]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
f361bf4a | 19 | #include <linux/sched/signal.h> |
edbd8d4e | 20 | #include <linux/pagemap.h> |
ec44a35c | 21 | #include <linux/writeback.h> |
21af804c | 22 | #include <linux/blkdev.h> |
b7a9f29f | 23 | #include <linux/sort.h> |
4184ea7f | 24 | #include <linux/rcupdate.h> |
817d52f8 | 25 | #include <linux/kthread.h> |
5a0e3ad6 | 26 | #include <linux/slab.h> |
dff51cd1 | 27 | #include <linux/ratelimit.h> |
b150a4f1 | 28 | #include <linux/percpu_counter.h> |
74493f7a | 29 | #include "hash.h" |
995946dd | 30 | #include "tree-log.h" |
fec577fb CM |
31 | #include "disk-io.h" |
32 | #include "print-tree.h" | |
0b86a832 | 33 | #include "volumes.h" |
53b381b3 | 34 | #include "raid56.h" |
925baedd | 35 | #include "locking.h" |
fa9c0d79 | 36 | #include "free-space-cache.h" |
1e144fb8 | 37 | #include "free-space-tree.h" |
3fed40cc | 38 | #include "math.h" |
6ab0a202 | 39 | #include "sysfs.h" |
fcebe456 | 40 | #include "qgroup.h" |
fec577fb | 41 | |
709c0486 AJ |
42 | #undef SCRAMBLE_DELAYED_REFS |
43 | ||
9e622d6b MX |
44 | /* |
45 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
46 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
47 | * if we really need one. | |
48 | * | |
0e4f8f88 CM |
49 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
50 | * if we have very few chunks already allocated. This is | |
51 | * used as part of the clustering code to help make sure | |
52 | * we have a good pool of storage to cluster in, without | |
53 | * filling the FS with empty chunks | |
54 | * | |
9e622d6b MX |
55 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
56 | * | |
0e4f8f88 CM |
57 | */ |
58 | enum { | |
59 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
60 | CHUNK_ALLOC_LIMITED = 1, |
61 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
62 | }; |
63 | ||
ce93ec54 | 64 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 65 | struct btrfs_fs_info *fs_info, u64 bytenr, |
ce93ec54 | 66 | u64 num_bytes, int alloc); |
5d4f98a2 | 67 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 68 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 69 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
70 | u64 root_objectid, u64 owner_objectid, |
71 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 72 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
73 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
74 | struct extent_buffer *leaf, | |
75 | struct btrfs_extent_item *ei); | |
76 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
2ff7e61e | 77 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
78 | u64 parent, u64 root_objectid, |
79 | u64 flags, u64 owner, u64 offset, | |
80 | struct btrfs_key *ins, int ref_mod); | |
81 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
2ff7e61e | 82 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
83 | u64 parent, u64 root_objectid, |
84 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 85 | int level, struct btrfs_key *ins); |
6a63209f | 86 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 87 | struct btrfs_fs_info *fs_info, u64 flags, |
698d0082 | 88 | int force); |
11833d66 YZ |
89 | static int find_next_key(struct btrfs_path *path, int level, |
90 | struct btrfs_key *key); | |
ab8d0fc4 JM |
91 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
92 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 93 | int dump_block_groups); |
4824f1f4 | 94 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 95 | u64 ram_bytes, u64 num_bytes, int delalloc); |
4824f1f4 WX |
96 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
97 | u64 num_bytes, int delalloc); | |
5d80366e JB |
98 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
99 | u64 num_bytes); | |
c1c4919b | 100 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
101 | struct btrfs_space_info *space_info, |
102 | u64 orig_bytes, | |
c1c4919b JM |
103 | enum btrfs_reserve_flush_enum flush, |
104 | bool system_chunk); | |
957780eb JB |
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); | |
0966a7b1 QW |
135 | |
136 | /* | |
137 | * If not empty, someone is still holding mutex of | |
138 | * full_stripe_lock, which can only be released by caller. | |
139 | * And it will definitely cause use-after-free when caller | |
140 | * tries to release full stripe lock. | |
141 | * | |
142 | * No better way to resolve, but only to warn. | |
143 | */ | |
144 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 145 | kfree(cache->free_space_ctl); |
11dfe35a | 146 | kfree(cache); |
f0486c68 | 147 | } |
11dfe35a JB |
148 | } |
149 | ||
0f9dd46c JB |
150 | /* |
151 | * this adds the block group to the fs_info rb tree for the block group | |
152 | * cache | |
153 | */ | |
b2950863 | 154 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
155 | struct btrfs_block_group_cache *block_group) |
156 | { | |
157 | struct rb_node **p; | |
158 | struct rb_node *parent = NULL; | |
159 | struct btrfs_block_group_cache *cache; | |
160 | ||
161 | spin_lock(&info->block_group_cache_lock); | |
162 | p = &info->block_group_cache_tree.rb_node; | |
163 | ||
164 | while (*p) { | |
165 | parent = *p; | |
166 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
167 | cache_node); | |
168 | if (block_group->key.objectid < cache->key.objectid) { | |
169 | p = &(*p)->rb_left; | |
170 | } else if (block_group->key.objectid > cache->key.objectid) { | |
171 | p = &(*p)->rb_right; | |
172 | } else { | |
173 | spin_unlock(&info->block_group_cache_lock); | |
174 | return -EEXIST; | |
175 | } | |
176 | } | |
177 | ||
178 | rb_link_node(&block_group->cache_node, parent, p); | |
179 | rb_insert_color(&block_group->cache_node, | |
180 | &info->block_group_cache_tree); | |
a1897fdd LB |
181 | |
182 | if (info->first_logical_byte > block_group->key.objectid) | |
183 | info->first_logical_byte = block_group->key.objectid; | |
184 | ||
0f9dd46c JB |
185 | spin_unlock(&info->block_group_cache_lock); |
186 | ||
187 | return 0; | |
188 | } | |
189 | ||
190 | /* | |
191 | * This will return the block group at or after bytenr if contains is 0, else | |
192 | * it will return the block group that contains the bytenr | |
193 | */ | |
194 | static struct btrfs_block_group_cache * | |
195 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
196 | int contains) | |
197 | { | |
198 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
199 | struct rb_node *n; | |
200 | u64 end, start; | |
201 | ||
202 | spin_lock(&info->block_group_cache_lock); | |
203 | n = info->block_group_cache_tree.rb_node; | |
204 | ||
205 | while (n) { | |
206 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
207 | cache_node); | |
208 | end = cache->key.objectid + cache->key.offset - 1; | |
209 | start = cache->key.objectid; | |
210 | ||
211 | if (bytenr < start) { | |
212 | if (!contains && (!ret || start < ret->key.objectid)) | |
213 | ret = cache; | |
214 | n = n->rb_left; | |
215 | } else if (bytenr > start) { | |
216 | if (contains && bytenr <= end) { | |
217 | ret = cache; | |
218 | break; | |
219 | } | |
220 | n = n->rb_right; | |
221 | } else { | |
222 | ret = cache; | |
223 | break; | |
224 | } | |
225 | } | |
a1897fdd | 226 | if (ret) { |
11dfe35a | 227 | btrfs_get_block_group(ret); |
a1897fdd LB |
228 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
229 | info->first_logical_byte = ret->key.objectid; | |
230 | } | |
0f9dd46c JB |
231 | spin_unlock(&info->block_group_cache_lock); |
232 | ||
233 | return ret; | |
234 | } | |
235 | ||
2ff7e61e | 236 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 237 | u64 start, u64 num_bytes) |
817d52f8 | 238 | { |
11833d66 | 239 | u64 end = start + num_bytes - 1; |
0b246afa | 240 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 241 | start, end, EXTENT_UPTODATE); |
0b246afa | 242 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 243 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
244 | return 0; |
245 | } | |
817d52f8 | 246 | |
2ff7e61e | 247 | static void free_excluded_extents(struct btrfs_fs_info *fs_info, |
11833d66 YZ |
248 | struct btrfs_block_group_cache *cache) |
249 | { | |
250 | u64 start, end; | |
817d52f8 | 251 | |
11833d66 YZ |
252 | start = cache->key.objectid; |
253 | end = start + cache->key.offset - 1; | |
254 | ||
0b246afa | 255 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 256 | start, end, EXTENT_UPTODATE); |
0b246afa | 257 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 258 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
259 | } |
260 | ||
2ff7e61e | 261 | static int exclude_super_stripes(struct btrfs_fs_info *fs_info, |
11833d66 | 262 | struct btrfs_block_group_cache *cache) |
817d52f8 | 263 | { |
817d52f8 JB |
264 | u64 bytenr; |
265 | u64 *logical; | |
266 | int stripe_len; | |
267 | int i, nr, ret; | |
268 | ||
06b2331f YZ |
269 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
270 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
271 | cache->bytes_super += stripe_len; | |
2ff7e61e | 272 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 273 | stripe_len); |
835d974f JB |
274 | if (ret) |
275 | return ret; | |
06b2331f YZ |
276 | } |
277 | ||
817d52f8 JB |
278 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
279 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 280 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
ab8d0fc4 | 281 | bytenr, 0, &logical, &nr, &stripe_len); |
835d974f JB |
282 | if (ret) |
283 | return ret; | |
11833d66 | 284 | |
817d52f8 | 285 | while (nr--) { |
51bf5f0b JB |
286 | u64 start, len; |
287 | ||
288 | if (logical[nr] > cache->key.objectid + | |
289 | cache->key.offset) | |
290 | continue; | |
291 | ||
292 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
293 | continue; | |
294 | ||
295 | start = logical[nr]; | |
296 | if (start < cache->key.objectid) { | |
297 | start = cache->key.objectid; | |
298 | len = (logical[nr] + stripe_len) - start; | |
299 | } else { | |
300 | len = min_t(u64, stripe_len, | |
301 | cache->key.objectid + | |
302 | cache->key.offset - start); | |
303 | } | |
304 | ||
305 | cache->bytes_super += len; | |
2ff7e61e | 306 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
307 | if (ret) { |
308 | kfree(logical); | |
309 | return ret; | |
310 | } | |
817d52f8 | 311 | } |
11833d66 | 312 | |
817d52f8 JB |
313 | kfree(logical); |
314 | } | |
817d52f8 JB |
315 | return 0; |
316 | } | |
317 | ||
11833d66 YZ |
318 | static struct btrfs_caching_control * |
319 | get_caching_control(struct btrfs_block_group_cache *cache) | |
320 | { | |
321 | struct btrfs_caching_control *ctl; | |
322 | ||
323 | spin_lock(&cache->lock); | |
dde5abee JB |
324 | if (!cache->caching_ctl) { |
325 | spin_unlock(&cache->lock); | |
11833d66 YZ |
326 | return NULL; |
327 | } | |
328 | ||
329 | ctl = cache->caching_ctl; | |
1e4f4714 | 330 | refcount_inc(&ctl->count); |
11833d66 YZ |
331 | spin_unlock(&cache->lock); |
332 | return ctl; | |
333 | } | |
334 | ||
335 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
336 | { | |
1e4f4714 | 337 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
338 | kfree(ctl); |
339 | } | |
340 | ||
d0bd4560 | 341 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 342 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 343 | { |
2ff7e61e | 344 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
345 | u64 start = block_group->key.objectid; |
346 | u64 len = block_group->key.offset; | |
347 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 348 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
349 | u64 step = chunk << 1; |
350 | ||
351 | while (len > chunk) { | |
352 | btrfs_remove_free_space(block_group, start, chunk); | |
353 | start += step; | |
354 | if (len < step) | |
355 | len = 0; | |
356 | else | |
357 | len -= step; | |
358 | } | |
359 | } | |
360 | #endif | |
361 | ||
0f9dd46c JB |
362 | /* |
363 | * this is only called by cache_block_group, since we could have freed extents | |
364 | * we need to check the pinned_extents for any extents that can't be used yet | |
365 | * since their free space will be released as soon as the transaction commits. | |
366 | */ | |
a5ed9182 OS |
367 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
368 | struct btrfs_fs_info *info, u64 start, u64 end) | |
0f9dd46c | 369 | { |
817d52f8 | 370 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
371 | int ret; |
372 | ||
373 | while (start < end) { | |
11833d66 | 374 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 375 | &extent_start, &extent_end, |
e6138876 JB |
376 | EXTENT_DIRTY | EXTENT_UPTODATE, |
377 | NULL); | |
0f9dd46c JB |
378 | if (ret) |
379 | break; | |
380 | ||
06b2331f | 381 | if (extent_start <= start) { |
0f9dd46c JB |
382 | start = extent_end + 1; |
383 | } else if (extent_start > start && extent_start < end) { | |
384 | size = extent_start - start; | |
817d52f8 | 385 | total_added += size; |
ea6a478e JB |
386 | ret = btrfs_add_free_space(block_group, start, |
387 | size); | |
79787eaa | 388 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
389 | start = extent_end + 1; |
390 | } else { | |
391 | break; | |
392 | } | |
393 | } | |
394 | ||
395 | if (start < end) { | |
396 | size = end - start; | |
817d52f8 | 397 | total_added += size; |
ea6a478e | 398 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 399 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
400 | } |
401 | ||
817d52f8 | 402 | return total_added; |
0f9dd46c JB |
403 | } |
404 | ||
73fa48b6 | 405 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 406 | { |
0b246afa JM |
407 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
408 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
409 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 410 | struct btrfs_path *path; |
5f39d397 | 411 | struct extent_buffer *leaf; |
11833d66 | 412 | struct btrfs_key key; |
817d52f8 | 413 | u64 total_found = 0; |
11833d66 YZ |
414 | u64 last = 0; |
415 | u32 nritems; | |
73fa48b6 | 416 | int ret; |
d0bd4560 | 417 | bool wakeup = true; |
f510cfec | 418 | |
e37c9e69 CM |
419 | path = btrfs_alloc_path(); |
420 | if (!path) | |
73fa48b6 | 421 | return -ENOMEM; |
7d7d6068 | 422 | |
817d52f8 | 423 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 424 | |
d0bd4560 JB |
425 | #ifdef CONFIG_BTRFS_DEBUG |
426 | /* | |
427 | * If we're fragmenting we don't want to make anybody think we can | |
428 | * allocate from this block group until we've had a chance to fragment | |
429 | * the free space. | |
430 | */ | |
2ff7e61e | 431 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
432 | wakeup = false; |
433 | #endif | |
5cd57b2c | 434 | /* |
817d52f8 JB |
435 | * We don't want to deadlock with somebody trying to allocate a new |
436 | * extent for the extent root while also trying to search the extent | |
437 | * root to add free space. So we skip locking and search the commit | |
438 | * root, since its read-only | |
5cd57b2c CM |
439 | */ |
440 | path->skip_locking = 1; | |
817d52f8 | 441 | path->search_commit_root = 1; |
e4058b54 | 442 | path->reada = READA_FORWARD; |
817d52f8 | 443 | |
e4404d6e | 444 | key.objectid = last; |
e37c9e69 | 445 | key.offset = 0; |
11833d66 | 446 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 447 | |
52ee28d2 | 448 | next: |
11833d66 | 449 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 450 | if (ret < 0) |
73fa48b6 | 451 | goto out; |
a512bbf8 | 452 | |
11833d66 YZ |
453 | leaf = path->nodes[0]; |
454 | nritems = btrfs_header_nritems(leaf); | |
455 | ||
d397712b | 456 | while (1) { |
7841cb28 | 457 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 458 | last = (u64)-1; |
817d52f8 | 459 | break; |
f25784b3 | 460 | } |
817d52f8 | 461 | |
11833d66 YZ |
462 | if (path->slots[0] < nritems) { |
463 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
464 | } else { | |
465 | ret = find_next_key(path, 0, &key); | |
466 | if (ret) | |
e37c9e69 | 467 | break; |
817d52f8 | 468 | |
c9ea7b24 | 469 | if (need_resched() || |
9e351cc8 | 470 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
471 | if (wakeup) |
472 | caching_ctl->progress = last; | |
ff5714cc | 473 | btrfs_release_path(path); |
9e351cc8 | 474 | up_read(&fs_info->commit_root_sem); |
589d8ade | 475 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 476 | cond_resched(); |
73fa48b6 OS |
477 | mutex_lock(&caching_ctl->mutex); |
478 | down_read(&fs_info->commit_root_sem); | |
479 | goto next; | |
589d8ade | 480 | } |
0a3896d0 JB |
481 | |
482 | ret = btrfs_next_leaf(extent_root, path); | |
483 | if (ret < 0) | |
73fa48b6 | 484 | goto out; |
0a3896d0 JB |
485 | if (ret) |
486 | break; | |
589d8ade JB |
487 | leaf = path->nodes[0]; |
488 | nritems = btrfs_header_nritems(leaf); | |
489 | continue; | |
11833d66 | 490 | } |
817d52f8 | 491 | |
52ee28d2 LB |
492 | if (key.objectid < last) { |
493 | key.objectid = last; | |
494 | key.offset = 0; | |
495 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
496 | ||
d0bd4560 JB |
497 | if (wakeup) |
498 | caching_ctl->progress = last; | |
52ee28d2 LB |
499 | btrfs_release_path(path); |
500 | goto next; | |
501 | } | |
502 | ||
11833d66 YZ |
503 | if (key.objectid < block_group->key.objectid) { |
504 | path->slots[0]++; | |
817d52f8 | 505 | continue; |
e37c9e69 | 506 | } |
0f9dd46c | 507 | |
e37c9e69 | 508 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 509 | block_group->key.offset) |
e37c9e69 | 510 | break; |
7d7d6068 | 511 | |
3173a18f JB |
512 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
513 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
514 | total_found += add_new_free_space(block_group, |
515 | fs_info, last, | |
516 | key.objectid); | |
3173a18f JB |
517 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
518 | last = key.objectid + | |
da17066c | 519 | fs_info->nodesize; |
3173a18f JB |
520 | else |
521 | last = key.objectid + key.offset; | |
817d52f8 | 522 | |
73fa48b6 | 523 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 524 | total_found = 0; |
d0bd4560 JB |
525 | if (wakeup) |
526 | wake_up(&caching_ctl->wait); | |
11833d66 | 527 | } |
817d52f8 | 528 | } |
e37c9e69 CM |
529 | path->slots[0]++; |
530 | } | |
817d52f8 | 531 | ret = 0; |
e37c9e69 | 532 | |
817d52f8 JB |
533 | total_found += add_new_free_space(block_group, fs_info, last, |
534 | block_group->key.objectid + | |
535 | block_group->key.offset); | |
11833d66 | 536 | caching_ctl->progress = (u64)-1; |
817d52f8 | 537 | |
73fa48b6 OS |
538 | out: |
539 | btrfs_free_path(path); | |
540 | return ret; | |
541 | } | |
542 | ||
543 | static noinline void caching_thread(struct btrfs_work *work) | |
544 | { | |
545 | struct btrfs_block_group_cache *block_group; | |
546 | struct btrfs_fs_info *fs_info; | |
547 | struct btrfs_caching_control *caching_ctl; | |
b4570aa9 | 548 | struct btrfs_root *extent_root; |
73fa48b6 OS |
549 | int ret; |
550 | ||
551 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
552 | block_group = caching_ctl->block_group; | |
553 | fs_info = block_group->fs_info; | |
b4570aa9 | 554 | extent_root = fs_info->extent_root; |
73fa48b6 OS |
555 | |
556 | mutex_lock(&caching_ctl->mutex); | |
557 | down_read(&fs_info->commit_root_sem); | |
558 | ||
1e144fb8 OS |
559 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
560 | ret = load_free_space_tree(caching_ctl); | |
561 | else | |
562 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 563 | |
817d52f8 | 564 | spin_lock(&block_group->lock); |
11833d66 | 565 | block_group->caching_ctl = NULL; |
73fa48b6 | 566 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 567 | spin_unlock(&block_group->lock); |
0f9dd46c | 568 | |
d0bd4560 | 569 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 570 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
571 | u64 bytes_used; |
572 | ||
573 | spin_lock(&block_group->space_info->lock); | |
574 | spin_lock(&block_group->lock); | |
575 | bytes_used = block_group->key.offset - | |
576 | btrfs_block_group_used(&block_group->item); | |
577 | block_group->space_info->bytes_used += bytes_used >> 1; | |
578 | spin_unlock(&block_group->lock); | |
579 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 580 | fragment_free_space(block_group); |
d0bd4560 JB |
581 | } |
582 | #endif | |
583 | ||
584 | caching_ctl->progress = (u64)-1; | |
11833d66 | 585 | |
9e351cc8 | 586 | up_read(&fs_info->commit_root_sem); |
2ff7e61e | 587 | free_excluded_extents(fs_info, block_group); |
11833d66 | 588 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 589 | |
11833d66 YZ |
590 | wake_up(&caching_ctl->wait); |
591 | ||
592 | put_caching_control(caching_ctl); | |
11dfe35a | 593 | btrfs_put_block_group(block_group); |
817d52f8 JB |
594 | } |
595 | ||
9d66e233 | 596 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 597 | int load_cache_only) |
817d52f8 | 598 | { |
291c7d2f | 599 | DEFINE_WAIT(wait); |
11833d66 YZ |
600 | struct btrfs_fs_info *fs_info = cache->fs_info; |
601 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
602 | int ret = 0; |
603 | ||
291c7d2f | 604 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
605 | if (!caching_ctl) |
606 | return -ENOMEM; | |
291c7d2f JB |
607 | |
608 | INIT_LIST_HEAD(&caching_ctl->list); | |
609 | mutex_init(&caching_ctl->mutex); | |
610 | init_waitqueue_head(&caching_ctl->wait); | |
611 | caching_ctl->block_group = cache; | |
612 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 613 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
614 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
615 | caching_thread, NULL, NULL); | |
291c7d2f JB |
616 | |
617 | spin_lock(&cache->lock); | |
618 | /* | |
619 | * This should be a rare occasion, but this could happen I think in the | |
620 | * case where one thread starts to load the space cache info, and then | |
621 | * some other thread starts a transaction commit which tries to do an | |
622 | * allocation while the other thread is still loading the space cache | |
623 | * info. The previous loop should have kept us from choosing this block | |
624 | * group, but if we've moved to the state where we will wait on caching | |
625 | * block groups we need to first check if we're doing a fast load here, | |
626 | * so we can wait for it to finish, otherwise we could end up allocating | |
627 | * from a block group who's cache gets evicted for one reason or | |
628 | * another. | |
629 | */ | |
630 | while (cache->cached == BTRFS_CACHE_FAST) { | |
631 | struct btrfs_caching_control *ctl; | |
632 | ||
633 | ctl = cache->caching_ctl; | |
1e4f4714 | 634 | refcount_inc(&ctl->count); |
291c7d2f JB |
635 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
636 | spin_unlock(&cache->lock); | |
637 | ||
638 | schedule(); | |
639 | ||
640 | finish_wait(&ctl->wait, &wait); | |
641 | put_caching_control(ctl); | |
642 | spin_lock(&cache->lock); | |
643 | } | |
644 | ||
645 | if (cache->cached != BTRFS_CACHE_NO) { | |
646 | spin_unlock(&cache->lock); | |
647 | kfree(caching_ctl); | |
11833d66 | 648 | return 0; |
291c7d2f JB |
649 | } |
650 | WARN_ON(cache->caching_ctl); | |
651 | cache->caching_ctl = caching_ctl; | |
652 | cache->cached = BTRFS_CACHE_FAST; | |
653 | spin_unlock(&cache->lock); | |
11833d66 | 654 | |
d53ba474 | 655 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
cb83b7b8 | 656 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
657 | ret = load_free_space_cache(fs_info, cache); |
658 | ||
659 | spin_lock(&cache->lock); | |
660 | if (ret == 1) { | |
291c7d2f | 661 | cache->caching_ctl = NULL; |
9d66e233 JB |
662 | cache->cached = BTRFS_CACHE_FINISHED; |
663 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 664 | caching_ctl->progress = (u64)-1; |
9d66e233 | 665 | } else { |
291c7d2f JB |
666 | if (load_cache_only) { |
667 | cache->caching_ctl = NULL; | |
668 | cache->cached = BTRFS_CACHE_NO; | |
669 | } else { | |
670 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 671 | cache->has_caching_ctl = 1; |
291c7d2f | 672 | } |
9d66e233 JB |
673 | } |
674 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
675 | #ifdef CONFIG_BTRFS_DEBUG |
676 | if (ret == 1 && | |
2ff7e61e | 677 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
678 | u64 bytes_used; |
679 | ||
680 | spin_lock(&cache->space_info->lock); | |
681 | spin_lock(&cache->lock); | |
682 | bytes_used = cache->key.offset - | |
683 | btrfs_block_group_used(&cache->item); | |
684 | cache->space_info->bytes_used += bytes_used >> 1; | |
685 | spin_unlock(&cache->lock); | |
686 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 687 | fragment_free_space(cache); |
d0bd4560 JB |
688 | } |
689 | #endif | |
cb83b7b8 JB |
690 | mutex_unlock(&caching_ctl->mutex); |
691 | ||
291c7d2f | 692 | wake_up(&caching_ctl->wait); |
3c14874a | 693 | if (ret == 1) { |
291c7d2f | 694 | put_caching_control(caching_ctl); |
2ff7e61e | 695 | free_excluded_extents(fs_info, cache); |
9d66e233 | 696 | return 0; |
3c14874a | 697 | } |
291c7d2f JB |
698 | } else { |
699 | /* | |
1e144fb8 OS |
700 | * We're either using the free space tree or no caching at all. |
701 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
702 | */ |
703 | spin_lock(&cache->lock); | |
704 | if (load_cache_only) { | |
705 | cache->caching_ctl = NULL; | |
706 | cache->cached = BTRFS_CACHE_NO; | |
707 | } else { | |
708 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 709 | cache->has_caching_ctl = 1; |
291c7d2f JB |
710 | } |
711 | spin_unlock(&cache->lock); | |
712 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
713 | } |
714 | ||
291c7d2f JB |
715 | if (load_cache_only) { |
716 | put_caching_control(caching_ctl); | |
11833d66 | 717 | return 0; |
817d52f8 | 718 | } |
817d52f8 | 719 | |
9e351cc8 | 720 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 721 | refcount_inc(&caching_ctl->count); |
11833d66 | 722 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 723 | up_write(&fs_info->commit_root_sem); |
11833d66 | 724 | |
11dfe35a | 725 | btrfs_get_block_group(cache); |
11833d66 | 726 | |
e66f0bb1 | 727 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 728 | |
ef8bbdfe | 729 | return ret; |
e37c9e69 CM |
730 | } |
731 | ||
0f9dd46c JB |
732 | /* |
733 | * return the block group that starts at or after bytenr | |
734 | */ | |
d397712b CM |
735 | static struct btrfs_block_group_cache * |
736 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 737 | { |
e2c89907 | 738 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
739 | } |
740 | ||
0f9dd46c | 741 | /* |
9f55684c | 742 | * return the block group that contains the given bytenr |
0f9dd46c | 743 | */ |
d397712b CM |
744 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
745 | struct btrfs_fs_info *info, | |
746 | u64 bytenr) | |
be744175 | 747 | { |
e2c89907 | 748 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 749 | } |
0b86a832 | 750 | |
0f9dd46c JB |
751 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
752 | u64 flags) | |
6324fbf3 | 753 | { |
0f9dd46c | 754 | struct list_head *head = &info->space_info; |
0f9dd46c | 755 | struct btrfs_space_info *found; |
4184ea7f | 756 | |
52ba6929 | 757 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 758 | |
4184ea7f CM |
759 | rcu_read_lock(); |
760 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 761 | if (found->flags & flags) { |
4184ea7f | 762 | rcu_read_unlock(); |
0f9dd46c | 763 | return found; |
4184ea7f | 764 | } |
0f9dd46c | 765 | } |
4184ea7f | 766 | rcu_read_unlock(); |
0f9dd46c | 767 | return NULL; |
6324fbf3 CM |
768 | } |
769 | ||
4184ea7f CM |
770 | /* |
771 | * after adding space to the filesystem, we need to clear the full flags | |
772 | * on all the space infos. | |
773 | */ | |
774 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
775 | { | |
776 | struct list_head *head = &info->space_info; | |
777 | struct btrfs_space_info *found; | |
778 | ||
779 | rcu_read_lock(); | |
780 | list_for_each_entry_rcu(found, head, list) | |
781 | found->full = 0; | |
782 | rcu_read_unlock(); | |
783 | } | |
784 | ||
1a4ed8fd | 785 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 786 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
787 | { |
788 | int ret; | |
789 | struct btrfs_key key; | |
31840ae1 | 790 | struct btrfs_path *path; |
e02119d5 | 791 | |
31840ae1 | 792 | path = btrfs_alloc_path(); |
d8926bb3 MF |
793 | if (!path) |
794 | return -ENOMEM; | |
795 | ||
e02119d5 CM |
796 | key.objectid = start; |
797 | key.offset = len; | |
3173a18f | 798 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 799 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 800 | btrfs_free_path(path); |
7bb86316 CM |
801 | return ret; |
802 | } | |
803 | ||
a22285a6 | 804 | /* |
3173a18f | 805 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
806 | * |
807 | * the head node for delayed ref is used to store the sum of all the | |
808 | * reference count modifications queued up in the rbtree. the head | |
809 | * node may also store the extent flags to set. This way you can check | |
810 | * to see what the reference count and extent flags would be if all of | |
811 | * the delayed refs are not processed. | |
812 | */ | |
813 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 814 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 815 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
816 | { |
817 | struct btrfs_delayed_ref_head *head; | |
818 | struct btrfs_delayed_ref_root *delayed_refs; | |
819 | struct btrfs_path *path; | |
820 | struct btrfs_extent_item *ei; | |
821 | struct extent_buffer *leaf; | |
822 | struct btrfs_key key; | |
823 | u32 item_size; | |
824 | u64 num_refs; | |
825 | u64 extent_flags; | |
826 | int ret; | |
827 | ||
3173a18f JB |
828 | /* |
829 | * If we don't have skinny metadata, don't bother doing anything | |
830 | * different | |
831 | */ | |
0b246afa JM |
832 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
833 | offset = fs_info->nodesize; | |
3173a18f JB |
834 | metadata = 0; |
835 | } | |
836 | ||
a22285a6 YZ |
837 | path = btrfs_alloc_path(); |
838 | if (!path) | |
839 | return -ENOMEM; | |
840 | ||
a22285a6 YZ |
841 | if (!trans) { |
842 | path->skip_locking = 1; | |
843 | path->search_commit_root = 1; | |
844 | } | |
639eefc8 FDBM |
845 | |
846 | search_again: | |
847 | key.objectid = bytenr; | |
848 | key.offset = offset; | |
849 | if (metadata) | |
850 | key.type = BTRFS_METADATA_ITEM_KEY; | |
851 | else | |
852 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
853 | ||
0b246afa | 854 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
855 | if (ret < 0) |
856 | goto out_free; | |
857 | ||
3173a18f | 858 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
859 | if (path->slots[0]) { |
860 | path->slots[0]--; | |
861 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
862 | path->slots[0]); | |
863 | if (key.objectid == bytenr && | |
864 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 865 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
866 | ret = 0; |
867 | } | |
3173a18f JB |
868 | } |
869 | ||
a22285a6 YZ |
870 | if (ret == 0) { |
871 | leaf = path->nodes[0]; | |
872 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
873 | if (item_size >= sizeof(*ei)) { | |
874 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
875 | struct btrfs_extent_item); | |
876 | num_refs = btrfs_extent_refs(leaf, ei); | |
877 | extent_flags = btrfs_extent_flags(leaf, ei); | |
878 | } else { | |
879 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
880 | struct btrfs_extent_item_v0 *ei0; | |
881 | BUG_ON(item_size != sizeof(*ei0)); | |
882 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
883 | struct btrfs_extent_item_v0); | |
884 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
885 | /* FIXME: this isn't correct for data */ | |
886 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
887 | #else | |
888 | BUG(); | |
889 | #endif | |
890 | } | |
891 | BUG_ON(num_refs == 0); | |
892 | } else { | |
893 | num_refs = 0; | |
894 | extent_flags = 0; | |
895 | ret = 0; | |
896 | } | |
897 | ||
898 | if (!trans) | |
899 | goto out; | |
900 | ||
901 | delayed_refs = &trans->transaction->delayed_refs; | |
902 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 903 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
904 | if (head) { |
905 | if (!mutex_trylock(&head->mutex)) { | |
6df8cdf5 | 906 | refcount_inc(&head->node.refs); |
a22285a6 YZ |
907 | spin_unlock(&delayed_refs->lock); |
908 | ||
b3b4aa74 | 909 | btrfs_release_path(path); |
a22285a6 | 910 | |
8cc33e5c DS |
911 | /* |
912 | * Mutex was contended, block until it's released and try | |
913 | * again | |
914 | */ | |
a22285a6 YZ |
915 | mutex_lock(&head->mutex); |
916 | mutex_unlock(&head->mutex); | |
917 | btrfs_put_delayed_ref(&head->node); | |
639eefc8 | 918 | goto search_again; |
a22285a6 | 919 | } |
d7df2c79 | 920 | spin_lock(&head->lock); |
a22285a6 YZ |
921 | if (head->extent_op && head->extent_op->update_flags) |
922 | extent_flags |= head->extent_op->flags_to_set; | |
923 | else | |
924 | BUG_ON(num_refs == 0); | |
925 | ||
926 | num_refs += head->node.ref_mod; | |
d7df2c79 | 927 | spin_unlock(&head->lock); |
a22285a6 YZ |
928 | mutex_unlock(&head->mutex); |
929 | } | |
930 | spin_unlock(&delayed_refs->lock); | |
931 | out: | |
932 | WARN_ON(num_refs == 0); | |
933 | if (refs) | |
934 | *refs = num_refs; | |
935 | if (flags) | |
936 | *flags = extent_flags; | |
937 | out_free: | |
938 | btrfs_free_path(path); | |
939 | return ret; | |
940 | } | |
941 | ||
d8d5f3e1 CM |
942 | /* |
943 | * Back reference rules. Back refs have three main goals: | |
944 | * | |
945 | * 1) differentiate between all holders of references to an extent so that | |
946 | * when a reference is dropped we can make sure it was a valid reference | |
947 | * before freeing the extent. | |
948 | * | |
949 | * 2) Provide enough information to quickly find the holders of an extent | |
950 | * if we notice a given block is corrupted or bad. | |
951 | * | |
952 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
953 | * maintenance. This is actually the same as #2, but with a slightly | |
954 | * different use case. | |
955 | * | |
5d4f98a2 YZ |
956 | * There are two kinds of back refs. The implicit back refs is optimized |
957 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
958 | * back refs of this kind provide information about the block's owner tree | |
959 | * and the pointer's key. These information allow us to find the block by | |
960 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
961 | * referenced by their owner trees. The location of tree block is recorded | |
962 | * in the back refs. Actually the full back refs is generic, and can be | |
963 | * used in all cases the implicit back refs is used. The major shortcoming | |
964 | * of the full back refs is its overhead. Every time a tree block gets | |
965 | * COWed, we have to update back refs entry for all pointers in it. | |
966 | * | |
967 | * For a newly allocated tree block, we use implicit back refs for | |
968 | * pointers in it. This means most tree related operations only involve | |
969 | * implicit back refs. For a tree block created in old transaction, the | |
970 | * only way to drop a reference to it is COW it. So we can detect the | |
971 | * event that tree block loses its owner tree's reference and do the | |
972 | * back refs conversion. | |
973 | * | |
01327610 | 974 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
975 | * |
976 | * The reference count of the block is one and the tree is the block's | |
977 | * owner tree. Nothing to do in this case. | |
978 | * | |
979 | * The reference count of the block is one and the tree is not the | |
980 | * block's owner tree. In this case, full back refs is used for pointers | |
981 | * in the block. Remove these full back refs, add implicit back refs for | |
982 | * every pointers in the new block. | |
983 | * | |
984 | * The reference count of the block is greater than one and the tree is | |
985 | * the block's owner tree. In this case, implicit back refs is used for | |
986 | * pointers in the block. Add full back refs for every pointers in the | |
987 | * block, increase lower level extents' reference counts. The original | |
988 | * implicit back refs are entailed to the new block. | |
989 | * | |
990 | * The reference count of the block is greater than one and the tree is | |
991 | * not the block's owner tree. Add implicit back refs for every pointer in | |
992 | * the new block, increase lower level extents' reference count. | |
993 | * | |
994 | * Back Reference Key composing: | |
995 | * | |
996 | * The key objectid corresponds to the first byte in the extent, | |
997 | * The key type is used to differentiate between types of back refs. | |
998 | * There are different meanings of the key offset for different types | |
999 | * of back refs. | |
1000 | * | |
d8d5f3e1 CM |
1001 | * File extents can be referenced by: |
1002 | * | |
1003 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1004 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1005 | * - different offsets inside a file (bookend extents in file.c) |
1006 | * | |
5d4f98a2 | 1007 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1008 | * |
1009 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1010 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1011 | * - original offset in the file |
1012 | * - how many bookend extents | |
d8d5f3e1 | 1013 | * |
5d4f98a2 YZ |
1014 | * The key offset for the implicit back refs is hash of the first |
1015 | * three fields. | |
d8d5f3e1 | 1016 | * |
5d4f98a2 | 1017 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1018 | * |
5d4f98a2 | 1019 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1020 | * |
5d4f98a2 YZ |
1021 | * The key offset for the implicit back refs is the first byte of |
1022 | * the tree leaf | |
d8d5f3e1 | 1023 | * |
5d4f98a2 YZ |
1024 | * When a file extent is allocated, The implicit back refs is used. |
1025 | * the fields are filled in: | |
d8d5f3e1 | 1026 | * |
5d4f98a2 | 1027 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1028 | * |
5d4f98a2 YZ |
1029 | * When a file extent is removed file truncation, we find the |
1030 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1031 | * |
5d4f98a2 | 1032 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1033 | * |
5d4f98a2 | 1034 | * Btree extents can be referenced by: |
d8d5f3e1 | 1035 | * |
5d4f98a2 | 1036 | * - Different subvolumes |
d8d5f3e1 | 1037 | * |
5d4f98a2 YZ |
1038 | * Both the implicit back refs and the full back refs for tree blocks |
1039 | * only consist of key. The key offset for the implicit back refs is | |
1040 | * objectid of block's owner tree. The key offset for the full back refs | |
1041 | * is the first byte of parent block. | |
d8d5f3e1 | 1042 | * |
5d4f98a2 YZ |
1043 | * When implicit back refs is used, information about the lowest key and |
1044 | * level of the tree block are required. These information are stored in | |
1045 | * tree block info structure. | |
d8d5f3e1 | 1046 | */ |
31840ae1 | 1047 | |
5d4f98a2 YZ |
1048 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1049 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
87bde3cd | 1050 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1051 | struct btrfs_path *path, |
1052 | u64 owner, u32 extra_size) | |
7bb86316 | 1053 | { |
87bde3cd | 1054 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1055 | struct btrfs_extent_item *item; |
1056 | struct btrfs_extent_item_v0 *ei0; | |
1057 | struct btrfs_extent_ref_v0 *ref0; | |
1058 | struct btrfs_tree_block_info *bi; | |
1059 | struct extent_buffer *leaf; | |
7bb86316 | 1060 | struct btrfs_key key; |
5d4f98a2 YZ |
1061 | struct btrfs_key found_key; |
1062 | u32 new_size = sizeof(*item); | |
1063 | u64 refs; | |
1064 | int ret; | |
1065 | ||
1066 | leaf = path->nodes[0]; | |
1067 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1068 | ||
1069 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1070 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1071 | struct btrfs_extent_item_v0); | |
1072 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1073 | ||
1074 | if (owner == (u64)-1) { | |
1075 | while (1) { | |
1076 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1077 | ret = btrfs_next_leaf(root, path); | |
1078 | if (ret < 0) | |
1079 | return ret; | |
79787eaa | 1080 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1081 | leaf = path->nodes[0]; |
1082 | } | |
1083 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1084 | path->slots[0]); | |
1085 | BUG_ON(key.objectid != found_key.objectid); | |
1086 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1087 | path->slots[0]++; | |
1088 | continue; | |
1089 | } | |
1090 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1091 | struct btrfs_extent_ref_v0); | |
1092 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1093 | break; | |
1094 | } | |
1095 | } | |
b3b4aa74 | 1096 | btrfs_release_path(path); |
5d4f98a2 YZ |
1097 | |
1098 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1099 | new_size += sizeof(*bi); | |
1100 | ||
1101 | new_size -= sizeof(*ei0); | |
1102 | ret = btrfs_search_slot(trans, root, &key, path, | |
1103 | new_size + extra_size, 1); | |
1104 | if (ret < 0) | |
1105 | return ret; | |
79787eaa | 1106 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1107 | |
87bde3cd | 1108 | btrfs_extend_item(fs_info, path, new_size); |
5d4f98a2 YZ |
1109 | |
1110 | leaf = path->nodes[0]; | |
1111 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1112 | btrfs_set_extent_refs(leaf, item, refs); | |
1113 | /* FIXME: get real generation */ | |
1114 | btrfs_set_extent_generation(leaf, item, 0); | |
1115 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1116 | btrfs_set_extent_flags(leaf, item, | |
1117 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1118 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1119 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1120 | /* FIXME: get first key of the block */ | |
b159fa28 | 1121 | memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi)); |
5d4f98a2 YZ |
1122 | btrfs_set_tree_block_level(leaf, bi, (int)owner); |
1123 | } else { | |
1124 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1125 | } | |
1126 | btrfs_mark_buffer_dirty(leaf); | |
1127 | return 0; | |
1128 | } | |
1129 | #endif | |
1130 | ||
1131 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1132 | { | |
1133 | u32 high_crc = ~(u32)0; | |
1134 | u32 low_crc = ~(u32)0; | |
1135 | __le64 lenum; | |
1136 | ||
1137 | lenum = cpu_to_le64(root_objectid); | |
14a958e6 | 1138 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1139 | lenum = cpu_to_le64(owner); |
14a958e6 | 1140 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1141 | lenum = cpu_to_le64(offset); |
14a958e6 | 1142 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1143 | |
1144 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1145 | } | |
1146 | ||
1147 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1148 | struct btrfs_extent_data_ref *ref) | |
1149 | { | |
1150 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1151 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1152 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1153 | } | |
1154 | ||
1155 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1156 | struct btrfs_extent_data_ref *ref, | |
1157 | u64 root_objectid, u64 owner, u64 offset) | |
1158 | { | |
1159 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1160 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1161 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1162 | return 0; | |
1163 | return 1; | |
1164 | } | |
1165 | ||
1166 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1167 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1168 | struct btrfs_path *path, |
1169 | u64 bytenr, u64 parent, | |
1170 | u64 root_objectid, | |
1171 | u64 owner, u64 offset) | |
1172 | { | |
87bde3cd | 1173 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1174 | struct btrfs_key key; |
1175 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1176 | struct extent_buffer *leaf; |
5d4f98a2 | 1177 | u32 nritems; |
74493f7a | 1178 | int ret; |
5d4f98a2 YZ |
1179 | int recow; |
1180 | int err = -ENOENT; | |
74493f7a | 1181 | |
31840ae1 | 1182 | key.objectid = bytenr; |
5d4f98a2 YZ |
1183 | if (parent) { |
1184 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1185 | key.offset = parent; | |
1186 | } else { | |
1187 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1188 | key.offset = hash_extent_data_ref(root_objectid, | |
1189 | owner, offset); | |
1190 | } | |
1191 | again: | |
1192 | recow = 0; | |
1193 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1194 | if (ret < 0) { | |
1195 | err = ret; | |
1196 | goto fail; | |
1197 | } | |
31840ae1 | 1198 | |
5d4f98a2 YZ |
1199 | if (parent) { |
1200 | if (!ret) | |
1201 | return 0; | |
1202 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1203 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1204 | btrfs_release_path(path); |
5d4f98a2 YZ |
1205 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1206 | if (ret < 0) { | |
1207 | err = ret; | |
1208 | goto fail; | |
1209 | } | |
1210 | if (!ret) | |
1211 | return 0; | |
1212 | #endif | |
1213 | goto fail; | |
31840ae1 ZY |
1214 | } |
1215 | ||
1216 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1217 | nritems = btrfs_header_nritems(leaf); |
1218 | while (1) { | |
1219 | if (path->slots[0] >= nritems) { | |
1220 | ret = btrfs_next_leaf(root, path); | |
1221 | if (ret < 0) | |
1222 | err = ret; | |
1223 | if (ret) | |
1224 | goto fail; | |
1225 | ||
1226 | leaf = path->nodes[0]; | |
1227 | nritems = btrfs_header_nritems(leaf); | |
1228 | recow = 1; | |
1229 | } | |
1230 | ||
1231 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1232 | if (key.objectid != bytenr || | |
1233 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1234 | goto fail; | |
1235 | ||
1236 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1237 | struct btrfs_extent_data_ref); | |
1238 | ||
1239 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1240 | owner, offset)) { | |
1241 | if (recow) { | |
b3b4aa74 | 1242 | btrfs_release_path(path); |
5d4f98a2 YZ |
1243 | goto again; |
1244 | } | |
1245 | err = 0; | |
1246 | break; | |
1247 | } | |
1248 | path->slots[0]++; | |
31840ae1 | 1249 | } |
5d4f98a2 YZ |
1250 | fail: |
1251 | return err; | |
31840ae1 ZY |
1252 | } |
1253 | ||
5d4f98a2 | 1254 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1255 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1256 | struct btrfs_path *path, |
1257 | u64 bytenr, u64 parent, | |
1258 | u64 root_objectid, u64 owner, | |
1259 | u64 offset, int refs_to_add) | |
31840ae1 | 1260 | { |
87bde3cd | 1261 | struct btrfs_root *root = fs_info->extent_root; |
31840ae1 ZY |
1262 | struct btrfs_key key; |
1263 | struct extent_buffer *leaf; | |
5d4f98a2 | 1264 | u32 size; |
31840ae1 ZY |
1265 | u32 num_refs; |
1266 | int ret; | |
74493f7a | 1267 | |
74493f7a | 1268 | key.objectid = bytenr; |
5d4f98a2 YZ |
1269 | if (parent) { |
1270 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1271 | key.offset = parent; | |
1272 | size = sizeof(struct btrfs_shared_data_ref); | |
1273 | } else { | |
1274 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1275 | key.offset = hash_extent_data_ref(root_objectid, | |
1276 | owner, offset); | |
1277 | size = sizeof(struct btrfs_extent_data_ref); | |
1278 | } | |
74493f7a | 1279 | |
5d4f98a2 YZ |
1280 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1281 | if (ret && ret != -EEXIST) | |
1282 | goto fail; | |
1283 | ||
1284 | leaf = path->nodes[0]; | |
1285 | if (parent) { | |
1286 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1287 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1288 | struct btrfs_shared_data_ref); |
1289 | if (ret == 0) { | |
1290 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1291 | } else { | |
1292 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1293 | num_refs += refs_to_add; | |
1294 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1295 | } |
5d4f98a2 YZ |
1296 | } else { |
1297 | struct btrfs_extent_data_ref *ref; | |
1298 | while (ret == -EEXIST) { | |
1299 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1300 | struct btrfs_extent_data_ref); | |
1301 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1302 | owner, offset)) | |
1303 | break; | |
b3b4aa74 | 1304 | btrfs_release_path(path); |
5d4f98a2 YZ |
1305 | key.offset++; |
1306 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1307 | size); | |
1308 | if (ret && ret != -EEXIST) | |
1309 | goto fail; | |
31840ae1 | 1310 | |
5d4f98a2 YZ |
1311 | leaf = path->nodes[0]; |
1312 | } | |
1313 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1314 | struct btrfs_extent_data_ref); | |
1315 | if (ret == 0) { | |
1316 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1317 | root_objectid); | |
1318 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1319 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1320 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1321 | } else { | |
1322 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1323 | num_refs += refs_to_add; | |
1324 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1325 | } |
31840ae1 | 1326 | } |
5d4f98a2 YZ |
1327 | btrfs_mark_buffer_dirty(leaf); |
1328 | ret = 0; | |
1329 | fail: | |
b3b4aa74 | 1330 | btrfs_release_path(path); |
7bb86316 | 1331 | return ret; |
74493f7a CM |
1332 | } |
1333 | ||
5d4f98a2 | 1334 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1335 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 1336 | struct btrfs_path *path, |
fcebe456 | 1337 | int refs_to_drop, int *last_ref) |
31840ae1 | 1338 | { |
5d4f98a2 YZ |
1339 | struct btrfs_key key; |
1340 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1341 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1342 | struct extent_buffer *leaf; |
5d4f98a2 | 1343 | u32 num_refs = 0; |
31840ae1 ZY |
1344 | int ret = 0; |
1345 | ||
1346 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1347 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1348 | ||
1349 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1350 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1351 | struct btrfs_extent_data_ref); | |
1352 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1353 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1354 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1355 | struct btrfs_shared_data_ref); | |
1356 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1357 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1358 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1359 | struct btrfs_extent_ref_v0 *ref0; | |
1360 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1361 | struct btrfs_extent_ref_v0); | |
1362 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1363 | #endif | |
1364 | } else { | |
1365 | BUG(); | |
1366 | } | |
1367 | ||
56bec294 CM |
1368 | BUG_ON(num_refs < refs_to_drop); |
1369 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1370 | |
31840ae1 | 1371 | if (num_refs == 0) { |
87bde3cd | 1372 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
fcebe456 | 1373 | *last_ref = 1; |
31840ae1 | 1374 | } else { |
5d4f98a2 YZ |
1375 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1376 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1377 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1378 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1379 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1380 | else { | |
1381 | struct btrfs_extent_ref_v0 *ref0; | |
1382 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1383 | struct btrfs_extent_ref_v0); | |
1384 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1385 | } | |
1386 | #endif | |
31840ae1 ZY |
1387 | btrfs_mark_buffer_dirty(leaf); |
1388 | } | |
31840ae1 ZY |
1389 | return ret; |
1390 | } | |
1391 | ||
9ed0dea0 | 1392 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1393 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1394 | { |
5d4f98a2 YZ |
1395 | struct btrfs_key key; |
1396 | struct extent_buffer *leaf; | |
1397 | struct btrfs_extent_data_ref *ref1; | |
1398 | struct btrfs_shared_data_ref *ref2; | |
1399 | u32 num_refs = 0; | |
1400 | ||
1401 | leaf = path->nodes[0]; | |
1402 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1403 | if (iref) { | |
1404 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1405 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1406 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1407 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1408 | } else { | |
1409 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1410 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1411 | } | |
1412 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1413 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1414 | struct btrfs_extent_data_ref); | |
1415 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1416 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1417 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1418 | struct btrfs_shared_data_ref); | |
1419 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1420 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1421 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1422 | struct btrfs_extent_ref_v0 *ref0; | |
1423 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1424 | struct btrfs_extent_ref_v0); | |
1425 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1426 | #endif |
5d4f98a2 YZ |
1427 | } else { |
1428 | WARN_ON(1); | |
1429 | } | |
1430 | return num_refs; | |
1431 | } | |
15916de8 | 1432 | |
5d4f98a2 | 1433 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1434 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1435 | struct btrfs_path *path, |
1436 | u64 bytenr, u64 parent, | |
1437 | u64 root_objectid) | |
1f3c79a2 | 1438 | { |
87bde3cd | 1439 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 | 1440 | struct btrfs_key key; |
1f3c79a2 | 1441 | int ret; |
1f3c79a2 | 1442 | |
5d4f98a2 YZ |
1443 | key.objectid = bytenr; |
1444 | if (parent) { | |
1445 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1446 | key.offset = parent; | |
1447 | } else { | |
1448 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1449 | key.offset = root_objectid; | |
1f3c79a2 LH |
1450 | } |
1451 | ||
5d4f98a2 YZ |
1452 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1453 | if (ret > 0) | |
1454 | ret = -ENOENT; | |
1455 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1456 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1457 | btrfs_release_path(path); |
5d4f98a2 YZ |
1458 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1459 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1460 | if (ret > 0) | |
1461 | ret = -ENOENT; | |
1462 | } | |
1f3c79a2 | 1463 | #endif |
5d4f98a2 | 1464 | return ret; |
1f3c79a2 LH |
1465 | } |
1466 | ||
5d4f98a2 | 1467 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1468 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1469 | struct btrfs_path *path, |
1470 | u64 bytenr, u64 parent, | |
1471 | u64 root_objectid) | |
31840ae1 | 1472 | { |
5d4f98a2 | 1473 | struct btrfs_key key; |
31840ae1 | 1474 | int ret; |
31840ae1 | 1475 | |
5d4f98a2 YZ |
1476 | key.objectid = bytenr; |
1477 | if (parent) { | |
1478 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1479 | key.offset = parent; | |
1480 | } else { | |
1481 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1482 | key.offset = root_objectid; | |
1483 | } | |
1484 | ||
87bde3cd JM |
1485 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, |
1486 | path, &key, 0); | |
b3b4aa74 | 1487 | btrfs_release_path(path); |
31840ae1 ZY |
1488 | return ret; |
1489 | } | |
1490 | ||
5d4f98a2 | 1491 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1492 | { |
5d4f98a2 YZ |
1493 | int type; |
1494 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1495 | if (parent > 0) | |
1496 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1497 | else | |
1498 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1499 | } else { | |
1500 | if (parent > 0) | |
1501 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1502 | else | |
1503 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1504 | } | |
1505 | return type; | |
31840ae1 | 1506 | } |
56bec294 | 1507 | |
2c47e605 YZ |
1508 | static int find_next_key(struct btrfs_path *path, int level, |
1509 | struct btrfs_key *key) | |
56bec294 | 1510 | |
02217ed2 | 1511 | { |
2c47e605 | 1512 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1513 | if (!path->nodes[level]) |
1514 | break; | |
5d4f98a2 YZ |
1515 | if (path->slots[level] + 1 >= |
1516 | btrfs_header_nritems(path->nodes[level])) | |
1517 | continue; | |
1518 | if (level == 0) | |
1519 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1520 | path->slots[level] + 1); | |
1521 | else | |
1522 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1523 | path->slots[level] + 1); | |
1524 | return 0; | |
1525 | } | |
1526 | return 1; | |
1527 | } | |
037e6390 | 1528 | |
5d4f98a2 YZ |
1529 | /* |
1530 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1531 | * to the address of inline back ref, and 0 is returned. | |
1532 | * | |
1533 | * if back ref isn't found, *ref_ret is set to the address where it | |
1534 | * should be inserted, and -ENOENT is returned. | |
1535 | * | |
1536 | * if insert is true and there are too many inline back refs, the path | |
1537 | * points to the extent item, and -EAGAIN is returned. | |
1538 | * | |
1539 | * NOTE: inline back refs are ordered in the same way that back ref | |
1540 | * items in the tree are ordered. | |
1541 | */ | |
1542 | static noinline_for_stack | |
1543 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1544 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1545 | struct btrfs_path *path, |
1546 | struct btrfs_extent_inline_ref **ref_ret, | |
1547 | u64 bytenr, u64 num_bytes, | |
1548 | u64 parent, u64 root_objectid, | |
1549 | u64 owner, u64 offset, int insert) | |
1550 | { | |
87bde3cd | 1551 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1552 | struct btrfs_key key; |
1553 | struct extent_buffer *leaf; | |
1554 | struct btrfs_extent_item *ei; | |
1555 | struct btrfs_extent_inline_ref *iref; | |
1556 | u64 flags; | |
1557 | u64 item_size; | |
1558 | unsigned long ptr; | |
1559 | unsigned long end; | |
1560 | int extra_size; | |
1561 | int type; | |
1562 | int want; | |
1563 | int ret; | |
1564 | int err = 0; | |
0b246afa | 1565 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
26b8003f | 1566 | |
db94535d | 1567 | key.objectid = bytenr; |
31840ae1 | 1568 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1569 | key.offset = num_bytes; |
31840ae1 | 1570 | |
5d4f98a2 YZ |
1571 | want = extent_ref_type(parent, owner); |
1572 | if (insert) { | |
1573 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1574 | path->keep_locks = 1; |
5d4f98a2 YZ |
1575 | } else |
1576 | extra_size = -1; | |
3173a18f JB |
1577 | |
1578 | /* | |
1579 | * Owner is our parent level, so we can just add one to get the level | |
1580 | * for the block we are interested in. | |
1581 | */ | |
1582 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1583 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1584 | key.offset = owner; | |
1585 | } | |
1586 | ||
1587 | again: | |
5d4f98a2 | 1588 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1589 | if (ret < 0) { |
5d4f98a2 YZ |
1590 | err = ret; |
1591 | goto out; | |
1592 | } | |
3173a18f JB |
1593 | |
1594 | /* | |
1595 | * We may be a newly converted file system which still has the old fat | |
1596 | * extent entries for metadata, so try and see if we have one of those. | |
1597 | */ | |
1598 | if (ret > 0 && skinny_metadata) { | |
1599 | skinny_metadata = false; | |
1600 | if (path->slots[0]) { | |
1601 | path->slots[0]--; | |
1602 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1603 | path->slots[0]); | |
1604 | if (key.objectid == bytenr && | |
1605 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1606 | key.offset == num_bytes) | |
1607 | ret = 0; | |
1608 | } | |
1609 | if (ret) { | |
9ce49a0b | 1610 | key.objectid = bytenr; |
3173a18f JB |
1611 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1612 | key.offset = num_bytes; | |
1613 | btrfs_release_path(path); | |
1614 | goto again; | |
1615 | } | |
1616 | } | |
1617 | ||
79787eaa JM |
1618 | if (ret && !insert) { |
1619 | err = -ENOENT; | |
1620 | goto out; | |
fae7f21c | 1621 | } else if (WARN_ON(ret)) { |
492104c8 | 1622 | err = -EIO; |
492104c8 | 1623 | goto out; |
79787eaa | 1624 | } |
5d4f98a2 YZ |
1625 | |
1626 | leaf = path->nodes[0]; | |
1627 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1628 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1629 | if (item_size < sizeof(*ei)) { | |
1630 | if (!insert) { | |
1631 | err = -ENOENT; | |
1632 | goto out; | |
1633 | } | |
87bde3cd | 1634 | ret = convert_extent_item_v0(trans, fs_info, path, owner, |
5d4f98a2 YZ |
1635 | extra_size); |
1636 | if (ret < 0) { | |
1637 | err = ret; | |
1638 | goto out; | |
1639 | } | |
1640 | leaf = path->nodes[0]; | |
1641 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1642 | } | |
1643 | #endif | |
1644 | BUG_ON(item_size < sizeof(*ei)); | |
1645 | ||
5d4f98a2 YZ |
1646 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1647 | flags = btrfs_extent_flags(leaf, ei); | |
1648 | ||
1649 | ptr = (unsigned long)(ei + 1); | |
1650 | end = (unsigned long)ei + item_size; | |
1651 | ||
3173a18f | 1652 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1653 | ptr += sizeof(struct btrfs_tree_block_info); |
1654 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1655 | } |
1656 | ||
1657 | err = -ENOENT; | |
1658 | while (1) { | |
1659 | if (ptr >= end) { | |
1660 | WARN_ON(ptr > end); | |
1661 | break; | |
1662 | } | |
1663 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1664 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1665 | if (want < type) | |
1666 | break; | |
1667 | if (want > type) { | |
1668 | ptr += btrfs_extent_inline_ref_size(type); | |
1669 | continue; | |
1670 | } | |
1671 | ||
1672 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1673 | struct btrfs_extent_data_ref *dref; | |
1674 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1675 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1676 | owner, offset)) { | |
1677 | err = 0; | |
1678 | break; | |
1679 | } | |
1680 | if (hash_extent_data_ref_item(leaf, dref) < | |
1681 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1682 | break; | |
1683 | } else { | |
1684 | u64 ref_offset; | |
1685 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1686 | if (parent > 0) { | |
1687 | if (parent == ref_offset) { | |
1688 | err = 0; | |
1689 | break; | |
1690 | } | |
1691 | if (ref_offset < parent) | |
1692 | break; | |
1693 | } else { | |
1694 | if (root_objectid == ref_offset) { | |
1695 | err = 0; | |
1696 | break; | |
1697 | } | |
1698 | if (ref_offset < root_objectid) | |
1699 | break; | |
1700 | } | |
1701 | } | |
1702 | ptr += btrfs_extent_inline_ref_size(type); | |
1703 | } | |
1704 | if (err == -ENOENT && insert) { | |
1705 | if (item_size + extra_size >= | |
1706 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1707 | err = -EAGAIN; | |
1708 | goto out; | |
1709 | } | |
1710 | /* | |
1711 | * To add new inline back ref, we have to make sure | |
1712 | * there is no corresponding back ref item. | |
1713 | * For simplicity, we just do not add new inline back | |
1714 | * ref if there is any kind of item for this block | |
1715 | */ | |
2c47e605 YZ |
1716 | if (find_next_key(path, 0, &key) == 0 && |
1717 | key.objectid == bytenr && | |
85d4198e | 1718 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1719 | err = -EAGAIN; |
1720 | goto out; | |
1721 | } | |
1722 | } | |
1723 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1724 | out: | |
85d4198e | 1725 | if (insert) { |
5d4f98a2 YZ |
1726 | path->keep_locks = 0; |
1727 | btrfs_unlock_up_safe(path, 1); | |
1728 | } | |
1729 | return err; | |
1730 | } | |
1731 | ||
1732 | /* | |
1733 | * helper to add new inline back ref | |
1734 | */ | |
1735 | static noinline_for_stack | |
87bde3cd | 1736 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1737 | struct btrfs_path *path, |
1738 | struct btrfs_extent_inline_ref *iref, | |
1739 | u64 parent, u64 root_objectid, | |
1740 | u64 owner, u64 offset, int refs_to_add, | |
1741 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1742 | { |
1743 | struct extent_buffer *leaf; | |
1744 | struct btrfs_extent_item *ei; | |
1745 | unsigned long ptr; | |
1746 | unsigned long end; | |
1747 | unsigned long item_offset; | |
1748 | u64 refs; | |
1749 | int size; | |
1750 | int type; | |
5d4f98a2 YZ |
1751 | |
1752 | leaf = path->nodes[0]; | |
1753 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1754 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1755 | ||
1756 | type = extent_ref_type(parent, owner); | |
1757 | size = btrfs_extent_inline_ref_size(type); | |
1758 | ||
87bde3cd | 1759 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1760 | |
1761 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1762 | refs = btrfs_extent_refs(leaf, ei); | |
1763 | refs += refs_to_add; | |
1764 | btrfs_set_extent_refs(leaf, ei, refs); | |
1765 | if (extent_op) | |
1766 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1767 | ||
1768 | ptr = (unsigned long)ei + item_offset; | |
1769 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1770 | if (ptr < end - size) | |
1771 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1772 | end - size - ptr); | |
1773 | ||
1774 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1775 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1776 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1777 | struct btrfs_extent_data_ref *dref; | |
1778 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1779 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1780 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1781 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1782 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1783 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1784 | struct btrfs_shared_data_ref *sref; | |
1785 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1786 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1787 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1788 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1789 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1790 | } else { | |
1791 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1792 | } | |
1793 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1794 | } |
1795 | ||
1796 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1797 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1798 | struct btrfs_path *path, |
1799 | struct btrfs_extent_inline_ref **ref_ret, | |
1800 | u64 bytenr, u64 num_bytes, u64 parent, | |
1801 | u64 root_objectid, u64 owner, u64 offset) | |
1802 | { | |
1803 | int ret; | |
1804 | ||
87bde3cd | 1805 | ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret, |
5d4f98a2 YZ |
1806 | bytenr, num_bytes, parent, |
1807 | root_objectid, owner, offset, 0); | |
1808 | if (ret != -ENOENT) | |
54aa1f4d | 1809 | return ret; |
5d4f98a2 | 1810 | |
b3b4aa74 | 1811 | btrfs_release_path(path); |
5d4f98a2 YZ |
1812 | *ref_ret = NULL; |
1813 | ||
1814 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
87bde3cd JM |
1815 | ret = lookup_tree_block_ref(trans, fs_info, path, bytenr, |
1816 | parent, root_objectid); | |
5d4f98a2 | 1817 | } else { |
87bde3cd JM |
1818 | ret = lookup_extent_data_ref(trans, fs_info, path, bytenr, |
1819 | parent, root_objectid, owner, | |
1820 | offset); | |
b9473439 | 1821 | } |
5d4f98a2 YZ |
1822 | return ret; |
1823 | } | |
31840ae1 | 1824 | |
5d4f98a2 YZ |
1825 | /* |
1826 | * helper to update/remove inline back ref | |
1827 | */ | |
1828 | static noinline_for_stack | |
87bde3cd | 1829 | void update_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1830 | struct btrfs_path *path, |
1831 | struct btrfs_extent_inline_ref *iref, | |
1832 | int refs_to_mod, | |
fcebe456 JB |
1833 | struct btrfs_delayed_extent_op *extent_op, |
1834 | int *last_ref) | |
5d4f98a2 YZ |
1835 | { |
1836 | struct extent_buffer *leaf; | |
1837 | struct btrfs_extent_item *ei; | |
1838 | struct btrfs_extent_data_ref *dref = NULL; | |
1839 | struct btrfs_shared_data_ref *sref = NULL; | |
1840 | unsigned long ptr; | |
1841 | unsigned long end; | |
1842 | u32 item_size; | |
1843 | int size; | |
1844 | int type; | |
5d4f98a2 YZ |
1845 | u64 refs; |
1846 | ||
1847 | leaf = path->nodes[0]; | |
1848 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1849 | refs = btrfs_extent_refs(leaf, ei); | |
1850 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1851 | refs += refs_to_mod; | |
1852 | btrfs_set_extent_refs(leaf, ei, refs); | |
1853 | if (extent_op) | |
1854 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1855 | ||
1856 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1857 | ||
1858 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1859 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1860 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1861 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1862 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1863 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1864 | } else { | |
1865 | refs = 1; | |
1866 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1867 | } |
31840ae1 | 1868 | |
5d4f98a2 YZ |
1869 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1870 | refs += refs_to_mod; | |
1871 | ||
1872 | if (refs > 0) { | |
1873 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1874 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1875 | else | |
1876 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1877 | } else { | |
fcebe456 | 1878 | *last_ref = 1; |
5d4f98a2 YZ |
1879 | size = btrfs_extent_inline_ref_size(type); |
1880 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1881 | ptr = (unsigned long)iref; | |
1882 | end = (unsigned long)ei + item_size; | |
1883 | if (ptr + size < end) | |
1884 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1885 | end - ptr - size); | |
1886 | item_size -= size; | |
87bde3cd | 1887 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1888 | } |
1889 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1890 | } |
1891 | ||
1892 | static noinline_for_stack | |
1893 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1894 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1895 | struct btrfs_path *path, |
1896 | u64 bytenr, u64 num_bytes, u64 parent, | |
1897 | u64 root_objectid, u64 owner, | |
1898 | u64 offset, int refs_to_add, | |
1899 | struct btrfs_delayed_extent_op *extent_op) | |
1900 | { | |
1901 | struct btrfs_extent_inline_ref *iref; | |
1902 | int ret; | |
1903 | ||
87bde3cd | 1904 | ret = lookup_inline_extent_backref(trans, fs_info, path, &iref, |
5d4f98a2 YZ |
1905 | bytenr, num_bytes, parent, |
1906 | root_objectid, owner, offset, 1); | |
1907 | if (ret == 0) { | |
1908 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
87bde3cd | 1909 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 1910 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1911 | } else if (ret == -ENOENT) { |
87bde3cd | 1912 | setup_inline_extent_backref(fs_info, path, iref, parent, |
143bede5 JM |
1913 | root_objectid, owner, offset, |
1914 | refs_to_add, extent_op); | |
1915 | ret = 0; | |
771ed689 | 1916 | } |
5d4f98a2 YZ |
1917 | return ret; |
1918 | } | |
31840ae1 | 1919 | |
5d4f98a2 | 1920 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 1921 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1922 | struct btrfs_path *path, |
1923 | u64 bytenr, u64 parent, u64 root_objectid, | |
1924 | u64 owner, u64 offset, int refs_to_add) | |
1925 | { | |
1926 | int ret; | |
1927 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1928 | BUG_ON(refs_to_add != 1); | |
87bde3cd | 1929 | ret = insert_tree_block_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
1930 | parent, root_objectid); |
1931 | } else { | |
87bde3cd | 1932 | ret = insert_extent_data_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
1933 | parent, root_objectid, |
1934 | owner, offset, refs_to_add); | |
1935 | } | |
1936 | return ret; | |
1937 | } | |
56bec294 | 1938 | |
5d4f98a2 | 1939 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 1940 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1941 | struct btrfs_path *path, |
1942 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1943 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1944 | { |
143bede5 | 1945 | int ret = 0; |
b9473439 | 1946 | |
5d4f98a2 YZ |
1947 | BUG_ON(!is_data && refs_to_drop != 1); |
1948 | if (iref) { | |
87bde3cd | 1949 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 1950 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 1951 | } else if (is_data) { |
87bde3cd | 1952 | ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop, |
fcebe456 | 1953 | last_ref); |
5d4f98a2 | 1954 | } else { |
fcebe456 | 1955 | *last_ref = 1; |
87bde3cd | 1956 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
5d4f98a2 YZ |
1957 | } |
1958 | return ret; | |
1959 | } | |
1960 | ||
86557861 | 1961 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
1962 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1963 | u64 *discarded_bytes) | |
5d4f98a2 | 1964 | { |
86557861 JM |
1965 | int j, ret = 0; |
1966 | u64 bytes_left, end; | |
4d89d377 | 1967 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1968 | |
4d89d377 JM |
1969 | if (WARN_ON(start != aligned_start)) { |
1970 | len -= aligned_start - start; | |
1971 | len = round_down(len, 1 << 9); | |
1972 | start = aligned_start; | |
1973 | } | |
d04c6b88 | 1974 | |
4d89d377 | 1975 | *discarded_bytes = 0; |
86557861 JM |
1976 | |
1977 | if (!len) | |
1978 | return 0; | |
1979 | ||
1980 | end = start + len; | |
1981 | bytes_left = len; | |
1982 | ||
1983 | /* Skip any superblocks on this device. */ | |
1984 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1985 | u64 sb_start = btrfs_sb_offset(j); | |
1986 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1987 | u64 size = sb_start - start; | |
1988 | ||
1989 | if (!in_range(sb_start, start, bytes_left) && | |
1990 | !in_range(sb_end, start, bytes_left) && | |
1991 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1992 | continue; | |
1993 | ||
1994 | /* | |
1995 | * Superblock spans beginning of range. Adjust start and | |
1996 | * try again. | |
1997 | */ | |
1998 | if (sb_start <= start) { | |
1999 | start += sb_end - start; | |
2000 | if (start > end) { | |
2001 | bytes_left = 0; | |
2002 | break; | |
2003 | } | |
2004 | bytes_left = end - start; | |
2005 | continue; | |
2006 | } | |
2007 | ||
2008 | if (size) { | |
2009 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2010 | GFP_NOFS, 0); | |
2011 | if (!ret) | |
2012 | *discarded_bytes += size; | |
2013 | else if (ret != -EOPNOTSUPP) | |
2014 | return ret; | |
2015 | } | |
2016 | ||
2017 | start = sb_end; | |
2018 | if (start > end) { | |
2019 | bytes_left = 0; | |
2020 | break; | |
2021 | } | |
2022 | bytes_left = end - start; | |
2023 | } | |
2024 | ||
2025 | if (bytes_left) { | |
2026 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2027 | GFP_NOFS, 0); |
2028 | if (!ret) | |
86557861 | 2029 | *discarded_bytes += bytes_left; |
4d89d377 | 2030 | } |
d04c6b88 | 2031 | return ret; |
5d4f98a2 | 2032 | } |
5d4f98a2 | 2033 | |
2ff7e61e | 2034 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 2035 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 2036 | { |
5d4f98a2 | 2037 | int ret; |
5378e607 | 2038 | u64 discarded_bytes = 0; |
a1d3c478 | 2039 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2040 | |
e244a0ae | 2041 | |
2999241d FM |
2042 | /* |
2043 | * Avoid races with device replace and make sure our bbio has devices | |
2044 | * associated to its stripes that don't go away while we are discarding. | |
2045 | */ | |
0b246afa | 2046 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 2047 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
2048 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
2049 | &bbio, 0); | |
79787eaa | 2050 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2051 | if (!ret) { |
a1d3c478 | 2052 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2053 | int i; |
2054 | ||
5d4f98a2 | 2055 | |
a1d3c478 | 2056 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2057 | u64 bytes; |
d5e2003c JB |
2058 | if (!stripe->dev->can_discard) |
2059 | continue; | |
2060 | ||
5378e607 LD |
2061 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2062 | stripe->physical, | |
d04c6b88 JM |
2063 | stripe->length, |
2064 | &bytes); | |
5378e607 | 2065 | if (!ret) |
d04c6b88 | 2066 | discarded_bytes += bytes; |
5378e607 | 2067 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2068 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2069 | |
2070 | /* | |
2071 | * Just in case we get back EOPNOTSUPP for some reason, | |
2072 | * just ignore the return value so we don't screw up | |
2073 | * people calling discard_extent. | |
2074 | */ | |
2075 | ret = 0; | |
5d4f98a2 | 2076 | } |
6e9606d2 | 2077 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2078 | } |
0b246afa | 2079 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2080 | |
2081 | if (actual_bytes) | |
2082 | *actual_bytes = discarded_bytes; | |
2083 | ||
5d4f98a2 | 2084 | |
53b381b3 DW |
2085 | if (ret == -EOPNOTSUPP) |
2086 | ret = 0; | |
5d4f98a2 | 2087 | return ret; |
5d4f98a2 YZ |
2088 | } |
2089 | ||
79787eaa | 2090 | /* Can return -ENOMEM */ |
5d4f98a2 | 2091 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2092 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 2093 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2094 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
2095 | { |
2096 | int ret; | |
66d7e7f0 | 2097 | |
5d4f98a2 YZ |
2098 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2099 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2100 | ||
2101 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
2102 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
2103 | num_bytes, | |
5d4f98a2 | 2104 | parent, root_objectid, (int)owner, |
b06c4bf5 | 2105 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 | 2106 | } else { |
66d7e7f0 | 2107 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5846a3c2 QW |
2108 | num_bytes, parent, root_objectid, |
2109 | owner, offset, 0, | |
fef394f7 | 2110 | BTRFS_ADD_DELAYED_REF); |
5d4f98a2 YZ |
2111 | } |
2112 | return ret; | |
2113 | } | |
2114 | ||
2115 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2116 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 2117 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2118 | u64 parent, u64 root_objectid, |
2119 | u64 owner, u64 offset, int refs_to_add, | |
2120 | struct btrfs_delayed_extent_op *extent_op) | |
2121 | { | |
2122 | struct btrfs_path *path; | |
2123 | struct extent_buffer *leaf; | |
2124 | struct btrfs_extent_item *item; | |
fcebe456 | 2125 | struct btrfs_key key; |
c682f9b3 QW |
2126 | u64 bytenr = node->bytenr; |
2127 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2128 | u64 refs; |
2129 | int ret; | |
5d4f98a2 YZ |
2130 | |
2131 | path = btrfs_alloc_path(); | |
2132 | if (!path) | |
2133 | return -ENOMEM; | |
2134 | ||
e4058b54 | 2135 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2136 | path->leave_spinning = 1; |
2137 | /* this will setup the path even if it fails to insert the back ref */ | |
87bde3cd JM |
2138 | ret = insert_inline_extent_backref(trans, fs_info, path, bytenr, |
2139 | num_bytes, parent, root_objectid, | |
2140 | owner, offset, | |
5d4f98a2 | 2141 | refs_to_add, extent_op); |
0ed4792a | 2142 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2143 | goto out; |
fcebe456 JB |
2144 | |
2145 | /* | |
2146 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2147 | * inline extent ref, so just update the reference count and add a | |
2148 | * normal backref. | |
2149 | */ | |
5d4f98a2 | 2150 | leaf = path->nodes[0]; |
fcebe456 | 2151 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2152 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2153 | refs = btrfs_extent_refs(leaf, item); | |
2154 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2155 | if (extent_op) | |
2156 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2157 | |
5d4f98a2 | 2158 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2159 | btrfs_release_path(path); |
56bec294 | 2160 | |
e4058b54 | 2161 | path->reada = READA_FORWARD; |
b9473439 | 2162 | path->leave_spinning = 1; |
56bec294 | 2163 | /* now insert the actual backref */ |
87bde3cd JM |
2164 | ret = insert_extent_backref(trans, fs_info, path, bytenr, parent, |
2165 | root_objectid, owner, offset, refs_to_add); | |
79787eaa | 2166 | if (ret) |
66642832 | 2167 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2168 | out: |
56bec294 | 2169 | btrfs_free_path(path); |
30d133fc | 2170 | return ret; |
56bec294 CM |
2171 | } |
2172 | ||
5d4f98a2 | 2173 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2174 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2175 | struct btrfs_delayed_ref_node *node, |
2176 | struct btrfs_delayed_extent_op *extent_op, | |
2177 | int insert_reserved) | |
56bec294 | 2178 | { |
5d4f98a2 YZ |
2179 | int ret = 0; |
2180 | struct btrfs_delayed_data_ref *ref; | |
2181 | struct btrfs_key ins; | |
2182 | u64 parent = 0; | |
2183 | u64 ref_root = 0; | |
2184 | u64 flags = 0; | |
2185 | ||
2186 | ins.objectid = node->bytenr; | |
2187 | ins.offset = node->num_bytes; | |
2188 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2189 | ||
2190 | ref = btrfs_delayed_node_to_data_ref(node); | |
0b246afa | 2191 | trace_run_delayed_data_ref(fs_info, node, ref, node->action); |
599c75ec | 2192 | |
5d4f98a2 YZ |
2193 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2194 | parent = ref->parent; | |
fcebe456 | 2195 | ref_root = ref->root; |
5d4f98a2 YZ |
2196 | |
2197 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2198 | if (extent_op) |
5d4f98a2 | 2199 | flags |= extent_op->flags_to_set; |
2ff7e61e | 2200 | ret = alloc_reserved_file_extent(trans, fs_info, |
5d4f98a2 YZ |
2201 | parent, ref_root, flags, |
2202 | ref->objectid, ref->offset, | |
2203 | &ins, node->ref_mod); | |
5d4f98a2 | 2204 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2205 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2206 | ref_root, ref->objectid, |
2207 | ref->offset, node->ref_mod, | |
c682f9b3 | 2208 | extent_op); |
5d4f98a2 | 2209 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2210 | ret = __btrfs_free_extent(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2211 | ref_root, ref->objectid, |
2212 | ref->offset, node->ref_mod, | |
c682f9b3 | 2213 | extent_op); |
5d4f98a2 YZ |
2214 | } else { |
2215 | BUG(); | |
2216 | } | |
2217 | return ret; | |
2218 | } | |
2219 | ||
2220 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2221 | struct extent_buffer *leaf, | |
2222 | struct btrfs_extent_item *ei) | |
2223 | { | |
2224 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2225 | if (extent_op->update_flags) { | |
2226 | flags |= extent_op->flags_to_set; | |
2227 | btrfs_set_extent_flags(leaf, ei, flags); | |
2228 | } | |
2229 | ||
2230 | if (extent_op->update_key) { | |
2231 | struct btrfs_tree_block_info *bi; | |
2232 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2233 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2234 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2235 | } | |
2236 | } | |
2237 | ||
2238 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2239 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2240 | struct btrfs_delayed_ref_node *node, |
2241 | struct btrfs_delayed_extent_op *extent_op) | |
2242 | { | |
2243 | struct btrfs_key key; | |
2244 | struct btrfs_path *path; | |
2245 | struct btrfs_extent_item *ei; | |
2246 | struct extent_buffer *leaf; | |
2247 | u32 item_size; | |
56bec294 | 2248 | int ret; |
5d4f98a2 | 2249 | int err = 0; |
b1c79e09 | 2250 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2251 | |
79787eaa JM |
2252 | if (trans->aborted) |
2253 | return 0; | |
2254 | ||
0b246afa | 2255 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2256 | metadata = 0; |
2257 | ||
5d4f98a2 YZ |
2258 | path = btrfs_alloc_path(); |
2259 | if (!path) | |
2260 | return -ENOMEM; | |
2261 | ||
2262 | key.objectid = node->bytenr; | |
5d4f98a2 | 2263 | |
3173a18f | 2264 | if (metadata) { |
3173a18f | 2265 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2266 | key.offset = extent_op->level; |
3173a18f JB |
2267 | } else { |
2268 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2269 | key.offset = node->num_bytes; | |
2270 | } | |
2271 | ||
2272 | again: | |
e4058b54 | 2273 | path->reada = READA_FORWARD; |
5d4f98a2 | 2274 | path->leave_spinning = 1; |
0b246afa | 2275 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2276 | if (ret < 0) { |
2277 | err = ret; | |
2278 | goto out; | |
2279 | } | |
2280 | if (ret > 0) { | |
3173a18f | 2281 | if (metadata) { |
55994887 FDBM |
2282 | if (path->slots[0] > 0) { |
2283 | path->slots[0]--; | |
2284 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2285 | path->slots[0]); | |
2286 | if (key.objectid == node->bytenr && | |
2287 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2288 | key.offset == node->num_bytes) | |
2289 | ret = 0; | |
2290 | } | |
2291 | if (ret > 0) { | |
2292 | btrfs_release_path(path); | |
2293 | metadata = 0; | |
3173a18f | 2294 | |
55994887 FDBM |
2295 | key.objectid = node->bytenr; |
2296 | key.offset = node->num_bytes; | |
2297 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2298 | goto again; | |
2299 | } | |
2300 | } else { | |
2301 | err = -EIO; | |
2302 | goto out; | |
3173a18f | 2303 | } |
5d4f98a2 YZ |
2304 | } |
2305 | ||
2306 | leaf = path->nodes[0]; | |
2307 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2308 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2309 | if (item_size < sizeof(*ei)) { | |
87bde3cd | 2310 | ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0); |
5d4f98a2 YZ |
2311 | if (ret < 0) { |
2312 | err = ret; | |
2313 | goto out; | |
2314 | } | |
2315 | leaf = path->nodes[0]; | |
2316 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2317 | } | |
2318 | #endif | |
2319 | BUG_ON(item_size < sizeof(*ei)); | |
2320 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2321 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2322 | |
5d4f98a2 YZ |
2323 | btrfs_mark_buffer_dirty(leaf); |
2324 | out: | |
2325 | btrfs_free_path(path); | |
2326 | return err; | |
56bec294 CM |
2327 | } |
2328 | ||
5d4f98a2 | 2329 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2330 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2331 | struct btrfs_delayed_ref_node *node, |
2332 | struct btrfs_delayed_extent_op *extent_op, | |
2333 | int insert_reserved) | |
56bec294 CM |
2334 | { |
2335 | int ret = 0; | |
5d4f98a2 YZ |
2336 | struct btrfs_delayed_tree_ref *ref; |
2337 | struct btrfs_key ins; | |
2338 | u64 parent = 0; | |
2339 | u64 ref_root = 0; | |
0b246afa | 2340 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
56bec294 | 2341 | |
5d4f98a2 | 2342 | ref = btrfs_delayed_node_to_tree_ref(node); |
0b246afa | 2343 | trace_run_delayed_tree_ref(fs_info, node, ref, node->action); |
599c75ec | 2344 | |
5d4f98a2 YZ |
2345 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2346 | parent = ref->parent; | |
fcebe456 | 2347 | ref_root = ref->root; |
5d4f98a2 | 2348 | |
3173a18f JB |
2349 | ins.objectid = node->bytenr; |
2350 | if (skinny_metadata) { | |
2351 | ins.offset = ref->level; | |
2352 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2353 | } else { | |
2354 | ins.offset = node->num_bytes; | |
2355 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2356 | } | |
2357 | ||
02794222 | 2358 | if (node->ref_mod != 1) { |
2ff7e61e | 2359 | btrfs_err(fs_info, |
02794222 LB |
2360 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2361 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2362 | parent); | |
2363 | return -EIO; | |
2364 | } | |
5d4f98a2 | 2365 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2366 | BUG_ON(!extent_op || !extent_op->update_flags); |
2ff7e61e | 2367 | ret = alloc_reserved_tree_block(trans, fs_info, |
5d4f98a2 YZ |
2368 | parent, ref_root, |
2369 | extent_op->flags_to_set, | |
2370 | &extent_op->key, | |
b06c4bf5 | 2371 | ref->level, &ins); |
5d4f98a2 | 2372 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2373 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, |
c682f9b3 QW |
2374 | parent, ref_root, |
2375 | ref->level, 0, 1, | |
fcebe456 | 2376 | extent_op); |
5d4f98a2 | 2377 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2378 | ret = __btrfs_free_extent(trans, fs_info, node, |
c682f9b3 QW |
2379 | parent, ref_root, |
2380 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2381 | } else { |
2382 | BUG(); | |
2383 | } | |
56bec294 CM |
2384 | return ret; |
2385 | } | |
2386 | ||
2387 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2388 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2389 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2390 | struct btrfs_delayed_ref_node *node, |
2391 | struct btrfs_delayed_extent_op *extent_op, | |
2392 | int insert_reserved) | |
56bec294 | 2393 | { |
79787eaa JM |
2394 | int ret = 0; |
2395 | ||
857cc2fc JB |
2396 | if (trans->aborted) { |
2397 | if (insert_reserved) | |
2ff7e61e | 2398 | btrfs_pin_extent(fs_info, node->bytenr, |
857cc2fc | 2399 | node->num_bytes, 1); |
79787eaa | 2400 | return 0; |
857cc2fc | 2401 | } |
79787eaa | 2402 | |
5d4f98a2 | 2403 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2404 | struct btrfs_delayed_ref_head *head; |
2405 | /* | |
2406 | * we've hit the end of the chain and we were supposed | |
2407 | * to insert this extent into the tree. But, it got | |
2408 | * deleted before we ever needed to insert it, so all | |
2409 | * we have to do is clean up the accounting | |
2410 | */ | |
5d4f98a2 YZ |
2411 | BUG_ON(extent_op); |
2412 | head = btrfs_delayed_node_to_head(node); | |
0b246afa | 2413 | trace_run_delayed_ref_head(fs_info, node, head, node->action); |
599c75ec | 2414 | |
56bec294 | 2415 | if (insert_reserved) { |
2ff7e61e | 2416 | btrfs_pin_extent(fs_info, node->bytenr, |
f0486c68 | 2417 | node->num_bytes, 1); |
5d4f98a2 | 2418 | if (head->is_data) { |
0b246afa | 2419 | ret = btrfs_del_csums(trans, fs_info, |
5d4f98a2 YZ |
2420 | node->bytenr, |
2421 | node->num_bytes); | |
5d4f98a2 | 2422 | } |
56bec294 | 2423 | } |
297d750b QW |
2424 | |
2425 | /* Also free its reserved qgroup space */ | |
0b246afa | 2426 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, |
297d750b | 2427 | head->qgroup_reserved); |
79787eaa | 2428 | return ret; |
56bec294 CM |
2429 | } |
2430 | ||
5d4f98a2 YZ |
2431 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2432 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2ff7e61e | 2433 | ret = run_delayed_tree_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2434 | insert_reserved); |
2435 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2436 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2ff7e61e | 2437 | ret = run_delayed_data_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2438 | insert_reserved); |
2439 | else | |
2440 | BUG(); | |
2441 | return ret; | |
56bec294 CM |
2442 | } |
2443 | ||
c6fc2454 | 2444 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2445 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2446 | { | |
cffc3374 FM |
2447 | struct btrfs_delayed_ref_node *ref; |
2448 | ||
c6fc2454 QW |
2449 | if (list_empty(&head->ref_list)) |
2450 | return NULL; | |
d7df2c79 | 2451 | |
cffc3374 FM |
2452 | /* |
2453 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2454 | * This is to prevent a ref count from going down to zero, which deletes | |
2455 | * the extent item from the extent tree, when there still are references | |
2456 | * to add, which would fail because they would not find the extent item. | |
2457 | */ | |
1d57ee94 WX |
2458 | if (!list_empty(&head->ref_add_list)) |
2459 | return list_first_entry(&head->ref_add_list, | |
2460 | struct btrfs_delayed_ref_node, add_list); | |
2461 | ||
2462 | ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node, | |
2463 | list); | |
2464 | ASSERT(list_empty(&ref->add_list)); | |
2465 | return ref; | |
56bec294 CM |
2466 | } |
2467 | ||
79787eaa JM |
2468 | /* |
2469 | * Returns 0 on success or if called with an already aborted transaction. | |
2470 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2471 | */ | |
d7df2c79 | 2472 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2473 | struct btrfs_fs_info *fs_info, |
d7df2c79 | 2474 | unsigned long nr) |
56bec294 | 2475 | { |
56bec294 CM |
2476 | struct btrfs_delayed_ref_root *delayed_refs; |
2477 | struct btrfs_delayed_ref_node *ref; | |
2478 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2479 | struct btrfs_delayed_extent_op *extent_op; |
0a2b2a84 | 2480 | ktime_t start = ktime_get(); |
56bec294 | 2481 | int ret; |
d7df2c79 | 2482 | unsigned long count = 0; |
0a2b2a84 | 2483 | unsigned long actual_count = 0; |
56bec294 | 2484 | int must_insert_reserved = 0; |
56bec294 CM |
2485 | |
2486 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2487 | while (1) { |
2488 | if (!locked_ref) { | |
d7df2c79 | 2489 | if (count >= nr) |
56bec294 | 2490 | break; |
56bec294 | 2491 | |
d7df2c79 JB |
2492 | spin_lock(&delayed_refs->lock); |
2493 | locked_ref = btrfs_select_ref_head(trans); | |
2494 | if (!locked_ref) { | |
2495 | spin_unlock(&delayed_refs->lock); | |
2496 | break; | |
2497 | } | |
c3e69d58 CM |
2498 | |
2499 | /* grab the lock that says we are going to process | |
2500 | * all the refs for this head */ | |
2501 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2502 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2503 | /* |
2504 | * we may have dropped the spin lock to get the head | |
2505 | * mutex lock, and that might have given someone else | |
2506 | * time to free the head. If that's true, it has been | |
2507 | * removed from our list and we can move on. | |
2508 | */ | |
2509 | if (ret == -EAGAIN) { | |
2510 | locked_ref = NULL; | |
2511 | count++; | |
2512 | continue; | |
56bec294 CM |
2513 | } |
2514 | } | |
a28ec197 | 2515 | |
2c3cf7d5 FM |
2516 | /* |
2517 | * We need to try and merge add/drops of the same ref since we | |
2518 | * can run into issues with relocate dropping the implicit ref | |
2519 | * and then it being added back again before the drop can | |
2520 | * finish. If we merged anything we need to re-loop so we can | |
2521 | * get a good ref. | |
2522 | * Or we can get node references of the same type that weren't | |
2523 | * merged when created due to bumps in the tree mod seq, and | |
2524 | * we need to merge them to prevent adding an inline extent | |
2525 | * backref before dropping it (triggering a BUG_ON at | |
2526 | * insert_inline_extent_backref()). | |
2527 | */ | |
d7df2c79 | 2528 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2529 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2530 | locked_ref); | |
ae1e206b | 2531 | |
d1270cd9 AJ |
2532 | /* |
2533 | * locked_ref is the head node, so we have to go one | |
2534 | * node back for any delayed ref updates | |
2535 | */ | |
2536 | ref = select_delayed_ref(locked_ref); | |
2537 | ||
2538 | if (ref && ref->seq && | |
097b8a7c | 2539 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2540 | spin_unlock(&locked_ref->lock); |
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); | |
d0280996 | 2545 | btrfs_delayed_ref_unlock(locked_ref); |
d7df2c79 | 2546 | locked_ref = NULL; |
d1270cd9 | 2547 | cond_resched(); |
27a377db | 2548 | count++; |
d1270cd9 AJ |
2549 | continue; |
2550 | } | |
2551 | ||
56bec294 CM |
2552 | /* |
2553 | * record the must insert reserved flag before we | |
2554 | * drop the spin lock. | |
2555 | */ | |
2556 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2557 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2558 | |
5d4f98a2 YZ |
2559 | extent_op = locked_ref->extent_op; |
2560 | locked_ref->extent_op = NULL; | |
2561 | ||
56bec294 | 2562 | if (!ref) { |
d7df2c79 JB |
2563 | |
2564 | ||
56bec294 CM |
2565 | /* All delayed refs have been processed, Go ahead |
2566 | * and send the head node to run_one_delayed_ref, | |
2567 | * so that any accounting fixes can happen | |
2568 | */ | |
2569 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2570 | |
2571 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2572 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2573 | extent_op = NULL; |
2574 | } | |
2575 | ||
2576 | if (extent_op) { | |
d7df2c79 | 2577 | spin_unlock(&locked_ref->lock); |
2ff7e61e | 2578 | ret = run_delayed_extent_op(trans, fs_info, |
5d4f98a2 | 2579 | ref, extent_op); |
78a6184a | 2580 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2581 | |
79787eaa | 2582 | if (ret) { |
857cc2fc JB |
2583 | /* |
2584 | * Need to reset must_insert_reserved if | |
2585 | * there was an error so the abort stuff | |
2586 | * can cleanup the reserved space | |
2587 | * properly. | |
2588 | */ | |
2589 | if (must_insert_reserved) | |
2590 | locked_ref->must_insert_reserved = 1; | |
aa7c8da3 | 2591 | spin_lock(&delayed_refs->lock); |
d7df2c79 | 2592 | locked_ref->processing = 0; |
aa7c8da3 JM |
2593 | delayed_refs->num_heads_ready++; |
2594 | spin_unlock(&delayed_refs->lock); | |
5d163e0e JM |
2595 | btrfs_debug(fs_info, |
2596 | "run_delayed_extent_op returned %d", | |
2597 | ret); | |
093486c4 | 2598 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2599 | return ret; |
2600 | } | |
d7df2c79 | 2601 | continue; |
5d4f98a2 | 2602 | } |
02217ed2 | 2603 | |
d7df2c79 | 2604 | /* |
01327610 | 2605 | * Need to drop our head ref lock and re-acquire the |
d7df2c79 JB |
2606 | * delayed ref lock and then re-check to make sure |
2607 | * nobody got added. | |
2608 | */ | |
2609 | spin_unlock(&locked_ref->lock); | |
2610 | spin_lock(&delayed_refs->lock); | |
2611 | spin_lock(&locked_ref->lock); | |
c6fc2454 | 2612 | if (!list_empty(&locked_ref->ref_list) || |
573a0755 | 2613 | locked_ref->extent_op) { |
d7df2c79 JB |
2614 | spin_unlock(&locked_ref->lock); |
2615 | spin_unlock(&delayed_refs->lock); | |
2616 | continue; | |
2617 | } | |
2618 | ref->in_tree = 0; | |
2619 | delayed_refs->num_heads--; | |
c46effa6 LB |
2620 | rb_erase(&locked_ref->href_node, |
2621 | &delayed_refs->href_root); | |
d7df2c79 JB |
2622 | spin_unlock(&delayed_refs->lock); |
2623 | } else { | |
0a2b2a84 | 2624 | actual_count++; |
d7df2c79 | 2625 | ref->in_tree = 0; |
c6fc2454 | 2626 | list_del(&ref->list); |
1d57ee94 WX |
2627 | if (!list_empty(&ref->add_list)) |
2628 | list_del(&ref->add_list); | |
c46effa6 | 2629 | } |
d7df2c79 JB |
2630 | atomic_dec(&delayed_refs->num_entries); |
2631 | ||
093486c4 | 2632 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2633 | /* |
2634 | * when we play the delayed ref, also correct the | |
2635 | * ref_mod on head | |
2636 | */ | |
2637 | switch (ref->action) { | |
2638 | case BTRFS_ADD_DELAYED_REF: | |
2639 | case BTRFS_ADD_DELAYED_EXTENT: | |
2640 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2641 | break; | |
2642 | case BTRFS_DROP_DELAYED_REF: | |
2643 | locked_ref->node.ref_mod += ref->ref_mod; | |
2644 | break; | |
2645 | default: | |
2646 | WARN_ON(1); | |
2647 | } | |
2648 | } | |
d7df2c79 | 2649 | spin_unlock(&locked_ref->lock); |
925baedd | 2650 | |
2ff7e61e | 2651 | ret = run_one_delayed_ref(trans, fs_info, ref, extent_op, |
56bec294 | 2652 | must_insert_reserved); |
eb099670 | 2653 | |
78a6184a | 2654 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2655 | if (ret) { |
9d1032cc | 2656 | spin_lock(&delayed_refs->lock); |
d7df2c79 | 2657 | locked_ref->processing = 0; |
9d1032cc WX |
2658 | delayed_refs->num_heads_ready++; |
2659 | spin_unlock(&delayed_refs->lock); | |
093486c4 MX |
2660 | btrfs_delayed_ref_unlock(locked_ref); |
2661 | btrfs_put_delayed_ref(ref); | |
5d163e0e JM |
2662 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
2663 | ret); | |
79787eaa JM |
2664 | return ret; |
2665 | } | |
2666 | ||
093486c4 MX |
2667 | /* |
2668 | * If this node is a head, that means all the refs in this head | |
2669 | * have been dealt with, and we will pick the next head to deal | |
2670 | * with, so we must unlock the head and drop it from the cluster | |
2671 | * list before we release it. | |
2672 | */ | |
2673 | if (btrfs_delayed_ref_is_head(ref)) { | |
1262133b JB |
2674 | if (locked_ref->is_data && |
2675 | locked_ref->total_ref_mod < 0) { | |
2676 | spin_lock(&delayed_refs->lock); | |
2677 | delayed_refs->pending_csums -= ref->num_bytes; | |
2678 | spin_unlock(&delayed_refs->lock); | |
2679 | } | |
093486c4 MX |
2680 | btrfs_delayed_ref_unlock(locked_ref); |
2681 | locked_ref = NULL; | |
2682 | } | |
2683 | btrfs_put_delayed_ref(ref); | |
2684 | count++; | |
c3e69d58 | 2685 | cond_resched(); |
c3e69d58 | 2686 | } |
0a2b2a84 JB |
2687 | |
2688 | /* | |
2689 | * We don't want to include ref heads since we can have empty ref heads | |
2690 | * and those will drastically skew our runtime down since we just do | |
2691 | * accounting, no actual extent tree updates. | |
2692 | */ | |
2693 | if (actual_count > 0) { | |
2694 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2695 | u64 avg; | |
2696 | ||
2697 | /* | |
2698 | * We weigh the current average higher than our current runtime | |
2699 | * to avoid large swings in the average. | |
2700 | */ | |
2701 | spin_lock(&delayed_refs->lock); | |
2702 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2703 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2704 | spin_unlock(&delayed_refs->lock); |
2705 | } | |
d7df2c79 | 2706 | return 0; |
c3e69d58 CM |
2707 | } |
2708 | ||
709c0486 AJ |
2709 | #ifdef SCRAMBLE_DELAYED_REFS |
2710 | /* | |
2711 | * Normally delayed refs get processed in ascending bytenr order. This | |
2712 | * correlates in most cases to the order added. To expose dependencies on this | |
2713 | * order, we start to process the tree in the middle instead of the beginning | |
2714 | */ | |
2715 | static u64 find_middle(struct rb_root *root) | |
2716 | { | |
2717 | struct rb_node *n = root->rb_node; | |
2718 | struct btrfs_delayed_ref_node *entry; | |
2719 | int alt = 1; | |
2720 | u64 middle; | |
2721 | u64 first = 0, last = 0; | |
2722 | ||
2723 | n = rb_first(root); | |
2724 | if (n) { | |
2725 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2726 | first = entry->bytenr; | |
2727 | } | |
2728 | n = rb_last(root); | |
2729 | if (n) { | |
2730 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2731 | last = entry->bytenr; | |
2732 | } | |
2733 | n = root->rb_node; | |
2734 | ||
2735 | while (n) { | |
2736 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2737 | WARN_ON(!entry->in_tree); | |
2738 | ||
2739 | middle = entry->bytenr; | |
2740 | ||
2741 | if (alt) | |
2742 | n = n->rb_left; | |
2743 | else | |
2744 | n = n->rb_right; | |
2745 | ||
2746 | alt = 1 - alt; | |
2747 | } | |
2748 | return middle; | |
2749 | } | |
2750 | #endif | |
2751 | ||
2ff7e61e | 2752 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2753 | { |
2754 | u64 num_bytes; | |
2755 | ||
2756 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2757 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2758 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2759 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2760 | ||
2761 | /* | |
2762 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2763 | * closer to what we're really going to want to use. |
1be41b78 | 2764 | */ |
0b246afa | 2765 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2766 | } |
2767 | ||
1262133b JB |
2768 | /* |
2769 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2770 | * would require to store the csums for that many bytes. | |
2771 | */ | |
2ff7e61e | 2772 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2773 | { |
2774 | u64 csum_size; | |
2775 | u64 num_csums_per_leaf; | |
2776 | u64 num_csums; | |
2777 | ||
0b246afa | 2778 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2779 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2780 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2781 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2782 | num_csums += num_csums_per_leaf - 1; |
2783 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2784 | return num_csums; | |
2785 | } | |
2786 | ||
0a2b2a84 | 2787 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2788 | struct btrfs_fs_info *fs_info) |
1be41b78 JB |
2789 | { |
2790 | struct btrfs_block_rsv *global_rsv; | |
2791 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2792 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
cb723e49 JB |
2793 | u64 num_dirty_bgs = trans->transaction->num_dirty_bgs; |
2794 | u64 num_bytes, num_dirty_bgs_bytes; | |
1be41b78 JB |
2795 | int ret = 0; |
2796 | ||
0b246afa | 2797 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2798 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2799 | if (num_heads > 1) |
0b246afa | 2800 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2801 | num_bytes <<= 1; |
2ff7e61e JM |
2802 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2803 | fs_info->nodesize; | |
0b246afa | 2804 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2805 | num_dirty_bgs); |
0b246afa | 2806 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2807 | |
2808 | /* | |
2809 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2810 | * wiggle room since running delayed refs can create more delayed refs. | |
2811 | */ | |
cb723e49 JB |
2812 | if (global_rsv->space_info->full) { |
2813 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2814 | num_bytes <<= 1; |
cb723e49 | 2815 | } |
1be41b78 JB |
2816 | |
2817 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2818 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2819 | ret = 1; |
2820 | spin_unlock(&global_rsv->lock); | |
2821 | return ret; | |
2822 | } | |
2823 | ||
0a2b2a84 | 2824 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2825 | struct btrfs_fs_info *fs_info) |
0a2b2a84 | 2826 | { |
0a2b2a84 JB |
2827 | u64 num_entries = |
2828 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2829 | u64 avg_runtime; | |
a79b7d4b | 2830 | u64 val; |
0a2b2a84 JB |
2831 | |
2832 | smp_mb(); | |
2833 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2834 | val = num_entries * avg_runtime; |
dc1a90c6 | 2835 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2836 | return 1; |
a79b7d4b CM |
2837 | if (val >= NSEC_PER_SEC / 2) |
2838 | return 2; | |
0a2b2a84 | 2839 | |
2ff7e61e | 2840 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
0a2b2a84 JB |
2841 | } |
2842 | ||
a79b7d4b CM |
2843 | struct async_delayed_refs { |
2844 | struct btrfs_root *root; | |
31b9655f | 2845 | u64 transid; |
a79b7d4b CM |
2846 | int count; |
2847 | int error; | |
2848 | int sync; | |
2849 | struct completion wait; | |
2850 | struct btrfs_work work; | |
2851 | }; | |
2852 | ||
2ff7e61e JM |
2853 | static inline struct async_delayed_refs * |
2854 | to_async_delayed_refs(struct btrfs_work *work) | |
2855 | { | |
2856 | return container_of(work, struct async_delayed_refs, work); | |
2857 | } | |
2858 | ||
a79b7d4b CM |
2859 | static void delayed_ref_async_start(struct btrfs_work *work) |
2860 | { | |
2ff7e61e | 2861 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2862 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2863 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2864 | int ret; |
2865 | ||
0f873eca | 2866 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2867 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2868 | goto done; |
31b9655f | 2869 | |
0f873eca CM |
2870 | trans = btrfs_join_transaction(async->root); |
2871 | if (IS_ERR(trans)) { | |
2872 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2873 | goto done; |
2874 | } | |
2875 | ||
2876 | /* | |
01327610 | 2877 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2878 | * wait on delayed refs |
2879 | */ | |
2880 | trans->sync = true; | |
0f873eca CM |
2881 | |
2882 | /* Don't bother flushing if we got into a different transaction */ | |
2883 | if (trans->transid > async->transid) | |
2884 | goto end; | |
2885 | ||
2ff7e61e | 2886 | ret = btrfs_run_delayed_refs(trans, fs_info, async->count); |
a79b7d4b CM |
2887 | if (ret) |
2888 | async->error = ret; | |
0f873eca | 2889 | end: |
3a45bb20 | 2890 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
2891 | if (ret && !async->error) |
2892 | async->error = ret; | |
2893 | done: | |
2894 | if (async->sync) | |
2895 | complete(&async->wait); | |
2896 | else | |
2897 | kfree(async); | |
2898 | } | |
2899 | ||
2ff7e61e | 2900 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 2901 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
2902 | { |
2903 | struct async_delayed_refs *async; | |
2904 | int ret; | |
2905 | ||
2906 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2907 | if (!async) | |
2908 | return -ENOMEM; | |
2909 | ||
0b246afa | 2910 | async->root = fs_info->tree_root; |
a79b7d4b CM |
2911 | async->count = count; |
2912 | async->error = 0; | |
31b9655f | 2913 | async->transid = transid; |
a79b7d4b CM |
2914 | if (wait) |
2915 | async->sync = 1; | |
2916 | else | |
2917 | async->sync = 0; | |
2918 | init_completion(&async->wait); | |
2919 | ||
9e0af237 LB |
2920 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
2921 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 2922 | |
0b246afa | 2923 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
2924 | |
2925 | if (wait) { | |
2926 | wait_for_completion(&async->wait); | |
2927 | ret = async->error; | |
2928 | kfree(async); | |
2929 | return ret; | |
2930 | } | |
2931 | return 0; | |
2932 | } | |
2933 | ||
c3e69d58 CM |
2934 | /* |
2935 | * this starts processing the delayed reference count updates and | |
2936 | * extent insertions we have queued up so far. count can be | |
2937 | * 0, which means to process everything in the tree at the start | |
2938 | * of the run (but not newly added entries), or it can be some target | |
2939 | * number you'd like to process. | |
79787eaa JM |
2940 | * |
2941 | * Returns 0 on success or if called with an aborted transaction | |
2942 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2943 | */ |
2944 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2945 | struct btrfs_fs_info *fs_info, unsigned long count) |
c3e69d58 CM |
2946 | { |
2947 | struct rb_node *node; | |
2948 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2949 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2950 | int ret; |
2951 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 2952 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 2953 | |
79787eaa JM |
2954 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2955 | if (trans->aborted) | |
2956 | return 0; | |
2957 | ||
0b246afa | 2958 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2959 | return 0; |
2960 | ||
c3e69d58 | 2961 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2962 | if (count == 0) |
d7df2c79 | 2963 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2964 | |
c3e69d58 | 2965 | again: |
709c0486 AJ |
2966 | #ifdef SCRAMBLE_DELAYED_REFS |
2967 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2968 | #endif | |
d9a0540a | 2969 | trans->can_flush_pending_bgs = false; |
2ff7e61e | 2970 | ret = __btrfs_run_delayed_refs(trans, fs_info, count); |
d7df2c79 | 2971 | if (ret < 0) { |
66642832 | 2972 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2973 | return ret; |
eb099670 | 2974 | } |
c3e69d58 | 2975 | |
56bec294 | 2976 | if (run_all) { |
d7df2c79 | 2977 | if (!list_empty(&trans->new_bgs)) |
2ff7e61e | 2978 | btrfs_create_pending_block_groups(trans, fs_info); |
ea658bad | 2979 | |
d7df2c79 | 2980 | spin_lock(&delayed_refs->lock); |
c46effa6 | 2981 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
2982 | if (!node) { |
2983 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2984 | goto out; |
d7df2c79 | 2985 | } |
e9d0b13b | 2986 | |
56bec294 | 2987 | while (node) { |
c46effa6 LB |
2988 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2989 | href_node); | |
2990 | if (btrfs_delayed_ref_is_head(&head->node)) { | |
2991 | struct btrfs_delayed_ref_node *ref; | |
5caf2a00 | 2992 | |
c46effa6 | 2993 | ref = &head->node; |
6df8cdf5 | 2994 | refcount_inc(&ref->refs); |
56bec294 CM |
2995 | |
2996 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2997 | /* |
2998 | * Mutex was contended, block until it's | |
2999 | * released and try again | |
3000 | */ | |
56bec294 CM |
3001 | mutex_lock(&head->mutex); |
3002 | mutex_unlock(&head->mutex); | |
3003 | ||
3004 | btrfs_put_delayed_ref(ref); | |
1887be66 | 3005 | cond_resched(); |
56bec294 | 3006 | goto again; |
c46effa6 LB |
3007 | } else { |
3008 | WARN_ON(1); | |
56bec294 CM |
3009 | } |
3010 | node = rb_next(node); | |
3011 | } | |
3012 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 3013 | cond_resched(); |
56bec294 | 3014 | goto again; |
5f39d397 | 3015 | } |
54aa1f4d | 3016 | out: |
d9a0540a | 3017 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3018 | return 0; |
3019 | } | |
3020 | ||
5d4f98a2 | 3021 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 3022 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 3023 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 3024 | int level, int is_data) |
5d4f98a2 YZ |
3025 | { |
3026 | struct btrfs_delayed_extent_op *extent_op; | |
3027 | int ret; | |
3028 | ||
78a6184a | 3029 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3030 | if (!extent_op) |
3031 | return -ENOMEM; | |
3032 | ||
3033 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3034 | extent_op->update_flags = true; |
3035 | extent_op->update_key = false; | |
3036 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3037 | extent_op->level = level; |
5d4f98a2 | 3038 | |
0b246afa | 3039 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 3040 | num_bytes, extent_op); |
5d4f98a2 | 3041 | if (ret) |
78a6184a | 3042 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3043 | return ret; |
3044 | } | |
3045 | ||
e4c3b2dc | 3046 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3047 | struct btrfs_path *path, |
3048 | u64 objectid, u64 offset, u64 bytenr) | |
3049 | { | |
3050 | struct btrfs_delayed_ref_head *head; | |
3051 | struct btrfs_delayed_ref_node *ref; | |
3052 | struct btrfs_delayed_data_ref *data_ref; | |
3053 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 3054 | struct btrfs_transaction *cur_trans; |
5d4f98a2 YZ |
3055 | int ret = 0; |
3056 | ||
e4c3b2dc LB |
3057 | cur_trans = root->fs_info->running_transaction; |
3058 | if (!cur_trans) | |
3059 | return 0; | |
3060 | ||
3061 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3062 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3063 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3064 | if (!head) { |
3065 | spin_unlock(&delayed_refs->lock); | |
3066 | return 0; | |
3067 | } | |
5d4f98a2 YZ |
3068 | |
3069 | if (!mutex_trylock(&head->mutex)) { | |
6df8cdf5 | 3070 | refcount_inc(&head->node.refs); |
5d4f98a2 YZ |
3071 | spin_unlock(&delayed_refs->lock); |
3072 | ||
b3b4aa74 | 3073 | btrfs_release_path(path); |
5d4f98a2 | 3074 | |
8cc33e5c DS |
3075 | /* |
3076 | * Mutex was contended, block until it's released and let | |
3077 | * caller try again | |
3078 | */ | |
5d4f98a2 YZ |
3079 | mutex_lock(&head->mutex); |
3080 | mutex_unlock(&head->mutex); | |
3081 | btrfs_put_delayed_ref(&head->node); | |
3082 | return -EAGAIN; | |
3083 | } | |
d7df2c79 | 3084 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3085 | |
d7df2c79 | 3086 | spin_lock(&head->lock); |
c6fc2454 | 3087 | list_for_each_entry(ref, &head->ref_list, list) { |
d7df2c79 JB |
3088 | /* If it's a shared ref we know a cross reference exists */ |
3089 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3090 | ret = 1; | |
3091 | break; | |
3092 | } | |
5d4f98a2 | 3093 | |
d7df2c79 | 3094 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3095 | |
d7df2c79 JB |
3096 | /* |
3097 | * If our ref doesn't match the one we're currently looking at | |
3098 | * then we have a cross reference. | |
3099 | */ | |
3100 | if (data_ref->root != root->root_key.objectid || | |
3101 | data_ref->objectid != objectid || | |
3102 | data_ref->offset != offset) { | |
3103 | ret = 1; | |
3104 | break; | |
3105 | } | |
5d4f98a2 | 3106 | } |
d7df2c79 | 3107 | spin_unlock(&head->lock); |
5d4f98a2 | 3108 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
3109 | return ret; |
3110 | } | |
3111 | ||
e4c3b2dc | 3112 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3113 | struct btrfs_path *path, |
3114 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3115 | { |
0b246afa JM |
3116 | struct btrfs_fs_info *fs_info = root->fs_info; |
3117 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3118 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3119 | struct btrfs_extent_data_ref *ref; |
3120 | struct btrfs_extent_inline_ref *iref; | |
3121 | struct btrfs_extent_item *ei; | |
f321e491 | 3122 | struct btrfs_key key; |
5d4f98a2 | 3123 | u32 item_size; |
be20aa9d | 3124 | int ret; |
925baedd | 3125 | |
be20aa9d | 3126 | key.objectid = bytenr; |
31840ae1 | 3127 | key.offset = (u64)-1; |
f321e491 | 3128 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3129 | |
be20aa9d CM |
3130 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3131 | if (ret < 0) | |
3132 | goto out; | |
79787eaa | 3133 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3134 | |
3135 | ret = -ENOENT; | |
3136 | if (path->slots[0] == 0) | |
31840ae1 | 3137 | goto out; |
be20aa9d | 3138 | |
31840ae1 | 3139 | path->slots[0]--; |
f321e491 | 3140 | leaf = path->nodes[0]; |
5d4f98a2 | 3141 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3142 | |
5d4f98a2 | 3143 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3144 | goto out; |
f321e491 | 3145 | |
5d4f98a2 YZ |
3146 | ret = 1; |
3147 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3148 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3149 | if (item_size < sizeof(*ei)) { | |
3150 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3151 | goto out; | |
3152 | } | |
3153 | #endif | |
3154 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3155 | |
5d4f98a2 YZ |
3156 | if (item_size != sizeof(*ei) + |
3157 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3158 | goto out; | |
be20aa9d | 3159 | |
5d4f98a2 YZ |
3160 | if (btrfs_extent_generation(leaf, ei) <= |
3161 | btrfs_root_last_snapshot(&root->root_item)) | |
3162 | goto out; | |
3163 | ||
3164 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3165 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
3166 | BTRFS_EXTENT_DATA_REF_KEY) | |
3167 | goto out; | |
3168 | ||
3169 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3170 | if (btrfs_extent_refs(leaf, ei) != | |
3171 | btrfs_extent_data_ref_count(leaf, ref) || | |
3172 | btrfs_extent_data_ref_root(leaf, ref) != | |
3173 | root->root_key.objectid || | |
3174 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3175 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3176 | goto out; | |
3177 | ||
3178 | ret = 0; | |
3179 | out: | |
3180 | return ret; | |
3181 | } | |
3182 | ||
e4c3b2dc LB |
3183 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3184 | u64 bytenr) | |
5d4f98a2 YZ |
3185 | { |
3186 | struct btrfs_path *path; | |
3187 | int ret; | |
3188 | int ret2; | |
3189 | ||
3190 | path = btrfs_alloc_path(); | |
3191 | if (!path) | |
3192 | return -ENOENT; | |
3193 | ||
3194 | do { | |
e4c3b2dc | 3195 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3196 | offset, bytenr); |
3197 | if (ret && ret != -ENOENT) | |
f321e491 | 3198 | goto out; |
80ff3856 | 3199 | |
e4c3b2dc | 3200 | ret2 = check_delayed_ref(root, path, objectid, |
5d4f98a2 YZ |
3201 | offset, bytenr); |
3202 | } while (ret2 == -EAGAIN); | |
3203 | ||
3204 | if (ret2 && ret2 != -ENOENT) { | |
3205 | ret = ret2; | |
3206 | goto out; | |
f321e491 | 3207 | } |
5d4f98a2 YZ |
3208 | |
3209 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3210 | ret = 0; | |
be20aa9d | 3211 | out: |
80ff3856 | 3212 | btrfs_free_path(path); |
f0486c68 YZ |
3213 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3214 | WARN_ON(ret > 0); | |
f321e491 | 3215 | return ret; |
be20aa9d | 3216 | } |
c5739bba | 3217 | |
5d4f98a2 | 3218 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3219 | struct btrfs_root *root, |
5d4f98a2 | 3220 | struct extent_buffer *buf, |
e339a6b0 | 3221 | int full_backref, int inc) |
31840ae1 | 3222 | { |
0b246afa | 3223 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3224 | u64 bytenr; |
5d4f98a2 YZ |
3225 | u64 num_bytes; |
3226 | u64 parent; | |
31840ae1 | 3227 | u64 ref_root; |
31840ae1 | 3228 | u32 nritems; |
31840ae1 ZY |
3229 | struct btrfs_key key; |
3230 | struct btrfs_file_extent_item *fi; | |
3231 | int i; | |
3232 | int level; | |
3233 | int ret = 0; | |
2ff7e61e JM |
3234 | int (*process_func)(struct btrfs_trans_handle *, |
3235 | struct btrfs_fs_info *, | |
b06c4bf5 | 3236 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3237 | |
fccb84c9 | 3238 | |
0b246afa | 3239 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3240 | return 0; |
fccb84c9 | 3241 | |
31840ae1 | 3242 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3243 | nritems = btrfs_header_nritems(buf); |
3244 | level = btrfs_header_level(buf); | |
3245 | ||
27cdeb70 | 3246 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3247 | return 0; |
31840ae1 | 3248 | |
5d4f98a2 YZ |
3249 | if (inc) |
3250 | process_func = btrfs_inc_extent_ref; | |
3251 | else | |
3252 | process_func = btrfs_free_extent; | |
31840ae1 | 3253 | |
5d4f98a2 YZ |
3254 | if (full_backref) |
3255 | parent = buf->start; | |
3256 | else | |
3257 | parent = 0; | |
3258 | ||
3259 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3260 | if (level == 0) { |
5d4f98a2 | 3261 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3262 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3263 | continue; |
5d4f98a2 | 3264 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3265 | struct btrfs_file_extent_item); |
3266 | if (btrfs_file_extent_type(buf, fi) == | |
3267 | BTRFS_FILE_EXTENT_INLINE) | |
3268 | continue; | |
3269 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3270 | if (bytenr == 0) | |
3271 | continue; | |
5d4f98a2 YZ |
3272 | |
3273 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3274 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2ff7e61e | 3275 | ret = process_func(trans, fs_info, bytenr, num_bytes, |
5d4f98a2 | 3276 | parent, ref_root, key.objectid, |
b06c4bf5 | 3277 | key.offset); |
31840ae1 ZY |
3278 | if (ret) |
3279 | goto fail; | |
3280 | } else { | |
5d4f98a2 | 3281 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3282 | num_bytes = fs_info->nodesize; |
2ff7e61e | 3283 | ret = process_func(trans, fs_info, bytenr, num_bytes, |
b06c4bf5 | 3284 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3285 | if (ret) |
3286 | goto fail; | |
3287 | } | |
3288 | } | |
3289 | return 0; | |
3290 | fail: | |
5d4f98a2 YZ |
3291 | return ret; |
3292 | } | |
3293 | ||
3294 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3295 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3296 | { |
e339a6b0 | 3297 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3298 | } |
3299 | ||
3300 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3301 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3302 | { |
e339a6b0 | 3303 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3304 | } |
3305 | ||
9078a3e1 | 3306 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3307 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3308 | struct btrfs_path *path, |
3309 | struct btrfs_block_group_cache *cache) | |
3310 | { | |
3311 | int ret; | |
0b246afa | 3312 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3313 | unsigned long bi; |
3314 | struct extent_buffer *leaf; | |
9078a3e1 | 3315 | |
9078a3e1 | 3316 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3317 | if (ret) { |
3318 | if (ret > 0) | |
3319 | ret = -ENOENT; | |
54aa1f4d | 3320 | goto fail; |
df95e7f0 | 3321 | } |
5f39d397 CM |
3322 | |
3323 | leaf = path->nodes[0]; | |
3324 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3325 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3326 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3327 | fail: |
24b89d08 | 3328 | btrfs_release_path(path); |
df95e7f0 | 3329 | return ret; |
9078a3e1 CM |
3330 | |
3331 | } | |
3332 | ||
4a8c9a62 | 3333 | static struct btrfs_block_group_cache * |
2ff7e61e | 3334 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3335 | struct btrfs_block_group_cache *cache) |
3336 | { | |
3337 | struct rb_node *node; | |
292cbd51 | 3338 | |
0b246afa | 3339 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3340 | |
3341 | /* If our block group was removed, we need a full search. */ | |
3342 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3343 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3344 | ||
0b246afa | 3345 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3346 | btrfs_put_block_group(cache); |
0b246afa | 3347 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3348 | } |
4a8c9a62 YZ |
3349 | node = rb_next(&cache->cache_node); |
3350 | btrfs_put_block_group(cache); | |
3351 | if (node) { | |
3352 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3353 | cache_node); | |
11dfe35a | 3354 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3355 | } else |
3356 | cache = NULL; | |
0b246afa | 3357 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3358 | return cache; |
3359 | } | |
3360 | ||
0af3d00b JB |
3361 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3362 | struct btrfs_trans_handle *trans, | |
3363 | struct btrfs_path *path) | |
3364 | { | |
0b246afa JM |
3365 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3366 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b JB |
3367 | struct inode *inode = NULL; |
3368 | u64 alloc_hint = 0; | |
2b20982e | 3369 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3370 | u64 num_pages = 0; |
0af3d00b JB |
3371 | int retries = 0; |
3372 | int ret = 0; | |
3373 | ||
3374 | /* | |
3375 | * If this block group is smaller than 100 megs don't bother caching the | |
3376 | * block group. | |
3377 | */ | |
ee22184b | 3378 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3379 | spin_lock(&block_group->lock); |
3380 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3381 | spin_unlock(&block_group->lock); | |
3382 | return 0; | |
3383 | } | |
3384 | ||
0c0ef4bc JB |
3385 | if (trans->aborted) |
3386 | return 0; | |
0af3d00b | 3387 | again: |
77ab86bf | 3388 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3389 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3390 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3391 | btrfs_release_path(path); |
0af3d00b JB |
3392 | goto out; |
3393 | } | |
3394 | ||
3395 | if (IS_ERR(inode)) { | |
3396 | BUG_ON(retries); | |
3397 | retries++; | |
3398 | ||
3399 | if (block_group->ro) | |
3400 | goto out_free; | |
3401 | ||
77ab86bf JM |
3402 | ret = create_free_space_inode(fs_info, trans, block_group, |
3403 | path); | |
0af3d00b JB |
3404 | if (ret) |
3405 | goto out_free; | |
3406 | goto again; | |
3407 | } | |
3408 | ||
5b0e95bf JB |
3409 | /* We've already setup this transaction, go ahead and exit */ |
3410 | if (block_group->cache_generation == trans->transid && | |
3411 | i_size_read(inode)) { | |
3412 | dcs = BTRFS_DC_SETUP; | |
3413 | goto out_put; | |
3414 | } | |
3415 | ||
0af3d00b JB |
3416 | /* |
3417 | * We want to set the generation to 0, that way if anything goes wrong | |
3418 | * from here on out we know not to trust this cache when we load up next | |
3419 | * time. | |
3420 | */ | |
3421 | BTRFS_I(inode)->generation = 0; | |
3422 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3423 | if (ret) { |
3424 | /* | |
3425 | * So theoretically we could recover from this, simply set the | |
3426 | * super cache generation to 0 so we know to invalidate the | |
3427 | * cache, but then we'd have to keep track of the block groups | |
3428 | * that fail this way so we know we _have_ to reset this cache | |
3429 | * before the next commit or risk reading stale cache. So to | |
3430 | * limit our exposure to horrible edge cases lets just abort the | |
3431 | * transaction, this only happens in really bad situations | |
3432 | * anyway. | |
3433 | */ | |
66642832 | 3434 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3435 | goto out_put; |
3436 | } | |
0af3d00b JB |
3437 | WARN_ON(ret); |
3438 | ||
3439 | if (i_size_read(inode) > 0) { | |
2ff7e61e | 3440 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3441 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3442 | if (ret) |
3443 | goto out_put; | |
3444 | ||
77ab86bf | 3445 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3446 | if (ret) |
3447 | goto out_put; | |
3448 | } | |
3449 | ||
3450 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3451 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3452 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3453 | /* |
3454 | * don't bother trying to write stuff out _if_ | |
3455 | * a) we're not cached, | |
1a79c1f2 LB |
3456 | * b) we're with nospace_cache mount option, |
3457 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3458 | */ |
2b20982e | 3459 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3460 | spin_unlock(&block_group->lock); |
3461 | goto out_put; | |
3462 | } | |
3463 | spin_unlock(&block_group->lock); | |
3464 | ||
2968b1f4 JB |
3465 | /* |
3466 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3467 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3468 | */ | |
3469 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3470 | ret = -ENOSPC; | |
3471 | goto out_put; | |
3472 | } | |
3473 | ||
6fc823b1 JB |
3474 | /* |
3475 | * Try to preallocate enough space based on how big the block group is. | |
3476 | * Keep in mind this has to include any pinned space which could end up | |
3477 | * taking up quite a bit since it's not folded into the other space | |
3478 | * cache. | |
3479 | */ | |
ee22184b | 3480 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3481 | if (!num_pages) |
3482 | num_pages = 1; | |
3483 | ||
0af3d00b | 3484 | num_pages *= 16; |
09cbfeaf | 3485 | num_pages *= PAGE_SIZE; |
0af3d00b | 3486 | |
7cf5b976 | 3487 | ret = btrfs_check_data_free_space(inode, 0, num_pages); |
0af3d00b JB |
3488 | if (ret) |
3489 | goto out_put; | |
3490 | ||
3491 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3492 | num_pages, num_pages, | |
3493 | &alloc_hint); | |
2968b1f4 JB |
3494 | /* |
3495 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3496 | * of metadata or split extents when writing the cache out, which means | |
3497 | * we can enospc if we are heavily fragmented in addition to just normal | |
3498 | * out of space conditions. So if we hit this just skip setting up any | |
3499 | * other block groups for this transaction, maybe we'll unpin enough | |
3500 | * space the next time around. | |
3501 | */ | |
2b20982e JB |
3502 | if (!ret) |
3503 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3504 | else if (ret == -ENOSPC) |
3505 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3506 | |
0af3d00b JB |
3507 | out_put: |
3508 | iput(inode); | |
3509 | out_free: | |
b3b4aa74 | 3510 | btrfs_release_path(path); |
0af3d00b JB |
3511 | out: |
3512 | spin_lock(&block_group->lock); | |
e65cbb94 | 3513 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3514 | block_group->cache_generation = trans->transid; |
2b20982e | 3515 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3516 | spin_unlock(&block_group->lock); |
3517 | ||
3518 | return ret; | |
3519 | } | |
3520 | ||
dcdf7f6d | 3521 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3522 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3523 | { |
3524 | struct btrfs_block_group_cache *cache, *tmp; | |
3525 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3526 | struct btrfs_path *path; | |
3527 | ||
3528 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3529 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3530 | return 0; |
3531 | ||
3532 | path = btrfs_alloc_path(); | |
3533 | if (!path) | |
3534 | return -ENOMEM; | |
3535 | ||
3536 | /* Could add new block groups, use _safe just in case */ | |
3537 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3538 | dirty_list) { | |
3539 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3540 | cache_save_setup(cache, trans, path); | |
3541 | } | |
3542 | ||
3543 | btrfs_free_path(path); | |
3544 | return 0; | |
3545 | } | |
3546 | ||
1bbc621e CM |
3547 | /* |
3548 | * transaction commit does final block group cache writeback during a | |
3549 | * critical section where nothing is allowed to change the FS. This is | |
3550 | * required in order for the cache to actually match the block group, | |
3551 | * but can introduce a lot of latency into the commit. | |
3552 | * | |
3553 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3554 | * cache IO. There's a chance we'll have to redo some of it if the | |
3555 | * block group changes again during the commit, but it greatly reduces | |
3556 | * the commit latency by getting rid of the easy block groups while | |
3557 | * we're still allowing others to join the commit. | |
3558 | */ | |
3559 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3560 | struct btrfs_fs_info *fs_info) |
9078a3e1 | 3561 | { |
4a8c9a62 | 3562 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3563 | struct btrfs_transaction *cur_trans = trans->transaction; |
3564 | int ret = 0; | |
c9dc4c65 | 3565 | int should_put; |
1bbc621e CM |
3566 | struct btrfs_path *path = NULL; |
3567 | LIST_HEAD(dirty); | |
3568 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3569 | int num_started = 0; |
1bbc621e CM |
3570 | int loops = 0; |
3571 | ||
3572 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3573 | if (list_empty(&cur_trans->dirty_bgs)) { |
3574 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3575 | return 0; | |
1bbc621e | 3576 | } |
b58d1a9e | 3577 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3578 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3579 | |
1bbc621e | 3580 | again: |
1bbc621e CM |
3581 | /* |
3582 | * make sure all the block groups on our dirty list actually | |
3583 | * exist | |
3584 | */ | |
2ff7e61e | 3585 | btrfs_create_pending_block_groups(trans, fs_info); |
1bbc621e CM |
3586 | |
3587 | if (!path) { | |
3588 | path = btrfs_alloc_path(); | |
3589 | if (!path) | |
3590 | return -ENOMEM; | |
3591 | } | |
3592 | ||
b58d1a9e FM |
3593 | /* |
3594 | * cache_write_mutex is here only to save us from balance or automatic | |
3595 | * removal of empty block groups deleting this block group while we are | |
3596 | * writing out the cache | |
3597 | */ | |
3598 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3599 | while (!list_empty(&dirty)) { |
3600 | cache = list_first_entry(&dirty, | |
3601 | struct btrfs_block_group_cache, | |
3602 | dirty_list); | |
1bbc621e CM |
3603 | /* |
3604 | * this can happen if something re-dirties a block | |
3605 | * group that is already under IO. Just wait for it to | |
3606 | * finish and then do it all again | |
3607 | */ | |
3608 | if (!list_empty(&cache->io_list)) { | |
3609 | list_del_init(&cache->io_list); | |
afdb5718 | 3610 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3611 | btrfs_put_block_group(cache); |
3612 | } | |
3613 | ||
3614 | ||
3615 | /* | |
3616 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3617 | * if it should update the cache_state. Don't delete | |
3618 | * until after we wait. | |
3619 | * | |
3620 | * Since we're not running in the commit critical section | |
3621 | * we need the dirty_bgs_lock to protect from update_block_group | |
3622 | */ | |
3623 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3624 | list_del_init(&cache->dirty_list); | |
3625 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3626 | ||
3627 | should_put = 1; | |
3628 | ||
3629 | cache_save_setup(cache, trans, path); | |
3630 | ||
3631 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3632 | cache->io_ctl.inode = NULL; | |
0b246afa | 3633 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3634 | cache, path); |
1bbc621e CM |
3635 | if (ret == 0 && cache->io_ctl.inode) { |
3636 | num_started++; | |
3637 | should_put = 0; | |
3638 | ||
3639 | /* | |
3640 | * the cache_write_mutex is protecting | |
3641 | * the io_list | |
3642 | */ | |
3643 | list_add_tail(&cache->io_list, io); | |
3644 | } else { | |
3645 | /* | |
3646 | * if we failed to write the cache, the | |
3647 | * generation will be bad and life goes on | |
3648 | */ | |
3649 | ret = 0; | |
3650 | } | |
3651 | } | |
ff1f8250 | 3652 | if (!ret) { |
2ff7e61e JM |
3653 | ret = write_one_cache_group(trans, fs_info, |
3654 | path, cache); | |
ff1f8250 FM |
3655 | /* |
3656 | * Our block group might still be attached to the list | |
3657 | * of new block groups in the transaction handle of some | |
3658 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3659 | * means its block group item isn't yet in the extent | |
3660 | * tree. If this happens ignore the error, as we will | |
3661 | * try again later in the critical section of the | |
3662 | * transaction commit. | |
3663 | */ | |
3664 | if (ret == -ENOENT) { | |
3665 | ret = 0; | |
3666 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3667 | if (list_empty(&cache->dirty_list)) { | |
3668 | list_add_tail(&cache->dirty_list, | |
3669 | &cur_trans->dirty_bgs); | |
3670 | btrfs_get_block_group(cache); | |
3671 | } | |
3672 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3673 | } else if (ret) { | |
66642832 | 3674 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3675 | } |
3676 | } | |
1bbc621e CM |
3677 | |
3678 | /* if its not on the io list, we need to put the block group */ | |
3679 | if (should_put) | |
3680 | btrfs_put_block_group(cache); | |
3681 | ||
3682 | if (ret) | |
3683 | break; | |
b58d1a9e FM |
3684 | |
3685 | /* | |
3686 | * Avoid blocking other tasks for too long. It might even save | |
3687 | * us from writing caches for block groups that are going to be | |
3688 | * removed. | |
3689 | */ | |
3690 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3691 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3692 | } |
b58d1a9e | 3693 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3694 | |
3695 | /* | |
3696 | * go through delayed refs for all the stuff we've just kicked off | |
3697 | * and then loop back (just once) | |
3698 | */ | |
2ff7e61e | 3699 | ret = btrfs_run_delayed_refs(trans, fs_info, 0); |
1bbc621e CM |
3700 | if (!ret && loops == 0) { |
3701 | loops++; | |
3702 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3703 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3704 | /* |
3705 | * dirty_bgs_lock protects us from concurrent block group | |
3706 | * deletes too (not just cache_write_mutex). | |
3707 | */ | |
3708 | if (!list_empty(&dirty)) { | |
3709 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3710 | goto again; | |
3711 | } | |
1bbc621e | 3712 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3713 | } else if (ret < 0) { |
2ff7e61e | 3714 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3715 | } |
3716 | ||
3717 | btrfs_free_path(path); | |
3718 | return ret; | |
3719 | } | |
3720 | ||
3721 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3722 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3723 | { |
3724 | struct btrfs_block_group_cache *cache; | |
3725 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3726 | int ret = 0; | |
3727 | int should_put; | |
3728 | struct btrfs_path *path; | |
3729 | struct list_head *io = &cur_trans->io_bgs; | |
3730 | int num_started = 0; | |
9078a3e1 CM |
3731 | |
3732 | path = btrfs_alloc_path(); | |
3733 | if (!path) | |
3734 | return -ENOMEM; | |
3735 | ||
ce93ec54 | 3736 | /* |
e44081ef FM |
3737 | * Even though we are in the critical section of the transaction commit, |
3738 | * we can still have concurrent tasks adding elements to this | |
3739 | * transaction's list of dirty block groups. These tasks correspond to | |
3740 | * endio free space workers started when writeback finishes for a | |
3741 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3742 | * allocate new block groups as a result of COWing nodes of the root | |
3743 | * tree when updating the free space inode. The writeback for the space | |
3744 | * caches is triggered by an earlier call to | |
3745 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3746 | * loop. | |
3747 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3748 | * delayed refs to make sure we have the best chance at doing this all |
3749 | * in one shot. | |
3750 | */ | |
e44081ef | 3751 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3752 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3753 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3754 | struct btrfs_block_group_cache, | |
3755 | dirty_list); | |
c9dc4c65 CM |
3756 | |
3757 | /* | |
3758 | * this can happen if cache_save_setup re-dirties a block | |
3759 | * group that is already under IO. Just wait for it to | |
3760 | * finish and then do it all again | |
3761 | */ | |
3762 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3763 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3764 | list_del_init(&cache->io_list); |
afdb5718 | 3765 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3766 | btrfs_put_block_group(cache); |
e44081ef | 3767 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3768 | } |
3769 | ||
1bbc621e CM |
3770 | /* |
3771 | * don't remove from the dirty list until after we've waited | |
3772 | * on any pending IO | |
3773 | */ | |
ce93ec54 | 3774 | list_del_init(&cache->dirty_list); |
e44081ef | 3775 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3776 | should_put = 1; |
3777 | ||
1bbc621e | 3778 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3779 | |
ce93ec54 | 3780 | if (!ret) |
2ff7e61e JM |
3781 | ret = btrfs_run_delayed_refs(trans, fs_info, |
3782 | (unsigned long) -1); | |
c9dc4c65 CM |
3783 | |
3784 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3785 | cache->io_ctl.inode = NULL; | |
0b246afa | 3786 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3787 | cache, path); |
c9dc4c65 CM |
3788 | if (ret == 0 && cache->io_ctl.inode) { |
3789 | num_started++; | |
3790 | should_put = 0; | |
1bbc621e | 3791 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3792 | } else { |
3793 | /* | |
3794 | * if we failed to write the cache, the | |
3795 | * generation will be bad and life goes on | |
3796 | */ | |
3797 | ret = 0; | |
3798 | } | |
3799 | } | |
ff1f8250 | 3800 | if (!ret) { |
2ff7e61e JM |
3801 | ret = write_one_cache_group(trans, fs_info, |
3802 | path, cache); | |
2bc0bb5f FM |
3803 | /* |
3804 | * One of the free space endio workers might have | |
3805 | * created a new block group while updating a free space | |
3806 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3807 | * and hasn't released its transaction handle yet, in | |
3808 | * which case the new block group is still attached to | |
3809 | * its transaction handle and its creation has not | |
3810 | * finished yet (no block group item in the extent tree | |
3811 | * yet, etc). If this is the case, wait for all free | |
3812 | * space endio workers to finish and retry. This is a | |
3813 | * a very rare case so no need for a more efficient and | |
3814 | * complex approach. | |
3815 | */ | |
3816 | if (ret == -ENOENT) { | |
3817 | wait_event(cur_trans->writer_wait, | |
3818 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3819 | ret = write_one_cache_group(trans, fs_info, |
3820 | path, cache); | |
2bc0bb5f | 3821 | } |
ff1f8250 | 3822 | if (ret) |
66642832 | 3823 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3824 | } |
c9dc4c65 CM |
3825 | |
3826 | /* if its not on the io list, we need to put the block group */ | |
3827 | if (should_put) | |
3828 | btrfs_put_block_group(cache); | |
e44081ef | 3829 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3830 | } |
e44081ef | 3831 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3832 | |
1bbc621e CM |
3833 | while (!list_empty(io)) { |
3834 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3835 | io_list); |
3836 | list_del_init(&cache->io_list); | |
afdb5718 | 3837 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3838 | btrfs_put_block_group(cache); |
3839 | } | |
3840 | ||
9078a3e1 | 3841 | btrfs_free_path(path); |
ce93ec54 | 3842 | return ret; |
9078a3e1 CM |
3843 | } |
3844 | ||
2ff7e61e | 3845 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3846 | { |
3847 | struct btrfs_block_group_cache *block_group; | |
3848 | int readonly = 0; | |
3849 | ||
0b246afa | 3850 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3851 | if (!block_group || block_group->ro) |
3852 | readonly = 1; | |
3853 | if (block_group) | |
fa9c0d79 | 3854 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3855 | return readonly; |
3856 | } | |
3857 | ||
f78c436c FM |
3858 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3859 | { | |
3860 | struct btrfs_block_group_cache *bg; | |
3861 | bool ret = true; | |
3862 | ||
3863 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3864 | if (!bg) | |
3865 | return false; | |
3866 | ||
3867 | spin_lock(&bg->lock); | |
3868 | if (bg->ro) | |
3869 | ret = false; | |
3870 | else | |
3871 | atomic_inc(&bg->nocow_writers); | |
3872 | spin_unlock(&bg->lock); | |
3873 | ||
3874 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3875 | if (!ret) | |
3876 | btrfs_put_block_group(bg); | |
3877 | ||
3878 | return ret; | |
3879 | ||
3880 | } | |
3881 | ||
3882 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3883 | { | |
3884 | struct btrfs_block_group_cache *bg; | |
3885 | ||
3886 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3887 | ASSERT(bg); | |
3888 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
3889 | wake_up_atomic_t(&bg->nocow_writers); | |
3890 | /* | |
3891 | * Once for our lookup and once for the lookup done by a previous call | |
3892 | * to btrfs_inc_nocow_writers() | |
3893 | */ | |
3894 | btrfs_put_block_group(bg); | |
3895 | btrfs_put_block_group(bg); | |
3896 | } | |
3897 | ||
3898 | static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a) | |
3899 | { | |
3900 | schedule(); | |
3901 | return 0; | |
3902 | } | |
3903 | ||
3904 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) | |
3905 | { | |
3906 | wait_on_atomic_t(&bg->nocow_writers, | |
3907 | btrfs_wait_nocow_writers_atomic_t, | |
3908 | TASK_UNINTERRUPTIBLE); | |
3909 | } | |
3910 | ||
6ab0a202 JM |
3911 | static const char *alloc_name(u64 flags) |
3912 | { | |
3913 | switch (flags) { | |
3914 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3915 | return "mixed"; | |
3916 | case BTRFS_BLOCK_GROUP_METADATA: | |
3917 | return "metadata"; | |
3918 | case BTRFS_BLOCK_GROUP_DATA: | |
3919 | return "data"; | |
3920 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3921 | return "system"; | |
3922 | default: | |
3923 | WARN_ON(1); | |
3924 | return "invalid-combination"; | |
3925 | }; | |
3926 | } | |
3927 | ||
593060d7 CM |
3928 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3929 | u64 total_bytes, u64 bytes_used, | |
e40edf2d | 3930 | u64 bytes_readonly, |
593060d7 CM |
3931 | struct btrfs_space_info **space_info) |
3932 | { | |
3933 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3934 | int i; |
3935 | int factor; | |
b150a4f1 | 3936 | int ret; |
b742bb82 YZ |
3937 | |
3938 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3939 | BTRFS_BLOCK_GROUP_RAID10)) | |
3940 | factor = 2; | |
3941 | else | |
3942 | factor = 1; | |
593060d7 CM |
3943 | |
3944 | found = __find_space_info(info, flags); | |
3945 | if (found) { | |
25179201 | 3946 | spin_lock(&found->lock); |
593060d7 | 3947 | found->total_bytes += total_bytes; |
89a55897 | 3948 | found->disk_total += total_bytes * factor; |
593060d7 | 3949 | found->bytes_used += bytes_used; |
b742bb82 | 3950 | found->disk_used += bytes_used * factor; |
e40edf2d | 3951 | found->bytes_readonly += bytes_readonly; |
2e6e5183 FM |
3952 | if (total_bytes > 0) |
3953 | found->full = 0; | |
957780eb JB |
3954 | space_info_add_new_bytes(info, found, total_bytes - |
3955 | bytes_used - bytes_readonly); | |
25179201 | 3956 | spin_unlock(&found->lock); |
593060d7 CM |
3957 | *space_info = found; |
3958 | return 0; | |
3959 | } | |
c146afad | 3960 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3961 | if (!found) |
3962 | return -ENOMEM; | |
3963 | ||
908c7f19 | 3964 | ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL); |
b150a4f1 JB |
3965 | if (ret) { |
3966 | kfree(found); | |
3967 | return ret; | |
3968 | } | |
3969 | ||
c1895442 | 3970 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
b742bb82 | 3971 | INIT_LIST_HEAD(&found->block_groups[i]); |
80eb234a | 3972 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3973 | spin_lock_init(&found->lock); |
52ba6929 | 3974 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3975 | found->total_bytes = total_bytes; |
89a55897 | 3976 | found->disk_total = total_bytes * factor; |
593060d7 | 3977 | found->bytes_used = bytes_used; |
b742bb82 | 3978 | found->disk_used = bytes_used * factor; |
593060d7 | 3979 | found->bytes_pinned = 0; |
e8569813 | 3980 | found->bytes_reserved = 0; |
e40edf2d | 3981 | found->bytes_readonly = bytes_readonly; |
f0486c68 | 3982 | found->bytes_may_use = 0; |
6af3e3ad | 3983 | found->full = 0; |
4f4db217 | 3984 | found->max_extent_size = 0; |
0e4f8f88 | 3985 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3986 | found->chunk_alloc = 0; |
fdb5effd JB |
3987 | found->flush = 0; |
3988 | init_waitqueue_head(&found->wait); | |
633c0aad | 3989 | INIT_LIST_HEAD(&found->ro_bgs); |
957780eb JB |
3990 | INIT_LIST_HEAD(&found->tickets); |
3991 | INIT_LIST_HEAD(&found->priority_tickets); | |
6ab0a202 JM |
3992 | |
3993 | ret = kobject_init_and_add(&found->kobj, &space_info_ktype, | |
3994 | info->space_info_kobj, "%s", | |
3995 | alloc_name(found->flags)); | |
3996 | if (ret) { | |
896533a7 | 3997 | percpu_counter_destroy(&found->total_bytes_pinned); |
6ab0a202 JM |
3998 | kfree(found); |
3999 | return ret; | |
4000 | } | |
4001 | ||
593060d7 | 4002 | *space_info = found; |
4184ea7f | 4003 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
4004 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4005 | info->data_sinfo = found; | |
6ab0a202 JM |
4006 | |
4007 | return ret; | |
593060d7 CM |
4008 | } |
4009 | ||
8790d502 CM |
4010 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
4011 | { | |
899c81ea ID |
4012 | u64 extra_flags = chunk_to_extended(flags) & |
4013 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4014 | |
de98ced9 | 4015 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4016 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4017 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4018 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4019 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4020 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4021 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4022 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4023 | } |
593060d7 | 4024 | |
fc67c450 ID |
4025 | /* |
4026 | * returns target flags in extended format or 0 if restripe for this | |
4027 | * chunk_type is not in progress | |
c6664b42 ID |
4028 | * |
4029 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
4030 | */ |
4031 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4032 | { | |
4033 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4034 | u64 target = 0; | |
4035 | ||
fc67c450 ID |
4036 | if (!bctl) |
4037 | return 0; | |
4038 | ||
4039 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4040 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4041 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4042 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4043 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4044 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4045 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4046 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4047 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4048 | } | |
4049 | ||
4050 | return target; | |
4051 | } | |
4052 | ||
a46d11a8 ID |
4053 | /* |
4054 | * @flags: available profiles in extended format (see ctree.h) | |
4055 | * | |
e4d8ec0f ID |
4056 | * Returns reduced profile in chunk format. If profile changing is in |
4057 | * progress (either running or paused) picks the target profile (if it's | |
4058 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4059 | */ |
2ff7e61e | 4060 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4061 | { |
0b246afa | 4062 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4063 | u64 target; |
9c170b26 ZL |
4064 | u64 raid_type; |
4065 | u64 allowed = 0; | |
a061fc8d | 4066 | |
fc67c450 ID |
4067 | /* |
4068 | * see if restripe for this chunk_type is in progress, if so | |
4069 | * try to reduce to the target profile | |
4070 | */ | |
0b246afa JM |
4071 | spin_lock(&fs_info->balance_lock); |
4072 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4073 | if (target) { |
4074 | /* pick target profile only if it's already available */ | |
4075 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4076 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4077 | return extended_to_chunk(target); |
e4d8ec0f ID |
4078 | } |
4079 | } | |
0b246afa | 4080 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4081 | |
53b381b3 | 4082 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4083 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4084 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
4085 | allowed |= btrfs_raid_group[raid_type]; | |
4086 | } | |
4087 | allowed &= flags; | |
4088 | ||
4089 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4090 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4091 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4092 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4093 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4094 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4095 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4096 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4097 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4098 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4099 | ||
4100 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4101 | ||
4102 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4103 | } |
4104 | ||
2ff7e61e | 4105 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4106 | { |
de98ced9 | 4107 | unsigned seq; |
f8213bdc | 4108 | u64 flags; |
de98ced9 MX |
4109 | |
4110 | do { | |
f8213bdc | 4111 | flags = orig_flags; |
0b246afa | 4112 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4113 | |
4114 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4115 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4116 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4117 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4118 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4119 | flags |= fs_info->avail_metadata_alloc_bits; |
4120 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4121 | |
2ff7e61e | 4122 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4123 | } |
4124 | ||
1b86826d | 4125 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4126 | { |
0b246afa | 4127 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4128 | u64 flags; |
53b381b3 | 4129 | u64 ret; |
9ed74f2d | 4130 | |
b742bb82 YZ |
4131 | if (data) |
4132 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4133 | else if (root == fs_info->chunk_root) |
b742bb82 | 4134 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4135 | else |
b742bb82 | 4136 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4137 | |
2ff7e61e | 4138 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4139 | return ret; |
6a63209f | 4140 | } |
9ed74f2d | 4141 | |
1b86826d JM |
4142 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4143 | { | |
4144 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4145 | } | |
4146 | ||
4147 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4148 | { | |
4149 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4150 | } | |
4151 | ||
4152 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4153 | { | |
4154 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4155 | } | |
4156 | ||
4136135b LB |
4157 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4158 | bool may_use_included) | |
4159 | { | |
4160 | ASSERT(s_info); | |
4161 | return s_info->bytes_used + s_info->bytes_reserved + | |
4162 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4163 | (may_use_included ? s_info->bytes_may_use : 0); | |
4164 | } | |
4165 | ||
04f4f916 | 4166 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4167 | { |
6a63209f | 4168 | struct btrfs_space_info *data_sinfo; |
04f4f916 | 4169 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4170 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 4171 | u64 used; |
94b947b2 | 4172 | int ret = 0; |
c99f1b0c ZL |
4173 | int need_commit = 2; |
4174 | int have_pinned_space; | |
6a63209f | 4175 | |
6a63209f | 4176 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4177 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4178 | |
9dced186 | 4179 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4180 | need_commit = 0; |
9dced186 | 4181 | ASSERT(current->journal_info); |
0af3d00b JB |
4182 | } |
4183 | ||
b4d7c3c9 | 4184 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
4185 | if (!data_sinfo) |
4186 | goto alloc; | |
9ed74f2d | 4187 | |
6a63209f JB |
4188 | again: |
4189 | /* make sure we have enough space to handle the data first */ | |
4190 | spin_lock(&data_sinfo->lock); | |
4136135b | 4191 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4192 | |
4193 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4194 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4195 | |
6a63209f JB |
4196 | /* |
4197 | * if we don't have enough free bytes in this space then we need | |
4198 | * to alloc a new chunk. | |
4199 | */ | |
b9fd47cd | 4200 | if (!data_sinfo->full) { |
6a63209f | 4201 | u64 alloc_target; |
9ed74f2d | 4202 | |
0e4f8f88 | 4203 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4204 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 4205 | alloc: |
1b86826d | 4206 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4207 | /* |
4208 | * It is ugly that we don't call nolock join | |
4209 | * transaction for the free space inode case here. | |
4210 | * But it is safe because we only do the data space | |
4211 | * reservation for the free space cache in the | |
4212 | * transaction context, the common join transaction | |
4213 | * just increase the counter of the current transaction | |
4214 | * handler, doesn't try to acquire the trans_lock of | |
4215 | * the fs. | |
4216 | */ | |
7a7eaa40 | 4217 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4218 | if (IS_ERR(trans)) |
4219 | return PTR_ERR(trans); | |
9ed74f2d | 4220 | |
2ff7e61e | 4221 | ret = do_chunk_alloc(trans, fs_info, alloc_target, |
0e4f8f88 | 4222 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4223 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4224 | if (ret < 0) { |
4225 | if (ret != -ENOSPC) | |
4226 | return ret; | |
c99f1b0c ZL |
4227 | else { |
4228 | have_pinned_space = 1; | |
d52a5b5f | 4229 | goto commit_trans; |
c99f1b0c | 4230 | } |
d52a5b5f | 4231 | } |
9ed74f2d | 4232 | |
b4d7c3c9 LZ |
4233 | if (!data_sinfo) |
4234 | data_sinfo = fs_info->data_sinfo; | |
4235 | ||
6a63209f JB |
4236 | goto again; |
4237 | } | |
f2bb8f5c JB |
4238 | |
4239 | /* | |
b150a4f1 | 4240 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4241 | * allocation, and no removed chunk in current transaction, |
4242 | * don't bother committing the transaction. | |
f2bb8f5c | 4243 | */ |
c99f1b0c ZL |
4244 | have_pinned_space = percpu_counter_compare( |
4245 | &data_sinfo->total_bytes_pinned, | |
4246 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4247 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4248 | |
4e06bdd6 | 4249 | /* commit the current transaction and try again */ |
d52a5b5f | 4250 | commit_trans: |
c99f1b0c | 4251 | if (need_commit && |
0b246afa | 4252 | !atomic_read(&fs_info->open_ioctl_trans)) { |
c99f1b0c | 4253 | need_commit--; |
b150a4f1 | 4254 | |
e1746e83 ZL |
4255 | if (need_commit > 0) { |
4256 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
0b246afa JM |
4257 | btrfs_wait_ordered_roots(fs_info, -1, 0, |
4258 | (u64)-1); | |
e1746e83 | 4259 | } |
9a4e7276 | 4260 | |
7a7eaa40 | 4261 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4262 | if (IS_ERR(trans)) |
4263 | return PTR_ERR(trans); | |
c99f1b0c | 4264 | if (have_pinned_space >= 0 || |
3204d33c JB |
4265 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4266 | &trans->transaction->flags) || | |
c99f1b0c | 4267 | need_commit > 0) { |
3a45bb20 | 4268 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4269 | if (ret) |
4270 | return ret; | |
d7c15171 | 4271 | /* |
c2d6cb16 FM |
4272 | * The cleaner kthread might still be doing iput |
4273 | * operations. Wait for it to finish so that | |
4274 | * more space is released. | |
d7c15171 | 4275 | */ |
0b246afa JM |
4276 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4277 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4278 | goto again; |
4279 | } else { | |
3a45bb20 | 4280 | btrfs_end_transaction(trans); |
94b947b2 | 4281 | } |
4e06bdd6 | 4282 | } |
9ed74f2d | 4283 | |
0b246afa | 4284 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4285 | "space_info:enospc", |
4286 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4287 | return -ENOSPC; |
4288 | } | |
4289 | data_sinfo->bytes_may_use += bytes; | |
0b246afa | 4290 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4291 | data_sinfo->flags, bytes, 1); |
6a63209f | 4292 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4293 | |
237c0e9f | 4294 | return ret; |
9ed74f2d | 4295 | } |
6a63209f | 4296 | |
4ceff079 QW |
4297 | /* |
4298 | * New check_data_free_space() with ability for precious data reservation | |
4299 | * Will replace old btrfs_check_data_free_space(), but for patch split, | |
4300 | * add a new function first and then replace it. | |
4301 | */ | |
7cf5b976 | 4302 | int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len) |
4ceff079 | 4303 | { |
0b246afa | 4304 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4305 | int ret; |
4306 | ||
4307 | /* align the range */ | |
0b246afa JM |
4308 | len = round_up(start + len, fs_info->sectorsize) - |
4309 | round_down(start, fs_info->sectorsize); | |
4310 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4311 | |
04f4f916 | 4312 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4313 | if (ret < 0) |
4314 | return ret; | |
4315 | ||
1e5ec2e7 | 4316 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
4ceff079 | 4317 | ret = btrfs_qgroup_reserve_data(inode, start, len); |
1e5ec2e7 JB |
4318 | if (ret) |
4319 | btrfs_free_reserved_data_space_noquota(inode, start, len); | |
4ceff079 QW |
4320 | return ret; |
4321 | } | |
4322 | ||
4ceff079 QW |
4323 | /* |
4324 | * Called if we need to clear a data reservation for this inode | |
4325 | * Normally in a error case. | |
4326 | * | |
51773bec QW |
4327 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4328 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4329 | * Like clear_bit_hook(). | |
4ceff079 | 4330 | */ |
51773bec QW |
4331 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4332 | u64 len) | |
4ceff079 | 4333 | { |
0b246afa | 4334 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4335 | struct btrfs_space_info *data_sinfo; |
4336 | ||
4337 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4338 | len = round_up(start + len, fs_info->sectorsize) - |
4339 | round_down(start, fs_info->sectorsize); | |
4340 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4341 | |
0b246afa | 4342 | data_sinfo = fs_info->data_sinfo; |
4ceff079 QW |
4343 | spin_lock(&data_sinfo->lock); |
4344 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4345 | data_sinfo->bytes_may_use = 0; | |
4346 | else | |
4347 | data_sinfo->bytes_may_use -= len; | |
0b246afa | 4348 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4349 | data_sinfo->flags, len, 0); |
4350 | spin_unlock(&data_sinfo->lock); | |
4351 | } | |
4352 | ||
51773bec QW |
4353 | /* |
4354 | * Called if we need to clear a data reservation for this inode | |
4355 | * Normally in a error case. | |
4356 | * | |
01327610 | 4357 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4358 | * space framework. |
4359 | */ | |
4360 | void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len) | |
4361 | { | |
0c476a5d JM |
4362 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4363 | ||
4364 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4365 | len = round_up(start + len, root->fs_info->sectorsize) - |
4366 | round_down(start, root->fs_info->sectorsize); | |
4367 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4368 | |
51773bec QW |
4369 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
4370 | btrfs_qgroup_free_data(inode, start, len); | |
4371 | } | |
4372 | ||
97e728d4 | 4373 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4374 | { |
97e728d4 JB |
4375 | struct list_head *head = &info->space_info; |
4376 | struct btrfs_space_info *found; | |
e3ccfa98 | 4377 | |
97e728d4 JB |
4378 | rcu_read_lock(); |
4379 | list_for_each_entry_rcu(found, head, list) { | |
4380 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4381 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4382 | } |
97e728d4 | 4383 | rcu_read_unlock(); |
e3ccfa98 JB |
4384 | } |
4385 | ||
3c76cd84 MX |
4386 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4387 | { | |
4388 | return (global->size << 1); | |
4389 | } | |
4390 | ||
2ff7e61e | 4391 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4392 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4393 | { |
0b246afa | 4394 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
424499db | 4395 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 4396 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 4397 | u64 thresh; |
e3ccfa98 | 4398 | |
0e4f8f88 CM |
4399 | if (force == CHUNK_ALLOC_FORCE) |
4400 | return 1; | |
4401 | ||
fb25e914 JB |
4402 | /* |
4403 | * We need to take into account the global rsv because for all intents | |
4404 | * and purposes it's used space. Don't worry about locking the | |
4405 | * global_rsv, it doesn't change except when the transaction commits. | |
4406 | */ | |
54338b5c | 4407 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3c76cd84 | 4408 | num_allocated += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4409 | |
0e4f8f88 CM |
4410 | /* |
4411 | * in limited mode, we want to have some free space up to | |
4412 | * about 1% of the FS size. | |
4413 | */ | |
4414 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4415 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4416 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 CM |
4417 | |
4418 | if (num_bytes - num_allocated < thresh) | |
4419 | return 1; | |
4420 | } | |
0e4f8f88 | 4421 | |
ee22184b | 4422 | if (num_allocated + SZ_2M < div_factor(num_bytes, 8)) |
14ed0ca6 | 4423 | return 0; |
424499db | 4424 | return 1; |
32c00aff JB |
4425 | } |
4426 | ||
2ff7e61e | 4427 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4428 | { |
4429 | u64 num_dev; | |
4430 | ||
53b381b3 DW |
4431 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4432 | BTRFS_BLOCK_GROUP_RAID0 | | |
4433 | BTRFS_BLOCK_GROUP_RAID5 | | |
4434 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4435 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4436 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4437 | num_dev = 2; | |
4438 | else | |
4439 | num_dev = 1; /* DUP or single */ | |
4440 | ||
39c2d7fa | 4441 | return num_dev; |
15d1ff81 LB |
4442 | } |
4443 | ||
39c2d7fa FM |
4444 | /* |
4445 | * If @is_allocation is true, reserve space in the system space info necessary | |
4446 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4447 | * removing a chunk. | |
4448 | */ | |
4449 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 4450 | struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4451 | { |
4452 | struct btrfs_space_info *info; | |
4453 | u64 left; | |
4454 | u64 thresh; | |
4fbcdf66 | 4455 | int ret = 0; |
39c2d7fa | 4456 | u64 num_devs; |
4fbcdf66 FM |
4457 | |
4458 | /* | |
4459 | * Needed because we can end up allocating a system chunk and for an | |
4460 | * atomic and race free space reservation in the chunk block reserve. | |
4461 | */ | |
0b246afa | 4462 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
15d1ff81 | 4463 | |
0b246afa | 4464 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4465 | spin_lock(&info->lock); |
4136135b | 4466 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4467 | spin_unlock(&info->lock); |
4468 | ||
2ff7e61e | 4469 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4470 | |
4471 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4472 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4473 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4474 | |
0b246afa JM |
4475 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4476 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4477 | left, thresh, type); | |
4478 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4479 | } |
4480 | ||
4481 | if (left < thresh) { | |
1b86826d | 4482 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4483 | |
4fbcdf66 FM |
4484 | /* |
4485 | * Ignore failure to create system chunk. We might end up not | |
4486 | * needing it, as we might not need to COW all nodes/leafs from | |
4487 | * the paths we visit in the chunk tree (they were already COWed | |
4488 | * or created in the current transaction for example). | |
4489 | */ | |
2ff7e61e | 4490 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
4fbcdf66 FM |
4491 | } |
4492 | ||
4493 | if (!ret) { | |
0b246afa JM |
4494 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4495 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4496 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4497 | if (!ret) | |
4498 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4499 | } |
4500 | } | |
4501 | ||
28b737f6 LB |
4502 | /* |
4503 | * If force is CHUNK_ALLOC_FORCE: | |
4504 | * - return 1 if it successfully allocates a chunk, | |
4505 | * - return errors including -ENOSPC otherwise. | |
4506 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4507 | * - return 0 if it doesn't need to allocate a new chunk, | |
4508 | * - return 1 if it successfully allocates a chunk, | |
4509 | * - return errors including -ENOSPC otherwise. | |
4510 | */ | |
6324fbf3 | 4511 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 4512 | struct btrfs_fs_info *fs_info, u64 flags, int force) |
9ed74f2d | 4513 | { |
6324fbf3 | 4514 | struct btrfs_space_info *space_info; |
6d74119f | 4515 | int wait_for_alloc = 0; |
9ed74f2d | 4516 | int ret = 0; |
9ed74f2d | 4517 | |
c6b305a8 JB |
4518 | /* Don't re-enter if we're already allocating a chunk */ |
4519 | if (trans->allocating_chunk) | |
4520 | return -ENOSPC; | |
4521 | ||
0b246afa | 4522 | space_info = __find_space_info(fs_info, flags); |
593060d7 | 4523 | if (!space_info) { |
0b246afa | 4524 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
79787eaa | 4525 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 4526 | } |
79787eaa | 4527 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 4528 | |
6d74119f | 4529 | again: |
25179201 | 4530 | spin_lock(&space_info->lock); |
9e622d6b | 4531 | if (force < space_info->force_alloc) |
0e4f8f88 | 4532 | force = space_info->force_alloc; |
25179201 | 4533 | if (space_info->full) { |
2ff7e61e | 4534 | if (should_alloc_chunk(fs_info, space_info, force)) |
09fb99a6 FDBM |
4535 | ret = -ENOSPC; |
4536 | else | |
4537 | ret = 0; | |
25179201 | 4538 | spin_unlock(&space_info->lock); |
09fb99a6 | 4539 | return ret; |
9ed74f2d JB |
4540 | } |
4541 | ||
2ff7e61e | 4542 | if (!should_alloc_chunk(fs_info, space_info, force)) { |
25179201 | 4543 | spin_unlock(&space_info->lock); |
6d74119f JB |
4544 | return 0; |
4545 | } else if (space_info->chunk_alloc) { | |
4546 | wait_for_alloc = 1; | |
4547 | } else { | |
4548 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4549 | } |
0e4f8f88 | 4550 | |
25179201 | 4551 | spin_unlock(&space_info->lock); |
9ed74f2d | 4552 | |
6d74119f JB |
4553 | mutex_lock(&fs_info->chunk_mutex); |
4554 | ||
4555 | /* | |
4556 | * The chunk_mutex is held throughout the entirety of a chunk | |
4557 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4558 | * other guy is done and we need to recheck and see if we should | |
4559 | * allocate. | |
4560 | */ | |
4561 | if (wait_for_alloc) { | |
4562 | mutex_unlock(&fs_info->chunk_mutex); | |
4563 | wait_for_alloc = 0; | |
4564 | goto again; | |
4565 | } | |
4566 | ||
c6b305a8 JB |
4567 | trans->allocating_chunk = true; |
4568 | ||
67377734 JB |
4569 | /* |
4570 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4571 | * allocating mixed chunks instead of individual chunks. | |
4572 | */ | |
4573 | if (btrfs_mixed_space_info(space_info)) | |
4574 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4575 | ||
97e728d4 JB |
4576 | /* |
4577 | * if we're doing a data chunk, go ahead and make sure that | |
4578 | * we keep a reasonable number of metadata chunks allocated in the | |
4579 | * FS as well. | |
4580 | */ | |
9ed74f2d | 4581 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4582 | fs_info->data_chunk_allocations++; |
4583 | if (!(fs_info->data_chunk_allocations % | |
4584 | fs_info->metadata_ratio)) | |
4585 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4586 | } |
4587 | ||
15d1ff81 LB |
4588 | /* |
4589 | * Check if we have enough space in SYSTEM chunk because we may need | |
4590 | * to update devices. | |
4591 | */ | |
2ff7e61e | 4592 | check_system_chunk(trans, fs_info, flags); |
15d1ff81 | 4593 | |
2ff7e61e | 4594 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
c6b305a8 | 4595 | trans->allocating_chunk = false; |
92b8e897 | 4596 | |
9ed74f2d | 4597 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4598 | if (ret < 0 && ret != -ENOSPC) |
4599 | goto out; | |
9ed74f2d | 4600 | if (ret) |
6324fbf3 | 4601 | space_info->full = 1; |
424499db YZ |
4602 | else |
4603 | ret = 1; | |
6d74119f | 4604 | |
0e4f8f88 | 4605 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4606 | out: |
6d74119f | 4607 | space_info->chunk_alloc = 0; |
9ed74f2d | 4608 | spin_unlock(&space_info->lock); |
a25c75d5 | 4609 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4610 | /* |
4611 | * When we allocate a new chunk we reserve space in the chunk block | |
4612 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4613 | * add new nodes/leafs to it if we end up needing to do it when | |
4614 | * inserting the chunk item and updating device items as part of the | |
4615 | * second phase of chunk allocation, performed by | |
4616 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4617 | * large number of new block groups to create in our transaction | |
4618 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4619 | * in extreme cases - like having a single transaction create many new | |
4620 | * block groups when starting to write out the free space caches of all | |
4621 | * the block groups that were made dirty during the lifetime of the | |
4622 | * transaction. | |
4623 | */ | |
d9a0540a | 4624 | if (trans->can_flush_pending_bgs && |
ee22184b | 4625 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
2ff7e61e | 4626 | btrfs_create_pending_block_groups(trans, fs_info); |
00d80e34 FM |
4627 | btrfs_trans_release_chunk_metadata(trans); |
4628 | } | |
0f9dd46c | 4629 | return ret; |
6324fbf3 | 4630 | } |
9ed74f2d | 4631 | |
c1c4919b | 4632 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4633 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4634 | enum btrfs_reserve_flush_enum flush, |
4635 | bool system_chunk) | |
a80c8dcf | 4636 | { |
0b246afa | 4637 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4638 | u64 profile; |
3c76cd84 | 4639 | u64 space_size; |
a80c8dcf JB |
4640 | u64 avail; |
4641 | u64 used; | |
4642 | ||
957780eb JB |
4643 | /* Don't overcommit when in mixed mode. */ |
4644 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4645 | return 0; | |
4646 | ||
c1c4919b JM |
4647 | if (system_chunk) |
4648 | profile = btrfs_system_alloc_profile(fs_info); | |
4649 | else | |
4650 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4651 | ||
4136135b | 4652 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4653 | |
96f1bb57 JB |
4654 | /* |
4655 | * We only want to allow over committing if we have lots of actual space | |
4656 | * free, but if we don't have enough space to handle the global reserve | |
4657 | * space then we could end up having a real enospc problem when trying | |
4658 | * to allocate a chunk or some other such important allocation. | |
4659 | */ | |
3c76cd84 MX |
4660 | spin_lock(&global_rsv->lock); |
4661 | space_size = calc_global_rsv_need_space(global_rsv); | |
4662 | spin_unlock(&global_rsv->lock); | |
4663 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4664 | return 0; |
4665 | ||
4666 | used += space_info->bytes_may_use; | |
a80c8dcf | 4667 | |
a5ed45f8 | 4668 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4669 | |
4670 | /* | |
4671 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4672 | * space is actually useable. For raid56, the space info used |
4673 | * doesn't include the parity drive, so we don't have to | |
4674 | * change the math | |
a80c8dcf JB |
4675 | */ |
4676 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4677 | BTRFS_BLOCK_GROUP_RAID1 | | |
4678 | BTRFS_BLOCK_GROUP_RAID10)) | |
4679 | avail >>= 1; | |
4680 | ||
4681 | /* | |
561c294d MX |
4682 | * If we aren't flushing all things, let us overcommit up to |
4683 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4684 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4685 | */ |
08e007d2 | 4686 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4687 | avail >>= 3; |
a80c8dcf | 4688 | else |
14575aef | 4689 | avail >>= 1; |
a80c8dcf | 4690 | |
14575aef | 4691 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4692 | return 1; |
4693 | return 0; | |
4694 | } | |
4695 | ||
2ff7e61e | 4696 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4697 | unsigned long nr_pages, int nr_items) |
da633a42 | 4698 | { |
0b246afa | 4699 | struct super_block *sb = fs_info->sb; |
da633a42 | 4700 | |
925a6efb JB |
4701 | if (down_read_trylock(&sb->s_umount)) { |
4702 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4703 | up_read(&sb->s_umount); | |
4704 | } else { | |
da633a42 MX |
4705 | /* |
4706 | * We needn't worry the filesystem going from r/w to r/o though | |
4707 | * we don't acquire ->s_umount mutex, because the filesystem | |
4708 | * should guarantee the delalloc inodes list be empty after | |
4709 | * the filesystem is readonly(all dirty pages are written to | |
4710 | * the disk). | |
4711 | */ | |
0b246afa | 4712 | btrfs_start_delalloc_roots(fs_info, 0, nr_items); |
98ad69cf | 4713 | if (!current->journal_info) |
0b246afa | 4714 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4715 | } |
4716 | } | |
4717 | ||
2ff7e61e JM |
4718 | static inline int calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
4719 | u64 to_reclaim) | |
18cd8ea6 MX |
4720 | { |
4721 | u64 bytes; | |
4722 | int nr; | |
4723 | ||
2ff7e61e | 4724 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
18cd8ea6 MX |
4725 | nr = (int)div64_u64(to_reclaim, bytes); |
4726 | if (!nr) | |
4727 | nr = 1; | |
4728 | return nr; | |
4729 | } | |
4730 | ||
ee22184b | 4731 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4732 | |
9ed74f2d | 4733 | /* |
5da9d01b | 4734 | * shrink metadata reservation for delalloc |
9ed74f2d | 4735 | */ |
c1c4919b JM |
4736 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4737 | u64 orig, bool wait_ordered) | |
5da9d01b | 4738 | { |
0ca1f7ce | 4739 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4740 | struct btrfs_space_info *space_info; |
663350ac | 4741 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4742 | u64 delalloc_bytes; |
5da9d01b | 4743 | u64 max_reclaim; |
b1953bce | 4744 | long time_left; |
d3ee29e3 MX |
4745 | unsigned long nr_pages; |
4746 | int loops; | |
b0244199 | 4747 | int items; |
08e007d2 | 4748 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4749 | |
c61a16a7 | 4750 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4751 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
8eb0dfdb | 4752 | to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4753 | |
663350ac | 4754 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0b246afa | 4755 | block_rsv = &fs_info->delalloc_block_rsv; |
0019f10d | 4756 | space_info = block_rsv->space_info; |
bf9022e0 | 4757 | |
963d678b | 4758 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4759 | &fs_info->delalloc_bytes); |
f4c738c2 | 4760 | if (delalloc_bytes == 0) { |
fdb5effd | 4761 | if (trans) |
f4c738c2 | 4762 | return; |
38c135af | 4763 | if (wait_ordered) |
0b246afa | 4764 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4765 | return; |
fdb5effd JB |
4766 | } |
4767 | ||
d3ee29e3 | 4768 | loops = 0; |
f4c738c2 JB |
4769 | while (delalloc_bytes && loops < 3) { |
4770 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4771 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4772 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4773 | /* |
4774 | * We need to wait for the async pages to actually start before | |
4775 | * we do anything. | |
4776 | */ | |
0b246afa | 4777 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4778 | if (!max_reclaim) |
4779 | goto skip_async; | |
4780 | ||
4781 | if (max_reclaim <= nr_pages) | |
4782 | max_reclaim = 0; | |
4783 | else | |
4784 | max_reclaim -= nr_pages; | |
dea31f52 | 4785 | |
0b246afa JM |
4786 | wait_event(fs_info->async_submit_wait, |
4787 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4788 | (int)max_reclaim); |
4789 | skip_async: | |
08e007d2 MX |
4790 | if (!trans) |
4791 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4792 | else | |
4793 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4794 | spin_lock(&space_info->lock); |
c1c4919b | 4795 | if (can_overcommit(fs_info, space_info, orig, flush, false)) { |
f4c738c2 JB |
4796 | spin_unlock(&space_info->lock); |
4797 | break; | |
4798 | } | |
957780eb JB |
4799 | if (list_empty(&space_info->tickets) && |
4800 | list_empty(&space_info->priority_tickets)) { | |
4801 | spin_unlock(&space_info->lock); | |
4802 | break; | |
4803 | } | |
0019f10d | 4804 | spin_unlock(&space_info->lock); |
5da9d01b | 4805 | |
36e39c40 | 4806 | loops++; |
f104d044 | 4807 | if (wait_ordered && !trans) { |
0b246afa | 4808 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4809 | } else { |
f4c738c2 | 4810 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4811 | if (time_left) |
4812 | break; | |
4813 | } | |
963d678b | 4814 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4815 | &fs_info->delalloc_bytes); |
5da9d01b | 4816 | } |
5da9d01b YZ |
4817 | } |
4818 | ||
663350ac JB |
4819 | /** |
4820 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4821 | * @root - the root we're allocating for | |
4822 | * @bytes - the number of bytes we want to reserve | |
4823 | * @force - force the commit | |
8bb8ab2e | 4824 | * |
663350ac JB |
4825 | * This will check to make sure that committing the transaction will actually |
4826 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4827 | * will return -ENOSPC. | |
8bb8ab2e | 4828 | */ |
0c9ab349 | 4829 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
663350ac JB |
4830 | struct btrfs_space_info *space_info, |
4831 | u64 bytes, int force) | |
4832 | { | |
0b246afa | 4833 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac JB |
4834 | struct btrfs_trans_handle *trans; |
4835 | ||
4836 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4837 | if (trans) | |
4838 | return -EAGAIN; | |
4839 | ||
4840 | if (force) | |
4841 | goto commit; | |
4842 | ||
4843 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4844 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4845 | bytes) >= 0) |
663350ac | 4846 | goto commit; |
663350ac JB |
4847 | |
4848 | /* | |
4849 | * See if there is some space in the delayed insertion reservation for | |
4850 | * this reservation. | |
4851 | */ | |
4852 | if (space_info != delayed_rsv->space_info) | |
4853 | return -ENOSPC; | |
4854 | ||
4855 | spin_lock(&delayed_rsv->lock); | |
b150a4f1 | 4856 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
28785f70 | 4857 | bytes - delayed_rsv->size) < 0) { |
663350ac JB |
4858 | spin_unlock(&delayed_rsv->lock); |
4859 | return -ENOSPC; | |
4860 | } | |
4861 | spin_unlock(&delayed_rsv->lock); | |
4862 | ||
4863 | commit: | |
a9b3311e | 4864 | trans = btrfs_join_transaction(fs_info->extent_root); |
663350ac JB |
4865 | if (IS_ERR(trans)) |
4866 | return -ENOSPC; | |
4867 | ||
3a45bb20 | 4868 | return btrfs_commit_transaction(trans); |
663350ac JB |
4869 | } |
4870 | ||
957780eb JB |
4871 | struct reserve_ticket { |
4872 | u64 bytes; | |
4873 | int error; | |
4874 | struct list_head list; | |
4875 | wait_queue_head_t wait; | |
96c3f433 JB |
4876 | }; |
4877 | ||
0c9ab349 | 4878 | static int flush_space(struct btrfs_fs_info *fs_info, |
96c3f433 JB |
4879 | struct btrfs_space_info *space_info, u64 num_bytes, |
4880 | u64 orig_bytes, int state) | |
4881 | { | |
a9b3311e | 4882 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4883 | struct btrfs_trans_handle *trans; |
4884 | int nr; | |
f4c738c2 | 4885 | int ret = 0; |
96c3f433 JB |
4886 | |
4887 | switch (state) { | |
96c3f433 JB |
4888 | case FLUSH_DELAYED_ITEMS_NR: |
4889 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4890 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4891 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4892 | else |
96c3f433 | 4893 | nr = -1; |
18cd8ea6 | 4894 | |
96c3f433 JB |
4895 | trans = btrfs_join_transaction(root); |
4896 | if (IS_ERR(trans)) { | |
4897 | ret = PTR_ERR(trans); | |
4898 | break; | |
4899 | } | |
2ff7e61e | 4900 | ret = btrfs_run_delayed_items_nr(trans, fs_info, nr); |
3a45bb20 | 4901 | btrfs_end_transaction(trans); |
96c3f433 | 4902 | break; |
67b0fd63 JB |
4903 | case FLUSH_DELALLOC: |
4904 | case FLUSH_DELALLOC_WAIT: | |
c1c4919b | 4905 | shrink_delalloc(fs_info, num_bytes * 2, orig_bytes, |
67b0fd63 JB |
4906 | state == FLUSH_DELALLOC_WAIT); |
4907 | break; | |
ea658bad JB |
4908 | case ALLOC_CHUNK: |
4909 | trans = btrfs_join_transaction(root); | |
4910 | if (IS_ERR(trans)) { | |
4911 | ret = PTR_ERR(trans); | |
4912 | break; | |
4913 | } | |
2ff7e61e | 4914 | ret = do_chunk_alloc(trans, fs_info, |
1b86826d | 4915 | btrfs_metadata_alloc_profile(fs_info), |
ea658bad | 4916 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4917 | btrfs_end_transaction(trans); |
eecba891 | 4918 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
4919 | ret = 0; |
4920 | break; | |
96c3f433 | 4921 | case COMMIT_TRANS: |
0c9ab349 JM |
4922 | ret = may_commit_transaction(fs_info, space_info, |
4923 | orig_bytes, 0); | |
96c3f433 JB |
4924 | break; |
4925 | default: | |
4926 | ret = -ENOSPC; | |
4927 | break; | |
4928 | } | |
4929 | ||
0b246afa | 4930 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, |
f376df2b | 4931 | orig_bytes, state, ret); |
96c3f433 JB |
4932 | return ret; |
4933 | } | |
21c7e756 MX |
4934 | |
4935 | static inline u64 | |
c1c4919b JM |
4936 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
4937 | struct btrfs_space_info *space_info, | |
4938 | bool system_chunk) | |
21c7e756 | 4939 | { |
957780eb | 4940 | struct reserve_ticket *ticket; |
21c7e756 MX |
4941 | u64 used; |
4942 | u64 expected; | |
957780eb | 4943 | u64 to_reclaim = 0; |
21c7e756 | 4944 | |
957780eb JB |
4945 | list_for_each_entry(ticket, &space_info->tickets, list) |
4946 | to_reclaim += ticket->bytes; | |
4947 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
4948 | to_reclaim += ticket->bytes; | |
4949 | if (to_reclaim) | |
4950 | return to_reclaim; | |
21c7e756 | 4951 | |
e0af2484 | 4952 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
4953 | if (can_overcommit(fs_info, space_info, to_reclaim, |
4954 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
4955 | return 0; |
4956 | ||
21c7e756 MX |
4957 | used = space_info->bytes_used + space_info->bytes_reserved + |
4958 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4959 | space_info->bytes_may_use; | |
c1c4919b JM |
4960 | if (can_overcommit(fs_info, space_info, SZ_1M, |
4961 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
4962 | expected = div_factor_fine(space_info->total_bytes, 95); |
4963 | else | |
4964 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4965 | ||
4966 | if (used > expected) | |
4967 | to_reclaim = used - expected; | |
4968 | else | |
4969 | to_reclaim = 0; | |
4970 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4971 | space_info->bytes_reserved); | |
21c7e756 MX |
4972 | return to_reclaim; |
4973 | } | |
4974 | ||
c1c4919b JM |
4975 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
4976 | struct btrfs_space_info *space_info, | |
4977 | u64 used, bool system_chunk) | |
21c7e756 | 4978 | { |
365c5313 JB |
4979 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4980 | ||
4981 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 4982 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
4983 | return 0; |
4984 | ||
c1c4919b JM |
4985 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4986 | system_chunk)) | |
d38b349c JB |
4987 | return 0; |
4988 | ||
0b246afa JM |
4989 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
4990 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
4991 | } |
4992 | ||
957780eb | 4993 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 4994 | { |
957780eb | 4995 | struct reserve_ticket *ticket; |
25ce459c | 4996 | |
957780eb JB |
4997 | while (!list_empty(head)) { |
4998 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
4999 | list_del_init(&ticket->list); | |
5000 | ticket->error = -ENOSPC; | |
5001 | wake_up(&ticket->wait); | |
21c7e756 | 5002 | } |
21c7e756 MX |
5003 | } |
5004 | ||
957780eb JB |
5005 | /* |
5006 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
5007 | * will loop and continuously try to flush as long as we are making progress. | |
5008 | * We count progress as clearing off tickets each time we have to loop. | |
5009 | */ | |
21c7e756 MX |
5010 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
5011 | { | |
5012 | struct btrfs_fs_info *fs_info; | |
5013 | struct btrfs_space_info *space_info; | |
5014 | u64 to_reclaim; | |
5015 | int flush_state; | |
957780eb | 5016 | int commit_cycles = 0; |
ce129655 | 5017 | u64 last_tickets_id; |
21c7e756 MX |
5018 | |
5019 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
5020 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5021 | ||
957780eb | 5022 | spin_lock(&space_info->lock); |
c1c4919b JM |
5023 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5024 | false); | |
957780eb JB |
5025 | if (!to_reclaim) { |
5026 | space_info->flush = 0; | |
5027 | spin_unlock(&space_info->lock); | |
21c7e756 | 5028 | return; |
957780eb | 5029 | } |
ce129655 | 5030 | last_tickets_id = space_info->tickets_id; |
957780eb | 5031 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5032 | |
5033 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb JB |
5034 | do { |
5035 | struct reserve_ticket *ticket; | |
5036 | int ret; | |
5037 | ||
0c9ab349 JM |
5038 | ret = flush_space(fs_info, space_info, to_reclaim, to_reclaim, |
5039 | flush_state); | |
957780eb JB |
5040 | spin_lock(&space_info->lock); |
5041 | if (list_empty(&space_info->tickets)) { | |
5042 | space_info->flush = 0; | |
5043 | spin_unlock(&space_info->lock); | |
5044 | return; | |
5045 | } | |
c1c4919b JM |
5046 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5047 | space_info, | |
5048 | false); | |
957780eb JB |
5049 | ticket = list_first_entry(&space_info->tickets, |
5050 | struct reserve_ticket, list); | |
ce129655 | 5051 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5052 | flush_state++; |
5053 | } else { | |
ce129655 | 5054 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5055 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5056 | if (commit_cycles) | |
5057 | commit_cycles--; | |
5058 | } | |
5059 | ||
5060 | if (flush_state > COMMIT_TRANS) { | |
5061 | commit_cycles++; | |
5062 | if (commit_cycles > 2) { | |
5063 | wake_all_tickets(&space_info->tickets); | |
5064 | space_info->flush = 0; | |
5065 | } else { | |
5066 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5067 | } | |
5068 | } | |
5069 | spin_unlock(&space_info->lock); | |
5070 | } while (flush_state <= COMMIT_TRANS); | |
5071 | } | |
5072 | ||
5073 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5074 | { | |
5075 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5076 | } | |
5077 | ||
5078 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5079 | struct btrfs_space_info *space_info, | |
5080 | struct reserve_ticket *ticket) | |
5081 | { | |
5082 | u64 to_reclaim; | |
5083 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5084 | ||
5085 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5086 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5087 | false); | |
957780eb JB |
5088 | if (!to_reclaim) { |
5089 | spin_unlock(&space_info->lock); | |
5090 | return; | |
5091 | } | |
5092 | spin_unlock(&space_info->lock); | |
5093 | ||
21c7e756 | 5094 | do { |
0c9ab349 JM |
5095 | flush_space(fs_info, space_info, to_reclaim, to_reclaim, |
5096 | flush_state); | |
21c7e756 | 5097 | flush_state++; |
957780eb JB |
5098 | spin_lock(&space_info->lock); |
5099 | if (ticket->bytes == 0) { | |
5100 | spin_unlock(&space_info->lock); | |
21c7e756 | 5101 | return; |
957780eb JB |
5102 | } |
5103 | spin_unlock(&space_info->lock); | |
5104 | ||
5105 | /* | |
5106 | * Priority flushers can't wait on delalloc without | |
5107 | * deadlocking. | |
5108 | */ | |
5109 | if (flush_state == FLUSH_DELALLOC || | |
5110 | flush_state == FLUSH_DELALLOC_WAIT) | |
5111 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5112 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5113 | } |
5114 | ||
957780eb JB |
5115 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5116 | struct btrfs_space_info *space_info, | |
5117 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5118 | ||
21c7e756 | 5119 | { |
957780eb JB |
5120 | DEFINE_WAIT(wait); |
5121 | int ret = 0; | |
5122 | ||
5123 | spin_lock(&space_info->lock); | |
5124 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5125 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5126 | if (ret) { | |
5127 | ret = -EINTR; | |
5128 | break; | |
5129 | } | |
5130 | spin_unlock(&space_info->lock); | |
5131 | ||
5132 | schedule(); | |
5133 | ||
5134 | finish_wait(&ticket->wait, &wait); | |
5135 | spin_lock(&space_info->lock); | |
5136 | } | |
5137 | if (!ret) | |
5138 | ret = ticket->error; | |
5139 | if (!list_empty(&ticket->list)) | |
5140 | list_del_init(&ticket->list); | |
5141 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5142 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5143 | space_info->bytes_may_use -= num_bytes; | |
5144 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5145 | space_info->flags, num_bytes, 0); | |
5146 | } | |
5147 | spin_unlock(&space_info->lock); | |
5148 | ||
5149 | return ret; | |
21c7e756 MX |
5150 | } |
5151 | ||
4a92b1b8 JB |
5152 | /** |
5153 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5154 | * @root - the root we're allocating for | |
957780eb | 5155 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5156 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5157 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5158 | * |
01327610 | 5159 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5160 | * with the block_rsv. If there is not enough space it will make an attempt to |
5161 | * flush out space to make room. It will do this by flushing delalloc if | |
5162 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5163 | * regain reservations will be made and this will fail if there is not enough | |
5164 | * space already. | |
8bb8ab2e | 5165 | */ |
c1c4919b | 5166 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5167 | struct btrfs_space_info *space_info, |
5168 | u64 orig_bytes, | |
c1c4919b JM |
5169 | enum btrfs_reserve_flush_enum flush, |
5170 | bool system_chunk) | |
9ed74f2d | 5171 | { |
957780eb | 5172 | struct reserve_ticket ticket; |
2bf64758 | 5173 | u64 used; |
8bb8ab2e | 5174 | int ret = 0; |
9ed74f2d | 5175 | |
957780eb | 5176 | ASSERT(orig_bytes); |
8ca17f0f | 5177 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5178 | |
8bb8ab2e | 5179 | spin_lock(&space_info->lock); |
fdb5effd | 5180 | ret = -ENOSPC; |
4136135b | 5181 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5182 | |
8bb8ab2e | 5183 | /* |
957780eb JB |
5184 | * If we have enough space then hooray, make our reservation and carry |
5185 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5186 | * If not things get more complicated. | |
8bb8ab2e | 5187 | */ |
957780eb JB |
5188 | if (used + orig_bytes <= space_info->total_bytes) { |
5189 | space_info->bytes_may_use += orig_bytes; | |
0b246afa JM |
5190 | trace_btrfs_space_reservation(fs_info, "space_info", |
5191 | space_info->flags, orig_bytes, 1); | |
957780eb | 5192 | ret = 0; |
c1c4919b JM |
5193 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5194 | system_chunk)) { | |
44734ed1 | 5195 | space_info->bytes_may_use += orig_bytes; |
0b246afa JM |
5196 | trace_btrfs_space_reservation(fs_info, "space_info", |
5197 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5198 | ret = 0; |
2bf64758 JB |
5199 | } |
5200 | ||
8bb8ab2e | 5201 | /* |
957780eb JB |
5202 | * If we couldn't make a reservation then setup our reservation ticket |
5203 | * and kick the async worker if it's not already running. | |
08e007d2 | 5204 | * |
957780eb JB |
5205 | * If we are a priority flusher then we just need to add our ticket to |
5206 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5207 | */ |
72bcd99d | 5208 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5209 | ticket.bytes = orig_bytes; |
5210 | ticket.error = 0; | |
5211 | init_waitqueue_head(&ticket.wait); | |
5212 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5213 | list_add_tail(&ticket.list, &space_info->tickets); | |
5214 | if (!space_info->flush) { | |
5215 | space_info->flush = 1; | |
0b246afa | 5216 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5217 | space_info->flags, |
5218 | orig_bytes, flush, | |
5219 | "enospc"); | |
957780eb | 5220 | queue_work(system_unbound_wq, |
c1c4919b | 5221 | &fs_info->async_reclaim_work); |
957780eb JB |
5222 | } |
5223 | } else { | |
5224 | list_add_tail(&ticket.list, | |
5225 | &space_info->priority_tickets); | |
5226 | } | |
21c7e756 MX |
5227 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5228 | used += orig_bytes; | |
f6acfd50 JB |
5229 | /* |
5230 | * We will do the space reservation dance during log replay, | |
5231 | * which means we won't have fs_info->fs_root set, so don't do | |
5232 | * the async reclaim as we will panic. | |
5233 | */ | |
0b246afa | 5234 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5235 | need_do_async_reclaim(fs_info, space_info, |
5236 | used, system_chunk) && | |
0b246afa JM |
5237 | !work_busy(&fs_info->async_reclaim_work)) { |
5238 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5239 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5240 | queue_work(system_unbound_wq, |
0b246afa | 5241 | &fs_info->async_reclaim_work); |
f376df2b | 5242 | } |
8bb8ab2e | 5243 | } |
f0486c68 | 5244 | spin_unlock(&space_info->lock); |
08e007d2 | 5245 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5246 | return ret; |
f0486c68 | 5247 | |
957780eb | 5248 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5249 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5250 | orig_bytes); |
08e007d2 | 5251 | |
957780eb | 5252 | ret = 0; |
0b246afa | 5253 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5254 | spin_lock(&space_info->lock); |
5255 | if (ticket.bytes) { | |
5256 | if (ticket.bytes < orig_bytes) { | |
5257 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5258 | space_info->bytes_may_use -= num_bytes; | |
0b246afa JM |
5259 | trace_btrfs_space_reservation(fs_info, "space_info", |
5260 | space_info->flags, | |
5261 | num_bytes, 0); | |
08e007d2 | 5262 | |
957780eb JB |
5263 | } |
5264 | list_del_init(&ticket.list); | |
5265 | ret = -ENOSPC; | |
5266 | } | |
5267 | spin_unlock(&space_info->lock); | |
5268 | ASSERT(list_empty(&ticket.list)); | |
5269 | return ret; | |
5270 | } | |
8bb8ab2e | 5271 | |
957780eb JB |
5272 | /** |
5273 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5274 | * @root - the root we're allocating for | |
5275 | * @block_rsv - the block_rsv we're allocating for | |
5276 | * @orig_bytes - the number of bytes we want | |
5277 | * @flush - whether or not we can flush to make our reservation | |
5278 | * | |
5279 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5280 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5281 | * flush out space to make room. It will do this by flushing delalloc if | |
5282 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5283 | * regain reservations will be made and this will fail if there is not enough | |
5284 | * space already. | |
5285 | */ | |
5286 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5287 | struct btrfs_block_rsv *block_rsv, | |
5288 | u64 orig_bytes, | |
5289 | enum btrfs_reserve_flush_enum flush) | |
5290 | { | |
0b246afa JM |
5291 | struct btrfs_fs_info *fs_info = root->fs_info; |
5292 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5293 | int ret; |
c1c4919b | 5294 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5295 | |
c1c4919b JM |
5296 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5297 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5298 | if (ret == -ENOSPC && |
5299 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5300 | if (block_rsv != global_rsv && |
5301 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5302 | ret = 0; | |
5303 | } | |
cab45e22 | 5304 | if (ret == -ENOSPC) |
0b246afa | 5305 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5306 | block_rsv->space_info->flags, |
5307 | orig_bytes, 1); | |
f0486c68 YZ |
5308 | return ret; |
5309 | } | |
5310 | ||
79787eaa JM |
5311 | static struct btrfs_block_rsv *get_block_rsv( |
5312 | const struct btrfs_trans_handle *trans, | |
5313 | const struct btrfs_root *root) | |
f0486c68 | 5314 | { |
0b246afa | 5315 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5316 | struct btrfs_block_rsv *block_rsv = NULL; |
5317 | ||
e9cf439f | 5318 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5319 | (root == fs_info->csum_root && trans->adding_csums) || |
5320 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5321 | block_rsv = trans->block_rsv; |
5322 | ||
4c13d758 | 5323 | if (!block_rsv) |
f0486c68 YZ |
5324 | block_rsv = root->block_rsv; |
5325 | ||
5326 | if (!block_rsv) | |
0b246afa | 5327 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5328 | |
5329 | return block_rsv; | |
5330 | } | |
5331 | ||
5332 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5333 | u64 num_bytes) | |
5334 | { | |
5335 | int ret = -ENOSPC; | |
5336 | spin_lock(&block_rsv->lock); | |
5337 | if (block_rsv->reserved >= num_bytes) { | |
5338 | block_rsv->reserved -= num_bytes; | |
5339 | if (block_rsv->reserved < block_rsv->size) | |
5340 | block_rsv->full = 0; | |
5341 | ret = 0; | |
5342 | } | |
5343 | spin_unlock(&block_rsv->lock); | |
5344 | return ret; | |
5345 | } | |
5346 | ||
5347 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5348 | u64 num_bytes, int update_size) | |
5349 | { | |
5350 | spin_lock(&block_rsv->lock); | |
5351 | block_rsv->reserved += num_bytes; | |
5352 | if (update_size) | |
5353 | block_rsv->size += num_bytes; | |
5354 | else if (block_rsv->reserved >= block_rsv->size) | |
5355 | block_rsv->full = 1; | |
5356 | spin_unlock(&block_rsv->lock); | |
5357 | } | |
5358 | ||
d52be818 JB |
5359 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5360 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5361 | int min_factor) | |
5362 | { | |
5363 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5364 | u64 min_bytes; | |
5365 | ||
5366 | if (global_rsv->space_info != dest->space_info) | |
5367 | return -ENOSPC; | |
5368 | ||
5369 | spin_lock(&global_rsv->lock); | |
5370 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5371 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5372 | spin_unlock(&global_rsv->lock); | |
5373 | return -ENOSPC; | |
5374 | } | |
5375 | global_rsv->reserved -= num_bytes; | |
5376 | if (global_rsv->reserved < global_rsv->size) | |
5377 | global_rsv->full = 0; | |
5378 | spin_unlock(&global_rsv->lock); | |
5379 | ||
5380 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5381 | return 0; | |
5382 | } | |
5383 | ||
957780eb JB |
5384 | /* |
5385 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5386 | * basically when we're returning space from block_rsv's. | |
5387 | */ | |
5388 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5389 | struct btrfs_space_info *space_info, | |
5390 | u64 num_bytes) | |
5391 | { | |
5392 | struct reserve_ticket *ticket; | |
5393 | struct list_head *head; | |
5394 | u64 used; | |
5395 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5396 | bool check_overcommit = false; | |
5397 | ||
5398 | spin_lock(&space_info->lock); | |
5399 | head = &space_info->priority_tickets; | |
5400 | ||
5401 | /* | |
5402 | * If we are over our limit then we need to check and see if we can | |
5403 | * overcommit, and if we can't then we just need to free up our space | |
5404 | * and not satisfy any requests. | |
5405 | */ | |
5406 | used = space_info->bytes_used + space_info->bytes_reserved + | |
5407 | space_info->bytes_pinned + space_info->bytes_readonly + | |
5408 | space_info->bytes_may_use; | |
5409 | if (used - num_bytes >= space_info->total_bytes) | |
5410 | check_overcommit = true; | |
5411 | again: | |
5412 | while (!list_empty(head) && num_bytes) { | |
5413 | ticket = list_first_entry(head, struct reserve_ticket, | |
5414 | list); | |
5415 | /* | |
5416 | * We use 0 bytes because this space is already reserved, so | |
5417 | * adding the ticket space would be a double count. | |
5418 | */ | |
5419 | if (check_overcommit && | |
c1c4919b | 5420 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5421 | break; |
5422 | if (num_bytes >= ticket->bytes) { | |
5423 | list_del_init(&ticket->list); | |
5424 | num_bytes -= ticket->bytes; | |
5425 | ticket->bytes = 0; | |
ce129655 | 5426 | space_info->tickets_id++; |
957780eb JB |
5427 | wake_up(&ticket->wait); |
5428 | } else { | |
5429 | ticket->bytes -= num_bytes; | |
5430 | num_bytes = 0; | |
5431 | } | |
5432 | } | |
5433 | ||
5434 | if (num_bytes && head == &space_info->priority_tickets) { | |
5435 | head = &space_info->tickets; | |
5436 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5437 | goto again; | |
5438 | } | |
5439 | space_info->bytes_may_use -= num_bytes; | |
5440 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5441 | space_info->flags, num_bytes, 0); | |
5442 | spin_unlock(&space_info->lock); | |
5443 | } | |
5444 | ||
5445 | /* | |
5446 | * This is for newly allocated space that isn't accounted in | |
5447 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5448 | * we use this helper. | |
5449 | */ | |
5450 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5451 | struct btrfs_space_info *space_info, | |
5452 | u64 num_bytes) | |
5453 | { | |
5454 | struct reserve_ticket *ticket; | |
5455 | struct list_head *head = &space_info->priority_tickets; | |
5456 | ||
5457 | again: | |
5458 | while (!list_empty(head) && num_bytes) { | |
5459 | ticket = list_first_entry(head, struct reserve_ticket, | |
5460 | list); | |
5461 | if (num_bytes >= ticket->bytes) { | |
5462 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5463 | space_info->flags, | |
5464 | ticket->bytes, 1); | |
5465 | list_del_init(&ticket->list); | |
5466 | num_bytes -= ticket->bytes; | |
5467 | space_info->bytes_may_use += ticket->bytes; | |
5468 | ticket->bytes = 0; | |
ce129655 | 5469 | space_info->tickets_id++; |
957780eb JB |
5470 | wake_up(&ticket->wait); |
5471 | } else { | |
5472 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5473 | space_info->flags, | |
5474 | num_bytes, 1); | |
5475 | space_info->bytes_may_use += num_bytes; | |
5476 | ticket->bytes -= num_bytes; | |
5477 | num_bytes = 0; | |
5478 | } | |
5479 | } | |
5480 | ||
5481 | if (num_bytes && head == &space_info->priority_tickets) { | |
5482 | head = &space_info->tickets; | |
5483 | goto again; | |
5484 | } | |
5485 | } | |
5486 | ||
8c2a3ca2 JB |
5487 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
5488 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 5489 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
5490 | { |
5491 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
5492 | ||
5493 | spin_lock(&block_rsv->lock); | |
5494 | if (num_bytes == (u64)-1) | |
5495 | num_bytes = block_rsv->size; | |
5496 | block_rsv->size -= num_bytes; | |
5497 | if (block_rsv->reserved >= block_rsv->size) { | |
5498 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5499 | block_rsv->reserved = block_rsv->size; | |
5500 | block_rsv->full = 1; | |
5501 | } else { | |
5502 | num_bytes = 0; | |
5503 | } | |
5504 | spin_unlock(&block_rsv->lock); | |
5505 | ||
5506 | if (num_bytes > 0) { | |
5507 | if (dest) { | |
e9e22899 JB |
5508 | spin_lock(&dest->lock); |
5509 | if (!dest->full) { | |
5510 | u64 bytes_to_add; | |
5511 | ||
5512 | bytes_to_add = dest->size - dest->reserved; | |
5513 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5514 | dest->reserved += bytes_to_add; | |
5515 | if (dest->reserved >= dest->size) | |
5516 | dest->full = 1; | |
5517 | num_bytes -= bytes_to_add; | |
5518 | } | |
5519 | spin_unlock(&dest->lock); | |
5520 | } | |
957780eb JB |
5521 | if (num_bytes) |
5522 | space_info_add_old_bytes(fs_info, space_info, | |
5523 | num_bytes); | |
9ed74f2d | 5524 | } |
f0486c68 | 5525 | } |
4e06bdd6 | 5526 | |
25d609f8 JB |
5527 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5528 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
5529 | int update_size) | |
f0486c68 YZ |
5530 | { |
5531 | int ret; | |
9ed74f2d | 5532 | |
f0486c68 YZ |
5533 | ret = block_rsv_use_bytes(src, num_bytes); |
5534 | if (ret) | |
5535 | return ret; | |
9ed74f2d | 5536 | |
25d609f8 | 5537 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5538 | return 0; |
5539 | } | |
5540 | ||
66d8f3dd | 5541 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5542 | { |
f0486c68 YZ |
5543 | memset(rsv, 0, sizeof(*rsv)); |
5544 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5545 | rsv->type = type; |
f0486c68 YZ |
5546 | } |
5547 | ||
2ff7e61e | 5548 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5549 | unsigned short type) |
f0486c68 YZ |
5550 | { |
5551 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5552 | |
f0486c68 YZ |
5553 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5554 | if (!block_rsv) | |
5555 | return NULL; | |
9ed74f2d | 5556 | |
66d8f3dd | 5557 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
5558 | block_rsv->space_info = __find_space_info(fs_info, |
5559 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
5560 | return block_rsv; |
5561 | } | |
9ed74f2d | 5562 | |
2ff7e61e | 5563 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5564 | struct btrfs_block_rsv *rsv) |
5565 | { | |
2aaa6655 JB |
5566 | if (!rsv) |
5567 | return; | |
2ff7e61e | 5568 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5569 | kfree(rsv); |
9ed74f2d JB |
5570 | } |
5571 | ||
cdfb080e CM |
5572 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5573 | { | |
5574 | kfree(rsv); | |
5575 | } | |
5576 | ||
08e007d2 MX |
5577 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5578 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5579 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5580 | { |
f0486c68 | 5581 | int ret; |
9ed74f2d | 5582 | |
f0486c68 YZ |
5583 | if (num_bytes == 0) |
5584 | return 0; | |
8bb8ab2e | 5585 | |
61b520a9 | 5586 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5587 | if (!ret) { |
5588 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5589 | return 0; | |
5590 | } | |
9ed74f2d | 5591 | |
f0486c68 | 5592 | return ret; |
f0486c68 | 5593 | } |
9ed74f2d | 5594 | |
2ff7e61e | 5595 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5596 | { |
5597 | u64 num_bytes = 0; | |
f0486c68 | 5598 | int ret = -ENOSPC; |
9ed74f2d | 5599 | |
f0486c68 YZ |
5600 | if (!block_rsv) |
5601 | return 0; | |
9ed74f2d | 5602 | |
f0486c68 | 5603 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5604 | num_bytes = div_factor(block_rsv->size, min_factor); |
5605 | if (block_rsv->reserved >= num_bytes) | |
5606 | ret = 0; | |
5607 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5608 | |
36ba022a JB |
5609 | return ret; |
5610 | } | |
5611 | ||
08e007d2 MX |
5612 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5613 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5614 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5615 | { |
5616 | u64 num_bytes = 0; | |
5617 | int ret = -ENOSPC; | |
5618 | ||
5619 | if (!block_rsv) | |
5620 | return 0; | |
5621 | ||
5622 | spin_lock(&block_rsv->lock); | |
5623 | num_bytes = min_reserved; | |
13553e52 | 5624 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5625 | ret = 0; |
13553e52 | 5626 | else |
f0486c68 | 5627 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5628 | spin_unlock(&block_rsv->lock); |
13553e52 | 5629 | |
f0486c68 YZ |
5630 | if (!ret) |
5631 | return 0; | |
5632 | ||
aa38a711 | 5633 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5634 | if (!ret) { |
5635 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5636 | return 0; |
6a63209f | 5637 | } |
9ed74f2d | 5638 | |
13553e52 | 5639 | return ret; |
f0486c68 YZ |
5640 | } |
5641 | ||
2ff7e61e | 5642 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5643 | struct btrfs_block_rsv *block_rsv, |
5644 | u64 num_bytes) | |
5645 | { | |
0b246afa JM |
5646 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5647 | ||
17504584 | 5648 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5649 | block_rsv->space_info != global_rsv->space_info) |
5650 | global_rsv = NULL; | |
0b246afa | 5651 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes); |
6a63209f JB |
5652 | } |
5653 | ||
8929ecfa YZ |
5654 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5655 | { | |
5656 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5657 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5658 | u64 num_bytes; | |
6a63209f | 5659 | |
ae2e4728 JB |
5660 | /* |
5661 | * The global block rsv is based on the size of the extent tree, the | |
5662 | * checksum tree and the root tree. If the fs is empty we want to set | |
5663 | * it to a minimal amount for safety. | |
5664 | */ | |
5665 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5666 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5667 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5668 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5669 | |
8929ecfa | 5670 | spin_lock(&sinfo->lock); |
1f699d38 | 5671 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5672 | |
ee22184b | 5673 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5674 | |
fb4b10e5 | 5675 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5676 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5677 | if (sinfo->total_bytes > num_bytes) { |
5678 | num_bytes = sinfo->total_bytes - num_bytes; | |
5679 | num_bytes = min(num_bytes, | |
5680 | block_rsv->size - block_rsv->reserved); | |
5681 | block_rsv->reserved += num_bytes; | |
5682 | sinfo->bytes_may_use += num_bytes; | |
5683 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5684 | sinfo->flags, num_bytes, | |
5685 | 1); | |
5686 | } | |
5687 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5688 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5689 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5690 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5691 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5692 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5693 | } |
182608c8 | 5694 | |
fb4b10e5 JB |
5695 | if (block_rsv->reserved == block_rsv->size) |
5696 | block_rsv->full = 1; | |
5697 | else | |
5698 | block_rsv->full = 0; | |
5699 | ||
8929ecfa | 5700 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5701 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5702 | } |
5703 | ||
f0486c68 | 5704 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5705 | { |
f0486c68 | 5706 | struct btrfs_space_info *space_info; |
6a63209f | 5707 | |
f0486c68 YZ |
5708 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5709 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5710 | |
f0486c68 | 5711 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5712 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 5713 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
5714 | fs_info->trans_block_rsv.space_info = space_info; |
5715 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5716 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5717 | |
8929ecfa YZ |
5718 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5719 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5720 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5721 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5722 | if (fs_info->quota_root) |
5723 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5724 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5725 | |
8929ecfa | 5726 | update_global_block_rsv(fs_info); |
6a63209f JB |
5727 | } |
5728 | ||
8929ecfa | 5729 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5730 | { |
8c2a3ca2 JB |
5731 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5732 | (u64)-1); | |
8929ecfa YZ |
5733 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
5734 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
5735 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
5736 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5737 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5738 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5739 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5740 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5741 | } |
5742 | ||
a22285a6 | 5743 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
2ff7e61e | 5744 | struct btrfs_fs_info *fs_info) |
6a63209f | 5745 | { |
0e721106 JB |
5746 | if (!trans->block_rsv) |
5747 | return; | |
5748 | ||
a22285a6 YZ |
5749 | if (!trans->bytes_reserved) |
5750 | return; | |
6a63209f | 5751 | |
0b246afa | 5752 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 5753 | trans->transid, trans->bytes_reserved, 0); |
2ff7e61e JM |
5754 | btrfs_block_rsv_release(fs_info, trans->block_rsv, |
5755 | trans->bytes_reserved); | |
a22285a6 YZ |
5756 | trans->bytes_reserved = 0; |
5757 | } | |
6a63209f | 5758 | |
4fbcdf66 FM |
5759 | /* |
5760 | * To be called after all the new block groups attached to the transaction | |
5761 | * handle have been created (btrfs_create_pending_block_groups()). | |
5762 | */ | |
5763 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5764 | { | |
64b63580 | 5765 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5766 | |
5767 | if (!trans->chunk_bytes_reserved) | |
5768 | return; | |
5769 | ||
5770 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5771 | ||
5772 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5773 | trans->chunk_bytes_reserved); | |
5774 | trans->chunk_bytes_reserved = 0; | |
5775 | } | |
5776 | ||
79787eaa | 5777 | /* Can only return 0 or -ENOSPC */ |
d68fc57b | 5778 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
8ed7a2a0 | 5779 | struct btrfs_inode *inode) |
d68fc57b | 5780 | { |
8ed7a2a0 NB |
5781 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5782 | struct btrfs_root *root = inode->root; | |
40acc3ee JB |
5783 | /* |
5784 | * We always use trans->block_rsv here as we will have reserved space | |
5785 | * for our orphan when starting the transaction, using get_block_rsv() | |
5786 | * here will sometimes make us choose the wrong block rsv as we could be | |
5787 | * doing a reloc inode for a non refcounted root. | |
5788 | */ | |
5789 | struct btrfs_block_rsv *src_rsv = trans->block_rsv; | |
d68fc57b YZ |
5790 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; |
5791 | ||
5792 | /* | |
fcb80c2a JB |
5793 | * We need to hold space in order to delete our orphan item once we've |
5794 | * added it, so this takes the reservation so we can release it later | |
5795 | * when we are truly done with the orphan item. | |
d68fc57b | 5796 | */ |
0b246afa JM |
5797 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5798 | ||
8ed7a2a0 NB |
5799 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5800 | num_bytes, 1); | |
25d609f8 | 5801 | return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); |
6a63209f JB |
5802 | } |
5803 | ||
703b391a | 5804 | void btrfs_orphan_release_metadata(struct btrfs_inode *inode) |
97e728d4 | 5805 | { |
703b391a NB |
5806 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5807 | struct btrfs_root *root = inode->root; | |
0b246afa JM |
5808 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5809 | ||
703b391a NB |
5810 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5811 | num_bytes, 0); | |
2ff7e61e | 5812 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes); |
d68fc57b | 5813 | } |
97e728d4 | 5814 | |
d5c12070 MX |
5815 | /* |
5816 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5817 | * root: the root of the parent directory | |
5818 | * rsv: block reservation | |
5819 | * items: the number of items that we need do reservation | |
5820 | * qgroup_reserved: used to return the reserved size in qgroup | |
5821 | * | |
5822 | * This function is used to reserve the space for snapshot/subvolume | |
5823 | * creation and deletion. Those operations are different with the | |
5824 | * common file/directory operations, they change two fs/file trees | |
5825 | * and root tree, the number of items that the qgroup reserves is | |
5826 | * different with the free space reservation. So we can not use | |
01327610 | 5827 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5828 | */ |
5829 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5830 | struct btrfs_block_rsv *rsv, | |
5831 | int items, | |
ee3441b4 JM |
5832 | u64 *qgroup_reserved, |
5833 | bool use_global_rsv) | |
a22285a6 | 5834 | { |
d5c12070 MX |
5835 | u64 num_bytes; |
5836 | int ret; | |
0b246afa JM |
5837 | struct btrfs_fs_info *fs_info = root->fs_info; |
5838 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5839 | |
0b246afa | 5840 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5841 | /* One for parent inode, two for dir entries */ |
0b246afa | 5842 | num_bytes = 3 * fs_info->nodesize; |
003d7c59 | 5843 | ret = btrfs_qgroup_reserve_meta(root, num_bytes, true); |
d5c12070 MX |
5844 | if (ret) |
5845 | return ret; | |
5846 | } else { | |
5847 | num_bytes = 0; | |
5848 | } | |
5849 | ||
5850 | *qgroup_reserved = num_bytes; | |
5851 | ||
0b246afa JM |
5852 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
5853 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
5854 | BTRFS_BLOCK_GROUP_METADATA); |
5855 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5856 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5857 | |
5858 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 5859 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 5860 | |
7174109c QW |
5861 | if (ret && *qgroup_reserved) |
5862 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
d5c12070 MX |
5863 | |
5864 | return ret; | |
5865 | } | |
5866 | ||
2ff7e61e | 5867 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 5868 | struct btrfs_block_rsv *rsv) |
d5c12070 | 5869 | { |
2ff7e61e | 5870 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
5871 | } |
5872 | ||
7709cde3 JB |
5873 | /** |
5874 | * drop_outstanding_extent - drop an outstanding extent | |
5875 | * @inode: the inode we're dropping the extent for | |
01327610 | 5876 | * @num_bytes: the number of bytes we're releasing. |
7709cde3 JB |
5877 | * |
5878 | * This is called when we are freeing up an outstanding extent, either called | |
5879 | * after an error or after an extent is written. This will return the number of | |
5880 | * reserved extents that need to be freed. This must be called with | |
5881 | * BTRFS_I(inode)->lock held. | |
5882 | */ | |
baa3ba39 NB |
5883 | static unsigned drop_outstanding_extent(struct btrfs_inode *inode, |
5884 | u64 num_bytes) | |
9e0baf60 | 5885 | { |
7fd2ae21 | 5886 | unsigned drop_inode_space = 0; |
9e0baf60 | 5887 | unsigned dropped_extents = 0; |
823bb20a | 5888 | unsigned num_extents; |
9e0baf60 | 5889 | |
823bb20a | 5890 | num_extents = count_max_extents(num_bytes); |
dcab6a3b | 5891 | ASSERT(num_extents); |
baa3ba39 NB |
5892 | ASSERT(inode->outstanding_extents >= num_extents); |
5893 | inode->outstanding_extents -= num_extents; | |
9e0baf60 | 5894 | |
baa3ba39 | 5895 | if (inode->outstanding_extents == 0 && |
72ac3c0d | 5896 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
baa3ba39 | 5897 | &inode->runtime_flags)) |
7fd2ae21 | 5898 | drop_inode_space = 1; |
7fd2ae21 | 5899 | |
9e0baf60 | 5900 | /* |
01327610 | 5901 | * If we have more or the same amount of outstanding extents than we have |
9e0baf60 JB |
5902 | * reserved then we need to leave the reserved extents count alone. |
5903 | */ | |
baa3ba39 | 5904 | if (inode->outstanding_extents >= inode->reserved_extents) |
7fd2ae21 | 5905 | return drop_inode_space; |
9e0baf60 | 5906 | |
baa3ba39 NB |
5907 | dropped_extents = inode->reserved_extents - inode->outstanding_extents; |
5908 | inode->reserved_extents -= dropped_extents; | |
7fd2ae21 | 5909 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
5910 | } |
5911 | ||
7709cde3 | 5912 | /** |
01327610 NS |
5913 | * calc_csum_metadata_size - return the amount of metadata space that must be |
5914 | * reserved/freed for the given bytes. | |
7709cde3 JB |
5915 | * @inode: the inode we're manipulating |
5916 | * @num_bytes: the number of bytes in question | |
5917 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
5918 | * | |
5919 | * This adjusts the number of csum_bytes in the inode and then returns the | |
5920 | * correct amount of metadata that must either be reserved or freed. We | |
5921 | * calculate how many checksums we can fit into one leaf and then divide the | |
5922 | * number of bytes that will need to be checksumed by this value to figure out | |
5923 | * how many checksums will be required. If we are adding bytes then the number | |
5924 | * may go up and we will return the number of additional bytes that must be | |
5925 | * reserved. If it is going down we will return the number of bytes that must | |
5926 | * be freed. | |
5927 | * | |
5928 | * This must be called with BTRFS_I(inode)->lock held. | |
5929 | */ | |
0e6bf9b1 | 5930 | static u64 calc_csum_metadata_size(struct btrfs_inode *inode, u64 num_bytes, |
7709cde3 | 5931 | int reserve) |
6324fbf3 | 5932 | { |
0e6bf9b1 | 5933 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
1262133b | 5934 | u64 old_csums, num_csums; |
7709cde3 | 5935 | |
0e6bf9b1 | 5936 | if (inode->flags & BTRFS_INODE_NODATASUM && inode->csum_bytes == 0) |
7709cde3 JB |
5937 | return 0; |
5938 | ||
0e6bf9b1 | 5939 | old_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes); |
7709cde3 | 5940 | if (reserve) |
0e6bf9b1 | 5941 | inode->csum_bytes += num_bytes; |
7709cde3 | 5942 | else |
0e6bf9b1 NB |
5943 | inode->csum_bytes -= num_bytes; |
5944 | num_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes); | |
7709cde3 JB |
5945 | |
5946 | /* No change, no need to reserve more */ | |
5947 | if (old_csums == num_csums) | |
5948 | return 0; | |
5949 | ||
5950 | if (reserve) | |
0b246afa | 5951 | return btrfs_calc_trans_metadata_size(fs_info, |
7709cde3 JB |
5952 | num_csums - old_csums); |
5953 | ||
0b246afa | 5954 | return btrfs_calc_trans_metadata_size(fs_info, old_csums - num_csums); |
0ca1f7ce | 5955 | } |
c146afad | 5956 | |
9f3db423 | 5957 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 5958 | { |
9f3db423 NB |
5959 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5960 | struct btrfs_root *root = inode->root; | |
0b246afa | 5961 | struct btrfs_block_rsv *block_rsv = &fs_info->delalloc_block_rsv; |
9e0baf60 | 5962 | u64 to_reserve = 0; |
660d3f6c | 5963 | u64 csum_bytes; |
823bb20a | 5964 | unsigned nr_extents; |
08e007d2 | 5965 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5966 | int ret = 0; |
c64c2bd8 | 5967 | bool delalloc_lock = true; |
88e081bf WS |
5968 | u64 to_free = 0; |
5969 | unsigned dropped; | |
48c3d480 | 5970 | bool release_extra = false; |
6324fbf3 | 5971 | |
c64c2bd8 JB |
5972 | /* If we are a free space inode we need to not flush since we will be in |
5973 | * the middle of a transaction commit. We also don't need the delalloc | |
5974 | * mutex since we won't race with anybody. We need this mostly to make | |
5975 | * lockdep shut its filthy mouth. | |
bac357dc JB |
5976 | * |
5977 | * If we have a transaction open (can happen if we call truncate_block | |
5978 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
5979 | */ |
5980 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 5981 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 5982 | delalloc_lock = false; |
bac357dc JB |
5983 | } else if (current->journal_info) { |
5984 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c64c2bd8 | 5985 | } |
c09544e0 | 5986 | |
08e007d2 | 5987 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
0b246afa | 5988 | btrfs_transaction_in_commit(fs_info)) |
0ca1f7ce | 5989 | schedule_timeout(1); |
ec44a35c | 5990 | |
c64c2bd8 | 5991 | if (delalloc_lock) |
9f3db423 | 5992 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 5993 | |
0b246afa | 5994 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
8bb8ab2e | 5995 | |
9f3db423 | 5996 | spin_lock(&inode->lock); |
823bb20a | 5997 | nr_extents = count_max_extents(num_bytes); |
9f3db423 | 5998 | inode->outstanding_extents += nr_extents; |
9e0baf60 | 5999 | |
48c3d480 | 6000 | nr_extents = 0; |
9f3db423 NB |
6001 | if (inode->outstanding_extents > inode->reserved_extents) |
6002 | nr_extents += inode->outstanding_extents - | |
6003 | inode->reserved_extents; | |
57a45ced | 6004 | |
48c3d480 | 6005 | /* We always want to reserve a slot for updating the inode. */ |
0b246afa | 6006 | to_reserve = btrfs_calc_trans_metadata_size(fs_info, nr_extents + 1); |
7709cde3 | 6007 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
9f3db423 NB |
6008 | csum_bytes = inode->csum_bytes; |
6009 | spin_unlock(&inode->lock); | |
57a45ced | 6010 | |
0b246afa | 6011 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
7174109c | 6012 | ret = btrfs_qgroup_reserve_meta(root, |
003d7c59 | 6013 | nr_extents * fs_info->nodesize, true); |
88e081bf WS |
6014 | if (ret) |
6015 | goto out_fail; | |
6016 | } | |
c5567237 | 6017 | |
48c3d480 | 6018 | ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush); |
88e081bf | 6019 | if (unlikely(ret)) { |
da17066c | 6020 | btrfs_qgroup_free_meta(root, |
0b246afa | 6021 | nr_extents * fs_info->nodesize); |
88e081bf | 6022 | goto out_fail; |
9e0baf60 | 6023 | } |
25179201 | 6024 | |
9f3db423 | 6025 | spin_lock(&inode->lock); |
48c3d480 | 6026 | if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
9f3db423 | 6027 | &inode->runtime_flags)) { |
0b246afa | 6028 | to_reserve -= btrfs_calc_trans_metadata_size(fs_info, 1); |
48c3d480 | 6029 | release_extra = true; |
660d3f6c | 6030 | } |
9f3db423 NB |
6031 | inode->reserved_extents += nr_extents; |
6032 | spin_unlock(&inode->lock); | |
c64c2bd8 JB |
6033 | |
6034 | if (delalloc_lock) | |
9f3db423 | 6035 | mutex_unlock(&inode->delalloc_mutex); |
660d3f6c | 6036 | |
8c2a3ca2 | 6037 | if (to_reserve) |
0b246afa | 6038 | trace_btrfs_space_reservation(fs_info, "delalloc", |
9f3db423 | 6039 | btrfs_ino(inode), to_reserve, 1); |
48c3d480 | 6040 | if (release_extra) |
2ff7e61e | 6041 | btrfs_block_rsv_release(fs_info, block_rsv, |
0b246afa | 6042 | btrfs_calc_trans_metadata_size(fs_info, 1)); |
0ca1f7ce | 6043 | return 0; |
88e081bf WS |
6044 | |
6045 | out_fail: | |
9f3db423 | 6046 | spin_lock(&inode->lock); |
dcab6a3b | 6047 | dropped = drop_outstanding_extent(inode, num_bytes); |
88e081bf WS |
6048 | /* |
6049 | * If the inodes csum_bytes is the same as the original | |
6050 | * csum_bytes then we know we haven't raced with any free()ers | |
6051 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 6052 | */ |
9f3db423 | 6053 | if (inode->csum_bytes == csum_bytes) { |
88e081bf | 6054 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 | 6055 | } else { |
9f3db423 | 6056 | u64 orig_csum_bytes = inode->csum_bytes; |
f4881bc7 JB |
6057 | u64 bytes; |
6058 | ||
6059 | /* | |
6060 | * This is tricky, but first we need to figure out how much we | |
01327610 | 6061 | * freed from any free-ers that occurred during this |
f4881bc7 JB |
6062 | * reservation, so we reset ->csum_bytes to the csum_bytes |
6063 | * before we dropped our lock, and then call the free for the | |
6064 | * number of bytes that were freed while we were trying our | |
6065 | * reservation. | |
6066 | */ | |
9f3db423 NB |
6067 | bytes = csum_bytes - inode->csum_bytes; |
6068 | inode->csum_bytes = csum_bytes; | |
f4881bc7 JB |
6069 | to_free = calc_csum_metadata_size(inode, bytes, 0); |
6070 | ||
6071 | ||
6072 | /* | |
6073 | * Now we need to see how much we would have freed had we not | |
6074 | * been making this reservation and our ->csum_bytes were not | |
6075 | * artificially inflated. | |
6076 | */ | |
9f3db423 | 6077 | inode->csum_bytes = csum_bytes - num_bytes; |
f4881bc7 JB |
6078 | bytes = csum_bytes - orig_csum_bytes; |
6079 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
6080 | ||
6081 | /* | |
6082 | * Now reset ->csum_bytes to what it should be. If bytes is | |
01327610 | 6083 | * more than to_free then we would have freed more space had we |
f4881bc7 JB |
6084 | * not had an artificially high ->csum_bytes, so we need to free |
6085 | * the remainder. If bytes is the same or less then we don't | |
6086 | * need to do anything, the other free-ers did the correct | |
6087 | * thing. | |
6088 | */ | |
9f3db423 | 6089 | inode->csum_bytes = orig_csum_bytes - num_bytes; |
f4881bc7 JB |
6090 | if (bytes > to_free) |
6091 | to_free = bytes - to_free; | |
6092 | else | |
6093 | to_free = 0; | |
6094 | } | |
9f3db423 | 6095 | spin_unlock(&inode->lock); |
e2d1f923 | 6096 | if (dropped) |
0b246afa | 6097 | to_free += btrfs_calc_trans_metadata_size(fs_info, dropped); |
88e081bf WS |
6098 | |
6099 | if (to_free) { | |
2ff7e61e | 6100 | btrfs_block_rsv_release(fs_info, block_rsv, to_free); |
0b246afa | 6101 | trace_btrfs_space_reservation(fs_info, "delalloc", |
9f3db423 | 6102 | btrfs_ino(inode), to_free, 0); |
88e081bf WS |
6103 | } |
6104 | if (delalloc_lock) | |
9f3db423 | 6105 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6106 | return ret; |
0ca1f7ce YZ |
6107 | } |
6108 | ||
7709cde3 JB |
6109 | /** |
6110 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
6111 | * @inode: the inode to release the reservation for | |
6112 | * @num_bytes: the number of bytes we're releasing | |
6113 | * | |
6114 | * This will release the metadata reservation for an inode. This can be called | |
6115 | * once we complete IO for a given set of bytes to release their metadata | |
6116 | * reservations. | |
6117 | */ | |
691fa059 | 6118 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6119 | { |
691fa059 | 6120 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
9e0baf60 JB |
6121 | u64 to_free = 0; |
6122 | unsigned dropped; | |
0ca1f7ce | 6123 | |
0b246afa | 6124 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6125 | spin_lock(&inode->lock); |
dcab6a3b | 6126 | dropped = drop_outstanding_extent(inode, num_bytes); |
97e728d4 | 6127 | |
0934856d MX |
6128 | if (num_bytes) |
6129 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
691fa059 | 6130 | spin_unlock(&inode->lock); |
9e0baf60 | 6131 | if (dropped > 0) |
0b246afa | 6132 | to_free += btrfs_calc_trans_metadata_size(fs_info, dropped); |
0ca1f7ce | 6133 | |
0b246afa | 6134 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6135 | return; |
6136 | ||
691fa059 NB |
6137 | trace_btrfs_space_reservation(fs_info, "delalloc", btrfs_ino(inode), |
6138 | to_free, 0); | |
c5567237 | 6139 | |
2ff7e61e | 6140 | btrfs_block_rsv_release(fs_info, &fs_info->delalloc_block_rsv, to_free); |
0ca1f7ce YZ |
6141 | } |
6142 | ||
1ada3a62 | 6143 | /** |
7cf5b976 | 6144 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6145 | * delalloc |
6146 | * @inode: inode we're writing to | |
6147 | * @start: start range we are writing to | |
6148 | * @len: how long the range we are writing to | |
6149 | * | |
1ada3a62 QW |
6150 | * This will do the following things |
6151 | * | |
6152 | * o reserve space in data space info for num bytes | |
6153 | * and reserve precious corresponding qgroup space | |
6154 | * (Done in check_data_free_space) | |
6155 | * | |
6156 | * o reserve space for metadata space, based on the number of outstanding | |
6157 | * extents and how much csums will be needed | |
6158 | * also reserve metadata space in a per root over-reserve method. | |
6159 | * o add to the inodes->delalloc_bytes | |
6160 | * o add it to the fs_info's delalloc inodes list. | |
6161 | * (Above 3 all done in delalloc_reserve_metadata) | |
6162 | * | |
6163 | * Return 0 for success | |
6164 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6165 | */ | |
7cf5b976 | 6166 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
6167 | { |
6168 | int ret; | |
6169 | ||
7cf5b976 | 6170 | ret = btrfs_check_data_free_space(inode, start, len); |
1ada3a62 QW |
6171 | if (ret < 0) |
6172 | return ret; | |
9f3db423 | 6173 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6174 | if (ret < 0) |
7cf5b976 | 6175 | btrfs_free_reserved_data_space(inode, start, len); |
1ada3a62 QW |
6176 | return ret; |
6177 | } | |
6178 | ||
7709cde3 | 6179 | /** |
7cf5b976 | 6180 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6181 | * @inode: inode we're releasing space for |
6182 | * @start: start position of the space already reserved | |
6183 | * @len: the len of the space already reserved | |
6184 | * | |
6185 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
6186 | * called in the case that we don't need the metadata AND data reservations | |
6187 | * anymore. So if there is an error or we insert an inline extent. | |
6188 | * | |
6189 | * This function will release the metadata space that was not used and will | |
6190 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6191 | * list if there are no delalloc bytes left. | |
6192 | * Also it will handle the qgroup reserved space. | |
6193 | */ | |
7cf5b976 | 6194 | void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 | 6195 | { |
691fa059 | 6196 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len); |
7cf5b976 | 6197 | btrfs_free_reserved_data_space(inode, start, len); |
6324fbf3 CM |
6198 | } |
6199 | ||
ce93ec54 | 6200 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6201 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6202 | u64 num_bytes, int alloc) |
9078a3e1 | 6203 | { |
0af3d00b | 6204 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6205 | u64 total = num_bytes; |
9078a3e1 | 6206 | u64 old_val; |
db94535d | 6207 | u64 byte_in_group; |
0af3d00b | 6208 | int factor; |
3e1ad54f | 6209 | |
5d4f98a2 | 6210 | /* block accounting for super block */ |
eb73c1b7 | 6211 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6212 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6213 | if (alloc) |
6214 | old_val += num_bytes; | |
6215 | else | |
6216 | old_val -= num_bytes; | |
6c41761f | 6217 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6218 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6219 | |
d397712b | 6220 | while (total) { |
db94535d | 6221 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6222 | if (!cache) |
79787eaa | 6223 | return -ENOENT; |
b742bb82 YZ |
6224 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6225 | BTRFS_BLOCK_GROUP_RAID1 | | |
6226 | BTRFS_BLOCK_GROUP_RAID10)) | |
6227 | factor = 2; | |
6228 | else | |
6229 | factor = 1; | |
9d66e233 JB |
6230 | /* |
6231 | * If this block group has free space cache written out, we | |
6232 | * need to make sure to load it if we are removing space. This | |
6233 | * is because we need the unpinning stage to actually add the | |
6234 | * space back to the block group, otherwise we will leak space. | |
6235 | */ | |
6236 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6237 | cache_block_group(cache, 1); |
0af3d00b | 6238 | |
db94535d CM |
6239 | byte_in_group = bytenr - cache->key.objectid; |
6240 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6241 | |
25179201 | 6242 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6243 | spin_lock(&cache->lock); |
0af3d00b | 6244 | |
6202df69 | 6245 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6246 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6247 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6248 | ||
9078a3e1 | 6249 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6250 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6251 | if (alloc) { |
db94535d | 6252 | old_val += num_bytes; |
11833d66 YZ |
6253 | btrfs_set_block_group_used(&cache->item, old_val); |
6254 | cache->reserved -= num_bytes; | |
11833d66 | 6255 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6256 | cache->space_info->bytes_used += num_bytes; |
6257 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6258 | spin_unlock(&cache->lock); |
25179201 | 6259 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6260 | } else { |
db94535d | 6261 | old_val -= num_bytes; |
ae0ab003 FM |
6262 | btrfs_set_block_group_used(&cache->item, old_val); |
6263 | cache->pinned += num_bytes; | |
6264 | cache->space_info->bytes_pinned += num_bytes; | |
6265 | cache->space_info->bytes_used -= num_bytes; | |
6266 | cache->space_info->disk_used -= num_bytes * factor; | |
6267 | spin_unlock(&cache->lock); | |
6268 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6269 | |
0b246afa | 6270 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6271 | cache->space_info->flags, |
6272 | num_bytes, 1); | |
ae0ab003 FM |
6273 | set_extent_dirty(info->pinned_extents, |
6274 | bytenr, bytenr + num_bytes - 1, | |
6275 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6276 | } |
1bbc621e CM |
6277 | |
6278 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6279 | if (list_empty(&cache->dirty_list)) { | |
6280 | list_add_tail(&cache->dirty_list, | |
6281 | &trans->transaction->dirty_bgs); | |
6282 | trans->transaction->num_dirty_bgs++; | |
6283 | btrfs_get_block_group(cache); | |
6284 | } | |
6285 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6286 | ||
036a9348 FM |
6287 | /* |
6288 | * No longer have used bytes in this block group, queue it for | |
6289 | * deletion. We do this after adding the block group to the | |
6290 | * dirty list to avoid races between cleaner kthread and space | |
6291 | * cache writeout. | |
6292 | */ | |
6293 | if (!alloc && old_val == 0) { | |
6294 | spin_lock(&info->unused_bgs_lock); | |
6295 | if (list_empty(&cache->bg_list)) { | |
6296 | btrfs_get_block_group(cache); | |
6297 | list_add_tail(&cache->bg_list, | |
6298 | &info->unused_bgs); | |
6299 | } | |
6300 | spin_unlock(&info->unused_bgs_lock); | |
6301 | } | |
6302 | ||
fa9c0d79 | 6303 | btrfs_put_block_group(cache); |
db94535d CM |
6304 | total -= num_bytes; |
6305 | bytenr += num_bytes; | |
9078a3e1 CM |
6306 | } |
6307 | return 0; | |
6308 | } | |
6324fbf3 | 6309 | |
2ff7e61e | 6310 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6311 | { |
0f9dd46c | 6312 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6313 | u64 bytenr; |
0f9dd46c | 6314 | |
0b246afa JM |
6315 | spin_lock(&fs_info->block_group_cache_lock); |
6316 | bytenr = fs_info->first_logical_byte; | |
6317 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6318 | |
6319 | if (bytenr < (u64)-1) | |
6320 | return bytenr; | |
6321 | ||
0b246afa | 6322 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6323 | if (!cache) |
a061fc8d | 6324 | return 0; |
0f9dd46c | 6325 | |
d2fb3437 | 6326 | bytenr = cache->key.objectid; |
fa9c0d79 | 6327 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6328 | |
6329 | return bytenr; | |
a061fc8d CM |
6330 | } |
6331 | ||
2ff7e61e | 6332 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6333 | struct btrfs_block_group_cache *cache, |
6334 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6335 | { |
11833d66 YZ |
6336 | spin_lock(&cache->space_info->lock); |
6337 | spin_lock(&cache->lock); | |
6338 | cache->pinned += num_bytes; | |
6339 | cache->space_info->bytes_pinned += num_bytes; | |
6340 | if (reserved) { | |
6341 | cache->reserved -= num_bytes; | |
6342 | cache->space_info->bytes_reserved -= num_bytes; | |
6343 | } | |
6344 | spin_unlock(&cache->lock); | |
6345 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6346 | |
0b246afa | 6347 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6348 | cache->space_info->flags, num_bytes, 1); |
0b246afa | 6349 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6350 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6351 | return 0; | |
6352 | } | |
68b38550 | 6353 | |
f0486c68 YZ |
6354 | /* |
6355 | * this function must be called within transaction | |
6356 | */ | |
2ff7e61e | 6357 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6358 | u64 bytenr, u64 num_bytes, int reserved) |
6359 | { | |
6360 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6361 | |
0b246afa | 6362 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6363 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6364 | |
2ff7e61e | 6365 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6366 | |
6367 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6368 | return 0; |
6369 | } | |
6370 | ||
f0486c68 | 6371 | /* |
e688b725 CM |
6372 | * this function must be called within transaction |
6373 | */ | |
2ff7e61e | 6374 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6375 | u64 bytenr, u64 num_bytes) |
6376 | { | |
6377 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6378 | int ret; |
e688b725 | 6379 | |
0b246afa | 6380 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6381 | if (!cache) |
6382 | return -EINVAL; | |
e688b725 CM |
6383 | |
6384 | /* | |
6385 | * pull in the free space cache (if any) so that our pin | |
6386 | * removes the free space from the cache. We have load_only set | |
6387 | * to one because the slow code to read in the free extents does check | |
6388 | * the pinned extents. | |
6389 | */ | |
f6373bf3 | 6390 | cache_block_group(cache, 1); |
e688b725 | 6391 | |
2ff7e61e | 6392 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6393 | |
6394 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6395 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6396 | btrfs_put_block_group(cache); |
b50c6e25 | 6397 | return ret; |
e688b725 CM |
6398 | } |
6399 | ||
2ff7e61e JM |
6400 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6401 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6402 | { |
6403 | int ret; | |
6404 | struct btrfs_block_group_cache *block_group; | |
6405 | struct btrfs_caching_control *caching_ctl; | |
6406 | ||
0b246afa | 6407 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6408 | if (!block_group) |
6409 | return -EINVAL; | |
6410 | ||
6411 | cache_block_group(block_group, 0); | |
6412 | caching_ctl = get_caching_control(block_group); | |
6413 | ||
6414 | if (!caching_ctl) { | |
6415 | /* Logic error */ | |
6416 | BUG_ON(!block_group_cache_done(block_group)); | |
6417 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6418 | } else { | |
6419 | mutex_lock(&caching_ctl->mutex); | |
6420 | ||
6421 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6422 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6423 | } else if (start + num_bytes <= caching_ctl->progress) { |
6424 | ret = btrfs_remove_free_space(block_group, | |
6425 | start, num_bytes); | |
6426 | } else { | |
6427 | num_bytes = caching_ctl->progress - start; | |
6428 | ret = btrfs_remove_free_space(block_group, | |
6429 | start, num_bytes); | |
6430 | if (ret) | |
6431 | goto out_lock; | |
6432 | ||
6433 | num_bytes = (start + num_bytes) - | |
6434 | caching_ctl->progress; | |
6435 | start = caching_ctl->progress; | |
2ff7e61e | 6436 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6437 | } |
6438 | out_lock: | |
6439 | mutex_unlock(&caching_ctl->mutex); | |
6440 | put_caching_control(caching_ctl); | |
6441 | } | |
6442 | btrfs_put_block_group(block_group); | |
6443 | return ret; | |
6444 | } | |
6445 | ||
2ff7e61e | 6446 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6447 | struct extent_buffer *eb) |
6448 | { | |
6449 | struct btrfs_file_extent_item *item; | |
6450 | struct btrfs_key key; | |
6451 | int found_type; | |
6452 | int i; | |
6453 | ||
2ff7e61e | 6454 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6455 | return 0; |
6456 | ||
6457 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6458 | btrfs_item_key_to_cpu(eb, &key, i); | |
6459 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6460 | continue; | |
6461 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6462 | found_type = btrfs_file_extent_type(eb, item); | |
6463 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6464 | continue; | |
6465 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6466 | continue; | |
6467 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6468 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
2ff7e61e | 6469 | __exclude_logged_extent(fs_info, key.objectid, key.offset); |
8c2a1a30 JB |
6470 | } |
6471 | ||
6472 | return 0; | |
6473 | } | |
6474 | ||
9cfa3e34 FM |
6475 | static void |
6476 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6477 | { | |
6478 | atomic_inc(&bg->reservations); | |
6479 | } | |
6480 | ||
6481 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6482 | const u64 start) | |
6483 | { | |
6484 | struct btrfs_block_group_cache *bg; | |
6485 | ||
6486 | bg = btrfs_lookup_block_group(fs_info, start); | |
6487 | ASSERT(bg); | |
6488 | if (atomic_dec_and_test(&bg->reservations)) | |
6489 | wake_up_atomic_t(&bg->reservations); | |
6490 | btrfs_put_block_group(bg); | |
6491 | } | |
6492 | ||
6493 | static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a) | |
6494 | { | |
6495 | schedule(); | |
6496 | return 0; | |
6497 | } | |
6498 | ||
6499 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6500 | { | |
6501 | struct btrfs_space_info *space_info = bg->space_info; | |
6502 | ||
6503 | ASSERT(bg->ro); | |
6504 | ||
6505 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6506 | return; | |
6507 | ||
6508 | /* | |
6509 | * Our block group is read only but before we set it to read only, | |
6510 | * some task might have had allocated an extent from it already, but it | |
6511 | * has not yet created a respective ordered extent (and added it to a | |
6512 | * root's list of ordered extents). | |
6513 | * Therefore wait for any task currently allocating extents, since the | |
6514 | * block group's reservations counter is incremented while a read lock | |
6515 | * on the groups' semaphore is held and decremented after releasing | |
6516 | * the read access on that semaphore and creating the ordered extent. | |
6517 | */ | |
6518 | down_write(&space_info->groups_sem); | |
6519 | up_write(&space_info->groups_sem); | |
6520 | ||
6521 | wait_on_atomic_t(&bg->reservations, | |
6522 | btrfs_wait_bg_reservations_atomic_t, | |
6523 | TASK_UNINTERRUPTIBLE); | |
6524 | } | |
6525 | ||
fb25e914 | 6526 | /** |
4824f1f4 | 6527 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6528 | * @cache: The cache we are manipulating |
18513091 WX |
6529 | * @ram_bytes: The number of bytes of file content, and will be same to |
6530 | * @num_bytes except for the compress path. | |
fb25e914 | 6531 | * @num_bytes: The number of bytes in question |
e570fd27 | 6532 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6533 | * |
745699ef XW |
6534 | * This is called by the allocator when it reserves space. If this is a |
6535 | * reservation and the block group has become read only we cannot make the | |
6536 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6537 | */ |
4824f1f4 | 6538 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6539 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6540 | { |
fb25e914 | 6541 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6542 | int ret = 0; |
79787eaa | 6543 | |
fb25e914 JB |
6544 | spin_lock(&space_info->lock); |
6545 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6546 | if (cache->ro) { |
6547 | ret = -EAGAIN; | |
fb25e914 | 6548 | } else { |
4824f1f4 WX |
6549 | cache->reserved += num_bytes; |
6550 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6551 | |
18513091 WX |
6552 | trace_btrfs_space_reservation(cache->fs_info, |
6553 | "space_info", space_info->flags, | |
6554 | ram_bytes, 0); | |
6555 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6556 | if (delalloc) |
4824f1f4 | 6557 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6558 | } |
fb25e914 JB |
6559 | spin_unlock(&cache->lock); |
6560 | spin_unlock(&space_info->lock); | |
f0486c68 | 6561 | return ret; |
324ae4df | 6562 | } |
9078a3e1 | 6563 | |
4824f1f4 WX |
6564 | /** |
6565 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6566 | * @cache: The cache we are manipulating | |
6567 | * @num_bytes: The number of bytes in question | |
6568 | * @delalloc: The blocks are allocated for the delalloc write | |
6569 | * | |
6570 | * This is called by somebody who is freeing space that was never actually used | |
6571 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6572 | * A and before transaction A commits you free that leaf, you call this with | |
6573 | * reserve set to 0 in order to clear the reservation. | |
6574 | */ | |
6575 | ||
6576 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6577 | u64 num_bytes, int delalloc) | |
6578 | { | |
6579 | struct btrfs_space_info *space_info = cache->space_info; | |
6580 | int ret = 0; | |
6581 | ||
6582 | spin_lock(&space_info->lock); | |
6583 | spin_lock(&cache->lock); | |
6584 | if (cache->ro) | |
6585 | space_info->bytes_readonly += num_bytes; | |
6586 | cache->reserved -= num_bytes; | |
6587 | space_info->bytes_reserved -= num_bytes; | |
6588 | ||
6589 | if (delalloc) | |
6590 | cache->delalloc_bytes -= num_bytes; | |
6591 | spin_unlock(&cache->lock); | |
6592 | spin_unlock(&space_info->lock); | |
6593 | return ret; | |
6594 | } | |
8b74c03e | 6595 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6596 | { |
11833d66 YZ |
6597 | struct btrfs_caching_control *next; |
6598 | struct btrfs_caching_control *caching_ctl; | |
6599 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6600 | |
9e351cc8 | 6601 | down_write(&fs_info->commit_root_sem); |
25179201 | 6602 | |
11833d66 YZ |
6603 | list_for_each_entry_safe(caching_ctl, next, |
6604 | &fs_info->caching_block_groups, list) { | |
6605 | cache = caching_ctl->block_group; | |
6606 | if (block_group_cache_done(cache)) { | |
6607 | cache->last_byte_to_unpin = (u64)-1; | |
6608 | list_del_init(&caching_ctl->list); | |
6609 | put_caching_control(caching_ctl); | |
e8569813 | 6610 | } else { |
11833d66 | 6611 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6612 | } |
e8569813 | 6613 | } |
11833d66 YZ |
6614 | |
6615 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6616 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6617 | else | |
6618 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6619 | ||
9e351cc8 | 6620 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6621 | |
6622 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6623 | } |
6624 | ||
c759c4e1 JB |
6625 | /* |
6626 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6627 | * what it should be based on the mount options. | |
6628 | */ | |
6629 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6630 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6631 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6632 | { |
6633 | struct btrfs_free_cluster *ret = NULL; | |
0b246afa | 6634 | bool ssd = btrfs_test_opt(fs_info, SSD); |
c759c4e1 JB |
6635 | |
6636 | *empty_cluster = 0; | |
6637 | if (btrfs_mixed_space_info(space_info)) | |
6638 | return ret; | |
6639 | ||
6640 | if (ssd) | |
ee22184b | 6641 | *empty_cluster = SZ_2M; |
c759c4e1 | 6642 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6643 | ret = &fs_info->meta_alloc_cluster; |
c759c4e1 | 6644 | if (!ssd) |
ee22184b | 6645 | *empty_cluster = SZ_64K; |
c759c4e1 | 6646 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) { |
0b246afa | 6647 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6648 | } |
6649 | ||
6650 | return ret; | |
6651 | } | |
6652 | ||
2ff7e61e JM |
6653 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6654 | u64 start, u64 end, | |
678886bd | 6655 | const bool return_free_space) |
ccd467d6 | 6656 | { |
11833d66 | 6657 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6658 | struct btrfs_space_info *space_info; |
6659 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6660 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6661 | u64 len; |
c759c4e1 JB |
6662 | u64 total_unpinned = 0; |
6663 | u64 empty_cluster = 0; | |
7b398f8e | 6664 | bool readonly; |
ccd467d6 | 6665 | |
11833d66 | 6666 | while (start <= end) { |
7b398f8e | 6667 | readonly = false; |
11833d66 YZ |
6668 | if (!cache || |
6669 | start >= cache->key.objectid + cache->key.offset) { | |
6670 | if (cache) | |
6671 | btrfs_put_block_group(cache); | |
c759c4e1 | 6672 | total_unpinned = 0; |
11833d66 | 6673 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6674 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6675 | |
2ff7e61e | 6676 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6677 | cache->space_info, |
6678 | &empty_cluster); | |
6679 | empty_cluster <<= 1; | |
11833d66 YZ |
6680 | } |
6681 | ||
6682 | len = cache->key.objectid + cache->key.offset - start; | |
6683 | len = min(len, end + 1 - start); | |
6684 | ||
6685 | if (start < cache->last_byte_to_unpin) { | |
6686 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6687 | if (return_free_space) |
6688 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6689 | } |
6690 | ||
f0486c68 | 6691 | start += len; |
c759c4e1 | 6692 | total_unpinned += len; |
7b398f8e | 6693 | space_info = cache->space_info; |
f0486c68 | 6694 | |
c759c4e1 JB |
6695 | /* |
6696 | * If this space cluster has been marked as fragmented and we've | |
6697 | * unpinned enough in this block group to potentially allow a | |
6698 | * cluster to be created inside of it go ahead and clear the | |
6699 | * fragmented check. | |
6700 | */ | |
6701 | if (cluster && cluster->fragmented && | |
6702 | total_unpinned > empty_cluster) { | |
6703 | spin_lock(&cluster->lock); | |
6704 | cluster->fragmented = 0; | |
6705 | spin_unlock(&cluster->lock); | |
6706 | } | |
6707 | ||
7b398f8e | 6708 | spin_lock(&space_info->lock); |
11833d66 YZ |
6709 | spin_lock(&cache->lock); |
6710 | cache->pinned -= len; | |
7b398f8e | 6711 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6712 | |
6713 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6714 | space_info->flags, len, 0); | |
4f4db217 | 6715 | space_info->max_extent_size = 0; |
d288db5d | 6716 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6717 | if (cache->ro) { |
6718 | space_info->bytes_readonly += len; | |
6719 | readonly = true; | |
6720 | } | |
11833d66 | 6721 | spin_unlock(&cache->lock); |
957780eb JB |
6722 | if (!readonly && return_free_space && |
6723 | global_rsv->space_info == space_info) { | |
6724 | u64 to_add = len; | |
6725 | WARN_ON(!return_free_space); | |
7b398f8e JB |
6726 | spin_lock(&global_rsv->lock); |
6727 | if (!global_rsv->full) { | |
957780eb JB |
6728 | to_add = min(len, global_rsv->size - |
6729 | global_rsv->reserved); | |
6730 | global_rsv->reserved += to_add; | |
6731 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6732 | if (global_rsv->reserved >= global_rsv->size) |
6733 | global_rsv->full = 1; | |
957780eb JB |
6734 | trace_btrfs_space_reservation(fs_info, |
6735 | "space_info", | |
6736 | space_info->flags, | |
6737 | to_add, 1); | |
6738 | len -= to_add; | |
7b398f8e JB |
6739 | } |
6740 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6741 | /* Add to any tickets we may have */ |
6742 | if (len) | |
6743 | space_info_add_new_bytes(fs_info, space_info, | |
6744 | len); | |
7b398f8e JB |
6745 | } |
6746 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6747 | } |
11833d66 YZ |
6748 | |
6749 | if (cache) | |
6750 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6751 | return 0; |
6752 | } | |
6753 | ||
6754 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
2ff7e61e | 6755 | struct btrfs_fs_info *fs_info) |
a28ec197 | 6756 | { |
e33e17ee JM |
6757 | struct btrfs_block_group_cache *block_group, *tmp; |
6758 | struct list_head *deleted_bgs; | |
11833d66 | 6759 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6760 | u64 start; |
6761 | u64 end; | |
a28ec197 | 6762 | int ret; |
a28ec197 | 6763 | |
11833d66 YZ |
6764 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6765 | unpin = &fs_info->freed_extents[1]; | |
6766 | else | |
6767 | unpin = &fs_info->freed_extents[0]; | |
6768 | ||
e33e17ee | 6769 | while (!trans->aborted) { |
d4b450cd | 6770 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6771 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6772 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6773 | if (ret) { |
6774 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6775 | break; |
d4b450cd | 6776 | } |
1f3c79a2 | 6777 | |
0b246afa | 6778 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6779 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6780 | end + 1 - start, NULL); |
1f3c79a2 | 6781 | |
af6f8f60 | 6782 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6783 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6784 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6785 | cond_resched(); |
a28ec197 | 6786 | } |
817d52f8 | 6787 | |
e33e17ee JM |
6788 | /* |
6789 | * Transaction is finished. We don't need the lock anymore. We | |
6790 | * do need to clean up the block groups in case of a transaction | |
6791 | * abort. | |
6792 | */ | |
6793 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6794 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6795 | u64 trimmed = 0; | |
6796 | ||
6797 | ret = -EROFS; | |
6798 | if (!trans->aborted) | |
2ff7e61e | 6799 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6800 | block_group->key.objectid, |
6801 | block_group->key.offset, | |
6802 | &trimmed); | |
6803 | ||
6804 | list_del_init(&block_group->bg_list); | |
6805 | btrfs_put_block_group_trimming(block_group); | |
6806 | btrfs_put_block_group(block_group); | |
6807 | ||
6808 | if (ret) { | |
6809 | const char *errstr = btrfs_decode_error(ret); | |
6810 | btrfs_warn(fs_info, | |
6811 | "Discard failed while removing blockgroup: errno=%d %s\n", | |
6812 | ret, errstr); | |
6813 | } | |
6814 | } | |
6815 | ||
e20d96d6 CM |
6816 | return 0; |
6817 | } | |
6818 | ||
b150a4f1 JB |
6819 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, |
6820 | u64 owner, u64 root_objectid) | |
6821 | { | |
6822 | struct btrfs_space_info *space_info; | |
6823 | u64 flags; | |
6824 | ||
6825 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
6826 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
6827 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
6828 | else | |
6829 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
6830 | } else { | |
6831 | flags = BTRFS_BLOCK_GROUP_DATA; | |
6832 | } | |
6833 | ||
6834 | space_info = __find_space_info(fs_info, flags); | |
6835 | BUG_ON(!space_info); /* Logic bug */ | |
6836 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); | |
6837 | } | |
6838 | ||
6839 | ||
5d4f98a2 | 6840 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 6841 | struct btrfs_fs_info *info, |
c682f9b3 | 6842 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6843 | u64 root_objectid, u64 owner_objectid, |
6844 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6845 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6846 | { |
e2fa7227 | 6847 | struct btrfs_key key; |
5d4f98a2 | 6848 | struct btrfs_path *path; |
1261ec42 | 6849 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6850 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6851 | struct btrfs_extent_item *ei; |
6852 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6853 | int ret; |
5d4f98a2 | 6854 | int is_data; |
952fccac CM |
6855 | int extent_slot = 0; |
6856 | int found_extent = 0; | |
6857 | int num_to_del = 1; | |
5d4f98a2 YZ |
6858 | u32 item_size; |
6859 | u64 refs; | |
c682f9b3 QW |
6860 | u64 bytenr = node->bytenr; |
6861 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6862 | int last_ref = 0; |
0b246afa | 6863 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6864 | |
5caf2a00 | 6865 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6866 | if (!path) |
6867 | return -ENOMEM; | |
5f26f772 | 6868 | |
e4058b54 | 6869 | path->reada = READA_FORWARD; |
b9473439 | 6870 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6871 | |
6872 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6873 | BUG_ON(!is_data && refs_to_drop != 1); | |
6874 | ||
3173a18f JB |
6875 | if (is_data) |
6876 | skinny_metadata = 0; | |
6877 | ||
87bde3cd | 6878 | ret = lookup_extent_backref(trans, info, path, &iref, |
5d4f98a2 YZ |
6879 | bytenr, num_bytes, parent, |
6880 | root_objectid, owner_objectid, | |
6881 | owner_offset); | |
7bb86316 | 6882 | if (ret == 0) { |
952fccac | 6883 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6884 | while (extent_slot >= 0) { |
6885 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6886 | extent_slot); |
5d4f98a2 | 6887 | if (key.objectid != bytenr) |
952fccac | 6888 | break; |
5d4f98a2 YZ |
6889 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6890 | key.offset == num_bytes) { | |
952fccac CM |
6891 | found_extent = 1; |
6892 | break; | |
6893 | } | |
3173a18f JB |
6894 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6895 | key.offset == owner_objectid) { | |
6896 | found_extent = 1; | |
6897 | break; | |
6898 | } | |
952fccac CM |
6899 | if (path->slots[0] - extent_slot > 5) |
6900 | break; | |
5d4f98a2 | 6901 | extent_slot--; |
952fccac | 6902 | } |
5d4f98a2 YZ |
6903 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6904 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6905 | if (found_extent && item_size < sizeof(*ei)) | |
6906 | found_extent = 0; | |
6907 | #endif | |
31840ae1 | 6908 | if (!found_extent) { |
5d4f98a2 | 6909 | BUG_ON(iref); |
87bde3cd JM |
6910 | ret = remove_extent_backref(trans, info, path, NULL, |
6911 | refs_to_drop, | |
fcebe456 | 6912 | is_data, &last_ref); |
005d6427 | 6913 | if (ret) { |
66642832 | 6914 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6915 | goto out; |
6916 | } | |
b3b4aa74 | 6917 | btrfs_release_path(path); |
b9473439 | 6918 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6919 | |
6920 | key.objectid = bytenr; | |
6921 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6922 | key.offset = num_bytes; | |
6923 | ||
3173a18f JB |
6924 | if (!is_data && skinny_metadata) { |
6925 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6926 | key.offset = owner_objectid; | |
6927 | } | |
6928 | ||
31840ae1 ZY |
6929 | ret = btrfs_search_slot(trans, extent_root, |
6930 | &key, path, -1, 1); | |
3173a18f JB |
6931 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6932 | /* | |
6933 | * Couldn't find our skinny metadata item, | |
6934 | * see if we have ye olde extent item. | |
6935 | */ | |
6936 | path->slots[0]--; | |
6937 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6938 | path->slots[0]); | |
6939 | if (key.objectid == bytenr && | |
6940 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6941 | key.offset == num_bytes) | |
6942 | ret = 0; | |
6943 | } | |
6944 | ||
6945 | if (ret > 0 && skinny_metadata) { | |
6946 | skinny_metadata = false; | |
9ce49a0b | 6947 | key.objectid = bytenr; |
3173a18f JB |
6948 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6949 | key.offset = num_bytes; | |
6950 | btrfs_release_path(path); | |
6951 | ret = btrfs_search_slot(trans, extent_root, | |
6952 | &key, path, -1, 1); | |
6953 | } | |
6954 | ||
f3465ca4 | 6955 | if (ret) { |
5d163e0e JM |
6956 | btrfs_err(info, |
6957 | "umm, got %d back from search, was looking for %llu", | |
6958 | ret, bytenr); | |
b783e62d | 6959 | if (ret > 0) |
2ff7e61e | 6960 | btrfs_print_leaf(info, path->nodes[0]); |
f3465ca4 | 6961 | } |
005d6427 | 6962 | if (ret < 0) { |
66642832 | 6963 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6964 | goto out; |
6965 | } | |
31840ae1 ZY |
6966 | extent_slot = path->slots[0]; |
6967 | } | |
fae7f21c | 6968 | } else if (WARN_ON(ret == -ENOENT)) { |
2ff7e61e | 6969 | btrfs_print_leaf(info, path->nodes[0]); |
c2cf52eb SK |
6970 | btrfs_err(info, |
6971 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6972 | bytenr, parent, root_objectid, owner_objectid, |
6973 | owner_offset); | |
66642832 | 6974 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6975 | goto out; |
79787eaa | 6976 | } else { |
66642832 | 6977 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6978 | goto out; |
7bb86316 | 6979 | } |
5f39d397 CM |
6980 | |
6981 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6982 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6983 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6984 | if (item_size < sizeof(*ei)) { | |
6985 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
87bde3cd JM |
6986 | ret = convert_extent_item_v0(trans, info, path, owner_objectid, |
6987 | 0); | |
005d6427 | 6988 | if (ret < 0) { |
66642832 | 6989 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6990 | goto out; |
6991 | } | |
5d4f98a2 | 6992 | |
b3b4aa74 | 6993 | btrfs_release_path(path); |
5d4f98a2 YZ |
6994 | path->leave_spinning = 1; |
6995 | ||
6996 | key.objectid = bytenr; | |
6997 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6998 | key.offset = num_bytes; | |
6999 | ||
7000 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
7001 | -1, 1); | |
7002 | if (ret) { | |
5d163e0e JM |
7003 | btrfs_err(info, |
7004 | "umm, got %d back from search, was looking for %llu", | |
c1c9ff7c | 7005 | ret, bytenr); |
2ff7e61e | 7006 | btrfs_print_leaf(info, path->nodes[0]); |
5d4f98a2 | 7007 | } |
005d6427 | 7008 | if (ret < 0) { |
66642832 | 7009 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7010 | goto out; |
7011 | } | |
7012 | ||
5d4f98a2 YZ |
7013 | extent_slot = path->slots[0]; |
7014 | leaf = path->nodes[0]; | |
7015 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
7016 | } | |
7017 | #endif | |
7018 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 7019 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 7020 | struct btrfs_extent_item); |
3173a18f JB |
7021 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7022 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7023 | struct btrfs_tree_block_info *bi; |
7024 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7025 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7026 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7027 | } | |
56bec294 | 7028 | |
5d4f98a2 | 7029 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7030 | if (refs < refs_to_drop) { |
5d163e0e JM |
7031 | btrfs_err(info, |
7032 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7033 | refs_to_drop, refs, bytenr); | |
32b02538 | 7034 | ret = -EINVAL; |
66642832 | 7035 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7036 | goto out; |
7037 | } | |
56bec294 | 7038 | refs -= refs_to_drop; |
5f39d397 | 7039 | |
5d4f98a2 YZ |
7040 | if (refs > 0) { |
7041 | if (extent_op) | |
7042 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7043 | /* | |
7044 | * In the case of inline back ref, reference count will | |
7045 | * be updated by remove_extent_backref | |
952fccac | 7046 | */ |
5d4f98a2 YZ |
7047 | if (iref) { |
7048 | BUG_ON(!found_extent); | |
7049 | } else { | |
7050 | btrfs_set_extent_refs(leaf, ei, refs); | |
7051 | btrfs_mark_buffer_dirty(leaf); | |
7052 | } | |
7053 | if (found_extent) { | |
87bde3cd | 7054 | ret = remove_extent_backref(trans, info, path, |
5d4f98a2 | 7055 | iref, refs_to_drop, |
fcebe456 | 7056 | is_data, &last_ref); |
005d6427 | 7057 | if (ret) { |
66642832 | 7058 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7059 | goto out; |
7060 | } | |
952fccac | 7061 | } |
0b246afa | 7062 | add_pinned_bytes(info, -num_bytes, owner_objectid, |
b150a4f1 | 7063 | root_objectid); |
5d4f98a2 | 7064 | } else { |
5d4f98a2 YZ |
7065 | if (found_extent) { |
7066 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7067 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7068 | if (iref) { |
7069 | BUG_ON(path->slots[0] != extent_slot); | |
7070 | } else { | |
7071 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7072 | path->slots[0] = extent_slot; | |
7073 | num_to_del = 2; | |
7074 | } | |
78fae27e | 7075 | } |
b9473439 | 7076 | |
fcebe456 | 7077 | last_ref = 1; |
952fccac CM |
7078 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7079 | num_to_del); | |
005d6427 | 7080 | if (ret) { |
66642832 | 7081 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7082 | goto out; |
7083 | } | |
b3b4aa74 | 7084 | btrfs_release_path(path); |
21af804c | 7085 | |
5d4f98a2 | 7086 | if (is_data) { |
5b4aacef | 7087 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7088 | if (ret) { |
66642832 | 7089 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7090 | goto out; |
7091 | } | |
459931ec CM |
7092 | } |
7093 | ||
0b246afa | 7094 | ret = add_to_free_space_tree(trans, info, bytenr, num_bytes); |
1e144fb8 | 7095 | if (ret) { |
66642832 | 7096 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7097 | goto out; |
7098 | } | |
7099 | ||
0b246afa | 7100 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 7101 | if (ret) { |
66642832 | 7102 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7103 | goto out; |
7104 | } | |
a28ec197 | 7105 | } |
fcebe456 JB |
7106 | btrfs_release_path(path); |
7107 | ||
79787eaa | 7108 | out: |
5caf2a00 | 7109 | btrfs_free_path(path); |
a28ec197 CM |
7110 | return ret; |
7111 | } | |
7112 | ||
1887be66 | 7113 | /* |
f0486c68 | 7114 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7115 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7116 | * a given extent, and if there are no other delayed refs to be processed, it | |
7117 | * removes it from the tree. | |
7118 | */ | |
7119 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7120 | u64 bytenr) |
1887be66 CM |
7121 | { |
7122 | struct btrfs_delayed_ref_head *head; | |
7123 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7124 | int ret = 0; |
1887be66 CM |
7125 | |
7126 | delayed_refs = &trans->transaction->delayed_refs; | |
7127 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7128 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7129 | if (!head) |
cf93da7b | 7130 | goto out_delayed_unlock; |
1887be66 | 7131 | |
d7df2c79 | 7132 | spin_lock(&head->lock); |
c6fc2454 | 7133 | if (!list_empty(&head->ref_list)) |
1887be66 CM |
7134 | goto out; |
7135 | ||
5d4f98a2 YZ |
7136 | if (head->extent_op) { |
7137 | if (!head->must_insert_reserved) | |
7138 | goto out; | |
78a6184a | 7139 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7140 | head->extent_op = NULL; |
7141 | } | |
7142 | ||
1887be66 CM |
7143 | /* |
7144 | * waiting for the lock here would deadlock. If someone else has it | |
7145 | * locked they are already in the process of dropping it anyway | |
7146 | */ | |
7147 | if (!mutex_trylock(&head->mutex)) | |
7148 | goto out; | |
7149 | ||
7150 | /* | |
7151 | * at this point we have a head with no other entries. Go | |
7152 | * ahead and process it. | |
7153 | */ | |
7154 | head->node.in_tree = 0; | |
c46effa6 | 7155 | rb_erase(&head->href_node, &delayed_refs->href_root); |
c3e69d58 | 7156 | |
d7df2c79 | 7157 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7158 | |
7159 | /* | |
7160 | * we don't take a ref on the node because we're removing it from the | |
7161 | * tree, so we just steal the ref the tree was holding. | |
7162 | */ | |
c3e69d58 | 7163 | delayed_refs->num_heads--; |
d7df2c79 | 7164 | if (head->processing == 0) |
c3e69d58 | 7165 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7166 | head->processing = 0; |
7167 | spin_unlock(&head->lock); | |
1887be66 CM |
7168 | spin_unlock(&delayed_refs->lock); |
7169 | ||
f0486c68 YZ |
7170 | BUG_ON(head->extent_op); |
7171 | if (head->must_insert_reserved) | |
7172 | ret = 1; | |
7173 | ||
7174 | mutex_unlock(&head->mutex); | |
1887be66 | 7175 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 7176 | return ret; |
1887be66 | 7177 | out: |
d7df2c79 | 7178 | spin_unlock(&head->lock); |
cf93da7b CM |
7179 | |
7180 | out_delayed_unlock: | |
1887be66 CM |
7181 | spin_unlock(&delayed_refs->lock); |
7182 | return 0; | |
7183 | } | |
7184 | ||
f0486c68 YZ |
7185 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7186 | struct btrfs_root *root, | |
7187 | struct extent_buffer *buf, | |
5581a51a | 7188 | u64 parent, int last_ref) |
f0486c68 | 7189 | { |
0b246afa | 7190 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7191 | int pin = 1; |
f0486c68 YZ |
7192 | int ret; |
7193 | ||
7194 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
0b246afa JM |
7195 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, |
7196 | buf->start, buf->len, | |
7197 | parent, | |
7198 | root->root_key.objectid, | |
7199 | btrfs_header_level(buf), | |
7200 | BTRFS_DROP_DELAYED_REF, NULL); | |
79787eaa | 7201 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
7202 | } |
7203 | ||
7204 | if (!last_ref) | |
7205 | return; | |
7206 | ||
f0486c68 | 7207 | if (btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7208 | struct btrfs_block_group_cache *cache; |
7209 | ||
f0486c68 | 7210 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7211 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7212 | if (!ret) |
37be25bc | 7213 | goto out; |
f0486c68 YZ |
7214 | } |
7215 | ||
0b246afa | 7216 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7217 | |
f0486c68 | 7218 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7219 | pin_down_extent(fs_info, cache, buf->start, |
7220 | buf->len, 1); | |
6219872d | 7221 | btrfs_put_block_group(cache); |
37be25bc | 7222 | goto out; |
f0486c68 YZ |
7223 | } |
7224 | ||
7225 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7226 | ||
7227 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7228 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7229 | btrfs_put_block_group(cache); |
71ff6437 | 7230 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
b150a4f1 | 7231 | pin = 0; |
f0486c68 YZ |
7232 | } |
7233 | out: | |
b150a4f1 | 7234 | if (pin) |
0b246afa | 7235 | add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf), |
b150a4f1 JB |
7236 | root->root_key.objectid); |
7237 | ||
a826d6dc JB |
7238 | /* |
7239 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
7240 | * anymore. | |
7241 | */ | |
7242 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
7243 | } |
7244 | ||
79787eaa | 7245 | /* Can return -ENOMEM */ |
2ff7e61e JM |
7246 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
7247 | struct btrfs_fs_info *fs_info, | |
66d7e7f0 | 7248 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7249 | u64 owner, u64 offset) |
925baedd CM |
7250 | { |
7251 | int ret; | |
7252 | ||
f5ee5c9a | 7253 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7254 | return 0; |
fccb84c9 | 7255 | |
0b246afa | 7256 | add_pinned_bytes(fs_info, num_bytes, owner, root_objectid); |
b150a4f1 | 7257 | |
56bec294 CM |
7258 | /* |
7259 | * tree log blocks never actually go into the extent allocation | |
7260 | * tree, just update pinning info and exit early. | |
56bec294 | 7261 | */ |
5d4f98a2 YZ |
7262 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7263 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7264 | /* unlocks the pinned mutex */ |
2ff7e61e | 7265 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
56bec294 | 7266 | ret = 0; |
5d4f98a2 | 7267 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
7268 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7269 | num_bytes, | |
5d4f98a2 | 7270 | parent, root_objectid, (int)owner, |
b06c4bf5 | 7271 | BTRFS_DROP_DELAYED_REF, NULL); |
5d4f98a2 | 7272 | } else { |
66d7e7f0 AJ |
7273 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7274 | num_bytes, | |
7275 | parent, root_objectid, owner, | |
5846a3c2 | 7276 | offset, 0, |
fef394f7 | 7277 | BTRFS_DROP_DELAYED_REF); |
56bec294 | 7278 | } |
925baedd CM |
7279 | return ret; |
7280 | } | |
7281 | ||
817d52f8 JB |
7282 | /* |
7283 | * when we wait for progress in the block group caching, its because | |
7284 | * our allocation attempt failed at least once. So, we must sleep | |
7285 | * and let some progress happen before we try again. | |
7286 | * | |
7287 | * This function will sleep at least once waiting for new free space to | |
7288 | * show up, and then it will check the block group free space numbers | |
7289 | * for our min num_bytes. Another option is to have it go ahead | |
7290 | * and look in the rbtree for a free extent of a given size, but this | |
7291 | * is a good start. | |
36cce922 JB |
7292 | * |
7293 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7294 | * any of the information in this block group. | |
817d52f8 | 7295 | */ |
36cce922 | 7296 | static noinline void |
817d52f8 JB |
7297 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7298 | u64 num_bytes) | |
7299 | { | |
11833d66 | 7300 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7301 | |
11833d66 YZ |
7302 | caching_ctl = get_caching_control(cache); |
7303 | if (!caching_ctl) | |
36cce922 | 7304 | return; |
817d52f8 | 7305 | |
11833d66 | 7306 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7307 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7308 | |
7309 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7310 | } |
7311 | ||
7312 | static noinline int | |
7313 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7314 | { | |
7315 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7316 | int ret = 0; |
11833d66 YZ |
7317 | |
7318 | caching_ctl = get_caching_control(cache); | |
7319 | if (!caching_ctl) | |
36cce922 | 7320 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7321 | |
7322 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7323 | if (cache->cached == BTRFS_CACHE_ERROR) |
7324 | ret = -EIO; | |
11833d66 | 7325 | put_caching_control(caching_ctl); |
36cce922 | 7326 | return ret; |
817d52f8 JB |
7327 | } |
7328 | ||
31e50229 | 7329 | int __get_raid_index(u64 flags) |
b742bb82 | 7330 | { |
7738a53a | 7331 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 7332 | return BTRFS_RAID_RAID10; |
7738a53a | 7333 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 7334 | return BTRFS_RAID_RAID1; |
7738a53a | 7335 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 7336 | return BTRFS_RAID_DUP; |
7738a53a | 7337 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 7338 | return BTRFS_RAID_RAID0; |
53b381b3 | 7339 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 7340 | return BTRFS_RAID_RAID5; |
53b381b3 | 7341 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 7342 | return BTRFS_RAID_RAID6; |
7738a53a | 7343 | |
e942f883 | 7344 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
7345 | } |
7346 | ||
6ab0a202 | 7347 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
7738a53a | 7348 | { |
31e50229 | 7349 | return __get_raid_index(cache->flags); |
7738a53a ID |
7350 | } |
7351 | ||
6ab0a202 JM |
7352 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
7353 | [BTRFS_RAID_RAID10] = "raid10", | |
7354 | [BTRFS_RAID_RAID1] = "raid1", | |
7355 | [BTRFS_RAID_DUP] = "dup", | |
7356 | [BTRFS_RAID_RAID0] = "raid0", | |
7357 | [BTRFS_RAID_SINGLE] = "single", | |
7358 | [BTRFS_RAID_RAID5] = "raid5", | |
7359 | [BTRFS_RAID_RAID6] = "raid6", | |
7360 | }; | |
7361 | ||
1b8e5df6 | 7362 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
7363 | { |
7364 | if (type >= BTRFS_NR_RAID_TYPES) | |
7365 | return NULL; | |
7366 | ||
7367 | return btrfs_raid_type_names[type]; | |
7368 | } | |
7369 | ||
817d52f8 | 7370 | enum btrfs_loop_type { |
285ff5af JB |
7371 | LOOP_CACHING_NOWAIT = 0, |
7372 | LOOP_CACHING_WAIT = 1, | |
7373 | LOOP_ALLOC_CHUNK = 2, | |
7374 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7375 | }; |
7376 | ||
e570fd27 MX |
7377 | static inline void |
7378 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7379 | int delalloc) | |
7380 | { | |
7381 | if (delalloc) | |
7382 | down_read(&cache->data_rwsem); | |
7383 | } | |
7384 | ||
7385 | static inline void | |
7386 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7387 | int delalloc) | |
7388 | { | |
7389 | btrfs_get_block_group(cache); | |
7390 | if (delalloc) | |
7391 | down_read(&cache->data_rwsem); | |
7392 | } | |
7393 | ||
7394 | static struct btrfs_block_group_cache * | |
7395 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7396 | struct btrfs_free_cluster *cluster, | |
7397 | int delalloc) | |
7398 | { | |
89771cc9 | 7399 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7400 | |
e570fd27 | 7401 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7402 | while (1) { |
7403 | used_bg = cluster->block_group; | |
7404 | if (!used_bg) | |
7405 | return NULL; | |
7406 | ||
7407 | if (used_bg == block_group) | |
e570fd27 MX |
7408 | return used_bg; |
7409 | ||
6719afdc | 7410 | btrfs_get_block_group(used_bg); |
e570fd27 | 7411 | |
6719afdc GU |
7412 | if (!delalloc) |
7413 | return used_bg; | |
e570fd27 | 7414 | |
6719afdc GU |
7415 | if (down_read_trylock(&used_bg->data_rwsem)) |
7416 | return used_bg; | |
e570fd27 | 7417 | |
6719afdc | 7418 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7419 | |
e321f8a8 LB |
7420 | /* We should only have one-level nested. */ |
7421 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7422 | |
6719afdc GU |
7423 | spin_lock(&cluster->refill_lock); |
7424 | if (used_bg == cluster->block_group) | |
7425 | return used_bg; | |
e570fd27 | 7426 | |
6719afdc GU |
7427 | up_read(&used_bg->data_rwsem); |
7428 | btrfs_put_block_group(used_bg); | |
7429 | } | |
e570fd27 MX |
7430 | } |
7431 | ||
7432 | static inline void | |
7433 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7434 | int delalloc) | |
7435 | { | |
7436 | if (delalloc) | |
7437 | up_read(&cache->data_rwsem); | |
7438 | btrfs_put_block_group(cache); | |
7439 | } | |
7440 | ||
fec577fb CM |
7441 | /* |
7442 | * walks the btree of allocated extents and find a hole of a given size. | |
7443 | * The key ins is changed to record the hole: | |
a4820398 | 7444 | * ins->objectid == start position |
62e2749e | 7445 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7446 | * ins->offset == the size of the hole. |
fec577fb | 7447 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7448 | * |
7449 | * If there is no suitable free space, we will record the max size of | |
7450 | * the free space extent currently. | |
fec577fb | 7451 | */ |
87bde3cd | 7452 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7453 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7454 | u64 hint_byte, struct btrfs_key *ins, | |
7455 | u64 flags, int delalloc) | |
fec577fb | 7456 | { |
80eb234a | 7457 | int ret = 0; |
0b246afa | 7458 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7459 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7460 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7461 | u64 search_start = 0; |
a4820398 | 7462 | u64 max_extent_size = 0; |
c759c4e1 | 7463 | u64 empty_cluster = 0; |
80eb234a | 7464 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7465 | int loop = 0; |
b6919a58 | 7466 | int index = __get_raid_index(flags); |
0a24325e | 7467 | bool failed_cluster_refill = false; |
1cdda9b8 | 7468 | bool failed_alloc = false; |
67377734 | 7469 | bool use_cluster = true; |
60d2adbb | 7470 | bool have_caching_bg = false; |
13a0db5a | 7471 | bool orig_have_caching_bg = false; |
a5e681d9 | 7472 | bool full_search = false; |
fec577fb | 7473 | |
0b246afa | 7474 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7475 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7476 | ins->objectid = 0; |
7477 | ins->offset = 0; | |
b1a4d965 | 7478 | |
71ff6437 | 7479 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7480 | |
0b246afa | 7481 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7482 | if (!space_info) { |
0b246afa | 7483 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7484 | return -ENOSPC; |
7485 | } | |
2552d17e | 7486 | |
67377734 | 7487 | /* |
4f4db217 JB |
7488 | * If our free space is heavily fragmented we may not be able to make |
7489 | * big contiguous allocations, so instead of doing the expensive search | |
7490 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7491 | * can go ahead and search for a more manageable chunk. | |
7492 | * | |
7493 | * If our max_extent_size is large enough for our allocation simply | |
7494 | * disable clustering since we will likely not be able to find enough | |
7495 | * space to create a cluster and induce latency trying. | |
67377734 | 7496 | */ |
4f4db217 JB |
7497 | if (unlikely(space_info->max_extent_size)) { |
7498 | spin_lock(&space_info->lock); | |
7499 | if (space_info->max_extent_size && | |
7500 | num_bytes > space_info->max_extent_size) { | |
7501 | ins->offset = space_info->max_extent_size; | |
7502 | spin_unlock(&space_info->lock); | |
7503 | return -ENOSPC; | |
7504 | } else if (space_info->max_extent_size) { | |
7505 | use_cluster = false; | |
7506 | } | |
7507 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7508 | } |
0f9dd46c | 7509 | |
2ff7e61e | 7510 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7511 | if (last_ptr) { |
fa9c0d79 CM |
7512 | spin_lock(&last_ptr->lock); |
7513 | if (last_ptr->block_group) | |
7514 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7515 | if (last_ptr->fragmented) { |
7516 | /* | |
7517 | * We still set window_start so we can keep track of the | |
7518 | * last place we found an allocation to try and save | |
7519 | * some time. | |
7520 | */ | |
7521 | hint_byte = last_ptr->window_start; | |
7522 | use_cluster = false; | |
7523 | } | |
fa9c0d79 | 7524 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7525 | } |
fa9c0d79 | 7526 | |
2ff7e61e | 7527 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7528 | search_start = max(search_start, hint_byte); |
2552d17e | 7529 | if (search_start == hint_byte) { |
0b246afa | 7530 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7531 | /* |
7532 | * we don't want to use the block group if it doesn't match our | |
7533 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7534 | * |
7535 | * However if we are re-searching with an ideal block group | |
7536 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7537 | */ |
b6919a58 | 7538 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7539 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7540 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7541 | if (list_empty(&block_group->list) || |
7542 | block_group->ro) { | |
7543 | /* | |
7544 | * someone is removing this block group, | |
7545 | * we can't jump into the have_block_group | |
7546 | * target because our list pointers are not | |
7547 | * valid | |
7548 | */ | |
7549 | btrfs_put_block_group(block_group); | |
7550 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7551 | } else { |
b742bb82 | 7552 | index = get_block_group_index(block_group); |
e570fd27 | 7553 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7554 | goto have_block_group; |
ccf0e725 | 7555 | } |
2552d17e | 7556 | } else if (block_group) { |
fa9c0d79 | 7557 | btrfs_put_block_group(block_group); |
2552d17e | 7558 | } |
42e70e7a | 7559 | } |
2552d17e | 7560 | search: |
60d2adbb | 7561 | have_caching_bg = false; |
a5e681d9 JB |
7562 | if (index == 0 || index == __get_raid_index(flags)) |
7563 | full_search = true; | |
80eb234a | 7564 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7565 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7566 | list) { | |
6226cb0a | 7567 | u64 offset; |
817d52f8 | 7568 | int cached; |
8a1413a2 | 7569 | |
e570fd27 | 7570 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7571 | search_start = block_group->key.objectid; |
42e70e7a | 7572 | |
83a50de9 CM |
7573 | /* |
7574 | * this can happen if we end up cycling through all the | |
7575 | * raid types, but we want to make sure we only allocate | |
7576 | * for the proper type. | |
7577 | */ | |
b6919a58 | 7578 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7579 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7580 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7581 | BTRFS_BLOCK_GROUP_RAID5 | |
7582 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7583 | BTRFS_BLOCK_GROUP_RAID10; |
7584 | ||
7585 | /* | |
7586 | * if they asked for extra copies and this block group | |
7587 | * doesn't provide them, bail. This does allow us to | |
7588 | * fill raid0 from raid1. | |
7589 | */ | |
b6919a58 | 7590 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7591 | goto loop; |
7592 | } | |
7593 | ||
2552d17e | 7594 | have_block_group: |
291c7d2f JB |
7595 | cached = block_group_cache_done(block_group); |
7596 | if (unlikely(!cached)) { | |
a5e681d9 | 7597 | have_caching_bg = true; |
f6373bf3 | 7598 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7599 | BUG_ON(ret < 0); |
7600 | ret = 0; | |
817d52f8 JB |
7601 | } |
7602 | ||
36cce922 JB |
7603 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7604 | goto loop; | |
ea6a478e | 7605 | if (unlikely(block_group->ro)) |
2552d17e | 7606 | goto loop; |
0f9dd46c | 7607 | |
0a24325e | 7608 | /* |
062c05c4 AO |
7609 | * Ok we want to try and use the cluster allocator, so |
7610 | * lets look there | |
0a24325e | 7611 | */ |
c759c4e1 | 7612 | if (last_ptr && use_cluster) { |
215a63d1 | 7613 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7614 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7615 | /* |
7616 | * the refill lock keeps out other | |
7617 | * people trying to start a new cluster | |
7618 | */ | |
e570fd27 MX |
7619 | used_block_group = btrfs_lock_cluster(block_group, |
7620 | last_ptr, | |
7621 | delalloc); | |
7622 | if (!used_block_group) | |
44fb5511 | 7623 | goto refill_cluster; |
274bd4fb | 7624 | |
e570fd27 MX |
7625 | if (used_block_group != block_group && |
7626 | (used_block_group->ro || | |
7627 | !block_group_bits(used_block_group, flags))) | |
7628 | goto release_cluster; | |
44fb5511 | 7629 | |
274bd4fb | 7630 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7631 | last_ptr, |
7632 | num_bytes, | |
7633 | used_block_group->key.objectid, | |
7634 | &max_extent_size); | |
fa9c0d79 CM |
7635 | if (offset) { |
7636 | /* we have a block, we're done */ | |
7637 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7638 | trace_btrfs_reserve_extent_cluster(fs_info, |
89d4346a MX |
7639 | used_block_group, |
7640 | search_start, num_bytes); | |
215a63d1 | 7641 | if (used_block_group != block_group) { |
e570fd27 MX |
7642 | btrfs_release_block_group(block_group, |
7643 | delalloc); | |
215a63d1 MX |
7644 | block_group = used_block_group; |
7645 | } | |
fa9c0d79 CM |
7646 | goto checks; |
7647 | } | |
7648 | ||
274bd4fb | 7649 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7650 | release_cluster: |
062c05c4 AO |
7651 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7652 | * set up a new clusters, so lets just skip it | |
7653 | * and let the allocator find whatever block | |
7654 | * it can find. If we reach this point, we | |
7655 | * will have tried the cluster allocator | |
7656 | * plenty of times and not have found | |
7657 | * anything, so we are likely way too | |
7658 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7659 | * anything. |
7660 | * | |
7661 | * However, if the cluster is taken from the | |
7662 | * current block group, release the cluster | |
7663 | * first, so that we stand a better chance of | |
7664 | * succeeding in the unclustered | |
7665 | * allocation. */ | |
7666 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7667 | used_block_group != block_group) { |
062c05c4 | 7668 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7669 | btrfs_release_block_group(used_block_group, |
7670 | delalloc); | |
062c05c4 AO |
7671 | goto unclustered_alloc; |
7672 | } | |
7673 | ||
fa9c0d79 CM |
7674 | /* |
7675 | * this cluster didn't work out, free it and | |
7676 | * start over | |
7677 | */ | |
7678 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7679 | ||
e570fd27 MX |
7680 | if (used_block_group != block_group) |
7681 | btrfs_release_block_group(used_block_group, | |
7682 | delalloc); | |
7683 | refill_cluster: | |
a5f6f719 AO |
7684 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7685 | spin_unlock(&last_ptr->refill_lock); | |
7686 | goto unclustered_alloc; | |
7687 | } | |
7688 | ||
8de972b4 CM |
7689 | aligned_cluster = max_t(unsigned long, |
7690 | empty_cluster + empty_size, | |
7691 | block_group->full_stripe_len); | |
7692 | ||
fa9c0d79 | 7693 | /* allocate a cluster in this block group */ |
2ff7e61e | 7694 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7695 | last_ptr, search_start, |
7696 | num_bytes, | |
7697 | aligned_cluster); | |
fa9c0d79 CM |
7698 | if (ret == 0) { |
7699 | /* | |
7700 | * now pull our allocation out of this | |
7701 | * cluster | |
7702 | */ | |
7703 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7704 | last_ptr, |
7705 | num_bytes, | |
7706 | search_start, | |
7707 | &max_extent_size); | |
fa9c0d79 CM |
7708 | if (offset) { |
7709 | /* we found one, proceed */ | |
7710 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7711 | trace_btrfs_reserve_extent_cluster(fs_info, |
3f7de037 JB |
7712 | block_group, search_start, |
7713 | num_bytes); | |
fa9c0d79 CM |
7714 | goto checks; |
7715 | } | |
0a24325e JB |
7716 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7717 | && !failed_cluster_refill) { | |
817d52f8 JB |
7718 | spin_unlock(&last_ptr->refill_lock); |
7719 | ||
0a24325e | 7720 | failed_cluster_refill = true; |
817d52f8 JB |
7721 | wait_block_group_cache_progress(block_group, |
7722 | num_bytes + empty_cluster + empty_size); | |
7723 | goto have_block_group; | |
fa9c0d79 | 7724 | } |
817d52f8 | 7725 | |
fa9c0d79 CM |
7726 | /* |
7727 | * at this point we either didn't find a cluster | |
7728 | * or we weren't able to allocate a block from our | |
7729 | * cluster. Free the cluster we've been trying | |
7730 | * to use, and go to the next block group | |
7731 | */ | |
0a24325e | 7732 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7733 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7734 | goto loop; |
fa9c0d79 CM |
7735 | } |
7736 | ||
062c05c4 | 7737 | unclustered_alloc: |
c759c4e1 JB |
7738 | /* |
7739 | * We are doing an unclustered alloc, set the fragmented flag so | |
7740 | * we don't bother trying to setup a cluster again until we get | |
7741 | * more space. | |
7742 | */ | |
7743 | if (unlikely(last_ptr)) { | |
7744 | spin_lock(&last_ptr->lock); | |
7745 | last_ptr->fragmented = 1; | |
7746 | spin_unlock(&last_ptr->lock); | |
7747 | } | |
0c9b36e0 LB |
7748 | if (cached) { |
7749 | struct btrfs_free_space_ctl *ctl = | |
7750 | block_group->free_space_ctl; | |
7751 | ||
7752 | spin_lock(&ctl->tree_lock); | |
7753 | if (ctl->free_space < | |
7754 | num_bytes + empty_cluster + empty_size) { | |
7755 | if (ctl->free_space > max_extent_size) | |
7756 | max_extent_size = ctl->free_space; | |
7757 | spin_unlock(&ctl->tree_lock); | |
7758 | goto loop; | |
7759 | } | |
7760 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7761 | } |
a5f6f719 | 7762 | |
6226cb0a | 7763 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7764 | num_bytes, empty_size, |
7765 | &max_extent_size); | |
1cdda9b8 JB |
7766 | /* |
7767 | * If we didn't find a chunk, and we haven't failed on this | |
7768 | * block group before, and this block group is in the middle of | |
7769 | * caching and we are ok with waiting, then go ahead and wait | |
7770 | * for progress to be made, and set failed_alloc to true. | |
7771 | * | |
7772 | * If failed_alloc is true then we've already waited on this | |
7773 | * block group once and should move on to the next block group. | |
7774 | */ | |
7775 | if (!offset && !failed_alloc && !cached && | |
7776 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7777 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7778 | num_bytes + empty_size); |
7779 | failed_alloc = true; | |
817d52f8 | 7780 | goto have_block_group; |
1cdda9b8 JB |
7781 | } else if (!offset) { |
7782 | goto loop; | |
817d52f8 | 7783 | } |
fa9c0d79 | 7784 | checks: |
0b246afa | 7785 | search_start = ALIGN(offset, fs_info->stripesize); |
25179201 | 7786 | |
2552d17e JB |
7787 | /* move on to the next group */ |
7788 | if (search_start + num_bytes > | |
215a63d1 MX |
7789 | block_group->key.objectid + block_group->key.offset) { |
7790 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7791 | goto loop; |
6226cb0a | 7792 | } |
f5a31e16 | 7793 | |
f0486c68 | 7794 | if (offset < search_start) |
215a63d1 | 7795 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7796 | search_start - offset); |
7797 | BUG_ON(offset > search_start); | |
2552d17e | 7798 | |
18513091 WX |
7799 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7800 | num_bytes, delalloc); | |
f0486c68 | 7801 | if (ret == -EAGAIN) { |
215a63d1 | 7802 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7803 | goto loop; |
0f9dd46c | 7804 | } |
9cfa3e34 | 7805 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7806 | |
f0486c68 | 7807 | /* we are all good, lets return */ |
2552d17e JB |
7808 | ins->objectid = search_start; |
7809 | ins->offset = num_bytes; | |
d2fb3437 | 7810 | |
71ff6437 | 7811 | trace_btrfs_reserve_extent(fs_info, block_group, |
3f7de037 | 7812 | search_start, num_bytes); |
e570fd27 | 7813 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7814 | break; |
7815 | loop: | |
0a24325e | 7816 | failed_cluster_refill = false; |
1cdda9b8 | 7817 | failed_alloc = false; |
b742bb82 | 7818 | BUG_ON(index != get_block_group_index(block_group)); |
e570fd27 | 7819 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7820 | } |
7821 | up_read(&space_info->groups_sem); | |
7822 | ||
13a0db5a | 7823 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7824 | && !orig_have_caching_bg) | |
7825 | orig_have_caching_bg = true; | |
7826 | ||
60d2adbb MX |
7827 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7828 | goto search; | |
7829 | ||
b742bb82 YZ |
7830 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7831 | goto search; | |
7832 | ||
285ff5af | 7833 | /* |
ccf0e725 JB |
7834 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7835 | * caching kthreads as we move along | |
817d52f8 JB |
7836 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7837 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7838 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7839 | * again | |
fa9c0d79 | 7840 | */ |
723bda20 | 7841 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7842 | index = 0; |
a5e681d9 JB |
7843 | if (loop == LOOP_CACHING_NOWAIT) { |
7844 | /* | |
7845 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7846 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7847 | * full search through. |
7848 | */ | |
13a0db5a | 7849 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7850 | loop = LOOP_CACHING_WAIT; |
7851 | else | |
7852 | loop = LOOP_ALLOC_CHUNK; | |
7853 | } else { | |
7854 | loop++; | |
7855 | } | |
7856 | ||
817d52f8 | 7857 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7858 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7859 | int exist = 0; |
7860 | ||
7861 | trans = current->journal_info; | |
7862 | if (trans) | |
7863 | exist = 1; | |
7864 | else | |
7865 | trans = btrfs_join_transaction(root); | |
00361589 | 7866 | |
00361589 JB |
7867 | if (IS_ERR(trans)) { |
7868 | ret = PTR_ERR(trans); | |
7869 | goto out; | |
7870 | } | |
7871 | ||
2ff7e61e | 7872 | ret = do_chunk_alloc(trans, fs_info, flags, |
ea658bad | 7873 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7874 | |
7875 | /* | |
7876 | * If we can't allocate a new chunk we've already looped | |
7877 | * through at least once, move on to the NO_EMPTY_SIZE | |
7878 | * case. | |
7879 | */ | |
7880 | if (ret == -ENOSPC) | |
7881 | loop = LOOP_NO_EMPTY_SIZE; | |
7882 | ||
ea658bad JB |
7883 | /* |
7884 | * Do not bail out on ENOSPC since we | |
7885 | * can do more things. | |
7886 | */ | |
00361589 | 7887 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7888 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7889 | else |
7890 | ret = 0; | |
f017f15f | 7891 | if (!exist) |
3a45bb20 | 7892 | btrfs_end_transaction(trans); |
00361589 | 7893 | if (ret) |
ea658bad | 7894 | goto out; |
2552d17e JB |
7895 | } |
7896 | ||
723bda20 | 7897 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7898 | /* |
7899 | * Don't loop again if we already have no empty_size and | |
7900 | * no empty_cluster. | |
7901 | */ | |
7902 | if (empty_size == 0 && | |
7903 | empty_cluster == 0) { | |
7904 | ret = -ENOSPC; | |
7905 | goto out; | |
7906 | } | |
723bda20 JB |
7907 | empty_size = 0; |
7908 | empty_cluster = 0; | |
fa9c0d79 | 7909 | } |
723bda20 JB |
7910 | |
7911 | goto search; | |
2552d17e JB |
7912 | } else if (!ins->objectid) { |
7913 | ret = -ENOSPC; | |
d82a6f1d | 7914 | } else if (ins->objectid) { |
c759c4e1 JB |
7915 | if (!use_cluster && last_ptr) { |
7916 | spin_lock(&last_ptr->lock); | |
7917 | last_ptr->window_start = ins->objectid; | |
7918 | spin_unlock(&last_ptr->lock); | |
7919 | } | |
80eb234a | 7920 | ret = 0; |
be744175 | 7921 | } |
79787eaa | 7922 | out: |
4f4db217 JB |
7923 | if (ret == -ENOSPC) { |
7924 | spin_lock(&space_info->lock); | |
7925 | space_info->max_extent_size = max_extent_size; | |
7926 | spin_unlock(&space_info->lock); | |
a4820398 | 7927 | ins->offset = max_extent_size; |
4f4db217 | 7928 | } |
0f70abe2 | 7929 | return ret; |
fec577fb | 7930 | } |
ec44a35c | 7931 | |
ab8d0fc4 JM |
7932 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7933 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7934 | int dump_block_groups) |
0f9dd46c JB |
7935 | { |
7936 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7937 | int index = 0; |
0f9dd46c | 7938 | |
9ed74f2d | 7939 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7940 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7941 | info->flags, | |
4136135b LB |
7942 | info->total_bytes - btrfs_space_info_used(info, true), |
7943 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7944 | btrfs_info(fs_info, |
7945 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7946 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7947 | info->bytes_reserved, info->bytes_may_use, | |
7948 | info->bytes_readonly); | |
9ed74f2d JB |
7949 | spin_unlock(&info->lock); |
7950 | ||
7951 | if (!dump_block_groups) | |
7952 | return; | |
0f9dd46c | 7953 | |
80eb234a | 7954 | down_read(&info->groups_sem); |
b742bb82 YZ |
7955 | again: |
7956 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7957 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7958 | btrfs_info(fs_info, |
7959 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7960 | cache->key.objectid, cache->key.offset, | |
7961 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7962 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7963 | btrfs_dump_free_space(cache, bytes); |
7964 | spin_unlock(&cache->lock); | |
7965 | } | |
b742bb82 YZ |
7966 | if (++index < BTRFS_NR_RAID_TYPES) |
7967 | goto again; | |
80eb234a | 7968 | up_read(&info->groups_sem); |
0f9dd46c | 7969 | } |
e8569813 | 7970 | |
18513091 | 7971 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
7972 | u64 num_bytes, u64 min_alloc_size, |
7973 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7974 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7975 | { |
ab8d0fc4 | 7976 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 7977 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7978 | u64 flags; |
fec577fb | 7979 | int ret; |
925baedd | 7980 | |
1b86826d | 7981 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 7982 | again: |
0b246afa | 7983 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 7984 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 7985 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 7986 | if (!ret && !is_data) { |
ab8d0fc4 | 7987 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 7988 | } else if (ret == -ENOSPC) { |
a4820398 MX |
7989 | if (!final_tried && ins->offset) { |
7990 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 7991 | num_bytes = round_down(num_bytes, |
0b246afa | 7992 | fs_info->sectorsize); |
9e622d6b | 7993 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 7994 | ram_bytes = num_bytes; |
9e622d6b MX |
7995 | if (num_bytes == min_alloc_size) |
7996 | final_tried = true; | |
7997 | goto again; | |
ab8d0fc4 | 7998 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
7999 | struct btrfs_space_info *sinfo; |
8000 | ||
ab8d0fc4 | 8001 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8002 | btrfs_err(fs_info, |
5d163e0e JM |
8003 | "allocation failed flags %llu, wanted %llu", |
8004 | flags, num_bytes); | |
53804280 | 8005 | if (sinfo) |
ab8d0fc4 | 8006 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8007 | } |
925baedd | 8008 | } |
0f9dd46c JB |
8009 | |
8010 | return ret; | |
e6dcd2dc CM |
8011 | } |
8012 | ||
2ff7e61e | 8013 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8014 | u64 start, u64 len, |
8015 | int pin, int delalloc) | |
65b51a00 | 8016 | { |
0f9dd46c | 8017 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8018 | int ret = 0; |
0f9dd46c | 8019 | |
0b246afa | 8020 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8021 | if (!cache) { |
0b246afa JM |
8022 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8023 | start); | |
0f9dd46c JB |
8024 | return -ENOSPC; |
8025 | } | |
1f3c79a2 | 8026 | |
e688b725 | 8027 | if (pin) |
2ff7e61e | 8028 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 8029 | else { |
0b246afa | 8030 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8031 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8032 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8033 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8034 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8035 | } |
31193213 | 8036 | |
fa9c0d79 | 8037 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8038 | return ret; |
8039 | } | |
8040 | ||
2ff7e61e | 8041 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8042 | u64 start, u64 len, int delalloc) |
e688b725 | 8043 | { |
2ff7e61e | 8044 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8045 | } |
8046 | ||
2ff7e61e | 8047 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8048 | u64 start, u64 len) |
8049 | { | |
2ff7e61e | 8050 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8051 | } |
8052 | ||
5d4f98a2 | 8053 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8054 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8055 | u64 parent, u64 root_objectid, |
8056 | u64 flags, u64 owner, u64 offset, | |
8057 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
8058 | { |
8059 | int ret; | |
e6dcd2dc | 8060 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8061 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8062 | struct btrfs_path *path; |
5d4f98a2 YZ |
8063 | struct extent_buffer *leaf; |
8064 | int type; | |
8065 | u32 size; | |
26b8003f | 8066 | |
5d4f98a2 YZ |
8067 | if (parent > 0) |
8068 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8069 | else | |
8070 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8071 | |
5d4f98a2 | 8072 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8073 | |
8074 | path = btrfs_alloc_path(); | |
db5b493a TI |
8075 | if (!path) |
8076 | return -ENOMEM; | |
47e4bb98 | 8077 | |
b9473439 | 8078 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8079 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8080 | ins, size); | |
79787eaa JM |
8081 | if (ret) { |
8082 | btrfs_free_path(path); | |
8083 | return ret; | |
8084 | } | |
0f9dd46c | 8085 | |
5d4f98a2 YZ |
8086 | leaf = path->nodes[0]; |
8087 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8088 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8089 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8090 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8091 | btrfs_set_extent_flags(leaf, extent_item, | |
8092 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8093 | ||
8094 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8095 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8096 | if (parent > 0) { | |
8097 | struct btrfs_shared_data_ref *ref; | |
8098 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8099 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8100 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8101 | } else { | |
8102 | struct btrfs_extent_data_ref *ref; | |
8103 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8104 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8105 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8106 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8107 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8108 | } | |
47e4bb98 CM |
8109 | |
8110 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8111 | btrfs_free_path(path); |
f510cfec | 8112 | |
1e144fb8 OS |
8113 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8114 | ins->offset); | |
8115 | if (ret) | |
8116 | return ret; | |
8117 | ||
6202df69 | 8118 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 8119 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8120 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8121 | ins->objectid, ins->offset); |
f5947066 CM |
8122 | BUG(); |
8123 | } | |
71ff6437 | 8124 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8125 | return ret; |
8126 | } | |
8127 | ||
5d4f98a2 | 8128 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
2ff7e61e | 8129 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8130 | u64 parent, u64 root_objectid, |
8131 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 8132 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
8133 | { |
8134 | int ret; | |
5d4f98a2 YZ |
8135 | struct btrfs_extent_item *extent_item; |
8136 | struct btrfs_tree_block_info *block_info; | |
8137 | struct btrfs_extent_inline_ref *iref; | |
8138 | struct btrfs_path *path; | |
8139 | struct extent_buffer *leaf; | |
3173a18f | 8140 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 8141 | u64 num_bytes = ins->offset; |
0b246afa | 8142 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f JB |
8143 | |
8144 | if (!skinny_metadata) | |
8145 | size += sizeof(*block_info); | |
1c2308f8 | 8146 | |
5d4f98a2 | 8147 | path = btrfs_alloc_path(); |
857cc2fc | 8148 | if (!path) { |
2ff7e61e | 8149 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8150 | fs_info->nodesize); |
d8926bb3 | 8151 | return -ENOMEM; |
857cc2fc | 8152 | } |
56bec294 | 8153 | |
5d4f98a2 YZ |
8154 | path->leave_spinning = 1; |
8155 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
8156 | ins, size); | |
79787eaa | 8157 | if (ret) { |
dd825259 | 8158 | btrfs_free_path(path); |
2ff7e61e | 8159 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8160 | fs_info->nodesize); |
79787eaa JM |
8161 | return ret; |
8162 | } | |
5d4f98a2 YZ |
8163 | |
8164 | leaf = path->nodes[0]; | |
8165 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8166 | struct btrfs_extent_item); | |
8167 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8168 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8169 | btrfs_set_extent_flags(leaf, extent_item, | |
8170 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8171 | |
3173a18f JB |
8172 | if (skinny_metadata) { |
8173 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
0b246afa | 8174 | num_bytes = fs_info->nodesize; |
3173a18f JB |
8175 | } else { |
8176 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
8177 | btrfs_set_tree_block_key(leaf, block_info, key); | |
8178 | btrfs_set_tree_block_level(leaf, block_info, level); | |
8179 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
8180 | } | |
5d4f98a2 | 8181 | |
5d4f98a2 YZ |
8182 | if (parent > 0) { |
8183 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
8184 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8185 | BTRFS_SHARED_BLOCK_REF_KEY); | |
8186 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8187 | } else { | |
8188 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8189 | BTRFS_TREE_BLOCK_REF_KEY); | |
8190 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
8191 | } | |
8192 | ||
8193 | btrfs_mark_buffer_dirty(leaf); | |
8194 | btrfs_free_path(path); | |
8195 | ||
1e144fb8 OS |
8196 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8197 | num_bytes); | |
8198 | if (ret) | |
8199 | return ret; | |
8200 | ||
6202df69 JM |
8201 | ret = update_block_group(trans, fs_info, ins->objectid, |
8202 | fs_info->nodesize, 1); | |
79787eaa | 8203 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8204 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8205 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
8206 | BUG(); |
8207 | } | |
0be5dc67 | 8208 | |
71ff6437 | 8209 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, |
0b246afa | 8210 | fs_info->nodesize); |
5d4f98a2 YZ |
8211 | return ret; |
8212 | } | |
8213 | ||
8214 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 | 8215 | u64 root_objectid, u64 owner, |
5846a3c2 QW |
8216 | u64 offset, u64 ram_bytes, |
8217 | struct btrfs_key *ins) | |
5d4f98a2 | 8218 | { |
2ff7e61e | 8219 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
8220 | int ret; |
8221 | ||
8222 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
8223 | ||
0b246afa | 8224 | ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid, |
66d7e7f0 AJ |
8225 | ins->offset, 0, |
8226 | root_objectid, owner, offset, | |
fef394f7 | 8227 | ram_bytes, BTRFS_ADD_DELAYED_EXTENT); |
e6dcd2dc CM |
8228 | return ret; |
8229 | } | |
e02119d5 CM |
8230 | |
8231 | /* | |
8232 | * this is used by the tree logging recovery code. It records that | |
8233 | * an extent has been allocated and makes sure to clear the free | |
8234 | * space cache bits as well | |
8235 | */ | |
5d4f98a2 | 8236 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8237 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8238 | u64 root_objectid, u64 owner, u64 offset, |
8239 | struct btrfs_key *ins) | |
e02119d5 CM |
8240 | { |
8241 | int ret; | |
8242 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8243 | struct btrfs_space_info *space_info; |
11833d66 | 8244 | |
8c2a1a30 JB |
8245 | /* |
8246 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8247 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8248 | */ |
0b246afa | 8249 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8250 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8251 | ins->offset); | |
b50c6e25 | 8252 | if (ret) |
8c2a1a30 | 8253 | return ret; |
11833d66 YZ |
8254 | } |
8255 | ||
0b246afa | 8256 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8257 | if (!block_group) |
8258 | return -EINVAL; | |
8259 | ||
ed7a6948 WX |
8260 | space_info = block_group->space_info; |
8261 | spin_lock(&space_info->lock); | |
8262 | spin_lock(&block_group->lock); | |
8263 | space_info->bytes_reserved += ins->offset; | |
8264 | block_group->reserved += ins->offset; | |
8265 | spin_unlock(&block_group->lock); | |
8266 | spin_unlock(&space_info->lock); | |
8267 | ||
2ff7e61e | 8268 | ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid, |
5d4f98a2 | 8269 | 0, owner, offset, ins, 1); |
b50c6e25 | 8270 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8271 | return ret; |
8272 | } | |
8273 | ||
48a3b636 ES |
8274 | static struct extent_buffer * |
8275 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8276 | u64 bytenr, int level) |
65b51a00 | 8277 | { |
0b246afa | 8278 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8279 | struct extent_buffer *buf; |
8280 | ||
2ff7e61e | 8281 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8282 | if (IS_ERR(buf)) |
8283 | return buf; | |
8284 | ||
65b51a00 | 8285 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8286 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8287 | btrfs_tree_lock(buf); |
7c302b49 | 8288 | clean_tree_block(fs_info, buf); |
3083ee2e | 8289 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8290 | |
8291 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8292 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8293 | |
d0c803c4 | 8294 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8295 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8296 | /* |
8297 | * we allow two log transactions at a time, use different | |
8298 | * EXENT bit to differentiate dirty pages. | |
8299 | */ | |
656f30db | 8300 | if (buf->log_index == 0) |
8cef4e16 YZ |
8301 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8302 | buf->start + buf->len - 1, GFP_NOFS); | |
8303 | else | |
8304 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8305 | buf->start + buf->len - 1); |
d0c803c4 | 8306 | } else { |
656f30db | 8307 | buf->log_index = -1; |
d0c803c4 | 8308 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8309 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8310 | } |
64c12921 | 8311 | trans->dirty = true; |
b4ce94de | 8312 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8313 | return buf; |
8314 | } | |
8315 | ||
f0486c68 YZ |
8316 | static struct btrfs_block_rsv * |
8317 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8318 | struct btrfs_root *root, u32 blocksize) | |
8319 | { | |
0b246afa | 8320 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8321 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8322 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8323 | int ret; |
d88033db | 8324 | bool global_updated = false; |
f0486c68 YZ |
8325 | |
8326 | block_rsv = get_block_rsv(trans, root); | |
8327 | ||
b586b323 MX |
8328 | if (unlikely(block_rsv->size == 0)) |
8329 | goto try_reserve; | |
d88033db | 8330 | again: |
f0486c68 YZ |
8331 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8332 | if (!ret) | |
8333 | return block_rsv; | |
8334 | ||
b586b323 MX |
8335 | if (block_rsv->failfast) |
8336 | return ERR_PTR(ret); | |
8337 | ||
d88033db MX |
8338 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8339 | global_updated = true; | |
0b246afa | 8340 | update_global_block_rsv(fs_info); |
d88033db MX |
8341 | goto again; |
8342 | } | |
8343 | ||
0b246afa | 8344 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8345 | static DEFINE_RATELIMIT_STATE(_rs, |
8346 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8347 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8348 | if (__ratelimit(&_rs)) | |
8349 | WARN(1, KERN_DEBUG | |
efe120a0 | 8350 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8351 | } |
8352 | try_reserve: | |
8353 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8354 | BTRFS_RESERVE_NO_FLUSH); | |
8355 | if (!ret) | |
8356 | return block_rsv; | |
8357 | /* | |
8358 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8359 | * the global reserve if its space type is the same as the global |
8360 | * reservation. | |
b586b323 | 8361 | */ |
5881cfc9 MX |
8362 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8363 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8364 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8365 | if (!ret) | |
8366 | return global_rsv; | |
8367 | } | |
8368 | return ERR_PTR(ret); | |
f0486c68 YZ |
8369 | } |
8370 | ||
8c2a3ca2 JB |
8371 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8372 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8373 | { |
8374 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 8375 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
8376 | } |
8377 | ||
fec577fb | 8378 | /* |
f0486c68 | 8379 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8380 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8381 | */ |
4d75f8a9 | 8382 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8383 | struct btrfs_root *root, |
8384 | u64 parent, u64 root_objectid, | |
8385 | const struct btrfs_disk_key *key, | |
8386 | int level, u64 hint, | |
8387 | u64 empty_size) | |
fec577fb | 8388 | { |
0b246afa | 8389 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8390 | struct btrfs_key ins; |
f0486c68 | 8391 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8392 | struct extent_buffer *buf; |
67b7859e | 8393 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8394 | u64 flags = 0; |
8395 | int ret; | |
0b246afa JM |
8396 | u32 blocksize = fs_info->nodesize; |
8397 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8398 | |
05653ef3 | 8399 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8400 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8401 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8402 | level); |
faa2dbf0 JB |
8403 | if (!IS_ERR(buf)) |
8404 | root->alloc_bytenr += blocksize; | |
8405 | return buf; | |
8406 | } | |
05653ef3 | 8407 | #endif |
fccb84c9 | 8408 | |
f0486c68 YZ |
8409 | block_rsv = use_block_rsv(trans, root, blocksize); |
8410 | if (IS_ERR(block_rsv)) | |
8411 | return ERR_CAST(block_rsv); | |
8412 | ||
18513091 | 8413 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8414 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8415 | if (ret) |
8416 | goto out_unuse; | |
55c69072 | 8417 | |
fe864576 | 8418 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8419 | if (IS_ERR(buf)) { |
8420 | ret = PTR_ERR(buf); | |
8421 | goto out_free_reserved; | |
8422 | } | |
f0486c68 YZ |
8423 | |
8424 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8425 | if (parent == 0) | |
8426 | parent = ins.objectid; | |
8427 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8428 | } else | |
8429 | BUG_ON(parent > 0); | |
8430 | ||
8431 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8432 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8433 | if (!extent_op) { |
8434 | ret = -ENOMEM; | |
8435 | goto out_free_buf; | |
8436 | } | |
f0486c68 YZ |
8437 | if (key) |
8438 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8439 | else | |
8440 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8441 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8442 | extent_op->update_key = skinny_metadata ? false : true; |
8443 | extent_op->update_flags = true; | |
8444 | extent_op->is_data = false; | |
b1c79e09 | 8445 | extent_op->level = level; |
f0486c68 | 8446 | |
0b246afa | 8447 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, |
67b7859e OS |
8448 | ins.objectid, ins.offset, |
8449 | parent, root_objectid, level, | |
8450 | BTRFS_ADD_DELAYED_EXTENT, | |
b06c4bf5 | 8451 | extent_op); |
67b7859e OS |
8452 | if (ret) |
8453 | goto out_free_delayed; | |
f0486c68 | 8454 | } |
fec577fb | 8455 | return buf; |
67b7859e OS |
8456 | |
8457 | out_free_delayed: | |
8458 | btrfs_free_delayed_extent_op(extent_op); | |
8459 | out_free_buf: | |
8460 | free_extent_buffer(buf); | |
8461 | out_free_reserved: | |
2ff7e61e | 8462 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8463 | out_unuse: |
0b246afa | 8464 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8465 | return ERR_PTR(ret); |
fec577fb | 8466 | } |
a28ec197 | 8467 | |
2c47e605 YZ |
8468 | struct walk_control { |
8469 | u64 refs[BTRFS_MAX_LEVEL]; | |
8470 | u64 flags[BTRFS_MAX_LEVEL]; | |
8471 | struct btrfs_key update_progress; | |
8472 | int stage; | |
8473 | int level; | |
8474 | int shared_level; | |
8475 | int update_ref; | |
8476 | int keep_locks; | |
1c4850e2 YZ |
8477 | int reada_slot; |
8478 | int reada_count; | |
66d7e7f0 | 8479 | int for_reloc; |
2c47e605 YZ |
8480 | }; |
8481 | ||
8482 | #define DROP_REFERENCE 1 | |
8483 | #define UPDATE_BACKREF 2 | |
8484 | ||
1c4850e2 YZ |
8485 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8486 | struct btrfs_root *root, | |
8487 | struct walk_control *wc, | |
8488 | struct btrfs_path *path) | |
6407bf6d | 8489 | { |
0b246afa | 8490 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8491 | u64 bytenr; |
8492 | u64 generation; | |
8493 | u64 refs; | |
94fcca9f | 8494 | u64 flags; |
5d4f98a2 | 8495 | u32 nritems; |
1c4850e2 YZ |
8496 | struct btrfs_key key; |
8497 | struct extent_buffer *eb; | |
6407bf6d | 8498 | int ret; |
1c4850e2 YZ |
8499 | int slot; |
8500 | int nread = 0; | |
6407bf6d | 8501 | |
1c4850e2 YZ |
8502 | if (path->slots[wc->level] < wc->reada_slot) { |
8503 | wc->reada_count = wc->reada_count * 2 / 3; | |
8504 | wc->reada_count = max(wc->reada_count, 2); | |
8505 | } else { | |
8506 | wc->reada_count = wc->reada_count * 3 / 2; | |
8507 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8508 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8509 | } |
7bb86316 | 8510 | |
1c4850e2 YZ |
8511 | eb = path->nodes[wc->level]; |
8512 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8513 | |
1c4850e2 YZ |
8514 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8515 | if (nread >= wc->reada_count) | |
8516 | break; | |
bd56b302 | 8517 | |
2dd3e67b | 8518 | cond_resched(); |
1c4850e2 YZ |
8519 | bytenr = btrfs_node_blockptr(eb, slot); |
8520 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8521 | |
1c4850e2 YZ |
8522 | if (slot == path->slots[wc->level]) |
8523 | goto reada; | |
5d4f98a2 | 8524 | |
1c4850e2 YZ |
8525 | if (wc->stage == UPDATE_BACKREF && |
8526 | generation <= root->root_key.offset) | |
bd56b302 CM |
8527 | continue; |
8528 | ||
94fcca9f | 8529 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8530 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8531 | wc->level - 1, 1, &refs, |
8532 | &flags); | |
79787eaa JM |
8533 | /* We don't care about errors in readahead. */ |
8534 | if (ret < 0) | |
8535 | continue; | |
94fcca9f YZ |
8536 | BUG_ON(refs == 0); |
8537 | ||
1c4850e2 | 8538 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8539 | if (refs == 1) |
8540 | goto reada; | |
bd56b302 | 8541 | |
94fcca9f YZ |
8542 | if (wc->level == 1 && |
8543 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8544 | continue; | |
1c4850e2 YZ |
8545 | if (!wc->update_ref || |
8546 | generation <= root->root_key.offset) | |
8547 | continue; | |
8548 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8549 | ret = btrfs_comp_cpu_keys(&key, | |
8550 | &wc->update_progress); | |
8551 | if (ret < 0) | |
8552 | continue; | |
94fcca9f YZ |
8553 | } else { |
8554 | if (wc->level == 1 && | |
8555 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8556 | continue; | |
6407bf6d | 8557 | } |
1c4850e2 | 8558 | reada: |
2ff7e61e | 8559 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8560 | nread++; |
20524f02 | 8561 | } |
1c4850e2 | 8562 | wc->reada_slot = slot; |
20524f02 | 8563 | } |
2c47e605 | 8564 | |
f82d02d9 | 8565 | /* |
2c016dc2 | 8566 | * helper to process tree block while walking down the tree. |
2c47e605 | 8567 | * |
2c47e605 YZ |
8568 | * when wc->stage == UPDATE_BACKREF, this function updates |
8569 | * back refs for pointers in the block. | |
8570 | * | |
8571 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8572 | */ |
2c47e605 | 8573 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8574 | struct btrfs_root *root, |
2c47e605 | 8575 | struct btrfs_path *path, |
94fcca9f | 8576 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8577 | { |
2ff7e61e | 8578 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8579 | int level = wc->level; |
8580 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8581 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8582 | int ret; |
8583 | ||
2c47e605 YZ |
8584 | if (wc->stage == UPDATE_BACKREF && |
8585 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8586 | return 1; | |
f82d02d9 | 8587 | |
2c47e605 YZ |
8588 | /* |
8589 | * when reference count of tree block is 1, it won't increase | |
8590 | * again. once full backref flag is set, we never clear it. | |
8591 | */ | |
94fcca9f YZ |
8592 | if (lookup_info && |
8593 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8594 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8595 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8596 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8597 | eb->start, level, 1, |
2c47e605 YZ |
8598 | &wc->refs[level], |
8599 | &wc->flags[level]); | |
79787eaa JM |
8600 | BUG_ON(ret == -ENOMEM); |
8601 | if (ret) | |
8602 | return ret; | |
2c47e605 YZ |
8603 | BUG_ON(wc->refs[level] == 0); |
8604 | } | |
5d4f98a2 | 8605 | |
2c47e605 YZ |
8606 | if (wc->stage == DROP_REFERENCE) { |
8607 | if (wc->refs[level] > 1) | |
8608 | return 1; | |
f82d02d9 | 8609 | |
2c47e605 | 8610 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8611 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8612 | path->locks[level] = 0; |
8613 | } | |
8614 | return 0; | |
8615 | } | |
f82d02d9 | 8616 | |
2c47e605 YZ |
8617 | /* wc->stage == UPDATE_BACKREF */ |
8618 | if (!(wc->flags[level] & flag)) { | |
8619 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8620 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8621 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8622 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8623 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8624 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8625 | eb->len, flag, |
8626 | btrfs_header_level(eb), 0); | |
79787eaa | 8627 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8628 | wc->flags[level] |= flag; |
8629 | } | |
8630 | ||
8631 | /* | |
8632 | * the block is shared by multiple trees, so it's not good to | |
8633 | * keep the tree lock | |
8634 | */ | |
8635 | if (path->locks[level] && level > 0) { | |
bd681513 | 8636 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8637 | path->locks[level] = 0; |
8638 | } | |
8639 | return 0; | |
8640 | } | |
8641 | ||
1c4850e2 | 8642 | /* |
2c016dc2 | 8643 | * helper to process tree block pointer. |
1c4850e2 YZ |
8644 | * |
8645 | * when wc->stage == DROP_REFERENCE, this function checks | |
8646 | * reference count of the block pointed to. if the block | |
8647 | * is shared and we need update back refs for the subtree | |
8648 | * rooted at the block, this function changes wc->stage to | |
8649 | * UPDATE_BACKREF. if the block is shared and there is no | |
8650 | * need to update back, this function drops the reference | |
8651 | * to the block. | |
8652 | * | |
8653 | * NOTE: return value 1 means we should stop walking down. | |
8654 | */ | |
8655 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8656 | struct btrfs_root *root, | |
8657 | struct btrfs_path *path, | |
94fcca9f | 8658 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8659 | { |
0b246afa | 8660 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8661 | u64 bytenr; |
8662 | u64 generation; | |
8663 | u64 parent; | |
8664 | u32 blocksize; | |
8665 | struct btrfs_key key; | |
8666 | struct extent_buffer *next; | |
8667 | int level = wc->level; | |
8668 | int reada = 0; | |
8669 | int ret = 0; | |
1152651a | 8670 | bool need_account = false; |
1c4850e2 YZ |
8671 | |
8672 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8673 | path->slots[level]); | |
8674 | /* | |
8675 | * if the lower level block was created before the snapshot | |
8676 | * was created, we know there is no need to update back refs | |
8677 | * for the subtree | |
8678 | */ | |
8679 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8680 | generation <= root->root_key.offset) { |
8681 | *lookup_info = 1; | |
1c4850e2 | 8682 | return 1; |
94fcca9f | 8683 | } |
1c4850e2 YZ |
8684 | |
8685 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
0b246afa | 8686 | blocksize = fs_info->nodesize; |
1c4850e2 | 8687 | |
0b246afa | 8688 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8689 | if (!next) { |
2ff7e61e | 8690 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8691 | if (IS_ERR(next)) |
8692 | return PTR_ERR(next); | |
8693 | ||
b2aaaa3b JB |
8694 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8695 | level - 1); | |
1c4850e2 YZ |
8696 | reada = 1; |
8697 | } | |
8698 | btrfs_tree_lock(next); | |
8699 | btrfs_set_lock_blocking(next); | |
8700 | ||
2ff7e61e | 8701 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8702 | &wc->refs[level - 1], |
8703 | &wc->flags[level - 1]); | |
4867268c JB |
8704 | if (ret < 0) |
8705 | goto out_unlock; | |
79787eaa | 8706 | |
c2cf52eb | 8707 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8708 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8709 | ret = -EIO; |
8710 | goto out_unlock; | |
c2cf52eb | 8711 | } |
94fcca9f | 8712 | *lookup_info = 0; |
1c4850e2 | 8713 | |
94fcca9f | 8714 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8715 | if (wc->refs[level - 1] > 1) { |
1152651a | 8716 | need_account = true; |
94fcca9f YZ |
8717 | if (level == 1 && |
8718 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8719 | goto skip; | |
8720 | ||
1c4850e2 YZ |
8721 | if (!wc->update_ref || |
8722 | generation <= root->root_key.offset) | |
8723 | goto skip; | |
8724 | ||
8725 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8726 | path->slots[level]); | |
8727 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8728 | if (ret < 0) | |
8729 | goto skip; | |
8730 | ||
8731 | wc->stage = UPDATE_BACKREF; | |
8732 | wc->shared_level = level - 1; | |
8733 | } | |
94fcca9f YZ |
8734 | } else { |
8735 | if (level == 1 && | |
8736 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8737 | goto skip; | |
1c4850e2 YZ |
8738 | } |
8739 | ||
b9fab919 | 8740 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8741 | btrfs_tree_unlock(next); |
8742 | free_extent_buffer(next); | |
8743 | next = NULL; | |
94fcca9f | 8744 | *lookup_info = 1; |
1c4850e2 YZ |
8745 | } |
8746 | ||
8747 | if (!next) { | |
8748 | if (reada && level == 1) | |
8749 | reada_walk_down(trans, root, wc, path); | |
2ff7e61e | 8750 | next = read_tree_block(fs_info, bytenr, generation); |
64c043de LB |
8751 | if (IS_ERR(next)) { |
8752 | return PTR_ERR(next); | |
8753 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8754 | free_extent_buffer(next); |
97d9a8a4 | 8755 | return -EIO; |
416bc658 | 8756 | } |
1c4850e2 YZ |
8757 | btrfs_tree_lock(next); |
8758 | btrfs_set_lock_blocking(next); | |
8759 | } | |
8760 | ||
8761 | level--; | |
4867268c JB |
8762 | ASSERT(level == btrfs_header_level(next)); |
8763 | if (level != btrfs_header_level(next)) { | |
8764 | btrfs_err(root->fs_info, "mismatched level"); | |
8765 | ret = -EIO; | |
8766 | goto out_unlock; | |
8767 | } | |
1c4850e2 YZ |
8768 | path->nodes[level] = next; |
8769 | path->slots[level] = 0; | |
bd681513 | 8770 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8771 | wc->level = level; |
8772 | if (wc->level == 1) | |
8773 | wc->reada_slot = 0; | |
8774 | return 0; | |
8775 | skip: | |
8776 | wc->refs[level - 1] = 0; | |
8777 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8778 | if (wc->stage == DROP_REFERENCE) { |
8779 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8780 | parent = path->nodes[level]->start; | |
8781 | } else { | |
4867268c | 8782 | ASSERT(root->root_key.objectid == |
94fcca9f | 8783 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8784 | if (root->root_key.objectid != |
8785 | btrfs_header_owner(path->nodes[level])) { | |
8786 | btrfs_err(root->fs_info, | |
8787 | "mismatched block owner"); | |
8788 | ret = -EIO; | |
8789 | goto out_unlock; | |
8790 | } | |
94fcca9f YZ |
8791 | parent = 0; |
8792 | } | |
1c4850e2 | 8793 | |
1152651a | 8794 | if (need_account) { |
33d1f05c QW |
8795 | ret = btrfs_qgroup_trace_subtree(trans, root, next, |
8796 | generation, level - 1); | |
1152651a | 8797 | if (ret) { |
0b246afa | 8798 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8799 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8800 | ret); | |
1152651a MF |
8801 | } |
8802 | } | |
2ff7e61e JM |
8803 | ret = btrfs_free_extent(trans, fs_info, bytenr, blocksize, |
8804 | parent, root->root_key.objectid, | |
8805 | level - 1, 0); | |
4867268c JB |
8806 | if (ret) |
8807 | goto out_unlock; | |
1c4850e2 | 8808 | } |
4867268c JB |
8809 | |
8810 | *lookup_info = 1; | |
8811 | ret = 1; | |
8812 | ||
8813 | out_unlock: | |
1c4850e2 YZ |
8814 | btrfs_tree_unlock(next); |
8815 | free_extent_buffer(next); | |
4867268c JB |
8816 | |
8817 | return ret; | |
1c4850e2 YZ |
8818 | } |
8819 | ||
2c47e605 | 8820 | /* |
2c016dc2 | 8821 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8822 | * |
8823 | * when wc->stage == DROP_REFERENCE, this function drops | |
8824 | * reference count on the block. | |
8825 | * | |
8826 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8827 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8828 | * to UPDATE_BACKREF previously while processing the block. | |
8829 | * | |
8830 | * NOTE: return value 1 means we should stop walking up. | |
8831 | */ | |
8832 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8833 | struct btrfs_root *root, | |
8834 | struct btrfs_path *path, | |
8835 | struct walk_control *wc) | |
8836 | { | |
0b246afa | 8837 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8838 | int ret; |
2c47e605 YZ |
8839 | int level = wc->level; |
8840 | struct extent_buffer *eb = path->nodes[level]; | |
8841 | u64 parent = 0; | |
8842 | ||
8843 | if (wc->stage == UPDATE_BACKREF) { | |
8844 | BUG_ON(wc->shared_level < level); | |
8845 | if (level < wc->shared_level) | |
8846 | goto out; | |
8847 | ||
2c47e605 YZ |
8848 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8849 | if (ret > 0) | |
8850 | wc->update_ref = 0; | |
8851 | ||
8852 | wc->stage = DROP_REFERENCE; | |
8853 | wc->shared_level = -1; | |
8854 | path->slots[level] = 0; | |
8855 | ||
8856 | /* | |
8857 | * check reference count again if the block isn't locked. | |
8858 | * we should start walking down the tree again if reference | |
8859 | * count is one. | |
8860 | */ | |
8861 | if (!path->locks[level]) { | |
8862 | BUG_ON(level == 0); | |
8863 | btrfs_tree_lock(eb); | |
8864 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8865 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8866 | |
2ff7e61e | 8867 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8868 | eb->start, level, 1, |
2c47e605 YZ |
8869 | &wc->refs[level], |
8870 | &wc->flags[level]); | |
79787eaa JM |
8871 | if (ret < 0) { |
8872 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8873 | path->locks[level] = 0; |
79787eaa JM |
8874 | return ret; |
8875 | } | |
2c47e605 YZ |
8876 | BUG_ON(wc->refs[level] == 0); |
8877 | if (wc->refs[level] == 1) { | |
bd681513 | 8878 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8879 | path->locks[level] = 0; |
2c47e605 YZ |
8880 | return 1; |
8881 | } | |
f82d02d9 | 8882 | } |
2c47e605 | 8883 | } |
f82d02d9 | 8884 | |
2c47e605 YZ |
8885 | /* wc->stage == DROP_REFERENCE */ |
8886 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8887 | |
2c47e605 YZ |
8888 | if (wc->refs[level] == 1) { |
8889 | if (level == 0) { | |
8890 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8891 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8892 | else |
e339a6b0 | 8893 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8894 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8895 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb); |
1152651a | 8896 | if (ret) { |
0b246afa | 8897 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8898 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
8899 | ret); | |
1152651a | 8900 | } |
2c47e605 YZ |
8901 | } |
8902 | /* make block locked assertion in clean_tree_block happy */ | |
8903 | if (!path->locks[level] && | |
8904 | btrfs_header_generation(eb) == trans->transid) { | |
8905 | btrfs_tree_lock(eb); | |
8906 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8907 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8908 | } |
7c302b49 | 8909 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
8910 | } |
8911 | ||
8912 | if (eb == root->node) { | |
8913 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8914 | parent = eb->start; | |
8915 | else | |
8916 | BUG_ON(root->root_key.objectid != | |
8917 | btrfs_header_owner(eb)); | |
8918 | } else { | |
8919 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8920 | parent = path->nodes[level + 1]->start; | |
8921 | else | |
8922 | BUG_ON(root->root_key.objectid != | |
8923 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8924 | } |
f82d02d9 | 8925 | |
5581a51a | 8926 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8927 | out: |
8928 | wc->refs[level] = 0; | |
8929 | wc->flags[level] = 0; | |
f0486c68 | 8930 | return 0; |
2c47e605 YZ |
8931 | } |
8932 | ||
8933 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8934 | struct btrfs_root *root, | |
8935 | struct btrfs_path *path, | |
8936 | struct walk_control *wc) | |
8937 | { | |
2c47e605 | 8938 | int level = wc->level; |
94fcca9f | 8939 | int lookup_info = 1; |
2c47e605 YZ |
8940 | int ret; |
8941 | ||
8942 | while (level >= 0) { | |
94fcca9f | 8943 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8944 | if (ret > 0) |
8945 | break; | |
8946 | ||
8947 | if (level == 0) | |
8948 | break; | |
8949 | ||
7a7965f8 YZ |
8950 | if (path->slots[level] >= |
8951 | btrfs_header_nritems(path->nodes[level])) | |
8952 | break; | |
8953 | ||
94fcca9f | 8954 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8955 | if (ret > 0) { |
8956 | path->slots[level]++; | |
8957 | continue; | |
90d2c51d MX |
8958 | } else if (ret < 0) |
8959 | return ret; | |
1c4850e2 | 8960 | level = wc->level; |
f82d02d9 | 8961 | } |
f82d02d9 YZ |
8962 | return 0; |
8963 | } | |
8964 | ||
d397712b | 8965 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 8966 | struct btrfs_root *root, |
f82d02d9 | 8967 | struct btrfs_path *path, |
2c47e605 | 8968 | struct walk_control *wc, int max_level) |
20524f02 | 8969 | { |
2c47e605 | 8970 | int level = wc->level; |
20524f02 | 8971 | int ret; |
9f3a7427 | 8972 | |
2c47e605 YZ |
8973 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
8974 | while (level < max_level && path->nodes[level]) { | |
8975 | wc->level = level; | |
8976 | if (path->slots[level] + 1 < | |
8977 | btrfs_header_nritems(path->nodes[level])) { | |
8978 | path->slots[level]++; | |
20524f02 CM |
8979 | return 0; |
8980 | } else { | |
2c47e605 YZ |
8981 | ret = walk_up_proc(trans, root, path, wc); |
8982 | if (ret > 0) | |
8983 | return 0; | |
bd56b302 | 8984 | |
2c47e605 | 8985 | if (path->locks[level]) { |
bd681513 CM |
8986 | btrfs_tree_unlock_rw(path->nodes[level], |
8987 | path->locks[level]); | |
2c47e605 | 8988 | path->locks[level] = 0; |
f82d02d9 | 8989 | } |
2c47e605 YZ |
8990 | free_extent_buffer(path->nodes[level]); |
8991 | path->nodes[level] = NULL; | |
8992 | level++; | |
20524f02 CM |
8993 | } |
8994 | } | |
8995 | return 1; | |
8996 | } | |
8997 | ||
9aca1d51 | 8998 | /* |
2c47e605 YZ |
8999 | * drop a subvolume tree. |
9000 | * | |
9001 | * this function traverses the tree freeing any blocks that only | |
9002 | * referenced by the tree. | |
9003 | * | |
9004 | * when a shared tree block is found. this function decreases its | |
9005 | * reference count by one. if update_ref is true, this function | |
9006 | * also make sure backrefs for the shared block and all lower level | |
9007 | * blocks are properly updated. | |
9d1a2a3a DS |
9008 | * |
9009 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9010 | */ |
2c536799 | 9011 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9012 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9013 | int for_reloc) | |
20524f02 | 9014 | { |
ab8d0fc4 | 9015 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9016 | struct btrfs_path *path; |
2c47e605 | 9017 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9018 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9019 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9020 | struct walk_control *wc; |
9021 | struct btrfs_key key; | |
9022 | int err = 0; | |
9023 | int ret; | |
9024 | int level; | |
d29a9f62 | 9025 | bool root_dropped = false; |
20524f02 | 9026 | |
ab8d0fc4 | 9027 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
1152651a | 9028 | |
5caf2a00 | 9029 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9030 | if (!path) { |
9031 | err = -ENOMEM; | |
9032 | goto out; | |
9033 | } | |
20524f02 | 9034 | |
2c47e605 | 9035 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9036 | if (!wc) { |
9037 | btrfs_free_path(path); | |
cb1b69f4 TI |
9038 | err = -ENOMEM; |
9039 | goto out; | |
38a1a919 | 9040 | } |
2c47e605 | 9041 | |
a22285a6 | 9042 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9043 | if (IS_ERR(trans)) { |
9044 | err = PTR_ERR(trans); | |
9045 | goto out_free; | |
9046 | } | |
98d5dc13 | 9047 | |
3fd0a558 YZ |
9048 | if (block_rsv) |
9049 | trans->block_rsv = block_rsv; | |
2c47e605 | 9050 | |
9f3a7427 | 9051 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9052 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9053 | path->nodes[level] = btrfs_lock_root_node(root); |
9054 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9055 | path->slots[level] = 0; |
bd681513 | 9056 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9057 | memset(&wc->update_progress, 0, |
9058 | sizeof(wc->update_progress)); | |
9f3a7427 | 9059 | } else { |
9f3a7427 | 9060 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9061 | memcpy(&wc->update_progress, &key, |
9062 | sizeof(wc->update_progress)); | |
9063 | ||
6702ed49 | 9064 | level = root_item->drop_level; |
2c47e605 | 9065 | BUG_ON(level == 0); |
6702ed49 | 9066 | path->lowest_level = level; |
2c47e605 YZ |
9067 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9068 | path->lowest_level = 0; | |
9069 | if (ret < 0) { | |
9070 | err = ret; | |
79787eaa | 9071 | goto out_end_trans; |
9f3a7427 | 9072 | } |
1c4850e2 | 9073 | WARN_ON(ret > 0); |
2c47e605 | 9074 | |
7d9eb12c CM |
9075 | /* |
9076 | * unlock our path, this is safe because only this | |
9077 | * function is allowed to delete this snapshot | |
9078 | */ | |
5d4f98a2 | 9079 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9080 | |
9081 | level = btrfs_header_level(root->node); | |
9082 | while (1) { | |
9083 | btrfs_tree_lock(path->nodes[level]); | |
9084 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9085 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9086 | |
2ff7e61e | 9087 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9088 | path->nodes[level]->start, |
3173a18f | 9089 | level, 1, &wc->refs[level], |
2c47e605 | 9090 | &wc->flags[level]); |
79787eaa JM |
9091 | if (ret < 0) { |
9092 | err = ret; | |
9093 | goto out_end_trans; | |
9094 | } | |
2c47e605 YZ |
9095 | BUG_ON(wc->refs[level] == 0); |
9096 | ||
9097 | if (level == root_item->drop_level) | |
9098 | break; | |
9099 | ||
9100 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9101 | path->locks[level] = 0; |
2c47e605 YZ |
9102 | WARN_ON(wc->refs[level] != 1); |
9103 | level--; | |
9104 | } | |
9f3a7427 | 9105 | } |
2c47e605 YZ |
9106 | |
9107 | wc->level = level; | |
9108 | wc->shared_level = -1; | |
9109 | wc->stage = DROP_REFERENCE; | |
9110 | wc->update_ref = update_ref; | |
9111 | wc->keep_locks = 0; | |
66d7e7f0 | 9112 | wc->for_reloc = for_reloc; |
0b246afa | 9113 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9114 | |
d397712b | 9115 | while (1) { |
9d1a2a3a | 9116 | |
2c47e605 YZ |
9117 | ret = walk_down_tree(trans, root, path, wc); |
9118 | if (ret < 0) { | |
9119 | err = ret; | |
20524f02 | 9120 | break; |
2c47e605 | 9121 | } |
9aca1d51 | 9122 | |
2c47e605 YZ |
9123 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9124 | if (ret < 0) { | |
9125 | err = ret; | |
20524f02 | 9126 | break; |
2c47e605 YZ |
9127 | } |
9128 | ||
9129 | if (ret > 0) { | |
9130 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9131 | break; |
9132 | } | |
2c47e605 YZ |
9133 | |
9134 | if (wc->stage == DROP_REFERENCE) { | |
9135 | level = wc->level; | |
9136 | btrfs_node_key(path->nodes[level], | |
9137 | &root_item->drop_progress, | |
9138 | path->slots[level]); | |
9139 | root_item->drop_level = level; | |
9140 | } | |
9141 | ||
9142 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9143 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9144 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9145 | ret = btrfs_update_root(trans, tree_root, |
9146 | &root->root_key, | |
9147 | root_item); | |
79787eaa | 9148 | if (ret) { |
66642832 | 9149 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9150 | err = ret; |
9151 | goto out_end_trans; | |
9152 | } | |
2c47e605 | 9153 | |
3a45bb20 | 9154 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9155 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9156 | btrfs_debug(fs_info, |
9157 | "drop snapshot early exit"); | |
3c8f2422 JB |
9158 | err = -EAGAIN; |
9159 | goto out_free; | |
9160 | } | |
9161 | ||
a22285a6 | 9162 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9163 | if (IS_ERR(trans)) { |
9164 | err = PTR_ERR(trans); | |
9165 | goto out_free; | |
9166 | } | |
3fd0a558 YZ |
9167 | if (block_rsv) |
9168 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9169 | } |
20524f02 | 9170 | } |
b3b4aa74 | 9171 | btrfs_release_path(path); |
79787eaa JM |
9172 | if (err) |
9173 | goto out_end_trans; | |
2c47e605 YZ |
9174 | |
9175 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa | 9176 | if (ret) { |
66642832 | 9177 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9178 | goto out_end_trans; |
9179 | } | |
2c47e605 | 9180 | |
76dda93c | 9181 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9182 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9183 | NULL, NULL); | |
79787eaa | 9184 | if (ret < 0) { |
66642832 | 9185 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9186 | err = ret; |
9187 | goto out_end_trans; | |
9188 | } else if (ret > 0) { | |
84cd948c JB |
9189 | /* if we fail to delete the orphan item this time |
9190 | * around, it'll get picked up the next time. | |
9191 | * | |
9192 | * The most common failure here is just -ENOENT. | |
9193 | */ | |
9194 | btrfs_del_orphan_item(trans, tree_root, | |
9195 | root->root_key.objectid); | |
76dda93c YZ |
9196 | } |
9197 | } | |
9198 | ||
27cdeb70 | 9199 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9200 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9201 | } else { |
9202 | free_extent_buffer(root->node); | |
9203 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9204 | btrfs_put_fs_root(root); |
76dda93c | 9205 | } |
d29a9f62 | 9206 | root_dropped = true; |
79787eaa | 9207 | out_end_trans: |
3a45bb20 | 9208 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9209 | out_free: |
2c47e605 | 9210 | kfree(wc); |
5caf2a00 | 9211 | btrfs_free_path(path); |
cb1b69f4 | 9212 | out: |
d29a9f62 JB |
9213 | /* |
9214 | * So if we need to stop dropping the snapshot for whatever reason we | |
9215 | * need to make sure to add it back to the dead root list so that we | |
9216 | * keep trying to do the work later. This also cleans up roots if we | |
9217 | * don't have it in the radix (like when we recover after a power fail | |
9218 | * or unmount) so we don't leak memory. | |
9219 | */ | |
b37b39cd | 9220 | if (!for_reloc && root_dropped == false) |
d29a9f62 | 9221 | btrfs_add_dead_root(root); |
90515e7f | 9222 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9223 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9224 | return err; |
20524f02 | 9225 | } |
9078a3e1 | 9226 | |
2c47e605 YZ |
9227 | /* |
9228 | * drop subtree rooted at tree block 'node'. | |
9229 | * | |
9230 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9231 | * only used by relocation code |
2c47e605 | 9232 | */ |
f82d02d9 YZ |
9233 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9234 | struct btrfs_root *root, | |
9235 | struct extent_buffer *node, | |
9236 | struct extent_buffer *parent) | |
9237 | { | |
0b246afa | 9238 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9239 | struct btrfs_path *path; |
2c47e605 | 9240 | struct walk_control *wc; |
f82d02d9 YZ |
9241 | int level; |
9242 | int parent_level; | |
9243 | int ret = 0; | |
9244 | int wret; | |
9245 | ||
2c47e605 YZ |
9246 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9247 | ||
f82d02d9 | 9248 | path = btrfs_alloc_path(); |
db5b493a TI |
9249 | if (!path) |
9250 | return -ENOMEM; | |
f82d02d9 | 9251 | |
2c47e605 | 9252 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9253 | if (!wc) { |
9254 | btrfs_free_path(path); | |
9255 | return -ENOMEM; | |
9256 | } | |
2c47e605 | 9257 | |
b9447ef8 | 9258 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9259 | parent_level = btrfs_header_level(parent); |
9260 | extent_buffer_get(parent); | |
9261 | path->nodes[parent_level] = parent; | |
9262 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9263 | ||
b9447ef8 | 9264 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9265 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9266 | path->nodes[level] = node; |
9267 | path->slots[level] = 0; | |
bd681513 | 9268 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9269 | |
9270 | wc->refs[parent_level] = 1; | |
9271 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9272 | wc->level = level; | |
9273 | wc->shared_level = -1; | |
9274 | wc->stage = DROP_REFERENCE; | |
9275 | wc->update_ref = 0; | |
9276 | wc->keep_locks = 1; | |
66d7e7f0 | 9277 | wc->for_reloc = 1; |
0b246afa | 9278 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9279 | |
9280 | while (1) { | |
2c47e605 YZ |
9281 | wret = walk_down_tree(trans, root, path, wc); |
9282 | if (wret < 0) { | |
f82d02d9 | 9283 | ret = wret; |
f82d02d9 | 9284 | break; |
2c47e605 | 9285 | } |
f82d02d9 | 9286 | |
2c47e605 | 9287 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9288 | if (wret < 0) |
9289 | ret = wret; | |
9290 | if (wret != 0) | |
9291 | break; | |
9292 | } | |
9293 | ||
2c47e605 | 9294 | kfree(wc); |
f82d02d9 YZ |
9295 | btrfs_free_path(path); |
9296 | return ret; | |
9297 | } | |
9298 | ||
6202df69 | 9299 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9300 | { |
9301 | u64 num_devices; | |
fc67c450 | 9302 | u64 stripped; |
e4d8ec0f | 9303 | |
fc67c450 ID |
9304 | /* |
9305 | * if restripe for this chunk_type is on pick target profile and | |
9306 | * return, otherwise do the usual balance | |
9307 | */ | |
6202df69 | 9308 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9309 | if (stripped) |
9310 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9311 | |
6202df69 | 9312 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9313 | |
fc67c450 | 9314 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9315 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9316 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9317 | ||
ec44a35c CM |
9318 | if (num_devices == 1) { |
9319 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9320 | stripped = flags & ~stripped; | |
9321 | ||
9322 | /* turn raid0 into single device chunks */ | |
9323 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9324 | return stripped; | |
9325 | ||
9326 | /* turn mirroring into duplication */ | |
9327 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9328 | BTRFS_BLOCK_GROUP_RAID10)) | |
9329 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9330 | } else { |
9331 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9332 | if (flags & stripped) |
9333 | return flags; | |
9334 | ||
9335 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9336 | stripped = flags & ~stripped; | |
9337 | ||
9338 | /* switch duplicated blocks with raid1 */ | |
9339 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9340 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9341 | ||
e3176ca2 | 9342 | /* this is drive concat, leave it alone */ |
ec44a35c | 9343 | } |
e3176ca2 | 9344 | |
ec44a35c CM |
9345 | return flags; |
9346 | } | |
9347 | ||
868f401a | 9348 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9349 | { |
f0486c68 YZ |
9350 | struct btrfs_space_info *sinfo = cache->space_info; |
9351 | u64 num_bytes; | |
199c36ea | 9352 | u64 min_allocable_bytes; |
f0486c68 | 9353 | int ret = -ENOSPC; |
0ef3e66b | 9354 | |
199c36ea MX |
9355 | /* |
9356 | * We need some metadata space and system metadata space for | |
9357 | * allocating chunks in some corner cases until we force to set | |
9358 | * it to be readonly. | |
9359 | */ | |
9360 | if ((sinfo->flags & | |
9361 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9362 | !force) | |
ee22184b | 9363 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9364 | else |
9365 | min_allocable_bytes = 0; | |
9366 | ||
f0486c68 YZ |
9367 | spin_lock(&sinfo->lock); |
9368 | spin_lock(&cache->lock); | |
61cfea9b W |
9369 | |
9370 | if (cache->ro) { | |
868f401a | 9371 | cache->ro++; |
61cfea9b W |
9372 | ret = 0; |
9373 | goto out; | |
9374 | } | |
9375 | ||
f0486c68 YZ |
9376 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9377 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9378 | ||
4136135b | 9379 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9380 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9381 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9382 | cache->ro++; |
633c0aad | 9383 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9384 | ret = 0; |
9385 | } | |
61cfea9b | 9386 | out: |
f0486c68 YZ |
9387 | spin_unlock(&cache->lock); |
9388 | spin_unlock(&sinfo->lock); | |
9389 | return ret; | |
9390 | } | |
7d9eb12c | 9391 | |
5e00f193 | 9392 | int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info, |
f0486c68 | 9393 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9394 | |
f0486c68 YZ |
9395 | { |
9396 | struct btrfs_trans_handle *trans; | |
9397 | u64 alloc_flags; | |
9398 | int ret; | |
7d9eb12c | 9399 | |
1bbc621e | 9400 | again: |
5e00f193 | 9401 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9402 | if (IS_ERR(trans)) |
9403 | return PTR_ERR(trans); | |
5d4f98a2 | 9404 | |
1bbc621e CM |
9405 | /* |
9406 | * we're not allowed to set block groups readonly after the dirty | |
9407 | * block groups cache has started writing. If it already started, | |
9408 | * back off and let this transaction commit | |
9409 | */ | |
0b246afa | 9410 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9411 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9412 | u64 transid = trans->transid; |
9413 | ||
0b246afa | 9414 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9415 | btrfs_end_transaction(trans); |
1bbc621e | 9416 | |
2ff7e61e | 9417 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9418 | if (ret) |
9419 | return ret; | |
9420 | goto again; | |
9421 | } | |
9422 | ||
153c35b6 CM |
9423 | /* |
9424 | * if we are changing raid levels, try to allocate a corresponding | |
9425 | * block group with the new raid level. | |
9426 | */ | |
0b246afa | 9427 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9428 | if (alloc_flags != cache->flags) { |
2ff7e61e | 9429 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, |
153c35b6 CM |
9430 | CHUNK_ALLOC_FORCE); |
9431 | /* | |
9432 | * ENOSPC is allowed here, we may have enough space | |
9433 | * already allocated at the new raid level to | |
9434 | * carry on | |
9435 | */ | |
9436 | if (ret == -ENOSPC) | |
9437 | ret = 0; | |
9438 | if (ret < 0) | |
9439 | goto out; | |
9440 | } | |
1bbc621e | 9441 | |
868f401a | 9442 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9443 | if (!ret) |
9444 | goto out; | |
2ff7e61e JM |
9445 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
9446 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, | |
0e4f8f88 | 9447 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9448 | if (ret < 0) |
9449 | goto out; | |
868f401a | 9450 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9451 | out: |
2f081088 | 9452 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9453 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9454 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 9455 | check_system_chunk(trans, fs_info, alloc_flags); |
34441361 | 9456 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9457 | } |
0b246afa | 9458 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9459 | |
3a45bb20 | 9460 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9461 | return ret; |
9462 | } | |
5d4f98a2 | 9463 | |
c87f08ca | 9464 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 9465 | struct btrfs_fs_info *fs_info, u64 type) |
c87f08ca | 9466 | { |
2ff7e61e JM |
9467 | u64 alloc_flags = get_alloc_profile(fs_info, type); |
9468 | ||
9469 | return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
9470 | } |
9471 | ||
6d07bcec MX |
9472 | /* |
9473 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9474 | * space_info. takes mirrors into account. |
6d07bcec | 9475 | */ |
633c0aad | 9476 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9477 | { |
9478 | struct btrfs_block_group_cache *block_group; | |
9479 | u64 free_bytes = 0; | |
9480 | int factor; | |
9481 | ||
01327610 | 9482 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9483 | if (list_empty(&sinfo->ro_bgs)) |
9484 | return 0; | |
9485 | ||
9486 | spin_lock(&sinfo->lock); | |
9487 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9488 | spin_lock(&block_group->lock); |
9489 | ||
9490 | if (!block_group->ro) { | |
9491 | spin_unlock(&block_group->lock); | |
9492 | continue; | |
9493 | } | |
9494 | ||
9495 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9496 | BTRFS_BLOCK_GROUP_RAID10 | | |
9497 | BTRFS_BLOCK_GROUP_DUP)) | |
9498 | factor = 2; | |
9499 | else | |
9500 | factor = 1; | |
9501 | ||
9502 | free_bytes += (block_group->key.offset - | |
9503 | btrfs_block_group_used(&block_group->item)) * | |
9504 | factor; | |
9505 | ||
9506 | spin_unlock(&block_group->lock); | |
9507 | } | |
6d07bcec MX |
9508 | spin_unlock(&sinfo->lock); |
9509 | ||
9510 | return free_bytes; | |
9511 | } | |
9512 | ||
2ff7e61e | 9513 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9514 | { |
f0486c68 YZ |
9515 | struct btrfs_space_info *sinfo = cache->space_info; |
9516 | u64 num_bytes; | |
9517 | ||
9518 | BUG_ON(!cache->ro); | |
9519 | ||
9520 | spin_lock(&sinfo->lock); | |
9521 | spin_lock(&cache->lock); | |
868f401a Z |
9522 | if (!--cache->ro) { |
9523 | num_bytes = cache->key.offset - cache->reserved - | |
9524 | cache->pinned - cache->bytes_super - | |
9525 | btrfs_block_group_used(&cache->item); | |
9526 | sinfo->bytes_readonly -= num_bytes; | |
9527 | list_del_init(&cache->ro_list); | |
9528 | } | |
f0486c68 YZ |
9529 | spin_unlock(&cache->lock); |
9530 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9531 | } |
9532 | ||
ba1bf481 JB |
9533 | /* |
9534 | * checks to see if its even possible to relocate this block group. | |
9535 | * | |
9536 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9537 | * ok to go ahead and try. | |
9538 | */ | |
6bccf3ab | 9539 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9540 | { |
6bccf3ab | 9541 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9542 | struct btrfs_block_group_cache *block_group; |
9543 | struct btrfs_space_info *space_info; | |
0b246afa | 9544 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9545 | struct btrfs_device *device; |
6df9a95e | 9546 | struct btrfs_trans_handle *trans; |
cdcb725c | 9547 | u64 min_free; |
6719db6a JB |
9548 | u64 dev_min = 1; |
9549 | u64 dev_nr = 0; | |
4a5e98f5 | 9550 | u64 target; |
0305bc27 | 9551 | int debug; |
cdcb725c | 9552 | int index; |
ba1bf481 JB |
9553 | int full = 0; |
9554 | int ret = 0; | |
1a40e23b | 9555 | |
0b246afa | 9556 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9557 | |
0b246afa | 9558 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9559 | |
ba1bf481 | 9560 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9561 | if (!block_group) { |
9562 | if (debug) | |
0b246afa | 9563 | btrfs_warn(fs_info, |
0305bc27 QW |
9564 | "can't find block group for bytenr %llu", |
9565 | bytenr); | |
ba1bf481 | 9566 | return -1; |
0305bc27 | 9567 | } |
1a40e23b | 9568 | |
cdcb725c | 9569 | min_free = btrfs_block_group_used(&block_group->item); |
9570 | ||
ba1bf481 | 9571 | /* no bytes used, we're good */ |
cdcb725c | 9572 | if (!min_free) |
1a40e23b ZY |
9573 | goto out; |
9574 | ||
ba1bf481 JB |
9575 | space_info = block_group->space_info; |
9576 | spin_lock(&space_info->lock); | |
17d217fe | 9577 | |
ba1bf481 | 9578 | full = space_info->full; |
17d217fe | 9579 | |
ba1bf481 JB |
9580 | /* |
9581 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9582 | * relocate it unless we're able to allocate a new chunk below. |
9583 | * | |
9584 | * Otherwise, we need to make sure we have room in the space to handle | |
9585 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9586 | */ |
7ce618db | 9587 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9588 | (btrfs_space_info_used(space_info, false) + min_free < |
9589 | space_info->total_bytes)) { | |
ba1bf481 JB |
9590 | spin_unlock(&space_info->lock); |
9591 | goto out; | |
17d217fe | 9592 | } |
ba1bf481 | 9593 | spin_unlock(&space_info->lock); |
ea8c2819 | 9594 | |
ba1bf481 JB |
9595 | /* |
9596 | * ok we don't have enough space, but maybe we have free space on our | |
9597 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9598 | * alloc devices and guess if we have enough space. if this block |
9599 | * group is going to be restriped, run checks against the target | |
9600 | * profile instead of the current one. | |
ba1bf481 JB |
9601 | */ |
9602 | ret = -1; | |
ea8c2819 | 9603 | |
cdcb725c | 9604 | /* |
9605 | * index: | |
9606 | * 0: raid10 | |
9607 | * 1: raid1 | |
9608 | * 2: dup | |
9609 | * 3: raid0 | |
9610 | * 4: single | |
9611 | */ | |
0b246afa | 9612 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9613 | if (target) { |
31e50229 | 9614 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9615 | } else { |
9616 | /* | |
9617 | * this is just a balance, so if we were marked as full | |
9618 | * we know there is no space for a new chunk | |
9619 | */ | |
0305bc27 QW |
9620 | if (full) { |
9621 | if (debug) | |
0b246afa JM |
9622 | btrfs_warn(fs_info, |
9623 | "no space to alloc new chunk for block group %llu", | |
9624 | block_group->key.objectid); | |
4a5e98f5 | 9625 | goto out; |
0305bc27 | 9626 | } |
4a5e98f5 ID |
9627 | |
9628 | index = get_block_group_index(block_group); | |
9629 | } | |
9630 | ||
e6ec716f | 9631 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9632 | dev_min = 4; |
6719db6a JB |
9633 | /* Divide by 2 */ |
9634 | min_free >>= 1; | |
e6ec716f | 9635 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9636 | dev_min = 2; |
e6ec716f | 9637 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9638 | /* Multiply by 2 */ |
9639 | min_free <<= 1; | |
e6ec716f | 9640 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9641 | dev_min = fs_devices->rw_devices; |
47c5713f | 9642 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9643 | } |
9644 | ||
6df9a95e JB |
9645 | /* We need to do this so that we can look at pending chunks */ |
9646 | trans = btrfs_join_transaction(root); | |
9647 | if (IS_ERR(trans)) { | |
9648 | ret = PTR_ERR(trans); | |
9649 | goto out; | |
9650 | } | |
9651 | ||
0b246afa | 9652 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9653 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9654 | u64 dev_offset; |
56bec294 | 9655 | |
ba1bf481 JB |
9656 | /* |
9657 | * check to make sure we can actually find a chunk with enough | |
9658 | * space to fit our block group in. | |
9659 | */ | |
63a212ab SB |
9660 | if (device->total_bytes > device->bytes_used + min_free && |
9661 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 9662 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9663 | &dev_offset, NULL); |
ba1bf481 | 9664 | if (!ret) |
cdcb725c | 9665 | dev_nr++; |
9666 | ||
9667 | if (dev_nr >= dev_min) | |
73e48b27 | 9668 | break; |
cdcb725c | 9669 | |
ba1bf481 | 9670 | ret = -1; |
725c8463 | 9671 | } |
edbd8d4e | 9672 | } |
0305bc27 | 9673 | if (debug && ret == -1) |
0b246afa JM |
9674 | btrfs_warn(fs_info, |
9675 | "no space to allocate a new chunk for block group %llu", | |
9676 | block_group->key.objectid); | |
9677 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9678 | btrfs_end_transaction(trans); |
edbd8d4e | 9679 | out: |
ba1bf481 | 9680 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9681 | return ret; |
9682 | } | |
9683 | ||
6bccf3ab JM |
9684 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9685 | struct btrfs_path *path, | |
9686 | struct btrfs_key *key) | |
0b86a832 | 9687 | { |
6bccf3ab | 9688 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9689 | int ret = 0; |
0b86a832 CM |
9690 | struct btrfs_key found_key; |
9691 | struct extent_buffer *leaf; | |
9692 | int slot; | |
edbd8d4e | 9693 | |
0b86a832 CM |
9694 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9695 | if (ret < 0) | |
925baedd CM |
9696 | goto out; |
9697 | ||
d397712b | 9698 | while (1) { |
0b86a832 | 9699 | slot = path->slots[0]; |
edbd8d4e | 9700 | leaf = path->nodes[0]; |
0b86a832 CM |
9701 | if (slot >= btrfs_header_nritems(leaf)) { |
9702 | ret = btrfs_next_leaf(root, path); | |
9703 | if (ret == 0) | |
9704 | continue; | |
9705 | if (ret < 0) | |
925baedd | 9706 | goto out; |
0b86a832 | 9707 | break; |
edbd8d4e | 9708 | } |
0b86a832 | 9709 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9710 | |
0b86a832 | 9711 | if (found_key.objectid >= key->objectid && |
925baedd | 9712 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9713 | struct extent_map_tree *em_tree; |
9714 | struct extent_map *em; | |
9715 | ||
9716 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9717 | read_lock(&em_tree->lock); | |
9718 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9719 | found_key.offset); | |
9720 | read_unlock(&em_tree->lock); | |
9721 | if (!em) { | |
0b246afa | 9722 | btrfs_err(fs_info, |
6fb37b75 LB |
9723 | "logical %llu len %llu found bg but no related chunk", |
9724 | found_key.objectid, found_key.offset); | |
9725 | ret = -ENOENT; | |
9726 | } else { | |
9727 | ret = 0; | |
9728 | } | |
187ee58c | 9729 | free_extent_map(em); |
925baedd CM |
9730 | goto out; |
9731 | } | |
0b86a832 | 9732 | path->slots[0]++; |
edbd8d4e | 9733 | } |
925baedd | 9734 | out: |
0b86a832 | 9735 | return ret; |
edbd8d4e CM |
9736 | } |
9737 | ||
0af3d00b JB |
9738 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9739 | { | |
9740 | struct btrfs_block_group_cache *block_group; | |
9741 | u64 last = 0; | |
9742 | ||
9743 | while (1) { | |
9744 | struct inode *inode; | |
9745 | ||
9746 | block_group = btrfs_lookup_first_block_group(info, last); | |
9747 | while (block_group) { | |
9748 | spin_lock(&block_group->lock); | |
9749 | if (block_group->iref) | |
9750 | break; | |
9751 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9752 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9753 | } |
9754 | if (!block_group) { | |
9755 | if (last == 0) | |
9756 | break; | |
9757 | last = 0; | |
9758 | continue; | |
9759 | } | |
9760 | ||
9761 | inode = block_group->inode; | |
9762 | block_group->iref = 0; | |
9763 | block_group->inode = NULL; | |
9764 | spin_unlock(&block_group->lock); | |
f3bca802 | 9765 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9766 | iput(inode); |
9767 | last = block_group->key.objectid + block_group->key.offset; | |
9768 | btrfs_put_block_group(block_group); | |
9769 | } | |
9770 | } | |
9771 | ||
5cdd7db6 FM |
9772 | /* |
9773 | * Must be called only after stopping all workers, since we could have block | |
9774 | * group caching kthreads running, and therefore they could race with us if we | |
9775 | * freed the block groups before stopping them. | |
9776 | */ | |
1a40e23b ZY |
9777 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9778 | { | |
9779 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9780 | struct btrfs_space_info *space_info; |
11833d66 | 9781 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9782 | struct rb_node *n; |
9783 | ||
9e351cc8 | 9784 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9785 | while (!list_empty(&info->caching_block_groups)) { |
9786 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9787 | struct btrfs_caching_control, list); | |
9788 | list_del(&caching_ctl->list); | |
9789 | put_caching_control(caching_ctl); | |
9790 | } | |
9e351cc8 | 9791 | up_write(&info->commit_root_sem); |
11833d66 | 9792 | |
47ab2a6c JB |
9793 | spin_lock(&info->unused_bgs_lock); |
9794 | while (!list_empty(&info->unused_bgs)) { | |
9795 | block_group = list_first_entry(&info->unused_bgs, | |
9796 | struct btrfs_block_group_cache, | |
9797 | bg_list); | |
9798 | list_del_init(&block_group->bg_list); | |
9799 | btrfs_put_block_group(block_group); | |
9800 | } | |
9801 | spin_unlock(&info->unused_bgs_lock); | |
9802 | ||
1a40e23b ZY |
9803 | spin_lock(&info->block_group_cache_lock); |
9804 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9805 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9806 | cache_node); | |
1a40e23b ZY |
9807 | rb_erase(&block_group->cache_node, |
9808 | &info->block_group_cache_tree); | |
01eacb27 | 9809 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9810 | spin_unlock(&info->block_group_cache_lock); |
9811 | ||
80eb234a | 9812 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9813 | list_del(&block_group->list); |
80eb234a | 9814 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9815 | |
3c14874a JB |
9816 | /* |
9817 | * We haven't cached this block group, which means we could | |
9818 | * possibly have excluded extents on this block group. | |
9819 | */ | |
36cce922 JB |
9820 | if (block_group->cached == BTRFS_CACHE_NO || |
9821 | block_group->cached == BTRFS_CACHE_ERROR) | |
2ff7e61e | 9822 | free_excluded_extents(info, block_group); |
3c14874a | 9823 | |
817d52f8 | 9824 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9825 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9826 | ASSERT(list_empty(&block_group->dirty_list)); |
9827 | ASSERT(list_empty(&block_group->io_list)); | |
9828 | ASSERT(list_empty(&block_group->bg_list)); | |
9829 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9830 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9831 | |
9832 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9833 | } |
9834 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9835 | |
9836 | /* now that all the block groups are freed, go through and | |
9837 | * free all the space_info structs. This is only called during | |
9838 | * the final stages of unmount, and so we know nobody is | |
9839 | * using them. We call synchronize_rcu() once before we start, | |
9840 | * just to be on the safe side. | |
9841 | */ | |
9842 | synchronize_rcu(); | |
9843 | ||
8929ecfa YZ |
9844 | release_global_block_rsv(info); |
9845 | ||
67871254 | 9846 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9847 | int i; |
9848 | ||
4184ea7f CM |
9849 | space_info = list_entry(info->space_info.next, |
9850 | struct btrfs_space_info, | |
9851 | list); | |
d555b6c3 JB |
9852 | |
9853 | /* | |
9854 | * Do not hide this behind enospc_debug, this is actually | |
9855 | * important and indicates a real bug if this happens. | |
9856 | */ | |
9857 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9858 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9859 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9860 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9861 | list_del(&space_info->list); |
6ab0a202 JM |
9862 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9863 | struct kobject *kobj; | |
c1895442 JM |
9864 | kobj = space_info->block_group_kobjs[i]; |
9865 | space_info->block_group_kobjs[i] = NULL; | |
9866 | if (kobj) { | |
6ab0a202 JM |
9867 | kobject_del(kobj); |
9868 | kobject_put(kobj); | |
9869 | } | |
9870 | } | |
9871 | kobject_del(&space_info->kobj); | |
9872 | kobject_put(&space_info->kobj); | |
4184ea7f | 9873 | } |
1a40e23b ZY |
9874 | return 0; |
9875 | } | |
9876 | ||
b742bb82 YZ |
9877 | static void __link_block_group(struct btrfs_space_info *space_info, |
9878 | struct btrfs_block_group_cache *cache) | |
9879 | { | |
9880 | int index = get_block_group_index(cache); | |
ed55b6ac | 9881 | bool first = false; |
b742bb82 YZ |
9882 | |
9883 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9884 | if (list_empty(&space_info->block_groups[index])) |
9885 | first = true; | |
9886 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9887 | up_write(&space_info->groups_sem); | |
9888 | ||
9889 | if (first) { | |
c1895442 | 9890 | struct raid_kobject *rkobj; |
6ab0a202 JM |
9891 | int ret; |
9892 | ||
c1895442 JM |
9893 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9894 | if (!rkobj) | |
9895 | goto out_err; | |
9896 | rkobj->raid_type = index; | |
9897 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9898 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9899 | "%s", get_raid_name(index)); | |
6ab0a202 | 9900 | if (ret) { |
c1895442 JM |
9901 | kobject_put(&rkobj->kobj); |
9902 | goto out_err; | |
6ab0a202 | 9903 | } |
c1895442 | 9904 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9905 | } |
c1895442 JM |
9906 | |
9907 | return; | |
9908 | out_err: | |
ab8d0fc4 JM |
9909 | btrfs_warn(cache->fs_info, |
9910 | "failed to add kobject for block cache, ignoring"); | |
b742bb82 YZ |
9911 | } |
9912 | ||
920e4a58 | 9913 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
9914 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
9915 | u64 start, u64 size) | |
920e4a58 MX |
9916 | { |
9917 | struct btrfs_block_group_cache *cache; | |
9918 | ||
9919 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9920 | if (!cache) | |
9921 | return NULL; | |
9922 | ||
9923 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9924 | GFP_NOFS); | |
9925 | if (!cache->free_space_ctl) { | |
9926 | kfree(cache); | |
9927 | return NULL; | |
9928 | } | |
9929 | ||
9930 | cache->key.objectid = start; | |
9931 | cache->key.offset = size; | |
9932 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9933 | ||
0b246afa JM |
9934 | cache->sectorsize = fs_info->sectorsize; |
9935 | cache->fs_info = fs_info; | |
2ff7e61e JM |
9936 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, |
9937 | &fs_info->mapping_tree, | |
9938 | start); | |
1e144fb8 OS |
9939 | set_free_space_tree_thresholds(cache); |
9940 | ||
920e4a58 MX |
9941 | atomic_set(&cache->count, 1); |
9942 | spin_lock_init(&cache->lock); | |
e570fd27 | 9943 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
9944 | INIT_LIST_HEAD(&cache->list); |
9945 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 9946 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 9947 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 9948 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 9949 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 9950 | btrfs_init_free_space_ctl(cache); |
04216820 | 9951 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 9952 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 9953 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
9954 | |
9955 | return cache; | |
9956 | } | |
9957 | ||
5b4aacef | 9958 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
9959 | { |
9960 | struct btrfs_path *path; | |
9961 | int ret; | |
9078a3e1 | 9962 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 9963 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
9964 | struct btrfs_key key; |
9965 | struct btrfs_key found_key; | |
5f39d397 | 9966 | struct extent_buffer *leaf; |
0af3d00b JB |
9967 | int need_clear = 0; |
9968 | u64 cache_gen; | |
49303381 LB |
9969 | u64 feature; |
9970 | int mixed; | |
9971 | ||
9972 | feature = btrfs_super_incompat_flags(info->super_copy); | |
9973 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 9974 | |
9078a3e1 | 9975 | key.objectid = 0; |
0b86a832 | 9976 | key.offset = 0; |
962a298f | 9977 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
9978 | path = btrfs_alloc_path(); |
9979 | if (!path) | |
9980 | return -ENOMEM; | |
e4058b54 | 9981 | path->reada = READA_FORWARD; |
9078a3e1 | 9982 | |
0b246afa JM |
9983 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
9984 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
9985 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 9986 | need_clear = 1; |
0b246afa | 9987 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 9988 | need_clear = 1; |
0af3d00b | 9989 | |
d397712b | 9990 | while (1) { |
6bccf3ab | 9991 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
9992 | if (ret > 0) |
9993 | break; | |
0b86a832 CM |
9994 | if (ret != 0) |
9995 | goto error; | |
920e4a58 | 9996 | |
5f39d397 CM |
9997 | leaf = path->nodes[0]; |
9998 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 9999 | |
2ff7e61e | 10000 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10001 | found_key.offset); |
9078a3e1 | 10002 | if (!cache) { |
0b86a832 | 10003 | ret = -ENOMEM; |
f0486c68 | 10004 | goto error; |
9078a3e1 | 10005 | } |
96303081 | 10006 | |
cf7c1ef6 LB |
10007 | if (need_clear) { |
10008 | /* | |
10009 | * When we mount with old space cache, we need to | |
10010 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10011 | * | |
10012 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10013 | * truncate the old free space cache inode and | |
10014 | * setup a new one. | |
10015 | * b) Setting 'dirty flag' makes sure that we flush | |
10016 | * the new space cache info onto disk. | |
10017 | */ | |
0b246afa | 10018 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10019 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10020 | } |
0af3d00b | 10021 | |
5f39d397 CM |
10022 | read_extent_buffer(leaf, &cache->item, |
10023 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10024 | sizeof(cache->item)); | |
920e4a58 | 10025 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10026 | if (!mixed && |
10027 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10028 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10029 | btrfs_err(info, | |
10030 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10031 | cache->key.objectid); | |
10032 | ret = -EINVAL; | |
10033 | goto error; | |
10034 | } | |
0b86a832 | 10035 | |
9078a3e1 | 10036 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10037 | btrfs_release_path(path); |
34d52cb6 | 10038 | |
3c14874a JB |
10039 | /* |
10040 | * We need to exclude the super stripes now so that the space | |
10041 | * info has super bytes accounted for, otherwise we'll think | |
10042 | * we have more space than we actually do. | |
10043 | */ | |
2ff7e61e | 10044 | ret = exclude_super_stripes(info, cache); |
835d974f JB |
10045 | if (ret) { |
10046 | /* | |
10047 | * We may have excluded something, so call this just in | |
10048 | * case. | |
10049 | */ | |
2ff7e61e | 10050 | free_excluded_extents(info, cache); |
920e4a58 | 10051 | btrfs_put_block_group(cache); |
835d974f JB |
10052 | goto error; |
10053 | } | |
3c14874a | 10054 | |
817d52f8 JB |
10055 | /* |
10056 | * check for two cases, either we are full, and therefore | |
10057 | * don't need to bother with the caching work since we won't | |
10058 | * find any space, or we are empty, and we can just add all | |
10059 | * the space in and be done with it. This saves us _alot_ of | |
10060 | * time, particularly in the full case. | |
10061 | */ | |
10062 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10063 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10064 | cache->cached = BTRFS_CACHE_FINISHED; |
2ff7e61e | 10065 | free_excluded_extents(info, cache); |
817d52f8 | 10066 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10067 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10068 | cache->cached = BTRFS_CACHE_FINISHED; |
0b246afa | 10069 | add_new_free_space(cache, info, |
817d52f8 JB |
10070 | found_key.objectid, |
10071 | found_key.objectid + | |
10072 | found_key.offset); | |
2ff7e61e | 10073 | free_excluded_extents(info, cache); |
817d52f8 | 10074 | } |
96b5179d | 10075 | |
0b246afa | 10076 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10077 | if (ret) { |
10078 | btrfs_remove_free_space_cache(cache); | |
10079 | btrfs_put_block_group(cache); | |
10080 | goto error; | |
10081 | } | |
10082 | ||
0b246afa | 10083 | trace_btrfs_add_block_group(info, cache, 0); |
6324fbf3 CM |
10084 | ret = update_space_info(info, cache->flags, found_key.offset, |
10085 | btrfs_block_group_used(&cache->item), | |
e40edf2d | 10086 | cache->bytes_super, &space_info); |
8c579fe7 JB |
10087 | if (ret) { |
10088 | btrfs_remove_free_space_cache(cache); | |
10089 | spin_lock(&info->block_group_cache_lock); | |
10090 | rb_erase(&cache->cache_node, | |
10091 | &info->block_group_cache_tree); | |
01eacb27 | 10092 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
10093 | spin_unlock(&info->block_group_cache_lock); |
10094 | btrfs_put_block_group(cache); | |
10095 | goto error; | |
10096 | } | |
10097 | ||
6324fbf3 | 10098 | cache->space_info = space_info; |
1b2da372 | 10099 | |
b742bb82 | 10100 | __link_block_group(space_info, cache); |
0f9dd46c | 10101 | |
0b246afa | 10102 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10103 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10104 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10105 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10106 | spin_lock(&info->unused_bgs_lock); | |
10107 | /* Should always be true but just in case. */ | |
10108 | if (list_empty(&cache->bg_list)) { | |
10109 | btrfs_get_block_group(cache); | |
10110 | list_add_tail(&cache->bg_list, | |
10111 | &info->unused_bgs); | |
10112 | } | |
10113 | spin_unlock(&info->unused_bgs_lock); | |
10114 | } | |
9078a3e1 | 10115 | } |
b742bb82 | 10116 | |
0b246afa | 10117 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10118 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10119 | (BTRFS_BLOCK_GROUP_RAID10 | |
10120 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10121 | BTRFS_BLOCK_GROUP_RAID5 | |
10122 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10123 | BTRFS_BLOCK_GROUP_DUP))) |
10124 | continue; | |
10125 | /* | |
10126 | * avoid allocating from un-mirrored block group if there are | |
10127 | * mirrored block groups. | |
10128 | */ | |
1095cc0d | 10129 | list_for_each_entry(cache, |
10130 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10131 | list) | |
868f401a | 10132 | inc_block_group_ro(cache, 1); |
1095cc0d | 10133 | list_for_each_entry(cache, |
10134 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10135 | list) | |
868f401a | 10136 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10137 | } |
f0486c68 YZ |
10138 | |
10139 | init_global_block_rsv(info); | |
0b86a832 CM |
10140 | ret = 0; |
10141 | error: | |
9078a3e1 | 10142 | btrfs_free_path(path); |
0b86a832 | 10143 | return ret; |
9078a3e1 | 10144 | } |
6324fbf3 | 10145 | |
ea658bad | 10146 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
2ff7e61e | 10147 | struct btrfs_fs_info *fs_info) |
ea658bad JB |
10148 | { |
10149 | struct btrfs_block_group_cache *block_group, *tmp; | |
0b246afa | 10150 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10151 | struct btrfs_block_group_item item; |
10152 | struct btrfs_key key; | |
10153 | int ret = 0; | |
d9a0540a | 10154 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10155 | |
d9a0540a | 10156 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10157 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10158 | if (ret) |
c92f6be3 | 10159 | goto next; |
ea658bad JB |
10160 | |
10161 | spin_lock(&block_group->lock); | |
10162 | memcpy(&item, &block_group->item, sizeof(item)); | |
10163 | memcpy(&key, &block_group->key, sizeof(key)); | |
10164 | spin_unlock(&block_group->lock); | |
10165 | ||
10166 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10167 | sizeof(item)); | |
10168 | if (ret) | |
66642832 | 10169 | btrfs_abort_transaction(trans, ret); |
0b246afa JM |
10170 | ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid, |
10171 | key.offset); | |
6df9a95e | 10172 | if (ret) |
66642832 | 10173 | btrfs_abort_transaction(trans, ret); |
0b246afa | 10174 | add_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 | 10175 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10176 | next: |
10177 | list_del_init(&block_group->bg_list); | |
ea658bad | 10178 | } |
d9a0540a | 10179 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10180 | } |
10181 | ||
6324fbf3 | 10182 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 10183 | struct btrfs_fs_info *fs_info, u64 bytes_used, |
e17cade2 | 10184 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
10185 | u64 size) |
10186 | { | |
6324fbf3 | 10187 | struct btrfs_block_group_cache *cache; |
0b246afa | 10188 | int ret; |
6324fbf3 | 10189 | |
0b246afa | 10190 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10191 | |
2ff7e61e | 10192 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10193 | if (!cache) |
10194 | return -ENOMEM; | |
34d52cb6 | 10195 | |
6324fbf3 | 10196 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 | 10197 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
6324fbf3 CM |
10198 | btrfs_set_block_group_flags(&cache->item, type); |
10199 | ||
920e4a58 | 10200 | cache->flags = type; |
11833d66 | 10201 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10202 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10203 | cache->needs_free_space = 1; |
2ff7e61e | 10204 | ret = exclude_super_stripes(fs_info, cache); |
835d974f JB |
10205 | if (ret) { |
10206 | /* | |
10207 | * We may have excluded something, so call this just in | |
10208 | * case. | |
10209 | */ | |
2ff7e61e | 10210 | free_excluded_extents(fs_info, cache); |
920e4a58 | 10211 | btrfs_put_block_group(cache); |
835d974f JB |
10212 | return ret; |
10213 | } | |
96303081 | 10214 | |
0b246afa | 10215 | add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size); |
817d52f8 | 10216 | |
2ff7e61e | 10217 | free_excluded_extents(fs_info, cache); |
11833d66 | 10218 | |
d0bd4560 | 10219 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10220 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10221 | u64 new_bytes_used = size - bytes_used; |
10222 | ||
10223 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10224 | fragment_free_space(cache); |
d0bd4560 JB |
10225 | } |
10226 | #endif | |
2e6e5183 FM |
10227 | /* |
10228 | * Call to ensure the corresponding space_info object is created and | |
10229 | * assigned to our block group, but don't update its counters just yet. | |
10230 | * We want our bg to be added to the rbtree with its ->space_info set. | |
10231 | */ | |
0b246afa | 10232 | ret = update_space_info(fs_info, cache->flags, 0, 0, 0, |
2e6e5183 FM |
10233 | &cache->space_info); |
10234 | if (ret) { | |
10235 | btrfs_remove_free_space_cache(cache); | |
10236 | btrfs_put_block_group(cache); | |
10237 | return ret; | |
10238 | } | |
10239 | ||
0b246afa | 10240 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10241 | if (ret) { |
10242 | btrfs_remove_free_space_cache(cache); | |
10243 | btrfs_put_block_group(cache); | |
10244 | return ret; | |
10245 | } | |
10246 | ||
2e6e5183 FM |
10247 | /* |
10248 | * Now that our block group has its ->space_info set and is inserted in | |
10249 | * the rbtree, update the space info's counters. | |
10250 | */ | |
0b246afa JM |
10251 | trace_btrfs_add_block_group(fs_info, cache, 1); |
10252 | ret = update_space_info(fs_info, cache->flags, size, bytes_used, | |
e40edf2d | 10253 | cache->bytes_super, &cache->space_info); |
8c579fe7 JB |
10254 | if (ret) { |
10255 | btrfs_remove_free_space_cache(cache); | |
0b246afa | 10256 | spin_lock(&fs_info->block_group_cache_lock); |
8c579fe7 | 10257 | rb_erase(&cache->cache_node, |
0b246afa | 10258 | &fs_info->block_group_cache_tree); |
01eacb27 | 10259 | RB_CLEAR_NODE(&cache->cache_node); |
0b246afa | 10260 | spin_unlock(&fs_info->block_group_cache_lock); |
8c579fe7 JB |
10261 | btrfs_put_block_group(cache); |
10262 | return ret; | |
10263 | } | |
0b246afa | 10264 | update_global_block_rsv(fs_info); |
1b2da372 | 10265 | |
b742bb82 | 10266 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 10267 | |
47ab2a6c | 10268 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10269 | |
0b246afa | 10270 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10271 | return 0; |
10272 | } | |
1a40e23b | 10273 | |
10ea00f5 ID |
10274 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10275 | { | |
899c81ea ID |
10276 | u64 extra_flags = chunk_to_extended(flags) & |
10277 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10278 | |
de98ced9 | 10279 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10280 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10281 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10282 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10283 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10284 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10285 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10286 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10287 | } |
10288 | ||
1a40e23b | 10289 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
6bccf3ab | 10290 | struct btrfs_fs_info *fs_info, u64 group_start, |
04216820 | 10291 | struct extent_map *em) |
1a40e23b | 10292 | { |
6bccf3ab | 10293 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10294 | struct btrfs_path *path; |
10295 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10296 | struct btrfs_free_cluster *cluster; |
0b246afa | 10297 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10298 | struct btrfs_key key; |
0af3d00b | 10299 | struct inode *inode; |
c1895442 | 10300 | struct kobject *kobj = NULL; |
1a40e23b | 10301 | int ret; |
10ea00f5 | 10302 | int index; |
89a55897 | 10303 | int factor; |
4f69cb98 | 10304 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10305 | bool remove_em; |
1a40e23b | 10306 | |
6bccf3ab | 10307 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10308 | BUG_ON(!block_group); |
c146afad | 10309 | BUG_ON(!block_group->ro); |
1a40e23b | 10310 | |
9f7c43c9 | 10311 | /* |
10312 | * Free the reserved super bytes from this block group before | |
10313 | * remove it. | |
10314 | */ | |
2ff7e61e | 10315 | free_excluded_extents(fs_info, block_group); |
9f7c43c9 | 10316 | |
1a40e23b | 10317 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 10318 | index = get_block_group_index(block_group); |
89a55897 JB |
10319 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10320 | BTRFS_BLOCK_GROUP_RAID1 | | |
10321 | BTRFS_BLOCK_GROUP_RAID10)) | |
10322 | factor = 2; | |
10323 | else | |
10324 | factor = 1; | |
1a40e23b | 10325 | |
44fb5511 | 10326 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10327 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10328 | spin_lock(&cluster->refill_lock); |
10329 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10330 | spin_unlock(&cluster->refill_lock); | |
10331 | ||
10332 | /* | |
10333 | * make sure this block group isn't part of a metadata | |
10334 | * allocation cluster | |
10335 | */ | |
0b246afa | 10336 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10337 | spin_lock(&cluster->refill_lock); |
10338 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10339 | spin_unlock(&cluster->refill_lock); | |
10340 | ||
1a40e23b | 10341 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10342 | if (!path) { |
10343 | ret = -ENOMEM; | |
10344 | goto out; | |
10345 | } | |
1a40e23b | 10346 | |
1bbc621e CM |
10347 | /* |
10348 | * get the inode first so any iput calls done for the io_list | |
10349 | * aren't the final iput (no unlinks allowed now) | |
10350 | */ | |
77ab86bf | 10351 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10352 | |
10353 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10354 | /* | |
10355 | * make sure our free spache cache IO is done before remove the | |
10356 | * free space inode | |
10357 | */ | |
10358 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10359 | if (!list_empty(&block_group->io_list)) { | |
10360 | list_del_init(&block_group->io_list); | |
10361 | ||
10362 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10363 | ||
10364 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10365 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10366 | btrfs_put_block_group(block_group); |
10367 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10368 | } | |
10369 | ||
10370 | if (!list_empty(&block_group->dirty_list)) { | |
10371 | list_del_init(&block_group->dirty_list); | |
10372 | btrfs_put_block_group(block_group); | |
10373 | } | |
10374 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10375 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10376 | ||
0af3d00b | 10377 | if (!IS_ERR(inode)) { |
73f2e545 | 10378 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10379 | if (ret) { |
10380 | btrfs_add_delayed_iput(inode); | |
10381 | goto out; | |
10382 | } | |
0af3d00b JB |
10383 | clear_nlink(inode); |
10384 | /* One for the block groups ref */ | |
10385 | spin_lock(&block_group->lock); | |
10386 | if (block_group->iref) { | |
10387 | block_group->iref = 0; | |
10388 | block_group->inode = NULL; | |
10389 | spin_unlock(&block_group->lock); | |
10390 | iput(inode); | |
10391 | } else { | |
10392 | spin_unlock(&block_group->lock); | |
10393 | } | |
10394 | /* One for our lookup ref */ | |
455757c3 | 10395 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10396 | } |
10397 | ||
10398 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10399 | key.offset = block_group->key.objectid; | |
10400 | key.type = 0; | |
10401 | ||
10402 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10403 | if (ret < 0) | |
10404 | goto out; | |
10405 | if (ret > 0) | |
b3b4aa74 | 10406 | btrfs_release_path(path); |
0af3d00b JB |
10407 | if (ret == 0) { |
10408 | ret = btrfs_del_item(trans, tree_root, path); | |
10409 | if (ret) | |
10410 | goto out; | |
b3b4aa74 | 10411 | btrfs_release_path(path); |
0af3d00b JB |
10412 | } |
10413 | ||
0b246afa | 10414 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10415 | rb_erase(&block_group->cache_node, |
0b246afa | 10416 | &fs_info->block_group_cache_tree); |
292cbd51 | 10417 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10418 | |
0b246afa JM |
10419 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10420 | fs_info->first_logical_byte = (u64)-1; | |
10421 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10422 | |
80eb234a | 10423 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10424 | /* |
10425 | * we must use list_del_init so people can check to see if they | |
10426 | * are still on the list after taking the semaphore | |
10427 | */ | |
10428 | list_del_init(&block_group->list); | |
6ab0a202 | 10429 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10430 | kobj = block_group->space_info->block_group_kobjs[index]; |
10431 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10432 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10433 | } |
80eb234a | 10434 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10435 | if (kobj) { |
10436 | kobject_del(kobj); | |
10437 | kobject_put(kobj); | |
10438 | } | |
1a40e23b | 10439 | |
4f69cb98 FM |
10440 | if (block_group->has_caching_ctl) |
10441 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10442 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10443 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10444 | if (block_group->has_caching_ctl) { |
0b246afa | 10445 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10446 | if (!caching_ctl) { |
10447 | struct btrfs_caching_control *ctl; | |
10448 | ||
10449 | list_for_each_entry(ctl, | |
0b246afa | 10450 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10451 | if (ctl->block_group == block_group) { |
10452 | caching_ctl = ctl; | |
1e4f4714 | 10453 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10454 | break; |
10455 | } | |
10456 | } | |
10457 | if (caching_ctl) | |
10458 | list_del_init(&caching_ctl->list); | |
0b246afa | 10459 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10460 | if (caching_ctl) { |
10461 | /* Once for the caching bgs list and once for us. */ | |
10462 | put_caching_control(caching_ctl); | |
10463 | put_caching_control(caching_ctl); | |
10464 | } | |
10465 | } | |
817d52f8 | 10466 | |
ce93ec54 JB |
10467 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10468 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10469 | WARN_ON(1); |
10470 | } | |
10471 | if (!list_empty(&block_group->io_list)) { | |
10472 | WARN_ON(1); | |
ce93ec54 JB |
10473 | } |
10474 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10475 | btrfs_remove_free_space_cache(block_group); |
10476 | ||
c146afad | 10477 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10478 | list_del_init(&block_group->ro_list); |
18d018ad | 10479 | |
0b246afa | 10480 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10481 | WARN_ON(block_group->space_info->total_bytes |
10482 | < block_group->key.offset); | |
10483 | WARN_ON(block_group->space_info->bytes_readonly | |
10484 | < block_group->key.offset); | |
10485 | WARN_ON(block_group->space_info->disk_total | |
10486 | < block_group->key.offset * factor); | |
10487 | } | |
c146afad YZ |
10488 | block_group->space_info->total_bytes -= block_group->key.offset; |
10489 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10490 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10491 | |
c146afad | 10492 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10493 | |
0af3d00b JB |
10494 | memcpy(&key, &block_group->key, sizeof(key)); |
10495 | ||
34441361 | 10496 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10497 | if (!list_empty(&em->list)) { |
10498 | /* We're in the transaction->pending_chunks list. */ | |
10499 | free_extent_map(em); | |
10500 | } | |
04216820 FM |
10501 | spin_lock(&block_group->lock); |
10502 | block_group->removed = 1; | |
10503 | /* | |
10504 | * At this point trimming can't start on this block group, because we | |
10505 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10506 | * so no one can't find it anymore and even if someone already got this | |
10507 | * block group before we removed it from the rbtree, they have already | |
10508 | * incremented block_group->trimming - if they didn't, they won't find | |
10509 | * any free space entries because we already removed them all when we | |
10510 | * called btrfs_remove_free_space_cache(). | |
10511 | * | |
10512 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10513 | * to prevent the same logical address range and physical device space | |
10514 | * ranges from being reused for a new block group. This is because our | |
10515 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10516 | * completely transactionless, so while it is trimming a range the | |
10517 | * currently running transaction might finish and a new one start, | |
10518 | * allowing for new block groups to be created that can reuse the same | |
10519 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10520 | * |
10521 | * There may also be an implicit trim operation if the file system | |
10522 | * is mounted with -odiscard. The same protections must remain | |
10523 | * in place until the extents have been discarded completely when | |
10524 | * the transaction commit has completed. | |
04216820 FM |
10525 | */ |
10526 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10527 | /* | |
10528 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10529 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10530 | * before checking block_group->removed). | |
10531 | */ | |
10532 | if (!remove_em) { | |
10533 | /* | |
10534 | * Our em might be in trans->transaction->pending_chunks which | |
10535 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10536 | * and so is the fs_info->pinned_chunks list. | |
10537 | * | |
10538 | * So at this point we must be holding the chunk_mutex to avoid | |
10539 | * any races with chunk allocation (more specifically at | |
10540 | * volumes.c:contains_pending_extent()), to ensure it always | |
10541 | * sees the em, either in the pending_chunks list or in the | |
10542 | * pinned_chunks list. | |
10543 | */ | |
0b246afa | 10544 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10545 | } |
10546 | spin_unlock(&block_group->lock); | |
04216820 FM |
10547 | |
10548 | if (remove_em) { | |
10549 | struct extent_map_tree *em_tree; | |
10550 | ||
0b246afa | 10551 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10552 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10553 | /* |
10554 | * The em might be in the pending_chunks list, so make sure the | |
10555 | * chunk mutex is locked, since remove_extent_mapping() will | |
10556 | * delete us from that list. | |
10557 | */ | |
04216820 FM |
10558 | remove_extent_mapping(em_tree, em); |
10559 | write_unlock(&em_tree->lock); | |
10560 | /* once for the tree */ | |
10561 | free_extent_map(em); | |
10562 | } | |
10563 | ||
34441361 | 10564 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10565 | |
0b246afa | 10566 | ret = remove_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 OS |
10567 | if (ret) |
10568 | goto out; | |
10569 | ||
fa9c0d79 CM |
10570 | btrfs_put_block_group(block_group); |
10571 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10572 | |
10573 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10574 | if (ret > 0) | |
10575 | ret = -EIO; | |
10576 | if (ret < 0) | |
10577 | goto out; | |
10578 | ||
10579 | ret = btrfs_del_item(trans, root, path); | |
10580 | out: | |
10581 | btrfs_free_path(path); | |
10582 | return ret; | |
10583 | } | |
acce952b | 10584 | |
8eab77ff | 10585 | struct btrfs_trans_handle * |
7fd01182 FM |
10586 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10587 | const u64 chunk_offset) | |
8eab77ff | 10588 | { |
7fd01182 FM |
10589 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10590 | struct extent_map *em; | |
10591 | struct map_lookup *map; | |
10592 | unsigned int num_items; | |
10593 | ||
10594 | read_lock(&em_tree->lock); | |
10595 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10596 | read_unlock(&em_tree->lock); | |
10597 | ASSERT(em && em->start == chunk_offset); | |
10598 | ||
8eab77ff | 10599 | /* |
7fd01182 FM |
10600 | * We need to reserve 3 + N units from the metadata space info in order |
10601 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10602 | * btrfs_remove_block_group()), which are used for: | |
10603 | * | |
8eab77ff FM |
10604 | * 1 unit for adding the free space inode's orphan (located in the tree |
10605 | * of tree roots). | |
7fd01182 FM |
10606 | * 1 unit for deleting the block group item (located in the extent |
10607 | * tree). | |
10608 | * 1 unit for deleting the free space item (located in tree of tree | |
10609 | * roots). | |
10610 | * N units for deleting N device extent items corresponding to each | |
10611 | * stripe (located in the device tree). | |
10612 | * | |
10613 | * In order to remove a block group we also need to reserve units in the | |
10614 | * system space info in order to update the chunk tree (update one or | |
10615 | * more device items and remove one chunk item), but this is done at | |
10616 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10617 | */ |
95617d69 | 10618 | map = em->map_lookup; |
7fd01182 FM |
10619 | num_items = 3 + map->num_stripes; |
10620 | free_extent_map(em); | |
10621 | ||
8eab77ff | 10622 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10623 | num_items, 1); |
8eab77ff FM |
10624 | } |
10625 | ||
47ab2a6c JB |
10626 | /* |
10627 | * Process the unused_bgs list and remove any that don't have any allocated | |
10628 | * space inside of them. | |
10629 | */ | |
10630 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10631 | { | |
10632 | struct btrfs_block_group_cache *block_group; | |
10633 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10634 | struct btrfs_trans_handle *trans; |
10635 | int ret = 0; | |
10636 | ||
afcdd129 | 10637 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10638 | return; |
10639 | ||
10640 | spin_lock(&fs_info->unused_bgs_lock); | |
10641 | while (!list_empty(&fs_info->unused_bgs)) { | |
10642 | u64 start, end; | |
e33e17ee | 10643 | int trimming; |
47ab2a6c JB |
10644 | |
10645 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10646 | struct btrfs_block_group_cache, | |
10647 | bg_list); | |
47ab2a6c | 10648 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10649 | |
10650 | space_info = block_group->space_info; | |
10651 | ||
47ab2a6c JB |
10652 | if (ret || btrfs_mixed_space_info(space_info)) { |
10653 | btrfs_put_block_group(block_group); | |
10654 | continue; | |
10655 | } | |
10656 | spin_unlock(&fs_info->unused_bgs_lock); | |
10657 | ||
d5f2e33b | 10658 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10659 | |
47ab2a6c JB |
10660 | /* Don't want to race with allocators so take the groups_sem */ |
10661 | down_write(&space_info->groups_sem); | |
10662 | spin_lock(&block_group->lock); | |
10663 | if (block_group->reserved || | |
10664 | btrfs_block_group_used(&block_group->item) || | |
19c4d2f9 | 10665 | block_group->ro || |
aefbe9a6 | 10666 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10667 | /* |
10668 | * We want to bail if we made new allocations or have | |
10669 | * outstanding allocations in this block group. We do | |
10670 | * the ro check in case balance is currently acting on | |
10671 | * this block group. | |
10672 | */ | |
10673 | spin_unlock(&block_group->lock); | |
10674 | up_write(&space_info->groups_sem); | |
10675 | goto next; | |
10676 | } | |
10677 | spin_unlock(&block_group->lock); | |
10678 | ||
10679 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10680 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10681 | up_write(&space_info->groups_sem); |
10682 | if (ret < 0) { | |
10683 | ret = 0; | |
10684 | goto next; | |
10685 | } | |
10686 | ||
10687 | /* | |
10688 | * Want to do this before we do anything else so we can recover | |
10689 | * properly if we fail to join the transaction. | |
10690 | */ | |
7fd01182 FM |
10691 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10692 | block_group->key.objectid); | |
47ab2a6c | 10693 | if (IS_ERR(trans)) { |
2ff7e61e | 10694 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10695 | ret = PTR_ERR(trans); |
10696 | goto next; | |
10697 | } | |
10698 | ||
10699 | /* | |
10700 | * We could have pending pinned extents for this block group, | |
10701 | * just delete them, we don't care about them anymore. | |
10702 | */ | |
10703 | start = block_group->key.objectid; | |
10704 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10705 | /* |
10706 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10707 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10708 | * another task might be running finish_extent_commit() for the | |
10709 | * previous transaction N - 1, and have seen a range belonging | |
10710 | * to the block group in freed_extents[] before we were able to | |
10711 | * clear the whole block group range from freed_extents[]. This | |
10712 | * means that task can lookup for the block group after we | |
10713 | * unpinned it from freed_extents[] and removed it, leading to | |
10714 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10715 | */ | |
10716 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10717 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10718 | EXTENT_DIRTY); |
758eb51e | 10719 | if (ret) { |
d4b450cd | 10720 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10721 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10722 | goto end_trans; |
10723 | } | |
10724 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10725 | EXTENT_DIRTY); |
758eb51e | 10726 | if (ret) { |
d4b450cd | 10727 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10728 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10729 | goto end_trans; |
10730 | } | |
d4b450cd | 10731 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10732 | |
10733 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10734 | spin_lock(&space_info->lock); |
10735 | spin_lock(&block_group->lock); | |
10736 | ||
10737 | space_info->bytes_pinned -= block_group->pinned; | |
10738 | space_info->bytes_readonly += block_group->pinned; | |
10739 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10740 | -block_group->pinned); | |
47ab2a6c JB |
10741 | block_group->pinned = 0; |
10742 | ||
c30666d4 ZL |
10743 | spin_unlock(&block_group->lock); |
10744 | spin_unlock(&space_info->lock); | |
10745 | ||
e33e17ee | 10746 | /* DISCARD can flip during remount */ |
0b246afa | 10747 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10748 | |
10749 | /* Implicit trim during transaction commit. */ | |
10750 | if (trimming) | |
10751 | btrfs_get_block_group_trimming(block_group); | |
10752 | ||
47ab2a6c JB |
10753 | /* |
10754 | * Btrfs_remove_chunk will abort the transaction if things go | |
10755 | * horribly wrong. | |
10756 | */ | |
5b4aacef | 10757 | ret = btrfs_remove_chunk(trans, fs_info, |
47ab2a6c | 10758 | block_group->key.objectid); |
e33e17ee JM |
10759 | |
10760 | if (ret) { | |
10761 | if (trimming) | |
10762 | btrfs_put_block_group_trimming(block_group); | |
10763 | goto end_trans; | |
10764 | } | |
10765 | ||
10766 | /* | |
10767 | * If we're not mounted with -odiscard, we can just forget | |
10768 | * about this block group. Otherwise we'll need to wait | |
10769 | * until transaction commit to do the actual discard. | |
10770 | */ | |
10771 | if (trimming) { | |
348a0013 FM |
10772 | spin_lock(&fs_info->unused_bgs_lock); |
10773 | /* | |
10774 | * A concurrent scrub might have added us to the list | |
10775 | * fs_info->unused_bgs, so use a list_move operation | |
10776 | * to add the block group to the deleted_bgs list. | |
10777 | */ | |
e33e17ee JM |
10778 | list_move(&block_group->bg_list, |
10779 | &trans->transaction->deleted_bgs); | |
348a0013 | 10780 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10781 | btrfs_get_block_group(block_group); |
10782 | } | |
758eb51e | 10783 | end_trans: |
3a45bb20 | 10784 | btrfs_end_transaction(trans); |
47ab2a6c | 10785 | next: |
d5f2e33b | 10786 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10787 | btrfs_put_block_group(block_group); |
10788 | spin_lock(&fs_info->unused_bgs_lock); | |
10789 | } | |
10790 | spin_unlock(&fs_info->unused_bgs_lock); | |
10791 | } | |
10792 | ||
c59021f8 | 10793 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10794 | { | |
10795 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10796 | struct btrfs_super_block *disk_super; |
10797 | u64 features; | |
10798 | u64 flags; | |
10799 | int mixed = 0; | |
c59021f8 | 10800 | int ret; |
10801 | ||
6c41761f | 10802 | disk_super = fs_info->super_copy; |
1aba86d6 | 10803 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10804 | return -EINVAL; |
c59021f8 | 10805 | |
1aba86d6 | 10806 | features = btrfs_super_incompat_flags(disk_super); |
10807 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10808 | mixed = 1; | |
c59021f8 | 10809 | |
1aba86d6 | 10810 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
e40edf2d | 10811 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
c59021f8 | 10812 | if (ret) |
1aba86d6 | 10813 | goto out; |
c59021f8 | 10814 | |
1aba86d6 | 10815 | if (mixed) { |
10816 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
e40edf2d | 10817 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10818 | } else { |
10819 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
e40edf2d | 10820 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10821 | if (ret) |
10822 | goto out; | |
10823 | ||
10824 | flags = BTRFS_BLOCK_GROUP_DATA; | |
e40edf2d | 10825 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10826 | } |
10827 | out: | |
c59021f8 | 10828 | return ret; |
10829 | } | |
10830 | ||
2ff7e61e JM |
10831 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10832 | u64 start, u64 end) | |
acce952b | 10833 | { |
2ff7e61e | 10834 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10835 | } |
10836 | ||
499f377f JM |
10837 | /* |
10838 | * It used to be that old block groups would be left around forever. | |
10839 | * Iterating over them would be enough to trim unused space. Since we | |
10840 | * now automatically remove them, we also need to iterate over unallocated | |
10841 | * space. | |
10842 | * | |
10843 | * We don't want a transaction for this since the discard may take a | |
10844 | * substantial amount of time. We don't require that a transaction be | |
10845 | * running, but we do need to take a running transaction into account | |
10846 | * to ensure that we're not discarding chunks that were released in | |
10847 | * the current transaction. | |
10848 | * | |
10849 | * Holding the chunks lock will prevent other threads from allocating | |
10850 | * or releasing chunks, but it won't prevent a running transaction | |
10851 | * from committing and releasing the memory that the pending chunks | |
10852 | * list head uses. For that, we need to take a reference to the | |
10853 | * transaction. | |
10854 | */ | |
10855 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10856 | u64 minlen, u64 *trimmed) | |
10857 | { | |
10858 | u64 start = 0, len = 0; | |
10859 | int ret; | |
10860 | ||
10861 | *trimmed = 0; | |
10862 | ||
10863 | /* Not writeable = nothing to do. */ | |
10864 | if (!device->writeable) | |
10865 | return 0; | |
10866 | ||
10867 | /* No free space = nothing to do. */ | |
10868 | if (device->total_bytes <= device->bytes_used) | |
10869 | return 0; | |
10870 | ||
10871 | ret = 0; | |
10872 | ||
10873 | while (1) { | |
fb456252 | 10874 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10875 | struct btrfs_transaction *trans; |
10876 | u64 bytes; | |
10877 | ||
10878 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10879 | if (ret) | |
10880 | return ret; | |
10881 | ||
10882 | down_read(&fs_info->commit_root_sem); | |
10883 | ||
10884 | spin_lock(&fs_info->trans_lock); | |
10885 | trans = fs_info->running_transaction; | |
10886 | if (trans) | |
9b64f57d | 10887 | refcount_inc(&trans->use_count); |
499f377f JM |
10888 | spin_unlock(&fs_info->trans_lock); |
10889 | ||
10890 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10891 | &start, &len); | |
10892 | if (trans) | |
10893 | btrfs_put_transaction(trans); | |
10894 | ||
10895 | if (ret) { | |
10896 | up_read(&fs_info->commit_root_sem); | |
10897 | mutex_unlock(&fs_info->chunk_mutex); | |
10898 | if (ret == -ENOSPC) | |
10899 | ret = 0; | |
10900 | break; | |
10901 | } | |
10902 | ||
10903 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10904 | up_read(&fs_info->commit_root_sem); | |
10905 | mutex_unlock(&fs_info->chunk_mutex); | |
10906 | ||
10907 | if (ret) | |
10908 | break; | |
10909 | ||
10910 | start += len; | |
10911 | *trimmed += bytes; | |
10912 | ||
10913 | if (fatal_signal_pending(current)) { | |
10914 | ret = -ERESTARTSYS; | |
10915 | break; | |
10916 | } | |
10917 | ||
10918 | cond_resched(); | |
10919 | } | |
10920 | ||
10921 | return ret; | |
10922 | } | |
10923 | ||
2ff7e61e | 10924 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 10925 | { |
f7039b1d | 10926 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
10927 | struct btrfs_device *device; |
10928 | struct list_head *devices; | |
f7039b1d LD |
10929 | u64 group_trimmed; |
10930 | u64 start; | |
10931 | u64 end; | |
10932 | u64 trimmed = 0; | |
2cac13e4 | 10933 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10934 | int ret = 0; |
10935 | ||
2cac13e4 LB |
10936 | /* |
10937 | * try to trim all FS space, our block group may start from non-zero. | |
10938 | */ | |
10939 | if (range->len == total_bytes) | |
10940 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10941 | else | |
10942 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10943 | |
10944 | while (cache) { | |
10945 | if (cache->key.objectid >= (range->start + range->len)) { | |
10946 | btrfs_put_block_group(cache); | |
10947 | break; | |
10948 | } | |
10949 | ||
10950 | start = max(range->start, cache->key.objectid); | |
10951 | end = min(range->start + range->len, | |
10952 | cache->key.objectid + cache->key.offset); | |
10953 | ||
10954 | if (end - start >= range->minlen) { | |
10955 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10956 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
10957 | if (ret) { |
10958 | btrfs_put_block_group(cache); | |
10959 | break; | |
10960 | } | |
10961 | ret = wait_block_group_cache_done(cache); | |
10962 | if (ret) { | |
10963 | btrfs_put_block_group(cache); | |
10964 | break; | |
10965 | } | |
f7039b1d LD |
10966 | } |
10967 | ret = btrfs_trim_block_group(cache, | |
10968 | &group_trimmed, | |
10969 | start, | |
10970 | end, | |
10971 | range->minlen); | |
10972 | ||
10973 | trimmed += group_trimmed; | |
10974 | if (ret) { | |
10975 | btrfs_put_block_group(cache); | |
10976 | break; | |
10977 | } | |
10978 | } | |
10979 | ||
2ff7e61e | 10980 | cache = next_block_group(fs_info, cache); |
f7039b1d LD |
10981 | } |
10982 | ||
0b246afa JM |
10983 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
10984 | devices = &fs_info->fs_devices->alloc_list; | |
499f377f JM |
10985 | list_for_each_entry(device, devices, dev_alloc_list) { |
10986 | ret = btrfs_trim_free_extents(device, range->minlen, | |
10987 | &group_trimmed); | |
10988 | if (ret) | |
10989 | break; | |
10990 | ||
10991 | trimmed += group_trimmed; | |
10992 | } | |
0b246afa | 10993 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 10994 | |
f7039b1d LD |
10995 | range->len = trimmed; |
10996 | return ret; | |
10997 | } | |
8257b2dc MX |
10998 | |
10999 | /* | |
9ea24bbe FM |
11000 | * btrfs_{start,end}_write_no_snapshoting() are similar to |
11001 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing | |
11002 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11003 | * but flush the data into disk after the snapshot creation, or to prevent | |
11004 | * operations while snapshoting is ongoing and that cause the snapshot to be | |
11005 | * inconsistent (writes followed by expanding truncates for example). | |
8257b2dc | 11006 | */ |
9ea24bbe | 11007 | void btrfs_end_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc MX |
11008 | { |
11009 | percpu_counter_dec(&root->subv_writers->counter); | |
11010 | /* | |
a83342aa | 11011 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
11012 | */ |
11013 | smp_mb(); | |
11014 | if (waitqueue_active(&root->subv_writers->wait)) | |
11015 | wake_up(&root->subv_writers->wait); | |
11016 | } | |
11017 | ||
9ea24bbe | 11018 | int btrfs_start_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc | 11019 | { |
ee39b432 | 11020 | if (atomic_read(&root->will_be_snapshoted)) |
8257b2dc MX |
11021 | return 0; |
11022 | ||
11023 | percpu_counter_inc(&root->subv_writers->counter); | |
11024 | /* | |
11025 | * Make sure counter is updated before we check for snapshot creation. | |
11026 | */ | |
11027 | smp_mb(); | |
ee39b432 | 11028 | if (atomic_read(&root->will_be_snapshoted)) { |
9ea24bbe | 11029 | btrfs_end_write_no_snapshoting(root); |
8257b2dc MX |
11030 | return 0; |
11031 | } | |
11032 | return 1; | |
11033 | } | |
0bc19f90 ZL |
11034 | |
11035 | static int wait_snapshoting_atomic_t(atomic_t *a) | |
11036 | { | |
11037 | schedule(); | |
11038 | return 0; | |
11039 | } | |
11040 | ||
11041 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) | |
11042 | { | |
11043 | while (true) { | |
11044 | int ret; | |
11045 | ||
11046 | ret = btrfs_start_write_no_snapshoting(root); | |
11047 | if (ret) | |
11048 | break; | |
11049 | wait_on_atomic_t(&root->will_be_snapshoted, | |
11050 | wait_snapshoting_atomic_t, | |
11051 | TASK_UNINTERRUPTIBLE); | |
11052 | } | |
11053 | } |