Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 | 4 | */ |
c1d7c514 | 5 | |
ec6b910f | 6 | #include <linux/sched.h> |
f361bf4a | 7 | #include <linux/sched/signal.h> |
edbd8d4e | 8 | #include <linux/pagemap.h> |
ec44a35c | 9 | #include <linux/writeback.h> |
21af804c | 10 | #include <linux/blkdev.h> |
b7a9f29f | 11 | #include <linux/sort.h> |
4184ea7f | 12 | #include <linux/rcupdate.h> |
817d52f8 | 13 | #include <linux/kthread.h> |
5a0e3ad6 | 14 | #include <linux/slab.h> |
dff51cd1 | 15 | #include <linux/ratelimit.h> |
b150a4f1 | 16 | #include <linux/percpu_counter.h> |
69fe2d75 | 17 | #include <linux/lockdep.h> |
9678c543 | 18 | #include <linux/crc32c.h> |
995946dd | 19 | #include "tree-log.h" |
fec577fb CM |
20 | #include "disk-io.h" |
21 | #include "print-tree.h" | |
0b86a832 | 22 | #include "volumes.h" |
53b381b3 | 23 | #include "raid56.h" |
925baedd | 24 | #include "locking.h" |
fa9c0d79 | 25 | #include "free-space-cache.h" |
1e144fb8 | 26 | #include "free-space-tree.h" |
3fed40cc | 27 | #include "math.h" |
6ab0a202 | 28 | #include "sysfs.h" |
fcebe456 | 29 | #include "qgroup.h" |
fd708b81 | 30 | #include "ref-verify.h" |
fec577fb | 31 | |
709c0486 AJ |
32 | #undef SCRAMBLE_DELAYED_REFS |
33 | ||
9e622d6b MX |
34 | /* |
35 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
36 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
37 | * if we really need one. | |
38 | * | |
0e4f8f88 CM |
39 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
40 | * if we have very few chunks already allocated. This is | |
41 | * used as part of the clustering code to help make sure | |
42 | * we have a good pool of storage to cluster in, without | |
43 | * filling the FS with empty chunks | |
44 | * | |
9e622d6b MX |
45 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
46 | * | |
0e4f8f88 CM |
47 | */ |
48 | enum { | |
49 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
50 | CHUNK_ALLOC_LIMITED = 1, |
51 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
52 | }; |
53 | ||
5d4f98a2 | 54 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 55 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 56 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
57 | u64 root_objectid, u64 owner_objectid, |
58 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 59 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
60 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
61 | struct extent_buffer *leaf, | |
62 | struct btrfs_extent_item *ei); | |
63 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
2ff7e61e | 64 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
65 | u64 parent, u64 root_objectid, |
66 | u64 flags, u64 owner, u64 offset, | |
67 | struct btrfs_key *ins, int ref_mod); | |
68 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 NB |
69 | struct btrfs_delayed_ref_node *node, |
70 | u64 flags, struct btrfs_disk_key *key); | |
6a63209f | 71 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 72 | struct btrfs_fs_info *fs_info, u64 flags, |
698d0082 | 73 | int force); |
11833d66 YZ |
74 | static int find_next_key(struct btrfs_path *path, int level, |
75 | struct btrfs_key *key); | |
ab8d0fc4 JM |
76 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
77 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 78 | int dump_block_groups); |
5d80366e JB |
79 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
80 | u64 num_bytes); | |
957780eb JB |
81 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
82 | struct btrfs_space_info *space_info, | |
83 | u64 num_bytes); | |
84 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
85 | struct btrfs_space_info *space_info, | |
86 | u64 num_bytes); | |
6a63209f | 87 | |
817d52f8 JB |
88 | static noinline int |
89 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
90 | { | |
91 | smp_mb(); | |
36cce922 JB |
92 | return cache->cached == BTRFS_CACHE_FINISHED || |
93 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
94 | } |
95 | ||
0f9dd46c JB |
96 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
97 | { | |
98 | return (cache->flags & bits) == bits; | |
99 | } | |
100 | ||
758f2dfc | 101 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
102 | { |
103 | atomic_inc(&cache->count); | |
104 | } | |
105 | ||
106 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
107 | { | |
f0486c68 YZ |
108 | if (atomic_dec_and_test(&cache->count)) { |
109 | WARN_ON(cache->pinned > 0); | |
110 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
111 | |
112 | /* | |
113 | * If not empty, someone is still holding mutex of | |
114 | * full_stripe_lock, which can only be released by caller. | |
115 | * And it will definitely cause use-after-free when caller | |
116 | * tries to release full stripe lock. | |
117 | * | |
118 | * No better way to resolve, but only to warn. | |
119 | */ | |
120 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 121 | kfree(cache->free_space_ctl); |
11dfe35a | 122 | kfree(cache); |
f0486c68 | 123 | } |
11dfe35a JB |
124 | } |
125 | ||
0f9dd46c JB |
126 | /* |
127 | * this adds the block group to the fs_info rb tree for the block group | |
128 | * cache | |
129 | */ | |
b2950863 | 130 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
131 | struct btrfs_block_group_cache *block_group) |
132 | { | |
133 | struct rb_node **p; | |
134 | struct rb_node *parent = NULL; | |
135 | struct btrfs_block_group_cache *cache; | |
136 | ||
137 | spin_lock(&info->block_group_cache_lock); | |
138 | p = &info->block_group_cache_tree.rb_node; | |
139 | ||
140 | while (*p) { | |
141 | parent = *p; | |
142 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
143 | cache_node); | |
144 | if (block_group->key.objectid < cache->key.objectid) { | |
145 | p = &(*p)->rb_left; | |
146 | } else if (block_group->key.objectid > cache->key.objectid) { | |
147 | p = &(*p)->rb_right; | |
148 | } else { | |
149 | spin_unlock(&info->block_group_cache_lock); | |
150 | return -EEXIST; | |
151 | } | |
152 | } | |
153 | ||
154 | rb_link_node(&block_group->cache_node, parent, p); | |
155 | rb_insert_color(&block_group->cache_node, | |
156 | &info->block_group_cache_tree); | |
a1897fdd LB |
157 | |
158 | if (info->first_logical_byte > block_group->key.objectid) | |
159 | info->first_logical_byte = block_group->key.objectid; | |
160 | ||
0f9dd46c JB |
161 | spin_unlock(&info->block_group_cache_lock); |
162 | ||
163 | return 0; | |
164 | } | |
165 | ||
166 | /* | |
167 | * This will return the block group at or after bytenr if contains is 0, else | |
168 | * it will return the block group that contains the bytenr | |
169 | */ | |
170 | static struct btrfs_block_group_cache * | |
171 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
172 | int contains) | |
173 | { | |
174 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
175 | struct rb_node *n; | |
176 | u64 end, start; | |
177 | ||
178 | spin_lock(&info->block_group_cache_lock); | |
179 | n = info->block_group_cache_tree.rb_node; | |
180 | ||
181 | while (n) { | |
182 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
183 | cache_node); | |
184 | end = cache->key.objectid + cache->key.offset - 1; | |
185 | start = cache->key.objectid; | |
186 | ||
187 | if (bytenr < start) { | |
188 | if (!contains && (!ret || start < ret->key.objectid)) | |
189 | ret = cache; | |
190 | n = n->rb_left; | |
191 | } else if (bytenr > start) { | |
192 | if (contains && bytenr <= end) { | |
193 | ret = cache; | |
194 | break; | |
195 | } | |
196 | n = n->rb_right; | |
197 | } else { | |
198 | ret = cache; | |
199 | break; | |
200 | } | |
201 | } | |
a1897fdd | 202 | if (ret) { |
11dfe35a | 203 | btrfs_get_block_group(ret); |
a1897fdd LB |
204 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
205 | info->first_logical_byte = ret->key.objectid; | |
206 | } | |
0f9dd46c JB |
207 | spin_unlock(&info->block_group_cache_lock); |
208 | ||
209 | return ret; | |
210 | } | |
211 | ||
2ff7e61e | 212 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 213 | u64 start, u64 num_bytes) |
817d52f8 | 214 | { |
11833d66 | 215 | u64 end = start + num_bytes - 1; |
0b246afa | 216 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 217 | start, end, EXTENT_UPTODATE); |
0b246afa | 218 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 219 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
220 | return 0; |
221 | } | |
817d52f8 | 222 | |
2ff7e61e | 223 | static void free_excluded_extents(struct btrfs_fs_info *fs_info, |
11833d66 YZ |
224 | struct btrfs_block_group_cache *cache) |
225 | { | |
226 | u64 start, end; | |
817d52f8 | 227 | |
11833d66 YZ |
228 | start = cache->key.objectid; |
229 | end = start + cache->key.offset - 1; | |
230 | ||
0b246afa | 231 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 232 | start, end, EXTENT_UPTODATE); |
0b246afa | 233 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 234 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
235 | } |
236 | ||
2ff7e61e | 237 | static int exclude_super_stripes(struct btrfs_fs_info *fs_info, |
11833d66 | 238 | struct btrfs_block_group_cache *cache) |
817d52f8 | 239 | { |
817d52f8 JB |
240 | u64 bytenr; |
241 | u64 *logical; | |
242 | int stripe_len; | |
243 | int i, nr, ret; | |
244 | ||
06b2331f YZ |
245 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
246 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
247 | cache->bytes_super += stripe_len; | |
2ff7e61e | 248 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 249 | stripe_len); |
835d974f JB |
250 | if (ret) |
251 | return ret; | |
06b2331f YZ |
252 | } |
253 | ||
817d52f8 JB |
254 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
255 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 256 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
63a9c7b9 | 257 | bytenr, &logical, &nr, &stripe_len); |
835d974f JB |
258 | if (ret) |
259 | return ret; | |
11833d66 | 260 | |
817d52f8 | 261 | while (nr--) { |
51bf5f0b JB |
262 | u64 start, len; |
263 | ||
264 | if (logical[nr] > cache->key.objectid + | |
265 | cache->key.offset) | |
266 | continue; | |
267 | ||
268 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
269 | continue; | |
270 | ||
271 | start = logical[nr]; | |
272 | if (start < cache->key.objectid) { | |
273 | start = cache->key.objectid; | |
274 | len = (logical[nr] + stripe_len) - start; | |
275 | } else { | |
276 | len = min_t(u64, stripe_len, | |
277 | cache->key.objectid + | |
278 | cache->key.offset - start); | |
279 | } | |
280 | ||
281 | cache->bytes_super += len; | |
2ff7e61e | 282 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
283 | if (ret) { |
284 | kfree(logical); | |
285 | return ret; | |
286 | } | |
817d52f8 | 287 | } |
11833d66 | 288 | |
817d52f8 JB |
289 | kfree(logical); |
290 | } | |
817d52f8 JB |
291 | return 0; |
292 | } | |
293 | ||
11833d66 YZ |
294 | static struct btrfs_caching_control * |
295 | get_caching_control(struct btrfs_block_group_cache *cache) | |
296 | { | |
297 | struct btrfs_caching_control *ctl; | |
298 | ||
299 | spin_lock(&cache->lock); | |
dde5abee JB |
300 | if (!cache->caching_ctl) { |
301 | spin_unlock(&cache->lock); | |
11833d66 YZ |
302 | return NULL; |
303 | } | |
304 | ||
305 | ctl = cache->caching_ctl; | |
1e4f4714 | 306 | refcount_inc(&ctl->count); |
11833d66 YZ |
307 | spin_unlock(&cache->lock); |
308 | return ctl; | |
309 | } | |
310 | ||
311 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
312 | { | |
1e4f4714 | 313 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
314 | kfree(ctl); |
315 | } | |
316 | ||
d0bd4560 | 317 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 318 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 319 | { |
2ff7e61e | 320 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
321 | u64 start = block_group->key.objectid; |
322 | u64 len = block_group->key.offset; | |
323 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 324 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
325 | u64 step = chunk << 1; |
326 | ||
327 | while (len > chunk) { | |
328 | btrfs_remove_free_space(block_group, start, chunk); | |
329 | start += step; | |
330 | if (len < step) | |
331 | len = 0; | |
332 | else | |
333 | len -= step; | |
334 | } | |
335 | } | |
336 | #endif | |
337 | ||
0f9dd46c JB |
338 | /* |
339 | * this is only called by cache_block_group, since we could have freed extents | |
340 | * we need to check the pinned_extents for any extents that can't be used yet | |
341 | * since their free space will be released as soon as the transaction commits. | |
342 | */ | |
a5ed9182 | 343 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
4457c1c7 | 344 | u64 start, u64 end) |
0f9dd46c | 345 | { |
4457c1c7 | 346 | struct btrfs_fs_info *info = block_group->fs_info; |
817d52f8 | 347 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
348 | int ret; |
349 | ||
350 | while (start < end) { | |
11833d66 | 351 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 352 | &extent_start, &extent_end, |
e6138876 JB |
353 | EXTENT_DIRTY | EXTENT_UPTODATE, |
354 | NULL); | |
0f9dd46c JB |
355 | if (ret) |
356 | break; | |
357 | ||
06b2331f | 358 | if (extent_start <= start) { |
0f9dd46c JB |
359 | start = extent_end + 1; |
360 | } else if (extent_start > start && extent_start < end) { | |
361 | size = extent_start - start; | |
817d52f8 | 362 | total_added += size; |
ea6a478e JB |
363 | ret = btrfs_add_free_space(block_group, start, |
364 | size); | |
79787eaa | 365 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
366 | start = extent_end + 1; |
367 | } else { | |
368 | break; | |
369 | } | |
370 | } | |
371 | ||
372 | if (start < end) { | |
373 | size = end - start; | |
817d52f8 | 374 | total_added += size; |
ea6a478e | 375 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 376 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
377 | } |
378 | ||
817d52f8 | 379 | return total_added; |
0f9dd46c JB |
380 | } |
381 | ||
73fa48b6 | 382 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 383 | { |
0b246afa JM |
384 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
385 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
386 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 387 | struct btrfs_path *path; |
5f39d397 | 388 | struct extent_buffer *leaf; |
11833d66 | 389 | struct btrfs_key key; |
817d52f8 | 390 | u64 total_found = 0; |
11833d66 YZ |
391 | u64 last = 0; |
392 | u32 nritems; | |
73fa48b6 | 393 | int ret; |
d0bd4560 | 394 | bool wakeup = true; |
f510cfec | 395 | |
e37c9e69 CM |
396 | path = btrfs_alloc_path(); |
397 | if (!path) | |
73fa48b6 | 398 | return -ENOMEM; |
7d7d6068 | 399 | |
817d52f8 | 400 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 401 | |
d0bd4560 JB |
402 | #ifdef CONFIG_BTRFS_DEBUG |
403 | /* | |
404 | * If we're fragmenting we don't want to make anybody think we can | |
405 | * allocate from this block group until we've had a chance to fragment | |
406 | * the free space. | |
407 | */ | |
2ff7e61e | 408 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
409 | wakeup = false; |
410 | #endif | |
5cd57b2c | 411 | /* |
817d52f8 JB |
412 | * We don't want to deadlock with somebody trying to allocate a new |
413 | * extent for the extent root while also trying to search the extent | |
414 | * root to add free space. So we skip locking and search the commit | |
415 | * root, since its read-only | |
5cd57b2c CM |
416 | */ |
417 | path->skip_locking = 1; | |
817d52f8 | 418 | path->search_commit_root = 1; |
e4058b54 | 419 | path->reada = READA_FORWARD; |
817d52f8 | 420 | |
e4404d6e | 421 | key.objectid = last; |
e37c9e69 | 422 | key.offset = 0; |
11833d66 | 423 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 424 | |
52ee28d2 | 425 | next: |
11833d66 | 426 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 427 | if (ret < 0) |
73fa48b6 | 428 | goto out; |
a512bbf8 | 429 | |
11833d66 YZ |
430 | leaf = path->nodes[0]; |
431 | nritems = btrfs_header_nritems(leaf); | |
432 | ||
d397712b | 433 | while (1) { |
7841cb28 | 434 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 435 | last = (u64)-1; |
817d52f8 | 436 | break; |
f25784b3 | 437 | } |
817d52f8 | 438 | |
11833d66 YZ |
439 | if (path->slots[0] < nritems) { |
440 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
441 | } else { | |
442 | ret = find_next_key(path, 0, &key); | |
443 | if (ret) | |
e37c9e69 | 444 | break; |
817d52f8 | 445 | |
c9ea7b24 | 446 | if (need_resched() || |
9e351cc8 | 447 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
448 | if (wakeup) |
449 | caching_ctl->progress = last; | |
ff5714cc | 450 | btrfs_release_path(path); |
9e351cc8 | 451 | up_read(&fs_info->commit_root_sem); |
589d8ade | 452 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 453 | cond_resched(); |
73fa48b6 OS |
454 | mutex_lock(&caching_ctl->mutex); |
455 | down_read(&fs_info->commit_root_sem); | |
456 | goto next; | |
589d8ade | 457 | } |
0a3896d0 JB |
458 | |
459 | ret = btrfs_next_leaf(extent_root, path); | |
460 | if (ret < 0) | |
73fa48b6 | 461 | goto out; |
0a3896d0 JB |
462 | if (ret) |
463 | break; | |
589d8ade JB |
464 | leaf = path->nodes[0]; |
465 | nritems = btrfs_header_nritems(leaf); | |
466 | continue; | |
11833d66 | 467 | } |
817d52f8 | 468 | |
52ee28d2 LB |
469 | if (key.objectid < last) { |
470 | key.objectid = last; | |
471 | key.offset = 0; | |
472 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
473 | ||
d0bd4560 JB |
474 | if (wakeup) |
475 | caching_ctl->progress = last; | |
52ee28d2 LB |
476 | btrfs_release_path(path); |
477 | goto next; | |
478 | } | |
479 | ||
11833d66 YZ |
480 | if (key.objectid < block_group->key.objectid) { |
481 | path->slots[0]++; | |
817d52f8 | 482 | continue; |
e37c9e69 | 483 | } |
0f9dd46c | 484 | |
e37c9e69 | 485 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 486 | block_group->key.offset) |
e37c9e69 | 487 | break; |
7d7d6068 | 488 | |
3173a18f JB |
489 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
490 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
4457c1c7 | 491 | total_found += add_new_free_space(block_group, last, |
817d52f8 | 492 | key.objectid); |
3173a18f JB |
493 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
494 | last = key.objectid + | |
da17066c | 495 | fs_info->nodesize; |
3173a18f JB |
496 | else |
497 | last = key.objectid + key.offset; | |
817d52f8 | 498 | |
73fa48b6 | 499 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 500 | total_found = 0; |
d0bd4560 JB |
501 | if (wakeup) |
502 | wake_up(&caching_ctl->wait); | |
11833d66 | 503 | } |
817d52f8 | 504 | } |
e37c9e69 CM |
505 | path->slots[0]++; |
506 | } | |
817d52f8 | 507 | ret = 0; |
e37c9e69 | 508 | |
4457c1c7 | 509 | total_found += add_new_free_space(block_group, last, |
817d52f8 JB |
510 | block_group->key.objectid + |
511 | block_group->key.offset); | |
11833d66 | 512 | caching_ctl->progress = (u64)-1; |
817d52f8 | 513 | |
73fa48b6 OS |
514 | out: |
515 | btrfs_free_path(path); | |
516 | return ret; | |
517 | } | |
518 | ||
519 | static noinline void caching_thread(struct btrfs_work *work) | |
520 | { | |
521 | struct btrfs_block_group_cache *block_group; | |
522 | struct btrfs_fs_info *fs_info; | |
523 | struct btrfs_caching_control *caching_ctl; | |
524 | int ret; | |
525 | ||
526 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
527 | block_group = caching_ctl->block_group; | |
528 | fs_info = block_group->fs_info; | |
529 | ||
530 | mutex_lock(&caching_ctl->mutex); | |
531 | down_read(&fs_info->commit_root_sem); | |
532 | ||
1e144fb8 OS |
533 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
534 | ret = load_free_space_tree(caching_ctl); | |
535 | else | |
536 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 537 | |
817d52f8 | 538 | spin_lock(&block_group->lock); |
11833d66 | 539 | block_group->caching_ctl = NULL; |
73fa48b6 | 540 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 541 | spin_unlock(&block_group->lock); |
0f9dd46c | 542 | |
d0bd4560 | 543 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 544 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
545 | u64 bytes_used; |
546 | ||
547 | spin_lock(&block_group->space_info->lock); | |
548 | spin_lock(&block_group->lock); | |
549 | bytes_used = block_group->key.offset - | |
550 | btrfs_block_group_used(&block_group->item); | |
551 | block_group->space_info->bytes_used += bytes_used >> 1; | |
552 | spin_unlock(&block_group->lock); | |
553 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 554 | fragment_free_space(block_group); |
d0bd4560 JB |
555 | } |
556 | #endif | |
557 | ||
558 | caching_ctl->progress = (u64)-1; | |
11833d66 | 559 | |
9e351cc8 | 560 | up_read(&fs_info->commit_root_sem); |
2ff7e61e | 561 | free_excluded_extents(fs_info, block_group); |
11833d66 | 562 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 563 | |
11833d66 YZ |
564 | wake_up(&caching_ctl->wait); |
565 | ||
566 | put_caching_control(caching_ctl); | |
11dfe35a | 567 | btrfs_put_block_group(block_group); |
817d52f8 JB |
568 | } |
569 | ||
9d66e233 | 570 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 571 | int load_cache_only) |
817d52f8 | 572 | { |
291c7d2f | 573 | DEFINE_WAIT(wait); |
11833d66 YZ |
574 | struct btrfs_fs_info *fs_info = cache->fs_info; |
575 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
576 | int ret = 0; |
577 | ||
291c7d2f | 578 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
579 | if (!caching_ctl) |
580 | return -ENOMEM; | |
291c7d2f JB |
581 | |
582 | INIT_LIST_HEAD(&caching_ctl->list); | |
583 | mutex_init(&caching_ctl->mutex); | |
584 | init_waitqueue_head(&caching_ctl->wait); | |
585 | caching_ctl->block_group = cache; | |
586 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 587 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
588 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
589 | caching_thread, NULL, NULL); | |
291c7d2f JB |
590 | |
591 | spin_lock(&cache->lock); | |
592 | /* | |
593 | * This should be a rare occasion, but this could happen I think in the | |
594 | * case where one thread starts to load the space cache info, and then | |
595 | * some other thread starts a transaction commit which tries to do an | |
596 | * allocation while the other thread is still loading the space cache | |
597 | * info. The previous loop should have kept us from choosing this block | |
598 | * group, but if we've moved to the state where we will wait on caching | |
599 | * block groups we need to first check if we're doing a fast load here, | |
600 | * so we can wait for it to finish, otherwise we could end up allocating | |
601 | * from a block group who's cache gets evicted for one reason or | |
602 | * another. | |
603 | */ | |
604 | while (cache->cached == BTRFS_CACHE_FAST) { | |
605 | struct btrfs_caching_control *ctl; | |
606 | ||
607 | ctl = cache->caching_ctl; | |
1e4f4714 | 608 | refcount_inc(&ctl->count); |
291c7d2f JB |
609 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
610 | spin_unlock(&cache->lock); | |
611 | ||
612 | schedule(); | |
613 | ||
614 | finish_wait(&ctl->wait, &wait); | |
615 | put_caching_control(ctl); | |
616 | spin_lock(&cache->lock); | |
617 | } | |
618 | ||
619 | if (cache->cached != BTRFS_CACHE_NO) { | |
620 | spin_unlock(&cache->lock); | |
621 | kfree(caching_ctl); | |
11833d66 | 622 | return 0; |
291c7d2f JB |
623 | } |
624 | WARN_ON(cache->caching_ctl); | |
625 | cache->caching_ctl = caching_ctl; | |
626 | cache->cached = BTRFS_CACHE_FAST; | |
627 | spin_unlock(&cache->lock); | |
11833d66 | 628 | |
d8953d69 | 629 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cb83b7b8 | 630 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
631 | ret = load_free_space_cache(fs_info, cache); |
632 | ||
633 | spin_lock(&cache->lock); | |
634 | if (ret == 1) { | |
291c7d2f | 635 | cache->caching_ctl = NULL; |
9d66e233 JB |
636 | cache->cached = BTRFS_CACHE_FINISHED; |
637 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 638 | caching_ctl->progress = (u64)-1; |
9d66e233 | 639 | } else { |
291c7d2f JB |
640 | if (load_cache_only) { |
641 | cache->caching_ctl = NULL; | |
642 | cache->cached = BTRFS_CACHE_NO; | |
643 | } else { | |
644 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 645 | cache->has_caching_ctl = 1; |
291c7d2f | 646 | } |
9d66e233 JB |
647 | } |
648 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
649 | #ifdef CONFIG_BTRFS_DEBUG |
650 | if (ret == 1 && | |
2ff7e61e | 651 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
652 | u64 bytes_used; |
653 | ||
654 | spin_lock(&cache->space_info->lock); | |
655 | spin_lock(&cache->lock); | |
656 | bytes_used = cache->key.offset - | |
657 | btrfs_block_group_used(&cache->item); | |
658 | cache->space_info->bytes_used += bytes_used >> 1; | |
659 | spin_unlock(&cache->lock); | |
660 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 661 | fragment_free_space(cache); |
d0bd4560 JB |
662 | } |
663 | #endif | |
cb83b7b8 JB |
664 | mutex_unlock(&caching_ctl->mutex); |
665 | ||
291c7d2f | 666 | wake_up(&caching_ctl->wait); |
3c14874a | 667 | if (ret == 1) { |
291c7d2f | 668 | put_caching_control(caching_ctl); |
2ff7e61e | 669 | free_excluded_extents(fs_info, cache); |
9d66e233 | 670 | return 0; |
3c14874a | 671 | } |
291c7d2f JB |
672 | } else { |
673 | /* | |
1e144fb8 OS |
674 | * We're either using the free space tree or no caching at all. |
675 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
676 | */ |
677 | spin_lock(&cache->lock); | |
678 | if (load_cache_only) { | |
679 | cache->caching_ctl = NULL; | |
680 | cache->cached = BTRFS_CACHE_NO; | |
681 | } else { | |
682 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 683 | cache->has_caching_ctl = 1; |
291c7d2f JB |
684 | } |
685 | spin_unlock(&cache->lock); | |
686 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
687 | } |
688 | ||
291c7d2f JB |
689 | if (load_cache_only) { |
690 | put_caching_control(caching_ctl); | |
11833d66 | 691 | return 0; |
817d52f8 | 692 | } |
817d52f8 | 693 | |
9e351cc8 | 694 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 695 | refcount_inc(&caching_ctl->count); |
11833d66 | 696 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 697 | up_write(&fs_info->commit_root_sem); |
11833d66 | 698 | |
11dfe35a | 699 | btrfs_get_block_group(cache); |
11833d66 | 700 | |
e66f0bb1 | 701 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 702 | |
ef8bbdfe | 703 | return ret; |
e37c9e69 CM |
704 | } |
705 | ||
0f9dd46c JB |
706 | /* |
707 | * return the block group that starts at or after bytenr | |
708 | */ | |
d397712b CM |
709 | static struct btrfs_block_group_cache * |
710 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 711 | { |
e2c89907 | 712 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
713 | } |
714 | ||
0f9dd46c | 715 | /* |
9f55684c | 716 | * return the block group that contains the given bytenr |
0f9dd46c | 717 | */ |
d397712b CM |
718 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
719 | struct btrfs_fs_info *info, | |
720 | u64 bytenr) | |
be744175 | 721 | { |
e2c89907 | 722 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 723 | } |
0b86a832 | 724 | |
0f9dd46c JB |
725 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
726 | u64 flags) | |
6324fbf3 | 727 | { |
0f9dd46c | 728 | struct list_head *head = &info->space_info; |
0f9dd46c | 729 | struct btrfs_space_info *found; |
4184ea7f | 730 | |
52ba6929 | 731 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 732 | |
4184ea7f CM |
733 | rcu_read_lock(); |
734 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 735 | if (found->flags & flags) { |
4184ea7f | 736 | rcu_read_unlock(); |
0f9dd46c | 737 | return found; |
4184ea7f | 738 | } |
0f9dd46c | 739 | } |
4184ea7f | 740 | rcu_read_unlock(); |
0f9dd46c | 741 | return NULL; |
6324fbf3 CM |
742 | } |
743 | ||
0d9f824d | 744 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes, |
29d2b84c | 745 | bool metadata, u64 root_objectid) |
0d9f824d OS |
746 | { |
747 | struct btrfs_space_info *space_info; | |
748 | u64 flags; | |
749 | ||
29d2b84c | 750 | if (metadata) { |
0d9f824d OS |
751 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) |
752 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
753 | else | |
754 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
755 | } else { | |
756 | flags = BTRFS_BLOCK_GROUP_DATA; | |
757 | } | |
758 | ||
759 | space_info = __find_space_info(fs_info, flags); | |
55e8196a | 760 | ASSERT(space_info); |
0d9f824d OS |
761 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); |
762 | } | |
763 | ||
4184ea7f CM |
764 | /* |
765 | * after adding space to the filesystem, we need to clear the full flags | |
766 | * on all the space infos. | |
767 | */ | |
768 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
769 | { | |
770 | struct list_head *head = &info->space_info; | |
771 | struct btrfs_space_info *found; | |
772 | ||
773 | rcu_read_lock(); | |
774 | list_for_each_entry_rcu(found, head, list) | |
775 | found->full = 0; | |
776 | rcu_read_unlock(); | |
777 | } | |
778 | ||
1a4ed8fd | 779 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 780 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
781 | { |
782 | int ret; | |
783 | struct btrfs_key key; | |
31840ae1 | 784 | struct btrfs_path *path; |
e02119d5 | 785 | |
31840ae1 | 786 | path = btrfs_alloc_path(); |
d8926bb3 MF |
787 | if (!path) |
788 | return -ENOMEM; | |
789 | ||
e02119d5 CM |
790 | key.objectid = start; |
791 | key.offset = len; | |
3173a18f | 792 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 793 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 794 | btrfs_free_path(path); |
7bb86316 CM |
795 | return ret; |
796 | } | |
797 | ||
a22285a6 | 798 | /* |
3173a18f | 799 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
800 | * |
801 | * the head node for delayed ref is used to store the sum of all the | |
802 | * reference count modifications queued up in the rbtree. the head | |
803 | * node may also store the extent flags to set. This way you can check | |
804 | * to see what the reference count and extent flags would be if all of | |
805 | * the delayed refs are not processed. | |
806 | */ | |
807 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 808 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 809 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
810 | { |
811 | struct btrfs_delayed_ref_head *head; | |
812 | struct btrfs_delayed_ref_root *delayed_refs; | |
813 | struct btrfs_path *path; | |
814 | struct btrfs_extent_item *ei; | |
815 | struct extent_buffer *leaf; | |
816 | struct btrfs_key key; | |
817 | u32 item_size; | |
818 | u64 num_refs; | |
819 | u64 extent_flags; | |
820 | int ret; | |
821 | ||
3173a18f JB |
822 | /* |
823 | * If we don't have skinny metadata, don't bother doing anything | |
824 | * different | |
825 | */ | |
0b246afa JM |
826 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
827 | offset = fs_info->nodesize; | |
3173a18f JB |
828 | metadata = 0; |
829 | } | |
830 | ||
a22285a6 YZ |
831 | path = btrfs_alloc_path(); |
832 | if (!path) | |
833 | return -ENOMEM; | |
834 | ||
a22285a6 YZ |
835 | if (!trans) { |
836 | path->skip_locking = 1; | |
837 | path->search_commit_root = 1; | |
838 | } | |
639eefc8 FDBM |
839 | |
840 | search_again: | |
841 | key.objectid = bytenr; | |
842 | key.offset = offset; | |
843 | if (metadata) | |
844 | key.type = BTRFS_METADATA_ITEM_KEY; | |
845 | else | |
846 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
847 | ||
0b246afa | 848 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
849 | if (ret < 0) |
850 | goto out_free; | |
851 | ||
3173a18f | 852 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
853 | if (path->slots[0]) { |
854 | path->slots[0]--; | |
855 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
856 | path->slots[0]); | |
857 | if (key.objectid == bytenr && | |
858 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 859 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
860 | ret = 0; |
861 | } | |
3173a18f JB |
862 | } |
863 | ||
a22285a6 YZ |
864 | if (ret == 0) { |
865 | leaf = path->nodes[0]; | |
866 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
867 | if (item_size >= sizeof(*ei)) { | |
868 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
869 | struct btrfs_extent_item); | |
870 | num_refs = btrfs_extent_refs(leaf, ei); | |
871 | extent_flags = btrfs_extent_flags(leaf, ei); | |
872 | } else { | |
873 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
874 | struct btrfs_extent_item_v0 *ei0; | |
875 | BUG_ON(item_size != sizeof(*ei0)); | |
876 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
877 | struct btrfs_extent_item_v0); | |
878 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
879 | /* FIXME: this isn't correct for data */ | |
880 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
881 | #else | |
882 | BUG(); | |
883 | #endif | |
884 | } | |
885 | BUG_ON(num_refs == 0); | |
886 | } else { | |
887 | num_refs = 0; | |
888 | extent_flags = 0; | |
889 | ret = 0; | |
890 | } | |
891 | ||
892 | if (!trans) | |
893 | goto out; | |
894 | ||
895 | delayed_refs = &trans->transaction->delayed_refs; | |
896 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 897 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
898 | if (head) { |
899 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 900 | refcount_inc(&head->refs); |
a22285a6 YZ |
901 | spin_unlock(&delayed_refs->lock); |
902 | ||
b3b4aa74 | 903 | btrfs_release_path(path); |
a22285a6 | 904 | |
8cc33e5c DS |
905 | /* |
906 | * Mutex was contended, block until it's released and try | |
907 | * again | |
908 | */ | |
a22285a6 YZ |
909 | mutex_lock(&head->mutex); |
910 | mutex_unlock(&head->mutex); | |
d278850e | 911 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 912 | goto search_again; |
a22285a6 | 913 | } |
d7df2c79 | 914 | spin_lock(&head->lock); |
a22285a6 YZ |
915 | if (head->extent_op && head->extent_op->update_flags) |
916 | extent_flags |= head->extent_op->flags_to_set; | |
917 | else | |
918 | BUG_ON(num_refs == 0); | |
919 | ||
d278850e | 920 | num_refs += head->ref_mod; |
d7df2c79 | 921 | spin_unlock(&head->lock); |
a22285a6 YZ |
922 | mutex_unlock(&head->mutex); |
923 | } | |
924 | spin_unlock(&delayed_refs->lock); | |
925 | out: | |
926 | WARN_ON(num_refs == 0); | |
927 | if (refs) | |
928 | *refs = num_refs; | |
929 | if (flags) | |
930 | *flags = extent_flags; | |
931 | out_free: | |
932 | btrfs_free_path(path); | |
933 | return ret; | |
934 | } | |
935 | ||
d8d5f3e1 CM |
936 | /* |
937 | * Back reference rules. Back refs have three main goals: | |
938 | * | |
939 | * 1) differentiate between all holders of references to an extent so that | |
940 | * when a reference is dropped we can make sure it was a valid reference | |
941 | * before freeing the extent. | |
942 | * | |
943 | * 2) Provide enough information to quickly find the holders of an extent | |
944 | * if we notice a given block is corrupted or bad. | |
945 | * | |
946 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
947 | * maintenance. This is actually the same as #2, but with a slightly | |
948 | * different use case. | |
949 | * | |
5d4f98a2 YZ |
950 | * There are two kinds of back refs. The implicit back refs is optimized |
951 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
952 | * back refs of this kind provide information about the block's owner tree | |
953 | * and the pointer's key. These information allow us to find the block by | |
954 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
955 | * referenced by their owner trees. The location of tree block is recorded | |
956 | * in the back refs. Actually the full back refs is generic, and can be | |
957 | * used in all cases the implicit back refs is used. The major shortcoming | |
958 | * of the full back refs is its overhead. Every time a tree block gets | |
959 | * COWed, we have to update back refs entry for all pointers in it. | |
960 | * | |
961 | * For a newly allocated tree block, we use implicit back refs for | |
962 | * pointers in it. This means most tree related operations only involve | |
963 | * implicit back refs. For a tree block created in old transaction, the | |
964 | * only way to drop a reference to it is COW it. So we can detect the | |
965 | * event that tree block loses its owner tree's reference and do the | |
966 | * back refs conversion. | |
967 | * | |
01327610 | 968 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
969 | * |
970 | * The reference count of the block is one and the tree is the block's | |
971 | * owner tree. Nothing to do in this case. | |
972 | * | |
973 | * The reference count of the block is one and the tree is not the | |
974 | * block's owner tree. In this case, full back refs is used for pointers | |
975 | * in the block. Remove these full back refs, add implicit back refs for | |
976 | * every pointers in the new block. | |
977 | * | |
978 | * The reference count of the block is greater than one and the tree is | |
979 | * the block's owner tree. In this case, implicit back refs is used for | |
980 | * pointers in the block. Add full back refs for every pointers in the | |
981 | * block, increase lower level extents' reference counts. The original | |
982 | * implicit back refs are entailed to the new block. | |
983 | * | |
984 | * The reference count of the block is greater than one and the tree is | |
985 | * not the block's owner tree. Add implicit back refs for every pointer in | |
986 | * the new block, increase lower level extents' reference count. | |
987 | * | |
988 | * Back Reference Key composing: | |
989 | * | |
990 | * The key objectid corresponds to the first byte in the extent, | |
991 | * The key type is used to differentiate between types of back refs. | |
992 | * There are different meanings of the key offset for different types | |
993 | * of back refs. | |
994 | * | |
d8d5f3e1 CM |
995 | * File extents can be referenced by: |
996 | * | |
997 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 998 | * - different files inside a single subvolume |
d8d5f3e1 CM |
999 | * - different offsets inside a file (bookend extents in file.c) |
1000 | * | |
5d4f98a2 | 1001 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1002 | * |
1003 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1004 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1005 | * - original offset in the file |
1006 | * - how many bookend extents | |
d8d5f3e1 | 1007 | * |
5d4f98a2 YZ |
1008 | * The key offset for the implicit back refs is hash of the first |
1009 | * three fields. | |
d8d5f3e1 | 1010 | * |
5d4f98a2 | 1011 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1012 | * |
5d4f98a2 | 1013 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1014 | * |
5d4f98a2 YZ |
1015 | * The key offset for the implicit back refs is the first byte of |
1016 | * the tree leaf | |
d8d5f3e1 | 1017 | * |
5d4f98a2 YZ |
1018 | * When a file extent is allocated, The implicit back refs is used. |
1019 | * the fields are filled in: | |
d8d5f3e1 | 1020 | * |
5d4f98a2 | 1021 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1022 | * |
5d4f98a2 YZ |
1023 | * When a file extent is removed file truncation, we find the |
1024 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1025 | * |
5d4f98a2 | 1026 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1027 | * |
5d4f98a2 | 1028 | * Btree extents can be referenced by: |
d8d5f3e1 | 1029 | * |
5d4f98a2 | 1030 | * - Different subvolumes |
d8d5f3e1 | 1031 | * |
5d4f98a2 YZ |
1032 | * Both the implicit back refs and the full back refs for tree blocks |
1033 | * only consist of key. The key offset for the implicit back refs is | |
1034 | * objectid of block's owner tree. The key offset for the full back refs | |
1035 | * is the first byte of parent block. | |
d8d5f3e1 | 1036 | * |
5d4f98a2 YZ |
1037 | * When implicit back refs is used, information about the lowest key and |
1038 | * level of the tree block are required. These information are stored in | |
1039 | * tree block info structure. | |
d8d5f3e1 | 1040 | */ |
31840ae1 | 1041 | |
5d4f98a2 YZ |
1042 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1043 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
87bde3cd | 1044 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1045 | struct btrfs_path *path, |
1046 | u64 owner, u32 extra_size) | |
7bb86316 | 1047 | { |
87bde3cd | 1048 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1049 | struct btrfs_extent_item *item; |
1050 | struct btrfs_extent_item_v0 *ei0; | |
1051 | struct btrfs_extent_ref_v0 *ref0; | |
1052 | struct btrfs_tree_block_info *bi; | |
1053 | struct extent_buffer *leaf; | |
7bb86316 | 1054 | struct btrfs_key key; |
5d4f98a2 YZ |
1055 | struct btrfs_key found_key; |
1056 | u32 new_size = sizeof(*item); | |
1057 | u64 refs; | |
1058 | int ret; | |
1059 | ||
1060 | leaf = path->nodes[0]; | |
1061 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1062 | ||
1063 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1064 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1065 | struct btrfs_extent_item_v0); | |
1066 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1067 | ||
1068 | if (owner == (u64)-1) { | |
1069 | while (1) { | |
1070 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1071 | ret = btrfs_next_leaf(root, path); | |
1072 | if (ret < 0) | |
1073 | return ret; | |
79787eaa | 1074 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1075 | leaf = path->nodes[0]; |
1076 | } | |
1077 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1078 | path->slots[0]); | |
1079 | BUG_ON(key.objectid != found_key.objectid); | |
1080 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1081 | path->slots[0]++; | |
1082 | continue; | |
1083 | } | |
1084 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1085 | struct btrfs_extent_ref_v0); | |
1086 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1087 | break; | |
1088 | } | |
1089 | } | |
b3b4aa74 | 1090 | btrfs_release_path(path); |
5d4f98a2 YZ |
1091 | |
1092 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1093 | new_size += sizeof(*bi); | |
1094 | ||
1095 | new_size -= sizeof(*ei0); | |
1096 | ret = btrfs_search_slot(trans, root, &key, path, | |
1097 | new_size + extra_size, 1); | |
1098 | if (ret < 0) | |
1099 | return ret; | |
79787eaa | 1100 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1101 | |
87bde3cd | 1102 | btrfs_extend_item(fs_info, path, new_size); |
5d4f98a2 YZ |
1103 | |
1104 | leaf = path->nodes[0]; | |
1105 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1106 | btrfs_set_extent_refs(leaf, item, refs); | |
1107 | /* FIXME: get real generation */ | |
1108 | btrfs_set_extent_generation(leaf, item, 0); | |
1109 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1110 | btrfs_set_extent_flags(leaf, item, | |
1111 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1112 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1113 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1114 | /* FIXME: get first key of the block */ | |
b159fa28 | 1115 | memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi)); |
5d4f98a2 YZ |
1116 | btrfs_set_tree_block_level(leaf, bi, (int)owner); |
1117 | } else { | |
1118 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1119 | } | |
1120 | btrfs_mark_buffer_dirty(leaf); | |
1121 | return 0; | |
1122 | } | |
1123 | #endif | |
1124 | ||
167ce953 LB |
1125 | /* |
1126 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
1127 | * is_data == BTRFS_REF_TYPE_DATA, data type is requried, | |
1128 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. | |
1129 | */ | |
1130 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
1131 | struct btrfs_extent_inline_ref *iref, | |
1132 | enum btrfs_inline_ref_type is_data) | |
1133 | { | |
1134 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1135 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1136 | |
1137 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1138 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1139 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1140 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1141 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1142 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1143 | return type; |
64ecdb64 LB |
1144 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1145 | ASSERT(eb->fs_info); | |
1146 | /* | |
1147 | * Every shared one has parent tree | |
1148 | * block, which must be aligned to | |
1149 | * nodesize. | |
1150 | */ | |
1151 | if (offset && | |
1152 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1153 | return type; | |
1154 | } | |
167ce953 | 1155 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1156 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1157 | return type; |
64ecdb64 LB |
1158 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1159 | ASSERT(eb->fs_info); | |
1160 | /* | |
1161 | * Every shared one has parent tree | |
1162 | * block, which must be aligned to | |
1163 | * nodesize. | |
1164 | */ | |
1165 | if (offset && | |
1166 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1167 | return type; | |
1168 | } | |
167ce953 LB |
1169 | } else { |
1170 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1171 | return type; | |
1172 | } | |
1173 | } | |
1174 | ||
1175 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1176 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1177 | eb->start, type); | |
1178 | WARN_ON(1); | |
1179 | ||
1180 | return BTRFS_REF_TYPE_INVALID; | |
1181 | } | |
1182 | ||
5d4f98a2 YZ |
1183 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1184 | { | |
1185 | u32 high_crc = ~(u32)0; | |
1186 | u32 low_crc = ~(u32)0; | |
1187 | __le64 lenum; | |
1188 | ||
1189 | lenum = cpu_to_le64(root_objectid); | |
9678c543 | 1190 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1191 | lenum = cpu_to_le64(owner); |
9678c543 | 1192 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1193 | lenum = cpu_to_le64(offset); |
9678c543 | 1194 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1195 | |
1196 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1197 | } | |
1198 | ||
1199 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1200 | struct btrfs_extent_data_ref *ref) | |
1201 | { | |
1202 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1203 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1204 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1205 | } | |
1206 | ||
1207 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1208 | struct btrfs_extent_data_ref *ref, | |
1209 | u64 root_objectid, u64 owner, u64 offset) | |
1210 | { | |
1211 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1212 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1213 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1214 | return 0; | |
1215 | return 1; | |
1216 | } | |
1217 | ||
1218 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1219 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1220 | struct btrfs_path *path, |
1221 | u64 bytenr, u64 parent, | |
1222 | u64 root_objectid, | |
1223 | u64 owner, u64 offset) | |
1224 | { | |
87bde3cd | 1225 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1226 | struct btrfs_key key; |
1227 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1228 | struct extent_buffer *leaf; |
5d4f98a2 | 1229 | u32 nritems; |
74493f7a | 1230 | int ret; |
5d4f98a2 YZ |
1231 | int recow; |
1232 | int err = -ENOENT; | |
74493f7a | 1233 | |
31840ae1 | 1234 | key.objectid = bytenr; |
5d4f98a2 YZ |
1235 | if (parent) { |
1236 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1237 | key.offset = parent; | |
1238 | } else { | |
1239 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1240 | key.offset = hash_extent_data_ref(root_objectid, | |
1241 | owner, offset); | |
1242 | } | |
1243 | again: | |
1244 | recow = 0; | |
1245 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1246 | if (ret < 0) { | |
1247 | err = ret; | |
1248 | goto fail; | |
1249 | } | |
31840ae1 | 1250 | |
5d4f98a2 YZ |
1251 | if (parent) { |
1252 | if (!ret) | |
1253 | return 0; | |
1254 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1255 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1256 | btrfs_release_path(path); |
5d4f98a2 YZ |
1257 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1258 | if (ret < 0) { | |
1259 | err = ret; | |
1260 | goto fail; | |
1261 | } | |
1262 | if (!ret) | |
1263 | return 0; | |
1264 | #endif | |
1265 | goto fail; | |
31840ae1 ZY |
1266 | } |
1267 | ||
1268 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1269 | nritems = btrfs_header_nritems(leaf); |
1270 | while (1) { | |
1271 | if (path->slots[0] >= nritems) { | |
1272 | ret = btrfs_next_leaf(root, path); | |
1273 | if (ret < 0) | |
1274 | err = ret; | |
1275 | if (ret) | |
1276 | goto fail; | |
1277 | ||
1278 | leaf = path->nodes[0]; | |
1279 | nritems = btrfs_header_nritems(leaf); | |
1280 | recow = 1; | |
1281 | } | |
1282 | ||
1283 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1284 | if (key.objectid != bytenr || | |
1285 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1286 | goto fail; | |
1287 | ||
1288 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1289 | struct btrfs_extent_data_ref); | |
1290 | ||
1291 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1292 | owner, offset)) { | |
1293 | if (recow) { | |
b3b4aa74 | 1294 | btrfs_release_path(path); |
5d4f98a2 YZ |
1295 | goto again; |
1296 | } | |
1297 | err = 0; | |
1298 | break; | |
1299 | } | |
1300 | path->slots[0]++; | |
31840ae1 | 1301 | } |
5d4f98a2 YZ |
1302 | fail: |
1303 | return err; | |
31840ae1 ZY |
1304 | } |
1305 | ||
5d4f98a2 | 1306 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1307 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1308 | struct btrfs_path *path, |
1309 | u64 bytenr, u64 parent, | |
1310 | u64 root_objectid, u64 owner, | |
1311 | u64 offset, int refs_to_add) | |
31840ae1 | 1312 | { |
87bde3cd | 1313 | struct btrfs_root *root = fs_info->extent_root; |
31840ae1 ZY |
1314 | struct btrfs_key key; |
1315 | struct extent_buffer *leaf; | |
5d4f98a2 | 1316 | u32 size; |
31840ae1 ZY |
1317 | u32 num_refs; |
1318 | int ret; | |
74493f7a | 1319 | |
74493f7a | 1320 | key.objectid = bytenr; |
5d4f98a2 YZ |
1321 | if (parent) { |
1322 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1323 | key.offset = parent; | |
1324 | size = sizeof(struct btrfs_shared_data_ref); | |
1325 | } else { | |
1326 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1327 | key.offset = hash_extent_data_ref(root_objectid, | |
1328 | owner, offset); | |
1329 | size = sizeof(struct btrfs_extent_data_ref); | |
1330 | } | |
74493f7a | 1331 | |
5d4f98a2 YZ |
1332 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1333 | if (ret && ret != -EEXIST) | |
1334 | goto fail; | |
1335 | ||
1336 | leaf = path->nodes[0]; | |
1337 | if (parent) { | |
1338 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1339 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1340 | struct btrfs_shared_data_ref); |
1341 | if (ret == 0) { | |
1342 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1343 | } else { | |
1344 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1345 | num_refs += refs_to_add; | |
1346 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1347 | } |
5d4f98a2 YZ |
1348 | } else { |
1349 | struct btrfs_extent_data_ref *ref; | |
1350 | while (ret == -EEXIST) { | |
1351 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1352 | struct btrfs_extent_data_ref); | |
1353 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1354 | owner, offset)) | |
1355 | break; | |
b3b4aa74 | 1356 | btrfs_release_path(path); |
5d4f98a2 YZ |
1357 | key.offset++; |
1358 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1359 | size); | |
1360 | if (ret && ret != -EEXIST) | |
1361 | goto fail; | |
31840ae1 | 1362 | |
5d4f98a2 YZ |
1363 | leaf = path->nodes[0]; |
1364 | } | |
1365 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1366 | struct btrfs_extent_data_ref); | |
1367 | if (ret == 0) { | |
1368 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1369 | root_objectid); | |
1370 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1371 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1372 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1373 | } else { | |
1374 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1375 | num_refs += refs_to_add; | |
1376 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1377 | } |
31840ae1 | 1378 | } |
5d4f98a2 YZ |
1379 | btrfs_mark_buffer_dirty(leaf); |
1380 | ret = 0; | |
1381 | fail: | |
b3b4aa74 | 1382 | btrfs_release_path(path); |
7bb86316 | 1383 | return ret; |
74493f7a CM |
1384 | } |
1385 | ||
5d4f98a2 | 1386 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1387 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 1388 | struct btrfs_path *path, |
fcebe456 | 1389 | int refs_to_drop, int *last_ref) |
31840ae1 | 1390 | { |
5d4f98a2 YZ |
1391 | struct btrfs_key key; |
1392 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1393 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1394 | struct extent_buffer *leaf; |
5d4f98a2 | 1395 | u32 num_refs = 0; |
31840ae1 ZY |
1396 | int ret = 0; |
1397 | ||
1398 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1399 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1400 | ||
1401 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1402 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1403 | struct btrfs_extent_data_ref); | |
1404 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1405 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1406 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1407 | struct btrfs_shared_data_ref); | |
1408 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1409 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1410 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1411 | struct btrfs_extent_ref_v0 *ref0; | |
1412 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1413 | struct btrfs_extent_ref_v0); | |
1414 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1415 | #endif | |
1416 | } else { | |
1417 | BUG(); | |
1418 | } | |
1419 | ||
56bec294 CM |
1420 | BUG_ON(num_refs < refs_to_drop); |
1421 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1422 | |
31840ae1 | 1423 | if (num_refs == 0) { |
87bde3cd | 1424 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
fcebe456 | 1425 | *last_ref = 1; |
31840ae1 | 1426 | } else { |
5d4f98a2 YZ |
1427 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1428 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1429 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1430 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1431 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1432 | else { | |
1433 | struct btrfs_extent_ref_v0 *ref0; | |
1434 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1435 | struct btrfs_extent_ref_v0); | |
1436 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1437 | } | |
1438 | #endif | |
31840ae1 ZY |
1439 | btrfs_mark_buffer_dirty(leaf); |
1440 | } | |
31840ae1 ZY |
1441 | return ret; |
1442 | } | |
1443 | ||
9ed0dea0 | 1444 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1445 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1446 | { |
5d4f98a2 YZ |
1447 | struct btrfs_key key; |
1448 | struct extent_buffer *leaf; | |
1449 | struct btrfs_extent_data_ref *ref1; | |
1450 | struct btrfs_shared_data_ref *ref2; | |
1451 | u32 num_refs = 0; | |
3de28d57 | 1452 | int type; |
5d4f98a2 YZ |
1453 | |
1454 | leaf = path->nodes[0]; | |
1455 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1456 | if (iref) { | |
3de28d57 LB |
1457 | /* |
1458 | * If type is invalid, we should have bailed out earlier than | |
1459 | * this call. | |
1460 | */ | |
1461 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1462 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1463 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1464 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1465 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1466 | } else { | |
1467 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1468 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1469 | } | |
1470 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1471 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1472 | struct btrfs_extent_data_ref); | |
1473 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1474 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1475 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1476 | struct btrfs_shared_data_ref); | |
1477 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1478 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1479 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1480 | struct btrfs_extent_ref_v0 *ref0; | |
1481 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1482 | struct btrfs_extent_ref_v0); | |
1483 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1484 | #endif |
5d4f98a2 YZ |
1485 | } else { |
1486 | WARN_ON(1); | |
1487 | } | |
1488 | return num_refs; | |
1489 | } | |
15916de8 | 1490 | |
5d4f98a2 | 1491 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1492 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1493 | struct btrfs_path *path, |
1494 | u64 bytenr, u64 parent, | |
1495 | u64 root_objectid) | |
1f3c79a2 | 1496 | { |
87bde3cd | 1497 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 | 1498 | struct btrfs_key key; |
1f3c79a2 | 1499 | int ret; |
1f3c79a2 | 1500 | |
5d4f98a2 YZ |
1501 | key.objectid = bytenr; |
1502 | if (parent) { | |
1503 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1504 | key.offset = parent; | |
1505 | } else { | |
1506 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1507 | key.offset = root_objectid; | |
1f3c79a2 LH |
1508 | } |
1509 | ||
5d4f98a2 YZ |
1510 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1511 | if (ret > 0) | |
1512 | ret = -ENOENT; | |
1513 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1514 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1515 | btrfs_release_path(path); |
5d4f98a2 YZ |
1516 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1517 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1518 | if (ret > 0) | |
1519 | ret = -ENOENT; | |
1520 | } | |
1f3c79a2 | 1521 | #endif |
5d4f98a2 | 1522 | return ret; |
1f3c79a2 LH |
1523 | } |
1524 | ||
5d4f98a2 | 1525 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1526 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1527 | struct btrfs_path *path, |
1528 | u64 bytenr, u64 parent, | |
1529 | u64 root_objectid) | |
31840ae1 | 1530 | { |
5d4f98a2 | 1531 | struct btrfs_key key; |
31840ae1 | 1532 | int ret; |
31840ae1 | 1533 | |
5d4f98a2 YZ |
1534 | key.objectid = bytenr; |
1535 | if (parent) { | |
1536 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1537 | key.offset = parent; | |
1538 | } else { | |
1539 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1540 | key.offset = root_objectid; | |
1541 | } | |
1542 | ||
87bde3cd JM |
1543 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, |
1544 | path, &key, 0); | |
b3b4aa74 | 1545 | btrfs_release_path(path); |
31840ae1 ZY |
1546 | return ret; |
1547 | } | |
1548 | ||
5d4f98a2 | 1549 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1550 | { |
5d4f98a2 YZ |
1551 | int type; |
1552 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1553 | if (parent > 0) | |
1554 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1555 | else | |
1556 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1557 | } else { | |
1558 | if (parent > 0) | |
1559 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1560 | else | |
1561 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1562 | } | |
1563 | return type; | |
31840ae1 | 1564 | } |
56bec294 | 1565 | |
2c47e605 YZ |
1566 | static int find_next_key(struct btrfs_path *path, int level, |
1567 | struct btrfs_key *key) | |
56bec294 | 1568 | |
02217ed2 | 1569 | { |
2c47e605 | 1570 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1571 | if (!path->nodes[level]) |
1572 | break; | |
5d4f98a2 YZ |
1573 | if (path->slots[level] + 1 >= |
1574 | btrfs_header_nritems(path->nodes[level])) | |
1575 | continue; | |
1576 | if (level == 0) | |
1577 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1578 | path->slots[level] + 1); | |
1579 | else | |
1580 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1581 | path->slots[level] + 1); | |
1582 | return 0; | |
1583 | } | |
1584 | return 1; | |
1585 | } | |
037e6390 | 1586 | |
5d4f98a2 YZ |
1587 | /* |
1588 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1589 | * to the address of inline back ref, and 0 is returned. | |
1590 | * | |
1591 | * if back ref isn't found, *ref_ret is set to the address where it | |
1592 | * should be inserted, and -ENOENT is returned. | |
1593 | * | |
1594 | * if insert is true and there are too many inline back refs, the path | |
1595 | * points to the extent item, and -EAGAIN is returned. | |
1596 | * | |
1597 | * NOTE: inline back refs are ordered in the same way that back ref | |
1598 | * items in the tree are ordered. | |
1599 | */ | |
1600 | static noinline_for_stack | |
1601 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1602 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1603 | struct btrfs_path *path, |
1604 | struct btrfs_extent_inline_ref **ref_ret, | |
1605 | u64 bytenr, u64 num_bytes, | |
1606 | u64 parent, u64 root_objectid, | |
1607 | u64 owner, u64 offset, int insert) | |
1608 | { | |
87bde3cd | 1609 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1610 | struct btrfs_key key; |
1611 | struct extent_buffer *leaf; | |
1612 | struct btrfs_extent_item *ei; | |
1613 | struct btrfs_extent_inline_ref *iref; | |
1614 | u64 flags; | |
1615 | u64 item_size; | |
1616 | unsigned long ptr; | |
1617 | unsigned long end; | |
1618 | int extra_size; | |
1619 | int type; | |
1620 | int want; | |
1621 | int ret; | |
1622 | int err = 0; | |
0b246afa | 1623 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1624 | int needed; |
26b8003f | 1625 | |
db94535d | 1626 | key.objectid = bytenr; |
31840ae1 | 1627 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1628 | key.offset = num_bytes; |
31840ae1 | 1629 | |
5d4f98a2 YZ |
1630 | want = extent_ref_type(parent, owner); |
1631 | if (insert) { | |
1632 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1633 | path->keep_locks = 1; |
5d4f98a2 YZ |
1634 | } else |
1635 | extra_size = -1; | |
3173a18f JB |
1636 | |
1637 | /* | |
1638 | * Owner is our parent level, so we can just add one to get the level | |
1639 | * for the block we are interested in. | |
1640 | */ | |
1641 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1642 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1643 | key.offset = owner; | |
1644 | } | |
1645 | ||
1646 | again: | |
5d4f98a2 | 1647 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1648 | if (ret < 0) { |
5d4f98a2 YZ |
1649 | err = ret; |
1650 | goto out; | |
1651 | } | |
3173a18f JB |
1652 | |
1653 | /* | |
1654 | * We may be a newly converted file system which still has the old fat | |
1655 | * extent entries for metadata, so try and see if we have one of those. | |
1656 | */ | |
1657 | if (ret > 0 && skinny_metadata) { | |
1658 | skinny_metadata = false; | |
1659 | if (path->slots[0]) { | |
1660 | path->slots[0]--; | |
1661 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1662 | path->slots[0]); | |
1663 | if (key.objectid == bytenr && | |
1664 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1665 | key.offset == num_bytes) | |
1666 | ret = 0; | |
1667 | } | |
1668 | if (ret) { | |
9ce49a0b | 1669 | key.objectid = bytenr; |
3173a18f JB |
1670 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1671 | key.offset = num_bytes; | |
1672 | btrfs_release_path(path); | |
1673 | goto again; | |
1674 | } | |
1675 | } | |
1676 | ||
79787eaa JM |
1677 | if (ret && !insert) { |
1678 | err = -ENOENT; | |
1679 | goto out; | |
fae7f21c | 1680 | } else if (WARN_ON(ret)) { |
492104c8 | 1681 | err = -EIO; |
492104c8 | 1682 | goto out; |
79787eaa | 1683 | } |
5d4f98a2 YZ |
1684 | |
1685 | leaf = path->nodes[0]; | |
1686 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1687 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1688 | if (item_size < sizeof(*ei)) { | |
1689 | if (!insert) { | |
1690 | err = -ENOENT; | |
1691 | goto out; | |
1692 | } | |
87bde3cd | 1693 | ret = convert_extent_item_v0(trans, fs_info, path, owner, |
5d4f98a2 YZ |
1694 | extra_size); |
1695 | if (ret < 0) { | |
1696 | err = ret; | |
1697 | goto out; | |
1698 | } | |
1699 | leaf = path->nodes[0]; | |
1700 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1701 | } | |
1702 | #endif | |
1703 | BUG_ON(item_size < sizeof(*ei)); | |
1704 | ||
5d4f98a2 YZ |
1705 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1706 | flags = btrfs_extent_flags(leaf, ei); | |
1707 | ||
1708 | ptr = (unsigned long)(ei + 1); | |
1709 | end = (unsigned long)ei + item_size; | |
1710 | ||
3173a18f | 1711 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1712 | ptr += sizeof(struct btrfs_tree_block_info); |
1713 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1714 | } |
1715 | ||
3de28d57 LB |
1716 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1717 | needed = BTRFS_REF_TYPE_DATA; | |
1718 | else | |
1719 | needed = BTRFS_REF_TYPE_BLOCK; | |
1720 | ||
5d4f98a2 YZ |
1721 | err = -ENOENT; |
1722 | while (1) { | |
1723 | if (ptr >= end) { | |
1724 | WARN_ON(ptr > end); | |
1725 | break; | |
1726 | } | |
1727 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1728 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1729 | if (type == BTRFS_REF_TYPE_INVALID) { | |
1730 | err = -EINVAL; | |
1731 | goto out; | |
1732 | } | |
1733 | ||
5d4f98a2 YZ |
1734 | if (want < type) |
1735 | break; | |
1736 | if (want > type) { | |
1737 | ptr += btrfs_extent_inline_ref_size(type); | |
1738 | continue; | |
1739 | } | |
1740 | ||
1741 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1742 | struct btrfs_extent_data_ref *dref; | |
1743 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1744 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1745 | owner, offset)) { | |
1746 | err = 0; | |
1747 | break; | |
1748 | } | |
1749 | if (hash_extent_data_ref_item(leaf, dref) < | |
1750 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1751 | break; | |
1752 | } else { | |
1753 | u64 ref_offset; | |
1754 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1755 | if (parent > 0) { | |
1756 | if (parent == ref_offset) { | |
1757 | err = 0; | |
1758 | break; | |
1759 | } | |
1760 | if (ref_offset < parent) | |
1761 | break; | |
1762 | } else { | |
1763 | if (root_objectid == ref_offset) { | |
1764 | err = 0; | |
1765 | break; | |
1766 | } | |
1767 | if (ref_offset < root_objectid) | |
1768 | break; | |
1769 | } | |
1770 | } | |
1771 | ptr += btrfs_extent_inline_ref_size(type); | |
1772 | } | |
1773 | if (err == -ENOENT && insert) { | |
1774 | if (item_size + extra_size >= | |
1775 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1776 | err = -EAGAIN; | |
1777 | goto out; | |
1778 | } | |
1779 | /* | |
1780 | * To add new inline back ref, we have to make sure | |
1781 | * there is no corresponding back ref item. | |
1782 | * For simplicity, we just do not add new inline back | |
1783 | * ref if there is any kind of item for this block | |
1784 | */ | |
2c47e605 YZ |
1785 | if (find_next_key(path, 0, &key) == 0 && |
1786 | key.objectid == bytenr && | |
85d4198e | 1787 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1788 | err = -EAGAIN; |
1789 | goto out; | |
1790 | } | |
1791 | } | |
1792 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1793 | out: | |
85d4198e | 1794 | if (insert) { |
5d4f98a2 YZ |
1795 | path->keep_locks = 0; |
1796 | btrfs_unlock_up_safe(path, 1); | |
1797 | } | |
1798 | return err; | |
1799 | } | |
1800 | ||
1801 | /* | |
1802 | * helper to add new inline back ref | |
1803 | */ | |
1804 | static noinline_for_stack | |
87bde3cd | 1805 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1806 | struct btrfs_path *path, |
1807 | struct btrfs_extent_inline_ref *iref, | |
1808 | u64 parent, u64 root_objectid, | |
1809 | u64 owner, u64 offset, int refs_to_add, | |
1810 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1811 | { |
1812 | struct extent_buffer *leaf; | |
1813 | struct btrfs_extent_item *ei; | |
1814 | unsigned long ptr; | |
1815 | unsigned long end; | |
1816 | unsigned long item_offset; | |
1817 | u64 refs; | |
1818 | int size; | |
1819 | int type; | |
5d4f98a2 YZ |
1820 | |
1821 | leaf = path->nodes[0]; | |
1822 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1823 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1824 | ||
1825 | type = extent_ref_type(parent, owner); | |
1826 | size = btrfs_extent_inline_ref_size(type); | |
1827 | ||
87bde3cd | 1828 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1829 | |
1830 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1831 | refs = btrfs_extent_refs(leaf, ei); | |
1832 | refs += refs_to_add; | |
1833 | btrfs_set_extent_refs(leaf, ei, refs); | |
1834 | if (extent_op) | |
1835 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1836 | ||
1837 | ptr = (unsigned long)ei + item_offset; | |
1838 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1839 | if (ptr < end - size) | |
1840 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1841 | end - size - ptr); | |
1842 | ||
1843 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1844 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1845 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1846 | struct btrfs_extent_data_ref *dref; | |
1847 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1848 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1849 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1850 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1851 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1852 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1853 | struct btrfs_shared_data_ref *sref; | |
1854 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1855 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1856 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1857 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1858 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1859 | } else { | |
1860 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1861 | } | |
1862 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1863 | } |
1864 | ||
1865 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1866 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1867 | struct btrfs_path *path, |
1868 | struct btrfs_extent_inline_ref **ref_ret, | |
1869 | u64 bytenr, u64 num_bytes, u64 parent, | |
1870 | u64 root_objectid, u64 owner, u64 offset) | |
1871 | { | |
1872 | int ret; | |
1873 | ||
87bde3cd | 1874 | ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret, |
5d4f98a2 YZ |
1875 | bytenr, num_bytes, parent, |
1876 | root_objectid, owner, offset, 0); | |
1877 | if (ret != -ENOENT) | |
54aa1f4d | 1878 | return ret; |
5d4f98a2 | 1879 | |
b3b4aa74 | 1880 | btrfs_release_path(path); |
5d4f98a2 YZ |
1881 | *ref_ret = NULL; |
1882 | ||
1883 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
87bde3cd JM |
1884 | ret = lookup_tree_block_ref(trans, fs_info, path, bytenr, |
1885 | parent, root_objectid); | |
5d4f98a2 | 1886 | } else { |
87bde3cd JM |
1887 | ret = lookup_extent_data_ref(trans, fs_info, path, bytenr, |
1888 | parent, root_objectid, owner, | |
1889 | offset); | |
b9473439 | 1890 | } |
5d4f98a2 YZ |
1891 | return ret; |
1892 | } | |
31840ae1 | 1893 | |
5d4f98a2 YZ |
1894 | /* |
1895 | * helper to update/remove inline back ref | |
1896 | */ | |
1897 | static noinline_for_stack | |
87bde3cd | 1898 | void update_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1899 | struct btrfs_path *path, |
1900 | struct btrfs_extent_inline_ref *iref, | |
1901 | int refs_to_mod, | |
fcebe456 JB |
1902 | struct btrfs_delayed_extent_op *extent_op, |
1903 | int *last_ref) | |
5d4f98a2 YZ |
1904 | { |
1905 | struct extent_buffer *leaf; | |
1906 | struct btrfs_extent_item *ei; | |
1907 | struct btrfs_extent_data_ref *dref = NULL; | |
1908 | struct btrfs_shared_data_ref *sref = NULL; | |
1909 | unsigned long ptr; | |
1910 | unsigned long end; | |
1911 | u32 item_size; | |
1912 | int size; | |
1913 | int type; | |
5d4f98a2 YZ |
1914 | u64 refs; |
1915 | ||
1916 | leaf = path->nodes[0]; | |
1917 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1918 | refs = btrfs_extent_refs(leaf, ei); | |
1919 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1920 | refs += refs_to_mod; | |
1921 | btrfs_set_extent_refs(leaf, ei, refs); | |
1922 | if (extent_op) | |
1923 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1924 | ||
3de28d57 LB |
1925 | /* |
1926 | * If type is invalid, we should have bailed out after | |
1927 | * lookup_inline_extent_backref(). | |
1928 | */ | |
1929 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1930 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1931 | |
1932 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1933 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1934 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1935 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1936 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1937 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1938 | } else { | |
1939 | refs = 1; | |
1940 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1941 | } |
31840ae1 | 1942 | |
5d4f98a2 YZ |
1943 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1944 | refs += refs_to_mod; | |
1945 | ||
1946 | if (refs > 0) { | |
1947 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1948 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1949 | else | |
1950 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1951 | } else { | |
fcebe456 | 1952 | *last_ref = 1; |
5d4f98a2 YZ |
1953 | size = btrfs_extent_inline_ref_size(type); |
1954 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1955 | ptr = (unsigned long)iref; | |
1956 | end = (unsigned long)ei + item_size; | |
1957 | if (ptr + size < end) | |
1958 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1959 | end - ptr - size); | |
1960 | item_size -= size; | |
87bde3cd | 1961 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1962 | } |
1963 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1964 | } |
1965 | ||
1966 | static noinline_for_stack | |
1967 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1968 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1969 | struct btrfs_path *path, |
1970 | u64 bytenr, u64 num_bytes, u64 parent, | |
1971 | u64 root_objectid, u64 owner, | |
1972 | u64 offset, int refs_to_add, | |
1973 | struct btrfs_delayed_extent_op *extent_op) | |
1974 | { | |
1975 | struct btrfs_extent_inline_ref *iref; | |
1976 | int ret; | |
1977 | ||
87bde3cd | 1978 | ret = lookup_inline_extent_backref(trans, fs_info, path, &iref, |
5d4f98a2 YZ |
1979 | bytenr, num_bytes, parent, |
1980 | root_objectid, owner, offset, 1); | |
1981 | if (ret == 0) { | |
1982 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
87bde3cd | 1983 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 1984 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1985 | } else if (ret == -ENOENT) { |
87bde3cd | 1986 | setup_inline_extent_backref(fs_info, path, iref, parent, |
143bede5 JM |
1987 | root_objectid, owner, offset, |
1988 | refs_to_add, extent_op); | |
1989 | ret = 0; | |
771ed689 | 1990 | } |
5d4f98a2 YZ |
1991 | return ret; |
1992 | } | |
31840ae1 | 1993 | |
5d4f98a2 | 1994 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 1995 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1996 | struct btrfs_path *path, |
1997 | u64 bytenr, u64 parent, u64 root_objectid, | |
1998 | u64 owner, u64 offset, int refs_to_add) | |
1999 | { | |
2000 | int ret; | |
2001 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
2002 | BUG_ON(refs_to_add != 1); | |
87bde3cd | 2003 | ret = insert_tree_block_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
2004 | parent, root_objectid); |
2005 | } else { | |
87bde3cd | 2006 | ret = insert_extent_data_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
2007 | parent, root_objectid, |
2008 | owner, offset, refs_to_add); | |
2009 | } | |
2010 | return ret; | |
2011 | } | |
56bec294 | 2012 | |
5d4f98a2 | 2013 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 2014 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2015 | struct btrfs_path *path, |
2016 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 2017 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 2018 | { |
143bede5 | 2019 | int ret = 0; |
b9473439 | 2020 | |
5d4f98a2 YZ |
2021 | BUG_ON(!is_data && refs_to_drop != 1); |
2022 | if (iref) { | |
87bde3cd | 2023 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 2024 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 2025 | } else if (is_data) { |
87bde3cd | 2026 | ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop, |
fcebe456 | 2027 | last_ref); |
5d4f98a2 | 2028 | } else { |
fcebe456 | 2029 | *last_ref = 1; |
87bde3cd | 2030 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
5d4f98a2 YZ |
2031 | } |
2032 | return ret; | |
2033 | } | |
2034 | ||
86557861 | 2035 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
2036 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
2037 | u64 *discarded_bytes) | |
5d4f98a2 | 2038 | { |
86557861 JM |
2039 | int j, ret = 0; |
2040 | u64 bytes_left, end; | |
4d89d377 | 2041 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 2042 | |
4d89d377 JM |
2043 | if (WARN_ON(start != aligned_start)) { |
2044 | len -= aligned_start - start; | |
2045 | len = round_down(len, 1 << 9); | |
2046 | start = aligned_start; | |
2047 | } | |
d04c6b88 | 2048 | |
4d89d377 | 2049 | *discarded_bytes = 0; |
86557861 JM |
2050 | |
2051 | if (!len) | |
2052 | return 0; | |
2053 | ||
2054 | end = start + len; | |
2055 | bytes_left = len; | |
2056 | ||
2057 | /* Skip any superblocks on this device. */ | |
2058 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
2059 | u64 sb_start = btrfs_sb_offset(j); | |
2060 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
2061 | u64 size = sb_start - start; | |
2062 | ||
2063 | if (!in_range(sb_start, start, bytes_left) && | |
2064 | !in_range(sb_end, start, bytes_left) && | |
2065 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
2066 | continue; | |
2067 | ||
2068 | /* | |
2069 | * Superblock spans beginning of range. Adjust start and | |
2070 | * try again. | |
2071 | */ | |
2072 | if (sb_start <= start) { | |
2073 | start += sb_end - start; | |
2074 | if (start > end) { | |
2075 | bytes_left = 0; | |
2076 | break; | |
2077 | } | |
2078 | bytes_left = end - start; | |
2079 | continue; | |
2080 | } | |
2081 | ||
2082 | if (size) { | |
2083 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2084 | GFP_NOFS, 0); | |
2085 | if (!ret) | |
2086 | *discarded_bytes += size; | |
2087 | else if (ret != -EOPNOTSUPP) | |
2088 | return ret; | |
2089 | } | |
2090 | ||
2091 | start = sb_end; | |
2092 | if (start > end) { | |
2093 | bytes_left = 0; | |
2094 | break; | |
2095 | } | |
2096 | bytes_left = end - start; | |
2097 | } | |
2098 | ||
2099 | if (bytes_left) { | |
2100 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2101 | GFP_NOFS, 0); |
2102 | if (!ret) | |
86557861 | 2103 | *discarded_bytes += bytes_left; |
4d89d377 | 2104 | } |
d04c6b88 | 2105 | return ret; |
5d4f98a2 | 2106 | } |
5d4f98a2 | 2107 | |
2ff7e61e | 2108 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 2109 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 2110 | { |
5d4f98a2 | 2111 | int ret; |
5378e607 | 2112 | u64 discarded_bytes = 0; |
a1d3c478 | 2113 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2114 | |
e244a0ae | 2115 | |
2999241d FM |
2116 | /* |
2117 | * Avoid races with device replace and make sure our bbio has devices | |
2118 | * associated to its stripes that don't go away while we are discarding. | |
2119 | */ | |
0b246afa | 2120 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 2121 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
2122 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
2123 | &bbio, 0); | |
79787eaa | 2124 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2125 | if (!ret) { |
a1d3c478 | 2126 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2127 | int i; |
2128 | ||
5d4f98a2 | 2129 | |
a1d3c478 | 2130 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2131 | u64 bytes; |
38b5f68e AJ |
2132 | struct request_queue *req_q; |
2133 | ||
627e0873 FM |
2134 | if (!stripe->dev->bdev) { |
2135 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
2136 | continue; | |
2137 | } | |
38b5f68e AJ |
2138 | req_q = bdev_get_queue(stripe->dev->bdev); |
2139 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
2140 | continue; |
2141 | ||
5378e607 LD |
2142 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2143 | stripe->physical, | |
d04c6b88 JM |
2144 | stripe->length, |
2145 | &bytes); | |
5378e607 | 2146 | if (!ret) |
d04c6b88 | 2147 | discarded_bytes += bytes; |
5378e607 | 2148 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2149 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2150 | |
2151 | /* | |
2152 | * Just in case we get back EOPNOTSUPP for some reason, | |
2153 | * just ignore the return value so we don't screw up | |
2154 | * people calling discard_extent. | |
2155 | */ | |
2156 | ret = 0; | |
5d4f98a2 | 2157 | } |
6e9606d2 | 2158 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2159 | } |
0b246afa | 2160 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2161 | |
2162 | if (actual_bytes) | |
2163 | *actual_bytes = discarded_bytes; | |
2164 | ||
5d4f98a2 | 2165 | |
53b381b3 DW |
2166 | if (ret == -EOPNOTSUPP) |
2167 | ret = 0; | |
5d4f98a2 | 2168 | return ret; |
5d4f98a2 YZ |
2169 | } |
2170 | ||
79787eaa | 2171 | /* Can return -ENOMEM */ |
5d4f98a2 | 2172 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
84f7d8e6 | 2173 | struct btrfs_root *root, |
5d4f98a2 | 2174 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2175 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 | 2176 | { |
84f7d8e6 | 2177 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 2178 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2179 | int ret; |
66d7e7f0 | 2180 | |
5d4f98a2 YZ |
2181 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2182 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2183 | ||
fd708b81 JB |
2184 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid, |
2185 | owner, offset, BTRFS_ADD_DELAYED_REF); | |
2186 | ||
5d4f98a2 | 2187 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 2188 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
2189 | num_bytes, parent, |
2190 | root_objectid, (int)owner, | |
2191 | BTRFS_ADD_DELAYED_REF, NULL, | |
d7eae340 | 2192 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 2193 | } else { |
66d7e7f0 | 2194 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
2195 | num_bytes, parent, |
2196 | root_objectid, owner, offset, | |
d7eae340 OS |
2197 | 0, BTRFS_ADD_DELAYED_REF, |
2198 | &old_ref_mod, &new_ref_mod); | |
5d4f98a2 | 2199 | } |
d7eae340 | 2200 | |
29d2b84c NB |
2201 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) { |
2202 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
2203 | ||
2204 | add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid); | |
2205 | } | |
d7eae340 | 2206 | |
5d4f98a2 YZ |
2207 | return ret; |
2208 | } | |
2209 | ||
2210 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2211 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 2212 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2213 | u64 parent, u64 root_objectid, |
2214 | u64 owner, u64 offset, int refs_to_add, | |
2215 | struct btrfs_delayed_extent_op *extent_op) | |
2216 | { | |
2217 | struct btrfs_path *path; | |
2218 | struct extent_buffer *leaf; | |
2219 | struct btrfs_extent_item *item; | |
fcebe456 | 2220 | struct btrfs_key key; |
c682f9b3 QW |
2221 | u64 bytenr = node->bytenr; |
2222 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2223 | u64 refs; |
2224 | int ret; | |
5d4f98a2 YZ |
2225 | |
2226 | path = btrfs_alloc_path(); | |
2227 | if (!path) | |
2228 | return -ENOMEM; | |
2229 | ||
e4058b54 | 2230 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2231 | path->leave_spinning = 1; |
2232 | /* this will setup the path even if it fails to insert the back ref */ | |
87bde3cd JM |
2233 | ret = insert_inline_extent_backref(trans, fs_info, path, bytenr, |
2234 | num_bytes, parent, root_objectid, | |
2235 | owner, offset, | |
5d4f98a2 | 2236 | refs_to_add, extent_op); |
0ed4792a | 2237 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2238 | goto out; |
fcebe456 JB |
2239 | |
2240 | /* | |
2241 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2242 | * inline extent ref, so just update the reference count and add a | |
2243 | * normal backref. | |
2244 | */ | |
5d4f98a2 | 2245 | leaf = path->nodes[0]; |
fcebe456 | 2246 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2247 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2248 | refs = btrfs_extent_refs(leaf, item); | |
2249 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2250 | if (extent_op) | |
2251 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2252 | |
5d4f98a2 | 2253 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2254 | btrfs_release_path(path); |
56bec294 | 2255 | |
e4058b54 | 2256 | path->reada = READA_FORWARD; |
b9473439 | 2257 | path->leave_spinning = 1; |
56bec294 | 2258 | /* now insert the actual backref */ |
87bde3cd JM |
2259 | ret = insert_extent_backref(trans, fs_info, path, bytenr, parent, |
2260 | root_objectid, owner, offset, refs_to_add); | |
79787eaa | 2261 | if (ret) |
66642832 | 2262 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2263 | out: |
56bec294 | 2264 | btrfs_free_path(path); |
30d133fc | 2265 | return ret; |
56bec294 CM |
2266 | } |
2267 | ||
5d4f98a2 | 2268 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2269 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2270 | struct btrfs_delayed_ref_node *node, |
2271 | struct btrfs_delayed_extent_op *extent_op, | |
2272 | int insert_reserved) | |
56bec294 | 2273 | { |
5d4f98a2 YZ |
2274 | int ret = 0; |
2275 | struct btrfs_delayed_data_ref *ref; | |
2276 | struct btrfs_key ins; | |
2277 | u64 parent = 0; | |
2278 | u64 ref_root = 0; | |
2279 | u64 flags = 0; | |
2280 | ||
2281 | ins.objectid = node->bytenr; | |
2282 | ins.offset = node->num_bytes; | |
2283 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2284 | ||
2285 | ref = btrfs_delayed_node_to_data_ref(node); | |
0b246afa | 2286 | trace_run_delayed_data_ref(fs_info, node, ref, node->action); |
599c75ec | 2287 | |
5d4f98a2 YZ |
2288 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2289 | parent = ref->parent; | |
fcebe456 | 2290 | ref_root = ref->root; |
5d4f98a2 YZ |
2291 | |
2292 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2293 | if (extent_op) |
5d4f98a2 | 2294 | flags |= extent_op->flags_to_set; |
2ff7e61e | 2295 | ret = alloc_reserved_file_extent(trans, fs_info, |
5d4f98a2 YZ |
2296 | parent, ref_root, flags, |
2297 | ref->objectid, ref->offset, | |
2298 | &ins, node->ref_mod); | |
5d4f98a2 | 2299 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2300 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2301 | ref_root, ref->objectid, |
2302 | ref->offset, node->ref_mod, | |
c682f9b3 | 2303 | extent_op); |
5d4f98a2 | 2304 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2305 | ret = __btrfs_free_extent(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2306 | ref_root, ref->objectid, |
2307 | ref->offset, node->ref_mod, | |
c682f9b3 | 2308 | extent_op); |
5d4f98a2 YZ |
2309 | } else { |
2310 | BUG(); | |
2311 | } | |
2312 | return ret; | |
2313 | } | |
2314 | ||
2315 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2316 | struct extent_buffer *leaf, | |
2317 | struct btrfs_extent_item *ei) | |
2318 | { | |
2319 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2320 | if (extent_op->update_flags) { | |
2321 | flags |= extent_op->flags_to_set; | |
2322 | btrfs_set_extent_flags(leaf, ei, flags); | |
2323 | } | |
2324 | ||
2325 | if (extent_op->update_key) { | |
2326 | struct btrfs_tree_block_info *bi; | |
2327 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2328 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2329 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2330 | } | |
2331 | } | |
2332 | ||
2333 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2334 | struct btrfs_fs_info *fs_info, |
d278850e | 2335 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2336 | struct btrfs_delayed_extent_op *extent_op) |
2337 | { | |
2338 | struct btrfs_key key; | |
2339 | struct btrfs_path *path; | |
2340 | struct btrfs_extent_item *ei; | |
2341 | struct extent_buffer *leaf; | |
2342 | u32 item_size; | |
56bec294 | 2343 | int ret; |
5d4f98a2 | 2344 | int err = 0; |
b1c79e09 | 2345 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2346 | |
79787eaa JM |
2347 | if (trans->aborted) |
2348 | return 0; | |
2349 | ||
0b246afa | 2350 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2351 | metadata = 0; |
2352 | ||
5d4f98a2 YZ |
2353 | path = btrfs_alloc_path(); |
2354 | if (!path) | |
2355 | return -ENOMEM; | |
2356 | ||
d278850e | 2357 | key.objectid = head->bytenr; |
5d4f98a2 | 2358 | |
3173a18f | 2359 | if (metadata) { |
3173a18f | 2360 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2361 | key.offset = extent_op->level; |
3173a18f JB |
2362 | } else { |
2363 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2364 | key.offset = head->num_bytes; |
3173a18f JB |
2365 | } |
2366 | ||
2367 | again: | |
e4058b54 | 2368 | path->reada = READA_FORWARD; |
5d4f98a2 | 2369 | path->leave_spinning = 1; |
0b246afa | 2370 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2371 | if (ret < 0) { |
2372 | err = ret; | |
2373 | goto out; | |
2374 | } | |
2375 | if (ret > 0) { | |
3173a18f | 2376 | if (metadata) { |
55994887 FDBM |
2377 | if (path->slots[0] > 0) { |
2378 | path->slots[0]--; | |
2379 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2380 | path->slots[0]); | |
d278850e | 2381 | if (key.objectid == head->bytenr && |
55994887 | 2382 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2383 | key.offset == head->num_bytes) |
55994887 FDBM |
2384 | ret = 0; |
2385 | } | |
2386 | if (ret > 0) { | |
2387 | btrfs_release_path(path); | |
2388 | metadata = 0; | |
3173a18f | 2389 | |
d278850e JB |
2390 | key.objectid = head->bytenr; |
2391 | key.offset = head->num_bytes; | |
55994887 FDBM |
2392 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2393 | goto again; | |
2394 | } | |
2395 | } else { | |
2396 | err = -EIO; | |
2397 | goto out; | |
3173a18f | 2398 | } |
5d4f98a2 YZ |
2399 | } |
2400 | ||
2401 | leaf = path->nodes[0]; | |
2402 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2403 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2404 | if (item_size < sizeof(*ei)) { | |
87bde3cd | 2405 | ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0); |
5d4f98a2 YZ |
2406 | if (ret < 0) { |
2407 | err = ret; | |
2408 | goto out; | |
2409 | } | |
2410 | leaf = path->nodes[0]; | |
2411 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2412 | } | |
2413 | #endif | |
2414 | BUG_ON(item_size < sizeof(*ei)); | |
2415 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2416 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2417 | |
5d4f98a2 YZ |
2418 | btrfs_mark_buffer_dirty(leaf); |
2419 | out: | |
2420 | btrfs_free_path(path); | |
2421 | return err; | |
56bec294 CM |
2422 | } |
2423 | ||
5d4f98a2 | 2424 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2425 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2426 | struct btrfs_delayed_ref_node *node, |
2427 | struct btrfs_delayed_extent_op *extent_op, | |
2428 | int insert_reserved) | |
56bec294 CM |
2429 | { |
2430 | int ret = 0; | |
5d4f98a2 | 2431 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
2432 | u64 parent = 0; |
2433 | u64 ref_root = 0; | |
56bec294 | 2434 | |
5d4f98a2 | 2435 | ref = btrfs_delayed_node_to_tree_ref(node); |
0b246afa | 2436 | trace_run_delayed_tree_ref(fs_info, node, ref, node->action); |
599c75ec | 2437 | |
5d4f98a2 YZ |
2438 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2439 | parent = ref->parent; | |
fcebe456 | 2440 | ref_root = ref->root; |
5d4f98a2 | 2441 | |
02794222 | 2442 | if (node->ref_mod != 1) { |
2ff7e61e | 2443 | btrfs_err(fs_info, |
02794222 LB |
2444 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2445 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2446 | parent); | |
2447 | return -EIO; | |
2448 | } | |
5d4f98a2 | 2449 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2450 | BUG_ON(!extent_op || !extent_op->update_flags); |
4e6bd4e0 | 2451 | ret = alloc_reserved_tree_block(trans, node, |
5d4f98a2 | 2452 | extent_op->flags_to_set, |
4e6bd4e0 | 2453 | &extent_op->key); |
5d4f98a2 | 2454 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2455 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, |
c682f9b3 QW |
2456 | parent, ref_root, |
2457 | ref->level, 0, 1, | |
fcebe456 | 2458 | extent_op); |
5d4f98a2 | 2459 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2460 | ret = __btrfs_free_extent(trans, fs_info, node, |
c682f9b3 QW |
2461 | parent, ref_root, |
2462 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2463 | } else { |
2464 | BUG(); | |
2465 | } | |
56bec294 CM |
2466 | return ret; |
2467 | } | |
2468 | ||
2469 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2470 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2471 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2472 | struct btrfs_delayed_ref_node *node, |
2473 | struct btrfs_delayed_extent_op *extent_op, | |
2474 | int insert_reserved) | |
56bec294 | 2475 | { |
79787eaa JM |
2476 | int ret = 0; |
2477 | ||
857cc2fc JB |
2478 | if (trans->aborted) { |
2479 | if (insert_reserved) | |
2ff7e61e | 2480 | btrfs_pin_extent(fs_info, node->bytenr, |
857cc2fc | 2481 | node->num_bytes, 1); |
79787eaa | 2482 | return 0; |
857cc2fc | 2483 | } |
79787eaa | 2484 | |
5d4f98a2 YZ |
2485 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2486 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2ff7e61e | 2487 | ret = run_delayed_tree_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2488 | insert_reserved); |
2489 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2490 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2ff7e61e | 2491 | ret = run_delayed_data_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2492 | insert_reserved); |
2493 | else | |
2494 | BUG(); | |
2495 | return ret; | |
56bec294 CM |
2496 | } |
2497 | ||
c6fc2454 | 2498 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2499 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2500 | { | |
cffc3374 FM |
2501 | struct btrfs_delayed_ref_node *ref; |
2502 | ||
0e0adbcf | 2503 | if (RB_EMPTY_ROOT(&head->ref_tree)) |
c6fc2454 | 2504 | return NULL; |
d7df2c79 | 2505 | |
cffc3374 FM |
2506 | /* |
2507 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2508 | * This is to prevent a ref count from going down to zero, which deletes | |
2509 | * the extent item from the extent tree, when there still are references | |
2510 | * to add, which would fail because they would not find the extent item. | |
2511 | */ | |
1d57ee94 WX |
2512 | if (!list_empty(&head->ref_add_list)) |
2513 | return list_first_entry(&head->ref_add_list, | |
2514 | struct btrfs_delayed_ref_node, add_list); | |
2515 | ||
0e0adbcf JB |
2516 | ref = rb_entry(rb_first(&head->ref_tree), |
2517 | struct btrfs_delayed_ref_node, ref_node); | |
1d57ee94 WX |
2518 | ASSERT(list_empty(&ref->add_list)); |
2519 | return ref; | |
56bec294 CM |
2520 | } |
2521 | ||
2eadaa22 JB |
2522 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2523 | struct btrfs_delayed_ref_head *head) | |
2524 | { | |
2525 | spin_lock(&delayed_refs->lock); | |
2526 | head->processing = 0; | |
2527 | delayed_refs->num_heads_ready++; | |
2528 | spin_unlock(&delayed_refs->lock); | |
2529 | btrfs_delayed_ref_unlock(head); | |
2530 | } | |
2531 | ||
b00e6250 JB |
2532 | static int cleanup_extent_op(struct btrfs_trans_handle *trans, |
2533 | struct btrfs_fs_info *fs_info, | |
2534 | struct btrfs_delayed_ref_head *head) | |
2535 | { | |
2536 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
2537 | int ret; | |
2538 | ||
2539 | if (!extent_op) | |
2540 | return 0; | |
2541 | head->extent_op = NULL; | |
2542 | if (head->must_insert_reserved) { | |
2543 | btrfs_free_delayed_extent_op(extent_op); | |
2544 | return 0; | |
2545 | } | |
2546 | spin_unlock(&head->lock); | |
d278850e | 2547 | ret = run_delayed_extent_op(trans, fs_info, head, extent_op); |
b00e6250 JB |
2548 | btrfs_free_delayed_extent_op(extent_op); |
2549 | return ret ? ret : 1; | |
2550 | } | |
2551 | ||
194ab0bc JB |
2552 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
2553 | struct btrfs_fs_info *fs_info, | |
2554 | struct btrfs_delayed_ref_head *head) | |
2555 | { | |
2556 | struct btrfs_delayed_ref_root *delayed_refs; | |
2557 | int ret; | |
2558 | ||
2559 | delayed_refs = &trans->transaction->delayed_refs; | |
2560 | ||
2561 | ret = cleanup_extent_op(trans, fs_info, head); | |
2562 | if (ret < 0) { | |
2563 | unselect_delayed_ref_head(delayed_refs, head); | |
2564 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2565 | return ret; | |
2566 | } else if (ret) { | |
2567 | return ret; | |
2568 | } | |
2569 | ||
2570 | /* | |
2571 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2572 | * and then re-check to make sure nobody got added. | |
2573 | */ | |
2574 | spin_unlock(&head->lock); | |
2575 | spin_lock(&delayed_refs->lock); | |
2576 | spin_lock(&head->lock); | |
0e0adbcf | 2577 | if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) { |
194ab0bc JB |
2578 | spin_unlock(&head->lock); |
2579 | spin_unlock(&delayed_refs->lock); | |
2580 | return 1; | |
2581 | } | |
194ab0bc JB |
2582 | delayed_refs->num_heads--; |
2583 | rb_erase(&head->href_node, &delayed_refs->href_root); | |
d278850e | 2584 | RB_CLEAR_NODE(&head->href_node); |
c1103f7a | 2585 | spin_unlock(&head->lock); |
1e7a1421 | 2586 | spin_unlock(&delayed_refs->lock); |
c1103f7a JB |
2587 | atomic_dec(&delayed_refs->num_entries); |
2588 | ||
d278850e | 2589 | trace_run_delayed_ref_head(fs_info, head, 0); |
c1103f7a JB |
2590 | |
2591 | if (head->total_ref_mod < 0) { | |
5e388e95 NB |
2592 | struct btrfs_space_info *space_info; |
2593 | u64 flags; | |
c1103f7a | 2594 | |
5e388e95 NB |
2595 | if (head->is_data) |
2596 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2597 | else if (head->is_system) | |
2598 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
2599 | else | |
2600 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2601 | space_info = __find_space_info(fs_info, flags); | |
2602 | ASSERT(space_info); | |
2603 | percpu_counter_add(&space_info->total_bytes_pinned, | |
d278850e | 2604 | -head->num_bytes); |
c1103f7a JB |
2605 | |
2606 | if (head->is_data) { | |
2607 | spin_lock(&delayed_refs->lock); | |
d278850e | 2608 | delayed_refs->pending_csums -= head->num_bytes; |
c1103f7a JB |
2609 | spin_unlock(&delayed_refs->lock); |
2610 | } | |
2611 | } | |
2612 | ||
2613 | if (head->must_insert_reserved) { | |
d278850e JB |
2614 | btrfs_pin_extent(fs_info, head->bytenr, |
2615 | head->num_bytes, 1); | |
c1103f7a | 2616 | if (head->is_data) { |
d278850e JB |
2617 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2618 | head->num_bytes); | |
c1103f7a JB |
2619 | } |
2620 | } | |
2621 | ||
2622 | /* Also free its reserved qgroup space */ | |
2623 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, | |
2624 | head->qgroup_reserved); | |
2625 | btrfs_delayed_ref_unlock(head); | |
d278850e | 2626 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2627 | return 0; |
2628 | } | |
2629 | ||
79787eaa JM |
2630 | /* |
2631 | * Returns 0 on success or if called with an already aborted transaction. | |
2632 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2633 | */ | |
d7df2c79 | 2634 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2635 | unsigned long nr) |
56bec294 | 2636 | { |
0a1e458a | 2637 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 CM |
2638 | struct btrfs_delayed_ref_root *delayed_refs; |
2639 | struct btrfs_delayed_ref_node *ref; | |
2640 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2641 | struct btrfs_delayed_extent_op *extent_op; |
0a2b2a84 | 2642 | ktime_t start = ktime_get(); |
56bec294 | 2643 | int ret; |
d7df2c79 | 2644 | unsigned long count = 0; |
0a2b2a84 | 2645 | unsigned long actual_count = 0; |
56bec294 | 2646 | int must_insert_reserved = 0; |
56bec294 CM |
2647 | |
2648 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2649 | while (1) { |
2650 | if (!locked_ref) { | |
d7df2c79 | 2651 | if (count >= nr) |
56bec294 | 2652 | break; |
56bec294 | 2653 | |
d7df2c79 JB |
2654 | spin_lock(&delayed_refs->lock); |
2655 | locked_ref = btrfs_select_ref_head(trans); | |
2656 | if (!locked_ref) { | |
2657 | spin_unlock(&delayed_refs->lock); | |
2658 | break; | |
2659 | } | |
c3e69d58 CM |
2660 | |
2661 | /* grab the lock that says we are going to process | |
2662 | * all the refs for this head */ | |
2663 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2664 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2665 | /* |
2666 | * we may have dropped the spin lock to get the head | |
2667 | * mutex lock, and that might have given someone else | |
2668 | * time to free the head. If that's true, it has been | |
2669 | * removed from our list and we can move on. | |
2670 | */ | |
2671 | if (ret == -EAGAIN) { | |
2672 | locked_ref = NULL; | |
2673 | count++; | |
2674 | continue; | |
56bec294 CM |
2675 | } |
2676 | } | |
a28ec197 | 2677 | |
2c3cf7d5 FM |
2678 | /* |
2679 | * We need to try and merge add/drops of the same ref since we | |
2680 | * can run into issues with relocate dropping the implicit ref | |
2681 | * and then it being added back again before the drop can | |
2682 | * finish. If we merged anything we need to re-loop so we can | |
2683 | * get a good ref. | |
2684 | * Or we can get node references of the same type that weren't | |
2685 | * merged when created due to bumps in the tree mod seq, and | |
2686 | * we need to merge them to prevent adding an inline extent | |
2687 | * backref before dropping it (triggering a BUG_ON at | |
2688 | * insert_inline_extent_backref()). | |
2689 | */ | |
d7df2c79 | 2690 | spin_lock(&locked_ref->lock); |
be97f133 | 2691 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2692 | |
d1270cd9 AJ |
2693 | ref = select_delayed_ref(locked_ref); |
2694 | ||
2695 | if (ref && ref->seq && | |
41d0bd3b | 2696 | btrfs_check_delayed_seq(fs_info, ref->seq)) { |
d7df2c79 | 2697 | spin_unlock(&locked_ref->lock); |
2eadaa22 | 2698 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
d7df2c79 | 2699 | locked_ref = NULL; |
d1270cd9 | 2700 | cond_resched(); |
27a377db | 2701 | count++; |
d1270cd9 AJ |
2702 | continue; |
2703 | } | |
2704 | ||
c1103f7a JB |
2705 | /* |
2706 | * We're done processing refs in this ref_head, clean everything | |
2707 | * up and move on to the next ref_head. | |
2708 | */ | |
56bec294 | 2709 | if (!ref) { |
194ab0bc JB |
2710 | ret = cleanup_ref_head(trans, fs_info, locked_ref); |
2711 | if (ret > 0 ) { | |
b00e6250 JB |
2712 | /* We dropped our lock, we need to loop. */ |
2713 | ret = 0; | |
d7df2c79 | 2714 | continue; |
194ab0bc JB |
2715 | } else if (ret) { |
2716 | return ret; | |
5d4f98a2 | 2717 | } |
c1103f7a JB |
2718 | locked_ref = NULL; |
2719 | count++; | |
2720 | continue; | |
2721 | } | |
02217ed2 | 2722 | |
c1103f7a JB |
2723 | actual_count++; |
2724 | ref->in_tree = 0; | |
0e0adbcf JB |
2725 | rb_erase(&ref->ref_node, &locked_ref->ref_tree); |
2726 | RB_CLEAR_NODE(&ref->ref_node); | |
c1103f7a JB |
2727 | if (!list_empty(&ref->add_list)) |
2728 | list_del(&ref->add_list); | |
2729 | /* | |
2730 | * When we play the delayed ref, also correct the ref_mod on | |
2731 | * head | |
2732 | */ | |
2733 | switch (ref->action) { | |
2734 | case BTRFS_ADD_DELAYED_REF: | |
2735 | case BTRFS_ADD_DELAYED_EXTENT: | |
d278850e | 2736 | locked_ref->ref_mod -= ref->ref_mod; |
c1103f7a JB |
2737 | break; |
2738 | case BTRFS_DROP_DELAYED_REF: | |
d278850e | 2739 | locked_ref->ref_mod += ref->ref_mod; |
c1103f7a JB |
2740 | break; |
2741 | default: | |
2742 | WARN_ON(1); | |
22cd2e7d | 2743 | } |
1ce7a5ec JB |
2744 | atomic_dec(&delayed_refs->num_entries); |
2745 | ||
b00e6250 JB |
2746 | /* |
2747 | * Record the must-insert_reserved flag before we drop the spin | |
2748 | * lock. | |
2749 | */ | |
2750 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2751 | locked_ref->must_insert_reserved = 0; | |
2752 | ||
2753 | extent_op = locked_ref->extent_op; | |
2754 | locked_ref->extent_op = NULL; | |
d7df2c79 | 2755 | spin_unlock(&locked_ref->lock); |
925baedd | 2756 | |
2ff7e61e | 2757 | ret = run_one_delayed_ref(trans, fs_info, ref, extent_op, |
56bec294 | 2758 | must_insert_reserved); |
eb099670 | 2759 | |
78a6184a | 2760 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2761 | if (ret) { |
2eadaa22 | 2762 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
093486c4 | 2763 | btrfs_put_delayed_ref(ref); |
5d163e0e JM |
2764 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
2765 | ret); | |
79787eaa JM |
2766 | return ret; |
2767 | } | |
2768 | ||
093486c4 MX |
2769 | btrfs_put_delayed_ref(ref); |
2770 | count++; | |
c3e69d58 | 2771 | cond_resched(); |
c3e69d58 | 2772 | } |
0a2b2a84 JB |
2773 | |
2774 | /* | |
2775 | * We don't want to include ref heads since we can have empty ref heads | |
2776 | * and those will drastically skew our runtime down since we just do | |
2777 | * accounting, no actual extent tree updates. | |
2778 | */ | |
2779 | if (actual_count > 0) { | |
2780 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2781 | u64 avg; | |
2782 | ||
2783 | /* | |
2784 | * We weigh the current average higher than our current runtime | |
2785 | * to avoid large swings in the average. | |
2786 | */ | |
2787 | spin_lock(&delayed_refs->lock); | |
2788 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2789 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2790 | spin_unlock(&delayed_refs->lock); |
2791 | } | |
d7df2c79 | 2792 | return 0; |
c3e69d58 CM |
2793 | } |
2794 | ||
709c0486 AJ |
2795 | #ifdef SCRAMBLE_DELAYED_REFS |
2796 | /* | |
2797 | * Normally delayed refs get processed in ascending bytenr order. This | |
2798 | * correlates in most cases to the order added. To expose dependencies on this | |
2799 | * order, we start to process the tree in the middle instead of the beginning | |
2800 | */ | |
2801 | static u64 find_middle(struct rb_root *root) | |
2802 | { | |
2803 | struct rb_node *n = root->rb_node; | |
2804 | struct btrfs_delayed_ref_node *entry; | |
2805 | int alt = 1; | |
2806 | u64 middle; | |
2807 | u64 first = 0, last = 0; | |
2808 | ||
2809 | n = rb_first(root); | |
2810 | if (n) { | |
2811 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2812 | first = entry->bytenr; | |
2813 | } | |
2814 | n = rb_last(root); | |
2815 | if (n) { | |
2816 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2817 | last = entry->bytenr; | |
2818 | } | |
2819 | n = root->rb_node; | |
2820 | ||
2821 | while (n) { | |
2822 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2823 | WARN_ON(!entry->in_tree); | |
2824 | ||
2825 | middle = entry->bytenr; | |
2826 | ||
2827 | if (alt) | |
2828 | n = n->rb_left; | |
2829 | else | |
2830 | n = n->rb_right; | |
2831 | ||
2832 | alt = 1 - alt; | |
2833 | } | |
2834 | return middle; | |
2835 | } | |
2836 | #endif | |
2837 | ||
2ff7e61e | 2838 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2839 | { |
2840 | u64 num_bytes; | |
2841 | ||
2842 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2843 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2844 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2845 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2846 | ||
2847 | /* | |
2848 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2849 | * closer to what we're really going to want to use. |
1be41b78 | 2850 | */ |
0b246afa | 2851 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2852 | } |
2853 | ||
1262133b JB |
2854 | /* |
2855 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2856 | * would require to store the csums for that many bytes. | |
2857 | */ | |
2ff7e61e | 2858 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2859 | { |
2860 | u64 csum_size; | |
2861 | u64 num_csums_per_leaf; | |
2862 | u64 num_csums; | |
2863 | ||
0b246afa | 2864 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2865 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2866 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2867 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2868 | num_csums += num_csums_per_leaf - 1; |
2869 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2870 | return num_csums; | |
2871 | } | |
2872 | ||
0a2b2a84 | 2873 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2874 | struct btrfs_fs_info *fs_info) |
1be41b78 JB |
2875 | { |
2876 | struct btrfs_block_rsv *global_rsv; | |
2877 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2878 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
165c8b02 | 2879 | unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs; |
cb723e49 | 2880 | u64 num_bytes, num_dirty_bgs_bytes; |
1be41b78 JB |
2881 | int ret = 0; |
2882 | ||
0b246afa | 2883 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2884 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2885 | if (num_heads > 1) |
0b246afa | 2886 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2887 | num_bytes <<= 1; |
2ff7e61e JM |
2888 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2889 | fs_info->nodesize; | |
0b246afa | 2890 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2891 | num_dirty_bgs); |
0b246afa | 2892 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2893 | |
2894 | /* | |
2895 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2896 | * wiggle room since running delayed refs can create more delayed refs. | |
2897 | */ | |
cb723e49 JB |
2898 | if (global_rsv->space_info->full) { |
2899 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2900 | num_bytes <<= 1; |
cb723e49 | 2901 | } |
1be41b78 JB |
2902 | |
2903 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2904 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2905 | ret = 1; |
2906 | spin_unlock(&global_rsv->lock); | |
2907 | return ret; | |
2908 | } | |
2909 | ||
0a2b2a84 | 2910 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2911 | struct btrfs_fs_info *fs_info) |
0a2b2a84 | 2912 | { |
0a2b2a84 JB |
2913 | u64 num_entries = |
2914 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2915 | u64 avg_runtime; | |
a79b7d4b | 2916 | u64 val; |
0a2b2a84 JB |
2917 | |
2918 | smp_mb(); | |
2919 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2920 | val = num_entries * avg_runtime; |
dc1a90c6 | 2921 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2922 | return 1; |
a79b7d4b CM |
2923 | if (val >= NSEC_PER_SEC / 2) |
2924 | return 2; | |
0a2b2a84 | 2925 | |
2ff7e61e | 2926 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
0a2b2a84 JB |
2927 | } |
2928 | ||
a79b7d4b CM |
2929 | struct async_delayed_refs { |
2930 | struct btrfs_root *root; | |
31b9655f | 2931 | u64 transid; |
a79b7d4b CM |
2932 | int count; |
2933 | int error; | |
2934 | int sync; | |
2935 | struct completion wait; | |
2936 | struct btrfs_work work; | |
2937 | }; | |
2938 | ||
2ff7e61e JM |
2939 | static inline struct async_delayed_refs * |
2940 | to_async_delayed_refs(struct btrfs_work *work) | |
2941 | { | |
2942 | return container_of(work, struct async_delayed_refs, work); | |
2943 | } | |
2944 | ||
a79b7d4b CM |
2945 | static void delayed_ref_async_start(struct btrfs_work *work) |
2946 | { | |
2ff7e61e | 2947 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2948 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2949 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2950 | int ret; |
2951 | ||
0f873eca | 2952 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2953 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2954 | goto done; |
31b9655f | 2955 | |
0f873eca CM |
2956 | trans = btrfs_join_transaction(async->root); |
2957 | if (IS_ERR(trans)) { | |
2958 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2959 | goto done; |
2960 | } | |
2961 | ||
2962 | /* | |
01327610 | 2963 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2964 | * wait on delayed refs |
2965 | */ | |
2966 | trans->sync = true; | |
0f873eca CM |
2967 | |
2968 | /* Don't bother flushing if we got into a different transaction */ | |
2969 | if (trans->transid > async->transid) | |
2970 | goto end; | |
2971 | ||
c79a70b1 | 2972 | ret = btrfs_run_delayed_refs(trans, async->count); |
a79b7d4b CM |
2973 | if (ret) |
2974 | async->error = ret; | |
0f873eca | 2975 | end: |
3a45bb20 | 2976 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
2977 | if (ret && !async->error) |
2978 | async->error = ret; | |
2979 | done: | |
2980 | if (async->sync) | |
2981 | complete(&async->wait); | |
2982 | else | |
2983 | kfree(async); | |
2984 | } | |
2985 | ||
2ff7e61e | 2986 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 2987 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
2988 | { |
2989 | struct async_delayed_refs *async; | |
2990 | int ret; | |
2991 | ||
2992 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2993 | if (!async) | |
2994 | return -ENOMEM; | |
2995 | ||
0b246afa | 2996 | async->root = fs_info->tree_root; |
a79b7d4b CM |
2997 | async->count = count; |
2998 | async->error = 0; | |
31b9655f | 2999 | async->transid = transid; |
a79b7d4b CM |
3000 | if (wait) |
3001 | async->sync = 1; | |
3002 | else | |
3003 | async->sync = 0; | |
3004 | init_completion(&async->wait); | |
3005 | ||
9e0af237 LB |
3006 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
3007 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 3008 | |
0b246afa | 3009 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
3010 | |
3011 | if (wait) { | |
3012 | wait_for_completion(&async->wait); | |
3013 | ret = async->error; | |
3014 | kfree(async); | |
3015 | return ret; | |
3016 | } | |
3017 | return 0; | |
3018 | } | |
3019 | ||
c3e69d58 CM |
3020 | /* |
3021 | * this starts processing the delayed reference count updates and | |
3022 | * extent insertions we have queued up so far. count can be | |
3023 | * 0, which means to process everything in the tree at the start | |
3024 | * of the run (but not newly added entries), or it can be some target | |
3025 | * number you'd like to process. | |
79787eaa JM |
3026 | * |
3027 | * Returns 0 on success or if called with an aborted transaction | |
3028 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
3029 | */ |
3030 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 3031 | unsigned long count) |
c3e69d58 | 3032 | { |
c79a70b1 | 3033 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
3034 | struct rb_node *node; |
3035 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 3036 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
3037 | int ret; |
3038 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 3039 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 3040 | |
79787eaa JM |
3041 | /* We'll clean this up in btrfs_cleanup_transaction */ |
3042 | if (trans->aborted) | |
3043 | return 0; | |
3044 | ||
0b246afa | 3045 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
3046 | return 0; |
3047 | ||
c3e69d58 | 3048 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 3049 | if (count == 0) |
d7df2c79 | 3050 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 3051 | |
c3e69d58 | 3052 | again: |
709c0486 AJ |
3053 | #ifdef SCRAMBLE_DELAYED_REFS |
3054 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
3055 | #endif | |
d9a0540a | 3056 | trans->can_flush_pending_bgs = false; |
0a1e458a | 3057 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 3058 | if (ret < 0) { |
66642832 | 3059 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 3060 | return ret; |
eb099670 | 3061 | } |
c3e69d58 | 3062 | |
56bec294 | 3063 | if (run_all) { |
d7df2c79 | 3064 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 3065 | btrfs_create_pending_block_groups(trans); |
ea658bad | 3066 | |
d7df2c79 | 3067 | spin_lock(&delayed_refs->lock); |
c46effa6 | 3068 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
3069 | if (!node) { |
3070 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 3071 | goto out; |
d7df2c79 | 3072 | } |
d278850e JB |
3073 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
3074 | href_node); | |
3075 | refcount_inc(&head->refs); | |
3076 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 3077 | |
d278850e JB |
3078 | /* Mutex was contended, block until it's released and retry. */ |
3079 | mutex_lock(&head->mutex); | |
3080 | mutex_unlock(&head->mutex); | |
56bec294 | 3081 | |
d278850e | 3082 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 3083 | cond_resched(); |
56bec294 | 3084 | goto again; |
5f39d397 | 3085 | } |
54aa1f4d | 3086 | out: |
d9a0540a | 3087 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3088 | return 0; |
3089 | } | |
3090 | ||
5d4f98a2 | 3091 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 3092 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 3093 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 3094 | int level, int is_data) |
5d4f98a2 YZ |
3095 | { |
3096 | struct btrfs_delayed_extent_op *extent_op; | |
3097 | int ret; | |
3098 | ||
78a6184a | 3099 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3100 | if (!extent_op) |
3101 | return -ENOMEM; | |
3102 | ||
3103 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3104 | extent_op->update_flags = true; |
3105 | extent_op->update_key = false; | |
3106 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3107 | extent_op->level = level; |
5d4f98a2 | 3108 | |
0b246afa | 3109 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 3110 | num_bytes, extent_op); |
5d4f98a2 | 3111 | if (ret) |
78a6184a | 3112 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3113 | return ret; |
3114 | } | |
3115 | ||
e4c3b2dc | 3116 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3117 | struct btrfs_path *path, |
3118 | u64 objectid, u64 offset, u64 bytenr) | |
3119 | { | |
3120 | struct btrfs_delayed_ref_head *head; | |
3121 | struct btrfs_delayed_ref_node *ref; | |
3122 | struct btrfs_delayed_data_ref *data_ref; | |
3123 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 3124 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 3125 | struct rb_node *node; |
5d4f98a2 YZ |
3126 | int ret = 0; |
3127 | ||
998ac6d2 | 3128 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 3129 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 3130 | if (cur_trans) |
3131 | refcount_inc(&cur_trans->use_count); | |
3132 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
3133 | if (!cur_trans) |
3134 | return 0; | |
3135 | ||
3136 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3137 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3138 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3139 | if (!head) { |
3140 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 3141 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
3142 | return 0; |
3143 | } | |
5d4f98a2 YZ |
3144 | |
3145 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 3146 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
3147 | spin_unlock(&delayed_refs->lock); |
3148 | ||
b3b4aa74 | 3149 | btrfs_release_path(path); |
5d4f98a2 | 3150 | |
8cc33e5c DS |
3151 | /* |
3152 | * Mutex was contended, block until it's released and let | |
3153 | * caller try again | |
3154 | */ | |
5d4f98a2 YZ |
3155 | mutex_lock(&head->mutex); |
3156 | mutex_unlock(&head->mutex); | |
d278850e | 3157 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 3158 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3159 | return -EAGAIN; |
3160 | } | |
d7df2c79 | 3161 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3162 | |
d7df2c79 | 3163 | spin_lock(&head->lock); |
0e0adbcf JB |
3164 | /* |
3165 | * XXX: We should replace this with a proper search function in the | |
3166 | * future. | |
3167 | */ | |
3168 | for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) { | |
3169 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); | |
d7df2c79 JB |
3170 | /* If it's a shared ref we know a cross reference exists */ |
3171 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3172 | ret = 1; | |
3173 | break; | |
3174 | } | |
5d4f98a2 | 3175 | |
d7df2c79 | 3176 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3177 | |
d7df2c79 JB |
3178 | /* |
3179 | * If our ref doesn't match the one we're currently looking at | |
3180 | * then we have a cross reference. | |
3181 | */ | |
3182 | if (data_ref->root != root->root_key.objectid || | |
3183 | data_ref->objectid != objectid || | |
3184 | data_ref->offset != offset) { | |
3185 | ret = 1; | |
3186 | break; | |
3187 | } | |
5d4f98a2 | 3188 | } |
d7df2c79 | 3189 | spin_unlock(&head->lock); |
5d4f98a2 | 3190 | mutex_unlock(&head->mutex); |
998ac6d2 | 3191 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3192 | return ret; |
3193 | } | |
3194 | ||
e4c3b2dc | 3195 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3196 | struct btrfs_path *path, |
3197 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3198 | { |
0b246afa JM |
3199 | struct btrfs_fs_info *fs_info = root->fs_info; |
3200 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3201 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3202 | struct btrfs_extent_data_ref *ref; |
3203 | struct btrfs_extent_inline_ref *iref; | |
3204 | struct btrfs_extent_item *ei; | |
f321e491 | 3205 | struct btrfs_key key; |
5d4f98a2 | 3206 | u32 item_size; |
3de28d57 | 3207 | int type; |
be20aa9d | 3208 | int ret; |
925baedd | 3209 | |
be20aa9d | 3210 | key.objectid = bytenr; |
31840ae1 | 3211 | key.offset = (u64)-1; |
f321e491 | 3212 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3213 | |
be20aa9d CM |
3214 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3215 | if (ret < 0) | |
3216 | goto out; | |
79787eaa | 3217 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3218 | |
3219 | ret = -ENOENT; | |
3220 | if (path->slots[0] == 0) | |
31840ae1 | 3221 | goto out; |
be20aa9d | 3222 | |
31840ae1 | 3223 | path->slots[0]--; |
f321e491 | 3224 | leaf = path->nodes[0]; |
5d4f98a2 | 3225 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3226 | |
5d4f98a2 | 3227 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3228 | goto out; |
f321e491 | 3229 | |
5d4f98a2 YZ |
3230 | ret = 1; |
3231 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3232 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3233 | if (item_size < sizeof(*ei)) { | |
3234 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3235 | goto out; | |
3236 | } | |
3237 | #endif | |
3238 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3239 | |
5d4f98a2 YZ |
3240 | if (item_size != sizeof(*ei) + |
3241 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3242 | goto out; | |
be20aa9d | 3243 | |
5d4f98a2 YZ |
3244 | if (btrfs_extent_generation(leaf, ei) <= |
3245 | btrfs_root_last_snapshot(&root->root_item)) | |
3246 | goto out; | |
3247 | ||
3248 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3249 | |
3250 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3251 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3252 | goto out; |
3253 | ||
3254 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3255 | if (btrfs_extent_refs(leaf, ei) != | |
3256 | btrfs_extent_data_ref_count(leaf, ref) || | |
3257 | btrfs_extent_data_ref_root(leaf, ref) != | |
3258 | root->root_key.objectid || | |
3259 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3260 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3261 | goto out; | |
3262 | ||
3263 | ret = 0; | |
3264 | out: | |
3265 | return ret; | |
3266 | } | |
3267 | ||
e4c3b2dc LB |
3268 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3269 | u64 bytenr) | |
5d4f98a2 YZ |
3270 | { |
3271 | struct btrfs_path *path; | |
3272 | int ret; | |
3273 | int ret2; | |
3274 | ||
3275 | path = btrfs_alloc_path(); | |
3276 | if (!path) | |
3277 | return -ENOENT; | |
3278 | ||
3279 | do { | |
e4c3b2dc | 3280 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3281 | offset, bytenr); |
3282 | if (ret && ret != -ENOENT) | |
f321e491 | 3283 | goto out; |
80ff3856 | 3284 | |
e4c3b2dc | 3285 | ret2 = check_delayed_ref(root, path, objectid, |
5d4f98a2 YZ |
3286 | offset, bytenr); |
3287 | } while (ret2 == -EAGAIN); | |
3288 | ||
3289 | if (ret2 && ret2 != -ENOENT) { | |
3290 | ret = ret2; | |
3291 | goto out; | |
f321e491 | 3292 | } |
5d4f98a2 YZ |
3293 | |
3294 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3295 | ret = 0; | |
be20aa9d | 3296 | out: |
80ff3856 | 3297 | btrfs_free_path(path); |
f0486c68 YZ |
3298 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3299 | WARN_ON(ret > 0); | |
f321e491 | 3300 | return ret; |
be20aa9d | 3301 | } |
c5739bba | 3302 | |
5d4f98a2 | 3303 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3304 | struct btrfs_root *root, |
5d4f98a2 | 3305 | struct extent_buffer *buf, |
e339a6b0 | 3306 | int full_backref, int inc) |
31840ae1 | 3307 | { |
0b246afa | 3308 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3309 | u64 bytenr; |
5d4f98a2 YZ |
3310 | u64 num_bytes; |
3311 | u64 parent; | |
31840ae1 | 3312 | u64 ref_root; |
31840ae1 | 3313 | u32 nritems; |
31840ae1 ZY |
3314 | struct btrfs_key key; |
3315 | struct btrfs_file_extent_item *fi; | |
3316 | int i; | |
3317 | int level; | |
3318 | int ret = 0; | |
2ff7e61e | 3319 | int (*process_func)(struct btrfs_trans_handle *, |
84f7d8e6 | 3320 | struct btrfs_root *, |
b06c4bf5 | 3321 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3322 | |
fccb84c9 | 3323 | |
0b246afa | 3324 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3325 | return 0; |
fccb84c9 | 3326 | |
31840ae1 | 3327 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3328 | nritems = btrfs_header_nritems(buf); |
3329 | level = btrfs_header_level(buf); | |
3330 | ||
27cdeb70 | 3331 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3332 | return 0; |
31840ae1 | 3333 | |
5d4f98a2 YZ |
3334 | if (inc) |
3335 | process_func = btrfs_inc_extent_ref; | |
3336 | else | |
3337 | process_func = btrfs_free_extent; | |
31840ae1 | 3338 | |
5d4f98a2 YZ |
3339 | if (full_backref) |
3340 | parent = buf->start; | |
3341 | else | |
3342 | parent = 0; | |
3343 | ||
3344 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3345 | if (level == 0) { |
5d4f98a2 | 3346 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3347 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3348 | continue; |
5d4f98a2 | 3349 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3350 | struct btrfs_file_extent_item); |
3351 | if (btrfs_file_extent_type(buf, fi) == | |
3352 | BTRFS_FILE_EXTENT_INLINE) | |
3353 | continue; | |
3354 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3355 | if (bytenr == 0) | |
3356 | continue; | |
5d4f98a2 YZ |
3357 | |
3358 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3359 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
84f7d8e6 | 3360 | ret = process_func(trans, root, bytenr, num_bytes, |
5d4f98a2 | 3361 | parent, ref_root, key.objectid, |
b06c4bf5 | 3362 | key.offset); |
31840ae1 ZY |
3363 | if (ret) |
3364 | goto fail; | |
3365 | } else { | |
5d4f98a2 | 3366 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3367 | num_bytes = fs_info->nodesize; |
84f7d8e6 | 3368 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3369 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3370 | if (ret) |
3371 | goto fail; | |
3372 | } | |
3373 | } | |
3374 | return 0; | |
3375 | fail: | |
5d4f98a2 YZ |
3376 | return ret; |
3377 | } | |
3378 | ||
3379 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3380 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3381 | { |
e339a6b0 | 3382 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3383 | } |
3384 | ||
3385 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3386 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3387 | { |
e339a6b0 | 3388 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3389 | } |
3390 | ||
9078a3e1 | 3391 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3392 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3393 | struct btrfs_path *path, |
3394 | struct btrfs_block_group_cache *cache) | |
3395 | { | |
3396 | int ret; | |
0b246afa | 3397 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3398 | unsigned long bi; |
3399 | struct extent_buffer *leaf; | |
9078a3e1 | 3400 | |
9078a3e1 | 3401 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3402 | if (ret) { |
3403 | if (ret > 0) | |
3404 | ret = -ENOENT; | |
54aa1f4d | 3405 | goto fail; |
df95e7f0 | 3406 | } |
5f39d397 CM |
3407 | |
3408 | leaf = path->nodes[0]; | |
3409 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3410 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3411 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3412 | fail: |
24b89d08 | 3413 | btrfs_release_path(path); |
df95e7f0 | 3414 | return ret; |
9078a3e1 CM |
3415 | |
3416 | } | |
3417 | ||
4a8c9a62 | 3418 | static struct btrfs_block_group_cache * |
2ff7e61e | 3419 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3420 | struct btrfs_block_group_cache *cache) |
3421 | { | |
3422 | struct rb_node *node; | |
292cbd51 | 3423 | |
0b246afa | 3424 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3425 | |
3426 | /* If our block group was removed, we need a full search. */ | |
3427 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3428 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3429 | ||
0b246afa | 3430 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3431 | btrfs_put_block_group(cache); |
0b246afa | 3432 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3433 | } |
4a8c9a62 YZ |
3434 | node = rb_next(&cache->cache_node); |
3435 | btrfs_put_block_group(cache); | |
3436 | if (node) { | |
3437 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3438 | cache_node); | |
11dfe35a | 3439 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3440 | } else |
3441 | cache = NULL; | |
0b246afa | 3442 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3443 | return cache; |
3444 | } | |
3445 | ||
0af3d00b JB |
3446 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3447 | struct btrfs_trans_handle *trans, | |
3448 | struct btrfs_path *path) | |
3449 | { | |
0b246afa JM |
3450 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3451 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3452 | struct inode *inode = NULL; |
364ecf36 | 3453 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3454 | u64 alloc_hint = 0; |
2b20982e | 3455 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3456 | u64 num_pages = 0; |
0af3d00b JB |
3457 | int retries = 0; |
3458 | int ret = 0; | |
3459 | ||
3460 | /* | |
3461 | * If this block group is smaller than 100 megs don't bother caching the | |
3462 | * block group. | |
3463 | */ | |
ee22184b | 3464 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3465 | spin_lock(&block_group->lock); |
3466 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3467 | spin_unlock(&block_group->lock); | |
3468 | return 0; | |
3469 | } | |
3470 | ||
0c0ef4bc JB |
3471 | if (trans->aborted) |
3472 | return 0; | |
0af3d00b | 3473 | again: |
77ab86bf | 3474 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3475 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3476 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3477 | btrfs_release_path(path); |
0af3d00b JB |
3478 | goto out; |
3479 | } | |
3480 | ||
3481 | if (IS_ERR(inode)) { | |
3482 | BUG_ON(retries); | |
3483 | retries++; | |
3484 | ||
3485 | if (block_group->ro) | |
3486 | goto out_free; | |
3487 | ||
77ab86bf JM |
3488 | ret = create_free_space_inode(fs_info, trans, block_group, |
3489 | path); | |
0af3d00b JB |
3490 | if (ret) |
3491 | goto out_free; | |
3492 | goto again; | |
3493 | } | |
3494 | ||
3495 | /* | |
3496 | * We want to set the generation to 0, that way if anything goes wrong | |
3497 | * from here on out we know not to trust this cache when we load up next | |
3498 | * time. | |
3499 | */ | |
3500 | BTRFS_I(inode)->generation = 0; | |
3501 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3502 | if (ret) { |
3503 | /* | |
3504 | * So theoretically we could recover from this, simply set the | |
3505 | * super cache generation to 0 so we know to invalidate the | |
3506 | * cache, but then we'd have to keep track of the block groups | |
3507 | * that fail this way so we know we _have_ to reset this cache | |
3508 | * before the next commit or risk reading stale cache. So to | |
3509 | * limit our exposure to horrible edge cases lets just abort the | |
3510 | * transaction, this only happens in really bad situations | |
3511 | * anyway. | |
3512 | */ | |
66642832 | 3513 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3514 | goto out_put; |
3515 | } | |
0af3d00b JB |
3516 | WARN_ON(ret); |
3517 | ||
8e138e0d JB |
3518 | /* We've already setup this transaction, go ahead and exit */ |
3519 | if (block_group->cache_generation == trans->transid && | |
3520 | i_size_read(inode)) { | |
3521 | dcs = BTRFS_DC_SETUP; | |
3522 | goto out_put; | |
3523 | } | |
3524 | ||
0af3d00b | 3525 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3526 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3527 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3528 | if (ret) |
3529 | goto out_put; | |
3530 | ||
77ab86bf | 3531 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3532 | if (ret) |
3533 | goto out_put; | |
3534 | } | |
3535 | ||
3536 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3537 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3538 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3539 | /* |
3540 | * don't bother trying to write stuff out _if_ | |
3541 | * a) we're not cached, | |
1a79c1f2 LB |
3542 | * b) we're with nospace_cache mount option, |
3543 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3544 | */ |
2b20982e | 3545 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3546 | spin_unlock(&block_group->lock); |
3547 | goto out_put; | |
3548 | } | |
3549 | spin_unlock(&block_group->lock); | |
3550 | ||
2968b1f4 JB |
3551 | /* |
3552 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3553 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3554 | */ | |
3555 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3556 | ret = -ENOSPC; | |
3557 | goto out_put; | |
3558 | } | |
3559 | ||
6fc823b1 JB |
3560 | /* |
3561 | * Try to preallocate enough space based on how big the block group is. | |
3562 | * Keep in mind this has to include any pinned space which could end up | |
3563 | * taking up quite a bit since it's not folded into the other space | |
3564 | * cache. | |
3565 | */ | |
ee22184b | 3566 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3567 | if (!num_pages) |
3568 | num_pages = 1; | |
3569 | ||
0af3d00b | 3570 | num_pages *= 16; |
09cbfeaf | 3571 | num_pages *= PAGE_SIZE; |
0af3d00b | 3572 | |
364ecf36 | 3573 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3574 | if (ret) |
3575 | goto out_put; | |
3576 | ||
3577 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3578 | num_pages, num_pages, | |
3579 | &alloc_hint); | |
2968b1f4 JB |
3580 | /* |
3581 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3582 | * of metadata or split extents when writing the cache out, which means | |
3583 | * we can enospc if we are heavily fragmented in addition to just normal | |
3584 | * out of space conditions. So if we hit this just skip setting up any | |
3585 | * other block groups for this transaction, maybe we'll unpin enough | |
3586 | * space the next time around. | |
3587 | */ | |
2b20982e JB |
3588 | if (!ret) |
3589 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3590 | else if (ret == -ENOSPC) |
3591 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3592 | |
0af3d00b JB |
3593 | out_put: |
3594 | iput(inode); | |
3595 | out_free: | |
b3b4aa74 | 3596 | btrfs_release_path(path); |
0af3d00b JB |
3597 | out: |
3598 | spin_lock(&block_group->lock); | |
e65cbb94 | 3599 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3600 | block_group->cache_generation = trans->transid; |
2b20982e | 3601 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3602 | spin_unlock(&block_group->lock); |
3603 | ||
364ecf36 | 3604 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3605 | return ret; |
3606 | } | |
3607 | ||
dcdf7f6d | 3608 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3609 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3610 | { |
3611 | struct btrfs_block_group_cache *cache, *tmp; | |
3612 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3613 | struct btrfs_path *path; | |
3614 | ||
3615 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3616 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3617 | return 0; |
3618 | ||
3619 | path = btrfs_alloc_path(); | |
3620 | if (!path) | |
3621 | return -ENOMEM; | |
3622 | ||
3623 | /* Could add new block groups, use _safe just in case */ | |
3624 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3625 | dirty_list) { | |
3626 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3627 | cache_save_setup(cache, trans, path); | |
3628 | } | |
3629 | ||
3630 | btrfs_free_path(path); | |
3631 | return 0; | |
3632 | } | |
3633 | ||
1bbc621e CM |
3634 | /* |
3635 | * transaction commit does final block group cache writeback during a | |
3636 | * critical section where nothing is allowed to change the FS. This is | |
3637 | * required in order for the cache to actually match the block group, | |
3638 | * but can introduce a lot of latency into the commit. | |
3639 | * | |
3640 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3641 | * cache IO. There's a chance we'll have to redo some of it if the | |
3642 | * block group changes again during the commit, but it greatly reduces | |
3643 | * the commit latency by getting rid of the easy block groups while | |
3644 | * we're still allowing others to join the commit. | |
3645 | */ | |
21217054 | 3646 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3647 | { |
21217054 | 3648 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3649 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3650 | struct btrfs_transaction *cur_trans = trans->transaction; |
3651 | int ret = 0; | |
c9dc4c65 | 3652 | int should_put; |
1bbc621e CM |
3653 | struct btrfs_path *path = NULL; |
3654 | LIST_HEAD(dirty); | |
3655 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3656 | int num_started = 0; |
1bbc621e CM |
3657 | int loops = 0; |
3658 | ||
3659 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3660 | if (list_empty(&cur_trans->dirty_bgs)) { |
3661 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3662 | return 0; | |
1bbc621e | 3663 | } |
b58d1a9e | 3664 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3665 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3666 | |
1bbc621e | 3667 | again: |
1bbc621e CM |
3668 | /* |
3669 | * make sure all the block groups on our dirty list actually | |
3670 | * exist | |
3671 | */ | |
6c686b35 | 3672 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3673 | |
3674 | if (!path) { | |
3675 | path = btrfs_alloc_path(); | |
3676 | if (!path) | |
3677 | return -ENOMEM; | |
3678 | } | |
3679 | ||
b58d1a9e FM |
3680 | /* |
3681 | * cache_write_mutex is here only to save us from balance or automatic | |
3682 | * removal of empty block groups deleting this block group while we are | |
3683 | * writing out the cache | |
3684 | */ | |
3685 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3686 | while (!list_empty(&dirty)) { |
3687 | cache = list_first_entry(&dirty, | |
3688 | struct btrfs_block_group_cache, | |
3689 | dirty_list); | |
1bbc621e CM |
3690 | /* |
3691 | * this can happen if something re-dirties a block | |
3692 | * group that is already under IO. Just wait for it to | |
3693 | * finish and then do it all again | |
3694 | */ | |
3695 | if (!list_empty(&cache->io_list)) { | |
3696 | list_del_init(&cache->io_list); | |
afdb5718 | 3697 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3698 | btrfs_put_block_group(cache); |
3699 | } | |
3700 | ||
3701 | ||
3702 | /* | |
3703 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3704 | * if it should update the cache_state. Don't delete | |
3705 | * until after we wait. | |
3706 | * | |
3707 | * Since we're not running in the commit critical section | |
3708 | * we need the dirty_bgs_lock to protect from update_block_group | |
3709 | */ | |
3710 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3711 | list_del_init(&cache->dirty_list); | |
3712 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3713 | ||
3714 | should_put = 1; | |
3715 | ||
3716 | cache_save_setup(cache, trans, path); | |
3717 | ||
3718 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3719 | cache->io_ctl.inode = NULL; | |
0b246afa | 3720 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3721 | cache, path); |
1bbc621e CM |
3722 | if (ret == 0 && cache->io_ctl.inode) { |
3723 | num_started++; | |
3724 | should_put = 0; | |
3725 | ||
3726 | /* | |
45ae2c18 NB |
3727 | * The cache_write_mutex is protecting the |
3728 | * io_list, also refer to the definition of | |
3729 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3730 | */ |
3731 | list_add_tail(&cache->io_list, io); | |
3732 | } else { | |
3733 | /* | |
3734 | * if we failed to write the cache, the | |
3735 | * generation will be bad and life goes on | |
3736 | */ | |
3737 | ret = 0; | |
3738 | } | |
3739 | } | |
ff1f8250 | 3740 | if (!ret) { |
2ff7e61e JM |
3741 | ret = write_one_cache_group(trans, fs_info, |
3742 | path, cache); | |
ff1f8250 FM |
3743 | /* |
3744 | * Our block group might still be attached to the list | |
3745 | * of new block groups in the transaction handle of some | |
3746 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3747 | * means its block group item isn't yet in the extent | |
3748 | * tree. If this happens ignore the error, as we will | |
3749 | * try again later in the critical section of the | |
3750 | * transaction commit. | |
3751 | */ | |
3752 | if (ret == -ENOENT) { | |
3753 | ret = 0; | |
3754 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3755 | if (list_empty(&cache->dirty_list)) { | |
3756 | list_add_tail(&cache->dirty_list, | |
3757 | &cur_trans->dirty_bgs); | |
3758 | btrfs_get_block_group(cache); | |
3759 | } | |
3760 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3761 | } else if (ret) { | |
66642832 | 3762 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3763 | } |
3764 | } | |
1bbc621e CM |
3765 | |
3766 | /* if its not on the io list, we need to put the block group */ | |
3767 | if (should_put) | |
3768 | btrfs_put_block_group(cache); | |
3769 | ||
3770 | if (ret) | |
3771 | break; | |
b58d1a9e FM |
3772 | |
3773 | /* | |
3774 | * Avoid blocking other tasks for too long. It might even save | |
3775 | * us from writing caches for block groups that are going to be | |
3776 | * removed. | |
3777 | */ | |
3778 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3779 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3780 | } |
b58d1a9e | 3781 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3782 | |
3783 | /* | |
3784 | * go through delayed refs for all the stuff we've just kicked off | |
3785 | * and then loop back (just once) | |
3786 | */ | |
c79a70b1 | 3787 | ret = btrfs_run_delayed_refs(trans, 0); |
1bbc621e CM |
3788 | if (!ret && loops == 0) { |
3789 | loops++; | |
3790 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3791 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3792 | /* |
3793 | * dirty_bgs_lock protects us from concurrent block group | |
3794 | * deletes too (not just cache_write_mutex). | |
3795 | */ | |
3796 | if (!list_empty(&dirty)) { | |
3797 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3798 | goto again; | |
3799 | } | |
1bbc621e | 3800 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3801 | } else if (ret < 0) { |
2ff7e61e | 3802 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3803 | } |
3804 | ||
3805 | btrfs_free_path(path); | |
3806 | return ret; | |
3807 | } | |
3808 | ||
3809 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3810 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3811 | { |
3812 | struct btrfs_block_group_cache *cache; | |
3813 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3814 | int ret = 0; | |
3815 | int should_put; | |
3816 | struct btrfs_path *path; | |
3817 | struct list_head *io = &cur_trans->io_bgs; | |
3818 | int num_started = 0; | |
9078a3e1 CM |
3819 | |
3820 | path = btrfs_alloc_path(); | |
3821 | if (!path) | |
3822 | return -ENOMEM; | |
3823 | ||
ce93ec54 | 3824 | /* |
e44081ef FM |
3825 | * Even though we are in the critical section of the transaction commit, |
3826 | * we can still have concurrent tasks adding elements to this | |
3827 | * transaction's list of dirty block groups. These tasks correspond to | |
3828 | * endio free space workers started when writeback finishes for a | |
3829 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3830 | * allocate new block groups as a result of COWing nodes of the root | |
3831 | * tree when updating the free space inode. The writeback for the space | |
3832 | * caches is triggered by an earlier call to | |
3833 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3834 | * loop. | |
3835 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3836 | * delayed refs to make sure we have the best chance at doing this all |
3837 | * in one shot. | |
3838 | */ | |
e44081ef | 3839 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3840 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3841 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3842 | struct btrfs_block_group_cache, | |
3843 | dirty_list); | |
c9dc4c65 CM |
3844 | |
3845 | /* | |
3846 | * this can happen if cache_save_setup re-dirties a block | |
3847 | * group that is already under IO. Just wait for it to | |
3848 | * finish and then do it all again | |
3849 | */ | |
3850 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3851 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3852 | list_del_init(&cache->io_list); |
afdb5718 | 3853 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3854 | btrfs_put_block_group(cache); |
e44081ef | 3855 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3856 | } |
3857 | ||
1bbc621e CM |
3858 | /* |
3859 | * don't remove from the dirty list until after we've waited | |
3860 | * on any pending IO | |
3861 | */ | |
ce93ec54 | 3862 | list_del_init(&cache->dirty_list); |
e44081ef | 3863 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3864 | should_put = 1; |
3865 | ||
1bbc621e | 3866 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3867 | |
ce93ec54 | 3868 | if (!ret) |
c79a70b1 | 3869 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 3870 | (unsigned long) -1); |
c9dc4c65 CM |
3871 | |
3872 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3873 | cache->io_ctl.inode = NULL; | |
0b246afa | 3874 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3875 | cache, path); |
c9dc4c65 CM |
3876 | if (ret == 0 && cache->io_ctl.inode) { |
3877 | num_started++; | |
3878 | should_put = 0; | |
1bbc621e | 3879 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3880 | } else { |
3881 | /* | |
3882 | * if we failed to write the cache, the | |
3883 | * generation will be bad and life goes on | |
3884 | */ | |
3885 | ret = 0; | |
3886 | } | |
3887 | } | |
ff1f8250 | 3888 | if (!ret) { |
2ff7e61e JM |
3889 | ret = write_one_cache_group(trans, fs_info, |
3890 | path, cache); | |
2bc0bb5f FM |
3891 | /* |
3892 | * One of the free space endio workers might have | |
3893 | * created a new block group while updating a free space | |
3894 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3895 | * and hasn't released its transaction handle yet, in | |
3896 | * which case the new block group is still attached to | |
3897 | * its transaction handle and its creation has not | |
3898 | * finished yet (no block group item in the extent tree | |
3899 | * yet, etc). If this is the case, wait for all free | |
3900 | * space endio workers to finish and retry. This is a | |
3901 | * a very rare case so no need for a more efficient and | |
3902 | * complex approach. | |
3903 | */ | |
3904 | if (ret == -ENOENT) { | |
3905 | wait_event(cur_trans->writer_wait, | |
3906 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3907 | ret = write_one_cache_group(trans, fs_info, |
3908 | path, cache); | |
2bc0bb5f | 3909 | } |
ff1f8250 | 3910 | if (ret) |
66642832 | 3911 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3912 | } |
c9dc4c65 CM |
3913 | |
3914 | /* if its not on the io list, we need to put the block group */ | |
3915 | if (should_put) | |
3916 | btrfs_put_block_group(cache); | |
e44081ef | 3917 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3918 | } |
e44081ef | 3919 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3920 | |
45ae2c18 NB |
3921 | /* |
3922 | * Refer to the definition of io_bgs member for details why it's safe | |
3923 | * to use it without any locking | |
3924 | */ | |
1bbc621e CM |
3925 | while (!list_empty(io)) { |
3926 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3927 | io_list); |
3928 | list_del_init(&cache->io_list); | |
afdb5718 | 3929 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3930 | btrfs_put_block_group(cache); |
3931 | } | |
3932 | ||
9078a3e1 | 3933 | btrfs_free_path(path); |
ce93ec54 | 3934 | return ret; |
9078a3e1 CM |
3935 | } |
3936 | ||
2ff7e61e | 3937 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3938 | { |
3939 | struct btrfs_block_group_cache *block_group; | |
3940 | int readonly = 0; | |
3941 | ||
0b246afa | 3942 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3943 | if (!block_group || block_group->ro) |
3944 | readonly = 1; | |
3945 | if (block_group) | |
fa9c0d79 | 3946 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3947 | return readonly; |
3948 | } | |
3949 | ||
f78c436c FM |
3950 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3951 | { | |
3952 | struct btrfs_block_group_cache *bg; | |
3953 | bool ret = true; | |
3954 | ||
3955 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3956 | if (!bg) | |
3957 | return false; | |
3958 | ||
3959 | spin_lock(&bg->lock); | |
3960 | if (bg->ro) | |
3961 | ret = false; | |
3962 | else | |
3963 | atomic_inc(&bg->nocow_writers); | |
3964 | spin_unlock(&bg->lock); | |
3965 | ||
3966 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3967 | if (!ret) | |
3968 | btrfs_put_block_group(bg); | |
3969 | ||
3970 | return ret; | |
3971 | ||
3972 | } | |
3973 | ||
3974 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3975 | { | |
3976 | struct btrfs_block_group_cache *bg; | |
3977 | ||
3978 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3979 | ASSERT(bg); | |
3980 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
4625956a | 3981 | wake_up_var(&bg->nocow_writers); |
f78c436c FM |
3982 | /* |
3983 | * Once for our lookup and once for the lookup done by a previous call | |
3984 | * to btrfs_inc_nocow_writers() | |
3985 | */ | |
3986 | btrfs_put_block_group(bg); | |
3987 | btrfs_put_block_group(bg); | |
3988 | } | |
3989 | ||
f78c436c FM |
3990 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
3991 | { | |
4625956a | 3992 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); |
f78c436c FM |
3993 | } |
3994 | ||
6ab0a202 JM |
3995 | static const char *alloc_name(u64 flags) |
3996 | { | |
3997 | switch (flags) { | |
3998 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3999 | return "mixed"; | |
4000 | case BTRFS_BLOCK_GROUP_METADATA: | |
4001 | return "metadata"; | |
4002 | case BTRFS_BLOCK_GROUP_DATA: | |
4003 | return "data"; | |
4004 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
4005 | return "system"; | |
4006 | default: | |
4007 | WARN_ON(1); | |
4008 | return "invalid-combination"; | |
4009 | }; | |
4010 | } | |
4011 | ||
2be12ef7 NB |
4012 | static int create_space_info(struct btrfs_fs_info *info, u64 flags, |
4013 | struct btrfs_space_info **new) | |
4014 | { | |
4015 | ||
4016 | struct btrfs_space_info *space_info; | |
4017 | int i; | |
4018 | int ret; | |
4019 | ||
4020 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
4021 | if (!space_info) | |
4022 | return -ENOMEM; | |
4023 | ||
4024 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
4025 | GFP_KERNEL); | |
4026 | if (ret) { | |
4027 | kfree(space_info); | |
4028 | return ret; | |
4029 | } | |
4030 | ||
4031 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
4032 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
4033 | init_rwsem(&space_info->groups_sem); | |
4034 | spin_lock_init(&space_info->lock); | |
4035 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
4036 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
4037 | init_waitqueue_head(&space_info->wait); | |
4038 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
4039 | INIT_LIST_HEAD(&space_info->tickets); | |
4040 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
4041 | ||
4042 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
4043 | info->space_info_kobj, "%s", | |
4044 | alloc_name(space_info->flags)); | |
4045 | if (ret) { | |
4046 | percpu_counter_destroy(&space_info->total_bytes_pinned); | |
4047 | kfree(space_info); | |
4048 | return ret; | |
4049 | } | |
4050 | ||
4051 | *new = space_info; | |
4052 | list_add_rcu(&space_info->list, &info->space_info); | |
4053 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4054 | info->data_sinfo = space_info; | |
4055 | ||
4056 | return ret; | |
4057 | } | |
4058 | ||
d2006e6d | 4059 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 4060 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 4061 | u64 bytes_readonly, |
593060d7 CM |
4062 | struct btrfs_space_info **space_info) |
4063 | { | |
4064 | struct btrfs_space_info *found; | |
b742bb82 YZ |
4065 | int factor; |
4066 | ||
4067 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
4068 | BTRFS_BLOCK_GROUP_RAID10)) | |
4069 | factor = 2; | |
4070 | else | |
4071 | factor = 1; | |
593060d7 CM |
4072 | |
4073 | found = __find_space_info(info, flags); | |
d2006e6d NB |
4074 | ASSERT(found); |
4075 | spin_lock(&found->lock); | |
4076 | found->total_bytes += total_bytes; | |
4077 | found->disk_total += total_bytes * factor; | |
4078 | found->bytes_used += bytes_used; | |
4079 | found->disk_used += bytes_used * factor; | |
4080 | found->bytes_readonly += bytes_readonly; | |
4081 | if (total_bytes > 0) | |
4082 | found->full = 0; | |
4083 | space_info_add_new_bytes(info, found, total_bytes - | |
4084 | bytes_used - bytes_readonly); | |
4085 | spin_unlock(&found->lock); | |
4086 | *space_info = found; | |
593060d7 CM |
4087 | } |
4088 | ||
8790d502 CM |
4089 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
4090 | { | |
899c81ea ID |
4091 | u64 extra_flags = chunk_to_extended(flags) & |
4092 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4093 | |
de98ced9 | 4094 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4095 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4096 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4097 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4098 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4099 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4100 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4101 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4102 | } |
593060d7 | 4103 | |
fc67c450 ID |
4104 | /* |
4105 | * returns target flags in extended format or 0 if restripe for this | |
4106 | * chunk_type is not in progress | |
c6664b42 | 4107 | * |
dccdb07b | 4108 | * should be called with balance_lock held |
fc67c450 ID |
4109 | */ |
4110 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4111 | { | |
4112 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4113 | u64 target = 0; | |
4114 | ||
fc67c450 ID |
4115 | if (!bctl) |
4116 | return 0; | |
4117 | ||
4118 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4119 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4120 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4121 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4122 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4123 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4124 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4125 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4126 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4127 | } | |
4128 | ||
4129 | return target; | |
4130 | } | |
4131 | ||
a46d11a8 ID |
4132 | /* |
4133 | * @flags: available profiles in extended format (see ctree.h) | |
4134 | * | |
e4d8ec0f ID |
4135 | * Returns reduced profile in chunk format. If profile changing is in |
4136 | * progress (either running or paused) picks the target profile (if it's | |
4137 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4138 | */ |
2ff7e61e | 4139 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4140 | { |
0b246afa | 4141 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4142 | u64 target; |
9c170b26 ZL |
4143 | u64 raid_type; |
4144 | u64 allowed = 0; | |
a061fc8d | 4145 | |
fc67c450 ID |
4146 | /* |
4147 | * see if restripe for this chunk_type is in progress, if so | |
4148 | * try to reduce to the target profile | |
4149 | */ | |
0b246afa JM |
4150 | spin_lock(&fs_info->balance_lock); |
4151 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4152 | if (target) { |
4153 | /* pick target profile only if it's already available */ | |
4154 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4155 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4156 | return extended_to_chunk(target); |
e4d8ec0f ID |
4157 | } |
4158 | } | |
0b246afa | 4159 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4160 | |
53b381b3 | 4161 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4162 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4163 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
41a6e891 | 4164 | allowed |= btrfs_raid_array[raid_type].bg_flag; |
9c170b26 ZL |
4165 | } |
4166 | allowed &= flags; | |
4167 | ||
4168 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4169 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4170 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4171 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4172 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4173 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4174 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4175 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4176 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4177 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4178 | ||
4179 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4180 | ||
4181 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4182 | } |
4183 | ||
2ff7e61e | 4184 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4185 | { |
de98ced9 | 4186 | unsigned seq; |
f8213bdc | 4187 | u64 flags; |
de98ced9 MX |
4188 | |
4189 | do { | |
f8213bdc | 4190 | flags = orig_flags; |
0b246afa | 4191 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4192 | |
4193 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4194 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4195 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4196 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4197 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4198 | flags |= fs_info->avail_metadata_alloc_bits; |
4199 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4200 | |
2ff7e61e | 4201 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4202 | } |
4203 | ||
1b86826d | 4204 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4205 | { |
0b246afa | 4206 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4207 | u64 flags; |
53b381b3 | 4208 | u64 ret; |
9ed74f2d | 4209 | |
b742bb82 YZ |
4210 | if (data) |
4211 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4212 | else if (root == fs_info->chunk_root) |
b742bb82 | 4213 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4214 | else |
b742bb82 | 4215 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4216 | |
2ff7e61e | 4217 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4218 | return ret; |
6a63209f | 4219 | } |
9ed74f2d | 4220 | |
1b86826d JM |
4221 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4222 | { | |
4223 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4224 | } | |
4225 | ||
4226 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4227 | { | |
4228 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4229 | } | |
4230 | ||
4231 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4232 | { | |
4233 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4234 | } | |
4235 | ||
4136135b LB |
4236 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4237 | bool may_use_included) | |
4238 | { | |
4239 | ASSERT(s_info); | |
4240 | return s_info->bytes_used + s_info->bytes_reserved + | |
4241 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4242 | (may_use_included ? s_info->bytes_may_use : 0); | |
4243 | } | |
4244 | ||
04f4f916 | 4245 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4246 | { |
04f4f916 | 4247 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4248 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4249 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4250 | u64 used; |
94b947b2 | 4251 | int ret = 0; |
c99f1b0c ZL |
4252 | int need_commit = 2; |
4253 | int have_pinned_space; | |
6a63209f | 4254 | |
6a63209f | 4255 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4256 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4257 | |
9dced186 | 4258 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4259 | need_commit = 0; |
9dced186 | 4260 | ASSERT(current->journal_info); |
0af3d00b JB |
4261 | } |
4262 | ||
6a63209f JB |
4263 | again: |
4264 | /* make sure we have enough space to handle the data first */ | |
4265 | spin_lock(&data_sinfo->lock); | |
4136135b | 4266 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4267 | |
4268 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4269 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4270 | |
6a63209f JB |
4271 | /* |
4272 | * if we don't have enough free bytes in this space then we need | |
4273 | * to alloc a new chunk. | |
4274 | */ | |
b9fd47cd | 4275 | if (!data_sinfo->full) { |
6a63209f | 4276 | u64 alloc_target; |
9ed74f2d | 4277 | |
0e4f8f88 | 4278 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4279 | spin_unlock(&data_sinfo->lock); |
1174cade | 4280 | |
1b86826d | 4281 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4282 | /* |
4283 | * It is ugly that we don't call nolock join | |
4284 | * transaction for the free space inode case here. | |
4285 | * But it is safe because we only do the data space | |
4286 | * reservation for the free space cache in the | |
4287 | * transaction context, the common join transaction | |
4288 | * just increase the counter of the current transaction | |
4289 | * handler, doesn't try to acquire the trans_lock of | |
4290 | * the fs. | |
4291 | */ | |
7a7eaa40 | 4292 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4293 | if (IS_ERR(trans)) |
4294 | return PTR_ERR(trans); | |
9ed74f2d | 4295 | |
2ff7e61e | 4296 | ret = do_chunk_alloc(trans, fs_info, alloc_target, |
0e4f8f88 | 4297 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4298 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4299 | if (ret < 0) { |
4300 | if (ret != -ENOSPC) | |
4301 | return ret; | |
c99f1b0c ZL |
4302 | else { |
4303 | have_pinned_space = 1; | |
d52a5b5f | 4304 | goto commit_trans; |
c99f1b0c | 4305 | } |
d52a5b5f | 4306 | } |
9ed74f2d | 4307 | |
6a63209f JB |
4308 | goto again; |
4309 | } | |
f2bb8f5c JB |
4310 | |
4311 | /* | |
b150a4f1 | 4312 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4313 | * allocation, and no removed chunk in current transaction, |
4314 | * don't bother committing the transaction. | |
f2bb8f5c | 4315 | */ |
c99f1b0c ZL |
4316 | have_pinned_space = percpu_counter_compare( |
4317 | &data_sinfo->total_bytes_pinned, | |
4318 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4319 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4320 | |
4e06bdd6 | 4321 | /* commit the current transaction and try again */ |
d52a5b5f | 4322 | commit_trans: |
92e2f7e3 | 4323 | if (need_commit) { |
c99f1b0c | 4324 | need_commit--; |
b150a4f1 | 4325 | |
e1746e83 | 4326 | if (need_commit > 0) { |
82b3e53b | 4327 | btrfs_start_delalloc_roots(fs_info, -1); |
6374e57a | 4328 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4329 | (u64)-1); |
e1746e83 | 4330 | } |
9a4e7276 | 4331 | |
7a7eaa40 | 4332 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4333 | if (IS_ERR(trans)) |
4334 | return PTR_ERR(trans); | |
c99f1b0c | 4335 | if (have_pinned_space >= 0 || |
3204d33c JB |
4336 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4337 | &trans->transaction->flags) || | |
c99f1b0c | 4338 | need_commit > 0) { |
3a45bb20 | 4339 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4340 | if (ret) |
4341 | return ret; | |
d7c15171 | 4342 | /* |
c2d6cb16 FM |
4343 | * The cleaner kthread might still be doing iput |
4344 | * operations. Wait for it to finish so that | |
4345 | * more space is released. | |
d7c15171 | 4346 | */ |
0b246afa JM |
4347 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4348 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4349 | goto again; |
4350 | } else { | |
3a45bb20 | 4351 | btrfs_end_transaction(trans); |
94b947b2 | 4352 | } |
4e06bdd6 | 4353 | } |
9ed74f2d | 4354 | |
0b246afa | 4355 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4356 | "space_info:enospc", |
4357 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4358 | return -ENOSPC; |
4359 | } | |
4360 | data_sinfo->bytes_may_use += bytes; | |
0b246afa | 4361 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4362 | data_sinfo->flags, bytes, 1); |
6a63209f | 4363 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4364 | |
237c0e9f | 4365 | return ret; |
9ed74f2d | 4366 | } |
6a63209f | 4367 | |
364ecf36 QW |
4368 | int btrfs_check_data_free_space(struct inode *inode, |
4369 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4370 | { |
0b246afa | 4371 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4372 | int ret; |
4373 | ||
4374 | /* align the range */ | |
0b246afa JM |
4375 | len = round_up(start + len, fs_info->sectorsize) - |
4376 | round_down(start, fs_info->sectorsize); | |
4377 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4378 | |
04f4f916 | 4379 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4380 | if (ret < 0) |
4381 | return ret; | |
4382 | ||
1e5ec2e7 | 4383 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4384 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4385 | if (ret < 0) |
1e5ec2e7 | 4386 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4387 | else |
4388 | ret = 0; | |
4ceff079 QW |
4389 | return ret; |
4390 | } | |
4391 | ||
4ceff079 QW |
4392 | /* |
4393 | * Called if we need to clear a data reservation for this inode | |
4394 | * Normally in a error case. | |
4395 | * | |
51773bec QW |
4396 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4397 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4398 | * Like clear_bit_hook(). | |
4ceff079 | 4399 | */ |
51773bec QW |
4400 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4401 | u64 len) | |
4ceff079 | 4402 | { |
0b246afa | 4403 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4404 | struct btrfs_space_info *data_sinfo; |
4405 | ||
4406 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4407 | len = round_up(start + len, fs_info->sectorsize) - |
4408 | round_down(start, fs_info->sectorsize); | |
4409 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4410 | |
0b246afa | 4411 | data_sinfo = fs_info->data_sinfo; |
4ceff079 QW |
4412 | spin_lock(&data_sinfo->lock); |
4413 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4414 | data_sinfo->bytes_may_use = 0; | |
4415 | else | |
4416 | data_sinfo->bytes_may_use -= len; | |
0b246afa | 4417 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4418 | data_sinfo->flags, len, 0); |
4419 | spin_unlock(&data_sinfo->lock); | |
4420 | } | |
4421 | ||
51773bec QW |
4422 | /* |
4423 | * Called if we need to clear a data reservation for this inode | |
4424 | * Normally in a error case. | |
4425 | * | |
01327610 | 4426 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4427 | * space framework. |
4428 | */ | |
bc42bda2 QW |
4429 | void btrfs_free_reserved_data_space(struct inode *inode, |
4430 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4431 | { |
0c476a5d JM |
4432 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4433 | ||
4434 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4435 | len = round_up(start + len, root->fs_info->sectorsize) - |
4436 | round_down(start, root->fs_info->sectorsize); | |
4437 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4438 | |
51773bec | 4439 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4440 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4441 | } |
4442 | ||
97e728d4 | 4443 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4444 | { |
97e728d4 JB |
4445 | struct list_head *head = &info->space_info; |
4446 | struct btrfs_space_info *found; | |
e3ccfa98 | 4447 | |
97e728d4 JB |
4448 | rcu_read_lock(); |
4449 | list_for_each_entry_rcu(found, head, list) { | |
4450 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4451 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4452 | } |
97e728d4 | 4453 | rcu_read_unlock(); |
e3ccfa98 JB |
4454 | } |
4455 | ||
3c76cd84 MX |
4456 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4457 | { | |
4458 | return (global->size << 1); | |
4459 | } | |
4460 | ||
2ff7e61e | 4461 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4462 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4463 | { |
0b246afa | 4464 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
8d8aafee | 4465 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4466 | u64 thresh; |
e3ccfa98 | 4467 | |
0e4f8f88 CM |
4468 | if (force == CHUNK_ALLOC_FORCE) |
4469 | return 1; | |
4470 | ||
fb25e914 JB |
4471 | /* |
4472 | * We need to take into account the global rsv because for all intents | |
4473 | * and purposes it's used space. Don't worry about locking the | |
4474 | * global_rsv, it doesn't change except when the transaction commits. | |
4475 | */ | |
54338b5c | 4476 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
8d8aafee | 4477 | bytes_used += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4478 | |
0e4f8f88 CM |
4479 | /* |
4480 | * in limited mode, we want to have some free space up to | |
4481 | * about 1% of the FS size. | |
4482 | */ | |
4483 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4484 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4485 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4486 | |
8d8aafee | 4487 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4488 | return 1; |
4489 | } | |
0e4f8f88 | 4490 | |
8d8aafee | 4491 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4492 | return 0; |
424499db | 4493 | return 1; |
32c00aff JB |
4494 | } |
4495 | ||
2ff7e61e | 4496 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4497 | { |
4498 | u64 num_dev; | |
4499 | ||
53b381b3 DW |
4500 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4501 | BTRFS_BLOCK_GROUP_RAID0 | | |
4502 | BTRFS_BLOCK_GROUP_RAID5 | | |
4503 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4504 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4505 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4506 | num_dev = 2; | |
4507 | else | |
4508 | num_dev = 1; /* DUP or single */ | |
4509 | ||
39c2d7fa | 4510 | return num_dev; |
15d1ff81 LB |
4511 | } |
4512 | ||
39c2d7fa FM |
4513 | /* |
4514 | * If @is_allocation is true, reserve space in the system space info necessary | |
4515 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4516 | * removing a chunk. | |
4517 | */ | |
4518 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 4519 | struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4520 | { |
4521 | struct btrfs_space_info *info; | |
4522 | u64 left; | |
4523 | u64 thresh; | |
4fbcdf66 | 4524 | int ret = 0; |
39c2d7fa | 4525 | u64 num_devs; |
4fbcdf66 FM |
4526 | |
4527 | /* | |
4528 | * Needed because we can end up allocating a system chunk and for an | |
4529 | * atomic and race free space reservation in the chunk block reserve. | |
4530 | */ | |
a32bf9a3 | 4531 | lockdep_assert_held(&fs_info->chunk_mutex); |
15d1ff81 | 4532 | |
0b246afa | 4533 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4534 | spin_lock(&info->lock); |
4136135b | 4535 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4536 | spin_unlock(&info->lock); |
4537 | ||
2ff7e61e | 4538 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4539 | |
4540 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4541 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4542 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4543 | |
0b246afa JM |
4544 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4545 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4546 | left, thresh, type); | |
4547 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4548 | } |
4549 | ||
4550 | if (left < thresh) { | |
1b86826d | 4551 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4552 | |
4fbcdf66 FM |
4553 | /* |
4554 | * Ignore failure to create system chunk. We might end up not | |
4555 | * needing it, as we might not need to COW all nodes/leafs from | |
4556 | * the paths we visit in the chunk tree (they were already COWed | |
4557 | * or created in the current transaction for example). | |
4558 | */ | |
2ff7e61e | 4559 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
4fbcdf66 FM |
4560 | } |
4561 | ||
4562 | if (!ret) { | |
0b246afa JM |
4563 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4564 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4565 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4566 | if (!ret) | |
4567 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4568 | } |
4569 | } | |
4570 | ||
28b737f6 LB |
4571 | /* |
4572 | * If force is CHUNK_ALLOC_FORCE: | |
4573 | * - return 1 if it successfully allocates a chunk, | |
4574 | * - return errors including -ENOSPC otherwise. | |
4575 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4576 | * - return 0 if it doesn't need to allocate a new chunk, | |
4577 | * - return 1 if it successfully allocates a chunk, | |
4578 | * - return errors including -ENOSPC otherwise. | |
4579 | */ | |
6324fbf3 | 4580 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 4581 | struct btrfs_fs_info *fs_info, u64 flags, int force) |
9ed74f2d | 4582 | { |
6324fbf3 | 4583 | struct btrfs_space_info *space_info; |
6d74119f | 4584 | int wait_for_alloc = 0; |
9ed74f2d | 4585 | int ret = 0; |
9ed74f2d | 4586 | |
c6b305a8 JB |
4587 | /* Don't re-enter if we're already allocating a chunk */ |
4588 | if (trans->allocating_chunk) | |
4589 | return -ENOSPC; | |
4590 | ||
0b246afa | 4591 | space_info = __find_space_info(fs_info, flags); |
dc2d3005 | 4592 | ASSERT(space_info); |
9ed74f2d | 4593 | |
6d74119f | 4594 | again: |
25179201 | 4595 | spin_lock(&space_info->lock); |
9e622d6b | 4596 | if (force < space_info->force_alloc) |
0e4f8f88 | 4597 | force = space_info->force_alloc; |
25179201 | 4598 | if (space_info->full) { |
2ff7e61e | 4599 | if (should_alloc_chunk(fs_info, space_info, force)) |
09fb99a6 FDBM |
4600 | ret = -ENOSPC; |
4601 | else | |
4602 | ret = 0; | |
25179201 | 4603 | spin_unlock(&space_info->lock); |
09fb99a6 | 4604 | return ret; |
9ed74f2d JB |
4605 | } |
4606 | ||
2ff7e61e | 4607 | if (!should_alloc_chunk(fs_info, space_info, force)) { |
25179201 | 4608 | spin_unlock(&space_info->lock); |
6d74119f JB |
4609 | return 0; |
4610 | } else if (space_info->chunk_alloc) { | |
4611 | wait_for_alloc = 1; | |
4612 | } else { | |
4613 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4614 | } |
0e4f8f88 | 4615 | |
25179201 | 4616 | spin_unlock(&space_info->lock); |
9ed74f2d | 4617 | |
6d74119f JB |
4618 | mutex_lock(&fs_info->chunk_mutex); |
4619 | ||
4620 | /* | |
4621 | * The chunk_mutex is held throughout the entirety of a chunk | |
4622 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4623 | * other guy is done and we need to recheck and see if we should | |
4624 | * allocate. | |
4625 | */ | |
4626 | if (wait_for_alloc) { | |
4627 | mutex_unlock(&fs_info->chunk_mutex); | |
4628 | wait_for_alloc = 0; | |
1e1c50a9 | 4629 | cond_resched(); |
6d74119f JB |
4630 | goto again; |
4631 | } | |
4632 | ||
c6b305a8 JB |
4633 | trans->allocating_chunk = true; |
4634 | ||
67377734 JB |
4635 | /* |
4636 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4637 | * allocating mixed chunks instead of individual chunks. | |
4638 | */ | |
4639 | if (btrfs_mixed_space_info(space_info)) | |
4640 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4641 | ||
97e728d4 JB |
4642 | /* |
4643 | * if we're doing a data chunk, go ahead and make sure that | |
4644 | * we keep a reasonable number of metadata chunks allocated in the | |
4645 | * FS as well. | |
4646 | */ | |
9ed74f2d | 4647 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4648 | fs_info->data_chunk_allocations++; |
4649 | if (!(fs_info->data_chunk_allocations % | |
4650 | fs_info->metadata_ratio)) | |
4651 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4652 | } |
4653 | ||
15d1ff81 LB |
4654 | /* |
4655 | * Check if we have enough space in SYSTEM chunk because we may need | |
4656 | * to update devices. | |
4657 | */ | |
2ff7e61e | 4658 | check_system_chunk(trans, fs_info, flags); |
15d1ff81 | 4659 | |
2ff7e61e | 4660 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
c6b305a8 | 4661 | trans->allocating_chunk = false; |
92b8e897 | 4662 | |
9ed74f2d | 4663 | spin_lock(&space_info->lock); |
57f1642e NB |
4664 | if (ret < 0) { |
4665 | if (ret == -ENOSPC) | |
4666 | space_info->full = 1; | |
4667 | else | |
4668 | goto out; | |
4669 | } else { | |
424499db | 4670 | ret = 1; |
57f1642e | 4671 | } |
6d74119f | 4672 | |
0e4f8f88 | 4673 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4674 | out: |
6d74119f | 4675 | space_info->chunk_alloc = 0; |
9ed74f2d | 4676 | spin_unlock(&space_info->lock); |
a25c75d5 | 4677 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4678 | /* |
4679 | * When we allocate a new chunk we reserve space in the chunk block | |
4680 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4681 | * add new nodes/leafs to it if we end up needing to do it when | |
4682 | * inserting the chunk item and updating device items as part of the | |
4683 | * second phase of chunk allocation, performed by | |
4684 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4685 | * large number of new block groups to create in our transaction | |
4686 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4687 | * in extreme cases - like having a single transaction create many new | |
4688 | * block groups when starting to write out the free space caches of all | |
4689 | * the block groups that were made dirty during the lifetime of the | |
4690 | * transaction. | |
4691 | */ | |
d9a0540a | 4692 | if (trans->can_flush_pending_bgs && |
ee22184b | 4693 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
6c686b35 | 4694 | btrfs_create_pending_block_groups(trans); |
00d80e34 FM |
4695 | btrfs_trans_release_chunk_metadata(trans); |
4696 | } | |
0f9dd46c | 4697 | return ret; |
6324fbf3 | 4698 | } |
9ed74f2d | 4699 | |
c1c4919b | 4700 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4701 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4702 | enum btrfs_reserve_flush_enum flush, |
4703 | bool system_chunk) | |
a80c8dcf | 4704 | { |
0b246afa | 4705 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4706 | u64 profile; |
3c76cd84 | 4707 | u64 space_size; |
a80c8dcf JB |
4708 | u64 avail; |
4709 | u64 used; | |
4710 | ||
957780eb JB |
4711 | /* Don't overcommit when in mixed mode. */ |
4712 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4713 | return 0; | |
4714 | ||
c1c4919b JM |
4715 | if (system_chunk) |
4716 | profile = btrfs_system_alloc_profile(fs_info); | |
4717 | else | |
4718 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4719 | ||
4136135b | 4720 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4721 | |
96f1bb57 JB |
4722 | /* |
4723 | * We only want to allow over committing if we have lots of actual space | |
4724 | * free, but if we don't have enough space to handle the global reserve | |
4725 | * space then we could end up having a real enospc problem when trying | |
4726 | * to allocate a chunk or some other such important allocation. | |
4727 | */ | |
3c76cd84 MX |
4728 | spin_lock(&global_rsv->lock); |
4729 | space_size = calc_global_rsv_need_space(global_rsv); | |
4730 | spin_unlock(&global_rsv->lock); | |
4731 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4732 | return 0; |
4733 | ||
4734 | used += space_info->bytes_may_use; | |
a80c8dcf | 4735 | |
a5ed45f8 | 4736 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4737 | |
4738 | /* | |
4739 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4740 | * space is actually useable. For raid56, the space info used |
4741 | * doesn't include the parity drive, so we don't have to | |
4742 | * change the math | |
a80c8dcf JB |
4743 | */ |
4744 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4745 | BTRFS_BLOCK_GROUP_RAID1 | | |
4746 | BTRFS_BLOCK_GROUP_RAID10)) | |
4747 | avail >>= 1; | |
4748 | ||
4749 | /* | |
561c294d MX |
4750 | * If we aren't flushing all things, let us overcommit up to |
4751 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4752 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4753 | */ |
08e007d2 | 4754 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4755 | avail >>= 3; |
a80c8dcf | 4756 | else |
14575aef | 4757 | avail >>= 1; |
a80c8dcf | 4758 | |
14575aef | 4759 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4760 | return 1; |
4761 | return 0; | |
4762 | } | |
4763 | ||
2ff7e61e | 4764 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4765 | unsigned long nr_pages, int nr_items) |
da633a42 | 4766 | { |
0b246afa | 4767 | struct super_block *sb = fs_info->sb; |
da633a42 | 4768 | |
925a6efb JB |
4769 | if (down_read_trylock(&sb->s_umount)) { |
4770 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4771 | up_read(&sb->s_umount); | |
4772 | } else { | |
da633a42 MX |
4773 | /* |
4774 | * We needn't worry the filesystem going from r/w to r/o though | |
4775 | * we don't acquire ->s_umount mutex, because the filesystem | |
4776 | * should guarantee the delalloc inodes list be empty after | |
4777 | * the filesystem is readonly(all dirty pages are written to | |
4778 | * the disk). | |
4779 | */ | |
82b3e53b | 4780 | btrfs_start_delalloc_roots(fs_info, nr_items); |
98ad69cf | 4781 | if (!current->journal_info) |
0b246afa | 4782 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4783 | } |
4784 | } | |
4785 | ||
6374e57a | 4786 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4787 | u64 to_reclaim) |
18cd8ea6 MX |
4788 | { |
4789 | u64 bytes; | |
6374e57a | 4790 | u64 nr; |
18cd8ea6 | 4791 | |
2ff7e61e | 4792 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4793 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4794 | if (!nr) |
4795 | nr = 1; | |
4796 | return nr; | |
4797 | } | |
4798 | ||
ee22184b | 4799 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4800 | |
9ed74f2d | 4801 | /* |
5da9d01b | 4802 | * shrink metadata reservation for delalloc |
9ed74f2d | 4803 | */ |
c1c4919b JM |
4804 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4805 | u64 orig, bool wait_ordered) | |
5da9d01b | 4806 | { |
0019f10d | 4807 | struct btrfs_space_info *space_info; |
663350ac | 4808 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4809 | u64 delalloc_bytes; |
5da9d01b | 4810 | u64 max_reclaim; |
6374e57a | 4811 | u64 items; |
b1953bce | 4812 | long time_left; |
d3ee29e3 MX |
4813 | unsigned long nr_pages; |
4814 | int loops; | |
5da9d01b | 4815 | |
c61a16a7 | 4816 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4817 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4818 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4819 | |
663350ac | 4820 | trans = (struct btrfs_trans_handle *)current->journal_info; |
69fe2d75 | 4821 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4822 | |
963d678b | 4823 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4824 | &fs_info->delalloc_bytes); |
f4c738c2 | 4825 | if (delalloc_bytes == 0) { |
fdb5effd | 4826 | if (trans) |
f4c738c2 | 4827 | return; |
38c135af | 4828 | if (wait_ordered) |
0b246afa | 4829 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4830 | return; |
fdb5effd JB |
4831 | } |
4832 | ||
d3ee29e3 | 4833 | loops = 0; |
f4c738c2 JB |
4834 | while (delalloc_bytes && loops < 3) { |
4835 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4836 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4837 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4838 | /* |
4839 | * We need to wait for the async pages to actually start before | |
4840 | * we do anything. | |
4841 | */ | |
0b246afa | 4842 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4843 | if (!max_reclaim) |
4844 | goto skip_async; | |
4845 | ||
4846 | if (max_reclaim <= nr_pages) | |
4847 | max_reclaim = 0; | |
4848 | else | |
4849 | max_reclaim -= nr_pages; | |
dea31f52 | 4850 | |
0b246afa JM |
4851 | wait_event(fs_info->async_submit_wait, |
4852 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4853 | (int)max_reclaim); |
4854 | skip_async: | |
0019f10d | 4855 | spin_lock(&space_info->lock); |
957780eb JB |
4856 | if (list_empty(&space_info->tickets) && |
4857 | list_empty(&space_info->priority_tickets)) { | |
4858 | spin_unlock(&space_info->lock); | |
4859 | break; | |
4860 | } | |
0019f10d | 4861 | spin_unlock(&space_info->lock); |
5da9d01b | 4862 | |
36e39c40 | 4863 | loops++; |
f104d044 | 4864 | if (wait_ordered && !trans) { |
0b246afa | 4865 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4866 | } else { |
f4c738c2 | 4867 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4868 | if (time_left) |
4869 | break; | |
4870 | } | |
963d678b | 4871 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4872 | &fs_info->delalloc_bytes); |
5da9d01b | 4873 | } |
5da9d01b YZ |
4874 | } |
4875 | ||
996478ca JB |
4876 | struct reserve_ticket { |
4877 | u64 bytes; | |
4878 | int error; | |
4879 | struct list_head list; | |
4880 | wait_queue_head_t wait; | |
4881 | }; | |
4882 | ||
663350ac JB |
4883 | /** |
4884 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4885 | * @root - the root we're allocating for | |
4886 | * @bytes - the number of bytes we want to reserve | |
4887 | * @force - force the commit | |
8bb8ab2e | 4888 | * |
663350ac JB |
4889 | * This will check to make sure that committing the transaction will actually |
4890 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4891 | * will return -ENOSPC. | |
8bb8ab2e | 4892 | */ |
0c9ab349 | 4893 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4894 | struct btrfs_space_info *space_info) |
663350ac | 4895 | { |
996478ca | 4896 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4897 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac | 4898 | struct btrfs_trans_handle *trans; |
996478ca | 4899 | u64 bytes; |
663350ac JB |
4900 | |
4901 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4902 | if (trans) | |
4903 | return -EAGAIN; | |
4904 | ||
996478ca JB |
4905 | spin_lock(&space_info->lock); |
4906 | if (!list_empty(&space_info->priority_tickets)) | |
4907 | ticket = list_first_entry(&space_info->priority_tickets, | |
4908 | struct reserve_ticket, list); | |
4909 | else if (!list_empty(&space_info->tickets)) | |
4910 | ticket = list_first_entry(&space_info->tickets, | |
4911 | struct reserve_ticket, list); | |
4912 | bytes = (ticket) ? ticket->bytes : 0; | |
4913 | spin_unlock(&space_info->lock); | |
4914 | ||
4915 | if (!bytes) | |
4916 | return 0; | |
663350ac JB |
4917 | |
4918 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4919 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4920 | bytes) >= 0) |
663350ac | 4921 | goto commit; |
663350ac JB |
4922 | |
4923 | /* | |
4924 | * See if there is some space in the delayed insertion reservation for | |
4925 | * this reservation. | |
4926 | */ | |
4927 | if (space_info != delayed_rsv->space_info) | |
4928 | return -ENOSPC; | |
4929 | ||
4930 | spin_lock(&delayed_rsv->lock); | |
996478ca JB |
4931 | if (delayed_rsv->size > bytes) |
4932 | bytes = 0; | |
4933 | else | |
4934 | bytes -= delayed_rsv->size; | |
057aac3e NB |
4935 | spin_unlock(&delayed_rsv->lock); |
4936 | ||
b150a4f1 | 4937 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
996478ca | 4938 | bytes) < 0) { |
663350ac JB |
4939 | return -ENOSPC; |
4940 | } | |
663350ac JB |
4941 | |
4942 | commit: | |
a9b3311e | 4943 | trans = btrfs_join_transaction(fs_info->extent_root); |
663350ac JB |
4944 | if (IS_ERR(trans)) |
4945 | return -ENOSPC; | |
4946 | ||
3a45bb20 | 4947 | return btrfs_commit_transaction(trans); |
663350ac JB |
4948 | } |
4949 | ||
e38ae7a0 NB |
4950 | /* |
4951 | * Try to flush some data based on policy set by @state. This is only advisory | |
4952 | * and may fail for various reasons. The caller is supposed to examine the | |
4953 | * state of @space_info to detect the outcome. | |
4954 | */ | |
4955 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4956 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4957 | int state) |
96c3f433 | 4958 | { |
a9b3311e | 4959 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4960 | struct btrfs_trans_handle *trans; |
4961 | int nr; | |
f4c738c2 | 4962 | int ret = 0; |
96c3f433 JB |
4963 | |
4964 | switch (state) { | |
96c3f433 JB |
4965 | case FLUSH_DELAYED_ITEMS_NR: |
4966 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4967 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4968 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4969 | else |
96c3f433 | 4970 | nr = -1; |
18cd8ea6 | 4971 | |
96c3f433 JB |
4972 | trans = btrfs_join_transaction(root); |
4973 | if (IS_ERR(trans)) { | |
4974 | ret = PTR_ERR(trans); | |
4975 | break; | |
4976 | } | |
e5c304e6 | 4977 | ret = btrfs_run_delayed_items_nr(trans, nr); |
3a45bb20 | 4978 | btrfs_end_transaction(trans); |
96c3f433 | 4979 | break; |
67b0fd63 JB |
4980 | case FLUSH_DELALLOC: |
4981 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 4982 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
4983 | state == FLUSH_DELALLOC_WAIT); |
4984 | break; | |
ea658bad JB |
4985 | case ALLOC_CHUNK: |
4986 | trans = btrfs_join_transaction(root); | |
4987 | if (IS_ERR(trans)) { | |
4988 | ret = PTR_ERR(trans); | |
4989 | break; | |
4990 | } | |
2ff7e61e | 4991 | ret = do_chunk_alloc(trans, fs_info, |
1b86826d | 4992 | btrfs_metadata_alloc_profile(fs_info), |
ea658bad | 4993 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4994 | btrfs_end_transaction(trans); |
eecba891 | 4995 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
4996 | ret = 0; |
4997 | break; | |
96c3f433 | 4998 | case COMMIT_TRANS: |
996478ca | 4999 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
5000 | break; |
5001 | default: | |
5002 | ret = -ENOSPC; | |
5003 | break; | |
5004 | } | |
5005 | ||
7bdd6277 NB |
5006 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
5007 | ret); | |
e38ae7a0 | 5008 | return; |
96c3f433 | 5009 | } |
21c7e756 MX |
5010 | |
5011 | static inline u64 | |
c1c4919b JM |
5012 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
5013 | struct btrfs_space_info *space_info, | |
5014 | bool system_chunk) | |
21c7e756 | 5015 | { |
957780eb | 5016 | struct reserve_ticket *ticket; |
21c7e756 MX |
5017 | u64 used; |
5018 | u64 expected; | |
957780eb | 5019 | u64 to_reclaim = 0; |
21c7e756 | 5020 | |
957780eb JB |
5021 | list_for_each_entry(ticket, &space_info->tickets, list) |
5022 | to_reclaim += ticket->bytes; | |
5023 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
5024 | to_reclaim += ticket->bytes; | |
5025 | if (to_reclaim) | |
5026 | return to_reclaim; | |
21c7e756 | 5027 | |
e0af2484 | 5028 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
5029 | if (can_overcommit(fs_info, space_info, to_reclaim, |
5030 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
5031 | return 0; |
5032 | ||
0eee8a49 NB |
5033 | used = btrfs_space_info_used(space_info, true); |
5034 | ||
c1c4919b JM |
5035 | if (can_overcommit(fs_info, space_info, SZ_1M, |
5036 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
5037 | expected = div_factor_fine(space_info->total_bytes, 95); |
5038 | else | |
5039 | expected = div_factor_fine(space_info->total_bytes, 90); | |
5040 | ||
5041 | if (used > expected) | |
5042 | to_reclaim = used - expected; | |
5043 | else | |
5044 | to_reclaim = 0; | |
5045 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
5046 | space_info->bytes_reserved); | |
21c7e756 MX |
5047 | return to_reclaim; |
5048 | } | |
5049 | ||
c1c4919b JM |
5050 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
5051 | struct btrfs_space_info *space_info, | |
5052 | u64 used, bool system_chunk) | |
21c7e756 | 5053 | { |
365c5313 JB |
5054 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
5055 | ||
5056 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 5057 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
5058 | return 0; |
5059 | ||
c1c4919b JM |
5060 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5061 | system_chunk)) | |
d38b349c JB |
5062 | return 0; |
5063 | ||
0b246afa JM |
5064 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
5065 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
5066 | } |
5067 | ||
957780eb | 5068 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 5069 | { |
957780eb | 5070 | struct reserve_ticket *ticket; |
25ce459c | 5071 | |
957780eb JB |
5072 | while (!list_empty(head)) { |
5073 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
5074 | list_del_init(&ticket->list); | |
5075 | ticket->error = -ENOSPC; | |
5076 | wake_up(&ticket->wait); | |
21c7e756 | 5077 | } |
21c7e756 MX |
5078 | } |
5079 | ||
957780eb JB |
5080 | /* |
5081 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
5082 | * will loop and continuously try to flush as long as we are making progress. | |
5083 | * We count progress as clearing off tickets each time we have to loop. | |
5084 | */ | |
21c7e756 MX |
5085 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
5086 | { | |
5087 | struct btrfs_fs_info *fs_info; | |
5088 | struct btrfs_space_info *space_info; | |
5089 | u64 to_reclaim; | |
5090 | int flush_state; | |
957780eb | 5091 | int commit_cycles = 0; |
ce129655 | 5092 | u64 last_tickets_id; |
21c7e756 MX |
5093 | |
5094 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
5095 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5096 | ||
957780eb | 5097 | spin_lock(&space_info->lock); |
c1c4919b JM |
5098 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5099 | false); | |
957780eb JB |
5100 | if (!to_reclaim) { |
5101 | space_info->flush = 0; | |
5102 | spin_unlock(&space_info->lock); | |
21c7e756 | 5103 | return; |
957780eb | 5104 | } |
ce129655 | 5105 | last_tickets_id = space_info->tickets_id; |
957780eb | 5106 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5107 | |
5108 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 5109 | do { |
e38ae7a0 | 5110 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
5111 | spin_lock(&space_info->lock); |
5112 | if (list_empty(&space_info->tickets)) { | |
5113 | space_info->flush = 0; | |
5114 | spin_unlock(&space_info->lock); | |
5115 | return; | |
5116 | } | |
c1c4919b JM |
5117 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5118 | space_info, | |
5119 | false); | |
ce129655 | 5120 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5121 | flush_state++; |
5122 | } else { | |
ce129655 | 5123 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5124 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5125 | if (commit_cycles) | |
5126 | commit_cycles--; | |
5127 | } | |
5128 | ||
5129 | if (flush_state > COMMIT_TRANS) { | |
5130 | commit_cycles++; | |
5131 | if (commit_cycles > 2) { | |
5132 | wake_all_tickets(&space_info->tickets); | |
5133 | space_info->flush = 0; | |
5134 | } else { | |
5135 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5136 | } | |
5137 | } | |
5138 | spin_unlock(&space_info->lock); | |
5139 | } while (flush_state <= COMMIT_TRANS); | |
5140 | } | |
5141 | ||
5142 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5143 | { | |
5144 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5145 | } | |
5146 | ||
5147 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5148 | struct btrfs_space_info *space_info, | |
5149 | struct reserve_ticket *ticket) | |
5150 | { | |
5151 | u64 to_reclaim; | |
5152 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5153 | ||
5154 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5155 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5156 | false); | |
957780eb JB |
5157 | if (!to_reclaim) { |
5158 | spin_unlock(&space_info->lock); | |
5159 | return; | |
5160 | } | |
5161 | spin_unlock(&space_info->lock); | |
5162 | ||
21c7e756 | 5163 | do { |
7bdd6277 | 5164 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
21c7e756 | 5165 | flush_state++; |
957780eb JB |
5166 | spin_lock(&space_info->lock); |
5167 | if (ticket->bytes == 0) { | |
5168 | spin_unlock(&space_info->lock); | |
21c7e756 | 5169 | return; |
957780eb JB |
5170 | } |
5171 | spin_unlock(&space_info->lock); | |
5172 | ||
5173 | /* | |
5174 | * Priority flushers can't wait on delalloc without | |
5175 | * deadlocking. | |
5176 | */ | |
5177 | if (flush_state == FLUSH_DELALLOC || | |
5178 | flush_state == FLUSH_DELALLOC_WAIT) | |
5179 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5180 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5181 | } |
5182 | ||
957780eb JB |
5183 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5184 | struct btrfs_space_info *space_info, | |
5185 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5186 | ||
21c7e756 | 5187 | { |
957780eb JB |
5188 | DEFINE_WAIT(wait); |
5189 | int ret = 0; | |
5190 | ||
5191 | spin_lock(&space_info->lock); | |
5192 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5193 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5194 | if (ret) { | |
5195 | ret = -EINTR; | |
5196 | break; | |
5197 | } | |
5198 | spin_unlock(&space_info->lock); | |
5199 | ||
5200 | schedule(); | |
5201 | ||
5202 | finish_wait(&ticket->wait, &wait); | |
5203 | spin_lock(&space_info->lock); | |
5204 | } | |
5205 | if (!ret) | |
5206 | ret = ticket->error; | |
5207 | if (!list_empty(&ticket->list)) | |
5208 | list_del_init(&ticket->list); | |
5209 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5210 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5211 | space_info->bytes_may_use -= num_bytes; | |
5212 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5213 | space_info->flags, num_bytes, 0); | |
5214 | } | |
5215 | spin_unlock(&space_info->lock); | |
5216 | ||
5217 | return ret; | |
21c7e756 MX |
5218 | } |
5219 | ||
4a92b1b8 JB |
5220 | /** |
5221 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5222 | * @root - the root we're allocating for | |
957780eb | 5223 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5224 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5225 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5226 | * |
01327610 | 5227 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5228 | * with the block_rsv. If there is not enough space it will make an attempt to |
5229 | * flush out space to make room. It will do this by flushing delalloc if | |
5230 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5231 | * regain reservations will be made and this will fail if there is not enough | |
5232 | * space already. | |
8bb8ab2e | 5233 | */ |
c1c4919b | 5234 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5235 | struct btrfs_space_info *space_info, |
5236 | u64 orig_bytes, | |
c1c4919b JM |
5237 | enum btrfs_reserve_flush_enum flush, |
5238 | bool system_chunk) | |
9ed74f2d | 5239 | { |
957780eb | 5240 | struct reserve_ticket ticket; |
2bf64758 | 5241 | u64 used; |
8bb8ab2e | 5242 | int ret = 0; |
9ed74f2d | 5243 | |
957780eb | 5244 | ASSERT(orig_bytes); |
8ca17f0f | 5245 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5246 | |
8bb8ab2e | 5247 | spin_lock(&space_info->lock); |
fdb5effd | 5248 | ret = -ENOSPC; |
4136135b | 5249 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5250 | |
8bb8ab2e | 5251 | /* |
957780eb JB |
5252 | * If we have enough space then hooray, make our reservation and carry |
5253 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5254 | * If not things get more complicated. | |
8bb8ab2e | 5255 | */ |
957780eb JB |
5256 | if (used + orig_bytes <= space_info->total_bytes) { |
5257 | space_info->bytes_may_use += orig_bytes; | |
0b246afa JM |
5258 | trace_btrfs_space_reservation(fs_info, "space_info", |
5259 | space_info->flags, orig_bytes, 1); | |
957780eb | 5260 | ret = 0; |
c1c4919b JM |
5261 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5262 | system_chunk)) { | |
44734ed1 | 5263 | space_info->bytes_may_use += orig_bytes; |
0b246afa JM |
5264 | trace_btrfs_space_reservation(fs_info, "space_info", |
5265 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5266 | ret = 0; |
2bf64758 JB |
5267 | } |
5268 | ||
8bb8ab2e | 5269 | /* |
957780eb JB |
5270 | * If we couldn't make a reservation then setup our reservation ticket |
5271 | * and kick the async worker if it's not already running. | |
08e007d2 | 5272 | * |
957780eb JB |
5273 | * If we are a priority flusher then we just need to add our ticket to |
5274 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5275 | */ |
72bcd99d | 5276 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5277 | ticket.bytes = orig_bytes; |
5278 | ticket.error = 0; | |
5279 | init_waitqueue_head(&ticket.wait); | |
5280 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5281 | list_add_tail(&ticket.list, &space_info->tickets); | |
5282 | if (!space_info->flush) { | |
5283 | space_info->flush = 1; | |
0b246afa | 5284 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5285 | space_info->flags, |
5286 | orig_bytes, flush, | |
5287 | "enospc"); | |
957780eb | 5288 | queue_work(system_unbound_wq, |
c1c4919b | 5289 | &fs_info->async_reclaim_work); |
957780eb JB |
5290 | } |
5291 | } else { | |
5292 | list_add_tail(&ticket.list, | |
5293 | &space_info->priority_tickets); | |
5294 | } | |
21c7e756 MX |
5295 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5296 | used += orig_bytes; | |
f6acfd50 JB |
5297 | /* |
5298 | * We will do the space reservation dance during log replay, | |
5299 | * which means we won't have fs_info->fs_root set, so don't do | |
5300 | * the async reclaim as we will panic. | |
5301 | */ | |
0b246afa | 5302 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5303 | need_do_async_reclaim(fs_info, space_info, |
5304 | used, system_chunk) && | |
0b246afa JM |
5305 | !work_busy(&fs_info->async_reclaim_work)) { |
5306 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5307 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5308 | queue_work(system_unbound_wq, |
0b246afa | 5309 | &fs_info->async_reclaim_work); |
f376df2b | 5310 | } |
8bb8ab2e | 5311 | } |
f0486c68 | 5312 | spin_unlock(&space_info->lock); |
08e007d2 | 5313 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5314 | return ret; |
f0486c68 | 5315 | |
957780eb | 5316 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5317 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5318 | orig_bytes); |
08e007d2 | 5319 | |
957780eb | 5320 | ret = 0; |
0b246afa | 5321 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5322 | spin_lock(&space_info->lock); |
5323 | if (ticket.bytes) { | |
5324 | if (ticket.bytes < orig_bytes) { | |
5325 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5326 | space_info->bytes_may_use -= num_bytes; | |
0b246afa JM |
5327 | trace_btrfs_space_reservation(fs_info, "space_info", |
5328 | space_info->flags, | |
5329 | num_bytes, 0); | |
08e007d2 | 5330 | |
957780eb JB |
5331 | } |
5332 | list_del_init(&ticket.list); | |
5333 | ret = -ENOSPC; | |
5334 | } | |
5335 | spin_unlock(&space_info->lock); | |
5336 | ASSERT(list_empty(&ticket.list)); | |
5337 | return ret; | |
5338 | } | |
8bb8ab2e | 5339 | |
957780eb JB |
5340 | /** |
5341 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5342 | * @root - the root we're allocating for | |
5343 | * @block_rsv - the block_rsv we're allocating for | |
5344 | * @orig_bytes - the number of bytes we want | |
5345 | * @flush - whether or not we can flush to make our reservation | |
5346 | * | |
5347 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5348 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5349 | * flush out space to make room. It will do this by flushing delalloc if | |
5350 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5351 | * regain reservations will be made and this will fail if there is not enough | |
5352 | * space already. | |
5353 | */ | |
5354 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5355 | struct btrfs_block_rsv *block_rsv, | |
5356 | u64 orig_bytes, | |
5357 | enum btrfs_reserve_flush_enum flush) | |
5358 | { | |
0b246afa JM |
5359 | struct btrfs_fs_info *fs_info = root->fs_info; |
5360 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5361 | int ret; |
c1c4919b | 5362 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5363 | |
c1c4919b JM |
5364 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5365 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5366 | if (ret == -ENOSPC && |
5367 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5368 | if (block_rsv != global_rsv && |
5369 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5370 | ret = 0; | |
5371 | } | |
9a3daff3 | 5372 | if (ret == -ENOSPC) { |
0b246afa | 5373 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5374 | block_rsv->space_info->flags, |
5375 | orig_bytes, 1); | |
9a3daff3 NB |
5376 | |
5377 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
5378 | dump_space_info(fs_info, block_rsv->space_info, | |
5379 | orig_bytes, 0); | |
5380 | } | |
f0486c68 YZ |
5381 | return ret; |
5382 | } | |
5383 | ||
79787eaa JM |
5384 | static struct btrfs_block_rsv *get_block_rsv( |
5385 | const struct btrfs_trans_handle *trans, | |
5386 | const struct btrfs_root *root) | |
f0486c68 | 5387 | { |
0b246afa | 5388 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5389 | struct btrfs_block_rsv *block_rsv = NULL; |
5390 | ||
e9cf439f | 5391 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5392 | (root == fs_info->csum_root && trans->adding_csums) || |
5393 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5394 | block_rsv = trans->block_rsv; |
5395 | ||
4c13d758 | 5396 | if (!block_rsv) |
f0486c68 YZ |
5397 | block_rsv = root->block_rsv; |
5398 | ||
5399 | if (!block_rsv) | |
0b246afa | 5400 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5401 | |
5402 | return block_rsv; | |
5403 | } | |
5404 | ||
5405 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5406 | u64 num_bytes) | |
5407 | { | |
5408 | int ret = -ENOSPC; | |
5409 | spin_lock(&block_rsv->lock); | |
5410 | if (block_rsv->reserved >= num_bytes) { | |
5411 | block_rsv->reserved -= num_bytes; | |
5412 | if (block_rsv->reserved < block_rsv->size) | |
5413 | block_rsv->full = 0; | |
5414 | ret = 0; | |
5415 | } | |
5416 | spin_unlock(&block_rsv->lock); | |
5417 | return ret; | |
5418 | } | |
5419 | ||
5420 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5421 | u64 num_bytes, int update_size) | |
5422 | { | |
5423 | spin_lock(&block_rsv->lock); | |
5424 | block_rsv->reserved += num_bytes; | |
5425 | if (update_size) | |
5426 | block_rsv->size += num_bytes; | |
5427 | else if (block_rsv->reserved >= block_rsv->size) | |
5428 | block_rsv->full = 1; | |
5429 | spin_unlock(&block_rsv->lock); | |
5430 | } | |
5431 | ||
d52be818 JB |
5432 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5433 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5434 | int min_factor) | |
5435 | { | |
5436 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5437 | u64 min_bytes; | |
5438 | ||
5439 | if (global_rsv->space_info != dest->space_info) | |
5440 | return -ENOSPC; | |
5441 | ||
5442 | spin_lock(&global_rsv->lock); | |
5443 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5444 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5445 | spin_unlock(&global_rsv->lock); | |
5446 | return -ENOSPC; | |
5447 | } | |
5448 | global_rsv->reserved -= num_bytes; | |
5449 | if (global_rsv->reserved < global_rsv->size) | |
5450 | global_rsv->full = 0; | |
5451 | spin_unlock(&global_rsv->lock); | |
5452 | ||
5453 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5454 | return 0; | |
5455 | } | |
5456 | ||
957780eb JB |
5457 | /* |
5458 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5459 | * basically when we're returning space from block_rsv's. | |
5460 | */ | |
5461 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5462 | struct btrfs_space_info *space_info, | |
5463 | u64 num_bytes) | |
5464 | { | |
5465 | struct reserve_ticket *ticket; | |
5466 | struct list_head *head; | |
5467 | u64 used; | |
5468 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5469 | bool check_overcommit = false; | |
5470 | ||
5471 | spin_lock(&space_info->lock); | |
5472 | head = &space_info->priority_tickets; | |
5473 | ||
5474 | /* | |
5475 | * If we are over our limit then we need to check and see if we can | |
5476 | * overcommit, and if we can't then we just need to free up our space | |
5477 | * and not satisfy any requests. | |
5478 | */ | |
0eee8a49 | 5479 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5480 | if (used - num_bytes >= space_info->total_bytes) |
5481 | check_overcommit = true; | |
5482 | again: | |
5483 | while (!list_empty(head) && num_bytes) { | |
5484 | ticket = list_first_entry(head, struct reserve_ticket, | |
5485 | list); | |
5486 | /* | |
5487 | * We use 0 bytes because this space is already reserved, so | |
5488 | * adding the ticket space would be a double count. | |
5489 | */ | |
5490 | if (check_overcommit && | |
c1c4919b | 5491 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5492 | break; |
5493 | if (num_bytes >= ticket->bytes) { | |
5494 | list_del_init(&ticket->list); | |
5495 | num_bytes -= ticket->bytes; | |
5496 | ticket->bytes = 0; | |
ce129655 | 5497 | space_info->tickets_id++; |
957780eb JB |
5498 | wake_up(&ticket->wait); |
5499 | } else { | |
5500 | ticket->bytes -= num_bytes; | |
5501 | num_bytes = 0; | |
5502 | } | |
5503 | } | |
5504 | ||
5505 | if (num_bytes && head == &space_info->priority_tickets) { | |
5506 | head = &space_info->tickets; | |
5507 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5508 | goto again; | |
5509 | } | |
5510 | space_info->bytes_may_use -= num_bytes; | |
5511 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5512 | space_info->flags, num_bytes, 0); | |
5513 | spin_unlock(&space_info->lock); | |
5514 | } | |
5515 | ||
5516 | /* | |
5517 | * This is for newly allocated space that isn't accounted in | |
5518 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5519 | * we use this helper. | |
5520 | */ | |
5521 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5522 | struct btrfs_space_info *space_info, | |
5523 | u64 num_bytes) | |
5524 | { | |
5525 | struct reserve_ticket *ticket; | |
5526 | struct list_head *head = &space_info->priority_tickets; | |
5527 | ||
5528 | again: | |
5529 | while (!list_empty(head) && num_bytes) { | |
5530 | ticket = list_first_entry(head, struct reserve_ticket, | |
5531 | list); | |
5532 | if (num_bytes >= ticket->bytes) { | |
5533 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5534 | space_info->flags, | |
5535 | ticket->bytes, 1); | |
5536 | list_del_init(&ticket->list); | |
5537 | num_bytes -= ticket->bytes; | |
5538 | space_info->bytes_may_use += ticket->bytes; | |
5539 | ticket->bytes = 0; | |
ce129655 | 5540 | space_info->tickets_id++; |
957780eb JB |
5541 | wake_up(&ticket->wait); |
5542 | } else { | |
5543 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5544 | space_info->flags, | |
5545 | num_bytes, 1); | |
5546 | space_info->bytes_may_use += num_bytes; | |
5547 | ticket->bytes -= num_bytes; | |
5548 | num_bytes = 0; | |
5549 | } | |
5550 | } | |
5551 | ||
5552 | if (num_bytes && head == &space_info->priority_tickets) { | |
5553 | head = &space_info->tickets; | |
5554 | goto again; | |
5555 | } | |
5556 | } | |
5557 | ||
69fe2d75 | 5558 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
8c2a3ca2 | 5559 | struct btrfs_block_rsv *block_rsv, |
ff6bc37e QW |
5560 | struct btrfs_block_rsv *dest, u64 num_bytes, |
5561 | u64 *qgroup_to_release_ret) | |
f0486c68 YZ |
5562 | { |
5563 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
ff6bc37e | 5564 | u64 qgroup_to_release = 0; |
69fe2d75 | 5565 | u64 ret; |
f0486c68 YZ |
5566 | |
5567 | spin_lock(&block_rsv->lock); | |
ff6bc37e | 5568 | if (num_bytes == (u64)-1) { |
f0486c68 | 5569 | num_bytes = block_rsv->size; |
ff6bc37e QW |
5570 | qgroup_to_release = block_rsv->qgroup_rsv_size; |
5571 | } | |
f0486c68 YZ |
5572 | block_rsv->size -= num_bytes; |
5573 | if (block_rsv->reserved >= block_rsv->size) { | |
5574 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5575 | block_rsv->reserved = block_rsv->size; | |
5576 | block_rsv->full = 1; | |
5577 | } else { | |
5578 | num_bytes = 0; | |
5579 | } | |
ff6bc37e QW |
5580 | if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) { |
5581 | qgroup_to_release = block_rsv->qgroup_rsv_reserved - | |
5582 | block_rsv->qgroup_rsv_size; | |
5583 | block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size; | |
5584 | } else { | |
5585 | qgroup_to_release = 0; | |
5586 | } | |
f0486c68 YZ |
5587 | spin_unlock(&block_rsv->lock); |
5588 | ||
69fe2d75 | 5589 | ret = num_bytes; |
f0486c68 YZ |
5590 | if (num_bytes > 0) { |
5591 | if (dest) { | |
e9e22899 JB |
5592 | spin_lock(&dest->lock); |
5593 | if (!dest->full) { | |
5594 | u64 bytes_to_add; | |
5595 | ||
5596 | bytes_to_add = dest->size - dest->reserved; | |
5597 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5598 | dest->reserved += bytes_to_add; | |
5599 | if (dest->reserved >= dest->size) | |
5600 | dest->full = 1; | |
5601 | num_bytes -= bytes_to_add; | |
5602 | } | |
5603 | spin_unlock(&dest->lock); | |
5604 | } | |
957780eb JB |
5605 | if (num_bytes) |
5606 | space_info_add_old_bytes(fs_info, space_info, | |
5607 | num_bytes); | |
9ed74f2d | 5608 | } |
ff6bc37e QW |
5609 | if (qgroup_to_release_ret) |
5610 | *qgroup_to_release_ret = qgroup_to_release; | |
69fe2d75 | 5611 | return ret; |
f0486c68 | 5612 | } |
4e06bdd6 | 5613 | |
25d609f8 JB |
5614 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5615 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
5616 | int update_size) | |
f0486c68 YZ |
5617 | { |
5618 | int ret; | |
9ed74f2d | 5619 | |
f0486c68 YZ |
5620 | ret = block_rsv_use_bytes(src, num_bytes); |
5621 | if (ret) | |
5622 | return ret; | |
9ed74f2d | 5623 | |
25d609f8 | 5624 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5625 | return 0; |
5626 | } | |
5627 | ||
66d8f3dd | 5628 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5629 | { |
f0486c68 YZ |
5630 | memset(rsv, 0, sizeof(*rsv)); |
5631 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5632 | rsv->type = type; |
f0486c68 YZ |
5633 | } |
5634 | ||
69fe2d75 JB |
5635 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
5636 | struct btrfs_block_rsv *rsv, | |
5637 | unsigned short type) | |
5638 | { | |
5639 | btrfs_init_block_rsv(rsv, type); | |
5640 | rsv->space_info = __find_space_info(fs_info, | |
5641 | BTRFS_BLOCK_GROUP_METADATA); | |
5642 | } | |
5643 | ||
2ff7e61e | 5644 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5645 | unsigned short type) |
f0486c68 YZ |
5646 | { |
5647 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5648 | |
f0486c68 YZ |
5649 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5650 | if (!block_rsv) | |
5651 | return NULL; | |
9ed74f2d | 5652 | |
69fe2d75 | 5653 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
f0486c68 YZ |
5654 | return block_rsv; |
5655 | } | |
9ed74f2d | 5656 | |
2ff7e61e | 5657 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5658 | struct btrfs_block_rsv *rsv) |
5659 | { | |
2aaa6655 JB |
5660 | if (!rsv) |
5661 | return; | |
2ff7e61e | 5662 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5663 | kfree(rsv); |
9ed74f2d JB |
5664 | } |
5665 | ||
cdfb080e CM |
5666 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5667 | { | |
5668 | kfree(rsv); | |
5669 | } | |
5670 | ||
08e007d2 MX |
5671 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5672 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5673 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5674 | { |
f0486c68 | 5675 | int ret; |
9ed74f2d | 5676 | |
f0486c68 YZ |
5677 | if (num_bytes == 0) |
5678 | return 0; | |
8bb8ab2e | 5679 | |
61b520a9 | 5680 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5681 | if (!ret) { |
5682 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5683 | return 0; | |
5684 | } | |
9ed74f2d | 5685 | |
f0486c68 | 5686 | return ret; |
f0486c68 | 5687 | } |
9ed74f2d | 5688 | |
2ff7e61e | 5689 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5690 | { |
5691 | u64 num_bytes = 0; | |
f0486c68 | 5692 | int ret = -ENOSPC; |
9ed74f2d | 5693 | |
f0486c68 YZ |
5694 | if (!block_rsv) |
5695 | return 0; | |
9ed74f2d | 5696 | |
f0486c68 | 5697 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5698 | num_bytes = div_factor(block_rsv->size, min_factor); |
5699 | if (block_rsv->reserved >= num_bytes) | |
5700 | ret = 0; | |
5701 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5702 | |
36ba022a JB |
5703 | return ret; |
5704 | } | |
5705 | ||
08e007d2 MX |
5706 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5707 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5708 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5709 | { |
5710 | u64 num_bytes = 0; | |
5711 | int ret = -ENOSPC; | |
5712 | ||
5713 | if (!block_rsv) | |
5714 | return 0; | |
5715 | ||
5716 | spin_lock(&block_rsv->lock); | |
5717 | num_bytes = min_reserved; | |
13553e52 | 5718 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5719 | ret = 0; |
13553e52 | 5720 | else |
f0486c68 | 5721 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5722 | spin_unlock(&block_rsv->lock); |
13553e52 | 5723 | |
f0486c68 YZ |
5724 | if (!ret) |
5725 | return 0; | |
5726 | ||
aa38a711 | 5727 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5728 | if (!ret) { |
5729 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5730 | return 0; |
6a63209f | 5731 | } |
9ed74f2d | 5732 | |
13553e52 | 5733 | return ret; |
f0486c68 YZ |
5734 | } |
5735 | ||
69fe2d75 JB |
5736 | /** |
5737 | * btrfs_inode_rsv_refill - refill the inode block rsv. | |
5738 | * @inode - the inode we are refilling. | |
5739 | * @flush - the flusing restriction. | |
5740 | * | |
5741 | * Essentially the same as btrfs_block_rsv_refill, except it uses the | |
5742 | * block_rsv->size as the minimum size. We'll either refill the missing amount | |
5743 | * or return if we already have enough space. This will also handle the resreve | |
5744 | * tracepoint for the reserved amount. | |
5745 | */ | |
3f2dd7a0 QW |
5746 | static int btrfs_inode_rsv_refill(struct btrfs_inode *inode, |
5747 | enum btrfs_reserve_flush_enum flush) | |
69fe2d75 JB |
5748 | { |
5749 | struct btrfs_root *root = inode->root; | |
5750 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5751 | u64 num_bytes = 0; | |
ff6bc37e | 5752 | u64 qgroup_num_bytes = 0; |
69fe2d75 JB |
5753 | int ret = -ENOSPC; |
5754 | ||
5755 | spin_lock(&block_rsv->lock); | |
5756 | if (block_rsv->reserved < block_rsv->size) | |
5757 | num_bytes = block_rsv->size - block_rsv->reserved; | |
ff6bc37e QW |
5758 | if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size) |
5759 | qgroup_num_bytes = block_rsv->qgroup_rsv_size - | |
5760 | block_rsv->qgroup_rsv_reserved; | |
69fe2d75 JB |
5761 | spin_unlock(&block_rsv->lock); |
5762 | ||
5763 | if (num_bytes == 0) | |
5764 | return 0; | |
5765 | ||
ff6bc37e | 5766 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true); |
43b18595 QW |
5767 | if (ret) |
5768 | return ret; | |
69fe2d75 JB |
5769 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5770 | if (!ret) { | |
5771 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
5772 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5773 | btrfs_ino(inode), num_bytes, 1); | |
ff6bc37e QW |
5774 | |
5775 | /* Don't forget to increase qgroup_rsv_reserved */ | |
5776 | spin_lock(&block_rsv->lock); | |
5777 | block_rsv->qgroup_rsv_reserved += qgroup_num_bytes; | |
5778 | spin_unlock(&block_rsv->lock); | |
5779 | } else | |
5780 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
69fe2d75 JB |
5781 | return ret; |
5782 | } | |
5783 | ||
5784 | /** | |
5785 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5786 | * @inode - the inode we need to release from. | |
43b18595 QW |
5787 | * @qgroup_free - free or convert qgroup meta. |
5788 | * Unlike normal operation, qgroup meta reservation needs to know if we are | |
5789 | * freeing qgroup reservation or just converting it into per-trans. Normally | |
5790 | * @qgroup_free is true for error handling, and false for normal release. | |
69fe2d75 JB |
5791 | * |
5792 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5793 | * tracepoint for the reservation. | |
5794 | */ | |
43b18595 | 5795 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
69fe2d75 JB |
5796 | { |
5797 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
5798 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5799 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5800 | u64 released = 0; | |
ff6bc37e | 5801 | u64 qgroup_to_release = 0; |
69fe2d75 JB |
5802 | |
5803 | /* | |
5804 | * Since we statically set the block_rsv->size we just want to say we | |
5805 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5806 | * the size free'd. | |
5807 | */ | |
ff6bc37e QW |
5808 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0, |
5809 | &qgroup_to_release); | |
69fe2d75 JB |
5810 | if (released > 0) |
5811 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5812 | btrfs_ino(inode), released, 0); | |
43b18595 | 5813 | if (qgroup_free) |
ff6bc37e | 5814 | btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); |
43b18595 | 5815 | else |
ff6bc37e QW |
5816 | btrfs_qgroup_convert_reserved_meta(inode->root, |
5817 | qgroup_to_release); | |
69fe2d75 JB |
5818 | } |
5819 | ||
2ff7e61e | 5820 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5821 | struct btrfs_block_rsv *block_rsv, |
5822 | u64 num_bytes) | |
5823 | { | |
0b246afa JM |
5824 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5825 | ||
17504584 | 5826 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5827 | block_rsv->space_info != global_rsv->space_info) |
5828 | global_rsv = NULL; | |
ff6bc37e | 5829 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL); |
6a63209f JB |
5830 | } |
5831 | ||
8929ecfa YZ |
5832 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5833 | { | |
5834 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5835 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5836 | u64 num_bytes; | |
6a63209f | 5837 | |
ae2e4728 JB |
5838 | /* |
5839 | * The global block rsv is based on the size of the extent tree, the | |
5840 | * checksum tree and the root tree. If the fs is empty we want to set | |
5841 | * it to a minimal amount for safety. | |
5842 | */ | |
5843 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5844 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5845 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5846 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5847 | |
8929ecfa | 5848 | spin_lock(&sinfo->lock); |
1f699d38 | 5849 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5850 | |
ee22184b | 5851 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5852 | |
fb4b10e5 | 5853 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5854 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5855 | if (sinfo->total_bytes > num_bytes) { |
5856 | num_bytes = sinfo->total_bytes - num_bytes; | |
5857 | num_bytes = min(num_bytes, | |
5858 | block_rsv->size - block_rsv->reserved); | |
5859 | block_rsv->reserved += num_bytes; | |
5860 | sinfo->bytes_may_use += num_bytes; | |
5861 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5862 | sinfo->flags, num_bytes, | |
5863 | 1); | |
5864 | } | |
5865 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5866 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5867 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5868 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5869 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5870 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5871 | } |
182608c8 | 5872 | |
fb4b10e5 JB |
5873 | if (block_rsv->reserved == block_rsv->size) |
5874 | block_rsv->full = 1; | |
5875 | else | |
5876 | block_rsv->full = 0; | |
5877 | ||
8929ecfa | 5878 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5879 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5880 | } |
5881 | ||
f0486c68 | 5882 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5883 | { |
f0486c68 | 5884 | struct btrfs_space_info *space_info; |
6a63209f | 5885 | |
f0486c68 YZ |
5886 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5887 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5888 | |
f0486c68 | 5889 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5890 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5891 | fs_info->trans_block_rsv.space_info = space_info; |
5892 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5893 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5894 | |
8929ecfa YZ |
5895 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5896 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5897 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5898 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5899 | if (fs_info->quota_root) |
5900 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5901 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5902 | |
8929ecfa | 5903 | update_global_block_rsv(fs_info); |
6a63209f JB |
5904 | } |
5905 | ||
8929ecfa | 5906 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5907 | { |
8c2a3ca2 | 5908 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
ff6bc37e | 5909 | (u64)-1, NULL); |
8929ecfa YZ |
5910 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5911 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5912 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5913 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5914 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5915 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5916 | } |
5917 | ||
6a63209f | 5918 | |
4fbcdf66 FM |
5919 | /* |
5920 | * To be called after all the new block groups attached to the transaction | |
5921 | * handle have been created (btrfs_create_pending_block_groups()). | |
5922 | */ | |
5923 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5924 | { | |
64b63580 | 5925 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5926 | |
5927 | if (!trans->chunk_bytes_reserved) | |
5928 | return; | |
5929 | ||
5930 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5931 | ||
5932 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
ff6bc37e | 5933 | trans->chunk_bytes_reserved, NULL); |
4fbcdf66 FM |
5934 | trans->chunk_bytes_reserved = 0; |
5935 | } | |
5936 | ||
d5c12070 MX |
5937 | /* |
5938 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5939 | * root: the root of the parent directory | |
5940 | * rsv: block reservation | |
5941 | * items: the number of items that we need do reservation | |
5942 | * qgroup_reserved: used to return the reserved size in qgroup | |
5943 | * | |
5944 | * This function is used to reserve the space for snapshot/subvolume | |
5945 | * creation and deletion. Those operations are different with the | |
5946 | * common file/directory operations, they change two fs/file trees | |
5947 | * and root tree, the number of items that the qgroup reserves is | |
5948 | * different with the free space reservation. So we can not use | |
01327610 | 5949 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5950 | */ |
5951 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5952 | struct btrfs_block_rsv *rsv, | |
5953 | int items, | |
ee3441b4 JM |
5954 | u64 *qgroup_reserved, |
5955 | bool use_global_rsv) | |
a22285a6 | 5956 | { |
d5c12070 MX |
5957 | u64 num_bytes; |
5958 | int ret; | |
0b246afa JM |
5959 | struct btrfs_fs_info *fs_info = root->fs_info; |
5960 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5961 | |
0b246afa | 5962 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5963 | /* One for parent inode, two for dir entries */ |
0b246afa | 5964 | num_bytes = 3 * fs_info->nodesize; |
733e03a0 | 5965 | ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true); |
d5c12070 MX |
5966 | if (ret) |
5967 | return ret; | |
5968 | } else { | |
5969 | num_bytes = 0; | |
5970 | } | |
5971 | ||
5972 | *qgroup_reserved = num_bytes; | |
5973 | ||
0b246afa JM |
5974 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
5975 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
5976 | BTRFS_BLOCK_GROUP_METADATA); |
5977 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5978 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5979 | |
5980 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 5981 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 5982 | |
7174109c | 5983 | if (ret && *qgroup_reserved) |
733e03a0 | 5984 | btrfs_qgroup_free_meta_prealloc(root, *qgroup_reserved); |
d5c12070 MX |
5985 | |
5986 | return ret; | |
5987 | } | |
5988 | ||
2ff7e61e | 5989 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 5990 | struct btrfs_block_rsv *rsv) |
d5c12070 | 5991 | { |
2ff7e61e | 5992 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
5993 | } |
5994 | ||
69fe2d75 JB |
5995 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
5996 | struct btrfs_inode *inode) | |
9e0baf60 | 5997 | { |
69fe2d75 JB |
5998 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
5999 | u64 reserve_size = 0; | |
ff6bc37e | 6000 | u64 qgroup_rsv_size = 0; |
69fe2d75 JB |
6001 | u64 csum_leaves; |
6002 | unsigned outstanding_extents; | |
9e0baf60 | 6003 | |
69fe2d75 JB |
6004 | lockdep_assert_held(&inode->lock); |
6005 | outstanding_extents = inode->outstanding_extents; | |
6006 | if (outstanding_extents) | |
6007 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
6008 | outstanding_extents + 1); | |
6009 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
6010 | inode->csum_bytes); | |
6011 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
6012 | csum_leaves); | |
ff6bc37e QW |
6013 | /* |
6014 | * For qgroup rsv, the calculation is very simple: | |
6015 | * account one nodesize for each outstanding extent | |
6016 | * | |
6017 | * This is overestimating in most cases. | |
6018 | */ | |
6019 | qgroup_rsv_size = outstanding_extents * fs_info->nodesize; | |
9e0baf60 | 6020 | |
69fe2d75 JB |
6021 | spin_lock(&block_rsv->lock); |
6022 | block_rsv->size = reserve_size; | |
ff6bc37e | 6023 | block_rsv->qgroup_rsv_size = qgroup_rsv_size; |
69fe2d75 | 6024 | spin_unlock(&block_rsv->lock); |
0ca1f7ce | 6025 | } |
c146afad | 6026 | |
9f3db423 | 6027 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6028 | { |
9f3db423 | 6029 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
69fe2d75 | 6030 | unsigned nr_extents; |
08e007d2 | 6031 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 6032 | int ret = 0; |
c64c2bd8 | 6033 | bool delalloc_lock = true; |
6324fbf3 | 6034 | |
c64c2bd8 JB |
6035 | /* If we are a free space inode we need to not flush since we will be in |
6036 | * the middle of a transaction commit. We also don't need the delalloc | |
6037 | * mutex since we won't race with anybody. We need this mostly to make | |
6038 | * lockdep shut its filthy mouth. | |
bac357dc JB |
6039 | * |
6040 | * If we have a transaction open (can happen if we call truncate_block | |
6041 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
6042 | */ |
6043 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 6044 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 6045 | delalloc_lock = false; |
da07d4ab NB |
6046 | } else { |
6047 | if (current->journal_info) | |
6048 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 6049 | |
da07d4ab NB |
6050 | if (btrfs_transaction_in_commit(fs_info)) |
6051 | schedule_timeout(1); | |
6052 | } | |
ec44a35c | 6053 | |
c64c2bd8 | 6054 | if (delalloc_lock) |
9f3db423 | 6055 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 6056 | |
0b246afa | 6057 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 JB |
6058 | |
6059 | /* Add our new extents and calculate the new rsv size. */ | |
9f3db423 | 6060 | spin_lock(&inode->lock); |
69fe2d75 | 6061 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 6062 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
6063 | inode->csum_bytes += num_bytes; |
6064 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6065 | spin_unlock(&inode->lock); |
57a45ced | 6066 | |
69fe2d75 | 6067 | ret = btrfs_inode_rsv_refill(inode, flush); |
43b18595 | 6068 | if (unlikely(ret)) |
88e081bf | 6069 | goto out_fail; |
25179201 | 6070 | |
c64c2bd8 | 6071 | if (delalloc_lock) |
9f3db423 | 6072 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 6073 | return 0; |
88e081bf WS |
6074 | |
6075 | out_fail: | |
9f3db423 | 6076 | spin_lock(&inode->lock); |
8b62f87b JB |
6077 | nr_extents = count_max_extents(num_bytes); |
6078 | btrfs_mod_outstanding_extents(inode, -nr_extents); | |
69fe2d75 JB |
6079 | inode->csum_bytes -= num_bytes; |
6080 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6081 | spin_unlock(&inode->lock); |
88e081bf | 6082 | |
43b18595 | 6083 | btrfs_inode_rsv_release(inode, true); |
88e081bf | 6084 | if (delalloc_lock) |
9f3db423 | 6085 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6086 | return ret; |
0ca1f7ce YZ |
6087 | } |
6088 | ||
7709cde3 JB |
6089 | /** |
6090 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
6091 | * @inode: the inode to release the reservation for. |
6092 | * @num_bytes: the number of bytes we are releasing. | |
43b18595 | 6093 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
7709cde3 JB |
6094 | * |
6095 | * This will release the metadata reservation for an inode. This can be called | |
6096 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 6097 | * reservations, or on error for the same reason. |
7709cde3 | 6098 | */ |
43b18595 QW |
6099 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
6100 | bool qgroup_free) | |
0ca1f7ce | 6101 | { |
691fa059 | 6102 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
0ca1f7ce | 6103 | |
0b246afa | 6104 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6105 | spin_lock(&inode->lock); |
69fe2d75 JB |
6106 | inode->csum_bytes -= num_bytes; |
6107 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 6108 | spin_unlock(&inode->lock); |
0ca1f7ce | 6109 | |
0b246afa | 6110 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6111 | return; |
6112 | ||
43b18595 | 6113 | btrfs_inode_rsv_release(inode, qgroup_free); |
0ca1f7ce YZ |
6114 | } |
6115 | ||
8b62f87b JB |
6116 | /** |
6117 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
6118 | * @inode: the inode to balance the reservation for. | |
6119 | * @num_bytes: the number of bytes we originally reserved with | |
43b18595 | 6120 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
8b62f87b JB |
6121 | * |
6122 | * When we reserve space we increase outstanding_extents for the extents we may | |
6123 | * add. Once we've set the range as delalloc or created our ordered extents we | |
6124 | * have outstanding_extents to track the real usage, so we use this to free our | |
6125 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
6126 | * with btrfs_delalloc_reserve_metadata. | |
6127 | */ | |
43b18595 QW |
6128 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
6129 | bool qgroup_free) | |
8b62f87b JB |
6130 | { |
6131 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); | |
6132 | unsigned num_extents; | |
8b62f87b JB |
6133 | |
6134 | spin_lock(&inode->lock); | |
6135 | num_extents = count_max_extents(num_bytes); | |
6136 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 6137 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
6138 | spin_unlock(&inode->lock); |
6139 | ||
8b62f87b JB |
6140 | if (btrfs_is_testing(fs_info)) |
6141 | return; | |
6142 | ||
43b18595 | 6143 | btrfs_inode_rsv_release(inode, qgroup_free); |
8b62f87b JB |
6144 | } |
6145 | ||
1ada3a62 | 6146 | /** |
7cf5b976 | 6147 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6148 | * delalloc |
6149 | * @inode: inode we're writing to | |
6150 | * @start: start range we are writing to | |
6151 | * @len: how long the range we are writing to | |
364ecf36 QW |
6152 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6153 | * current reservation. | |
1ada3a62 | 6154 | * |
1ada3a62 QW |
6155 | * This will do the following things |
6156 | * | |
6157 | * o reserve space in data space info for num bytes | |
6158 | * and reserve precious corresponding qgroup space | |
6159 | * (Done in check_data_free_space) | |
6160 | * | |
6161 | * o reserve space for metadata space, based on the number of outstanding | |
6162 | * extents and how much csums will be needed | |
6163 | * also reserve metadata space in a per root over-reserve method. | |
6164 | * o add to the inodes->delalloc_bytes | |
6165 | * o add it to the fs_info's delalloc inodes list. | |
6166 | * (Above 3 all done in delalloc_reserve_metadata) | |
6167 | * | |
6168 | * Return 0 for success | |
6169 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6170 | */ | |
364ecf36 QW |
6171 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6172 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6173 | { |
6174 | int ret; | |
6175 | ||
364ecf36 | 6176 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6177 | if (ret < 0) |
6178 | return ret; | |
9f3db423 | 6179 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6180 | if (ret < 0) |
bc42bda2 | 6181 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6182 | return ret; |
6183 | } | |
6184 | ||
7709cde3 | 6185 | /** |
7cf5b976 | 6186 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6187 | * @inode: inode we're releasing space for |
6188 | * @start: start position of the space already reserved | |
6189 | * @len: the len of the space already reserved | |
8b62f87b | 6190 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
6191 | * |
6192 | * This function will release the metadata space that was not used and will | |
6193 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6194 | * list if there are no delalloc bytes left. | |
6195 | * Also it will handle the qgroup reserved space. | |
6196 | */ | |
bc42bda2 | 6197 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b | 6198 | struct extent_changeset *reserved, |
43b18595 | 6199 | u64 start, u64 len, bool qgroup_free) |
1ada3a62 | 6200 | { |
43b18595 | 6201 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
bc42bda2 | 6202 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6203 | } |
6204 | ||
ce93ec54 | 6205 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6206 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6207 | u64 num_bytes, int alloc) |
9078a3e1 | 6208 | { |
0af3d00b | 6209 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6210 | u64 total = num_bytes; |
9078a3e1 | 6211 | u64 old_val; |
db94535d | 6212 | u64 byte_in_group; |
0af3d00b | 6213 | int factor; |
3e1ad54f | 6214 | |
5d4f98a2 | 6215 | /* block accounting for super block */ |
eb73c1b7 | 6216 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6217 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6218 | if (alloc) |
6219 | old_val += num_bytes; | |
6220 | else | |
6221 | old_val -= num_bytes; | |
6c41761f | 6222 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6223 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6224 | |
d397712b | 6225 | while (total) { |
db94535d | 6226 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6227 | if (!cache) |
79787eaa | 6228 | return -ENOENT; |
b742bb82 YZ |
6229 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6230 | BTRFS_BLOCK_GROUP_RAID1 | | |
6231 | BTRFS_BLOCK_GROUP_RAID10)) | |
6232 | factor = 2; | |
6233 | else | |
6234 | factor = 1; | |
9d66e233 JB |
6235 | /* |
6236 | * If this block group has free space cache written out, we | |
6237 | * need to make sure to load it if we are removing space. This | |
6238 | * is because we need the unpinning stage to actually add the | |
6239 | * space back to the block group, otherwise we will leak space. | |
6240 | */ | |
6241 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6242 | cache_block_group(cache, 1); |
0af3d00b | 6243 | |
db94535d CM |
6244 | byte_in_group = bytenr - cache->key.objectid; |
6245 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6246 | |
25179201 | 6247 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6248 | spin_lock(&cache->lock); |
0af3d00b | 6249 | |
6202df69 | 6250 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6251 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6252 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6253 | ||
9078a3e1 | 6254 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6255 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6256 | if (alloc) { |
db94535d | 6257 | old_val += num_bytes; |
11833d66 YZ |
6258 | btrfs_set_block_group_used(&cache->item, old_val); |
6259 | cache->reserved -= num_bytes; | |
11833d66 | 6260 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6261 | cache->space_info->bytes_used += num_bytes; |
6262 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6263 | spin_unlock(&cache->lock); |
25179201 | 6264 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6265 | } else { |
db94535d | 6266 | old_val -= num_bytes; |
ae0ab003 FM |
6267 | btrfs_set_block_group_used(&cache->item, old_val); |
6268 | cache->pinned += num_bytes; | |
6269 | cache->space_info->bytes_pinned += num_bytes; | |
6270 | cache->space_info->bytes_used -= num_bytes; | |
6271 | cache->space_info->disk_used -= num_bytes * factor; | |
6272 | spin_unlock(&cache->lock); | |
6273 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6274 | |
0b246afa | 6275 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6276 | cache->space_info->flags, |
6277 | num_bytes, 1); | |
d7eae340 OS |
6278 | percpu_counter_add(&cache->space_info->total_bytes_pinned, |
6279 | num_bytes); | |
ae0ab003 FM |
6280 | set_extent_dirty(info->pinned_extents, |
6281 | bytenr, bytenr + num_bytes - 1, | |
6282 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6283 | } |
1bbc621e CM |
6284 | |
6285 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6286 | if (list_empty(&cache->dirty_list)) { | |
6287 | list_add_tail(&cache->dirty_list, | |
6288 | &trans->transaction->dirty_bgs); | |
6289 | trans->transaction->num_dirty_bgs++; | |
6290 | btrfs_get_block_group(cache); | |
6291 | } | |
6292 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6293 | ||
036a9348 FM |
6294 | /* |
6295 | * No longer have used bytes in this block group, queue it for | |
6296 | * deletion. We do this after adding the block group to the | |
6297 | * dirty list to avoid races between cleaner kthread and space | |
6298 | * cache writeout. | |
6299 | */ | |
6300 | if (!alloc && old_val == 0) { | |
6301 | spin_lock(&info->unused_bgs_lock); | |
6302 | if (list_empty(&cache->bg_list)) { | |
6303 | btrfs_get_block_group(cache); | |
4ed0a7a3 | 6304 | trace_btrfs_add_unused_block_group(cache); |
036a9348 FM |
6305 | list_add_tail(&cache->bg_list, |
6306 | &info->unused_bgs); | |
6307 | } | |
6308 | spin_unlock(&info->unused_bgs_lock); | |
6309 | } | |
6310 | ||
fa9c0d79 | 6311 | btrfs_put_block_group(cache); |
db94535d CM |
6312 | total -= num_bytes; |
6313 | bytenr += num_bytes; | |
9078a3e1 CM |
6314 | } |
6315 | return 0; | |
6316 | } | |
6324fbf3 | 6317 | |
2ff7e61e | 6318 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6319 | { |
0f9dd46c | 6320 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6321 | u64 bytenr; |
0f9dd46c | 6322 | |
0b246afa JM |
6323 | spin_lock(&fs_info->block_group_cache_lock); |
6324 | bytenr = fs_info->first_logical_byte; | |
6325 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6326 | |
6327 | if (bytenr < (u64)-1) | |
6328 | return bytenr; | |
6329 | ||
0b246afa | 6330 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6331 | if (!cache) |
a061fc8d | 6332 | return 0; |
0f9dd46c | 6333 | |
d2fb3437 | 6334 | bytenr = cache->key.objectid; |
fa9c0d79 | 6335 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6336 | |
6337 | return bytenr; | |
a061fc8d CM |
6338 | } |
6339 | ||
2ff7e61e | 6340 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6341 | struct btrfs_block_group_cache *cache, |
6342 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6343 | { |
11833d66 YZ |
6344 | spin_lock(&cache->space_info->lock); |
6345 | spin_lock(&cache->lock); | |
6346 | cache->pinned += num_bytes; | |
6347 | cache->space_info->bytes_pinned += num_bytes; | |
6348 | if (reserved) { | |
6349 | cache->reserved -= num_bytes; | |
6350 | cache->space_info->bytes_reserved -= num_bytes; | |
6351 | } | |
6352 | spin_unlock(&cache->lock); | |
6353 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6354 | |
0b246afa | 6355 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6356 | cache->space_info->flags, num_bytes, 1); |
4da8b76d | 6357 | percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes); |
0b246afa | 6358 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6359 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6360 | return 0; | |
6361 | } | |
68b38550 | 6362 | |
f0486c68 YZ |
6363 | /* |
6364 | * this function must be called within transaction | |
6365 | */ | |
2ff7e61e | 6366 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6367 | u64 bytenr, u64 num_bytes, int reserved) |
6368 | { | |
6369 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6370 | |
0b246afa | 6371 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6372 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6373 | |
2ff7e61e | 6374 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6375 | |
6376 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6377 | return 0; |
6378 | } | |
6379 | ||
f0486c68 | 6380 | /* |
e688b725 CM |
6381 | * this function must be called within transaction |
6382 | */ | |
2ff7e61e | 6383 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6384 | u64 bytenr, u64 num_bytes) |
6385 | { | |
6386 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6387 | int ret; |
e688b725 | 6388 | |
0b246afa | 6389 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6390 | if (!cache) |
6391 | return -EINVAL; | |
e688b725 CM |
6392 | |
6393 | /* | |
6394 | * pull in the free space cache (if any) so that our pin | |
6395 | * removes the free space from the cache. We have load_only set | |
6396 | * to one because the slow code to read in the free extents does check | |
6397 | * the pinned extents. | |
6398 | */ | |
f6373bf3 | 6399 | cache_block_group(cache, 1); |
e688b725 | 6400 | |
2ff7e61e | 6401 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6402 | |
6403 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6404 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6405 | btrfs_put_block_group(cache); |
b50c6e25 | 6406 | return ret; |
e688b725 CM |
6407 | } |
6408 | ||
2ff7e61e JM |
6409 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6410 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6411 | { |
6412 | int ret; | |
6413 | struct btrfs_block_group_cache *block_group; | |
6414 | struct btrfs_caching_control *caching_ctl; | |
6415 | ||
0b246afa | 6416 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6417 | if (!block_group) |
6418 | return -EINVAL; | |
6419 | ||
6420 | cache_block_group(block_group, 0); | |
6421 | caching_ctl = get_caching_control(block_group); | |
6422 | ||
6423 | if (!caching_ctl) { | |
6424 | /* Logic error */ | |
6425 | BUG_ON(!block_group_cache_done(block_group)); | |
6426 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6427 | } else { | |
6428 | mutex_lock(&caching_ctl->mutex); | |
6429 | ||
6430 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6431 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6432 | } else if (start + num_bytes <= caching_ctl->progress) { |
6433 | ret = btrfs_remove_free_space(block_group, | |
6434 | start, num_bytes); | |
6435 | } else { | |
6436 | num_bytes = caching_ctl->progress - start; | |
6437 | ret = btrfs_remove_free_space(block_group, | |
6438 | start, num_bytes); | |
6439 | if (ret) | |
6440 | goto out_lock; | |
6441 | ||
6442 | num_bytes = (start + num_bytes) - | |
6443 | caching_ctl->progress; | |
6444 | start = caching_ctl->progress; | |
2ff7e61e | 6445 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6446 | } |
6447 | out_lock: | |
6448 | mutex_unlock(&caching_ctl->mutex); | |
6449 | put_caching_control(caching_ctl); | |
6450 | } | |
6451 | btrfs_put_block_group(block_group); | |
6452 | return ret; | |
6453 | } | |
6454 | ||
2ff7e61e | 6455 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6456 | struct extent_buffer *eb) |
6457 | { | |
6458 | struct btrfs_file_extent_item *item; | |
6459 | struct btrfs_key key; | |
6460 | int found_type; | |
6461 | int i; | |
6462 | ||
2ff7e61e | 6463 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6464 | return 0; |
6465 | ||
6466 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6467 | btrfs_item_key_to_cpu(eb, &key, i); | |
6468 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6469 | continue; | |
6470 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6471 | found_type = btrfs_file_extent_type(eb, item); | |
6472 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6473 | continue; | |
6474 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6475 | continue; | |
6476 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6477 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
2ff7e61e | 6478 | __exclude_logged_extent(fs_info, key.objectid, key.offset); |
8c2a1a30 JB |
6479 | } |
6480 | ||
6481 | return 0; | |
6482 | } | |
6483 | ||
9cfa3e34 FM |
6484 | static void |
6485 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6486 | { | |
6487 | atomic_inc(&bg->reservations); | |
6488 | } | |
6489 | ||
6490 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6491 | const u64 start) | |
6492 | { | |
6493 | struct btrfs_block_group_cache *bg; | |
6494 | ||
6495 | bg = btrfs_lookup_block_group(fs_info, start); | |
6496 | ASSERT(bg); | |
6497 | if (atomic_dec_and_test(&bg->reservations)) | |
4625956a | 6498 | wake_up_var(&bg->reservations); |
9cfa3e34 FM |
6499 | btrfs_put_block_group(bg); |
6500 | } | |
6501 | ||
9cfa3e34 FM |
6502 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
6503 | { | |
6504 | struct btrfs_space_info *space_info = bg->space_info; | |
6505 | ||
6506 | ASSERT(bg->ro); | |
6507 | ||
6508 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6509 | return; | |
6510 | ||
6511 | /* | |
6512 | * Our block group is read only but before we set it to read only, | |
6513 | * some task might have had allocated an extent from it already, but it | |
6514 | * has not yet created a respective ordered extent (and added it to a | |
6515 | * root's list of ordered extents). | |
6516 | * Therefore wait for any task currently allocating extents, since the | |
6517 | * block group's reservations counter is incremented while a read lock | |
6518 | * on the groups' semaphore is held and decremented after releasing | |
6519 | * the read access on that semaphore and creating the ordered extent. | |
6520 | */ | |
6521 | down_write(&space_info->groups_sem); | |
6522 | up_write(&space_info->groups_sem); | |
6523 | ||
4625956a | 6524 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); |
9cfa3e34 FM |
6525 | } |
6526 | ||
fb25e914 | 6527 | /** |
4824f1f4 | 6528 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6529 | * @cache: The cache we are manipulating |
18513091 WX |
6530 | * @ram_bytes: The number of bytes of file content, and will be same to |
6531 | * @num_bytes except for the compress path. | |
fb25e914 | 6532 | * @num_bytes: The number of bytes in question |
e570fd27 | 6533 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6534 | * |
745699ef XW |
6535 | * This is called by the allocator when it reserves space. If this is a |
6536 | * reservation and the block group has become read only we cannot make the | |
6537 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6538 | */ |
4824f1f4 | 6539 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6540 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6541 | { |
fb25e914 | 6542 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6543 | int ret = 0; |
79787eaa | 6544 | |
fb25e914 JB |
6545 | spin_lock(&space_info->lock); |
6546 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6547 | if (cache->ro) { |
6548 | ret = -EAGAIN; | |
fb25e914 | 6549 | } else { |
4824f1f4 WX |
6550 | cache->reserved += num_bytes; |
6551 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6552 | |
18513091 WX |
6553 | trace_btrfs_space_reservation(cache->fs_info, |
6554 | "space_info", space_info->flags, | |
6555 | ram_bytes, 0); | |
6556 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6557 | if (delalloc) |
4824f1f4 | 6558 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6559 | } |
fb25e914 JB |
6560 | spin_unlock(&cache->lock); |
6561 | spin_unlock(&space_info->lock); | |
f0486c68 | 6562 | return ret; |
324ae4df | 6563 | } |
9078a3e1 | 6564 | |
4824f1f4 WX |
6565 | /** |
6566 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6567 | * @cache: The cache we are manipulating | |
6568 | * @num_bytes: The number of bytes in question | |
6569 | * @delalloc: The blocks are allocated for the delalloc write | |
6570 | * | |
6571 | * This is called by somebody who is freeing space that was never actually used | |
6572 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6573 | * A and before transaction A commits you free that leaf, you call this with | |
6574 | * reserve set to 0 in order to clear the reservation. | |
6575 | */ | |
6576 | ||
6577 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6578 | u64 num_bytes, int delalloc) | |
6579 | { | |
6580 | struct btrfs_space_info *space_info = cache->space_info; | |
6581 | int ret = 0; | |
6582 | ||
6583 | spin_lock(&space_info->lock); | |
6584 | spin_lock(&cache->lock); | |
6585 | if (cache->ro) | |
6586 | space_info->bytes_readonly += num_bytes; | |
6587 | cache->reserved -= num_bytes; | |
6588 | space_info->bytes_reserved -= num_bytes; | |
6589 | ||
6590 | if (delalloc) | |
6591 | cache->delalloc_bytes -= num_bytes; | |
6592 | spin_unlock(&cache->lock); | |
6593 | spin_unlock(&space_info->lock); | |
6594 | return ret; | |
6595 | } | |
8b74c03e | 6596 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6597 | { |
11833d66 YZ |
6598 | struct btrfs_caching_control *next; |
6599 | struct btrfs_caching_control *caching_ctl; | |
6600 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6601 | |
9e351cc8 | 6602 | down_write(&fs_info->commit_root_sem); |
25179201 | 6603 | |
11833d66 YZ |
6604 | list_for_each_entry_safe(caching_ctl, next, |
6605 | &fs_info->caching_block_groups, list) { | |
6606 | cache = caching_ctl->block_group; | |
6607 | if (block_group_cache_done(cache)) { | |
6608 | cache->last_byte_to_unpin = (u64)-1; | |
6609 | list_del_init(&caching_ctl->list); | |
6610 | put_caching_control(caching_ctl); | |
e8569813 | 6611 | } else { |
11833d66 | 6612 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6613 | } |
e8569813 | 6614 | } |
11833d66 YZ |
6615 | |
6616 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6617 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6618 | else | |
6619 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6620 | ||
9e351cc8 | 6621 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6622 | |
6623 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6624 | } |
6625 | ||
c759c4e1 JB |
6626 | /* |
6627 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6628 | * what it should be based on the mount options. | |
6629 | */ | |
6630 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6631 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6632 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6633 | { |
6634 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6635 | |
6636 | *empty_cluster = 0; | |
6637 | if (btrfs_mixed_space_info(space_info)) | |
6638 | return ret; | |
6639 | ||
c759c4e1 | 6640 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6641 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6642 | if (btrfs_test_opt(fs_info, SSD)) |
6643 | *empty_cluster = SZ_2M; | |
6644 | else | |
ee22184b | 6645 | *empty_cluster = SZ_64K; |
583b7231 HK |
6646 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6647 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6648 | *empty_cluster = SZ_2M; | |
0b246afa | 6649 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6650 | } |
6651 | ||
6652 | return ret; | |
6653 | } | |
6654 | ||
2ff7e61e JM |
6655 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6656 | u64 start, u64 end, | |
678886bd | 6657 | const bool return_free_space) |
ccd467d6 | 6658 | { |
11833d66 | 6659 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6660 | struct btrfs_space_info *space_info; |
6661 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6662 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6663 | u64 len; |
c759c4e1 JB |
6664 | u64 total_unpinned = 0; |
6665 | u64 empty_cluster = 0; | |
7b398f8e | 6666 | bool readonly; |
ccd467d6 | 6667 | |
11833d66 | 6668 | while (start <= end) { |
7b398f8e | 6669 | readonly = false; |
11833d66 YZ |
6670 | if (!cache || |
6671 | start >= cache->key.objectid + cache->key.offset) { | |
6672 | if (cache) | |
6673 | btrfs_put_block_group(cache); | |
c759c4e1 | 6674 | total_unpinned = 0; |
11833d66 | 6675 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6676 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6677 | |
2ff7e61e | 6678 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6679 | cache->space_info, |
6680 | &empty_cluster); | |
6681 | empty_cluster <<= 1; | |
11833d66 YZ |
6682 | } |
6683 | ||
6684 | len = cache->key.objectid + cache->key.offset - start; | |
6685 | len = min(len, end + 1 - start); | |
6686 | ||
6687 | if (start < cache->last_byte_to_unpin) { | |
6688 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6689 | if (return_free_space) |
6690 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6691 | } |
6692 | ||
f0486c68 | 6693 | start += len; |
c759c4e1 | 6694 | total_unpinned += len; |
7b398f8e | 6695 | space_info = cache->space_info; |
f0486c68 | 6696 | |
c759c4e1 JB |
6697 | /* |
6698 | * If this space cluster has been marked as fragmented and we've | |
6699 | * unpinned enough in this block group to potentially allow a | |
6700 | * cluster to be created inside of it go ahead and clear the | |
6701 | * fragmented check. | |
6702 | */ | |
6703 | if (cluster && cluster->fragmented && | |
6704 | total_unpinned > empty_cluster) { | |
6705 | spin_lock(&cluster->lock); | |
6706 | cluster->fragmented = 0; | |
6707 | spin_unlock(&cluster->lock); | |
6708 | } | |
6709 | ||
7b398f8e | 6710 | spin_lock(&space_info->lock); |
11833d66 YZ |
6711 | spin_lock(&cache->lock); |
6712 | cache->pinned -= len; | |
7b398f8e | 6713 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6714 | |
6715 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6716 | space_info->flags, len, 0); | |
4f4db217 | 6717 | space_info->max_extent_size = 0; |
d288db5d | 6718 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6719 | if (cache->ro) { |
6720 | space_info->bytes_readonly += len; | |
6721 | readonly = true; | |
6722 | } | |
11833d66 | 6723 | spin_unlock(&cache->lock); |
957780eb JB |
6724 | if (!readonly && return_free_space && |
6725 | global_rsv->space_info == space_info) { | |
6726 | u64 to_add = len; | |
92ac58ec | 6727 | |
7b398f8e JB |
6728 | spin_lock(&global_rsv->lock); |
6729 | if (!global_rsv->full) { | |
957780eb JB |
6730 | to_add = min(len, global_rsv->size - |
6731 | global_rsv->reserved); | |
6732 | global_rsv->reserved += to_add; | |
6733 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6734 | if (global_rsv->reserved >= global_rsv->size) |
6735 | global_rsv->full = 1; | |
957780eb JB |
6736 | trace_btrfs_space_reservation(fs_info, |
6737 | "space_info", | |
6738 | space_info->flags, | |
6739 | to_add, 1); | |
6740 | len -= to_add; | |
7b398f8e JB |
6741 | } |
6742 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6743 | /* Add to any tickets we may have */ |
6744 | if (len) | |
6745 | space_info_add_new_bytes(fs_info, space_info, | |
6746 | len); | |
7b398f8e JB |
6747 | } |
6748 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6749 | } |
11833d66 YZ |
6750 | |
6751 | if (cache) | |
6752 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6753 | return 0; |
6754 | } | |
6755 | ||
5ead2dd0 | 6756 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 6757 | { |
5ead2dd0 | 6758 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
6759 | struct btrfs_block_group_cache *block_group, *tmp; |
6760 | struct list_head *deleted_bgs; | |
11833d66 | 6761 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6762 | u64 start; |
6763 | u64 end; | |
a28ec197 | 6764 | int ret; |
a28ec197 | 6765 | |
11833d66 YZ |
6766 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6767 | unpin = &fs_info->freed_extents[1]; | |
6768 | else | |
6769 | unpin = &fs_info->freed_extents[0]; | |
6770 | ||
e33e17ee | 6771 | while (!trans->aborted) { |
d4b450cd | 6772 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6773 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6774 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6775 | if (ret) { |
6776 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6777 | break; |
d4b450cd | 6778 | } |
1f3c79a2 | 6779 | |
0b246afa | 6780 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6781 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6782 | end + 1 - start, NULL); |
1f3c79a2 | 6783 | |
af6f8f60 | 6784 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6785 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6786 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6787 | cond_resched(); |
a28ec197 | 6788 | } |
817d52f8 | 6789 | |
e33e17ee JM |
6790 | /* |
6791 | * Transaction is finished. We don't need the lock anymore. We | |
6792 | * do need to clean up the block groups in case of a transaction | |
6793 | * abort. | |
6794 | */ | |
6795 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6796 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6797 | u64 trimmed = 0; | |
6798 | ||
6799 | ret = -EROFS; | |
6800 | if (!trans->aborted) | |
2ff7e61e | 6801 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6802 | block_group->key.objectid, |
6803 | block_group->key.offset, | |
6804 | &trimmed); | |
6805 | ||
6806 | list_del_init(&block_group->bg_list); | |
6807 | btrfs_put_block_group_trimming(block_group); | |
6808 | btrfs_put_block_group(block_group); | |
6809 | ||
6810 | if (ret) { | |
6811 | const char *errstr = btrfs_decode_error(ret); | |
6812 | btrfs_warn(fs_info, | |
913e1535 | 6813 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6814 | ret, errstr); |
6815 | } | |
6816 | } | |
6817 | ||
e20d96d6 CM |
6818 | return 0; |
6819 | } | |
6820 | ||
5d4f98a2 | 6821 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 6822 | struct btrfs_fs_info *info, |
c682f9b3 | 6823 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6824 | u64 root_objectid, u64 owner_objectid, |
6825 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6826 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6827 | { |
e2fa7227 | 6828 | struct btrfs_key key; |
5d4f98a2 | 6829 | struct btrfs_path *path; |
1261ec42 | 6830 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6831 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6832 | struct btrfs_extent_item *ei; |
6833 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6834 | int ret; |
5d4f98a2 | 6835 | int is_data; |
952fccac CM |
6836 | int extent_slot = 0; |
6837 | int found_extent = 0; | |
6838 | int num_to_del = 1; | |
5d4f98a2 YZ |
6839 | u32 item_size; |
6840 | u64 refs; | |
c682f9b3 QW |
6841 | u64 bytenr = node->bytenr; |
6842 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6843 | int last_ref = 0; |
0b246afa | 6844 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6845 | |
5caf2a00 | 6846 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6847 | if (!path) |
6848 | return -ENOMEM; | |
5f26f772 | 6849 | |
e4058b54 | 6850 | path->reada = READA_FORWARD; |
b9473439 | 6851 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6852 | |
6853 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6854 | BUG_ON(!is_data && refs_to_drop != 1); | |
6855 | ||
3173a18f | 6856 | if (is_data) |
897ca819 | 6857 | skinny_metadata = false; |
3173a18f | 6858 | |
87bde3cd | 6859 | ret = lookup_extent_backref(trans, info, path, &iref, |
5d4f98a2 YZ |
6860 | bytenr, num_bytes, parent, |
6861 | root_objectid, owner_objectid, | |
6862 | owner_offset); | |
7bb86316 | 6863 | if (ret == 0) { |
952fccac | 6864 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6865 | while (extent_slot >= 0) { |
6866 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6867 | extent_slot); |
5d4f98a2 | 6868 | if (key.objectid != bytenr) |
952fccac | 6869 | break; |
5d4f98a2 YZ |
6870 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6871 | key.offset == num_bytes) { | |
952fccac CM |
6872 | found_extent = 1; |
6873 | break; | |
6874 | } | |
3173a18f JB |
6875 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6876 | key.offset == owner_objectid) { | |
6877 | found_extent = 1; | |
6878 | break; | |
6879 | } | |
952fccac CM |
6880 | if (path->slots[0] - extent_slot > 5) |
6881 | break; | |
5d4f98a2 | 6882 | extent_slot--; |
952fccac | 6883 | } |
5d4f98a2 YZ |
6884 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6885 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6886 | if (found_extent && item_size < sizeof(*ei)) | |
6887 | found_extent = 0; | |
6888 | #endif | |
31840ae1 | 6889 | if (!found_extent) { |
5d4f98a2 | 6890 | BUG_ON(iref); |
87bde3cd JM |
6891 | ret = remove_extent_backref(trans, info, path, NULL, |
6892 | refs_to_drop, | |
fcebe456 | 6893 | is_data, &last_ref); |
005d6427 | 6894 | if (ret) { |
66642832 | 6895 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6896 | goto out; |
6897 | } | |
b3b4aa74 | 6898 | btrfs_release_path(path); |
b9473439 | 6899 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6900 | |
6901 | key.objectid = bytenr; | |
6902 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6903 | key.offset = num_bytes; | |
6904 | ||
3173a18f JB |
6905 | if (!is_data && skinny_metadata) { |
6906 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6907 | key.offset = owner_objectid; | |
6908 | } | |
6909 | ||
31840ae1 ZY |
6910 | ret = btrfs_search_slot(trans, extent_root, |
6911 | &key, path, -1, 1); | |
3173a18f JB |
6912 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6913 | /* | |
6914 | * Couldn't find our skinny metadata item, | |
6915 | * see if we have ye olde extent item. | |
6916 | */ | |
6917 | path->slots[0]--; | |
6918 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6919 | path->slots[0]); | |
6920 | if (key.objectid == bytenr && | |
6921 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6922 | key.offset == num_bytes) | |
6923 | ret = 0; | |
6924 | } | |
6925 | ||
6926 | if (ret > 0 && skinny_metadata) { | |
6927 | skinny_metadata = false; | |
9ce49a0b | 6928 | key.objectid = bytenr; |
3173a18f JB |
6929 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6930 | key.offset = num_bytes; | |
6931 | btrfs_release_path(path); | |
6932 | ret = btrfs_search_slot(trans, extent_root, | |
6933 | &key, path, -1, 1); | |
6934 | } | |
6935 | ||
f3465ca4 | 6936 | if (ret) { |
5d163e0e JM |
6937 | btrfs_err(info, |
6938 | "umm, got %d back from search, was looking for %llu", | |
6939 | ret, bytenr); | |
b783e62d | 6940 | if (ret > 0) |
a4f78750 | 6941 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6942 | } |
005d6427 | 6943 | if (ret < 0) { |
66642832 | 6944 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6945 | goto out; |
6946 | } | |
31840ae1 ZY |
6947 | extent_slot = path->slots[0]; |
6948 | } | |
fae7f21c | 6949 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 6950 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
6951 | btrfs_err(info, |
6952 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6953 | bytenr, parent, root_objectid, owner_objectid, |
6954 | owner_offset); | |
66642832 | 6955 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6956 | goto out; |
79787eaa | 6957 | } else { |
66642832 | 6958 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6959 | goto out; |
7bb86316 | 6960 | } |
5f39d397 CM |
6961 | |
6962 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6963 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6964 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6965 | if (item_size < sizeof(*ei)) { | |
6966 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
87bde3cd JM |
6967 | ret = convert_extent_item_v0(trans, info, path, owner_objectid, |
6968 | 0); | |
005d6427 | 6969 | if (ret < 0) { |
66642832 | 6970 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6971 | goto out; |
6972 | } | |
5d4f98a2 | 6973 | |
b3b4aa74 | 6974 | btrfs_release_path(path); |
5d4f98a2 YZ |
6975 | path->leave_spinning = 1; |
6976 | ||
6977 | key.objectid = bytenr; | |
6978 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6979 | key.offset = num_bytes; | |
6980 | ||
6981 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
6982 | -1, 1); | |
6983 | if (ret) { | |
5d163e0e JM |
6984 | btrfs_err(info, |
6985 | "umm, got %d back from search, was looking for %llu", | |
c1c9ff7c | 6986 | ret, bytenr); |
a4f78750 | 6987 | btrfs_print_leaf(path->nodes[0]); |
5d4f98a2 | 6988 | } |
005d6427 | 6989 | if (ret < 0) { |
66642832 | 6990 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6991 | goto out; |
6992 | } | |
6993 | ||
5d4f98a2 YZ |
6994 | extent_slot = path->slots[0]; |
6995 | leaf = path->nodes[0]; | |
6996 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
6997 | } | |
6998 | #endif | |
6999 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 7000 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 7001 | struct btrfs_extent_item); |
3173a18f JB |
7002 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7003 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7004 | struct btrfs_tree_block_info *bi; |
7005 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7006 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7007 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7008 | } | |
56bec294 | 7009 | |
5d4f98a2 | 7010 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7011 | if (refs < refs_to_drop) { |
5d163e0e JM |
7012 | btrfs_err(info, |
7013 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7014 | refs_to_drop, refs, bytenr); | |
32b02538 | 7015 | ret = -EINVAL; |
66642832 | 7016 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7017 | goto out; |
7018 | } | |
56bec294 | 7019 | refs -= refs_to_drop; |
5f39d397 | 7020 | |
5d4f98a2 YZ |
7021 | if (refs > 0) { |
7022 | if (extent_op) | |
7023 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7024 | /* | |
7025 | * In the case of inline back ref, reference count will | |
7026 | * be updated by remove_extent_backref | |
952fccac | 7027 | */ |
5d4f98a2 YZ |
7028 | if (iref) { |
7029 | BUG_ON(!found_extent); | |
7030 | } else { | |
7031 | btrfs_set_extent_refs(leaf, ei, refs); | |
7032 | btrfs_mark_buffer_dirty(leaf); | |
7033 | } | |
7034 | if (found_extent) { | |
87bde3cd | 7035 | ret = remove_extent_backref(trans, info, path, |
5d4f98a2 | 7036 | iref, refs_to_drop, |
fcebe456 | 7037 | is_data, &last_ref); |
005d6427 | 7038 | if (ret) { |
66642832 | 7039 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7040 | goto out; |
7041 | } | |
952fccac | 7042 | } |
5d4f98a2 | 7043 | } else { |
5d4f98a2 YZ |
7044 | if (found_extent) { |
7045 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7046 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7047 | if (iref) { |
7048 | BUG_ON(path->slots[0] != extent_slot); | |
7049 | } else { | |
7050 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7051 | path->slots[0] = extent_slot; | |
7052 | num_to_del = 2; | |
7053 | } | |
78fae27e | 7054 | } |
b9473439 | 7055 | |
fcebe456 | 7056 | last_ref = 1; |
952fccac CM |
7057 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7058 | num_to_del); | |
005d6427 | 7059 | if (ret) { |
66642832 | 7060 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7061 | goto out; |
7062 | } | |
b3b4aa74 | 7063 | btrfs_release_path(path); |
21af804c | 7064 | |
5d4f98a2 | 7065 | if (is_data) { |
5b4aacef | 7066 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7067 | if (ret) { |
66642832 | 7068 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7069 | goto out; |
7070 | } | |
459931ec CM |
7071 | } |
7072 | ||
e7355e50 | 7073 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 7074 | if (ret) { |
66642832 | 7075 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7076 | goto out; |
7077 | } | |
7078 | ||
0b246afa | 7079 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 7080 | if (ret) { |
66642832 | 7081 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7082 | goto out; |
7083 | } | |
a28ec197 | 7084 | } |
fcebe456 JB |
7085 | btrfs_release_path(path); |
7086 | ||
79787eaa | 7087 | out: |
5caf2a00 | 7088 | btrfs_free_path(path); |
a28ec197 CM |
7089 | return ret; |
7090 | } | |
7091 | ||
1887be66 | 7092 | /* |
f0486c68 | 7093 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7094 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7095 | * a given extent, and if there are no other delayed refs to be processed, it | |
7096 | * removes it from the tree. | |
7097 | */ | |
7098 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7099 | u64 bytenr) |
1887be66 CM |
7100 | { |
7101 | struct btrfs_delayed_ref_head *head; | |
7102 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7103 | int ret = 0; |
1887be66 CM |
7104 | |
7105 | delayed_refs = &trans->transaction->delayed_refs; | |
7106 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7107 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7108 | if (!head) |
cf93da7b | 7109 | goto out_delayed_unlock; |
1887be66 | 7110 | |
d7df2c79 | 7111 | spin_lock(&head->lock); |
0e0adbcf | 7112 | if (!RB_EMPTY_ROOT(&head->ref_tree)) |
1887be66 CM |
7113 | goto out; |
7114 | ||
5d4f98a2 YZ |
7115 | if (head->extent_op) { |
7116 | if (!head->must_insert_reserved) | |
7117 | goto out; | |
78a6184a | 7118 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7119 | head->extent_op = NULL; |
7120 | } | |
7121 | ||
1887be66 CM |
7122 | /* |
7123 | * waiting for the lock here would deadlock. If someone else has it | |
7124 | * locked they are already in the process of dropping it anyway | |
7125 | */ | |
7126 | if (!mutex_trylock(&head->mutex)) | |
7127 | goto out; | |
7128 | ||
7129 | /* | |
7130 | * at this point we have a head with no other entries. Go | |
7131 | * ahead and process it. | |
7132 | */ | |
c46effa6 | 7133 | rb_erase(&head->href_node, &delayed_refs->href_root); |
d278850e | 7134 | RB_CLEAR_NODE(&head->href_node); |
d7df2c79 | 7135 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7136 | |
7137 | /* | |
7138 | * we don't take a ref on the node because we're removing it from the | |
7139 | * tree, so we just steal the ref the tree was holding. | |
7140 | */ | |
c3e69d58 | 7141 | delayed_refs->num_heads--; |
d7df2c79 | 7142 | if (head->processing == 0) |
c3e69d58 | 7143 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7144 | head->processing = 0; |
7145 | spin_unlock(&head->lock); | |
1887be66 CM |
7146 | spin_unlock(&delayed_refs->lock); |
7147 | ||
f0486c68 YZ |
7148 | BUG_ON(head->extent_op); |
7149 | if (head->must_insert_reserved) | |
7150 | ret = 1; | |
7151 | ||
7152 | mutex_unlock(&head->mutex); | |
d278850e | 7153 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 7154 | return ret; |
1887be66 | 7155 | out: |
d7df2c79 | 7156 | spin_unlock(&head->lock); |
cf93da7b CM |
7157 | |
7158 | out_delayed_unlock: | |
1887be66 CM |
7159 | spin_unlock(&delayed_refs->lock); |
7160 | return 0; | |
7161 | } | |
7162 | ||
f0486c68 YZ |
7163 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7164 | struct btrfs_root *root, | |
7165 | struct extent_buffer *buf, | |
5581a51a | 7166 | u64 parent, int last_ref) |
f0486c68 | 7167 | { |
0b246afa | 7168 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7169 | int pin = 1; |
f0486c68 YZ |
7170 | int ret; |
7171 | ||
7172 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
d7eae340 OS |
7173 | int old_ref_mod, new_ref_mod; |
7174 | ||
fd708b81 JB |
7175 | btrfs_ref_tree_mod(root, buf->start, buf->len, parent, |
7176 | root->root_key.objectid, | |
7177 | btrfs_header_level(buf), 0, | |
7178 | BTRFS_DROP_DELAYED_REF); | |
7be07912 OS |
7179 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start, |
7180 | buf->len, parent, | |
0b246afa JM |
7181 | root->root_key.objectid, |
7182 | btrfs_header_level(buf), | |
7be07912 | 7183 | BTRFS_DROP_DELAYED_REF, NULL, |
d7eae340 | 7184 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7185 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7186 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7187 | } |
7188 | ||
0a16c7d7 | 7189 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7190 | struct btrfs_block_group_cache *cache; |
7191 | ||
f0486c68 | 7192 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7193 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7194 | if (!ret) |
37be25bc | 7195 | goto out; |
f0486c68 YZ |
7196 | } |
7197 | ||
4da8b76d | 7198 | pin = 0; |
0b246afa | 7199 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7200 | |
f0486c68 | 7201 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7202 | pin_down_extent(fs_info, cache, buf->start, |
7203 | buf->len, 1); | |
6219872d | 7204 | btrfs_put_block_group(cache); |
37be25bc | 7205 | goto out; |
f0486c68 YZ |
7206 | } |
7207 | ||
7208 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7209 | ||
7210 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7211 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7212 | btrfs_put_block_group(cache); |
71ff6437 | 7213 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7214 | } |
7215 | out: | |
b150a4f1 | 7216 | if (pin) |
29d2b84c | 7217 | add_pinned_bytes(fs_info, buf->len, true, |
b150a4f1 JB |
7218 | root->root_key.objectid); |
7219 | ||
0a16c7d7 OS |
7220 | if (last_ref) { |
7221 | /* | |
7222 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7223 | * matter anymore. | |
7224 | */ | |
7225 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7226 | } | |
f0486c68 YZ |
7227 | } |
7228 | ||
79787eaa | 7229 | /* Can return -ENOMEM */ |
2ff7e61e | 7230 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
84f7d8e6 | 7231 | struct btrfs_root *root, |
66d7e7f0 | 7232 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7233 | u64 owner, u64 offset) |
925baedd | 7234 | { |
84f7d8e6 | 7235 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 7236 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7237 | int ret; |
7238 | ||
f5ee5c9a | 7239 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7240 | return 0; |
fccb84c9 | 7241 | |
fd708b81 JB |
7242 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) |
7243 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, | |
7244 | root_objectid, owner, offset, | |
7245 | BTRFS_DROP_DELAYED_REF); | |
7246 | ||
56bec294 CM |
7247 | /* |
7248 | * tree log blocks never actually go into the extent allocation | |
7249 | * tree, just update pinning info and exit early. | |
56bec294 | 7250 | */ |
5d4f98a2 YZ |
7251 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7252 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7253 | /* unlocks the pinned mutex */ |
2ff7e61e | 7254 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
d7eae340 | 7255 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7256 | ret = 0; |
5d4f98a2 | 7257 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 7258 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
7259 | num_bytes, parent, |
7260 | root_objectid, (int)owner, | |
7261 | BTRFS_DROP_DELAYED_REF, NULL, | |
d7eae340 | 7262 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7263 | } else { |
66d7e7f0 | 7264 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
7265 | num_bytes, parent, |
7266 | root_objectid, owner, offset, | |
7267 | 0, BTRFS_DROP_DELAYED_REF, | |
d7eae340 | 7268 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7269 | } |
d7eae340 | 7270 | |
29d2b84c NB |
7271 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) { |
7272 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
7273 | ||
7274 | add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid); | |
7275 | } | |
d7eae340 | 7276 | |
925baedd CM |
7277 | return ret; |
7278 | } | |
7279 | ||
817d52f8 JB |
7280 | /* |
7281 | * when we wait for progress in the block group caching, its because | |
7282 | * our allocation attempt failed at least once. So, we must sleep | |
7283 | * and let some progress happen before we try again. | |
7284 | * | |
7285 | * This function will sleep at least once waiting for new free space to | |
7286 | * show up, and then it will check the block group free space numbers | |
7287 | * for our min num_bytes. Another option is to have it go ahead | |
7288 | * and look in the rbtree for a free extent of a given size, but this | |
7289 | * is a good start. | |
36cce922 JB |
7290 | * |
7291 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7292 | * any of the information in this block group. | |
817d52f8 | 7293 | */ |
36cce922 | 7294 | static noinline void |
817d52f8 JB |
7295 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7296 | u64 num_bytes) | |
7297 | { | |
11833d66 | 7298 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7299 | |
11833d66 YZ |
7300 | caching_ctl = get_caching_control(cache); |
7301 | if (!caching_ctl) | |
36cce922 | 7302 | return; |
817d52f8 | 7303 | |
11833d66 | 7304 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7305 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7306 | |
7307 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7308 | } |
7309 | ||
7310 | static noinline int | |
7311 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7312 | { | |
7313 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7314 | int ret = 0; |
11833d66 YZ |
7315 | |
7316 | caching_ctl = get_caching_control(cache); | |
7317 | if (!caching_ctl) | |
36cce922 | 7318 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7319 | |
7320 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7321 | if (cache->cached == BTRFS_CACHE_ERROR) |
7322 | ret = -EIO; | |
11833d66 | 7323 | put_caching_control(caching_ctl); |
36cce922 | 7324 | return ret; |
817d52f8 JB |
7325 | } |
7326 | ||
7327 | enum btrfs_loop_type { | |
285ff5af JB |
7328 | LOOP_CACHING_NOWAIT = 0, |
7329 | LOOP_CACHING_WAIT = 1, | |
7330 | LOOP_ALLOC_CHUNK = 2, | |
7331 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7332 | }; |
7333 | ||
e570fd27 MX |
7334 | static inline void |
7335 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7336 | int delalloc) | |
7337 | { | |
7338 | if (delalloc) | |
7339 | down_read(&cache->data_rwsem); | |
7340 | } | |
7341 | ||
7342 | static inline void | |
7343 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7344 | int delalloc) | |
7345 | { | |
7346 | btrfs_get_block_group(cache); | |
7347 | if (delalloc) | |
7348 | down_read(&cache->data_rwsem); | |
7349 | } | |
7350 | ||
7351 | static struct btrfs_block_group_cache * | |
7352 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7353 | struct btrfs_free_cluster *cluster, | |
7354 | int delalloc) | |
7355 | { | |
89771cc9 | 7356 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7357 | |
e570fd27 | 7358 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7359 | while (1) { |
7360 | used_bg = cluster->block_group; | |
7361 | if (!used_bg) | |
7362 | return NULL; | |
7363 | ||
7364 | if (used_bg == block_group) | |
e570fd27 MX |
7365 | return used_bg; |
7366 | ||
6719afdc | 7367 | btrfs_get_block_group(used_bg); |
e570fd27 | 7368 | |
6719afdc GU |
7369 | if (!delalloc) |
7370 | return used_bg; | |
e570fd27 | 7371 | |
6719afdc GU |
7372 | if (down_read_trylock(&used_bg->data_rwsem)) |
7373 | return used_bg; | |
e570fd27 | 7374 | |
6719afdc | 7375 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7376 | |
e321f8a8 LB |
7377 | /* We should only have one-level nested. */ |
7378 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7379 | |
6719afdc GU |
7380 | spin_lock(&cluster->refill_lock); |
7381 | if (used_bg == cluster->block_group) | |
7382 | return used_bg; | |
e570fd27 | 7383 | |
6719afdc GU |
7384 | up_read(&used_bg->data_rwsem); |
7385 | btrfs_put_block_group(used_bg); | |
7386 | } | |
e570fd27 MX |
7387 | } |
7388 | ||
7389 | static inline void | |
7390 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7391 | int delalloc) | |
7392 | { | |
7393 | if (delalloc) | |
7394 | up_read(&cache->data_rwsem); | |
7395 | btrfs_put_block_group(cache); | |
7396 | } | |
7397 | ||
fec577fb CM |
7398 | /* |
7399 | * walks the btree of allocated extents and find a hole of a given size. | |
7400 | * The key ins is changed to record the hole: | |
a4820398 | 7401 | * ins->objectid == start position |
62e2749e | 7402 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7403 | * ins->offset == the size of the hole. |
fec577fb | 7404 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7405 | * |
7406 | * If there is no suitable free space, we will record the max size of | |
7407 | * the free space extent currently. | |
fec577fb | 7408 | */ |
87bde3cd | 7409 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7410 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7411 | u64 hint_byte, struct btrfs_key *ins, | |
7412 | u64 flags, int delalloc) | |
fec577fb | 7413 | { |
80eb234a | 7414 | int ret = 0; |
0b246afa | 7415 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7416 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7417 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7418 | u64 search_start = 0; |
a4820398 | 7419 | u64 max_extent_size = 0; |
c759c4e1 | 7420 | u64 empty_cluster = 0; |
80eb234a | 7421 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7422 | int loop = 0; |
3e72ee88 | 7423 | int index = btrfs_bg_flags_to_raid_index(flags); |
0a24325e | 7424 | bool failed_cluster_refill = false; |
1cdda9b8 | 7425 | bool failed_alloc = false; |
67377734 | 7426 | bool use_cluster = true; |
60d2adbb | 7427 | bool have_caching_bg = false; |
13a0db5a | 7428 | bool orig_have_caching_bg = false; |
a5e681d9 | 7429 | bool full_search = false; |
fec577fb | 7430 | |
0b246afa | 7431 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7432 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7433 | ins->objectid = 0; |
7434 | ins->offset = 0; | |
b1a4d965 | 7435 | |
71ff6437 | 7436 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7437 | |
0b246afa | 7438 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7439 | if (!space_info) { |
0b246afa | 7440 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7441 | return -ENOSPC; |
7442 | } | |
2552d17e | 7443 | |
67377734 | 7444 | /* |
4f4db217 JB |
7445 | * If our free space is heavily fragmented we may not be able to make |
7446 | * big contiguous allocations, so instead of doing the expensive search | |
7447 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7448 | * can go ahead and search for a more manageable chunk. | |
7449 | * | |
7450 | * If our max_extent_size is large enough for our allocation simply | |
7451 | * disable clustering since we will likely not be able to find enough | |
7452 | * space to create a cluster and induce latency trying. | |
67377734 | 7453 | */ |
4f4db217 JB |
7454 | if (unlikely(space_info->max_extent_size)) { |
7455 | spin_lock(&space_info->lock); | |
7456 | if (space_info->max_extent_size && | |
7457 | num_bytes > space_info->max_extent_size) { | |
7458 | ins->offset = space_info->max_extent_size; | |
7459 | spin_unlock(&space_info->lock); | |
7460 | return -ENOSPC; | |
7461 | } else if (space_info->max_extent_size) { | |
7462 | use_cluster = false; | |
7463 | } | |
7464 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7465 | } |
0f9dd46c | 7466 | |
2ff7e61e | 7467 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7468 | if (last_ptr) { |
fa9c0d79 CM |
7469 | spin_lock(&last_ptr->lock); |
7470 | if (last_ptr->block_group) | |
7471 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7472 | if (last_ptr->fragmented) { |
7473 | /* | |
7474 | * We still set window_start so we can keep track of the | |
7475 | * last place we found an allocation to try and save | |
7476 | * some time. | |
7477 | */ | |
7478 | hint_byte = last_ptr->window_start; | |
7479 | use_cluster = false; | |
7480 | } | |
fa9c0d79 | 7481 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7482 | } |
fa9c0d79 | 7483 | |
2ff7e61e | 7484 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7485 | search_start = max(search_start, hint_byte); |
2552d17e | 7486 | if (search_start == hint_byte) { |
0b246afa | 7487 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7488 | /* |
7489 | * we don't want to use the block group if it doesn't match our | |
7490 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7491 | * |
7492 | * However if we are re-searching with an ideal block group | |
7493 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7494 | */ |
b6919a58 | 7495 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7496 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7497 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7498 | if (list_empty(&block_group->list) || |
7499 | block_group->ro) { | |
7500 | /* | |
7501 | * someone is removing this block group, | |
7502 | * we can't jump into the have_block_group | |
7503 | * target because our list pointers are not | |
7504 | * valid | |
7505 | */ | |
7506 | btrfs_put_block_group(block_group); | |
7507 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7508 | } else { |
3e72ee88 QW |
7509 | index = btrfs_bg_flags_to_raid_index( |
7510 | block_group->flags); | |
e570fd27 | 7511 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7512 | goto have_block_group; |
ccf0e725 | 7513 | } |
2552d17e | 7514 | } else if (block_group) { |
fa9c0d79 | 7515 | btrfs_put_block_group(block_group); |
2552d17e | 7516 | } |
42e70e7a | 7517 | } |
2552d17e | 7518 | search: |
60d2adbb | 7519 | have_caching_bg = false; |
3e72ee88 | 7520 | if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags)) |
a5e681d9 | 7521 | full_search = true; |
80eb234a | 7522 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7523 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7524 | list) { | |
6226cb0a | 7525 | u64 offset; |
817d52f8 | 7526 | int cached; |
8a1413a2 | 7527 | |
14443937 JM |
7528 | /* If the block group is read-only, we can skip it entirely. */ |
7529 | if (unlikely(block_group->ro)) | |
7530 | continue; | |
7531 | ||
e570fd27 | 7532 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7533 | search_start = block_group->key.objectid; |
42e70e7a | 7534 | |
83a50de9 CM |
7535 | /* |
7536 | * this can happen if we end up cycling through all the | |
7537 | * raid types, but we want to make sure we only allocate | |
7538 | * for the proper type. | |
7539 | */ | |
b6919a58 | 7540 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7541 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7542 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7543 | BTRFS_BLOCK_GROUP_RAID5 | |
7544 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7545 | BTRFS_BLOCK_GROUP_RAID10; |
7546 | ||
7547 | /* | |
7548 | * if they asked for extra copies and this block group | |
7549 | * doesn't provide them, bail. This does allow us to | |
7550 | * fill raid0 from raid1. | |
7551 | */ | |
b6919a58 | 7552 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7553 | goto loop; |
7554 | } | |
7555 | ||
2552d17e | 7556 | have_block_group: |
291c7d2f JB |
7557 | cached = block_group_cache_done(block_group); |
7558 | if (unlikely(!cached)) { | |
a5e681d9 | 7559 | have_caching_bg = true; |
f6373bf3 | 7560 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7561 | BUG_ON(ret < 0); |
7562 | ret = 0; | |
817d52f8 JB |
7563 | } |
7564 | ||
36cce922 JB |
7565 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7566 | goto loop; | |
0f9dd46c | 7567 | |
0a24325e | 7568 | /* |
062c05c4 AO |
7569 | * Ok we want to try and use the cluster allocator, so |
7570 | * lets look there | |
0a24325e | 7571 | */ |
c759c4e1 | 7572 | if (last_ptr && use_cluster) { |
215a63d1 | 7573 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7574 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7575 | /* |
7576 | * the refill lock keeps out other | |
7577 | * people trying to start a new cluster | |
7578 | */ | |
e570fd27 MX |
7579 | used_block_group = btrfs_lock_cluster(block_group, |
7580 | last_ptr, | |
7581 | delalloc); | |
7582 | if (!used_block_group) | |
44fb5511 | 7583 | goto refill_cluster; |
274bd4fb | 7584 | |
e570fd27 MX |
7585 | if (used_block_group != block_group && |
7586 | (used_block_group->ro || | |
7587 | !block_group_bits(used_block_group, flags))) | |
7588 | goto release_cluster; | |
44fb5511 | 7589 | |
274bd4fb | 7590 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7591 | last_ptr, |
7592 | num_bytes, | |
7593 | used_block_group->key.objectid, | |
7594 | &max_extent_size); | |
fa9c0d79 CM |
7595 | if (offset) { |
7596 | /* we have a block, we're done */ | |
7597 | spin_unlock(&last_ptr->refill_lock); | |
3dca5c94 | 7598 | trace_btrfs_reserve_extent_cluster( |
89d4346a MX |
7599 | used_block_group, |
7600 | search_start, num_bytes); | |
215a63d1 | 7601 | if (used_block_group != block_group) { |
e570fd27 MX |
7602 | btrfs_release_block_group(block_group, |
7603 | delalloc); | |
215a63d1 MX |
7604 | block_group = used_block_group; |
7605 | } | |
fa9c0d79 CM |
7606 | goto checks; |
7607 | } | |
7608 | ||
274bd4fb | 7609 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7610 | release_cluster: |
062c05c4 AO |
7611 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7612 | * set up a new clusters, so lets just skip it | |
7613 | * and let the allocator find whatever block | |
7614 | * it can find. If we reach this point, we | |
7615 | * will have tried the cluster allocator | |
7616 | * plenty of times and not have found | |
7617 | * anything, so we are likely way too | |
7618 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7619 | * anything. |
7620 | * | |
7621 | * However, if the cluster is taken from the | |
7622 | * current block group, release the cluster | |
7623 | * first, so that we stand a better chance of | |
7624 | * succeeding in the unclustered | |
7625 | * allocation. */ | |
7626 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7627 | used_block_group != block_group) { |
062c05c4 | 7628 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7629 | btrfs_release_block_group(used_block_group, |
7630 | delalloc); | |
062c05c4 AO |
7631 | goto unclustered_alloc; |
7632 | } | |
7633 | ||
fa9c0d79 CM |
7634 | /* |
7635 | * this cluster didn't work out, free it and | |
7636 | * start over | |
7637 | */ | |
7638 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7639 | ||
e570fd27 MX |
7640 | if (used_block_group != block_group) |
7641 | btrfs_release_block_group(used_block_group, | |
7642 | delalloc); | |
7643 | refill_cluster: | |
a5f6f719 AO |
7644 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7645 | spin_unlock(&last_ptr->refill_lock); | |
7646 | goto unclustered_alloc; | |
7647 | } | |
7648 | ||
8de972b4 CM |
7649 | aligned_cluster = max_t(unsigned long, |
7650 | empty_cluster + empty_size, | |
7651 | block_group->full_stripe_len); | |
7652 | ||
fa9c0d79 | 7653 | /* allocate a cluster in this block group */ |
2ff7e61e | 7654 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7655 | last_ptr, search_start, |
7656 | num_bytes, | |
7657 | aligned_cluster); | |
fa9c0d79 CM |
7658 | if (ret == 0) { |
7659 | /* | |
7660 | * now pull our allocation out of this | |
7661 | * cluster | |
7662 | */ | |
7663 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7664 | last_ptr, |
7665 | num_bytes, | |
7666 | search_start, | |
7667 | &max_extent_size); | |
fa9c0d79 CM |
7668 | if (offset) { |
7669 | /* we found one, proceed */ | |
7670 | spin_unlock(&last_ptr->refill_lock); | |
3dca5c94 | 7671 | trace_btrfs_reserve_extent_cluster( |
3f7de037 JB |
7672 | block_group, search_start, |
7673 | num_bytes); | |
fa9c0d79 CM |
7674 | goto checks; |
7675 | } | |
0a24325e JB |
7676 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7677 | && !failed_cluster_refill) { | |
817d52f8 JB |
7678 | spin_unlock(&last_ptr->refill_lock); |
7679 | ||
0a24325e | 7680 | failed_cluster_refill = true; |
817d52f8 JB |
7681 | wait_block_group_cache_progress(block_group, |
7682 | num_bytes + empty_cluster + empty_size); | |
7683 | goto have_block_group; | |
fa9c0d79 | 7684 | } |
817d52f8 | 7685 | |
fa9c0d79 CM |
7686 | /* |
7687 | * at this point we either didn't find a cluster | |
7688 | * or we weren't able to allocate a block from our | |
7689 | * cluster. Free the cluster we've been trying | |
7690 | * to use, and go to the next block group | |
7691 | */ | |
0a24325e | 7692 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7693 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7694 | goto loop; |
fa9c0d79 CM |
7695 | } |
7696 | ||
062c05c4 | 7697 | unclustered_alloc: |
c759c4e1 JB |
7698 | /* |
7699 | * We are doing an unclustered alloc, set the fragmented flag so | |
7700 | * we don't bother trying to setup a cluster again until we get | |
7701 | * more space. | |
7702 | */ | |
7703 | if (unlikely(last_ptr)) { | |
7704 | spin_lock(&last_ptr->lock); | |
7705 | last_ptr->fragmented = 1; | |
7706 | spin_unlock(&last_ptr->lock); | |
7707 | } | |
0c9b36e0 LB |
7708 | if (cached) { |
7709 | struct btrfs_free_space_ctl *ctl = | |
7710 | block_group->free_space_ctl; | |
7711 | ||
7712 | spin_lock(&ctl->tree_lock); | |
7713 | if (ctl->free_space < | |
7714 | num_bytes + empty_cluster + empty_size) { | |
7715 | if (ctl->free_space > max_extent_size) | |
7716 | max_extent_size = ctl->free_space; | |
7717 | spin_unlock(&ctl->tree_lock); | |
7718 | goto loop; | |
7719 | } | |
7720 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7721 | } |
a5f6f719 | 7722 | |
6226cb0a | 7723 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7724 | num_bytes, empty_size, |
7725 | &max_extent_size); | |
1cdda9b8 JB |
7726 | /* |
7727 | * If we didn't find a chunk, and we haven't failed on this | |
7728 | * block group before, and this block group is in the middle of | |
7729 | * caching and we are ok with waiting, then go ahead and wait | |
7730 | * for progress to be made, and set failed_alloc to true. | |
7731 | * | |
7732 | * If failed_alloc is true then we've already waited on this | |
7733 | * block group once and should move on to the next block group. | |
7734 | */ | |
7735 | if (!offset && !failed_alloc && !cached && | |
7736 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7737 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7738 | num_bytes + empty_size); |
7739 | failed_alloc = true; | |
817d52f8 | 7740 | goto have_block_group; |
1cdda9b8 JB |
7741 | } else if (!offset) { |
7742 | goto loop; | |
817d52f8 | 7743 | } |
fa9c0d79 | 7744 | checks: |
0b246afa | 7745 | search_start = ALIGN(offset, fs_info->stripesize); |
25179201 | 7746 | |
2552d17e JB |
7747 | /* move on to the next group */ |
7748 | if (search_start + num_bytes > | |
215a63d1 MX |
7749 | block_group->key.objectid + block_group->key.offset) { |
7750 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7751 | goto loop; |
6226cb0a | 7752 | } |
f5a31e16 | 7753 | |
f0486c68 | 7754 | if (offset < search_start) |
215a63d1 | 7755 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7756 | search_start - offset); |
7757 | BUG_ON(offset > search_start); | |
2552d17e | 7758 | |
18513091 WX |
7759 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7760 | num_bytes, delalloc); | |
f0486c68 | 7761 | if (ret == -EAGAIN) { |
215a63d1 | 7762 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7763 | goto loop; |
0f9dd46c | 7764 | } |
9cfa3e34 | 7765 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7766 | |
f0486c68 | 7767 | /* we are all good, lets return */ |
2552d17e JB |
7768 | ins->objectid = search_start; |
7769 | ins->offset = num_bytes; | |
d2fb3437 | 7770 | |
3dca5c94 | 7771 | trace_btrfs_reserve_extent(block_group, search_start, num_bytes); |
e570fd27 | 7772 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7773 | break; |
7774 | loop: | |
0a24325e | 7775 | failed_cluster_refill = false; |
1cdda9b8 | 7776 | failed_alloc = false; |
3e72ee88 QW |
7777 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
7778 | index); | |
e570fd27 | 7779 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7780 | cond_resched(); |
2552d17e JB |
7781 | } |
7782 | up_read(&space_info->groups_sem); | |
7783 | ||
13a0db5a | 7784 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7785 | && !orig_have_caching_bg) | |
7786 | orig_have_caching_bg = true; | |
7787 | ||
60d2adbb MX |
7788 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7789 | goto search; | |
7790 | ||
b742bb82 YZ |
7791 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7792 | goto search; | |
7793 | ||
285ff5af | 7794 | /* |
ccf0e725 JB |
7795 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7796 | * caching kthreads as we move along | |
817d52f8 JB |
7797 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7798 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7799 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7800 | * again | |
fa9c0d79 | 7801 | */ |
723bda20 | 7802 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7803 | index = 0; |
a5e681d9 JB |
7804 | if (loop == LOOP_CACHING_NOWAIT) { |
7805 | /* | |
7806 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7807 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7808 | * full search through. |
7809 | */ | |
13a0db5a | 7810 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7811 | loop = LOOP_CACHING_WAIT; |
7812 | else | |
7813 | loop = LOOP_ALLOC_CHUNK; | |
7814 | } else { | |
7815 | loop++; | |
7816 | } | |
7817 | ||
817d52f8 | 7818 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7819 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7820 | int exist = 0; |
7821 | ||
7822 | trans = current->journal_info; | |
7823 | if (trans) | |
7824 | exist = 1; | |
7825 | else | |
7826 | trans = btrfs_join_transaction(root); | |
00361589 | 7827 | |
00361589 JB |
7828 | if (IS_ERR(trans)) { |
7829 | ret = PTR_ERR(trans); | |
7830 | goto out; | |
7831 | } | |
7832 | ||
2ff7e61e | 7833 | ret = do_chunk_alloc(trans, fs_info, flags, |
ea658bad | 7834 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7835 | |
7836 | /* | |
7837 | * If we can't allocate a new chunk we've already looped | |
7838 | * through at least once, move on to the NO_EMPTY_SIZE | |
7839 | * case. | |
7840 | */ | |
7841 | if (ret == -ENOSPC) | |
7842 | loop = LOOP_NO_EMPTY_SIZE; | |
7843 | ||
ea658bad JB |
7844 | /* |
7845 | * Do not bail out on ENOSPC since we | |
7846 | * can do more things. | |
7847 | */ | |
00361589 | 7848 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7849 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7850 | else |
7851 | ret = 0; | |
f017f15f | 7852 | if (!exist) |
3a45bb20 | 7853 | btrfs_end_transaction(trans); |
00361589 | 7854 | if (ret) |
ea658bad | 7855 | goto out; |
2552d17e JB |
7856 | } |
7857 | ||
723bda20 | 7858 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7859 | /* |
7860 | * Don't loop again if we already have no empty_size and | |
7861 | * no empty_cluster. | |
7862 | */ | |
7863 | if (empty_size == 0 && | |
7864 | empty_cluster == 0) { | |
7865 | ret = -ENOSPC; | |
7866 | goto out; | |
7867 | } | |
723bda20 JB |
7868 | empty_size = 0; |
7869 | empty_cluster = 0; | |
fa9c0d79 | 7870 | } |
723bda20 JB |
7871 | |
7872 | goto search; | |
2552d17e JB |
7873 | } else if (!ins->objectid) { |
7874 | ret = -ENOSPC; | |
d82a6f1d | 7875 | } else if (ins->objectid) { |
c759c4e1 JB |
7876 | if (!use_cluster && last_ptr) { |
7877 | spin_lock(&last_ptr->lock); | |
7878 | last_ptr->window_start = ins->objectid; | |
7879 | spin_unlock(&last_ptr->lock); | |
7880 | } | |
80eb234a | 7881 | ret = 0; |
be744175 | 7882 | } |
79787eaa | 7883 | out: |
4f4db217 JB |
7884 | if (ret == -ENOSPC) { |
7885 | spin_lock(&space_info->lock); | |
7886 | space_info->max_extent_size = max_extent_size; | |
7887 | spin_unlock(&space_info->lock); | |
a4820398 | 7888 | ins->offset = max_extent_size; |
4f4db217 | 7889 | } |
0f70abe2 | 7890 | return ret; |
fec577fb | 7891 | } |
ec44a35c | 7892 | |
ab8d0fc4 JM |
7893 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7894 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7895 | int dump_block_groups) |
0f9dd46c JB |
7896 | { |
7897 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7898 | int index = 0; |
0f9dd46c | 7899 | |
9ed74f2d | 7900 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7901 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7902 | info->flags, | |
4136135b LB |
7903 | info->total_bytes - btrfs_space_info_used(info, true), |
7904 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7905 | btrfs_info(fs_info, |
7906 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7907 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7908 | info->bytes_reserved, info->bytes_may_use, | |
7909 | info->bytes_readonly); | |
9ed74f2d JB |
7910 | spin_unlock(&info->lock); |
7911 | ||
7912 | if (!dump_block_groups) | |
7913 | return; | |
0f9dd46c | 7914 | |
80eb234a | 7915 | down_read(&info->groups_sem); |
b742bb82 YZ |
7916 | again: |
7917 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7918 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7919 | btrfs_info(fs_info, |
7920 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7921 | cache->key.objectid, cache->key.offset, | |
7922 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7923 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7924 | btrfs_dump_free_space(cache, bytes); |
7925 | spin_unlock(&cache->lock); | |
7926 | } | |
b742bb82 YZ |
7927 | if (++index < BTRFS_NR_RAID_TYPES) |
7928 | goto again; | |
80eb234a | 7929 | up_read(&info->groups_sem); |
0f9dd46c | 7930 | } |
e8569813 | 7931 | |
6f47c706 NB |
7932 | /* |
7933 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
7934 | * hole that is at least as big as @num_bytes. | |
7935 | * | |
7936 | * @root - The root that will contain this extent | |
7937 | * | |
7938 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
7939 | * is used for accounting purposes. This value differs | |
7940 | * from @num_bytes only in the case of compressed extents. | |
7941 | * | |
7942 | * @num_bytes - Number of bytes to allocate on-disk. | |
7943 | * | |
7944 | * @min_alloc_size - Indicates the minimum amount of space that the | |
7945 | * allocator should try to satisfy. In some cases | |
7946 | * @num_bytes may be larger than what is required and if | |
7947 | * the filesystem is fragmented then allocation fails. | |
7948 | * However, the presence of @min_alloc_size gives a | |
7949 | * chance to try and satisfy the smaller allocation. | |
7950 | * | |
7951 | * @empty_size - A hint that you plan on doing more COW. This is the | |
7952 | * size in bytes the allocator should try to find free | |
7953 | * next to the block it returns. This is just a hint and | |
7954 | * may be ignored by the allocator. | |
7955 | * | |
7956 | * @hint_byte - Hint to the allocator to start searching above the byte | |
7957 | * address passed. It might be ignored. | |
7958 | * | |
7959 | * @ins - This key is modified to record the found hole. It will | |
7960 | * have the following values: | |
7961 | * ins->objectid == start position | |
7962 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
7963 | * ins->offset == the size of the hole. | |
7964 | * | |
7965 | * @is_data - Boolean flag indicating whether an extent is | |
7966 | * allocated for data (true) or metadata (false) | |
7967 | * | |
7968 | * @delalloc - Boolean flag indicating whether this allocation is for | |
7969 | * delalloc or not. If 'true' data_rwsem of block groups | |
7970 | * is going to be acquired. | |
7971 | * | |
7972 | * | |
7973 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
7974 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
7975 | * largest available hole the allocator managed to find. | |
7976 | */ | |
18513091 | 7977 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
7978 | u64 num_bytes, u64 min_alloc_size, |
7979 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7980 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7981 | { |
ab8d0fc4 | 7982 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 7983 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7984 | u64 flags; |
fec577fb | 7985 | int ret; |
925baedd | 7986 | |
1b86826d | 7987 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 7988 | again: |
0b246afa | 7989 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 7990 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 7991 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 7992 | if (!ret && !is_data) { |
ab8d0fc4 | 7993 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 7994 | } else if (ret == -ENOSPC) { |
a4820398 MX |
7995 | if (!final_tried && ins->offset) { |
7996 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 7997 | num_bytes = round_down(num_bytes, |
0b246afa | 7998 | fs_info->sectorsize); |
9e622d6b | 7999 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 8000 | ram_bytes = num_bytes; |
9e622d6b MX |
8001 | if (num_bytes == min_alloc_size) |
8002 | final_tried = true; | |
8003 | goto again; | |
ab8d0fc4 | 8004 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
8005 | struct btrfs_space_info *sinfo; |
8006 | ||
ab8d0fc4 | 8007 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8008 | btrfs_err(fs_info, |
5d163e0e JM |
8009 | "allocation failed flags %llu, wanted %llu", |
8010 | flags, num_bytes); | |
53804280 | 8011 | if (sinfo) |
ab8d0fc4 | 8012 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8013 | } |
925baedd | 8014 | } |
0f9dd46c JB |
8015 | |
8016 | return ret; | |
e6dcd2dc CM |
8017 | } |
8018 | ||
2ff7e61e | 8019 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8020 | u64 start, u64 len, |
8021 | int pin, int delalloc) | |
65b51a00 | 8022 | { |
0f9dd46c | 8023 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8024 | int ret = 0; |
0f9dd46c | 8025 | |
0b246afa | 8026 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8027 | if (!cache) { |
0b246afa JM |
8028 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8029 | start); | |
0f9dd46c JB |
8030 | return -ENOSPC; |
8031 | } | |
1f3c79a2 | 8032 | |
e688b725 | 8033 | if (pin) |
2ff7e61e | 8034 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 8035 | else { |
0b246afa | 8036 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8037 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8038 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8039 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8040 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8041 | } |
31193213 | 8042 | |
fa9c0d79 | 8043 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8044 | return ret; |
8045 | } | |
8046 | ||
2ff7e61e | 8047 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8048 | u64 start, u64 len, int delalloc) |
e688b725 | 8049 | { |
2ff7e61e | 8050 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8051 | } |
8052 | ||
2ff7e61e | 8053 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8054 | u64 start, u64 len) |
8055 | { | |
2ff7e61e | 8056 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8057 | } |
8058 | ||
5d4f98a2 | 8059 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8060 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8061 | u64 parent, u64 root_objectid, |
8062 | u64 flags, u64 owner, u64 offset, | |
8063 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
8064 | { |
8065 | int ret; | |
e6dcd2dc | 8066 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8067 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8068 | struct btrfs_path *path; |
5d4f98a2 YZ |
8069 | struct extent_buffer *leaf; |
8070 | int type; | |
8071 | u32 size; | |
26b8003f | 8072 | |
5d4f98a2 YZ |
8073 | if (parent > 0) |
8074 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8075 | else | |
8076 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8077 | |
5d4f98a2 | 8078 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8079 | |
8080 | path = btrfs_alloc_path(); | |
db5b493a TI |
8081 | if (!path) |
8082 | return -ENOMEM; | |
47e4bb98 | 8083 | |
b9473439 | 8084 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8085 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8086 | ins, size); | |
79787eaa JM |
8087 | if (ret) { |
8088 | btrfs_free_path(path); | |
8089 | return ret; | |
8090 | } | |
0f9dd46c | 8091 | |
5d4f98a2 YZ |
8092 | leaf = path->nodes[0]; |
8093 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8094 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8095 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8096 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8097 | btrfs_set_extent_flags(leaf, extent_item, | |
8098 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8099 | ||
8100 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8101 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8102 | if (parent > 0) { | |
8103 | struct btrfs_shared_data_ref *ref; | |
8104 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8105 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8106 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8107 | } else { | |
8108 | struct btrfs_extent_data_ref *ref; | |
8109 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8110 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8111 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8112 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8113 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8114 | } | |
47e4bb98 CM |
8115 | |
8116 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8117 | btrfs_free_path(path); |
f510cfec | 8118 | |
25a356d3 | 8119 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
8120 | if (ret) |
8121 | return ret; | |
8122 | ||
6202df69 | 8123 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 8124 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8125 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8126 | ins->objectid, ins->offset); |
f5947066 CM |
8127 | BUG(); |
8128 | } | |
71ff6437 | 8129 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8130 | return ret; |
8131 | } | |
8132 | ||
5d4f98a2 | 8133 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 NB |
8134 | struct btrfs_delayed_ref_node *node, |
8135 | u64 flags, struct btrfs_disk_key *key) | |
e6dcd2dc | 8136 | { |
9dcdbe01 | 8137 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 8138 | int ret; |
5d4f98a2 | 8139 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 8140 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
8141 | struct btrfs_tree_block_info *block_info; |
8142 | struct btrfs_extent_inline_ref *iref; | |
8143 | struct btrfs_path *path; | |
8144 | struct extent_buffer *leaf; | |
4e6bd4e0 | 8145 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 8146 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 NB |
8147 | u64 num_bytes; |
8148 | u64 parent; | |
0b246afa | 8149 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 8150 | |
4e6bd4e0 NB |
8151 | ref = btrfs_delayed_node_to_tree_ref(node); |
8152 | ||
8153 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
8154 | parent = ref->parent; | |
8155 | else | |
8156 | parent = 0; | |
8157 | ||
8158 | extent_key.objectid = node->bytenr; | |
8159 | if (skinny_metadata) { | |
8160 | extent_key.offset = ref->level; | |
8161 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
8162 | num_bytes = fs_info->nodesize; | |
8163 | } else { | |
8164 | extent_key.offset = node->num_bytes; | |
8165 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 8166 | size += sizeof(*block_info); |
4e6bd4e0 NB |
8167 | num_bytes = node->num_bytes; |
8168 | } | |
1c2308f8 | 8169 | |
5d4f98a2 | 8170 | path = btrfs_alloc_path(); |
857cc2fc | 8171 | if (!path) { |
4e6bd4e0 NB |
8172 | btrfs_free_and_pin_reserved_extent(fs_info, |
8173 | extent_key.objectid, | |
0b246afa | 8174 | fs_info->nodesize); |
d8926bb3 | 8175 | return -ENOMEM; |
857cc2fc | 8176 | } |
56bec294 | 8177 | |
5d4f98a2 YZ |
8178 | path->leave_spinning = 1; |
8179 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 8180 | &extent_key, size); |
79787eaa | 8181 | if (ret) { |
dd825259 | 8182 | btrfs_free_path(path); |
4e6bd4e0 NB |
8183 | btrfs_free_and_pin_reserved_extent(fs_info, |
8184 | extent_key.objectid, | |
0b246afa | 8185 | fs_info->nodesize); |
79787eaa JM |
8186 | return ret; |
8187 | } | |
5d4f98a2 YZ |
8188 | |
8189 | leaf = path->nodes[0]; | |
8190 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8191 | struct btrfs_extent_item); | |
8192 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8193 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8194 | btrfs_set_extent_flags(leaf, extent_item, | |
8195 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8196 | |
3173a18f JB |
8197 | if (skinny_metadata) { |
8198 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8199 | } else { | |
8200 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
8201 | btrfs_set_tree_block_key(leaf, block_info, key); | |
4e6bd4e0 | 8202 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
8203 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
8204 | } | |
5d4f98a2 | 8205 | |
5d4f98a2 YZ |
8206 | if (parent > 0) { |
8207 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
8208 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8209 | BTRFS_SHARED_BLOCK_REF_KEY); | |
8210 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8211 | } else { | |
8212 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8213 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 8214 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
8215 | } |
8216 | ||
8217 | btrfs_mark_buffer_dirty(leaf); | |
8218 | btrfs_free_path(path); | |
8219 | ||
4e6bd4e0 NB |
8220 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
8221 | num_bytes); | |
1e144fb8 OS |
8222 | if (ret) |
8223 | return ret; | |
8224 | ||
4e6bd4e0 | 8225 | ret = update_block_group(trans, fs_info, extent_key.objectid, |
6202df69 | 8226 | fs_info->nodesize, 1); |
79787eaa | 8227 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8228 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 8229 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
8230 | BUG(); |
8231 | } | |
0be5dc67 | 8232 | |
4e6bd4e0 | 8233 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 8234 | fs_info->nodesize); |
5d4f98a2 YZ |
8235 | return ret; |
8236 | } | |
8237 | ||
8238 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 8239 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
8240 | u64 offset, u64 ram_bytes, |
8241 | struct btrfs_key *ins) | |
5d4f98a2 | 8242 | { |
84f7d8e6 | 8243 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
8244 | int ret; |
8245 | ||
84f7d8e6 | 8246 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8247 | |
fd708b81 JB |
8248 | btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0, |
8249 | root->root_key.objectid, owner, offset, | |
8250 | BTRFS_ADD_DELAYED_EXTENT); | |
8251 | ||
0b246afa | 8252 | ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid, |
84f7d8e6 JB |
8253 | ins->offset, 0, |
8254 | root->root_key.objectid, owner, | |
7be07912 OS |
8255 | offset, ram_bytes, |
8256 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); | |
e6dcd2dc CM |
8257 | return ret; |
8258 | } | |
e02119d5 CM |
8259 | |
8260 | /* | |
8261 | * this is used by the tree logging recovery code. It records that | |
8262 | * an extent has been allocated and makes sure to clear the free | |
8263 | * space cache bits as well | |
8264 | */ | |
5d4f98a2 | 8265 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8266 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8267 | u64 root_objectid, u64 owner, u64 offset, |
8268 | struct btrfs_key *ins) | |
e02119d5 CM |
8269 | { |
8270 | int ret; | |
8271 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8272 | struct btrfs_space_info *space_info; |
11833d66 | 8273 | |
8c2a1a30 JB |
8274 | /* |
8275 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8276 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8277 | */ |
0b246afa | 8278 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8279 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8280 | ins->offset); | |
b50c6e25 | 8281 | if (ret) |
8c2a1a30 | 8282 | return ret; |
11833d66 YZ |
8283 | } |
8284 | ||
0b246afa | 8285 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8286 | if (!block_group) |
8287 | return -EINVAL; | |
8288 | ||
ed7a6948 WX |
8289 | space_info = block_group->space_info; |
8290 | spin_lock(&space_info->lock); | |
8291 | spin_lock(&block_group->lock); | |
8292 | space_info->bytes_reserved += ins->offset; | |
8293 | block_group->reserved += ins->offset; | |
8294 | spin_unlock(&block_group->lock); | |
8295 | spin_unlock(&space_info->lock); | |
8296 | ||
2ff7e61e | 8297 | ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid, |
5d4f98a2 | 8298 | 0, owner, offset, ins, 1); |
b50c6e25 | 8299 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8300 | return ret; |
8301 | } | |
8302 | ||
48a3b636 ES |
8303 | static struct extent_buffer * |
8304 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8305 | u64 bytenr, int level) |
65b51a00 | 8306 | { |
0b246afa | 8307 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8308 | struct extent_buffer *buf; |
8309 | ||
2ff7e61e | 8310 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8311 | if (IS_ERR(buf)) |
8312 | return buf; | |
8313 | ||
65b51a00 | 8314 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8315 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8316 | btrfs_tree_lock(buf); |
7c302b49 | 8317 | clean_tree_block(fs_info, buf); |
3083ee2e | 8318 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8319 | |
8320 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8321 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8322 | |
d0c803c4 | 8323 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8324 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8325 | /* |
8326 | * we allow two log transactions at a time, use different | |
8327 | * EXENT bit to differentiate dirty pages. | |
8328 | */ | |
656f30db | 8329 | if (buf->log_index == 0) |
8cef4e16 YZ |
8330 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8331 | buf->start + buf->len - 1, GFP_NOFS); | |
8332 | else | |
8333 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8334 | buf->start + buf->len - 1); |
d0c803c4 | 8335 | } else { |
656f30db | 8336 | buf->log_index = -1; |
d0c803c4 | 8337 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8338 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8339 | } |
64c12921 | 8340 | trans->dirty = true; |
b4ce94de | 8341 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8342 | return buf; |
8343 | } | |
8344 | ||
f0486c68 YZ |
8345 | static struct btrfs_block_rsv * |
8346 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8347 | struct btrfs_root *root, u32 blocksize) | |
8348 | { | |
0b246afa | 8349 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8350 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8351 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8352 | int ret; |
d88033db | 8353 | bool global_updated = false; |
f0486c68 YZ |
8354 | |
8355 | block_rsv = get_block_rsv(trans, root); | |
8356 | ||
b586b323 MX |
8357 | if (unlikely(block_rsv->size == 0)) |
8358 | goto try_reserve; | |
d88033db | 8359 | again: |
f0486c68 YZ |
8360 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8361 | if (!ret) | |
8362 | return block_rsv; | |
8363 | ||
b586b323 MX |
8364 | if (block_rsv->failfast) |
8365 | return ERR_PTR(ret); | |
8366 | ||
d88033db MX |
8367 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8368 | global_updated = true; | |
0b246afa | 8369 | update_global_block_rsv(fs_info); |
d88033db MX |
8370 | goto again; |
8371 | } | |
8372 | ||
0b246afa | 8373 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8374 | static DEFINE_RATELIMIT_STATE(_rs, |
8375 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8376 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8377 | if (__ratelimit(&_rs)) | |
8378 | WARN(1, KERN_DEBUG | |
efe120a0 | 8379 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8380 | } |
8381 | try_reserve: | |
8382 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8383 | BTRFS_RESERVE_NO_FLUSH); | |
8384 | if (!ret) | |
8385 | return block_rsv; | |
8386 | /* | |
8387 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8388 | * the global reserve if its space type is the same as the global |
8389 | * reservation. | |
b586b323 | 8390 | */ |
5881cfc9 MX |
8391 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8392 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8393 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8394 | if (!ret) | |
8395 | return global_rsv; | |
8396 | } | |
8397 | return ERR_PTR(ret); | |
f0486c68 YZ |
8398 | } |
8399 | ||
8c2a3ca2 JB |
8400 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8401 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8402 | { |
8403 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
ff6bc37e | 8404 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL); |
f0486c68 YZ |
8405 | } |
8406 | ||
fec577fb | 8407 | /* |
f0486c68 | 8408 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8409 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8410 | */ |
4d75f8a9 | 8411 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8412 | struct btrfs_root *root, |
8413 | u64 parent, u64 root_objectid, | |
8414 | const struct btrfs_disk_key *key, | |
8415 | int level, u64 hint, | |
8416 | u64 empty_size) | |
fec577fb | 8417 | { |
0b246afa | 8418 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8419 | struct btrfs_key ins; |
f0486c68 | 8420 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8421 | struct extent_buffer *buf; |
67b7859e | 8422 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8423 | u64 flags = 0; |
8424 | int ret; | |
0b246afa JM |
8425 | u32 blocksize = fs_info->nodesize; |
8426 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8427 | |
05653ef3 | 8428 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8429 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8430 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8431 | level); |
faa2dbf0 JB |
8432 | if (!IS_ERR(buf)) |
8433 | root->alloc_bytenr += blocksize; | |
8434 | return buf; | |
8435 | } | |
05653ef3 | 8436 | #endif |
fccb84c9 | 8437 | |
f0486c68 YZ |
8438 | block_rsv = use_block_rsv(trans, root, blocksize); |
8439 | if (IS_ERR(block_rsv)) | |
8440 | return ERR_CAST(block_rsv); | |
8441 | ||
18513091 | 8442 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8443 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8444 | if (ret) |
8445 | goto out_unuse; | |
55c69072 | 8446 | |
fe864576 | 8447 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8448 | if (IS_ERR(buf)) { |
8449 | ret = PTR_ERR(buf); | |
8450 | goto out_free_reserved; | |
8451 | } | |
f0486c68 YZ |
8452 | |
8453 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8454 | if (parent == 0) | |
8455 | parent = ins.objectid; | |
8456 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8457 | } else | |
8458 | BUG_ON(parent > 0); | |
8459 | ||
8460 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8461 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8462 | if (!extent_op) { |
8463 | ret = -ENOMEM; | |
8464 | goto out_free_buf; | |
8465 | } | |
f0486c68 YZ |
8466 | if (key) |
8467 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8468 | else | |
8469 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8470 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8471 | extent_op->update_key = skinny_metadata ? false : true; |
8472 | extent_op->update_flags = true; | |
8473 | extent_op->is_data = false; | |
b1c79e09 | 8474 | extent_op->level = level; |
f0486c68 | 8475 | |
fd708b81 JB |
8476 | btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent, |
8477 | root_objectid, level, 0, | |
8478 | BTRFS_ADD_DELAYED_EXTENT); | |
7be07912 OS |
8479 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid, |
8480 | ins.offset, parent, | |
8481 | root_objectid, level, | |
67b7859e | 8482 | BTRFS_ADD_DELAYED_EXTENT, |
7be07912 | 8483 | extent_op, NULL, NULL); |
67b7859e OS |
8484 | if (ret) |
8485 | goto out_free_delayed; | |
f0486c68 | 8486 | } |
fec577fb | 8487 | return buf; |
67b7859e OS |
8488 | |
8489 | out_free_delayed: | |
8490 | btrfs_free_delayed_extent_op(extent_op); | |
8491 | out_free_buf: | |
8492 | free_extent_buffer(buf); | |
8493 | out_free_reserved: | |
2ff7e61e | 8494 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8495 | out_unuse: |
0b246afa | 8496 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8497 | return ERR_PTR(ret); |
fec577fb | 8498 | } |
a28ec197 | 8499 | |
2c47e605 YZ |
8500 | struct walk_control { |
8501 | u64 refs[BTRFS_MAX_LEVEL]; | |
8502 | u64 flags[BTRFS_MAX_LEVEL]; | |
8503 | struct btrfs_key update_progress; | |
8504 | int stage; | |
8505 | int level; | |
8506 | int shared_level; | |
8507 | int update_ref; | |
8508 | int keep_locks; | |
1c4850e2 YZ |
8509 | int reada_slot; |
8510 | int reada_count; | |
66d7e7f0 | 8511 | int for_reloc; |
2c47e605 YZ |
8512 | }; |
8513 | ||
8514 | #define DROP_REFERENCE 1 | |
8515 | #define UPDATE_BACKREF 2 | |
8516 | ||
1c4850e2 YZ |
8517 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8518 | struct btrfs_root *root, | |
8519 | struct walk_control *wc, | |
8520 | struct btrfs_path *path) | |
6407bf6d | 8521 | { |
0b246afa | 8522 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8523 | u64 bytenr; |
8524 | u64 generation; | |
8525 | u64 refs; | |
94fcca9f | 8526 | u64 flags; |
5d4f98a2 | 8527 | u32 nritems; |
1c4850e2 YZ |
8528 | struct btrfs_key key; |
8529 | struct extent_buffer *eb; | |
6407bf6d | 8530 | int ret; |
1c4850e2 YZ |
8531 | int slot; |
8532 | int nread = 0; | |
6407bf6d | 8533 | |
1c4850e2 YZ |
8534 | if (path->slots[wc->level] < wc->reada_slot) { |
8535 | wc->reada_count = wc->reada_count * 2 / 3; | |
8536 | wc->reada_count = max(wc->reada_count, 2); | |
8537 | } else { | |
8538 | wc->reada_count = wc->reada_count * 3 / 2; | |
8539 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8540 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8541 | } |
7bb86316 | 8542 | |
1c4850e2 YZ |
8543 | eb = path->nodes[wc->level]; |
8544 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8545 | |
1c4850e2 YZ |
8546 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8547 | if (nread >= wc->reada_count) | |
8548 | break; | |
bd56b302 | 8549 | |
2dd3e67b | 8550 | cond_resched(); |
1c4850e2 YZ |
8551 | bytenr = btrfs_node_blockptr(eb, slot); |
8552 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8553 | |
1c4850e2 YZ |
8554 | if (slot == path->slots[wc->level]) |
8555 | goto reada; | |
5d4f98a2 | 8556 | |
1c4850e2 YZ |
8557 | if (wc->stage == UPDATE_BACKREF && |
8558 | generation <= root->root_key.offset) | |
bd56b302 CM |
8559 | continue; |
8560 | ||
94fcca9f | 8561 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8562 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8563 | wc->level - 1, 1, &refs, |
8564 | &flags); | |
79787eaa JM |
8565 | /* We don't care about errors in readahead. */ |
8566 | if (ret < 0) | |
8567 | continue; | |
94fcca9f YZ |
8568 | BUG_ON(refs == 0); |
8569 | ||
1c4850e2 | 8570 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8571 | if (refs == 1) |
8572 | goto reada; | |
bd56b302 | 8573 | |
94fcca9f YZ |
8574 | if (wc->level == 1 && |
8575 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8576 | continue; | |
1c4850e2 YZ |
8577 | if (!wc->update_ref || |
8578 | generation <= root->root_key.offset) | |
8579 | continue; | |
8580 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8581 | ret = btrfs_comp_cpu_keys(&key, | |
8582 | &wc->update_progress); | |
8583 | if (ret < 0) | |
8584 | continue; | |
94fcca9f YZ |
8585 | } else { |
8586 | if (wc->level == 1 && | |
8587 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8588 | continue; | |
6407bf6d | 8589 | } |
1c4850e2 | 8590 | reada: |
2ff7e61e | 8591 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8592 | nread++; |
20524f02 | 8593 | } |
1c4850e2 | 8594 | wc->reada_slot = slot; |
20524f02 | 8595 | } |
2c47e605 | 8596 | |
f82d02d9 | 8597 | /* |
2c016dc2 | 8598 | * helper to process tree block while walking down the tree. |
2c47e605 | 8599 | * |
2c47e605 YZ |
8600 | * when wc->stage == UPDATE_BACKREF, this function updates |
8601 | * back refs for pointers in the block. | |
8602 | * | |
8603 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8604 | */ |
2c47e605 | 8605 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8606 | struct btrfs_root *root, |
2c47e605 | 8607 | struct btrfs_path *path, |
94fcca9f | 8608 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8609 | { |
2ff7e61e | 8610 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8611 | int level = wc->level; |
8612 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8613 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8614 | int ret; |
8615 | ||
2c47e605 YZ |
8616 | if (wc->stage == UPDATE_BACKREF && |
8617 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8618 | return 1; | |
f82d02d9 | 8619 | |
2c47e605 YZ |
8620 | /* |
8621 | * when reference count of tree block is 1, it won't increase | |
8622 | * again. once full backref flag is set, we never clear it. | |
8623 | */ | |
94fcca9f YZ |
8624 | if (lookup_info && |
8625 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8626 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8627 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8628 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8629 | eb->start, level, 1, |
2c47e605 YZ |
8630 | &wc->refs[level], |
8631 | &wc->flags[level]); | |
79787eaa JM |
8632 | BUG_ON(ret == -ENOMEM); |
8633 | if (ret) | |
8634 | return ret; | |
2c47e605 YZ |
8635 | BUG_ON(wc->refs[level] == 0); |
8636 | } | |
5d4f98a2 | 8637 | |
2c47e605 YZ |
8638 | if (wc->stage == DROP_REFERENCE) { |
8639 | if (wc->refs[level] > 1) | |
8640 | return 1; | |
f82d02d9 | 8641 | |
2c47e605 | 8642 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8643 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8644 | path->locks[level] = 0; |
8645 | } | |
8646 | return 0; | |
8647 | } | |
f82d02d9 | 8648 | |
2c47e605 YZ |
8649 | /* wc->stage == UPDATE_BACKREF */ |
8650 | if (!(wc->flags[level] & flag)) { | |
8651 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8652 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8653 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8654 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8655 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8656 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8657 | eb->len, flag, |
8658 | btrfs_header_level(eb), 0); | |
79787eaa | 8659 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8660 | wc->flags[level] |= flag; |
8661 | } | |
8662 | ||
8663 | /* | |
8664 | * the block is shared by multiple trees, so it's not good to | |
8665 | * keep the tree lock | |
8666 | */ | |
8667 | if (path->locks[level] && level > 0) { | |
bd681513 | 8668 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8669 | path->locks[level] = 0; |
8670 | } | |
8671 | return 0; | |
8672 | } | |
8673 | ||
1c4850e2 | 8674 | /* |
2c016dc2 | 8675 | * helper to process tree block pointer. |
1c4850e2 YZ |
8676 | * |
8677 | * when wc->stage == DROP_REFERENCE, this function checks | |
8678 | * reference count of the block pointed to. if the block | |
8679 | * is shared and we need update back refs for the subtree | |
8680 | * rooted at the block, this function changes wc->stage to | |
8681 | * UPDATE_BACKREF. if the block is shared and there is no | |
8682 | * need to update back, this function drops the reference | |
8683 | * to the block. | |
8684 | * | |
8685 | * NOTE: return value 1 means we should stop walking down. | |
8686 | */ | |
8687 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8688 | struct btrfs_root *root, | |
8689 | struct btrfs_path *path, | |
94fcca9f | 8690 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8691 | { |
0b246afa | 8692 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8693 | u64 bytenr; |
8694 | u64 generation; | |
8695 | u64 parent; | |
8696 | u32 blocksize; | |
8697 | struct btrfs_key key; | |
581c1760 | 8698 | struct btrfs_key first_key; |
1c4850e2 YZ |
8699 | struct extent_buffer *next; |
8700 | int level = wc->level; | |
8701 | int reada = 0; | |
8702 | int ret = 0; | |
1152651a | 8703 | bool need_account = false; |
1c4850e2 YZ |
8704 | |
8705 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8706 | path->slots[level]); | |
8707 | /* | |
8708 | * if the lower level block was created before the snapshot | |
8709 | * was created, we know there is no need to update back refs | |
8710 | * for the subtree | |
8711 | */ | |
8712 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8713 | generation <= root->root_key.offset) { |
8714 | *lookup_info = 1; | |
1c4850e2 | 8715 | return 1; |
94fcca9f | 8716 | } |
1c4850e2 YZ |
8717 | |
8718 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
8719 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
8720 | path->slots[level]); | |
0b246afa | 8721 | blocksize = fs_info->nodesize; |
1c4850e2 | 8722 | |
0b246afa | 8723 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8724 | if (!next) { |
2ff7e61e | 8725 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8726 | if (IS_ERR(next)) |
8727 | return PTR_ERR(next); | |
8728 | ||
b2aaaa3b JB |
8729 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8730 | level - 1); | |
1c4850e2 YZ |
8731 | reada = 1; |
8732 | } | |
8733 | btrfs_tree_lock(next); | |
8734 | btrfs_set_lock_blocking(next); | |
8735 | ||
2ff7e61e | 8736 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8737 | &wc->refs[level - 1], |
8738 | &wc->flags[level - 1]); | |
4867268c JB |
8739 | if (ret < 0) |
8740 | goto out_unlock; | |
79787eaa | 8741 | |
c2cf52eb | 8742 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8743 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8744 | ret = -EIO; |
8745 | goto out_unlock; | |
c2cf52eb | 8746 | } |
94fcca9f | 8747 | *lookup_info = 0; |
1c4850e2 | 8748 | |
94fcca9f | 8749 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8750 | if (wc->refs[level - 1] > 1) { |
1152651a | 8751 | need_account = true; |
94fcca9f YZ |
8752 | if (level == 1 && |
8753 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8754 | goto skip; | |
8755 | ||
1c4850e2 YZ |
8756 | if (!wc->update_ref || |
8757 | generation <= root->root_key.offset) | |
8758 | goto skip; | |
8759 | ||
8760 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8761 | path->slots[level]); | |
8762 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8763 | if (ret < 0) | |
8764 | goto skip; | |
8765 | ||
8766 | wc->stage = UPDATE_BACKREF; | |
8767 | wc->shared_level = level - 1; | |
8768 | } | |
94fcca9f YZ |
8769 | } else { |
8770 | if (level == 1 && | |
8771 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8772 | goto skip; | |
1c4850e2 YZ |
8773 | } |
8774 | ||
b9fab919 | 8775 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8776 | btrfs_tree_unlock(next); |
8777 | free_extent_buffer(next); | |
8778 | next = NULL; | |
94fcca9f | 8779 | *lookup_info = 1; |
1c4850e2 YZ |
8780 | } |
8781 | ||
8782 | if (!next) { | |
8783 | if (reada && level == 1) | |
8784 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
8785 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
8786 | &first_key); | |
64c043de LB |
8787 | if (IS_ERR(next)) { |
8788 | return PTR_ERR(next); | |
8789 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8790 | free_extent_buffer(next); |
97d9a8a4 | 8791 | return -EIO; |
416bc658 | 8792 | } |
1c4850e2 YZ |
8793 | btrfs_tree_lock(next); |
8794 | btrfs_set_lock_blocking(next); | |
8795 | } | |
8796 | ||
8797 | level--; | |
4867268c JB |
8798 | ASSERT(level == btrfs_header_level(next)); |
8799 | if (level != btrfs_header_level(next)) { | |
8800 | btrfs_err(root->fs_info, "mismatched level"); | |
8801 | ret = -EIO; | |
8802 | goto out_unlock; | |
8803 | } | |
1c4850e2 YZ |
8804 | path->nodes[level] = next; |
8805 | path->slots[level] = 0; | |
bd681513 | 8806 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8807 | wc->level = level; |
8808 | if (wc->level == 1) | |
8809 | wc->reada_slot = 0; | |
8810 | return 0; | |
8811 | skip: | |
8812 | wc->refs[level - 1] = 0; | |
8813 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8814 | if (wc->stage == DROP_REFERENCE) { |
8815 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8816 | parent = path->nodes[level]->start; | |
8817 | } else { | |
4867268c | 8818 | ASSERT(root->root_key.objectid == |
94fcca9f | 8819 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8820 | if (root->root_key.objectid != |
8821 | btrfs_header_owner(path->nodes[level])) { | |
8822 | btrfs_err(root->fs_info, | |
8823 | "mismatched block owner"); | |
8824 | ret = -EIO; | |
8825 | goto out_unlock; | |
8826 | } | |
94fcca9f YZ |
8827 | parent = 0; |
8828 | } | |
1c4850e2 | 8829 | |
1152651a | 8830 | if (need_account) { |
33d1f05c QW |
8831 | ret = btrfs_qgroup_trace_subtree(trans, root, next, |
8832 | generation, level - 1); | |
1152651a | 8833 | if (ret) { |
0b246afa | 8834 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8835 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8836 | ret); | |
1152651a MF |
8837 | } |
8838 | } | |
84f7d8e6 | 8839 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
2ff7e61e JM |
8840 | parent, root->root_key.objectid, |
8841 | level - 1, 0); | |
4867268c JB |
8842 | if (ret) |
8843 | goto out_unlock; | |
1c4850e2 | 8844 | } |
4867268c JB |
8845 | |
8846 | *lookup_info = 1; | |
8847 | ret = 1; | |
8848 | ||
8849 | out_unlock: | |
1c4850e2 YZ |
8850 | btrfs_tree_unlock(next); |
8851 | free_extent_buffer(next); | |
4867268c JB |
8852 | |
8853 | return ret; | |
1c4850e2 YZ |
8854 | } |
8855 | ||
2c47e605 | 8856 | /* |
2c016dc2 | 8857 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8858 | * |
8859 | * when wc->stage == DROP_REFERENCE, this function drops | |
8860 | * reference count on the block. | |
8861 | * | |
8862 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8863 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8864 | * to UPDATE_BACKREF previously while processing the block. | |
8865 | * | |
8866 | * NOTE: return value 1 means we should stop walking up. | |
8867 | */ | |
8868 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8869 | struct btrfs_root *root, | |
8870 | struct btrfs_path *path, | |
8871 | struct walk_control *wc) | |
8872 | { | |
0b246afa | 8873 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8874 | int ret; |
2c47e605 YZ |
8875 | int level = wc->level; |
8876 | struct extent_buffer *eb = path->nodes[level]; | |
8877 | u64 parent = 0; | |
8878 | ||
8879 | if (wc->stage == UPDATE_BACKREF) { | |
8880 | BUG_ON(wc->shared_level < level); | |
8881 | if (level < wc->shared_level) | |
8882 | goto out; | |
8883 | ||
2c47e605 YZ |
8884 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8885 | if (ret > 0) | |
8886 | wc->update_ref = 0; | |
8887 | ||
8888 | wc->stage = DROP_REFERENCE; | |
8889 | wc->shared_level = -1; | |
8890 | path->slots[level] = 0; | |
8891 | ||
8892 | /* | |
8893 | * check reference count again if the block isn't locked. | |
8894 | * we should start walking down the tree again if reference | |
8895 | * count is one. | |
8896 | */ | |
8897 | if (!path->locks[level]) { | |
8898 | BUG_ON(level == 0); | |
8899 | btrfs_tree_lock(eb); | |
8900 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8901 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8902 | |
2ff7e61e | 8903 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8904 | eb->start, level, 1, |
2c47e605 YZ |
8905 | &wc->refs[level], |
8906 | &wc->flags[level]); | |
79787eaa JM |
8907 | if (ret < 0) { |
8908 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8909 | path->locks[level] = 0; |
79787eaa JM |
8910 | return ret; |
8911 | } | |
2c47e605 YZ |
8912 | BUG_ON(wc->refs[level] == 0); |
8913 | if (wc->refs[level] == 1) { | |
bd681513 | 8914 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8915 | path->locks[level] = 0; |
2c47e605 YZ |
8916 | return 1; |
8917 | } | |
f82d02d9 | 8918 | } |
2c47e605 | 8919 | } |
f82d02d9 | 8920 | |
2c47e605 YZ |
8921 | /* wc->stage == DROP_REFERENCE */ |
8922 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8923 | |
2c47e605 YZ |
8924 | if (wc->refs[level] == 1) { |
8925 | if (level == 0) { | |
8926 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8927 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8928 | else |
e339a6b0 | 8929 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8930 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8931 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb); |
1152651a | 8932 | if (ret) { |
0b246afa | 8933 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8934 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
8935 | ret); | |
1152651a | 8936 | } |
2c47e605 YZ |
8937 | } |
8938 | /* make block locked assertion in clean_tree_block happy */ | |
8939 | if (!path->locks[level] && | |
8940 | btrfs_header_generation(eb) == trans->transid) { | |
8941 | btrfs_tree_lock(eb); | |
8942 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8943 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8944 | } |
7c302b49 | 8945 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
8946 | } |
8947 | ||
8948 | if (eb == root->node) { | |
8949 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8950 | parent = eb->start; | |
8951 | else | |
8952 | BUG_ON(root->root_key.objectid != | |
8953 | btrfs_header_owner(eb)); | |
8954 | } else { | |
8955 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8956 | parent = path->nodes[level + 1]->start; | |
8957 | else | |
8958 | BUG_ON(root->root_key.objectid != | |
8959 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8960 | } |
f82d02d9 | 8961 | |
5581a51a | 8962 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8963 | out: |
8964 | wc->refs[level] = 0; | |
8965 | wc->flags[level] = 0; | |
f0486c68 | 8966 | return 0; |
2c47e605 YZ |
8967 | } |
8968 | ||
8969 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8970 | struct btrfs_root *root, | |
8971 | struct btrfs_path *path, | |
8972 | struct walk_control *wc) | |
8973 | { | |
2c47e605 | 8974 | int level = wc->level; |
94fcca9f | 8975 | int lookup_info = 1; |
2c47e605 YZ |
8976 | int ret; |
8977 | ||
8978 | while (level >= 0) { | |
94fcca9f | 8979 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8980 | if (ret > 0) |
8981 | break; | |
8982 | ||
8983 | if (level == 0) | |
8984 | break; | |
8985 | ||
7a7965f8 YZ |
8986 | if (path->slots[level] >= |
8987 | btrfs_header_nritems(path->nodes[level])) | |
8988 | break; | |
8989 | ||
94fcca9f | 8990 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8991 | if (ret > 0) { |
8992 | path->slots[level]++; | |
8993 | continue; | |
90d2c51d MX |
8994 | } else if (ret < 0) |
8995 | return ret; | |
1c4850e2 | 8996 | level = wc->level; |
f82d02d9 | 8997 | } |
f82d02d9 YZ |
8998 | return 0; |
8999 | } | |
9000 | ||
d397712b | 9001 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 9002 | struct btrfs_root *root, |
f82d02d9 | 9003 | struct btrfs_path *path, |
2c47e605 | 9004 | struct walk_control *wc, int max_level) |
20524f02 | 9005 | { |
2c47e605 | 9006 | int level = wc->level; |
20524f02 | 9007 | int ret; |
9f3a7427 | 9008 | |
2c47e605 YZ |
9009 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
9010 | while (level < max_level && path->nodes[level]) { | |
9011 | wc->level = level; | |
9012 | if (path->slots[level] + 1 < | |
9013 | btrfs_header_nritems(path->nodes[level])) { | |
9014 | path->slots[level]++; | |
20524f02 CM |
9015 | return 0; |
9016 | } else { | |
2c47e605 YZ |
9017 | ret = walk_up_proc(trans, root, path, wc); |
9018 | if (ret > 0) | |
9019 | return 0; | |
bd56b302 | 9020 | |
2c47e605 | 9021 | if (path->locks[level]) { |
bd681513 CM |
9022 | btrfs_tree_unlock_rw(path->nodes[level], |
9023 | path->locks[level]); | |
2c47e605 | 9024 | path->locks[level] = 0; |
f82d02d9 | 9025 | } |
2c47e605 YZ |
9026 | free_extent_buffer(path->nodes[level]); |
9027 | path->nodes[level] = NULL; | |
9028 | level++; | |
20524f02 CM |
9029 | } |
9030 | } | |
9031 | return 1; | |
9032 | } | |
9033 | ||
9aca1d51 | 9034 | /* |
2c47e605 YZ |
9035 | * drop a subvolume tree. |
9036 | * | |
9037 | * this function traverses the tree freeing any blocks that only | |
9038 | * referenced by the tree. | |
9039 | * | |
9040 | * when a shared tree block is found. this function decreases its | |
9041 | * reference count by one. if update_ref is true, this function | |
9042 | * also make sure backrefs for the shared block and all lower level | |
9043 | * blocks are properly updated. | |
9d1a2a3a DS |
9044 | * |
9045 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9046 | */ |
2c536799 | 9047 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9048 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9049 | int for_reloc) | |
20524f02 | 9050 | { |
ab8d0fc4 | 9051 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9052 | struct btrfs_path *path; |
2c47e605 | 9053 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9054 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9055 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9056 | struct walk_control *wc; |
9057 | struct btrfs_key key; | |
9058 | int err = 0; | |
9059 | int ret; | |
9060 | int level; | |
d29a9f62 | 9061 | bool root_dropped = false; |
20524f02 | 9062 | |
ab8d0fc4 | 9063 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
1152651a | 9064 | |
5caf2a00 | 9065 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9066 | if (!path) { |
9067 | err = -ENOMEM; | |
9068 | goto out; | |
9069 | } | |
20524f02 | 9070 | |
2c47e605 | 9071 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9072 | if (!wc) { |
9073 | btrfs_free_path(path); | |
cb1b69f4 TI |
9074 | err = -ENOMEM; |
9075 | goto out; | |
38a1a919 | 9076 | } |
2c47e605 | 9077 | |
a22285a6 | 9078 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9079 | if (IS_ERR(trans)) { |
9080 | err = PTR_ERR(trans); | |
9081 | goto out_free; | |
9082 | } | |
98d5dc13 | 9083 | |
3fd0a558 YZ |
9084 | if (block_rsv) |
9085 | trans->block_rsv = block_rsv; | |
2c47e605 | 9086 | |
9f3a7427 | 9087 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9088 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9089 | path->nodes[level] = btrfs_lock_root_node(root); |
9090 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9091 | path->slots[level] = 0; |
bd681513 | 9092 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9093 | memset(&wc->update_progress, 0, |
9094 | sizeof(wc->update_progress)); | |
9f3a7427 | 9095 | } else { |
9f3a7427 | 9096 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9097 | memcpy(&wc->update_progress, &key, |
9098 | sizeof(wc->update_progress)); | |
9099 | ||
6702ed49 | 9100 | level = root_item->drop_level; |
2c47e605 | 9101 | BUG_ON(level == 0); |
6702ed49 | 9102 | path->lowest_level = level; |
2c47e605 YZ |
9103 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9104 | path->lowest_level = 0; | |
9105 | if (ret < 0) { | |
9106 | err = ret; | |
79787eaa | 9107 | goto out_end_trans; |
9f3a7427 | 9108 | } |
1c4850e2 | 9109 | WARN_ON(ret > 0); |
2c47e605 | 9110 | |
7d9eb12c CM |
9111 | /* |
9112 | * unlock our path, this is safe because only this | |
9113 | * function is allowed to delete this snapshot | |
9114 | */ | |
5d4f98a2 | 9115 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9116 | |
9117 | level = btrfs_header_level(root->node); | |
9118 | while (1) { | |
9119 | btrfs_tree_lock(path->nodes[level]); | |
9120 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9121 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9122 | |
2ff7e61e | 9123 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9124 | path->nodes[level]->start, |
3173a18f | 9125 | level, 1, &wc->refs[level], |
2c47e605 | 9126 | &wc->flags[level]); |
79787eaa JM |
9127 | if (ret < 0) { |
9128 | err = ret; | |
9129 | goto out_end_trans; | |
9130 | } | |
2c47e605 YZ |
9131 | BUG_ON(wc->refs[level] == 0); |
9132 | ||
9133 | if (level == root_item->drop_level) | |
9134 | break; | |
9135 | ||
9136 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9137 | path->locks[level] = 0; |
2c47e605 YZ |
9138 | WARN_ON(wc->refs[level] != 1); |
9139 | level--; | |
9140 | } | |
9f3a7427 | 9141 | } |
2c47e605 YZ |
9142 | |
9143 | wc->level = level; | |
9144 | wc->shared_level = -1; | |
9145 | wc->stage = DROP_REFERENCE; | |
9146 | wc->update_ref = update_ref; | |
9147 | wc->keep_locks = 0; | |
66d7e7f0 | 9148 | wc->for_reloc = for_reloc; |
0b246afa | 9149 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9150 | |
d397712b | 9151 | while (1) { |
9d1a2a3a | 9152 | |
2c47e605 YZ |
9153 | ret = walk_down_tree(trans, root, path, wc); |
9154 | if (ret < 0) { | |
9155 | err = ret; | |
20524f02 | 9156 | break; |
2c47e605 | 9157 | } |
9aca1d51 | 9158 | |
2c47e605 YZ |
9159 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9160 | if (ret < 0) { | |
9161 | err = ret; | |
20524f02 | 9162 | break; |
2c47e605 YZ |
9163 | } |
9164 | ||
9165 | if (ret > 0) { | |
9166 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9167 | break; |
9168 | } | |
2c47e605 YZ |
9169 | |
9170 | if (wc->stage == DROP_REFERENCE) { | |
9171 | level = wc->level; | |
9172 | btrfs_node_key(path->nodes[level], | |
9173 | &root_item->drop_progress, | |
9174 | path->slots[level]); | |
9175 | root_item->drop_level = level; | |
9176 | } | |
9177 | ||
9178 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9179 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9180 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9181 | ret = btrfs_update_root(trans, tree_root, |
9182 | &root->root_key, | |
9183 | root_item); | |
79787eaa | 9184 | if (ret) { |
66642832 | 9185 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9186 | err = ret; |
9187 | goto out_end_trans; | |
9188 | } | |
2c47e605 | 9189 | |
3a45bb20 | 9190 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9191 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9192 | btrfs_debug(fs_info, |
9193 | "drop snapshot early exit"); | |
3c8f2422 JB |
9194 | err = -EAGAIN; |
9195 | goto out_free; | |
9196 | } | |
9197 | ||
a22285a6 | 9198 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9199 | if (IS_ERR(trans)) { |
9200 | err = PTR_ERR(trans); | |
9201 | goto out_free; | |
9202 | } | |
3fd0a558 YZ |
9203 | if (block_rsv) |
9204 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9205 | } |
20524f02 | 9206 | } |
b3b4aa74 | 9207 | btrfs_release_path(path); |
79787eaa JM |
9208 | if (err) |
9209 | goto out_end_trans; | |
2c47e605 | 9210 | |
1cd5447e | 9211 | ret = btrfs_del_root(trans, fs_info, &root->root_key); |
79787eaa | 9212 | if (ret) { |
66642832 | 9213 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 9214 | err = ret; |
79787eaa JM |
9215 | goto out_end_trans; |
9216 | } | |
2c47e605 | 9217 | |
76dda93c | 9218 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9219 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9220 | NULL, NULL); | |
79787eaa | 9221 | if (ret < 0) { |
66642832 | 9222 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9223 | err = ret; |
9224 | goto out_end_trans; | |
9225 | } else if (ret > 0) { | |
84cd948c JB |
9226 | /* if we fail to delete the orphan item this time |
9227 | * around, it'll get picked up the next time. | |
9228 | * | |
9229 | * The most common failure here is just -ENOENT. | |
9230 | */ | |
9231 | btrfs_del_orphan_item(trans, tree_root, | |
9232 | root->root_key.objectid); | |
76dda93c YZ |
9233 | } |
9234 | } | |
9235 | ||
27cdeb70 | 9236 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9237 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9238 | } else { |
9239 | free_extent_buffer(root->node); | |
9240 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9241 | btrfs_put_fs_root(root); |
76dda93c | 9242 | } |
d29a9f62 | 9243 | root_dropped = true; |
79787eaa | 9244 | out_end_trans: |
3a45bb20 | 9245 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9246 | out_free: |
2c47e605 | 9247 | kfree(wc); |
5caf2a00 | 9248 | btrfs_free_path(path); |
cb1b69f4 | 9249 | out: |
d29a9f62 JB |
9250 | /* |
9251 | * So if we need to stop dropping the snapshot for whatever reason we | |
9252 | * need to make sure to add it back to the dead root list so that we | |
9253 | * keep trying to do the work later. This also cleans up roots if we | |
9254 | * don't have it in the radix (like when we recover after a power fail | |
9255 | * or unmount) so we don't leak memory. | |
9256 | */ | |
897ca819 | 9257 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9258 | btrfs_add_dead_root(root); |
90515e7f | 9259 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9260 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9261 | return err; |
20524f02 | 9262 | } |
9078a3e1 | 9263 | |
2c47e605 YZ |
9264 | /* |
9265 | * drop subtree rooted at tree block 'node'. | |
9266 | * | |
9267 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9268 | * only used by relocation code |
2c47e605 | 9269 | */ |
f82d02d9 YZ |
9270 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9271 | struct btrfs_root *root, | |
9272 | struct extent_buffer *node, | |
9273 | struct extent_buffer *parent) | |
9274 | { | |
0b246afa | 9275 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9276 | struct btrfs_path *path; |
2c47e605 | 9277 | struct walk_control *wc; |
f82d02d9 YZ |
9278 | int level; |
9279 | int parent_level; | |
9280 | int ret = 0; | |
9281 | int wret; | |
9282 | ||
2c47e605 YZ |
9283 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9284 | ||
f82d02d9 | 9285 | path = btrfs_alloc_path(); |
db5b493a TI |
9286 | if (!path) |
9287 | return -ENOMEM; | |
f82d02d9 | 9288 | |
2c47e605 | 9289 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9290 | if (!wc) { |
9291 | btrfs_free_path(path); | |
9292 | return -ENOMEM; | |
9293 | } | |
2c47e605 | 9294 | |
b9447ef8 | 9295 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9296 | parent_level = btrfs_header_level(parent); |
9297 | extent_buffer_get(parent); | |
9298 | path->nodes[parent_level] = parent; | |
9299 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9300 | ||
b9447ef8 | 9301 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9302 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9303 | path->nodes[level] = node; |
9304 | path->slots[level] = 0; | |
bd681513 | 9305 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9306 | |
9307 | wc->refs[parent_level] = 1; | |
9308 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9309 | wc->level = level; | |
9310 | wc->shared_level = -1; | |
9311 | wc->stage = DROP_REFERENCE; | |
9312 | wc->update_ref = 0; | |
9313 | wc->keep_locks = 1; | |
66d7e7f0 | 9314 | wc->for_reloc = 1; |
0b246afa | 9315 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9316 | |
9317 | while (1) { | |
2c47e605 YZ |
9318 | wret = walk_down_tree(trans, root, path, wc); |
9319 | if (wret < 0) { | |
f82d02d9 | 9320 | ret = wret; |
f82d02d9 | 9321 | break; |
2c47e605 | 9322 | } |
f82d02d9 | 9323 | |
2c47e605 | 9324 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9325 | if (wret < 0) |
9326 | ret = wret; | |
9327 | if (wret != 0) | |
9328 | break; | |
9329 | } | |
9330 | ||
2c47e605 | 9331 | kfree(wc); |
f82d02d9 YZ |
9332 | btrfs_free_path(path); |
9333 | return ret; | |
9334 | } | |
9335 | ||
6202df69 | 9336 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9337 | { |
9338 | u64 num_devices; | |
fc67c450 | 9339 | u64 stripped; |
e4d8ec0f | 9340 | |
fc67c450 ID |
9341 | /* |
9342 | * if restripe for this chunk_type is on pick target profile and | |
9343 | * return, otherwise do the usual balance | |
9344 | */ | |
6202df69 | 9345 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9346 | if (stripped) |
9347 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9348 | |
6202df69 | 9349 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9350 | |
fc67c450 | 9351 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9352 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9353 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9354 | ||
ec44a35c CM |
9355 | if (num_devices == 1) { |
9356 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9357 | stripped = flags & ~stripped; | |
9358 | ||
9359 | /* turn raid0 into single device chunks */ | |
9360 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9361 | return stripped; | |
9362 | ||
9363 | /* turn mirroring into duplication */ | |
9364 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9365 | BTRFS_BLOCK_GROUP_RAID10)) | |
9366 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9367 | } else { |
9368 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9369 | if (flags & stripped) |
9370 | return flags; | |
9371 | ||
9372 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9373 | stripped = flags & ~stripped; | |
9374 | ||
9375 | /* switch duplicated blocks with raid1 */ | |
9376 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9377 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9378 | ||
e3176ca2 | 9379 | /* this is drive concat, leave it alone */ |
ec44a35c | 9380 | } |
e3176ca2 | 9381 | |
ec44a35c CM |
9382 | return flags; |
9383 | } | |
9384 | ||
868f401a | 9385 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9386 | { |
f0486c68 YZ |
9387 | struct btrfs_space_info *sinfo = cache->space_info; |
9388 | u64 num_bytes; | |
199c36ea | 9389 | u64 min_allocable_bytes; |
f0486c68 | 9390 | int ret = -ENOSPC; |
0ef3e66b | 9391 | |
199c36ea MX |
9392 | /* |
9393 | * We need some metadata space and system metadata space for | |
9394 | * allocating chunks in some corner cases until we force to set | |
9395 | * it to be readonly. | |
9396 | */ | |
9397 | if ((sinfo->flags & | |
9398 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9399 | !force) | |
ee22184b | 9400 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9401 | else |
9402 | min_allocable_bytes = 0; | |
9403 | ||
f0486c68 YZ |
9404 | spin_lock(&sinfo->lock); |
9405 | spin_lock(&cache->lock); | |
61cfea9b W |
9406 | |
9407 | if (cache->ro) { | |
868f401a | 9408 | cache->ro++; |
61cfea9b W |
9409 | ret = 0; |
9410 | goto out; | |
9411 | } | |
9412 | ||
f0486c68 YZ |
9413 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9414 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9415 | ||
4136135b | 9416 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9417 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9418 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9419 | cache->ro++; |
633c0aad | 9420 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9421 | ret = 0; |
9422 | } | |
61cfea9b | 9423 | out: |
f0486c68 YZ |
9424 | spin_unlock(&cache->lock); |
9425 | spin_unlock(&sinfo->lock); | |
9426 | return ret; | |
9427 | } | |
7d9eb12c | 9428 | |
5e00f193 | 9429 | int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info, |
f0486c68 | 9430 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9431 | |
f0486c68 YZ |
9432 | { |
9433 | struct btrfs_trans_handle *trans; | |
9434 | u64 alloc_flags; | |
9435 | int ret; | |
7d9eb12c | 9436 | |
1bbc621e | 9437 | again: |
5e00f193 | 9438 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9439 | if (IS_ERR(trans)) |
9440 | return PTR_ERR(trans); | |
5d4f98a2 | 9441 | |
1bbc621e CM |
9442 | /* |
9443 | * we're not allowed to set block groups readonly after the dirty | |
9444 | * block groups cache has started writing. If it already started, | |
9445 | * back off and let this transaction commit | |
9446 | */ | |
0b246afa | 9447 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9448 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9449 | u64 transid = trans->transid; |
9450 | ||
0b246afa | 9451 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9452 | btrfs_end_transaction(trans); |
1bbc621e | 9453 | |
2ff7e61e | 9454 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9455 | if (ret) |
9456 | return ret; | |
9457 | goto again; | |
9458 | } | |
9459 | ||
153c35b6 CM |
9460 | /* |
9461 | * if we are changing raid levels, try to allocate a corresponding | |
9462 | * block group with the new raid level. | |
9463 | */ | |
0b246afa | 9464 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9465 | if (alloc_flags != cache->flags) { |
2ff7e61e | 9466 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, |
153c35b6 CM |
9467 | CHUNK_ALLOC_FORCE); |
9468 | /* | |
9469 | * ENOSPC is allowed here, we may have enough space | |
9470 | * already allocated at the new raid level to | |
9471 | * carry on | |
9472 | */ | |
9473 | if (ret == -ENOSPC) | |
9474 | ret = 0; | |
9475 | if (ret < 0) | |
9476 | goto out; | |
9477 | } | |
1bbc621e | 9478 | |
868f401a | 9479 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9480 | if (!ret) |
9481 | goto out; | |
2ff7e61e JM |
9482 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
9483 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, | |
0e4f8f88 | 9484 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9485 | if (ret < 0) |
9486 | goto out; | |
868f401a | 9487 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9488 | out: |
2f081088 | 9489 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9490 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9491 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 9492 | check_system_chunk(trans, fs_info, alloc_flags); |
34441361 | 9493 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9494 | } |
0b246afa | 9495 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9496 | |
3a45bb20 | 9497 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9498 | return ret; |
9499 | } | |
5d4f98a2 | 9500 | |
c87f08ca | 9501 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 9502 | struct btrfs_fs_info *fs_info, u64 type) |
c87f08ca | 9503 | { |
2ff7e61e JM |
9504 | u64 alloc_flags = get_alloc_profile(fs_info, type); |
9505 | ||
9506 | return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
9507 | } |
9508 | ||
6d07bcec MX |
9509 | /* |
9510 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9511 | * space_info. takes mirrors into account. |
6d07bcec | 9512 | */ |
633c0aad | 9513 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9514 | { |
9515 | struct btrfs_block_group_cache *block_group; | |
9516 | u64 free_bytes = 0; | |
9517 | int factor; | |
9518 | ||
01327610 | 9519 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9520 | if (list_empty(&sinfo->ro_bgs)) |
9521 | return 0; | |
9522 | ||
9523 | spin_lock(&sinfo->lock); | |
9524 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9525 | spin_lock(&block_group->lock); |
9526 | ||
9527 | if (!block_group->ro) { | |
9528 | spin_unlock(&block_group->lock); | |
9529 | continue; | |
9530 | } | |
9531 | ||
9532 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9533 | BTRFS_BLOCK_GROUP_RAID10 | | |
9534 | BTRFS_BLOCK_GROUP_DUP)) | |
9535 | factor = 2; | |
9536 | else | |
9537 | factor = 1; | |
9538 | ||
9539 | free_bytes += (block_group->key.offset - | |
9540 | btrfs_block_group_used(&block_group->item)) * | |
9541 | factor; | |
9542 | ||
9543 | spin_unlock(&block_group->lock); | |
9544 | } | |
6d07bcec MX |
9545 | spin_unlock(&sinfo->lock); |
9546 | ||
9547 | return free_bytes; | |
9548 | } | |
9549 | ||
2ff7e61e | 9550 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9551 | { |
f0486c68 YZ |
9552 | struct btrfs_space_info *sinfo = cache->space_info; |
9553 | u64 num_bytes; | |
9554 | ||
9555 | BUG_ON(!cache->ro); | |
9556 | ||
9557 | spin_lock(&sinfo->lock); | |
9558 | spin_lock(&cache->lock); | |
868f401a Z |
9559 | if (!--cache->ro) { |
9560 | num_bytes = cache->key.offset - cache->reserved - | |
9561 | cache->pinned - cache->bytes_super - | |
9562 | btrfs_block_group_used(&cache->item); | |
9563 | sinfo->bytes_readonly -= num_bytes; | |
9564 | list_del_init(&cache->ro_list); | |
9565 | } | |
f0486c68 YZ |
9566 | spin_unlock(&cache->lock); |
9567 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9568 | } |
9569 | ||
ba1bf481 JB |
9570 | /* |
9571 | * checks to see if its even possible to relocate this block group. | |
9572 | * | |
9573 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9574 | * ok to go ahead and try. | |
9575 | */ | |
6bccf3ab | 9576 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9577 | { |
6bccf3ab | 9578 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9579 | struct btrfs_block_group_cache *block_group; |
9580 | struct btrfs_space_info *space_info; | |
0b246afa | 9581 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9582 | struct btrfs_device *device; |
6df9a95e | 9583 | struct btrfs_trans_handle *trans; |
cdcb725c | 9584 | u64 min_free; |
6719db6a JB |
9585 | u64 dev_min = 1; |
9586 | u64 dev_nr = 0; | |
4a5e98f5 | 9587 | u64 target; |
0305bc27 | 9588 | int debug; |
cdcb725c | 9589 | int index; |
ba1bf481 JB |
9590 | int full = 0; |
9591 | int ret = 0; | |
1a40e23b | 9592 | |
0b246afa | 9593 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9594 | |
0b246afa | 9595 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9596 | |
ba1bf481 | 9597 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9598 | if (!block_group) { |
9599 | if (debug) | |
0b246afa | 9600 | btrfs_warn(fs_info, |
0305bc27 QW |
9601 | "can't find block group for bytenr %llu", |
9602 | bytenr); | |
ba1bf481 | 9603 | return -1; |
0305bc27 | 9604 | } |
1a40e23b | 9605 | |
cdcb725c | 9606 | min_free = btrfs_block_group_used(&block_group->item); |
9607 | ||
ba1bf481 | 9608 | /* no bytes used, we're good */ |
cdcb725c | 9609 | if (!min_free) |
1a40e23b ZY |
9610 | goto out; |
9611 | ||
ba1bf481 JB |
9612 | space_info = block_group->space_info; |
9613 | spin_lock(&space_info->lock); | |
17d217fe | 9614 | |
ba1bf481 | 9615 | full = space_info->full; |
17d217fe | 9616 | |
ba1bf481 JB |
9617 | /* |
9618 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9619 | * relocate it unless we're able to allocate a new chunk below. |
9620 | * | |
9621 | * Otherwise, we need to make sure we have room in the space to handle | |
9622 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9623 | */ |
7ce618db | 9624 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9625 | (btrfs_space_info_used(space_info, false) + min_free < |
9626 | space_info->total_bytes)) { | |
ba1bf481 JB |
9627 | spin_unlock(&space_info->lock); |
9628 | goto out; | |
17d217fe | 9629 | } |
ba1bf481 | 9630 | spin_unlock(&space_info->lock); |
ea8c2819 | 9631 | |
ba1bf481 JB |
9632 | /* |
9633 | * ok we don't have enough space, but maybe we have free space on our | |
9634 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9635 | * alloc devices and guess if we have enough space. if this block |
9636 | * group is going to be restriped, run checks against the target | |
9637 | * profile instead of the current one. | |
ba1bf481 JB |
9638 | */ |
9639 | ret = -1; | |
ea8c2819 | 9640 | |
cdcb725c | 9641 | /* |
9642 | * index: | |
9643 | * 0: raid10 | |
9644 | * 1: raid1 | |
9645 | * 2: dup | |
9646 | * 3: raid0 | |
9647 | * 4: single | |
9648 | */ | |
0b246afa | 9649 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9650 | if (target) { |
3e72ee88 | 9651 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9652 | } else { |
9653 | /* | |
9654 | * this is just a balance, so if we were marked as full | |
9655 | * we know there is no space for a new chunk | |
9656 | */ | |
0305bc27 QW |
9657 | if (full) { |
9658 | if (debug) | |
0b246afa JM |
9659 | btrfs_warn(fs_info, |
9660 | "no space to alloc new chunk for block group %llu", | |
9661 | block_group->key.objectid); | |
4a5e98f5 | 9662 | goto out; |
0305bc27 | 9663 | } |
4a5e98f5 | 9664 | |
3e72ee88 | 9665 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
9666 | } |
9667 | ||
e6ec716f | 9668 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9669 | dev_min = 4; |
6719db6a JB |
9670 | /* Divide by 2 */ |
9671 | min_free >>= 1; | |
e6ec716f | 9672 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9673 | dev_min = 2; |
e6ec716f | 9674 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9675 | /* Multiply by 2 */ |
9676 | min_free <<= 1; | |
e6ec716f | 9677 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9678 | dev_min = fs_devices->rw_devices; |
47c5713f | 9679 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9680 | } |
9681 | ||
6df9a95e JB |
9682 | /* We need to do this so that we can look at pending chunks */ |
9683 | trans = btrfs_join_transaction(root); | |
9684 | if (IS_ERR(trans)) { | |
9685 | ret = PTR_ERR(trans); | |
9686 | goto out; | |
9687 | } | |
9688 | ||
0b246afa | 9689 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9690 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9691 | u64 dev_offset; |
56bec294 | 9692 | |
ba1bf481 JB |
9693 | /* |
9694 | * check to make sure we can actually find a chunk with enough | |
9695 | * space to fit our block group in. | |
9696 | */ | |
63a212ab | 9697 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 9698 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
6df9a95e | 9699 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9700 | &dev_offset, NULL); |
ba1bf481 | 9701 | if (!ret) |
cdcb725c | 9702 | dev_nr++; |
9703 | ||
9704 | if (dev_nr >= dev_min) | |
73e48b27 | 9705 | break; |
cdcb725c | 9706 | |
ba1bf481 | 9707 | ret = -1; |
725c8463 | 9708 | } |
edbd8d4e | 9709 | } |
0305bc27 | 9710 | if (debug && ret == -1) |
0b246afa JM |
9711 | btrfs_warn(fs_info, |
9712 | "no space to allocate a new chunk for block group %llu", | |
9713 | block_group->key.objectid); | |
9714 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9715 | btrfs_end_transaction(trans); |
edbd8d4e | 9716 | out: |
ba1bf481 | 9717 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9718 | return ret; |
9719 | } | |
9720 | ||
6bccf3ab JM |
9721 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9722 | struct btrfs_path *path, | |
9723 | struct btrfs_key *key) | |
0b86a832 | 9724 | { |
6bccf3ab | 9725 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9726 | int ret = 0; |
0b86a832 CM |
9727 | struct btrfs_key found_key; |
9728 | struct extent_buffer *leaf; | |
9729 | int slot; | |
edbd8d4e | 9730 | |
0b86a832 CM |
9731 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9732 | if (ret < 0) | |
925baedd CM |
9733 | goto out; |
9734 | ||
d397712b | 9735 | while (1) { |
0b86a832 | 9736 | slot = path->slots[0]; |
edbd8d4e | 9737 | leaf = path->nodes[0]; |
0b86a832 CM |
9738 | if (slot >= btrfs_header_nritems(leaf)) { |
9739 | ret = btrfs_next_leaf(root, path); | |
9740 | if (ret == 0) | |
9741 | continue; | |
9742 | if (ret < 0) | |
925baedd | 9743 | goto out; |
0b86a832 | 9744 | break; |
edbd8d4e | 9745 | } |
0b86a832 | 9746 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9747 | |
0b86a832 | 9748 | if (found_key.objectid >= key->objectid && |
925baedd | 9749 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9750 | struct extent_map_tree *em_tree; |
9751 | struct extent_map *em; | |
9752 | ||
9753 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9754 | read_lock(&em_tree->lock); | |
9755 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9756 | found_key.offset); | |
9757 | read_unlock(&em_tree->lock); | |
9758 | if (!em) { | |
0b246afa | 9759 | btrfs_err(fs_info, |
6fb37b75 LB |
9760 | "logical %llu len %llu found bg but no related chunk", |
9761 | found_key.objectid, found_key.offset); | |
9762 | ret = -ENOENT; | |
9763 | } else { | |
9764 | ret = 0; | |
9765 | } | |
187ee58c | 9766 | free_extent_map(em); |
925baedd CM |
9767 | goto out; |
9768 | } | |
0b86a832 | 9769 | path->slots[0]++; |
edbd8d4e | 9770 | } |
925baedd | 9771 | out: |
0b86a832 | 9772 | return ret; |
edbd8d4e CM |
9773 | } |
9774 | ||
0af3d00b JB |
9775 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9776 | { | |
9777 | struct btrfs_block_group_cache *block_group; | |
9778 | u64 last = 0; | |
9779 | ||
9780 | while (1) { | |
9781 | struct inode *inode; | |
9782 | ||
9783 | block_group = btrfs_lookup_first_block_group(info, last); | |
9784 | while (block_group) { | |
9785 | spin_lock(&block_group->lock); | |
9786 | if (block_group->iref) | |
9787 | break; | |
9788 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9789 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9790 | } |
9791 | if (!block_group) { | |
9792 | if (last == 0) | |
9793 | break; | |
9794 | last = 0; | |
9795 | continue; | |
9796 | } | |
9797 | ||
9798 | inode = block_group->inode; | |
9799 | block_group->iref = 0; | |
9800 | block_group->inode = NULL; | |
9801 | spin_unlock(&block_group->lock); | |
f3bca802 | 9802 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9803 | iput(inode); |
9804 | last = block_group->key.objectid + block_group->key.offset; | |
9805 | btrfs_put_block_group(block_group); | |
9806 | } | |
9807 | } | |
9808 | ||
5cdd7db6 FM |
9809 | /* |
9810 | * Must be called only after stopping all workers, since we could have block | |
9811 | * group caching kthreads running, and therefore they could race with us if we | |
9812 | * freed the block groups before stopping them. | |
9813 | */ | |
1a40e23b ZY |
9814 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9815 | { | |
9816 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9817 | struct btrfs_space_info *space_info; |
11833d66 | 9818 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9819 | struct rb_node *n; |
9820 | ||
9e351cc8 | 9821 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9822 | while (!list_empty(&info->caching_block_groups)) { |
9823 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9824 | struct btrfs_caching_control, list); | |
9825 | list_del(&caching_ctl->list); | |
9826 | put_caching_control(caching_ctl); | |
9827 | } | |
9e351cc8 | 9828 | up_write(&info->commit_root_sem); |
11833d66 | 9829 | |
47ab2a6c JB |
9830 | spin_lock(&info->unused_bgs_lock); |
9831 | while (!list_empty(&info->unused_bgs)) { | |
9832 | block_group = list_first_entry(&info->unused_bgs, | |
9833 | struct btrfs_block_group_cache, | |
9834 | bg_list); | |
9835 | list_del_init(&block_group->bg_list); | |
9836 | btrfs_put_block_group(block_group); | |
9837 | } | |
9838 | spin_unlock(&info->unused_bgs_lock); | |
9839 | ||
1a40e23b ZY |
9840 | spin_lock(&info->block_group_cache_lock); |
9841 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9842 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9843 | cache_node); | |
1a40e23b ZY |
9844 | rb_erase(&block_group->cache_node, |
9845 | &info->block_group_cache_tree); | |
01eacb27 | 9846 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9847 | spin_unlock(&info->block_group_cache_lock); |
9848 | ||
80eb234a | 9849 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9850 | list_del(&block_group->list); |
80eb234a | 9851 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9852 | |
3c14874a JB |
9853 | /* |
9854 | * We haven't cached this block group, which means we could | |
9855 | * possibly have excluded extents on this block group. | |
9856 | */ | |
36cce922 JB |
9857 | if (block_group->cached == BTRFS_CACHE_NO || |
9858 | block_group->cached == BTRFS_CACHE_ERROR) | |
2ff7e61e | 9859 | free_excluded_extents(info, block_group); |
3c14874a | 9860 | |
817d52f8 | 9861 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9862 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9863 | ASSERT(list_empty(&block_group->dirty_list)); |
9864 | ASSERT(list_empty(&block_group->io_list)); | |
9865 | ASSERT(list_empty(&block_group->bg_list)); | |
9866 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9867 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9868 | |
9869 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9870 | } |
9871 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9872 | |
9873 | /* now that all the block groups are freed, go through and | |
9874 | * free all the space_info structs. This is only called during | |
9875 | * the final stages of unmount, and so we know nobody is | |
9876 | * using them. We call synchronize_rcu() once before we start, | |
9877 | * just to be on the safe side. | |
9878 | */ | |
9879 | synchronize_rcu(); | |
9880 | ||
8929ecfa YZ |
9881 | release_global_block_rsv(info); |
9882 | ||
67871254 | 9883 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9884 | int i; |
9885 | ||
4184ea7f CM |
9886 | space_info = list_entry(info->space_info.next, |
9887 | struct btrfs_space_info, | |
9888 | list); | |
d555b6c3 JB |
9889 | |
9890 | /* | |
9891 | * Do not hide this behind enospc_debug, this is actually | |
9892 | * important and indicates a real bug if this happens. | |
9893 | */ | |
9894 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9895 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9896 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9897 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9898 | list_del(&space_info->list); |
6ab0a202 JM |
9899 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9900 | struct kobject *kobj; | |
c1895442 JM |
9901 | kobj = space_info->block_group_kobjs[i]; |
9902 | space_info->block_group_kobjs[i] = NULL; | |
9903 | if (kobj) { | |
6ab0a202 JM |
9904 | kobject_del(kobj); |
9905 | kobject_put(kobj); | |
9906 | } | |
9907 | } | |
9908 | kobject_del(&space_info->kobj); | |
9909 | kobject_put(&space_info->kobj); | |
4184ea7f | 9910 | } |
1a40e23b ZY |
9911 | return 0; |
9912 | } | |
9913 | ||
75cb379d JM |
9914 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
9915 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) | |
9916 | { | |
9917 | struct btrfs_space_info *space_info; | |
9918 | struct raid_kobject *rkobj; | |
9919 | LIST_HEAD(list); | |
9920 | int index; | |
9921 | int ret = 0; | |
9922 | ||
9923 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9924 | list_splice_init(&fs_info->pending_raid_kobjs, &list); | |
9925 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
9926 | ||
9927 | list_for_each_entry(rkobj, &list, list) { | |
9928 | space_info = __find_space_info(fs_info, rkobj->flags); | |
9929 | index = btrfs_bg_flags_to_raid_index(rkobj->flags); | |
9930 | ||
9931 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9932 | "%s", get_raid_name(index)); | |
9933 | if (ret) { | |
9934 | kobject_put(&rkobj->kobj); | |
9935 | break; | |
9936 | } | |
9937 | } | |
9938 | if (ret) | |
9939 | btrfs_warn(fs_info, | |
9940 | "failed to add kobject for block cache, ignoring"); | |
9941 | } | |
9942 | ||
c434d21c | 9943 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 9944 | { |
c434d21c | 9945 | struct btrfs_space_info *space_info = cache->space_info; |
75cb379d | 9946 | struct btrfs_fs_info *fs_info = cache->fs_info; |
3e72ee88 | 9947 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 9948 | bool first = false; |
b742bb82 YZ |
9949 | |
9950 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9951 | if (list_empty(&space_info->block_groups[index])) |
9952 | first = true; | |
9953 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9954 | up_write(&space_info->groups_sem); | |
9955 | ||
9956 | if (first) { | |
75cb379d JM |
9957 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9958 | if (!rkobj) { | |
9959 | btrfs_warn(cache->fs_info, | |
9960 | "couldn't alloc memory for raid level kobject"); | |
9961 | return; | |
6ab0a202 | 9962 | } |
75cb379d JM |
9963 | rkobj->flags = cache->flags; |
9964 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9965 | ||
9966 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9967 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); | |
9968 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
c1895442 | 9969 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9970 | } |
b742bb82 YZ |
9971 | } |
9972 | ||
920e4a58 | 9973 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
9974 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
9975 | u64 start, u64 size) | |
920e4a58 MX |
9976 | { |
9977 | struct btrfs_block_group_cache *cache; | |
9978 | ||
9979 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9980 | if (!cache) | |
9981 | return NULL; | |
9982 | ||
9983 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9984 | GFP_NOFS); | |
9985 | if (!cache->free_space_ctl) { | |
9986 | kfree(cache); | |
9987 | return NULL; | |
9988 | } | |
9989 | ||
9990 | cache->key.objectid = start; | |
9991 | cache->key.offset = size; | |
9992 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9993 | ||
0b246afa | 9994 | cache->fs_info = fs_info; |
e4ff5fb5 | 9995 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
9996 | set_free_space_tree_thresholds(cache); |
9997 | ||
920e4a58 MX |
9998 | atomic_set(&cache->count, 1); |
9999 | spin_lock_init(&cache->lock); | |
e570fd27 | 10000 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
10001 | INIT_LIST_HEAD(&cache->list); |
10002 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 10003 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 10004 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 10005 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 10006 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 10007 | btrfs_init_free_space_ctl(cache); |
04216820 | 10008 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 10009 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 10010 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
10011 | |
10012 | return cache; | |
10013 | } | |
10014 | ||
5b4aacef | 10015 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
10016 | { |
10017 | struct btrfs_path *path; | |
10018 | int ret; | |
9078a3e1 | 10019 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 10020 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
10021 | struct btrfs_key key; |
10022 | struct btrfs_key found_key; | |
5f39d397 | 10023 | struct extent_buffer *leaf; |
0af3d00b JB |
10024 | int need_clear = 0; |
10025 | u64 cache_gen; | |
49303381 LB |
10026 | u64 feature; |
10027 | int mixed; | |
10028 | ||
10029 | feature = btrfs_super_incompat_flags(info->super_copy); | |
10030 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 10031 | |
9078a3e1 | 10032 | key.objectid = 0; |
0b86a832 | 10033 | key.offset = 0; |
962a298f | 10034 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
10035 | path = btrfs_alloc_path(); |
10036 | if (!path) | |
10037 | return -ENOMEM; | |
e4058b54 | 10038 | path->reada = READA_FORWARD; |
9078a3e1 | 10039 | |
0b246afa JM |
10040 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
10041 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
10042 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 10043 | need_clear = 1; |
0b246afa | 10044 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 10045 | need_clear = 1; |
0af3d00b | 10046 | |
d397712b | 10047 | while (1) { |
6bccf3ab | 10048 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
10049 | if (ret > 0) |
10050 | break; | |
0b86a832 CM |
10051 | if (ret != 0) |
10052 | goto error; | |
920e4a58 | 10053 | |
5f39d397 CM |
10054 | leaf = path->nodes[0]; |
10055 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10056 | |
2ff7e61e | 10057 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10058 | found_key.offset); |
9078a3e1 | 10059 | if (!cache) { |
0b86a832 | 10060 | ret = -ENOMEM; |
f0486c68 | 10061 | goto error; |
9078a3e1 | 10062 | } |
96303081 | 10063 | |
cf7c1ef6 LB |
10064 | if (need_clear) { |
10065 | /* | |
10066 | * When we mount with old space cache, we need to | |
10067 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10068 | * | |
10069 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10070 | * truncate the old free space cache inode and | |
10071 | * setup a new one. | |
10072 | * b) Setting 'dirty flag' makes sure that we flush | |
10073 | * the new space cache info onto disk. | |
10074 | */ | |
0b246afa | 10075 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10076 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10077 | } |
0af3d00b | 10078 | |
5f39d397 CM |
10079 | read_extent_buffer(leaf, &cache->item, |
10080 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10081 | sizeof(cache->item)); | |
920e4a58 | 10082 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10083 | if (!mixed && |
10084 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10085 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10086 | btrfs_err(info, | |
10087 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10088 | cache->key.objectid); | |
10089 | ret = -EINVAL; | |
10090 | goto error; | |
10091 | } | |
0b86a832 | 10092 | |
9078a3e1 | 10093 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10094 | btrfs_release_path(path); |
34d52cb6 | 10095 | |
3c14874a JB |
10096 | /* |
10097 | * We need to exclude the super stripes now so that the space | |
10098 | * info has super bytes accounted for, otherwise we'll think | |
10099 | * we have more space than we actually do. | |
10100 | */ | |
2ff7e61e | 10101 | ret = exclude_super_stripes(info, cache); |
835d974f JB |
10102 | if (ret) { |
10103 | /* | |
10104 | * We may have excluded something, so call this just in | |
10105 | * case. | |
10106 | */ | |
2ff7e61e | 10107 | free_excluded_extents(info, cache); |
920e4a58 | 10108 | btrfs_put_block_group(cache); |
835d974f JB |
10109 | goto error; |
10110 | } | |
3c14874a | 10111 | |
817d52f8 JB |
10112 | /* |
10113 | * check for two cases, either we are full, and therefore | |
10114 | * don't need to bother with the caching work since we won't | |
10115 | * find any space, or we are empty, and we can just add all | |
10116 | * the space in and be done with it. This saves us _alot_ of | |
10117 | * time, particularly in the full case. | |
10118 | */ | |
10119 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10120 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10121 | cache->cached = BTRFS_CACHE_FINISHED; |
2ff7e61e | 10122 | free_excluded_extents(info, cache); |
817d52f8 | 10123 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10124 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10125 | cache->cached = BTRFS_CACHE_FINISHED; |
4457c1c7 | 10126 | add_new_free_space(cache, found_key.objectid, |
817d52f8 JB |
10127 | found_key.objectid + |
10128 | found_key.offset); | |
2ff7e61e | 10129 | free_excluded_extents(info, cache); |
817d52f8 | 10130 | } |
96b5179d | 10131 | |
0b246afa | 10132 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10133 | if (ret) { |
10134 | btrfs_remove_free_space_cache(cache); | |
10135 | btrfs_put_block_group(cache); | |
10136 | goto error; | |
10137 | } | |
10138 | ||
0b246afa | 10139 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10140 | update_space_info(info, cache->flags, found_key.offset, |
10141 | btrfs_block_group_used(&cache->item), | |
10142 | cache->bytes_super, &space_info); | |
8c579fe7 | 10143 | |
6324fbf3 | 10144 | cache->space_info = space_info; |
1b2da372 | 10145 | |
c434d21c | 10146 | link_block_group(cache); |
0f9dd46c | 10147 | |
0b246afa | 10148 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10149 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10150 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10151 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10152 | spin_lock(&info->unused_bgs_lock); | |
10153 | /* Should always be true but just in case. */ | |
10154 | if (list_empty(&cache->bg_list)) { | |
10155 | btrfs_get_block_group(cache); | |
4ed0a7a3 | 10156 | trace_btrfs_add_unused_block_group(cache); |
47ab2a6c JB |
10157 | list_add_tail(&cache->bg_list, |
10158 | &info->unused_bgs); | |
10159 | } | |
10160 | spin_unlock(&info->unused_bgs_lock); | |
10161 | } | |
9078a3e1 | 10162 | } |
b742bb82 | 10163 | |
0b246afa | 10164 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10165 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10166 | (BTRFS_BLOCK_GROUP_RAID10 | |
10167 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10168 | BTRFS_BLOCK_GROUP_RAID5 | |
10169 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10170 | BTRFS_BLOCK_GROUP_DUP))) |
10171 | continue; | |
10172 | /* | |
10173 | * avoid allocating from un-mirrored block group if there are | |
10174 | * mirrored block groups. | |
10175 | */ | |
1095cc0d | 10176 | list_for_each_entry(cache, |
10177 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10178 | list) | |
868f401a | 10179 | inc_block_group_ro(cache, 1); |
1095cc0d | 10180 | list_for_each_entry(cache, |
10181 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10182 | list) | |
868f401a | 10183 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10184 | } |
f0486c68 | 10185 | |
75cb379d | 10186 | btrfs_add_raid_kobjects(info); |
f0486c68 | 10187 | init_global_block_rsv(info); |
0b86a832 CM |
10188 | ret = 0; |
10189 | error: | |
9078a3e1 | 10190 | btrfs_free_path(path); |
0b86a832 | 10191 | return ret; |
9078a3e1 | 10192 | } |
6324fbf3 | 10193 | |
6c686b35 | 10194 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 10195 | { |
6c686b35 | 10196 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea658bad | 10197 | struct btrfs_block_group_cache *block_group, *tmp; |
0b246afa | 10198 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10199 | struct btrfs_block_group_item item; |
10200 | struct btrfs_key key; | |
10201 | int ret = 0; | |
d9a0540a | 10202 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10203 | |
d9a0540a | 10204 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10205 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10206 | if (ret) |
c92f6be3 | 10207 | goto next; |
ea658bad JB |
10208 | |
10209 | spin_lock(&block_group->lock); | |
10210 | memcpy(&item, &block_group->item, sizeof(item)); | |
10211 | memcpy(&key, &block_group->key, sizeof(key)); | |
10212 | spin_unlock(&block_group->lock); | |
10213 | ||
10214 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10215 | sizeof(item)); | |
10216 | if (ret) | |
66642832 | 10217 | btrfs_abort_transaction(trans, ret); |
0b246afa JM |
10218 | ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid, |
10219 | key.offset); | |
6df9a95e | 10220 | if (ret) |
66642832 | 10221 | btrfs_abort_transaction(trans, ret); |
e4e0711c | 10222 | add_block_group_free_space(trans, block_group); |
1e144fb8 | 10223 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10224 | next: |
10225 | list_del_init(&block_group->bg_list); | |
ea658bad | 10226 | } |
d9a0540a | 10227 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10228 | } |
10229 | ||
6324fbf3 | 10230 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 10231 | struct btrfs_fs_info *fs_info, u64 bytes_used, |
0174484d | 10232 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10233 | { |
6324fbf3 | 10234 | struct btrfs_block_group_cache *cache; |
0b246afa | 10235 | int ret; |
6324fbf3 | 10236 | |
0b246afa | 10237 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10238 | |
2ff7e61e | 10239 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10240 | if (!cache) |
10241 | return -ENOMEM; | |
34d52cb6 | 10242 | |
6324fbf3 | 10243 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10244 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10245 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10246 | btrfs_set_block_group_flags(&cache->item, type); |
10247 | ||
920e4a58 | 10248 | cache->flags = type; |
11833d66 | 10249 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10250 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10251 | cache->needs_free_space = 1; |
2ff7e61e | 10252 | ret = exclude_super_stripes(fs_info, cache); |
835d974f JB |
10253 | if (ret) { |
10254 | /* | |
10255 | * We may have excluded something, so call this just in | |
10256 | * case. | |
10257 | */ | |
2ff7e61e | 10258 | free_excluded_extents(fs_info, cache); |
920e4a58 | 10259 | btrfs_put_block_group(cache); |
835d974f JB |
10260 | return ret; |
10261 | } | |
96303081 | 10262 | |
4457c1c7 | 10263 | add_new_free_space(cache, chunk_offset, chunk_offset + size); |
817d52f8 | 10264 | |
2ff7e61e | 10265 | free_excluded_extents(fs_info, cache); |
11833d66 | 10266 | |
d0bd4560 | 10267 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10268 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10269 | u64 new_bytes_used = size - bytes_used; |
10270 | ||
10271 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10272 | fragment_free_space(cache); |
d0bd4560 JB |
10273 | } |
10274 | #endif | |
2e6e5183 | 10275 | /* |
2be12ef7 NB |
10276 | * Ensure the corresponding space_info object is created and |
10277 | * assigned to our block group. We want our bg to be added to the rbtree | |
10278 | * with its ->space_info set. | |
2e6e5183 | 10279 | */ |
2be12ef7 | 10280 | cache->space_info = __find_space_info(fs_info, cache->flags); |
dc2d3005 | 10281 | ASSERT(cache->space_info); |
2e6e5183 | 10282 | |
0b246afa | 10283 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10284 | if (ret) { |
10285 | btrfs_remove_free_space_cache(cache); | |
10286 | btrfs_put_block_group(cache); | |
10287 | return ret; | |
10288 | } | |
10289 | ||
2e6e5183 FM |
10290 | /* |
10291 | * Now that our block group has its ->space_info set and is inserted in | |
10292 | * the rbtree, update the space info's counters. | |
10293 | */ | |
0b246afa | 10294 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10295 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10296 | cache->bytes_super, &cache->space_info); |
0b246afa | 10297 | update_global_block_rsv(fs_info); |
1b2da372 | 10298 | |
c434d21c | 10299 | link_block_group(cache); |
6324fbf3 | 10300 | |
47ab2a6c | 10301 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10302 | |
0b246afa | 10303 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10304 | return 0; |
10305 | } | |
1a40e23b | 10306 | |
10ea00f5 ID |
10307 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10308 | { | |
899c81ea ID |
10309 | u64 extra_flags = chunk_to_extended(flags) & |
10310 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10311 | |
de98ced9 | 10312 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10313 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10314 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10315 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10316 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10317 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10318 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10319 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10320 | } |
10321 | ||
1a40e23b | 10322 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
6bccf3ab | 10323 | struct btrfs_fs_info *fs_info, u64 group_start, |
04216820 | 10324 | struct extent_map *em) |
1a40e23b | 10325 | { |
6bccf3ab | 10326 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10327 | struct btrfs_path *path; |
10328 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10329 | struct btrfs_free_cluster *cluster; |
0b246afa | 10330 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10331 | struct btrfs_key key; |
0af3d00b | 10332 | struct inode *inode; |
c1895442 | 10333 | struct kobject *kobj = NULL; |
1a40e23b | 10334 | int ret; |
10ea00f5 | 10335 | int index; |
89a55897 | 10336 | int factor; |
4f69cb98 | 10337 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10338 | bool remove_em; |
1a40e23b | 10339 | |
6bccf3ab | 10340 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10341 | BUG_ON(!block_group); |
c146afad | 10342 | BUG_ON(!block_group->ro); |
1a40e23b | 10343 | |
4ed0a7a3 | 10344 | trace_btrfs_remove_block_group(block_group); |
9f7c43c9 | 10345 | /* |
10346 | * Free the reserved super bytes from this block group before | |
10347 | * remove it. | |
10348 | */ | |
2ff7e61e | 10349 | free_excluded_extents(fs_info, block_group); |
fd708b81 JB |
10350 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10351 | block_group->key.offset); | |
9f7c43c9 | 10352 | |
1a40e23b | 10353 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 10354 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
89a55897 JB |
10355 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10356 | BTRFS_BLOCK_GROUP_RAID1 | | |
10357 | BTRFS_BLOCK_GROUP_RAID10)) | |
10358 | factor = 2; | |
10359 | else | |
10360 | factor = 1; | |
1a40e23b | 10361 | |
44fb5511 | 10362 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10363 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10364 | spin_lock(&cluster->refill_lock); |
10365 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10366 | spin_unlock(&cluster->refill_lock); | |
10367 | ||
10368 | /* | |
10369 | * make sure this block group isn't part of a metadata | |
10370 | * allocation cluster | |
10371 | */ | |
0b246afa | 10372 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10373 | spin_lock(&cluster->refill_lock); |
10374 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10375 | spin_unlock(&cluster->refill_lock); | |
10376 | ||
1a40e23b | 10377 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10378 | if (!path) { |
10379 | ret = -ENOMEM; | |
10380 | goto out; | |
10381 | } | |
1a40e23b | 10382 | |
1bbc621e CM |
10383 | /* |
10384 | * get the inode first so any iput calls done for the io_list | |
10385 | * aren't the final iput (no unlinks allowed now) | |
10386 | */ | |
77ab86bf | 10387 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10388 | |
10389 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10390 | /* | |
10391 | * make sure our free spache cache IO is done before remove the | |
10392 | * free space inode | |
10393 | */ | |
10394 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10395 | if (!list_empty(&block_group->io_list)) { | |
10396 | list_del_init(&block_group->io_list); | |
10397 | ||
10398 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10399 | ||
10400 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10401 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10402 | btrfs_put_block_group(block_group); |
10403 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10404 | } | |
10405 | ||
10406 | if (!list_empty(&block_group->dirty_list)) { | |
10407 | list_del_init(&block_group->dirty_list); | |
10408 | btrfs_put_block_group(block_group); | |
10409 | } | |
10410 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10411 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10412 | ||
0af3d00b | 10413 | if (!IS_ERR(inode)) { |
73f2e545 | 10414 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10415 | if (ret) { |
10416 | btrfs_add_delayed_iput(inode); | |
10417 | goto out; | |
10418 | } | |
0af3d00b JB |
10419 | clear_nlink(inode); |
10420 | /* One for the block groups ref */ | |
10421 | spin_lock(&block_group->lock); | |
10422 | if (block_group->iref) { | |
10423 | block_group->iref = 0; | |
10424 | block_group->inode = NULL; | |
10425 | spin_unlock(&block_group->lock); | |
10426 | iput(inode); | |
10427 | } else { | |
10428 | spin_unlock(&block_group->lock); | |
10429 | } | |
10430 | /* One for our lookup ref */ | |
455757c3 | 10431 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10432 | } |
10433 | ||
10434 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10435 | key.offset = block_group->key.objectid; | |
10436 | key.type = 0; | |
10437 | ||
10438 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10439 | if (ret < 0) | |
10440 | goto out; | |
10441 | if (ret > 0) | |
b3b4aa74 | 10442 | btrfs_release_path(path); |
0af3d00b JB |
10443 | if (ret == 0) { |
10444 | ret = btrfs_del_item(trans, tree_root, path); | |
10445 | if (ret) | |
10446 | goto out; | |
b3b4aa74 | 10447 | btrfs_release_path(path); |
0af3d00b JB |
10448 | } |
10449 | ||
0b246afa | 10450 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10451 | rb_erase(&block_group->cache_node, |
0b246afa | 10452 | &fs_info->block_group_cache_tree); |
292cbd51 | 10453 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10454 | |
0b246afa JM |
10455 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10456 | fs_info->first_logical_byte = (u64)-1; | |
10457 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10458 | |
80eb234a | 10459 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10460 | /* |
10461 | * we must use list_del_init so people can check to see if they | |
10462 | * are still on the list after taking the semaphore | |
10463 | */ | |
10464 | list_del_init(&block_group->list); | |
6ab0a202 | 10465 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10466 | kobj = block_group->space_info->block_group_kobjs[index]; |
10467 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10468 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10469 | } |
80eb234a | 10470 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10471 | if (kobj) { |
10472 | kobject_del(kobj); | |
10473 | kobject_put(kobj); | |
10474 | } | |
1a40e23b | 10475 | |
4f69cb98 FM |
10476 | if (block_group->has_caching_ctl) |
10477 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10478 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10479 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10480 | if (block_group->has_caching_ctl) { |
0b246afa | 10481 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10482 | if (!caching_ctl) { |
10483 | struct btrfs_caching_control *ctl; | |
10484 | ||
10485 | list_for_each_entry(ctl, | |
0b246afa | 10486 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10487 | if (ctl->block_group == block_group) { |
10488 | caching_ctl = ctl; | |
1e4f4714 | 10489 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10490 | break; |
10491 | } | |
10492 | } | |
10493 | if (caching_ctl) | |
10494 | list_del_init(&caching_ctl->list); | |
0b246afa | 10495 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10496 | if (caching_ctl) { |
10497 | /* Once for the caching bgs list and once for us. */ | |
10498 | put_caching_control(caching_ctl); | |
10499 | put_caching_control(caching_ctl); | |
10500 | } | |
10501 | } | |
817d52f8 | 10502 | |
ce93ec54 JB |
10503 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10504 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10505 | WARN_ON(1); |
10506 | } | |
10507 | if (!list_empty(&block_group->io_list)) { | |
10508 | WARN_ON(1); | |
ce93ec54 JB |
10509 | } |
10510 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10511 | btrfs_remove_free_space_cache(block_group); |
10512 | ||
c146afad | 10513 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10514 | list_del_init(&block_group->ro_list); |
18d018ad | 10515 | |
0b246afa | 10516 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10517 | WARN_ON(block_group->space_info->total_bytes |
10518 | < block_group->key.offset); | |
10519 | WARN_ON(block_group->space_info->bytes_readonly | |
10520 | < block_group->key.offset); | |
10521 | WARN_ON(block_group->space_info->disk_total | |
10522 | < block_group->key.offset * factor); | |
10523 | } | |
c146afad YZ |
10524 | block_group->space_info->total_bytes -= block_group->key.offset; |
10525 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10526 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10527 | |
c146afad | 10528 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10529 | |
0af3d00b JB |
10530 | memcpy(&key, &block_group->key, sizeof(key)); |
10531 | ||
34441361 | 10532 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10533 | if (!list_empty(&em->list)) { |
10534 | /* We're in the transaction->pending_chunks list. */ | |
10535 | free_extent_map(em); | |
10536 | } | |
04216820 FM |
10537 | spin_lock(&block_group->lock); |
10538 | block_group->removed = 1; | |
10539 | /* | |
10540 | * At this point trimming can't start on this block group, because we | |
10541 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10542 | * so no one can't find it anymore and even if someone already got this | |
10543 | * block group before we removed it from the rbtree, they have already | |
10544 | * incremented block_group->trimming - if they didn't, they won't find | |
10545 | * any free space entries because we already removed them all when we | |
10546 | * called btrfs_remove_free_space_cache(). | |
10547 | * | |
10548 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10549 | * to prevent the same logical address range and physical device space | |
10550 | * ranges from being reused for a new block group. This is because our | |
10551 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10552 | * completely transactionless, so while it is trimming a range the | |
10553 | * currently running transaction might finish and a new one start, | |
10554 | * allowing for new block groups to be created that can reuse the same | |
10555 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10556 | * |
10557 | * There may also be an implicit trim operation if the file system | |
10558 | * is mounted with -odiscard. The same protections must remain | |
10559 | * in place until the extents have been discarded completely when | |
10560 | * the transaction commit has completed. | |
04216820 FM |
10561 | */ |
10562 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10563 | /* | |
10564 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10565 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10566 | * before checking block_group->removed). | |
10567 | */ | |
10568 | if (!remove_em) { | |
10569 | /* | |
10570 | * Our em might be in trans->transaction->pending_chunks which | |
10571 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10572 | * and so is the fs_info->pinned_chunks list. | |
10573 | * | |
10574 | * So at this point we must be holding the chunk_mutex to avoid | |
10575 | * any races with chunk allocation (more specifically at | |
10576 | * volumes.c:contains_pending_extent()), to ensure it always | |
10577 | * sees the em, either in the pending_chunks list or in the | |
10578 | * pinned_chunks list. | |
10579 | */ | |
0b246afa | 10580 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10581 | } |
10582 | spin_unlock(&block_group->lock); | |
04216820 FM |
10583 | |
10584 | if (remove_em) { | |
10585 | struct extent_map_tree *em_tree; | |
10586 | ||
0b246afa | 10587 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10588 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10589 | /* |
10590 | * The em might be in the pending_chunks list, so make sure the | |
10591 | * chunk mutex is locked, since remove_extent_mapping() will | |
10592 | * delete us from that list. | |
10593 | */ | |
04216820 FM |
10594 | remove_extent_mapping(em_tree, em); |
10595 | write_unlock(&em_tree->lock); | |
10596 | /* once for the tree */ | |
10597 | free_extent_map(em); | |
10598 | } | |
10599 | ||
34441361 | 10600 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10601 | |
f3f72779 | 10602 | ret = remove_block_group_free_space(trans, block_group); |
1e144fb8 OS |
10603 | if (ret) |
10604 | goto out; | |
10605 | ||
fa9c0d79 CM |
10606 | btrfs_put_block_group(block_group); |
10607 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10608 | |
10609 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10610 | if (ret > 0) | |
10611 | ret = -EIO; | |
10612 | if (ret < 0) | |
10613 | goto out; | |
10614 | ||
10615 | ret = btrfs_del_item(trans, root, path); | |
10616 | out: | |
10617 | btrfs_free_path(path); | |
10618 | return ret; | |
10619 | } | |
acce952b | 10620 | |
8eab77ff | 10621 | struct btrfs_trans_handle * |
7fd01182 FM |
10622 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10623 | const u64 chunk_offset) | |
8eab77ff | 10624 | { |
7fd01182 FM |
10625 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10626 | struct extent_map *em; | |
10627 | struct map_lookup *map; | |
10628 | unsigned int num_items; | |
10629 | ||
10630 | read_lock(&em_tree->lock); | |
10631 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10632 | read_unlock(&em_tree->lock); | |
10633 | ASSERT(em && em->start == chunk_offset); | |
10634 | ||
8eab77ff | 10635 | /* |
7fd01182 FM |
10636 | * We need to reserve 3 + N units from the metadata space info in order |
10637 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10638 | * btrfs_remove_block_group()), which are used for: | |
10639 | * | |
8eab77ff FM |
10640 | * 1 unit for adding the free space inode's orphan (located in the tree |
10641 | * of tree roots). | |
7fd01182 FM |
10642 | * 1 unit for deleting the block group item (located in the extent |
10643 | * tree). | |
10644 | * 1 unit for deleting the free space item (located in tree of tree | |
10645 | * roots). | |
10646 | * N units for deleting N device extent items corresponding to each | |
10647 | * stripe (located in the device tree). | |
10648 | * | |
10649 | * In order to remove a block group we also need to reserve units in the | |
10650 | * system space info in order to update the chunk tree (update one or | |
10651 | * more device items and remove one chunk item), but this is done at | |
10652 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10653 | */ |
95617d69 | 10654 | map = em->map_lookup; |
7fd01182 FM |
10655 | num_items = 3 + map->num_stripes; |
10656 | free_extent_map(em); | |
10657 | ||
8eab77ff | 10658 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10659 | num_items, 1); |
8eab77ff FM |
10660 | } |
10661 | ||
47ab2a6c JB |
10662 | /* |
10663 | * Process the unused_bgs list and remove any that don't have any allocated | |
10664 | * space inside of them. | |
10665 | */ | |
10666 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10667 | { | |
10668 | struct btrfs_block_group_cache *block_group; | |
10669 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10670 | struct btrfs_trans_handle *trans; |
10671 | int ret = 0; | |
10672 | ||
afcdd129 | 10673 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10674 | return; |
10675 | ||
10676 | spin_lock(&fs_info->unused_bgs_lock); | |
10677 | while (!list_empty(&fs_info->unused_bgs)) { | |
10678 | u64 start, end; | |
e33e17ee | 10679 | int trimming; |
47ab2a6c JB |
10680 | |
10681 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10682 | struct btrfs_block_group_cache, | |
10683 | bg_list); | |
47ab2a6c | 10684 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10685 | |
10686 | space_info = block_group->space_info; | |
10687 | ||
47ab2a6c JB |
10688 | if (ret || btrfs_mixed_space_info(space_info)) { |
10689 | btrfs_put_block_group(block_group); | |
10690 | continue; | |
10691 | } | |
10692 | spin_unlock(&fs_info->unused_bgs_lock); | |
10693 | ||
d5f2e33b | 10694 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10695 | |
47ab2a6c JB |
10696 | /* Don't want to race with allocators so take the groups_sem */ |
10697 | down_write(&space_info->groups_sem); | |
10698 | spin_lock(&block_group->lock); | |
10699 | if (block_group->reserved || | |
10700 | btrfs_block_group_used(&block_group->item) || | |
19c4d2f9 | 10701 | block_group->ro || |
aefbe9a6 | 10702 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10703 | /* |
10704 | * We want to bail if we made new allocations or have | |
10705 | * outstanding allocations in this block group. We do | |
10706 | * the ro check in case balance is currently acting on | |
10707 | * this block group. | |
10708 | */ | |
4ed0a7a3 | 10709 | trace_btrfs_skip_unused_block_group(block_group); |
47ab2a6c JB |
10710 | spin_unlock(&block_group->lock); |
10711 | up_write(&space_info->groups_sem); | |
10712 | goto next; | |
10713 | } | |
10714 | spin_unlock(&block_group->lock); | |
10715 | ||
10716 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10717 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10718 | up_write(&space_info->groups_sem); |
10719 | if (ret < 0) { | |
10720 | ret = 0; | |
10721 | goto next; | |
10722 | } | |
10723 | ||
10724 | /* | |
10725 | * Want to do this before we do anything else so we can recover | |
10726 | * properly if we fail to join the transaction. | |
10727 | */ | |
7fd01182 FM |
10728 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10729 | block_group->key.objectid); | |
47ab2a6c | 10730 | if (IS_ERR(trans)) { |
2ff7e61e | 10731 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10732 | ret = PTR_ERR(trans); |
10733 | goto next; | |
10734 | } | |
10735 | ||
10736 | /* | |
10737 | * We could have pending pinned extents for this block group, | |
10738 | * just delete them, we don't care about them anymore. | |
10739 | */ | |
10740 | start = block_group->key.objectid; | |
10741 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10742 | /* |
10743 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10744 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10745 | * another task might be running finish_extent_commit() for the | |
10746 | * previous transaction N - 1, and have seen a range belonging | |
10747 | * to the block group in freed_extents[] before we were able to | |
10748 | * clear the whole block group range from freed_extents[]. This | |
10749 | * means that task can lookup for the block group after we | |
10750 | * unpinned it from freed_extents[] and removed it, leading to | |
10751 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10752 | */ | |
10753 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10754 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10755 | EXTENT_DIRTY); |
758eb51e | 10756 | if (ret) { |
d4b450cd | 10757 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10758 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10759 | goto end_trans; |
10760 | } | |
10761 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10762 | EXTENT_DIRTY); |
758eb51e | 10763 | if (ret) { |
d4b450cd | 10764 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10765 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10766 | goto end_trans; |
10767 | } | |
d4b450cd | 10768 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10769 | |
10770 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10771 | spin_lock(&space_info->lock); |
10772 | spin_lock(&block_group->lock); | |
10773 | ||
10774 | space_info->bytes_pinned -= block_group->pinned; | |
10775 | space_info->bytes_readonly += block_group->pinned; | |
10776 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10777 | -block_group->pinned); | |
47ab2a6c JB |
10778 | block_group->pinned = 0; |
10779 | ||
c30666d4 ZL |
10780 | spin_unlock(&block_group->lock); |
10781 | spin_unlock(&space_info->lock); | |
10782 | ||
e33e17ee | 10783 | /* DISCARD can flip during remount */ |
0b246afa | 10784 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10785 | |
10786 | /* Implicit trim during transaction commit. */ | |
10787 | if (trimming) | |
10788 | btrfs_get_block_group_trimming(block_group); | |
10789 | ||
47ab2a6c JB |
10790 | /* |
10791 | * Btrfs_remove_chunk will abort the transaction if things go | |
10792 | * horribly wrong. | |
10793 | */ | |
5b4aacef | 10794 | ret = btrfs_remove_chunk(trans, fs_info, |
47ab2a6c | 10795 | block_group->key.objectid); |
e33e17ee JM |
10796 | |
10797 | if (ret) { | |
10798 | if (trimming) | |
10799 | btrfs_put_block_group_trimming(block_group); | |
10800 | goto end_trans; | |
10801 | } | |
10802 | ||
10803 | /* | |
10804 | * If we're not mounted with -odiscard, we can just forget | |
10805 | * about this block group. Otherwise we'll need to wait | |
10806 | * until transaction commit to do the actual discard. | |
10807 | */ | |
10808 | if (trimming) { | |
348a0013 FM |
10809 | spin_lock(&fs_info->unused_bgs_lock); |
10810 | /* | |
10811 | * A concurrent scrub might have added us to the list | |
10812 | * fs_info->unused_bgs, so use a list_move operation | |
10813 | * to add the block group to the deleted_bgs list. | |
10814 | */ | |
e33e17ee JM |
10815 | list_move(&block_group->bg_list, |
10816 | &trans->transaction->deleted_bgs); | |
348a0013 | 10817 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10818 | btrfs_get_block_group(block_group); |
10819 | } | |
758eb51e | 10820 | end_trans: |
3a45bb20 | 10821 | btrfs_end_transaction(trans); |
47ab2a6c | 10822 | next: |
d5f2e33b | 10823 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10824 | btrfs_put_block_group(block_group); |
10825 | spin_lock(&fs_info->unused_bgs_lock); | |
10826 | } | |
10827 | spin_unlock(&fs_info->unused_bgs_lock); | |
10828 | } | |
10829 | ||
c59021f8 | 10830 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10831 | { | |
10832 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10833 | struct btrfs_super_block *disk_super; |
10834 | u64 features; | |
10835 | u64 flags; | |
10836 | int mixed = 0; | |
c59021f8 | 10837 | int ret; |
10838 | ||
6c41761f | 10839 | disk_super = fs_info->super_copy; |
1aba86d6 | 10840 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10841 | return -EINVAL; |
c59021f8 | 10842 | |
1aba86d6 | 10843 | features = btrfs_super_incompat_flags(disk_super); |
10844 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10845 | mixed = 1; | |
c59021f8 | 10846 | |
1aba86d6 | 10847 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
2be12ef7 | 10848 | ret = create_space_info(fs_info, flags, &space_info); |
c59021f8 | 10849 | if (ret) |
1aba86d6 | 10850 | goto out; |
c59021f8 | 10851 | |
1aba86d6 | 10852 | if (mixed) { |
10853 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10854 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10855 | } else { |
10856 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2be12ef7 | 10857 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10858 | if (ret) |
10859 | goto out; | |
10860 | ||
10861 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10862 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10863 | } |
10864 | out: | |
c59021f8 | 10865 | return ret; |
10866 | } | |
10867 | ||
2ff7e61e JM |
10868 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10869 | u64 start, u64 end) | |
acce952b | 10870 | { |
2ff7e61e | 10871 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10872 | } |
10873 | ||
499f377f JM |
10874 | /* |
10875 | * It used to be that old block groups would be left around forever. | |
10876 | * Iterating over them would be enough to trim unused space. Since we | |
10877 | * now automatically remove them, we also need to iterate over unallocated | |
10878 | * space. | |
10879 | * | |
10880 | * We don't want a transaction for this since the discard may take a | |
10881 | * substantial amount of time. We don't require that a transaction be | |
10882 | * running, but we do need to take a running transaction into account | |
10883 | * to ensure that we're not discarding chunks that were released in | |
10884 | * the current transaction. | |
10885 | * | |
10886 | * Holding the chunks lock will prevent other threads from allocating | |
10887 | * or releasing chunks, but it won't prevent a running transaction | |
10888 | * from committing and releasing the memory that the pending chunks | |
10889 | * list head uses. For that, we need to take a reference to the | |
10890 | * transaction. | |
10891 | */ | |
10892 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10893 | u64 minlen, u64 *trimmed) | |
10894 | { | |
10895 | u64 start = 0, len = 0; | |
10896 | int ret; | |
10897 | ||
10898 | *trimmed = 0; | |
10899 | ||
10900 | /* Not writeable = nothing to do. */ | |
ebbede42 | 10901 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
10902 | return 0; |
10903 | ||
10904 | /* No free space = nothing to do. */ | |
10905 | if (device->total_bytes <= device->bytes_used) | |
10906 | return 0; | |
10907 | ||
10908 | ret = 0; | |
10909 | ||
10910 | while (1) { | |
fb456252 | 10911 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10912 | struct btrfs_transaction *trans; |
10913 | u64 bytes; | |
10914 | ||
10915 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10916 | if (ret) | |
10917 | return ret; | |
10918 | ||
10919 | down_read(&fs_info->commit_root_sem); | |
10920 | ||
10921 | spin_lock(&fs_info->trans_lock); | |
10922 | trans = fs_info->running_transaction; | |
10923 | if (trans) | |
9b64f57d | 10924 | refcount_inc(&trans->use_count); |
499f377f JM |
10925 | spin_unlock(&fs_info->trans_lock); |
10926 | ||
10927 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10928 | &start, &len); | |
10929 | if (trans) | |
10930 | btrfs_put_transaction(trans); | |
10931 | ||
10932 | if (ret) { | |
10933 | up_read(&fs_info->commit_root_sem); | |
10934 | mutex_unlock(&fs_info->chunk_mutex); | |
10935 | if (ret == -ENOSPC) | |
10936 | ret = 0; | |
10937 | break; | |
10938 | } | |
10939 | ||
10940 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10941 | up_read(&fs_info->commit_root_sem); | |
10942 | mutex_unlock(&fs_info->chunk_mutex); | |
10943 | ||
10944 | if (ret) | |
10945 | break; | |
10946 | ||
10947 | start += len; | |
10948 | *trimmed += bytes; | |
10949 | ||
10950 | if (fatal_signal_pending(current)) { | |
10951 | ret = -ERESTARTSYS; | |
10952 | break; | |
10953 | } | |
10954 | ||
10955 | cond_resched(); | |
10956 | } | |
10957 | ||
10958 | return ret; | |
10959 | } | |
10960 | ||
2ff7e61e | 10961 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 10962 | { |
f7039b1d | 10963 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
10964 | struct btrfs_device *device; |
10965 | struct list_head *devices; | |
f7039b1d LD |
10966 | u64 group_trimmed; |
10967 | u64 start; | |
10968 | u64 end; | |
10969 | u64 trimmed = 0; | |
2cac13e4 | 10970 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10971 | int ret = 0; |
10972 | ||
2cac13e4 LB |
10973 | /* |
10974 | * try to trim all FS space, our block group may start from non-zero. | |
10975 | */ | |
10976 | if (range->len == total_bytes) | |
10977 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10978 | else | |
10979 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10980 | |
10981 | while (cache) { | |
10982 | if (cache->key.objectid >= (range->start + range->len)) { | |
10983 | btrfs_put_block_group(cache); | |
10984 | break; | |
10985 | } | |
10986 | ||
10987 | start = max(range->start, cache->key.objectid); | |
10988 | end = min(range->start + range->len, | |
10989 | cache->key.objectid + cache->key.offset); | |
10990 | ||
10991 | if (end - start >= range->minlen) { | |
10992 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10993 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
10994 | if (ret) { |
10995 | btrfs_put_block_group(cache); | |
10996 | break; | |
10997 | } | |
10998 | ret = wait_block_group_cache_done(cache); | |
10999 | if (ret) { | |
11000 | btrfs_put_block_group(cache); | |
11001 | break; | |
11002 | } | |
f7039b1d LD |
11003 | } |
11004 | ret = btrfs_trim_block_group(cache, | |
11005 | &group_trimmed, | |
11006 | start, | |
11007 | end, | |
11008 | range->minlen); | |
11009 | ||
11010 | trimmed += group_trimmed; | |
11011 | if (ret) { | |
11012 | btrfs_put_block_group(cache); | |
11013 | break; | |
11014 | } | |
11015 | } | |
11016 | ||
2ff7e61e | 11017 | cache = next_block_group(fs_info, cache); |
f7039b1d LD |
11018 | } |
11019 | ||
0b246afa JM |
11020 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
11021 | devices = &fs_info->fs_devices->alloc_list; | |
499f377f JM |
11022 | list_for_each_entry(device, devices, dev_alloc_list) { |
11023 | ret = btrfs_trim_free_extents(device, range->minlen, | |
11024 | &group_trimmed); | |
11025 | if (ret) | |
11026 | break; | |
11027 | ||
11028 | trimmed += group_trimmed; | |
11029 | } | |
0b246afa | 11030 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 11031 | |
f7039b1d LD |
11032 | range->len = trimmed; |
11033 | return ret; | |
11034 | } | |
8257b2dc MX |
11035 | |
11036 | /* | |
ea14b57f | 11037 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
11038 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
11039 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11040 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 11041 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 11042 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 11043 | */ |
ea14b57f | 11044 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
11045 | { |
11046 | percpu_counter_dec(&root->subv_writers->counter); | |
093258e6 | 11047 | cond_wake_up(&root->subv_writers->wait); |
8257b2dc MX |
11048 | } |
11049 | ||
ea14b57f | 11050 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 11051 | { |
ea14b57f | 11052 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
11053 | return 0; |
11054 | ||
11055 | percpu_counter_inc(&root->subv_writers->counter); | |
11056 | /* | |
11057 | * Make sure counter is updated before we check for snapshot creation. | |
11058 | */ | |
11059 | smp_mb(); | |
ea14b57f DS |
11060 | if (atomic_read(&root->will_be_snapshotted)) { |
11061 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
11062 | return 0; |
11063 | } | |
11064 | return 1; | |
11065 | } | |
0bc19f90 | 11066 | |
0bc19f90 ZL |
11067 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
11068 | { | |
11069 | while (true) { | |
11070 | int ret; | |
11071 | ||
ea14b57f | 11072 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
11073 | if (ret) |
11074 | break; | |
4625956a PZ |
11075 | wait_var_event(&root->will_be_snapshotted, |
11076 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
11077 | } |
11078 | } |