]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 | 4 | */ |
c1d7c514 | 5 | |
ec6b910f | 6 | #include <linux/sched.h> |
f361bf4a | 7 | #include <linux/sched/signal.h> |
edbd8d4e | 8 | #include <linux/pagemap.h> |
ec44a35c | 9 | #include <linux/writeback.h> |
21af804c | 10 | #include <linux/blkdev.h> |
b7a9f29f | 11 | #include <linux/sort.h> |
4184ea7f | 12 | #include <linux/rcupdate.h> |
817d52f8 | 13 | #include <linux/kthread.h> |
5a0e3ad6 | 14 | #include <linux/slab.h> |
dff51cd1 | 15 | #include <linux/ratelimit.h> |
b150a4f1 | 16 | #include <linux/percpu_counter.h> |
69fe2d75 | 17 | #include <linux/lockdep.h> |
9678c543 | 18 | #include <linux/crc32c.h> |
995946dd | 19 | #include "tree-log.h" |
fec577fb CM |
20 | #include "disk-io.h" |
21 | #include "print-tree.h" | |
0b86a832 | 22 | #include "volumes.h" |
53b381b3 | 23 | #include "raid56.h" |
925baedd | 24 | #include "locking.h" |
fa9c0d79 | 25 | #include "free-space-cache.h" |
1e144fb8 | 26 | #include "free-space-tree.h" |
3fed40cc | 27 | #include "math.h" |
6ab0a202 | 28 | #include "sysfs.h" |
fcebe456 | 29 | #include "qgroup.h" |
fd708b81 | 30 | #include "ref-verify.h" |
fec577fb | 31 | |
709c0486 AJ |
32 | #undef SCRAMBLE_DELAYED_REFS |
33 | ||
9e622d6b MX |
34 | /* |
35 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
36 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
37 | * if we really need one. | |
38 | * | |
0e4f8f88 CM |
39 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
40 | * if we have very few chunks already allocated. This is | |
41 | * used as part of the clustering code to help make sure | |
42 | * we have a good pool of storage to cluster in, without | |
43 | * filling the FS with empty chunks | |
44 | * | |
9e622d6b MX |
45 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
46 | * | |
0e4f8f88 CM |
47 | */ |
48 | enum { | |
49 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
50 | CHUNK_ALLOC_LIMITED = 1, |
51 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
52 | }; |
53 | ||
9f9b8e8d QW |
54 | /* |
55 | * Declare a helper function to detect underflow of various space info members | |
56 | */ | |
57 | #define DECLARE_SPACE_INFO_UPDATE(name) \ | |
58 | static inline void update_##name(struct btrfs_space_info *sinfo, \ | |
59 | s64 bytes) \ | |
60 | { \ | |
61 | if (bytes < 0 && sinfo->name < -bytes) { \ | |
62 | WARN_ON(1); \ | |
63 | sinfo->name = 0; \ | |
64 | return; \ | |
65 | } \ | |
66 | sinfo->name += bytes; \ | |
67 | } | |
68 | ||
69 | DECLARE_SPACE_INFO_UPDATE(bytes_may_use); | |
e2907c1a | 70 | DECLARE_SPACE_INFO_UPDATE(bytes_pinned); |
9f9b8e8d | 71 | |
5d4f98a2 | 72 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
73 | struct btrfs_delayed_ref_node *node, u64 parent, |
74 | u64 root_objectid, u64 owner_objectid, | |
75 | u64 owner_offset, int refs_to_drop, | |
76 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
77 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
78 | struct extent_buffer *leaf, | |
79 | struct btrfs_extent_item *ei); | |
80 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
81 | u64 parent, u64 root_objectid, |
82 | u64 flags, u64 owner, u64 offset, | |
83 | struct btrfs_key *ins, int ref_mod); | |
84 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 85 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 86 | struct btrfs_delayed_extent_op *extent_op); |
01458828 | 87 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
698d0082 | 88 | int force); |
11833d66 YZ |
89 | static int find_next_key(struct btrfs_path *path, int level, |
90 | struct btrfs_key *key); | |
ab8d0fc4 JM |
91 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
92 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 93 | int dump_block_groups); |
5d80366e JB |
94 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
95 | u64 num_bytes); | |
957780eb JB |
96 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
97 | struct btrfs_space_info *space_info, | |
98 | u64 num_bytes); | |
99 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
100 | struct btrfs_space_info *space_info, | |
101 | u64 num_bytes); | |
6a63209f | 102 | |
817d52f8 JB |
103 | static noinline int |
104 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
105 | { | |
106 | smp_mb(); | |
36cce922 JB |
107 | return cache->cached == BTRFS_CACHE_FINISHED || |
108 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
109 | } |
110 | ||
0f9dd46c JB |
111 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
112 | { | |
113 | return (cache->flags & bits) == bits; | |
114 | } | |
115 | ||
758f2dfc | 116 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
117 | { |
118 | atomic_inc(&cache->count); | |
119 | } | |
120 | ||
121 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
122 | { | |
f0486c68 YZ |
123 | if (atomic_dec_and_test(&cache->count)) { |
124 | WARN_ON(cache->pinned > 0); | |
125 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
126 | |
127 | /* | |
128 | * If not empty, someone is still holding mutex of | |
129 | * full_stripe_lock, which can only be released by caller. | |
130 | * And it will definitely cause use-after-free when caller | |
131 | * tries to release full stripe lock. | |
132 | * | |
133 | * No better way to resolve, but only to warn. | |
134 | */ | |
135 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 136 | kfree(cache->free_space_ctl); |
11dfe35a | 137 | kfree(cache); |
f0486c68 | 138 | } |
11dfe35a JB |
139 | } |
140 | ||
0f9dd46c JB |
141 | /* |
142 | * this adds the block group to the fs_info rb tree for the block group | |
143 | * cache | |
144 | */ | |
b2950863 | 145 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
146 | struct btrfs_block_group_cache *block_group) |
147 | { | |
148 | struct rb_node **p; | |
149 | struct rb_node *parent = NULL; | |
150 | struct btrfs_block_group_cache *cache; | |
151 | ||
152 | spin_lock(&info->block_group_cache_lock); | |
153 | p = &info->block_group_cache_tree.rb_node; | |
154 | ||
155 | while (*p) { | |
156 | parent = *p; | |
157 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
158 | cache_node); | |
159 | if (block_group->key.objectid < cache->key.objectid) { | |
160 | p = &(*p)->rb_left; | |
161 | } else if (block_group->key.objectid > cache->key.objectid) { | |
162 | p = &(*p)->rb_right; | |
163 | } else { | |
164 | spin_unlock(&info->block_group_cache_lock); | |
165 | return -EEXIST; | |
166 | } | |
167 | } | |
168 | ||
169 | rb_link_node(&block_group->cache_node, parent, p); | |
170 | rb_insert_color(&block_group->cache_node, | |
171 | &info->block_group_cache_tree); | |
a1897fdd LB |
172 | |
173 | if (info->first_logical_byte > block_group->key.objectid) | |
174 | info->first_logical_byte = block_group->key.objectid; | |
175 | ||
0f9dd46c JB |
176 | spin_unlock(&info->block_group_cache_lock); |
177 | ||
178 | return 0; | |
179 | } | |
180 | ||
181 | /* | |
182 | * This will return the block group at or after bytenr if contains is 0, else | |
183 | * it will return the block group that contains the bytenr | |
184 | */ | |
185 | static struct btrfs_block_group_cache * | |
186 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
187 | int contains) | |
188 | { | |
189 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
190 | struct rb_node *n; | |
191 | u64 end, start; | |
192 | ||
193 | spin_lock(&info->block_group_cache_lock); | |
194 | n = info->block_group_cache_tree.rb_node; | |
195 | ||
196 | while (n) { | |
197 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
198 | cache_node); | |
199 | end = cache->key.objectid + cache->key.offset - 1; | |
200 | start = cache->key.objectid; | |
201 | ||
202 | if (bytenr < start) { | |
203 | if (!contains && (!ret || start < ret->key.objectid)) | |
204 | ret = cache; | |
205 | n = n->rb_left; | |
206 | } else if (bytenr > start) { | |
207 | if (contains && bytenr <= end) { | |
208 | ret = cache; | |
209 | break; | |
210 | } | |
211 | n = n->rb_right; | |
212 | } else { | |
213 | ret = cache; | |
214 | break; | |
215 | } | |
216 | } | |
a1897fdd | 217 | if (ret) { |
11dfe35a | 218 | btrfs_get_block_group(ret); |
a1897fdd LB |
219 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
220 | info->first_logical_byte = ret->key.objectid; | |
221 | } | |
0f9dd46c JB |
222 | spin_unlock(&info->block_group_cache_lock); |
223 | ||
224 | return ret; | |
225 | } | |
226 | ||
2ff7e61e | 227 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 228 | u64 start, u64 num_bytes) |
817d52f8 | 229 | { |
11833d66 | 230 | u64 end = start + num_bytes - 1; |
0b246afa | 231 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 232 | start, end, EXTENT_UPTODATE); |
0b246afa | 233 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 234 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
235 | return 0; |
236 | } | |
817d52f8 | 237 | |
9e715da8 | 238 | static void free_excluded_extents(struct btrfs_block_group_cache *cache) |
11833d66 | 239 | { |
9e715da8 | 240 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 241 | u64 start, end; |
817d52f8 | 242 | |
11833d66 YZ |
243 | start = cache->key.objectid; |
244 | end = start + cache->key.offset - 1; | |
245 | ||
0b246afa | 246 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 247 | start, end, EXTENT_UPTODATE); |
0b246afa | 248 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 249 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
250 | } |
251 | ||
3c4da657 | 252 | static int exclude_super_stripes(struct btrfs_block_group_cache *cache) |
817d52f8 | 253 | { |
3c4da657 | 254 | struct btrfs_fs_info *fs_info = cache->fs_info; |
817d52f8 JB |
255 | u64 bytenr; |
256 | u64 *logical; | |
257 | int stripe_len; | |
258 | int i, nr, ret; | |
259 | ||
06b2331f YZ |
260 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
261 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
262 | cache->bytes_super += stripe_len; | |
2ff7e61e | 263 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 264 | stripe_len); |
835d974f JB |
265 | if (ret) |
266 | return ret; | |
06b2331f YZ |
267 | } |
268 | ||
817d52f8 JB |
269 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
270 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 271 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
63a9c7b9 | 272 | bytenr, &logical, &nr, &stripe_len); |
835d974f JB |
273 | if (ret) |
274 | return ret; | |
11833d66 | 275 | |
817d52f8 | 276 | while (nr--) { |
51bf5f0b JB |
277 | u64 start, len; |
278 | ||
279 | if (logical[nr] > cache->key.objectid + | |
280 | cache->key.offset) | |
281 | continue; | |
282 | ||
283 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
284 | continue; | |
285 | ||
286 | start = logical[nr]; | |
287 | if (start < cache->key.objectid) { | |
288 | start = cache->key.objectid; | |
289 | len = (logical[nr] + stripe_len) - start; | |
290 | } else { | |
291 | len = min_t(u64, stripe_len, | |
292 | cache->key.objectid + | |
293 | cache->key.offset - start); | |
294 | } | |
295 | ||
296 | cache->bytes_super += len; | |
2ff7e61e | 297 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
298 | if (ret) { |
299 | kfree(logical); | |
300 | return ret; | |
301 | } | |
817d52f8 | 302 | } |
11833d66 | 303 | |
817d52f8 JB |
304 | kfree(logical); |
305 | } | |
817d52f8 JB |
306 | return 0; |
307 | } | |
308 | ||
11833d66 YZ |
309 | static struct btrfs_caching_control * |
310 | get_caching_control(struct btrfs_block_group_cache *cache) | |
311 | { | |
312 | struct btrfs_caching_control *ctl; | |
313 | ||
314 | spin_lock(&cache->lock); | |
dde5abee JB |
315 | if (!cache->caching_ctl) { |
316 | spin_unlock(&cache->lock); | |
11833d66 YZ |
317 | return NULL; |
318 | } | |
319 | ||
320 | ctl = cache->caching_ctl; | |
1e4f4714 | 321 | refcount_inc(&ctl->count); |
11833d66 YZ |
322 | spin_unlock(&cache->lock); |
323 | return ctl; | |
324 | } | |
325 | ||
326 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
327 | { | |
1e4f4714 | 328 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
329 | kfree(ctl); |
330 | } | |
331 | ||
d0bd4560 | 332 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 333 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 334 | { |
2ff7e61e | 335 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
336 | u64 start = block_group->key.objectid; |
337 | u64 len = block_group->key.offset; | |
338 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 339 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
340 | u64 step = chunk << 1; |
341 | ||
342 | while (len > chunk) { | |
343 | btrfs_remove_free_space(block_group, start, chunk); | |
344 | start += step; | |
345 | if (len < step) | |
346 | len = 0; | |
347 | else | |
348 | len -= step; | |
349 | } | |
350 | } | |
351 | #endif | |
352 | ||
0f9dd46c JB |
353 | /* |
354 | * this is only called by cache_block_group, since we could have freed extents | |
355 | * we need to check the pinned_extents for any extents that can't be used yet | |
356 | * since their free space will be released as soon as the transaction commits. | |
357 | */ | |
a5ed9182 | 358 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
4457c1c7 | 359 | u64 start, u64 end) |
0f9dd46c | 360 | { |
4457c1c7 | 361 | struct btrfs_fs_info *info = block_group->fs_info; |
817d52f8 | 362 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
363 | int ret; |
364 | ||
365 | while (start < end) { | |
11833d66 | 366 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 367 | &extent_start, &extent_end, |
e6138876 JB |
368 | EXTENT_DIRTY | EXTENT_UPTODATE, |
369 | NULL); | |
0f9dd46c JB |
370 | if (ret) |
371 | break; | |
372 | ||
06b2331f | 373 | if (extent_start <= start) { |
0f9dd46c JB |
374 | start = extent_end + 1; |
375 | } else if (extent_start > start && extent_start < end) { | |
376 | size = extent_start - start; | |
817d52f8 | 377 | total_added += size; |
ea6a478e JB |
378 | ret = btrfs_add_free_space(block_group, start, |
379 | size); | |
79787eaa | 380 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
381 | start = extent_end + 1; |
382 | } else { | |
383 | break; | |
384 | } | |
385 | } | |
386 | ||
387 | if (start < end) { | |
388 | size = end - start; | |
817d52f8 | 389 | total_added += size; |
ea6a478e | 390 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 391 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
392 | } |
393 | ||
817d52f8 | 394 | return total_added; |
0f9dd46c JB |
395 | } |
396 | ||
73fa48b6 | 397 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 398 | { |
0b246afa JM |
399 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
400 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
401 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 402 | struct btrfs_path *path; |
5f39d397 | 403 | struct extent_buffer *leaf; |
11833d66 | 404 | struct btrfs_key key; |
817d52f8 | 405 | u64 total_found = 0; |
11833d66 YZ |
406 | u64 last = 0; |
407 | u32 nritems; | |
73fa48b6 | 408 | int ret; |
d0bd4560 | 409 | bool wakeup = true; |
f510cfec | 410 | |
e37c9e69 CM |
411 | path = btrfs_alloc_path(); |
412 | if (!path) | |
73fa48b6 | 413 | return -ENOMEM; |
7d7d6068 | 414 | |
817d52f8 | 415 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 416 | |
d0bd4560 JB |
417 | #ifdef CONFIG_BTRFS_DEBUG |
418 | /* | |
419 | * If we're fragmenting we don't want to make anybody think we can | |
420 | * allocate from this block group until we've had a chance to fragment | |
421 | * the free space. | |
422 | */ | |
2ff7e61e | 423 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
424 | wakeup = false; |
425 | #endif | |
5cd57b2c | 426 | /* |
817d52f8 JB |
427 | * We don't want to deadlock with somebody trying to allocate a new |
428 | * extent for the extent root while also trying to search the extent | |
429 | * root to add free space. So we skip locking and search the commit | |
430 | * root, since its read-only | |
5cd57b2c CM |
431 | */ |
432 | path->skip_locking = 1; | |
817d52f8 | 433 | path->search_commit_root = 1; |
e4058b54 | 434 | path->reada = READA_FORWARD; |
817d52f8 | 435 | |
e4404d6e | 436 | key.objectid = last; |
e37c9e69 | 437 | key.offset = 0; |
11833d66 | 438 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 439 | |
52ee28d2 | 440 | next: |
11833d66 | 441 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 442 | if (ret < 0) |
73fa48b6 | 443 | goto out; |
a512bbf8 | 444 | |
11833d66 YZ |
445 | leaf = path->nodes[0]; |
446 | nritems = btrfs_header_nritems(leaf); | |
447 | ||
d397712b | 448 | while (1) { |
7841cb28 | 449 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 450 | last = (u64)-1; |
817d52f8 | 451 | break; |
f25784b3 | 452 | } |
817d52f8 | 453 | |
11833d66 YZ |
454 | if (path->slots[0] < nritems) { |
455 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
456 | } else { | |
457 | ret = find_next_key(path, 0, &key); | |
458 | if (ret) | |
e37c9e69 | 459 | break; |
817d52f8 | 460 | |
c9ea7b24 | 461 | if (need_resched() || |
9e351cc8 | 462 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
463 | if (wakeup) |
464 | caching_ctl->progress = last; | |
ff5714cc | 465 | btrfs_release_path(path); |
9e351cc8 | 466 | up_read(&fs_info->commit_root_sem); |
589d8ade | 467 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 468 | cond_resched(); |
73fa48b6 OS |
469 | mutex_lock(&caching_ctl->mutex); |
470 | down_read(&fs_info->commit_root_sem); | |
471 | goto next; | |
589d8ade | 472 | } |
0a3896d0 JB |
473 | |
474 | ret = btrfs_next_leaf(extent_root, path); | |
475 | if (ret < 0) | |
73fa48b6 | 476 | goto out; |
0a3896d0 JB |
477 | if (ret) |
478 | break; | |
589d8ade JB |
479 | leaf = path->nodes[0]; |
480 | nritems = btrfs_header_nritems(leaf); | |
481 | continue; | |
11833d66 | 482 | } |
817d52f8 | 483 | |
52ee28d2 LB |
484 | if (key.objectid < last) { |
485 | key.objectid = last; | |
486 | key.offset = 0; | |
487 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
488 | ||
d0bd4560 JB |
489 | if (wakeup) |
490 | caching_ctl->progress = last; | |
52ee28d2 LB |
491 | btrfs_release_path(path); |
492 | goto next; | |
493 | } | |
494 | ||
11833d66 YZ |
495 | if (key.objectid < block_group->key.objectid) { |
496 | path->slots[0]++; | |
817d52f8 | 497 | continue; |
e37c9e69 | 498 | } |
0f9dd46c | 499 | |
e37c9e69 | 500 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 501 | block_group->key.offset) |
e37c9e69 | 502 | break; |
7d7d6068 | 503 | |
3173a18f JB |
504 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
505 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
4457c1c7 | 506 | total_found += add_new_free_space(block_group, last, |
817d52f8 | 507 | key.objectid); |
3173a18f JB |
508 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
509 | last = key.objectid + | |
da17066c | 510 | fs_info->nodesize; |
3173a18f JB |
511 | else |
512 | last = key.objectid + key.offset; | |
817d52f8 | 513 | |
73fa48b6 | 514 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 515 | total_found = 0; |
d0bd4560 JB |
516 | if (wakeup) |
517 | wake_up(&caching_ctl->wait); | |
11833d66 | 518 | } |
817d52f8 | 519 | } |
e37c9e69 CM |
520 | path->slots[0]++; |
521 | } | |
817d52f8 | 522 | ret = 0; |
e37c9e69 | 523 | |
4457c1c7 | 524 | total_found += add_new_free_space(block_group, last, |
817d52f8 JB |
525 | block_group->key.objectid + |
526 | block_group->key.offset); | |
11833d66 | 527 | caching_ctl->progress = (u64)-1; |
817d52f8 | 528 | |
73fa48b6 OS |
529 | out: |
530 | btrfs_free_path(path); | |
531 | return ret; | |
532 | } | |
533 | ||
534 | static noinline void caching_thread(struct btrfs_work *work) | |
535 | { | |
536 | struct btrfs_block_group_cache *block_group; | |
537 | struct btrfs_fs_info *fs_info; | |
538 | struct btrfs_caching_control *caching_ctl; | |
539 | int ret; | |
540 | ||
541 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
542 | block_group = caching_ctl->block_group; | |
543 | fs_info = block_group->fs_info; | |
544 | ||
545 | mutex_lock(&caching_ctl->mutex); | |
546 | down_read(&fs_info->commit_root_sem); | |
547 | ||
1e144fb8 OS |
548 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
549 | ret = load_free_space_tree(caching_ctl); | |
550 | else | |
551 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 552 | |
817d52f8 | 553 | spin_lock(&block_group->lock); |
11833d66 | 554 | block_group->caching_ctl = NULL; |
73fa48b6 | 555 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 556 | spin_unlock(&block_group->lock); |
0f9dd46c | 557 | |
d0bd4560 | 558 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 559 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
560 | u64 bytes_used; |
561 | ||
562 | spin_lock(&block_group->space_info->lock); | |
563 | spin_lock(&block_group->lock); | |
564 | bytes_used = block_group->key.offset - | |
565 | btrfs_block_group_used(&block_group->item); | |
566 | block_group->space_info->bytes_used += bytes_used >> 1; | |
567 | spin_unlock(&block_group->lock); | |
568 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 569 | fragment_free_space(block_group); |
d0bd4560 JB |
570 | } |
571 | #endif | |
572 | ||
573 | caching_ctl->progress = (u64)-1; | |
11833d66 | 574 | |
9e351cc8 | 575 | up_read(&fs_info->commit_root_sem); |
9e715da8 | 576 | free_excluded_extents(block_group); |
11833d66 | 577 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 578 | |
11833d66 YZ |
579 | wake_up(&caching_ctl->wait); |
580 | ||
581 | put_caching_control(caching_ctl); | |
11dfe35a | 582 | btrfs_put_block_group(block_group); |
817d52f8 JB |
583 | } |
584 | ||
9d66e233 | 585 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 586 | int load_cache_only) |
817d52f8 | 587 | { |
291c7d2f | 588 | DEFINE_WAIT(wait); |
11833d66 YZ |
589 | struct btrfs_fs_info *fs_info = cache->fs_info; |
590 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
591 | int ret = 0; |
592 | ||
291c7d2f | 593 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
594 | if (!caching_ctl) |
595 | return -ENOMEM; | |
291c7d2f JB |
596 | |
597 | INIT_LIST_HEAD(&caching_ctl->list); | |
598 | mutex_init(&caching_ctl->mutex); | |
599 | init_waitqueue_head(&caching_ctl->wait); | |
600 | caching_ctl->block_group = cache; | |
601 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 602 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
603 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
604 | caching_thread, NULL, NULL); | |
291c7d2f JB |
605 | |
606 | spin_lock(&cache->lock); | |
607 | /* | |
608 | * This should be a rare occasion, but this could happen I think in the | |
609 | * case where one thread starts to load the space cache info, and then | |
610 | * some other thread starts a transaction commit which tries to do an | |
611 | * allocation while the other thread is still loading the space cache | |
612 | * info. The previous loop should have kept us from choosing this block | |
613 | * group, but if we've moved to the state where we will wait on caching | |
614 | * block groups we need to first check if we're doing a fast load here, | |
615 | * so we can wait for it to finish, otherwise we could end up allocating | |
616 | * from a block group who's cache gets evicted for one reason or | |
617 | * another. | |
618 | */ | |
619 | while (cache->cached == BTRFS_CACHE_FAST) { | |
620 | struct btrfs_caching_control *ctl; | |
621 | ||
622 | ctl = cache->caching_ctl; | |
1e4f4714 | 623 | refcount_inc(&ctl->count); |
291c7d2f JB |
624 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
625 | spin_unlock(&cache->lock); | |
626 | ||
627 | schedule(); | |
628 | ||
629 | finish_wait(&ctl->wait, &wait); | |
630 | put_caching_control(ctl); | |
631 | spin_lock(&cache->lock); | |
632 | } | |
633 | ||
634 | if (cache->cached != BTRFS_CACHE_NO) { | |
635 | spin_unlock(&cache->lock); | |
636 | kfree(caching_ctl); | |
11833d66 | 637 | return 0; |
291c7d2f JB |
638 | } |
639 | WARN_ON(cache->caching_ctl); | |
640 | cache->caching_ctl = caching_ctl; | |
641 | cache->cached = BTRFS_CACHE_FAST; | |
642 | spin_unlock(&cache->lock); | |
11833d66 | 643 | |
d8953d69 | 644 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cb83b7b8 | 645 | mutex_lock(&caching_ctl->mutex); |
bb6cb1c5 | 646 | ret = load_free_space_cache(cache); |
9d66e233 JB |
647 | |
648 | spin_lock(&cache->lock); | |
649 | if (ret == 1) { | |
291c7d2f | 650 | cache->caching_ctl = NULL; |
9d66e233 JB |
651 | cache->cached = BTRFS_CACHE_FINISHED; |
652 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 653 | caching_ctl->progress = (u64)-1; |
9d66e233 | 654 | } else { |
291c7d2f JB |
655 | if (load_cache_only) { |
656 | cache->caching_ctl = NULL; | |
657 | cache->cached = BTRFS_CACHE_NO; | |
658 | } else { | |
659 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 660 | cache->has_caching_ctl = 1; |
291c7d2f | 661 | } |
9d66e233 JB |
662 | } |
663 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
664 | #ifdef CONFIG_BTRFS_DEBUG |
665 | if (ret == 1 && | |
2ff7e61e | 666 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
667 | u64 bytes_used; |
668 | ||
669 | spin_lock(&cache->space_info->lock); | |
670 | spin_lock(&cache->lock); | |
671 | bytes_used = cache->key.offset - | |
672 | btrfs_block_group_used(&cache->item); | |
673 | cache->space_info->bytes_used += bytes_used >> 1; | |
674 | spin_unlock(&cache->lock); | |
675 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 676 | fragment_free_space(cache); |
d0bd4560 JB |
677 | } |
678 | #endif | |
cb83b7b8 JB |
679 | mutex_unlock(&caching_ctl->mutex); |
680 | ||
291c7d2f | 681 | wake_up(&caching_ctl->wait); |
3c14874a | 682 | if (ret == 1) { |
291c7d2f | 683 | put_caching_control(caching_ctl); |
9e715da8 | 684 | free_excluded_extents(cache); |
9d66e233 | 685 | return 0; |
3c14874a | 686 | } |
291c7d2f JB |
687 | } else { |
688 | /* | |
1e144fb8 OS |
689 | * We're either using the free space tree or no caching at all. |
690 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
691 | */ |
692 | spin_lock(&cache->lock); | |
693 | if (load_cache_only) { | |
694 | cache->caching_ctl = NULL; | |
695 | cache->cached = BTRFS_CACHE_NO; | |
696 | } else { | |
697 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 698 | cache->has_caching_ctl = 1; |
291c7d2f JB |
699 | } |
700 | spin_unlock(&cache->lock); | |
701 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
702 | } |
703 | ||
291c7d2f JB |
704 | if (load_cache_only) { |
705 | put_caching_control(caching_ctl); | |
11833d66 | 706 | return 0; |
817d52f8 | 707 | } |
817d52f8 | 708 | |
9e351cc8 | 709 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 710 | refcount_inc(&caching_ctl->count); |
11833d66 | 711 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 712 | up_write(&fs_info->commit_root_sem); |
11833d66 | 713 | |
11dfe35a | 714 | btrfs_get_block_group(cache); |
11833d66 | 715 | |
e66f0bb1 | 716 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 717 | |
ef8bbdfe | 718 | return ret; |
e37c9e69 CM |
719 | } |
720 | ||
0f9dd46c JB |
721 | /* |
722 | * return the block group that starts at or after bytenr | |
723 | */ | |
d397712b CM |
724 | static struct btrfs_block_group_cache * |
725 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 726 | { |
e2c89907 | 727 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
728 | } |
729 | ||
0f9dd46c | 730 | /* |
9f55684c | 731 | * return the block group that contains the given bytenr |
0f9dd46c | 732 | */ |
d397712b CM |
733 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
734 | struct btrfs_fs_info *info, | |
735 | u64 bytenr) | |
be744175 | 736 | { |
e2c89907 | 737 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 738 | } |
0b86a832 | 739 | |
0f9dd46c JB |
740 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
741 | u64 flags) | |
6324fbf3 | 742 | { |
0f9dd46c | 743 | struct list_head *head = &info->space_info; |
0f9dd46c | 744 | struct btrfs_space_info *found; |
4184ea7f | 745 | |
52ba6929 | 746 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 747 | |
4184ea7f CM |
748 | rcu_read_lock(); |
749 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 750 | if (found->flags & flags) { |
4184ea7f | 751 | rcu_read_unlock(); |
0f9dd46c | 752 | return found; |
4184ea7f | 753 | } |
0f9dd46c | 754 | } |
4184ea7f | 755 | rcu_read_unlock(); |
0f9dd46c | 756 | return NULL; |
6324fbf3 CM |
757 | } |
758 | ||
0d9f824d | 759 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes, |
29d2b84c | 760 | bool metadata, u64 root_objectid) |
0d9f824d OS |
761 | { |
762 | struct btrfs_space_info *space_info; | |
763 | u64 flags; | |
764 | ||
29d2b84c | 765 | if (metadata) { |
0d9f824d OS |
766 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) |
767 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
768 | else | |
769 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
770 | } else { | |
771 | flags = BTRFS_BLOCK_GROUP_DATA; | |
772 | } | |
773 | ||
774 | space_info = __find_space_info(fs_info, flags); | |
55e8196a | 775 | ASSERT(space_info); |
dec59fa3 EL |
776 | percpu_counter_add_batch(&space_info->total_bytes_pinned, num_bytes, |
777 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0d9f824d OS |
778 | } |
779 | ||
4184ea7f CM |
780 | /* |
781 | * after adding space to the filesystem, we need to clear the full flags | |
782 | * on all the space infos. | |
783 | */ | |
784 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
785 | { | |
786 | struct list_head *head = &info->space_info; | |
787 | struct btrfs_space_info *found; | |
788 | ||
789 | rcu_read_lock(); | |
790 | list_for_each_entry_rcu(found, head, list) | |
791 | found->full = 0; | |
792 | rcu_read_unlock(); | |
793 | } | |
794 | ||
1a4ed8fd | 795 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 796 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
797 | { |
798 | int ret; | |
799 | struct btrfs_key key; | |
31840ae1 | 800 | struct btrfs_path *path; |
e02119d5 | 801 | |
31840ae1 | 802 | path = btrfs_alloc_path(); |
d8926bb3 MF |
803 | if (!path) |
804 | return -ENOMEM; | |
805 | ||
e02119d5 CM |
806 | key.objectid = start; |
807 | key.offset = len; | |
3173a18f | 808 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 809 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 810 | btrfs_free_path(path); |
7bb86316 CM |
811 | return ret; |
812 | } | |
813 | ||
a22285a6 | 814 | /* |
3173a18f | 815 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
816 | * |
817 | * the head node for delayed ref is used to store the sum of all the | |
818 | * reference count modifications queued up in the rbtree. the head | |
819 | * node may also store the extent flags to set. This way you can check | |
820 | * to see what the reference count and extent flags would be if all of | |
821 | * the delayed refs are not processed. | |
822 | */ | |
823 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 824 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 825 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
826 | { |
827 | struct btrfs_delayed_ref_head *head; | |
828 | struct btrfs_delayed_ref_root *delayed_refs; | |
829 | struct btrfs_path *path; | |
830 | struct btrfs_extent_item *ei; | |
831 | struct extent_buffer *leaf; | |
832 | struct btrfs_key key; | |
833 | u32 item_size; | |
834 | u64 num_refs; | |
835 | u64 extent_flags; | |
836 | int ret; | |
837 | ||
3173a18f JB |
838 | /* |
839 | * If we don't have skinny metadata, don't bother doing anything | |
840 | * different | |
841 | */ | |
0b246afa JM |
842 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
843 | offset = fs_info->nodesize; | |
3173a18f JB |
844 | metadata = 0; |
845 | } | |
846 | ||
a22285a6 YZ |
847 | path = btrfs_alloc_path(); |
848 | if (!path) | |
849 | return -ENOMEM; | |
850 | ||
a22285a6 YZ |
851 | if (!trans) { |
852 | path->skip_locking = 1; | |
853 | path->search_commit_root = 1; | |
854 | } | |
639eefc8 FDBM |
855 | |
856 | search_again: | |
857 | key.objectid = bytenr; | |
858 | key.offset = offset; | |
859 | if (metadata) | |
860 | key.type = BTRFS_METADATA_ITEM_KEY; | |
861 | else | |
862 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
863 | ||
0b246afa | 864 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
865 | if (ret < 0) |
866 | goto out_free; | |
867 | ||
3173a18f | 868 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
869 | if (path->slots[0]) { |
870 | path->slots[0]--; | |
871 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
872 | path->slots[0]); | |
873 | if (key.objectid == bytenr && | |
874 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 875 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
876 | ret = 0; |
877 | } | |
3173a18f JB |
878 | } |
879 | ||
a22285a6 YZ |
880 | if (ret == 0) { |
881 | leaf = path->nodes[0]; | |
882 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
883 | if (item_size >= sizeof(*ei)) { | |
884 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
885 | struct btrfs_extent_item); | |
886 | num_refs = btrfs_extent_refs(leaf, ei); | |
887 | extent_flags = btrfs_extent_flags(leaf, ei); | |
888 | } else { | |
ba3c2b19 NB |
889 | ret = -EINVAL; |
890 | btrfs_print_v0_err(fs_info); | |
891 | if (trans) | |
892 | btrfs_abort_transaction(trans, ret); | |
893 | else | |
894 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
895 | ||
896 | goto out_free; | |
a22285a6 | 897 | } |
ba3c2b19 | 898 | |
a22285a6 YZ |
899 | BUG_ON(num_refs == 0); |
900 | } else { | |
901 | num_refs = 0; | |
902 | extent_flags = 0; | |
903 | ret = 0; | |
904 | } | |
905 | ||
906 | if (!trans) | |
907 | goto out; | |
908 | ||
909 | delayed_refs = &trans->transaction->delayed_refs; | |
910 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 911 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
912 | if (head) { |
913 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 914 | refcount_inc(&head->refs); |
a22285a6 YZ |
915 | spin_unlock(&delayed_refs->lock); |
916 | ||
b3b4aa74 | 917 | btrfs_release_path(path); |
a22285a6 | 918 | |
8cc33e5c DS |
919 | /* |
920 | * Mutex was contended, block until it's released and try | |
921 | * again | |
922 | */ | |
a22285a6 YZ |
923 | mutex_lock(&head->mutex); |
924 | mutex_unlock(&head->mutex); | |
d278850e | 925 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 926 | goto search_again; |
a22285a6 | 927 | } |
d7df2c79 | 928 | spin_lock(&head->lock); |
a22285a6 YZ |
929 | if (head->extent_op && head->extent_op->update_flags) |
930 | extent_flags |= head->extent_op->flags_to_set; | |
931 | else | |
932 | BUG_ON(num_refs == 0); | |
933 | ||
d278850e | 934 | num_refs += head->ref_mod; |
d7df2c79 | 935 | spin_unlock(&head->lock); |
a22285a6 YZ |
936 | mutex_unlock(&head->mutex); |
937 | } | |
938 | spin_unlock(&delayed_refs->lock); | |
939 | out: | |
940 | WARN_ON(num_refs == 0); | |
941 | if (refs) | |
942 | *refs = num_refs; | |
943 | if (flags) | |
944 | *flags = extent_flags; | |
945 | out_free: | |
946 | btrfs_free_path(path); | |
947 | return ret; | |
948 | } | |
949 | ||
d8d5f3e1 CM |
950 | /* |
951 | * Back reference rules. Back refs have three main goals: | |
952 | * | |
953 | * 1) differentiate between all holders of references to an extent so that | |
954 | * when a reference is dropped we can make sure it was a valid reference | |
955 | * before freeing the extent. | |
956 | * | |
957 | * 2) Provide enough information to quickly find the holders of an extent | |
958 | * if we notice a given block is corrupted or bad. | |
959 | * | |
960 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
961 | * maintenance. This is actually the same as #2, but with a slightly | |
962 | * different use case. | |
963 | * | |
5d4f98a2 YZ |
964 | * There are two kinds of back refs. The implicit back refs is optimized |
965 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
966 | * back refs of this kind provide information about the block's owner tree | |
967 | * and the pointer's key. These information allow us to find the block by | |
968 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
969 | * referenced by their owner trees. The location of tree block is recorded | |
970 | * in the back refs. Actually the full back refs is generic, and can be | |
971 | * used in all cases the implicit back refs is used. The major shortcoming | |
972 | * of the full back refs is its overhead. Every time a tree block gets | |
973 | * COWed, we have to update back refs entry for all pointers in it. | |
974 | * | |
975 | * For a newly allocated tree block, we use implicit back refs for | |
976 | * pointers in it. This means most tree related operations only involve | |
977 | * implicit back refs. For a tree block created in old transaction, the | |
978 | * only way to drop a reference to it is COW it. So we can detect the | |
979 | * event that tree block loses its owner tree's reference and do the | |
980 | * back refs conversion. | |
981 | * | |
01327610 | 982 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
983 | * |
984 | * The reference count of the block is one and the tree is the block's | |
985 | * owner tree. Nothing to do in this case. | |
986 | * | |
987 | * The reference count of the block is one and the tree is not the | |
988 | * block's owner tree. In this case, full back refs is used for pointers | |
989 | * in the block. Remove these full back refs, add implicit back refs for | |
990 | * every pointers in the new block. | |
991 | * | |
992 | * The reference count of the block is greater than one and the tree is | |
993 | * the block's owner tree. In this case, implicit back refs is used for | |
994 | * pointers in the block. Add full back refs for every pointers in the | |
995 | * block, increase lower level extents' reference counts. The original | |
996 | * implicit back refs are entailed to the new block. | |
997 | * | |
998 | * The reference count of the block is greater than one and the tree is | |
999 | * not the block's owner tree. Add implicit back refs for every pointer in | |
1000 | * the new block, increase lower level extents' reference count. | |
1001 | * | |
1002 | * Back Reference Key composing: | |
1003 | * | |
1004 | * The key objectid corresponds to the first byte in the extent, | |
1005 | * The key type is used to differentiate between types of back refs. | |
1006 | * There are different meanings of the key offset for different types | |
1007 | * of back refs. | |
1008 | * | |
d8d5f3e1 CM |
1009 | * File extents can be referenced by: |
1010 | * | |
1011 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1012 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1013 | * - different offsets inside a file (bookend extents in file.c) |
1014 | * | |
5d4f98a2 | 1015 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1016 | * |
1017 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1018 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1019 | * - original offset in the file |
1020 | * - how many bookend extents | |
d8d5f3e1 | 1021 | * |
5d4f98a2 YZ |
1022 | * The key offset for the implicit back refs is hash of the first |
1023 | * three fields. | |
d8d5f3e1 | 1024 | * |
5d4f98a2 | 1025 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1026 | * |
5d4f98a2 | 1027 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1028 | * |
5d4f98a2 YZ |
1029 | * The key offset for the implicit back refs is the first byte of |
1030 | * the tree leaf | |
d8d5f3e1 | 1031 | * |
5d4f98a2 YZ |
1032 | * When a file extent is allocated, The implicit back refs is used. |
1033 | * the fields are filled in: | |
d8d5f3e1 | 1034 | * |
5d4f98a2 | 1035 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1036 | * |
5d4f98a2 YZ |
1037 | * When a file extent is removed file truncation, we find the |
1038 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1039 | * |
5d4f98a2 | 1040 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1041 | * |
5d4f98a2 | 1042 | * Btree extents can be referenced by: |
d8d5f3e1 | 1043 | * |
5d4f98a2 | 1044 | * - Different subvolumes |
d8d5f3e1 | 1045 | * |
5d4f98a2 YZ |
1046 | * Both the implicit back refs and the full back refs for tree blocks |
1047 | * only consist of key. The key offset for the implicit back refs is | |
1048 | * objectid of block's owner tree. The key offset for the full back refs | |
1049 | * is the first byte of parent block. | |
d8d5f3e1 | 1050 | * |
5d4f98a2 YZ |
1051 | * When implicit back refs is used, information about the lowest key and |
1052 | * level of the tree block are required. These information are stored in | |
1053 | * tree block info structure. | |
d8d5f3e1 | 1054 | */ |
31840ae1 | 1055 | |
167ce953 LB |
1056 | /* |
1057 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 1058 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
1059 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
1060 | */ | |
1061 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
1062 | struct btrfs_extent_inline_ref *iref, | |
1063 | enum btrfs_inline_ref_type is_data) | |
1064 | { | |
1065 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1066 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1067 | |
1068 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1069 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1070 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1071 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1072 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1073 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1074 | return type; |
64ecdb64 LB |
1075 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1076 | ASSERT(eb->fs_info); | |
1077 | /* | |
1078 | * Every shared one has parent tree | |
1079 | * block, which must be aligned to | |
1080 | * nodesize. | |
1081 | */ | |
1082 | if (offset && | |
1083 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1084 | return type; | |
1085 | } | |
167ce953 | 1086 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1087 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1088 | return type; |
64ecdb64 LB |
1089 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1090 | ASSERT(eb->fs_info); | |
1091 | /* | |
1092 | * Every shared one has parent tree | |
1093 | * block, which must be aligned to | |
1094 | * nodesize. | |
1095 | */ | |
1096 | if (offset && | |
1097 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1098 | return type; | |
1099 | } | |
167ce953 LB |
1100 | } else { |
1101 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1102 | return type; | |
1103 | } | |
1104 | } | |
1105 | ||
1106 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1107 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1108 | eb->start, type); | |
1109 | WARN_ON(1); | |
1110 | ||
1111 | return BTRFS_REF_TYPE_INVALID; | |
1112 | } | |
1113 | ||
5d4f98a2 YZ |
1114 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1115 | { | |
1116 | u32 high_crc = ~(u32)0; | |
1117 | u32 low_crc = ~(u32)0; | |
1118 | __le64 lenum; | |
1119 | ||
1120 | lenum = cpu_to_le64(root_objectid); | |
9678c543 | 1121 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1122 | lenum = cpu_to_le64(owner); |
9678c543 | 1123 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1124 | lenum = cpu_to_le64(offset); |
9678c543 | 1125 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1126 | |
1127 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1128 | } | |
1129 | ||
1130 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1131 | struct btrfs_extent_data_ref *ref) | |
1132 | { | |
1133 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1134 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1135 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1136 | } | |
1137 | ||
1138 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1139 | struct btrfs_extent_data_ref *ref, | |
1140 | u64 root_objectid, u64 owner, u64 offset) | |
1141 | { | |
1142 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1143 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1144 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1145 | return 0; | |
1146 | return 1; | |
1147 | } | |
1148 | ||
1149 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1150 | struct btrfs_path *path, |
1151 | u64 bytenr, u64 parent, | |
1152 | u64 root_objectid, | |
1153 | u64 owner, u64 offset) | |
1154 | { | |
bd1d53ef | 1155 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
1156 | struct btrfs_key key; |
1157 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1158 | struct extent_buffer *leaf; |
5d4f98a2 | 1159 | u32 nritems; |
74493f7a | 1160 | int ret; |
5d4f98a2 YZ |
1161 | int recow; |
1162 | int err = -ENOENT; | |
74493f7a | 1163 | |
31840ae1 | 1164 | key.objectid = bytenr; |
5d4f98a2 YZ |
1165 | if (parent) { |
1166 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1167 | key.offset = parent; | |
1168 | } else { | |
1169 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1170 | key.offset = hash_extent_data_ref(root_objectid, | |
1171 | owner, offset); | |
1172 | } | |
1173 | again: | |
1174 | recow = 0; | |
1175 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1176 | if (ret < 0) { | |
1177 | err = ret; | |
1178 | goto fail; | |
1179 | } | |
31840ae1 | 1180 | |
5d4f98a2 YZ |
1181 | if (parent) { |
1182 | if (!ret) | |
1183 | return 0; | |
5d4f98a2 | 1184 | goto fail; |
31840ae1 ZY |
1185 | } |
1186 | ||
1187 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1188 | nritems = btrfs_header_nritems(leaf); |
1189 | while (1) { | |
1190 | if (path->slots[0] >= nritems) { | |
1191 | ret = btrfs_next_leaf(root, path); | |
1192 | if (ret < 0) | |
1193 | err = ret; | |
1194 | if (ret) | |
1195 | goto fail; | |
1196 | ||
1197 | leaf = path->nodes[0]; | |
1198 | nritems = btrfs_header_nritems(leaf); | |
1199 | recow = 1; | |
1200 | } | |
1201 | ||
1202 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1203 | if (key.objectid != bytenr || | |
1204 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1205 | goto fail; | |
1206 | ||
1207 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1208 | struct btrfs_extent_data_ref); | |
1209 | ||
1210 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1211 | owner, offset)) { | |
1212 | if (recow) { | |
b3b4aa74 | 1213 | btrfs_release_path(path); |
5d4f98a2 YZ |
1214 | goto again; |
1215 | } | |
1216 | err = 0; | |
1217 | break; | |
1218 | } | |
1219 | path->slots[0]++; | |
31840ae1 | 1220 | } |
5d4f98a2 YZ |
1221 | fail: |
1222 | return err; | |
31840ae1 ZY |
1223 | } |
1224 | ||
5d4f98a2 | 1225 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1226 | struct btrfs_path *path, |
1227 | u64 bytenr, u64 parent, | |
1228 | u64 root_objectid, u64 owner, | |
1229 | u64 offset, int refs_to_add) | |
31840ae1 | 1230 | { |
62b895af | 1231 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
1232 | struct btrfs_key key; |
1233 | struct extent_buffer *leaf; | |
5d4f98a2 | 1234 | u32 size; |
31840ae1 ZY |
1235 | u32 num_refs; |
1236 | int ret; | |
74493f7a | 1237 | |
74493f7a | 1238 | key.objectid = bytenr; |
5d4f98a2 YZ |
1239 | if (parent) { |
1240 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1241 | key.offset = parent; | |
1242 | size = sizeof(struct btrfs_shared_data_ref); | |
1243 | } else { | |
1244 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1245 | key.offset = hash_extent_data_ref(root_objectid, | |
1246 | owner, offset); | |
1247 | size = sizeof(struct btrfs_extent_data_ref); | |
1248 | } | |
74493f7a | 1249 | |
5d4f98a2 YZ |
1250 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1251 | if (ret && ret != -EEXIST) | |
1252 | goto fail; | |
1253 | ||
1254 | leaf = path->nodes[0]; | |
1255 | if (parent) { | |
1256 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1257 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1258 | struct btrfs_shared_data_ref); |
1259 | if (ret == 0) { | |
1260 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1261 | } else { | |
1262 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1263 | num_refs += refs_to_add; | |
1264 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1265 | } |
5d4f98a2 YZ |
1266 | } else { |
1267 | struct btrfs_extent_data_ref *ref; | |
1268 | while (ret == -EEXIST) { | |
1269 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1270 | struct btrfs_extent_data_ref); | |
1271 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1272 | owner, offset)) | |
1273 | break; | |
b3b4aa74 | 1274 | btrfs_release_path(path); |
5d4f98a2 YZ |
1275 | key.offset++; |
1276 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1277 | size); | |
1278 | if (ret && ret != -EEXIST) | |
1279 | goto fail; | |
31840ae1 | 1280 | |
5d4f98a2 YZ |
1281 | leaf = path->nodes[0]; |
1282 | } | |
1283 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1284 | struct btrfs_extent_data_ref); | |
1285 | if (ret == 0) { | |
1286 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1287 | root_objectid); | |
1288 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1289 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1290 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1291 | } else { | |
1292 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1293 | num_refs += refs_to_add; | |
1294 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1295 | } |
31840ae1 | 1296 | } |
5d4f98a2 YZ |
1297 | btrfs_mark_buffer_dirty(leaf); |
1298 | ret = 0; | |
1299 | fail: | |
b3b4aa74 | 1300 | btrfs_release_path(path); |
7bb86316 | 1301 | return ret; |
74493f7a CM |
1302 | } |
1303 | ||
5d4f98a2 | 1304 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 1305 | struct btrfs_path *path, |
fcebe456 | 1306 | int refs_to_drop, int *last_ref) |
31840ae1 | 1307 | { |
5d4f98a2 YZ |
1308 | struct btrfs_key key; |
1309 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1310 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1311 | struct extent_buffer *leaf; |
5d4f98a2 | 1312 | u32 num_refs = 0; |
31840ae1 ZY |
1313 | int ret = 0; |
1314 | ||
1315 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1316 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1317 | ||
1318 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1319 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1320 | struct btrfs_extent_data_ref); | |
1321 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1322 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1323 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1324 | struct btrfs_shared_data_ref); | |
1325 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 1326 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
1327 | btrfs_print_v0_err(trans->fs_info); |
1328 | btrfs_abort_transaction(trans, -EINVAL); | |
1329 | return -EINVAL; | |
5d4f98a2 YZ |
1330 | } else { |
1331 | BUG(); | |
1332 | } | |
1333 | ||
56bec294 CM |
1334 | BUG_ON(num_refs < refs_to_drop); |
1335 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1336 | |
31840ae1 | 1337 | if (num_refs == 0) { |
e9f6290d | 1338 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 1339 | *last_ref = 1; |
31840ae1 | 1340 | } else { |
5d4f98a2 YZ |
1341 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1342 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1343 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1344 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
1345 | btrfs_mark_buffer_dirty(leaf); |
1346 | } | |
31840ae1 ZY |
1347 | return ret; |
1348 | } | |
1349 | ||
9ed0dea0 | 1350 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1351 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1352 | { |
5d4f98a2 YZ |
1353 | struct btrfs_key key; |
1354 | struct extent_buffer *leaf; | |
1355 | struct btrfs_extent_data_ref *ref1; | |
1356 | struct btrfs_shared_data_ref *ref2; | |
1357 | u32 num_refs = 0; | |
3de28d57 | 1358 | int type; |
5d4f98a2 YZ |
1359 | |
1360 | leaf = path->nodes[0]; | |
1361 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
1362 | |
1363 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 1364 | if (iref) { |
3de28d57 LB |
1365 | /* |
1366 | * If type is invalid, we should have bailed out earlier than | |
1367 | * this call. | |
1368 | */ | |
1369 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1370 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1371 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1372 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1373 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1374 | } else { | |
1375 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1376 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1377 | } | |
1378 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1379 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1380 | struct btrfs_extent_data_ref); | |
1381 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1382 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1383 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1384 | struct btrfs_shared_data_ref); | |
1385 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
1386 | } else { |
1387 | WARN_ON(1); | |
1388 | } | |
1389 | return num_refs; | |
1390 | } | |
15916de8 | 1391 | |
5d4f98a2 | 1392 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1393 | struct btrfs_path *path, |
1394 | u64 bytenr, u64 parent, | |
1395 | u64 root_objectid) | |
1f3c79a2 | 1396 | { |
b8582eea | 1397 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 1398 | struct btrfs_key key; |
1f3c79a2 | 1399 | int ret; |
1f3c79a2 | 1400 | |
5d4f98a2 YZ |
1401 | key.objectid = bytenr; |
1402 | if (parent) { | |
1403 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1404 | key.offset = parent; | |
1405 | } else { | |
1406 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1407 | key.offset = root_objectid; | |
1f3c79a2 LH |
1408 | } |
1409 | ||
5d4f98a2 YZ |
1410 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1411 | if (ret > 0) | |
1412 | ret = -ENOENT; | |
5d4f98a2 | 1413 | return ret; |
1f3c79a2 LH |
1414 | } |
1415 | ||
5d4f98a2 | 1416 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1417 | struct btrfs_path *path, |
1418 | u64 bytenr, u64 parent, | |
1419 | u64 root_objectid) | |
31840ae1 | 1420 | { |
5d4f98a2 | 1421 | struct btrfs_key key; |
31840ae1 | 1422 | int ret; |
31840ae1 | 1423 | |
5d4f98a2 YZ |
1424 | key.objectid = bytenr; |
1425 | if (parent) { | |
1426 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1427 | key.offset = parent; | |
1428 | } else { | |
1429 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1430 | key.offset = root_objectid; | |
1431 | } | |
1432 | ||
10728404 | 1433 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 1434 | path, &key, 0); |
b3b4aa74 | 1435 | btrfs_release_path(path); |
31840ae1 ZY |
1436 | return ret; |
1437 | } | |
1438 | ||
5d4f98a2 | 1439 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1440 | { |
5d4f98a2 YZ |
1441 | int type; |
1442 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1443 | if (parent > 0) | |
1444 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1445 | else | |
1446 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1447 | } else { | |
1448 | if (parent > 0) | |
1449 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1450 | else | |
1451 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1452 | } | |
1453 | return type; | |
31840ae1 | 1454 | } |
56bec294 | 1455 | |
2c47e605 YZ |
1456 | static int find_next_key(struct btrfs_path *path, int level, |
1457 | struct btrfs_key *key) | |
56bec294 | 1458 | |
02217ed2 | 1459 | { |
2c47e605 | 1460 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1461 | if (!path->nodes[level]) |
1462 | break; | |
5d4f98a2 YZ |
1463 | if (path->slots[level] + 1 >= |
1464 | btrfs_header_nritems(path->nodes[level])) | |
1465 | continue; | |
1466 | if (level == 0) | |
1467 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1468 | path->slots[level] + 1); | |
1469 | else | |
1470 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1471 | path->slots[level] + 1); | |
1472 | return 0; | |
1473 | } | |
1474 | return 1; | |
1475 | } | |
037e6390 | 1476 | |
5d4f98a2 YZ |
1477 | /* |
1478 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1479 | * to the address of inline back ref, and 0 is returned. | |
1480 | * | |
1481 | * if back ref isn't found, *ref_ret is set to the address where it | |
1482 | * should be inserted, and -ENOENT is returned. | |
1483 | * | |
1484 | * if insert is true and there are too many inline back refs, the path | |
1485 | * points to the extent item, and -EAGAIN is returned. | |
1486 | * | |
1487 | * NOTE: inline back refs are ordered in the same way that back ref | |
1488 | * items in the tree are ordered. | |
1489 | */ | |
1490 | static noinline_for_stack | |
1491 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1492 | struct btrfs_path *path, |
1493 | struct btrfs_extent_inline_ref **ref_ret, | |
1494 | u64 bytenr, u64 num_bytes, | |
1495 | u64 parent, u64 root_objectid, | |
1496 | u64 owner, u64 offset, int insert) | |
1497 | { | |
867cc1fb | 1498 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 1499 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1500 | struct btrfs_key key; |
1501 | struct extent_buffer *leaf; | |
1502 | struct btrfs_extent_item *ei; | |
1503 | struct btrfs_extent_inline_ref *iref; | |
1504 | u64 flags; | |
1505 | u64 item_size; | |
1506 | unsigned long ptr; | |
1507 | unsigned long end; | |
1508 | int extra_size; | |
1509 | int type; | |
1510 | int want; | |
1511 | int ret; | |
1512 | int err = 0; | |
0b246afa | 1513 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1514 | int needed; |
26b8003f | 1515 | |
db94535d | 1516 | key.objectid = bytenr; |
31840ae1 | 1517 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1518 | key.offset = num_bytes; |
31840ae1 | 1519 | |
5d4f98a2 YZ |
1520 | want = extent_ref_type(parent, owner); |
1521 | if (insert) { | |
1522 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1523 | path->keep_locks = 1; |
5d4f98a2 YZ |
1524 | } else |
1525 | extra_size = -1; | |
3173a18f JB |
1526 | |
1527 | /* | |
16d1c062 NB |
1528 | * Owner is our level, so we can just add one to get the level for the |
1529 | * block we are interested in. | |
3173a18f JB |
1530 | */ |
1531 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1532 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1533 | key.offset = owner; | |
1534 | } | |
1535 | ||
1536 | again: | |
5d4f98a2 | 1537 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1538 | if (ret < 0) { |
5d4f98a2 YZ |
1539 | err = ret; |
1540 | goto out; | |
1541 | } | |
3173a18f JB |
1542 | |
1543 | /* | |
1544 | * We may be a newly converted file system which still has the old fat | |
1545 | * extent entries for metadata, so try and see if we have one of those. | |
1546 | */ | |
1547 | if (ret > 0 && skinny_metadata) { | |
1548 | skinny_metadata = false; | |
1549 | if (path->slots[0]) { | |
1550 | path->slots[0]--; | |
1551 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1552 | path->slots[0]); | |
1553 | if (key.objectid == bytenr && | |
1554 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1555 | key.offset == num_bytes) | |
1556 | ret = 0; | |
1557 | } | |
1558 | if (ret) { | |
9ce49a0b | 1559 | key.objectid = bytenr; |
3173a18f JB |
1560 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1561 | key.offset = num_bytes; | |
1562 | btrfs_release_path(path); | |
1563 | goto again; | |
1564 | } | |
1565 | } | |
1566 | ||
79787eaa JM |
1567 | if (ret && !insert) { |
1568 | err = -ENOENT; | |
1569 | goto out; | |
fae7f21c | 1570 | } else if (WARN_ON(ret)) { |
492104c8 | 1571 | err = -EIO; |
492104c8 | 1572 | goto out; |
79787eaa | 1573 | } |
5d4f98a2 YZ |
1574 | |
1575 | leaf = path->nodes[0]; | |
1576 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 1577 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1578 | err = -EINVAL; |
1579 | btrfs_print_v0_err(fs_info); | |
1580 | btrfs_abort_transaction(trans, err); | |
1581 | goto out; | |
1582 | } | |
5d4f98a2 | 1583 | |
5d4f98a2 YZ |
1584 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1585 | flags = btrfs_extent_flags(leaf, ei); | |
1586 | ||
1587 | ptr = (unsigned long)(ei + 1); | |
1588 | end = (unsigned long)ei + item_size; | |
1589 | ||
3173a18f | 1590 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1591 | ptr += sizeof(struct btrfs_tree_block_info); |
1592 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1593 | } |
1594 | ||
3de28d57 LB |
1595 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1596 | needed = BTRFS_REF_TYPE_DATA; | |
1597 | else | |
1598 | needed = BTRFS_REF_TYPE_BLOCK; | |
1599 | ||
5d4f98a2 YZ |
1600 | err = -ENOENT; |
1601 | while (1) { | |
1602 | if (ptr >= end) { | |
1603 | WARN_ON(ptr > end); | |
1604 | break; | |
1605 | } | |
1606 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1607 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1608 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 1609 | err = -EUCLEAN; |
3de28d57 LB |
1610 | goto out; |
1611 | } | |
1612 | ||
5d4f98a2 YZ |
1613 | if (want < type) |
1614 | break; | |
1615 | if (want > type) { | |
1616 | ptr += btrfs_extent_inline_ref_size(type); | |
1617 | continue; | |
1618 | } | |
1619 | ||
1620 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1621 | struct btrfs_extent_data_ref *dref; | |
1622 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1623 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1624 | owner, offset)) { | |
1625 | err = 0; | |
1626 | break; | |
1627 | } | |
1628 | if (hash_extent_data_ref_item(leaf, dref) < | |
1629 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1630 | break; | |
1631 | } else { | |
1632 | u64 ref_offset; | |
1633 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1634 | if (parent > 0) { | |
1635 | if (parent == ref_offset) { | |
1636 | err = 0; | |
1637 | break; | |
1638 | } | |
1639 | if (ref_offset < parent) | |
1640 | break; | |
1641 | } else { | |
1642 | if (root_objectid == ref_offset) { | |
1643 | err = 0; | |
1644 | break; | |
1645 | } | |
1646 | if (ref_offset < root_objectid) | |
1647 | break; | |
1648 | } | |
1649 | } | |
1650 | ptr += btrfs_extent_inline_ref_size(type); | |
1651 | } | |
1652 | if (err == -ENOENT && insert) { | |
1653 | if (item_size + extra_size >= | |
1654 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1655 | err = -EAGAIN; | |
1656 | goto out; | |
1657 | } | |
1658 | /* | |
1659 | * To add new inline back ref, we have to make sure | |
1660 | * there is no corresponding back ref item. | |
1661 | * For simplicity, we just do not add new inline back | |
1662 | * ref if there is any kind of item for this block | |
1663 | */ | |
2c47e605 YZ |
1664 | if (find_next_key(path, 0, &key) == 0 && |
1665 | key.objectid == bytenr && | |
85d4198e | 1666 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1667 | err = -EAGAIN; |
1668 | goto out; | |
1669 | } | |
1670 | } | |
1671 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1672 | out: | |
85d4198e | 1673 | if (insert) { |
5d4f98a2 YZ |
1674 | path->keep_locks = 0; |
1675 | btrfs_unlock_up_safe(path, 1); | |
1676 | } | |
1677 | return err; | |
1678 | } | |
1679 | ||
1680 | /* | |
1681 | * helper to add new inline back ref | |
1682 | */ | |
1683 | static noinline_for_stack | |
87bde3cd | 1684 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1685 | struct btrfs_path *path, |
1686 | struct btrfs_extent_inline_ref *iref, | |
1687 | u64 parent, u64 root_objectid, | |
1688 | u64 owner, u64 offset, int refs_to_add, | |
1689 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1690 | { |
1691 | struct extent_buffer *leaf; | |
1692 | struct btrfs_extent_item *ei; | |
1693 | unsigned long ptr; | |
1694 | unsigned long end; | |
1695 | unsigned long item_offset; | |
1696 | u64 refs; | |
1697 | int size; | |
1698 | int type; | |
5d4f98a2 YZ |
1699 | |
1700 | leaf = path->nodes[0]; | |
1701 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1702 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1703 | ||
1704 | type = extent_ref_type(parent, owner); | |
1705 | size = btrfs_extent_inline_ref_size(type); | |
1706 | ||
87bde3cd | 1707 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1708 | |
1709 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1710 | refs = btrfs_extent_refs(leaf, ei); | |
1711 | refs += refs_to_add; | |
1712 | btrfs_set_extent_refs(leaf, ei, refs); | |
1713 | if (extent_op) | |
1714 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1715 | ||
1716 | ptr = (unsigned long)ei + item_offset; | |
1717 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1718 | if (ptr < end - size) | |
1719 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1720 | end - size - ptr); | |
1721 | ||
1722 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1723 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1724 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1725 | struct btrfs_extent_data_ref *dref; | |
1726 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1727 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1728 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1729 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1730 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1731 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1732 | struct btrfs_shared_data_ref *sref; | |
1733 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1734 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1735 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1736 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1737 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1738 | } else { | |
1739 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1740 | } | |
1741 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1742 | } |
1743 | ||
1744 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1745 | struct btrfs_path *path, |
1746 | struct btrfs_extent_inline_ref **ref_ret, | |
1747 | u64 bytenr, u64 num_bytes, u64 parent, | |
1748 | u64 root_objectid, u64 owner, u64 offset) | |
1749 | { | |
1750 | int ret; | |
1751 | ||
867cc1fb NB |
1752 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1753 | num_bytes, parent, root_objectid, | |
1754 | owner, offset, 0); | |
5d4f98a2 | 1755 | if (ret != -ENOENT) |
54aa1f4d | 1756 | return ret; |
5d4f98a2 | 1757 | |
b3b4aa74 | 1758 | btrfs_release_path(path); |
5d4f98a2 YZ |
1759 | *ref_ret = NULL; |
1760 | ||
1761 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1762 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1763 | root_objectid); | |
5d4f98a2 | 1764 | } else { |
bd1d53ef NB |
1765 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1766 | root_objectid, owner, offset); | |
b9473439 | 1767 | } |
5d4f98a2 YZ |
1768 | return ret; |
1769 | } | |
31840ae1 | 1770 | |
5d4f98a2 YZ |
1771 | /* |
1772 | * helper to update/remove inline back ref | |
1773 | */ | |
1774 | static noinline_for_stack | |
61a18f1c | 1775 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1776 | struct btrfs_extent_inline_ref *iref, |
1777 | int refs_to_mod, | |
fcebe456 JB |
1778 | struct btrfs_delayed_extent_op *extent_op, |
1779 | int *last_ref) | |
5d4f98a2 | 1780 | { |
61a18f1c NB |
1781 | struct extent_buffer *leaf = path->nodes[0]; |
1782 | struct btrfs_fs_info *fs_info = leaf->fs_info; | |
5d4f98a2 YZ |
1783 | struct btrfs_extent_item *ei; |
1784 | struct btrfs_extent_data_ref *dref = NULL; | |
1785 | struct btrfs_shared_data_ref *sref = NULL; | |
1786 | unsigned long ptr; | |
1787 | unsigned long end; | |
1788 | u32 item_size; | |
1789 | int size; | |
1790 | int type; | |
5d4f98a2 YZ |
1791 | u64 refs; |
1792 | ||
5d4f98a2 YZ |
1793 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1794 | refs = btrfs_extent_refs(leaf, ei); | |
1795 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1796 | refs += refs_to_mod; | |
1797 | btrfs_set_extent_refs(leaf, ei, refs); | |
1798 | if (extent_op) | |
1799 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1800 | ||
3de28d57 LB |
1801 | /* |
1802 | * If type is invalid, we should have bailed out after | |
1803 | * lookup_inline_extent_backref(). | |
1804 | */ | |
1805 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1806 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1807 | |
1808 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1809 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1810 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1811 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1812 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1813 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1814 | } else { | |
1815 | refs = 1; | |
1816 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1817 | } |
31840ae1 | 1818 | |
5d4f98a2 YZ |
1819 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1820 | refs += refs_to_mod; | |
1821 | ||
1822 | if (refs > 0) { | |
1823 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1824 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1825 | else | |
1826 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1827 | } else { | |
fcebe456 | 1828 | *last_ref = 1; |
5d4f98a2 YZ |
1829 | size = btrfs_extent_inline_ref_size(type); |
1830 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1831 | ptr = (unsigned long)iref; | |
1832 | end = (unsigned long)ei + item_size; | |
1833 | if (ptr + size < end) | |
1834 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1835 | end - ptr - size); | |
1836 | item_size -= size; | |
87bde3cd | 1837 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1838 | } |
1839 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1840 | } |
1841 | ||
1842 | static noinline_for_stack | |
1843 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1844 | struct btrfs_path *path, |
1845 | u64 bytenr, u64 num_bytes, u64 parent, | |
1846 | u64 root_objectid, u64 owner, | |
1847 | u64 offset, int refs_to_add, | |
1848 | struct btrfs_delayed_extent_op *extent_op) | |
1849 | { | |
1850 | struct btrfs_extent_inline_ref *iref; | |
1851 | int ret; | |
1852 | ||
867cc1fb NB |
1853 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1854 | num_bytes, parent, root_objectid, | |
1855 | owner, offset, 1); | |
5d4f98a2 YZ |
1856 | if (ret == 0) { |
1857 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
61a18f1c NB |
1858 | update_inline_extent_backref(path, iref, refs_to_add, |
1859 | extent_op, NULL); | |
5d4f98a2 | 1860 | } else if (ret == -ENOENT) { |
a639cdeb | 1861 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1862 | root_objectid, owner, offset, |
1863 | refs_to_add, extent_op); | |
1864 | ret = 0; | |
771ed689 | 1865 | } |
5d4f98a2 YZ |
1866 | return ret; |
1867 | } | |
31840ae1 | 1868 | |
5d4f98a2 | 1869 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1870 | struct btrfs_path *path, |
1871 | u64 bytenr, u64 parent, u64 root_objectid, | |
1872 | u64 owner, u64 offset, int refs_to_add) | |
1873 | { | |
1874 | int ret; | |
1875 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1876 | BUG_ON(refs_to_add != 1); | |
10728404 NB |
1877 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
1878 | root_objectid); | |
5d4f98a2 | 1879 | } else { |
62b895af NB |
1880 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
1881 | root_objectid, owner, offset, | |
1882 | refs_to_add); | |
5d4f98a2 YZ |
1883 | } |
1884 | return ret; | |
1885 | } | |
56bec294 | 1886 | |
5d4f98a2 | 1887 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1888 | struct btrfs_path *path, |
1889 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1890 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1891 | { |
143bede5 | 1892 | int ret = 0; |
b9473439 | 1893 | |
5d4f98a2 YZ |
1894 | BUG_ON(!is_data && refs_to_drop != 1); |
1895 | if (iref) { | |
61a18f1c NB |
1896 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1897 | last_ref); | |
5d4f98a2 | 1898 | } else if (is_data) { |
e9f6290d | 1899 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1900 | last_ref); |
5d4f98a2 | 1901 | } else { |
fcebe456 | 1902 | *last_ref = 1; |
87cc7a8a | 1903 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1904 | } |
1905 | return ret; | |
1906 | } | |
1907 | ||
d04c6b88 JM |
1908 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1909 | u64 *discarded_bytes) | |
5d4f98a2 | 1910 | { |
86557861 JM |
1911 | int j, ret = 0; |
1912 | u64 bytes_left, end; | |
4d89d377 | 1913 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1914 | |
4d89d377 JM |
1915 | if (WARN_ON(start != aligned_start)) { |
1916 | len -= aligned_start - start; | |
1917 | len = round_down(len, 1 << 9); | |
1918 | start = aligned_start; | |
1919 | } | |
d04c6b88 | 1920 | |
4d89d377 | 1921 | *discarded_bytes = 0; |
86557861 JM |
1922 | |
1923 | if (!len) | |
1924 | return 0; | |
1925 | ||
1926 | end = start + len; | |
1927 | bytes_left = len; | |
1928 | ||
1929 | /* Skip any superblocks on this device. */ | |
1930 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1931 | u64 sb_start = btrfs_sb_offset(j); | |
1932 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1933 | u64 size = sb_start - start; | |
1934 | ||
1935 | if (!in_range(sb_start, start, bytes_left) && | |
1936 | !in_range(sb_end, start, bytes_left) && | |
1937 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1938 | continue; | |
1939 | ||
1940 | /* | |
1941 | * Superblock spans beginning of range. Adjust start and | |
1942 | * try again. | |
1943 | */ | |
1944 | if (sb_start <= start) { | |
1945 | start += sb_end - start; | |
1946 | if (start > end) { | |
1947 | bytes_left = 0; | |
1948 | break; | |
1949 | } | |
1950 | bytes_left = end - start; | |
1951 | continue; | |
1952 | } | |
1953 | ||
1954 | if (size) { | |
1955 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1956 | GFP_NOFS, 0); | |
1957 | if (!ret) | |
1958 | *discarded_bytes += size; | |
1959 | else if (ret != -EOPNOTSUPP) | |
1960 | return ret; | |
1961 | } | |
1962 | ||
1963 | start = sb_end; | |
1964 | if (start > end) { | |
1965 | bytes_left = 0; | |
1966 | break; | |
1967 | } | |
1968 | bytes_left = end - start; | |
1969 | } | |
1970 | ||
1971 | if (bytes_left) { | |
1972 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1973 | GFP_NOFS, 0); |
1974 | if (!ret) | |
86557861 | 1975 | *discarded_bytes += bytes_left; |
4d89d377 | 1976 | } |
d04c6b88 | 1977 | return ret; |
5d4f98a2 | 1978 | } |
5d4f98a2 | 1979 | |
2ff7e61e | 1980 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1981 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1982 | { |
5d4f98a2 | 1983 | int ret; |
5378e607 | 1984 | u64 discarded_bytes = 0; |
a1d3c478 | 1985 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1986 | |
e244a0ae | 1987 | |
2999241d FM |
1988 | /* |
1989 | * Avoid races with device replace and make sure our bbio has devices | |
1990 | * associated to its stripes that don't go away while we are discarding. | |
1991 | */ | |
0b246afa | 1992 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 1993 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
1994 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
1995 | &bbio, 0); | |
79787eaa | 1996 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1997 | if (!ret) { |
a1d3c478 | 1998 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1999 | int i; |
2000 | ||
5d4f98a2 | 2001 | |
a1d3c478 | 2002 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2003 | u64 bytes; |
38b5f68e AJ |
2004 | struct request_queue *req_q; |
2005 | ||
627e0873 FM |
2006 | if (!stripe->dev->bdev) { |
2007 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
2008 | continue; | |
2009 | } | |
38b5f68e AJ |
2010 | req_q = bdev_get_queue(stripe->dev->bdev); |
2011 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
2012 | continue; |
2013 | ||
5378e607 LD |
2014 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2015 | stripe->physical, | |
d04c6b88 JM |
2016 | stripe->length, |
2017 | &bytes); | |
5378e607 | 2018 | if (!ret) |
d04c6b88 | 2019 | discarded_bytes += bytes; |
5378e607 | 2020 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2021 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2022 | |
2023 | /* | |
2024 | * Just in case we get back EOPNOTSUPP for some reason, | |
2025 | * just ignore the return value so we don't screw up | |
2026 | * people calling discard_extent. | |
2027 | */ | |
2028 | ret = 0; | |
5d4f98a2 | 2029 | } |
6e9606d2 | 2030 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2031 | } |
0b246afa | 2032 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2033 | |
2034 | if (actual_bytes) | |
2035 | *actual_bytes = discarded_bytes; | |
2036 | ||
5d4f98a2 | 2037 | |
53b381b3 DW |
2038 | if (ret == -EOPNOTSUPP) |
2039 | ret = 0; | |
5d4f98a2 | 2040 | return ret; |
5d4f98a2 YZ |
2041 | } |
2042 | ||
79787eaa | 2043 | /* Can return -ENOMEM */ |
5d4f98a2 | 2044 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
84f7d8e6 | 2045 | struct btrfs_root *root, |
5d4f98a2 | 2046 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2047 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 | 2048 | { |
84f7d8e6 | 2049 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 2050 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2051 | int ret; |
66d7e7f0 | 2052 | |
5d4f98a2 YZ |
2053 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2054 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2055 | ||
fd708b81 JB |
2056 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid, |
2057 | owner, offset, BTRFS_ADD_DELAYED_REF); | |
2058 | ||
5d4f98a2 | 2059 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
44e1c47d | 2060 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, |
7be07912 OS |
2061 | num_bytes, parent, |
2062 | root_objectid, (int)owner, | |
2063 | BTRFS_ADD_DELAYED_REF, NULL, | |
d7eae340 | 2064 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 2065 | } else { |
88a979c6 | 2066 | ret = btrfs_add_delayed_data_ref(trans, bytenr, |
7be07912 OS |
2067 | num_bytes, parent, |
2068 | root_objectid, owner, offset, | |
d7eae340 OS |
2069 | 0, BTRFS_ADD_DELAYED_REF, |
2070 | &old_ref_mod, &new_ref_mod); | |
5d4f98a2 | 2071 | } |
d7eae340 | 2072 | |
29d2b84c NB |
2073 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) { |
2074 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
2075 | ||
2076 | add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid); | |
2077 | } | |
d7eae340 | 2078 | |
5d4f98a2 YZ |
2079 | return ret; |
2080 | } | |
2081 | ||
bd3c685e NB |
2082 | /* |
2083 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
2084 | * | |
2085 | * @trans: Handle of transaction | |
2086 | * | |
2087 | * @node: The delayed ref node used to get the bytenr/length for | |
2088 | * extent whose references are incremented. | |
2089 | * | |
2090 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
2091 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
2092 | * bytenr of the parent block. Since new extents are always | |
2093 | * created with indirect references, this will only be the case | |
2094 | * when relocating a shared extent. In that case, root_objectid | |
2095 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
2096 | * be 0 | |
2097 | * | |
2098 | * @root_objectid: The id of the root where this modification has originated, | |
2099 | * this can be either one of the well-known metadata trees or | |
2100 | * the subvolume id which references this extent. | |
2101 | * | |
2102 | * @owner: For data extents it is the inode number of the owning file. | |
2103 | * For metadata extents this parameter holds the level in the | |
2104 | * tree of the extent. | |
2105 | * | |
2106 | * @offset: For metadata extents the offset is ignored and is currently | |
2107 | * always passed as 0. For data extents it is the fileoffset | |
2108 | * this extent belongs to. | |
2109 | * | |
2110 | * @refs_to_add Number of references to add | |
2111 | * | |
2112 | * @extent_op Pointer to a structure, holding information necessary when | |
2113 | * updating a tree block's flags | |
2114 | * | |
2115 | */ | |
5d4f98a2 | 2116 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 2117 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2118 | u64 parent, u64 root_objectid, |
2119 | u64 owner, u64 offset, int refs_to_add, | |
2120 | struct btrfs_delayed_extent_op *extent_op) | |
2121 | { | |
2122 | struct btrfs_path *path; | |
2123 | struct extent_buffer *leaf; | |
2124 | struct btrfs_extent_item *item; | |
fcebe456 | 2125 | struct btrfs_key key; |
c682f9b3 QW |
2126 | u64 bytenr = node->bytenr; |
2127 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2128 | u64 refs; |
2129 | int ret; | |
5d4f98a2 YZ |
2130 | |
2131 | path = btrfs_alloc_path(); | |
2132 | if (!path) | |
2133 | return -ENOMEM; | |
2134 | ||
e4058b54 | 2135 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2136 | path->leave_spinning = 1; |
2137 | /* this will setup the path even if it fails to insert the back ref */ | |
a639cdeb NB |
2138 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
2139 | parent, root_objectid, owner, | |
2140 | offset, refs_to_add, extent_op); | |
0ed4792a | 2141 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2142 | goto out; |
fcebe456 JB |
2143 | |
2144 | /* | |
2145 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2146 | * inline extent ref, so just update the reference count and add a | |
2147 | * normal backref. | |
2148 | */ | |
5d4f98a2 | 2149 | leaf = path->nodes[0]; |
fcebe456 | 2150 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2151 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2152 | refs = btrfs_extent_refs(leaf, item); | |
2153 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2154 | if (extent_op) | |
2155 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2156 | |
5d4f98a2 | 2157 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2158 | btrfs_release_path(path); |
56bec294 | 2159 | |
e4058b54 | 2160 | path->reada = READA_FORWARD; |
b9473439 | 2161 | path->leave_spinning = 1; |
56bec294 | 2162 | /* now insert the actual backref */ |
37593410 NB |
2163 | ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid, |
2164 | owner, offset, refs_to_add); | |
79787eaa | 2165 | if (ret) |
66642832 | 2166 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2167 | out: |
56bec294 | 2168 | btrfs_free_path(path); |
30d133fc | 2169 | return ret; |
56bec294 CM |
2170 | } |
2171 | ||
5d4f98a2 | 2172 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2173 | struct btrfs_delayed_ref_node *node, |
2174 | struct btrfs_delayed_extent_op *extent_op, | |
2175 | int insert_reserved) | |
56bec294 | 2176 | { |
5d4f98a2 YZ |
2177 | int ret = 0; |
2178 | struct btrfs_delayed_data_ref *ref; | |
2179 | struct btrfs_key ins; | |
2180 | u64 parent = 0; | |
2181 | u64 ref_root = 0; | |
2182 | u64 flags = 0; | |
2183 | ||
2184 | ins.objectid = node->bytenr; | |
2185 | ins.offset = node->num_bytes; | |
2186 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2187 | ||
2188 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 2189 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2190 | |
5d4f98a2 YZ |
2191 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2192 | parent = ref->parent; | |
fcebe456 | 2193 | ref_root = ref->root; |
5d4f98a2 YZ |
2194 | |
2195 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2196 | if (extent_op) |
5d4f98a2 | 2197 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
2198 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
2199 | flags, ref->objectid, | |
2200 | ref->offset, &ins, | |
2201 | node->ref_mod); | |
5d4f98a2 | 2202 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2203 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2204 | ref->objectid, ref->offset, | |
2205 | node->ref_mod, extent_op); | |
5d4f98a2 | 2206 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2207 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
2208 | ref_root, ref->objectid, |
2209 | ref->offset, node->ref_mod, | |
c682f9b3 | 2210 | extent_op); |
5d4f98a2 YZ |
2211 | } else { |
2212 | BUG(); | |
2213 | } | |
2214 | return ret; | |
2215 | } | |
2216 | ||
2217 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2218 | struct extent_buffer *leaf, | |
2219 | struct btrfs_extent_item *ei) | |
2220 | { | |
2221 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2222 | if (extent_op->update_flags) { | |
2223 | flags |= extent_op->flags_to_set; | |
2224 | btrfs_set_extent_flags(leaf, ei, flags); | |
2225 | } | |
2226 | ||
2227 | if (extent_op->update_key) { | |
2228 | struct btrfs_tree_block_info *bi; | |
2229 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2230 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2231 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2232 | } | |
2233 | } | |
2234 | ||
2235 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 2236 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2237 | struct btrfs_delayed_extent_op *extent_op) |
2238 | { | |
20b9a2d6 | 2239 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
2240 | struct btrfs_key key; |
2241 | struct btrfs_path *path; | |
2242 | struct btrfs_extent_item *ei; | |
2243 | struct extent_buffer *leaf; | |
2244 | u32 item_size; | |
56bec294 | 2245 | int ret; |
5d4f98a2 | 2246 | int err = 0; |
b1c79e09 | 2247 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2248 | |
79787eaa JM |
2249 | if (trans->aborted) |
2250 | return 0; | |
2251 | ||
0b246afa | 2252 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2253 | metadata = 0; |
2254 | ||
5d4f98a2 YZ |
2255 | path = btrfs_alloc_path(); |
2256 | if (!path) | |
2257 | return -ENOMEM; | |
2258 | ||
d278850e | 2259 | key.objectid = head->bytenr; |
5d4f98a2 | 2260 | |
3173a18f | 2261 | if (metadata) { |
3173a18f | 2262 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2263 | key.offset = extent_op->level; |
3173a18f JB |
2264 | } else { |
2265 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2266 | key.offset = head->num_bytes; |
3173a18f JB |
2267 | } |
2268 | ||
2269 | again: | |
e4058b54 | 2270 | path->reada = READA_FORWARD; |
5d4f98a2 | 2271 | path->leave_spinning = 1; |
0b246afa | 2272 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2273 | if (ret < 0) { |
2274 | err = ret; | |
2275 | goto out; | |
2276 | } | |
2277 | if (ret > 0) { | |
3173a18f | 2278 | if (metadata) { |
55994887 FDBM |
2279 | if (path->slots[0] > 0) { |
2280 | path->slots[0]--; | |
2281 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2282 | path->slots[0]); | |
d278850e | 2283 | if (key.objectid == head->bytenr && |
55994887 | 2284 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2285 | key.offset == head->num_bytes) |
55994887 FDBM |
2286 | ret = 0; |
2287 | } | |
2288 | if (ret > 0) { | |
2289 | btrfs_release_path(path); | |
2290 | metadata = 0; | |
3173a18f | 2291 | |
d278850e JB |
2292 | key.objectid = head->bytenr; |
2293 | key.offset = head->num_bytes; | |
55994887 FDBM |
2294 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2295 | goto again; | |
2296 | } | |
2297 | } else { | |
2298 | err = -EIO; | |
2299 | goto out; | |
3173a18f | 2300 | } |
5d4f98a2 YZ |
2301 | } |
2302 | ||
2303 | leaf = path->nodes[0]; | |
2304 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 2305 | |
6d8ff4e4 | 2306 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
2307 | err = -EINVAL; |
2308 | btrfs_print_v0_err(fs_info); | |
2309 | btrfs_abort_transaction(trans, err); | |
2310 | goto out; | |
2311 | } | |
2312 | ||
5d4f98a2 YZ |
2313 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2314 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2315 | |
5d4f98a2 YZ |
2316 | btrfs_mark_buffer_dirty(leaf); |
2317 | out: | |
2318 | btrfs_free_path(path); | |
2319 | return err; | |
56bec294 CM |
2320 | } |
2321 | ||
5d4f98a2 | 2322 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2323 | struct btrfs_delayed_ref_node *node, |
2324 | struct btrfs_delayed_extent_op *extent_op, | |
2325 | int insert_reserved) | |
56bec294 CM |
2326 | { |
2327 | int ret = 0; | |
5d4f98a2 | 2328 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
2329 | u64 parent = 0; |
2330 | u64 ref_root = 0; | |
56bec294 | 2331 | |
5d4f98a2 | 2332 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 2333 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2334 | |
5d4f98a2 YZ |
2335 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2336 | parent = ref->parent; | |
fcebe456 | 2337 | ref_root = ref->root; |
5d4f98a2 | 2338 | |
02794222 | 2339 | if (node->ref_mod != 1) { |
f97806f2 | 2340 | btrfs_err(trans->fs_info, |
02794222 LB |
2341 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2342 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2343 | parent); | |
2344 | return -EIO; | |
2345 | } | |
5d4f98a2 | 2346 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2347 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 2348 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 2349 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2350 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2351 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 2352 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2353 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 2354 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
2355 | } else { |
2356 | BUG(); | |
2357 | } | |
56bec294 CM |
2358 | return ret; |
2359 | } | |
2360 | ||
2361 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2362 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2363 | struct btrfs_delayed_ref_node *node, |
2364 | struct btrfs_delayed_extent_op *extent_op, | |
2365 | int insert_reserved) | |
56bec294 | 2366 | { |
79787eaa JM |
2367 | int ret = 0; |
2368 | ||
857cc2fc JB |
2369 | if (trans->aborted) { |
2370 | if (insert_reserved) | |
5fac7f9e | 2371 | btrfs_pin_extent(trans->fs_info, node->bytenr, |
857cc2fc | 2372 | node->num_bytes, 1); |
79787eaa | 2373 | return 0; |
857cc2fc | 2374 | } |
79787eaa | 2375 | |
5d4f98a2 YZ |
2376 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2377 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 2378 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2379 | insert_reserved); |
2380 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2381 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 2382 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2383 | insert_reserved); |
2384 | else | |
2385 | BUG(); | |
80ee54bf JB |
2386 | if (ret && insert_reserved) |
2387 | btrfs_pin_extent(trans->fs_info, node->bytenr, | |
2388 | node->num_bytes, 1); | |
5d4f98a2 | 2389 | return ret; |
56bec294 CM |
2390 | } |
2391 | ||
c6fc2454 | 2392 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2393 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2394 | { | |
cffc3374 FM |
2395 | struct btrfs_delayed_ref_node *ref; |
2396 | ||
e3d03965 | 2397 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 2398 | return NULL; |
d7df2c79 | 2399 | |
cffc3374 FM |
2400 | /* |
2401 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2402 | * This is to prevent a ref count from going down to zero, which deletes | |
2403 | * the extent item from the extent tree, when there still are references | |
2404 | * to add, which would fail because they would not find the extent item. | |
2405 | */ | |
1d57ee94 WX |
2406 | if (!list_empty(&head->ref_add_list)) |
2407 | return list_first_entry(&head->ref_add_list, | |
2408 | struct btrfs_delayed_ref_node, add_list); | |
2409 | ||
e3d03965 | 2410 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 2411 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
2412 | ASSERT(list_empty(&ref->add_list)); |
2413 | return ref; | |
56bec294 CM |
2414 | } |
2415 | ||
2eadaa22 JB |
2416 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2417 | struct btrfs_delayed_ref_head *head) | |
2418 | { | |
2419 | spin_lock(&delayed_refs->lock); | |
2420 | head->processing = 0; | |
2421 | delayed_refs->num_heads_ready++; | |
2422 | spin_unlock(&delayed_refs->lock); | |
2423 | btrfs_delayed_ref_unlock(head); | |
2424 | } | |
2425 | ||
bedc6617 JB |
2426 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
2427 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
2428 | { |
2429 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
2430 | |
2431 | if (!extent_op) | |
bedc6617 JB |
2432 | return NULL; |
2433 | ||
b00e6250 | 2434 | if (head->must_insert_reserved) { |
bedc6617 | 2435 | head->extent_op = NULL; |
b00e6250 | 2436 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 2437 | return NULL; |
b00e6250 | 2438 | } |
bedc6617 JB |
2439 | return extent_op; |
2440 | } | |
2441 | ||
2442 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
2443 | struct btrfs_delayed_ref_head *head) | |
2444 | { | |
2445 | struct btrfs_delayed_extent_op *extent_op; | |
2446 | int ret; | |
2447 | ||
2448 | extent_op = cleanup_extent_op(head); | |
2449 | if (!extent_op) | |
2450 | return 0; | |
2451 | head->extent_op = NULL; | |
b00e6250 | 2452 | spin_unlock(&head->lock); |
20b9a2d6 | 2453 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
2454 | btrfs_free_delayed_extent_op(extent_op); |
2455 | return ret ? ret : 1; | |
2456 | } | |
2457 | ||
31890da0 JB |
2458 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
2459 | struct btrfs_delayed_ref_root *delayed_refs, | |
2460 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 2461 | { |
ba2c4d4e | 2462 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 JB |
2463 | |
2464 | if (head->total_ref_mod < 0) { | |
2465 | struct btrfs_space_info *space_info; | |
2466 | u64 flags; | |
2467 | ||
2468 | if (head->is_data) | |
2469 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2470 | else if (head->is_system) | |
2471 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
2472 | else | |
2473 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2474 | space_info = __find_space_info(fs_info, flags); | |
2475 | ASSERT(space_info); | |
2476 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
2477 | -head->num_bytes, | |
2478 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
2479 | ||
ba2c4d4e JB |
2480 | /* |
2481 | * We had csum deletions accounted for in our delayed refs rsv, | |
2482 | * we need to drop the csum leaves for this update from our | |
2483 | * delayed_refs_rsv. | |
2484 | */ | |
07c47775 JB |
2485 | if (head->is_data) { |
2486 | spin_lock(&delayed_refs->lock); | |
2487 | delayed_refs->pending_csums -= head->num_bytes; | |
2488 | spin_unlock(&delayed_refs->lock); | |
ba2c4d4e JB |
2489 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, |
2490 | head->num_bytes); | |
07c47775 JB |
2491 | } |
2492 | } | |
2493 | ||
ba2c4d4e | 2494 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
2495 | } |
2496 | ||
194ab0bc | 2497 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
2498 | struct btrfs_delayed_ref_head *head) |
2499 | { | |
f9871edd NB |
2500 | |
2501 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
2502 | struct btrfs_delayed_ref_root *delayed_refs; |
2503 | int ret; | |
2504 | ||
2505 | delayed_refs = &trans->transaction->delayed_refs; | |
2506 | ||
bedc6617 | 2507 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
2508 | if (ret < 0) { |
2509 | unselect_delayed_ref_head(delayed_refs, head); | |
2510 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2511 | return ret; | |
2512 | } else if (ret) { | |
2513 | return ret; | |
2514 | } | |
2515 | ||
2516 | /* | |
2517 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2518 | * and then re-check to make sure nobody got added. | |
2519 | */ | |
2520 | spin_unlock(&head->lock); | |
2521 | spin_lock(&delayed_refs->lock); | |
2522 | spin_lock(&head->lock); | |
e3d03965 | 2523 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
2524 | spin_unlock(&head->lock); |
2525 | spin_unlock(&delayed_refs->lock); | |
2526 | return 1; | |
2527 | } | |
d7baffda | 2528 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 2529 | spin_unlock(&head->lock); |
1e7a1421 | 2530 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 2531 | |
c1103f7a | 2532 | if (head->must_insert_reserved) { |
d278850e JB |
2533 | btrfs_pin_extent(fs_info, head->bytenr, |
2534 | head->num_bytes, 1); | |
c1103f7a | 2535 | if (head->is_data) { |
d278850e JB |
2536 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2537 | head->num_bytes); | |
c1103f7a JB |
2538 | } |
2539 | } | |
2540 | ||
31890da0 | 2541 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
2542 | |
2543 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 2544 | btrfs_delayed_ref_unlock(head); |
d278850e | 2545 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2546 | return 0; |
2547 | } | |
2548 | ||
b1cdbcb5 NB |
2549 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
2550 | struct btrfs_trans_handle *trans) | |
2551 | { | |
2552 | struct btrfs_delayed_ref_root *delayed_refs = | |
2553 | &trans->transaction->delayed_refs; | |
2554 | struct btrfs_delayed_ref_head *head = NULL; | |
2555 | int ret; | |
2556 | ||
2557 | spin_lock(&delayed_refs->lock); | |
5637c74b | 2558 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
2559 | if (!head) { |
2560 | spin_unlock(&delayed_refs->lock); | |
2561 | return head; | |
2562 | } | |
2563 | ||
2564 | /* | |
2565 | * Grab the lock that says we are going to process all the refs for | |
2566 | * this head | |
2567 | */ | |
9e920a6f | 2568 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
2569 | spin_unlock(&delayed_refs->lock); |
2570 | ||
2571 | /* | |
2572 | * We may have dropped the spin lock to get the head mutex lock, and | |
2573 | * that might have given someone else time to free the head. If that's | |
2574 | * true, it has been removed from our list and we can move on. | |
2575 | */ | |
2576 | if (ret == -EAGAIN) | |
2577 | head = ERR_PTR(-EAGAIN); | |
2578 | ||
2579 | return head; | |
2580 | } | |
2581 | ||
e7261386 NB |
2582 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
2583 | struct btrfs_delayed_ref_head *locked_ref, | |
2584 | unsigned long *run_refs) | |
2585 | { | |
2586 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2587 | struct btrfs_delayed_ref_root *delayed_refs; | |
2588 | struct btrfs_delayed_extent_op *extent_op; | |
2589 | struct btrfs_delayed_ref_node *ref; | |
2590 | int must_insert_reserved = 0; | |
2591 | int ret; | |
2592 | ||
2593 | delayed_refs = &trans->transaction->delayed_refs; | |
2594 | ||
0110a4c4 NB |
2595 | lockdep_assert_held(&locked_ref->mutex); |
2596 | lockdep_assert_held(&locked_ref->lock); | |
2597 | ||
e7261386 NB |
2598 | while ((ref = select_delayed_ref(locked_ref))) { |
2599 | if (ref->seq && | |
2600 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
2601 | spin_unlock(&locked_ref->lock); | |
2602 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2603 | return -EAGAIN; | |
2604 | } | |
2605 | ||
2606 | (*run_refs)++; | |
2607 | ref->in_tree = 0; | |
2608 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
2609 | RB_CLEAR_NODE(&ref->ref_node); | |
2610 | if (!list_empty(&ref->add_list)) | |
2611 | list_del(&ref->add_list); | |
2612 | /* | |
2613 | * When we play the delayed ref, also correct the ref_mod on | |
2614 | * head | |
2615 | */ | |
2616 | switch (ref->action) { | |
2617 | case BTRFS_ADD_DELAYED_REF: | |
2618 | case BTRFS_ADD_DELAYED_EXTENT: | |
2619 | locked_ref->ref_mod -= ref->ref_mod; | |
2620 | break; | |
2621 | case BTRFS_DROP_DELAYED_REF: | |
2622 | locked_ref->ref_mod += ref->ref_mod; | |
2623 | break; | |
2624 | default: | |
2625 | WARN_ON(1); | |
2626 | } | |
2627 | atomic_dec(&delayed_refs->num_entries); | |
2628 | ||
2629 | /* | |
2630 | * Record the must_insert_reserved flag before we drop the | |
2631 | * spin lock. | |
2632 | */ | |
2633 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2634 | locked_ref->must_insert_reserved = 0; | |
2635 | ||
2636 | extent_op = locked_ref->extent_op; | |
2637 | locked_ref->extent_op = NULL; | |
2638 | spin_unlock(&locked_ref->lock); | |
2639 | ||
2640 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
2641 | must_insert_reserved); | |
2642 | ||
2643 | btrfs_free_delayed_extent_op(extent_op); | |
2644 | if (ret) { | |
2645 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2646 | btrfs_put_delayed_ref(ref); | |
2647 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
2648 | ret); | |
2649 | return ret; | |
2650 | } | |
2651 | ||
2652 | btrfs_put_delayed_ref(ref); | |
2653 | cond_resched(); | |
2654 | ||
2655 | spin_lock(&locked_ref->lock); | |
2656 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
2657 | } | |
2658 | ||
2659 | return 0; | |
2660 | } | |
2661 | ||
79787eaa JM |
2662 | /* |
2663 | * Returns 0 on success or if called with an already aborted transaction. | |
2664 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2665 | */ | |
d7df2c79 | 2666 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2667 | unsigned long nr) |
56bec294 | 2668 | { |
0a1e458a | 2669 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2670 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 2671 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 2672 | ktime_t start = ktime_get(); |
56bec294 | 2673 | int ret; |
d7df2c79 | 2674 | unsigned long count = 0; |
0a2b2a84 | 2675 | unsigned long actual_count = 0; |
56bec294 CM |
2676 | |
2677 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 2678 | do { |
56bec294 | 2679 | if (!locked_ref) { |
b1cdbcb5 | 2680 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2681 | if (IS_ERR_OR_NULL(locked_ref)) { |
2682 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2683 | continue; | |
2684 | } else { | |
2685 | break; | |
2686 | } | |
56bec294 | 2687 | } |
0110a4c4 | 2688 | count++; |
56bec294 | 2689 | } |
2c3cf7d5 FM |
2690 | /* |
2691 | * We need to try and merge add/drops of the same ref since we | |
2692 | * can run into issues with relocate dropping the implicit ref | |
2693 | * and then it being added back again before the drop can | |
2694 | * finish. If we merged anything we need to re-loop so we can | |
2695 | * get a good ref. | |
2696 | * Or we can get node references of the same type that weren't | |
2697 | * merged when created due to bumps in the tree mod seq, and | |
2698 | * we need to merge them to prevent adding an inline extent | |
2699 | * backref before dropping it (triggering a BUG_ON at | |
2700 | * insert_inline_extent_backref()). | |
2701 | */ | |
d7df2c79 | 2702 | spin_lock(&locked_ref->lock); |
be97f133 | 2703 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2704 | |
0110a4c4 NB |
2705 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2706 | &actual_count); | |
2707 | if (ret < 0 && ret != -EAGAIN) { | |
2708 | /* | |
2709 | * Error, btrfs_run_delayed_refs_for_head already | |
2710 | * unlocked everything so just bail out | |
2711 | */ | |
2712 | return ret; | |
2713 | } else if (!ret) { | |
2714 | /* | |
2715 | * Success, perform the usual cleanup of a processed | |
2716 | * head | |
2717 | */ | |
f9871edd | 2718 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2719 | if (ret > 0 ) { |
b00e6250 JB |
2720 | /* We dropped our lock, we need to loop. */ |
2721 | ret = 0; | |
d7df2c79 | 2722 | continue; |
194ab0bc JB |
2723 | } else if (ret) { |
2724 | return ret; | |
5d4f98a2 | 2725 | } |
22cd2e7d | 2726 | } |
1ce7a5ec | 2727 | |
b00e6250 | 2728 | /* |
0110a4c4 NB |
2729 | * Either success case or btrfs_run_delayed_refs_for_head |
2730 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2731 | */ |
b00e6250 | 2732 | |
0110a4c4 | 2733 | locked_ref = NULL; |
c3e69d58 | 2734 | cond_resched(); |
0110a4c4 | 2735 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2736 | |
2737 | /* | |
2738 | * We don't want to include ref heads since we can have empty ref heads | |
2739 | * and those will drastically skew our runtime down since we just do | |
2740 | * accounting, no actual extent tree updates. | |
2741 | */ | |
2742 | if (actual_count > 0) { | |
2743 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2744 | u64 avg; | |
2745 | ||
2746 | /* | |
2747 | * We weigh the current average higher than our current runtime | |
2748 | * to avoid large swings in the average. | |
2749 | */ | |
2750 | spin_lock(&delayed_refs->lock); | |
2751 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2752 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2753 | spin_unlock(&delayed_refs->lock); |
2754 | } | |
d7df2c79 | 2755 | return 0; |
c3e69d58 CM |
2756 | } |
2757 | ||
709c0486 AJ |
2758 | #ifdef SCRAMBLE_DELAYED_REFS |
2759 | /* | |
2760 | * Normally delayed refs get processed in ascending bytenr order. This | |
2761 | * correlates in most cases to the order added. To expose dependencies on this | |
2762 | * order, we start to process the tree in the middle instead of the beginning | |
2763 | */ | |
2764 | static u64 find_middle(struct rb_root *root) | |
2765 | { | |
2766 | struct rb_node *n = root->rb_node; | |
2767 | struct btrfs_delayed_ref_node *entry; | |
2768 | int alt = 1; | |
2769 | u64 middle; | |
2770 | u64 first = 0, last = 0; | |
2771 | ||
2772 | n = rb_first(root); | |
2773 | if (n) { | |
2774 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2775 | first = entry->bytenr; | |
2776 | } | |
2777 | n = rb_last(root); | |
2778 | if (n) { | |
2779 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2780 | last = entry->bytenr; | |
2781 | } | |
2782 | n = root->rb_node; | |
2783 | ||
2784 | while (n) { | |
2785 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2786 | WARN_ON(!entry->in_tree); | |
2787 | ||
2788 | middle = entry->bytenr; | |
2789 | ||
2790 | if (alt) | |
2791 | n = n->rb_left; | |
2792 | else | |
2793 | n = n->rb_right; | |
2794 | ||
2795 | alt = 1 - alt; | |
2796 | } | |
2797 | return middle; | |
2798 | } | |
2799 | #endif | |
2800 | ||
2ff7e61e | 2801 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2802 | { |
2803 | u64 num_bytes; | |
2804 | ||
2805 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2806 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2807 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2808 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2809 | ||
2810 | /* | |
2811 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2812 | * closer to what we're really going to want to use. |
1be41b78 | 2813 | */ |
0b246afa | 2814 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2815 | } |
2816 | ||
1262133b JB |
2817 | /* |
2818 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2819 | * would require to store the csums for that many bytes. | |
2820 | */ | |
2ff7e61e | 2821 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2822 | { |
2823 | u64 csum_size; | |
2824 | u64 num_csums_per_leaf; | |
2825 | u64 num_csums; | |
2826 | ||
0b246afa | 2827 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2828 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2829 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2830 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2831 | num_csums += num_csums_per_leaf - 1; |
2832 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2833 | return num_csums; | |
2834 | } | |
2835 | ||
64403612 | 2836 | bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info) |
1be41b78 | 2837 | { |
64403612 JB |
2838 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
2839 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
2840 | bool ret = false; | |
2841 | u64 reserved; | |
1be41b78 | 2842 | |
64403612 JB |
2843 | spin_lock(&global_rsv->lock); |
2844 | reserved = global_rsv->reserved; | |
2845 | spin_unlock(&global_rsv->lock); | |
1be41b78 JB |
2846 | |
2847 | /* | |
64403612 JB |
2848 | * Since the global reserve is just kind of magic we don't really want |
2849 | * to rely on it to save our bacon, so if our size is more than the | |
2850 | * delayed_refs_rsv and the global rsv then it's time to think about | |
2851 | * bailing. | |
1be41b78 | 2852 | */ |
64403612 JB |
2853 | spin_lock(&delayed_refs_rsv->lock); |
2854 | reserved += delayed_refs_rsv->reserved; | |
2855 | if (delayed_refs_rsv->size >= reserved) | |
2856 | ret = true; | |
2857 | spin_unlock(&delayed_refs_rsv->lock); | |
1be41b78 JB |
2858 | return ret; |
2859 | } | |
2860 | ||
7c861627 | 2861 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans) |
0a2b2a84 | 2862 | { |
0a2b2a84 JB |
2863 | u64 num_entries = |
2864 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2865 | u64 avg_runtime; | |
a79b7d4b | 2866 | u64 val; |
0a2b2a84 JB |
2867 | |
2868 | smp_mb(); | |
7c861627 | 2869 | avg_runtime = trans->fs_info->avg_delayed_ref_runtime; |
a79b7d4b | 2870 | val = num_entries * avg_runtime; |
dc1a90c6 | 2871 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2872 | return 1; |
a79b7d4b CM |
2873 | if (val >= NSEC_PER_SEC / 2) |
2874 | return 2; | |
0a2b2a84 | 2875 | |
64403612 | 2876 | return btrfs_check_space_for_delayed_refs(trans->fs_info); |
0a2b2a84 JB |
2877 | } |
2878 | ||
c3e69d58 CM |
2879 | /* |
2880 | * this starts processing the delayed reference count updates and | |
2881 | * extent insertions we have queued up so far. count can be | |
2882 | * 0, which means to process everything in the tree at the start | |
2883 | * of the run (but not newly added entries), or it can be some target | |
2884 | * number you'd like to process. | |
79787eaa JM |
2885 | * |
2886 | * Returns 0 on success or if called with an aborted transaction | |
2887 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2888 | */ |
2889 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2890 | unsigned long count) |
c3e69d58 | 2891 | { |
c79a70b1 | 2892 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2893 | struct rb_node *node; |
2894 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2895 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2896 | int ret; |
2897 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2898 | |
79787eaa JM |
2899 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2900 | if (trans->aborted) | |
2901 | return 0; | |
2902 | ||
0b246afa | 2903 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2904 | return 0; |
2905 | ||
c3e69d58 | 2906 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2907 | if (count == 0) |
d7df2c79 | 2908 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2909 | |
c3e69d58 | 2910 | again: |
709c0486 AJ |
2911 | #ifdef SCRAMBLE_DELAYED_REFS |
2912 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2913 | #endif | |
0a1e458a | 2914 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2915 | if (ret < 0) { |
66642832 | 2916 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2917 | return ret; |
eb099670 | 2918 | } |
c3e69d58 | 2919 | |
56bec294 | 2920 | if (run_all) { |
119e80df | 2921 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2922 | |
d7df2c79 | 2923 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2924 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2925 | if (!node) { |
2926 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2927 | goto out; |
d7df2c79 | 2928 | } |
d278850e JB |
2929 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2930 | href_node); | |
2931 | refcount_inc(&head->refs); | |
2932 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2933 | |
d278850e JB |
2934 | /* Mutex was contended, block until it's released and retry. */ |
2935 | mutex_lock(&head->mutex); | |
2936 | mutex_unlock(&head->mutex); | |
56bec294 | 2937 | |
d278850e | 2938 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2939 | cond_resched(); |
56bec294 | 2940 | goto again; |
5f39d397 | 2941 | } |
54aa1f4d | 2942 | out: |
a28ec197 CM |
2943 | return 0; |
2944 | } | |
2945 | ||
5d4f98a2 | 2946 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 2947 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 2948 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 2949 | int level, int is_data) |
5d4f98a2 YZ |
2950 | { |
2951 | struct btrfs_delayed_extent_op *extent_op; | |
2952 | int ret; | |
2953 | ||
78a6184a | 2954 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2955 | if (!extent_op) |
2956 | return -ENOMEM; | |
2957 | ||
2958 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2959 | extent_op->update_flags = true; |
2960 | extent_op->update_key = false; | |
2961 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2962 | extent_op->level = level; |
5d4f98a2 | 2963 | |
0b246afa | 2964 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 2965 | num_bytes, extent_op); |
5d4f98a2 | 2966 | if (ret) |
78a6184a | 2967 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2968 | return ret; |
2969 | } | |
2970 | ||
e4c3b2dc | 2971 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2972 | struct btrfs_path *path, |
2973 | u64 objectid, u64 offset, u64 bytenr) | |
2974 | { | |
2975 | struct btrfs_delayed_ref_head *head; | |
2976 | struct btrfs_delayed_ref_node *ref; | |
2977 | struct btrfs_delayed_data_ref *data_ref; | |
2978 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2979 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2980 | struct rb_node *node; |
5d4f98a2 YZ |
2981 | int ret = 0; |
2982 | ||
998ac6d2 | 2983 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2984 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2985 | if (cur_trans) |
2986 | refcount_inc(&cur_trans->use_count); | |
2987 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2988 | if (!cur_trans) |
2989 | return 0; | |
2990 | ||
2991 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2992 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2993 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2994 | if (!head) { |
2995 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2996 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2997 | return 0; |
2998 | } | |
5d4f98a2 YZ |
2999 | |
3000 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 3001 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
3002 | spin_unlock(&delayed_refs->lock); |
3003 | ||
b3b4aa74 | 3004 | btrfs_release_path(path); |
5d4f98a2 | 3005 | |
8cc33e5c DS |
3006 | /* |
3007 | * Mutex was contended, block until it's released and let | |
3008 | * caller try again | |
3009 | */ | |
5d4f98a2 YZ |
3010 | mutex_lock(&head->mutex); |
3011 | mutex_unlock(&head->mutex); | |
d278850e | 3012 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 3013 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3014 | return -EAGAIN; |
3015 | } | |
d7df2c79 | 3016 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3017 | |
d7df2c79 | 3018 | spin_lock(&head->lock); |
0e0adbcf JB |
3019 | /* |
3020 | * XXX: We should replace this with a proper search function in the | |
3021 | * future. | |
3022 | */ | |
e3d03965 LB |
3023 | for (node = rb_first_cached(&head->ref_tree); node; |
3024 | node = rb_next(node)) { | |
0e0adbcf | 3025 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
3026 | /* If it's a shared ref we know a cross reference exists */ |
3027 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3028 | ret = 1; | |
3029 | break; | |
3030 | } | |
5d4f98a2 | 3031 | |
d7df2c79 | 3032 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3033 | |
d7df2c79 JB |
3034 | /* |
3035 | * If our ref doesn't match the one we're currently looking at | |
3036 | * then we have a cross reference. | |
3037 | */ | |
3038 | if (data_ref->root != root->root_key.objectid || | |
3039 | data_ref->objectid != objectid || | |
3040 | data_ref->offset != offset) { | |
3041 | ret = 1; | |
3042 | break; | |
3043 | } | |
5d4f98a2 | 3044 | } |
d7df2c79 | 3045 | spin_unlock(&head->lock); |
5d4f98a2 | 3046 | mutex_unlock(&head->mutex); |
998ac6d2 | 3047 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3048 | return ret; |
3049 | } | |
3050 | ||
e4c3b2dc | 3051 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3052 | struct btrfs_path *path, |
3053 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3054 | { |
0b246afa JM |
3055 | struct btrfs_fs_info *fs_info = root->fs_info; |
3056 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3057 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3058 | struct btrfs_extent_data_ref *ref; |
3059 | struct btrfs_extent_inline_ref *iref; | |
3060 | struct btrfs_extent_item *ei; | |
f321e491 | 3061 | struct btrfs_key key; |
5d4f98a2 | 3062 | u32 item_size; |
3de28d57 | 3063 | int type; |
be20aa9d | 3064 | int ret; |
925baedd | 3065 | |
be20aa9d | 3066 | key.objectid = bytenr; |
31840ae1 | 3067 | key.offset = (u64)-1; |
f321e491 | 3068 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3069 | |
be20aa9d CM |
3070 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3071 | if (ret < 0) | |
3072 | goto out; | |
79787eaa | 3073 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3074 | |
3075 | ret = -ENOENT; | |
3076 | if (path->slots[0] == 0) | |
31840ae1 | 3077 | goto out; |
be20aa9d | 3078 | |
31840ae1 | 3079 | path->slots[0]--; |
f321e491 | 3080 | leaf = path->nodes[0]; |
5d4f98a2 | 3081 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3082 | |
5d4f98a2 | 3083 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3084 | goto out; |
f321e491 | 3085 | |
5d4f98a2 YZ |
3086 | ret = 1; |
3087 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 3088 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 3089 | |
5d4f98a2 YZ |
3090 | if (item_size != sizeof(*ei) + |
3091 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3092 | goto out; | |
be20aa9d | 3093 | |
5d4f98a2 YZ |
3094 | if (btrfs_extent_generation(leaf, ei) <= |
3095 | btrfs_root_last_snapshot(&root->root_item)) | |
3096 | goto out; | |
3097 | ||
3098 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3099 | |
3100 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3101 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3102 | goto out; |
3103 | ||
3104 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3105 | if (btrfs_extent_refs(leaf, ei) != | |
3106 | btrfs_extent_data_ref_count(leaf, ref) || | |
3107 | btrfs_extent_data_ref_root(leaf, ref) != | |
3108 | root->root_key.objectid || | |
3109 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3110 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3111 | goto out; | |
3112 | ||
3113 | ret = 0; | |
3114 | out: | |
3115 | return ret; | |
3116 | } | |
3117 | ||
e4c3b2dc LB |
3118 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3119 | u64 bytenr) | |
5d4f98a2 YZ |
3120 | { |
3121 | struct btrfs_path *path; | |
3122 | int ret; | |
5d4f98a2 YZ |
3123 | |
3124 | path = btrfs_alloc_path(); | |
3125 | if (!path) | |
9132c4ff | 3126 | return -ENOMEM; |
5d4f98a2 YZ |
3127 | |
3128 | do { | |
e4c3b2dc | 3129 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3130 | offset, bytenr); |
3131 | if (ret && ret != -ENOENT) | |
f321e491 | 3132 | goto out; |
80ff3856 | 3133 | |
380fd066 MT |
3134 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
3135 | } while (ret == -EAGAIN); | |
5d4f98a2 | 3136 | |
be20aa9d | 3137 | out: |
80ff3856 | 3138 | btrfs_free_path(path); |
f0486c68 YZ |
3139 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3140 | WARN_ON(ret > 0); | |
f321e491 | 3141 | return ret; |
be20aa9d | 3142 | } |
c5739bba | 3143 | |
5d4f98a2 | 3144 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3145 | struct btrfs_root *root, |
5d4f98a2 | 3146 | struct extent_buffer *buf, |
e339a6b0 | 3147 | int full_backref, int inc) |
31840ae1 | 3148 | { |
0b246afa | 3149 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3150 | u64 bytenr; |
5d4f98a2 YZ |
3151 | u64 num_bytes; |
3152 | u64 parent; | |
31840ae1 | 3153 | u64 ref_root; |
31840ae1 | 3154 | u32 nritems; |
31840ae1 ZY |
3155 | struct btrfs_key key; |
3156 | struct btrfs_file_extent_item *fi; | |
3157 | int i; | |
3158 | int level; | |
3159 | int ret = 0; | |
fccb84c9 | 3160 | |
0b246afa | 3161 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3162 | return 0; |
fccb84c9 | 3163 | |
31840ae1 | 3164 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3165 | nritems = btrfs_header_nritems(buf); |
3166 | level = btrfs_header_level(buf); | |
3167 | ||
27cdeb70 | 3168 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3169 | return 0; |
31840ae1 | 3170 | |
5d4f98a2 YZ |
3171 | if (full_backref) |
3172 | parent = buf->start; | |
3173 | else | |
3174 | parent = 0; | |
3175 | ||
3176 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3177 | if (level == 0) { |
5d4f98a2 | 3178 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3179 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3180 | continue; |
5d4f98a2 | 3181 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3182 | struct btrfs_file_extent_item); |
3183 | if (btrfs_file_extent_type(buf, fi) == | |
3184 | BTRFS_FILE_EXTENT_INLINE) | |
3185 | continue; | |
3186 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3187 | if (bytenr == 0) | |
3188 | continue; | |
5d4f98a2 YZ |
3189 | |
3190 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3191 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
dd28b6a5 QW |
3192 | if (inc) |
3193 | ret = btrfs_inc_extent_ref(trans, root, bytenr, | |
3194 | num_bytes, parent, ref_root, | |
3195 | key.objectid, key.offset); | |
3196 | else | |
3197 | ret = btrfs_free_extent(trans, root, bytenr, | |
3198 | num_bytes, parent, ref_root, | |
3199 | key.objectid, key.offset); | |
31840ae1 ZY |
3200 | if (ret) |
3201 | goto fail; | |
3202 | } else { | |
5d4f98a2 | 3203 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3204 | num_bytes = fs_info->nodesize; |
dd28b6a5 QW |
3205 | if (inc) |
3206 | ret = btrfs_inc_extent_ref(trans, root, bytenr, | |
3207 | num_bytes, parent, ref_root, | |
3208 | level - 1, 0); | |
3209 | else | |
3210 | ret = btrfs_free_extent(trans, root, bytenr, | |
3211 | num_bytes, parent, ref_root, | |
3212 | level - 1, 0); | |
31840ae1 ZY |
3213 | if (ret) |
3214 | goto fail; | |
3215 | } | |
3216 | } | |
3217 | return 0; | |
3218 | fail: | |
5d4f98a2 YZ |
3219 | return ret; |
3220 | } | |
3221 | ||
3222 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3223 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3224 | { |
e339a6b0 | 3225 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3226 | } |
3227 | ||
3228 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3229 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3230 | { |
e339a6b0 | 3231 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3232 | } |
3233 | ||
9078a3e1 | 3234 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
9078a3e1 CM |
3235 | struct btrfs_path *path, |
3236 | struct btrfs_block_group_cache *cache) | |
3237 | { | |
39db232d | 3238 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9078a3e1 | 3239 | int ret; |
0b246afa | 3240 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3241 | unsigned long bi; |
3242 | struct extent_buffer *leaf; | |
9078a3e1 | 3243 | |
9078a3e1 | 3244 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3245 | if (ret) { |
3246 | if (ret > 0) | |
3247 | ret = -ENOENT; | |
54aa1f4d | 3248 | goto fail; |
df95e7f0 | 3249 | } |
5f39d397 CM |
3250 | |
3251 | leaf = path->nodes[0]; | |
3252 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3253 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3254 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3255 | fail: |
24b89d08 | 3256 | btrfs_release_path(path); |
df95e7f0 | 3257 | return ret; |
9078a3e1 CM |
3258 | |
3259 | } | |
3260 | ||
f87b7eb8 DS |
3261 | static struct btrfs_block_group_cache *next_block_group( |
3262 | struct btrfs_block_group_cache *cache) | |
4a8c9a62 | 3263 | { |
f87b7eb8 | 3264 | struct btrfs_fs_info *fs_info = cache->fs_info; |
4a8c9a62 | 3265 | struct rb_node *node; |
292cbd51 | 3266 | |
0b246afa | 3267 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3268 | |
3269 | /* If our block group was removed, we need a full search. */ | |
3270 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3271 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3272 | ||
0b246afa | 3273 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3274 | btrfs_put_block_group(cache); |
0b246afa | 3275 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3276 | } |
4a8c9a62 YZ |
3277 | node = rb_next(&cache->cache_node); |
3278 | btrfs_put_block_group(cache); | |
3279 | if (node) { | |
3280 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3281 | cache_node); | |
11dfe35a | 3282 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3283 | } else |
3284 | cache = NULL; | |
0b246afa | 3285 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3286 | return cache; |
3287 | } | |
3288 | ||
0af3d00b JB |
3289 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3290 | struct btrfs_trans_handle *trans, | |
3291 | struct btrfs_path *path) | |
3292 | { | |
0b246afa JM |
3293 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3294 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3295 | struct inode *inode = NULL; |
364ecf36 | 3296 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3297 | u64 alloc_hint = 0; |
2b20982e | 3298 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3299 | u64 num_pages = 0; |
0af3d00b JB |
3300 | int retries = 0; |
3301 | int ret = 0; | |
3302 | ||
3303 | /* | |
3304 | * If this block group is smaller than 100 megs don't bother caching the | |
3305 | * block group. | |
3306 | */ | |
ee22184b | 3307 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3308 | spin_lock(&block_group->lock); |
3309 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3310 | spin_unlock(&block_group->lock); | |
3311 | return 0; | |
3312 | } | |
3313 | ||
0c0ef4bc JB |
3314 | if (trans->aborted) |
3315 | return 0; | |
0af3d00b | 3316 | again: |
7949f339 | 3317 | inode = lookup_free_space_inode(block_group, path); |
0af3d00b JB |
3318 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3319 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3320 | btrfs_release_path(path); |
0af3d00b JB |
3321 | goto out; |
3322 | } | |
3323 | ||
3324 | if (IS_ERR(inode)) { | |
3325 | BUG_ON(retries); | |
3326 | retries++; | |
3327 | ||
3328 | if (block_group->ro) | |
3329 | goto out_free; | |
3330 | ||
4ca75f1b | 3331 | ret = create_free_space_inode(trans, block_group, path); |
0af3d00b JB |
3332 | if (ret) |
3333 | goto out_free; | |
3334 | goto again; | |
3335 | } | |
3336 | ||
3337 | /* | |
3338 | * We want to set the generation to 0, that way if anything goes wrong | |
3339 | * from here on out we know not to trust this cache when we load up next | |
3340 | * time. | |
3341 | */ | |
3342 | BTRFS_I(inode)->generation = 0; | |
3343 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3344 | if (ret) { |
3345 | /* | |
3346 | * So theoretically we could recover from this, simply set the | |
3347 | * super cache generation to 0 so we know to invalidate the | |
3348 | * cache, but then we'd have to keep track of the block groups | |
3349 | * that fail this way so we know we _have_ to reset this cache | |
3350 | * before the next commit or risk reading stale cache. So to | |
3351 | * limit our exposure to horrible edge cases lets just abort the | |
3352 | * transaction, this only happens in really bad situations | |
3353 | * anyway. | |
3354 | */ | |
66642832 | 3355 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3356 | goto out_put; |
3357 | } | |
0af3d00b JB |
3358 | WARN_ON(ret); |
3359 | ||
8e138e0d JB |
3360 | /* We've already setup this transaction, go ahead and exit */ |
3361 | if (block_group->cache_generation == trans->transid && | |
3362 | i_size_read(inode)) { | |
3363 | dcs = BTRFS_DC_SETUP; | |
3364 | goto out_put; | |
3365 | } | |
3366 | ||
0af3d00b | 3367 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3368 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3369 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3370 | if (ret) |
3371 | goto out_put; | |
3372 | ||
77ab86bf | 3373 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3374 | if (ret) |
3375 | goto out_put; | |
3376 | } | |
3377 | ||
3378 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3379 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3380 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3381 | /* |
3382 | * don't bother trying to write stuff out _if_ | |
3383 | * a) we're not cached, | |
1a79c1f2 LB |
3384 | * b) we're with nospace_cache mount option, |
3385 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3386 | */ |
2b20982e | 3387 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3388 | spin_unlock(&block_group->lock); |
3389 | goto out_put; | |
3390 | } | |
3391 | spin_unlock(&block_group->lock); | |
3392 | ||
2968b1f4 JB |
3393 | /* |
3394 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3395 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3396 | */ | |
3397 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3398 | ret = -ENOSPC; | |
3399 | goto out_put; | |
3400 | } | |
3401 | ||
6fc823b1 JB |
3402 | /* |
3403 | * Try to preallocate enough space based on how big the block group is. | |
3404 | * Keep in mind this has to include any pinned space which could end up | |
3405 | * taking up quite a bit since it's not folded into the other space | |
3406 | * cache. | |
3407 | */ | |
ee22184b | 3408 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3409 | if (!num_pages) |
3410 | num_pages = 1; | |
3411 | ||
0af3d00b | 3412 | num_pages *= 16; |
09cbfeaf | 3413 | num_pages *= PAGE_SIZE; |
0af3d00b | 3414 | |
364ecf36 | 3415 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3416 | if (ret) |
3417 | goto out_put; | |
3418 | ||
3419 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3420 | num_pages, num_pages, | |
3421 | &alloc_hint); | |
2968b1f4 JB |
3422 | /* |
3423 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3424 | * of metadata or split extents when writing the cache out, which means | |
3425 | * we can enospc if we are heavily fragmented in addition to just normal | |
3426 | * out of space conditions. So if we hit this just skip setting up any | |
3427 | * other block groups for this transaction, maybe we'll unpin enough | |
3428 | * space the next time around. | |
3429 | */ | |
2b20982e JB |
3430 | if (!ret) |
3431 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3432 | else if (ret == -ENOSPC) |
3433 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3434 | |
0af3d00b JB |
3435 | out_put: |
3436 | iput(inode); | |
3437 | out_free: | |
b3b4aa74 | 3438 | btrfs_release_path(path); |
0af3d00b JB |
3439 | out: |
3440 | spin_lock(&block_group->lock); | |
e65cbb94 | 3441 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3442 | block_group->cache_generation = trans->transid; |
2b20982e | 3443 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3444 | spin_unlock(&block_group->lock); |
3445 | ||
364ecf36 | 3446 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3447 | return ret; |
3448 | } | |
3449 | ||
bbebb3e0 | 3450 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) |
dcdf7f6d | 3451 | { |
bbebb3e0 | 3452 | struct btrfs_fs_info *fs_info = trans->fs_info; |
dcdf7f6d JB |
3453 | struct btrfs_block_group_cache *cache, *tmp; |
3454 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3455 | struct btrfs_path *path; | |
3456 | ||
3457 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3458 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3459 | return 0; |
3460 | ||
3461 | path = btrfs_alloc_path(); | |
3462 | if (!path) | |
3463 | return -ENOMEM; | |
3464 | ||
3465 | /* Could add new block groups, use _safe just in case */ | |
3466 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3467 | dirty_list) { | |
3468 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3469 | cache_save_setup(cache, trans, path); | |
3470 | } | |
3471 | ||
3472 | btrfs_free_path(path); | |
3473 | return 0; | |
3474 | } | |
3475 | ||
1bbc621e CM |
3476 | /* |
3477 | * transaction commit does final block group cache writeback during a | |
3478 | * critical section where nothing is allowed to change the FS. This is | |
3479 | * required in order for the cache to actually match the block group, | |
3480 | * but can introduce a lot of latency into the commit. | |
3481 | * | |
3482 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3483 | * cache IO. There's a chance we'll have to redo some of it if the | |
3484 | * block group changes again during the commit, but it greatly reduces | |
3485 | * the commit latency by getting rid of the easy block groups while | |
3486 | * we're still allowing others to join the commit. | |
3487 | */ | |
21217054 | 3488 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3489 | { |
21217054 | 3490 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3491 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3492 | struct btrfs_transaction *cur_trans = trans->transaction; |
3493 | int ret = 0; | |
c9dc4c65 | 3494 | int should_put; |
1bbc621e CM |
3495 | struct btrfs_path *path = NULL; |
3496 | LIST_HEAD(dirty); | |
3497 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3498 | int num_started = 0; |
1bbc621e CM |
3499 | int loops = 0; |
3500 | ||
3501 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3502 | if (list_empty(&cur_trans->dirty_bgs)) { |
3503 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3504 | return 0; | |
1bbc621e | 3505 | } |
b58d1a9e | 3506 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3507 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3508 | |
1bbc621e | 3509 | again: |
1bbc621e CM |
3510 | /* |
3511 | * make sure all the block groups on our dirty list actually | |
3512 | * exist | |
3513 | */ | |
6c686b35 | 3514 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3515 | |
3516 | if (!path) { | |
3517 | path = btrfs_alloc_path(); | |
3518 | if (!path) | |
3519 | return -ENOMEM; | |
3520 | } | |
3521 | ||
b58d1a9e FM |
3522 | /* |
3523 | * cache_write_mutex is here only to save us from balance or automatic | |
3524 | * removal of empty block groups deleting this block group while we are | |
3525 | * writing out the cache | |
3526 | */ | |
3527 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3528 | while (!list_empty(&dirty)) { |
ba2c4d4e JB |
3529 | bool drop_reserve = true; |
3530 | ||
1bbc621e CM |
3531 | cache = list_first_entry(&dirty, |
3532 | struct btrfs_block_group_cache, | |
3533 | dirty_list); | |
1bbc621e CM |
3534 | /* |
3535 | * this can happen if something re-dirties a block | |
3536 | * group that is already under IO. Just wait for it to | |
3537 | * finish and then do it all again | |
3538 | */ | |
3539 | if (!list_empty(&cache->io_list)) { | |
3540 | list_del_init(&cache->io_list); | |
afdb5718 | 3541 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3542 | btrfs_put_block_group(cache); |
3543 | } | |
3544 | ||
3545 | ||
3546 | /* | |
3547 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3548 | * if it should update the cache_state. Don't delete | |
3549 | * until after we wait. | |
3550 | * | |
3551 | * Since we're not running in the commit critical section | |
3552 | * we need the dirty_bgs_lock to protect from update_block_group | |
3553 | */ | |
3554 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3555 | list_del_init(&cache->dirty_list); | |
3556 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3557 | ||
3558 | should_put = 1; | |
3559 | ||
3560 | cache_save_setup(cache, trans, path); | |
3561 | ||
3562 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3563 | cache->io_ctl.inode = NULL; | |
fe041534 | 3564 | ret = btrfs_write_out_cache(trans, cache, path); |
1bbc621e CM |
3565 | if (ret == 0 && cache->io_ctl.inode) { |
3566 | num_started++; | |
3567 | should_put = 0; | |
3568 | ||
3569 | /* | |
45ae2c18 NB |
3570 | * The cache_write_mutex is protecting the |
3571 | * io_list, also refer to the definition of | |
3572 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3573 | */ |
3574 | list_add_tail(&cache->io_list, io); | |
3575 | } else { | |
3576 | /* | |
3577 | * if we failed to write the cache, the | |
3578 | * generation will be bad and life goes on | |
3579 | */ | |
3580 | ret = 0; | |
3581 | } | |
3582 | } | |
ff1f8250 | 3583 | if (!ret) { |
39db232d | 3584 | ret = write_one_cache_group(trans, path, cache); |
ff1f8250 FM |
3585 | /* |
3586 | * Our block group might still be attached to the list | |
3587 | * of new block groups in the transaction handle of some | |
3588 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3589 | * means its block group item isn't yet in the extent | |
3590 | * tree. If this happens ignore the error, as we will | |
3591 | * try again later in the critical section of the | |
3592 | * transaction commit. | |
3593 | */ | |
3594 | if (ret == -ENOENT) { | |
3595 | ret = 0; | |
3596 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3597 | if (list_empty(&cache->dirty_list)) { | |
3598 | list_add_tail(&cache->dirty_list, | |
3599 | &cur_trans->dirty_bgs); | |
3600 | btrfs_get_block_group(cache); | |
ba2c4d4e | 3601 | drop_reserve = false; |
ff1f8250 FM |
3602 | } |
3603 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3604 | } else if (ret) { | |
66642832 | 3605 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3606 | } |
3607 | } | |
1bbc621e | 3608 | |
52042d8e | 3609 | /* if it's not on the io list, we need to put the block group */ |
1bbc621e CM |
3610 | if (should_put) |
3611 | btrfs_put_block_group(cache); | |
ba2c4d4e JB |
3612 | if (drop_reserve) |
3613 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1bbc621e CM |
3614 | |
3615 | if (ret) | |
3616 | break; | |
b58d1a9e FM |
3617 | |
3618 | /* | |
3619 | * Avoid blocking other tasks for too long. It might even save | |
3620 | * us from writing caches for block groups that are going to be | |
3621 | * removed. | |
3622 | */ | |
3623 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3624 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3625 | } |
b58d1a9e | 3626 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3627 | |
3628 | /* | |
3629 | * go through delayed refs for all the stuff we've just kicked off | |
3630 | * and then loop back (just once) | |
3631 | */ | |
c79a70b1 | 3632 | ret = btrfs_run_delayed_refs(trans, 0); |
1bbc621e CM |
3633 | if (!ret && loops == 0) { |
3634 | loops++; | |
3635 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3636 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3637 | /* |
3638 | * dirty_bgs_lock protects us from concurrent block group | |
3639 | * deletes too (not just cache_write_mutex). | |
3640 | */ | |
3641 | if (!list_empty(&dirty)) { | |
3642 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3643 | goto again; | |
3644 | } | |
1bbc621e | 3645 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3646 | } else if (ret < 0) { |
2ff7e61e | 3647 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3648 | } |
3649 | ||
3650 | btrfs_free_path(path); | |
3651 | return ret; | |
3652 | } | |
3653 | ||
5742d15f | 3654 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) |
1bbc621e | 3655 | { |
5742d15f | 3656 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1bbc621e CM |
3657 | struct btrfs_block_group_cache *cache; |
3658 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3659 | int ret = 0; | |
3660 | int should_put; | |
3661 | struct btrfs_path *path; | |
3662 | struct list_head *io = &cur_trans->io_bgs; | |
3663 | int num_started = 0; | |
9078a3e1 CM |
3664 | |
3665 | path = btrfs_alloc_path(); | |
3666 | if (!path) | |
3667 | return -ENOMEM; | |
3668 | ||
ce93ec54 | 3669 | /* |
e44081ef FM |
3670 | * Even though we are in the critical section of the transaction commit, |
3671 | * we can still have concurrent tasks adding elements to this | |
3672 | * transaction's list of dirty block groups. These tasks correspond to | |
3673 | * endio free space workers started when writeback finishes for a | |
3674 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3675 | * allocate new block groups as a result of COWing nodes of the root | |
3676 | * tree when updating the free space inode. The writeback for the space | |
3677 | * caches is triggered by an earlier call to | |
3678 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3679 | * loop. | |
3680 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3681 | * delayed refs to make sure we have the best chance at doing this all |
3682 | * in one shot. | |
3683 | */ | |
e44081ef | 3684 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3685 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3686 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3687 | struct btrfs_block_group_cache, | |
3688 | dirty_list); | |
c9dc4c65 CM |
3689 | |
3690 | /* | |
3691 | * this can happen if cache_save_setup re-dirties a block | |
3692 | * group that is already under IO. Just wait for it to | |
3693 | * finish and then do it all again | |
3694 | */ | |
3695 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3696 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3697 | list_del_init(&cache->io_list); |
afdb5718 | 3698 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3699 | btrfs_put_block_group(cache); |
e44081ef | 3700 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3701 | } |
3702 | ||
1bbc621e CM |
3703 | /* |
3704 | * don't remove from the dirty list until after we've waited | |
3705 | * on any pending IO | |
3706 | */ | |
ce93ec54 | 3707 | list_del_init(&cache->dirty_list); |
e44081ef | 3708 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3709 | should_put = 1; |
3710 | ||
1bbc621e | 3711 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3712 | |
ce93ec54 | 3713 | if (!ret) |
c79a70b1 | 3714 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 3715 | (unsigned long) -1); |
c9dc4c65 CM |
3716 | |
3717 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3718 | cache->io_ctl.inode = NULL; | |
fe041534 | 3719 | ret = btrfs_write_out_cache(trans, cache, path); |
c9dc4c65 CM |
3720 | if (ret == 0 && cache->io_ctl.inode) { |
3721 | num_started++; | |
3722 | should_put = 0; | |
1bbc621e | 3723 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3724 | } else { |
3725 | /* | |
3726 | * if we failed to write the cache, the | |
3727 | * generation will be bad and life goes on | |
3728 | */ | |
3729 | ret = 0; | |
3730 | } | |
3731 | } | |
ff1f8250 | 3732 | if (!ret) { |
39db232d | 3733 | ret = write_one_cache_group(trans, path, cache); |
2bc0bb5f FM |
3734 | /* |
3735 | * One of the free space endio workers might have | |
3736 | * created a new block group while updating a free space | |
3737 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3738 | * and hasn't released its transaction handle yet, in | |
3739 | * which case the new block group is still attached to | |
3740 | * its transaction handle and its creation has not | |
3741 | * finished yet (no block group item in the extent tree | |
3742 | * yet, etc). If this is the case, wait for all free | |
3743 | * space endio workers to finish and retry. This is a | |
3744 | * a very rare case so no need for a more efficient and | |
3745 | * complex approach. | |
3746 | */ | |
3747 | if (ret == -ENOENT) { | |
3748 | wait_event(cur_trans->writer_wait, | |
3749 | atomic_read(&cur_trans->num_writers) == 1); | |
39db232d | 3750 | ret = write_one_cache_group(trans, path, cache); |
2bc0bb5f | 3751 | } |
ff1f8250 | 3752 | if (ret) |
66642832 | 3753 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3754 | } |
c9dc4c65 CM |
3755 | |
3756 | /* if its not on the io list, we need to put the block group */ | |
3757 | if (should_put) | |
3758 | btrfs_put_block_group(cache); | |
ba2c4d4e | 3759 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
e44081ef | 3760 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3761 | } |
e44081ef | 3762 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3763 | |
45ae2c18 NB |
3764 | /* |
3765 | * Refer to the definition of io_bgs member for details why it's safe | |
3766 | * to use it without any locking | |
3767 | */ | |
1bbc621e CM |
3768 | while (!list_empty(io)) { |
3769 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3770 | io_list); |
3771 | list_del_init(&cache->io_list); | |
afdb5718 | 3772 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3773 | btrfs_put_block_group(cache); |
3774 | } | |
3775 | ||
9078a3e1 | 3776 | btrfs_free_path(path); |
ce93ec54 | 3777 | return ret; |
9078a3e1 CM |
3778 | } |
3779 | ||
2ff7e61e | 3780 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3781 | { |
3782 | struct btrfs_block_group_cache *block_group; | |
3783 | int readonly = 0; | |
3784 | ||
0b246afa | 3785 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3786 | if (!block_group || block_group->ro) |
3787 | readonly = 1; | |
3788 | if (block_group) | |
fa9c0d79 | 3789 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3790 | return readonly; |
3791 | } | |
3792 | ||
f78c436c FM |
3793 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3794 | { | |
3795 | struct btrfs_block_group_cache *bg; | |
3796 | bool ret = true; | |
3797 | ||
3798 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3799 | if (!bg) | |
3800 | return false; | |
3801 | ||
3802 | spin_lock(&bg->lock); | |
3803 | if (bg->ro) | |
3804 | ret = false; | |
3805 | else | |
3806 | atomic_inc(&bg->nocow_writers); | |
3807 | spin_unlock(&bg->lock); | |
3808 | ||
3809 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3810 | if (!ret) | |
3811 | btrfs_put_block_group(bg); | |
3812 | ||
3813 | return ret; | |
3814 | ||
3815 | } | |
3816 | ||
3817 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3818 | { | |
3819 | struct btrfs_block_group_cache *bg; | |
3820 | ||
3821 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3822 | ASSERT(bg); | |
3823 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
4625956a | 3824 | wake_up_var(&bg->nocow_writers); |
f78c436c FM |
3825 | /* |
3826 | * Once for our lookup and once for the lookup done by a previous call | |
3827 | * to btrfs_inc_nocow_writers() | |
3828 | */ | |
3829 | btrfs_put_block_group(bg); | |
3830 | btrfs_put_block_group(bg); | |
3831 | } | |
3832 | ||
f78c436c FM |
3833 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
3834 | { | |
4625956a | 3835 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); |
f78c436c FM |
3836 | } |
3837 | ||
6ab0a202 JM |
3838 | static const char *alloc_name(u64 flags) |
3839 | { | |
3840 | switch (flags) { | |
3841 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3842 | return "mixed"; | |
3843 | case BTRFS_BLOCK_GROUP_METADATA: | |
3844 | return "metadata"; | |
3845 | case BTRFS_BLOCK_GROUP_DATA: | |
3846 | return "data"; | |
3847 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3848 | return "system"; | |
3849 | default: | |
3850 | WARN_ON(1); | |
3851 | return "invalid-combination"; | |
3852 | }; | |
3853 | } | |
3854 | ||
4ca61683 | 3855 | static int create_space_info(struct btrfs_fs_info *info, u64 flags) |
2be12ef7 NB |
3856 | { |
3857 | ||
3858 | struct btrfs_space_info *space_info; | |
3859 | int i; | |
3860 | int ret; | |
3861 | ||
3862 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
3863 | if (!space_info) | |
3864 | return -ENOMEM; | |
3865 | ||
3866 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
3867 | GFP_KERNEL); | |
3868 | if (ret) { | |
3869 | kfree(space_info); | |
3870 | return ret; | |
3871 | } | |
3872 | ||
3873 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
3874 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
3875 | init_rwsem(&space_info->groups_sem); | |
3876 | spin_lock_init(&space_info->lock); | |
3877 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3878 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
3879 | init_waitqueue_head(&space_info->wait); | |
3880 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
3881 | INIT_LIST_HEAD(&space_info->tickets); | |
3882 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
3883 | ||
3884 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
3885 | info->space_info_kobj, "%s", | |
3886 | alloc_name(space_info->flags)); | |
3887 | if (ret) { | |
3888 | percpu_counter_destroy(&space_info->total_bytes_pinned); | |
3889 | kfree(space_info); | |
3890 | return ret; | |
3891 | } | |
3892 | ||
2be12ef7 NB |
3893 | list_add_rcu(&space_info->list, &info->space_info); |
3894 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3895 | info->data_sinfo = space_info; | |
3896 | ||
3897 | return ret; | |
3898 | } | |
3899 | ||
d2006e6d | 3900 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 3901 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 3902 | u64 bytes_readonly, |
593060d7 CM |
3903 | struct btrfs_space_info **space_info) |
3904 | { | |
3905 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3906 | int factor; |
3907 | ||
46df06b8 | 3908 | factor = btrfs_bg_type_to_factor(flags); |
593060d7 CM |
3909 | |
3910 | found = __find_space_info(info, flags); | |
d2006e6d NB |
3911 | ASSERT(found); |
3912 | spin_lock(&found->lock); | |
3913 | found->total_bytes += total_bytes; | |
3914 | found->disk_total += total_bytes * factor; | |
3915 | found->bytes_used += bytes_used; | |
3916 | found->disk_used += bytes_used * factor; | |
3917 | found->bytes_readonly += bytes_readonly; | |
3918 | if (total_bytes > 0) | |
3919 | found->full = 0; | |
3920 | space_info_add_new_bytes(info, found, total_bytes - | |
3921 | bytes_used - bytes_readonly); | |
3922 | spin_unlock(&found->lock); | |
3923 | *space_info = found; | |
593060d7 CM |
3924 | } |
3925 | ||
8790d502 CM |
3926 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3927 | { | |
899c81ea ID |
3928 | u64 extra_flags = chunk_to_extended(flags) & |
3929 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3930 | |
de98ced9 | 3931 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3932 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3933 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3934 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3935 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3936 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3937 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3938 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3939 | } |
593060d7 | 3940 | |
fc67c450 ID |
3941 | /* |
3942 | * returns target flags in extended format or 0 if restripe for this | |
3943 | * chunk_type is not in progress | |
c6664b42 | 3944 | * |
dccdb07b | 3945 | * should be called with balance_lock held |
fc67c450 ID |
3946 | */ |
3947 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3948 | { | |
3949 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3950 | u64 target = 0; | |
3951 | ||
fc67c450 ID |
3952 | if (!bctl) |
3953 | return 0; | |
3954 | ||
3955 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3956 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3957 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3958 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3959 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3960 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3961 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3962 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3963 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3964 | } | |
3965 | ||
3966 | return target; | |
3967 | } | |
3968 | ||
a46d11a8 ID |
3969 | /* |
3970 | * @flags: available profiles in extended format (see ctree.h) | |
3971 | * | |
e4d8ec0f ID |
3972 | * Returns reduced profile in chunk format. If profile changing is in |
3973 | * progress (either running or paused) picks the target profile (if it's | |
3974 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3975 | */ |
2ff7e61e | 3976 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 3977 | { |
0b246afa | 3978 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 3979 | u64 target; |
9c170b26 ZL |
3980 | u64 raid_type; |
3981 | u64 allowed = 0; | |
a061fc8d | 3982 | |
fc67c450 ID |
3983 | /* |
3984 | * see if restripe for this chunk_type is in progress, if so | |
3985 | * try to reduce to the target profile | |
3986 | */ | |
0b246afa JM |
3987 | spin_lock(&fs_info->balance_lock); |
3988 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
3989 | if (target) { |
3990 | /* pick target profile only if it's already available */ | |
3991 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 3992 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 3993 | return extended_to_chunk(target); |
e4d8ec0f ID |
3994 | } |
3995 | } | |
0b246afa | 3996 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 3997 | |
53b381b3 | 3998 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
3999 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4000 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
41a6e891 | 4001 | allowed |= btrfs_raid_array[raid_type].bg_flag; |
9c170b26 ZL |
4002 | } |
4003 | allowed &= flags; | |
4004 | ||
4005 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4006 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4007 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4008 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4009 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4010 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4011 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4012 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4013 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4014 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4015 | ||
4016 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4017 | ||
4018 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4019 | } |
4020 | ||
2ff7e61e | 4021 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4022 | { |
de98ced9 | 4023 | unsigned seq; |
f8213bdc | 4024 | u64 flags; |
de98ced9 MX |
4025 | |
4026 | do { | |
f8213bdc | 4027 | flags = orig_flags; |
0b246afa | 4028 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4029 | |
4030 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4031 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4032 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4033 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4034 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4035 | flags |= fs_info->avail_metadata_alloc_bits; |
4036 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4037 | |
2ff7e61e | 4038 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4039 | } |
4040 | ||
1b86826d | 4041 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4042 | { |
0b246afa | 4043 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4044 | u64 flags; |
53b381b3 | 4045 | u64 ret; |
9ed74f2d | 4046 | |
b742bb82 YZ |
4047 | if (data) |
4048 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4049 | else if (root == fs_info->chunk_root) |
b742bb82 | 4050 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4051 | else |
b742bb82 | 4052 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4053 | |
2ff7e61e | 4054 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4055 | return ret; |
6a63209f | 4056 | } |
9ed74f2d | 4057 | |
1b86826d JM |
4058 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4059 | { | |
4060 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4061 | } | |
4062 | ||
4063 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4064 | { | |
4065 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4066 | } | |
4067 | ||
4068 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4069 | { | |
4070 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4071 | } | |
4072 | ||
4136135b LB |
4073 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4074 | bool may_use_included) | |
4075 | { | |
4076 | ASSERT(s_info); | |
4077 | return s_info->bytes_used + s_info->bytes_reserved + | |
4078 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4079 | (may_use_included ? s_info->bytes_may_use : 0); | |
4080 | } | |
4081 | ||
04f4f916 | 4082 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4083 | { |
04f4f916 | 4084 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4085 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4086 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4087 | u64 used; |
94b947b2 | 4088 | int ret = 0; |
c99f1b0c ZL |
4089 | int need_commit = 2; |
4090 | int have_pinned_space; | |
6a63209f | 4091 | |
6a63209f | 4092 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4093 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4094 | |
9dced186 | 4095 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4096 | need_commit = 0; |
9dced186 | 4097 | ASSERT(current->journal_info); |
0af3d00b JB |
4098 | } |
4099 | ||
6a63209f JB |
4100 | again: |
4101 | /* make sure we have enough space to handle the data first */ | |
4102 | spin_lock(&data_sinfo->lock); | |
4136135b | 4103 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4104 | |
4105 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4106 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4107 | |
6a63209f JB |
4108 | /* |
4109 | * if we don't have enough free bytes in this space then we need | |
4110 | * to alloc a new chunk. | |
4111 | */ | |
b9fd47cd | 4112 | if (!data_sinfo->full) { |
6a63209f | 4113 | u64 alloc_target; |
9ed74f2d | 4114 | |
0e4f8f88 | 4115 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4116 | spin_unlock(&data_sinfo->lock); |
1174cade | 4117 | |
1b86826d | 4118 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4119 | /* |
4120 | * It is ugly that we don't call nolock join | |
4121 | * transaction for the free space inode case here. | |
4122 | * But it is safe because we only do the data space | |
4123 | * reservation for the free space cache in the | |
4124 | * transaction context, the common join transaction | |
4125 | * just increase the counter of the current transaction | |
4126 | * handler, doesn't try to acquire the trans_lock of | |
4127 | * the fs. | |
4128 | */ | |
7a7eaa40 | 4129 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4130 | if (IS_ERR(trans)) |
4131 | return PTR_ERR(trans); | |
9ed74f2d | 4132 | |
01458828 | 4133 | ret = do_chunk_alloc(trans, alloc_target, |
0e4f8f88 | 4134 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4135 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4136 | if (ret < 0) { |
4137 | if (ret != -ENOSPC) | |
4138 | return ret; | |
c99f1b0c ZL |
4139 | else { |
4140 | have_pinned_space = 1; | |
d52a5b5f | 4141 | goto commit_trans; |
c99f1b0c | 4142 | } |
d52a5b5f | 4143 | } |
9ed74f2d | 4144 | |
6a63209f JB |
4145 | goto again; |
4146 | } | |
f2bb8f5c JB |
4147 | |
4148 | /* | |
b150a4f1 | 4149 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4150 | * allocation, and no removed chunk in current transaction, |
4151 | * don't bother committing the transaction. | |
f2bb8f5c | 4152 | */ |
dec59fa3 | 4153 | have_pinned_space = __percpu_counter_compare( |
c99f1b0c | 4154 | &data_sinfo->total_bytes_pinned, |
dec59fa3 EL |
4155 | used + bytes - data_sinfo->total_bytes, |
4156 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
6a63209f | 4157 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4158 | |
4e06bdd6 | 4159 | /* commit the current transaction and try again */ |
d52a5b5f | 4160 | commit_trans: |
92e2f7e3 | 4161 | if (need_commit) { |
c99f1b0c | 4162 | need_commit--; |
b150a4f1 | 4163 | |
e1746e83 | 4164 | if (need_commit > 0) { |
82b3e53b | 4165 | btrfs_start_delalloc_roots(fs_info, -1); |
6374e57a | 4166 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4167 | (u64)-1); |
e1746e83 | 4168 | } |
9a4e7276 | 4169 | |
7a7eaa40 | 4170 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4171 | if (IS_ERR(trans)) |
4172 | return PTR_ERR(trans); | |
c99f1b0c | 4173 | if (have_pinned_space >= 0 || |
3204d33c JB |
4174 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4175 | &trans->transaction->flags) || | |
c99f1b0c | 4176 | need_commit > 0) { |
3a45bb20 | 4177 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4178 | if (ret) |
4179 | return ret; | |
d7c15171 | 4180 | /* |
c2d6cb16 FM |
4181 | * The cleaner kthread might still be doing iput |
4182 | * operations. Wait for it to finish so that | |
034f784d JB |
4183 | * more space is released. We don't need to |
4184 | * explicitly run the delayed iputs here because | |
4185 | * the commit_transaction would have woken up | |
4186 | * the cleaner. | |
d7c15171 | 4187 | */ |
034f784d JB |
4188 | ret = btrfs_wait_on_delayed_iputs(fs_info); |
4189 | if (ret) | |
4190 | return ret; | |
94b947b2 ZL |
4191 | goto again; |
4192 | } else { | |
3a45bb20 | 4193 | btrfs_end_transaction(trans); |
94b947b2 | 4194 | } |
4e06bdd6 | 4195 | } |
9ed74f2d | 4196 | |
0b246afa | 4197 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4198 | "space_info:enospc", |
4199 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4200 | return -ENOSPC; |
4201 | } | |
9f9b8e8d | 4202 | update_bytes_may_use(data_sinfo, bytes); |
0b246afa | 4203 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4204 | data_sinfo->flags, bytes, 1); |
6a63209f | 4205 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4206 | |
4559b0a7 | 4207 | return 0; |
9ed74f2d | 4208 | } |
6a63209f | 4209 | |
364ecf36 QW |
4210 | int btrfs_check_data_free_space(struct inode *inode, |
4211 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4212 | { |
0b246afa | 4213 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4214 | int ret; |
4215 | ||
4216 | /* align the range */ | |
0b246afa JM |
4217 | len = round_up(start + len, fs_info->sectorsize) - |
4218 | round_down(start, fs_info->sectorsize); | |
4219 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4220 | |
04f4f916 | 4221 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4222 | if (ret < 0) |
4223 | return ret; | |
4224 | ||
1e5ec2e7 | 4225 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4226 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4227 | if (ret < 0) |
1e5ec2e7 | 4228 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4229 | else |
4230 | ret = 0; | |
4ceff079 QW |
4231 | return ret; |
4232 | } | |
4233 | ||
4ceff079 QW |
4234 | /* |
4235 | * Called if we need to clear a data reservation for this inode | |
4236 | * Normally in a error case. | |
4237 | * | |
51773bec QW |
4238 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4239 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4240 | * Like clear_bit_hook(). | |
4ceff079 | 4241 | */ |
51773bec QW |
4242 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4243 | u64 len) | |
4ceff079 | 4244 | { |
0b246afa | 4245 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4246 | struct btrfs_space_info *data_sinfo; |
4247 | ||
4248 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4249 | len = round_up(start + len, fs_info->sectorsize) - |
4250 | round_down(start, fs_info->sectorsize); | |
4251 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4252 | |
0b246afa | 4253 | data_sinfo = fs_info->data_sinfo; |
4ceff079 | 4254 | spin_lock(&data_sinfo->lock); |
9f9b8e8d | 4255 | update_bytes_may_use(data_sinfo, -len); |
0b246afa | 4256 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4257 | data_sinfo->flags, len, 0); |
4258 | spin_unlock(&data_sinfo->lock); | |
4259 | } | |
4260 | ||
51773bec QW |
4261 | /* |
4262 | * Called if we need to clear a data reservation for this inode | |
4263 | * Normally in a error case. | |
4264 | * | |
01327610 | 4265 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4266 | * space framework. |
4267 | */ | |
bc42bda2 QW |
4268 | void btrfs_free_reserved_data_space(struct inode *inode, |
4269 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4270 | { |
0c476a5d JM |
4271 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4272 | ||
4273 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4274 | len = round_up(start + len, root->fs_info->sectorsize) - |
4275 | round_down(start, root->fs_info->sectorsize); | |
4276 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4277 | |
51773bec | 4278 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4279 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4280 | } |
4281 | ||
97e728d4 | 4282 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4283 | { |
97e728d4 JB |
4284 | struct list_head *head = &info->space_info; |
4285 | struct btrfs_space_info *found; | |
e3ccfa98 | 4286 | |
97e728d4 JB |
4287 | rcu_read_lock(); |
4288 | list_for_each_entry_rcu(found, head, list) { | |
4289 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4290 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4291 | } |
97e728d4 | 4292 | rcu_read_unlock(); |
e3ccfa98 JB |
4293 | } |
4294 | ||
3c76cd84 MX |
4295 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4296 | { | |
4297 | return (global->size << 1); | |
4298 | } | |
4299 | ||
2ff7e61e | 4300 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4301 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4302 | { |
8d8aafee | 4303 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4304 | u64 thresh; |
e3ccfa98 | 4305 | |
0e4f8f88 CM |
4306 | if (force == CHUNK_ALLOC_FORCE) |
4307 | return 1; | |
4308 | ||
4309 | /* | |
4310 | * in limited mode, we want to have some free space up to | |
4311 | * about 1% of the FS size. | |
4312 | */ | |
4313 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4314 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4315 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4316 | |
8d8aafee | 4317 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4318 | return 1; |
4319 | } | |
0e4f8f88 | 4320 | |
8d8aafee | 4321 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4322 | return 0; |
424499db | 4323 | return 1; |
32c00aff JB |
4324 | } |
4325 | ||
2ff7e61e | 4326 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4327 | { |
4328 | u64 num_dev; | |
4329 | ||
53b381b3 DW |
4330 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4331 | BTRFS_BLOCK_GROUP_RAID0 | | |
4332 | BTRFS_BLOCK_GROUP_RAID5 | | |
4333 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4334 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4335 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4336 | num_dev = 2; | |
4337 | else | |
4338 | num_dev = 1; /* DUP or single */ | |
4339 | ||
39c2d7fa | 4340 | return num_dev; |
15d1ff81 LB |
4341 | } |
4342 | ||
39c2d7fa FM |
4343 | /* |
4344 | * If @is_allocation is true, reserve space in the system space info necessary | |
4345 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4346 | * removing a chunk. | |
4347 | */ | |
451a2c13 | 4348 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) |
15d1ff81 | 4349 | { |
451a2c13 | 4350 | struct btrfs_fs_info *fs_info = trans->fs_info; |
15d1ff81 LB |
4351 | struct btrfs_space_info *info; |
4352 | u64 left; | |
4353 | u64 thresh; | |
4fbcdf66 | 4354 | int ret = 0; |
39c2d7fa | 4355 | u64 num_devs; |
4fbcdf66 FM |
4356 | |
4357 | /* | |
4358 | * Needed because we can end up allocating a system chunk and for an | |
4359 | * atomic and race free space reservation in the chunk block reserve. | |
4360 | */ | |
a32bf9a3 | 4361 | lockdep_assert_held(&fs_info->chunk_mutex); |
15d1ff81 | 4362 | |
0b246afa | 4363 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4364 | spin_lock(&info->lock); |
4136135b | 4365 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4366 | spin_unlock(&info->lock); |
4367 | ||
2ff7e61e | 4368 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4369 | |
4370 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4371 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4372 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4373 | |
0b246afa JM |
4374 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4375 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4376 | left, thresh, type); | |
4377 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4378 | } |
4379 | ||
4380 | if (left < thresh) { | |
1b86826d | 4381 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4382 | |
4fbcdf66 FM |
4383 | /* |
4384 | * Ignore failure to create system chunk. We might end up not | |
4385 | * needing it, as we might not need to COW all nodes/leafs from | |
4386 | * the paths we visit in the chunk tree (they were already COWed | |
4387 | * or created in the current transaction for example). | |
4388 | */ | |
c216b203 | 4389 | ret = btrfs_alloc_chunk(trans, flags); |
4fbcdf66 FM |
4390 | } |
4391 | ||
4392 | if (!ret) { | |
0b246afa JM |
4393 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4394 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4395 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4396 | if (!ret) | |
4397 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4398 | } |
4399 | } | |
4400 | ||
28b737f6 LB |
4401 | /* |
4402 | * If force is CHUNK_ALLOC_FORCE: | |
4403 | * - return 1 if it successfully allocates a chunk, | |
4404 | * - return errors including -ENOSPC otherwise. | |
4405 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4406 | * - return 0 if it doesn't need to allocate a new chunk, | |
4407 | * - return 1 if it successfully allocates a chunk, | |
4408 | * - return errors including -ENOSPC otherwise. | |
4409 | */ | |
01458828 NB |
4410 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
4411 | int force) | |
9ed74f2d | 4412 | { |
01458828 | 4413 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 4414 | struct btrfs_space_info *space_info; |
2556fbb0 NB |
4415 | bool wait_for_alloc = false; |
4416 | bool should_alloc = false; | |
9ed74f2d | 4417 | int ret = 0; |
9ed74f2d | 4418 | |
c6b305a8 JB |
4419 | /* Don't re-enter if we're already allocating a chunk */ |
4420 | if (trans->allocating_chunk) | |
4421 | return -ENOSPC; | |
4422 | ||
0b246afa | 4423 | space_info = __find_space_info(fs_info, flags); |
dc2d3005 | 4424 | ASSERT(space_info); |
9ed74f2d | 4425 | |
2556fbb0 NB |
4426 | do { |
4427 | spin_lock(&space_info->lock); | |
4428 | if (force < space_info->force_alloc) | |
4429 | force = space_info->force_alloc; | |
4430 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
4431 | if (space_info->full) { | |
4432 | /* No more free physical space */ | |
4433 | if (should_alloc) | |
4434 | ret = -ENOSPC; | |
4435 | else | |
4436 | ret = 0; | |
4437 | spin_unlock(&space_info->lock); | |
4438 | return ret; | |
4439 | } else if (!should_alloc) { | |
4440 | spin_unlock(&space_info->lock); | |
4441 | return 0; | |
4442 | } else if (space_info->chunk_alloc) { | |
4443 | /* | |
4444 | * Someone is already allocating, so we need to block | |
4445 | * until this someone is finished and then loop to | |
4446 | * recheck if we should continue with our allocation | |
4447 | * attempt. | |
4448 | */ | |
4449 | wait_for_alloc = true; | |
4450 | spin_unlock(&space_info->lock); | |
4451 | mutex_lock(&fs_info->chunk_mutex); | |
4452 | mutex_unlock(&fs_info->chunk_mutex); | |
4453 | } else { | |
4454 | /* Proceed with allocation */ | |
4455 | space_info->chunk_alloc = 1; | |
4456 | wait_for_alloc = false; | |
4457 | spin_unlock(&space_info->lock); | |
4458 | } | |
6d74119f | 4459 | |
1e1c50a9 | 4460 | cond_resched(); |
2556fbb0 | 4461 | } while (wait_for_alloc); |
6d74119f | 4462 | |
2556fbb0 | 4463 | mutex_lock(&fs_info->chunk_mutex); |
c6b305a8 JB |
4464 | trans->allocating_chunk = true; |
4465 | ||
67377734 JB |
4466 | /* |
4467 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4468 | * allocating mixed chunks instead of individual chunks. | |
4469 | */ | |
4470 | if (btrfs_mixed_space_info(space_info)) | |
4471 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4472 | ||
97e728d4 JB |
4473 | /* |
4474 | * if we're doing a data chunk, go ahead and make sure that | |
4475 | * we keep a reasonable number of metadata chunks allocated in the | |
4476 | * FS as well. | |
4477 | */ | |
9ed74f2d | 4478 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4479 | fs_info->data_chunk_allocations++; |
4480 | if (!(fs_info->data_chunk_allocations % | |
4481 | fs_info->metadata_ratio)) | |
4482 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4483 | } |
4484 | ||
15d1ff81 LB |
4485 | /* |
4486 | * Check if we have enough space in SYSTEM chunk because we may need | |
4487 | * to update devices. | |
4488 | */ | |
451a2c13 | 4489 | check_system_chunk(trans, flags); |
15d1ff81 | 4490 | |
c216b203 | 4491 | ret = btrfs_alloc_chunk(trans, flags); |
c6b305a8 | 4492 | trans->allocating_chunk = false; |
92b8e897 | 4493 | |
9ed74f2d | 4494 | spin_lock(&space_info->lock); |
57f1642e NB |
4495 | if (ret < 0) { |
4496 | if (ret == -ENOSPC) | |
4497 | space_info->full = 1; | |
4498 | else | |
4499 | goto out; | |
4500 | } else { | |
424499db | 4501 | ret = 1; |
21a94f7a | 4502 | space_info->max_extent_size = 0; |
57f1642e | 4503 | } |
6d74119f | 4504 | |
0e4f8f88 | 4505 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4506 | out: |
6d74119f | 4507 | space_info->chunk_alloc = 0; |
9ed74f2d | 4508 | spin_unlock(&space_info->lock); |
a25c75d5 | 4509 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4510 | /* |
4511 | * When we allocate a new chunk we reserve space in the chunk block | |
4512 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4513 | * add new nodes/leafs to it if we end up needing to do it when | |
4514 | * inserting the chunk item and updating device items as part of the | |
4515 | * second phase of chunk allocation, performed by | |
4516 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4517 | * large number of new block groups to create in our transaction | |
4518 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4519 | * in extreme cases - like having a single transaction create many new | |
4520 | * block groups when starting to write out the free space caches of all | |
4521 | * the block groups that were made dirty during the lifetime of the | |
4522 | * transaction. | |
4523 | */ | |
5ce55557 | 4524 | if (trans->chunk_bytes_reserved >= (u64)SZ_2M) |
6c686b35 | 4525 | btrfs_create_pending_block_groups(trans); |
5ce55557 | 4526 | |
0f9dd46c | 4527 | return ret; |
6324fbf3 | 4528 | } |
9ed74f2d | 4529 | |
c1c4919b | 4530 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4531 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4532 | enum btrfs_reserve_flush_enum flush, |
4533 | bool system_chunk) | |
a80c8dcf | 4534 | { |
0b246afa | 4535 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4536 | u64 profile; |
3c76cd84 | 4537 | u64 space_size; |
a80c8dcf JB |
4538 | u64 avail; |
4539 | u64 used; | |
46df06b8 | 4540 | int factor; |
a80c8dcf | 4541 | |
957780eb JB |
4542 | /* Don't overcommit when in mixed mode. */ |
4543 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4544 | return 0; | |
4545 | ||
c1c4919b JM |
4546 | if (system_chunk) |
4547 | profile = btrfs_system_alloc_profile(fs_info); | |
4548 | else | |
4549 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4550 | ||
4136135b | 4551 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4552 | |
96f1bb57 JB |
4553 | /* |
4554 | * We only want to allow over committing if we have lots of actual space | |
4555 | * free, but if we don't have enough space to handle the global reserve | |
4556 | * space then we could end up having a real enospc problem when trying | |
4557 | * to allocate a chunk or some other such important allocation. | |
4558 | */ | |
3c76cd84 MX |
4559 | spin_lock(&global_rsv->lock); |
4560 | space_size = calc_global_rsv_need_space(global_rsv); | |
4561 | spin_unlock(&global_rsv->lock); | |
4562 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4563 | return 0; |
4564 | ||
4565 | used += space_info->bytes_may_use; | |
a80c8dcf | 4566 | |
a5ed45f8 | 4567 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4568 | |
4569 | /* | |
4570 | * If we have dup, raid1 or raid10 then only half of the free | |
52042d8e | 4571 | * space is actually usable. For raid56, the space info used |
53b381b3 DW |
4572 | * doesn't include the parity drive, so we don't have to |
4573 | * change the math | |
a80c8dcf | 4574 | */ |
46df06b8 DS |
4575 | factor = btrfs_bg_type_to_factor(profile); |
4576 | avail = div_u64(avail, factor); | |
a80c8dcf JB |
4577 | |
4578 | /* | |
561c294d MX |
4579 | * If we aren't flushing all things, let us overcommit up to |
4580 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4581 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4582 | */ |
08e007d2 | 4583 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4584 | avail >>= 3; |
a80c8dcf | 4585 | else |
14575aef | 4586 | avail >>= 1; |
a80c8dcf | 4587 | |
14575aef | 4588 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4589 | return 1; |
4590 | return 0; | |
4591 | } | |
4592 | ||
2ff7e61e | 4593 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4594 | unsigned long nr_pages, int nr_items) |
da633a42 | 4595 | { |
0b246afa | 4596 | struct super_block *sb = fs_info->sb; |
da633a42 | 4597 | |
925a6efb JB |
4598 | if (down_read_trylock(&sb->s_umount)) { |
4599 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4600 | up_read(&sb->s_umount); | |
4601 | } else { | |
da633a42 MX |
4602 | /* |
4603 | * We needn't worry the filesystem going from r/w to r/o though | |
4604 | * we don't acquire ->s_umount mutex, because the filesystem | |
4605 | * should guarantee the delalloc inodes list be empty after | |
4606 | * the filesystem is readonly(all dirty pages are written to | |
4607 | * the disk). | |
4608 | */ | |
82b3e53b | 4609 | btrfs_start_delalloc_roots(fs_info, nr_items); |
98ad69cf | 4610 | if (!current->journal_info) |
0b246afa | 4611 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4612 | } |
4613 | } | |
4614 | ||
6374e57a | 4615 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4616 | u64 to_reclaim) |
18cd8ea6 MX |
4617 | { |
4618 | u64 bytes; | |
6374e57a | 4619 | u64 nr; |
18cd8ea6 | 4620 | |
2ff7e61e | 4621 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4622 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4623 | if (!nr) |
4624 | nr = 1; | |
4625 | return nr; | |
4626 | } | |
4627 | ||
ee22184b | 4628 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4629 | |
9ed74f2d | 4630 | /* |
5da9d01b | 4631 | * shrink metadata reservation for delalloc |
9ed74f2d | 4632 | */ |
c1c4919b JM |
4633 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4634 | u64 orig, bool wait_ordered) | |
5da9d01b | 4635 | { |
0019f10d | 4636 | struct btrfs_space_info *space_info; |
663350ac | 4637 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4638 | u64 delalloc_bytes; |
420829d8 | 4639 | u64 async_pages; |
6374e57a | 4640 | u64 items; |
b1953bce | 4641 | long time_left; |
d3ee29e3 MX |
4642 | unsigned long nr_pages; |
4643 | int loops; | |
5da9d01b | 4644 | |
c61a16a7 | 4645 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4646 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4647 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4648 | |
663350ac | 4649 | trans = (struct btrfs_trans_handle *)current->journal_info; |
69fe2d75 | 4650 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4651 | |
963d678b | 4652 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4653 | &fs_info->delalloc_bytes); |
f4c738c2 | 4654 | if (delalloc_bytes == 0) { |
fdb5effd | 4655 | if (trans) |
f4c738c2 | 4656 | return; |
38c135af | 4657 | if (wait_ordered) |
0b246afa | 4658 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4659 | return; |
fdb5effd JB |
4660 | } |
4661 | ||
d3ee29e3 | 4662 | loops = 0; |
f4c738c2 | 4663 | while (delalloc_bytes && loops < 3) { |
420829d8 NB |
4664 | nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT; |
4665 | ||
4666 | /* | |
4667 | * Triggers inode writeback for up to nr_pages. This will invoke | |
4668 | * ->writepages callback and trigger delalloc filling | |
4669 | * (btrfs_run_delalloc_range()). | |
4670 | */ | |
2ff7e61e | 4671 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
420829d8 | 4672 | |
dea31f52 | 4673 | /* |
420829d8 NB |
4674 | * We need to wait for the compressed pages to start before |
4675 | * we continue. | |
dea31f52 | 4676 | */ |
420829d8 NB |
4677 | async_pages = atomic_read(&fs_info->async_delalloc_pages); |
4678 | if (!async_pages) | |
9f3a074d MX |
4679 | goto skip_async; |
4680 | ||
420829d8 NB |
4681 | /* |
4682 | * Calculate how many compressed pages we want to be written | |
4683 | * before we continue. I.e if there are more async pages than we | |
4684 | * require wait_event will wait until nr_pages are written. | |
4685 | */ | |
4686 | if (async_pages <= nr_pages) | |
4687 | async_pages = 0; | |
9f3a074d | 4688 | else |
420829d8 | 4689 | async_pages -= nr_pages; |
dea31f52 | 4690 | |
0b246afa JM |
4691 | wait_event(fs_info->async_submit_wait, |
4692 | atomic_read(&fs_info->async_delalloc_pages) <= | |
420829d8 | 4693 | (int)async_pages); |
9f3a074d | 4694 | skip_async: |
0019f10d | 4695 | spin_lock(&space_info->lock); |
957780eb JB |
4696 | if (list_empty(&space_info->tickets) && |
4697 | list_empty(&space_info->priority_tickets)) { | |
4698 | spin_unlock(&space_info->lock); | |
4699 | break; | |
4700 | } | |
0019f10d | 4701 | spin_unlock(&space_info->lock); |
5da9d01b | 4702 | |
36e39c40 | 4703 | loops++; |
f104d044 | 4704 | if (wait_ordered && !trans) { |
0b246afa | 4705 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4706 | } else { |
f4c738c2 | 4707 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4708 | if (time_left) |
4709 | break; | |
4710 | } | |
963d678b | 4711 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4712 | &fs_info->delalloc_bytes); |
5da9d01b | 4713 | } |
5da9d01b YZ |
4714 | } |
4715 | ||
996478ca | 4716 | struct reserve_ticket { |
f91587e4 | 4717 | u64 orig_bytes; |
996478ca JB |
4718 | u64 bytes; |
4719 | int error; | |
4720 | struct list_head list; | |
4721 | wait_queue_head_t wait; | |
4722 | }; | |
4723 | ||
663350ac JB |
4724 | /** |
4725 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4726 | * @root - the root we're allocating for | |
4727 | * @bytes - the number of bytes we want to reserve | |
4728 | * @force - force the commit | |
8bb8ab2e | 4729 | * |
663350ac JB |
4730 | * This will check to make sure that committing the transaction will actually |
4731 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4732 | * will return -ENOSPC. | |
8bb8ab2e | 4733 | */ |
0c9ab349 | 4734 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4735 | struct btrfs_space_info *space_info) |
663350ac | 4736 | { |
996478ca | 4737 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4738 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
4c8edbc7 | 4739 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
663350ac | 4740 | struct btrfs_trans_handle *trans; |
4c8edbc7 JB |
4741 | u64 bytes_needed; |
4742 | u64 reclaim_bytes = 0; | |
663350ac JB |
4743 | |
4744 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4745 | if (trans) | |
4746 | return -EAGAIN; | |
4747 | ||
996478ca JB |
4748 | spin_lock(&space_info->lock); |
4749 | if (!list_empty(&space_info->priority_tickets)) | |
4750 | ticket = list_first_entry(&space_info->priority_tickets, | |
4751 | struct reserve_ticket, list); | |
4752 | else if (!list_empty(&space_info->tickets)) | |
4753 | ticket = list_first_entry(&space_info->tickets, | |
4754 | struct reserve_ticket, list); | |
4c8edbc7 | 4755 | bytes_needed = (ticket) ? ticket->bytes : 0; |
996478ca JB |
4756 | spin_unlock(&space_info->lock); |
4757 | ||
4c8edbc7 | 4758 | if (!bytes_needed) |
996478ca | 4759 | return 0; |
663350ac | 4760 | |
d89dbefb JB |
4761 | trans = btrfs_join_transaction(fs_info->extent_root); |
4762 | if (IS_ERR(trans)) | |
4763 | return PTR_ERR(trans); | |
4764 | ||
4765 | /* | |
4766 | * See if there is enough pinned space to make this reservation, or if | |
4767 | * we have block groups that are going to be freed, allowing us to | |
4768 | * possibly do a chunk allocation the next loop through. | |
4769 | */ | |
4770 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) || | |
4771 | __percpu_counter_compare(&space_info->total_bytes_pinned, | |
4772 | bytes_needed, | |
4773 | BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0) | |
663350ac | 4774 | goto commit; |
663350ac JB |
4775 | |
4776 | /* | |
4777 | * See if there is some space in the delayed insertion reservation for | |
4778 | * this reservation. | |
4779 | */ | |
4780 | if (space_info != delayed_rsv->space_info) | |
d89dbefb | 4781 | goto enospc; |
663350ac JB |
4782 | |
4783 | spin_lock(&delayed_rsv->lock); | |
4c8edbc7 | 4784 | reclaim_bytes += delayed_rsv->reserved; |
057aac3e NB |
4785 | spin_unlock(&delayed_rsv->lock); |
4786 | ||
4c8edbc7 JB |
4787 | spin_lock(&delayed_refs_rsv->lock); |
4788 | reclaim_bytes += delayed_refs_rsv->reserved; | |
4789 | spin_unlock(&delayed_refs_rsv->lock); | |
4790 | if (reclaim_bytes >= bytes_needed) | |
4791 | goto commit; | |
4792 | bytes_needed -= reclaim_bytes; | |
4793 | ||
dec59fa3 | 4794 | if (__percpu_counter_compare(&space_info->total_bytes_pinned, |
4c8edbc7 | 4795 | bytes_needed, |
d89dbefb JB |
4796 | BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) |
4797 | goto enospc; | |
663350ac JB |
4798 | |
4799 | commit: | |
3a45bb20 | 4800 | return btrfs_commit_transaction(trans); |
d89dbefb JB |
4801 | enospc: |
4802 | btrfs_end_transaction(trans); | |
4803 | return -ENOSPC; | |
663350ac JB |
4804 | } |
4805 | ||
e38ae7a0 NB |
4806 | /* |
4807 | * Try to flush some data based on policy set by @state. This is only advisory | |
4808 | * and may fail for various reasons. The caller is supposed to examine the | |
4809 | * state of @space_info to detect the outcome. | |
4810 | */ | |
4811 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4812 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4813 | int state) |
96c3f433 | 4814 | { |
a9b3311e | 4815 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4816 | struct btrfs_trans_handle *trans; |
4817 | int nr; | |
f4c738c2 | 4818 | int ret = 0; |
96c3f433 JB |
4819 | |
4820 | switch (state) { | |
96c3f433 JB |
4821 | case FLUSH_DELAYED_ITEMS_NR: |
4822 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4823 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4824 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4825 | else |
96c3f433 | 4826 | nr = -1; |
18cd8ea6 | 4827 | |
96c3f433 JB |
4828 | trans = btrfs_join_transaction(root); |
4829 | if (IS_ERR(trans)) { | |
4830 | ret = PTR_ERR(trans); | |
4831 | break; | |
4832 | } | |
e5c304e6 | 4833 | ret = btrfs_run_delayed_items_nr(trans, nr); |
3a45bb20 | 4834 | btrfs_end_transaction(trans); |
96c3f433 | 4835 | break; |
67b0fd63 JB |
4836 | case FLUSH_DELALLOC: |
4837 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 4838 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
4839 | state == FLUSH_DELALLOC_WAIT); |
4840 | break; | |
413df725 JB |
4841 | case FLUSH_DELAYED_REFS_NR: |
4842 | case FLUSH_DELAYED_REFS: | |
4843 | trans = btrfs_join_transaction(root); | |
4844 | if (IS_ERR(trans)) { | |
4845 | ret = PTR_ERR(trans); | |
4846 | break; | |
4847 | } | |
4848 | if (state == FLUSH_DELAYED_REFS_NR) | |
4849 | nr = calc_reclaim_items_nr(fs_info, num_bytes); | |
4850 | else | |
4851 | nr = 0; | |
4852 | btrfs_run_delayed_refs(trans, nr); | |
4853 | btrfs_end_transaction(trans); | |
4854 | break; | |
ea658bad | 4855 | case ALLOC_CHUNK: |
450114fc | 4856 | case ALLOC_CHUNK_FORCE: |
ea658bad JB |
4857 | trans = btrfs_join_transaction(root); |
4858 | if (IS_ERR(trans)) { | |
4859 | ret = PTR_ERR(trans); | |
4860 | break; | |
4861 | } | |
01458828 | 4862 | ret = do_chunk_alloc(trans, |
1b86826d | 4863 | btrfs_metadata_alloc_profile(fs_info), |
450114fc JB |
4864 | (state == ALLOC_CHUNK) ? |
4865 | CHUNK_ALLOC_NO_FORCE : CHUNK_ALLOC_FORCE); | |
3a45bb20 | 4866 | btrfs_end_transaction(trans); |
eecba891 | 4867 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
4868 | ret = 0; |
4869 | break; | |
96c3f433 | 4870 | case COMMIT_TRANS: |
3ec9a4c8 JB |
4871 | /* |
4872 | * If we have pending delayed iputs then we could free up a | |
4873 | * bunch of pinned space, so make sure we run the iputs before | |
4874 | * we do our pinned bytes check below. | |
4875 | */ | |
3ec9a4c8 | 4876 | btrfs_run_delayed_iputs(fs_info); |
034f784d | 4877 | btrfs_wait_on_delayed_iputs(fs_info); |
3ec9a4c8 | 4878 | |
996478ca | 4879 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
4880 | break; |
4881 | default: | |
4882 | ret = -ENOSPC; | |
4883 | break; | |
4884 | } | |
4885 | ||
7bdd6277 NB |
4886 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
4887 | ret); | |
e38ae7a0 | 4888 | return; |
96c3f433 | 4889 | } |
21c7e756 MX |
4890 | |
4891 | static inline u64 | |
c1c4919b JM |
4892 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
4893 | struct btrfs_space_info *space_info, | |
4894 | bool system_chunk) | |
21c7e756 | 4895 | { |
957780eb | 4896 | struct reserve_ticket *ticket; |
21c7e756 MX |
4897 | u64 used; |
4898 | u64 expected; | |
957780eb | 4899 | u64 to_reclaim = 0; |
21c7e756 | 4900 | |
957780eb JB |
4901 | list_for_each_entry(ticket, &space_info->tickets, list) |
4902 | to_reclaim += ticket->bytes; | |
4903 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
4904 | to_reclaim += ticket->bytes; | |
4905 | if (to_reclaim) | |
4906 | return to_reclaim; | |
21c7e756 | 4907 | |
e0af2484 | 4908 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
4909 | if (can_overcommit(fs_info, space_info, to_reclaim, |
4910 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
4911 | return 0; |
4912 | ||
0eee8a49 NB |
4913 | used = btrfs_space_info_used(space_info, true); |
4914 | ||
c1c4919b JM |
4915 | if (can_overcommit(fs_info, space_info, SZ_1M, |
4916 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
4917 | expected = div_factor_fine(space_info->total_bytes, 95); |
4918 | else | |
4919 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4920 | ||
4921 | if (used > expected) | |
4922 | to_reclaim = used - expected; | |
4923 | else | |
4924 | to_reclaim = 0; | |
4925 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4926 | space_info->bytes_reserved); | |
21c7e756 MX |
4927 | return to_reclaim; |
4928 | } | |
4929 | ||
c1c4919b JM |
4930 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
4931 | struct btrfs_space_info *space_info, | |
4932 | u64 used, bool system_chunk) | |
21c7e756 | 4933 | { |
365c5313 JB |
4934 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4935 | ||
4936 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 4937 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
4938 | return 0; |
4939 | ||
c1c4919b JM |
4940 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4941 | system_chunk)) | |
d38b349c JB |
4942 | return 0; |
4943 | ||
0b246afa JM |
4944 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
4945 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
4946 | } |
4947 | ||
f91587e4 | 4948 | static bool wake_all_tickets(struct list_head *head) |
21c7e756 | 4949 | { |
957780eb | 4950 | struct reserve_ticket *ticket; |
25ce459c | 4951 | |
957780eb JB |
4952 | while (!list_empty(head)) { |
4953 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
4954 | list_del_init(&ticket->list); | |
4955 | ticket->error = -ENOSPC; | |
4956 | wake_up(&ticket->wait); | |
f91587e4 JB |
4957 | if (ticket->bytes != ticket->orig_bytes) |
4958 | return true; | |
21c7e756 | 4959 | } |
f91587e4 | 4960 | return false; |
21c7e756 MX |
4961 | } |
4962 | ||
957780eb JB |
4963 | /* |
4964 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
4965 | * will loop and continuously try to flush as long as we are making progress. | |
4966 | * We count progress as clearing off tickets each time we have to loop. | |
4967 | */ | |
21c7e756 MX |
4968 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
4969 | { | |
4970 | struct btrfs_fs_info *fs_info; | |
4971 | struct btrfs_space_info *space_info; | |
4972 | u64 to_reclaim; | |
4973 | int flush_state; | |
957780eb | 4974 | int commit_cycles = 0; |
ce129655 | 4975 | u64 last_tickets_id; |
21c7e756 MX |
4976 | |
4977 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4978 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4979 | ||
957780eb | 4980 | spin_lock(&space_info->lock); |
c1c4919b JM |
4981 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4982 | false); | |
957780eb JB |
4983 | if (!to_reclaim) { |
4984 | space_info->flush = 0; | |
4985 | spin_unlock(&space_info->lock); | |
21c7e756 | 4986 | return; |
957780eb | 4987 | } |
ce129655 | 4988 | last_tickets_id = space_info->tickets_id; |
957780eb | 4989 | spin_unlock(&space_info->lock); |
21c7e756 MX |
4990 | |
4991 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 4992 | do { |
e38ae7a0 | 4993 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
4994 | spin_lock(&space_info->lock); |
4995 | if (list_empty(&space_info->tickets)) { | |
4996 | space_info->flush = 0; | |
4997 | spin_unlock(&space_info->lock); | |
4998 | return; | |
4999 | } | |
c1c4919b JM |
5000 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5001 | space_info, | |
5002 | false); | |
ce129655 | 5003 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5004 | flush_state++; |
5005 | } else { | |
ce129655 | 5006 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5007 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5008 | if (commit_cycles) | |
5009 | commit_cycles--; | |
5010 | } | |
5011 | ||
450114fc JB |
5012 | /* |
5013 | * We don't want to force a chunk allocation until we've tried | |
5014 | * pretty hard to reclaim space. Think of the case where we | |
5015 | * freed up a bunch of space and so have a lot of pinned space | |
5016 | * to reclaim. We would rather use that than possibly create a | |
5017 | * underutilized metadata chunk. So if this is our first run | |
5018 | * through the flushing state machine skip ALLOC_CHUNK_FORCE and | |
5019 | * commit the transaction. If nothing has changed the next go | |
5020 | * around then we can force a chunk allocation. | |
5021 | */ | |
5022 | if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles) | |
5023 | flush_state++; | |
5024 | ||
957780eb JB |
5025 | if (flush_state > COMMIT_TRANS) { |
5026 | commit_cycles++; | |
5027 | if (commit_cycles > 2) { | |
f91587e4 JB |
5028 | if (wake_all_tickets(&space_info->tickets)) { |
5029 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5030 | commit_cycles--; | |
5031 | } else { | |
5032 | space_info->flush = 0; | |
5033 | } | |
957780eb JB |
5034 | } else { |
5035 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5036 | } | |
5037 | } | |
5038 | spin_unlock(&space_info->lock); | |
5039 | } while (flush_state <= COMMIT_TRANS); | |
5040 | } | |
5041 | ||
5042 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5043 | { | |
5044 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5045 | } | |
5046 | ||
8a1bbe1d JB |
5047 | static const enum btrfs_flush_state priority_flush_states[] = { |
5048 | FLUSH_DELAYED_ITEMS_NR, | |
5049 | FLUSH_DELAYED_ITEMS, | |
5050 | ALLOC_CHUNK, | |
5051 | }; | |
5052 | ||
957780eb JB |
5053 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, |
5054 | struct btrfs_space_info *space_info, | |
5055 | struct reserve_ticket *ticket) | |
5056 | { | |
5057 | u64 to_reclaim; | |
8a1bbe1d | 5058 | int flush_state; |
957780eb JB |
5059 | |
5060 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5061 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5062 | false); | |
957780eb JB |
5063 | if (!to_reclaim) { |
5064 | spin_unlock(&space_info->lock); | |
5065 | return; | |
5066 | } | |
5067 | spin_unlock(&space_info->lock); | |
5068 | ||
8a1bbe1d | 5069 | flush_state = 0; |
21c7e756 | 5070 | do { |
8a1bbe1d JB |
5071 | flush_space(fs_info, space_info, to_reclaim, |
5072 | priority_flush_states[flush_state]); | |
21c7e756 | 5073 | flush_state++; |
957780eb JB |
5074 | spin_lock(&space_info->lock); |
5075 | if (ticket->bytes == 0) { | |
5076 | spin_unlock(&space_info->lock); | |
21c7e756 | 5077 | return; |
957780eb JB |
5078 | } |
5079 | spin_unlock(&space_info->lock); | |
8a1bbe1d | 5080 | } while (flush_state < ARRAY_SIZE(priority_flush_states)); |
21c7e756 MX |
5081 | } |
5082 | ||
957780eb JB |
5083 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5084 | struct btrfs_space_info *space_info, | |
f91587e4 | 5085 | struct reserve_ticket *ticket) |
957780eb | 5086 | |
21c7e756 | 5087 | { |
957780eb | 5088 | DEFINE_WAIT(wait); |
f91587e4 | 5089 | u64 reclaim_bytes = 0; |
957780eb JB |
5090 | int ret = 0; |
5091 | ||
5092 | spin_lock(&space_info->lock); | |
5093 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5094 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5095 | if (ret) { | |
5096 | ret = -EINTR; | |
5097 | break; | |
5098 | } | |
5099 | spin_unlock(&space_info->lock); | |
5100 | ||
5101 | schedule(); | |
5102 | ||
5103 | finish_wait(&ticket->wait, &wait); | |
5104 | spin_lock(&space_info->lock); | |
5105 | } | |
5106 | if (!ret) | |
5107 | ret = ticket->error; | |
5108 | if (!list_empty(&ticket->list)) | |
5109 | list_del_init(&ticket->list); | |
f91587e4 JB |
5110 | if (ticket->bytes && ticket->bytes < ticket->orig_bytes) |
5111 | reclaim_bytes = ticket->orig_bytes - ticket->bytes; | |
957780eb JB |
5112 | spin_unlock(&space_info->lock); |
5113 | ||
f91587e4 JB |
5114 | if (reclaim_bytes) |
5115 | space_info_add_old_bytes(fs_info, space_info, reclaim_bytes); | |
957780eb | 5116 | return ret; |
21c7e756 MX |
5117 | } |
5118 | ||
4a92b1b8 JB |
5119 | /** |
5120 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5121 | * @root - the root we're allocating for | |
957780eb | 5122 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5123 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5124 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5125 | * |
01327610 | 5126 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5127 | * with the block_rsv. If there is not enough space it will make an attempt to |
5128 | * flush out space to make room. It will do this by flushing delalloc if | |
5129 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5130 | * regain reservations will be made and this will fail if there is not enough | |
5131 | * space already. | |
8bb8ab2e | 5132 | */ |
c1c4919b | 5133 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5134 | struct btrfs_space_info *space_info, |
5135 | u64 orig_bytes, | |
c1c4919b JM |
5136 | enum btrfs_reserve_flush_enum flush, |
5137 | bool system_chunk) | |
9ed74f2d | 5138 | { |
957780eb | 5139 | struct reserve_ticket ticket; |
2bf64758 | 5140 | u64 used; |
f91587e4 | 5141 | u64 reclaim_bytes = 0; |
8bb8ab2e | 5142 | int ret = 0; |
9ed74f2d | 5143 | |
957780eb | 5144 | ASSERT(orig_bytes); |
8ca17f0f | 5145 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5146 | |
8bb8ab2e | 5147 | spin_lock(&space_info->lock); |
fdb5effd | 5148 | ret = -ENOSPC; |
4136135b | 5149 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5150 | |
8bb8ab2e | 5151 | /* |
957780eb JB |
5152 | * If we have enough space then hooray, make our reservation and carry |
5153 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5154 | * If not things get more complicated. | |
8bb8ab2e | 5155 | */ |
957780eb | 5156 | if (used + orig_bytes <= space_info->total_bytes) { |
9f9b8e8d | 5157 | update_bytes_may_use(space_info, orig_bytes); |
0b246afa JM |
5158 | trace_btrfs_space_reservation(fs_info, "space_info", |
5159 | space_info->flags, orig_bytes, 1); | |
957780eb | 5160 | ret = 0; |
c1c4919b JM |
5161 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5162 | system_chunk)) { | |
9f9b8e8d | 5163 | update_bytes_may_use(space_info, orig_bytes); |
0b246afa JM |
5164 | trace_btrfs_space_reservation(fs_info, "space_info", |
5165 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5166 | ret = 0; |
2bf64758 JB |
5167 | } |
5168 | ||
8bb8ab2e | 5169 | /* |
957780eb JB |
5170 | * If we couldn't make a reservation then setup our reservation ticket |
5171 | * and kick the async worker if it's not already running. | |
08e007d2 | 5172 | * |
957780eb JB |
5173 | * If we are a priority flusher then we just need to add our ticket to |
5174 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5175 | */ |
72bcd99d | 5176 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
f91587e4 | 5177 | ticket.orig_bytes = orig_bytes; |
957780eb JB |
5178 | ticket.bytes = orig_bytes; |
5179 | ticket.error = 0; | |
5180 | init_waitqueue_head(&ticket.wait); | |
5181 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5182 | list_add_tail(&ticket.list, &space_info->tickets); | |
5183 | if (!space_info->flush) { | |
5184 | space_info->flush = 1; | |
0b246afa | 5185 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5186 | space_info->flags, |
5187 | orig_bytes, flush, | |
5188 | "enospc"); | |
957780eb | 5189 | queue_work(system_unbound_wq, |
c1c4919b | 5190 | &fs_info->async_reclaim_work); |
957780eb JB |
5191 | } |
5192 | } else { | |
5193 | list_add_tail(&ticket.list, | |
5194 | &space_info->priority_tickets); | |
5195 | } | |
21c7e756 MX |
5196 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5197 | used += orig_bytes; | |
f6acfd50 JB |
5198 | /* |
5199 | * We will do the space reservation dance during log replay, | |
5200 | * which means we won't have fs_info->fs_root set, so don't do | |
5201 | * the async reclaim as we will panic. | |
5202 | */ | |
0b246afa | 5203 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5204 | need_do_async_reclaim(fs_info, space_info, |
5205 | used, system_chunk) && | |
0b246afa JM |
5206 | !work_busy(&fs_info->async_reclaim_work)) { |
5207 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5208 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5209 | queue_work(system_unbound_wq, |
0b246afa | 5210 | &fs_info->async_reclaim_work); |
f376df2b | 5211 | } |
8bb8ab2e | 5212 | } |
f0486c68 | 5213 | spin_unlock(&space_info->lock); |
08e007d2 | 5214 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5215 | return ret; |
f0486c68 | 5216 | |
957780eb | 5217 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
f91587e4 | 5218 | return wait_reserve_ticket(fs_info, space_info, &ticket); |
08e007d2 | 5219 | |
957780eb | 5220 | ret = 0; |
0b246afa | 5221 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5222 | spin_lock(&space_info->lock); |
5223 | if (ticket.bytes) { | |
f91587e4 JB |
5224 | if (ticket.bytes < orig_bytes) |
5225 | reclaim_bytes = orig_bytes - ticket.bytes; | |
957780eb JB |
5226 | list_del_init(&ticket.list); |
5227 | ret = -ENOSPC; | |
5228 | } | |
5229 | spin_unlock(&space_info->lock); | |
f91587e4 JB |
5230 | |
5231 | if (reclaim_bytes) | |
5232 | space_info_add_old_bytes(fs_info, space_info, reclaim_bytes); | |
957780eb JB |
5233 | ASSERT(list_empty(&ticket.list)); |
5234 | return ret; | |
5235 | } | |
8bb8ab2e | 5236 | |
957780eb JB |
5237 | /** |
5238 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5239 | * @root - the root we're allocating for | |
5240 | * @block_rsv - the block_rsv we're allocating for | |
5241 | * @orig_bytes - the number of bytes we want | |
5242 | * @flush - whether or not we can flush to make our reservation | |
5243 | * | |
52042d8e | 5244 | * This will reserve orig_bytes number of bytes from the space info associated |
957780eb JB |
5245 | * with the block_rsv. If there is not enough space it will make an attempt to |
5246 | * flush out space to make room. It will do this by flushing delalloc if | |
5247 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5248 | * regain reservations will be made and this will fail if there is not enough | |
5249 | * space already. | |
5250 | */ | |
5251 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5252 | struct btrfs_block_rsv *block_rsv, | |
5253 | u64 orig_bytes, | |
5254 | enum btrfs_reserve_flush_enum flush) | |
5255 | { | |
0b246afa JM |
5256 | struct btrfs_fs_info *fs_info = root->fs_info; |
5257 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5258 | int ret; |
c1c4919b | 5259 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5260 | |
c1c4919b JM |
5261 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5262 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5263 | if (ret == -ENOSPC && |
5264 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5265 | if (block_rsv != global_rsv && |
5266 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5267 | ret = 0; | |
5268 | } | |
9a3daff3 | 5269 | if (ret == -ENOSPC) { |
0b246afa | 5270 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5271 | block_rsv->space_info->flags, |
5272 | orig_bytes, 1); | |
9a3daff3 NB |
5273 | |
5274 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
5275 | dump_space_info(fs_info, block_rsv->space_info, | |
5276 | orig_bytes, 0); | |
5277 | } | |
f0486c68 YZ |
5278 | return ret; |
5279 | } | |
5280 | ||
79787eaa JM |
5281 | static struct btrfs_block_rsv *get_block_rsv( |
5282 | const struct btrfs_trans_handle *trans, | |
5283 | const struct btrfs_root *root) | |
f0486c68 | 5284 | { |
0b246afa | 5285 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5286 | struct btrfs_block_rsv *block_rsv = NULL; |
5287 | ||
e9cf439f | 5288 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5289 | (root == fs_info->csum_root && trans->adding_csums) || |
5290 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5291 | block_rsv = trans->block_rsv; |
5292 | ||
4c13d758 | 5293 | if (!block_rsv) |
f0486c68 YZ |
5294 | block_rsv = root->block_rsv; |
5295 | ||
5296 | if (!block_rsv) | |
0b246afa | 5297 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5298 | |
5299 | return block_rsv; | |
5300 | } | |
5301 | ||
5302 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5303 | u64 num_bytes) | |
5304 | { | |
5305 | int ret = -ENOSPC; | |
5306 | spin_lock(&block_rsv->lock); | |
5307 | if (block_rsv->reserved >= num_bytes) { | |
5308 | block_rsv->reserved -= num_bytes; | |
5309 | if (block_rsv->reserved < block_rsv->size) | |
5310 | block_rsv->full = 0; | |
5311 | ret = 0; | |
5312 | } | |
5313 | spin_unlock(&block_rsv->lock); | |
5314 | return ret; | |
5315 | } | |
5316 | ||
5317 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3a584174 | 5318 | u64 num_bytes, bool update_size) |
f0486c68 YZ |
5319 | { |
5320 | spin_lock(&block_rsv->lock); | |
5321 | block_rsv->reserved += num_bytes; | |
5322 | if (update_size) | |
5323 | block_rsv->size += num_bytes; | |
5324 | else if (block_rsv->reserved >= block_rsv->size) | |
5325 | block_rsv->full = 1; | |
5326 | spin_unlock(&block_rsv->lock); | |
5327 | } | |
5328 | ||
d52be818 JB |
5329 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5330 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5331 | int min_factor) | |
5332 | { | |
5333 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5334 | u64 min_bytes; | |
5335 | ||
5336 | if (global_rsv->space_info != dest->space_info) | |
5337 | return -ENOSPC; | |
5338 | ||
5339 | spin_lock(&global_rsv->lock); | |
5340 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5341 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5342 | spin_unlock(&global_rsv->lock); | |
5343 | return -ENOSPC; | |
5344 | } | |
5345 | global_rsv->reserved -= num_bytes; | |
5346 | if (global_rsv->reserved < global_rsv->size) | |
5347 | global_rsv->full = 0; | |
5348 | spin_unlock(&global_rsv->lock); | |
5349 | ||
3a584174 | 5350 | block_rsv_add_bytes(dest, num_bytes, true); |
d52be818 JB |
5351 | return 0; |
5352 | } | |
5353 | ||
ba2c4d4e JB |
5354 | /** |
5355 | * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv. | |
5356 | * @fs_info - the fs info for our fs. | |
5357 | * @src - the source block rsv to transfer from. | |
5358 | * @num_bytes - the number of bytes to transfer. | |
5359 | * | |
5360 | * This transfers up to the num_bytes amount from the src rsv to the | |
5361 | * delayed_refs_rsv. Any extra bytes are returned to the space info. | |
5362 | */ | |
5363 | void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info, | |
5364 | struct btrfs_block_rsv *src, | |
5365 | u64 num_bytes) | |
5366 | { | |
5367 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; | |
5368 | u64 to_free = 0; | |
5369 | ||
5370 | spin_lock(&src->lock); | |
5371 | src->reserved -= num_bytes; | |
5372 | src->size -= num_bytes; | |
5373 | spin_unlock(&src->lock); | |
5374 | ||
5375 | spin_lock(&delayed_refs_rsv->lock); | |
5376 | if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) { | |
5377 | u64 delta = delayed_refs_rsv->size - | |
5378 | delayed_refs_rsv->reserved; | |
5379 | if (num_bytes > delta) { | |
5380 | to_free = num_bytes - delta; | |
5381 | num_bytes = delta; | |
5382 | } | |
5383 | } else { | |
5384 | to_free = num_bytes; | |
5385 | num_bytes = 0; | |
5386 | } | |
5387 | ||
5388 | if (num_bytes) | |
5389 | delayed_refs_rsv->reserved += num_bytes; | |
5390 | if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size) | |
5391 | delayed_refs_rsv->full = 1; | |
5392 | spin_unlock(&delayed_refs_rsv->lock); | |
5393 | ||
5394 | if (num_bytes) | |
5395 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5396 | 0, num_bytes, 1); | |
5397 | if (to_free) | |
5398 | space_info_add_old_bytes(fs_info, delayed_refs_rsv->space_info, | |
5399 | to_free); | |
5400 | } | |
5401 | ||
5402 | /** | |
5403 | * btrfs_delayed_refs_rsv_refill - refill based on our delayed refs usage. | |
5404 | * @fs_info - the fs_info for our fs. | |
5405 | * @flush - control how we can flush for this reservation. | |
5406 | * | |
5407 | * This will refill the delayed block_rsv up to 1 items size worth of space and | |
5408 | * will return -ENOSPC if we can't make the reservation. | |
5409 | */ | |
5410 | int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info, | |
5411 | enum btrfs_reserve_flush_enum flush) | |
5412 | { | |
5413 | struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; | |
5414 | u64 limit = btrfs_calc_trans_metadata_size(fs_info, 1); | |
5415 | u64 num_bytes = 0; | |
5416 | int ret = -ENOSPC; | |
5417 | ||
5418 | spin_lock(&block_rsv->lock); | |
5419 | if (block_rsv->reserved < block_rsv->size) { | |
5420 | num_bytes = block_rsv->size - block_rsv->reserved; | |
5421 | num_bytes = min(num_bytes, limit); | |
5422 | } | |
5423 | spin_unlock(&block_rsv->lock); | |
5424 | ||
5425 | if (!num_bytes) | |
5426 | return 0; | |
5427 | ||
5428 | ret = reserve_metadata_bytes(fs_info->extent_root, block_rsv, | |
5429 | num_bytes, flush); | |
5430 | if (ret) | |
5431 | return ret; | |
5432 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
5433 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5434 | 0, num_bytes, 1); | |
5435 | return 0; | |
5436 | } | |
5437 | ||
957780eb JB |
5438 | /* |
5439 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5440 | * basically when we're returning space from block_rsv's. | |
5441 | */ | |
5442 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5443 | struct btrfs_space_info *space_info, | |
5444 | u64 num_bytes) | |
5445 | { | |
5446 | struct reserve_ticket *ticket; | |
5447 | struct list_head *head; | |
5448 | u64 used; | |
5449 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5450 | bool check_overcommit = false; | |
5451 | ||
5452 | spin_lock(&space_info->lock); | |
5453 | head = &space_info->priority_tickets; | |
5454 | ||
5455 | /* | |
5456 | * If we are over our limit then we need to check and see if we can | |
5457 | * overcommit, and if we can't then we just need to free up our space | |
5458 | * and not satisfy any requests. | |
5459 | */ | |
0eee8a49 | 5460 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5461 | if (used - num_bytes >= space_info->total_bytes) |
5462 | check_overcommit = true; | |
5463 | again: | |
5464 | while (!list_empty(head) && num_bytes) { | |
5465 | ticket = list_first_entry(head, struct reserve_ticket, | |
5466 | list); | |
5467 | /* | |
5468 | * We use 0 bytes because this space is already reserved, so | |
5469 | * adding the ticket space would be a double count. | |
5470 | */ | |
5471 | if (check_overcommit && | |
c1c4919b | 5472 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5473 | break; |
5474 | if (num_bytes >= ticket->bytes) { | |
5475 | list_del_init(&ticket->list); | |
5476 | num_bytes -= ticket->bytes; | |
5477 | ticket->bytes = 0; | |
ce129655 | 5478 | space_info->tickets_id++; |
957780eb JB |
5479 | wake_up(&ticket->wait); |
5480 | } else { | |
5481 | ticket->bytes -= num_bytes; | |
5482 | num_bytes = 0; | |
5483 | } | |
5484 | } | |
5485 | ||
5486 | if (num_bytes && head == &space_info->priority_tickets) { | |
5487 | head = &space_info->tickets; | |
5488 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5489 | goto again; | |
5490 | } | |
9f9b8e8d | 5491 | update_bytes_may_use(space_info, -num_bytes); |
957780eb JB |
5492 | trace_btrfs_space_reservation(fs_info, "space_info", |
5493 | space_info->flags, num_bytes, 0); | |
5494 | spin_unlock(&space_info->lock); | |
5495 | } | |
5496 | ||
5497 | /* | |
5498 | * This is for newly allocated space that isn't accounted in | |
5499 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5500 | * we use this helper. | |
5501 | */ | |
5502 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5503 | struct btrfs_space_info *space_info, | |
5504 | u64 num_bytes) | |
5505 | { | |
5506 | struct reserve_ticket *ticket; | |
5507 | struct list_head *head = &space_info->priority_tickets; | |
5508 | ||
5509 | again: | |
5510 | while (!list_empty(head) && num_bytes) { | |
5511 | ticket = list_first_entry(head, struct reserve_ticket, | |
5512 | list); | |
5513 | if (num_bytes >= ticket->bytes) { | |
5514 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5515 | space_info->flags, | |
5516 | ticket->bytes, 1); | |
5517 | list_del_init(&ticket->list); | |
5518 | num_bytes -= ticket->bytes; | |
9f9b8e8d | 5519 | update_bytes_may_use(space_info, ticket->bytes); |
957780eb | 5520 | ticket->bytes = 0; |
ce129655 | 5521 | space_info->tickets_id++; |
957780eb JB |
5522 | wake_up(&ticket->wait); |
5523 | } else { | |
5524 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5525 | space_info->flags, | |
5526 | num_bytes, 1); | |
9f9b8e8d | 5527 | update_bytes_may_use(space_info, num_bytes); |
957780eb JB |
5528 | ticket->bytes -= num_bytes; |
5529 | num_bytes = 0; | |
5530 | } | |
5531 | } | |
5532 | ||
5533 | if (num_bytes && head == &space_info->priority_tickets) { | |
5534 | head = &space_info->tickets; | |
5535 | goto again; | |
5536 | } | |
5537 | } | |
5538 | ||
69fe2d75 | 5539 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
8c2a3ca2 | 5540 | struct btrfs_block_rsv *block_rsv, |
ff6bc37e QW |
5541 | struct btrfs_block_rsv *dest, u64 num_bytes, |
5542 | u64 *qgroup_to_release_ret) | |
f0486c68 YZ |
5543 | { |
5544 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
ff6bc37e | 5545 | u64 qgroup_to_release = 0; |
69fe2d75 | 5546 | u64 ret; |
f0486c68 YZ |
5547 | |
5548 | spin_lock(&block_rsv->lock); | |
ff6bc37e | 5549 | if (num_bytes == (u64)-1) { |
f0486c68 | 5550 | num_bytes = block_rsv->size; |
ff6bc37e QW |
5551 | qgroup_to_release = block_rsv->qgroup_rsv_size; |
5552 | } | |
f0486c68 YZ |
5553 | block_rsv->size -= num_bytes; |
5554 | if (block_rsv->reserved >= block_rsv->size) { | |
5555 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5556 | block_rsv->reserved = block_rsv->size; | |
5557 | block_rsv->full = 1; | |
5558 | } else { | |
5559 | num_bytes = 0; | |
5560 | } | |
ff6bc37e QW |
5561 | if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) { |
5562 | qgroup_to_release = block_rsv->qgroup_rsv_reserved - | |
5563 | block_rsv->qgroup_rsv_size; | |
5564 | block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size; | |
5565 | } else { | |
5566 | qgroup_to_release = 0; | |
5567 | } | |
f0486c68 YZ |
5568 | spin_unlock(&block_rsv->lock); |
5569 | ||
69fe2d75 | 5570 | ret = num_bytes; |
f0486c68 YZ |
5571 | if (num_bytes > 0) { |
5572 | if (dest) { | |
e9e22899 JB |
5573 | spin_lock(&dest->lock); |
5574 | if (!dest->full) { | |
5575 | u64 bytes_to_add; | |
5576 | ||
5577 | bytes_to_add = dest->size - dest->reserved; | |
5578 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5579 | dest->reserved += bytes_to_add; | |
5580 | if (dest->reserved >= dest->size) | |
5581 | dest->full = 1; | |
5582 | num_bytes -= bytes_to_add; | |
5583 | } | |
5584 | spin_unlock(&dest->lock); | |
5585 | } | |
957780eb JB |
5586 | if (num_bytes) |
5587 | space_info_add_old_bytes(fs_info, space_info, | |
5588 | num_bytes); | |
9ed74f2d | 5589 | } |
ff6bc37e QW |
5590 | if (qgroup_to_release_ret) |
5591 | *qgroup_to_release_ret = qgroup_to_release; | |
69fe2d75 | 5592 | return ret; |
f0486c68 | 5593 | } |
4e06bdd6 | 5594 | |
25d609f8 JB |
5595 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5596 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
3a584174 | 5597 | bool update_size) |
f0486c68 YZ |
5598 | { |
5599 | int ret; | |
9ed74f2d | 5600 | |
f0486c68 YZ |
5601 | ret = block_rsv_use_bytes(src, num_bytes); |
5602 | if (ret) | |
5603 | return ret; | |
9ed74f2d | 5604 | |
25d609f8 | 5605 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5606 | return 0; |
5607 | } | |
5608 | ||
66d8f3dd | 5609 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5610 | { |
f0486c68 YZ |
5611 | memset(rsv, 0, sizeof(*rsv)); |
5612 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5613 | rsv->type = type; |
f0486c68 YZ |
5614 | } |
5615 | ||
69fe2d75 JB |
5616 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
5617 | struct btrfs_block_rsv *rsv, | |
5618 | unsigned short type) | |
5619 | { | |
5620 | btrfs_init_block_rsv(rsv, type); | |
5621 | rsv->space_info = __find_space_info(fs_info, | |
5622 | BTRFS_BLOCK_GROUP_METADATA); | |
5623 | } | |
5624 | ||
2ff7e61e | 5625 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5626 | unsigned short type) |
f0486c68 YZ |
5627 | { |
5628 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5629 | |
f0486c68 YZ |
5630 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5631 | if (!block_rsv) | |
5632 | return NULL; | |
9ed74f2d | 5633 | |
69fe2d75 | 5634 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
f0486c68 YZ |
5635 | return block_rsv; |
5636 | } | |
9ed74f2d | 5637 | |
2ff7e61e | 5638 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5639 | struct btrfs_block_rsv *rsv) |
5640 | { | |
2aaa6655 JB |
5641 | if (!rsv) |
5642 | return; | |
2ff7e61e | 5643 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5644 | kfree(rsv); |
9ed74f2d JB |
5645 | } |
5646 | ||
08e007d2 MX |
5647 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5648 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5649 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5650 | { |
f0486c68 | 5651 | int ret; |
9ed74f2d | 5652 | |
f0486c68 YZ |
5653 | if (num_bytes == 0) |
5654 | return 0; | |
8bb8ab2e | 5655 | |
61b520a9 | 5656 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5a2cb25a | 5657 | if (!ret) |
3a584174 | 5658 | block_rsv_add_bytes(block_rsv, num_bytes, true); |
9ed74f2d | 5659 | |
f0486c68 | 5660 | return ret; |
f0486c68 | 5661 | } |
9ed74f2d | 5662 | |
2ff7e61e | 5663 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5664 | { |
5665 | u64 num_bytes = 0; | |
f0486c68 | 5666 | int ret = -ENOSPC; |
9ed74f2d | 5667 | |
f0486c68 YZ |
5668 | if (!block_rsv) |
5669 | return 0; | |
9ed74f2d | 5670 | |
f0486c68 | 5671 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5672 | num_bytes = div_factor(block_rsv->size, min_factor); |
5673 | if (block_rsv->reserved >= num_bytes) | |
5674 | ret = 0; | |
5675 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5676 | |
36ba022a JB |
5677 | return ret; |
5678 | } | |
5679 | ||
08e007d2 MX |
5680 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5681 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5682 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5683 | { |
5684 | u64 num_bytes = 0; | |
5685 | int ret = -ENOSPC; | |
5686 | ||
5687 | if (!block_rsv) | |
5688 | return 0; | |
5689 | ||
5690 | spin_lock(&block_rsv->lock); | |
5691 | num_bytes = min_reserved; | |
13553e52 | 5692 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5693 | ret = 0; |
13553e52 | 5694 | else |
f0486c68 | 5695 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5696 | spin_unlock(&block_rsv->lock); |
13553e52 | 5697 | |
f0486c68 YZ |
5698 | if (!ret) |
5699 | return 0; | |
5700 | ||
aa38a711 | 5701 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 | 5702 | if (!ret) { |
3a584174 | 5703 | block_rsv_add_bytes(block_rsv, num_bytes, false); |
f0486c68 | 5704 | return 0; |
6a63209f | 5705 | } |
9ed74f2d | 5706 | |
13553e52 | 5707 | return ret; |
f0486c68 YZ |
5708 | } |
5709 | ||
5df11363 JB |
5710 | static void calc_refill_bytes(struct btrfs_block_rsv *block_rsv, |
5711 | u64 *metadata_bytes, u64 *qgroup_bytes) | |
5712 | { | |
5713 | *metadata_bytes = 0; | |
5714 | *qgroup_bytes = 0; | |
5715 | ||
5716 | spin_lock(&block_rsv->lock); | |
5717 | if (block_rsv->reserved < block_rsv->size) | |
5718 | *metadata_bytes = block_rsv->size - block_rsv->reserved; | |
5719 | if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size) | |
5720 | *qgroup_bytes = block_rsv->qgroup_rsv_size - | |
5721 | block_rsv->qgroup_rsv_reserved; | |
5722 | spin_unlock(&block_rsv->lock); | |
5723 | } | |
5724 | ||
69fe2d75 JB |
5725 | /** |
5726 | * btrfs_inode_rsv_refill - refill the inode block rsv. | |
5727 | * @inode - the inode we are refilling. | |
52042d8e | 5728 | * @flush - the flushing restriction. |
69fe2d75 JB |
5729 | * |
5730 | * Essentially the same as btrfs_block_rsv_refill, except it uses the | |
5731 | * block_rsv->size as the minimum size. We'll either refill the missing amount | |
52042d8e | 5732 | * or return if we already have enough space. This will also handle the reserve |
69fe2d75 JB |
5733 | * tracepoint for the reserved amount. |
5734 | */ | |
3f2dd7a0 QW |
5735 | static int btrfs_inode_rsv_refill(struct btrfs_inode *inode, |
5736 | enum btrfs_reserve_flush_enum flush) | |
69fe2d75 JB |
5737 | { |
5738 | struct btrfs_root *root = inode->root; | |
5739 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5df11363 JB |
5740 | u64 num_bytes, last = 0; |
5741 | u64 qgroup_num_bytes; | |
69fe2d75 JB |
5742 | int ret = -ENOSPC; |
5743 | ||
5df11363 | 5744 | calc_refill_bytes(block_rsv, &num_bytes, &qgroup_num_bytes); |
69fe2d75 JB |
5745 | if (num_bytes == 0) |
5746 | return 0; | |
5747 | ||
5df11363 JB |
5748 | do { |
5749 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, | |
5750 | true); | |
5751 | if (ret) | |
5752 | return ret; | |
5753 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); | |
5754 | if (ret) { | |
5755 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
5756 | last = num_bytes; | |
5757 | /* | |
5758 | * If we are fragmented we can end up with a lot of | |
5759 | * outstanding extents which will make our size be much | |
5760 | * larger than our reserved amount. | |
5761 | * | |
5762 | * If the reservation happens here, it might be very | |
5763 | * big though not needed in the end, if the delalloc | |
5764 | * flushing happens. | |
5765 | * | |
5766 | * If this is the case try and do the reserve again. | |
5767 | */ | |
5768 | if (flush == BTRFS_RESERVE_FLUSH_ALL) | |
5769 | calc_refill_bytes(block_rsv, &num_bytes, | |
5770 | &qgroup_num_bytes); | |
5771 | if (num_bytes == 0) | |
5772 | return 0; | |
5773 | } | |
5774 | } while (ret && last != num_bytes); | |
5775 | ||
69fe2d75 | 5776 | if (!ret) { |
3a584174 | 5777 | block_rsv_add_bytes(block_rsv, num_bytes, false); |
69fe2d75 JB |
5778 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
5779 | btrfs_ino(inode), num_bytes, 1); | |
ff6bc37e QW |
5780 | |
5781 | /* Don't forget to increase qgroup_rsv_reserved */ | |
5782 | spin_lock(&block_rsv->lock); | |
5783 | block_rsv->qgroup_rsv_reserved += qgroup_num_bytes; | |
5784 | spin_unlock(&block_rsv->lock); | |
5df11363 | 5785 | } |
69fe2d75 JB |
5786 | return ret; |
5787 | } | |
5788 | ||
ba2c4d4e JB |
5789 | static u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
5790 | struct btrfs_block_rsv *block_rsv, | |
5791 | u64 num_bytes, u64 *qgroup_to_release) | |
5792 | { | |
5793 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5794 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv; | |
5795 | struct btrfs_block_rsv *target = delayed_rsv; | |
5796 | ||
5797 | if (target->full || target == block_rsv) | |
5798 | target = global_rsv; | |
5799 | ||
5800 | if (block_rsv->space_info != target->space_info) | |
5801 | target = NULL; | |
5802 | ||
5803 | return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes, | |
5804 | qgroup_to_release); | |
5805 | } | |
5806 | ||
5807 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, | |
5808 | struct btrfs_block_rsv *block_rsv, | |
5809 | u64 num_bytes) | |
5810 | { | |
5811 | __btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL); | |
5812 | } | |
5813 | ||
69fe2d75 JB |
5814 | /** |
5815 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5816 | * @inode - the inode we need to release from. | |
43b18595 QW |
5817 | * @qgroup_free - free or convert qgroup meta. |
5818 | * Unlike normal operation, qgroup meta reservation needs to know if we are | |
5819 | * freeing qgroup reservation or just converting it into per-trans. Normally | |
5820 | * @qgroup_free is true for error handling, and false for normal release. | |
69fe2d75 JB |
5821 | * |
5822 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5823 | * tracepoint for the reservation. | |
5824 | */ | |
43b18595 | 5825 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
69fe2d75 JB |
5826 | { |
5827 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
69fe2d75 JB |
5828 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
5829 | u64 released = 0; | |
ff6bc37e | 5830 | u64 qgroup_to_release = 0; |
69fe2d75 JB |
5831 | |
5832 | /* | |
5833 | * Since we statically set the block_rsv->size we just want to say we | |
5834 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5835 | * the size free'd. | |
5836 | */ | |
ba2c4d4e JB |
5837 | released = __btrfs_block_rsv_release(fs_info, block_rsv, 0, |
5838 | &qgroup_to_release); | |
69fe2d75 JB |
5839 | if (released > 0) |
5840 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5841 | btrfs_ino(inode), released, 0); | |
43b18595 | 5842 | if (qgroup_free) |
ff6bc37e | 5843 | btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); |
43b18595 | 5844 | else |
ff6bc37e QW |
5845 | btrfs_qgroup_convert_reserved_meta(inode->root, |
5846 | qgroup_to_release); | |
69fe2d75 JB |
5847 | } |
5848 | ||
ba2c4d4e JB |
5849 | /** |
5850 | * btrfs_delayed_refs_rsv_release - release a ref head's reservation. | |
5851 | * @fs_info - the fs_info for our fs. | |
5852 | * @nr - the number of items to drop. | |
5853 | * | |
5854 | * This drops the delayed ref head's count from the delayed refs rsv and frees | |
5855 | * any excess reservation we had. | |
5856 | */ | |
5857 | void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr) | |
f0486c68 | 5858 | { |
ba2c4d4e | 5859 | struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; |
0b246afa | 5860 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
ba2c4d4e JB |
5861 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, nr); |
5862 | u64 released = 0; | |
0b246afa | 5863 | |
ba2c4d4e JB |
5864 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, |
5865 | num_bytes, NULL); | |
5866 | if (released) | |
5867 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5868 | 0, released, 0); | |
6a63209f JB |
5869 | } |
5870 | ||
8929ecfa YZ |
5871 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5872 | { | |
5873 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5874 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5875 | u64 num_bytes; | |
6a63209f | 5876 | |
ae2e4728 JB |
5877 | /* |
5878 | * The global block rsv is based on the size of the extent tree, the | |
5879 | * checksum tree and the root tree. If the fs is empty we want to set | |
5880 | * it to a minimal amount for safety. | |
5881 | */ | |
5882 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5883 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5884 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5885 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5886 | |
8929ecfa | 5887 | spin_lock(&sinfo->lock); |
1f699d38 | 5888 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5889 | |
ee22184b | 5890 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5891 | |
fb4b10e5 | 5892 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5893 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5894 | if (sinfo->total_bytes > num_bytes) { |
5895 | num_bytes = sinfo->total_bytes - num_bytes; | |
5896 | num_bytes = min(num_bytes, | |
5897 | block_rsv->size - block_rsv->reserved); | |
5898 | block_rsv->reserved += num_bytes; | |
9f9b8e8d | 5899 | update_bytes_may_use(sinfo, num_bytes); |
fb4b10e5 JB |
5900 | trace_btrfs_space_reservation(fs_info, "space_info", |
5901 | sinfo->flags, num_bytes, | |
5902 | 1); | |
5903 | } | |
5904 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5905 | num_bytes = block_rsv->reserved - block_rsv->size; |
9f9b8e8d | 5906 | update_bytes_may_use(sinfo, -num_bytes); |
8c2a3ca2 | 5907 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5908 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5909 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5910 | } |
182608c8 | 5911 | |
fb4b10e5 JB |
5912 | if (block_rsv->reserved == block_rsv->size) |
5913 | block_rsv->full = 1; | |
5914 | else | |
5915 | block_rsv->full = 0; | |
5916 | ||
8929ecfa | 5917 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5918 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5919 | } |
5920 | ||
f0486c68 | 5921 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5922 | { |
f0486c68 | 5923 | struct btrfs_space_info *space_info; |
6a63209f | 5924 | |
f0486c68 YZ |
5925 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5926 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5927 | |
f0486c68 | 5928 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5929 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5930 | fs_info->trans_block_rsv.space_info = space_info; |
5931 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5932 | fs_info->delayed_block_rsv.space_info = space_info; |
ba2c4d4e | 5933 | fs_info->delayed_refs_rsv.space_info = space_info; |
f0486c68 | 5934 | |
ba2c4d4e JB |
5935 | fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv; |
5936 | fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv; | |
8929ecfa YZ |
5937 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; |
5938 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5939 | if (fs_info->quota_root) |
5940 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5941 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5942 | |
8929ecfa | 5943 | update_global_block_rsv(fs_info); |
6a63209f JB |
5944 | } |
5945 | ||
8929ecfa | 5946 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5947 | { |
8c2a3ca2 | 5948 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
ff6bc37e | 5949 | (u64)-1, NULL); |
8929ecfa YZ |
5950 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5951 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5952 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5953 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5954 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5955 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
ba2c4d4e JB |
5956 | WARN_ON(fs_info->delayed_refs_rsv.reserved > 0); |
5957 | WARN_ON(fs_info->delayed_refs_rsv.size > 0); | |
fcb80c2a JB |
5958 | } |
5959 | ||
ba2c4d4e JB |
5960 | /* |
5961 | * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv | |
5962 | * @trans - the trans that may have generated delayed refs | |
5963 | * | |
5964 | * This is to be called anytime we may have adjusted trans->delayed_ref_updates, | |
5965 | * it'll calculate the additional size and add it to the delayed_refs_rsv. | |
5966 | */ | |
5967 | void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans) | |
5968 | { | |
5969 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
5970 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv; | |
5971 | u64 num_bytes; | |
5972 | ||
5973 | if (!trans->delayed_ref_updates) | |
5974 | return; | |
5975 | ||
5976 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, | |
5977 | trans->delayed_ref_updates); | |
5978 | spin_lock(&delayed_rsv->lock); | |
5979 | delayed_rsv->size += num_bytes; | |
5980 | delayed_rsv->full = 0; | |
5981 | spin_unlock(&delayed_rsv->lock); | |
5982 | trans->delayed_ref_updates = 0; | |
5983 | } | |
6a63209f | 5984 | |
4fbcdf66 FM |
5985 | /* |
5986 | * To be called after all the new block groups attached to the transaction | |
5987 | * handle have been created (btrfs_create_pending_block_groups()). | |
5988 | */ | |
5989 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5990 | { | |
64b63580 | 5991 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5992 | |
5993 | if (!trans->chunk_bytes_reserved) | |
5994 | return; | |
5995 | ||
5996 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5997 | ||
5998 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
ff6bc37e | 5999 | trans->chunk_bytes_reserved, NULL); |
4fbcdf66 FM |
6000 | trans->chunk_bytes_reserved = 0; |
6001 | } | |
6002 | ||
d5c12070 MX |
6003 | /* |
6004 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
6005 | * root: the root of the parent directory | |
6006 | * rsv: block reservation | |
6007 | * items: the number of items that we need do reservation | |
a5b7f429 | 6008 | * use_global_rsv: allow fallback to the global block reservation |
d5c12070 MX |
6009 | * |
6010 | * This function is used to reserve the space for snapshot/subvolume | |
6011 | * creation and deletion. Those operations are different with the | |
6012 | * common file/directory operations, they change two fs/file trees | |
6013 | * and root tree, the number of items that the qgroup reserves is | |
6014 | * different with the free space reservation. So we can not use | |
01327610 | 6015 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
6016 | */ |
6017 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
a5b7f429 | 6018 | struct btrfs_block_rsv *rsv, int items, |
ee3441b4 | 6019 | bool use_global_rsv) |
a22285a6 | 6020 | { |
a5b7f429 | 6021 | u64 qgroup_num_bytes = 0; |
d5c12070 MX |
6022 | u64 num_bytes; |
6023 | int ret; | |
0b246afa JM |
6024 | struct btrfs_fs_info *fs_info = root->fs_info; |
6025 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 6026 | |
0b246afa | 6027 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 6028 | /* One for parent inode, two for dir entries */ |
a5b7f429 LF |
6029 | qgroup_num_bytes = 3 * fs_info->nodesize; |
6030 | ret = btrfs_qgroup_reserve_meta_prealloc(root, | |
6031 | qgroup_num_bytes, true); | |
d5c12070 MX |
6032 | if (ret) |
6033 | return ret; | |
d5c12070 MX |
6034 | } |
6035 | ||
0b246afa JM |
6036 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
6037 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
6038 | BTRFS_BLOCK_GROUP_METADATA); |
6039 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
6040 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
6041 | |
6042 | if (ret == -ENOSPC && use_global_rsv) | |
3a584174 | 6043 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true); |
ee3441b4 | 6044 | |
a5b7f429 LF |
6045 | if (ret && qgroup_num_bytes) |
6046 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
d5c12070 MX |
6047 | |
6048 | return ret; | |
6049 | } | |
6050 | ||
2ff7e61e | 6051 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 6052 | struct btrfs_block_rsv *rsv) |
d5c12070 | 6053 | { |
2ff7e61e | 6054 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
6055 | } |
6056 | ||
69fe2d75 JB |
6057 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
6058 | struct btrfs_inode *inode) | |
9e0baf60 | 6059 | { |
69fe2d75 JB |
6060 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
6061 | u64 reserve_size = 0; | |
ff6bc37e | 6062 | u64 qgroup_rsv_size = 0; |
69fe2d75 JB |
6063 | u64 csum_leaves; |
6064 | unsigned outstanding_extents; | |
9e0baf60 | 6065 | |
69fe2d75 JB |
6066 | lockdep_assert_held(&inode->lock); |
6067 | outstanding_extents = inode->outstanding_extents; | |
6068 | if (outstanding_extents) | |
6069 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
6070 | outstanding_extents + 1); | |
6071 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
6072 | inode->csum_bytes); | |
6073 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
6074 | csum_leaves); | |
ff6bc37e QW |
6075 | /* |
6076 | * For qgroup rsv, the calculation is very simple: | |
6077 | * account one nodesize for each outstanding extent | |
6078 | * | |
6079 | * This is overestimating in most cases. | |
6080 | */ | |
139a5617 | 6081 | qgroup_rsv_size = (u64)outstanding_extents * fs_info->nodesize; |
9e0baf60 | 6082 | |
69fe2d75 JB |
6083 | spin_lock(&block_rsv->lock); |
6084 | block_rsv->size = reserve_size; | |
ff6bc37e | 6085 | block_rsv->qgroup_rsv_size = qgroup_rsv_size; |
69fe2d75 | 6086 | spin_unlock(&block_rsv->lock); |
0ca1f7ce | 6087 | } |
c146afad | 6088 | |
9f3db423 | 6089 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6090 | { |
3ffbd68c | 6091 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
69fe2d75 | 6092 | unsigned nr_extents; |
08e007d2 | 6093 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 6094 | int ret = 0; |
c64c2bd8 | 6095 | bool delalloc_lock = true; |
6324fbf3 | 6096 | |
c64c2bd8 JB |
6097 | /* If we are a free space inode we need to not flush since we will be in |
6098 | * the middle of a transaction commit. We also don't need the delalloc | |
6099 | * mutex since we won't race with anybody. We need this mostly to make | |
6100 | * lockdep shut its filthy mouth. | |
bac357dc JB |
6101 | * |
6102 | * If we have a transaction open (can happen if we call truncate_block | |
6103 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
6104 | */ |
6105 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 6106 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 6107 | delalloc_lock = false; |
da07d4ab NB |
6108 | } else { |
6109 | if (current->journal_info) | |
6110 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 6111 | |
da07d4ab NB |
6112 | if (btrfs_transaction_in_commit(fs_info)) |
6113 | schedule_timeout(1); | |
6114 | } | |
ec44a35c | 6115 | |
c64c2bd8 | 6116 | if (delalloc_lock) |
9f3db423 | 6117 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 6118 | |
0b246afa | 6119 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 JB |
6120 | |
6121 | /* Add our new extents and calculate the new rsv size. */ | |
9f3db423 | 6122 | spin_lock(&inode->lock); |
69fe2d75 | 6123 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 6124 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
6125 | inode->csum_bytes += num_bytes; |
6126 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6127 | spin_unlock(&inode->lock); |
57a45ced | 6128 | |
69fe2d75 | 6129 | ret = btrfs_inode_rsv_refill(inode, flush); |
43b18595 | 6130 | if (unlikely(ret)) |
88e081bf | 6131 | goto out_fail; |
25179201 | 6132 | |
c64c2bd8 | 6133 | if (delalloc_lock) |
9f3db423 | 6134 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 6135 | return 0; |
88e081bf WS |
6136 | |
6137 | out_fail: | |
9f3db423 | 6138 | spin_lock(&inode->lock); |
8b62f87b JB |
6139 | nr_extents = count_max_extents(num_bytes); |
6140 | btrfs_mod_outstanding_extents(inode, -nr_extents); | |
69fe2d75 JB |
6141 | inode->csum_bytes -= num_bytes; |
6142 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6143 | spin_unlock(&inode->lock); |
88e081bf | 6144 | |
43b18595 | 6145 | btrfs_inode_rsv_release(inode, true); |
88e081bf | 6146 | if (delalloc_lock) |
9f3db423 | 6147 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6148 | return ret; |
0ca1f7ce YZ |
6149 | } |
6150 | ||
7709cde3 JB |
6151 | /** |
6152 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
6153 | * @inode: the inode to release the reservation for. |
6154 | * @num_bytes: the number of bytes we are releasing. | |
43b18595 | 6155 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
7709cde3 JB |
6156 | * |
6157 | * This will release the metadata reservation for an inode. This can be called | |
6158 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 6159 | * reservations, or on error for the same reason. |
7709cde3 | 6160 | */ |
43b18595 QW |
6161 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
6162 | bool qgroup_free) | |
0ca1f7ce | 6163 | { |
3ffbd68c | 6164 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
0ca1f7ce | 6165 | |
0b246afa | 6166 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6167 | spin_lock(&inode->lock); |
69fe2d75 JB |
6168 | inode->csum_bytes -= num_bytes; |
6169 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 6170 | spin_unlock(&inode->lock); |
0ca1f7ce | 6171 | |
0b246afa | 6172 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6173 | return; |
6174 | ||
43b18595 | 6175 | btrfs_inode_rsv_release(inode, qgroup_free); |
0ca1f7ce YZ |
6176 | } |
6177 | ||
8b62f87b JB |
6178 | /** |
6179 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
6180 | * @inode: the inode to balance the reservation for. | |
6181 | * @num_bytes: the number of bytes we originally reserved with | |
43b18595 | 6182 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
8b62f87b JB |
6183 | * |
6184 | * When we reserve space we increase outstanding_extents for the extents we may | |
6185 | * add. Once we've set the range as delalloc or created our ordered extents we | |
6186 | * have outstanding_extents to track the real usage, so we use this to free our | |
6187 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
6188 | * with btrfs_delalloc_reserve_metadata. | |
6189 | */ | |
43b18595 QW |
6190 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
6191 | bool qgroup_free) | |
8b62f87b | 6192 | { |
3ffbd68c | 6193 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
8b62f87b | 6194 | unsigned num_extents; |
8b62f87b JB |
6195 | |
6196 | spin_lock(&inode->lock); | |
6197 | num_extents = count_max_extents(num_bytes); | |
6198 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 6199 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
6200 | spin_unlock(&inode->lock); |
6201 | ||
8b62f87b JB |
6202 | if (btrfs_is_testing(fs_info)) |
6203 | return; | |
6204 | ||
43b18595 | 6205 | btrfs_inode_rsv_release(inode, qgroup_free); |
8b62f87b JB |
6206 | } |
6207 | ||
1ada3a62 | 6208 | /** |
7cf5b976 | 6209 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6210 | * delalloc |
6211 | * @inode: inode we're writing to | |
6212 | * @start: start range we are writing to | |
6213 | * @len: how long the range we are writing to | |
364ecf36 QW |
6214 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6215 | * current reservation. | |
1ada3a62 | 6216 | * |
1ada3a62 QW |
6217 | * This will do the following things |
6218 | * | |
6219 | * o reserve space in data space info for num bytes | |
6220 | * and reserve precious corresponding qgroup space | |
6221 | * (Done in check_data_free_space) | |
6222 | * | |
6223 | * o reserve space for metadata space, based on the number of outstanding | |
6224 | * extents and how much csums will be needed | |
6225 | * also reserve metadata space in a per root over-reserve method. | |
6226 | * o add to the inodes->delalloc_bytes | |
6227 | * o add it to the fs_info's delalloc inodes list. | |
6228 | * (Above 3 all done in delalloc_reserve_metadata) | |
6229 | * | |
6230 | * Return 0 for success | |
6231 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6232 | */ | |
364ecf36 QW |
6233 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6234 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6235 | { |
6236 | int ret; | |
6237 | ||
364ecf36 | 6238 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6239 | if (ret < 0) |
6240 | return ret; | |
9f3db423 | 6241 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6242 | if (ret < 0) |
bc42bda2 | 6243 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6244 | return ret; |
6245 | } | |
6246 | ||
7709cde3 | 6247 | /** |
7cf5b976 | 6248 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6249 | * @inode: inode we're releasing space for |
6250 | * @start: start position of the space already reserved | |
6251 | * @len: the len of the space already reserved | |
8b62f87b | 6252 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
6253 | * |
6254 | * This function will release the metadata space that was not used and will | |
6255 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6256 | * list if there are no delalloc bytes left. | |
6257 | * Also it will handle the qgroup reserved space. | |
6258 | */ | |
bc42bda2 | 6259 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b | 6260 | struct extent_changeset *reserved, |
43b18595 | 6261 | u64 start, u64 len, bool qgroup_free) |
1ada3a62 | 6262 | { |
43b18595 | 6263 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
bc42bda2 | 6264 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6265 | } |
6266 | ||
ce93ec54 | 6267 | static int update_block_group(struct btrfs_trans_handle *trans, |
6b279408 | 6268 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 6269 | { |
6b279408 | 6270 | struct btrfs_fs_info *info = trans->fs_info; |
0af3d00b | 6271 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6272 | u64 total = num_bytes; |
9078a3e1 | 6273 | u64 old_val; |
db94535d | 6274 | u64 byte_in_group; |
0af3d00b | 6275 | int factor; |
ba2c4d4e | 6276 | int ret = 0; |
3e1ad54f | 6277 | |
5d4f98a2 | 6278 | /* block accounting for super block */ |
eb73c1b7 | 6279 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6280 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6281 | if (alloc) |
6282 | old_val += num_bytes; | |
6283 | else | |
6284 | old_val -= num_bytes; | |
6c41761f | 6285 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6286 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6287 | |
d397712b | 6288 | while (total) { |
db94535d | 6289 | cache = btrfs_lookup_block_group(info, bytenr); |
ba2c4d4e JB |
6290 | if (!cache) { |
6291 | ret = -ENOENT; | |
6292 | break; | |
6293 | } | |
46df06b8 DS |
6294 | factor = btrfs_bg_type_to_factor(cache->flags); |
6295 | ||
9d66e233 JB |
6296 | /* |
6297 | * If this block group has free space cache written out, we | |
6298 | * need to make sure to load it if we are removing space. This | |
6299 | * is because we need the unpinning stage to actually add the | |
6300 | * space back to the block group, otherwise we will leak space. | |
6301 | */ | |
6302 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6303 | cache_block_group(cache, 1); |
0af3d00b | 6304 | |
db94535d CM |
6305 | byte_in_group = bytenr - cache->key.objectid; |
6306 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6307 | |
25179201 | 6308 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6309 | spin_lock(&cache->lock); |
0af3d00b | 6310 | |
6202df69 | 6311 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6312 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6313 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6314 | ||
9078a3e1 | 6315 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6316 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6317 | if (alloc) { |
db94535d | 6318 | old_val += num_bytes; |
11833d66 YZ |
6319 | btrfs_set_block_group_used(&cache->item, old_val); |
6320 | cache->reserved -= num_bytes; | |
11833d66 | 6321 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6322 | cache->space_info->bytes_used += num_bytes; |
6323 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6324 | spin_unlock(&cache->lock); |
25179201 | 6325 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6326 | } else { |
db94535d | 6327 | old_val -= num_bytes; |
ae0ab003 FM |
6328 | btrfs_set_block_group_used(&cache->item, old_val); |
6329 | cache->pinned += num_bytes; | |
e2907c1a | 6330 | update_bytes_pinned(cache->space_info, num_bytes); |
ae0ab003 FM |
6331 | cache->space_info->bytes_used -= num_bytes; |
6332 | cache->space_info->disk_used -= num_bytes * factor; | |
6333 | spin_unlock(&cache->lock); | |
6334 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6335 | |
0b246afa | 6336 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6337 | cache->space_info->flags, |
6338 | num_bytes, 1); | |
dec59fa3 EL |
6339 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
6340 | num_bytes, | |
6341 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
ae0ab003 FM |
6342 | set_extent_dirty(info->pinned_extents, |
6343 | bytenr, bytenr + num_bytes - 1, | |
6344 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6345 | } |
1bbc621e CM |
6346 | |
6347 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6348 | if (list_empty(&cache->dirty_list)) { | |
6349 | list_add_tail(&cache->dirty_list, | |
6350 | &trans->transaction->dirty_bgs); | |
ba2c4d4e | 6351 | trans->delayed_ref_updates++; |
1bbc621e CM |
6352 | btrfs_get_block_group(cache); |
6353 | } | |
6354 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6355 | ||
036a9348 FM |
6356 | /* |
6357 | * No longer have used bytes in this block group, queue it for | |
6358 | * deletion. We do this after adding the block group to the | |
6359 | * dirty list to avoid races between cleaner kthread and space | |
6360 | * cache writeout. | |
6361 | */ | |
031f24da QW |
6362 | if (!alloc && old_val == 0) |
6363 | btrfs_mark_bg_unused(cache); | |
036a9348 | 6364 | |
fa9c0d79 | 6365 | btrfs_put_block_group(cache); |
db94535d CM |
6366 | total -= num_bytes; |
6367 | bytenr += num_bytes; | |
9078a3e1 | 6368 | } |
ba2c4d4e JB |
6369 | |
6370 | /* Modified block groups are accounted for in the delayed_refs_rsv. */ | |
6371 | btrfs_update_delayed_refs_rsv(trans); | |
6372 | return ret; | |
9078a3e1 | 6373 | } |
6324fbf3 | 6374 | |
2ff7e61e | 6375 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6376 | { |
0f9dd46c | 6377 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6378 | u64 bytenr; |
0f9dd46c | 6379 | |
0b246afa JM |
6380 | spin_lock(&fs_info->block_group_cache_lock); |
6381 | bytenr = fs_info->first_logical_byte; | |
6382 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6383 | |
6384 | if (bytenr < (u64)-1) | |
6385 | return bytenr; | |
6386 | ||
0b246afa | 6387 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6388 | if (!cache) |
a061fc8d | 6389 | return 0; |
0f9dd46c | 6390 | |
d2fb3437 | 6391 | bytenr = cache->key.objectid; |
fa9c0d79 | 6392 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6393 | |
6394 | return bytenr; | |
a061fc8d CM |
6395 | } |
6396 | ||
fdf08605 | 6397 | static int pin_down_extent(struct btrfs_block_group_cache *cache, |
f0486c68 | 6398 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 6399 | { |
fdf08605 DS |
6400 | struct btrfs_fs_info *fs_info = cache->fs_info; |
6401 | ||
11833d66 YZ |
6402 | spin_lock(&cache->space_info->lock); |
6403 | spin_lock(&cache->lock); | |
6404 | cache->pinned += num_bytes; | |
e2907c1a | 6405 | update_bytes_pinned(cache->space_info, num_bytes); |
11833d66 YZ |
6406 | if (reserved) { |
6407 | cache->reserved -= num_bytes; | |
6408 | cache->space_info->bytes_reserved -= num_bytes; | |
6409 | } | |
6410 | spin_unlock(&cache->lock); | |
6411 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6412 | |
0b246afa | 6413 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6414 | cache->space_info->flags, num_bytes, 1); |
dec59fa3 EL |
6415 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
6416 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0b246afa | 6417 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6418 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6419 | return 0; | |
6420 | } | |
68b38550 | 6421 | |
f0486c68 YZ |
6422 | /* |
6423 | * this function must be called within transaction | |
6424 | */ | |
2ff7e61e | 6425 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6426 | u64 bytenr, u64 num_bytes, int reserved) |
6427 | { | |
6428 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6429 | |
0b246afa | 6430 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6431 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6432 | |
fdf08605 | 6433 | pin_down_extent(cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6434 | |
6435 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6436 | return 0; |
6437 | } | |
6438 | ||
f0486c68 | 6439 | /* |
e688b725 CM |
6440 | * this function must be called within transaction |
6441 | */ | |
2ff7e61e | 6442 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6443 | u64 bytenr, u64 num_bytes) |
6444 | { | |
6445 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6446 | int ret; |
e688b725 | 6447 | |
0b246afa | 6448 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6449 | if (!cache) |
6450 | return -EINVAL; | |
e688b725 CM |
6451 | |
6452 | /* | |
6453 | * pull in the free space cache (if any) so that our pin | |
6454 | * removes the free space from the cache. We have load_only set | |
6455 | * to one because the slow code to read in the free extents does check | |
6456 | * the pinned extents. | |
6457 | */ | |
f6373bf3 | 6458 | cache_block_group(cache, 1); |
e688b725 | 6459 | |
fdf08605 | 6460 | pin_down_extent(cache, bytenr, num_bytes, 0); |
e688b725 CM |
6461 | |
6462 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6463 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6464 | btrfs_put_block_group(cache); |
b50c6e25 | 6465 | return ret; |
e688b725 CM |
6466 | } |
6467 | ||
2ff7e61e JM |
6468 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6469 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6470 | { |
6471 | int ret; | |
6472 | struct btrfs_block_group_cache *block_group; | |
6473 | struct btrfs_caching_control *caching_ctl; | |
6474 | ||
0b246afa | 6475 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6476 | if (!block_group) |
6477 | return -EINVAL; | |
6478 | ||
6479 | cache_block_group(block_group, 0); | |
6480 | caching_ctl = get_caching_control(block_group); | |
6481 | ||
6482 | if (!caching_ctl) { | |
6483 | /* Logic error */ | |
6484 | BUG_ON(!block_group_cache_done(block_group)); | |
6485 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6486 | } else { | |
6487 | mutex_lock(&caching_ctl->mutex); | |
6488 | ||
6489 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6490 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6491 | } else if (start + num_bytes <= caching_ctl->progress) { |
6492 | ret = btrfs_remove_free_space(block_group, | |
6493 | start, num_bytes); | |
6494 | } else { | |
6495 | num_bytes = caching_ctl->progress - start; | |
6496 | ret = btrfs_remove_free_space(block_group, | |
6497 | start, num_bytes); | |
6498 | if (ret) | |
6499 | goto out_lock; | |
6500 | ||
6501 | num_bytes = (start + num_bytes) - | |
6502 | caching_ctl->progress; | |
6503 | start = caching_ctl->progress; | |
2ff7e61e | 6504 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6505 | } |
6506 | out_lock: | |
6507 | mutex_unlock(&caching_ctl->mutex); | |
6508 | put_caching_control(caching_ctl); | |
6509 | } | |
6510 | btrfs_put_block_group(block_group); | |
6511 | return ret; | |
6512 | } | |
6513 | ||
bcdc428c | 6514 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 6515 | { |
bcdc428c | 6516 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
6517 | struct btrfs_file_extent_item *item; |
6518 | struct btrfs_key key; | |
6519 | int found_type; | |
6520 | int i; | |
b89311ef | 6521 | int ret = 0; |
8c2a1a30 | 6522 | |
2ff7e61e | 6523 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6524 | return 0; |
6525 | ||
6526 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6527 | btrfs_item_key_to_cpu(eb, &key, i); | |
6528 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6529 | continue; | |
6530 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6531 | found_type = btrfs_file_extent_type(eb, item); | |
6532 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6533 | continue; | |
6534 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6535 | continue; | |
6536 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6537 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
6538 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
6539 | if (ret) | |
6540 | break; | |
8c2a1a30 JB |
6541 | } |
6542 | ||
b89311ef | 6543 | return ret; |
8c2a1a30 JB |
6544 | } |
6545 | ||
9cfa3e34 FM |
6546 | static void |
6547 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6548 | { | |
6549 | atomic_inc(&bg->reservations); | |
6550 | } | |
6551 | ||
6552 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6553 | const u64 start) | |
6554 | { | |
6555 | struct btrfs_block_group_cache *bg; | |
6556 | ||
6557 | bg = btrfs_lookup_block_group(fs_info, start); | |
6558 | ASSERT(bg); | |
6559 | if (atomic_dec_and_test(&bg->reservations)) | |
4625956a | 6560 | wake_up_var(&bg->reservations); |
9cfa3e34 FM |
6561 | btrfs_put_block_group(bg); |
6562 | } | |
6563 | ||
9cfa3e34 FM |
6564 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
6565 | { | |
6566 | struct btrfs_space_info *space_info = bg->space_info; | |
6567 | ||
6568 | ASSERT(bg->ro); | |
6569 | ||
6570 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6571 | return; | |
6572 | ||
6573 | /* | |
6574 | * Our block group is read only but before we set it to read only, | |
6575 | * some task might have had allocated an extent from it already, but it | |
6576 | * has not yet created a respective ordered extent (and added it to a | |
6577 | * root's list of ordered extents). | |
6578 | * Therefore wait for any task currently allocating extents, since the | |
6579 | * block group's reservations counter is incremented while a read lock | |
6580 | * on the groups' semaphore is held and decremented after releasing | |
6581 | * the read access on that semaphore and creating the ordered extent. | |
6582 | */ | |
6583 | down_write(&space_info->groups_sem); | |
6584 | up_write(&space_info->groups_sem); | |
6585 | ||
4625956a | 6586 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); |
9cfa3e34 FM |
6587 | } |
6588 | ||
fb25e914 | 6589 | /** |
4824f1f4 | 6590 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6591 | * @cache: The cache we are manipulating |
18513091 WX |
6592 | * @ram_bytes: The number of bytes of file content, and will be same to |
6593 | * @num_bytes except for the compress path. | |
fb25e914 | 6594 | * @num_bytes: The number of bytes in question |
e570fd27 | 6595 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6596 | * |
745699ef XW |
6597 | * This is called by the allocator when it reserves space. If this is a |
6598 | * reservation and the block group has become read only we cannot make the | |
6599 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6600 | */ |
4824f1f4 | 6601 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6602 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6603 | { |
fb25e914 | 6604 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6605 | int ret = 0; |
79787eaa | 6606 | |
fb25e914 JB |
6607 | spin_lock(&space_info->lock); |
6608 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6609 | if (cache->ro) { |
6610 | ret = -EAGAIN; | |
fb25e914 | 6611 | } else { |
4824f1f4 WX |
6612 | cache->reserved += num_bytes; |
6613 | space_info->bytes_reserved += num_bytes; | |
9f9b8e8d | 6614 | update_bytes_may_use(space_info, -ram_bytes); |
e570fd27 | 6615 | if (delalloc) |
4824f1f4 | 6616 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6617 | } |
fb25e914 JB |
6618 | spin_unlock(&cache->lock); |
6619 | spin_unlock(&space_info->lock); | |
f0486c68 | 6620 | return ret; |
324ae4df | 6621 | } |
9078a3e1 | 6622 | |
4824f1f4 WX |
6623 | /** |
6624 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6625 | * @cache: The cache we are manipulating | |
6626 | * @num_bytes: The number of bytes in question | |
6627 | * @delalloc: The blocks are allocated for the delalloc write | |
6628 | * | |
6629 | * This is called by somebody who is freeing space that was never actually used | |
6630 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6631 | * A and before transaction A commits you free that leaf, you call this with | |
6632 | * reserve set to 0 in order to clear the reservation. | |
6633 | */ | |
6634 | ||
556f3ca8 | 6635 | static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
6636 | u64 num_bytes, int delalloc) | |
4824f1f4 WX |
6637 | { |
6638 | struct btrfs_space_info *space_info = cache->space_info; | |
4824f1f4 WX |
6639 | |
6640 | spin_lock(&space_info->lock); | |
6641 | spin_lock(&cache->lock); | |
6642 | if (cache->ro) | |
6643 | space_info->bytes_readonly += num_bytes; | |
6644 | cache->reserved -= num_bytes; | |
6645 | space_info->bytes_reserved -= num_bytes; | |
21a94f7a | 6646 | space_info->max_extent_size = 0; |
4824f1f4 WX |
6647 | |
6648 | if (delalloc) | |
6649 | cache->delalloc_bytes -= num_bytes; | |
6650 | spin_unlock(&cache->lock); | |
6651 | spin_unlock(&space_info->lock); | |
4824f1f4 | 6652 | } |
8b74c03e | 6653 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6654 | { |
11833d66 YZ |
6655 | struct btrfs_caching_control *next; |
6656 | struct btrfs_caching_control *caching_ctl; | |
6657 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6658 | |
9e351cc8 | 6659 | down_write(&fs_info->commit_root_sem); |
25179201 | 6660 | |
11833d66 YZ |
6661 | list_for_each_entry_safe(caching_ctl, next, |
6662 | &fs_info->caching_block_groups, list) { | |
6663 | cache = caching_ctl->block_group; | |
6664 | if (block_group_cache_done(cache)) { | |
6665 | cache->last_byte_to_unpin = (u64)-1; | |
6666 | list_del_init(&caching_ctl->list); | |
6667 | put_caching_control(caching_ctl); | |
e8569813 | 6668 | } else { |
11833d66 | 6669 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6670 | } |
e8569813 | 6671 | } |
11833d66 YZ |
6672 | |
6673 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6674 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6675 | else | |
6676 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6677 | ||
9e351cc8 | 6678 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6679 | |
6680 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6681 | } |
6682 | ||
c759c4e1 JB |
6683 | /* |
6684 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6685 | * what it should be based on the mount options. | |
6686 | */ | |
6687 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6688 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6689 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6690 | { |
6691 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6692 | |
6693 | *empty_cluster = 0; | |
6694 | if (btrfs_mixed_space_info(space_info)) | |
6695 | return ret; | |
6696 | ||
c759c4e1 | 6697 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6698 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6699 | if (btrfs_test_opt(fs_info, SSD)) |
6700 | *empty_cluster = SZ_2M; | |
6701 | else | |
ee22184b | 6702 | *empty_cluster = SZ_64K; |
583b7231 HK |
6703 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6704 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6705 | *empty_cluster = SZ_2M; | |
0b246afa | 6706 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6707 | } |
6708 | ||
6709 | return ret; | |
6710 | } | |
6711 | ||
2ff7e61e JM |
6712 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6713 | u64 start, u64 end, | |
678886bd | 6714 | const bool return_free_space) |
ccd467d6 | 6715 | { |
11833d66 | 6716 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6717 | struct btrfs_space_info *space_info; |
6718 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6719 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6720 | u64 len; |
c759c4e1 JB |
6721 | u64 total_unpinned = 0; |
6722 | u64 empty_cluster = 0; | |
7b398f8e | 6723 | bool readonly; |
ccd467d6 | 6724 | |
11833d66 | 6725 | while (start <= end) { |
7b398f8e | 6726 | readonly = false; |
11833d66 YZ |
6727 | if (!cache || |
6728 | start >= cache->key.objectid + cache->key.offset) { | |
6729 | if (cache) | |
6730 | btrfs_put_block_group(cache); | |
c759c4e1 | 6731 | total_unpinned = 0; |
11833d66 | 6732 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6733 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6734 | |
2ff7e61e | 6735 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6736 | cache->space_info, |
6737 | &empty_cluster); | |
6738 | empty_cluster <<= 1; | |
11833d66 YZ |
6739 | } |
6740 | ||
6741 | len = cache->key.objectid + cache->key.offset - start; | |
6742 | len = min(len, end + 1 - start); | |
6743 | ||
6744 | if (start < cache->last_byte_to_unpin) { | |
6745 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6746 | if (return_free_space) |
6747 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6748 | } |
6749 | ||
f0486c68 | 6750 | start += len; |
c759c4e1 | 6751 | total_unpinned += len; |
7b398f8e | 6752 | space_info = cache->space_info; |
f0486c68 | 6753 | |
c759c4e1 JB |
6754 | /* |
6755 | * If this space cluster has been marked as fragmented and we've | |
6756 | * unpinned enough in this block group to potentially allow a | |
6757 | * cluster to be created inside of it go ahead and clear the | |
6758 | * fragmented check. | |
6759 | */ | |
6760 | if (cluster && cluster->fragmented && | |
6761 | total_unpinned > empty_cluster) { | |
6762 | spin_lock(&cluster->lock); | |
6763 | cluster->fragmented = 0; | |
6764 | spin_unlock(&cluster->lock); | |
6765 | } | |
6766 | ||
7b398f8e | 6767 | spin_lock(&space_info->lock); |
11833d66 YZ |
6768 | spin_lock(&cache->lock); |
6769 | cache->pinned -= len; | |
e2907c1a | 6770 | update_bytes_pinned(space_info, -len); |
c51e7bb1 JB |
6771 | |
6772 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6773 | space_info->flags, len, 0); | |
4f4db217 | 6774 | space_info->max_extent_size = 0; |
dec59fa3 EL |
6775 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
6776 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
6777 | if (cache->ro) { |
6778 | space_info->bytes_readonly += len; | |
6779 | readonly = true; | |
6780 | } | |
11833d66 | 6781 | spin_unlock(&cache->lock); |
957780eb JB |
6782 | if (!readonly && return_free_space && |
6783 | global_rsv->space_info == space_info) { | |
6784 | u64 to_add = len; | |
92ac58ec | 6785 | |
7b398f8e JB |
6786 | spin_lock(&global_rsv->lock); |
6787 | if (!global_rsv->full) { | |
957780eb JB |
6788 | to_add = min(len, global_rsv->size - |
6789 | global_rsv->reserved); | |
6790 | global_rsv->reserved += to_add; | |
9f9b8e8d | 6791 | update_bytes_may_use(space_info, to_add); |
7b398f8e JB |
6792 | if (global_rsv->reserved >= global_rsv->size) |
6793 | global_rsv->full = 1; | |
957780eb JB |
6794 | trace_btrfs_space_reservation(fs_info, |
6795 | "space_info", | |
6796 | space_info->flags, | |
6797 | to_add, 1); | |
6798 | len -= to_add; | |
7b398f8e JB |
6799 | } |
6800 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6801 | /* Add to any tickets we may have */ |
6802 | if (len) | |
6803 | space_info_add_new_bytes(fs_info, space_info, | |
6804 | len); | |
7b398f8e JB |
6805 | } |
6806 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6807 | } |
11833d66 YZ |
6808 | |
6809 | if (cache) | |
6810 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6811 | return 0; |
6812 | } | |
6813 | ||
5ead2dd0 | 6814 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 6815 | { |
5ead2dd0 | 6816 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
6817 | struct btrfs_block_group_cache *block_group, *tmp; |
6818 | struct list_head *deleted_bgs; | |
11833d66 | 6819 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6820 | u64 start; |
6821 | u64 end; | |
a28ec197 | 6822 | int ret; |
a28ec197 | 6823 | |
11833d66 YZ |
6824 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6825 | unpin = &fs_info->freed_extents[1]; | |
6826 | else | |
6827 | unpin = &fs_info->freed_extents[0]; | |
6828 | ||
e33e17ee | 6829 | while (!trans->aborted) { |
0e6ec385 FM |
6830 | struct extent_state *cached_state = NULL; |
6831 | ||
d4b450cd | 6832 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6833 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 6834 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
6835 | if (ret) { |
6836 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6837 | break; |
d4b450cd | 6838 | } |
1f3c79a2 | 6839 | |
0b246afa | 6840 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6841 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6842 | end + 1 - start, NULL); |
1f3c79a2 | 6843 | |
0e6ec385 | 6844 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 6845 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6846 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 6847 | free_extent_state(cached_state); |
b9473439 | 6848 | cond_resched(); |
a28ec197 | 6849 | } |
817d52f8 | 6850 | |
e33e17ee JM |
6851 | /* |
6852 | * Transaction is finished. We don't need the lock anymore. We | |
6853 | * do need to clean up the block groups in case of a transaction | |
6854 | * abort. | |
6855 | */ | |
6856 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6857 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6858 | u64 trimmed = 0; | |
6859 | ||
6860 | ret = -EROFS; | |
6861 | if (!trans->aborted) | |
2ff7e61e | 6862 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6863 | block_group->key.objectid, |
6864 | block_group->key.offset, | |
6865 | &trimmed); | |
6866 | ||
6867 | list_del_init(&block_group->bg_list); | |
6868 | btrfs_put_block_group_trimming(block_group); | |
6869 | btrfs_put_block_group(block_group); | |
6870 | ||
6871 | if (ret) { | |
6872 | const char *errstr = btrfs_decode_error(ret); | |
6873 | btrfs_warn(fs_info, | |
913e1535 | 6874 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6875 | ret, errstr); |
6876 | } | |
6877 | } | |
6878 | ||
e20d96d6 CM |
6879 | return 0; |
6880 | } | |
6881 | ||
5d4f98a2 | 6882 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
6883 | struct btrfs_delayed_ref_node *node, u64 parent, |
6884 | u64 root_objectid, u64 owner_objectid, | |
6885 | u64 owner_offset, int refs_to_drop, | |
6886 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 6887 | { |
e72cb923 | 6888 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 6889 | struct btrfs_key key; |
5d4f98a2 | 6890 | struct btrfs_path *path; |
1261ec42 | 6891 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6892 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6893 | struct btrfs_extent_item *ei; |
6894 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6895 | int ret; |
5d4f98a2 | 6896 | int is_data; |
952fccac CM |
6897 | int extent_slot = 0; |
6898 | int found_extent = 0; | |
6899 | int num_to_del = 1; | |
5d4f98a2 YZ |
6900 | u32 item_size; |
6901 | u64 refs; | |
c682f9b3 QW |
6902 | u64 bytenr = node->bytenr; |
6903 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6904 | int last_ref = 0; |
0b246afa | 6905 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6906 | |
5caf2a00 | 6907 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6908 | if (!path) |
6909 | return -ENOMEM; | |
5f26f772 | 6910 | |
e4058b54 | 6911 | path->reada = READA_FORWARD; |
b9473439 | 6912 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6913 | |
6914 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6915 | BUG_ON(!is_data && refs_to_drop != 1); | |
6916 | ||
3173a18f | 6917 | if (is_data) |
897ca819 | 6918 | skinny_metadata = false; |
3173a18f | 6919 | |
fbe4801b NB |
6920 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
6921 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 6922 | owner_offset); |
7bb86316 | 6923 | if (ret == 0) { |
952fccac | 6924 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6925 | while (extent_slot >= 0) { |
6926 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6927 | extent_slot); |
5d4f98a2 | 6928 | if (key.objectid != bytenr) |
952fccac | 6929 | break; |
5d4f98a2 YZ |
6930 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6931 | key.offset == num_bytes) { | |
952fccac CM |
6932 | found_extent = 1; |
6933 | break; | |
6934 | } | |
3173a18f JB |
6935 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6936 | key.offset == owner_objectid) { | |
6937 | found_extent = 1; | |
6938 | break; | |
6939 | } | |
952fccac CM |
6940 | if (path->slots[0] - extent_slot > 5) |
6941 | break; | |
5d4f98a2 | 6942 | extent_slot--; |
952fccac | 6943 | } |
a79865c6 | 6944 | |
31840ae1 | 6945 | if (!found_extent) { |
5d4f98a2 | 6946 | BUG_ON(iref); |
87cc7a8a | 6947 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 6948 | refs_to_drop, |
fcebe456 | 6949 | is_data, &last_ref); |
005d6427 | 6950 | if (ret) { |
66642832 | 6951 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6952 | goto out; |
6953 | } | |
b3b4aa74 | 6954 | btrfs_release_path(path); |
b9473439 | 6955 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6956 | |
6957 | key.objectid = bytenr; | |
6958 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6959 | key.offset = num_bytes; | |
6960 | ||
3173a18f JB |
6961 | if (!is_data && skinny_metadata) { |
6962 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6963 | key.offset = owner_objectid; | |
6964 | } | |
6965 | ||
31840ae1 ZY |
6966 | ret = btrfs_search_slot(trans, extent_root, |
6967 | &key, path, -1, 1); | |
3173a18f JB |
6968 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6969 | /* | |
6970 | * Couldn't find our skinny metadata item, | |
6971 | * see if we have ye olde extent item. | |
6972 | */ | |
6973 | path->slots[0]--; | |
6974 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6975 | path->slots[0]); | |
6976 | if (key.objectid == bytenr && | |
6977 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6978 | key.offset == num_bytes) | |
6979 | ret = 0; | |
6980 | } | |
6981 | ||
6982 | if (ret > 0 && skinny_metadata) { | |
6983 | skinny_metadata = false; | |
9ce49a0b | 6984 | key.objectid = bytenr; |
3173a18f JB |
6985 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6986 | key.offset = num_bytes; | |
6987 | btrfs_release_path(path); | |
6988 | ret = btrfs_search_slot(trans, extent_root, | |
6989 | &key, path, -1, 1); | |
6990 | } | |
6991 | ||
f3465ca4 | 6992 | if (ret) { |
5d163e0e JM |
6993 | btrfs_err(info, |
6994 | "umm, got %d back from search, was looking for %llu", | |
6995 | ret, bytenr); | |
b783e62d | 6996 | if (ret > 0) |
a4f78750 | 6997 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6998 | } |
005d6427 | 6999 | if (ret < 0) { |
66642832 | 7000 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7001 | goto out; |
7002 | } | |
31840ae1 ZY |
7003 | extent_slot = path->slots[0]; |
7004 | } | |
fae7f21c | 7005 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 7006 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
7007 | btrfs_err(info, |
7008 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
7009 | bytenr, parent, root_objectid, owner_objectid, |
7010 | owner_offset); | |
66642832 | 7011 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 7012 | goto out; |
79787eaa | 7013 | } else { |
66642832 | 7014 | btrfs_abort_transaction(trans, ret); |
005d6427 | 7015 | goto out; |
7bb86316 | 7016 | } |
5f39d397 CM |
7017 | |
7018 | leaf = path->nodes[0]; | |
5d4f98a2 | 7019 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 7020 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
7021 | ret = -EINVAL; |
7022 | btrfs_print_v0_err(info); | |
7023 | btrfs_abort_transaction(trans, ret); | |
7024 | goto out; | |
7025 | } | |
952fccac | 7026 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 7027 | struct btrfs_extent_item); |
3173a18f JB |
7028 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7029 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7030 | struct btrfs_tree_block_info *bi; |
7031 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7032 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7033 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7034 | } | |
56bec294 | 7035 | |
5d4f98a2 | 7036 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7037 | if (refs < refs_to_drop) { |
5d163e0e JM |
7038 | btrfs_err(info, |
7039 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7040 | refs_to_drop, refs, bytenr); | |
32b02538 | 7041 | ret = -EINVAL; |
66642832 | 7042 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7043 | goto out; |
7044 | } | |
56bec294 | 7045 | refs -= refs_to_drop; |
5f39d397 | 7046 | |
5d4f98a2 YZ |
7047 | if (refs > 0) { |
7048 | if (extent_op) | |
7049 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7050 | /* | |
7051 | * In the case of inline back ref, reference count will | |
7052 | * be updated by remove_extent_backref | |
952fccac | 7053 | */ |
5d4f98a2 YZ |
7054 | if (iref) { |
7055 | BUG_ON(!found_extent); | |
7056 | } else { | |
7057 | btrfs_set_extent_refs(leaf, ei, refs); | |
7058 | btrfs_mark_buffer_dirty(leaf); | |
7059 | } | |
7060 | if (found_extent) { | |
87cc7a8a NB |
7061 | ret = remove_extent_backref(trans, path, iref, |
7062 | refs_to_drop, is_data, | |
7063 | &last_ref); | |
005d6427 | 7064 | if (ret) { |
66642832 | 7065 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7066 | goto out; |
7067 | } | |
952fccac | 7068 | } |
5d4f98a2 | 7069 | } else { |
5d4f98a2 YZ |
7070 | if (found_extent) { |
7071 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7072 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7073 | if (iref) { |
7074 | BUG_ON(path->slots[0] != extent_slot); | |
7075 | } else { | |
7076 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7077 | path->slots[0] = extent_slot; | |
7078 | num_to_del = 2; | |
7079 | } | |
78fae27e | 7080 | } |
b9473439 | 7081 | |
fcebe456 | 7082 | last_ref = 1; |
952fccac CM |
7083 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7084 | num_to_del); | |
005d6427 | 7085 | if (ret) { |
66642832 | 7086 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7087 | goto out; |
7088 | } | |
b3b4aa74 | 7089 | btrfs_release_path(path); |
21af804c | 7090 | |
5d4f98a2 | 7091 | if (is_data) { |
5b4aacef | 7092 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7093 | if (ret) { |
66642832 | 7094 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7095 | goto out; |
7096 | } | |
459931ec CM |
7097 | } |
7098 | ||
e7355e50 | 7099 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 7100 | if (ret) { |
66642832 | 7101 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7102 | goto out; |
7103 | } | |
7104 | ||
6b279408 | 7105 | ret = update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 7106 | if (ret) { |
66642832 | 7107 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7108 | goto out; |
7109 | } | |
a28ec197 | 7110 | } |
fcebe456 JB |
7111 | btrfs_release_path(path); |
7112 | ||
79787eaa | 7113 | out: |
5caf2a00 | 7114 | btrfs_free_path(path); |
a28ec197 CM |
7115 | return ret; |
7116 | } | |
7117 | ||
1887be66 | 7118 | /* |
f0486c68 | 7119 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7120 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7121 | * a given extent, and if there are no other delayed refs to be processed, it | |
7122 | * removes it from the tree. | |
7123 | */ | |
7124 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7125 | u64 bytenr) |
1887be66 CM |
7126 | { |
7127 | struct btrfs_delayed_ref_head *head; | |
7128 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7129 | int ret = 0; |
1887be66 CM |
7130 | |
7131 | delayed_refs = &trans->transaction->delayed_refs; | |
7132 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7133 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7134 | if (!head) |
cf93da7b | 7135 | goto out_delayed_unlock; |
1887be66 | 7136 | |
d7df2c79 | 7137 | spin_lock(&head->lock); |
e3d03965 | 7138 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
7139 | goto out; |
7140 | ||
bedc6617 JB |
7141 | if (cleanup_extent_op(head) != NULL) |
7142 | goto out; | |
5d4f98a2 | 7143 | |
1887be66 CM |
7144 | /* |
7145 | * waiting for the lock here would deadlock. If someone else has it | |
7146 | * locked they are already in the process of dropping it anyway | |
7147 | */ | |
7148 | if (!mutex_trylock(&head->mutex)) | |
7149 | goto out; | |
7150 | ||
d7baffda | 7151 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 7152 | head->processing = 0; |
d7baffda | 7153 | |
d7df2c79 | 7154 | spin_unlock(&head->lock); |
1887be66 CM |
7155 | spin_unlock(&delayed_refs->lock); |
7156 | ||
f0486c68 YZ |
7157 | BUG_ON(head->extent_op); |
7158 | if (head->must_insert_reserved) | |
7159 | ret = 1; | |
7160 | ||
31890da0 | 7161 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 7162 | mutex_unlock(&head->mutex); |
d278850e | 7163 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 7164 | return ret; |
1887be66 | 7165 | out: |
d7df2c79 | 7166 | spin_unlock(&head->lock); |
cf93da7b CM |
7167 | |
7168 | out_delayed_unlock: | |
1887be66 CM |
7169 | spin_unlock(&delayed_refs->lock); |
7170 | return 0; | |
7171 | } | |
7172 | ||
f0486c68 YZ |
7173 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7174 | struct btrfs_root *root, | |
7175 | struct extent_buffer *buf, | |
5581a51a | 7176 | u64 parent, int last_ref) |
f0486c68 | 7177 | { |
0b246afa | 7178 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7179 | int pin = 1; |
f0486c68 YZ |
7180 | int ret; |
7181 | ||
7182 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
d7eae340 OS |
7183 | int old_ref_mod, new_ref_mod; |
7184 | ||
fd708b81 JB |
7185 | btrfs_ref_tree_mod(root, buf->start, buf->len, parent, |
7186 | root->root_key.objectid, | |
7187 | btrfs_header_level(buf), 0, | |
7188 | BTRFS_DROP_DELAYED_REF); | |
44e1c47d | 7189 | ret = btrfs_add_delayed_tree_ref(trans, buf->start, |
7be07912 | 7190 | buf->len, parent, |
0b246afa JM |
7191 | root->root_key.objectid, |
7192 | btrfs_header_level(buf), | |
7be07912 | 7193 | BTRFS_DROP_DELAYED_REF, NULL, |
d7eae340 | 7194 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7195 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7196 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7197 | } |
7198 | ||
0a16c7d7 | 7199 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7200 | struct btrfs_block_group_cache *cache; |
7201 | ||
f0486c68 | 7202 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7203 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7204 | if (!ret) |
37be25bc | 7205 | goto out; |
f0486c68 YZ |
7206 | } |
7207 | ||
4da8b76d | 7208 | pin = 0; |
0b246afa | 7209 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7210 | |
f0486c68 | 7211 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
fdf08605 | 7212 | pin_down_extent(cache, buf->start, buf->len, 1); |
6219872d | 7213 | btrfs_put_block_group(cache); |
37be25bc | 7214 | goto out; |
f0486c68 YZ |
7215 | } |
7216 | ||
7217 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7218 | ||
7219 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7220 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7221 | btrfs_put_block_group(cache); |
71ff6437 | 7222 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7223 | } |
7224 | out: | |
b150a4f1 | 7225 | if (pin) |
29d2b84c | 7226 | add_pinned_bytes(fs_info, buf->len, true, |
b150a4f1 JB |
7227 | root->root_key.objectid); |
7228 | ||
0a16c7d7 OS |
7229 | if (last_ref) { |
7230 | /* | |
7231 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7232 | * matter anymore. | |
7233 | */ | |
7234 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7235 | } | |
f0486c68 YZ |
7236 | } |
7237 | ||
79787eaa | 7238 | /* Can return -ENOMEM */ |
2ff7e61e | 7239 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
84f7d8e6 | 7240 | struct btrfs_root *root, |
66d7e7f0 | 7241 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7242 | u64 owner, u64 offset) |
925baedd | 7243 | { |
84f7d8e6 | 7244 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 7245 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7246 | int ret; |
7247 | ||
f5ee5c9a | 7248 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7249 | return 0; |
fccb84c9 | 7250 | |
fd708b81 JB |
7251 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) |
7252 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, | |
7253 | root_objectid, owner, offset, | |
7254 | BTRFS_DROP_DELAYED_REF); | |
7255 | ||
56bec294 CM |
7256 | /* |
7257 | * tree log blocks never actually go into the extent allocation | |
7258 | * tree, just update pinning info and exit early. | |
56bec294 | 7259 | */ |
5d4f98a2 YZ |
7260 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7261 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7262 | /* unlocks the pinned mutex */ |
2ff7e61e | 7263 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
d7eae340 | 7264 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7265 | ret = 0; |
5d4f98a2 | 7266 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
44e1c47d | 7267 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, |
7be07912 OS |
7268 | num_bytes, parent, |
7269 | root_objectid, (int)owner, | |
7270 | BTRFS_DROP_DELAYED_REF, NULL, | |
d7eae340 | 7271 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7272 | } else { |
88a979c6 | 7273 | ret = btrfs_add_delayed_data_ref(trans, bytenr, |
7be07912 OS |
7274 | num_bytes, parent, |
7275 | root_objectid, owner, offset, | |
7276 | 0, BTRFS_DROP_DELAYED_REF, | |
d7eae340 | 7277 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7278 | } |
d7eae340 | 7279 | |
29d2b84c NB |
7280 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) { |
7281 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
7282 | ||
7283 | add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid); | |
7284 | } | |
d7eae340 | 7285 | |
925baedd CM |
7286 | return ret; |
7287 | } | |
7288 | ||
817d52f8 JB |
7289 | /* |
7290 | * when we wait for progress in the block group caching, its because | |
7291 | * our allocation attempt failed at least once. So, we must sleep | |
7292 | * and let some progress happen before we try again. | |
7293 | * | |
7294 | * This function will sleep at least once waiting for new free space to | |
7295 | * show up, and then it will check the block group free space numbers | |
7296 | * for our min num_bytes. Another option is to have it go ahead | |
7297 | * and look in the rbtree for a free extent of a given size, but this | |
7298 | * is a good start. | |
36cce922 JB |
7299 | * |
7300 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7301 | * any of the information in this block group. | |
817d52f8 | 7302 | */ |
36cce922 | 7303 | static noinline void |
817d52f8 JB |
7304 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7305 | u64 num_bytes) | |
7306 | { | |
11833d66 | 7307 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7308 | |
11833d66 YZ |
7309 | caching_ctl = get_caching_control(cache); |
7310 | if (!caching_ctl) | |
36cce922 | 7311 | return; |
817d52f8 | 7312 | |
11833d66 | 7313 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7314 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7315 | |
7316 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7317 | } |
7318 | ||
7319 | static noinline int | |
7320 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7321 | { | |
7322 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7323 | int ret = 0; |
11833d66 YZ |
7324 | |
7325 | caching_ctl = get_caching_control(cache); | |
7326 | if (!caching_ctl) | |
36cce922 | 7327 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7328 | |
7329 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7330 | if (cache->cached == BTRFS_CACHE_ERROR) |
7331 | ret = -EIO; | |
11833d66 | 7332 | put_caching_control(caching_ctl); |
36cce922 | 7333 | return ret; |
817d52f8 JB |
7334 | } |
7335 | ||
7336 | enum btrfs_loop_type { | |
285ff5af JB |
7337 | LOOP_CACHING_NOWAIT = 0, |
7338 | LOOP_CACHING_WAIT = 1, | |
7339 | LOOP_ALLOC_CHUNK = 2, | |
7340 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7341 | }; |
7342 | ||
e570fd27 MX |
7343 | static inline void |
7344 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7345 | int delalloc) | |
7346 | { | |
7347 | if (delalloc) | |
7348 | down_read(&cache->data_rwsem); | |
7349 | } | |
7350 | ||
7351 | static inline void | |
7352 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7353 | int delalloc) | |
7354 | { | |
7355 | btrfs_get_block_group(cache); | |
7356 | if (delalloc) | |
7357 | down_read(&cache->data_rwsem); | |
7358 | } | |
7359 | ||
7360 | static struct btrfs_block_group_cache * | |
7361 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7362 | struct btrfs_free_cluster *cluster, | |
7363 | int delalloc) | |
7364 | { | |
89771cc9 | 7365 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7366 | |
e570fd27 | 7367 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7368 | while (1) { |
7369 | used_bg = cluster->block_group; | |
7370 | if (!used_bg) | |
7371 | return NULL; | |
7372 | ||
7373 | if (used_bg == block_group) | |
e570fd27 MX |
7374 | return used_bg; |
7375 | ||
6719afdc | 7376 | btrfs_get_block_group(used_bg); |
e570fd27 | 7377 | |
6719afdc GU |
7378 | if (!delalloc) |
7379 | return used_bg; | |
e570fd27 | 7380 | |
6719afdc GU |
7381 | if (down_read_trylock(&used_bg->data_rwsem)) |
7382 | return used_bg; | |
e570fd27 | 7383 | |
6719afdc | 7384 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7385 | |
e321f8a8 LB |
7386 | /* We should only have one-level nested. */ |
7387 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7388 | |
6719afdc GU |
7389 | spin_lock(&cluster->refill_lock); |
7390 | if (used_bg == cluster->block_group) | |
7391 | return used_bg; | |
e570fd27 | 7392 | |
6719afdc GU |
7393 | up_read(&used_bg->data_rwsem); |
7394 | btrfs_put_block_group(used_bg); | |
7395 | } | |
e570fd27 MX |
7396 | } |
7397 | ||
7398 | static inline void | |
7399 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7400 | int delalloc) | |
7401 | { | |
7402 | if (delalloc) | |
7403 | up_read(&cache->data_rwsem); | |
7404 | btrfs_put_block_group(cache); | |
7405 | } | |
7406 | ||
b4bd745d QW |
7407 | /* |
7408 | * Structure used internally for find_free_extent() function. Wraps needed | |
7409 | * parameters. | |
7410 | */ | |
7411 | struct find_free_extent_ctl { | |
7412 | /* Basic allocation info */ | |
7413 | u64 ram_bytes; | |
7414 | u64 num_bytes; | |
7415 | u64 empty_size; | |
7416 | u64 flags; | |
7417 | int delalloc; | |
7418 | ||
7419 | /* Where to start the search inside the bg */ | |
7420 | u64 search_start; | |
7421 | ||
7422 | /* For clustered allocation */ | |
7423 | u64 empty_cluster; | |
7424 | ||
7425 | bool have_caching_bg; | |
7426 | bool orig_have_caching_bg; | |
7427 | ||
7428 | /* RAID index, converted from flags */ | |
7429 | int index; | |
7430 | ||
e72d79d6 QW |
7431 | /* |
7432 | * Current loop number, check find_free_extent_update_loop() for details | |
7433 | */ | |
b4bd745d QW |
7434 | int loop; |
7435 | ||
7436 | /* | |
7437 | * Whether we're refilling a cluster, if true we need to re-search | |
7438 | * current block group but don't try to refill the cluster again. | |
7439 | */ | |
7440 | bool retry_clustered; | |
7441 | ||
7442 | /* | |
7443 | * Whether we're updating free space cache, if true we need to re-search | |
7444 | * current block group but don't try updating free space cache again. | |
7445 | */ | |
7446 | bool retry_unclustered; | |
7447 | ||
7448 | /* If current block group is cached */ | |
7449 | int cached; | |
7450 | ||
7451 | /* Max contiguous hole found */ | |
7452 | u64 max_extent_size; | |
7453 | ||
7454 | /* Total free space from free space cache, not always contiguous */ | |
7455 | u64 total_free_space; | |
7456 | ||
7457 | /* Found result */ | |
7458 | u64 found_offset; | |
7459 | }; | |
7460 | ||
d06e3bb6 QW |
7461 | |
7462 | /* | |
7463 | * Helper function for find_free_extent(). | |
7464 | * | |
7465 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
7466 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
7467 | * Return >0 to inform caller that we find nothing | |
7468 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
7469 | */ | |
7470 | static int find_free_extent_clustered(struct btrfs_block_group_cache *bg, | |
7471 | struct btrfs_free_cluster *last_ptr, | |
7472 | struct find_free_extent_ctl *ffe_ctl, | |
7473 | struct btrfs_block_group_cache **cluster_bg_ret) | |
7474 | { | |
d06e3bb6 QW |
7475 | struct btrfs_block_group_cache *cluster_bg; |
7476 | u64 aligned_cluster; | |
7477 | u64 offset; | |
7478 | int ret; | |
7479 | ||
7480 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
7481 | if (!cluster_bg) | |
7482 | goto refill_cluster; | |
7483 | if (cluster_bg != bg && (cluster_bg->ro || | |
7484 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
7485 | goto release_cluster; | |
7486 | ||
7487 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
7488 | ffe_ctl->num_bytes, cluster_bg->key.objectid, | |
7489 | &ffe_ctl->max_extent_size); | |
7490 | if (offset) { | |
7491 | /* We have a block, we're done */ | |
7492 | spin_unlock(&last_ptr->refill_lock); | |
7493 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
7494 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
7495 | *cluster_bg_ret = cluster_bg; | |
7496 | ffe_ctl->found_offset = offset; | |
7497 | return 0; | |
7498 | } | |
7499 | WARN_ON(last_ptr->block_group != cluster_bg); | |
7500 | ||
7501 | release_cluster: | |
7502 | /* | |
7503 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
7504 | * lets just skip it and let the allocator find whatever block it can | |
7505 | * find. If we reach this point, we will have tried the cluster | |
7506 | * allocator plenty of times and not have found anything, so we are | |
7507 | * likely way too fragmented for the clustering stuff to find anything. | |
7508 | * | |
7509 | * However, if the cluster is taken from the current block group, | |
7510 | * release the cluster first, so that we stand a better chance of | |
7511 | * succeeding in the unclustered allocation. | |
7512 | */ | |
7513 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
7514 | spin_unlock(&last_ptr->refill_lock); | |
7515 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
7516 | return -ENOENT; | |
7517 | } | |
7518 | ||
7519 | /* This cluster didn't work out, free it and start over */ | |
7520 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7521 | ||
7522 | if (cluster_bg != bg) | |
7523 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
7524 | ||
7525 | refill_cluster: | |
7526 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
7527 | spin_unlock(&last_ptr->refill_lock); | |
7528 | return -ENOENT; | |
7529 | } | |
7530 | ||
7531 | aligned_cluster = max_t(u64, | |
7532 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
7533 | bg->full_stripe_len); | |
2ceeae2e DS |
7534 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
7535 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
7536 | if (ret == 0) { |
7537 | /* Now pull our allocation out of this cluster */ | |
7538 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
7539 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
7540 | &ffe_ctl->max_extent_size); | |
7541 | if (offset) { | |
7542 | /* We found one, proceed */ | |
7543 | spin_unlock(&last_ptr->refill_lock); | |
7544 | trace_btrfs_reserve_extent_cluster(bg, | |
7545 | ffe_ctl->search_start, | |
7546 | ffe_ctl->num_bytes); | |
7547 | ffe_ctl->found_offset = offset; | |
7548 | return 0; | |
7549 | } | |
7550 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
7551 | !ffe_ctl->retry_clustered) { | |
7552 | spin_unlock(&last_ptr->refill_lock); | |
7553 | ||
7554 | ffe_ctl->retry_clustered = true; | |
7555 | wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + | |
7556 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); | |
7557 | return -EAGAIN; | |
7558 | } | |
7559 | /* | |
7560 | * At this point we either didn't find a cluster or we weren't able to | |
7561 | * allocate a block from our cluster. Free the cluster we've been | |
7562 | * trying to use, and go to the next block group. | |
7563 | */ | |
7564 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7565 | spin_unlock(&last_ptr->refill_lock); | |
7566 | return 1; | |
7567 | } | |
7568 | ||
e1a41848 QW |
7569 | /* |
7570 | * Return >0 to inform caller that we find nothing | |
7571 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
7572 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
7573 | */ | |
7574 | static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg, | |
7575 | struct btrfs_free_cluster *last_ptr, | |
7576 | struct find_free_extent_ctl *ffe_ctl) | |
7577 | { | |
7578 | u64 offset; | |
7579 | ||
7580 | /* | |
7581 | * We are doing an unclustered allocation, set the fragmented flag so | |
7582 | * we don't bother trying to setup a cluster again until we get more | |
7583 | * space. | |
7584 | */ | |
7585 | if (unlikely(last_ptr)) { | |
7586 | spin_lock(&last_ptr->lock); | |
7587 | last_ptr->fragmented = 1; | |
7588 | spin_unlock(&last_ptr->lock); | |
7589 | } | |
7590 | if (ffe_ctl->cached) { | |
7591 | struct btrfs_free_space_ctl *free_space_ctl; | |
7592 | ||
7593 | free_space_ctl = bg->free_space_ctl; | |
7594 | spin_lock(&free_space_ctl->tree_lock); | |
7595 | if (free_space_ctl->free_space < | |
7596 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
7597 | ffe_ctl->empty_size) { | |
7598 | ffe_ctl->total_free_space = max_t(u64, | |
7599 | ffe_ctl->total_free_space, | |
7600 | free_space_ctl->free_space); | |
7601 | spin_unlock(&free_space_ctl->tree_lock); | |
7602 | return 1; | |
7603 | } | |
7604 | spin_unlock(&free_space_ctl->tree_lock); | |
7605 | } | |
7606 | ||
7607 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
7608 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
7609 | &ffe_ctl->max_extent_size); | |
7610 | ||
7611 | /* | |
7612 | * If we didn't find a chunk, and we haven't failed on this block group | |
7613 | * before, and this block group is in the middle of caching and we are | |
7614 | * ok with waiting, then go ahead and wait for progress to be made, and | |
7615 | * set @retry_unclustered to true. | |
7616 | * | |
7617 | * If @retry_unclustered is true then we've already waited on this | |
7618 | * block group once and should move on to the next block group. | |
7619 | */ | |
7620 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
7621 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
7622 | wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + | |
7623 | ffe_ctl->empty_size); | |
7624 | ffe_ctl->retry_unclustered = true; | |
7625 | return -EAGAIN; | |
7626 | } else if (!offset) { | |
7627 | return 1; | |
7628 | } | |
7629 | ffe_ctl->found_offset = offset; | |
7630 | return 0; | |
7631 | } | |
7632 | ||
e72d79d6 QW |
7633 | /* |
7634 | * Return >0 means caller needs to re-search for free extent | |
7635 | * Return 0 means we have the needed free extent. | |
7636 | * Return <0 means we failed to locate any free extent. | |
7637 | */ | |
7638 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
7639 | struct btrfs_free_cluster *last_ptr, | |
7640 | struct btrfs_key *ins, | |
7641 | struct find_free_extent_ctl *ffe_ctl, | |
7642 | int full_search, bool use_cluster) | |
7643 | { | |
7644 | struct btrfs_root *root = fs_info->extent_root; | |
7645 | int ret; | |
7646 | ||
7647 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
7648 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
7649 | ffe_ctl->orig_have_caching_bg = true; | |
7650 | ||
7651 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
7652 | ffe_ctl->have_caching_bg) | |
7653 | return 1; | |
7654 | ||
7655 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
7656 | return 1; | |
7657 | ||
7658 | if (ins->objectid) { | |
7659 | if (!use_cluster && last_ptr) { | |
7660 | spin_lock(&last_ptr->lock); | |
7661 | last_ptr->window_start = ins->objectid; | |
7662 | spin_unlock(&last_ptr->lock); | |
7663 | } | |
7664 | return 0; | |
7665 | } | |
7666 | ||
7667 | /* | |
7668 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
7669 | * caching kthreads as we move along | |
7670 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
7671 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7672 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7673 | * again | |
7674 | */ | |
7675 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
7676 | ffe_ctl->index = 0; | |
7677 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
7678 | /* | |
7679 | * We want to skip the LOOP_CACHING_WAIT step if we | |
7680 | * don't have any uncached bgs and we've already done a | |
7681 | * full search through. | |
7682 | */ | |
7683 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
7684 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
7685 | else | |
7686 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
7687 | } else { | |
7688 | ffe_ctl->loop++; | |
7689 | } | |
7690 | ||
7691 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
7692 | struct btrfs_trans_handle *trans; | |
7693 | int exist = 0; | |
7694 | ||
7695 | trans = current->journal_info; | |
7696 | if (trans) | |
7697 | exist = 1; | |
7698 | else | |
7699 | trans = btrfs_join_transaction(root); | |
7700 | ||
7701 | if (IS_ERR(trans)) { | |
7702 | ret = PTR_ERR(trans); | |
7703 | return ret; | |
7704 | } | |
7705 | ||
7706 | ret = do_chunk_alloc(trans, ffe_ctl->flags, | |
7707 | CHUNK_ALLOC_FORCE); | |
7708 | ||
7709 | /* | |
7710 | * If we can't allocate a new chunk we've already looped | |
7711 | * through at least once, move on to the NO_EMPTY_SIZE | |
7712 | * case. | |
7713 | */ | |
7714 | if (ret == -ENOSPC) | |
7715 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
7716 | ||
7717 | /* Do not bail out on ENOSPC since we can do more. */ | |
7718 | if (ret < 0 && ret != -ENOSPC) | |
7719 | btrfs_abort_transaction(trans, ret); | |
7720 | else | |
7721 | ret = 0; | |
7722 | if (!exist) | |
7723 | btrfs_end_transaction(trans); | |
7724 | if (ret) | |
7725 | return ret; | |
7726 | } | |
7727 | ||
7728 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
7729 | /* | |
7730 | * Don't loop again if we already have no empty_size and | |
7731 | * no empty_cluster. | |
7732 | */ | |
7733 | if (ffe_ctl->empty_size == 0 && | |
7734 | ffe_ctl->empty_cluster == 0) | |
7735 | return -ENOSPC; | |
7736 | ffe_ctl->empty_size = 0; | |
7737 | ffe_ctl->empty_cluster = 0; | |
7738 | } | |
7739 | return 1; | |
7740 | } | |
7741 | return -ENOSPC; | |
7742 | } | |
7743 | ||
fec577fb CM |
7744 | /* |
7745 | * walks the btree of allocated extents and find a hole of a given size. | |
7746 | * The key ins is changed to record the hole: | |
a4820398 | 7747 | * ins->objectid == start position |
62e2749e | 7748 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7749 | * ins->offset == the size of the hole. |
fec577fb | 7750 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7751 | * |
7752 | * If there is no suitable free space, we will record the max size of | |
7753 | * the free space extent currently. | |
e72d79d6 QW |
7754 | * |
7755 | * The overall logic and call chain: | |
7756 | * | |
7757 | * find_free_extent() | |
7758 | * |- Iterate through all block groups | |
7759 | * | |- Get a valid block group | |
7760 | * | |- Try to do clustered allocation in that block group | |
7761 | * | |- Try to do unclustered allocation in that block group | |
7762 | * | |- Check if the result is valid | |
7763 | * | | |- If valid, then exit | |
7764 | * | |- Jump to next block group | |
7765 | * | | |
7766 | * |- Push harder to find free extents | |
7767 | * |- If not found, re-iterate all block groups | |
fec577fb | 7768 | */ |
87bde3cd | 7769 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7770 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7771 | u64 hint_byte, struct btrfs_key *ins, | |
7772 | u64 flags, int delalloc) | |
fec577fb | 7773 | { |
80eb234a | 7774 | int ret = 0; |
fa9c0d79 | 7775 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7776 | struct btrfs_block_group_cache *block_group = NULL; |
b4bd745d | 7777 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 7778 | struct btrfs_space_info *space_info; |
67377734 | 7779 | bool use_cluster = true; |
a5e681d9 | 7780 | bool full_search = false; |
fec577fb | 7781 | |
0b246afa | 7782 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d QW |
7783 | |
7784 | ffe_ctl.ram_bytes = ram_bytes; | |
7785 | ffe_ctl.num_bytes = num_bytes; | |
7786 | ffe_ctl.empty_size = empty_size; | |
7787 | ffe_ctl.flags = flags; | |
7788 | ffe_ctl.search_start = 0; | |
7789 | ffe_ctl.retry_clustered = false; | |
7790 | ffe_ctl.retry_unclustered = false; | |
7791 | ffe_ctl.delalloc = delalloc; | |
7792 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
7793 | ffe_ctl.have_caching_bg = false; | |
7794 | ffe_ctl.orig_have_caching_bg = false; | |
7795 | ffe_ctl.found_offset = 0; | |
7796 | ||
962a298f | 7797 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7798 | ins->objectid = 0; |
7799 | ins->offset = 0; | |
b1a4d965 | 7800 | |
71ff6437 | 7801 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7802 | |
0b246afa | 7803 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7804 | if (!space_info) { |
0b246afa | 7805 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7806 | return -ENOSPC; |
7807 | } | |
2552d17e | 7808 | |
67377734 | 7809 | /* |
4f4db217 JB |
7810 | * If our free space is heavily fragmented we may not be able to make |
7811 | * big contiguous allocations, so instead of doing the expensive search | |
7812 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7813 | * can go ahead and search for a more manageable chunk. | |
7814 | * | |
7815 | * If our max_extent_size is large enough for our allocation simply | |
7816 | * disable clustering since we will likely not be able to find enough | |
7817 | * space to create a cluster and induce latency trying. | |
67377734 | 7818 | */ |
4f4db217 JB |
7819 | if (unlikely(space_info->max_extent_size)) { |
7820 | spin_lock(&space_info->lock); | |
7821 | if (space_info->max_extent_size && | |
7822 | num_bytes > space_info->max_extent_size) { | |
7823 | ins->offset = space_info->max_extent_size; | |
7824 | spin_unlock(&space_info->lock); | |
7825 | return -ENOSPC; | |
7826 | } else if (space_info->max_extent_size) { | |
7827 | use_cluster = false; | |
7828 | } | |
7829 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7830 | } |
0f9dd46c | 7831 | |
b4bd745d QW |
7832 | last_ptr = fetch_cluster_info(fs_info, space_info, |
7833 | &ffe_ctl.empty_cluster); | |
239b14b3 | 7834 | if (last_ptr) { |
fa9c0d79 CM |
7835 | spin_lock(&last_ptr->lock); |
7836 | if (last_ptr->block_group) | |
7837 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7838 | if (last_ptr->fragmented) { |
7839 | /* | |
7840 | * We still set window_start so we can keep track of the | |
7841 | * last place we found an allocation to try and save | |
7842 | * some time. | |
7843 | */ | |
7844 | hint_byte = last_ptr->window_start; | |
7845 | use_cluster = false; | |
7846 | } | |
fa9c0d79 | 7847 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7848 | } |
fa9c0d79 | 7849 | |
b4bd745d QW |
7850 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
7851 | first_logical_byte(fs_info, 0)); | |
7852 | ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte); | |
7853 | if (ffe_ctl.search_start == hint_byte) { | |
7854 | block_group = btrfs_lookup_block_group(fs_info, | |
7855 | ffe_ctl.search_start); | |
817d52f8 JB |
7856 | /* |
7857 | * we don't want to use the block group if it doesn't match our | |
7858 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7859 | * |
7860 | * However if we are re-searching with an ideal block group | |
7861 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7862 | */ |
b6919a58 | 7863 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7864 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7865 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7866 | if (list_empty(&block_group->list) || |
7867 | block_group->ro) { | |
7868 | /* | |
7869 | * someone is removing this block group, | |
7870 | * we can't jump into the have_block_group | |
7871 | * target because our list pointers are not | |
7872 | * valid | |
7873 | */ | |
7874 | btrfs_put_block_group(block_group); | |
7875 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7876 | } else { |
b4bd745d | 7877 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 7878 | block_group->flags); |
e570fd27 | 7879 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7880 | goto have_block_group; |
ccf0e725 | 7881 | } |
2552d17e | 7882 | } else if (block_group) { |
fa9c0d79 | 7883 | btrfs_put_block_group(block_group); |
2552d17e | 7884 | } |
42e70e7a | 7885 | } |
2552d17e | 7886 | search: |
b4bd745d QW |
7887 | ffe_ctl.have_caching_bg = false; |
7888 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
7889 | ffe_ctl.index == 0) | |
a5e681d9 | 7890 | full_search = true; |
80eb234a | 7891 | down_read(&space_info->groups_sem); |
b4bd745d QW |
7892 | list_for_each_entry(block_group, |
7893 | &space_info->block_groups[ffe_ctl.index], list) { | |
14443937 JM |
7894 | /* If the block group is read-only, we can skip it entirely. */ |
7895 | if (unlikely(block_group->ro)) | |
7896 | continue; | |
7897 | ||
e570fd27 | 7898 | btrfs_grab_block_group(block_group, delalloc); |
b4bd745d | 7899 | ffe_ctl.search_start = block_group->key.objectid; |
42e70e7a | 7900 | |
83a50de9 CM |
7901 | /* |
7902 | * this can happen if we end up cycling through all the | |
7903 | * raid types, but we want to make sure we only allocate | |
7904 | * for the proper type. | |
7905 | */ | |
b6919a58 | 7906 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 7907 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
83a50de9 | 7908 | BTRFS_BLOCK_GROUP_RAID1 | |
53b381b3 DW |
7909 | BTRFS_BLOCK_GROUP_RAID5 | |
7910 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7911 | BTRFS_BLOCK_GROUP_RAID10; |
7912 | ||
7913 | /* | |
7914 | * if they asked for extra copies and this block group | |
7915 | * doesn't provide them, bail. This does allow us to | |
7916 | * fill raid0 from raid1. | |
7917 | */ | |
b6919a58 | 7918 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7919 | goto loop; |
7920 | } | |
7921 | ||
2552d17e | 7922 | have_block_group: |
b4bd745d QW |
7923 | ffe_ctl.cached = block_group_cache_done(block_group); |
7924 | if (unlikely(!ffe_ctl.cached)) { | |
7925 | ffe_ctl.have_caching_bg = true; | |
f6373bf3 | 7926 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7927 | BUG_ON(ret < 0); |
7928 | ret = 0; | |
817d52f8 JB |
7929 | } |
7930 | ||
36cce922 JB |
7931 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7932 | goto loop; | |
0f9dd46c | 7933 | |
0a24325e | 7934 | /* |
062c05c4 AO |
7935 | * Ok we want to try and use the cluster allocator, so |
7936 | * lets look there | |
0a24325e | 7937 | */ |
c759c4e1 | 7938 | if (last_ptr && use_cluster) { |
d06e3bb6 | 7939 | struct btrfs_block_group_cache *cluster_bg = NULL; |
fa9c0d79 | 7940 | |
d06e3bb6 QW |
7941 | ret = find_free_extent_clustered(block_group, last_ptr, |
7942 | &ffe_ctl, &cluster_bg); | |
062c05c4 | 7943 | |
fa9c0d79 | 7944 | if (ret == 0) { |
d06e3bb6 QW |
7945 | if (cluster_bg && cluster_bg != block_group) { |
7946 | btrfs_release_block_group(block_group, | |
7947 | delalloc); | |
7948 | block_group = cluster_bg; | |
fa9c0d79 | 7949 | } |
d06e3bb6 QW |
7950 | goto checks; |
7951 | } else if (ret == -EAGAIN) { | |
817d52f8 | 7952 | goto have_block_group; |
d06e3bb6 QW |
7953 | } else if (ret > 0) { |
7954 | goto loop; | |
fa9c0d79 | 7955 | } |
d06e3bb6 | 7956 | /* ret == -ENOENT case falls through */ |
fa9c0d79 CM |
7957 | } |
7958 | ||
e1a41848 QW |
7959 | ret = find_free_extent_unclustered(block_group, last_ptr, |
7960 | &ffe_ctl); | |
7961 | if (ret == -EAGAIN) | |
817d52f8 | 7962 | goto have_block_group; |
e1a41848 | 7963 | else if (ret > 0) |
1cdda9b8 | 7964 | goto loop; |
e1a41848 | 7965 | /* ret == 0 case falls through */ |
fa9c0d79 | 7966 | checks: |
b4bd745d QW |
7967 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
7968 | fs_info->stripesize); | |
25179201 | 7969 | |
2552d17e | 7970 | /* move on to the next group */ |
b4bd745d | 7971 | if (ffe_ctl.search_start + num_bytes > |
215a63d1 | 7972 | block_group->key.objectid + block_group->key.offset) { |
b4bd745d QW |
7973 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
7974 | num_bytes); | |
2552d17e | 7975 | goto loop; |
6226cb0a | 7976 | } |
f5a31e16 | 7977 | |
b4bd745d QW |
7978 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
7979 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, | |
7980 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 7981 | |
18513091 WX |
7982 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7983 | num_bytes, delalloc); | |
f0486c68 | 7984 | if (ret == -EAGAIN) { |
b4bd745d QW |
7985 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
7986 | num_bytes); | |
2552d17e | 7987 | goto loop; |
0f9dd46c | 7988 | } |
9cfa3e34 | 7989 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7990 | |
f0486c68 | 7991 | /* we are all good, lets return */ |
b4bd745d | 7992 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 7993 | ins->offset = num_bytes; |
d2fb3437 | 7994 | |
b4bd745d QW |
7995 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
7996 | num_bytes); | |
e570fd27 | 7997 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7998 | break; |
7999 | loop: | |
b4bd745d QW |
8000 | ffe_ctl.retry_clustered = false; |
8001 | ffe_ctl.retry_unclustered = false; | |
3e72ee88 | 8002 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
b4bd745d | 8003 | ffe_ctl.index); |
e570fd27 | 8004 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 8005 | cond_resched(); |
2552d17e JB |
8006 | } |
8007 | up_read(&space_info->groups_sem); | |
8008 | ||
e72d79d6 QW |
8009 | ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl, |
8010 | full_search, use_cluster); | |
8011 | if (ret > 0) | |
b742bb82 YZ |
8012 | goto search; |
8013 | ||
4f4db217 | 8014 | if (ret == -ENOSPC) { |
b4bd745d QW |
8015 | /* |
8016 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
8017 | * any contiguous hole. | |
8018 | */ | |
8019 | if (!ffe_ctl.max_extent_size) | |
8020 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 8021 | spin_lock(&space_info->lock); |
b4bd745d | 8022 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 8023 | spin_unlock(&space_info->lock); |
b4bd745d | 8024 | ins->offset = ffe_ctl.max_extent_size; |
4f4db217 | 8025 | } |
0f70abe2 | 8026 | return ret; |
fec577fb | 8027 | } |
ec44a35c | 8028 | |
b78e5616 JB |
8029 | #define DUMP_BLOCK_RSV(fs_info, rsv_name) \ |
8030 | do { \ | |
8031 | struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name; \ | |
8032 | spin_lock(&__rsv->lock); \ | |
8033 | btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu", \ | |
8034 | __rsv->size, __rsv->reserved); \ | |
8035 | spin_unlock(&__rsv->lock); \ | |
8036 | } while (0) | |
8037 | ||
ab8d0fc4 JM |
8038 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
8039 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 8040 | int dump_block_groups) |
0f9dd46c JB |
8041 | { |
8042 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 8043 | int index = 0; |
0f9dd46c | 8044 | |
9ed74f2d | 8045 | spin_lock(&info->lock); |
ab8d0fc4 JM |
8046 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
8047 | info->flags, | |
4136135b LB |
8048 | info->total_bytes - btrfs_space_info_used(info, true), |
8049 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
8050 | btrfs_info(fs_info, |
8051 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
8052 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
8053 | info->bytes_reserved, info->bytes_may_use, | |
8054 | info->bytes_readonly); | |
9ed74f2d JB |
8055 | spin_unlock(&info->lock); |
8056 | ||
b78e5616 JB |
8057 | DUMP_BLOCK_RSV(fs_info, global_block_rsv); |
8058 | DUMP_BLOCK_RSV(fs_info, trans_block_rsv); | |
8059 | DUMP_BLOCK_RSV(fs_info, chunk_block_rsv); | |
8060 | DUMP_BLOCK_RSV(fs_info, delayed_block_rsv); | |
8061 | DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv); | |
8062 | ||
9ed74f2d JB |
8063 | if (!dump_block_groups) |
8064 | return; | |
0f9dd46c | 8065 | |
80eb234a | 8066 | down_read(&info->groups_sem); |
b742bb82 YZ |
8067 | again: |
8068 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 8069 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
8070 | btrfs_info(fs_info, |
8071 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
8072 | cache->key.objectid, cache->key.offset, | |
8073 | btrfs_block_group_used(&cache->item), cache->pinned, | |
8074 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
8075 | btrfs_dump_free_space(cache, bytes); |
8076 | spin_unlock(&cache->lock); | |
8077 | } | |
b742bb82 YZ |
8078 | if (++index < BTRFS_NR_RAID_TYPES) |
8079 | goto again; | |
80eb234a | 8080 | up_read(&info->groups_sem); |
0f9dd46c | 8081 | } |
e8569813 | 8082 | |
6f47c706 NB |
8083 | /* |
8084 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
8085 | * hole that is at least as big as @num_bytes. | |
8086 | * | |
8087 | * @root - The root that will contain this extent | |
8088 | * | |
8089 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
8090 | * is used for accounting purposes. This value differs | |
8091 | * from @num_bytes only in the case of compressed extents. | |
8092 | * | |
8093 | * @num_bytes - Number of bytes to allocate on-disk. | |
8094 | * | |
8095 | * @min_alloc_size - Indicates the minimum amount of space that the | |
8096 | * allocator should try to satisfy. In some cases | |
8097 | * @num_bytes may be larger than what is required and if | |
8098 | * the filesystem is fragmented then allocation fails. | |
8099 | * However, the presence of @min_alloc_size gives a | |
8100 | * chance to try and satisfy the smaller allocation. | |
8101 | * | |
8102 | * @empty_size - A hint that you plan on doing more COW. This is the | |
8103 | * size in bytes the allocator should try to find free | |
8104 | * next to the block it returns. This is just a hint and | |
8105 | * may be ignored by the allocator. | |
8106 | * | |
8107 | * @hint_byte - Hint to the allocator to start searching above the byte | |
8108 | * address passed. It might be ignored. | |
8109 | * | |
8110 | * @ins - This key is modified to record the found hole. It will | |
8111 | * have the following values: | |
8112 | * ins->objectid == start position | |
8113 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
8114 | * ins->offset == the size of the hole. | |
8115 | * | |
8116 | * @is_data - Boolean flag indicating whether an extent is | |
8117 | * allocated for data (true) or metadata (false) | |
8118 | * | |
8119 | * @delalloc - Boolean flag indicating whether this allocation is for | |
8120 | * delalloc or not. If 'true' data_rwsem of block groups | |
8121 | * is going to be acquired. | |
8122 | * | |
8123 | * | |
8124 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
8125 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
8126 | * largest available hole the allocator managed to find. | |
8127 | */ | |
18513091 | 8128 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
8129 | u64 num_bytes, u64 min_alloc_size, |
8130 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 8131 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 8132 | { |
ab8d0fc4 | 8133 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 8134 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 8135 | u64 flags; |
fec577fb | 8136 | int ret; |
925baedd | 8137 | |
1b86826d | 8138 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 8139 | again: |
0b246afa | 8140 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 8141 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 8142 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 8143 | if (!ret && !is_data) { |
ab8d0fc4 | 8144 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 8145 | } else if (ret == -ENOSPC) { |
a4820398 MX |
8146 | if (!final_tried && ins->offset) { |
8147 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 8148 | num_bytes = round_down(num_bytes, |
0b246afa | 8149 | fs_info->sectorsize); |
9e622d6b | 8150 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 8151 | ram_bytes = num_bytes; |
9e622d6b MX |
8152 | if (num_bytes == min_alloc_size) |
8153 | final_tried = true; | |
8154 | goto again; | |
ab8d0fc4 | 8155 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
8156 | struct btrfs_space_info *sinfo; |
8157 | ||
ab8d0fc4 | 8158 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8159 | btrfs_err(fs_info, |
5d163e0e JM |
8160 | "allocation failed flags %llu, wanted %llu", |
8161 | flags, num_bytes); | |
53804280 | 8162 | if (sinfo) |
ab8d0fc4 | 8163 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8164 | } |
925baedd | 8165 | } |
0f9dd46c JB |
8166 | |
8167 | return ret; | |
e6dcd2dc CM |
8168 | } |
8169 | ||
2ff7e61e | 8170 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8171 | u64 start, u64 len, |
8172 | int pin, int delalloc) | |
65b51a00 | 8173 | { |
0f9dd46c | 8174 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8175 | int ret = 0; |
0f9dd46c | 8176 | |
0b246afa | 8177 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8178 | if (!cache) { |
0b246afa JM |
8179 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8180 | start); | |
0f9dd46c JB |
8181 | return -ENOSPC; |
8182 | } | |
1f3c79a2 | 8183 | |
e688b725 | 8184 | if (pin) |
fdf08605 | 8185 | pin_down_extent(cache, start, len, 1); |
e688b725 | 8186 | else { |
0b246afa | 8187 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8188 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8189 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8190 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8191 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8192 | } |
31193213 | 8193 | |
fa9c0d79 | 8194 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8195 | return ret; |
8196 | } | |
8197 | ||
2ff7e61e | 8198 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8199 | u64 start, u64 len, int delalloc) |
e688b725 | 8200 | { |
2ff7e61e | 8201 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8202 | } |
8203 | ||
2ff7e61e | 8204 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8205 | u64 start, u64 len) |
8206 | { | |
2ff7e61e | 8207 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8208 | } |
8209 | ||
5d4f98a2 | 8210 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
8211 | u64 parent, u64 root_objectid, |
8212 | u64 flags, u64 owner, u64 offset, | |
8213 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 8214 | { |
ef89b824 | 8215 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 8216 | int ret; |
e6dcd2dc | 8217 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8218 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8219 | struct btrfs_path *path; |
5d4f98a2 YZ |
8220 | struct extent_buffer *leaf; |
8221 | int type; | |
8222 | u32 size; | |
26b8003f | 8223 | |
5d4f98a2 YZ |
8224 | if (parent > 0) |
8225 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8226 | else | |
8227 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8228 | |
5d4f98a2 | 8229 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8230 | |
8231 | path = btrfs_alloc_path(); | |
db5b493a TI |
8232 | if (!path) |
8233 | return -ENOMEM; | |
47e4bb98 | 8234 | |
b9473439 | 8235 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8236 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8237 | ins, size); | |
79787eaa JM |
8238 | if (ret) { |
8239 | btrfs_free_path(path); | |
8240 | return ret; | |
8241 | } | |
0f9dd46c | 8242 | |
5d4f98a2 YZ |
8243 | leaf = path->nodes[0]; |
8244 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8245 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8246 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8247 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8248 | btrfs_set_extent_flags(leaf, extent_item, | |
8249 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8250 | ||
8251 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8252 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8253 | if (parent > 0) { | |
8254 | struct btrfs_shared_data_ref *ref; | |
8255 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8256 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8257 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8258 | } else { | |
8259 | struct btrfs_extent_data_ref *ref; | |
8260 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8261 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8262 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8263 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8264 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8265 | } | |
47e4bb98 CM |
8266 | |
8267 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8268 | btrfs_free_path(path); |
f510cfec | 8269 | |
25a356d3 | 8270 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
8271 | if (ret) |
8272 | return ret; | |
8273 | ||
6b279408 | 8274 | ret = update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 8275 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8276 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8277 | ins->objectid, ins->offset); |
f5947066 CM |
8278 | BUG(); |
8279 | } | |
71ff6437 | 8280 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8281 | return ret; |
8282 | } | |
8283 | ||
5d4f98a2 | 8284 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 8285 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 8286 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 8287 | { |
9dcdbe01 | 8288 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 8289 | int ret; |
5d4f98a2 | 8290 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 8291 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
8292 | struct btrfs_tree_block_info *block_info; |
8293 | struct btrfs_extent_inline_ref *iref; | |
8294 | struct btrfs_path *path; | |
8295 | struct extent_buffer *leaf; | |
4e6bd4e0 | 8296 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 8297 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 8298 | u64 num_bytes; |
21ebfbe7 | 8299 | u64 flags = extent_op->flags_to_set; |
0b246afa | 8300 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 8301 | |
4e6bd4e0 NB |
8302 | ref = btrfs_delayed_node_to_tree_ref(node); |
8303 | ||
4e6bd4e0 NB |
8304 | extent_key.objectid = node->bytenr; |
8305 | if (skinny_metadata) { | |
8306 | extent_key.offset = ref->level; | |
8307 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
8308 | num_bytes = fs_info->nodesize; | |
8309 | } else { | |
8310 | extent_key.offset = node->num_bytes; | |
8311 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 8312 | size += sizeof(*block_info); |
4e6bd4e0 NB |
8313 | num_bytes = node->num_bytes; |
8314 | } | |
1c2308f8 | 8315 | |
5d4f98a2 | 8316 | path = btrfs_alloc_path(); |
80ee54bf | 8317 | if (!path) |
d8926bb3 | 8318 | return -ENOMEM; |
56bec294 | 8319 | |
5d4f98a2 YZ |
8320 | path->leave_spinning = 1; |
8321 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 8322 | &extent_key, size); |
79787eaa | 8323 | if (ret) { |
dd825259 | 8324 | btrfs_free_path(path); |
79787eaa JM |
8325 | return ret; |
8326 | } | |
5d4f98a2 YZ |
8327 | |
8328 | leaf = path->nodes[0]; | |
8329 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8330 | struct btrfs_extent_item); | |
8331 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8332 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8333 | btrfs_set_extent_flags(leaf, extent_item, | |
8334 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8335 | |
3173a18f JB |
8336 | if (skinny_metadata) { |
8337 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8338 | } else { | |
8339 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 8340 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 8341 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
8342 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
8343 | } | |
5d4f98a2 | 8344 | |
d4b20733 | 8345 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
8346 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
8347 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8348 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 8349 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
8350 | } else { |
8351 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8352 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 8353 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
8354 | } |
8355 | ||
8356 | btrfs_mark_buffer_dirty(leaf); | |
8357 | btrfs_free_path(path); | |
8358 | ||
4e6bd4e0 NB |
8359 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
8360 | num_bytes); | |
1e144fb8 OS |
8361 | if (ret) |
8362 | return ret; | |
8363 | ||
6b279408 | 8364 | ret = update_block_group(trans, extent_key.objectid, |
6202df69 | 8365 | fs_info->nodesize, 1); |
79787eaa | 8366 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8367 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 8368 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
8369 | BUG(); |
8370 | } | |
0be5dc67 | 8371 | |
4e6bd4e0 | 8372 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 8373 | fs_info->nodesize); |
5d4f98a2 YZ |
8374 | return ret; |
8375 | } | |
8376 | ||
8377 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 8378 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
8379 | u64 offset, u64 ram_bytes, |
8380 | struct btrfs_key *ins) | |
5d4f98a2 YZ |
8381 | { |
8382 | int ret; | |
8383 | ||
84f7d8e6 | 8384 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8385 | |
fd708b81 JB |
8386 | btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0, |
8387 | root->root_key.objectid, owner, offset, | |
8388 | BTRFS_ADD_DELAYED_EXTENT); | |
8389 | ||
88a979c6 | 8390 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, |
84f7d8e6 JB |
8391 | ins->offset, 0, |
8392 | root->root_key.objectid, owner, | |
7be07912 OS |
8393 | offset, ram_bytes, |
8394 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); | |
e6dcd2dc CM |
8395 | return ret; |
8396 | } | |
e02119d5 CM |
8397 | |
8398 | /* | |
8399 | * this is used by the tree logging recovery code. It records that | |
8400 | * an extent has been allocated and makes sure to clear the free | |
8401 | * space cache bits as well | |
8402 | */ | |
5d4f98a2 | 8403 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
8404 | u64 root_objectid, u64 owner, u64 offset, |
8405 | struct btrfs_key *ins) | |
e02119d5 | 8406 | { |
61da2abf | 8407 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 CM |
8408 | int ret; |
8409 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8410 | struct btrfs_space_info *space_info; |
11833d66 | 8411 | |
8c2a1a30 JB |
8412 | /* |
8413 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8414 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8415 | */ |
0b246afa | 8416 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8417 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8418 | ins->offset); | |
b50c6e25 | 8419 | if (ret) |
8c2a1a30 | 8420 | return ret; |
11833d66 YZ |
8421 | } |
8422 | ||
0b246afa | 8423 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8424 | if (!block_group) |
8425 | return -EINVAL; | |
8426 | ||
ed7a6948 WX |
8427 | space_info = block_group->space_info; |
8428 | spin_lock(&space_info->lock); | |
8429 | spin_lock(&block_group->lock); | |
8430 | space_info->bytes_reserved += ins->offset; | |
8431 | block_group->reserved += ins->offset; | |
8432 | spin_unlock(&block_group->lock); | |
8433 | spin_unlock(&space_info->lock); | |
8434 | ||
ef89b824 NB |
8435 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
8436 | offset, ins, 1); | |
b50c6e25 | 8437 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8438 | return ret; |
8439 | } | |
8440 | ||
48a3b636 ES |
8441 | static struct extent_buffer * |
8442 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
bc877d28 | 8443 | u64 bytenr, int level, u64 owner) |
65b51a00 | 8444 | { |
0b246afa | 8445 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8446 | struct extent_buffer *buf; |
8447 | ||
2ff7e61e | 8448 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8449 | if (IS_ERR(buf)) |
8450 | return buf; | |
8451 | ||
b72c3aba QW |
8452 | /* |
8453 | * Extra safety check in case the extent tree is corrupted and extent | |
8454 | * allocator chooses to use a tree block which is already used and | |
8455 | * locked. | |
8456 | */ | |
8457 | if (buf->lock_owner == current->pid) { | |
8458 | btrfs_err_rl(fs_info, | |
8459 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
8460 | buf->start, btrfs_header_owner(buf), current->pid); | |
8461 | free_extent_buffer(buf); | |
8462 | return ERR_PTR(-EUCLEAN); | |
8463 | } | |
8464 | ||
85d4e461 | 8465 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8466 | btrfs_tree_lock(buf); |
6a884d7d | 8467 | btrfs_clean_tree_block(buf); |
3083ee2e | 8468 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 8469 | |
8bead258 | 8470 | btrfs_set_lock_blocking_write(buf); |
4db8c528 | 8471 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8472 | |
bc877d28 NB |
8473 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
8474 | btrfs_set_header_level(buf, level); | |
8475 | btrfs_set_header_bytenr(buf, buf->start); | |
8476 | btrfs_set_header_generation(buf, trans->transid); | |
8477 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
8478 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 8479 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 8480 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 8481 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8482 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8483 | /* |
8484 | * we allow two log transactions at a time, use different | |
52042d8e | 8485 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 8486 | */ |
656f30db | 8487 | if (buf->log_index == 0) |
8cef4e16 YZ |
8488 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8489 | buf->start + buf->len - 1, GFP_NOFS); | |
8490 | else | |
8491 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8492 | buf->start + buf->len - 1); |
d0c803c4 | 8493 | } else { |
656f30db | 8494 | buf->log_index = -1; |
d0c803c4 | 8495 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8496 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8497 | } |
64c12921 | 8498 | trans->dirty = true; |
b4ce94de | 8499 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8500 | return buf; |
8501 | } | |
8502 | ||
f0486c68 YZ |
8503 | static struct btrfs_block_rsv * |
8504 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8505 | struct btrfs_root *root, u32 blocksize) | |
8506 | { | |
0b246afa | 8507 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8508 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8509 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8510 | int ret; |
d88033db | 8511 | bool global_updated = false; |
f0486c68 YZ |
8512 | |
8513 | block_rsv = get_block_rsv(trans, root); | |
8514 | ||
b586b323 MX |
8515 | if (unlikely(block_rsv->size == 0)) |
8516 | goto try_reserve; | |
d88033db | 8517 | again: |
f0486c68 YZ |
8518 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8519 | if (!ret) | |
8520 | return block_rsv; | |
8521 | ||
b586b323 MX |
8522 | if (block_rsv->failfast) |
8523 | return ERR_PTR(ret); | |
8524 | ||
d88033db MX |
8525 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8526 | global_updated = true; | |
0b246afa | 8527 | update_global_block_rsv(fs_info); |
d88033db MX |
8528 | goto again; |
8529 | } | |
8530 | ||
ba2c4d4e JB |
8531 | /* |
8532 | * The global reserve still exists to save us from ourselves, so don't | |
8533 | * warn_on if we are short on our delayed refs reserve. | |
8534 | */ | |
8535 | if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS && | |
8536 | btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
b586b323 MX |
8537 | static DEFINE_RATELIMIT_STATE(_rs, |
8538 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8539 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8540 | if (__ratelimit(&_rs)) | |
8541 | WARN(1, KERN_DEBUG | |
efe120a0 | 8542 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8543 | } |
8544 | try_reserve: | |
8545 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8546 | BTRFS_RESERVE_NO_FLUSH); | |
8547 | if (!ret) | |
8548 | return block_rsv; | |
8549 | /* | |
8550 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8551 | * the global reserve if its space type is the same as the global |
8552 | * reservation. | |
b586b323 | 8553 | */ |
5881cfc9 MX |
8554 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8555 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8556 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8557 | if (!ret) | |
8558 | return global_rsv; | |
8559 | } | |
8560 | return ERR_PTR(ret); | |
f0486c68 YZ |
8561 | } |
8562 | ||
8c2a3ca2 JB |
8563 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8564 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 | 8565 | { |
3a584174 | 8566 | block_rsv_add_bytes(block_rsv, blocksize, false); |
ff6bc37e | 8567 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL); |
f0486c68 YZ |
8568 | } |
8569 | ||
fec577fb | 8570 | /* |
f0486c68 | 8571 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8572 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8573 | */ |
4d75f8a9 | 8574 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8575 | struct btrfs_root *root, |
8576 | u64 parent, u64 root_objectid, | |
8577 | const struct btrfs_disk_key *key, | |
8578 | int level, u64 hint, | |
8579 | u64 empty_size) | |
fec577fb | 8580 | { |
0b246afa | 8581 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8582 | struct btrfs_key ins; |
f0486c68 | 8583 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8584 | struct extent_buffer *buf; |
67b7859e | 8585 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8586 | u64 flags = 0; |
8587 | int ret; | |
0b246afa JM |
8588 | u32 blocksize = fs_info->nodesize; |
8589 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8590 | |
05653ef3 | 8591 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8592 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8593 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
bc877d28 | 8594 | level, root_objectid); |
faa2dbf0 JB |
8595 | if (!IS_ERR(buf)) |
8596 | root->alloc_bytenr += blocksize; | |
8597 | return buf; | |
8598 | } | |
05653ef3 | 8599 | #endif |
fccb84c9 | 8600 | |
f0486c68 YZ |
8601 | block_rsv = use_block_rsv(trans, root, blocksize); |
8602 | if (IS_ERR(block_rsv)) | |
8603 | return ERR_CAST(block_rsv); | |
8604 | ||
18513091 | 8605 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8606 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8607 | if (ret) |
8608 | goto out_unuse; | |
55c69072 | 8609 | |
bc877d28 NB |
8610 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
8611 | root_objectid); | |
67b7859e OS |
8612 | if (IS_ERR(buf)) { |
8613 | ret = PTR_ERR(buf); | |
8614 | goto out_free_reserved; | |
8615 | } | |
f0486c68 YZ |
8616 | |
8617 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8618 | if (parent == 0) | |
8619 | parent = ins.objectid; | |
8620 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8621 | } else | |
8622 | BUG_ON(parent > 0); | |
8623 | ||
8624 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8625 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8626 | if (!extent_op) { |
8627 | ret = -ENOMEM; | |
8628 | goto out_free_buf; | |
8629 | } | |
f0486c68 YZ |
8630 | if (key) |
8631 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8632 | else | |
8633 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8634 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8635 | extent_op->update_key = skinny_metadata ? false : true; |
8636 | extent_op->update_flags = true; | |
8637 | extent_op->is_data = false; | |
b1c79e09 | 8638 | extent_op->level = level; |
f0486c68 | 8639 | |
fd708b81 JB |
8640 | btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent, |
8641 | root_objectid, level, 0, | |
8642 | BTRFS_ADD_DELAYED_EXTENT); | |
44e1c47d | 8643 | ret = btrfs_add_delayed_tree_ref(trans, ins.objectid, |
7be07912 OS |
8644 | ins.offset, parent, |
8645 | root_objectid, level, | |
67b7859e | 8646 | BTRFS_ADD_DELAYED_EXTENT, |
7be07912 | 8647 | extent_op, NULL, NULL); |
67b7859e OS |
8648 | if (ret) |
8649 | goto out_free_delayed; | |
f0486c68 | 8650 | } |
fec577fb | 8651 | return buf; |
67b7859e OS |
8652 | |
8653 | out_free_delayed: | |
8654 | btrfs_free_delayed_extent_op(extent_op); | |
8655 | out_free_buf: | |
8656 | free_extent_buffer(buf); | |
8657 | out_free_reserved: | |
2ff7e61e | 8658 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8659 | out_unuse: |
0b246afa | 8660 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8661 | return ERR_PTR(ret); |
fec577fb | 8662 | } |
a28ec197 | 8663 | |
2c47e605 YZ |
8664 | struct walk_control { |
8665 | u64 refs[BTRFS_MAX_LEVEL]; | |
8666 | u64 flags[BTRFS_MAX_LEVEL]; | |
8667 | struct btrfs_key update_progress; | |
aea6f028 JB |
8668 | struct btrfs_key drop_progress; |
8669 | int drop_level; | |
2c47e605 YZ |
8670 | int stage; |
8671 | int level; | |
8672 | int shared_level; | |
8673 | int update_ref; | |
8674 | int keep_locks; | |
1c4850e2 YZ |
8675 | int reada_slot; |
8676 | int reada_count; | |
78c52d9e | 8677 | int restarted; |
2c47e605 YZ |
8678 | }; |
8679 | ||
8680 | #define DROP_REFERENCE 1 | |
8681 | #define UPDATE_BACKREF 2 | |
8682 | ||
1c4850e2 YZ |
8683 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8684 | struct btrfs_root *root, | |
8685 | struct walk_control *wc, | |
8686 | struct btrfs_path *path) | |
6407bf6d | 8687 | { |
0b246afa | 8688 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8689 | u64 bytenr; |
8690 | u64 generation; | |
8691 | u64 refs; | |
94fcca9f | 8692 | u64 flags; |
5d4f98a2 | 8693 | u32 nritems; |
1c4850e2 YZ |
8694 | struct btrfs_key key; |
8695 | struct extent_buffer *eb; | |
6407bf6d | 8696 | int ret; |
1c4850e2 YZ |
8697 | int slot; |
8698 | int nread = 0; | |
6407bf6d | 8699 | |
1c4850e2 YZ |
8700 | if (path->slots[wc->level] < wc->reada_slot) { |
8701 | wc->reada_count = wc->reada_count * 2 / 3; | |
8702 | wc->reada_count = max(wc->reada_count, 2); | |
8703 | } else { | |
8704 | wc->reada_count = wc->reada_count * 3 / 2; | |
8705 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8706 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8707 | } |
7bb86316 | 8708 | |
1c4850e2 YZ |
8709 | eb = path->nodes[wc->level]; |
8710 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8711 | |
1c4850e2 YZ |
8712 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8713 | if (nread >= wc->reada_count) | |
8714 | break; | |
bd56b302 | 8715 | |
2dd3e67b | 8716 | cond_resched(); |
1c4850e2 YZ |
8717 | bytenr = btrfs_node_blockptr(eb, slot); |
8718 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8719 | |
1c4850e2 YZ |
8720 | if (slot == path->slots[wc->level]) |
8721 | goto reada; | |
5d4f98a2 | 8722 | |
1c4850e2 YZ |
8723 | if (wc->stage == UPDATE_BACKREF && |
8724 | generation <= root->root_key.offset) | |
bd56b302 CM |
8725 | continue; |
8726 | ||
94fcca9f | 8727 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8728 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8729 | wc->level - 1, 1, &refs, |
8730 | &flags); | |
79787eaa JM |
8731 | /* We don't care about errors in readahead. */ |
8732 | if (ret < 0) | |
8733 | continue; | |
94fcca9f YZ |
8734 | BUG_ON(refs == 0); |
8735 | ||
1c4850e2 | 8736 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8737 | if (refs == 1) |
8738 | goto reada; | |
bd56b302 | 8739 | |
94fcca9f YZ |
8740 | if (wc->level == 1 && |
8741 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8742 | continue; | |
1c4850e2 YZ |
8743 | if (!wc->update_ref || |
8744 | generation <= root->root_key.offset) | |
8745 | continue; | |
8746 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8747 | ret = btrfs_comp_cpu_keys(&key, | |
8748 | &wc->update_progress); | |
8749 | if (ret < 0) | |
8750 | continue; | |
94fcca9f YZ |
8751 | } else { |
8752 | if (wc->level == 1 && | |
8753 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8754 | continue; | |
6407bf6d | 8755 | } |
1c4850e2 | 8756 | reada: |
2ff7e61e | 8757 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8758 | nread++; |
20524f02 | 8759 | } |
1c4850e2 | 8760 | wc->reada_slot = slot; |
20524f02 | 8761 | } |
2c47e605 | 8762 | |
f82d02d9 | 8763 | /* |
2c016dc2 | 8764 | * helper to process tree block while walking down the tree. |
2c47e605 | 8765 | * |
2c47e605 YZ |
8766 | * when wc->stage == UPDATE_BACKREF, this function updates |
8767 | * back refs for pointers in the block. | |
8768 | * | |
8769 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8770 | */ |
2c47e605 | 8771 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8772 | struct btrfs_root *root, |
2c47e605 | 8773 | struct btrfs_path *path, |
94fcca9f | 8774 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8775 | { |
2ff7e61e | 8776 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8777 | int level = wc->level; |
8778 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8779 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8780 | int ret; |
8781 | ||
2c47e605 YZ |
8782 | if (wc->stage == UPDATE_BACKREF && |
8783 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8784 | return 1; | |
f82d02d9 | 8785 | |
2c47e605 YZ |
8786 | /* |
8787 | * when reference count of tree block is 1, it won't increase | |
8788 | * again. once full backref flag is set, we never clear it. | |
8789 | */ | |
94fcca9f YZ |
8790 | if (lookup_info && |
8791 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8792 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8793 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8794 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8795 | eb->start, level, 1, |
2c47e605 YZ |
8796 | &wc->refs[level], |
8797 | &wc->flags[level]); | |
79787eaa JM |
8798 | BUG_ON(ret == -ENOMEM); |
8799 | if (ret) | |
8800 | return ret; | |
2c47e605 YZ |
8801 | BUG_ON(wc->refs[level] == 0); |
8802 | } | |
5d4f98a2 | 8803 | |
2c47e605 YZ |
8804 | if (wc->stage == DROP_REFERENCE) { |
8805 | if (wc->refs[level] > 1) | |
8806 | return 1; | |
f82d02d9 | 8807 | |
2c47e605 | 8808 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8809 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8810 | path->locks[level] = 0; |
8811 | } | |
8812 | return 0; | |
8813 | } | |
f82d02d9 | 8814 | |
2c47e605 YZ |
8815 | /* wc->stage == UPDATE_BACKREF */ |
8816 | if (!(wc->flags[level] & flag)) { | |
8817 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8818 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8819 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8820 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8821 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8822 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8823 | eb->len, flag, |
8824 | btrfs_header_level(eb), 0); | |
79787eaa | 8825 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8826 | wc->flags[level] |= flag; |
8827 | } | |
8828 | ||
8829 | /* | |
8830 | * the block is shared by multiple trees, so it's not good to | |
8831 | * keep the tree lock | |
8832 | */ | |
8833 | if (path->locks[level] && level > 0) { | |
bd681513 | 8834 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8835 | path->locks[level] = 0; |
8836 | } | |
8837 | return 0; | |
8838 | } | |
8839 | ||
78c52d9e JB |
8840 | /* |
8841 | * This is used to verify a ref exists for this root to deal with a bug where we | |
8842 | * would have a drop_progress key that hadn't been updated properly. | |
8843 | */ | |
8844 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
8845 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
8846 | int level) | |
8847 | { | |
8848 | struct btrfs_path *path; | |
8849 | struct btrfs_extent_inline_ref *iref; | |
8850 | int ret; | |
8851 | ||
8852 | path = btrfs_alloc_path(); | |
8853 | if (!path) | |
8854 | return -ENOMEM; | |
8855 | ||
8856 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
8857 | root->fs_info->nodesize, parent, | |
8858 | root->root_key.objectid, level, 0); | |
8859 | btrfs_free_path(path); | |
8860 | if (ret == -ENOENT) | |
8861 | return 0; | |
8862 | if (ret < 0) | |
8863 | return ret; | |
8864 | return 1; | |
8865 | } | |
8866 | ||
1c4850e2 | 8867 | /* |
2c016dc2 | 8868 | * helper to process tree block pointer. |
1c4850e2 YZ |
8869 | * |
8870 | * when wc->stage == DROP_REFERENCE, this function checks | |
8871 | * reference count of the block pointed to. if the block | |
8872 | * is shared and we need update back refs for the subtree | |
8873 | * rooted at the block, this function changes wc->stage to | |
8874 | * UPDATE_BACKREF. if the block is shared and there is no | |
8875 | * need to update back, this function drops the reference | |
8876 | * to the block. | |
8877 | * | |
8878 | * NOTE: return value 1 means we should stop walking down. | |
8879 | */ | |
8880 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8881 | struct btrfs_root *root, | |
8882 | struct btrfs_path *path, | |
94fcca9f | 8883 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8884 | { |
0b246afa | 8885 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8886 | u64 bytenr; |
8887 | u64 generation; | |
8888 | u64 parent; | |
1c4850e2 | 8889 | struct btrfs_key key; |
581c1760 | 8890 | struct btrfs_key first_key; |
1c4850e2 YZ |
8891 | struct extent_buffer *next; |
8892 | int level = wc->level; | |
8893 | int reada = 0; | |
8894 | int ret = 0; | |
1152651a | 8895 | bool need_account = false; |
1c4850e2 YZ |
8896 | |
8897 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8898 | path->slots[level]); | |
8899 | /* | |
8900 | * if the lower level block was created before the snapshot | |
8901 | * was created, we know there is no need to update back refs | |
8902 | * for the subtree | |
8903 | */ | |
8904 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8905 | generation <= root->root_key.offset) { |
8906 | *lookup_info = 1; | |
1c4850e2 | 8907 | return 1; |
94fcca9f | 8908 | } |
1c4850e2 YZ |
8909 | |
8910 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
8911 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
8912 | path->slots[level]); | |
1c4850e2 | 8913 | |
0b246afa | 8914 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8915 | if (!next) { |
2ff7e61e | 8916 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8917 | if (IS_ERR(next)) |
8918 | return PTR_ERR(next); | |
8919 | ||
b2aaaa3b JB |
8920 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8921 | level - 1); | |
1c4850e2 YZ |
8922 | reada = 1; |
8923 | } | |
8924 | btrfs_tree_lock(next); | |
8bead258 | 8925 | btrfs_set_lock_blocking_write(next); |
1c4850e2 | 8926 | |
2ff7e61e | 8927 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8928 | &wc->refs[level - 1], |
8929 | &wc->flags[level - 1]); | |
4867268c JB |
8930 | if (ret < 0) |
8931 | goto out_unlock; | |
79787eaa | 8932 | |
c2cf52eb | 8933 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8934 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8935 | ret = -EIO; |
8936 | goto out_unlock; | |
c2cf52eb | 8937 | } |
94fcca9f | 8938 | *lookup_info = 0; |
1c4850e2 | 8939 | |
94fcca9f | 8940 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8941 | if (wc->refs[level - 1] > 1) { |
1152651a | 8942 | need_account = true; |
94fcca9f YZ |
8943 | if (level == 1 && |
8944 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8945 | goto skip; | |
8946 | ||
1c4850e2 YZ |
8947 | if (!wc->update_ref || |
8948 | generation <= root->root_key.offset) | |
8949 | goto skip; | |
8950 | ||
8951 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8952 | path->slots[level]); | |
8953 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8954 | if (ret < 0) | |
8955 | goto skip; | |
8956 | ||
8957 | wc->stage = UPDATE_BACKREF; | |
8958 | wc->shared_level = level - 1; | |
8959 | } | |
94fcca9f YZ |
8960 | } else { |
8961 | if (level == 1 && | |
8962 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8963 | goto skip; | |
1c4850e2 YZ |
8964 | } |
8965 | ||
b9fab919 | 8966 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8967 | btrfs_tree_unlock(next); |
8968 | free_extent_buffer(next); | |
8969 | next = NULL; | |
94fcca9f | 8970 | *lookup_info = 1; |
1c4850e2 YZ |
8971 | } |
8972 | ||
8973 | if (!next) { | |
8974 | if (reada && level == 1) | |
8975 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
8976 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
8977 | &first_key); | |
64c043de LB |
8978 | if (IS_ERR(next)) { |
8979 | return PTR_ERR(next); | |
8980 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8981 | free_extent_buffer(next); |
97d9a8a4 | 8982 | return -EIO; |
416bc658 | 8983 | } |
1c4850e2 | 8984 | btrfs_tree_lock(next); |
8bead258 | 8985 | btrfs_set_lock_blocking_write(next); |
1c4850e2 YZ |
8986 | } |
8987 | ||
8988 | level--; | |
4867268c JB |
8989 | ASSERT(level == btrfs_header_level(next)); |
8990 | if (level != btrfs_header_level(next)) { | |
8991 | btrfs_err(root->fs_info, "mismatched level"); | |
8992 | ret = -EIO; | |
8993 | goto out_unlock; | |
8994 | } | |
1c4850e2 YZ |
8995 | path->nodes[level] = next; |
8996 | path->slots[level] = 0; | |
bd681513 | 8997 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8998 | wc->level = level; |
8999 | if (wc->level == 1) | |
9000 | wc->reada_slot = 0; | |
9001 | return 0; | |
9002 | skip: | |
9003 | wc->refs[level - 1] = 0; | |
9004 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
9005 | if (wc->stage == DROP_REFERENCE) { |
9006 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
9007 | parent = path->nodes[level]->start; | |
9008 | } else { | |
4867268c | 9009 | ASSERT(root->root_key.objectid == |
94fcca9f | 9010 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
9011 | if (root->root_key.objectid != |
9012 | btrfs_header_owner(path->nodes[level])) { | |
9013 | btrfs_err(root->fs_info, | |
9014 | "mismatched block owner"); | |
9015 | ret = -EIO; | |
9016 | goto out_unlock; | |
9017 | } | |
94fcca9f YZ |
9018 | parent = 0; |
9019 | } | |
1c4850e2 | 9020 | |
78c52d9e JB |
9021 | /* |
9022 | * If we had a drop_progress we need to verify the refs are set | |
9023 | * as expected. If we find our ref then we know that from here | |
9024 | * on out everything should be correct, and we can clear the | |
9025 | * ->restarted flag. | |
9026 | */ | |
9027 | if (wc->restarted) { | |
9028 | ret = check_ref_exists(trans, root, bytenr, parent, | |
9029 | level - 1); | |
9030 | if (ret < 0) | |
9031 | goto out_unlock; | |
9032 | if (ret == 0) | |
9033 | goto no_delete; | |
9034 | ret = 0; | |
9035 | wc->restarted = 0; | |
9036 | } | |
9037 | ||
2cd86d30 QW |
9038 | /* |
9039 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
9040 | * already accounted them at merge time (replace_path), | |
9041 | * thus we could skip expensive subtree trace here. | |
9042 | */ | |
9043 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
9044 | need_account) { | |
deb40627 | 9045 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 9046 | generation, level - 1); |
1152651a | 9047 | if (ret) { |
0b246afa | 9048 | btrfs_err_rl(fs_info, |
5d163e0e JM |
9049 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
9050 | ret); | |
1152651a MF |
9051 | } |
9052 | } | |
aea6f028 JB |
9053 | |
9054 | /* | |
9055 | * We need to update the next key in our walk control so we can | |
9056 | * update the drop_progress key accordingly. We don't care if | |
9057 | * find_next_key doesn't find a key because that means we're at | |
9058 | * the end and are going to clean up now. | |
9059 | */ | |
9060 | wc->drop_level = level; | |
9061 | find_next_key(path, level, &wc->drop_progress); | |
9062 | ||
01e0da48 | 9063 | ret = btrfs_free_extent(trans, root, bytenr, fs_info->nodesize, |
2ff7e61e JM |
9064 | parent, root->root_key.objectid, |
9065 | level - 1, 0); | |
4867268c JB |
9066 | if (ret) |
9067 | goto out_unlock; | |
1c4850e2 | 9068 | } |
78c52d9e | 9069 | no_delete: |
4867268c JB |
9070 | *lookup_info = 1; |
9071 | ret = 1; | |
9072 | ||
9073 | out_unlock: | |
1c4850e2 YZ |
9074 | btrfs_tree_unlock(next); |
9075 | free_extent_buffer(next); | |
4867268c JB |
9076 | |
9077 | return ret; | |
1c4850e2 YZ |
9078 | } |
9079 | ||
2c47e605 | 9080 | /* |
2c016dc2 | 9081 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
9082 | * |
9083 | * when wc->stage == DROP_REFERENCE, this function drops | |
9084 | * reference count on the block. | |
9085 | * | |
9086 | * when wc->stage == UPDATE_BACKREF, this function changes | |
9087 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
9088 | * to UPDATE_BACKREF previously while processing the block. | |
9089 | * | |
9090 | * NOTE: return value 1 means we should stop walking up. | |
9091 | */ | |
9092 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
9093 | struct btrfs_root *root, | |
9094 | struct btrfs_path *path, | |
9095 | struct walk_control *wc) | |
9096 | { | |
0b246afa | 9097 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 9098 | int ret; |
2c47e605 YZ |
9099 | int level = wc->level; |
9100 | struct extent_buffer *eb = path->nodes[level]; | |
9101 | u64 parent = 0; | |
9102 | ||
9103 | if (wc->stage == UPDATE_BACKREF) { | |
9104 | BUG_ON(wc->shared_level < level); | |
9105 | if (level < wc->shared_level) | |
9106 | goto out; | |
9107 | ||
2c47e605 YZ |
9108 | ret = find_next_key(path, level + 1, &wc->update_progress); |
9109 | if (ret > 0) | |
9110 | wc->update_ref = 0; | |
9111 | ||
9112 | wc->stage = DROP_REFERENCE; | |
9113 | wc->shared_level = -1; | |
9114 | path->slots[level] = 0; | |
9115 | ||
9116 | /* | |
9117 | * check reference count again if the block isn't locked. | |
9118 | * we should start walking down the tree again if reference | |
9119 | * count is one. | |
9120 | */ | |
9121 | if (!path->locks[level]) { | |
9122 | BUG_ON(level == 0); | |
9123 | btrfs_tree_lock(eb); | |
8bead258 | 9124 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 9125 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9126 | |
2ff7e61e | 9127 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 9128 | eb->start, level, 1, |
2c47e605 YZ |
9129 | &wc->refs[level], |
9130 | &wc->flags[level]); | |
79787eaa JM |
9131 | if (ret < 0) { |
9132 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 9133 | path->locks[level] = 0; |
79787eaa JM |
9134 | return ret; |
9135 | } | |
2c47e605 YZ |
9136 | BUG_ON(wc->refs[level] == 0); |
9137 | if (wc->refs[level] == 1) { | |
bd681513 | 9138 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 9139 | path->locks[level] = 0; |
2c47e605 YZ |
9140 | return 1; |
9141 | } | |
f82d02d9 | 9142 | } |
2c47e605 | 9143 | } |
f82d02d9 | 9144 | |
2c47e605 YZ |
9145 | /* wc->stage == DROP_REFERENCE */ |
9146 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 9147 | |
2c47e605 YZ |
9148 | if (wc->refs[level] == 1) { |
9149 | if (level == 0) { | |
9150 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 9151 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 9152 | else |
e339a6b0 | 9153 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 9154 | BUG_ON(ret); /* -ENOMEM */ |
8d38d7eb | 9155 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); |
1152651a | 9156 | if (ret) { |
0b246afa | 9157 | btrfs_err_rl(fs_info, |
5d163e0e JM |
9158 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
9159 | ret); | |
1152651a | 9160 | } |
2c47e605 | 9161 | } |
6a884d7d | 9162 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
9163 | if (!path->locks[level] && |
9164 | btrfs_header_generation(eb) == trans->transid) { | |
9165 | btrfs_tree_lock(eb); | |
8bead258 | 9166 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 9167 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9168 | } |
6a884d7d | 9169 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
9170 | } |
9171 | ||
9172 | if (eb == root->node) { | |
9173 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9174 | parent = eb->start; | |
65c6e82b QW |
9175 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
9176 | goto owner_mismatch; | |
2c47e605 YZ |
9177 | } else { |
9178 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9179 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
9180 | else if (root->root_key.objectid != |
9181 | btrfs_header_owner(path->nodes[level + 1])) | |
9182 | goto owner_mismatch; | |
f82d02d9 | 9183 | } |
f82d02d9 | 9184 | |
5581a51a | 9185 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
9186 | out: |
9187 | wc->refs[level] = 0; | |
9188 | wc->flags[level] = 0; | |
f0486c68 | 9189 | return 0; |
65c6e82b QW |
9190 | |
9191 | owner_mismatch: | |
9192 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
9193 | btrfs_header_owner(eb), root->root_key.objectid); | |
9194 | return -EUCLEAN; | |
2c47e605 YZ |
9195 | } |
9196 | ||
9197 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
9198 | struct btrfs_root *root, | |
9199 | struct btrfs_path *path, | |
9200 | struct walk_control *wc) | |
9201 | { | |
2c47e605 | 9202 | int level = wc->level; |
94fcca9f | 9203 | int lookup_info = 1; |
2c47e605 YZ |
9204 | int ret; |
9205 | ||
9206 | while (level >= 0) { | |
94fcca9f | 9207 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
9208 | if (ret > 0) |
9209 | break; | |
9210 | ||
9211 | if (level == 0) | |
9212 | break; | |
9213 | ||
7a7965f8 YZ |
9214 | if (path->slots[level] >= |
9215 | btrfs_header_nritems(path->nodes[level])) | |
9216 | break; | |
9217 | ||
94fcca9f | 9218 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
9219 | if (ret > 0) { |
9220 | path->slots[level]++; | |
9221 | continue; | |
90d2c51d MX |
9222 | } else if (ret < 0) |
9223 | return ret; | |
1c4850e2 | 9224 | level = wc->level; |
f82d02d9 | 9225 | } |
f82d02d9 YZ |
9226 | return 0; |
9227 | } | |
9228 | ||
d397712b | 9229 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 9230 | struct btrfs_root *root, |
f82d02d9 | 9231 | struct btrfs_path *path, |
2c47e605 | 9232 | struct walk_control *wc, int max_level) |
20524f02 | 9233 | { |
2c47e605 | 9234 | int level = wc->level; |
20524f02 | 9235 | int ret; |
9f3a7427 | 9236 | |
2c47e605 YZ |
9237 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
9238 | while (level < max_level && path->nodes[level]) { | |
9239 | wc->level = level; | |
9240 | if (path->slots[level] + 1 < | |
9241 | btrfs_header_nritems(path->nodes[level])) { | |
9242 | path->slots[level]++; | |
20524f02 CM |
9243 | return 0; |
9244 | } else { | |
2c47e605 YZ |
9245 | ret = walk_up_proc(trans, root, path, wc); |
9246 | if (ret > 0) | |
9247 | return 0; | |
65c6e82b QW |
9248 | if (ret < 0) |
9249 | return ret; | |
bd56b302 | 9250 | |
2c47e605 | 9251 | if (path->locks[level]) { |
bd681513 CM |
9252 | btrfs_tree_unlock_rw(path->nodes[level], |
9253 | path->locks[level]); | |
2c47e605 | 9254 | path->locks[level] = 0; |
f82d02d9 | 9255 | } |
2c47e605 YZ |
9256 | free_extent_buffer(path->nodes[level]); |
9257 | path->nodes[level] = NULL; | |
9258 | level++; | |
20524f02 CM |
9259 | } |
9260 | } | |
9261 | return 1; | |
9262 | } | |
9263 | ||
9aca1d51 | 9264 | /* |
2c47e605 YZ |
9265 | * drop a subvolume tree. |
9266 | * | |
9267 | * this function traverses the tree freeing any blocks that only | |
9268 | * referenced by the tree. | |
9269 | * | |
9270 | * when a shared tree block is found. this function decreases its | |
9271 | * reference count by one. if update_ref is true, this function | |
9272 | * also make sure backrefs for the shared block and all lower level | |
9273 | * blocks are properly updated. | |
9d1a2a3a DS |
9274 | * |
9275 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9276 | */ |
2c536799 | 9277 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9278 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9279 | int for_reloc) | |
20524f02 | 9280 | { |
ab8d0fc4 | 9281 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9282 | struct btrfs_path *path; |
2c47e605 | 9283 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9284 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9285 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9286 | struct walk_control *wc; |
9287 | struct btrfs_key key; | |
9288 | int err = 0; | |
9289 | int ret; | |
9290 | int level; | |
d29a9f62 | 9291 | bool root_dropped = false; |
20524f02 | 9292 | |
4fd786e6 | 9293 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 9294 | |
5caf2a00 | 9295 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9296 | if (!path) { |
9297 | err = -ENOMEM; | |
9298 | goto out; | |
9299 | } | |
20524f02 | 9300 | |
2c47e605 | 9301 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9302 | if (!wc) { |
9303 | btrfs_free_path(path); | |
cb1b69f4 TI |
9304 | err = -ENOMEM; |
9305 | goto out; | |
38a1a919 | 9306 | } |
2c47e605 | 9307 | |
a22285a6 | 9308 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9309 | if (IS_ERR(trans)) { |
9310 | err = PTR_ERR(trans); | |
9311 | goto out_free; | |
9312 | } | |
98d5dc13 | 9313 | |
0568e82d JB |
9314 | err = btrfs_run_delayed_items(trans); |
9315 | if (err) | |
9316 | goto out_end_trans; | |
9317 | ||
3fd0a558 YZ |
9318 | if (block_rsv) |
9319 | trans->block_rsv = block_rsv; | |
2c47e605 | 9320 | |
83354f07 JB |
9321 | /* |
9322 | * This will help us catch people modifying the fs tree while we're | |
9323 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
9324 | * dropped as we unlock the root node and parent nodes as we walk down | |
9325 | * the tree, assuming nothing will change. If something does change | |
9326 | * then we'll have stale information and drop references to blocks we've | |
9327 | * already dropped. | |
9328 | */ | |
9329 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 9330 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9331 | level = btrfs_header_level(root->node); |
5d4f98a2 | 9332 | path->nodes[level] = btrfs_lock_root_node(root); |
8bead258 | 9333 | btrfs_set_lock_blocking_write(path->nodes[level]); |
9f3a7427 | 9334 | path->slots[level] = 0; |
bd681513 | 9335 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9336 | memset(&wc->update_progress, 0, |
9337 | sizeof(wc->update_progress)); | |
9f3a7427 | 9338 | } else { |
9f3a7427 | 9339 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9340 | memcpy(&wc->update_progress, &key, |
9341 | sizeof(wc->update_progress)); | |
9342 | ||
6702ed49 | 9343 | level = root_item->drop_level; |
2c47e605 | 9344 | BUG_ON(level == 0); |
6702ed49 | 9345 | path->lowest_level = level; |
2c47e605 YZ |
9346 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9347 | path->lowest_level = 0; | |
9348 | if (ret < 0) { | |
9349 | err = ret; | |
79787eaa | 9350 | goto out_end_trans; |
9f3a7427 | 9351 | } |
1c4850e2 | 9352 | WARN_ON(ret > 0); |
2c47e605 | 9353 | |
7d9eb12c CM |
9354 | /* |
9355 | * unlock our path, this is safe because only this | |
9356 | * function is allowed to delete this snapshot | |
9357 | */ | |
5d4f98a2 | 9358 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9359 | |
9360 | level = btrfs_header_level(root->node); | |
9361 | while (1) { | |
9362 | btrfs_tree_lock(path->nodes[level]); | |
8bead258 | 9363 | btrfs_set_lock_blocking_write(path->nodes[level]); |
fec386ac | 9364 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9365 | |
2ff7e61e | 9366 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9367 | path->nodes[level]->start, |
3173a18f | 9368 | level, 1, &wc->refs[level], |
2c47e605 | 9369 | &wc->flags[level]); |
79787eaa JM |
9370 | if (ret < 0) { |
9371 | err = ret; | |
9372 | goto out_end_trans; | |
9373 | } | |
2c47e605 YZ |
9374 | BUG_ON(wc->refs[level] == 0); |
9375 | ||
9376 | if (level == root_item->drop_level) | |
9377 | break; | |
9378 | ||
9379 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9380 | path->locks[level] = 0; |
2c47e605 YZ |
9381 | WARN_ON(wc->refs[level] != 1); |
9382 | level--; | |
9383 | } | |
9f3a7427 | 9384 | } |
2c47e605 | 9385 | |
78c52d9e | 9386 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
9387 | wc->level = level; |
9388 | wc->shared_level = -1; | |
9389 | wc->stage = DROP_REFERENCE; | |
9390 | wc->update_ref = update_ref; | |
9391 | wc->keep_locks = 0; | |
0b246afa | 9392 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9393 | |
d397712b | 9394 | while (1) { |
9d1a2a3a | 9395 | |
2c47e605 YZ |
9396 | ret = walk_down_tree(trans, root, path, wc); |
9397 | if (ret < 0) { | |
9398 | err = ret; | |
20524f02 | 9399 | break; |
2c47e605 | 9400 | } |
9aca1d51 | 9401 | |
2c47e605 YZ |
9402 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9403 | if (ret < 0) { | |
9404 | err = ret; | |
20524f02 | 9405 | break; |
2c47e605 YZ |
9406 | } |
9407 | ||
9408 | if (ret > 0) { | |
9409 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9410 | break; |
9411 | } | |
2c47e605 YZ |
9412 | |
9413 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
9414 | wc->drop_level = wc->level; |
9415 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
9416 | &wc->drop_progress, | |
9417 | path->slots[wc->drop_level]); | |
9418 | } | |
9419 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
9420 | &wc->drop_progress); | |
9421 | root_item->drop_level = wc->drop_level; | |
2c47e605 YZ |
9422 | |
9423 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9424 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9425 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9426 | ret = btrfs_update_root(trans, tree_root, |
9427 | &root->root_key, | |
9428 | root_item); | |
79787eaa | 9429 | if (ret) { |
66642832 | 9430 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9431 | err = ret; |
9432 | goto out_end_trans; | |
9433 | } | |
2c47e605 | 9434 | |
3a45bb20 | 9435 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9436 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9437 | btrfs_debug(fs_info, |
9438 | "drop snapshot early exit"); | |
3c8f2422 JB |
9439 | err = -EAGAIN; |
9440 | goto out_free; | |
9441 | } | |
9442 | ||
a22285a6 | 9443 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9444 | if (IS_ERR(trans)) { |
9445 | err = PTR_ERR(trans); | |
9446 | goto out_free; | |
9447 | } | |
3fd0a558 YZ |
9448 | if (block_rsv) |
9449 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9450 | } |
20524f02 | 9451 | } |
b3b4aa74 | 9452 | btrfs_release_path(path); |
79787eaa JM |
9453 | if (err) |
9454 | goto out_end_trans; | |
2c47e605 | 9455 | |
ab9ce7d4 | 9456 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 9457 | if (ret) { |
66642832 | 9458 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 9459 | err = ret; |
79787eaa JM |
9460 | goto out_end_trans; |
9461 | } | |
2c47e605 | 9462 | |
76dda93c | 9463 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9464 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9465 | NULL, NULL); | |
79787eaa | 9466 | if (ret < 0) { |
66642832 | 9467 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9468 | err = ret; |
9469 | goto out_end_trans; | |
9470 | } else if (ret > 0) { | |
84cd948c JB |
9471 | /* if we fail to delete the orphan item this time |
9472 | * around, it'll get picked up the next time. | |
9473 | * | |
9474 | * The most common failure here is just -ENOENT. | |
9475 | */ | |
9476 | btrfs_del_orphan_item(trans, tree_root, | |
9477 | root->root_key.objectid); | |
76dda93c YZ |
9478 | } |
9479 | } | |
9480 | ||
27cdeb70 | 9481 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9482 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9483 | } else { |
9484 | free_extent_buffer(root->node); | |
9485 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9486 | btrfs_put_fs_root(root); |
76dda93c | 9487 | } |
d29a9f62 | 9488 | root_dropped = true; |
79787eaa | 9489 | out_end_trans: |
3a45bb20 | 9490 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9491 | out_free: |
2c47e605 | 9492 | kfree(wc); |
5caf2a00 | 9493 | btrfs_free_path(path); |
cb1b69f4 | 9494 | out: |
d29a9f62 JB |
9495 | /* |
9496 | * So if we need to stop dropping the snapshot for whatever reason we | |
9497 | * need to make sure to add it back to the dead root list so that we | |
9498 | * keep trying to do the work later. This also cleans up roots if we | |
9499 | * don't have it in the radix (like when we recover after a power fail | |
9500 | * or unmount) so we don't leak memory. | |
9501 | */ | |
897ca819 | 9502 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9503 | btrfs_add_dead_root(root); |
90515e7f | 9504 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9505 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9506 | return err; |
20524f02 | 9507 | } |
9078a3e1 | 9508 | |
2c47e605 YZ |
9509 | /* |
9510 | * drop subtree rooted at tree block 'node'. | |
9511 | * | |
9512 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9513 | * only used by relocation code |
2c47e605 | 9514 | */ |
f82d02d9 YZ |
9515 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9516 | struct btrfs_root *root, | |
9517 | struct extent_buffer *node, | |
9518 | struct extent_buffer *parent) | |
9519 | { | |
0b246afa | 9520 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9521 | struct btrfs_path *path; |
2c47e605 | 9522 | struct walk_control *wc; |
f82d02d9 YZ |
9523 | int level; |
9524 | int parent_level; | |
9525 | int ret = 0; | |
9526 | int wret; | |
9527 | ||
2c47e605 YZ |
9528 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9529 | ||
f82d02d9 | 9530 | path = btrfs_alloc_path(); |
db5b493a TI |
9531 | if (!path) |
9532 | return -ENOMEM; | |
f82d02d9 | 9533 | |
2c47e605 | 9534 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9535 | if (!wc) { |
9536 | btrfs_free_path(path); | |
9537 | return -ENOMEM; | |
9538 | } | |
2c47e605 | 9539 | |
b9447ef8 | 9540 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9541 | parent_level = btrfs_header_level(parent); |
9542 | extent_buffer_get(parent); | |
9543 | path->nodes[parent_level] = parent; | |
9544 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9545 | ||
b9447ef8 | 9546 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9547 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9548 | path->nodes[level] = node; |
9549 | path->slots[level] = 0; | |
bd681513 | 9550 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9551 | |
9552 | wc->refs[parent_level] = 1; | |
9553 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9554 | wc->level = level; | |
9555 | wc->shared_level = -1; | |
9556 | wc->stage = DROP_REFERENCE; | |
9557 | wc->update_ref = 0; | |
9558 | wc->keep_locks = 1; | |
0b246afa | 9559 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9560 | |
9561 | while (1) { | |
2c47e605 YZ |
9562 | wret = walk_down_tree(trans, root, path, wc); |
9563 | if (wret < 0) { | |
f82d02d9 | 9564 | ret = wret; |
f82d02d9 | 9565 | break; |
2c47e605 | 9566 | } |
f82d02d9 | 9567 | |
2c47e605 | 9568 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9569 | if (wret < 0) |
9570 | ret = wret; | |
9571 | if (wret != 0) | |
9572 | break; | |
9573 | } | |
9574 | ||
2c47e605 | 9575 | kfree(wc); |
f82d02d9 YZ |
9576 | btrfs_free_path(path); |
9577 | return ret; | |
9578 | } | |
9579 | ||
6202df69 | 9580 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9581 | { |
9582 | u64 num_devices; | |
fc67c450 | 9583 | u64 stripped; |
e4d8ec0f | 9584 | |
fc67c450 ID |
9585 | /* |
9586 | * if restripe for this chunk_type is on pick target profile and | |
9587 | * return, otherwise do the usual balance | |
9588 | */ | |
6202df69 | 9589 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9590 | if (stripped) |
9591 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9592 | |
6202df69 | 9593 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9594 | |
fc67c450 | 9595 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9596 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9597 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9598 | ||
ec44a35c CM |
9599 | if (num_devices == 1) { |
9600 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9601 | stripped = flags & ~stripped; | |
9602 | ||
9603 | /* turn raid0 into single device chunks */ | |
9604 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9605 | return stripped; | |
9606 | ||
9607 | /* turn mirroring into duplication */ | |
9608 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9609 | BTRFS_BLOCK_GROUP_RAID10)) | |
9610 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9611 | } else { |
9612 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9613 | if (flags & stripped) |
9614 | return flags; | |
9615 | ||
9616 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9617 | stripped = flags & ~stripped; | |
9618 | ||
9619 | /* switch duplicated blocks with raid1 */ | |
9620 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9621 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9622 | ||
e3176ca2 | 9623 | /* this is drive concat, leave it alone */ |
ec44a35c | 9624 | } |
e3176ca2 | 9625 | |
ec44a35c CM |
9626 | return flags; |
9627 | } | |
9628 | ||
868f401a | 9629 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9630 | { |
f0486c68 YZ |
9631 | struct btrfs_space_info *sinfo = cache->space_info; |
9632 | u64 num_bytes; | |
3ece54e5 | 9633 | u64 sinfo_used; |
199c36ea | 9634 | u64 min_allocable_bytes; |
f0486c68 | 9635 | int ret = -ENOSPC; |
0ef3e66b | 9636 | |
199c36ea MX |
9637 | /* |
9638 | * We need some metadata space and system metadata space for | |
9639 | * allocating chunks in some corner cases until we force to set | |
9640 | * it to be readonly. | |
9641 | */ | |
9642 | if ((sinfo->flags & | |
9643 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9644 | !force) | |
ee22184b | 9645 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9646 | else |
9647 | min_allocable_bytes = 0; | |
9648 | ||
f0486c68 YZ |
9649 | spin_lock(&sinfo->lock); |
9650 | spin_lock(&cache->lock); | |
61cfea9b W |
9651 | |
9652 | if (cache->ro) { | |
868f401a | 9653 | cache->ro++; |
61cfea9b W |
9654 | ret = 0; |
9655 | goto out; | |
9656 | } | |
9657 | ||
f0486c68 YZ |
9658 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9659 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
3ece54e5 | 9660 | sinfo_used = btrfs_space_info_used(sinfo, true); |
f0486c68 | 9661 | |
3ece54e5 QW |
9662 | if (sinfo_used + num_bytes + min_allocable_bytes <= |
9663 | sinfo->total_bytes) { | |
f0486c68 | 9664 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9665 | cache->ro++; |
633c0aad | 9666 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9667 | ret = 0; |
9668 | } | |
61cfea9b | 9669 | out: |
f0486c68 YZ |
9670 | spin_unlock(&cache->lock); |
9671 | spin_unlock(&sinfo->lock); | |
3ece54e5 QW |
9672 | if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { |
9673 | btrfs_info(cache->fs_info, | |
9674 | "unable to make block group %llu ro", | |
9675 | cache->key.objectid); | |
9676 | btrfs_info(cache->fs_info, | |
9677 | "sinfo_used=%llu bg_num_bytes=%llu min_allocable=%llu", | |
9678 | sinfo_used, num_bytes, min_allocable_bytes); | |
9679 | dump_space_info(cache->fs_info, cache->space_info, 0, 0); | |
9680 | } | |
f0486c68 YZ |
9681 | return ret; |
9682 | } | |
7d9eb12c | 9683 | |
c83488af | 9684 | int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache) |
c286ac48 | 9685 | |
f0486c68 | 9686 | { |
c83488af | 9687 | struct btrfs_fs_info *fs_info = cache->fs_info; |
f0486c68 YZ |
9688 | struct btrfs_trans_handle *trans; |
9689 | u64 alloc_flags; | |
9690 | int ret; | |
7d9eb12c | 9691 | |
1bbc621e | 9692 | again: |
5e00f193 | 9693 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9694 | if (IS_ERR(trans)) |
9695 | return PTR_ERR(trans); | |
5d4f98a2 | 9696 | |
1bbc621e CM |
9697 | /* |
9698 | * we're not allowed to set block groups readonly after the dirty | |
9699 | * block groups cache has started writing. If it already started, | |
9700 | * back off and let this transaction commit | |
9701 | */ | |
0b246afa | 9702 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9703 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9704 | u64 transid = trans->transid; |
9705 | ||
0b246afa | 9706 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9707 | btrfs_end_transaction(trans); |
1bbc621e | 9708 | |
2ff7e61e | 9709 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9710 | if (ret) |
9711 | return ret; | |
9712 | goto again; | |
9713 | } | |
9714 | ||
153c35b6 CM |
9715 | /* |
9716 | * if we are changing raid levels, try to allocate a corresponding | |
9717 | * block group with the new raid level. | |
9718 | */ | |
0b246afa | 9719 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9720 | if (alloc_flags != cache->flags) { |
01458828 | 9721 | ret = do_chunk_alloc(trans, alloc_flags, |
153c35b6 CM |
9722 | CHUNK_ALLOC_FORCE); |
9723 | /* | |
9724 | * ENOSPC is allowed here, we may have enough space | |
9725 | * already allocated at the new raid level to | |
9726 | * carry on | |
9727 | */ | |
9728 | if (ret == -ENOSPC) | |
9729 | ret = 0; | |
9730 | if (ret < 0) | |
9731 | goto out; | |
9732 | } | |
1bbc621e | 9733 | |
868f401a | 9734 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9735 | if (!ret) |
9736 | goto out; | |
2ff7e61e | 9737 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
01458828 | 9738 | ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9739 | if (ret < 0) |
9740 | goto out; | |
868f401a | 9741 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9742 | out: |
2f081088 | 9743 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9744 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9745 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 9746 | check_system_chunk(trans, alloc_flags); |
34441361 | 9747 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9748 | } |
0b246afa | 9749 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9750 | |
3a45bb20 | 9751 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9752 | return ret; |
9753 | } | |
5d4f98a2 | 9754 | |
43a7e99d | 9755 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) |
c87f08ca | 9756 | { |
43a7e99d | 9757 | u64 alloc_flags = get_alloc_profile(trans->fs_info, type); |
2ff7e61e | 9758 | |
01458828 | 9759 | return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9760 | } |
9761 | ||
6d07bcec MX |
9762 | /* |
9763 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9764 | * space_info. takes mirrors into account. |
6d07bcec | 9765 | */ |
633c0aad | 9766 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9767 | { |
9768 | struct btrfs_block_group_cache *block_group; | |
9769 | u64 free_bytes = 0; | |
9770 | int factor; | |
9771 | ||
01327610 | 9772 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9773 | if (list_empty(&sinfo->ro_bgs)) |
9774 | return 0; | |
9775 | ||
9776 | spin_lock(&sinfo->lock); | |
9777 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9778 | spin_lock(&block_group->lock); |
9779 | ||
9780 | if (!block_group->ro) { | |
9781 | spin_unlock(&block_group->lock); | |
9782 | continue; | |
9783 | } | |
9784 | ||
46df06b8 | 9785 | factor = btrfs_bg_type_to_factor(block_group->flags); |
6d07bcec MX |
9786 | free_bytes += (block_group->key.offset - |
9787 | btrfs_block_group_used(&block_group->item)) * | |
9788 | factor; | |
9789 | ||
9790 | spin_unlock(&block_group->lock); | |
9791 | } | |
6d07bcec MX |
9792 | spin_unlock(&sinfo->lock); |
9793 | ||
9794 | return free_bytes; | |
9795 | } | |
9796 | ||
2ff7e61e | 9797 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9798 | { |
f0486c68 YZ |
9799 | struct btrfs_space_info *sinfo = cache->space_info; |
9800 | u64 num_bytes; | |
9801 | ||
9802 | BUG_ON(!cache->ro); | |
9803 | ||
9804 | spin_lock(&sinfo->lock); | |
9805 | spin_lock(&cache->lock); | |
868f401a Z |
9806 | if (!--cache->ro) { |
9807 | num_bytes = cache->key.offset - cache->reserved - | |
9808 | cache->pinned - cache->bytes_super - | |
9809 | btrfs_block_group_used(&cache->item); | |
9810 | sinfo->bytes_readonly -= num_bytes; | |
9811 | list_del_init(&cache->ro_list); | |
9812 | } | |
f0486c68 YZ |
9813 | spin_unlock(&cache->lock); |
9814 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9815 | } |
9816 | ||
ba1bf481 | 9817 | /* |
52042d8e | 9818 | * Checks to see if it's even possible to relocate this block group. |
ba1bf481 JB |
9819 | * |
9820 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9821 | * ok to go ahead and try. | |
9822 | */ | |
6bccf3ab | 9823 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9824 | { |
ba1bf481 JB |
9825 | struct btrfs_block_group_cache *block_group; |
9826 | struct btrfs_space_info *space_info; | |
0b246afa | 9827 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9828 | struct btrfs_device *device; |
cdcb725c | 9829 | u64 min_free; |
6719db6a JB |
9830 | u64 dev_min = 1; |
9831 | u64 dev_nr = 0; | |
4a5e98f5 | 9832 | u64 target; |
0305bc27 | 9833 | int debug; |
cdcb725c | 9834 | int index; |
ba1bf481 JB |
9835 | int full = 0; |
9836 | int ret = 0; | |
1a40e23b | 9837 | |
0b246afa | 9838 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9839 | |
0b246afa | 9840 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9841 | |
ba1bf481 | 9842 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9843 | if (!block_group) { |
9844 | if (debug) | |
0b246afa | 9845 | btrfs_warn(fs_info, |
0305bc27 QW |
9846 | "can't find block group for bytenr %llu", |
9847 | bytenr); | |
ba1bf481 | 9848 | return -1; |
0305bc27 | 9849 | } |
1a40e23b | 9850 | |
cdcb725c | 9851 | min_free = btrfs_block_group_used(&block_group->item); |
9852 | ||
ba1bf481 | 9853 | /* no bytes used, we're good */ |
cdcb725c | 9854 | if (!min_free) |
1a40e23b ZY |
9855 | goto out; |
9856 | ||
ba1bf481 JB |
9857 | space_info = block_group->space_info; |
9858 | spin_lock(&space_info->lock); | |
17d217fe | 9859 | |
ba1bf481 | 9860 | full = space_info->full; |
17d217fe | 9861 | |
ba1bf481 JB |
9862 | /* |
9863 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9864 | * relocate it unless we're able to allocate a new chunk below. |
9865 | * | |
9866 | * Otherwise, we need to make sure we have room in the space to handle | |
9867 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9868 | */ |
7ce618db | 9869 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9870 | (btrfs_space_info_used(space_info, false) + min_free < |
9871 | space_info->total_bytes)) { | |
ba1bf481 JB |
9872 | spin_unlock(&space_info->lock); |
9873 | goto out; | |
17d217fe | 9874 | } |
ba1bf481 | 9875 | spin_unlock(&space_info->lock); |
ea8c2819 | 9876 | |
ba1bf481 JB |
9877 | /* |
9878 | * ok we don't have enough space, but maybe we have free space on our | |
9879 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9880 | * alloc devices and guess if we have enough space. if this block |
9881 | * group is going to be restriped, run checks against the target | |
9882 | * profile instead of the current one. | |
ba1bf481 JB |
9883 | */ |
9884 | ret = -1; | |
ea8c2819 | 9885 | |
cdcb725c | 9886 | /* |
9887 | * index: | |
9888 | * 0: raid10 | |
9889 | * 1: raid1 | |
9890 | * 2: dup | |
9891 | * 3: raid0 | |
9892 | * 4: single | |
9893 | */ | |
0b246afa | 9894 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9895 | if (target) { |
3e72ee88 | 9896 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9897 | } else { |
9898 | /* | |
9899 | * this is just a balance, so if we were marked as full | |
9900 | * we know there is no space for a new chunk | |
9901 | */ | |
0305bc27 QW |
9902 | if (full) { |
9903 | if (debug) | |
0b246afa JM |
9904 | btrfs_warn(fs_info, |
9905 | "no space to alloc new chunk for block group %llu", | |
9906 | block_group->key.objectid); | |
4a5e98f5 | 9907 | goto out; |
0305bc27 | 9908 | } |
4a5e98f5 | 9909 | |
3e72ee88 | 9910 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
9911 | } |
9912 | ||
e6ec716f | 9913 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9914 | dev_min = 4; |
6719db6a JB |
9915 | /* Divide by 2 */ |
9916 | min_free >>= 1; | |
e6ec716f | 9917 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9918 | dev_min = 2; |
e6ec716f | 9919 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9920 | /* Multiply by 2 */ |
9921 | min_free <<= 1; | |
e6ec716f | 9922 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9923 | dev_min = fs_devices->rw_devices; |
47c5713f | 9924 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9925 | } |
9926 | ||
0b246afa | 9927 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9928 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9929 | u64 dev_offset; |
56bec294 | 9930 | |
ba1bf481 JB |
9931 | /* |
9932 | * check to make sure we can actually find a chunk with enough | |
9933 | * space to fit our block group in. | |
9934 | */ | |
63a212ab | 9935 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 9936 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
60dfdf25 | 9937 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 9938 | &dev_offset, NULL); |
ba1bf481 | 9939 | if (!ret) |
cdcb725c | 9940 | dev_nr++; |
9941 | ||
9942 | if (dev_nr >= dev_min) | |
73e48b27 | 9943 | break; |
cdcb725c | 9944 | |
ba1bf481 | 9945 | ret = -1; |
725c8463 | 9946 | } |
edbd8d4e | 9947 | } |
0305bc27 | 9948 | if (debug && ret == -1) |
0b246afa JM |
9949 | btrfs_warn(fs_info, |
9950 | "no space to allocate a new chunk for block group %llu", | |
9951 | block_group->key.objectid); | |
9952 | mutex_unlock(&fs_info->chunk_mutex); | |
edbd8d4e | 9953 | out: |
ba1bf481 | 9954 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9955 | return ret; |
9956 | } | |
9957 | ||
6bccf3ab JM |
9958 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9959 | struct btrfs_path *path, | |
9960 | struct btrfs_key *key) | |
0b86a832 | 9961 | { |
6bccf3ab | 9962 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9963 | int ret = 0; |
0b86a832 CM |
9964 | struct btrfs_key found_key; |
9965 | struct extent_buffer *leaf; | |
514c7dca QW |
9966 | struct btrfs_block_group_item bg; |
9967 | u64 flags; | |
0b86a832 | 9968 | int slot; |
edbd8d4e | 9969 | |
0b86a832 CM |
9970 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9971 | if (ret < 0) | |
925baedd CM |
9972 | goto out; |
9973 | ||
d397712b | 9974 | while (1) { |
0b86a832 | 9975 | slot = path->slots[0]; |
edbd8d4e | 9976 | leaf = path->nodes[0]; |
0b86a832 CM |
9977 | if (slot >= btrfs_header_nritems(leaf)) { |
9978 | ret = btrfs_next_leaf(root, path); | |
9979 | if (ret == 0) | |
9980 | continue; | |
9981 | if (ret < 0) | |
925baedd | 9982 | goto out; |
0b86a832 | 9983 | break; |
edbd8d4e | 9984 | } |
0b86a832 | 9985 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9986 | |
0b86a832 | 9987 | if (found_key.objectid >= key->objectid && |
925baedd | 9988 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9989 | struct extent_map_tree *em_tree; |
9990 | struct extent_map *em; | |
9991 | ||
9992 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9993 | read_lock(&em_tree->lock); | |
9994 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9995 | found_key.offset); | |
9996 | read_unlock(&em_tree->lock); | |
9997 | if (!em) { | |
0b246afa | 9998 | btrfs_err(fs_info, |
6fb37b75 LB |
9999 | "logical %llu len %llu found bg but no related chunk", |
10000 | found_key.objectid, found_key.offset); | |
10001 | ret = -ENOENT; | |
514c7dca QW |
10002 | } else if (em->start != found_key.objectid || |
10003 | em->len != found_key.offset) { | |
10004 | btrfs_err(fs_info, | |
10005 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
10006 | found_key.objectid, found_key.offset, | |
10007 | em->start, em->len); | |
10008 | ret = -EUCLEAN; | |
6fb37b75 | 10009 | } else { |
514c7dca QW |
10010 | read_extent_buffer(leaf, &bg, |
10011 | btrfs_item_ptr_offset(leaf, slot), | |
10012 | sizeof(bg)); | |
10013 | flags = btrfs_block_group_flags(&bg) & | |
10014 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
10015 | ||
10016 | if (flags != (em->map_lookup->type & | |
10017 | BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
10018 | btrfs_err(fs_info, | |
10019 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
10020 | found_key.objectid, | |
10021 | found_key.offset, flags, | |
10022 | (BTRFS_BLOCK_GROUP_TYPE_MASK & | |
10023 | em->map_lookup->type)); | |
10024 | ret = -EUCLEAN; | |
10025 | } else { | |
10026 | ret = 0; | |
10027 | } | |
6fb37b75 | 10028 | } |
187ee58c | 10029 | free_extent_map(em); |
925baedd CM |
10030 | goto out; |
10031 | } | |
0b86a832 | 10032 | path->slots[0]++; |
edbd8d4e | 10033 | } |
925baedd | 10034 | out: |
0b86a832 | 10035 | return ret; |
edbd8d4e CM |
10036 | } |
10037 | ||
0af3d00b JB |
10038 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
10039 | { | |
10040 | struct btrfs_block_group_cache *block_group; | |
10041 | u64 last = 0; | |
10042 | ||
10043 | while (1) { | |
10044 | struct inode *inode; | |
10045 | ||
10046 | block_group = btrfs_lookup_first_block_group(info, last); | |
10047 | while (block_group) { | |
3aa7c7a3 | 10048 | wait_block_group_cache_done(block_group); |
0af3d00b JB |
10049 | spin_lock(&block_group->lock); |
10050 | if (block_group->iref) | |
10051 | break; | |
10052 | spin_unlock(&block_group->lock); | |
f87b7eb8 | 10053 | block_group = next_block_group(block_group); |
0af3d00b JB |
10054 | } |
10055 | if (!block_group) { | |
10056 | if (last == 0) | |
10057 | break; | |
10058 | last = 0; | |
10059 | continue; | |
10060 | } | |
10061 | ||
10062 | inode = block_group->inode; | |
10063 | block_group->iref = 0; | |
10064 | block_group->inode = NULL; | |
10065 | spin_unlock(&block_group->lock); | |
f3bca802 | 10066 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
10067 | iput(inode); |
10068 | last = block_group->key.objectid + block_group->key.offset; | |
10069 | btrfs_put_block_group(block_group); | |
10070 | } | |
10071 | } | |
10072 | ||
5cdd7db6 FM |
10073 | /* |
10074 | * Must be called only after stopping all workers, since we could have block | |
10075 | * group caching kthreads running, and therefore they could race with us if we | |
10076 | * freed the block groups before stopping them. | |
10077 | */ | |
1a40e23b ZY |
10078 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
10079 | { | |
10080 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 10081 | struct btrfs_space_info *space_info; |
11833d66 | 10082 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
10083 | struct rb_node *n; |
10084 | ||
9e351cc8 | 10085 | down_write(&info->commit_root_sem); |
11833d66 YZ |
10086 | while (!list_empty(&info->caching_block_groups)) { |
10087 | caching_ctl = list_entry(info->caching_block_groups.next, | |
10088 | struct btrfs_caching_control, list); | |
10089 | list_del(&caching_ctl->list); | |
10090 | put_caching_control(caching_ctl); | |
10091 | } | |
9e351cc8 | 10092 | up_write(&info->commit_root_sem); |
11833d66 | 10093 | |
47ab2a6c JB |
10094 | spin_lock(&info->unused_bgs_lock); |
10095 | while (!list_empty(&info->unused_bgs)) { | |
10096 | block_group = list_first_entry(&info->unused_bgs, | |
10097 | struct btrfs_block_group_cache, | |
10098 | bg_list); | |
10099 | list_del_init(&block_group->bg_list); | |
10100 | btrfs_put_block_group(block_group); | |
10101 | } | |
10102 | spin_unlock(&info->unused_bgs_lock); | |
10103 | ||
1a40e23b ZY |
10104 | spin_lock(&info->block_group_cache_lock); |
10105 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
10106 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
10107 | cache_node); | |
1a40e23b ZY |
10108 | rb_erase(&block_group->cache_node, |
10109 | &info->block_group_cache_tree); | |
01eacb27 | 10110 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
10111 | spin_unlock(&info->block_group_cache_lock); |
10112 | ||
80eb234a | 10113 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 10114 | list_del(&block_group->list); |
80eb234a | 10115 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 10116 | |
3c14874a JB |
10117 | /* |
10118 | * We haven't cached this block group, which means we could | |
10119 | * possibly have excluded extents on this block group. | |
10120 | */ | |
36cce922 JB |
10121 | if (block_group->cached == BTRFS_CACHE_NO || |
10122 | block_group->cached == BTRFS_CACHE_ERROR) | |
9e715da8 | 10123 | free_excluded_extents(block_group); |
3c14874a | 10124 | |
817d52f8 | 10125 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 10126 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
10127 | ASSERT(list_empty(&block_group->dirty_list)); |
10128 | ASSERT(list_empty(&block_group->io_list)); | |
10129 | ASSERT(list_empty(&block_group->bg_list)); | |
10130 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 10131 | btrfs_put_block_group(block_group); |
d899e052 YZ |
10132 | |
10133 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
10134 | } |
10135 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
10136 | |
10137 | /* now that all the block groups are freed, go through and | |
10138 | * free all the space_info structs. This is only called during | |
10139 | * the final stages of unmount, and so we know nobody is | |
10140 | * using them. We call synchronize_rcu() once before we start, | |
10141 | * just to be on the safe side. | |
10142 | */ | |
10143 | synchronize_rcu(); | |
10144 | ||
8929ecfa YZ |
10145 | release_global_block_rsv(info); |
10146 | ||
67871254 | 10147 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
10148 | int i; |
10149 | ||
4184ea7f CM |
10150 | space_info = list_entry(info->space_info.next, |
10151 | struct btrfs_space_info, | |
10152 | list); | |
d555b6c3 JB |
10153 | |
10154 | /* | |
10155 | * Do not hide this behind enospc_debug, this is actually | |
10156 | * important and indicates a real bug if this happens. | |
10157 | */ | |
10158 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 10159 | space_info->bytes_reserved > 0 || |
d555b6c3 | 10160 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 10161 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 10162 | list_del(&space_info->list); |
6ab0a202 JM |
10163 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
10164 | struct kobject *kobj; | |
c1895442 JM |
10165 | kobj = space_info->block_group_kobjs[i]; |
10166 | space_info->block_group_kobjs[i] = NULL; | |
10167 | if (kobj) { | |
6ab0a202 JM |
10168 | kobject_del(kobj); |
10169 | kobject_put(kobj); | |
10170 | } | |
10171 | } | |
10172 | kobject_del(&space_info->kobj); | |
10173 | kobject_put(&space_info->kobj); | |
4184ea7f | 10174 | } |
1a40e23b ZY |
10175 | return 0; |
10176 | } | |
10177 | ||
75cb379d JM |
10178 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
10179 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) | |
10180 | { | |
10181 | struct btrfs_space_info *space_info; | |
10182 | struct raid_kobject *rkobj; | |
10183 | LIST_HEAD(list); | |
10184 | int index; | |
10185 | int ret = 0; | |
10186 | ||
10187 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
10188 | list_splice_init(&fs_info->pending_raid_kobjs, &list); | |
10189 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
10190 | ||
10191 | list_for_each_entry(rkobj, &list, list) { | |
10192 | space_info = __find_space_info(fs_info, rkobj->flags); | |
10193 | index = btrfs_bg_flags_to_raid_index(rkobj->flags); | |
10194 | ||
10195 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
10196 | "%s", get_raid_name(index)); | |
10197 | if (ret) { | |
10198 | kobject_put(&rkobj->kobj); | |
10199 | break; | |
10200 | } | |
10201 | } | |
10202 | if (ret) | |
10203 | btrfs_warn(fs_info, | |
10204 | "failed to add kobject for block cache, ignoring"); | |
10205 | } | |
10206 | ||
c434d21c | 10207 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 10208 | { |
c434d21c | 10209 | struct btrfs_space_info *space_info = cache->space_info; |
75cb379d | 10210 | struct btrfs_fs_info *fs_info = cache->fs_info; |
3e72ee88 | 10211 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 10212 | bool first = false; |
b742bb82 YZ |
10213 | |
10214 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
10215 | if (list_empty(&space_info->block_groups[index])) |
10216 | first = true; | |
10217 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
10218 | up_write(&space_info->groups_sem); | |
10219 | ||
10220 | if (first) { | |
75cb379d JM |
10221 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
10222 | if (!rkobj) { | |
10223 | btrfs_warn(cache->fs_info, | |
10224 | "couldn't alloc memory for raid level kobject"); | |
10225 | return; | |
6ab0a202 | 10226 | } |
75cb379d JM |
10227 | rkobj->flags = cache->flags; |
10228 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
10229 | ||
10230 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
10231 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); | |
10232 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
c1895442 | 10233 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 10234 | } |
b742bb82 YZ |
10235 | } |
10236 | ||
920e4a58 | 10237 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
10238 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
10239 | u64 start, u64 size) | |
920e4a58 MX |
10240 | { |
10241 | struct btrfs_block_group_cache *cache; | |
10242 | ||
10243 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
10244 | if (!cache) | |
10245 | return NULL; | |
10246 | ||
10247 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
10248 | GFP_NOFS); | |
10249 | if (!cache->free_space_ctl) { | |
10250 | kfree(cache); | |
10251 | return NULL; | |
10252 | } | |
10253 | ||
10254 | cache->key.objectid = start; | |
10255 | cache->key.offset = size; | |
10256 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
10257 | ||
0b246afa | 10258 | cache->fs_info = fs_info; |
e4ff5fb5 | 10259 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
10260 | set_free_space_tree_thresholds(cache); |
10261 | ||
920e4a58 MX |
10262 | atomic_set(&cache->count, 1); |
10263 | spin_lock_init(&cache->lock); | |
e570fd27 | 10264 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
10265 | INIT_LIST_HEAD(&cache->list); |
10266 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 10267 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 10268 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 10269 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 10270 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 10271 | btrfs_init_free_space_ctl(cache); |
04216820 | 10272 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 10273 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 10274 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
10275 | |
10276 | return cache; | |
10277 | } | |
10278 | ||
7ef49515 QW |
10279 | |
10280 | /* | |
10281 | * Iterate all chunks and verify that each of them has the corresponding block | |
10282 | * group | |
10283 | */ | |
10284 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
10285 | { | |
10286 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
10287 | struct extent_map *em; | |
10288 | struct btrfs_block_group_cache *bg; | |
10289 | u64 start = 0; | |
10290 | int ret = 0; | |
10291 | ||
10292 | while (1) { | |
10293 | read_lock(&map_tree->map_tree.lock); | |
10294 | /* | |
10295 | * lookup_extent_mapping will return the first extent map | |
10296 | * intersecting the range, so setting @len to 1 is enough to | |
10297 | * get the first chunk. | |
10298 | */ | |
10299 | em = lookup_extent_mapping(&map_tree->map_tree, start, 1); | |
10300 | read_unlock(&map_tree->map_tree.lock); | |
10301 | if (!em) | |
10302 | break; | |
10303 | ||
10304 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
10305 | if (!bg) { | |
10306 | btrfs_err(fs_info, | |
10307 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
10308 | em->start, em->len); | |
10309 | ret = -EUCLEAN; | |
10310 | free_extent_map(em); | |
10311 | break; | |
10312 | } | |
10313 | if (bg->key.objectid != em->start || | |
10314 | bg->key.offset != em->len || | |
10315 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != | |
10316 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
10317 | btrfs_err(fs_info, | |
10318 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
10319 | em->start, em->len, | |
10320 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
10321 | bg->key.objectid, bg->key.offset, | |
10322 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); | |
10323 | ret = -EUCLEAN; | |
10324 | free_extent_map(em); | |
10325 | btrfs_put_block_group(bg); | |
10326 | break; | |
10327 | } | |
10328 | start = em->start + em->len; | |
10329 | free_extent_map(em); | |
10330 | btrfs_put_block_group(bg); | |
10331 | } | |
10332 | return ret; | |
10333 | } | |
10334 | ||
5b4aacef | 10335 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
10336 | { |
10337 | struct btrfs_path *path; | |
10338 | int ret; | |
9078a3e1 | 10339 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 10340 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
10341 | struct btrfs_key key; |
10342 | struct btrfs_key found_key; | |
5f39d397 | 10343 | struct extent_buffer *leaf; |
0af3d00b JB |
10344 | int need_clear = 0; |
10345 | u64 cache_gen; | |
49303381 LB |
10346 | u64 feature; |
10347 | int mixed; | |
10348 | ||
10349 | feature = btrfs_super_incompat_flags(info->super_copy); | |
10350 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 10351 | |
9078a3e1 | 10352 | key.objectid = 0; |
0b86a832 | 10353 | key.offset = 0; |
962a298f | 10354 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
10355 | path = btrfs_alloc_path(); |
10356 | if (!path) | |
10357 | return -ENOMEM; | |
e4058b54 | 10358 | path->reada = READA_FORWARD; |
9078a3e1 | 10359 | |
0b246afa JM |
10360 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
10361 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
10362 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 10363 | need_clear = 1; |
0b246afa | 10364 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 10365 | need_clear = 1; |
0af3d00b | 10366 | |
d397712b | 10367 | while (1) { |
6bccf3ab | 10368 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
10369 | if (ret > 0) |
10370 | break; | |
0b86a832 CM |
10371 | if (ret != 0) |
10372 | goto error; | |
920e4a58 | 10373 | |
5f39d397 CM |
10374 | leaf = path->nodes[0]; |
10375 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10376 | |
2ff7e61e | 10377 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10378 | found_key.offset); |
9078a3e1 | 10379 | if (!cache) { |
0b86a832 | 10380 | ret = -ENOMEM; |
f0486c68 | 10381 | goto error; |
9078a3e1 | 10382 | } |
96303081 | 10383 | |
cf7c1ef6 LB |
10384 | if (need_clear) { |
10385 | /* | |
10386 | * When we mount with old space cache, we need to | |
10387 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10388 | * | |
10389 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10390 | * truncate the old free space cache inode and | |
10391 | * setup a new one. | |
10392 | * b) Setting 'dirty flag' makes sure that we flush | |
10393 | * the new space cache info onto disk. | |
10394 | */ | |
0b246afa | 10395 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10396 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10397 | } |
0af3d00b | 10398 | |
5f39d397 CM |
10399 | read_extent_buffer(leaf, &cache->item, |
10400 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10401 | sizeof(cache->item)); | |
920e4a58 | 10402 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10403 | if (!mixed && |
10404 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10405 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10406 | btrfs_err(info, | |
10407 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10408 | cache->key.objectid); | |
10409 | ret = -EINVAL; | |
10410 | goto error; | |
10411 | } | |
0b86a832 | 10412 | |
9078a3e1 | 10413 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10414 | btrfs_release_path(path); |
34d52cb6 | 10415 | |
3c14874a JB |
10416 | /* |
10417 | * We need to exclude the super stripes now so that the space | |
10418 | * info has super bytes accounted for, otherwise we'll think | |
10419 | * we have more space than we actually do. | |
10420 | */ | |
3c4da657 | 10421 | ret = exclude_super_stripes(cache); |
835d974f JB |
10422 | if (ret) { |
10423 | /* | |
10424 | * We may have excluded something, so call this just in | |
10425 | * case. | |
10426 | */ | |
9e715da8 | 10427 | free_excluded_extents(cache); |
920e4a58 | 10428 | btrfs_put_block_group(cache); |
835d974f JB |
10429 | goto error; |
10430 | } | |
3c14874a | 10431 | |
817d52f8 JB |
10432 | /* |
10433 | * check for two cases, either we are full, and therefore | |
10434 | * don't need to bother with the caching work since we won't | |
10435 | * find any space, or we are empty, and we can just add all | |
52042d8e | 10436 | * the space in and be done with it. This saves us _a_lot_ of |
817d52f8 JB |
10437 | * time, particularly in the full case. |
10438 | */ | |
10439 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10440 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10441 | cache->cached = BTRFS_CACHE_FINISHED; |
9e715da8 | 10442 | free_excluded_extents(cache); |
817d52f8 | 10443 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10444 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10445 | cache->cached = BTRFS_CACHE_FINISHED; |
4457c1c7 | 10446 | add_new_free_space(cache, found_key.objectid, |
817d52f8 JB |
10447 | found_key.objectid + |
10448 | found_key.offset); | |
9e715da8 | 10449 | free_excluded_extents(cache); |
817d52f8 | 10450 | } |
96b5179d | 10451 | |
0b246afa | 10452 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10453 | if (ret) { |
10454 | btrfs_remove_free_space_cache(cache); | |
10455 | btrfs_put_block_group(cache); | |
10456 | goto error; | |
10457 | } | |
10458 | ||
0b246afa | 10459 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10460 | update_space_info(info, cache->flags, found_key.offset, |
10461 | btrfs_block_group_used(&cache->item), | |
10462 | cache->bytes_super, &space_info); | |
8c579fe7 | 10463 | |
6324fbf3 | 10464 | cache->space_info = space_info; |
1b2da372 | 10465 | |
c434d21c | 10466 | link_block_group(cache); |
0f9dd46c | 10467 | |
0b246afa | 10468 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10469 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10470 | inc_block_group_ro(cache, 1); |
47ab2a6c | 10471 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
031f24da QW |
10472 | ASSERT(list_empty(&cache->bg_list)); |
10473 | btrfs_mark_bg_unused(cache); | |
47ab2a6c | 10474 | } |
9078a3e1 | 10475 | } |
b742bb82 | 10476 | |
0b246afa | 10477 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10478 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10479 | (BTRFS_BLOCK_GROUP_RAID10 | |
10480 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10481 | BTRFS_BLOCK_GROUP_RAID5 | |
10482 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10483 | BTRFS_BLOCK_GROUP_DUP))) |
10484 | continue; | |
10485 | /* | |
10486 | * avoid allocating from un-mirrored block group if there are | |
10487 | * mirrored block groups. | |
10488 | */ | |
1095cc0d | 10489 | list_for_each_entry(cache, |
10490 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10491 | list) | |
868f401a | 10492 | inc_block_group_ro(cache, 1); |
1095cc0d | 10493 | list_for_each_entry(cache, |
10494 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10495 | list) | |
868f401a | 10496 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10497 | } |
f0486c68 | 10498 | |
75cb379d | 10499 | btrfs_add_raid_kobjects(info); |
f0486c68 | 10500 | init_global_block_rsv(info); |
7ef49515 | 10501 | ret = check_chunk_block_group_mappings(info); |
0b86a832 | 10502 | error: |
9078a3e1 | 10503 | btrfs_free_path(path); |
0b86a832 | 10504 | return ret; |
9078a3e1 | 10505 | } |
6324fbf3 | 10506 | |
6c686b35 | 10507 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 10508 | { |
6c686b35 | 10509 | struct btrfs_fs_info *fs_info = trans->fs_info; |
545e3366 | 10510 | struct btrfs_block_group_cache *block_group; |
0b246afa | 10511 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10512 | struct btrfs_block_group_item item; |
10513 | struct btrfs_key key; | |
10514 | int ret = 0; | |
10515 | ||
5ce55557 FM |
10516 | if (!trans->can_flush_pending_bgs) |
10517 | return; | |
10518 | ||
545e3366 JB |
10519 | while (!list_empty(&trans->new_bgs)) { |
10520 | block_group = list_first_entry(&trans->new_bgs, | |
10521 | struct btrfs_block_group_cache, | |
10522 | bg_list); | |
ea658bad | 10523 | if (ret) |
c92f6be3 | 10524 | goto next; |
ea658bad JB |
10525 | |
10526 | spin_lock(&block_group->lock); | |
10527 | memcpy(&item, &block_group->item, sizeof(item)); | |
10528 | memcpy(&key, &block_group->key, sizeof(key)); | |
10529 | spin_unlock(&block_group->lock); | |
10530 | ||
10531 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10532 | sizeof(item)); | |
10533 | if (ret) | |
66642832 | 10534 | btrfs_abort_transaction(trans, ret); |
97aff912 | 10535 | ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset); |
6df9a95e | 10536 | if (ret) |
66642832 | 10537 | btrfs_abort_transaction(trans, ret); |
e4e0711c | 10538 | add_block_group_free_space(trans, block_group); |
1e144fb8 | 10539 | /* already aborted the transaction if it failed. */ |
c92f6be3 | 10540 | next: |
ba2c4d4e | 10541 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
c92f6be3 | 10542 | list_del_init(&block_group->bg_list); |
ea658bad | 10543 | } |
5ce55557 | 10544 | btrfs_trans_release_chunk_metadata(trans); |
ea658bad JB |
10545 | } |
10546 | ||
e7e02096 | 10547 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, |
0174484d | 10548 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10549 | { |
e7e02096 | 10550 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 10551 | struct btrfs_block_group_cache *cache; |
0b246afa | 10552 | int ret; |
6324fbf3 | 10553 | |
90787766 | 10554 | btrfs_set_log_full_commit(trans); |
e02119d5 | 10555 | |
2ff7e61e | 10556 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10557 | if (!cache) |
10558 | return -ENOMEM; | |
34d52cb6 | 10559 | |
6324fbf3 | 10560 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10561 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10562 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10563 | btrfs_set_block_group_flags(&cache->item, type); |
10564 | ||
920e4a58 | 10565 | cache->flags = type; |
11833d66 | 10566 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10567 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10568 | cache->needs_free_space = 1; |
3c4da657 | 10569 | ret = exclude_super_stripes(cache); |
835d974f JB |
10570 | if (ret) { |
10571 | /* | |
10572 | * We may have excluded something, so call this just in | |
10573 | * case. | |
10574 | */ | |
9e715da8 | 10575 | free_excluded_extents(cache); |
920e4a58 | 10576 | btrfs_put_block_group(cache); |
835d974f JB |
10577 | return ret; |
10578 | } | |
96303081 | 10579 | |
4457c1c7 | 10580 | add_new_free_space(cache, chunk_offset, chunk_offset + size); |
817d52f8 | 10581 | |
9e715da8 | 10582 | free_excluded_extents(cache); |
11833d66 | 10583 | |
d0bd4560 | 10584 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10585 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10586 | u64 new_bytes_used = size - bytes_used; |
10587 | ||
10588 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10589 | fragment_free_space(cache); |
d0bd4560 JB |
10590 | } |
10591 | #endif | |
2e6e5183 | 10592 | /* |
2be12ef7 NB |
10593 | * Ensure the corresponding space_info object is created and |
10594 | * assigned to our block group. We want our bg to be added to the rbtree | |
10595 | * with its ->space_info set. | |
2e6e5183 | 10596 | */ |
2be12ef7 | 10597 | cache->space_info = __find_space_info(fs_info, cache->flags); |
dc2d3005 | 10598 | ASSERT(cache->space_info); |
2e6e5183 | 10599 | |
0b246afa | 10600 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10601 | if (ret) { |
10602 | btrfs_remove_free_space_cache(cache); | |
10603 | btrfs_put_block_group(cache); | |
10604 | return ret; | |
10605 | } | |
10606 | ||
2e6e5183 FM |
10607 | /* |
10608 | * Now that our block group has its ->space_info set and is inserted in | |
10609 | * the rbtree, update the space info's counters. | |
10610 | */ | |
0b246afa | 10611 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10612 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10613 | cache->bytes_super, &cache->space_info); |
0b246afa | 10614 | update_global_block_rsv(fs_info); |
1b2da372 | 10615 | |
c434d21c | 10616 | link_block_group(cache); |
6324fbf3 | 10617 | |
47ab2a6c | 10618 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
ba2c4d4e JB |
10619 | trans->delayed_ref_updates++; |
10620 | btrfs_update_delayed_refs_rsv(trans); | |
6324fbf3 | 10621 | |
0b246afa | 10622 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10623 | return 0; |
10624 | } | |
1a40e23b | 10625 | |
10ea00f5 ID |
10626 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10627 | { | |
899c81ea ID |
10628 | u64 extra_flags = chunk_to_extended(flags) & |
10629 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10630 | |
de98ced9 | 10631 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10632 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10633 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10634 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10635 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10636 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10637 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10638 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10639 | } |
10640 | ||
1a40e23b | 10641 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
5a98ec01 | 10642 | u64 group_start, struct extent_map *em) |
1a40e23b | 10643 | { |
5a98ec01 | 10644 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 10645 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10646 | struct btrfs_path *path; |
10647 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10648 | struct btrfs_free_cluster *cluster; |
0b246afa | 10649 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10650 | struct btrfs_key key; |
0af3d00b | 10651 | struct inode *inode; |
c1895442 | 10652 | struct kobject *kobj = NULL; |
1a40e23b | 10653 | int ret; |
10ea00f5 | 10654 | int index; |
89a55897 | 10655 | int factor; |
4f69cb98 | 10656 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10657 | bool remove_em; |
ba2c4d4e | 10658 | bool remove_rsv = false; |
1a40e23b | 10659 | |
6bccf3ab | 10660 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10661 | BUG_ON(!block_group); |
c146afad | 10662 | BUG_ON(!block_group->ro); |
1a40e23b | 10663 | |
4ed0a7a3 | 10664 | trace_btrfs_remove_block_group(block_group); |
9f7c43c9 | 10665 | /* |
10666 | * Free the reserved super bytes from this block group before | |
10667 | * remove it. | |
10668 | */ | |
9e715da8 | 10669 | free_excluded_extents(block_group); |
fd708b81 JB |
10670 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10671 | block_group->key.offset); | |
9f7c43c9 | 10672 | |
1a40e23b | 10673 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 10674 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
46df06b8 | 10675 | factor = btrfs_bg_type_to_factor(block_group->flags); |
1a40e23b | 10676 | |
44fb5511 | 10677 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10678 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10679 | spin_lock(&cluster->refill_lock); |
10680 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10681 | spin_unlock(&cluster->refill_lock); | |
10682 | ||
10683 | /* | |
10684 | * make sure this block group isn't part of a metadata | |
10685 | * allocation cluster | |
10686 | */ | |
0b246afa | 10687 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10688 | spin_lock(&cluster->refill_lock); |
10689 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10690 | spin_unlock(&cluster->refill_lock); | |
10691 | ||
1a40e23b | 10692 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10693 | if (!path) { |
10694 | ret = -ENOMEM; | |
10695 | goto out; | |
10696 | } | |
1a40e23b | 10697 | |
1bbc621e CM |
10698 | /* |
10699 | * get the inode first so any iput calls done for the io_list | |
10700 | * aren't the final iput (no unlinks allowed now) | |
10701 | */ | |
7949f339 | 10702 | inode = lookup_free_space_inode(block_group, path); |
1bbc621e CM |
10703 | |
10704 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10705 | /* | |
52042d8e | 10706 | * Make sure our free space cache IO is done before removing the |
1bbc621e CM |
10707 | * free space inode |
10708 | */ | |
10709 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10710 | if (!list_empty(&block_group->io_list)) { | |
10711 | list_del_init(&block_group->io_list); | |
10712 | ||
10713 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10714 | ||
10715 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10716 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10717 | btrfs_put_block_group(block_group); |
10718 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10719 | } | |
10720 | ||
10721 | if (!list_empty(&block_group->dirty_list)) { | |
10722 | list_del_init(&block_group->dirty_list); | |
ba2c4d4e | 10723 | remove_rsv = true; |
1bbc621e CM |
10724 | btrfs_put_block_group(block_group); |
10725 | } | |
10726 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10727 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10728 | ||
0af3d00b | 10729 | if (!IS_ERR(inode)) { |
73f2e545 | 10730 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10731 | if (ret) { |
10732 | btrfs_add_delayed_iput(inode); | |
10733 | goto out; | |
10734 | } | |
0af3d00b JB |
10735 | clear_nlink(inode); |
10736 | /* One for the block groups ref */ | |
10737 | spin_lock(&block_group->lock); | |
10738 | if (block_group->iref) { | |
10739 | block_group->iref = 0; | |
10740 | block_group->inode = NULL; | |
10741 | spin_unlock(&block_group->lock); | |
10742 | iput(inode); | |
10743 | } else { | |
10744 | spin_unlock(&block_group->lock); | |
10745 | } | |
10746 | /* One for our lookup ref */ | |
455757c3 | 10747 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10748 | } |
10749 | ||
10750 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10751 | key.offset = block_group->key.objectid; | |
10752 | key.type = 0; | |
10753 | ||
10754 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10755 | if (ret < 0) | |
10756 | goto out; | |
10757 | if (ret > 0) | |
b3b4aa74 | 10758 | btrfs_release_path(path); |
0af3d00b JB |
10759 | if (ret == 0) { |
10760 | ret = btrfs_del_item(trans, tree_root, path); | |
10761 | if (ret) | |
10762 | goto out; | |
b3b4aa74 | 10763 | btrfs_release_path(path); |
0af3d00b JB |
10764 | } |
10765 | ||
0b246afa | 10766 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10767 | rb_erase(&block_group->cache_node, |
0b246afa | 10768 | &fs_info->block_group_cache_tree); |
292cbd51 | 10769 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10770 | |
0b246afa JM |
10771 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10772 | fs_info->first_logical_byte = (u64)-1; | |
10773 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10774 | |
80eb234a | 10775 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10776 | /* |
10777 | * we must use list_del_init so people can check to see if they | |
10778 | * are still on the list after taking the semaphore | |
10779 | */ | |
10780 | list_del_init(&block_group->list); | |
6ab0a202 | 10781 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10782 | kobj = block_group->space_info->block_group_kobjs[index]; |
10783 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10784 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10785 | } |
80eb234a | 10786 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10787 | if (kobj) { |
10788 | kobject_del(kobj); | |
10789 | kobject_put(kobj); | |
10790 | } | |
1a40e23b | 10791 | |
4f69cb98 FM |
10792 | if (block_group->has_caching_ctl) |
10793 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10794 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10795 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10796 | if (block_group->has_caching_ctl) { |
0b246afa | 10797 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10798 | if (!caching_ctl) { |
10799 | struct btrfs_caching_control *ctl; | |
10800 | ||
10801 | list_for_each_entry(ctl, | |
0b246afa | 10802 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10803 | if (ctl->block_group == block_group) { |
10804 | caching_ctl = ctl; | |
1e4f4714 | 10805 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10806 | break; |
10807 | } | |
10808 | } | |
10809 | if (caching_ctl) | |
10810 | list_del_init(&caching_ctl->list); | |
0b246afa | 10811 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10812 | if (caching_ctl) { |
10813 | /* Once for the caching bgs list and once for us. */ | |
10814 | put_caching_control(caching_ctl); | |
10815 | put_caching_control(caching_ctl); | |
10816 | } | |
10817 | } | |
817d52f8 | 10818 | |
ce93ec54 | 10819 | spin_lock(&trans->transaction->dirty_bgs_lock); |
9a0ec83d NB |
10820 | WARN_ON(!list_empty(&block_group->dirty_list)); |
10821 | WARN_ON(!list_empty(&block_group->io_list)); | |
ce93ec54 | 10822 | spin_unlock(&trans->transaction->dirty_bgs_lock); |
9a0ec83d | 10823 | |
817d52f8 JB |
10824 | btrfs_remove_free_space_cache(block_group); |
10825 | ||
c146afad | 10826 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10827 | list_del_init(&block_group->ro_list); |
18d018ad | 10828 | |
0b246afa | 10829 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10830 | WARN_ON(block_group->space_info->total_bytes |
10831 | < block_group->key.offset); | |
10832 | WARN_ON(block_group->space_info->bytes_readonly | |
10833 | < block_group->key.offset); | |
10834 | WARN_ON(block_group->space_info->disk_total | |
10835 | < block_group->key.offset * factor); | |
10836 | } | |
c146afad YZ |
10837 | block_group->space_info->total_bytes -= block_group->key.offset; |
10838 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10839 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10840 | |
c146afad | 10841 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10842 | |
0af3d00b JB |
10843 | memcpy(&key, &block_group->key, sizeof(key)); |
10844 | ||
34441361 | 10845 | mutex_lock(&fs_info->chunk_mutex); |
04216820 FM |
10846 | spin_lock(&block_group->lock); |
10847 | block_group->removed = 1; | |
10848 | /* | |
10849 | * At this point trimming can't start on this block group, because we | |
10850 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10851 | * so no one can't find it anymore and even if someone already got this | |
10852 | * block group before we removed it from the rbtree, they have already | |
10853 | * incremented block_group->trimming - if they didn't, they won't find | |
10854 | * any free space entries because we already removed them all when we | |
10855 | * called btrfs_remove_free_space_cache(). | |
10856 | * | |
10857 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10858 | * to prevent the same logical address range and physical device space | |
10859 | * ranges from being reused for a new block group. This is because our | |
10860 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10861 | * completely transactionless, so while it is trimming a range the | |
10862 | * currently running transaction might finish and a new one start, | |
10863 | * allowing for new block groups to be created that can reuse the same | |
10864 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10865 | * |
10866 | * There may also be an implicit trim operation if the file system | |
10867 | * is mounted with -odiscard. The same protections must remain | |
10868 | * in place until the extents have been discarded completely when | |
10869 | * the transaction commit has completed. | |
04216820 FM |
10870 | */ |
10871 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
04216820 | 10872 | spin_unlock(&block_group->lock); |
04216820 FM |
10873 | |
10874 | if (remove_em) { | |
10875 | struct extent_map_tree *em_tree; | |
10876 | ||
0b246afa | 10877 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 FM |
10878 | write_lock(&em_tree->lock); |
10879 | remove_extent_mapping(em_tree, em); | |
10880 | write_unlock(&em_tree->lock); | |
10881 | /* once for the tree */ | |
10882 | free_extent_map(em); | |
10883 | } | |
10884 | ||
34441361 | 10885 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10886 | |
f3f72779 | 10887 | ret = remove_block_group_free_space(trans, block_group); |
1e144fb8 OS |
10888 | if (ret) |
10889 | goto out; | |
10890 | ||
fa9c0d79 CM |
10891 | btrfs_put_block_group(block_group); |
10892 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10893 | |
10894 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10895 | if (ret > 0) | |
10896 | ret = -EIO; | |
10897 | if (ret < 0) | |
10898 | goto out; | |
10899 | ||
10900 | ret = btrfs_del_item(trans, root, path); | |
10901 | out: | |
ba2c4d4e JB |
10902 | if (remove_rsv) |
10903 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1a40e23b ZY |
10904 | btrfs_free_path(path); |
10905 | return ret; | |
10906 | } | |
acce952b | 10907 | |
8eab77ff | 10908 | struct btrfs_trans_handle * |
7fd01182 FM |
10909 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10910 | const u64 chunk_offset) | |
8eab77ff | 10911 | { |
7fd01182 FM |
10912 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10913 | struct extent_map *em; | |
10914 | struct map_lookup *map; | |
10915 | unsigned int num_items; | |
10916 | ||
10917 | read_lock(&em_tree->lock); | |
10918 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10919 | read_unlock(&em_tree->lock); | |
10920 | ASSERT(em && em->start == chunk_offset); | |
10921 | ||
8eab77ff | 10922 | /* |
7fd01182 FM |
10923 | * We need to reserve 3 + N units from the metadata space info in order |
10924 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10925 | * btrfs_remove_block_group()), which are used for: | |
10926 | * | |
8eab77ff FM |
10927 | * 1 unit for adding the free space inode's orphan (located in the tree |
10928 | * of tree roots). | |
7fd01182 FM |
10929 | * 1 unit for deleting the block group item (located in the extent |
10930 | * tree). | |
10931 | * 1 unit for deleting the free space item (located in tree of tree | |
10932 | * roots). | |
10933 | * N units for deleting N device extent items corresponding to each | |
10934 | * stripe (located in the device tree). | |
10935 | * | |
10936 | * In order to remove a block group we also need to reserve units in the | |
10937 | * system space info in order to update the chunk tree (update one or | |
10938 | * more device items and remove one chunk item), but this is done at | |
10939 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10940 | */ |
95617d69 | 10941 | map = em->map_lookup; |
7fd01182 FM |
10942 | num_items = 3 + map->num_stripes; |
10943 | free_extent_map(em); | |
10944 | ||
8eab77ff | 10945 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10946 | num_items, 1); |
8eab77ff FM |
10947 | } |
10948 | ||
47ab2a6c JB |
10949 | /* |
10950 | * Process the unused_bgs list and remove any that don't have any allocated | |
10951 | * space inside of them. | |
10952 | */ | |
10953 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10954 | { | |
10955 | struct btrfs_block_group_cache *block_group; | |
10956 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10957 | struct btrfs_trans_handle *trans; |
10958 | int ret = 0; | |
10959 | ||
afcdd129 | 10960 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10961 | return; |
10962 | ||
10963 | spin_lock(&fs_info->unused_bgs_lock); | |
10964 | while (!list_empty(&fs_info->unused_bgs)) { | |
10965 | u64 start, end; | |
e33e17ee | 10966 | int trimming; |
47ab2a6c JB |
10967 | |
10968 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10969 | struct btrfs_block_group_cache, | |
10970 | bg_list); | |
47ab2a6c | 10971 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10972 | |
10973 | space_info = block_group->space_info; | |
10974 | ||
47ab2a6c JB |
10975 | if (ret || btrfs_mixed_space_info(space_info)) { |
10976 | btrfs_put_block_group(block_group); | |
10977 | continue; | |
10978 | } | |
10979 | spin_unlock(&fs_info->unused_bgs_lock); | |
10980 | ||
d5f2e33b | 10981 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10982 | |
47ab2a6c JB |
10983 | /* Don't want to race with allocators so take the groups_sem */ |
10984 | down_write(&space_info->groups_sem); | |
10985 | spin_lock(&block_group->lock); | |
43794446 | 10986 | if (block_group->reserved || block_group->pinned || |
47ab2a6c | 10987 | btrfs_block_group_used(&block_group->item) || |
19c4d2f9 | 10988 | block_group->ro || |
aefbe9a6 | 10989 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10990 | /* |
10991 | * We want to bail if we made new allocations or have | |
10992 | * outstanding allocations in this block group. We do | |
10993 | * the ro check in case balance is currently acting on | |
10994 | * this block group. | |
10995 | */ | |
4ed0a7a3 | 10996 | trace_btrfs_skip_unused_block_group(block_group); |
47ab2a6c JB |
10997 | spin_unlock(&block_group->lock); |
10998 | up_write(&space_info->groups_sem); | |
10999 | goto next; | |
11000 | } | |
11001 | spin_unlock(&block_group->lock); | |
11002 | ||
11003 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 11004 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
11005 | up_write(&space_info->groups_sem); |
11006 | if (ret < 0) { | |
11007 | ret = 0; | |
11008 | goto next; | |
11009 | } | |
11010 | ||
11011 | /* | |
11012 | * Want to do this before we do anything else so we can recover | |
11013 | * properly if we fail to join the transaction. | |
11014 | */ | |
7fd01182 FM |
11015 | trans = btrfs_start_trans_remove_block_group(fs_info, |
11016 | block_group->key.objectid); | |
47ab2a6c | 11017 | if (IS_ERR(trans)) { |
2ff7e61e | 11018 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
11019 | ret = PTR_ERR(trans); |
11020 | goto next; | |
11021 | } | |
11022 | ||
11023 | /* | |
11024 | * We could have pending pinned extents for this block group, | |
11025 | * just delete them, we don't care about them anymore. | |
11026 | */ | |
11027 | start = block_group->key.objectid; | |
11028 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
11029 | /* |
11030 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
11031 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
11032 | * another task might be running finish_extent_commit() for the | |
11033 | * previous transaction N - 1, and have seen a range belonging | |
11034 | * to the block group in freed_extents[] before we were able to | |
11035 | * clear the whole block group range from freed_extents[]. This | |
11036 | * means that task can lookup for the block group after we | |
11037 | * unpinned it from freed_extents[] and removed it, leading to | |
11038 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
11039 | */ | |
11040 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 11041 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 11042 | EXTENT_DIRTY); |
758eb51e | 11043 | if (ret) { |
d4b450cd | 11044 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 11045 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
11046 | goto end_trans; |
11047 | } | |
11048 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 11049 | EXTENT_DIRTY); |
758eb51e | 11050 | if (ret) { |
d4b450cd | 11051 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 11052 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
11053 | goto end_trans; |
11054 | } | |
d4b450cd | 11055 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
11056 | |
11057 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
11058 | spin_lock(&space_info->lock); |
11059 | spin_lock(&block_group->lock); | |
11060 | ||
e2907c1a | 11061 | update_bytes_pinned(space_info, -block_group->pinned); |
c30666d4 | 11062 | space_info->bytes_readonly += block_group->pinned; |
dec59fa3 EL |
11063 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
11064 | -block_group->pinned, | |
11065 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
47ab2a6c JB |
11066 | block_group->pinned = 0; |
11067 | ||
c30666d4 ZL |
11068 | spin_unlock(&block_group->lock); |
11069 | spin_unlock(&space_info->lock); | |
11070 | ||
e33e17ee | 11071 | /* DISCARD can flip during remount */ |
0b246afa | 11072 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
11073 | |
11074 | /* Implicit trim during transaction commit. */ | |
11075 | if (trimming) | |
11076 | btrfs_get_block_group_trimming(block_group); | |
11077 | ||
47ab2a6c JB |
11078 | /* |
11079 | * Btrfs_remove_chunk will abort the transaction if things go | |
11080 | * horribly wrong. | |
11081 | */ | |
97aff912 | 11082 | ret = btrfs_remove_chunk(trans, block_group->key.objectid); |
e33e17ee JM |
11083 | |
11084 | if (ret) { | |
11085 | if (trimming) | |
11086 | btrfs_put_block_group_trimming(block_group); | |
11087 | goto end_trans; | |
11088 | } | |
11089 | ||
11090 | /* | |
11091 | * If we're not mounted with -odiscard, we can just forget | |
11092 | * about this block group. Otherwise we'll need to wait | |
11093 | * until transaction commit to do the actual discard. | |
11094 | */ | |
11095 | if (trimming) { | |
348a0013 FM |
11096 | spin_lock(&fs_info->unused_bgs_lock); |
11097 | /* | |
11098 | * A concurrent scrub might have added us to the list | |
11099 | * fs_info->unused_bgs, so use a list_move operation | |
11100 | * to add the block group to the deleted_bgs list. | |
11101 | */ | |
e33e17ee JM |
11102 | list_move(&block_group->bg_list, |
11103 | &trans->transaction->deleted_bgs); | |
348a0013 | 11104 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
11105 | btrfs_get_block_group(block_group); |
11106 | } | |
758eb51e | 11107 | end_trans: |
3a45bb20 | 11108 | btrfs_end_transaction(trans); |
47ab2a6c | 11109 | next: |
d5f2e33b | 11110 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
11111 | btrfs_put_block_group(block_group); |
11112 | spin_lock(&fs_info->unused_bgs_lock); | |
11113 | } | |
11114 | spin_unlock(&fs_info->unused_bgs_lock); | |
11115 | } | |
11116 | ||
c59021f8 | 11117 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
11118 | { | |
1aba86d6 | 11119 | struct btrfs_super_block *disk_super; |
11120 | u64 features; | |
11121 | u64 flags; | |
11122 | int mixed = 0; | |
c59021f8 | 11123 | int ret; |
11124 | ||
6c41761f | 11125 | disk_super = fs_info->super_copy; |
1aba86d6 | 11126 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 11127 | return -EINVAL; |
c59021f8 | 11128 | |
1aba86d6 | 11129 | features = btrfs_super_incompat_flags(disk_super); |
11130 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
11131 | mixed = 1; | |
c59021f8 | 11132 | |
1aba86d6 | 11133 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
4ca61683 | 11134 | ret = create_space_info(fs_info, flags); |
c59021f8 | 11135 | if (ret) |
1aba86d6 | 11136 | goto out; |
c59021f8 | 11137 | |
1aba86d6 | 11138 | if (mixed) { |
11139 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 11140 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 11141 | } else { |
11142 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
4ca61683 | 11143 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 11144 | if (ret) |
11145 | goto out; | |
11146 | ||
11147 | flags = BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 11148 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 11149 | } |
11150 | out: | |
c59021f8 | 11151 | return ret; |
11152 | } | |
11153 | ||
2ff7e61e JM |
11154 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
11155 | u64 start, u64 end) | |
acce952b | 11156 | { |
2ff7e61e | 11157 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 11158 | } |
11159 | ||
499f377f JM |
11160 | /* |
11161 | * It used to be that old block groups would be left around forever. | |
11162 | * Iterating over them would be enough to trim unused space. Since we | |
11163 | * now automatically remove them, we also need to iterate over unallocated | |
11164 | * space. | |
11165 | * | |
11166 | * We don't want a transaction for this since the discard may take a | |
11167 | * substantial amount of time. We don't require that a transaction be | |
11168 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
11169 | * to ensure that we're not discarding chunks that were released or |
11170 | * allocated in the current transaction. | |
499f377f JM |
11171 | * |
11172 | * Holding the chunks lock will prevent other threads from allocating | |
11173 | * or releasing chunks, but it won't prevent a running transaction | |
11174 | * from committing and releasing the memory that the pending chunks | |
11175 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
11176 | * transaction and hold the commit root sem. We only need to hold |
11177 | * it while performing the free space search since we have already | |
11178 | * held back allocations. | |
499f377f JM |
11179 | */ |
11180 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
c2d1b3aa | 11181 | struct fstrim_range *range, u64 *trimmed) |
499f377f | 11182 | { |
929be17a | 11183 | u64 start, len = 0, end = 0; |
499f377f JM |
11184 | int ret; |
11185 | ||
929be17a | 11186 | start = max_t(u64, range->start, SZ_1M); |
499f377f JM |
11187 | *trimmed = 0; |
11188 | ||
0be88e36 JM |
11189 | /* Discard not supported = nothing to do. */ |
11190 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
11191 | return 0; | |
11192 | ||
52042d8e | 11193 | /* Not writable = nothing to do. */ |
ebbede42 | 11194 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
11195 | return 0; |
11196 | ||
11197 | /* No free space = nothing to do. */ | |
11198 | if (device->total_bytes <= device->bytes_used) | |
11199 | return 0; | |
11200 | ||
11201 | ret = 0; | |
11202 | ||
11203 | while (1) { | |
fb456252 | 11204 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
11205 | u64 bytes; |
11206 | ||
11207 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
11208 | if (ret) | |
fee7acc3 | 11209 | break; |
499f377f | 11210 | |
929be17a NB |
11211 | find_first_clear_extent_bit(&device->alloc_state, start, |
11212 | &start, &end, | |
11213 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
11214 | /* | |
11215 | * If find_first_clear_extent_bit find a range that spans the | |
11216 | * end of the device it will set end to -1, in this case it's up | |
11217 | * to the caller to trim the value to the size of the device. | |
11218 | */ | |
11219 | end = min(end, device->total_bytes - 1); | |
11220 | len = end - start + 1; | |
499f377f | 11221 | |
929be17a NB |
11222 | /* We didn't find any extents */ |
11223 | if (!len) { | |
499f377f | 11224 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 11225 | ret = 0; |
499f377f JM |
11226 | break; |
11227 | } | |
11228 | ||
929be17a NB |
11229 | /* Keep going until we satisfy minlen or reach end of space */ |
11230 | if (len < range->minlen) { | |
11231 | mutex_unlock(&fs_info->chunk_mutex); | |
11232 | start += len; | |
11233 | continue; | |
11234 | } | |
11235 | ||
c2d1b3aa NB |
11236 | /* If we are out of the passed range break */ |
11237 | if (start > range->start + range->len - 1) { | |
11238 | mutex_unlock(&fs_info->chunk_mutex); | |
c2d1b3aa NB |
11239 | break; |
11240 | } | |
11241 | ||
11242 | start = max(range->start, start); | |
11243 | len = min(range->len, len); | |
11244 | ||
929be17a NB |
11245 | ret = btrfs_issue_discard(device->bdev, start, len, |
11246 | &bytes); | |
11247 | if (!ret) | |
11248 | set_extent_bits(&device->alloc_state, start, | |
11249 | start + bytes - 1, | |
11250 | CHUNK_TRIMMED); | |
499f377f JM |
11251 | mutex_unlock(&fs_info->chunk_mutex); |
11252 | ||
11253 | if (ret) | |
11254 | break; | |
11255 | ||
11256 | start += len; | |
11257 | *trimmed += bytes; | |
11258 | ||
c2d1b3aa NB |
11259 | /* We've trimmed enough */ |
11260 | if (*trimmed >= range->len) | |
11261 | break; | |
11262 | ||
499f377f JM |
11263 | if (fatal_signal_pending(current)) { |
11264 | ret = -ERESTARTSYS; | |
11265 | break; | |
11266 | } | |
11267 | ||
11268 | cond_resched(); | |
11269 | } | |
11270 | ||
11271 | return ret; | |
11272 | } | |
11273 | ||
93bba24d QW |
11274 | /* |
11275 | * Trim the whole filesystem by: | |
11276 | * 1) trimming the free space in each block group | |
11277 | * 2) trimming the unallocated space on each device | |
11278 | * | |
11279 | * This will also continue trimming even if a block group or device encounters | |
11280 | * an error. The return value will be the last error, or 0 if nothing bad | |
11281 | * happens. | |
11282 | */ | |
2ff7e61e | 11283 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 11284 | { |
f7039b1d | 11285 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
11286 | struct btrfs_device *device; |
11287 | struct list_head *devices; | |
f7039b1d LD |
11288 | u64 group_trimmed; |
11289 | u64 start; | |
11290 | u64 end; | |
11291 | u64 trimmed = 0; | |
93bba24d QW |
11292 | u64 bg_failed = 0; |
11293 | u64 dev_failed = 0; | |
11294 | int bg_ret = 0; | |
11295 | int dev_ret = 0; | |
f7039b1d LD |
11296 | int ret = 0; |
11297 | ||
6ba9fc8e | 11298 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
f87b7eb8 | 11299 | for (; cache; cache = next_block_group(cache)) { |
f7039b1d LD |
11300 | if (cache->key.objectid >= (range->start + range->len)) { |
11301 | btrfs_put_block_group(cache); | |
11302 | break; | |
11303 | } | |
11304 | ||
11305 | start = max(range->start, cache->key.objectid); | |
11306 | end = min(range->start + range->len, | |
11307 | cache->key.objectid + cache->key.offset); | |
11308 | ||
11309 | if (end - start >= range->minlen) { | |
11310 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 11311 | ret = cache_block_group(cache, 0); |
1be41b78 | 11312 | if (ret) { |
93bba24d QW |
11313 | bg_failed++; |
11314 | bg_ret = ret; | |
11315 | continue; | |
1be41b78 JB |
11316 | } |
11317 | ret = wait_block_group_cache_done(cache); | |
11318 | if (ret) { | |
93bba24d QW |
11319 | bg_failed++; |
11320 | bg_ret = ret; | |
11321 | continue; | |
1be41b78 | 11322 | } |
f7039b1d LD |
11323 | } |
11324 | ret = btrfs_trim_block_group(cache, | |
11325 | &group_trimmed, | |
11326 | start, | |
11327 | end, | |
11328 | range->minlen); | |
11329 | ||
11330 | trimmed += group_trimmed; | |
11331 | if (ret) { | |
93bba24d QW |
11332 | bg_failed++; |
11333 | bg_ret = ret; | |
11334 | continue; | |
f7039b1d LD |
11335 | } |
11336 | } | |
f7039b1d LD |
11337 | } |
11338 | ||
93bba24d QW |
11339 | if (bg_failed) |
11340 | btrfs_warn(fs_info, | |
11341 | "failed to trim %llu block group(s), last error %d", | |
11342 | bg_failed, bg_ret); | |
0b246afa | 11343 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
11344 | devices = &fs_info->fs_devices->devices; |
11345 | list_for_each_entry(device, devices, dev_list) { | |
c2d1b3aa | 11346 | ret = btrfs_trim_free_extents(device, range, &group_trimmed); |
93bba24d QW |
11347 | if (ret) { |
11348 | dev_failed++; | |
11349 | dev_ret = ret; | |
499f377f | 11350 | break; |
93bba24d | 11351 | } |
499f377f JM |
11352 | |
11353 | trimmed += group_trimmed; | |
11354 | } | |
0b246afa | 11355 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 11356 | |
93bba24d QW |
11357 | if (dev_failed) |
11358 | btrfs_warn(fs_info, | |
11359 | "failed to trim %llu device(s), last error %d", | |
11360 | dev_failed, dev_ret); | |
f7039b1d | 11361 | range->len = trimmed; |
93bba24d QW |
11362 | if (bg_ret) |
11363 | return bg_ret; | |
11364 | return dev_ret; | |
f7039b1d | 11365 | } |
8257b2dc MX |
11366 | |
11367 | /* | |
ea14b57f | 11368 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
11369 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
11370 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11371 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 11372 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 11373 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 11374 | */ |
ea14b57f | 11375 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
11376 | { |
11377 | percpu_counter_dec(&root->subv_writers->counter); | |
093258e6 | 11378 | cond_wake_up(&root->subv_writers->wait); |
8257b2dc MX |
11379 | } |
11380 | ||
ea14b57f | 11381 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 11382 | { |
ea14b57f | 11383 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
11384 | return 0; |
11385 | ||
11386 | percpu_counter_inc(&root->subv_writers->counter); | |
11387 | /* | |
11388 | * Make sure counter is updated before we check for snapshot creation. | |
11389 | */ | |
11390 | smp_mb(); | |
ea14b57f DS |
11391 | if (atomic_read(&root->will_be_snapshotted)) { |
11392 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
11393 | return 0; |
11394 | } | |
11395 | return 1; | |
11396 | } | |
0bc19f90 | 11397 | |
0bc19f90 ZL |
11398 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
11399 | { | |
11400 | while (true) { | |
11401 | int ret; | |
11402 | ||
ea14b57f | 11403 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
11404 | if (ret) |
11405 | break; | |
4625956a PZ |
11406 | wait_var_event(&root->will_be_snapshotted, |
11407 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
11408 | } |
11409 | } | |
031f24da QW |
11410 | |
11411 | void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg) | |
11412 | { | |
11413 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
11414 | ||
11415 | spin_lock(&fs_info->unused_bgs_lock); | |
11416 | if (list_empty(&bg->bg_list)) { | |
11417 | btrfs_get_block_group(bg); | |
11418 | trace_btrfs_add_unused_block_group(bg); | |
11419 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
11420 | } | |
11421 | spin_unlock(&fs_info->unused_bgs_lock); | |
11422 | } |