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" |
8719aaae | 31 | #include "space-info.h" |
d12ffdd1 | 32 | #include "block-rsv.h" |
fec577fb | 33 | |
709c0486 AJ |
34 | #undef SCRAMBLE_DELAYED_REFS |
35 | ||
9f9b8e8d | 36 | |
5d4f98a2 | 37 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
38 | struct btrfs_delayed_ref_node *node, u64 parent, |
39 | u64 root_objectid, u64 owner_objectid, | |
40 | u64 owner_offset, int refs_to_drop, | |
41 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
42 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
43 | struct extent_buffer *leaf, | |
44 | struct btrfs_extent_item *ei); | |
45 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
46 | u64 parent, u64 root_objectid, |
47 | u64 flags, u64 owner, u64 offset, | |
48 | struct btrfs_key *ins, int ref_mod); | |
49 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 50 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 51 | struct btrfs_delayed_extent_op *extent_op); |
11833d66 YZ |
52 | static int find_next_key(struct btrfs_path *path, int level, |
53 | struct btrfs_key *key); | |
6a63209f | 54 | |
817d52f8 JB |
55 | static noinline int |
56 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
57 | { | |
58 | smp_mb(); | |
36cce922 JB |
59 | return cache->cached == BTRFS_CACHE_FINISHED || |
60 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
61 | } |
62 | ||
0f9dd46c JB |
63 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
64 | { | |
65 | return (cache->flags & bits) == bits; | |
66 | } | |
67 | ||
758f2dfc | 68 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
69 | { |
70 | atomic_inc(&cache->count); | |
71 | } | |
72 | ||
73 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
74 | { | |
f0486c68 YZ |
75 | if (atomic_dec_and_test(&cache->count)) { |
76 | WARN_ON(cache->pinned > 0); | |
77 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
78 | |
79 | /* | |
80 | * If not empty, someone is still holding mutex of | |
81 | * full_stripe_lock, which can only be released by caller. | |
82 | * And it will definitely cause use-after-free when caller | |
83 | * tries to release full stripe lock. | |
84 | * | |
85 | * No better way to resolve, but only to warn. | |
86 | */ | |
87 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 88 | kfree(cache->free_space_ctl); |
11dfe35a | 89 | kfree(cache); |
f0486c68 | 90 | } |
11dfe35a JB |
91 | } |
92 | ||
0f9dd46c JB |
93 | /* |
94 | * this adds the block group to the fs_info rb tree for the block group | |
95 | * cache | |
96 | */ | |
b2950863 | 97 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
98 | struct btrfs_block_group_cache *block_group) |
99 | { | |
100 | struct rb_node **p; | |
101 | struct rb_node *parent = NULL; | |
102 | struct btrfs_block_group_cache *cache; | |
103 | ||
104 | spin_lock(&info->block_group_cache_lock); | |
105 | p = &info->block_group_cache_tree.rb_node; | |
106 | ||
107 | while (*p) { | |
108 | parent = *p; | |
109 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
110 | cache_node); | |
111 | if (block_group->key.objectid < cache->key.objectid) { | |
112 | p = &(*p)->rb_left; | |
113 | } else if (block_group->key.objectid > cache->key.objectid) { | |
114 | p = &(*p)->rb_right; | |
115 | } else { | |
116 | spin_unlock(&info->block_group_cache_lock); | |
117 | return -EEXIST; | |
118 | } | |
119 | } | |
120 | ||
121 | rb_link_node(&block_group->cache_node, parent, p); | |
122 | rb_insert_color(&block_group->cache_node, | |
123 | &info->block_group_cache_tree); | |
a1897fdd LB |
124 | |
125 | if (info->first_logical_byte > block_group->key.objectid) | |
126 | info->first_logical_byte = block_group->key.objectid; | |
127 | ||
0f9dd46c JB |
128 | spin_unlock(&info->block_group_cache_lock); |
129 | ||
130 | return 0; | |
131 | } | |
132 | ||
133 | /* | |
134 | * This will return the block group at or after bytenr if contains is 0, else | |
135 | * it will return the block group that contains the bytenr | |
136 | */ | |
137 | static struct btrfs_block_group_cache * | |
138 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
139 | int contains) | |
140 | { | |
141 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
142 | struct rb_node *n; | |
143 | u64 end, start; | |
144 | ||
145 | spin_lock(&info->block_group_cache_lock); | |
146 | n = info->block_group_cache_tree.rb_node; | |
147 | ||
148 | while (n) { | |
149 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
150 | cache_node); | |
151 | end = cache->key.objectid + cache->key.offset - 1; | |
152 | start = cache->key.objectid; | |
153 | ||
154 | if (bytenr < start) { | |
155 | if (!contains && (!ret || start < ret->key.objectid)) | |
156 | ret = cache; | |
157 | n = n->rb_left; | |
158 | } else if (bytenr > start) { | |
159 | if (contains && bytenr <= end) { | |
160 | ret = cache; | |
161 | break; | |
162 | } | |
163 | n = n->rb_right; | |
164 | } else { | |
165 | ret = cache; | |
166 | break; | |
167 | } | |
168 | } | |
a1897fdd | 169 | if (ret) { |
11dfe35a | 170 | btrfs_get_block_group(ret); |
a1897fdd LB |
171 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
172 | info->first_logical_byte = ret->key.objectid; | |
173 | } | |
0f9dd46c JB |
174 | spin_unlock(&info->block_group_cache_lock); |
175 | ||
176 | return ret; | |
177 | } | |
178 | ||
2ff7e61e | 179 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 180 | u64 start, u64 num_bytes) |
817d52f8 | 181 | { |
11833d66 | 182 | u64 end = start + num_bytes - 1; |
0b246afa | 183 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 184 | start, end, EXTENT_UPTODATE); |
0b246afa | 185 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 186 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
187 | return 0; |
188 | } | |
817d52f8 | 189 | |
9e715da8 | 190 | static void free_excluded_extents(struct btrfs_block_group_cache *cache) |
11833d66 | 191 | { |
9e715da8 | 192 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 193 | u64 start, end; |
817d52f8 | 194 | |
11833d66 YZ |
195 | start = cache->key.objectid; |
196 | end = start + cache->key.offset - 1; | |
197 | ||
0b246afa | 198 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 199 | start, end, EXTENT_UPTODATE); |
0b246afa | 200 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 201 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
202 | } |
203 | ||
3c4da657 | 204 | static int exclude_super_stripes(struct btrfs_block_group_cache *cache) |
817d52f8 | 205 | { |
3c4da657 | 206 | struct btrfs_fs_info *fs_info = cache->fs_info; |
817d52f8 JB |
207 | u64 bytenr; |
208 | u64 *logical; | |
209 | int stripe_len; | |
210 | int i, nr, ret; | |
211 | ||
06b2331f YZ |
212 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
213 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
214 | cache->bytes_super += stripe_len; | |
2ff7e61e | 215 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 216 | stripe_len); |
835d974f JB |
217 | if (ret) |
218 | return ret; | |
06b2331f YZ |
219 | } |
220 | ||
817d52f8 JB |
221 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
222 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 223 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
63a9c7b9 | 224 | bytenr, &logical, &nr, &stripe_len); |
835d974f JB |
225 | if (ret) |
226 | return ret; | |
11833d66 | 227 | |
817d52f8 | 228 | while (nr--) { |
51bf5f0b JB |
229 | u64 start, len; |
230 | ||
231 | if (logical[nr] > cache->key.objectid + | |
232 | cache->key.offset) | |
233 | continue; | |
234 | ||
235 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
236 | continue; | |
237 | ||
238 | start = logical[nr]; | |
239 | if (start < cache->key.objectid) { | |
240 | start = cache->key.objectid; | |
241 | len = (logical[nr] + stripe_len) - start; | |
242 | } else { | |
243 | len = min_t(u64, stripe_len, | |
244 | cache->key.objectid + | |
245 | cache->key.offset - start); | |
246 | } | |
247 | ||
248 | cache->bytes_super += len; | |
2ff7e61e | 249 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
250 | if (ret) { |
251 | kfree(logical); | |
252 | return ret; | |
253 | } | |
817d52f8 | 254 | } |
11833d66 | 255 | |
817d52f8 JB |
256 | kfree(logical); |
257 | } | |
817d52f8 JB |
258 | return 0; |
259 | } | |
260 | ||
11833d66 YZ |
261 | static struct btrfs_caching_control * |
262 | get_caching_control(struct btrfs_block_group_cache *cache) | |
263 | { | |
264 | struct btrfs_caching_control *ctl; | |
265 | ||
266 | spin_lock(&cache->lock); | |
dde5abee JB |
267 | if (!cache->caching_ctl) { |
268 | spin_unlock(&cache->lock); | |
11833d66 YZ |
269 | return NULL; |
270 | } | |
271 | ||
272 | ctl = cache->caching_ctl; | |
1e4f4714 | 273 | refcount_inc(&ctl->count); |
11833d66 YZ |
274 | spin_unlock(&cache->lock); |
275 | return ctl; | |
276 | } | |
277 | ||
278 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
279 | { | |
1e4f4714 | 280 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
281 | kfree(ctl); |
282 | } | |
283 | ||
d0bd4560 | 284 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 285 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 286 | { |
2ff7e61e | 287 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
288 | u64 start = block_group->key.objectid; |
289 | u64 len = block_group->key.offset; | |
290 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 291 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
292 | u64 step = chunk << 1; |
293 | ||
294 | while (len > chunk) { | |
295 | btrfs_remove_free_space(block_group, start, chunk); | |
296 | start += step; | |
297 | if (len < step) | |
298 | len = 0; | |
299 | else | |
300 | len -= step; | |
301 | } | |
302 | } | |
303 | #endif | |
304 | ||
0f9dd46c JB |
305 | /* |
306 | * this is only called by cache_block_group, since we could have freed extents | |
307 | * we need to check the pinned_extents for any extents that can't be used yet | |
308 | * since their free space will be released as soon as the transaction commits. | |
309 | */ | |
a5ed9182 | 310 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
4457c1c7 | 311 | u64 start, u64 end) |
0f9dd46c | 312 | { |
4457c1c7 | 313 | struct btrfs_fs_info *info = block_group->fs_info; |
817d52f8 | 314 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
315 | int ret; |
316 | ||
317 | while (start < end) { | |
11833d66 | 318 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 319 | &extent_start, &extent_end, |
e6138876 JB |
320 | EXTENT_DIRTY | EXTENT_UPTODATE, |
321 | NULL); | |
0f9dd46c JB |
322 | if (ret) |
323 | break; | |
324 | ||
06b2331f | 325 | if (extent_start <= start) { |
0f9dd46c JB |
326 | start = extent_end + 1; |
327 | } else if (extent_start > start && extent_start < end) { | |
328 | size = extent_start - start; | |
817d52f8 | 329 | total_added += size; |
ea6a478e JB |
330 | ret = btrfs_add_free_space(block_group, start, |
331 | size); | |
79787eaa | 332 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
333 | start = extent_end + 1; |
334 | } else { | |
335 | break; | |
336 | } | |
337 | } | |
338 | ||
339 | if (start < end) { | |
340 | size = end - start; | |
817d52f8 | 341 | total_added += size; |
ea6a478e | 342 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 343 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
344 | } |
345 | ||
817d52f8 | 346 | return total_added; |
0f9dd46c JB |
347 | } |
348 | ||
73fa48b6 | 349 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 350 | { |
0b246afa JM |
351 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
352 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
353 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 354 | struct btrfs_path *path; |
5f39d397 | 355 | struct extent_buffer *leaf; |
11833d66 | 356 | struct btrfs_key key; |
817d52f8 | 357 | u64 total_found = 0; |
11833d66 YZ |
358 | u64 last = 0; |
359 | u32 nritems; | |
73fa48b6 | 360 | int ret; |
d0bd4560 | 361 | bool wakeup = true; |
f510cfec | 362 | |
e37c9e69 CM |
363 | path = btrfs_alloc_path(); |
364 | if (!path) | |
73fa48b6 | 365 | return -ENOMEM; |
7d7d6068 | 366 | |
817d52f8 | 367 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 368 | |
d0bd4560 JB |
369 | #ifdef CONFIG_BTRFS_DEBUG |
370 | /* | |
371 | * If we're fragmenting we don't want to make anybody think we can | |
372 | * allocate from this block group until we've had a chance to fragment | |
373 | * the free space. | |
374 | */ | |
2ff7e61e | 375 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
376 | wakeup = false; |
377 | #endif | |
5cd57b2c | 378 | /* |
817d52f8 JB |
379 | * We don't want to deadlock with somebody trying to allocate a new |
380 | * extent for the extent root while also trying to search the extent | |
381 | * root to add free space. So we skip locking and search the commit | |
382 | * root, since its read-only | |
5cd57b2c CM |
383 | */ |
384 | path->skip_locking = 1; | |
817d52f8 | 385 | path->search_commit_root = 1; |
e4058b54 | 386 | path->reada = READA_FORWARD; |
817d52f8 | 387 | |
e4404d6e | 388 | key.objectid = last; |
e37c9e69 | 389 | key.offset = 0; |
11833d66 | 390 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 391 | |
52ee28d2 | 392 | next: |
11833d66 | 393 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 394 | if (ret < 0) |
73fa48b6 | 395 | goto out; |
a512bbf8 | 396 | |
11833d66 YZ |
397 | leaf = path->nodes[0]; |
398 | nritems = btrfs_header_nritems(leaf); | |
399 | ||
d397712b | 400 | while (1) { |
7841cb28 | 401 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 402 | last = (u64)-1; |
817d52f8 | 403 | break; |
f25784b3 | 404 | } |
817d52f8 | 405 | |
11833d66 YZ |
406 | if (path->slots[0] < nritems) { |
407 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
408 | } else { | |
409 | ret = find_next_key(path, 0, &key); | |
410 | if (ret) | |
e37c9e69 | 411 | break; |
817d52f8 | 412 | |
c9ea7b24 | 413 | if (need_resched() || |
9e351cc8 | 414 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
415 | if (wakeup) |
416 | caching_ctl->progress = last; | |
ff5714cc | 417 | btrfs_release_path(path); |
9e351cc8 | 418 | up_read(&fs_info->commit_root_sem); |
589d8ade | 419 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 420 | cond_resched(); |
73fa48b6 OS |
421 | mutex_lock(&caching_ctl->mutex); |
422 | down_read(&fs_info->commit_root_sem); | |
423 | goto next; | |
589d8ade | 424 | } |
0a3896d0 JB |
425 | |
426 | ret = btrfs_next_leaf(extent_root, path); | |
427 | if (ret < 0) | |
73fa48b6 | 428 | goto out; |
0a3896d0 JB |
429 | if (ret) |
430 | break; | |
589d8ade JB |
431 | leaf = path->nodes[0]; |
432 | nritems = btrfs_header_nritems(leaf); | |
433 | continue; | |
11833d66 | 434 | } |
817d52f8 | 435 | |
52ee28d2 LB |
436 | if (key.objectid < last) { |
437 | key.objectid = last; | |
438 | key.offset = 0; | |
439 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
440 | ||
d0bd4560 JB |
441 | if (wakeup) |
442 | caching_ctl->progress = last; | |
52ee28d2 LB |
443 | btrfs_release_path(path); |
444 | goto next; | |
445 | } | |
446 | ||
11833d66 YZ |
447 | if (key.objectid < block_group->key.objectid) { |
448 | path->slots[0]++; | |
817d52f8 | 449 | continue; |
e37c9e69 | 450 | } |
0f9dd46c | 451 | |
e37c9e69 | 452 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 453 | block_group->key.offset) |
e37c9e69 | 454 | break; |
7d7d6068 | 455 | |
3173a18f JB |
456 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
457 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
4457c1c7 | 458 | total_found += add_new_free_space(block_group, last, |
817d52f8 | 459 | key.objectid); |
3173a18f JB |
460 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
461 | last = key.objectid + | |
da17066c | 462 | fs_info->nodesize; |
3173a18f JB |
463 | else |
464 | last = key.objectid + key.offset; | |
817d52f8 | 465 | |
73fa48b6 | 466 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 467 | total_found = 0; |
d0bd4560 JB |
468 | if (wakeup) |
469 | wake_up(&caching_ctl->wait); | |
11833d66 | 470 | } |
817d52f8 | 471 | } |
e37c9e69 CM |
472 | path->slots[0]++; |
473 | } | |
817d52f8 | 474 | ret = 0; |
e37c9e69 | 475 | |
4457c1c7 | 476 | total_found += add_new_free_space(block_group, last, |
817d52f8 JB |
477 | block_group->key.objectid + |
478 | block_group->key.offset); | |
11833d66 | 479 | caching_ctl->progress = (u64)-1; |
817d52f8 | 480 | |
73fa48b6 OS |
481 | out: |
482 | btrfs_free_path(path); | |
483 | return ret; | |
484 | } | |
485 | ||
486 | static noinline void caching_thread(struct btrfs_work *work) | |
487 | { | |
488 | struct btrfs_block_group_cache *block_group; | |
489 | struct btrfs_fs_info *fs_info; | |
490 | struct btrfs_caching_control *caching_ctl; | |
491 | int ret; | |
492 | ||
493 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
494 | block_group = caching_ctl->block_group; | |
495 | fs_info = block_group->fs_info; | |
496 | ||
497 | mutex_lock(&caching_ctl->mutex); | |
498 | down_read(&fs_info->commit_root_sem); | |
499 | ||
1e144fb8 OS |
500 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
501 | ret = load_free_space_tree(caching_ctl); | |
502 | else | |
503 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 504 | |
817d52f8 | 505 | spin_lock(&block_group->lock); |
11833d66 | 506 | block_group->caching_ctl = NULL; |
73fa48b6 | 507 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 508 | spin_unlock(&block_group->lock); |
0f9dd46c | 509 | |
d0bd4560 | 510 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 511 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
512 | u64 bytes_used; |
513 | ||
514 | spin_lock(&block_group->space_info->lock); | |
515 | spin_lock(&block_group->lock); | |
516 | bytes_used = block_group->key.offset - | |
517 | btrfs_block_group_used(&block_group->item); | |
518 | block_group->space_info->bytes_used += bytes_used >> 1; | |
519 | spin_unlock(&block_group->lock); | |
520 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 521 | fragment_free_space(block_group); |
d0bd4560 JB |
522 | } |
523 | #endif | |
524 | ||
525 | caching_ctl->progress = (u64)-1; | |
11833d66 | 526 | |
9e351cc8 | 527 | up_read(&fs_info->commit_root_sem); |
9e715da8 | 528 | free_excluded_extents(block_group); |
11833d66 | 529 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 530 | |
11833d66 YZ |
531 | wake_up(&caching_ctl->wait); |
532 | ||
533 | put_caching_control(caching_ctl); | |
11dfe35a | 534 | btrfs_put_block_group(block_group); |
817d52f8 JB |
535 | } |
536 | ||
9d66e233 | 537 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 538 | int load_cache_only) |
817d52f8 | 539 | { |
291c7d2f | 540 | DEFINE_WAIT(wait); |
11833d66 YZ |
541 | struct btrfs_fs_info *fs_info = cache->fs_info; |
542 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
543 | int ret = 0; |
544 | ||
291c7d2f | 545 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
546 | if (!caching_ctl) |
547 | return -ENOMEM; | |
291c7d2f JB |
548 | |
549 | INIT_LIST_HEAD(&caching_ctl->list); | |
550 | mutex_init(&caching_ctl->mutex); | |
551 | init_waitqueue_head(&caching_ctl->wait); | |
552 | caching_ctl->block_group = cache; | |
553 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 554 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
555 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
556 | caching_thread, NULL, NULL); | |
291c7d2f JB |
557 | |
558 | spin_lock(&cache->lock); | |
559 | /* | |
560 | * This should be a rare occasion, but this could happen I think in the | |
561 | * case where one thread starts to load the space cache info, and then | |
562 | * some other thread starts a transaction commit which tries to do an | |
563 | * allocation while the other thread is still loading the space cache | |
564 | * info. The previous loop should have kept us from choosing this block | |
565 | * group, but if we've moved to the state where we will wait on caching | |
566 | * block groups we need to first check if we're doing a fast load here, | |
567 | * so we can wait for it to finish, otherwise we could end up allocating | |
568 | * from a block group who's cache gets evicted for one reason or | |
569 | * another. | |
570 | */ | |
571 | while (cache->cached == BTRFS_CACHE_FAST) { | |
572 | struct btrfs_caching_control *ctl; | |
573 | ||
574 | ctl = cache->caching_ctl; | |
1e4f4714 | 575 | refcount_inc(&ctl->count); |
291c7d2f JB |
576 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
577 | spin_unlock(&cache->lock); | |
578 | ||
579 | schedule(); | |
580 | ||
581 | finish_wait(&ctl->wait, &wait); | |
582 | put_caching_control(ctl); | |
583 | spin_lock(&cache->lock); | |
584 | } | |
585 | ||
586 | if (cache->cached != BTRFS_CACHE_NO) { | |
587 | spin_unlock(&cache->lock); | |
588 | kfree(caching_ctl); | |
11833d66 | 589 | return 0; |
291c7d2f JB |
590 | } |
591 | WARN_ON(cache->caching_ctl); | |
592 | cache->caching_ctl = caching_ctl; | |
593 | cache->cached = BTRFS_CACHE_FAST; | |
594 | spin_unlock(&cache->lock); | |
11833d66 | 595 | |
d8953d69 | 596 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cb83b7b8 | 597 | mutex_lock(&caching_ctl->mutex); |
bb6cb1c5 | 598 | ret = load_free_space_cache(cache); |
9d66e233 JB |
599 | |
600 | spin_lock(&cache->lock); | |
601 | if (ret == 1) { | |
291c7d2f | 602 | cache->caching_ctl = NULL; |
9d66e233 JB |
603 | cache->cached = BTRFS_CACHE_FINISHED; |
604 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 605 | caching_ctl->progress = (u64)-1; |
9d66e233 | 606 | } else { |
291c7d2f JB |
607 | if (load_cache_only) { |
608 | cache->caching_ctl = NULL; | |
609 | cache->cached = BTRFS_CACHE_NO; | |
610 | } else { | |
611 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 612 | cache->has_caching_ctl = 1; |
291c7d2f | 613 | } |
9d66e233 JB |
614 | } |
615 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
616 | #ifdef CONFIG_BTRFS_DEBUG |
617 | if (ret == 1 && | |
2ff7e61e | 618 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
619 | u64 bytes_used; |
620 | ||
621 | spin_lock(&cache->space_info->lock); | |
622 | spin_lock(&cache->lock); | |
623 | bytes_used = cache->key.offset - | |
624 | btrfs_block_group_used(&cache->item); | |
625 | cache->space_info->bytes_used += bytes_used >> 1; | |
626 | spin_unlock(&cache->lock); | |
627 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 628 | fragment_free_space(cache); |
d0bd4560 JB |
629 | } |
630 | #endif | |
cb83b7b8 JB |
631 | mutex_unlock(&caching_ctl->mutex); |
632 | ||
291c7d2f | 633 | wake_up(&caching_ctl->wait); |
3c14874a | 634 | if (ret == 1) { |
291c7d2f | 635 | put_caching_control(caching_ctl); |
9e715da8 | 636 | free_excluded_extents(cache); |
9d66e233 | 637 | return 0; |
3c14874a | 638 | } |
291c7d2f JB |
639 | } else { |
640 | /* | |
1e144fb8 OS |
641 | * We're either using the free space tree or no caching at all. |
642 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
643 | */ |
644 | spin_lock(&cache->lock); | |
645 | if (load_cache_only) { | |
646 | cache->caching_ctl = NULL; | |
647 | cache->cached = BTRFS_CACHE_NO; | |
648 | } else { | |
649 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 650 | cache->has_caching_ctl = 1; |
291c7d2f JB |
651 | } |
652 | spin_unlock(&cache->lock); | |
653 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
654 | } |
655 | ||
291c7d2f JB |
656 | if (load_cache_only) { |
657 | put_caching_control(caching_ctl); | |
11833d66 | 658 | return 0; |
817d52f8 | 659 | } |
817d52f8 | 660 | |
9e351cc8 | 661 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 662 | refcount_inc(&caching_ctl->count); |
11833d66 | 663 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 664 | up_write(&fs_info->commit_root_sem); |
11833d66 | 665 | |
11dfe35a | 666 | btrfs_get_block_group(cache); |
11833d66 | 667 | |
e66f0bb1 | 668 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 669 | |
ef8bbdfe | 670 | return ret; |
e37c9e69 CM |
671 | } |
672 | ||
0f9dd46c JB |
673 | /* |
674 | * return the block group that starts at or after bytenr | |
675 | */ | |
d397712b CM |
676 | static struct btrfs_block_group_cache * |
677 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 678 | { |
e2c89907 | 679 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
680 | } |
681 | ||
0f9dd46c | 682 | /* |
9f55684c | 683 | * return the block group that contains the given bytenr |
0f9dd46c | 684 | */ |
d397712b CM |
685 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
686 | struct btrfs_fs_info *info, | |
687 | u64 bytenr) | |
be744175 | 688 | { |
e2c89907 | 689 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 690 | } |
0b86a832 | 691 | |
78192442 | 692 | static u64 generic_ref_to_space_flags(struct btrfs_ref *ref) |
0d9f824d | 693 | { |
ddf30cf0 QW |
694 | if (ref->type == BTRFS_REF_METADATA) { |
695 | if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID) | |
78192442 | 696 | return BTRFS_BLOCK_GROUP_SYSTEM; |
0d9f824d | 697 | else |
78192442 | 698 | return BTRFS_BLOCK_GROUP_METADATA; |
0d9f824d | 699 | } |
78192442 QW |
700 | return BTRFS_BLOCK_GROUP_DATA; |
701 | } | |
702 | ||
703 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, | |
704 | struct btrfs_ref *ref) | |
705 | { | |
706 | struct btrfs_space_info *space_info; | |
707 | u64 flags = generic_ref_to_space_flags(ref); | |
708 | ||
280c2908 | 709 | space_info = btrfs_find_space_info(fs_info, flags); |
78192442 QW |
710 | ASSERT(space_info); |
711 | percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len, | |
712 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
713 | } | |
714 | ||
715 | static void sub_pinned_bytes(struct btrfs_fs_info *fs_info, | |
716 | struct btrfs_ref *ref) | |
717 | { | |
718 | struct btrfs_space_info *space_info; | |
719 | u64 flags = generic_ref_to_space_flags(ref); | |
0d9f824d | 720 | |
280c2908 | 721 | space_info = btrfs_find_space_info(fs_info, flags); |
55e8196a | 722 | ASSERT(space_info); |
78192442 | 723 | percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len, |
dec59fa3 | 724 | BTRFS_TOTAL_BYTES_PINNED_BATCH); |
0d9f824d OS |
725 | } |
726 | ||
1a4ed8fd | 727 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 728 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
729 | { |
730 | int ret; | |
731 | struct btrfs_key key; | |
31840ae1 | 732 | struct btrfs_path *path; |
e02119d5 | 733 | |
31840ae1 | 734 | path = btrfs_alloc_path(); |
d8926bb3 MF |
735 | if (!path) |
736 | return -ENOMEM; | |
737 | ||
e02119d5 CM |
738 | key.objectid = start; |
739 | key.offset = len; | |
3173a18f | 740 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 741 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 742 | btrfs_free_path(path); |
7bb86316 CM |
743 | return ret; |
744 | } | |
745 | ||
a22285a6 | 746 | /* |
3173a18f | 747 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
748 | * |
749 | * the head node for delayed ref is used to store the sum of all the | |
750 | * reference count modifications queued up in the rbtree. the head | |
751 | * node may also store the extent flags to set. This way you can check | |
752 | * to see what the reference count and extent flags would be if all of | |
753 | * the delayed refs are not processed. | |
754 | */ | |
755 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 756 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 757 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
758 | { |
759 | struct btrfs_delayed_ref_head *head; | |
760 | struct btrfs_delayed_ref_root *delayed_refs; | |
761 | struct btrfs_path *path; | |
762 | struct btrfs_extent_item *ei; | |
763 | struct extent_buffer *leaf; | |
764 | struct btrfs_key key; | |
765 | u32 item_size; | |
766 | u64 num_refs; | |
767 | u64 extent_flags; | |
768 | int ret; | |
769 | ||
3173a18f JB |
770 | /* |
771 | * If we don't have skinny metadata, don't bother doing anything | |
772 | * different | |
773 | */ | |
0b246afa JM |
774 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
775 | offset = fs_info->nodesize; | |
3173a18f JB |
776 | metadata = 0; |
777 | } | |
778 | ||
a22285a6 YZ |
779 | path = btrfs_alloc_path(); |
780 | if (!path) | |
781 | return -ENOMEM; | |
782 | ||
a22285a6 YZ |
783 | if (!trans) { |
784 | path->skip_locking = 1; | |
785 | path->search_commit_root = 1; | |
786 | } | |
639eefc8 FDBM |
787 | |
788 | search_again: | |
789 | key.objectid = bytenr; | |
790 | key.offset = offset; | |
791 | if (metadata) | |
792 | key.type = BTRFS_METADATA_ITEM_KEY; | |
793 | else | |
794 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
795 | ||
0b246afa | 796 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
797 | if (ret < 0) |
798 | goto out_free; | |
799 | ||
3173a18f | 800 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
801 | if (path->slots[0]) { |
802 | path->slots[0]--; | |
803 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
804 | path->slots[0]); | |
805 | if (key.objectid == bytenr && | |
806 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 807 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
808 | ret = 0; |
809 | } | |
3173a18f JB |
810 | } |
811 | ||
a22285a6 YZ |
812 | if (ret == 0) { |
813 | leaf = path->nodes[0]; | |
814 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
815 | if (item_size >= sizeof(*ei)) { | |
816 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
817 | struct btrfs_extent_item); | |
818 | num_refs = btrfs_extent_refs(leaf, ei); | |
819 | extent_flags = btrfs_extent_flags(leaf, ei); | |
820 | } else { | |
ba3c2b19 NB |
821 | ret = -EINVAL; |
822 | btrfs_print_v0_err(fs_info); | |
823 | if (trans) | |
824 | btrfs_abort_transaction(trans, ret); | |
825 | else | |
826 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
827 | ||
828 | goto out_free; | |
a22285a6 | 829 | } |
ba3c2b19 | 830 | |
a22285a6 YZ |
831 | BUG_ON(num_refs == 0); |
832 | } else { | |
833 | num_refs = 0; | |
834 | extent_flags = 0; | |
835 | ret = 0; | |
836 | } | |
837 | ||
838 | if (!trans) | |
839 | goto out; | |
840 | ||
841 | delayed_refs = &trans->transaction->delayed_refs; | |
842 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 843 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
844 | if (head) { |
845 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 846 | refcount_inc(&head->refs); |
a22285a6 YZ |
847 | spin_unlock(&delayed_refs->lock); |
848 | ||
b3b4aa74 | 849 | btrfs_release_path(path); |
a22285a6 | 850 | |
8cc33e5c DS |
851 | /* |
852 | * Mutex was contended, block until it's released and try | |
853 | * again | |
854 | */ | |
a22285a6 YZ |
855 | mutex_lock(&head->mutex); |
856 | mutex_unlock(&head->mutex); | |
d278850e | 857 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 858 | goto search_again; |
a22285a6 | 859 | } |
d7df2c79 | 860 | spin_lock(&head->lock); |
a22285a6 YZ |
861 | if (head->extent_op && head->extent_op->update_flags) |
862 | extent_flags |= head->extent_op->flags_to_set; | |
863 | else | |
864 | BUG_ON(num_refs == 0); | |
865 | ||
d278850e | 866 | num_refs += head->ref_mod; |
d7df2c79 | 867 | spin_unlock(&head->lock); |
a22285a6 YZ |
868 | mutex_unlock(&head->mutex); |
869 | } | |
870 | spin_unlock(&delayed_refs->lock); | |
871 | out: | |
872 | WARN_ON(num_refs == 0); | |
873 | if (refs) | |
874 | *refs = num_refs; | |
875 | if (flags) | |
876 | *flags = extent_flags; | |
877 | out_free: | |
878 | btrfs_free_path(path); | |
879 | return ret; | |
880 | } | |
881 | ||
d8d5f3e1 CM |
882 | /* |
883 | * Back reference rules. Back refs have three main goals: | |
884 | * | |
885 | * 1) differentiate between all holders of references to an extent so that | |
886 | * when a reference is dropped we can make sure it was a valid reference | |
887 | * before freeing the extent. | |
888 | * | |
889 | * 2) Provide enough information to quickly find the holders of an extent | |
890 | * if we notice a given block is corrupted or bad. | |
891 | * | |
892 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
893 | * maintenance. This is actually the same as #2, but with a slightly | |
894 | * different use case. | |
895 | * | |
5d4f98a2 YZ |
896 | * There are two kinds of back refs. The implicit back refs is optimized |
897 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
898 | * back refs of this kind provide information about the block's owner tree | |
899 | * and the pointer's key. These information allow us to find the block by | |
900 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
901 | * referenced by their owner trees. The location of tree block is recorded | |
902 | * in the back refs. Actually the full back refs is generic, and can be | |
903 | * used in all cases the implicit back refs is used. The major shortcoming | |
904 | * of the full back refs is its overhead. Every time a tree block gets | |
905 | * COWed, we have to update back refs entry for all pointers in it. | |
906 | * | |
907 | * For a newly allocated tree block, we use implicit back refs for | |
908 | * pointers in it. This means most tree related operations only involve | |
909 | * implicit back refs. For a tree block created in old transaction, the | |
910 | * only way to drop a reference to it is COW it. So we can detect the | |
911 | * event that tree block loses its owner tree's reference and do the | |
912 | * back refs conversion. | |
913 | * | |
01327610 | 914 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
915 | * |
916 | * The reference count of the block is one and the tree is the block's | |
917 | * owner tree. Nothing to do in this case. | |
918 | * | |
919 | * The reference count of the block is one and the tree is not the | |
920 | * block's owner tree. In this case, full back refs is used for pointers | |
921 | * in the block. Remove these full back refs, add implicit back refs for | |
922 | * every pointers in the new block. | |
923 | * | |
924 | * The reference count of the block is greater than one and the tree is | |
925 | * the block's owner tree. In this case, implicit back refs is used for | |
926 | * pointers in the block. Add full back refs for every pointers in the | |
927 | * block, increase lower level extents' reference counts. The original | |
928 | * implicit back refs are entailed to the new block. | |
929 | * | |
930 | * The reference count of the block is greater than one and the tree is | |
931 | * not the block's owner tree. Add implicit back refs for every pointer in | |
932 | * the new block, increase lower level extents' reference count. | |
933 | * | |
934 | * Back Reference Key composing: | |
935 | * | |
936 | * The key objectid corresponds to the first byte in the extent, | |
937 | * The key type is used to differentiate between types of back refs. | |
938 | * There are different meanings of the key offset for different types | |
939 | * of back refs. | |
940 | * | |
d8d5f3e1 CM |
941 | * File extents can be referenced by: |
942 | * | |
943 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 944 | * - different files inside a single subvolume |
d8d5f3e1 CM |
945 | * - different offsets inside a file (bookend extents in file.c) |
946 | * | |
5d4f98a2 | 947 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
948 | * |
949 | * - Objectid of the subvolume root | |
d8d5f3e1 | 950 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
951 | * - original offset in the file |
952 | * - how many bookend extents | |
d8d5f3e1 | 953 | * |
5d4f98a2 YZ |
954 | * The key offset for the implicit back refs is hash of the first |
955 | * three fields. | |
d8d5f3e1 | 956 | * |
5d4f98a2 | 957 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 958 | * |
5d4f98a2 | 959 | * - number of pointers in the tree leaf |
d8d5f3e1 | 960 | * |
5d4f98a2 YZ |
961 | * The key offset for the implicit back refs is the first byte of |
962 | * the tree leaf | |
d8d5f3e1 | 963 | * |
5d4f98a2 YZ |
964 | * When a file extent is allocated, The implicit back refs is used. |
965 | * the fields are filled in: | |
d8d5f3e1 | 966 | * |
5d4f98a2 | 967 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 968 | * |
5d4f98a2 YZ |
969 | * When a file extent is removed file truncation, we find the |
970 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 971 | * |
5d4f98a2 | 972 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 973 | * |
5d4f98a2 | 974 | * Btree extents can be referenced by: |
d8d5f3e1 | 975 | * |
5d4f98a2 | 976 | * - Different subvolumes |
d8d5f3e1 | 977 | * |
5d4f98a2 YZ |
978 | * Both the implicit back refs and the full back refs for tree blocks |
979 | * only consist of key. The key offset for the implicit back refs is | |
980 | * objectid of block's owner tree. The key offset for the full back refs | |
981 | * is the first byte of parent block. | |
d8d5f3e1 | 982 | * |
5d4f98a2 YZ |
983 | * When implicit back refs is used, information about the lowest key and |
984 | * level of the tree block are required. These information are stored in | |
985 | * tree block info structure. | |
d8d5f3e1 | 986 | */ |
31840ae1 | 987 | |
167ce953 LB |
988 | /* |
989 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 990 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
991 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
992 | */ | |
993 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
994 | struct btrfs_extent_inline_ref *iref, | |
995 | enum btrfs_inline_ref_type is_data) | |
996 | { | |
997 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 998 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
999 | |
1000 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1001 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1002 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1003 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1004 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1005 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1006 | return type; |
64ecdb64 LB |
1007 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1008 | ASSERT(eb->fs_info); | |
1009 | /* | |
1010 | * Every shared one has parent tree | |
1011 | * block, which must be aligned to | |
1012 | * nodesize. | |
1013 | */ | |
1014 | if (offset && | |
1015 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1016 | return type; | |
1017 | } | |
167ce953 | 1018 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1019 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1020 | return type; |
64ecdb64 LB |
1021 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1022 | ASSERT(eb->fs_info); | |
1023 | /* | |
1024 | * Every shared one has parent tree | |
1025 | * block, which must be aligned to | |
1026 | * nodesize. | |
1027 | */ | |
1028 | if (offset && | |
1029 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1030 | return type; | |
1031 | } | |
167ce953 LB |
1032 | } else { |
1033 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1034 | return type; | |
1035 | } | |
1036 | } | |
1037 | ||
1038 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1039 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1040 | eb->start, type); | |
1041 | WARN_ON(1); | |
1042 | ||
1043 | return BTRFS_REF_TYPE_INVALID; | |
1044 | } | |
1045 | ||
5d4f98a2 YZ |
1046 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1047 | { | |
1048 | u32 high_crc = ~(u32)0; | |
1049 | u32 low_crc = ~(u32)0; | |
1050 | __le64 lenum; | |
1051 | ||
1052 | lenum = cpu_to_le64(root_objectid); | |
65019df8 | 1053 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1054 | lenum = cpu_to_le64(owner); |
65019df8 | 1055 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1056 | lenum = cpu_to_le64(offset); |
65019df8 | 1057 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1058 | |
1059 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1060 | } | |
1061 | ||
1062 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1063 | struct btrfs_extent_data_ref *ref) | |
1064 | { | |
1065 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1066 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1067 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1068 | } | |
1069 | ||
1070 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1071 | struct btrfs_extent_data_ref *ref, | |
1072 | u64 root_objectid, u64 owner, u64 offset) | |
1073 | { | |
1074 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1075 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1076 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1077 | return 0; | |
1078 | return 1; | |
1079 | } | |
1080 | ||
1081 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1082 | struct btrfs_path *path, |
1083 | u64 bytenr, u64 parent, | |
1084 | u64 root_objectid, | |
1085 | u64 owner, u64 offset) | |
1086 | { | |
bd1d53ef | 1087 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
1088 | struct btrfs_key key; |
1089 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1090 | struct extent_buffer *leaf; |
5d4f98a2 | 1091 | u32 nritems; |
74493f7a | 1092 | int ret; |
5d4f98a2 YZ |
1093 | int recow; |
1094 | int err = -ENOENT; | |
74493f7a | 1095 | |
31840ae1 | 1096 | key.objectid = bytenr; |
5d4f98a2 YZ |
1097 | if (parent) { |
1098 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1099 | key.offset = parent; | |
1100 | } else { | |
1101 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1102 | key.offset = hash_extent_data_ref(root_objectid, | |
1103 | owner, offset); | |
1104 | } | |
1105 | again: | |
1106 | recow = 0; | |
1107 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1108 | if (ret < 0) { | |
1109 | err = ret; | |
1110 | goto fail; | |
1111 | } | |
31840ae1 | 1112 | |
5d4f98a2 YZ |
1113 | if (parent) { |
1114 | if (!ret) | |
1115 | return 0; | |
5d4f98a2 | 1116 | goto fail; |
31840ae1 ZY |
1117 | } |
1118 | ||
1119 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1120 | nritems = btrfs_header_nritems(leaf); |
1121 | while (1) { | |
1122 | if (path->slots[0] >= nritems) { | |
1123 | ret = btrfs_next_leaf(root, path); | |
1124 | if (ret < 0) | |
1125 | err = ret; | |
1126 | if (ret) | |
1127 | goto fail; | |
1128 | ||
1129 | leaf = path->nodes[0]; | |
1130 | nritems = btrfs_header_nritems(leaf); | |
1131 | recow = 1; | |
1132 | } | |
1133 | ||
1134 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1135 | if (key.objectid != bytenr || | |
1136 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1137 | goto fail; | |
1138 | ||
1139 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1140 | struct btrfs_extent_data_ref); | |
1141 | ||
1142 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1143 | owner, offset)) { | |
1144 | if (recow) { | |
b3b4aa74 | 1145 | btrfs_release_path(path); |
5d4f98a2 YZ |
1146 | goto again; |
1147 | } | |
1148 | err = 0; | |
1149 | break; | |
1150 | } | |
1151 | path->slots[0]++; | |
31840ae1 | 1152 | } |
5d4f98a2 YZ |
1153 | fail: |
1154 | return err; | |
31840ae1 ZY |
1155 | } |
1156 | ||
5d4f98a2 | 1157 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1158 | struct btrfs_path *path, |
1159 | u64 bytenr, u64 parent, | |
1160 | u64 root_objectid, u64 owner, | |
1161 | u64 offset, int refs_to_add) | |
31840ae1 | 1162 | { |
62b895af | 1163 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
1164 | struct btrfs_key key; |
1165 | struct extent_buffer *leaf; | |
5d4f98a2 | 1166 | u32 size; |
31840ae1 ZY |
1167 | u32 num_refs; |
1168 | int ret; | |
74493f7a | 1169 | |
74493f7a | 1170 | key.objectid = bytenr; |
5d4f98a2 YZ |
1171 | if (parent) { |
1172 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1173 | key.offset = parent; | |
1174 | size = sizeof(struct btrfs_shared_data_ref); | |
1175 | } else { | |
1176 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1177 | key.offset = hash_extent_data_ref(root_objectid, | |
1178 | owner, offset); | |
1179 | size = sizeof(struct btrfs_extent_data_ref); | |
1180 | } | |
74493f7a | 1181 | |
5d4f98a2 YZ |
1182 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1183 | if (ret && ret != -EEXIST) | |
1184 | goto fail; | |
1185 | ||
1186 | leaf = path->nodes[0]; | |
1187 | if (parent) { | |
1188 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1189 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1190 | struct btrfs_shared_data_ref); |
1191 | if (ret == 0) { | |
1192 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1193 | } else { | |
1194 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1195 | num_refs += refs_to_add; | |
1196 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1197 | } |
5d4f98a2 YZ |
1198 | } else { |
1199 | struct btrfs_extent_data_ref *ref; | |
1200 | while (ret == -EEXIST) { | |
1201 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1202 | struct btrfs_extent_data_ref); | |
1203 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1204 | owner, offset)) | |
1205 | break; | |
b3b4aa74 | 1206 | btrfs_release_path(path); |
5d4f98a2 YZ |
1207 | key.offset++; |
1208 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1209 | size); | |
1210 | if (ret && ret != -EEXIST) | |
1211 | goto fail; | |
31840ae1 | 1212 | |
5d4f98a2 YZ |
1213 | leaf = path->nodes[0]; |
1214 | } | |
1215 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1216 | struct btrfs_extent_data_ref); | |
1217 | if (ret == 0) { | |
1218 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1219 | root_objectid); | |
1220 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1221 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1222 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1223 | } else { | |
1224 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1225 | num_refs += refs_to_add; | |
1226 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1227 | } |
31840ae1 | 1228 | } |
5d4f98a2 YZ |
1229 | btrfs_mark_buffer_dirty(leaf); |
1230 | ret = 0; | |
1231 | fail: | |
b3b4aa74 | 1232 | btrfs_release_path(path); |
7bb86316 | 1233 | return ret; |
74493f7a CM |
1234 | } |
1235 | ||
5d4f98a2 | 1236 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 1237 | struct btrfs_path *path, |
fcebe456 | 1238 | int refs_to_drop, int *last_ref) |
31840ae1 | 1239 | { |
5d4f98a2 YZ |
1240 | struct btrfs_key key; |
1241 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1242 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1243 | struct extent_buffer *leaf; |
5d4f98a2 | 1244 | u32 num_refs = 0; |
31840ae1 ZY |
1245 | int ret = 0; |
1246 | ||
1247 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1248 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1249 | ||
1250 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1251 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1252 | struct btrfs_extent_data_ref); | |
1253 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1254 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1255 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1256 | struct btrfs_shared_data_ref); | |
1257 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 1258 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
1259 | btrfs_print_v0_err(trans->fs_info); |
1260 | btrfs_abort_transaction(trans, -EINVAL); | |
1261 | return -EINVAL; | |
5d4f98a2 YZ |
1262 | } else { |
1263 | BUG(); | |
1264 | } | |
1265 | ||
56bec294 CM |
1266 | BUG_ON(num_refs < refs_to_drop); |
1267 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1268 | |
31840ae1 | 1269 | if (num_refs == 0) { |
e9f6290d | 1270 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 1271 | *last_ref = 1; |
31840ae1 | 1272 | } else { |
5d4f98a2 YZ |
1273 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1274 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1275 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1276 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
1277 | btrfs_mark_buffer_dirty(leaf); |
1278 | } | |
31840ae1 ZY |
1279 | return ret; |
1280 | } | |
1281 | ||
9ed0dea0 | 1282 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1283 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1284 | { |
5d4f98a2 YZ |
1285 | struct btrfs_key key; |
1286 | struct extent_buffer *leaf; | |
1287 | struct btrfs_extent_data_ref *ref1; | |
1288 | struct btrfs_shared_data_ref *ref2; | |
1289 | u32 num_refs = 0; | |
3de28d57 | 1290 | int type; |
5d4f98a2 YZ |
1291 | |
1292 | leaf = path->nodes[0]; | |
1293 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
1294 | |
1295 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 1296 | if (iref) { |
3de28d57 LB |
1297 | /* |
1298 | * If type is invalid, we should have bailed out earlier than | |
1299 | * this call. | |
1300 | */ | |
1301 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1302 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1303 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1304 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1305 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1306 | } else { | |
1307 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1308 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1309 | } | |
1310 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1311 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1312 | struct btrfs_extent_data_ref); | |
1313 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1314 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1315 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1316 | struct btrfs_shared_data_ref); | |
1317 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
1318 | } else { |
1319 | WARN_ON(1); | |
1320 | } | |
1321 | return num_refs; | |
1322 | } | |
15916de8 | 1323 | |
5d4f98a2 | 1324 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1325 | struct btrfs_path *path, |
1326 | u64 bytenr, u64 parent, | |
1327 | u64 root_objectid) | |
1f3c79a2 | 1328 | { |
b8582eea | 1329 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 1330 | struct btrfs_key key; |
1f3c79a2 | 1331 | int ret; |
1f3c79a2 | 1332 | |
5d4f98a2 YZ |
1333 | key.objectid = bytenr; |
1334 | if (parent) { | |
1335 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1336 | key.offset = parent; | |
1337 | } else { | |
1338 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1339 | key.offset = root_objectid; | |
1f3c79a2 LH |
1340 | } |
1341 | ||
5d4f98a2 YZ |
1342 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1343 | if (ret > 0) | |
1344 | ret = -ENOENT; | |
5d4f98a2 | 1345 | return ret; |
1f3c79a2 LH |
1346 | } |
1347 | ||
5d4f98a2 | 1348 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1349 | struct btrfs_path *path, |
1350 | u64 bytenr, u64 parent, | |
1351 | u64 root_objectid) | |
31840ae1 | 1352 | { |
5d4f98a2 | 1353 | struct btrfs_key key; |
31840ae1 | 1354 | int ret; |
31840ae1 | 1355 | |
5d4f98a2 YZ |
1356 | key.objectid = bytenr; |
1357 | if (parent) { | |
1358 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1359 | key.offset = parent; | |
1360 | } else { | |
1361 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1362 | key.offset = root_objectid; | |
1363 | } | |
1364 | ||
10728404 | 1365 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 1366 | path, &key, 0); |
b3b4aa74 | 1367 | btrfs_release_path(path); |
31840ae1 ZY |
1368 | return ret; |
1369 | } | |
1370 | ||
5d4f98a2 | 1371 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1372 | { |
5d4f98a2 YZ |
1373 | int type; |
1374 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1375 | if (parent > 0) | |
1376 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1377 | else | |
1378 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1379 | } else { | |
1380 | if (parent > 0) | |
1381 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1382 | else | |
1383 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1384 | } | |
1385 | return type; | |
31840ae1 | 1386 | } |
56bec294 | 1387 | |
2c47e605 YZ |
1388 | static int find_next_key(struct btrfs_path *path, int level, |
1389 | struct btrfs_key *key) | |
56bec294 | 1390 | |
02217ed2 | 1391 | { |
2c47e605 | 1392 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1393 | if (!path->nodes[level]) |
1394 | break; | |
5d4f98a2 YZ |
1395 | if (path->slots[level] + 1 >= |
1396 | btrfs_header_nritems(path->nodes[level])) | |
1397 | continue; | |
1398 | if (level == 0) | |
1399 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1400 | path->slots[level] + 1); | |
1401 | else | |
1402 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1403 | path->slots[level] + 1); | |
1404 | return 0; | |
1405 | } | |
1406 | return 1; | |
1407 | } | |
037e6390 | 1408 | |
5d4f98a2 YZ |
1409 | /* |
1410 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1411 | * to the address of inline back ref, and 0 is returned. | |
1412 | * | |
1413 | * if back ref isn't found, *ref_ret is set to the address where it | |
1414 | * should be inserted, and -ENOENT is returned. | |
1415 | * | |
1416 | * if insert is true and there are too many inline back refs, the path | |
1417 | * points to the extent item, and -EAGAIN is returned. | |
1418 | * | |
1419 | * NOTE: inline back refs are ordered in the same way that back ref | |
1420 | * items in the tree are ordered. | |
1421 | */ | |
1422 | static noinline_for_stack | |
1423 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1424 | struct btrfs_path *path, |
1425 | struct btrfs_extent_inline_ref **ref_ret, | |
1426 | u64 bytenr, u64 num_bytes, | |
1427 | u64 parent, u64 root_objectid, | |
1428 | u64 owner, u64 offset, int insert) | |
1429 | { | |
867cc1fb | 1430 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 1431 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1432 | struct btrfs_key key; |
1433 | struct extent_buffer *leaf; | |
1434 | struct btrfs_extent_item *ei; | |
1435 | struct btrfs_extent_inline_ref *iref; | |
1436 | u64 flags; | |
1437 | u64 item_size; | |
1438 | unsigned long ptr; | |
1439 | unsigned long end; | |
1440 | int extra_size; | |
1441 | int type; | |
1442 | int want; | |
1443 | int ret; | |
1444 | int err = 0; | |
0b246afa | 1445 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1446 | int needed; |
26b8003f | 1447 | |
db94535d | 1448 | key.objectid = bytenr; |
31840ae1 | 1449 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1450 | key.offset = num_bytes; |
31840ae1 | 1451 | |
5d4f98a2 YZ |
1452 | want = extent_ref_type(parent, owner); |
1453 | if (insert) { | |
1454 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1455 | path->keep_locks = 1; |
5d4f98a2 YZ |
1456 | } else |
1457 | extra_size = -1; | |
3173a18f JB |
1458 | |
1459 | /* | |
16d1c062 NB |
1460 | * Owner is our level, so we can just add one to get the level for the |
1461 | * block we are interested in. | |
3173a18f JB |
1462 | */ |
1463 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1464 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1465 | key.offset = owner; | |
1466 | } | |
1467 | ||
1468 | again: | |
5d4f98a2 | 1469 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1470 | if (ret < 0) { |
5d4f98a2 YZ |
1471 | err = ret; |
1472 | goto out; | |
1473 | } | |
3173a18f JB |
1474 | |
1475 | /* | |
1476 | * We may be a newly converted file system which still has the old fat | |
1477 | * extent entries for metadata, so try and see if we have one of those. | |
1478 | */ | |
1479 | if (ret > 0 && skinny_metadata) { | |
1480 | skinny_metadata = false; | |
1481 | if (path->slots[0]) { | |
1482 | path->slots[0]--; | |
1483 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1484 | path->slots[0]); | |
1485 | if (key.objectid == bytenr && | |
1486 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1487 | key.offset == num_bytes) | |
1488 | ret = 0; | |
1489 | } | |
1490 | if (ret) { | |
9ce49a0b | 1491 | key.objectid = bytenr; |
3173a18f JB |
1492 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1493 | key.offset = num_bytes; | |
1494 | btrfs_release_path(path); | |
1495 | goto again; | |
1496 | } | |
1497 | } | |
1498 | ||
79787eaa JM |
1499 | if (ret && !insert) { |
1500 | err = -ENOENT; | |
1501 | goto out; | |
fae7f21c | 1502 | } else if (WARN_ON(ret)) { |
492104c8 | 1503 | err = -EIO; |
492104c8 | 1504 | goto out; |
79787eaa | 1505 | } |
5d4f98a2 YZ |
1506 | |
1507 | leaf = path->nodes[0]; | |
1508 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 1509 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1510 | err = -EINVAL; |
1511 | btrfs_print_v0_err(fs_info); | |
1512 | btrfs_abort_transaction(trans, err); | |
1513 | goto out; | |
1514 | } | |
5d4f98a2 | 1515 | |
5d4f98a2 YZ |
1516 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1517 | flags = btrfs_extent_flags(leaf, ei); | |
1518 | ||
1519 | ptr = (unsigned long)(ei + 1); | |
1520 | end = (unsigned long)ei + item_size; | |
1521 | ||
3173a18f | 1522 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1523 | ptr += sizeof(struct btrfs_tree_block_info); |
1524 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1525 | } |
1526 | ||
3de28d57 LB |
1527 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1528 | needed = BTRFS_REF_TYPE_DATA; | |
1529 | else | |
1530 | needed = BTRFS_REF_TYPE_BLOCK; | |
1531 | ||
5d4f98a2 YZ |
1532 | err = -ENOENT; |
1533 | while (1) { | |
1534 | if (ptr >= end) { | |
1535 | WARN_ON(ptr > end); | |
1536 | break; | |
1537 | } | |
1538 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1539 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1540 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 1541 | err = -EUCLEAN; |
3de28d57 LB |
1542 | goto out; |
1543 | } | |
1544 | ||
5d4f98a2 YZ |
1545 | if (want < type) |
1546 | break; | |
1547 | if (want > type) { | |
1548 | ptr += btrfs_extent_inline_ref_size(type); | |
1549 | continue; | |
1550 | } | |
1551 | ||
1552 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1553 | struct btrfs_extent_data_ref *dref; | |
1554 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1555 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1556 | owner, offset)) { | |
1557 | err = 0; | |
1558 | break; | |
1559 | } | |
1560 | if (hash_extent_data_ref_item(leaf, dref) < | |
1561 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1562 | break; | |
1563 | } else { | |
1564 | u64 ref_offset; | |
1565 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1566 | if (parent > 0) { | |
1567 | if (parent == ref_offset) { | |
1568 | err = 0; | |
1569 | break; | |
1570 | } | |
1571 | if (ref_offset < parent) | |
1572 | break; | |
1573 | } else { | |
1574 | if (root_objectid == ref_offset) { | |
1575 | err = 0; | |
1576 | break; | |
1577 | } | |
1578 | if (ref_offset < root_objectid) | |
1579 | break; | |
1580 | } | |
1581 | } | |
1582 | ptr += btrfs_extent_inline_ref_size(type); | |
1583 | } | |
1584 | if (err == -ENOENT && insert) { | |
1585 | if (item_size + extra_size >= | |
1586 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1587 | err = -EAGAIN; | |
1588 | goto out; | |
1589 | } | |
1590 | /* | |
1591 | * To add new inline back ref, we have to make sure | |
1592 | * there is no corresponding back ref item. | |
1593 | * For simplicity, we just do not add new inline back | |
1594 | * ref if there is any kind of item for this block | |
1595 | */ | |
2c47e605 YZ |
1596 | if (find_next_key(path, 0, &key) == 0 && |
1597 | key.objectid == bytenr && | |
85d4198e | 1598 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1599 | err = -EAGAIN; |
1600 | goto out; | |
1601 | } | |
1602 | } | |
1603 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1604 | out: | |
85d4198e | 1605 | if (insert) { |
5d4f98a2 YZ |
1606 | path->keep_locks = 0; |
1607 | btrfs_unlock_up_safe(path, 1); | |
1608 | } | |
1609 | return err; | |
1610 | } | |
1611 | ||
1612 | /* | |
1613 | * helper to add new inline back ref | |
1614 | */ | |
1615 | static noinline_for_stack | |
87bde3cd | 1616 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1617 | struct btrfs_path *path, |
1618 | struct btrfs_extent_inline_ref *iref, | |
1619 | u64 parent, u64 root_objectid, | |
1620 | u64 owner, u64 offset, int refs_to_add, | |
1621 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1622 | { |
1623 | struct extent_buffer *leaf; | |
1624 | struct btrfs_extent_item *ei; | |
1625 | unsigned long ptr; | |
1626 | unsigned long end; | |
1627 | unsigned long item_offset; | |
1628 | u64 refs; | |
1629 | int size; | |
1630 | int type; | |
5d4f98a2 YZ |
1631 | |
1632 | leaf = path->nodes[0]; | |
1633 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1634 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1635 | ||
1636 | type = extent_ref_type(parent, owner); | |
1637 | size = btrfs_extent_inline_ref_size(type); | |
1638 | ||
c71dd880 | 1639 | btrfs_extend_item(path, size); |
5d4f98a2 YZ |
1640 | |
1641 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1642 | refs = btrfs_extent_refs(leaf, ei); | |
1643 | refs += refs_to_add; | |
1644 | btrfs_set_extent_refs(leaf, ei, refs); | |
1645 | if (extent_op) | |
1646 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1647 | ||
1648 | ptr = (unsigned long)ei + item_offset; | |
1649 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1650 | if (ptr < end - size) | |
1651 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1652 | end - size - ptr); | |
1653 | ||
1654 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1655 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1656 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1657 | struct btrfs_extent_data_ref *dref; | |
1658 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1659 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1660 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1661 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1662 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1663 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1664 | struct btrfs_shared_data_ref *sref; | |
1665 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1666 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1667 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1668 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1669 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1670 | } else { | |
1671 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1672 | } | |
1673 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1674 | } |
1675 | ||
1676 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1677 | struct btrfs_path *path, |
1678 | struct btrfs_extent_inline_ref **ref_ret, | |
1679 | u64 bytenr, u64 num_bytes, u64 parent, | |
1680 | u64 root_objectid, u64 owner, u64 offset) | |
1681 | { | |
1682 | int ret; | |
1683 | ||
867cc1fb NB |
1684 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1685 | num_bytes, parent, root_objectid, | |
1686 | owner, offset, 0); | |
5d4f98a2 | 1687 | if (ret != -ENOENT) |
54aa1f4d | 1688 | return ret; |
5d4f98a2 | 1689 | |
b3b4aa74 | 1690 | btrfs_release_path(path); |
5d4f98a2 YZ |
1691 | *ref_ret = NULL; |
1692 | ||
1693 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1694 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1695 | root_objectid); | |
5d4f98a2 | 1696 | } else { |
bd1d53ef NB |
1697 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1698 | root_objectid, owner, offset); | |
b9473439 | 1699 | } |
5d4f98a2 YZ |
1700 | return ret; |
1701 | } | |
31840ae1 | 1702 | |
5d4f98a2 YZ |
1703 | /* |
1704 | * helper to update/remove inline back ref | |
1705 | */ | |
1706 | static noinline_for_stack | |
61a18f1c | 1707 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1708 | struct btrfs_extent_inline_ref *iref, |
1709 | int refs_to_mod, | |
fcebe456 JB |
1710 | struct btrfs_delayed_extent_op *extent_op, |
1711 | int *last_ref) | |
5d4f98a2 | 1712 | { |
61a18f1c | 1713 | struct extent_buffer *leaf = path->nodes[0]; |
5d4f98a2 YZ |
1714 | struct btrfs_extent_item *ei; |
1715 | struct btrfs_extent_data_ref *dref = NULL; | |
1716 | struct btrfs_shared_data_ref *sref = NULL; | |
1717 | unsigned long ptr; | |
1718 | unsigned long end; | |
1719 | u32 item_size; | |
1720 | int size; | |
1721 | int type; | |
5d4f98a2 YZ |
1722 | u64 refs; |
1723 | ||
5d4f98a2 YZ |
1724 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1725 | refs = btrfs_extent_refs(leaf, ei); | |
1726 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1727 | refs += refs_to_mod; | |
1728 | btrfs_set_extent_refs(leaf, ei, refs); | |
1729 | if (extent_op) | |
1730 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1731 | ||
3de28d57 LB |
1732 | /* |
1733 | * If type is invalid, we should have bailed out after | |
1734 | * lookup_inline_extent_backref(). | |
1735 | */ | |
1736 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1737 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1738 | |
1739 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1740 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1741 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1742 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1743 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1744 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1745 | } else { | |
1746 | refs = 1; | |
1747 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1748 | } |
31840ae1 | 1749 | |
5d4f98a2 YZ |
1750 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1751 | refs += refs_to_mod; | |
1752 | ||
1753 | if (refs > 0) { | |
1754 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1755 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1756 | else | |
1757 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1758 | } else { | |
fcebe456 | 1759 | *last_ref = 1; |
5d4f98a2 YZ |
1760 | size = btrfs_extent_inline_ref_size(type); |
1761 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1762 | ptr = (unsigned long)iref; | |
1763 | end = (unsigned long)ei + item_size; | |
1764 | if (ptr + size < end) | |
1765 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1766 | end - ptr - size); | |
1767 | item_size -= size; | |
78ac4f9e | 1768 | btrfs_truncate_item(path, item_size, 1); |
5d4f98a2 YZ |
1769 | } |
1770 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1771 | } |
1772 | ||
1773 | static noinline_for_stack | |
1774 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1775 | struct btrfs_path *path, |
1776 | u64 bytenr, u64 num_bytes, u64 parent, | |
1777 | u64 root_objectid, u64 owner, | |
1778 | u64 offset, int refs_to_add, | |
1779 | struct btrfs_delayed_extent_op *extent_op) | |
1780 | { | |
1781 | struct btrfs_extent_inline_ref *iref; | |
1782 | int ret; | |
1783 | ||
867cc1fb NB |
1784 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1785 | num_bytes, parent, root_objectid, | |
1786 | owner, offset, 1); | |
5d4f98a2 YZ |
1787 | if (ret == 0) { |
1788 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
61a18f1c NB |
1789 | update_inline_extent_backref(path, iref, refs_to_add, |
1790 | extent_op, NULL); | |
5d4f98a2 | 1791 | } else if (ret == -ENOENT) { |
a639cdeb | 1792 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1793 | root_objectid, owner, offset, |
1794 | refs_to_add, extent_op); | |
1795 | ret = 0; | |
771ed689 | 1796 | } |
5d4f98a2 YZ |
1797 | return ret; |
1798 | } | |
31840ae1 | 1799 | |
5d4f98a2 | 1800 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1801 | struct btrfs_path *path, |
1802 | u64 bytenr, u64 parent, u64 root_objectid, | |
1803 | u64 owner, u64 offset, int refs_to_add) | |
1804 | { | |
1805 | int ret; | |
1806 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1807 | BUG_ON(refs_to_add != 1); | |
10728404 NB |
1808 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
1809 | root_objectid); | |
5d4f98a2 | 1810 | } else { |
62b895af NB |
1811 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
1812 | root_objectid, owner, offset, | |
1813 | refs_to_add); | |
5d4f98a2 YZ |
1814 | } |
1815 | return ret; | |
1816 | } | |
56bec294 | 1817 | |
5d4f98a2 | 1818 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1819 | struct btrfs_path *path, |
1820 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1821 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1822 | { |
143bede5 | 1823 | int ret = 0; |
b9473439 | 1824 | |
5d4f98a2 YZ |
1825 | BUG_ON(!is_data && refs_to_drop != 1); |
1826 | if (iref) { | |
61a18f1c NB |
1827 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1828 | last_ref); | |
5d4f98a2 | 1829 | } else if (is_data) { |
e9f6290d | 1830 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1831 | last_ref); |
5d4f98a2 | 1832 | } else { |
fcebe456 | 1833 | *last_ref = 1; |
87cc7a8a | 1834 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1835 | } |
1836 | return ret; | |
1837 | } | |
1838 | ||
d04c6b88 JM |
1839 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1840 | u64 *discarded_bytes) | |
5d4f98a2 | 1841 | { |
86557861 JM |
1842 | int j, ret = 0; |
1843 | u64 bytes_left, end; | |
4d89d377 | 1844 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1845 | |
4d89d377 JM |
1846 | if (WARN_ON(start != aligned_start)) { |
1847 | len -= aligned_start - start; | |
1848 | len = round_down(len, 1 << 9); | |
1849 | start = aligned_start; | |
1850 | } | |
d04c6b88 | 1851 | |
4d89d377 | 1852 | *discarded_bytes = 0; |
86557861 JM |
1853 | |
1854 | if (!len) | |
1855 | return 0; | |
1856 | ||
1857 | end = start + len; | |
1858 | bytes_left = len; | |
1859 | ||
1860 | /* Skip any superblocks on this device. */ | |
1861 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1862 | u64 sb_start = btrfs_sb_offset(j); | |
1863 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1864 | u64 size = sb_start - start; | |
1865 | ||
1866 | if (!in_range(sb_start, start, bytes_left) && | |
1867 | !in_range(sb_end, start, bytes_left) && | |
1868 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1869 | continue; | |
1870 | ||
1871 | /* | |
1872 | * Superblock spans beginning of range. Adjust start and | |
1873 | * try again. | |
1874 | */ | |
1875 | if (sb_start <= start) { | |
1876 | start += sb_end - start; | |
1877 | if (start > end) { | |
1878 | bytes_left = 0; | |
1879 | break; | |
1880 | } | |
1881 | bytes_left = end - start; | |
1882 | continue; | |
1883 | } | |
1884 | ||
1885 | if (size) { | |
1886 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1887 | GFP_NOFS, 0); | |
1888 | if (!ret) | |
1889 | *discarded_bytes += size; | |
1890 | else if (ret != -EOPNOTSUPP) | |
1891 | return ret; | |
1892 | } | |
1893 | ||
1894 | start = sb_end; | |
1895 | if (start > end) { | |
1896 | bytes_left = 0; | |
1897 | break; | |
1898 | } | |
1899 | bytes_left = end - start; | |
1900 | } | |
1901 | ||
1902 | if (bytes_left) { | |
1903 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1904 | GFP_NOFS, 0); |
1905 | if (!ret) | |
86557861 | 1906 | *discarded_bytes += bytes_left; |
4d89d377 | 1907 | } |
d04c6b88 | 1908 | return ret; |
5d4f98a2 | 1909 | } |
5d4f98a2 | 1910 | |
2ff7e61e | 1911 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1912 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1913 | { |
5d4f98a2 | 1914 | int ret; |
5378e607 | 1915 | u64 discarded_bytes = 0; |
a1d3c478 | 1916 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1917 | |
e244a0ae | 1918 | |
2999241d FM |
1919 | /* |
1920 | * Avoid races with device replace and make sure our bbio has devices | |
1921 | * associated to its stripes that don't go away while we are discarding. | |
1922 | */ | |
0b246afa | 1923 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 1924 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
1925 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
1926 | &bbio, 0); | |
79787eaa | 1927 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1928 | if (!ret) { |
a1d3c478 | 1929 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1930 | int i; |
1931 | ||
5d4f98a2 | 1932 | |
a1d3c478 | 1933 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 1934 | u64 bytes; |
38b5f68e AJ |
1935 | struct request_queue *req_q; |
1936 | ||
627e0873 FM |
1937 | if (!stripe->dev->bdev) { |
1938 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
1939 | continue; | |
1940 | } | |
38b5f68e AJ |
1941 | req_q = bdev_get_queue(stripe->dev->bdev); |
1942 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
1943 | continue; |
1944 | ||
5378e607 LD |
1945 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1946 | stripe->physical, | |
d04c6b88 JM |
1947 | stripe->length, |
1948 | &bytes); | |
5378e607 | 1949 | if (!ret) |
d04c6b88 | 1950 | discarded_bytes += bytes; |
5378e607 | 1951 | else if (ret != -EOPNOTSUPP) |
79787eaa | 1952 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
1953 | |
1954 | /* | |
1955 | * Just in case we get back EOPNOTSUPP for some reason, | |
1956 | * just ignore the return value so we don't screw up | |
1957 | * people calling discard_extent. | |
1958 | */ | |
1959 | ret = 0; | |
5d4f98a2 | 1960 | } |
6e9606d2 | 1961 | btrfs_put_bbio(bbio); |
5d4f98a2 | 1962 | } |
0b246afa | 1963 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1964 | |
1965 | if (actual_bytes) | |
1966 | *actual_bytes = discarded_bytes; | |
1967 | ||
5d4f98a2 | 1968 | |
53b381b3 DW |
1969 | if (ret == -EOPNOTSUPP) |
1970 | ret = 0; | |
5d4f98a2 | 1971 | return ret; |
5d4f98a2 YZ |
1972 | } |
1973 | ||
79787eaa | 1974 | /* Can return -ENOMEM */ |
5d4f98a2 | 1975 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1976 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1977 | { |
82fa113f | 1978 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 1979 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 1980 | int ret; |
66d7e7f0 | 1981 | |
82fa113f QW |
1982 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1983 | generic_ref->action); | |
1984 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
1985 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 1986 | |
82fa113f QW |
1987 | if (generic_ref->type == BTRFS_REF_METADATA) |
1988 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, | |
ed4f255b | 1989 | NULL, &old_ref_mod, &new_ref_mod); |
82fa113f QW |
1990 | else |
1991 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0, | |
d7eae340 | 1992 | &old_ref_mod, &new_ref_mod); |
d7eae340 | 1993 | |
82fa113f | 1994 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1995 | |
ddf30cf0 | 1996 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) |
78192442 | 1997 | sub_pinned_bytes(fs_info, generic_ref); |
d7eae340 | 1998 | |
5d4f98a2 YZ |
1999 | return ret; |
2000 | } | |
2001 | ||
bd3c685e NB |
2002 | /* |
2003 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
2004 | * | |
2005 | * @trans: Handle of transaction | |
2006 | * | |
2007 | * @node: The delayed ref node used to get the bytenr/length for | |
2008 | * extent whose references are incremented. | |
2009 | * | |
2010 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
2011 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
2012 | * bytenr of the parent block. Since new extents are always | |
2013 | * created with indirect references, this will only be the case | |
2014 | * when relocating a shared extent. In that case, root_objectid | |
2015 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
2016 | * be 0 | |
2017 | * | |
2018 | * @root_objectid: The id of the root where this modification has originated, | |
2019 | * this can be either one of the well-known metadata trees or | |
2020 | * the subvolume id which references this extent. | |
2021 | * | |
2022 | * @owner: For data extents it is the inode number of the owning file. | |
2023 | * For metadata extents this parameter holds the level in the | |
2024 | * tree of the extent. | |
2025 | * | |
2026 | * @offset: For metadata extents the offset is ignored and is currently | |
2027 | * always passed as 0. For data extents it is the fileoffset | |
2028 | * this extent belongs to. | |
2029 | * | |
2030 | * @refs_to_add Number of references to add | |
2031 | * | |
2032 | * @extent_op Pointer to a structure, holding information necessary when | |
2033 | * updating a tree block's flags | |
2034 | * | |
2035 | */ | |
5d4f98a2 | 2036 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 2037 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2038 | u64 parent, u64 root_objectid, |
2039 | u64 owner, u64 offset, int refs_to_add, | |
2040 | struct btrfs_delayed_extent_op *extent_op) | |
2041 | { | |
2042 | struct btrfs_path *path; | |
2043 | struct extent_buffer *leaf; | |
2044 | struct btrfs_extent_item *item; | |
fcebe456 | 2045 | struct btrfs_key key; |
c682f9b3 QW |
2046 | u64 bytenr = node->bytenr; |
2047 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2048 | u64 refs; |
2049 | int ret; | |
5d4f98a2 YZ |
2050 | |
2051 | path = btrfs_alloc_path(); | |
2052 | if (!path) | |
2053 | return -ENOMEM; | |
2054 | ||
e4058b54 | 2055 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2056 | path->leave_spinning = 1; |
2057 | /* this will setup the path even if it fails to insert the back ref */ | |
a639cdeb NB |
2058 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
2059 | parent, root_objectid, owner, | |
2060 | offset, refs_to_add, extent_op); | |
0ed4792a | 2061 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2062 | goto out; |
fcebe456 JB |
2063 | |
2064 | /* | |
2065 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2066 | * inline extent ref, so just update the reference count and add a | |
2067 | * normal backref. | |
2068 | */ | |
5d4f98a2 | 2069 | leaf = path->nodes[0]; |
fcebe456 | 2070 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2071 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2072 | refs = btrfs_extent_refs(leaf, item); | |
2073 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2074 | if (extent_op) | |
2075 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2076 | |
5d4f98a2 | 2077 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2078 | btrfs_release_path(path); |
56bec294 | 2079 | |
e4058b54 | 2080 | path->reada = READA_FORWARD; |
b9473439 | 2081 | path->leave_spinning = 1; |
56bec294 | 2082 | /* now insert the actual backref */ |
37593410 NB |
2083 | ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid, |
2084 | owner, offset, refs_to_add); | |
79787eaa | 2085 | if (ret) |
66642832 | 2086 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2087 | out: |
56bec294 | 2088 | btrfs_free_path(path); |
30d133fc | 2089 | return ret; |
56bec294 CM |
2090 | } |
2091 | ||
5d4f98a2 | 2092 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2093 | struct btrfs_delayed_ref_node *node, |
2094 | struct btrfs_delayed_extent_op *extent_op, | |
2095 | int insert_reserved) | |
56bec294 | 2096 | { |
5d4f98a2 YZ |
2097 | int ret = 0; |
2098 | struct btrfs_delayed_data_ref *ref; | |
2099 | struct btrfs_key ins; | |
2100 | u64 parent = 0; | |
2101 | u64 ref_root = 0; | |
2102 | u64 flags = 0; | |
2103 | ||
2104 | ins.objectid = node->bytenr; | |
2105 | ins.offset = node->num_bytes; | |
2106 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2107 | ||
2108 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 2109 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2110 | |
5d4f98a2 YZ |
2111 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2112 | parent = ref->parent; | |
fcebe456 | 2113 | ref_root = ref->root; |
5d4f98a2 YZ |
2114 | |
2115 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2116 | if (extent_op) |
5d4f98a2 | 2117 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
2118 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
2119 | flags, ref->objectid, | |
2120 | ref->offset, &ins, | |
2121 | node->ref_mod); | |
5d4f98a2 | 2122 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2123 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2124 | ref->objectid, ref->offset, | |
2125 | node->ref_mod, extent_op); | |
5d4f98a2 | 2126 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2127 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
2128 | ref_root, ref->objectid, |
2129 | ref->offset, node->ref_mod, | |
c682f9b3 | 2130 | extent_op); |
5d4f98a2 YZ |
2131 | } else { |
2132 | BUG(); | |
2133 | } | |
2134 | return ret; | |
2135 | } | |
2136 | ||
2137 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2138 | struct extent_buffer *leaf, | |
2139 | struct btrfs_extent_item *ei) | |
2140 | { | |
2141 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2142 | if (extent_op->update_flags) { | |
2143 | flags |= extent_op->flags_to_set; | |
2144 | btrfs_set_extent_flags(leaf, ei, flags); | |
2145 | } | |
2146 | ||
2147 | if (extent_op->update_key) { | |
2148 | struct btrfs_tree_block_info *bi; | |
2149 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2150 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2151 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2152 | } | |
2153 | } | |
2154 | ||
2155 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 2156 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2157 | struct btrfs_delayed_extent_op *extent_op) |
2158 | { | |
20b9a2d6 | 2159 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
2160 | struct btrfs_key key; |
2161 | struct btrfs_path *path; | |
2162 | struct btrfs_extent_item *ei; | |
2163 | struct extent_buffer *leaf; | |
2164 | u32 item_size; | |
56bec294 | 2165 | int ret; |
5d4f98a2 | 2166 | int err = 0; |
b1c79e09 | 2167 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2168 | |
79787eaa JM |
2169 | if (trans->aborted) |
2170 | return 0; | |
2171 | ||
0b246afa | 2172 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2173 | metadata = 0; |
2174 | ||
5d4f98a2 YZ |
2175 | path = btrfs_alloc_path(); |
2176 | if (!path) | |
2177 | return -ENOMEM; | |
2178 | ||
d278850e | 2179 | key.objectid = head->bytenr; |
5d4f98a2 | 2180 | |
3173a18f | 2181 | if (metadata) { |
3173a18f | 2182 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2183 | key.offset = extent_op->level; |
3173a18f JB |
2184 | } else { |
2185 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2186 | key.offset = head->num_bytes; |
3173a18f JB |
2187 | } |
2188 | ||
2189 | again: | |
e4058b54 | 2190 | path->reada = READA_FORWARD; |
5d4f98a2 | 2191 | path->leave_spinning = 1; |
0b246afa | 2192 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2193 | if (ret < 0) { |
2194 | err = ret; | |
2195 | goto out; | |
2196 | } | |
2197 | if (ret > 0) { | |
3173a18f | 2198 | if (metadata) { |
55994887 FDBM |
2199 | if (path->slots[0] > 0) { |
2200 | path->slots[0]--; | |
2201 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2202 | path->slots[0]); | |
d278850e | 2203 | if (key.objectid == head->bytenr && |
55994887 | 2204 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2205 | key.offset == head->num_bytes) |
55994887 FDBM |
2206 | ret = 0; |
2207 | } | |
2208 | if (ret > 0) { | |
2209 | btrfs_release_path(path); | |
2210 | metadata = 0; | |
3173a18f | 2211 | |
d278850e JB |
2212 | key.objectid = head->bytenr; |
2213 | key.offset = head->num_bytes; | |
55994887 FDBM |
2214 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2215 | goto again; | |
2216 | } | |
2217 | } else { | |
2218 | err = -EIO; | |
2219 | goto out; | |
3173a18f | 2220 | } |
5d4f98a2 YZ |
2221 | } |
2222 | ||
2223 | leaf = path->nodes[0]; | |
2224 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 2225 | |
6d8ff4e4 | 2226 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
2227 | err = -EINVAL; |
2228 | btrfs_print_v0_err(fs_info); | |
2229 | btrfs_abort_transaction(trans, err); | |
2230 | goto out; | |
2231 | } | |
2232 | ||
5d4f98a2 YZ |
2233 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2234 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2235 | |
5d4f98a2 YZ |
2236 | btrfs_mark_buffer_dirty(leaf); |
2237 | out: | |
2238 | btrfs_free_path(path); | |
2239 | return err; | |
56bec294 CM |
2240 | } |
2241 | ||
5d4f98a2 | 2242 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2243 | struct btrfs_delayed_ref_node *node, |
2244 | struct btrfs_delayed_extent_op *extent_op, | |
2245 | int insert_reserved) | |
56bec294 CM |
2246 | { |
2247 | int ret = 0; | |
5d4f98a2 | 2248 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
2249 | u64 parent = 0; |
2250 | u64 ref_root = 0; | |
56bec294 | 2251 | |
5d4f98a2 | 2252 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 2253 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2254 | |
5d4f98a2 YZ |
2255 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2256 | parent = ref->parent; | |
fcebe456 | 2257 | ref_root = ref->root; |
5d4f98a2 | 2258 | |
02794222 | 2259 | if (node->ref_mod != 1) { |
f97806f2 | 2260 | btrfs_err(trans->fs_info, |
02794222 LB |
2261 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2262 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2263 | parent); | |
2264 | return -EIO; | |
2265 | } | |
5d4f98a2 | 2266 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2267 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 2268 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 2269 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2270 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2271 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 2272 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2273 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 2274 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
2275 | } else { |
2276 | BUG(); | |
2277 | } | |
56bec294 CM |
2278 | return ret; |
2279 | } | |
2280 | ||
2281 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2282 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2283 | struct btrfs_delayed_ref_node *node, |
2284 | struct btrfs_delayed_extent_op *extent_op, | |
2285 | int insert_reserved) | |
56bec294 | 2286 | { |
79787eaa JM |
2287 | int ret = 0; |
2288 | ||
857cc2fc JB |
2289 | if (trans->aborted) { |
2290 | if (insert_reserved) | |
5fac7f9e | 2291 | btrfs_pin_extent(trans->fs_info, node->bytenr, |
857cc2fc | 2292 | node->num_bytes, 1); |
79787eaa | 2293 | return 0; |
857cc2fc | 2294 | } |
79787eaa | 2295 | |
5d4f98a2 YZ |
2296 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2297 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 2298 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2299 | insert_reserved); |
2300 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2301 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 2302 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2303 | insert_reserved); |
2304 | else | |
2305 | BUG(); | |
80ee54bf JB |
2306 | if (ret && insert_reserved) |
2307 | btrfs_pin_extent(trans->fs_info, node->bytenr, | |
2308 | node->num_bytes, 1); | |
5d4f98a2 | 2309 | return ret; |
56bec294 CM |
2310 | } |
2311 | ||
c6fc2454 | 2312 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2313 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2314 | { | |
cffc3374 FM |
2315 | struct btrfs_delayed_ref_node *ref; |
2316 | ||
e3d03965 | 2317 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 2318 | return NULL; |
d7df2c79 | 2319 | |
cffc3374 FM |
2320 | /* |
2321 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2322 | * This is to prevent a ref count from going down to zero, which deletes | |
2323 | * the extent item from the extent tree, when there still are references | |
2324 | * to add, which would fail because they would not find the extent item. | |
2325 | */ | |
1d57ee94 WX |
2326 | if (!list_empty(&head->ref_add_list)) |
2327 | return list_first_entry(&head->ref_add_list, | |
2328 | struct btrfs_delayed_ref_node, add_list); | |
2329 | ||
e3d03965 | 2330 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 2331 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
2332 | ASSERT(list_empty(&ref->add_list)); |
2333 | return ref; | |
56bec294 CM |
2334 | } |
2335 | ||
2eadaa22 JB |
2336 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2337 | struct btrfs_delayed_ref_head *head) | |
2338 | { | |
2339 | spin_lock(&delayed_refs->lock); | |
2340 | head->processing = 0; | |
2341 | delayed_refs->num_heads_ready++; | |
2342 | spin_unlock(&delayed_refs->lock); | |
2343 | btrfs_delayed_ref_unlock(head); | |
2344 | } | |
2345 | ||
bedc6617 JB |
2346 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
2347 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
2348 | { |
2349 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
2350 | |
2351 | if (!extent_op) | |
bedc6617 JB |
2352 | return NULL; |
2353 | ||
b00e6250 | 2354 | if (head->must_insert_reserved) { |
bedc6617 | 2355 | head->extent_op = NULL; |
b00e6250 | 2356 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 2357 | return NULL; |
b00e6250 | 2358 | } |
bedc6617 JB |
2359 | return extent_op; |
2360 | } | |
2361 | ||
2362 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
2363 | struct btrfs_delayed_ref_head *head) | |
2364 | { | |
2365 | struct btrfs_delayed_extent_op *extent_op; | |
2366 | int ret; | |
2367 | ||
2368 | extent_op = cleanup_extent_op(head); | |
2369 | if (!extent_op) | |
2370 | return 0; | |
2371 | head->extent_op = NULL; | |
b00e6250 | 2372 | spin_unlock(&head->lock); |
20b9a2d6 | 2373 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
2374 | btrfs_free_delayed_extent_op(extent_op); |
2375 | return ret ? ret : 1; | |
2376 | } | |
2377 | ||
31890da0 JB |
2378 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
2379 | struct btrfs_delayed_ref_root *delayed_refs, | |
2380 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 2381 | { |
ba2c4d4e | 2382 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 JB |
2383 | |
2384 | if (head->total_ref_mod < 0) { | |
2385 | struct btrfs_space_info *space_info; | |
2386 | u64 flags; | |
2387 | ||
2388 | if (head->is_data) | |
2389 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2390 | else if (head->is_system) | |
2391 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
2392 | else | |
2393 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
280c2908 | 2394 | space_info = btrfs_find_space_info(fs_info, flags); |
07c47775 JB |
2395 | ASSERT(space_info); |
2396 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
2397 | -head->num_bytes, | |
2398 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
2399 | ||
ba2c4d4e JB |
2400 | /* |
2401 | * We had csum deletions accounted for in our delayed refs rsv, | |
2402 | * we need to drop the csum leaves for this update from our | |
2403 | * delayed_refs_rsv. | |
2404 | */ | |
07c47775 JB |
2405 | if (head->is_data) { |
2406 | spin_lock(&delayed_refs->lock); | |
2407 | delayed_refs->pending_csums -= head->num_bytes; | |
2408 | spin_unlock(&delayed_refs->lock); | |
ba2c4d4e JB |
2409 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, |
2410 | head->num_bytes); | |
07c47775 JB |
2411 | } |
2412 | } | |
2413 | ||
ba2c4d4e | 2414 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
2415 | } |
2416 | ||
194ab0bc | 2417 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
2418 | struct btrfs_delayed_ref_head *head) |
2419 | { | |
f9871edd NB |
2420 | |
2421 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
2422 | struct btrfs_delayed_ref_root *delayed_refs; |
2423 | int ret; | |
2424 | ||
2425 | delayed_refs = &trans->transaction->delayed_refs; | |
2426 | ||
bedc6617 | 2427 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
2428 | if (ret < 0) { |
2429 | unselect_delayed_ref_head(delayed_refs, head); | |
2430 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2431 | return ret; | |
2432 | } else if (ret) { | |
2433 | return ret; | |
2434 | } | |
2435 | ||
2436 | /* | |
2437 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2438 | * and then re-check to make sure nobody got added. | |
2439 | */ | |
2440 | spin_unlock(&head->lock); | |
2441 | spin_lock(&delayed_refs->lock); | |
2442 | spin_lock(&head->lock); | |
e3d03965 | 2443 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
2444 | spin_unlock(&head->lock); |
2445 | spin_unlock(&delayed_refs->lock); | |
2446 | return 1; | |
2447 | } | |
d7baffda | 2448 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 2449 | spin_unlock(&head->lock); |
1e7a1421 | 2450 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 2451 | |
c1103f7a | 2452 | if (head->must_insert_reserved) { |
d278850e JB |
2453 | btrfs_pin_extent(fs_info, head->bytenr, |
2454 | head->num_bytes, 1); | |
c1103f7a | 2455 | if (head->is_data) { |
d278850e JB |
2456 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2457 | head->num_bytes); | |
c1103f7a JB |
2458 | } |
2459 | } | |
2460 | ||
31890da0 | 2461 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
2462 | |
2463 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 2464 | btrfs_delayed_ref_unlock(head); |
d278850e | 2465 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2466 | return 0; |
2467 | } | |
2468 | ||
b1cdbcb5 NB |
2469 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
2470 | struct btrfs_trans_handle *trans) | |
2471 | { | |
2472 | struct btrfs_delayed_ref_root *delayed_refs = | |
2473 | &trans->transaction->delayed_refs; | |
2474 | struct btrfs_delayed_ref_head *head = NULL; | |
2475 | int ret; | |
2476 | ||
2477 | spin_lock(&delayed_refs->lock); | |
5637c74b | 2478 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
2479 | if (!head) { |
2480 | spin_unlock(&delayed_refs->lock); | |
2481 | return head; | |
2482 | } | |
2483 | ||
2484 | /* | |
2485 | * Grab the lock that says we are going to process all the refs for | |
2486 | * this head | |
2487 | */ | |
9e920a6f | 2488 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
2489 | spin_unlock(&delayed_refs->lock); |
2490 | ||
2491 | /* | |
2492 | * We may have dropped the spin lock to get the head mutex lock, and | |
2493 | * that might have given someone else time to free the head. If that's | |
2494 | * true, it has been removed from our list and we can move on. | |
2495 | */ | |
2496 | if (ret == -EAGAIN) | |
2497 | head = ERR_PTR(-EAGAIN); | |
2498 | ||
2499 | return head; | |
2500 | } | |
2501 | ||
e7261386 NB |
2502 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
2503 | struct btrfs_delayed_ref_head *locked_ref, | |
2504 | unsigned long *run_refs) | |
2505 | { | |
2506 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2507 | struct btrfs_delayed_ref_root *delayed_refs; | |
2508 | struct btrfs_delayed_extent_op *extent_op; | |
2509 | struct btrfs_delayed_ref_node *ref; | |
2510 | int must_insert_reserved = 0; | |
2511 | int ret; | |
2512 | ||
2513 | delayed_refs = &trans->transaction->delayed_refs; | |
2514 | ||
0110a4c4 NB |
2515 | lockdep_assert_held(&locked_ref->mutex); |
2516 | lockdep_assert_held(&locked_ref->lock); | |
2517 | ||
e7261386 NB |
2518 | while ((ref = select_delayed_ref(locked_ref))) { |
2519 | if (ref->seq && | |
2520 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
2521 | spin_unlock(&locked_ref->lock); | |
2522 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2523 | return -EAGAIN; | |
2524 | } | |
2525 | ||
2526 | (*run_refs)++; | |
2527 | ref->in_tree = 0; | |
2528 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
2529 | RB_CLEAR_NODE(&ref->ref_node); | |
2530 | if (!list_empty(&ref->add_list)) | |
2531 | list_del(&ref->add_list); | |
2532 | /* | |
2533 | * When we play the delayed ref, also correct the ref_mod on | |
2534 | * head | |
2535 | */ | |
2536 | switch (ref->action) { | |
2537 | case BTRFS_ADD_DELAYED_REF: | |
2538 | case BTRFS_ADD_DELAYED_EXTENT: | |
2539 | locked_ref->ref_mod -= ref->ref_mod; | |
2540 | break; | |
2541 | case BTRFS_DROP_DELAYED_REF: | |
2542 | locked_ref->ref_mod += ref->ref_mod; | |
2543 | break; | |
2544 | default: | |
2545 | WARN_ON(1); | |
2546 | } | |
2547 | atomic_dec(&delayed_refs->num_entries); | |
2548 | ||
2549 | /* | |
2550 | * Record the must_insert_reserved flag before we drop the | |
2551 | * spin lock. | |
2552 | */ | |
2553 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2554 | locked_ref->must_insert_reserved = 0; | |
2555 | ||
2556 | extent_op = locked_ref->extent_op; | |
2557 | locked_ref->extent_op = NULL; | |
2558 | spin_unlock(&locked_ref->lock); | |
2559 | ||
2560 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
2561 | must_insert_reserved); | |
2562 | ||
2563 | btrfs_free_delayed_extent_op(extent_op); | |
2564 | if (ret) { | |
2565 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2566 | btrfs_put_delayed_ref(ref); | |
2567 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
2568 | ret); | |
2569 | return ret; | |
2570 | } | |
2571 | ||
2572 | btrfs_put_delayed_ref(ref); | |
2573 | cond_resched(); | |
2574 | ||
2575 | spin_lock(&locked_ref->lock); | |
2576 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
2577 | } | |
2578 | ||
2579 | return 0; | |
2580 | } | |
2581 | ||
79787eaa JM |
2582 | /* |
2583 | * Returns 0 on success or if called with an already aborted transaction. | |
2584 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2585 | */ | |
d7df2c79 | 2586 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2587 | unsigned long nr) |
56bec294 | 2588 | { |
0a1e458a | 2589 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2590 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 2591 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 2592 | ktime_t start = ktime_get(); |
56bec294 | 2593 | int ret; |
d7df2c79 | 2594 | unsigned long count = 0; |
0a2b2a84 | 2595 | unsigned long actual_count = 0; |
56bec294 CM |
2596 | |
2597 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 2598 | do { |
56bec294 | 2599 | if (!locked_ref) { |
b1cdbcb5 | 2600 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2601 | if (IS_ERR_OR_NULL(locked_ref)) { |
2602 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2603 | continue; | |
2604 | } else { | |
2605 | break; | |
2606 | } | |
56bec294 | 2607 | } |
0110a4c4 | 2608 | count++; |
56bec294 | 2609 | } |
2c3cf7d5 FM |
2610 | /* |
2611 | * We need to try and merge add/drops of the same ref since we | |
2612 | * can run into issues with relocate dropping the implicit ref | |
2613 | * and then it being added back again before the drop can | |
2614 | * finish. If we merged anything we need to re-loop so we can | |
2615 | * get a good ref. | |
2616 | * Or we can get node references of the same type that weren't | |
2617 | * merged when created due to bumps in the tree mod seq, and | |
2618 | * we need to merge them to prevent adding an inline extent | |
2619 | * backref before dropping it (triggering a BUG_ON at | |
2620 | * insert_inline_extent_backref()). | |
2621 | */ | |
d7df2c79 | 2622 | spin_lock(&locked_ref->lock); |
be97f133 | 2623 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2624 | |
0110a4c4 NB |
2625 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2626 | &actual_count); | |
2627 | if (ret < 0 && ret != -EAGAIN) { | |
2628 | /* | |
2629 | * Error, btrfs_run_delayed_refs_for_head already | |
2630 | * unlocked everything so just bail out | |
2631 | */ | |
2632 | return ret; | |
2633 | } else if (!ret) { | |
2634 | /* | |
2635 | * Success, perform the usual cleanup of a processed | |
2636 | * head | |
2637 | */ | |
f9871edd | 2638 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2639 | if (ret > 0 ) { |
b00e6250 JB |
2640 | /* We dropped our lock, we need to loop. */ |
2641 | ret = 0; | |
d7df2c79 | 2642 | continue; |
194ab0bc JB |
2643 | } else if (ret) { |
2644 | return ret; | |
5d4f98a2 | 2645 | } |
22cd2e7d | 2646 | } |
1ce7a5ec | 2647 | |
b00e6250 | 2648 | /* |
0110a4c4 NB |
2649 | * Either success case or btrfs_run_delayed_refs_for_head |
2650 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2651 | */ |
b00e6250 | 2652 | |
0110a4c4 | 2653 | locked_ref = NULL; |
c3e69d58 | 2654 | cond_resched(); |
0110a4c4 | 2655 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2656 | |
2657 | /* | |
2658 | * We don't want to include ref heads since we can have empty ref heads | |
2659 | * and those will drastically skew our runtime down since we just do | |
2660 | * accounting, no actual extent tree updates. | |
2661 | */ | |
2662 | if (actual_count > 0) { | |
2663 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2664 | u64 avg; | |
2665 | ||
2666 | /* | |
2667 | * We weigh the current average higher than our current runtime | |
2668 | * to avoid large swings in the average. | |
2669 | */ | |
2670 | spin_lock(&delayed_refs->lock); | |
2671 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2672 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2673 | spin_unlock(&delayed_refs->lock); |
2674 | } | |
d7df2c79 | 2675 | return 0; |
c3e69d58 CM |
2676 | } |
2677 | ||
709c0486 AJ |
2678 | #ifdef SCRAMBLE_DELAYED_REFS |
2679 | /* | |
2680 | * Normally delayed refs get processed in ascending bytenr order. This | |
2681 | * correlates in most cases to the order added. To expose dependencies on this | |
2682 | * order, we start to process the tree in the middle instead of the beginning | |
2683 | */ | |
2684 | static u64 find_middle(struct rb_root *root) | |
2685 | { | |
2686 | struct rb_node *n = root->rb_node; | |
2687 | struct btrfs_delayed_ref_node *entry; | |
2688 | int alt = 1; | |
2689 | u64 middle; | |
2690 | u64 first = 0, last = 0; | |
2691 | ||
2692 | n = rb_first(root); | |
2693 | if (n) { | |
2694 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2695 | first = entry->bytenr; | |
2696 | } | |
2697 | n = rb_last(root); | |
2698 | if (n) { | |
2699 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2700 | last = entry->bytenr; | |
2701 | } | |
2702 | n = root->rb_node; | |
2703 | ||
2704 | while (n) { | |
2705 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2706 | WARN_ON(!entry->in_tree); | |
2707 | ||
2708 | middle = entry->bytenr; | |
2709 | ||
2710 | if (alt) | |
2711 | n = n->rb_left; | |
2712 | else | |
2713 | n = n->rb_right; | |
2714 | ||
2715 | alt = 1 - alt; | |
2716 | } | |
2717 | return middle; | |
2718 | } | |
2719 | #endif | |
2720 | ||
2ff7e61e | 2721 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2722 | { |
2723 | u64 num_bytes; | |
2724 | ||
2725 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2726 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2727 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2728 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2729 | ||
2730 | /* | |
2731 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2732 | * closer to what we're really going to want to use. |
1be41b78 | 2733 | */ |
0b246afa | 2734 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2735 | } |
2736 | ||
1262133b JB |
2737 | /* |
2738 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2739 | * would require to store the csums for that many bytes. | |
2740 | */ | |
2ff7e61e | 2741 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2742 | { |
2743 | u64 csum_size; | |
2744 | u64 num_csums_per_leaf; | |
2745 | u64 num_csums; | |
2746 | ||
0b246afa | 2747 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2748 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2749 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2750 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2751 | num_csums += num_csums_per_leaf - 1; |
2752 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2753 | return num_csums; | |
2754 | } | |
2755 | ||
c3e69d58 CM |
2756 | /* |
2757 | * this starts processing the delayed reference count updates and | |
2758 | * extent insertions we have queued up so far. count can be | |
2759 | * 0, which means to process everything in the tree at the start | |
2760 | * of the run (but not newly added entries), or it can be some target | |
2761 | * number you'd like to process. | |
79787eaa JM |
2762 | * |
2763 | * Returns 0 on success or if called with an aborted transaction | |
2764 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2765 | */ |
2766 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2767 | unsigned long count) |
c3e69d58 | 2768 | { |
c79a70b1 | 2769 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2770 | struct rb_node *node; |
2771 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2772 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2773 | int ret; |
2774 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2775 | |
79787eaa JM |
2776 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2777 | if (trans->aborted) | |
2778 | return 0; | |
2779 | ||
0b246afa | 2780 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2781 | return 0; |
2782 | ||
c3e69d58 | 2783 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2784 | if (count == 0) |
d7df2c79 | 2785 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2786 | |
c3e69d58 | 2787 | again: |
709c0486 AJ |
2788 | #ifdef SCRAMBLE_DELAYED_REFS |
2789 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2790 | #endif | |
0a1e458a | 2791 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2792 | if (ret < 0) { |
66642832 | 2793 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2794 | return ret; |
eb099670 | 2795 | } |
c3e69d58 | 2796 | |
56bec294 | 2797 | if (run_all) { |
119e80df | 2798 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2799 | |
d7df2c79 | 2800 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2801 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2802 | if (!node) { |
2803 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2804 | goto out; |
d7df2c79 | 2805 | } |
d278850e JB |
2806 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2807 | href_node); | |
2808 | refcount_inc(&head->refs); | |
2809 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2810 | |
d278850e JB |
2811 | /* Mutex was contended, block until it's released and retry. */ |
2812 | mutex_lock(&head->mutex); | |
2813 | mutex_unlock(&head->mutex); | |
56bec294 | 2814 | |
d278850e | 2815 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2816 | cond_resched(); |
56bec294 | 2817 | goto again; |
5f39d397 | 2818 | } |
54aa1f4d | 2819 | out: |
a28ec197 CM |
2820 | return 0; |
2821 | } | |
2822 | ||
5d4f98a2 | 2823 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
5d4f98a2 | 2824 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 2825 | int level, int is_data) |
5d4f98a2 YZ |
2826 | { |
2827 | struct btrfs_delayed_extent_op *extent_op; | |
2828 | int ret; | |
2829 | ||
78a6184a | 2830 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2831 | if (!extent_op) |
2832 | return -ENOMEM; | |
2833 | ||
2834 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2835 | extent_op->update_flags = true; |
2836 | extent_op->update_key = false; | |
2837 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2838 | extent_op->level = level; |
5d4f98a2 | 2839 | |
c6e340bc | 2840 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); |
5d4f98a2 | 2841 | if (ret) |
78a6184a | 2842 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2843 | return ret; |
2844 | } | |
2845 | ||
e4c3b2dc | 2846 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2847 | struct btrfs_path *path, |
2848 | u64 objectid, u64 offset, u64 bytenr) | |
2849 | { | |
2850 | struct btrfs_delayed_ref_head *head; | |
2851 | struct btrfs_delayed_ref_node *ref; | |
2852 | struct btrfs_delayed_data_ref *data_ref; | |
2853 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2854 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2855 | struct rb_node *node; |
5d4f98a2 YZ |
2856 | int ret = 0; |
2857 | ||
998ac6d2 | 2858 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2859 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2860 | if (cur_trans) |
2861 | refcount_inc(&cur_trans->use_count); | |
2862 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2863 | if (!cur_trans) |
2864 | return 0; | |
2865 | ||
2866 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2867 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2868 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2869 | if (!head) { |
2870 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2871 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2872 | return 0; |
2873 | } | |
5d4f98a2 YZ |
2874 | |
2875 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2876 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2877 | spin_unlock(&delayed_refs->lock); |
2878 | ||
b3b4aa74 | 2879 | btrfs_release_path(path); |
5d4f98a2 | 2880 | |
8cc33e5c DS |
2881 | /* |
2882 | * Mutex was contended, block until it's released and let | |
2883 | * caller try again | |
2884 | */ | |
5d4f98a2 YZ |
2885 | mutex_lock(&head->mutex); |
2886 | mutex_unlock(&head->mutex); | |
d278850e | 2887 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2888 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2889 | return -EAGAIN; |
2890 | } | |
d7df2c79 | 2891 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2892 | |
d7df2c79 | 2893 | spin_lock(&head->lock); |
0e0adbcf JB |
2894 | /* |
2895 | * XXX: We should replace this with a proper search function in the | |
2896 | * future. | |
2897 | */ | |
e3d03965 LB |
2898 | for (node = rb_first_cached(&head->ref_tree); node; |
2899 | node = rb_next(node)) { | |
0e0adbcf | 2900 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2901 | /* If it's a shared ref we know a cross reference exists */ |
2902 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2903 | ret = 1; | |
2904 | break; | |
2905 | } | |
5d4f98a2 | 2906 | |
d7df2c79 | 2907 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2908 | |
d7df2c79 JB |
2909 | /* |
2910 | * If our ref doesn't match the one we're currently looking at | |
2911 | * then we have a cross reference. | |
2912 | */ | |
2913 | if (data_ref->root != root->root_key.objectid || | |
2914 | data_ref->objectid != objectid || | |
2915 | data_ref->offset != offset) { | |
2916 | ret = 1; | |
2917 | break; | |
2918 | } | |
5d4f98a2 | 2919 | } |
d7df2c79 | 2920 | spin_unlock(&head->lock); |
5d4f98a2 | 2921 | mutex_unlock(&head->mutex); |
998ac6d2 | 2922 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2923 | return ret; |
2924 | } | |
2925 | ||
e4c3b2dc | 2926 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2927 | struct btrfs_path *path, |
2928 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 2929 | { |
0b246afa JM |
2930 | struct btrfs_fs_info *fs_info = root->fs_info; |
2931 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 2932 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2933 | struct btrfs_extent_data_ref *ref; |
2934 | struct btrfs_extent_inline_ref *iref; | |
2935 | struct btrfs_extent_item *ei; | |
f321e491 | 2936 | struct btrfs_key key; |
5d4f98a2 | 2937 | u32 item_size; |
3de28d57 | 2938 | int type; |
be20aa9d | 2939 | int ret; |
925baedd | 2940 | |
be20aa9d | 2941 | key.objectid = bytenr; |
31840ae1 | 2942 | key.offset = (u64)-1; |
f321e491 | 2943 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2944 | |
be20aa9d CM |
2945 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2946 | if (ret < 0) | |
2947 | goto out; | |
79787eaa | 2948 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2949 | |
2950 | ret = -ENOENT; | |
2951 | if (path->slots[0] == 0) | |
31840ae1 | 2952 | goto out; |
be20aa9d | 2953 | |
31840ae1 | 2954 | path->slots[0]--; |
f321e491 | 2955 | leaf = path->nodes[0]; |
5d4f98a2 | 2956 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2957 | |
5d4f98a2 | 2958 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2959 | goto out; |
f321e491 | 2960 | |
5d4f98a2 YZ |
2961 | ret = 1; |
2962 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 2963 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 2964 | |
5d4f98a2 YZ |
2965 | if (item_size != sizeof(*ei) + |
2966 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2967 | goto out; | |
be20aa9d | 2968 | |
5d4f98a2 YZ |
2969 | if (btrfs_extent_generation(leaf, ei) <= |
2970 | btrfs_root_last_snapshot(&root->root_item)) | |
2971 | goto out; | |
2972 | ||
2973 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
2974 | |
2975 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
2976 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2977 | goto out; |
2978 | ||
2979 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2980 | if (btrfs_extent_refs(leaf, ei) != | |
2981 | btrfs_extent_data_ref_count(leaf, ref) || | |
2982 | btrfs_extent_data_ref_root(leaf, ref) != | |
2983 | root->root_key.objectid || | |
2984 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2985 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2986 | goto out; | |
2987 | ||
2988 | ret = 0; | |
2989 | out: | |
2990 | return ret; | |
2991 | } | |
2992 | ||
e4c3b2dc LB |
2993 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
2994 | u64 bytenr) | |
5d4f98a2 YZ |
2995 | { |
2996 | struct btrfs_path *path; | |
2997 | int ret; | |
5d4f98a2 YZ |
2998 | |
2999 | path = btrfs_alloc_path(); | |
3000 | if (!path) | |
9132c4ff | 3001 | return -ENOMEM; |
5d4f98a2 YZ |
3002 | |
3003 | do { | |
e4c3b2dc | 3004 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3005 | offset, bytenr); |
3006 | if (ret && ret != -ENOENT) | |
f321e491 | 3007 | goto out; |
80ff3856 | 3008 | |
380fd066 MT |
3009 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
3010 | } while (ret == -EAGAIN); | |
5d4f98a2 | 3011 | |
be20aa9d | 3012 | out: |
80ff3856 | 3013 | btrfs_free_path(path); |
f0486c68 YZ |
3014 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3015 | WARN_ON(ret > 0); | |
f321e491 | 3016 | return ret; |
be20aa9d | 3017 | } |
c5739bba | 3018 | |
5d4f98a2 | 3019 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3020 | struct btrfs_root *root, |
5d4f98a2 | 3021 | struct extent_buffer *buf, |
e339a6b0 | 3022 | int full_backref, int inc) |
31840ae1 | 3023 | { |
0b246afa | 3024 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3025 | u64 bytenr; |
5d4f98a2 YZ |
3026 | u64 num_bytes; |
3027 | u64 parent; | |
31840ae1 | 3028 | u64 ref_root; |
31840ae1 | 3029 | u32 nritems; |
31840ae1 ZY |
3030 | struct btrfs_key key; |
3031 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
3032 | struct btrfs_ref generic_ref = { 0 }; |
3033 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 3034 | int i; |
82fa113f | 3035 | int action; |
31840ae1 ZY |
3036 | int level; |
3037 | int ret = 0; | |
fccb84c9 | 3038 | |
0b246afa | 3039 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3040 | return 0; |
fccb84c9 | 3041 | |
31840ae1 | 3042 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3043 | nritems = btrfs_header_nritems(buf); |
3044 | level = btrfs_header_level(buf); | |
3045 | ||
27cdeb70 | 3046 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3047 | return 0; |
31840ae1 | 3048 | |
5d4f98a2 YZ |
3049 | if (full_backref) |
3050 | parent = buf->start; | |
3051 | else | |
3052 | parent = 0; | |
82fa113f QW |
3053 | if (inc) |
3054 | action = BTRFS_ADD_DELAYED_REF; | |
3055 | else | |
3056 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
3057 | |
3058 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3059 | if (level == 0) { |
5d4f98a2 | 3060 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3061 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3062 | continue; |
5d4f98a2 | 3063 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3064 | struct btrfs_file_extent_item); |
3065 | if (btrfs_file_extent_type(buf, fi) == | |
3066 | BTRFS_FILE_EXTENT_INLINE) | |
3067 | continue; | |
3068 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3069 | if (bytenr == 0) | |
3070 | continue; | |
5d4f98a2 YZ |
3071 | |
3072 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3073 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
3074 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
3075 | num_bytes, parent); | |
3076 | generic_ref.real_root = root->root_key.objectid; | |
3077 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
3078 | key.offset); | |
3079 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 3080 | if (inc) |
82fa113f | 3081 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 3082 | else |
ffd4bb2a | 3083 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
3084 | if (ret) |
3085 | goto fail; | |
3086 | } else { | |
5d4f98a2 | 3087 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3088 | num_bytes = fs_info->nodesize; |
82fa113f QW |
3089 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
3090 | num_bytes, parent); | |
3091 | generic_ref.real_root = root->root_key.objectid; | |
3092 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
3093 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 3094 | if (inc) |
82fa113f | 3095 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 3096 | else |
ffd4bb2a | 3097 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
3098 | if (ret) |
3099 | goto fail; | |
3100 | } | |
3101 | } | |
3102 | return 0; | |
3103 | fail: | |
5d4f98a2 YZ |
3104 | return ret; |
3105 | } | |
3106 | ||
3107 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3108 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3109 | { |
e339a6b0 | 3110 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3111 | } |
3112 | ||
3113 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3114 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3115 | { |
e339a6b0 | 3116 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3117 | } |
3118 | ||
9078a3e1 | 3119 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
9078a3e1 CM |
3120 | struct btrfs_path *path, |
3121 | struct btrfs_block_group_cache *cache) | |
3122 | { | |
39db232d | 3123 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9078a3e1 | 3124 | int ret; |
0b246afa | 3125 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3126 | unsigned long bi; |
3127 | struct extent_buffer *leaf; | |
9078a3e1 | 3128 | |
9078a3e1 | 3129 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3130 | if (ret) { |
3131 | if (ret > 0) | |
3132 | ret = -ENOENT; | |
54aa1f4d | 3133 | goto fail; |
df95e7f0 | 3134 | } |
5f39d397 CM |
3135 | |
3136 | leaf = path->nodes[0]; | |
3137 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3138 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3139 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3140 | fail: |
24b89d08 | 3141 | btrfs_release_path(path); |
df95e7f0 | 3142 | return ret; |
9078a3e1 CM |
3143 | |
3144 | } | |
3145 | ||
f87b7eb8 DS |
3146 | static struct btrfs_block_group_cache *next_block_group( |
3147 | struct btrfs_block_group_cache *cache) | |
4a8c9a62 | 3148 | { |
f87b7eb8 | 3149 | struct btrfs_fs_info *fs_info = cache->fs_info; |
4a8c9a62 | 3150 | struct rb_node *node; |
292cbd51 | 3151 | |
0b246afa | 3152 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3153 | |
3154 | /* If our block group was removed, we need a full search. */ | |
3155 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3156 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3157 | ||
0b246afa | 3158 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3159 | btrfs_put_block_group(cache); |
0b246afa | 3160 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3161 | } |
4a8c9a62 YZ |
3162 | node = rb_next(&cache->cache_node); |
3163 | btrfs_put_block_group(cache); | |
3164 | if (node) { | |
3165 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3166 | cache_node); | |
11dfe35a | 3167 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3168 | } else |
3169 | cache = NULL; | |
0b246afa | 3170 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3171 | return cache; |
3172 | } | |
3173 | ||
0af3d00b JB |
3174 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3175 | struct btrfs_trans_handle *trans, | |
3176 | struct btrfs_path *path) | |
3177 | { | |
0b246afa JM |
3178 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3179 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3180 | struct inode *inode = NULL; |
364ecf36 | 3181 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3182 | u64 alloc_hint = 0; |
2b20982e | 3183 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3184 | u64 num_pages = 0; |
0af3d00b JB |
3185 | int retries = 0; |
3186 | int ret = 0; | |
3187 | ||
3188 | /* | |
3189 | * If this block group is smaller than 100 megs don't bother caching the | |
3190 | * block group. | |
3191 | */ | |
ee22184b | 3192 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3193 | spin_lock(&block_group->lock); |
3194 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3195 | spin_unlock(&block_group->lock); | |
3196 | return 0; | |
3197 | } | |
3198 | ||
0c0ef4bc JB |
3199 | if (trans->aborted) |
3200 | return 0; | |
0af3d00b | 3201 | again: |
7949f339 | 3202 | inode = lookup_free_space_inode(block_group, path); |
0af3d00b JB |
3203 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3204 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3205 | btrfs_release_path(path); |
0af3d00b JB |
3206 | goto out; |
3207 | } | |
3208 | ||
3209 | if (IS_ERR(inode)) { | |
3210 | BUG_ON(retries); | |
3211 | retries++; | |
3212 | ||
3213 | if (block_group->ro) | |
3214 | goto out_free; | |
3215 | ||
4ca75f1b | 3216 | ret = create_free_space_inode(trans, block_group, path); |
0af3d00b JB |
3217 | if (ret) |
3218 | goto out_free; | |
3219 | goto again; | |
3220 | } | |
3221 | ||
3222 | /* | |
3223 | * We want to set the generation to 0, that way if anything goes wrong | |
3224 | * from here on out we know not to trust this cache when we load up next | |
3225 | * time. | |
3226 | */ | |
3227 | BTRFS_I(inode)->generation = 0; | |
3228 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3229 | if (ret) { |
3230 | /* | |
3231 | * So theoretically we could recover from this, simply set the | |
3232 | * super cache generation to 0 so we know to invalidate the | |
3233 | * cache, but then we'd have to keep track of the block groups | |
3234 | * that fail this way so we know we _have_ to reset this cache | |
3235 | * before the next commit or risk reading stale cache. So to | |
3236 | * limit our exposure to horrible edge cases lets just abort the | |
3237 | * transaction, this only happens in really bad situations | |
3238 | * anyway. | |
3239 | */ | |
66642832 | 3240 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3241 | goto out_put; |
3242 | } | |
0af3d00b JB |
3243 | WARN_ON(ret); |
3244 | ||
8e138e0d JB |
3245 | /* We've already setup this transaction, go ahead and exit */ |
3246 | if (block_group->cache_generation == trans->transid && | |
3247 | i_size_read(inode)) { | |
3248 | dcs = BTRFS_DC_SETUP; | |
3249 | goto out_put; | |
3250 | } | |
3251 | ||
0af3d00b | 3252 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3253 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3254 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3255 | if (ret) |
3256 | goto out_put; | |
3257 | ||
77ab86bf | 3258 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3259 | if (ret) |
3260 | goto out_put; | |
3261 | } | |
3262 | ||
3263 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3264 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3265 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3266 | /* |
3267 | * don't bother trying to write stuff out _if_ | |
3268 | * a) we're not cached, | |
1a79c1f2 LB |
3269 | * b) we're with nospace_cache mount option, |
3270 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3271 | */ |
2b20982e | 3272 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3273 | spin_unlock(&block_group->lock); |
3274 | goto out_put; | |
3275 | } | |
3276 | spin_unlock(&block_group->lock); | |
3277 | ||
2968b1f4 JB |
3278 | /* |
3279 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3280 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3281 | */ | |
3282 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3283 | ret = -ENOSPC; | |
3284 | goto out_put; | |
3285 | } | |
3286 | ||
6fc823b1 JB |
3287 | /* |
3288 | * Try to preallocate enough space based on how big the block group is. | |
3289 | * Keep in mind this has to include any pinned space which could end up | |
3290 | * taking up quite a bit since it's not folded into the other space | |
3291 | * cache. | |
3292 | */ | |
ee22184b | 3293 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3294 | if (!num_pages) |
3295 | num_pages = 1; | |
3296 | ||
0af3d00b | 3297 | num_pages *= 16; |
09cbfeaf | 3298 | num_pages *= PAGE_SIZE; |
0af3d00b | 3299 | |
364ecf36 | 3300 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3301 | if (ret) |
3302 | goto out_put; | |
3303 | ||
3304 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3305 | num_pages, num_pages, | |
3306 | &alloc_hint); | |
2968b1f4 JB |
3307 | /* |
3308 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3309 | * of metadata or split extents when writing the cache out, which means | |
3310 | * we can enospc if we are heavily fragmented in addition to just normal | |
3311 | * out of space conditions. So if we hit this just skip setting up any | |
3312 | * other block groups for this transaction, maybe we'll unpin enough | |
3313 | * space the next time around. | |
3314 | */ | |
2b20982e JB |
3315 | if (!ret) |
3316 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3317 | else if (ret == -ENOSPC) |
3318 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3319 | |
0af3d00b JB |
3320 | out_put: |
3321 | iput(inode); | |
3322 | out_free: | |
b3b4aa74 | 3323 | btrfs_release_path(path); |
0af3d00b JB |
3324 | out: |
3325 | spin_lock(&block_group->lock); | |
e65cbb94 | 3326 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3327 | block_group->cache_generation = trans->transid; |
2b20982e | 3328 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3329 | spin_unlock(&block_group->lock); |
3330 | ||
364ecf36 | 3331 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3332 | return ret; |
3333 | } | |
3334 | ||
bbebb3e0 | 3335 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) |
dcdf7f6d | 3336 | { |
bbebb3e0 | 3337 | struct btrfs_fs_info *fs_info = trans->fs_info; |
dcdf7f6d JB |
3338 | struct btrfs_block_group_cache *cache, *tmp; |
3339 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3340 | struct btrfs_path *path; | |
3341 | ||
3342 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3343 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3344 | return 0; |
3345 | ||
3346 | path = btrfs_alloc_path(); | |
3347 | if (!path) | |
3348 | return -ENOMEM; | |
3349 | ||
3350 | /* Could add new block groups, use _safe just in case */ | |
3351 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3352 | dirty_list) { | |
3353 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3354 | cache_save_setup(cache, trans, path); | |
3355 | } | |
3356 | ||
3357 | btrfs_free_path(path); | |
3358 | return 0; | |
3359 | } | |
3360 | ||
1bbc621e CM |
3361 | /* |
3362 | * transaction commit does final block group cache writeback during a | |
3363 | * critical section where nothing is allowed to change the FS. This is | |
3364 | * required in order for the cache to actually match the block group, | |
3365 | * but can introduce a lot of latency into the commit. | |
3366 | * | |
3367 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3368 | * cache IO. There's a chance we'll have to redo some of it if the | |
3369 | * block group changes again during the commit, but it greatly reduces | |
3370 | * the commit latency by getting rid of the easy block groups while | |
3371 | * we're still allowing others to join the commit. | |
3372 | */ | |
21217054 | 3373 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3374 | { |
21217054 | 3375 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3376 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3377 | struct btrfs_transaction *cur_trans = trans->transaction; |
3378 | int ret = 0; | |
c9dc4c65 | 3379 | int should_put; |
1bbc621e CM |
3380 | struct btrfs_path *path = NULL; |
3381 | LIST_HEAD(dirty); | |
3382 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3383 | int num_started = 0; |
1bbc621e CM |
3384 | int loops = 0; |
3385 | ||
3386 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3387 | if (list_empty(&cur_trans->dirty_bgs)) { |
3388 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3389 | return 0; | |
1bbc621e | 3390 | } |
b58d1a9e | 3391 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3392 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3393 | |
1bbc621e | 3394 | again: |
1bbc621e CM |
3395 | /* |
3396 | * make sure all the block groups on our dirty list actually | |
3397 | * exist | |
3398 | */ | |
6c686b35 | 3399 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3400 | |
3401 | if (!path) { | |
3402 | path = btrfs_alloc_path(); | |
3403 | if (!path) | |
3404 | return -ENOMEM; | |
3405 | } | |
3406 | ||
b58d1a9e FM |
3407 | /* |
3408 | * cache_write_mutex is here only to save us from balance or automatic | |
3409 | * removal of empty block groups deleting this block group while we are | |
3410 | * writing out the cache | |
3411 | */ | |
3412 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3413 | while (!list_empty(&dirty)) { |
ba2c4d4e JB |
3414 | bool drop_reserve = true; |
3415 | ||
1bbc621e CM |
3416 | cache = list_first_entry(&dirty, |
3417 | struct btrfs_block_group_cache, | |
3418 | dirty_list); | |
1bbc621e CM |
3419 | /* |
3420 | * this can happen if something re-dirties a block | |
3421 | * group that is already under IO. Just wait for it to | |
3422 | * finish and then do it all again | |
3423 | */ | |
3424 | if (!list_empty(&cache->io_list)) { | |
3425 | list_del_init(&cache->io_list); | |
afdb5718 | 3426 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3427 | btrfs_put_block_group(cache); |
3428 | } | |
3429 | ||
3430 | ||
3431 | /* | |
3432 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3433 | * if it should update the cache_state. Don't delete | |
3434 | * until after we wait. | |
3435 | * | |
3436 | * Since we're not running in the commit critical section | |
3437 | * we need the dirty_bgs_lock to protect from update_block_group | |
3438 | */ | |
3439 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3440 | list_del_init(&cache->dirty_list); | |
3441 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3442 | ||
3443 | should_put = 1; | |
3444 | ||
3445 | cache_save_setup(cache, trans, path); | |
3446 | ||
3447 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3448 | cache->io_ctl.inode = NULL; | |
fe041534 | 3449 | ret = btrfs_write_out_cache(trans, cache, path); |
1bbc621e CM |
3450 | if (ret == 0 && cache->io_ctl.inode) { |
3451 | num_started++; | |
3452 | should_put = 0; | |
3453 | ||
3454 | /* | |
45ae2c18 NB |
3455 | * The cache_write_mutex is protecting the |
3456 | * io_list, also refer to the definition of | |
3457 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3458 | */ |
3459 | list_add_tail(&cache->io_list, io); | |
3460 | } else { | |
3461 | /* | |
3462 | * if we failed to write the cache, the | |
3463 | * generation will be bad and life goes on | |
3464 | */ | |
3465 | ret = 0; | |
3466 | } | |
3467 | } | |
ff1f8250 | 3468 | if (!ret) { |
39db232d | 3469 | ret = write_one_cache_group(trans, path, cache); |
ff1f8250 FM |
3470 | /* |
3471 | * Our block group might still be attached to the list | |
3472 | * of new block groups in the transaction handle of some | |
3473 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3474 | * means its block group item isn't yet in the extent | |
3475 | * tree. If this happens ignore the error, as we will | |
3476 | * try again later in the critical section of the | |
3477 | * transaction commit. | |
3478 | */ | |
3479 | if (ret == -ENOENT) { | |
3480 | ret = 0; | |
3481 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3482 | if (list_empty(&cache->dirty_list)) { | |
3483 | list_add_tail(&cache->dirty_list, | |
3484 | &cur_trans->dirty_bgs); | |
3485 | btrfs_get_block_group(cache); | |
ba2c4d4e | 3486 | drop_reserve = false; |
ff1f8250 FM |
3487 | } |
3488 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3489 | } else if (ret) { | |
66642832 | 3490 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3491 | } |
3492 | } | |
1bbc621e | 3493 | |
52042d8e | 3494 | /* if it's not on the io list, we need to put the block group */ |
1bbc621e CM |
3495 | if (should_put) |
3496 | btrfs_put_block_group(cache); | |
ba2c4d4e JB |
3497 | if (drop_reserve) |
3498 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1bbc621e CM |
3499 | |
3500 | if (ret) | |
3501 | break; | |
b58d1a9e FM |
3502 | |
3503 | /* | |
3504 | * Avoid blocking other tasks for too long. It might even save | |
3505 | * us from writing caches for block groups that are going to be | |
3506 | * removed. | |
3507 | */ | |
3508 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3509 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3510 | } |
b58d1a9e | 3511 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3512 | |
3513 | /* | |
3514 | * go through delayed refs for all the stuff we've just kicked off | |
3515 | * and then loop back (just once) | |
3516 | */ | |
c79a70b1 | 3517 | ret = btrfs_run_delayed_refs(trans, 0); |
1bbc621e CM |
3518 | if (!ret && loops == 0) { |
3519 | loops++; | |
3520 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3521 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3522 | /* |
3523 | * dirty_bgs_lock protects us from concurrent block group | |
3524 | * deletes too (not just cache_write_mutex). | |
3525 | */ | |
3526 | if (!list_empty(&dirty)) { | |
3527 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3528 | goto again; | |
3529 | } | |
1bbc621e | 3530 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3531 | } else if (ret < 0) { |
2ff7e61e | 3532 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3533 | } |
3534 | ||
3535 | btrfs_free_path(path); | |
3536 | return ret; | |
3537 | } | |
3538 | ||
5742d15f | 3539 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) |
1bbc621e | 3540 | { |
5742d15f | 3541 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1bbc621e CM |
3542 | struct btrfs_block_group_cache *cache; |
3543 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3544 | int ret = 0; | |
3545 | int should_put; | |
3546 | struct btrfs_path *path; | |
3547 | struct list_head *io = &cur_trans->io_bgs; | |
3548 | int num_started = 0; | |
9078a3e1 CM |
3549 | |
3550 | path = btrfs_alloc_path(); | |
3551 | if (!path) | |
3552 | return -ENOMEM; | |
3553 | ||
ce93ec54 | 3554 | /* |
e44081ef FM |
3555 | * Even though we are in the critical section of the transaction commit, |
3556 | * we can still have concurrent tasks adding elements to this | |
3557 | * transaction's list of dirty block groups. These tasks correspond to | |
3558 | * endio free space workers started when writeback finishes for a | |
3559 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3560 | * allocate new block groups as a result of COWing nodes of the root | |
3561 | * tree when updating the free space inode. The writeback for the space | |
3562 | * caches is triggered by an earlier call to | |
3563 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3564 | * loop. | |
3565 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3566 | * delayed refs to make sure we have the best chance at doing this all |
3567 | * in one shot. | |
3568 | */ | |
e44081ef | 3569 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3570 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3571 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3572 | struct btrfs_block_group_cache, | |
3573 | dirty_list); | |
c9dc4c65 CM |
3574 | |
3575 | /* | |
3576 | * this can happen if cache_save_setup re-dirties a block | |
3577 | * group that is already under IO. Just wait for it to | |
3578 | * finish and then do it all again | |
3579 | */ | |
3580 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3581 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3582 | list_del_init(&cache->io_list); |
afdb5718 | 3583 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3584 | btrfs_put_block_group(cache); |
e44081ef | 3585 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3586 | } |
3587 | ||
1bbc621e CM |
3588 | /* |
3589 | * don't remove from the dirty list until after we've waited | |
3590 | * on any pending IO | |
3591 | */ | |
ce93ec54 | 3592 | list_del_init(&cache->dirty_list); |
e44081ef | 3593 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3594 | should_put = 1; |
3595 | ||
1bbc621e | 3596 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3597 | |
ce93ec54 | 3598 | if (!ret) |
c79a70b1 | 3599 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 3600 | (unsigned long) -1); |
c9dc4c65 CM |
3601 | |
3602 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3603 | cache->io_ctl.inode = NULL; | |
fe041534 | 3604 | ret = btrfs_write_out_cache(trans, cache, path); |
c9dc4c65 CM |
3605 | if (ret == 0 && cache->io_ctl.inode) { |
3606 | num_started++; | |
3607 | should_put = 0; | |
1bbc621e | 3608 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3609 | } else { |
3610 | /* | |
3611 | * if we failed to write the cache, the | |
3612 | * generation will be bad and life goes on | |
3613 | */ | |
3614 | ret = 0; | |
3615 | } | |
3616 | } | |
ff1f8250 | 3617 | if (!ret) { |
39db232d | 3618 | ret = write_one_cache_group(trans, path, cache); |
2bc0bb5f FM |
3619 | /* |
3620 | * One of the free space endio workers might have | |
3621 | * created a new block group while updating a free space | |
3622 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3623 | * and hasn't released its transaction handle yet, in | |
3624 | * which case the new block group is still attached to | |
3625 | * its transaction handle and its creation has not | |
3626 | * finished yet (no block group item in the extent tree | |
3627 | * yet, etc). If this is the case, wait for all free | |
3628 | * space endio workers to finish and retry. This is a | |
3629 | * a very rare case so no need for a more efficient and | |
3630 | * complex approach. | |
3631 | */ | |
3632 | if (ret == -ENOENT) { | |
3633 | wait_event(cur_trans->writer_wait, | |
3634 | atomic_read(&cur_trans->num_writers) == 1); | |
39db232d | 3635 | ret = write_one_cache_group(trans, path, cache); |
2bc0bb5f | 3636 | } |
ff1f8250 | 3637 | if (ret) |
66642832 | 3638 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3639 | } |
c9dc4c65 CM |
3640 | |
3641 | /* if its not on the io list, we need to put the block group */ | |
3642 | if (should_put) | |
3643 | btrfs_put_block_group(cache); | |
ba2c4d4e | 3644 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
e44081ef | 3645 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3646 | } |
e44081ef | 3647 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3648 | |
45ae2c18 NB |
3649 | /* |
3650 | * Refer to the definition of io_bgs member for details why it's safe | |
3651 | * to use it without any locking | |
3652 | */ | |
1bbc621e CM |
3653 | while (!list_empty(io)) { |
3654 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3655 | io_list); |
3656 | list_del_init(&cache->io_list); | |
afdb5718 | 3657 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3658 | btrfs_put_block_group(cache); |
3659 | } | |
3660 | ||
9078a3e1 | 3661 | btrfs_free_path(path); |
ce93ec54 | 3662 | return ret; |
9078a3e1 CM |
3663 | } |
3664 | ||
2ff7e61e | 3665 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3666 | { |
3667 | struct btrfs_block_group_cache *block_group; | |
3668 | int readonly = 0; | |
3669 | ||
0b246afa | 3670 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3671 | if (!block_group || block_group->ro) |
3672 | readonly = 1; | |
3673 | if (block_group) | |
fa9c0d79 | 3674 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3675 | return readonly; |
3676 | } | |
3677 | ||
f78c436c FM |
3678 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3679 | { | |
3680 | struct btrfs_block_group_cache *bg; | |
3681 | bool ret = true; | |
3682 | ||
3683 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3684 | if (!bg) | |
3685 | return false; | |
3686 | ||
3687 | spin_lock(&bg->lock); | |
3688 | if (bg->ro) | |
3689 | ret = false; | |
3690 | else | |
3691 | atomic_inc(&bg->nocow_writers); | |
3692 | spin_unlock(&bg->lock); | |
3693 | ||
3694 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3695 | if (!ret) | |
3696 | btrfs_put_block_group(bg); | |
3697 | ||
3698 | return ret; | |
3699 | ||
3700 | } | |
3701 | ||
3702 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3703 | { | |
3704 | struct btrfs_block_group_cache *bg; | |
3705 | ||
3706 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3707 | ASSERT(bg); | |
3708 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
4625956a | 3709 | wake_up_var(&bg->nocow_writers); |
f78c436c FM |
3710 | /* |
3711 | * Once for our lookup and once for the lookup done by a previous call | |
3712 | * to btrfs_inc_nocow_writers() | |
3713 | */ | |
3714 | btrfs_put_block_group(bg); | |
3715 | btrfs_put_block_group(bg); | |
3716 | } | |
3717 | ||
f78c436c FM |
3718 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
3719 | { | |
4625956a | 3720 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); |
f78c436c FM |
3721 | } |
3722 | ||
8790d502 CM |
3723 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3724 | { | |
899c81ea ID |
3725 | u64 extra_flags = chunk_to_extended(flags) & |
3726 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3727 | |
de98ced9 | 3728 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3729 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3730 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3731 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3732 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3733 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3734 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3735 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3736 | } |
593060d7 | 3737 | |
fc67c450 ID |
3738 | /* |
3739 | * returns target flags in extended format or 0 if restripe for this | |
3740 | * chunk_type is not in progress | |
c6664b42 | 3741 | * |
dccdb07b | 3742 | * should be called with balance_lock held |
fc67c450 ID |
3743 | */ |
3744 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3745 | { | |
3746 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3747 | u64 target = 0; | |
3748 | ||
fc67c450 ID |
3749 | if (!bctl) |
3750 | return 0; | |
3751 | ||
3752 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3753 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3754 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3755 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3756 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3757 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3758 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3759 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3760 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3761 | } | |
3762 | ||
3763 | return target; | |
3764 | } | |
3765 | ||
a46d11a8 ID |
3766 | /* |
3767 | * @flags: available profiles in extended format (see ctree.h) | |
3768 | * | |
e4d8ec0f ID |
3769 | * Returns reduced profile in chunk format. If profile changing is in |
3770 | * progress (either running or paused) picks the target profile (if it's | |
3771 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3772 | */ |
2ff7e61e | 3773 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 3774 | { |
0b246afa | 3775 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 3776 | u64 target; |
9c170b26 ZL |
3777 | u64 raid_type; |
3778 | u64 allowed = 0; | |
a061fc8d | 3779 | |
fc67c450 ID |
3780 | /* |
3781 | * see if restripe for this chunk_type is in progress, if so | |
3782 | * try to reduce to the target profile | |
3783 | */ | |
0b246afa JM |
3784 | spin_lock(&fs_info->balance_lock); |
3785 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
3786 | if (target) { |
3787 | /* pick target profile only if it's already available */ | |
3788 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 3789 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 3790 | return extended_to_chunk(target); |
e4d8ec0f ID |
3791 | } |
3792 | } | |
0b246afa | 3793 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 3794 | |
53b381b3 | 3795 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
3796 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
3797 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
41a6e891 | 3798 | allowed |= btrfs_raid_array[raid_type].bg_flag; |
9c170b26 ZL |
3799 | } |
3800 | allowed &= flags; | |
3801 | ||
3802 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
3803 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
3804 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
3805 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
3806 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
3807 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
3808 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
3809 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
3810 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
3811 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
3812 | ||
3813 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
3814 | ||
3815 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
3816 | } |
3817 | ||
2ff7e61e | 3818 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 3819 | { |
de98ced9 | 3820 | unsigned seq; |
f8213bdc | 3821 | u64 flags; |
de98ced9 MX |
3822 | |
3823 | do { | |
f8213bdc | 3824 | flags = orig_flags; |
0b246afa | 3825 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
3826 | |
3827 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 3828 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 3829 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 3830 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 3831 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
3832 | flags |= fs_info->avail_metadata_alloc_bits; |
3833 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 3834 | |
2ff7e61e | 3835 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
3836 | } |
3837 | ||
1b86826d | 3838 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 3839 | { |
0b246afa | 3840 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 3841 | u64 flags; |
53b381b3 | 3842 | u64 ret; |
9ed74f2d | 3843 | |
b742bb82 YZ |
3844 | if (data) |
3845 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 3846 | else if (root == fs_info->chunk_root) |
b742bb82 | 3847 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 3848 | else |
b742bb82 | 3849 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3850 | |
2ff7e61e | 3851 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 3852 | return ret; |
6a63209f | 3853 | } |
9ed74f2d | 3854 | |
1b86826d JM |
3855 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
3856 | { | |
3857 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
3858 | } | |
3859 | ||
3860 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
3861 | { | |
3862 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
3863 | } | |
3864 | ||
3865 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
3866 | { | |
3867 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3868 | } | |
3869 | ||
04f4f916 | 3870 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 3871 | { |
04f4f916 | 3872 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 3873 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 3874 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 3875 | u64 used; |
94b947b2 | 3876 | int ret = 0; |
c99f1b0c ZL |
3877 | int need_commit = 2; |
3878 | int have_pinned_space; | |
6a63209f | 3879 | |
6a63209f | 3880 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 3881 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 3882 | |
9dced186 | 3883 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 3884 | need_commit = 0; |
9dced186 | 3885 | ASSERT(current->journal_info); |
0af3d00b JB |
3886 | } |
3887 | ||
6a63209f JB |
3888 | again: |
3889 | /* make sure we have enough space to handle the data first */ | |
3890 | spin_lock(&data_sinfo->lock); | |
4136135b | 3891 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
3892 | |
3893 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3894 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3895 | |
6a63209f JB |
3896 | /* |
3897 | * if we don't have enough free bytes in this space then we need | |
3898 | * to alloc a new chunk. | |
3899 | */ | |
b9fd47cd | 3900 | if (!data_sinfo->full) { |
6a63209f | 3901 | u64 alloc_target; |
9ed74f2d | 3902 | |
0e4f8f88 | 3903 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3904 | spin_unlock(&data_sinfo->lock); |
1174cade | 3905 | |
1b86826d | 3906 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
3907 | /* |
3908 | * It is ugly that we don't call nolock join | |
3909 | * transaction for the free space inode case here. | |
3910 | * But it is safe because we only do the data space | |
3911 | * reservation for the free space cache in the | |
3912 | * transaction context, the common join transaction | |
3913 | * just increase the counter of the current transaction | |
3914 | * handler, doesn't try to acquire the trans_lock of | |
3915 | * the fs. | |
3916 | */ | |
7a7eaa40 | 3917 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3918 | if (IS_ERR(trans)) |
3919 | return PTR_ERR(trans); | |
9ed74f2d | 3920 | |
fc471cb0 JB |
3921 | ret = btrfs_chunk_alloc(trans, alloc_target, |
3922 | CHUNK_ALLOC_NO_FORCE); | |
3a45bb20 | 3923 | btrfs_end_transaction(trans); |
d52a5b5f MX |
3924 | if (ret < 0) { |
3925 | if (ret != -ENOSPC) | |
3926 | return ret; | |
c99f1b0c ZL |
3927 | else { |
3928 | have_pinned_space = 1; | |
d52a5b5f | 3929 | goto commit_trans; |
c99f1b0c | 3930 | } |
d52a5b5f | 3931 | } |
9ed74f2d | 3932 | |
6a63209f JB |
3933 | goto again; |
3934 | } | |
f2bb8f5c JB |
3935 | |
3936 | /* | |
b150a4f1 | 3937 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
3938 | * allocation, and no removed chunk in current transaction, |
3939 | * don't bother committing the transaction. | |
f2bb8f5c | 3940 | */ |
dec59fa3 | 3941 | have_pinned_space = __percpu_counter_compare( |
c99f1b0c | 3942 | &data_sinfo->total_bytes_pinned, |
dec59fa3 EL |
3943 | used + bytes - data_sinfo->total_bytes, |
3944 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
6a63209f | 3945 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3946 | |
4e06bdd6 | 3947 | /* commit the current transaction and try again */ |
d52a5b5f | 3948 | commit_trans: |
92e2f7e3 | 3949 | if (need_commit) { |
c99f1b0c | 3950 | need_commit--; |
b150a4f1 | 3951 | |
e1746e83 | 3952 | if (need_commit > 0) { |
82b3e53b | 3953 | btrfs_start_delalloc_roots(fs_info, -1); |
6374e57a | 3954 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 3955 | (u64)-1); |
e1746e83 | 3956 | } |
9a4e7276 | 3957 | |
7a7eaa40 | 3958 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3959 | if (IS_ERR(trans)) |
3960 | return PTR_ERR(trans); | |
c99f1b0c | 3961 | if (have_pinned_space >= 0 || |
3204d33c JB |
3962 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
3963 | &trans->transaction->flags) || | |
c99f1b0c | 3964 | need_commit > 0) { |
3a45bb20 | 3965 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
3966 | if (ret) |
3967 | return ret; | |
d7c15171 | 3968 | /* |
c2d6cb16 FM |
3969 | * The cleaner kthread might still be doing iput |
3970 | * operations. Wait for it to finish so that | |
034f784d JB |
3971 | * more space is released. We don't need to |
3972 | * explicitly run the delayed iputs here because | |
3973 | * the commit_transaction would have woken up | |
3974 | * the cleaner. | |
d7c15171 | 3975 | */ |
034f784d JB |
3976 | ret = btrfs_wait_on_delayed_iputs(fs_info); |
3977 | if (ret) | |
3978 | return ret; | |
94b947b2 ZL |
3979 | goto again; |
3980 | } else { | |
3a45bb20 | 3981 | btrfs_end_transaction(trans); |
94b947b2 | 3982 | } |
4e06bdd6 | 3983 | } |
9ed74f2d | 3984 | |
0b246afa | 3985 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
3986 | "space_info:enospc", |
3987 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
3988 | return -ENOSPC; |
3989 | } | |
bb96c4e5 | 3990 | btrfs_space_info_update_bytes_may_use(fs_info, data_sinfo, bytes); |
0b246afa | 3991 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 3992 | data_sinfo->flags, bytes, 1); |
6a63209f | 3993 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3994 | |
4559b0a7 | 3995 | return 0; |
9ed74f2d | 3996 | } |
6a63209f | 3997 | |
364ecf36 QW |
3998 | int btrfs_check_data_free_space(struct inode *inode, |
3999 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4000 | { |
0b246afa | 4001 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4002 | int ret; |
4003 | ||
4004 | /* align the range */ | |
0b246afa JM |
4005 | len = round_up(start + len, fs_info->sectorsize) - |
4006 | round_down(start, fs_info->sectorsize); | |
4007 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4008 | |
04f4f916 | 4009 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4010 | if (ret < 0) |
4011 | return ret; | |
4012 | ||
1e5ec2e7 | 4013 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4014 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4015 | if (ret < 0) |
1e5ec2e7 | 4016 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4017 | else |
4018 | ret = 0; | |
4ceff079 QW |
4019 | return ret; |
4020 | } | |
4021 | ||
4ceff079 QW |
4022 | /* |
4023 | * Called if we need to clear a data reservation for this inode | |
4024 | * Normally in a error case. | |
4025 | * | |
51773bec QW |
4026 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4027 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4028 | * Like clear_bit_hook(). | |
4ceff079 | 4029 | */ |
51773bec QW |
4030 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4031 | u64 len) | |
4ceff079 | 4032 | { |
0b246afa | 4033 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4034 | struct btrfs_space_info *data_sinfo; |
4035 | ||
4036 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4037 | len = round_up(start + len, fs_info->sectorsize) - |
4038 | round_down(start, fs_info->sectorsize); | |
4039 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4040 | |
0b246afa | 4041 | data_sinfo = fs_info->data_sinfo; |
4ceff079 | 4042 | spin_lock(&data_sinfo->lock); |
bb96c4e5 | 4043 | btrfs_space_info_update_bytes_may_use(fs_info, data_sinfo, -len); |
0b246afa | 4044 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4045 | data_sinfo->flags, len, 0); |
4046 | spin_unlock(&data_sinfo->lock); | |
4047 | } | |
4048 | ||
51773bec QW |
4049 | /* |
4050 | * Called if we need to clear a data reservation for this inode | |
4051 | * Normally in a error case. | |
4052 | * | |
01327610 | 4053 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4054 | * space framework. |
4055 | */ | |
bc42bda2 QW |
4056 | void btrfs_free_reserved_data_space(struct inode *inode, |
4057 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4058 | { |
0c476a5d JM |
4059 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4060 | ||
4061 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4062 | len = round_up(start + len, root->fs_info->sectorsize) - |
4063 | round_down(start, root->fs_info->sectorsize); | |
4064 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4065 | |
51773bec | 4066 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4067 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4068 | } |
4069 | ||
97e728d4 | 4070 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4071 | { |
97e728d4 JB |
4072 | struct list_head *head = &info->space_info; |
4073 | struct btrfs_space_info *found; | |
e3ccfa98 | 4074 | |
97e728d4 JB |
4075 | rcu_read_lock(); |
4076 | list_for_each_entry_rcu(found, head, list) { | |
4077 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4078 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4079 | } |
97e728d4 | 4080 | rcu_read_unlock(); |
e3ccfa98 JB |
4081 | } |
4082 | ||
2ff7e61e | 4083 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4084 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4085 | { |
8d8aafee | 4086 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4087 | u64 thresh; |
e3ccfa98 | 4088 | |
0e4f8f88 CM |
4089 | if (force == CHUNK_ALLOC_FORCE) |
4090 | return 1; | |
4091 | ||
4092 | /* | |
4093 | * in limited mode, we want to have some free space up to | |
4094 | * about 1% of the FS size. | |
4095 | */ | |
4096 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4097 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4098 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4099 | |
8d8aafee | 4100 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4101 | return 1; |
4102 | } | |
0e4f8f88 | 4103 | |
8d8aafee | 4104 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4105 | return 0; |
424499db | 4106 | return 1; |
32c00aff JB |
4107 | } |
4108 | ||
2ff7e61e | 4109 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4110 | { |
4111 | u64 num_dev; | |
4112 | ||
9fa02ac7 DS |
4113 | num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; |
4114 | if (!num_dev) | |
0b246afa | 4115 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 | 4116 | |
39c2d7fa | 4117 | return num_dev; |
15d1ff81 LB |
4118 | } |
4119 | ||
39c2d7fa FM |
4120 | /* |
4121 | * If @is_allocation is true, reserve space in the system space info necessary | |
4122 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4123 | * removing a chunk. | |
4124 | */ | |
451a2c13 | 4125 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) |
15d1ff81 | 4126 | { |
451a2c13 | 4127 | struct btrfs_fs_info *fs_info = trans->fs_info; |
15d1ff81 LB |
4128 | struct btrfs_space_info *info; |
4129 | u64 left; | |
4130 | u64 thresh; | |
4fbcdf66 | 4131 | int ret = 0; |
39c2d7fa | 4132 | u64 num_devs; |
4fbcdf66 FM |
4133 | |
4134 | /* | |
4135 | * Needed because we can end up allocating a system chunk and for an | |
4136 | * atomic and race free space reservation in the chunk block reserve. | |
4137 | */ | |
a32bf9a3 | 4138 | lockdep_assert_held(&fs_info->chunk_mutex); |
15d1ff81 | 4139 | |
280c2908 | 4140 | info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4141 | spin_lock(&info->lock); |
4136135b | 4142 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4143 | spin_unlock(&info->lock); |
4144 | ||
2ff7e61e | 4145 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4146 | |
4147 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4148 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4149 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4150 | |
0b246afa JM |
4151 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4152 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4153 | left, thresh, type); | |
5da6afeb | 4154 | btrfs_dump_space_info(fs_info, info, 0, 0); |
15d1ff81 LB |
4155 | } |
4156 | ||
4157 | if (left < thresh) { | |
1b86826d | 4158 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4159 | |
4fbcdf66 FM |
4160 | /* |
4161 | * Ignore failure to create system chunk. We might end up not | |
4162 | * needing it, as we might not need to COW all nodes/leafs from | |
4163 | * the paths we visit in the chunk tree (they were already COWed | |
4164 | * or created in the current transaction for example). | |
4165 | */ | |
c216b203 | 4166 | ret = btrfs_alloc_chunk(trans, flags); |
4fbcdf66 FM |
4167 | } |
4168 | ||
4169 | if (!ret) { | |
0b246afa JM |
4170 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4171 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4172 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4173 | if (!ret) | |
4174 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4175 | } |
4176 | } | |
4177 | ||
28b737f6 LB |
4178 | /* |
4179 | * If force is CHUNK_ALLOC_FORCE: | |
4180 | * - return 1 if it successfully allocates a chunk, | |
4181 | * - return errors including -ENOSPC otherwise. | |
4182 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4183 | * - return 0 if it doesn't need to allocate a new chunk, | |
4184 | * - return 1 if it successfully allocates a chunk, | |
4185 | * - return errors including -ENOSPC otherwise. | |
4186 | */ | |
fc471cb0 JB |
4187 | int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
4188 | enum btrfs_chunk_alloc_enum force) | |
9ed74f2d | 4189 | { |
01458828 | 4190 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 4191 | struct btrfs_space_info *space_info; |
2556fbb0 NB |
4192 | bool wait_for_alloc = false; |
4193 | bool should_alloc = false; | |
9ed74f2d | 4194 | int ret = 0; |
9ed74f2d | 4195 | |
c6b305a8 JB |
4196 | /* Don't re-enter if we're already allocating a chunk */ |
4197 | if (trans->allocating_chunk) | |
4198 | return -ENOSPC; | |
4199 | ||
280c2908 | 4200 | space_info = btrfs_find_space_info(fs_info, flags); |
dc2d3005 | 4201 | ASSERT(space_info); |
9ed74f2d | 4202 | |
2556fbb0 NB |
4203 | do { |
4204 | spin_lock(&space_info->lock); | |
4205 | if (force < space_info->force_alloc) | |
4206 | force = space_info->force_alloc; | |
4207 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
4208 | if (space_info->full) { | |
4209 | /* No more free physical space */ | |
4210 | if (should_alloc) | |
4211 | ret = -ENOSPC; | |
4212 | else | |
4213 | ret = 0; | |
4214 | spin_unlock(&space_info->lock); | |
4215 | return ret; | |
4216 | } else if (!should_alloc) { | |
4217 | spin_unlock(&space_info->lock); | |
4218 | return 0; | |
4219 | } else if (space_info->chunk_alloc) { | |
4220 | /* | |
4221 | * Someone is already allocating, so we need to block | |
4222 | * until this someone is finished and then loop to | |
4223 | * recheck if we should continue with our allocation | |
4224 | * attempt. | |
4225 | */ | |
4226 | wait_for_alloc = true; | |
4227 | spin_unlock(&space_info->lock); | |
4228 | mutex_lock(&fs_info->chunk_mutex); | |
4229 | mutex_unlock(&fs_info->chunk_mutex); | |
4230 | } else { | |
4231 | /* Proceed with allocation */ | |
4232 | space_info->chunk_alloc = 1; | |
4233 | wait_for_alloc = false; | |
4234 | spin_unlock(&space_info->lock); | |
4235 | } | |
6d74119f | 4236 | |
1e1c50a9 | 4237 | cond_resched(); |
2556fbb0 | 4238 | } while (wait_for_alloc); |
6d74119f | 4239 | |
2556fbb0 | 4240 | mutex_lock(&fs_info->chunk_mutex); |
c6b305a8 JB |
4241 | trans->allocating_chunk = true; |
4242 | ||
67377734 JB |
4243 | /* |
4244 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4245 | * allocating mixed chunks instead of individual chunks. | |
4246 | */ | |
4247 | if (btrfs_mixed_space_info(space_info)) | |
4248 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4249 | ||
97e728d4 JB |
4250 | /* |
4251 | * if we're doing a data chunk, go ahead and make sure that | |
4252 | * we keep a reasonable number of metadata chunks allocated in the | |
4253 | * FS as well. | |
4254 | */ | |
9ed74f2d | 4255 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4256 | fs_info->data_chunk_allocations++; |
4257 | if (!(fs_info->data_chunk_allocations % | |
4258 | fs_info->metadata_ratio)) | |
4259 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4260 | } |
4261 | ||
15d1ff81 LB |
4262 | /* |
4263 | * Check if we have enough space in SYSTEM chunk because we may need | |
4264 | * to update devices. | |
4265 | */ | |
451a2c13 | 4266 | check_system_chunk(trans, flags); |
15d1ff81 | 4267 | |
c216b203 | 4268 | ret = btrfs_alloc_chunk(trans, flags); |
c6b305a8 | 4269 | trans->allocating_chunk = false; |
92b8e897 | 4270 | |
9ed74f2d | 4271 | spin_lock(&space_info->lock); |
57f1642e NB |
4272 | if (ret < 0) { |
4273 | if (ret == -ENOSPC) | |
4274 | space_info->full = 1; | |
4275 | else | |
4276 | goto out; | |
4277 | } else { | |
424499db | 4278 | ret = 1; |
21a94f7a | 4279 | space_info->max_extent_size = 0; |
57f1642e | 4280 | } |
6d74119f | 4281 | |
0e4f8f88 | 4282 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4283 | out: |
6d74119f | 4284 | space_info->chunk_alloc = 0; |
9ed74f2d | 4285 | spin_unlock(&space_info->lock); |
a25c75d5 | 4286 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4287 | /* |
4288 | * When we allocate a new chunk we reserve space in the chunk block | |
4289 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4290 | * add new nodes/leafs to it if we end up needing to do it when | |
4291 | * inserting the chunk item and updating device items as part of the | |
4292 | * second phase of chunk allocation, performed by | |
4293 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4294 | * large number of new block groups to create in our transaction | |
4295 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4296 | * in extreme cases - like having a single transaction create many new | |
4297 | * block groups when starting to write out the free space caches of all | |
4298 | * the block groups that were made dirty during the lifetime of the | |
4299 | * transaction. | |
4300 | */ | |
5ce55557 | 4301 | if (trans->chunk_bytes_reserved >= (u64)SZ_2M) |
6c686b35 | 4302 | btrfs_create_pending_block_groups(trans); |
5ce55557 | 4303 | |
0f9dd46c | 4304 | return ret; |
6324fbf3 | 4305 | } |
9ed74f2d | 4306 | |
69fe2d75 JB |
4307 | /** |
4308 | * btrfs_inode_rsv_release - release any excessive reservation. | |
4309 | * @inode - the inode we need to release from. | |
43b18595 QW |
4310 | * @qgroup_free - free or convert qgroup meta. |
4311 | * Unlike normal operation, qgroup meta reservation needs to know if we are | |
4312 | * freeing qgroup reservation or just converting it into per-trans. Normally | |
4313 | * @qgroup_free is true for error handling, and false for normal release. | |
69fe2d75 JB |
4314 | * |
4315 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
4316 | * tracepoint for the reservation. | |
4317 | */ | |
43b18595 | 4318 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
69fe2d75 JB |
4319 | { |
4320 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
69fe2d75 JB |
4321 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
4322 | u64 released = 0; | |
ff6bc37e | 4323 | u64 qgroup_to_release = 0; |
69fe2d75 JB |
4324 | |
4325 | /* | |
4326 | * Since we statically set the block_rsv->size we just want to say we | |
4327 | * are releasing 0 bytes, and then we'll just get the reservation over | |
4328 | * the size free'd. | |
4329 | */ | |
ba2c4d4e JB |
4330 | released = __btrfs_block_rsv_release(fs_info, block_rsv, 0, |
4331 | &qgroup_to_release); | |
69fe2d75 JB |
4332 | if (released > 0) |
4333 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
4334 | btrfs_ino(inode), released, 0); | |
43b18595 | 4335 | if (qgroup_free) |
ff6bc37e | 4336 | btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); |
43b18595 | 4337 | else |
ff6bc37e QW |
4338 | btrfs_qgroup_convert_reserved_meta(inode->root, |
4339 | qgroup_to_release); | |
69fe2d75 JB |
4340 | } |
4341 | ||
d5c12070 MX |
4342 | /* |
4343 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
4344 | * root: the root of the parent directory | |
4345 | * rsv: block reservation | |
4346 | * items: the number of items that we need do reservation | |
a5b7f429 | 4347 | * use_global_rsv: allow fallback to the global block reservation |
d5c12070 MX |
4348 | * |
4349 | * This function is used to reserve the space for snapshot/subvolume | |
4350 | * creation and deletion. Those operations are different with the | |
4351 | * common file/directory operations, they change two fs/file trees | |
4352 | * and root tree, the number of items that the qgroup reserves is | |
4353 | * different with the free space reservation. So we can not use | |
01327610 | 4354 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
4355 | */ |
4356 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
a5b7f429 | 4357 | struct btrfs_block_rsv *rsv, int items, |
ee3441b4 | 4358 | bool use_global_rsv) |
a22285a6 | 4359 | { |
a5b7f429 | 4360 | u64 qgroup_num_bytes = 0; |
d5c12070 MX |
4361 | u64 num_bytes; |
4362 | int ret; | |
0b246afa JM |
4363 | struct btrfs_fs_info *fs_info = root->fs_info; |
4364 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 4365 | |
0b246afa | 4366 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 4367 | /* One for parent inode, two for dir entries */ |
a5b7f429 LF |
4368 | qgroup_num_bytes = 3 * fs_info->nodesize; |
4369 | ret = btrfs_qgroup_reserve_meta_prealloc(root, | |
4370 | qgroup_num_bytes, true); | |
d5c12070 MX |
4371 | if (ret) |
4372 | return ret; | |
d5c12070 MX |
4373 | } |
4374 | ||
0b246afa | 4375 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
280c2908 | 4376 | rsv->space_info = btrfs_find_space_info(fs_info, |
d5c12070 MX |
4377 | BTRFS_BLOCK_GROUP_METADATA); |
4378 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
4379 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
4380 | |
4381 | if (ret == -ENOSPC && use_global_rsv) | |
3a584174 | 4382 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true); |
ee3441b4 | 4383 | |
a5b7f429 LF |
4384 | if (ret && qgroup_num_bytes) |
4385 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
d5c12070 MX |
4386 | |
4387 | return ret; | |
4388 | } | |
4389 | ||
2ff7e61e | 4390 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 4391 | struct btrfs_block_rsv *rsv) |
d5c12070 | 4392 | { |
2ff7e61e | 4393 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
4394 | } |
4395 | ||
69fe2d75 JB |
4396 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
4397 | struct btrfs_inode *inode) | |
9e0baf60 | 4398 | { |
69fe2d75 JB |
4399 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
4400 | u64 reserve_size = 0; | |
ff6bc37e | 4401 | u64 qgroup_rsv_size = 0; |
69fe2d75 JB |
4402 | u64 csum_leaves; |
4403 | unsigned outstanding_extents; | |
9e0baf60 | 4404 | |
69fe2d75 JB |
4405 | lockdep_assert_held(&inode->lock); |
4406 | outstanding_extents = inode->outstanding_extents; | |
4407 | if (outstanding_extents) | |
4408 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
4409 | outstanding_extents + 1); | |
4410 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
4411 | inode->csum_bytes); | |
4412 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
4413 | csum_leaves); | |
ff6bc37e QW |
4414 | /* |
4415 | * For qgroup rsv, the calculation is very simple: | |
4416 | * account one nodesize for each outstanding extent | |
4417 | * | |
4418 | * This is overestimating in most cases. | |
4419 | */ | |
139a5617 | 4420 | qgroup_rsv_size = (u64)outstanding_extents * fs_info->nodesize; |
9e0baf60 | 4421 | |
69fe2d75 JB |
4422 | spin_lock(&block_rsv->lock); |
4423 | block_rsv->size = reserve_size; | |
ff6bc37e | 4424 | block_rsv->qgroup_rsv_size = qgroup_rsv_size; |
69fe2d75 | 4425 | spin_unlock(&block_rsv->lock); |
0ca1f7ce | 4426 | } |
c146afad | 4427 | |
c8eaeac7 JB |
4428 | static void calc_inode_reservations(struct btrfs_fs_info *fs_info, |
4429 | u64 num_bytes, u64 *meta_reserve, | |
4430 | u64 *qgroup_reserve) | |
4431 | { | |
4432 | u64 nr_extents = count_max_extents(num_bytes); | |
4433 | u64 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, num_bytes); | |
4434 | ||
4435 | /* We add one for the inode update at finish ordered time */ | |
4436 | *meta_reserve = btrfs_calc_trans_metadata_size(fs_info, | |
4437 | nr_extents + csum_leaves + 1); | |
4438 | *qgroup_reserve = nr_extents * fs_info->nodesize; | |
4439 | } | |
4440 | ||
9f3db423 | 4441 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 4442 | { |
c8eaeac7 JB |
4443 | struct btrfs_root *root = inode->root; |
4444 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4445 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
4446 | u64 meta_reserve, qgroup_reserve; | |
69fe2d75 | 4447 | unsigned nr_extents; |
08e007d2 | 4448 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 4449 | int ret = 0; |
c64c2bd8 | 4450 | bool delalloc_lock = true; |
6324fbf3 | 4451 | |
c64c2bd8 JB |
4452 | /* If we are a free space inode we need to not flush since we will be in |
4453 | * the middle of a transaction commit. We also don't need the delalloc | |
4454 | * mutex since we won't race with anybody. We need this mostly to make | |
4455 | * lockdep shut its filthy mouth. | |
bac357dc JB |
4456 | * |
4457 | * If we have a transaction open (can happen if we call truncate_block | |
4458 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
4459 | */ |
4460 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 4461 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 4462 | delalloc_lock = false; |
da07d4ab NB |
4463 | } else { |
4464 | if (current->journal_info) | |
4465 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 4466 | |
da07d4ab NB |
4467 | if (btrfs_transaction_in_commit(fs_info)) |
4468 | schedule_timeout(1); | |
4469 | } | |
ec44a35c | 4470 | |
c64c2bd8 | 4471 | if (delalloc_lock) |
9f3db423 | 4472 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 4473 | |
0b246afa | 4474 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 | 4475 | |
c8eaeac7 JB |
4476 | /* |
4477 | * We always want to do it this way, every other way is wrong and ends | |
4478 | * in tears. Pre-reserving the amount we are going to add will always | |
4479 | * be the right way, because otherwise if we have enough parallelism we | |
4480 | * could end up with thousands of inodes all holding little bits of | |
4481 | * reservations they were able to make previously and the only way to | |
4482 | * reclaim that space is to ENOSPC out the operations and clear | |
4483 | * everything out and try again, which is bad. This way we just | |
4484 | * over-reserve slightly, and clean up the mess when we are done. | |
4485 | */ | |
4486 | calc_inode_reservations(fs_info, num_bytes, &meta_reserve, | |
4487 | &qgroup_reserve); | |
4488 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserve, true); | |
4489 | if (ret) | |
4490 | goto out_fail; | |
0d9764f6 | 4491 | ret = btrfs_reserve_metadata_bytes(root, block_rsv, meta_reserve, flush); |
c8eaeac7 JB |
4492 | if (ret) |
4493 | goto out_qgroup; | |
4494 | ||
4495 | /* | |
4496 | * Now we need to update our outstanding extents and csum bytes _first_ | |
4497 | * and then add the reservation to the block_rsv. This keeps us from | |
4498 | * racing with an ordered completion or some such that would think it | |
4499 | * needs to free the reservation we just made. | |
4500 | */ | |
9f3db423 | 4501 | spin_lock(&inode->lock); |
69fe2d75 | 4502 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 4503 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
4504 | inode->csum_bytes += num_bytes; |
4505 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 4506 | spin_unlock(&inode->lock); |
57a45ced | 4507 | |
c8eaeac7 | 4508 | /* Now we can safely add our space to our block rsv */ |
0b50174a | 4509 | btrfs_block_rsv_add_bytes(block_rsv, meta_reserve, false); |
c8eaeac7 JB |
4510 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
4511 | btrfs_ino(inode), meta_reserve, 1); | |
4512 | ||
4513 | spin_lock(&block_rsv->lock); | |
4514 | block_rsv->qgroup_rsv_reserved += qgroup_reserve; | |
4515 | spin_unlock(&block_rsv->lock); | |
25179201 | 4516 | |
c64c2bd8 | 4517 | if (delalloc_lock) |
9f3db423 | 4518 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 4519 | return 0; |
c8eaeac7 JB |
4520 | out_qgroup: |
4521 | btrfs_qgroup_free_meta_prealloc(root, qgroup_reserve); | |
88e081bf | 4522 | out_fail: |
43b18595 | 4523 | btrfs_inode_rsv_release(inode, true); |
88e081bf | 4524 | if (delalloc_lock) |
9f3db423 | 4525 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 4526 | return ret; |
0ca1f7ce YZ |
4527 | } |
4528 | ||
7709cde3 JB |
4529 | /** |
4530 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
4531 | * @inode: the inode to release the reservation for. |
4532 | * @num_bytes: the number of bytes we are releasing. | |
43b18595 | 4533 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
7709cde3 JB |
4534 | * |
4535 | * This will release the metadata reservation for an inode. This can be called | |
4536 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 4537 | * reservations, or on error for the same reason. |
7709cde3 | 4538 | */ |
43b18595 QW |
4539 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
4540 | bool qgroup_free) | |
0ca1f7ce | 4541 | { |
3ffbd68c | 4542 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
0ca1f7ce | 4543 | |
0b246afa | 4544 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 4545 | spin_lock(&inode->lock); |
69fe2d75 JB |
4546 | inode->csum_bytes -= num_bytes; |
4547 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 4548 | spin_unlock(&inode->lock); |
0ca1f7ce | 4549 | |
0b246afa | 4550 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
4551 | return; |
4552 | ||
43b18595 | 4553 | btrfs_inode_rsv_release(inode, qgroup_free); |
0ca1f7ce YZ |
4554 | } |
4555 | ||
8b62f87b JB |
4556 | /** |
4557 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
4558 | * @inode: the inode to balance the reservation for. | |
4559 | * @num_bytes: the number of bytes we originally reserved with | |
43b18595 | 4560 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
8b62f87b JB |
4561 | * |
4562 | * When we reserve space we increase outstanding_extents for the extents we may | |
4563 | * add. Once we've set the range as delalloc or created our ordered extents we | |
4564 | * have outstanding_extents to track the real usage, so we use this to free our | |
4565 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
4566 | * with btrfs_delalloc_reserve_metadata. | |
4567 | */ | |
43b18595 QW |
4568 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
4569 | bool qgroup_free) | |
8b62f87b | 4570 | { |
3ffbd68c | 4571 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
8b62f87b | 4572 | unsigned num_extents; |
8b62f87b JB |
4573 | |
4574 | spin_lock(&inode->lock); | |
4575 | num_extents = count_max_extents(num_bytes); | |
4576 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 4577 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
4578 | spin_unlock(&inode->lock); |
4579 | ||
8b62f87b JB |
4580 | if (btrfs_is_testing(fs_info)) |
4581 | return; | |
4582 | ||
43b18595 | 4583 | btrfs_inode_rsv_release(inode, qgroup_free); |
8b62f87b JB |
4584 | } |
4585 | ||
1ada3a62 | 4586 | /** |
7cf5b976 | 4587 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
4588 | * delalloc |
4589 | * @inode: inode we're writing to | |
4590 | * @start: start range we are writing to | |
4591 | * @len: how long the range we are writing to | |
364ecf36 QW |
4592 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
4593 | * current reservation. | |
1ada3a62 | 4594 | * |
1ada3a62 QW |
4595 | * This will do the following things |
4596 | * | |
4597 | * o reserve space in data space info for num bytes | |
4598 | * and reserve precious corresponding qgroup space | |
4599 | * (Done in check_data_free_space) | |
4600 | * | |
4601 | * o reserve space for metadata space, based on the number of outstanding | |
4602 | * extents and how much csums will be needed | |
4603 | * also reserve metadata space in a per root over-reserve method. | |
4604 | * o add to the inodes->delalloc_bytes | |
4605 | * o add it to the fs_info's delalloc inodes list. | |
4606 | * (Above 3 all done in delalloc_reserve_metadata) | |
4607 | * | |
4608 | * Return 0 for success | |
4609 | * Return <0 for error(-ENOSPC or -EQUOT) | |
4610 | */ | |
364ecf36 QW |
4611 | int btrfs_delalloc_reserve_space(struct inode *inode, |
4612 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
4613 | { |
4614 | int ret; | |
4615 | ||
364ecf36 | 4616 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
4617 | if (ret < 0) |
4618 | return ret; | |
9f3db423 | 4619 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 4620 | if (ret < 0) |
bc42bda2 | 4621 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
4622 | return ret; |
4623 | } | |
4624 | ||
7709cde3 | 4625 | /** |
7cf5b976 | 4626 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
4627 | * @inode: inode we're releasing space for |
4628 | * @start: start position of the space already reserved | |
4629 | * @len: the len of the space already reserved | |
8b62f87b | 4630 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
4631 | * |
4632 | * This function will release the metadata space that was not used and will | |
4633 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
4634 | * list if there are no delalloc bytes left. | |
4635 | * Also it will handle the qgroup reserved space. | |
4636 | */ | |
bc42bda2 | 4637 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b | 4638 | struct extent_changeset *reserved, |
43b18595 | 4639 | u64 start, u64 len, bool qgroup_free) |
1ada3a62 | 4640 | { |
43b18595 | 4641 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
bc42bda2 | 4642 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
4643 | } |
4644 | ||
ce93ec54 | 4645 | static int update_block_group(struct btrfs_trans_handle *trans, |
6b279408 | 4646 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4647 | { |
6b279408 | 4648 | struct btrfs_fs_info *info = trans->fs_info; |
0af3d00b | 4649 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 4650 | u64 total = num_bytes; |
9078a3e1 | 4651 | u64 old_val; |
db94535d | 4652 | u64 byte_in_group; |
0af3d00b | 4653 | int factor; |
ba2c4d4e | 4654 | int ret = 0; |
3e1ad54f | 4655 | |
5d4f98a2 | 4656 | /* block accounting for super block */ |
eb73c1b7 | 4657 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 4658 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
4659 | if (alloc) |
4660 | old_val += num_bytes; | |
4661 | else | |
4662 | old_val -= num_bytes; | |
6c41761f | 4663 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 4664 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 4665 | |
d397712b | 4666 | while (total) { |
db94535d | 4667 | cache = btrfs_lookup_block_group(info, bytenr); |
ba2c4d4e JB |
4668 | if (!cache) { |
4669 | ret = -ENOENT; | |
4670 | break; | |
4671 | } | |
46df06b8 DS |
4672 | factor = btrfs_bg_type_to_factor(cache->flags); |
4673 | ||
9d66e233 JB |
4674 | /* |
4675 | * If this block group has free space cache written out, we | |
4676 | * need to make sure to load it if we are removing space. This | |
4677 | * is because we need the unpinning stage to actually add the | |
4678 | * space back to the block group, otherwise we will leak space. | |
4679 | */ | |
4680 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 4681 | cache_block_group(cache, 1); |
0af3d00b | 4682 | |
db94535d CM |
4683 | byte_in_group = bytenr - cache->key.objectid; |
4684 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4685 | |
25179201 | 4686 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4687 | spin_lock(&cache->lock); |
0af3d00b | 4688 | |
6202df69 | 4689 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
4690 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
4691 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4692 | ||
9078a3e1 | 4693 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4694 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4695 | if (alloc) { |
db94535d | 4696 | old_val += num_bytes; |
11833d66 YZ |
4697 | btrfs_set_block_group_used(&cache->item, old_val); |
4698 | cache->reserved -= num_bytes; | |
11833d66 | 4699 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
4700 | cache->space_info->bytes_used += num_bytes; |
4701 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4702 | spin_unlock(&cache->lock); |
25179201 | 4703 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4704 | } else { |
db94535d | 4705 | old_val -= num_bytes; |
ae0ab003 FM |
4706 | btrfs_set_block_group_used(&cache->item, old_val); |
4707 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
4708 | btrfs_space_info_update_bytes_pinned(info, |
4709 | cache->space_info, num_bytes); | |
ae0ab003 FM |
4710 | cache->space_info->bytes_used -= num_bytes; |
4711 | cache->space_info->disk_used -= num_bytes * factor; | |
4712 | spin_unlock(&cache->lock); | |
4713 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 4714 | |
0b246afa | 4715 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
4716 | cache->space_info->flags, |
4717 | num_bytes, 1); | |
dec59fa3 EL |
4718 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
4719 | num_bytes, | |
4720 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
ae0ab003 FM |
4721 | set_extent_dirty(info->pinned_extents, |
4722 | bytenr, bytenr + num_bytes - 1, | |
4723 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4724 | } |
1bbc621e CM |
4725 | |
4726 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
4727 | if (list_empty(&cache->dirty_list)) { | |
4728 | list_add_tail(&cache->dirty_list, | |
4729 | &trans->transaction->dirty_bgs); | |
ba2c4d4e | 4730 | trans->delayed_ref_updates++; |
1bbc621e CM |
4731 | btrfs_get_block_group(cache); |
4732 | } | |
4733 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
4734 | ||
036a9348 FM |
4735 | /* |
4736 | * No longer have used bytes in this block group, queue it for | |
4737 | * deletion. We do this after adding the block group to the | |
4738 | * dirty list to avoid races between cleaner kthread and space | |
4739 | * cache writeout. | |
4740 | */ | |
031f24da QW |
4741 | if (!alloc && old_val == 0) |
4742 | btrfs_mark_bg_unused(cache); | |
036a9348 | 4743 | |
fa9c0d79 | 4744 | btrfs_put_block_group(cache); |
db94535d CM |
4745 | total -= num_bytes; |
4746 | bytenr += num_bytes; | |
9078a3e1 | 4747 | } |
ba2c4d4e JB |
4748 | |
4749 | /* Modified block groups are accounted for in the delayed_refs_rsv. */ | |
4750 | btrfs_update_delayed_refs_rsv(trans); | |
4751 | return ret; | |
9078a3e1 | 4752 | } |
6324fbf3 | 4753 | |
2ff7e61e | 4754 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 4755 | { |
0f9dd46c | 4756 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4757 | u64 bytenr; |
0f9dd46c | 4758 | |
0b246afa JM |
4759 | spin_lock(&fs_info->block_group_cache_lock); |
4760 | bytenr = fs_info->first_logical_byte; | |
4761 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
4762 | |
4763 | if (bytenr < (u64)-1) | |
4764 | return bytenr; | |
4765 | ||
0b246afa | 4766 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 4767 | if (!cache) |
a061fc8d | 4768 | return 0; |
0f9dd46c | 4769 | |
d2fb3437 | 4770 | bytenr = cache->key.objectid; |
fa9c0d79 | 4771 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4772 | |
4773 | return bytenr; | |
a061fc8d CM |
4774 | } |
4775 | ||
fdf08605 | 4776 | static int pin_down_extent(struct btrfs_block_group_cache *cache, |
f0486c68 | 4777 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 4778 | { |
fdf08605 DS |
4779 | struct btrfs_fs_info *fs_info = cache->fs_info; |
4780 | ||
11833d66 YZ |
4781 | spin_lock(&cache->space_info->lock); |
4782 | spin_lock(&cache->lock); | |
4783 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
4784 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
4785 | num_bytes); | |
11833d66 YZ |
4786 | if (reserved) { |
4787 | cache->reserved -= num_bytes; | |
4788 | cache->space_info->bytes_reserved -= num_bytes; | |
4789 | } | |
4790 | spin_unlock(&cache->lock); | |
4791 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4792 | |
0b246afa | 4793 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 4794 | cache->space_info->flags, num_bytes, 1); |
dec59fa3 EL |
4795 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
4796 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0b246afa | 4797 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
4798 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
4799 | return 0; | |
4800 | } | |
68b38550 | 4801 | |
f0486c68 YZ |
4802 | /* |
4803 | * this function must be called within transaction | |
4804 | */ | |
2ff7e61e | 4805 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
4806 | u64 bytenr, u64 num_bytes, int reserved) |
4807 | { | |
4808 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4809 | |
0b246afa | 4810 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 4811 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 4812 | |
fdf08605 | 4813 | pin_down_extent(cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
4814 | |
4815 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4816 | return 0; |
4817 | } | |
4818 | ||
f0486c68 | 4819 | /* |
e688b725 CM |
4820 | * this function must be called within transaction |
4821 | */ | |
2ff7e61e | 4822 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
4823 | u64 bytenr, u64 num_bytes) |
4824 | { | |
4825 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 4826 | int ret; |
e688b725 | 4827 | |
0b246afa | 4828 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
4829 | if (!cache) |
4830 | return -EINVAL; | |
e688b725 CM |
4831 | |
4832 | /* | |
4833 | * pull in the free space cache (if any) so that our pin | |
4834 | * removes the free space from the cache. We have load_only set | |
4835 | * to one because the slow code to read in the free extents does check | |
4836 | * the pinned extents. | |
4837 | */ | |
f6373bf3 | 4838 | cache_block_group(cache, 1); |
e688b725 | 4839 | |
fdf08605 | 4840 | pin_down_extent(cache, bytenr, num_bytes, 0); |
e688b725 CM |
4841 | |
4842 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 4843 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 4844 | btrfs_put_block_group(cache); |
b50c6e25 | 4845 | return ret; |
e688b725 CM |
4846 | } |
4847 | ||
2ff7e61e JM |
4848 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
4849 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
4850 | { |
4851 | int ret; | |
4852 | struct btrfs_block_group_cache *block_group; | |
4853 | struct btrfs_caching_control *caching_ctl; | |
4854 | ||
0b246afa | 4855 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
4856 | if (!block_group) |
4857 | return -EINVAL; | |
4858 | ||
4859 | cache_block_group(block_group, 0); | |
4860 | caching_ctl = get_caching_control(block_group); | |
4861 | ||
4862 | if (!caching_ctl) { | |
4863 | /* Logic error */ | |
4864 | BUG_ON(!block_group_cache_done(block_group)); | |
4865 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
4866 | } else { | |
4867 | mutex_lock(&caching_ctl->mutex); | |
4868 | ||
4869 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 4870 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
4871 | } else if (start + num_bytes <= caching_ctl->progress) { |
4872 | ret = btrfs_remove_free_space(block_group, | |
4873 | start, num_bytes); | |
4874 | } else { | |
4875 | num_bytes = caching_ctl->progress - start; | |
4876 | ret = btrfs_remove_free_space(block_group, | |
4877 | start, num_bytes); | |
4878 | if (ret) | |
4879 | goto out_lock; | |
4880 | ||
4881 | num_bytes = (start + num_bytes) - | |
4882 | caching_ctl->progress; | |
4883 | start = caching_ctl->progress; | |
2ff7e61e | 4884 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
4885 | } |
4886 | out_lock: | |
4887 | mutex_unlock(&caching_ctl->mutex); | |
4888 | put_caching_control(caching_ctl); | |
4889 | } | |
4890 | btrfs_put_block_group(block_group); | |
4891 | return ret; | |
4892 | } | |
4893 | ||
bcdc428c | 4894 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 4895 | { |
bcdc428c | 4896 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
4897 | struct btrfs_file_extent_item *item; |
4898 | struct btrfs_key key; | |
4899 | int found_type; | |
4900 | int i; | |
b89311ef | 4901 | int ret = 0; |
8c2a1a30 | 4902 | |
2ff7e61e | 4903 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
4904 | return 0; |
4905 | ||
4906 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
4907 | btrfs_item_key_to_cpu(eb, &key, i); | |
4908 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
4909 | continue; | |
4910 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
4911 | found_type = btrfs_file_extent_type(eb, item); | |
4912 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
4913 | continue; | |
4914 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
4915 | continue; | |
4916 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
4917 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
4918 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
4919 | if (ret) | |
4920 | break; | |
8c2a1a30 JB |
4921 | } |
4922 | ||
b89311ef | 4923 | return ret; |
8c2a1a30 JB |
4924 | } |
4925 | ||
9cfa3e34 FM |
4926 | static void |
4927 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
4928 | { | |
4929 | atomic_inc(&bg->reservations); | |
4930 | } | |
4931 | ||
4932 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
4933 | const u64 start) | |
4934 | { | |
4935 | struct btrfs_block_group_cache *bg; | |
4936 | ||
4937 | bg = btrfs_lookup_block_group(fs_info, start); | |
4938 | ASSERT(bg); | |
4939 | if (atomic_dec_and_test(&bg->reservations)) | |
4625956a | 4940 | wake_up_var(&bg->reservations); |
9cfa3e34 FM |
4941 | btrfs_put_block_group(bg); |
4942 | } | |
4943 | ||
9cfa3e34 FM |
4944 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
4945 | { | |
4946 | struct btrfs_space_info *space_info = bg->space_info; | |
4947 | ||
4948 | ASSERT(bg->ro); | |
4949 | ||
4950 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
4951 | return; | |
4952 | ||
4953 | /* | |
4954 | * Our block group is read only but before we set it to read only, | |
4955 | * some task might have had allocated an extent from it already, but it | |
4956 | * has not yet created a respective ordered extent (and added it to a | |
4957 | * root's list of ordered extents). | |
4958 | * Therefore wait for any task currently allocating extents, since the | |
4959 | * block group's reservations counter is incremented while a read lock | |
4960 | * on the groups' semaphore is held and decremented after releasing | |
4961 | * the read access on that semaphore and creating the ordered extent. | |
4962 | */ | |
4963 | down_write(&space_info->groups_sem); | |
4964 | up_write(&space_info->groups_sem); | |
4965 | ||
4625956a | 4966 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); |
9cfa3e34 FM |
4967 | } |
4968 | ||
fb25e914 | 4969 | /** |
4824f1f4 | 4970 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 4971 | * @cache: The cache we are manipulating |
18513091 WX |
4972 | * @ram_bytes: The number of bytes of file content, and will be same to |
4973 | * @num_bytes except for the compress path. | |
fb25e914 | 4974 | * @num_bytes: The number of bytes in question |
e570fd27 | 4975 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 4976 | * |
745699ef XW |
4977 | * This is called by the allocator when it reserves space. If this is a |
4978 | * reservation and the block group has become read only we cannot make the | |
4979 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 4980 | */ |
4824f1f4 | 4981 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 4982 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 4983 | { |
fb25e914 | 4984 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 4985 | int ret = 0; |
79787eaa | 4986 | |
fb25e914 JB |
4987 | spin_lock(&space_info->lock); |
4988 | spin_lock(&cache->lock); | |
4824f1f4 WX |
4989 | if (cache->ro) { |
4990 | ret = -EAGAIN; | |
fb25e914 | 4991 | } else { |
4824f1f4 WX |
4992 | cache->reserved += num_bytes; |
4993 | space_info->bytes_reserved += num_bytes; | |
bb96c4e5 JB |
4994 | btrfs_space_info_update_bytes_may_use(cache->fs_info, |
4995 | space_info, -ram_bytes); | |
e570fd27 | 4996 | if (delalloc) |
4824f1f4 | 4997 | cache->delalloc_bytes += num_bytes; |
324ae4df | 4998 | } |
fb25e914 JB |
4999 | spin_unlock(&cache->lock); |
5000 | spin_unlock(&space_info->lock); | |
f0486c68 | 5001 | return ret; |
324ae4df | 5002 | } |
9078a3e1 | 5003 | |
4824f1f4 WX |
5004 | /** |
5005 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
5006 | * @cache: The cache we are manipulating | |
5007 | * @num_bytes: The number of bytes in question | |
5008 | * @delalloc: The blocks are allocated for the delalloc write | |
5009 | * | |
5010 | * This is called by somebody who is freeing space that was never actually used | |
5011 | * on disk. For example if you reserve some space for a new leaf in transaction | |
5012 | * A and before transaction A commits you free that leaf, you call this with | |
5013 | * reserve set to 0 in order to clear the reservation. | |
5014 | */ | |
5015 | ||
556f3ca8 | 5016 | static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
5017 | u64 num_bytes, int delalloc) | |
4824f1f4 WX |
5018 | { |
5019 | struct btrfs_space_info *space_info = cache->space_info; | |
4824f1f4 WX |
5020 | |
5021 | spin_lock(&space_info->lock); | |
5022 | spin_lock(&cache->lock); | |
5023 | if (cache->ro) | |
5024 | space_info->bytes_readonly += num_bytes; | |
5025 | cache->reserved -= num_bytes; | |
5026 | space_info->bytes_reserved -= num_bytes; | |
21a94f7a | 5027 | space_info->max_extent_size = 0; |
4824f1f4 WX |
5028 | |
5029 | if (delalloc) | |
5030 | cache->delalloc_bytes -= num_bytes; | |
5031 | spin_unlock(&cache->lock); | |
5032 | spin_unlock(&space_info->lock); | |
4824f1f4 | 5033 | } |
8b74c03e | 5034 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 5035 | { |
11833d66 YZ |
5036 | struct btrfs_caching_control *next; |
5037 | struct btrfs_caching_control *caching_ctl; | |
5038 | struct btrfs_block_group_cache *cache; | |
e8569813 | 5039 | |
9e351cc8 | 5040 | down_write(&fs_info->commit_root_sem); |
25179201 | 5041 | |
11833d66 YZ |
5042 | list_for_each_entry_safe(caching_ctl, next, |
5043 | &fs_info->caching_block_groups, list) { | |
5044 | cache = caching_ctl->block_group; | |
5045 | if (block_group_cache_done(cache)) { | |
5046 | cache->last_byte_to_unpin = (u64)-1; | |
5047 | list_del_init(&caching_ctl->list); | |
5048 | put_caching_control(caching_ctl); | |
e8569813 | 5049 | } else { |
11833d66 | 5050 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 5051 | } |
e8569813 | 5052 | } |
11833d66 YZ |
5053 | |
5054 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
5055 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
5056 | else | |
5057 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
5058 | ||
9e351cc8 | 5059 | up_write(&fs_info->commit_root_sem); |
8929ecfa | 5060 | |
67f9c220 | 5061 | btrfs_update_global_block_rsv(fs_info); |
e8569813 ZY |
5062 | } |
5063 | ||
c759c4e1 JB |
5064 | /* |
5065 | * Returns the free cluster for the given space info and sets empty_cluster to | |
5066 | * what it should be based on the mount options. | |
5067 | */ | |
5068 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
5069 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
5070 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
5071 | { |
5072 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
5073 | |
5074 | *empty_cluster = 0; | |
5075 | if (btrfs_mixed_space_info(space_info)) | |
5076 | return ret; | |
5077 | ||
c759c4e1 | 5078 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 5079 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
5080 | if (btrfs_test_opt(fs_info, SSD)) |
5081 | *empty_cluster = SZ_2M; | |
5082 | else | |
ee22184b | 5083 | *empty_cluster = SZ_64K; |
583b7231 HK |
5084 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
5085 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
5086 | *empty_cluster = SZ_2M; | |
0b246afa | 5087 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
5088 | } |
5089 | ||
5090 | return ret; | |
5091 | } | |
5092 | ||
2ff7e61e JM |
5093 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
5094 | u64 start, u64 end, | |
678886bd | 5095 | const bool return_free_space) |
ccd467d6 | 5096 | { |
11833d66 | 5097 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
5098 | struct btrfs_space_info *space_info; |
5099 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 5100 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 5101 | u64 len; |
c759c4e1 JB |
5102 | u64 total_unpinned = 0; |
5103 | u64 empty_cluster = 0; | |
7b398f8e | 5104 | bool readonly; |
ccd467d6 | 5105 | |
11833d66 | 5106 | while (start <= end) { |
7b398f8e | 5107 | readonly = false; |
11833d66 YZ |
5108 | if (!cache || |
5109 | start >= cache->key.objectid + cache->key.offset) { | |
5110 | if (cache) | |
5111 | btrfs_put_block_group(cache); | |
c759c4e1 | 5112 | total_unpinned = 0; |
11833d66 | 5113 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 5114 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 5115 | |
2ff7e61e | 5116 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
5117 | cache->space_info, |
5118 | &empty_cluster); | |
5119 | empty_cluster <<= 1; | |
11833d66 YZ |
5120 | } |
5121 | ||
5122 | len = cache->key.objectid + cache->key.offset - start; | |
5123 | len = min(len, end + 1 - start); | |
5124 | ||
5125 | if (start < cache->last_byte_to_unpin) { | |
5126 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
5127 | if (return_free_space) |
5128 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
5129 | } |
5130 | ||
f0486c68 | 5131 | start += len; |
c759c4e1 | 5132 | total_unpinned += len; |
7b398f8e | 5133 | space_info = cache->space_info; |
f0486c68 | 5134 | |
c759c4e1 JB |
5135 | /* |
5136 | * If this space cluster has been marked as fragmented and we've | |
5137 | * unpinned enough in this block group to potentially allow a | |
5138 | * cluster to be created inside of it go ahead and clear the | |
5139 | * fragmented check. | |
5140 | */ | |
5141 | if (cluster && cluster->fragmented && | |
5142 | total_unpinned > empty_cluster) { | |
5143 | spin_lock(&cluster->lock); | |
5144 | cluster->fragmented = 0; | |
5145 | spin_unlock(&cluster->lock); | |
5146 | } | |
5147 | ||
7b398f8e | 5148 | spin_lock(&space_info->lock); |
11833d66 YZ |
5149 | spin_lock(&cache->lock); |
5150 | cache->pinned -= len; | |
bb96c4e5 | 5151 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
c51e7bb1 JB |
5152 | |
5153 | trace_btrfs_space_reservation(fs_info, "pinned", | |
5154 | space_info->flags, len, 0); | |
4f4db217 | 5155 | space_info->max_extent_size = 0; |
dec59fa3 EL |
5156 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
5157 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
5158 | if (cache->ro) { |
5159 | space_info->bytes_readonly += len; | |
5160 | readonly = true; | |
5161 | } | |
11833d66 | 5162 | spin_unlock(&cache->lock); |
957780eb JB |
5163 | if (!readonly && return_free_space && |
5164 | global_rsv->space_info == space_info) { | |
5165 | u64 to_add = len; | |
92ac58ec | 5166 | |
7b398f8e JB |
5167 | spin_lock(&global_rsv->lock); |
5168 | if (!global_rsv->full) { | |
957780eb JB |
5169 | to_add = min(len, global_rsv->size - |
5170 | global_rsv->reserved); | |
5171 | global_rsv->reserved += to_add; | |
bb96c4e5 JB |
5172 | btrfs_space_info_update_bytes_may_use(fs_info, |
5173 | space_info, to_add); | |
7b398f8e JB |
5174 | if (global_rsv->reserved >= global_rsv->size) |
5175 | global_rsv->full = 1; | |
957780eb JB |
5176 | trace_btrfs_space_reservation(fs_info, |
5177 | "space_info", | |
5178 | space_info->flags, | |
5179 | to_add, 1); | |
5180 | len -= to_add; | |
7b398f8e JB |
5181 | } |
5182 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
5183 | /* Add to any tickets we may have */ |
5184 | if (len) | |
d44b72aa JB |
5185 | btrfs_space_info_add_new_bytes(fs_info, |
5186 | space_info, len); | |
7b398f8e JB |
5187 | } |
5188 | spin_unlock(&space_info->lock); | |
ccd467d6 | 5189 | } |
11833d66 YZ |
5190 | |
5191 | if (cache) | |
5192 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
5193 | return 0; |
5194 | } | |
5195 | ||
5ead2dd0 | 5196 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 5197 | { |
5ead2dd0 | 5198 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
5199 | struct btrfs_block_group_cache *block_group, *tmp; |
5200 | struct list_head *deleted_bgs; | |
11833d66 | 5201 | struct extent_io_tree *unpin; |
1a5bc167 CM |
5202 | u64 start; |
5203 | u64 end; | |
a28ec197 | 5204 | int ret; |
a28ec197 | 5205 | |
11833d66 YZ |
5206 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
5207 | unpin = &fs_info->freed_extents[1]; | |
5208 | else | |
5209 | unpin = &fs_info->freed_extents[0]; | |
5210 | ||
e33e17ee | 5211 | while (!trans->aborted) { |
0e6ec385 FM |
5212 | struct extent_state *cached_state = NULL; |
5213 | ||
d4b450cd | 5214 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 5215 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 5216 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
5217 | if (ret) { |
5218 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 5219 | break; |
d4b450cd | 5220 | } |
1f3c79a2 | 5221 | |
0b246afa | 5222 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 5223 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 5224 | end + 1 - start, NULL); |
1f3c79a2 | 5225 | |
0e6ec385 | 5226 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 5227 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 5228 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 5229 | free_extent_state(cached_state); |
b9473439 | 5230 | cond_resched(); |
a28ec197 | 5231 | } |
817d52f8 | 5232 | |
e33e17ee JM |
5233 | /* |
5234 | * Transaction is finished. We don't need the lock anymore. We | |
5235 | * do need to clean up the block groups in case of a transaction | |
5236 | * abort. | |
5237 | */ | |
5238 | deleted_bgs = &trans->transaction->deleted_bgs; | |
5239 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
5240 | u64 trimmed = 0; | |
5241 | ||
5242 | ret = -EROFS; | |
5243 | if (!trans->aborted) | |
2ff7e61e | 5244 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
5245 | block_group->key.objectid, |
5246 | block_group->key.offset, | |
5247 | &trimmed); | |
5248 | ||
5249 | list_del_init(&block_group->bg_list); | |
5250 | btrfs_put_block_group_trimming(block_group); | |
5251 | btrfs_put_block_group(block_group); | |
5252 | ||
5253 | if (ret) { | |
5254 | const char *errstr = btrfs_decode_error(ret); | |
5255 | btrfs_warn(fs_info, | |
913e1535 | 5256 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
5257 | ret, errstr); |
5258 | } | |
5259 | } | |
5260 | ||
e20d96d6 CM |
5261 | return 0; |
5262 | } | |
5263 | ||
5d4f98a2 | 5264 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
5265 | struct btrfs_delayed_ref_node *node, u64 parent, |
5266 | u64 root_objectid, u64 owner_objectid, | |
5267 | u64 owner_offset, int refs_to_drop, | |
5268 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 5269 | { |
e72cb923 | 5270 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 5271 | struct btrfs_key key; |
5d4f98a2 | 5272 | struct btrfs_path *path; |
1261ec42 | 5273 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 5274 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
5275 | struct btrfs_extent_item *ei; |
5276 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 5277 | int ret; |
5d4f98a2 | 5278 | int is_data; |
952fccac CM |
5279 | int extent_slot = 0; |
5280 | int found_extent = 0; | |
5281 | int num_to_del = 1; | |
5d4f98a2 YZ |
5282 | u32 item_size; |
5283 | u64 refs; | |
c682f9b3 QW |
5284 | u64 bytenr = node->bytenr; |
5285 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 5286 | int last_ref = 0; |
0b246afa | 5287 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 5288 | |
5caf2a00 | 5289 | path = btrfs_alloc_path(); |
54aa1f4d CM |
5290 | if (!path) |
5291 | return -ENOMEM; | |
5f26f772 | 5292 | |
e4058b54 | 5293 | path->reada = READA_FORWARD; |
b9473439 | 5294 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5295 | |
5296 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
5297 | BUG_ON(!is_data && refs_to_drop != 1); | |
5298 | ||
3173a18f | 5299 | if (is_data) |
897ca819 | 5300 | skinny_metadata = false; |
3173a18f | 5301 | |
fbe4801b NB |
5302 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
5303 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 5304 | owner_offset); |
7bb86316 | 5305 | if (ret == 0) { |
952fccac | 5306 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
5307 | while (extent_slot >= 0) { |
5308 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 5309 | extent_slot); |
5d4f98a2 | 5310 | if (key.objectid != bytenr) |
952fccac | 5311 | break; |
5d4f98a2 YZ |
5312 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
5313 | key.offset == num_bytes) { | |
952fccac CM |
5314 | found_extent = 1; |
5315 | break; | |
5316 | } | |
3173a18f JB |
5317 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
5318 | key.offset == owner_objectid) { | |
5319 | found_extent = 1; | |
5320 | break; | |
5321 | } | |
952fccac CM |
5322 | if (path->slots[0] - extent_slot > 5) |
5323 | break; | |
5d4f98a2 | 5324 | extent_slot--; |
952fccac | 5325 | } |
a79865c6 | 5326 | |
31840ae1 | 5327 | if (!found_extent) { |
5d4f98a2 | 5328 | BUG_ON(iref); |
87cc7a8a | 5329 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 5330 | refs_to_drop, |
fcebe456 | 5331 | is_data, &last_ref); |
005d6427 | 5332 | if (ret) { |
66642832 | 5333 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
5334 | goto out; |
5335 | } | |
b3b4aa74 | 5336 | btrfs_release_path(path); |
b9473439 | 5337 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5338 | |
5339 | key.objectid = bytenr; | |
5340 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5341 | key.offset = num_bytes; | |
5342 | ||
3173a18f JB |
5343 | if (!is_data && skinny_metadata) { |
5344 | key.type = BTRFS_METADATA_ITEM_KEY; | |
5345 | key.offset = owner_objectid; | |
5346 | } | |
5347 | ||
31840ae1 ZY |
5348 | ret = btrfs_search_slot(trans, extent_root, |
5349 | &key, path, -1, 1); | |
3173a18f JB |
5350 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
5351 | /* | |
5352 | * Couldn't find our skinny metadata item, | |
5353 | * see if we have ye olde extent item. | |
5354 | */ | |
5355 | path->slots[0]--; | |
5356 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
5357 | path->slots[0]); | |
5358 | if (key.objectid == bytenr && | |
5359 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
5360 | key.offset == num_bytes) | |
5361 | ret = 0; | |
5362 | } | |
5363 | ||
5364 | if (ret > 0 && skinny_metadata) { | |
5365 | skinny_metadata = false; | |
9ce49a0b | 5366 | key.objectid = bytenr; |
3173a18f JB |
5367 | key.type = BTRFS_EXTENT_ITEM_KEY; |
5368 | key.offset = num_bytes; | |
5369 | btrfs_release_path(path); | |
5370 | ret = btrfs_search_slot(trans, extent_root, | |
5371 | &key, path, -1, 1); | |
5372 | } | |
5373 | ||
f3465ca4 | 5374 | if (ret) { |
5d163e0e JM |
5375 | btrfs_err(info, |
5376 | "umm, got %d back from search, was looking for %llu", | |
5377 | ret, bytenr); | |
b783e62d | 5378 | if (ret > 0) |
a4f78750 | 5379 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 5380 | } |
005d6427 | 5381 | if (ret < 0) { |
66642832 | 5382 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
5383 | goto out; |
5384 | } | |
31840ae1 ZY |
5385 | extent_slot = path->slots[0]; |
5386 | } | |
fae7f21c | 5387 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 5388 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
5389 | btrfs_err(info, |
5390 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
5391 | bytenr, parent, root_objectid, owner_objectid, |
5392 | owner_offset); | |
66642832 | 5393 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 5394 | goto out; |
79787eaa | 5395 | } else { |
66642832 | 5396 | btrfs_abort_transaction(trans, ret); |
005d6427 | 5397 | goto out; |
7bb86316 | 5398 | } |
5f39d397 CM |
5399 | |
5400 | leaf = path->nodes[0]; | |
5d4f98a2 | 5401 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 5402 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
5403 | ret = -EINVAL; |
5404 | btrfs_print_v0_err(info); | |
5405 | btrfs_abort_transaction(trans, ret); | |
5406 | goto out; | |
5407 | } | |
952fccac | 5408 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 5409 | struct btrfs_extent_item); |
3173a18f JB |
5410 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
5411 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
5412 | struct btrfs_tree_block_info *bi; |
5413 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
5414 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
5415 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
5416 | } | |
56bec294 | 5417 | |
5d4f98a2 | 5418 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 5419 | if (refs < refs_to_drop) { |
5d163e0e JM |
5420 | btrfs_err(info, |
5421 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
5422 | refs_to_drop, refs, bytenr); | |
32b02538 | 5423 | ret = -EINVAL; |
66642832 | 5424 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
5425 | goto out; |
5426 | } | |
56bec294 | 5427 | refs -= refs_to_drop; |
5f39d397 | 5428 | |
5d4f98a2 YZ |
5429 | if (refs > 0) { |
5430 | if (extent_op) | |
5431 | __run_delayed_extent_op(extent_op, leaf, ei); | |
5432 | /* | |
5433 | * In the case of inline back ref, reference count will | |
5434 | * be updated by remove_extent_backref | |
952fccac | 5435 | */ |
5d4f98a2 YZ |
5436 | if (iref) { |
5437 | BUG_ON(!found_extent); | |
5438 | } else { | |
5439 | btrfs_set_extent_refs(leaf, ei, refs); | |
5440 | btrfs_mark_buffer_dirty(leaf); | |
5441 | } | |
5442 | if (found_extent) { | |
87cc7a8a NB |
5443 | ret = remove_extent_backref(trans, path, iref, |
5444 | refs_to_drop, is_data, | |
5445 | &last_ref); | |
005d6427 | 5446 | if (ret) { |
66642832 | 5447 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
5448 | goto out; |
5449 | } | |
952fccac | 5450 | } |
5d4f98a2 | 5451 | } else { |
5d4f98a2 YZ |
5452 | if (found_extent) { |
5453 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 5454 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
5455 | if (iref) { |
5456 | BUG_ON(path->slots[0] != extent_slot); | |
5457 | } else { | |
5458 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5459 | path->slots[0] = extent_slot; | |
5460 | num_to_del = 2; | |
5461 | } | |
78fae27e | 5462 | } |
b9473439 | 5463 | |
fcebe456 | 5464 | last_ref = 1; |
952fccac CM |
5465 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5466 | num_to_del); | |
005d6427 | 5467 | if (ret) { |
66642832 | 5468 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
5469 | goto out; |
5470 | } | |
b3b4aa74 | 5471 | btrfs_release_path(path); |
21af804c | 5472 | |
5d4f98a2 | 5473 | if (is_data) { |
5b4aacef | 5474 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 5475 | if (ret) { |
66642832 | 5476 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
5477 | goto out; |
5478 | } | |
459931ec CM |
5479 | } |
5480 | ||
e7355e50 | 5481 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 5482 | if (ret) { |
66642832 | 5483 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
5484 | goto out; |
5485 | } | |
5486 | ||
6b279408 | 5487 | ret = update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 5488 | if (ret) { |
66642832 | 5489 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
5490 | goto out; |
5491 | } | |
a28ec197 | 5492 | } |
fcebe456 JB |
5493 | btrfs_release_path(path); |
5494 | ||
79787eaa | 5495 | out: |
5caf2a00 | 5496 | btrfs_free_path(path); |
a28ec197 CM |
5497 | return ret; |
5498 | } | |
5499 | ||
1887be66 | 5500 | /* |
f0486c68 | 5501 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5502 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5503 | * a given extent, and if there are no other delayed refs to be processed, it | |
5504 | * removes it from the tree. | |
5505 | */ | |
5506 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 5507 | u64 bytenr) |
1887be66 CM |
5508 | { |
5509 | struct btrfs_delayed_ref_head *head; | |
5510 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 5511 | int ret = 0; |
1887be66 CM |
5512 | |
5513 | delayed_refs = &trans->transaction->delayed_refs; | |
5514 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 5515 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 5516 | if (!head) |
cf93da7b | 5517 | goto out_delayed_unlock; |
1887be66 | 5518 | |
d7df2c79 | 5519 | spin_lock(&head->lock); |
e3d03965 | 5520 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
5521 | goto out; |
5522 | ||
bedc6617 JB |
5523 | if (cleanup_extent_op(head) != NULL) |
5524 | goto out; | |
5d4f98a2 | 5525 | |
1887be66 CM |
5526 | /* |
5527 | * waiting for the lock here would deadlock. If someone else has it | |
5528 | * locked they are already in the process of dropping it anyway | |
5529 | */ | |
5530 | if (!mutex_trylock(&head->mutex)) | |
5531 | goto out; | |
5532 | ||
d7baffda | 5533 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 5534 | head->processing = 0; |
d7baffda | 5535 | |
d7df2c79 | 5536 | spin_unlock(&head->lock); |
1887be66 CM |
5537 | spin_unlock(&delayed_refs->lock); |
5538 | ||
f0486c68 YZ |
5539 | BUG_ON(head->extent_op); |
5540 | if (head->must_insert_reserved) | |
5541 | ret = 1; | |
5542 | ||
31890da0 | 5543 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 5544 | mutex_unlock(&head->mutex); |
d278850e | 5545 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 5546 | return ret; |
1887be66 | 5547 | out: |
d7df2c79 | 5548 | spin_unlock(&head->lock); |
cf93da7b CM |
5549 | |
5550 | out_delayed_unlock: | |
1887be66 CM |
5551 | spin_unlock(&delayed_refs->lock); |
5552 | return 0; | |
5553 | } | |
5554 | ||
f0486c68 YZ |
5555 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5556 | struct btrfs_root *root, | |
5557 | struct extent_buffer *buf, | |
5581a51a | 5558 | u64 parent, int last_ref) |
f0486c68 | 5559 | { |
0b246afa | 5560 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 5561 | struct btrfs_ref generic_ref = { 0 }; |
b150a4f1 | 5562 | int pin = 1; |
f0486c68 YZ |
5563 | int ret; |
5564 | ||
ed4f255b QW |
5565 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
5566 | buf->start, buf->len, parent); | |
5567 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
5568 | root->root_key.objectid); | |
5569 | ||
f0486c68 | 5570 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
d7eae340 OS |
5571 | int old_ref_mod, new_ref_mod; |
5572 | ||
8a5040f7 | 5573 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 5574 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL, |
d7eae340 | 5575 | &old_ref_mod, &new_ref_mod); |
79787eaa | 5576 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 5577 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
5578 | } |
5579 | ||
0a16c7d7 | 5580 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
5581 | struct btrfs_block_group_cache *cache; |
5582 | ||
f0486c68 | 5583 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 5584 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 5585 | if (!ret) |
37be25bc | 5586 | goto out; |
f0486c68 YZ |
5587 | } |
5588 | ||
4da8b76d | 5589 | pin = 0; |
0b246afa | 5590 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 5591 | |
f0486c68 | 5592 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
fdf08605 | 5593 | pin_down_extent(cache, buf->start, buf->len, 1); |
6219872d | 5594 | btrfs_put_block_group(cache); |
37be25bc | 5595 | goto out; |
f0486c68 YZ |
5596 | } |
5597 | ||
5598 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5599 | ||
5600 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 5601 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 5602 | btrfs_put_block_group(cache); |
71ff6437 | 5603 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
5604 | } |
5605 | out: | |
b150a4f1 | 5606 | if (pin) |
78192442 | 5607 | add_pinned_bytes(fs_info, &generic_ref); |
b150a4f1 | 5608 | |
0a16c7d7 OS |
5609 | if (last_ref) { |
5610 | /* | |
5611 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
5612 | * matter anymore. | |
5613 | */ | |
5614 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
5615 | } | |
f0486c68 YZ |
5616 | } |
5617 | ||
79787eaa | 5618 | /* Can return -ENOMEM */ |
ffd4bb2a | 5619 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 5620 | { |
ffd4bb2a | 5621 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 5622 | int old_ref_mod, new_ref_mod; |
925baedd CM |
5623 | int ret; |
5624 | ||
f5ee5c9a | 5625 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 5626 | return 0; |
fccb84c9 | 5627 | |
56bec294 CM |
5628 | /* |
5629 | * tree log blocks never actually go into the extent allocation | |
5630 | * tree, just update pinning info and exit early. | |
56bec294 | 5631 | */ |
ffd4bb2a QW |
5632 | if ((ref->type == BTRFS_REF_METADATA && |
5633 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
5634 | (ref->type == BTRFS_REF_DATA && | |
5635 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 5636 | /* unlocks the pinned mutex */ |
ffd4bb2a | 5637 | btrfs_pin_extent(fs_info, ref->bytenr, ref->len, 1); |
d7eae340 | 5638 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 5639 | ret = 0; |
ffd4bb2a QW |
5640 | } else if (ref->type == BTRFS_REF_METADATA) { |
5641 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL, | |
d7eae340 | 5642 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 5643 | } else { |
ffd4bb2a | 5644 | ret = btrfs_add_delayed_data_ref(trans, ref, 0, |
d7eae340 | 5645 | &old_ref_mod, &new_ref_mod); |
56bec294 | 5646 | } |
d7eae340 | 5647 | |
ffd4bb2a QW |
5648 | if (!((ref->type == BTRFS_REF_METADATA && |
5649 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
5650 | (ref->type == BTRFS_REF_DATA && | |
5651 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
5652 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 5653 | |
ddf30cf0 | 5654 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) |
78192442 | 5655 | add_pinned_bytes(fs_info, ref); |
d7eae340 | 5656 | |
925baedd CM |
5657 | return ret; |
5658 | } | |
5659 | ||
817d52f8 JB |
5660 | /* |
5661 | * when we wait for progress in the block group caching, its because | |
5662 | * our allocation attempt failed at least once. So, we must sleep | |
5663 | * and let some progress happen before we try again. | |
5664 | * | |
5665 | * This function will sleep at least once waiting for new free space to | |
5666 | * show up, and then it will check the block group free space numbers | |
5667 | * for our min num_bytes. Another option is to have it go ahead | |
5668 | * and look in the rbtree for a free extent of a given size, but this | |
5669 | * is a good start. | |
36cce922 JB |
5670 | * |
5671 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
5672 | * any of the information in this block group. | |
817d52f8 | 5673 | */ |
36cce922 | 5674 | static noinline void |
817d52f8 JB |
5675 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
5676 | u64 num_bytes) | |
5677 | { | |
11833d66 | 5678 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 5679 | |
11833d66 YZ |
5680 | caching_ctl = get_caching_control(cache); |
5681 | if (!caching_ctl) | |
36cce922 | 5682 | return; |
817d52f8 | 5683 | |
11833d66 | 5684 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5685 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5686 | |
5687 | put_caching_control(caching_ctl); | |
11833d66 YZ |
5688 | } |
5689 | ||
5690 | static noinline int | |
5691 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5692 | { | |
5693 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 5694 | int ret = 0; |
11833d66 YZ |
5695 | |
5696 | caching_ctl = get_caching_control(cache); | |
5697 | if (!caching_ctl) | |
36cce922 | 5698 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
5699 | |
5700 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
5701 | if (cache->cached == BTRFS_CACHE_ERROR) |
5702 | ret = -EIO; | |
11833d66 | 5703 | put_caching_control(caching_ctl); |
36cce922 | 5704 | return ret; |
817d52f8 JB |
5705 | } |
5706 | ||
5707 | enum btrfs_loop_type { | |
f262fa8d DS |
5708 | LOOP_CACHING_NOWAIT, |
5709 | LOOP_CACHING_WAIT, | |
5710 | LOOP_ALLOC_CHUNK, | |
5711 | LOOP_NO_EMPTY_SIZE, | |
817d52f8 JB |
5712 | }; |
5713 | ||
e570fd27 MX |
5714 | static inline void |
5715 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
5716 | int delalloc) | |
5717 | { | |
5718 | if (delalloc) | |
5719 | down_read(&cache->data_rwsem); | |
5720 | } | |
5721 | ||
5722 | static inline void | |
5723 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
5724 | int delalloc) | |
5725 | { | |
5726 | btrfs_get_block_group(cache); | |
5727 | if (delalloc) | |
5728 | down_read(&cache->data_rwsem); | |
5729 | } | |
5730 | ||
5731 | static struct btrfs_block_group_cache * | |
5732 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
5733 | struct btrfs_free_cluster *cluster, | |
5734 | int delalloc) | |
5735 | { | |
89771cc9 | 5736 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 5737 | |
e570fd27 | 5738 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
5739 | while (1) { |
5740 | used_bg = cluster->block_group; | |
5741 | if (!used_bg) | |
5742 | return NULL; | |
5743 | ||
5744 | if (used_bg == block_group) | |
e570fd27 MX |
5745 | return used_bg; |
5746 | ||
6719afdc | 5747 | btrfs_get_block_group(used_bg); |
e570fd27 | 5748 | |
6719afdc GU |
5749 | if (!delalloc) |
5750 | return used_bg; | |
e570fd27 | 5751 | |
6719afdc GU |
5752 | if (down_read_trylock(&used_bg->data_rwsem)) |
5753 | return used_bg; | |
e570fd27 | 5754 | |
6719afdc | 5755 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 5756 | |
e321f8a8 LB |
5757 | /* We should only have one-level nested. */ |
5758 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 5759 | |
6719afdc GU |
5760 | spin_lock(&cluster->refill_lock); |
5761 | if (used_bg == cluster->block_group) | |
5762 | return used_bg; | |
e570fd27 | 5763 | |
6719afdc GU |
5764 | up_read(&used_bg->data_rwsem); |
5765 | btrfs_put_block_group(used_bg); | |
5766 | } | |
e570fd27 MX |
5767 | } |
5768 | ||
5769 | static inline void | |
5770 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
5771 | int delalloc) | |
5772 | { | |
5773 | if (delalloc) | |
5774 | up_read(&cache->data_rwsem); | |
5775 | btrfs_put_block_group(cache); | |
5776 | } | |
5777 | ||
b4bd745d QW |
5778 | /* |
5779 | * Structure used internally for find_free_extent() function. Wraps needed | |
5780 | * parameters. | |
5781 | */ | |
5782 | struct find_free_extent_ctl { | |
5783 | /* Basic allocation info */ | |
5784 | u64 ram_bytes; | |
5785 | u64 num_bytes; | |
5786 | u64 empty_size; | |
5787 | u64 flags; | |
5788 | int delalloc; | |
5789 | ||
5790 | /* Where to start the search inside the bg */ | |
5791 | u64 search_start; | |
5792 | ||
5793 | /* For clustered allocation */ | |
5794 | u64 empty_cluster; | |
5795 | ||
5796 | bool have_caching_bg; | |
5797 | bool orig_have_caching_bg; | |
5798 | ||
5799 | /* RAID index, converted from flags */ | |
5800 | int index; | |
5801 | ||
e72d79d6 QW |
5802 | /* |
5803 | * Current loop number, check find_free_extent_update_loop() for details | |
5804 | */ | |
b4bd745d QW |
5805 | int loop; |
5806 | ||
5807 | /* | |
5808 | * Whether we're refilling a cluster, if true we need to re-search | |
5809 | * current block group but don't try to refill the cluster again. | |
5810 | */ | |
5811 | bool retry_clustered; | |
5812 | ||
5813 | /* | |
5814 | * Whether we're updating free space cache, if true we need to re-search | |
5815 | * current block group but don't try updating free space cache again. | |
5816 | */ | |
5817 | bool retry_unclustered; | |
5818 | ||
5819 | /* If current block group is cached */ | |
5820 | int cached; | |
5821 | ||
5822 | /* Max contiguous hole found */ | |
5823 | u64 max_extent_size; | |
5824 | ||
5825 | /* Total free space from free space cache, not always contiguous */ | |
5826 | u64 total_free_space; | |
5827 | ||
5828 | /* Found result */ | |
5829 | u64 found_offset; | |
5830 | }; | |
5831 | ||
d06e3bb6 QW |
5832 | |
5833 | /* | |
5834 | * Helper function for find_free_extent(). | |
5835 | * | |
5836 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
5837 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
5838 | * Return >0 to inform caller that we find nothing | |
5839 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
5840 | */ | |
5841 | static int find_free_extent_clustered(struct btrfs_block_group_cache *bg, | |
5842 | struct btrfs_free_cluster *last_ptr, | |
5843 | struct find_free_extent_ctl *ffe_ctl, | |
5844 | struct btrfs_block_group_cache **cluster_bg_ret) | |
5845 | { | |
d06e3bb6 QW |
5846 | struct btrfs_block_group_cache *cluster_bg; |
5847 | u64 aligned_cluster; | |
5848 | u64 offset; | |
5849 | int ret; | |
5850 | ||
5851 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
5852 | if (!cluster_bg) | |
5853 | goto refill_cluster; | |
5854 | if (cluster_bg != bg && (cluster_bg->ro || | |
5855 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
5856 | goto release_cluster; | |
5857 | ||
5858 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
5859 | ffe_ctl->num_bytes, cluster_bg->key.objectid, | |
5860 | &ffe_ctl->max_extent_size); | |
5861 | if (offset) { | |
5862 | /* We have a block, we're done */ | |
5863 | spin_unlock(&last_ptr->refill_lock); | |
5864 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
5865 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
5866 | *cluster_bg_ret = cluster_bg; | |
5867 | ffe_ctl->found_offset = offset; | |
5868 | return 0; | |
5869 | } | |
5870 | WARN_ON(last_ptr->block_group != cluster_bg); | |
5871 | ||
5872 | release_cluster: | |
5873 | /* | |
5874 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
5875 | * lets just skip it and let the allocator find whatever block it can | |
5876 | * find. If we reach this point, we will have tried the cluster | |
5877 | * allocator plenty of times and not have found anything, so we are | |
5878 | * likely way too fragmented for the clustering stuff to find anything. | |
5879 | * | |
5880 | * However, if the cluster is taken from the current block group, | |
5881 | * release the cluster first, so that we stand a better chance of | |
5882 | * succeeding in the unclustered allocation. | |
5883 | */ | |
5884 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
5885 | spin_unlock(&last_ptr->refill_lock); | |
5886 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
5887 | return -ENOENT; | |
5888 | } | |
5889 | ||
5890 | /* This cluster didn't work out, free it and start over */ | |
5891 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5892 | ||
5893 | if (cluster_bg != bg) | |
5894 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
5895 | ||
5896 | refill_cluster: | |
5897 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
5898 | spin_unlock(&last_ptr->refill_lock); | |
5899 | return -ENOENT; | |
5900 | } | |
5901 | ||
5902 | aligned_cluster = max_t(u64, | |
5903 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
5904 | bg->full_stripe_len); | |
2ceeae2e DS |
5905 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
5906 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
5907 | if (ret == 0) { |
5908 | /* Now pull our allocation out of this cluster */ | |
5909 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
5910 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
5911 | &ffe_ctl->max_extent_size); | |
5912 | if (offset) { | |
5913 | /* We found one, proceed */ | |
5914 | spin_unlock(&last_ptr->refill_lock); | |
5915 | trace_btrfs_reserve_extent_cluster(bg, | |
5916 | ffe_ctl->search_start, | |
5917 | ffe_ctl->num_bytes); | |
5918 | ffe_ctl->found_offset = offset; | |
5919 | return 0; | |
5920 | } | |
5921 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
5922 | !ffe_ctl->retry_clustered) { | |
5923 | spin_unlock(&last_ptr->refill_lock); | |
5924 | ||
5925 | ffe_ctl->retry_clustered = true; | |
5926 | wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + | |
5927 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); | |
5928 | return -EAGAIN; | |
5929 | } | |
5930 | /* | |
5931 | * At this point we either didn't find a cluster or we weren't able to | |
5932 | * allocate a block from our cluster. Free the cluster we've been | |
5933 | * trying to use, and go to the next block group. | |
5934 | */ | |
5935 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5936 | spin_unlock(&last_ptr->refill_lock); | |
5937 | return 1; | |
5938 | } | |
5939 | ||
e1a41848 QW |
5940 | /* |
5941 | * Return >0 to inform caller that we find nothing | |
5942 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
5943 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
5944 | */ | |
5945 | static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg, | |
5946 | struct btrfs_free_cluster *last_ptr, | |
5947 | struct find_free_extent_ctl *ffe_ctl) | |
5948 | { | |
5949 | u64 offset; | |
5950 | ||
5951 | /* | |
5952 | * We are doing an unclustered allocation, set the fragmented flag so | |
5953 | * we don't bother trying to setup a cluster again until we get more | |
5954 | * space. | |
5955 | */ | |
5956 | if (unlikely(last_ptr)) { | |
5957 | spin_lock(&last_ptr->lock); | |
5958 | last_ptr->fragmented = 1; | |
5959 | spin_unlock(&last_ptr->lock); | |
5960 | } | |
5961 | if (ffe_ctl->cached) { | |
5962 | struct btrfs_free_space_ctl *free_space_ctl; | |
5963 | ||
5964 | free_space_ctl = bg->free_space_ctl; | |
5965 | spin_lock(&free_space_ctl->tree_lock); | |
5966 | if (free_space_ctl->free_space < | |
5967 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
5968 | ffe_ctl->empty_size) { | |
5969 | ffe_ctl->total_free_space = max_t(u64, | |
5970 | ffe_ctl->total_free_space, | |
5971 | free_space_ctl->free_space); | |
5972 | spin_unlock(&free_space_ctl->tree_lock); | |
5973 | return 1; | |
5974 | } | |
5975 | spin_unlock(&free_space_ctl->tree_lock); | |
5976 | } | |
5977 | ||
5978 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
5979 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
5980 | &ffe_ctl->max_extent_size); | |
5981 | ||
5982 | /* | |
5983 | * If we didn't find a chunk, and we haven't failed on this block group | |
5984 | * before, and this block group is in the middle of caching and we are | |
5985 | * ok with waiting, then go ahead and wait for progress to be made, and | |
5986 | * set @retry_unclustered to true. | |
5987 | * | |
5988 | * If @retry_unclustered is true then we've already waited on this | |
5989 | * block group once and should move on to the next block group. | |
5990 | */ | |
5991 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
5992 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
5993 | wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + | |
5994 | ffe_ctl->empty_size); | |
5995 | ffe_ctl->retry_unclustered = true; | |
5996 | return -EAGAIN; | |
5997 | } else if (!offset) { | |
5998 | return 1; | |
5999 | } | |
6000 | ffe_ctl->found_offset = offset; | |
6001 | return 0; | |
6002 | } | |
6003 | ||
e72d79d6 QW |
6004 | /* |
6005 | * Return >0 means caller needs to re-search for free extent | |
6006 | * Return 0 means we have the needed free extent. | |
6007 | * Return <0 means we failed to locate any free extent. | |
6008 | */ | |
6009 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
6010 | struct btrfs_free_cluster *last_ptr, | |
6011 | struct btrfs_key *ins, | |
6012 | struct find_free_extent_ctl *ffe_ctl, | |
6013 | int full_search, bool use_cluster) | |
6014 | { | |
6015 | struct btrfs_root *root = fs_info->extent_root; | |
6016 | int ret; | |
6017 | ||
6018 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
6019 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
6020 | ffe_ctl->orig_have_caching_bg = true; | |
6021 | ||
6022 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
6023 | ffe_ctl->have_caching_bg) | |
6024 | return 1; | |
6025 | ||
6026 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
6027 | return 1; | |
6028 | ||
6029 | if (ins->objectid) { | |
6030 | if (!use_cluster && last_ptr) { | |
6031 | spin_lock(&last_ptr->lock); | |
6032 | last_ptr->window_start = ins->objectid; | |
6033 | spin_unlock(&last_ptr->lock); | |
6034 | } | |
6035 | return 0; | |
6036 | } | |
6037 | ||
6038 | /* | |
6039 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
6040 | * caching kthreads as we move along | |
6041 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
6042 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
6043 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
6044 | * again | |
6045 | */ | |
6046 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
6047 | ffe_ctl->index = 0; | |
6048 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
6049 | /* | |
6050 | * We want to skip the LOOP_CACHING_WAIT step if we | |
6051 | * don't have any uncached bgs and we've already done a | |
6052 | * full search through. | |
6053 | */ | |
6054 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
6055 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
6056 | else | |
6057 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
6058 | } else { | |
6059 | ffe_ctl->loop++; | |
6060 | } | |
6061 | ||
6062 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
6063 | struct btrfs_trans_handle *trans; | |
6064 | int exist = 0; | |
6065 | ||
6066 | trans = current->journal_info; | |
6067 | if (trans) | |
6068 | exist = 1; | |
6069 | else | |
6070 | trans = btrfs_join_transaction(root); | |
6071 | ||
6072 | if (IS_ERR(trans)) { | |
6073 | ret = PTR_ERR(trans); | |
6074 | return ret; | |
6075 | } | |
6076 | ||
fc471cb0 JB |
6077 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
6078 | CHUNK_ALLOC_FORCE); | |
e72d79d6 QW |
6079 | |
6080 | /* | |
6081 | * If we can't allocate a new chunk we've already looped | |
6082 | * through at least once, move on to the NO_EMPTY_SIZE | |
6083 | * case. | |
6084 | */ | |
6085 | if (ret == -ENOSPC) | |
6086 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
6087 | ||
6088 | /* Do not bail out on ENOSPC since we can do more. */ | |
6089 | if (ret < 0 && ret != -ENOSPC) | |
6090 | btrfs_abort_transaction(trans, ret); | |
6091 | else | |
6092 | ret = 0; | |
6093 | if (!exist) | |
6094 | btrfs_end_transaction(trans); | |
6095 | if (ret) | |
6096 | return ret; | |
6097 | } | |
6098 | ||
6099 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
6100 | /* | |
6101 | * Don't loop again if we already have no empty_size and | |
6102 | * no empty_cluster. | |
6103 | */ | |
6104 | if (ffe_ctl->empty_size == 0 && | |
6105 | ffe_ctl->empty_cluster == 0) | |
6106 | return -ENOSPC; | |
6107 | ffe_ctl->empty_size = 0; | |
6108 | ffe_ctl->empty_cluster = 0; | |
6109 | } | |
6110 | return 1; | |
6111 | } | |
6112 | return -ENOSPC; | |
6113 | } | |
6114 | ||
fec577fb CM |
6115 | /* |
6116 | * walks the btree of allocated extents and find a hole of a given size. | |
6117 | * The key ins is changed to record the hole: | |
a4820398 | 6118 | * ins->objectid == start position |
62e2749e | 6119 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 6120 | * ins->offset == the size of the hole. |
fec577fb | 6121 | * Any available blocks before search_start are skipped. |
a4820398 MX |
6122 | * |
6123 | * If there is no suitable free space, we will record the max size of | |
6124 | * the free space extent currently. | |
e72d79d6 QW |
6125 | * |
6126 | * The overall logic and call chain: | |
6127 | * | |
6128 | * find_free_extent() | |
6129 | * |- Iterate through all block groups | |
6130 | * | |- Get a valid block group | |
6131 | * | |- Try to do clustered allocation in that block group | |
6132 | * | |- Try to do unclustered allocation in that block group | |
6133 | * | |- Check if the result is valid | |
6134 | * | | |- If valid, then exit | |
6135 | * | |- Jump to next block group | |
6136 | * | | |
6137 | * |- Push harder to find free extents | |
6138 | * |- If not found, re-iterate all block groups | |
fec577fb | 6139 | */ |
87bde3cd | 6140 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
6141 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
6142 | u64 hint_byte, struct btrfs_key *ins, | |
6143 | u64 flags, int delalloc) | |
fec577fb | 6144 | { |
80eb234a | 6145 | int ret = 0; |
fa9c0d79 | 6146 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 6147 | struct btrfs_block_group_cache *block_group = NULL; |
b4bd745d | 6148 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 6149 | struct btrfs_space_info *space_info; |
67377734 | 6150 | bool use_cluster = true; |
a5e681d9 | 6151 | bool full_search = false; |
fec577fb | 6152 | |
0b246afa | 6153 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d QW |
6154 | |
6155 | ffe_ctl.ram_bytes = ram_bytes; | |
6156 | ffe_ctl.num_bytes = num_bytes; | |
6157 | ffe_ctl.empty_size = empty_size; | |
6158 | ffe_ctl.flags = flags; | |
6159 | ffe_ctl.search_start = 0; | |
6160 | ffe_ctl.retry_clustered = false; | |
6161 | ffe_ctl.retry_unclustered = false; | |
6162 | ffe_ctl.delalloc = delalloc; | |
6163 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
6164 | ffe_ctl.have_caching_bg = false; | |
6165 | ffe_ctl.orig_have_caching_bg = false; | |
6166 | ffe_ctl.found_offset = 0; | |
6167 | ||
962a298f | 6168 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
6169 | ins->objectid = 0; |
6170 | ins->offset = 0; | |
b1a4d965 | 6171 | |
71ff6437 | 6172 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 6173 | |
280c2908 | 6174 | space_info = btrfs_find_space_info(fs_info, flags); |
1b1d1f66 | 6175 | if (!space_info) { |
0b246afa | 6176 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
6177 | return -ENOSPC; |
6178 | } | |
2552d17e | 6179 | |
67377734 | 6180 | /* |
4f4db217 JB |
6181 | * If our free space is heavily fragmented we may not be able to make |
6182 | * big contiguous allocations, so instead of doing the expensive search | |
6183 | * for free space, simply return ENOSPC with our max_extent_size so we | |
6184 | * can go ahead and search for a more manageable chunk. | |
6185 | * | |
6186 | * If our max_extent_size is large enough for our allocation simply | |
6187 | * disable clustering since we will likely not be able to find enough | |
6188 | * space to create a cluster and induce latency trying. | |
67377734 | 6189 | */ |
4f4db217 JB |
6190 | if (unlikely(space_info->max_extent_size)) { |
6191 | spin_lock(&space_info->lock); | |
6192 | if (space_info->max_extent_size && | |
6193 | num_bytes > space_info->max_extent_size) { | |
6194 | ins->offset = space_info->max_extent_size; | |
6195 | spin_unlock(&space_info->lock); | |
6196 | return -ENOSPC; | |
6197 | } else if (space_info->max_extent_size) { | |
6198 | use_cluster = false; | |
6199 | } | |
6200 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 6201 | } |
0f9dd46c | 6202 | |
b4bd745d QW |
6203 | last_ptr = fetch_cluster_info(fs_info, space_info, |
6204 | &ffe_ctl.empty_cluster); | |
239b14b3 | 6205 | if (last_ptr) { |
fa9c0d79 CM |
6206 | spin_lock(&last_ptr->lock); |
6207 | if (last_ptr->block_group) | |
6208 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
6209 | if (last_ptr->fragmented) { |
6210 | /* | |
6211 | * We still set window_start so we can keep track of the | |
6212 | * last place we found an allocation to try and save | |
6213 | * some time. | |
6214 | */ | |
6215 | hint_byte = last_ptr->window_start; | |
6216 | use_cluster = false; | |
6217 | } | |
fa9c0d79 | 6218 | spin_unlock(&last_ptr->lock); |
239b14b3 | 6219 | } |
fa9c0d79 | 6220 | |
b4bd745d QW |
6221 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
6222 | first_logical_byte(fs_info, 0)); | |
6223 | ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte); | |
6224 | if (ffe_ctl.search_start == hint_byte) { | |
6225 | block_group = btrfs_lookup_block_group(fs_info, | |
6226 | ffe_ctl.search_start); | |
817d52f8 JB |
6227 | /* |
6228 | * we don't want to use the block group if it doesn't match our | |
6229 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
6230 | * |
6231 | * However if we are re-searching with an ideal block group | |
6232 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 6233 | */ |
b6919a58 | 6234 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 6235 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 6236 | down_read(&space_info->groups_sem); |
44fb5511 CM |
6237 | if (list_empty(&block_group->list) || |
6238 | block_group->ro) { | |
6239 | /* | |
6240 | * someone is removing this block group, | |
6241 | * we can't jump into the have_block_group | |
6242 | * target because our list pointers are not | |
6243 | * valid | |
6244 | */ | |
6245 | btrfs_put_block_group(block_group); | |
6246 | up_read(&space_info->groups_sem); | |
ccf0e725 | 6247 | } else { |
b4bd745d | 6248 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 6249 | block_group->flags); |
e570fd27 | 6250 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 6251 | goto have_block_group; |
ccf0e725 | 6252 | } |
2552d17e | 6253 | } else if (block_group) { |
fa9c0d79 | 6254 | btrfs_put_block_group(block_group); |
2552d17e | 6255 | } |
42e70e7a | 6256 | } |
2552d17e | 6257 | search: |
b4bd745d QW |
6258 | ffe_ctl.have_caching_bg = false; |
6259 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
6260 | ffe_ctl.index == 0) | |
a5e681d9 | 6261 | full_search = true; |
80eb234a | 6262 | down_read(&space_info->groups_sem); |
b4bd745d QW |
6263 | list_for_each_entry(block_group, |
6264 | &space_info->block_groups[ffe_ctl.index], list) { | |
14443937 JM |
6265 | /* If the block group is read-only, we can skip it entirely. */ |
6266 | if (unlikely(block_group->ro)) | |
6267 | continue; | |
6268 | ||
e570fd27 | 6269 | btrfs_grab_block_group(block_group, delalloc); |
b4bd745d | 6270 | ffe_ctl.search_start = block_group->key.objectid; |
42e70e7a | 6271 | |
83a50de9 CM |
6272 | /* |
6273 | * this can happen if we end up cycling through all the | |
6274 | * raid types, but we want to make sure we only allocate | |
6275 | * for the proper type. | |
6276 | */ | |
b6919a58 | 6277 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 6278 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 6279 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 6280 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
6281 | BTRFS_BLOCK_GROUP_RAID10; |
6282 | ||
6283 | /* | |
6284 | * if they asked for extra copies and this block group | |
6285 | * doesn't provide them, bail. This does allow us to | |
6286 | * fill raid0 from raid1. | |
6287 | */ | |
b6919a58 | 6288 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
6289 | goto loop; |
6290 | } | |
6291 | ||
2552d17e | 6292 | have_block_group: |
b4bd745d QW |
6293 | ffe_ctl.cached = block_group_cache_done(block_group); |
6294 | if (unlikely(!ffe_ctl.cached)) { | |
6295 | ffe_ctl.have_caching_bg = true; | |
f6373bf3 | 6296 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
6297 | BUG_ON(ret < 0); |
6298 | ret = 0; | |
817d52f8 JB |
6299 | } |
6300 | ||
36cce922 JB |
6301 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
6302 | goto loop; | |
0f9dd46c | 6303 | |
0a24325e | 6304 | /* |
062c05c4 AO |
6305 | * Ok we want to try and use the cluster allocator, so |
6306 | * lets look there | |
0a24325e | 6307 | */ |
c759c4e1 | 6308 | if (last_ptr && use_cluster) { |
d06e3bb6 | 6309 | struct btrfs_block_group_cache *cluster_bg = NULL; |
fa9c0d79 | 6310 | |
d06e3bb6 QW |
6311 | ret = find_free_extent_clustered(block_group, last_ptr, |
6312 | &ffe_ctl, &cluster_bg); | |
062c05c4 | 6313 | |
fa9c0d79 | 6314 | if (ret == 0) { |
d06e3bb6 QW |
6315 | if (cluster_bg && cluster_bg != block_group) { |
6316 | btrfs_release_block_group(block_group, | |
6317 | delalloc); | |
6318 | block_group = cluster_bg; | |
fa9c0d79 | 6319 | } |
d06e3bb6 QW |
6320 | goto checks; |
6321 | } else if (ret == -EAGAIN) { | |
817d52f8 | 6322 | goto have_block_group; |
d06e3bb6 QW |
6323 | } else if (ret > 0) { |
6324 | goto loop; | |
fa9c0d79 | 6325 | } |
d06e3bb6 | 6326 | /* ret == -ENOENT case falls through */ |
fa9c0d79 CM |
6327 | } |
6328 | ||
e1a41848 QW |
6329 | ret = find_free_extent_unclustered(block_group, last_ptr, |
6330 | &ffe_ctl); | |
6331 | if (ret == -EAGAIN) | |
817d52f8 | 6332 | goto have_block_group; |
e1a41848 | 6333 | else if (ret > 0) |
1cdda9b8 | 6334 | goto loop; |
e1a41848 | 6335 | /* ret == 0 case falls through */ |
fa9c0d79 | 6336 | checks: |
b4bd745d QW |
6337 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
6338 | fs_info->stripesize); | |
25179201 | 6339 | |
2552d17e | 6340 | /* move on to the next group */ |
b4bd745d | 6341 | if (ffe_ctl.search_start + num_bytes > |
215a63d1 | 6342 | block_group->key.objectid + block_group->key.offset) { |
b4bd745d QW |
6343 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
6344 | num_bytes); | |
2552d17e | 6345 | goto loop; |
6226cb0a | 6346 | } |
f5a31e16 | 6347 | |
b4bd745d QW |
6348 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
6349 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, | |
6350 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 6351 | |
18513091 WX |
6352 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
6353 | num_bytes, delalloc); | |
f0486c68 | 6354 | if (ret == -EAGAIN) { |
b4bd745d QW |
6355 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
6356 | num_bytes); | |
2552d17e | 6357 | goto loop; |
0f9dd46c | 6358 | } |
9cfa3e34 | 6359 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 6360 | |
f0486c68 | 6361 | /* we are all good, lets return */ |
b4bd745d | 6362 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 6363 | ins->offset = num_bytes; |
d2fb3437 | 6364 | |
b4bd745d QW |
6365 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
6366 | num_bytes); | |
e570fd27 | 6367 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
6368 | break; |
6369 | loop: | |
b4bd745d QW |
6370 | ffe_ctl.retry_clustered = false; |
6371 | ffe_ctl.retry_unclustered = false; | |
3e72ee88 | 6372 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
b4bd745d | 6373 | ffe_ctl.index); |
e570fd27 | 6374 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 6375 | cond_resched(); |
2552d17e JB |
6376 | } |
6377 | up_read(&space_info->groups_sem); | |
6378 | ||
e72d79d6 QW |
6379 | ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl, |
6380 | full_search, use_cluster); | |
6381 | if (ret > 0) | |
b742bb82 YZ |
6382 | goto search; |
6383 | ||
4f4db217 | 6384 | if (ret == -ENOSPC) { |
b4bd745d QW |
6385 | /* |
6386 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
6387 | * any contiguous hole. | |
6388 | */ | |
6389 | if (!ffe_ctl.max_extent_size) | |
6390 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 6391 | spin_lock(&space_info->lock); |
b4bd745d | 6392 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 6393 | spin_unlock(&space_info->lock); |
b4bd745d | 6394 | ins->offset = ffe_ctl.max_extent_size; |
4f4db217 | 6395 | } |
0f70abe2 | 6396 | return ret; |
fec577fb | 6397 | } |
ec44a35c | 6398 | |
6f47c706 NB |
6399 | /* |
6400 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
6401 | * hole that is at least as big as @num_bytes. | |
6402 | * | |
6403 | * @root - The root that will contain this extent | |
6404 | * | |
6405 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
6406 | * is used for accounting purposes. This value differs | |
6407 | * from @num_bytes only in the case of compressed extents. | |
6408 | * | |
6409 | * @num_bytes - Number of bytes to allocate on-disk. | |
6410 | * | |
6411 | * @min_alloc_size - Indicates the minimum amount of space that the | |
6412 | * allocator should try to satisfy. In some cases | |
6413 | * @num_bytes may be larger than what is required and if | |
6414 | * the filesystem is fragmented then allocation fails. | |
6415 | * However, the presence of @min_alloc_size gives a | |
6416 | * chance to try and satisfy the smaller allocation. | |
6417 | * | |
6418 | * @empty_size - A hint that you plan on doing more COW. This is the | |
6419 | * size in bytes the allocator should try to find free | |
6420 | * next to the block it returns. This is just a hint and | |
6421 | * may be ignored by the allocator. | |
6422 | * | |
6423 | * @hint_byte - Hint to the allocator to start searching above the byte | |
6424 | * address passed. It might be ignored. | |
6425 | * | |
6426 | * @ins - This key is modified to record the found hole. It will | |
6427 | * have the following values: | |
6428 | * ins->objectid == start position | |
6429 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
6430 | * ins->offset == the size of the hole. | |
6431 | * | |
6432 | * @is_data - Boolean flag indicating whether an extent is | |
6433 | * allocated for data (true) or metadata (false) | |
6434 | * | |
6435 | * @delalloc - Boolean flag indicating whether this allocation is for | |
6436 | * delalloc or not. If 'true' data_rwsem of block groups | |
6437 | * is going to be acquired. | |
6438 | * | |
6439 | * | |
6440 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
6441 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
6442 | * largest available hole the allocator managed to find. | |
6443 | */ | |
18513091 | 6444 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
6445 | u64 num_bytes, u64 min_alloc_size, |
6446 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 6447 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 6448 | { |
ab8d0fc4 | 6449 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 6450 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 6451 | u64 flags; |
fec577fb | 6452 | int ret; |
925baedd | 6453 | |
1b86826d | 6454 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 6455 | again: |
0b246afa | 6456 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 6457 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 6458 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 6459 | if (!ret && !is_data) { |
ab8d0fc4 | 6460 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 6461 | } else if (ret == -ENOSPC) { |
a4820398 MX |
6462 | if (!final_tried && ins->offset) { |
6463 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 6464 | num_bytes = round_down(num_bytes, |
0b246afa | 6465 | fs_info->sectorsize); |
9e622d6b | 6466 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 6467 | ram_bytes = num_bytes; |
9e622d6b MX |
6468 | if (num_bytes == min_alloc_size) |
6469 | final_tried = true; | |
6470 | goto again; | |
ab8d0fc4 | 6471 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
6472 | struct btrfs_space_info *sinfo; |
6473 | ||
280c2908 | 6474 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 6475 | btrfs_err(fs_info, |
5d163e0e JM |
6476 | "allocation failed flags %llu, wanted %llu", |
6477 | flags, num_bytes); | |
53804280 | 6478 | if (sinfo) |
5da6afeb JB |
6479 | btrfs_dump_space_info(fs_info, sinfo, |
6480 | num_bytes, 1); | |
9e622d6b | 6481 | } |
925baedd | 6482 | } |
0f9dd46c JB |
6483 | |
6484 | return ret; | |
e6dcd2dc CM |
6485 | } |
6486 | ||
2ff7e61e | 6487 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
6488 | u64 start, u64 len, |
6489 | int pin, int delalloc) | |
65b51a00 | 6490 | { |
0f9dd46c | 6491 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 6492 | int ret = 0; |
0f9dd46c | 6493 | |
0b246afa | 6494 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 6495 | if (!cache) { |
0b246afa JM |
6496 | btrfs_err(fs_info, "Unable to find block group for %llu", |
6497 | start); | |
0f9dd46c JB |
6498 | return -ENOSPC; |
6499 | } | |
1f3c79a2 | 6500 | |
e688b725 | 6501 | if (pin) |
fdf08605 | 6502 | pin_down_extent(cache, start, len, 1); |
e688b725 | 6503 | else { |
0b246afa | 6504 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6505 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 6506 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 6507 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 6508 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 6509 | } |
31193213 | 6510 | |
fa9c0d79 | 6511 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
6512 | return ret; |
6513 | } | |
6514 | ||
2ff7e61e | 6515 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 6516 | u64 start, u64 len, int delalloc) |
e688b725 | 6517 | { |
2ff7e61e | 6518 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
6519 | } |
6520 | ||
2ff7e61e | 6521 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6522 | u64 start, u64 len) |
6523 | { | |
2ff7e61e | 6524 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
6525 | } |
6526 | ||
5d4f98a2 | 6527 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6528 | u64 parent, u64 root_objectid, |
6529 | u64 flags, u64 owner, u64 offset, | |
6530 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 6531 | { |
ef89b824 | 6532 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 6533 | int ret; |
e6dcd2dc | 6534 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 6535 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 6536 | struct btrfs_path *path; |
5d4f98a2 YZ |
6537 | struct extent_buffer *leaf; |
6538 | int type; | |
6539 | u32 size; | |
26b8003f | 6540 | |
5d4f98a2 YZ |
6541 | if (parent > 0) |
6542 | type = BTRFS_SHARED_DATA_REF_KEY; | |
6543 | else | |
6544 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 6545 | |
5d4f98a2 | 6546 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
6547 | |
6548 | path = btrfs_alloc_path(); | |
db5b493a TI |
6549 | if (!path) |
6550 | return -ENOMEM; | |
47e4bb98 | 6551 | |
b9473439 | 6552 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6553 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
6554 | ins, size); | |
79787eaa JM |
6555 | if (ret) { |
6556 | btrfs_free_path(path); | |
6557 | return ret; | |
6558 | } | |
0f9dd46c | 6559 | |
5d4f98a2 YZ |
6560 | leaf = path->nodes[0]; |
6561 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 6562 | struct btrfs_extent_item); |
5d4f98a2 YZ |
6563 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
6564 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6565 | btrfs_set_extent_flags(leaf, extent_item, | |
6566 | flags | BTRFS_EXTENT_FLAG_DATA); | |
6567 | ||
6568 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6569 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
6570 | if (parent > 0) { | |
6571 | struct btrfs_shared_data_ref *ref; | |
6572 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
6573 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6574 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
6575 | } else { | |
6576 | struct btrfs_extent_data_ref *ref; | |
6577 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
6578 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
6579 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
6580 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
6581 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
6582 | } | |
47e4bb98 CM |
6583 | |
6584 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 6585 | btrfs_free_path(path); |
f510cfec | 6586 | |
25a356d3 | 6587 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
6588 | if (ret) |
6589 | return ret; | |
6590 | ||
6b279408 | 6591 | ret = update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 6592 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 6593 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 6594 | ins->objectid, ins->offset); |
f5947066 CM |
6595 | BUG(); |
6596 | } | |
71ff6437 | 6597 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
6598 | return ret; |
6599 | } | |
6600 | ||
5d4f98a2 | 6601 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 6602 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 6603 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 6604 | { |
9dcdbe01 | 6605 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 6606 | int ret; |
5d4f98a2 | 6607 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 6608 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
6609 | struct btrfs_tree_block_info *block_info; |
6610 | struct btrfs_extent_inline_ref *iref; | |
6611 | struct btrfs_path *path; | |
6612 | struct extent_buffer *leaf; | |
4e6bd4e0 | 6613 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 6614 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 6615 | u64 num_bytes; |
21ebfbe7 | 6616 | u64 flags = extent_op->flags_to_set; |
0b246afa | 6617 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 6618 | |
4e6bd4e0 NB |
6619 | ref = btrfs_delayed_node_to_tree_ref(node); |
6620 | ||
4e6bd4e0 NB |
6621 | extent_key.objectid = node->bytenr; |
6622 | if (skinny_metadata) { | |
6623 | extent_key.offset = ref->level; | |
6624 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
6625 | num_bytes = fs_info->nodesize; | |
6626 | } else { | |
6627 | extent_key.offset = node->num_bytes; | |
6628 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 6629 | size += sizeof(*block_info); |
4e6bd4e0 NB |
6630 | num_bytes = node->num_bytes; |
6631 | } | |
1c2308f8 | 6632 | |
5d4f98a2 | 6633 | path = btrfs_alloc_path(); |
80ee54bf | 6634 | if (!path) |
d8926bb3 | 6635 | return -ENOMEM; |
56bec294 | 6636 | |
5d4f98a2 YZ |
6637 | path->leave_spinning = 1; |
6638 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 6639 | &extent_key, size); |
79787eaa | 6640 | if (ret) { |
dd825259 | 6641 | btrfs_free_path(path); |
79787eaa JM |
6642 | return ret; |
6643 | } | |
5d4f98a2 YZ |
6644 | |
6645 | leaf = path->nodes[0]; | |
6646 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
6647 | struct btrfs_extent_item); | |
6648 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
6649 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6650 | btrfs_set_extent_flags(leaf, extent_item, | |
6651 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 6652 | |
3173a18f JB |
6653 | if (skinny_metadata) { |
6654 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6655 | } else { | |
6656 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 6657 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 6658 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
6659 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
6660 | } | |
5d4f98a2 | 6661 | |
d4b20733 | 6662 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
6663 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
6664 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6665 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 6666 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
6667 | } else { |
6668 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6669 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 6670 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
6671 | } |
6672 | ||
6673 | btrfs_mark_buffer_dirty(leaf); | |
6674 | btrfs_free_path(path); | |
6675 | ||
4e6bd4e0 NB |
6676 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
6677 | num_bytes); | |
1e144fb8 OS |
6678 | if (ret) |
6679 | return ret; | |
6680 | ||
6b279408 | 6681 | ret = update_block_group(trans, extent_key.objectid, |
6202df69 | 6682 | fs_info->nodesize, 1); |
79787eaa | 6683 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 6684 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 6685 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
6686 | BUG(); |
6687 | } | |
0be5dc67 | 6688 | |
4e6bd4e0 | 6689 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 6690 | fs_info->nodesize); |
5d4f98a2 YZ |
6691 | return ret; |
6692 | } | |
6693 | ||
6694 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 6695 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
6696 | u64 offset, u64 ram_bytes, |
6697 | struct btrfs_key *ins) | |
5d4f98a2 | 6698 | { |
76675593 | 6699 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 YZ |
6700 | int ret; |
6701 | ||
84f7d8e6 | 6702 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 6703 | |
76675593 QW |
6704 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
6705 | ins->objectid, ins->offset, 0); | |
6706 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 6707 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
76675593 QW |
6708 | ret = btrfs_add_delayed_data_ref(trans, &generic_ref, |
6709 | ram_bytes, NULL, NULL); | |
e6dcd2dc CM |
6710 | return ret; |
6711 | } | |
e02119d5 CM |
6712 | |
6713 | /* | |
6714 | * this is used by the tree logging recovery code. It records that | |
6715 | * an extent has been allocated and makes sure to clear the free | |
6716 | * space cache bits as well | |
6717 | */ | |
5d4f98a2 | 6718 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6719 | u64 root_objectid, u64 owner, u64 offset, |
6720 | struct btrfs_key *ins) | |
e02119d5 | 6721 | { |
61da2abf | 6722 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 CM |
6723 | int ret; |
6724 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 6725 | struct btrfs_space_info *space_info; |
11833d66 | 6726 | |
8c2a1a30 JB |
6727 | /* |
6728 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 6729 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 6730 | */ |
0b246afa | 6731 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
6732 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
6733 | ins->offset); | |
b50c6e25 | 6734 | if (ret) |
8c2a1a30 | 6735 | return ret; |
11833d66 YZ |
6736 | } |
6737 | ||
0b246afa | 6738 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
6739 | if (!block_group) |
6740 | return -EINVAL; | |
6741 | ||
ed7a6948 WX |
6742 | space_info = block_group->space_info; |
6743 | spin_lock(&space_info->lock); | |
6744 | spin_lock(&block_group->lock); | |
6745 | space_info->bytes_reserved += ins->offset; | |
6746 | block_group->reserved += ins->offset; | |
6747 | spin_unlock(&block_group->lock); | |
6748 | spin_unlock(&space_info->lock); | |
6749 | ||
ef89b824 NB |
6750 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
6751 | offset, ins, 1); | |
b50c6e25 | 6752 | btrfs_put_block_group(block_group); |
e02119d5 CM |
6753 | return ret; |
6754 | } | |
6755 | ||
48a3b636 ES |
6756 | static struct extent_buffer * |
6757 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
bc877d28 | 6758 | u64 bytenr, int level, u64 owner) |
65b51a00 | 6759 | { |
0b246afa | 6760 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
6761 | struct extent_buffer *buf; |
6762 | ||
2ff7e61e | 6763 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
6764 | if (IS_ERR(buf)) |
6765 | return buf; | |
6766 | ||
b72c3aba QW |
6767 | /* |
6768 | * Extra safety check in case the extent tree is corrupted and extent | |
6769 | * allocator chooses to use a tree block which is already used and | |
6770 | * locked. | |
6771 | */ | |
6772 | if (buf->lock_owner == current->pid) { | |
6773 | btrfs_err_rl(fs_info, | |
6774 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
6775 | buf->start, btrfs_header_owner(buf), current->pid); | |
6776 | free_extent_buffer(buf); | |
6777 | return ERR_PTR(-EUCLEAN); | |
6778 | } | |
6779 | ||
85d4e461 | 6780 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 6781 | btrfs_tree_lock(buf); |
6a884d7d | 6782 | btrfs_clean_tree_block(buf); |
3083ee2e | 6783 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 6784 | |
8bead258 | 6785 | btrfs_set_lock_blocking_write(buf); |
4db8c528 | 6786 | set_extent_buffer_uptodate(buf); |
b4ce94de | 6787 | |
bc877d28 NB |
6788 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
6789 | btrfs_set_header_level(buf, level); | |
6790 | btrfs_set_header_bytenr(buf, buf->start); | |
6791 | btrfs_set_header_generation(buf, trans->transid); | |
6792 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
6793 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 6794 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 6795 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 6796 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 6797 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
6798 | /* |
6799 | * we allow two log transactions at a time, use different | |
52042d8e | 6800 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 6801 | */ |
656f30db | 6802 | if (buf->log_index == 0) |
8cef4e16 YZ |
6803 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
6804 | buf->start + buf->len - 1, GFP_NOFS); | |
6805 | else | |
6806 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 6807 | buf->start + buf->len - 1); |
d0c803c4 | 6808 | } else { |
656f30db | 6809 | buf->log_index = -1; |
d0c803c4 | 6810 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 6811 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6812 | } |
64c12921 | 6813 | trans->dirty = true; |
b4ce94de | 6814 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6815 | return buf; |
6816 | } | |
6817 | ||
fec577fb | 6818 | /* |
f0486c68 | 6819 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 6820 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 6821 | */ |
4d75f8a9 | 6822 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
6823 | struct btrfs_root *root, |
6824 | u64 parent, u64 root_objectid, | |
6825 | const struct btrfs_disk_key *key, | |
6826 | int level, u64 hint, | |
6827 | u64 empty_size) | |
fec577fb | 6828 | { |
0b246afa | 6829 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 6830 | struct btrfs_key ins; |
f0486c68 | 6831 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6832 | struct extent_buffer *buf; |
67b7859e | 6833 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 6834 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
6835 | u64 flags = 0; |
6836 | int ret; | |
0b246afa JM |
6837 | u32 blocksize = fs_info->nodesize; |
6838 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 6839 | |
05653ef3 | 6840 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 6841 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 6842 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
bc877d28 | 6843 | level, root_objectid); |
faa2dbf0 JB |
6844 | if (!IS_ERR(buf)) |
6845 | root->alloc_bytenr += blocksize; | |
6846 | return buf; | |
6847 | } | |
05653ef3 | 6848 | #endif |
fccb84c9 | 6849 | |
67f9c220 | 6850 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
6851 | if (IS_ERR(block_rsv)) |
6852 | return ERR_CAST(block_rsv); | |
6853 | ||
18513091 | 6854 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 6855 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
6856 | if (ret) |
6857 | goto out_unuse; | |
55c69072 | 6858 | |
bc877d28 NB |
6859 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
6860 | root_objectid); | |
67b7859e OS |
6861 | if (IS_ERR(buf)) { |
6862 | ret = PTR_ERR(buf); | |
6863 | goto out_free_reserved; | |
6864 | } | |
f0486c68 YZ |
6865 | |
6866 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6867 | if (parent == 0) | |
6868 | parent = ins.objectid; | |
6869 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6870 | } else | |
6871 | BUG_ON(parent > 0); | |
6872 | ||
6873 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 6874 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
6875 | if (!extent_op) { |
6876 | ret = -ENOMEM; | |
6877 | goto out_free_buf; | |
6878 | } | |
f0486c68 YZ |
6879 | if (key) |
6880 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6881 | else | |
6882 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6883 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
6884 | extent_op->update_key = skinny_metadata ? false : true; |
6885 | extent_op->update_flags = true; | |
6886 | extent_op->is_data = false; | |
b1c79e09 | 6887 | extent_op->level = level; |
f0486c68 | 6888 | |
ed4f255b QW |
6889 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
6890 | ins.objectid, ins.offset, parent); | |
6891 | generic_ref.real_root = root->root_key.objectid; | |
6892 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 6893 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 6894 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, |
7be07912 | 6895 | extent_op, NULL, NULL); |
67b7859e OS |
6896 | if (ret) |
6897 | goto out_free_delayed; | |
f0486c68 | 6898 | } |
fec577fb | 6899 | return buf; |
67b7859e OS |
6900 | |
6901 | out_free_delayed: | |
6902 | btrfs_free_delayed_extent_op(extent_op); | |
6903 | out_free_buf: | |
6904 | free_extent_buffer(buf); | |
6905 | out_free_reserved: | |
2ff7e61e | 6906 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 6907 | out_unuse: |
67f9c220 | 6908 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 6909 | return ERR_PTR(ret); |
fec577fb | 6910 | } |
a28ec197 | 6911 | |
2c47e605 YZ |
6912 | struct walk_control { |
6913 | u64 refs[BTRFS_MAX_LEVEL]; | |
6914 | u64 flags[BTRFS_MAX_LEVEL]; | |
6915 | struct btrfs_key update_progress; | |
aea6f028 JB |
6916 | struct btrfs_key drop_progress; |
6917 | int drop_level; | |
2c47e605 YZ |
6918 | int stage; |
6919 | int level; | |
6920 | int shared_level; | |
6921 | int update_ref; | |
6922 | int keep_locks; | |
1c4850e2 YZ |
6923 | int reada_slot; |
6924 | int reada_count; | |
78c52d9e | 6925 | int restarted; |
2c47e605 YZ |
6926 | }; |
6927 | ||
6928 | #define DROP_REFERENCE 1 | |
6929 | #define UPDATE_BACKREF 2 | |
6930 | ||
1c4850e2 YZ |
6931 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6932 | struct btrfs_root *root, | |
6933 | struct walk_control *wc, | |
6934 | struct btrfs_path *path) | |
6407bf6d | 6935 | { |
0b246afa | 6936 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
6937 | u64 bytenr; |
6938 | u64 generation; | |
6939 | u64 refs; | |
94fcca9f | 6940 | u64 flags; |
5d4f98a2 | 6941 | u32 nritems; |
1c4850e2 YZ |
6942 | struct btrfs_key key; |
6943 | struct extent_buffer *eb; | |
6407bf6d | 6944 | int ret; |
1c4850e2 YZ |
6945 | int slot; |
6946 | int nread = 0; | |
6407bf6d | 6947 | |
1c4850e2 YZ |
6948 | if (path->slots[wc->level] < wc->reada_slot) { |
6949 | wc->reada_count = wc->reada_count * 2 / 3; | |
6950 | wc->reada_count = max(wc->reada_count, 2); | |
6951 | } else { | |
6952 | wc->reada_count = wc->reada_count * 3 / 2; | |
6953 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 6954 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 6955 | } |
7bb86316 | 6956 | |
1c4850e2 YZ |
6957 | eb = path->nodes[wc->level]; |
6958 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 6959 | |
1c4850e2 YZ |
6960 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6961 | if (nread >= wc->reada_count) | |
6962 | break; | |
bd56b302 | 6963 | |
2dd3e67b | 6964 | cond_resched(); |
1c4850e2 YZ |
6965 | bytenr = btrfs_node_blockptr(eb, slot); |
6966 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6967 | |
1c4850e2 YZ |
6968 | if (slot == path->slots[wc->level]) |
6969 | goto reada; | |
5d4f98a2 | 6970 | |
1c4850e2 YZ |
6971 | if (wc->stage == UPDATE_BACKREF && |
6972 | generation <= root->root_key.offset) | |
bd56b302 CM |
6973 | continue; |
6974 | ||
94fcca9f | 6975 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 6976 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
6977 | wc->level - 1, 1, &refs, |
6978 | &flags); | |
79787eaa JM |
6979 | /* We don't care about errors in readahead. */ |
6980 | if (ret < 0) | |
6981 | continue; | |
94fcca9f YZ |
6982 | BUG_ON(refs == 0); |
6983 | ||
1c4850e2 | 6984 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6985 | if (refs == 1) |
6986 | goto reada; | |
bd56b302 | 6987 | |
94fcca9f YZ |
6988 | if (wc->level == 1 && |
6989 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6990 | continue; | |
1c4850e2 YZ |
6991 | if (!wc->update_ref || |
6992 | generation <= root->root_key.offset) | |
6993 | continue; | |
6994 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6995 | ret = btrfs_comp_cpu_keys(&key, | |
6996 | &wc->update_progress); | |
6997 | if (ret < 0) | |
6998 | continue; | |
94fcca9f YZ |
6999 | } else { |
7000 | if (wc->level == 1 && | |
7001 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7002 | continue; | |
6407bf6d | 7003 | } |
1c4850e2 | 7004 | reada: |
2ff7e61e | 7005 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 7006 | nread++; |
20524f02 | 7007 | } |
1c4850e2 | 7008 | wc->reada_slot = slot; |
20524f02 | 7009 | } |
2c47e605 | 7010 | |
f82d02d9 | 7011 | /* |
2c016dc2 | 7012 | * helper to process tree block while walking down the tree. |
2c47e605 | 7013 | * |
2c47e605 YZ |
7014 | * when wc->stage == UPDATE_BACKREF, this function updates |
7015 | * back refs for pointers in the block. | |
7016 | * | |
7017 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 7018 | */ |
2c47e605 | 7019 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 7020 | struct btrfs_root *root, |
2c47e605 | 7021 | struct btrfs_path *path, |
94fcca9f | 7022 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 7023 | { |
2ff7e61e | 7024 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
7025 | int level = wc->level; |
7026 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 7027 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
7028 | int ret; |
7029 | ||
2c47e605 YZ |
7030 | if (wc->stage == UPDATE_BACKREF && |
7031 | btrfs_header_owner(eb) != root->root_key.objectid) | |
7032 | return 1; | |
f82d02d9 | 7033 | |
2c47e605 YZ |
7034 | /* |
7035 | * when reference count of tree block is 1, it won't increase | |
7036 | * again. once full backref flag is set, we never clear it. | |
7037 | */ | |
94fcca9f YZ |
7038 | if (lookup_info && |
7039 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
7040 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 7041 | BUG_ON(!path->locks[level]); |
2ff7e61e | 7042 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 7043 | eb->start, level, 1, |
2c47e605 YZ |
7044 | &wc->refs[level], |
7045 | &wc->flags[level]); | |
79787eaa JM |
7046 | BUG_ON(ret == -ENOMEM); |
7047 | if (ret) | |
7048 | return ret; | |
2c47e605 YZ |
7049 | BUG_ON(wc->refs[level] == 0); |
7050 | } | |
5d4f98a2 | 7051 | |
2c47e605 YZ |
7052 | if (wc->stage == DROP_REFERENCE) { |
7053 | if (wc->refs[level] > 1) | |
7054 | return 1; | |
f82d02d9 | 7055 | |
2c47e605 | 7056 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 7057 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
7058 | path->locks[level] = 0; |
7059 | } | |
7060 | return 0; | |
7061 | } | |
f82d02d9 | 7062 | |
2c47e605 YZ |
7063 | /* wc->stage == UPDATE_BACKREF */ |
7064 | if (!(wc->flags[level] & flag)) { | |
7065 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 7066 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 7067 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 7068 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 7069 | BUG_ON(ret); /* -ENOMEM */ |
f5c8daa5 | 7070 | ret = btrfs_set_disk_extent_flags(trans, eb->start, |
b1c79e09 JB |
7071 | eb->len, flag, |
7072 | btrfs_header_level(eb), 0); | |
79787eaa | 7073 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
7074 | wc->flags[level] |= flag; |
7075 | } | |
7076 | ||
7077 | /* | |
7078 | * the block is shared by multiple trees, so it's not good to | |
7079 | * keep the tree lock | |
7080 | */ | |
7081 | if (path->locks[level] && level > 0) { | |
bd681513 | 7082 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
7083 | path->locks[level] = 0; |
7084 | } | |
7085 | return 0; | |
7086 | } | |
7087 | ||
78c52d9e JB |
7088 | /* |
7089 | * This is used to verify a ref exists for this root to deal with a bug where we | |
7090 | * would have a drop_progress key that hadn't been updated properly. | |
7091 | */ | |
7092 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
7093 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
7094 | int level) | |
7095 | { | |
7096 | struct btrfs_path *path; | |
7097 | struct btrfs_extent_inline_ref *iref; | |
7098 | int ret; | |
7099 | ||
7100 | path = btrfs_alloc_path(); | |
7101 | if (!path) | |
7102 | return -ENOMEM; | |
7103 | ||
7104 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
7105 | root->fs_info->nodesize, parent, | |
7106 | root->root_key.objectid, level, 0); | |
7107 | btrfs_free_path(path); | |
7108 | if (ret == -ENOENT) | |
7109 | return 0; | |
7110 | if (ret < 0) | |
7111 | return ret; | |
7112 | return 1; | |
7113 | } | |
7114 | ||
1c4850e2 | 7115 | /* |
2c016dc2 | 7116 | * helper to process tree block pointer. |
1c4850e2 YZ |
7117 | * |
7118 | * when wc->stage == DROP_REFERENCE, this function checks | |
7119 | * reference count of the block pointed to. if the block | |
7120 | * is shared and we need update back refs for the subtree | |
7121 | * rooted at the block, this function changes wc->stage to | |
7122 | * UPDATE_BACKREF. if the block is shared and there is no | |
7123 | * need to update back, this function drops the reference | |
7124 | * to the block. | |
7125 | * | |
7126 | * NOTE: return value 1 means we should stop walking down. | |
7127 | */ | |
7128 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
7129 | struct btrfs_root *root, | |
7130 | struct btrfs_path *path, | |
94fcca9f | 7131 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 7132 | { |
0b246afa | 7133 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
7134 | u64 bytenr; |
7135 | u64 generation; | |
7136 | u64 parent; | |
1c4850e2 | 7137 | struct btrfs_key key; |
581c1760 | 7138 | struct btrfs_key first_key; |
ffd4bb2a | 7139 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
7140 | struct extent_buffer *next; |
7141 | int level = wc->level; | |
7142 | int reada = 0; | |
7143 | int ret = 0; | |
1152651a | 7144 | bool need_account = false; |
1c4850e2 YZ |
7145 | |
7146 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
7147 | path->slots[level]); | |
7148 | /* | |
7149 | * if the lower level block was created before the snapshot | |
7150 | * was created, we know there is no need to update back refs | |
7151 | * for the subtree | |
7152 | */ | |
7153 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
7154 | generation <= root->root_key.offset) { |
7155 | *lookup_info = 1; | |
1c4850e2 | 7156 | return 1; |
94fcca9f | 7157 | } |
1c4850e2 YZ |
7158 | |
7159 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
7160 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
7161 | path->slots[level]); | |
1c4850e2 | 7162 | |
0b246afa | 7163 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 7164 | if (!next) { |
2ff7e61e | 7165 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
7166 | if (IS_ERR(next)) |
7167 | return PTR_ERR(next); | |
7168 | ||
b2aaaa3b JB |
7169 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
7170 | level - 1); | |
1c4850e2 YZ |
7171 | reada = 1; |
7172 | } | |
7173 | btrfs_tree_lock(next); | |
8bead258 | 7174 | btrfs_set_lock_blocking_write(next); |
1c4850e2 | 7175 | |
2ff7e61e | 7176 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
7177 | &wc->refs[level - 1], |
7178 | &wc->flags[level - 1]); | |
4867268c JB |
7179 | if (ret < 0) |
7180 | goto out_unlock; | |
79787eaa | 7181 | |
c2cf52eb | 7182 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 7183 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
7184 | ret = -EIO; |
7185 | goto out_unlock; | |
c2cf52eb | 7186 | } |
94fcca9f | 7187 | *lookup_info = 0; |
1c4850e2 | 7188 | |
94fcca9f | 7189 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 7190 | if (wc->refs[level - 1] > 1) { |
1152651a | 7191 | need_account = true; |
94fcca9f YZ |
7192 | if (level == 1 && |
7193 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7194 | goto skip; | |
7195 | ||
1c4850e2 YZ |
7196 | if (!wc->update_ref || |
7197 | generation <= root->root_key.offset) | |
7198 | goto skip; | |
7199 | ||
7200 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
7201 | path->slots[level]); | |
7202 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
7203 | if (ret < 0) | |
7204 | goto skip; | |
7205 | ||
7206 | wc->stage = UPDATE_BACKREF; | |
7207 | wc->shared_level = level - 1; | |
7208 | } | |
94fcca9f YZ |
7209 | } else { |
7210 | if (level == 1 && | |
7211 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7212 | goto skip; | |
1c4850e2 YZ |
7213 | } |
7214 | ||
b9fab919 | 7215 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
7216 | btrfs_tree_unlock(next); |
7217 | free_extent_buffer(next); | |
7218 | next = NULL; | |
94fcca9f | 7219 | *lookup_info = 1; |
1c4850e2 YZ |
7220 | } |
7221 | ||
7222 | if (!next) { | |
7223 | if (reada && level == 1) | |
7224 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
7225 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
7226 | &first_key); | |
64c043de LB |
7227 | if (IS_ERR(next)) { |
7228 | return PTR_ERR(next); | |
7229 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 7230 | free_extent_buffer(next); |
97d9a8a4 | 7231 | return -EIO; |
416bc658 | 7232 | } |
1c4850e2 | 7233 | btrfs_tree_lock(next); |
8bead258 | 7234 | btrfs_set_lock_blocking_write(next); |
1c4850e2 YZ |
7235 | } |
7236 | ||
7237 | level--; | |
4867268c JB |
7238 | ASSERT(level == btrfs_header_level(next)); |
7239 | if (level != btrfs_header_level(next)) { | |
7240 | btrfs_err(root->fs_info, "mismatched level"); | |
7241 | ret = -EIO; | |
7242 | goto out_unlock; | |
7243 | } | |
1c4850e2 YZ |
7244 | path->nodes[level] = next; |
7245 | path->slots[level] = 0; | |
bd681513 | 7246 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
7247 | wc->level = level; |
7248 | if (wc->level == 1) | |
7249 | wc->reada_slot = 0; | |
7250 | return 0; | |
7251 | skip: | |
7252 | wc->refs[level - 1] = 0; | |
7253 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
7254 | if (wc->stage == DROP_REFERENCE) { |
7255 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
7256 | parent = path->nodes[level]->start; | |
7257 | } else { | |
4867268c | 7258 | ASSERT(root->root_key.objectid == |
94fcca9f | 7259 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
7260 | if (root->root_key.objectid != |
7261 | btrfs_header_owner(path->nodes[level])) { | |
7262 | btrfs_err(root->fs_info, | |
7263 | "mismatched block owner"); | |
7264 | ret = -EIO; | |
7265 | goto out_unlock; | |
7266 | } | |
94fcca9f YZ |
7267 | parent = 0; |
7268 | } | |
1c4850e2 | 7269 | |
78c52d9e JB |
7270 | /* |
7271 | * If we had a drop_progress we need to verify the refs are set | |
7272 | * as expected. If we find our ref then we know that from here | |
7273 | * on out everything should be correct, and we can clear the | |
7274 | * ->restarted flag. | |
7275 | */ | |
7276 | if (wc->restarted) { | |
7277 | ret = check_ref_exists(trans, root, bytenr, parent, | |
7278 | level - 1); | |
7279 | if (ret < 0) | |
7280 | goto out_unlock; | |
7281 | if (ret == 0) | |
7282 | goto no_delete; | |
7283 | ret = 0; | |
7284 | wc->restarted = 0; | |
7285 | } | |
7286 | ||
2cd86d30 QW |
7287 | /* |
7288 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
7289 | * already accounted them at merge time (replace_path), | |
7290 | * thus we could skip expensive subtree trace here. | |
7291 | */ | |
7292 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
7293 | need_account) { | |
deb40627 | 7294 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 7295 | generation, level - 1); |
1152651a | 7296 | if (ret) { |
0b246afa | 7297 | btrfs_err_rl(fs_info, |
5d163e0e JM |
7298 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
7299 | ret); | |
1152651a MF |
7300 | } |
7301 | } | |
aea6f028 JB |
7302 | |
7303 | /* | |
7304 | * We need to update the next key in our walk control so we can | |
7305 | * update the drop_progress key accordingly. We don't care if | |
7306 | * find_next_key doesn't find a key because that means we're at | |
7307 | * the end and are going to clean up now. | |
7308 | */ | |
7309 | wc->drop_level = level; | |
7310 | find_next_key(path, level, &wc->drop_progress); | |
7311 | ||
ffd4bb2a QW |
7312 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
7313 | fs_info->nodesize, parent); | |
7314 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
7315 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
7316 | if (ret) |
7317 | goto out_unlock; | |
1c4850e2 | 7318 | } |
78c52d9e | 7319 | no_delete: |
4867268c JB |
7320 | *lookup_info = 1; |
7321 | ret = 1; | |
7322 | ||
7323 | out_unlock: | |
1c4850e2 YZ |
7324 | btrfs_tree_unlock(next); |
7325 | free_extent_buffer(next); | |
4867268c JB |
7326 | |
7327 | return ret; | |
1c4850e2 YZ |
7328 | } |
7329 | ||
2c47e605 | 7330 | /* |
2c016dc2 | 7331 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
7332 | * |
7333 | * when wc->stage == DROP_REFERENCE, this function drops | |
7334 | * reference count on the block. | |
7335 | * | |
7336 | * when wc->stage == UPDATE_BACKREF, this function changes | |
7337 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
7338 | * to UPDATE_BACKREF previously while processing the block. | |
7339 | * | |
7340 | * NOTE: return value 1 means we should stop walking up. | |
7341 | */ | |
7342 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
7343 | struct btrfs_root *root, | |
7344 | struct btrfs_path *path, | |
7345 | struct walk_control *wc) | |
7346 | { | |
0b246afa | 7347 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 7348 | int ret; |
2c47e605 YZ |
7349 | int level = wc->level; |
7350 | struct extent_buffer *eb = path->nodes[level]; | |
7351 | u64 parent = 0; | |
7352 | ||
7353 | if (wc->stage == UPDATE_BACKREF) { | |
7354 | BUG_ON(wc->shared_level < level); | |
7355 | if (level < wc->shared_level) | |
7356 | goto out; | |
7357 | ||
2c47e605 YZ |
7358 | ret = find_next_key(path, level + 1, &wc->update_progress); |
7359 | if (ret > 0) | |
7360 | wc->update_ref = 0; | |
7361 | ||
7362 | wc->stage = DROP_REFERENCE; | |
7363 | wc->shared_level = -1; | |
7364 | path->slots[level] = 0; | |
7365 | ||
7366 | /* | |
7367 | * check reference count again if the block isn't locked. | |
7368 | * we should start walking down the tree again if reference | |
7369 | * count is one. | |
7370 | */ | |
7371 | if (!path->locks[level]) { | |
7372 | BUG_ON(level == 0); | |
7373 | btrfs_tree_lock(eb); | |
8bead258 | 7374 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 7375 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 7376 | |
2ff7e61e | 7377 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 7378 | eb->start, level, 1, |
2c47e605 YZ |
7379 | &wc->refs[level], |
7380 | &wc->flags[level]); | |
79787eaa JM |
7381 | if (ret < 0) { |
7382 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 7383 | path->locks[level] = 0; |
79787eaa JM |
7384 | return ret; |
7385 | } | |
2c47e605 YZ |
7386 | BUG_ON(wc->refs[level] == 0); |
7387 | if (wc->refs[level] == 1) { | |
bd681513 | 7388 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 7389 | path->locks[level] = 0; |
2c47e605 YZ |
7390 | return 1; |
7391 | } | |
f82d02d9 | 7392 | } |
2c47e605 | 7393 | } |
f82d02d9 | 7394 | |
2c47e605 YZ |
7395 | /* wc->stage == DROP_REFERENCE */ |
7396 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 7397 | |
2c47e605 YZ |
7398 | if (wc->refs[level] == 1) { |
7399 | if (level == 0) { | |
7400 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 7401 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 7402 | else |
e339a6b0 | 7403 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 7404 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
7405 | if (is_fstree(root->root_key.objectid)) { |
7406 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
7407 | if (ret) { | |
7408 | btrfs_err_rl(fs_info, | |
7409 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 7410 | ret); |
c4140cbf | 7411 | } |
1152651a | 7412 | } |
2c47e605 | 7413 | } |
6a884d7d | 7414 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
7415 | if (!path->locks[level] && |
7416 | btrfs_header_generation(eb) == trans->transid) { | |
7417 | btrfs_tree_lock(eb); | |
8bead258 | 7418 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 7419 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 7420 | } |
6a884d7d | 7421 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
7422 | } |
7423 | ||
7424 | if (eb == root->node) { | |
7425 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
7426 | parent = eb->start; | |
65c6e82b QW |
7427 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
7428 | goto owner_mismatch; | |
2c47e605 YZ |
7429 | } else { |
7430 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
7431 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
7432 | else if (root->root_key.objectid != |
7433 | btrfs_header_owner(path->nodes[level + 1])) | |
7434 | goto owner_mismatch; | |
f82d02d9 | 7435 | } |
f82d02d9 | 7436 | |
5581a51a | 7437 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
7438 | out: |
7439 | wc->refs[level] = 0; | |
7440 | wc->flags[level] = 0; | |
f0486c68 | 7441 | return 0; |
65c6e82b QW |
7442 | |
7443 | owner_mismatch: | |
7444 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
7445 | btrfs_header_owner(eb), root->root_key.objectid); | |
7446 | return -EUCLEAN; | |
2c47e605 YZ |
7447 | } |
7448 | ||
7449 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
7450 | struct btrfs_root *root, | |
7451 | struct btrfs_path *path, | |
7452 | struct walk_control *wc) | |
7453 | { | |
2c47e605 | 7454 | int level = wc->level; |
94fcca9f | 7455 | int lookup_info = 1; |
2c47e605 YZ |
7456 | int ret; |
7457 | ||
7458 | while (level >= 0) { | |
94fcca9f | 7459 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
7460 | if (ret > 0) |
7461 | break; | |
7462 | ||
7463 | if (level == 0) | |
7464 | break; | |
7465 | ||
7a7965f8 YZ |
7466 | if (path->slots[level] >= |
7467 | btrfs_header_nritems(path->nodes[level])) | |
7468 | break; | |
7469 | ||
94fcca9f | 7470 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
7471 | if (ret > 0) { |
7472 | path->slots[level]++; | |
7473 | continue; | |
90d2c51d MX |
7474 | } else if (ret < 0) |
7475 | return ret; | |
1c4850e2 | 7476 | level = wc->level; |
f82d02d9 | 7477 | } |
f82d02d9 YZ |
7478 | return 0; |
7479 | } | |
7480 | ||
d397712b | 7481 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 7482 | struct btrfs_root *root, |
f82d02d9 | 7483 | struct btrfs_path *path, |
2c47e605 | 7484 | struct walk_control *wc, int max_level) |
20524f02 | 7485 | { |
2c47e605 | 7486 | int level = wc->level; |
20524f02 | 7487 | int ret; |
9f3a7427 | 7488 | |
2c47e605 YZ |
7489 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
7490 | while (level < max_level && path->nodes[level]) { | |
7491 | wc->level = level; | |
7492 | if (path->slots[level] + 1 < | |
7493 | btrfs_header_nritems(path->nodes[level])) { | |
7494 | path->slots[level]++; | |
20524f02 CM |
7495 | return 0; |
7496 | } else { | |
2c47e605 YZ |
7497 | ret = walk_up_proc(trans, root, path, wc); |
7498 | if (ret > 0) | |
7499 | return 0; | |
65c6e82b QW |
7500 | if (ret < 0) |
7501 | return ret; | |
bd56b302 | 7502 | |
2c47e605 | 7503 | if (path->locks[level]) { |
bd681513 CM |
7504 | btrfs_tree_unlock_rw(path->nodes[level], |
7505 | path->locks[level]); | |
2c47e605 | 7506 | path->locks[level] = 0; |
f82d02d9 | 7507 | } |
2c47e605 YZ |
7508 | free_extent_buffer(path->nodes[level]); |
7509 | path->nodes[level] = NULL; | |
7510 | level++; | |
20524f02 CM |
7511 | } |
7512 | } | |
7513 | return 1; | |
7514 | } | |
7515 | ||
9aca1d51 | 7516 | /* |
2c47e605 YZ |
7517 | * drop a subvolume tree. |
7518 | * | |
7519 | * this function traverses the tree freeing any blocks that only | |
7520 | * referenced by the tree. | |
7521 | * | |
7522 | * when a shared tree block is found. this function decreases its | |
7523 | * reference count by one. if update_ref is true, this function | |
7524 | * also make sure backrefs for the shared block and all lower level | |
7525 | * blocks are properly updated. | |
9d1a2a3a DS |
7526 | * |
7527 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 7528 | */ |
2c536799 | 7529 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
7530 | struct btrfs_block_rsv *block_rsv, int update_ref, |
7531 | int for_reloc) | |
20524f02 | 7532 | { |
ab8d0fc4 | 7533 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 7534 | struct btrfs_path *path; |
2c47e605 | 7535 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 7536 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 7537 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
7538 | struct walk_control *wc; |
7539 | struct btrfs_key key; | |
7540 | int err = 0; | |
7541 | int ret; | |
7542 | int level; | |
d29a9f62 | 7543 | bool root_dropped = false; |
20524f02 | 7544 | |
4fd786e6 | 7545 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 7546 | |
5caf2a00 | 7547 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
7548 | if (!path) { |
7549 | err = -ENOMEM; | |
7550 | goto out; | |
7551 | } | |
20524f02 | 7552 | |
2c47e605 | 7553 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
7554 | if (!wc) { |
7555 | btrfs_free_path(path); | |
cb1b69f4 TI |
7556 | err = -ENOMEM; |
7557 | goto out; | |
38a1a919 | 7558 | } |
2c47e605 | 7559 | |
a22285a6 | 7560 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7561 | if (IS_ERR(trans)) { |
7562 | err = PTR_ERR(trans); | |
7563 | goto out_free; | |
7564 | } | |
98d5dc13 | 7565 | |
0568e82d JB |
7566 | err = btrfs_run_delayed_items(trans); |
7567 | if (err) | |
7568 | goto out_end_trans; | |
7569 | ||
3fd0a558 YZ |
7570 | if (block_rsv) |
7571 | trans->block_rsv = block_rsv; | |
2c47e605 | 7572 | |
83354f07 JB |
7573 | /* |
7574 | * This will help us catch people modifying the fs tree while we're | |
7575 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
7576 | * dropped as we unlock the root node and parent nodes as we walk down | |
7577 | * the tree, assuming nothing will change. If something does change | |
7578 | * then we'll have stale information and drop references to blocks we've | |
7579 | * already dropped. | |
7580 | */ | |
7581 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 7582 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 7583 | level = btrfs_header_level(root->node); |
5d4f98a2 | 7584 | path->nodes[level] = btrfs_lock_root_node(root); |
8bead258 | 7585 | btrfs_set_lock_blocking_write(path->nodes[level]); |
9f3a7427 | 7586 | path->slots[level] = 0; |
bd681513 | 7587 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7588 | memset(&wc->update_progress, 0, |
7589 | sizeof(wc->update_progress)); | |
9f3a7427 | 7590 | } else { |
9f3a7427 | 7591 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
7592 | memcpy(&wc->update_progress, &key, |
7593 | sizeof(wc->update_progress)); | |
7594 | ||
6702ed49 | 7595 | level = root_item->drop_level; |
2c47e605 | 7596 | BUG_ON(level == 0); |
6702ed49 | 7597 | path->lowest_level = level; |
2c47e605 YZ |
7598 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
7599 | path->lowest_level = 0; | |
7600 | if (ret < 0) { | |
7601 | err = ret; | |
79787eaa | 7602 | goto out_end_trans; |
9f3a7427 | 7603 | } |
1c4850e2 | 7604 | WARN_ON(ret > 0); |
2c47e605 | 7605 | |
7d9eb12c CM |
7606 | /* |
7607 | * unlock our path, this is safe because only this | |
7608 | * function is allowed to delete this snapshot | |
7609 | */ | |
5d4f98a2 | 7610 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
7611 | |
7612 | level = btrfs_header_level(root->node); | |
7613 | while (1) { | |
7614 | btrfs_tree_lock(path->nodes[level]); | |
8bead258 | 7615 | btrfs_set_lock_blocking_write(path->nodes[level]); |
fec386ac | 7616 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 7617 | |
2ff7e61e | 7618 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 7619 | path->nodes[level]->start, |
3173a18f | 7620 | level, 1, &wc->refs[level], |
2c47e605 | 7621 | &wc->flags[level]); |
79787eaa JM |
7622 | if (ret < 0) { |
7623 | err = ret; | |
7624 | goto out_end_trans; | |
7625 | } | |
2c47e605 YZ |
7626 | BUG_ON(wc->refs[level] == 0); |
7627 | ||
7628 | if (level == root_item->drop_level) | |
7629 | break; | |
7630 | ||
7631 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 7632 | path->locks[level] = 0; |
2c47e605 YZ |
7633 | WARN_ON(wc->refs[level] != 1); |
7634 | level--; | |
7635 | } | |
9f3a7427 | 7636 | } |
2c47e605 | 7637 | |
78c52d9e | 7638 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
7639 | wc->level = level; |
7640 | wc->shared_level = -1; | |
7641 | wc->stage = DROP_REFERENCE; | |
7642 | wc->update_ref = update_ref; | |
7643 | wc->keep_locks = 0; | |
0b246afa | 7644 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 7645 | |
d397712b | 7646 | while (1) { |
9d1a2a3a | 7647 | |
2c47e605 YZ |
7648 | ret = walk_down_tree(trans, root, path, wc); |
7649 | if (ret < 0) { | |
7650 | err = ret; | |
20524f02 | 7651 | break; |
2c47e605 | 7652 | } |
9aca1d51 | 7653 | |
2c47e605 YZ |
7654 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
7655 | if (ret < 0) { | |
7656 | err = ret; | |
20524f02 | 7657 | break; |
2c47e605 YZ |
7658 | } |
7659 | ||
7660 | if (ret > 0) { | |
7661 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
7662 | break; |
7663 | } | |
2c47e605 YZ |
7664 | |
7665 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
7666 | wc->drop_level = wc->level; |
7667 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
7668 | &wc->drop_progress, | |
7669 | path->slots[wc->drop_level]); | |
7670 | } | |
7671 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
7672 | &wc->drop_progress); | |
7673 | root_item->drop_level = wc->drop_level; | |
2c47e605 YZ |
7674 | |
7675 | BUG_ON(wc->level == 0); | |
3a45bb20 | 7676 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 7677 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
7678 | ret = btrfs_update_root(trans, tree_root, |
7679 | &root->root_key, | |
7680 | root_item); | |
79787eaa | 7681 | if (ret) { |
66642832 | 7682 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
7683 | err = ret; |
7684 | goto out_end_trans; | |
7685 | } | |
2c47e605 | 7686 | |
3a45bb20 | 7687 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 7688 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
7689 | btrfs_debug(fs_info, |
7690 | "drop snapshot early exit"); | |
3c8f2422 JB |
7691 | err = -EAGAIN; |
7692 | goto out_free; | |
7693 | } | |
7694 | ||
a22285a6 | 7695 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7696 | if (IS_ERR(trans)) { |
7697 | err = PTR_ERR(trans); | |
7698 | goto out_free; | |
7699 | } | |
3fd0a558 YZ |
7700 | if (block_rsv) |
7701 | trans->block_rsv = block_rsv; | |
c3e69d58 | 7702 | } |
20524f02 | 7703 | } |
b3b4aa74 | 7704 | btrfs_release_path(path); |
79787eaa JM |
7705 | if (err) |
7706 | goto out_end_trans; | |
2c47e605 | 7707 | |
ab9ce7d4 | 7708 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 7709 | if (ret) { |
66642832 | 7710 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 7711 | err = ret; |
79787eaa JM |
7712 | goto out_end_trans; |
7713 | } | |
2c47e605 | 7714 | |
76dda93c | 7715 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
7716 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
7717 | NULL, NULL); | |
79787eaa | 7718 | if (ret < 0) { |
66642832 | 7719 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
7720 | err = ret; |
7721 | goto out_end_trans; | |
7722 | } else if (ret > 0) { | |
84cd948c JB |
7723 | /* if we fail to delete the orphan item this time |
7724 | * around, it'll get picked up the next time. | |
7725 | * | |
7726 | * The most common failure here is just -ENOENT. | |
7727 | */ | |
7728 | btrfs_del_orphan_item(trans, tree_root, | |
7729 | root->root_key.objectid); | |
76dda93c YZ |
7730 | } |
7731 | } | |
7732 | ||
27cdeb70 | 7733 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 7734 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
7735 | } else { |
7736 | free_extent_buffer(root->node); | |
7737 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 7738 | btrfs_put_fs_root(root); |
76dda93c | 7739 | } |
d29a9f62 | 7740 | root_dropped = true; |
79787eaa | 7741 | out_end_trans: |
3a45bb20 | 7742 | btrfs_end_transaction_throttle(trans); |
79787eaa | 7743 | out_free: |
2c47e605 | 7744 | kfree(wc); |
5caf2a00 | 7745 | btrfs_free_path(path); |
cb1b69f4 | 7746 | out: |
d29a9f62 JB |
7747 | /* |
7748 | * So if we need to stop dropping the snapshot for whatever reason we | |
7749 | * need to make sure to add it back to the dead root list so that we | |
7750 | * keep trying to do the work later. This also cleans up roots if we | |
7751 | * don't have it in the radix (like when we recover after a power fail | |
7752 | * or unmount) so we don't leak memory. | |
7753 | */ | |
897ca819 | 7754 | if (!for_reloc && !root_dropped) |
d29a9f62 | 7755 | btrfs_add_dead_root(root); |
90515e7f | 7756 | if (err && err != -EAGAIN) |
ab8d0fc4 | 7757 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 7758 | return err; |
20524f02 | 7759 | } |
9078a3e1 | 7760 | |
2c47e605 YZ |
7761 | /* |
7762 | * drop subtree rooted at tree block 'node'. | |
7763 | * | |
7764 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 7765 | * only used by relocation code |
2c47e605 | 7766 | */ |
f82d02d9 YZ |
7767 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
7768 | struct btrfs_root *root, | |
7769 | struct extent_buffer *node, | |
7770 | struct extent_buffer *parent) | |
7771 | { | |
0b246afa | 7772 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 7773 | struct btrfs_path *path; |
2c47e605 | 7774 | struct walk_control *wc; |
f82d02d9 YZ |
7775 | int level; |
7776 | int parent_level; | |
7777 | int ret = 0; | |
7778 | int wret; | |
7779 | ||
2c47e605 YZ |
7780 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
7781 | ||
f82d02d9 | 7782 | path = btrfs_alloc_path(); |
db5b493a TI |
7783 | if (!path) |
7784 | return -ENOMEM; | |
f82d02d9 | 7785 | |
2c47e605 | 7786 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
7787 | if (!wc) { |
7788 | btrfs_free_path(path); | |
7789 | return -ENOMEM; | |
7790 | } | |
2c47e605 | 7791 | |
b9447ef8 | 7792 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
7793 | parent_level = btrfs_header_level(parent); |
7794 | extent_buffer_get(parent); | |
7795 | path->nodes[parent_level] = parent; | |
7796 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
7797 | ||
b9447ef8 | 7798 | btrfs_assert_tree_locked(node); |
f82d02d9 | 7799 | level = btrfs_header_level(node); |
f82d02d9 YZ |
7800 | path->nodes[level] = node; |
7801 | path->slots[level] = 0; | |
bd681513 | 7802 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7803 | |
7804 | wc->refs[parent_level] = 1; | |
7805 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7806 | wc->level = level; | |
7807 | wc->shared_level = -1; | |
7808 | wc->stage = DROP_REFERENCE; | |
7809 | wc->update_ref = 0; | |
7810 | wc->keep_locks = 1; | |
0b246afa | 7811 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
7812 | |
7813 | while (1) { | |
2c47e605 YZ |
7814 | wret = walk_down_tree(trans, root, path, wc); |
7815 | if (wret < 0) { | |
f82d02d9 | 7816 | ret = wret; |
f82d02d9 | 7817 | break; |
2c47e605 | 7818 | } |
f82d02d9 | 7819 | |
2c47e605 | 7820 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
7821 | if (wret < 0) |
7822 | ret = wret; | |
7823 | if (wret != 0) | |
7824 | break; | |
7825 | } | |
7826 | ||
2c47e605 | 7827 | kfree(wc); |
f82d02d9 YZ |
7828 | btrfs_free_path(path); |
7829 | return ret; | |
7830 | } | |
7831 | ||
6202df69 | 7832 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
7833 | { |
7834 | u64 num_devices; | |
fc67c450 | 7835 | u64 stripped; |
e4d8ec0f | 7836 | |
fc67c450 ID |
7837 | /* |
7838 | * if restripe for this chunk_type is on pick target profile and | |
7839 | * return, otherwise do the usual balance | |
7840 | */ | |
6202df69 | 7841 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
7842 | if (stripped) |
7843 | return extended_to_chunk(stripped); | |
e4d8ec0f | 7844 | |
6202df69 | 7845 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 7846 | |
a07e8a46 | 7847 | stripped = BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
c7369b3f | 7848 | BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10; |
fc67c450 | 7849 | |
ec44a35c CM |
7850 | if (num_devices == 1) { |
7851 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7852 | stripped = flags & ~stripped; | |
7853 | ||
7854 | /* turn raid0 into single device chunks */ | |
7855 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7856 | return stripped; | |
7857 | ||
7858 | /* turn mirroring into duplication */ | |
c7369b3f | 7859 | if (flags & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
ec44a35c CM |
7860 | BTRFS_BLOCK_GROUP_RAID10)) |
7861 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
7862 | } else { |
7863 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7864 | if (flags & stripped) |
7865 | return flags; | |
7866 | ||
7867 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7868 | stripped = flags & ~stripped; | |
7869 | ||
7870 | /* switch duplicated blocks with raid1 */ | |
7871 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7872 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7873 | ||
e3176ca2 | 7874 | /* this is drive concat, leave it alone */ |
ec44a35c | 7875 | } |
e3176ca2 | 7876 | |
ec44a35c CM |
7877 | return flags; |
7878 | } | |
7879 | ||
868f401a | 7880 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 7881 | { |
f0486c68 YZ |
7882 | struct btrfs_space_info *sinfo = cache->space_info; |
7883 | u64 num_bytes; | |
3ece54e5 | 7884 | u64 sinfo_used; |
199c36ea | 7885 | u64 min_allocable_bytes; |
f0486c68 | 7886 | int ret = -ENOSPC; |
0ef3e66b | 7887 | |
199c36ea MX |
7888 | /* |
7889 | * We need some metadata space and system metadata space for | |
7890 | * allocating chunks in some corner cases until we force to set | |
7891 | * it to be readonly. | |
7892 | */ | |
7893 | if ((sinfo->flags & | |
7894 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
7895 | !force) | |
ee22184b | 7896 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
7897 | else |
7898 | min_allocable_bytes = 0; | |
7899 | ||
f0486c68 YZ |
7900 | spin_lock(&sinfo->lock); |
7901 | spin_lock(&cache->lock); | |
61cfea9b W |
7902 | |
7903 | if (cache->ro) { | |
868f401a | 7904 | cache->ro++; |
61cfea9b W |
7905 | ret = 0; |
7906 | goto out; | |
7907 | } | |
7908 | ||
f0486c68 YZ |
7909 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
7910 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
3ece54e5 | 7911 | sinfo_used = btrfs_space_info_used(sinfo, true); |
f0486c68 | 7912 | |
3ece54e5 QW |
7913 | if (sinfo_used + num_bytes + min_allocable_bytes <= |
7914 | sinfo->total_bytes) { | |
f0486c68 | 7915 | sinfo->bytes_readonly += num_bytes; |
868f401a | 7916 | cache->ro++; |
633c0aad | 7917 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
7918 | ret = 0; |
7919 | } | |
61cfea9b | 7920 | out: |
f0486c68 YZ |
7921 | spin_unlock(&cache->lock); |
7922 | spin_unlock(&sinfo->lock); | |
3ece54e5 QW |
7923 | if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { |
7924 | btrfs_info(cache->fs_info, | |
7925 | "unable to make block group %llu ro", | |
7926 | cache->key.objectid); | |
7927 | btrfs_info(cache->fs_info, | |
7928 | "sinfo_used=%llu bg_num_bytes=%llu min_allocable=%llu", | |
7929 | sinfo_used, num_bytes, min_allocable_bytes); | |
5da6afeb | 7930 | btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0); |
3ece54e5 | 7931 | } |
f0486c68 YZ |
7932 | return ret; |
7933 | } | |
7d9eb12c | 7934 | |
c83488af | 7935 | int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache) |
c286ac48 | 7936 | |
f0486c68 | 7937 | { |
c83488af | 7938 | struct btrfs_fs_info *fs_info = cache->fs_info; |
f0486c68 YZ |
7939 | struct btrfs_trans_handle *trans; |
7940 | u64 alloc_flags; | |
7941 | int ret; | |
7d9eb12c | 7942 | |
1bbc621e | 7943 | again: |
5e00f193 | 7944 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
7945 | if (IS_ERR(trans)) |
7946 | return PTR_ERR(trans); | |
5d4f98a2 | 7947 | |
1bbc621e CM |
7948 | /* |
7949 | * we're not allowed to set block groups readonly after the dirty | |
7950 | * block groups cache has started writing. If it already started, | |
7951 | * back off and let this transaction commit | |
7952 | */ | |
0b246afa | 7953 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 7954 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
7955 | u64 transid = trans->transid; |
7956 | ||
0b246afa | 7957 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 7958 | btrfs_end_transaction(trans); |
1bbc621e | 7959 | |
2ff7e61e | 7960 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
7961 | if (ret) |
7962 | return ret; | |
7963 | goto again; | |
7964 | } | |
7965 | ||
153c35b6 CM |
7966 | /* |
7967 | * if we are changing raid levels, try to allocate a corresponding | |
7968 | * block group with the new raid level. | |
7969 | */ | |
0b246afa | 7970 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 7971 | if (alloc_flags != cache->flags) { |
fc471cb0 | 7972 | ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
153c35b6 CM |
7973 | /* |
7974 | * ENOSPC is allowed here, we may have enough space | |
7975 | * already allocated at the new raid level to | |
7976 | * carry on | |
7977 | */ | |
7978 | if (ret == -ENOSPC) | |
7979 | ret = 0; | |
7980 | if (ret < 0) | |
7981 | goto out; | |
7982 | } | |
1bbc621e | 7983 | |
868f401a | 7984 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
7985 | if (!ret) |
7986 | goto out; | |
2ff7e61e | 7987 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
fc471cb0 | 7988 | ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
7989 | if (ret < 0) |
7990 | goto out; | |
868f401a | 7991 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 7992 | out: |
2f081088 | 7993 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 7994 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 7995 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 7996 | check_system_chunk(trans, alloc_flags); |
34441361 | 7997 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 7998 | } |
0b246afa | 7999 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 8000 | |
3a45bb20 | 8001 | btrfs_end_transaction(trans); |
f0486c68 YZ |
8002 | return ret; |
8003 | } | |
5d4f98a2 | 8004 | |
43a7e99d | 8005 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) |
c87f08ca | 8006 | { |
43a7e99d | 8007 | u64 alloc_flags = get_alloc_profile(trans->fs_info, type); |
2ff7e61e | 8008 | |
fc471cb0 | 8009 | return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
c87f08ca CM |
8010 | } |
8011 | ||
6d07bcec MX |
8012 | /* |
8013 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 8014 | * space_info. takes mirrors into account. |
6d07bcec | 8015 | */ |
633c0aad | 8016 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
8017 | { |
8018 | struct btrfs_block_group_cache *block_group; | |
8019 | u64 free_bytes = 0; | |
8020 | int factor; | |
8021 | ||
01327610 | 8022 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
8023 | if (list_empty(&sinfo->ro_bgs)) |
8024 | return 0; | |
8025 | ||
8026 | spin_lock(&sinfo->lock); | |
8027 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
8028 | spin_lock(&block_group->lock); |
8029 | ||
8030 | if (!block_group->ro) { | |
8031 | spin_unlock(&block_group->lock); | |
8032 | continue; | |
8033 | } | |
8034 | ||
46df06b8 | 8035 | factor = btrfs_bg_type_to_factor(block_group->flags); |
6d07bcec MX |
8036 | free_bytes += (block_group->key.offset - |
8037 | btrfs_block_group_used(&block_group->item)) * | |
8038 | factor; | |
8039 | ||
8040 | spin_unlock(&block_group->lock); | |
8041 | } | |
6d07bcec MX |
8042 | spin_unlock(&sinfo->lock); |
8043 | ||
8044 | return free_bytes; | |
8045 | } | |
8046 | ||
2ff7e61e | 8047 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 8048 | { |
f0486c68 YZ |
8049 | struct btrfs_space_info *sinfo = cache->space_info; |
8050 | u64 num_bytes; | |
8051 | ||
8052 | BUG_ON(!cache->ro); | |
8053 | ||
8054 | spin_lock(&sinfo->lock); | |
8055 | spin_lock(&cache->lock); | |
868f401a Z |
8056 | if (!--cache->ro) { |
8057 | num_bytes = cache->key.offset - cache->reserved - | |
8058 | cache->pinned - cache->bytes_super - | |
8059 | btrfs_block_group_used(&cache->item); | |
8060 | sinfo->bytes_readonly -= num_bytes; | |
8061 | list_del_init(&cache->ro_list); | |
8062 | } | |
f0486c68 YZ |
8063 | spin_unlock(&cache->lock); |
8064 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
8065 | } |
8066 | ||
ba1bf481 | 8067 | /* |
52042d8e | 8068 | * Checks to see if it's even possible to relocate this block group. |
ba1bf481 JB |
8069 | * |
8070 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
8071 | * ok to go ahead and try. | |
8072 | */ | |
6bccf3ab | 8073 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 8074 | { |
ba1bf481 JB |
8075 | struct btrfs_block_group_cache *block_group; |
8076 | struct btrfs_space_info *space_info; | |
0b246afa | 8077 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 8078 | struct btrfs_device *device; |
cdcb725c | 8079 | u64 min_free; |
6719db6a JB |
8080 | u64 dev_min = 1; |
8081 | u64 dev_nr = 0; | |
4a5e98f5 | 8082 | u64 target; |
0305bc27 | 8083 | int debug; |
cdcb725c | 8084 | int index; |
ba1bf481 JB |
8085 | int full = 0; |
8086 | int ret = 0; | |
1a40e23b | 8087 | |
0b246afa | 8088 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 8089 | |
0b246afa | 8090 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 8091 | |
ba1bf481 | 8092 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
8093 | if (!block_group) { |
8094 | if (debug) | |
0b246afa | 8095 | btrfs_warn(fs_info, |
0305bc27 QW |
8096 | "can't find block group for bytenr %llu", |
8097 | bytenr); | |
ba1bf481 | 8098 | return -1; |
0305bc27 | 8099 | } |
1a40e23b | 8100 | |
cdcb725c | 8101 | min_free = btrfs_block_group_used(&block_group->item); |
8102 | ||
ba1bf481 | 8103 | /* no bytes used, we're good */ |
cdcb725c | 8104 | if (!min_free) |
1a40e23b ZY |
8105 | goto out; |
8106 | ||
ba1bf481 JB |
8107 | space_info = block_group->space_info; |
8108 | spin_lock(&space_info->lock); | |
17d217fe | 8109 | |
ba1bf481 | 8110 | full = space_info->full; |
17d217fe | 8111 | |
ba1bf481 JB |
8112 | /* |
8113 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
8114 | * relocate it unless we're able to allocate a new chunk below. |
8115 | * | |
8116 | * Otherwise, we need to make sure we have room in the space to handle | |
8117 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 8118 | */ |
7ce618db | 8119 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
8120 | (btrfs_space_info_used(space_info, false) + min_free < |
8121 | space_info->total_bytes)) { | |
ba1bf481 JB |
8122 | spin_unlock(&space_info->lock); |
8123 | goto out; | |
17d217fe | 8124 | } |
ba1bf481 | 8125 | spin_unlock(&space_info->lock); |
ea8c2819 | 8126 | |
ba1bf481 JB |
8127 | /* |
8128 | * ok we don't have enough space, but maybe we have free space on our | |
8129 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
8130 | * alloc devices and guess if we have enough space. if this block |
8131 | * group is going to be restriped, run checks against the target | |
8132 | * profile instead of the current one. | |
ba1bf481 JB |
8133 | */ |
8134 | ret = -1; | |
ea8c2819 | 8135 | |
cdcb725c | 8136 | /* |
8137 | * index: | |
8138 | * 0: raid10 | |
8139 | * 1: raid1 | |
8140 | * 2: dup | |
8141 | * 3: raid0 | |
8142 | * 4: single | |
8143 | */ | |
0b246afa | 8144 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 8145 | if (target) { |
3e72ee88 | 8146 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
8147 | } else { |
8148 | /* | |
8149 | * this is just a balance, so if we were marked as full | |
8150 | * we know there is no space for a new chunk | |
8151 | */ | |
0305bc27 QW |
8152 | if (full) { |
8153 | if (debug) | |
0b246afa JM |
8154 | btrfs_warn(fs_info, |
8155 | "no space to alloc new chunk for block group %llu", | |
8156 | block_group->key.objectid); | |
4a5e98f5 | 8157 | goto out; |
0305bc27 | 8158 | } |
4a5e98f5 | 8159 | |
3e72ee88 | 8160 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
8161 | } |
8162 | ||
e6ec716f | 8163 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 8164 | dev_min = 4; |
6719db6a JB |
8165 | /* Divide by 2 */ |
8166 | min_free >>= 1; | |
e6ec716f | 8167 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 8168 | dev_min = 2; |
e6ec716f | 8169 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
8170 | /* Multiply by 2 */ |
8171 | min_free <<= 1; | |
e6ec716f | 8172 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 8173 | dev_min = fs_devices->rw_devices; |
47c5713f | 8174 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 8175 | } |
8176 | ||
0b246afa | 8177 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 8178 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 8179 | u64 dev_offset; |
56bec294 | 8180 | |
ba1bf481 JB |
8181 | /* |
8182 | * check to make sure we can actually find a chunk with enough | |
8183 | * space to fit our block group in. | |
8184 | */ | |
63a212ab | 8185 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 8186 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
60dfdf25 | 8187 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 8188 | &dev_offset, NULL); |
ba1bf481 | 8189 | if (!ret) |
cdcb725c | 8190 | dev_nr++; |
8191 | ||
8192 | if (dev_nr >= dev_min) | |
73e48b27 | 8193 | break; |
cdcb725c | 8194 | |
ba1bf481 | 8195 | ret = -1; |
725c8463 | 8196 | } |
edbd8d4e | 8197 | } |
0305bc27 | 8198 | if (debug && ret == -1) |
0b246afa JM |
8199 | btrfs_warn(fs_info, |
8200 | "no space to allocate a new chunk for block group %llu", | |
8201 | block_group->key.objectid); | |
8202 | mutex_unlock(&fs_info->chunk_mutex); | |
edbd8d4e | 8203 | out: |
ba1bf481 | 8204 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
8205 | return ret; |
8206 | } | |
8207 | ||
6bccf3ab JM |
8208 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
8209 | struct btrfs_path *path, | |
8210 | struct btrfs_key *key) | |
0b86a832 | 8211 | { |
6bccf3ab | 8212 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 8213 | int ret = 0; |
0b86a832 CM |
8214 | struct btrfs_key found_key; |
8215 | struct extent_buffer *leaf; | |
514c7dca QW |
8216 | struct btrfs_block_group_item bg; |
8217 | u64 flags; | |
0b86a832 | 8218 | int slot; |
edbd8d4e | 8219 | |
0b86a832 CM |
8220 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
8221 | if (ret < 0) | |
925baedd CM |
8222 | goto out; |
8223 | ||
d397712b | 8224 | while (1) { |
0b86a832 | 8225 | slot = path->slots[0]; |
edbd8d4e | 8226 | leaf = path->nodes[0]; |
0b86a832 CM |
8227 | if (slot >= btrfs_header_nritems(leaf)) { |
8228 | ret = btrfs_next_leaf(root, path); | |
8229 | if (ret == 0) | |
8230 | continue; | |
8231 | if (ret < 0) | |
925baedd | 8232 | goto out; |
0b86a832 | 8233 | break; |
edbd8d4e | 8234 | } |
0b86a832 | 8235 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 8236 | |
0b86a832 | 8237 | if (found_key.objectid >= key->objectid && |
925baedd | 8238 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
8239 | struct extent_map_tree *em_tree; |
8240 | struct extent_map *em; | |
8241 | ||
c8bf1b67 | 8242 | em_tree = &root->fs_info->mapping_tree; |
6fb37b75 LB |
8243 | read_lock(&em_tree->lock); |
8244 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
8245 | found_key.offset); | |
8246 | read_unlock(&em_tree->lock); | |
8247 | if (!em) { | |
0b246afa | 8248 | btrfs_err(fs_info, |
6fb37b75 LB |
8249 | "logical %llu len %llu found bg but no related chunk", |
8250 | found_key.objectid, found_key.offset); | |
8251 | ret = -ENOENT; | |
514c7dca QW |
8252 | } else if (em->start != found_key.objectid || |
8253 | em->len != found_key.offset) { | |
8254 | btrfs_err(fs_info, | |
8255 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
8256 | found_key.objectid, found_key.offset, | |
8257 | em->start, em->len); | |
8258 | ret = -EUCLEAN; | |
6fb37b75 | 8259 | } else { |
514c7dca QW |
8260 | read_extent_buffer(leaf, &bg, |
8261 | btrfs_item_ptr_offset(leaf, slot), | |
8262 | sizeof(bg)); | |
8263 | flags = btrfs_block_group_flags(&bg) & | |
8264 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
8265 | ||
8266 | if (flags != (em->map_lookup->type & | |
8267 | BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
8268 | btrfs_err(fs_info, | |
8269 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
8270 | found_key.objectid, | |
8271 | found_key.offset, flags, | |
8272 | (BTRFS_BLOCK_GROUP_TYPE_MASK & | |
8273 | em->map_lookup->type)); | |
8274 | ret = -EUCLEAN; | |
8275 | } else { | |
8276 | ret = 0; | |
8277 | } | |
6fb37b75 | 8278 | } |
187ee58c | 8279 | free_extent_map(em); |
925baedd CM |
8280 | goto out; |
8281 | } | |
0b86a832 | 8282 | path->slots[0]++; |
edbd8d4e | 8283 | } |
925baedd | 8284 | out: |
0b86a832 | 8285 | return ret; |
edbd8d4e CM |
8286 | } |
8287 | ||
0af3d00b JB |
8288 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
8289 | { | |
8290 | struct btrfs_block_group_cache *block_group; | |
8291 | u64 last = 0; | |
8292 | ||
8293 | while (1) { | |
8294 | struct inode *inode; | |
8295 | ||
8296 | block_group = btrfs_lookup_first_block_group(info, last); | |
8297 | while (block_group) { | |
3aa7c7a3 | 8298 | wait_block_group_cache_done(block_group); |
0af3d00b JB |
8299 | spin_lock(&block_group->lock); |
8300 | if (block_group->iref) | |
8301 | break; | |
8302 | spin_unlock(&block_group->lock); | |
f87b7eb8 | 8303 | block_group = next_block_group(block_group); |
0af3d00b JB |
8304 | } |
8305 | if (!block_group) { | |
8306 | if (last == 0) | |
8307 | break; | |
8308 | last = 0; | |
8309 | continue; | |
8310 | } | |
8311 | ||
8312 | inode = block_group->inode; | |
8313 | block_group->iref = 0; | |
8314 | block_group->inode = NULL; | |
8315 | spin_unlock(&block_group->lock); | |
f3bca802 | 8316 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
8317 | iput(inode); |
8318 | last = block_group->key.objectid + block_group->key.offset; | |
8319 | btrfs_put_block_group(block_group); | |
8320 | } | |
8321 | } | |
8322 | ||
5cdd7db6 FM |
8323 | /* |
8324 | * Must be called only after stopping all workers, since we could have block | |
8325 | * group caching kthreads running, and therefore they could race with us if we | |
8326 | * freed the block groups before stopping them. | |
8327 | */ | |
1a40e23b ZY |
8328 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
8329 | { | |
8330 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 8331 | struct btrfs_space_info *space_info; |
11833d66 | 8332 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
8333 | struct rb_node *n; |
8334 | ||
9e351cc8 | 8335 | down_write(&info->commit_root_sem); |
11833d66 YZ |
8336 | while (!list_empty(&info->caching_block_groups)) { |
8337 | caching_ctl = list_entry(info->caching_block_groups.next, | |
8338 | struct btrfs_caching_control, list); | |
8339 | list_del(&caching_ctl->list); | |
8340 | put_caching_control(caching_ctl); | |
8341 | } | |
9e351cc8 | 8342 | up_write(&info->commit_root_sem); |
11833d66 | 8343 | |
47ab2a6c JB |
8344 | spin_lock(&info->unused_bgs_lock); |
8345 | while (!list_empty(&info->unused_bgs)) { | |
8346 | block_group = list_first_entry(&info->unused_bgs, | |
8347 | struct btrfs_block_group_cache, | |
8348 | bg_list); | |
8349 | list_del_init(&block_group->bg_list); | |
8350 | btrfs_put_block_group(block_group); | |
8351 | } | |
8352 | spin_unlock(&info->unused_bgs_lock); | |
8353 | ||
1a40e23b ZY |
8354 | spin_lock(&info->block_group_cache_lock); |
8355 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
8356 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
8357 | cache_node); | |
1a40e23b ZY |
8358 | rb_erase(&block_group->cache_node, |
8359 | &info->block_group_cache_tree); | |
01eacb27 | 8360 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
8361 | spin_unlock(&info->block_group_cache_lock); |
8362 | ||
80eb234a | 8363 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 8364 | list_del(&block_group->list); |
80eb234a | 8365 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 8366 | |
3c14874a JB |
8367 | /* |
8368 | * We haven't cached this block group, which means we could | |
8369 | * possibly have excluded extents on this block group. | |
8370 | */ | |
36cce922 JB |
8371 | if (block_group->cached == BTRFS_CACHE_NO || |
8372 | block_group->cached == BTRFS_CACHE_ERROR) | |
9e715da8 | 8373 | free_excluded_extents(block_group); |
3c14874a | 8374 | |
817d52f8 | 8375 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 8376 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
8377 | ASSERT(list_empty(&block_group->dirty_list)); |
8378 | ASSERT(list_empty(&block_group->io_list)); | |
8379 | ASSERT(list_empty(&block_group->bg_list)); | |
8380 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 8381 | btrfs_put_block_group(block_group); |
d899e052 YZ |
8382 | |
8383 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
8384 | } |
8385 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
8386 | |
8387 | /* now that all the block groups are freed, go through and | |
8388 | * free all the space_info structs. This is only called during | |
8389 | * the final stages of unmount, and so we know nobody is | |
8390 | * using them. We call synchronize_rcu() once before we start, | |
8391 | * just to be on the safe side. | |
8392 | */ | |
8393 | synchronize_rcu(); | |
8394 | ||
67f9c220 | 8395 | btrfs_release_global_block_rsv(info); |
8929ecfa | 8396 | |
67871254 | 8397 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
8398 | int i; |
8399 | ||
4184ea7f CM |
8400 | space_info = list_entry(info->space_info.next, |
8401 | struct btrfs_space_info, | |
8402 | list); | |
d555b6c3 JB |
8403 | |
8404 | /* | |
8405 | * Do not hide this behind enospc_debug, this is actually | |
8406 | * important and indicates a real bug if this happens. | |
8407 | */ | |
8408 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 8409 | space_info->bytes_reserved > 0 || |
d555b6c3 | 8410 | space_info->bytes_may_use > 0)) |
5da6afeb | 8411 | btrfs_dump_space_info(info, space_info, 0, 0); |
4184ea7f | 8412 | list_del(&space_info->list); |
6ab0a202 JM |
8413 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
8414 | struct kobject *kobj; | |
c1895442 JM |
8415 | kobj = space_info->block_group_kobjs[i]; |
8416 | space_info->block_group_kobjs[i] = NULL; | |
8417 | if (kobj) { | |
6ab0a202 JM |
8418 | kobject_del(kobj); |
8419 | kobject_put(kobj); | |
8420 | } | |
8421 | } | |
8422 | kobject_del(&space_info->kobj); | |
8423 | kobject_put(&space_info->kobj); | |
4184ea7f | 8424 | } |
1a40e23b ZY |
8425 | return 0; |
8426 | } | |
8427 | ||
75cb379d JM |
8428 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
8429 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) | |
8430 | { | |
8431 | struct btrfs_space_info *space_info; | |
8432 | struct raid_kobject *rkobj; | |
8433 | LIST_HEAD(list); | |
75cb379d JM |
8434 | int ret = 0; |
8435 | ||
8436 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
8437 | list_splice_init(&fs_info->pending_raid_kobjs, &list); | |
8438 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
8439 | ||
8440 | list_for_each_entry(rkobj, &list, list) { | |
280c2908 | 8441 | space_info = btrfs_find_space_info(fs_info, rkobj->flags); |
75cb379d JM |
8442 | |
8443 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
158da513 | 8444 | "%s", btrfs_bg_type_to_raid_name(rkobj->flags)); |
75cb379d JM |
8445 | if (ret) { |
8446 | kobject_put(&rkobj->kobj); | |
8447 | break; | |
8448 | } | |
8449 | } | |
8450 | if (ret) | |
8451 | btrfs_warn(fs_info, | |
8452 | "failed to add kobject for block cache, ignoring"); | |
8453 | } | |
8454 | ||
c434d21c | 8455 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 8456 | { |
c434d21c | 8457 | struct btrfs_space_info *space_info = cache->space_info; |
75cb379d | 8458 | struct btrfs_fs_info *fs_info = cache->fs_info; |
3e72ee88 | 8459 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 8460 | bool first = false; |
b742bb82 YZ |
8461 | |
8462 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
8463 | if (list_empty(&space_info->block_groups[index])) |
8464 | first = true; | |
8465 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
8466 | up_write(&space_info->groups_sem); | |
8467 | ||
8468 | if (first) { | |
75cb379d JM |
8469 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
8470 | if (!rkobj) { | |
8471 | btrfs_warn(cache->fs_info, | |
8472 | "couldn't alloc memory for raid level kobject"); | |
8473 | return; | |
6ab0a202 | 8474 | } |
75cb379d JM |
8475 | rkobj->flags = cache->flags; |
8476 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
8477 | ||
8478 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
8479 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); | |
8480 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
c1895442 | 8481 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 8482 | } |
b742bb82 YZ |
8483 | } |
8484 | ||
920e4a58 | 8485 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
8486 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
8487 | u64 start, u64 size) | |
920e4a58 MX |
8488 | { |
8489 | struct btrfs_block_group_cache *cache; | |
8490 | ||
8491 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
8492 | if (!cache) | |
8493 | return NULL; | |
8494 | ||
8495 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
8496 | GFP_NOFS); | |
8497 | if (!cache->free_space_ctl) { | |
8498 | kfree(cache); | |
8499 | return NULL; | |
8500 | } | |
8501 | ||
8502 | cache->key.objectid = start; | |
8503 | cache->key.offset = size; | |
8504 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
8505 | ||
0b246afa | 8506 | cache->fs_info = fs_info; |
e4ff5fb5 | 8507 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
8508 | set_free_space_tree_thresholds(cache); |
8509 | ||
920e4a58 MX |
8510 | atomic_set(&cache->count, 1); |
8511 | spin_lock_init(&cache->lock); | |
e570fd27 | 8512 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
8513 | INIT_LIST_HEAD(&cache->list); |
8514 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 8515 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 8516 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 8517 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 8518 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 8519 | btrfs_init_free_space_ctl(cache); |
04216820 | 8520 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 8521 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 8522 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
8523 | |
8524 | return cache; | |
8525 | } | |
8526 | ||
7ef49515 QW |
8527 | |
8528 | /* | |
8529 | * Iterate all chunks and verify that each of them has the corresponding block | |
8530 | * group | |
8531 | */ | |
8532 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
8533 | { | |
c8bf1b67 | 8534 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
7ef49515 QW |
8535 | struct extent_map *em; |
8536 | struct btrfs_block_group_cache *bg; | |
8537 | u64 start = 0; | |
8538 | int ret = 0; | |
8539 | ||
8540 | while (1) { | |
c8bf1b67 | 8541 | read_lock(&map_tree->lock); |
7ef49515 QW |
8542 | /* |
8543 | * lookup_extent_mapping will return the first extent map | |
8544 | * intersecting the range, so setting @len to 1 is enough to | |
8545 | * get the first chunk. | |
8546 | */ | |
c8bf1b67 DS |
8547 | em = lookup_extent_mapping(map_tree, start, 1); |
8548 | read_unlock(&map_tree->lock); | |
7ef49515 QW |
8549 | if (!em) |
8550 | break; | |
8551 | ||
8552 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
8553 | if (!bg) { | |
8554 | btrfs_err(fs_info, | |
8555 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
8556 | em->start, em->len); | |
8557 | ret = -EUCLEAN; | |
8558 | free_extent_map(em); | |
8559 | break; | |
8560 | } | |
8561 | if (bg->key.objectid != em->start || | |
8562 | bg->key.offset != em->len || | |
8563 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != | |
8564 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
8565 | btrfs_err(fs_info, | |
8566 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
8567 | em->start, em->len, | |
8568 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
8569 | bg->key.objectid, bg->key.offset, | |
8570 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); | |
8571 | ret = -EUCLEAN; | |
8572 | free_extent_map(em); | |
8573 | btrfs_put_block_group(bg); | |
8574 | break; | |
8575 | } | |
8576 | start = em->start + em->len; | |
8577 | free_extent_map(em); | |
8578 | btrfs_put_block_group(bg); | |
8579 | } | |
8580 | return ret; | |
8581 | } | |
8582 | ||
5b4aacef | 8583 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
8584 | { |
8585 | struct btrfs_path *path; | |
8586 | int ret; | |
9078a3e1 | 8587 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 8588 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
8589 | struct btrfs_key key; |
8590 | struct btrfs_key found_key; | |
5f39d397 | 8591 | struct extent_buffer *leaf; |
0af3d00b JB |
8592 | int need_clear = 0; |
8593 | u64 cache_gen; | |
49303381 LB |
8594 | u64 feature; |
8595 | int mixed; | |
8596 | ||
8597 | feature = btrfs_super_incompat_flags(info->super_copy); | |
8598 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 8599 | |
9078a3e1 | 8600 | key.objectid = 0; |
0b86a832 | 8601 | key.offset = 0; |
962a298f | 8602 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
8603 | path = btrfs_alloc_path(); |
8604 | if (!path) | |
8605 | return -ENOMEM; | |
e4058b54 | 8606 | path->reada = READA_FORWARD; |
9078a3e1 | 8607 | |
0b246afa JM |
8608 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
8609 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
8610 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 8611 | need_clear = 1; |
0b246afa | 8612 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 8613 | need_clear = 1; |
0af3d00b | 8614 | |
d397712b | 8615 | while (1) { |
6bccf3ab | 8616 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
8617 | if (ret > 0) |
8618 | break; | |
0b86a832 CM |
8619 | if (ret != 0) |
8620 | goto error; | |
920e4a58 | 8621 | |
5f39d397 CM |
8622 | leaf = path->nodes[0]; |
8623 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 8624 | |
2ff7e61e | 8625 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 8626 | found_key.offset); |
9078a3e1 | 8627 | if (!cache) { |
0b86a832 | 8628 | ret = -ENOMEM; |
f0486c68 | 8629 | goto error; |
9078a3e1 | 8630 | } |
96303081 | 8631 | |
cf7c1ef6 LB |
8632 | if (need_clear) { |
8633 | /* | |
8634 | * When we mount with old space cache, we need to | |
8635 | * set BTRFS_DC_CLEAR and set dirty flag. | |
8636 | * | |
8637 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
8638 | * truncate the old free space cache inode and | |
8639 | * setup a new one. | |
8640 | * b) Setting 'dirty flag' makes sure that we flush | |
8641 | * the new space cache info onto disk. | |
8642 | */ | |
0b246afa | 8643 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 8644 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 8645 | } |
0af3d00b | 8646 | |
5f39d397 CM |
8647 | read_extent_buffer(leaf, &cache->item, |
8648 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
8649 | sizeof(cache->item)); | |
920e4a58 | 8650 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
8651 | if (!mixed && |
8652 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
8653 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
8654 | btrfs_err(info, | |
8655 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
8656 | cache->key.objectid); | |
8657 | ret = -EINVAL; | |
8658 | goto error; | |
8659 | } | |
0b86a832 | 8660 | |
9078a3e1 | 8661 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 8662 | btrfs_release_path(path); |
34d52cb6 | 8663 | |
3c14874a JB |
8664 | /* |
8665 | * We need to exclude the super stripes now so that the space | |
8666 | * info has super bytes accounted for, otherwise we'll think | |
8667 | * we have more space than we actually do. | |
8668 | */ | |
3c4da657 | 8669 | ret = exclude_super_stripes(cache); |
835d974f JB |
8670 | if (ret) { |
8671 | /* | |
8672 | * We may have excluded something, so call this just in | |
8673 | * case. | |
8674 | */ | |
9e715da8 | 8675 | free_excluded_extents(cache); |
920e4a58 | 8676 | btrfs_put_block_group(cache); |
835d974f JB |
8677 | goto error; |
8678 | } | |
3c14874a | 8679 | |
817d52f8 JB |
8680 | /* |
8681 | * check for two cases, either we are full, and therefore | |
8682 | * don't need to bother with the caching work since we won't | |
8683 | * find any space, or we are empty, and we can just add all | |
52042d8e | 8684 | * the space in and be done with it. This saves us _a_lot_ of |
817d52f8 JB |
8685 | * time, particularly in the full case. |
8686 | */ | |
8687 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 8688 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8689 | cache->cached = BTRFS_CACHE_FINISHED; |
9e715da8 | 8690 | free_excluded_extents(cache); |
817d52f8 | 8691 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 8692 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8693 | cache->cached = BTRFS_CACHE_FINISHED; |
4457c1c7 | 8694 | add_new_free_space(cache, found_key.objectid, |
817d52f8 JB |
8695 | found_key.objectid + |
8696 | found_key.offset); | |
9e715da8 | 8697 | free_excluded_extents(cache); |
817d52f8 | 8698 | } |
96b5179d | 8699 | |
0b246afa | 8700 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
8701 | if (ret) { |
8702 | btrfs_remove_free_space_cache(cache); | |
8703 | btrfs_put_block_group(cache); | |
8704 | goto error; | |
8705 | } | |
8706 | ||
0b246afa | 8707 | trace_btrfs_add_block_group(info, cache, 0); |
280c2908 JB |
8708 | btrfs_update_space_info(info, cache->flags, found_key.offset, |
8709 | btrfs_block_group_used(&cache->item), | |
8710 | cache->bytes_super, &space_info); | |
8c579fe7 | 8711 | |
6324fbf3 | 8712 | cache->space_info = space_info; |
1b2da372 | 8713 | |
c434d21c | 8714 | link_block_group(cache); |
0f9dd46c | 8715 | |
0b246afa | 8716 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 8717 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 8718 | inc_block_group_ro(cache, 1); |
47ab2a6c | 8719 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
031f24da QW |
8720 | ASSERT(list_empty(&cache->bg_list)); |
8721 | btrfs_mark_bg_unused(cache); | |
47ab2a6c | 8722 | } |
9078a3e1 | 8723 | } |
b742bb82 | 8724 | |
0b246afa | 8725 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 8726 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 | 8727 | (BTRFS_BLOCK_GROUP_RAID10 | |
c7369b3f | 8728 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 8729 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
b742bb82 YZ |
8730 | BTRFS_BLOCK_GROUP_DUP))) |
8731 | continue; | |
8732 | /* | |
8733 | * avoid allocating from un-mirrored block group if there are | |
8734 | * mirrored block groups. | |
8735 | */ | |
1095cc0d | 8736 | list_for_each_entry(cache, |
8737 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
8738 | list) | |
868f401a | 8739 | inc_block_group_ro(cache, 1); |
1095cc0d | 8740 | list_for_each_entry(cache, |
8741 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
8742 | list) | |
868f401a | 8743 | inc_block_group_ro(cache, 1); |
9078a3e1 | 8744 | } |
f0486c68 | 8745 | |
75cb379d | 8746 | btrfs_add_raid_kobjects(info); |
67f9c220 | 8747 | btrfs_init_global_block_rsv(info); |
7ef49515 | 8748 | ret = check_chunk_block_group_mappings(info); |
0b86a832 | 8749 | error: |
9078a3e1 | 8750 | btrfs_free_path(path); |
0b86a832 | 8751 | return ret; |
9078a3e1 | 8752 | } |
6324fbf3 | 8753 | |
6c686b35 | 8754 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 8755 | { |
6c686b35 | 8756 | struct btrfs_fs_info *fs_info = trans->fs_info; |
545e3366 | 8757 | struct btrfs_block_group_cache *block_group; |
0b246afa | 8758 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
8759 | struct btrfs_block_group_item item; |
8760 | struct btrfs_key key; | |
8761 | int ret = 0; | |
8762 | ||
5ce55557 FM |
8763 | if (!trans->can_flush_pending_bgs) |
8764 | return; | |
8765 | ||
545e3366 JB |
8766 | while (!list_empty(&trans->new_bgs)) { |
8767 | block_group = list_first_entry(&trans->new_bgs, | |
8768 | struct btrfs_block_group_cache, | |
8769 | bg_list); | |
ea658bad | 8770 | if (ret) |
c92f6be3 | 8771 | goto next; |
ea658bad JB |
8772 | |
8773 | spin_lock(&block_group->lock); | |
8774 | memcpy(&item, &block_group->item, sizeof(item)); | |
8775 | memcpy(&key, &block_group->key, sizeof(key)); | |
8776 | spin_unlock(&block_group->lock); | |
8777 | ||
8778 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
8779 | sizeof(item)); | |
8780 | if (ret) | |
66642832 | 8781 | btrfs_abort_transaction(trans, ret); |
97aff912 | 8782 | ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset); |
6df9a95e | 8783 | if (ret) |
66642832 | 8784 | btrfs_abort_transaction(trans, ret); |
e4e0711c | 8785 | add_block_group_free_space(trans, block_group); |
1e144fb8 | 8786 | /* already aborted the transaction if it failed. */ |
c92f6be3 | 8787 | next: |
ba2c4d4e | 8788 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
c92f6be3 | 8789 | list_del_init(&block_group->bg_list); |
ea658bad | 8790 | } |
5ce55557 | 8791 | btrfs_trans_release_chunk_metadata(trans); |
ea658bad JB |
8792 | } |
8793 | ||
e7e02096 | 8794 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, |
0174484d | 8795 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 8796 | { |
e7e02096 | 8797 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 8798 | struct btrfs_block_group_cache *cache; |
0b246afa | 8799 | int ret; |
6324fbf3 | 8800 | |
90787766 | 8801 | btrfs_set_log_full_commit(trans); |
e02119d5 | 8802 | |
2ff7e61e | 8803 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
8804 | if (!cache) |
8805 | return -ENOMEM; | |
34d52cb6 | 8806 | |
6324fbf3 | 8807 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
8808 | btrfs_set_block_group_chunk_objectid(&cache->item, |
8809 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
8810 | btrfs_set_block_group_flags(&cache->item, type); |
8811 | ||
920e4a58 | 8812 | cache->flags = type; |
11833d66 | 8813 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8814 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 8815 | cache->needs_free_space = 1; |
3c4da657 | 8816 | ret = exclude_super_stripes(cache); |
835d974f JB |
8817 | if (ret) { |
8818 | /* | |
8819 | * We may have excluded something, so call this just in | |
8820 | * case. | |
8821 | */ | |
9e715da8 | 8822 | free_excluded_extents(cache); |
920e4a58 | 8823 | btrfs_put_block_group(cache); |
835d974f JB |
8824 | return ret; |
8825 | } | |
96303081 | 8826 | |
4457c1c7 | 8827 | add_new_free_space(cache, chunk_offset, chunk_offset + size); |
817d52f8 | 8828 | |
9e715da8 | 8829 | free_excluded_extents(cache); |
11833d66 | 8830 | |
d0bd4560 | 8831 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 8832 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
8833 | u64 new_bytes_used = size - bytes_used; |
8834 | ||
8835 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 8836 | fragment_free_space(cache); |
d0bd4560 JB |
8837 | } |
8838 | #endif | |
2e6e5183 | 8839 | /* |
2be12ef7 NB |
8840 | * Ensure the corresponding space_info object is created and |
8841 | * assigned to our block group. We want our bg to be added to the rbtree | |
8842 | * with its ->space_info set. | |
2e6e5183 | 8843 | */ |
280c2908 | 8844 | cache->space_info = btrfs_find_space_info(fs_info, cache->flags); |
dc2d3005 | 8845 | ASSERT(cache->space_info); |
2e6e5183 | 8846 | |
0b246afa | 8847 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
8848 | if (ret) { |
8849 | btrfs_remove_free_space_cache(cache); | |
8850 | btrfs_put_block_group(cache); | |
8851 | return ret; | |
8852 | } | |
8853 | ||
2e6e5183 FM |
8854 | /* |
8855 | * Now that our block group has its ->space_info set and is inserted in | |
8856 | * the rbtree, update the space info's counters. | |
8857 | */ | |
0b246afa | 8858 | trace_btrfs_add_block_group(fs_info, cache, 1); |
280c2908 | 8859 | btrfs_update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 8860 | cache->bytes_super, &cache->space_info); |
67f9c220 | 8861 | btrfs_update_global_block_rsv(fs_info); |
1b2da372 | 8862 | |
c434d21c | 8863 | link_block_group(cache); |
6324fbf3 | 8864 | |
47ab2a6c | 8865 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
ba2c4d4e JB |
8866 | trans->delayed_ref_updates++; |
8867 | btrfs_update_delayed_refs_rsv(trans); | |
6324fbf3 | 8868 | |
0b246afa | 8869 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
8870 | return 0; |
8871 | } | |
1a40e23b | 8872 | |
10ea00f5 ID |
8873 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
8874 | { | |
899c81ea ID |
8875 | u64 extra_flags = chunk_to_extended(flags) & |
8876 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 8877 | |
de98ced9 | 8878 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
8879 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
8880 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
8881 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
8882 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
8883 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
8884 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 8885 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
8886 | } |
8887 | ||
6d58a55a DS |
8888 | /* |
8889 | * Clear incompat bits for the following feature(s): | |
8890 | * | |
8891 | * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group | |
8892 | * in the whole filesystem | |
8893 | */ | |
8894 | static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
8895 | { | |
8896 | if (flags & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
8897 | struct list_head *head = &fs_info->space_info; | |
8898 | struct btrfs_space_info *sinfo; | |
8899 | ||
8900 | list_for_each_entry_rcu(sinfo, head, list) { | |
8901 | bool found = false; | |
8902 | ||
8903 | down_read(&sinfo->groups_sem); | |
8904 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5])) | |
8905 | found = true; | |
8906 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6])) | |
8907 | found = true; | |
8908 | up_read(&sinfo->groups_sem); | |
8909 | ||
8910 | if (found) | |
8911 | return; | |
8912 | } | |
8913 | btrfs_clear_fs_incompat(fs_info, RAID56); | |
8914 | } | |
8915 | } | |
8916 | ||
1a40e23b | 8917 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
5a98ec01 | 8918 | u64 group_start, struct extent_map *em) |
1a40e23b | 8919 | { |
5a98ec01 | 8920 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 8921 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
8922 | struct btrfs_path *path; |
8923 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 8924 | struct btrfs_free_cluster *cluster; |
0b246afa | 8925 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 8926 | struct btrfs_key key; |
0af3d00b | 8927 | struct inode *inode; |
c1895442 | 8928 | struct kobject *kobj = NULL; |
1a40e23b | 8929 | int ret; |
10ea00f5 | 8930 | int index; |
89a55897 | 8931 | int factor; |
4f69cb98 | 8932 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 8933 | bool remove_em; |
ba2c4d4e | 8934 | bool remove_rsv = false; |
1a40e23b | 8935 | |
6bccf3ab | 8936 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 8937 | BUG_ON(!block_group); |
c146afad | 8938 | BUG_ON(!block_group->ro); |
1a40e23b | 8939 | |
4ed0a7a3 | 8940 | trace_btrfs_remove_block_group(block_group); |
9f7c43c9 | 8941 | /* |
8942 | * Free the reserved super bytes from this block group before | |
8943 | * remove it. | |
8944 | */ | |
9e715da8 | 8945 | free_excluded_extents(block_group); |
fd708b81 JB |
8946 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
8947 | block_group->key.offset); | |
9f7c43c9 | 8948 | |
1a40e23b | 8949 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 8950 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
46df06b8 | 8951 | factor = btrfs_bg_type_to_factor(block_group->flags); |
1a40e23b | 8952 | |
44fb5511 | 8953 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 8954 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
8955 | spin_lock(&cluster->refill_lock); |
8956 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8957 | spin_unlock(&cluster->refill_lock); | |
8958 | ||
8959 | /* | |
8960 | * make sure this block group isn't part of a metadata | |
8961 | * allocation cluster | |
8962 | */ | |
0b246afa | 8963 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
8964 | spin_lock(&cluster->refill_lock); |
8965 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8966 | spin_unlock(&cluster->refill_lock); | |
8967 | ||
1a40e23b | 8968 | path = btrfs_alloc_path(); |
d8926bb3 MF |
8969 | if (!path) { |
8970 | ret = -ENOMEM; | |
8971 | goto out; | |
8972 | } | |
1a40e23b | 8973 | |
1bbc621e CM |
8974 | /* |
8975 | * get the inode first so any iput calls done for the io_list | |
8976 | * aren't the final iput (no unlinks allowed now) | |
8977 | */ | |
7949f339 | 8978 | inode = lookup_free_space_inode(block_group, path); |
1bbc621e CM |
8979 | |
8980 | mutex_lock(&trans->transaction->cache_write_mutex); | |
8981 | /* | |
52042d8e | 8982 | * Make sure our free space cache IO is done before removing the |
1bbc621e CM |
8983 | * free space inode |
8984 | */ | |
8985 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
8986 | if (!list_empty(&block_group->io_list)) { | |
8987 | list_del_init(&block_group->io_list); | |
8988 | ||
8989 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
8990 | ||
8991 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 8992 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
8993 | btrfs_put_block_group(block_group); |
8994 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
8995 | } | |
8996 | ||
8997 | if (!list_empty(&block_group->dirty_list)) { | |
8998 | list_del_init(&block_group->dirty_list); | |
ba2c4d4e | 8999 | remove_rsv = true; |
1bbc621e CM |
9000 | btrfs_put_block_group(block_group); |
9001 | } | |
9002 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
9003 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
9004 | ||
0af3d00b | 9005 | if (!IS_ERR(inode)) { |
73f2e545 | 9006 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
9007 | if (ret) { |
9008 | btrfs_add_delayed_iput(inode); | |
9009 | goto out; | |
9010 | } | |
0af3d00b JB |
9011 | clear_nlink(inode); |
9012 | /* One for the block groups ref */ | |
9013 | spin_lock(&block_group->lock); | |
9014 | if (block_group->iref) { | |
9015 | block_group->iref = 0; | |
9016 | block_group->inode = NULL; | |
9017 | spin_unlock(&block_group->lock); | |
9018 | iput(inode); | |
9019 | } else { | |
9020 | spin_unlock(&block_group->lock); | |
9021 | } | |
9022 | /* One for our lookup ref */ | |
455757c3 | 9023 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
9024 | } |
9025 | ||
9026 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
9027 | key.offset = block_group->key.objectid; | |
9028 | key.type = 0; | |
9029 | ||
9030 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
9031 | if (ret < 0) | |
9032 | goto out; | |
9033 | if (ret > 0) | |
b3b4aa74 | 9034 | btrfs_release_path(path); |
0af3d00b JB |
9035 | if (ret == 0) { |
9036 | ret = btrfs_del_item(trans, tree_root, path); | |
9037 | if (ret) | |
9038 | goto out; | |
b3b4aa74 | 9039 | btrfs_release_path(path); |
0af3d00b JB |
9040 | } |
9041 | ||
0b246afa | 9042 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 9043 | rb_erase(&block_group->cache_node, |
0b246afa | 9044 | &fs_info->block_group_cache_tree); |
292cbd51 | 9045 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 9046 | |
0b246afa JM |
9047 | if (fs_info->first_logical_byte == block_group->key.objectid) |
9048 | fs_info->first_logical_byte = (u64)-1; | |
9049 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 9050 | |
80eb234a | 9051 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
9052 | /* |
9053 | * we must use list_del_init so people can check to see if they | |
9054 | * are still on the list after taking the semaphore | |
9055 | */ | |
9056 | list_del_init(&block_group->list); | |
6ab0a202 | 9057 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
9058 | kobj = block_group->space_info->block_group_kobjs[index]; |
9059 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 9060 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 9061 | } |
80eb234a | 9062 | up_write(&block_group->space_info->groups_sem); |
6d58a55a | 9063 | clear_incompat_bg_bits(fs_info, block_group->flags); |
c1895442 JM |
9064 | if (kobj) { |
9065 | kobject_del(kobj); | |
9066 | kobject_put(kobj); | |
9067 | } | |
1a40e23b | 9068 | |
4f69cb98 FM |
9069 | if (block_group->has_caching_ctl) |
9070 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 9071 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 9072 | wait_block_group_cache_done(block_group); |
4f69cb98 | 9073 | if (block_group->has_caching_ctl) { |
0b246afa | 9074 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
9075 | if (!caching_ctl) { |
9076 | struct btrfs_caching_control *ctl; | |
9077 | ||
9078 | list_for_each_entry(ctl, | |
0b246afa | 9079 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
9080 | if (ctl->block_group == block_group) { |
9081 | caching_ctl = ctl; | |
1e4f4714 | 9082 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
9083 | break; |
9084 | } | |
9085 | } | |
9086 | if (caching_ctl) | |
9087 | list_del_init(&caching_ctl->list); | |
0b246afa | 9088 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
9089 | if (caching_ctl) { |
9090 | /* Once for the caching bgs list and once for us. */ | |
9091 | put_caching_control(caching_ctl); | |
9092 | put_caching_control(caching_ctl); | |
9093 | } | |
9094 | } | |
817d52f8 | 9095 | |
ce93ec54 | 9096 | spin_lock(&trans->transaction->dirty_bgs_lock); |
9a0ec83d NB |
9097 | WARN_ON(!list_empty(&block_group->dirty_list)); |
9098 | WARN_ON(!list_empty(&block_group->io_list)); | |
ce93ec54 | 9099 | spin_unlock(&trans->transaction->dirty_bgs_lock); |
9a0ec83d | 9100 | |
817d52f8 JB |
9101 | btrfs_remove_free_space_cache(block_group); |
9102 | ||
c146afad | 9103 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 9104 | list_del_init(&block_group->ro_list); |
18d018ad | 9105 | |
0b246afa | 9106 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
9107 | WARN_ON(block_group->space_info->total_bytes |
9108 | < block_group->key.offset); | |
9109 | WARN_ON(block_group->space_info->bytes_readonly | |
9110 | < block_group->key.offset); | |
9111 | WARN_ON(block_group->space_info->disk_total | |
9112 | < block_group->key.offset * factor); | |
9113 | } | |
c146afad YZ |
9114 | block_group->space_info->total_bytes -= block_group->key.offset; |
9115 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 9116 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 9117 | |
c146afad | 9118 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 9119 | |
0af3d00b JB |
9120 | memcpy(&key, &block_group->key, sizeof(key)); |
9121 | ||
34441361 | 9122 | mutex_lock(&fs_info->chunk_mutex); |
04216820 FM |
9123 | spin_lock(&block_group->lock); |
9124 | block_group->removed = 1; | |
9125 | /* | |
9126 | * At this point trimming can't start on this block group, because we | |
9127 | * removed the block group from the tree fs_info->block_group_cache_tree | |
9128 | * so no one can't find it anymore and even if someone already got this | |
9129 | * block group before we removed it from the rbtree, they have already | |
9130 | * incremented block_group->trimming - if they didn't, they won't find | |
9131 | * any free space entries because we already removed them all when we | |
9132 | * called btrfs_remove_free_space_cache(). | |
9133 | * | |
9134 | * And we must not remove the extent map from the fs_info->mapping_tree | |
9135 | * to prevent the same logical address range and physical device space | |
9136 | * ranges from being reused for a new block group. This is because our | |
9137 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
9138 | * completely transactionless, so while it is trimming a range the | |
9139 | * currently running transaction might finish and a new one start, | |
9140 | * allowing for new block groups to be created that can reuse the same | |
9141 | * physical device locations unless we take this special care. | |
e33e17ee JM |
9142 | * |
9143 | * There may also be an implicit trim operation if the file system | |
9144 | * is mounted with -odiscard. The same protections must remain | |
9145 | * in place until the extents have been discarded completely when | |
9146 | * the transaction commit has completed. | |
04216820 FM |
9147 | */ |
9148 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
04216820 | 9149 | spin_unlock(&block_group->lock); |
04216820 | 9150 | |
34441361 | 9151 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 9152 | |
f3f72779 | 9153 | ret = remove_block_group_free_space(trans, block_group); |
1e144fb8 OS |
9154 | if (ret) |
9155 | goto out; | |
9156 | ||
fa9c0d79 CM |
9157 | btrfs_put_block_group(block_group); |
9158 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
9159 | |
9160 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
9161 | if (ret > 0) | |
9162 | ret = -EIO; | |
9163 | if (ret < 0) | |
9164 | goto out; | |
9165 | ||
9166 | ret = btrfs_del_item(trans, root, path); | |
8eaf40c0 FM |
9167 | if (ret) |
9168 | goto out; | |
9169 | ||
9170 | if (remove_em) { | |
9171 | struct extent_map_tree *em_tree; | |
9172 | ||
c8bf1b67 | 9173 | em_tree = &fs_info->mapping_tree; |
8eaf40c0 FM |
9174 | write_lock(&em_tree->lock); |
9175 | remove_extent_mapping(em_tree, em); | |
9176 | write_unlock(&em_tree->lock); | |
9177 | /* once for the tree */ | |
9178 | free_extent_map(em); | |
9179 | } | |
1a40e23b | 9180 | out: |
ba2c4d4e JB |
9181 | if (remove_rsv) |
9182 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1a40e23b ZY |
9183 | btrfs_free_path(path); |
9184 | return ret; | |
9185 | } | |
acce952b | 9186 | |
8eab77ff | 9187 | struct btrfs_trans_handle * |
7fd01182 FM |
9188 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
9189 | const u64 chunk_offset) | |
8eab77ff | 9190 | { |
c8bf1b67 | 9191 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
7fd01182 FM |
9192 | struct extent_map *em; |
9193 | struct map_lookup *map; | |
9194 | unsigned int num_items; | |
9195 | ||
9196 | read_lock(&em_tree->lock); | |
9197 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
9198 | read_unlock(&em_tree->lock); | |
9199 | ASSERT(em && em->start == chunk_offset); | |
9200 | ||
8eab77ff | 9201 | /* |
7fd01182 FM |
9202 | * We need to reserve 3 + N units from the metadata space info in order |
9203 | * to remove a block group (done at btrfs_remove_chunk() and at | |
9204 | * btrfs_remove_block_group()), which are used for: | |
9205 | * | |
8eab77ff FM |
9206 | * 1 unit for adding the free space inode's orphan (located in the tree |
9207 | * of tree roots). | |
7fd01182 FM |
9208 | * 1 unit for deleting the block group item (located in the extent |
9209 | * tree). | |
9210 | * 1 unit for deleting the free space item (located in tree of tree | |
9211 | * roots). | |
9212 | * N units for deleting N device extent items corresponding to each | |
9213 | * stripe (located in the device tree). | |
9214 | * | |
9215 | * In order to remove a block group we also need to reserve units in the | |
9216 | * system space info in order to update the chunk tree (update one or | |
9217 | * more device items and remove one chunk item), but this is done at | |
9218 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 9219 | */ |
95617d69 | 9220 | map = em->map_lookup; |
7fd01182 FM |
9221 | num_items = 3 + map->num_stripes; |
9222 | free_extent_map(em); | |
9223 | ||
8eab77ff | 9224 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 9225 | num_items, 1); |
8eab77ff FM |
9226 | } |
9227 | ||
47ab2a6c JB |
9228 | /* |
9229 | * Process the unused_bgs list and remove any that don't have any allocated | |
9230 | * space inside of them. | |
9231 | */ | |
9232 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
9233 | { | |
9234 | struct btrfs_block_group_cache *block_group; | |
9235 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
9236 | struct btrfs_trans_handle *trans; |
9237 | int ret = 0; | |
9238 | ||
afcdd129 | 9239 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
9240 | return; |
9241 | ||
9242 | spin_lock(&fs_info->unused_bgs_lock); | |
9243 | while (!list_empty(&fs_info->unused_bgs)) { | |
9244 | u64 start, end; | |
e33e17ee | 9245 | int trimming; |
47ab2a6c JB |
9246 | |
9247 | block_group = list_first_entry(&fs_info->unused_bgs, | |
9248 | struct btrfs_block_group_cache, | |
9249 | bg_list); | |
47ab2a6c | 9250 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
9251 | |
9252 | space_info = block_group->space_info; | |
9253 | ||
47ab2a6c JB |
9254 | if (ret || btrfs_mixed_space_info(space_info)) { |
9255 | btrfs_put_block_group(block_group); | |
9256 | continue; | |
9257 | } | |
9258 | spin_unlock(&fs_info->unused_bgs_lock); | |
9259 | ||
d5f2e33b | 9260 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 9261 | |
47ab2a6c JB |
9262 | /* Don't want to race with allocators so take the groups_sem */ |
9263 | down_write(&space_info->groups_sem); | |
9264 | spin_lock(&block_group->lock); | |
43794446 | 9265 | if (block_group->reserved || block_group->pinned || |
47ab2a6c | 9266 | btrfs_block_group_used(&block_group->item) || |
19c4d2f9 | 9267 | block_group->ro || |
aefbe9a6 | 9268 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
9269 | /* |
9270 | * We want to bail if we made new allocations or have | |
9271 | * outstanding allocations in this block group. We do | |
9272 | * the ro check in case balance is currently acting on | |
9273 | * this block group. | |
9274 | */ | |
4ed0a7a3 | 9275 | trace_btrfs_skip_unused_block_group(block_group); |
47ab2a6c JB |
9276 | spin_unlock(&block_group->lock); |
9277 | up_write(&space_info->groups_sem); | |
9278 | goto next; | |
9279 | } | |
9280 | spin_unlock(&block_group->lock); | |
9281 | ||
9282 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 9283 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
9284 | up_write(&space_info->groups_sem); |
9285 | if (ret < 0) { | |
9286 | ret = 0; | |
9287 | goto next; | |
9288 | } | |
9289 | ||
9290 | /* | |
9291 | * Want to do this before we do anything else so we can recover | |
9292 | * properly if we fail to join the transaction. | |
9293 | */ | |
7fd01182 FM |
9294 | trans = btrfs_start_trans_remove_block_group(fs_info, |
9295 | block_group->key.objectid); | |
47ab2a6c | 9296 | if (IS_ERR(trans)) { |
2ff7e61e | 9297 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
9298 | ret = PTR_ERR(trans); |
9299 | goto next; | |
9300 | } | |
9301 | ||
9302 | /* | |
9303 | * We could have pending pinned extents for this block group, | |
9304 | * just delete them, we don't care about them anymore. | |
9305 | */ | |
9306 | start = block_group->key.objectid; | |
9307 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
9308 | /* |
9309 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
9310 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
9311 | * another task might be running finish_extent_commit() for the | |
9312 | * previous transaction N - 1, and have seen a range belonging | |
9313 | * to the block group in freed_extents[] before we were able to | |
9314 | * clear the whole block group range from freed_extents[]. This | |
9315 | * means that task can lookup for the block group after we | |
9316 | * unpinned it from freed_extents[] and removed it, leading to | |
9317 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
9318 | */ | |
9319 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 9320 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 9321 | EXTENT_DIRTY); |
758eb51e | 9322 | if (ret) { |
d4b450cd | 9323 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 9324 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
9325 | goto end_trans; |
9326 | } | |
9327 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 9328 | EXTENT_DIRTY); |
758eb51e | 9329 | if (ret) { |
d4b450cd | 9330 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 9331 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
9332 | goto end_trans; |
9333 | } | |
d4b450cd | 9334 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
9335 | |
9336 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
9337 | spin_lock(&space_info->lock); |
9338 | spin_lock(&block_group->lock); | |
9339 | ||
bb96c4e5 JB |
9340 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, |
9341 | -block_group->pinned); | |
c30666d4 | 9342 | space_info->bytes_readonly += block_group->pinned; |
dec59fa3 EL |
9343 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
9344 | -block_group->pinned, | |
9345 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
47ab2a6c JB |
9346 | block_group->pinned = 0; |
9347 | ||
c30666d4 ZL |
9348 | spin_unlock(&block_group->lock); |
9349 | spin_unlock(&space_info->lock); | |
9350 | ||
e33e17ee | 9351 | /* DISCARD can flip during remount */ |
0b246afa | 9352 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
9353 | |
9354 | /* Implicit trim during transaction commit. */ | |
9355 | if (trimming) | |
9356 | btrfs_get_block_group_trimming(block_group); | |
9357 | ||
47ab2a6c JB |
9358 | /* |
9359 | * Btrfs_remove_chunk will abort the transaction if things go | |
9360 | * horribly wrong. | |
9361 | */ | |
97aff912 | 9362 | ret = btrfs_remove_chunk(trans, block_group->key.objectid); |
e33e17ee JM |
9363 | |
9364 | if (ret) { | |
9365 | if (trimming) | |
9366 | btrfs_put_block_group_trimming(block_group); | |
9367 | goto end_trans; | |
9368 | } | |
9369 | ||
9370 | /* | |
9371 | * If we're not mounted with -odiscard, we can just forget | |
9372 | * about this block group. Otherwise we'll need to wait | |
9373 | * until transaction commit to do the actual discard. | |
9374 | */ | |
9375 | if (trimming) { | |
348a0013 FM |
9376 | spin_lock(&fs_info->unused_bgs_lock); |
9377 | /* | |
9378 | * A concurrent scrub might have added us to the list | |
9379 | * fs_info->unused_bgs, so use a list_move operation | |
9380 | * to add the block group to the deleted_bgs list. | |
9381 | */ | |
e33e17ee JM |
9382 | list_move(&block_group->bg_list, |
9383 | &trans->transaction->deleted_bgs); | |
348a0013 | 9384 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
9385 | btrfs_get_block_group(block_group); |
9386 | } | |
758eb51e | 9387 | end_trans: |
3a45bb20 | 9388 | btrfs_end_transaction(trans); |
47ab2a6c | 9389 | next: |
d5f2e33b | 9390 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
9391 | btrfs_put_block_group(block_group); |
9392 | spin_lock(&fs_info->unused_bgs_lock); | |
9393 | } | |
9394 | spin_unlock(&fs_info->unused_bgs_lock); | |
9395 | } | |
9396 | ||
2ff7e61e JM |
9397 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
9398 | u64 start, u64 end) | |
acce952b | 9399 | { |
2ff7e61e | 9400 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 9401 | } |
9402 | ||
499f377f JM |
9403 | /* |
9404 | * It used to be that old block groups would be left around forever. | |
9405 | * Iterating over them would be enough to trim unused space. Since we | |
9406 | * now automatically remove them, we also need to iterate over unallocated | |
9407 | * space. | |
9408 | * | |
9409 | * We don't want a transaction for this since the discard may take a | |
9410 | * substantial amount of time. We don't require that a transaction be | |
9411 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
9412 | * to ensure that we're not discarding chunks that were released or |
9413 | * allocated in the current transaction. | |
499f377f JM |
9414 | * |
9415 | * Holding the chunks lock will prevent other threads from allocating | |
9416 | * or releasing chunks, but it won't prevent a running transaction | |
9417 | * from committing and releasing the memory that the pending chunks | |
9418 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
9419 | * transaction and hold the commit root sem. We only need to hold |
9420 | * it while performing the free space search since we have already | |
9421 | * held back allocations. | |
499f377f | 9422 | */ |
8103d10b | 9423 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 9424 | { |
8103d10b | 9425 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
9426 | int ret; |
9427 | ||
9428 | *trimmed = 0; | |
9429 | ||
0be88e36 JM |
9430 | /* Discard not supported = nothing to do. */ |
9431 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
9432 | return 0; | |
9433 | ||
52042d8e | 9434 | /* Not writable = nothing to do. */ |
ebbede42 | 9435 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
9436 | return 0; |
9437 | ||
9438 | /* No free space = nothing to do. */ | |
9439 | if (device->total_bytes <= device->bytes_used) | |
9440 | return 0; | |
9441 | ||
9442 | ret = 0; | |
9443 | ||
9444 | while (1) { | |
fb456252 | 9445 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
9446 | u64 bytes; |
9447 | ||
9448 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
9449 | if (ret) | |
fee7acc3 | 9450 | break; |
499f377f | 9451 | |
929be17a NB |
9452 | find_first_clear_extent_bit(&device->alloc_state, start, |
9453 | &start, &end, | |
9454 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 NB |
9455 | |
9456 | /* Ensure we skip the reserved area in the first 1M */ | |
9457 | start = max_t(u64, start, SZ_1M); | |
9458 | ||
929be17a NB |
9459 | /* |
9460 | * If find_first_clear_extent_bit find a range that spans the | |
9461 | * end of the device it will set end to -1, in this case it's up | |
9462 | * to the caller to trim the value to the size of the device. | |
9463 | */ | |
9464 | end = min(end, device->total_bytes - 1); | |
53460a45 | 9465 | |
929be17a | 9466 | len = end - start + 1; |
499f377f | 9467 | |
929be17a NB |
9468 | /* We didn't find any extents */ |
9469 | if (!len) { | |
499f377f | 9470 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 9471 | ret = 0; |
499f377f JM |
9472 | break; |
9473 | } | |
9474 | ||
929be17a NB |
9475 | ret = btrfs_issue_discard(device->bdev, start, len, |
9476 | &bytes); | |
9477 | if (!ret) | |
9478 | set_extent_bits(&device->alloc_state, start, | |
9479 | start + bytes - 1, | |
9480 | CHUNK_TRIMMED); | |
499f377f JM |
9481 | mutex_unlock(&fs_info->chunk_mutex); |
9482 | ||
9483 | if (ret) | |
9484 | break; | |
9485 | ||
9486 | start += len; | |
9487 | *trimmed += bytes; | |
9488 | ||
9489 | if (fatal_signal_pending(current)) { | |
9490 | ret = -ERESTARTSYS; | |
9491 | break; | |
9492 | } | |
9493 | ||
9494 | cond_resched(); | |
9495 | } | |
9496 | ||
9497 | return ret; | |
9498 | } | |
9499 | ||
93bba24d QW |
9500 | /* |
9501 | * Trim the whole filesystem by: | |
9502 | * 1) trimming the free space in each block group | |
9503 | * 2) trimming the unallocated space on each device | |
9504 | * | |
9505 | * This will also continue trimming even if a block group or device encounters | |
9506 | * an error. The return value will be the last error, or 0 if nothing bad | |
9507 | * happens. | |
9508 | */ | |
2ff7e61e | 9509 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 9510 | { |
f7039b1d | 9511 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
9512 | struct btrfs_device *device; |
9513 | struct list_head *devices; | |
f7039b1d LD |
9514 | u64 group_trimmed; |
9515 | u64 start; | |
9516 | u64 end; | |
9517 | u64 trimmed = 0; | |
93bba24d QW |
9518 | u64 bg_failed = 0; |
9519 | u64 dev_failed = 0; | |
9520 | int bg_ret = 0; | |
9521 | int dev_ret = 0; | |
f7039b1d LD |
9522 | int ret = 0; |
9523 | ||
6ba9fc8e | 9524 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
f87b7eb8 | 9525 | for (; cache; cache = next_block_group(cache)) { |
f7039b1d LD |
9526 | if (cache->key.objectid >= (range->start + range->len)) { |
9527 | btrfs_put_block_group(cache); | |
9528 | break; | |
9529 | } | |
9530 | ||
9531 | start = max(range->start, cache->key.objectid); | |
9532 | end = min(range->start + range->len, | |
9533 | cache->key.objectid + cache->key.offset); | |
9534 | ||
9535 | if (end - start >= range->minlen) { | |
9536 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 9537 | ret = cache_block_group(cache, 0); |
1be41b78 | 9538 | if (ret) { |
93bba24d QW |
9539 | bg_failed++; |
9540 | bg_ret = ret; | |
9541 | continue; | |
1be41b78 JB |
9542 | } |
9543 | ret = wait_block_group_cache_done(cache); | |
9544 | if (ret) { | |
93bba24d QW |
9545 | bg_failed++; |
9546 | bg_ret = ret; | |
9547 | continue; | |
1be41b78 | 9548 | } |
f7039b1d LD |
9549 | } |
9550 | ret = btrfs_trim_block_group(cache, | |
9551 | &group_trimmed, | |
9552 | start, | |
9553 | end, | |
9554 | range->minlen); | |
9555 | ||
9556 | trimmed += group_trimmed; | |
9557 | if (ret) { | |
93bba24d QW |
9558 | bg_failed++; |
9559 | bg_ret = ret; | |
9560 | continue; | |
f7039b1d LD |
9561 | } |
9562 | } | |
f7039b1d LD |
9563 | } |
9564 | ||
93bba24d QW |
9565 | if (bg_failed) |
9566 | btrfs_warn(fs_info, | |
9567 | "failed to trim %llu block group(s), last error %d", | |
9568 | bg_failed, bg_ret); | |
0b246afa | 9569 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
9570 | devices = &fs_info->fs_devices->devices; |
9571 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 9572 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
9573 | if (ret) { |
9574 | dev_failed++; | |
9575 | dev_ret = ret; | |
499f377f | 9576 | break; |
93bba24d | 9577 | } |
499f377f JM |
9578 | |
9579 | trimmed += group_trimmed; | |
9580 | } | |
0b246afa | 9581 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 9582 | |
93bba24d QW |
9583 | if (dev_failed) |
9584 | btrfs_warn(fs_info, | |
9585 | "failed to trim %llu device(s), last error %d", | |
9586 | dev_failed, dev_ret); | |
f7039b1d | 9587 | range->len = trimmed; |
93bba24d QW |
9588 | if (bg_ret) |
9589 | return bg_ret; | |
9590 | return dev_ret; | |
f7039b1d | 9591 | } |
8257b2dc MX |
9592 | |
9593 | /* | |
ea14b57f | 9594 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
9595 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
9596 | * data into the page cache through nocow before the subvolume is snapshoted, | |
9597 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 9598 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 9599 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 9600 | */ |
ea14b57f | 9601 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
9602 | { |
9603 | percpu_counter_dec(&root->subv_writers->counter); | |
093258e6 | 9604 | cond_wake_up(&root->subv_writers->wait); |
8257b2dc MX |
9605 | } |
9606 | ||
ea14b57f | 9607 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 9608 | { |
ea14b57f | 9609 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
9610 | return 0; |
9611 | ||
9612 | percpu_counter_inc(&root->subv_writers->counter); | |
9613 | /* | |
9614 | * Make sure counter is updated before we check for snapshot creation. | |
9615 | */ | |
9616 | smp_mb(); | |
ea14b57f DS |
9617 | if (atomic_read(&root->will_be_snapshotted)) { |
9618 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
9619 | return 0; |
9620 | } | |
9621 | return 1; | |
9622 | } | |
0bc19f90 | 9623 | |
0bc19f90 ZL |
9624 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
9625 | { | |
9626 | while (true) { | |
9627 | int ret; | |
9628 | ||
ea14b57f | 9629 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
9630 | if (ret) |
9631 | break; | |
4625956a PZ |
9632 | wait_var_event(&root->will_be_snapshotted, |
9633 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
9634 | } |
9635 | } | |
031f24da QW |
9636 | |
9637 | void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg) | |
9638 | { | |
9639 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
9640 | ||
9641 | spin_lock(&fs_info->unused_bgs_lock); | |
9642 | if (list_empty(&bg->bg_list)) { | |
9643 | btrfs_get_block_group(bg); | |
9644 | trace_btrfs_add_unused_block_group(bg); | |
9645 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
9646 | } | |
9647 | spin_unlock(&fs_info->unused_bgs_lock); | |
9648 | } |