]>
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" |
3fed40cc | 36 | #include "math.h" |
6ab0a202 | 37 | #include "sysfs.h" |
fcebe456 | 38 | #include "qgroup.h" |
fec577fb | 39 | |
709c0486 AJ |
40 | #undef SCRAMBLE_DELAYED_REFS |
41 | ||
9e622d6b MX |
42 | /* |
43 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
44 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
45 | * if we really need one. | |
46 | * | |
0e4f8f88 CM |
47 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
48 | * if we have very few chunks already allocated. This is | |
49 | * used as part of the clustering code to help make sure | |
50 | * we have a good pool of storage to cluster in, without | |
51 | * filling the FS with empty chunks | |
52 | * | |
9e622d6b MX |
53 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
54 | * | |
0e4f8f88 CM |
55 | */ |
56 | enum { | |
57 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
58 | CHUNK_ALLOC_LIMITED = 1, |
59 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
60 | }; |
61 | ||
fb25e914 JB |
62 | /* |
63 | * Control how reservations are dealt with. | |
64 | * | |
65 | * RESERVE_FREE - freeing a reservation. | |
66 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
67 | * ENOSPC accounting | |
68 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
69 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
70 | */ | |
71 | enum { | |
72 | RESERVE_FREE = 0, | |
73 | RESERVE_ALLOC = 1, | |
74 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
75 | }; | |
76 | ||
ce93ec54 JB |
77 | static int update_block_group(struct btrfs_trans_handle *trans, |
78 | struct btrfs_root *root, u64 bytenr, | |
79 | u64 num_bytes, int alloc); | |
5d4f98a2 YZ |
80 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
81 | struct btrfs_root *root, | |
c682f9b3 | 82 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
83 | u64 root_objectid, u64 owner_objectid, |
84 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 85 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
86 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
87 | struct extent_buffer *leaf, | |
88 | struct btrfs_extent_item *ei); | |
89 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
90 | struct btrfs_root *root, | |
91 | u64 parent, u64 root_objectid, | |
92 | u64 flags, u64 owner, u64 offset, | |
93 | struct btrfs_key *ins, int ref_mod); | |
94 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
95 | struct btrfs_root *root, | |
96 | u64 parent, u64 root_objectid, | |
97 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 98 | int level, struct btrfs_key *ins); |
6a63209f | 99 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 JB |
100 | struct btrfs_root *extent_root, u64 flags, |
101 | int force); | |
11833d66 YZ |
102 | static int find_next_key(struct btrfs_path *path, int level, |
103 | struct btrfs_key *key); | |
9ed74f2d JB |
104 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
105 | int dump_block_groups); | |
fb25e914 | 106 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
e570fd27 MX |
107 | u64 num_bytes, int reserve, |
108 | int delalloc); | |
5d80366e JB |
109 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
110 | u64 num_bytes); | |
48a3b636 ES |
111 | int btrfs_pin_extent(struct btrfs_root *root, |
112 | u64 bytenr, u64 num_bytes, int reserved); | |
6a63209f | 113 | |
817d52f8 JB |
114 | static noinline int |
115 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
116 | { | |
117 | smp_mb(); | |
36cce922 JB |
118 | return cache->cached == BTRFS_CACHE_FINISHED || |
119 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
120 | } |
121 | ||
0f9dd46c JB |
122 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
123 | { | |
124 | return (cache->flags & bits) == bits; | |
125 | } | |
126 | ||
62a45b60 | 127 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
128 | { |
129 | atomic_inc(&cache->count); | |
130 | } | |
131 | ||
132 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
133 | { | |
f0486c68 YZ |
134 | if (atomic_dec_and_test(&cache->count)) { |
135 | WARN_ON(cache->pinned > 0); | |
136 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 137 | kfree(cache->free_space_ctl); |
11dfe35a | 138 | kfree(cache); |
f0486c68 | 139 | } |
11dfe35a JB |
140 | } |
141 | ||
0f9dd46c JB |
142 | /* |
143 | * this adds the block group to the fs_info rb tree for the block group | |
144 | * cache | |
145 | */ | |
b2950863 | 146 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
147 | struct btrfs_block_group_cache *block_group) |
148 | { | |
149 | struct rb_node **p; | |
150 | struct rb_node *parent = NULL; | |
151 | struct btrfs_block_group_cache *cache; | |
152 | ||
153 | spin_lock(&info->block_group_cache_lock); | |
154 | p = &info->block_group_cache_tree.rb_node; | |
155 | ||
156 | while (*p) { | |
157 | parent = *p; | |
158 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
159 | cache_node); | |
160 | if (block_group->key.objectid < cache->key.objectid) { | |
161 | p = &(*p)->rb_left; | |
162 | } else if (block_group->key.objectid > cache->key.objectid) { | |
163 | p = &(*p)->rb_right; | |
164 | } else { | |
165 | spin_unlock(&info->block_group_cache_lock); | |
166 | return -EEXIST; | |
167 | } | |
168 | } | |
169 | ||
170 | rb_link_node(&block_group->cache_node, parent, p); | |
171 | rb_insert_color(&block_group->cache_node, | |
172 | &info->block_group_cache_tree); | |
a1897fdd LB |
173 | |
174 | if (info->first_logical_byte > block_group->key.objectid) | |
175 | info->first_logical_byte = block_group->key.objectid; | |
176 | ||
0f9dd46c JB |
177 | spin_unlock(&info->block_group_cache_lock); |
178 | ||
179 | return 0; | |
180 | } | |
181 | ||
182 | /* | |
183 | * This will return the block group at or after bytenr if contains is 0, else | |
184 | * it will return the block group that contains the bytenr | |
185 | */ | |
186 | static struct btrfs_block_group_cache * | |
187 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
188 | int contains) | |
189 | { | |
190 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
191 | struct rb_node *n; | |
192 | u64 end, start; | |
193 | ||
194 | spin_lock(&info->block_group_cache_lock); | |
195 | n = info->block_group_cache_tree.rb_node; | |
196 | ||
197 | while (n) { | |
198 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
199 | cache_node); | |
200 | end = cache->key.objectid + cache->key.offset - 1; | |
201 | start = cache->key.objectid; | |
202 | ||
203 | if (bytenr < start) { | |
204 | if (!contains && (!ret || start < ret->key.objectid)) | |
205 | ret = cache; | |
206 | n = n->rb_left; | |
207 | } else if (bytenr > start) { | |
208 | if (contains && bytenr <= end) { | |
209 | ret = cache; | |
210 | break; | |
211 | } | |
212 | n = n->rb_right; | |
213 | } else { | |
214 | ret = cache; | |
215 | break; | |
216 | } | |
217 | } | |
a1897fdd | 218 | if (ret) { |
11dfe35a | 219 | btrfs_get_block_group(ret); |
a1897fdd LB |
220 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
221 | info->first_logical_byte = ret->key.objectid; | |
222 | } | |
0f9dd46c JB |
223 | spin_unlock(&info->block_group_cache_lock); |
224 | ||
225 | return ret; | |
226 | } | |
227 | ||
11833d66 YZ |
228 | static int add_excluded_extent(struct btrfs_root *root, |
229 | u64 start, u64 num_bytes) | |
817d52f8 | 230 | { |
11833d66 YZ |
231 | u64 end = start + num_bytes - 1; |
232 | set_extent_bits(&root->fs_info->freed_extents[0], | |
233 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
234 | set_extent_bits(&root->fs_info->freed_extents[1], | |
235 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
236 | return 0; | |
237 | } | |
817d52f8 | 238 | |
11833d66 YZ |
239 | static void free_excluded_extents(struct btrfs_root *root, |
240 | struct btrfs_block_group_cache *cache) | |
241 | { | |
242 | u64 start, end; | |
817d52f8 | 243 | |
11833d66 YZ |
244 | start = cache->key.objectid; |
245 | end = start + cache->key.offset - 1; | |
246 | ||
247 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
248 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
249 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
250 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
251 | } |
252 | ||
11833d66 YZ |
253 | static int exclude_super_stripes(struct btrfs_root *root, |
254 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 255 | { |
817d52f8 JB |
256 | u64 bytenr; |
257 | u64 *logical; | |
258 | int stripe_len; | |
259 | int i, nr, ret; | |
260 | ||
06b2331f YZ |
261 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
262 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
263 | cache->bytes_super += stripe_len; | |
264 | ret = add_excluded_extent(root, cache->key.objectid, | |
265 | stripe_len); | |
835d974f JB |
266 | if (ret) |
267 | return ret; | |
06b2331f YZ |
268 | } |
269 | ||
817d52f8 JB |
270 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
271 | bytenr = btrfs_sb_offset(i); | |
272 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
273 | cache->key.objectid, bytenr, | |
274 | 0, &logical, &nr, &stripe_len); | |
835d974f JB |
275 | if (ret) |
276 | return ret; | |
11833d66 | 277 | |
817d52f8 | 278 | while (nr--) { |
51bf5f0b JB |
279 | u64 start, len; |
280 | ||
281 | if (logical[nr] > cache->key.objectid + | |
282 | cache->key.offset) | |
283 | continue; | |
284 | ||
285 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
286 | continue; | |
287 | ||
288 | start = logical[nr]; | |
289 | if (start < cache->key.objectid) { | |
290 | start = cache->key.objectid; | |
291 | len = (logical[nr] + stripe_len) - start; | |
292 | } else { | |
293 | len = min_t(u64, stripe_len, | |
294 | cache->key.objectid + | |
295 | cache->key.offset - start); | |
296 | } | |
297 | ||
298 | cache->bytes_super += len; | |
299 | ret = add_excluded_extent(root, start, len); | |
835d974f JB |
300 | if (ret) { |
301 | kfree(logical); | |
302 | return ret; | |
303 | } | |
817d52f8 | 304 | } |
11833d66 | 305 | |
817d52f8 JB |
306 | kfree(logical); |
307 | } | |
817d52f8 JB |
308 | return 0; |
309 | } | |
310 | ||
11833d66 YZ |
311 | static struct btrfs_caching_control * |
312 | get_caching_control(struct btrfs_block_group_cache *cache) | |
313 | { | |
314 | struct btrfs_caching_control *ctl; | |
315 | ||
316 | spin_lock(&cache->lock); | |
dde5abee JB |
317 | if (!cache->caching_ctl) { |
318 | spin_unlock(&cache->lock); | |
11833d66 YZ |
319 | return NULL; |
320 | } | |
321 | ||
322 | ctl = cache->caching_ctl; | |
323 | atomic_inc(&ctl->count); | |
324 | spin_unlock(&cache->lock); | |
325 | return ctl; | |
326 | } | |
327 | ||
328 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
329 | { | |
330 | if (atomic_dec_and_test(&ctl->count)) | |
331 | kfree(ctl); | |
332 | } | |
333 | ||
d0bd4560 JB |
334 | #ifdef CONFIG_BTRFS_DEBUG |
335 | static void fragment_free_space(struct btrfs_root *root, | |
336 | struct btrfs_block_group_cache *block_group) | |
337 | { | |
338 | u64 start = block_group->key.objectid; | |
339 | u64 len = block_group->key.offset; | |
340 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
341 | root->nodesize : root->sectorsize; | |
342 | u64 step = chunk << 1; | |
343 | ||
344 | while (len > chunk) { | |
345 | btrfs_remove_free_space(block_group, start, chunk); | |
346 | start += step; | |
347 | if (len < step) | |
348 | len = 0; | |
349 | else | |
350 | len -= step; | |
351 | } | |
352 | } | |
353 | #endif | |
354 | ||
0f9dd46c JB |
355 | /* |
356 | * this is only called by cache_block_group, since we could have freed extents | |
357 | * we need to check the pinned_extents for any extents that can't be used yet | |
358 | * since their free space will be released as soon as the transaction commits. | |
359 | */ | |
817d52f8 | 360 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
361 | struct btrfs_fs_info *info, u64 start, u64 end) |
362 | { | |
817d52f8 | 363 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
364 | int ret; |
365 | ||
366 | while (start < end) { | |
11833d66 | 367 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 368 | &extent_start, &extent_end, |
e6138876 JB |
369 | EXTENT_DIRTY | EXTENT_UPTODATE, |
370 | NULL); | |
0f9dd46c JB |
371 | if (ret) |
372 | break; | |
373 | ||
06b2331f | 374 | if (extent_start <= start) { |
0f9dd46c JB |
375 | start = extent_end + 1; |
376 | } else if (extent_start > start && extent_start < end) { | |
377 | size = extent_start - start; | |
817d52f8 | 378 | total_added += size; |
ea6a478e JB |
379 | ret = btrfs_add_free_space(block_group, start, |
380 | size); | |
79787eaa | 381 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
382 | start = extent_end + 1; |
383 | } else { | |
384 | break; | |
385 | } | |
386 | } | |
387 | ||
388 | if (start < end) { | |
389 | size = end - start; | |
817d52f8 | 390 | total_added += size; |
ea6a478e | 391 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 392 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
393 | } |
394 | ||
817d52f8 | 395 | return total_added; |
0f9dd46c JB |
396 | } |
397 | ||
d458b054 | 398 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 399 | { |
bab39bf9 JB |
400 | struct btrfs_block_group_cache *block_group; |
401 | struct btrfs_fs_info *fs_info; | |
402 | struct btrfs_caching_control *caching_ctl; | |
403 | struct btrfs_root *extent_root; | |
e37c9e69 | 404 | struct btrfs_path *path; |
5f39d397 | 405 | struct extent_buffer *leaf; |
11833d66 | 406 | struct btrfs_key key; |
817d52f8 | 407 | u64 total_found = 0; |
11833d66 YZ |
408 | u64 last = 0; |
409 | u32 nritems; | |
36cce922 | 410 | int ret = -ENOMEM; |
d0bd4560 | 411 | bool wakeup = true; |
f510cfec | 412 | |
bab39bf9 JB |
413 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
414 | block_group = caching_ctl->block_group; | |
415 | fs_info = block_group->fs_info; | |
416 | extent_root = fs_info->extent_root; | |
417 | ||
e37c9e69 CM |
418 | path = btrfs_alloc_path(); |
419 | if (!path) | |
bab39bf9 | 420 | goto out; |
7d7d6068 | 421 | |
817d52f8 | 422 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 423 | |
d0bd4560 JB |
424 | #ifdef CONFIG_BTRFS_DEBUG |
425 | /* | |
426 | * If we're fragmenting we don't want to make anybody think we can | |
427 | * allocate from this block group until we've had a chance to fragment | |
428 | * the free space. | |
429 | */ | |
430 | if (btrfs_should_fragment_free_space(extent_root, block_group)) | |
431 | wakeup = false; | |
432 | #endif | |
5cd57b2c | 433 | /* |
817d52f8 JB |
434 | * We don't want to deadlock with somebody trying to allocate a new |
435 | * extent for the extent root while also trying to search the extent | |
436 | * root to add free space. So we skip locking and search the commit | |
437 | * root, since its read-only | |
5cd57b2c CM |
438 | */ |
439 | path->skip_locking = 1; | |
817d52f8 | 440 | path->search_commit_root = 1; |
026fd317 | 441 | path->reada = 1; |
817d52f8 | 442 | |
e4404d6e | 443 | key.objectid = last; |
e37c9e69 | 444 | key.offset = 0; |
11833d66 | 445 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 446 | again: |
11833d66 | 447 | mutex_lock(&caching_ctl->mutex); |
013f1b12 | 448 | /* need to make sure the commit_root doesn't disappear */ |
9e351cc8 | 449 | down_read(&fs_info->commit_root_sem); |
013f1b12 | 450 | |
52ee28d2 | 451 | next: |
11833d66 | 452 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 453 | if (ret < 0) |
ef8bbdfe | 454 | goto err; |
a512bbf8 | 455 | |
11833d66 YZ |
456 | leaf = path->nodes[0]; |
457 | nritems = btrfs_header_nritems(leaf); | |
458 | ||
d397712b | 459 | while (1) { |
7841cb28 | 460 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 461 | last = (u64)-1; |
817d52f8 | 462 | break; |
f25784b3 | 463 | } |
817d52f8 | 464 | |
11833d66 YZ |
465 | if (path->slots[0] < nritems) { |
466 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
467 | } else { | |
468 | ret = find_next_key(path, 0, &key); | |
469 | if (ret) | |
e37c9e69 | 470 | break; |
817d52f8 | 471 | |
c9ea7b24 | 472 | if (need_resched() || |
9e351cc8 | 473 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
474 | if (wakeup) |
475 | caching_ctl->progress = last; | |
ff5714cc | 476 | btrfs_release_path(path); |
9e351cc8 | 477 | up_read(&fs_info->commit_root_sem); |
589d8ade | 478 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 479 | cond_resched(); |
589d8ade JB |
480 | goto again; |
481 | } | |
0a3896d0 JB |
482 | |
483 | ret = btrfs_next_leaf(extent_root, path); | |
484 | if (ret < 0) | |
485 | goto err; | |
486 | if (ret) | |
487 | break; | |
589d8ade JB |
488 | leaf = path->nodes[0]; |
489 | nritems = btrfs_header_nritems(leaf); | |
490 | continue; | |
11833d66 | 491 | } |
817d52f8 | 492 | |
52ee28d2 LB |
493 | if (key.objectid < last) { |
494 | key.objectid = last; | |
495 | key.offset = 0; | |
496 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
497 | ||
d0bd4560 JB |
498 | if (wakeup) |
499 | caching_ctl->progress = last; | |
52ee28d2 LB |
500 | btrfs_release_path(path); |
501 | goto next; | |
502 | } | |
503 | ||
11833d66 YZ |
504 | if (key.objectid < block_group->key.objectid) { |
505 | path->slots[0]++; | |
817d52f8 | 506 | continue; |
e37c9e69 | 507 | } |
0f9dd46c | 508 | |
e37c9e69 | 509 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 510 | block_group->key.offset) |
e37c9e69 | 511 | break; |
7d7d6068 | 512 | |
3173a18f JB |
513 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
514 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
515 | total_found += add_new_free_space(block_group, |
516 | fs_info, last, | |
517 | key.objectid); | |
3173a18f JB |
518 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
519 | last = key.objectid + | |
707e8a07 | 520 | fs_info->tree_root->nodesize; |
3173a18f JB |
521 | else |
522 | last = key.objectid + key.offset; | |
817d52f8 | 523 | |
11833d66 YZ |
524 | if (total_found > (1024 * 1024 * 2)) { |
525 | total_found = 0; | |
d0bd4560 JB |
526 | if (wakeup) |
527 | wake_up(&caching_ctl->wait); | |
11833d66 | 528 | } |
817d52f8 | 529 | } |
e37c9e69 CM |
530 | path->slots[0]++; |
531 | } | |
817d52f8 | 532 | ret = 0; |
e37c9e69 | 533 | |
817d52f8 JB |
534 | total_found += add_new_free_space(block_group, fs_info, last, |
535 | block_group->key.objectid + | |
536 | block_group->key.offset); | |
817d52f8 | 537 | spin_lock(&block_group->lock); |
11833d66 | 538 | block_group->caching_ctl = NULL; |
817d52f8 JB |
539 | block_group->cached = BTRFS_CACHE_FINISHED; |
540 | spin_unlock(&block_group->lock); | |
0f9dd46c | 541 | |
d0bd4560 JB |
542 | #ifdef CONFIG_BTRFS_DEBUG |
543 | if (btrfs_should_fragment_free_space(extent_root, block_group)) { | |
544 | u64 bytes_used; | |
545 | ||
546 | spin_lock(&block_group->space_info->lock); | |
547 | spin_lock(&block_group->lock); | |
548 | bytes_used = block_group->key.offset - | |
549 | btrfs_block_group_used(&block_group->item); | |
550 | block_group->space_info->bytes_used += bytes_used >> 1; | |
551 | spin_unlock(&block_group->lock); | |
552 | spin_unlock(&block_group->space_info->lock); | |
553 | fragment_free_space(extent_root, block_group); | |
554 | } | |
555 | #endif | |
556 | ||
557 | caching_ctl->progress = (u64)-1; | |
54aa1f4d | 558 | err: |
e37c9e69 | 559 | btrfs_free_path(path); |
9e351cc8 | 560 | up_read(&fs_info->commit_root_sem); |
817d52f8 | 561 | |
11833d66 YZ |
562 | free_excluded_extents(extent_root, block_group); |
563 | ||
564 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 565 | out: |
36cce922 JB |
566 | if (ret) { |
567 | spin_lock(&block_group->lock); | |
568 | block_group->caching_ctl = NULL; | |
569 | block_group->cached = BTRFS_CACHE_ERROR; | |
570 | spin_unlock(&block_group->lock); | |
571 | } | |
11833d66 YZ |
572 | wake_up(&caching_ctl->wait); |
573 | ||
574 | put_caching_control(caching_ctl); | |
11dfe35a | 575 | btrfs_put_block_group(block_group); |
817d52f8 JB |
576 | } |
577 | ||
9d66e233 | 578 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 579 | int load_cache_only) |
817d52f8 | 580 | { |
291c7d2f | 581 | DEFINE_WAIT(wait); |
11833d66 YZ |
582 | struct btrfs_fs_info *fs_info = cache->fs_info; |
583 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
584 | int ret = 0; |
585 | ||
291c7d2f | 586 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
587 | if (!caching_ctl) |
588 | return -ENOMEM; | |
291c7d2f JB |
589 | |
590 | INIT_LIST_HEAD(&caching_ctl->list); | |
591 | mutex_init(&caching_ctl->mutex); | |
592 | init_waitqueue_head(&caching_ctl->wait); | |
593 | caching_ctl->block_group = cache; | |
594 | caching_ctl->progress = cache->key.objectid; | |
595 | atomic_set(&caching_ctl->count, 1); | |
9e0af237 LB |
596 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
597 | caching_thread, NULL, NULL); | |
291c7d2f JB |
598 | |
599 | spin_lock(&cache->lock); | |
600 | /* | |
601 | * This should be a rare occasion, but this could happen I think in the | |
602 | * case where one thread starts to load the space cache info, and then | |
603 | * some other thread starts a transaction commit which tries to do an | |
604 | * allocation while the other thread is still loading the space cache | |
605 | * info. The previous loop should have kept us from choosing this block | |
606 | * group, but if we've moved to the state where we will wait on caching | |
607 | * block groups we need to first check if we're doing a fast load here, | |
608 | * so we can wait for it to finish, otherwise we could end up allocating | |
609 | * from a block group who's cache gets evicted for one reason or | |
610 | * another. | |
611 | */ | |
612 | while (cache->cached == BTRFS_CACHE_FAST) { | |
613 | struct btrfs_caching_control *ctl; | |
614 | ||
615 | ctl = cache->caching_ctl; | |
616 | atomic_inc(&ctl->count); | |
617 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
618 | spin_unlock(&cache->lock); | |
619 | ||
620 | schedule(); | |
621 | ||
622 | finish_wait(&ctl->wait, &wait); | |
623 | put_caching_control(ctl); | |
624 | spin_lock(&cache->lock); | |
625 | } | |
626 | ||
627 | if (cache->cached != BTRFS_CACHE_NO) { | |
628 | spin_unlock(&cache->lock); | |
629 | kfree(caching_ctl); | |
11833d66 | 630 | return 0; |
291c7d2f JB |
631 | } |
632 | WARN_ON(cache->caching_ctl); | |
633 | cache->caching_ctl = caching_ctl; | |
634 | cache->cached = BTRFS_CACHE_FAST; | |
635 | spin_unlock(&cache->lock); | |
11833d66 | 636 | |
d53ba474 | 637 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
cb83b7b8 | 638 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
639 | ret = load_free_space_cache(fs_info, cache); |
640 | ||
641 | spin_lock(&cache->lock); | |
642 | if (ret == 1) { | |
291c7d2f | 643 | cache->caching_ctl = NULL; |
9d66e233 JB |
644 | cache->cached = BTRFS_CACHE_FINISHED; |
645 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 646 | caching_ctl->progress = (u64)-1; |
9d66e233 | 647 | } else { |
291c7d2f JB |
648 | if (load_cache_only) { |
649 | cache->caching_ctl = NULL; | |
650 | cache->cached = BTRFS_CACHE_NO; | |
651 | } else { | |
652 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 653 | cache->has_caching_ctl = 1; |
291c7d2f | 654 | } |
9d66e233 JB |
655 | } |
656 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
657 | #ifdef CONFIG_BTRFS_DEBUG |
658 | if (ret == 1 && | |
659 | btrfs_should_fragment_free_space(fs_info->extent_root, | |
660 | cache)) { | |
661 | u64 bytes_used; | |
662 | ||
663 | spin_lock(&cache->space_info->lock); | |
664 | spin_lock(&cache->lock); | |
665 | bytes_used = cache->key.offset - | |
666 | btrfs_block_group_used(&cache->item); | |
667 | cache->space_info->bytes_used += bytes_used >> 1; | |
668 | spin_unlock(&cache->lock); | |
669 | spin_unlock(&cache->space_info->lock); | |
670 | fragment_free_space(fs_info->extent_root, cache); | |
671 | } | |
672 | #endif | |
cb83b7b8 JB |
673 | mutex_unlock(&caching_ctl->mutex); |
674 | ||
291c7d2f | 675 | wake_up(&caching_ctl->wait); |
3c14874a | 676 | if (ret == 1) { |
291c7d2f | 677 | put_caching_control(caching_ctl); |
3c14874a | 678 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 679 | return 0; |
3c14874a | 680 | } |
291c7d2f JB |
681 | } else { |
682 | /* | |
683 | * We are not going to do the fast caching, set cached to the | |
684 | * appropriate value and wakeup any waiters. | |
685 | */ | |
686 | spin_lock(&cache->lock); | |
687 | if (load_cache_only) { | |
688 | cache->caching_ctl = NULL; | |
689 | cache->cached = BTRFS_CACHE_NO; | |
690 | } else { | |
691 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 692 | cache->has_caching_ctl = 1; |
291c7d2f JB |
693 | } |
694 | spin_unlock(&cache->lock); | |
695 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
696 | } |
697 | ||
291c7d2f JB |
698 | if (load_cache_only) { |
699 | put_caching_control(caching_ctl); | |
11833d66 | 700 | return 0; |
817d52f8 | 701 | } |
817d52f8 | 702 | |
9e351cc8 | 703 | down_write(&fs_info->commit_root_sem); |
291c7d2f | 704 | atomic_inc(&caching_ctl->count); |
11833d66 | 705 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 706 | up_write(&fs_info->commit_root_sem); |
11833d66 | 707 | |
11dfe35a | 708 | btrfs_get_block_group(cache); |
11833d66 | 709 | |
e66f0bb1 | 710 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 711 | |
ef8bbdfe | 712 | return ret; |
e37c9e69 CM |
713 | } |
714 | ||
0f9dd46c JB |
715 | /* |
716 | * return the block group that starts at or after bytenr | |
717 | */ | |
d397712b CM |
718 | static struct btrfs_block_group_cache * |
719 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 720 | { |
0f9dd46c | 721 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 722 | |
0f9dd46c | 723 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 724 | |
0f9dd46c | 725 | return cache; |
0ef3e66b CM |
726 | } |
727 | ||
0f9dd46c | 728 | /* |
9f55684c | 729 | * return the block group that contains the given bytenr |
0f9dd46c | 730 | */ |
d397712b CM |
731 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
732 | struct btrfs_fs_info *info, | |
733 | u64 bytenr) | |
be744175 | 734 | { |
0f9dd46c | 735 | struct btrfs_block_group_cache *cache; |
be744175 | 736 | |
0f9dd46c | 737 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 738 | |
0f9dd46c | 739 | return cache; |
be744175 | 740 | } |
0b86a832 | 741 | |
0f9dd46c JB |
742 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
743 | u64 flags) | |
6324fbf3 | 744 | { |
0f9dd46c | 745 | struct list_head *head = &info->space_info; |
0f9dd46c | 746 | struct btrfs_space_info *found; |
4184ea7f | 747 | |
52ba6929 | 748 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 749 | |
4184ea7f CM |
750 | rcu_read_lock(); |
751 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 752 | if (found->flags & flags) { |
4184ea7f | 753 | rcu_read_unlock(); |
0f9dd46c | 754 | return found; |
4184ea7f | 755 | } |
0f9dd46c | 756 | } |
4184ea7f | 757 | rcu_read_unlock(); |
0f9dd46c | 758 | return NULL; |
6324fbf3 CM |
759 | } |
760 | ||
4184ea7f CM |
761 | /* |
762 | * after adding space to the filesystem, we need to clear the full flags | |
763 | * on all the space infos. | |
764 | */ | |
765 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
766 | { | |
767 | struct list_head *head = &info->space_info; | |
768 | struct btrfs_space_info *found; | |
769 | ||
770 | rcu_read_lock(); | |
771 | list_for_each_entry_rcu(found, head, list) | |
772 | found->full = 0; | |
773 | rcu_read_unlock(); | |
774 | } | |
775 | ||
1a4ed8fd FM |
776 | /* simple helper to search for an existing data extent at a given offset */ |
777 | int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len) | |
e02119d5 CM |
778 | { |
779 | int ret; | |
780 | struct btrfs_key key; | |
31840ae1 | 781 | struct btrfs_path *path; |
e02119d5 | 782 | |
31840ae1 | 783 | path = btrfs_alloc_path(); |
d8926bb3 MF |
784 | if (!path) |
785 | return -ENOMEM; | |
786 | ||
e02119d5 CM |
787 | key.objectid = start; |
788 | key.offset = len; | |
3173a18f | 789 | key.type = BTRFS_EXTENT_ITEM_KEY; |
e02119d5 CM |
790 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, |
791 | 0, 0); | |
31840ae1 | 792 | btrfs_free_path(path); |
7bb86316 CM |
793 | return ret; |
794 | } | |
795 | ||
a22285a6 | 796 | /* |
3173a18f | 797 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
798 | * |
799 | * the head node for delayed ref is used to store the sum of all the | |
800 | * reference count modifications queued up in the rbtree. the head | |
801 | * node may also store the extent flags to set. This way you can check | |
802 | * to see what the reference count and extent flags would be if all of | |
803 | * the delayed refs are not processed. | |
804 | */ | |
805 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
806 | struct btrfs_root *root, u64 bytenr, | |
3173a18f | 807 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
808 | { |
809 | struct btrfs_delayed_ref_head *head; | |
810 | struct btrfs_delayed_ref_root *delayed_refs; | |
811 | struct btrfs_path *path; | |
812 | struct btrfs_extent_item *ei; | |
813 | struct extent_buffer *leaf; | |
814 | struct btrfs_key key; | |
815 | u32 item_size; | |
816 | u64 num_refs; | |
817 | u64 extent_flags; | |
818 | int ret; | |
819 | ||
3173a18f JB |
820 | /* |
821 | * If we don't have skinny metadata, don't bother doing anything | |
822 | * different | |
823 | */ | |
824 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { | |
707e8a07 | 825 | offset = root->nodesize; |
3173a18f JB |
826 | metadata = 0; |
827 | } | |
828 | ||
a22285a6 YZ |
829 | path = btrfs_alloc_path(); |
830 | if (!path) | |
831 | return -ENOMEM; | |
832 | ||
a22285a6 YZ |
833 | if (!trans) { |
834 | path->skip_locking = 1; | |
835 | path->search_commit_root = 1; | |
836 | } | |
639eefc8 FDBM |
837 | |
838 | search_again: | |
839 | key.objectid = bytenr; | |
840 | key.offset = offset; | |
841 | if (metadata) | |
842 | key.type = BTRFS_METADATA_ITEM_KEY; | |
843 | else | |
844 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
845 | ||
a22285a6 YZ |
846 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, |
847 | &key, path, 0, 0); | |
848 | if (ret < 0) | |
849 | goto out_free; | |
850 | ||
3173a18f | 851 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
852 | if (path->slots[0]) { |
853 | path->slots[0]--; | |
854 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
855 | path->slots[0]); | |
856 | if (key.objectid == bytenr && | |
857 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
707e8a07 | 858 | key.offset == root->nodesize) |
74be9510 FDBM |
859 | ret = 0; |
860 | } | |
3173a18f JB |
861 | } |
862 | ||
a22285a6 YZ |
863 | if (ret == 0) { |
864 | leaf = path->nodes[0]; | |
865 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
866 | if (item_size >= sizeof(*ei)) { | |
867 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
868 | struct btrfs_extent_item); | |
869 | num_refs = btrfs_extent_refs(leaf, ei); | |
870 | extent_flags = btrfs_extent_flags(leaf, ei); | |
871 | } else { | |
872 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
873 | struct btrfs_extent_item_v0 *ei0; | |
874 | BUG_ON(item_size != sizeof(*ei0)); | |
875 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
876 | struct btrfs_extent_item_v0); | |
877 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
878 | /* FIXME: this isn't correct for data */ | |
879 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
880 | #else | |
881 | BUG(); | |
882 | #endif | |
883 | } | |
884 | BUG_ON(num_refs == 0); | |
885 | } else { | |
886 | num_refs = 0; | |
887 | extent_flags = 0; | |
888 | ret = 0; | |
889 | } | |
890 | ||
891 | if (!trans) | |
892 | goto out; | |
893 | ||
894 | delayed_refs = &trans->transaction->delayed_refs; | |
895 | spin_lock(&delayed_refs->lock); | |
896 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
897 | if (head) { | |
898 | if (!mutex_trylock(&head->mutex)) { | |
899 | atomic_inc(&head->node.refs); | |
900 | spin_unlock(&delayed_refs->lock); | |
901 | ||
b3b4aa74 | 902 | btrfs_release_path(path); |
a22285a6 | 903 | |
8cc33e5c DS |
904 | /* |
905 | * Mutex was contended, block until it's released and try | |
906 | * again | |
907 | */ | |
a22285a6 YZ |
908 | mutex_lock(&head->mutex); |
909 | mutex_unlock(&head->mutex); | |
910 | btrfs_put_delayed_ref(&head->node); | |
639eefc8 | 911 | goto search_again; |
a22285a6 | 912 | } |
d7df2c79 | 913 | spin_lock(&head->lock); |
a22285a6 YZ |
914 | if (head->extent_op && head->extent_op->update_flags) |
915 | extent_flags |= head->extent_op->flags_to_set; | |
916 | else | |
917 | BUG_ON(num_refs == 0); | |
918 | ||
919 | num_refs += head->node.ref_mod; | |
d7df2c79 | 920 | spin_unlock(&head->lock); |
a22285a6 YZ |
921 | mutex_unlock(&head->mutex); |
922 | } | |
923 | spin_unlock(&delayed_refs->lock); | |
924 | out: | |
925 | WARN_ON(num_refs == 0); | |
926 | if (refs) | |
927 | *refs = num_refs; | |
928 | if (flags) | |
929 | *flags = extent_flags; | |
930 | out_free: | |
931 | btrfs_free_path(path); | |
932 | return ret; | |
933 | } | |
934 | ||
d8d5f3e1 CM |
935 | /* |
936 | * Back reference rules. Back refs have three main goals: | |
937 | * | |
938 | * 1) differentiate between all holders of references to an extent so that | |
939 | * when a reference is dropped we can make sure it was a valid reference | |
940 | * before freeing the extent. | |
941 | * | |
942 | * 2) Provide enough information to quickly find the holders of an extent | |
943 | * if we notice a given block is corrupted or bad. | |
944 | * | |
945 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
946 | * maintenance. This is actually the same as #2, but with a slightly | |
947 | * different use case. | |
948 | * | |
5d4f98a2 YZ |
949 | * There are two kinds of back refs. The implicit back refs is optimized |
950 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
951 | * back refs of this kind provide information about the block's owner tree | |
952 | * and the pointer's key. These information allow us to find the block by | |
953 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
954 | * referenced by their owner trees. The location of tree block is recorded | |
955 | * in the back refs. Actually the full back refs is generic, and can be | |
956 | * used in all cases the implicit back refs is used. The major shortcoming | |
957 | * of the full back refs is its overhead. Every time a tree block gets | |
958 | * COWed, we have to update back refs entry for all pointers in it. | |
959 | * | |
960 | * For a newly allocated tree block, we use implicit back refs for | |
961 | * pointers in it. This means most tree related operations only involve | |
962 | * implicit back refs. For a tree block created in old transaction, the | |
963 | * only way to drop a reference to it is COW it. So we can detect the | |
964 | * event that tree block loses its owner tree's reference and do the | |
965 | * back refs conversion. | |
966 | * | |
967 | * When a tree block is COW'd through a tree, there are four cases: | |
968 | * | |
969 | * The reference count of the block is one and the tree is the block's | |
970 | * owner tree. Nothing to do in this case. | |
971 | * | |
972 | * The reference count of the block is one and the tree is not the | |
973 | * block's owner tree. In this case, full back refs is used for pointers | |
974 | * in the block. Remove these full back refs, add implicit back refs for | |
975 | * every pointers in the new block. | |
976 | * | |
977 | * The reference count of the block is greater than one and the tree is | |
978 | * the block's owner tree. In this case, implicit back refs is used for | |
979 | * pointers in the block. Add full back refs for every pointers in the | |
980 | * block, increase lower level extents' reference counts. The original | |
981 | * implicit back refs are entailed to the new block. | |
982 | * | |
983 | * The reference count of the block is greater than one and the tree is | |
984 | * not the block's owner tree. Add implicit back refs for every pointer in | |
985 | * the new block, increase lower level extents' reference count. | |
986 | * | |
987 | * Back Reference Key composing: | |
988 | * | |
989 | * The key objectid corresponds to the first byte in the extent, | |
990 | * The key type is used to differentiate between types of back refs. | |
991 | * There are different meanings of the key offset for different types | |
992 | * of back refs. | |
993 | * | |
d8d5f3e1 CM |
994 | * File extents can be referenced by: |
995 | * | |
996 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 997 | * - different files inside a single subvolume |
d8d5f3e1 CM |
998 | * - different offsets inside a file (bookend extents in file.c) |
999 | * | |
5d4f98a2 | 1000 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1001 | * |
1002 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1003 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1004 | * - original offset in the file |
1005 | * - how many bookend extents | |
d8d5f3e1 | 1006 | * |
5d4f98a2 YZ |
1007 | * The key offset for the implicit back refs is hash of the first |
1008 | * three fields. | |
d8d5f3e1 | 1009 | * |
5d4f98a2 | 1010 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1011 | * |
5d4f98a2 | 1012 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1013 | * |
5d4f98a2 YZ |
1014 | * The key offset for the implicit back refs is the first byte of |
1015 | * the tree leaf | |
d8d5f3e1 | 1016 | * |
5d4f98a2 YZ |
1017 | * When a file extent is allocated, The implicit back refs is used. |
1018 | * the fields are filled in: | |
d8d5f3e1 | 1019 | * |
5d4f98a2 | 1020 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1021 | * |
5d4f98a2 YZ |
1022 | * When a file extent is removed file truncation, we find the |
1023 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1024 | * |
5d4f98a2 | 1025 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1026 | * |
5d4f98a2 | 1027 | * Btree extents can be referenced by: |
d8d5f3e1 | 1028 | * |
5d4f98a2 | 1029 | * - Different subvolumes |
d8d5f3e1 | 1030 | * |
5d4f98a2 YZ |
1031 | * Both the implicit back refs and the full back refs for tree blocks |
1032 | * only consist of key. The key offset for the implicit back refs is | |
1033 | * objectid of block's owner tree. The key offset for the full back refs | |
1034 | * is the first byte of parent block. | |
d8d5f3e1 | 1035 | * |
5d4f98a2 YZ |
1036 | * When implicit back refs is used, information about the lowest key and |
1037 | * level of the tree block are required. These information are stored in | |
1038 | * tree block info structure. | |
d8d5f3e1 | 1039 | */ |
31840ae1 | 1040 | |
5d4f98a2 YZ |
1041 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1042 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
1043 | struct btrfs_root *root, | |
1044 | struct btrfs_path *path, | |
1045 | u64 owner, u32 extra_size) | |
7bb86316 | 1046 | { |
5d4f98a2 YZ |
1047 | struct btrfs_extent_item *item; |
1048 | struct btrfs_extent_item_v0 *ei0; | |
1049 | struct btrfs_extent_ref_v0 *ref0; | |
1050 | struct btrfs_tree_block_info *bi; | |
1051 | struct extent_buffer *leaf; | |
7bb86316 | 1052 | struct btrfs_key key; |
5d4f98a2 YZ |
1053 | struct btrfs_key found_key; |
1054 | u32 new_size = sizeof(*item); | |
1055 | u64 refs; | |
1056 | int ret; | |
1057 | ||
1058 | leaf = path->nodes[0]; | |
1059 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1060 | ||
1061 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1062 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1063 | struct btrfs_extent_item_v0); | |
1064 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1065 | ||
1066 | if (owner == (u64)-1) { | |
1067 | while (1) { | |
1068 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1069 | ret = btrfs_next_leaf(root, path); | |
1070 | if (ret < 0) | |
1071 | return ret; | |
79787eaa | 1072 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1073 | leaf = path->nodes[0]; |
1074 | } | |
1075 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1076 | path->slots[0]); | |
1077 | BUG_ON(key.objectid != found_key.objectid); | |
1078 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1079 | path->slots[0]++; | |
1080 | continue; | |
1081 | } | |
1082 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1083 | struct btrfs_extent_ref_v0); | |
1084 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1085 | break; | |
1086 | } | |
1087 | } | |
b3b4aa74 | 1088 | btrfs_release_path(path); |
5d4f98a2 YZ |
1089 | |
1090 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1091 | new_size += sizeof(*bi); | |
1092 | ||
1093 | new_size -= sizeof(*ei0); | |
1094 | ret = btrfs_search_slot(trans, root, &key, path, | |
1095 | new_size + extra_size, 1); | |
1096 | if (ret < 0) | |
1097 | return ret; | |
79787eaa | 1098 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1099 | |
4b90c680 | 1100 | btrfs_extend_item(root, path, new_size); |
5d4f98a2 YZ |
1101 | |
1102 | leaf = path->nodes[0]; | |
1103 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1104 | btrfs_set_extent_refs(leaf, item, refs); | |
1105 | /* FIXME: get real generation */ | |
1106 | btrfs_set_extent_generation(leaf, item, 0); | |
1107 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1108 | btrfs_set_extent_flags(leaf, item, | |
1109 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1110 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1111 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1112 | /* FIXME: get first key of the block */ | |
1113 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1114 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1115 | } else { | |
1116 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1117 | } | |
1118 | btrfs_mark_buffer_dirty(leaf); | |
1119 | return 0; | |
1120 | } | |
1121 | #endif | |
1122 | ||
1123 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1124 | { | |
1125 | u32 high_crc = ~(u32)0; | |
1126 | u32 low_crc = ~(u32)0; | |
1127 | __le64 lenum; | |
1128 | ||
1129 | lenum = cpu_to_le64(root_objectid); | |
14a958e6 | 1130 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1131 | lenum = cpu_to_le64(owner); |
14a958e6 | 1132 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1133 | lenum = cpu_to_le64(offset); |
14a958e6 | 1134 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1135 | |
1136 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1137 | } | |
1138 | ||
1139 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1140 | struct btrfs_extent_data_ref *ref) | |
1141 | { | |
1142 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1143 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1144 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1145 | } | |
1146 | ||
1147 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1148 | struct btrfs_extent_data_ref *ref, | |
1149 | u64 root_objectid, u64 owner, u64 offset) | |
1150 | { | |
1151 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1152 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1153 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1154 | return 0; | |
1155 | return 1; | |
1156 | } | |
1157 | ||
1158 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1159 | struct btrfs_root *root, | |
1160 | struct btrfs_path *path, | |
1161 | u64 bytenr, u64 parent, | |
1162 | u64 root_objectid, | |
1163 | u64 owner, u64 offset) | |
1164 | { | |
1165 | struct btrfs_key key; | |
1166 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1167 | struct extent_buffer *leaf; |
5d4f98a2 | 1168 | u32 nritems; |
74493f7a | 1169 | int ret; |
5d4f98a2 YZ |
1170 | int recow; |
1171 | int err = -ENOENT; | |
74493f7a | 1172 | |
31840ae1 | 1173 | key.objectid = bytenr; |
5d4f98a2 YZ |
1174 | if (parent) { |
1175 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1176 | key.offset = parent; | |
1177 | } else { | |
1178 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1179 | key.offset = hash_extent_data_ref(root_objectid, | |
1180 | owner, offset); | |
1181 | } | |
1182 | again: | |
1183 | recow = 0; | |
1184 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1185 | if (ret < 0) { | |
1186 | err = ret; | |
1187 | goto fail; | |
1188 | } | |
31840ae1 | 1189 | |
5d4f98a2 YZ |
1190 | if (parent) { |
1191 | if (!ret) | |
1192 | return 0; | |
1193 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1194 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1195 | btrfs_release_path(path); |
5d4f98a2 YZ |
1196 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1197 | if (ret < 0) { | |
1198 | err = ret; | |
1199 | goto fail; | |
1200 | } | |
1201 | if (!ret) | |
1202 | return 0; | |
1203 | #endif | |
1204 | goto fail; | |
31840ae1 ZY |
1205 | } |
1206 | ||
1207 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1208 | nritems = btrfs_header_nritems(leaf); |
1209 | while (1) { | |
1210 | if (path->slots[0] >= nritems) { | |
1211 | ret = btrfs_next_leaf(root, path); | |
1212 | if (ret < 0) | |
1213 | err = ret; | |
1214 | if (ret) | |
1215 | goto fail; | |
1216 | ||
1217 | leaf = path->nodes[0]; | |
1218 | nritems = btrfs_header_nritems(leaf); | |
1219 | recow = 1; | |
1220 | } | |
1221 | ||
1222 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1223 | if (key.objectid != bytenr || | |
1224 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1225 | goto fail; | |
1226 | ||
1227 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1228 | struct btrfs_extent_data_ref); | |
1229 | ||
1230 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1231 | owner, offset)) { | |
1232 | if (recow) { | |
b3b4aa74 | 1233 | btrfs_release_path(path); |
5d4f98a2 YZ |
1234 | goto again; |
1235 | } | |
1236 | err = 0; | |
1237 | break; | |
1238 | } | |
1239 | path->slots[0]++; | |
31840ae1 | 1240 | } |
5d4f98a2 YZ |
1241 | fail: |
1242 | return err; | |
31840ae1 ZY |
1243 | } |
1244 | ||
5d4f98a2 YZ |
1245 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1246 | struct btrfs_root *root, | |
1247 | struct btrfs_path *path, | |
1248 | u64 bytenr, u64 parent, | |
1249 | u64 root_objectid, u64 owner, | |
1250 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1251 | { |
1252 | struct btrfs_key key; | |
1253 | struct extent_buffer *leaf; | |
5d4f98a2 | 1254 | u32 size; |
31840ae1 ZY |
1255 | u32 num_refs; |
1256 | int ret; | |
74493f7a | 1257 | |
74493f7a | 1258 | key.objectid = bytenr; |
5d4f98a2 YZ |
1259 | if (parent) { |
1260 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1261 | key.offset = parent; | |
1262 | size = sizeof(struct btrfs_shared_data_ref); | |
1263 | } else { | |
1264 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1265 | key.offset = hash_extent_data_ref(root_objectid, | |
1266 | owner, offset); | |
1267 | size = sizeof(struct btrfs_extent_data_ref); | |
1268 | } | |
74493f7a | 1269 | |
5d4f98a2 YZ |
1270 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1271 | if (ret && ret != -EEXIST) | |
1272 | goto fail; | |
1273 | ||
1274 | leaf = path->nodes[0]; | |
1275 | if (parent) { | |
1276 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1277 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1278 | struct btrfs_shared_data_ref); |
1279 | if (ret == 0) { | |
1280 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1281 | } else { | |
1282 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1283 | num_refs += refs_to_add; | |
1284 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1285 | } |
5d4f98a2 YZ |
1286 | } else { |
1287 | struct btrfs_extent_data_ref *ref; | |
1288 | while (ret == -EEXIST) { | |
1289 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1290 | struct btrfs_extent_data_ref); | |
1291 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1292 | owner, offset)) | |
1293 | break; | |
b3b4aa74 | 1294 | btrfs_release_path(path); |
5d4f98a2 YZ |
1295 | key.offset++; |
1296 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1297 | size); | |
1298 | if (ret && ret != -EEXIST) | |
1299 | goto fail; | |
31840ae1 | 1300 | |
5d4f98a2 YZ |
1301 | leaf = path->nodes[0]; |
1302 | } | |
1303 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1304 | struct btrfs_extent_data_ref); | |
1305 | if (ret == 0) { | |
1306 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1307 | root_objectid); | |
1308 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1309 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1310 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1311 | } else { | |
1312 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1313 | num_refs += refs_to_add; | |
1314 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1315 | } |
31840ae1 | 1316 | } |
5d4f98a2 YZ |
1317 | btrfs_mark_buffer_dirty(leaf); |
1318 | ret = 0; | |
1319 | fail: | |
b3b4aa74 | 1320 | btrfs_release_path(path); |
7bb86316 | 1321 | return ret; |
74493f7a CM |
1322 | } |
1323 | ||
5d4f98a2 YZ |
1324 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1325 | struct btrfs_root *root, | |
1326 | struct btrfs_path *path, | |
fcebe456 | 1327 | int refs_to_drop, int *last_ref) |
31840ae1 | 1328 | { |
5d4f98a2 YZ |
1329 | struct btrfs_key key; |
1330 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1331 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1332 | struct extent_buffer *leaf; |
5d4f98a2 | 1333 | u32 num_refs = 0; |
31840ae1 ZY |
1334 | int ret = 0; |
1335 | ||
1336 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1337 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1338 | ||
1339 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1340 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1341 | struct btrfs_extent_data_ref); | |
1342 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1343 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1344 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1345 | struct btrfs_shared_data_ref); | |
1346 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1347 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1348 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1349 | struct btrfs_extent_ref_v0 *ref0; | |
1350 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1351 | struct btrfs_extent_ref_v0); | |
1352 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1353 | #endif | |
1354 | } else { | |
1355 | BUG(); | |
1356 | } | |
1357 | ||
56bec294 CM |
1358 | BUG_ON(num_refs < refs_to_drop); |
1359 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1360 | |
31840ae1 ZY |
1361 | if (num_refs == 0) { |
1362 | ret = btrfs_del_item(trans, root, path); | |
fcebe456 | 1363 | *last_ref = 1; |
31840ae1 | 1364 | } else { |
5d4f98a2 YZ |
1365 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1366 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1367 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1368 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1369 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1370 | else { | |
1371 | struct btrfs_extent_ref_v0 *ref0; | |
1372 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1373 | struct btrfs_extent_ref_v0); | |
1374 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1375 | } | |
1376 | #endif | |
31840ae1 ZY |
1377 | btrfs_mark_buffer_dirty(leaf); |
1378 | } | |
31840ae1 ZY |
1379 | return ret; |
1380 | } | |
1381 | ||
9ed0dea0 | 1382 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1383 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1384 | { |
5d4f98a2 YZ |
1385 | struct btrfs_key key; |
1386 | struct extent_buffer *leaf; | |
1387 | struct btrfs_extent_data_ref *ref1; | |
1388 | struct btrfs_shared_data_ref *ref2; | |
1389 | u32 num_refs = 0; | |
1390 | ||
1391 | leaf = path->nodes[0]; | |
1392 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1393 | if (iref) { | |
1394 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1395 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1396 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1397 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1398 | } else { | |
1399 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1400 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1401 | } | |
1402 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1403 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1404 | struct btrfs_extent_data_ref); | |
1405 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1406 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1407 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1408 | struct btrfs_shared_data_ref); | |
1409 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1410 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1411 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1412 | struct btrfs_extent_ref_v0 *ref0; | |
1413 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1414 | struct btrfs_extent_ref_v0); | |
1415 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1416 | #endif |
5d4f98a2 YZ |
1417 | } else { |
1418 | WARN_ON(1); | |
1419 | } | |
1420 | return num_refs; | |
1421 | } | |
15916de8 | 1422 | |
5d4f98a2 YZ |
1423 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1424 | struct btrfs_root *root, | |
1425 | struct btrfs_path *path, | |
1426 | u64 bytenr, u64 parent, | |
1427 | u64 root_objectid) | |
1f3c79a2 | 1428 | { |
5d4f98a2 | 1429 | struct btrfs_key key; |
1f3c79a2 | 1430 | int ret; |
1f3c79a2 | 1431 | |
5d4f98a2 YZ |
1432 | key.objectid = bytenr; |
1433 | if (parent) { | |
1434 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1435 | key.offset = parent; | |
1436 | } else { | |
1437 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1438 | key.offset = root_objectid; | |
1f3c79a2 LH |
1439 | } |
1440 | ||
5d4f98a2 YZ |
1441 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1442 | if (ret > 0) | |
1443 | ret = -ENOENT; | |
1444 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1445 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1446 | btrfs_release_path(path); |
5d4f98a2 YZ |
1447 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1448 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1449 | if (ret > 0) | |
1450 | ret = -ENOENT; | |
1451 | } | |
1f3c79a2 | 1452 | #endif |
5d4f98a2 | 1453 | return ret; |
1f3c79a2 LH |
1454 | } |
1455 | ||
5d4f98a2 YZ |
1456 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1457 | struct btrfs_root *root, | |
1458 | struct btrfs_path *path, | |
1459 | u64 bytenr, u64 parent, | |
1460 | u64 root_objectid) | |
31840ae1 | 1461 | { |
5d4f98a2 | 1462 | struct btrfs_key key; |
31840ae1 | 1463 | int ret; |
31840ae1 | 1464 | |
5d4f98a2 YZ |
1465 | key.objectid = bytenr; |
1466 | if (parent) { | |
1467 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1468 | key.offset = parent; | |
1469 | } else { | |
1470 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1471 | key.offset = root_objectid; | |
1472 | } | |
1473 | ||
1474 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1475 | btrfs_release_path(path); |
31840ae1 ZY |
1476 | return ret; |
1477 | } | |
1478 | ||
5d4f98a2 | 1479 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1480 | { |
5d4f98a2 YZ |
1481 | int type; |
1482 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1483 | if (parent > 0) | |
1484 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1485 | else | |
1486 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1487 | } else { | |
1488 | if (parent > 0) | |
1489 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1490 | else | |
1491 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1492 | } | |
1493 | return type; | |
31840ae1 | 1494 | } |
56bec294 | 1495 | |
2c47e605 YZ |
1496 | static int find_next_key(struct btrfs_path *path, int level, |
1497 | struct btrfs_key *key) | |
56bec294 | 1498 | |
02217ed2 | 1499 | { |
2c47e605 | 1500 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1501 | if (!path->nodes[level]) |
1502 | break; | |
5d4f98a2 YZ |
1503 | if (path->slots[level] + 1 >= |
1504 | btrfs_header_nritems(path->nodes[level])) | |
1505 | continue; | |
1506 | if (level == 0) | |
1507 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1508 | path->slots[level] + 1); | |
1509 | else | |
1510 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1511 | path->slots[level] + 1); | |
1512 | return 0; | |
1513 | } | |
1514 | return 1; | |
1515 | } | |
037e6390 | 1516 | |
5d4f98a2 YZ |
1517 | /* |
1518 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1519 | * to the address of inline back ref, and 0 is returned. | |
1520 | * | |
1521 | * if back ref isn't found, *ref_ret is set to the address where it | |
1522 | * should be inserted, and -ENOENT is returned. | |
1523 | * | |
1524 | * if insert is true and there are too many inline back refs, the path | |
1525 | * points to the extent item, and -EAGAIN is returned. | |
1526 | * | |
1527 | * NOTE: inline back refs are ordered in the same way that back ref | |
1528 | * items in the tree are ordered. | |
1529 | */ | |
1530 | static noinline_for_stack | |
1531 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1532 | struct btrfs_root *root, | |
1533 | struct btrfs_path *path, | |
1534 | struct btrfs_extent_inline_ref **ref_ret, | |
1535 | u64 bytenr, u64 num_bytes, | |
1536 | u64 parent, u64 root_objectid, | |
1537 | u64 owner, u64 offset, int insert) | |
1538 | { | |
1539 | struct btrfs_key key; | |
1540 | struct extent_buffer *leaf; | |
1541 | struct btrfs_extent_item *ei; | |
1542 | struct btrfs_extent_inline_ref *iref; | |
1543 | u64 flags; | |
1544 | u64 item_size; | |
1545 | unsigned long ptr; | |
1546 | unsigned long end; | |
1547 | int extra_size; | |
1548 | int type; | |
1549 | int want; | |
1550 | int ret; | |
1551 | int err = 0; | |
3173a18f JB |
1552 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
1553 | SKINNY_METADATA); | |
26b8003f | 1554 | |
db94535d | 1555 | key.objectid = bytenr; |
31840ae1 | 1556 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1557 | key.offset = num_bytes; |
31840ae1 | 1558 | |
5d4f98a2 YZ |
1559 | want = extent_ref_type(parent, owner); |
1560 | if (insert) { | |
1561 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1562 | path->keep_locks = 1; |
5d4f98a2 YZ |
1563 | } else |
1564 | extra_size = -1; | |
3173a18f JB |
1565 | |
1566 | /* | |
1567 | * Owner is our parent level, so we can just add one to get the level | |
1568 | * for the block we are interested in. | |
1569 | */ | |
1570 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1571 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1572 | key.offset = owner; | |
1573 | } | |
1574 | ||
1575 | again: | |
5d4f98a2 | 1576 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1577 | if (ret < 0) { |
5d4f98a2 YZ |
1578 | err = ret; |
1579 | goto out; | |
1580 | } | |
3173a18f JB |
1581 | |
1582 | /* | |
1583 | * We may be a newly converted file system which still has the old fat | |
1584 | * extent entries for metadata, so try and see if we have one of those. | |
1585 | */ | |
1586 | if (ret > 0 && skinny_metadata) { | |
1587 | skinny_metadata = false; | |
1588 | if (path->slots[0]) { | |
1589 | path->slots[0]--; | |
1590 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1591 | path->slots[0]); | |
1592 | if (key.objectid == bytenr && | |
1593 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1594 | key.offset == num_bytes) | |
1595 | ret = 0; | |
1596 | } | |
1597 | if (ret) { | |
9ce49a0b | 1598 | key.objectid = bytenr; |
3173a18f JB |
1599 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1600 | key.offset = num_bytes; | |
1601 | btrfs_release_path(path); | |
1602 | goto again; | |
1603 | } | |
1604 | } | |
1605 | ||
79787eaa JM |
1606 | if (ret && !insert) { |
1607 | err = -ENOENT; | |
1608 | goto out; | |
fae7f21c | 1609 | } else if (WARN_ON(ret)) { |
492104c8 | 1610 | err = -EIO; |
492104c8 | 1611 | goto out; |
79787eaa | 1612 | } |
5d4f98a2 YZ |
1613 | |
1614 | leaf = path->nodes[0]; | |
1615 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1616 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1617 | if (item_size < sizeof(*ei)) { | |
1618 | if (!insert) { | |
1619 | err = -ENOENT; | |
1620 | goto out; | |
1621 | } | |
1622 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1623 | extra_size); | |
1624 | if (ret < 0) { | |
1625 | err = ret; | |
1626 | goto out; | |
1627 | } | |
1628 | leaf = path->nodes[0]; | |
1629 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1630 | } | |
1631 | #endif | |
1632 | BUG_ON(item_size < sizeof(*ei)); | |
1633 | ||
5d4f98a2 YZ |
1634 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1635 | flags = btrfs_extent_flags(leaf, ei); | |
1636 | ||
1637 | ptr = (unsigned long)(ei + 1); | |
1638 | end = (unsigned long)ei + item_size; | |
1639 | ||
3173a18f | 1640 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1641 | ptr += sizeof(struct btrfs_tree_block_info); |
1642 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1643 | } |
1644 | ||
1645 | err = -ENOENT; | |
1646 | while (1) { | |
1647 | if (ptr >= end) { | |
1648 | WARN_ON(ptr > end); | |
1649 | break; | |
1650 | } | |
1651 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1652 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1653 | if (want < type) | |
1654 | break; | |
1655 | if (want > type) { | |
1656 | ptr += btrfs_extent_inline_ref_size(type); | |
1657 | continue; | |
1658 | } | |
1659 | ||
1660 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1661 | struct btrfs_extent_data_ref *dref; | |
1662 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1663 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1664 | owner, offset)) { | |
1665 | err = 0; | |
1666 | break; | |
1667 | } | |
1668 | if (hash_extent_data_ref_item(leaf, dref) < | |
1669 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1670 | break; | |
1671 | } else { | |
1672 | u64 ref_offset; | |
1673 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1674 | if (parent > 0) { | |
1675 | if (parent == ref_offset) { | |
1676 | err = 0; | |
1677 | break; | |
1678 | } | |
1679 | if (ref_offset < parent) | |
1680 | break; | |
1681 | } else { | |
1682 | if (root_objectid == ref_offset) { | |
1683 | err = 0; | |
1684 | break; | |
1685 | } | |
1686 | if (ref_offset < root_objectid) | |
1687 | break; | |
1688 | } | |
1689 | } | |
1690 | ptr += btrfs_extent_inline_ref_size(type); | |
1691 | } | |
1692 | if (err == -ENOENT && insert) { | |
1693 | if (item_size + extra_size >= | |
1694 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1695 | err = -EAGAIN; | |
1696 | goto out; | |
1697 | } | |
1698 | /* | |
1699 | * To add new inline back ref, we have to make sure | |
1700 | * there is no corresponding back ref item. | |
1701 | * For simplicity, we just do not add new inline back | |
1702 | * ref if there is any kind of item for this block | |
1703 | */ | |
2c47e605 YZ |
1704 | if (find_next_key(path, 0, &key) == 0 && |
1705 | key.objectid == bytenr && | |
85d4198e | 1706 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1707 | err = -EAGAIN; |
1708 | goto out; | |
1709 | } | |
1710 | } | |
1711 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1712 | out: | |
85d4198e | 1713 | if (insert) { |
5d4f98a2 YZ |
1714 | path->keep_locks = 0; |
1715 | btrfs_unlock_up_safe(path, 1); | |
1716 | } | |
1717 | return err; | |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * helper to add new inline back ref | |
1722 | */ | |
1723 | static noinline_for_stack | |
fd279fae | 1724 | void setup_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1725 | struct btrfs_path *path, |
1726 | struct btrfs_extent_inline_ref *iref, | |
1727 | u64 parent, u64 root_objectid, | |
1728 | u64 owner, u64 offset, int refs_to_add, | |
1729 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1730 | { |
1731 | struct extent_buffer *leaf; | |
1732 | struct btrfs_extent_item *ei; | |
1733 | unsigned long ptr; | |
1734 | unsigned long end; | |
1735 | unsigned long item_offset; | |
1736 | u64 refs; | |
1737 | int size; | |
1738 | int type; | |
5d4f98a2 YZ |
1739 | |
1740 | leaf = path->nodes[0]; | |
1741 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1742 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1743 | ||
1744 | type = extent_ref_type(parent, owner); | |
1745 | size = btrfs_extent_inline_ref_size(type); | |
1746 | ||
4b90c680 | 1747 | btrfs_extend_item(root, path, size); |
5d4f98a2 YZ |
1748 | |
1749 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1750 | refs = btrfs_extent_refs(leaf, ei); | |
1751 | refs += refs_to_add; | |
1752 | btrfs_set_extent_refs(leaf, ei, refs); | |
1753 | if (extent_op) | |
1754 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1755 | ||
1756 | ptr = (unsigned long)ei + item_offset; | |
1757 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1758 | if (ptr < end - size) | |
1759 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1760 | end - size - ptr); | |
1761 | ||
1762 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1763 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1764 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1765 | struct btrfs_extent_data_ref *dref; | |
1766 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1767 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1768 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1769 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1770 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1771 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1772 | struct btrfs_shared_data_ref *sref; | |
1773 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1774 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1775 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1776 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1777 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1778 | } else { | |
1779 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1780 | } | |
1781 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1782 | } |
1783 | ||
1784 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1785 | struct btrfs_root *root, | |
1786 | struct btrfs_path *path, | |
1787 | struct btrfs_extent_inline_ref **ref_ret, | |
1788 | u64 bytenr, u64 num_bytes, u64 parent, | |
1789 | u64 root_objectid, u64 owner, u64 offset) | |
1790 | { | |
1791 | int ret; | |
1792 | ||
1793 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1794 | bytenr, num_bytes, parent, | |
1795 | root_objectid, owner, offset, 0); | |
1796 | if (ret != -ENOENT) | |
54aa1f4d | 1797 | return ret; |
5d4f98a2 | 1798 | |
b3b4aa74 | 1799 | btrfs_release_path(path); |
5d4f98a2 YZ |
1800 | *ref_ret = NULL; |
1801 | ||
1802 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1803 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1804 | root_objectid); | |
1805 | } else { | |
1806 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1807 | root_objectid, owner, offset); | |
b9473439 | 1808 | } |
5d4f98a2 YZ |
1809 | return ret; |
1810 | } | |
31840ae1 | 1811 | |
5d4f98a2 YZ |
1812 | /* |
1813 | * helper to update/remove inline back ref | |
1814 | */ | |
1815 | static noinline_for_stack | |
afe5fea7 | 1816 | void update_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1817 | struct btrfs_path *path, |
1818 | struct btrfs_extent_inline_ref *iref, | |
1819 | int refs_to_mod, | |
fcebe456 JB |
1820 | struct btrfs_delayed_extent_op *extent_op, |
1821 | int *last_ref) | |
5d4f98a2 YZ |
1822 | { |
1823 | struct extent_buffer *leaf; | |
1824 | struct btrfs_extent_item *ei; | |
1825 | struct btrfs_extent_data_ref *dref = NULL; | |
1826 | struct btrfs_shared_data_ref *sref = NULL; | |
1827 | unsigned long ptr; | |
1828 | unsigned long end; | |
1829 | u32 item_size; | |
1830 | int size; | |
1831 | int type; | |
5d4f98a2 YZ |
1832 | u64 refs; |
1833 | ||
1834 | leaf = path->nodes[0]; | |
1835 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1836 | refs = btrfs_extent_refs(leaf, ei); | |
1837 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1838 | refs += refs_to_mod; | |
1839 | btrfs_set_extent_refs(leaf, ei, refs); | |
1840 | if (extent_op) | |
1841 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1842 | ||
1843 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1844 | ||
1845 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1846 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1847 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1848 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1849 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1850 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1851 | } else { | |
1852 | refs = 1; | |
1853 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1854 | } |
31840ae1 | 1855 | |
5d4f98a2 YZ |
1856 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1857 | refs += refs_to_mod; | |
1858 | ||
1859 | if (refs > 0) { | |
1860 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1861 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1862 | else | |
1863 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1864 | } else { | |
fcebe456 | 1865 | *last_ref = 1; |
5d4f98a2 YZ |
1866 | size = btrfs_extent_inline_ref_size(type); |
1867 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1868 | ptr = (unsigned long)iref; | |
1869 | end = (unsigned long)ei + item_size; | |
1870 | if (ptr + size < end) | |
1871 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1872 | end - ptr - size); | |
1873 | item_size -= size; | |
afe5fea7 | 1874 | btrfs_truncate_item(root, path, item_size, 1); |
5d4f98a2 YZ |
1875 | } |
1876 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1877 | } |
1878 | ||
1879 | static noinline_for_stack | |
1880 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1881 | struct btrfs_root *root, | |
1882 | struct btrfs_path *path, | |
1883 | u64 bytenr, u64 num_bytes, u64 parent, | |
1884 | u64 root_objectid, u64 owner, | |
1885 | u64 offset, int refs_to_add, | |
1886 | struct btrfs_delayed_extent_op *extent_op) | |
1887 | { | |
1888 | struct btrfs_extent_inline_ref *iref; | |
1889 | int ret; | |
1890 | ||
1891 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1892 | bytenr, num_bytes, parent, | |
1893 | root_objectid, owner, offset, 1); | |
1894 | if (ret == 0) { | |
1895 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
afe5fea7 | 1896 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1897 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1898 | } else if (ret == -ENOENT) { |
fd279fae | 1899 | setup_inline_extent_backref(root, path, iref, parent, |
143bede5 JM |
1900 | root_objectid, owner, offset, |
1901 | refs_to_add, extent_op); | |
1902 | ret = 0; | |
771ed689 | 1903 | } |
5d4f98a2 YZ |
1904 | return ret; |
1905 | } | |
31840ae1 | 1906 | |
5d4f98a2 YZ |
1907 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1908 | struct btrfs_root *root, | |
1909 | struct btrfs_path *path, | |
1910 | u64 bytenr, u64 parent, u64 root_objectid, | |
1911 | u64 owner, u64 offset, int refs_to_add) | |
1912 | { | |
1913 | int ret; | |
1914 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1915 | BUG_ON(refs_to_add != 1); | |
1916 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1917 | parent, root_objectid); | |
1918 | } else { | |
1919 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1920 | parent, root_objectid, | |
1921 | owner, offset, refs_to_add); | |
1922 | } | |
1923 | return ret; | |
1924 | } | |
56bec294 | 1925 | |
5d4f98a2 YZ |
1926 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1927 | struct btrfs_root *root, | |
1928 | struct btrfs_path *path, | |
1929 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1930 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1931 | { |
143bede5 | 1932 | int ret = 0; |
b9473439 | 1933 | |
5d4f98a2 YZ |
1934 | BUG_ON(!is_data && refs_to_drop != 1); |
1935 | if (iref) { | |
afe5fea7 | 1936 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1937 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 1938 | } else if (is_data) { |
fcebe456 JB |
1939 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop, |
1940 | last_ref); | |
5d4f98a2 | 1941 | } else { |
fcebe456 | 1942 | *last_ref = 1; |
5d4f98a2 YZ |
1943 | ret = btrfs_del_item(trans, root, path); |
1944 | } | |
1945 | return ret; | |
1946 | } | |
1947 | ||
86557861 | 1948 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
1949 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1950 | u64 *discarded_bytes) | |
5d4f98a2 | 1951 | { |
86557861 JM |
1952 | int j, ret = 0; |
1953 | u64 bytes_left, end; | |
4d89d377 | 1954 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1955 | |
4d89d377 JM |
1956 | if (WARN_ON(start != aligned_start)) { |
1957 | len -= aligned_start - start; | |
1958 | len = round_down(len, 1 << 9); | |
1959 | start = aligned_start; | |
1960 | } | |
d04c6b88 | 1961 | |
4d89d377 | 1962 | *discarded_bytes = 0; |
86557861 JM |
1963 | |
1964 | if (!len) | |
1965 | return 0; | |
1966 | ||
1967 | end = start + len; | |
1968 | bytes_left = len; | |
1969 | ||
1970 | /* Skip any superblocks on this device. */ | |
1971 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1972 | u64 sb_start = btrfs_sb_offset(j); | |
1973 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1974 | u64 size = sb_start - start; | |
1975 | ||
1976 | if (!in_range(sb_start, start, bytes_left) && | |
1977 | !in_range(sb_end, start, bytes_left) && | |
1978 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1979 | continue; | |
1980 | ||
1981 | /* | |
1982 | * Superblock spans beginning of range. Adjust start and | |
1983 | * try again. | |
1984 | */ | |
1985 | if (sb_start <= start) { | |
1986 | start += sb_end - start; | |
1987 | if (start > end) { | |
1988 | bytes_left = 0; | |
1989 | break; | |
1990 | } | |
1991 | bytes_left = end - start; | |
1992 | continue; | |
1993 | } | |
1994 | ||
1995 | if (size) { | |
1996 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1997 | GFP_NOFS, 0); | |
1998 | if (!ret) | |
1999 | *discarded_bytes += size; | |
2000 | else if (ret != -EOPNOTSUPP) | |
2001 | return ret; | |
2002 | } | |
2003 | ||
2004 | start = sb_end; | |
2005 | if (start > end) { | |
2006 | bytes_left = 0; | |
2007 | break; | |
2008 | } | |
2009 | bytes_left = end - start; | |
2010 | } | |
2011 | ||
2012 | if (bytes_left) { | |
2013 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2014 | GFP_NOFS, 0); |
2015 | if (!ret) | |
86557861 | 2016 | *discarded_bytes += bytes_left; |
4d89d377 | 2017 | } |
d04c6b88 | 2018 | return ret; |
5d4f98a2 | 2019 | } |
5d4f98a2 | 2020 | |
1edb647b FM |
2021 | int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, |
2022 | u64 num_bytes, u64 *actual_bytes) | |
5d4f98a2 | 2023 | { |
5d4f98a2 | 2024 | int ret; |
5378e607 | 2025 | u64 discarded_bytes = 0; |
a1d3c478 | 2026 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2027 | |
e244a0ae | 2028 | |
5d4f98a2 | 2029 | /* Tell the block device(s) that the sectors can be discarded */ |
3ec706c8 | 2030 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
a1d3c478 | 2031 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 2032 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2033 | if (!ret) { |
a1d3c478 | 2034 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2035 | int i; |
2036 | ||
5d4f98a2 | 2037 | |
a1d3c478 | 2038 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2039 | u64 bytes; |
d5e2003c JB |
2040 | if (!stripe->dev->can_discard) |
2041 | continue; | |
2042 | ||
5378e607 LD |
2043 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2044 | stripe->physical, | |
d04c6b88 JM |
2045 | stripe->length, |
2046 | &bytes); | |
5378e607 | 2047 | if (!ret) |
d04c6b88 | 2048 | discarded_bytes += bytes; |
5378e607 | 2049 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2050 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2051 | |
2052 | /* | |
2053 | * Just in case we get back EOPNOTSUPP for some reason, | |
2054 | * just ignore the return value so we don't screw up | |
2055 | * people calling discard_extent. | |
2056 | */ | |
2057 | ret = 0; | |
5d4f98a2 | 2058 | } |
6e9606d2 | 2059 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2060 | } |
5378e607 LD |
2061 | |
2062 | if (actual_bytes) | |
2063 | *actual_bytes = discarded_bytes; | |
2064 | ||
5d4f98a2 | 2065 | |
53b381b3 DW |
2066 | if (ret == -EOPNOTSUPP) |
2067 | ret = 0; | |
5d4f98a2 | 2068 | return ret; |
5d4f98a2 YZ |
2069 | } |
2070 | ||
79787eaa | 2071 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
2072 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
2073 | struct btrfs_root *root, | |
2074 | u64 bytenr, u64 num_bytes, u64 parent, | |
b06c4bf5 | 2075 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
2076 | { |
2077 | int ret; | |
66d7e7f0 AJ |
2078 | struct btrfs_fs_info *fs_info = root->fs_info; |
2079 | ||
5d4f98a2 YZ |
2080 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2081 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2082 | ||
2083 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
2084 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
2085 | num_bytes, | |
5d4f98a2 | 2086 | parent, root_objectid, (int)owner, |
b06c4bf5 | 2087 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 | 2088 | } else { |
66d7e7f0 | 2089 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5846a3c2 QW |
2090 | num_bytes, parent, root_objectid, |
2091 | owner, offset, 0, | |
b06c4bf5 | 2092 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 YZ |
2093 | } |
2094 | return ret; | |
2095 | } | |
2096 | ||
2097 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2098 | struct btrfs_root *root, | |
c682f9b3 | 2099 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2100 | u64 parent, u64 root_objectid, |
2101 | u64 owner, u64 offset, int refs_to_add, | |
2102 | struct btrfs_delayed_extent_op *extent_op) | |
2103 | { | |
fcebe456 | 2104 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
2105 | struct btrfs_path *path; |
2106 | struct extent_buffer *leaf; | |
2107 | struct btrfs_extent_item *item; | |
fcebe456 | 2108 | struct btrfs_key key; |
c682f9b3 QW |
2109 | u64 bytenr = node->bytenr; |
2110 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2111 | u64 refs; |
2112 | int ret; | |
5d4f98a2 YZ |
2113 | |
2114 | path = btrfs_alloc_path(); | |
2115 | if (!path) | |
2116 | return -ENOMEM; | |
2117 | ||
2118 | path->reada = 1; | |
2119 | path->leave_spinning = 1; | |
2120 | /* this will setup the path even if it fails to insert the back ref */ | |
fcebe456 JB |
2121 | ret = insert_inline_extent_backref(trans, fs_info->extent_root, path, |
2122 | bytenr, num_bytes, parent, | |
5d4f98a2 YZ |
2123 | root_objectid, owner, offset, |
2124 | refs_to_add, extent_op); | |
0ed4792a | 2125 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2126 | goto out; |
fcebe456 JB |
2127 | |
2128 | /* | |
2129 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2130 | * inline extent ref, so just update the reference count and add a | |
2131 | * normal backref. | |
2132 | */ | |
5d4f98a2 | 2133 | leaf = path->nodes[0]; |
fcebe456 | 2134 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2135 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2136 | refs = btrfs_extent_refs(leaf, item); | |
2137 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2138 | if (extent_op) | |
2139 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2140 | |
5d4f98a2 | 2141 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2142 | btrfs_release_path(path); |
56bec294 CM |
2143 | |
2144 | path->reada = 1; | |
b9473439 | 2145 | path->leave_spinning = 1; |
56bec294 CM |
2146 | /* now insert the actual backref */ |
2147 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
2148 | path, bytenr, parent, root_objectid, |
2149 | owner, offset, refs_to_add); | |
79787eaa JM |
2150 | if (ret) |
2151 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 2152 | out: |
56bec294 | 2153 | btrfs_free_path(path); |
30d133fc | 2154 | return ret; |
56bec294 CM |
2155 | } |
2156 | ||
5d4f98a2 YZ |
2157 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2158 | struct btrfs_root *root, | |
2159 | struct btrfs_delayed_ref_node *node, | |
2160 | struct btrfs_delayed_extent_op *extent_op, | |
2161 | int insert_reserved) | |
56bec294 | 2162 | { |
5d4f98a2 YZ |
2163 | int ret = 0; |
2164 | struct btrfs_delayed_data_ref *ref; | |
2165 | struct btrfs_key ins; | |
2166 | u64 parent = 0; | |
2167 | u64 ref_root = 0; | |
2168 | u64 flags = 0; | |
2169 | ||
2170 | ins.objectid = node->bytenr; | |
2171 | ins.offset = node->num_bytes; | |
2172 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2173 | ||
2174 | ref = btrfs_delayed_node_to_data_ref(node); | |
599c75ec LB |
2175 | trace_run_delayed_data_ref(node, ref, node->action); |
2176 | ||
5d4f98a2 YZ |
2177 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2178 | parent = ref->parent; | |
fcebe456 | 2179 | ref_root = ref->root; |
5d4f98a2 YZ |
2180 | |
2181 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2182 | if (extent_op) |
5d4f98a2 | 2183 | flags |= extent_op->flags_to_set; |
5d4f98a2 YZ |
2184 | ret = alloc_reserved_file_extent(trans, root, |
2185 | parent, ref_root, flags, | |
2186 | ref->objectid, ref->offset, | |
2187 | &ins, node->ref_mod); | |
5d4f98a2 | 2188 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
c682f9b3 | 2189 | ret = __btrfs_inc_extent_ref(trans, root, node, parent, |
5d4f98a2 YZ |
2190 | ref_root, ref->objectid, |
2191 | ref->offset, node->ref_mod, | |
c682f9b3 | 2192 | extent_op); |
5d4f98a2 | 2193 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
c682f9b3 | 2194 | ret = __btrfs_free_extent(trans, root, node, parent, |
5d4f98a2 YZ |
2195 | ref_root, ref->objectid, |
2196 | ref->offset, node->ref_mod, | |
c682f9b3 | 2197 | extent_op); |
5d4f98a2 YZ |
2198 | } else { |
2199 | BUG(); | |
2200 | } | |
2201 | return ret; | |
2202 | } | |
2203 | ||
2204 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2205 | struct extent_buffer *leaf, | |
2206 | struct btrfs_extent_item *ei) | |
2207 | { | |
2208 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2209 | if (extent_op->update_flags) { | |
2210 | flags |= extent_op->flags_to_set; | |
2211 | btrfs_set_extent_flags(leaf, ei, flags); | |
2212 | } | |
2213 | ||
2214 | if (extent_op->update_key) { | |
2215 | struct btrfs_tree_block_info *bi; | |
2216 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2217 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2218 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2219 | } | |
2220 | } | |
2221 | ||
2222 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2223 | struct btrfs_root *root, | |
2224 | struct btrfs_delayed_ref_node *node, | |
2225 | struct btrfs_delayed_extent_op *extent_op) | |
2226 | { | |
2227 | struct btrfs_key key; | |
2228 | struct btrfs_path *path; | |
2229 | struct btrfs_extent_item *ei; | |
2230 | struct extent_buffer *leaf; | |
2231 | u32 item_size; | |
56bec294 | 2232 | int ret; |
5d4f98a2 | 2233 | int err = 0; |
b1c79e09 | 2234 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2235 | |
79787eaa JM |
2236 | if (trans->aborted) |
2237 | return 0; | |
2238 | ||
3173a18f JB |
2239 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
2240 | metadata = 0; | |
2241 | ||
5d4f98a2 YZ |
2242 | path = btrfs_alloc_path(); |
2243 | if (!path) | |
2244 | return -ENOMEM; | |
2245 | ||
2246 | key.objectid = node->bytenr; | |
5d4f98a2 | 2247 | |
3173a18f | 2248 | if (metadata) { |
3173a18f | 2249 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2250 | key.offset = extent_op->level; |
3173a18f JB |
2251 | } else { |
2252 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2253 | key.offset = node->num_bytes; | |
2254 | } | |
2255 | ||
2256 | again: | |
5d4f98a2 YZ |
2257 | path->reada = 1; |
2258 | path->leave_spinning = 1; | |
2259 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2260 | path, 0, 1); | |
2261 | if (ret < 0) { | |
2262 | err = ret; | |
2263 | goto out; | |
2264 | } | |
2265 | if (ret > 0) { | |
3173a18f | 2266 | if (metadata) { |
55994887 FDBM |
2267 | if (path->slots[0] > 0) { |
2268 | path->slots[0]--; | |
2269 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2270 | path->slots[0]); | |
2271 | if (key.objectid == node->bytenr && | |
2272 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2273 | key.offset == node->num_bytes) | |
2274 | ret = 0; | |
2275 | } | |
2276 | if (ret > 0) { | |
2277 | btrfs_release_path(path); | |
2278 | metadata = 0; | |
3173a18f | 2279 | |
55994887 FDBM |
2280 | key.objectid = node->bytenr; |
2281 | key.offset = node->num_bytes; | |
2282 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2283 | goto again; | |
2284 | } | |
2285 | } else { | |
2286 | err = -EIO; | |
2287 | goto out; | |
3173a18f | 2288 | } |
5d4f98a2 YZ |
2289 | } |
2290 | ||
2291 | leaf = path->nodes[0]; | |
2292 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2293 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2294 | if (item_size < sizeof(*ei)) { | |
2295 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2296 | path, (u64)-1, 0); | |
2297 | if (ret < 0) { | |
2298 | err = ret; | |
2299 | goto out; | |
2300 | } | |
2301 | leaf = path->nodes[0]; | |
2302 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2303 | } | |
2304 | #endif | |
2305 | BUG_ON(item_size < sizeof(*ei)); | |
2306 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2307 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2308 | |
5d4f98a2 YZ |
2309 | btrfs_mark_buffer_dirty(leaf); |
2310 | out: | |
2311 | btrfs_free_path(path); | |
2312 | return err; | |
56bec294 CM |
2313 | } |
2314 | ||
5d4f98a2 YZ |
2315 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2316 | struct btrfs_root *root, | |
2317 | struct btrfs_delayed_ref_node *node, | |
2318 | struct btrfs_delayed_extent_op *extent_op, | |
2319 | int insert_reserved) | |
56bec294 CM |
2320 | { |
2321 | int ret = 0; | |
5d4f98a2 YZ |
2322 | struct btrfs_delayed_tree_ref *ref; |
2323 | struct btrfs_key ins; | |
2324 | u64 parent = 0; | |
2325 | u64 ref_root = 0; | |
3173a18f JB |
2326 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
2327 | SKINNY_METADATA); | |
56bec294 | 2328 | |
5d4f98a2 | 2329 | ref = btrfs_delayed_node_to_tree_ref(node); |
599c75ec LB |
2330 | trace_run_delayed_tree_ref(node, ref, node->action); |
2331 | ||
5d4f98a2 YZ |
2332 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2333 | parent = ref->parent; | |
fcebe456 | 2334 | ref_root = ref->root; |
5d4f98a2 | 2335 | |
3173a18f JB |
2336 | ins.objectid = node->bytenr; |
2337 | if (skinny_metadata) { | |
2338 | ins.offset = ref->level; | |
2339 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2340 | } else { | |
2341 | ins.offset = node->num_bytes; | |
2342 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2343 | } | |
2344 | ||
5d4f98a2 YZ |
2345 | BUG_ON(node->ref_mod != 1); |
2346 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2347 | BUG_ON(!extent_op || !extent_op->update_flags); |
5d4f98a2 YZ |
2348 | ret = alloc_reserved_tree_block(trans, root, |
2349 | parent, ref_root, | |
2350 | extent_op->flags_to_set, | |
2351 | &extent_op->key, | |
b06c4bf5 | 2352 | ref->level, &ins); |
5d4f98a2 | 2353 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
c682f9b3 QW |
2354 | ret = __btrfs_inc_extent_ref(trans, root, node, |
2355 | parent, ref_root, | |
2356 | ref->level, 0, 1, | |
fcebe456 | 2357 | extent_op); |
5d4f98a2 | 2358 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
c682f9b3 QW |
2359 | ret = __btrfs_free_extent(trans, root, node, |
2360 | parent, ref_root, | |
2361 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2362 | } else { |
2363 | BUG(); | |
2364 | } | |
56bec294 CM |
2365 | return ret; |
2366 | } | |
2367 | ||
2368 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2369 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2370 | struct btrfs_root *root, | |
2371 | struct btrfs_delayed_ref_node *node, | |
2372 | struct btrfs_delayed_extent_op *extent_op, | |
2373 | int insert_reserved) | |
56bec294 | 2374 | { |
79787eaa JM |
2375 | int ret = 0; |
2376 | ||
857cc2fc JB |
2377 | if (trans->aborted) { |
2378 | if (insert_reserved) | |
2379 | btrfs_pin_extent(root, node->bytenr, | |
2380 | node->num_bytes, 1); | |
79787eaa | 2381 | return 0; |
857cc2fc | 2382 | } |
79787eaa | 2383 | |
5d4f98a2 | 2384 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2385 | struct btrfs_delayed_ref_head *head; |
2386 | /* | |
2387 | * we've hit the end of the chain and we were supposed | |
2388 | * to insert this extent into the tree. But, it got | |
2389 | * deleted before we ever needed to insert it, so all | |
2390 | * we have to do is clean up the accounting | |
2391 | */ | |
5d4f98a2 YZ |
2392 | BUG_ON(extent_op); |
2393 | head = btrfs_delayed_node_to_head(node); | |
599c75ec LB |
2394 | trace_run_delayed_ref_head(node, head, node->action); |
2395 | ||
56bec294 | 2396 | if (insert_reserved) { |
f0486c68 YZ |
2397 | btrfs_pin_extent(root, node->bytenr, |
2398 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2399 | if (head->is_data) { |
2400 | ret = btrfs_del_csums(trans, root, | |
2401 | node->bytenr, | |
2402 | node->num_bytes); | |
5d4f98a2 | 2403 | } |
56bec294 | 2404 | } |
297d750b QW |
2405 | |
2406 | /* Also free its reserved qgroup space */ | |
2407 | btrfs_qgroup_free_delayed_ref(root->fs_info, | |
2408 | head->qgroup_ref_root, | |
2409 | head->qgroup_reserved); | |
79787eaa | 2410 | return ret; |
56bec294 CM |
2411 | } |
2412 | ||
5d4f98a2 YZ |
2413 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2414 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2415 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2416 | insert_reserved); | |
2417 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2418 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2419 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2420 | insert_reserved); | |
2421 | else | |
2422 | BUG(); | |
2423 | return ret; | |
56bec294 CM |
2424 | } |
2425 | ||
c6fc2454 | 2426 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2427 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2428 | { | |
cffc3374 FM |
2429 | struct btrfs_delayed_ref_node *ref; |
2430 | ||
c6fc2454 QW |
2431 | if (list_empty(&head->ref_list)) |
2432 | return NULL; | |
d7df2c79 | 2433 | |
cffc3374 FM |
2434 | /* |
2435 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2436 | * This is to prevent a ref count from going down to zero, which deletes | |
2437 | * the extent item from the extent tree, when there still are references | |
2438 | * to add, which would fail because they would not find the extent item. | |
2439 | */ | |
2440 | list_for_each_entry(ref, &head->ref_list, list) { | |
2441 | if (ref->action == BTRFS_ADD_DELAYED_REF) | |
2442 | return ref; | |
2443 | } | |
2444 | ||
c6fc2454 QW |
2445 | return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node, |
2446 | list); | |
56bec294 CM |
2447 | } |
2448 | ||
79787eaa JM |
2449 | /* |
2450 | * Returns 0 on success or if called with an already aborted transaction. | |
2451 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2452 | */ | |
d7df2c79 JB |
2453 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2454 | struct btrfs_root *root, | |
2455 | unsigned long nr) | |
56bec294 | 2456 | { |
56bec294 CM |
2457 | struct btrfs_delayed_ref_root *delayed_refs; |
2458 | struct btrfs_delayed_ref_node *ref; | |
2459 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2460 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2461 | struct btrfs_fs_info *fs_info = root->fs_info; |
0a2b2a84 | 2462 | ktime_t start = ktime_get(); |
56bec294 | 2463 | int ret; |
d7df2c79 | 2464 | unsigned long count = 0; |
0a2b2a84 | 2465 | unsigned long actual_count = 0; |
56bec294 | 2466 | int must_insert_reserved = 0; |
56bec294 CM |
2467 | |
2468 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2469 | while (1) { |
2470 | if (!locked_ref) { | |
d7df2c79 | 2471 | if (count >= nr) |
56bec294 | 2472 | break; |
56bec294 | 2473 | |
d7df2c79 JB |
2474 | spin_lock(&delayed_refs->lock); |
2475 | locked_ref = btrfs_select_ref_head(trans); | |
2476 | if (!locked_ref) { | |
2477 | spin_unlock(&delayed_refs->lock); | |
2478 | break; | |
2479 | } | |
c3e69d58 CM |
2480 | |
2481 | /* grab the lock that says we are going to process | |
2482 | * all the refs for this head */ | |
2483 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2484 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2485 | /* |
2486 | * we may have dropped the spin lock to get the head | |
2487 | * mutex lock, and that might have given someone else | |
2488 | * time to free the head. If that's true, it has been | |
2489 | * removed from our list and we can move on. | |
2490 | */ | |
2491 | if (ret == -EAGAIN) { | |
2492 | locked_ref = NULL; | |
2493 | count++; | |
2494 | continue; | |
56bec294 CM |
2495 | } |
2496 | } | |
a28ec197 | 2497 | |
2c3cf7d5 FM |
2498 | /* |
2499 | * We need to try and merge add/drops of the same ref since we | |
2500 | * can run into issues with relocate dropping the implicit ref | |
2501 | * and then it being added back again before the drop can | |
2502 | * finish. If we merged anything we need to re-loop so we can | |
2503 | * get a good ref. | |
2504 | * Or we can get node references of the same type that weren't | |
2505 | * merged when created due to bumps in the tree mod seq, and | |
2506 | * we need to merge them to prevent adding an inline extent | |
2507 | * backref before dropping it (triggering a BUG_ON at | |
2508 | * insert_inline_extent_backref()). | |
2509 | */ | |
d7df2c79 | 2510 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2511 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2512 | locked_ref); | |
ae1e206b | 2513 | |
d1270cd9 AJ |
2514 | /* |
2515 | * locked_ref is the head node, so we have to go one | |
2516 | * node back for any delayed ref updates | |
2517 | */ | |
2518 | ref = select_delayed_ref(locked_ref); | |
2519 | ||
2520 | if (ref && ref->seq && | |
097b8a7c | 2521 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2522 | spin_unlock(&locked_ref->lock); |
093486c4 | 2523 | btrfs_delayed_ref_unlock(locked_ref); |
d7df2c79 JB |
2524 | spin_lock(&delayed_refs->lock); |
2525 | locked_ref->processing = 0; | |
d1270cd9 AJ |
2526 | delayed_refs->num_heads_ready++; |
2527 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 2528 | locked_ref = NULL; |
d1270cd9 | 2529 | cond_resched(); |
27a377db | 2530 | count++; |
d1270cd9 AJ |
2531 | continue; |
2532 | } | |
2533 | ||
56bec294 CM |
2534 | /* |
2535 | * record the must insert reserved flag before we | |
2536 | * drop the spin lock. | |
2537 | */ | |
2538 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2539 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2540 | |
5d4f98a2 YZ |
2541 | extent_op = locked_ref->extent_op; |
2542 | locked_ref->extent_op = NULL; | |
2543 | ||
56bec294 | 2544 | if (!ref) { |
d7df2c79 JB |
2545 | |
2546 | ||
56bec294 CM |
2547 | /* All delayed refs have been processed, Go ahead |
2548 | * and send the head node to run_one_delayed_ref, | |
2549 | * so that any accounting fixes can happen | |
2550 | */ | |
2551 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2552 | |
2553 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2554 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2555 | extent_op = NULL; |
2556 | } | |
2557 | ||
2558 | if (extent_op) { | |
d7df2c79 | 2559 | spin_unlock(&locked_ref->lock); |
5d4f98a2 YZ |
2560 | ret = run_delayed_extent_op(trans, root, |
2561 | ref, extent_op); | |
78a6184a | 2562 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2563 | |
79787eaa | 2564 | if (ret) { |
857cc2fc JB |
2565 | /* |
2566 | * Need to reset must_insert_reserved if | |
2567 | * there was an error so the abort stuff | |
2568 | * can cleanup the reserved space | |
2569 | * properly. | |
2570 | */ | |
2571 | if (must_insert_reserved) | |
2572 | locked_ref->must_insert_reserved = 1; | |
d7df2c79 | 2573 | locked_ref->processing = 0; |
c2cf52eb | 2574 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); |
093486c4 | 2575 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2576 | return ret; |
2577 | } | |
d7df2c79 | 2578 | continue; |
5d4f98a2 | 2579 | } |
02217ed2 | 2580 | |
d7df2c79 JB |
2581 | /* |
2582 | * Need to drop our head ref lock and re-aqcuire the | |
2583 | * delayed ref lock and then re-check to make sure | |
2584 | * nobody got added. | |
2585 | */ | |
2586 | spin_unlock(&locked_ref->lock); | |
2587 | spin_lock(&delayed_refs->lock); | |
2588 | spin_lock(&locked_ref->lock); | |
c6fc2454 | 2589 | if (!list_empty(&locked_ref->ref_list) || |
573a0755 | 2590 | locked_ref->extent_op) { |
d7df2c79 JB |
2591 | spin_unlock(&locked_ref->lock); |
2592 | spin_unlock(&delayed_refs->lock); | |
2593 | continue; | |
2594 | } | |
2595 | ref->in_tree = 0; | |
2596 | delayed_refs->num_heads--; | |
c46effa6 LB |
2597 | rb_erase(&locked_ref->href_node, |
2598 | &delayed_refs->href_root); | |
d7df2c79 JB |
2599 | spin_unlock(&delayed_refs->lock); |
2600 | } else { | |
0a2b2a84 | 2601 | actual_count++; |
d7df2c79 | 2602 | ref->in_tree = 0; |
c6fc2454 | 2603 | list_del(&ref->list); |
c46effa6 | 2604 | } |
d7df2c79 JB |
2605 | atomic_dec(&delayed_refs->num_entries); |
2606 | ||
093486c4 | 2607 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2608 | /* |
2609 | * when we play the delayed ref, also correct the | |
2610 | * ref_mod on head | |
2611 | */ | |
2612 | switch (ref->action) { | |
2613 | case BTRFS_ADD_DELAYED_REF: | |
2614 | case BTRFS_ADD_DELAYED_EXTENT: | |
2615 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2616 | break; | |
2617 | case BTRFS_DROP_DELAYED_REF: | |
2618 | locked_ref->node.ref_mod += ref->ref_mod; | |
2619 | break; | |
2620 | default: | |
2621 | WARN_ON(1); | |
2622 | } | |
2623 | } | |
d7df2c79 | 2624 | spin_unlock(&locked_ref->lock); |
925baedd | 2625 | |
5d4f98a2 | 2626 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2627 | must_insert_reserved); |
eb099670 | 2628 | |
78a6184a | 2629 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2630 | if (ret) { |
d7df2c79 | 2631 | locked_ref->processing = 0; |
093486c4 MX |
2632 | btrfs_delayed_ref_unlock(locked_ref); |
2633 | btrfs_put_delayed_ref(ref); | |
c2cf52eb | 2634 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret); |
79787eaa JM |
2635 | return ret; |
2636 | } | |
2637 | ||
093486c4 MX |
2638 | /* |
2639 | * If this node is a head, that means all the refs in this head | |
2640 | * have been dealt with, and we will pick the next head to deal | |
2641 | * with, so we must unlock the head and drop it from the cluster | |
2642 | * list before we release it. | |
2643 | */ | |
2644 | if (btrfs_delayed_ref_is_head(ref)) { | |
1262133b JB |
2645 | if (locked_ref->is_data && |
2646 | locked_ref->total_ref_mod < 0) { | |
2647 | spin_lock(&delayed_refs->lock); | |
2648 | delayed_refs->pending_csums -= ref->num_bytes; | |
2649 | spin_unlock(&delayed_refs->lock); | |
2650 | } | |
093486c4 MX |
2651 | btrfs_delayed_ref_unlock(locked_ref); |
2652 | locked_ref = NULL; | |
2653 | } | |
2654 | btrfs_put_delayed_ref(ref); | |
2655 | count++; | |
c3e69d58 | 2656 | cond_resched(); |
c3e69d58 | 2657 | } |
0a2b2a84 JB |
2658 | |
2659 | /* | |
2660 | * We don't want to include ref heads since we can have empty ref heads | |
2661 | * and those will drastically skew our runtime down since we just do | |
2662 | * accounting, no actual extent tree updates. | |
2663 | */ | |
2664 | if (actual_count > 0) { | |
2665 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2666 | u64 avg; | |
2667 | ||
2668 | /* | |
2669 | * We weigh the current average higher than our current runtime | |
2670 | * to avoid large swings in the average. | |
2671 | */ | |
2672 | spin_lock(&delayed_refs->lock); | |
2673 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2674 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2675 | spin_unlock(&delayed_refs->lock); |
2676 | } | |
d7df2c79 | 2677 | return 0; |
c3e69d58 CM |
2678 | } |
2679 | ||
709c0486 AJ |
2680 | #ifdef SCRAMBLE_DELAYED_REFS |
2681 | /* | |
2682 | * Normally delayed refs get processed in ascending bytenr order. This | |
2683 | * correlates in most cases to the order added. To expose dependencies on this | |
2684 | * order, we start to process the tree in the middle instead of the beginning | |
2685 | */ | |
2686 | static u64 find_middle(struct rb_root *root) | |
2687 | { | |
2688 | struct rb_node *n = root->rb_node; | |
2689 | struct btrfs_delayed_ref_node *entry; | |
2690 | int alt = 1; | |
2691 | u64 middle; | |
2692 | u64 first = 0, last = 0; | |
2693 | ||
2694 | n = rb_first(root); | |
2695 | if (n) { | |
2696 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2697 | first = entry->bytenr; | |
2698 | } | |
2699 | n = rb_last(root); | |
2700 | if (n) { | |
2701 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2702 | last = entry->bytenr; | |
2703 | } | |
2704 | n = root->rb_node; | |
2705 | ||
2706 | while (n) { | |
2707 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2708 | WARN_ON(!entry->in_tree); | |
2709 | ||
2710 | middle = entry->bytenr; | |
2711 | ||
2712 | if (alt) | |
2713 | n = n->rb_left; | |
2714 | else | |
2715 | n = n->rb_right; | |
2716 | ||
2717 | alt = 1 - alt; | |
2718 | } | |
2719 | return middle; | |
2720 | } | |
2721 | #endif | |
2722 | ||
1be41b78 JB |
2723 | static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads) |
2724 | { | |
2725 | u64 num_bytes; | |
2726 | ||
2727 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2728 | sizeof(struct btrfs_extent_inline_ref)); | |
2729 | if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) | |
2730 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); | |
2731 | ||
2732 | /* | |
2733 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
2734 | * closer to what we're really going to want to ouse. | |
2735 | */ | |
f8c269d7 | 2736 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root)); |
1be41b78 JB |
2737 | } |
2738 | ||
1262133b JB |
2739 | /* |
2740 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2741 | * would require to store the csums for that many bytes. | |
2742 | */ | |
28f75a0e | 2743 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes) |
1262133b JB |
2744 | { |
2745 | u64 csum_size; | |
2746 | u64 num_csums_per_leaf; | |
2747 | u64 num_csums; | |
2748 | ||
2749 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
2750 | num_csums_per_leaf = div64_u64(csum_size, | |
2751 | (u64)btrfs_super_csum_size(root->fs_info->super_copy)); | |
2752 | num_csums = div64_u64(csum_bytes, root->sectorsize); | |
2753 | num_csums += num_csums_per_leaf - 1; | |
2754 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2755 | return num_csums; | |
2756 | } | |
2757 | ||
0a2b2a84 | 2758 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
1be41b78 JB |
2759 | struct btrfs_root *root) |
2760 | { | |
2761 | struct btrfs_block_rsv *global_rsv; | |
2762 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2763 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
cb723e49 JB |
2764 | u64 num_dirty_bgs = trans->transaction->num_dirty_bgs; |
2765 | u64 num_bytes, num_dirty_bgs_bytes; | |
1be41b78 JB |
2766 | int ret = 0; |
2767 | ||
2768 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
2769 | num_heads = heads_to_leaves(root, num_heads); | |
2770 | if (num_heads > 1) | |
707e8a07 | 2771 | num_bytes += (num_heads - 1) * root->nodesize; |
1be41b78 | 2772 | num_bytes <<= 1; |
28f75a0e | 2773 | num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize; |
cb723e49 JB |
2774 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root, |
2775 | num_dirty_bgs); | |
1be41b78 JB |
2776 | global_rsv = &root->fs_info->global_block_rsv; |
2777 | ||
2778 | /* | |
2779 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2780 | * wiggle room since running delayed refs can create more delayed refs. | |
2781 | */ | |
cb723e49 JB |
2782 | if (global_rsv->space_info->full) { |
2783 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2784 | num_bytes <<= 1; |
cb723e49 | 2785 | } |
1be41b78 JB |
2786 | |
2787 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2788 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2789 | ret = 1; |
2790 | spin_unlock(&global_rsv->lock); | |
2791 | return ret; | |
2792 | } | |
2793 | ||
0a2b2a84 JB |
2794 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2795 | struct btrfs_root *root) | |
2796 | { | |
2797 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2798 | u64 num_entries = | |
2799 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2800 | u64 avg_runtime; | |
a79b7d4b | 2801 | u64 val; |
0a2b2a84 JB |
2802 | |
2803 | smp_mb(); | |
2804 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2805 | val = num_entries * avg_runtime; |
0a2b2a84 JB |
2806 | if (num_entries * avg_runtime >= NSEC_PER_SEC) |
2807 | return 1; | |
a79b7d4b CM |
2808 | if (val >= NSEC_PER_SEC / 2) |
2809 | return 2; | |
0a2b2a84 JB |
2810 | |
2811 | return btrfs_check_space_for_delayed_refs(trans, root); | |
2812 | } | |
2813 | ||
a79b7d4b CM |
2814 | struct async_delayed_refs { |
2815 | struct btrfs_root *root; | |
2816 | int count; | |
2817 | int error; | |
2818 | int sync; | |
2819 | struct completion wait; | |
2820 | struct btrfs_work work; | |
2821 | }; | |
2822 | ||
2823 | static void delayed_ref_async_start(struct btrfs_work *work) | |
2824 | { | |
2825 | struct async_delayed_refs *async; | |
2826 | struct btrfs_trans_handle *trans; | |
2827 | int ret; | |
2828 | ||
2829 | async = container_of(work, struct async_delayed_refs, work); | |
2830 | ||
2831 | trans = btrfs_join_transaction(async->root); | |
2832 | if (IS_ERR(trans)) { | |
2833 | async->error = PTR_ERR(trans); | |
2834 | goto done; | |
2835 | } | |
2836 | ||
2837 | /* | |
2838 | * trans->sync means that when we call end_transaciton, we won't | |
2839 | * wait on delayed refs | |
2840 | */ | |
2841 | trans->sync = true; | |
2842 | ret = btrfs_run_delayed_refs(trans, async->root, async->count); | |
2843 | if (ret) | |
2844 | async->error = ret; | |
2845 | ||
2846 | ret = btrfs_end_transaction(trans, async->root); | |
2847 | if (ret && !async->error) | |
2848 | async->error = ret; | |
2849 | done: | |
2850 | if (async->sync) | |
2851 | complete(&async->wait); | |
2852 | else | |
2853 | kfree(async); | |
2854 | } | |
2855 | ||
2856 | int btrfs_async_run_delayed_refs(struct btrfs_root *root, | |
2857 | unsigned long count, int wait) | |
2858 | { | |
2859 | struct async_delayed_refs *async; | |
2860 | int ret; | |
2861 | ||
2862 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2863 | if (!async) | |
2864 | return -ENOMEM; | |
2865 | ||
2866 | async->root = root->fs_info->tree_root; | |
2867 | async->count = count; | |
2868 | async->error = 0; | |
2869 | if (wait) | |
2870 | async->sync = 1; | |
2871 | else | |
2872 | async->sync = 0; | |
2873 | init_completion(&async->wait); | |
2874 | ||
9e0af237 LB |
2875 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
2876 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b CM |
2877 | |
2878 | btrfs_queue_work(root->fs_info->extent_workers, &async->work); | |
2879 | ||
2880 | if (wait) { | |
2881 | wait_for_completion(&async->wait); | |
2882 | ret = async->error; | |
2883 | kfree(async); | |
2884 | return ret; | |
2885 | } | |
2886 | return 0; | |
2887 | } | |
2888 | ||
c3e69d58 CM |
2889 | /* |
2890 | * this starts processing the delayed reference count updates and | |
2891 | * extent insertions we have queued up so far. count can be | |
2892 | * 0, which means to process everything in the tree at the start | |
2893 | * of the run (but not newly added entries), or it can be some target | |
2894 | * number you'd like to process. | |
79787eaa JM |
2895 | * |
2896 | * Returns 0 on success or if called with an aborted transaction | |
2897 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2898 | */ |
2899 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2900 | struct btrfs_root *root, unsigned long count) | |
2901 | { | |
2902 | struct rb_node *node; | |
2903 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2904 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2905 | int ret; |
2906 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 2907 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 2908 | |
79787eaa JM |
2909 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2910 | if (trans->aborted) | |
2911 | return 0; | |
2912 | ||
c3e69d58 CM |
2913 | if (root == root->fs_info->extent_root) |
2914 | root = root->fs_info->tree_root; | |
2915 | ||
2916 | delayed_refs = &trans->transaction->delayed_refs; | |
26455d33 | 2917 | if (count == 0) |
d7df2c79 | 2918 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2919 | |
c3e69d58 | 2920 | again: |
709c0486 AJ |
2921 | #ifdef SCRAMBLE_DELAYED_REFS |
2922 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2923 | #endif | |
d9a0540a | 2924 | trans->can_flush_pending_bgs = false; |
d7df2c79 JB |
2925 | ret = __btrfs_run_delayed_refs(trans, root, count); |
2926 | if (ret < 0) { | |
2927 | btrfs_abort_transaction(trans, root, ret); | |
2928 | return ret; | |
eb099670 | 2929 | } |
c3e69d58 | 2930 | |
56bec294 | 2931 | if (run_all) { |
d7df2c79 | 2932 | if (!list_empty(&trans->new_bgs)) |
ea658bad | 2933 | btrfs_create_pending_block_groups(trans, root); |
ea658bad | 2934 | |
d7df2c79 | 2935 | spin_lock(&delayed_refs->lock); |
c46effa6 | 2936 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
2937 | if (!node) { |
2938 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2939 | goto out; |
d7df2c79 | 2940 | } |
c3e69d58 | 2941 | count = (unsigned long)-1; |
e9d0b13b | 2942 | |
56bec294 | 2943 | while (node) { |
c46effa6 LB |
2944 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2945 | href_node); | |
2946 | if (btrfs_delayed_ref_is_head(&head->node)) { | |
2947 | struct btrfs_delayed_ref_node *ref; | |
5caf2a00 | 2948 | |
c46effa6 | 2949 | ref = &head->node; |
56bec294 CM |
2950 | atomic_inc(&ref->refs); |
2951 | ||
2952 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2953 | /* |
2954 | * Mutex was contended, block until it's | |
2955 | * released and try again | |
2956 | */ | |
56bec294 CM |
2957 | mutex_lock(&head->mutex); |
2958 | mutex_unlock(&head->mutex); | |
2959 | ||
2960 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2961 | cond_resched(); |
56bec294 | 2962 | goto again; |
c46effa6 LB |
2963 | } else { |
2964 | WARN_ON(1); | |
56bec294 CM |
2965 | } |
2966 | node = rb_next(node); | |
2967 | } | |
2968 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 2969 | cond_resched(); |
56bec294 | 2970 | goto again; |
5f39d397 | 2971 | } |
54aa1f4d | 2972 | out: |
edf39272 | 2973 | assert_qgroups_uptodate(trans); |
d9a0540a | 2974 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
2975 | return 0; |
2976 | } | |
2977 | ||
5d4f98a2 YZ |
2978 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2979 | struct btrfs_root *root, | |
2980 | u64 bytenr, u64 num_bytes, u64 flags, | |
b1c79e09 | 2981 | int level, int is_data) |
5d4f98a2 YZ |
2982 | { |
2983 | struct btrfs_delayed_extent_op *extent_op; | |
2984 | int ret; | |
2985 | ||
78a6184a | 2986 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2987 | if (!extent_op) |
2988 | return -ENOMEM; | |
2989 | ||
2990 | extent_op->flags_to_set = flags; | |
2991 | extent_op->update_flags = 1; | |
2992 | extent_op->update_key = 0; | |
2993 | extent_op->is_data = is_data ? 1 : 0; | |
b1c79e09 | 2994 | extent_op->level = level; |
5d4f98a2 | 2995 | |
66d7e7f0 AJ |
2996 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2997 | num_bytes, extent_op); | |
5d4f98a2 | 2998 | if (ret) |
78a6184a | 2999 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3000 | return ret; |
3001 | } | |
3002 | ||
3003 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
3004 | struct btrfs_root *root, | |
3005 | struct btrfs_path *path, | |
3006 | u64 objectid, u64 offset, u64 bytenr) | |
3007 | { | |
3008 | struct btrfs_delayed_ref_head *head; | |
3009 | struct btrfs_delayed_ref_node *ref; | |
3010 | struct btrfs_delayed_data_ref *data_ref; | |
3011 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 YZ |
3012 | int ret = 0; |
3013 | ||
5d4f98a2 YZ |
3014 | delayed_refs = &trans->transaction->delayed_refs; |
3015 | spin_lock(&delayed_refs->lock); | |
3016 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
d7df2c79 JB |
3017 | if (!head) { |
3018 | spin_unlock(&delayed_refs->lock); | |
3019 | return 0; | |
3020 | } | |
5d4f98a2 YZ |
3021 | |
3022 | if (!mutex_trylock(&head->mutex)) { | |
3023 | atomic_inc(&head->node.refs); | |
3024 | spin_unlock(&delayed_refs->lock); | |
3025 | ||
b3b4aa74 | 3026 | btrfs_release_path(path); |
5d4f98a2 | 3027 | |
8cc33e5c DS |
3028 | /* |
3029 | * Mutex was contended, block until it's released and let | |
3030 | * caller try again | |
3031 | */ | |
5d4f98a2 YZ |
3032 | mutex_lock(&head->mutex); |
3033 | mutex_unlock(&head->mutex); | |
3034 | btrfs_put_delayed_ref(&head->node); | |
3035 | return -EAGAIN; | |
3036 | } | |
d7df2c79 | 3037 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3038 | |
d7df2c79 | 3039 | spin_lock(&head->lock); |
c6fc2454 | 3040 | list_for_each_entry(ref, &head->ref_list, list) { |
d7df2c79 JB |
3041 | /* If it's a shared ref we know a cross reference exists */ |
3042 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3043 | ret = 1; | |
3044 | break; | |
3045 | } | |
5d4f98a2 | 3046 | |
d7df2c79 | 3047 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3048 | |
d7df2c79 JB |
3049 | /* |
3050 | * If our ref doesn't match the one we're currently looking at | |
3051 | * then we have a cross reference. | |
3052 | */ | |
3053 | if (data_ref->root != root->root_key.objectid || | |
3054 | data_ref->objectid != objectid || | |
3055 | data_ref->offset != offset) { | |
3056 | ret = 1; | |
3057 | break; | |
3058 | } | |
5d4f98a2 | 3059 | } |
d7df2c79 | 3060 | spin_unlock(&head->lock); |
5d4f98a2 | 3061 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
3062 | return ret; |
3063 | } | |
3064 | ||
3065 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
3066 | struct btrfs_root *root, | |
3067 | struct btrfs_path *path, | |
3068 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
3069 | { |
3070 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 3071 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3072 | struct btrfs_extent_data_ref *ref; |
3073 | struct btrfs_extent_inline_ref *iref; | |
3074 | struct btrfs_extent_item *ei; | |
f321e491 | 3075 | struct btrfs_key key; |
5d4f98a2 | 3076 | u32 item_size; |
be20aa9d | 3077 | int ret; |
925baedd | 3078 | |
be20aa9d | 3079 | key.objectid = bytenr; |
31840ae1 | 3080 | key.offset = (u64)-1; |
f321e491 | 3081 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3082 | |
be20aa9d CM |
3083 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3084 | if (ret < 0) | |
3085 | goto out; | |
79787eaa | 3086 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3087 | |
3088 | ret = -ENOENT; | |
3089 | if (path->slots[0] == 0) | |
31840ae1 | 3090 | goto out; |
be20aa9d | 3091 | |
31840ae1 | 3092 | path->slots[0]--; |
f321e491 | 3093 | leaf = path->nodes[0]; |
5d4f98a2 | 3094 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3095 | |
5d4f98a2 | 3096 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3097 | goto out; |
f321e491 | 3098 | |
5d4f98a2 YZ |
3099 | ret = 1; |
3100 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3101 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3102 | if (item_size < sizeof(*ei)) { | |
3103 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3104 | goto out; | |
3105 | } | |
3106 | #endif | |
3107 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3108 | |
5d4f98a2 YZ |
3109 | if (item_size != sizeof(*ei) + |
3110 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3111 | goto out; | |
be20aa9d | 3112 | |
5d4f98a2 YZ |
3113 | if (btrfs_extent_generation(leaf, ei) <= |
3114 | btrfs_root_last_snapshot(&root->root_item)) | |
3115 | goto out; | |
3116 | ||
3117 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3118 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
3119 | BTRFS_EXTENT_DATA_REF_KEY) | |
3120 | goto out; | |
3121 | ||
3122 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3123 | if (btrfs_extent_refs(leaf, ei) != | |
3124 | btrfs_extent_data_ref_count(leaf, ref) || | |
3125 | btrfs_extent_data_ref_root(leaf, ref) != | |
3126 | root->root_key.objectid || | |
3127 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3128 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3129 | goto out; | |
3130 | ||
3131 | ret = 0; | |
3132 | out: | |
3133 | return ret; | |
3134 | } | |
3135 | ||
3136 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
3137 | struct btrfs_root *root, | |
3138 | u64 objectid, u64 offset, u64 bytenr) | |
3139 | { | |
3140 | struct btrfs_path *path; | |
3141 | int ret; | |
3142 | int ret2; | |
3143 | ||
3144 | path = btrfs_alloc_path(); | |
3145 | if (!path) | |
3146 | return -ENOENT; | |
3147 | ||
3148 | do { | |
3149 | ret = check_committed_ref(trans, root, path, objectid, | |
3150 | offset, bytenr); | |
3151 | if (ret && ret != -ENOENT) | |
f321e491 | 3152 | goto out; |
80ff3856 | 3153 | |
5d4f98a2 YZ |
3154 | ret2 = check_delayed_ref(trans, root, path, objectid, |
3155 | offset, bytenr); | |
3156 | } while (ret2 == -EAGAIN); | |
3157 | ||
3158 | if (ret2 && ret2 != -ENOENT) { | |
3159 | ret = ret2; | |
3160 | goto out; | |
f321e491 | 3161 | } |
5d4f98a2 YZ |
3162 | |
3163 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3164 | ret = 0; | |
be20aa9d | 3165 | out: |
80ff3856 | 3166 | btrfs_free_path(path); |
f0486c68 YZ |
3167 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3168 | WARN_ON(ret > 0); | |
f321e491 | 3169 | return ret; |
be20aa9d | 3170 | } |
c5739bba | 3171 | |
5d4f98a2 | 3172 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3173 | struct btrfs_root *root, |
5d4f98a2 | 3174 | struct extent_buffer *buf, |
e339a6b0 | 3175 | int full_backref, int inc) |
31840ae1 ZY |
3176 | { |
3177 | u64 bytenr; | |
5d4f98a2 YZ |
3178 | u64 num_bytes; |
3179 | u64 parent; | |
31840ae1 | 3180 | u64 ref_root; |
31840ae1 | 3181 | u32 nritems; |
31840ae1 ZY |
3182 | struct btrfs_key key; |
3183 | struct btrfs_file_extent_item *fi; | |
3184 | int i; | |
3185 | int level; | |
3186 | int ret = 0; | |
31840ae1 | 3187 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
b06c4bf5 | 3188 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3189 | |
fccb84c9 DS |
3190 | |
3191 | if (btrfs_test_is_dummy_root(root)) | |
faa2dbf0 | 3192 | return 0; |
fccb84c9 | 3193 | |
31840ae1 | 3194 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3195 | nritems = btrfs_header_nritems(buf); |
3196 | level = btrfs_header_level(buf); | |
3197 | ||
27cdeb70 | 3198 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3199 | return 0; |
31840ae1 | 3200 | |
5d4f98a2 YZ |
3201 | if (inc) |
3202 | process_func = btrfs_inc_extent_ref; | |
3203 | else | |
3204 | process_func = btrfs_free_extent; | |
31840ae1 | 3205 | |
5d4f98a2 YZ |
3206 | if (full_backref) |
3207 | parent = buf->start; | |
3208 | else | |
3209 | parent = 0; | |
3210 | ||
3211 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3212 | if (level == 0) { |
5d4f98a2 | 3213 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3214 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3215 | continue; |
5d4f98a2 | 3216 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3217 | struct btrfs_file_extent_item); |
3218 | if (btrfs_file_extent_type(buf, fi) == | |
3219 | BTRFS_FILE_EXTENT_INLINE) | |
3220 | continue; | |
3221 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3222 | if (bytenr == 0) | |
3223 | continue; | |
5d4f98a2 YZ |
3224 | |
3225 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3226 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
3227 | ret = process_func(trans, root, bytenr, num_bytes, | |
3228 | parent, ref_root, key.objectid, | |
b06c4bf5 | 3229 | key.offset); |
31840ae1 ZY |
3230 | if (ret) |
3231 | goto fail; | |
3232 | } else { | |
5d4f98a2 | 3233 | bytenr = btrfs_node_blockptr(buf, i); |
707e8a07 | 3234 | num_bytes = root->nodesize; |
5d4f98a2 | 3235 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3236 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3237 | if (ret) |
3238 | goto fail; | |
3239 | } | |
3240 | } | |
3241 | return 0; | |
3242 | fail: | |
5d4f98a2 YZ |
3243 | return ret; |
3244 | } | |
3245 | ||
3246 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3247 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3248 | { |
e339a6b0 | 3249 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3250 | } |
3251 | ||
3252 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3253 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3254 | { |
e339a6b0 | 3255 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3256 | } |
3257 | ||
9078a3e1 CM |
3258 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
3259 | struct btrfs_root *root, | |
3260 | struct btrfs_path *path, | |
3261 | struct btrfs_block_group_cache *cache) | |
3262 | { | |
3263 | int ret; | |
9078a3e1 | 3264 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
3265 | unsigned long bi; |
3266 | struct extent_buffer *leaf; | |
9078a3e1 | 3267 | |
9078a3e1 | 3268 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3269 | if (ret) { |
3270 | if (ret > 0) | |
3271 | ret = -ENOENT; | |
54aa1f4d | 3272 | goto fail; |
df95e7f0 | 3273 | } |
5f39d397 CM |
3274 | |
3275 | leaf = path->nodes[0]; | |
3276 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3277 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3278 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3279 | fail: |
24b89d08 | 3280 | btrfs_release_path(path); |
df95e7f0 | 3281 | return ret; |
9078a3e1 CM |
3282 | |
3283 | } | |
3284 | ||
4a8c9a62 YZ |
3285 | static struct btrfs_block_group_cache * |
3286 | next_block_group(struct btrfs_root *root, | |
3287 | struct btrfs_block_group_cache *cache) | |
3288 | { | |
3289 | struct rb_node *node; | |
292cbd51 | 3290 | |
4a8c9a62 | 3291 | spin_lock(&root->fs_info->block_group_cache_lock); |
292cbd51 FM |
3292 | |
3293 | /* If our block group was removed, we need a full search. */ | |
3294 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3295 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3296 | ||
3297 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3298 | btrfs_put_block_group(cache); | |
3299 | cache = btrfs_lookup_first_block_group(root->fs_info, | |
3300 | next_bytenr); | |
3301 | return cache; | |
3302 | } | |
4a8c9a62 YZ |
3303 | node = rb_next(&cache->cache_node); |
3304 | btrfs_put_block_group(cache); | |
3305 | if (node) { | |
3306 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3307 | cache_node); | |
11dfe35a | 3308 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3309 | } else |
3310 | cache = NULL; | |
3311 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3312 | return cache; | |
3313 | } | |
3314 | ||
0af3d00b JB |
3315 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3316 | struct btrfs_trans_handle *trans, | |
3317 | struct btrfs_path *path) | |
3318 | { | |
3319 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
3320 | struct inode *inode = NULL; | |
3321 | u64 alloc_hint = 0; | |
2b20982e | 3322 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3323 | u64 num_pages = 0; |
0af3d00b JB |
3324 | int retries = 0; |
3325 | int ret = 0; | |
3326 | ||
3327 | /* | |
3328 | * If this block group is smaller than 100 megs don't bother caching the | |
3329 | * block group. | |
3330 | */ | |
3331 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
3332 | spin_lock(&block_group->lock); | |
3333 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3334 | spin_unlock(&block_group->lock); | |
3335 | return 0; | |
3336 | } | |
3337 | ||
0c0ef4bc JB |
3338 | if (trans->aborted) |
3339 | return 0; | |
0af3d00b JB |
3340 | again: |
3341 | inode = lookup_free_space_inode(root, block_group, path); | |
3342 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
3343 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3344 | btrfs_release_path(path); |
0af3d00b JB |
3345 | goto out; |
3346 | } | |
3347 | ||
3348 | if (IS_ERR(inode)) { | |
3349 | BUG_ON(retries); | |
3350 | retries++; | |
3351 | ||
3352 | if (block_group->ro) | |
3353 | goto out_free; | |
3354 | ||
3355 | ret = create_free_space_inode(root, trans, block_group, path); | |
3356 | if (ret) | |
3357 | goto out_free; | |
3358 | goto again; | |
3359 | } | |
3360 | ||
5b0e95bf JB |
3361 | /* We've already setup this transaction, go ahead and exit */ |
3362 | if (block_group->cache_generation == trans->transid && | |
3363 | i_size_read(inode)) { | |
3364 | dcs = BTRFS_DC_SETUP; | |
3365 | goto out_put; | |
3366 | } | |
3367 | ||
0af3d00b JB |
3368 | /* |
3369 | * We want to set the generation to 0, that way if anything goes wrong | |
3370 | * from here on out we know not to trust this cache when we load up next | |
3371 | * time. | |
3372 | */ | |
3373 | BTRFS_I(inode)->generation = 0; | |
3374 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3375 | if (ret) { |
3376 | /* | |
3377 | * So theoretically we could recover from this, simply set the | |
3378 | * super cache generation to 0 so we know to invalidate the | |
3379 | * cache, but then we'd have to keep track of the block groups | |
3380 | * that fail this way so we know we _have_ to reset this cache | |
3381 | * before the next commit or risk reading stale cache. So to | |
3382 | * limit our exposure to horrible edge cases lets just abort the | |
3383 | * transaction, this only happens in really bad situations | |
3384 | * anyway. | |
3385 | */ | |
3386 | btrfs_abort_transaction(trans, root, ret); | |
3387 | goto out_put; | |
3388 | } | |
0af3d00b JB |
3389 | WARN_ON(ret); |
3390 | ||
3391 | if (i_size_read(inode) > 0) { | |
7b61cd92 MX |
3392 | ret = btrfs_check_trunc_cache_free_space(root, |
3393 | &root->fs_info->global_block_rsv); | |
3394 | if (ret) | |
3395 | goto out_put; | |
3396 | ||
1bbc621e | 3397 | ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode); |
0af3d00b JB |
3398 | if (ret) |
3399 | goto out_put; | |
3400 | } | |
3401 | ||
3402 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3403 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
e4c88f00 | 3404 | !btrfs_test_opt(root, SPACE_CACHE)) { |
cf7c1ef6 LB |
3405 | /* |
3406 | * don't bother trying to write stuff out _if_ | |
3407 | * a) we're not cached, | |
3408 | * b) we're with nospace_cache mount option. | |
3409 | */ | |
2b20982e | 3410 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3411 | spin_unlock(&block_group->lock); |
3412 | goto out_put; | |
3413 | } | |
3414 | spin_unlock(&block_group->lock); | |
3415 | ||
2968b1f4 JB |
3416 | /* |
3417 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3418 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3419 | */ | |
3420 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3421 | ret = -ENOSPC; | |
3422 | goto out_put; | |
3423 | } | |
3424 | ||
6fc823b1 JB |
3425 | /* |
3426 | * Try to preallocate enough space based on how big the block group is. | |
3427 | * Keep in mind this has to include any pinned space which could end up | |
3428 | * taking up quite a bit since it's not folded into the other space | |
3429 | * cache. | |
3430 | */ | |
f8c269d7 | 3431 | num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024); |
0af3d00b JB |
3432 | if (!num_pages) |
3433 | num_pages = 1; | |
3434 | ||
0af3d00b JB |
3435 | num_pages *= 16; |
3436 | num_pages *= PAGE_CACHE_SIZE; | |
3437 | ||
7cf5b976 | 3438 | ret = btrfs_check_data_free_space(inode, 0, num_pages); |
0af3d00b JB |
3439 | if (ret) |
3440 | goto out_put; | |
3441 | ||
3442 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3443 | num_pages, num_pages, | |
3444 | &alloc_hint); | |
2968b1f4 JB |
3445 | /* |
3446 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3447 | * of metadata or split extents when writing the cache out, which means | |
3448 | * we can enospc if we are heavily fragmented in addition to just normal | |
3449 | * out of space conditions. So if we hit this just skip setting up any | |
3450 | * other block groups for this transaction, maybe we'll unpin enough | |
3451 | * space the next time around. | |
3452 | */ | |
2b20982e JB |
3453 | if (!ret) |
3454 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3455 | else if (ret == -ENOSPC) |
3456 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
7cf5b976 | 3457 | btrfs_free_reserved_data_space(inode, 0, num_pages); |
c09544e0 | 3458 | |
0af3d00b JB |
3459 | out_put: |
3460 | iput(inode); | |
3461 | out_free: | |
b3b4aa74 | 3462 | btrfs_release_path(path); |
0af3d00b JB |
3463 | out: |
3464 | spin_lock(&block_group->lock); | |
e65cbb94 | 3465 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3466 | block_group->cache_generation = trans->transid; |
2b20982e | 3467 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3468 | spin_unlock(&block_group->lock); |
3469 | ||
3470 | return ret; | |
3471 | } | |
3472 | ||
dcdf7f6d JB |
3473 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
3474 | struct btrfs_root *root) | |
3475 | { | |
3476 | struct btrfs_block_group_cache *cache, *tmp; | |
3477 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3478 | struct btrfs_path *path; | |
3479 | ||
3480 | if (list_empty(&cur_trans->dirty_bgs) || | |
3481 | !btrfs_test_opt(root, SPACE_CACHE)) | |
3482 | return 0; | |
3483 | ||
3484 | path = btrfs_alloc_path(); | |
3485 | if (!path) | |
3486 | return -ENOMEM; | |
3487 | ||
3488 | /* Could add new block groups, use _safe just in case */ | |
3489 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3490 | dirty_list) { | |
3491 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3492 | cache_save_setup(cache, trans, path); | |
3493 | } | |
3494 | ||
3495 | btrfs_free_path(path); | |
3496 | return 0; | |
3497 | } | |
3498 | ||
1bbc621e CM |
3499 | /* |
3500 | * transaction commit does final block group cache writeback during a | |
3501 | * critical section where nothing is allowed to change the FS. This is | |
3502 | * required in order for the cache to actually match the block group, | |
3503 | * but can introduce a lot of latency into the commit. | |
3504 | * | |
3505 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3506 | * cache IO. There's a chance we'll have to redo some of it if the | |
3507 | * block group changes again during the commit, but it greatly reduces | |
3508 | * the commit latency by getting rid of the easy block groups while | |
3509 | * we're still allowing others to join the commit. | |
3510 | */ | |
3511 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans, | |
96b5179d | 3512 | struct btrfs_root *root) |
9078a3e1 | 3513 | { |
4a8c9a62 | 3514 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3515 | struct btrfs_transaction *cur_trans = trans->transaction; |
3516 | int ret = 0; | |
c9dc4c65 | 3517 | int should_put; |
1bbc621e CM |
3518 | struct btrfs_path *path = NULL; |
3519 | LIST_HEAD(dirty); | |
3520 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3521 | int num_started = 0; |
1bbc621e CM |
3522 | int loops = 0; |
3523 | ||
3524 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3525 | if (list_empty(&cur_trans->dirty_bgs)) { |
3526 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3527 | return 0; | |
1bbc621e | 3528 | } |
b58d1a9e | 3529 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3530 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3531 | |
1bbc621e | 3532 | again: |
1bbc621e CM |
3533 | /* |
3534 | * make sure all the block groups on our dirty list actually | |
3535 | * exist | |
3536 | */ | |
3537 | btrfs_create_pending_block_groups(trans, root); | |
3538 | ||
3539 | if (!path) { | |
3540 | path = btrfs_alloc_path(); | |
3541 | if (!path) | |
3542 | return -ENOMEM; | |
3543 | } | |
3544 | ||
b58d1a9e FM |
3545 | /* |
3546 | * cache_write_mutex is here only to save us from balance or automatic | |
3547 | * removal of empty block groups deleting this block group while we are | |
3548 | * writing out the cache | |
3549 | */ | |
3550 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3551 | while (!list_empty(&dirty)) { |
3552 | cache = list_first_entry(&dirty, | |
3553 | struct btrfs_block_group_cache, | |
3554 | dirty_list); | |
1bbc621e CM |
3555 | /* |
3556 | * this can happen if something re-dirties a block | |
3557 | * group that is already under IO. Just wait for it to | |
3558 | * finish and then do it all again | |
3559 | */ | |
3560 | if (!list_empty(&cache->io_list)) { | |
3561 | list_del_init(&cache->io_list); | |
3562 | btrfs_wait_cache_io(root, trans, cache, | |
3563 | &cache->io_ctl, path, | |
3564 | cache->key.objectid); | |
3565 | btrfs_put_block_group(cache); | |
3566 | } | |
3567 | ||
3568 | ||
3569 | /* | |
3570 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3571 | * if it should update the cache_state. Don't delete | |
3572 | * until after we wait. | |
3573 | * | |
3574 | * Since we're not running in the commit critical section | |
3575 | * we need the dirty_bgs_lock to protect from update_block_group | |
3576 | */ | |
3577 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3578 | list_del_init(&cache->dirty_list); | |
3579 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3580 | ||
3581 | should_put = 1; | |
3582 | ||
3583 | cache_save_setup(cache, trans, path); | |
3584 | ||
3585 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3586 | cache->io_ctl.inode = NULL; | |
3587 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3588 | if (ret == 0 && cache->io_ctl.inode) { | |
3589 | num_started++; | |
3590 | should_put = 0; | |
3591 | ||
3592 | /* | |
3593 | * the cache_write_mutex is protecting | |
3594 | * the io_list | |
3595 | */ | |
3596 | list_add_tail(&cache->io_list, io); | |
3597 | } else { | |
3598 | /* | |
3599 | * if we failed to write the cache, the | |
3600 | * generation will be bad and life goes on | |
3601 | */ | |
3602 | ret = 0; | |
3603 | } | |
3604 | } | |
ff1f8250 | 3605 | if (!ret) { |
1bbc621e | 3606 | ret = write_one_cache_group(trans, root, path, cache); |
ff1f8250 FM |
3607 | /* |
3608 | * Our block group might still be attached to the list | |
3609 | * of new block groups in the transaction handle of some | |
3610 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3611 | * means its block group item isn't yet in the extent | |
3612 | * tree. If this happens ignore the error, as we will | |
3613 | * try again later in the critical section of the | |
3614 | * transaction commit. | |
3615 | */ | |
3616 | if (ret == -ENOENT) { | |
3617 | ret = 0; | |
3618 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3619 | if (list_empty(&cache->dirty_list)) { | |
3620 | list_add_tail(&cache->dirty_list, | |
3621 | &cur_trans->dirty_bgs); | |
3622 | btrfs_get_block_group(cache); | |
3623 | } | |
3624 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3625 | } else if (ret) { | |
3626 | btrfs_abort_transaction(trans, root, ret); | |
3627 | } | |
3628 | } | |
1bbc621e CM |
3629 | |
3630 | /* if its not on the io list, we need to put the block group */ | |
3631 | if (should_put) | |
3632 | btrfs_put_block_group(cache); | |
3633 | ||
3634 | if (ret) | |
3635 | break; | |
b58d1a9e FM |
3636 | |
3637 | /* | |
3638 | * Avoid blocking other tasks for too long. It might even save | |
3639 | * us from writing caches for block groups that are going to be | |
3640 | * removed. | |
3641 | */ | |
3642 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3643 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3644 | } |
b58d1a9e | 3645 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3646 | |
3647 | /* | |
3648 | * go through delayed refs for all the stuff we've just kicked off | |
3649 | * and then loop back (just once) | |
3650 | */ | |
3651 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
3652 | if (!ret && loops == 0) { | |
3653 | loops++; | |
3654 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3655 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3656 | /* |
3657 | * dirty_bgs_lock protects us from concurrent block group | |
3658 | * deletes too (not just cache_write_mutex). | |
3659 | */ | |
3660 | if (!list_empty(&dirty)) { | |
3661 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3662 | goto again; | |
3663 | } | |
1bbc621e | 3664 | spin_unlock(&cur_trans->dirty_bgs_lock); |
1bbc621e CM |
3665 | } |
3666 | ||
3667 | btrfs_free_path(path); | |
3668 | return ret; | |
3669 | } | |
3670 | ||
3671 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
3672 | struct btrfs_root *root) | |
3673 | { | |
3674 | struct btrfs_block_group_cache *cache; | |
3675 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3676 | int ret = 0; | |
3677 | int should_put; | |
3678 | struct btrfs_path *path; | |
3679 | struct list_head *io = &cur_trans->io_bgs; | |
3680 | int num_started = 0; | |
9078a3e1 CM |
3681 | |
3682 | path = btrfs_alloc_path(); | |
3683 | if (!path) | |
3684 | return -ENOMEM; | |
3685 | ||
ce93ec54 JB |
3686 | /* |
3687 | * We don't need the lock here since we are protected by the transaction | |
3688 | * commit. We want to do the cache_save_setup first and then run the | |
3689 | * delayed refs to make sure we have the best chance at doing this all | |
3690 | * in one shot. | |
3691 | */ | |
3692 | while (!list_empty(&cur_trans->dirty_bgs)) { | |
3693 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3694 | struct btrfs_block_group_cache, | |
3695 | dirty_list); | |
c9dc4c65 CM |
3696 | |
3697 | /* | |
3698 | * this can happen if cache_save_setup re-dirties a block | |
3699 | * group that is already under IO. Just wait for it to | |
3700 | * finish and then do it all again | |
3701 | */ | |
3702 | if (!list_empty(&cache->io_list)) { | |
3703 | list_del_init(&cache->io_list); | |
3704 | btrfs_wait_cache_io(root, trans, cache, | |
3705 | &cache->io_ctl, path, | |
3706 | cache->key.objectid); | |
3707 | btrfs_put_block_group(cache); | |
c9dc4c65 CM |
3708 | } |
3709 | ||
1bbc621e CM |
3710 | /* |
3711 | * don't remove from the dirty list until after we've waited | |
3712 | * on any pending IO | |
3713 | */ | |
ce93ec54 | 3714 | list_del_init(&cache->dirty_list); |
c9dc4c65 CM |
3715 | should_put = 1; |
3716 | ||
1bbc621e | 3717 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3718 | |
ce93ec54 | 3719 | if (!ret) |
c9dc4c65 CM |
3720 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1); |
3721 | ||
3722 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3723 | cache->io_ctl.inode = NULL; | |
3724 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3725 | if (ret == 0 && cache->io_ctl.inode) { | |
3726 | num_started++; | |
3727 | should_put = 0; | |
1bbc621e | 3728 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3729 | } else { |
3730 | /* | |
3731 | * if we failed to write the cache, the | |
3732 | * generation will be bad and life goes on | |
3733 | */ | |
3734 | ret = 0; | |
3735 | } | |
3736 | } | |
ff1f8250 | 3737 | if (!ret) { |
ce93ec54 | 3738 | ret = write_one_cache_group(trans, root, path, cache); |
ff1f8250 FM |
3739 | if (ret) |
3740 | btrfs_abort_transaction(trans, root, ret); | |
3741 | } | |
c9dc4c65 CM |
3742 | |
3743 | /* if its not on the io list, we need to put the block group */ | |
3744 | if (should_put) | |
3745 | btrfs_put_block_group(cache); | |
3746 | } | |
3747 | ||
1bbc621e CM |
3748 | while (!list_empty(io)) { |
3749 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3750 | io_list); |
3751 | list_del_init(&cache->io_list); | |
c9dc4c65 CM |
3752 | btrfs_wait_cache_io(root, trans, cache, |
3753 | &cache->io_ctl, path, cache->key.objectid); | |
0cb59c99 JB |
3754 | btrfs_put_block_group(cache); |
3755 | } | |
3756 | ||
9078a3e1 | 3757 | btrfs_free_path(path); |
ce93ec54 | 3758 | return ret; |
9078a3e1 CM |
3759 | } |
3760 | ||
d2fb3437 YZ |
3761 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3762 | { | |
3763 | struct btrfs_block_group_cache *block_group; | |
3764 | int readonly = 0; | |
3765 | ||
3766 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3767 | if (!block_group || block_group->ro) | |
3768 | readonly = 1; | |
3769 | if (block_group) | |
fa9c0d79 | 3770 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3771 | return readonly; |
3772 | } | |
3773 | ||
6ab0a202 JM |
3774 | static const char *alloc_name(u64 flags) |
3775 | { | |
3776 | switch (flags) { | |
3777 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3778 | return "mixed"; | |
3779 | case BTRFS_BLOCK_GROUP_METADATA: | |
3780 | return "metadata"; | |
3781 | case BTRFS_BLOCK_GROUP_DATA: | |
3782 | return "data"; | |
3783 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3784 | return "system"; | |
3785 | default: | |
3786 | WARN_ON(1); | |
3787 | return "invalid-combination"; | |
3788 | }; | |
3789 | } | |
3790 | ||
593060d7 CM |
3791 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3792 | u64 total_bytes, u64 bytes_used, | |
3793 | struct btrfs_space_info **space_info) | |
3794 | { | |
3795 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3796 | int i; |
3797 | int factor; | |
b150a4f1 | 3798 | int ret; |
b742bb82 YZ |
3799 | |
3800 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3801 | BTRFS_BLOCK_GROUP_RAID10)) | |
3802 | factor = 2; | |
3803 | else | |
3804 | factor = 1; | |
593060d7 CM |
3805 | |
3806 | found = __find_space_info(info, flags); | |
3807 | if (found) { | |
25179201 | 3808 | spin_lock(&found->lock); |
593060d7 | 3809 | found->total_bytes += total_bytes; |
89a55897 | 3810 | found->disk_total += total_bytes * factor; |
593060d7 | 3811 | found->bytes_used += bytes_used; |
b742bb82 | 3812 | found->disk_used += bytes_used * factor; |
2e6e5183 FM |
3813 | if (total_bytes > 0) |
3814 | found->full = 0; | |
25179201 | 3815 | spin_unlock(&found->lock); |
593060d7 CM |
3816 | *space_info = found; |
3817 | return 0; | |
3818 | } | |
c146afad | 3819 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3820 | if (!found) |
3821 | return -ENOMEM; | |
3822 | ||
908c7f19 | 3823 | ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL); |
b150a4f1 JB |
3824 | if (ret) { |
3825 | kfree(found); | |
3826 | return ret; | |
3827 | } | |
3828 | ||
c1895442 | 3829 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
b742bb82 | 3830 | INIT_LIST_HEAD(&found->block_groups[i]); |
80eb234a | 3831 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3832 | spin_lock_init(&found->lock); |
52ba6929 | 3833 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3834 | found->total_bytes = total_bytes; |
89a55897 | 3835 | found->disk_total = total_bytes * factor; |
593060d7 | 3836 | found->bytes_used = bytes_used; |
b742bb82 | 3837 | found->disk_used = bytes_used * factor; |
593060d7 | 3838 | found->bytes_pinned = 0; |
e8569813 | 3839 | found->bytes_reserved = 0; |
c146afad | 3840 | found->bytes_readonly = 0; |
f0486c68 | 3841 | found->bytes_may_use = 0; |
6af3e3ad | 3842 | found->full = 0; |
4f4db217 | 3843 | found->max_extent_size = 0; |
0e4f8f88 | 3844 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3845 | found->chunk_alloc = 0; |
fdb5effd JB |
3846 | found->flush = 0; |
3847 | init_waitqueue_head(&found->wait); | |
633c0aad | 3848 | INIT_LIST_HEAD(&found->ro_bgs); |
6ab0a202 JM |
3849 | |
3850 | ret = kobject_init_and_add(&found->kobj, &space_info_ktype, | |
3851 | info->space_info_kobj, "%s", | |
3852 | alloc_name(found->flags)); | |
3853 | if (ret) { | |
3854 | kfree(found); | |
3855 | return ret; | |
3856 | } | |
3857 | ||
593060d7 | 3858 | *space_info = found; |
4184ea7f | 3859 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3860 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3861 | info->data_sinfo = found; | |
6ab0a202 JM |
3862 | |
3863 | return ret; | |
593060d7 CM |
3864 | } |
3865 | ||
8790d502 CM |
3866 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3867 | { | |
899c81ea ID |
3868 | u64 extra_flags = chunk_to_extended(flags) & |
3869 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3870 | |
de98ced9 | 3871 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3872 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3873 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3874 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3875 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3876 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3877 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3878 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3879 | } |
593060d7 | 3880 | |
fc67c450 ID |
3881 | /* |
3882 | * returns target flags in extended format or 0 if restripe for this | |
3883 | * chunk_type is not in progress | |
c6664b42 ID |
3884 | * |
3885 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3886 | */ |
3887 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3888 | { | |
3889 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3890 | u64 target = 0; | |
3891 | ||
fc67c450 ID |
3892 | if (!bctl) |
3893 | return 0; | |
3894 | ||
3895 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3896 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3897 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3898 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3899 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3900 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3901 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3902 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3903 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3904 | } | |
3905 | ||
3906 | return target; | |
3907 | } | |
3908 | ||
a46d11a8 ID |
3909 | /* |
3910 | * @flags: available profiles in extended format (see ctree.h) | |
3911 | * | |
e4d8ec0f ID |
3912 | * Returns reduced profile in chunk format. If profile changing is in |
3913 | * progress (either running or paused) picks the target profile (if it's | |
3914 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3915 | */ |
48a3b636 | 3916 | static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3917 | { |
95669976 | 3918 | u64 num_devices = root->fs_info->fs_devices->rw_devices; |
fc67c450 | 3919 | u64 target; |
9c170b26 ZL |
3920 | u64 raid_type; |
3921 | u64 allowed = 0; | |
a061fc8d | 3922 | |
fc67c450 ID |
3923 | /* |
3924 | * see if restripe for this chunk_type is in progress, if so | |
3925 | * try to reduce to the target profile | |
3926 | */ | |
e4d8ec0f | 3927 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3928 | target = get_restripe_target(root->fs_info, flags); |
3929 | if (target) { | |
3930 | /* pick target profile only if it's already available */ | |
3931 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3932 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3933 | return extended_to_chunk(target); |
e4d8ec0f ID |
3934 | } |
3935 | } | |
3936 | spin_unlock(&root->fs_info->balance_lock); | |
3937 | ||
53b381b3 | 3938 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
3939 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
3940 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
3941 | allowed |= btrfs_raid_group[raid_type]; | |
3942 | } | |
3943 | allowed &= flags; | |
3944 | ||
3945 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
3946 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
3947 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
3948 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
3949 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
3950 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
3951 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
3952 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
3953 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
3954 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
3955 | ||
3956 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
3957 | ||
3958 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
3959 | } |
3960 | ||
f8213bdc | 3961 | static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags) |
6a63209f | 3962 | { |
de98ced9 | 3963 | unsigned seq; |
f8213bdc | 3964 | u64 flags; |
de98ced9 MX |
3965 | |
3966 | do { | |
f8213bdc | 3967 | flags = orig_flags; |
de98ced9 MX |
3968 | seq = read_seqbegin(&root->fs_info->profiles_lock); |
3969 | ||
3970 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3971 | flags |= root->fs_info->avail_data_alloc_bits; | |
3972 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3973 | flags |= root->fs_info->avail_system_alloc_bits; | |
3974 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3975 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
3976 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
6fef8df1 | 3977 | |
b742bb82 | 3978 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3979 | } |
3980 | ||
6d07bcec | 3981 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3982 | { |
b742bb82 | 3983 | u64 flags; |
53b381b3 | 3984 | u64 ret; |
9ed74f2d | 3985 | |
b742bb82 YZ |
3986 | if (data) |
3987 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3988 | else if (root == root->fs_info->chunk_root) | |
3989 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3990 | else |
b742bb82 | 3991 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3992 | |
53b381b3 DW |
3993 | ret = get_alloc_profile(root, flags); |
3994 | return ret; | |
6a63209f | 3995 | } |
9ed74f2d | 3996 | |
4ceff079 | 3997 | int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes) |
6a63209f | 3998 | { |
6a63209f | 3999 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 4000 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 4001 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 4002 | u64 used; |
94b947b2 | 4003 | int ret = 0; |
c99f1b0c ZL |
4004 | int need_commit = 2; |
4005 | int have_pinned_space; | |
6a63209f | 4006 | |
6a63209f | 4007 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 4008 | bytes = ALIGN(bytes, root->sectorsize); |
6a63209f | 4009 | |
9dced186 | 4010 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4011 | need_commit = 0; |
9dced186 | 4012 | ASSERT(current->journal_info); |
0af3d00b JB |
4013 | } |
4014 | ||
b4d7c3c9 | 4015 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
4016 | if (!data_sinfo) |
4017 | goto alloc; | |
9ed74f2d | 4018 | |
6a63209f JB |
4019 | again: |
4020 | /* make sure we have enough space to handle the data first */ | |
4021 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
4022 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
4023 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
4024 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
4025 | |
4026 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4027 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4028 | |
6a63209f JB |
4029 | /* |
4030 | * if we don't have enough free bytes in this space then we need | |
4031 | * to alloc a new chunk. | |
4032 | */ | |
b9fd47cd | 4033 | if (!data_sinfo->full) { |
6a63209f | 4034 | u64 alloc_target; |
9ed74f2d | 4035 | |
0e4f8f88 | 4036 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4037 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 4038 | alloc: |
6a63209f | 4039 | alloc_target = btrfs_get_alloc_profile(root, 1); |
9dced186 MX |
4040 | /* |
4041 | * It is ugly that we don't call nolock join | |
4042 | * transaction for the free space inode case here. | |
4043 | * But it is safe because we only do the data space | |
4044 | * reservation for the free space cache in the | |
4045 | * transaction context, the common join transaction | |
4046 | * just increase the counter of the current transaction | |
4047 | * handler, doesn't try to acquire the trans_lock of | |
4048 | * the fs. | |
4049 | */ | |
7a7eaa40 | 4050 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4051 | if (IS_ERR(trans)) |
4052 | return PTR_ERR(trans); | |
9ed74f2d | 4053 | |
6a63209f | 4054 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
4055 | alloc_target, |
4056 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 4057 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
4058 | if (ret < 0) { |
4059 | if (ret != -ENOSPC) | |
4060 | return ret; | |
c99f1b0c ZL |
4061 | else { |
4062 | have_pinned_space = 1; | |
d52a5b5f | 4063 | goto commit_trans; |
c99f1b0c | 4064 | } |
d52a5b5f | 4065 | } |
9ed74f2d | 4066 | |
b4d7c3c9 LZ |
4067 | if (!data_sinfo) |
4068 | data_sinfo = fs_info->data_sinfo; | |
4069 | ||
6a63209f JB |
4070 | goto again; |
4071 | } | |
f2bb8f5c JB |
4072 | |
4073 | /* | |
b150a4f1 | 4074 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4075 | * allocation, and no removed chunk in current transaction, |
4076 | * don't bother committing the transaction. | |
f2bb8f5c | 4077 | */ |
c99f1b0c ZL |
4078 | have_pinned_space = percpu_counter_compare( |
4079 | &data_sinfo->total_bytes_pinned, | |
4080 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4081 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4082 | |
4e06bdd6 | 4083 | /* commit the current transaction and try again */ |
d52a5b5f | 4084 | commit_trans: |
c99f1b0c | 4085 | if (need_commit && |
a4abeea4 | 4086 | !atomic_read(&root->fs_info->open_ioctl_trans)) { |
c99f1b0c | 4087 | need_commit--; |
b150a4f1 | 4088 | |
9a4e7276 ZL |
4089 | if (need_commit > 0) |
4090 | btrfs_wait_ordered_roots(fs_info, -1); | |
4091 | ||
7a7eaa40 | 4092 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4093 | if (IS_ERR(trans)) |
4094 | return PTR_ERR(trans); | |
c99f1b0c | 4095 | if (have_pinned_space >= 0 || |
3204d33c JB |
4096 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4097 | &trans->transaction->flags) || | |
c99f1b0c | 4098 | need_commit > 0) { |
94b947b2 ZL |
4099 | ret = btrfs_commit_transaction(trans, root); |
4100 | if (ret) | |
4101 | return ret; | |
d7c15171 ZL |
4102 | /* |
4103 | * make sure that all running delayed iput are | |
4104 | * done | |
4105 | */ | |
4106 | down_write(&root->fs_info->delayed_iput_sem); | |
4107 | up_write(&root->fs_info->delayed_iput_sem); | |
94b947b2 ZL |
4108 | goto again; |
4109 | } else { | |
4110 | btrfs_end_transaction(trans, root); | |
4111 | } | |
4e06bdd6 | 4112 | } |
9ed74f2d | 4113 | |
cab45e22 JM |
4114 | trace_btrfs_space_reservation(root->fs_info, |
4115 | "space_info:enospc", | |
4116 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4117 | return -ENOSPC; |
4118 | } | |
4119 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 4120 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 4121 | data_sinfo->flags, bytes, 1); |
6a63209f | 4122 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4123 | |
237c0e9f | 4124 | return ret; |
9ed74f2d | 4125 | } |
6a63209f | 4126 | |
4ceff079 QW |
4127 | /* |
4128 | * New check_data_free_space() with ability for precious data reservation | |
4129 | * Will replace old btrfs_check_data_free_space(), but for patch split, | |
4130 | * add a new function first and then replace it. | |
4131 | */ | |
7cf5b976 | 4132 | int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len) |
4ceff079 QW |
4133 | { |
4134 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4135 | int ret; | |
4136 | ||
4137 | /* align the range */ | |
4138 | len = round_up(start + len, root->sectorsize) - | |
4139 | round_down(start, root->sectorsize); | |
4140 | start = round_down(start, root->sectorsize); | |
4141 | ||
4142 | ret = btrfs_alloc_data_chunk_ondemand(inode, len); | |
4143 | if (ret < 0) | |
4144 | return ret; | |
4145 | ||
94ed938a QW |
4146 | /* |
4147 | * Use new btrfs_qgroup_reserve_data to reserve precious data space | |
4148 | * | |
4149 | * TODO: Find a good method to avoid reserve data space for NOCOW | |
4150 | * range, but don't impact performance on quota disable case. | |
4151 | */ | |
4ceff079 QW |
4152 | ret = btrfs_qgroup_reserve_data(inode, start, len); |
4153 | return ret; | |
4154 | } | |
4155 | ||
4ceff079 QW |
4156 | /* |
4157 | * Called if we need to clear a data reservation for this inode | |
4158 | * Normally in a error case. | |
4159 | * | |
51773bec QW |
4160 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4161 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4162 | * Like clear_bit_hook(). | |
4ceff079 | 4163 | */ |
51773bec QW |
4164 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4165 | u64 len) | |
4ceff079 QW |
4166 | { |
4167 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4168 | struct btrfs_space_info *data_sinfo; | |
4169 | ||
4170 | /* Make sure the range is aligned to sectorsize */ | |
4171 | len = round_up(start + len, root->sectorsize) - | |
4172 | round_down(start, root->sectorsize); | |
4173 | start = round_down(start, root->sectorsize); | |
4174 | ||
4ceff079 QW |
4175 | data_sinfo = root->fs_info->data_sinfo; |
4176 | spin_lock(&data_sinfo->lock); | |
4177 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4178 | data_sinfo->bytes_may_use = 0; | |
4179 | else | |
4180 | data_sinfo->bytes_may_use -= len; | |
4181 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4182 | data_sinfo->flags, len, 0); | |
4183 | spin_unlock(&data_sinfo->lock); | |
4184 | } | |
4185 | ||
51773bec QW |
4186 | /* |
4187 | * Called if we need to clear a data reservation for this inode | |
4188 | * Normally in a error case. | |
4189 | * | |
4190 | * This one will handle the per-indoe data rsv map for accurate reserved | |
4191 | * space framework. | |
4192 | */ | |
4193 | void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len) | |
4194 | { | |
4195 | btrfs_free_reserved_data_space_noquota(inode, start, len); | |
4196 | btrfs_qgroup_free_data(inode, start, len); | |
4197 | } | |
4198 | ||
97e728d4 | 4199 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4200 | { |
97e728d4 JB |
4201 | struct list_head *head = &info->space_info; |
4202 | struct btrfs_space_info *found; | |
e3ccfa98 | 4203 | |
97e728d4 JB |
4204 | rcu_read_lock(); |
4205 | list_for_each_entry_rcu(found, head, list) { | |
4206 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4207 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4208 | } |
97e728d4 | 4209 | rcu_read_unlock(); |
e3ccfa98 JB |
4210 | } |
4211 | ||
3c76cd84 MX |
4212 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4213 | { | |
4214 | return (global->size << 1); | |
4215 | } | |
4216 | ||
e5bc2458 | 4217 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 4218 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4219 | { |
fb25e914 | 4220 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 4221 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 4222 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 4223 | u64 thresh; |
e3ccfa98 | 4224 | |
0e4f8f88 CM |
4225 | if (force == CHUNK_ALLOC_FORCE) |
4226 | return 1; | |
4227 | ||
fb25e914 JB |
4228 | /* |
4229 | * We need to take into account the global rsv because for all intents | |
4230 | * and purposes it's used space. Don't worry about locking the | |
4231 | * global_rsv, it doesn't change except when the transaction commits. | |
4232 | */ | |
54338b5c | 4233 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3c76cd84 | 4234 | num_allocated += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4235 | |
0e4f8f88 CM |
4236 | /* |
4237 | * in limited mode, we want to have some free space up to | |
4238 | * about 1% of the FS size. | |
4239 | */ | |
4240 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 4241 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
4242 | thresh = max_t(u64, 64 * 1024 * 1024, |
4243 | div_factor_fine(thresh, 1)); | |
4244 | ||
4245 | if (num_bytes - num_allocated < thresh) | |
4246 | return 1; | |
4247 | } | |
0e4f8f88 | 4248 | |
698d0082 | 4249 | if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8)) |
14ed0ca6 | 4250 | return 0; |
424499db | 4251 | return 1; |
32c00aff JB |
4252 | } |
4253 | ||
39c2d7fa | 4254 | static u64 get_profile_num_devs(struct btrfs_root *root, u64 type) |
15d1ff81 LB |
4255 | { |
4256 | u64 num_dev; | |
4257 | ||
53b381b3 DW |
4258 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4259 | BTRFS_BLOCK_GROUP_RAID0 | | |
4260 | BTRFS_BLOCK_GROUP_RAID5 | | |
4261 | BTRFS_BLOCK_GROUP_RAID6)) | |
15d1ff81 LB |
4262 | num_dev = root->fs_info->fs_devices->rw_devices; |
4263 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
4264 | num_dev = 2; | |
4265 | else | |
4266 | num_dev = 1; /* DUP or single */ | |
4267 | ||
39c2d7fa | 4268 | return num_dev; |
15d1ff81 LB |
4269 | } |
4270 | ||
39c2d7fa FM |
4271 | /* |
4272 | * If @is_allocation is true, reserve space in the system space info necessary | |
4273 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4274 | * removing a chunk. | |
4275 | */ | |
4276 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
4277 | struct btrfs_root *root, | |
4617ea3a | 4278 | u64 type) |
15d1ff81 LB |
4279 | { |
4280 | struct btrfs_space_info *info; | |
4281 | u64 left; | |
4282 | u64 thresh; | |
4fbcdf66 | 4283 | int ret = 0; |
39c2d7fa | 4284 | u64 num_devs; |
4fbcdf66 FM |
4285 | |
4286 | /* | |
4287 | * Needed because we can end up allocating a system chunk and for an | |
4288 | * atomic and race free space reservation in the chunk block reserve. | |
4289 | */ | |
4290 | ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex)); | |
15d1ff81 LB |
4291 | |
4292 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4293 | spin_lock(&info->lock); | |
4294 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
4fbcdf66 FM |
4295 | info->bytes_reserved - info->bytes_readonly - |
4296 | info->bytes_may_use; | |
15d1ff81 LB |
4297 | spin_unlock(&info->lock); |
4298 | ||
39c2d7fa FM |
4299 | num_devs = get_profile_num_devs(root, type); |
4300 | ||
4301 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
4617ea3a FM |
4302 | thresh = btrfs_calc_trunc_metadata_size(root, num_devs) + |
4303 | btrfs_calc_trans_metadata_size(root, 1); | |
39c2d7fa | 4304 | |
15d1ff81 | 4305 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { |
c2cf52eb SK |
4306 | btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu", |
4307 | left, thresh, type); | |
15d1ff81 LB |
4308 | dump_space_info(info, 0, 0); |
4309 | } | |
4310 | ||
4311 | if (left < thresh) { | |
4312 | u64 flags; | |
4313 | ||
4314 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
4fbcdf66 FM |
4315 | /* |
4316 | * Ignore failure to create system chunk. We might end up not | |
4317 | * needing it, as we might not need to COW all nodes/leafs from | |
4318 | * the paths we visit in the chunk tree (they were already COWed | |
4319 | * or created in the current transaction for example). | |
4320 | */ | |
4321 | ret = btrfs_alloc_chunk(trans, root, flags); | |
4322 | } | |
4323 | ||
4324 | if (!ret) { | |
4325 | ret = btrfs_block_rsv_add(root->fs_info->chunk_root, | |
4326 | &root->fs_info->chunk_block_rsv, | |
4327 | thresh, BTRFS_RESERVE_NO_FLUSH); | |
4328 | if (!ret) | |
4329 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4330 | } |
4331 | } | |
4332 | ||
6324fbf3 | 4333 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 4334 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 4335 | { |
6324fbf3 | 4336 | struct btrfs_space_info *space_info; |
97e728d4 | 4337 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 4338 | int wait_for_alloc = 0; |
9ed74f2d | 4339 | int ret = 0; |
9ed74f2d | 4340 | |
c6b305a8 JB |
4341 | /* Don't re-enter if we're already allocating a chunk */ |
4342 | if (trans->allocating_chunk) | |
4343 | return -ENOSPC; | |
4344 | ||
6324fbf3 | 4345 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
4346 | if (!space_info) { |
4347 | ret = update_space_info(extent_root->fs_info, flags, | |
4348 | 0, 0, &space_info); | |
79787eaa | 4349 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 4350 | } |
79787eaa | 4351 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 4352 | |
6d74119f | 4353 | again: |
25179201 | 4354 | spin_lock(&space_info->lock); |
9e622d6b | 4355 | if (force < space_info->force_alloc) |
0e4f8f88 | 4356 | force = space_info->force_alloc; |
25179201 | 4357 | if (space_info->full) { |
09fb99a6 FDBM |
4358 | if (should_alloc_chunk(extent_root, space_info, force)) |
4359 | ret = -ENOSPC; | |
4360 | else | |
4361 | ret = 0; | |
25179201 | 4362 | spin_unlock(&space_info->lock); |
09fb99a6 | 4363 | return ret; |
9ed74f2d JB |
4364 | } |
4365 | ||
698d0082 | 4366 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 4367 | spin_unlock(&space_info->lock); |
6d74119f JB |
4368 | return 0; |
4369 | } else if (space_info->chunk_alloc) { | |
4370 | wait_for_alloc = 1; | |
4371 | } else { | |
4372 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4373 | } |
0e4f8f88 | 4374 | |
25179201 | 4375 | spin_unlock(&space_info->lock); |
9ed74f2d | 4376 | |
6d74119f JB |
4377 | mutex_lock(&fs_info->chunk_mutex); |
4378 | ||
4379 | /* | |
4380 | * The chunk_mutex is held throughout the entirety of a chunk | |
4381 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4382 | * other guy is done and we need to recheck and see if we should | |
4383 | * allocate. | |
4384 | */ | |
4385 | if (wait_for_alloc) { | |
4386 | mutex_unlock(&fs_info->chunk_mutex); | |
4387 | wait_for_alloc = 0; | |
4388 | goto again; | |
4389 | } | |
4390 | ||
c6b305a8 JB |
4391 | trans->allocating_chunk = true; |
4392 | ||
67377734 JB |
4393 | /* |
4394 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4395 | * allocating mixed chunks instead of individual chunks. | |
4396 | */ | |
4397 | if (btrfs_mixed_space_info(space_info)) | |
4398 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4399 | ||
97e728d4 JB |
4400 | /* |
4401 | * if we're doing a data chunk, go ahead and make sure that | |
4402 | * we keep a reasonable number of metadata chunks allocated in the | |
4403 | * FS as well. | |
4404 | */ | |
9ed74f2d | 4405 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4406 | fs_info->data_chunk_allocations++; |
4407 | if (!(fs_info->data_chunk_allocations % | |
4408 | fs_info->metadata_ratio)) | |
4409 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4410 | } |
4411 | ||
15d1ff81 LB |
4412 | /* |
4413 | * Check if we have enough space in SYSTEM chunk because we may need | |
4414 | * to update devices. | |
4415 | */ | |
4617ea3a | 4416 | check_system_chunk(trans, extent_root, flags); |
15d1ff81 | 4417 | |
2b82032c | 4418 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 4419 | trans->allocating_chunk = false; |
92b8e897 | 4420 | |
9ed74f2d | 4421 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4422 | if (ret < 0 && ret != -ENOSPC) |
4423 | goto out; | |
9ed74f2d | 4424 | if (ret) |
6324fbf3 | 4425 | space_info->full = 1; |
424499db YZ |
4426 | else |
4427 | ret = 1; | |
6d74119f | 4428 | |
0e4f8f88 | 4429 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4430 | out: |
6d74119f | 4431 | space_info->chunk_alloc = 0; |
9ed74f2d | 4432 | spin_unlock(&space_info->lock); |
a25c75d5 | 4433 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4434 | /* |
4435 | * When we allocate a new chunk we reserve space in the chunk block | |
4436 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4437 | * add new nodes/leafs to it if we end up needing to do it when | |
4438 | * inserting the chunk item and updating device items as part of the | |
4439 | * second phase of chunk allocation, performed by | |
4440 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4441 | * large number of new block groups to create in our transaction | |
4442 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4443 | * in extreme cases - like having a single transaction create many new | |
4444 | * block groups when starting to write out the free space caches of all | |
4445 | * the block groups that were made dirty during the lifetime of the | |
4446 | * transaction. | |
4447 | */ | |
d9a0540a FM |
4448 | if (trans->can_flush_pending_bgs && |
4449 | trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) { | |
00d80e34 FM |
4450 | btrfs_create_pending_block_groups(trans, trans->root); |
4451 | btrfs_trans_release_chunk_metadata(trans); | |
4452 | } | |
0f9dd46c | 4453 | return ret; |
6324fbf3 | 4454 | } |
9ed74f2d | 4455 | |
a80c8dcf JB |
4456 | static int can_overcommit(struct btrfs_root *root, |
4457 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 4458 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf | 4459 | { |
96f1bb57 | 4460 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
a80c8dcf | 4461 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3c76cd84 | 4462 | u64 space_size; |
a80c8dcf JB |
4463 | u64 avail; |
4464 | u64 used; | |
4465 | ||
4466 | used = space_info->bytes_used + space_info->bytes_reserved + | |
96f1bb57 JB |
4467 | space_info->bytes_pinned + space_info->bytes_readonly; |
4468 | ||
96f1bb57 JB |
4469 | /* |
4470 | * We only want to allow over committing if we have lots of actual space | |
4471 | * free, but if we don't have enough space to handle the global reserve | |
4472 | * space then we could end up having a real enospc problem when trying | |
4473 | * to allocate a chunk or some other such important allocation. | |
4474 | */ | |
3c76cd84 MX |
4475 | spin_lock(&global_rsv->lock); |
4476 | space_size = calc_global_rsv_need_space(global_rsv); | |
4477 | spin_unlock(&global_rsv->lock); | |
4478 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4479 | return 0; |
4480 | ||
4481 | used += space_info->bytes_may_use; | |
a80c8dcf JB |
4482 | |
4483 | spin_lock(&root->fs_info->free_chunk_lock); | |
4484 | avail = root->fs_info->free_chunk_space; | |
4485 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4486 | ||
4487 | /* | |
4488 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4489 | * space is actually useable. For raid56, the space info used |
4490 | * doesn't include the parity drive, so we don't have to | |
4491 | * change the math | |
a80c8dcf JB |
4492 | */ |
4493 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4494 | BTRFS_BLOCK_GROUP_RAID1 | | |
4495 | BTRFS_BLOCK_GROUP_RAID10)) | |
4496 | avail >>= 1; | |
4497 | ||
4498 | /* | |
561c294d MX |
4499 | * If we aren't flushing all things, let us overcommit up to |
4500 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4501 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4502 | */ |
08e007d2 | 4503 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4504 | avail >>= 3; |
a80c8dcf | 4505 | else |
14575aef | 4506 | avail >>= 1; |
a80c8dcf | 4507 | |
14575aef | 4508 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4509 | return 1; |
4510 | return 0; | |
4511 | } | |
4512 | ||
48a3b636 | 4513 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
6c255e67 | 4514 | unsigned long nr_pages, int nr_items) |
da633a42 MX |
4515 | { |
4516 | struct super_block *sb = root->fs_info->sb; | |
da633a42 | 4517 | |
925a6efb JB |
4518 | if (down_read_trylock(&sb->s_umount)) { |
4519 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4520 | up_read(&sb->s_umount); | |
4521 | } else { | |
da633a42 MX |
4522 | /* |
4523 | * We needn't worry the filesystem going from r/w to r/o though | |
4524 | * we don't acquire ->s_umount mutex, because the filesystem | |
4525 | * should guarantee the delalloc inodes list be empty after | |
4526 | * the filesystem is readonly(all dirty pages are written to | |
4527 | * the disk). | |
4528 | */ | |
6c255e67 | 4529 | btrfs_start_delalloc_roots(root->fs_info, 0, nr_items); |
98ad69cf | 4530 | if (!current->journal_info) |
6c255e67 | 4531 | btrfs_wait_ordered_roots(root->fs_info, nr_items); |
da633a42 MX |
4532 | } |
4533 | } | |
4534 | ||
18cd8ea6 MX |
4535 | static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim) |
4536 | { | |
4537 | u64 bytes; | |
4538 | int nr; | |
4539 | ||
4540 | bytes = btrfs_calc_trans_metadata_size(root, 1); | |
4541 | nr = (int)div64_u64(to_reclaim, bytes); | |
4542 | if (!nr) | |
4543 | nr = 1; | |
4544 | return nr; | |
4545 | } | |
4546 | ||
c61a16a7 MX |
4547 | #define EXTENT_SIZE_PER_ITEM (256 * 1024) |
4548 | ||
9ed74f2d | 4549 | /* |
5da9d01b | 4550 | * shrink metadata reservation for delalloc |
9ed74f2d | 4551 | */ |
f4c738c2 JB |
4552 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
4553 | bool wait_ordered) | |
5da9d01b | 4554 | { |
0ca1f7ce | 4555 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4556 | struct btrfs_space_info *space_info; |
663350ac | 4557 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4558 | u64 delalloc_bytes; |
5da9d01b | 4559 | u64 max_reclaim; |
b1953bce | 4560 | long time_left; |
d3ee29e3 MX |
4561 | unsigned long nr_pages; |
4562 | int loops; | |
b0244199 | 4563 | int items; |
08e007d2 | 4564 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4565 | |
c61a16a7 | 4566 | /* Calc the number of the pages we need flush for space reservation */ |
b0244199 MX |
4567 | items = calc_reclaim_items_nr(root, to_reclaim); |
4568 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; | |
c61a16a7 | 4569 | |
663350ac | 4570 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 4571 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 4572 | space_info = block_rsv->space_info; |
bf9022e0 | 4573 | |
963d678b MX |
4574 | delalloc_bytes = percpu_counter_sum_positive( |
4575 | &root->fs_info->delalloc_bytes); | |
f4c738c2 | 4576 | if (delalloc_bytes == 0) { |
fdb5effd | 4577 | if (trans) |
f4c738c2 | 4578 | return; |
38c135af | 4579 | if (wait_ordered) |
b0244199 | 4580 | btrfs_wait_ordered_roots(root->fs_info, items); |
f4c738c2 | 4581 | return; |
fdb5effd JB |
4582 | } |
4583 | ||
d3ee29e3 | 4584 | loops = 0; |
f4c738c2 JB |
4585 | while (delalloc_bytes && loops < 3) { |
4586 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
4587 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; | |
6c255e67 | 4588 | btrfs_writeback_inodes_sb_nr(root, nr_pages, items); |
dea31f52 JB |
4589 | /* |
4590 | * We need to wait for the async pages to actually start before | |
4591 | * we do anything. | |
4592 | */ | |
9f3a074d MX |
4593 | max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages); |
4594 | if (!max_reclaim) | |
4595 | goto skip_async; | |
4596 | ||
4597 | if (max_reclaim <= nr_pages) | |
4598 | max_reclaim = 0; | |
4599 | else | |
4600 | max_reclaim -= nr_pages; | |
dea31f52 | 4601 | |
9f3a074d MX |
4602 | wait_event(root->fs_info->async_submit_wait, |
4603 | atomic_read(&root->fs_info->async_delalloc_pages) <= | |
4604 | (int)max_reclaim); | |
4605 | skip_async: | |
08e007d2 MX |
4606 | if (!trans) |
4607 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4608 | else | |
4609 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4610 | spin_lock(&space_info->lock); |
08e007d2 | 4611 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
4612 | spin_unlock(&space_info->lock); |
4613 | break; | |
4614 | } | |
0019f10d | 4615 | spin_unlock(&space_info->lock); |
5da9d01b | 4616 | |
36e39c40 | 4617 | loops++; |
f104d044 | 4618 | if (wait_ordered && !trans) { |
b0244199 | 4619 | btrfs_wait_ordered_roots(root->fs_info, items); |
f104d044 | 4620 | } else { |
f4c738c2 | 4621 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4622 | if (time_left) |
4623 | break; | |
4624 | } | |
963d678b MX |
4625 | delalloc_bytes = percpu_counter_sum_positive( |
4626 | &root->fs_info->delalloc_bytes); | |
5da9d01b | 4627 | } |
5da9d01b YZ |
4628 | } |
4629 | ||
663350ac JB |
4630 | /** |
4631 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4632 | * @root - the root we're allocating for | |
4633 | * @bytes - the number of bytes we want to reserve | |
4634 | * @force - force the commit | |
8bb8ab2e | 4635 | * |
663350ac JB |
4636 | * This will check to make sure that committing the transaction will actually |
4637 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4638 | * will return -ENOSPC. | |
8bb8ab2e | 4639 | */ |
663350ac JB |
4640 | static int may_commit_transaction(struct btrfs_root *root, |
4641 | struct btrfs_space_info *space_info, | |
4642 | u64 bytes, int force) | |
4643 | { | |
4644 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
4645 | struct btrfs_trans_handle *trans; | |
4646 | ||
4647 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4648 | if (trans) | |
4649 | return -EAGAIN; | |
4650 | ||
4651 | if (force) | |
4652 | goto commit; | |
4653 | ||
4654 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4655 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4656 | bytes) >= 0) |
663350ac | 4657 | goto commit; |
663350ac JB |
4658 | |
4659 | /* | |
4660 | * See if there is some space in the delayed insertion reservation for | |
4661 | * this reservation. | |
4662 | */ | |
4663 | if (space_info != delayed_rsv->space_info) | |
4664 | return -ENOSPC; | |
4665 | ||
4666 | spin_lock(&delayed_rsv->lock); | |
b150a4f1 JB |
4667 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
4668 | bytes - delayed_rsv->size) >= 0) { | |
663350ac JB |
4669 | spin_unlock(&delayed_rsv->lock); |
4670 | return -ENOSPC; | |
4671 | } | |
4672 | spin_unlock(&delayed_rsv->lock); | |
4673 | ||
4674 | commit: | |
4675 | trans = btrfs_join_transaction(root); | |
4676 | if (IS_ERR(trans)) | |
4677 | return -ENOSPC; | |
4678 | ||
4679 | return btrfs_commit_transaction(trans, root); | |
4680 | } | |
4681 | ||
96c3f433 | 4682 | enum flush_state { |
67b0fd63 JB |
4683 | FLUSH_DELAYED_ITEMS_NR = 1, |
4684 | FLUSH_DELAYED_ITEMS = 2, | |
4685 | FLUSH_DELALLOC = 3, | |
4686 | FLUSH_DELALLOC_WAIT = 4, | |
ea658bad JB |
4687 | ALLOC_CHUNK = 5, |
4688 | COMMIT_TRANS = 6, | |
96c3f433 JB |
4689 | }; |
4690 | ||
4691 | static int flush_space(struct btrfs_root *root, | |
4692 | struct btrfs_space_info *space_info, u64 num_bytes, | |
4693 | u64 orig_bytes, int state) | |
4694 | { | |
4695 | struct btrfs_trans_handle *trans; | |
4696 | int nr; | |
f4c738c2 | 4697 | int ret = 0; |
96c3f433 JB |
4698 | |
4699 | switch (state) { | |
96c3f433 JB |
4700 | case FLUSH_DELAYED_ITEMS_NR: |
4701 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 MX |
4702 | if (state == FLUSH_DELAYED_ITEMS_NR) |
4703 | nr = calc_reclaim_items_nr(root, num_bytes) * 2; | |
4704 | else | |
96c3f433 | 4705 | nr = -1; |
18cd8ea6 | 4706 | |
96c3f433 JB |
4707 | trans = btrfs_join_transaction(root); |
4708 | if (IS_ERR(trans)) { | |
4709 | ret = PTR_ERR(trans); | |
4710 | break; | |
4711 | } | |
4712 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
4713 | btrfs_end_transaction(trans, root); | |
4714 | break; | |
67b0fd63 JB |
4715 | case FLUSH_DELALLOC: |
4716 | case FLUSH_DELALLOC_WAIT: | |
24af7dd1 | 4717 | shrink_delalloc(root, num_bytes * 2, orig_bytes, |
67b0fd63 JB |
4718 | state == FLUSH_DELALLOC_WAIT); |
4719 | break; | |
ea658bad JB |
4720 | case ALLOC_CHUNK: |
4721 | trans = btrfs_join_transaction(root); | |
4722 | if (IS_ERR(trans)) { | |
4723 | ret = PTR_ERR(trans); | |
4724 | break; | |
4725 | } | |
4726 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
4727 | btrfs_get_alloc_profile(root, 0), |
4728 | CHUNK_ALLOC_NO_FORCE); | |
4729 | btrfs_end_transaction(trans, root); | |
4730 | if (ret == -ENOSPC) | |
4731 | ret = 0; | |
4732 | break; | |
96c3f433 JB |
4733 | case COMMIT_TRANS: |
4734 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
4735 | break; | |
4736 | default: | |
4737 | ret = -ENOSPC; | |
4738 | break; | |
4739 | } | |
4740 | ||
4741 | return ret; | |
4742 | } | |
21c7e756 MX |
4743 | |
4744 | static inline u64 | |
4745 | btrfs_calc_reclaim_metadata_size(struct btrfs_root *root, | |
4746 | struct btrfs_space_info *space_info) | |
4747 | { | |
4748 | u64 used; | |
4749 | u64 expected; | |
4750 | u64 to_reclaim; | |
4751 | ||
4752 | to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024, | |
4753 | 16 * 1024 * 1024); | |
4754 | spin_lock(&space_info->lock); | |
4755 | if (can_overcommit(root, space_info, to_reclaim, | |
4756 | BTRFS_RESERVE_FLUSH_ALL)) { | |
4757 | to_reclaim = 0; | |
4758 | goto out; | |
4759 | } | |
4760 | ||
4761 | used = space_info->bytes_used + space_info->bytes_reserved + | |
4762 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4763 | space_info->bytes_may_use; | |
4764 | if (can_overcommit(root, space_info, 1024 * 1024, | |
4765 | BTRFS_RESERVE_FLUSH_ALL)) | |
4766 | expected = div_factor_fine(space_info->total_bytes, 95); | |
4767 | else | |
4768 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4769 | ||
4770 | if (used > expected) | |
4771 | to_reclaim = used - expected; | |
4772 | else | |
4773 | to_reclaim = 0; | |
4774 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4775 | space_info->bytes_reserved); | |
4776 | out: | |
4777 | spin_unlock(&space_info->lock); | |
4778 | ||
4779 | return to_reclaim; | |
4780 | } | |
4781 | ||
4782 | static inline int need_do_async_reclaim(struct btrfs_space_info *space_info, | |
4783 | struct btrfs_fs_info *fs_info, u64 used) | |
4784 | { | |
365c5313 JB |
4785 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4786 | ||
4787 | /* If we're just plain full then async reclaim just slows us down. */ | |
4788 | if (space_info->bytes_used >= thresh) | |
4789 | return 0; | |
4790 | ||
4791 | return (used >= thresh && !btrfs_fs_closing(fs_info) && | |
21c7e756 MX |
4792 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); |
4793 | } | |
4794 | ||
4795 | static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info, | |
25ce459c LB |
4796 | struct btrfs_fs_info *fs_info, |
4797 | int flush_state) | |
21c7e756 MX |
4798 | { |
4799 | u64 used; | |
4800 | ||
4801 | spin_lock(&space_info->lock); | |
25ce459c LB |
4802 | /* |
4803 | * We run out of space and have not got any free space via flush_space, | |
4804 | * so don't bother doing async reclaim. | |
4805 | */ | |
4806 | if (flush_state > COMMIT_TRANS && space_info->full) { | |
4807 | spin_unlock(&space_info->lock); | |
4808 | return 0; | |
4809 | } | |
4810 | ||
21c7e756 MX |
4811 | used = space_info->bytes_used + space_info->bytes_reserved + |
4812 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4813 | space_info->bytes_may_use; | |
4814 | if (need_do_async_reclaim(space_info, fs_info, used)) { | |
4815 | spin_unlock(&space_info->lock); | |
4816 | return 1; | |
4817 | } | |
4818 | spin_unlock(&space_info->lock); | |
4819 | ||
4820 | return 0; | |
4821 | } | |
4822 | ||
4823 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) | |
4824 | { | |
4825 | struct btrfs_fs_info *fs_info; | |
4826 | struct btrfs_space_info *space_info; | |
4827 | u64 to_reclaim; | |
4828 | int flush_state; | |
4829 | ||
4830 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4831 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4832 | ||
4833 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, | |
4834 | space_info); | |
4835 | if (!to_reclaim) | |
4836 | return; | |
4837 | ||
4838 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
4839 | do { | |
4840 | flush_space(fs_info->fs_root, space_info, to_reclaim, | |
4841 | to_reclaim, flush_state); | |
4842 | flush_state++; | |
25ce459c LB |
4843 | if (!btrfs_need_do_async_reclaim(space_info, fs_info, |
4844 | flush_state)) | |
21c7e756 | 4845 | return; |
365c5313 | 4846 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
4847 | } |
4848 | ||
4849 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
4850 | { | |
4851 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
4852 | } | |
4853 | ||
4a92b1b8 JB |
4854 | /** |
4855 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
4856 | * @root - the root we're allocating for | |
4857 | * @block_rsv - the block_rsv we're allocating for | |
4858 | * @orig_bytes - the number of bytes we want | |
48fc7f7e | 4859 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 4860 | * |
4a92b1b8 JB |
4861 | * This will reserve orgi_bytes number of bytes from the space info associated |
4862 | * with the block_rsv. If there is not enough space it will make an attempt to | |
4863 | * flush out space to make room. It will do this by flushing delalloc if | |
4864 | * possible or committing the transaction. If flush is 0 then no attempts to | |
4865 | * regain reservations will be made and this will fail if there is not enough | |
4866 | * space already. | |
8bb8ab2e | 4867 | */ |
4a92b1b8 | 4868 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e | 4869 | struct btrfs_block_rsv *block_rsv, |
08e007d2 MX |
4870 | u64 orig_bytes, |
4871 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4872 | { |
f0486c68 | 4873 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 4874 | u64 used; |
8bb8ab2e | 4875 | u64 num_bytes = orig_bytes; |
67b0fd63 | 4876 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
8bb8ab2e | 4877 | int ret = 0; |
fdb5effd | 4878 | bool flushing = false; |
9ed74f2d | 4879 | |
8bb8ab2e | 4880 | again: |
fdb5effd | 4881 | ret = 0; |
8bb8ab2e | 4882 | spin_lock(&space_info->lock); |
fdb5effd | 4883 | /* |
08e007d2 MX |
4884 | * We only want to wait if somebody other than us is flushing and we |
4885 | * are actually allowed to flush all things. | |
fdb5effd | 4886 | */ |
08e007d2 MX |
4887 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && |
4888 | space_info->flush) { | |
fdb5effd JB |
4889 | spin_unlock(&space_info->lock); |
4890 | /* | |
4891 | * If we have a trans handle we can't wait because the flusher | |
4892 | * may have to commit the transaction, which would mean we would | |
4893 | * deadlock since we are waiting for the flusher to finish, but | |
4894 | * hold the current transaction open. | |
4895 | */ | |
663350ac | 4896 | if (current->journal_info) |
fdb5effd | 4897 | return -EAGAIN; |
b9688bb8 AJ |
4898 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
4899 | /* Must have been killed, return */ | |
4900 | if (ret) | |
fdb5effd JB |
4901 | return -EINTR; |
4902 | ||
4903 | spin_lock(&space_info->lock); | |
4904 | } | |
4905 | ||
4906 | ret = -ENOSPC; | |
2bf64758 JB |
4907 | used = space_info->bytes_used + space_info->bytes_reserved + |
4908 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4909 | space_info->bytes_may_use; | |
9ed74f2d | 4910 | |
8bb8ab2e JB |
4911 | /* |
4912 | * The idea here is that we've not already over-reserved the block group | |
4913 | * then we can go ahead and save our reservation first and then start | |
4914 | * flushing if we need to. Otherwise if we've already overcommitted | |
4915 | * lets start flushing stuff first and then come back and try to make | |
4916 | * our reservation. | |
4917 | */ | |
2bf64758 JB |
4918 | if (used <= space_info->total_bytes) { |
4919 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 4920 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 4921 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 4922 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
4923 | ret = 0; |
4924 | } else { | |
4925 | /* | |
4926 | * Ok set num_bytes to orig_bytes since we aren't | |
4927 | * overocmmitted, this way we only try and reclaim what | |
4928 | * we need. | |
4929 | */ | |
4930 | num_bytes = orig_bytes; | |
4931 | } | |
4932 | } else { | |
4933 | /* | |
4934 | * Ok we're over committed, set num_bytes to the overcommitted | |
4935 | * amount plus the amount of bytes that we need for this | |
4936 | * reservation. | |
4937 | */ | |
2bf64758 | 4938 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 4939 | (orig_bytes * 2); |
8bb8ab2e | 4940 | } |
9ed74f2d | 4941 | |
44734ed1 JB |
4942 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { |
4943 | space_info->bytes_may_use += orig_bytes; | |
4944 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4945 | space_info->flags, orig_bytes, | |
4946 | 1); | |
4947 | ret = 0; | |
2bf64758 JB |
4948 | } |
4949 | ||
8bb8ab2e JB |
4950 | /* |
4951 | * Couldn't make our reservation, save our place so while we're trying | |
4952 | * to reclaim space we can actually use it instead of somebody else | |
4953 | * stealing it from us. | |
08e007d2 MX |
4954 | * |
4955 | * We make the other tasks wait for the flush only when we can flush | |
4956 | * all things. | |
8bb8ab2e | 4957 | */ |
72bcd99d | 4958 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
fdb5effd JB |
4959 | flushing = true; |
4960 | space_info->flush = 1; | |
21c7e756 MX |
4961 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
4962 | used += orig_bytes; | |
f6acfd50 JB |
4963 | /* |
4964 | * We will do the space reservation dance during log replay, | |
4965 | * which means we won't have fs_info->fs_root set, so don't do | |
4966 | * the async reclaim as we will panic. | |
4967 | */ | |
4968 | if (!root->fs_info->log_root_recovering && | |
4969 | need_do_async_reclaim(space_info, root->fs_info, used) && | |
21c7e756 MX |
4970 | !work_busy(&root->fs_info->async_reclaim_work)) |
4971 | queue_work(system_unbound_wq, | |
4972 | &root->fs_info->async_reclaim_work); | |
8bb8ab2e | 4973 | } |
f0486c68 | 4974 | spin_unlock(&space_info->lock); |
9ed74f2d | 4975 | |
08e007d2 | 4976 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
8bb8ab2e | 4977 | goto out; |
f0486c68 | 4978 | |
96c3f433 JB |
4979 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
4980 | flush_state); | |
4981 | flush_state++; | |
08e007d2 MX |
4982 | |
4983 | /* | |
4984 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock | |
4985 | * would happen. So skip delalloc flush. | |
4986 | */ | |
4987 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
4988 | (flush_state == FLUSH_DELALLOC || | |
4989 | flush_state == FLUSH_DELALLOC_WAIT)) | |
4990 | flush_state = ALLOC_CHUNK; | |
4991 | ||
96c3f433 | 4992 | if (!ret) |
8bb8ab2e | 4993 | goto again; |
08e007d2 MX |
4994 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
4995 | flush_state < COMMIT_TRANS) | |
4996 | goto again; | |
4997 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
4998 | flush_state <= COMMIT_TRANS) | |
8bb8ab2e JB |
4999 | goto again; |
5000 | ||
5001 | out: | |
5d80366e JB |
5002 | if (ret == -ENOSPC && |
5003 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5004 | struct btrfs_block_rsv *global_rsv = | |
5005 | &root->fs_info->global_block_rsv; | |
5006 | ||
5007 | if (block_rsv != global_rsv && | |
5008 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5009 | ret = 0; | |
5010 | } | |
cab45e22 JM |
5011 | if (ret == -ENOSPC) |
5012 | trace_btrfs_space_reservation(root->fs_info, | |
5013 | "space_info:enospc", | |
5014 | space_info->flags, orig_bytes, 1); | |
fdb5effd | 5015 | if (flushing) { |
8bb8ab2e | 5016 | spin_lock(&space_info->lock); |
fdb5effd JB |
5017 | space_info->flush = 0; |
5018 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 5019 | spin_unlock(&space_info->lock); |
f0486c68 | 5020 | } |
f0486c68 YZ |
5021 | return ret; |
5022 | } | |
5023 | ||
79787eaa JM |
5024 | static struct btrfs_block_rsv *get_block_rsv( |
5025 | const struct btrfs_trans_handle *trans, | |
5026 | const struct btrfs_root *root) | |
f0486c68 | 5027 | { |
4c13d758 JB |
5028 | struct btrfs_block_rsv *block_rsv = NULL; |
5029 | ||
e9cf439f AM |
5030 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
5031 | (root == root->fs_info->csum_root && trans->adding_csums) || | |
5032 | (root == root->fs_info->uuid_root)) | |
f7a81ea4 SB |
5033 | block_rsv = trans->block_rsv; |
5034 | ||
4c13d758 | 5035 | if (!block_rsv) |
f0486c68 YZ |
5036 | block_rsv = root->block_rsv; |
5037 | ||
5038 | if (!block_rsv) | |
5039 | block_rsv = &root->fs_info->empty_block_rsv; | |
5040 | ||
5041 | return block_rsv; | |
5042 | } | |
5043 | ||
5044 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5045 | u64 num_bytes) | |
5046 | { | |
5047 | int ret = -ENOSPC; | |
5048 | spin_lock(&block_rsv->lock); | |
5049 | if (block_rsv->reserved >= num_bytes) { | |
5050 | block_rsv->reserved -= num_bytes; | |
5051 | if (block_rsv->reserved < block_rsv->size) | |
5052 | block_rsv->full = 0; | |
5053 | ret = 0; | |
5054 | } | |
5055 | spin_unlock(&block_rsv->lock); | |
5056 | return ret; | |
5057 | } | |
5058 | ||
5059 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5060 | u64 num_bytes, int update_size) | |
5061 | { | |
5062 | spin_lock(&block_rsv->lock); | |
5063 | block_rsv->reserved += num_bytes; | |
5064 | if (update_size) | |
5065 | block_rsv->size += num_bytes; | |
5066 | else if (block_rsv->reserved >= block_rsv->size) | |
5067 | block_rsv->full = 1; | |
5068 | spin_unlock(&block_rsv->lock); | |
5069 | } | |
5070 | ||
d52be818 JB |
5071 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5072 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5073 | int min_factor) | |
5074 | { | |
5075 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5076 | u64 min_bytes; | |
5077 | ||
5078 | if (global_rsv->space_info != dest->space_info) | |
5079 | return -ENOSPC; | |
5080 | ||
5081 | spin_lock(&global_rsv->lock); | |
5082 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5083 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5084 | spin_unlock(&global_rsv->lock); | |
5085 | return -ENOSPC; | |
5086 | } | |
5087 | global_rsv->reserved -= num_bytes; | |
5088 | if (global_rsv->reserved < global_rsv->size) | |
5089 | global_rsv->full = 0; | |
5090 | spin_unlock(&global_rsv->lock); | |
5091 | ||
5092 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5093 | return 0; | |
5094 | } | |
5095 | ||
8c2a3ca2 JB |
5096 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
5097 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 5098 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
5099 | { |
5100 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
5101 | ||
5102 | spin_lock(&block_rsv->lock); | |
5103 | if (num_bytes == (u64)-1) | |
5104 | num_bytes = block_rsv->size; | |
5105 | block_rsv->size -= num_bytes; | |
5106 | if (block_rsv->reserved >= block_rsv->size) { | |
5107 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5108 | block_rsv->reserved = block_rsv->size; | |
5109 | block_rsv->full = 1; | |
5110 | } else { | |
5111 | num_bytes = 0; | |
5112 | } | |
5113 | spin_unlock(&block_rsv->lock); | |
5114 | ||
5115 | if (num_bytes > 0) { | |
5116 | if (dest) { | |
e9e22899 JB |
5117 | spin_lock(&dest->lock); |
5118 | if (!dest->full) { | |
5119 | u64 bytes_to_add; | |
5120 | ||
5121 | bytes_to_add = dest->size - dest->reserved; | |
5122 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5123 | dest->reserved += bytes_to_add; | |
5124 | if (dest->reserved >= dest->size) | |
5125 | dest->full = 1; | |
5126 | num_bytes -= bytes_to_add; | |
5127 | } | |
5128 | spin_unlock(&dest->lock); | |
5129 | } | |
5130 | if (num_bytes) { | |
f0486c68 | 5131 | spin_lock(&space_info->lock); |
fb25e914 | 5132 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5133 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5134 | space_info->flags, num_bytes, 0); |
f0486c68 | 5135 | spin_unlock(&space_info->lock); |
4e06bdd6 | 5136 | } |
9ed74f2d | 5137 | } |
f0486c68 | 5138 | } |
4e06bdd6 | 5139 | |
f0486c68 YZ |
5140 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
5141 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
5142 | { | |
5143 | int ret; | |
9ed74f2d | 5144 | |
f0486c68 YZ |
5145 | ret = block_rsv_use_bytes(src, num_bytes); |
5146 | if (ret) | |
5147 | return ret; | |
9ed74f2d | 5148 | |
f0486c68 | 5149 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
5150 | return 0; |
5151 | } | |
5152 | ||
66d8f3dd | 5153 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5154 | { |
f0486c68 YZ |
5155 | memset(rsv, 0, sizeof(*rsv)); |
5156 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5157 | rsv->type = type; |
f0486c68 YZ |
5158 | } |
5159 | ||
66d8f3dd MX |
5160 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
5161 | unsigned short type) | |
f0486c68 YZ |
5162 | { |
5163 | struct btrfs_block_rsv *block_rsv; | |
5164 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 5165 | |
f0486c68 YZ |
5166 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5167 | if (!block_rsv) | |
5168 | return NULL; | |
9ed74f2d | 5169 | |
66d8f3dd | 5170 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
5171 | block_rsv->space_info = __find_space_info(fs_info, |
5172 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
5173 | return block_rsv; |
5174 | } | |
9ed74f2d | 5175 | |
f0486c68 YZ |
5176 | void btrfs_free_block_rsv(struct btrfs_root *root, |
5177 | struct btrfs_block_rsv *rsv) | |
5178 | { | |
2aaa6655 JB |
5179 | if (!rsv) |
5180 | return; | |
dabdb640 JB |
5181 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
5182 | kfree(rsv); | |
9ed74f2d JB |
5183 | } |
5184 | ||
cdfb080e CM |
5185 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5186 | { | |
5187 | kfree(rsv); | |
5188 | } | |
5189 | ||
08e007d2 MX |
5190 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5191 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5192 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5193 | { |
f0486c68 | 5194 | int ret; |
9ed74f2d | 5195 | |
f0486c68 YZ |
5196 | if (num_bytes == 0) |
5197 | return 0; | |
8bb8ab2e | 5198 | |
61b520a9 | 5199 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5200 | if (!ret) { |
5201 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5202 | return 0; | |
5203 | } | |
9ed74f2d | 5204 | |
f0486c68 | 5205 | return ret; |
f0486c68 | 5206 | } |
9ed74f2d | 5207 | |
4a92b1b8 | 5208 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 5209 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5210 | { |
5211 | u64 num_bytes = 0; | |
f0486c68 | 5212 | int ret = -ENOSPC; |
9ed74f2d | 5213 | |
f0486c68 YZ |
5214 | if (!block_rsv) |
5215 | return 0; | |
9ed74f2d | 5216 | |
f0486c68 | 5217 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5218 | num_bytes = div_factor(block_rsv->size, min_factor); |
5219 | if (block_rsv->reserved >= num_bytes) | |
5220 | ret = 0; | |
5221 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5222 | |
36ba022a JB |
5223 | return ret; |
5224 | } | |
5225 | ||
08e007d2 MX |
5226 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5227 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5228 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5229 | { |
5230 | u64 num_bytes = 0; | |
5231 | int ret = -ENOSPC; | |
5232 | ||
5233 | if (!block_rsv) | |
5234 | return 0; | |
5235 | ||
5236 | spin_lock(&block_rsv->lock); | |
5237 | num_bytes = min_reserved; | |
13553e52 | 5238 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5239 | ret = 0; |
13553e52 | 5240 | else |
f0486c68 | 5241 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5242 | spin_unlock(&block_rsv->lock); |
13553e52 | 5243 | |
f0486c68 YZ |
5244 | if (!ret) |
5245 | return 0; | |
5246 | ||
aa38a711 | 5247 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5248 | if (!ret) { |
5249 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5250 | return 0; |
6a63209f | 5251 | } |
9ed74f2d | 5252 | |
13553e52 | 5253 | return ret; |
f0486c68 YZ |
5254 | } |
5255 | ||
5256 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
5257 | struct btrfs_block_rsv *dst_rsv, | |
5258 | u64 num_bytes) | |
5259 | { | |
5260 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
5261 | } | |
5262 | ||
5263 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
5264 | struct btrfs_block_rsv *block_rsv, | |
5265 | u64 num_bytes) | |
5266 | { | |
5267 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
17504584 | 5268 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5269 | block_rsv->space_info != global_rsv->space_info) |
5270 | global_rsv = NULL; | |
8c2a3ca2 JB |
5271 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
5272 | num_bytes); | |
6a63209f JB |
5273 | } |
5274 | ||
5275 | /* | |
8929ecfa YZ |
5276 | * helper to calculate size of global block reservation. |
5277 | * the desired value is sum of space used by extent tree, | |
5278 | * checksum tree and root tree | |
6a63209f | 5279 | */ |
8929ecfa | 5280 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 5281 | { |
8929ecfa YZ |
5282 | struct btrfs_space_info *sinfo; |
5283 | u64 num_bytes; | |
5284 | u64 meta_used; | |
5285 | u64 data_used; | |
6c41761f | 5286 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 5287 | |
8929ecfa YZ |
5288 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
5289 | spin_lock(&sinfo->lock); | |
5290 | data_used = sinfo->bytes_used; | |
5291 | spin_unlock(&sinfo->lock); | |
33b4d47f | 5292 | |
8929ecfa YZ |
5293 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
5294 | spin_lock(&sinfo->lock); | |
6d48755d JB |
5295 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
5296 | data_used = 0; | |
8929ecfa YZ |
5297 | meta_used = sinfo->bytes_used; |
5298 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 5299 | |
8929ecfa YZ |
5300 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
5301 | csum_size * 2; | |
f8c269d7 | 5302 | num_bytes += div_u64(data_used + meta_used, 50); |
4e06bdd6 | 5303 | |
8929ecfa | 5304 | if (num_bytes * 3 > meta_used) |
f8c269d7 | 5305 | num_bytes = div_u64(meta_used, 3); |
ab6e2410 | 5306 | |
707e8a07 | 5307 | return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10); |
8929ecfa | 5308 | } |
6a63209f | 5309 | |
8929ecfa YZ |
5310 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5311 | { | |
5312 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5313 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5314 | u64 num_bytes; | |
6a63209f | 5315 | |
8929ecfa | 5316 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 5317 | |
8929ecfa | 5318 | spin_lock(&sinfo->lock); |
1f699d38 | 5319 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5320 | |
fdf30d1c | 5321 | block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024); |
4e06bdd6 | 5322 | |
8929ecfa | 5323 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
5324 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
5325 | sinfo->bytes_may_use; | |
8929ecfa YZ |
5326 | |
5327 | if (sinfo->total_bytes > num_bytes) { | |
5328 | num_bytes = sinfo->total_bytes - num_bytes; | |
5329 | block_rsv->reserved += num_bytes; | |
fb25e914 | 5330 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 5331 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5332 | sinfo->flags, num_bytes, 1); |
6a63209f | 5333 | } |
6a63209f | 5334 | |
8929ecfa YZ |
5335 | if (block_rsv->reserved >= block_rsv->size) { |
5336 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 5337 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5338 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5339 | sinfo->flags, num_bytes, 0); |
8929ecfa YZ |
5340 | block_rsv->reserved = block_rsv->size; |
5341 | block_rsv->full = 1; | |
5342 | } | |
182608c8 | 5343 | |
8929ecfa | 5344 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5345 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5346 | } |
5347 | ||
f0486c68 | 5348 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5349 | { |
f0486c68 | 5350 | struct btrfs_space_info *space_info; |
6a63209f | 5351 | |
f0486c68 YZ |
5352 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5353 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5354 | |
f0486c68 | 5355 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5356 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 5357 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
5358 | fs_info->trans_block_rsv.space_info = space_info; |
5359 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5360 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5361 | |
8929ecfa YZ |
5362 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5363 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5364 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5365 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5366 | if (fs_info->quota_root) |
5367 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5368 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5369 | |
8929ecfa | 5370 | update_global_block_rsv(fs_info); |
6a63209f JB |
5371 | } |
5372 | ||
8929ecfa | 5373 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5374 | { |
8c2a3ca2 JB |
5375 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5376 | (u64)-1); | |
8929ecfa YZ |
5377 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
5378 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
5379 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
5380 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5381 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5382 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5383 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5384 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5385 | } |
5386 | ||
a22285a6 YZ |
5387 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
5388 | struct btrfs_root *root) | |
6a63209f | 5389 | { |
0e721106 JB |
5390 | if (!trans->block_rsv) |
5391 | return; | |
5392 | ||
a22285a6 YZ |
5393 | if (!trans->bytes_reserved) |
5394 | return; | |
6a63209f | 5395 | |
e77266e4 | 5396 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 5397 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 5398 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
5399 | trans->bytes_reserved = 0; |
5400 | } | |
6a63209f | 5401 | |
4fbcdf66 FM |
5402 | /* |
5403 | * To be called after all the new block groups attached to the transaction | |
5404 | * handle have been created (btrfs_create_pending_block_groups()). | |
5405 | */ | |
5406 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5407 | { | |
5408 | struct btrfs_fs_info *fs_info = trans->root->fs_info; | |
5409 | ||
5410 | if (!trans->chunk_bytes_reserved) | |
5411 | return; | |
5412 | ||
5413 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5414 | ||
5415 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5416 | trans->chunk_bytes_reserved); | |
5417 | trans->chunk_bytes_reserved = 0; | |
5418 | } | |
5419 | ||
79787eaa | 5420 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
5421 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
5422 | struct inode *inode) | |
5423 | { | |
5424 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5425 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
5426 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
5427 | ||
5428 | /* | |
fcb80c2a JB |
5429 | * We need to hold space in order to delete our orphan item once we've |
5430 | * added it, so this takes the reservation so we can release it later | |
5431 | * when we are truly done with the orphan item. | |
d68fc57b | 5432 | */ |
ff5714cc | 5433 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
5434 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
5435 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 5436 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
5437 | } |
5438 | ||
d68fc57b | 5439 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 5440 | { |
d68fc57b | 5441 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 5442 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
5443 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
5444 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
5445 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
5446 | } | |
97e728d4 | 5447 | |
d5c12070 MX |
5448 | /* |
5449 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5450 | * root: the root of the parent directory | |
5451 | * rsv: block reservation | |
5452 | * items: the number of items that we need do reservation | |
5453 | * qgroup_reserved: used to return the reserved size in qgroup | |
5454 | * | |
5455 | * This function is used to reserve the space for snapshot/subvolume | |
5456 | * creation and deletion. Those operations are different with the | |
5457 | * common file/directory operations, they change two fs/file trees | |
5458 | * and root tree, the number of items that the qgroup reserves is | |
5459 | * different with the free space reservation. So we can not use | |
5460 | * the space reseravtion mechanism in start_transaction(). | |
5461 | */ | |
5462 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5463 | struct btrfs_block_rsv *rsv, | |
5464 | int items, | |
ee3441b4 JM |
5465 | u64 *qgroup_reserved, |
5466 | bool use_global_rsv) | |
a22285a6 | 5467 | { |
d5c12070 MX |
5468 | u64 num_bytes; |
5469 | int ret; | |
ee3441b4 | 5470 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
d5c12070 MX |
5471 | |
5472 | if (root->fs_info->quota_enabled) { | |
5473 | /* One for parent inode, two for dir entries */ | |
707e8a07 | 5474 | num_bytes = 3 * root->nodesize; |
7174109c | 5475 | ret = btrfs_qgroup_reserve_meta(root, num_bytes); |
d5c12070 MX |
5476 | if (ret) |
5477 | return ret; | |
5478 | } else { | |
5479 | num_bytes = 0; | |
5480 | } | |
5481 | ||
5482 | *qgroup_reserved = num_bytes; | |
5483 | ||
5484 | num_bytes = btrfs_calc_trans_metadata_size(root, items); | |
5485 | rsv->space_info = __find_space_info(root->fs_info, | |
5486 | BTRFS_BLOCK_GROUP_METADATA); | |
5487 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5488 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5489 | |
5490 | if (ret == -ENOSPC && use_global_rsv) | |
5491 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes); | |
5492 | ||
7174109c QW |
5493 | if (ret && *qgroup_reserved) |
5494 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
d5c12070 MX |
5495 | |
5496 | return ret; | |
5497 | } | |
5498 | ||
5499 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, | |
5500 | struct btrfs_block_rsv *rsv, | |
5501 | u64 qgroup_reserved) | |
5502 | { | |
5503 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
97e728d4 JB |
5504 | } |
5505 | ||
7709cde3 JB |
5506 | /** |
5507 | * drop_outstanding_extent - drop an outstanding extent | |
5508 | * @inode: the inode we're dropping the extent for | |
dcab6a3b | 5509 | * @num_bytes: the number of bytes we're relaseing. |
7709cde3 JB |
5510 | * |
5511 | * This is called when we are freeing up an outstanding extent, either called | |
5512 | * after an error or after an extent is written. This will return the number of | |
5513 | * reserved extents that need to be freed. This must be called with | |
5514 | * BTRFS_I(inode)->lock held. | |
5515 | */ | |
dcab6a3b | 5516 | static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes) |
9e0baf60 | 5517 | { |
7fd2ae21 | 5518 | unsigned drop_inode_space = 0; |
9e0baf60 | 5519 | unsigned dropped_extents = 0; |
dcab6a3b | 5520 | unsigned num_extents = 0; |
9e0baf60 | 5521 | |
dcab6a3b JB |
5522 | num_extents = (unsigned)div64_u64(num_bytes + |
5523 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5524 | BTRFS_MAX_EXTENT_SIZE); | |
5525 | ASSERT(num_extents); | |
5526 | ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents); | |
5527 | BTRFS_I(inode)->outstanding_extents -= num_extents; | |
9e0baf60 | 5528 | |
7fd2ae21 | 5529 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
5530 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5531 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 5532 | drop_inode_space = 1; |
7fd2ae21 | 5533 | |
9e0baf60 JB |
5534 | /* |
5535 | * If we have more or the same amount of outsanding extents than we have | |
5536 | * reserved then we need to leave the reserved extents count alone. | |
5537 | */ | |
5538 | if (BTRFS_I(inode)->outstanding_extents >= | |
5539 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 5540 | return drop_inode_space; |
9e0baf60 JB |
5541 | |
5542 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
5543 | BTRFS_I(inode)->outstanding_extents; | |
5544 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 5545 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
5546 | } |
5547 | ||
7709cde3 JB |
5548 | /** |
5549 | * calc_csum_metadata_size - return the amount of metada space that must be | |
5550 | * reserved/free'd for the given bytes. | |
5551 | * @inode: the inode we're manipulating | |
5552 | * @num_bytes: the number of bytes in question | |
5553 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
5554 | * | |
5555 | * This adjusts the number of csum_bytes in the inode and then returns the | |
5556 | * correct amount of metadata that must either be reserved or freed. We | |
5557 | * calculate how many checksums we can fit into one leaf and then divide the | |
5558 | * number of bytes that will need to be checksumed by this value to figure out | |
5559 | * how many checksums will be required. If we are adding bytes then the number | |
5560 | * may go up and we will return the number of additional bytes that must be | |
5561 | * reserved. If it is going down we will return the number of bytes that must | |
5562 | * be freed. | |
5563 | * | |
5564 | * This must be called with BTRFS_I(inode)->lock held. | |
5565 | */ | |
5566 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
5567 | int reserve) | |
6324fbf3 | 5568 | { |
7709cde3 | 5569 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1262133b | 5570 | u64 old_csums, num_csums; |
7709cde3 JB |
5571 | |
5572 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
5573 | BTRFS_I(inode)->csum_bytes == 0) | |
5574 | return 0; | |
5575 | ||
28f75a0e | 5576 | old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); |
7709cde3 JB |
5577 | if (reserve) |
5578 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
5579 | else | |
5580 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
28f75a0e | 5581 | num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); |
7709cde3 JB |
5582 | |
5583 | /* No change, no need to reserve more */ | |
5584 | if (old_csums == num_csums) | |
5585 | return 0; | |
5586 | ||
5587 | if (reserve) | |
5588 | return btrfs_calc_trans_metadata_size(root, | |
5589 | num_csums - old_csums); | |
5590 | ||
5591 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 5592 | } |
c146afad | 5593 | |
0ca1f7ce YZ |
5594 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
5595 | { | |
5596 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5597 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 5598 | u64 to_reserve = 0; |
660d3f6c | 5599 | u64 csum_bytes; |
9e0baf60 | 5600 | unsigned nr_extents = 0; |
660d3f6c | 5601 | int extra_reserve = 0; |
08e007d2 | 5602 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5603 | int ret = 0; |
c64c2bd8 | 5604 | bool delalloc_lock = true; |
88e081bf WS |
5605 | u64 to_free = 0; |
5606 | unsigned dropped; | |
6324fbf3 | 5607 | |
c64c2bd8 JB |
5608 | /* If we are a free space inode we need to not flush since we will be in |
5609 | * the middle of a transaction commit. We also don't need the delalloc | |
5610 | * mutex since we won't race with anybody. We need this mostly to make | |
5611 | * lockdep shut its filthy mouth. | |
5612 | */ | |
5613 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 5614 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 JB |
5615 | delalloc_lock = false; |
5616 | } | |
c09544e0 | 5617 | |
08e007d2 MX |
5618 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
5619 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 5620 | schedule_timeout(1); |
ec44a35c | 5621 | |
c64c2bd8 JB |
5622 | if (delalloc_lock) |
5623 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
5624 | ||
0ca1f7ce | 5625 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 5626 | |
9e0baf60 | 5627 | spin_lock(&BTRFS_I(inode)->lock); |
6a41dd09 JB |
5628 | nr_extents = (unsigned)div64_u64(num_bytes + |
5629 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5630 | BTRFS_MAX_EXTENT_SIZE); | |
5631 | BTRFS_I(inode)->outstanding_extents += nr_extents; | |
5632 | nr_extents = 0; | |
9e0baf60 JB |
5633 | |
5634 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 5635 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
5636 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
5637 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 5638 | |
7fd2ae21 JB |
5639 | /* |
5640 | * Add an item to reserve for updating the inode when we complete the | |
5641 | * delalloc io. | |
5642 | */ | |
72ac3c0d JB |
5643 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5644 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 5645 | nr_extents++; |
660d3f6c | 5646 | extra_reserve = 1; |
593060d7 | 5647 | } |
7fd2ae21 JB |
5648 | |
5649 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 5650 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 5651 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 5652 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 5653 | |
88e081bf | 5654 | if (root->fs_info->quota_enabled) { |
7174109c QW |
5655 | ret = btrfs_qgroup_reserve_meta(root, |
5656 | nr_extents * root->nodesize); | |
88e081bf WS |
5657 | if (ret) |
5658 | goto out_fail; | |
5659 | } | |
c5567237 | 5660 | |
88e081bf WS |
5661 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
5662 | if (unlikely(ret)) { | |
7174109c | 5663 | btrfs_qgroup_free_meta(root, nr_extents * root->nodesize); |
88e081bf | 5664 | goto out_fail; |
9e0baf60 | 5665 | } |
25179201 | 5666 | |
660d3f6c JB |
5667 | spin_lock(&BTRFS_I(inode)->lock); |
5668 | if (extra_reserve) { | |
72ac3c0d JB |
5669 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5670 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
5671 | nr_extents--; |
5672 | } | |
5673 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
5674 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
5675 | |
5676 | if (delalloc_lock) | |
5677 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 5678 | |
8c2a3ca2 | 5679 | if (to_reserve) |
67871254 | 5680 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
8c2a3ca2 | 5681 | btrfs_ino(inode), to_reserve, 1); |
0ca1f7ce YZ |
5682 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
5683 | ||
0ca1f7ce | 5684 | return 0; |
88e081bf WS |
5685 | |
5686 | out_fail: | |
5687 | spin_lock(&BTRFS_I(inode)->lock); | |
dcab6a3b | 5688 | dropped = drop_outstanding_extent(inode, num_bytes); |
88e081bf WS |
5689 | /* |
5690 | * If the inodes csum_bytes is the same as the original | |
5691 | * csum_bytes then we know we haven't raced with any free()ers | |
5692 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 5693 | */ |
f4881bc7 | 5694 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { |
88e081bf | 5695 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 JB |
5696 | } else { |
5697 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; | |
5698 | u64 bytes; | |
5699 | ||
5700 | /* | |
5701 | * This is tricky, but first we need to figure out how much we | |
5702 | * free'd from any free-ers that occured during this | |
5703 | * reservation, so we reset ->csum_bytes to the csum_bytes | |
5704 | * before we dropped our lock, and then call the free for the | |
5705 | * number of bytes that were freed while we were trying our | |
5706 | * reservation. | |
5707 | */ | |
5708 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; | |
5709 | BTRFS_I(inode)->csum_bytes = csum_bytes; | |
5710 | to_free = calc_csum_metadata_size(inode, bytes, 0); | |
5711 | ||
5712 | ||
5713 | /* | |
5714 | * Now we need to see how much we would have freed had we not | |
5715 | * been making this reservation and our ->csum_bytes were not | |
5716 | * artificially inflated. | |
5717 | */ | |
5718 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; | |
5719 | bytes = csum_bytes - orig_csum_bytes; | |
5720 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
5721 | ||
5722 | /* | |
5723 | * Now reset ->csum_bytes to what it should be. If bytes is | |
5724 | * more than to_free then we would have free'd more space had we | |
5725 | * not had an artificially high ->csum_bytes, so we need to free | |
5726 | * the remainder. If bytes is the same or less then we don't | |
5727 | * need to do anything, the other free-ers did the correct | |
5728 | * thing. | |
5729 | */ | |
5730 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; | |
5731 | if (bytes > to_free) | |
5732 | to_free = bytes - to_free; | |
5733 | else | |
5734 | to_free = 0; | |
5735 | } | |
88e081bf | 5736 | spin_unlock(&BTRFS_I(inode)->lock); |
e2d1f923 | 5737 | if (dropped) |
88e081bf WS |
5738 | to_free += btrfs_calc_trans_metadata_size(root, dropped); |
5739 | ||
5740 | if (to_free) { | |
5741 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
5742 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5743 | btrfs_ino(inode), to_free, 0); | |
5744 | } | |
5745 | if (delalloc_lock) | |
5746 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
5747 | return ret; | |
0ca1f7ce YZ |
5748 | } |
5749 | ||
7709cde3 JB |
5750 | /** |
5751 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
5752 | * @inode: the inode to release the reservation for | |
5753 | * @num_bytes: the number of bytes we're releasing | |
5754 | * | |
5755 | * This will release the metadata reservation for an inode. This can be called | |
5756 | * once we complete IO for a given set of bytes to release their metadata | |
5757 | * reservations. | |
5758 | */ | |
0ca1f7ce YZ |
5759 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
5760 | { | |
5761 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
5762 | u64 to_free = 0; |
5763 | unsigned dropped; | |
0ca1f7ce YZ |
5764 | |
5765 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 5766 | spin_lock(&BTRFS_I(inode)->lock); |
dcab6a3b | 5767 | dropped = drop_outstanding_extent(inode, num_bytes); |
97e728d4 | 5768 | |
0934856d MX |
5769 | if (num_bytes) |
5770 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
7709cde3 | 5771 | spin_unlock(&BTRFS_I(inode)->lock); |
9e0baf60 JB |
5772 | if (dropped > 0) |
5773 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 5774 | |
6a3891c5 JB |
5775 | if (btrfs_test_is_dummy_root(root)) |
5776 | return; | |
5777 | ||
8c2a3ca2 JB |
5778 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
5779 | btrfs_ino(inode), to_free, 0); | |
c5567237 | 5780 | |
0ca1f7ce YZ |
5781 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
5782 | to_free); | |
5783 | } | |
5784 | ||
1ada3a62 | 5785 | /** |
7cf5b976 | 5786 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
5787 | * delalloc |
5788 | * @inode: inode we're writing to | |
5789 | * @start: start range we are writing to | |
5790 | * @len: how long the range we are writing to | |
5791 | * | |
5792 | * TODO: This function will finally replace old btrfs_delalloc_reserve_space() | |
5793 | * | |
5794 | * This will do the following things | |
5795 | * | |
5796 | * o reserve space in data space info for num bytes | |
5797 | * and reserve precious corresponding qgroup space | |
5798 | * (Done in check_data_free_space) | |
5799 | * | |
5800 | * o reserve space for metadata space, based on the number of outstanding | |
5801 | * extents and how much csums will be needed | |
5802 | * also reserve metadata space in a per root over-reserve method. | |
5803 | * o add to the inodes->delalloc_bytes | |
5804 | * o add it to the fs_info's delalloc inodes list. | |
5805 | * (Above 3 all done in delalloc_reserve_metadata) | |
5806 | * | |
5807 | * Return 0 for success | |
5808 | * Return <0 for error(-ENOSPC or -EQUOT) | |
5809 | */ | |
7cf5b976 | 5810 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
5811 | { |
5812 | int ret; | |
5813 | ||
7cf5b976 | 5814 | ret = btrfs_check_data_free_space(inode, start, len); |
1ada3a62 QW |
5815 | if (ret < 0) |
5816 | return ret; | |
5817 | ret = btrfs_delalloc_reserve_metadata(inode, len); | |
5818 | if (ret < 0) | |
7cf5b976 | 5819 | btrfs_free_reserved_data_space(inode, start, len); |
1ada3a62 QW |
5820 | return ret; |
5821 | } | |
5822 | ||
7709cde3 | 5823 | /** |
7cf5b976 | 5824 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
5825 | * @inode: inode we're releasing space for |
5826 | * @start: start position of the space already reserved | |
5827 | * @len: the len of the space already reserved | |
5828 | * | |
5829 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
5830 | * called in the case that we don't need the metadata AND data reservations | |
5831 | * anymore. So if there is an error or we insert an inline extent. | |
5832 | * | |
5833 | * This function will release the metadata space that was not used and will | |
5834 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
5835 | * list if there are no delalloc bytes left. | |
5836 | * Also it will handle the qgroup reserved space. | |
5837 | */ | |
7cf5b976 | 5838 | void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
5839 | { |
5840 | btrfs_delalloc_release_metadata(inode, len); | |
7cf5b976 | 5841 | btrfs_free_reserved_data_space(inode, start, len); |
6324fbf3 CM |
5842 | } |
5843 | ||
ce93ec54 JB |
5844 | static int update_block_group(struct btrfs_trans_handle *trans, |
5845 | struct btrfs_root *root, u64 bytenr, | |
5846 | u64 num_bytes, int alloc) | |
9078a3e1 | 5847 | { |
0af3d00b | 5848 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 5849 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 5850 | u64 total = num_bytes; |
9078a3e1 | 5851 | u64 old_val; |
db94535d | 5852 | u64 byte_in_group; |
0af3d00b | 5853 | int factor; |
3e1ad54f | 5854 | |
5d4f98a2 | 5855 | /* block accounting for super block */ |
eb73c1b7 | 5856 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 5857 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
5858 | if (alloc) |
5859 | old_val += num_bytes; | |
5860 | else | |
5861 | old_val -= num_bytes; | |
6c41761f | 5862 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 5863 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 5864 | |
d397712b | 5865 | while (total) { |
db94535d | 5866 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 5867 | if (!cache) |
79787eaa | 5868 | return -ENOENT; |
b742bb82 YZ |
5869 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
5870 | BTRFS_BLOCK_GROUP_RAID1 | | |
5871 | BTRFS_BLOCK_GROUP_RAID10)) | |
5872 | factor = 2; | |
5873 | else | |
5874 | factor = 1; | |
9d66e233 JB |
5875 | /* |
5876 | * If this block group has free space cache written out, we | |
5877 | * need to make sure to load it if we are removing space. This | |
5878 | * is because we need the unpinning stage to actually add the | |
5879 | * space back to the block group, otherwise we will leak space. | |
5880 | */ | |
5881 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 5882 | cache_block_group(cache, 1); |
0af3d00b | 5883 | |
db94535d CM |
5884 | byte_in_group = bytenr - cache->key.objectid; |
5885 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 5886 | |
25179201 | 5887 | spin_lock(&cache->space_info->lock); |
c286ac48 | 5888 | spin_lock(&cache->lock); |
0af3d00b | 5889 | |
73bc1876 | 5890 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
5891 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
5892 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
5893 | ||
9078a3e1 | 5894 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 5895 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 5896 | if (alloc) { |
db94535d | 5897 | old_val += num_bytes; |
11833d66 YZ |
5898 | btrfs_set_block_group_used(&cache->item, old_val); |
5899 | cache->reserved -= num_bytes; | |
11833d66 | 5900 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
5901 | cache->space_info->bytes_used += num_bytes; |
5902 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 5903 | spin_unlock(&cache->lock); |
25179201 | 5904 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 5905 | } else { |
db94535d | 5906 | old_val -= num_bytes; |
ae0ab003 FM |
5907 | btrfs_set_block_group_used(&cache->item, old_val); |
5908 | cache->pinned += num_bytes; | |
5909 | cache->space_info->bytes_pinned += num_bytes; | |
5910 | cache->space_info->bytes_used -= num_bytes; | |
5911 | cache->space_info->disk_used -= num_bytes * factor; | |
5912 | spin_unlock(&cache->lock); | |
5913 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 5914 | |
ae0ab003 FM |
5915 | set_extent_dirty(info->pinned_extents, |
5916 | bytenr, bytenr + num_bytes - 1, | |
5917 | GFP_NOFS | __GFP_NOFAIL); | |
47ab2a6c JB |
5918 | /* |
5919 | * No longer have used bytes in this block group, queue | |
5920 | * it for deletion. | |
5921 | */ | |
5922 | if (old_val == 0) { | |
5923 | spin_lock(&info->unused_bgs_lock); | |
5924 | if (list_empty(&cache->bg_list)) { | |
5925 | btrfs_get_block_group(cache); | |
5926 | list_add_tail(&cache->bg_list, | |
5927 | &info->unused_bgs); | |
5928 | } | |
5929 | spin_unlock(&info->unused_bgs_lock); | |
5930 | } | |
cd1bc465 | 5931 | } |
1bbc621e CM |
5932 | |
5933 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
5934 | if (list_empty(&cache->dirty_list)) { | |
5935 | list_add_tail(&cache->dirty_list, | |
5936 | &trans->transaction->dirty_bgs); | |
5937 | trans->transaction->num_dirty_bgs++; | |
5938 | btrfs_get_block_group(cache); | |
5939 | } | |
5940 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
5941 | ||
fa9c0d79 | 5942 | btrfs_put_block_group(cache); |
db94535d CM |
5943 | total -= num_bytes; |
5944 | bytenr += num_bytes; | |
9078a3e1 CM |
5945 | } |
5946 | return 0; | |
5947 | } | |
6324fbf3 | 5948 | |
a061fc8d CM |
5949 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
5950 | { | |
0f9dd46c | 5951 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 5952 | u64 bytenr; |
0f9dd46c | 5953 | |
a1897fdd LB |
5954 | spin_lock(&root->fs_info->block_group_cache_lock); |
5955 | bytenr = root->fs_info->first_logical_byte; | |
5956 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
5957 | ||
5958 | if (bytenr < (u64)-1) | |
5959 | return bytenr; | |
5960 | ||
0f9dd46c JB |
5961 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
5962 | if (!cache) | |
a061fc8d | 5963 | return 0; |
0f9dd46c | 5964 | |
d2fb3437 | 5965 | bytenr = cache->key.objectid; |
fa9c0d79 | 5966 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
5967 | |
5968 | return bytenr; | |
a061fc8d CM |
5969 | } |
5970 | ||
f0486c68 YZ |
5971 | static int pin_down_extent(struct btrfs_root *root, |
5972 | struct btrfs_block_group_cache *cache, | |
5973 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 5974 | { |
11833d66 YZ |
5975 | spin_lock(&cache->space_info->lock); |
5976 | spin_lock(&cache->lock); | |
5977 | cache->pinned += num_bytes; | |
5978 | cache->space_info->bytes_pinned += num_bytes; | |
5979 | if (reserved) { | |
5980 | cache->reserved -= num_bytes; | |
5981 | cache->space_info->bytes_reserved -= num_bytes; | |
5982 | } | |
5983 | spin_unlock(&cache->lock); | |
5984 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 5985 | |
f0486c68 YZ |
5986 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
5987 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
e2d1f923 | 5988 | if (reserved) |
0be5dc67 | 5989 | trace_btrfs_reserved_extent_free(root, bytenr, num_bytes); |
f0486c68 YZ |
5990 | return 0; |
5991 | } | |
68b38550 | 5992 | |
f0486c68 YZ |
5993 | /* |
5994 | * this function must be called within transaction | |
5995 | */ | |
5996 | int btrfs_pin_extent(struct btrfs_root *root, | |
5997 | u64 bytenr, u64 num_bytes, int reserved) | |
5998 | { | |
5999 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6000 | |
f0486c68 | 6001 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 6002 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
6003 | |
6004 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
6005 | ||
6006 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6007 | return 0; |
6008 | } | |
6009 | ||
f0486c68 | 6010 | /* |
e688b725 CM |
6011 | * this function must be called within transaction |
6012 | */ | |
dcfac415 | 6013 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
e688b725 CM |
6014 | u64 bytenr, u64 num_bytes) |
6015 | { | |
6016 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6017 | int ret; |
e688b725 CM |
6018 | |
6019 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
b50c6e25 JB |
6020 | if (!cache) |
6021 | return -EINVAL; | |
e688b725 CM |
6022 | |
6023 | /* | |
6024 | * pull in the free space cache (if any) so that our pin | |
6025 | * removes the free space from the cache. We have load_only set | |
6026 | * to one because the slow code to read in the free extents does check | |
6027 | * the pinned extents. | |
6028 | */ | |
f6373bf3 | 6029 | cache_block_group(cache, 1); |
e688b725 CM |
6030 | |
6031 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
6032 | ||
6033 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6034 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6035 | btrfs_put_block_group(cache); |
b50c6e25 | 6036 | return ret; |
e688b725 CM |
6037 | } |
6038 | ||
8c2a1a30 JB |
6039 | static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes) |
6040 | { | |
6041 | int ret; | |
6042 | struct btrfs_block_group_cache *block_group; | |
6043 | struct btrfs_caching_control *caching_ctl; | |
6044 | ||
6045 | block_group = btrfs_lookup_block_group(root->fs_info, start); | |
6046 | if (!block_group) | |
6047 | return -EINVAL; | |
6048 | ||
6049 | cache_block_group(block_group, 0); | |
6050 | caching_ctl = get_caching_control(block_group); | |
6051 | ||
6052 | if (!caching_ctl) { | |
6053 | /* Logic error */ | |
6054 | BUG_ON(!block_group_cache_done(block_group)); | |
6055 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6056 | } else { | |
6057 | mutex_lock(&caching_ctl->mutex); | |
6058 | ||
6059 | if (start >= caching_ctl->progress) { | |
6060 | ret = add_excluded_extent(root, start, num_bytes); | |
6061 | } else if (start + num_bytes <= caching_ctl->progress) { | |
6062 | ret = btrfs_remove_free_space(block_group, | |
6063 | start, num_bytes); | |
6064 | } else { | |
6065 | num_bytes = caching_ctl->progress - start; | |
6066 | ret = btrfs_remove_free_space(block_group, | |
6067 | start, num_bytes); | |
6068 | if (ret) | |
6069 | goto out_lock; | |
6070 | ||
6071 | num_bytes = (start + num_bytes) - | |
6072 | caching_ctl->progress; | |
6073 | start = caching_ctl->progress; | |
6074 | ret = add_excluded_extent(root, start, num_bytes); | |
6075 | } | |
6076 | out_lock: | |
6077 | mutex_unlock(&caching_ctl->mutex); | |
6078 | put_caching_control(caching_ctl); | |
6079 | } | |
6080 | btrfs_put_block_group(block_group); | |
6081 | return ret; | |
6082 | } | |
6083 | ||
6084 | int btrfs_exclude_logged_extents(struct btrfs_root *log, | |
6085 | struct extent_buffer *eb) | |
6086 | { | |
6087 | struct btrfs_file_extent_item *item; | |
6088 | struct btrfs_key key; | |
6089 | int found_type; | |
6090 | int i; | |
6091 | ||
6092 | if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) | |
6093 | return 0; | |
6094 | ||
6095 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6096 | btrfs_item_key_to_cpu(eb, &key, i); | |
6097 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6098 | continue; | |
6099 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6100 | found_type = btrfs_file_extent_type(eb, item); | |
6101 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6102 | continue; | |
6103 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6104 | continue; | |
6105 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6106 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
6107 | __exclude_logged_extent(log, key.objectid, key.offset); | |
6108 | } | |
6109 | ||
6110 | return 0; | |
6111 | } | |
6112 | ||
fb25e914 JB |
6113 | /** |
6114 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
6115 | * @cache: The cache we are manipulating | |
6116 | * @num_bytes: The number of bytes in question | |
6117 | * @reserve: One of the reservation enums | |
e570fd27 | 6118 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 JB |
6119 | * |
6120 | * This is called by the allocator when it reserves space, or by somebody who is | |
6121 | * freeing space that was never actually used on disk. For example if you | |
6122 | * reserve some space for a new leaf in transaction A and before transaction A | |
6123 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
6124 | * clear the reservation. | |
6125 | * | |
6126 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
6127 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
6128 | * delalloc bits in the io_tree. We have to do this since we could end up | |
6129 | * allocating less disk space for the amount of data we have reserved in the | |
6130 | * case of compression. | |
6131 | * | |
6132 | * If this is a reservation and the block group has become read only we cannot | |
6133 | * make the reservation and return -EAGAIN, otherwise this function always | |
6134 | * succeeds. | |
f0486c68 | 6135 | */ |
fb25e914 | 6136 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
e570fd27 | 6137 | u64 num_bytes, int reserve, int delalloc) |
11833d66 | 6138 | { |
fb25e914 | 6139 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6140 | int ret = 0; |
79787eaa | 6141 | |
fb25e914 JB |
6142 | spin_lock(&space_info->lock); |
6143 | spin_lock(&cache->lock); | |
6144 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
6145 | if (cache->ro) { |
6146 | ret = -EAGAIN; | |
6147 | } else { | |
fb25e914 JB |
6148 | cache->reserved += num_bytes; |
6149 | space_info->bytes_reserved += num_bytes; | |
6150 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 6151 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
6152 | "space_info", space_info->flags, |
6153 | num_bytes, 0); | |
fb25e914 JB |
6154 | space_info->bytes_may_use -= num_bytes; |
6155 | } | |
e570fd27 MX |
6156 | |
6157 | if (delalloc) | |
6158 | cache->delalloc_bytes += num_bytes; | |
f0486c68 | 6159 | } |
fb25e914 JB |
6160 | } else { |
6161 | if (cache->ro) | |
6162 | space_info->bytes_readonly += num_bytes; | |
6163 | cache->reserved -= num_bytes; | |
6164 | space_info->bytes_reserved -= num_bytes; | |
e570fd27 MX |
6165 | |
6166 | if (delalloc) | |
6167 | cache->delalloc_bytes -= num_bytes; | |
324ae4df | 6168 | } |
fb25e914 JB |
6169 | spin_unlock(&cache->lock); |
6170 | spin_unlock(&space_info->lock); | |
f0486c68 | 6171 | return ret; |
324ae4df | 6172 | } |
9078a3e1 | 6173 | |
143bede5 | 6174 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 6175 | struct btrfs_root *root) |
e8569813 | 6176 | { |
e8569813 | 6177 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
6178 | struct btrfs_caching_control *next; |
6179 | struct btrfs_caching_control *caching_ctl; | |
6180 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6181 | |
9e351cc8 | 6182 | down_write(&fs_info->commit_root_sem); |
25179201 | 6183 | |
11833d66 YZ |
6184 | list_for_each_entry_safe(caching_ctl, next, |
6185 | &fs_info->caching_block_groups, list) { | |
6186 | cache = caching_ctl->block_group; | |
6187 | if (block_group_cache_done(cache)) { | |
6188 | cache->last_byte_to_unpin = (u64)-1; | |
6189 | list_del_init(&caching_ctl->list); | |
6190 | put_caching_control(caching_ctl); | |
e8569813 | 6191 | } else { |
11833d66 | 6192 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6193 | } |
e8569813 | 6194 | } |
11833d66 YZ |
6195 | |
6196 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6197 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6198 | else | |
6199 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6200 | ||
9e351cc8 | 6201 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6202 | |
6203 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6204 | } |
6205 | ||
c759c4e1 JB |
6206 | /* |
6207 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6208 | * what it should be based on the mount options. | |
6209 | */ | |
6210 | static struct btrfs_free_cluster * | |
6211 | fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info, | |
6212 | u64 *empty_cluster) | |
6213 | { | |
6214 | struct btrfs_free_cluster *ret = NULL; | |
6215 | bool ssd = btrfs_test_opt(root, SSD); | |
6216 | ||
6217 | *empty_cluster = 0; | |
6218 | if (btrfs_mixed_space_info(space_info)) | |
6219 | return ret; | |
6220 | ||
6221 | if (ssd) | |
6222 | *empty_cluster = 2 * 1024 * 1024; | |
6223 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
6224 | ret = &root->fs_info->meta_alloc_cluster; | |
6225 | if (!ssd) | |
6226 | *empty_cluster = 64 * 1024; | |
6227 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) { | |
6228 | ret = &root->fs_info->data_alloc_cluster; | |
6229 | } | |
6230 | ||
6231 | return ret; | |
6232 | } | |
6233 | ||
678886bd FM |
6234 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end, |
6235 | const bool return_free_space) | |
ccd467d6 | 6236 | { |
11833d66 YZ |
6237 | struct btrfs_fs_info *fs_info = root->fs_info; |
6238 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
6239 | struct btrfs_space_info *space_info; |
6240 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6241 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6242 | u64 len; |
c759c4e1 JB |
6243 | u64 total_unpinned = 0; |
6244 | u64 empty_cluster = 0; | |
7b398f8e | 6245 | bool readonly; |
ccd467d6 | 6246 | |
11833d66 | 6247 | while (start <= end) { |
7b398f8e | 6248 | readonly = false; |
11833d66 YZ |
6249 | if (!cache || |
6250 | start >= cache->key.objectid + cache->key.offset) { | |
6251 | if (cache) | |
6252 | btrfs_put_block_group(cache); | |
c759c4e1 | 6253 | total_unpinned = 0; |
11833d66 | 6254 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6255 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 JB |
6256 | |
6257 | cluster = fetch_cluster_info(root, | |
6258 | cache->space_info, | |
6259 | &empty_cluster); | |
6260 | empty_cluster <<= 1; | |
11833d66 YZ |
6261 | } |
6262 | ||
6263 | len = cache->key.objectid + cache->key.offset - start; | |
6264 | len = min(len, end + 1 - start); | |
6265 | ||
6266 | if (start < cache->last_byte_to_unpin) { | |
6267 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6268 | if (return_free_space) |
6269 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6270 | } |
6271 | ||
f0486c68 | 6272 | start += len; |
c759c4e1 | 6273 | total_unpinned += len; |
7b398f8e | 6274 | space_info = cache->space_info; |
f0486c68 | 6275 | |
c759c4e1 JB |
6276 | /* |
6277 | * If this space cluster has been marked as fragmented and we've | |
6278 | * unpinned enough in this block group to potentially allow a | |
6279 | * cluster to be created inside of it go ahead and clear the | |
6280 | * fragmented check. | |
6281 | */ | |
6282 | if (cluster && cluster->fragmented && | |
6283 | total_unpinned > empty_cluster) { | |
6284 | spin_lock(&cluster->lock); | |
6285 | cluster->fragmented = 0; | |
6286 | spin_unlock(&cluster->lock); | |
6287 | } | |
6288 | ||
7b398f8e | 6289 | spin_lock(&space_info->lock); |
11833d66 YZ |
6290 | spin_lock(&cache->lock); |
6291 | cache->pinned -= len; | |
7b398f8e | 6292 | space_info->bytes_pinned -= len; |
4f4db217 | 6293 | space_info->max_extent_size = 0; |
d288db5d | 6294 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6295 | if (cache->ro) { |
6296 | space_info->bytes_readonly += len; | |
6297 | readonly = true; | |
6298 | } | |
11833d66 | 6299 | spin_unlock(&cache->lock); |
7b398f8e JB |
6300 | if (!readonly && global_rsv->space_info == space_info) { |
6301 | spin_lock(&global_rsv->lock); | |
6302 | if (!global_rsv->full) { | |
6303 | len = min(len, global_rsv->size - | |
6304 | global_rsv->reserved); | |
6305 | global_rsv->reserved += len; | |
6306 | space_info->bytes_may_use += len; | |
6307 | if (global_rsv->reserved >= global_rsv->size) | |
6308 | global_rsv->full = 1; | |
6309 | } | |
6310 | spin_unlock(&global_rsv->lock); | |
6311 | } | |
6312 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6313 | } |
11833d66 YZ |
6314 | |
6315 | if (cache) | |
6316 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6317 | return 0; |
6318 | } | |
6319 | ||
6320 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 6321 | struct btrfs_root *root) |
a28ec197 | 6322 | { |
11833d66 | 6323 | struct btrfs_fs_info *fs_info = root->fs_info; |
e33e17ee JM |
6324 | struct btrfs_block_group_cache *block_group, *tmp; |
6325 | struct list_head *deleted_bgs; | |
11833d66 | 6326 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6327 | u64 start; |
6328 | u64 end; | |
a28ec197 | 6329 | int ret; |
a28ec197 | 6330 | |
11833d66 YZ |
6331 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6332 | unpin = &fs_info->freed_extents[1]; | |
6333 | else | |
6334 | unpin = &fs_info->freed_extents[0]; | |
6335 | ||
e33e17ee | 6336 | while (!trans->aborted) { |
d4b450cd | 6337 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6338 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6339 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6340 | if (ret) { |
6341 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6342 | break; |
d4b450cd | 6343 | } |
1f3c79a2 | 6344 | |
5378e607 LD |
6345 | if (btrfs_test_opt(root, DISCARD)) |
6346 | ret = btrfs_discard_extent(root, start, | |
6347 | end + 1 - start, NULL); | |
1f3c79a2 | 6348 | |
1a5bc167 | 6349 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
678886bd | 6350 | unpin_extent_range(root, start, end, true); |
d4b450cd | 6351 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6352 | cond_resched(); |
a28ec197 | 6353 | } |
817d52f8 | 6354 | |
e33e17ee JM |
6355 | /* |
6356 | * Transaction is finished. We don't need the lock anymore. We | |
6357 | * do need to clean up the block groups in case of a transaction | |
6358 | * abort. | |
6359 | */ | |
6360 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6361 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6362 | u64 trimmed = 0; | |
6363 | ||
6364 | ret = -EROFS; | |
6365 | if (!trans->aborted) | |
6366 | ret = btrfs_discard_extent(root, | |
6367 | block_group->key.objectid, | |
6368 | block_group->key.offset, | |
6369 | &trimmed); | |
6370 | ||
6371 | list_del_init(&block_group->bg_list); | |
6372 | btrfs_put_block_group_trimming(block_group); | |
6373 | btrfs_put_block_group(block_group); | |
6374 | ||
6375 | if (ret) { | |
6376 | const char *errstr = btrfs_decode_error(ret); | |
6377 | btrfs_warn(fs_info, | |
6378 | "Discard failed while removing blockgroup: errno=%d %s\n", | |
6379 | ret, errstr); | |
6380 | } | |
6381 | } | |
6382 | ||
e20d96d6 CM |
6383 | return 0; |
6384 | } | |
6385 | ||
b150a4f1 JB |
6386 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, |
6387 | u64 owner, u64 root_objectid) | |
6388 | { | |
6389 | struct btrfs_space_info *space_info; | |
6390 | u64 flags; | |
6391 | ||
6392 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
6393 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
6394 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
6395 | else | |
6396 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
6397 | } else { | |
6398 | flags = BTRFS_BLOCK_GROUP_DATA; | |
6399 | } | |
6400 | ||
6401 | space_info = __find_space_info(fs_info, flags); | |
6402 | BUG_ON(!space_info); /* Logic bug */ | |
6403 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); | |
6404 | } | |
6405 | ||
6406 | ||
5d4f98a2 YZ |
6407 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
6408 | struct btrfs_root *root, | |
c682f9b3 | 6409 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6410 | u64 root_objectid, u64 owner_objectid, |
6411 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6412 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6413 | { |
e2fa7227 | 6414 | struct btrfs_key key; |
5d4f98a2 | 6415 | struct btrfs_path *path; |
1261ec42 CM |
6416 | struct btrfs_fs_info *info = root->fs_info; |
6417 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 6418 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6419 | struct btrfs_extent_item *ei; |
6420 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6421 | int ret; |
5d4f98a2 | 6422 | int is_data; |
952fccac CM |
6423 | int extent_slot = 0; |
6424 | int found_extent = 0; | |
6425 | int num_to_del = 1; | |
5d4f98a2 YZ |
6426 | u32 item_size; |
6427 | u64 refs; | |
c682f9b3 QW |
6428 | u64 bytenr = node->bytenr; |
6429 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6430 | int last_ref = 0; |
3173a18f JB |
6431 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
6432 | SKINNY_METADATA); | |
037e6390 | 6433 | |
5caf2a00 | 6434 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6435 | if (!path) |
6436 | return -ENOMEM; | |
5f26f772 | 6437 | |
3c12ac72 | 6438 | path->reada = 1; |
b9473439 | 6439 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6440 | |
6441 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6442 | BUG_ON(!is_data && refs_to_drop != 1); | |
6443 | ||
3173a18f JB |
6444 | if (is_data) |
6445 | skinny_metadata = 0; | |
6446 | ||
5d4f98a2 YZ |
6447 | ret = lookup_extent_backref(trans, extent_root, path, &iref, |
6448 | bytenr, num_bytes, parent, | |
6449 | root_objectid, owner_objectid, | |
6450 | owner_offset); | |
7bb86316 | 6451 | if (ret == 0) { |
952fccac | 6452 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6453 | while (extent_slot >= 0) { |
6454 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6455 | extent_slot); |
5d4f98a2 | 6456 | if (key.objectid != bytenr) |
952fccac | 6457 | break; |
5d4f98a2 YZ |
6458 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6459 | key.offset == num_bytes) { | |
952fccac CM |
6460 | found_extent = 1; |
6461 | break; | |
6462 | } | |
3173a18f JB |
6463 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6464 | key.offset == owner_objectid) { | |
6465 | found_extent = 1; | |
6466 | break; | |
6467 | } | |
952fccac CM |
6468 | if (path->slots[0] - extent_slot > 5) |
6469 | break; | |
5d4f98a2 | 6470 | extent_slot--; |
952fccac | 6471 | } |
5d4f98a2 YZ |
6472 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6473 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6474 | if (found_extent && item_size < sizeof(*ei)) | |
6475 | found_extent = 0; | |
6476 | #endif | |
31840ae1 | 6477 | if (!found_extent) { |
5d4f98a2 | 6478 | BUG_ON(iref); |
56bec294 | 6479 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 | 6480 | NULL, refs_to_drop, |
fcebe456 | 6481 | is_data, &last_ref); |
005d6427 DS |
6482 | if (ret) { |
6483 | btrfs_abort_transaction(trans, extent_root, ret); | |
6484 | goto out; | |
6485 | } | |
b3b4aa74 | 6486 | btrfs_release_path(path); |
b9473439 | 6487 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6488 | |
6489 | key.objectid = bytenr; | |
6490 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6491 | key.offset = num_bytes; | |
6492 | ||
3173a18f JB |
6493 | if (!is_data && skinny_metadata) { |
6494 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6495 | key.offset = owner_objectid; | |
6496 | } | |
6497 | ||
31840ae1 ZY |
6498 | ret = btrfs_search_slot(trans, extent_root, |
6499 | &key, path, -1, 1); | |
3173a18f JB |
6500 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6501 | /* | |
6502 | * Couldn't find our skinny metadata item, | |
6503 | * see if we have ye olde extent item. | |
6504 | */ | |
6505 | path->slots[0]--; | |
6506 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6507 | path->slots[0]); | |
6508 | if (key.objectid == bytenr && | |
6509 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6510 | key.offset == num_bytes) | |
6511 | ret = 0; | |
6512 | } | |
6513 | ||
6514 | if (ret > 0 && skinny_metadata) { | |
6515 | skinny_metadata = false; | |
9ce49a0b | 6516 | key.objectid = bytenr; |
3173a18f JB |
6517 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6518 | key.offset = num_bytes; | |
6519 | btrfs_release_path(path); | |
6520 | ret = btrfs_search_slot(trans, extent_root, | |
6521 | &key, path, -1, 1); | |
6522 | } | |
6523 | ||
f3465ca4 | 6524 | if (ret) { |
c2cf52eb | 6525 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 6526 | ret, bytenr); |
b783e62d JB |
6527 | if (ret > 0) |
6528 | btrfs_print_leaf(extent_root, | |
6529 | path->nodes[0]); | |
f3465ca4 | 6530 | } |
005d6427 DS |
6531 | if (ret < 0) { |
6532 | btrfs_abort_transaction(trans, extent_root, ret); | |
6533 | goto out; | |
6534 | } | |
31840ae1 ZY |
6535 | extent_slot = path->slots[0]; |
6536 | } | |
fae7f21c | 6537 | } else if (WARN_ON(ret == -ENOENT)) { |
7bb86316 | 6538 | btrfs_print_leaf(extent_root, path->nodes[0]); |
c2cf52eb SK |
6539 | btrfs_err(info, |
6540 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6541 | bytenr, parent, root_objectid, owner_objectid, |
6542 | owner_offset); | |
c4a050bb JB |
6543 | btrfs_abort_transaction(trans, extent_root, ret); |
6544 | goto out; | |
79787eaa | 6545 | } else { |
005d6427 DS |
6546 | btrfs_abort_transaction(trans, extent_root, ret); |
6547 | goto out; | |
7bb86316 | 6548 | } |
5f39d397 CM |
6549 | |
6550 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6551 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6552 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6553 | if (item_size < sizeof(*ei)) { | |
6554 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
6555 | ret = convert_extent_item_v0(trans, extent_root, path, | |
6556 | owner_objectid, 0); | |
005d6427 DS |
6557 | if (ret < 0) { |
6558 | btrfs_abort_transaction(trans, extent_root, ret); | |
6559 | goto out; | |
6560 | } | |
5d4f98a2 | 6561 | |
b3b4aa74 | 6562 | btrfs_release_path(path); |
5d4f98a2 YZ |
6563 | path->leave_spinning = 1; |
6564 | ||
6565 | key.objectid = bytenr; | |
6566 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6567 | key.offset = num_bytes; | |
6568 | ||
6569 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
6570 | -1, 1); | |
6571 | if (ret) { | |
c2cf52eb | 6572 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 6573 | ret, bytenr); |
5d4f98a2 YZ |
6574 | btrfs_print_leaf(extent_root, path->nodes[0]); |
6575 | } | |
005d6427 DS |
6576 | if (ret < 0) { |
6577 | btrfs_abort_transaction(trans, extent_root, ret); | |
6578 | goto out; | |
6579 | } | |
6580 | ||
5d4f98a2 YZ |
6581 | extent_slot = path->slots[0]; |
6582 | leaf = path->nodes[0]; | |
6583 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
6584 | } | |
6585 | #endif | |
6586 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 6587 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 6588 | struct btrfs_extent_item); |
3173a18f JB |
6589 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
6590 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
6591 | struct btrfs_tree_block_info *bi; |
6592 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
6593 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
6594 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
6595 | } | |
56bec294 | 6596 | |
5d4f98a2 | 6597 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 JB |
6598 | if (refs < refs_to_drop) { |
6599 | btrfs_err(info, "trying to drop %d refs but we only have %Lu " | |
351fd353 | 6600 | "for bytenr %Lu", refs_to_drop, refs, bytenr); |
32b02538 JB |
6601 | ret = -EINVAL; |
6602 | btrfs_abort_transaction(trans, extent_root, ret); | |
6603 | goto out; | |
6604 | } | |
56bec294 | 6605 | refs -= refs_to_drop; |
5f39d397 | 6606 | |
5d4f98a2 YZ |
6607 | if (refs > 0) { |
6608 | if (extent_op) | |
6609 | __run_delayed_extent_op(extent_op, leaf, ei); | |
6610 | /* | |
6611 | * In the case of inline back ref, reference count will | |
6612 | * be updated by remove_extent_backref | |
952fccac | 6613 | */ |
5d4f98a2 YZ |
6614 | if (iref) { |
6615 | BUG_ON(!found_extent); | |
6616 | } else { | |
6617 | btrfs_set_extent_refs(leaf, ei, refs); | |
6618 | btrfs_mark_buffer_dirty(leaf); | |
6619 | } | |
6620 | if (found_extent) { | |
6621 | ret = remove_extent_backref(trans, extent_root, path, | |
6622 | iref, refs_to_drop, | |
fcebe456 | 6623 | is_data, &last_ref); |
005d6427 DS |
6624 | if (ret) { |
6625 | btrfs_abort_transaction(trans, extent_root, ret); | |
6626 | goto out; | |
6627 | } | |
952fccac | 6628 | } |
b150a4f1 JB |
6629 | add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid, |
6630 | root_objectid); | |
5d4f98a2 | 6631 | } else { |
5d4f98a2 YZ |
6632 | if (found_extent) { |
6633 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 6634 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
6635 | if (iref) { |
6636 | BUG_ON(path->slots[0] != extent_slot); | |
6637 | } else { | |
6638 | BUG_ON(path->slots[0] != extent_slot + 1); | |
6639 | path->slots[0] = extent_slot; | |
6640 | num_to_del = 2; | |
6641 | } | |
78fae27e | 6642 | } |
b9473439 | 6643 | |
fcebe456 | 6644 | last_ref = 1; |
952fccac CM |
6645 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
6646 | num_to_del); | |
005d6427 DS |
6647 | if (ret) { |
6648 | btrfs_abort_transaction(trans, extent_root, ret); | |
6649 | goto out; | |
6650 | } | |
b3b4aa74 | 6651 | btrfs_release_path(path); |
21af804c | 6652 | |
5d4f98a2 | 6653 | if (is_data) { |
459931ec | 6654 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 DS |
6655 | if (ret) { |
6656 | btrfs_abort_transaction(trans, extent_root, ret); | |
6657 | goto out; | |
6658 | } | |
459931ec CM |
6659 | } |
6660 | ||
ce93ec54 | 6661 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
005d6427 DS |
6662 | if (ret) { |
6663 | btrfs_abort_transaction(trans, extent_root, ret); | |
6664 | goto out; | |
6665 | } | |
a28ec197 | 6666 | } |
fcebe456 JB |
6667 | btrfs_release_path(path); |
6668 | ||
79787eaa | 6669 | out: |
5caf2a00 | 6670 | btrfs_free_path(path); |
a28ec197 CM |
6671 | return ret; |
6672 | } | |
6673 | ||
1887be66 | 6674 | /* |
f0486c68 | 6675 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
6676 | * delayed ref for that extent as well. This searches the delayed ref tree for |
6677 | * a given extent, and if there are no other delayed refs to be processed, it | |
6678 | * removes it from the tree. | |
6679 | */ | |
6680 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
6681 | struct btrfs_root *root, u64 bytenr) | |
6682 | { | |
6683 | struct btrfs_delayed_ref_head *head; | |
6684 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 6685 | int ret = 0; |
1887be66 CM |
6686 | |
6687 | delayed_refs = &trans->transaction->delayed_refs; | |
6688 | spin_lock(&delayed_refs->lock); | |
6689 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
6690 | if (!head) | |
cf93da7b | 6691 | goto out_delayed_unlock; |
1887be66 | 6692 | |
d7df2c79 | 6693 | spin_lock(&head->lock); |
c6fc2454 | 6694 | if (!list_empty(&head->ref_list)) |
1887be66 CM |
6695 | goto out; |
6696 | ||
5d4f98a2 YZ |
6697 | if (head->extent_op) { |
6698 | if (!head->must_insert_reserved) | |
6699 | goto out; | |
78a6184a | 6700 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
6701 | head->extent_op = NULL; |
6702 | } | |
6703 | ||
1887be66 CM |
6704 | /* |
6705 | * waiting for the lock here would deadlock. If someone else has it | |
6706 | * locked they are already in the process of dropping it anyway | |
6707 | */ | |
6708 | if (!mutex_trylock(&head->mutex)) | |
6709 | goto out; | |
6710 | ||
6711 | /* | |
6712 | * at this point we have a head with no other entries. Go | |
6713 | * ahead and process it. | |
6714 | */ | |
6715 | head->node.in_tree = 0; | |
c46effa6 | 6716 | rb_erase(&head->href_node, &delayed_refs->href_root); |
c3e69d58 | 6717 | |
d7df2c79 | 6718 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
6719 | |
6720 | /* | |
6721 | * we don't take a ref on the node because we're removing it from the | |
6722 | * tree, so we just steal the ref the tree was holding. | |
6723 | */ | |
c3e69d58 | 6724 | delayed_refs->num_heads--; |
d7df2c79 | 6725 | if (head->processing == 0) |
c3e69d58 | 6726 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
6727 | head->processing = 0; |
6728 | spin_unlock(&head->lock); | |
1887be66 CM |
6729 | spin_unlock(&delayed_refs->lock); |
6730 | ||
f0486c68 YZ |
6731 | BUG_ON(head->extent_op); |
6732 | if (head->must_insert_reserved) | |
6733 | ret = 1; | |
6734 | ||
6735 | mutex_unlock(&head->mutex); | |
1887be66 | 6736 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 6737 | return ret; |
1887be66 | 6738 | out: |
d7df2c79 | 6739 | spin_unlock(&head->lock); |
cf93da7b CM |
6740 | |
6741 | out_delayed_unlock: | |
1887be66 CM |
6742 | spin_unlock(&delayed_refs->lock); |
6743 | return 0; | |
6744 | } | |
6745 | ||
f0486c68 YZ |
6746 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
6747 | struct btrfs_root *root, | |
6748 | struct extent_buffer *buf, | |
5581a51a | 6749 | u64 parent, int last_ref) |
f0486c68 | 6750 | { |
b150a4f1 | 6751 | int pin = 1; |
f0486c68 YZ |
6752 | int ret; |
6753 | ||
6754 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
6755 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6756 | buf->start, buf->len, | |
6757 | parent, root->root_key.objectid, | |
6758 | btrfs_header_level(buf), | |
b06c4bf5 | 6759 | BTRFS_DROP_DELAYED_REF, NULL); |
79787eaa | 6760 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
6761 | } |
6762 | ||
6763 | if (!last_ref) | |
6764 | return; | |
6765 | ||
f0486c68 | 6766 | if (btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
6767 | struct btrfs_block_group_cache *cache; |
6768 | ||
f0486c68 YZ |
6769 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
6770 | ret = check_ref_cleanup(trans, root, buf->start); | |
6771 | if (!ret) | |
37be25bc | 6772 | goto out; |
f0486c68 YZ |
6773 | } |
6774 | ||
6219872d FM |
6775 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
6776 | ||
f0486c68 YZ |
6777 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6778 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
6219872d | 6779 | btrfs_put_block_group(cache); |
37be25bc | 6780 | goto out; |
f0486c68 YZ |
6781 | } |
6782 | ||
6783 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
6784 | ||
6785 | btrfs_add_free_space(cache, buf->start, buf->len); | |
e570fd27 | 6786 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0); |
6219872d | 6787 | btrfs_put_block_group(cache); |
0be5dc67 | 6788 | trace_btrfs_reserved_extent_free(root, buf->start, buf->len); |
b150a4f1 | 6789 | pin = 0; |
f0486c68 YZ |
6790 | } |
6791 | out: | |
b150a4f1 JB |
6792 | if (pin) |
6793 | add_pinned_bytes(root->fs_info, buf->len, | |
6794 | btrfs_header_level(buf), | |
6795 | root->root_key.objectid); | |
6796 | ||
a826d6dc JB |
6797 | /* |
6798 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
6799 | * anymore. | |
6800 | */ | |
6801 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
6802 | } |
6803 | ||
79787eaa | 6804 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
6805 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
6806 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
b06c4bf5 | 6807 | u64 owner, u64 offset) |
925baedd CM |
6808 | { |
6809 | int ret; | |
66d7e7f0 | 6810 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 6811 | |
fccb84c9 | 6812 | if (btrfs_test_is_dummy_root(root)) |
faa2dbf0 | 6813 | return 0; |
fccb84c9 | 6814 | |
b150a4f1 JB |
6815 | add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid); |
6816 | ||
56bec294 CM |
6817 | /* |
6818 | * tree log blocks never actually go into the extent allocation | |
6819 | * tree, just update pinning info and exit early. | |
56bec294 | 6820 | */ |
5d4f98a2 YZ |
6821 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
6822 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 6823 | /* unlocks the pinned mutex */ |
11833d66 | 6824 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 6825 | ret = 0; |
5d4f98a2 | 6826 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
6827 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
6828 | num_bytes, | |
5d4f98a2 | 6829 | parent, root_objectid, (int)owner, |
b06c4bf5 | 6830 | BTRFS_DROP_DELAYED_REF, NULL); |
5d4f98a2 | 6831 | } else { |
66d7e7f0 AJ |
6832 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
6833 | num_bytes, | |
6834 | parent, root_objectid, owner, | |
5846a3c2 QW |
6835 | offset, 0, |
6836 | BTRFS_DROP_DELAYED_REF, NULL); | |
56bec294 | 6837 | } |
925baedd CM |
6838 | return ret; |
6839 | } | |
6840 | ||
817d52f8 JB |
6841 | /* |
6842 | * when we wait for progress in the block group caching, its because | |
6843 | * our allocation attempt failed at least once. So, we must sleep | |
6844 | * and let some progress happen before we try again. | |
6845 | * | |
6846 | * This function will sleep at least once waiting for new free space to | |
6847 | * show up, and then it will check the block group free space numbers | |
6848 | * for our min num_bytes. Another option is to have it go ahead | |
6849 | * and look in the rbtree for a free extent of a given size, but this | |
6850 | * is a good start. | |
36cce922 JB |
6851 | * |
6852 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
6853 | * any of the information in this block group. | |
817d52f8 | 6854 | */ |
36cce922 | 6855 | static noinline void |
817d52f8 JB |
6856 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
6857 | u64 num_bytes) | |
6858 | { | |
11833d66 | 6859 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 6860 | |
11833d66 YZ |
6861 | caching_ctl = get_caching_control(cache); |
6862 | if (!caching_ctl) | |
36cce922 | 6863 | return; |
817d52f8 | 6864 | |
11833d66 | 6865 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 6866 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
6867 | |
6868 | put_caching_control(caching_ctl); | |
11833d66 YZ |
6869 | } |
6870 | ||
6871 | static noinline int | |
6872 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
6873 | { | |
6874 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 6875 | int ret = 0; |
11833d66 YZ |
6876 | |
6877 | caching_ctl = get_caching_control(cache); | |
6878 | if (!caching_ctl) | |
36cce922 | 6879 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
6880 | |
6881 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
6882 | if (cache->cached == BTRFS_CACHE_ERROR) |
6883 | ret = -EIO; | |
11833d66 | 6884 | put_caching_control(caching_ctl); |
36cce922 | 6885 | return ret; |
817d52f8 JB |
6886 | } |
6887 | ||
31e50229 | 6888 | int __get_raid_index(u64 flags) |
b742bb82 | 6889 | { |
7738a53a | 6890 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 6891 | return BTRFS_RAID_RAID10; |
7738a53a | 6892 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 6893 | return BTRFS_RAID_RAID1; |
7738a53a | 6894 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 6895 | return BTRFS_RAID_DUP; |
7738a53a | 6896 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 6897 | return BTRFS_RAID_RAID0; |
53b381b3 | 6898 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 6899 | return BTRFS_RAID_RAID5; |
53b381b3 | 6900 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 6901 | return BTRFS_RAID_RAID6; |
7738a53a | 6902 | |
e942f883 | 6903 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
6904 | } |
6905 | ||
6ab0a202 | 6906 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
7738a53a | 6907 | { |
31e50229 | 6908 | return __get_raid_index(cache->flags); |
7738a53a ID |
6909 | } |
6910 | ||
6ab0a202 JM |
6911 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
6912 | [BTRFS_RAID_RAID10] = "raid10", | |
6913 | [BTRFS_RAID_RAID1] = "raid1", | |
6914 | [BTRFS_RAID_DUP] = "dup", | |
6915 | [BTRFS_RAID_RAID0] = "raid0", | |
6916 | [BTRFS_RAID_SINGLE] = "single", | |
6917 | [BTRFS_RAID_RAID5] = "raid5", | |
6918 | [BTRFS_RAID_RAID6] = "raid6", | |
6919 | }; | |
6920 | ||
1b8e5df6 | 6921 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
6922 | { |
6923 | if (type >= BTRFS_NR_RAID_TYPES) | |
6924 | return NULL; | |
6925 | ||
6926 | return btrfs_raid_type_names[type]; | |
6927 | } | |
6928 | ||
817d52f8 | 6929 | enum btrfs_loop_type { |
285ff5af JB |
6930 | LOOP_CACHING_NOWAIT = 0, |
6931 | LOOP_CACHING_WAIT = 1, | |
6932 | LOOP_ALLOC_CHUNK = 2, | |
6933 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
6934 | }; |
6935 | ||
e570fd27 MX |
6936 | static inline void |
6937 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
6938 | int delalloc) | |
6939 | { | |
6940 | if (delalloc) | |
6941 | down_read(&cache->data_rwsem); | |
6942 | } | |
6943 | ||
6944 | static inline void | |
6945 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
6946 | int delalloc) | |
6947 | { | |
6948 | btrfs_get_block_group(cache); | |
6949 | if (delalloc) | |
6950 | down_read(&cache->data_rwsem); | |
6951 | } | |
6952 | ||
6953 | static struct btrfs_block_group_cache * | |
6954 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
6955 | struct btrfs_free_cluster *cluster, | |
6956 | int delalloc) | |
6957 | { | |
6958 | struct btrfs_block_group_cache *used_bg; | |
6959 | bool locked = false; | |
6960 | again: | |
6961 | spin_lock(&cluster->refill_lock); | |
6962 | if (locked) { | |
6963 | if (used_bg == cluster->block_group) | |
6964 | return used_bg; | |
6965 | ||
6966 | up_read(&used_bg->data_rwsem); | |
6967 | btrfs_put_block_group(used_bg); | |
6968 | } | |
6969 | ||
6970 | used_bg = cluster->block_group; | |
6971 | if (!used_bg) | |
6972 | return NULL; | |
6973 | ||
6974 | if (used_bg == block_group) | |
6975 | return used_bg; | |
6976 | ||
6977 | btrfs_get_block_group(used_bg); | |
6978 | ||
6979 | if (!delalloc) | |
6980 | return used_bg; | |
6981 | ||
6982 | if (down_read_trylock(&used_bg->data_rwsem)) | |
6983 | return used_bg; | |
6984 | ||
6985 | spin_unlock(&cluster->refill_lock); | |
6986 | down_read(&used_bg->data_rwsem); | |
6987 | locked = true; | |
6988 | goto again; | |
6989 | } | |
6990 | ||
6991 | static inline void | |
6992 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
6993 | int delalloc) | |
6994 | { | |
6995 | if (delalloc) | |
6996 | up_read(&cache->data_rwsem); | |
6997 | btrfs_put_block_group(cache); | |
6998 | } | |
6999 | ||
fec577fb CM |
7000 | /* |
7001 | * walks the btree of allocated extents and find a hole of a given size. | |
7002 | * The key ins is changed to record the hole: | |
a4820398 | 7003 | * ins->objectid == start position |
62e2749e | 7004 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7005 | * ins->offset == the size of the hole. |
fec577fb | 7006 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7007 | * |
7008 | * If there is no suitable free space, we will record the max size of | |
7009 | * the free space extent currently. | |
fec577fb | 7010 | */ |
00361589 | 7011 | static noinline int find_free_extent(struct btrfs_root *orig_root, |
98ed5174 | 7012 | u64 num_bytes, u64 empty_size, |
98ed5174 | 7013 | u64 hint_byte, struct btrfs_key *ins, |
e570fd27 | 7014 | u64 flags, int delalloc) |
fec577fb | 7015 | { |
80eb234a | 7016 | int ret = 0; |
d397712b | 7017 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 7018 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7019 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7020 | u64 search_start = 0; |
a4820398 | 7021 | u64 max_extent_size = 0; |
c759c4e1 | 7022 | u64 empty_cluster = 0; |
80eb234a | 7023 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7024 | int loop = 0; |
b6919a58 DS |
7025 | int index = __get_raid_index(flags); |
7026 | int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ? | |
fb25e914 | 7027 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; |
0a24325e | 7028 | bool failed_cluster_refill = false; |
1cdda9b8 | 7029 | bool failed_alloc = false; |
67377734 | 7030 | bool use_cluster = true; |
60d2adbb | 7031 | bool have_caching_bg = false; |
13a0db5a | 7032 | bool orig_have_caching_bg = false; |
a5e681d9 | 7033 | bool full_search = false; |
fec577fb | 7034 | |
db94535d | 7035 | WARN_ON(num_bytes < root->sectorsize); |
962a298f | 7036 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7037 | ins->objectid = 0; |
7038 | ins->offset = 0; | |
b1a4d965 | 7039 | |
b6919a58 | 7040 | trace_find_free_extent(orig_root, num_bytes, empty_size, flags); |
3f7de037 | 7041 | |
b6919a58 | 7042 | space_info = __find_space_info(root->fs_info, flags); |
1b1d1f66 | 7043 | if (!space_info) { |
b6919a58 | 7044 | btrfs_err(root->fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7045 | return -ENOSPC; |
7046 | } | |
2552d17e | 7047 | |
67377734 | 7048 | /* |
4f4db217 JB |
7049 | * If our free space is heavily fragmented we may not be able to make |
7050 | * big contiguous allocations, so instead of doing the expensive search | |
7051 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7052 | * can go ahead and search for a more manageable chunk. | |
7053 | * | |
7054 | * If our max_extent_size is large enough for our allocation simply | |
7055 | * disable clustering since we will likely not be able to find enough | |
7056 | * space to create a cluster and induce latency trying. | |
67377734 | 7057 | */ |
4f4db217 JB |
7058 | if (unlikely(space_info->max_extent_size)) { |
7059 | spin_lock(&space_info->lock); | |
7060 | if (space_info->max_extent_size && | |
7061 | num_bytes > space_info->max_extent_size) { | |
7062 | ins->offset = space_info->max_extent_size; | |
7063 | spin_unlock(&space_info->lock); | |
7064 | return -ENOSPC; | |
7065 | } else if (space_info->max_extent_size) { | |
7066 | use_cluster = false; | |
7067 | } | |
7068 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7069 | } |
0f9dd46c | 7070 | |
c759c4e1 | 7071 | last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster); |
239b14b3 | 7072 | if (last_ptr) { |
fa9c0d79 CM |
7073 | spin_lock(&last_ptr->lock); |
7074 | if (last_ptr->block_group) | |
7075 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7076 | if (last_ptr->fragmented) { |
7077 | /* | |
7078 | * We still set window_start so we can keep track of the | |
7079 | * last place we found an allocation to try and save | |
7080 | * some time. | |
7081 | */ | |
7082 | hint_byte = last_ptr->window_start; | |
7083 | use_cluster = false; | |
7084 | } | |
fa9c0d79 | 7085 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7086 | } |
fa9c0d79 | 7087 | |
a061fc8d | 7088 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 7089 | search_start = max(search_start, hint_byte); |
2552d17e | 7090 | if (search_start == hint_byte) { |
2552d17e JB |
7091 | block_group = btrfs_lookup_block_group(root->fs_info, |
7092 | search_start); | |
817d52f8 JB |
7093 | /* |
7094 | * we don't want to use the block group if it doesn't match our | |
7095 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7096 | * |
7097 | * However if we are re-searching with an ideal block group | |
7098 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7099 | */ |
b6919a58 | 7100 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7101 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7102 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7103 | if (list_empty(&block_group->list) || |
7104 | block_group->ro) { | |
7105 | /* | |
7106 | * someone is removing this block group, | |
7107 | * we can't jump into the have_block_group | |
7108 | * target because our list pointers are not | |
7109 | * valid | |
7110 | */ | |
7111 | btrfs_put_block_group(block_group); | |
7112 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7113 | } else { |
b742bb82 | 7114 | index = get_block_group_index(block_group); |
e570fd27 | 7115 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7116 | goto have_block_group; |
ccf0e725 | 7117 | } |
2552d17e | 7118 | } else if (block_group) { |
fa9c0d79 | 7119 | btrfs_put_block_group(block_group); |
2552d17e | 7120 | } |
42e70e7a | 7121 | } |
2552d17e | 7122 | search: |
60d2adbb | 7123 | have_caching_bg = false; |
a5e681d9 JB |
7124 | if (index == 0 || index == __get_raid_index(flags)) |
7125 | full_search = true; | |
80eb234a | 7126 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7127 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7128 | list) { | |
6226cb0a | 7129 | u64 offset; |
817d52f8 | 7130 | int cached; |
8a1413a2 | 7131 | |
e570fd27 | 7132 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7133 | search_start = block_group->key.objectid; |
42e70e7a | 7134 | |
83a50de9 CM |
7135 | /* |
7136 | * this can happen if we end up cycling through all the | |
7137 | * raid types, but we want to make sure we only allocate | |
7138 | * for the proper type. | |
7139 | */ | |
b6919a58 | 7140 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7141 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7142 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7143 | BTRFS_BLOCK_GROUP_RAID5 | |
7144 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7145 | BTRFS_BLOCK_GROUP_RAID10; |
7146 | ||
7147 | /* | |
7148 | * if they asked for extra copies and this block group | |
7149 | * doesn't provide them, bail. This does allow us to | |
7150 | * fill raid0 from raid1. | |
7151 | */ | |
b6919a58 | 7152 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7153 | goto loop; |
7154 | } | |
7155 | ||
2552d17e | 7156 | have_block_group: |
291c7d2f JB |
7157 | cached = block_group_cache_done(block_group); |
7158 | if (unlikely(!cached)) { | |
a5e681d9 | 7159 | have_caching_bg = true; |
f6373bf3 | 7160 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7161 | BUG_ON(ret < 0); |
7162 | ret = 0; | |
817d52f8 JB |
7163 | } |
7164 | ||
36cce922 JB |
7165 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7166 | goto loop; | |
ea6a478e | 7167 | if (unlikely(block_group->ro)) |
2552d17e | 7168 | goto loop; |
0f9dd46c | 7169 | |
0a24325e | 7170 | /* |
062c05c4 AO |
7171 | * Ok we want to try and use the cluster allocator, so |
7172 | * lets look there | |
0a24325e | 7173 | */ |
c759c4e1 | 7174 | if (last_ptr && use_cluster) { |
215a63d1 | 7175 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7176 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7177 | /* |
7178 | * the refill lock keeps out other | |
7179 | * people trying to start a new cluster | |
7180 | */ | |
e570fd27 MX |
7181 | used_block_group = btrfs_lock_cluster(block_group, |
7182 | last_ptr, | |
7183 | delalloc); | |
7184 | if (!used_block_group) | |
44fb5511 | 7185 | goto refill_cluster; |
274bd4fb | 7186 | |
e570fd27 MX |
7187 | if (used_block_group != block_group && |
7188 | (used_block_group->ro || | |
7189 | !block_group_bits(used_block_group, flags))) | |
7190 | goto release_cluster; | |
44fb5511 | 7191 | |
274bd4fb | 7192 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7193 | last_ptr, |
7194 | num_bytes, | |
7195 | used_block_group->key.objectid, | |
7196 | &max_extent_size); | |
fa9c0d79 CM |
7197 | if (offset) { |
7198 | /* we have a block, we're done */ | |
7199 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 | 7200 | trace_btrfs_reserve_extent_cluster(root, |
89d4346a MX |
7201 | used_block_group, |
7202 | search_start, num_bytes); | |
215a63d1 | 7203 | if (used_block_group != block_group) { |
e570fd27 MX |
7204 | btrfs_release_block_group(block_group, |
7205 | delalloc); | |
215a63d1 MX |
7206 | block_group = used_block_group; |
7207 | } | |
fa9c0d79 CM |
7208 | goto checks; |
7209 | } | |
7210 | ||
274bd4fb | 7211 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7212 | release_cluster: |
062c05c4 AO |
7213 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7214 | * set up a new clusters, so lets just skip it | |
7215 | * and let the allocator find whatever block | |
7216 | * it can find. If we reach this point, we | |
7217 | * will have tried the cluster allocator | |
7218 | * plenty of times and not have found | |
7219 | * anything, so we are likely way too | |
7220 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7221 | * anything. |
7222 | * | |
7223 | * However, if the cluster is taken from the | |
7224 | * current block group, release the cluster | |
7225 | * first, so that we stand a better chance of | |
7226 | * succeeding in the unclustered | |
7227 | * allocation. */ | |
7228 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7229 | used_block_group != block_group) { |
062c05c4 | 7230 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7231 | btrfs_release_block_group(used_block_group, |
7232 | delalloc); | |
062c05c4 AO |
7233 | goto unclustered_alloc; |
7234 | } | |
7235 | ||
fa9c0d79 CM |
7236 | /* |
7237 | * this cluster didn't work out, free it and | |
7238 | * start over | |
7239 | */ | |
7240 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7241 | ||
e570fd27 MX |
7242 | if (used_block_group != block_group) |
7243 | btrfs_release_block_group(used_block_group, | |
7244 | delalloc); | |
7245 | refill_cluster: | |
a5f6f719 AO |
7246 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7247 | spin_unlock(&last_ptr->refill_lock); | |
7248 | goto unclustered_alloc; | |
7249 | } | |
7250 | ||
8de972b4 CM |
7251 | aligned_cluster = max_t(unsigned long, |
7252 | empty_cluster + empty_size, | |
7253 | block_group->full_stripe_len); | |
7254 | ||
fa9c0d79 | 7255 | /* allocate a cluster in this block group */ |
00361589 JB |
7256 | ret = btrfs_find_space_cluster(root, block_group, |
7257 | last_ptr, search_start, | |
7258 | num_bytes, | |
7259 | aligned_cluster); | |
fa9c0d79 CM |
7260 | if (ret == 0) { |
7261 | /* | |
7262 | * now pull our allocation out of this | |
7263 | * cluster | |
7264 | */ | |
7265 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7266 | last_ptr, |
7267 | num_bytes, | |
7268 | search_start, | |
7269 | &max_extent_size); | |
fa9c0d79 CM |
7270 | if (offset) { |
7271 | /* we found one, proceed */ | |
7272 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
7273 | trace_btrfs_reserve_extent_cluster(root, |
7274 | block_group, search_start, | |
7275 | num_bytes); | |
fa9c0d79 CM |
7276 | goto checks; |
7277 | } | |
0a24325e JB |
7278 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7279 | && !failed_cluster_refill) { | |
817d52f8 JB |
7280 | spin_unlock(&last_ptr->refill_lock); |
7281 | ||
0a24325e | 7282 | failed_cluster_refill = true; |
817d52f8 JB |
7283 | wait_block_group_cache_progress(block_group, |
7284 | num_bytes + empty_cluster + empty_size); | |
7285 | goto have_block_group; | |
fa9c0d79 | 7286 | } |
817d52f8 | 7287 | |
fa9c0d79 CM |
7288 | /* |
7289 | * at this point we either didn't find a cluster | |
7290 | * or we weren't able to allocate a block from our | |
7291 | * cluster. Free the cluster we've been trying | |
7292 | * to use, and go to the next block group | |
7293 | */ | |
0a24325e | 7294 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7295 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7296 | goto loop; |
fa9c0d79 CM |
7297 | } |
7298 | ||
062c05c4 | 7299 | unclustered_alloc: |
c759c4e1 JB |
7300 | /* |
7301 | * We are doing an unclustered alloc, set the fragmented flag so | |
7302 | * we don't bother trying to setup a cluster again until we get | |
7303 | * more space. | |
7304 | */ | |
7305 | if (unlikely(last_ptr)) { | |
7306 | spin_lock(&last_ptr->lock); | |
7307 | last_ptr->fragmented = 1; | |
7308 | spin_unlock(&last_ptr->lock); | |
7309 | } | |
a5f6f719 AO |
7310 | spin_lock(&block_group->free_space_ctl->tree_lock); |
7311 | if (cached && | |
7312 | block_group->free_space_ctl->free_space < | |
7313 | num_bytes + empty_cluster + empty_size) { | |
a4820398 MX |
7314 | if (block_group->free_space_ctl->free_space > |
7315 | max_extent_size) | |
7316 | max_extent_size = | |
7317 | block_group->free_space_ctl->free_space; | |
a5f6f719 AO |
7318 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
7319 | goto loop; | |
7320 | } | |
7321 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
7322 | ||
6226cb0a | 7323 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7324 | num_bytes, empty_size, |
7325 | &max_extent_size); | |
1cdda9b8 JB |
7326 | /* |
7327 | * If we didn't find a chunk, and we haven't failed on this | |
7328 | * block group before, and this block group is in the middle of | |
7329 | * caching and we are ok with waiting, then go ahead and wait | |
7330 | * for progress to be made, and set failed_alloc to true. | |
7331 | * | |
7332 | * If failed_alloc is true then we've already waited on this | |
7333 | * block group once and should move on to the next block group. | |
7334 | */ | |
7335 | if (!offset && !failed_alloc && !cached && | |
7336 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7337 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7338 | num_bytes + empty_size); |
7339 | failed_alloc = true; | |
817d52f8 | 7340 | goto have_block_group; |
1cdda9b8 JB |
7341 | } else if (!offset) { |
7342 | goto loop; | |
817d52f8 | 7343 | } |
fa9c0d79 | 7344 | checks: |
4e54b17a | 7345 | search_start = ALIGN(offset, root->stripesize); |
25179201 | 7346 | |
2552d17e JB |
7347 | /* move on to the next group */ |
7348 | if (search_start + num_bytes > | |
215a63d1 MX |
7349 | block_group->key.objectid + block_group->key.offset) { |
7350 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7351 | goto loop; |
6226cb0a | 7352 | } |
f5a31e16 | 7353 | |
f0486c68 | 7354 | if (offset < search_start) |
215a63d1 | 7355 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7356 | search_start - offset); |
7357 | BUG_ON(offset > search_start); | |
2552d17e | 7358 | |
215a63d1 | 7359 | ret = btrfs_update_reserved_bytes(block_group, num_bytes, |
e570fd27 | 7360 | alloc_type, delalloc); |
f0486c68 | 7361 | if (ret == -EAGAIN) { |
215a63d1 | 7362 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7363 | goto loop; |
0f9dd46c | 7364 | } |
0b86a832 | 7365 | |
f0486c68 | 7366 | /* we are all good, lets return */ |
2552d17e JB |
7367 | ins->objectid = search_start; |
7368 | ins->offset = num_bytes; | |
d2fb3437 | 7369 | |
3f7de037 JB |
7370 | trace_btrfs_reserve_extent(orig_root, block_group, |
7371 | search_start, num_bytes); | |
e570fd27 | 7372 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7373 | break; |
7374 | loop: | |
0a24325e | 7375 | failed_cluster_refill = false; |
1cdda9b8 | 7376 | failed_alloc = false; |
b742bb82 | 7377 | BUG_ON(index != get_block_group_index(block_group)); |
e570fd27 | 7378 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7379 | } |
7380 | up_read(&space_info->groups_sem); | |
7381 | ||
13a0db5a | 7382 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7383 | && !orig_have_caching_bg) | |
7384 | orig_have_caching_bg = true; | |
7385 | ||
60d2adbb MX |
7386 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7387 | goto search; | |
7388 | ||
b742bb82 YZ |
7389 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7390 | goto search; | |
7391 | ||
285ff5af | 7392 | /* |
ccf0e725 JB |
7393 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7394 | * caching kthreads as we move along | |
817d52f8 JB |
7395 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7396 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7397 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7398 | * again | |
fa9c0d79 | 7399 | */ |
723bda20 | 7400 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7401 | index = 0; |
a5e681d9 JB |
7402 | if (loop == LOOP_CACHING_NOWAIT) { |
7403 | /* | |
7404 | * We want to skip the LOOP_CACHING_WAIT step if we | |
7405 | * don't have any unached bgs and we've alrelady done a | |
7406 | * full search through. | |
7407 | */ | |
13a0db5a | 7408 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7409 | loop = LOOP_CACHING_WAIT; |
7410 | else | |
7411 | loop = LOOP_ALLOC_CHUNK; | |
7412 | } else { | |
7413 | loop++; | |
7414 | } | |
7415 | ||
817d52f8 | 7416 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7417 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7418 | int exist = 0; |
7419 | ||
7420 | trans = current->journal_info; | |
7421 | if (trans) | |
7422 | exist = 1; | |
7423 | else | |
7424 | trans = btrfs_join_transaction(root); | |
00361589 | 7425 | |
00361589 JB |
7426 | if (IS_ERR(trans)) { |
7427 | ret = PTR_ERR(trans); | |
7428 | goto out; | |
7429 | } | |
7430 | ||
b6919a58 | 7431 | ret = do_chunk_alloc(trans, root, flags, |
ea658bad | 7432 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7433 | |
7434 | /* | |
7435 | * If we can't allocate a new chunk we've already looped | |
7436 | * through at least once, move on to the NO_EMPTY_SIZE | |
7437 | * case. | |
7438 | */ | |
7439 | if (ret == -ENOSPC) | |
7440 | loop = LOOP_NO_EMPTY_SIZE; | |
7441 | ||
ea658bad JB |
7442 | /* |
7443 | * Do not bail out on ENOSPC since we | |
7444 | * can do more things. | |
7445 | */ | |
00361589 | 7446 | if (ret < 0 && ret != -ENOSPC) |
ea658bad JB |
7447 | btrfs_abort_transaction(trans, |
7448 | root, ret); | |
00361589 JB |
7449 | else |
7450 | ret = 0; | |
f017f15f WS |
7451 | if (!exist) |
7452 | btrfs_end_transaction(trans, root); | |
00361589 | 7453 | if (ret) |
ea658bad | 7454 | goto out; |
2552d17e JB |
7455 | } |
7456 | ||
723bda20 | 7457 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7458 | /* |
7459 | * Don't loop again if we already have no empty_size and | |
7460 | * no empty_cluster. | |
7461 | */ | |
7462 | if (empty_size == 0 && | |
7463 | empty_cluster == 0) { | |
7464 | ret = -ENOSPC; | |
7465 | goto out; | |
7466 | } | |
723bda20 JB |
7467 | empty_size = 0; |
7468 | empty_cluster = 0; | |
fa9c0d79 | 7469 | } |
723bda20 JB |
7470 | |
7471 | goto search; | |
2552d17e JB |
7472 | } else if (!ins->objectid) { |
7473 | ret = -ENOSPC; | |
d82a6f1d | 7474 | } else if (ins->objectid) { |
c759c4e1 JB |
7475 | if (!use_cluster && last_ptr) { |
7476 | spin_lock(&last_ptr->lock); | |
7477 | last_ptr->window_start = ins->objectid; | |
7478 | spin_unlock(&last_ptr->lock); | |
7479 | } | |
80eb234a | 7480 | ret = 0; |
be744175 | 7481 | } |
79787eaa | 7482 | out: |
4f4db217 JB |
7483 | if (ret == -ENOSPC) { |
7484 | spin_lock(&space_info->lock); | |
7485 | space_info->max_extent_size = max_extent_size; | |
7486 | spin_unlock(&space_info->lock); | |
a4820398 | 7487 | ins->offset = max_extent_size; |
4f4db217 | 7488 | } |
0f70abe2 | 7489 | return ret; |
fec577fb | 7490 | } |
ec44a35c | 7491 | |
9ed74f2d JB |
7492 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
7493 | int dump_block_groups) | |
0f9dd46c JB |
7494 | { |
7495 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7496 | int index = 0; |
0f9dd46c | 7497 | |
9ed74f2d | 7498 | spin_lock(&info->lock); |
efe120a0 | 7499 | printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n", |
c1c9ff7c GU |
7500 | info->flags, |
7501 | info->total_bytes - info->bytes_used - info->bytes_pinned - | |
7502 | info->bytes_reserved - info->bytes_readonly, | |
d397712b | 7503 | (info->full) ? "" : "not "); |
efe120a0 | 7504 | printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, " |
8929ecfa | 7505 | "reserved=%llu, may_use=%llu, readonly=%llu\n", |
c1c9ff7c GU |
7506 | info->total_bytes, info->bytes_used, info->bytes_pinned, |
7507 | info->bytes_reserved, info->bytes_may_use, | |
7508 | info->bytes_readonly); | |
9ed74f2d JB |
7509 | spin_unlock(&info->lock); |
7510 | ||
7511 | if (!dump_block_groups) | |
7512 | return; | |
0f9dd46c | 7513 | |
80eb234a | 7514 | down_read(&info->groups_sem); |
b742bb82 YZ |
7515 | again: |
7516 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7517 | spin_lock(&cache->lock); |
efe120a0 FH |
7518 | printk(KERN_INFO "BTRFS: " |
7519 | "block group %llu has %llu bytes, " | |
7520 | "%llu used %llu pinned %llu reserved %s\n", | |
c1c9ff7c GU |
7521 | cache->key.objectid, cache->key.offset, |
7522 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7523 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7524 | btrfs_dump_free_space(cache, bytes); |
7525 | spin_unlock(&cache->lock); | |
7526 | } | |
b742bb82 YZ |
7527 | if (++index < BTRFS_NR_RAID_TYPES) |
7528 | goto again; | |
80eb234a | 7529 | up_read(&info->groups_sem); |
0f9dd46c | 7530 | } |
e8569813 | 7531 | |
00361589 | 7532 | int btrfs_reserve_extent(struct btrfs_root *root, |
11833d66 YZ |
7533 | u64 num_bytes, u64 min_alloc_size, |
7534 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7535 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7536 | { |
36af4e07 | 7537 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7538 | u64 flags; |
fec577fb | 7539 | int ret; |
925baedd | 7540 | |
b6919a58 | 7541 | flags = btrfs_get_alloc_profile(root, is_data); |
98d20f67 | 7542 | again: |
db94535d | 7543 | WARN_ON(num_bytes < root->sectorsize); |
00361589 | 7544 | ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins, |
e570fd27 | 7545 | flags, delalloc); |
3b951516 | 7546 | |
9e622d6b | 7547 | if (ret == -ENOSPC) { |
a4820398 MX |
7548 | if (!final_tried && ins->offset) { |
7549 | num_bytes = min(num_bytes >> 1, ins->offset); | |
24542bf7 | 7550 | num_bytes = round_down(num_bytes, root->sectorsize); |
9e622d6b | 7551 | num_bytes = max(num_bytes, min_alloc_size); |
9e622d6b MX |
7552 | if (num_bytes == min_alloc_size) |
7553 | final_tried = true; | |
7554 | goto again; | |
7555 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
7556 | struct btrfs_space_info *sinfo; | |
7557 | ||
b6919a58 | 7558 | sinfo = __find_space_info(root->fs_info, flags); |
c2cf52eb | 7559 | btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu", |
c1c9ff7c | 7560 | flags, num_bytes); |
53804280 JM |
7561 | if (sinfo) |
7562 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 7563 | } |
925baedd | 7564 | } |
0f9dd46c JB |
7565 | |
7566 | return ret; | |
e6dcd2dc CM |
7567 | } |
7568 | ||
e688b725 | 7569 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 MX |
7570 | u64 start, u64 len, |
7571 | int pin, int delalloc) | |
65b51a00 | 7572 | { |
0f9dd46c | 7573 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 7574 | int ret = 0; |
0f9dd46c | 7575 | |
0f9dd46c JB |
7576 | cache = btrfs_lookup_block_group(root->fs_info, start); |
7577 | if (!cache) { | |
c2cf52eb | 7578 | btrfs_err(root->fs_info, "Unable to find block group for %llu", |
c1c9ff7c | 7579 | start); |
0f9dd46c JB |
7580 | return -ENOSPC; |
7581 | } | |
1f3c79a2 | 7582 | |
e688b725 CM |
7583 | if (pin) |
7584 | pin_down_extent(root, cache, start, len, 1); | |
7585 | else { | |
dcc82f47 FM |
7586 | if (btrfs_test_opt(root, DISCARD)) |
7587 | ret = btrfs_discard_extent(root, start, len, NULL); | |
e688b725 | 7588 | btrfs_add_free_space(cache, start, len); |
e570fd27 | 7589 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc); |
e688b725 | 7590 | } |
31193213 | 7591 | |
fa9c0d79 | 7592 | btrfs_put_block_group(cache); |
817d52f8 | 7593 | |
1abe9b8a | 7594 | trace_btrfs_reserved_extent_free(root, start, len); |
7595 | ||
e6dcd2dc CM |
7596 | return ret; |
7597 | } | |
7598 | ||
e688b725 | 7599 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 | 7600 | u64 start, u64 len, int delalloc) |
e688b725 | 7601 | { |
e570fd27 | 7602 | return __btrfs_free_reserved_extent(root, start, len, 0, delalloc); |
e688b725 CM |
7603 | } |
7604 | ||
7605 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
7606 | u64 start, u64 len) | |
7607 | { | |
e570fd27 | 7608 | return __btrfs_free_reserved_extent(root, start, len, 1, 0); |
e688b725 CM |
7609 | } |
7610 | ||
5d4f98a2 YZ |
7611 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
7612 | struct btrfs_root *root, | |
7613 | u64 parent, u64 root_objectid, | |
7614 | u64 flags, u64 owner, u64 offset, | |
7615 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
7616 | { |
7617 | int ret; | |
5d4f98a2 | 7618 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 7619 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 7620 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 7621 | struct btrfs_path *path; |
5d4f98a2 YZ |
7622 | struct extent_buffer *leaf; |
7623 | int type; | |
7624 | u32 size; | |
26b8003f | 7625 | |
5d4f98a2 YZ |
7626 | if (parent > 0) |
7627 | type = BTRFS_SHARED_DATA_REF_KEY; | |
7628 | else | |
7629 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 7630 | |
5d4f98a2 | 7631 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
7632 | |
7633 | path = btrfs_alloc_path(); | |
db5b493a TI |
7634 | if (!path) |
7635 | return -ENOMEM; | |
47e4bb98 | 7636 | |
b9473439 | 7637 | path->leave_spinning = 1; |
5d4f98a2 YZ |
7638 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
7639 | ins, size); | |
79787eaa JM |
7640 | if (ret) { |
7641 | btrfs_free_path(path); | |
7642 | return ret; | |
7643 | } | |
0f9dd46c | 7644 | |
5d4f98a2 YZ |
7645 | leaf = path->nodes[0]; |
7646 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 7647 | struct btrfs_extent_item); |
5d4f98a2 YZ |
7648 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
7649 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7650 | btrfs_set_extent_flags(leaf, extent_item, | |
7651 | flags | BTRFS_EXTENT_FLAG_DATA); | |
7652 | ||
7653 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
7654 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
7655 | if (parent > 0) { | |
7656 | struct btrfs_shared_data_ref *ref; | |
7657 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
7658 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7659 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
7660 | } else { | |
7661 | struct btrfs_extent_data_ref *ref; | |
7662 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
7663 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
7664 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
7665 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
7666 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
7667 | } | |
47e4bb98 CM |
7668 | |
7669 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 7670 | btrfs_free_path(path); |
f510cfec | 7671 | |
ce93ec54 | 7672 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 7673 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7674 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 7675 | ins->objectid, ins->offset); |
f5947066 CM |
7676 | BUG(); |
7677 | } | |
0be5dc67 | 7678 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
e6dcd2dc CM |
7679 | return ret; |
7680 | } | |
7681 | ||
5d4f98a2 YZ |
7682 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
7683 | struct btrfs_root *root, | |
7684 | u64 parent, u64 root_objectid, | |
7685 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 7686 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
7687 | { |
7688 | int ret; | |
5d4f98a2 YZ |
7689 | struct btrfs_fs_info *fs_info = root->fs_info; |
7690 | struct btrfs_extent_item *extent_item; | |
7691 | struct btrfs_tree_block_info *block_info; | |
7692 | struct btrfs_extent_inline_ref *iref; | |
7693 | struct btrfs_path *path; | |
7694 | struct extent_buffer *leaf; | |
3173a18f | 7695 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 7696 | u64 num_bytes = ins->offset; |
3173a18f JB |
7697 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
7698 | SKINNY_METADATA); | |
7699 | ||
7700 | if (!skinny_metadata) | |
7701 | size += sizeof(*block_info); | |
1c2308f8 | 7702 | |
5d4f98a2 | 7703 | path = btrfs_alloc_path(); |
857cc2fc JB |
7704 | if (!path) { |
7705 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, | |
707e8a07 | 7706 | root->nodesize); |
d8926bb3 | 7707 | return -ENOMEM; |
857cc2fc | 7708 | } |
56bec294 | 7709 | |
5d4f98a2 YZ |
7710 | path->leave_spinning = 1; |
7711 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
7712 | ins, size); | |
79787eaa | 7713 | if (ret) { |
dd825259 | 7714 | btrfs_free_path(path); |
857cc2fc | 7715 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, |
707e8a07 | 7716 | root->nodesize); |
79787eaa JM |
7717 | return ret; |
7718 | } | |
5d4f98a2 YZ |
7719 | |
7720 | leaf = path->nodes[0]; | |
7721 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
7722 | struct btrfs_extent_item); | |
7723 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
7724 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7725 | btrfs_set_extent_flags(leaf, extent_item, | |
7726 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 7727 | |
3173a18f JB |
7728 | if (skinny_metadata) { |
7729 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
707e8a07 | 7730 | num_bytes = root->nodesize; |
3173a18f JB |
7731 | } else { |
7732 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
7733 | btrfs_set_tree_block_key(leaf, block_info, key); | |
7734 | btrfs_set_tree_block_level(leaf, block_info, level); | |
7735 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
7736 | } | |
5d4f98a2 | 7737 | |
5d4f98a2 YZ |
7738 | if (parent > 0) { |
7739 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
7740 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7741 | BTRFS_SHARED_BLOCK_REF_KEY); | |
7742 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7743 | } else { | |
7744 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7745 | BTRFS_TREE_BLOCK_REF_KEY); | |
7746 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
7747 | } | |
7748 | ||
7749 | btrfs_mark_buffer_dirty(leaf); | |
7750 | btrfs_free_path(path); | |
7751 | ||
ce93ec54 JB |
7752 | ret = update_block_group(trans, root, ins->objectid, root->nodesize, |
7753 | 1); | |
79787eaa | 7754 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7755 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 7756 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
7757 | BUG(); |
7758 | } | |
0be5dc67 | 7759 | |
707e8a07 | 7760 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize); |
5d4f98a2 YZ |
7761 | return ret; |
7762 | } | |
7763 | ||
7764 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
7765 | struct btrfs_root *root, | |
7766 | u64 root_objectid, u64 owner, | |
5846a3c2 QW |
7767 | u64 offset, u64 ram_bytes, |
7768 | struct btrfs_key *ins) | |
5d4f98a2 YZ |
7769 | { |
7770 | int ret; | |
7771 | ||
7772 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
7773 | ||
66d7e7f0 AJ |
7774 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
7775 | ins->offset, 0, | |
7776 | root_objectid, owner, offset, | |
5846a3c2 QW |
7777 | ram_bytes, BTRFS_ADD_DELAYED_EXTENT, |
7778 | NULL); | |
e6dcd2dc CM |
7779 | return ret; |
7780 | } | |
e02119d5 CM |
7781 | |
7782 | /* | |
7783 | * this is used by the tree logging recovery code. It records that | |
7784 | * an extent has been allocated and makes sure to clear the free | |
7785 | * space cache bits as well | |
7786 | */ | |
5d4f98a2 YZ |
7787 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
7788 | struct btrfs_root *root, | |
7789 | u64 root_objectid, u64 owner, u64 offset, | |
7790 | struct btrfs_key *ins) | |
e02119d5 CM |
7791 | { |
7792 | int ret; | |
7793 | struct btrfs_block_group_cache *block_group; | |
11833d66 | 7794 | |
8c2a1a30 JB |
7795 | /* |
7796 | * Mixed block groups will exclude before processing the log so we only | |
7797 | * need to do the exlude dance if this fs isn't mixed. | |
7798 | */ | |
7799 | if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) { | |
7800 | ret = __exclude_logged_extent(root, ins->objectid, ins->offset); | |
b50c6e25 | 7801 | if (ret) |
8c2a1a30 | 7802 | return ret; |
11833d66 YZ |
7803 | } |
7804 | ||
8c2a1a30 JB |
7805 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
7806 | if (!block_group) | |
7807 | return -EINVAL; | |
7808 | ||
fb25e914 | 7809 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
e570fd27 | 7810 | RESERVE_ALLOC_NO_ACCOUNT, 0); |
79787eaa | 7811 | BUG_ON(ret); /* logic error */ |
5d4f98a2 YZ |
7812 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
7813 | 0, owner, offset, ins, 1); | |
b50c6e25 | 7814 | btrfs_put_block_group(block_group); |
e02119d5 CM |
7815 | return ret; |
7816 | } | |
7817 | ||
48a3b636 ES |
7818 | static struct extent_buffer * |
7819 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 7820 | u64 bytenr, int level) |
65b51a00 CM |
7821 | { |
7822 | struct extent_buffer *buf; | |
7823 | ||
a83fffb7 | 7824 | buf = btrfs_find_create_tree_block(root, bytenr); |
65b51a00 CM |
7825 | if (!buf) |
7826 | return ERR_PTR(-ENOMEM); | |
7827 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 7828 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 7829 | btrfs_tree_lock(buf); |
01d58472 | 7830 | clean_tree_block(trans, root->fs_info, buf); |
3083ee2e | 7831 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
7832 | |
7833 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 7834 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 7835 | |
d0c803c4 | 7836 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 7837 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
7838 | /* |
7839 | * we allow two log transactions at a time, use different | |
7840 | * EXENT bit to differentiate dirty pages. | |
7841 | */ | |
656f30db | 7842 | if (buf->log_index == 0) |
8cef4e16 YZ |
7843 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
7844 | buf->start + buf->len - 1, GFP_NOFS); | |
7845 | else | |
7846 | set_extent_new(&root->dirty_log_pages, buf->start, | |
7847 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 | 7848 | } else { |
656f30db | 7849 | buf->log_index = -1; |
d0c803c4 | 7850 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 7851 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 7852 | } |
65b51a00 | 7853 | trans->blocks_used++; |
b4ce94de | 7854 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
7855 | return buf; |
7856 | } | |
7857 | ||
f0486c68 YZ |
7858 | static struct btrfs_block_rsv * |
7859 | use_block_rsv(struct btrfs_trans_handle *trans, | |
7860 | struct btrfs_root *root, u32 blocksize) | |
7861 | { | |
7862 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 7863 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 | 7864 | int ret; |
d88033db | 7865 | bool global_updated = false; |
f0486c68 YZ |
7866 | |
7867 | block_rsv = get_block_rsv(trans, root); | |
7868 | ||
b586b323 MX |
7869 | if (unlikely(block_rsv->size == 0)) |
7870 | goto try_reserve; | |
d88033db | 7871 | again: |
f0486c68 YZ |
7872 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
7873 | if (!ret) | |
7874 | return block_rsv; | |
7875 | ||
b586b323 MX |
7876 | if (block_rsv->failfast) |
7877 | return ERR_PTR(ret); | |
7878 | ||
d88033db MX |
7879 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
7880 | global_updated = true; | |
7881 | update_global_block_rsv(root->fs_info); | |
7882 | goto again; | |
7883 | } | |
7884 | ||
b586b323 MX |
7885 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { |
7886 | static DEFINE_RATELIMIT_STATE(_rs, | |
7887 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
7888 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
7889 | if (__ratelimit(&_rs)) | |
7890 | WARN(1, KERN_DEBUG | |
efe120a0 | 7891 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
7892 | } |
7893 | try_reserve: | |
7894 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
7895 | BTRFS_RESERVE_NO_FLUSH); | |
7896 | if (!ret) | |
7897 | return block_rsv; | |
7898 | /* | |
7899 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
7900 | * the global reserve if its space type is the same as the global |
7901 | * reservation. | |
b586b323 | 7902 | */ |
5881cfc9 MX |
7903 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
7904 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
7905 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
7906 | if (!ret) | |
7907 | return global_rsv; | |
7908 | } | |
7909 | return ERR_PTR(ret); | |
f0486c68 YZ |
7910 | } |
7911 | ||
8c2a3ca2 JB |
7912 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
7913 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
7914 | { |
7915 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 7916 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
7917 | } |
7918 | ||
fec577fb | 7919 | /* |
f0486c68 | 7920 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 7921 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 7922 | */ |
4d75f8a9 DS |
7923 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
7924 | struct btrfs_root *root, | |
5d4f98a2 YZ |
7925 | u64 parent, u64 root_objectid, |
7926 | struct btrfs_disk_key *key, int level, | |
5581a51a | 7927 | u64 hint, u64 empty_size) |
fec577fb | 7928 | { |
e2fa7227 | 7929 | struct btrfs_key ins; |
f0486c68 | 7930 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 7931 | struct extent_buffer *buf; |
67b7859e | 7932 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
7933 | u64 flags = 0; |
7934 | int ret; | |
4d75f8a9 | 7935 | u32 blocksize = root->nodesize; |
3173a18f JB |
7936 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
7937 | SKINNY_METADATA); | |
fec577fb | 7938 | |
fccb84c9 | 7939 | if (btrfs_test_is_dummy_root(root)) { |
faa2dbf0 | 7940 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 7941 | level); |
faa2dbf0 JB |
7942 | if (!IS_ERR(buf)) |
7943 | root->alloc_bytenr += blocksize; | |
7944 | return buf; | |
7945 | } | |
fccb84c9 | 7946 | |
f0486c68 YZ |
7947 | block_rsv = use_block_rsv(trans, root, blocksize); |
7948 | if (IS_ERR(block_rsv)) | |
7949 | return ERR_CAST(block_rsv); | |
7950 | ||
00361589 | 7951 | ret = btrfs_reserve_extent(root, blocksize, blocksize, |
e570fd27 | 7952 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
7953 | if (ret) |
7954 | goto out_unuse; | |
55c69072 | 7955 | |
fe864576 | 7956 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
7957 | if (IS_ERR(buf)) { |
7958 | ret = PTR_ERR(buf); | |
7959 | goto out_free_reserved; | |
7960 | } | |
f0486c68 YZ |
7961 | |
7962 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
7963 | if (parent == 0) | |
7964 | parent = ins.objectid; | |
7965 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7966 | } else | |
7967 | BUG_ON(parent > 0); | |
7968 | ||
7969 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 7970 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
7971 | if (!extent_op) { |
7972 | ret = -ENOMEM; | |
7973 | goto out_free_buf; | |
7974 | } | |
f0486c68 YZ |
7975 | if (key) |
7976 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
7977 | else | |
7978 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
7979 | extent_op->flags_to_set = flags; | |
3173a18f JB |
7980 | if (skinny_metadata) |
7981 | extent_op->update_key = 0; | |
7982 | else | |
7983 | extent_op->update_key = 1; | |
f0486c68 YZ |
7984 | extent_op->update_flags = 1; |
7985 | extent_op->is_data = 0; | |
b1c79e09 | 7986 | extent_op->level = level; |
f0486c68 | 7987 | |
66d7e7f0 | 7988 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
67b7859e OS |
7989 | ins.objectid, ins.offset, |
7990 | parent, root_objectid, level, | |
7991 | BTRFS_ADD_DELAYED_EXTENT, | |
b06c4bf5 | 7992 | extent_op); |
67b7859e OS |
7993 | if (ret) |
7994 | goto out_free_delayed; | |
f0486c68 | 7995 | } |
fec577fb | 7996 | return buf; |
67b7859e OS |
7997 | |
7998 | out_free_delayed: | |
7999 | btrfs_free_delayed_extent_op(extent_op); | |
8000 | out_free_buf: | |
8001 | free_extent_buffer(buf); | |
8002 | out_free_reserved: | |
8003 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0); | |
8004 | out_unuse: | |
8005 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); | |
8006 | return ERR_PTR(ret); | |
fec577fb | 8007 | } |
a28ec197 | 8008 | |
2c47e605 YZ |
8009 | struct walk_control { |
8010 | u64 refs[BTRFS_MAX_LEVEL]; | |
8011 | u64 flags[BTRFS_MAX_LEVEL]; | |
8012 | struct btrfs_key update_progress; | |
8013 | int stage; | |
8014 | int level; | |
8015 | int shared_level; | |
8016 | int update_ref; | |
8017 | int keep_locks; | |
1c4850e2 YZ |
8018 | int reada_slot; |
8019 | int reada_count; | |
66d7e7f0 | 8020 | int for_reloc; |
2c47e605 YZ |
8021 | }; |
8022 | ||
8023 | #define DROP_REFERENCE 1 | |
8024 | #define UPDATE_BACKREF 2 | |
8025 | ||
1c4850e2 YZ |
8026 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8027 | struct btrfs_root *root, | |
8028 | struct walk_control *wc, | |
8029 | struct btrfs_path *path) | |
6407bf6d | 8030 | { |
1c4850e2 YZ |
8031 | u64 bytenr; |
8032 | u64 generation; | |
8033 | u64 refs; | |
94fcca9f | 8034 | u64 flags; |
5d4f98a2 | 8035 | u32 nritems; |
1c4850e2 YZ |
8036 | u32 blocksize; |
8037 | struct btrfs_key key; | |
8038 | struct extent_buffer *eb; | |
6407bf6d | 8039 | int ret; |
1c4850e2 YZ |
8040 | int slot; |
8041 | int nread = 0; | |
6407bf6d | 8042 | |
1c4850e2 YZ |
8043 | if (path->slots[wc->level] < wc->reada_slot) { |
8044 | wc->reada_count = wc->reada_count * 2 / 3; | |
8045 | wc->reada_count = max(wc->reada_count, 2); | |
8046 | } else { | |
8047 | wc->reada_count = wc->reada_count * 3 / 2; | |
8048 | wc->reada_count = min_t(int, wc->reada_count, | |
8049 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
8050 | } | |
7bb86316 | 8051 | |
1c4850e2 YZ |
8052 | eb = path->nodes[wc->level]; |
8053 | nritems = btrfs_header_nritems(eb); | |
707e8a07 | 8054 | blocksize = root->nodesize; |
bd56b302 | 8055 | |
1c4850e2 YZ |
8056 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8057 | if (nread >= wc->reada_count) | |
8058 | break; | |
bd56b302 | 8059 | |
2dd3e67b | 8060 | cond_resched(); |
1c4850e2 YZ |
8061 | bytenr = btrfs_node_blockptr(eb, slot); |
8062 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8063 | |
1c4850e2 YZ |
8064 | if (slot == path->slots[wc->level]) |
8065 | goto reada; | |
5d4f98a2 | 8066 | |
1c4850e2 YZ |
8067 | if (wc->stage == UPDATE_BACKREF && |
8068 | generation <= root->root_key.offset) | |
bd56b302 CM |
8069 | continue; |
8070 | ||
94fcca9f | 8071 | /* We don't lock the tree block, it's OK to be racy here */ |
3173a18f JB |
8072 | ret = btrfs_lookup_extent_info(trans, root, bytenr, |
8073 | wc->level - 1, 1, &refs, | |
8074 | &flags); | |
79787eaa JM |
8075 | /* We don't care about errors in readahead. */ |
8076 | if (ret < 0) | |
8077 | continue; | |
94fcca9f YZ |
8078 | BUG_ON(refs == 0); |
8079 | ||
1c4850e2 | 8080 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8081 | if (refs == 1) |
8082 | goto reada; | |
bd56b302 | 8083 | |
94fcca9f YZ |
8084 | if (wc->level == 1 && |
8085 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8086 | continue; | |
1c4850e2 YZ |
8087 | if (!wc->update_ref || |
8088 | generation <= root->root_key.offset) | |
8089 | continue; | |
8090 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8091 | ret = btrfs_comp_cpu_keys(&key, | |
8092 | &wc->update_progress); | |
8093 | if (ret < 0) | |
8094 | continue; | |
94fcca9f YZ |
8095 | } else { |
8096 | if (wc->level == 1 && | |
8097 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8098 | continue; | |
6407bf6d | 8099 | } |
1c4850e2 | 8100 | reada: |
d3e46fea | 8101 | readahead_tree_block(root, bytenr); |
1c4850e2 | 8102 | nread++; |
20524f02 | 8103 | } |
1c4850e2 | 8104 | wc->reada_slot = slot; |
20524f02 | 8105 | } |
2c47e605 | 8106 | |
0ed4792a QW |
8107 | /* |
8108 | * TODO: Modify related function to add related node/leaf to dirty_extent_root, | |
8109 | * for later qgroup accounting. | |
8110 | * | |
8111 | * Current, this function does nothing. | |
8112 | */ | |
1152651a MF |
8113 | static int account_leaf_items(struct btrfs_trans_handle *trans, |
8114 | struct btrfs_root *root, | |
8115 | struct extent_buffer *eb) | |
8116 | { | |
8117 | int nr = btrfs_header_nritems(eb); | |
0ed4792a | 8118 | int i, extent_type; |
1152651a MF |
8119 | struct btrfs_key key; |
8120 | struct btrfs_file_extent_item *fi; | |
8121 | u64 bytenr, num_bytes; | |
8122 | ||
8123 | for (i = 0; i < nr; i++) { | |
8124 | btrfs_item_key_to_cpu(eb, &key, i); | |
8125 | ||
8126 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
8127 | continue; | |
8128 | ||
8129 | fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
8130 | /* filter out non qgroup-accountable extents */ | |
8131 | extent_type = btrfs_file_extent_type(eb, fi); | |
8132 | ||
8133 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
8134 | continue; | |
8135 | ||
8136 | bytenr = btrfs_file_extent_disk_bytenr(eb, fi); | |
8137 | if (!bytenr) | |
8138 | continue; | |
8139 | ||
8140 | num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi); | |
1152651a MF |
8141 | } |
8142 | return 0; | |
8143 | } | |
8144 | ||
8145 | /* | |
8146 | * Walk up the tree from the bottom, freeing leaves and any interior | |
8147 | * nodes which have had all slots visited. If a node (leaf or | |
8148 | * interior) is freed, the node above it will have it's slot | |
8149 | * incremented. The root node will never be freed. | |
8150 | * | |
8151 | * At the end of this function, we should have a path which has all | |
8152 | * slots incremented to the next position for a search. If we need to | |
8153 | * read a new node it will be NULL and the node above it will have the | |
8154 | * correct slot selected for a later read. | |
8155 | * | |
8156 | * If we increment the root nodes slot counter past the number of | |
8157 | * elements, 1 is returned to signal completion of the search. | |
8158 | */ | |
8159 | static int adjust_slots_upwards(struct btrfs_root *root, | |
8160 | struct btrfs_path *path, int root_level) | |
8161 | { | |
8162 | int level = 0; | |
8163 | int nr, slot; | |
8164 | struct extent_buffer *eb; | |
8165 | ||
8166 | if (root_level == 0) | |
8167 | return 1; | |
8168 | ||
8169 | while (level <= root_level) { | |
8170 | eb = path->nodes[level]; | |
8171 | nr = btrfs_header_nritems(eb); | |
8172 | path->slots[level]++; | |
8173 | slot = path->slots[level]; | |
8174 | if (slot >= nr || level == 0) { | |
8175 | /* | |
8176 | * Don't free the root - we will detect this | |
8177 | * condition after our loop and return a | |
8178 | * positive value for caller to stop walking the tree. | |
8179 | */ | |
8180 | if (level != root_level) { | |
8181 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
8182 | path->locks[level] = 0; | |
8183 | ||
8184 | free_extent_buffer(eb); | |
8185 | path->nodes[level] = NULL; | |
8186 | path->slots[level] = 0; | |
8187 | } | |
8188 | } else { | |
8189 | /* | |
8190 | * We have a valid slot to walk back down | |
8191 | * from. Stop here so caller can process these | |
8192 | * new nodes. | |
8193 | */ | |
8194 | break; | |
8195 | } | |
8196 | ||
8197 | level++; | |
8198 | } | |
8199 | ||
8200 | eb = path->nodes[root_level]; | |
8201 | if (path->slots[root_level] >= btrfs_header_nritems(eb)) | |
8202 | return 1; | |
8203 | ||
8204 | return 0; | |
8205 | } | |
8206 | ||
8207 | /* | |
8208 | * root_eb is the subtree root and is locked before this function is called. | |
0ed4792a QW |
8209 | * TODO: Modify this function to mark all (including complete shared node) |
8210 | * to dirty_extent_root to allow it get accounted in qgroup. | |
1152651a MF |
8211 | */ |
8212 | static int account_shared_subtree(struct btrfs_trans_handle *trans, | |
8213 | struct btrfs_root *root, | |
8214 | struct extent_buffer *root_eb, | |
8215 | u64 root_gen, | |
8216 | int root_level) | |
8217 | { | |
8218 | int ret = 0; | |
8219 | int level; | |
8220 | struct extent_buffer *eb = root_eb; | |
8221 | struct btrfs_path *path = NULL; | |
8222 | ||
8223 | BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL); | |
8224 | BUG_ON(root_eb == NULL); | |
8225 | ||
8226 | if (!root->fs_info->quota_enabled) | |
8227 | return 0; | |
8228 | ||
8229 | if (!extent_buffer_uptodate(root_eb)) { | |
8230 | ret = btrfs_read_buffer(root_eb, root_gen); | |
8231 | if (ret) | |
8232 | goto out; | |
8233 | } | |
8234 | ||
8235 | if (root_level == 0) { | |
8236 | ret = account_leaf_items(trans, root, root_eb); | |
8237 | goto out; | |
8238 | } | |
8239 | ||
8240 | path = btrfs_alloc_path(); | |
8241 | if (!path) | |
8242 | return -ENOMEM; | |
8243 | ||
8244 | /* | |
8245 | * Walk down the tree. Missing extent blocks are filled in as | |
8246 | * we go. Metadata is accounted every time we read a new | |
8247 | * extent block. | |
8248 | * | |
8249 | * When we reach a leaf, we account for file extent items in it, | |
8250 | * walk back up the tree (adjusting slot pointers as we go) | |
8251 | * and restart the search process. | |
8252 | */ | |
8253 | extent_buffer_get(root_eb); /* For path */ | |
8254 | path->nodes[root_level] = root_eb; | |
8255 | path->slots[root_level] = 0; | |
8256 | path->locks[root_level] = 0; /* so release_path doesn't try to unlock */ | |
8257 | walk_down: | |
8258 | level = root_level; | |
8259 | while (level >= 0) { | |
8260 | if (path->nodes[level] == NULL) { | |
1152651a MF |
8261 | int parent_slot; |
8262 | u64 child_gen; | |
8263 | u64 child_bytenr; | |
8264 | ||
8265 | /* We need to get child blockptr/gen from | |
8266 | * parent before we can read it. */ | |
8267 | eb = path->nodes[level + 1]; | |
8268 | parent_slot = path->slots[level + 1]; | |
8269 | child_bytenr = btrfs_node_blockptr(eb, parent_slot); | |
8270 | child_gen = btrfs_node_ptr_generation(eb, parent_slot); | |
8271 | ||
ce86cd59 | 8272 | eb = read_tree_block(root, child_bytenr, child_gen); |
64c043de LB |
8273 | if (IS_ERR(eb)) { |
8274 | ret = PTR_ERR(eb); | |
8275 | goto out; | |
8276 | } else if (!extent_buffer_uptodate(eb)) { | |
8635eda9 | 8277 | free_extent_buffer(eb); |
64c043de | 8278 | ret = -EIO; |
1152651a MF |
8279 | goto out; |
8280 | } | |
8281 | ||
8282 | path->nodes[level] = eb; | |
8283 | path->slots[level] = 0; | |
8284 | ||
8285 | btrfs_tree_read_lock(eb); | |
8286 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
8287 | path->locks[level] = BTRFS_READ_LOCK_BLOCKING; | |
1152651a MF |
8288 | } |
8289 | ||
8290 | if (level == 0) { | |
8291 | ret = account_leaf_items(trans, root, path->nodes[level]); | |
8292 | if (ret) | |
8293 | goto out; | |
8294 | ||
8295 | /* Nonzero return here means we completed our search */ | |
8296 | ret = adjust_slots_upwards(root, path, root_level); | |
8297 | if (ret) | |
8298 | break; | |
8299 | ||
8300 | /* Restart search with new slots */ | |
8301 | goto walk_down; | |
8302 | } | |
8303 | ||
8304 | level--; | |
8305 | } | |
8306 | ||
8307 | ret = 0; | |
8308 | out: | |
8309 | btrfs_free_path(path); | |
8310 | ||
8311 | return ret; | |
8312 | } | |
8313 | ||
f82d02d9 | 8314 | /* |
2c016dc2 | 8315 | * helper to process tree block while walking down the tree. |
2c47e605 | 8316 | * |
2c47e605 YZ |
8317 | * when wc->stage == UPDATE_BACKREF, this function updates |
8318 | * back refs for pointers in the block. | |
8319 | * | |
8320 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8321 | */ |
2c47e605 | 8322 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8323 | struct btrfs_root *root, |
2c47e605 | 8324 | struct btrfs_path *path, |
94fcca9f | 8325 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8326 | { |
2c47e605 YZ |
8327 | int level = wc->level; |
8328 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8329 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8330 | int ret; |
8331 | ||
2c47e605 YZ |
8332 | if (wc->stage == UPDATE_BACKREF && |
8333 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8334 | return 1; | |
f82d02d9 | 8335 | |
2c47e605 YZ |
8336 | /* |
8337 | * when reference count of tree block is 1, it won't increase | |
8338 | * again. once full backref flag is set, we never clear it. | |
8339 | */ | |
94fcca9f YZ |
8340 | if (lookup_info && |
8341 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8342 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
8343 | BUG_ON(!path->locks[level]); |
8344 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 8345 | eb->start, level, 1, |
2c47e605 YZ |
8346 | &wc->refs[level], |
8347 | &wc->flags[level]); | |
79787eaa JM |
8348 | BUG_ON(ret == -ENOMEM); |
8349 | if (ret) | |
8350 | return ret; | |
2c47e605 YZ |
8351 | BUG_ON(wc->refs[level] == 0); |
8352 | } | |
5d4f98a2 | 8353 | |
2c47e605 YZ |
8354 | if (wc->stage == DROP_REFERENCE) { |
8355 | if (wc->refs[level] > 1) | |
8356 | return 1; | |
f82d02d9 | 8357 | |
2c47e605 | 8358 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8359 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8360 | path->locks[level] = 0; |
8361 | } | |
8362 | return 0; | |
8363 | } | |
f82d02d9 | 8364 | |
2c47e605 YZ |
8365 | /* wc->stage == UPDATE_BACKREF */ |
8366 | if (!(wc->flags[level] & flag)) { | |
8367 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8368 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8369 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8370 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8371 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 | 8372 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
b1c79e09 JB |
8373 | eb->len, flag, |
8374 | btrfs_header_level(eb), 0); | |
79787eaa | 8375 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8376 | wc->flags[level] |= flag; |
8377 | } | |
8378 | ||
8379 | /* | |
8380 | * the block is shared by multiple trees, so it's not good to | |
8381 | * keep the tree lock | |
8382 | */ | |
8383 | if (path->locks[level] && level > 0) { | |
bd681513 | 8384 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8385 | path->locks[level] = 0; |
8386 | } | |
8387 | return 0; | |
8388 | } | |
8389 | ||
1c4850e2 | 8390 | /* |
2c016dc2 | 8391 | * helper to process tree block pointer. |
1c4850e2 YZ |
8392 | * |
8393 | * when wc->stage == DROP_REFERENCE, this function checks | |
8394 | * reference count of the block pointed to. if the block | |
8395 | * is shared and we need update back refs for the subtree | |
8396 | * rooted at the block, this function changes wc->stage to | |
8397 | * UPDATE_BACKREF. if the block is shared and there is no | |
8398 | * need to update back, this function drops the reference | |
8399 | * to the block. | |
8400 | * | |
8401 | * NOTE: return value 1 means we should stop walking down. | |
8402 | */ | |
8403 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8404 | struct btrfs_root *root, | |
8405 | struct btrfs_path *path, | |
94fcca9f | 8406 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
8407 | { |
8408 | u64 bytenr; | |
8409 | u64 generation; | |
8410 | u64 parent; | |
8411 | u32 blocksize; | |
8412 | struct btrfs_key key; | |
8413 | struct extent_buffer *next; | |
8414 | int level = wc->level; | |
8415 | int reada = 0; | |
8416 | int ret = 0; | |
1152651a | 8417 | bool need_account = false; |
1c4850e2 YZ |
8418 | |
8419 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8420 | path->slots[level]); | |
8421 | /* | |
8422 | * if the lower level block was created before the snapshot | |
8423 | * was created, we know there is no need to update back refs | |
8424 | * for the subtree | |
8425 | */ | |
8426 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8427 | generation <= root->root_key.offset) { |
8428 | *lookup_info = 1; | |
1c4850e2 | 8429 | return 1; |
94fcca9f | 8430 | } |
1c4850e2 YZ |
8431 | |
8432 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
707e8a07 | 8433 | blocksize = root->nodesize; |
1c4850e2 | 8434 | |
01d58472 | 8435 | next = btrfs_find_tree_block(root->fs_info, bytenr); |
1c4850e2 | 8436 | if (!next) { |
a83fffb7 | 8437 | next = btrfs_find_create_tree_block(root, bytenr); |
90d2c51d MX |
8438 | if (!next) |
8439 | return -ENOMEM; | |
b2aaaa3b JB |
8440 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8441 | level - 1); | |
1c4850e2 YZ |
8442 | reada = 1; |
8443 | } | |
8444 | btrfs_tree_lock(next); | |
8445 | btrfs_set_lock_blocking(next); | |
8446 | ||
3173a18f | 8447 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, |
94fcca9f YZ |
8448 | &wc->refs[level - 1], |
8449 | &wc->flags[level - 1]); | |
79787eaa JM |
8450 | if (ret < 0) { |
8451 | btrfs_tree_unlock(next); | |
8452 | return ret; | |
8453 | } | |
8454 | ||
c2cf52eb SK |
8455 | if (unlikely(wc->refs[level - 1] == 0)) { |
8456 | btrfs_err(root->fs_info, "Missing references."); | |
8457 | BUG(); | |
8458 | } | |
94fcca9f | 8459 | *lookup_info = 0; |
1c4850e2 | 8460 | |
94fcca9f | 8461 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8462 | if (wc->refs[level - 1] > 1) { |
1152651a | 8463 | need_account = true; |
94fcca9f YZ |
8464 | if (level == 1 && |
8465 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8466 | goto skip; | |
8467 | ||
1c4850e2 YZ |
8468 | if (!wc->update_ref || |
8469 | generation <= root->root_key.offset) | |
8470 | goto skip; | |
8471 | ||
8472 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8473 | path->slots[level]); | |
8474 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8475 | if (ret < 0) | |
8476 | goto skip; | |
8477 | ||
8478 | wc->stage = UPDATE_BACKREF; | |
8479 | wc->shared_level = level - 1; | |
8480 | } | |
94fcca9f YZ |
8481 | } else { |
8482 | if (level == 1 && | |
8483 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8484 | goto skip; | |
1c4850e2 YZ |
8485 | } |
8486 | ||
b9fab919 | 8487 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8488 | btrfs_tree_unlock(next); |
8489 | free_extent_buffer(next); | |
8490 | next = NULL; | |
94fcca9f | 8491 | *lookup_info = 1; |
1c4850e2 YZ |
8492 | } |
8493 | ||
8494 | if (!next) { | |
8495 | if (reada && level == 1) | |
8496 | reada_walk_down(trans, root, wc, path); | |
ce86cd59 | 8497 | next = read_tree_block(root, bytenr, generation); |
64c043de LB |
8498 | if (IS_ERR(next)) { |
8499 | return PTR_ERR(next); | |
8500 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8501 | free_extent_buffer(next); |
97d9a8a4 | 8502 | return -EIO; |
416bc658 | 8503 | } |
1c4850e2 YZ |
8504 | btrfs_tree_lock(next); |
8505 | btrfs_set_lock_blocking(next); | |
8506 | } | |
8507 | ||
8508 | level--; | |
8509 | BUG_ON(level != btrfs_header_level(next)); | |
8510 | path->nodes[level] = next; | |
8511 | path->slots[level] = 0; | |
bd681513 | 8512 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8513 | wc->level = level; |
8514 | if (wc->level == 1) | |
8515 | wc->reada_slot = 0; | |
8516 | return 0; | |
8517 | skip: | |
8518 | wc->refs[level - 1] = 0; | |
8519 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8520 | if (wc->stage == DROP_REFERENCE) { |
8521 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8522 | parent = path->nodes[level]->start; | |
8523 | } else { | |
8524 | BUG_ON(root->root_key.objectid != | |
8525 | btrfs_header_owner(path->nodes[level])); | |
8526 | parent = 0; | |
8527 | } | |
1c4850e2 | 8528 | |
1152651a MF |
8529 | if (need_account) { |
8530 | ret = account_shared_subtree(trans, root, next, | |
8531 | generation, level - 1); | |
8532 | if (ret) { | |
94647322 DS |
8533 | btrfs_err_rl(root->fs_info, |
8534 | "Error " | |
1152651a | 8535 | "%d accounting shared subtree. Quota " |
94647322 DS |
8536 | "is out of sync, rescan required.", |
8537 | ret); | |
1152651a MF |
8538 | } |
8539 | } | |
94fcca9f | 8540 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
b06c4bf5 | 8541 | root->root_key.objectid, level - 1, 0); |
79787eaa | 8542 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 8543 | } |
1c4850e2 YZ |
8544 | btrfs_tree_unlock(next); |
8545 | free_extent_buffer(next); | |
94fcca9f | 8546 | *lookup_info = 1; |
1c4850e2 YZ |
8547 | return 1; |
8548 | } | |
8549 | ||
2c47e605 | 8550 | /* |
2c016dc2 | 8551 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8552 | * |
8553 | * when wc->stage == DROP_REFERENCE, this function drops | |
8554 | * reference count on the block. | |
8555 | * | |
8556 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8557 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8558 | * to UPDATE_BACKREF previously while processing the block. | |
8559 | * | |
8560 | * NOTE: return value 1 means we should stop walking up. | |
8561 | */ | |
8562 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8563 | struct btrfs_root *root, | |
8564 | struct btrfs_path *path, | |
8565 | struct walk_control *wc) | |
8566 | { | |
f0486c68 | 8567 | int ret; |
2c47e605 YZ |
8568 | int level = wc->level; |
8569 | struct extent_buffer *eb = path->nodes[level]; | |
8570 | u64 parent = 0; | |
8571 | ||
8572 | if (wc->stage == UPDATE_BACKREF) { | |
8573 | BUG_ON(wc->shared_level < level); | |
8574 | if (level < wc->shared_level) | |
8575 | goto out; | |
8576 | ||
2c47e605 YZ |
8577 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8578 | if (ret > 0) | |
8579 | wc->update_ref = 0; | |
8580 | ||
8581 | wc->stage = DROP_REFERENCE; | |
8582 | wc->shared_level = -1; | |
8583 | path->slots[level] = 0; | |
8584 | ||
8585 | /* | |
8586 | * check reference count again if the block isn't locked. | |
8587 | * we should start walking down the tree again if reference | |
8588 | * count is one. | |
8589 | */ | |
8590 | if (!path->locks[level]) { | |
8591 | BUG_ON(level == 0); | |
8592 | btrfs_tree_lock(eb); | |
8593 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8594 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8595 | |
8596 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 8597 | eb->start, level, 1, |
2c47e605 YZ |
8598 | &wc->refs[level], |
8599 | &wc->flags[level]); | |
79787eaa JM |
8600 | if (ret < 0) { |
8601 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8602 | path->locks[level] = 0; |
79787eaa JM |
8603 | return ret; |
8604 | } | |
2c47e605 YZ |
8605 | BUG_ON(wc->refs[level] == 0); |
8606 | if (wc->refs[level] == 1) { | |
bd681513 | 8607 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8608 | path->locks[level] = 0; |
2c47e605 YZ |
8609 | return 1; |
8610 | } | |
f82d02d9 | 8611 | } |
2c47e605 | 8612 | } |
f82d02d9 | 8613 | |
2c47e605 YZ |
8614 | /* wc->stage == DROP_REFERENCE */ |
8615 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8616 | |
2c47e605 YZ |
8617 | if (wc->refs[level] == 1) { |
8618 | if (level == 0) { | |
8619 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8620 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8621 | else |
e339a6b0 | 8622 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8623 | BUG_ON(ret); /* -ENOMEM */ |
1152651a MF |
8624 | ret = account_leaf_items(trans, root, eb); |
8625 | if (ret) { | |
94647322 DS |
8626 | btrfs_err_rl(root->fs_info, |
8627 | "error " | |
1152651a | 8628 | "%d accounting leaf items. Quota " |
94647322 DS |
8629 | "is out of sync, rescan required.", |
8630 | ret); | |
1152651a | 8631 | } |
2c47e605 YZ |
8632 | } |
8633 | /* make block locked assertion in clean_tree_block happy */ | |
8634 | if (!path->locks[level] && | |
8635 | btrfs_header_generation(eb) == trans->transid) { | |
8636 | btrfs_tree_lock(eb); | |
8637 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8638 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8639 | } |
01d58472 | 8640 | clean_tree_block(trans, root->fs_info, eb); |
2c47e605 YZ |
8641 | } |
8642 | ||
8643 | if (eb == root->node) { | |
8644 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8645 | parent = eb->start; | |
8646 | else | |
8647 | BUG_ON(root->root_key.objectid != | |
8648 | btrfs_header_owner(eb)); | |
8649 | } else { | |
8650 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8651 | parent = path->nodes[level + 1]->start; | |
8652 | else | |
8653 | BUG_ON(root->root_key.objectid != | |
8654 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8655 | } |
f82d02d9 | 8656 | |
5581a51a | 8657 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8658 | out: |
8659 | wc->refs[level] = 0; | |
8660 | wc->flags[level] = 0; | |
f0486c68 | 8661 | return 0; |
2c47e605 YZ |
8662 | } |
8663 | ||
8664 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8665 | struct btrfs_root *root, | |
8666 | struct btrfs_path *path, | |
8667 | struct walk_control *wc) | |
8668 | { | |
2c47e605 | 8669 | int level = wc->level; |
94fcca9f | 8670 | int lookup_info = 1; |
2c47e605 YZ |
8671 | int ret; |
8672 | ||
8673 | while (level >= 0) { | |
94fcca9f | 8674 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8675 | if (ret > 0) |
8676 | break; | |
8677 | ||
8678 | if (level == 0) | |
8679 | break; | |
8680 | ||
7a7965f8 YZ |
8681 | if (path->slots[level] >= |
8682 | btrfs_header_nritems(path->nodes[level])) | |
8683 | break; | |
8684 | ||
94fcca9f | 8685 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8686 | if (ret > 0) { |
8687 | path->slots[level]++; | |
8688 | continue; | |
90d2c51d MX |
8689 | } else if (ret < 0) |
8690 | return ret; | |
1c4850e2 | 8691 | level = wc->level; |
f82d02d9 | 8692 | } |
f82d02d9 YZ |
8693 | return 0; |
8694 | } | |
8695 | ||
d397712b | 8696 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 8697 | struct btrfs_root *root, |
f82d02d9 | 8698 | struct btrfs_path *path, |
2c47e605 | 8699 | struct walk_control *wc, int max_level) |
20524f02 | 8700 | { |
2c47e605 | 8701 | int level = wc->level; |
20524f02 | 8702 | int ret; |
9f3a7427 | 8703 | |
2c47e605 YZ |
8704 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
8705 | while (level < max_level && path->nodes[level]) { | |
8706 | wc->level = level; | |
8707 | if (path->slots[level] + 1 < | |
8708 | btrfs_header_nritems(path->nodes[level])) { | |
8709 | path->slots[level]++; | |
20524f02 CM |
8710 | return 0; |
8711 | } else { | |
2c47e605 YZ |
8712 | ret = walk_up_proc(trans, root, path, wc); |
8713 | if (ret > 0) | |
8714 | return 0; | |
bd56b302 | 8715 | |
2c47e605 | 8716 | if (path->locks[level]) { |
bd681513 CM |
8717 | btrfs_tree_unlock_rw(path->nodes[level], |
8718 | path->locks[level]); | |
2c47e605 | 8719 | path->locks[level] = 0; |
f82d02d9 | 8720 | } |
2c47e605 YZ |
8721 | free_extent_buffer(path->nodes[level]); |
8722 | path->nodes[level] = NULL; | |
8723 | level++; | |
20524f02 CM |
8724 | } |
8725 | } | |
8726 | return 1; | |
8727 | } | |
8728 | ||
9aca1d51 | 8729 | /* |
2c47e605 YZ |
8730 | * drop a subvolume tree. |
8731 | * | |
8732 | * this function traverses the tree freeing any blocks that only | |
8733 | * referenced by the tree. | |
8734 | * | |
8735 | * when a shared tree block is found. this function decreases its | |
8736 | * reference count by one. if update_ref is true, this function | |
8737 | * also make sure backrefs for the shared block and all lower level | |
8738 | * blocks are properly updated. | |
9d1a2a3a DS |
8739 | * |
8740 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 8741 | */ |
2c536799 | 8742 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
8743 | struct btrfs_block_rsv *block_rsv, int update_ref, |
8744 | int for_reloc) | |
20524f02 | 8745 | { |
5caf2a00 | 8746 | struct btrfs_path *path; |
2c47e605 YZ |
8747 | struct btrfs_trans_handle *trans; |
8748 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 8749 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
8750 | struct walk_control *wc; |
8751 | struct btrfs_key key; | |
8752 | int err = 0; | |
8753 | int ret; | |
8754 | int level; | |
d29a9f62 | 8755 | bool root_dropped = false; |
20524f02 | 8756 | |
1152651a MF |
8757 | btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid); |
8758 | ||
5caf2a00 | 8759 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
8760 | if (!path) { |
8761 | err = -ENOMEM; | |
8762 | goto out; | |
8763 | } | |
20524f02 | 8764 | |
2c47e605 | 8765 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
8766 | if (!wc) { |
8767 | btrfs_free_path(path); | |
cb1b69f4 TI |
8768 | err = -ENOMEM; |
8769 | goto out; | |
38a1a919 | 8770 | } |
2c47e605 | 8771 | |
a22285a6 | 8772 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
8773 | if (IS_ERR(trans)) { |
8774 | err = PTR_ERR(trans); | |
8775 | goto out_free; | |
8776 | } | |
98d5dc13 | 8777 | |
3fd0a558 YZ |
8778 | if (block_rsv) |
8779 | trans->block_rsv = block_rsv; | |
2c47e605 | 8780 | |
9f3a7427 | 8781 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 8782 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
8783 | path->nodes[level] = btrfs_lock_root_node(root); |
8784 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 8785 | path->slots[level] = 0; |
bd681513 | 8786 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8787 | memset(&wc->update_progress, 0, |
8788 | sizeof(wc->update_progress)); | |
9f3a7427 | 8789 | } else { |
9f3a7427 | 8790 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
8791 | memcpy(&wc->update_progress, &key, |
8792 | sizeof(wc->update_progress)); | |
8793 | ||
6702ed49 | 8794 | level = root_item->drop_level; |
2c47e605 | 8795 | BUG_ON(level == 0); |
6702ed49 | 8796 | path->lowest_level = level; |
2c47e605 YZ |
8797 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
8798 | path->lowest_level = 0; | |
8799 | if (ret < 0) { | |
8800 | err = ret; | |
79787eaa | 8801 | goto out_end_trans; |
9f3a7427 | 8802 | } |
1c4850e2 | 8803 | WARN_ON(ret > 0); |
2c47e605 | 8804 | |
7d9eb12c CM |
8805 | /* |
8806 | * unlock our path, this is safe because only this | |
8807 | * function is allowed to delete this snapshot | |
8808 | */ | |
5d4f98a2 | 8809 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
8810 | |
8811 | level = btrfs_header_level(root->node); | |
8812 | while (1) { | |
8813 | btrfs_tree_lock(path->nodes[level]); | |
8814 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 8815 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8816 | |
8817 | ret = btrfs_lookup_extent_info(trans, root, | |
8818 | path->nodes[level]->start, | |
3173a18f | 8819 | level, 1, &wc->refs[level], |
2c47e605 | 8820 | &wc->flags[level]); |
79787eaa JM |
8821 | if (ret < 0) { |
8822 | err = ret; | |
8823 | goto out_end_trans; | |
8824 | } | |
2c47e605 YZ |
8825 | BUG_ON(wc->refs[level] == 0); |
8826 | ||
8827 | if (level == root_item->drop_level) | |
8828 | break; | |
8829 | ||
8830 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 8831 | path->locks[level] = 0; |
2c47e605 YZ |
8832 | WARN_ON(wc->refs[level] != 1); |
8833 | level--; | |
8834 | } | |
9f3a7427 | 8835 | } |
2c47e605 YZ |
8836 | |
8837 | wc->level = level; | |
8838 | wc->shared_level = -1; | |
8839 | wc->stage = DROP_REFERENCE; | |
8840 | wc->update_ref = update_ref; | |
8841 | wc->keep_locks = 0; | |
66d7e7f0 | 8842 | wc->for_reloc = for_reloc; |
1c4850e2 | 8843 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 8844 | |
d397712b | 8845 | while (1) { |
9d1a2a3a | 8846 | |
2c47e605 YZ |
8847 | ret = walk_down_tree(trans, root, path, wc); |
8848 | if (ret < 0) { | |
8849 | err = ret; | |
20524f02 | 8850 | break; |
2c47e605 | 8851 | } |
9aca1d51 | 8852 | |
2c47e605 YZ |
8853 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
8854 | if (ret < 0) { | |
8855 | err = ret; | |
20524f02 | 8856 | break; |
2c47e605 YZ |
8857 | } |
8858 | ||
8859 | if (ret > 0) { | |
8860 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
8861 | break; |
8862 | } | |
2c47e605 YZ |
8863 | |
8864 | if (wc->stage == DROP_REFERENCE) { | |
8865 | level = wc->level; | |
8866 | btrfs_node_key(path->nodes[level], | |
8867 | &root_item->drop_progress, | |
8868 | path->slots[level]); | |
8869 | root_item->drop_level = level; | |
8870 | } | |
8871 | ||
8872 | BUG_ON(wc->level == 0); | |
3c8f2422 JB |
8873 | if (btrfs_should_end_transaction(trans, tree_root) || |
8874 | (!for_reloc && btrfs_need_cleaner_sleep(root))) { | |
2c47e605 YZ |
8875 | ret = btrfs_update_root(trans, tree_root, |
8876 | &root->root_key, | |
8877 | root_item); | |
79787eaa JM |
8878 | if (ret) { |
8879 | btrfs_abort_transaction(trans, tree_root, ret); | |
8880 | err = ret; | |
8881 | goto out_end_trans; | |
8882 | } | |
2c47e605 | 8883 | |
3fd0a558 | 8884 | btrfs_end_transaction_throttle(trans, tree_root); |
3c8f2422 | 8885 | if (!for_reloc && btrfs_need_cleaner_sleep(root)) { |
efe120a0 | 8886 | pr_debug("BTRFS: drop snapshot early exit\n"); |
3c8f2422 JB |
8887 | err = -EAGAIN; |
8888 | goto out_free; | |
8889 | } | |
8890 | ||
a22285a6 | 8891 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
8892 | if (IS_ERR(trans)) { |
8893 | err = PTR_ERR(trans); | |
8894 | goto out_free; | |
8895 | } | |
3fd0a558 YZ |
8896 | if (block_rsv) |
8897 | trans->block_rsv = block_rsv; | |
c3e69d58 | 8898 | } |
20524f02 | 8899 | } |
b3b4aa74 | 8900 | btrfs_release_path(path); |
79787eaa JM |
8901 | if (err) |
8902 | goto out_end_trans; | |
2c47e605 YZ |
8903 | |
8904 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
8905 | if (ret) { |
8906 | btrfs_abort_transaction(trans, tree_root, ret); | |
8907 | goto out_end_trans; | |
8908 | } | |
2c47e605 | 8909 | |
76dda93c | 8910 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
8911 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
8912 | NULL, NULL); | |
79787eaa JM |
8913 | if (ret < 0) { |
8914 | btrfs_abort_transaction(trans, tree_root, ret); | |
8915 | err = ret; | |
8916 | goto out_end_trans; | |
8917 | } else if (ret > 0) { | |
84cd948c JB |
8918 | /* if we fail to delete the orphan item this time |
8919 | * around, it'll get picked up the next time. | |
8920 | * | |
8921 | * The most common failure here is just -ENOENT. | |
8922 | */ | |
8923 | btrfs_del_orphan_item(trans, tree_root, | |
8924 | root->root_key.objectid); | |
76dda93c YZ |
8925 | } |
8926 | } | |
8927 | ||
27cdeb70 | 8928 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 8929 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
8930 | } else { |
8931 | free_extent_buffer(root->node); | |
8932 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 8933 | btrfs_put_fs_root(root); |
76dda93c | 8934 | } |
d29a9f62 | 8935 | root_dropped = true; |
79787eaa | 8936 | out_end_trans: |
3fd0a558 | 8937 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 8938 | out_free: |
2c47e605 | 8939 | kfree(wc); |
5caf2a00 | 8940 | btrfs_free_path(path); |
cb1b69f4 | 8941 | out: |
d29a9f62 JB |
8942 | /* |
8943 | * So if we need to stop dropping the snapshot for whatever reason we | |
8944 | * need to make sure to add it back to the dead root list so that we | |
8945 | * keep trying to do the work later. This also cleans up roots if we | |
8946 | * don't have it in the radix (like when we recover after a power fail | |
8947 | * or unmount) so we don't leak memory. | |
8948 | */ | |
b37b39cd | 8949 | if (!for_reloc && root_dropped == false) |
d29a9f62 | 8950 | btrfs_add_dead_root(root); |
90515e7f | 8951 | if (err && err != -EAGAIN) |
a4553fef | 8952 | btrfs_std_error(root->fs_info, err, NULL); |
2c536799 | 8953 | return err; |
20524f02 | 8954 | } |
9078a3e1 | 8955 | |
2c47e605 YZ |
8956 | /* |
8957 | * drop subtree rooted at tree block 'node'. | |
8958 | * | |
8959 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 8960 | * only used by relocation code |
2c47e605 | 8961 | */ |
f82d02d9 YZ |
8962 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
8963 | struct btrfs_root *root, | |
8964 | struct extent_buffer *node, | |
8965 | struct extent_buffer *parent) | |
8966 | { | |
8967 | struct btrfs_path *path; | |
2c47e605 | 8968 | struct walk_control *wc; |
f82d02d9 YZ |
8969 | int level; |
8970 | int parent_level; | |
8971 | int ret = 0; | |
8972 | int wret; | |
8973 | ||
2c47e605 YZ |
8974 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
8975 | ||
f82d02d9 | 8976 | path = btrfs_alloc_path(); |
db5b493a TI |
8977 | if (!path) |
8978 | return -ENOMEM; | |
f82d02d9 | 8979 | |
2c47e605 | 8980 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
8981 | if (!wc) { |
8982 | btrfs_free_path(path); | |
8983 | return -ENOMEM; | |
8984 | } | |
2c47e605 | 8985 | |
b9447ef8 | 8986 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
8987 | parent_level = btrfs_header_level(parent); |
8988 | extent_buffer_get(parent); | |
8989 | path->nodes[parent_level] = parent; | |
8990 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
8991 | ||
b9447ef8 | 8992 | btrfs_assert_tree_locked(node); |
f82d02d9 | 8993 | level = btrfs_header_level(node); |
f82d02d9 YZ |
8994 | path->nodes[level] = node; |
8995 | path->slots[level] = 0; | |
bd681513 | 8996 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8997 | |
8998 | wc->refs[parent_level] = 1; | |
8999 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9000 | wc->level = level; | |
9001 | wc->shared_level = -1; | |
9002 | wc->stage = DROP_REFERENCE; | |
9003 | wc->update_ref = 0; | |
9004 | wc->keep_locks = 1; | |
66d7e7f0 | 9005 | wc->for_reloc = 1; |
1c4850e2 | 9006 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
9007 | |
9008 | while (1) { | |
2c47e605 YZ |
9009 | wret = walk_down_tree(trans, root, path, wc); |
9010 | if (wret < 0) { | |
f82d02d9 | 9011 | ret = wret; |
f82d02d9 | 9012 | break; |
2c47e605 | 9013 | } |
f82d02d9 | 9014 | |
2c47e605 | 9015 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9016 | if (wret < 0) |
9017 | ret = wret; | |
9018 | if (wret != 0) | |
9019 | break; | |
9020 | } | |
9021 | ||
2c47e605 | 9022 | kfree(wc); |
f82d02d9 YZ |
9023 | btrfs_free_path(path); |
9024 | return ret; | |
9025 | } | |
9026 | ||
ec44a35c CM |
9027 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
9028 | { | |
9029 | u64 num_devices; | |
fc67c450 | 9030 | u64 stripped; |
e4d8ec0f | 9031 | |
fc67c450 ID |
9032 | /* |
9033 | * if restripe for this chunk_type is on pick target profile and | |
9034 | * return, otherwise do the usual balance | |
9035 | */ | |
9036 | stripped = get_restripe_target(root->fs_info, flags); | |
9037 | if (stripped) | |
9038 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9039 | |
95669976 | 9040 | num_devices = root->fs_info->fs_devices->rw_devices; |
cd02dca5 | 9041 | |
fc67c450 | 9042 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9043 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9044 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9045 | ||
ec44a35c CM |
9046 | if (num_devices == 1) { |
9047 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9048 | stripped = flags & ~stripped; | |
9049 | ||
9050 | /* turn raid0 into single device chunks */ | |
9051 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9052 | return stripped; | |
9053 | ||
9054 | /* turn mirroring into duplication */ | |
9055 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9056 | BTRFS_BLOCK_GROUP_RAID10)) | |
9057 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9058 | } else { |
9059 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9060 | if (flags & stripped) |
9061 | return flags; | |
9062 | ||
9063 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9064 | stripped = flags & ~stripped; | |
9065 | ||
9066 | /* switch duplicated blocks with raid1 */ | |
9067 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9068 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9069 | ||
e3176ca2 | 9070 | /* this is drive concat, leave it alone */ |
ec44a35c | 9071 | } |
e3176ca2 | 9072 | |
ec44a35c CM |
9073 | return flags; |
9074 | } | |
9075 | ||
868f401a | 9076 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9077 | { |
f0486c68 YZ |
9078 | struct btrfs_space_info *sinfo = cache->space_info; |
9079 | u64 num_bytes; | |
199c36ea | 9080 | u64 min_allocable_bytes; |
f0486c68 | 9081 | int ret = -ENOSPC; |
0ef3e66b | 9082 | |
199c36ea MX |
9083 | /* |
9084 | * We need some metadata space and system metadata space for | |
9085 | * allocating chunks in some corner cases until we force to set | |
9086 | * it to be readonly. | |
9087 | */ | |
9088 | if ((sinfo->flags & | |
9089 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9090 | !force) | |
9091 | min_allocable_bytes = 1 * 1024 * 1024; | |
9092 | else | |
9093 | min_allocable_bytes = 0; | |
9094 | ||
f0486c68 YZ |
9095 | spin_lock(&sinfo->lock); |
9096 | spin_lock(&cache->lock); | |
61cfea9b W |
9097 | |
9098 | if (cache->ro) { | |
868f401a | 9099 | cache->ro++; |
61cfea9b W |
9100 | ret = 0; |
9101 | goto out; | |
9102 | } | |
9103 | ||
f0486c68 YZ |
9104 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9105 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9106 | ||
9107 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
9108 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
9109 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 9110 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9111 | cache->ro++; |
633c0aad | 9112 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9113 | ret = 0; |
9114 | } | |
61cfea9b | 9115 | out: |
f0486c68 YZ |
9116 | spin_unlock(&cache->lock); |
9117 | spin_unlock(&sinfo->lock); | |
9118 | return ret; | |
9119 | } | |
7d9eb12c | 9120 | |
868f401a | 9121 | int btrfs_inc_block_group_ro(struct btrfs_root *root, |
f0486c68 | 9122 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9123 | |
f0486c68 YZ |
9124 | { |
9125 | struct btrfs_trans_handle *trans; | |
9126 | u64 alloc_flags; | |
9127 | int ret; | |
7d9eb12c | 9128 | |
1bbc621e | 9129 | again: |
ff5714cc | 9130 | trans = btrfs_join_transaction(root); |
79787eaa JM |
9131 | if (IS_ERR(trans)) |
9132 | return PTR_ERR(trans); | |
5d4f98a2 | 9133 | |
1bbc621e CM |
9134 | /* |
9135 | * we're not allowed to set block groups readonly after the dirty | |
9136 | * block groups cache has started writing. If it already started, | |
9137 | * back off and let this transaction commit | |
9138 | */ | |
9139 | mutex_lock(&root->fs_info->ro_block_group_mutex); | |
3204d33c | 9140 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9141 | u64 transid = trans->transid; |
9142 | ||
9143 | mutex_unlock(&root->fs_info->ro_block_group_mutex); | |
9144 | btrfs_end_transaction(trans, root); | |
9145 | ||
9146 | ret = btrfs_wait_for_commit(root, transid); | |
9147 | if (ret) | |
9148 | return ret; | |
9149 | goto again; | |
9150 | } | |
9151 | ||
153c35b6 CM |
9152 | /* |
9153 | * if we are changing raid levels, try to allocate a corresponding | |
9154 | * block group with the new raid level. | |
9155 | */ | |
9156 | alloc_flags = update_block_group_flags(root, cache->flags); | |
9157 | if (alloc_flags != cache->flags) { | |
9158 | ret = do_chunk_alloc(trans, root, alloc_flags, | |
9159 | CHUNK_ALLOC_FORCE); | |
9160 | /* | |
9161 | * ENOSPC is allowed here, we may have enough space | |
9162 | * already allocated at the new raid level to | |
9163 | * carry on | |
9164 | */ | |
9165 | if (ret == -ENOSPC) | |
9166 | ret = 0; | |
9167 | if (ret < 0) | |
9168 | goto out; | |
9169 | } | |
1bbc621e | 9170 | |
868f401a | 9171 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9172 | if (!ret) |
9173 | goto out; | |
9174 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 9175 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 9176 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9177 | if (ret < 0) |
9178 | goto out; | |
868f401a | 9179 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9180 | out: |
2f081088 SL |
9181 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
9182 | alloc_flags = update_block_group_flags(root, cache->flags); | |
a9629596 | 9183 | lock_chunks(root->fs_info->chunk_root); |
4617ea3a | 9184 | check_system_chunk(trans, root, alloc_flags); |
a9629596 | 9185 | unlock_chunks(root->fs_info->chunk_root); |
2f081088 | 9186 | } |
1bbc621e | 9187 | mutex_unlock(&root->fs_info->ro_block_group_mutex); |
2f081088 | 9188 | |
f0486c68 YZ |
9189 | btrfs_end_transaction(trans, root); |
9190 | return ret; | |
9191 | } | |
5d4f98a2 | 9192 | |
c87f08ca CM |
9193 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
9194 | struct btrfs_root *root, u64 type) | |
9195 | { | |
9196 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 9197 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 9198 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9199 | } |
9200 | ||
6d07bcec MX |
9201 | /* |
9202 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9203 | * space_info. takes mirrors into account. |
6d07bcec | 9204 | */ |
633c0aad | 9205 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9206 | { |
9207 | struct btrfs_block_group_cache *block_group; | |
9208 | u64 free_bytes = 0; | |
9209 | int factor; | |
9210 | ||
633c0aad JB |
9211 | /* It's df, we don't care if it's racey */ |
9212 | if (list_empty(&sinfo->ro_bgs)) | |
9213 | return 0; | |
9214 | ||
9215 | spin_lock(&sinfo->lock); | |
9216 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9217 | spin_lock(&block_group->lock); |
9218 | ||
9219 | if (!block_group->ro) { | |
9220 | spin_unlock(&block_group->lock); | |
9221 | continue; | |
9222 | } | |
9223 | ||
9224 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9225 | BTRFS_BLOCK_GROUP_RAID10 | | |
9226 | BTRFS_BLOCK_GROUP_DUP)) | |
9227 | factor = 2; | |
9228 | else | |
9229 | factor = 1; | |
9230 | ||
9231 | free_bytes += (block_group->key.offset - | |
9232 | btrfs_block_group_used(&block_group->item)) * | |
9233 | factor; | |
9234 | ||
9235 | spin_unlock(&block_group->lock); | |
9236 | } | |
6d07bcec MX |
9237 | spin_unlock(&sinfo->lock); |
9238 | ||
9239 | return free_bytes; | |
9240 | } | |
9241 | ||
868f401a | 9242 | void btrfs_dec_block_group_ro(struct btrfs_root *root, |
f0486c68 | 9243 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9244 | { |
f0486c68 YZ |
9245 | struct btrfs_space_info *sinfo = cache->space_info; |
9246 | u64 num_bytes; | |
9247 | ||
9248 | BUG_ON(!cache->ro); | |
9249 | ||
9250 | spin_lock(&sinfo->lock); | |
9251 | spin_lock(&cache->lock); | |
868f401a Z |
9252 | if (!--cache->ro) { |
9253 | num_bytes = cache->key.offset - cache->reserved - | |
9254 | cache->pinned - cache->bytes_super - | |
9255 | btrfs_block_group_used(&cache->item); | |
9256 | sinfo->bytes_readonly -= num_bytes; | |
9257 | list_del_init(&cache->ro_list); | |
9258 | } | |
f0486c68 YZ |
9259 | spin_unlock(&cache->lock); |
9260 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9261 | } |
9262 | ||
ba1bf481 JB |
9263 | /* |
9264 | * checks to see if its even possible to relocate this block group. | |
9265 | * | |
9266 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9267 | * ok to go ahead and try. | |
9268 | */ | |
9269 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 9270 | { |
ba1bf481 JB |
9271 | struct btrfs_block_group_cache *block_group; |
9272 | struct btrfs_space_info *space_info; | |
9273 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
9274 | struct btrfs_device *device; | |
6df9a95e | 9275 | struct btrfs_trans_handle *trans; |
cdcb725c | 9276 | u64 min_free; |
6719db6a JB |
9277 | u64 dev_min = 1; |
9278 | u64 dev_nr = 0; | |
4a5e98f5 | 9279 | u64 target; |
cdcb725c | 9280 | int index; |
ba1bf481 JB |
9281 | int full = 0; |
9282 | int ret = 0; | |
1a40e23b | 9283 | |
ba1bf481 | 9284 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 9285 | |
ba1bf481 JB |
9286 | /* odd, couldn't find the block group, leave it alone */ |
9287 | if (!block_group) | |
9288 | return -1; | |
1a40e23b | 9289 | |
cdcb725c | 9290 | min_free = btrfs_block_group_used(&block_group->item); |
9291 | ||
ba1bf481 | 9292 | /* no bytes used, we're good */ |
cdcb725c | 9293 | if (!min_free) |
1a40e23b ZY |
9294 | goto out; |
9295 | ||
ba1bf481 JB |
9296 | space_info = block_group->space_info; |
9297 | spin_lock(&space_info->lock); | |
17d217fe | 9298 | |
ba1bf481 | 9299 | full = space_info->full; |
17d217fe | 9300 | |
ba1bf481 JB |
9301 | /* |
9302 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9303 | * relocate it unless we're able to allocate a new chunk below. |
9304 | * | |
9305 | * Otherwise, we need to make sure we have room in the space to handle | |
9306 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9307 | */ |
7ce618db | 9308 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 9309 | (space_info->bytes_used + space_info->bytes_reserved + |
9310 | space_info->bytes_pinned + space_info->bytes_readonly + | |
9311 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
9312 | spin_unlock(&space_info->lock); |
9313 | goto out; | |
17d217fe | 9314 | } |
ba1bf481 | 9315 | spin_unlock(&space_info->lock); |
ea8c2819 | 9316 | |
ba1bf481 JB |
9317 | /* |
9318 | * ok we don't have enough space, but maybe we have free space on our | |
9319 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9320 | * alloc devices and guess if we have enough space. if this block |
9321 | * group is going to be restriped, run checks against the target | |
9322 | * profile instead of the current one. | |
ba1bf481 JB |
9323 | */ |
9324 | ret = -1; | |
ea8c2819 | 9325 | |
cdcb725c | 9326 | /* |
9327 | * index: | |
9328 | * 0: raid10 | |
9329 | * 1: raid1 | |
9330 | * 2: dup | |
9331 | * 3: raid0 | |
9332 | * 4: single | |
9333 | */ | |
4a5e98f5 ID |
9334 | target = get_restripe_target(root->fs_info, block_group->flags); |
9335 | if (target) { | |
31e50229 | 9336 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9337 | } else { |
9338 | /* | |
9339 | * this is just a balance, so if we were marked as full | |
9340 | * we know there is no space for a new chunk | |
9341 | */ | |
9342 | if (full) | |
9343 | goto out; | |
9344 | ||
9345 | index = get_block_group_index(block_group); | |
9346 | } | |
9347 | ||
e6ec716f | 9348 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9349 | dev_min = 4; |
6719db6a JB |
9350 | /* Divide by 2 */ |
9351 | min_free >>= 1; | |
e6ec716f | 9352 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9353 | dev_min = 2; |
e6ec716f | 9354 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9355 | /* Multiply by 2 */ |
9356 | min_free <<= 1; | |
e6ec716f | 9357 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9358 | dev_min = fs_devices->rw_devices; |
47c5713f | 9359 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9360 | } |
9361 | ||
6df9a95e JB |
9362 | /* We need to do this so that we can look at pending chunks */ |
9363 | trans = btrfs_join_transaction(root); | |
9364 | if (IS_ERR(trans)) { | |
9365 | ret = PTR_ERR(trans); | |
9366 | goto out; | |
9367 | } | |
9368 | ||
ba1bf481 JB |
9369 | mutex_lock(&root->fs_info->chunk_mutex); |
9370 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 9371 | u64 dev_offset; |
56bec294 | 9372 | |
ba1bf481 JB |
9373 | /* |
9374 | * check to make sure we can actually find a chunk with enough | |
9375 | * space to fit our block group in. | |
9376 | */ | |
63a212ab SB |
9377 | if (device->total_bytes > device->bytes_used + min_free && |
9378 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 9379 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9380 | &dev_offset, NULL); |
ba1bf481 | 9381 | if (!ret) |
cdcb725c | 9382 | dev_nr++; |
9383 | ||
9384 | if (dev_nr >= dev_min) | |
73e48b27 | 9385 | break; |
cdcb725c | 9386 | |
ba1bf481 | 9387 | ret = -1; |
725c8463 | 9388 | } |
edbd8d4e | 9389 | } |
ba1bf481 | 9390 | mutex_unlock(&root->fs_info->chunk_mutex); |
6df9a95e | 9391 | btrfs_end_transaction(trans, root); |
edbd8d4e | 9392 | out: |
ba1bf481 | 9393 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9394 | return ret; |
9395 | } | |
9396 | ||
b2950863 CH |
9397 | static int find_first_block_group(struct btrfs_root *root, |
9398 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 9399 | { |
925baedd | 9400 | int ret = 0; |
0b86a832 CM |
9401 | struct btrfs_key found_key; |
9402 | struct extent_buffer *leaf; | |
9403 | int slot; | |
edbd8d4e | 9404 | |
0b86a832 CM |
9405 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9406 | if (ret < 0) | |
925baedd CM |
9407 | goto out; |
9408 | ||
d397712b | 9409 | while (1) { |
0b86a832 | 9410 | slot = path->slots[0]; |
edbd8d4e | 9411 | leaf = path->nodes[0]; |
0b86a832 CM |
9412 | if (slot >= btrfs_header_nritems(leaf)) { |
9413 | ret = btrfs_next_leaf(root, path); | |
9414 | if (ret == 0) | |
9415 | continue; | |
9416 | if (ret < 0) | |
925baedd | 9417 | goto out; |
0b86a832 | 9418 | break; |
edbd8d4e | 9419 | } |
0b86a832 | 9420 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9421 | |
0b86a832 | 9422 | if (found_key.objectid >= key->objectid && |
925baedd CM |
9423 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
9424 | ret = 0; | |
9425 | goto out; | |
9426 | } | |
0b86a832 | 9427 | path->slots[0]++; |
edbd8d4e | 9428 | } |
925baedd | 9429 | out: |
0b86a832 | 9430 | return ret; |
edbd8d4e CM |
9431 | } |
9432 | ||
0af3d00b JB |
9433 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9434 | { | |
9435 | struct btrfs_block_group_cache *block_group; | |
9436 | u64 last = 0; | |
9437 | ||
9438 | while (1) { | |
9439 | struct inode *inode; | |
9440 | ||
9441 | block_group = btrfs_lookup_first_block_group(info, last); | |
9442 | while (block_group) { | |
9443 | spin_lock(&block_group->lock); | |
9444 | if (block_group->iref) | |
9445 | break; | |
9446 | spin_unlock(&block_group->lock); | |
9447 | block_group = next_block_group(info->tree_root, | |
9448 | block_group); | |
9449 | } | |
9450 | if (!block_group) { | |
9451 | if (last == 0) | |
9452 | break; | |
9453 | last = 0; | |
9454 | continue; | |
9455 | } | |
9456 | ||
9457 | inode = block_group->inode; | |
9458 | block_group->iref = 0; | |
9459 | block_group->inode = NULL; | |
9460 | spin_unlock(&block_group->lock); | |
9461 | iput(inode); | |
9462 | last = block_group->key.objectid + block_group->key.offset; | |
9463 | btrfs_put_block_group(block_group); | |
9464 | } | |
9465 | } | |
9466 | ||
1a40e23b ZY |
9467 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9468 | { | |
9469 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9470 | struct btrfs_space_info *space_info; |
11833d66 | 9471 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9472 | struct rb_node *n; |
9473 | ||
9e351cc8 | 9474 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9475 | while (!list_empty(&info->caching_block_groups)) { |
9476 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9477 | struct btrfs_caching_control, list); | |
9478 | list_del(&caching_ctl->list); | |
9479 | put_caching_control(caching_ctl); | |
9480 | } | |
9e351cc8 | 9481 | up_write(&info->commit_root_sem); |
11833d66 | 9482 | |
47ab2a6c JB |
9483 | spin_lock(&info->unused_bgs_lock); |
9484 | while (!list_empty(&info->unused_bgs)) { | |
9485 | block_group = list_first_entry(&info->unused_bgs, | |
9486 | struct btrfs_block_group_cache, | |
9487 | bg_list); | |
9488 | list_del_init(&block_group->bg_list); | |
9489 | btrfs_put_block_group(block_group); | |
9490 | } | |
9491 | spin_unlock(&info->unused_bgs_lock); | |
9492 | ||
1a40e23b ZY |
9493 | spin_lock(&info->block_group_cache_lock); |
9494 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9495 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9496 | cache_node); | |
1a40e23b ZY |
9497 | rb_erase(&block_group->cache_node, |
9498 | &info->block_group_cache_tree); | |
01eacb27 | 9499 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9500 | spin_unlock(&info->block_group_cache_lock); |
9501 | ||
80eb234a | 9502 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9503 | list_del(&block_group->list); |
80eb234a | 9504 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9505 | |
817d52f8 | 9506 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 9507 | wait_block_group_cache_done(block_group); |
817d52f8 | 9508 | |
3c14874a JB |
9509 | /* |
9510 | * We haven't cached this block group, which means we could | |
9511 | * possibly have excluded extents on this block group. | |
9512 | */ | |
36cce922 JB |
9513 | if (block_group->cached == BTRFS_CACHE_NO || |
9514 | block_group->cached == BTRFS_CACHE_ERROR) | |
3c14874a JB |
9515 | free_excluded_extents(info->extent_root, block_group); |
9516 | ||
817d52f8 | 9517 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 9518 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9519 | |
9520 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9521 | } |
9522 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9523 | |
9524 | /* now that all the block groups are freed, go through and | |
9525 | * free all the space_info structs. This is only called during | |
9526 | * the final stages of unmount, and so we know nobody is | |
9527 | * using them. We call synchronize_rcu() once before we start, | |
9528 | * just to be on the safe side. | |
9529 | */ | |
9530 | synchronize_rcu(); | |
9531 | ||
8929ecfa YZ |
9532 | release_global_block_rsv(info); |
9533 | ||
67871254 | 9534 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9535 | int i; |
9536 | ||
4184ea7f CM |
9537 | space_info = list_entry(info->space_info.next, |
9538 | struct btrfs_space_info, | |
9539 | list); | |
b069e0c3 | 9540 | if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) { |
fae7f21c | 9541 | if (WARN_ON(space_info->bytes_pinned > 0 || |
b069e0c3 | 9542 | space_info->bytes_reserved > 0 || |
fae7f21c | 9543 | space_info->bytes_may_use > 0)) { |
b069e0c3 DS |
9544 | dump_space_info(space_info, 0, 0); |
9545 | } | |
f0486c68 | 9546 | } |
4184ea7f | 9547 | list_del(&space_info->list); |
6ab0a202 JM |
9548 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9549 | struct kobject *kobj; | |
c1895442 JM |
9550 | kobj = space_info->block_group_kobjs[i]; |
9551 | space_info->block_group_kobjs[i] = NULL; | |
9552 | if (kobj) { | |
6ab0a202 JM |
9553 | kobject_del(kobj); |
9554 | kobject_put(kobj); | |
9555 | } | |
9556 | } | |
9557 | kobject_del(&space_info->kobj); | |
9558 | kobject_put(&space_info->kobj); | |
4184ea7f | 9559 | } |
1a40e23b ZY |
9560 | return 0; |
9561 | } | |
9562 | ||
b742bb82 YZ |
9563 | static void __link_block_group(struct btrfs_space_info *space_info, |
9564 | struct btrfs_block_group_cache *cache) | |
9565 | { | |
9566 | int index = get_block_group_index(cache); | |
ed55b6ac | 9567 | bool first = false; |
b742bb82 YZ |
9568 | |
9569 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9570 | if (list_empty(&space_info->block_groups[index])) |
9571 | first = true; | |
9572 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9573 | up_write(&space_info->groups_sem); | |
9574 | ||
9575 | if (first) { | |
c1895442 | 9576 | struct raid_kobject *rkobj; |
6ab0a202 JM |
9577 | int ret; |
9578 | ||
c1895442 JM |
9579 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9580 | if (!rkobj) | |
9581 | goto out_err; | |
9582 | rkobj->raid_type = index; | |
9583 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9584 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9585 | "%s", get_raid_name(index)); | |
6ab0a202 | 9586 | if (ret) { |
c1895442 JM |
9587 | kobject_put(&rkobj->kobj); |
9588 | goto out_err; | |
6ab0a202 | 9589 | } |
c1895442 | 9590 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9591 | } |
c1895442 JM |
9592 | |
9593 | return; | |
9594 | out_err: | |
9595 | pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n"); | |
b742bb82 YZ |
9596 | } |
9597 | ||
920e4a58 MX |
9598 | static struct btrfs_block_group_cache * |
9599 | btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size) | |
9600 | { | |
9601 | struct btrfs_block_group_cache *cache; | |
9602 | ||
9603 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9604 | if (!cache) | |
9605 | return NULL; | |
9606 | ||
9607 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9608 | GFP_NOFS); | |
9609 | if (!cache->free_space_ctl) { | |
9610 | kfree(cache); | |
9611 | return NULL; | |
9612 | } | |
9613 | ||
9614 | cache->key.objectid = start; | |
9615 | cache->key.offset = size; | |
9616 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9617 | ||
9618 | cache->sectorsize = root->sectorsize; | |
9619 | cache->fs_info = root->fs_info; | |
9620 | cache->full_stripe_len = btrfs_full_stripe_len(root, | |
9621 | &root->fs_info->mapping_tree, | |
9622 | start); | |
9623 | atomic_set(&cache->count, 1); | |
9624 | spin_lock_init(&cache->lock); | |
e570fd27 | 9625 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
9626 | INIT_LIST_HEAD(&cache->list); |
9627 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 9628 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 9629 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 9630 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 9631 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 9632 | btrfs_init_free_space_ctl(cache); |
04216820 | 9633 | atomic_set(&cache->trimming, 0); |
920e4a58 MX |
9634 | |
9635 | return cache; | |
9636 | } | |
9637 | ||
9078a3e1 CM |
9638 | int btrfs_read_block_groups(struct btrfs_root *root) |
9639 | { | |
9640 | struct btrfs_path *path; | |
9641 | int ret; | |
9078a3e1 | 9642 | struct btrfs_block_group_cache *cache; |
be744175 | 9643 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 9644 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
9645 | struct btrfs_key key; |
9646 | struct btrfs_key found_key; | |
5f39d397 | 9647 | struct extent_buffer *leaf; |
0af3d00b JB |
9648 | int need_clear = 0; |
9649 | u64 cache_gen; | |
96b5179d | 9650 | |
be744175 | 9651 | root = info->extent_root; |
9078a3e1 | 9652 | key.objectid = 0; |
0b86a832 | 9653 | key.offset = 0; |
962a298f | 9654 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
9655 | path = btrfs_alloc_path(); |
9656 | if (!path) | |
9657 | return -ENOMEM; | |
026fd317 | 9658 | path->reada = 1; |
9078a3e1 | 9659 | |
6c41761f | 9660 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 9661 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 9662 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 9663 | need_clear = 1; |
88c2ba3b JB |
9664 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
9665 | need_clear = 1; | |
0af3d00b | 9666 | |
d397712b | 9667 | while (1) { |
0b86a832 | 9668 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
9669 | if (ret > 0) |
9670 | break; | |
0b86a832 CM |
9671 | if (ret != 0) |
9672 | goto error; | |
920e4a58 | 9673 | |
5f39d397 CM |
9674 | leaf = path->nodes[0]; |
9675 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 MX |
9676 | |
9677 | cache = btrfs_create_block_group_cache(root, found_key.objectid, | |
9678 | found_key.offset); | |
9078a3e1 | 9679 | if (!cache) { |
0b86a832 | 9680 | ret = -ENOMEM; |
f0486c68 | 9681 | goto error; |
9078a3e1 | 9682 | } |
96303081 | 9683 | |
cf7c1ef6 LB |
9684 | if (need_clear) { |
9685 | /* | |
9686 | * When we mount with old space cache, we need to | |
9687 | * set BTRFS_DC_CLEAR and set dirty flag. | |
9688 | * | |
9689 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
9690 | * truncate the old free space cache inode and | |
9691 | * setup a new one. | |
9692 | * b) Setting 'dirty flag' makes sure that we flush | |
9693 | * the new space cache info onto disk. | |
9694 | */ | |
cf7c1ef6 | 9695 | if (btrfs_test_opt(root, SPACE_CACHE)) |
ce93ec54 | 9696 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 9697 | } |
0af3d00b | 9698 | |
5f39d397 CM |
9699 | read_extent_buffer(leaf, &cache->item, |
9700 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
9701 | sizeof(cache->item)); | |
920e4a58 | 9702 | cache->flags = btrfs_block_group_flags(&cache->item); |
0b86a832 | 9703 | |
9078a3e1 | 9704 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 9705 | btrfs_release_path(path); |
34d52cb6 | 9706 | |
3c14874a JB |
9707 | /* |
9708 | * We need to exclude the super stripes now so that the space | |
9709 | * info has super bytes accounted for, otherwise we'll think | |
9710 | * we have more space than we actually do. | |
9711 | */ | |
835d974f JB |
9712 | ret = exclude_super_stripes(root, cache); |
9713 | if (ret) { | |
9714 | /* | |
9715 | * We may have excluded something, so call this just in | |
9716 | * case. | |
9717 | */ | |
9718 | free_excluded_extents(root, cache); | |
920e4a58 | 9719 | btrfs_put_block_group(cache); |
835d974f JB |
9720 | goto error; |
9721 | } | |
3c14874a | 9722 | |
817d52f8 JB |
9723 | /* |
9724 | * check for two cases, either we are full, and therefore | |
9725 | * don't need to bother with the caching work since we won't | |
9726 | * find any space, or we are empty, and we can just add all | |
9727 | * the space in and be done with it. This saves us _alot_ of | |
9728 | * time, particularly in the full case. | |
9729 | */ | |
9730 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 9731 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 9732 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 9733 | free_excluded_extents(root, cache); |
817d52f8 | 9734 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 9735 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
9736 | cache->cached = BTRFS_CACHE_FINISHED; |
9737 | add_new_free_space(cache, root->fs_info, | |
9738 | found_key.objectid, | |
9739 | found_key.objectid + | |
9740 | found_key.offset); | |
11833d66 | 9741 | free_excluded_extents(root, cache); |
817d52f8 | 9742 | } |
96b5179d | 9743 | |
8c579fe7 JB |
9744 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
9745 | if (ret) { | |
9746 | btrfs_remove_free_space_cache(cache); | |
9747 | btrfs_put_block_group(cache); | |
9748 | goto error; | |
9749 | } | |
9750 | ||
6324fbf3 CM |
9751 | ret = update_space_info(info, cache->flags, found_key.offset, |
9752 | btrfs_block_group_used(&cache->item), | |
9753 | &space_info); | |
8c579fe7 JB |
9754 | if (ret) { |
9755 | btrfs_remove_free_space_cache(cache); | |
9756 | spin_lock(&info->block_group_cache_lock); | |
9757 | rb_erase(&cache->cache_node, | |
9758 | &info->block_group_cache_tree); | |
01eacb27 | 9759 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
9760 | spin_unlock(&info->block_group_cache_lock); |
9761 | btrfs_put_block_group(cache); | |
9762 | goto error; | |
9763 | } | |
9764 | ||
6324fbf3 | 9765 | cache->space_info = space_info; |
1b2da372 | 9766 | spin_lock(&cache->space_info->lock); |
f0486c68 | 9767 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
9768 | spin_unlock(&cache->space_info->lock); |
9769 | ||
b742bb82 | 9770 | __link_block_group(space_info, cache); |
0f9dd46c | 9771 | |
75ccf47d | 9772 | set_avail_alloc_bits(root->fs_info, cache->flags); |
47ab2a6c | 9773 | if (btrfs_chunk_readonly(root, cache->key.objectid)) { |
868f401a | 9774 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
9775 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
9776 | spin_lock(&info->unused_bgs_lock); | |
9777 | /* Should always be true but just in case. */ | |
9778 | if (list_empty(&cache->bg_list)) { | |
9779 | btrfs_get_block_group(cache); | |
9780 | list_add_tail(&cache->bg_list, | |
9781 | &info->unused_bgs); | |
9782 | } | |
9783 | spin_unlock(&info->unused_bgs_lock); | |
9784 | } | |
9078a3e1 | 9785 | } |
b742bb82 YZ |
9786 | |
9787 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
9788 | if (!(get_alloc_profile(root, space_info->flags) & | |
9789 | (BTRFS_BLOCK_GROUP_RAID10 | | |
9790 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
9791 | BTRFS_BLOCK_GROUP_RAID5 | |
9792 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
9793 | BTRFS_BLOCK_GROUP_DUP))) |
9794 | continue; | |
9795 | /* | |
9796 | * avoid allocating from un-mirrored block group if there are | |
9797 | * mirrored block groups. | |
9798 | */ | |
1095cc0d | 9799 | list_for_each_entry(cache, |
9800 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
9801 | list) | |
868f401a | 9802 | inc_block_group_ro(cache, 1); |
1095cc0d | 9803 | list_for_each_entry(cache, |
9804 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
9805 | list) | |
868f401a | 9806 | inc_block_group_ro(cache, 1); |
9078a3e1 | 9807 | } |
f0486c68 YZ |
9808 | |
9809 | init_global_block_rsv(info); | |
0b86a832 CM |
9810 | ret = 0; |
9811 | error: | |
9078a3e1 | 9812 | btrfs_free_path(path); |
0b86a832 | 9813 | return ret; |
9078a3e1 | 9814 | } |
6324fbf3 | 9815 | |
ea658bad JB |
9816 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
9817 | struct btrfs_root *root) | |
9818 | { | |
9819 | struct btrfs_block_group_cache *block_group, *tmp; | |
9820 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
9821 | struct btrfs_block_group_item item; | |
9822 | struct btrfs_key key; | |
9823 | int ret = 0; | |
d9a0540a | 9824 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 9825 | |
d9a0540a | 9826 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 9827 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 9828 | if (ret) |
c92f6be3 | 9829 | goto next; |
ea658bad JB |
9830 | |
9831 | spin_lock(&block_group->lock); | |
9832 | memcpy(&item, &block_group->item, sizeof(item)); | |
9833 | memcpy(&key, &block_group->key, sizeof(key)); | |
9834 | spin_unlock(&block_group->lock); | |
9835 | ||
9836 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
9837 | sizeof(item)); | |
9838 | if (ret) | |
9839 | btrfs_abort_transaction(trans, extent_root, ret); | |
6df9a95e JB |
9840 | ret = btrfs_finish_chunk_alloc(trans, extent_root, |
9841 | key.objectid, key.offset); | |
9842 | if (ret) | |
9843 | btrfs_abort_transaction(trans, extent_root, ret); | |
c92f6be3 FM |
9844 | next: |
9845 | list_del_init(&block_group->bg_list); | |
ea658bad | 9846 | } |
d9a0540a | 9847 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
9848 | } |
9849 | ||
6324fbf3 CM |
9850 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
9851 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 9852 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
9853 | u64 size) |
9854 | { | |
9855 | int ret; | |
6324fbf3 CM |
9856 | struct btrfs_root *extent_root; |
9857 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
9858 | |
9859 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 9860 | |
995946dd | 9861 | btrfs_set_log_full_commit(root->fs_info, trans); |
e02119d5 | 9862 | |
920e4a58 | 9863 | cache = btrfs_create_block_group_cache(root, chunk_offset, size); |
0f9dd46c JB |
9864 | if (!cache) |
9865 | return -ENOMEM; | |
34d52cb6 | 9866 | |
6324fbf3 | 9867 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 | 9868 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
6324fbf3 CM |
9869 | btrfs_set_block_group_flags(&cache->item, type); |
9870 | ||
920e4a58 | 9871 | cache->flags = type; |
11833d66 | 9872 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 9873 | cache->cached = BTRFS_CACHE_FINISHED; |
835d974f JB |
9874 | ret = exclude_super_stripes(root, cache); |
9875 | if (ret) { | |
9876 | /* | |
9877 | * We may have excluded something, so call this just in | |
9878 | * case. | |
9879 | */ | |
9880 | free_excluded_extents(root, cache); | |
920e4a58 | 9881 | btrfs_put_block_group(cache); |
835d974f JB |
9882 | return ret; |
9883 | } | |
96303081 | 9884 | |
817d52f8 JB |
9885 | add_new_free_space(cache, root->fs_info, chunk_offset, |
9886 | chunk_offset + size); | |
9887 | ||
11833d66 YZ |
9888 | free_excluded_extents(root, cache); |
9889 | ||
d0bd4560 JB |
9890 | #ifdef CONFIG_BTRFS_DEBUG |
9891 | if (btrfs_should_fragment_free_space(root, cache)) { | |
9892 | u64 new_bytes_used = size - bytes_used; | |
9893 | ||
9894 | bytes_used += new_bytes_used >> 1; | |
9895 | fragment_free_space(root, cache); | |
9896 | } | |
9897 | #endif | |
2e6e5183 FM |
9898 | /* |
9899 | * Call to ensure the corresponding space_info object is created and | |
9900 | * assigned to our block group, but don't update its counters just yet. | |
9901 | * We want our bg to be added to the rbtree with its ->space_info set. | |
9902 | */ | |
9903 | ret = update_space_info(root->fs_info, cache->flags, 0, 0, | |
9904 | &cache->space_info); | |
9905 | if (ret) { | |
9906 | btrfs_remove_free_space_cache(cache); | |
9907 | btrfs_put_block_group(cache); | |
9908 | return ret; | |
9909 | } | |
9910 | ||
8c579fe7 JB |
9911 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
9912 | if (ret) { | |
9913 | btrfs_remove_free_space_cache(cache); | |
9914 | btrfs_put_block_group(cache); | |
9915 | return ret; | |
9916 | } | |
9917 | ||
2e6e5183 FM |
9918 | /* |
9919 | * Now that our block group has its ->space_info set and is inserted in | |
9920 | * the rbtree, update the space info's counters. | |
9921 | */ | |
6324fbf3 CM |
9922 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
9923 | &cache->space_info); | |
8c579fe7 JB |
9924 | if (ret) { |
9925 | btrfs_remove_free_space_cache(cache); | |
9926 | spin_lock(&root->fs_info->block_group_cache_lock); | |
9927 | rb_erase(&cache->cache_node, | |
9928 | &root->fs_info->block_group_cache_tree); | |
01eacb27 | 9929 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
9930 | spin_unlock(&root->fs_info->block_group_cache_lock); |
9931 | btrfs_put_block_group(cache); | |
9932 | return ret; | |
9933 | } | |
c7c144db | 9934 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
9935 | |
9936 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 9937 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
9938 | spin_unlock(&cache->space_info->lock); |
9939 | ||
b742bb82 | 9940 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 9941 | |
47ab2a6c | 9942 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 9943 | |
d18a2c44 | 9944 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 9945 | |
6324fbf3 CM |
9946 | return 0; |
9947 | } | |
1a40e23b | 9948 | |
10ea00f5 ID |
9949 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
9950 | { | |
899c81ea ID |
9951 | u64 extra_flags = chunk_to_extended(flags) & |
9952 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 9953 | |
de98ced9 | 9954 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
9955 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
9956 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
9957 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
9958 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
9959 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
9960 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 9961 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
9962 | } |
9963 | ||
1a40e23b | 9964 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
04216820 FM |
9965 | struct btrfs_root *root, u64 group_start, |
9966 | struct extent_map *em) | |
1a40e23b ZY |
9967 | { |
9968 | struct btrfs_path *path; | |
9969 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 9970 | struct btrfs_free_cluster *cluster; |
0af3d00b | 9971 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 9972 | struct btrfs_key key; |
0af3d00b | 9973 | struct inode *inode; |
c1895442 | 9974 | struct kobject *kobj = NULL; |
1a40e23b | 9975 | int ret; |
10ea00f5 | 9976 | int index; |
89a55897 | 9977 | int factor; |
4f69cb98 | 9978 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 9979 | bool remove_em; |
1a40e23b | 9980 | |
1a40e23b ZY |
9981 | root = root->fs_info->extent_root; |
9982 | ||
9983 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
9984 | BUG_ON(!block_group); | |
c146afad | 9985 | BUG_ON(!block_group->ro); |
1a40e23b | 9986 | |
9f7c43c9 | 9987 | /* |
9988 | * Free the reserved super bytes from this block group before | |
9989 | * remove it. | |
9990 | */ | |
9991 | free_excluded_extents(root, block_group); | |
9992 | ||
1a40e23b | 9993 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 9994 | index = get_block_group_index(block_group); |
89a55897 JB |
9995 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
9996 | BTRFS_BLOCK_GROUP_RAID1 | | |
9997 | BTRFS_BLOCK_GROUP_RAID10)) | |
9998 | factor = 2; | |
9999 | else | |
10000 | factor = 1; | |
1a40e23b | 10001 | |
44fb5511 CM |
10002 | /* make sure this block group isn't part of an allocation cluster */ |
10003 | cluster = &root->fs_info->data_alloc_cluster; | |
10004 | spin_lock(&cluster->refill_lock); | |
10005 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10006 | spin_unlock(&cluster->refill_lock); | |
10007 | ||
10008 | /* | |
10009 | * make sure this block group isn't part of a metadata | |
10010 | * allocation cluster | |
10011 | */ | |
10012 | cluster = &root->fs_info->meta_alloc_cluster; | |
10013 | spin_lock(&cluster->refill_lock); | |
10014 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10015 | spin_unlock(&cluster->refill_lock); | |
10016 | ||
1a40e23b | 10017 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10018 | if (!path) { |
10019 | ret = -ENOMEM; | |
10020 | goto out; | |
10021 | } | |
1a40e23b | 10022 | |
1bbc621e CM |
10023 | /* |
10024 | * get the inode first so any iput calls done for the io_list | |
10025 | * aren't the final iput (no unlinks allowed now) | |
10026 | */ | |
10b2f34d | 10027 | inode = lookup_free_space_inode(tree_root, block_group, path); |
1bbc621e CM |
10028 | |
10029 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10030 | /* | |
10031 | * make sure our free spache cache IO is done before remove the | |
10032 | * free space inode | |
10033 | */ | |
10034 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10035 | if (!list_empty(&block_group->io_list)) { | |
10036 | list_del_init(&block_group->io_list); | |
10037 | ||
10038 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10039 | ||
10040 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10041 | btrfs_wait_cache_io(root, trans, block_group, | |
10042 | &block_group->io_ctl, path, | |
10043 | block_group->key.objectid); | |
10044 | btrfs_put_block_group(block_group); | |
10045 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10046 | } | |
10047 | ||
10048 | if (!list_empty(&block_group->dirty_list)) { | |
10049 | list_del_init(&block_group->dirty_list); | |
10050 | btrfs_put_block_group(block_group); | |
10051 | } | |
10052 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10053 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10054 | ||
0af3d00b | 10055 | if (!IS_ERR(inode)) { |
b532402e | 10056 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
10057 | if (ret) { |
10058 | btrfs_add_delayed_iput(inode); | |
10059 | goto out; | |
10060 | } | |
0af3d00b JB |
10061 | clear_nlink(inode); |
10062 | /* One for the block groups ref */ | |
10063 | spin_lock(&block_group->lock); | |
10064 | if (block_group->iref) { | |
10065 | block_group->iref = 0; | |
10066 | block_group->inode = NULL; | |
10067 | spin_unlock(&block_group->lock); | |
10068 | iput(inode); | |
10069 | } else { | |
10070 | spin_unlock(&block_group->lock); | |
10071 | } | |
10072 | /* One for our lookup ref */ | |
455757c3 | 10073 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10074 | } |
10075 | ||
10076 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10077 | key.offset = block_group->key.objectid; | |
10078 | key.type = 0; | |
10079 | ||
10080 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10081 | if (ret < 0) | |
10082 | goto out; | |
10083 | if (ret > 0) | |
b3b4aa74 | 10084 | btrfs_release_path(path); |
0af3d00b JB |
10085 | if (ret == 0) { |
10086 | ret = btrfs_del_item(trans, tree_root, path); | |
10087 | if (ret) | |
10088 | goto out; | |
b3b4aa74 | 10089 | btrfs_release_path(path); |
0af3d00b JB |
10090 | } |
10091 | ||
3dfdb934 | 10092 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
10093 | rb_erase(&block_group->cache_node, |
10094 | &root->fs_info->block_group_cache_tree); | |
292cbd51 | 10095 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd LB |
10096 | |
10097 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
10098 | root->fs_info->first_logical_byte = (u64)-1; | |
3dfdb934 | 10099 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 10100 | |
80eb234a | 10101 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10102 | /* |
10103 | * we must use list_del_init so people can check to see if they | |
10104 | * are still on the list after taking the semaphore | |
10105 | */ | |
10106 | list_del_init(&block_group->list); | |
6ab0a202 | 10107 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10108 | kobj = block_group->space_info->block_group_kobjs[index]; |
10109 | block_group->space_info->block_group_kobjs[index] = NULL; | |
10ea00f5 | 10110 | clear_avail_alloc_bits(root->fs_info, block_group->flags); |
6ab0a202 | 10111 | } |
80eb234a | 10112 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10113 | if (kobj) { |
10114 | kobject_del(kobj); | |
10115 | kobject_put(kobj); | |
10116 | } | |
1a40e23b | 10117 | |
4f69cb98 FM |
10118 | if (block_group->has_caching_ctl) |
10119 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10120 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10121 | wait_block_group_cache_done(block_group); |
4f69cb98 FM |
10122 | if (block_group->has_caching_ctl) { |
10123 | down_write(&root->fs_info->commit_root_sem); | |
10124 | if (!caching_ctl) { | |
10125 | struct btrfs_caching_control *ctl; | |
10126 | ||
10127 | list_for_each_entry(ctl, | |
10128 | &root->fs_info->caching_block_groups, list) | |
10129 | if (ctl->block_group == block_group) { | |
10130 | caching_ctl = ctl; | |
10131 | atomic_inc(&caching_ctl->count); | |
10132 | break; | |
10133 | } | |
10134 | } | |
10135 | if (caching_ctl) | |
10136 | list_del_init(&caching_ctl->list); | |
10137 | up_write(&root->fs_info->commit_root_sem); | |
10138 | if (caching_ctl) { | |
10139 | /* Once for the caching bgs list and once for us. */ | |
10140 | put_caching_control(caching_ctl); | |
10141 | put_caching_control(caching_ctl); | |
10142 | } | |
10143 | } | |
817d52f8 | 10144 | |
ce93ec54 JB |
10145 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10146 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10147 | WARN_ON(1); |
10148 | } | |
10149 | if (!list_empty(&block_group->io_list)) { | |
10150 | WARN_ON(1); | |
ce93ec54 JB |
10151 | } |
10152 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10153 | btrfs_remove_free_space_cache(block_group); |
10154 | ||
c146afad | 10155 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10156 | list_del_init(&block_group->ro_list); |
18d018ad ZL |
10157 | |
10158 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
10159 | WARN_ON(block_group->space_info->total_bytes | |
10160 | < block_group->key.offset); | |
10161 | WARN_ON(block_group->space_info->bytes_readonly | |
10162 | < block_group->key.offset); | |
10163 | WARN_ON(block_group->space_info->disk_total | |
10164 | < block_group->key.offset * factor); | |
10165 | } | |
c146afad YZ |
10166 | block_group->space_info->total_bytes -= block_group->key.offset; |
10167 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10168 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10169 | |
c146afad | 10170 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10171 | |
0af3d00b JB |
10172 | memcpy(&key, &block_group->key, sizeof(key)); |
10173 | ||
04216820 | 10174 | lock_chunks(root); |
495e64f4 FM |
10175 | if (!list_empty(&em->list)) { |
10176 | /* We're in the transaction->pending_chunks list. */ | |
10177 | free_extent_map(em); | |
10178 | } | |
04216820 FM |
10179 | spin_lock(&block_group->lock); |
10180 | block_group->removed = 1; | |
10181 | /* | |
10182 | * At this point trimming can't start on this block group, because we | |
10183 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10184 | * so no one can't find it anymore and even if someone already got this | |
10185 | * block group before we removed it from the rbtree, they have already | |
10186 | * incremented block_group->trimming - if they didn't, they won't find | |
10187 | * any free space entries because we already removed them all when we | |
10188 | * called btrfs_remove_free_space_cache(). | |
10189 | * | |
10190 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10191 | * to prevent the same logical address range and physical device space | |
10192 | * ranges from being reused for a new block group. This is because our | |
10193 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10194 | * completely transactionless, so while it is trimming a range the | |
10195 | * currently running transaction might finish and a new one start, | |
10196 | * allowing for new block groups to be created that can reuse the same | |
10197 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10198 | * |
10199 | * There may also be an implicit trim operation if the file system | |
10200 | * is mounted with -odiscard. The same protections must remain | |
10201 | * in place until the extents have been discarded completely when | |
10202 | * the transaction commit has completed. | |
04216820 FM |
10203 | */ |
10204 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10205 | /* | |
10206 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10207 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10208 | * before checking block_group->removed). | |
10209 | */ | |
10210 | if (!remove_em) { | |
10211 | /* | |
10212 | * Our em might be in trans->transaction->pending_chunks which | |
10213 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10214 | * and so is the fs_info->pinned_chunks list. | |
10215 | * | |
10216 | * So at this point we must be holding the chunk_mutex to avoid | |
10217 | * any races with chunk allocation (more specifically at | |
10218 | * volumes.c:contains_pending_extent()), to ensure it always | |
10219 | * sees the em, either in the pending_chunks list or in the | |
10220 | * pinned_chunks list. | |
10221 | */ | |
10222 | list_move_tail(&em->list, &root->fs_info->pinned_chunks); | |
10223 | } | |
10224 | spin_unlock(&block_group->lock); | |
04216820 FM |
10225 | |
10226 | if (remove_em) { | |
10227 | struct extent_map_tree *em_tree; | |
10228 | ||
10229 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
10230 | write_lock(&em_tree->lock); | |
8dbcd10f FM |
10231 | /* |
10232 | * The em might be in the pending_chunks list, so make sure the | |
10233 | * chunk mutex is locked, since remove_extent_mapping() will | |
10234 | * delete us from that list. | |
10235 | */ | |
04216820 FM |
10236 | remove_extent_mapping(em_tree, em); |
10237 | write_unlock(&em_tree->lock); | |
10238 | /* once for the tree */ | |
10239 | free_extent_map(em); | |
10240 | } | |
10241 | ||
8dbcd10f FM |
10242 | unlock_chunks(root); |
10243 | ||
fa9c0d79 CM |
10244 | btrfs_put_block_group(block_group); |
10245 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10246 | |
10247 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10248 | if (ret > 0) | |
10249 | ret = -EIO; | |
10250 | if (ret < 0) | |
10251 | goto out; | |
10252 | ||
10253 | ret = btrfs_del_item(trans, root, path); | |
10254 | out: | |
10255 | btrfs_free_path(path); | |
10256 | return ret; | |
10257 | } | |
acce952b | 10258 | |
8eab77ff | 10259 | struct btrfs_trans_handle * |
7fd01182 FM |
10260 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10261 | const u64 chunk_offset) | |
8eab77ff | 10262 | { |
7fd01182 FM |
10263 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10264 | struct extent_map *em; | |
10265 | struct map_lookup *map; | |
10266 | unsigned int num_items; | |
10267 | ||
10268 | read_lock(&em_tree->lock); | |
10269 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10270 | read_unlock(&em_tree->lock); | |
10271 | ASSERT(em && em->start == chunk_offset); | |
10272 | ||
8eab77ff | 10273 | /* |
7fd01182 FM |
10274 | * We need to reserve 3 + N units from the metadata space info in order |
10275 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10276 | * btrfs_remove_block_group()), which are used for: | |
10277 | * | |
8eab77ff FM |
10278 | * 1 unit for adding the free space inode's orphan (located in the tree |
10279 | * of tree roots). | |
7fd01182 FM |
10280 | * 1 unit for deleting the block group item (located in the extent |
10281 | * tree). | |
10282 | * 1 unit for deleting the free space item (located in tree of tree | |
10283 | * roots). | |
10284 | * N units for deleting N device extent items corresponding to each | |
10285 | * stripe (located in the device tree). | |
10286 | * | |
10287 | * In order to remove a block group we also need to reserve units in the | |
10288 | * system space info in order to update the chunk tree (update one or | |
10289 | * more device items and remove one chunk item), but this is done at | |
10290 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10291 | */ |
7fd01182 FM |
10292 | map = (struct map_lookup *)em->bdev; |
10293 | num_items = 3 + map->num_stripes; | |
10294 | free_extent_map(em); | |
10295 | ||
8eab77ff | 10296 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10297 | num_items, 1); |
8eab77ff FM |
10298 | } |
10299 | ||
47ab2a6c JB |
10300 | /* |
10301 | * Process the unused_bgs list and remove any that don't have any allocated | |
10302 | * space inside of them. | |
10303 | */ | |
10304 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10305 | { | |
10306 | struct btrfs_block_group_cache *block_group; | |
10307 | struct btrfs_space_info *space_info; | |
10308 | struct btrfs_root *root = fs_info->extent_root; | |
10309 | struct btrfs_trans_handle *trans; | |
10310 | int ret = 0; | |
10311 | ||
10312 | if (!fs_info->open) | |
10313 | return; | |
10314 | ||
10315 | spin_lock(&fs_info->unused_bgs_lock); | |
10316 | while (!list_empty(&fs_info->unused_bgs)) { | |
10317 | u64 start, end; | |
e33e17ee | 10318 | int trimming; |
47ab2a6c JB |
10319 | |
10320 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10321 | struct btrfs_block_group_cache, | |
10322 | bg_list); | |
47ab2a6c | 10323 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10324 | |
10325 | space_info = block_group->space_info; | |
10326 | ||
47ab2a6c JB |
10327 | if (ret || btrfs_mixed_space_info(space_info)) { |
10328 | btrfs_put_block_group(block_group); | |
10329 | continue; | |
10330 | } | |
10331 | spin_unlock(&fs_info->unused_bgs_lock); | |
10332 | ||
d5f2e33b | 10333 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10334 | |
47ab2a6c JB |
10335 | /* Don't want to race with allocators so take the groups_sem */ |
10336 | down_write(&space_info->groups_sem); | |
10337 | spin_lock(&block_group->lock); | |
10338 | if (block_group->reserved || | |
10339 | btrfs_block_group_used(&block_group->item) || | |
aefbe9a6 ZL |
10340 | block_group->ro || |
10341 | list_is_singular(&block_group->list)) { | |
47ab2a6c JB |
10342 | /* |
10343 | * We want to bail if we made new allocations or have | |
10344 | * outstanding allocations in this block group. We do | |
10345 | * the ro check in case balance is currently acting on | |
10346 | * this block group. | |
10347 | */ | |
10348 | spin_unlock(&block_group->lock); | |
10349 | up_write(&space_info->groups_sem); | |
10350 | goto next; | |
10351 | } | |
10352 | spin_unlock(&block_group->lock); | |
10353 | ||
10354 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10355 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10356 | up_write(&space_info->groups_sem); |
10357 | if (ret < 0) { | |
10358 | ret = 0; | |
10359 | goto next; | |
10360 | } | |
10361 | ||
10362 | /* | |
10363 | * Want to do this before we do anything else so we can recover | |
10364 | * properly if we fail to join the transaction. | |
10365 | */ | |
7fd01182 FM |
10366 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10367 | block_group->key.objectid); | |
47ab2a6c | 10368 | if (IS_ERR(trans)) { |
868f401a | 10369 | btrfs_dec_block_group_ro(root, block_group); |
47ab2a6c JB |
10370 | ret = PTR_ERR(trans); |
10371 | goto next; | |
10372 | } | |
10373 | ||
10374 | /* | |
10375 | * We could have pending pinned extents for this block group, | |
10376 | * just delete them, we don't care about them anymore. | |
10377 | */ | |
10378 | start = block_group->key.objectid; | |
10379 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10380 | /* |
10381 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10382 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10383 | * another task might be running finish_extent_commit() for the | |
10384 | * previous transaction N - 1, and have seen a range belonging | |
10385 | * to the block group in freed_extents[] before we were able to | |
10386 | * clear the whole block group range from freed_extents[]. This | |
10387 | * means that task can lookup for the block group after we | |
10388 | * unpinned it from freed_extents[] and removed it, leading to | |
10389 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10390 | */ | |
10391 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10392 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
47ab2a6c | 10393 | EXTENT_DIRTY, GFP_NOFS); |
758eb51e | 10394 | if (ret) { |
d4b450cd | 10395 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
868f401a | 10396 | btrfs_dec_block_group_ro(root, block_group); |
758eb51e FM |
10397 | goto end_trans; |
10398 | } | |
10399 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
47ab2a6c | 10400 | EXTENT_DIRTY, GFP_NOFS); |
758eb51e | 10401 | if (ret) { |
d4b450cd | 10402 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
868f401a | 10403 | btrfs_dec_block_group_ro(root, block_group); |
758eb51e FM |
10404 | goto end_trans; |
10405 | } | |
d4b450cd | 10406 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10407 | |
10408 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10409 | spin_lock(&space_info->lock); |
10410 | spin_lock(&block_group->lock); | |
10411 | ||
10412 | space_info->bytes_pinned -= block_group->pinned; | |
10413 | space_info->bytes_readonly += block_group->pinned; | |
10414 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10415 | -block_group->pinned); | |
47ab2a6c JB |
10416 | block_group->pinned = 0; |
10417 | ||
c30666d4 ZL |
10418 | spin_unlock(&block_group->lock); |
10419 | spin_unlock(&space_info->lock); | |
10420 | ||
e33e17ee JM |
10421 | /* DISCARD can flip during remount */ |
10422 | trimming = btrfs_test_opt(root, DISCARD); | |
10423 | ||
10424 | /* Implicit trim during transaction commit. */ | |
10425 | if (trimming) | |
10426 | btrfs_get_block_group_trimming(block_group); | |
10427 | ||
47ab2a6c JB |
10428 | /* |
10429 | * Btrfs_remove_chunk will abort the transaction if things go | |
10430 | * horribly wrong. | |
10431 | */ | |
10432 | ret = btrfs_remove_chunk(trans, root, | |
10433 | block_group->key.objectid); | |
e33e17ee JM |
10434 | |
10435 | if (ret) { | |
10436 | if (trimming) | |
10437 | btrfs_put_block_group_trimming(block_group); | |
10438 | goto end_trans; | |
10439 | } | |
10440 | ||
10441 | /* | |
10442 | * If we're not mounted with -odiscard, we can just forget | |
10443 | * about this block group. Otherwise we'll need to wait | |
10444 | * until transaction commit to do the actual discard. | |
10445 | */ | |
10446 | if (trimming) { | |
10447 | WARN_ON(!list_empty(&block_group->bg_list)); | |
10448 | spin_lock(&trans->transaction->deleted_bgs_lock); | |
10449 | list_move(&block_group->bg_list, | |
10450 | &trans->transaction->deleted_bgs); | |
10451 | spin_unlock(&trans->transaction->deleted_bgs_lock); | |
10452 | btrfs_get_block_group(block_group); | |
10453 | } | |
758eb51e | 10454 | end_trans: |
47ab2a6c JB |
10455 | btrfs_end_transaction(trans, root); |
10456 | next: | |
d5f2e33b | 10457 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10458 | btrfs_put_block_group(block_group); |
10459 | spin_lock(&fs_info->unused_bgs_lock); | |
10460 | } | |
10461 | spin_unlock(&fs_info->unused_bgs_lock); | |
10462 | } | |
10463 | ||
c59021f8 | 10464 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10465 | { | |
10466 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10467 | struct btrfs_super_block *disk_super; |
10468 | u64 features; | |
10469 | u64 flags; | |
10470 | int mixed = 0; | |
c59021f8 | 10471 | int ret; |
10472 | ||
6c41761f | 10473 | disk_super = fs_info->super_copy; |
1aba86d6 | 10474 | if (!btrfs_super_root(disk_super)) |
10475 | return 1; | |
c59021f8 | 10476 | |
1aba86d6 | 10477 | features = btrfs_super_incompat_flags(disk_super); |
10478 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10479 | mixed = 1; | |
c59021f8 | 10480 | |
1aba86d6 | 10481 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
10482 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 10483 | if (ret) |
1aba86d6 | 10484 | goto out; |
c59021f8 | 10485 | |
1aba86d6 | 10486 | if (mixed) { |
10487 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
10488 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
10489 | } else { | |
10490 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
10491 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
10492 | if (ret) | |
10493 | goto out; | |
10494 | ||
10495 | flags = BTRFS_BLOCK_GROUP_DATA; | |
10496 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
10497 | } | |
10498 | out: | |
c59021f8 | 10499 | return ret; |
10500 | } | |
10501 | ||
acce952b | 10502 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
10503 | { | |
678886bd | 10504 | return unpin_extent_range(root, start, end, false); |
acce952b | 10505 | } |
10506 | ||
499f377f JM |
10507 | /* |
10508 | * It used to be that old block groups would be left around forever. | |
10509 | * Iterating over them would be enough to trim unused space. Since we | |
10510 | * now automatically remove them, we also need to iterate over unallocated | |
10511 | * space. | |
10512 | * | |
10513 | * We don't want a transaction for this since the discard may take a | |
10514 | * substantial amount of time. We don't require that a transaction be | |
10515 | * running, but we do need to take a running transaction into account | |
10516 | * to ensure that we're not discarding chunks that were released in | |
10517 | * the current transaction. | |
10518 | * | |
10519 | * Holding the chunks lock will prevent other threads from allocating | |
10520 | * or releasing chunks, but it won't prevent a running transaction | |
10521 | * from committing and releasing the memory that the pending chunks | |
10522 | * list head uses. For that, we need to take a reference to the | |
10523 | * transaction. | |
10524 | */ | |
10525 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10526 | u64 minlen, u64 *trimmed) | |
10527 | { | |
10528 | u64 start = 0, len = 0; | |
10529 | int ret; | |
10530 | ||
10531 | *trimmed = 0; | |
10532 | ||
10533 | /* Not writeable = nothing to do. */ | |
10534 | if (!device->writeable) | |
10535 | return 0; | |
10536 | ||
10537 | /* No free space = nothing to do. */ | |
10538 | if (device->total_bytes <= device->bytes_used) | |
10539 | return 0; | |
10540 | ||
10541 | ret = 0; | |
10542 | ||
10543 | while (1) { | |
10544 | struct btrfs_fs_info *fs_info = device->dev_root->fs_info; | |
10545 | struct btrfs_transaction *trans; | |
10546 | u64 bytes; | |
10547 | ||
10548 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10549 | if (ret) | |
10550 | return ret; | |
10551 | ||
10552 | down_read(&fs_info->commit_root_sem); | |
10553 | ||
10554 | spin_lock(&fs_info->trans_lock); | |
10555 | trans = fs_info->running_transaction; | |
10556 | if (trans) | |
10557 | atomic_inc(&trans->use_count); | |
10558 | spin_unlock(&fs_info->trans_lock); | |
10559 | ||
10560 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10561 | &start, &len); | |
10562 | if (trans) | |
10563 | btrfs_put_transaction(trans); | |
10564 | ||
10565 | if (ret) { | |
10566 | up_read(&fs_info->commit_root_sem); | |
10567 | mutex_unlock(&fs_info->chunk_mutex); | |
10568 | if (ret == -ENOSPC) | |
10569 | ret = 0; | |
10570 | break; | |
10571 | } | |
10572 | ||
10573 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10574 | up_read(&fs_info->commit_root_sem); | |
10575 | mutex_unlock(&fs_info->chunk_mutex); | |
10576 | ||
10577 | if (ret) | |
10578 | break; | |
10579 | ||
10580 | start += len; | |
10581 | *trimmed += bytes; | |
10582 | ||
10583 | if (fatal_signal_pending(current)) { | |
10584 | ret = -ERESTARTSYS; | |
10585 | break; | |
10586 | } | |
10587 | ||
10588 | cond_resched(); | |
10589 | } | |
10590 | ||
10591 | return ret; | |
10592 | } | |
10593 | ||
f7039b1d LD |
10594 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) |
10595 | { | |
10596 | struct btrfs_fs_info *fs_info = root->fs_info; | |
10597 | struct btrfs_block_group_cache *cache = NULL; | |
499f377f JM |
10598 | struct btrfs_device *device; |
10599 | struct list_head *devices; | |
f7039b1d LD |
10600 | u64 group_trimmed; |
10601 | u64 start; | |
10602 | u64 end; | |
10603 | u64 trimmed = 0; | |
2cac13e4 | 10604 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10605 | int ret = 0; |
10606 | ||
2cac13e4 LB |
10607 | /* |
10608 | * try to trim all FS space, our block group may start from non-zero. | |
10609 | */ | |
10610 | if (range->len == total_bytes) | |
10611 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10612 | else | |
10613 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10614 | |
10615 | while (cache) { | |
10616 | if (cache->key.objectid >= (range->start + range->len)) { | |
10617 | btrfs_put_block_group(cache); | |
10618 | break; | |
10619 | } | |
10620 | ||
10621 | start = max(range->start, cache->key.objectid); | |
10622 | end = min(range->start + range->len, | |
10623 | cache->key.objectid + cache->key.offset); | |
10624 | ||
10625 | if (end - start >= range->minlen) { | |
10626 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10627 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
10628 | if (ret) { |
10629 | btrfs_put_block_group(cache); | |
10630 | break; | |
10631 | } | |
10632 | ret = wait_block_group_cache_done(cache); | |
10633 | if (ret) { | |
10634 | btrfs_put_block_group(cache); | |
10635 | break; | |
10636 | } | |
f7039b1d LD |
10637 | } |
10638 | ret = btrfs_trim_block_group(cache, | |
10639 | &group_trimmed, | |
10640 | start, | |
10641 | end, | |
10642 | range->minlen); | |
10643 | ||
10644 | trimmed += group_trimmed; | |
10645 | if (ret) { | |
10646 | btrfs_put_block_group(cache); | |
10647 | break; | |
10648 | } | |
10649 | } | |
10650 | ||
10651 | cache = next_block_group(fs_info->tree_root, cache); | |
10652 | } | |
10653 | ||
499f377f JM |
10654 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
10655 | devices = &root->fs_info->fs_devices->alloc_list; | |
10656 | list_for_each_entry(device, devices, dev_alloc_list) { | |
10657 | ret = btrfs_trim_free_extents(device, range->minlen, | |
10658 | &group_trimmed); | |
10659 | if (ret) | |
10660 | break; | |
10661 | ||
10662 | trimmed += group_trimmed; | |
10663 | } | |
10664 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
10665 | ||
f7039b1d LD |
10666 | range->len = trimmed; |
10667 | return ret; | |
10668 | } | |
8257b2dc MX |
10669 | |
10670 | /* | |
9ea24bbe FM |
10671 | * btrfs_{start,end}_write_no_snapshoting() are similar to |
10672 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing | |
10673 | * data into the page cache through nocow before the subvolume is snapshoted, | |
10674 | * but flush the data into disk after the snapshot creation, or to prevent | |
10675 | * operations while snapshoting is ongoing and that cause the snapshot to be | |
10676 | * inconsistent (writes followed by expanding truncates for example). | |
8257b2dc | 10677 | */ |
9ea24bbe | 10678 | void btrfs_end_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc MX |
10679 | { |
10680 | percpu_counter_dec(&root->subv_writers->counter); | |
10681 | /* | |
a83342aa | 10682 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
10683 | */ |
10684 | smp_mb(); | |
10685 | if (waitqueue_active(&root->subv_writers->wait)) | |
10686 | wake_up(&root->subv_writers->wait); | |
10687 | } | |
10688 | ||
9ea24bbe | 10689 | int btrfs_start_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc | 10690 | { |
ee39b432 | 10691 | if (atomic_read(&root->will_be_snapshoted)) |
8257b2dc MX |
10692 | return 0; |
10693 | ||
10694 | percpu_counter_inc(&root->subv_writers->counter); | |
10695 | /* | |
10696 | * Make sure counter is updated before we check for snapshot creation. | |
10697 | */ | |
10698 | smp_mb(); | |
ee39b432 | 10699 | if (atomic_read(&root->will_be_snapshoted)) { |
9ea24bbe | 10700 | btrfs_end_write_no_snapshoting(root); |
8257b2dc MX |
10701 | return 0; |
10702 | } | |
10703 | return 1; | |
10704 | } |