1c4850e2 YZ |
8498 | if (wc->stage == UPDATE_BACKREF && |
8499 | generation <= root->root_key.offset) | |
bd56b302 CM |
8500 | continue; |
8501 | ||
94fcca9f | 8502 | /* We don't lock the tree block, it's OK to be racy here */ |
3173a18f JB |
8503 | ret = btrfs_lookup_extent_info(trans, root, bytenr, |
8504 | wc->level - 1, 1, &refs, | |
8505 | &flags); | |
79787eaa JM |
8506 | /* We don't care about errors in readahead. */ |
8507 | if (ret < 0) | |
8508 | continue; | |
94fcca9f YZ |
8509 | BUG_ON(refs == 0); |
8510 | ||
1c4850e2 | 8511 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8512 | if (refs == 1) |
8513 | goto reada; | |
bd56b302 | 8514 | |
94fcca9f YZ |
8515 | if (wc->level == 1 && |
8516 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8517 | continue; | |
1c4850e2 YZ |
8518 | if (!wc->update_ref || |
8519 | generation <= root->root_key.offset) | |
8520 | continue; | |
8521 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8522 | ret = btrfs_comp_cpu_keys(&key, | |
8523 | &wc->update_progress); | |
8524 | if (ret < 0) | |
8525 | continue; | |
94fcca9f YZ |
8526 | } else { |
8527 | if (wc->level == 1 && | |
8528 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8529 | continue; | |
6407bf6d | 8530 | } |
1c4850e2 | 8531 | reada: |
d3e46fea | 8532 | readahead_tree_block(root, bytenr); |
1c4850e2 | 8533 | nread++; |
20524f02 | 8534 | } |
1c4850e2 | 8535 | wc->reada_slot = slot; |
20524f02 | 8536 | } |
2c47e605 | 8537 | |
1152651a MF |
8538 | static int account_leaf_items(struct btrfs_trans_handle *trans, |
8539 | struct btrfs_root *root, | |
8540 | struct extent_buffer *eb) | |
8541 | { | |
8542 | int nr = btrfs_header_nritems(eb); | |
82bd101b | 8543 | int i, extent_type, ret; |
1152651a MF |
8544 | struct btrfs_key key; |
8545 | struct btrfs_file_extent_item *fi; | |
8546 | u64 bytenr, num_bytes; | |
8547 | ||
82bd101b | 8548 | /* We can be called directly from walk_up_proc() */ |
afcdd129 | 8549 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags)) |
82bd101b MF |
8550 | return 0; |
8551 | ||
1152651a MF |
8552 | for (i = 0; i < nr; i++) { |
8553 | btrfs_item_key_to_cpu(eb, &key, i); | |
8554 | ||
8555 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
8556 | continue; | |
8557 | ||
8558 | fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
8559 | /* filter out non qgroup-accountable extents */ | |
8560 | extent_type = btrfs_file_extent_type(eb, fi); | |
8561 | ||
8562 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
8563 | continue; | |
8564 | ||
8565 | bytenr = btrfs_file_extent_disk_bytenr(eb, fi); | |
8566 | if (!bytenr) | |
8567 | continue; | |
8568 | ||
8569 | num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi); | |
82bd101b | 8570 | |
cb93b52c QW |
8571 | ret = btrfs_qgroup_insert_dirty_extent(trans, root->fs_info, |
8572 | bytenr, num_bytes, GFP_NOFS); | |
82bd101b MF |
8573 | if (ret) |
8574 | return ret; | |
1152651a MF |
8575 | } |
8576 | return 0; | |
8577 | } | |
8578 | ||
8579 | /* | |
8580 | * Walk up the tree from the bottom, freeing leaves and any interior | |
8581 | * nodes which have had all slots visited. If a node (leaf or | |
8582 | * interior) is freed, the node above it will have it's slot | |
8583 | * incremented. The root node will never be freed. | |
8584 | * | |
8585 | * At the end of this function, we should have a path which has all | |
8586 | * slots incremented to the next position for a search. If we need to | |
8587 | * read a new node it will be NULL and the node above it will have the | |
8588 | * correct slot selected for a later read. | |
8589 | * | |
8590 | * If we increment the root nodes slot counter past the number of | |
8591 | * elements, 1 is returned to signal completion of the search. | |
8592 | */ | |
8593 | static int adjust_slots_upwards(struct btrfs_root *root, | |
8594 | struct btrfs_path *path, int root_level) | |
8595 | { | |
8596 | int level = 0; | |
8597 | int nr, slot; | |
8598 | struct extent_buffer *eb; | |
8599 | ||
8600 | if (root_level == 0) | |
8601 | return 1; | |
8602 | ||
8603 | while (level <= root_level) { | |
8604 | eb = path->nodes[level]; | |
8605 | nr = btrfs_header_nritems(eb); | |
8606 | path->slots[level]++; | |
8607 | slot = path->slots[level]; | |
8608 | if (slot >= nr || level == 0) { | |
8609 | /* | |
8610 | * Don't free the root - we will detect this | |
8611 | * condition after our loop and return a | |
8612 | * positive value for caller to stop walking the tree. | |
8613 | */ | |
8614 | if (level != root_level) { | |
8615 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
8616 | path->locks[level] = 0; | |
8617 | ||
8618 | free_extent_buffer(eb); | |
8619 | path->nodes[level] = NULL; | |
8620 | path->slots[level] = 0; | |
8621 | } | |
8622 | } else { | |
8623 | /* | |
8624 | * We have a valid slot to walk back down | |
8625 | * from. Stop here so caller can process these | |
8626 | * new nodes. | |
8627 | */ | |
8628 | break; | |
8629 | } | |
8630 | ||
8631 | level++; | |
8632 | } | |
8633 | ||
8634 | eb = path->nodes[root_level]; | |
8635 | if (path->slots[root_level] >= btrfs_header_nritems(eb)) | |
8636 | return 1; | |
8637 | ||
8638 | return 0; | |
8639 | } | |
8640 | ||
8641 | /* | |
8642 | * root_eb is the subtree root and is locked before this function is called. | |
8643 | */ | |
8644 | static int account_shared_subtree(struct btrfs_trans_handle *trans, | |
8645 | struct btrfs_root *root, | |
8646 | struct extent_buffer *root_eb, | |
8647 | u64 root_gen, | |
8648 | int root_level) | |
8649 | { | |
8650 | int ret = 0; | |
8651 | int level; | |
8652 | struct extent_buffer *eb = root_eb; | |
8653 | struct btrfs_path *path = NULL; | |
8654 | ||
8655 | BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL); | |
8656 | BUG_ON(root_eb == NULL); | |
8657 | ||
afcdd129 | 8658 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags)) |
1152651a MF |
8659 | return 0; |
8660 | ||
8661 | if (!extent_buffer_uptodate(root_eb)) { | |
8662 | ret = btrfs_read_buffer(root_eb, root_gen); | |
8663 | if (ret) | |
8664 | goto out; | |
8665 | } | |
8666 | ||
8667 | if (root_level == 0) { | |
8668 | ret = account_leaf_items(trans, root, root_eb); | |
8669 | goto out; | |
8670 | } | |
8671 | ||
8672 | path = btrfs_alloc_path(); | |
8673 | if (!path) | |
8674 | return -ENOMEM; | |
8675 | ||
8676 | /* | |
8677 | * Walk down the tree. Missing extent blocks are filled in as | |
8678 | * we go. Metadata is accounted every time we read a new | |
8679 | * extent block. | |
8680 | * | |
8681 | * When we reach a leaf, we account for file extent items in it, | |
8682 | * walk back up the tree (adjusting slot pointers as we go) | |
8683 | * and restart the search process. | |
8684 | */ | |
8685 | extent_buffer_get(root_eb); /* For path */ | |
8686 | path->nodes[root_level] = root_eb; | |
8687 | path->slots[root_level] = 0; | |
8688 | path->locks[root_level] = 0; /* so release_path doesn't try to unlock */ | |
8689 | walk_down: | |
8690 | level = root_level; | |
8691 | while (level >= 0) { | |
8692 | if (path->nodes[level] == NULL) { | |
1152651a MF |
8693 | int parent_slot; |
8694 | u64 child_gen; | |
8695 | u64 child_bytenr; | |
8696 | ||
8697 | /* We need to get child blockptr/gen from | |
8698 | * parent before we can read it. */ | |
8699 | eb = path->nodes[level + 1]; | |
8700 | parent_slot = path->slots[level + 1]; | |
8701 | child_bytenr = btrfs_node_blockptr(eb, parent_slot); | |
8702 | child_gen = btrfs_node_ptr_generation(eb, parent_slot); | |
8703 | ||
ce86cd59 | 8704 | eb = read_tree_block(root, child_bytenr, child_gen); |
64c043de LB |
8705 | if (IS_ERR(eb)) { |
8706 | ret = PTR_ERR(eb); | |
8707 | goto out; | |
8708 | } else if (!extent_buffer_uptodate(eb)) { | |
8635eda9 | 8709 | free_extent_buffer(eb); |
64c043de | 8710 | ret = -EIO; |
1152651a MF |
8711 | goto out; |
8712 | } | |
8713 | ||
8714 | path->nodes[level] = eb; | |
8715 | path->slots[level] = 0; | |
8716 | ||
8717 | btrfs_tree_read_lock(eb); | |
8718 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
8719 | path->locks[level] = BTRFS_READ_LOCK_BLOCKING; | |
82bd101b | 8720 | |
cb93b52c QW |
8721 | ret = btrfs_qgroup_insert_dirty_extent(trans, |
8722 | root->fs_info, child_bytenr, | |
8723 | root->nodesize, GFP_NOFS); | |
82bd101b MF |
8724 | if (ret) |
8725 | goto out; | |
1152651a MF |
8726 | } |
8727 | ||
8728 | if (level == 0) { | |
8729 | ret = account_leaf_items(trans, root, path->nodes[level]); | |
8730 | if (ret) | |
8731 | goto out; | |
8732 | ||
8733 | /* Nonzero return here means we completed our search */ | |
8734 | ret = adjust_slots_upwards(root, path, root_level); | |
8735 | if (ret) | |
8736 | break; | |
8737 | ||
8738 | /* Restart search with new slots */ | |
8739 | goto walk_down; | |
8740 | } | |
8741 | ||
8742 | level--; | |
8743 | } | |
8744 | ||
8745 | ret = 0; | |
8746 | out: | |
8747 | btrfs_free_path(path); | |
8748 | ||
8749 | return ret; | |
8750 | } | |
8751 | ||
f82d02d9 | 8752 | /* |
2c016dc2 | 8753 | * helper to process tree block while walking down the tree. |
2c47e605 | 8754 | * |
2c47e605 YZ |
8755 | * when wc->stage == UPDATE_BACKREF, this function updates |
8756 | * back refs for pointers in the block. | |
8757 | * | |
8758 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8759 | */ |
2c47e605 | 8760 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8761 | struct btrfs_root *root, |
2c47e605 | 8762 | struct btrfs_path *path, |
94fcca9f | 8763 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8764 | { |
2c47e605 YZ |
8765 | int level = wc->level; |
8766 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8767 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8768 | int ret; |
8769 | ||
2c47e605 YZ |
8770 | if (wc->stage == UPDATE_BACKREF && |
8771 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8772 | return 1; | |
f82d02d9 | 8773 | |
2c47e605 YZ |
8774 | /* |
8775 | * when reference count of tree block is 1, it won't increase | |
8776 | * again. once full backref flag is set, we never clear it. | |
8777 | */ | |
94fcca9f YZ |
8778 | if (lookup_info && |
8779 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8780 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
8781 | BUG_ON(!path->locks[level]); |
8782 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 8783 | eb->start, level, 1, |
2c47e605 YZ |
8784 | &wc->refs[level], |
8785 | &wc->flags[level]); | |
79787eaa JM |
8786 | BUG_ON(ret == -ENOMEM); |
8787 | if (ret) | |
8788 | return ret; | |
2c47e605 YZ |
8789 | BUG_ON(wc->refs[level] == 0); |
8790 | } | |
5d4f98a2 | 8791 | |
2c47e605 YZ |
8792 | if (wc->stage == DROP_REFERENCE) { |
8793 | if (wc->refs[level] > 1) | |
8794 | return 1; | |
f82d02d9 | 8795 | |
2c47e605 | 8796 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8797 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8798 | path->locks[level] = 0; |
8799 | } | |
8800 | return 0; | |
8801 | } | |
f82d02d9 | 8802 | |
2c47e605 YZ |
8803 | /* wc->stage == UPDATE_BACKREF */ |
8804 | if (!(wc->flags[level] & flag)) { | |
8805 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8806 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8807 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8808 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8809 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 | 8810 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
b1c79e09 JB |
8811 | eb->len, flag, |
8812 | btrfs_header_level(eb), 0); | |
79787eaa | 8813 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8814 | wc->flags[level] |= flag; |
8815 | } | |
8816 | ||
8817 | /* | |
8818 | * the block is shared by multiple trees, so it's not good to | |
8819 | * keep the tree lock | |
8820 | */ | |
8821 | if (path->locks[level] && level > 0) { | |
bd681513 | 8822 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8823 | path->locks[level] = 0; |
8824 | } | |
8825 | return 0; | |
8826 | } | |
8827 | ||
1c4850e2 | 8828 | /* |
2c016dc2 | 8829 | * helper to process tree block pointer. |
1c4850e2 YZ |
8830 | * |
8831 | * when wc->stage == DROP_REFERENCE, this function checks | |
8832 | * reference count of the block pointed to. if the block | |
8833 | * is shared and we need update back refs for the subtree | |
8834 | * rooted at the block, this function changes wc->stage to | |
8835 | * UPDATE_BACKREF. if the block is shared and there is no | |
8836 | * need to update back, this function drops the reference | |
8837 | * to the block. | |
8838 | * | |
8839 | * NOTE: return value 1 means we should stop walking down. | |
8840 | */ | |
8841 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8842 | struct btrfs_root *root, | |
8843 | struct btrfs_path *path, | |
94fcca9f | 8844 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
8845 | { |
8846 | u64 bytenr; | |
8847 | u64 generation; | |
8848 | u64 parent; | |
8849 | u32 blocksize; | |
8850 | struct btrfs_key key; | |
8851 | struct extent_buffer *next; | |
8852 | int level = wc->level; | |
8853 | int reada = 0; | |
8854 | int ret = 0; | |
1152651a | 8855 | bool need_account = false; |
1c4850e2 YZ |
8856 | |
8857 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8858 | path->slots[level]); | |
8859 | /* | |
8860 | * if the lower level block was created before the snapshot | |
8861 | * was created, we know there is no need to update back refs | |
8862 | * for the subtree | |
8863 | */ | |
8864 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8865 | generation <= root->root_key.offset) { |
8866 | *lookup_info = 1; | |
1c4850e2 | 8867 | return 1; |
94fcca9f | 8868 | } |
1c4850e2 YZ |
8869 | |
8870 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
707e8a07 | 8871 | blocksize = root->nodesize; |
1c4850e2 | 8872 | |
01d58472 | 8873 | next = btrfs_find_tree_block(root->fs_info, bytenr); |
1c4850e2 | 8874 | if (!next) { |
a83fffb7 | 8875 | next = btrfs_find_create_tree_block(root, bytenr); |
c871b0f2 LB |
8876 | if (IS_ERR(next)) |
8877 | return PTR_ERR(next); | |
8878 | ||
b2aaaa3b JB |
8879 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8880 | level - 1); | |
1c4850e2 YZ |
8881 | reada = 1; |
8882 | } | |
8883 | btrfs_tree_lock(next); | |
8884 | btrfs_set_lock_blocking(next); | |
8885 | ||
3173a18f | 8886 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, |
94fcca9f YZ |
8887 | &wc->refs[level - 1], |
8888 | &wc->flags[level - 1]); | |
79787eaa JM |
8889 | if (ret < 0) { |
8890 | btrfs_tree_unlock(next); | |
a958eab0 | 8891 | free_extent_buffer(next); |
79787eaa JM |
8892 | return ret; |
8893 | } | |
8894 | ||
c2cf52eb SK |
8895 | if (unlikely(wc->refs[level - 1] == 0)) { |
8896 | btrfs_err(root->fs_info, "Missing references."); | |
8897 | BUG(); | |
8898 | } | |
94fcca9f | 8899 | *lookup_info = 0; |
1c4850e2 | 8900 | |
94fcca9f | 8901 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8902 | if (wc->refs[level - 1] > 1) { |
1152651a | 8903 | need_account = true; |
94fcca9f YZ |
8904 | if (level == 1 && |
8905 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8906 | goto skip; | |
8907 | ||
1c4850e2 YZ |
8908 | if (!wc->update_ref || |
8909 | generation <= root->root_key.offset) | |
8910 | goto skip; | |
8911 | ||
8912 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8913 | path->slots[level]); | |
8914 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8915 | if (ret < 0) | |
8916 | goto skip; | |
8917 | ||
8918 | wc->stage = UPDATE_BACKREF; | |
8919 | wc->shared_level = level - 1; | |
8920 | } | |
94fcca9f YZ |
8921 | } else { |
8922 | if (level == 1 && | |
8923 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8924 | goto skip; | |
1c4850e2 YZ |
8925 | } |
8926 | ||
b9fab919 | 8927 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8928 | btrfs_tree_unlock(next); |
8929 | free_extent_buffer(next); | |
8930 | next = NULL; | |
94fcca9f | 8931 | *lookup_info = 1; |
1c4850e2 YZ |
8932 | } |
8933 | ||
8934 | if (!next) { | |
8935 | if (reada && level == 1) | |
8936 | reada_walk_down(trans, root, wc, path); | |
ce86cd59 | 8937 | next = read_tree_block(root, bytenr, generation); |
64c043de LB |
8938 | if (IS_ERR(next)) { |
8939 | return PTR_ERR(next); | |
8940 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8941 | free_extent_buffer(next); |
97d9a8a4 | 8942 | return -EIO; |
416bc658 | 8943 | } |
1c4850e2 YZ |
8944 | btrfs_tree_lock(next); |
8945 | btrfs_set_lock_blocking(next); | |
8946 | } | |
8947 | ||
8948 | level--; | |
8949 | BUG_ON(level != btrfs_header_level(next)); | |
8950 | path->nodes[level] = next; | |
8951 | path->slots[level] = 0; | |
bd681513 | 8952 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8953 | wc->level = level; |
8954 | if (wc->level == 1) | |
8955 | wc->reada_slot = 0; | |
8956 | return 0; | |
8957 | skip: | |
8958 | wc->refs[level - 1] = 0; | |
8959 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8960 | if (wc->stage == DROP_REFERENCE) { |
8961 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8962 | parent = path->nodes[level]->start; | |
8963 | } else { | |
8964 | BUG_ON(root->root_key.objectid != | |
8965 | btrfs_header_owner(path->nodes[level])); | |
8966 | parent = 0; | |
8967 | } | |
1c4850e2 | 8968 | |
1152651a MF |
8969 | if (need_account) { |
8970 | ret = account_shared_subtree(trans, root, next, | |
8971 | generation, level - 1); | |
8972 | if (ret) { | |
94647322 | 8973 | btrfs_err_rl(root->fs_info, |
5d163e0e JM |
8974 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8975 | ret); | |
1152651a MF |
8976 | } |
8977 | } | |
94fcca9f | 8978 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
b06c4bf5 | 8979 | root->root_key.objectid, level - 1, 0); |
79787eaa | 8980 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 8981 | } |
1c4850e2 YZ |
8982 | btrfs_tree_unlock(next); |
8983 | free_extent_buffer(next); | |
94fcca9f | 8984 | *lookup_info = 1; |
1c4850e2 YZ |
8985 | return 1; |
8986 | } | |
8987 | ||
2c47e605 | 8988 | /* |
2c016dc2 | 8989 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8990 | * |
8991 | * when wc->stage == DROP_REFERENCE, this function drops | |
8992 | * reference count on the block. | |
8993 | * | |
8994 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8995 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8996 | * to UPDATE_BACKREF previously while processing the block. | |
8997 | * | |
8998 | * NOTE: return value 1 means we should stop walking up. | |
8999 | */ | |
9000 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
9001 | struct btrfs_root *root, | |
9002 | struct btrfs_path *path, | |
9003 | struct walk_control *wc) | |
9004 | { | |
f0486c68 | 9005 | int ret; |
2c47e605 YZ |
9006 | int level = wc->level; |
9007 | struct extent_buffer *eb = path->nodes[level]; | |
9008 | u64 parent = 0; | |
9009 | ||
9010 | if (wc->stage == UPDATE_BACKREF) { | |
9011 | BUG_ON(wc->shared_level < level); | |
9012 | if (level < wc->shared_level) | |
9013 | goto out; | |
9014 | ||
2c47e605 YZ |
9015 | ret = find_next_key(path, level + 1, &wc->update_progress); |
9016 | if (ret > 0) | |
9017 | wc->update_ref = 0; | |
9018 | ||
9019 | wc->stage = DROP_REFERENCE; | |
9020 | wc->shared_level = -1; | |
9021 | path->slots[level] = 0; | |
9022 | ||
9023 | /* | |
9024 | * check reference count again if the block isn't locked. | |
9025 | * we should start walking down the tree again if reference | |
9026 | * count is one. | |
9027 | */ | |
9028 | if (!path->locks[level]) { | |
9029 | BUG_ON(level == 0); | |
9030 | btrfs_tree_lock(eb); | |
9031 | btrfs_set_lock_blocking(eb); | |
bd681513 | 9032 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9033 | |
9034 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 9035 | eb->start, level, 1, |
2c47e605 YZ |
9036 | &wc->refs[level], |
9037 | &wc->flags[level]); | |
79787eaa JM |
9038 | if (ret < 0) { |
9039 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 9040 | path->locks[level] = 0; |
79787eaa JM |
9041 | return ret; |
9042 | } | |
2c47e605 YZ |
9043 | BUG_ON(wc->refs[level] == 0); |
9044 | if (wc->refs[level] == 1) { | |
bd681513 | 9045 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 9046 | path->locks[level] = 0; |
2c47e605 YZ |
9047 | return 1; |
9048 | } | |
f82d02d9 | 9049 | } |
2c47e605 | 9050 | } |
f82d02d9 | 9051 | |
2c47e605 YZ |
9052 | /* wc->stage == DROP_REFERENCE */ |
9053 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 9054 | |
2c47e605 YZ |
9055 | if (wc->refs[level] == 1) { |
9056 | if (level == 0) { | |
9057 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 9058 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 9059 | else |
e339a6b0 | 9060 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 9061 | BUG_ON(ret); /* -ENOMEM */ |
1152651a MF |
9062 | ret = account_leaf_items(trans, root, eb); |
9063 | if (ret) { | |
94647322 | 9064 | btrfs_err_rl(root->fs_info, |
5d163e0e JM |
9065 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
9066 | ret); | |
1152651a | 9067 | } |
2c47e605 YZ |
9068 | } |
9069 | /* make block locked assertion in clean_tree_block happy */ | |
9070 | if (!path->locks[level] && | |
9071 | btrfs_header_generation(eb) == trans->transid) { | |
9072 | btrfs_tree_lock(eb); | |
9073 | btrfs_set_lock_blocking(eb); | |
bd681513 | 9074 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9075 | } |
01d58472 | 9076 | clean_tree_block(trans, root->fs_info, eb); |
2c47e605 YZ |
9077 | } |
9078 | ||
9079 | if (eb == root->node) { | |
9080 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9081 | parent = eb->start; | |
9082 | else | |
9083 | BUG_ON(root->root_key.objectid != | |
9084 | btrfs_header_owner(eb)); | |
9085 | } else { | |
9086 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9087 | parent = path->nodes[level + 1]->start; | |
9088 | else | |
9089 | BUG_ON(root->root_key.objectid != | |
9090 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 9091 | } |
f82d02d9 | 9092 | |
5581a51a | 9093 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
9094 | out: |
9095 | wc->refs[level] = 0; | |
9096 | wc->flags[level] = 0; | |
f0486c68 | 9097 | return 0; |
2c47e605 YZ |
9098 | } |
9099 | ||
9100 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
9101 | struct btrfs_root *root, | |
9102 | struct btrfs_path *path, | |
9103 | struct walk_control *wc) | |
9104 | { | |
2c47e605 | 9105 | int level = wc->level; |
94fcca9f | 9106 | int lookup_info = 1; |
2c47e605 YZ |
9107 | int ret; |
9108 | ||
9109 | while (level >= 0) { | |
94fcca9f | 9110 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
9111 | if (ret > 0) |
9112 | break; | |
9113 | ||
9114 | if (level == 0) | |
9115 | break; | |
9116 | ||
7a7965f8 YZ |
9117 | if (path->slots[level] >= |
9118 | btrfs_header_nritems(path->nodes[level])) | |
9119 | break; | |
9120 | ||
94fcca9f | 9121 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
9122 | if (ret > 0) { |
9123 | path->slots[level]++; | |
9124 | continue; | |
90d2c51d MX |
9125 | } else if (ret < 0) |
9126 | return ret; | |
1c4850e2 | 9127 | level = wc->level; |
f82d02d9 | 9128 | } |
f82d02d9 YZ |
9129 | return 0; |
9130 | } | |
9131 | ||
d397712b | 9132 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 9133 | struct btrfs_root *root, |
f82d02d9 | 9134 | struct btrfs_path *path, |
2c47e605 | 9135 | struct walk_control *wc, int max_level) |
20524f02 | 9136 | { |
2c47e605 | 9137 | int level = wc->level; |
20524f02 | 9138 | int ret; |
9f3a7427 | 9139 | |
2c47e605 YZ |
9140 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
9141 | while (level < max_level && path->nodes[level]) { | |
9142 | wc->level = level; | |
9143 | if (path->slots[level] + 1 < | |
9144 | btrfs_header_nritems(path->nodes[level])) { | |
9145 | path->slots[level]++; | |
20524f02 CM |
9146 | return 0; |
9147 | } else { | |
2c47e605 YZ |
9148 | ret = walk_up_proc(trans, root, path, wc); |
9149 | if (ret > 0) | |
9150 | return 0; | |
bd56b302 | 9151 | |
2c47e605 | 9152 | if (path->locks[level]) { |
bd681513 CM |
9153 | btrfs_tree_unlock_rw(path->nodes[level], |
9154 | path->locks[level]); | |
2c47e605 | 9155 | path->locks[level] = 0; |
f82d02d9 | 9156 | } |
2c47e605 YZ |
9157 | free_extent_buffer(path->nodes[level]); |
9158 | path->nodes[level] = NULL; | |
9159 | level++; | |
20524f02 CM |
9160 | } |
9161 | } | |
9162 | return 1; | |
9163 | } | |
9164 | ||
9aca1d51 | 9165 | /* |
2c47e605 YZ |
9166 | * drop a subvolume tree. |
9167 | * | |
9168 | * this function traverses the tree freeing any blocks that only | |
9169 | * referenced by the tree. | |
9170 | * | |
9171 | * when a shared tree block is found. this function decreases its | |
9172 | * reference count by one. if update_ref is true, this function | |
9173 | * also make sure backrefs for the shared block and all lower level | |
9174 | * blocks are properly updated. | |
9d1a2a3a DS |
9175 | * |
9176 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9177 | */ |
2c536799 | 9178 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9179 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9180 | int for_reloc) | |
20524f02 | 9181 | { |
ab8d0fc4 | 9182 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9183 | struct btrfs_path *path; |
2c47e605 | 9184 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9185 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9186 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9187 | struct walk_control *wc; |
9188 | struct btrfs_key key; | |
9189 | int err = 0; | |
9190 | int ret; | |
9191 | int level; | |
d29a9f62 | 9192 | bool root_dropped = false; |
20524f02 | 9193 | |
ab8d0fc4 | 9194 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
1152651a | 9195 | |
5caf2a00 | 9196 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9197 | if (!path) { |
9198 | err = -ENOMEM; | |
9199 | goto out; | |
9200 | } | |
20524f02 | 9201 | |
2c47e605 | 9202 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9203 | if (!wc) { |
9204 | btrfs_free_path(path); | |
cb1b69f4 TI |
9205 | err = -ENOMEM; |
9206 | goto out; | |
38a1a919 | 9207 | } |
2c47e605 | 9208 | |
a22285a6 | 9209 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9210 | if (IS_ERR(trans)) { |
9211 | err = PTR_ERR(trans); | |
9212 | goto out_free; | |
9213 | } | |
98d5dc13 | 9214 | |
3fd0a558 YZ |
9215 | if (block_rsv) |
9216 | trans->block_rsv = block_rsv; | |
2c47e605 | 9217 | |
9f3a7427 | 9218 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9219 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9220 | path->nodes[level] = btrfs_lock_root_node(root); |
9221 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9222 | path->slots[level] = 0; |
bd681513 | 9223 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9224 | memset(&wc->update_progress, 0, |
9225 | sizeof(wc->update_progress)); | |
9f3a7427 | 9226 | } else { |
9f3a7427 | 9227 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9228 | memcpy(&wc->update_progress, &key, |
9229 | sizeof(wc->update_progress)); | |
9230 | ||
6702ed49 | 9231 | level = root_item->drop_level; |
2c47e605 | 9232 | BUG_ON(level == 0); |
6702ed49 | 9233 | path->lowest_level = level; |
2c47e605 YZ |
9234 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9235 | path->lowest_level = 0; | |
9236 | if (ret < 0) { | |
9237 | err = ret; | |
79787eaa | 9238 | goto out_end_trans; |
9f3a7427 | 9239 | } |
1c4850e2 | 9240 | WARN_ON(ret > 0); |
2c47e605 | 9241 | |
7d9eb12c CM |
9242 | /* |
9243 | * unlock our path, this is safe because only this | |
9244 | * function is allowed to delete this snapshot | |
9245 | */ | |
5d4f98a2 | 9246 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9247 | |
9248 | level = btrfs_header_level(root->node); | |
9249 | while (1) { | |
9250 | btrfs_tree_lock(path->nodes[level]); | |
9251 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9252 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9253 | |
9254 | ret = btrfs_lookup_extent_info(trans, root, | |
9255 | path->nodes[level]->start, | |
3173a18f | 9256 | level, 1, &wc->refs[level], |
2c47e605 | 9257 | &wc->flags[level]); |
79787eaa JM |
9258 | if (ret < 0) { |
9259 | err = ret; | |
9260 | goto out_end_trans; | |
9261 | } | |
2c47e605 YZ |
9262 | BUG_ON(wc->refs[level] == 0); |
9263 | ||
9264 | if (level == root_item->drop_level) | |
9265 | break; | |
9266 | ||
9267 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9268 | path->locks[level] = 0; |
2c47e605 YZ |
9269 | WARN_ON(wc->refs[level] != 1); |
9270 | level--; | |
9271 | } | |
9f3a7427 | 9272 | } |
2c47e605 YZ |
9273 | |
9274 | wc->level = level; | |
9275 | wc->shared_level = -1; | |
9276 | wc->stage = DROP_REFERENCE; | |
9277 | wc->update_ref = update_ref; | |
9278 | wc->keep_locks = 0; | |
66d7e7f0 | 9279 | wc->for_reloc = for_reloc; |
1c4850e2 | 9280 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 9281 | |
d397712b | 9282 | while (1) { |
9d1a2a3a | 9283 | |
2c47e605 YZ |
9284 | ret = walk_down_tree(trans, root, path, wc); |
9285 | if (ret < 0) { | |
9286 | err = ret; | |
20524f02 | 9287 | break; |
2c47e605 | 9288 | } |
9aca1d51 | 9289 | |
2c47e605 YZ |
9290 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9291 | if (ret < 0) { | |
9292 | err = ret; | |
20524f02 | 9293 | break; |
2c47e605 YZ |
9294 | } |
9295 | ||
9296 | if (ret > 0) { | |
9297 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9298 | break; |
9299 | } | |
2c47e605 YZ |
9300 | |
9301 | if (wc->stage == DROP_REFERENCE) { | |
9302 | level = wc->level; | |
9303 | btrfs_node_key(path->nodes[level], | |
9304 | &root_item->drop_progress, | |
9305 | path->slots[level]); | |
9306 | root_item->drop_level = level; | |
9307 | } | |
9308 | ||
9309 | BUG_ON(wc->level == 0); | |
3c8f2422 JB |
9310 | if (btrfs_should_end_transaction(trans, tree_root) || |
9311 | (!for_reloc && btrfs_need_cleaner_sleep(root))) { | |
2c47e605 YZ |
9312 | ret = btrfs_update_root(trans, tree_root, |
9313 | &root->root_key, | |
9314 | root_item); | |
79787eaa | 9315 | if (ret) { |
66642832 | 9316 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9317 | err = ret; |
9318 | goto out_end_trans; | |
9319 | } | |
2c47e605 | 9320 | |
3fd0a558 | 9321 | btrfs_end_transaction_throttle(trans, tree_root); |
3c8f2422 | 9322 | if (!for_reloc && btrfs_need_cleaner_sleep(root)) { |
ab8d0fc4 JM |
9323 | btrfs_debug(fs_info, |
9324 | "drop snapshot early exit"); | |
3c8f2422 JB |
9325 | err = -EAGAIN; |
9326 | goto out_free; | |
9327 | } | |
9328 | ||
a22285a6 | 9329 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9330 | if (IS_ERR(trans)) { |
9331 | err = PTR_ERR(trans); | |
9332 | goto out_free; | |
9333 | } | |
3fd0a558 YZ |
9334 | if (block_rsv) |
9335 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9336 | } |
20524f02 | 9337 | } |
b3b4aa74 | 9338 | btrfs_release_path(path); |
79787eaa JM |
9339 | if (err) |
9340 | goto out_end_trans; | |
2c47e605 YZ |
9341 | |
9342 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa | 9343 | if (ret) { |
66642832 | 9344 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9345 | goto out_end_trans; |
9346 | } | |
2c47e605 | 9347 | |
76dda93c | 9348 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9349 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9350 | NULL, NULL); | |
79787eaa | 9351 | if (ret < 0) { |
66642832 | 9352 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9353 | err = ret; |
9354 | goto out_end_trans; | |
9355 | } else if (ret > 0) { | |
84cd948c JB |
9356 | /* if we fail to delete the orphan item this time |
9357 | * around, it'll get picked up the next time. | |
9358 | * | |
9359 | * The most common failure here is just -ENOENT. | |
9360 | */ | |
9361 | btrfs_del_orphan_item(trans, tree_root, | |
9362 | root->root_key.objectid); | |
76dda93c YZ |
9363 | } |
9364 | } | |
9365 | ||
27cdeb70 | 9366 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9367 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9368 | } else { |
9369 | free_extent_buffer(root->node); | |
9370 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9371 | btrfs_put_fs_root(root); |
76dda93c | 9372 | } |
d29a9f62 | 9373 | root_dropped = true; |
79787eaa | 9374 | out_end_trans: |
3fd0a558 | 9375 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 9376 | out_free: |
2c47e605 | 9377 | kfree(wc); |
5caf2a00 | 9378 | btrfs_free_path(path); |
cb1b69f4 | 9379 | out: |
d29a9f62 JB |
9380 | /* |
9381 | * So if we need to stop dropping the snapshot for whatever reason we | |
9382 | * need to make sure to add it back to the dead root list so that we | |
9383 | * keep trying to do the work later. This also cleans up roots if we | |
9384 | * don't have it in the radix (like when we recover after a power fail | |
9385 | * or unmount) so we don't leak memory. | |
9386 | */ | |
b37b39cd | 9387 | if (!for_reloc && root_dropped == false) |
d29a9f62 | 9388 | btrfs_add_dead_root(root); |
90515e7f | 9389 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9390 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9391 | return err; |
20524f02 | 9392 | } |
9078a3e1 | 9393 | |
2c47e605 YZ |
9394 | /* |
9395 | * drop subtree rooted at tree block 'node'. | |
9396 | * | |
9397 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9398 | * only used by relocation code |
2c47e605 | 9399 | */ |
f82d02d9 YZ |
9400 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9401 | struct btrfs_root *root, | |
9402 | struct extent_buffer *node, | |
9403 | struct extent_buffer *parent) | |
9404 | { | |
9405 | struct btrfs_path *path; | |
2c47e605 | 9406 | struct walk_control *wc; |
f82d02d9 YZ |
9407 | int level; |
9408 | int parent_level; | |
9409 | int ret = 0; | |
9410 | int wret; | |
9411 | ||
2c47e605 YZ |
9412 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9413 | ||
f82d02d9 | 9414 | path = btrfs_alloc_path(); |
db5b493a TI |
9415 | if (!path) |
9416 | return -ENOMEM; | |
f82d02d9 | 9417 | |
2c47e605 | 9418 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9419 | if (!wc) { |
9420 | btrfs_free_path(path); | |
9421 | return -ENOMEM; | |
9422 | } | |
2c47e605 | 9423 | |
b9447ef8 | 9424 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9425 | parent_level = btrfs_header_level(parent); |
9426 | extent_buffer_get(parent); | |
9427 | path->nodes[parent_level] = parent; | |
9428 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9429 | ||
b9447ef8 | 9430 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9431 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9432 | path->nodes[level] = node; |
9433 | path->slots[level] = 0; | |
bd681513 | 9434 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9435 | |
9436 | wc->refs[parent_level] = 1; | |
9437 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9438 | wc->level = level; | |
9439 | wc->shared_level = -1; | |
9440 | wc->stage = DROP_REFERENCE; | |
9441 | wc->update_ref = 0; | |
9442 | wc->keep_locks = 1; | |
66d7e7f0 | 9443 | wc->for_reloc = 1; |
1c4850e2 | 9444 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
9445 | |
9446 | while (1) { | |
2c47e605 YZ |
9447 | wret = walk_down_tree(trans, root, path, wc); |
9448 | if (wret < 0) { | |
f82d02d9 | 9449 | ret = wret; |
f82d02d9 | 9450 | break; |
2c47e605 | 9451 | } |
f82d02d9 | 9452 | |
2c47e605 | 9453 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9454 | if (wret < 0) |
9455 | ret = wret; | |
9456 | if (wret != 0) | |
9457 | break; | |
9458 | } | |
9459 | ||
2c47e605 | 9460 | kfree(wc); |
f82d02d9 YZ |
9461 | btrfs_free_path(path); |
9462 | return ret; | |
9463 | } | |
9464 | ||
ec44a35c CM |
9465 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
9466 | { | |
9467 | u64 num_devices; | |
fc67c450 | 9468 | u64 stripped; |
e4d8ec0f | 9469 | |
fc67c450 ID |
9470 | /* |
9471 | * if restripe for this chunk_type is on pick target profile and | |
9472 | * return, otherwise do the usual balance | |
9473 | */ | |
9474 | stripped = get_restripe_target(root->fs_info, flags); | |
9475 | if (stripped) | |
9476 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9477 | |
95669976 | 9478 | num_devices = root->fs_info->fs_devices->rw_devices; |
cd02dca5 | 9479 | |
fc67c450 | 9480 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9481 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9482 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9483 | ||
ec44a35c CM |
9484 | if (num_devices == 1) { |
9485 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9486 | stripped = flags & ~stripped; | |
9487 | ||
9488 | /* turn raid0 into single device chunks */ | |
9489 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9490 | return stripped; | |
9491 | ||
9492 | /* turn mirroring into duplication */ | |
9493 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9494 | BTRFS_BLOCK_GROUP_RAID10)) | |
9495 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9496 | } else { |
9497 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9498 | if (flags & stripped) |
9499 | return flags; | |
9500 | ||
9501 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9502 | stripped = flags & ~stripped; | |
9503 | ||
9504 | /* switch duplicated blocks with raid1 */ | |
9505 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9506 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9507 | ||
e3176ca2 | 9508 | /* this is drive concat, leave it alone */ |
ec44a35c | 9509 | } |
e3176ca2 | 9510 | |
ec44a35c CM |
9511 | return flags; |
9512 | } | |
9513 | ||
868f401a | 9514 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9515 | { |
f0486c68 YZ |
9516 | struct btrfs_space_info *sinfo = cache->space_info; |
9517 | u64 num_bytes; | |
199c36ea | 9518 | u64 min_allocable_bytes; |
f0486c68 | 9519 | int ret = -ENOSPC; |
0ef3e66b | 9520 | |
199c36ea MX |
9521 | /* |
9522 | * We need some metadata space and system metadata space for | |
9523 | * allocating chunks in some corner cases until we force to set | |
9524 | * it to be readonly. | |
9525 | */ | |
9526 | if ((sinfo->flags & | |
9527 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9528 | !force) | |
ee22184b | 9529 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9530 | else |
9531 | min_allocable_bytes = 0; | |
9532 | ||
f0486c68 YZ |
9533 | spin_lock(&sinfo->lock); |
9534 | spin_lock(&cache->lock); | |
61cfea9b W |
9535 | |
9536 | if (cache->ro) { | |
868f401a | 9537 | cache->ro++; |
61cfea9b W |
9538 | ret = 0; |
9539 | goto out; | |
9540 | } | |
9541 | ||
f0486c68 YZ |
9542 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9543 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9544 | ||
9545 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
9546 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
9547 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 9548 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9549 | cache->ro++; |
633c0aad | 9550 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9551 | ret = 0; |
9552 | } | |
61cfea9b | 9553 | out: |
f0486c68 YZ |
9554 | spin_unlock(&cache->lock); |
9555 | spin_unlock(&sinfo->lock); | |
9556 | return ret; | |
9557 | } | |
7d9eb12c | 9558 | |
868f401a | 9559 | int btrfs_inc_block_group_ro(struct btrfs_root *root, |
f0486c68 | 9560 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9561 | |
f0486c68 YZ |
9562 | { |
9563 | struct btrfs_trans_handle *trans; | |
9564 | u64 alloc_flags; | |
9565 | int ret; | |
7d9eb12c | 9566 | |
1bbc621e | 9567 | again: |
ff5714cc | 9568 | trans = btrfs_join_transaction(root); |
79787eaa JM |
9569 | if (IS_ERR(trans)) |
9570 | return PTR_ERR(trans); | |
5d4f98a2 | 9571 | |
1bbc621e CM |
9572 | /* |
9573 | * we're not allowed to set block groups readonly after the dirty | |
9574 | * block groups cache has started writing. If it already started, | |
9575 | * back off and let this transaction commit | |
9576 | */ | |
9577 | mutex_lock(&root->fs_info->ro_block_group_mutex); | |
3204d33c | 9578 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9579 | u64 transid = trans->transid; |
9580 | ||
9581 | mutex_unlock(&root->fs_info->ro_block_group_mutex); | |
9582 | btrfs_end_transaction(trans, root); | |
9583 | ||
9584 | ret = btrfs_wait_for_commit(root, transid); | |
9585 | if (ret) | |
9586 | return ret; | |
9587 | goto again; | |
9588 | } | |
9589 | ||
153c35b6 CM |
9590 | /* |
9591 | * if we are changing raid levels, try to allocate a corresponding | |
9592 | * block group with the new raid level. | |
9593 | */ | |
9594 | alloc_flags = update_block_group_flags(root, cache->flags); | |
9595 | if (alloc_flags != cache->flags) { | |
9596 | ret = do_chunk_alloc(trans, root, alloc_flags, | |
9597 | CHUNK_ALLOC_FORCE); | |
9598 | /* | |
9599 | * ENOSPC is allowed here, we may have enough space | |
9600 | * already allocated at the new raid level to | |
9601 | * carry on | |
9602 | */ | |
9603 | if (ret == -ENOSPC) | |
9604 | ret = 0; | |
9605 | if (ret < 0) | |
9606 | goto out; | |
9607 | } | |
1bbc621e | 9608 | |
868f401a | 9609 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9610 | if (!ret) |
9611 | goto out; | |
9612 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 9613 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 9614 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9615 | if (ret < 0) |
9616 | goto out; | |
868f401a | 9617 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9618 | out: |
2f081088 SL |
9619 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
9620 | alloc_flags = update_block_group_flags(root, cache->flags); | |
a9629596 | 9621 | lock_chunks(root->fs_info->chunk_root); |
4617ea3a | 9622 | check_system_chunk(trans, root, alloc_flags); |
a9629596 | 9623 | unlock_chunks(root->fs_info->chunk_root); |
2f081088 | 9624 | } |
1bbc621e | 9625 | mutex_unlock(&root->fs_info->ro_block_group_mutex); |
2f081088 | 9626 | |
f0486c68 YZ |
9627 | btrfs_end_transaction(trans, root); |
9628 | return ret; | |
9629 | } | |
5d4f98a2 | 9630 | |
c87f08ca CM |
9631 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
9632 | struct btrfs_root *root, u64 type) | |
9633 | { | |
9634 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 9635 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 9636 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9637 | } |
9638 | ||
6d07bcec MX |
9639 | /* |
9640 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9641 | * space_info. takes mirrors into account. |
6d07bcec | 9642 | */ |
633c0aad | 9643 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9644 | { |
9645 | struct btrfs_block_group_cache *block_group; | |
9646 | u64 free_bytes = 0; | |
9647 | int factor; | |
9648 | ||
01327610 | 9649 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9650 | if (list_empty(&sinfo->ro_bgs)) |
9651 | return 0; | |
9652 | ||
9653 | spin_lock(&sinfo->lock); | |
9654 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9655 | spin_lock(&block_group->lock); |
9656 | ||
9657 | if (!block_group->ro) { | |
9658 | spin_unlock(&block_group->lock); | |
9659 | continue; | |
9660 | } | |
9661 | ||
9662 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9663 | BTRFS_BLOCK_GROUP_RAID10 | | |
9664 | BTRFS_BLOCK_GROUP_DUP)) | |
9665 | factor = 2; | |
9666 | else | |
9667 | factor = 1; | |
9668 | ||
9669 | free_bytes += (block_group->key.offset - | |
9670 | btrfs_block_group_used(&block_group->item)) * | |
9671 | factor; | |
9672 | ||
9673 | spin_unlock(&block_group->lock); | |
9674 | } | |
6d07bcec MX |
9675 | spin_unlock(&sinfo->lock); |
9676 | ||
9677 | return free_bytes; | |
9678 | } | |
9679 | ||
868f401a | 9680 | void btrfs_dec_block_group_ro(struct btrfs_root *root, |
f0486c68 | 9681 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9682 | { |
f0486c68 YZ |
9683 | struct btrfs_space_info *sinfo = cache->space_info; |
9684 | u64 num_bytes; | |
9685 | ||
9686 | BUG_ON(!cache->ro); | |
9687 | ||
9688 | spin_lock(&sinfo->lock); | |
9689 | spin_lock(&cache->lock); | |
868f401a Z |
9690 | if (!--cache->ro) { |
9691 | num_bytes = cache->key.offset - cache->reserved - | |
9692 | cache->pinned - cache->bytes_super - | |
9693 | btrfs_block_group_used(&cache->item); | |
9694 | sinfo->bytes_readonly -= num_bytes; | |
9695 | list_del_init(&cache->ro_list); | |
9696 | } | |
f0486c68 YZ |
9697 | spin_unlock(&cache->lock); |
9698 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9699 | } |
9700 | ||
ba1bf481 JB |
9701 | /* |
9702 | * checks to see if its even possible to relocate this block group. | |
9703 | * | |
9704 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9705 | * ok to go ahead and try. | |
9706 | */ | |
9707 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 9708 | { |
ba1bf481 JB |
9709 | struct btrfs_block_group_cache *block_group; |
9710 | struct btrfs_space_info *space_info; | |
9711 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
9712 | struct btrfs_device *device; | |
6df9a95e | 9713 | struct btrfs_trans_handle *trans; |
cdcb725c | 9714 | u64 min_free; |
6719db6a JB |
9715 | u64 dev_min = 1; |
9716 | u64 dev_nr = 0; | |
4a5e98f5 | 9717 | u64 target; |
0305bc27 | 9718 | int debug; |
cdcb725c | 9719 | int index; |
ba1bf481 JB |
9720 | int full = 0; |
9721 | int ret = 0; | |
1a40e23b | 9722 | |
3cdde224 | 9723 | debug = btrfs_test_opt(root->fs_info, ENOSPC_DEBUG); |
0305bc27 | 9724 | |
ba1bf481 | 9725 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 9726 | |
ba1bf481 | 9727 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9728 | if (!block_group) { |
9729 | if (debug) | |
9730 | btrfs_warn(root->fs_info, | |
9731 | "can't find block group for bytenr %llu", | |
9732 | bytenr); | |
ba1bf481 | 9733 | return -1; |
0305bc27 | 9734 | } |
1a40e23b | 9735 | |
cdcb725c | 9736 | min_free = btrfs_block_group_used(&block_group->item); |
9737 | ||
ba1bf481 | 9738 | /* no bytes used, we're good */ |
cdcb725c | 9739 | if (!min_free) |
1a40e23b ZY |
9740 | goto out; |
9741 | ||
ba1bf481 JB |
9742 | space_info = block_group->space_info; |
9743 | spin_lock(&space_info->lock); | |
17d217fe | 9744 | |
ba1bf481 | 9745 | full = space_info->full; |
17d217fe | 9746 | |
ba1bf481 JB |
9747 | /* |
9748 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9749 | * relocate it unless we're able to allocate a new chunk below. |
9750 | * | |
9751 | * Otherwise, we need to make sure we have room in the space to handle | |
9752 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9753 | */ |
7ce618db | 9754 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 9755 | (space_info->bytes_used + space_info->bytes_reserved + |
9756 | space_info->bytes_pinned + space_info->bytes_readonly + | |
9757 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
9758 | spin_unlock(&space_info->lock); |
9759 | goto out; | |
17d217fe | 9760 | } |
ba1bf481 | 9761 | spin_unlock(&space_info->lock); |
ea8c2819 | 9762 | |
ba1bf481 JB |
9763 | /* |
9764 | * ok we don't have enough space, but maybe we have free space on our | |
9765 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9766 | * alloc devices and guess if we have enough space. if this block |
9767 | * group is going to be restriped, run checks against the target | |
9768 | * profile instead of the current one. | |
ba1bf481 JB |
9769 | */ |
9770 | ret = -1; | |
ea8c2819 | 9771 | |
cdcb725c | 9772 | /* |
9773 | * index: | |
9774 | * 0: raid10 | |
9775 | * 1: raid1 | |
9776 | * 2: dup | |
9777 | * 3: raid0 | |
9778 | * 4: single | |
9779 | */ | |
4a5e98f5 ID |
9780 | target = get_restripe_target(root->fs_info, block_group->flags); |
9781 | if (target) { | |
31e50229 | 9782 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9783 | } else { |
9784 | /* | |
9785 | * this is just a balance, so if we were marked as full | |
9786 | * we know there is no space for a new chunk | |
9787 | */ | |
0305bc27 QW |
9788 | if (full) { |
9789 | if (debug) | |
9790 | btrfs_warn(root->fs_info, | |
9791 | "no space to alloc new chunk for block group %llu", | |
9792 | block_group->key.objectid); | |
4a5e98f5 | 9793 | goto out; |
0305bc27 | 9794 | } |
4a5e98f5 ID |
9795 | |
9796 | index = get_block_group_index(block_group); | |
9797 | } | |
9798 | ||
e6ec716f | 9799 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9800 | dev_min = 4; |
6719db6a JB |
9801 | /* Divide by 2 */ |
9802 | min_free >>= 1; | |
e6ec716f | 9803 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9804 | dev_min = 2; |
e6ec716f | 9805 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9806 | /* Multiply by 2 */ |
9807 | min_free <<= 1; | |
e6ec716f | 9808 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9809 | dev_min = fs_devices->rw_devices; |
47c5713f | 9810 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9811 | } |
9812 | ||
6df9a95e JB |
9813 | /* We need to do this so that we can look at pending chunks */ |
9814 | trans = btrfs_join_transaction(root); | |
9815 | if (IS_ERR(trans)) { | |
9816 | ret = PTR_ERR(trans); | |
9817 | goto out; | |
9818 | } | |
9819 | ||
ba1bf481 JB |
9820 | mutex_lock(&root->fs_info->chunk_mutex); |
9821 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 9822 | u64 dev_offset; |
56bec294 | 9823 | |
ba1bf481 JB |
9824 | /* |
9825 | * check to make sure we can actually find a chunk with enough | |
9826 | * space to fit our block group in. | |
9827 | */ | |
63a212ab SB |
9828 | if (device->total_bytes > device->bytes_used + min_free && |
9829 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 9830 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9831 | &dev_offset, NULL); |
ba1bf481 | 9832 | if (!ret) |
cdcb725c | 9833 | dev_nr++; |
9834 | ||
9835 | if (dev_nr >= dev_min) | |
73e48b27 | 9836 | break; |
cdcb725c | 9837 | |
ba1bf481 | 9838 | ret = -1; |
725c8463 | 9839 | } |
edbd8d4e | 9840 | } |
0305bc27 QW |
9841 | if (debug && ret == -1) |
9842 | btrfs_warn(root->fs_info, | |
9843 | "no space to allocate a new chunk for block group %llu", | |
9844 | block_group->key.objectid); | |
ba1bf481 | 9845 | mutex_unlock(&root->fs_info->chunk_mutex); |
6df9a95e | 9846 | btrfs_end_transaction(trans, root); |
edbd8d4e | 9847 | out: |
ba1bf481 | 9848 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9849 | return ret; |
9850 | } | |
9851 | ||
b2950863 CH |
9852 | static int find_first_block_group(struct btrfs_root *root, |
9853 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 9854 | { |
925baedd | 9855 | int ret = 0; |
0b86a832 CM |
9856 | struct btrfs_key found_key; |
9857 | struct extent_buffer *leaf; | |
9858 | int slot; | |
edbd8d4e | 9859 | |
0b86a832 CM |
9860 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9861 | if (ret < 0) | |
925baedd CM |
9862 | goto out; |
9863 | ||
d397712b | 9864 | while (1) { |
0b86a832 | 9865 | slot = path->slots[0]; |
edbd8d4e | 9866 | leaf = path->nodes[0]; |
0b86a832 CM |
9867 | if (slot >= btrfs_header_nritems(leaf)) { |
9868 | ret = btrfs_next_leaf(root, path); | |
9869 | if (ret == 0) | |
9870 | continue; | |
9871 | if (ret < 0) | |
925baedd | 9872 | goto out; |
0b86a832 | 9873 | break; |
edbd8d4e | 9874 | } |
0b86a832 | 9875 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9876 | |
0b86a832 | 9877 | if (found_key.objectid >= key->objectid && |
925baedd | 9878 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9879 | struct extent_map_tree *em_tree; |
9880 | struct extent_map *em; | |
9881 | ||
9882 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9883 | read_lock(&em_tree->lock); | |
9884 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9885 | found_key.offset); | |
9886 | read_unlock(&em_tree->lock); | |
9887 | if (!em) { | |
9888 | btrfs_err(root->fs_info, | |
9889 | "logical %llu len %llu found bg but no related chunk", | |
9890 | found_key.objectid, found_key.offset); | |
9891 | ret = -ENOENT; | |
9892 | } else { | |
9893 | ret = 0; | |
9894 | } | |
187ee58c | 9895 | free_extent_map(em); |
925baedd CM |
9896 | goto out; |
9897 | } | |
0b86a832 | 9898 | path->slots[0]++; |
edbd8d4e | 9899 | } |
925baedd | 9900 | out: |
0b86a832 | 9901 | return ret; |
edbd8d4e CM |
9902 | } |
9903 | ||
0af3d00b JB |
9904 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9905 | { | |
9906 | struct btrfs_block_group_cache *block_group; | |
9907 | u64 last = 0; | |
9908 | ||
9909 | while (1) { | |
9910 | struct inode *inode; | |
9911 | ||
9912 | block_group = btrfs_lookup_first_block_group(info, last); | |
9913 | while (block_group) { | |
9914 | spin_lock(&block_group->lock); | |
9915 | if (block_group->iref) | |
9916 | break; | |
9917 | spin_unlock(&block_group->lock); | |
9918 | block_group = next_block_group(info->tree_root, | |
9919 | block_group); | |
9920 | } | |
9921 | if (!block_group) { | |
9922 | if (last == 0) | |
9923 | break; | |
9924 | last = 0; | |
9925 | continue; | |
9926 | } | |
9927 | ||
9928 | inode = block_group->inode; | |
9929 | block_group->iref = 0; | |
9930 | block_group->inode = NULL; | |
9931 | spin_unlock(&block_group->lock); | |
f3bca802 | 9932 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9933 | iput(inode); |
9934 | last = block_group->key.objectid + block_group->key.offset; | |
9935 | btrfs_put_block_group(block_group); | |
9936 | } | |
9937 | } | |
9938 | ||
1a40e23b ZY |
9939 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9940 | { | |
9941 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9942 | struct btrfs_space_info *space_info; |
11833d66 | 9943 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9944 | struct rb_node *n; |
9945 | ||
9e351cc8 | 9946 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9947 | while (!list_empty(&info->caching_block_groups)) { |
9948 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9949 | struct btrfs_caching_control, list); | |
9950 | list_del(&caching_ctl->list); | |
9951 | put_caching_control(caching_ctl); | |
9952 | } | |
9e351cc8 | 9953 | up_write(&info->commit_root_sem); |
11833d66 | 9954 | |
47ab2a6c JB |
9955 | spin_lock(&info->unused_bgs_lock); |
9956 | while (!list_empty(&info->unused_bgs)) { | |
9957 | block_group = list_first_entry(&info->unused_bgs, | |
9958 | struct btrfs_block_group_cache, | |
9959 | bg_list); | |
9960 | list_del_init(&block_group->bg_list); | |
9961 | btrfs_put_block_group(block_group); | |
9962 | } | |
9963 | spin_unlock(&info->unused_bgs_lock); | |
9964 | ||
1a40e23b ZY |
9965 | spin_lock(&info->block_group_cache_lock); |
9966 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9967 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9968 | cache_node); | |
1a40e23b ZY |
9969 | rb_erase(&block_group->cache_node, |
9970 | &info->block_group_cache_tree); | |
01eacb27 | 9971 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9972 | spin_unlock(&info->block_group_cache_lock); |
9973 | ||
80eb234a | 9974 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9975 | list_del(&block_group->list); |
80eb234a | 9976 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9977 | |
817d52f8 | 9978 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 9979 | wait_block_group_cache_done(block_group); |
817d52f8 | 9980 | |
3c14874a JB |
9981 | /* |
9982 | * We haven't cached this block group, which means we could | |
9983 | * possibly have excluded extents on this block group. | |
9984 | */ | |
36cce922 JB |
9985 | if (block_group->cached == BTRFS_CACHE_NO || |
9986 | block_group->cached == BTRFS_CACHE_ERROR) | |
3c14874a JB |
9987 | free_excluded_extents(info->extent_root, block_group); |
9988 | ||
817d52f8 | 9989 | btrfs_remove_free_space_cache(block_group); |
f3bca802 LB |
9990 | ASSERT(list_empty(&block_group->dirty_list)); |
9991 | ASSERT(list_empty(&block_group->io_list)); | |
9992 | ASSERT(list_empty(&block_group->bg_list)); | |
9993 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9994 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9995 | |
9996 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9997 | } |
9998 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9999 | |
10000 | /* now that all the block groups are freed, go through and | |
10001 | * free all the space_info structs. This is only called during | |
10002 | * the final stages of unmount, and so we know nobody is | |
10003 | * using them. We call synchronize_rcu() once before we start, | |
10004 | * just to be on the safe side. | |
10005 | */ | |
10006 | synchronize_rcu(); | |
10007 | ||
8929ecfa YZ |
10008 | release_global_block_rsv(info); |
10009 | ||
67871254 | 10010 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
10011 | int i; |
10012 | ||
4184ea7f CM |
10013 | space_info = list_entry(info->space_info.next, |
10014 | struct btrfs_space_info, | |
10015 | list); | |
d555b6c3 JB |
10016 | |
10017 | /* | |
10018 | * Do not hide this behind enospc_debug, this is actually | |
10019 | * important and indicates a real bug if this happens. | |
10020 | */ | |
10021 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 10022 | space_info->bytes_reserved > 0 || |
d555b6c3 | 10023 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 10024 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 10025 | list_del(&space_info->list); |
6ab0a202 JM |
10026 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
10027 | struct kobject *kobj; | |
c1895442 JM |
10028 | kobj = space_info->block_group_kobjs[i]; |
10029 | space_info->block_group_kobjs[i] = NULL; | |
10030 | if (kobj) { | |
6ab0a202 JM |
10031 | kobject_del(kobj); |
10032 | kobject_put(kobj); | |
10033 | } | |
10034 | } | |
10035 | kobject_del(&space_info->kobj); | |
10036 | kobject_put(&space_info->kobj); | |
4184ea7f | 10037 | } |
1a40e23b ZY |
10038 | return 0; |
10039 | } | |
10040 | ||
b742bb82 YZ |
10041 | static void __link_block_group(struct btrfs_space_info *space_info, |
10042 | struct btrfs_block_group_cache *cache) | |
10043 | { | |
10044 | int index = get_block_group_index(cache); | |
ed55b6ac | 10045 | bool first = false; |
b742bb82 YZ |
10046 | |
10047 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
10048 | if (list_empty(&space_info->block_groups[index])) |
10049 | first = true; | |
10050 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
10051 | up_write(&space_info->groups_sem); | |
10052 | ||
10053 | if (first) { | |
c1895442 | 10054 | struct raid_kobject *rkobj; |
6ab0a202 JM |
10055 | int ret; |
10056 | ||
c1895442 JM |
10057 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
10058 | if (!rkobj) | |
10059 | goto out_err; | |
10060 | rkobj->raid_type = index; | |
10061 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
10062 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
10063 | "%s", get_raid_name(index)); | |
6ab0a202 | 10064 | if (ret) { |
c1895442 JM |
10065 | kobject_put(&rkobj->kobj); |
10066 | goto out_err; | |
6ab0a202 | 10067 | } |
c1895442 | 10068 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 10069 | } |
c1895442 JM |
10070 | |
10071 | return; | |
10072 | out_err: | |
ab8d0fc4 JM |
10073 | btrfs_warn(cache->fs_info, |
10074 | "failed to add kobject for block cache, ignoring"); | |
b742bb82 YZ |
10075 | } |
10076 | ||
920e4a58 MX |
10077 | static struct btrfs_block_group_cache * |
10078 | btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size) | |
10079 | { | |
10080 | struct btrfs_block_group_cache *cache; | |
10081 | ||
10082 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
10083 | if (!cache) | |
10084 | return NULL; | |
10085 | ||
10086 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
10087 | GFP_NOFS); | |
10088 | if (!cache->free_space_ctl) { | |
10089 | kfree(cache); | |
10090 | return NULL; | |
10091 | } | |
10092 | ||
10093 | cache->key.objectid = start; | |
10094 | cache->key.offset = size; | |
10095 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
10096 | ||
10097 | cache->sectorsize = root->sectorsize; | |
10098 | cache->fs_info = root->fs_info; | |
10099 | cache->full_stripe_len = btrfs_full_stripe_len(root, | |
10100 | &root->fs_info->mapping_tree, | |
10101 | start); | |
1e144fb8 OS |
10102 | set_free_space_tree_thresholds(cache); |
10103 | ||
920e4a58 MX |
10104 | atomic_set(&cache->count, 1); |
10105 | spin_lock_init(&cache->lock); | |
e570fd27 | 10106 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
10107 | INIT_LIST_HEAD(&cache->list); |
10108 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 10109 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 10110 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 10111 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 10112 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 10113 | btrfs_init_free_space_ctl(cache); |
04216820 | 10114 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 10115 | mutex_init(&cache->free_space_lock); |
920e4a58 MX |
10116 | |
10117 | return cache; | |
10118 | } | |
10119 | ||
9078a3e1 CM |
10120 | int btrfs_read_block_groups(struct btrfs_root *root) |
10121 | { | |
10122 | struct btrfs_path *path; | |
10123 | int ret; | |
9078a3e1 | 10124 | struct btrfs_block_group_cache *cache; |
be744175 | 10125 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 10126 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
10127 | struct btrfs_key key; |
10128 | struct btrfs_key found_key; | |
5f39d397 | 10129 | struct extent_buffer *leaf; |
0af3d00b JB |
10130 | int need_clear = 0; |
10131 | u64 cache_gen; | |
49303381 LB |
10132 | u64 feature; |
10133 | int mixed; | |
10134 | ||
10135 | feature = btrfs_super_incompat_flags(info->super_copy); | |
10136 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 10137 | |
be744175 | 10138 | root = info->extent_root; |
9078a3e1 | 10139 | key.objectid = 0; |
0b86a832 | 10140 | key.offset = 0; |
962a298f | 10141 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
10142 | path = btrfs_alloc_path(); |
10143 | if (!path) | |
10144 | return -ENOMEM; | |
e4058b54 | 10145 | path->reada = READA_FORWARD; |
9078a3e1 | 10146 | |
6c41761f | 10147 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
3cdde224 | 10148 | if (btrfs_test_opt(root->fs_info, SPACE_CACHE) && |
6c41761f | 10149 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 10150 | need_clear = 1; |
3cdde224 | 10151 | if (btrfs_test_opt(root->fs_info, CLEAR_CACHE)) |
88c2ba3b | 10152 | need_clear = 1; |
0af3d00b | 10153 | |
d397712b | 10154 | while (1) { |
0b86a832 | 10155 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
10156 | if (ret > 0) |
10157 | break; | |
0b86a832 CM |
10158 | if (ret != 0) |
10159 | goto error; | |
920e4a58 | 10160 | |
5f39d397 CM |
10161 | leaf = path->nodes[0]; |
10162 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 MX |
10163 | |
10164 | cache = btrfs_create_block_group_cache(root, found_key.objectid, | |
10165 | found_key.offset); | |
9078a3e1 | 10166 | if (!cache) { |
0b86a832 | 10167 | ret = -ENOMEM; |
f0486c68 | 10168 | goto error; |
9078a3e1 | 10169 | } |
96303081 | 10170 | |
cf7c1ef6 LB |
10171 | if (need_clear) { |
10172 | /* | |
10173 | * When we mount with old space cache, we need to | |
10174 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10175 | * | |
10176 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10177 | * truncate the old free space cache inode and | |
10178 | * setup a new one. | |
10179 | * b) Setting 'dirty flag' makes sure that we flush | |
10180 | * the new space cache info onto disk. | |
10181 | */ | |
3cdde224 | 10182 | if (btrfs_test_opt(root->fs_info, SPACE_CACHE)) |
ce93ec54 | 10183 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10184 | } |
0af3d00b | 10185 | |
5f39d397 CM |
10186 | read_extent_buffer(leaf, &cache->item, |
10187 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10188 | sizeof(cache->item)); | |
920e4a58 | 10189 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10190 | if (!mixed && |
10191 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10192 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10193 | btrfs_err(info, | |
10194 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10195 | cache->key.objectid); | |
10196 | ret = -EINVAL; | |
10197 | goto error; | |
10198 | } | |
0b86a832 | 10199 | |
9078a3e1 | 10200 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10201 | btrfs_release_path(path); |
34d52cb6 | 10202 | |
3c14874a JB |
10203 | /* |
10204 | * We need to exclude the super stripes now so that the space | |
10205 | * info has super bytes accounted for, otherwise we'll think | |
10206 | * we have more space than we actually do. | |
10207 | */ | |
835d974f JB |
10208 | ret = exclude_super_stripes(root, cache); |
10209 | if (ret) { | |
10210 | /* | |
10211 | * We may have excluded something, so call this just in | |
10212 | * case. | |
10213 | */ | |
10214 | free_excluded_extents(root, cache); | |
920e4a58 | 10215 | btrfs_put_block_group(cache); |
835d974f JB |
10216 | goto error; |
10217 | } | |
3c14874a | 10218 | |
817d52f8 JB |
10219 | /* |
10220 | * check for two cases, either we are full, and therefore | |
10221 | * don't need to bother with the caching work since we won't | |
10222 | * find any space, or we are empty, and we can just add all | |
10223 | * the space in and be done with it. This saves us _alot_ of | |
10224 | * time, particularly in the full case. | |
10225 | */ | |
10226 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10227 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10228 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 10229 | free_excluded_extents(root, cache); |
817d52f8 | 10230 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10231 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
10232 | cache->cached = BTRFS_CACHE_FINISHED; |
10233 | add_new_free_space(cache, root->fs_info, | |
10234 | found_key.objectid, | |
10235 | found_key.objectid + | |
10236 | found_key.offset); | |
11833d66 | 10237 | free_excluded_extents(root, cache); |
817d52f8 | 10238 | } |
96b5179d | 10239 | |
8c579fe7 JB |
10240 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
10241 | if (ret) { | |
10242 | btrfs_remove_free_space_cache(cache); | |
10243 | btrfs_put_block_group(cache); | |
10244 | goto error; | |
10245 | } | |
10246 | ||
c83f8eff | 10247 | trace_btrfs_add_block_group(root->fs_info, cache, 0); |
6324fbf3 CM |
10248 | ret = update_space_info(info, cache->flags, found_key.offset, |
10249 | btrfs_block_group_used(&cache->item), | |
e40edf2d | 10250 | cache->bytes_super, &space_info); |
8c579fe7 JB |
10251 | if (ret) { |
10252 | btrfs_remove_free_space_cache(cache); | |
10253 | spin_lock(&info->block_group_cache_lock); | |
10254 | rb_erase(&cache->cache_node, | |
10255 | &info->block_group_cache_tree); | |
01eacb27 | 10256 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
10257 | spin_unlock(&info->block_group_cache_lock); |
10258 | btrfs_put_block_group(cache); | |
10259 | goto error; | |
10260 | } | |
10261 | ||
6324fbf3 | 10262 | cache->space_info = space_info; |
1b2da372 | 10263 | |
b742bb82 | 10264 | __link_block_group(space_info, cache); |
0f9dd46c | 10265 | |
75ccf47d | 10266 | set_avail_alloc_bits(root->fs_info, cache->flags); |
47ab2a6c | 10267 | if (btrfs_chunk_readonly(root, cache->key.objectid)) { |
868f401a | 10268 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10269 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10270 | spin_lock(&info->unused_bgs_lock); | |
10271 | /* Should always be true but just in case. */ | |
10272 | if (list_empty(&cache->bg_list)) { | |
10273 | btrfs_get_block_group(cache); | |
10274 | list_add_tail(&cache->bg_list, | |
10275 | &info->unused_bgs); | |
10276 | } | |
10277 | spin_unlock(&info->unused_bgs_lock); | |
10278 | } | |
9078a3e1 | 10279 | } |
b742bb82 YZ |
10280 | |
10281 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
10282 | if (!(get_alloc_profile(root, space_info->flags) & | |
10283 | (BTRFS_BLOCK_GROUP_RAID10 | | |
10284 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10285 | BTRFS_BLOCK_GROUP_RAID5 | |
10286 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10287 | BTRFS_BLOCK_GROUP_DUP))) |
10288 | continue; | |
10289 | /* | |
10290 | * avoid allocating from un-mirrored block group if there are | |
10291 | * mirrored block groups. | |
10292 | */ | |
1095cc0d | 10293 | list_for_each_entry(cache, |
10294 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10295 | list) | |
868f401a | 10296 | inc_block_group_ro(cache, 1); |
1095cc0d | 10297 | list_for_each_entry(cache, |
10298 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10299 | list) | |
868f401a | 10300 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10301 | } |
f0486c68 YZ |
10302 | |
10303 | init_global_block_rsv(info); | |
0b86a832 CM |
10304 | ret = 0; |
10305 | error: | |
9078a3e1 | 10306 | btrfs_free_path(path); |
0b86a832 | 10307 | return ret; |
9078a3e1 | 10308 | } |
6324fbf3 | 10309 | |
ea658bad JB |
10310 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
10311 | struct btrfs_root *root) | |
10312 | { | |
10313 | struct btrfs_block_group_cache *block_group, *tmp; | |
10314 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
10315 | struct btrfs_block_group_item item; | |
10316 | struct btrfs_key key; | |
10317 | int ret = 0; | |
d9a0540a | 10318 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10319 | |
d9a0540a | 10320 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10321 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10322 | if (ret) |
c92f6be3 | 10323 | goto next; |
ea658bad JB |
10324 | |
10325 | spin_lock(&block_group->lock); | |
10326 | memcpy(&item, &block_group->item, sizeof(item)); | |
10327 | memcpy(&key, &block_group->key, sizeof(key)); | |
10328 | spin_unlock(&block_group->lock); | |
10329 | ||
10330 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10331 | sizeof(item)); | |
10332 | if (ret) | |
66642832 | 10333 | btrfs_abort_transaction(trans, ret); |
6df9a95e JB |
10334 | ret = btrfs_finish_chunk_alloc(trans, extent_root, |
10335 | key.objectid, key.offset); | |
10336 | if (ret) | |
66642832 | 10337 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
10338 | add_block_group_free_space(trans, root->fs_info, block_group); |
10339 | /* already aborted the transaction if it failed. */ | |
c92f6be3 FM |
10340 | next: |
10341 | list_del_init(&block_group->bg_list); | |
ea658bad | 10342 | } |
d9a0540a | 10343 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10344 | } |
10345 | ||
6324fbf3 CM |
10346 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
10347 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 10348 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
10349 | u64 size) |
10350 | { | |
10351 | int ret; | |
6324fbf3 CM |
10352 | struct btrfs_root *extent_root; |
10353 | struct btrfs_block_group_cache *cache; | |
6324fbf3 | 10354 | extent_root = root->fs_info->extent_root; |
6324fbf3 | 10355 | |
995946dd | 10356 | btrfs_set_log_full_commit(root->fs_info, trans); |
e02119d5 | 10357 | |
920e4a58 | 10358 | cache = btrfs_create_block_group_cache(root, chunk_offset, size); |
0f9dd46c JB |
10359 | if (!cache) |
10360 | return -ENOMEM; | |
34d52cb6 | 10361 | |
6324fbf3 | 10362 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 | 10363 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
6324fbf3 CM |
10364 | btrfs_set_block_group_flags(&cache->item, type); |
10365 | ||
920e4a58 | 10366 | cache->flags = type; |
11833d66 | 10367 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10368 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10369 | cache->needs_free_space = 1; |
835d974f JB |
10370 | ret = exclude_super_stripes(root, cache); |
10371 | if (ret) { | |
10372 | /* | |
10373 | * We may have excluded something, so call this just in | |
10374 | * case. | |
10375 | */ | |
10376 | free_excluded_extents(root, cache); | |
920e4a58 | 10377 | btrfs_put_block_group(cache); |
835d974f JB |
10378 | return ret; |
10379 | } | |
96303081 | 10380 | |
817d52f8 JB |
10381 | add_new_free_space(cache, root->fs_info, chunk_offset, |
10382 | chunk_offset + size); | |
10383 | ||
11833d66 YZ |
10384 | free_excluded_extents(root, cache); |
10385 | ||
d0bd4560 JB |
10386 | #ifdef CONFIG_BTRFS_DEBUG |
10387 | if (btrfs_should_fragment_free_space(root, cache)) { | |
10388 | u64 new_bytes_used = size - bytes_used; | |
10389 | ||
10390 | bytes_used += new_bytes_used >> 1; | |
10391 | fragment_free_space(root, cache); | |
10392 | } | |
10393 | #endif | |
2e6e5183 FM |
10394 | /* |
10395 | * Call to ensure the corresponding space_info object is created and | |
10396 | * assigned to our block group, but don't update its counters just yet. | |
10397 | * We want our bg to be added to the rbtree with its ->space_info set. | |
10398 | */ | |
e40edf2d | 10399 | ret = update_space_info(root->fs_info, cache->flags, 0, 0, 0, |
2e6e5183 FM |
10400 | &cache->space_info); |
10401 | if (ret) { | |
10402 | btrfs_remove_free_space_cache(cache); | |
10403 | btrfs_put_block_group(cache); | |
10404 | return ret; | |
10405 | } | |
10406 | ||
8c579fe7 JB |
10407 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
10408 | if (ret) { | |
10409 | btrfs_remove_free_space_cache(cache); | |
10410 | btrfs_put_block_group(cache); | |
10411 | return ret; | |
10412 | } | |
10413 | ||
2e6e5183 FM |
10414 | /* |
10415 | * Now that our block group has its ->space_info set and is inserted in | |
10416 | * the rbtree, update the space info's counters. | |
10417 | */ | |
c83f8eff | 10418 | trace_btrfs_add_block_group(root->fs_info, cache, 1); |
6324fbf3 | 10419 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10420 | cache->bytes_super, &cache->space_info); |
8c579fe7 JB |
10421 | if (ret) { |
10422 | btrfs_remove_free_space_cache(cache); | |
10423 | spin_lock(&root->fs_info->block_group_cache_lock); | |
10424 | rb_erase(&cache->cache_node, | |
10425 | &root->fs_info->block_group_cache_tree); | |
01eacb27 | 10426 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
10427 | spin_unlock(&root->fs_info->block_group_cache_lock); |
10428 | btrfs_put_block_group(cache); | |
10429 | return ret; | |
10430 | } | |
c7c144db | 10431 | update_global_block_rsv(root->fs_info); |
1b2da372 | 10432 | |
b742bb82 | 10433 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 10434 | |
47ab2a6c | 10435 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10436 | |
d18a2c44 | 10437 | set_avail_alloc_bits(extent_root->fs_info, type); |
6324fbf3 CM |
10438 | return 0; |
10439 | } | |
1a40e23b | 10440 | |
10ea00f5 ID |
10441 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10442 | { | |
899c81ea ID |
10443 | u64 extra_flags = chunk_to_extended(flags) & |
10444 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10445 | |
de98ced9 | 10446 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10447 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10448 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10449 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10450 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10451 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10452 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10453 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10454 | } |
10455 | ||
1a40e23b | 10456 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
04216820 FM |
10457 | struct btrfs_root *root, u64 group_start, |
10458 | struct extent_map *em) | |
1a40e23b ZY |
10459 | { |
10460 | struct btrfs_path *path; | |
10461 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10462 | struct btrfs_free_cluster *cluster; |
0af3d00b | 10463 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 10464 | struct btrfs_key key; |
0af3d00b | 10465 | struct inode *inode; |
c1895442 | 10466 | struct kobject *kobj = NULL; |
1a40e23b | 10467 | int ret; |
10ea00f5 | 10468 | int index; |
89a55897 | 10469 | int factor; |
4f69cb98 | 10470 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10471 | bool remove_em; |
1a40e23b | 10472 | |
1a40e23b ZY |
10473 | root = root->fs_info->extent_root; |
10474 | ||
10475 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
10476 | BUG_ON(!block_group); | |
c146afad | 10477 | BUG_ON(!block_group->ro); |
1a40e23b | 10478 | |
9f7c43c9 | 10479 | /* |
10480 | * Free the reserved super bytes from this block group before | |
10481 | * remove it. | |
10482 | */ | |
10483 | free_excluded_extents(root, block_group); | |
10484 | ||
1a40e23b | 10485 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 10486 | index = get_block_group_index(block_group); |
89a55897 JB |
10487 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10488 | BTRFS_BLOCK_GROUP_RAID1 | | |
10489 | BTRFS_BLOCK_GROUP_RAID10)) | |
10490 | factor = 2; | |
10491 | else | |
10492 | factor = 1; | |
1a40e23b | 10493 | |
44fb5511 CM |
10494 | /* make sure this block group isn't part of an allocation cluster */ |
10495 | cluster = &root->fs_info->data_alloc_cluster; | |
10496 | spin_lock(&cluster->refill_lock); | |
10497 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10498 | spin_unlock(&cluster->refill_lock); | |
10499 | ||
10500 | /* | |
10501 | * make sure this block group isn't part of a metadata | |
10502 | * allocation cluster | |
10503 | */ | |
10504 | cluster = &root->fs_info->meta_alloc_cluster; | |
10505 | spin_lock(&cluster->refill_lock); | |
10506 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10507 | spin_unlock(&cluster->refill_lock); | |
10508 | ||
1a40e23b | 10509 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10510 | if (!path) { |
10511 | ret = -ENOMEM; | |
10512 | goto out; | |
10513 | } | |
1a40e23b | 10514 | |
1bbc621e CM |
10515 | /* |
10516 | * get the inode first so any iput calls done for the io_list | |
10517 | * aren't the final iput (no unlinks allowed now) | |
10518 | */ | |
10b2f34d | 10519 | inode = lookup_free_space_inode(tree_root, block_group, path); |
1bbc621e CM |
10520 | |
10521 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10522 | /* | |
10523 | * make sure our free spache cache IO is done before remove the | |
10524 | * free space inode | |
10525 | */ | |
10526 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10527 | if (!list_empty(&block_group->io_list)) { | |
10528 | list_del_init(&block_group->io_list); | |
10529 | ||
10530 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10531 | ||
10532 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10533 | btrfs_wait_cache_io(root, trans, block_group, | |
10534 | &block_group->io_ctl, path, | |
10535 | block_group->key.objectid); | |
10536 | btrfs_put_block_group(block_group); | |
10537 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10538 | } | |
10539 | ||
10540 | if (!list_empty(&block_group->dirty_list)) { | |
10541 | list_del_init(&block_group->dirty_list); | |
10542 | btrfs_put_block_group(block_group); | |
10543 | } | |
10544 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10545 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10546 | ||
0af3d00b | 10547 | if (!IS_ERR(inode)) { |
b532402e | 10548 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
10549 | if (ret) { |
10550 | btrfs_add_delayed_iput(inode); | |
10551 | goto out; | |
10552 | } | |
0af3d00b JB |
10553 | clear_nlink(inode); |
10554 | /* One for the block groups ref */ | |
10555 | spin_lock(&block_group->lock); | |
10556 | if (block_group->iref) { | |
10557 | block_group->iref = 0; | |
10558 | block_group->inode = NULL; | |
10559 | spin_unlock(&block_group->lock); | |
10560 | iput(inode); | |
10561 | } else { | |
10562 | spin_unlock(&block_group->lock); | |
10563 | } | |
10564 | /* One for our lookup ref */ | |
455757c3 | 10565 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10566 | } |
10567 | ||
10568 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10569 | key.offset = block_group->key.objectid; | |
10570 | key.type = 0; | |
10571 | ||
10572 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10573 | if (ret < 0) | |
10574 | goto out; | |
10575 | if (ret > 0) | |
b3b4aa74 | 10576 | btrfs_release_path(path); |
0af3d00b JB |
10577 | if (ret == 0) { |
10578 | ret = btrfs_del_item(trans, tree_root, path); | |
10579 | if (ret) | |
10580 | goto out; | |
b3b4aa74 | 10581 | btrfs_release_path(path); |
0af3d00b JB |
10582 | } |
10583 | ||
3dfdb934 | 10584 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
10585 | rb_erase(&block_group->cache_node, |
10586 | &root->fs_info->block_group_cache_tree); | |
292cbd51 | 10587 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd LB |
10588 | |
10589 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
10590 | root->fs_info->first_logical_byte = (u64)-1; | |
3dfdb934 | 10591 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 10592 | |
80eb234a | 10593 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10594 | /* |
10595 | * we must use list_del_init so people can check to see if they | |
10596 | * are still on the list after taking the semaphore | |
10597 | */ | |
10598 | list_del_init(&block_group->list); | |
6ab0a202 | 10599 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10600 | kobj = block_group->space_info->block_group_kobjs[index]; |
10601 | block_group->space_info->block_group_kobjs[index] = NULL; | |
10ea00f5 | 10602 | clear_avail_alloc_bits(root->fs_info, block_group->flags); |
6ab0a202 | 10603 | } |
80eb234a | 10604 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10605 | if (kobj) { |
10606 | kobject_del(kobj); | |
10607 | kobject_put(kobj); | |
10608 | } | |
1a40e23b | 10609 | |
4f69cb98 FM |
10610 | if (block_group->has_caching_ctl) |
10611 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10612 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10613 | wait_block_group_cache_done(block_group); |
4f69cb98 FM |
10614 | if (block_group->has_caching_ctl) { |
10615 | down_write(&root->fs_info->commit_root_sem); | |
10616 | if (!caching_ctl) { | |
10617 | struct btrfs_caching_control *ctl; | |
10618 | ||
10619 | list_for_each_entry(ctl, | |
10620 | &root->fs_info->caching_block_groups, list) | |
10621 | if (ctl->block_group == block_group) { | |
10622 | caching_ctl = ctl; | |
10623 | atomic_inc(&caching_ctl->count); | |
10624 | break; | |
10625 | } | |
10626 | } | |
10627 | if (caching_ctl) | |
10628 | list_del_init(&caching_ctl->list); | |
10629 | up_write(&root->fs_info->commit_root_sem); | |
10630 | if (caching_ctl) { | |
10631 | /* Once for the caching bgs list and once for us. */ | |
10632 | put_caching_control(caching_ctl); | |
10633 | put_caching_control(caching_ctl); | |
10634 | } | |
10635 | } | |
817d52f8 | 10636 | |
ce93ec54 JB |
10637 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10638 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10639 | WARN_ON(1); |
10640 | } | |
10641 | if (!list_empty(&block_group->io_list)) { | |
10642 | WARN_ON(1); | |
ce93ec54 JB |
10643 | } |
10644 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10645 | btrfs_remove_free_space_cache(block_group); |
10646 | ||
c146afad | 10647 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10648 | list_del_init(&block_group->ro_list); |
18d018ad | 10649 | |
3cdde224 | 10650 | if (btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10651 | WARN_ON(block_group->space_info->total_bytes |
10652 | < block_group->key.offset); | |
10653 | WARN_ON(block_group->space_info->bytes_readonly | |
10654 | < block_group->key.offset); | |
10655 | WARN_ON(block_group->space_info->disk_total | |
10656 | < block_group->key.offset * factor); | |
10657 | } | |
c146afad YZ |
10658 | block_group->space_info->total_bytes -= block_group->key.offset; |
10659 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10660 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10661 | |
c146afad | 10662 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10663 | |
0af3d00b JB |
10664 | memcpy(&key, &block_group->key, sizeof(key)); |
10665 | ||
04216820 | 10666 | lock_chunks(root); |
495e64f4 FM |
10667 | if (!list_empty(&em->list)) { |
10668 | /* We're in the transaction->pending_chunks list. */ | |
10669 | free_extent_map(em); | |
10670 | } | |
04216820 FM |
10671 | spin_lock(&block_group->lock); |
10672 | block_group->removed = 1; | |
10673 | /* | |
10674 | * At this point trimming can't start on this block group, because we | |
10675 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10676 | * so no one can't find it anymore and even if someone already got this | |
10677 | * block group before we removed it from the rbtree, they have already | |
10678 | * incremented block_group->trimming - if they didn't, they won't find | |
10679 | * any free space entries because we already removed them all when we | |
10680 | * called btrfs_remove_free_space_cache(). | |
10681 | * | |
10682 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10683 | * to prevent the same logical address range and physical device space | |
10684 | * ranges from being reused for a new block group. This is because our | |
10685 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10686 | * completely transactionless, so while it is trimming a range the | |
10687 | * currently running transaction might finish and a new one start, | |
10688 | * allowing for new block groups to be created that can reuse the same | |
10689 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10690 | * |
10691 | * There may also be an implicit trim operation if the file system | |
10692 | * is mounted with -odiscard. The same protections must remain | |
10693 | * in place until the extents have been discarded completely when | |
10694 | * the transaction commit has completed. | |
04216820 FM |
10695 | */ |
10696 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10697 | /* | |
10698 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10699 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10700 | * before checking block_group->removed). | |
10701 | */ | |
10702 | if (!remove_em) { | |
10703 | /* | |
10704 | * Our em might be in trans->transaction->pending_chunks which | |
10705 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10706 | * and so is the fs_info->pinned_chunks list. | |
10707 | * | |
10708 | * So at this point we must be holding the chunk_mutex to avoid | |
10709 | * any races with chunk allocation (more specifically at | |
10710 | * volumes.c:contains_pending_extent()), to ensure it always | |
10711 | * sees the em, either in the pending_chunks list or in the | |
10712 | * pinned_chunks list. | |
10713 | */ | |
10714 | list_move_tail(&em->list, &root->fs_info->pinned_chunks); | |
10715 | } | |
10716 | spin_unlock(&block_group->lock); | |
04216820 FM |
10717 | |
10718 | if (remove_em) { | |
10719 | struct extent_map_tree *em_tree; | |
10720 | ||
10721 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
10722 | write_lock(&em_tree->lock); | |
8dbcd10f FM |
10723 | /* |
10724 | * The em might be in the pending_chunks list, so make sure the | |
10725 | * chunk mutex is locked, since remove_extent_mapping() will | |
10726 | * delete us from that list. | |
10727 | */ | |
04216820 FM |
10728 | remove_extent_mapping(em_tree, em); |
10729 | write_unlock(&em_tree->lock); | |
10730 | /* once for the tree */ | |
10731 | free_extent_map(em); | |
10732 | } | |
10733 | ||
8dbcd10f FM |
10734 | unlock_chunks(root); |
10735 | ||
1e144fb8 OS |
10736 | ret = remove_block_group_free_space(trans, root->fs_info, block_group); |
10737 | if (ret) | |
10738 | goto out; | |
10739 | ||
fa9c0d79 CM |
10740 | btrfs_put_block_group(block_group); |
10741 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10742 | |
10743 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10744 | if (ret > 0) | |
10745 | ret = -EIO; | |
10746 | if (ret < 0) | |
10747 | goto out; | |
10748 | ||
10749 | ret = btrfs_del_item(trans, root, path); | |
10750 | out: | |
10751 | btrfs_free_path(path); | |
10752 | return ret; | |
10753 | } | |
acce952b | 10754 | |
8eab77ff | 10755 | struct btrfs_trans_handle * |
7fd01182 FM |
10756 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10757 | const u64 chunk_offset) | |
8eab77ff | 10758 | { |
7fd01182 FM |
10759 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10760 | struct extent_map *em; | |
10761 | struct map_lookup *map; | |
10762 | unsigned int num_items; | |
10763 | ||
10764 | read_lock(&em_tree->lock); | |
10765 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10766 | read_unlock(&em_tree->lock); | |
10767 | ASSERT(em && em->start == chunk_offset); | |
10768 | ||
8eab77ff | 10769 | /* |
7fd01182 FM |
10770 | * We need to reserve 3 + N units from the metadata space info in order |
10771 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10772 | * btrfs_remove_block_group()), which are used for: | |
10773 | * | |
8eab77ff FM |
10774 | * 1 unit for adding the free space inode's orphan (located in the tree |
10775 | * of tree roots). | |
7fd01182 FM |
10776 | * 1 unit for deleting the block group item (located in the extent |
10777 | * tree). | |
10778 | * 1 unit for deleting the free space item (located in tree of tree | |
10779 | * roots). | |
10780 | * N units for deleting N device extent items corresponding to each | |
10781 | * stripe (located in the device tree). | |
10782 | * | |
10783 | * In order to remove a block group we also need to reserve units in the | |
10784 | * system space info in order to update the chunk tree (update one or | |
10785 | * more device items and remove one chunk item), but this is done at | |
10786 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10787 | */ |
95617d69 | 10788 | map = em->map_lookup; |
7fd01182 FM |
10789 | num_items = 3 + map->num_stripes; |
10790 | free_extent_map(em); | |
10791 | ||
8eab77ff | 10792 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10793 | num_items, 1); |
8eab77ff FM |
10794 | } |
10795 | ||
47ab2a6c JB |
10796 | /* |
10797 | * Process the unused_bgs list and remove any that don't have any allocated | |
10798 | * space inside of them. | |
10799 | */ | |
10800 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10801 | { | |
10802 | struct btrfs_block_group_cache *block_group; | |
10803 | struct btrfs_space_info *space_info; | |
10804 | struct btrfs_root *root = fs_info->extent_root; | |
10805 | struct btrfs_trans_handle *trans; | |
10806 | int ret = 0; | |
10807 | ||
afcdd129 | 10808 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10809 | return; |
10810 | ||
10811 | spin_lock(&fs_info->unused_bgs_lock); | |
10812 | while (!list_empty(&fs_info->unused_bgs)) { | |
10813 | u64 start, end; | |
e33e17ee | 10814 | int trimming; |
47ab2a6c JB |
10815 | |
10816 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10817 | struct btrfs_block_group_cache, | |
10818 | bg_list); | |
47ab2a6c | 10819 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10820 | |
10821 | space_info = block_group->space_info; | |
10822 | ||
47ab2a6c JB |
10823 | if (ret || btrfs_mixed_space_info(space_info)) { |
10824 | btrfs_put_block_group(block_group); | |
10825 | continue; | |
10826 | } | |
10827 | spin_unlock(&fs_info->unused_bgs_lock); | |
10828 | ||
d5f2e33b | 10829 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10830 | |
47ab2a6c JB |
10831 | /* Don't want to race with allocators so take the groups_sem */ |
10832 | down_write(&space_info->groups_sem); | |
10833 | spin_lock(&block_group->lock); | |
10834 | if (block_group->reserved || | |
10835 | btrfs_block_group_used(&block_group->item) || | |
5d8eb6fe | 10836 | (block_group->ro && !block_group->removed) || |
aefbe9a6 | 10837 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10838 | /* |
10839 | * We want to bail if we made new allocations or have | |
10840 | * outstanding allocations in this block group. We do | |
10841 | * the ro check in case balance is currently acting on | |
10842 | * this block group. | |
10843 | */ | |
10844 | spin_unlock(&block_group->lock); | |
10845 | up_write(&space_info->groups_sem); | |
10846 | goto next; | |
10847 | } | |
10848 | spin_unlock(&block_group->lock); | |
10849 | ||
10850 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10851 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10852 | up_write(&space_info->groups_sem); |
10853 | if (ret < 0) { | |
10854 | ret = 0; | |
10855 | goto next; | |
10856 | } | |
10857 | ||
10858 | /* | |
10859 | * Want to do this before we do anything else so we can recover | |
10860 | * properly if we fail to join the transaction. | |
10861 | */ | |
7fd01182 FM |
10862 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10863 | block_group->key.objectid); | |
47ab2a6c | 10864 | if (IS_ERR(trans)) { |
868f401a | 10865 | btrfs_dec_block_group_ro(root, block_group); |
47ab2a6c JB |
10866 | ret = PTR_ERR(trans); |
10867 | goto next; | |
10868 | } | |
10869 | ||
10870 | /* | |
10871 | * We could have pending pinned extents for this block group, | |
10872 | * just delete them, we don't care about them anymore. | |
10873 | */ | |
10874 | start = block_group->key.objectid; | |
10875 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10876 | /* |
10877 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10878 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10879 | * another task might be running finish_extent_commit() for the | |
10880 | * previous transaction N - 1, and have seen a range belonging | |
10881 | * to the block group in freed_extents[] before we were able to | |
10882 | * clear the whole block group range from freed_extents[]. This | |
10883 | * means that task can lookup for the block group after we | |
10884 | * unpinned it from freed_extents[] and removed it, leading to | |
10885 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10886 | */ | |
10887 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10888 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10889 | EXTENT_DIRTY); |
758eb51e | 10890 | if (ret) { |
d4b450cd | 10891 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
868f401a | 10892 | btrfs_dec_block_group_ro(root, block_group); |
758eb51e FM |
10893 | goto end_trans; |
10894 | } | |
10895 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10896 | EXTENT_DIRTY); |
758eb51e | 10897 | if (ret) { |
d4b450cd | 10898 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
868f401a | 10899 | btrfs_dec_block_group_ro(root, block_group); |
758eb51e FM |
10900 | goto end_trans; |
10901 | } | |
d4b450cd | 10902 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10903 | |
10904 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10905 | spin_lock(&space_info->lock); |
10906 | spin_lock(&block_group->lock); | |
10907 | ||
10908 | space_info->bytes_pinned -= block_group->pinned; | |
10909 | space_info->bytes_readonly += block_group->pinned; | |
10910 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10911 | -block_group->pinned); | |
47ab2a6c JB |
10912 | block_group->pinned = 0; |
10913 | ||
c30666d4 ZL |
10914 | spin_unlock(&block_group->lock); |
10915 | spin_unlock(&space_info->lock); | |
10916 | ||
e33e17ee | 10917 | /* DISCARD can flip during remount */ |
3cdde224 | 10918 | trimming = btrfs_test_opt(root->fs_info, DISCARD); |
e33e17ee JM |
10919 | |
10920 | /* Implicit trim during transaction commit. */ | |
10921 | if (trimming) | |
10922 | btrfs_get_block_group_trimming(block_group); | |
10923 | ||
47ab2a6c JB |
10924 | /* |
10925 | * Btrfs_remove_chunk will abort the transaction if things go | |
10926 | * horribly wrong. | |
10927 | */ | |
10928 | ret = btrfs_remove_chunk(trans, root, | |
10929 | block_group->key.objectid); | |
e33e17ee JM |
10930 | |
10931 | if (ret) { | |
10932 | if (trimming) | |
10933 | btrfs_put_block_group_trimming(block_group); | |
10934 | goto end_trans; | |
10935 | } | |
10936 | ||
10937 | /* | |
10938 | * If we're not mounted with -odiscard, we can just forget | |
10939 | * about this block group. Otherwise we'll need to wait | |
10940 | * until transaction commit to do the actual discard. | |
10941 | */ | |
10942 | if (trimming) { | |
348a0013 FM |
10943 | spin_lock(&fs_info->unused_bgs_lock); |
10944 | /* | |
10945 | * A concurrent scrub might have added us to the list | |
10946 | * fs_info->unused_bgs, so use a list_move operation | |
10947 | * to add the block group to the deleted_bgs list. | |
10948 | */ | |
e33e17ee JM |
10949 | list_move(&block_group->bg_list, |
10950 | &trans->transaction->deleted_bgs); | |
348a0013 | 10951 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10952 | btrfs_get_block_group(block_group); |
10953 | } | |
758eb51e | 10954 | end_trans: |
47ab2a6c JB |
10955 | btrfs_end_transaction(trans, root); |
10956 | next: | |
d5f2e33b | 10957 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10958 | btrfs_put_block_group(block_group); |
10959 | spin_lock(&fs_info->unused_bgs_lock); | |
10960 | } | |
10961 | spin_unlock(&fs_info->unused_bgs_lock); | |
10962 | } | |
10963 | ||
c59021f8 | 10964 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10965 | { | |
10966 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10967 | struct btrfs_super_block *disk_super; |
10968 | u64 features; | |
10969 | u64 flags; | |
10970 | int mixed = 0; | |
c59021f8 | 10971 | int ret; |
10972 | ||
6c41761f | 10973 | disk_super = fs_info->super_copy; |
1aba86d6 | 10974 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10975 | return -EINVAL; |
c59021f8 | 10976 | |
1aba86d6 | 10977 | features = btrfs_super_incompat_flags(disk_super); |
10978 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10979 | mixed = 1; | |
c59021f8 | 10980 | |
1aba86d6 | 10981 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
e40edf2d | 10982 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
c59021f8 | 10983 | if (ret) |
1aba86d6 | 10984 | goto out; |
c59021f8 | 10985 | |
1aba86d6 | 10986 | if (mixed) { |
10987 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
e40edf2d | 10988 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10989 | } else { |
10990 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
e40edf2d | 10991 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10992 | if (ret) |
10993 | goto out; | |
10994 | ||
10995 | flags = BTRFS_BLOCK_GROUP_DATA; | |
e40edf2d | 10996 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10997 | } |
10998 | out: | |
c59021f8 | 10999 | return ret; |
11000 | } | |
11001 | ||
acce952b | 11002 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
11003 | { | |
678886bd | 11004 | return unpin_extent_range(root, start, end, false); |
acce952b | 11005 | } |
11006 | ||
499f377f JM |
11007 | /* |
11008 | * It used to be that old block groups would be left around forever. | |
11009 | * Iterating over them would be enough to trim unused space. Since we | |
11010 | * now automatically remove them, we also need to iterate over unallocated | |
11011 | * space. | |
11012 | * | |
11013 | * We don't want a transaction for this since the discard may take a | |
11014 | * substantial amount of time. We don't require that a transaction be | |
11015 | * running, but we do need to take a running transaction into account | |
11016 | * to ensure that we're not discarding chunks that were released in | |
11017 | * the current transaction. | |
11018 | * | |
11019 | * Holding the chunks lock will prevent other threads from allocating | |
11020 | * or releasing chunks, but it won't prevent a running transaction | |
11021 | * from committing and releasing the memory that the pending chunks | |
11022 | * list head uses. For that, we need to take a reference to the | |
11023 | * transaction. | |
11024 | */ | |
11025 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
11026 | u64 minlen, u64 *trimmed) | |
11027 | { | |
11028 | u64 start = 0, len = 0; | |
11029 | int ret; | |
11030 | ||
11031 | *trimmed = 0; | |
11032 | ||
11033 | /* Not writeable = nothing to do. */ | |
11034 | if (!device->writeable) | |
11035 | return 0; | |
11036 | ||
11037 | /* No free space = nothing to do. */ | |
11038 | if (device->total_bytes <= device->bytes_used) | |
11039 | return 0; | |
11040 | ||
11041 | ret = 0; | |
11042 | ||
11043 | while (1) { | |
11044 | struct btrfs_fs_info *fs_info = device->dev_root->fs_info; | |
11045 | struct btrfs_transaction *trans; | |
11046 | u64 bytes; | |
11047 | ||
11048 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
11049 | if (ret) | |
11050 | return ret; | |
11051 | ||
11052 | down_read(&fs_info->commit_root_sem); | |
11053 | ||
11054 | spin_lock(&fs_info->trans_lock); | |
11055 | trans = fs_info->running_transaction; | |
11056 | if (trans) | |
11057 | atomic_inc(&trans->use_count); | |
11058 | spin_unlock(&fs_info->trans_lock); | |
11059 | ||
11060 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
11061 | &start, &len); | |
11062 | if (trans) | |
11063 | btrfs_put_transaction(trans); | |
11064 | ||
11065 | if (ret) { | |
11066 | up_read(&fs_info->commit_root_sem); | |
11067 | mutex_unlock(&fs_info->chunk_mutex); | |
11068 | if (ret == -ENOSPC) | |
11069 | ret = 0; | |
11070 | break; | |
11071 | } | |
11072 | ||
11073 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
11074 | up_read(&fs_info->commit_root_sem); | |
11075 | mutex_unlock(&fs_info->chunk_mutex); | |
11076 | ||
11077 | if (ret) | |
11078 | break; | |
11079 | ||
11080 | start += len; | |
11081 | *trimmed += bytes; | |
11082 | ||
11083 | if (fatal_signal_pending(current)) { | |
11084 | ret = -ERESTARTSYS; | |
11085 | break; | |
11086 | } | |
11087 | ||
11088 | cond_resched(); | |
11089 | } | |
11090 | ||
11091 | return ret; | |
11092 | } | |
11093 | ||
f7039b1d LD |
11094 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) |
11095 | { | |
11096 | struct btrfs_fs_info *fs_info = root->fs_info; | |
11097 | struct btrfs_block_group_cache *cache = NULL; | |
499f377f JM |
11098 | struct btrfs_device *device; |
11099 | struct list_head *devices; | |
f7039b1d LD |
11100 | u64 group_trimmed; |
11101 | u64 start; | |
11102 | u64 end; | |
11103 | u64 trimmed = 0; | |
2cac13e4 | 11104 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
11105 | int ret = 0; |
11106 | ||
2cac13e4 LB |
11107 | /* |
11108 | * try to trim all FS space, our block group may start from non-zero. | |
11109 | */ | |
11110 | if (range->len == total_bytes) | |
11111 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
11112 | else | |
11113 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
11114 | |
11115 | while (cache) { | |
11116 | if (cache->key.objectid >= (range->start + range->len)) { | |
11117 | btrfs_put_block_group(cache); | |
11118 | break; | |
11119 | } | |
11120 | ||
11121 | start = max(range->start, cache->key.objectid); | |
11122 | end = min(range->start + range->len, | |
11123 | cache->key.objectid + cache->key.offset); | |
11124 | ||
11125 | if (end - start >= range->minlen) { | |
11126 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 11127 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
11128 | if (ret) { |
11129 | btrfs_put_block_group(cache); | |
11130 | break; | |
11131 | } | |
11132 | ret = wait_block_group_cache_done(cache); | |
11133 | if (ret) { | |
11134 | btrfs_put_block_group(cache); | |
11135 | break; | |
11136 | } | |
f7039b1d LD |
11137 | } |
11138 | ret = btrfs_trim_block_group(cache, | |
11139 | &group_trimmed, | |
11140 | start, | |
11141 | end, | |
11142 | range->minlen); | |
11143 | ||
11144 | trimmed += group_trimmed; | |
11145 | if (ret) { | |
11146 | btrfs_put_block_group(cache); | |
11147 | break; | |
11148 | } | |
11149 | } | |
11150 | ||
11151 | cache = next_block_group(fs_info->tree_root, cache); | |
11152 | } | |
11153 | ||
499f377f JM |
11154 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
11155 | devices = &root->fs_info->fs_devices->alloc_list; | |
11156 | list_for_each_entry(device, devices, dev_alloc_list) { | |
11157 | ret = btrfs_trim_free_extents(device, range->minlen, | |
11158 | &group_trimmed); | |
11159 | if (ret) | |
11160 | break; | |
11161 | ||
11162 | trimmed += group_trimmed; | |
11163 | } | |
11164 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
11165 | ||
f7039b1d LD |
11166 | range->len = trimmed; |
11167 | return ret; | |
11168 | } | |
8257b2dc MX |
11169 | |
11170 | /* | |
9ea24bbe FM |
11171 | * btrfs_{start,end}_write_no_snapshoting() are similar to |
11172 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing | |
11173 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11174 | * but flush the data into disk after the snapshot creation, or to prevent | |
11175 | * operations while snapshoting is ongoing and that cause the snapshot to be | |
11176 | * inconsistent (writes followed by expanding truncates for example). | |
8257b2dc | 11177 | */ |
9ea24bbe | 11178 | void btrfs_end_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc MX |
11179 | { |
11180 | percpu_counter_dec(&root->subv_writers->counter); | |
11181 | /* | |
a83342aa | 11182 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
11183 | */ |
11184 | smp_mb(); | |
11185 | if (waitqueue_active(&root->subv_writers->wait)) | |
11186 | wake_up(&root->subv_writers->wait); | |
11187 | } | |
11188 | ||
9ea24bbe | 11189 | int btrfs_start_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc | 11190 | { |
ee39b432 | 11191 | if (atomic_read(&root->will_be_snapshoted)) |
8257b2dc MX |
11192 | return 0; |
11193 | ||
11194 | percpu_counter_inc(&root->subv_writers->counter); | |
11195 | /* | |
11196 | * Make sure counter is updated before we check for snapshot creation. | |
11197 | */ | |
11198 | smp_mb(); | |
ee39b432 | 11199 | if (atomic_read(&root->will_be_snapshoted)) { |
9ea24bbe | 11200 | btrfs_end_write_no_snapshoting(root); |
8257b2dc MX |
11201 | return 0; |
11202 | } | |
11203 | return 1; | |
11204 | } | |
0bc19f90 ZL |
11205 | |
11206 | static int wait_snapshoting_atomic_t(atomic_t *a) | |
11207 | { | |
11208 | schedule(); | |
11209 | return 0; | |
11210 | } | |
11211 | ||
11212 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) | |
11213 | { | |
11214 | while (true) { | |
11215 | int ret; | |
11216 | ||
11217 | ret = btrfs_start_write_no_snapshoting(root); | |
11218 | if (ret) | |
11219 | break; | |
11220 | wait_on_atomic_t(&root->will_be_snapshoted, | |
11221 | wait_snapshoting_atomic_t, | |
11222 | TASK_UNINTERRUPTIBLE); | |
11223 | } | |
11224 | } |