]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
edbd8d4e | 19 | #include <linux/pagemap.h> |
ec44a35c | 20 | #include <linux/writeback.h> |
21af804c | 21 | #include <linux/blkdev.h> |
b7a9f29f | 22 | #include <linux/sort.h> |
4184ea7f | 23 | #include <linux/rcupdate.h> |
817d52f8 | 24 | #include <linux/kthread.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
dff51cd1 | 26 | #include <linux/ratelimit.h> |
b150a4f1 | 27 | #include <linux/percpu_counter.h> |
74493f7a | 28 | #include "hash.h" |
995946dd | 29 | #include "tree-log.h" |
fec577fb CM |
30 | #include "disk-io.h" |
31 | #include "print-tree.h" | |
0b86a832 | 32 | #include "volumes.h" |
53b381b3 | 33 | #include "raid56.h" |
925baedd | 34 | #include "locking.h" |
fa9c0d79 | 35 | #include "free-space-cache.h" |
1e144fb8 | 36 | #include "free-space-tree.h" |
3fed40cc | 37 | #include "math.h" |
6ab0a202 | 38 | #include "sysfs.h" |
fcebe456 | 39 | #include "qgroup.h" |
fec577fb | 40 | |
709c0486 AJ |
41 | #undef SCRAMBLE_DELAYED_REFS |
42 | ||
9e622d6b MX |
43 | /* |
44 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
45 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
46 | * if we really need one. | |
47 | * | |
0e4f8f88 CM |
48 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
49 | * if we have very few chunks already allocated. This is | |
50 | * used as part of the clustering code to help make sure | |
51 | * we have a good pool of storage to cluster in, without | |
52 | * filling the FS with empty chunks | |
53 | * | |
9e622d6b MX |
54 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
55 | * | |
0e4f8f88 CM |
56 | */ |
57 | enum { | |
58 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
59 | CHUNK_ALLOC_LIMITED = 1, |
60 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
61 | }; |
62 | ||
fb25e914 JB |
63 | /* |
64 | * Control how reservations are dealt with. | |
65 | * | |
66 | * RESERVE_FREE - freeing a reservation. | |
67 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
68 | * ENOSPC accounting | |
69 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
70 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
71 | */ | |
72 | enum { | |
73 | RESERVE_FREE = 0, | |
74 | RESERVE_ALLOC = 1, | |
75 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
76 | }; | |
77 | ||
ce93ec54 JB |
78 | static int update_block_group(struct btrfs_trans_handle *trans, |
79 | struct btrfs_root *root, u64 bytenr, | |
80 | u64 num_bytes, int alloc); | |
5d4f98a2 YZ |
81 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
82 | struct btrfs_root *root, | |
c682f9b3 | 83 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
84 | u64 root_objectid, u64 owner_objectid, |
85 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 86 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
87 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
88 | struct extent_buffer *leaf, | |
89 | struct btrfs_extent_item *ei); | |
90 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
91 | struct btrfs_root *root, | |
92 | u64 parent, u64 root_objectid, | |
93 | u64 flags, u64 owner, u64 offset, | |
94 | struct btrfs_key *ins, int ref_mod); | |
95 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
96 | struct btrfs_root *root, | |
97 | u64 parent, u64 root_objectid, | |
98 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 99 | int level, struct btrfs_key *ins); |
6a63209f | 100 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 JB |
101 | struct btrfs_root *extent_root, u64 flags, |
102 | int force); | |
11833d66 YZ |
103 | static int find_next_key(struct btrfs_path *path, int level, |
104 | struct btrfs_key *key); | |
9ed74f2d JB |
105 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
106 | int dump_block_groups); | |
fb25e914 | 107 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
e570fd27 MX |
108 | u64 num_bytes, int reserve, |
109 | int delalloc); | |
5d80366e JB |
110 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
111 | u64 num_bytes); | |
48a3b636 ES |
112 | int btrfs_pin_extent(struct btrfs_root *root, |
113 | u64 bytenr, u64 num_bytes, int reserved); | |
957780eb JB |
114 | static int __reserve_metadata_bytes(struct btrfs_root *root, |
115 | struct btrfs_space_info *space_info, | |
116 | u64 orig_bytes, | |
117 | enum btrfs_reserve_flush_enum flush); | |
118 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
119 | struct btrfs_space_info *space_info, | |
120 | u64 num_bytes); | |
121 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
122 | struct btrfs_space_info *space_info, | |
123 | u64 num_bytes); | |
6a63209f | 124 | |
817d52f8 JB |
125 | static noinline int |
126 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
127 | { | |
128 | smp_mb(); | |
36cce922 JB |
129 | return cache->cached == BTRFS_CACHE_FINISHED || |
130 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
131 | } |
132 | ||
0f9dd46c JB |
133 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
134 | { | |
135 | return (cache->flags & bits) == bits; | |
136 | } | |
137 | ||
758f2dfc | 138 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
139 | { |
140 | atomic_inc(&cache->count); | |
141 | } | |
142 | ||
143 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
144 | { | |
f0486c68 YZ |
145 | if (atomic_dec_and_test(&cache->count)) { |
146 | WARN_ON(cache->pinned > 0); | |
147 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 148 | kfree(cache->free_space_ctl); |
11dfe35a | 149 | kfree(cache); |
f0486c68 | 150 | } |
11dfe35a JB |
151 | } |
152 | ||
0f9dd46c JB |
153 | /* |
154 | * this adds the block group to the fs_info rb tree for the block group | |
155 | * cache | |
156 | */ | |
b2950863 | 157 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
158 | struct btrfs_block_group_cache *block_group) |
159 | { | |
160 | struct rb_node **p; | |
161 | struct rb_node *parent = NULL; | |
162 | struct btrfs_block_group_cache *cache; | |
163 | ||
164 | spin_lock(&info->block_group_cache_lock); | |
165 | p = &info->block_group_cache_tree.rb_node; | |
166 | ||
167 | while (*p) { | |
168 | parent = *p; | |
169 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
170 | cache_node); | |
171 | if (block_group->key.objectid < cache->key.objectid) { | |
172 | p = &(*p)->rb_left; | |
173 | } else if (block_group->key.objectid > cache->key.objectid) { | |
174 | p = &(*p)->rb_right; | |
175 | } else { | |
176 | spin_unlock(&info->block_group_cache_lock); | |
177 | return -EEXIST; | |
178 | } | |
179 | } | |
180 | ||
181 | rb_link_node(&block_group->cache_node, parent, p); | |
182 | rb_insert_color(&block_group->cache_node, | |
183 | &info->block_group_cache_tree); | |
a1897fdd LB |
184 | |
185 | if (info->first_logical_byte > block_group->key.objectid) | |
186 | info->first_logical_byte = block_group->key.objectid; | |
187 | ||
0f9dd46c JB |
188 | spin_unlock(&info->block_group_cache_lock); |
189 | ||
190 | return 0; | |
191 | } | |
192 | ||
193 | /* | |
194 | * This will return the block group at or after bytenr if contains is 0, else | |
195 | * it will return the block group that contains the bytenr | |
196 | */ | |
197 | static struct btrfs_block_group_cache * | |
198 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
199 | int contains) | |
200 | { | |
201 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
202 | struct rb_node *n; | |
203 | u64 end, start; | |
204 | ||
205 | spin_lock(&info->block_group_cache_lock); | |
206 | n = info->block_group_cache_tree.rb_node; | |
207 | ||
208 | while (n) { | |
209 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
210 | cache_node); | |
211 | end = cache->key.objectid + cache->key.offset - 1; | |
212 | start = cache->key.objectid; | |
213 | ||
214 | if (bytenr < start) { | |
215 | if (!contains && (!ret || start < ret->key.objectid)) | |
216 | ret = cache; | |
217 | n = n->rb_left; | |
218 | } else if (bytenr > start) { | |
219 | if (contains && bytenr <= end) { | |
220 | ret = cache; | |
221 | break; | |
222 | } | |
223 | n = n->rb_right; | |
224 | } else { | |
225 | ret = cache; | |
226 | break; | |
227 | } | |
228 | } | |
a1897fdd | 229 | if (ret) { |
11dfe35a | 230 | btrfs_get_block_group(ret); |
a1897fdd LB |
231 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
232 | info->first_logical_byte = ret->key.objectid; | |
233 | } | |
0f9dd46c JB |
234 | spin_unlock(&info->block_group_cache_lock); |
235 | ||
236 | return ret; | |
237 | } | |
238 | ||
11833d66 YZ |
239 | static int add_excluded_extent(struct btrfs_root *root, |
240 | u64 start, u64 num_bytes) | |
817d52f8 | 241 | { |
11833d66 YZ |
242 | u64 end = start + num_bytes - 1; |
243 | set_extent_bits(&root->fs_info->freed_extents[0], | |
ceeb0ae7 | 244 | start, end, EXTENT_UPTODATE); |
11833d66 | 245 | set_extent_bits(&root->fs_info->freed_extents[1], |
ceeb0ae7 | 246 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
247 | return 0; |
248 | } | |
817d52f8 | 249 | |
11833d66 YZ |
250 | static void free_excluded_extents(struct btrfs_root *root, |
251 | struct btrfs_block_group_cache *cache) | |
252 | { | |
253 | u64 start, end; | |
817d52f8 | 254 | |
11833d66 YZ |
255 | start = cache->key.objectid; |
256 | end = start + cache->key.offset - 1; | |
257 | ||
258 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
91166212 | 259 | start, end, EXTENT_UPTODATE); |
11833d66 | 260 | clear_extent_bits(&root->fs_info->freed_extents[1], |
91166212 | 261 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
262 | } |
263 | ||
11833d66 YZ |
264 | static int exclude_super_stripes(struct btrfs_root *root, |
265 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 266 | { |
817d52f8 JB |
267 | u64 bytenr; |
268 | u64 *logical; | |
269 | int stripe_len; | |
270 | int i, nr, ret; | |
271 | ||
06b2331f YZ |
272 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
273 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
274 | cache->bytes_super += stripe_len; | |
275 | ret = add_excluded_extent(root, cache->key.objectid, | |
276 | stripe_len); | |
835d974f JB |
277 | if (ret) |
278 | return ret; | |
06b2331f YZ |
279 | } |
280 | ||
817d52f8 JB |
281 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
282 | bytenr = btrfs_sb_offset(i); | |
283 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
284 | cache->key.objectid, bytenr, | |
285 | 0, &logical, &nr, &stripe_len); | |
835d974f JB |
286 | if (ret) |
287 | return ret; | |
11833d66 | 288 | |
817d52f8 | 289 | while (nr--) { |
51bf5f0b JB |
290 | u64 start, len; |
291 | ||
292 | if (logical[nr] > cache->key.objectid + | |
293 | cache->key.offset) | |
294 | continue; | |
295 | ||
296 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
297 | continue; | |
298 | ||
299 | start = logical[nr]; | |
300 | if (start < cache->key.objectid) { | |
301 | start = cache->key.objectid; | |
302 | len = (logical[nr] + stripe_len) - start; | |
303 | } else { | |
304 | len = min_t(u64, stripe_len, | |
305 | cache->key.objectid + | |
306 | cache->key.offset - start); | |
307 | } | |
308 | ||
309 | cache->bytes_super += len; | |
310 | ret = add_excluded_extent(root, start, len); | |
835d974f JB |
311 | if (ret) { |
312 | kfree(logical); | |
313 | return ret; | |
314 | } | |
817d52f8 | 315 | } |
11833d66 | 316 | |
817d52f8 JB |
317 | kfree(logical); |
318 | } | |
817d52f8 JB |
319 | return 0; |
320 | } | |
321 | ||
11833d66 YZ |
322 | static struct btrfs_caching_control * |
323 | get_caching_control(struct btrfs_block_group_cache *cache) | |
324 | { | |
325 | struct btrfs_caching_control *ctl; | |
326 | ||
327 | spin_lock(&cache->lock); | |
dde5abee JB |
328 | if (!cache->caching_ctl) { |
329 | spin_unlock(&cache->lock); | |
11833d66 YZ |
330 | return NULL; |
331 | } | |
332 | ||
333 | ctl = cache->caching_ctl; | |
334 | atomic_inc(&ctl->count); | |
335 | spin_unlock(&cache->lock); | |
336 | return ctl; | |
337 | } | |
338 | ||
339 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
340 | { | |
341 | if (atomic_dec_and_test(&ctl->count)) | |
342 | kfree(ctl); | |
343 | } | |
344 | ||
d0bd4560 JB |
345 | #ifdef CONFIG_BTRFS_DEBUG |
346 | static void fragment_free_space(struct btrfs_root *root, | |
347 | struct btrfs_block_group_cache *block_group) | |
348 | { | |
349 | u64 start = block_group->key.objectid; | |
350 | u64 len = block_group->key.offset; | |
351 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
352 | root->nodesize : root->sectorsize; | |
353 | u64 step = chunk << 1; | |
354 | ||
355 | while (len > chunk) { | |
356 | btrfs_remove_free_space(block_group, start, chunk); | |
357 | start += step; | |
358 | if (len < step) | |
359 | len = 0; | |
360 | else | |
361 | len -= step; | |
362 | } | |
363 | } | |
364 | #endif | |
365 | ||
0f9dd46c JB |
366 | /* |
367 | * this is only called by cache_block_group, since we could have freed extents | |
368 | * we need to check the pinned_extents for any extents that can't be used yet | |
369 | * since their free space will be released as soon as the transaction commits. | |
370 | */ | |
a5ed9182 OS |
371 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
372 | struct btrfs_fs_info *info, u64 start, u64 end) | |
0f9dd46c | 373 | { |
817d52f8 | 374 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
375 | int ret; |
376 | ||
377 | while (start < end) { | |
11833d66 | 378 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 379 | &extent_start, &extent_end, |
e6138876 JB |
380 | EXTENT_DIRTY | EXTENT_UPTODATE, |
381 | NULL); | |
0f9dd46c JB |
382 | if (ret) |
383 | break; | |
384 | ||
06b2331f | 385 | if (extent_start <= start) { |
0f9dd46c JB |
386 | start = extent_end + 1; |
387 | } else if (extent_start > start && extent_start < end) { | |
388 | size = extent_start - start; | |
817d52f8 | 389 | total_added += size; |
ea6a478e JB |
390 | ret = btrfs_add_free_space(block_group, start, |
391 | size); | |
79787eaa | 392 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
393 | start = extent_end + 1; |
394 | } else { | |
395 | break; | |
396 | } | |
397 | } | |
398 | ||
399 | if (start < end) { | |
400 | size = end - start; | |
817d52f8 | 401 | total_added += size; |
ea6a478e | 402 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 403 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
404 | } |
405 | ||
817d52f8 | 406 | return total_added; |
0f9dd46c JB |
407 | } |
408 | ||
73fa48b6 | 409 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 410 | { |
bab39bf9 JB |
411 | struct btrfs_block_group_cache *block_group; |
412 | struct btrfs_fs_info *fs_info; | |
bab39bf9 | 413 | struct btrfs_root *extent_root; |
e37c9e69 | 414 | struct btrfs_path *path; |
5f39d397 | 415 | struct extent_buffer *leaf; |
11833d66 | 416 | struct btrfs_key key; |
817d52f8 | 417 | u64 total_found = 0; |
11833d66 YZ |
418 | u64 last = 0; |
419 | u32 nritems; | |
73fa48b6 | 420 | int ret; |
d0bd4560 | 421 | bool wakeup = true; |
f510cfec | 422 | |
bab39bf9 JB |
423 | block_group = caching_ctl->block_group; |
424 | fs_info = block_group->fs_info; | |
425 | extent_root = fs_info->extent_root; | |
426 | ||
e37c9e69 CM |
427 | path = btrfs_alloc_path(); |
428 | if (!path) | |
73fa48b6 | 429 | return -ENOMEM; |
7d7d6068 | 430 | |
817d52f8 | 431 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 432 | |
d0bd4560 JB |
433 | #ifdef CONFIG_BTRFS_DEBUG |
434 | /* | |
435 | * If we're fragmenting we don't want to make anybody think we can | |
436 | * allocate from this block group until we've had a chance to fragment | |
437 | * the free space. | |
438 | */ | |
439 | if (btrfs_should_fragment_free_space(extent_root, block_group)) | |
440 | wakeup = false; | |
441 | #endif | |
5cd57b2c | 442 | /* |
817d52f8 JB |
443 | * We don't want to deadlock with somebody trying to allocate a new |
444 | * extent for the extent root while also trying to search the extent | |
445 | * root to add free space. So we skip locking and search the commit | |
446 | * root, since its read-only | |
5cd57b2c CM |
447 | */ |
448 | path->skip_locking = 1; | |
817d52f8 | 449 | path->search_commit_root = 1; |
e4058b54 | 450 | path->reada = READA_FORWARD; |
817d52f8 | 451 | |
e4404d6e | 452 | key.objectid = last; |
e37c9e69 | 453 | key.offset = 0; |
11833d66 | 454 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 455 | |
52ee28d2 | 456 | next: |
11833d66 | 457 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 458 | if (ret < 0) |
73fa48b6 | 459 | goto out; |
a512bbf8 | 460 | |
11833d66 YZ |
461 | leaf = path->nodes[0]; |
462 | nritems = btrfs_header_nritems(leaf); | |
463 | ||
d397712b | 464 | while (1) { |
7841cb28 | 465 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 466 | last = (u64)-1; |
817d52f8 | 467 | break; |
f25784b3 | 468 | } |
817d52f8 | 469 | |
11833d66 YZ |
470 | if (path->slots[0] < nritems) { |
471 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
472 | } else { | |
473 | ret = find_next_key(path, 0, &key); | |
474 | if (ret) | |
e37c9e69 | 475 | break; |
817d52f8 | 476 | |
c9ea7b24 | 477 | if (need_resched() || |
9e351cc8 | 478 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
479 | if (wakeup) |
480 | caching_ctl->progress = last; | |
ff5714cc | 481 | btrfs_release_path(path); |
9e351cc8 | 482 | up_read(&fs_info->commit_root_sem); |
589d8ade | 483 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 484 | cond_resched(); |
73fa48b6 OS |
485 | mutex_lock(&caching_ctl->mutex); |
486 | down_read(&fs_info->commit_root_sem); | |
487 | goto next; | |
589d8ade | 488 | } |
0a3896d0 JB |
489 | |
490 | ret = btrfs_next_leaf(extent_root, path); | |
491 | if (ret < 0) | |
73fa48b6 | 492 | goto out; |
0a3896d0 JB |
493 | if (ret) |
494 | break; | |
589d8ade JB |
495 | leaf = path->nodes[0]; |
496 | nritems = btrfs_header_nritems(leaf); | |
497 | continue; | |
11833d66 | 498 | } |
817d52f8 | 499 | |
52ee28d2 LB |
500 | if (key.objectid < last) { |
501 | key.objectid = last; | |
502 | key.offset = 0; | |
503 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
504 | ||
d0bd4560 JB |
505 | if (wakeup) |
506 | caching_ctl->progress = last; | |
52ee28d2 LB |
507 | btrfs_release_path(path); |
508 | goto next; | |
509 | } | |
510 | ||
11833d66 YZ |
511 | if (key.objectid < block_group->key.objectid) { |
512 | path->slots[0]++; | |
817d52f8 | 513 | continue; |
e37c9e69 | 514 | } |
0f9dd46c | 515 | |
e37c9e69 | 516 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 517 | block_group->key.offset) |
e37c9e69 | 518 | break; |
7d7d6068 | 519 | |
3173a18f JB |
520 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
521 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
522 | total_found += add_new_free_space(block_group, |
523 | fs_info, last, | |
524 | key.objectid); | |
3173a18f JB |
525 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
526 | last = key.objectid + | |
707e8a07 | 527 | fs_info->tree_root->nodesize; |
3173a18f JB |
528 | else |
529 | last = key.objectid + key.offset; | |
817d52f8 | 530 | |
73fa48b6 | 531 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 532 | total_found = 0; |
d0bd4560 JB |
533 | if (wakeup) |
534 | wake_up(&caching_ctl->wait); | |
11833d66 | 535 | } |
817d52f8 | 536 | } |
e37c9e69 CM |
537 | path->slots[0]++; |
538 | } | |
817d52f8 | 539 | ret = 0; |
e37c9e69 | 540 | |
817d52f8 JB |
541 | total_found += add_new_free_space(block_group, fs_info, last, |
542 | block_group->key.objectid + | |
543 | block_group->key.offset); | |
11833d66 | 544 | caching_ctl->progress = (u64)-1; |
817d52f8 | 545 | |
73fa48b6 OS |
546 | out: |
547 | btrfs_free_path(path); | |
548 | return ret; | |
549 | } | |
550 | ||
551 | static noinline void caching_thread(struct btrfs_work *work) | |
552 | { | |
553 | struct btrfs_block_group_cache *block_group; | |
554 | struct btrfs_fs_info *fs_info; | |
555 | struct btrfs_caching_control *caching_ctl; | |
b4570aa9 | 556 | struct btrfs_root *extent_root; |
73fa48b6 OS |
557 | int ret; |
558 | ||
559 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
560 | block_group = caching_ctl->block_group; | |
561 | fs_info = block_group->fs_info; | |
b4570aa9 | 562 | extent_root = fs_info->extent_root; |
73fa48b6 OS |
563 | |
564 | mutex_lock(&caching_ctl->mutex); | |
565 | down_read(&fs_info->commit_root_sem); | |
566 | ||
1e144fb8 OS |
567 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
568 | ret = load_free_space_tree(caching_ctl); | |
569 | else | |
570 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 571 | |
817d52f8 | 572 | spin_lock(&block_group->lock); |
11833d66 | 573 | block_group->caching_ctl = NULL; |
73fa48b6 | 574 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 575 | spin_unlock(&block_group->lock); |
0f9dd46c | 576 | |
d0bd4560 JB |
577 | #ifdef CONFIG_BTRFS_DEBUG |
578 | if (btrfs_should_fragment_free_space(extent_root, block_group)) { | |
579 | u64 bytes_used; | |
580 | ||
581 | spin_lock(&block_group->space_info->lock); | |
582 | spin_lock(&block_group->lock); | |
583 | bytes_used = block_group->key.offset - | |
584 | btrfs_block_group_used(&block_group->item); | |
585 | block_group->space_info->bytes_used += bytes_used >> 1; | |
586 | spin_unlock(&block_group->lock); | |
587 | spin_unlock(&block_group->space_info->lock); | |
588 | fragment_free_space(extent_root, block_group); | |
589 | } | |
590 | #endif | |
591 | ||
592 | caching_ctl->progress = (u64)-1; | |
11833d66 | 593 | |
9e351cc8 | 594 | up_read(&fs_info->commit_root_sem); |
73fa48b6 | 595 | free_excluded_extents(fs_info->extent_root, block_group); |
11833d66 | 596 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 597 | |
11833d66 YZ |
598 | wake_up(&caching_ctl->wait); |
599 | ||
600 | put_caching_control(caching_ctl); | |
11dfe35a | 601 | btrfs_put_block_group(block_group); |
817d52f8 JB |
602 | } |
603 | ||
9d66e233 | 604 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 605 | int load_cache_only) |
817d52f8 | 606 | { |
291c7d2f | 607 | DEFINE_WAIT(wait); |
11833d66 YZ |
608 | struct btrfs_fs_info *fs_info = cache->fs_info; |
609 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
610 | int ret = 0; |
611 | ||
291c7d2f | 612 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
613 | if (!caching_ctl) |
614 | return -ENOMEM; | |
291c7d2f JB |
615 | |
616 | INIT_LIST_HEAD(&caching_ctl->list); | |
617 | mutex_init(&caching_ctl->mutex); | |
618 | init_waitqueue_head(&caching_ctl->wait); | |
619 | caching_ctl->block_group = cache; | |
620 | caching_ctl->progress = cache->key.objectid; | |
621 | atomic_set(&caching_ctl->count, 1); | |
9e0af237 LB |
622 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
623 | caching_thread, NULL, NULL); | |
291c7d2f JB |
624 | |
625 | spin_lock(&cache->lock); | |
626 | /* | |
627 | * This should be a rare occasion, but this could happen I think in the | |
628 | * case where one thread starts to load the space cache info, and then | |
629 | * some other thread starts a transaction commit which tries to do an | |
630 | * allocation while the other thread is still loading the space cache | |
631 | * info. The previous loop should have kept us from choosing this block | |
632 | * group, but if we've moved to the state where we will wait on caching | |
633 | * block groups we need to first check if we're doing a fast load here, | |
634 | * so we can wait for it to finish, otherwise we could end up allocating | |
635 | * from a block group who's cache gets evicted for one reason or | |
636 | * another. | |
637 | */ | |
638 | while (cache->cached == BTRFS_CACHE_FAST) { | |
639 | struct btrfs_caching_control *ctl; | |
640 | ||
641 | ctl = cache->caching_ctl; | |
642 | atomic_inc(&ctl->count); | |
643 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
644 | spin_unlock(&cache->lock); | |
645 | ||
646 | schedule(); | |
647 | ||
648 | finish_wait(&ctl->wait, &wait); | |
649 | put_caching_control(ctl); | |
650 | spin_lock(&cache->lock); | |
651 | } | |
652 | ||
653 | if (cache->cached != BTRFS_CACHE_NO) { | |
654 | spin_unlock(&cache->lock); | |
655 | kfree(caching_ctl); | |
11833d66 | 656 | return 0; |
291c7d2f JB |
657 | } |
658 | WARN_ON(cache->caching_ctl); | |
659 | cache->caching_ctl = caching_ctl; | |
660 | cache->cached = BTRFS_CACHE_FAST; | |
661 | spin_unlock(&cache->lock); | |
11833d66 | 662 | |
d53ba474 | 663 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
cb83b7b8 | 664 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
665 | ret = load_free_space_cache(fs_info, cache); |
666 | ||
667 | spin_lock(&cache->lock); | |
668 | if (ret == 1) { | |
291c7d2f | 669 | cache->caching_ctl = NULL; |
9d66e233 JB |
670 | cache->cached = BTRFS_CACHE_FINISHED; |
671 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 672 | caching_ctl->progress = (u64)-1; |
9d66e233 | 673 | } else { |
291c7d2f JB |
674 | if (load_cache_only) { |
675 | cache->caching_ctl = NULL; | |
676 | cache->cached = BTRFS_CACHE_NO; | |
677 | } else { | |
678 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 679 | cache->has_caching_ctl = 1; |
291c7d2f | 680 | } |
9d66e233 JB |
681 | } |
682 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
683 | #ifdef CONFIG_BTRFS_DEBUG |
684 | if (ret == 1 && | |
685 | btrfs_should_fragment_free_space(fs_info->extent_root, | |
686 | cache)) { | |
687 | u64 bytes_used; | |
688 | ||
689 | spin_lock(&cache->space_info->lock); | |
690 | spin_lock(&cache->lock); | |
691 | bytes_used = cache->key.offset - | |
692 | btrfs_block_group_used(&cache->item); | |
693 | cache->space_info->bytes_used += bytes_used >> 1; | |
694 | spin_unlock(&cache->lock); | |
695 | spin_unlock(&cache->space_info->lock); | |
696 | fragment_free_space(fs_info->extent_root, cache); | |
697 | } | |
698 | #endif | |
cb83b7b8 JB |
699 | mutex_unlock(&caching_ctl->mutex); |
700 | ||
291c7d2f | 701 | wake_up(&caching_ctl->wait); |
3c14874a | 702 | if (ret == 1) { |
291c7d2f | 703 | put_caching_control(caching_ctl); |
3c14874a | 704 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 705 | return 0; |
3c14874a | 706 | } |
291c7d2f JB |
707 | } else { |
708 | /* | |
1e144fb8 OS |
709 | * We're either using the free space tree or no caching at all. |
710 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
711 | */ |
712 | spin_lock(&cache->lock); | |
713 | if (load_cache_only) { | |
714 | cache->caching_ctl = NULL; | |
715 | cache->cached = BTRFS_CACHE_NO; | |
716 | } else { | |
717 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 718 | cache->has_caching_ctl = 1; |
291c7d2f JB |
719 | } |
720 | spin_unlock(&cache->lock); | |
721 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
722 | } |
723 | ||
291c7d2f JB |
724 | if (load_cache_only) { |
725 | put_caching_control(caching_ctl); | |
11833d66 | 726 | return 0; |
817d52f8 | 727 | } |
817d52f8 | 728 | |
9e351cc8 | 729 | down_write(&fs_info->commit_root_sem); |
291c7d2f | 730 | atomic_inc(&caching_ctl->count); |
11833d66 | 731 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 732 | up_write(&fs_info->commit_root_sem); |
11833d66 | 733 | |
11dfe35a | 734 | btrfs_get_block_group(cache); |
11833d66 | 735 | |
e66f0bb1 | 736 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 737 | |
ef8bbdfe | 738 | return ret; |
e37c9e69 CM |
739 | } |
740 | ||
0f9dd46c JB |
741 | /* |
742 | * return the block group that starts at or after bytenr | |
743 | */ | |
d397712b CM |
744 | static struct btrfs_block_group_cache * |
745 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 746 | { |
0f9dd46c | 747 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 748 | |
0f9dd46c | 749 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 750 | |
0f9dd46c | 751 | return cache; |
0ef3e66b CM |
752 | } |
753 | ||
0f9dd46c | 754 | /* |
9f55684c | 755 | * return the block group that contains the given bytenr |
0f9dd46c | 756 | */ |
d397712b CM |
757 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
758 | struct btrfs_fs_info *info, | |
759 | u64 bytenr) | |
be744175 | 760 | { |
0f9dd46c | 761 | struct btrfs_block_group_cache *cache; |
be744175 | 762 | |
0f9dd46c | 763 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 764 | |
0f9dd46c | 765 | return cache; |
be744175 | 766 | } |
0b86a832 | 767 | |
0f9dd46c JB |
768 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
769 | u64 flags) | |
6324fbf3 | 770 | { |
0f9dd46c | 771 | struct list_head *head = &info->space_info; |
0f9dd46c | 772 | struct btrfs_space_info *found; |
4184ea7f | 773 | |
52ba6929 | 774 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 775 | |
4184ea7f CM |
776 | rcu_read_lock(); |
777 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 778 | if (found->flags & flags) { |
4184ea7f | 779 | rcu_read_unlock(); |
0f9dd46c | 780 | return found; |
4184ea7f | 781 | } |
0f9dd46c | 782 | } |
4184ea7f | 783 | rcu_read_unlock(); |
0f9dd46c | 784 | return NULL; |
6324fbf3 CM |
785 | } |
786 | ||
4184ea7f CM |
787 | /* |
788 | * after adding space to the filesystem, we need to clear the full flags | |
789 | * on all the space infos. | |
790 | */ | |
791 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
792 | { | |
793 | struct list_head *head = &info->space_info; | |
794 | struct btrfs_space_info *found; | |
795 | ||
796 | rcu_read_lock(); | |
797 | list_for_each_entry_rcu(found, head, list) | |
798 | found->full = 0; | |
799 | rcu_read_unlock(); | |
800 | } | |
801 | ||
1a4ed8fd FM |
802 | /* simple helper to search for an existing data extent at a given offset */ |
803 | int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len) | |
e02119d5 CM |
804 | { |
805 | int ret; | |
806 | struct btrfs_key key; | |
31840ae1 | 807 | struct btrfs_path *path; |
e02119d5 | 808 | |
31840ae1 | 809 | path = btrfs_alloc_path(); |
d8926bb3 MF |
810 | if (!path) |
811 | return -ENOMEM; | |
812 | ||
e02119d5 CM |
813 | key.objectid = start; |
814 | key.offset = len; | |
3173a18f | 815 | key.type = BTRFS_EXTENT_ITEM_KEY; |
e02119d5 CM |
816 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, |
817 | 0, 0); | |
31840ae1 | 818 | btrfs_free_path(path); |
7bb86316 CM |
819 | return ret; |
820 | } | |
821 | ||
a22285a6 | 822 | /* |
3173a18f | 823 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
824 | * |
825 | * the head node for delayed ref is used to store the sum of all the | |
826 | * reference count modifications queued up in the rbtree. the head | |
827 | * node may also store the extent flags to set. This way you can check | |
828 | * to see what the reference count and extent flags would be if all of | |
829 | * the delayed refs are not processed. | |
830 | */ | |
831 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
832 | struct btrfs_root *root, u64 bytenr, | |
3173a18f | 833 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
834 | { |
835 | struct btrfs_delayed_ref_head *head; | |
836 | struct btrfs_delayed_ref_root *delayed_refs; | |
837 | struct btrfs_path *path; | |
838 | struct btrfs_extent_item *ei; | |
839 | struct extent_buffer *leaf; | |
840 | struct btrfs_key key; | |
841 | u32 item_size; | |
842 | u64 num_refs; | |
843 | u64 extent_flags; | |
844 | int ret; | |
845 | ||
3173a18f JB |
846 | /* |
847 | * If we don't have skinny metadata, don't bother doing anything | |
848 | * different | |
849 | */ | |
850 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { | |
707e8a07 | 851 | offset = root->nodesize; |
3173a18f JB |
852 | metadata = 0; |
853 | } | |
854 | ||
a22285a6 YZ |
855 | path = btrfs_alloc_path(); |
856 | if (!path) | |
857 | return -ENOMEM; | |
858 | ||
a22285a6 YZ |
859 | if (!trans) { |
860 | path->skip_locking = 1; | |
861 | path->search_commit_root = 1; | |
862 | } | |
639eefc8 FDBM |
863 | |
864 | search_again: | |
865 | key.objectid = bytenr; | |
866 | key.offset = offset; | |
867 | if (metadata) | |
868 | key.type = BTRFS_METADATA_ITEM_KEY; | |
869 | else | |
870 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
871 | ||
a22285a6 YZ |
872 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, |
873 | &key, path, 0, 0); | |
874 | if (ret < 0) | |
875 | goto out_free; | |
876 | ||
3173a18f | 877 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
878 | if (path->slots[0]) { |
879 | path->slots[0]--; | |
880 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
881 | path->slots[0]); | |
882 | if (key.objectid == bytenr && | |
883 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
707e8a07 | 884 | key.offset == root->nodesize) |
74be9510 FDBM |
885 | ret = 0; |
886 | } | |
3173a18f JB |
887 | } |
888 | ||
a22285a6 YZ |
889 | if (ret == 0) { |
890 | leaf = path->nodes[0]; | |
891 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
892 | if (item_size >= sizeof(*ei)) { | |
893 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
894 | struct btrfs_extent_item); | |
895 | num_refs = btrfs_extent_refs(leaf, ei); | |
896 | extent_flags = btrfs_extent_flags(leaf, ei); | |
897 | } else { | |
898 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
899 | struct btrfs_extent_item_v0 *ei0; | |
900 | BUG_ON(item_size != sizeof(*ei0)); | |
901 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
902 | struct btrfs_extent_item_v0); | |
903 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
904 | /* FIXME: this isn't correct for data */ | |
905 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
906 | #else | |
907 | BUG(); | |
908 | #endif | |
909 | } | |
910 | BUG_ON(num_refs == 0); | |
911 | } else { | |
912 | num_refs = 0; | |
913 | extent_flags = 0; | |
914 | ret = 0; | |
915 | } | |
916 | ||
917 | if (!trans) | |
918 | goto out; | |
919 | ||
920 | delayed_refs = &trans->transaction->delayed_refs; | |
921 | spin_lock(&delayed_refs->lock); | |
922 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
923 | if (head) { | |
924 | if (!mutex_trylock(&head->mutex)) { | |
925 | atomic_inc(&head->node.refs); | |
926 | spin_unlock(&delayed_refs->lock); | |
927 | ||
b3b4aa74 | 928 | btrfs_release_path(path); |
a22285a6 | 929 | |
8cc33e5c DS |
930 | /* |
931 | * Mutex was contended, block until it's released and try | |
932 | * again | |
933 | */ | |
a22285a6 YZ |
934 | mutex_lock(&head->mutex); |
935 | mutex_unlock(&head->mutex); | |
936 | btrfs_put_delayed_ref(&head->node); | |
639eefc8 | 937 | goto search_again; |
a22285a6 | 938 | } |
d7df2c79 | 939 | spin_lock(&head->lock); |
a22285a6 YZ |
940 | if (head->extent_op && head->extent_op->update_flags) |
941 | extent_flags |= head->extent_op->flags_to_set; | |
942 | else | |
943 | BUG_ON(num_refs == 0); | |
944 | ||
945 | num_refs += head->node.ref_mod; | |
d7df2c79 | 946 | spin_unlock(&head->lock); |
a22285a6 YZ |
947 | mutex_unlock(&head->mutex); |
948 | } | |
949 | spin_unlock(&delayed_refs->lock); | |
950 | out: | |
951 | WARN_ON(num_refs == 0); | |
952 | if (refs) | |
953 | *refs = num_refs; | |
954 | if (flags) | |
955 | *flags = extent_flags; | |
956 | out_free: | |
957 | btrfs_free_path(path); | |
958 | return ret; | |
959 | } | |
960 | ||
d8d5f3e1 CM |
961 | /* |
962 | * Back reference rules. Back refs have three main goals: | |
963 | * | |
964 | * 1) differentiate between all holders of references to an extent so that | |
965 | * when a reference is dropped we can make sure it was a valid reference | |
966 | * before freeing the extent. | |
967 | * | |
968 | * 2) Provide enough information to quickly find the holders of an extent | |
969 | * if we notice a given block is corrupted or bad. | |
970 | * | |
971 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
972 | * maintenance. This is actually the same as #2, but with a slightly | |
973 | * different use case. | |
974 | * | |
5d4f98a2 YZ |
975 | * There are two kinds of back refs. The implicit back refs is optimized |
976 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
977 | * back refs of this kind provide information about the block's owner tree | |
978 | * and the pointer's key. These information allow us to find the block by | |
979 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
980 | * referenced by their owner trees. The location of tree block is recorded | |
981 | * in the back refs. Actually the full back refs is generic, and can be | |
982 | * used in all cases the implicit back refs is used. The major shortcoming | |
983 | * of the full back refs is its overhead. Every time a tree block gets | |
984 | * COWed, we have to update back refs entry for all pointers in it. | |
985 | * | |
986 | * For a newly allocated tree block, we use implicit back refs for | |
987 | * pointers in it. This means most tree related operations only involve | |
988 | * implicit back refs. For a tree block created in old transaction, the | |
989 | * only way to drop a reference to it is COW it. So we can detect the | |
990 | * event that tree block loses its owner tree's reference and do the | |
991 | * back refs conversion. | |
992 | * | |
01327610 | 993 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
994 | * |
995 | * The reference count of the block is one and the tree is the block's | |
996 | * owner tree. Nothing to do in this case. | |
997 | * | |
998 | * The reference count of the block is one and the tree is not the | |
999 | * block's owner tree. In this case, full back refs is used for pointers | |
1000 | * in the block. Remove these full back refs, add implicit back refs for | |
1001 | * every pointers in the new block. | |
1002 | * | |
1003 | * The reference count of the block is greater than one and the tree is | |
1004 | * the block's owner tree. In this case, implicit back refs is used for | |
1005 | * pointers in the block. Add full back refs for every pointers in the | |
1006 | * block, increase lower level extents' reference counts. The original | |
1007 | * implicit back refs are entailed to the new block. | |
1008 | * | |
1009 | * The reference count of the block is greater than one and the tree is | |
1010 | * not the block's owner tree. Add implicit back refs for every pointer in | |
1011 | * the new block, increase lower level extents' reference count. | |
1012 | * | |
1013 | * Back Reference Key composing: | |
1014 | * | |
1015 | * The key objectid corresponds to the first byte in the extent, | |
1016 | * The key type is used to differentiate between types of back refs. | |
1017 | * There are different meanings of the key offset for different types | |
1018 | * of back refs. | |
1019 | * | |
d8d5f3e1 CM |
1020 | * File extents can be referenced by: |
1021 | * | |
1022 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1023 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1024 | * - different offsets inside a file (bookend extents in file.c) |
1025 | * | |
5d4f98a2 | 1026 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1027 | * |
1028 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1029 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1030 | * - original offset in the file |
1031 | * - how many bookend extents | |
d8d5f3e1 | 1032 | * |
5d4f98a2 YZ |
1033 | * The key offset for the implicit back refs is hash of the first |
1034 | * three fields. | |
d8d5f3e1 | 1035 | * |
5d4f98a2 | 1036 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1037 | * |
5d4f98a2 | 1038 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1039 | * |
5d4f98a2 YZ |
1040 | * The key offset for the implicit back refs is the first byte of |
1041 | * the tree leaf | |
d8d5f3e1 | 1042 | * |
5d4f98a2 YZ |
1043 | * When a file extent is allocated, The implicit back refs is used. |
1044 | * the fields are filled in: | |
d8d5f3e1 | 1045 | * |
5d4f98a2 | 1046 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1047 | * |
5d4f98a2 YZ |
1048 | * When a file extent is removed file truncation, we find the |
1049 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1050 | * |
5d4f98a2 | 1051 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1052 | * |
5d4f98a2 | 1053 | * Btree extents can be referenced by: |
d8d5f3e1 | 1054 | * |
5d4f98a2 | 1055 | * - Different subvolumes |
d8d5f3e1 | 1056 | * |
5d4f98a2 YZ |
1057 | * Both the implicit back refs and the full back refs for tree blocks |
1058 | * only consist of key. The key offset for the implicit back refs is | |
1059 | * objectid of block's owner tree. The key offset for the full back refs | |
1060 | * is the first byte of parent block. | |
d8d5f3e1 | 1061 | * |
5d4f98a2 YZ |
1062 | * When implicit back refs is used, information about the lowest key and |
1063 | * level of the tree block are required. These information are stored in | |
1064 | * tree block info structure. | |
d8d5f3e1 | 1065 | */ |
31840ae1 | 1066 | |
5d4f98a2 YZ |
1067 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1068 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
1069 | struct btrfs_root *root, | |
1070 | struct btrfs_path *path, | |
1071 | u64 owner, u32 extra_size) | |
7bb86316 | 1072 | { |
5d4f98a2 YZ |
1073 | struct btrfs_extent_item *item; |
1074 | struct btrfs_extent_item_v0 *ei0; | |
1075 | struct btrfs_extent_ref_v0 *ref0; | |
1076 | struct btrfs_tree_block_info *bi; | |
1077 | struct extent_buffer *leaf; | |
7bb86316 | 1078 | struct btrfs_key key; |
5d4f98a2 YZ |
1079 | struct btrfs_key found_key; |
1080 | u32 new_size = sizeof(*item); | |
1081 | u64 refs; | |
1082 | int ret; | |
1083 | ||
1084 | leaf = path->nodes[0]; | |
1085 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1086 | ||
1087 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1088 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1089 | struct btrfs_extent_item_v0); | |
1090 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1091 | ||
1092 | if (owner == (u64)-1) { | |
1093 | while (1) { | |
1094 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1095 | ret = btrfs_next_leaf(root, path); | |
1096 | if (ret < 0) | |
1097 | return ret; | |
79787eaa | 1098 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1099 | leaf = path->nodes[0]; |
1100 | } | |
1101 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1102 | path->slots[0]); | |
1103 | BUG_ON(key.objectid != found_key.objectid); | |
1104 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1105 | path->slots[0]++; | |
1106 | continue; | |
1107 | } | |
1108 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1109 | struct btrfs_extent_ref_v0); | |
1110 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1111 | break; | |
1112 | } | |
1113 | } | |
b3b4aa74 | 1114 | btrfs_release_path(path); |
5d4f98a2 YZ |
1115 | |
1116 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1117 | new_size += sizeof(*bi); | |
1118 | ||
1119 | new_size -= sizeof(*ei0); | |
1120 | ret = btrfs_search_slot(trans, root, &key, path, | |
1121 | new_size + extra_size, 1); | |
1122 | if (ret < 0) | |
1123 | return ret; | |
79787eaa | 1124 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1125 | |
4b90c680 | 1126 | btrfs_extend_item(root, path, new_size); |
5d4f98a2 YZ |
1127 | |
1128 | leaf = path->nodes[0]; | |
1129 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1130 | btrfs_set_extent_refs(leaf, item, refs); | |
1131 | /* FIXME: get real generation */ | |
1132 | btrfs_set_extent_generation(leaf, item, 0); | |
1133 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1134 | btrfs_set_extent_flags(leaf, item, | |
1135 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1136 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1137 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1138 | /* FIXME: get first key of the block */ | |
1139 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1140 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1141 | } else { | |
1142 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1143 | } | |
1144 | btrfs_mark_buffer_dirty(leaf); | |
1145 | return 0; | |
1146 | } | |
1147 | #endif | |
1148 | ||
1149 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1150 | { | |
1151 | u32 high_crc = ~(u32)0; | |
1152 | u32 low_crc = ~(u32)0; | |
1153 | __le64 lenum; | |
1154 | ||
1155 | lenum = cpu_to_le64(root_objectid); | |
14a958e6 | 1156 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1157 | lenum = cpu_to_le64(owner); |
14a958e6 | 1158 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1159 | lenum = cpu_to_le64(offset); |
14a958e6 | 1160 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1161 | |
1162 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1163 | } | |
1164 | ||
1165 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1166 | struct btrfs_extent_data_ref *ref) | |
1167 | { | |
1168 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1169 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1170 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1171 | } | |
1172 | ||
1173 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1174 | struct btrfs_extent_data_ref *ref, | |
1175 | u64 root_objectid, u64 owner, u64 offset) | |
1176 | { | |
1177 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1178 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1179 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1180 | return 0; | |
1181 | return 1; | |
1182 | } | |
1183 | ||
1184 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1185 | struct btrfs_root *root, | |
1186 | struct btrfs_path *path, | |
1187 | u64 bytenr, u64 parent, | |
1188 | u64 root_objectid, | |
1189 | u64 owner, u64 offset) | |
1190 | { | |
1191 | struct btrfs_key key; | |
1192 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1193 | struct extent_buffer *leaf; |
5d4f98a2 | 1194 | u32 nritems; |
74493f7a | 1195 | int ret; |
5d4f98a2 YZ |
1196 | int recow; |
1197 | int err = -ENOENT; | |
74493f7a | 1198 | |
31840ae1 | 1199 | key.objectid = bytenr; |
5d4f98a2 YZ |
1200 | if (parent) { |
1201 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1202 | key.offset = parent; | |
1203 | } else { | |
1204 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1205 | key.offset = hash_extent_data_ref(root_objectid, | |
1206 | owner, offset); | |
1207 | } | |
1208 | again: | |
1209 | recow = 0; | |
1210 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1211 | if (ret < 0) { | |
1212 | err = ret; | |
1213 | goto fail; | |
1214 | } | |
31840ae1 | 1215 | |
5d4f98a2 YZ |
1216 | if (parent) { |
1217 | if (!ret) | |
1218 | return 0; | |
1219 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1220 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1221 | btrfs_release_path(path); |
5d4f98a2 YZ |
1222 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1223 | if (ret < 0) { | |
1224 | err = ret; | |
1225 | goto fail; | |
1226 | } | |
1227 | if (!ret) | |
1228 | return 0; | |
1229 | #endif | |
1230 | goto fail; | |
31840ae1 ZY |
1231 | } |
1232 | ||
1233 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1234 | nritems = btrfs_header_nritems(leaf); |
1235 | while (1) { | |
1236 | if (path->slots[0] >= nritems) { | |
1237 | ret = btrfs_next_leaf(root, path); | |
1238 | if (ret < 0) | |
1239 | err = ret; | |
1240 | if (ret) | |
1241 | goto fail; | |
1242 | ||
1243 | leaf = path->nodes[0]; | |
1244 | nritems = btrfs_header_nritems(leaf); | |
1245 | recow = 1; | |
1246 | } | |
1247 | ||
1248 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1249 | if (key.objectid != bytenr || | |
1250 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1251 | goto fail; | |
1252 | ||
1253 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1254 | struct btrfs_extent_data_ref); | |
1255 | ||
1256 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1257 | owner, offset)) { | |
1258 | if (recow) { | |
b3b4aa74 | 1259 | btrfs_release_path(path); |
5d4f98a2 YZ |
1260 | goto again; |
1261 | } | |
1262 | err = 0; | |
1263 | break; | |
1264 | } | |
1265 | path->slots[0]++; | |
31840ae1 | 1266 | } |
5d4f98a2 YZ |
1267 | fail: |
1268 | return err; | |
31840ae1 ZY |
1269 | } |
1270 | ||
5d4f98a2 YZ |
1271 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1272 | struct btrfs_root *root, | |
1273 | struct btrfs_path *path, | |
1274 | u64 bytenr, u64 parent, | |
1275 | u64 root_objectid, u64 owner, | |
1276 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1277 | { |
1278 | struct btrfs_key key; | |
1279 | struct extent_buffer *leaf; | |
5d4f98a2 | 1280 | u32 size; |
31840ae1 ZY |
1281 | u32 num_refs; |
1282 | int ret; | |
74493f7a | 1283 | |
74493f7a | 1284 | key.objectid = bytenr; |
5d4f98a2 YZ |
1285 | if (parent) { |
1286 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1287 | key.offset = parent; | |
1288 | size = sizeof(struct btrfs_shared_data_ref); | |
1289 | } else { | |
1290 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1291 | key.offset = hash_extent_data_ref(root_objectid, | |
1292 | owner, offset); | |
1293 | size = sizeof(struct btrfs_extent_data_ref); | |
1294 | } | |
74493f7a | 1295 | |
5d4f98a2 YZ |
1296 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1297 | if (ret && ret != -EEXIST) | |
1298 | goto fail; | |
1299 | ||
1300 | leaf = path->nodes[0]; | |
1301 | if (parent) { | |
1302 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1303 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1304 | struct btrfs_shared_data_ref); |
1305 | if (ret == 0) { | |
1306 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1307 | } else { | |
1308 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1309 | num_refs += refs_to_add; | |
1310 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1311 | } |
5d4f98a2 YZ |
1312 | } else { |
1313 | struct btrfs_extent_data_ref *ref; | |
1314 | while (ret == -EEXIST) { | |
1315 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1316 | struct btrfs_extent_data_ref); | |
1317 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1318 | owner, offset)) | |
1319 | break; | |
b3b4aa74 | 1320 | btrfs_release_path(path); |
5d4f98a2 YZ |
1321 | key.offset++; |
1322 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1323 | size); | |
1324 | if (ret && ret != -EEXIST) | |
1325 | goto fail; | |
31840ae1 | 1326 | |
5d4f98a2 YZ |
1327 | leaf = path->nodes[0]; |
1328 | } | |
1329 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1330 | struct btrfs_extent_data_ref); | |
1331 | if (ret == 0) { | |
1332 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1333 | root_objectid); | |
1334 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1335 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1336 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1337 | } else { | |
1338 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1339 | num_refs += refs_to_add; | |
1340 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1341 | } |
31840ae1 | 1342 | } |
5d4f98a2 YZ |
1343 | btrfs_mark_buffer_dirty(leaf); |
1344 | ret = 0; | |
1345 | fail: | |
b3b4aa74 | 1346 | btrfs_release_path(path); |
7bb86316 | 1347 | return ret; |
74493f7a CM |
1348 | } |
1349 | ||
5d4f98a2 YZ |
1350 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1351 | struct btrfs_root *root, | |
1352 | struct btrfs_path *path, | |
fcebe456 | 1353 | int refs_to_drop, int *last_ref) |
31840ae1 | 1354 | { |
5d4f98a2 YZ |
1355 | struct btrfs_key key; |
1356 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1357 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1358 | struct extent_buffer *leaf; |
5d4f98a2 | 1359 | u32 num_refs = 0; |
31840ae1 ZY |
1360 | int ret = 0; |
1361 | ||
1362 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1363 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1364 | ||
1365 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1366 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1367 | struct btrfs_extent_data_ref); | |
1368 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1369 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1370 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1371 | struct btrfs_shared_data_ref); | |
1372 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1373 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1374 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1375 | struct btrfs_extent_ref_v0 *ref0; | |
1376 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1377 | struct btrfs_extent_ref_v0); | |
1378 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1379 | #endif | |
1380 | } else { | |
1381 | BUG(); | |
1382 | } | |
1383 | ||
56bec294 CM |
1384 | BUG_ON(num_refs < refs_to_drop); |
1385 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1386 | |
31840ae1 ZY |
1387 | if (num_refs == 0) { |
1388 | ret = btrfs_del_item(trans, root, path); | |
fcebe456 | 1389 | *last_ref = 1; |
31840ae1 | 1390 | } else { |
5d4f98a2 YZ |
1391 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1392 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1393 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1394 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1395 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1396 | else { | |
1397 | struct btrfs_extent_ref_v0 *ref0; | |
1398 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1399 | struct btrfs_extent_ref_v0); | |
1400 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1401 | } | |
1402 | #endif | |
31840ae1 ZY |
1403 | btrfs_mark_buffer_dirty(leaf); |
1404 | } | |
31840ae1 ZY |
1405 | return ret; |
1406 | } | |
1407 | ||
9ed0dea0 | 1408 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1409 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1410 | { |
5d4f98a2 YZ |
1411 | struct btrfs_key key; |
1412 | struct extent_buffer *leaf; | |
1413 | struct btrfs_extent_data_ref *ref1; | |
1414 | struct btrfs_shared_data_ref *ref2; | |
1415 | u32 num_refs = 0; | |
1416 | ||
1417 | leaf = path->nodes[0]; | |
1418 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1419 | if (iref) { | |
1420 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1421 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1422 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1423 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1424 | } else { | |
1425 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1426 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1427 | } | |
1428 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1429 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1430 | struct btrfs_extent_data_ref); | |
1431 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1432 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1433 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1434 | struct btrfs_shared_data_ref); | |
1435 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1436 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1437 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1438 | struct btrfs_extent_ref_v0 *ref0; | |
1439 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1440 | struct btrfs_extent_ref_v0); | |
1441 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1442 | #endif |
5d4f98a2 YZ |
1443 | } else { |
1444 | WARN_ON(1); | |
1445 | } | |
1446 | return num_refs; | |
1447 | } | |
15916de8 | 1448 | |
5d4f98a2 YZ |
1449 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1450 | struct btrfs_root *root, | |
1451 | struct btrfs_path *path, | |
1452 | u64 bytenr, u64 parent, | |
1453 | u64 root_objectid) | |
1f3c79a2 | 1454 | { |
5d4f98a2 | 1455 | struct btrfs_key key; |
1f3c79a2 | 1456 | int ret; |
1f3c79a2 | 1457 | |
5d4f98a2 YZ |
1458 | key.objectid = bytenr; |
1459 | if (parent) { | |
1460 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1461 | key.offset = parent; | |
1462 | } else { | |
1463 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1464 | key.offset = root_objectid; | |
1f3c79a2 LH |
1465 | } |
1466 | ||
5d4f98a2 YZ |
1467 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1468 | if (ret > 0) | |
1469 | ret = -ENOENT; | |
1470 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1471 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1472 | btrfs_release_path(path); |
5d4f98a2 YZ |
1473 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1474 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1475 | if (ret > 0) | |
1476 | ret = -ENOENT; | |
1477 | } | |
1f3c79a2 | 1478 | #endif |
5d4f98a2 | 1479 | return ret; |
1f3c79a2 LH |
1480 | } |
1481 | ||
5d4f98a2 YZ |
1482 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1483 | struct btrfs_root *root, | |
1484 | struct btrfs_path *path, | |
1485 | u64 bytenr, u64 parent, | |
1486 | u64 root_objectid) | |
31840ae1 | 1487 | { |
5d4f98a2 | 1488 | struct btrfs_key key; |
31840ae1 | 1489 | int ret; |
31840ae1 | 1490 | |
5d4f98a2 YZ |
1491 | key.objectid = bytenr; |
1492 | if (parent) { | |
1493 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1494 | key.offset = parent; | |
1495 | } else { | |
1496 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1497 | key.offset = root_objectid; | |
1498 | } | |
1499 | ||
1500 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1501 | btrfs_release_path(path); |
31840ae1 ZY |
1502 | return ret; |
1503 | } | |
1504 | ||
5d4f98a2 | 1505 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1506 | { |
5d4f98a2 YZ |
1507 | int type; |
1508 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1509 | if (parent > 0) | |
1510 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1511 | else | |
1512 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1513 | } else { | |
1514 | if (parent > 0) | |
1515 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1516 | else | |
1517 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1518 | } | |
1519 | return type; | |
31840ae1 | 1520 | } |
56bec294 | 1521 | |
2c47e605 YZ |
1522 | static int find_next_key(struct btrfs_path *path, int level, |
1523 | struct btrfs_key *key) | |
56bec294 | 1524 | |
02217ed2 | 1525 | { |
2c47e605 | 1526 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1527 | if (!path->nodes[level]) |
1528 | break; | |
5d4f98a2 YZ |
1529 | if (path->slots[level] + 1 >= |
1530 | btrfs_header_nritems(path->nodes[level])) | |
1531 | continue; | |
1532 | if (level == 0) | |
1533 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1534 | path->slots[level] + 1); | |
1535 | else | |
1536 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1537 | path->slots[level] + 1); | |
1538 | return 0; | |
1539 | } | |
1540 | return 1; | |
1541 | } | |
037e6390 | 1542 | |
5d4f98a2 YZ |
1543 | /* |
1544 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1545 | * to the address of inline back ref, and 0 is returned. | |
1546 | * | |
1547 | * if back ref isn't found, *ref_ret is set to the address where it | |
1548 | * should be inserted, and -ENOENT is returned. | |
1549 | * | |
1550 | * if insert is true and there are too many inline back refs, the path | |
1551 | * points to the extent item, and -EAGAIN is returned. | |
1552 | * | |
1553 | * NOTE: inline back refs are ordered in the same way that back ref | |
1554 | * items in the tree are ordered. | |
1555 | */ | |
1556 | static noinline_for_stack | |
1557 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1558 | struct btrfs_root *root, | |
1559 | struct btrfs_path *path, | |
1560 | struct btrfs_extent_inline_ref **ref_ret, | |
1561 | u64 bytenr, u64 num_bytes, | |
1562 | u64 parent, u64 root_objectid, | |
1563 | u64 owner, u64 offset, int insert) | |
1564 | { | |
1565 | struct btrfs_key key; | |
1566 | struct extent_buffer *leaf; | |
1567 | struct btrfs_extent_item *ei; | |
1568 | struct btrfs_extent_inline_ref *iref; | |
1569 | u64 flags; | |
1570 | u64 item_size; | |
1571 | unsigned long ptr; | |
1572 | unsigned long end; | |
1573 | int extra_size; | |
1574 | int type; | |
1575 | int want; | |
1576 | int ret; | |
1577 | int err = 0; | |
3173a18f JB |
1578 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
1579 | SKINNY_METADATA); | |
26b8003f | 1580 | |
db94535d | 1581 | key.objectid = bytenr; |
31840ae1 | 1582 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1583 | key.offset = num_bytes; |
31840ae1 | 1584 | |
5d4f98a2 YZ |
1585 | want = extent_ref_type(parent, owner); |
1586 | if (insert) { | |
1587 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1588 | path->keep_locks = 1; |
5d4f98a2 YZ |
1589 | } else |
1590 | extra_size = -1; | |
3173a18f JB |
1591 | |
1592 | /* | |
1593 | * Owner is our parent level, so we can just add one to get the level | |
1594 | * for the block we are interested in. | |
1595 | */ | |
1596 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1597 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1598 | key.offset = owner; | |
1599 | } | |
1600 | ||
1601 | again: | |
5d4f98a2 | 1602 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1603 | if (ret < 0) { |
5d4f98a2 YZ |
1604 | err = ret; |
1605 | goto out; | |
1606 | } | |
3173a18f JB |
1607 | |
1608 | /* | |
1609 | * We may be a newly converted file system which still has the old fat | |
1610 | * extent entries for metadata, so try and see if we have one of those. | |
1611 | */ | |
1612 | if (ret > 0 && skinny_metadata) { | |
1613 | skinny_metadata = false; | |
1614 | if (path->slots[0]) { | |
1615 | path->slots[0]--; | |
1616 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1617 | path->slots[0]); | |
1618 | if (key.objectid == bytenr && | |
1619 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1620 | key.offset == num_bytes) | |
1621 | ret = 0; | |
1622 | } | |
1623 | if (ret) { | |
9ce49a0b | 1624 | key.objectid = bytenr; |
3173a18f JB |
1625 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1626 | key.offset = num_bytes; | |
1627 | btrfs_release_path(path); | |
1628 | goto again; | |
1629 | } | |
1630 | } | |
1631 | ||
79787eaa JM |
1632 | if (ret && !insert) { |
1633 | err = -ENOENT; | |
1634 | goto out; | |
fae7f21c | 1635 | } else if (WARN_ON(ret)) { |
492104c8 | 1636 | err = -EIO; |
492104c8 | 1637 | goto out; |
79787eaa | 1638 | } |
5d4f98a2 YZ |
1639 | |
1640 | leaf = path->nodes[0]; | |
1641 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1642 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1643 | if (item_size < sizeof(*ei)) { | |
1644 | if (!insert) { | |
1645 | err = -ENOENT; | |
1646 | goto out; | |
1647 | } | |
1648 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1649 | extra_size); | |
1650 | if (ret < 0) { | |
1651 | err = ret; | |
1652 | goto out; | |
1653 | } | |
1654 | leaf = path->nodes[0]; | |
1655 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1656 | } | |
1657 | #endif | |
1658 | BUG_ON(item_size < sizeof(*ei)); | |
1659 | ||
5d4f98a2 YZ |
1660 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1661 | flags = btrfs_extent_flags(leaf, ei); | |
1662 | ||
1663 | ptr = (unsigned long)(ei + 1); | |
1664 | end = (unsigned long)ei + item_size; | |
1665 | ||
3173a18f | 1666 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1667 | ptr += sizeof(struct btrfs_tree_block_info); |
1668 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1669 | } |
1670 | ||
1671 | err = -ENOENT; | |
1672 | while (1) { | |
1673 | if (ptr >= end) { | |
1674 | WARN_ON(ptr > end); | |
1675 | break; | |
1676 | } | |
1677 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1678 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1679 | if (want < type) | |
1680 | break; | |
1681 | if (want > type) { | |
1682 | ptr += btrfs_extent_inline_ref_size(type); | |
1683 | continue; | |
1684 | } | |
1685 | ||
1686 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1687 | struct btrfs_extent_data_ref *dref; | |
1688 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1689 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1690 | owner, offset)) { | |
1691 | err = 0; | |
1692 | break; | |
1693 | } | |
1694 | if (hash_extent_data_ref_item(leaf, dref) < | |
1695 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1696 | break; | |
1697 | } else { | |
1698 | u64 ref_offset; | |
1699 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1700 | if (parent > 0) { | |
1701 | if (parent == ref_offset) { | |
1702 | err = 0; | |
1703 | break; | |
1704 | } | |
1705 | if (ref_offset < parent) | |
1706 | break; | |
1707 | } else { | |
1708 | if (root_objectid == ref_offset) { | |
1709 | err = 0; | |
1710 | break; | |
1711 | } | |
1712 | if (ref_offset < root_objectid) | |
1713 | break; | |
1714 | } | |
1715 | } | |
1716 | ptr += btrfs_extent_inline_ref_size(type); | |
1717 | } | |
1718 | if (err == -ENOENT && insert) { | |
1719 | if (item_size + extra_size >= | |
1720 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1721 | err = -EAGAIN; | |
1722 | goto out; | |
1723 | } | |
1724 | /* | |
1725 | * To add new inline back ref, we have to make sure | |
1726 | * there is no corresponding back ref item. | |
1727 | * For simplicity, we just do not add new inline back | |
1728 | * ref if there is any kind of item for this block | |
1729 | */ | |
2c47e605 YZ |
1730 | if (find_next_key(path, 0, &key) == 0 && |
1731 | key.objectid == bytenr && | |
85d4198e | 1732 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1733 | err = -EAGAIN; |
1734 | goto out; | |
1735 | } | |
1736 | } | |
1737 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1738 | out: | |
85d4198e | 1739 | if (insert) { |
5d4f98a2 YZ |
1740 | path->keep_locks = 0; |
1741 | btrfs_unlock_up_safe(path, 1); | |
1742 | } | |
1743 | return err; | |
1744 | } | |
1745 | ||
1746 | /* | |
1747 | * helper to add new inline back ref | |
1748 | */ | |
1749 | static noinline_for_stack | |
fd279fae | 1750 | void setup_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1751 | struct btrfs_path *path, |
1752 | struct btrfs_extent_inline_ref *iref, | |
1753 | u64 parent, u64 root_objectid, | |
1754 | u64 owner, u64 offset, int refs_to_add, | |
1755 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1756 | { |
1757 | struct extent_buffer *leaf; | |
1758 | struct btrfs_extent_item *ei; | |
1759 | unsigned long ptr; | |
1760 | unsigned long end; | |
1761 | unsigned long item_offset; | |
1762 | u64 refs; | |
1763 | int size; | |
1764 | int type; | |
5d4f98a2 YZ |
1765 | |
1766 | leaf = path->nodes[0]; | |
1767 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1768 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1769 | ||
1770 | type = extent_ref_type(parent, owner); | |
1771 | size = btrfs_extent_inline_ref_size(type); | |
1772 | ||
4b90c680 | 1773 | btrfs_extend_item(root, path, size); |
5d4f98a2 YZ |
1774 | |
1775 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1776 | refs = btrfs_extent_refs(leaf, ei); | |
1777 | refs += refs_to_add; | |
1778 | btrfs_set_extent_refs(leaf, ei, refs); | |
1779 | if (extent_op) | |
1780 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1781 | ||
1782 | ptr = (unsigned long)ei + item_offset; | |
1783 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1784 | if (ptr < end - size) | |
1785 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1786 | end - size - ptr); | |
1787 | ||
1788 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1789 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1790 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1791 | struct btrfs_extent_data_ref *dref; | |
1792 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1793 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1794 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1795 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1796 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1797 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1798 | struct btrfs_shared_data_ref *sref; | |
1799 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1800 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1801 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1802 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1803 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1804 | } else { | |
1805 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1806 | } | |
1807 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1808 | } |
1809 | ||
1810 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1811 | struct btrfs_root *root, | |
1812 | struct btrfs_path *path, | |
1813 | struct btrfs_extent_inline_ref **ref_ret, | |
1814 | u64 bytenr, u64 num_bytes, u64 parent, | |
1815 | u64 root_objectid, u64 owner, u64 offset) | |
1816 | { | |
1817 | int ret; | |
1818 | ||
1819 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1820 | bytenr, num_bytes, parent, | |
1821 | root_objectid, owner, offset, 0); | |
1822 | if (ret != -ENOENT) | |
54aa1f4d | 1823 | return ret; |
5d4f98a2 | 1824 | |
b3b4aa74 | 1825 | btrfs_release_path(path); |
5d4f98a2 YZ |
1826 | *ref_ret = NULL; |
1827 | ||
1828 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1829 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1830 | root_objectid); | |
1831 | } else { | |
1832 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1833 | root_objectid, owner, offset); | |
b9473439 | 1834 | } |
5d4f98a2 YZ |
1835 | return ret; |
1836 | } | |
31840ae1 | 1837 | |
5d4f98a2 YZ |
1838 | /* |
1839 | * helper to update/remove inline back ref | |
1840 | */ | |
1841 | static noinline_for_stack | |
afe5fea7 | 1842 | void update_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1843 | struct btrfs_path *path, |
1844 | struct btrfs_extent_inline_ref *iref, | |
1845 | int refs_to_mod, | |
fcebe456 JB |
1846 | struct btrfs_delayed_extent_op *extent_op, |
1847 | int *last_ref) | |
5d4f98a2 YZ |
1848 | { |
1849 | struct extent_buffer *leaf; | |
1850 | struct btrfs_extent_item *ei; | |
1851 | struct btrfs_extent_data_ref *dref = NULL; | |
1852 | struct btrfs_shared_data_ref *sref = NULL; | |
1853 | unsigned long ptr; | |
1854 | unsigned long end; | |
1855 | u32 item_size; | |
1856 | int size; | |
1857 | int type; | |
5d4f98a2 YZ |
1858 | u64 refs; |
1859 | ||
1860 | leaf = path->nodes[0]; | |
1861 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1862 | refs = btrfs_extent_refs(leaf, ei); | |
1863 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1864 | refs += refs_to_mod; | |
1865 | btrfs_set_extent_refs(leaf, ei, refs); | |
1866 | if (extent_op) | |
1867 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1868 | ||
1869 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1870 | ||
1871 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1872 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1873 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1874 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1875 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1876 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1877 | } else { | |
1878 | refs = 1; | |
1879 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1880 | } |
31840ae1 | 1881 | |
5d4f98a2 YZ |
1882 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1883 | refs += refs_to_mod; | |
1884 | ||
1885 | if (refs > 0) { | |
1886 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1887 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1888 | else | |
1889 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1890 | } else { | |
fcebe456 | 1891 | *last_ref = 1; |
5d4f98a2 YZ |
1892 | size = btrfs_extent_inline_ref_size(type); |
1893 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1894 | ptr = (unsigned long)iref; | |
1895 | end = (unsigned long)ei + item_size; | |
1896 | if (ptr + size < end) | |
1897 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1898 | end - ptr - size); | |
1899 | item_size -= size; | |
afe5fea7 | 1900 | btrfs_truncate_item(root, path, item_size, 1); |
5d4f98a2 YZ |
1901 | } |
1902 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1903 | } |
1904 | ||
1905 | static noinline_for_stack | |
1906 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1907 | struct btrfs_root *root, | |
1908 | struct btrfs_path *path, | |
1909 | u64 bytenr, u64 num_bytes, u64 parent, | |
1910 | u64 root_objectid, u64 owner, | |
1911 | u64 offset, int refs_to_add, | |
1912 | struct btrfs_delayed_extent_op *extent_op) | |
1913 | { | |
1914 | struct btrfs_extent_inline_ref *iref; | |
1915 | int ret; | |
1916 | ||
1917 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1918 | bytenr, num_bytes, parent, | |
1919 | root_objectid, owner, offset, 1); | |
1920 | if (ret == 0) { | |
1921 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
afe5fea7 | 1922 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1923 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1924 | } else if (ret == -ENOENT) { |
fd279fae | 1925 | setup_inline_extent_backref(root, path, iref, parent, |
143bede5 JM |
1926 | root_objectid, owner, offset, |
1927 | refs_to_add, extent_op); | |
1928 | ret = 0; | |
771ed689 | 1929 | } |
5d4f98a2 YZ |
1930 | return ret; |
1931 | } | |
31840ae1 | 1932 | |
5d4f98a2 YZ |
1933 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1934 | struct btrfs_root *root, | |
1935 | struct btrfs_path *path, | |
1936 | u64 bytenr, u64 parent, u64 root_objectid, | |
1937 | u64 owner, u64 offset, int refs_to_add) | |
1938 | { | |
1939 | int ret; | |
1940 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1941 | BUG_ON(refs_to_add != 1); | |
1942 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1943 | parent, root_objectid); | |
1944 | } else { | |
1945 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1946 | parent, root_objectid, | |
1947 | owner, offset, refs_to_add); | |
1948 | } | |
1949 | return ret; | |
1950 | } | |
56bec294 | 1951 | |
5d4f98a2 YZ |
1952 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1953 | struct btrfs_root *root, | |
1954 | struct btrfs_path *path, | |
1955 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1956 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1957 | { |
143bede5 | 1958 | int ret = 0; |
b9473439 | 1959 | |
5d4f98a2 YZ |
1960 | BUG_ON(!is_data && refs_to_drop != 1); |
1961 | if (iref) { | |
afe5fea7 | 1962 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1963 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 1964 | } else if (is_data) { |
fcebe456 JB |
1965 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop, |
1966 | last_ref); | |
5d4f98a2 | 1967 | } else { |
fcebe456 | 1968 | *last_ref = 1; |
5d4f98a2 YZ |
1969 | ret = btrfs_del_item(trans, root, path); |
1970 | } | |
1971 | return ret; | |
1972 | } | |
1973 | ||
86557861 | 1974 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
1975 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1976 | u64 *discarded_bytes) | |
5d4f98a2 | 1977 | { |
86557861 JM |
1978 | int j, ret = 0; |
1979 | u64 bytes_left, end; | |
4d89d377 | 1980 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1981 | |
4d89d377 JM |
1982 | if (WARN_ON(start != aligned_start)) { |
1983 | len -= aligned_start - start; | |
1984 | len = round_down(len, 1 << 9); | |
1985 | start = aligned_start; | |
1986 | } | |
d04c6b88 | 1987 | |
4d89d377 | 1988 | *discarded_bytes = 0; |
86557861 JM |
1989 | |
1990 | if (!len) | |
1991 | return 0; | |
1992 | ||
1993 | end = start + len; | |
1994 | bytes_left = len; | |
1995 | ||
1996 | /* Skip any superblocks on this device. */ | |
1997 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1998 | u64 sb_start = btrfs_sb_offset(j); | |
1999 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
2000 | u64 size = sb_start - start; | |
2001 | ||
2002 | if (!in_range(sb_start, start, bytes_left) && | |
2003 | !in_range(sb_end, start, bytes_left) && | |
2004 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
2005 | continue; | |
2006 | ||
2007 | /* | |
2008 | * Superblock spans beginning of range. Adjust start and | |
2009 | * try again. | |
2010 | */ | |
2011 | if (sb_start <= start) { | |
2012 | start += sb_end - start; | |
2013 | if (start > end) { | |
2014 | bytes_left = 0; | |
2015 | break; | |
2016 | } | |
2017 | bytes_left = end - start; | |
2018 | continue; | |
2019 | } | |
2020 | ||
2021 | if (size) { | |
2022 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2023 | GFP_NOFS, 0); | |
2024 | if (!ret) | |
2025 | *discarded_bytes += size; | |
2026 | else if (ret != -EOPNOTSUPP) | |
2027 | return ret; | |
2028 | } | |
2029 | ||
2030 | start = sb_end; | |
2031 | if (start > end) { | |
2032 | bytes_left = 0; | |
2033 | break; | |
2034 | } | |
2035 | bytes_left = end - start; | |
2036 | } | |
2037 | ||
2038 | if (bytes_left) { | |
2039 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2040 | GFP_NOFS, 0); |
2041 | if (!ret) | |
86557861 | 2042 | *discarded_bytes += bytes_left; |
4d89d377 | 2043 | } |
d04c6b88 | 2044 | return ret; |
5d4f98a2 | 2045 | } |
5d4f98a2 | 2046 | |
1edb647b FM |
2047 | int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, |
2048 | u64 num_bytes, u64 *actual_bytes) | |
5d4f98a2 | 2049 | { |
5d4f98a2 | 2050 | int ret; |
5378e607 | 2051 | u64 discarded_bytes = 0; |
a1d3c478 | 2052 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2053 | |
e244a0ae | 2054 | |
2999241d FM |
2055 | /* |
2056 | * Avoid races with device replace and make sure our bbio has devices | |
2057 | * associated to its stripes that don't go away while we are discarding. | |
2058 | */ | |
2059 | btrfs_bio_counter_inc_blocked(root->fs_info); | |
5d4f98a2 | 2060 | /* Tell the block device(s) that the sectors can be discarded */ |
3ec706c8 | 2061 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
a1d3c478 | 2062 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 2063 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2064 | if (!ret) { |
a1d3c478 | 2065 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2066 | int i; |
2067 | ||
5d4f98a2 | 2068 | |
a1d3c478 | 2069 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2070 | u64 bytes; |
d5e2003c JB |
2071 | if (!stripe->dev->can_discard) |
2072 | continue; | |
2073 | ||
5378e607 LD |
2074 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2075 | stripe->physical, | |
d04c6b88 JM |
2076 | stripe->length, |
2077 | &bytes); | |
5378e607 | 2078 | if (!ret) |
d04c6b88 | 2079 | discarded_bytes += bytes; |
5378e607 | 2080 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2081 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2082 | |
2083 | /* | |
2084 | * Just in case we get back EOPNOTSUPP for some reason, | |
2085 | * just ignore the return value so we don't screw up | |
2086 | * people calling discard_extent. | |
2087 | */ | |
2088 | ret = 0; | |
5d4f98a2 | 2089 | } |
6e9606d2 | 2090 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2091 | } |
2999241d | 2092 | btrfs_bio_counter_dec(root->fs_info); |
5378e607 LD |
2093 | |
2094 | if (actual_bytes) | |
2095 | *actual_bytes = discarded_bytes; | |
2096 | ||
5d4f98a2 | 2097 | |
53b381b3 DW |
2098 | if (ret == -EOPNOTSUPP) |
2099 | ret = 0; | |
5d4f98a2 | 2100 | return ret; |
5d4f98a2 YZ |
2101 | } |
2102 | ||
79787eaa | 2103 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
2104 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
2105 | struct btrfs_root *root, | |
2106 | u64 bytenr, u64 num_bytes, u64 parent, | |
b06c4bf5 | 2107 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
2108 | { |
2109 | int ret; | |
66d7e7f0 AJ |
2110 | struct btrfs_fs_info *fs_info = root->fs_info; |
2111 | ||
5d4f98a2 YZ |
2112 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2113 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2114 | ||
2115 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
2116 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
2117 | num_bytes, | |
5d4f98a2 | 2118 | parent, root_objectid, (int)owner, |
b06c4bf5 | 2119 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 | 2120 | } else { |
66d7e7f0 | 2121 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5846a3c2 QW |
2122 | num_bytes, parent, root_objectid, |
2123 | owner, offset, 0, | |
b06c4bf5 | 2124 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 YZ |
2125 | } |
2126 | return ret; | |
2127 | } | |
2128 | ||
2129 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2130 | struct btrfs_root *root, | |
c682f9b3 | 2131 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2132 | u64 parent, u64 root_objectid, |
2133 | u64 owner, u64 offset, int refs_to_add, | |
2134 | struct btrfs_delayed_extent_op *extent_op) | |
2135 | { | |
fcebe456 | 2136 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
2137 | struct btrfs_path *path; |
2138 | struct extent_buffer *leaf; | |
2139 | struct btrfs_extent_item *item; | |
fcebe456 | 2140 | struct btrfs_key key; |
c682f9b3 QW |
2141 | u64 bytenr = node->bytenr; |
2142 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2143 | u64 refs; |
2144 | int ret; | |
5d4f98a2 YZ |
2145 | |
2146 | path = btrfs_alloc_path(); | |
2147 | if (!path) | |
2148 | return -ENOMEM; | |
2149 | ||
e4058b54 | 2150 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2151 | path->leave_spinning = 1; |
2152 | /* this will setup the path even if it fails to insert the back ref */ | |
fcebe456 JB |
2153 | ret = insert_inline_extent_backref(trans, fs_info->extent_root, path, |
2154 | bytenr, num_bytes, parent, | |
5d4f98a2 YZ |
2155 | root_objectid, owner, offset, |
2156 | refs_to_add, extent_op); | |
0ed4792a | 2157 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2158 | goto out; |
fcebe456 JB |
2159 | |
2160 | /* | |
2161 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2162 | * inline extent ref, so just update the reference count and add a | |
2163 | * normal backref. | |
2164 | */ | |
5d4f98a2 | 2165 | leaf = path->nodes[0]; |
fcebe456 | 2166 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2167 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2168 | refs = btrfs_extent_refs(leaf, item); | |
2169 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2170 | if (extent_op) | |
2171 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2172 | |
5d4f98a2 | 2173 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2174 | btrfs_release_path(path); |
56bec294 | 2175 | |
e4058b54 | 2176 | path->reada = READA_FORWARD; |
b9473439 | 2177 | path->leave_spinning = 1; |
56bec294 CM |
2178 | /* now insert the actual backref */ |
2179 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
2180 | path, bytenr, parent, root_objectid, |
2181 | owner, offset, refs_to_add); | |
79787eaa JM |
2182 | if (ret) |
2183 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 2184 | out: |
56bec294 | 2185 | btrfs_free_path(path); |
30d133fc | 2186 | return ret; |
56bec294 CM |
2187 | } |
2188 | ||
5d4f98a2 YZ |
2189 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2190 | struct btrfs_root *root, | |
2191 | struct btrfs_delayed_ref_node *node, | |
2192 | struct btrfs_delayed_extent_op *extent_op, | |
2193 | int insert_reserved) | |
56bec294 | 2194 | { |
5d4f98a2 YZ |
2195 | int ret = 0; |
2196 | struct btrfs_delayed_data_ref *ref; | |
2197 | struct btrfs_key ins; | |
2198 | u64 parent = 0; | |
2199 | u64 ref_root = 0; | |
2200 | u64 flags = 0; | |
2201 | ||
2202 | ins.objectid = node->bytenr; | |
2203 | ins.offset = node->num_bytes; | |
2204 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2205 | ||
2206 | ref = btrfs_delayed_node_to_data_ref(node); | |
599c75ec LB |
2207 | trace_run_delayed_data_ref(node, ref, node->action); |
2208 | ||
5d4f98a2 YZ |
2209 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2210 | parent = ref->parent; | |
fcebe456 | 2211 | ref_root = ref->root; |
5d4f98a2 YZ |
2212 | |
2213 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2214 | if (extent_op) |
5d4f98a2 | 2215 | flags |= extent_op->flags_to_set; |
5d4f98a2 YZ |
2216 | ret = alloc_reserved_file_extent(trans, root, |
2217 | parent, ref_root, flags, | |
2218 | ref->objectid, ref->offset, | |
2219 | &ins, node->ref_mod); | |
5d4f98a2 | 2220 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
c682f9b3 | 2221 | ret = __btrfs_inc_extent_ref(trans, root, node, parent, |
5d4f98a2 YZ |
2222 | ref_root, ref->objectid, |
2223 | ref->offset, node->ref_mod, | |
c682f9b3 | 2224 | extent_op); |
5d4f98a2 | 2225 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
c682f9b3 | 2226 | ret = __btrfs_free_extent(trans, root, node, parent, |
5d4f98a2 YZ |
2227 | ref_root, ref->objectid, |
2228 | ref->offset, node->ref_mod, | |
c682f9b3 | 2229 | extent_op); |
5d4f98a2 YZ |
2230 | } else { |
2231 | BUG(); | |
2232 | } | |
2233 | return ret; | |
2234 | } | |
2235 | ||
2236 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2237 | struct extent_buffer *leaf, | |
2238 | struct btrfs_extent_item *ei) | |
2239 | { | |
2240 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2241 | if (extent_op->update_flags) { | |
2242 | flags |= extent_op->flags_to_set; | |
2243 | btrfs_set_extent_flags(leaf, ei, flags); | |
2244 | } | |
2245 | ||
2246 | if (extent_op->update_key) { | |
2247 | struct btrfs_tree_block_info *bi; | |
2248 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2249 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2250 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2251 | } | |
2252 | } | |
2253 | ||
2254 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2255 | struct btrfs_root *root, | |
2256 | struct btrfs_delayed_ref_node *node, | |
2257 | struct btrfs_delayed_extent_op *extent_op) | |
2258 | { | |
2259 | struct btrfs_key key; | |
2260 | struct btrfs_path *path; | |
2261 | struct btrfs_extent_item *ei; | |
2262 | struct extent_buffer *leaf; | |
2263 | u32 item_size; | |
56bec294 | 2264 | int ret; |
5d4f98a2 | 2265 | int err = 0; |
b1c79e09 | 2266 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2267 | |
79787eaa JM |
2268 | if (trans->aborted) |
2269 | return 0; | |
2270 | ||
3173a18f JB |
2271 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
2272 | metadata = 0; | |
2273 | ||
5d4f98a2 YZ |
2274 | path = btrfs_alloc_path(); |
2275 | if (!path) | |
2276 | return -ENOMEM; | |
2277 | ||
2278 | key.objectid = node->bytenr; | |
5d4f98a2 | 2279 | |
3173a18f | 2280 | if (metadata) { |
3173a18f | 2281 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2282 | key.offset = extent_op->level; |
3173a18f JB |
2283 | } else { |
2284 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2285 | key.offset = node->num_bytes; | |
2286 | } | |
2287 | ||
2288 | again: | |
e4058b54 | 2289 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2290 | path->leave_spinning = 1; |
2291 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2292 | path, 0, 1); | |
2293 | if (ret < 0) { | |
2294 | err = ret; | |
2295 | goto out; | |
2296 | } | |
2297 | if (ret > 0) { | |
3173a18f | 2298 | if (metadata) { |
55994887 FDBM |
2299 | if (path->slots[0] > 0) { |
2300 | path->slots[0]--; | |
2301 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2302 | path->slots[0]); | |
2303 | if (key.objectid == node->bytenr && | |
2304 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2305 | key.offset == node->num_bytes) | |
2306 | ret = 0; | |
2307 | } | |
2308 | if (ret > 0) { | |
2309 | btrfs_release_path(path); | |
2310 | metadata = 0; | |
3173a18f | 2311 | |
55994887 FDBM |
2312 | key.objectid = node->bytenr; |
2313 | key.offset = node->num_bytes; | |
2314 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2315 | goto again; | |
2316 | } | |
2317 | } else { | |
2318 | err = -EIO; | |
2319 | goto out; | |
3173a18f | 2320 | } |
5d4f98a2 YZ |
2321 | } |
2322 | ||
2323 | leaf = path->nodes[0]; | |
2324 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2325 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2326 | if (item_size < sizeof(*ei)) { | |
2327 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2328 | path, (u64)-1, 0); | |
2329 | if (ret < 0) { | |
2330 | err = ret; | |
2331 | goto out; | |
2332 | } | |
2333 | leaf = path->nodes[0]; | |
2334 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2335 | } | |
2336 | #endif | |
2337 | BUG_ON(item_size < sizeof(*ei)); | |
2338 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2339 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2340 | |
5d4f98a2 YZ |
2341 | btrfs_mark_buffer_dirty(leaf); |
2342 | out: | |
2343 | btrfs_free_path(path); | |
2344 | return err; | |
56bec294 CM |
2345 | } |
2346 | ||
5d4f98a2 YZ |
2347 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2348 | struct btrfs_root *root, | |
2349 | struct btrfs_delayed_ref_node *node, | |
2350 | struct btrfs_delayed_extent_op *extent_op, | |
2351 | int insert_reserved) | |
56bec294 CM |
2352 | { |
2353 | int ret = 0; | |
5d4f98a2 YZ |
2354 | struct btrfs_delayed_tree_ref *ref; |
2355 | struct btrfs_key ins; | |
2356 | u64 parent = 0; | |
2357 | u64 ref_root = 0; | |
3173a18f JB |
2358 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
2359 | SKINNY_METADATA); | |
56bec294 | 2360 | |
5d4f98a2 | 2361 | ref = btrfs_delayed_node_to_tree_ref(node); |
599c75ec LB |
2362 | trace_run_delayed_tree_ref(node, ref, node->action); |
2363 | ||
5d4f98a2 YZ |
2364 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2365 | parent = ref->parent; | |
fcebe456 | 2366 | ref_root = ref->root; |
5d4f98a2 | 2367 | |
3173a18f JB |
2368 | ins.objectid = node->bytenr; |
2369 | if (skinny_metadata) { | |
2370 | ins.offset = ref->level; | |
2371 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2372 | } else { | |
2373 | ins.offset = node->num_bytes; | |
2374 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2375 | } | |
2376 | ||
5d4f98a2 YZ |
2377 | BUG_ON(node->ref_mod != 1); |
2378 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2379 | BUG_ON(!extent_op || !extent_op->update_flags); |
5d4f98a2 YZ |
2380 | ret = alloc_reserved_tree_block(trans, root, |
2381 | parent, ref_root, | |
2382 | extent_op->flags_to_set, | |
2383 | &extent_op->key, | |
b06c4bf5 | 2384 | ref->level, &ins); |
5d4f98a2 | 2385 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
c682f9b3 QW |
2386 | ret = __btrfs_inc_extent_ref(trans, root, node, |
2387 | parent, ref_root, | |
2388 | ref->level, 0, 1, | |
fcebe456 | 2389 | extent_op); |
5d4f98a2 | 2390 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
c682f9b3 QW |
2391 | ret = __btrfs_free_extent(trans, root, node, |
2392 | parent, ref_root, | |
2393 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2394 | } else { |
2395 | BUG(); | |
2396 | } | |
56bec294 CM |
2397 | return ret; |
2398 | } | |
2399 | ||
2400 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2401 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2402 | struct btrfs_root *root, | |
2403 | struct btrfs_delayed_ref_node *node, | |
2404 | struct btrfs_delayed_extent_op *extent_op, | |
2405 | int insert_reserved) | |
56bec294 | 2406 | { |
79787eaa JM |
2407 | int ret = 0; |
2408 | ||
857cc2fc JB |
2409 | if (trans->aborted) { |
2410 | if (insert_reserved) | |
2411 | btrfs_pin_extent(root, node->bytenr, | |
2412 | node->num_bytes, 1); | |
79787eaa | 2413 | return 0; |
857cc2fc | 2414 | } |
79787eaa | 2415 | |
5d4f98a2 | 2416 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2417 | struct btrfs_delayed_ref_head *head; |
2418 | /* | |
2419 | * we've hit the end of the chain and we were supposed | |
2420 | * to insert this extent into the tree. But, it got | |
2421 | * deleted before we ever needed to insert it, so all | |
2422 | * we have to do is clean up the accounting | |
2423 | */ | |
5d4f98a2 YZ |
2424 | BUG_ON(extent_op); |
2425 | head = btrfs_delayed_node_to_head(node); | |
599c75ec LB |
2426 | trace_run_delayed_ref_head(node, head, node->action); |
2427 | ||
56bec294 | 2428 | if (insert_reserved) { |
f0486c68 YZ |
2429 | btrfs_pin_extent(root, node->bytenr, |
2430 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2431 | if (head->is_data) { |
2432 | ret = btrfs_del_csums(trans, root, | |
2433 | node->bytenr, | |
2434 | node->num_bytes); | |
5d4f98a2 | 2435 | } |
56bec294 | 2436 | } |
297d750b QW |
2437 | |
2438 | /* Also free its reserved qgroup space */ | |
2439 | btrfs_qgroup_free_delayed_ref(root->fs_info, | |
2440 | head->qgroup_ref_root, | |
2441 | head->qgroup_reserved); | |
79787eaa | 2442 | return ret; |
56bec294 CM |
2443 | } |
2444 | ||
5d4f98a2 YZ |
2445 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2446 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2447 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2448 | insert_reserved); | |
2449 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2450 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2451 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2452 | insert_reserved); | |
2453 | else | |
2454 | BUG(); | |
2455 | return ret; | |
56bec294 CM |
2456 | } |
2457 | ||
c6fc2454 | 2458 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2459 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2460 | { | |
cffc3374 FM |
2461 | struct btrfs_delayed_ref_node *ref; |
2462 | ||
c6fc2454 QW |
2463 | if (list_empty(&head->ref_list)) |
2464 | return NULL; | |
d7df2c79 | 2465 | |
cffc3374 FM |
2466 | /* |
2467 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2468 | * This is to prevent a ref count from going down to zero, which deletes | |
2469 | * the extent item from the extent tree, when there still are references | |
2470 | * to add, which would fail because they would not find the extent item. | |
2471 | */ | |
2472 | list_for_each_entry(ref, &head->ref_list, list) { | |
2473 | if (ref->action == BTRFS_ADD_DELAYED_REF) | |
2474 | return ref; | |
2475 | } | |
2476 | ||
c6fc2454 QW |
2477 | return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node, |
2478 | list); | |
56bec294 CM |
2479 | } |
2480 | ||
79787eaa JM |
2481 | /* |
2482 | * Returns 0 on success or if called with an already aborted transaction. | |
2483 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2484 | */ | |
d7df2c79 JB |
2485 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2486 | struct btrfs_root *root, | |
2487 | unsigned long nr) | |
56bec294 | 2488 | { |
56bec294 CM |
2489 | struct btrfs_delayed_ref_root *delayed_refs; |
2490 | struct btrfs_delayed_ref_node *ref; | |
2491 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2492 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2493 | struct btrfs_fs_info *fs_info = root->fs_info; |
0a2b2a84 | 2494 | ktime_t start = ktime_get(); |
56bec294 | 2495 | int ret; |
d7df2c79 | 2496 | unsigned long count = 0; |
0a2b2a84 | 2497 | unsigned long actual_count = 0; |
56bec294 | 2498 | int must_insert_reserved = 0; |
56bec294 CM |
2499 | |
2500 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2501 | while (1) { |
2502 | if (!locked_ref) { | |
d7df2c79 | 2503 | if (count >= nr) |
56bec294 | 2504 | break; |
56bec294 | 2505 | |
d7df2c79 JB |
2506 | spin_lock(&delayed_refs->lock); |
2507 | locked_ref = btrfs_select_ref_head(trans); | |
2508 | if (!locked_ref) { | |
2509 | spin_unlock(&delayed_refs->lock); | |
2510 | break; | |
2511 | } | |
c3e69d58 CM |
2512 | |
2513 | /* grab the lock that says we are going to process | |
2514 | * all the refs for this head */ | |
2515 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2516 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2517 | /* |
2518 | * we may have dropped the spin lock to get the head | |
2519 | * mutex lock, and that might have given someone else | |
2520 | * time to free the head. If that's true, it has been | |
2521 | * removed from our list and we can move on. | |
2522 | */ | |
2523 | if (ret == -EAGAIN) { | |
2524 | locked_ref = NULL; | |
2525 | count++; | |
2526 | continue; | |
56bec294 CM |
2527 | } |
2528 | } | |
a28ec197 | 2529 | |
2c3cf7d5 FM |
2530 | /* |
2531 | * We need to try and merge add/drops of the same ref since we | |
2532 | * can run into issues with relocate dropping the implicit ref | |
2533 | * and then it being added back again before the drop can | |
2534 | * finish. If we merged anything we need to re-loop so we can | |
2535 | * get a good ref. | |
2536 | * Or we can get node references of the same type that weren't | |
2537 | * merged when created due to bumps in the tree mod seq, and | |
2538 | * we need to merge them to prevent adding an inline extent | |
2539 | * backref before dropping it (triggering a BUG_ON at | |
2540 | * insert_inline_extent_backref()). | |
2541 | */ | |
d7df2c79 | 2542 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2543 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2544 | locked_ref); | |
ae1e206b | 2545 | |
d1270cd9 AJ |
2546 | /* |
2547 | * locked_ref is the head node, so we have to go one | |
2548 | * node back for any delayed ref updates | |
2549 | */ | |
2550 | ref = select_delayed_ref(locked_ref); | |
2551 | ||
2552 | if (ref && ref->seq && | |
097b8a7c | 2553 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2554 | spin_unlock(&locked_ref->lock); |
093486c4 | 2555 | btrfs_delayed_ref_unlock(locked_ref); |
d7df2c79 JB |
2556 | spin_lock(&delayed_refs->lock); |
2557 | locked_ref->processing = 0; | |
d1270cd9 AJ |
2558 | delayed_refs->num_heads_ready++; |
2559 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 2560 | locked_ref = NULL; |
d1270cd9 | 2561 | cond_resched(); |
27a377db | 2562 | count++; |
d1270cd9 AJ |
2563 | continue; |
2564 | } | |
2565 | ||
56bec294 CM |
2566 | /* |
2567 | * record the must insert reserved flag before we | |
2568 | * drop the spin lock. | |
2569 | */ | |
2570 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2571 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2572 | |
5d4f98a2 YZ |
2573 | extent_op = locked_ref->extent_op; |
2574 | locked_ref->extent_op = NULL; | |
2575 | ||
56bec294 | 2576 | if (!ref) { |
d7df2c79 JB |
2577 | |
2578 | ||
56bec294 CM |
2579 | /* All delayed refs have been processed, Go ahead |
2580 | * and send the head node to run_one_delayed_ref, | |
2581 | * so that any accounting fixes can happen | |
2582 | */ | |
2583 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2584 | |
2585 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2586 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2587 | extent_op = NULL; |
2588 | } | |
2589 | ||
2590 | if (extent_op) { | |
d7df2c79 | 2591 | spin_unlock(&locked_ref->lock); |
5d4f98a2 YZ |
2592 | ret = run_delayed_extent_op(trans, root, |
2593 | ref, extent_op); | |
78a6184a | 2594 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2595 | |
79787eaa | 2596 | if (ret) { |
857cc2fc JB |
2597 | /* |
2598 | * Need to reset must_insert_reserved if | |
2599 | * there was an error so the abort stuff | |
2600 | * can cleanup the reserved space | |
2601 | * properly. | |
2602 | */ | |
2603 | if (must_insert_reserved) | |
2604 | locked_ref->must_insert_reserved = 1; | |
d7df2c79 | 2605 | locked_ref->processing = 0; |
c2cf52eb | 2606 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); |
093486c4 | 2607 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2608 | return ret; |
2609 | } | |
d7df2c79 | 2610 | continue; |
5d4f98a2 | 2611 | } |
02217ed2 | 2612 | |
d7df2c79 | 2613 | /* |
01327610 | 2614 | * Need to drop our head ref lock and re-acquire the |
d7df2c79 JB |
2615 | * delayed ref lock and then re-check to make sure |
2616 | * nobody got added. | |
2617 | */ | |
2618 | spin_unlock(&locked_ref->lock); | |
2619 | spin_lock(&delayed_refs->lock); | |
2620 | spin_lock(&locked_ref->lock); | |
c6fc2454 | 2621 | if (!list_empty(&locked_ref->ref_list) || |
573a0755 | 2622 | locked_ref->extent_op) { |
d7df2c79 JB |
2623 | spin_unlock(&locked_ref->lock); |
2624 | spin_unlock(&delayed_refs->lock); | |
2625 | continue; | |
2626 | } | |
2627 | ref->in_tree = 0; | |
2628 | delayed_refs->num_heads--; | |
c46effa6 LB |
2629 | rb_erase(&locked_ref->href_node, |
2630 | &delayed_refs->href_root); | |
d7df2c79 JB |
2631 | spin_unlock(&delayed_refs->lock); |
2632 | } else { | |
0a2b2a84 | 2633 | actual_count++; |
d7df2c79 | 2634 | ref->in_tree = 0; |
c6fc2454 | 2635 | list_del(&ref->list); |
c46effa6 | 2636 | } |
d7df2c79 JB |
2637 | atomic_dec(&delayed_refs->num_entries); |
2638 | ||
093486c4 | 2639 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2640 | /* |
2641 | * when we play the delayed ref, also correct the | |
2642 | * ref_mod on head | |
2643 | */ | |
2644 | switch (ref->action) { | |
2645 | case BTRFS_ADD_DELAYED_REF: | |
2646 | case BTRFS_ADD_DELAYED_EXTENT: | |
2647 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2648 | break; | |
2649 | case BTRFS_DROP_DELAYED_REF: | |
2650 | locked_ref->node.ref_mod += ref->ref_mod; | |
2651 | break; | |
2652 | default: | |
2653 | WARN_ON(1); | |
2654 | } | |
2655 | } | |
d7df2c79 | 2656 | spin_unlock(&locked_ref->lock); |
925baedd | 2657 | |
5d4f98a2 | 2658 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2659 | must_insert_reserved); |
eb099670 | 2660 | |
78a6184a | 2661 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2662 | if (ret) { |
d7df2c79 | 2663 | locked_ref->processing = 0; |
093486c4 MX |
2664 | btrfs_delayed_ref_unlock(locked_ref); |
2665 | btrfs_put_delayed_ref(ref); | |
c2cf52eb | 2666 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret); |
79787eaa JM |
2667 | return ret; |
2668 | } | |
2669 | ||
093486c4 MX |
2670 | /* |
2671 | * If this node is a head, that means all the refs in this head | |
2672 | * have been dealt with, and we will pick the next head to deal | |
2673 | * with, so we must unlock the head and drop it from the cluster | |
2674 | * list before we release it. | |
2675 | */ | |
2676 | if (btrfs_delayed_ref_is_head(ref)) { | |
1262133b JB |
2677 | if (locked_ref->is_data && |
2678 | locked_ref->total_ref_mod < 0) { | |
2679 | spin_lock(&delayed_refs->lock); | |
2680 | delayed_refs->pending_csums -= ref->num_bytes; | |
2681 | spin_unlock(&delayed_refs->lock); | |
2682 | } | |
093486c4 MX |
2683 | btrfs_delayed_ref_unlock(locked_ref); |
2684 | locked_ref = NULL; | |
2685 | } | |
2686 | btrfs_put_delayed_ref(ref); | |
2687 | count++; | |
c3e69d58 | 2688 | cond_resched(); |
c3e69d58 | 2689 | } |
0a2b2a84 JB |
2690 | |
2691 | /* | |
2692 | * We don't want to include ref heads since we can have empty ref heads | |
2693 | * and those will drastically skew our runtime down since we just do | |
2694 | * accounting, no actual extent tree updates. | |
2695 | */ | |
2696 | if (actual_count > 0) { | |
2697 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2698 | u64 avg; | |
2699 | ||
2700 | /* | |
2701 | * We weigh the current average higher than our current runtime | |
2702 | * to avoid large swings in the average. | |
2703 | */ | |
2704 | spin_lock(&delayed_refs->lock); | |
2705 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2706 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2707 | spin_unlock(&delayed_refs->lock); |
2708 | } | |
d7df2c79 | 2709 | return 0; |
c3e69d58 CM |
2710 | } |
2711 | ||
709c0486 AJ |
2712 | #ifdef SCRAMBLE_DELAYED_REFS |
2713 | /* | |
2714 | * Normally delayed refs get processed in ascending bytenr order. This | |
2715 | * correlates in most cases to the order added. To expose dependencies on this | |
2716 | * order, we start to process the tree in the middle instead of the beginning | |
2717 | */ | |
2718 | static u64 find_middle(struct rb_root *root) | |
2719 | { | |
2720 | struct rb_node *n = root->rb_node; | |
2721 | struct btrfs_delayed_ref_node *entry; | |
2722 | int alt = 1; | |
2723 | u64 middle; | |
2724 | u64 first = 0, last = 0; | |
2725 | ||
2726 | n = rb_first(root); | |
2727 | if (n) { | |
2728 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2729 | first = entry->bytenr; | |
2730 | } | |
2731 | n = rb_last(root); | |
2732 | if (n) { | |
2733 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2734 | last = entry->bytenr; | |
2735 | } | |
2736 | n = root->rb_node; | |
2737 | ||
2738 | while (n) { | |
2739 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2740 | WARN_ON(!entry->in_tree); | |
2741 | ||
2742 | middle = entry->bytenr; | |
2743 | ||
2744 | if (alt) | |
2745 | n = n->rb_left; | |
2746 | else | |
2747 | n = n->rb_right; | |
2748 | ||
2749 | alt = 1 - alt; | |
2750 | } | |
2751 | return middle; | |
2752 | } | |
2753 | #endif | |
2754 | ||
1be41b78 JB |
2755 | static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads) |
2756 | { | |
2757 | u64 num_bytes; | |
2758 | ||
2759 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2760 | sizeof(struct btrfs_extent_inline_ref)); | |
2761 | if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) | |
2762 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); | |
2763 | ||
2764 | /* | |
2765 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2766 | * closer to what we're really going to want to use. |
1be41b78 | 2767 | */ |
f8c269d7 | 2768 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root)); |
1be41b78 JB |
2769 | } |
2770 | ||
1262133b JB |
2771 | /* |
2772 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2773 | * would require to store the csums for that many bytes. | |
2774 | */ | |
28f75a0e | 2775 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes) |
1262133b JB |
2776 | { |
2777 | u64 csum_size; | |
2778 | u64 num_csums_per_leaf; | |
2779 | u64 num_csums; | |
2780 | ||
2781 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
2782 | num_csums_per_leaf = div64_u64(csum_size, | |
2783 | (u64)btrfs_super_csum_size(root->fs_info->super_copy)); | |
2784 | num_csums = div64_u64(csum_bytes, root->sectorsize); | |
2785 | num_csums += num_csums_per_leaf - 1; | |
2786 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2787 | return num_csums; | |
2788 | } | |
2789 | ||
0a2b2a84 | 2790 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
1be41b78 JB |
2791 | struct btrfs_root *root) |
2792 | { | |
2793 | struct btrfs_block_rsv *global_rsv; | |
2794 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2795 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
cb723e49 JB |
2796 | u64 num_dirty_bgs = trans->transaction->num_dirty_bgs; |
2797 | u64 num_bytes, num_dirty_bgs_bytes; | |
1be41b78 JB |
2798 | int ret = 0; |
2799 | ||
2800 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
2801 | num_heads = heads_to_leaves(root, num_heads); | |
2802 | if (num_heads > 1) | |
707e8a07 | 2803 | num_bytes += (num_heads - 1) * root->nodesize; |
1be41b78 | 2804 | num_bytes <<= 1; |
28f75a0e | 2805 | num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize; |
cb723e49 JB |
2806 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root, |
2807 | num_dirty_bgs); | |
1be41b78 JB |
2808 | global_rsv = &root->fs_info->global_block_rsv; |
2809 | ||
2810 | /* | |
2811 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2812 | * wiggle room since running delayed refs can create more delayed refs. | |
2813 | */ | |
cb723e49 JB |
2814 | if (global_rsv->space_info->full) { |
2815 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2816 | num_bytes <<= 1; |
cb723e49 | 2817 | } |
1be41b78 JB |
2818 | |
2819 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2820 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2821 | ret = 1; |
2822 | spin_unlock(&global_rsv->lock); | |
2823 | return ret; | |
2824 | } | |
2825 | ||
0a2b2a84 JB |
2826 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2827 | struct btrfs_root *root) | |
2828 | { | |
2829 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2830 | u64 num_entries = | |
2831 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2832 | u64 avg_runtime; | |
a79b7d4b | 2833 | u64 val; |
0a2b2a84 JB |
2834 | |
2835 | smp_mb(); | |
2836 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2837 | val = num_entries * avg_runtime; |
0a2b2a84 JB |
2838 | if (num_entries * avg_runtime >= NSEC_PER_SEC) |
2839 | return 1; | |
a79b7d4b CM |
2840 | if (val >= NSEC_PER_SEC / 2) |
2841 | return 2; | |
0a2b2a84 JB |
2842 | |
2843 | return btrfs_check_space_for_delayed_refs(trans, root); | |
2844 | } | |
2845 | ||
a79b7d4b CM |
2846 | struct async_delayed_refs { |
2847 | struct btrfs_root *root; | |
31b9655f | 2848 | u64 transid; |
a79b7d4b CM |
2849 | int count; |
2850 | int error; | |
2851 | int sync; | |
2852 | struct completion wait; | |
2853 | struct btrfs_work work; | |
2854 | }; | |
2855 | ||
2856 | static void delayed_ref_async_start(struct btrfs_work *work) | |
2857 | { | |
2858 | struct async_delayed_refs *async; | |
2859 | struct btrfs_trans_handle *trans; | |
2860 | int ret; | |
2861 | ||
2862 | async = container_of(work, struct async_delayed_refs, work); | |
2863 | ||
0f873eca CM |
2864 | /* if the commit is already started, we don't need to wait here */ |
2865 | if (btrfs_transaction_blocked(async->root->fs_info)) | |
31b9655f | 2866 | goto done; |
31b9655f | 2867 | |
0f873eca CM |
2868 | trans = btrfs_join_transaction(async->root); |
2869 | if (IS_ERR(trans)) { | |
2870 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2871 | goto done; |
2872 | } | |
2873 | ||
2874 | /* | |
01327610 | 2875 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2876 | * wait on delayed refs |
2877 | */ | |
2878 | trans->sync = true; | |
0f873eca CM |
2879 | |
2880 | /* Don't bother flushing if we got into a different transaction */ | |
2881 | if (trans->transid > async->transid) | |
2882 | goto end; | |
2883 | ||
a79b7d4b CM |
2884 | ret = btrfs_run_delayed_refs(trans, async->root, async->count); |
2885 | if (ret) | |
2886 | async->error = ret; | |
0f873eca | 2887 | end: |
a79b7d4b CM |
2888 | ret = btrfs_end_transaction(trans, async->root); |
2889 | if (ret && !async->error) | |
2890 | async->error = ret; | |
2891 | done: | |
2892 | if (async->sync) | |
2893 | complete(&async->wait); | |
2894 | else | |
2895 | kfree(async); | |
2896 | } | |
2897 | ||
2898 | int btrfs_async_run_delayed_refs(struct btrfs_root *root, | |
31b9655f | 2899 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
2900 | { |
2901 | struct async_delayed_refs *async; | |
2902 | int ret; | |
2903 | ||
2904 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2905 | if (!async) | |
2906 | return -ENOMEM; | |
2907 | ||
2908 | async->root = root->fs_info->tree_root; | |
2909 | async->count = count; | |
2910 | async->error = 0; | |
31b9655f | 2911 | async->transid = transid; |
a79b7d4b CM |
2912 | if (wait) |
2913 | async->sync = 1; | |
2914 | else | |
2915 | async->sync = 0; | |
2916 | init_completion(&async->wait); | |
2917 | ||
9e0af237 LB |
2918 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
2919 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b CM |
2920 | |
2921 | btrfs_queue_work(root->fs_info->extent_workers, &async->work); | |
2922 | ||
2923 | if (wait) { | |
2924 | wait_for_completion(&async->wait); | |
2925 | ret = async->error; | |
2926 | kfree(async); | |
2927 | return ret; | |
2928 | } | |
2929 | return 0; | |
2930 | } | |
2931 | ||
c3e69d58 CM |
2932 | /* |
2933 | * this starts processing the delayed reference count updates and | |
2934 | * extent insertions we have queued up so far. count can be | |
2935 | * 0, which means to process everything in the tree at the start | |
2936 | * of the run (but not newly added entries), or it can be some target | |
2937 | * number you'd like to process. | |
79787eaa JM |
2938 | * |
2939 | * Returns 0 on success or if called with an aborted transaction | |
2940 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2941 | */ |
2942 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2943 | struct btrfs_root *root, unsigned long count) | |
2944 | { | |
2945 | struct rb_node *node; | |
2946 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2947 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2948 | int ret; |
2949 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 2950 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 2951 | |
79787eaa JM |
2952 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2953 | if (trans->aborted) | |
2954 | return 0; | |
2955 | ||
511711af CM |
2956 | if (root->fs_info->creating_free_space_tree) |
2957 | return 0; | |
2958 | ||
c3e69d58 CM |
2959 | if (root == root->fs_info->extent_root) |
2960 | root = root->fs_info->tree_root; | |
2961 | ||
2962 | delayed_refs = &trans->transaction->delayed_refs; | |
26455d33 | 2963 | if (count == 0) |
d7df2c79 | 2964 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2965 | |
c3e69d58 | 2966 | again: |
709c0486 AJ |
2967 | #ifdef SCRAMBLE_DELAYED_REFS |
2968 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2969 | #endif | |
d9a0540a | 2970 | trans->can_flush_pending_bgs = false; |
d7df2c79 JB |
2971 | ret = __btrfs_run_delayed_refs(trans, root, count); |
2972 | if (ret < 0) { | |
2973 | btrfs_abort_transaction(trans, root, ret); | |
2974 | return ret; | |
eb099670 | 2975 | } |
c3e69d58 | 2976 | |
56bec294 | 2977 | if (run_all) { |
d7df2c79 | 2978 | if (!list_empty(&trans->new_bgs)) |
ea658bad | 2979 | btrfs_create_pending_block_groups(trans, root); |
ea658bad | 2980 | |
d7df2c79 | 2981 | spin_lock(&delayed_refs->lock); |
c46effa6 | 2982 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
2983 | if (!node) { |
2984 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2985 | goto out; |
d7df2c79 | 2986 | } |
c3e69d58 | 2987 | count = (unsigned long)-1; |
e9d0b13b | 2988 | |
56bec294 | 2989 | while (node) { |
c46effa6 LB |
2990 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2991 | href_node); | |
2992 | if (btrfs_delayed_ref_is_head(&head->node)) { | |
2993 | struct btrfs_delayed_ref_node *ref; | |
5caf2a00 | 2994 | |
c46effa6 | 2995 | ref = &head->node; |
56bec294 CM |
2996 | atomic_inc(&ref->refs); |
2997 | ||
2998 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2999 | /* |
3000 | * Mutex was contended, block until it's | |
3001 | * released and try again | |
3002 | */ | |
56bec294 CM |
3003 | mutex_lock(&head->mutex); |
3004 | mutex_unlock(&head->mutex); | |
3005 | ||
3006 | btrfs_put_delayed_ref(ref); | |
1887be66 | 3007 | cond_resched(); |
56bec294 | 3008 | goto again; |
c46effa6 LB |
3009 | } else { |
3010 | WARN_ON(1); | |
56bec294 CM |
3011 | } |
3012 | node = rb_next(node); | |
3013 | } | |
3014 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 3015 | cond_resched(); |
56bec294 | 3016 | goto again; |
5f39d397 | 3017 | } |
54aa1f4d | 3018 | out: |
edf39272 | 3019 | assert_qgroups_uptodate(trans); |
d9a0540a | 3020 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3021 | return 0; |
3022 | } | |
3023 | ||
5d4f98a2 YZ |
3024 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
3025 | struct btrfs_root *root, | |
3026 | u64 bytenr, u64 num_bytes, u64 flags, | |
b1c79e09 | 3027 | int level, int is_data) |
5d4f98a2 YZ |
3028 | { |
3029 | struct btrfs_delayed_extent_op *extent_op; | |
3030 | int ret; | |
3031 | ||
78a6184a | 3032 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3033 | if (!extent_op) |
3034 | return -ENOMEM; | |
3035 | ||
3036 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3037 | extent_op->update_flags = true; |
3038 | extent_op->update_key = false; | |
3039 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3040 | extent_op->level = level; |
5d4f98a2 | 3041 | |
66d7e7f0 AJ |
3042 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
3043 | num_bytes, extent_op); | |
5d4f98a2 | 3044 | if (ret) |
78a6184a | 3045 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3046 | return ret; |
3047 | } | |
3048 | ||
3049 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
3050 | struct btrfs_root *root, | |
3051 | struct btrfs_path *path, | |
3052 | u64 objectid, u64 offset, u64 bytenr) | |
3053 | { | |
3054 | struct btrfs_delayed_ref_head *head; | |
3055 | struct btrfs_delayed_ref_node *ref; | |
3056 | struct btrfs_delayed_data_ref *data_ref; | |
3057 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 YZ |
3058 | int ret = 0; |
3059 | ||
5d4f98a2 YZ |
3060 | delayed_refs = &trans->transaction->delayed_refs; |
3061 | spin_lock(&delayed_refs->lock); | |
3062 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
d7df2c79 JB |
3063 | if (!head) { |
3064 | spin_unlock(&delayed_refs->lock); | |
3065 | return 0; | |
3066 | } | |
5d4f98a2 YZ |
3067 | |
3068 | if (!mutex_trylock(&head->mutex)) { | |
3069 | atomic_inc(&head->node.refs); | |
3070 | spin_unlock(&delayed_refs->lock); | |
3071 | ||
b3b4aa74 | 3072 | btrfs_release_path(path); |
5d4f98a2 | 3073 | |
8cc33e5c DS |
3074 | /* |
3075 | * Mutex was contended, block until it's released and let | |
3076 | * caller try again | |
3077 | */ | |
5d4f98a2 YZ |
3078 | mutex_lock(&head->mutex); |
3079 | mutex_unlock(&head->mutex); | |
3080 | btrfs_put_delayed_ref(&head->node); | |
3081 | return -EAGAIN; | |
3082 | } | |
d7df2c79 | 3083 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3084 | |
d7df2c79 | 3085 | spin_lock(&head->lock); |
c6fc2454 | 3086 | list_for_each_entry(ref, &head->ref_list, list) { |
d7df2c79 JB |
3087 | /* If it's a shared ref we know a cross reference exists */ |
3088 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3089 | ret = 1; | |
3090 | break; | |
3091 | } | |
5d4f98a2 | 3092 | |
d7df2c79 | 3093 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3094 | |
d7df2c79 JB |
3095 | /* |
3096 | * If our ref doesn't match the one we're currently looking at | |
3097 | * then we have a cross reference. | |
3098 | */ | |
3099 | if (data_ref->root != root->root_key.objectid || | |
3100 | data_ref->objectid != objectid || | |
3101 | data_ref->offset != offset) { | |
3102 | ret = 1; | |
3103 | break; | |
3104 | } | |
5d4f98a2 | 3105 | } |
d7df2c79 | 3106 | spin_unlock(&head->lock); |
5d4f98a2 | 3107 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
3108 | return ret; |
3109 | } | |
3110 | ||
3111 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
3112 | struct btrfs_root *root, | |
3113 | struct btrfs_path *path, | |
3114 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
3115 | { |
3116 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 3117 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3118 | struct btrfs_extent_data_ref *ref; |
3119 | struct btrfs_extent_inline_ref *iref; | |
3120 | struct btrfs_extent_item *ei; | |
f321e491 | 3121 | struct btrfs_key key; |
5d4f98a2 | 3122 | u32 item_size; |
be20aa9d | 3123 | int ret; |
925baedd | 3124 | |
be20aa9d | 3125 | key.objectid = bytenr; |
31840ae1 | 3126 | key.offset = (u64)-1; |
f321e491 | 3127 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3128 | |
be20aa9d CM |
3129 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3130 | if (ret < 0) | |
3131 | goto out; | |
79787eaa | 3132 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3133 | |
3134 | ret = -ENOENT; | |
3135 | if (path->slots[0] == 0) | |
31840ae1 | 3136 | goto out; |
be20aa9d | 3137 | |
31840ae1 | 3138 | path->slots[0]--; |
f321e491 | 3139 | leaf = path->nodes[0]; |
5d4f98a2 | 3140 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3141 | |
5d4f98a2 | 3142 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3143 | goto out; |
f321e491 | 3144 | |
5d4f98a2 YZ |
3145 | ret = 1; |
3146 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3147 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3148 | if (item_size < sizeof(*ei)) { | |
3149 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3150 | goto out; | |
3151 | } | |
3152 | #endif | |
3153 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3154 | |
5d4f98a2 YZ |
3155 | if (item_size != sizeof(*ei) + |
3156 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3157 | goto out; | |
be20aa9d | 3158 | |
5d4f98a2 YZ |
3159 | if (btrfs_extent_generation(leaf, ei) <= |
3160 | btrfs_root_last_snapshot(&root->root_item)) | |
3161 | goto out; | |
3162 | ||
3163 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3164 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
3165 | BTRFS_EXTENT_DATA_REF_KEY) | |
3166 | goto out; | |
3167 | ||
3168 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3169 | if (btrfs_extent_refs(leaf, ei) != | |
3170 | btrfs_extent_data_ref_count(leaf, ref) || | |
3171 | btrfs_extent_data_ref_root(leaf, ref) != | |
3172 | root->root_key.objectid || | |
3173 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3174 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3175 | goto out; | |
3176 | ||
3177 | ret = 0; | |
3178 | out: | |
3179 | return ret; | |
3180 | } | |
3181 | ||
3182 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
3183 | struct btrfs_root *root, | |
3184 | u64 objectid, u64 offset, u64 bytenr) | |
3185 | { | |
3186 | struct btrfs_path *path; | |
3187 | int ret; | |
3188 | int ret2; | |
3189 | ||
3190 | path = btrfs_alloc_path(); | |
3191 | if (!path) | |
3192 | return -ENOENT; | |
3193 | ||
3194 | do { | |
3195 | ret = check_committed_ref(trans, root, path, objectid, | |
3196 | offset, bytenr); | |
3197 | if (ret && ret != -ENOENT) | |
f321e491 | 3198 | goto out; |
80ff3856 | 3199 | |
5d4f98a2 YZ |
3200 | ret2 = check_delayed_ref(trans, root, path, objectid, |
3201 | offset, bytenr); | |
3202 | } while (ret2 == -EAGAIN); | |
3203 | ||
3204 | if (ret2 && ret2 != -ENOENT) { | |
3205 | ret = ret2; | |
3206 | goto out; | |
f321e491 | 3207 | } |
5d4f98a2 YZ |
3208 | |
3209 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3210 | ret = 0; | |
be20aa9d | 3211 | out: |
80ff3856 | 3212 | btrfs_free_path(path); |
f0486c68 YZ |
3213 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3214 | WARN_ON(ret > 0); | |
f321e491 | 3215 | return ret; |
be20aa9d | 3216 | } |
c5739bba | 3217 | |
5d4f98a2 | 3218 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3219 | struct btrfs_root *root, |
5d4f98a2 | 3220 | struct extent_buffer *buf, |
e339a6b0 | 3221 | int full_backref, int inc) |
31840ae1 ZY |
3222 | { |
3223 | u64 bytenr; | |
5d4f98a2 YZ |
3224 | u64 num_bytes; |
3225 | u64 parent; | |
31840ae1 | 3226 | u64 ref_root; |
31840ae1 | 3227 | u32 nritems; |
31840ae1 ZY |
3228 | struct btrfs_key key; |
3229 | struct btrfs_file_extent_item *fi; | |
3230 | int i; | |
3231 | int level; | |
3232 | int ret = 0; | |
31840ae1 | 3233 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
b06c4bf5 | 3234 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3235 | |
fccb84c9 DS |
3236 | |
3237 | if (btrfs_test_is_dummy_root(root)) | |
faa2dbf0 | 3238 | return 0; |
fccb84c9 | 3239 | |
31840ae1 | 3240 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3241 | nritems = btrfs_header_nritems(buf); |
3242 | level = btrfs_header_level(buf); | |
3243 | ||
27cdeb70 | 3244 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3245 | return 0; |
31840ae1 | 3246 | |
5d4f98a2 YZ |
3247 | if (inc) |
3248 | process_func = btrfs_inc_extent_ref; | |
3249 | else | |
3250 | process_func = btrfs_free_extent; | |
31840ae1 | 3251 | |
5d4f98a2 YZ |
3252 | if (full_backref) |
3253 | parent = buf->start; | |
3254 | else | |
3255 | parent = 0; | |
3256 | ||
3257 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3258 | if (level == 0) { |
5d4f98a2 | 3259 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3260 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3261 | continue; |
5d4f98a2 | 3262 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3263 | struct btrfs_file_extent_item); |
3264 | if (btrfs_file_extent_type(buf, fi) == | |
3265 | BTRFS_FILE_EXTENT_INLINE) | |
3266 | continue; | |
3267 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3268 | if (bytenr == 0) | |
3269 | continue; | |
5d4f98a2 YZ |
3270 | |
3271 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3272 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
3273 | ret = process_func(trans, root, bytenr, num_bytes, | |
3274 | parent, ref_root, key.objectid, | |
b06c4bf5 | 3275 | key.offset); |
31840ae1 ZY |
3276 | if (ret) |
3277 | goto fail; | |
3278 | } else { | |
5d4f98a2 | 3279 | bytenr = btrfs_node_blockptr(buf, i); |
707e8a07 | 3280 | num_bytes = root->nodesize; |
5d4f98a2 | 3281 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3282 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3283 | if (ret) |
3284 | goto fail; | |
3285 | } | |
3286 | } | |
3287 | return 0; | |
3288 | fail: | |
5d4f98a2 YZ |
3289 | return ret; |
3290 | } | |
3291 | ||
3292 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3293 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3294 | { |
e339a6b0 | 3295 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3296 | } |
3297 | ||
3298 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3299 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3300 | { |
e339a6b0 | 3301 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3302 | } |
3303 | ||
9078a3e1 CM |
3304 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
3305 | struct btrfs_root *root, | |
3306 | struct btrfs_path *path, | |
3307 | struct btrfs_block_group_cache *cache) | |
3308 | { | |
3309 | int ret; | |
9078a3e1 | 3310 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
3311 | unsigned long bi; |
3312 | struct extent_buffer *leaf; | |
9078a3e1 | 3313 | |
9078a3e1 | 3314 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3315 | if (ret) { |
3316 | if (ret > 0) | |
3317 | ret = -ENOENT; | |
54aa1f4d | 3318 | goto fail; |
df95e7f0 | 3319 | } |
5f39d397 CM |
3320 | |
3321 | leaf = path->nodes[0]; | |
3322 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3323 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3324 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3325 | fail: |
24b89d08 | 3326 | btrfs_release_path(path); |
df95e7f0 | 3327 | return ret; |
9078a3e1 CM |
3328 | |
3329 | } | |
3330 | ||
4a8c9a62 YZ |
3331 | static struct btrfs_block_group_cache * |
3332 | next_block_group(struct btrfs_root *root, | |
3333 | struct btrfs_block_group_cache *cache) | |
3334 | { | |
3335 | struct rb_node *node; | |
292cbd51 | 3336 | |
4a8c9a62 | 3337 | spin_lock(&root->fs_info->block_group_cache_lock); |
292cbd51 FM |
3338 | |
3339 | /* If our block group was removed, we need a full search. */ | |
3340 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3341 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3342 | ||
3343 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3344 | btrfs_put_block_group(cache); | |
3345 | cache = btrfs_lookup_first_block_group(root->fs_info, | |
3346 | next_bytenr); | |
3347 | return cache; | |
3348 | } | |
4a8c9a62 YZ |
3349 | node = rb_next(&cache->cache_node); |
3350 | btrfs_put_block_group(cache); | |
3351 | if (node) { | |
3352 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3353 | cache_node); | |
11dfe35a | 3354 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3355 | } else |
3356 | cache = NULL; | |
3357 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3358 | return cache; | |
3359 | } | |
3360 | ||
0af3d00b JB |
3361 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3362 | struct btrfs_trans_handle *trans, | |
3363 | struct btrfs_path *path) | |
3364 | { | |
3365 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
3366 | struct inode *inode = NULL; | |
3367 | u64 alloc_hint = 0; | |
2b20982e | 3368 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3369 | u64 num_pages = 0; |
0af3d00b JB |
3370 | int retries = 0; |
3371 | int ret = 0; | |
3372 | ||
3373 | /* | |
3374 | * If this block group is smaller than 100 megs don't bother caching the | |
3375 | * block group. | |
3376 | */ | |
ee22184b | 3377 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3378 | spin_lock(&block_group->lock); |
3379 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3380 | spin_unlock(&block_group->lock); | |
3381 | return 0; | |
3382 | } | |
3383 | ||
0c0ef4bc JB |
3384 | if (trans->aborted) |
3385 | return 0; | |
0af3d00b JB |
3386 | again: |
3387 | inode = lookup_free_space_inode(root, block_group, path); | |
3388 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
3389 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3390 | btrfs_release_path(path); |
0af3d00b JB |
3391 | goto out; |
3392 | } | |
3393 | ||
3394 | if (IS_ERR(inode)) { | |
3395 | BUG_ON(retries); | |
3396 | retries++; | |
3397 | ||
3398 | if (block_group->ro) | |
3399 | goto out_free; | |
3400 | ||
3401 | ret = create_free_space_inode(root, trans, block_group, path); | |
3402 | if (ret) | |
3403 | goto out_free; | |
3404 | goto again; | |
3405 | } | |
3406 | ||
5b0e95bf JB |
3407 | /* We've already setup this transaction, go ahead and exit */ |
3408 | if (block_group->cache_generation == trans->transid && | |
3409 | i_size_read(inode)) { | |
3410 | dcs = BTRFS_DC_SETUP; | |
3411 | goto out_put; | |
3412 | } | |
3413 | ||
0af3d00b JB |
3414 | /* |
3415 | * We want to set the generation to 0, that way if anything goes wrong | |
3416 | * from here on out we know not to trust this cache when we load up next | |
3417 | * time. | |
3418 | */ | |
3419 | BTRFS_I(inode)->generation = 0; | |
3420 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3421 | if (ret) { |
3422 | /* | |
3423 | * So theoretically we could recover from this, simply set the | |
3424 | * super cache generation to 0 so we know to invalidate the | |
3425 | * cache, but then we'd have to keep track of the block groups | |
3426 | * that fail this way so we know we _have_ to reset this cache | |
3427 | * before the next commit or risk reading stale cache. So to | |
3428 | * limit our exposure to horrible edge cases lets just abort the | |
3429 | * transaction, this only happens in really bad situations | |
3430 | * anyway. | |
3431 | */ | |
3432 | btrfs_abort_transaction(trans, root, ret); | |
3433 | goto out_put; | |
3434 | } | |
0af3d00b JB |
3435 | WARN_ON(ret); |
3436 | ||
3437 | if (i_size_read(inode) > 0) { | |
7b61cd92 MX |
3438 | ret = btrfs_check_trunc_cache_free_space(root, |
3439 | &root->fs_info->global_block_rsv); | |
3440 | if (ret) | |
3441 | goto out_put; | |
3442 | ||
1bbc621e | 3443 | ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode); |
0af3d00b JB |
3444 | if (ret) |
3445 | goto out_put; | |
3446 | } | |
3447 | ||
3448 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3449 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
e4c88f00 | 3450 | !btrfs_test_opt(root, SPACE_CACHE)) { |
cf7c1ef6 LB |
3451 | /* |
3452 | * don't bother trying to write stuff out _if_ | |
3453 | * a) we're not cached, | |
3454 | * b) we're with nospace_cache mount option. | |
3455 | */ | |
2b20982e | 3456 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3457 | spin_unlock(&block_group->lock); |
3458 | goto out_put; | |
3459 | } | |
3460 | spin_unlock(&block_group->lock); | |
3461 | ||
2968b1f4 JB |
3462 | /* |
3463 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3464 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3465 | */ | |
3466 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3467 | ret = -ENOSPC; | |
3468 | goto out_put; | |
3469 | } | |
3470 | ||
6fc823b1 JB |
3471 | /* |
3472 | * Try to preallocate enough space based on how big the block group is. | |
3473 | * Keep in mind this has to include any pinned space which could end up | |
3474 | * taking up quite a bit since it's not folded into the other space | |
3475 | * cache. | |
3476 | */ | |
ee22184b | 3477 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3478 | if (!num_pages) |
3479 | num_pages = 1; | |
3480 | ||
0af3d00b | 3481 | num_pages *= 16; |
09cbfeaf | 3482 | num_pages *= PAGE_SIZE; |
0af3d00b | 3483 | |
7cf5b976 | 3484 | ret = btrfs_check_data_free_space(inode, 0, num_pages); |
0af3d00b JB |
3485 | if (ret) |
3486 | goto out_put; | |
3487 | ||
3488 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3489 | num_pages, num_pages, | |
3490 | &alloc_hint); | |
2968b1f4 JB |
3491 | /* |
3492 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3493 | * of metadata or split extents when writing the cache out, which means | |
3494 | * we can enospc if we are heavily fragmented in addition to just normal | |
3495 | * out of space conditions. So if we hit this just skip setting up any | |
3496 | * other block groups for this transaction, maybe we'll unpin enough | |
3497 | * space the next time around. | |
3498 | */ | |
2b20982e JB |
3499 | if (!ret) |
3500 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3501 | else if (ret == -ENOSPC) |
3502 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
7cf5b976 | 3503 | btrfs_free_reserved_data_space(inode, 0, num_pages); |
c09544e0 | 3504 | |
0af3d00b JB |
3505 | out_put: |
3506 | iput(inode); | |
3507 | out_free: | |
b3b4aa74 | 3508 | btrfs_release_path(path); |
0af3d00b JB |
3509 | out: |
3510 | spin_lock(&block_group->lock); | |
e65cbb94 | 3511 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3512 | block_group->cache_generation = trans->transid; |
2b20982e | 3513 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3514 | spin_unlock(&block_group->lock); |
3515 | ||
3516 | return ret; | |
3517 | } | |
3518 | ||
dcdf7f6d JB |
3519 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
3520 | struct btrfs_root *root) | |
3521 | { | |
3522 | struct btrfs_block_group_cache *cache, *tmp; | |
3523 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3524 | struct btrfs_path *path; | |
3525 | ||
3526 | if (list_empty(&cur_trans->dirty_bgs) || | |
3527 | !btrfs_test_opt(root, SPACE_CACHE)) | |
3528 | return 0; | |
3529 | ||
3530 | path = btrfs_alloc_path(); | |
3531 | if (!path) | |
3532 | return -ENOMEM; | |
3533 | ||
3534 | /* Could add new block groups, use _safe just in case */ | |
3535 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3536 | dirty_list) { | |
3537 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3538 | cache_save_setup(cache, trans, path); | |
3539 | } | |
3540 | ||
3541 | btrfs_free_path(path); | |
3542 | return 0; | |
3543 | } | |
3544 | ||
1bbc621e CM |
3545 | /* |
3546 | * transaction commit does final block group cache writeback during a | |
3547 | * critical section where nothing is allowed to change the FS. This is | |
3548 | * required in order for the cache to actually match the block group, | |
3549 | * but can introduce a lot of latency into the commit. | |
3550 | * | |
3551 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3552 | * cache IO. There's a chance we'll have to redo some of it if the | |
3553 | * block group changes again during the commit, but it greatly reduces | |
3554 | * the commit latency by getting rid of the easy block groups while | |
3555 | * we're still allowing others to join the commit. | |
3556 | */ | |
3557 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans, | |
96b5179d | 3558 | struct btrfs_root *root) |
9078a3e1 | 3559 | { |
4a8c9a62 | 3560 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3561 | struct btrfs_transaction *cur_trans = trans->transaction; |
3562 | int ret = 0; | |
c9dc4c65 | 3563 | int should_put; |
1bbc621e CM |
3564 | struct btrfs_path *path = NULL; |
3565 | LIST_HEAD(dirty); | |
3566 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3567 | int num_started = 0; |
1bbc621e CM |
3568 | int loops = 0; |
3569 | ||
3570 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3571 | if (list_empty(&cur_trans->dirty_bgs)) { |
3572 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3573 | return 0; | |
1bbc621e | 3574 | } |
b58d1a9e | 3575 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3576 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3577 | |
1bbc621e | 3578 | again: |
1bbc621e CM |
3579 | /* |
3580 | * make sure all the block groups on our dirty list actually | |
3581 | * exist | |
3582 | */ | |
3583 | btrfs_create_pending_block_groups(trans, root); | |
3584 | ||
3585 | if (!path) { | |
3586 | path = btrfs_alloc_path(); | |
3587 | if (!path) | |
3588 | return -ENOMEM; | |
3589 | } | |
3590 | ||
b58d1a9e FM |
3591 | /* |
3592 | * cache_write_mutex is here only to save us from balance or automatic | |
3593 | * removal of empty block groups deleting this block group while we are | |
3594 | * writing out the cache | |
3595 | */ | |
3596 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3597 | while (!list_empty(&dirty)) { |
3598 | cache = list_first_entry(&dirty, | |
3599 | struct btrfs_block_group_cache, | |
3600 | dirty_list); | |
1bbc621e CM |
3601 | /* |
3602 | * this can happen if something re-dirties a block | |
3603 | * group that is already under IO. Just wait for it to | |
3604 | * finish and then do it all again | |
3605 | */ | |
3606 | if (!list_empty(&cache->io_list)) { | |
3607 | list_del_init(&cache->io_list); | |
3608 | btrfs_wait_cache_io(root, trans, cache, | |
3609 | &cache->io_ctl, path, | |
3610 | cache->key.objectid); | |
3611 | btrfs_put_block_group(cache); | |
3612 | } | |
3613 | ||
3614 | ||
3615 | /* | |
3616 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3617 | * if it should update the cache_state. Don't delete | |
3618 | * until after we wait. | |
3619 | * | |
3620 | * Since we're not running in the commit critical section | |
3621 | * we need the dirty_bgs_lock to protect from update_block_group | |
3622 | */ | |
3623 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3624 | list_del_init(&cache->dirty_list); | |
3625 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3626 | ||
3627 | should_put = 1; | |
3628 | ||
3629 | cache_save_setup(cache, trans, path); | |
3630 | ||
3631 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3632 | cache->io_ctl.inode = NULL; | |
3633 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3634 | if (ret == 0 && cache->io_ctl.inode) { | |
3635 | num_started++; | |
3636 | should_put = 0; | |
3637 | ||
3638 | /* | |
3639 | * the cache_write_mutex is protecting | |
3640 | * the io_list | |
3641 | */ | |
3642 | list_add_tail(&cache->io_list, io); | |
3643 | } else { | |
3644 | /* | |
3645 | * if we failed to write the cache, the | |
3646 | * generation will be bad and life goes on | |
3647 | */ | |
3648 | ret = 0; | |
3649 | } | |
3650 | } | |
ff1f8250 | 3651 | if (!ret) { |
1bbc621e | 3652 | ret = write_one_cache_group(trans, root, path, cache); |
ff1f8250 FM |
3653 | /* |
3654 | * Our block group might still be attached to the list | |
3655 | * of new block groups in the transaction handle of some | |
3656 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3657 | * means its block group item isn't yet in the extent | |
3658 | * tree. If this happens ignore the error, as we will | |
3659 | * try again later in the critical section of the | |
3660 | * transaction commit. | |
3661 | */ | |
3662 | if (ret == -ENOENT) { | |
3663 | ret = 0; | |
3664 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3665 | if (list_empty(&cache->dirty_list)) { | |
3666 | list_add_tail(&cache->dirty_list, | |
3667 | &cur_trans->dirty_bgs); | |
3668 | btrfs_get_block_group(cache); | |
3669 | } | |
3670 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3671 | } else if (ret) { | |
3672 | btrfs_abort_transaction(trans, root, ret); | |
3673 | } | |
3674 | } | |
1bbc621e CM |
3675 | |
3676 | /* if its not on the io list, we need to put the block group */ | |
3677 | if (should_put) | |
3678 | btrfs_put_block_group(cache); | |
3679 | ||
3680 | if (ret) | |
3681 | break; | |
b58d1a9e FM |
3682 | |
3683 | /* | |
3684 | * Avoid blocking other tasks for too long. It might even save | |
3685 | * us from writing caches for block groups that are going to be | |
3686 | * removed. | |
3687 | */ | |
3688 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3689 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3690 | } |
b58d1a9e | 3691 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3692 | |
3693 | /* | |
3694 | * go through delayed refs for all the stuff we've just kicked off | |
3695 | * and then loop back (just once) | |
3696 | */ | |
3697 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
3698 | if (!ret && loops == 0) { | |
3699 | loops++; | |
3700 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3701 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3702 | /* |
3703 | * dirty_bgs_lock protects us from concurrent block group | |
3704 | * deletes too (not just cache_write_mutex). | |
3705 | */ | |
3706 | if (!list_empty(&dirty)) { | |
3707 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3708 | goto again; | |
3709 | } | |
1bbc621e | 3710 | spin_unlock(&cur_trans->dirty_bgs_lock); |
1bbc621e CM |
3711 | } |
3712 | ||
3713 | btrfs_free_path(path); | |
3714 | return ret; | |
3715 | } | |
3716 | ||
3717 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
3718 | struct btrfs_root *root) | |
3719 | { | |
3720 | struct btrfs_block_group_cache *cache; | |
3721 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3722 | int ret = 0; | |
3723 | int should_put; | |
3724 | struct btrfs_path *path; | |
3725 | struct list_head *io = &cur_trans->io_bgs; | |
3726 | int num_started = 0; | |
9078a3e1 CM |
3727 | |
3728 | path = btrfs_alloc_path(); | |
3729 | if (!path) | |
3730 | return -ENOMEM; | |
3731 | ||
ce93ec54 | 3732 | /* |
e44081ef FM |
3733 | * Even though we are in the critical section of the transaction commit, |
3734 | * we can still have concurrent tasks adding elements to this | |
3735 | * transaction's list of dirty block groups. These tasks correspond to | |
3736 | * endio free space workers started when writeback finishes for a | |
3737 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3738 | * allocate new block groups as a result of COWing nodes of the root | |
3739 | * tree when updating the free space inode. The writeback for the space | |
3740 | * caches is triggered by an earlier call to | |
3741 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3742 | * loop. | |
3743 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3744 | * delayed refs to make sure we have the best chance at doing this all |
3745 | * in one shot. | |
3746 | */ | |
e44081ef | 3747 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3748 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3749 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3750 | struct btrfs_block_group_cache, | |
3751 | dirty_list); | |
c9dc4c65 CM |
3752 | |
3753 | /* | |
3754 | * this can happen if cache_save_setup re-dirties a block | |
3755 | * group that is already under IO. Just wait for it to | |
3756 | * finish and then do it all again | |
3757 | */ | |
3758 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3759 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3760 | list_del_init(&cache->io_list); |
3761 | btrfs_wait_cache_io(root, trans, cache, | |
3762 | &cache->io_ctl, path, | |
3763 | cache->key.objectid); | |
3764 | btrfs_put_block_group(cache); | |
e44081ef | 3765 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3766 | } |
3767 | ||
1bbc621e CM |
3768 | /* |
3769 | * don't remove from the dirty list until after we've waited | |
3770 | * on any pending IO | |
3771 | */ | |
ce93ec54 | 3772 | list_del_init(&cache->dirty_list); |
e44081ef | 3773 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3774 | should_put = 1; |
3775 | ||
1bbc621e | 3776 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3777 | |
ce93ec54 | 3778 | if (!ret) |
c9dc4c65 CM |
3779 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1); |
3780 | ||
3781 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3782 | cache->io_ctl.inode = NULL; | |
3783 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3784 | if (ret == 0 && cache->io_ctl.inode) { | |
3785 | num_started++; | |
3786 | should_put = 0; | |
1bbc621e | 3787 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3788 | } else { |
3789 | /* | |
3790 | * if we failed to write the cache, the | |
3791 | * generation will be bad and life goes on | |
3792 | */ | |
3793 | ret = 0; | |
3794 | } | |
3795 | } | |
ff1f8250 | 3796 | if (!ret) { |
ce93ec54 | 3797 | ret = write_one_cache_group(trans, root, path, cache); |
2bc0bb5f FM |
3798 | /* |
3799 | * One of the free space endio workers might have | |
3800 | * created a new block group while updating a free space | |
3801 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3802 | * and hasn't released its transaction handle yet, in | |
3803 | * which case the new block group is still attached to | |
3804 | * its transaction handle and its creation has not | |
3805 | * finished yet (no block group item in the extent tree | |
3806 | * yet, etc). If this is the case, wait for all free | |
3807 | * space endio workers to finish and retry. This is a | |
3808 | * a very rare case so no need for a more efficient and | |
3809 | * complex approach. | |
3810 | */ | |
3811 | if (ret == -ENOENT) { | |
3812 | wait_event(cur_trans->writer_wait, | |
3813 | atomic_read(&cur_trans->num_writers) == 1); | |
3814 | ret = write_one_cache_group(trans, root, path, | |
3815 | cache); | |
3816 | } | |
ff1f8250 FM |
3817 | if (ret) |
3818 | btrfs_abort_transaction(trans, root, ret); | |
3819 | } | |
c9dc4c65 CM |
3820 | |
3821 | /* if its not on the io list, we need to put the block group */ | |
3822 | if (should_put) | |
3823 | btrfs_put_block_group(cache); | |
e44081ef | 3824 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3825 | } |
e44081ef | 3826 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3827 | |
1bbc621e CM |
3828 | while (!list_empty(io)) { |
3829 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3830 | io_list); |
3831 | list_del_init(&cache->io_list); | |
c9dc4c65 CM |
3832 | btrfs_wait_cache_io(root, trans, cache, |
3833 | &cache->io_ctl, path, cache->key.objectid); | |
0cb59c99 JB |
3834 | btrfs_put_block_group(cache); |
3835 | } | |
3836 | ||
9078a3e1 | 3837 | btrfs_free_path(path); |
ce93ec54 | 3838 | return ret; |
9078a3e1 CM |
3839 | } |
3840 | ||
d2fb3437 YZ |
3841 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3842 | { | |
3843 | struct btrfs_block_group_cache *block_group; | |
3844 | int readonly = 0; | |
3845 | ||
3846 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3847 | if (!block_group || block_group->ro) | |
3848 | readonly = 1; | |
3849 | if (block_group) | |
fa9c0d79 | 3850 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3851 | return readonly; |
3852 | } | |
3853 | ||
f78c436c FM |
3854 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3855 | { | |
3856 | struct btrfs_block_group_cache *bg; | |
3857 | bool ret = true; | |
3858 | ||
3859 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3860 | if (!bg) | |
3861 | return false; | |
3862 | ||
3863 | spin_lock(&bg->lock); | |
3864 | if (bg->ro) | |
3865 | ret = false; | |
3866 | else | |
3867 | atomic_inc(&bg->nocow_writers); | |
3868 | spin_unlock(&bg->lock); | |
3869 | ||
3870 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3871 | if (!ret) | |
3872 | btrfs_put_block_group(bg); | |
3873 | ||
3874 | return ret; | |
3875 | ||
3876 | } | |
3877 | ||
3878 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3879 | { | |
3880 | struct btrfs_block_group_cache *bg; | |
3881 | ||
3882 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3883 | ASSERT(bg); | |
3884 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
3885 | wake_up_atomic_t(&bg->nocow_writers); | |
3886 | /* | |
3887 | * Once for our lookup and once for the lookup done by a previous call | |
3888 | * to btrfs_inc_nocow_writers() | |
3889 | */ | |
3890 | btrfs_put_block_group(bg); | |
3891 | btrfs_put_block_group(bg); | |
3892 | } | |
3893 | ||
3894 | static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a) | |
3895 | { | |
3896 | schedule(); | |
3897 | return 0; | |
3898 | } | |
3899 | ||
3900 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) | |
3901 | { | |
3902 | wait_on_atomic_t(&bg->nocow_writers, | |
3903 | btrfs_wait_nocow_writers_atomic_t, | |
3904 | TASK_UNINTERRUPTIBLE); | |
3905 | } | |
3906 | ||
6ab0a202 JM |
3907 | static const char *alloc_name(u64 flags) |
3908 | { | |
3909 | switch (flags) { | |
3910 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3911 | return "mixed"; | |
3912 | case BTRFS_BLOCK_GROUP_METADATA: | |
3913 | return "metadata"; | |
3914 | case BTRFS_BLOCK_GROUP_DATA: | |
3915 | return "data"; | |
3916 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3917 | return "system"; | |
3918 | default: | |
3919 | WARN_ON(1); | |
3920 | return "invalid-combination"; | |
3921 | }; | |
3922 | } | |
3923 | ||
593060d7 CM |
3924 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3925 | u64 total_bytes, u64 bytes_used, | |
e40edf2d | 3926 | u64 bytes_readonly, |
593060d7 CM |
3927 | struct btrfs_space_info **space_info) |
3928 | { | |
3929 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3930 | int i; |
3931 | int factor; | |
b150a4f1 | 3932 | int ret; |
b742bb82 YZ |
3933 | |
3934 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3935 | BTRFS_BLOCK_GROUP_RAID10)) | |
3936 | factor = 2; | |
3937 | else | |
3938 | factor = 1; | |
593060d7 CM |
3939 | |
3940 | found = __find_space_info(info, flags); | |
3941 | if (found) { | |
25179201 | 3942 | spin_lock(&found->lock); |
593060d7 | 3943 | found->total_bytes += total_bytes; |
89a55897 | 3944 | found->disk_total += total_bytes * factor; |
593060d7 | 3945 | found->bytes_used += bytes_used; |
b742bb82 | 3946 | found->disk_used += bytes_used * factor; |
e40edf2d | 3947 | found->bytes_readonly += bytes_readonly; |
2e6e5183 FM |
3948 | if (total_bytes > 0) |
3949 | found->full = 0; | |
957780eb JB |
3950 | space_info_add_new_bytes(info, found, total_bytes - |
3951 | bytes_used - bytes_readonly); | |
25179201 | 3952 | spin_unlock(&found->lock); |
593060d7 CM |
3953 | *space_info = found; |
3954 | return 0; | |
3955 | } | |
c146afad | 3956 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3957 | if (!found) |
3958 | return -ENOMEM; | |
3959 | ||
908c7f19 | 3960 | ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL); |
b150a4f1 JB |
3961 | if (ret) { |
3962 | kfree(found); | |
3963 | return ret; | |
3964 | } | |
3965 | ||
c1895442 | 3966 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
b742bb82 | 3967 | INIT_LIST_HEAD(&found->block_groups[i]); |
80eb234a | 3968 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3969 | spin_lock_init(&found->lock); |
52ba6929 | 3970 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3971 | found->total_bytes = total_bytes; |
89a55897 | 3972 | found->disk_total = total_bytes * factor; |
593060d7 | 3973 | found->bytes_used = bytes_used; |
b742bb82 | 3974 | found->disk_used = bytes_used * factor; |
593060d7 | 3975 | found->bytes_pinned = 0; |
e8569813 | 3976 | found->bytes_reserved = 0; |
e40edf2d | 3977 | found->bytes_readonly = bytes_readonly; |
f0486c68 | 3978 | found->bytes_may_use = 0; |
6af3e3ad | 3979 | found->full = 0; |
4f4db217 | 3980 | found->max_extent_size = 0; |
0e4f8f88 | 3981 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3982 | found->chunk_alloc = 0; |
fdb5effd JB |
3983 | found->flush = 0; |
3984 | init_waitqueue_head(&found->wait); | |
633c0aad | 3985 | INIT_LIST_HEAD(&found->ro_bgs); |
957780eb JB |
3986 | INIT_LIST_HEAD(&found->tickets); |
3987 | INIT_LIST_HEAD(&found->priority_tickets); | |
6ab0a202 JM |
3988 | |
3989 | ret = kobject_init_and_add(&found->kobj, &space_info_ktype, | |
3990 | info->space_info_kobj, "%s", | |
3991 | alloc_name(found->flags)); | |
3992 | if (ret) { | |
3993 | kfree(found); | |
3994 | return ret; | |
3995 | } | |
3996 | ||
593060d7 | 3997 | *space_info = found; |
4184ea7f | 3998 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3999 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4000 | info->data_sinfo = found; | |
6ab0a202 JM |
4001 | |
4002 | return ret; | |
593060d7 CM |
4003 | } |
4004 | ||
8790d502 CM |
4005 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
4006 | { | |
899c81ea ID |
4007 | u64 extra_flags = chunk_to_extended(flags) & |
4008 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4009 | |
de98ced9 | 4010 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4011 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4012 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4013 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4014 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4015 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4016 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4017 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4018 | } |
593060d7 | 4019 | |
fc67c450 ID |
4020 | /* |
4021 | * returns target flags in extended format or 0 if restripe for this | |
4022 | * chunk_type is not in progress | |
c6664b42 ID |
4023 | * |
4024 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
4025 | */ |
4026 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4027 | { | |
4028 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4029 | u64 target = 0; | |
4030 | ||
fc67c450 ID |
4031 | if (!bctl) |
4032 | return 0; | |
4033 | ||
4034 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4035 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4036 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4037 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4038 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4039 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4040 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4041 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4042 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4043 | } | |
4044 | ||
4045 | return target; | |
4046 | } | |
4047 | ||
a46d11a8 ID |
4048 | /* |
4049 | * @flags: available profiles in extended format (see ctree.h) | |
4050 | * | |
e4d8ec0f ID |
4051 | * Returns reduced profile in chunk format. If profile changing is in |
4052 | * progress (either running or paused) picks the target profile (if it's | |
4053 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4054 | */ |
48a3b636 | 4055 | static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 4056 | { |
95669976 | 4057 | u64 num_devices = root->fs_info->fs_devices->rw_devices; |
fc67c450 | 4058 | u64 target; |
9c170b26 ZL |
4059 | u64 raid_type; |
4060 | u64 allowed = 0; | |
a061fc8d | 4061 | |
fc67c450 ID |
4062 | /* |
4063 | * see if restripe for this chunk_type is in progress, if so | |
4064 | * try to reduce to the target profile | |
4065 | */ | |
e4d8ec0f | 4066 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
4067 | target = get_restripe_target(root->fs_info, flags); |
4068 | if (target) { | |
4069 | /* pick target profile only if it's already available */ | |
4070 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 4071 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 4072 | return extended_to_chunk(target); |
e4d8ec0f ID |
4073 | } |
4074 | } | |
4075 | spin_unlock(&root->fs_info->balance_lock); | |
4076 | ||
53b381b3 | 4077 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4078 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4079 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
4080 | allowed |= btrfs_raid_group[raid_type]; | |
4081 | } | |
4082 | allowed &= flags; | |
4083 | ||
4084 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4085 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4086 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4087 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4088 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4089 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4090 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4091 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4092 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4093 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4094 | ||
4095 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4096 | ||
4097 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4098 | } |
4099 | ||
f8213bdc | 4100 | static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags) |
6a63209f | 4101 | { |
de98ced9 | 4102 | unsigned seq; |
f8213bdc | 4103 | u64 flags; |
de98ced9 MX |
4104 | |
4105 | do { | |
f8213bdc | 4106 | flags = orig_flags; |
de98ced9 MX |
4107 | seq = read_seqbegin(&root->fs_info->profiles_lock); |
4108 | ||
4109 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4110 | flags |= root->fs_info->avail_data_alloc_bits; | |
4111 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4112 | flags |= root->fs_info->avail_system_alloc_bits; | |
4113 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4114 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
4115 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
6fef8df1 | 4116 | |
b742bb82 | 4117 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
4118 | } |
4119 | ||
6d07bcec | 4120 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 4121 | { |
b742bb82 | 4122 | u64 flags; |
53b381b3 | 4123 | u64 ret; |
9ed74f2d | 4124 | |
b742bb82 YZ |
4125 | if (data) |
4126 | flags = BTRFS_BLOCK_GROUP_DATA; | |
4127 | else if (root == root->fs_info->chunk_root) | |
4128 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 4129 | else |
b742bb82 | 4130 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4131 | |
53b381b3 DW |
4132 | ret = get_alloc_profile(root, flags); |
4133 | return ret; | |
6a63209f | 4134 | } |
9ed74f2d | 4135 | |
4ceff079 | 4136 | int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes) |
6a63209f | 4137 | { |
6a63209f | 4138 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 4139 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 4140 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 4141 | u64 used; |
94b947b2 | 4142 | int ret = 0; |
c99f1b0c ZL |
4143 | int need_commit = 2; |
4144 | int have_pinned_space; | |
6a63209f | 4145 | |
6a63209f | 4146 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 4147 | bytes = ALIGN(bytes, root->sectorsize); |
6a63209f | 4148 | |
9dced186 | 4149 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4150 | need_commit = 0; |
9dced186 | 4151 | ASSERT(current->journal_info); |
0af3d00b JB |
4152 | } |
4153 | ||
b4d7c3c9 | 4154 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
4155 | if (!data_sinfo) |
4156 | goto alloc; | |
9ed74f2d | 4157 | |
6a63209f JB |
4158 | again: |
4159 | /* make sure we have enough space to handle the data first */ | |
4160 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
4161 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
4162 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
4163 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
4164 | |
4165 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4166 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4167 | |
6a63209f JB |
4168 | /* |
4169 | * if we don't have enough free bytes in this space then we need | |
4170 | * to alloc a new chunk. | |
4171 | */ | |
b9fd47cd | 4172 | if (!data_sinfo->full) { |
6a63209f | 4173 | u64 alloc_target; |
9ed74f2d | 4174 | |
0e4f8f88 | 4175 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4176 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 4177 | alloc: |
6a63209f | 4178 | alloc_target = btrfs_get_alloc_profile(root, 1); |
9dced186 MX |
4179 | /* |
4180 | * It is ugly that we don't call nolock join | |
4181 | * transaction for the free space inode case here. | |
4182 | * But it is safe because we only do the data space | |
4183 | * reservation for the free space cache in the | |
4184 | * transaction context, the common join transaction | |
4185 | * just increase the counter of the current transaction | |
4186 | * handler, doesn't try to acquire the trans_lock of | |
4187 | * the fs. | |
4188 | */ | |
7a7eaa40 | 4189 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4190 | if (IS_ERR(trans)) |
4191 | return PTR_ERR(trans); | |
9ed74f2d | 4192 | |
6a63209f | 4193 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
4194 | alloc_target, |
4195 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 4196 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
4197 | if (ret < 0) { |
4198 | if (ret != -ENOSPC) | |
4199 | return ret; | |
c99f1b0c ZL |
4200 | else { |
4201 | have_pinned_space = 1; | |
d52a5b5f | 4202 | goto commit_trans; |
c99f1b0c | 4203 | } |
d52a5b5f | 4204 | } |
9ed74f2d | 4205 | |
b4d7c3c9 LZ |
4206 | if (!data_sinfo) |
4207 | data_sinfo = fs_info->data_sinfo; | |
4208 | ||
6a63209f JB |
4209 | goto again; |
4210 | } | |
f2bb8f5c JB |
4211 | |
4212 | /* | |
b150a4f1 | 4213 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4214 | * allocation, and no removed chunk in current transaction, |
4215 | * don't bother committing the transaction. | |
f2bb8f5c | 4216 | */ |
c99f1b0c ZL |
4217 | have_pinned_space = percpu_counter_compare( |
4218 | &data_sinfo->total_bytes_pinned, | |
4219 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4220 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4221 | |
4e06bdd6 | 4222 | /* commit the current transaction and try again */ |
d52a5b5f | 4223 | commit_trans: |
c99f1b0c | 4224 | if (need_commit && |
a4abeea4 | 4225 | !atomic_read(&root->fs_info->open_ioctl_trans)) { |
c99f1b0c | 4226 | need_commit--; |
b150a4f1 | 4227 | |
e1746e83 ZL |
4228 | if (need_commit > 0) { |
4229 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
578def7c | 4230 | btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1); |
e1746e83 | 4231 | } |
9a4e7276 | 4232 | |
7a7eaa40 | 4233 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4234 | if (IS_ERR(trans)) |
4235 | return PTR_ERR(trans); | |
c99f1b0c | 4236 | if (have_pinned_space >= 0 || |
3204d33c JB |
4237 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4238 | &trans->transaction->flags) || | |
c99f1b0c | 4239 | need_commit > 0) { |
94b947b2 ZL |
4240 | ret = btrfs_commit_transaction(trans, root); |
4241 | if (ret) | |
4242 | return ret; | |
d7c15171 | 4243 | /* |
c2d6cb16 FM |
4244 | * The cleaner kthread might still be doing iput |
4245 | * operations. Wait for it to finish so that | |
4246 | * more space is released. | |
d7c15171 | 4247 | */ |
c2d6cb16 FM |
4248 | mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex); |
4249 | mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4250 | goto again; |
4251 | } else { | |
4252 | btrfs_end_transaction(trans, root); | |
4253 | } | |
4e06bdd6 | 4254 | } |
9ed74f2d | 4255 | |
cab45e22 JM |
4256 | trace_btrfs_space_reservation(root->fs_info, |
4257 | "space_info:enospc", | |
4258 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4259 | return -ENOSPC; |
4260 | } | |
4261 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 4262 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 4263 | data_sinfo->flags, bytes, 1); |
6a63209f | 4264 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4265 | |
237c0e9f | 4266 | return ret; |
9ed74f2d | 4267 | } |
6a63209f | 4268 | |
4ceff079 QW |
4269 | /* |
4270 | * New check_data_free_space() with ability for precious data reservation | |
4271 | * Will replace old btrfs_check_data_free_space(), but for patch split, | |
4272 | * add a new function first and then replace it. | |
4273 | */ | |
7cf5b976 | 4274 | int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len) |
4ceff079 QW |
4275 | { |
4276 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4277 | int ret; | |
4278 | ||
4279 | /* align the range */ | |
4280 | len = round_up(start + len, root->sectorsize) - | |
4281 | round_down(start, root->sectorsize); | |
4282 | start = round_down(start, root->sectorsize); | |
4283 | ||
4284 | ret = btrfs_alloc_data_chunk_ondemand(inode, len); | |
4285 | if (ret < 0) | |
4286 | return ret; | |
4287 | ||
94ed938a QW |
4288 | /* |
4289 | * Use new btrfs_qgroup_reserve_data to reserve precious data space | |
4290 | * | |
4291 | * TODO: Find a good method to avoid reserve data space for NOCOW | |
4292 | * range, but don't impact performance on quota disable case. | |
4293 | */ | |
4ceff079 QW |
4294 | ret = btrfs_qgroup_reserve_data(inode, start, len); |
4295 | return ret; | |
4296 | } | |
4297 | ||
4ceff079 QW |
4298 | /* |
4299 | * Called if we need to clear a data reservation for this inode | |
4300 | * Normally in a error case. | |
4301 | * | |
51773bec QW |
4302 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4303 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4304 | * Like clear_bit_hook(). | |
4ceff079 | 4305 | */ |
51773bec QW |
4306 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4307 | u64 len) | |
4ceff079 QW |
4308 | { |
4309 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4310 | struct btrfs_space_info *data_sinfo; | |
4311 | ||
4312 | /* Make sure the range is aligned to sectorsize */ | |
4313 | len = round_up(start + len, root->sectorsize) - | |
4314 | round_down(start, root->sectorsize); | |
4315 | start = round_down(start, root->sectorsize); | |
4316 | ||
4ceff079 QW |
4317 | data_sinfo = root->fs_info->data_sinfo; |
4318 | spin_lock(&data_sinfo->lock); | |
4319 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4320 | data_sinfo->bytes_may_use = 0; | |
4321 | else | |
4322 | data_sinfo->bytes_may_use -= len; | |
4323 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4324 | data_sinfo->flags, len, 0); | |
4325 | spin_unlock(&data_sinfo->lock); | |
4326 | } | |
4327 | ||
51773bec QW |
4328 | /* |
4329 | * Called if we need to clear a data reservation for this inode | |
4330 | * Normally in a error case. | |
4331 | * | |
01327610 | 4332 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4333 | * space framework. |
4334 | */ | |
4335 | void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len) | |
4336 | { | |
4337 | btrfs_free_reserved_data_space_noquota(inode, start, len); | |
4338 | btrfs_qgroup_free_data(inode, start, len); | |
4339 | } | |
4340 | ||
97e728d4 | 4341 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4342 | { |
97e728d4 JB |
4343 | struct list_head *head = &info->space_info; |
4344 | struct btrfs_space_info *found; | |
e3ccfa98 | 4345 | |
97e728d4 JB |
4346 | rcu_read_lock(); |
4347 | list_for_each_entry_rcu(found, head, list) { | |
4348 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4349 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4350 | } |
97e728d4 | 4351 | rcu_read_unlock(); |
e3ccfa98 JB |
4352 | } |
4353 | ||
3c76cd84 MX |
4354 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4355 | { | |
4356 | return (global->size << 1); | |
4357 | } | |
4358 | ||
e5bc2458 | 4359 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 4360 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4361 | { |
fb25e914 | 4362 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 4363 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 4364 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 4365 | u64 thresh; |
e3ccfa98 | 4366 | |
0e4f8f88 CM |
4367 | if (force == CHUNK_ALLOC_FORCE) |
4368 | return 1; | |
4369 | ||
fb25e914 JB |
4370 | /* |
4371 | * We need to take into account the global rsv because for all intents | |
4372 | * and purposes it's used space. Don't worry about locking the | |
4373 | * global_rsv, it doesn't change except when the transaction commits. | |
4374 | */ | |
54338b5c | 4375 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3c76cd84 | 4376 | num_allocated += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4377 | |
0e4f8f88 CM |
4378 | /* |
4379 | * in limited mode, we want to have some free space up to | |
4380 | * about 1% of the FS size. | |
4381 | */ | |
4382 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 4383 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
ee22184b | 4384 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 CM |
4385 | |
4386 | if (num_bytes - num_allocated < thresh) | |
4387 | return 1; | |
4388 | } | |
0e4f8f88 | 4389 | |
ee22184b | 4390 | if (num_allocated + SZ_2M < div_factor(num_bytes, 8)) |
14ed0ca6 | 4391 | return 0; |
424499db | 4392 | return 1; |
32c00aff JB |
4393 | } |
4394 | ||
39c2d7fa | 4395 | static u64 get_profile_num_devs(struct btrfs_root *root, u64 type) |
15d1ff81 LB |
4396 | { |
4397 | u64 num_dev; | |
4398 | ||
53b381b3 DW |
4399 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4400 | BTRFS_BLOCK_GROUP_RAID0 | | |
4401 | BTRFS_BLOCK_GROUP_RAID5 | | |
4402 | BTRFS_BLOCK_GROUP_RAID6)) | |
15d1ff81 LB |
4403 | num_dev = root->fs_info->fs_devices->rw_devices; |
4404 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
4405 | num_dev = 2; | |
4406 | else | |
4407 | num_dev = 1; /* DUP or single */ | |
4408 | ||
39c2d7fa | 4409 | return num_dev; |
15d1ff81 LB |
4410 | } |
4411 | ||
39c2d7fa FM |
4412 | /* |
4413 | * If @is_allocation is true, reserve space in the system space info necessary | |
4414 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4415 | * removing a chunk. | |
4416 | */ | |
4417 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
4418 | struct btrfs_root *root, | |
4617ea3a | 4419 | u64 type) |
15d1ff81 LB |
4420 | { |
4421 | struct btrfs_space_info *info; | |
4422 | u64 left; | |
4423 | u64 thresh; | |
4fbcdf66 | 4424 | int ret = 0; |
39c2d7fa | 4425 | u64 num_devs; |
4fbcdf66 FM |
4426 | |
4427 | /* | |
4428 | * Needed because we can end up allocating a system chunk and for an | |
4429 | * atomic and race free space reservation in the chunk block reserve. | |
4430 | */ | |
4431 | ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex)); | |
15d1ff81 LB |
4432 | |
4433 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4434 | spin_lock(&info->lock); | |
4435 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
4fbcdf66 FM |
4436 | info->bytes_reserved - info->bytes_readonly - |
4437 | info->bytes_may_use; | |
15d1ff81 LB |
4438 | spin_unlock(&info->lock); |
4439 | ||
39c2d7fa FM |
4440 | num_devs = get_profile_num_devs(root, type); |
4441 | ||
4442 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
4617ea3a FM |
4443 | thresh = btrfs_calc_trunc_metadata_size(root, num_devs) + |
4444 | btrfs_calc_trans_metadata_size(root, 1); | |
39c2d7fa | 4445 | |
15d1ff81 | 4446 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { |
c2cf52eb SK |
4447 | btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu", |
4448 | left, thresh, type); | |
15d1ff81 LB |
4449 | dump_space_info(info, 0, 0); |
4450 | } | |
4451 | ||
4452 | if (left < thresh) { | |
4453 | u64 flags; | |
4454 | ||
4455 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
4fbcdf66 FM |
4456 | /* |
4457 | * Ignore failure to create system chunk. We might end up not | |
4458 | * needing it, as we might not need to COW all nodes/leafs from | |
4459 | * the paths we visit in the chunk tree (they were already COWed | |
4460 | * or created in the current transaction for example). | |
4461 | */ | |
4462 | ret = btrfs_alloc_chunk(trans, root, flags); | |
4463 | } | |
4464 | ||
4465 | if (!ret) { | |
4466 | ret = btrfs_block_rsv_add(root->fs_info->chunk_root, | |
4467 | &root->fs_info->chunk_block_rsv, | |
4468 | thresh, BTRFS_RESERVE_NO_FLUSH); | |
4469 | if (!ret) | |
4470 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4471 | } |
4472 | } | |
4473 | ||
6324fbf3 | 4474 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 4475 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 4476 | { |
6324fbf3 | 4477 | struct btrfs_space_info *space_info; |
97e728d4 | 4478 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 4479 | int wait_for_alloc = 0; |
9ed74f2d | 4480 | int ret = 0; |
9ed74f2d | 4481 | |
c6b305a8 JB |
4482 | /* Don't re-enter if we're already allocating a chunk */ |
4483 | if (trans->allocating_chunk) | |
4484 | return -ENOSPC; | |
4485 | ||
6324fbf3 | 4486 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
4487 | if (!space_info) { |
4488 | ret = update_space_info(extent_root->fs_info, flags, | |
e40edf2d | 4489 | 0, 0, 0, &space_info); |
79787eaa | 4490 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 4491 | } |
79787eaa | 4492 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 4493 | |
6d74119f | 4494 | again: |
25179201 | 4495 | spin_lock(&space_info->lock); |
9e622d6b | 4496 | if (force < space_info->force_alloc) |
0e4f8f88 | 4497 | force = space_info->force_alloc; |
25179201 | 4498 | if (space_info->full) { |
09fb99a6 FDBM |
4499 | if (should_alloc_chunk(extent_root, space_info, force)) |
4500 | ret = -ENOSPC; | |
4501 | else | |
4502 | ret = 0; | |
25179201 | 4503 | spin_unlock(&space_info->lock); |
09fb99a6 | 4504 | return ret; |
9ed74f2d JB |
4505 | } |
4506 | ||
698d0082 | 4507 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 4508 | spin_unlock(&space_info->lock); |
6d74119f JB |
4509 | return 0; |
4510 | } else if (space_info->chunk_alloc) { | |
4511 | wait_for_alloc = 1; | |
4512 | } else { | |
4513 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4514 | } |
0e4f8f88 | 4515 | |
25179201 | 4516 | spin_unlock(&space_info->lock); |
9ed74f2d | 4517 | |
6d74119f JB |
4518 | mutex_lock(&fs_info->chunk_mutex); |
4519 | ||
4520 | /* | |
4521 | * The chunk_mutex is held throughout the entirety of a chunk | |
4522 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4523 | * other guy is done and we need to recheck and see if we should | |
4524 | * allocate. | |
4525 | */ | |
4526 | if (wait_for_alloc) { | |
4527 | mutex_unlock(&fs_info->chunk_mutex); | |
4528 | wait_for_alloc = 0; | |
4529 | goto again; | |
4530 | } | |
4531 | ||
c6b305a8 JB |
4532 | trans->allocating_chunk = true; |
4533 | ||
67377734 JB |
4534 | /* |
4535 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4536 | * allocating mixed chunks instead of individual chunks. | |
4537 | */ | |
4538 | if (btrfs_mixed_space_info(space_info)) | |
4539 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4540 | ||
97e728d4 JB |
4541 | /* |
4542 | * if we're doing a data chunk, go ahead and make sure that | |
4543 | * we keep a reasonable number of metadata chunks allocated in the | |
4544 | * FS as well. | |
4545 | */ | |
9ed74f2d | 4546 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4547 | fs_info->data_chunk_allocations++; |
4548 | if (!(fs_info->data_chunk_allocations % | |
4549 | fs_info->metadata_ratio)) | |
4550 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4551 | } |
4552 | ||
15d1ff81 LB |
4553 | /* |
4554 | * Check if we have enough space in SYSTEM chunk because we may need | |
4555 | * to update devices. | |
4556 | */ | |
4617ea3a | 4557 | check_system_chunk(trans, extent_root, flags); |
15d1ff81 | 4558 | |
2b82032c | 4559 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 4560 | trans->allocating_chunk = false; |
92b8e897 | 4561 | |
9ed74f2d | 4562 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4563 | if (ret < 0 && ret != -ENOSPC) |
4564 | goto out; | |
9ed74f2d | 4565 | if (ret) |
6324fbf3 | 4566 | space_info->full = 1; |
424499db YZ |
4567 | else |
4568 | ret = 1; | |
6d74119f | 4569 | |
0e4f8f88 | 4570 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4571 | out: |
6d74119f | 4572 | space_info->chunk_alloc = 0; |
9ed74f2d | 4573 | spin_unlock(&space_info->lock); |
a25c75d5 | 4574 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4575 | /* |
4576 | * When we allocate a new chunk we reserve space in the chunk block | |
4577 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4578 | * add new nodes/leafs to it if we end up needing to do it when | |
4579 | * inserting the chunk item and updating device items as part of the | |
4580 | * second phase of chunk allocation, performed by | |
4581 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4582 | * large number of new block groups to create in our transaction | |
4583 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4584 | * in extreme cases - like having a single transaction create many new | |
4585 | * block groups when starting to write out the free space caches of all | |
4586 | * the block groups that were made dirty during the lifetime of the | |
4587 | * transaction. | |
4588 | */ | |
d9a0540a | 4589 | if (trans->can_flush_pending_bgs && |
ee22184b | 4590 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
00d80e34 FM |
4591 | btrfs_create_pending_block_groups(trans, trans->root); |
4592 | btrfs_trans_release_chunk_metadata(trans); | |
4593 | } | |
0f9dd46c | 4594 | return ret; |
6324fbf3 | 4595 | } |
9ed74f2d | 4596 | |
a80c8dcf JB |
4597 | static int can_overcommit(struct btrfs_root *root, |
4598 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 4599 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf | 4600 | { |
957780eb JB |
4601 | struct btrfs_block_rsv *global_rsv; |
4602 | u64 profile; | |
3c76cd84 | 4603 | u64 space_size; |
a80c8dcf JB |
4604 | u64 avail; |
4605 | u64 used; | |
4606 | ||
957780eb JB |
4607 | /* Don't overcommit when in mixed mode. */ |
4608 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4609 | return 0; | |
4610 | ||
4611 | BUG_ON(root->fs_info == NULL); | |
4612 | global_rsv = &root->fs_info->global_block_rsv; | |
4613 | profile = btrfs_get_alloc_profile(root, 0); | |
a80c8dcf | 4614 | used = space_info->bytes_used + space_info->bytes_reserved + |
96f1bb57 JB |
4615 | space_info->bytes_pinned + space_info->bytes_readonly; |
4616 | ||
96f1bb57 JB |
4617 | /* |
4618 | * We only want to allow over committing if we have lots of actual space | |
4619 | * free, but if we don't have enough space to handle the global reserve | |
4620 | * space then we could end up having a real enospc problem when trying | |
4621 | * to allocate a chunk or some other such important allocation. | |
4622 | */ | |
3c76cd84 MX |
4623 | spin_lock(&global_rsv->lock); |
4624 | space_size = calc_global_rsv_need_space(global_rsv); | |
4625 | spin_unlock(&global_rsv->lock); | |
4626 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4627 | return 0; |
4628 | ||
4629 | used += space_info->bytes_may_use; | |
a80c8dcf JB |
4630 | |
4631 | spin_lock(&root->fs_info->free_chunk_lock); | |
4632 | avail = root->fs_info->free_chunk_space; | |
4633 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4634 | ||
4635 | /* | |
4636 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4637 | * space is actually useable. For raid56, the space info used |
4638 | * doesn't include the parity drive, so we don't have to | |
4639 | * change the math | |
a80c8dcf JB |
4640 | */ |
4641 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4642 | BTRFS_BLOCK_GROUP_RAID1 | | |
4643 | BTRFS_BLOCK_GROUP_RAID10)) | |
4644 | avail >>= 1; | |
4645 | ||
4646 | /* | |
561c294d MX |
4647 | * If we aren't flushing all things, let us overcommit up to |
4648 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4649 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4650 | */ |
08e007d2 | 4651 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4652 | avail >>= 3; |
a80c8dcf | 4653 | else |
14575aef | 4654 | avail >>= 1; |
a80c8dcf | 4655 | |
14575aef | 4656 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4657 | return 1; |
4658 | return 0; | |
4659 | } | |
4660 | ||
48a3b636 | 4661 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
6c255e67 | 4662 | unsigned long nr_pages, int nr_items) |
da633a42 MX |
4663 | { |
4664 | struct super_block *sb = root->fs_info->sb; | |
da633a42 | 4665 | |
925a6efb JB |
4666 | if (down_read_trylock(&sb->s_umount)) { |
4667 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4668 | up_read(&sb->s_umount); | |
4669 | } else { | |
da633a42 MX |
4670 | /* |
4671 | * We needn't worry the filesystem going from r/w to r/o though | |
4672 | * we don't acquire ->s_umount mutex, because the filesystem | |
4673 | * should guarantee the delalloc inodes list be empty after | |
4674 | * the filesystem is readonly(all dirty pages are written to | |
4675 | * the disk). | |
4676 | */ | |
6c255e67 | 4677 | btrfs_start_delalloc_roots(root->fs_info, 0, nr_items); |
98ad69cf | 4678 | if (!current->journal_info) |
578def7c FM |
4679 | btrfs_wait_ordered_roots(root->fs_info, nr_items, |
4680 | 0, (u64)-1); | |
da633a42 MX |
4681 | } |
4682 | } | |
4683 | ||
18cd8ea6 MX |
4684 | static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim) |
4685 | { | |
4686 | u64 bytes; | |
4687 | int nr; | |
4688 | ||
4689 | bytes = btrfs_calc_trans_metadata_size(root, 1); | |
4690 | nr = (int)div64_u64(to_reclaim, bytes); | |
4691 | if (!nr) | |
4692 | nr = 1; | |
4693 | return nr; | |
4694 | } | |
4695 | ||
ee22184b | 4696 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4697 | |
9ed74f2d | 4698 | /* |
5da9d01b | 4699 | * shrink metadata reservation for delalloc |
9ed74f2d | 4700 | */ |
f4c738c2 JB |
4701 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
4702 | bool wait_ordered) | |
5da9d01b | 4703 | { |
0ca1f7ce | 4704 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4705 | struct btrfs_space_info *space_info; |
663350ac | 4706 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4707 | u64 delalloc_bytes; |
5da9d01b | 4708 | u64 max_reclaim; |
b1953bce | 4709 | long time_left; |
d3ee29e3 MX |
4710 | unsigned long nr_pages; |
4711 | int loops; | |
b0244199 | 4712 | int items; |
08e007d2 | 4713 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4714 | |
c61a16a7 | 4715 | /* Calc the number of the pages we need flush for space reservation */ |
b0244199 | 4716 | items = calc_reclaim_items_nr(root, to_reclaim); |
8eb0dfdb | 4717 | to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4718 | |
663350ac | 4719 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 4720 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 4721 | space_info = block_rsv->space_info; |
bf9022e0 | 4722 | |
963d678b MX |
4723 | delalloc_bytes = percpu_counter_sum_positive( |
4724 | &root->fs_info->delalloc_bytes); | |
f4c738c2 | 4725 | if (delalloc_bytes == 0) { |
fdb5effd | 4726 | if (trans) |
f4c738c2 | 4727 | return; |
38c135af | 4728 | if (wait_ordered) |
578def7c FM |
4729 | btrfs_wait_ordered_roots(root->fs_info, items, |
4730 | 0, (u64)-1); | |
f4c738c2 | 4731 | return; |
fdb5effd JB |
4732 | } |
4733 | ||
d3ee29e3 | 4734 | loops = 0; |
f4c738c2 JB |
4735 | while (delalloc_bytes && loops < 3) { |
4736 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4737 | nr_pages = max_reclaim >> PAGE_SHIFT; |
6c255e67 | 4738 | btrfs_writeback_inodes_sb_nr(root, nr_pages, items); |
dea31f52 JB |
4739 | /* |
4740 | * We need to wait for the async pages to actually start before | |
4741 | * we do anything. | |
4742 | */ | |
9f3a074d MX |
4743 | max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages); |
4744 | if (!max_reclaim) | |
4745 | goto skip_async; | |
4746 | ||
4747 | if (max_reclaim <= nr_pages) | |
4748 | max_reclaim = 0; | |
4749 | else | |
4750 | max_reclaim -= nr_pages; | |
dea31f52 | 4751 | |
9f3a074d MX |
4752 | wait_event(root->fs_info->async_submit_wait, |
4753 | atomic_read(&root->fs_info->async_delalloc_pages) <= | |
4754 | (int)max_reclaim); | |
4755 | skip_async: | |
08e007d2 MX |
4756 | if (!trans) |
4757 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4758 | else | |
4759 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4760 | spin_lock(&space_info->lock); |
08e007d2 | 4761 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
4762 | spin_unlock(&space_info->lock); |
4763 | break; | |
4764 | } | |
957780eb JB |
4765 | if (list_empty(&space_info->tickets) && |
4766 | list_empty(&space_info->priority_tickets)) { | |
4767 | spin_unlock(&space_info->lock); | |
4768 | break; | |
4769 | } | |
0019f10d | 4770 | spin_unlock(&space_info->lock); |
5da9d01b | 4771 | |
36e39c40 | 4772 | loops++; |
f104d044 | 4773 | if (wait_ordered && !trans) { |
578def7c FM |
4774 | btrfs_wait_ordered_roots(root->fs_info, items, |
4775 | 0, (u64)-1); | |
f104d044 | 4776 | } else { |
f4c738c2 | 4777 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4778 | if (time_left) |
4779 | break; | |
4780 | } | |
963d678b MX |
4781 | delalloc_bytes = percpu_counter_sum_positive( |
4782 | &root->fs_info->delalloc_bytes); | |
5da9d01b | 4783 | } |
5da9d01b YZ |
4784 | } |
4785 | ||
663350ac JB |
4786 | /** |
4787 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4788 | * @root - the root we're allocating for | |
4789 | * @bytes - the number of bytes we want to reserve | |
4790 | * @force - force the commit | |
8bb8ab2e | 4791 | * |
663350ac JB |
4792 | * This will check to make sure that committing the transaction will actually |
4793 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4794 | * will return -ENOSPC. | |
8bb8ab2e | 4795 | */ |
663350ac JB |
4796 | static int may_commit_transaction(struct btrfs_root *root, |
4797 | struct btrfs_space_info *space_info, | |
4798 | u64 bytes, int force) | |
4799 | { | |
4800 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
4801 | struct btrfs_trans_handle *trans; | |
4802 | ||
4803 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4804 | if (trans) | |
4805 | return -EAGAIN; | |
4806 | ||
4807 | if (force) | |
4808 | goto commit; | |
4809 | ||
4810 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4811 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4812 | bytes) >= 0) |
663350ac | 4813 | goto commit; |
663350ac JB |
4814 | |
4815 | /* | |
4816 | * See if there is some space in the delayed insertion reservation for | |
4817 | * this reservation. | |
4818 | */ | |
4819 | if (space_info != delayed_rsv->space_info) | |
4820 | return -ENOSPC; | |
4821 | ||
4822 | spin_lock(&delayed_rsv->lock); | |
b150a4f1 JB |
4823 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
4824 | bytes - delayed_rsv->size) >= 0) { | |
663350ac JB |
4825 | spin_unlock(&delayed_rsv->lock); |
4826 | return -ENOSPC; | |
4827 | } | |
4828 | spin_unlock(&delayed_rsv->lock); | |
4829 | ||
4830 | commit: | |
4831 | trans = btrfs_join_transaction(root); | |
4832 | if (IS_ERR(trans)) | |
4833 | return -ENOSPC; | |
4834 | ||
4835 | return btrfs_commit_transaction(trans, root); | |
4836 | } | |
4837 | ||
96c3f433 | 4838 | enum flush_state { |
67b0fd63 JB |
4839 | FLUSH_DELAYED_ITEMS_NR = 1, |
4840 | FLUSH_DELAYED_ITEMS = 2, | |
4841 | FLUSH_DELALLOC = 3, | |
4842 | FLUSH_DELALLOC_WAIT = 4, | |
ea658bad JB |
4843 | ALLOC_CHUNK = 5, |
4844 | COMMIT_TRANS = 6, | |
96c3f433 JB |
4845 | }; |
4846 | ||
957780eb JB |
4847 | struct reserve_ticket { |
4848 | u64 bytes; | |
4849 | int error; | |
4850 | struct list_head list; | |
4851 | wait_queue_head_t wait; | |
4852 | }; | |
4853 | ||
96c3f433 JB |
4854 | static int flush_space(struct btrfs_root *root, |
4855 | struct btrfs_space_info *space_info, u64 num_bytes, | |
4856 | u64 orig_bytes, int state) | |
4857 | { | |
4858 | struct btrfs_trans_handle *trans; | |
4859 | int nr; | |
f4c738c2 | 4860 | int ret = 0; |
96c3f433 JB |
4861 | |
4862 | switch (state) { | |
96c3f433 JB |
4863 | case FLUSH_DELAYED_ITEMS_NR: |
4864 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 MX |
4865 | if (state == FLUSH_DELAYED_ITEMS_NR) |
4866 | nr = calc_reclaim_items_nr(root, num_bytes) * 2; | |
4867 | else | |
96c3f433 | 4868 | nr = -1; |
18cd8ea6 | 4869 | |
96c3f433 JB |
4870 | trans = btrfs_join_transaction(root); |
4871 | if (IS_ERR(trans)) { | |
4872 | ret = PTR_ERR(trans); | |
4873 | break; | |
4874 | } | |
4875 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
4876 | btrfs_end_transaction(trans, root); | |
4877 | break; | |
67b0fd63 JB |
4878 | case FLUSH_DELALLOC: |
4879 | case FLUSH_DELALLOC_WAIT: | |
24af7dd1 | 4880 | shrink_delalloc(root, num_bytes * 2, orig_bytes, |
67b0fd63 JB |
4881 | state == FLUSH_DELALLOC_WAIT); |
4882 | break; | |
ea658bad JB |
4883 | case ALLOC_CHUNK: |
4884 | trans = btrfs_join_transaction(root); | |
4885 | if (IS_ERR(trans)) { | |
4886 | ret = PTR_ERR(trans); | |
4887 | break; | |
4888 | } | |
4889 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
4890 | btrfs_get_alloc_profile(root, 0), |
4891 | CHUNK_ALLOC_NO_FORCE); | |
4892 | btrfs_end_transaction(trans, root); | |
4893 | if (ret == -ENOSPC) | |
4894 | ret = 0; | |
4895 | break; | |
96c3f433 JB |
4896 | case COMMIT_TRANS: |
4897 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
4898 | break; | |
4899 | default: | |
4900 | ret = -ENOSPC; | |
4901 | break; | |
4902 | } | |
4903 | ||
4904 | return ret; | |
4905 | } | |
21c7e756 MX |
4906 | |
4907 | static inline u64 | |
4908 | btrfs_calc_reclaim_metadata_size(struct btrfs_root *root, | |
4909 | struct btrfs_space_info *space_info) | |
4910 | { | |
957780eb | 4911 | struct reserve_ticket *ticket; |
21c7e756 MX |
4912 | u64 used; |
4913 | u64 expected; | |
957780eb | 4914 | u64 to_reclaim = 0; |
21c7e756 | 4915 | |
ee22184b | 4916 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
21c7e756 | 4917 | if (can_overcommit(root, space_info, to_reclaim, |
957780eb JB |
4918 | BTRFS_RESERVE_FLUSH_ALL)) |
4919 | return 0; | |
4920 | ||
4921 | list_for_each_entry(ticket, &space_info->tickets, list) | |
4922 | to_reclaim += ticket->bytes; | |
4923 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
4924 | to_reclaim += ticket->bytes; | |
4925 | if (to_reclaim) | |
4926 | return to_reclaim; | |
21c7e756 MX |
4927 | |
4928 | used = space_info->bytes_used + space_info->bytes_reserved + | |
4929 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4930 | space_info->bytes_may_use; | |
ee22184b | 4931 | if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL)) |
21c7e756 MX |
4932 | expected = div_factor_fine(space_info->total_bytes, 95); |
4933 | else | |
4934 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4935 | ||
4936 | if (used > expected) | |
4937 | to_reclaim = used - expected; | |
4938 | else | |
4939 | to_reclaim = 0; | |
4940 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4941 | space_info->bytes_reserved); | |
21c7e756 MX |
4942 | return to_reclaim; |
4943 | } | |
4944 | ||
4945 | static inline int need_do_async_reclaim(struct btrfs_space_info *space_info, | |
4946 | struct btrfs_fs_info *fs_info, u64 used) | |
4947 | { | |
365c5313 JB |
4948 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4949 | ||
4950 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 4951 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
4952 | return 0; |
4953 | ||
4954 | return (used >= thresh && !btrfs_fs_closing(fs_info) && | |
21c7e756 MX |
4955 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); |
4956 | } | |
4957 | ||
957780eb | 4958 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 4959 | { |
957780eb | 4960 | struct reserve_ticket *ticket; |
21c7e756 | 4961 | |
957780eb JB |
4962 | while (!list_empty(head)) { |
4963 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
4964 | list_del_init(&ticket->list); | |
4965 | ticket->error = -ENOSPC; | |
4966 | wake_up(&ticket->wait); | |
25ce459c | 4967 | } |
21c7e756 MX |
4968 | } |
4969 | ||
957780eb JB |
4970 | /* |
4971 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
4972 | * will loop and continuously try to flush as long as we are making progress. | |
4973 | * We count progress as clearing off tickets each time we have to loop. | |
4974 | */ | |
21c7e756 MX |
4975 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
4976 | { | |
957780eb | 4977 | struct reserve_ticket *last_ticket = NULL; |
21c7e756 MX |
4978 | struct btrfs_fs_info *fs_info; |
4979 | struct btrfs_space_info *space_info; | |
4980 | u64 to_reclaim; | |
4981 | int flush_state; | |
957780eb | 4982 | int commit_cycles = 0; |
21c7e756 MX |
4983 | |
4984 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4985 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4986 | ||
957780eb | 4987 | spin_lock(&space_info->lock); |
21c7e756 MX |
4988 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, |
4989 | space_info); | |
957780eb JB |
4990 | if (!to_reclaim) { |
4991 | space_info->flush = 0; | |
4992 | spin_unlock(&space_info->lock); | |
21c7e756 | 4993 | return; |
957780eb JB |
4994 | } |
4995 | last_ticket = list_first_entry(&space_info->tickets, | |
4996 | struct reserve_ticket, list); | |
4997 | spin_unlock(&space_info->lock); | |
21c7e756 MX |
4998 | |
4999 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb JB |
5000 | do { |
5001 | struct reserve_ticket *ticket; | |
5002 | int ret; | |
5003 | ||
5004 | ret = flush_space(fs_info->fs_root, space_info, to_reclaim, | |
5005 | to_reclaim, flush_state); | |
5006 | spin_lock(&space_info->lock); | |
5007 | if (list_empty(&space_info->tickets)) { | |
5008 | space_info->flush = 0; | |
5009 | spin_unlock(&space_info->lock); | |
5010 | return; | |
5011 | } | |
5012 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, | |
5013 | space_info); | |
5014 | ticket = list_first_entry(&space_info->tickets, | |
5015 | struct reserve_ticket, list); | |
5016 | if (last_ticket == ticket) { | |
5017 | flush_state++; | |
5018 | } else { | |
5019 | last_ticket = ticket; | |
5020 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5021 | if (commit_cycles) | |
5022 | commit_cycles--; | |
5023 | } | |
5024 | ||
5025 | if (flush_state > COMMIT_TRANS) { | |
5026 | commit_cycles++; | |
5027 | if (commit_cycles > 2) { | |
5028 | wake_all_tickets(&space_info->tickets); | |
5029 | space_info->flush = 0; | |
5030 | } else { | |
5031 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5032 | } | |
5033 | } | |
5034 | spin_unlock(&space_info->lock); | |
5035 | } while (flush_state <= COMMIT_TRANS); | |
5036 | } | |
5037 | ||
5038 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5039 | { | |
5040 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5041 | } | |
5042 | ||
5043 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5044 | struct btrfs_space_info *space_info, | |
5045 | struct reserve_ticket *ticket) | |
5046 | { | |
5047 | u64 to_reclaim; | |
5048 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5049 | ||
5050 | spin_lock(&space_info->lock); | |
5051 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, | |
5052 | space_info); | |
5053 | if (!to_reclaim) { | |
5054 | spin_unlock(&space_info->lock); | |
5055 | return; | |
5056 | } | |
5057 | spin_unlock(&space_info->lock); | |
5058 | ||
21c7e756 MX |
5059 | do { |
5060 | flush_space(fs_info->fs_root, space_info, to_reclaim, | |
5061 | to_reclaim, flush_state); | |
5062 | flush_state++; | |
957780eb JB |
5063 | spin_lock(&space_info->lock); |
5064 | if (ticket->bytes == 0) { | |
5065 | spin_unlock(&space_info->lock); | |
21c7e756 | 5066 | return; |
957780eb JB |
5067 | } |
5068 | spin_unlock(&space_info->lock); | |
5069 | ||
5070 | /* | |
5071 | * Priority flushers can't wait on delalloc without | |
5072 | * deadlocking. | |
5073 | */ | |
5074 | if (flush_state == FLUSH_DELALLOC || | |
5075 | flush_state == FLUSH_DELALLOC_WAIT) | |
5076 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5077 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5078 | } |
5079 | ||
957780eb JB |
5080 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5081 | struct btrfs_space_info *space_info, | |
5082 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5083 | ||
21c7e756 | 5084 | { |
957780eb JB |
5085 | DEFINE_WAIT(wait); |
5086 | int ret = 0; | |
5087 | ||
5088 | spin_lock(&space_info->lock); | |
5089 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5090 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5091 | if (ret) { | |
5092 | ret = -EINTR; | |
5093 | break; | |
5094 | } | |
5095 | spin_unlock(&space_info->lock); | |
5096 | ||
5097 | schedule(); | |
5098 | ||
5099 | finish_wait(&ticket->wait, &wait); | |
5100 | spin_lock(&space_info->lock); | |
5101 | } | |
5102 | if (!ret) | |
5103 | ret = ticket->error; | |
5104 | if (!list_empty(&ticket->list)) | |
5105 | list_del_init(&ticket->list); | |
5106 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5107 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5108 | space_info->bytes_may_use -= num_bytes; | |
5109 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5110 | space_info->flags, num_bytes, 0); | |
5111 | } | |
5112 | spin_unlock(&space_info->lock); | |
5113 | ||
5114 | return ret; | |
21c7e756 MX |
5115 | } |
5116 | ||
4a92b1b8 JB |
5117 | /** |
5118 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5119 | * @root - the root we're allocating for | |
957780eb | 5120 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5121 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5122 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5123 | * |
01327610 | 5124 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5125 | * with the block_rsv. If there is not enough space it will make an attempt to |
5126 | * flush out space to make room. It will do this by flushing delalloc if | |
5127 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5128 | * regain reservations will be made and this will fail if there is not enough | |
5129 | * space already. | |
8bb8ab2e | 5130 | */ |
957780eb JB |
5131 | static int __reserve_metadata_bytes(struct btrfs_root *root, |
5132 | struct btrfs_space_info *space_info, | |
5133 | u64 orig_bytes, | |
5134 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5135 | { |
957780eb | 5136 | struct reserve_ticket ticket; |
2bf64758 | 5137 | u64 used; |
8bb8ab2e | 5138 | int ret = 0; |
9ed74f2d | 5139 | |
957780eb | 5140 | ASSERT(orig_bytes); |
8bb8ab2e | 5141 | spin_lock(&space_info->lock); |
fdb5effd | 5142 | ret = -ENOSPC; |
2bf64758 JB |
5143 | used = space_info->bytes_used + space_info->bytes_reserved + |
5144 | space_info->bytes_pinned + space_info->bytes_readonly + | |
5145 | space_info->bytes_may_use; | |
9ed74f2d | 5146 | |
8bb8ab2e | 5147 | /* |
957780eb JB |
5148 | * If we have enough space then hooray, make our reservation and carry |
5149 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5150 | * If not things get more complicated. | |
8bb8ab2e | 5151 | */ |
957780eb JB |
5152 | if (used + orig_bytes <= space_info->total_bytes) { |
5153 | space_info->bytes_may_use += orig_bytes; | |
5154 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
5155 | space_info->flags, orig_bytes, | |
5156 | 1); | |
5157 | ret = 0; | |
5158 | } else if (can_overcommit(root, space_info, orig_bytes, flush)) { | |
44734ed1 JB |
5159 | space_info->bytes_may_use += orig_bytes; |
5160 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
5161 | space_info->flags, orig_bytes, | |
5162 | 1); | |
5163 | ret = 0; | |
2bf64758 JB |
5164 | } |
5165 | ||
8bb8ab2e | 5166 | /* |
957780eb JB |
5167 | * If we couldn't make a reservation then setup our reservation ticket |
5168 | * and kick the async worker if it's not already running. | |
08e007d2 | 5169 | * |
957780eb JB |
5170 | * If we are a priority flusher then we just need to add our ticket to |
5171 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5172 | */ |
72bcd99d | 5173 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5174 | ticket.bytes = orig_bytes; |
5175 | ticket.error = 0; | |
5176 | init_waitqueue_head(&ticket.wait); | |
5177 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5178 | list_add_tail(&ticket.list, &space_info->tickets); | |
5179 | if (!space_info->flush) { | |
5180 | space_info->flush = 1; | |
5181 | queue_work(system_unbound_wq, | |
5182 | &root->fs_info->async_reclaim_work); | |
5183 | } | |
5184 | } else { | |
5185 | list_add_tail(&ticket.list, | |
5186 | &space_info->priority_tickets); | |
5187 | } | |
21c7e756 MX |
5188 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5189 | used += orig_bytes; | |
f6acfd50 JB |
5190 | /* |
5191 | * We will do the space reservation dance during log replay, | |
5192 | * which means we won't have fs_info->fs_root set, so don't do | |
5193 | * the async reclaim as we will panic. | |
5194 | */ | |
5195 | if (!root->fs_info->log_root_recovering && | |
5196 | need_do_async_reclaim(space_info, root->fs_info, used) && | |
21c7e756 MX |
5197 | !work_busy(&root->fs_info->async_reclaim_work)) |
5198 | queue_work(system_unbound_wq, | |
5199 | &root->fs_info->async_reclaim_work); | |
8bb8ab2e | 5200 | } |
f0486c68 | 5201 | spin_unlock(&space_info->lock); |
08e007d2 | 5202 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5203 | return ret; |
f0486c68 | 5204 | |
957780eb JB |
5205 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
5206 | return wait_reserve_ticket(root->fs_info, space_info, &ticket, | |
5207 | orig_bytes); | |
08e007d2 | 5208 | |
957780eb JB |
5209 | ret = 0; |
5210 | priority_reclaim_metadata_space(root->fs_info, space_info, &ticket); | |
5211 | spin_lock(&space_info->lock); | |
5212 | if (ticket.bytes) { | |
5213 | if (ticket.bytes < orig_bytes) { | |
5214 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5215 | space_info->bytes_may_use -= num_bytes; | |
5216 | trace_btrfs_space_reservation(root->fs_info, | |
5217 | "space_info", space_info->flags, | |
5218 | num_bytes, 0); | |
08e007d2 | 5219 | |
957780eb JB |
5220 | } |
5221 | list_del_init(&ticket.list); | |
5222 | ret = -ENOSPC; | |
5223 | } | |
5224 | spin_unlock(&space_info->lock); | |
5225 | ASSERT(list_empty(&ticket.list)); | |
5226 | return ret; | |
5227 | } | |
8bb8ab2e | 5228 | |
957780eb JB |
5229 | /** |
5230 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5231 | * @root - the root we're allocating for | |
5232 | * @block_rsv - the block_rsv we're allocating for | |
5233 | * @orig_bytes - the number of bytes we want | |
5234 | * @flush - whether or not we can flush to make our reservation | |
5235 | * | |
5236 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5237 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5238 | * flush out space to make room. It will do this by flushing delalloc if | |
5239 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5240 | * regain reservations will be made and this will fail if there is not enough | |
5241 | * space already. | |
5242 | */ | |
5243 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5244 | struct btrfs_block_rsv *block_rsv, | |
5245 | u64 orig_bytes, | |
5246 | enum btrfs_reserve_flush_enum flush) | |
5247 | { | |
5248 | int ret; | |
5249 | ||
5250 | ret = __reserve_metadata_bytes(root, block_rsv->space_info, orig_bytes, | |
5251 | flush); | |
5d80366e JB |
5252 | if (ret == -ENOSPC && |
5253 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5254 | struct btrfs_block_rsv *global_rsv = | |
5255 | &root->fs_info->global_block_rsv; | |
5256 | ||
5257 | if (block_rsv != global_rsv && | |
5258 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5259 | ret = 0; | |
5260 | } | |
cab45e22 JM |
5261 | if (ret == -ENOSPC) |
5262 | trace_btrfs_space_reservation(root->fs_info, | |
5263 | "space_info:enospc", | |
957780eb JB |
5264 | block_rsv->space_info->flags, |
5265 | orig_bytes, 1); | |
f0486c68 YZ |
5266 | return ret; |
5267 | } | |
5268 | ||
79787eaa JM |
5269 | static struct btrfs_block_rsv *get_block_rsv( |
5270 | const struct btrfs_trans_handle *trans, | |
5271 | const struct btrfs_root *root) | |
f0486c68 | 5272 | { |
4c13d758 JB |
5273 | struct btrfs_block_rsv *block_rsv = NULL; |
5274 | ||
e9cf439f AM |
5275 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
5276 | (root == root->fs_info->csum_root && trans->adding_csums) || | |
5277 | (root == root->fs_info->uuid_root)) | |
f7a81ea4 SB |
5278 | block_rsv = trans->block_rsv; |
5279 | ||
4c13d758 | 5280 | if (!block_rsv) |
f0486c68 YZ |
5281 | block_rsv = root->block_rsv; |
5282 | ||
5283 | if (!block_rsv) | |
5284 | block_rsv = &root->fs_info->empty_block_rsv; | |
5285 | ||
5286 | return block_rsv; | |
5287 | } | |
5288 | ||
5289 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5290 | u64 num_bytes) | |
5291 | { | |
5292 | int ret = -ENOSPC; | |
5293 | spin_lock(&block_rsv->lock); | |
5294 | if (block_rsv->reserved >= num_bytes) { | |
5295 | block_rsv->reserved -= num_bytes; | |
5296 | if (block_rsv->reserved < block_rsv->size) | |
5297 | block_rsv->full = 0; | |
5298 | ret = 0; | |
5299 | } | |
5300 | spin_unlock(&block_rsv->lock); | |
5301 | return ret; | |
5302 | } | |
5303 | ||
5304 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5305 | u64 num_bytes, int update_size) | |
5306 | { | |
5307 | spin_lock(&block_rsv->lock); | |
5308 | block_rsv->reserved += num_bytes; | |
5309 | if (update_size) | |
5310 | block_rsv->size += num_bytes; | |
5311 | else if (block_rsv->reserved >= block_rsv->size) | |
5312 | block_rsv->full = 1; | |
5313 | spin_unlock(&block_rsv->lock); | |
5314 | } | |
5315 | ||
d52be818 JB |
5316 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5317 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5318 | int min_factor) | |
5319 | { | |
5320 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5321 | u64 min_bytes; | |
5322 | ||
5323 | if (global_rsv->space_info != dest->space_info) | |
5324 | return -ENOSPC; | |
5325 | ||
5326 | spin_lock(&global_rsv->lock); | |
5327 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5328 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5329 | spin_unlock(&global_rsv->lock); | |
5330 | return -ENOSPC; | |
5331 | } | |
5332 | global_rsv->reserved -= num_bytes; | |
5333 | if (global_rsv->reserved < global_rsv->size) | |
5334 | global_rsv->full = 0; | |
5335 | spin_unlock(&global_rsv->lock); | |
5336 | ||
5337 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5338 | return 0; | |
5339 | } | |
5340 | ||
957780eb JB |
5341 | /* |
5342 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5343 | * basically when we're returning space from block_rsv's. | |
5344 | */ | |
5345 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5346 | struct btrfs_space_info *space_info, | |
5347 | u64 num_bytes) | |
5348 | { | |
5349 | struct reserve_ticket *ticket; | |
5350 | struct list_head *head; | |
5351 | u64 used; | |
5352 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5353 | bool check_overcommit = false; | |
5354 | ||
5355 | spin_lock(&space_info->lock); | |
5356 | head = &space_info->priority_tickets; | |
5357 | ||
5358 | /* | |
5359 | * If we are over our limit then we need to check and see if we can | |
5360 | * overcommit, and if we can't then we just need to free up our space | |
5361 | * and not satisfy any requests. | |
5362 | */ | |
5363 | used = space_info->bytes_used + space_info->bytes_reserved + | |
5364 | space_info->bytes_pinned + space_info->bytes_readonly + | |
5365 | space_info->bytes_may_use; | |
5366 | if (used - num_bytes >= space_info->total_bytes) | |
5367 | check_overcommit = true; | |
5368 | again: | |
5369 | while (!list_empty(head) && num_bytes) { | |
5370 | ticket = list_first_entry(head, struct reserve_ticket, | |
5371 | list); | |
5372 | /* | |
5373 | * We use 0 bytes because this space is already reserved, so | |
5374 | * adding the ticket space would be a double count. | |
5375 | */ | |
5376 | if (check_overcommit && | |
5377 | !can_overcommit(fs_info->extent_root, space_info, 0, | |
5378 | flush)) | |
5379 | break; | |
5380 | if (num_bytes >= ticket->bytes) { | |
5381 | list_del_init(&ticket->list); | |
5382 | num_bytes -= ticket->bytes; | |
5383 | ticket->bytes = 0; | |
5384 | wake_up(&ticket->wait); | |
5385 | } else { | |
5386 | ticket->bytes -= num_bytes; | |
5387 | num_bytes = 0; | |
5388 | } | |
5389 | } | |
5390 | ||
5391 | if (num_bytes && head == &space_info->priority_tickets) { | |
5392 | head = &space_info->tickets; | |
5393 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5394 | goto again; | |
5395 | } | |
5396 | space_info->bytes_may_use -= num_bytes; | |
5397 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5398 | space_info->flags, num_bytes, 0); | |
5399 | spin_unlock(&space_info->lock); | |
5400 | } | |
5401 | ||
5402 | /* | |
5403 | * This is for newly allocated space that isn't accounted in | |
5404 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5405 | * we use this helper. | |
5406 | */ | |
5407 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5408 | struct btrfs_space_info *space_info, | |
5409 | u64 num_bytes) | |
5410 | { | |
5411 | struct reserve_ticket *ticket; | |
5412 | struct list_head *head = &space_info->priority_tickets; | |
5413 | ||
5414 | again: | |
5415 | while (!list_empty(head) && num_bytes) { | |
5416 | ticket = list_first_entry(head, struct reserve_ticket, | |
5417 | list); | |
5418 | if (num_bytes >= ticket->bytes) { | |
5419 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5420 | space_info->flags, | |
5421 | ticket->bytes, 1); | |
5422 | list_del_init(&ticket->list); | |
5423 | num_bytes -= ticket->bytes; | |
5424 | space_info->bytes_may_use += ticket->bytes; | |
5425 | ticket->bytes = 0; | |
5426 | wake_up(&ticket->wait); | |
5427 | } else { | |
5428 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5429 | space_info->flags, | |
5430 | num_bytes, 1); | |
5431 | space_info->bytes_may_use += num_bytes; | |
5432 | ticket->bytes -= num_bytes; | |
5433 | num_bytes = 0; | |
5434 | } | |
5435 | } | |
5436 | ||
5437 | if (num_bytes && head == &space_info->priority_tickets) { | |
5438 | head = &space_info->tickets; | |
5439 | goto again; | |
5440 | } | |
5441 | } | |
5442 | ||
8c2a3ca2 JB |
5443 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
5444 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 5445 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
5446 | { |
5447 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
5448 | ||
5449 | spin_lock(&block_rsv->lock); | |
5450 | if (num_bytes == (u64)-1) | |
5451 | num_bytes = block_rsv->size; | |
5452 | block_rsv->size -= num_bytes; | |
5453 | if (block_rsv->reserved >= block_rsv->size) { | |
5454 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5455 | block_rsv->reserved = block_rsv->size; | |
5456 | block_rsv->full = 1; | |
5457 | } else { | |
5458 | num_bytes = 0; | |
5459 | } | |
5460 | spin_unlock(&block_rsv->lock); | |
5461 | ||
5462 | if (num_bytes > 0) { | |
5463 | if (dest) { | |
e9e22899 JB |
5464 | spin_lock(&dest->lock); |
5465 | if (!dest->full) { | |
5466 | u64 bytes_to_add; | |
5467 | ||
5468 | bytes_to_add = dest->size - dest->reserved; | |
5469 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5470 | dest->reserved += bytes_to_add; | |
5471 | if (dest->reserved >= dest->size) | |
5472 | dest->full = 1; | |
5473 | num_bytes -= bytes_to_add; | |
5474 | } | |
5475 | spin_unlock(&dest->lock); | |
5476 | } | |
957780eb JB |
5477 | if (num_bytes) |
5478 | space_info_add_old_bytes(fs_info, space_info, | |
5479 | num_bytes); | |
9ed74f2d | 5480 | } |
f0486c68 | 5481 | } |
4e06bdd6 | 5482 | |
25d609f8 JB |
5483 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5484 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
5485 | int update_size) | |
f0486c68 YZ |
5486 | { |
5487 | int ret; | |
9ed74f2d | 5488 | |
f0486c68 YZ |
5489 | ret = block_rsv_use_bytes(src, num_bytes); |
5490 | if (ret) | |
5491 | return ret; | |
9ed74f2d | 5492 | |
25d609f8 | 5493 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5494 | return 0; |
5495 | } | |
5496 | ||
66d8f3dd | 5497 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5498 | { |
f0486c68 YZ |
5499 | memset(rsv, 0, sizeof(*rsv)); |
5500 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5501 | rsv->type = type; |
f0486c68 YZ |
5502 | } |
5503 | ||
66d8f3dd MX |
5504 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
5505 | unsigned short type) | |
f0486c68 YZ |
5506 | { |
5507 | struct btrfs_block_rsv *block_rsv; | |
5508 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 5509 | |
f0486c68 YZ |
5510 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5511 | if (!block_rsv) | |
5512 | return NULL; | |
9ed74f2d | 5513 | |
66d8f3dd | 5514 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
5515 | block_rsv->space_info = __find_space_info(fs_info, |
5516 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
5517 | return block_rsv; |
5518 | } | |
9ed74f2d | 5519 | |
f0486c68 YZ |
5520 | void btrfs_free_block_rsv(struct btrfs_root *root, |
5521 | struct btrfs_block_rsv *rsv) | |
5522 | { | |
2aaa6655 JB |
5523 | if (!rsv) |
5524 | return; | |
dabdb640 JB |
5525 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
5526 | kfree(rsv); | |
9ed74f2d JB |
5527 | } |
5528 | ||
cdfb080e CM |
5529 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5530 | { | |
5531 | kfree(rsv); | |
5532 | } | |
5533 | ||
08e007d2 MX |
5534 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5535 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5536 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5537 | { |
f0486c68 | 5538 | int ret; |
9ed74f2d | 5539 | |
f0486c68 YZ |
5540 | if (num_bytes == 0) |
5541 | return 0; | |
8bb8ab2e | 5542 | |
61b520a9 | 5543 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5544 | if (!ret) { |
5545 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5546 | return 0; | |
5547 | } | |
9ed74f2d | 5548 | |
f0486c68 | 5549 | return ret; |
f0486c68 | 5550 | } |
9ed74f2d | 5551 | |
4a92b1b8 | 5552 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 5553 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5554 | { |
5555 | u64 num_bytes = 0; | |
f0486c68 | 5556 | int ret = -ENOSPC; |
9ed74f2d | 5557 | |
f0486c68 YZ |
5558 | if (!block_rsv) |
5559 | return 0; | |
9ed74f2d | 5560 | |
f0486c68 | 5561 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5562 | num_bytes = div_factor(block_rsv->size, min_factor); |
5563 | if (block_rsv->reserved >= num_bytes) | |
5564 | ret = 0; | |
5565 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5566 | |
36ba022a JB |
5567 | return ret; |
5568 | } | |
5569 | ||
08e007d2 MX |
5570 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5571 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5572 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5573 | { |
5574 | u64 num_bytes = 0; | |
5575 | int ret = -ENOSPC; | |
5576 | ||
5577 | if (!block_rsv) | |
5578 | return 0; | |
5579 | ||
5580 | spin_lock(&block_rsv->lock); | |
5581 | num_bytes = min_reserved; | |
13553e52 | 5582 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5583 | ret = 0; |
13553e52 | 5584 | else |
f0486c68 | 5585 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5586 | spin_unlock(&block_rsv->lock); |
13553e52 | 5587 | |
f0486c68 YZ |
5588 | if (!ret) |
5589 | return 0; | |
5590 | ||
aa38a711 | 5591 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5592 | if (!ret) { |
5593 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5594 | return 0; |
6a63209f | 5595 | } |
9ed74f2d | 5596 | |
13553e52 | 5597 | return ret; |
f0486c68 YZ |
5598 | } |
5599 | ||
f0486c68 YZ |
5600 | void btrfs_block_rsv_release(struct btrfs_root *root, |
5601 | struct btrfs_block_rsv *block_rsv, | |
5602 | u64 num_bytes) | |
5603 | { | |
5604 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
17504584 | 5605 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5606 | block_rsv->space_info != global_rsv->space_info) |
5607 | global_rsv = NULL; | |
8c2a3ca2 JB |
5608 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
5609 | num_bytes); | |
6a63209f JB |
5610 | } |
5611 | ||
5612 | /* | |
8929ecfa YZ |
5613 | * helper to calculate size of global block reservation. |
5614 | * the desired value is sum of space used by extent tree, | |
5615 | * checksum tree and root tree | |
6a63209f | 5616 | */ |
8929ecfa | 5617 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 5618 | { |
8929ecfa YZ |
5619 | struct btrfs_space_info *sinfo; |
5620 | u64 num_bytes; | |
5621 | u64 meta_used; | |
5622 | u64 data_used; | |
6c41761f | 5623 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 5624 | |
8929ecfa YZ |
5625 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
5626 | spin_lock(&sinfo->lock); | |
5627 | data_used = sinfo->bytes_used; | |
5628 | spin_unlock(&sinfo->lock); | |
33b4d47f | 5629 | |
8929ecfa YZ |
5630 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
5631 | spin_lock(&sinfo->lock); | |
6d48755d JB |
5632 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
5633 | data_used = 0; | |
8929ecfa YZ |
5634 | meta_used = sinfo->bytes_used; |
5635 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 5636 | |
8929ecfa YZ |
5637 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
5638 | csum_size * 2; | |
f8c269d7 | 5639 | num_bytes += div_u64(data_used + meta_used, 50); |
4e06bdd6 | 5640 | |
8929ecfa | 5641 | if (num_bytes * 3 > meta_used) |
f8c269d7 | 5642 | num_bytes = div_u64(meta_used, 3); |
ab6e2410 | 5643 | |
707e8a07 | 5644 | return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10); |
8929ecfa | 5645 | } |
6a63209f | 5646 | |
8929ecfa YZ |
5647 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5648 | { | |
5649 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5650 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5651 | u64 num_bytes; | |
6a63209f | 5652 | |
8929ecfa | 5653 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 5654 | |
8929ecfa | 5655 | spin_lock(&sinfo->lock); |
1f699d38 | 5656 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5657 | |
ee22184b | 5658 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5659 | |
fb4b10e5 JB |
5660 | if (block_rsv->reserved < block_rsv->size) { |
5661 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + | |
5662 | sinfo->bytes_reserved + sinfo->bytes_readonly + | |
5663 | sinfo->bytes_may_use; | |
5664 | if (sinfo->total_bytes > num_bytes) { | |
5665 | num_bytes = sinfo->total_bytes - num_bytes; | |
5666 | num_bytes = min(num_bytes, | |
5667 | block_rsv->size - block_rsv->reserved); | |
5668 | block_rsv->reserved += num_bytes; | |
5669 | sinfo->bytes_may_use += num_bytes; | |
5670 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5671 | sinfo->flags, num_bytes, | |
5672 | 1); | |
5673 | } | |
5674 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5675 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5676 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5677 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5678 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5679 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5680 | } |
182608c8 | 5681 | |
fb4b10e5 JB |
5682 | if (block_rsv->reserved == block_rsv->size) |
5683 | block_rsv->full = 1; | |
5684 | else | |
5685 | block_rsv->full = 0; | |
5686 | ||
8929ecfa | 5687 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5688 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5689 | } |
5690 | ||
f0486c68 | 5691 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5692 | { |
f0486c68 | 5693 | struct btrfs_space_info *space_info; |
6a63209f | 5694 | |
f0486c68 YZ |
5695 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5696 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5697 | |
f0486c68 | 5698 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5699 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 5700 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
5701 | fs_info->trans_block_rsv.space_info = space_info; |
5702 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5703 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5704 | |
8929ecfa YZ |
5705 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5706 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5707 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5708 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5709 | if (fs_info->quota_root) |
5710 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5711 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5712 | |
8929ecfa | 5713 | update_global_block_rsv(fs_info); |
6a63209f JB |
5714 | } |
5715 | ||
8929ecfa | 5716 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5717 | { |
8c2a3ca2 JB |
5718 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5719 | (u64)-1); | |
8929ecfa YZ |
5720 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
5721 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
5722 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
5723 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5724 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5725 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5726 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5727 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5728 | } |
5729 | ||
a22285a6 YZ |
5730 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
5731 | struct btrfs_root *root) | |
6a63209f | 5732 | { |
0e721106 JB |
5733 | if (!trans->block_rsv) |
5734 | return; | |
5735 | ||
a22285a6 YZ |
5736 | if (!trans->bytes_reserved) |
5737 | return; | |
6a63209f | 5738 | |
e77266e4 | 5739 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 5740 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 5741 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
5742 | trans->bytes_reserved = 0; |
5743 | } | |
6a63209f | 5744 | |
4fbcdf66 FM |
5745 | /* |
5746 | * To be called after all the new block groups attached to the transaction | |
5747 | * handle have been created (btrfs_create_pending_block_groups()). | |
5748 | */ | |
5749 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5750 | { | |
5751 | struct btrfs_fs_info *fs_info = trans->root->fs_info; | |
5752 | ||
5753 | if (!trans->chunk_bytes_reserved) | |
5754 | return; | |
5755 | ||
5756 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5757 | ||
5758 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5759 | trans->chunk_bytes_reserved); | |
5760 | trans->chunk_bytes_reserved = 0; | |
5761 | } | |
5762 | ||
79787eaa | 5763 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
5764 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
5765 | struct inode *inode) | |
5766 | { | |
5767 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5768 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
5769 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
5770 | ||
5771 | /* | |
fcb80c2a JB |
5772 | * We need to hold space in order to delete our orphan item once we've |
5773 | * added it, so this takes the reservation so we can release it later | |
5774 | * when we are truly done with the orphan item. | |
d68fc57b | 5775 | */ |
ff5714cc | 5776 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
5777 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
5778 | btrfs_ino(inode), num_bytes, 1); | |
25d609f8 | 5779 | return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); |
6a63209f JB |
5780 | } |
5781 | ||
d68fc57b | 5782 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 5783 | { |
d68fc57b | 5784 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 5785 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
5786 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
5787 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
5788 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
5789 | } | |
97e728d4 | 5790 | |
d5c12070 MX |
5791 | /* |
5792 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5793 | * root: the root of the parent directory | |
5794 | * rsv: block reservation | |
5795 | * items: the number of items that we need do reservation | |
5796 | * qgroup_reserved: used to return the reserved size in qgroup | |
5797 | * | |
5798 | * This function is used to reserve the space for snapshot/subvolume | |
5799 | * creation and deletion. Those operations are different with the | |
5800 | * common file/directory operations, they change two fs/file trees | |
5801 | * and root tree, the number of items that the qgroup reserves is | |
5802 | * different with the free space reservation. So we can not use | |
01327610 | 5803 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5804 | */ |
5805 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5806 | struct btrfs_block_rsv *rsv, | |
5807 | int items, | |
ee3441b4 JM |
5808 | u64 *qgroup_reserved, |
5809 | bool use_global_rsv) | |
a22285a6 | 5810 | { |
d5c12070 MX |
5811 | u64 num_bytes; |
5812 | int ret; | |
ee3441b4 | 5813 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
d5c12070 MX |
5814 | |
5815 | if (root->fs_info->quota_enabled) { | |
5816 | /* One for parent inode, two for dir entries */ | |
707e8a07 | 5817 | num_bytes = 3 * root->nodesize; |
7174109c | 5818 | ret = btrfs_qgroup_reserve_meta(root, num_bytes); |
d5c12070 MX |
5819 | if (ret) |
5820 | return ret; | |
5821 | } else { | |
5822 | num_bytes = 0; | |
5823 | } | |
5824 | ||
5825 | *qgroup_reserved = num_bytes; | |
5826 | ||
5827 | num_bytes = btrfs_calc_trans_metadata_size(root, items); | |
5828 | rsv->space_info = __find_space_info(root->fs_info, | |
5829 | BTRFS_BLOCK_GROUP_METADATA); | |
5830 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5831 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5832 | |
5833 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 5834 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 5835 | |
7174109c QW |
5836 | if (ret && *qgroup_reserved) |
5837 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
d5c12070 MX |
5838 | |
5839 | return ret; | |
5840 | } | |
5841 | ||
5842 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, | |
5843 | struct btrfs_block_rsv *rsv, | |
5844 | u64 qgroup_reserved) | |
5845 | { | |
5846 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
97e728d4 JB |
5847 | } |
5848 | ||
7709cde3 JB |
5849 | /** |
5850 | * drop_outstanding_extent - drop an outstanding extent | |
5851 | * @inode: the inode we're dropping the extent for | |
01327610 | 5852 | * @num_bytes: the number of bytes we're releasing. |
7709cde3 JB |
5853 | * |
5854 | * This is called when we are freeing up an outstanding extent, either called | |
5855 | * after an error or after an extent is written. This will return the number of | |
5856 | * reserved extents that need to be freed. This must be called with | |
5857 | * BTRFS_I(inode)->lock held. | |
5858 | */ | |
dcab6a3b | 5859 | static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes) |
9e0baf60 | 5860 | { |
7fd2ae21 | 5861 | unsigned drop_inode_space = 0; |
9e0baf60 | 5862 | unsigned dropped_extents = 0; |
dcab6a3b | 5863 | unsigned num_extents = 0; |
9e0baf60 | 5864 | |
dcab6a3b JB |
5865 | num_extents = (unsigned)div64_u64(num_bytes + |
5866 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5867 | BTRFS_MAX_EXTENT_SIZE); | |
5868 | ASSERT(num_extents); | |
5869 | ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents); | |
5870 | BTRFS_I(inode)->outstanding_extents -= num_extents; | |
9e0baf60 | 5871 | |
7fd2ae21 | 5872 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
5873 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5874 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 5875 | drop_inode_space = 1; |
7fd2ae21 | 5876 | |
9e0baf60 | 5877 | /* |
01327610 | 5878 | * If we have more or the same amount of outstanding extents than we have |
9e0baf60 JB |
5879 | * reserved then we need to leave the reserved extents count alone. |
5880 | */ | |
5881 | if (BTRFS_I(inode)->outstanding_extents >= | |
5882 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 5883 | return drop_inode_space; |
9e0baf60 JB |
5884 | |
5885 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
5886 | BTRFS_I(inode)->outstanding_extents; | |
5887 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 5888 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
5889 | } |
5890 | ||
7709cde3 | 5891 | /** |
01327610 NS |
5892 | * calc_csum_metadata_size - return the amount of metadata space that must be |
5893 | * reserved/freed for the given bytes. | |
7709cde3 JB |
5894 | * @inode: the inode we're manipulating |
5895 | * @num_bytes: the number of bytes in question | |
5896 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
5897 | * | |
5898 | * This adjusts the number of csum_bytes in the inode and then returns the | |
5899 | * correct amount of metadata that must either be reserved or freed. We | |
5900 | * calculate how many checksums we can fit into one leaf and then divide the | |
5901 | * number of bytes that will need to be checksumed by this value to figure out | |
5902 | * how many checksums will be required. If we are adding bytes then the number | |
5903 | * may go up and we will return the number of additional bytes that must be | |
5904 | * reserved. If it is going down we will return the number of bytes that must | |
5905 | * be freed. | |
5906 | * | |
5907 | * This must be called with BTRFS_I(inode)->lock held. | |
5908 | */ | |
5909 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
5910 | int reserve) | |
6324fbf3 | 5911 | { |
7709cde3 | 5912 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1262133b | 5913 | u64 old_csums, num_csums; |
7709cde3 JB |
5914 | |
5915 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
5916 | BTRFS_I(inode)->csum_bytes == 0) | |
5917 | return 0; | |
5918 | ||
28f75a0e | 5919 | old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); |
7709cde3 JB |
5920 | if (reserve) |
5921 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
5922 | else | |
5923 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
28f75a0e | 5924 | num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); |
7709cde3 JB |
5925 | |
5926 | /* No change, no need to reserve more */ | |
5927 | if (old_csums == num_csums) | |
5928 | return 0; | |
5929 | ||
5930 | if (reserve) | |
5931 | return btrfs_calc_trans_metadata_size(root, | |
5932 | num_csums - old_csums); | |
5933 | ||
5934 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 5935 | } |
c146afad | 5936 | |
0ca1f7ce YZ |
5937 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
5938 | { | |
5939 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5940 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 5941 | u64 to_reserve = 0; |
660d3f6c | 5942 | u64 csum_bytes; |
9e0baf60 | 5943 | unsigned nr_extents = 0; |
08e007d2 | 5944 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5945 | int ret = 0; |
c64c2bd8 | 5946 | bool delalloc_lock = true; |
88e081bf WS |
5947 | u64 to_free = 0; |
5948 | unsigned dropped; | |
48c3d480 | 5949 | bool release_extra = false; |
6324fbf3 | 5950 | |
c64c2bd8 JB |
5951 | /* If we are a free space inode we need to not flush since we will be in |
5952 | * the middle of a transaction commit. We also don't need the delalloc | |
5953 | * mutex since we won't race with anybody. We need this mostly to make | |
5954 | * lockdep shut its filthy mouth. | |
5955 | */ | |
5956 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 5957 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 JB |
5958 | delalloc_lock = false; |
5959 | } | |
c09544e0 | 5960 | |
08e007d2 MX |
5961 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
5962 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 5963 | schedule_timeout(1); |
ec44a35c | 5964 | |
c64c2bd8 JB |
5965 | if (delalloc_lock) |
5966 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
5967 | ||
0ca1f7ce | 5968 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 5969 | |
9e0baf60 | 5970 | spin_lock(&BTRFS_I(inode)->lock); |
6a41dd09 JB |
5971 | nr_extents = (unsigned)div64_u64(num_bytes + |
5972 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5973 | BTRFS_MAX_EXTENT_SIZE); | |
5974 | BTRFS_I(inode)->outstanding_extents += nr_extents; | |
9e0baf60 | 5975 | |
48c3d480 | 5976 | nr_extents = 0; |
9e0baf60 | 5977 | if (BTRFS_I(inode)->outstanding_extents > |
660d3f6c | 5978 | BTRFS_I(inode)->reserved_extents) |
48c3d480 | 5979 | nr_extents += BTRFS_I(inode)->outstanding_extents - |
9e0baf60 | 5980 | BTRFS_I(inode)->reserved_extents; |
57a45ced | 5981 | |
48c3d480 JB |
5982 | /* We always want to reserve a slot for updating the inode. */ |
5983 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents + 1); | |
7709cde3 | 5984 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 5985 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 5986 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 5987 | |
88e081bf | 5988 | if (root->fs_info->quota_enabled) { |
7174109c QW |
5989 | ret = btrfs_qgroup_reserve_meta(root, |
5990 | nr_extents * root->nodesize); | |
88e081bf WS |
5991 | if (ret) |
5992 | goto out_fail; | |
5993 | } | |
c5567237 | 5994 | |
48c3d480 | 5995 | ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush); |
88e081bf | 5996 | if (unlikely(ret)) { |
7174109c | 5997 | btrfs_qgroup_free_meta(root, nr_extents * root->nodesize); |
88e081bf | 5998 | goto out_fail; |
9e0baf60 | 5999 | } |
25179201 | 6000 | |
660d3f6c | 6001 | spin_lock(&BTRFS_I(inode)->lock); |
48c3d480 JB |
6002 | if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
6003 | &BTRFS_I(inode)->runtime_flags)) | |
6004 | release_extra = true; | |
660d3f6c JB |
6005 | BTRFS_I(inode)->reserved_extents += nr_extents; |
6006 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
6007 | |
6008 | if (delalloc_lock) | |
6009 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 6010 | |
8c2a3ca2 | 6011 | if (to_reserve) |
67871254 | 6012 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
8c2a3ca2 | 6013 | btrfs_ino(inode), to_reserve, 1); |
48c3d480 JB |
6014 | if (release_extra) |
6015 | btrfs_block_rsv_release(root, block_rsv, | |
6016 | btrfs_calc_trans_metadata_size(root, | |
6017 | 1)); | |
0ca1f7ce | 6018 | return 0; |
88e081bf WS |
6019 | |
6020 | out_fail: | |
6021 | spin_lock(&BTRFS_I(inode)->lock); | |
dcab6a3b | 6022 | dropped = drop_outstanding_extent(inode, num_bytes); |
88e081bf WS |
6023 | /* |
6024 | * If the inodes csum_bytes is the same as the original | |
6025 | * csum_bytes then we know we haven't raced with any free()ers | |
6026 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 6027 | */ |
f4881bc7 | 6028 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { |
88e081bf | 6029 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 JB |
6030 | } else { |
6031 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; | |
6032 | u64 bytes; | |
6033 | ||
6034 | /* | |
6035 | * This is tricky, but first we need to figure out how much we | |
01327610 | 6036 | * freed from any free-ers that occurred during this |
f4881bc7 JB |
6037 | * reservation, so we reset ->csum_bytes to the csum_bytes |
6038 | * before we dropped our lock, and then call the free for the | |
6039 | * number of bytes that were freed while we were trying our | |
6040 | * reservation. | |
6041 | */ | |
6042 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; | |
6043 | BTRFS_I(inode)->csum_bytes = csum_bytes; | |
6044 | to_free = calc_csum_metadata_size(inode, bytes, 0); | |
6045 | ||
6046 | ||
6047 | /* | |
6048 | * Now we need to see how much we would have freed had we not | |
6049 | * been making this reservation and our ->csum_bytes were not | |
6050 | * artificially inflated. | |
6051 | */ | |
6052 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; | |
6053 | bytes = csum_bytes - orig_csum_bytes; | |
6054 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
6055 | ||
6056 | /* | |
6057 | * Now reset ->csum_bytes to what it should be. If bytes is | |
01327610 | 6058 | * more than to_free then we would have freed more space had we |
f4881bc7 JB |
6059 | * not had an artificially high ->csum_bytes, so we need to free |
6060 | * the remainder. If bytes is the same or less then we don't | |
6061 | * need to do anything, the other free-ers did the correct | |
6062 | * thing. | |
6063 | */ | |
6064 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; | |
6065 | if (bytes > to_free) | |
6066 | to_free = bytes - to_free; | |
6067 | else | |
6068 | to_free = 0; | |
6069 | } | |
88e081bf | 6070 | spin_unlock(&BTRFS_I(inode)->lock); |
e2d1f923 | 6071 | if (dropped) |
88e081bf WS |
6072 | to_free += btrfs_calc_trans_metadata_size(root, dropped); |
6073 | ||
6074 | if (to_free) { | |
6075 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
6076 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
6077 | btrfs_ino(inode), to_free, 0); | |
6078 | } | |
6079 | if (delalloc_lock) | |
6080 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
6081 | return ret; | |
0ca1f7ce YZ |
6082 | } |
6083 | ||
7709cde3 JB |
6084 | /** |
6085 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
6086 | * @inode: the inode to release the reservation for | |
6087 | * @num_bytes: the number of bytes we're releasing | |
6088 | * | |
6089 | * This will release the metadata reservation for an inode. This can be called | |
6090 | * once we complete IO for a given set of bytes to release their metadata | |
6091 | * reservations. | |
6092 | */ | |
0ca1f7ce YZ |
6093 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
6094 | { | |
6095 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
6096 | u64 to_free = 0; |
6097 | unsigned dropped; | |
0ca1f7ce YZ |
6098 | |
6099 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 6100 | spin_lock(&BTRFS_I(inode)->lock); |
dcab6a3b | 6101 | dropped = drop_outstanding_extent(inode, num_bytes); |
97e728d4 | 6102 | |
0934856d MX |
6103 | if (num_bytes) |
6104 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
7709cde3 | 6105 | spin_unlock(&BTRFS_I(inode)->lock); |
9e0baf60 JB |
6106 | if (dropped > 0) |
6107 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 6108 | |
6a3891c5 JB |
6109 | if (btrfs_test_is_dummy_root(root)) |
6110 | return; | |
6111 | ||
8c2a3ca2 JB |
6112 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
6113 | btrfs_ino(inode), to_free, 0); | |
c5567237 | 6114 | |
0ca1f7ce YZ |
6115 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
6116 | to_free); | |
6117 | } | |
6118 | ||
1ada3a62 | 6119 | /** |
7cf5b976 | 6120 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6121 | * delalloc |
6122 | * @inode: inode we're writing to | |
6123 | * @start: start range we are writing to | |
6124 | * @len: how long the range we are writing to | |
6125 | * | |
6126 | * TODO: This function will finally replace old btrfs_delalloc_reserve_space() | |
6127 | * | |
6128 | * This will do the following things | |
6129 | * | |
6130 | * o reserve space in data space info for num bytes | |
6131 | * and reserve precious corresponding qgroup space | |
6132 | * (Done in check_data_free_space) | |
6133 | * | |
6134 | * o reserve space for metadata space, based on the number of outstanding | |
6135 | * extents and how much csums will be needed | |
6136 | * also reserve metadata space in a per root over-reserve method. | |
6137 | * o add to the inodes->delalloc_bytes | |
6138 | * o add it to the fs_info's delalloc inodes list. | |
6139 | * (Above 3 all done in delalloc_reserve_metadata) | |
6140 | * | |
6141 | * Return 0 for success | |
6142 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6143 | */ | |
7cf5b976 | 6144 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
6145 | { |
6146 | int ret; | |
6147 | ||
7cf5b976 | 6148 | ret = btrfs_check_data_free_space(inode, start, len); |
1ada3a62 QW |
6149 | if (ret < 0) |
6150 | return ret; | |
6151 | ret = btrfs_delalloc_reserve_metadata(inode, len); | |
6152 | if (ret < 0) | |
7cf5b976 | 6153 | btrfs_free_reserved_data_space(inode, start, len); |
1ada3a62 QW |
6154 | return ret; |
6155 | } | |
6156 | ||
7709cde3 | 6157 | /** |
7cf5b976 | 6158 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6159 | * @inode: inode we're releasing space for |
6160 | * @start: start position of the space already reserved | |
6161 | * @len: the len of the space already reserved | |
6162 | * | |
6163 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
6164 | * called in the case that we don't need the metadata AND data reservations | |
6165 | * anymore. So if there is an error or we insert an inline extent. | |
6166 | * | |
6167 | * This function will release the metadata space that was not used and will | |
6168 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6169 | * list if there are no delalloc bytes left. | |
6170 | * Also it will handle the qgroup reserved space. | |
6171 | */ | |
7cf5b976 | 6172 | void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
6173 | { |
6174 | btrfs_delalloc_release_metadata(inode, len); | |
7cf5b976 | 6175 | btrfs_free_reserved_data_space(inode, start, len); |
6324fbf3 CM |
6176 | } |
6177 | ||
ce93ec54 JB |
6178 | static int update_block_group(struct btrfs_trans_handle *trans, |
6179 | struct btrfs_root *root, u64 bytenr, | |
6180 | u64 num_bytes, int alloc) | |
9078a3e1 | 6181 | { |
0af3d00b | 6182 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 6183 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 6184 | u64 total = num_bytes; |
9078a3e1 | 6185 | u64 old_val; |
db94535d | 6186 | u64 byte_in_group; |
0af3d00b | 6187 | int factor; |
3e1ad54f | 6188 | |
5d4f98a2 | 6189 | /* block accounting for super block */ |
eb73c1b7 | 6190 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6191 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6192 | if (alloc) |
6193 | old_val += num_bytes; | |
6194 | else | |
6195 | old_val -= num_bytes; | |
6c41761f | 6196 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6197 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6198 | |
d397712b | 6199 | while (total) { |
db94535d | 6200 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6201 | if (!cache) |
79787eaa | 6202 | return -ENOENT; |
b742bb82 YZ |
6203 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6204 | BTRFS_BLOCK_GROUP_RAID1 | | |
6205 | BTRFS_BLOCK_GROUP_RAID10)) | |
6206 | factor = 2; | |
6207 | else | |
6208 | factor = 1; | |
9d66e233 JB |
6209 | /* |
6210 | * If this block group has free space cache written out, we | |
6211 | * need to make sure to load it if we are removing space. This | |
6212 | * is because we need the unpinning stage to actually add the | |
6213 | * space back to the block group, otherwise we will leak space. | |
6214 | */ | |
6215 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6216 | cache_block_group(cache, 1); |
0af3d00b | 6217 | |
db94535d CM |
6218 | byte_in_group = bytenr - cache->key.objectid; |
6219 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6220 | |
25179201 | 6221 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6222 | spin_lock(&cache->lock); |
0af3d00b | 6223 | |
73bc1876 | 6224 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
6225 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6226 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6227 | ||
9078a3e1 | 6228 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6229 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6230 | if (alloc) { |
db94535d | 6231 | old_val += num_bytes; |
11833d66 YZ |
6232 | btrfs_set_block_group_used(&cache->item, old_val); |
6233 | cache->reserved -= num_bytes; | |
11833d66 | 6234 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6235 | cache->space_info->bytes_used += num_bytes; |
6236 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6237 | spin_unlock(&cache->lock); |
25179201 | 6238 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6239 | } else { |
db94535d | 6240 | old_val -= num_bytes; |
ae0ab003 FM |
6241 | btrfs_set_block_group_used(&cache->item, old_val); |
6242 | cache->pinned += num_bytes; | |
6243 | cache->space_info->bytes_pinned += num_bytes; | |
6244 | cache->space_info->bytes_used -= num_bytes; | |
6245 | cache->space_info->disk_used -= num_bytes * factor; | |
6246 | spin_unlock(&cache->lock); | |
6247 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6248 | |
ae0ab003 FM |
6249 | set_extent_dirty(info->pinned_extents, |
6250 | bytenr, bytenr + num_bytes - 1, | |
6251 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6252 | } |
1bbc621e CM |
6253 | |
6254 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6255 | if (list_empty(&cache->dirty_list)) { | |
6256 | list_add_tail(&cache->dirty_list, | |
6257 | &trans->transaction->dirty_bgs); | |
6258 | trans->transaction->num_dirty_bgs++; | |
6259 | btrfs_get_block_group(cache); | |
6260 | } | |
6261 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6262 | ||
036a9348 FM |
6263 | /* |
6264 | * No longer have used bytes in this block group, queue it for | |
6265 | * deletion. We do this after adding the block group to the | |
6266 | * dirty list to avoid races between cleaner kthread and space | |
6267 | * cache writeout. | |
6268 | */ | |
6269 | if (!alloc && old_val == 0) { | |
6270 | spin_lock(&info->unused_bgs_lock); | |
6271 | if (list_empty(&cache->bg_list)) { | |
6272 | btrfs_get_block_group(cache); | |
6273 | list_add_tail(&cache->bg_list, | |
6274 | &info->unused_bgs); | |
6275 | } | |
6276 | spin_unlock(&info->unused_bgs_lock); | |
6277 | } | |
6278 | ||
fa9c0d79 | 6279 | btrfs_put_block_group(cache); |
db94535d CM |
6280 | total -= num_bytes; |
6281 | bytenr += num_bytes; | |
9078a3e1 CM |
6282 | } |
6283 | return 0; | |
6284 | } | |
6324fbf3 | 6285 | |
a061fc8d CM |
6286 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
6287 | { | |
0f9dd46c | 6288 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6289 | u64 bytenr; |
0f9dd46c | 6290 | |
a1897fdd LB |
6291 | spin_lock(&root->fs_info->block_group_cache_lock); |
6292 | bytenr = root->fs_info->first_logical_byte; | |
6293 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
6294 | ||
6295 | if (bytenr < (u64)-1) | |
6296 | return bytenr; | |
6297 | ||
0f9dd46c JB |
6298 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
6299 | if (!cache) | |
a061fc8d | 6300 | return 0; |
0f9dd46c | 6301 | |
d2fb3437 | 6302 | bytenr = cache->key.objectid; |
fa9c0d79 | 6303 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6304 | |
6305 | return bytenr; | |
a061fc8d CM |
6306 | } |
6307 | ||
f0486c68 YZ |
6308 | static int pin_down_extent(struct btrfs_root *root, |
6309 | struct btrfs_block_group_cache *cache, | |
6310 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6311 | { |
11833d66 YZ |
6312 | spin_lock(&cache->space_info->lock); |
6313 | spin_lock(&cache->lock); | |
6314 | cache->pinned += num_bytes; | |
6315 | cache->space_info->bytes_pinned += num_bytes; | |
6316 | if (reserved) { | |
6317 | cache->reserved -= num_bytes; | |
6318 | cache->space_info->bytes_reserved -= num_bytes; | |
6319 | } | |
6320 | spin_unlock(&cache->lock); | |
6321 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6322 | |
f0486c68 YZ |
6323 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
6324 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
e2d1f923 | 6325 | if (reserved) |
0be5dc67 | 6326 | trace_btrfs_reserved_extent_free(root, bytenr, num_bytes); |
f0486c68 YZ |
6327 | return 0; |
6328 | } | |
68b38550 | 6329 | |
f0486c68 YZ |
6330 | /* |
6331 | * this function must be called within transaction | |
6332 | */ | |
6333 | int btrfs_pin_extent(struct btrfs_root *root, | |
6334 | u64 bytenr, u64 num_bytes, int reserved) | |
6335 | { | |
6336 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6337 | |
f0486c68 | 6338 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 6339 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
6340 | |
6341 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
6342 | ||
6343 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6344 | return 0; |
6345 | } | |
6346 | ||
f0486c68 | 6347 | /* |
e688b725 CM |
6348 | * this function must be called within transaction |
6349 | */ | |
dcfac415 | 6350 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
e688b725 CM |
6351 | u64 bytenr, u64 num_bytes) |
6352 | { | |
6353 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6354 | int ret; |
e688b725 CM |
6355 | |
6356 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
b50c6e25 JB |
6357 | if (!cache) |
6358 | return -EINVAL; | |
e688b725 CM |
6359 | |
6360 | /* | |
6361 | * pull in the free space cache (if any) so that our pin | |
6362 | * removes the free space from the cache. We have load_only set | |
6363 | * to one because the slow code to read in the free extents does check | |
6364 | * the pinned extents. | |
6365 | */ | |
f6373bf3 | 6366 | cache_block_group(cache, 1); |
e688b725 CM |
6367 | |
6368 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
6369 | ||
6370 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6371 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6372 | btrfs_put_block_group(cache); |
b50c6e25 | 6373 | return ret; |
e688b725 CM |
6374 | } |
6375 | ||
8c2a1a30 JB |
6376 | static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes) |
6377 | { | |
6378 | int ret; | |
6379 | struct btrfs_block_group_cache *block_group; | |
6380 | struct btrfs_caching_control *caching_ctl; | |
6381 | ||
6382 | block_group = btrfs_lookup_block_group(root->fs_info, start); | |
6383 | if (!block_group) | |
6384 | return -EINVAL; | |
6385 | ||
6386 | cache_block_group(block_group, 0); | |
6387 | caching_ctl = get_caching_control(block_group); | |
6388 | ||
6389 | if (!caching_ctl) { | |
6390 | /* Logic error */ | |
6391 | BUG_ON(!block_group_cache_done(block_group)); | |
6392 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6393 | } else { | |
6394 | mutex_lock(&caching_ctl->mutex); | |
6395 | ||
6396 | if (start >= caching_ctl->progress) { | |
6397 | ret = add_excluded_extent(root, start, num_bytes); | |
6398 | } else if (start + num_bytes <= caching_ctl->progress) { | |
6399 | ret = btrfs_remove_free_space(block_group, | |
6400 | start, num_bytes); | |
6401 | } else { | |
6402 | num_bytes = caching_ctl->progress - start; | |
6403 | ret = btrfs_remove_free_space(block_group, | |
6404 | start, num_bytes); | |
6405 | if (ret) | |
6406 | goto out_lock; | |
6407 | ||
6408 | num_bytes = (start + num_bytes) - | |
6409 | caching_ctl->progress; | |
6410 | start = caching_ctl->progress; | |
6411 | ret = add_excluded_extent(root, start, num_bytes); | |
6412 | } | |
6413 | out_lock: | |
6414 | mutex_unlock(&caching_ctl->mutex); | |
6415 | put_caching_control(caching_ctl); | |
6416 | } | |
6417 | btrfs_put_block_group(block_group); | |
6418 | return ret; | |
6419 | } | |
6420 | ||
6421 | int btrfs_exclude_logged_extents(struct btrfs_root *log, | |
6422 | struct extent_buffer *eb) | |
6423 | { | |
6424 | struct btrfs_file_extent_item *item; | |
6425 | struct btrfs_key key; | |
6426 | int found_type; | |
6427 | int i; | |
6428 | ||
6429 | if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) | |
6430 | return 0; | |
6431 | ||
6432 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6433 | btrfs_item_key_to_cpu(eb, &key, i); | |
6434 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6435 | continue; | |
6436 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6437 | found_type = btrfs_file_extent_type(eb, item); | |
6438 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6439 | continue; | |
6440 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6441 | continue; | |
6442 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6443 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
6444 | __exclude_logged_extent(log, key.objectid, key.offset); | |
6445 | } | |
6446 | ||
6447 | return 0; | |
6448 | } | |
6449 | ||
9cfa3e34 FM |
6450 | static void |
6451 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6452 | { | |
6453 | atomic_inc(&bg->reservations); | |
6454 | } | |
6455 | ||
6456 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6457 | const u64 start) | |
6458 | { | |
6459 | struct btrfs_block_group_cache *bg; | |
6460 | ||
6461 | bg = btrfs_lookup_block_group(fs_info, start); | |
6462 | ASSERT(bg); | |
6463 | if (atomic_dec_and_test(&bg->reservations)) | |
6464 | wake_up_atomic_t(&bg->reservations); | |
6465 | btrfs_put_block_group(bg); | |
6466 | } | |
6467 | ||
6468 | static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a) | |
6469 | { | |
6470 | schedule(); | |
6471 | return 0; | |
6472 | } | |
6473 | ||
6474 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6475 | { | |
6476 | struct btrfs_space_info *space_info = bg->space_info; | |
6477 | ||
6478 | ASSERT(bg->ro); | |
6479 | ||
6480 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6481 | return; | |
6482 | ||
6483 | /* | |
6484 | * Our block group is read only but before we set it to read only, | |
6485 | * some task might have had allocated an extent from it already, but it | |
6486 | * has not yet created a respective ordered extent (and added it to a | |
6487 | * root's list of ordered extents). | |
6488 | * Therefore wait for any task currently allocating extents, since the | |
6489 | * block group's reservations counter is incremented while a read lock | |
6490 | * on the groups' semaphore is held and decremented after releasing | |
6491 | * the read access on that semaphore and creating the ordered extent. | |
6492 | */ | |
6493 | down_write(&space_info->groups_sem); | |
6494 | up_write(&space_info->groups_sem); | |
6495 | ||
6496 | wait_on_atomic_t(&bg->reservations, | |
6497 | btrfs_wait_bg_reservations_atomic_t, | |
6498 | TASK_UNINTERRUPTIBLE); | |
6499 | } | |
6500 | ||
fb25e914 JB |
6501 | /** |
6502 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
6503 | * @cache: The cache we are manipulating | |
6504 | * @num_bytes: The number of bytes in question | |
6505 | * @reserve: One of the reservation enums | |
e570fd27 | 6506 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 JB |
6507 | * |
6508 | * This is called by the allocator when it reserves space, or by somebody who is | |
6509 | * freeing space that was never actually used on disk. For example if you | |
6510 | * reserve some space for a new leaf in transaction A and before transaction A | |
6511 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
6512 | * clear the reservation. | |
6513 | * | |
6514 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
6515 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
6516 | * delalloc bits in the io_tree. We have to do this since we could end up | |
6517 | * allocating less disk space for the amount of data we have reserved in the | |
6518 | * case of compression. | |
6519 | * | |
6520 | * If this is a reservation and the block group has become read only we cannot | |
6521 | * make the reservation and return -EAGAIN, otherwise this function always | |
6522 | * succeeds. | |
f0486c68 | 6523 | */ |
fb25e914 | 6524 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
e570fd27 | 6525 | u64 num_bytes, int reserve, int delalloc) |
11833d66 | 6526 | { |
fb25e914 | 6527 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6528 | int ret = 0; |
79787eaa | 6529 | |
fb25e914 JB |
6530 | spin_lock(&space_info->lock); |
6531 | spin_lock(&cache->lock); | |
6532 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
6533 | if (cache->ro) { |
6534 | ret = -EAGAIN; | |
6535 | } else { | |
fb25e914 JB |
6536 | cache->reserved += num_bytes; |
6537 | space_info->bytes_reserved += num_bytes; | |
6538 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 6539 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
6540 | "space_info", space_info->flags, |
6541 | num_bytes, 0); | |
fb25e914 JB |
6542 | space_info->bytes_may_use -= num_bytes; |
6543 | } | |
e570fd27 MX |
6544 | |
6545 | if (delalloc) | |
6546 | cache->delalloc_bytes += num_bytes; | |
f0486c68 | 6547 | } |
fb25e914 JB |
6548 | } else { |
6549 | if (cache->ro) | |
6550 | space_info->bytes_readonly += num_bytes; | |
6551 | cache->reserved -= num_bytes; | |
6552 | space_info->bytes_reserved -= num_bytes; | |
e570fd27 MX |
6553 | |
6554 | if (delalloc) | |
6555 | cache->delalloc_bytes -= num_bytes; | |
324ae4df | 6556 | } |
fb25e914 JB |
6557 | spin_unlock(&cache->lock); |
6558 | spin_unlock(&space_info->lock); | |
f0486c68 | 6559 | return ret; |
324ae4df | 6560 | } |
9078a3e1 | 6561 | |
143bede5 | 6562 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 6563 | struct btrfs_root *root) |
e8569813 | 6564 | { |
e8569813 | 6565 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
6566 | struct btrfs_caching_control *next; |
6567 | struct btrfs_caching_control *caching_ctl; | |
6568 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6569 | |
9e351cc8 | 6570 | down_write(&fs_info->commit_root_sem); |
25179201 | 6571 | |
11833d66 YZ |
6572 | list_for_each_entry_safe(caching_ctl, next, |
6573 | &fs_info->caching_block_groups, list) { | |
6574 | cache = caching_ctl->block_group; | |
6575 | if (block_group_cache_done(cache)) { | |
6576 | cache->last_byte_to_unpin = (u64)-1; | |
6577 | list_del_init(&caching_ctl->list); | |
6578 | put_caching_control(caching_ctl); | |
e8569813 | 6579 | } else { |
11833d66 | 6580 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6581 | } |
e8569813 | 6582 | } |
11833d66 YZ |
6583 | |
6584 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6585 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6586 | else | |
6587 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6588 | ||
9e351cc8 | 6589 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6590 | |
6591 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6592 | } |
6593 | ||
c759c4e1 JB |
6594 | /* |
6595 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6596 | * what it should be based on the mount options. | |
6597 | */ | |
6598 | static struct btrfs_free_cluster * | |
6599 | fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info, | |
6600 | u64 *empty_cluster) | |
6601 | { | |
6602 | struct btrfs_free_cluster *ret = NULL; | |
6603 | bool ssd = btrfs_test_opt(root, SSD); | |
6604 | ||
6605 | *empty_cluster = 0; | |
6606 | if (btrfs_mixed_space_info(space_info)) | |
6607 | return ret; | |
6608 | ||
6609 | if (ssd) | |
ee22184b | 6610 | *empty_cluster = SZ_2M; |
c759c4e1 JB |
6611 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
6612 | ret = &root->fs_info->meta_alloc_cluster; | |
6613 | if (!ssd) | |
ee22184b | 6614 | *empty_cluster = SZ_64K; |
c759c4e1 JB |
6615 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) { |
6616 | ret = &root->fs_info->data_alloc_cluster; | |
6617 | } | |
6618 | ||
6619 | return ret; | |
6620 | } | |
6621 | ||
678886bd FM |
6622 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end, |
6623 | const bool return_free_space) | |
ccd467d6 | 6624 | { |
11833d66 YZ |
6625 | struct btrfs_fs_info *fs_info = root->fs_info; |
6626 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
6627 | struct btrfs_space_info *space_info; |
6628 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6629 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6630 | u64 len; |
c759c4e1 JB |
6631 | u64 total_unpinned = 0; |
6632 | u64 empty_cluster = 0; | |
7b398f8e | 6633 | bool readonly; |
ccd467d6 | 6634 | |
11833d66 | 6635 | while (start <= end) { |
7b398f8e | 6636 | readonly = false; |
11833d66 YZ |
6637 | if (!cache || |
6638 | start >= cache->key.objectid + cache->key.offset) { | |
6639 | if (cache) | |
6640 | btrfs_put_block_group(cache); | |
c759c4e1 | 6641 | total_unpinned = 0; |
11833d66 | 6642 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6643 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 JB |
6644 | |
6645 | cluster = fetch_cluster_info(root, | |
6646 | cache->space_info, | |
6647 | &empty_cluster); | |
6648 | empty_cluster <<= 1; | |
11833d66 YZ |
6649 | } |
6650 | ||
6651 | len = cache->key.objectid + cache->key.offset - start; | |
6652 | len = min(len, end + 1 - start); | |
6653 | ||
6654 | if (start < cache->last_byte_to_unpin) { | |
6655 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6656 | if (return_free_space) |
6657 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6658 | } |
6659 | ||
f0486c68 | 6660 | start += len; |
c759c4e1 | 6661 | total_unpinned += len; |
7b398f8e | 6662 | space_info = cache->space_info; |
f0486c68 | 6663 | |
c759c4e1 JB |
6664 | /* |
6665 | * If this space cluster has been marked as fragmented and we've | |
6666 | * unpinned enough in this block group to potentially allow a | |
6667 | * cluster to be created inside of it go ahead and clear the | |
6668 | * fragmented check. | |
6669 | */ | |
6670 | if (cluster && cluster->fragmented && | |
6671 | total_unpinned > empty_cluster) { | |
6672 | spin_lock(&cluster->lock); | |
6673 | cluster->fragmented = 0; | |
6674 | spin_unlock(&cluster->lock); | |
6675 | } | |
6676 | ||
7b398f8e | 6677 | spin_lock(&space_info->lock); |
11833d66 YZ |
6678 | spin_lock(&cache->lock); |
6679 | cache->pinned -= len; | |
7b398f8e | 6680 | space_info->bytes_pinned -= len; |
4f4db217 | 6681 | space_info->max_extent_size = 0; |
d288db5d | 6682 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6683 | if (cache->ro) { |
6684 | space_info->bytes_readonly += len; | |
6685 | readonly = true; | |
6686 | } | |
11833d66 | 6687 | spin_unlock(&cache->lock); |
957780eb JB |
6688 | if (!readonly && return_free_space && |
6689 | global_rsv->space_info == space_info) { | |
6690 | u64 to_add = len; | |
6691 | WARN_ON(!return_free_space); | |
7b398f8e JB |
6692 | spin_lock(&global_rsv->lock); |
6693 | if (!global_rsv->full) { | |
957780eb JB |
6694 | to_add = min(len, global_rsv->size - |
6695 | global_rsv->reserved); | |
6696 | global_rsv->reserved += to_add; | |
6697 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6698 | if (global_rsv->reserved >= global_rsv->size) |
6699 | global_rsv->full = 1; | |
957780eb JB |
6700 | trace_btrfs_space_reservation(fs_info, |
6701 | "space_info", | |
6702 | space_info->flags, | |
6703 | to_add, 1); | |
6704 | len -= to_add; | |
7b398f8e JB |
6705 | } |
6706 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6707 | /* Add to any tickets we may have */ |
6708 | if (len) | |
6709 | space_info_add_new_bytes(fs_info, space_info, | |
6710 | len); | |
7b398f8e JB |
6711 | } |
6712 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6713 | } |
11833d66 YZ |
6714 | |
6715 | if (cache) | |
6716 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6717 | return 0; |
6718 | } | |
6719 | ||
6720 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 6721 | struct btrfs_root *root) |
a28ec197 | 6722 | { |
11833d66 | 6723 | struct btrfs_fs_info *fs_info = root->fs_info; |
e33e17ee JM |
6724 | struct btrfs_block_group_cache *block_group, *tmp; |
6725 | struct list_head *deleted_bgs; | |
11833d66 | 6726 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6727 | u64 start; |
6728 | u64 end; | |
a28ec197 | 6729 | int ret; |
a28ec197 | 6730 | |
11833d66 YZ |
6731 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6732 | unpin = &fs_info->freed_extents[1]; | |
6733 | else | |
6734 | unpin = &fs_info->freed_extents[0]; | |
6735 | ||
e33e17ee | 6736 | while (!trans->aborted) { |
d4b450cd | 6737 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6738 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6739 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6740 | if (ret) { |
6741 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6742 | break; |
d4b450cd | 6743 | } |
1f3c79a2 | 6744 | |
5378e607 LD |
6745 | if (btrfs_test_opt(root, DISCARD)) |
6746 | ret = btrfs_discard_extent(root, start, | |
6747 | end + 1 - start, NULL); | |
1f3c79a2 | 6748 | |
af6f8f60 | 6749 | clear_extent_dirty(unpin, start, end); |
678886bd | 6750 | unpin_extent_range(root, start, end, true); |
d4b450cd | 6751 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6752 | cond_resched(); |
a28ec197 | 6753 | } |
817d52f8 | 6754 | |
e33e17ee JM |
6755 | /* |
6756 | * Transaction is finished. We don't need the lock anymore. We | |
6757 | * do need to clean up the block groups in case of a transaction | |
6758 | * abort. | |
6759 | */ | |
6760 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6761 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6762 | u64 trimmed = 0; | |
6763 | ||
6764 | ret = -EROFS; | |
6765 | if (!trans->aborted) | |
6766 | ret = btrfs_discard_extent(root, | |
6767 | block_group->key.objectid, | |
6768 | block_group->key.offset, | |
6769 | &trimmed); | |
6770 | ||
6771 | list_del_init(&block_group->bg_list); | |
6772 | btrfs_put_block_group_trimming(block_group); | |
6773 | btrfs_put_block_group(block_group); | |
6774 | ||
6775 | if (ret) { | |
6776 | const char *errstr = btrfs_decode_error(ret); | |
6777 | btrfs_warn(fs_info, | |
6778 | "Discard failed while removing blockgroup: errno=%d %s\n", | |
6779 | ret, errstr); | |
6780 | } | |
6781 | } | |
6782 | ||
e20d96d6 CM |
6783 | return 0; |
6784 | } | |
6785 | ||
b150a4f1 JB |
6786 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, |
6787 | u64 owner, u64 root_objectid) | |
6788 | { | |
6789 | struct btrfs_space_info *space_info; | |
6790 | u64 flags; | |
6791 | ||
6792 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
6793 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
6794 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
6795 | else | |
6796 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
6797 | } else { | |
6798 | flags = BTRFS_BLOCK_GROUP_DATA; | |
6799 | } | |
6800 | ||
6801 | space_info = __find_space_info(fs_info, flags); | |
6802 | BUG_ON(!space_info); /* Logic bug */ | |
6803 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); | |
6804 | } | |
6805 | ||
6806 | ||
5d4f98a2 YZ |
6807 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
6808 | struct btrfs_root *root, | |
c682f9b3 | 6809 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6810 | u64 root_objectid, u64 owner_objectid, |
6811 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6812 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6813 | { |
e2fa7227 | 6814 | struct btrfs_key key; |
5d4f98a2 | 6815 | struct btrfs_path *path; |
1261ec42 CM |
6816 | struct btrfs_fs_info *info = root->fs_info; |
6817 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 6818 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6819 | struct btrfs_extent_item *ei; |
6820 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6821 | int ret; |
5d4f98a2 | 6822 | int is_data; |
952fccac CM |
6823 | int extent_slot = 0; |
6824 | int found_extent = 0; | |
6825 | int num_to_del = 1; | |
5d4f98a2 YZ |
6826 | u32 item_size; |
6827 | u64 refs; | |
c682f9b3 QW |
6828 | u64 bytenr = node->bytenr; |
6829 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6830 | int last_ref = 0; |
3173a18f JB |
6831 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
6832 | SKINNY_METADATA); | |
037e6390 | 6833 | |
5caf2a00 | 6834 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6835 | if (!path) |
6836 | return -ENOMEM; | |
5f26f772 | 6837 | |
e4058b54 | 6838 | path->reada = READA_FORWARD; |
b9473439 | 6839 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6840 | |
6841 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6842 | BUG_ON(!is_data && refs_to_drop != 1); | |
6843 | ||
3173a18f JB |
6844 | if (is_data) |
6845 | skinny_metadata = 0; | |
6846 | ||
5d4f98a2 YZ |
6847 | ret = lookup_extent_backref(trans, extent_root, path, &iref, |
6848 | bytenr, num_bytes, parent, | |
6849 | root_objectid, owner_objectid, | |
6850 | owner_offset); | |
7bb86316 | 6851 | if (ret == 0) { |
952fccac | 6852 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6853 | while (extent_slot >= 0) { |
6854 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6855 | extent_slot); |
5d4f98a2 | 6856 | if (key.objectid != bytenr) |
952fccac | 6857 | break; |
5d4f98a2 YZ |
6858 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6859 | key.offset == num_bytes) { | |
952fccac CM |
6860 | found_extent = 1; |
6861 | break; | |
6862 | } | |
3173a18f JB |
6863 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6864 | key.offset == owner_objectid) { | |
6865 | found_extent = 1; | |
6866 | break; | |
6867 | } | |
952fccac CM |
6868 | if (path->slots[0] - extent_slot > 5) |
6869 | break; | |
5d4f98a2 | 6870 | extent_slot--; |
952fccac | 6871 | } |
5d4f98a2 YZ |
6872 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6873 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6874 | if (found_extent && item_size < sizeof(*ei)) | |
6875 | found_extent = 0; | |
6876 | #endif | |
31840ae1 | 6877 | if (!found_extent) { |
5d4f98a2 | 6878 | BUG_ON(iref); |
56bec294 | 6879 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 | 6880 | NULL, refs_to_drop, |
fcebe456 | 6881 | is_data, &last_ref); |
005d6427 DS |
6882 | if (ret) { |
6883 | btrfs_abort_transaction(trans, extent_root, ret); | |
6884 | goto out; | |
6885 | } | |
b3b4aa74 | 6886 | btrfs_release_path(path); |
b9473439 | 6887 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6888 | |
6889 | key.objectid = bytenr; | |
6890 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6891 | key.offset = num_bytes; | |
6892 | ||
3173a18f JB |
6893 | if (!is_data && skinny_metadata) { |
6894 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6895 | key.offset = owner_objectid; | |
6896 | } | |
6897 | ||
31840ae1 ZY |
6898 | ret = btrfs_search_slot(trans, extent_root, |
6899 | &key, path, -1, 1); | |
3173a18f JB |
6900 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6901 | /* | |
6902 | * Couldn't find our skinny metadata item, | |
6903 | * see if we have ye olde extent item. | |
6904 | */ | |
6905 | path->slots[0]--; | |
6906 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6907 | path->slots[0]); | |
6908 | if (key.objectid == bytenr && | |
6909 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6910 | key.offset == num_bytes) | |
6911 | ret = 0; | |
6912 | } | |
6913 | ||
6914 | if (ret > 0 && skinny_metadata) { | |
6915 | skinny_metadata = false; | |
9ce49a0b | 6916 | key.objectid = bytenr; |
3173a18f JB |
6917 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6918 | key.offset = num_bytes; | |
6919 | btrfs_release_path(path); | |
6920 | ret = btrfs_search_slot(trans, extent_root, | |
6921 | &key, path, -1, 1); | |
6922 | } | |
6923 | ||
f3465ca4 | 6924 | if (ret) { |
c2cf52eb | 6925 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 6926 | ret, bytenr); |
b783e62d JB |
6927 | if (ret > 0) |
6928 | btrfs_print_leaf(extent_root, | |
6929 | path->nodes[0]); | |
f3465ca4 | 6930 | } |
005d6427 DS |
6931 | if (ret < 0) { |
6932 | btrfs_abort_transaction(trans, extent_root, ret); | |
6933 | goto out; | |
6934 | } | |
31840ae1 ZY |
6935 | extent_slot = path->slots[0]; |
6936 | } | |
fae7f21c | 6937 | } else if (WARN_ON(ret == -ENOENT)) { |
7bb86316 | 6938 | btrfs_print_leaf(extent_root, path->nodes[0]); |
c2cf52eb SK |
6939 | btrfs_err(info, |
6940 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6941 | bytenr, parent, root_objectid, owner_objectid, |
6942 | owner_offset); | |
c4a050bb JB |
6943 | btrfs_abort_transaction(trans, extent_root, ret); |
6944 | goto out; | |
79787eaa | 6945 | } else { |
005d6427 DS |
6946 | btrfs_abort_transaction(trans, extent_root, ret); |
6947 | goto out; | |
7bb86316 | 6948 | } |
5f39d397 CM |
6949 | |
6950 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6951 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6952 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6953 | if (item_size < sizeof(*ei)) { | |
6954 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
6955 | ret = convert_extent_item_v0(trans, extent_root, path, | |
6956 | owner_objectid, 0); | |
005d6427 DS |
6957 | if (ret < 0) { |
6958 | btrfs_abort_transaction(trans, extent_root, ret); | |
6959 | goto out; | |
6960 | } | |
5d4f98a2 | 6961 | |
b3b4aa74 | 6962 | btrfs_release_path(path); |
5d4f98a2 YZ |
6963 | path->leave_spinning = 1; |
6964 | ||
6965 | key.objectid = bytenr; | |
6966 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6967 | key.offset = num_bytes; | |
6968 | ||
6969 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
6970 | -1, 1); | |
6971 | if (ret) { | |
c2cf52eb | 6972 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 6973 | ret, bytenr); |
5d4f98a2 YZ |
6974 | btrfs_print_leaf(extent_root, path->nodes[0]); |
6975 | } | |
005d6427 DS |
6976 | if (ret < 0) { |
6977 | btrfs_abort_transaction(trans, extent_root, ret); | |
6978 | goto out; | |
6979 | } | |
6980 | ||
5d4f98a2 YZ |
6981 | extent_slot = path->slots[0]; |
6982 | leaf = path->nodes[0]; | |
6983 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
6984 | } | |
6985 | #endif | |
6986 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 6987 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 6988 | struct btrfs_extent_item); |
3173a18f JB |
6989 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
6990 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
6991 | struct btrfs_tree_block_info *bi; |
6992 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
6993 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
6994 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
6995 | } | |
56bec294 | 6996 | |
5d4f98a2 | 6997 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 JB |
6998 | if (refs < refs_to_drop) { |
6999 | btrfs_err(info, "trying to drop %d refs but we only have %Lu " | |
351fd353 | 7000 | "for bytenr %Lu", refs_to_drop, refs, bytenr); |
32b02538 JB |
7001 | ret = -EINVAL; |
7002 | btrfs_abort_transaction(trans, extent_root, ret); | |
7003 | goto out; | |
7004 | } | |
56bec294 | 7005 | refs -= refs_to_drop; |
5f39d397 | 7006 | |
5d4f98a2 YZ |
7007 | if (refs > 0) { |
7008 | if (extent_op) | |
7009 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7010 | /* | |
7011 | * In the case of inline back ref, reference count will | |
7012 | * be updated by remove_extent_backref | |
952fccac | 7013 | */ |
5d4f98a2 YZ |
7014 | if (iref) { |
7015 | BUG_ON(!found_extent); | |
7016 | } else { | |
7017 | btrfs_set_extent_refs(leaf, ei, refs); | |
7018 | btrfs_mark_buffer_dirty(leaf); | |
7019 | } | |
7020 | if (found_extent) { | |
7021 | ret = remove_extent_backref(trans, extent_root, path, | |
7022 | iref, refs_to_drop, | |
fcebe456 | 7023 | is_data, &last_ref); |
005d6427 DS |
7024 | if (ret) { |
7025 | btrfs_abort_transaction(trans, extent_root, ret); | |
7026 | goto out; | |
7027 | } | |
952fccac | 7028 | } |
b150a4f1 JB |
7029 | add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid, |
7030 | root_objectid); | |
5d4f98a2 | 7031 | } else { |
5d4f98a2 YZ |
7032 | if (found_extent) { |
7033 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7034 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7035 | if (iref) { |
7036 | BUG_ON(path->slots[0] != extent_slot); | |
7037 | } else { | |
7038 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7039 | path->slots[0] = extent_slot; | |
7040 | num_to_del = 2; | |
7041 | } | |
78fae27e | 7042 | } |
b9473439 | 7043 | |
fcebe456 | 7044 | last_ref = 1; |
952fccac CM |
7045 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7046 | num_to_del); | |
005d6427 DS |
7047 | if (ret) { |
7048 | btrfs_abort_transaction(trans, extent_root, ret); | |
7049 | goto out; | |
7050 | } | |
b3b4aa74 | 7051 | btrfs_release_path(path); |
21af804c | 7052 | |
5d4f98a2 | 7053 | if (is_data) { |
459931ec | 7054 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 DS |
7055 | if (ret) { |
7056 | btrfs_abort_transaction(trans, extent_root, ret); | |
7057 | goto out; | |
7058 | } | |
459931ec CM |
7059 | } |
7060 | ||
1e144fb8 OS |
7061 | ret = add_to_free_space_tree(trans, root->fs_info, bytenr, |
7062 | num_bytes); | |
7063 | if (ret) { | |
7064 | btrfs_abort_transaction(trans, extent_root, ret); | |
7065 | goto out; | |
7066 | } | |
7067 | ||
ce93ec54 | 7068 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
005d6427 DS |
7069 | if (ret) { |
7070 | btrfs_abort_transaction(trans, extent_root, ret); | |
7071 | goto out; | |
7072 | } | |
a28ec197 | 7073 | } |
fcebe456 JB |
7074 | btrfs_release_path(path); |
7075 | ||
79787eaa | 7076 | out: |
5caf2a00 | 7077 | btrfs_free_path(path); |
a28ec197 CM |
7078 | return ret; |
7079 | } | |
7080 | ||
1887be66 | 7081 | /* |
f0486c68 | 7082 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7083 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7084 | * a given extent, and if there are no other delayed refs to be processed, it | |
7085 | * removes it from the tree. | |
7086 | */ | |
7087 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
7088 | struct btrfs_root *root, u64 bytenr) | |
7089 | { | |
7090 | struct btrfs_delayed_ref_head *head; | |
7091 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7092 | int ret = 0; |
1887be66 CM |
7093 | |
7094 | delayed_refs = &trans->transaction->delayed_refs; | |
7095 | spin_lock(&delayed_refs->lock); | |
7096 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
7097 | if (!head) | |
cf93da7b | 7098 | goto out_delayed_unlock; |
1887be66 | 7099 | |
d7df2c79 | 7100 | spin_lock(&head->lock); |
c6fc2454 | 7101 | if (!list_empty(&head->ref_list)) |
1887be66 CM |
7102 | goto out; |
7103 | ||
5d4f98a2 YZ |
7104 | if (head->extent_op) { |
7105 | if (!head->must_insert_reserved) | |
7106 | goto out; | |
78a6184a | 7107 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7108 | head->extent_op = NULL; |
7109 | } | |
7110 | ||
1887be66 CM |
7111 | /* |
7112 | * waiting for the lock here would deadlock. If someone else has it | |
7113 | * locked they are already in the process of dropping it anyway | |
7114 | */ | |
7115 | if (!mutex_trylock(&head->mutex)) | |
7116 | goto out; | |
7117 | ||
7118 | /* | |
7119 | * at this point we have a head with no other entries. Go | |
7120 | * ahead and process it. | |
7121 | */ | |
7122 | head->node.in_tree = 0; | |
c46effa6 | 7123 | rb_erase(&head->href_node, &delayed_refs->href_root); |
c3e69d58 | 7124 | |
d7df2c79 | 7125 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7126 | |
7127 | /* | |
7128 | * we don't take a ref on the node because we're removing it from the | |
7129 | * tree, so we just steal the ref the tree was holding. | |
7130 | */ | |
c3e69d58 | 7131 | delayed_refs->num_heads--; |
d7df2c79 | 7132 | if (head->processing == 0) |
c3e69d58 | 7133 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7134 | head->processing = 0; |
7135 | spin_unlock(&head->lock); | |
1887be66 CM |
7136 | spin_unlock(&delayed_refs->lock); |
7137 | ||
f0486c68 YZ |
7138 | BUG_ON(head->extent_op); |
7139 | if (head->must_insert_reserved) | |
7140 | ret = 1; | |
7141 | ||
7142 | mutex_unlock(&head->mutex); | |
1887be66 | 7143 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 7144 | return ret; |
1887be66 | 7145 | out: |
d7df2c79 | 7146 | spin_unlock(&head->lock); |
cf93da7b CM |
7147 | |
7148 | out_delayed_unlock: | |
1887be66 CM |
7149 | spin_unlock(&delayed_refs->lock); |
7150 | return 0; | |
7151 | } | |
7152 | ||
f0486c68 YZ |
7153 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7154 | struct btrfs_root *root, | |
7155 | struct extent_buffer *buf, | |
5581a51a | 7156 | u64 parent, int last_ref) |
f0486c68 | 7157 | { |
b150a4f1 | 7158 | int pin = 1; |
f0486c68 YZ |
7159 | int ret; |
7160 | ||
7161 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
7162 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
7163 | buf->start, buf->len, | |
7164 | parent, root->root_key.objectid, | |
7165 | btrfs_header_level(buf), | |
b06c4bf5 | 7166 | BTRFS_DROP_DELAYED_REF, NULL); |
79787eaa | 7167 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
7168 | } |
7169 | ||
7170 | if (!last_ref) | |
7171 | return; | |
7172 | ||
f0486c68 | 7173 | if (btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7174 | struct btrfs_block_group_cache *cache; |
7175 | ||
f0486c68 YZ |
7176 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
7177 | ret = check_ref_cleanup(trans, root, buf->start); | |
7178 | if (!ret) | |
37be25bc | 7179 | goto out; |
f0486c68 YZ |
7180 | } |
7181 | ||
6219872d FM |
7182 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
7183 | ||
f0486c68 YZ |
7184 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
7185 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
6219872d | 7186 | btrfs_put_block_group(cache); |
37be25bc | 7187 | goto out; |
f0486c68 YZ |
7188 | } |
7189 | ||
7190 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7191 | ||
7192 | btrfs_add_free_space(cache, buf->start, buf->len); | |
e570fd27 | 7193 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0); |
6219872d | 7194 | btrfs_put_block_group(cache); |
0be5dc67 | 7195 | trace_btrfs_reserved_extent_free(root, buf->start, buf->len); |
b150a4f1 | 7196 | pin = 0; |
f0486c68 YZ |
7197 | } |
7198 | out: | |
b150a4f1 JB |
7199 | if (pin) |
7200 | add_pinned_bytes(root->fs_info, buf->len, | |
7201 | btrfs_header_level(buf), | |
7202 | root->root_key.objectid); | |
7203 | ||
a826d6dc JB |
7204 | /* |
7205 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
7206 | * anymore. | |
7207 | */ | |
7208 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
7209 | } |
7210 | ||
79787eaa | 7211 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
7212 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
7213 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
b06c4bf5 | 7214 | u64 owner, u64 offset) |
925baedd CM |
7215 | { |
7216 | int ret; | |
66d7e7f0 | 7217 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 7218 | |
fccb84c9 | 7219 | if (btrfs_test_is_dummy_root(root)) |
faa2dbf0 | 7220 | return 0; |
fccb84c9 | 7221 | |
b150a4f1 JB |
7222 | add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid); |
7223 | ||
56bec294 CM |
7224 | /* |
7225 | * tree log blocks never actually go into the extent allocation | |
7226 | * tree, just update pinning info and exit early. | |
56bec294 | 7227 | */ |
5d4f98a2 YZ |
7228 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7229 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7230 | /* unlocks the pinned mutex */ |
11833d66 | 7231 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 7232 | ret = 0; |
5d4f98a2 | 7233 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
7234 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7235 | num_bytes, | |
5d4f98a2 | 7236 | parent, root_objectid, (int)owner, |
b06c4bf5 | 7237 | BTRFS_DROP_DELAYED_REF, NULL); |
5d4f98a2 | 7238 | } else { |
66d7e7f0 AJ |
7239 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7240 | num_bytes, | |
7241 | parent, root_objectid, owner, | |
5846a3c2 QW |
7242 | offset, 0, |
7243 | BTRFS_DROP_DELAYED_REF, NULL); | |
56bec294 | 7244 | } |
925baedd CM |
7245 | return ret; |
7246 | } | |
7247 | ||
817d52f8 JB |
7248 | /* |
7249 | * when we wait for progress in the block group caching, its because | |
7250 | * our allocation attempt failed at least once. So, we must sleep | |
7251 | * and let some progress happen before we try again. | |
7252 | * | |
7253 | * This function will sleep at least once waiting for new free space to | |
7254 | * show up, and then it will check the block group free space numbers | |
7255 | * for our min num_bytes. Another option is to have it go ahead | |
7256 | * and look in the rbtree for a free extent of a given size, but this | |
7257 | * is a good start. | |
36cce922 JB |
7258 | * |
7259 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7260 | * any of the information in this block group. | |
817d52f8 | 7261 | */ |
36cce922 | 7262 | static noinline void |
817d52f8 JB |
7263 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7264 | u64 num_bytes) | |
7265 | { | |
11833d66 | 7266 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7267 | |
11833d66 YZ |
7268 | caching_ctl = get_caching_control(cache); |
7269 | if (!caching_ctl) | |
36cce922 | 7270 | return; |
817d52f8 | 7271 | |
11833d66 | 7272 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7273 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7274 | |
7275 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7276 | } |
7277 | ||
7278 | static noinline int | |
7279 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7280 | { | |
7281 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7282 | int ret = 0; |
11833d66 YZ |
7283 | |
7284 | caching_ctl = get_caching_control(cache); | |
7285 | if (!caching_ctl) | |
36cce922 | 7286 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7287 | |
7288 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7289 | if (cache->cached == BTRFS_CACHE_ERROR) |
7290 | ret = -EIO; | |
11833d66 | 7291 | put_caching_control(caching_ctl); |
36cce922 | 7292 | return ret; |
817d52f8 JB |
7293 | } |
7294 | ||
31e50229 | 7295 | int __get_raid_index(u64 flags) |
b742bb82 | 7296 | { |
7738a53a | 7297 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 7298 | return BTRFS_RAID_RAID10; |
7738a53a | 7299 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 7300 | return BTRFS_RAID_RAID1; |
7738a53a | 7301 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 7302 | return BTRFS_RAID_DUP; |
7738a53a | 7303 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 7304 | return BTRFS_RAID_RAID0; |
53b381b3 | 7305 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 7306 | return BTRFS_RAID_RAID5; |
53b381b3 | 7307 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 7308 | return BTRFS_RAID_RAID6; |
7738a53a | 7309 | |
e942f883 | 7310 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
7311 | } |
7312 | ||
6ab0a202 | 7313 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
7738a53a | 7314 | { |
31e50229 | 7315 | return __get_raid_index(cache->flags); |
7738a53a ID |
7316 | } |
7317 | ||
6ab0a202 JM |
7318 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
7319 | [BTRFS_RAID_RAID10] = "raid10", | |
7320 | [BTRFS_RAID_RAID1] = "raid1", | |
7321 | [BTRFS_RAID_DUP] = "dup", | |
7322 | [BTRFS_RAID_RAID0] = "raid0", | |
7323 | [BTRFS_RAID_SINGLE] = "single", | |
7324 | [BTRFS_RAID_RAID5] = "raid5", | |
7325 | [BTRFS_RAID_RAID6] = "raid6", | |
7326 | }; | |
7327 | ||
1b8e5df6 | 7328 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
7329 | { |
7330 | if (type >= BTRFS_NR_RAID_TYPES) | |
7331 | return NULL; | |
7332 | ||
7333 | return btrfs_raid_type_names[type]; | |
7334 | } | |
7335 | ||
817d52f8 | 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 | |
6719afdc | 7386 | down_read(&used_bg->data_rwsem); |
e570fd27 | 7387 | |
6719afdc GU |
7388 | spin_lock(&cluster->refill_lock); |
7389 | if (used_bg == cluster->block_group) | |
7390 | return used_bg; | |
e570fd27 | 7391 | |
6719afdc GU |
7392 | up_read(&used_bg->data_rwsem); |
7393 | btrfs_put_block_group(used_bg); | |
7394 | } | |
e570fd27 MX |
7395 | } |
7396 | ||
7397 | static inline void | |
7398 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7399 | int delalloc) | |
7400 | { | |
7401 | if (delalloc) | |
7402 | up_read(&cache->data_rwsem); | |
7403 | btrfs_put_block_group(cache); | |
7404 | } | |
7405 | ||
fec577fb CM |
7406 | /* |
7407 | * walks the btree of allocated extents and find a hole of a given size. | |
7408 | * The key ins is changed to record the hole: | |
a4820398 | 7409 | * ins->objectid == start position |
62e2749e | 7410 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7411 | * ins->offset == the size of the hole. |
fec577fb | 7412 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7413 | * |
7414 | * If there is no suitable free space, we will record the max size of | |
7415 | * the free space extent currently. | |
fec577fb | 7416 | */ |
00361589 | 7417 | static noinline int find_free_extent(struct btrfs_root *orig_root, |
98ed5174 | 7418 | u64 num_bytes, u64 empty_size, |
98ed5174 | 7419 | u64 hint_byte, struct btrfs_key *ins, |
e570fd27 | 7420 | u64 flags, int delalloc) |
fec577fb | 7421 | { |
80eb234a | 7422 | int ret = 0; |
d397712b | 7423 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 7424 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7425 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7426 | u64 search_start = 0; |
a4820398 | 7427 | u64 max_extent_size = 0; |
c759c4e1 | 7428 | u64 empty_cluster = 0; |
80eb234a | 7429 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7430 | int loop = 0; |
b6919a58 DS |
7431 | int index = __get_raid_index(flags); |
7432 | int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ? | |
fb25e914 | 7433 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; |
0a24325e | 7434 | bool failed_cluster_refill = false; |
1cdda9b8 | 7435 | bool failed_alloc = false; |
67377734 | 7436 | bool use_cluster = true; |
60d2adbb | 7437 | bool have_caching_bg = false; |
13a0db5a | 7438 | bool orig_have_caching_bg = false; |
a5e681d9 | 7439 | bool full_search = false; |
fec577fb | 7440 | |
db94535d | 7441 | WARN_ON(num_bytes < root->sectorsize); |
962a298f | 7442 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7443 | ins->objectid = 0; |
7444 | ins->offset = 0; | |
b1a4d965 | 7445 | |
b6919a58 | 7446 | trace_find_free_extent(orig_root, num_bytes, empty_size, flags); |
3f7de037 | 7447 | |
b6919a58 | 7448 | space_info = __find_space_info(root->fs_info, flags); |
1b1d1f66 | 7449 | if (!space_info) { |
b6919a58 | 7450 | btrfs_err(root->fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7451 | return -ENOSPC; |
7452 | } | |
2552d17e | 7453 | |
67377734 | 7454 | /* |
4f4db217 JB |
7455 | * If our free space is heavily fragmented we may not be able to make |
7456 | * big contiguous allocations, so instead of doing the expensive search | |
7457 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7458 | * can go ahead and search for a more manageable chunk. | |
7459 | * | |
7460 | * If our max_extent_size is large enough for our allocation simply | |
7461 | * disable clustering since we will likely not be able to find enough | |
7462 | * space to create a cluster and induce latency trying. | |
67377734 | 7463 | */ |
4f4db217 JB |
7464 | if (unlikely(space_info->max_extent_size)) { |
7465 | spin_lock(&space_info->lock); | |
7466 | if (space_info->max_extent_size && | |
7467 | num_bytes > space_info->max_extent_size) { | |
7468 | ins->offset = space_info->max_extent_size; | |
7469 | spin_unlock(&space_info->lock); | |
7470 | return -ENOSPC; | |
7471 | } else if (space_info->max_extent_size) { | |
7472 | use_cluster = false; | |
7473 | } | |
7474 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7475 | } |
0f9dd46c | 7476 | |
c759c4e1 | 7477 | last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster); |
239b14b3 | 7478 | if (last_ptr) { |
fa9c0d79 CM |
7479 | spin_lock(&last_ptr->lock); |
7480 | if (last_ptr->block_group) | |
7481 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7482 | if (last_ptr->fragmented) { |
7483 | /* | |
7484 | * We still set window_start so we can keep track of the | |
7485 | * last place we found an allocation to try and save | |
7486 | * some time. | |
7487 | */ | |
7488 | hint_byte = last_ptr->window_start; | |
7489 | use_cluster = false; | |
7490 | } | |
fa9c0d79 | 7491 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7492 | } |
fa9c0d79 | 7493 | |
a061fc8d | 7494 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 7495 | search_start = max(search_start, hint_byte); |
2552d17e | 7496 | if (search_start == hint_byte) { |
2552d17e JB |
7497 | block_group = btrfs_lookup_block_group(root->fs_info, |
7498 | search_start); | |
817d52f8 JB |
7499 | /* |
7500 | * we don't want to use the block group if it doesn't match our | |
7501 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7502 | * |
7503 | * However if we are re-searching with an ideal block group | |
7504 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7505 | */ |
b6919a58 | 7506 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7507 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7508 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7509 | if (list_empty(&block_group->list) || |
7510 | block_group->ro) { | |
7511 | /* | |
7512 | * someone is removing this block group, | |
7513 | * we can't jump into the have_block_group | |
7514 | * target because our list pointers are not | |
7515 | * valid | |
7516 | */ | |
7517 | btrfs_put_block_group(block_group); | |
7518 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7519 | } else { |
b742bb82 | 7520 | index = get_block_group_index(block_group); |
e570fd27 | 7521 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7522 | goto have_block_group; |
ccf0e725 | 7523 | } |
2552d17e | 7524 | } else if (block_group) { |
fa9c0d79 | 7525 | btrfs_put_block_group(block_group); |
2552d17e | 7526 | } |
42e70e7a | 7527 | } |
2552d17e | 7528 | search: |
60d2adbb | 7529 | have_caching_bg = false; |
a5e681d9 JB |
7530 | if (index == 0 || index == __get_raid_index(flags)) |
7531 | full_search = true; | |
80eb234a | 7532 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7533 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7534 | list) { | |
6226cb0a | 7535 | u64 offset; |
817d52f8 | 7536 | int cached; |
8a1413a2 | 7537 | |
e570fd27 | 7538 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7539 | search_start = block_group->key.objectid; |
42e70e7a | 7540 | |
83a50de9 CM |
7541 | /* |
7542 | * this can happen if we end up cycling through all the | |
7543 | * raid types, but we want to make sure we only allocate | |
7544 | * for the proper type. | |
7545 | */ | |
b6919a58 | 7546 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7547 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7548 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7549 | BTRFS_BLOCK_GROUP_RAID5 | |
7550 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7551 | BTRFS_BLOCK_GROUP_RAID10; |
7552 | ||
7553 | /* | |
7554 | * if they asked for extra copies and this block group | |
7555 | * doesn't provide them, bail. This does allow us to | |
7556 | * fill raid0 from raid1. | |
7557 | */ | |
b6919a58 | 7558 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7559 | goto loop; |
7560 | } | |
7561 | ||
2552d17e | 7562 | have_block_group: |
291c7d2f JB |
7563 | cached = block_group_cache_done(block_group); |
7564 | if (unlikely(!cached)) { | |
a5e681d9 | 7565 | have_caching_bg = true; |
f6373bf3 | 7566 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7567 | BUG_ON(ret < 0); |
7568 | ret = 0; | |
817d52f8 JB |
7569 | } |
7570 | ||
36cce922 JB |
7571 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7572 | goto loop; | |
ea6a478e | 7573 | if (unlikely(block_group->ro)) |
2552d17e | 7574 | goto loop; |
0f9dd46c | 7575 | |
0a24325e | 7576 | /* |
062c05c4 AO |
7577 | * Ok we want to try and use the cluster allocator, so |
7578 | * lets look there | |
0a24325e | 7579 | */ |
c759c4e1 | 7580 | if (last_ptr && use_cluster) { |
215a63d1 | 7581 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7582 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7583 | /* |
7584 | * the refill lock keeps out other | |
7585 | * people trying to start a new cluster | |
7586 | */ | |
e570fd27 MX |
7587 | used_block_group = btrfs_lock_cluster(block_group, |
7588 | last_ptr, | |
7589 | delalloc); | |
7590 | if (!used_block_group) | |
44fb5511 | 7591 | goto refill_cluster; |
274bd4fb | 7592 | |
e570fd27 MX |
7593 | if (used_block_group != block_group && |
7594 | (used_block_group->ro || | |
7595 | !block_group_bits(used_block_group, flags))) | |
7596 | goto release_cluster; | |
44fb5511 | 7597 | |
274bd4fb | 7598 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7599 | last_ptr, |
7600 | num_bytes, | |
7601 | used_block_group->key.objectid, | |
7602 | &max_extent_size); | |
fa9c0d79 CM |
7603 | if (offset) { |
7604 | /* we have a block, we're done */ | |
7605 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 | 7606 | trace_btrfs_reserve_extent_cluster(root, |
89d4346a MX |
7607 | used_block_group, |
7608 | search_start, num_bytes); | |
215a63d1 | 7609 | if (used_block_group != block_group) { |
e570fd27 MX |
7610 | btrfs_release_block_group(block_group, |
7611 | delalloc); | |
215a63d1 MX |
7612 | block_group = used_block_group; |
7613 | } | |
fa9c0d79 CM |
7614 | goto checks; |
7615 | } | |
7616 | ||
274bd4fb | 7617 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7618 | release_cluster: |
062c05c4 AO |
7619 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7620 | * set up a new clusters, so lets just skip it | |
7621 | * and let the allocator find whatever block | |
7622 | * it can find. If we reach this point, we | |
7623 | * will have tried the cluster allocator | |
7624 | * plenty of times and not have found | |
7625 | * anything, so we are likely way too | |
7626 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7627 | * anything. |
7628 | * | |
7629 | * However, if the cluster is taken from the | |
7630 | * current block group, release the cluster | |
7631 | * first, so that we stand a better chance of | |
7632 | * succeeding in the unclustered | |
7633 | * allocation. */ | |
7634 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7635 | used_block_group != block_group) { |
062c05c4 | 7636 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7637 | btrfs_release_block_group(used_block_group, |
7638 | delalloc); | |
062c05c4 AO |
7639 | goto unclustered_alloc; |
7640 | } | |
7641 | ||
fa9c0d79 CM |
7642 | /* |
7643 | * this cluster didn't work out, free it and | |
7644 | * start over | |
7645 | */ | |
7646 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7647 | ||
e570fd27 MX |
7648 | if (used_block_group != block_group) |
7649 | btrfs_release_block_group(used_block_group, | |
7650 | delalloc); | |
7651 | refill_cluster: | |
a5f6f719 AO |
7652 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7653 | spin_unlock(&last_ptr->refill_lock); | |
7654 | goto unclustered_alloc; | |
7655 | } | |
7656 | ||
8de972b4 CM |
7657 | aligned_cluster = max_t(unsigned long, |
7658 | empty_cluster + empty_size, | |
7659 | block_group->full_stripe_len); | |
7660 | ||
fa9c0d79 | 7661 | /* allocate a cluster in this block group */ |
00361589 JB |
7662 | ret = btrfs_find_space_cluster(root, block_group, |
7663 | last_ptr, search_start, | |
7664 | num_bytes, | |
7665 | aligned_cluster); | |
fa9c0d79 CM |
7666 | if (ret == 0) { |
7667 | /* | |
7668 | * now pull our allocation out of this | |
7669 | * cluster | |
7670 | */ | |
7671 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7672 | last_ptr, |
7673 | num_bytes, | |
7674 | search_start, | |
7675 | &max_extent_size); | |
fa9c0d79 CM |
7676 | if (offset) { |
7677 | /* we found one, proceed */ | |
7678 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
7679 | trace_btrfs_reserve_extent_cluster(root, |
7680 | block_group, search_start, | |
7681 | num_bytes); | |
fa9c0d79 CM |
7682 | goto checks; |
7683 | } | |
0a24325e JB |
7684 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7685 | && !failed_cluster_refill) { | |
817d52f8 JB |
7686 | spin_unlock(&last_ptr->refill_lock); |
7687 | ||
0a24325e | 7688 | failed_cluster_refill = true; |
817d52f8 JB |
7689 | wait_block_group_cache_progress(block_group, |
7690 | num_bytes + empty_cluster + empty_size); | |
7691 | goto have_block_group; | |
fa9c0d79 | 7692 | } |
817d52f8 | 7693 | |
fa9c0d79 CM |
7694 | /* |
7695 | * at this point we either didn't find a cluster | |
7696 | * or we weren't able to allocate a block from our | |
7697 | * cluster. Free the cluster we've been trying | |
7698 | * to use, and go to the next block group | |
7699 | */ | |
0a24325e | 7700 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7701 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7702 | goto loop; |
fa9c0d79 CM |
7703 | } |
7704 | ||
062c05c4 | 7705 | unclustered_alloc: |
c759c4e1 JB |
7706 | /* |
7707 | * We are doing an unclustered alloc, set the fragmented flag so | |
7708 | * we don't bother trying to setup a cluster again until we get | |
7709 | * more space. | |
7710 | */ | |
7711 | if (unlikely(last_ptr)) { | |
7712 | spin_lock(&last_ptr->lock); | |
7713 | last_ptr->fragmented = 1; | |
7714 | spin_unlock(&last_ptr->lock); | |
7715 | } | |
a5f6f719 AO |
7716 | spin_lock(&block_group->free_space_ctl->tree_lock); |
7717 | if (cached && | |
7718 | block_group->free_space_ctl->free_space < | |
7719 | num_bytes + empty_cluster + empty_size) { | |
a4820398 MX |
7720 | if (block_group->free_space_ctl->free_space > |
7721 | max_extent_size) | |
7722 | max_extent_size = | |
7723 | block_group->free_space_ctl->free_space; | |
a5f6f719 AO |
7724 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
7725 | goto loop; | |
7726 | } | |
7727 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
7728 | ||
6226cb0a | 7729 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7730 | num_bytes, empty_size, |
7731 | &max_extent_size); | |
1cdda9b8 JB |
7732 | /* |
7733 | * If we didn't find a chunk, and we haven't failed on this | |
7734 | * block group before, and this block group is in the middle of | |
7735 | * caching and we are ok with waiting, then go ahead and wait | |
7736 | * for progress to be made, and set failed_alloc to true. | |
7737 | * | |
7738 | * If failed_alloc is true then we've already waited on this | |
7739 | * block group once and should move on to the next block group. | |
7740 | */ | |
7741 | if (!offset && !failed_alloc && !cached && | |
7742 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7743 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7744 | num_bytes + empty_size); |
7745 | failed_alloc = true; | |
817d52f8 | 7746 | goto have_block_group; |
1cdda9b8 JB |
7747 | } else if (!offset) { |
7748 | goto loop; | |
817d52f8 | 7749 | } |
fa9c0d79 | 7750 | checks: |
4e54b17a | 7751 | search_start = ALIGN(offset, root->stripesize); |
25179201 | 7752 | |
2552d17e JB |
7753 | /* move on to the next group */ |
7754 | if (search_start + num_bytes > | |
215a63d1 MX |
7755 | block_group->key.objectid + block_group->key.offset) { |
7756 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7757 | goto loop; |
6226cb0a | 7758 | } |
f5a31e16 | 7759 | |
f0486c68 | 7760 | if (offset < search_start) |
215a63d1 | 7761 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7762 | search_start - offset); |
7763 | BUG_ON(offset > search_start); | |
2552d17e | 7764 | |
215a63d1 | 7765 | ret = btrfs_update_reserved_bytes(block_group, num_bytes, |
e570fd27 | 7766 | alloc_type, delalloc); |
f0486c68 | 7767 | if (ret == -EAGAIN) { |
215a63d1 | 7768 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7769 | goto loop; |
0f9dd46c | 7770 | } |
9cfa3e34 | 7771 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7772 | |
f0486c68 | 7773 | /* we are all good, lets return */ |
2552d17e JB |
7774 | ins->objectid = search_start; |
7775 | ins->offset = num_bytes; | |
d2fb3437 | 7776 | |
3f7de037 JB |
7777 | trace_btrfs_reserve_extent(orig_root, block_group, |
7778 | search_start, num_bytes); | |
e570fd27 | 7779 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7780 | break; |
7781 | loop: | |
0a24325e | 7782 | failed_cluster_refill = false; |
1cdda9b8 | 7783 | failed_alloc = false; |
b742bb82 | 7784 | BUG_ON(index != get_block_group_index(block_group)); |
e570fd27 | 7785 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7786 | } |
7787 | up_read(&space_info->groups_sem); | |
7788 | ||
13a0db5a | 7789 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7790 | && !orig_have_caching_bg) | |
7791 | orig_have_caching_bg = true; | |
7792 | ||
60d2adbb MX |
7793 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7794 | goto search; | |
7795 | ||
b742bb82 YZ |
7796 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7797 | goto search; | |
7798 | ||
285ff5af | 7799 | /* |
ccf0e725 JB |
7800 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7801 | * caching kthreads as we move along | |
817d52f8 JB |
7802 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7803 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7804 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7805 | * again | |
fa9c0d79 | 7806 | */ |
723bda20 | 7807 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7808 | index = 0; |
a5e681d9 JB |
7809 | if (loop == LOOP_CACHING_NOWAIT) { |
7810 | /* | |
7811 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7812 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7813 | * full search through. |
7814 | */ | |
13a0db5a | 7815 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7816 | loop = LOOP_CACHING_WAIT; |
7817 | else | |
7818 | loop = LOOP_ALLOC_CHUNK; | |
7819 | } else { | |
7820 | loop++; | |
7821 | } | |
7822 | ||
817d52f8 | 7823 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7824 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7825 | int exist = 0; |
7826 | ||
7827 | trans = current->journal_info; | |
7828 | if (trans) | |
7829 | exist = 1; | |
7830 | else | |
7831 | trans = btrfs_join_transaction(root); | |
00361589 | 7832 | |
00361589 JB |
7833 | if (IS_ERR(trans)) { |
7834 | ret = PTR_ERR(trans); | |
7835 | goto out; | |
7836 | } | |
7837 | ||
b6919a58 | 7838 | ret = do_chunk_alloc(trans, root, flags, |
ea658bad | 7839 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7840 | |
7841 | /* | |
7842 | * If we can't allocate a new chunk we've already looped | |
7843 | * through at least once, move on to the NO_EMPTY_SIZE | |
7844 | * case. | |
7845 | */ | |
7846 | if (ret == -ENOSPC) | |
7847 | loop = LOOP_NO_EMPTY_SIZE; | |
7848 | ||
ea658bad JB |
7849 | /* |
7850 | * Do not bail out on ENOSPC since we | |
7851 | * can do more things. | |
7852 | */ | |
00361589 | 7853 | if (ret < 0 && ret != -ENOSPC) |
ea658bad JB |
7854 | btrfs_abort_transaction(trans, |
7855 | root, ret); | |
00361589 JB |
7856 | else |
7857 | ret = 0; | |
f017f15f WS |
7858 | if (!exist) |
7859 | btrfs_end_transaction(trans, root); | |
00361589 | 7860 | if (ret) |
ea658bad | 7861 | goto out; |
2552d17e JB |
7862 | } |
7863 | ||
723bda20 | 7864 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7865 | /* |
7866 | * Don't loop again if we already have no empty_size and | |
7867 | * no empty_cluster. | |
7868 | */ | |
7869 | if (empty_size == 0 && | |
7870 | empty_cluster == 0) { | |
7871 | ret = -ENOSPC; | |
7872 | goto out; | |
7873 | } | |
723bda20 JB |
7874 | empty_size = 0; |
7875 | empty_cluster = 0; | |
fa9c0d79 | 7876 | } |
723bda20 JB |
7877 | |
7878 | goto search; | |
2552d17e JB |
7879 | } else if (!ins->objectid) { |
7880 | ret = -ENOSPC; | |
d82a6f1d | 7881 | } else if (ins->objectid) { |
c759c4e1 JB |
7882 | if (!use_cluster && last_ptr) { |
7883 | spin_lock(&last_ptr->lock); | |
7884 | last_ptr->window_start = ins->objectid; | |
7885 | spin_unlock(&last_ptr->lock); | |
7886 | } | |
80eb234a | 7887 | ret = 0; |
be744175 | 7888 | } |
79787eaa | 7889 | out: |
4f4db217 JB |
7890 | if (ret == -ENOSPC) { |
7891 | spin_lock(&space_info->lock); | |
7892 | space_info->max_extent_size = max_extent_size; | |
7893 | spin_unlock(&space_info->lock); | |
a4820398 | 7894 | ins->offset = max_extent_size; |
4f4db217 | 7895 | } |
0f70abe2 | 7896 | return ret; |
fec577fb | 7897 | } |
ec44a35c | 7898 | |
9ed74f2d JB |
7899 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
7900 | int dump_block_groups) | |
0f9dd46c JB |
7901 | { |
7902 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7903 | int index = 0; |
0f9dd46c | 7904 | |
9ed74f2d | 7905 | spin_lock(&info->lock); |
efe120a0 | 7906 | printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n", |
c1c9ff7c GU |
7907 | info->flags, |
7908 | info->total_bytes - info->bytes_used - info->bytes_pinned - | |
7909 | info->bytes_reserved - info->bytes_readonly, | |
d397712b | 7910 | (info->full) ? "" : "not "); |
efe120a0 | 7911 | printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, " |
8929ecfa | 7912 | "reserved=%llu, may_use=%llu, readonly=%llu\n", |
c1c9ff7c GU |
7913 | info->total_bytes, info->bytes_used, info->bytes_pinned, |
7914 | info->bytes_reserved, info->bytes_may_use, | |
7915 | info->bytes_readonly); | |
9ed74f2d JB |
7916 | spin_unlock(&info->lock); |
7917 | ||
7918 | if (!dump_block_groups) | |
7919 | return; | |
0f9dd46c | 7920 | |
80eb234a | 7921 | down_read(&info->groups_sem); |
b742bb82 YZ |
7922 | again: |
7923 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7924 | spin_lock(&cache->lock); |
efe120a0 FH |
7925 | printk(KERN_INFO "BTRFS: " |
7926 | "block group %llu has %llu bytes, " | |
7927 | "%llu used %llu pinned %llu reserved %s\n", | |
c1c9ff7c GU |
7928 | cache->key.objectid, cache->key.offset, |
7929 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7930 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7931 | btrfs_dump_free_space(cache, bytes); |
7932 | spin_unlock(&cache->lock); | |
7933 | } | |
b742bb82 YZ |
7934 | if (++index < BTRFS_NR_RAID_TYPES) |
7935 | goto again; | |
80eb234a | 7936 | up_read(&info->groups_sem); |
0f9dd46c | 7937 | } |
e8569813 | 7938 | |
00361589 | 7939 | int btrfs_reserve_extent(struct btrfs_root *root, |
11833d66 YZ |
7940 | u64 num_bytes, u64 min_alloc_size, |
7941 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7942 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7943 | { |
36af4e07 | 7944 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7945 | u64 flags; |
fec577fb | 7946 | int ret; |
925baedd | 7947 | |
b6919a58 | 7948 | flags = btrfs_get_alloc_profile(root, is_data); |
98d20f67 | 7949 | again: |
db94535d | 7950 | WARN_ON(num_bytes < root->sectorsize); |
00361589 | 7951 | ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins, |
e570fd27 | 7952 | flags, delalloc); |
9cfa3e34 FM |
7953 | if (!ret && !is_data) { |
7954 | btrfs_dec_block_group_reservations(root->fs_info, | |
7955 | ins->objectid); | |
7956 | } else if (ret == -ENOSPC) { | |
a4820398 MX |
7957 | if (!final_tried && ins->offset) { |
7958 | num_bytes = min(num_bytes >> 1, ins->offset); | |
24542bf7 | 7959 | num_bytes = round_down(num_bytes, root->sectorsize); |
9e622d6b | 7960 | num_bytes = max(num_bytes, min_alloc_size); |
9e622d6b MX |
7961 | if (num_bytes == min_alloc_size) |
7962 | final_tried = true; | |
7963 | goto again; | |
7964 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
7965 | struct btrfs_space_info *sinfo; | |
7966 | ||
b6919a58 | 7967 | sinfo = __find_space_info(root->fs_info, flags); |
c2cf52eb | 7968 | btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu", |
c1c9ff7c | 7969 | flags, num_bytes); |
53804280 JM |
7970 | if (sinfo) |
7971 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 7972 | } |
925baedd | 7973 | } |
0f9dd46c JB |
7974 | |
7975 | return ret; | |
e6dcd2dc CM |
7976 | } |
7977 | ||
e688b725 | 7978 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 MX |
7979 | u64 start, u64 len, |
7980 | int pin, int delalloc) | |
65b51a00 | 7981 | { |
0f9dd46c | 7982 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 7983 | int ret = 0; |
0f9dd46c | 7984 | |
0f9dd46c JB |
7985 | cache = btrfs_lookup_block_group(root->fs_info, start); |
7986 | if (!cache) { | |
c2cf52eb | 7987 | btrfs_err(root->fs_info, "Unable to find block group for %llu", |
c1c9ff7c | 7988 | start); |
0f9dd46c JB |
7989 | return -ENOSPC; |
7990 | } | |
1f3c79a2 | 7991 | |
e688b725 CM |
7992 | if (pin) |
7993 | pin_down_extent(root, cache, start, len, 1); | |
7994 | else { | |
dcc82f47 FM |
7995 | if (btrfs_test_opt(root, DISCARD)) |
7996 | ret = btrfs_discard_extent(root, start, len, NULL); | |
e688b725 | 7997 | btrfs_add_free_space(cache, start, len); |
e570fd27 | 7998 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc); |
e688b725 | 7999 | } |
31193213 | 8000 | |
fa9c0d79 | 8001 | btrfs_put_block_group(cache); |
817d52f8 | 8002 | |
1abe9b8a | 8003 | trace_btrfs_reserved_extent_free(root, start, len); |
8004 | ||
e6dcd2dc CM |
8005 | return ret; |
8006 | } | |
8007 | ||
e688b725 | 8008 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 | 8009 | u64 start, u64 len, int delalloc) |
e688b725 | 8010 | { |
e570fd27 | 8011 | return __btrfs_free_reserved_extent(root, start, len, 0, delalloc); |
e688b725 CM |
8012 | } |
8013 | ||
8014 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
8015 | u64 start, u64 len) | |
8016 | { | |
e570fd27 | 8017 | return __btrfs_free_reserved_extent(root, start, len, 1, 0); |
e688b725 CM |
8018 | } |
8019 | ||
5d4f98a2 YZ |
8020 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
8021 | struct btrfs_root *root, | |
8022 | u64 parent, u64 root_objectid, | |
8023 | u64 flags, u64 owner, u64 offset, | |
8024 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
8025 | { |
8026 | int ret; | |
5d4f98a2 | 8027 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 8028 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8029 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8030 | struct btrfs_path *path; |
5d4f98a2 YZ |
8031 | struct extent_buffer *leaf; |
8032 | int type; | |
8033 | u32 size; | |
26b8003f | 8034 | |
5d4f98a2 YZ |
8035 | if (parent > 0) |
8036 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8037 | else | |
8038 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8039 | |
5d4f98a2 | 8040 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8041 | |
8042 | path = btrfs_alloc_path(); | |
db5b493a TI |
8043 | if (!path) |
8044 | return -ENOMEM; | |
47e4bb98 | 8045 | |
b9473439 | 8046 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8047 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8048 | ins, size); | |
79787eaa JM |
8049 | if (ret) { |
8050 | btrfs_free_path(path); | |
8051 | return ret; | |
8052 | } | |
0f9dd46c | 8053 | |
5d4f98a2 YZ |
8054 | leaf = path->nodes[0]; |
8055 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8056 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8057 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8058 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8059 | btrfs_set_extent_flags(leaf, extent_item, | |
8060 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8061 | ||
8062 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8063 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8064 | if (parent > 0) { | |
8065 | struct btrfs_shared_data_ref *ref; | |
8066 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8067 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8068 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8069 | } else { | |
8070 | struct btrfs_extent_data_ref *ref; | |
8071 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8072 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8073 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8074 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8075 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8076 | } | |
47e4bb98 CM |
8077 | |
8078 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8079 | btrfs_free_path(path); |
f510cfec | 8080 | |
1e144fb8 OS |
8081 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8082 | ins->offset); | |
8083 | if (ret) | |
8084 | return ret; | |
8085 | ||
ce93ec54 | 8086 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 8087 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8088 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8089 | ins->objectid, ins->offset); |
f5947066 CM |
8090 | BUG(); |
8091 | } | |
0be5dc67 | 8092 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
e6dcd2dc CM |
8093 | return ret; |
8094 | } | |
8095 | ||
5d4f98a2 YZ |
8096 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
8097 | struct btrfs_root *root, | |
8098 | u64 parent, u64 root_objectid, | |
8099 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 8100 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
8101 | { |
8102 | int ret; | |
5d4f98a2 YZ |
8103 | struct btrfs_fs_info *fs_info = root->fs_info; |
8104 | struct btrfs_extent_item *extent_item; | |
8105 | struct btrfs_tree_block_info *block_info; | |
8106 | struct btrfs_extent_inline_ref *iref; | |
8107 | struct btrfs_path *path; | |
8108 | struct extent_buffer *leaf; | |
3173a18f | 8109 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 8110 | u64 num_bytes = ins->offset; |
3173a18f JB |
8111 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
8112 | SKINNY_METADATA); | |
8113 | ||
8114 | if (!skinny_metadata) | |
8115 | size += sizeof(*block_info); | |
1c2308f8 | 8116 | |
5d4f98a2 | 8117 | path = btrfs_alloc_path(); |
857cc2fc JB |
8118 | if (!path) { |
8119 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, | |
707e8a07 | 8120 | root->nodesize); |
d8926bb3 | 8121 | return -ENOMEM; |
857cc2fc | 8122 | } |
56bec294 | 8123 | |
5d4f98a2 YZ |
8124 | path->leave_spinning = 1; |
8125 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
8126 | ins, size); | |
79787eaa | 8127 | if (ret) { |
dd825259 | 8128 | btrfs_free_path(path); |
857cc2fc | 8129 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, |
707e8a07 | 8130 | root->nodesize); |
79787eaa JM |
8131 | return ret; |
8132 | } | |
5d4f98a2 YZ |
8133 | |
8134 | leaf = path->nodes[0]; | |
8135 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8136 | struct btrfs_extent_item); | |
8137 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8138 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8139 | btrfs_set_extent_flags(leaf, extent_item, | |
8140 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8141 | |
3173a18f JB |
8142 | if (skinny_metadata) { |
8143 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
707e8a07 | 8144 | num_bytes = root->nodesize; |
3173a18f JB |
8145 | } else { |
8146 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
8147 | btrfs_set_tree_block_key(leaf, block_info, key); | |
8148 | btrfs_set_tree_block_level(leaf, block_info, level); | |
8149 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
8150 | } | |
5d4f98a2 | 8151 | |
5d4f98a2 YZ |
8152 | if (parent > 0) { |
8153 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
8154 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8155 | BTRFS_SHARED_BLOCK_REF_KEY); | |
8156 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8157 | } else { | |
8158 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8159 | BTRFS_TREE_BLOCK_REF_KEY); | |
8160 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
8161 | } | |
8162 | ||
8163 | btrfs_mark_buffer_dirty(leaf); | |
8164 | btrfs_free_path(path); | |
8165 | ||
1e144fb8 OS |
8166 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8167 | num_bytes); | |
8168 | if (ret) | |
8169 | return ret; | |
8170 | ||
ce93ec54 JB |
8171 | ret = update_block_group(trans, root, ins->objectid, root->nodesize, |
8172 | 1); | |
79787eaa | 8173 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8174 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8175 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
8176 | BUG(); |
8177 | } | |
0be5dc67 | 8178 | |
707e8a07 | 8179 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize); |
5d4f98a2 YZ |
8180 | return ret; |
8181 | } | |
8182 | ||
8183 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
8184 | struct btrfs_root *root, | |
8185 | u64 root_objectid, u64 owner, | |
5846a3c2 QW |
8186 | u64 offset, u64 ram_bytes, |
8187 | struct btrfs_key *ins) | |
5d4f98a2 YZ |
8188 | { |
8189 | int ret; | |
8190 | ||
8191 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
8192 | ||
66d7e7f0 AJ |
8193 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
8194 | ins->offset, 0, | |
8195 | root_objectid, owner, offset, | |
5846a3c2 QW |
8196 | ram_bytes, BTRFS_ADD_DELAYED_EXTENT, |
8197 | NULL); | |
e6dcd2dc CM |
8198 | return ret; |
8199 | } | |
e02119d5 CM |
8200 | |
8201 | /* | |
8202 | * this is used by the tree logging recovery code. It records that | |
8203 | * an extent has been allocated and makes sure to clear the free | |
8204 | * space cache bits as well | |
8205 | */ | |
5d4f98a2 YZ |
8206 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
8207 | struct btrfs_root *root, | |
8208 | u64 root_objectid, u64 owner, u64 offset, | |
8209 | struct btrfs_key *ins) | |
e02119d5 CM |
8210 | { |
8211 | int ret; | |
8212 | struct btrfs_block_group_cache *block_group; | |
11833d66 | 8213 | |
8c2a1a30 JB |
8214 | /* |
8215 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8216 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 JB |
8217 | */ |
8218 | if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) { | |
8219 | ret = __exclude_logged_extent(root, ins->objectid, ins->offset); | |
b50c6e25 | 8220 | if (ret) |
8c2a1a30 | 8221 | return ret; |
11833d66 YZ |
8222 | } |
8223 | ||
8c2a1a30 JB |
8224 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
8225 | if (!block_group) | |
8226 | return -EINVAL; | |
8227 | ||
fb25e914 | 8228 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
e570fd27 | 8229 | RESERVE_ALLOC_NO_ACCOUNT, 0); |
79787eaa | 8230 | BUG_ON(ret); /* logic error */ |
5d4f98a2 YZ |
8231 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
8232 | 0, owner, offset, ins, 1); | |
b50c6e25 | 8233 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8234 | return ret; |
8235 | } | |
8236 | ||
48a3b636 ES |
8237 | static struct extent_buffer * |
8238 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8239 | u64 bytenr, int level) |
65b51a00 CM |
8240 | { |
8241 | struct extent_buffer *buf; | |
8242 | ||
a83fffb7 | 8243 | buf = btrfs_find_create_tree_block(root, bytenr); |
c871b0f2 LB |
8244 | if (IS_ERR(buf)) |
8245 | return buf; | |
8246 | ||
65b51a00 | 8247 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8248 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8249 | btrfs_tree_lock(buf); |
01d58472 | 8250 | clean_tree_block(trans, root->fs_info, buf); |
3083ee2e | 8251 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8252 | |
8253 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8254 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8255 | |
d0c803c4 | 8256 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8257 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8258 | /* |
8259 | * we allow two log transactions at a time, use different | |
8260 | * EXENT bit to differentiate dirty pages. | |
8261 | */ | |
656f30db | 8262 | if (buf->log_index == 0) |
8cef4e16 YZ |
8263 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8264 | buf->start + buf->len - 1, GFP_NOFS); | |
8265 | else | |
8266 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8267 | buf->start + buf->len - 1); |
d0c803c4 | 8268 | } else { |
656f30db | 8269 | buf->log_index = -1; |
d0c803c4 | 8270 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8271 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8272 | } |
64c12921 | 8273 | trans->dirty = true; |
b4ce94de | 8274 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8275 | return buf; |
8276 | } | |
8277 | ||
f0486c68 YZ |
8278 | static struct btrfs_block_rsv * |
8279 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8280 | struct btrfs_root *root, u32 blocksize) | |
8281 | { | |
8282 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 8283 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 | 8284 | int ret; |
d88033db | 8285 | bool global_updated = false; |
f0486c68 YZ |
8286 | |
8287 | block_rsv = get_block_rsv(trans, root); | |
8288 | ||
b586b323 MX |
8289 | if (unlikely(block_rsv->size == 0)) |
8290 | goto try_reserve; | |
d88033db | 8291 | again: |
f0486c68 YZ |
8292 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8293 | if (!ret) | |
8294 | return block_rsv; | |
8295 | ||
b586b323 MX |
8296 | if (block_rsv->failfast) |
8297 | return ERR_PTR(ret); | |
8298 | ||
d88033db MX |
8299 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8300 | global_updated = true; | |
8301 | update_global_block_rsv(root->fs_info); | |
8302 | goto again; | |
8303 | } | |
8304 | ||
b586b323 MX |
8305 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { |
8306 | static DEFINE_RATELIMIT_STATE(_rs, | |
8307 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8308 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8309 | if (__ratelimit(&_rs)) | |
8310 | WARN(1, KERN_DEBUG | |
efe120a0 | 8311 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8312 | } |
8313 | try_reserve: | |
8314 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8315 | BTRFS_RESERVE_NO_FLUSH); | |
8316 | if (!ret) | |
8317 | return block_rsv; | |
8318 | /* | |
8319 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8320 | * the global reserve if its space type is the same as the global |
8321 | * reservation. | |
b586b323 | 8322 | */ |
5881cfc9 MX |
8323 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8324 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8325 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8326 | if (!ret) | |
8327 | return global_rsv; | |
8328 | } | |
8329 | return ERR_PTR(ret); | |
f0486c68 YZ |
8330 | } |
8331 | ||
8c2a3ca2 JB |
8332 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8333 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8334 | { |
8335 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 8336 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
8337 | } |
8338 | ||
fec577fb | 8339 | /* |
f0486c68 | 8340 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8341 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8342 | */ |
4d75f8a9 DS |
8343 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
8344 | struct btrfs_root *root, | |
5d4f98a2 YZ |
8345 | u64 parent, u64 root_objectid, |
8346 | struct btrfs_disk_key *key, int level, | |
5581a51a | 8347 | u64 hint, u64 empty_size) |
fec577fb | 8348 | { |
e2fa7227 | 8349 | struct btrfs_key ins; |
f0486c68 | 8350 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8351 | struct extent_buffer *buf; |
67b7859e | 8352 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8353 | u64 flags = 0; |
8354 | int ret; | |
4d75f8a9 | 8355 | u32 blocksize = root->nodesize; |
3173a18f JB |
8356 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
8357 | SKINNY_METADATA); | |
fec577fb | 8358 | |
fccb84c9 | 8359 | if (btrfs_test_is_dummy_root(root)) { |
faa2dbf0 | 8360 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8361 | level); |
faa2dbf0 JB |
8362 | if (!IS_ERR(buf)) |
8363 | root->alloc_bytenr += blocksize; | |
8364 | return buf; | |
8365 | } | |
fccb84c9 | 8366 | |
f0486c68 YZ |
8367 | block_rsv = use_block_rsv(trans, root, blocksize); |
8368 | if (IS_ERR(block_rsv)) | |
8369 | return ERR_CAST(block_rsv); | |
8370 | ||
00361589 | 8371 | ret = btrfs_reserve_extent(root, blocksize, blocksize, |
e570fd27 | 8372 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8373 | if (ret) |
8374 | goto out_unuse; | |
55c69072 | 8375 | |
fe864576 | 8376 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8377 | if (IS_ERR(buf)) { |
8378 | ret = PTR_ERR(buf); | |
8379 | goto out_free_reserved; | |
8380 | } | |
f0486c68 YZ |
8381 | |
8382 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8383 | if (parent == 0) | |
8384 | parent = ins.objectid; | |
8385 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8386 | } else | |
8387 | BUG_ON(parent > 0); | |
8388 | ||
8389 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8390 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8391 | if (!extent_op) { |
8392 | ret = -ENOMEM; | |
8393 | goto out_free_buf; | |
8394 | } | |
f0486c68 YZ |
8395 | if (key) |
8396 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8397 | else | |
8398 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8399 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8400 | extent_op->update_key = skinny_metadata ? false : true; |
8401 | extent_op->update_flags = true; | |
8402 | extent_op->is_data = false; | |
b1c79e09 | 8403 | extent_op->level = level; |
f0486c68 | 8404 | |
66d7e7f0 | 8405 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
67b7859e OS |
8406 | ins.objectid, ins.offset, |
8407 | parent, root_objectid, level, | |
8408 | BTRFS_ADD_DELAYED_EXTENT, | |
b06c4bf5 | 8409 | extent_op); |
67b7859e OS |
8410 | if (ret) |
8411 | goto out_free_delayed; | |
f0486c68 | 8412 | } |
fec577fb | 8413 | return buf; |
67b7859e OS |
8414 | |
8415 | out_free_delayed: | |
8416 | btrfs_free_delayed_extent_op(extent_op); | |
8417 | out_free_buf: | |
8418 | free_extent_buffer(buf); | |
8419 | out_free_reserved: | |
8420 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0); | |
8421 | out_unuse: | |
8422 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); | |
8423 | return ERR_PTR(ret); | |
fec577fb | 8424 | } |
a28ec197 | 8425 | |
2c47e605 YZ |
8426 | struct walk_control { |
8427 | u64 refs[BTRFS_MAX_LEVEL]; | |
8428 | u64 flags[BTRFS_MAX_LEVEL]; | |
8429 | struct btrfs_key update_progress; | |
8430 | int stage; | |
8431 | int level; | |
8432 | int shared_level; | |
8433 | int update_ref; | |
8434 | int keep_locks; | |
1c4850e2 YZ |
8435 | int reada_slot; |
8436 | int reada_count; | |
66d7e7f0 | 8437 | int for_reloc; |
2c47e605 YZ |
8438 | }; |
8439 | ||
8440 | #define DROP_REFERENCE 1 | |
8441 | #define UPDATE_BACKREF 2 | |
8442 | ||
1c4850e2 YZ |
8443 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8444 | struct btrfs_root *root, | |
8445 | struct walk_control *wc, | |
8446 | struct btrfs_path *path) | |
6407bf6d | 8447 | { |
1c4850e2 YZ |
8448 | u64 bytenr; |
8449 | u64 generation; | |
8450 | u64 refs; | |
94fcca9f | 8451 | u64 flags; |
5d4f98a2 | 8452 | u32 nritems; |
1c4850e2 YZ |
8453 | u32 blocksize; |
8454 | struct btrfs_key key; | |
8455 | struct extent_buffer *eb; | |
6407bf6d | 8456 | int ret; |
1c4850e2 YZ |
8457 | int slot; |
8458 | int nread = 0; | |
6407bf6d | 8459 | |
1c4850e2 YZ |
8460 | if (path->slots[wc->level] < wc->reada_slot) { |
8461 | wc->reada_count = wc->reada_count * 2 / 3; | |
8462 | wc->reada_count = max(wc->reada_count, 2); | |
8463 | } else { | |
8464 | wc->reada_count = wc->reada_count * 3 / 2; | |
8465 | wc->reada_count = min_t(int, wc->reada_count, | |
8466 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
8467 | } | |
7bb86316 | 8468 | |
1c4850e2 YZ |
8469 | eb = path->nodes[wc->level]; |
8470 | nritems = btrfs_header_nritems(eb); | |
707e8a07 | 8471 | blocksize = root->nodesize; |
bd56b302 | 8472 | |
1c4850e2 YZ |
8473 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8474 | if (nread >= wc->reada_count) | |
8475 | break; | |
bd56b302 | 8476 | |
2dd3e67b | 8477 | cond_resched(); |
1c4850e2 YZ |
8478 | bytenr = btrfs_node_blockptr(eb, slot); |
8479 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8480 | |
1c4850e2 YZ |
8481 | if (slot == path->slots[wc->level]) |
8482 | goto reada; | |
5d4f98a2 | 8483 | |
1c4850e2 YZ |
8484 | if (wc->stage == UPDATE_BACKREF && |
8485 | generation <= root->root_key.offset) | |
bd56b302 CM |
8486 | continue; |
8487 | ||
94fcca9f | 8488 | /* We don't lock the tree block, it's OK to be racy here */ |
3173a18f JB |
8489 | ret = btrfs_lookup_extent_info(trans, root, bytenr, |
8490 | wc->level - 1, 1, &refs, | |
8491 | &flags); | |
79787eaa JM |
8492 | /* We don't care about errors in readahead. */ |
8493 | if (ret < 0) | |
8494 | continue; | |
94fcca9f YZ |
8495 | BUG_ON(refs == 0); |
8496 | ||
1c4850e2 | 8497 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8498 | if (refs == 1) |
8499 | goto reada; | |
bd56b302 | 8500 | |
94fcca9f YZ |
8501 | if (wc->level == 1 && |
8502 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8503 | continue; | |
1c4850e2 YZ |
8504 | if (!wc->update_ref || |
8505 | generation <= root->root_key.offset) | |
8506 | continue; | |
8507 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8508 | ret = btrfs_comp_cpu_keys(&key, | |
8509 | &wc->update_progress); | |
8510 | if (ret < 0) | |
8511 | continue; | |
94fcca9f YZ |
8512 | } else { |
8513 | if (wc->level == 1 && | |
8514 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8515 | continue; | |
6407bf6d | 8516 | } |
1c4850e2 | 8517 | reada: |
d3e46fea | 8518 | readahead_tree_block(root, bytenr); |
1c4850e2 | 8519 | nread++; |
20524f02 | 8520 | } |
1c4850e2 | 8521 | wc->reada_slot = slot; |
20524f02 | 8522 | } |
2c47e605 | 8523 | |
0ed4792a | 8524 | /* |
82bd101b MF |
8525 | * These may not be seen by the usual inc/dec ref code so we have to |
8526 | * add them here. | |
0ed4792a | 8527 | */ |
82bd101b MF |
8528 | static int record_one_subtree_extent(struct btrfs_trans_handle *trans, |
8529 | struct btrfs_root *root, u64 bytenr, | |
8530 | u64 num_bytes) | |
8531 | { | |
8532 | struct btrfs_qgroup_extent_record *qrecord; | |
8533 | struct btrfs_delayed_ref_root *delayed_refs; | |
8534 | ||
8535 | qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS); | |
8536 | if (!qrecord) | |
8537 | return -ENOMEM; | |
8538 | ||
8539 | qrecord->bytenr = bytenr; | |
8540 | qrecord->num_bytes = num_bytes; | |
8541 | qrecord->old_roots = NULL; | |
8542 | ||
8543 | delayed_refs = &trans->transaction->delayed_refs; | |
8544 | spin_lock(&delayed_refs->lock); | |
8545 | if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord)) | |
8546 | kfree(qrecord); | |
8547 | spin_unlock(&delayed_refs->lock); | |
8548 | ||
8549 | return 0; | |
8550 | } | |
8551 | ||
1152651a MF |
8552 | static int account_leaf_items(struct btrfs_trans_handle *trans, |
8553 | struct btrfs_root *root, | |
8554 | struct extent_buffer *eb) | |
8555 | { | |
8556 | int nr = btrfs_header_nritems(eb); | |
82bd101b | 8557 | int i, extent_type, ret; |
1152651a MF |
8558 | struct btrfs_key key; |
8559 | struct btrfs_file_extent_item *fi; | |
8560 | u64 bytenr, num_bytes; | |
8561 | ||
82bd101b MF |
8562 | /* We can be called directly from walk_up_proc() */ |
8563 | if (!root->fs_info->quota_enabled) | |
8564 | return 0; | |
8565 | ||
1152651a MF |
8566 | for (i = 0; i < nr; i++) { |
8567 | btrfs_item_key_to_cpu(eb, &key, i); | |
8568 | ||
8569 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
8570 | continue; | |
8571 | ||
8572 | fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
8573 | /* filter out non qgroup-accountable extents */ | |
8574 | extent_type = btrfs_file_extent_type(eb, fi); | |
8575 | ||
8576 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
8577 | continue; | |
8578 | ||
8579 | bytenr = btrfs_file_extent_disk_bytenr(eb, fi); | |
8580 | if (!bytenr) | |
8581 | continue; | |
8582 | ||
8583 | num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi); | |
82bd101b MF |
8584 | |
8585 | ret = record_one_subtree_extent(trans, root, bytenr, num_bytes); | |
8586 | if (ret) | |
8587 | return ret; | |
1152651a MF |
8588 | } |
8589 | return 0; | |
8590 | } | |
8591 | ||
8592 | /* | |
8593 | * Walk up the tree from the bottom, freeing leaves and any interior | |
8594 | * nodes which have had all slots visited. If a node (leaf or | |
8595 | * interior) is freed, the node above it will have it's slot | |
8596 | * incremented. The root node will never be freed. | |
8597 | * | |
8598 | * At the end of this function, we should have a path which has all | |
8599 | * slots incremented to the next position for a search. If we need to | |
8600 | * read a new node it will be NULL and the node above it will have the | |
8601 | * correct slot selected for a later read. | |
8602 | * | |
8603 | * If we increment the root nodes slot counter past the number of | |
8604 | * elements, 1 is returned to signal completion of the search. | |
8605 | */ | |
8606 | static int adjust_slots_upwards(struct btrfs_root *root, | |
8607 | struct btrfs_path *path, int root_level) | |
8608 | { | |
8609 | int level = 0; | |
8610 | int nr, slot; | |
8611 | struct extent_buffer *eb; | |
8612 | ||
8613 | if (root_level == 0) | |
8614 | return 1; | |
8615 | ||
8616 | while (level <= root_level) { | |
8617 | eb = path->nodes[level]; | |
8618 | nr = btrfs_header_nritems(eb); | |
8619 | path->slots[level]++; | |
8620 | slot = path->slots[level]; | |
8621 | if (slot >= nr || level == 0) { | |
8622 | /* | |
8623 | * Don't free the root - we will detect this | |
8624 | * condition after our loop and return a | |
8625 | * positive value for caller to stop walking the tree. | |
8626 | */ | |
8627 | if (level != root_level) { | |
8628 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
8629 | path->locks[level] = 0; | |
8630 | ||
8631 | free_extent_buffer(eb); | |
8632 | path->nodes[level] = NULL; | |
8633 | path->slots[level] = 0; | |
8634 | } | |
8635 | } else { | |
8636 | /* | |
8637 | * We have a valid slot to walk back down | |
8638 | * from. Stop here so caller can process these | |
8639 | * new nodes. | |
8640 | */ | |
8641 | break; | |
8642 | } | |
8643 | ||
8644 | level++; | |
8645 | } | |
8646 | ||
8647 | eb = path->nodes[root_level]; | |
8648 | if (path->slots[root_level] >= btrfs_header_nritems(eb)) | |
8649 | return 1; | |
8650 | ||
8651 | return 0; | |
8652 | } | |
8653 | ||
8654 | /* | |
8655 | * root_eb is the subtree root and is locked before this function is called. | |
8656 | */ | |
8657 | static int account_shared_subtree(struct btrfs_trans_handle *trans, | |
8658 | struct btrfs_root *root, | |
8659 | struct extent_buffer *root_eb, | |
8660 | u64 root_gen, | |
8661 | int root_level) | |
8662 | { | |
8663 | int ret = 0; | |
8664 | int level; | |
8665 | struct extent_buffer *eb = root_eb; | |
8666 | struct btrfs_path *path = NULL; | |
8667 | ||
8668 | BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL); | |
8669 | BUG_ON(root_eb == NULL); | |
8670 | ||
8671 | if (!root->fs_info->quota_enabled) | |
8672 | return 0; | |
8673 | ||
8674 | if (!extent_buffer_uptodate(root_eb)) { | |
8675 | ret = btrfs_read_buffer(root_eb, root_gen); | |
8676 | if (ret) | |
8677 | goto out; | |
8678 | } | |
8679 | ||
8680 | if (root_level == 0) { | |
8681 | ret = account_leaf_items(trans, root, root_eb); | |
8682 | goto out; | |
8683 | } | |
8684 | ||
8685 | path = btrfs_alloc_path(); | |
8686 | if (!path) | |
8687 | return -ENOMEM; | |
8688 | ||
8689 | /* | |
8690 | * Walk down the tree. Missing extent blocks are filled in as | |
8691 | * we go. Metadata is accounted every time we read a new | |
8692 | * extent block. | |
8693 | * | |
8694 | * When we reach a leaf, we account for file extent items in it, | |
8695 | * walk back up the tree (adjusting slot pointers as we go) | |
8696 | * and restart the search process. | |
8697 | */ | |
8698 | extent_buffer_get(root_eb); /* For path */ | |
8699 | path->nodes[root_level] = root_eb; | |
8700 | path->slots[root_level] = 0; | |
8701 | path->locks[root_level] = 0; /* so release_path doesn't try to unlock */ | |
8702 | walk_down: | |
8703 | level = root_level; | |
8704 | while (level >= 0) { | |
8705 | if (path->nodes[level] == NULL) { | |
1152651a MF |
8706 | int parent_slot; |
8707 | u64 child_gen; | |
8708 | u64 child_bytenr; | |
8709 | ||
8710 | /* We need to get child blockptr/gen from | |
8711 | * parent before we can read it. */ | |
8712 | eb = path->nodes[level + 1]; | |
8713 | parent_slot = path->slots[level + 1]; | |
8714 | child_bytenr = btrfs_node_blockptr(eb, parent_slot); | |
8715 | child_gen = btrfs_node_ptr_generation(eb, parent_slot); | |
8716 | ||
ce86cd59 | 8717 | eb = read_tree_block(root, child_bytenr, child_gen); |
64c043de LB |
8718 | if (IS_ERR(eb)) { |
8719 | ret = PTR_ERR(eb); | |
8720 | goto out; | |
8721 | } else if (!extent_buffer_uptodate(eb)) { | |
8635eda9 | 8722 | free_extent_buffer(eb); |
64c043de | 8723 | ret = -EIO; |
1152651a MF |
8724 | goto out; |
8725 | } | |
8726 | ||
8727 | path->nodes[level] = eb; | |
8728 | path->slots[level] = 0; | |
8729 | ||
8730 | btrfs_tree_read_lock(eb); | |
8731 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
8732 | path->locks[level] = BTRFS_READ_LOCK_BLOCKING; | |
82bd101b MF |
8733 | |
8734 | ret = record_one_subtree_extent(trans, root, child_bytenr, | |
8735 | root->nodesize); | |
8736 | if (ret) | |
8737 | goto out; | |
1152651a MF |
8738 | } |
8739 | ||
8740 | if (level == 0) { | |
8741 | ret = account_leaf_items(trans, root, path->nodes[level]); | |
8742 | if (ret) | |
8743 | goto out; | |
8744 | ||
8745 | /* Nonzero return here means we completed our search */ | |
8746 | ret = adjust_slots_upwards(root, path, root_level); | |
8747 | if (ret) | |
8748 | break; | |
8749 | ||
8750 | /* Restart search with new slots */ | |
8751 | goto walk_down; | |
8752 | } | |
8753 | ||
8754 | level--; | |
8755 | } | |
8756 | ||
8757 | ret = 0; | |
8758 | out: | |
8759 | btrfs_free_path(path); | |
8760 | ||
8761 | return ret; | |
8762 | } | |
8763 | ||
f82d02d9 | 8764 | /* |
2c016dc2 | 8765 | * helper to process tree block while walking down the tree. |
2c47e605 | 8766 | * |
2c47e605 YZ |
8767 | * when wc->stage == UPDATE_BACKREF, this function updates |
8768 | * back refs for pointers in the block. | |
8769 | * | |
8770 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8771 | */ |
2c47e605 | 8772 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8773 | struct btrfs_root *root, |
2c47e605 | 8774 | struct btrfs_path *path, |
94fcca9f | 8775 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8776 | { |
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 YZ |
8793 | BUG_ON(!path->locks[level]); |
8794 | ret = btrfs_lookup_extent_info(trans, root, | |
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 */ |
2c47e605 | 8822 | ret = btrfs_set_disk_extent_flags(trans, root, 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 | ||
1c4850e2 | 8840 | /* |
2c016dc2 | 8841 | * helper to process tree block pointer. |
1c4850e2 YZ |
8842 | * |
8843 | * when wc->stage == DROP_REFERENCE, this function checks | |
8844 | * reference count of the block pointed to. if the block | |
8845 | * is shared and we need update back refs for the subtree | |
8846 | * rooted at the block, this function changes wc->stage to | |
8847 | * UPDATE_BACKREF. if the block is shared and there is no | |
8848 | * need to update back, this function drops the reference | |
8849 | * to the block. | |
8850 | * | |
8851 | * NOTE: return value 1 means we should stop walking down. | |
8852 | */ | |
8853 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8854 | struct btrfs_root *root, | |
8855 | struct btrfs_path *path, | |
94fcca9f | 8856 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
8857 | { |
8858 | u64 bytenr; | |
8859 | u64 generation; | |
8860 | u64 parent; | |
8861 | u32 blocksize; | |
8862 | struct btrfs_key key; | |
8863 | struct extent_buffer *next; | |
8864 | int level = wc->level; | |
8865 | int reada = 0; | |
8866 | int ret = 0; | |
1152651a | 8867 | bool need_account = false; |
1c4850e2 YZ |
8868 | |
8869 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8870 | path->slots[level]); | |
8871 | /* | |
8872 | * if the lower level block was created before the snapshot | |
8873 | * was created, we know there is no need to update back refs | |
8874 | * for the subtree | |
8875 | */ | |
8876 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8877 | generation <= root->root_key.offset) { |
8878 | *lookup_info = 1; | |
1c4850e2 | 8879 | return 1; |
94fcca9f | 8880 | } |
1c4850e2 YZ |
8881 | |
8882 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
707e8a07 | 8883 | blocksize = root->nodesize; |
1c4850e2 | 8884 | |
01d58472 | 8885 | next = btrfs_find_tree_block(root->fs_info, bytenr); |
1c4850e2 | 8886 | if (!next) { |
a83fffb7 | 8887 | next = btrfs_find_create_tree_block(root, bytenr); |
c871b0f2 LB |
8888 | if (IS_ERR(next)) |
8889 | return PTR_ERR(next); | |
8890 | ||
b2aaaa3b JB |
8891 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8892 | level - 1); | |
1c4850e2 YZ |
8893 | reada = 1; |
8894 | } | |
8895 | btrfs_tree_lock(next); | |
8896 | btrfs_set_lock_blocking(next); | |
8897 | ||
3173a18f | 8898 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, |
94fcca9f YZ |
8899 | &wc->refs[level - 1], |
8900 | &wc->flags[level - 1]); | |
79787eaa JM |
8901 | if (ret < 0) { |
8902 | btrfs_tree_unlock(next); | |
8903 | return ret; | |
8904 | } | |
8905 | ||
c2cf52eb SK |
8906 | if (unlikely(wc->refs[level - 1] == 0)) { |
8907 | btrfs_err(root->fs_info, "Missing references."); | |
8908 | BUG(); | |
8909 | } | |
94fcca9f | 8910 | *lookup_info = 0; |
1c4850e2 | 8911 | |
94fcca9f | 8912 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8913 | if (wc->refs[level - 1] > 1) { |
1152651a | 8914 | need_account = true; |
94fcca9f YZ |
8915 | if (level == 1 && |
8916 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8917 | goto skip; | |
8918 | ||
1c4850e2 YZ |
8919 | if (!wc->update_ref || |
8920 | generation <= root->root_key.offset) | |
8921 | goto skip; | |
8922 | ||
8923 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8924 | path->slots[level]); | |
8925 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8926 | if (ret < 0) | |
8927 | goto skip; | |
8928 | ||
8929 | wc->stage = UPDATE_BACKREF; | |
8930 | wc->shared_level = level - 1; | |
8931 | } | |
94fcca9f YZ |
8932 | } else { |
8933 | if (level == 1 && | |
8934 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8935 | goto skip; | |
1c4850e2 YZ |
8936 | } |
8937 | ||
b9fab919 | 8938 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8939 | btrfs_tree_unlock(next); |
8940 | free_extent_buffer(next); | |
8941 | next = NULL; | |
94fcca9f | 8942 | *lookup_info = 1; |
1c4850e2 YZ |
8943 | } |
8944 | ||
8945 | if (!next) { | |
8946 | if (reada && level == 1) | |
8947 | reada_walk_down(trans, root, wc, path); | |
ce86cd59 | 8948 | next = read_tree_block(root, bytenr, generation); |
64c043de LB |
8949 | if (IS_ERR(next)) { |
8950 | return PTR_ERR(next); | |
8951 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8952 | free_extent_buffer(next); |
97d9a8a4 | 8953 | return -EIO; |
416bc658 | 8954 | } |
1c4850e2 YZ |
8955 | btrfs_tree_lock(next); |
8956 | btrfs_set_lock_blocking(next); | |
8957 | } | |
8958 | ||
8959 | level--; | |
8960 | BUG_ON(level != btrfs_header_level(next)); | |
8961 | path->nodes[level] = next; | |
8962 | path->slots[level] = 0; | |
bd681513 | 8963 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8964 | wc->level = level; |
8965 | if (wc->level == 1) | |
8966 | wc->reada_slot = 0; | |
8967 | return 0; | |
8968 | skip: | |
8969 | wc->refs[level - 1] = 0; | |
8970 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8971 | if (wc->stage == DROP_REFERENCE) { |
8972 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8973 | parent = path->nodes[level]->start; | |
8974 | } else { | |
8975 | BUG_ON(root->root_key.objectid != | |
8976 | btrfs_header_owner(path->nodes[level])); | |
8977 | parent = 0; | |
8978 | } | |
1c4850e2 | 8979 | |
1152651a MF |
8980 | if (need_account) { |
8981 | ret = account_shared_subtree(trans, root, next, | |
8982 | generation, level - 1); | |
8983 | if (ret) { | |
94647322 DS |
8984 | btrfs_err_rl(root->fs_info, |
8985 | "Error " | |
1152651a | 8986 | "%d accounting shared subtree. Quota " |
94647322 DS |
8987 | "is out of sync, rescan required.", |
8988 | ret); | |
1152651a MF |
8989 | } |
8990 | } | |
94fcca9f | 8991 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
b06c4bf5 | 8992 | root->root_key.objectid, level - 1, 0); |
79787eaa | 8993 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 8994 | } |
1c4850e2 YZ |
8995 | btrfs_tree_unlock(next); |
8996 | free_extent_buffer(next); | |
94fcca9f | 8997 | *lookup_info = 1; |
1c4850e2 YZ |
8998 | return 1; |
8999 | } | |
9000 | ||
2c47e605 | 9001 | /* |
2c016dc2 | 9002 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
9003 | * |
9004 | * when wc->stage == DROP_REFERENCE, this function drops | |
9005 | * reference count on the block. | |
9006 | * | |
9007 | * when wc->stage == UPDATE_BACKREF, this function changes | |
9008 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
9009 | * to UPDATE_BACKREF previously while processing the block. | |
9010 | * | |
9011 | * NOTE: return value 1 means we should stop walking up. | |
9012 | */ | |
9013 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
9014 | struct btrfs_root *root, | |
9015 | struct btrfs_path *path, | |
9016 | struct walk_control *wc) | |
9017 | { | |
f0486c68 | 9018 | int ret; |
2c47e605 YZ |
9019 | int level = wc->level; |
9020 | struct extent_buffer *eb = path->nodes[level]; | |
9021 | u64 parent = 0; | |
9022 | ||
9023 | if (wc->stage == UPDATE_BACKREF) { | |
9024 | BUG_ON(wc->shared_level < level); | |
9025 | if (level < wc->shared_level) | |
9026 | goto out; | |
9027 | ||
2c47e605 YZ |
9028 | ret = find_next_key(path, level + 1, &wc->update_progress); |
9029 | if (ret > 0) | |
9030 | wc->update_ref = 0; | |
9031 | ||
9032 | wc->stage = DROP_REFERENCE; | |
9033 | wc->shared_level = -1; | |
9034 | path->slots[level] = 0; | |
9035 | ||
9036 | /* | |
9037 | * check reference count again if the block isn't locked. | |
9038 | * we should start walking down the tree again if reference | |
9039 | * count is one. | |
9040 | */ | |
9041 | if (!path->locks[level]) { | |
9042 | BUG_ON(level == 0); | |
9043 | btrfs_tree_lock(eb); | |
9044 | btrfs_set_lock_blocking(eb); | |
bd681513 | 9045 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9046 | |
9047 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 9048 | eb->start, level, 1, |
2c47e605 YZ |
9049 | &wc->refs[level], |
9050 | &wc->flags[level]); | |
79787eaa JM |
9051 | if (ret < 0) { |
9052 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 9053 | path->locks[level] = 0; |
79787eaa JM |
9054 | return ret; |
9055 | } | |
2c47e605 YZ |
9056 | BUG_ON(wc->refs[level] == 0); |
9057 | if (wc->refs[level] == 1) { | |
bd681513 | 9058 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 9059 | path->locks[level] = 0; |
2c47e605 YZ |
9060 | return 1; |
9061 | } | |
f82d02d9 | 9062 | } |
2c47e605 | 9063 | } |
f82d02d9 | 9064 | |
2c47e605 YZ |
9065 | /* wc->stage == DROP_REFERENCE */ |
9066 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 9067 | |
2c47e605 YZ |
9068 | if (wc->refs[level] == 1) { |
9069 | if (level == 0) { | |
9070 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 9071 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 9072 | else |
e339a6b0 | 9073 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 9074 | BUG_ON(ret); /* -ENOMEM */ |
1152651a MF |
9075 | ret = account_leaf_items(trans, root, eb); |
9076 | if (ret) { | |
94647322 DS |
9077 | btrfs_err_rl(root->fs_info, |
9078 | "error " | |
1152651a | 9079 | "%d accounting leaf items. Quota " |
94647322 DS |
9080 | "is out of sync, rescan required.", |
9081 | ret); | |
1152651a | 9082 | } |
2c47e605 YZ |
9083 | } |
9084 | /* make block locked assertion in clean_tree_block happy */ | |
9085 | if (!path->locks[level] && | |
9086 | btrfs_header_generation(eb) == trans->transid) { | |
9087 | btrfs_tree_lock(eb); | |
9088 | btrfs_set_lock_blocking(eb); | |
bd681513 | 9089 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9090 | } |
01d58472 | 9091 | clean_tree_block(trans, root->fs_info, eb); |
2c47e605 YZ |
9092 | } |
9093 | ||
9094 | if (eb == root->node) { | |
9095 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9096 | parent = eb->start; | |
9097 | else | |
9098 | BUG_ON(root->root_key.objectid != | |
9099 | btrfs_header_owner(eb)); | |
9100 | } else { | |
9101 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9102 | parent = path->nodes[level + 1]->start; | |
9103 | else | |
9104 | BUG_ON(root->root_key.objectid != | |
9105 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 9106 | } |
f82d02d9 | 9107 | |
5581a51a | 9108 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
9109 | out: |
9110 | wc->refs[level] = 0; | |
9111 | wc->flags[level] = 0; | |
f0486c68 | 9112 | return 0; |
2c47e605 YZ |
9113 | } |
9114 | ||
9115 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
9116 | struct btrfs_root *root, | |
9117 | struct btrfs_path *path, | |
9118 | struct walk_control *wc) | |
9119 | { | |
2c47e605 | 9120 | int level = wc->level; |
94fcca9f | 9121 | int lookup_info = 1; |
2c47e605 YZ |
9122 | int ret; |
9123 | ||
9124 | while (level >= 0) { | |
94fcca9f | 9125 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
9126 | if (ret > 0) |
9127 | break; | |
9128 | ||
9129 | if (level == 0) | |
9130 | break; | |
9131 | ||
7a7965f8 YZ |
9132 | if (path->slots[level] >= |
9133 | btrfs_header_nritems(path->nodes[level])) | |
9134 | break; | |
9135 | ||
94fcca9f | 9136 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
9137 | if (ret > 0) { |
9138 | path->slots[level]++; | |
9139 | continue; | |
90d2c51d MX |
9140 | } else if (ret < 0) |
9141 | return ret; | |
1c4850e2 | 9142 | level = wc->level; |
f82d02d9 | 9143 | } |
f82d02d9 YZ |
9144 | return 0; |
9145 | } | |
9146 | ||
d397712b | 9147 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 9148 | struct btrfs_root *root, |
f82d02d9 | 9149 | struct btrfs_path *path, |
2c47e605 | 9150 | struct walk_control *wc, int max_level) |
20524f02 | 9151 | { |
2c47e605 | 9152 | int level = wc->level; |
20524f02 | 9153 | int ret; |
9f3a7427 | 9154 | |
2c47e605 YZ |
9155 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
9156 | while (level < max_level && path->nodes[level]) { | |
9157 | wc->level = level; | |
9158 | if (path->slots[level] + 1 < | |
9159 | btrfs_header_nritems(path->nodes[level])) { | |
9160 | path->slots[level]++; | |
20524f02 CM |
9161 | return 0; |
9162 | } else { | |
2c47e605 YZ |
9163 | ret = walk_up_proc(trans, root, path, wc); |
9164 | if (ret > 0) | |
9165 | return 0; | |
bd56b302 | 9166 | |
2c47e605 | 9167 | if (path->locks[level]) { |
bd681513 CM |
9168 | btrfs_tree_unlock_rw(path->nodes[level], |
9169 | path->locks[level]); | |
2c47e605 | 9170 | path->locks[level] = 0; |
f82d02d9 | 9171 | } |
2c47e605 YZ |
9172 | free_extent_buffer(path->nodes[level]); |
9173 | path->nodes[level] = NULL; | |
9174 | level++; | |
20524f02 CM |
9175 | } |
9176 | } | |
9177 | return 1; | |
9178 | } | |
9179 | ||
9aca1d51 | 9180 | /* |
2c47e605 YZ |
9181 | * drop a subvolume tree. |
9182 | * | |
9183 | * this function traverses the tree freeing any blocks that only | |
9184 | * referenced by the tree. | |
9185 | * | |
9186 | * when a shared tree block is found. this function decreases its | |
9187 | * reference count by one. if update_ref is true, this function | |
9188 | * also make sure backrefs for the shared block and all lower level | |
9189 | * blocks are properly updated. | |
9d1a2a3a DS |
9190 | * |
9191 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9192 | */ |
2c536799 | 9193 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9194 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9195 | int for_reloc) | |
20524f02 | 9196 | { |
5caf2a00 | 9197 | struct btrfs_path *path; |
2c47e605 YZ |
9198 | struct btrfs_trans_handle *trans; |
9199 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 9200 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9201 | struct walk_control *wc; |
9202 | struct btrfs_key key; | |
9203 | int err = 0; | |
9204 | int ret; | |
9205 | int level; | |
d29a9f62 | 9206 | bool root_dropped = false; |
20524f02 | 9207 | |
1152651a MF |
9208 | btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid); |
9209 | ||
5caf2a00 | 9210 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9211 | if (!path) { |
9212 | err = -ENOMEM; | |
9213 | goto out; | |
9214 | } | |
20524f02 | 9215 | |
2c47e605 | 9216 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9217 | if (!wc) { |
9218 | btrfs_free_path(path); | |
cb1b69f4 TI |
9219 | err = -ENOMEM; |
9220 | goto out; | |
38a1a919 | 9221 | } |
2c47e605 | 9222 | |
a22285a6 | 9223 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9224 | if (IS_ERR(trans)) { |
9225 | err = PTR_ERR(trans); | |
9226 | goto out_free; | |
9227 | } | |
98d5dc13 | 9228 | |
3fd0a558 YZ |
9229 | if (block_rsv) |
9230 | trans->block_rsv = block_rsv; | |
2c47e605 | 9231 | |
9f3a7427 | 9232 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9233 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9234 | path->nodes[level] = btrfs_lock_root_node(root); |
9235 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9236 | path->slots[level] = 0; |
bd681513 | 9237 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9238 | memset(&wc->update_progress, 0, |
9239 | sizeof(wc->update_progress)); | |
9f3a7427 | 9240 | } else { |
9f3a7427 | 9241 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9242 | memcpy(&wc->update_progress, &key, |
9243 | sizeof(wc->update_progress)); | |
9244 | ||
6702ed49 | 9245 | level = root_item->drop_level; |
2c47e605 | 9246 | BUG_ON(level == 0); |
6702ed49 | 9247 | path->lowest_level = level; |
2c47e605 YZ |
9248 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9249 | path->lowest_level = 0; | |
9250 | if (ret < 0) { | |
9251 | err = ret; | |
79787eaa | 9252 | goto out_end_trans; |
9f3a7427 | 9253 | } |
1c4850e2 | 9254 | WARN_ON(ret > 0); |
2c47e605 | 9255 | |
7d9eb12c CM |
9256 | /* |
9257 | * unlock our path, this is safe because only this | |
9258 | * function is allowed to delete this snapshot | |
9259 | */ | |
5d4f98a2 | 9260 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9261 | |
9262 | level = btrfs_header_level(root->node); | |
9263 | while (1) { | |
9264 | btrfs_tree_lock(path->nodes[level]); | |
9265 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9266 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9267 | |
9268 | ret = btrfs_lookup_extent_info(trans, root, | |
9269 | path->nodes[level]->start, | |
3173a18f | 9270 | level, 1, &wc->refs[level], |
2c47e605 | 9271 | &wc->flags[level]); |
79787eaa JM |
9272 | if (ret < 0) { |
9273 | err = ret; | |
9274 | goto out_end_trans; | |
9275 | } | |
2c47e605 YZ |
9276 | BUG_ON(wc->refs[level] == 0); |
9277 | ||
9278 | if (level == root_item->drop_level) | |
9279 | break; | |
9280 | ||
9281 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9282 | path->locks[level] = 0; |
2c47e605 YZ |
9283 | WARN_ON(wc->refs[level] != 1); |
9284 | level--; | |
9285 | } | |
9f3a7427 | 9286 | } |
2c47e605 YZ |
9287 | |
9288 | wc->level = level; | |
9289 | wc->shared_level = -1; | |
9290 | wc->stage = DROP_REFERENCE; | |
9291 | wc->update_ref = update_ref; | |
9292 | wc->keep_locks = 0; | |
66d7e7f0 | 9293 | wc->for_reloc = for_reloc; |
1c4850e2 | 9294 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 9295 | |
d397712b | 9296 | while (1) { |
9d1a2a3a | 9297 | |
2c47e605 YZ |
9298 | ret = walk_down_tree(trans, root, path, wc); |
9299 | if (ret < 0) { | |
9300 | err = ret; | |
20524f02 | 9301 | break; |
2c47e605 | 9302 | } |
9aca1d51 | 9303 | |
2c47e605 YZ |
9304 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9305 | if (ret < 0) { | |
9306 | err = ret; | |
20524f02 | 9307 | break; |
2c47e605 YZ |
9308 | } |
9309 | ||
9310 | if (ret > 0) { | |
9311 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9312 | break; |
9313 | } | |
2c47e605 YZ |
9314 | |
9315 | if (wc->stage == DROP_REFERENCE) { | |
9316 | level = wc->level; | |
9317 | btrfs_node_key(path->nodes[level], | |
9318 | &root_item->drop_progress, | |
9319 | path->slots[level]); | |
9320 | root_item->drop_level = level; | |
9321 | } | |
9322 | ||
9323 | BUG_ON(wc->level == 0); | |
3c8f2422 JB |
9324 | if (btrfs_should_end_transaction(trans, tree_root) || |
9325 | (!for_reloc && btrfs_need_cleaner_sleep(root))) { | |
2c47e605 YZ |
9326 | ret = btrfs_update_root(trans, tree_root, |
9327 | &root->root_key, | |
9328 | root_item); | |
79787eaa JM |
9329 | if (ret) { |
9330 | btrfs_abort_transaction(trans, tree_root, ret); | |
9331 | err = ret; | |
9332 | goto out_end_trans; | |
9333 | } | |
2c47e605 | 9334 | |
3fd0a558 | 9335 | btrfs_end_transaction_throttle(trans, tree_root); |
3c8f2422 | 9336 | if (!for_reloc && btrfs_need_cleaner_sleep(root)) { |
efe120a0 | 9337 | pr_debug("BTRFS: drop snapshot early exit\n"); |
3c8f2422 JB |
9338 | err = -EAGAIN; |
9339 | goto out_free; | |
9340 | } | |
9341 | ||
a22285a6 | 9342 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9343 | if (IS_ERR(trans)) { |
9344 | err = PTR_ERR(trans); | |
9345 | goto out_free; | |
9346 | } | |
3fd0a558 YZ |
9347 | if (block_rsv) |
9348 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9349 | } |
20524f02 | 9350 | } |
b3b4aa74 | 9351 | btrfs_release_path(path); |
79787eaa JM |
9352 | if (err) |
9353 | goto out_end_trans; | |
2c47e605 YZ |
9354 | |
9355 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
9356 | if (ret) { |
9357 | btrfs_abort_transaction(trans, tree_root, ret); | |
9358 | goto out_end_trans; | |
9359 | } | |
2c47e605 | 9360 | |
76dda93c | 9361 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9362 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9363 | NULL, NULL); | |
79787eaa JM |
9364 | if (ret < 0) { |
9365 | btrfs_abort_transaction(trans, tree_root, ret); | |
9366 | err = ret; | |
9367 | goto out_end_trans; | |
9368 | } else if (ret > 0) { | |
84cd948c JB |
9369 | /* if we fail to delete the orphan item this time |
9370 | * around, it'll get picked up the next time. | |
9371 | * | |
9372 | * The most common failure here is just -ENOENT. | |
9373 | */ | |
9374 | btrfs_del_orphan_item(trans, tree_root, | |
9375 | root->root_key.objectid); | |
76dda93c YZ |
9376 | } |
9377 | } | |
9378 | ||
27cdeb70 | 9379 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9380 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9381 | } else { |
9382 | free_extent_buffer(root->node); | |
9383 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9384 | btrfs_put_fs_root(root); |
76dda93c | 9385 | } |
d29a9f62 | 9386 | root_dropped = true; |
79787eaa | 9387 | out_end_trans: |
3fd0a558 | 9388 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 9389 | out_free: |
2c47e605 | 9390 | kfree(wc); |
5caf2a00 | 9391 | btrfs_free_path(path); |
cb1b69f4 | 9392 | out: |
d29a9f62 JB |
9393 | /* |
9394 | * So if we need to stop dropping the snapshot for whatever reason we | |
9395 | * need to make sure to add it back to the dead root list so that we | |
9396 | * keep trying to do the work later. This also cleans up roots if we | |
9397 | * don't have it in the radix (like when we recover after a power fail | |
9398 | * or unmount) so we don't leak memory. | |
9399 | */ | |
b37b39cd | 9400 | if (!for_reloc && root_dropped == false) |
d29a9f62 | 9401 | btrfs_add_dead_root(root); |
90515e7f | 9402 | if (err && err != -EAGAIN) |
34d97007 | 9403 | btrfs_handle_fs_error(root->fs_info, err, NULL); |
2c536799 | 9404 | return err; |
20524f02 | 9405 | } |
9078a3e1 | 9406 | |
2c47e605 YZ |
9407 | /* |
9408 | * drop subtree rooted at tree block 'node'. | |
9409 | * | |
9410 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9411 | * only used by relocation code |
2c47e605 | 9412 | */ |
f82d02d9 YZ |
9413 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9414 | struct btrfs_root *root, | |
9415 | struct extent_buffer *node, | |
9416 | struct extent_buffer *parent) | |
9417 | { | |
9418 | struct btrfs_path *path; | |
2c47e605 | 9419 | struct walk_control *wc; |
f82d02d9 YZ |
9420 | int level; |
9421 | int parent_level; | |
9422 | int ret = 0; | |
9423 | int wret; | |
9424 | ||
2c47e605 YZ |
9425 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9426 | ||
f82d02d9 | 9427 | path = btrfs_alloc_path(); |
db5b493a TI |
9428 | if (!path) |
9429 | return -ENOMEM; | |
f82d02d9 | 9430 | |
2c47e605 | 9431 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9432 | if (!wc) { |
9433 | btrfs_free_path(path); | |
9434 | return -ENOMEM; | |
9435 | } | |
2c47e605 | 9436 | |
b9447ef8 | 9437 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9438 | parent_level = btrfs_header_level(parent); |
9439 | extent_buffer_get(parent); | |
9440 | path->nodes[parent_level] = parent; | |
9441 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9442 | ||
b9447ef8 | 9443 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9444 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9445 | path->nodes[level] = node; |
9446 | path->slots[level] = 0; | |
bd681513 | 9447 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9448 | |
9449 | wc->refs[parent_level] = 1; | |
9450 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9451 | wc->level = level; | |
9452 | wc->shared_level = -1; | |
9453 | wc->stage = DROP_REFERENCE; | |
9454 | wc->update_ref = 0; | |
9455 | wc->keep_locks = 1; | |
66d7e7f0 | 9456 | wc->for_reloc = 1; |
1c4850e2 | 9457 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
9458 | |
9459 | while (1) { | |
2c47e605 YZ |
9460 | wret = walk_down_tree(trans, root, path, wc); |
9461 | if (wret < 0) { | |
f82d02d9 | 9462 | ret = wret; |
f82d02d9 | 9463 | break; |
2c47e605 | 9464 | } |
f82d02d9 | 9465 | |
2c47e605 | 9466 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9467 | if (wret < 0) |
9468 | ret = wret; | |
9469 | if (wret != 0) | |
9470 | break; | |
9471 | } | |
9472 | ||
2c47e605 | 9473 | kfree(wc); |
f82d02d9 YZ |
9474 | btrfs_free_path(path); |
9475 | return ret; | |
9476 | } | |
9477 | ||
ec44a35c CM |
9478 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
9479 | { | |
9480 | u64 num_devices; | |
fc67c450 | 9481 | u64 stripped; |
e4d8ec0f | 9482 | |
fc67c450 ID |
9483 | /* |
9484 | * if restripe for this chunk_type is on pick target profile and | |
9485 | * return, otherwise do the usual balance | |
9486 | */ | |
9487 | stripped = get_restripe_target(root->fs_info, flags); | |
9488 | if (stripped) | |
9489 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9490 | |
95669976 | 9491 | num_devices = root->fs_info->fs_devices->rw_devices; |
cd02dca5 | 9492 | |
fc67c450 | 9493 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9494 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9495 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9496 | ||
ec44a35c CM |
9497 | if (num_devices == 1) { |
9498 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9499 | stripped = flags & ~stripped; | |
9500 | ||
9501 | /* turn raid0 into single device chunks */ | |
9502 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9503 | return stripped; | |
9504 | ||
9505 | /* turn mirroring into duplication */ | |
9506 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9507 | BTRFS_BLOCK_GROUP_RAID10)) | |
9508 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9509 | } else { |
9510 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9511 | if (flags & stripped) |
9512 | return flags; | |
9513 | ||
9514 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9515 | stripped = flags & ~stripped; | |
9516 | ||
9517 | /* switch duplicated blocks with raid1 */ | |
9518 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9519 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9520 | ||
e3176ca2 | 9521 | /* this is drive concat, leave it alone */ |
ec44a35c | 9522 | } |
e3176ca2 | 9523 | |
ec44a35c CM |
9524 | return flags; |
9525 | } | |
9526 | ||
868f401a | 9527 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9528 | { |
f0486c68 YZ |
9529 | struct btrfs_space_info *sinfo = cache->space_info; |
9530 | u64 num_bytes; | |
199c36ea | 9531 | u64 min_allocable_bytes; |
f0486c68 | 9532 | int ret = -ENOSPC; |
0ef3e66b | 9533 | |
199c36ea MX |
9534 | /* |
9535 | * We need some metadata space and system metadata space for | |
9536 | * allocating chunks in some corner cases until we force to set | |
9537 | * it to be readonly. | |
9538 | */ | |
9539 | if ((sinfo->flags & | |
9540 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9541 | !force) | |
ee22184b | 9542 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9543 | else |
9544 | min_allocable_bytes = 0; | |
9545 | ||
f0486c68 YZ |
9546 | spin_lock(&sinfo->lock); |
9547 | spin_lock(&cache->lock); | |
61cfea9b W |
9548 | |
9549 | if (cache->ro) { | |
868f401a | 9550 | cache->ro++; |
61cfea9b W |
9551 | ret = 0; |
9552 | goto out; | |
9553 | } | |
9554 | ||
f0486c68 YZ |
9555 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9556 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9557 | ||
9558 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
9559 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
9560 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 9561 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9562 | cache->ro++; |
633c0aad | 9563 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9564 | ret = 0; |
9565 | } | |
61cfea9b | 9566 | out: |
f0486c68 YZ |
9567 | spin_unlock(&cache->lock); |
9568 | spin_unlock(&sinfo->lock); | |
9569 | return ret; | |
9570 | } | |
7d9eb12c | 9571 | |
868f401a | 9572 | int btrfs_inc_block_group_ro(struct btrfs_root *root, |
f0486c68 | 9573 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9574 | |
f0486c68 YZ |
9575 | { |
9576 | struct btrfs_trans_handle *trans; | |
9577 | u64 alloc_flags; | |
9578 | int ret; | |
7d9eb12c | 9579 | |
1bbc621e | 9580 | again: |
ff5714cc | 9581 | trans = btrfs_join_transaction(root); |
79787eaa JM |
9582 | if (IS_ERR(trans)) |
9583 | return PTR_ERR(trans); | |
5d4f98a2 | 9584 | |
1bbc621e CM |
9585 | /* |
9586 | * we're not allowed to set block groups readonly after the dirty | |
9587 | * block groups cache has started writing. If it already started, | |
9588 | * back off and let this transaction commit | |
9589 | */ | |
9590 | mutex_lock(&root->fs_info->ro_block_group_mutex); | |
3204d33c | 9591 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9592 | u64 transid = trans->transid; |
9593 | ||
9594 | mutex_unlock(&root->fs_info->ro_block_group_mutex); | |
9595 | btrfs_end_transaction(trans, root); | |
9596 | ||
9597 | ret = btrfs_wait_for_commit(root, transid); | |
9598 | if (ret) | |
9599 | return ret; | |
9600 | goto again; | |
9601 | } | |
9602 | ||
153c35b6 CM |
9603 | /* |
9604 | * if we are changing raid levels, try to allocate a corresponding | |
9605 | * block group with the new raid level. | |
9606 | */ | |
9607 | alloc_flags = update_block_group_flags(root, cache->flags); | |
9608 | if (alloc_flags != cache->flags) { | |
9609 | ret = do_chunk_alloc(trans, root, alloc_flags, | |
9610 | CHUNK_ALLOC_FORCE); | |
9611 | /* | |
9612 | * ENOSPC is allowed here, we may have enough space | |
9613 | * already allocated at the new raid level to | |
9614 | * carry on | |
9615 | */ | |
9616 | if (ret == -ENOSPC) | |
9617 | ret = 0; | |
9618 | if (ret < 0) | |
9619 | goto out; | |
9620 | } | |
1bbc621e | 9621 | |
868f401a | 9622 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9623 | if (!ret) |
9624 | goto out; | |
9625 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 9626 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 9627 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9628 | if (ret < 0) |
9629 | goto out; | |
868f401a | 9630 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9631 | out: |
2f081088 SL |
9632 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
9633 | alloc_flags = update_block_group_flags(root, cache->flags); | |
a9629596 | 9634 | lock_chunks(root->fs_info->chunk_root); |
4617ea3a | 9635 | check_system_chunk(trans, root, alloc_flags); |
a9629596 | 9636 | unlock_chunks(root->fs_info->chunk_root); |
2f081088 | 9637 | } |
1bbc621e | 9638 | mutex_unlock(&root->fs_info->ro_block_group_mutex); |
2f081088 | 9639 | |
f0486c68 YZ |
9640 | btrfs_end_transaction(trans, root); |
9641 | return ret; | |
9642 | } | |
5d4f98a2 | 9643 | |
c87f08ca CM |
9644 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
9645 | struct btrfs_root *root, u64 type) | |
9646 | { | |
9647 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 9648 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 9649 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9650 | } |
9651 | ||
6d07bcec MX |
9652 | /* |
9653 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9654 | * space_info. takes mirrors into account. |
6d07bcec | 9655 | */ |
633c0aad | 9656 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9657 | { |
9658 | struct btrfs_block_group_cache *block_group; | |
9659 | u64 free_bytes = 0; | |
9660 | int factor; | |
9661 | ||
01327610 | 9662 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9663 | if (list_empty(&sinfo->ro_bgs)) |
9664 | return 0; | |
9665 | ||
9666 | spin_lock(&sinfo->lock); | |
9667 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9668 | spin_lock(&block_group->lock); |
9669 | ||
9670 | if (!block_group->ro) { | |
9671 | spin_unlock(&block_group->lock); | |
9672 | continue; | |
9673 | } | |
9674 | ||
9675 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9676 | BTRFS_BLOCK_GROUP_RAID10 | | |
9677 | BTRFS_BLOCK_GROUP_DUP)) | |
9678 | factor = 2; | |
9679 | else | |
9680 | factor = 1; | |
9681 | ||
9682 | free_bytes += (block_group->key.offset - | |
9683 | btrfs_block_group_used(&block_group->item)) * | |
9684 | factor; | |
9685 | ||
9686 | spin_unlock(&block_group->lock); | |
9687 | } | |
6d07bcec MX |
9688 | spin_unlock(&sinfo->lock); |
9689 | ||
9690 | return free_bytes; | |
9691 | } | |
9692 | ||
868f401a | 9693 | void btrfs_dec_block_group_ro(struct btrfs_root *root, |
f0486c68 | 9694 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9695 | { |
f0486c68 YZ |
9696 | struct btrfs_space_info *sinfo = cache->space_info; |
9697 | u64 num_bytes; | |
9698 | ||
9699 | BUG_ON(!cache->ro); | |
9700 | ||
9701 | spin_lock(&sinfo->lock); | |
9702 | spin_lock(&cache->lock); | |
868f401a Z |
9703 | if (!--cache->ro) { |
9704 | num_bytes = cache->key.offset - cache->reserved - | |
9705 | cache->pinned - cache->bytes_super - | |
9706 | btrfs_block_group_used(&cache->item); | |
9707 | sinfo->bytes_readonly -= num_bytes; | |
9708 | list_del_init(&cache->ro_list); | |
9709 | } | |
f0486c68 YZ |
9710 | spin_unlock(&cache->lock); |
9711 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9712 | } |
9713 | ||
ba1bf481 JB |
9714 | /* |
9715 | * checks to see if its even possible to relocate this block group. | |
9716 | * | |
9717 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9718 | * ok to go ahead and try. | |
9719 | */ | |
9720 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 9721 | { |
ba1bf481 JB |
9722 | struct btrfs_block_group_cache *block_group; |
9723 | struct btrfs_space_info *space_info; | |
9724 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
9725 | struct btrfs_device *device; | |
6df9a95e | 9726 | struct btrfs_trans_handle *trans; |
cdcb725c | 9727 | u64 min_free; |
6719db6a JB |
9728 | u64 dev_min = 1; |
9729 | u64 dev_nr = 0; | |
4a5e98f5 | 9730 | u64 target; |
0305bc27 | 9731 | int debug; |
cdcb725c | 9732 | int index; |
ba1bf481 JB |
9733 | int full = 0; |
9734 | int ret = 0; | |
1a40e23b | 9735 | |
0305bc27 QW |
9736 | debug = btrfs_test_opt(root, ENOSPC_DEBUG); |
9737 | ||
ba1bf481 | 9738 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 9739 | |
ba1bf481 | 9740 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9741 | if (!block_group) { |
9742 | if (debug) | |
9743 | btrfs_warn(root->fs_info, | |
9744 | "can't find block group for bytenr %llu", | |
9745 | bytenr); | |
ba1bf481 | 9746 | return -1; |
0305bc27 | 9747 | } |
1a40e23b | 9748 | |
cdcb725c | 9749 | min_free = btrfs_block_group_used(&block_group->item); |
9750 | ||
ba1bf481 | 9751 | /* no bytes used, we're good */ |
cdcb725c | 9752 | if (!min_free) |
1a40e23b ZY |
9753 | goto out; |
9754 | ||
ba1bf481 JB |
9755 | space_info = block_group->space_info; |
9756 | spin_lock(&space_info->lock); | |
17d217fe | 9757 | |
ba1bf481 | 9758 | full = space_info->full; |
17d217fe | 9759 | |
ba1bf481 JB |
9760 | /* |
9761 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9762 | * relocate it unless we're able to allocate a new chunk below. |
9763 | * | |
9764 | * Otherwise, we need to make sure we have room in the space to handle | |
9765 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9766 | */ |
7ce618db | 9767 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 9768 | (space_info->bytes_used + space_info->bytes_reserved + |
9769 | space_info->bytes_pinned + space_info->bytes_readonly + | |
9770 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
9771 | spin_unlock(&space_info->lock); |
9772 | goto out; | |
17d217fe | 9773 | } |
ba1bf481 | 9774 | spin_unlock(&space_info->lock); |
ea8c2819 | 9775 | |
ba1bf481 JB |
9776 | /* |
9777 | * ok we don't have enough space, but maybe we have free space on our | |
9778 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9779 | * alloc devices and guess if we have enough space. if this block |
9780 | * group is going to be restriped, run checks against the target | |
9781 | * profile instead of the current one. | |
ba1bf481 JB |
9782 | */ |
9783 | ret = -1; | |
ea8c2819 | 9784 | |
cdcb725c | 9785 | /* |
9786 | * index: | |
9787 | * 0: raid10 | |
9788 | * 1: raid1 | |
9789 | * 2: dup | |
9790 | * 3: raid0 | |
9791 | * 4: single | |
9792 | */ | |
4a5e98f5 ID |
9793 | target = get_restripe_target(root->fs_info, block_group->flags); |
9794 | if (target) { | |
31e50229 | 9795 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9796 | } else { |
9797 | /* | |
9798 | * this is just a balance, so if we were marked as full | |
9799 | * we know there is no space for a new chunk | |
9800 | */ | |
0305bc27 QW |
9801 | if (full) { |
9802 | if (debug) | |
9803 | btrfs_warn(root->fs_info, | |
9804 | "no space to alloc new chunk for block group %llu", | |
9805 | block_group->key.objectid); | |
4a5e98f5 | 9806 | goto out; |
0305bc27 | 9807 | } |
4a5e98f5 ID |
9808 | |
9809 | index = get_block_group_index(block_group); | |
9810 | } | |
9811 | ||
e6ec716f | 9812 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9813 | dev_min = 4; |
6719db6a JB |
9814 | /* Divide by 2 */ |
9815 | min_free >>= 1; | |
e6ec716f | 9816 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9817 | dev_min = 2; |
e6ec716f | 9818 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9819 | /* Multiply by 2 */ |
9820 | min_free <<= 1; | |
e6ec716f | 9821 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9822 | dev_min = fs_devices->rw_devices; |
47c5713f | 9823 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9824 | } |
9825 | ||
6df9a95e JB |
9826 | /* We need to do this so that we can look at pending chunks */ |
9827 | trans = btrfs_join_transaction(root); | |
9828 | if (IS_ERR(trans)) { | |
9829 | ret = PTR_ERR(trans); | |
9830 | goto out; | |
9831 | } | |
9832 | ||
ba1bf481 JB |
9833 | mutex_lock(&root->fs_info->chunk_mutex); |
9834 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 9835 | u64 dev_offset; |
56bec294 | 9836 | |
ba1bf481 JB |
9837 | /* |
9838 | * check to make sure we can actually find a chunk with enough | |
9839 | * space to fit our block group in. | |
9840 | */ | |
63a212ab SB |
9841 | if (device->total_bytes > device->bytes_used + min_free && |
9842 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 9843 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9844 | &dev_offset, NULL); |
ba1bf481 | 9845 | if (!ret) |
cdcb725c | 9846 | dev_nr++; |
9847 | ||
9848 | if (dev_nr >= dev_min) | |
73e48b27 | 9849 | break; |
cdcb725c | 9850 | |
ba1bf481 | 9851 | ret = -1; |
725c8463 | 9852 | } |
edbd8d4e | 9853 | } |
0305bc27 QW |
9854 | if (debug && ret == -1) |
9855 | btrfs_warn(root->fs_info, | |
9856 | "no space to allocate a new chunk for block group %llu", | |
9857 | block_group->key.objectid); | |
ba1bf481 | 9858 | mutex_unlock(&root->fs_info->chunk_mutex); |
6df9a95e | 9859 | btrfs_end_transaction(trans, root); |
edbd8d4e | 9860 | out: |
ba1bf481 | 9861 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9862 | return ret; |
9863 | } | |
9864 | ||
b2950863 CH |
9865 | static int find_first_block_group(struct btrfs_root *root, |
9866 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 9867 | { |
925baedd | 9868 | int ret = 0; |
0b86a832 CM |
9869 | struct btrfs_key found_key; |
9870 | struct extent_buffer *leaf; | |
9871 | int slot; | |
edbd8d4e | 9872 | |
0b86a832 CM |
9873 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9874 | if (ret < 0) | |
925baedd CM |
9875 | goto out; |
9876 | ||
d397712b | 9877 | while (1) { |
0b86a832 | 9878 | slot = path->slots[0]; |
edbd8d4e | 9879 | leaf = path->nodes[0]; |
0b86a832 CM |
9880 | if (slot >= btrfs_header_nritems(leaf)) { |
9881 | ret = btrfs_next_leaf(root, path); | |
9882 | if (ret == 0) | |
9883 | continue; | |
9884 | if (ret < 0) | |
925baedd | 9885 | goto out; |
0b86a832 | 9886 | break; |
edbd8d4e | 9887 | } |
0b86a832 | 9888 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9889 | |
0b86a832 | 9890 | if (found_key.objectid >= key->objectid && |
925baedd CM |
9891 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
9892 | ret = 0; | |
9893 | goto out; | |
9894 | } | |
0b86a832 | 9895 | path->slots[0]++; |
edbd8d4e | 9896 | } |
925baedd | 9897 | out: |
0b86a832 | 9898 | return ret; |
edbd8d4e CM |
9899 | } |
9900 | ||
0af3d00b JB |
9901 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9902 | { | |
9903 | struct btrfs_block_group_cache *block_group; | |
9904 | u64 last = 0; | |
9905 | ||
9906 | while (1) { | |
9907 | struct inode *inode; | |
9908 | ||
9909 | block_group = btrfs_lookup_first_block_group(info, last); | |
9910 | while (block_group) { | |
9911 | spin_lock(&block_group->lock); | |
9912 | if (block_group->iref) | |
9913 | break; | |
9914 | spin_unlock(&block_group->lock); | |
9915 | block_group = next_block_group(info->tree_root, | |
9916 | block_group); | |
9917 | } | |
9918 | if (!block_group) { | |
9919 | if (last == 0) | |
9920 | break; | |
9921 | last = 0; | |
9922 | continue; | |
9923 | } | |
9924 | ||
9925 | inode = block_group->inode; | |
9926 | block_group->iref = 0; | |
9927 | block_group->inode = NULL; | |
9928 | spin_unlock(&block_group->lock); | |
9929 | iput(inode); | |
9930 | last = block_group->key.objectid + block_group->key.offset; | |
9931 | btrfs_put_block_group(block_group); | |
9932 | } | |
9933 | } | |
9934 | ||
1a40e23b ZY |
9935 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9936 | { | |
9937 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9938 | struct btrfs_space_info *space_info; |
11833d66 | 9939 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9940 | struct rb_node *n; |
9941 | ||
9e351cc8 | 9942 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9943 | while (!list_empty(&info->caching_block_groups)) { |
9944 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9945 | struct btrfs_caching_control, list); | |
9946 | list_del(&caching_ctl->list); | |
9947 | put_caching_control(caching_ctl); | |
9948 | } | |
9e351cc8 | 9949 | up_write(&info->commit_root_sem); |
11833d66 | 9950 | |
47ab2a6c JB |
9951 | spin_lock(&info->unused_bgs_lock); |
9952 | while (!list_empty(&info->unused_bgs)) { | |
9953 | block_group = list_first_entry(&info->unused_bgs, | |
9954 | struct btrfs_block_group_cache, | |
9955 | bg_list); | |
9956 | list_del_init(&block_group->bg_list); | |
9957 | btrfs_put_block_group(block_group); | |
9958 | } | |
9959 | spin_unlock(&info->unused_bgs_lock); | |
9960 | ||
1a40e23b ZY |
9961 | spin_lock(&info->block_group_cache_lock); |
9962 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9963 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9964 | cache_node); | |
1a40e23b ZY |
9965 | rb_erase(&block_group->cache_node, |
9966 | &info->block_group_cache_tree); | |
01eacb27 | 9967 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9968 | spin_unlock(&info->block_group_cache_lock); |
9969 | ||
80eb234a | 9970 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9971 | list_del(&block_group->list); |
80eb234a | 9972 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9973 | |
817d52f8 | 9974 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 9975 | wait_block_group_cache_done(block_group); |
817d52f8 | 9976 | |
3c14874a JB |
9977 | /* |
9978 | * We haven't cached this block group, which means we could | |
9979 | * possibly have excluded extents on this block group. | |
9980 | */ | |
36cce922 JB |
9981 | if (block_group->cached == BTRFS_CACHE_NO || |
9982 | block_group->cached == BTRFS_CACHE_ERROR) | |
3c14874a JB |
9983 | free_excluded_extents(info->extent_root, block_group); |
9984 | ||
817d52f8 | 9985 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 9986 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9987 | |
9988 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9989 | } |
9990 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9991 | |
9992 | /* now that all the block groups are freed, go through and | |
9993 | * free all the space_info structs. This is only called during | |
9994 | * the final stages of unmount, and so we know nobody is | |
9995 | * using them. We call synchronize_rcu() once before we start, | |
9996 | * just to be on the safe side. | |
9997 | */ | |
9998 | synchronize_rcu(); | |
9999 | ||
8929ecfa YZ |
10000 | release_global_block_rsv(info); |
10001 | ||
67871254 | 10002 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
10003 | int i; |
10004 | ||
4184ea7f CM |
10005 | space_info = list_entry(info->space_info.next, |
10006 | struct btrfs_space_info, | |
10007 | list); | |
d555b6c3 JB |
10008 | |
10009 | /* | |
10010 | * Do not hide this behind enospc_debug, this is actually | |
10011 | * important and indicates a real bug if this happens. | |
10012 | */ | |
10013 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 10014 | space_info->bytes_reserved > 0 || |
d555b6c3 JB |
10015 | space_info->bytes_may_use > 0)) |
10016 | dump_space_info(space_info, 0, 0); | |
4184ea7f | 10017 | list_del(&space_info->list); |
6ab0a202 JM |
10018 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
10019 | struct kobject *kobj; | |
c1895442 JM |
10020 | kobj = space_info->block_group_kobjs[i]; |
10021 | space_info->block_group_kobjs[i] = NULL; | |
10022 | if (kobj) { | |
6ab0a202 JM |
10023 | kobject_del(kobj); |
10024 | kobject_put(kobj); | |
10025 | } | |
10026 | } | |
10027 | kobject_del(&space_info->kobj); | |
10028 | kobject_put(&space_info->kobj); | |
4184ea7f | 10029 | } |
1a40e23b ZY |
10030 | return 0; |
10031 | } | |
10032 | ||
b742bb82 YZ |
10033 | static void __link_block_group(struct btrfs_space_info *space_info, |
10034 | struct btrfs_block_group_cache *cache) | |
10035 | { | |
10036 | int index = get_block_group_index(cache); | |
ed55b6ac | 10037 | bool first = false; |
b742bb82 YZ |
10038 | |
10039 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
10040 | if (list_empty(&space_info->block_groups[index])) |
10041 | first = true; | |
10042 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
10043 | up_write(&space_info->groups_sem); | |
10044 | ||
10045 | if (first) { | |
c1895442 | 10046 | struct raid_kobject *rkobj; |
6ab0a202 JM |
10047 | int ret; |
10048 | ||
c1895442 JM |
10049 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
10050 | if (!rkobj) | |
10051 | goto out_err; | |
10052 | rkobj->raid_type = index; | |
10053 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
10054 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
10055 | "%s", get_raid_name(index)); | |
6ab0a202 | 10056 | if (ret) { |
c1895442 JM |
10057 | kobject_put(&rkobj->kobj); |
10058 | goto out_err; | |
6ab0a202 | 10059 | } |
c1895442 | 10060 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 10061 | } |
c1895442 JM |
10062 | |
10063 | return; | |
10064 | out_err: | |
10065 | pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n"); | |
b742bb82 YZ |
10066 | } |
10067 | ||
920e4a58 MX |
10068 | static struct btrfs_block_group_cache * |
10069 | btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size) | |
10070 | { | |
10071 | struct btrfs_block_group_cache *cache; | |
10072 | ||
10073 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
10074 | if (!cache) | |
10075 | return NULL; | |
10076 | ||
10077 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
10078 | GFP_NOFS); | |
10079 | if (!cache->free_space_ctl) { | |
10080 | kfree(cache); | |
10081 | return NULL; | |
10082 | } | |
10083 | ||
10084 | cache->key.objectid = start; | |
10085 | cache->key.offset = size; | |
10086 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
10087 | ||
10088 | cache->sectorsize = root->sectorsize; | |
10089 | cache->fs_info = root->fs_info; | |
10090 | cache->full_stripe_len = btrfs_full_stripe_len(root, | |
10091 | &root->fs_info->mapping_tree, | |
10092 | start); | |
1e144fb8 OS |
10093 | set_free_space_tree_thresholds(cache); |
10094 | ||
920e4a58 MX |
10095 | atomic_set(&cache->count, 1); |
10096 | spin_lock_init(&cache->lock); | |
e570fd27 | 10097 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
10098 | INIT_LIST_HEAD(&cache->list); |
10099 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 10100 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 10101 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 10102 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 10103 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 10104 | btrfs_init_free_space_ctl(cache); |
04216820 | 10105 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 10106 | mutex_init(&cache->free_space_lock); |
920e4a58 MX |
10107 | |
10108 | return cache; | |
10109 | } | |
10110 | ||
9078a3e1 CM |
10111 | int btrfs_read_block_groups(struct btrfs_root *root) |
10112 | { | |
10113 | struct btrfs_path *path; | |
10114 | int ret; | |
9078a3e1 | 10115 | struct btrfs_block_group_cache *cache; |
be744175 | 10116 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 10117 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
10118 | struct btrfs_key key; |
10119 | struct btrfs_key found_key; | |
5f39d397 | 10120 | struct extent_buffer *leaf; |
0af3d00b JB |
10121 | int need_clear = 0; |
10122 | u64 cache_gen; | |
96b5179d | 10123 | |
be744175 | 10124 | root = info->extent_root; |
9078a3e1 | 10125 | key.objectid = 0; |
0b86a832 | 10126 | key.offset = 0; |
962a298f | 10127 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
10128 | path = btrfs_alloc_path(); |
10129 | if (!path) | |
10130 | return -ENOMEM; | |
e4058b54 | 10131 | path->reada = READA_FORWARD; |
9078a3e1 | 10132 | |
6c41761f | 10133 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 10134 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 10135 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 10136 | need_clear = 1; |
88c2ba3b JB |
10137 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
10138 | need_clear = 1; | |
0af3d00b | 10139 | |
d397712b | 10140 | while (1) { |
0b86a832 | 10141 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
10142 | if (ret > 0) |
10143 | break; | |
0b86a832 CM |
10144 | if (ret != 0) |
10145 | goto error; | |
920e4a58 | 10146 | |
5f39d397 CM |
10147 | leaf = path->nodes[0]; |
10148 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 MX |
10149 | |
10150 | cache = btrfs_create_block_group_cache(root, found_key.objectid, | |
10151 | found_key.offset); | |
9078a3e1 | 10152 | if (!cache) { |
0b86a832 | 10153 | ret = -ENOMEM; |
f0486c68 | 10154 | goto error; |
9078a3e1 | 10155 | } |
96303081 | 10156 | |
cf7c1ef6 LB |
10157 | if (need_clear) { |
10158 | /* | |
10159 | * When we mount with old space cache, we need to | |
10160 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10161 | * | |
10162 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10163 | * truncate the old free space cache inode and | |
10164 | * setup a new one. | |
10165 | * b) Setting 'dirty flag' makes sure that we flush | |
10166 | * the new space cache info onto disk. | |
10167 | */ | |
cf7c1ef6 | 10168 | if (btrfs_test_opt(root, SPACE_CACHE)) |
ce93ec54 | 10169 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10170 | } |
0af3d00b | 10171 | |
5f39d397 CM |
10172 | read_extent_buffer(leaf, &cache->item, |
10173 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10174 | sizeof(cache->item)); | |
920e4a58 | 10175 | cache->flags = btrfs_block_group_flags(&cache->item); |
0b86a832 | 10176 | |
9078a3e1 | 10177 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10178 | btrfs_release_path(path); |
34d52cb6 | 10179 | |
3c14874a JB |
10180 | /* |
10181 | * We need to exclude the super stripes now so that the space | |
10182 | * info has super bytes accounted for, otherwise we'll think | |
10183 | * we have more space than we actually do. | |
10184 | */ | |
835d974f JB |
10185 | ret = exclude_super_stripes(root, cache); |
10186 | if (ret) { | |
10187 | /* | |
10188 | * We may have excluded something, so call this just in | |
10189 | * case. | |
10190 | */ | |
10191 | free_excluded_extents(root, cache); | |
920e4a58 | 10192 | btrfs_put_block_group(cache); |
835d974f JB |
10193 | goto error; |
10194 | } | |
3c14874a | 10195 | |
817d52f8 JB |
10196 | /* |
10197 | * check for two cases, either we are full, and therefore | |
10198 | * don't need to bother with the caching work since we won't | |
10199 | * find any space, or we are empty, and we can just add all | |
10200 | * the space in and be done with it. This saves us _alot_ of | |
10201 | * time, particularly in the full case. | |
10202 | */ | |
10203 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10204 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10205 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 10206 | free_excluded_extents(root, cache); |
817d52f8 | 10207 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10208 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
10209 | cache->cached = BTRFS_CACHE_FINISHED; |
10210 | add_new_free_space(cache, root->fs_info, | |
10211 | found_key.objectid, | |
10212 | found_key.objectid + | |
10213 | found_key.offset); | |
11833d66 | 10214 | free_excluded_extents(root, cache); |
817d52f8 | 10215 | } |
96b5179d | 10216 | |
8c579fe7 JB |
10217 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
10218 | if (ret) { | |
10219 | btrfs_remove_free_space_cache(cache); | |
10220 | btrfs_put_block_group(cache); | |
10221 | goto error; | |
10222 | } | |
10223 | ||
c83f8eff | 10224 | trace_btrfs_add_block_group(root->fs_info, cache, 0); |
6324fbf3 CM |
10225 | ret = update_space_info(info, cache->flags, found_key.offset, |
10226 | btrfs_block_group_used(&cache->item), | |
e40edf2d | 10227 | cache->bytes_super, &space_info); |
8c579fe7 JB |
10228 | if (ret) { |
10229 | btrfs_remove_free_space_cache(cache); | |
10230 | spin_lock(&info->block_group_cache_lock); | |
10231 | rb_erase(&cache->cache_node, | |
10232 | &info->block_group_cache_tree); | |
01eacb27 | 10233 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
10234 | spin_unlock(&info->block_group_cache_lock); |
10235 | btrfs_put_block_group(cache); | |
10236 | goto error; | |
10237 | } | |
10238 | ||
6324fbf3 | 10239 | cache->space_info = space_info; |
1b2da372 | 10240 | |
b742bb82 | 10241 | __link_block_group(space_info, cache); |
0f9dd46c | 10242 | |
75ccf47d | 10243 | set_avail_alloc_bits(root->fs_info, cache->flags); |
47ab2a6c | 10244 | if (btrfs_chunk_readonly(root, cache->key.objectid)) { |
868f401a | 10245 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10246 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10247 | spin_lock(&info->unused_bgs_lock); | |
10248 | /* Should always be true but just in case. */ | |
10249 | if (list_empty(&cache->bg_list)) { | |
10250 | btrfs_get_block_group(cache); | |
10251 | list_add_tail(&cache->bg_list, | |
10252 | &info->unused_bgs); | |
10253 | } | |
10254 | spin_unlock(&info->unused_bgs_lock); | |
10255 | } | |
9078a3e1 | 10256 | } |
b742bb82 YZ |
10257 | |
10258 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
10259 | if (!(get_alloc_profile(root, space_info->flags) & | |
10260 | (BTRFS_BLOCK_GROUP_RAID10 | | |
10261 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10262 | BTRFS_BLOCK_GROUP_RAID5 | |
10263 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10264 | BTRFS_BLOCK_GROUP_DUP))) |
10265 | continue; | |
10266 | /* | |
10267 | * avoid allocating from un-mirrored block group if there are | |
10268 | * mirrored block groups. | |
10269 | */ | |
1095cc0d | 10270 | list_for_each_entry(cache, |
10271 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10272 | list) | |
868f401a | 10273 | inc_block_group_ro(cache, 1); |
1095cc0d | 10274 | list_for_each_entry(cache, |
10275 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10276 | list) | |
868f401a | 10277 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10278 | } |
f0486c68 YZ |
10279 | |
10280 | init_global_block_rsv(info); | |
0b86a832 CM |
10281 | ret = 0; |
10282 | error: | |
9078a3e1 | 10283 | btrfs_free_path(path); |
0b86a832 | 10284 | return ret; |
9078a3e1 | 10285 | } |
6324fbf3 | 10286 | |
ea658bad JB |
10287 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
10288 | struct btrfs_root *root) | |
10289 | { | |
10290 | struct btrfs_block_group_cache *block_group, *tmp; | |
10291 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
10292 | struct btrfs_block_group_item item; | |
10293 | struct btrfs_key key; | |
10294 | int ret = 0; | |
d9a0540a | 10295 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10296 | |
d9a0540a | 10297 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10298 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10299 | if (ret) |
c92f6be3 | 10300 | goto next; |
ea658bad JB |
10301 | |
10302 | spin_lock(&block_group->lock); | |
10303 | memcpy(&item, &block_group->item, sizeof(item)); | |
10304 | memcpy(&key, &block_group->key, sizeof(key)); | |
10305 | spin_unlock(&block_group->lock); | |
10306 | ||
10307 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10308 | sizeof(item)); | |
10309 | if (ret) | |
10310 | btrfs_abort_transaction(trans, extent_root, ret); | |
6df9a95e JB |
10311 | ret = btrfs_finish_chunk_alloc(trans, extent_root, |
10312 | key.objectid, key.offset); | |
10313 | if (ret) | |
10314 | btrfs_abort_transaction(trans, extent_root, ret); | |
1e144fb8 OS |
10315 | add_block_group_free_space(trans, root->fs_info, block_group); |
10316 | /* already aborted the transaction if it failed. */ | |
c92f6be3 FM |
10317 | next: |
10318 | list_del_init(&block_group->bg_list); | |
ea658bad | 10319 | } |
d9a0540a | 10320 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10321 | } |
10322 | ||
6324fbf3 CM |
10323 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
10324 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 10325 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
10326 | u64 size) |
10327 | { | |
10328 | int ret; | |
6324fbf3 CM |
10329 | struct btrfs_root *extent_root; |
10330 | struct btrfs_block_group_cache *cache; | |
6324fbf3 | 10331 | extent_root = root->fs_info->extent_root; |
6324fbf3 | 10332 | |
995946dd | 10333 | btrfs_set_log_full_commit(root->fs_info, trans); |
e02119d5 | 10334 | |
920e4a58 | 10335 | cache = btrfs_create_block_group_cache(root, chunk_offset, size); |
0f9dd46c JB |
10336 | if (!cache) |
10337 | return -ENOMEM; | |
34d52cb6 | 10338 | |
6324fbf3 | 10339 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 | 10340 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
6324fbf3 CM |
10341 | btrfs_set_block_group_flags(&cache->item, type); |
10342 | ||
920e4a58 | 10343 | cache->flags = type; |
11833d66 | 10344 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10345 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10346 | cache->needs_free_space = 1; |
835d974f JB |
10347 | ret = exclude_super_stripes(root, cache); |
10348 | if (ret) { | |
10349 | /* | |
10350 | * We may have excluded something, so call this just in | |
10351 | * case. | |
10352 | */ | |
10353 | free_excluded_extents(root, cache); | |
920e4a58 | 10354 | btrfs_put_block_group(cache); |
835d974f JB |
10355 | return ret; |
10356 | } | |
96303081 | 10357 | |
817d52f8 JB |
10358 | add_new_free_space(cache, root->fs_info, chunk_offset, |
10359 | chunk_offset + size); | |
10360 | ||
11833d66 YZ |
10361 | free_excluded_extents(root, cache); |
10362 | ||
d0bd4560 JB |
10363 | #ifdef CONFIG_BTRFS_DEBUG |
10364 | if (btrfs_should_fragment_free_space(root, cache)) { | |
10365 | u64 new_bytes_used = size - bytes_used; | |
10366 | ||
10367 | bytes_used += new_bytes_used >> 1; | |
10368 | fragment_free_space(root, cache); | |
10369 | } | |
10370 | #endif | |
2e6e5183 FM |
10371 | /* |
10372 | * Call to ensure the corresponding space_info object is created and | |
10373 | * assigned to our block group, but don't update its counters just yet. | |
10374 | * We want our bg to be added to the rbtree with its ->space_info set. | |
10375 | */ | |
e40edf2d | 10376 | ret = update_space_info(root->fs_info, cache->flags, 0, 0, 0, |
2e6e5183 FM |
10377 | &cache->space_info); |
10378 | if (ret) { | |
10379 | btrfs_remove_free_space_cache(cache); | |
10380 | btrfs_put_block_group(cache); | |
10381 | return ret; | |
10382 | } | |
10383 | ||
8c579fe7 JB |
10384 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
10385 | if (ret) { | |
10386 | btrfs_remove_free_space_cache(cache); | |
10387 | btrfs_put_block_group(cache); | |
10388 | return ret; | |
10389 | } | |
10390 | ||
2e6e5183 FM |
10391 | /* |
10392 | * Now that our block group has its ->space_info set and is inserted in | |
10393 | * the rbtree, update the space info's counters. | |
10394 | */ | |
c83f8eff | 10395 | trace_btrfs_add_block_group(root->fs_info, cache, 1); |
6324fbf3 | 10396 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10397 | cache->bytes_super, &cache->space_info); |
8c579fe7 JB |
10398 | if (ret) { |
10399 | btrfs_remove_free_space_cache(cache); | |
10400 | spin_lock(&root->fs_info->block_group_cache_lock); | |
10401 | rb_erase(&cache->cache_node, | |
10402 | &root->fs_info->block_group_cache_tree); | |
01eacb27 | 10403 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
10404 | spin_unlock(&root->fs_info->block_group_cache_lock); |
10405 | btrfs_put_block_group(cache); | |
10406 | return ret; | |
10407 | } | |
c7c144db | 10408 | update_global_block_rsv(root->fs_info); |
1b2da372 | 10409 | |
b742bb82 | 10410 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 10411 | |
47ab2a6c | 10412 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10413 | |
d18a2c44 | 10414 | set_avail_alloc_bits(extent_root->fs_info, type); |
6324fbf3 CM |
10415 | return 0; |
10416 | } | |
1a40e23b | 10417 | |
10ea00f5 ID |
10418 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10419 | { | |
899c81ea ID |
10420 | u64 extra_flags = chunk_to_extended(flags) & |
10421 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10422 | |
de98ced9 | 10423 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10424 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10425 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10426 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10427 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10428 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10429 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10430 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10431 | } |
10432 | ||
1a40e23b | 10433 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
04216820 FM |
10434 | struct btrfs_root *root, u64 group_start, |
10435 | struct extent_map *em) | |
1a40e23b ZY |
10436 | { |
10437 | struct btrfs_path *path; | |
10438 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10439 | struct btrfs_free_cluster *cluster; |
0af3d00b | 10440 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 10441 | struct btrfs_key key; |
0af3d00b | 10442 | struct inode *inode; |
c1895442 | 10443 | struct kobject *kobj = NULL; |
1a40e23b | 10444 | int ret; |
10ea00f5 | 10445 | int index; |
89a55897 | 10446 | int factor; |
4f69cb98 | 10447 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10448 | bool remove_em; |
1a40e23b | 10449 | |
1a40e23b ZY |
10450 | root = root->fs_info->extent_root; |
10451 | ||
10452 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
10453 | BUG_ON(!block_group); | |
c146afad | 10454 | BUG_ON(!block_group->ro); |
1a40e23b | 10455 | |
9f7c43c9 | 10456 | /* |
10457 | * Free the reserved super bytes from this block group before | |
10458 | * remove it. | |
10459 | */ | |
10460 | free_excluded_extents(root, block_group); | |
10461 | ||
1a40e23b | 10462 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 10463 | index = get_block_group_index(block_group); |
89a55897 JB |
10464 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10465 | BTRFS_BLOCK_GROUP_RAID1 | | |
10466 | BTRFS_BLOCK_GROUP_RAID10)) | |
10467 | factor = 2; | |
10468 | else | |
10469 | factor = 1; | |
1a40e23b | 10470 | |
44fb5511 CM |
10471 | /* make sure this block group isn't part of an allocation cluster */ |
10472 | cluster = &root->fs_info->data_alloc_cluster; | |
10473 | spin_lock(&cluster->refill_lock); | |
10474 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10475 | spin_unlock(&cluster->refill_lock); | |
10476 | ||
10477 | /* | |
10478 | * make sure this block group isn't part of a metadata | |
10479 | * allocation cluster | |
10480 | */ | |
10481 | cluster = &root->fs_info->meta_alloc_cluster; | |
10482 | spin_lock(&cluster->refill_lock); | |
10483 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10484 | spin_unlock(&cluster->refill_lock); | |
10485 | ||
1a40e23b | 10486 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10487 | if (!path) { |
10488 | ret = -ENOMEM; | |
10489 | goto out; | |
10490 | } | |
1a40e23b | 10491 | |
1bbc621e CM |
10492 | /* |
10493 | * get the inode first so any iput calls done for the io_list | |
10494 | * aren't the final iput (no unlinks allowed now) | |
10495 | */ | |
10b2f34d | 10496 | inode = lookup_free_space_inode(tree_root, block_group, path); |
1bbc621e CM |
10497 | |
10498 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10499 | /* | |
10500 | * make sure our free spache cache IO is done before remove the | |
10501 | * free space inode | |
10502 | */ | |
10503 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10504 | if (!list_empty(&block_group->io_list)) { | |
10505 | list_del_init(&block_group->io_list); | |
10506 | ||
10507 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10508 | ||
10509 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10510 | btrfs_wait_cache_io(root, trans, block_group, | |
10511 | &block_group->io_ctl, path, | |
10512 | block_group->key.objectid); | |
10513 | btrfs_put_block_group(block_group); | |
10514 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10515 | } | |
10516 | ||
10517 | if (!list_empty(&block_group->dirty_list)) { | |
10518 | list_del_init(&block_group->dirty_list); | |
10519 | btrfs_put_block_group(block_group); | |
10520 | } | |
10521 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10522 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10523 | ||
0af3d00b | 10524 | if (!IS_ERR(inode)) { |
b532402e | 10525 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
10526 | if (ret) { |
10527 | btrfs_add_delayed_iput(inode); | |
10528 | goto out; | |
10529 | } | |
0af3d00b JB |
10530 | clear_nlink(inode); |
10531 | /* One for the block groups ref */ | |
10532 | spin_lock(&block_group->lock); | |
10533 | if (block_group->iref) { | |
10534 | block_group->iref = 0; | |
10535 | block_group->inode = NULL; | |
10536 | spin_unlock(&block_group->lock); | |
10537 | iput(inode); | |
10538 | } else { | |
10539 | spin_unlock(&block_group->lock); | |
10540 | } | |
10541 | /* One for our lookup ref */ | |
455757c3 | 10542 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10543 | } |
10544 | ||
10545 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10546 | key.offset = block_group->key.objectid; | |
10547 | key.type = 0; | |
10548 | ||
10549 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10550 | if (ret < 0) | |
10551 | goto out; | |
10552 | if (ret > 0) | |
b3b4aa74 | 10553 | btrfs_release_path(path); |
0af3d00b JB |
10554 | if (ret == 0) { |
10555 | ret = btrfs_del_item(trans, tree_root, path); | |
10556 | if (ret) | |
10557 | goto out; | |
b3b4aa74 | 10558 | btrfs_release_path(path); |
0af3d00b JB |
10559 | } |
10560 | ||
3dfdb934 | 10561 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
10562 | rb_erase(&block_group->cache_node, |
10563 | &root->fs_info->block_group_cache_tree); | |
292cbd51 | 10564 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd LB |
10565 | |
10566 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
10567 | root->fs_info->first_logical_byte = (u64)-1; | |
3dfdb934 | 10568 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 10569 | |
80eb234a | 10570 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10571 | /* |
10572 | * we must use list_del_init so people can check to see if they | |
10573 | * are still on the list after taking the semaphore | |
10574 | */ | |
10575 | list_del_init(&block_group->list); | |
6ab0a202 | 10576 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10577 | kobj = block_group->space_info->block_group_kobjs[index]; |
10578 | block_group->space_info->block_group_kobjs[index] = NULL; | |
10ea00f5 | 10579 | clear_avail_alloc_bits(root->fs_info, block_group->flags); |
6ab0a202 | 10580 | } |
80eb234a | 10581 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10582 | if (kobj) { |
10583 | kobject_del(kobj); | |
10584 | kobject_put(kobj); | |
10585 | } | |
1a40e23b | 10586 | |
4f69cb98 FM |
10587 | if (block_group->has_caching_ctl) |
10588 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10589 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10590 | wait_block_group_cache_done(block_group); |
4f69cb98 FM |
10591 | if (block_group->has_caching_ctl) { |
10592 | down_write(&root->fs_info->commit_root_sem); | |
10593 | if (!caching_ctl) { | |
10594 | struct btrfs_caching_control *ctl; | |
10595 | ||
10596 | list_for_each_entry(ctl, | |
10597 | &root->fs_info->caching_block_groups, list) | |
10598 | if (ctl->block_group == block_group) { | |
10599 | caching_ctl = ctl; | |
10600 | atomic_inc(&caching_ctl->count); | |
10601 | break; | |
10602 | } | |
10603 | } | |
10604 | if (caching_ctl) | |
10605 | list_del_init(&caching_ctl->list); | |
10606 | up_write(&root->fs_info->commit_root_sem); | |
10607 | if (caching_ctl) { | |
10608 | /* Once for the caching bgs list and once for us. */ | |
10609 | put_caching_control(caching_ctl); | |
10610 | put_caching_control(caching_ctl); | |
10611 | } | |
10612 | } | |
817d52f8 | 10613 | |
ce93ec54 JB |
10614 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10615 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10616 | WARN_ON(1); |
10617 | } | |
10618 | if (!list_empty(&block_group->io_list)) { | |
10619 | WARN_ON(1); | |
ce93ec54 JB |
10620 | } |
10621 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10622 | btrfs_remove_free_space_cache(block_group); |
10623 | ||
c146afad | 10624 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10625 | list_del_init(&block_group->ro_list); |
18d018ad ZL |
10626 | |
10627 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
10628 | WARN_ON(block_group->space_info->total_bytes | |
10629 | < block_group->key.offset); | |
10630 | WARN_ON(block_group->space_info->bytes_readonly | |
10631 | < block_group->key.offset); | |
10632 | WARN_ON(block_group->space_info->disk_total | |
10633 | < block_group->key.offset * factor); | |
10634 | } | |
c146afad YZ |
10635 | block_group->space_info->total_bytes -= block_group->key.offset; |
10636 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10637 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10638 | |
c146afad | 10639 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10640 | |
0af3d00b JB |
10641 | memcpy(&key, &block_group->key, sizeof(key)); |
10642 | ||
04216820 | 10643 | lock_chunks(root); |
495e64f4 FM |
10644 | if (!list_empty(&em->list)) { |
10645 | /* We're in the transaction->pending_chunks list. */ | |
10646 | free_extent_map(em); | |
10647 | } | |
04216820 FM |
10648 | spin_lock(&block_group->lock); |
10649 | block_group->removed = 1; | |
10650 | /* | |
10651 | * At this point trimming can't start on this block group, because we | |
10652 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10653 | * so no one can't find it anymore and even if someone already got this | |
10654 | * block group before we removed it from the rbtree, they have already | |
10655 | * incremented block_group->trimming - if they didn't, they won't find | |
10656 | * any free space entries because we already removed them all when we | |
10657 | * called btrfs_remove_free_space_cache(). | |
10658 | * | |
10659 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10660 | * to prevent the same logical address range and physical device space | |
10661 | * ranges from being reused for a new block group. This is because our | |
10662 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10663 | * completely transactionless, so while it is trimming a range the | |
10664 | * currently running transaction might finish and a new one start, | |
10665 | * allowing for new block groups to be created that can reuse the same | |
10666 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10667 | * |
10668 | * There may also be an implicit trim operation if the file system | |
10669 | * is mounted with -odiscard. The same protections must remain | |
10670 | * in place until the extents have been discarded completely when | |
10671 | * the transaction commit has completed. | |
04216820 FM |
10672 | */ |
10673 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10674 | /* | |
10675 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10676 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10677 | * before checking block_group->removed). | |
10678 | */ | |
10679 | if (!remove_em) { | |
10680 | /* | |
10681 | * Our em might be in trans->transaction->pending_chunks which | |
10682 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10683 | * and so is the fs_info->pinned_chunks list. | |
10684 | * | |
10685 | * So at this point we must be holding the chunk_mutex to avoid | |
10686 | * any races with chunk allocation (more specifically at | |
10687 | * volumes.c:contains_pending_extent()), to ensure it always | |
10688 | * sees the em, either in the pending_chunks list or in the | |
10689 | * pinned_chunks list. | |
10690 | */ | |
10691 | list_move_tail(&em->list, &root->fs_info->pinned_chunks); | |
10692 | } | |
10693 | spin_unlock(&block_group->lock); | |
04216820 FM |
10694 | |
10695 | if (remove_em) { | |
10696 | struct extent_map_tree *em_tree; | |
10697 | ||
10698 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
10699 | write_lock(&em_tree->lock); | |
8dbcd10f FM |
10700 | /* |
10701 | * The em might be in the pending_chunks list, so make sure the | |
10702 | * chunk mutex is locked, since remove_extent_mapping() will | |
10703 | * delete us from that list. | |
10704 | */ | |
04216820 FM |
10705 | remove_extent_mapping(em_tree, em); |
10706 | write_unlock(&em_tree->lock); | |
10707 | /* once for the tree */ | |
10708 | free_extent_map(em); | |
10709 | } | |
10710 | ||
8dbcd10f FM |
10711 | unlock_chunks(root); |
10712 | ||
1e144fb8 OS |
10713 | ret = remove_block_group_free_space(trans, root->fs_info, block_group); |
10714 | if (ret) | |
10715 | goto out; | |
10716 | ||
fa9c0d79 CM |
10717 | btrfs_put_block_group(block_group); |
10718 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10719 | |
10720 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10721 | if (ret > 0) | |
10722 | ret = -EIO; | |
10723 | if (ret < 0) | |
10724 | goto out; | |
10725 | ||
10726 | ret = btrfs_del_item(trans, root, path); | |
10727 | out: | |
10728 | btrfs_free_path(path); | |
10729 | return ret; | |
10730 | } | |
acce952b | 10731 | |
8eab77ff | 10732 | struct btrfs_trans_handle * |
7fd01182 FM |
10733 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10734 | const u64 chunk_offset) | |
8eab77ff | 10735 | { |
7fd01182 FM |
10736 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10737 | struct extent_map *em; | |
10738 | struct map_lookup *map; | |
10739 | unsigned int num_items; | |
10740 | ||
10741 | read_lock(&em_tree->lock); | |
10742 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10743 | read_unlock(&em_tree->lock); | |
10744 | ASSERT(em && em->start == chunk_offset); | |
10745 | ||
8eab77ff | 10746 | /* |
7fd01182 FM |
10747 | * We need to reserve 3 + N units from the metadata space info in order |
10748 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10749 | * btrfs_remove_block_group()), which are used for: | |
10750 | * | |
8eab77ff FM |
10751 | * 1 unit for adding the free space inode's orphan (located in the tree |
10752 | * of tree roots). | |
7fd01182 FM |
10753 | * 1 unit for deleting the block group item (located in the extent |
10754 | * tree). | |
10755 | * 1 unit for deleting the free space item (located in tree of tree | |
10756 | * roots). | |
10757 | * N units for deleting N device extent items corresponding to each | |
10758 | * stripe (located in the device tree). | |
10759 | * | |
10760 | * In order to remove a block group we also need to reserve units in the | |
10761 | * system space info in order to update the chunk tree (update one or | |
10762 | * more device items and remove one chunk item), but this is done at | |
10763 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10764 | */ |
95617d69 | 10765 | map = em->map_lookup; |
7fd01182 FM |
10766 | num_items = 3 + map->num_stripes; |
10767 | free_extent_map(em); | |
10768 | ||
8eab77ff | 10769 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10770 | num_items, 1); |
8eab77ff FM |
10771 | } |
10772 | ||
47ab2a6c JB |
10773 | /* |
10774 | * Process the unused_bgs list and remove any that don't have any allocated | |
10775 | * space inside of them. | |
10776 | */ | |
10777 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10778 | { | |
10779 | struct btrfs_block_group_cache *block_group; | |
10780 | struct btrfs_space_info *space_info; | |
10781 | struct btrfs_root *root = fs_info->extent_root; | |
10782 | struct btrfs_trans_handle *trans; | |
10783 | int ret = 0; | |
10784 | ||
10785 | if (!fs_info->open) | |
10786 | return; | |
10787 | ||
10788 | spin_lock(&fs_info->unused_bgs_lock); | |
10789 | while (!list_empty(&fs_info->unused_bgs)) { | |
10790 | u64 start, end; | |
e33e17ee | 10791 | int trimming; |
47ab2a6c JB |
10792 | |
10793 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10794 | struct btrfs_block_group_cache, | |
10795 | bg_list); | |
47ab2a6c | 10796 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10797 | |
10798 | space_info = block_group->space_info; | |
10799 | ||
47ab2a6c JB |
10800 | if (ret || btrfs_mixed_space_info(space_info)) { |
10801 | btrfs_put_block_group(block_group); | |
10802 | continue; | |
10803 | } | |
10804 | spin_unlock(&fs_info->unused_bgs_lock); | |
10805 | ||
d5f2e33b | 10806 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10807 | |
47ab2a6c JB |
10808 | /* Don't want to race with allocators so take the groups_sem */ |
10809 | down_write(&space_info->groups_sem); | |
10810 | spin_lock(&block_group->lock); | |
10811 | if (block_group->reserved || | |
10812 | btrfs_block_group_used(&block_group->item) || | |
aefbe9a6 ZL |
10813 | block_group->ro || |
10814 | list_is_singular(&block_group->list)) { | |
47ab2a6c JB |
10815 | /* |
10816 | * We want to bail if we made new allocations or have | |
10817 | * outstanding allocations in this block group. We do | |
10818 | * the ro check in case balance is currently acting on | |
10819 | * this block group. | |
10820 | */ | |
10821 | spin_unlock(&block_group->lock); | |
10822 | up_write(&space_info->groups_sem); | |
10823 | goto next; | |
10824 | } | |
10825 | spin_unlock(&block_group->lock); | |
10826 | ||
10827 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10828 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10829 | up_write(&space_info->groups_sem); |
10830 | if (ret < 0) { | |
10831 | ret = 0; | |
10832 | goto next; | |
10833 | } | |
10834 | ||
10835 | /* | |
10836 | * Want to do this before we do anything else so we can recover | |
10837 | * properly if we fail to join the transaction. | |
10838 | */ | |
7fd01182 FM |
10839 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10840 | block_group->key.objectid); | |
47ab2a6c | 10841 | if (IS_ERR(trans)) { |
868f401a | 10842 | btrfs_dec_block_group_ro(root, block_group); |
47ab2a6c JB |
10843 | ret = PTR_ERR(trans); |
10844 | goto next; | |
10845 | } | |
10846 | ||
10847 | /* | |
10848 | * We could have pending pinned extents for this block group, | |
10849 | * just delete them, we don't care about them anymore. | |
10850 | */ | |
10851 | start = block_group->key.objectid; | |
10852 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10853 | /* |
10854 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10855 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10856 | * another task might be running finish_extent_commit() for the | |
10857 | * previous transaction N - 1, and have seen a range belonging | |
10858 | * to the block group in freed_extents[] before we were able to | |
10859 | * clear the whole block group range from freed_extents[]. This | |
10860 | * means that task can lookup for the block group after we | |
10861 | * unpinned it from freed_extents[] and removed it, leading to | |
10862 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10863 | */ | |
10864 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10865 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10866 | EXTENT_DIRTY); |
758eb51e | 10867 | if (ret) { |
d4b450cd | 10868 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
868f401a | 10869 | btrfs_dec_block_group_ro(root, block_group); |
758eb51e FM |
10870 | goto end_trans; |
10871 | } | |
10872 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10873 | EXTENT_DIRTY); |
758eb51e | 10874 | if (ret) { |
d4b450cd | 10875 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
868f401a | 10876 | btrfs_dec_block_group_ro(root, block_group); |
758eb51e FM |
10877 | goto end_trans; |
10878 | } | |
d4b450cd | 10879 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10880 | |
10881 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10882 | spin_lock(&space_info->lock); |
10883 | spin_lock(&block_group->lock); | |
10884 | ||
10885 | space_info->bytes_pinned -= block_group->pinned; | |
10886 | space_info->bytes_readonly += block_group->pinned; | |
10887 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10888 | -block_group->pinned); | |
47ab2a6c JB |
10889 | block_group->pinned = 0; |
10890 | ||
c30666d4 ZL |
10891 | spin_unlock(&block_group->lock); |
10892 | spin_unlock(&space_info->lock); | |
10893 | ||
e33e17ee JM |
10894 | /* DISCARD can flip during remount */ |
10895 | trimming = btrfs_test_opt(root, DISCARD); | |
10896 | ||
10897 | /* Implicit trim during transaction commit. */ | |
10898 | if (trimming) | |
10899 | btrfs_get_block_group_trimming(block_group); | |
10900 | ||
47ab2a6c JB |
10901 | /* |
10902 | * Btrfs_remove_chunk will abort the transaction if things go | |
10903 | * horribly wrong. | |
10904 | */ | |
10905 | ret = btrfs_remove_chunk(trans, root, | |
10906 | block_group->key.objectid); | |
e33e17ee JM |
10907 | |
10908 | if (ret) { | |
10909 | if (trimming) | |
10910 | btrfs_put_block_group_trimming(block_group); | |
10911 | goto end_trans; | |
10912 | } | |
10913 | ||
10914 | /* | |
10915 | * If we're not mounted with -odiscard, we can just forget | |
10916 | * about this block group. Otherwise we'll need to wait | |
10917 | * until transaction commit to do the actual discard. | |
10918 | */ | |
10919 | if (trimming) { | |
348a0013 FM |
10920 | spin_lock(&fs_info->unused_bgs_lock); |
10921 | /* | |
10922 | * A concurrent scrub might have added us to the list | |
10923 | * fs_info->unused_bgs, so use a list_move operation | |
10924 | * to add the block group to the deleted_bgs list. | |
10925 | */ | |
e33e17ee JM |
10926 | list_move(&block_group->bg_list, |
10927 | &trans->transaction->deleted_bgs); | |
348a0013 | 10928 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10929 | btrfs_get_block_group(block_group); |
10930 | } | |
758eb51e | 10931 | end_trans: |
47ab2a6c JB |
10932 | btrfs_end_transaction(trans, root); |
10933 | next: | |
d5f2e33b | 10934 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10935 | btrfs_put_block_group(block_group); |
10936 | spin_lock(&fs_info->unused_bgs_lock); | |
10937 | } | |
10938 | spin_unlock(&fs_info->unused_bgs_lock); | |
10939 | } | |
10940 | ||
c59021f8 | 10941 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10942 | { | |
10943 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10944 | struct btrfs_super_block *disk_super; |
10945 | u64 features; | |
10946 | u64 flags; | |
10947 | int mixed = 0; | |
c59021f8 | 10948 | int ret; |
10949 | ||
6c41761f | 10950 | disk_super = fs_info->super_copy; |
1aba86d6 | 10951 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10952 | return -EINVAL; |
c59021f8 | 10953 | |
1aba86d6 | 10954 | features = btrfs_super_incompat_flags(disk_super); |
10955 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10956 | mixed = 1; | |
c59021f8 | 10957 | |
1aba86d6 | 10958 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
e40edf2d | 10959 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
c59021f8 | 10960 | if (ret) |
1aba86d6 | 10961 | goto out; |
c59021f8 | 10962 | |
1aba86d6 | 10963 | if (mixed) { |
10964 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
e40edf2d | 10965 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10966 | } else { |
10967 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
e40edf2d | 10968 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10969 | if (ret) |
10970 | goto out; | |
10971 | ||
10972 | flags = BTRFS_BLOCK_GROUP_DATA; | |
e40edf2d | 10973 | ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info); |
1aba86d6 | 10974 | } |
10975 | out: | |
c59021f8 | 10976 | return ret; |
10977 | } | |
10978 | ||
acce952b | 10979 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
10980 | { | |
678886bd | 10981 | return unpin_extent_range(root, start, end, false); |
acce952b | 10982 | } |
10983 | ||
499f377f JM |
10984 | /* |
10985 | * It used to be that old block groups would be left around forever. | |
10986 | * Iterating over them would be enough to trim unused space. Since we | |
10987 | * now automatically remove them, we also need to iterate over unallocated | |
10988 | * space. | |
10989 | * | |
10990 | * We don't want a transaction for this since the discard may take a | |
10991 | * substantial amount of time. We don't require that a transaction be | |
10992 | * running, but we do need to take a running transaction into account | |
10993 | * to ensure that we're not discarding chunks that were released in | |
10994 | * the current transaction. | |
10995 | * | |
10996 | * Holding the chunks lock will prevent other threads from allocating | |
10997 | * or releasing chunks, but it won't prevent a running transaction | |
10998 | * from committing and releasing the memory that the pending chunks | |
10999 | * list head uses. For that, we need to take a reference to the | |
11000 | * transaction. | |
11001 | */ | |
11002 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
11003 | u64 minlen, u64 *trimmed) | |
11004 | { | |
11005 | u64 start = 0, len = 0; | |
11006 | int ret; | |
11007 | ||
11008 | *trimmed = 0; | |
11009 | ||
11010 | /* Not writeable = nothing to do. */ | |
11011 | if (!device->writeable) | |
11012 | return 0; | |
11013 | ||
11014 | /* No free space = nothing to do. */ | |
11015 | if (device->total_bytes <= device->bytes_used) | |
11016 | return 0; | |
11017 | ||
11018 | ret = 0; | |
11019 | ||
11020 | while (1) { | |
11021 | struct btrfs_fs_info *fs_info = device->dev_root->fs_info; | |
11022 | struct btrfs_transaction *trans; | |
11023 | u64 bytes; | |
11024 | ||
11025 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
11026 | if (ret) | |
11027 | return ret; | |
11028 | ||
11029 | down_read(&fs_info->commit_root_sem); | |
11030 | ||
11031 | spin_lock(&fs_info->trans_lock); | |
11032 | trans = fs_info->running_transaction; | |
11033 | if (trans) | |
11034 | atomic_inc(&trans->use_count); | |
11035 | spin_unlock(&fs_info->trans_lock); | |
11036 | ||
11037 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
11038 | &start, &len); | |
11039 | if (trans) | |
11040 | btrfs_put_transaction(trans); | |
11041 | ||
11042 | if (ret) { | |
11043 | up_read(&fs_info->commit_root_sem); | |
11044 | mutex_unlock(&fs_info->chunk_mutex); | |
11045 | if (ret == -ENOSPC) | |
11046 | ret = 0; | |
11047 | break; | |
11048 | } | |
11049 | ||
11050 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
11051 | up_read(&fs_info->commit_root_sem); | |
11052 | mutex_unlock(&fs_info->chunk_mutex); | |
11053 | ||
11054 | if (ret) | |
11055 | break; | |
11056 | ||
11057 | start += len; | |
11058 | *trimmed += bytes; | |
11059 | ||
11060 | if (fatal_signal_pending(current)) { | |
11061 | ret = -ERESTARTSYS; | |
11062 | break; | |
11063 | } | |
11064 | ||
11065 | cond_resched(); | |
11066 | } | |
11067 | ||
11068 | return ret; | |
11069 | } | |
11070 | ||
f7039b1d LD |
11071 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) |
11072 | { | |
11073 | struct btrfs_fs_info *fs_info = root->fs_info; | |
11074 | struct btrfs_block_group_cache *cache = NULL; | |
499f377f JM |
11075 | struct btrfs_device *device; |
11076 | struct list_head *devices; | |
f7039b1d LD |
11077 | u64 group_trimmed; |
11078 | u64 start; | |
11079 | u64 end; | |
11080 | u64 trimmed = 0; | |
2cac13e4 | 11081 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
11082 | int ret = 0; |
11083 | ||
2cac13e4 LB |
11084 | /* |
11085 | * try to trim all FS space, our block group may start from non-zero. | |
11086 | */ | |
11087 | if (range->len == total_bytes) | |
11088 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
11089 | else | |
11090 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
11091 | |
11092 | while (cache) { | |
11093 | if (cache->key.objectid >= (range->start + range->len)) { | |
11094 | btrfs_put_block_group(cache); | |
11095 | break; | |
11096 | } | |
11097 | ||
11098 | start = max(range->start, cache->key.objectid); | |
11099 | end = min(range->start + range->len, | |
11100 | cache->key.objectid + cache->key.offset); | |
11101 | ||
11102 | if (end - start >= range->minlen) { | |
11103 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 11104 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
11105 | if (ret) { |
11106 | btrfs_put_block_group(cache); | |
11107 | break; | |
11108 | } | |
11109 | ret = wait_block_group_cache_done(cache); | |
11110 | if (ret) { | |
11111 | btrfs_put_block_group(cache); | |
11112 | break; | |
11113 | } | |
f7039b1d LD |
11114 | } |
11115 | ret = btrfs_trim_block_group(cache, | |
11116 | &group_trimmed, | |
11117 | start, | |
11118 | end, | |
11119 | range->minlen); | |
11120 | ||
11121 | trimmed += group_trimmed; | |
11122 | if (ret) { | |
11123 | btrfs_put_block_group(cache); | |
11124 | break; | |
11125 | } | |
11126 | } | |
11127 | ||
11128 | cache = next_block_group(fs_info->tree_root, cache); | |
11129 | } | |
11130 | ||
499f377f JM |
11131 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
11132 | devices = &root->fs_info->fs_devices->alloc_list; | |
11133 | list_for_each_entry(device, devices, dev_alloc_list) { | |
11134 | ret = btrfs_trim_free_extents(device, range->minlen, | |
11135 | &group_trimmed); | |
11136 | if (ret) | |
11137 | break; | |
11138 | ||
11139 | trimmed += group_trimmed; | |
11140 | } | |
11141 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
11142 | ||
f7039b1d LD |
11143 | range->len = trimmed; |
11144 | return ret; | |
11145 | } | |
8257b2dc MX |
11146 | |
11147 | /* | |
9ea24bbe FM |
11148 | * btrfs_{start,end}_write_no_snapshoting() are similar to |
11149 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing | |
11150 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11151 | * but flush the data into disk after the snapshot creation, or to prevent | |
11152 | * operations while snapshoting is ongoing and that cause the snapshot to be | |
11153 | * inconsistent (writes followed by expanding truncates for example). | |
8257b2dc | 11154 | */ |
9ea24bbe | 11155 | void btrfs_end_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc MX |
11156 | { |
11157 | percpu_counter_dec(&root->subv_writers->counter); | |
11158 | /* | |
a83342aa | 11159 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
11160 | */ |
11161 | smp_mb(); | |
11162 | if (waitqueue_active(&root->subv_writers->wait)) | |
11163 | wake_up(&root->subv_writers->wait); | |
11164 | } | |
11165 | ||
9ea24bbe | 11166 | int btrfs_start_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc | 11167 | { |
ee39b432 | 11168 | if (atomic_read(&root->will_be_snapshoted)) |
8257b2dc MX |
11169 | return 0; |
11170 | ||
11171 | percpu_counter_inc(&root->subv_writers->counter); | |
11172 | /* | |
11173 | * Make sure counter is updated before we check for snapshot creation. | |
11174 | */ | |
11175 | smp_mb(); | |
ee39b432 | 11176 | if (atomic_read(&root->will_be_snapshoted)) { |
9ea24bbe | 11177 | btrfs_end_write_no_snapshoting(root); |
8257b2dc MX |
11178 | return 0; |
11179 | } | |
11180 | return 1; | |
11181 | } | |
0bc19f90 ZL |
11182 | |
11183 | static int wait_snapshoting_atomic_t(atomic_t *a) | |
11184 | { | |
11185 | schedule(); | |
11186 | return 0; | |
11187 | } | |
11188 | ||
11189 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) | |
11190 | { | |
11191 | while (true) { | |
11192 | int ret; | |
11193 | ||
11194 | ret = btrfs_start_write_no_snapshoting(root); | |
11195 | if (ret) | |
11196 | break; | |
11197 | wait_on_atomic_t(&root->will_be_snapshoted, | |
11198 | wait_snapshoting_atomic_t, | |
11199 | TASK_UNINTERRUPTIBLE); | |
11200 | } | |
11201 | } |