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> |
4b4e25f2 | 25 | #include "compat.h" |
74493f7a | 26 | #include "hash.h" |
fec577fb CM |
27 | #include "ctree.h" |
28 | #include "disk-io.h" | |
29 | #include "print-tree.h" | |
e089f05c | 30 | #include "transaction.h" |
0b86a832 | 31 | #include "volumes.h" |
925baedd | 32 | #include "locking.h" |
fa9c0d79 | 33 | #include "free-space-cache.h" |
fec577fb | 34 | |
f3465ca4 JB |
35 | static int update_block_group(struct btrfs_trans_handle *trans, |
36 | struct btrfs_root *root, | |
37 | u64 bytenr, u64 num_bytes, int alloc, | |
38 | int mark_free); | |
11833d66 YZ |
39 | static int update_reserved_extents(struct btrfs_block_group_cache *cache, |
40 | u64 num_bytes, int reserve); | |
5d4f98a2 YZ |
41 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
42 | struct btrfs_root *root, | |
43 | u64 bytenr, u64 num_bytes, u64 parent, | |
44 | u64 root_objectid, u64 owner_objectid, | |
45 | u64 owner_offset, int refs_to_drop, | |
46 | struct btrfs_delayed_extent_op *extra_op); | |
47 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
48 | struct extent_buffer *leaf, | |
49 | struct btrfs_extent_item *ei); | |
50 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
51 | struct btrfs_root *root, | |
52 | u64 parent, u64 root_objectid, | |
53 | u64 flags, u64 owner, u64 offset, | |
54 | struct btrfs_key *ins, int ref_mod); | |
55 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
56 | struct btrfs_root *root, | |
57 | u64 parent, u64 root_objectid, | |
58 | u64 flags, struct btrfs_disk_key *key, | |
59 | int level, struct btrfs_key *ins); | |
6a63209f JB |
60 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
61 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
62 | u64 flags, int force); | |
11833d66 YZ |
63 | static int pin_down_bytes(struct btrfs_trans_handle *trans, |
64 | struct btrfs_root *root, | |
65 | struct btrfs_path *path, | |
66 | u64 bytenr, u64 num_bytes, | |
67 | int is_data, int reserved, | |
68 | struct extent_buffer **must_clean); | |
69 | static int find_next_key(struct btrfs_path *path, int level, | |
70 | struct btrfs_key *key); | |
9ed74f2d JB |
71 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
72 | int dump_block_groups); | |
6a63209f | 73 | |
817d52f8 JB |
74 | static noinline int |
75 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
76 | { | |
77 | smp_mb(); | |
78 | return cache->cached == BTRFS_CACHE_FINISHED; | |
79 | } | |
80 | ||
0f9dd46c JB |
81 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
82 | { | |
83 | return (cache->flags & bits) == bits; | |
84 | } | |
85 | ||
86 | /* | |
87 | * this adds the block group to the fs_info rb tree for the block group | |
88 | * cache | |
89 | */ | |
b2950863 | 90 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
91 | struct btrfs_block_group_cache *block_group) |
92 | { | |
93 | struct rb_node **p; | |
94 | struct rb_node *parent = NULL; | |
95 | struct btrfs_block_group_cache *cache; | |
96 | ||
97 | spin_lock(&info->block_group_cache_lock); | |
98 | p = &info->block_group_cache_tree.rb_node; | |
99 | ||
100 | while (*p) { | |
101 | parent = *p; | |
102 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
103 | cache_node); | |
104 | if (block_group->key.objectid < cache->key.objectid) { | |
105 | p = &(*p)->rb_left; | |
106 | } else if (block_group->key.objectid > cache->key.objectid) { | |
107 | p = &(*p)->rb_right; | |
108 | } else { | |
109 | spin_unlock(&info->block_group_cache_lock); | |
110 | return -EEXIST; | |
111 | } | |
112 | } | |
113 | ||
114 | rb_link_node(&block_group->cache_node, parent, p); | |
115 | rb_insert_color(&block_group->cache_node, | |
116 | &info->block_group_cache_tree); | |
117 | spin_unlock(&info->block_group_cache_lock); | |
118 | ||
119 | return 0; | |
120 | } | |
121 | ||
122 | /* | |
123 | * This will return the block group at or after bytenr if contains is 0, else | |
124 | * it will return the block group that contains the bytenr | |
125 | */ | |
126 | static struct btrfs_block_group_cache * | |
127 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
128 | int contains) | |
129 | { | |
130 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
131 | struct rb_node *n; | |
132 | u64 end, start; | |
133 | ||
134 | spin_lock(&info->block_group_cache_lock); | |
135 | n = info->block_group_cache_tree.rb_node; | |
136 | ||
137 | while (n) { | |
138 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
139 | cache_node); | |
140 | end = cache->key.objectid + cache->key.offset - 1; | |
141 | start = cache->key.objectid; | |
142 | ||
143 | if (bytenr < start) { | |
144 | if (!contains && (!ret || start < ret->key.objectid)) | |
145 | ret = cache; | |
146 | n = n->rb_left; | |
147 | } else if (bytenr > start) { | |
148 | if (contains && bytenr <= end) { | |
149 | ret = cache; | |
150 | break; | |
151 | } | |
152 | n = n->rb_right; | |
153 | } else { | |
154 | ret = cache; | |
155 | break; | |
156 | } | |
157 | } | |
d2fb3437 YZ |
158 | if (ret) |
159 | atomic_inc(&ret->count); | |
0f9dd46c JB |
160 | spin_unlock(&info->block_group_cache_lock); |
161 | ||
162 | return ret; | |
163 | } | |
164 | ||
11833d66 YZ |
165 | static int add_excluded_extent(struct btrfs_root *root, |
166 | u64 start, u64 num_bytes) | |
817d52f8 | 167 | { |
11833d66 YZ |
168 | u64 end = start + num_bytes - 1; |
169 | set_extent_bits(&root->fs_info->freed_extents[0], | |
170 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
171 | set_extent_bits(&root->fs_info->freed_extents[1], | |
172 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
173 | return 0; | |
174 | } | |
817d52f8 | 175 | |
11833d66 YZ |
176 | static void free_excluded_extents(struct btrfs_root *root, |
177 | struct btrfs_block_group_cache *cache) | |
178 | { | |
179 | u64 start, end; | |
817d52f8 | 180 | |
11833d66 YZ |
181 | start = cache->key.objectid; |
182 | end = start + cache->key.offset - 1; | |
183 | ||
184 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
185 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
186 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
187 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
188 | } |
189 | ||
11833d66 YZ |
190 | static int exclude_super_stripes(struct btrfs_root *root, |
191 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 192 | { |
817d52f8 JB |
193 | u64 bytenr; |
194 | u64 *logical; | |
195 | int stripe_len; | |
196 | int i, nr, ret; | |
197 | ||
06b2331f YZ |
198 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
199 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
200 | cache->bytes_super += stripe_len; | |
201 | ret = add_excluded_extent(root, cache->key.objectid, | |
202 | stripe_len); | |
203 | BUG_ON(ret); | |
204 | } | |
205 | ||
817d52f8 JB |
206 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
207 | bytenr = btrfs_sb_offset(i); | |
208 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
209 | cache->key.objectid, bytenr, | |
210 | 0, &logical, &nr, &stripe_len); | |
211 | BUG_ON(ret); | |
11833d66 | 212 | |
817d52f8 | 213 | while (nr--) { |
1b2da372 | 214 | cache->bytes_super += stripe_len; |
11833d66 YZ |
215 | ret = add_excluded_extent(root, logical[nr], |
216 | stripe_len); | |
217 | BUG_ON(ret); | |
817d52f8 | 218 | } |
11833d66 | 219 | |
817d52f8 JB |
220 | kfree(logical); |
221 | } | |
817d52f8 JB |
222 | return 0; |
223 | } | |
224 | ||
11833d66 YZ |
225 | static struct btrfs_caching_control * |
226 | get_caching_control(struct btrfs_block_group_cache *cache) | |
227 | { | |
228 | struct btrfs_caching_control *ctl; | |
229 | ||
230 | spin_lock(&cache->lock); | |
231 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
232 | spin_unlock(&cache->lock); | |
233 | return NULL; | |
234 | } | |
235 | ||
236 | ctl = cache->caching_ctl; | |
237 | atomic_inc(&ctl->count); | |
238 | spin_unlock(&cache->lock); | |
239 | return ctl; | |
240 | } | |
241 | ||
242 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
243 | { | |
244 | if (atomic_dec_and_test(&ctl->count)) | |
245 | kfree(ctl); | |
246 | } | |
247 | ||
0f9dd46c JB |
248 | /* |
249 | * this is only called by cache_block_group, since we could have freed extents | |
250 | * we need to check the pinned_extents for any extents that can't be used yet | |
251 | * since their free space will be released as soon as the transaction commits. | |
252 | */ | |
817d52f8 | 253 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
254 | struct btrfs_fs_info *info, u64 start, u64 end) |
255 | { | |
817d52f8 | 256 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
257 | int ret; |
258 | ||
259 | while (start < end) { | |
11833d66 | 260 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 261 | &extent_start, &extent_end, |
11833d66 | 262 | EXTENT_DIRTY | EXTENT_UPTODATE); |
0f9dd46c JB |
263 | if (ret) |
264 | break; | |
265 | ||
06b2331f | 266 | if (extent_start <= start) { |
0f9dd46c JB |
267 | start = extent_end + 1; |
268 | } else if (extent_start > start && extent_start < end) { | |
269 | size = extent_start - start; | |
817d52f8 | 270 | total_added += size; |
ea6a478e JB |
271 | ret = btrfs_add_free_space(block_group, start, |
272 | size); | |
0f9dd46c JB |
273 | BUG_ON(ret); |
274 | start = extent_end + 1; | |
275 | } else { | |
276 | break; | |
277 | } | |
278 | } | |
279 | ||
280 | if (start < end) { | |
281 | size = end - start; | |
817d52f8 | 282 | total_added += size; |
ea6a478e | 283 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
284 | BUG_ON(ret); |
285 | } | |
286 | ||
817d52f8 | 287 | return total_added; |
0f9dd46c JB |
288 | } |
289 | ||
817d52f8 | 290 | static int caching_kthread(void *data) |
e37c9e69 | 291 | { |
817d52f8 JB |
292 | struct btrfs_block_group_cache *block_group = data; |
293 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
11833d66 YZ |
294 | struct btrfs_caching_control *caching_ctl = block_group->caching_ctl; |
295 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 296 | struct btrfs_path *path; |
5f39d397 | 297 | struct extent_buffer *leaf; |
11833d66 | 298 | struct btrfs_key key; |
817d52f8 | 299 | u64 total_found = 0; |
11833d66 YZ |
300 | u64 last = 0; |
301 | u32 nritems; | |
302 | int ret = 0; | |
f510cfec | 303 | |
e37c9e69 CM |
304 | path = btrfs_alloc_path(); |
305 | if (!path) | |
306 | return -ENOMEM; | |
7d7d6068 | 307 | |
11833d66 | 308 | exclude_super_stripes(extent_root, block_group); |
1b2da372 JB |
309 | spin_lock(&block_group->space_info->lock); |
310 | block_group->space_info->bytes_super += block_group->bytes_super; | |
311 | spin_unlock(&block_group->space_info->lock); | |
11833d66 | 312 | |
817d52f8 | 313 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 314 | |
5cd57b2c | 315 | /* |
817d52f8 JB |
316 | * We don't want to deadlock with somebody trying to allocate a new |
317 | * extent for the extent root while also trying to search the extent | |
318 | * root to add free space. So we skip locking and search the commit | |
319 | * root, since its read-only | |
5cd57b2c CM |
320 | */ |
321 | path->skip_locking = 1; | |
817d52f8 JB |
322 | path->search_commit_root = 1; |
323 | path->reada = 2; | |
324 | ||
e4404d6e | 325 | key.objectid = last; |
e37c9e69 | 326 | key.offset = 0; |
11833d66 | 327 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 328 | again: |
11833d66 | 329 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
330 | /* need to make sure the commit_root doesn't disappear */ |
331 | down_read(&fs_info->extent_commit_sem); | |
332 | ||
11833d66 | 333 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 334 | if (ret < 0) |
ef8bbdfe | 335 | goto err; |
a512bbf8 | 336 | |
11833d66 YZ |
337 | leaf = path->nodes[0]; |
338 | nritems = btrfs_header_nritems(leaf); | |
339 | ||
d397712b | 340 | while (1) { |
817d52f8 | 341 | smp_mb(); |
11833d66 | 342 | if (fs_info->closing > 1) { |
f25784b3 | 343 | last = (u64)-1; |
817d52f8 | 344 | break; |
f25784b3 | 345 | } |
817d52f8 | 346 | |
11833d66 YZ |
347 | if (path->slots[0] < nritems) { |
348 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
349 | } else { | |
350 | ret = find_next_key(path, 0, &key); | |
351 | if (ret) | |
e37c9e69 | 352 | break; |
817d52f8 | 353 | |
11833d66 YZ |
354 | caching_ctl->progress = last; |
355 | btrfs_release_path(extent_root, path); | |
356 | up_read(&fs_info->extent_commit_sem); | |
357 | mutex_unlock(&caching_ctl->mutex); | |
358 | if (btrfs_transaction_in_commit(fs_info)) | |
f36f3042 | 359 | schedule_timeout(1); |
11833d66 YZ |
360 | else |
361 | cond_resched(); | |
362 | goto again; | |
363 | } | |
817d52f8 | 364 | |
11833d66 YZ |
365 | if (key.objectid < block_group->key.objectid) { |
366 | path->slots[0]++; | |
817d52f8 | 367 | continue; |
e37c9e69 | 368 | } |
0f9dd46c | 369 | |
e37c9e69 | 370 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 371 | block_group->key.offset) |
e37c9e69 | 372 | break; |
7d7d6068 | 373 | |
11833d66 | 374 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
375 | total_found += add_new_free_space(block_group, |
376 | fs_info, last, | |
377 | key.objectid); | |
7d7d6068 | 378 | last = key.objectid + key.offset; |
817d52f8 | 379 | |
11833d66 YZ |
380 | if (total_found > (1024 * 1024 * 2)) { |
381 | total_found = 0; | |
382 | wake_up(&caching_ctl->wait); | |
383 | } | |
817d52f8 | 384 | } |
e37c9e69 CM |
385 | path->slots[0]++; |
386 | } | |
817d52f8 | 387 | ret = 0; |
e37c9e69 | 388 | |
817d52f8 JB |
389 | total_found += add_new_free_space(block_group, fs_info, last, |
390 | block_group->key.objectid + | |
391 | block_group->key.offset); | |
11833d66 | 392 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
393 | |
394 | spin_lock(&block_group->lock); | |
11833d66 | 395 | block_group->caching_ctl = NULL; |
817d52f8 JB |
396 | block_group->cached = BTRFS_CACHE_FINISHED; |
397 | spin_unlock(&block_group->lock); | |
0f9dd46c | 398 | |
54aa1f4d | 399 | err: |
e37c9e69 | 400 | btrfs_free_path(path); |
276e680d | 401 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 402 | |
11833d66 YZ |
403 | free_excluded_extents(extent_root, block_group); |
404 | ||
405 | mutex_unlock(&caching_ctl->mutex); | |
406 | wake_up(&caching_ctl->wait); | |
407 | ||
408 | put_caching_control(caching_ctl); | |
409 | atomic_dec(&block_group->space_info->caching_threads); | |
817d52f8 JB |
410 | return 0; |
411 | } | |
412 | ||
413 | static int cache_block_group(struct btrfs_block_group_cache *cache) | |
414 | { | |
11833d66 YZ |
415 | struct btrfs_fs_info *fs_info = cache->fs_info; |
416 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
417 | struct task_struct *tsk; |
418 | int ret = 0; | |
419 | ||
11833d66 YZ |
420 | smp_mb(); |
421 | if (cache->cached != BTRFS_CACHE_NO) | |
422 | return 0; | |
423 | ||
424 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL); | |
425 | BUG_ON(!caching_ctl); | |
426 | ||
427 | INIT_LIST_HEAD(&caching_ctl->list); | |
428 | mutex_init(&caching_ctl->mutex); | |
429 | init_waitqueue_head(&caching_ctl->wait); | |
430 | caching_ctl->block_group = cache; | |
431 | caching_ctl->progress = cache->key.objectid; | |
432 | /* one for caching kthread, one for caching block group list */ | |
433 | atomic_set(&caching_ctl->count, 2); | |
434 | ||
817d52f8 JB |
435 | spin_lock(&cache->lock); |
436 | if (cache->cached != BTRFS_CACHE_NO) { | |
437 | spin_unlock(&cache->lock); | |
11833d66 YZ |
438 | kfree(caching_ctl); |
439 | return 0; | |
817d52f8 | 440 | } |
11833d66 | 441 | cache->caching_ctl = caching_ctl; |
817d52f8 JB |
442 | cache->cached = BTRFS_CACHE_STARTED; |
443 | spin_unlock(&cache->lock); | |
444 | ||
11833d66 YZ |
445 | down_write(&fs_info->extent_commit_sem); |
446 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
447 | up_write(&fs_info->extent_commit_sem); | |
448 | ||
449 | atomic_inc(&cache->space_info->caching_threads); | |
450 | ||
817d52f8 JB |
451 | tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n", |
452 | cache->key.objectid); | |
453 | if (IS_ERR(tsk)) { | |
454 | ret = PTR_ERR(tsk); | |
455 | printk(KERN_ERR "error running thread %d\n", ret); | |
456 | BUG(); | |
457 | } | |
458 | ||
ef8bbdfe | 459 | return ret; |
e37c9e69 CM |
460 | } |
461 | ||
0f9dd46c JB |
462 | /* |
463 | * return the block group that starts at or after bytenr | |
464 | */ | |
d397712b CM |
465 | static struct btrfs_block_group_cache * |
466 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 467 | { |
0f9dd46c | 468 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 469 | |
0f9dd46c | 470 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 471 | |
0f9dd46c | 472 | return cache; |
0ef3e66b CM |
473 | } |
474 | ||
0f9dd46c | 475 | /* |
9f55684c | 476 | * return the block group that contains the given bytenr |
0f9dd46c | 477 | */ |
d397712b CM |
478 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
479 | struct btrfs_fs_info *info, | |
480 | u64 bytenr) | |
be744175 | 481 | { |
0f9dd46c | 482 | struct btrfs_block_group_cache *cache; |
be744175 | 483 | |
0f9dd46c | 484 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 485 | |
0f9dd46c | 486 | return cache; |
be744175 | 487 | } |
0b86a832 | 488 | |
fa9c0d79 | 489 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) |
d2fb3437 YZ |
490 | { |
491 | if (atomic_dec_and_test(&cache->count)) | |
492 | kfree(cache); | |
493 | } | |
494 | ||
0f9dd46c JB |
495 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
496 | u64 flags) | |
6324fbf3 | 497 | { |
0f9dd46c | 498 | struct list_head *head = &info->space_info; |
0f9dd46c | 499 | struct btrfs_space_info *found; |
4184ea7f CM |
500 | |
501 | rcu_read_lock(); | |
502 | list_for_each_entry_rcu(found, head, list) { | |
503 | if (found->flags == flags) { | |
504 | rcu_read_unlock(); | |
0f9dd46c | 505 | return found; |
4184ea7f | 506 | } |
0f9dd46c | 507 | } |
4184ea7f | 508 | rcu_read_unlock(); |
0f9dd46c | 509 | return NULL; |
6324fbf3 CM |
510 | } |
511 | ||
4184ea7f CM |
512 | /* |
513 | * after adding space to the filesystem, we need to clear the full flags | |
514 | * on all the space infos. | |
515 | */ | |
516 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
517 | { | |
518 | struct list_head *head = &info->space_info; | |
519 | struct btrfs_space_info *found; | |
520 | ||
521 | rcu_read_lock(); | |
522 | list_for_each_entry_rcu(found, head, list) | |
523 | found->full = 0; | |
524 | rcu_read_unlock(); | |
525 | } | |
526 | ||
80eb234a JB |
527 | static u64 div_factor(u64 num, int factor) |
528 | { | |
529 | if (factor == 10) | |
530 | return num; | |
531 | num *= factor; | |
532 | do_div(num, 10); | |
533 | return num; | |
534 | } | |
535 | ||
d2fb3437 YZ |
536 | u64 btrfs_find_block_group(struct btrfs_root *root, |
537 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 538 | { |
96b5179d | 539 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 540 | u64 used; |
d2fb3437 YZ |
541 | u64 last = max(search_hint, search_start); |
542 | u64 group_start = 0; | |
31f3c99b | 543 | int full_search = 0; |
d2fb3437 | 544 | int factor = 9; |
0ef3e66b | 545 | int wrapped = 0; |
31f3c99b | 546 | again: |
e8569813 ZY |
547 | while (1) { |
548 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
549 | if (!cache) |
550 | break; | |
96b5179d | 551 | |
c286ac48 | 552 | spin_lock(&cache->lock); |
96b5179d CM |
553 | last = cache->key.objectid + cache->key.offset; |
554 | used = btrfs_block_group_used(&cache->item); | |
555 | ||
d2fb3437 YZ |
556 | if ((full_search || !cache->ro) && |
557 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 558 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
559 | div_factor(cache->key.offset, factor)) { |
560 | group_start = cache->key.objectid; | |
c286ac48 | 561 | spin_unlock(&cache->lock); |
fa9c0d79 | 562 | btrfs_put_block_group(cache); |
8790d502 CM |
563 | goto found; |
564 | } | |
6324fbf3 | 565 | } |
c286ac48 | 566 | spin_unlock(&cache->lock); |
fa9c0d79 | 567 | btrfs_put_block_group(cache); |
de428b63 | 568 | cond_resched(); |
cd1bc465 | 569 | } |
0ef3e66b CM |
570 | if (!wrapped) { |
571 | last = search_start; | |
572 | wrapped = 1; | |
573 | goto again; | |
574 | } | |
575 | if (!full_search && factor < 10) { | |
be744175 | 576 | last = search_start; |
31f3c99b | 577 | full_search = 1; |
0ef3e66b | 578 | factor = 10; |
31f3c99b CM |
579 | goto again; |
580 | } | |
be744175 | 581 | found: |
d2fb3437 | 582 | return group_start; |
925baedd | 583 | } |
0f9dd46c | 584 | |
e02119d5 | 585 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 586 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
587 | { |
588 | int ret; | |
589 | struct btrfs_key key; | |
31840ae1 | 590 | struct btrfs_path *path; |
e02119d5 | 591 | |
31840ae1 ZY |
592 | path = btrfs_alloc_path(); |
593 | BUG_ON(!path); | |
e02119d5 CM |
594 | key.objectid = start; |
595 | key.offset = len; | |
596 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
597 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
598 | 0, 0); | |
31840ae1 | 599 | btrfs_free_path(path); |
7bb86316 CM |
600 | return ret; |
601 | } | |
602 | ||
d8d5f3e1 CM |
603 | /* |
604 | * Back reference rules. Back refs have three main goals: | |
605 | * | |
606 | * 1) differentiate between all holders of references to an extent so that | |
607 | * when a reference is dropped we can make sure it was a valid reference | |
608 | * before freeing the extent. | |
609 | * | |
610 | * 2) Provide enough information to quickly find the holders of an extent | |
611 | * if we notice a given block is corrupted or bad. | |
612 | * | |
613 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
614 | * maintenance. This is actually the same as #2, but with a slightly | |
615 | * different use case. | |
616 | * | |
5d4f98a2 YZ |
617 | * There are two kinds of back refs. The implicit back refs is optimized |
618 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
619 | * back refs of this kind provide information about the block's owner tree | |
620 | * and the pointer's key. These information allow us to find the block by | |
621 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
622 | * referenced by their owner trees. The location of tree block is recorded | |
623 | * in the back refs. Actually the full back refs is generic, and can be | |
624 | * used in all cases the implicit back refs is used. The major shortcoming | |
625 | * of the full back refs is its overhead. Every time a tree block gets | |
626 | * COWed, we have to update back refs entry for all pointers in it. | |
627 | * | |
628 | * For a newly allocated tree block, we use implicit back refs for | |
629 | * pointers in it. This means most tree related operations only involve | |
630 | * implicit back refs. For a tree block created in old transaction, the | |
631 | * only way to drop a reference to it is COW it. So we can detect the | |
632 | * event that tree block loses its owner tree's reference and do the | |
633 | * back refs conversion. | |
634 | * | |
635 | * When a tree block is COW'd through a tree, there are four cases: | |
636 | * | |
637 | * The reference count of the block is one and the tree is the block's | |
638 | * owner tree. Nothing to do in this case. | |
639 | * | |
640 | * The reference count of the block is one and the tree is not the | |
641 | * block's owner tree. In this case, full back refs is used for pointers | |
642 | * in the block. Remove these full back refs, add implicit back refs for | |
643 | * every pointers in the new block. | |
644 | * | |
645 | * The reference count of the block is greater than one and the tree is | |
646 | * the block's owner tree. In this case, implicit back refs is used for | |
647 | * pointers in the block. Add full back refs for every pointers in the | |
648 | * block, increase lower level extents' reference counts. The original | |
649 | * implicit back refs are entailed to the new block. | |
650 | * | |
651 | * The reference count of the block is greater than one and the tree is | |
652 | * not the block's owner tree. Add implicit back refs for every pointer in | |
653 | * the new block, increase lower level extents' reference count. | |
654 | * | |
655 | * Back Reference Key composing: | |
656 | * | |
657 | * The key objectid corresponds to the first byte in the extent, | |
658 | * The key type is used to differentiate between types of back refs. | |
659 | * There are different meanings of the key offset for different types | |
660 | * of back refs. | |
661 | * | |
d8d5f3e1 CM |
662 | * File extents can be referenced by: |
663 | * | |
664 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 665 | * - different files inside a single subvolume |
d8d5f3e1 CM |
666 | * - different offsets inside a file (bookend extents in file.c) |
667 | * | |
5d4f98a2 | 668 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
669 | * |
670 | * - Objectid of the subvolume root | |
d8d5f3e1 | 671 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
672 | * - original offset in the file |
673 | * - how many bookend extents | |
d8d5f3e1 | 674 | * |
5d4f98a2 YZ |
675 | * The key offset for the implicit back refs is hash of the first |
676 | * three fields. | |
d8d5f3e1 | 677 | * |
5d4f98a2 | 678 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 679 | * |
5d4f98a2 | 680 | * - number of pointers in the tree leaf |
d8d5f3e1 | 681 | * |
5d4f98a2 YZ |
682 | * The key offset for the implicit back refs is the first byte of |
683 | * the tree leaf | |
d8d5f3e1 | 684 | * |
5d4f98a2 YZ |
685 | * When a file extent is allocated, The implicit back refs is used. |
686 | * the fields are filled in: | |
d8d5f3e1 | 687 | * |
5d4f98a2 | 688 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 689 | * |
5d4f98a2 YZ |
690 | * When a file extent is removed file truncation, we find the |
691 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 692 | * |
5d4f98a2 | 693 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 694 | * |
5d4f98a2 | 695 | * Btree extents can be referenced by: |
d8d5f3e1 | 696 | * |
5d4f98a2 | 697 | * - Different subvolumes |
d8d5f3e1 | 698 | * |
5d4f98a2 YZ |
699 | * Both the implicit back refs and the full back refs for tree blocks |
700 | * only consist of key. The key offset for the implicit back refs is | |
701 | * objectid of block's owner tree. The key offset for the full back refs | |
702 | * is the first byte of parent block. | |
d8d5f3e1 | 703 | * |
5d4f98a2 YZ |
704 | * When implicit back refs is used, information about the lowest key and |
705 | * level of the tree block are required. These information are stored in | |
706 | * tree block info structure. | |
d8d5f3e1 | 707 | */ |
31840ae1 | 708 | |
5d4f98a2 YZ |
709 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
710 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
711 | struct btrfs_root *root, | |
712 | struct btrfs_path *path, | |
713 | u64 owner, u32 extra_size) | |
7bb86316 | 714 | { |
5d4f98a2 YZ |
715 | struct btrfs_extent_item *item; |
716 | struct btrfs_extent_item_v0 *ei0; | |
717 | struct btrfs_extent_ref_v0 *ref0; | |
718 | struct btrfs_tree_block_info *bi; | |
719 | struct extent_buffer *leaf; | |
7bb86316 | 720 | struct btrfs_key key; |
5d4f98a2 YZ |
721 | struct btrfs_key found_key; |
722 | u32 new_size = sizeof(*item); | |
723 | u64 refs; | |
724 | int ret; | |
725 | ||
726 | leaf = path->nodes[0]; | |
727 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
728 | ||
729 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
730 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
731 | struct btrfs_extent_item_v0); | |
732 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
733 | ||
734 | if (owner == (u64)-1) { | |
735 | while (1) { | |
736 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
737 | ret = btrfs_next_leaf(root, path); | |
738 | if (ret < 0) | |
739 | return ret; | |
740 | BUG_ON(ret > 0); | |
741 | leaf = path->nodes[0]; | |
742 | } | |
743 | btrfs_item_key_to_cpu(leaf, &found_key, | |
744 | path->slots[0]); | |
745 | BUG_ON(key.objectid != found_key.objectid); | |
746 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
747 | path->slots[0]++; | |
748 | continue; | |
749 | } | |
750 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
751 | struct btrfs_extent_ref_v0); | |
752 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
753 | break; | |
754 | } | |
755 | } | |
756 | btrfs_release_path(root, path); | |
757 | ||
758 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
759 | new_size += sizeof(*bi); | |
760 | ||
761 | new_size -= sizeof(*ei0); | |
762 | ret = btrfs_search_slot(trans, root, &key, path, | |
763 | new_size + extra_size, 1); | |
764 | if (ret < 0) | |
765 | return ret; | |
766 | BUG_ON(ret); | |
767 | ||
768 | ret = btrfs_extend_item(trans, root, path, new_size); | |
769 | BUG_ON(ret); | |
770 | ||
771 | leaf = path->nodes[0]; | |
772 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
773 | btrfs_set_extent_refs(leaf, item, refs); | |
774 | /* FIXME: get real generation */ | |
775 | btrfs_set_extent_generation(leaf, item, 0); | |
776 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
777 | btrfs_set_extent_flags(leaf, item, | |
778 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
779 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
780 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
781 | /* FIXME: get first key of the block */ | |
782 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
783 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
784 | } else { | |
785 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
786 | } | |
787 | btrfs_mark_buffer_dirty(leaf); | |
788 | return 0; | |
789 | } | |
790 | #endif | |
791 | ||
792 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
793 | { | |
794 | u32 high_crc = ~(u32)0; | |
795 | u32 low_crc = ~(u32)0; | |
796 | __le64 lenum; | |
797 | ||
798 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 799 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 800 | lenum = cpu_to_le64(owner); |
163e783e | 801 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 802 | lenum = cpu_to_le64(offset); |
163e783e | 803 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
804 | |
805 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
806 | } | |
807 | ||
808 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
809 | struct btrfs_extent_data_ref *ref) | |
810 | { | |
811 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
812 | btrfs_extent_data_ref_objectid(leaf, ref), | |
813 | btrfs_extent_data_ref_offset(leaf, ref)); | |
814 | } | |
815 | ||
816 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
817 | struct btrfs_extent_data_ref *ref, | |
818 | u64 root_objectid, u64 owner, u64 offset) | |
819 | { | |
820 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
821 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
822 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
823 | return 0; | |
824 | return 1; | |
825 | } | |
826 | ||
827 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
828 | struct btrfs_root *root, | |
829 | struct btrfs_path *path, | |
830 | u64 bytenr, u64 parent, | |
831 | u64 root_objectid, | |
832 | u64 owner, u64 offset) | |
833 | { | |
834 | struct btrfs_key key; | |
835 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 836 | struct extent_buffer *leaf; |
5d4f98a2 | 837 | u32 nritems; |
74493f7a | 838 | int ret; |
5d4f98a2 YZ |
839 | int recow; |
840 | int err = -ENOENT; | |
74493f7a | 841 | |
31840ae1 | 842 | key.objectid = bytenr; |
5d4f98a2 YZ |
843 | if (parent) { |
844 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
845 | key.offset = parent; | |
846 | } else { | |
847 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
848 | key.offset = hash_extent_data_ref(root_objectid, | |
849 | owner, offset); | |
850 | } | |
851 | again: | |
852 | recow = 0; | |
853 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
854 | if (ret < 0) { | |
855 | err = ret; | |
856 | goto fail; | |
857 | } | |
31840ae1 | 858 | |
5d4f98a2 YZ |
859 | if (parent) { |
860 | if (!ret) | |
861 | return 0; | |
862 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
863 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
864 | btrfs_release_path(root, path); | |
865 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
866 | if (ret < 0) { | |
867 | err = ret; | |
868 | goto fail; | |
869 | } | |
870 | if (!ret) | |
871 | return 0; | |
872 | #endif | |
873 | goto fail; | |
31840ae1 ZY |
874 | } |
875 | ||
876 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
877 | nritems = btrfs_header_nritems(leaf); |
878 | while (1) { | |
879 | if (path->slots[0] >= nritems) { | |
880 | ret = btrfs_next_leaf(root, path); | |
881 | if (ret < 0) | |
882 | err = ret; | |
883 | if (ret) | |
884 | goto fail; | |
885 | ||
886 | leaf = path->nodes[0]; | |
887 | nritems = btrfs_header_nritems(leaf); | |
888 | recow = 1; | |
889 | } | |
890 | ||
891 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
892 | if (key.objectid != bytenr || | |
893 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
894 | goto fail; | |
895 | ||
896 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
897 | struct btrfs_extent_data_ref); | |
898 | ||
899 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
900 | owner, offset)) { | |
901 | if (recow) { | |
902 | btrfs_release_path(root, path); | |
903 | goto again; | |
904 | } | |
905 | err = 0; | |
906 | break; | |
907 | } | |
908 | path->slots[0]++; | |
31840ae1 | 909 | } |
5d4f98a2 YZ |
910 | fail: |
911 | return err; | |
31840ae1 ZY |
912 | } |
913 | ||
5d4f98a2 YZ |
914 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
915 | struct btrfs_root *root, | |
916 | struct btrfs_path *path, | |
917 | u64 bytenr, u64 parent, | |
918 | u64 root_objectid, u64 owner, | |
919 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
920 | { |
921 | struct btrfs_key key; | |
922 | struct extent_buffer *leaf; | |
5d4f98a2 | 923 | u32 size; |
31840ae1 ZY |
924 | u32 num_refs; |
925 | int ret; | |
74493f7a | 926 | |
74493f7a | 927 | key.objectid = bytenr; |
5d4f98a2 YZ |
928 | if (parent) { |
929 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
930 | key.offset = parent; | |
931 | size = sizeof(struct btrfs_shared_data_ref); | |
932 | } else { | |
933 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
934 | key.offset = hash_extent_data_ref(root_objectid, | |
935 | owner, offset); | |
936 | size = sizeof(struct btrfs_extent_data_ref); | |
937 | } | |
74493f7a | 938 | |
5d4f98a2 YZ |
939 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
940 | if (ret && ret != -EEXIST) | |
941 | goto fail; | |
942 | ||
943 | leaf = path->nodes[0]; | |
944 | if (parent) { | |
945 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 946 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
947 | struct btrfs_shared_data_ref); |
948 | if (ret == 0) { | |
949 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
950 | } else { | |
951 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
952 | num_refs += refs_to_add; | |
953 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 954 | } |
5d4f98a2 YZ |
955 | } else { |
956 | struct btrfs_extent_data_ref *ref; | |
957 | while (ret == -EEXIST) { | |
958 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
959 | struct btrfs_extent_data_ref); | |
960 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
961 | owner, offset)) | |
962 | break; | |
963 | btrfs_release_path(root, path); | |
964 | key.offset++; | |
965 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
966 | size); | |
967 | if (ret && ret != -EEXIST) | |
968 | goto fail; | |
31840ae1 | 969 | |
5d4f98a2 YZ |
970 | leaf = path->nodes[0]; |
971 | } | |
972 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
973 | struct btrfs_extent_data_ref); | |
974 | if (ret == 0) { | |
975 | btrfs_set_extent_data_ref_root(leaf, ref, | |
976 | root_objectid); | |
977 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
978 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
979 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
980 | } else { | |
981 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
982 | num_refs += refs_to_add; | |
983 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 984 | } |
31840ae1 | 985 | } |
5d4f98a2 YZ |
986 | btrfs_mark_buffer_dirty(leaf); |
987 | ret = 0; | |
988 | fail: | |
7bb86316 CM |
989 | btrfs_release_path(root, path); |
990 | return ret; | |
74493f7a CM |
991 | } |
992 | ||
5d4f98a2 YZ |
993 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
994 | struct btrfs_root *root, | |
995 | struct btrfs_path *path, | |
996 | int refs_to_drop) | |
31840ae1 | 997 | { |
5d4f98a2 YZ |
998 | struct btrfs_key key; |
999 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1000 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1001 | struct extent_buffer *leaf; |
5d4f98a2 | 1002 | u32 num_refs = 0; |
31840ae1 ZY |
1003 | int ret = 0; |
1004 | ||
1005 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1006 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1007 | ||
1008 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1009 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1010 | struct btrfs_extent_data_ref); | |
1011 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1012 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1013 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1014 | struct btrfs_shared_data_ref); | |
1015 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1016 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1017 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1018 | struct btrfs_extent_ref_v0 *ref0; | |
1019 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1020 | struct btrfs_extent_ref_v0); | |
1021 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1022 | #endif | |
1023 | } else { | |
1024 | BUG(); | |
1025 | } | |
1026 | ||
56bec294 CM |
1027 | BUG_ON(num_refs < refs_to_drop); |
1028 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1029 | |
31840ae1 ZY |
1030 | if (num_refs == 0) { |
1031 | ret = btrfs_del_item(trans, root, path); | |
1032 | } else { | |
5d4f98a2 YZ |
1033 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1034 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1035 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1036 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1037 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1038 | else { | |
1039 | struct btrfs_extent_ref_v0 *ref0; | |
1040 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1041 | struct btrfs_extent_ref_v0); | |
1042 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1043 | } | |
1044 | #endif | |
31840ae1 ZY |
1045 | btrfs_mark_buffer_dirty(leaf); |
1046 | } | |
31840ae1 ZY |
1047 | return ret; |
1048 | } | |
1049 | ||
5d4f98a2 YZ |
1050 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1051 | struct btrfs_path *path, | |
1052 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1053 | { |
5d4f98a2 YZ |
1054 | struct btrfs_key key; |
1055 | struct extent_buffer *leaf; | |
1056 | struct btrfs_extent_data_ref *ref1; | |
1057 | struct btrfs_shared_data_ref *ref2; | |
1058 | u32 num_refs = 0; | |
1059 | ||
1060 | leaf = path->nodes[0]; | |
1061 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1062 | if (iref) { | |
1063 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1064 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1065 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1066 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1067 | } else { | |
1068 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1069 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1070 | } | |
1071 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1072 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1073 | struct btrfs_extent_data_ref); | |
1074 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1075 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1076 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1077 | struct btrfs_shared_data_ref); | |
1078 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1079 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1080 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1081 | struct btrfs_extent_ref_v0 *ref0; | |
1082 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1083 | struct btrfs_extent_ref_v0); | |
1084 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1085 | #endif |
5d4f98a2 YZ |
1086 | } else { |
1087 | WARN_ON(1); | |
1088 | } | |
1089 | return num_refs; | |
1090 | } | |
15916de8 | 1091 | |
5d4f98a2 YZ |
1092 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1093 | struct btrfs_root *root, | |
1094 | struct btrfs_path *path, | |
1095 | u64 bytenr, u64 parent, | |
1096 | u64 root_objectid) | |
1f3c79a2 | 1097 | { |
5d4f98a2 | 1098 | struct btrfs_key key; |
1f3c79a2 | 1099 | int ret; |
1f3c79a2 | 1100 | |
5d4f98a2 YZ |
1101 | key.objectid = bytenr; |
1102 | if (parent) { | |
1103 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1104 | key.offset = parent; | |
1105 | } else { | |
1106 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1107 | key.offset = root_objectid; | |
1f3c79a2 LH |
1108 | } |
1109 | ||
5d4f98a2 YZ |
1110 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1111 | if (ret > 0) | |
1112 | ret = -ENOENT; | |
1113 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1114 | if (ret == -ENOENT && parent) { | |
1115 | btrfs_release_path(root, path); | |
1116 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
1117 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1118 | if (ret > 0) | |
1119 | ret = -ENOENT; | |
1120 | } | |
1f3c79a2 | 1121 | #endif |
5d4f98a2 | 1122 | return ret; |
1f3c79a2 LH |
1123 | } |
1124 | ||
5d4f98a2 YZ |
1125 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1126 | struct btrfs_root *root, | |
1127 | struct btrfs_path *path, | |
1128 | u64 bytenr, u64 parent, | |
1129 | u64 root_objectid) | |
31840ae1 | 1130 | { |
5d4f98a2 | 1131 | struct btrfs_key key; |
31840ae1 | 1132 | int ret; |
31840ae1 | 1133 | |
5d4f98a2 YZ |
1134 | key.objectid = bytenr; |
1135 | if (parent) { | |
1136 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1137 | key.offset = parent; | |
1138 | } else { | |
1139 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1140 | key.offset = root_objectid; | |
1141 | } | |
1142 | ||
1143 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1144 | btrfs_release_path(root, path); | |
31840ae1 ZY |
1145 | return ret; |
1146 | } | |
1147 | ||
5d4f98a2 | 1148 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1149 | { |
5d4f98a2 YZ |
1150 | int type; |
1151 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1152 | if (parent > 0) | |
1153 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1154 | else | |
1155 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1156 | } else { | |
1157 | if (parent > 0) | |
1158 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1159 | else | |
1160 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1161 | } | |
1162 | return type; | |
31840ae1 | 1163 | } |
56bec294 | 1164 | |
2c47e605 YZ |
1165 | static int find_next_key(struct btrfs_path *path, int level, |
1166 | struct btrfs_key *key) | |
56bec294 | 1167 | |
02217ed2 | 1168 | { |
2c47e605 | 1169 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1170 | if (!path->nodes[level]) |
1171 | break; | |
5d4f98a2 YZ |
1172 | if (path->slots[level] + 1 >= |
1173 | btrfs_header_nritems(path->nodes[level])) | |
1174 | continue; | |
1175 | if (level == 0) | |
1176 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1177 | path->slots[level] + 1); | |
1178 | else | |
1179 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1180 | path->slots[level] + 1); | |
1181 | return 0; | |
1182 | } | |
1183 | return 1; | |
1184 | } | |
037e6390 | 1185 | |
5d4f98a2 YZ |
1186 | /* |
1187 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1188 | * to the address of inline back ref, and 0 is returned. | |
1189 | * | |
1190 | * if back ref isn't found, *ref_ret is set to the address where it | |
1191 | * should be inserted, and -ENOENT is returned. | |
1192 | * | |
1193 | * if insert is true and there are too many inline back refs, the path | |
1194 | * points to the extent item, and -EAGAIN is returned. | |
1195 | * | |
1196 | * NOTE: inline back refs are ordered in the same way that back ref | |
1197 | * items in the tree are ordered. | |
1198 | */ | |
1199 | static noinline_for_stack | |
1200 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1201 | struct btrfs_root *root, | |
1202 | struct btrfs_path *path, | |
1203 | struct btrfs_extent_inline_ref **ref_ret, | |
1204 | u64 bytenr, u64 num_bytes, | |
1205 | u64 parent, u64 root_objectid, | |
1206 | u64 owner, u64 offset, int insert) | |
1207 | { | |
1208 | struct btrfs_key key; | |
1209 | struct extent_buffer *leaf; | |
1210 | struct btrfs_extent_item *ei; | |
1211 | struct btrfs_extent_inline_ref *iref; | |
1212 | u64 flags; | |
1213 | u64 item_size; | |
1214 | unsigned long ptr; | |
1215 | unsigned long end; | |
1216 | int extra_size; | |
1217 | int type; | |
1218 | int want; | |
1219 | int ret; | |
1220 | int err = 0; | |
26b8003f | 1221 | |
db94535d | 1222 | key.objectid = bytenr; |
31840ae1 | 1223 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1224 | key.offset = num_bytes; |
31840ae1 | 1225 | |
5d4f98a2 YZ |
1226 | want = extent_ref_type(parent, owner); |
1227 | if (insert) { | |
1228 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1229 | path->keep_locks = 1; |
5d4f98a2 YZ |
1230 | } else |
1231 | extra_size = -1; | |
1232 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1233 | if (ret < 0) { |
5d4f98a2 YZ |
1234 | err = ret; |
1235 | goto out; | |
1236 | } | |
1237 | BUG_ON(ret); | |
1238 | ||
1239 | leaf = path->nodes[0]; | |
1240 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1241 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1242 | if (item_size < sizeof(*ei)) { | |
1243 | if (!insert) { | |
1244 | err = -ENOENT; | |
1245 | goto out; | |
1246 | } | |
1247 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1248 | extra_size); | |
1249 | if (ret < 0) { | |
1250 | err = ret; | |
1251 | goto out; | |
1252 | } | |
1253 | leaf = path->nodes[0]; | |
1254 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1255 | } | |
1256 | #endif | |
1257 | BUG_ON(item_size < sizeof(*ei)); | |
1258 | ||
5d4f98a2 YZ |
1259 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1260 | flags = btrfs_extent_flags(leaf, ei); | |
1261 | ||
1262 | ptr = (unsigned long)(ei + 1); | |
1263 | end = (unsigned long)ei + item_size; | |
1264 | ||
1265 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1266 | ptr += sizeof(struct btrfs_tree_block_info); | |
1267 | BUG_ON(ptr > end); | |
1268 | } else { | |
1269 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1270 | } | |
1271 | ||
1272 | err = -ENOENT; | |
1273 | while (1) { | |
1274 | if (ptr >= end) { | |
1275 | WARN_ON(ptr > end); | |
1276 | break; | |
1277 | } | |
1278 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1279 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1280 | if (want < type) | |
1281 | break; | |
1282 | if (want > type) { | |
1283 | ptr += btrfs_extent_inline_ref_size(type); | |
1284 | continue; | |
1285 | } | |
1286 | ||
1287 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1288 | struct btrfs_extent_data_ref *dref; | |
1289 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1290 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1291 | owner, offset)) { | |
1292 | err = 0; | |
1293 | break; | |
1294 | } | |
1295 | if (hash_extent_data_ref_item(leaf, dref) < | |
1296 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1297 | break; | |
1298 | } else { | |
1299 | u64 ref_offset; | |
1300 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1301 | if (parent > 0) { | |
1302 | if (parent == ref_offset) { | |
1303 | err = 0; | |
1304 | break; | |
1305 | } | |
1306 | if (ref_offset < parent) | |
1307 | break; | |
1308 | } else { | |
1309 | if (root_objectid == ref_offset) { | |
1310 | err = 0; | |
1311 | break; | |
1312 | } | |
1313 | if (ref_offset < root_objectid) | |
1314 | break; | |
1315 | } | |
1316 | } | |
1317 | ptr += btrfs_extent_inline_ref_size(type); | |
1318 | } | |
1319 | if (err == -ENOENT && insert) { | |
1320 | if (item_size + extra_size >= | |
1321 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1322 | err = -EAGAIN; | |
1323 | goto out; | |
1324 | } | |
1325 | /* | |
1326 | * To add new inline back ref, we have to make sure | |
1327 | * there is no corresponding back ref item. | |
1328 | * For simplicity, we just do not add new inline back | |
1329 | * ref if there is any kind of item for this block | |
1330 | */ | |
2c47e605 YZ |
1331 | if (find_next_key(path, 0, &key) == 0 && |
1332 | key.objectid == bytenr && | |
85d4198e | 1333 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1334 | err = -EAGAIN; |
1335 | goto out; | |
1336 | } | |
1337 | } | |
1338 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1339 | out: | |
85d4198e | 1340 | if (insert) { |
5d4f98a2 YZ |
1341 | path->keep_locks = 0; |
1342 | btrfs_unlock_up_safe(path, 1); | |
1343 | } | |
1344 | return err; | |
1345 | } | |
1346 | ||
1347 | /* | |
1348 | * helper to add new inline back ref | |
1349 | */ | |
1350 | static noinline_for_stack | |
1351 | int setup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1352 | struct btrfs_root *root, | |
1353 | struct btrfs_path *path, | |
1354 | struct btrfs_extent_inline_ref *iref, | |
1355 | u64 parent, u64 root_objectid, | |
1356 | u64 owner, u64 offset, int refs_to_add, | |
1357 | struct btrfs_delayed_extent_op *extent_op) | |
1358 | { | |
1359 | struct extent_buffer *leaf; | |
1360 | struct btrfs_extent_item *ei; | |
1361 | unsigned long ptr; | |
1362 | unsigned long end; | |
1363 | unsigned long item_offset; | |
1364 | u64 refs; | |
1365 | int size; | |
1366 | int type; | |
1367 | int ret; | |
1368 | ||
1369 | leaf = path->nodes[0]; | |
1370 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1371 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1372 | ||
1373 | type = extent_ref_type(parent, owner); | |
1374 | size = btrfs_extent_inline_ref_size(type); | |
1375 | ||
1376 | ret = btrfs_extend_item(trans, root, path, size); | |
1377 | BUG_ON(ret); | |
1378 | ||
1379 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1380 | refs = btrfs_extent_refs(leaf, ei); | |
1381 | refs += refs_to_add; | |
1382 | btrfs_set_extent_refs(leaf, ei, refs); | |
1383 | if (extent_op) | |
1384 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1385 | ||
1386 | ptr = (unsigned long)ei + item_offset; | |
1387 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1388 | if (ptr < end - size) | |
1389 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1390 | end - size - ptr); | |
1391 | ||
1392 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1393 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1394 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1395 | struct btrfs_extent_data_ref *dref; | |
1396 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1397 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1398 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1399 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1400 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1401 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1402 | struct btrfs_shared_data_ref *sref; | |
1403 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1404 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1405 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1406 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1407 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1408 | } else { | |
1409 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1410 | } | |
1411 | btrfs_mark_buffer_dirty(leaf); | |
1412 | return 0; | |
1413 | } | |
1414 | ||
1415 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1416 | struct btrfs_root *root, | |
1417 | struct btrfs_path *path, | |
1418 | struct btrfs_extent_inline_ref **ref_ret, | |
1419 | u64 bytenr, u64 num_bytes, u64 parent, | |
1420 | u64 root_objectid, u64 owner, u64 offset) | |
1421 | { | |
1422 | int ret; | |
1423 | ||
1424 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1425 | bytenr, num_bytes, parent, | |
1426 | root_objectid, owner, offset, 0); | |
1427 | if (ret != -ENOENT) | |
54aa1f4d | 1428 | return ret; |
5d4f98a2 YZ |
1429 | |
1430 | btrfs_release_path(root, path); | |
1431 | *ref_ret = NULL; | |
1432 | ||
1433 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1434 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1435 | root_objectid); | |
1436 | } else { | |
1437 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1438 | root_objectid, owner, offset); | |
b9473439 | 1439 | } |
5d4f98a2 YZ |
1440 | return ret; |
1441 | } | |
31840ae1 | 1442 | |
5d4f98a2 YZ |
1443 | /* |
1444 | * helper to update/remove inline back ref | |
1445 | */ | |
1446 | static noinline_for_stack | |
1447 | int update_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1448 | struct btrfs_root *root, | |
1449 | struct btrfs_path *path, | |
1450 | struct btrfs_extent_inline_ref *iref, | |
1451 | int refs_to_mod, | |
1452 | struct btrfs_delayed_extent_op *extent_op) | |
1453 | { | |
1454 | struct extent_buffer *leaf; | |
1455 | struct btrfs_extent_item *ei; | |
1456 | struct btrfs_extent_data_ref *dref = NULL; | |
1457 | struct btrfs_shared_data_ref *sref = NULL; | |
1458 | unsigned long ptr; | |
1459 | unsigned long end; | |
1460 | u32 item_size; | |
1461 | int size; | |
1462 | int type; | |
1463 | int ret; | |
1464 | u64 refs; | |
1465 | ||
1466 | leaf = path->nodes[0]; | |
1467 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1468 | refs = btrfs_extent_refs(leaf, ei); | |
1469 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1470 | refs += refs_to_mod; | |
1471 | btrfs_set_extent_refs(leaf, ei, refs); | |
1472 | if (extent_op) | |
1473 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1474 | ||
1475 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1476 | ||
1477 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1478 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1479 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1480 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1481 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1482 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1483 | } else { | |
1484 | refs = 1; | |
1485 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1486 | } |
31840ae1 | 1487 | |
5d4f98a2 YZ |
1488 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1489 | refs += refs_to_mod; | |
1490 | ||
1491 | if (refs > 0) { | |
1492 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1493 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1494 | else | |
1495 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1496 | } else { | |
1497 | size = btrfs_extent_inline_ref_size(type); | |
1498 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1499 | ptr = (unsigned long)iref; | |
1500 | end = (unsigned long)ei + item_size; | |
1501 | if (ptr + size < end) | |
1502 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1503 | end - ptr - size); | |
1504 | item_size -= size; | |
1505 | ret = btrfs_truncate_item(trans, root, path, item_size, 1); | |
1506 | BUG_ON(ret); | |
1507 | } | |
1508 | btrfs_mark_buffer_dirty(leaf); | |
1509 | return 0; | |
1510 | } | |
1511 | ||
1512 | static noinline_for_stack | |
1513 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1514 | struct btrfs_root *root, | |
1515 | struct btrfs_path *path, | |
1516 | u64 bytenr, u64 num_bytes, u64 parent, | |
1517 | u64 root_objectid, u64 owner, | |
1518 | u64 offset, int refs_to_add, | |
1519 | struct btrfs_delayed_extent_op *extent_op) | |
1520 | { | |
1521 | struct btrfs_extent_inline_ref *iref; | |
1522 | int ret; | |
1523 | ||
1524 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1525 | bytenr, num_bytes, parent, | |
1526 | root_objectid, owner, offset, 1); | |
1527 | if (ret == 0) { | |
1528 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
1529 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1530 | refs_to_add, extent_op); | |
1531 | } else if (ret == -ENOENT) { | |
1532 | ret = setup_inline_extent_backref(trans, root, path, iref, | |
1533 | parent, root_objectid, | |
1534 | owner, offset, refs_to_add, | |
1535 | extent_op); | |
771ed689 | 1536 | } |
5d4f98a2 YZ |
1537 | return ret; |
1538 | } | |
31840ae1 | 1539 | |
5d4f98a2 YZ |
1540 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1541 | struct btrfs_root *root, | |
1542 | struct btrfs_path *path, | |
1543 | u64 bytenr, u64 parent, u64 root_objectid, | |
1544 | u64 owner, u64 offset, int refs_to_add) | |
1545 | { | |
1546 | int ret; | |
1547 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1548 | BUG_ON(refs_to_add != 1); | |
1549 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1550 | parent, root_objectid); | |
1551 | } else { | |
1552 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1553 | parent, root_objectid, | |
1554 | owner, offset, refs_to_add); | |
1555 | } | |
1556 | return ret; | |
1557 | } | |
56bec294 | 1558 | |
5d4f98a2 YZ |
1559 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1560 | struct btrfs_root *root, | |
1561 | struct btrfs_path *path, | |
1562 | struct btrfs_extent_inline_ref *iref, | |
1563 | int refs_to_drop, int is_data) | |
1564 | { | |
1565 | int ret; | |
b9473439 | 1566 | |
5d4f98a2 YZ |
1567 | BUG_ON(!is_data && refs_to_drop != 1); |
1568 | if (iref) { | |
1569 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1570 | -refs_to_drop, NULL); | |
1571 | } else if (is_data) { | |
1572 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1573 | } else { | |
1574 | ret = btrfs_del_item(trans, root, path); | |
1575 | } | |
1576 | return ret; | |
1577 | } | |
1578 | ||
5d4f98a2 YZ |
1579 | static void btrfs_issue_discard(struct block_device *bdev, |
1580 | u64 start, u64 len) | |
1581 | { | |
746cd1e7 CH |
1582 | blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL, |
1583 | DISCARD_FL_BARRIER); | |
5d4f98a2 | 1584 | } |
5d4f98a2 YZ |
1585 | |
1586 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
1587 | u64 num_bytes) | |
1588 | { | |
5d4f98a2 YZ |
1589 | int ret; |
1590 | u64 map_length = num_bytes; | |
1591 | struct btrfs_multi_bio *multi = NULL; | |
1592 | ||
e244a0ae CH |
1593 | if (!btrfs_test_opt(root, DISCARD)) |
1594 | return 0; | |
1595 | ||
5d4f98a2 YZ |
1596 | /* Tell the block device(s) that the sectors can be discarded */ |
1597 | ret = btrfs_map_block(&root->fs_info->mapping_tree, READ, | |
1598 | bytenr, &map_length, &multi, 0); | |
1599 | if (!ret) { | |
1600 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
1601 | int i; | |
1602 | ||
1603 | if (map_length > num_bytes) | |
1604 | map_length = num_bytes; | |
1605 | ||
1606 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
1607 | btrfs_issue_discard(stripe->dev->bdev, | |
1608 | stripe->physical, | |
1609 | map_length); | |
1610 | } | |
1611 | kfree(multi); | |
1612 | } | |
1613 | ||
1614 | return ret; | |
5d4f98a2 YZ |
1615 | } |
1616 | ||
1617 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1618 | struct btrfs_root *root, | |
1619 | u64 bytenr, u64 num_bytes, u64 parent, | |
1620 | u64 root_objectid, u64 owner, u64 offset) | |
1621 | { | |
1622 | int ret; | |
1623 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && | |
1624 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1625 | ||
1626 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1627 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
1628 | parent, root_objectid, (int)owner, | |
1629 | BTRFS_ADD_DELAYED_REF, NULL); | |
1630 | } else { | |
1631 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
1632 | parent, root_objectid, owner, offset, | |
1633 | BTRFS_ADD_DELAYED_REF, NULL); | |
1634 | } | |
1635 | return ret; | |
1636 | } | |
1637 | ||
1638 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1639 | struct btrfs_root *root, | |
1640 | u64 bytenr, u64 num_bytes, | |
1641 | u64 parent, u64 root_objectid, | |
1642 | u64 owner, u64 offset, int refs_to_add, | |
1643 | struct btrfs_delayed_extent_op *extent_op) | |
1644 | { | |
1645 | struct btrfs_path *path; | |
1646 | struct extent_buffer *leaf; | |
1647 | struct btrfs_extent_item *item; | |
1648 | u64 refs; | |
1649 | int ret; | |
1650 | int err = 0; | |
1651 | ||
1652 | path = btrfs_alloc_path(); | |
1653 | if (!path) | |
1654 | return -ENOMEM; | |
1655 | ||
1656 | path->reada = 1; | |
1657 | path->leave_spinning = 1; | |
1658 | /* this will setup the path even if it fails to insert the back ref */ | |
1659 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1660 | path, bytenr, num_bytes, parent, | |
1661 | root_objectid, owner, offset, | |
1662 | refs_to_add, extent_op); | |
1663 | if (ret == 0) | |
1664 | goto out; | |
1665 | ||
1666 | if (ret != -EAGAIN) { | |
1667 | err = ret; | |
1668 | goto out; | |
1669 | } | |
1670 | ||
1671 | leaf = path->nodes[0]; | |
1672 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1673 | refs = btrfs_extent_refs(leaf, item); | |
1674 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1675 | if (extent_op) | |
1676 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1677 | |
5d4f98a2 | 1678 | btrfs_mark_buffer_dirty(leaf); |
56bec294 CM |
1679 | btrfs_release_path(root->fs_info->extent_root, path); |
1680 | ||
1681 | path->reada = 1; | |
b9473439 CM |
1682 | path->leave_spinning = 1; |
1683 | ||
56bec294 CM |
1684 | /* now insert the actual backref */ |
1685 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1686 | path, bytenr, parent, root_objectid, |
1687 | owner, offset, refs_to_add); | |
56bec294 | 1688 | BUG_ON(ret); |
5d4f98a2 | 1689 | out: |
56bec294 | 1690 | btrfs_free_path(path); |
5d4f98a2 | 1691 | return err; |
56bec294 CM |
1692 | } |
1693 | ||
5d4f98a2 YZ |
1694 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1695 | struct btrfs_root *root, | |
1696 | struct btrfs_delayed_ref_node *node, | |
1697 | struct btrfs_delayed_extent_op *extent_op, | |
1698 | int insert_reserved) | |
56bec294 | 1699 | { |
5d4f98a2 YZ |
1700 | int ret = 0; |
1701 | struct btrfs_delayed_data_ref *ref; | |
1702 | struct btrfs_key ins; | |
1703 | u64 parent = 0; | |
1704 | u64 ref_root = 0; | |
1705 | u64 flags = 0; | |
1706 | ||
1707 | ins.objectid = node->bytenr; | |
1708 | ins.offset = node->num_bytes; | |
1709 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1710 | ||
1711 | ref = btrfs_delayed_node_to_data_ref(node); | |
1712 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1713 | parent = ref->parent; | |
1714 | else | |
1715 | ref_root = ref->root; | |
1716 | ||
1717 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1718 | if (extent_op) { | |
1719 | BUG_ON(extent_op->update_key); | |
1720 | flags |= extent_op->flags_to_set; | |
1721 | } | |
1722 | ret = alloc_reserved_file_extent(trans, root, | |
1723 | parent, ref_root, flags, | |
1724 | ref->objectid, ref->offset, | |
1725 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1726 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1727 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1728 | node->num_bytes, parent, | |
1729 | ref_root, ref->objectid, | |
1730 | ref->offset, node->ref_mod, | |
1731 | extent_op); | |
1732 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1733 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1734 | node->num_bytes, parent, | |
1735 | ref_root, ref->objectid, | |
1736 | ref->offset, node->ref_mod, | |
1737 | extent_op); | |
1738 | } else { | |
1739 | BUG(); | |
1740 | } | |
1741 | return ret; | |
1742 | } | |
1743 | ||
1744 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1745 | struct extent_buffer *leaf, | |
1746 | struct btrfs_extent_item *ei) | |
1747 | { | |
1748 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1749 | if (extent_op->update_flags) { | |
1750 | flags |= extent_op->flags_to_set; | |
1751 | btrfs_set_extent_flags(leaf, ei, flags); | |
1752 | } | |
1753 | ||
1754 | if (extent_op->update_key) { | |
1755 | struct btrfs_tree_block_info *bi; | |
1756 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1757 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1758 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1759 | } | |
1760 | } | |
1761 | ||
1762 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
1763 | struct btrfs_root *root, | |
1764 | struct btrfs_delayed_ref_node *node, | |
1765 | struct btrfs_delayed_extent_op *extent_op) | |
1766 | { | |
1767 | struct btrfs_key key; | |
1768 | struct btrfs_path *path; | |
1769 | struct btrfs_extent_item *ei; | |
1770 | struct extent_buffer *leaf; | |
1771 | u32 item_size; | |
56bec294 | 1772 | int ret; |
5d4f98a2 YZ |
1773 | int err = 0; |
1774 | ||
1775 | path = btrfs_alloc_path(); | |
1776 | if (!path) | |
1777 | return -ENOMEM; | |
1778 | ||
1779 | key.objectid = node->bytenr; | |
1780 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1781 | key.offset = node->num_bytes; | |
1782 | ||
1783 | path->reada = 1; | |
1784 | path->leave_spinning = 1; | |
1785 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
1786 | path, 0, 1); | |
1787 | if (ret < 0) { | |
1788 | err = ret; | |
1789 | goto out; | |
1790 | } | |
1791 | if (ret > 0) { | |
1792 | err = -EIO; | |
1793 | goto out; | |
1794 | } | |
1795 | ||
1796 | leaf = path->nodes[0]; | |
1797 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1798 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1799 | if (item_size < sizeof(*ei)) { | |
1800 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
1801 | path, (u64)-1, 0); | |
1802 | if (ret < 0) { | |
1803 | err = ret; | |
1804 | goto out; | |
1805 | } | |
1806 | leaf = path->nodes[0]; | |
1807 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1808 | } | |
1809 | #endif | |
1810 | BUG_ON(item_size < sizeof(*ei)); | |
1811 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1812 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1813 | |
5d4f98a2 YZ |
1814 | btrfs_mark_buffer_dirty(leaf); |
1815 | out: | |
1816 | btrfs_free_path(path); | |
1817 | return err; | |
56bec294 CM |
1818 | } |
1819 | ||
5d4f98a2 YZ |
1820 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
1821 | struct btrfs_root *root, | |
1822 | struct btrfs_delayed_ref_node *node, | |
1823 | struct btrfs_delayed_extent_op *extent_op, | |
1824 | int insert_reserved) | |
56bec294 CM |
1825 | { |
1826 | int ret = 0; | |
5d4f98a2 YZ |
1827 | struct btrfs_delayed_tree_ref *ref; |
1828 | struct btrfs_key ins; | |
1829 | u64 parent = 0; | |
1830 | u64 ref_root = 0; | |
56bec294 | 1831 | |
5d4f98a2 YZ |
1832 | ins.objectid = node->bytenr; |
1833 | ins.offset = node->num_bytes; | |
1834 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 1835 | |
5d4f98a2 YZ |
1836 | ref = btrfs_delayed_node_to_tree_ref(node); |
1837 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
1838 | parent = ref->parent; | |
1839 | else | |
1840 | ref_root = ref->root; | |
1841 | ||
1842 | BUG_ON(node->ref_mod != 1); | |
1843 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1844 | BUG_ON(!extent_op || !extent_op->update_flags || | |
1845 | !extent_op->update_key); | |
1846 | ret = alloc_reserved_tree_block(trans, root, | |
1847 | parent, ref_root, | |
1848 | extent_op->flags_to_set, | |
1849 | &extent_op->key, | |
1850 | ref->level, &ins); | |
5d4f98a2 YZ |
1851 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1852 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1853 | node->num_bytes, parent, ref_root, | |
1854 | ref->level, 0, 1, extent_op); | |
1855 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1856 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1857 | node->num_bytes, parent, ref_root, | |
1858 | ref->level, 0, 1, extent_op); | |
1859 | } else { | |
1860 | BUG(); | |
1861 | } | |
56bec294 CM |
1862 | return ret; |
1863 | } | |
1864 | ||
5d4f98a2 | 1865 | |
56bec294 | 1866 | /* helper function to actually process a single delayed ref entry */ |
5d4f98a2 YZ |
1867 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
1868 | struct btrfs_root *root, | |
1869 | struct btrfs_delayed_ref_node *node, | |
1870 | struct btrfs_delayed_extent_op *extent_op, | |
1871 | int insert_reserved) | |
56bec294 CM |
1872 | { |
1873 | int ret; | |
5d4f98a2 | 1874 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
1875 | struct btrfs_delayed_ref_head *head; |
1876 | /* | |
1877 | * we've hit the end of the chain and we were supposed | |
1878 | * to insert this extent into the tree. But, it got | |
1879 | * deleted before we ever needed to insert it, so all | |
1880 | * we have to do is clean up the accounting | |
1881 | */ | |
5d4f98a2 YZ |
1882 | BUG_ON(extent_op); |
1883 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 1884 | if (insert_reserved) { |
11833d66 YZ |
1885 | int mark_free = 0; |
1886 | struct extent_buffer *must_clean = NULL; | |
1887 | ||
1888 | ret = pin_down_bytes(trans, root, NULL, | |
1889 | node->bytenr, node->num_bytes, | |
1890 | head->is_data, 1, &must_clean); | |
1891 | if (ret > 0) | |
1892 | mark_free = 1; | |
1893 | ||
1894 | if (must_clean) { | |
1895 | clean_tree_block(NULL, root, must_clean); | |
1896 | btrfs_tree_unlock(must_clean); | |
1897 | free_extent_buffer(must_clean); | |
1898 | } | |
5d4f98a2 YZ |
1899 | if (head->is_data) { |
1900 | ret = btrfs_del_csums(trans, root, | |
1901 | node->bytenr, | |
1902 | node->num_bytes); | |
1903 | BUG_ON(ret); | |
1904 | } | |
11833d66 YZ |
1905 | if (mark_free) { |
1906 | ret = btrfs_free_reserved_extent(root, | |
1907 | node->bytenr, | |
1908 | node->num_bytes); | |
1909 | BUG_ON(ret); | |
1910 | } | |
56bec294 | 1911 | } |
56bec294 CM |
1912 | mutex_unlock(&head->mutex); |
1913 | return 0; | |
1914 | } | |
1915 | ||
5d4f98a2 YZ |
1916 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1917 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
1918 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
1919 | insert_reserved); | |
1920 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1921 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1922 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
1923 | insert_reserved); | |
1924 | else | |
1925 | BUG(); | |
1926 | return ret; | |
56bec294 CM |
1927 | } |
1928 | ||
1929 | static noinline struct btrfs_delayed_ref_node * | |
1930 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
1931 | { | |
1932 | struct rb_node *node; | |
1933 | struct btrfs_delayed_ref_node *ref; | |
1934 | int action = BTRFS_ADD_DELAYED_REF; | |
1935 | again: | |
1936 | /* | |
1937 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1938 | * this prevents ref count from going down to zero when | |
1939 | * there still are pending delayed ref. | |
1940 | */ | |
1941 | node = rb_prev(&head->node.rb_node); | |
1942 | while (1) { | |
1943 | if (!node) | |
1944 | break; | |
1945 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
1946 | rb_node); | |
1947 | if (ref->bytenr != head->node.bytenr) | |
1948 | break; | |
5d4f98a2 | 1949 | if (ref->action == action) |
56bec294 CM |
1950 | return ref; |
1951 | node = rb_prev(node); | |
1952 | } | |
1953 | if (action == BTRFS_ADD_DELAYED_REF) { | |
1954 | action = BTRFS_DROP_DELAYED_REF; | |
1955 | goto again; | |
1956 | } | |
1957 | return NULL; | |
1958 | } | |
1959 | ||
c3e69d58 CM |
1960 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
1961 | struct btrfs_root *root, | |
1962 | struct list_head *cluster) | |
56bec294 | 1963 | { |
56bec294 CM |
1964 | struct btrfs_delayed_ref_root *delayed_refs; |
1965 | struct btrfs_delayed_ref_node *ref; | |
1966 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 1967 | struct btrfs_delayed_extent_op *extent_op; |
56bec294 | 1968 | int ret; |
c3e69d58 | 1969 | int count = 0; |
56bec294 | 1970 | int must_insert_reserved = 0; |
56bec294 CM |
1971 | |
1972 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
1973 | while (1) { |
1974 | if (!locked_ref) { | |
c3e69d58 CM |
1975 | /* pick a new head ref from the cluster list */ |
1976 | if (list_empty(cluster)) | |
56bec294 | 1977 | break; |
56bec294 | 1978 | |
c3e69d58 CM |
1979 | locked_ref = list_entry(cluster->next, |
1980 | struct btrfs_delayed_ref_head, cluster); | |
1981 | ||
1982 | /* grab the lock that says we are going to process | |
1983 | * all the refs for this head */ | |
1984 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
1985 | ||
1986 | /* | |
1987 | * we may have dropped the spin lock to get the head | |
1988 | * mutex lock, and that might have given someone else | |
1989 | * time to free the head. If that's true, it has been | |
1990 | * removed from our list and we can move on. | |
1991 | */ | |
1992 | if (ret == -EAGAIN) { | |
1993 | locked_ref = NULL; | |
1994 | count++; | |
1995 | continue; | |
56bec294 CM |
1996 | } |
1997 | } | |
a28ec197 | 1998 | |
56bec294 CM |
1999 | /* |
2000 | * record the must insert reserved flag before we | |
2001 | * drop the spin lock. | |
2002 | */ | |
2003 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2004 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2005 | |
5d4f98a2 YZ |
2006 | extent_op = locked_ref->extent_op; |
2007 | locked_ref->extent_op = NULL; | |
2008 | ||
56bec294 CM |
2009 | /* |
2010 | * locked_ref is the head node, so we have to go one | |
2011 | * node back for any delayed ref updates | |
2012 | */ | |
56bec294 CM |
2013 | ref = select_delayed_ref(locked_ref); |
2014 | if (!ref) { | |
2015 | /* All delayed refs have been processed, Go ahead | |
2016 | * and send the head node to run_one_delayed_ref, | |
2017 | * so that any accounting fixes can happen | |
2018 | */ | |
2019 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2020 | |
2021 | if (extent_op && must_insert_reserved) { | |
2022 | kfree(extent_op); | |
2023 | extent_op = NULL; | |
2024 | } | |
2025 | ||
2026 | if (extent_op) { | |
2027 | spin_unlock(&delayed_refs->lock); | |
2028 | ||
2029 | ret = run_delayed_extent_op(trans, root, | |
2030 | ref, extent_op); | |
2031 | BUG_ON(ret); | |
2032 | kfree(extent_op); | |
2033 | ||
2034 | cond_resched(); | |
2035 | spin_lock(&delayed_refs->lock); | |
2036 | continue; | |
2037 | } | |
2038 | ||
c3e69d58 | 2039 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2040 | locked_ref = NULL; |
2041 | } | |
02217ed2 | 2042 | |
56bec294 CM |
2043 | ref->in_tree = 0; |
2044 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2045 | delayed_refs->num_entries--; | |
5d4f98a2 | 2046 | |
56bec294 | 2047 | spin_unlock(&delayed_refs->lock); |
925baedd | 2048 | |
5d4f98a2 | 2049 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 CM |
2050 | must_insert_reserved); |
2051 | BUG_ON(ret); | |
eb099670 | 2052 | |
5d4f98a2 YZ |
2053 | btrfs_put_delayed_ref(ref); |
2054 | kfree(extent_op); | |
c3e69d58 | 2055 | count++; |
5d4f98a2 | 2056 | |
c3e69d58 CM |
2057 | cond_resched(); |
2058 | spin_lock(&delayed_refs->lock); | |
2059 | } | |
2060 | return count; | |
2061 | } | |
2062 | ||
2063 | /* | |
2064 | * this starts processing the delayed reference count updates and | |
2065 | * extent insertions we have queued up so far. count can be | |
2066 | * 0, which means to process everything in the tree at the start | |
2067 | * of the run (but not newly added entries), or it can be some target | |
2068 | * number you'd like to process. | |
2069 | */ | |
2070 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2071 | struct btrfs_root *root, unsigned long count) | |
2072 | { | |
2073 | struct rb_node *node; | |
2074 | struct btrfs_delayed_ref_root *delayed_refs; | |
2075 | struct btrfs_delayed_ref_node *ref; | |
2076 | struct list_head cluster; | |
2077 | int ret; | |
2078 | int run_all = count == (unsigned long)-1; | |
2079 | int run_most = 0; | |
2080 | ||
2081 | if (root == root->fs_info->extent_root) | |
2082 | root = root->fs_info->tree_root; | |
2083 | ||
2084 | delayed_refs = &trans->transaction->delayed_refs; | |
2085 | INIT_LIST_HEAD(&cluster); | |
2086 | again: | |
2087 | spin_lock(&delayed_refs->lock); | |
2088 | if (count == 0) { | |
2089 | count = delayed_refs->num_entries * 2; | |
2090 | run_most = 1; | |
2091 | } | |
2092 | while (1) { | |
2093 | if (!(run_all || run_most) && | |
2094 | delayed_refs->num_heads_ready < 64) | |
2095 | break; | |
eb099670 | 2096 | |
56bec294 | 2097 | /* |
c3e69d58 CM |
2098 | * go find something we can process in the rbtree. We start at |
2099 | * the beginning of the tree, and then build a cluster | |
2100 | * of refs to process starting at the first one we are able to | |
2101 | * lock | |
56bec294 | 2102 | */ |
c3e69d58 CM |
2103 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2104 | delayed_refs->run_delayed_start); | |
2105 | if (ret) | |
56bec294 CM |
2106 | break; |
2107 | ||
c3e69d58 CM |
2108 | ret = run_clustered_refs(trans, root, &cluster); |
2109 | BUG_ON(ret < 0); | |
2110 | ||
2111 | count -= min_t(unsigned long, ret, count); | |
2112 | ||
2113 | if (count == 0) | |
2114 | break; | |
eb099670 | 2115 | } |
c3e69d58 | 2116 | |
56bec294 | 2117 | if (run_all) { |
56bec294 | 2118 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2119 | if (!node) |
56bec294 | 2120 | goto out; |
c3e69d58 | 2121 | count = (unsigned long)-1; |
e9d0b13b | 2122 | |
56bec294 CM |
2123 | while (node) { |
2124 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2125 | rb_node); | |
2126 | if (btrfs_delayed_ref_is_head(ref)) { | |
2127 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2128 | |
56bec294 CM |
2129 | head = btrfs_delayed_node_to_head(ref); |
2130 | atomic_inc(&ref->refs); | |
2131 | ||
2132 | spin_unlock(&delayed_refs->lock); | |
2133 | mutex_lock(&head->mutex); | |
2134 | mutex_unlock(&head->mutex); | |
2135 | ||
2136 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2137 | cond_resched(); |
56bec294 CM |
2138 | goto again; |
2139 | } | |
2140 | node = rb_next(node); | |
2141 | } | |
2142 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2143 | schedule_timeout(1); |
2144 | goto again; | |
5f39d397 | 2145 | } |
54aa1f4d | 2146 | out: |
c3e69d58 | 2147 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2148 | return 0; |
2149 | } | |
2150 | ||
5d4f98a2 YZ |
2151 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2152 | struct btrfs_root *root, | |
2153 | u64 bytenr, u64 num_bytes, u64 flags, | |
2154 | int is_data) | |
2155 | { | |
2156 | struct btrfs_delayed_extent_op *extent_op; | |
2157 | int ret; | |
2158 | ||
2159 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2160 | if (!extent_op) | |
2161 | return -ENOMEM; | |
2162 | ||
2163 | extent_op->flags_to_set = flags; | |
2164 | extent_op->update_flags = 1; | |
2165 | extent_op->update_key = 0; | |
2166 | extent_op->is_data = is_data ? 1 : 0; | |
2167 | ||
2168 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); | |
2169 | if (ret) | |
2170 | kfree(extent_op); | |
2171 | return ret; | |
2172 | } | |
2173 | ||
2174 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2175 | struct btrfs_root *root, | |
2176 | struct btrfs_path *path, | |
2177 | u64 objectid, u64 offset, u64 bytenr) | |
2178 | { | |
2179 | struct btrfs_delayed_ref_head *head; | |
2180 | struct btrfs_delayed_ref_node *ref; | |
2181 | struct btrfs_delayed_data_ref *data_ref; | |
2182 | struct btrfs_delayed_ref_root *delayed_refs; | |
2183 | struct rb_node *node; | |
2184 | int ret = 0; | |
2185 | ||
2186 | ret = -ENOENT; | |
2187 | delayed_refs = &trans->transaction->delayed_refs; | |
2188 | spin_lock(&delayed_refs->lock); | |
2189 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2190 | if (!head) | |
2191 | goto out; | |
2192 | ||
2193 | if (!mutex_trylock(&head->mutex)) { | |
2194 | atomic_inc(&head->node.refs); | |
2195 | spin_unlock(&delayed_refs->lock); | |
2196 | ||
2197 | btrfs_release_path(root->fs_info->extent_root, path); | |
2198 | ||
2199 | mutex_lock(&head->mutex); | |
2200 | mutex_unlock(&head->mutex); | |
2201 | btrfs_put_delayed_ref(&head->node); | |
2202 | return -EAGAIN; | |
2203 | } | |
2204 | ||
2205 | node = rb_prev(&head->node.rb_node); | |
2206 | if (!node) | |
2207 | goto out_unlock; | |
2208 | ||
2209 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2210 | ||
2211 | if (ref->bytenr != bytenr) | |
2212 | goto out_unlock; | |
2213 | ||
2214 | ret = 1; | |
2215 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2216 | goto out_unlock; | |
2217 | ||
2218 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2219 | ||
2220 | node = rb_prev(node); | |
2221 | if (node) { | |
2222 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2223 | if (ref->bytenr == bytenr) | |
2224 | goto out_unlock; | |
2225 | } | |
2226 | ||
2227 | if (data_ref->root != root->root_key.objectid || | |
2228 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2229 | goto out_unlock; | |
2230 | ||
2231 | ret = 0; | |
2232 | out_unlock: | |
2233 | mutex_unlock(&head->mutex); | |
2234 | out: | |
2235 | spin_unlock(&delayed_refs->lock); | |
2236 | return ret; | |
2237 | } | |
2238 | ||
2239 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2240 | struct btrfs_root *root, | |
2241 | struct btrfs_path *path, | |
2242 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2243 | { |
2244 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2245 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2246 | struct btrfs_extent_data_ref *ref; |
2247 | struct btrfs_extent_inline_ref *iref; | |
2248 | struct btrfs_extent_item *ei; | |
f321e491 | 2249 | struct btrfs_key key; |
5d4f98a2 | 2250 | u32 item_size; |
be20aa9d | 2251 | int ret; |
925baedd | 2252 | |
be20aa9d | 2253 | key.objectid = bytenr; |
31840ae1 | 2254 | key.offset = (u64)-1; |
f321e491 | 2255 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2256 | |
be20aa9d CM |
2257 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2258 | if (ret < 0) | |
2259 | goto out; | |
2260 | BUG_ON(ret == 0); | |
80ff3856 YZ |
2261 | |
2262 | ret = -ENOENT; | |
2263 | if (path->slots[0] == 0) | |
31840ae1 | 2264 | goto out; |
be20aa9d | 2265 | |
31840ae1 | 2266 | path->slots[0]--; |
f321e491 | 2267 | leaf = path->nodes[0]; |
5d4f98a2 | 2268 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2269 | |
5d4f98a2 | 2270 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2271 | goto out; |
f321e491 | 2272 | |
5d4f98a2 YZ |
2273 | ret = 1; |
2274 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2275 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2276 | if (item_size < sizeof(*ei)) { | |
2277 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2278 | goto out; | |
2279 | } | |
2280 | #endif | |
2281 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2282 | |
5d4f98a2 YZ |
2283 | if (item_size != sizeof(*ei) + |
2284 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2285 | goto out; | |
be20aa9d | 2286 | |
5d4f98a2 YZ |
2287 | if (btrfs_extent_generation(leaf, ei) <= |
2288 | btrfs_root_last_snapshot(&root->root_item)) | |
2289 | goto out; | |
2290 | ||
2291 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2292 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2293 | BTRFS_EXTENT_DATA_REF_KEY) | |
2294 | goto out; | |
2295 | ||
2296 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2297 | if (btrfs_extent_refs(leaf, ei) != | |
2298 | btrfs_extent_data_ref_count(leaf, ref) || | |
2299 | btrfs_extent_data_ref_root(leaf, ref) != | |
2300 | root->root_key.objectid || | |
2301 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2302 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2303 | goto out; | |
2304 | ||
2305 | ret = 0; | |
2306 | out: | |
2307 | return ret; | |
2308 | } | |
2309 | ||
2310 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2311 | struct btrfs_root *root, | |
2312 | u64 objectid, u64 offset, u64 bytenr) | |
2313 | { | |
2314 | struct btrfs_path *path; | |
2315 | int ret; | |
2316 | int ret2; | |
2317 | ||
2318 | path = btrfs_alloc_path(); | |
2319 | if (!path) | |
2320 | return -ENOENT; | |
2321 | ||
2322 | do { | |
2323 | ret = check_committed_ref(trans, root, path, objectid, | |
2324 | offset, bytenr); | |
2325 | if (ret && ret != -ENOENT) | |
f321e491 | 2326 | goto out; |
80ff3856 | 2327 | |
5d4f98a2 YZ |
2328 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2329 | offset, bytenr); | |
2330 | } while (ret2 == -EAGAIN); | |
2331 | ||
2332 | if (ret2 && ret2 != -ENOENT) { | |
2333 | ret = ret2; | |
2334 | goto out; | |
f321e491 | 2335 | } |
5d4f98a2 YZ |
2336 | |
2337 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2338 | ret = 0; | |
be20aa9d | 2339 | out: |
80ff3856 | 2340 | btrfs_free_path(path); |
f321e491 | 2341 | return ret; |
be20aa9d | 2342 | } |
c5739bba | 2343 | |
5d4f98a2 | 2344 | #if 0 |
31840ae1 ZY |
2345 | int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2346 | struct extent_buffer *buf, u32 nr_extents) | |
02217ed2 | 2347 | { |
5f39d397 | 2348 | struct btrfs_key key; |
6407bf6d | 2349 | struct btrfs_file_extent_item *fi; |
e4657689 ZY |
2350 | u64 root_gen; |
2351 | u32 nritems; | |
02217ed2 | 2352 | int i; |
db94535d | 2353 | int level; |
31840ae1 | 2354 | int ret = 0; |
e4657689 | 2355 | int shared = 0; |
a28ec197 | 2356 | |
3768f368 | 2357 | if (!root->ref_cows) |
a28ec197 | 2358 | return 0; |
5f39d397 | 2359 | |
e4657689 ZY |
2360 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
2361 | shared = 0; | |
2362 | root_gen = root->root_key.offset; | |
2363 | } else { | |
2364 | shared = 1; | |
2365 | root_gen = trans->transid - 1; | |
2366 | } | |
2367 | ||
db94535d | 2368 | level = btrfs_header_level(buf); |
5f39d397 | 2369 | nritems = btrfs_header_nritems(buf); |
4a096752 | 2370 | |
31840ae1 | 2371 | if (level == 0) { |
31153d81 YZ |
2372 | struct btrfs_leaf_ref *ref; |
2373 | struct btrfs_extent_info *info; | |
2374 | ||
31840ae1 | 2375 | ref = btrfs_alloc_leaf_ref(root, nr_extents); |
31153d81 | 2376 | if (!ref) { |
31840ae1 | 2377 | ret = -ENOMEM; |
31153d81 YZ |
2378 | goto out; |
2379 | } | |
2380 | ||
e4657689 | 2381 | ref->root_gen = root_gen; |
31153d81 YZ |
2382 | ref->bytenr = buf->start; |
2383 | ref->owner = btrfs_header_owner(buf); | |
2384 | ref->generation = btrfs_header_generation(buf); | |
31840ae1 | 2385 | ref->nritems = nr_extents; |
31153d81 | 2386 | info = ref->extents; |
bcc63abb | 2387 | |
31840ae1 | 2388 | for (i = 0; nr_extents > 0 && i < nritems; i++) { |
31153d81 YZ |
2389 | u64 disk_bytenr; |
2390 | btrfs_item_key_to_cpu(buf, &key, i); | |
2391 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
2392 | continue; | |
2393 | fi = btrfs_item_ptr(buf, i, | |
2394 | struct btrfs_file_extent_item); | |
2395 | if (btrfs_file_extent_type(buf, fi) == | |
2396 | BTRFS_FILE_EXTENT_INLINE) | |
2397 | continue; | |
2398 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2399 | if (disk_bytenr == 0) | |
2400 | continue; | |
2401 | ||
2402 | info->bytenr = disk_bytenr; | |
2403 | info->num_bytes = | |
2404 | btrfs_file_extent_disk_num_bytes(buf, fi); | |
2405 | info->objectid = key.objectid; | |
2406 | info->offset = key.offset; | |
2407 | info++; | |
2408 | } | |
2409 | ||
e4657689 | 2410 | ret = btrfs_add_leaf_ref(root, ref, shared); |
5b84e8d6 YZ |
2411 | if (ret == -EEXIST && shared) { |
2412 | struct btrfs_leaf_ref *old; | |
2413 | old = btrfs_lookup_leaf_ref(root, ref->bytenr); | |
2414 | BUG_ON(!old); | |
2415 | btrfs_remove_leaf_ref(root, old); | |
2416 | btrfs_free_leaf_ref(root, old); | |
2417 | ret = btrfs_add_leaf_ref(root, ref, shared); | |
2418 | } | |
31153d81 | 2419 | WARN_ON(ret); |
bcc63abb | 2420 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
2421 | } |
2422 | out: | |
31840ae1 ZY |
2423 | return ret; |
2424 | } | |
2425 | ||
b7a9f29f CM |
2426 | /* when a block goes through cow, we update the reference counts of |
2427 | * everything that block points to. The internal pointers of the block | |
2428 | * can be in just about any order, and it is likely to have clusters of | |
2429 | * things that are close together and clusters of things that are not. | |
2430 | * | |
2431 | * To help reduce the seeks that come with updating all of these reference | |
2432 | * counts, sort them by byte number before actual updates are done. | |
2433 | * | |
2434 | * struct refsort is used to match byte number to slot in the btree block. | |
2435 | * we sort based on the byte number and then use the slot to actually | |
2436 | * find the item. | |
bd56b302 CM |
2437 | * |
2438 | * struct refsort is smaller than strcut btrfs_item and smaller than | |
2439 | * struct btrfs_key_ptr. Since we're currently limited to the page size | |
2440 | * for a btree block, there's no way for a kmalloc of refsorts for a | |
2441 | * single node to be bigger than a page. | |
b7a9f29f CM |
2442 | */ |
2443 | struct refsort { | |
2444 | u64 bytenr; | |
2445 | u32 slot; | |
2446 | }; | |
2447 | ||
2448 | /* | |
2449 | * for passing into sort() | |
2450 | */ | |
2451 | static int refsort_cmp(const void *a_void, const void *b_void) | |
2452 | { | |
2453 | const struct refsort *a = a_void; | |
2454 | const struct refsort *b = b_void; | |
2455 | ||
2456 | if (a->bytenr < b->bytenr) | |
2457 | return -1; | |
2458 | if (a->bytenr > b->bytenr) | |
2459 | return 1; | |
2460 | return 0; | |
2461 | } | |
5d4f98a2 | 2462 | #endif |
b7a9f29f | 2463 | |
5d4f98a2 | 2464 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2465 | struct btrfs_root *root, |
5d4f98a2 YZ |
2466 | struct extent_buffer *buf, |
2467 | int full_backref, int inc) | |
31840ae1 ZY |
2468 | { |
2469 | u64 bytenr; | |
5d4f98a2 YZ |
2470 | u64 num_bytes; |
2471 | u64 parent; | |
31840ae1 | 2472 | u64 ref_root; |
31840ae1 | 2473 | u32 nritems; |
31840ae1 ZY |
2474 | struct btrfs_key key; |
2475 | struct btrfs_file_extent_item *fi; | |
2476 | int i; | |
2477 | int level; | |
2478 | int ret = 0; | |
31840ae1 | 2479 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
5d4f98a2 | 2480 | u64, u64, u64, u64, u64, u64); |
31840ae1 ZY |
2481 | |
2482 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2483 | nritems = btrfs_header_nritems(buf); |
2484 | level = btrfs_header_level(buf); | |
2485 | ||
5d4f98a2 YZ |
2486 | if (!root->ref_cows && level == 0) |
2487 | return 0; | |
31840ae1 | 2488 | |
5d4f98a2 YZ |
2489 | if (inc) |
2490 | process_func = btrfs_inc_extent_ref; | |
2491 | else | |
2492 | process_func = btrfs_free_extent; | |
31840ae1 | 2493 | |
5d4f98a2 YZ |
2494 | if (full_backref) |
2495 | parent = buf->start; | |
2496 | else | |
2497 | parent = 0; | |
2498 | ||
2499 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2500 | if (level == 0) { |
5d4f98a2 | 2501 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2502 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2503 | continue; | |
5d4f98a2 | 2504 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2505 | struct btrfs_file_extent_item); |
2506 | if (btrfs_file_extent_type(buf, fi) == | |
2507 | BTRFS_FILE_EXTENT_INLINE) | |
2508 | continue; | |
2509 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2510 | if (bytenr == 0) | |
2511 | continue; | |
5d4f98a2 YZ |
2512 | |
2513 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2514 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2515 | ret = process_func(trans, root, bytenr, num_bytes, | |
2516 | parent, ref_root, key.objectid, | |
2517 | key.offset); | |
31840ae1 ZY |
2518 | if (ret) |
2519 | goto fail; | |
2520 | } else { | |
5d4f98a2 YZ |
2521 | bytenr = btrfs_node_blockptr(buf, i); |
2522 | num_bytes = btrfs_level_size(root, level - 1); | |
2523 | ret = process_func(trans, root, bytenr, num_bytes, | |
2524 | parent, ref_root, level - 1, 0); | |
31840ae1 ZY |
2525 | if (ret) |
2526 | goto fail; | |
2527 | } | |
2528 | } | |
2529 | return 0; | |
2530 | fail: | |
5d4f98a2 YZ |
2531 | BUG(); |
2532 | return ret; | |
2533 | } | |
2534 | ||
2535 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2536 | struct extent_buffer *buf, int full_backref) | |
2537 | { | |
2538 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); | |
2539 | } | |
2540 | ||
2541 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2542 | struct extent_buffer *buf, int full_backref) | |
2543 | { | |
2544 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); | |
31840ae1 ZY |
2545 | } |
2546 | ||
9078a3e1 CM |
2547 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2548 | struct btrfs_root *root, | |
2549 | struct btrfs_path *path, | |
2550 | struct btrfs_block_group_cache *cache) | |
2551 | { | |
2552 | int ret; | |
9078a3e1 | 2553 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2554 | unsigned long bi; |
2555 | struct extent_buffer *leaf; | |
9078a3e1 | 2556 | |
9078a3e1 | 2557 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2558 | if (ret < 0) |
2559 | goto fail; | |
9078a3e1 | 2560 | BUG_ON(ret); |
5f39d397 CM |
2561 | |
2562 | leaf = path->nodes[0]; | |
2563 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2564 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2565 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 2566 | btrfs_release_path(extent_root, path); |
54aa1f4d | 2567 | fail: |
9078a3e1 CM |
2568 | if (ret) |
2569 | return ret; | |
9078a3e1 CM |
2570 | return 0; |
2571 | ||
2572 | } | |
2573 | ||
4a8c9a62 YZ |
2574 | static struct btrfs_block_group_cache * |
2575 | next_block_group(struct btrfs_root *root, | |
2576 | struct btrfs_block_group_cache *cache) | |
2577 | { | |
2578 | struct rb_node *node; | |
2579 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2580 | node = rb_next(&cache->cache_node); | |
2581 | btrfs_put_block_group(cache); | |
2582 | if (node) { | |
2583 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2584 | cache_node); | |
2585 | atomic_inc(&cache->count); | |
2586 | } else | |
2587 | cache = NULL; | |
2588 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2589 | return cache; | |
2590 | } | |
2591 | ||
96b5179d CM |
2592 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2593 | struct btrfs_root *root) | |
9078a3e1 | 2594 | { |
4a8c9a62 | 2595 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2596 | int err = 0; |
9078a3e1 | 2597 | struct btrfs_path *path; |
96b5179d | 2598 | u64 last = 0; |
9078a3e1 CM |
2599 | |
2600 | path = btrfs_alloc_path(); | |
2601 | if (!path) | |
2602 | return -ENOMEM; | |
2603 | ||
d397712b | 2604 | while (1) { |
4a8c9a62 YZ |
2605 | if (last == 0) { |
2606 | err = btrfs_run_delayed_refs(trans, root, | |
2607 | (unsigned long)-1); | |
2608 | BUG_ON(err); | |
0f9dd46c | 2609 | } |
54aa1f4d | 2610 | |
4a8c9a62 YZ |
2611 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2612 | while (cache) { | |
2613 | if (cache->dirty) | |
2614 | break; | |
2615 | cache = next_block_group(root, cache); | |
2616 | } | |
2617 | if (!cache) { | |
2618 | if (last == 0) | |
2619 | break; | |
2620 | last = 0; | |
2621 | continue; | |
2622 | } | |
0f9dd46c | 2623 | |
e8569813 | 2624 | cache->dirty = 0; |
4a8c9a62 | 2625 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 2626 | |
4a8c9a62 YZ |
2627 | err = write_one_cache_group(trans, root, path, cache); |
2628 | BUG_ON(err); | |
2629 | btrfs_put_block_group(cache); | |
9078a3e1 | 2630 | } |
4a8c9a62 | 2631 | |
9078a3e1 | 2632 | btrfs_free_path(path); |
4a8c9a62 | 2633 | return 0; |
9078a3e1 CM |
2634 | } |
2635 | ||
d2fb3437 YZ |
2636 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
2637 | { | |
2638 | struct btrfs_block_group_cache *block_group; | |
2639 | int readonly = 0; | |
2640 | ||
2641 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
2642 | if (!block_group || block_group->ro) | |
2643 | readonly = 1; | |
2644 | if (block_group) | |
fa9c0d79 | 2645 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2646 | return readonly; |
2647 | } | |
2648 | ||
593060d7 CM |
2649 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
2650 | u64 total_bytes, u64 bytes_used, | |
2651 | struct btrfs_space_info **space_info) | |
2652 | { | |
2653 | struct btrfs_space_info *found; | |
2654 | ||
2655 | found = __find_space_info(info, flags); | |
2656 | if (found) { | |
25179201 | 2657 | spin_lock(&found->lock); |
593060d7 CM |
2658 | found->total_bytes += total_bytes; |
2659 | found->bytes_used += bytes_used; | |
8f18cf13 | 2660 | found->full = 0; |
25179201 | 2661 | spin_unlock(&found->lock); |
593060d7 CM |
2662 | *space_info = found; |
2663 | return 0; | |
2664 | } | |
c146afad | 2665 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
2666 | if (!found) |
2667 | return -ENOMEM; | |
2668 | ||
0f9dd46c | 2669 | INIT_LIST_HEAD(&found->block_groups); |
80eb234a | 2670 | init_rwsem(&found->groups_sem); |
0f9dd46c | 2671 | spin_lock_init(&found->lock); |
593060d7 CM |
2672 | found->flags = flags; |
2673 | found->total_bytes = total_bytes; | |
2674 | found->bytes_used = bytes_used; | |
2675 | found->bytes_pinned = 0; | |
e8569813 | 2676 | found->bytes_reserved = 0; |
c146afad | 2677 | found->bytes_readonly = 0; |
6a63209f | 2678 | found->bytes_delalloc = 0; |
593060d7 | 2679 | found->full = 0; |
0ef3e66b | 2680 | found->force_alloc = 0; |
593060d7 | 2681 | *space_info = found; |
4184ea7f | 2682 | list_add_rcu(&found->list, &info->space_info); |
817d52f8 | 2683 | atomic_set(&found->caching_threads, 0); |
593060d7 CM |
2684 | return 0; |
2685 | } | |
2686 | ||
8790d502 CM |
2687 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
2688 | { | |
2689 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 2690 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 2691 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 2692 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
2693 | if (extra_flags) { |
2694 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
2695 | fs_info->avail_data_alloc_bits |= extra_flags; | |
2696 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
2697 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
2698 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
2699 | fs_info->avail_system_alloc_bits |= extra_flags; | |
2700 | } | |
2701 | } | |
593060d7 | 2702 | |
c146afad YZ |
2703 | static void set_block_group_readonly(struct btrfs_block_group_cache *cache) |
2704 | { | |
2705 | spin_lock(&cache->space_info->lock); | |
2706 | spin_lock(&cache->lock); | |
2707 | if (!cache->ro) { | |
2708 | cache->space_info->bytes_readonly += cache->key.offset - | |
2709 | btrfs_block_group_used(&cache->item); | |
2710 | cache->ro = 1; | |
2711 | } | |
2712 | spin_unlock(&cache->lock); | |
2713 | spin_unlock(&cache->space_info->lock); | |
2714 | } | |
2715 | ||
2b82032c | 2716 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 2717 | { |
2b82032c | 2718 | u64 num_devices = root->fs_info->fs_devices->rw_devices; |
a061fc8d CM |
2719 | |
2720 | if (num_devices == 1) | |
2721 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
2722 | if (num_devices < 4) | |
2723 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
2724 | ||
ec44a35c CM |
2725 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
2726 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 2727 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 2728 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 2729 | } |
ec44a35c CM |
2730 | |
2731 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 2732 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 2733 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 2734 | } |
ec44a35c CM |
2735 | |
2736 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
2737 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
2738 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
2739 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
2740 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
2741 | return flags; | |
2742 | } | |
2743 | ||
6a63209f JB |
2744 | static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data) |
2745 | { | |
2746 | struct btrfs_fs_info *info = root->fs_info; | |
2747 | u64 alloc_profile; | |
2748 | ||
2749 | if (data) { | |
2750 | alloc_profile = info->avail_data_alloc_bits & | |
2751 | info->data_alloc_profile; | |
2752 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
2753 | } else if (root == root->fs_info->chunk_root) { | |
2754 | alloc_profile = info->avail_system_alloc_bits & | |
2755 | info->system_alloc_profile; | |
2756 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
2757 | } else { | |
2758 | alloc_profile = info->avail_metadata_alloc_bits & | |
2759 | info->metadata_alloc_profile; | |
2760 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
2761 | } | |
2762 | ||
2763 | return btrfs_reduce_alloc_profile(root, data); | |
2764 | } | |
2765 | ||
2766 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) | |
2767 | { | |
2768 | u64 alloc_target; | |
2769 | ||
2770 | alloc_target = btrfs_get_alloc_profile(root, 1); | |
2771 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, | |
2772 | alloc_target); | |
2773 | } | |
2774 | ||
9ed74f2d JB |
2775 | static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items) |
2776 | { | |
2777 | u64 num_bytes; | |
2778 | int level; | |
2779 | ||
2780 | level = BTRFS_MAX_LEVEL - 2; | |
2781 | /* | |
2782 | * NOTE: these calculations are absolutely the worst possible case. | |
2783 | * This assumes that _every_ item we insert will require a new leaf, and | |
2784 | * that the tree has grown to its maximum level size. | |
2785 | */ | |
2786 | ||
2787 | /* | |
2788 | * for every item we insert we could insert both an extent item and a | |
2789 | * extent ref item. Then for ever item we insert, we will need to cow | |
2790 | * both the original leaf, plus the leaf to the left and right of it. | |
2791 | * | |
2792 | * Unless we are talking about the extent root, then we just want the | |
2793 | * number of items * 2, since we just need the extent item plus its ref. | |
2794 | */ | |
2795 | if (root == root->fs_info->extent_root) | |
2796 | num_bytes = num_items * 2; | |
2797 | else | |
2798 | num_bytes = (num_items + (2 * num_items)) * 3; | |
2799 | ||
2800 | /* | |
2801 | * num_bytes is total number of leaves we could need times the leaf | |
2802 | * size, and then for every leaf we could end up cow'ing 2 nodes per | |
2803 | * level, down to the leaf level. | |
2804 | */ | |
2805 | num_bytes = (num_bytes * root->leafsize) + | |
2806 | (num_bytes * (level * 2)) * root->nodesize; | |
2807 | ||
2808 | return num_bytes; | |
2809 | } | |
2810 | ||
6a63209f | 2811 | /* |
9ed74f2d JB |
2812 | * Unreserve metadata space for delalloc. If we have less reserved credits than |
2813 | * we have extents, this function does nothing. | |
6a63209f | 2814 | */ |
9ed74f2d JB |
2815 | int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root, |
2816 | struct inode *inode, int num_items) | |
6a63209f JB |
2817 | { |
2818 | struct btrfs_fs_info *info = root->fs_info; | |
2819 | struct btrfs_space_info *meta_sinfo; | |
9ed74f2d JB |
2820 | u64 num_bytes; |
2821 | u64 alloc_target; | |
2822 | bool bug = false; | |
6a63209f JB |
2823 | |
2824 | /* get the space info for where the metadata will live */ | |
2825 | alloc_target = btrfs_get_alloc_profile(root, 0); | |
2826 | meta_sinfo = __find_space_info(info, alloc_target); | |
2827 | ||
9ed74f2d JB |
2828 | num_bytes = calculate_bytes_needed(root->fs_info->extent_root, |
2829 | num_items); | |
2830 | ||
6a63209f | 2831 | spin_lock(&meta_sinfo->lock); |
32c00aff JB |
2832 | spin_lock(&BTRFS_I(inode)->accounting_lock); |
2833 | if (BTRFS_I(inode)->reserved_extents <= | |
2834 | BTRFS_I(inode)->outstanding_extents) { | |
2835 | spin_unlock(&BTRFS_I(inode)->accounting_lock); | |
9ed74f2d JB |
2836 | spin_unlock(&meta_sinfo->lock); |
2837 | return 0; | |
2838 | } | |
32c00aff | 2839 | spin_unlock(&BTRFS_I(inode)->accounting_lock); |
9ed74f2d | 2840 | |
32c00aff JB |
2841 | BTRFS_I(inode)->reserved_extents--; |
2842 | BUG_ON(BTRFS_I(inode)->reserved_extents < 0); | |
9ed74f2d JB |
2843 | |
2844 | if (meta_sinfo->bytes_delalloc < num_bytes) { | |
2845 | bug = true; | |
2846 | meta_sinfo->bytes_delalloc = 0; | |
2847 | } else { | |
2848 | meta_sinfo->bytes_delalloc -= num_bytes; | |
2849 | } | |
2850 | spin_unlock(&meta_sinfo->lock); | |
2851 | ||
2852 | BUG_ON(bug); | |
2853 | ||
2854 | return 0; | |
2855 | } | |
6a63209f | 2856 | |
9ed74f2d JB |
2857 | static void check_force_delalloc(struct btrfs_space_info *meta_sinfo) |
2858 | { | |
2859 | u64 thresh; | |
2860 | ||
2861 | thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + | |
2862 | meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly + | |
2863 | meta_sinfo->bytes_super + meta_sinfo->bytes_root + | |
2864 | meta_sinfo->bytes_may_use; | |
6a63209f | 2865 | |
9ed74f2d JB |
2866 | thresh = meta_sinfo->total_bytes - thresh; |
2867 | thresh *= 80; | |
6a63209f | 2868 | do_div(thresh, 100); |
9ed74f2d JB |
2869 | if (thresh <= meta_sinfo->bytes_delalloc) |
2870 | meta_sinfo->force_delalloc = 1; | |
2871 | else | |
2872 | meta_sinfo->force_delalloc = 0; | |
2873 | } | |
6a63209f | 2874 | |
e3ccfa98 JB |
2875 | struct async_flush { |
2876 | struct btrfs_root *root; | |
2877 | struct btrfs_space_info *info; | |
2878 | struct btrfs_work work; | |
2879 | }; | |
2880 | ||
2881 | static noinline void flush_delalloc_async(struct btrfs_work *work) | |
2882 | { | |
2883 | struct async_flush *async; | |
2884 | struct btrfs_root *root; | |
2885 | struct btrfs_space_info *info; | |
2886 | ||
2887 | async = container_of(work, struct async_flush, work); | |
2888 | root = async->root; | |
2889 | info = async->info; | |
2890 | ||
24bbcf04 | 2891 | btrfs_start_delalloc_inodes(root, 0); |
e3ccfa98 | 2892 | wake_up(&info->flush_wait); |
24bbcf04 | 2893 | btrfs_wait_ordered_extents(root, 0, 0); |
e3ccfa98 JB |
2894 | |
2895 | spin_lock(&info->lock); | |
2896 | info->flushing = 0; | |
2897 | spin_unlock(&info->lock); | |
2898 | wake_up(&info->flush_wait); | |
2899 | ||
2900 | kfree(async); | |
2901 | } | |
2902 | ||
2903 | static void wait_on_flush(struct btrfs_space_info *info) | |
2904 | { | |
2905 | DEFINE_WAIT(wait); | |
2906 | u64 used; | |
2907 | ||
2908 | while (1) { | |
2909 | prepare_to_wait(&info->flush_wait, &wait, | |
2910 | TASK_UNINTERRUPTIBLE); | |
2911 | spin_lock(&info->lock); | |
2912 | if (!info->flushing) { | |
2913 | spin_unlock(&info->lock); | |
2914 | break; | |
2915 | } | |
2916 | ||
2917 | used = info->bytes_used + info->bytes_reserved + | |
2918 | info->bytes_pinned + info->bytes_readonly + | |
2919 | info->bytes_super + info->bytes_root + | |
2920 | info->bytes_may_use + info->bytes_delalloc; | |
2921 | if (used < info->total_bytes) { | |
2922 | spin_unlock(&info->lock); | |
2923 | break; | |
2924 | } | |
2925 | spin_unlock(&info->lock); | |
2926 | schedule(); | |
2927 | } | |
2928 | finish_wait(&info->flush_wait, &wait); | |
2929 | } | |
2930 | ||
32c00aff JB |
2931 | static void flush_delalloc(struct btrfs_root *root, |
2932 | struct btrfs_space_info *info) | |
2933 | { | |
e3ccfa98 | 2934 | struct async_flush *async; |
32c00aff JB |
2935 | bool wait = false; |
2936 | ||
2937 | spin_lock(&info->lock); | |
2938 | ||
2939 | if (!info->flushing) { | |
2940 | info->flushing = 1; | |
2941 | init_waitqueue_head(&info->flush_wait); | |
2942 | } else { | |
2943 | wait = true; | |
2944 | } | |
2945 | ||
2946 | spin_unlock(&info->lock); | |
2947 | ||
2948 | if (wait) { | |
e3ccfa98 | 2949 | wait_on_flush(info); |
32c00aff JB |
2950 | return; |
2951 | } | |
2952 | ||
e3ccfa98 JB |
2953 | async = kzalloc(sizeof(*async), GFP_NOFS); |
2954 | if (!async) | |
2955 | goto flush; | |
2956 | ||
2957 | async->root = root; | |
2958 | async->info = info; | |
2959 | async->work.func = flush_delalloc_async; | |
2960 | ||
2961 | btrfs_queue_worker(&root->fs_info->enospc_workers, | |
2962 | &async->work); | |
2963 | wait_on_flush(info); | |
2964 | return; | |
2965 | ||
2966 | flush: | |
24bbcf04 YZ |
2967 | btrfs_start_delalloc_inodes(root, 0); |
2968 | btrfs_wait_ordered_extents(root, 0, 0); | |
32c00aff JB |
2969 | |
2970 | spin_lock(&info->lock); | |
2971 | info->flushing = 0; | |
2972 | spin_unlock(&info->lock); | |
2973 | wake_up(&info->flush_wait); | |
2974 | } | |
2975 | ||
9ed74f2d JB |
2976 | static int maybe_allocate_chunk(struct btrfs_root *root, |
2977 | struct btrfs_space_info *info) | |
2978 | { | |
2979 | struct btrfs_super_block *disk_super = &root->fs_info->super_copy; | |
2980 | struct btrfs_trans_handle *trans; | |
2981 | bool wait = false; | |
2982 | int ret = 0; | |
2983 | u64 min_metadata; | |
2984 | u64 free_space; | |
2985 | ||
2986 | free_space = btrfs_super_total_bytes(disk_super); | |
2987 | /* | |
33b25808 | 2988 | * we allow the metadata to grow to a max of either 10gb or 5% of the |
9ed74f2d JB |
2989 | * space in the volume. |
2990 | */ | |
33b25808 | 2991 | min_metadata = min((u64)10 * 1024 * 1024 * 1024, |
9ed74f2d JB |
2992 | div64_u64(free_space * 5, 100)); |
2993 | if (info->total_bytes >= min_metadata) { | |
2994 | spin_unlock(&info->lock); | |
2995 | return 0; | |
2996 | } | |
2997 | ||
2998 | if (info->full) { | |
2999 | spin_unlock(&info->lock); | |
3000 | return 0; | |
3001 | } | |
3002 | ||
3003 | if (!info->allocating_chunk) { | |
3004 | info->force_alloc = 1; | |
3005 | info->allocating_chunk = 1; | |
e3ccfa98 | 3006 | init_waitqueue_head(&info->allocate_wait); |
9ed74f2d JB |
3007 | } else { |
3008 | wait = true; | |
3009 | } | |
3010 | ||
3011 | spin_unlock(&info->lock); | |
3012 | ||
3013 | if (wait) { | |
e3ccfa98 | 3014 | wait_event(info->allocate_wait, |
9ed74f2d JB |
3015 | !info->allocating_chunk); |
3016 | return 1; | |
3017 | } | |
3018 | ||
3019 | trans = btrfs_start_transaction(root, 1); | |
3020 | if (!trans) { | |
3021 | ret = -ENOMEM; | |
3022 | goto out; | |
3023 | } | |
3024 | ||
3025 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
3026 | 4096 + 2 * 1024 * 1024, | |
3027 | info->flags, 0); | |
3028 | btrfs_end_transaction(trans, root); | |
3029 | if (ret) | |
3030 | goto out; | |
3031 | out: | |
3032 | spin_lock(&info->lock); | |
3033 | info->allocating_chunk = 0; | |
3034 | spin_unlock(&info->lock); | |
e3ccfa98 | 3035 | wake_up(&info->allocate_wait); |
9ed74f2d JB |
3036 | |
3037 | if (ret) | |
3038 | return 0; | |
3039 | return 1; | |
3040 | } | |
3041 | ||
3042 | /* | |
3043 | * Reserve metadata space for delalloc. | |
3044 | */ | |
3045 | int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root, | |
3046 | struct inode *inode, int num_items) | |
3047 | { | |
3048 | struct btrfs_fs_info *info = root->fs_info; | |
3049 | struct btrfs_space_info *meta_sinfo; | |
3050 | u64 num_bytes; | |
3051 | u64 used; | |
3052 | u64 alloc_target; | |
3053 | int flushed = 0; | |
3054 | int force_delalloc; | |
3055 | ||
3056 | /* get the space info for where the metadata will live */ | |
3057 | alloc_target = btrfs_get_alloc_profile(root, 0); | |
3058 | meta_sinfo = __find_space_info(info, alloc_target); | |
3059 | ||
3060 | num_bytes = calculate_bytes_needed(root->fs_info->extent_root, | |
3061 | num_items); | |
3062 | again: | |
3063 | spin_lock(&meta_sinfo->lock); | |
3064 | ||
3065 | force_delalloc = meta_sinfo->force_delalloc; | |
3066 | ||
3067 | if (unlikely(!meta_sinfo->bytes_root)) | |
3068 | meta_sinfo->bytes_root = calculate_bytes_needed(root, 6); | |
3069 | ||
3070 | if (!flushed) | |
3071 | meta_sinfo->bytes_delalloc += num_bytes; | |
3072 | ||
3073 | used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + | |
3074 | meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly + | |
3075 | meta_sinfo->bytes_super + meta_sinfo->bytes_root + | |
3076 | meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc; | |
3077 | ||
3078 | if (used > meta_sinfo->total_bytes) { | |
3079 | flushed++; | |
3080 | ||
3081 | if (flushed == 1) { | |
3082 | if (maybe_allocate_chunk(root, meta_sinfo)) | |
3083 | goto again; | |
3084 | flushed++; | |
3085 | } else { | |
4e06bdd6 | 3086 | spin_unlock(&meta_sinfo->lock); |
9ed74f2d | 3087 | } |
4e06bdd6 | 3088 | |
9ed74f2d JB |
3089 | if (flushed == 2) { |
3090 | filemap_flush(inode->i_mapping); | |
3091 | goto again; | |
3092 | } else if (flushed == 3) { | |
32c00aff | 3093 | flush_delalloc(root, meta_sinfo); |
4e06bdd6 JB |
3094 | goto again; |
3095 | } | |
9ed74f2d JB |
3096 | spin_lock(&meta_sinfo->lock); |
3097 | meta_sinfo->bytes_delalloc -= num_bytes; | |
6a63209f | 3098 | spin_unlock(&meta_sinfo->lock); |
9ed74f2d | 3099 | printk(KERN_ERR "enospc, has %d, reserved %d\n", |
32c00aff JB |
3100 | BTRFS_I(inode)->outstanding_extents, |
3101 | BTRFS_I(inode)->reserved_extents); | |
9ed74f2d JB |
3102 | dump_space_info(meta_sinfo, 0, 0); |
3103 | return -ENOSPC; | |
3104 | } | |
4e06bdd6 | 3105 | |
32c00aff | 3106 | BTRFS_I(inode)->reserved_extents++; |
9ed74f2d JB |
3107 | check_force_delalloc(meta_sinfo); |
3108 | spin_unlock(&meta_sinfo->lock); | |
3109 | ||
3110 | if (!flushed && force_delalloc) | |
3111 | filemap_flush(inode->i_mapping); | |
3112 | ||
3113 | return 0; | |
3114 | } | |
3115 | ||
3116 | /* | |
3117 | * unreserve num_items number of items worth of metadata space. This needs to | |
3118 | * be paired with btrfs_reserve_metadata_space. | |
3119 | * | |
3120 | * NOTE: if you have the option, run this _AFTER_ you do a | |
3121 | * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref | |
3122 | * oprations which will result in more used metadata, so we want to make sure we | |
3123 | * can do that without issue. | |
3124 | */ | |
3125 | int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items) | |
3126 | { | |
3127 | struct btrfs_fs_info *info = root->fs_info; | |
3128 | struct btrfs_space_info *meta_sinfo; | |
3129 | u64 num_bytes; | |
3130 | u64 alloc_target; | |
3131 | bool bug = false; | |
3132 | ||
3133 | /* get the space info for where the metadata will live */ | |
3134 | alloc_target = btrfs_get_alloc_profile(root, 0); | |
3135 | meta_sinfo = __find_space_info(info, alloc_target); | |
3136 | ||
3137 | num_bytes = calculate_bytes_needed(root, num_items); | |
3138 | ||
3139 | spin_lock(&meta_sinfo->lock); | |
3140 | if (meta_sinfo->bytes_may_use < num_bytes) { | |
3141 | bug = true; | |
3142 | meta_sinfo->bytes_may_use = 0; | |
3143 | } else { | |
3144 | meta_sinfo->bytes_may_use -= num_bytes; | |
3145 | } | |
3146 | spin_unlock(&meta_sinfo->lock); | |
3147 | ||
3148 | BUG_ON(bug); | |
3149 | ||
3150 | return 0; | |
3151 | } | |
3152 | ||
3153 | /* | |
3154 | * Reserve some metadata space for use. We'll calculate the worste case number | |
3155 | * of bytes that would be needed to modify num_items number of items. If we | |
3156 | * have space, fantastic, if not, you get -ENOSPC. Please call | |
3157 | * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of | |
3158 | * items you reserved, since whatever metadata you needed should have already | |
3159 | * been allocated. | |
3160 | * | |
3161 | * This will commit the transaction to make more space if we don't have enough | |
3162 | * metadata space. THe only time we don't do this is if we're reserving space | |
3163 | * inside of a transaction, then we will just return -ENOSPC and it is the | |
3164 | * callers responsibility to handle it properly. | |
3165 | */ | |
3166 | int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items) | |
3167 | { | |
3168 | struct btrfs_fs_info *info = root->fs_info; | |
3169 | struct btrfs_space_info *meta_sinfo; | |
3170 | u64 num_bytes; | |
3171 | u64 used; | |
3172 | u64 alloc_target; | |
3173 | int retries = 0; | |
3174 | ||
3175 | /* get the space info for where the metadata will live */ | |
3176 | alloc_target = btrfs_get_alloc_profile(root, 0); | |
3177 | meta_sinfo = __find_space_info(info, alloc_target); | |
3178 | ||
3179 | num_bytes = calculate_bytes_needed(root, num_items); | |
3180 | again: | |
3181 | spin_lock(&meta_sinfo->lock); | |
3182 | ||
3183 | if (unlikely(!meta_sinfo->bytes_root)) | |
3184 | meta_sinfo->bytes_root = calculate_bytes_needed(root, 6); | |
3185 | ||
3186 | if (!retries) | |
3187 | meta_sinfo->bytes_may_use += num_bytes; | |
3188 | ||
3189 | used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + | |
3190 | meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly + | |
3191 | meta_sinfo->bytes_super + meta_sinfo->bytes_root + | |
3192 | meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc; | |
3193 | ||
3194 | if (used > meta_sinfo->total_bytes) { | |
3195 | retries++; | |
3196 | if (retries == 1) { | |
3197 | if (maybe_allocate_chunk(root, meta_sinfo)) | |
3198 | goto again; | |
3199 | retries++; | |
3200 | } else { | |
3201 | spin_unlock(&meta_sinfo->lock); | |
3202 | } | |
3203 | ||
3204 | if (retries == 2) { | |
32c00aff | 3205 | flush_delalloc(root, meta_sinfo); |
4e06bdd6 JB |
3206 | goto again; |
3207 | } | |
9ed74f2d JB |
3208 | spin_lock(&meta_sinfo->lock); |
3209 | meta_sinfo->bytes_may_use -= num_bytes; | |
3210 | spin_unlock(&meta_sinfo->lock); | |
3211 | ||
3212 | dump_space_info(meta_sinfo, 0, 0); | |
6a63209f JB |
3213 | return -ENOSPC; |
3214 | } | |
9ed74f2d JB |
3215 | |
3216 | check_force_delalloc(meta_sinfo); | |
6a63209f JB |
3217 | spin_unlock(&meta_sinfo->lock); |
3218 | ||
3219 | return 0; | |
3220 | } | |
3221 | ||
3222 | /* | |
3223 | * This will check the space that the inode allocates from to make sure we have | |
3224 | * enough space for bytes. | |
3225 | */ | |
3226 | int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode, | |
3227 | u64 bytes) | |
3228 | { | |
3229 | struct btrfs_space_info *data_sinfo; | |
4e06bdd6 | 3230 | int ret = 0, committed = 0; |
6a63209f JB |
3231 | |
3232 | /* make sure bytes are sectorsize aligned */ | |
3233 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
3234 | ||
3235 | data_sinfo = BTRFS_I(inode)->space_info; | |
33b4d47f CM |
3236 | if (!data_sinfo) |
3237 | goto alloc; | |
3238 | ||
6a63209f JB |
3239 | again: |
3240 | /* make sure we have enough space to handle the data first */ | |
3241 | spin_lock(&data_sinfo->lock); | |
3242 | if (data_sinfo->total_bytes - data_sinfo->bytes_used - | |
3243 | data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved - | |
3244 | data_sinfo->bytes_pinned - data_sinfo->bytes_readonly - | |
1b2da372 | 3245 | data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) { |
4e06bdd6 JB |
3246 | struct btrfs_trans_handle *trans; |
3247 | ||
6a63209f JB |
3248 | /* |
3249 | * if we don't have enough free bytes in this space then we need | |
3250 | * to alloc a new chunk. | |
3251 | */ | |
3252 | if (!data_sinfo->full) { | |
3253 | u64 alloc_target; | |
6a63209f JB |
3254 | |
3255 | data_sinfo->force_alloc = 1; | |
3256 | spin_unlock(&data_sinfo->lock); | |
33b4d47f | 3257 | alloc: |
6a63209f JB |
3258 | alloc_target = btrfs_get_alloc_profile(root, 1); |
3259 | trans = btrfs_start_transaction(root, 1); | |
3260 | if (!trans) | |
3261 | return -ENOMEM; | |
3262 | ||
3263 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
3264 | bytes + 2 * 1024 * 1024, | |
3265 | alloc_target, 0); | |
3266 | btrfs_end_transaction(trans, root); | |
3267 | if (ret) | |
3268 | return ret; | |
33b4d47f CM |
3269 | |
3270 | if (!data_sinfo) { | |
3271 | btrfs_set_inode_space_info(root, inode); | |
3272 | data_sinfo = BTRFS_I(inode)->space_info; | |
3273 | } | |
6a63209f JB |
3274 | goto again; |
3275 | } | |
3276 | spin_unlock(&data_sinfo->lock); | |
4e06bdd6 JB |
3277 | |
3278 | /* commit the current transaction and try again */ | |
dd7e0b7b | 3279 | if (!committed && !root->fs_info->open_ioctl_trans) { |
4e06bdd6 JB |
3280 | committed = 1; |
3281 | trans = btrfs_join_transaction(root, 1); | |
3282 | if (!trans) | |
3283 | return -ENOMEM; | |
3284 | ret = btrfs_commit_transaction(trans, root); | |
3285 | if (ret) | |
3286 | return ret; | |
3287 | goto again; | |
3288 | } | |
3289 | ||
6a63209f JB |
3290 | printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes" |
3291 | ", %llu bytes_used, %llu bytes_reserved, " | |
68f5a38c | 3292 | "%llu bytes_pinned, %llu bytes_readonly, %llu may use " |
21380931 JB |
3293 | "%llu total\n", (unsigned long long)bytes, |
3294 | (unsigned long long)data_sinfo->bytes_delalloc, | |
3295 | (unsigned long long)data_sinfo->bytes_used, | |
3296 | (unsigned long long)data_sinfo->bytes_reserved, | |
3297 | (unsigned long long)data_sinfo->bytes_pinned, | |
3298 | (unsigned long long)data_sinfo->bytes_readonly, | |
3299 | (unsigned long long)data_sinfo->bytes_may_use, | |
3300 | (unsigned long long)data_sinfo->total_bytes); | |
6a63209f JB |
3301 | return -ENOSPC; |
3302 | } | |
3303 | data_sinfo->bytes_may_use += bytes; | |
3304 | BTRFS_I(inode)->reserved_bytes += bytes; | |
3305 | spin_unlock(&data_sinfo->lock); | |
3306 | ||
9ed74f2d | 3307 | return 0; |
6a63209f JB |
3308 | } |
3309 | ||
3310 | /* | |
3311 | * if there was an error for whatever reason after calling | |
3312 | * btrfs_check_data_free_space, call this so we can cleanup the counters. | |
3313 | */ | |
3314 | void btrfs_free_reserved_data_space(struct btrfs_root *root, | |
3315 | struct inode *inode, u64 bytes) | |
3316 | { | |
3317 | struct btrfs_space_info *data_sinfo; | |
3318 | ||
3319 | /* make sure bytes are sectorsize aligned */ | |
3320 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
3321 | ||
3322 | data_sinfo = BTRFS_I(inode)->space_info; | |
3323 | spin_lock(&data_sinfo->lock); | |
3324 | data_sinfo->bytes_may_use -= bytes; | |
3325 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
3326 | spin_unlock(&data_sinfo->lock); | |
3327 | } | |
3328 | ||
3329 | /* called when we are adding a delalloc extent to the inode's io_tree */ | |
3330 | void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode, | |
3331 | u64 bytes) | |
3332 | { | |
3333 | struct btrfs_space_info *data_sinfo; | |
3334 | ||
3335 | /* get the space info for where this inode will be storing its data */ | |
3336 | data_sinfo = BTRFS_I(inode)->space_info; | |
3337 | ||
3338 | /* make sure we have enough space to handle the data first */ | |
3339 | spin_lock(&data_sinfo->lock); | |
3340 | data_sinfo->bytes_delalloc += bytes; | |
3341 | ||
3342 | /* | |
3343 | * we are adding a delalloc extent without calling | |
3344 | * btrfs_check_data_free_space first. This happens on a weird | |
3345 | * writepage condition, but shouldn't hurt our accounting | |
3346 | */ | |
3347 | if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) { | |
3348 | data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes; | |
3349 | BTRFS_I(inode)->reserved_bytes = 0; | |
3350 | } else { | |
3351 | data_sinfo->bytes_may_use -= bytes; | |
3352 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
3353 | } | |
3354 | ||
3355 | spin_unlock(&data_sinfo->lock); | |
3356 | } | |
3357 | ||
3358 | /* called when we are clearing an delalloc extent from the inode's io_tree */ | |
3359 | void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode, | |
3360 | u64 bytes) | |
3361 | { | |
3362 | struct btrfs_space_info *info; | |
3363 | ||
3364 | info = BTRFS_I(inode)->space_info; | |
3365 | ||
3366 | spin_lock(&info->lock); | |
3367 | info->bytes_delalloc -= bytes; | |
3368 | spin_unlock(&info->lock); | |
3369 | } | |
3370 | ||
97e728d4 JB |
3371 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
3372 | { | |
3373 | struct list_head *head = &info->space_info; | |
3374 | struct btrfs_space_info *found; | |
3375 | ||
3376 | rcu_read_lock(); | |
3377 | list_for_each_entry_rcu(found, head, list) { | |
3378 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
3379 | found->force_alloc = 1; | |
3380 | } | |
3381 | rcu_read_unlock(); | |
3382 | } | |
3383 | ||
6324fbf3 CM |
3384 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3385 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3386 | u64 flags, int force) |
6324fbf3 CM |
3387 | { |
3388 | struct btrfs_space_info *space_info; | |
97e728d4 | 3389 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6324fbf3 | 3390 | u64 thresh; |
c146afad YZ |
3391 | int ret = 0; |
3392 | ||
97e728d4 | 3393 | mutex_lock(&fs_info->chunk_mutex); |
6324fbf3 | 3394 | |
2b82032c | 3395 | flags = btrfs_reduce_alloc_profile(extent_root, flags); |
ec44a35c | 3396 | |
6324fbf3 | 3397 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3398 | if (!space_info) { |
3399 | ret = update_space_info(extent_root->fs_info, flags, | |
3400 | 0, 0, &space_info); | |
3401 | BUG_ON(ret); | |
3402 | } | |
6324fbf3 CM |
3403 | BUG_ON(!space_info); |
3404 | ||
25179201 | 3405 | spin_lock(&space_info->lock); |
9ed74f2d | 3406 | if (space_info->force_alloc) |
0ef3e66b | 3407 | force = 1; |
25179201 JB |
3408 | if (space_info->full) { |
3409 | spin_unlock(&space_info->lock); | |
925baedd | 3410 | goto out; |
25179201 | 3411 | } |
6324fbf3 | 3412 | |
c146afad | 3413 | thresh = space_info->total_bytes - space_info->bytes_readonly; |
9ed74f2d | 3414 | thresh = div_factor(thresh, 8); |
0ef3e66b | 3415 | if (!force && |
e8569813 | 3416 | (space_info->bytes_used + space_info->bytes_pinned + |
25179201 JB |
3417 | space_info->bytes_reserved + alloc_bytes) < thresh) { |
3418 | spin_unlock(&space_info->lock); | |
925baedd | 3419 | goto out; |
25179201 | 3420 | } |
25179201 JB |
3421 | spin_unlock(&space_info->lock); |
3422 | ||
97e728d4 JB |
3423 | /* |
3424 | * if we're doing a data chunk, go ahead and make sure that | |
3425 | * we keep a reasonable number of metadata chunks allocated in the | |
3426 | * FS as well. | |
3427 | */ | |
9ed74f2d | 3428 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3429 | fs_info->data_chunk_allocations++; |
3430 | if (!(fs_info->data_chunk_allocations % | |
3431 | fs_info->metadata_ratio)) | |
3432 | force_metadata_allocation(fs_info); | |
3433 | } | |
3434 | ||
2b82032c | 3435 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
9ed74f2d | 3436 | spin_lock(&space_info->lock); |
d397712b | 3437 | if (ret) |
6324fbf3 | 3438 | space_info->full = 1; |
9ed74f2d JB |
3439 | space_info->force_alloc = 0; |
3440 | spin_unlock(&space_info->lock); | |
a74a4b97 | 3441 | out: |
c146afad | 3442 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 3443 | return ret; |
6324fbf3 CM |
3444 | } |
3445 | ||
9078a3e1 CM |
3446 | static int update_block_group(struct btrfs_trans_handle *trans, |
3447 | struct btrfs_root *root, | |
db94535d | 3448 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 3449 | int mark_free) |
9078a3e1 CM |
3450 | { |
3451 | struct btrfs_block_group_cache *cache; | |
3452 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 3453 | u64 total = num_bytes; |
9078a3e1 | 3454 | u64 old_val; |
db94535d | 3455 | u64 byte_in_group; |
3e1ad54f | 3456 | |
5d4f98a2 YZ |
3457 | /* block accounting for super block */ |
3458 | spin_lock(&info->delalloc_lock); | |
3459 | old_val = btrfs_super_bytes_used(&info->super_copy); | |
3460 | if (alloc) | |
3461 | old_val += num_bytes; | |
3462 | else | |
3463 | old_val -= num_bytes; | |
3464 | btrfs_set_super_bytes_used(&info->super_copy, old_val); | |
5d4f98a2 YZ |
3465 | spin_unlock(&info->delalloc_lock); |
3466 | ||
d397712b | 3467 | while (total) { |
db94535d | 3468 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 3469 | if (!cache) |
9078a3e1 | 3470 | return -1; |
db94535d CM |
3471 | byte_in_group = bytenr - cache->key.objectid; |
3472 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 3473 | |
25179201 | 3474 | spin_lock(&cache->space_info->lock); |
c286ac48 | 3475 | spin_lock(&cache->lock); |
0f9dd46c | 3476 | cache->dirty = 1; |
9078a3e1 | 3477 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 3478 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 3479 | if (alloc) { |
db94535d | 3480 | old_val += num_bytes; |
11833d66 YZ |
3481 | btrfs_set_block_group_used(&cache->item, old_val); |
3482 | cache->reserved -= num_bytes; | |
6324fbf3 | 3483 | cache->space_info->bytes_used += num_bytes; |
11833d66 | 3484 | cache->space_info->bytes_reserved -= num_bytes; |
a512bbf8 | 3485 | if (cache->ro) |
c146afad | 3486 | cache->space_info->bytes_readonly -= num_bytes; |
c286ac48 | 3487 | spin_unlock(&cache->lock); |
25179201 | 3488 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 3489 | } else { |
db94535d | 3490 | old_val -= num_bytes; |
6324fbf3 | 3491 | cache->space_info->bytes_used -= num_bytes; |
c146afad YZ |
3492 | if (cache->ro) |
3493 | cache->space_info->bytes_readonly += num_bytes; | |
c286ac48 CM |
3494 | btrfs_set_block_group_used(&cache->item, old_val); |
3495 | spin_unlock(&cache->lock); | |
25179201 | 3496 | spin_unlock(&cache->space_info->lock); |
f510cfec | 3497 | if (mark_free) { |
0f9dd46c | 3498 | int ret; |
1f3c79a2 LH |
3499 | |
3500 | ret = btrfs_discard_extent(root, bytenr, | |
3501 | num_bytes); | |
3502 | WARN_ON(ret); | |
3503 | ||
0f9dd46c JB |
3504 | ret = btrfs_add_free_space(cache, bytenr, |
3505 | num_bytes); | |
d2fb3437 | 3506 | WARN_ON(ret); |
e37c9e69 | 3507 | } |
cd1bc465 | 3508 | } |
fa9c0d79 | 3509 | btrfs_put_block_group(cache); |
db94535d CM |
3510 | total -= num_bytes; |
3511 | bytenr += num_bytes; | |
9078a3e1 CM |
3512 | } |
3513 | return 0; | |
3514 | } | |
6324fbf3 | 3515 | |
a061fc8d CM |
3516 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
3517 | { | |
0f9dd46c | 3518 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 3519 | u64 bytenr; |
0f9dd46c JB |
3520 | |
3521 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
3522 | if (!cache) | |
a061fc8d | 3523 | return 0; |
0f9dd46c | 3524 | |
d2fb3437 | 3525 | bytenr = cache->key.objectid; |
fa9c0d79 | 3526 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
3527 | |
3528 | return bytenr; | |
a061fc8d CM |
3529 | } |
3530 | ||
11833d66 YZ |
3531 | /* |
3532 | * this function must be called within transaction | |
3533 | */ | |
3534 | int btrfs_pin_extent(struct btrfs_root *root, | |
3535 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 3536 | { |
324ae4df | 3537 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 | 3538 | struct btrfs_block_group_cache *cache; |
324ae4df | 3539 | |
11833d66 YZ |
3540 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
3541 | BUG_ON(!cache); | |
68b38550 | 3542 | |
11833d66 YZ |
3543 | spin_lock(&cache->space_info->lock); |
3544 | spin_lock(&cache->lock); | |
3545 | cache->pinned += num_bytes; | |
3546 | cache->space_info->bytes_pinned += num_bytes; | |
3547 | if (reserved) { | |
3548 | cache->reserved -= num_bytes; | |
3549 | cache->space_info->bytes_reserved -= num_bytes; | |
3550 | } | |
3551 | spin_unlock(&cache->lock); | |
3552 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 3553 | |
11833d66 | 3554 | btrfs_put_block_group(cache); |
68b38550 | 3555 | |
11833d66 YZ |
3556 | set_extent_dirty(fs_info->pinned_extents, |
3557 | bytenr, bytenr + num_bytes - 1, GFP_NOFS); | |
3558 | return 0; | |
3559 | } | |
3560 | ||
3561 | static int update_reserved_extents(struct btrfs_block_group_cache *cache, | |
3562 | u64 num_bytes, int reserve) | |
3563 | { | |
3564 | spin_lock(&cache->space_info->lock); | |
3565 | spin_lock(&cache->lock); | |
3566 | if (reserve) { | |
3567 | cache->reserved += num_bytes; | |
3568 | cache->space_info->bytes_reserved += num_bytes; | |
3569 | } else { | |
3570 | cache->reserved -= num_bytes; | |
3571 | cache->space_info->bytes_reserved -= num_bytes; | |
324ae4df | 3572 | } |
11833d66 YZ |
3573 | spin_unlock(&cache->lock); |
3574 | spin_unlock(&cache->space_info->lock); | |
324ae4df Y |
3575 | return 0; |
3576 | } | |
9078a3e1 | 3577 | |
11833d66 YZ |
3578 | int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
3579 | struct btrfs_root *root) | |
e8569813 | 3580 | { |
e8569813 | 3581 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
3582 | struct btrfs_caching_control *next; |
3583 | struct btrfs_caching_control *caching_ctl; | |
3584 | struct btrfs_block_group_cache *cache; | |
e8569813 | 3585 | |
11833d66 | 3586 | down_write(&fs_info->extent_commit_sem); |
25179201 | 3587 | |
11833d66 YZ |
3588 | list_for_each_entry_safe(caching_ctl, next, |
3589 | &fs_info->caching_block_groups, list) { | |
3590 | cache = caching_ctl->block_group; | |
3591 | if (block_group_cache_done(cache)) { | |
3592 | cache->last_byte_to_unpin = (u64)-1; | |
3593 | list_del_init(&caching_ctl->list); | |
3594 | put_caching_control(caching_ctl); | |
e8569813 | 3595 | } else { |
11833d66 | 3596 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 3597 | } |
e8569813 | 3598 | } |
11833d66 YZ |
3599 | |
3600 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
3601 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
3602 | else | |
3603 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
3604 | ||
3605 | up_write(&fs_info->extent_commit_sem); | |
e8569813 ZY |
3606 | return 0; |
3607 | } | |
3608 | ||
11833d66 | 3609 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 3610 | { |
11833d66 YZ |
3611 | struct btrfs_fs_info *fs_info = root->fs_info; |
3612 | struct btrfs_block_group_cache *cache = NULL; | |
3613 | u64 len; | |
ccd467d6 | 3614 | |
11833d66 YZ |
3615 | while (start <= end) { |
3616 | if (!cache || | |
3617 | start >= cache->key.objectid + cache->key.offset) { | |
3618 | if (cache) | |
3619 | btrfs_put_block_group(cache); | |
3620 | cache = btrfs_lookup_block_group(fs_info, start); | |
3621 | BUG_ON(!cache); | |
3622 | } | |
3623 | ||
3624 | len = cache->key.objectid + cache->key.offset - start; | |
3625 | len = min(len, end + 1 - start); | |
3626 | ||
3627 | if (start < cache->last_byte_to_unpin) { | |
3628 | len = min(len, cache->last_byte_to_unpin - start); | |
3629 | btrfs_add_free_space(cache, start, len); | |
3630 | } | |
3631 | ||
3632 | spin_lock(&cache->space_info->lock); | |
3633 | spin_lock(&cache->lock); | |
3634 | cache->pinned -= len; | |
3635 | cache->space_info->bytes_pinned -= len; | |
3636 | spin_unlock(&cache->lock); | |
3637 | spin_unlock(&cache->space_info->lock); | |
817d52f8 | 3638 | |
11833d66 | 3639 | start += len; |
ccd467d6 | 3640 | } |
11833d66 YZ |
3641 | |
3642 | if (cache) | |
3643 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
3644 | return 0; |
3645 | } | |
3646 | ||
3647 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 3648 | struct btrfs_root *root) |
a28ec197 | 3649 | { |
11833d66 YZ |
3650 | struct btrfs_fs_info *fs_info = root->fs_info; |
3651 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
3652 | u64 start; |
3653 | u64 end; | |
a28ec197 | 3654 | int ret; |
a28ec197 | 3655 | |
11833d66 YZ |
3656 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
3657 | unpin = &fs_info->freed_extents[1]; | |
3658 | else | |
3659 | unpin = &fs_info->freed_extents[0]; | |
3660 | ||
d397712b | 3661 | while (1) { |
1a5bc167 CM |
3662 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
3663 | EXTENT_DIRTY); | |
3664 | if (ret) | |
a28ec197 | 3665 | break; |
1f3c79a2 LH |
3666 | |
3667 | ret = btrfs_discard_extent(root, start, end + 1 - start); | |
3668 | ||
1a5bc167 | 3669 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 3670 | unpin_extent_range(root, start, end); |
b9473439 | 3671 | cond_resched(); |
a28ec197 | 3672 | } |
817d52f8 | 3673 | |
1f3c79a2 | 3674 | return ret; |
a28ec197 CM |
3675 | } |
3676 | ||
31840ae1 ZY |
3677 | static int pin_down_bytes(struct btrfs_trans_handle *trans, |
3678 | struct btrfs_root *root, | |
b9473439 | 3679 | struct btrfs_path *path, |
11833d66 YZ |
3680 | u64 bytenr, u64 num_bytes, |
3681 | int is_data, int reserved, | |
b9473439 | 3682 | struct extent_buffer **must_clean) |
e20d96d6 | 3683 | { |
1a5bc167 | 3684 | int err = 0; |
31840ae1 | 3685 | struct extent_buffer *buf; |
8ef97622 | 3686 | |
31840ae1 ZY |
3687 | if (is_data) |
3688 | goto pinit; | |
3689 | ||
444528b3 CM |
3690 | /* |
3691 | * discard is sloooow, and so triggering discards on | |
3692 | * individual btree blocks isn't a good plan. Just | |
3693 | * pin everything in discard mode. | |
3694 | */ | |
3695 | if (btrfs_test_opt(root, DISCARD)) | |
3696 | goto pinit; | |
3697 | ||
31840ae1 ZY |
3698 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); |
3699 | if (!buf) | |
3700 | goto pinit; | |
3701 | ||
3702 | /* we can reuse a block if it hasn't been written | |
3703 | * and it is from this transaction. We can't | |
3704 | * reuse anything from the tree log root because | |
3705 | * it has tiny sub-transactions. | |
3706 | */ | |
3707 | if (btrfs_buffer_uptodate(buf, 0) && | |
3708 | btrfs_try_tree_lock(buf)) { | |
3709 | u64 header_owner = btrfs_header_owner(buf); | |
3710 | u64 header_transid = btrfs_header_generation(buf); | |
3711 | if (header_owner != BTRFS_TREE_LOG_OBJECTID && | |
3712 | header_transid == trans->transid && | |
3713 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
b9473439 | 3714 | *must_clean = buf; |
31840ae1 | 3715 | return 1; |
8ef97622 | 3716 | } |
31840ae1 | 3717 | btrfs_tree_unlock(buf); |
f4b9aa8d | 3718 | } |
31840ae1 ZY |
3719 | free_extent_buffer(buf); |
3720 | pinit: | |
11833d66 YZ |
3721 | if (path) |
3722 | btrfs_set_path_blocking(path); | |
b9473439 | 3723 | /* unlocks the pinned mutex */ |
11833d66 | 3724 | btrfs_pin_extent(root, bytenr, num_bytes, reserved); |
31840ae1 | 3725 | |
be744175 | 3726 | BUG_ON(err < 0); |
e20d96d6 CM |
3727 | return 0; |
3728 | } | |
3729 | ||
5d4f98a2 YZ |
3730 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
3731 | struct btrfs_root *root, | |
3732 | u64 bytenr, u64 num_bytes, u64 parent, | |
3733 | u64 root_objectid, u64 owner_objectid, | |
3734 | u64 owner_offset, int refs_to_drop, | |
3735 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 3736 | { |
e2fa7227 | 3737 | struct btrfs_key key; |
5d4f98a2 | 3738 | struct btrfs_path *path; |
1261ec42 CM |
3739 | struct btrfs_fs_info *info = root->fs_info; |
3740 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 3741 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3742 | struct btrfs_extent_item *ei; |
3743 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 3744 | int ret; |
5d4f98a2 | 3745 | int is_data; |
952fccac CM |
3746 | int extent_slot = 0; |
3747 | int found_extent = 0; | |
3748 | int num_to_del = 1; | |
5d4f98a2 YZ |
3749 | u32 item_size; |
3750 | u64 refs; | |
037e6390 | 3751 | |
5caf2a00 | 3752 | path = btrfs_alloc_path(); |
54aa1f4d CM |
3753 | if (!path) |
3754 | return -ENOMEM; | |
5f26f772 | 3755 | |
3c12ac72 | 3756 | path->reada = 1; |
b9473439 | 3757 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3758 | |
3759 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
3760 | BUG_ON(!is_data && refs_to_drop != 1); | |
3761 | ||
3762 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
3763 | bytenr, num_bytes, parent, | |
3764 | root_objectid, owner_objectid, | |
3765 | owner_offset); | |
7bb86316 | 3766 | if (ret == 0) { |
952fccac | 3767 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
3768 | while (extent_slot >= 0) { |
3769 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 3770 | extent_slot); |
5d4f98a2 | 3771 | if (key.objectid != bytenr) |
952fccac | 3772 | break; |
5d4f98a2 YZ |
3773 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
3774 | key.offset == num_bytes) { | |
952fccac CM |
3775 | found_extent = 1; |
3776 | break; | |
3777 | } | |
3778 | if (path->slots[0] - extent_slot > 5) | |
3779 | break; | |
5d4f98a2 | 3780 | extent_slot--; |
952fccac | 3781 | } |
5d4f98a2 YZ |
3782 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
3783 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
3784 | if (found_extent && item_size < sizeof(*ei)) | |
3785 | found_extent = 0; | |
3786 | #endif | |
31840ae1 | 3787 | if (!found_extent) { |
5d4f98a2 | 3788 | BUG_ON(iref); |
56bec294 | 3789 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
3790 | NULL, refs_to_drop, |
3791 | is_data); | |
31840ae1 ZY |
3792 | BUG_ON(ret); |
3793 | btrfs_release_path(extent_root, path); | |
b9473439 | 3794 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3795 | |
3796 | key.objectid = bytenr; | |
3797 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3798 | key.offset = num_bytes; | |
3799 | ||
31840ae1 ZY |
3800 | ret = btrfs_search_slot(trans, extent_root, |
3801 | &key, path, -1, 1); | |
f3465ca4 JB |
3802 | if (ret) { |
3803 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
3804 | ", was looking for %llu\n", ret, |
3805 | (unsigned long long)bytenr); | |
f3465ca4 JB |
3806 | btrfs_print_leaf(extent_root, path->nodes[0]); |
3807 | } | |
31840ae1 ZY |
3808 | BUG_ON(ret); |
3809 | extent_slot = path->slots[0]; | |
3810 | } | |
7bb86316 CM |
3811 | } else { |
3812 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
3813 | WARN_ON(1); | |
d397712b | 3814 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 3815 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 3816 | (unsigned long long)bytenr, |
56bec294 | 3817 | (unsigned long long)parent, |
d397712b | 3818 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
3819 | (unsigned long long)owner_objectid, |
3820 | (unsigned long long)owner_offset); | |
7bb86316 | 3821 | } |
5f39d397 CM |
3822 | |
3823 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
3824 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
3825 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3826 | if (item_size < sizeof(*ei)) { | |
3827 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
3828 | ret = convert_extent_item_v0(trans, extent_root, path, | |
3829 | owner_objectid, 0); | |
3830 | BUG_ON(ret < 0); | |
3831 | ||
3832 | btrfs_release_path(extent_root, path); | |
3833 | path->leave_spinning = 1; | |
3834 | ||
3835 | key.objectid = bytenr; | |
3836 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3837 | key.offset = num_bytes; | |
3838 | ||
3839 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
3840 | -1, 1); | |
3841 | if (ret) { | |
3842 | printk(KERN_ERR "umm, got %d back from search" | |
3843 | ", was looking for %llu\n", ret, | |
3844 | (unsigned long long)bytenr); | |
3845 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
3846 | } | |
3847 | BUG_ON(ret); | |
3848 | extent_slot = path->slots[0]; | |
3849 | leaf = path->nodes[0]; | |
3850 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
3851 | } | |
3852 | #endif | |
3853 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 3854 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3855 | struct btrfs_extent_item); |
5d4f98a2 YZ |
3856 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
3857 | struct btrfs_tree_block_info *bi; | |
3858 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
3859 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
3860 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3861 | } | |
56bec294 | 3862 | |
5d4f98a2 | 3863 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
3864 | BUG_ON(refs < refs_to_drop); |
3865 | refs -= refs_to_drop; | |
5f39d397 | 3866 | |
5d4f98a2 YZ |
3867 | if (refs > 0) { |
3868 | if (extent_op) | |
3869 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3870 | /* | |
3871 | * In the case of inline back ref, reference count will | |
3872 | * be updated by remove_extent_backref | |
952fccac | 3873 | */ |
5d4f98a2 YZ |
3874 | if (iref) { |
3875 | BUG_ON(!found_extent); | |
3876 | } else { | |
3877 | btrfs_set_extent_refs(leaf, ei, refs); | |
3878 | btrfs_mark_buffer_dirty(leaf); | |
3879 | } | |
3880 | if (found_extent) { | |
3881 | ret = remove_extent_backref(trans, extent_root, path, | |
3882 | iref, refs_to_drop, | |
3883 | is_data); | |
952fccac CM |
3884 | BUG_ON(ret); |
3885 | } | |
5d4f98a2 YZ |
3886 | } else { |
3887 | int mark_free = 0; | |
b9473439 | 3888 | struct extent_buffer *must_clean = NULL; |
78fae27e | 3889 | |
5d4f98a2 YZ |
3890 | if (found_extent) { |
3891 | BUG_ON(is_data && refs_to_drop != | |
3892 | extent_data_ref_count(root, path, iref)); | |
3893 | if (iref) { | |
3894 | BUG_ON(path->slots[0] != extent_slot); | |
3895 | } else { | |
3896 | BUG_ON(path->slots[0] != extent_slot + 1); | |
3897 | path->slots[0] = extent_slot; | |
3898 | num_to_del = 2; | |
3899 | } | |
78fae27e | 3900 | } |
b9473439 | 3901 | |
5d4f98a2 | 3902 | ret = pin_down_bytes(trans, root, path, bytenr, |
11833d66 | 3903 | num_bytes, is_data, 0, &must_clean); |
5d4f98a2 YZ |
3904 | if (ret > 0) |
3905 | mark_free = 1; | |
3906 | BUG_ON(ret < 0); | |
b9473439 CM |
3907 | /* |
3908 | * it is going to be very rare for someone to be waiting | |
3909 | * on the block we're freeing. del_items might need to | |
3910 | * schedule, so rather than get fancy, just force it | |
3911 | * to blocking here | |
3912 | */ | |
3913 | if (must_clean) | |
3914 | btrfs_set_lock_blocking(must_clean); | |
3915 | ||
952fccac CM |
3916 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3917 | num_to_del); | |
31840ae1 | 3918 | BUG_ON(ret); |
25179201 | 3919 | btrfs_release_path(extent_root, path); |
21af804c | 3920 | |
b9473439 CM |
3921 | if (must_clean) { |
3922 | clean_tree_block(NULL, root, must_clean); | |
3923 | btrfs_tree_unlock(must_clean); | |
3924 | free_extent_buffer(must_clean); | |
3925 | } | |
3926 | ||
5d4f98a2 | 3927 | if (is_data) { |
459931ec CM |
3928 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
3929 | BUG_ON(ret); | |
d57e62b8 CM |
3930 | } else { |
3931 | invalidate_mapping_pages(info->btree_inode->i_mapping, | |
3932 | bytenr >> PAGE_CACHE_SHIFT, | |
3933 | (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT); | |
459931ec CM |
3934 | } |
3935 | ||
dcbdd4dc CM |
3936 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
3937 | mark_free); | |
3938 | BUG_ON(ret); | |
a28ec197 | 3939 | } |
5caf2a00 | 3940 | btrfs_free_path(path); |
a28ec197 CM |
3941 | return ret; |
3942 | } | |
3943 | ||
1887be66 CM |
3944 | /* |
3945 | * when we free an extent, it is possible (and likely) that we free the last | |
3946 | * delayed ref for that extent as well. This searches the delayed ref tree for | |
3947 | * a given extent, and if there are no other delayed refs to be processed, it | |
3948 | * removes it from the tree. | |
3949 | */ | |
3950 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
3951 | struct btrfs_root *root, u64 bytenr) | |
3952 | { | |
3953 | struct btrfs_delayed_ref_head *head; | |
3954 | struct btrfs_delayed_ref_root *delayed_refs; | |
3955 | struct btrfs_delayed_ref_node *ref; | |
3956 | struct rb_node *node; | |
3957 | int ret; | |
3958 | ||
3959 | delayed_refs = &trans->transaction->delayed_refs; | |
3960 | spin_lock(&delayed_refs->lock); | |
3961 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
3962 | if (!head) | |
3963 | goto out; | |
3964 | ||
3965 | node = rb_prev(&head->node.rb_node); | |
3966 | if (!node) | |
3967 | goto out; | |
3968 | ||
3969 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
3970 | ||
3971 | /* there are still entries for this ref, we can't drop it */ | |
3972 | if (ref->bytenr == bytenr) | |
3973 | goto out; | |
3974 | ||
5d4f98a2 YZ |
3975 | if (head->extent_op) { |
3976 | if (!head->must_insert_reserved) | |
3977 | goto out; | |
3978 | kfree(head->extent_op); | |
3979 | head->extent_op = NULL; | |
3980 | } | |
3981 | ||
1887be66 CM |
3982 | /* |
3983 | * waiting for the lock here would deadlock. If someone else has it | |
3984 | * locked they are already in the process of dropping it anyway | |
3985 | */ | |
3986 | if (!mutex_trylock(&head->mutex)) | |
3987 | goto out; | |
3988 | ||
3989 | /* | |
3990 | * at this point we have a head with no other entries. Go | |
3991 | * ahead and process it. | |
3992 | */ | |
3993 | head->node.in_tree = 0; | |
3994 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 3995 | |
1887be66 CM |
3996 | delayed_refs->num_entries--; |
3997 | ||
3998 | /* | |
3999 | * we don't take a ref on the node because we're removing it from the | |
4000 | * tree, so we just steal the ref the tree was holding. | |
4001 | */ | |
c3e69d58 CM |
4002 | delayed_refs->num_heads--; |
4003 | if (list_empty(&head->cluster)) | |
4004 | delayed_refs->num_heads_ready--; | |
4005 | ||
4006 | list_del_init(&head->cluster); | |
1887be66 CM |
4007 | spin_unlock(&delayed_refs->lock); |
4008 | ||
4009 | ret = run_one_delayed_ref(trans, root->fs_info->tree_root, | |
5d4f98a2 YZ |
4010 | &head->node, head->extent_op, |
4011 | head->must_insert_reserved); | |
1887be66 CM |
4012 | BUG_ON(ret); |
4013 | btrfs_put_delayed_ref(&head->node); | |
4014 | return 0; | |
4015 | out: | |
4016 | spin_unlock(&delayed_refs->lock); | |
4017 | return 0; | |
4018 | } | |
4019 | ||
925baedd | 4020 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
4021 | struct btrfs_root *root, |
4022 | u64 bytenr, u64 num_bytes, u64 parent, | |
5d4f98a2 | 4023 | u64 root_objectid, u64 owner, u64 offset) |
925baedd CM |
4024 | { |
4025 | int ret; | |
4026 | ||
56bec294 CM |
4027 | /* |
4028 | * tree log blocks never actually go into the extent allocation | |
4029 | * tree, just update pinning info and exit early. | |
56bec294 | 4030 | */ |
5d4f98a2 YZ |
4031 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
4032 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 4033 | /* unlocks the pinned mutex */ |
11833d66 | 4034 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 4035 | ret = 0; |
5d4f98a2 YZ |
4036 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
4037 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
4038 | parent, root_objectid, (int)owner, | |
4039 | BTRFS_DROP_DELAYED_REF, NULL); | |
1887be66 CM |
4040 | BUG_ON(ret); |
4041 | ret = check_ref_cleanup(trans, root, bytenr); | |
4042 | BUG_ON(ret); | |
5d4f98a2 YZ |
4043 | } else { |
4044 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
4045 | parent, root_objectid, owner, | |
4046 | offset, BTRFS_DROP_DELAYED_REF, NULL); | |
4047 | BUG_ON(ret); | |
56bec294 | 4048 | } |
925baedd CM |
4049 | return ret; |
4050 | } | |
4051 | ||
86b9f2ec YZ |
4052 | int btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
4053 | struct btrfs_root *root, | |
4054 | u64 bytenr, u32 blocksize, | |
4055 | u64 parent, u64 root_objectid, int level) | |
4056 | { | |
4057 | u64 used; | |
4058 | spin_lock(&root->node_lock); | |
4059 | used = btrfs_root_used(&root->root_item) - blocksize; | |
4060 | btrfs_set_root_used(&root->root_item, used); | |
4061 | spin_unlock(&root->node_lock); | |
4062 | ||
4063 | return btrfs_free_extent(trans, root, bytenr, blocksize, | |
4064 | parent, root_objectid, level, 0); | |
4065 | } | |
4066 | ||
87ee04eb CM |
4067 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
4068 | { | |
4069 | u64 mask = ((u64)root->stripesize - 1); | |
4070 | u64 ret = (val + mask) & ~mask; | |
4071 | return ret; | |
4072 | } | |
4073 | ||
817d52f8 JB |
4074 | /* |
4075 | * when we wait for progress in the block group caching, its because | |
4076 | * our allocation attempt failed at least once. So, we must sleep | |
4077 | * and let some progress happen before we try again. | |
4078 | * | |
4079 | * This function will sleep at least once waiting for new free space to | |
4080 | * show up, and then it will check the block group free space numbers | |
4081 | * for our min num_bytes. Another option is to have it go ahead | |
4082 | * and look in the rbtree for a free extent of a given size, but this | |
4083 | * is a good start. | |
4084 | */ | |
4085 | static noinline int | |
4086 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
4087 | u64 num_bytes) | |
4088 | { | |
11833d66 | 4089 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
4090 | DEFINE_WAIT(wait); |
4091 | ||
11833d66 YZ |
4092 | caching_ctl = get_caching_control(cache); |
4093 | if (!caching_ctl) | |
817d52f8 | 4094 | return 0; |
817d52f8 | 4095 | |
11833d66 | 4096 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
817d52f8 | 4097 | (cache->free_space >= num_bytes)); |
11833d66 YZ |
4098 | |
4099 | put_caching_control(caching_ctl); | |
4100 | return 0; | |
4101 | } | |
4102 | ||
4103 | static noinline int | |
4104 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
4105 | { | |
4106 | struct btrfs_caching_control *caching_ctl; | |
4107 | DEFINE_WAIT(wait); | |
4108 | ||
4109 | caching_ctl = get_caching_control(cache); | |
4110 | if (!caching_ctl) | |
4111 | return 0; | |
4112 | ||
4113 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
4114 | ||
4115 | put_caching_control(caching_ctl); | |
817d52f8 JB |
4116 | return 0; |
4117 | } | |
4118 | ||
4119 | enum btrfs_loop_type { | |
ccf0e725 | 4120 | LOOP_FIND_IDEAL = 0, |
817d52f8 JB |
4121 | LOOP_CACHING_NOWAIT = 1, |
4122 | LOOP_CACHING_WAIT = 2, | |
4123 | LOOP_ALLOC_CHUNK = 3, | |
4124 | LOOP_NO_EMPTY_SIZE = 4, | |
4125 | }; | |
4126 | ||
fec577fb CM |
4127 | /* |
4128 | * walks the btree of allocated extents and find a hole of a given size. | |
4129 | * The key ins is changed to record the hole: | |
4130 | * ins->objectid == block start | |
62e2749e | 4131 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
4132 | * ins->offset == number of blocks |
4133 | * Any available blocks before search_start are skipped. | |
4134 | */ | |
d397712b | 4135 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
4136 | struct btrfs_root *orig_root, |
4137 | u64 num_bytes, u64 empty_size, | |
4138 | u64 search_start, u64 search_end, | |
4139 | u64 hint_byte, struct btrfs_key *ins, | |
4140 | u64 exclude_start, u64 exclude_nr, | |
4141 | int data) | |
fec577fb | 4142 | { |
80eb234a | 4143 | int ret = 0; |
d397712b | 4144 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 4145 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 4146 | struct btrfs_block_group_cache *block_group = NULL; |
239b14b3 | 4147 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 4148 | int allowed_chunk_alloc = 0; |
ccf0e725 | 4149 | int done_chunk_alloc = 0; |
80eb234a | 4150 | struct btrfs_space_info *space_info; |
fa9c0d79 CM |
4151 | int last_ptr_loop = 0; |
4152 | int loop = 0; | |
817d52f8 | 4153 | bool found_uncached_bg = false; |
0a24325e | 4154 | bool failed_cluster_refill = false; |
1cdda9b8 | 4155 | bool failed_alloc = false; |
ccf0e725 JB |
4156 | u64 ideal_cache_percent = 0; |
4157 | u64 ideal_cache_offset = 0; | |
fec577fb | 4158 | |
db94535d | 4159 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 4160 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
4161 | ins->objectid = 0; |
4162 | ins->offset = 0; | |
b1a4d965 | 4163 | |
2552d17e JB |
4164 | space_info = __find_space_info(root->fs_info, data); |
4165 | ||
0ef3e66b CM |
4166 | if (orig_root->ref_cows || empty_size) |
4167 | allowed_chunk_alloc = 1; | |
4168 | ||
239b14b3 | 4169 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
fa9c0d79 | 4170 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
4171 | if (!btrfs_test_opt(root, SSD)) |
4172 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
4173 | } |
4174 | ||
fa9c0d79 CM |
4175 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { |
4176 | last_ptr = &root->fs_info->data_alloc_cluster; | |
4177 | } | |
0f9dd46c | 4178 | |
239b14b3 | 4179 | if (last_ptr) { |
fa9c0d79 CM |
4180 | spin_lock(&last_ptr->lock); |
4181 | if (last_ptr->block_group) | |
4182 | hint_byte = last_ptr->window_start; | |
4183 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 4184 | } |
fa9c0d79 | 4185 | |
a061fc8d | 4186 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 4187 | search_start = max(search_start, hint_byte); |
0b86a832 | 4188 | |
817d52f8 | 4189 | if (!last_ptr) |
fa9c0d79 | 4190 | empty_cluster = 0; |
fa9c0d79 | 4191 | |
2552d17e | 4192 | if (search_start == hint_byte) { |
ccf0e725 | 4193 | ideal_cache: |
2552d17e JB |
4194 | block_group = btrfs_lookup_block_group(root->fs_info, |
4195 | search_start); | |
817d52f8 JB |
4196 | /* |
4197 | * we don't want to use the block group if it doesn't match our | |
4198 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4199 | * |
4200 | * However if we are re-searching with an ideal block group | |
4201 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
4202 | */ |
4203 | if (block_group && block_group_bits(block_group, data) && | |
ccf0e725 JB |
4204 | (block_group->cached != BTRFS_CACHE_NO || |
4205 | search_start == ideal_cache_offset)) { | |
2552d17e | 4206 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4207 | if (list_empty(&block_group->list) || |
4208 | block_group->ro) { | |
4209 | /* | |
4210 | * someone is removing this block group, | |
4211 | * we can't jump into the have_block_group | |
4212 | * target because our list pointers are not | |
4213 | * valid | |
4214 | */ | |
4215 | btrfs_put_block_group(block_group); | |
4216 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4217 | } else { |
44fb5511 | 4218 | goto have_block_group; |
ccf0e725 | 4219 | } |
2552d17e | 4220 | } else if (block_group) { |
fa9c0d79 | 4221 | btrfs_put_block_group(block_group); |
2552d17e | 4222 | } |
42e70e7a | 4223 | } |
2552d17e | 4224 | search: |
80eb234a | 4225 | down_read(&space_info->groups_sem); |
2552d17e | 4226 | list_for_each_entry(block_group, &space_info->block_groups, list) { |
6226cb0a | 4227 | u64 offset; |
817d52f8 | 4228 | int cached; |
8a1413a2 | 4229 | |
2552d17e JB |
4230 | atomic_inc(&block_group->count); |
4231 | search_start = block_group->key.objectid; | |
42e70e7a | 4232 | |
2552d17e | 4233 | have_block_group: |
817d52f8 | 4234 | if (unlikely(block_group->cached == BTRFS_CACHE_NO)) { |
ccf0e725 JB |
4235 | u64 free_percent; |
4236 | ||
4237 | free_percent = btrfs_block_group_used(&block_group->item); | |
4238 | free_percent *= 100; | |
4239 | free_percent = div64_u64(free_percent, | |
4240 | block_group->key.offset); | |
4241 | free_percent = 100 - free_percent; | |
4242 | if (free_percent > ideal_cache_percent && | |
4243 | likely(!block_group->ro)) { | |
4244 | ideal_cache_offset = block_group->key.objectid; | |
4245 | ideal_cache_percent = free_percent; | |
4246 | } | |
4247 | ||
817d52f8 | 4248 | /* |
ccf0e725 JB |
4249 | * We only want to start kthread caching if we are at |
4250 | * the point where we will wait for caching to make | |
4251 | * progress, or if our ideal search is over and we've | |
4252 | * found somebody to start caching. | |
817d52f8 JB |
4253 | */ |
4254 | if (loop > LOOP_CACHING_NOWAIT || | |
ccf0e725 JB |
4255 | (loop > LOOP_FIND_IDEAL && |
4256 | atomic_read(&space_info->caching_threads) < 2)) { | |
817d52f8 JB |
4257 | ret = cache_block_group(block_group); |
4258 | BUG_ON(ret); | |
2552d17e | 4259 | } |
817d52f8 JB |
4260 | found_uncached_bg = true; |
4261 | ||
ccf0e725 JB |
4262 | /* |
4263 | * If loop is set for cached only, try the next block | |
4264 | * group. | |
4265 | */ | |
4266 | if (loop == LOOP_FIND_IDEAL) | |
817d52f8 JB |
4267 | goto loop; |
4268 | } | |
4269 | ||
ccf0e725 JB |
4270 | cached = block_group_cache_done(block_group); |
4271 | if (unlikely(!cached)) | |
4272 | found_uncached_bg = true; | |
4273 | ||
ea6a478e | 4274 | if (unlikely(block_group->ro)) |
2552d17e | 4275 | goto loop; |
0f9dd46c | 4276 | |
0a24325e JB |
4277 | /* |
4278 | * Ok we want to try and use the cluster allocator, so lets look | |
4279 | * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will | |
4280 | * have tried the cluster allocator plenty of times at this | |
4281 | * point and not have found anything, so we are likely way too | |
4282 | * fragmented for the clustering stuff to find anything, so lets | |
4283 | * just skip it and let the allocator find whatever block it can | |
4284 | * find | |
4285 | */ | |
4286 | if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) { | |
fa9c0d79 CM |
4287 | /* |
4288 | * the refill lock keeps out other | |
4289 | * people trying to start a new cluster | |
4290 | */ | |
4291 | spin_lock(&last_ptr->refill_lock); | |
44fb5511 CM |
4292 | if (last_ptr->block_group && |
4293 | (last_ptr->block_group->ro || | |
4294 | !block_group_bits(last_ptr->block_group, data))) { | |
4295 | offset = 0; | |
4296 | goto refill_cluster; | |
4297 | } | |
4298 | ||
fa9c0d79 CM |
4299 | offset = btrfs_alloc_from_cluster(block_group, last_ptr, |
4300 | num_bytes, search_start); | |
4301 | if (offset) { | |
4302 | /* we have a block, we're done */ | |
4303 | spin_unlock(&last_ptr->refill_lock); | |
4304 | goto checks; | |
4305 | } | |
4306 | ||
4307 | spin_lock(&last_ptr->lock); | |
4308 | /* | |
4309 | * whoops, this cluster doesn't actually point to | |
4310 | * this block group. Get a ref on the block | |
4311 | * group is does point to and try again | |
4312 | */ | |
4313 | if (!last_ptr_loop && last_ptr->block_group && | |
4314 | last_ptr->block_group != block_group) { | |
4315 | ||
4316 | btrfs_put_block_group(block_group); | |
4317 | block_group = last_ptr->block_group; | |
4318 | atomic_inc(&block_group->count); | |
4319 | spin_unlock(&last_ptr->lock); | |
4320 | spin_unlock(&last_ptr->refill_lock); | |
4321 | ||
4322 | last_ptr_loop = 1; | |
4323 | search_start = block_group->key.objectid; | |
44fb5511 CM |
4324 | /* |
4325 | * we know this block group is properly | |
4326 | * in the list because | |
4327 | * btrfs_remove_block_group, drops the | |
4328 | * cluster before it removes the block | |
4329 | * group from the list | |
4330 | */ | |
fa9c0d79 CM |
4331 | goto have_block_group; |
4332 | } | |
4333 | spin_unlock(&last_ptr->lock); | |
44fb5511 | 4334 | refill_cluster: |
fa9c0d79 CM |
4335 | /* |
4336 | * this cluster didn't work out, free it and | |
4337 | * start over | |
4338 | */ | |
4339 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
4340 | ||
4341 | last_ptr_loop = 0; | |
4342 | ||
4343 | /* allocate a cluster in this block group */ | |
451d7585 | 4344 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 CM |
4345 | block_group, last_ptr, |
4346 | offset, num_bytes, | |
4347 | empty_cluster + empty_size); | |
4348 | if (ret == 0) { | |
4349 | /* | |
4350 | * now pull our allocation out of this | |
4351 | * cluster | |
4352 | */ | |
4353 | offset = btrfs_alloc_from_cluster(block_group, | |
4354 | last_ptr, num_bytes, | |
4355 | search_start); | |
4356 | if (offset) { | |
4357 | /* we found one, proceed */ | |
4358 | spin_unlock(&last_ptr->refill_lock); | |
4359 | goto checks; | |
4360 | } | |
0a24325e JB |
4361 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
4362 | && !failed_cluster_refill) { | |
817d52f8 JB |
4363 | spin_unlock(&last_ptr->refill_lock); |
4364 | ||
0a24325e | 4365 | failed_cluster_refill = true; |
817d52f8 JB |
4366 | wait_block_group_cache_progress(block_group, |
4367 | num_bytes + empty_cluster + empty_size); | |
4368 | goto have_block_group; | |
fa9c0d79 | 4369 | } |
817d52f8 | 4370 | |
fa9c0d79 CM |
4371 | /* |
4372 | * at this point we either didn't find a cluster | |
4373 | * or we weren't able to allocate a block from our | |
4374 | * cluster. Free the cluster we've been trying | |
4375 | * to use, and go to the next block group | |
4376 | */ | |
0a24325e | 4377 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 4378 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 4379 | goto loop; |
fa9c0d79 CM |
4380 | } |
4381 | ||
6226cb0a JB |
4382 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
4383 | num_bytes, empty_size); | |
1cdda9b8 JB |
4384 | /* |
4385 | * If we didn't find a chunk, and we haven't failed on this | |
4386 | * block group before, and this block group is in the middle of | |
4387 | * caching and we are ok with waiting, then go ahead and wait | |
4388 | * for progress to be made, and set failed_alloc to true. | |
4389 | * | |
4390 | * If failed_alloc is true then we've already waited on this | |
4391 | * block group once and should move on to the next block group. | |
4392 | */ | |
4393 | if (!offset && !failed_alloc && !cached && | |
4394 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 4395 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
4396 | num_bytes + empty_size); |
4397 | failed_alloc = true; | |
817d52f8 | 4398 | goto have_block_group; |
1cdda9b8 JB |
4399 | } else if (!offset) { |
4400 | goto loop; | |
817d52f8 | 4401 | } |
fa9c0d79 | 4402 | checks: |
6226cb0a | 4403 | search_start = stripe_align(root, offset); |
2552d17e | 4404 | /* move on to the next group */ |
6226cb0a JB |
4405 | if (search_start + num_bytes >= search_end) { |
4406 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 4407 | goto loop; |
6226cb0a | 4408 | } |
25179201 | 4409 | |
2552d17e JB |
4410 | /* move on to the next group */ |
4411 | if (search_start + num_bytes > | |
6226cb0a JB |
4412 | block_group->key.objectid + block_group->key.offset) { |
4413 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 4414 | goto loop; |
6226cb0a | 4415 | } |
f5a31e16 | 4416 | |
2552d17e JB |
4417 | if (exclude_nr > 0 && |
4418 | (search_start + num_bytes > exclude_start && | |
4419 | search_start < exclude_start + exclude_nr)) { | |
4420 | search_start = exclude_start + exclude_nr; | |
4421 | ||
6226cb0a | 4422 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e JB |
4423 | /* |
4424 | * if search_start is still in this block group | |
4425 | * then we just re-search this block group | |
f5a31e16 | 4426 | */ |
2552d17e JB |
4427 | if (search_start >= block_group->key.objectid && |
4428 | search_start < (block_group->key.objectid + | |
6226cb0a | 4429 | block_group->key.offset)) |
2552d17e | 4430 | goto have_block_group; |
2552d17e | 4431 | goto loop; |
0f9dd46c | 4432 | } |
0b86a832 | 4433 | |
2552d17e JB |
4434 | ins->objectid = search_start; |
4435 | ins->offset = num_bytes; | |
d2fb3437 | 4436 | |
6226cb0a JB |
4437 | if (offset < search_start) |
4438 | btrfs_add_free_space(block_group, offset, | |
4439 | search_start - offset); | |
4440 | BUG_ON(offset > search_start); | |
4441 | ||
11833d66 YZ |
4442 | update_reserved_extents(block_group, num_bytes, 1); |
4443 | ||
2552d17e | 4444 | /* we are all good, lets return */ |
2552d17e JB |
4445 | break; |
4446 | loop: | |
0a24325e | 4447 | failed_cluster_refill = false; |
1cdda9b8 | 4448 | failed_alloc = false; |
fa9c0d79 | 4449 | btrfs_put_block_group(block_group); |
2552d17e JB |
4450 | } |
4451 | up_read(&space_info->groups_sem); | |
4452 | ||
ccf0e725 JB |
4453 | /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for |
4454 | * for them to make caching progress. Also | |
4455 | * determine the best possible bg to cache | |
4456 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
4457 | * caching kthreads as we move along | |
817d52f8 JB |
4458 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
4459 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
4460 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
4461 | * again | |
fa9c0d79 | 4462 | */ |
817d52f8 JB |
4463 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE && |
4464 | (found_uncached_bg || empty_size || empty_cluster || | |
4465 | allowed_chunk_alloc)) { | |
ccf0e725 | 4466 | if (loop == LOOP_FIND_IDEAL && found_uncached_bg) { |
817d52f8 | 4467 | found_uncached_bg = false; |
ccf0e725 JB |
4468 | loop++; |
4469 | if (!ideal_cache_percent && | |
4470 | atomic_read(&space_info->caching_threads)) | |
817d52f8 | 4471 | goto search; |
ccf0e725 JB |
4472 | |
4473 | /* | |
4474 | * 1 of the following 2 things have happened so far | |
4475 | * | |
4476 | * 1) We found an ideal block group for caching that | |
4477 | * is mostly full and will cache quickly, so we might | |
4478 | * as well wait for it. | |
4479 | * | |
4480 | * 2) We searched for cached only and we didn't find | |
4481 | * anything, and we didn't start any caching kthreads | |
4482 | * either, so chances are we will loop through and | |
4483 | * start a couple caching kthreads, and then come back | |
4484 | * around and just wait for them. This will be slower | |
4485 | * because we will have 2 caching kthreads reading at | |
4486 | * the same time when we could have just started one | |
4487 | * and waited for it to get far enough to give us an | |
4488 | * allocation, so go ahead and go to the wait caching | |
4489 | * loop. | |
4490 | */ | |
4491 | loop = LOOP_CACHING_WAIT; | |
4492 | search_start = ideal_cache_offset; | |
4493 | ideal_cache_percent = 0; | |
4494 | goto ideal_cache; | |
4495 | } else if (loop == LOOP_FIND_IDEAL) { | |
4496 | /* | |
4497 | * Didn't find a uncached bg, wait on anything we find | |
4498 | * next. | |
4499 | */ | |
4500 | loop = LOOP_CACHING_WAIT; | |
4501 | goto search; | |
4502 | } | |
4503 | ||
4504 | if (loop < LOOP_CACHING_WAIT) { | |
4505 | loop++; | |
4506 | goto search; | |
817d52f8 JB |
4507 | } |
4508 | ||
4509 | if (loop == LOOP_ALLOC_CHUNK) { | |
fa9c0d79 CM |
4510 | empty_size = 0; |
4511 | empty_cluster = 0; | |
4512 | } | |
2552d17e JB |
4513 | |
4514 | if (allowed_chunk_alloc) { | |
4515 | ret = do_chunk_alloc(trans, root, num_bytes + | |
4516 | 2 * 1024 * 1024, data, 1); | |
2552d17e | 4517 | allowed_chunk_alloc = 0; |
ccf0e725 JB |
4518 | done_chunk_alloc = 1; |
4519 | } else if (!done_chunk_alloc) { | |
2552d17e JB |
4520 | space_info->force_alloc = 1; |
4521 | } | |
4522 | ||
817d52f8 | 4523 | if (loop < LOOP_NO_EMPTY_SIZE) { |
fa9c0d79 | 4524 | loop++; |
2552d17e | 4525 | goto search; |
fa9c0d79 | 4526 | } |
2552d17e JB |
4527 | ret = -ENOSPC; |
4528 | } else if (!ins->objectid) { | |
4529 | ret = -ENOSPC; | |
f2654de4 | 4530 | } |
0b86a832 | 4531 | |
80eb234a JB |
4532 | /* we found what we needed */ |
4533 | if (ins->objectid) { | |
4534 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) | |
d2fb3437 | 4535 | trans->block_group = block_group->key.objectid; |
0f9dd46c | 4536 | |
fa9c0d79 | 4537 | btrfs_put_block_group(block_group); |
80eb234a | 4538 | ret = 0; |
be744175 | 4539 | } |
be744175 | 4540 | |
0f70abe2 | 4541 | return ret; |
fec577fb | 4542 | } |
ec44a35c | 4543 | |
9ed74f2d JB |
4544 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
4545 | int dump_block_groups) | |
0f9dd46c JB |
4546 | { |
4547 | struct btrfs_block_group_cache *cache; | |
0f9dd46c | 4548 | |
9ed74f2d | 4549 | spin_lock(&info->lock); |
d397712b CM |
4550 | printk(KERN_INFO "space_info has %llu free, is %sfull\n", |
4551 | (unsigned long long)(info->total_bytes - info->bytes_used - | |
9ed74f2d JB |
4552 | info->bytes_pinned - info->bytes_reserved - |
4553 | info->bytes_super), | |
d397712b | 4554 | (info->full) ? "" : "not "); |
6a63209f | 4555 | printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu," |
9ed74f2d JB |
4556 | " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu" |
4557 | "\n", | |
21380931 JB |
4558 | (unsigned long long)info->total_bytes, |
4559 | (unsigned long long)info->bytes_pinned, | |
4560 | (unsigned long long)info->bytes_delalloc, | |
4561 | (unsigned long long)info->bytes_may_use, | |
9ed74f2d JB |
4562 | (unsigned long long)info->bytes_used, |
4563 | (unsigned long long)info->bytes_root, | |
4564 | (unsigned long long)info->bytes_super, | |
4565 | (unsigned long long)info->bytes_reserved); | |
4566 | spin_unlock(&info->lock); | |
4567 | ||
4568 | if (!dump_block_groups) | |
4569 | return; | |
0f9dd46c | 4570 | |
80eb234a | 4571 | down_read(&info->groups_sem); |
c6e30871 | 4572 | list_for_each_entry(cache, &info->block_groups, list) { |
0f9dd46c | 4573 | spin_lock(&cache->lock); |
d397712b CM |
4574 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
4575 | "%llu pinned %llu reserved\n", | |
4576 | (unsigned long long)cache->key.objectid, | |
4577 | (unsigned long long)cache->key.offset, | |
4578 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
4579 | (unsigned long long)cache->pinned, | |
4580 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
4581 | btrfs_dump_free_space(cache, bytes); |
4582 | spin_unlock(&cache->lock); | |
4583 | } | |
80eb234a | 4584 | up_read(&info->groups_sem); |
0f9dd46c | 4585 | } |
e8569813 | 4586 | |
11833d66 YZ |
4587 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
4588 | struct btrfs_root *root, | |
4589 | u64 num_bytes, u64 min_alloc_size, | |
4590 | u64 empty_size, u64 hint_byte, | |
4591 | u64 search_end, struct btrfs_key *ins, | |
4592 | u64 data) | |
fec577fb CM |
4593 | { |
4594 | int ret; | |
fbdc762b | 4595 | u64 search_start = 0; |
1261ec42 | 4596 | struct btrfs_fs_info *info = root->fs_info; |
925baedd | 4597 | |
6a63209f | 4598 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 4599 | again: |
0ef3e66b CM |
4600 | /* |
4601 | * the only place that sets empty_size is btrfs_realloc_node, which | |
4602 | * is not called recursively on allocations | |
4603 | */ | |
4604 | if (empty_size || root->ref_cows) { | |
593060d7 | 4605 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 4606 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b CM |
4607 | 2 * 1024 * 1024, |
4608 | BTRFS_BLOCK_GROUP_METADATA | | |
4609 | (info->metadata_alloc_profile & | |
4610 | info->avail_metadata_alloc_bits), 0); | |
6324fbf3 CM |
4611 | } |
4612 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
0ef3e66b | 4613 | num_bytes + 2 * 1024 * 1024, data, 0); |
6324fbf3 | 4614 | } |
0b86a832 | 4615 | |
db94535d CM |
4616 | WARN_ON(num_bytes < root->sectorsize); |
4617 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
4618 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
4619 | trans->alloc_exclude_start, |
4620 | trans->alloc_exclude_nr, data); | |
3b951516 | 4621 | |
98d20f67 CM |
4622 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
4623 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 4624 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 4625 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b CM |
4626 | do_chunk_alloc(trans, root->fs_info->extent_root, |
4627 | num_bytes, data, 1); | |
98d20f67 CM |
4628 | goto again; |
4629 | } | |
817d52f8 | 4630 | if (ret == -ENOSPC) { |
0f9dd46c JB |
4631 | struct btrfs_space_info *sinfo; |
4632 | ||
4633 | sinfo = __find_space_info(root->fs_info, data); | |
d397712b CM |
4634 | printk(KERN_ERR "btrfs allocation failed flags %llu, " |
4635 | "wanted %llu\n", (unsigned long long)data, | |
4636 | (unsigned long long)num_bytes); | |
9ed74f2d | 4637 | dump_space_info(sinfo, num_bytes, 1); |
925baedd | 4638 | } |
0f9dd46c JB |
4639 | |
4640 | return ret; | |
e6dcd2dc CM |
4641 | } |
4642 | ||
65b51a00 CM |
4643 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
4644 | { | |
0f9dd46c | 4645 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 4646 | int ret = 0; |
0f9dd46c | 4647 | |
0f9dd46c JB |
4648 | cache = btrfs_lookup_block_group(root->fs_info, start); |
4649 | if (!cache) { | |
d397712b CM |
4650 | printk(KERN_ERR "Unable to find block group for %llu\n", |
4651 | (unsigned long long)start); | |
0f9dd46c JB |
4652 | return -ENOSPC; |
4653 | } | |
1f3c79a2 LH |
4654 | |
4655 | ret = btrfs_discard_extent(root, start, len); | |
4656 | ||
0f9dd46c | 4657 | btrfs_add_free_space(cache, start, len); |
11833d66 | 4658 | update_reserved_extents(cache, len, 0); |
fa9c0d79 | 4659 | btrfs_put_block_group(cache); |
817d52f8 | 4660 | |
e6dcd2dc CM |
4661 | return ret; |
4662 | } | |
4663 | ||
5d4f98a2 YZ |
4664 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
4665 | struct btrfs_root *root, | |
4666 | u64 parent, u64 root_objectid, | |
4667 | u64 flags, u64 owner, u64 offset, | |
4668 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
4669 | { |
4670 | int ret; | |
5d4f98a2 | 4671 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 4672 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4673 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4674 | struct btrfs_path *path; |
5d4f98a2 YZ |
4675 | struct extent_buffer *leaf; |
4676 | int type; | |
4677 | u32 size; | |
26b8003f | 4678 | |
5d4f98a2 YZ |
4679 | if (parent > 0) |
4680 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4681 | else | |
4682 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4683 | |
5d4f98a2 | 4684 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4685 | |
4686 | path = btrfs_alloc_path(); | |
4687 | BUG_ON(!path); | |
47e4bb98 | 4688 | |
b9473439 | 4689 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4690 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4691 | ins, size); | |
ccd467d6 | 4692 | BUG_ON(ret); |
0f9dd46c | 4693 | |
5d4f98a2 YZ |
4694 | leaf = path->nodes[0]; |
4695 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4696 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4697 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4698 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4699 | btrfs_set_extent_flags(leaf, extent_item, | |
4700 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4701 | ||
4702 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4703 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4704 | if (parent > 0) { | |
4705 | struct btrfs_shared_data_ref *ref; | |
4706 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4707 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4708 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4709 | } else { | |
4710 | struct btrfs_extent_data_ref *ref; | |
4711 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4712 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4713 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4714 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4715 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4716 | } | |
47e4bb98 CM |
4717 | |
4718 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4719 | btrfs_free_path(path); |
f510cfec | 4720 | |
5d4f98a2 YZ |
4721 | ret = update_block_group(trans, root, ins->objectid, ins->offset, |
4722 | 1, 0); | |
f5947066 | 4723 | if (ret) { |
d397712b CM |
4724 | printk(KERN_ERR "btrfs update block group failed for %llu " |
4725 | "%llu\n", (unsigned long long)ins->objectid, | |
4726 | (unsigned long long)ins->offset); | |
f5947066 CM |
4727 | BUG(); |
4728 | } | |
e6dcd2dc CM |
4729 | return ret; |
4730 | } | |
4731 | ||
5d4f98a2 YZ |
4732 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4733 | struct btrfs_root *root, | |
4734 | u64 parent, u64 root_objectid, | |
4735 | u64 flags, struct btrfs_disk_key *key, | |
4736 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
4737 | { |
4738 | int ret; | |
5d4f98a2 YZ |
4739 | struct btrfs_fs_info *fs_info = root->fs_info; |
4740 | struct btrfs_extent_item *extent_item; | |
4741 | struct btrfs_tree_block_info *block_info; | |
4742 | struct btrfs_extent_inline_ref *iref; | |
4743 | struct btrfs_path *path; | |
4744 | struct extent_buffer *leaf; | |
4745 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 4746 | |
5d4f98a2 YZ |
4747 | path = btrfs_alloc_path(); |
4748 | BUG_ON(!path); | |
56bec294 | 4749 | |
5d4f98a2 YZ |
4750 | path->leave_spinning = 1; |
4751 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4752 | ins, size); | |
56bec294 | 4753 | BUG_ON(ret); |
5d4f98a2 YZ |
4754 | |
4755 | leaf = path->nodes[0]; | |
4756 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4757 | struct btrfs_extent_item); | |
4758 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4759 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4760 | btrfs_set_extent_flags(leaf, extent_item, | |
4761 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
4762 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
4763 | ||
4764 | btrfs_set_tree_block_key(leaf, block_info, key); | |
4765 | btrfs_set_tree_block_level(leaf, block_info, level); | |
4766 | ||
4767 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
4768 | if (parent > 0) { | |
4769 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
4770 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4771 | BTRFS_SHARED_BLOCK_REF_KEY); | |
4772 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4773 | } else { | |
4774 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4775 | BTRFS_TREE_BLOCK_REF_KEY); | |
4776 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
4777 | } | |
4778 | ||
4779 | btrfs_mark_buffer_dirty(leaf); | |
4780 | btrfs_free_path(path); | |
4781 | ||
4782 | ret = update_block_group(trans, root, ins->objectid, ins->offset, | |
4783 | 1, 0); | |
4784 | if (ret) { | |
4785 | printk(KERN_ERR "btrfs update block group failed for %llu " | |
4786 | "%llu\n", (unsigned long long)ins->objectid, | |
4787 | (unsigned long long)ins->offset); | |
4788 | BUG(); | |
4789 | } | |
4790 | return ret; | |
4791 | } | |
4792 | ||
4793 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
4794 | struct btrfs_root *root, | |
4795 | u64 root_objectid, u64 owner, | |
4796 | u64 offset, struct btrfs_key *ins) | |
4797 | { | |
4798 | int ret; | |
4799 | ||
4800 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
4801 | ||
4802 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset, | |
4803 | 0, root_objectid, owner, offset, | |
4804 | BTRFS_ADD_DELAYED_EXTENT, NULL); | |
e6dcd2dc CM |
4805 | return ret; |
4806 | } | |
e02119d5 CM |
4807 | |
4808 | /* | |
4809 | * this is used by the tree logging recovery code. It records that | |
4810 | * an extent has been allocated and makes sure to clear the free | |
4811 | * space cache bits as well | |
4812 | */ | |
5d4f98a2 YZ |
4813 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
4814 | struct btrfs_root *root, | |
4815 | u64 root_objectid, u64 owner, u64 offset, | |
4816 | struct btrfs_key *ins) | |
e02119d5 CM |
4817 | { |
4818 | int ret; | |
4819 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
4820 | struct btrfs_caching_control *caching_ctl; |
4821 | u64 start = ins->objectid; | |
4822 | u64 num_bytes = ins->offset; | |
e02119d5 | 4823 | |
e02119d5 | 4824 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
817d52f8 | 4825 | cache_block_group(block_group); |
11833d66 | 4826 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 4827 | |
11833d66 YZ |
4828 | if (!caching_ctl) { |
4829 | BUG_ON(!block_group_cache_done(block_group)); | |
4830 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
4831 | BUG_ON(ret); | |
4832 | } else { | |
4833 | mutex_lock(&caching_ctl->mutex); | |
4834 | ||
4835 | if (start >= caching_ctl->progress) { | |
4836 | ret = add_excluded_extent(root, start, num_bytes); | |
4837 | BUG_ON(ret); | |
4838 | } else if (start + num_bytes <= caching_ctl->progress) { | |
4839 | ret = btrfs_remove_free_space(block_group, | |
4840 | start, num_bytes); | |
4841 | BUG_ON(ret); | |
4842 | } else { | |
4843 | num_bytes = caching_ctl->progress - start; | |
4844 | ret = btrfs_remove_free_space(block_group, | |
4845 | start, num_bytes); | |
4846 | BUG_ON(ret); | |
4847 | ||
4848 | start = caching_ctl->progress; | |
4849 | num_bytes = ins->objectid + ins->offset - | |
4850 | caching_ctl->progress; | |
4851 | ret = add_excluded_extent(root, start, num_bytes); | |
4852 | BUG_ON(ret); | |
4853 | } | |
4854 | ||
4855 | mutex_unlock(&caching_ctl->mutex); | |
4856 | put_caching_control(caching_ctl); | |
4857 | } | |
4858 | ||
4859 | update_reserved_extents(block_group, ins->offset, 1); | |
fa9c0d79 | 4860 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
4861 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
4862 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
4863 | return ret; |
4864 | } | |
4865 | ||
e6dcd2dc CM |
4866 | /* |
4867 | * finds a free extent and does all the dirty work required for allocation | |
4868 | * returns the key for the extent through ins, and a tree buffer for | |
4869 | * the first block of the extent through buf. | |
4870 | * | |
4871 | * returns 0 if everything worked, non-zero otherwise. | |
4872 | */ | |
5d4f98a2 YZ |
4873 | static int alloc_tree_block(struct btrfs_trans_handle *trans, |
4874 | struct btrfs_root *root, | |
4875 | u64 num_bytes, u64 parent, u64 root_objectid, | |
4876 | struct btrfs_disk_key *key, int level, | |
4877 | u64 empty_size, u64 hint_byte, u64 search_end, | |
4878 | struct btrfs_key *ins) | |
e6dcd2dc CM |
4879 | { |
4880 | int ret; | |
5d4f98a2 YZ |
4881 | u64 flags = 0; |
4882 | ||
11833d66 YZ |
4883 | ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes, |
4884 | empty_size, hint_byte, search_end, | |
4885 | ins, 0); | |
817d52f8 JB |
4886 | if (ret) |
4887 | return ret; | |
5d4f98a2 YZ |
4888 | |
4889 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4890 | if (parent == 0) | |
4891 | parent = ins->objectid; | |
4892 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4893 | } else | |
4894 | BUG_ON(parent > 0); | |
4895 | ||
d00aff00 | 4896 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { |
5d4f98a2 YZ |
4897 | struct btrfs_delayed_extent_op *extent_op; |
4898 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
4899 | BUG_ON(!extent_op); | |
4900 | if (key) | |
4901 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4902 | else | |
4903 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4904 | extent_op->flags_to_set = flags; | |
4905 | extent_op->update_key = 1; | |
4906 | extent_op->update_flags = 1; | |
4907 | extent_op->is_data = 0; | |
4908 | ||
4909 | ret = btrfs_add_delayed_tree_ref(trans, ins->objectid, | |
4910 | ins->offset, parent, root_objectid, | |
4911 | level, BTRFS_ADD_DELAYED_EXTENT, | |
4912 | extent_op); | |
d00aff00 | 4913 | BUG_ON(ret); |
d00aff00 | 4914 | } |
86b9f2ec YZ |
4915 | |
4916 | if (root_objectid == root->root_key.objectid) { | |
4917 | u64 used; | |
4918 | spin_lock(&root->node_lock); | |
4919 | used = btrfs_root_used(&root->root_item) + num_bytes; | |
4920 | btrfs_set_root_used(&root->root_item, used); | |
4921 | spin_unlock(&root->node_lock); | |
4922 | } | |
925baedd | 4923 | return ret; |
fec577fb | 4924 | } |
65b51a00 CM |
4925 | |
4926 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, | |
4927 | struct btrfs_root *root, | |
4008c04a CM |
4928 | u64 bytenr, u32 blocksize, |
4929 | int level) | |
65b51a00 CM |
4930 | { |
4931 | struct extent_buffer *buf; | |
4932 | ||
4933 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
4934 | if (!buf) | |
4935 | return ERR_PTR(-ENOMEM); | |
4936 | btrfs_set_header_generation(buf, trans->transid); | |
4008c04a | 4937 | btrfs_set_buffer_lockdep_class(buf, level); |
65b51a00 CM |
4938 | btrfs_tree_lock(buf); |
4939 | clean_tree_block(trans, root, buf); | |
b4ce94de CM |
4940 | |
4941 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 4942 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 4943 | |
d0c803c4 | 4944 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
4945 | /* |
4946 | * we allow two log transactions at a time, use different | |
4947 | * EXENT bit to differentiate dirty pages. | |
4948 | */ | |
4949 | if (root->log_transid % 2 == 0) | |
4950 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
4951 | buf->start + buf->len - 1, GFP_NOFS); | |
4952 | else | |
4953 | set_extent_new(&root->dirty_log_pages, buf->start, | |
4954 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
4955 | } else { |
4956 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 4957 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4958 | } |
65b51a00 | 4959 | trans->blocks_used++; |
b4ce94de | 4960 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4961 | return buf; |
4962 | } | |
4963 | ||
fec577fb CM |
4964 | /* |
4965 | * helper function to allocate a block for a given tree | |
4966 | * returns the tree buffer or NULL. | |
4967 | */ | |
5f39d397 | 4968 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4969 | struct btrfs_root *root, u32 blocksize, |
4970 | u64 parent, u64 root_objectid, | |
4971 | struct btrfs_disk_key *key, int level, | |
4972 | u64 hint, u64 empty_size) | |
fec577fb | 4973 | { |
e2fa7227 | 4974 | struct btrfs_key ins; |
fec577fb | 4975 | int ret; |
5f39d397 | 4976 | struct extent_buffer *buf; |
fec577fb | 4977 | |
5d4f98a2 YZ |
4978 | ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid, |
4979 | key, level, empty_size, hint, (u64)-1, &ins); | |
fec577fb | 4980 | if (ret) { |
54aa1f4d CM |
4981 | BUG_ON(ret > 0); |
4982 | return ERR_PTR(ret); | |
fec577fb | 4983 | } |
55c69072 | 4984 | |
4008c04a CM |
4985 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
4986 | blocksize, level); | |
fec577fb CM |
4987 | return buf; |
4988 | } | |
a28ec197 | 4989 | |
2c47e605 YZ |
4990 | struct walk_control { |
4991 | u64 refs[BTRFS_MAX_LEVEL]; | |
4992 | u64 flags[BTRFS_MAX_LEVEL]; | |
4993 | struct btrfs_key update_progress; | |
4994 | int stage; | |
4995 | int level; | |
4996 | int shared_level; | |
4997 | int update_ref; | |
4998 | int keep_locks; | |
1c4850e2 YZ |
4999 | int reada_slot; |
5000 | int reada_count; | |
2c47e605 YZ |
5001 | }; |
5002 | ||
5003 | #define DROP_REFERENCE 1 | |
5004 | #define UPDATE_BACKREF 2 | |
5005 | ||
1c4850e2 YZ |
5006 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
5007 | struct btrfs_root *root, | |
5008 | struct walk_control *wc, | |
5009 | struct btrfs_path *path) | |
6407bf6d | 5010 | { |
1c4850e2 YZ |
5011 | u64 bytenr; |
5012 | u64 generation; | |
5013 | u64 refs; | |
94fcca9f | 5014 | u64 flags; |
1c4850e2 | 5015 | u64 last = 0; |
5d4f98a2 | 5016 | u32 nritems; |
1c4850e2 YZ |
5017 | u32 blocksize; |
5018 | struct btrfs_key key; | |
5019 | struct extent_buffer *eb; | |
6407bf6d | 5020 | int ret; |
1c4850e2 YZ |
5021 | int slot; |
5022 | int nread = 0; | |
6407bf6d | 5023 | |
1c4850e2 YZ |
5024 | if (path->slots[wc->level] < wc->reada_slot) { |
5025 | wc->reada_count = wc->reada_count * 2 / 3; | |
5026 | wc->reada_count = max(wc->reada_count, 2); | |
5027 | } else { | |
5028 | wc->reada_count = wc->reada_count * 3 / 2; | |
5029 | wc->reada_count = min_t(int, wc->reada_count, | |
5030 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
5031 | } | |
7bb86316 | 5032 | |
1c4850e2 YZ |
5033 | eb = path->nodes[wc->level]; |
5034 | nritems = btrfs_header_nritems(eb); | |
5035 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 5036 | |
1c4850e2 YZ |
5037 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
5038 | if (nread >= wc->reada_count) | |
5039 | break; | |
bd56b302 | 5040 | |
2dd3e67b | 5041 | cond_resched(); |
1c4850e2 YZ |
5042 | bytenr = btrfs_node_blockptr(eb, slot); |
5043 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 5044 | |
1c4850e2 YZ |
5045 | if (slot == path->slots[wc->level]) |
5046 | goto reada; | |
5d4f98a2 | 5047 | |
1c4850e2 YZ |
5048 | if (wc->stage == UPDATE_BACKREF && |
5049 | generation <= root->root_key.offset) | |
bd56b302 CM |
5050 | continue; |
5051 | ||
94fcca9f YZ |
5052 | /* We don't lock the tree block, it's OK to be racy here */ |
5053 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
5054 | &refs, &flags); | |
5055 | BUG_ON(ret); | |
5056 | BUG_ON(refs == 0); | |
5057 | ||
1c4850e2 | 5058 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
5059 | if (refs == 1) |
5060 | goto reada; | |
bd56b302 | 5061 | |
94fcca9f YZ |
5062 | if (wc->level == 1 && |
5063 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5064 | continue; | |
1c4850e2 YZ |
5065 | if (!wc->update_ref || |
5066 | generation <= root->root_key.offset) | |
5067 | continue; | |
5068 | btrfs_node_key_to_cpu(eb, &key, slot); | |
5069 | ret = btrfs_comp_cpu_keys(&key, | |
5070 | &wc->update_progress); | |
5071 | if (ret < 0) | |
5072 | continue; | |
94fcca9f YZ |
5073 | } else { |
5074 | if (wc->level == 1 && | |
5075 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5076 | continue; | |
6407bf6d | 5077 | } |
1c4850e2 YZ |
5078 | reada: |
5079 | ret = readahead_tree_block(root, bytenr, blocksize, | |
5080 | generation); | |
5081 | if (ret) | |
bd56b302 | 5082 | break; |
1c4850e2 YZ |
5083 | last = bytenr + blocksize; |
5084 | nread++; | |
20524f02 | 5085 | } |
1c4850e2 | 5086 | wc->reada_slot = slot; |
20524f02 | 5087 | } |
2c47e605 | 5088 | |
f82d02d9 | 5089 | /* |
2c47e605 YZ |
5090 | * hepler to process tree block while walking down the tree. |
5091 | * | |
2c47e605 YZ |
5092 | * when wc->stage == UPDATE_BACKREF, this function updates |
5093 | * back refs for pointers in the block. | |
5094 | * | |
5095 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 5096 | */ |
2c47e605 | 5097 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 5098 | struct btrfs_root *root, |
2c47e605 | 5099 | struct btrfs_path *path, |
94fcca9f | 5100 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 5101 | { |
2c47e605 YZ |
5102 | int level = wc->level; |
5103 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 5104 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
5105 | int ret; |
5106 | ||
2c47e605 YZ |
5107 | if (wc->stage == UPDATE_BACKREF && |
5108 | btrfs_header_owner(eb) != root->root_key.objectid) | |
5109 | return 1; | |
f82d02d9 | 5110 | |
2c47e605 YZ |
5111 | /* |
5112 | * when reference count of tree block is 1, it won't increase | |
5113 | * again. once full backref flag is set, we never clear it. | |
5114 | */ | |
94fcca9f YZ |
5115 | if (lookup_info && |
5116 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
5117 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
5118 | BUG_ON(!path->locks[level]); |
5119 | ret = btrfs_lookup_extent_info(trans, root, | |
5120 | eb->start, eb->len, | |
5121 | &wc->refs[level], | |
5122 | &wc->flags[level]); | |
5123 | BUG_ON(ret); | |
5124 | BUG_ON(wc->refs[level] == 0); | |
5125 | } | |
5d4f98a2 | 5126 | |
2c47e605 YZ |
5127 | if (wc->stage == DROP_REFERENCE) { |
5128 | if (wc->refs[level] > 1) | |
5129 | return 1; | |
f82d02d9 | 5130 | |
2c47e605 YZ |
5131 | if (path->locks[level] && !wc->keep_locks) { |
5132 | btrfs_tree_unlock(eb); | |
5133 | path->locks[level] = 0; | |
5134 | } | |
5135 | return 0; | |
5136 | } | |
f82d02d9 | 5137 | |
2c47e605 YZ |
5138 | /* wc->stage == UPDATE_BACKREF */ |
5139 | if (!(wc->flags[level] & flag)) { | |
5140 | BUG_ON(!path->locks[level]); | |
5141 | ret = btrfs_inc_ref(trans, root, eb, 1); | |
f82d02d9 | 5142 | BUG_ON(ret); |
2c47e605 YZ |
5143 | ret = btrfs_dec_ref(trans, root, eb, 0); |
5144 | BUG_ON(ret); | |
5145 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, | |
5146 | eb->len, flag, 0); | |
5147 | BUG_ON(ret); | |
5148 | wc->flags[level] |= flag; | |
5149 | } | |
5150 | ||
5151 | /* | |
5152 | * the block is shared by multiple trees, so it's not good to | |
5153 | * keep the tree lock | |
5154 | */ | |
5155 | if (path->locks[level] && level > 0) { | |
5156 | btrfs_tree_unlock(eb); | |
5157 | path->locks[level] = 0; | |
5158 | } | |
5159 | return 0; | |
5160 | } | |
5161 | ||
1c4850e2 YZ |
5162 | /* |
5163 | * hepler to process tree block pointer. | |
5164 | * | |
5165 | * when wc->stage == DROP_REFERENCE, this function checks | |
5166 | * reference count of the block pointed to. if the block | |
5167 | * is shared and we need update back refs for the subtree | |
5168 | * rooted at the block, this function changes wc->stage to | |
5169 | * UPDATE_BACKREF. if the block is shared and there is no | |
5170 | * need to update back, this function drops the reference | |
5171 | * to the block. | |
5172 | * | |
5173 | * NOTE: return value 1 means we should stop walking down. | |
5174 | */ | |
5175 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
5176 | struct btrfs_root *root, | |
5177 | struct btrfs_path *path, | |
94fcca9f | 5178 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
5179 | { |
5180 | u64 bytenr; | |
5181 | u64 generation; | |
5182 | u64 parent; | |
5183 | u32 blocksize; | |
5184 | struct btrfs_key key; | |
5185 | struct extent_buffer *next; | |
5186 | int level = wc->level; | |
5187 | int reada = 0; | |
5188 | int ret = 0; | |
5189 | ||
5190 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
5191 | path->slots[level]); | |
5192 | /* | |
5193 | * if the lower level block was created before the snapshot | |
5194 | * was created, we know there is no need to update back refs | |
5195 | * for the subtree | |
5196 | */ | |
5197 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5198 | generation <= root->root_key.offset) { |
5199 | *lookup_info = 1; | |
1c4850e2 | 5200 | return 1; |
94fcca9f | 5201 | } |
1c4850e2 YZ |
5202 | |
5203 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
5204 | blocksize = btrfs_level_size(root, level - 1); | |
5205 | ||
5206 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
5207 | if (!next) { | |
5208 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
5209 | reada = 1; | |
5210 | } | |
5211 | btrfs_tree_lock(next); | |
5212 | btrfs_set_lock_blocking(next); | |
5213 | ||
94fcca9f YZ |
5214 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
5215 | &wc->refs[level - 1], | |
5216 | &wc->flags[level - 1]); | |
5217 | BUG_ON(ret); | |
5218 | BUG_ON(wc->refs[level - 1] == 0); | |
5219 | *lookup_info = 0; | |
1c4850e2 | 5220 | |
94fcca9f | 5221 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 5222 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
5223 | if (level == 1 && |
5224 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5225 | goto skip; | |
5226 | ||
1c4850e2 YZ |
5227 | if (!wc->update_ref || |
5228 | generation <= root->root_key.offset) | |
5229 | goto skip; | |
5230 | ||
5231 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
5232 | path->slots[level]); | |
5233 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
5234 | if (ret < 0) | |
5235 | goto skip; | |
5236 | ||
5237 | wc->stage = UPDATE_BACKREF; | |
5238 | wc->shared_level = level - 1; | |
5239 | } | |
94fcca9f YZ |
5240 | } else { |
5241 | if (level == 1 && | |
5242 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5243 | goto skip; | |
1c4850e2 YZ |
5244 | } |
5245 | ||
5246 | if (!btrfs_buffer_uptodate(next, generation)) { | |
5247 | btrfs_tree_unlock(next); | |
5248 | free_extent_buffer(next); | |
5249 | next = NULL; | |
94fcca9f | 5250 | *lookup_info = 1; |
1c4850e2 YZ |
5251 | } |
5252 | ||
5253 | if (!next) { | |
5254 | if (reada && level == 1) | |
5255 | reada_walk_down(trans, root, wc, path); | |
5256 | next = read_tree_block(root, bytenr, blocksize, generation); | |
5257 | btrfs_tree_lock(next); | |
5258 | btrfs_set_lock_blocking(next); | |
5259 | } | |
5260 | ||
5261 | level--; | |
5262 | BUG_ON(level != btrfs_header_level(next)); | |
5263 | path->nodes[level] = next; | |
5264 | path->slots[level] = 0; | |
5265 | path->locks[level] = 1; | |
5266 | wc->level = level; | |
5267 | if (wc->level == 1) | |
5268 | wc->reada_slot = 0; | |
5269 | return 0; | |
5270 | skip: | |
5271 | wc->refs[level - 1] = 0; | |
5272 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
5273 | if (wc->stage == DROP_REFERENCE) { |
5274 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
5275 | parent = path->nodes[level]->start; | |
5276 | } else { | |
5277 | BUG_ON(root->root_key.objectid != | |
5278 | btrfs_header_owner(path->nodes[level])); | |
5279 | parent = 0; | |
5280 | } | |
1c4850e2 | 5281 | |
94fcca9f YZ |
5282 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
5283 | root->root_key.objectid, level - 1, 0); | |
5284 | BUG_ON(ret); | |
1c4850e2 | 5285 | } |
1c4850e2 YZ |
5286 | btrfs_tree_unlock(next); |
5287 | free_extent_buffer(next); | |
94fcca9f | 5288 | *lookup_info = 1; |
1c4850e2 YZ |
5289 | return 1; |
5290 | } | |
5291 | ||
2c47e605 YZ |
5292 | /* |
5293 | * hepler to process tree block while walking up the tree. | |
5294 | * | |
5295 | * when wc->stage == DROP_REFERENCE, this function drops | |
5296 | * reference count on the block. | |
5297 | * | |
5298 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5299 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5300 | * to UPDATE_BACKREF previously while processing the block. | |
5301 | * | |
5302 | * NOTE: return value 1 means we should stop walking up. | |
5303 | */ | |
5304 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5305 | struct btrfs_root *root, | |
5306 | struct btrfs_path *path, | |
5307 | struct walk_control *wc) | |
5308 | { | |
5309 | int ret = 0; | |
5310 | int level = wc->level; | |
5311 | struct extent_buffer *eb = path->nodes[level]; | |
5312 | u64 parent = 0; | |
5313 | ||
5314 | if (wc->stage == UPDATE_BACKREF) { | |
5315 | BUG_ON(wc->shared_level < level); | |
5316 | if (level < wc->shared_level) | |
5317 | goto out; | |
5318 | ||
2c47e605 YZ |
5319 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5320 | if (ret > 0) | |
5321 | wc->update_ref = 0; | |
5322 | ||
5323 | wc->stage = DROP_REFERENCE; | |
5324 | wc->shared_level = -1; | |
5325 | path->slots[level] = 0; | |
5326 | ||
5327 | /* | |
5328 | * check reference count again if the block isn't locked. | |
5329 | * we should start walking down the tree again if reference | |
5330 | * count is one. | |
5331 | */ | |
5332 | if (!path->locks[level]) { | |
5333 | BUG_ON(level == 0); | |
5334 | btrfs_tree_lock(eb); | |
5335 | btrfs_set_lock_blocking(eb); | |
5336 | path->locks[level] = 1; | |
5337 | ||
5338 | ret = btrfs_lookup_extent_info(trans, root, | |
5339 | eb->start, eb->len, | |
5340 | &wc->refs[level], | |
5341 | &wc->flags[level]); | |
f82d02d9 | 5342 | BUG_ON(ret); |
2c47e605 YZ |
5343 | BUG_ON(wc->refs[level] == 0); |
5344 | if (wc->refs[level] == 1) { | |
5345 | btrfs_tree_unlock(eb); | |
5346 | path->locks[level] = 0; | |
5347 | return 1; | |
5348 | } | |
f82d02d9 | 5349 | } |
2c47e605 | 5350 | } |
f82d02d9 | 5351 | |
2c47e605 YZ |
5352 | /* wc->stage == DROP_REFERENCE */ |
5353 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5354 | |
2c47e605 YZ |
5355 | if (wc->refs[level] == 1) { |
5356 | if (level == 0) { | |
5357 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5358 | ret = btrfs_dec_ref(trans, root, eb, 1); | |
5359 | else | |
5360 | ret = btrfs_dec_ref(trans, root, eb, 0); | |
5361 | BUG_ON(ret); | |
5362 | } | |
5363 | /* make block locked assertion in clean_tree_block happy */ | |
5364 | if (!path->locks[level] && | |
5365 | btrfs_header_generation(eb) == trans->transid) { | |
5366 | btrfs_tree_lock(eb); | |
5367 | btrfs_set_lock_blocking(eb); | |
5368 | path->locks[level] = 1; | |
5369 | } | |
5370 | clean_tree_block(trans, root, eb); | |
5371 | } | |
5372 | ||
5373 | if (eb == root->node) { | |
5374 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5375 | parent = eb->start; | |
5376 | else | |
5377 | BUG_ON(root->root_key.objectid != | |
5378 | btrfs_header_owner(eb)); | |
5379 | } else { | |
5380 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5381 | parent = path->nodes[level + 1]->start; | |
5382 | else | |
5383 | BUG_ON(root->root_key.objectid != | |
5384 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 5385 | } |
f82d02d9 | 5386 | |
2c47e605 YZ |
5387 | ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent, |
5388 | root->root_key.objectid, level, 0); | |
f82d02d9 | 5389 | BUG_ON(ret); |
2c47e605 YZ |
5390 | out: |
5391 | wc->refs[level] = 0; | |
5392 | wc->flags[level] = 0; | |
5393 | return ret; | |
5394 | } | |
5395 | ||
5396 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5397 | struct btrfs_root *root, | |
5398 | struct btrfs_path *path, | |
5399 | struct walk_control *wc) | |
5400 | { | |
2c47e605 | 5401 | int level = wc->level; |
94fcca9f | 5402 | int lookup_info = 1; |
2c47e605 YZ |
5403 | int ret; |
5404 | ||
5405 | while (level >= 0) { | |
1c4850e2 YZ |
5406 | if (path->slots[level] >= |
5407 | btrfs_header_nritems(path->nodes[level])) | |
5408 | break; | |
f82d02d9 | 5409 | |
94fcca9f | 5410 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
5411 | if (ret > 0) |
5412 | break; | |
5413 | ||
5414 | if (level == 0) | |
5415 | break; | |
5416 | ||
94fcca9f | 5417 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5418 | if (ret > 0) { |
5419 | path->slots[level]++; | |
5420 | continue; | |
5421 | } | |
5422 | level = wc->level; | |
f82d02d9 | 5423 | } |
f82d02d9 YZ |
5424 | return 0; |
5425 | } | |
5426 | ||
d397712b | 5427 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5428 | struct btrfs_root *root, |
f82d02d9 | 5429 | struct btrfs_path *path, |
2c47e605 | 5430 | struct walk_control *wc, int max_level) |
20524f02 | 5431 | { |
2c47e605 | 5432 | int level = wc->level; |
20524f02 | 5433 | int ret; |
9f3a7427 | 5434 | |
2c47e605 YZ |
5435 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5436 | while (level < max_level && path->nodes[level]) { | |
5437 | wc->level = level; | |
5438 | if (path->slots[level] + 1 < | |
5439 | btrfs_header_nritems(path->nodes[level])) { | |
5440 | path->slots[level]++; | |
20524f02 CM |
5441 | return 0; |
5442 | } else { | |
2c47e605 YZ |
5443 | ret = walk_up_proc(trans, root, path, wc); |
5444 | if (ret > 0) | |
5445 | return 0; | |
bd56b302 | 5446 | |
2c47e605 YZ |
5447 | if (path->locks[level]) { |
5448 | btrfs_tree_unlock(path->nodes[level]); | |
5449 | path->locks[level] = 0; | |
f82d02d9 | 5450 | } |
2c47e605 YZ |
5451 | free_extent_buffer(path->nodes[level]); |
5452 | path->nodes[level] = NULL; | |
5453 | level++; | |
20524f02 CM |
5454 | } |
5455 | } | |
5456 | return 1; | |
5457 | } | |
5458 | ||
9aca1d51 | 5459 | /* |
2c47e605 YZ |
5460 | * drop a subvolume tree. |
5461 | * | |
5462 | * this function traverses the tree freeing any blocks that only | |
5463 | * referenced by the tree. | |
5464 | * | |
5465 | * when a shared tree block is found. this function decreases its | |
5466 | * reference count by one. if update_ref is true, this function | |
5467 | * also make sure backrefs for the shared block and all lower level | |
5468 | * blocks are properly updated. | |
9aca1d51 | 5469 | */ |
2c47e605 | 5470 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref) |
20524f02 | 5471 | { |
5caf2a00 | 5472 | struct btrfs_path *path; |
2c47e605 YZ |
5473 | struct btrfs_trans_handle *trans; |
5474 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 5475 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5476 | struct walk_control *wc; |
5477 | struct btrfs_key key; | |
5478 | int err = 0; | |
5479 | int ret; | |
5480 | int level; | |
20524f02 | 5481 | |
5caf2a00 CM |
5482 | path = btrfs_alloc_path(); |
5483 | BUG_ON(!path); | |
20524f02 | 5484 | |
2c47e605 YZ |
5485 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
5486 | BUG_ON(!wc); | |
5487 | ||
5488 | trans = btrfs_start_transaction(tree_root, 1); | |
5489 | ||
9f3a7427 | 5490 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5491 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
5492 | path->nodes[level] = btrfs_lock_root_node(root); |
5493 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 5494 | path->slots[level] = 0; |
5d4f98a2 | 5495 | path->locks[level] = 1; |
2c47e605 YZ |
5496 | memset(&wc->update_progress, 0, |
5497 | sizeof(wc->update_progress)); | |
9f3a7427 | 5498 | } else { |
9f3a7427 | 5499 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5500 | memcpy(&wc->update_progress, &key, |
5501 | sizeof(wc->update_progress)); | |
5502 | ||
6702ed49 | 5503 | level = root_item->drop_level; |
2c47e605 | 5504 | BUG_ON(level == 0); |
6702ed49 | 5505 | path->lowest_level = level; |
2c47e605 YZ |
5506 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5507 | path->lowest_level = 0; | |
5508 | if (ret < 0) { | |
5509 | err = ret; | |
9f3a7427 CM |
5510 | goto out; |
5511 | } | |
1c4850e2 | 5512 | WARN_ON(ret > 0); |
2c47e605 | 5513 | |
7d9eb12c CM |
5514 | /* |
5515 | * unlock our path, this is safe because only this | |
5516 | * function is allowed to delete this snapshot | |
5517 | */ | |
5d4f98a2 | 5518 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5519 | |
5520 | level = btrfs_header_level(root->node); | |
5521 | while (1) { | |
5522 | btrfs_tree_lock(path->nodes[level]); | |
5523 | btrfs_set_lock_blocking(path->nodes[level]); | |
5524 | ||
5525 | ret = btrfs_lookup_extent_info(trans, root, | |
5526 | path->nodes[level]->start, | |
5527 | path->nodes[level]->len, | |
5528 | &wc->refs[level], | |
5529 | &wc->flags[level]); | |
5530 | BUG_ON(ret); | |
5531 | BUG_ON(wc->refs[level] == 0); | |
5532 | ||
5533 | if (level == root_item->drop_level) | |
5534 | break; | |
5535 | ||
5536 | btrfs_tree_unlock(path->nodes[level]); | |
5537 | WARN_ON(wc->refs[level] != 1); | |
5538 | level--; | |
5539 | } | |
9f3a7427 | 5540 | } |
2c47e605 YZ |
5541 | |
5542 | wc->level = level; | |
5543 | wc->shared_level = -1; | |
5544 | wc->stage = DROP_REFERENCE; | |
5545 | wc->update_ref = update_ref; | |
5546 | wc->keep_locks = 0; | |
1c4850e2 | 5547 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 5548 | |
d397712b | 5549 | while (1) { |
2c47e605 YZ |
5550 | ret = walk_down_tree(trans, root, path, wc); |
5551 | if (ret < 0) { | |
5552 | err = ret; | |
20524f02 | 5553 | break; |
2c47e605 | 5554 | } |
9aca1d51 | 5555 | |
2c47e605 YZ |
5556 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5557 | if (ret < 0) { | |
5558 | err = ret; | |
20524f02 | 5559 | break; |
2c47e605 YZ |
5560 | } |
5561 | ||
5562 | if (ret > 0) { | |
5563 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5564 | break; |
5565 | } | |
2c47e605 YZ |
5566 | |
5567 | if (wc->stage == DROP_REFERENCE) { | |
5568 | level = wc->level; | |
5569 | btrfs_node_key(path->nodes[level], | |
5570 | &root_item->drop_progress, | |
5571 | path->slots[level]); | |
5572 | root_item->drop_level = level; | |
5573 | } | |
5574 | ||
5575 | BUG_ON(wc->level == 0); | |
5576 | if (trans->transaction->in_commit || | |
5577 | trans->transaction->delayed_refs.flushing) { | |
5578 | ret = btrfs_update_root(trans, tree_root, | |
5579 | &root->root_key, | |
5580 | root_item); | |
5581 | BUG_ON(ret); | |
5582 | ||
5583 | btrfs_end_transaction(trans, tree_root); | |
5584 | trans = btrfs_start_transaction(tree_root, 1); | |
5585 | } else { | |
5586 | unsigned long update; | |
c3e69d58 CM |
5587 | update = trans->delayed_ref_updates; |
5588 | trans->delayed_ref_updates = 0; | |
5589 | if (update) | |
2c47e605 YZ |
5590 | btrfs_run_delayed_refs(trans, tree_root, |
5591 | update); | |
c3e69d58 | 5592 | } |
20524f02 | 5593 | } |
2c47e605 YZ |
5594 | btrfs_release_path(root, path); |
5595 | BUG_ON(err); | |
5596 | ||
5597 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
5598 | BUG_ON(ret); | |
5599 | ||
76dda93c YZ |
5600 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
5601 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
5602 | NULL, NULL); | |
5603 | BUG_ON(ret < 0); | |
5604 | if (ret > 0) { | |
5605 | ret = btrfs_del_orphan_item(trans, tree_root, | |
5606 | root->root_key.objectid); | |
5607 | BUG_ON(ret); | |
5608 | } | |
5609 | } | |
5610 | ||
5611 | if (root->in_radix) { | |
5612 | btrfs_free_fs_root(tree_root->fs_info, root); | |
5613 | } else { | |
5614 | free_extent_buffer(root->node); | |
5615 | free_extent_buffer(root->commit_root); | |
5616 | kfree(root); | |
5617 | } | |
9f3a7427 | 5618 | out: |
2c47e605 YZ |
5619 | btrfs_end_transaction(trans, tree_root); |
5620 | kfree(wc); | |
5caf2a00 | 5621 | btrfs_free_path(path); |
2c47e605 | 5622 | return err; |
20524f02 | 5623 | } |
9078a3e1 | 5624 | |
2c47e605 YZ |
5625 | /* |
5626 | * drop subtree rooted at tree block 'node'. | |
5627 | * | |
5628 | * NOTE: this function will unlock and release tree block 'node' | |
5629 | */ | |
f82d02d9 YZ |
5630 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5631 | struct btrfs_root *root, | |
5632 | struct extent_buffer *node, | |
5633 | struct extent_buffer *parent) | |
5634 | { | |
5635 | struct btrfs_path *path; | |
2c47e605 | 5636 | struct walk_control *wc; |
f82d02d9 YZ |
5637 | int level; |
5638 | int parent_level; | |
5639 | int ret = 0; | |
5640 | int wret; | |
5641 | ||
2c47e605 YZ |
5642 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5643 | ||
f82d02d9 YZ |
5644 | path = btrfs_alloc_path(); |
5645 | BUG_ON(!path); | |
5646 | ||
2c47e605 YZ |
5647 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
5648 | BUG_ON(!wc); | |
5649 | ||
b9447ef8 | 5650 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
5651 | parent_level = btrfs_header_level(parent); |
5652 | extent_buffer_get(parent); | |
5653 | path->nodes[parent_level] = parent; | |
5654 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5655 | ||
b9447ef8 | 5656 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5657 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5658 | path->nodes[level] = node; |
5659 | path->slots[level] = 0; | |
2c47e605 YZ |
5660 | path->locks[level] = 1; |
5661 | ||
5662 | wc->refs[parent_level] = 1; | |
5663 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5664 | wc->level = level; | |
5665 | wc->shared_level = -1; | |
5666 | wc->stage = DROP_REFERENCE; | |
5667 | wc->update_ref = 0; | |
5668 | wc->keep_locks = 1; | |
1c4850e2 | 5669 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
5670 | |
5671 | while (1) { | |
2c47e605 YZ |
5672 | wret = walk_down_tree(trans, root, path, wc); |
5673 | if (wret < 0) { | |
f82d02d9 | 5674 | ret = wret; |
f82d02d9 | 5675 | break; |
2c47e605 | 5676 | } |
f82d02d9 | 5677 | |
2c47e605 | 5678 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5679 | if (wret < 0) |
5680 | ret = wret; | |
5681 | if (wret != 0) | |
5682 | break; | |
5683 | } | |
5684 | ||
2c47e605 | 5685 | kfree(wc); |
f82d02d9 YZ |
5686 | btrfs_free_path(path); |
5687 | return ret; | |
5688 | } | |
5689 | ||
5d4f98a2 | 5690 | #if 0 |
8e7bf94f CM |
5691 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
5692 | unsigned long nr) | |
5693 | { | |
5694 | return min(last, start + nr - 1); | |
5695 | } | |
5696 | ||
d397712b | 5697 | static noinline int relocate_inode_pages(struct inode *inode, u64 start, |
98ed5174 | 5698 | u64 len) |
edbd8d4e CM |
5699 | { |
5700 | u64 page_start; | |
5701 | u64 page_end; | |
1a40e23b | 5702 | unsigned long first_index; |
edbd8d4e | 5703 | unsigned long last_index; |
edbd8d4e CM |
5704 | unsigned long i; |
5705 | struct page *page; | |
d1310b2e | 5706 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 5707 | struct file_ra_state *ra; |
3eaa2885 | 5708 | struct btrfs_ordered_extent *ordered; |
1a40e23b ZY |
5709 | unsigned int total_read = 0; |
5710 | unsigned int total_dirty = 0; | |
5711 | int ret = 0; | |
4313b399 CM |
5712 | |
5713 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
5714 | |
5715 | mutex_lock(&inode->i_mutex); | |
1a40e23b | 5716 | first_index = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
5717 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
5718 | ||
1a40e23b ZY |
5719 | /* make sure the dirty trick played by the caller work */ |
5720 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
5721 | first_index, last_index); | |
5722 | if (ret) | |
5723 | goto out_unlock; | |
8e7bf94f | 5724 | |
4313b399 | 5725 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 5726 | |
1a40e23b ZY |
5727 | for (i = first_index ; i <= last_index; i++) { |
5728 | if (total_read % ra->ra_pages == 0) { | |
8e7bf94f | 5729 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, |
1a40e23b | 5730 | calc_ra(i, last_index, ra->ra_pages)); |
8e7bf94f CM |
5731 | } |
5732 | total_read++; | |
3eaa2885 CM |
5733 | again: |
5734 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
1a40e23b | 5735 | BUG_ON(1); |
edbd8d4e | 5736 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 5737 | if (!page) { |
1a40e23b | 5738 | ret = -ENOMEM; |
edbd8d4e | 5739 | goto out_unlock; |
a061fc8d | 5740 | } |
edbd8d4e CM |
5741 | if (!PageUptodate(page)) { |
5742 | btrfs_readpage(NULL, page); | |
5743 | lock_page(page); | |
5744 | if (!PageUptodate(page)) { | |
5745 | unlock_page(page); | |
5746 | page_cache_release(page); | |
1a40e23b | 5747 | ret = -EIO; |
edbd8d4e CM |
5748 | goto out_unlock; |
5749 | } | |
5750 | } | |
ec44a35c | 5751 | wait_on_page_writeback(page); |
3eaa2885 | 5752 | |
edbd8d4e CM |
5753 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
5754 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
d1310b2e | 5755 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 5756 | |
3eaa2885 CM |
5757 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
5758 | if (ordered) { | |
5759 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
5760 | unlock_page(page); | |
5761 | page_cache_release(page); | |
5762 | btrfs_start_ordered_extent(inode, ordered, 1); | |
5763 | btrfs_put_ordered_extent(ordered); | |
5764 | goto again; | |
5765 | } | |
5766 | set_page_extent_mapped(page); | |
5767 | ||
1a40e23b ZY |
5768 | if (i == first_index) |
5769 | set_extent_bits(io_tree, page_start, page_end, | |
5770 | EXTENT_BOUNDARY, GFP_NOFS); | |
1f80e4db | 5771 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
1a40e23b | 5772 | |
a061fc8d | 5773 | set_page_dirty(page); |
1a40e23b | 5774 | total_dirty++; |
edbd8d4e | 5775 | |
d1310b2e | 5776 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
5777 | unlock_page(page); |
5778 | page_cache_release(page); | |
5779 | } | |
5780 | ||
5781 | out_unlock: | |
ec44a35c | 5782 | kfree(ra); |
edbd8d4e | 5783 | mutex_unlock(&inode->i_mutex); |
1a40e23b ZY |
5784 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty); |
5785 | return ret; | |
edbd8d4e CM |
5786 | } |
5787 | ||
d397712b | 5788 | static noinline int relocate_data_extent(struct inode *reloc_inode, |
1a40e23b ZY |
5789 | struct btrfs_key *extent_key, |
5790 | u64 offset) | |
5791 | { | |
5792 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
5793 | struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree; | |
5794 | struct extent_map *em; | |
6643558d YZ |
5795 | u64 start = extent_key->objectid - offset; |
5796 | u64 end = start + extent_key->offset - 1; | |
bf4ef679 | 5797 | |
1a40e23b ZY |
5798 | em = alloc_extent_map(GFP_NOFS); |
5799 | BUG_ON(!em || IS_ERR(em)); | |
bf4ef679 | 5800 | |
6643558d | 5801 | em->start = start; |
1a40e23b | 5802 | em->len = extent_key->offset; |
c8b97818 | 5803 | em->block_len = extent_key->offset; |
1a40e23b ZY |
5804 | em->block_start = extent_key->objectid; |
5805 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
5806 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
5807 | ||
5808 | /* setup extent map to cheat btrfs_readpage */ | |
6643558d | 5809 | lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
1a40e23b ZY |
5810 | while (1) { |
5811 | int ret; | |
890871be | 5812 | write_lock(&em_tree->lock); |
1a40e23b | 5813 | ret = add_extent_mapping(em_tree, em); |
890871be | 5814 | write_unlock(&em_tree->lock); |
1a40e23b ZY |
5815 | if (ret != -EEXIST) { |
5816 | free_extent_map(em); | |
bf4ef679 CM |
5817 | break; |
5818 | } | |
6643558d | 5819 | btrfs_drop_extent_cache(reloc_inode, start, end, 0); |
bf4ef679 | 5820 | } |
6643558d | 5821 | unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
bf4ef679 | 5822 | |
6643558d | 5823 | return relocate_inode_pages(reloc_inode, start, extent_key->offset); |
1a40e23b | 5824 | } |
edbd8d4e | 5825 | |
1a40e23b ZY |
5826 | struct btrfs_ref_path { |
5827 | u64 extent_start; | |
5828 | u64 nodes[BTRFS_MAX_LEVEL]; | |
5829 | u64 root_objectid; | |
5830 | u64 root_generation; | |
5831 | u64 owner_objectid; | |
1a40e23b ZY |
5832 | u32 num_refs; |
5833 | int lowest_level; | |
5834 | int current_level; | |
f82d02d9 YZ |
5835 | int shared_level; |
5836 | ||
5837 | struct btrfs_key node_keys[BTRFS_MAX_LEVEL]; | |
5838 | u64 new_nodes[BTRFS_MAX_LEVEL]; | |
1a40e23b | 5839 | }; |
7d9eb12c | 5840 | |
1a40e23b | 5841 | struct disk_extent { |
c8b97818 | 5842 | u64 ram_bytes; |
1a40e23b ZY |
5843 | u64 disk_bytenr; |
5844 | u64 disk_num_bytes; | |
5845 | u64 offset; | |
5846 | u64 num_bytes; | |
c8b97818 CM |
5847 | u8 compression; |
5848 | u8 encryption; | |
5849 | u16 other_encoding; | |
1a40e23b | 5850 | }; |
4313b399 | 5851 | |
1a40e23b ZY |
5852 | static int is_cowonly_root(u64 root_objectid) |
5853 | { | |
5854 | if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || | |
5855 | root_objectid == BTRFS_EXTENT_TREE_OBJECTID || | |
5856 | root_objectid == BTRFS_CHUNK_TREE_OBJECTID || | |
5857 | root_objectid == BTRFS_DEV_TREE_OBJECTID || | |
0403e47e YZ |
5858 | root_objectid == BTRFS_TREE_LOG_OBJECTID || |
5859 | root_objectid == BTRFS_CSUM_TREE_OBJECTID) | |
1a40e23b ZY |
5860 | return 1; |
5861 | return 0; | |
5862 | } | |
edbd8d4e | 5863 | |
d397712b | 5864 | static noinline int __next_ref_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
5865 | struct btrfs_root *extent_root, |
5866 | struct btrfs_ref_path *ref_path, | |
5867 | int first_time) | |
5868 | { | |
5869 | struct extent_buffer *leaf; | |
5870 | struct btrfs_path *path; | |
5871 | struct btrfs_extent_ref *ref; | |
5872 | struct btrfs_key key; | |
5873 | struct btrfs_key found_key; | |
5874 | u64 bytenr; | |
5875 | u32 nritems; | |
5876 | int level; | |
5877 | int ret = 1; | |
edbd8d4e | 5878 | |
1a40e23b ZY |
5879 | path = btrfs_alloc_path(); |
5880 | if (!path) | |
5881 | return -ENOMEM; | |
bf4ef679 | 5882 | |
1a40e23b ZY |
5883 | if (first_time) { |
5884 | ref_path->lowest_level = -1; | |
5885 | ref_path->current_level = -1; | |
f82d02d9 | 5886 | ref_path->shared_level = -1; |
1a40e23b ZY |
5887 | goto walk_up; |
5888 | } | |
5889 | walk_down: | |
5890 | level = ref_path->current_level - 1; | |
5891 | while (level >= -1) { | |
5892 | u64 parent; | |
5893 | if (level < ref_path->lowest_level) | |
5894 | break; | |
bf4ef679 | 5895 | |
d397712b | 5896 | if (level >= 0) |
1a40e23b | 5897 | bytenr = ref_path->nodes[level]; |
d397712b | 5898 | else |
1a40e23b | 5899 | bytenr = ref_path->extent_start; |
1a40e23b | 5900 | BUG_ON(bytenr == 0); |
bf4ef679 | 5901 | |
1a40e23b ZY |
5902 | parent = ref_path->nodes[level + 1]; |
5903 | ref_path->nodes[level + 1] = 0; | |
5904 | ref_path->current_level = level; | |
5905 | BUG_ON(parent == 0); | |
0ef3e66b | 5906 | |
1a40e23b ZY |
5907 | key.objectid = bytenr; |
5908 | key.offset = parent + 1; | |
5909 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 5910 | |
1a40e23b ZY |
5911 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
5912 | if (ret < 0) | |
edbd8d4e | 5913 | goto out; |
1a40e23b | 5914 | BUG_ON(ret == 0); |
7d9eb12c | 5915 | |
1a40e23b ZY |
5916 | leaf = path->nodes[0]; |
5917 | nritems = btrfs_header_nritems(leaf); | |
5918 | if (path->slots[0] >= nritems) { | |
5919 | ret = btrfs_next_leaf(extent_root, path); | |
5920 | if (ret < 0) | |
5921 | goto out; | |
5922 | if (ret > 0) | |
5923 | goto next; | |
5924 | leaf = path->nodes[0]; | |
5925 | } | |
0ef3e66b | 5926 | |
1a40e23b ZY |
5927 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5928 | if (found_key.objectid == bytenr && | |
f82d02d9 YZ |
5929 | found_key.type == BTRFS_EXTENT_REF_KEY) { |
5930 | if (level < ref_path->shared_level) | |
5931 | ref_path->shared_level = level; | |
1a40e23b | 5932 | goto found; |
f82d02d9 | 5933 | } |
1a40e23b ZY |
5934 | next: |
5935 | level--; | |
5936 | btrfs_release_path(extent_root, path); | |
d899e052 | 5937 | cond_resched(); |
1a40e23b ZY |
5938 | } |
5939 | /* reached lowest level */ | |
5940 | ret = 1; | |
5941 | goto out; | |
5942 | walk_up: | |
5943 | level = ref_path->current_level; | |
5944 | while (level < BTRFS_MAX_LEVEL - 1) { | |
5945 | u64 ref_objectid; | |
d397712b CM |
5946 | |
5947 | if (level >= 0) | |
1a40e23b | 5948 | bytenr = ref_path->nodes[level]; |
d397712b | 5949 | else |
1a40e23b | 5950 | bytenr = ref_path->extent_start; |
d397712b | 5951 | |
1a40e23b | 5952 | BUG_ON(bytenr == 0); |
edbd8d4e | 5953 | |
1a40e23b ZY |
5954 | key.objectid = bytenr; |
5955 | key.offset = 0; | |
5956 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 5957 | |
1a40e23b ZY |
5958 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
5959 | if (ret < 0) | |
5960 | goto out; | |
edbd8d4e | 5961 | |
1a40e23b ZY |
5962 | leaf = path->nodes[0]; |
5963 | nritems = btrfs_header_nritems(leaf); | |
5964 | if (path->slots[0] >= nritems) { | |
5965 | ret = btrfs_next_leaf(extent_root, path); | |
5966 | if (ret < 0) | |
5967 | goto out; | |
5968 | if (ret > 0) { | |
5969 | /* the extent was freed by someone */ | |
5970 | if (ref_path->lowest_level == level) | |
5971 | goto out; | |
5972 | btrfs_release_path(extent_root, path); | |
5973 | goto walk_down; | |
5974 | } | |
5975 | leaf = path->nodes[0]; | |
5976 | } | |
edbd8d4e | 5977 | |
1a40e23b ZY |
5978 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5979 | if (found_key.objectid != bytenr || | |
5980 | found_key.type != BTRFS_EXTENT_REF_KEY) { | |
5981 | /* the extent was freed by someone */ | |
5982 | if (ref_path->lowest_level == level) { | |
5983 | ret = 1; | |
5984 | goto out; | |
5985 | } | |
5986 | btrfs_release_path(extent_root, path); | |
5987 | goto walk_down; | |
5988 | } | |
5989 | found: | |
5990 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
5991 | struct btrfs_extent_ref); | |
5992 | ref_objectid = btrfs_ref_objectid(leaf, ref); | |
5993 | if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
5994 | if (first_time) { | |
5995 | level = (int)ref_objectid; | |
5996 | BUG_ON(level >= BTRFS_MAX_LEVEL); | |
5997 | ref_path->lowest_level = level; | |
5998 | ref_path->current_level = level; | |
5999 | ref_path->nodes[level] = bytenr; | |
6000 | } else { | |
6001 | WARN_ON(ref_objectid != level); | |
6002 | } | |
6003 | } else { | |
6004 | WARN_ON(level != -1); | |
6005 | } | |
6006 | first_time = 0; | |
bf4ef679 | 6007 | |
1a40e23b ZY |
6008 | if (ref_path->lowest_level == level) { |
6009 | ref_path->owner_objectid = ref_objectid; | |
1a40e23b ZY |
6010 | ref_path->num_refs = btrfs_ref_num_refs(leaf, ref); |
6011 | } | |
bf4ef679 | 6012 | |
7d9eb12c | 6013 | /* |
1a40e23b ZY |
6014 | * the block is tree root or the block isn't in reference |
6015 | * counted tree. | |
7d9eb12c | 6016 | */ |
1a40e23b ZY |
6017 | if (found_key.objectid == found_key.offset || |
6018 | is_cowonly_root(btrfs_ref_root(leaf, ref))) { | |
6019 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
6020 | ref_path->root_generation = | |
6021 | btrfs_ref_generation(leaf, ref); | |
6022 | if (level < 0) { | |
6023 | /* special reference from the tree log */ | |
6024 | ref_path->nodes[0] = found_key.offset; | |
6025 | ref_path->current_level = 0; | |
6026 | } | |
6027 | ret = 0; | |
6028 | goto out; | |
6029 | } | |
7d9eb12c | 6030 | |
1a40e23b ZY |
6031 | level++; |
6032 | BUG_ON(ref_path->nodes[level] != 0); | |
6033 | ref_path->nodes[level] = found_key.offset; | |
6034 | ref_path->current_level = level; | |
bf4ef679 | 6035 | |
1a40e23b ZY |
6036 | /* |
6037 | * the reference was created in the running transaction, | |
6038 | * no need to continue walking up. | |
6039 | */ | |
6040 | if (btrfs_ref_generation(leaf, ref) == trans->transid) { | |
6041 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
6042 | ref_path->root_generation = | |
6043 | btrfs_ref_generation(leaf, ref); | |
6044 | ret = 0; | |
6045 | goto out; | |
7d9eb12c CM |
6046 | } |
6047 | ||
1a40e23b | 6048 | btrfs_release_path(extent_root, path); |
d899e052 | 6049 | cond_resched(); |
7d9eb12c | 6050 | } |
1a40e23b ZY |
6051 | /* reached max tree level, but no tree root found. */ |
6052 | BUG(); | |
edbd8d4e | 6053 | out: |
1a40e23b ZY |
6054 | btrfs_free_path(path); |
6055 | return ret; | |
edbd8d4e CM |
6056 | } |
6057 | ||
1a40e23b ZY |
6058 | static int btrfs_first_ref_path(struct btrfs_trans_handle *trans, |
6059 | struct btrfs_root *extent_root, | |
6060 | struct btrfs_ref_path *ref_path, | |
6061 | u64 extent_start) | |
a061fc8d | 6062 | { |
1a40e23b ZY |
6063 | memset(ref_path, 0, sizeof(*ref_path)); |
6064 | ref_path->extent_start = extent_start; | |
a061fc8d | 6065 | |
1a40e23b | 6066 | return __next_ref_path(trans, extent_root, ref_path, 1); |
a061fc8d CM |
6067 | } |
6068 | ||
1a40e23b ZY |
6069 | static int btrfs_next_ref_path(struct btrfs_trans_handle *trans, |
6070 | struct btrfs_root *extent_root, | |
6071 | struct btrfs_ref_path *ref_path) | |
edbd8d4e | 6072 | { |
1a40e23b ZY |
6073 | return __next_ref_path(trans, extent_root, ref_path, 0); |
6074 | } | |
6075 | ||
d397712b | 6076 | static noinline int get_new_locations(struct inode *reloc_inode, |
1a40e23b ZY |
6077 | struct btrfs_key *extent_key, |
6078 | u64 offset, int no_fragment, | |
6079 | struct disk_extent **extents, | |
6080 | int *nr_extents) | |
6081 | { | |
6082 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
6083 | struct btrfs_path *path; | |
6084 | struct btrfs_file_extent_item *fi; | |
edbd8d4e | 6085 | struct extent_buffer *leaf; |
1a40e23b ZY |
6086 | struct disk_extent *exts = *extents; |
6087 | struct btrfs_key found_key; | |
6088 | u64 cur_pos; | |
6089 | u64 last_byte; | |
edbd8d4e | 6090 | u32 nritems; |
1a40e23b ZY |
6091 | int nr = 0; |
6092 | int max = *nr_extents; | |
6093 | int ret; | |
edbd8d4e | 6094 | |
1a40e23b ZY |
6095 | WARN_ON(!no_fragment && *extents); |
6096 | if (!exts) { | |
6097 | max = 1; | |
6098 | exts = kmalloc(sizeof(*exts) * max, GFP_NOFS); | |
6099 | if (!exts) | |
6100 | return -ENOMEM; | |
a061fc8d | 6101 | } |
edbd8d4e | 6102 | |
1a40e23b ZY |
6103 | path = btrfs_alloc_path(); |
6104 | BUG_ON(!path); | |
edbd8d4e | 6105 | |
1a40e23b ZY |
6106 | cur_pos = extent_key->objectid - offset; |
6107 | last_byte = extent_key->objectid + extent_key->offset; | |
6108 | ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino, | |
6109 | cur_pos, 0); | |
6110 | if (ret < 0) | |
6111 | goto out; | |
6112 | if (ret > 0) { | |
6113 | ret = -ENOENT; | |
6114 | goto out; | |
6115 | } | |
edbd8d4e | 6116 | |
1a40e23b | 6117 | while (1) { |
edbd8d4e CM |
6118 | leaf = path->nodes[0]; |
6119 | nritems = btrfs_header_nritems(leaf); | |
1a40e23b ZY |
6120 | if (path->slots[0] >= nritems) { |
6121 | ret = btrfs_next_leaf(root, path); | |
a061fc8d CM |
6122 | if (ret < 0) |
6123 | goto out; | |
1a40e23b ZY |
6124 | if (ret > 0) |
6125 | break; | |
bf4ef679 | 6126 | leaf = path->nodes[0]; |
a061fc8d | 6127 | } |
edbd8d4e CM |
6128 | |
6129 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b ZY |
6130 | if (found_key.offset != cur_pos || |
6131 | found_key.type != BTRFS_EXTENT_DATA_KEY || | |
6132 | found_key.objectid != reloc_inode->i_ino) | |
edbd8d4e CM |
6133 | break; |
6134 | ||
1a40e23b ZY |
6135 | fi = btrfs_item_ptr(leaf, path->slots[0], |
6136 | struct btrfs_file_extent_item); | |
6137 | if (btrfs_file_extent_type(leaf, fi) != | |
6138 | BTRFS_FILE_EXTENT_REG || | |
6139 | btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
edbd8d4e | 6140 | break; |
1a40e23b ZY |
6141 | |
6142 | if (nr == max) { | |
6143 | struct disk_extent *old = exts; | |
6144 | max *= 2; | |
6145 | exts = kzalloc(sizeof(*exts) * max, GFP_NOFS); | |
6146 | memcpy(exts, old, sizeof(*exts) * nr); | |
6147 | if (old != *extents) | |
6148 | kfree(old); | |
a061fc8d | 6149 | } |
edbd8d4e | 6150 | |
1a40e23b ZY |
6151 | exts[nr].disk_bytenr = |
6152 | btrfs_file_extent_disk_bytenr(leaf, fi); | |
6153 | exts[nr].disk_num_bytes = | |
6154 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6155 | exts[nr].offset = btrfs_file_extent_offset(leaf, fi); | |
6156 | exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
c8b97818 CM |
6157 | exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
6158 | exts[nr].compression = btrfs_file_extent_compression(leaf, fi); | |
6159 | exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi); | |
6160 | exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf, | |
6161 | fi); | |
d899e052 YZ |
6162 | BUG_ON(exts[nr].offset > 0); |
6163 | BUG_ON(exts[nr].compression || exts[nr].encryption); | |
6164 | BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes); | |
edbd8d4e | 6165 | |
1a40e23b ZY |
6166 | cur_pos += exts[nr].num_bytes; |
6167 | nr++; | |
6168 | ||
6169 | if (cur_pos + offset >= last_byte) | |
6170 | break; | |
6171 | ||
6172 | if (no_fragment) { | |
6173 | ret = 1; | |
edbd8d4e | 6174 | goto out; |
1a40e23b ZY |
6175 | } |
6176 | path->slots[0]++; | |
6177 | } | |
6178 | ||
1f80e4db | 6179 | BUG_ON(cur_pos + offset > last_byte); |
1a40e23b ZY |
6180 | if (cur_pos + offset < last_byte) { |
6181 | ret = -ENOENT; | |
6182 | goto out; | |
edbd8d4e CM |
6183 | } |
6184 | ret = 0; | |
6185 | out: | |
1a40e23b ZY |
6186 | btrfs_free_path(path); |
6187 | if (ret) { | |
6188 | if (exts != *extents) | |
6189 | kfree(exts); | |
6190 | } else { | |
6191 | *extents = exts; | |
6192 | *nr_extents = nr; | |
6193 | } | |
6194 | return ret; | |
6195 | } | |
6196 | ||
d397712b | 6197 | static noinline int replace_one_extent(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6198 | struct btrfs_root *root, |
6199 | struct btrfs_path *path, | |
6200 | struct btrfs_key *extent_key, | |
6201 | struct btrfs_key *leaf_key, | |
6202 | struct btrfs_ref_path *ref_path, | |
6203 | struct disk_extent *new_extents, | |
6204 | int nr_extents) | |
6205 | { | |
6206 | struct extent_buffer *leaf; | |
6207 | struct btrfs_file_extent_item *fi; | |
6208 | struct inode *inode = NULL; | |
6209 | struct btrfs_key key; | |
6210 | u64 lock_start = 0; | |
6211 | u64 lock_end = 0; | |
6212 | u64 num_bytes; | |
6213 | u64 ext_offset; | |
86288a19 | 6214 | u64 search_end = (u64)-1; |
1a40e23b | 6215 | u32 nritems; |
3bb1a1bc | 6216 | int nr_scaned = 0; |
1a40e23b | 6217 | int extent_locked = 0; |
d899e052 | 6218 | int extent_type; |
1a40e23b ZY |
6219 | int ret; |
6220 | ||
3bb1a1bc | 6221 | memcpy(&key, leaf_key, sizeof(key)); |
1a40e23b | 6222 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { |
3bb1a1bc YZ |
6223 | if (key.objectid < ref_path->owner_objectid || |
6224 | (key.objectid == ref_path->owner_objectid && | |
6225 | key.type < BTRFS_EXTENT_DATA_KEY)) { | |
6226 | key.objectid = ref_path->owner_objectid; | |
6227 | key.type = BTRFS_EXTENT_DATA_KEY; | |
6228 | key.offset = 0; | |
6229 | } | |
1a40e23b ZY |
6230 | } |
6231 | ||
6232 | while (1) { | |
6233 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
6234 | if (ret < 0) | |
6235 | goto out; | |
6236 | ||
6237 | leaf = path->nodes[0]; | |
6238 | nritems = btrfs_header_nritems(leaf); | |
6239 | next: | |
6240 | if (extent_locked && ret > 0) { | |
6241 | /* | |
6242 | * the file extent item was modified by someone | |
6243 | * before the extent got locked. | |
6244 | */ | |
1a40e23b ZY |
6245 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
6246 | lock_end, GFP_NOFS); | |
6247 | extent_locked = 0; | |
6248 | } | |
6249 | ||
6250 | if (path->slots[0] >= nritems) { | |
3bb1a1bc | 6251 | if (++nr_scaned > 2) |
1a40e23b ZY |
6252 | break; |
6253 | ||
6254 | BUG_ON(extent_locked); | |
6255 | ret = btrfs_next_leaf(root, path); | |
6256 | if (ret < 0) | |
6257 | goto out; | |
6258 | if (ret > 0) | |
6259 | break; | |
6260 | leaf = path->nodes[0]; | |
6261 | nritems = btrfs_header_nritems(leaf); | |
6262 | } | |
6263 | ||
6264 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
6265 | ||
6266 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { | |
6267 | if ((key.objectid > ref_path->owner_objectid) || | |
6268 | (key.objectid == ref_path->owner_objectid && | |
6269 | key.type > BTRFS_EXTENT_DATA_KEY) || | |
86288a19 | 6270 | key.offset >= search_end) |
1a40e23b ZY |
6271 | break; |
6272 | } | |
6273 | ||
6274 | if (inode && key.objectid != inode->i_ino) { | |
6275 | BUG_ON(extent_locked); | |
6276 | btrfs_release_path(root, path); | |
6277 | mutex_unlock(&inode->i_mutex); | |
6278 | iput(inode); | |
6279 | inode = NULL; | |
6280 | continue; | |
6281 | } | |
6282 | ||
6283 | if (key.type != BTRFS_EXTENT_DATA_KEY) { | |
6284 | path->slots[0]++; | |
6285 | ret = 1; | |
6286 | goto next; | |
6287 | } | |
6288 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
6289 | struct btrfs_file_extent_item); | |
d899e052 YZ |
6290 | extent_type = btrfs_file_extent_type(leaf, fi); |
6291 | if ((extent_type != BTRFS_FILE_EXTENT_REG && | |
6292 | extent_type != BTRFS_FILE_EXTENT_PREALLOC) || | |
1a40e23b ZY |
6293 | (btrfs_file_extent_disk_bytenr(leaf, fi) != |
6294 | extent_key->objectid)) { | |
6295 | path->slots[0]++; | |
6296 | ret = 1; | |
6297 | goto next; | |
6298 | } | |
6299 | ||
6300 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
6301 | ext_offset = btrfs_file_extent_offset(leaf, fi); | |
6302 | ||
86288a19 YZ |
6303 | if (search_end == (u64)-1) { |
6304 | search_end = key.offset - ext_offset + | |
6305 | btrfs_file_extent_ram_bytes(leaf, fi); | |
6306 | } | |
1a40e23b ZY |
6307 | |
6308 | if (!extent_locked) { | |
6309 | lock_start = key.offset; | |
6310 | lock_end = lock_start + num_bytes - 1; | |
6311 | } else { | |
6643558d YZ |
6312 | if (lock_start > key.offset || |
6313 | lock_end + 1 < key.offset + num_bytes) { | |
6314 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
6315 | lock_start, lock_end, GFP_NOFS); | |
6316 | extent_locked = 0; | |
6317 | } | |
1a40e23b ZY |
6318 | } |
6319 | ||
6320 | if (!inode) { | |
6321 | btrfs_release_path(root, path); | |
6322 | ||
6323 | inode = btrfs_iget_locked(root->fs_info->sb, | |
6324 | key.objectid, root); | |
6325 | if (inode->i_state & I_NEW) { | |
6326 | BTRFS_I(inode)->root = root; | |
6327 | BTRFS_I(inode)->location.objectid = | |
6328 | key.objectid; | |
6329 | BTRFS_I(inode)->location.type = | |
6330 | BTRFS_INODE_ITEM_KEY; | |
6331 | BTRFS_I(inode)->location.offset = 0; | |
6332 | btrfs_read_locked_inode(inode); | |
6333 | unlock_new_inode(inode); | |
6334 | } | |
6335 | /* | |
6336 | * some code call btrfs_commit_transaction while | |
6337 | * holding the i_mutex, so we can't use mutex_lock | |
6338 | * here. | |
6339 | */ | |
6340 | if (is_bad_inode(inode) || | |
6341 | !mutex_trylock(&inode->i_mutex)) { | |
6342 | iput(inode); | |
6343 | inode = NULL; | |
6344 | key.offset = (u64)-1; | |
6345 | goto skip; | |
6346 | } | |
6347 | } | |
6348 | ||
6349 | if (!extent_locked) { | |
6350 | struct btrfs_ordered_extent *ordered; | |
6351 | ||
6352 | btrfs_release_path(root, path); | |
6353 | ||
6354 | lock_extent(&BTRFS_I(inode)->io_tree, lock_start, | |
6355 | lock_end, GFP_NOFS); | |
6356 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
6357 | lock_end); | |
6358 | if (ordered && | |
6359 | ordered->file_offset <= lock_end && | |
6360 | ordered->file_offset + ordered->len > lock_start) { | |
6361 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
6362 | lock_start, lock_end, GFP_NOFS); | |
6363 | btrfs_start_ordered_extent(inode, ordered, 1); | |
6364 | btrfs_put_ordered_extent(ordered); | |
6365 | key.offset += num_bytes; | |
6366 | goto skip; | |
6367 | } | |
6368 | if (ordered) | |
6369 | btrfs_put_ordered_extent(ordered); | |
6370 | ||
1a40e23b ZY |
6371 | extent_locked = 1; |
6372 | continue; | |
6373 | } | |
6374 | ||
6375 | if (nr_extents == 1) { | |
6376 | /* update extent pointer in place */ | |
1a40e23b ZY |
6377 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
6378 | new_extents[0].disk_bytenr); | |
6379 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
6380 | new_extents[0].disk_num_bytes); | |
1a40e23b ZY |
6381 | btrfs_mark_buffer_dirty(leaf); |
6382 | ||
6383 | btrfs_drop_extent_cache(inode, key.offset, | |
6384 | key.offset + num_bytes - 1, 0); | |
6385 | ||
6386 | ret = btrfs_inc_extent_ref(trans, root, | |
6387 | new_extents[0].disk_bytenr, | |
6388 | new_extents[0].disk_num_bytes, | |
6389 | leaf->start, | |
6390 | root->root_key.objectid, | |
6391 | trans->transid, | |
3bb1a1bc | 6392 | key.objectid); |
1a40e23b ZY |
6393 | BUG_ON(ret); |
6394 | ||
6395 | ret = btrfs_free_extent(trans, root, | |
6396 | extent_key->objectid, | |
6397 | extent_key->offset, | |
6398 | leaf->start, | |
6399 | btrfs_header_owner(leaf), | |
6400 | btrfs_header_generation(leaf), | |
3bb1a1bc | 6401 | key.objectid, 0); |
1a40e23b ZY |
6402 | BUG_ON(ret); |
6403 | ||
6404 | btrfs_release_path(root, path); | |
6405 | key.offset += num_bytes; | |
6406 | } else { | |
d899e052 YZ |
6407 | BUG_ON(1); |
6408 | #if 0 | |
1a40e23b ZY |
6409 | u64 alloc_hint; |
6410 | u64 extent_len; | |
6411 | int i; | |
6412 | /* | |
6413 | * drop old extent pointer at first, then insert the | |
6414 | * new pointers one bye one | |
6415 | */ | |
6416 | btrfs_release_path(root, path); | |
6417 | ret = btrfs_drop_extents(trans, root, inode, key.offset, | |
6418 | key.offset + num_bytes, | |
6419 | key.offset, &alloc_hint); | |
6420 | BUG_ON(ret); | |
6421 | ||
6422 | for (i = 0; i < nr_extents; i++) { | |
6423 | if (ext_offset >= new_extents[i].num_bytes) { | |
6424 | ext_offset -= new_extents[i].num_bytes; | |
6425 | continue; | |
6426 | } | |
6427 | extent_len = min(new_extents[i].num_bytes - | |
6428 | ext_offset, num_bytes); | |
6429 | ||
6430 | ret = btrfs_insert_empty_item(trans, root, | |
6431 | path, &key, | |
6432 | sizeof(*fi)); | |
6433 | BUG_ON(ret); | |
6434 | ||
6435 | leaf = path->nodes[0]; | |
6436 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
6437 | struct btrfs_file_extent_item); | |
6438 | btrfs_set_file_extent_generation(leaf, fi, | |
6439 | trans->transid); | |
6440 | btrfs_set_file_extent_type(leaf, fi, | |
6441 | BTRFS_FILE_EXTENT_REG); | |
6442 | btrfs_set_file_extent_disk_bytenr(leaf, fi, | |
6443 | new_extents[i].disk_bytenr); | |
6444 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
6445 | new_extents[i].disk_num_bytes); | |
c8b97818 CM |
6446 | btrfs_set_file_extent_ram_bytes(leaf, fi, |
6447 | new_extents[i].ram_bytes); | |
6448 | ||
6449 | btrfs_set_file_extent_compression(leaf, fi, | |
6450 | new_extents[i].compression); | |
6451 | btrfs_set_file_extent_encryption(leaf, fi, | |
6452 | new_extents[i].encryption); | |
6453 | btrfs_set_file_extent_other_encoding(leaf, fi, | |
6454 | new_extents[i].other_encoding); | |
6455 | ||
1a40e23b ZY |
6456 | btrfs_set_file_extent_num_bytes(leaf, fi, |
6457 | extent_len); | |
6458 | ext_offset += new_extents[i].offset; | |
6459 | btrfs_set_file_extent_offset(leaf, fi, | |
6460 | ext_offset); | |
6461 | btrfs_mark_buffer_dirty(leaf); | |
6462 | ||
6463 | btrfs_drop_extent_cache(inode, key.offset, | |
6464 | key.offset + extent_len - 1, 0); | |
6465 | ||
6466 | ret = btrfs_inc_extent_ref(trans, root, | |
6467 | new_extents[i].disk_bytenr, | |
6468 | new_extents[i].disk_num_bytes, | |
6469 | leaf->start, | |
6470 | root->root_key.objectid, | |
3bb1a1bc | 6471 | trans->transid, key.objectid); |
1a40e23b ZY |
6472 | BUG_ON(ret); |
6473 | btrfs_release_path(root, path); | |
6474 | ||
a76a3cd4 | 6475 | inode_add_bytes(inode, extent_len); |
1a40e23b ZY |
6476 | |
6477 | ext_offset = 0; | |
6478 | num_bytes -= extent_len; | |
6479 | key.offset += extent_len; | |
6480 | ||
6481 | if (num_bytes == 0) | |
6482 | break; | |
6483 | } | |
6484 | BUG_ON(i >= nr_extents); | |
d899e052 | 6485 | #endif |
1a40e23b ZY |
6486 | } |
6487 | ||
6488 | if (extent_locked) { | |
1a40e23b ZY |
6489 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
6490 | lock_end, GFP_NOFS); | |
6491 | extent_locked = 0; | |
6492 | } | |
6493 | skip: | |
6494 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS && | |
86288a19 | 6495 | key.offset >= search_end) |
1a40e23b ZY |
6496 | break; |
6497 | ||
6498 | cond_resched(); | |
6499 | } | |
6500 | ret = 0; | |
6501 | out: | |
6502 | btrfs_release_path(root, path); | |
6503 | if (inode) { | |
6504 | mutex_unlock(&inode->i_mutex); | |
6505 | if (extent_locked) { | |
1a40e23b ZY |
6506 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
6507 | lock_end, GFP_NOFS); | |
6508 | } | |
6509 | iput(inode); | |
6510 | } | |
6511 | return ret; | |
6512 | } | |
6513 | ||
1a40e23b ZY |
6514 | int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, |
6515 | struct btrfs_root *root, | |
6516 | struct extent_buffer *buf, u64 orig_start) | |
6517 | { | |
6518 | int level; | |
6519 | int ret; | |
6520 | ||
6521 | BUG_ON(btrfs_header_generation(buf) != trans->transid); | |
6522 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
6523 | ||
6524 | level = btrfs_header_level(buf); | |
6525 | if (level == 0) { | |
6526 | struct btrfs_leaf_ref *ref; | |
6527 | struct btrfs_leaf_ref *orig_ref; | |
6528 | ||
6529 | orig_ref = btrfs_lookup_leaf_ref(root, orig_start); | |
6530 | if (!orig_ref) | |
6531 | return -ENOENT; | |
6532 | ||
6533 | ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems); | |
6534 | if (!ref) { | |
6535 | btrfs_free_leaf_ref(root, orig_ref); | |
6536 | return -ENOMEM; | |
6537 | } | |
6538 | ||
6539 | ref->nritems = orig_ref->nritems; | |
6540 | memcpy(ref->extents, orig_ref->extents, | |
6541 | sizeof(ref->extents[0]) * ref->nritems); | |
6542 | ||
6543 | btrfs_free_leaf_ref(root, orig_ref); | |
6544 | ||
6545 | ref->root_gen = trans->transid; | |
6546 | ref->bytenr = buf->start; | |
6547 | ref->owner = btrfs_header_owner(buf); | |
6548 | ref->generation = btrfs_header_generation(buf); | |
bd56b302 | 6549 | |
1a40e23b ZY |
6550 | ret = btrfs_add_leaf_ref(root, ref, 0); |
6551 | WARN_ON(ret); | |
6552 | btrfs_free_leaf_ref(root, ref); | |
6553 | } | |
6554 | return 0; | |
6555 | } | |
6556 | ||
d397712b | 6557 | static noinline int invalidate_extent_cache(struct btrfs_root *root, |
1a40e23b ZY |
6558 | struct extent_buffer *leaf, |
6559 | struct btrfs_block_group_cache *group, | |
6560 | struct btrfs_root *target_root) | |
6561 | { | |
6562 | struct btrfs_key key; | |
6563 | struct inode *inode = NULL; | |
6564 | struct btrfs_file_extent_item *fi; | |
6565 | u64 num_bytes; | |
6566 | u64 skip_objectid = 0; | |
6567 | u32 nritems; | |
6568 | u32 i; | |
6569 | ||
6570 | nritems = btrfs_header_nritems(leaf); | |
6571 | for (i = 0; i < nritems; i++) { | |
6572 | btrfs_item_key_to_cpu(leaf, &key, i); | |
6573 | if (key.objectid == skip_objectid || | |
6574 | key.type != BTRFS_EXTENT_DATA_KEY) | |
6575 | continue; | |
6576 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
6577 | if (btrfs_file_extent_type(leaf, fi) == | |
6578 | BTRFS_FILE_EXTENT_INLINE) | |
6579 | continue; | |
6580 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
6581 | continue; | |
6582 | if (!inode || inode->i_ino != key.objectid) { | |
6583 | iput(inode); | |
6584 | inode = btrfs_ilookup(target_root->fs_info->sb, | |
6585 | key.objectid, target_root, 1); | |
6586 | } | |
6587 | if (!inode) { | |
6588 | skip_objectid = key.objectid; | |
6589 | continue; | |
6590 | } | |
6591 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
6592 | ||
6593 | lock_extent(&BTRFS_I(inode)->io_tree, key.offset, | |
6594 | key.offset + num_bytes - 1, GFP_NOFS); | |
1a40e23b ZY |
6595 | btrfs_drop_extent_cache(inode, key.offset, |
6596 | key.offset + num_bytes - 1, 1); | |
1a40e23b ZY |
6597 | unlock_extent(&BTRFS_I(inode)->io_tree, key.offset, |
6598 | key.offset + num_bytes - 1, GFP_NOFS); | |
6599 | cond_resched(); | |
6600 | } | |
6601 | iput(inode); | |
6602 | return 0; | |
6603 | } | |
6604 | ||
d397712b | 6605 | static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6606 | struct btrfs_root *root, |
6607 | struct extent_buffer *leaf, | |
6608 | struct btrfs_block_group_cache *group, | |
6609 | struct inode *reloc_inode) | |
6610 | { | |
6611 | struct btrfs_key key; | |
6612 | struct btrfs_key extent_key; | |
6613 | struct btrfs_file_extent_item *fi; | |
6614 | struct btrfs_leaf_ref *ref; | |
6615 | struct disk_extent *new_extent; | |
6616 | u64 bytenr; | |
6617 | u64 num_bytes; | |
6618 | u32 nritems; | |
6619 | u32 i; | |
6620 | int ext_index; | |
6621 | int nr_extent; | |
6622 | int ret; | |
6623 | ||
6624 | new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS); | |
6625 | BUG_ON(!new_extent); | |
6626 | ||
6627 | ref = btrfs_lookup_leaf_ref(root, leaf->start); | |
6628 | BUG_ON(!ref); | |
6629 | ||
6630 | ext_index = -1; | |
6631 | nritems = btrfs_header_nritems(leaf); | |
6632 | for (i = 0; i < nritems; i++) { | |
6633 | btrfs_item_key_to_cpu(leaf, &key, i); | |
6634 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6635 | continue; | |
6636 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
6637 | if (btrfs_file_extent_type(leaf, fi) == | |
6638 | BTRFS_FILE_EXTENT_INLINE) | |
6639 | continue; | |
6640 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
6641 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
6642 | if (bytenr == 0) | |
6643 | continue; | |
6644 | ||
6645 | ext_index++; | |
6646 | if (bytenr >= group->key.objectid + group->key.offset || | |
6647 | bytenr + num_bytes <= group->key.objectid) | |
6648 | continue; | |
6649 | ||
6650 | extent_key.objectid = bytenr; | |
6651 | extent_key.offset = num_bytes; | |
6652 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
6653 | nr_extent = 1; | |
6654 | ret = get_new_locations(reloc_inode, &extent_key, | |
6655 | group->key.objectid, 1, | |
6656 | &new_extent, &nr_extent); | |
6657 | if (ret > 0) | |
6658 | continue; | |
6659 | BUG_ON(ret < 0); | |
6660 | ||
6661 | BUG_ON(ref->extents[ext_index].bytenr != bytenr); | |
6662 | BUG_ON(ref->extents[ext_index].num_bytes != num_bytes); | |
6663 | ref->extents[ext_index].bytenr = new_extent->disk_bytenr; | |
6664 | ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes; | |
6665 | ||
1a40e23b ZY |
6666 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
6667 | new_extent->disk_bytenr); | |
6668 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
6669 | new_extent->disk_num_bytes); | |
1a40e23b ZY |
6670 | btrfs_mark_buffer_dirty(leaf); |
6671 | ||
6672 | ret = btrfs_inc_extent_ref(trans, root, | |
6673 | new_extent->disk_bytenr, | |
6674 | new_extent->disk_num_bytes, | |
6675 | leaf->start, | |
6676 | root->root_key.objectid, | |
3bb1a1bc | 6677 | trans->transid, key.objectid); |
1a40e23b | 6678 | BUG_ON(ret); |
56bec294 | 6679 | |
1a40e23b ZY |
6680 | ret = btrfs_free_extent(trans, root, |
6681 | bytenr, num_bytes, leaf->start, | |
6682 | btrfs_header_owner(leaf), | |
6683 | btrfs_header_generation(leaf), | |
3bb1a1bc | 6684 | key.objectid, 0); |
1a40e23b ZY |
6685 | BUG_ON(ret); |
6686 | cond_resched(); | |
6687 | } | |
6688 | kfree(new_extent); | |
6689 | BUG_ON(ext_index + 1 != ref->nritems); | |
6690 | btrfs_free_leaf_ref(root, ref); | |
6691 | return 0; | |
6692 | } | |
6693 | ||
f82d02d9 YZ |
6694 | int btrfs_free_reloc_root(struct btrfs_trans_handle *trans, |
6695 | struct btrfs_root *root) | |
1a40e23b ZY |
6696 | { |
6697 | struct btrfs_root *reloc_root; | |
f82d02d9 | 6698 | int ret; |
1a40e23b ZY |
6699 | |
6700 | if (root->reloc_root) { | |
6701 | reloc_root = root->reloc_root; | |
6702 | root->reloc_root = NULL; | |
6703 | list_add(&reloc_root->dead_list, | |
6704 | &root->fs_info->dead_reloc_roots); | |
f82d02d9 YZ |
6705 | |
6706 | btrfs_set_root_bytenr(&reloc_root->root_item, | |
6707 | reloc_root->node->start); | |
6708 | btrfs_set_root_level(&root->root_item, | |
6709 | btrfs_header_level(reloc_root->node)); | |
6710 | memset(&reloc_root->root_item.drop_progress, 0, | |
6711 | sizeof(struct btrfs_disk_key)); | |
6712 | reloc_root->root_item.drop_level = 0; | |
6713 | ||
6714 | ret = btrfs_update_root(trans, root->fs_info->tree_root, | |
6715 | &reloc_root->root_key, | |
6716 | &reloc_root->root_item); | |
6717 | BUG_ON(ret); | |
1a40e23b ZY |
6718 | } |
6719 | return 0; | |
6720 | } | |
6721 | ||
6722 | int btrfs_drop_dead_reloc_roots(struct btrfs_root *root) | |
6723 | { | |
6724 | struct btrfs_trans_handle *trans; | |
6725 | struct btrfs_root *reloc_root; | |
6726 | struct btrfs_root *prev_root = NULL; | |
6727 | struct list_head dead_roots; | |
6728 | int ret; | |
6729 | unsigned long nr; | |
6730 | ||
6731 | INIT_LIST_HEAD(&dead_roots); | |
6732 | list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots); | |
6733 | ||
6734 | while (!list_empty(&dead_roots)) { | |
6735 | reloc_root = list_entry(dead_roots.prev, | |
6736 | struct btrfs_root, dead_list); | |
6737 | list_del_init(&reloc_root->dead_list); | |
6738 | ||
6739 | BUG_ON(reloc_root->commit_root != NULL); | |
6740 | while (1) { | |
6741 | trans = btrfs_join_transaction(root, 1); | |
6742 | BUG_ON(!trans); | |
6743 | ||
6744 | mutex_lock(&root->fs_info->drop_mutex); | |
6745 | ret = btrfs_drop_snapshot(trans, reloc_root); | |
6746 | if (ret != -EAGAIN) | |
6747 | break; | |
6748 | mutex_unlock(&root->fs_info->drop_mutex); | |
6749 | ||
6750 | nr = trans->blocks_used; | |
6751 | ret = btrfs_end_transaction(trans, root); | |
6752 | BUG_ON(ret); | |
6753 | btrfs_btree_balance_dirty(root, nr); | |
6754 | } | |
6755 | ||
6756 | free_extent_buffer(reloc_root->node); | |
6757 | ||
6758 | ret = btrfs_del_root(trans, root->fs_info->tree_root, | |
6759 | &reloc_root->root_key); | |
6760 | BUG_ON(ret); | |
6761 | mutex_unlock(&root->fs_info->drop_mutex); | |
6762 | ||
6763 | nr = trans->blocks_used; | |
6764 | ret = btrfs_end_transaction(trans, root); | |
6765 | BUG_ON(ret); | |
6766 | btrfs_btree_balance_dirty(root, nr); | |
6767 | ||
6768 | kfree(prev_root); | |
6769 | prev_root = reloc_root; | |
6770 | } | |
6771 | if (prev_root) { | |
6772 | btrfs_remove_leaf_refs(prev_root, (u64)-1, 0); | |
6773 | kfree(prev_root); | |
6774 | } | |
6775 | return 0; | |
6776 | } | |
6777 | ||
6778 | int btrfs_add_dead_reloc_root(struct btrfs_root *root) | |
6779 | { | |
6780 | list_add(&root->dead_list, &root->fs_info->dead_reloc_roots); | |
6781 | return 0; | |
6782 | } | |
6783 | ||
6784 | int btrfs_cleanup_reloc_trees(struct btrfs_root *root) | |
6785 | { | |
6786 | struct btrfs_root *reloc_root; | |
6787 | struct btrfs_trans_handle *trans; | |
6788 | struct btrfs_key location; | |
6789 | int found; | |
6790 | int ret; | |
6791 | ||
6792 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
6793 | ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL); | |
6794 | BUG_ON(ret); | |
6795 | found = !list_empty(&root->fs_info->dead_reloc_roots); | |
6796 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
6797 | ||
6798 | if (found) { | |
6799 | trans = btrfs_start_transaction(root, 1); | |
6800 | BUG_ON(!trans); | |
6801 | ret = btrfs_commit_transaction(trans, root); | |
6802 | BUG_ON(ret); | |
6803 | } | |
6804 | ||
6805 | location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
6806 | location.offset = (u64)-1; | |
6807 | location.type = BTRFS_ROOT_ITEM_KEY; | |
6808 | ||
6809 | reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location); | |
6810 | BUG_ON(!reloc_root); | |
6811 | btrfs_orphan_cleanup(reloc_root); | |
6812 | return 0; | |
6813 | } | |
6814 | ||
d397712b | 6815 | static noinline int init_reloc_tree(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6816 | struct btrfs_root *root) |
6817 | { | |
6818 | struct btrfs_root *reloc_root; | |
6819 | struct extent_buffer *eb; | |
6820 | struct btrfs_root_item *root_item; | |
6821 | struct btrfs_key root_key; | |
6822 | int ret; | |
6823 | ||
6824 | BUG_ON(!root->ref_cows); | |
6825 | if (root->reloc_root) | |
6826 | return 0; | |
6827 | ||
6828 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); | |
6829 | BUG_ON(!root_item); | |
6830 | ||
6831 | ret = btrfs_copy_root(trans, root, root->commit_root, | |
6832 | &eb, BTRFS_TREE_RELOC_OBJECTID); | |
6833 | BUG_ON(ret); | |
6834 | ||
6835 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
6836 | root_key.offset = root->root_key.objectid; | |
6837 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
6838 | ||
6839 | memcpy(root_item, &root->root_item, sizeof(root_item)); | |
6840 | btrfs_set_root_refs(root_item, 0); | |
6841 | btrfs_set_root_bytenr(root_item, eb->start); | |
6842 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
84234f3a | 6843 | btrfs_set_root_generation(root_item, trans->transid); |
1a40e23b ZY |
6844 | |
6845 | btrfs_tree_unlock(eb); | |
6846 | free_extent_buffer(eb); | |
6847 | ||
6848 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, | |
6849 | &root_key, root_item); | |
6850 | BUG_ON(ret); | |
6851 | kfree(root_item); | |
6852 | ||
6853 | reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root, | |
6854 | &root_key); | |
6855 | BUG_ON(!reloc_root); | |
6856 | reloc_root->last_trans = trans->transid; | |
6857 | reloc_root->commit_root = NULL; | |
6858 | reloc_root->ref_tree = &root->fs_info->reloc_ref_tree; | |
6859 | ||
6860 | root->reloc_root = reloc_root; | |
6861 | return 0; | |
6862 | } | |
6863 | ||
6864 | /* | |
6865 | * Core function of space balance. | |
6866 | * | |
6867 | * The idea is using reloc trees to relocate tree blocks in reference | |
f82d02d9 YZ |
6868 | * counted roots. There is one reloc tree for each subvol, and all |
6869 | * reloc trees share same root key objectid. Reloc trees are snapshots | |
6870 | * of the latest committed roots of subvols (root->commit_root). | |
6871 | * | |
6872 | * To relocate a tree block referenced by a subvol, there are two steps. | |
6873 | * COW the block through subvol's reloc tree, then update block pointer | |
6874 | * in the subvol to point to the new block. Since all reloc trees share | |
6875 | * same root key objectid, doing special handing for tree blocks owned | |
6876 | * by them is easy. Once a tree block has been COWed in one reloc tree, | |
6877 | * we can use the resulting new block directly when the same block is | |
6878 | * required to COW again through other reloc trees. By this way, relocated | |
6879 | * tree blocks are shared between reloc trees, so they are also shared | |
6880 | * between subvols. | |
1a40e23b | 6881 | */ |
d397712b | 6882 | static noinline int relocate_one_path(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6883 | struct btrfs_root *root, |
6884 | struct btrfs_path *path, | |
6885 | struct btrfs_key *first_key, | |
6886 | struct btrfs_ref_path *ref_path, | |
6887 | struct btrfs_block_group_cache *group, | |
6888 | struct inode *reloc_inode) | |
6889 | { | |
6890 | struct btrfs_root *reloc_root; | |
6891 | struct extent_buffer *eb = NULL; | |
6892 | struct btrfs_key *keys; | |
6893 | u64 *nodes; | |
6894 | int level; | |
f82d02d9 | 6895 | int shared_level; |
1a40e23b | 6896 | int lowest_level = 0; |
1a40e23b ZY |
6897 | int ret; |
6898 | ||
6899 | if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
6900 | lowest_level = ref_path->owner_objectid; | |
6901 | ||
f82d02d9 | 6902 | if (!root->ref_cows) { |
1a40e23b ZY |
6903 | path->lowest_level = lowest_level; |
6904 | ret = btrfs_search_slot(trans, root, first_key, path, 0, 1); | |
6905 | BUG_ON(ret < 0); | |
6906 | path->lowest_level = 0; | |
6907 | btrfs_release_path(root, path); | |
6908 | return 0; | |
6909 | } | |
6910 | ||
1a40e23b ZY |
6911 | mutex_lock(&root->fs_info->tree_reloc_mutex); |
6912 | ret = init_reloc_tree(trans, root); | |
6913 | BUG_ON(ret); | |
6914 | reloc_root = root->reloc_root; | |
6915 | ||
f82d02d9 YZ |
6916 | shared_level = ref_path->shared_level; |
6917 | ref_path->shared_level = BTRFS_MAX_LEVEL - 1; | |
1a40e23b | 6918 | |
f82d02d9 YZ |
6919 | keys = ref_path->node_keys; |
6920 | nodes = ref_path->new_nodes; | |
6921 | memset(&keys[shared_level + 1], 0, | |
6922 | sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
6923 | memset(&nodes[shared_level + 1], 0, | |
6924 | sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
1a40e23b | 6925 | |
f82d02d9 YZ |
6926 | if (nodes[lowest_level] == 0) { |
6927 | path->lowest_level = lowest_level; | |
6928 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, | |
6929 | 0, 1); | |
6930 | BUG_ON(ret); | |
6931 | for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) { | |
6932 | eb = path->nodes[level]; | |
6933 | if (!eb || eb == reloc_root->node) | |
6934 | break; | |
6935 | nodes[level] = eb->start; | |
6936 | if (level == 0) | |
6937 | btrfs_item_key_to_cpu(eb, &keys[level], 0); | |
6938 | else | |
6939 | btrfs_node_key_to_cpu(eb, &keys[level], 0); | |
6940 | } | |
2b82032c YZ |
6941 | if (nodes[0] && |
6942 | ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
6943 | eb = path->nodes[0]; |
6944 | ret = replace_extents_in_leaf(trans, reloc_root, eb, | |
6945 | group, reloc_inode); | |
6946 | BUG_ON(ret); | |
6947 | } | |
6948 | btrfs_release_path(reloc_root, path); | |
6949 | } else { | |
1a40e23b | 6950 | ret = btrfs_merge_path(trans, reloc_root, keys, nodes, |
f82d02d9 | 6951 | lowest_level); |
1a40e23b ZY |
6952 | BUG_ON(ret); |
6953 | } | |
6954 | ||
1a40e23b ZY |
6955 | /* |
6956 | * replace tree blocks in the fs tree with tree blocks in | |
6957 | * the reloc tree. | |
6958 | */ | |
6959 | ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level); | |
6960 | BUG_ON(ret < 0); | |
6961 | ||
6962 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
6963 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, |
6964 | 0, 0); | |
6965 | BUG_ON(ret); | |
6966 | extent_buffer_get(path->nodes[0]); | |
6967 | eb = path->nodes[0]; | |
6968 | btrfs_release_path(reloc_root, path); | |
1a40e23b ZY |
6969 | ret = invalidate_extent_cache(reloc_root, eb, group, root); |
6970 | BUG_ON(ret); | |
6971 | free_extent_buffer(eb); | |
6972 | } | |
1a40e23b | 6973 | |
f82d02d9 | 6974 | mutex_unlock(&root->fs_info->tree_reloc_mutex); |
1a40e23b | 6975 | path->lowest_level = 0; |
1a40e23b ZY |
6976 | return 0; |
6977 | } | |
6978 | ||
d397712b | 6979 | static noinline int relocate_tree_block(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6980 | struct btrfs_root *root, |
6981 | struct btrfs_path *path, | |
6982 | struct btrfs_key *first_key, | |
6983 | struct btrfs_ref_path *ref_path) | |
6984 | { | |
6985 | int ret; | |
1a40e23b ZY |
6986 | |
6987 | ret = relocate_one_path(trans, root, path, first_key, | |
6988 | ref_path, NULL, NULL); | |
6989 | BUG_ON(ret); | |
6990 | ||
1a40e23b ZY |
6991 | return 0; |
6992 | } | |
6993 | ||
d397712b | 6994 | static noinline int del_extent_zero(struct btrfs_trans_handle *trans, |
1a40e23b ZY |
6995 | struct btrfs_root *extent_root, |
6996 | struct btrfs_path *path, | |
6997 | struct btrfs_key *extent_key) | |
6998 | { | |
6999 | int ret; | |
7000 | ||
1a40e23b ZY |
7001 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); |
7002 | if (ret) | |
7003 | goto out; | |
7004 | ret = btrfs_del_item(trans, extent_root, path); | |
7005 | out: | |
7006 | btrfs_release_path(extent_root, path); | |
1a40e23b ZY |
7007 | return ret; |
7008 | } | |
7009 | ||
d397712b | 7010 | static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info, |
1a40e23b ZY |
7011 | struct btrfs_ref_path *ref_path) |
7012 | { | |
7013 | struct btrfs_key root_key; | |
7014 | ||
7015 | root_key.objectid = ref_path->root_objectid; | |
7016 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
7017 | if (is_cowonly_root(ref_path->root_objectid)) | |
7018 | root_key.offset = 0; | |
7019 | else | |
7020 | root_key.offset = (u64)-1; | |
7021 | ||
7022 | return btrfs_read_fs_root_no_name(fs_info, &root_key); | |
7023 | } | |
7024 | ||
d397712b | 7025 | static noinline int relocate_one_extent(struct btrfs_root *extent_root, |
1a40e23b ZY |
7026 | struct btrfs_path *path, |
7027 | struct btrfs_key *extent_key, | |
7028 | struct btrfs_block_group_cache *group, | |
7029 | struct inode *reloc_inode, int pass) | |
7030 | { | |
7031 | struct btrfs_trans_handle *trans; | |
7032 | struct btrfs_root *found_root; | |
7033 | struct btrfs_ref_path *ref_path = NULL; | |
7034 | struct disk_extent *new_extents = NULL; | |
7035 | int nr_extents = 0; | |
7036 | int loops; | |
7037 | int ret; | |
7038 | int level; | |
7039 | struct btrfs_key first_key; | |
7040 | u64 prev_block = 0; | |
7041 | ||
1a40e23b ZY |
7042 | |
7043 | trans = btrfs_start_transaction(extent_root, 1); | |
7044 | BUG_ON(!trans); | |
7045 | ||
7046 | if (extent_key->objectid == 0) { | |
7047 | ret = del_extent_zero(trans, extent_root, path, extent_key); | |
7048 | goto out; | |
7049 | } | |
7050 | ||
7051 | ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS); | |
7052 | if (!ref_path) { | |
d397712b CM |
7053 | ret = -ENOMEM; |
7054 | goto out; | |
1a40e23b ZY |
7055 | } |
7056 | ||
7057 | for (loops = 0; ; loops++) { | |
7058 | if (loops == 0) { | |
7059 | ret = btrfs_first_ref_path(trans, extent_root, ref_path, | |
7060 | extent_key->objectid); | |
7061 | } else { | |
7062 | ret = btrfs_next_ref_path(trans, extent_root, ref_path); | |
7063 | } | |
7064 | if (ret < 0) | |
7065 | goto out; | |
7066 | if (ret > 0) | |
7067 | break; | |
7068 | ||
7069 | if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID || | |
7070 | ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
7071 | continue; | |
7072 | ||
7073 | found_root = read_ref_root(extent_root->fs_info, ref_path); | |
7074 | BUG_ON(!found_root); | |
7075 | /* | |
7076 | * for reference counted tree, only process reference paths | |
7077 | * rooted at the latest committed root. | |
7078 | */ | |
7079 | if (found_root->ref_cows && | |
7080 | ref_path->root_generation != found_root->root_key.offset) | |
7081 | continue; | |
7082 | ||
7083 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
7084 | if (pass == 0) { | |
7085 | /* | |
7086 | * copy data extents to new locations | |
7087 | */ | |
7088 | u64 group_start = group->key.objectid; | |
7089 | ret = relocate_data_extent(reloc_inode, | |
7090 | extent_key, | |
7091 | group_start); | |
7092 | if (ret < 0) | |
7093 | goto out; | |
7094 | break; | |
7095 | } | |
7096 | level = 0; | |
7097 | } else { | |
7098 | level = ref_path->owner_objectid; | |
7099 | } | |
7100 | ||
7101 | if (prev_block != ref_path->nodes[level]) { | |
7102 | struct extent_buffer *eb; | |
7103 | u64 block_start = ref_path->nodes[level]; | |
7104 | u64 block_size = btrfs_level_size(found_root, level); | |
7105 | ||
7106 | eb = read_tree_block(found_root, block_start, | |
7107 | block_size, 0); | |
7108 | btrfs_tree_lock(eb); | |
7109 | BUG_ON(level != btrfs_header_level(eb)); | |
7110 | ||
7111 | if (level == 0) | |
7112 | btrfs_item_key_to_cpu(eb, &first_key, 0); | |
7113 | else | |
7114 | btrfs_node_key_to_cpu(eb, &first_key, 0); | |
7115 | ||
7116 | btrfs_tree_unlock(eb); | |
7117 | free_extent_buffer(eb); | |
7118 | prev_block = block_start; | |
7119 | } | |
7120 | ||
24562425 | 7121 | mutex_lock(&extent_root->fs_info->trans_mutex); |
e4404d6e | 7122 | btrfs_record_root_in_trans(found_root); |
24562425 | 7123 | mutex_unlock(&extent_root->fs_info->trans_mutex); |
e4404d6e YZ |
7124 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { |
7125 | /* | |
7126 | * try to update data extent references while | |
7127 | * keeping metadata shared between snapshots. | |
7128 | */ | |
7129 | if (pass == 1) { | |
7130 | ret = relocate_one_path(trans, found_root, | |
7131 | path, &first_key, ref_path, | |
7132 | group, reloc_inode); | |
7133 | if (ret < 0) | |
7134 | goto out; | |
7135 | continue; | |
7136 | } | |
1a40e23b ZY |
7137 | /* |
7138 | * use fallback method to process the remaining | |
7139 | * references. | |
7140 | */ | |
7141 | if (!new_extents) { | |
7142 | u64 group_start = group->key.objectid; | |
d899e052 YZ |
7143 | new_extents = kmalloc(sizeof(*new_extents), |
7144 | GFP_NOFS); | |
7145 | nr_extents = 1; | |
1a40e23b ZY |
7146 | ret = get_new_locations(reloc_inode, |
7147 | extent_key, | |
d899e052 | 7148 | group_start, 1, |
1a40e23b ZY |
7149 | &new_extents, |
7150 | &nr_extents); | |
d899e052 | 7151 | if (ret) |
1a40e23b ZY |
7152 | goto out; |
7153 | } | |
1a40e23b ZY |
7154 | ret = replace_one_extent(trans, found_root, |
7155 | path, extent_key, | |
7156 | &first_key, ref_path, | |
7157 | new_extents, nr_extents); | |
e4404d6e | 7158 | } else { |
1a40e23b ZY |
7159 | ret = relocate_tree_block(trans, found_root, path, |
7160 | &first_key, ref_path); | |
1a40e23b ZY |
7161 | } |
7162 | if (ret < 0) | |
7163 | goto out; | |
7164 | } | |
7165 | ret = 0; | |
7166 | out: | |
7167 | btrfs_end_transaction(trans, extent_root); | |
7168 | kfree(new_extents); | |
7169 | kfree(ref_path); | |
1a40e23b ZY |
7170 | return ret; |
7171 | } | |
5d4f98a2 | 7172 | #endif |
1a40e23b | 7173 | |
ec44a35c CM |
7174 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7175 | { | |
7176 | u64 num_devices; | |
7177 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
7178 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
7179 | ||
2b82032c | 7180 | num_devices = root->fs_info->fs_devices->rw_devices; |
ec44a35c CM |
7181 | if (num_devices == 1) { |
7182 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7183 | stripped = flags & ~stripped; | |
7184 | ||
7185 | /* turn raid0 into single device chunks */ | |
7186 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7187 | return stripped; | |
7188 | ||
7189 | /* turn mirroring into duplication */ | |
7190 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7191 | BTRFS_BLOCK_GROUP_RAID10)) | |
7192 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
7193 | return flags; | |
7194 | } else { | |
7195 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7196 | if (flags & stripped) |
7197 | return flags; | |
7198 | ||
7199 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7200 | stripped = flags & ~stripped; | |
7201 | ||
7202 | /* switch duplicated blocks with raid1 */ | |
7203 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7204 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7205 | ||
7206 | /* turn single device chunks into raid0 */ | |
7207 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
7208 | } | |
7209 | return flags; | |
7210 | } | |
7211 | ||
b2950863 | 7212 | static int __alloc_chunk_for_shrink(struct btrfs_root *root, |
0ef3e66b CM |
7213 | struct btrfs_block_group_cache *shrink_block_group, |
7214 | int force) | |
7215 | { | |
7216 | struct btrfs_trans_handle *trans; | |
7217 | u64 new_alloc_flags; | |
7218 | u64 calc; | |
7219 | ||
c286ac48 | 7220 | spin_lock(&shrink_block_group->lock); |
5d4f98a2 YZ |
7221 | if (btrfs_block_group_used(&shrink_block_group->item) + |
7222 | shrink_block_group->reserved > 0) { | |
c286ac48 | 7223 | spin_unlock(&shrink_block_group->lock); |
c286ac48 | 7224 | |
0ef3e66b | 7225 | trans = btrfs_start_transaction(root, 1); |
c286ac48 | 7226 | spin_lock(&shrink_block_group->lock); |
7d9eb12c | 7227 | |
0ef3e66b CM |
7228 | new_alloc_flags = update_block_group_flags(root, |
7229 | shrink_block_group->flags); | |
7230 | if (new_alloc_flags != shrink_block_group->flags) { | |
7231 | calc = | |
7232 | btrfs_block_group_used(&shrink_block_group->item); | |
7233 | } else { | |
7234 | calc = shrink_block_group->key.offset; | |
7235 | } | |
c286ac48 CM |
7236 | spin_unlock(&shrink_block_group->lock); |
7237 | ||
0ef3e66b CM |
7238 | do_chunk_alloc(trans, root->fs_info->extent_root, |
7239 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
7d9eb12c | 7240 | |
0ef3e66b | 7241 | btrfs_end_transaction(trans, root); |
c286ac48 CM |
7242 | } else |
7243 | spin_unlock(&shrink_block_group->lock); | |
0ef3e66b CM |
7244 | return 0; |
7245 | } | |
7246 | ||
5d4f98a2 YZ |
7247 | |
7248 | int btrfs_prepare_block_group_relocation(struct btrfs_root *root, | |
7249 | struct btrfs_block_group_cache *group) | |
7250 | ||
7251 | { | |
7252 | __alloc_chunk_for_shrink(root, group, 1); | |
7253 | set_block_group_readonly(group); | |
7254 | return 0; | |
7255 | } | |
7256 | ||
ba1bf481 JB |
7257 | /* |
7258 | * checks to see if its even possible to relocate this block group. | |
7259 | * | |
7260 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7261 | * ok to go ahead and try. | |
7262 | */ | |
7263 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7264 | { |
ba1bf481 JB |
7265 | struct btrfs_block_group_cache *block_group; |
7266 | struct btrfs_space_info *space_info; | |
7267 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7268 | struct btrfs_device *device; | |
7269 | int full = 0; | |
7270 | int ret = 0; | |
1a40e23b | 7271 | |
ba1bf481 | 7272 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7273 | |
ba1bf481 JB |
7274 | /* odd, couldn't find the block group, leave it alone */ |
7275 | if (!block_group) | |
7276 | return -1; | |
1a40e23b | 7277 | |
ba1bf481 JB |
7278 | /* no bytes used, we're good */ |
7279 | if (!btrfs_block_group_used(&block_group->item)) | |
1a40e23b ZY |
7280 | goto out; |
7281 | ||
ba1bf481 JB |
7282 | space_info = block_group->space_info; |
7283 | spin_lock(&space_info->lock); | |
17d217fe | 7284 | |
ba1bf481 | 7285 | full = space_info->full; |
17d217fe | 7286 | |
ba1bf481 JB |
7287 | /* |
7288 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7289 | * relocate it unless we're able to allocate a new chunk below. |
7290 | * | |
7291 | * Otherwise, we need to make sure we have room in the space to handle | |
7292 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7293 | */ |
7ce618db CM |
7294 | if ((space_info->total_bytes != block_group->key.offset) && |
7295 | (space_info->bytes_used + space_info->bytes_reserved + | |
ba1bf481 JB |
7296 | space_info->bytes_pinned + space_info->bytes_readonly + |
7297 | btrfs_block_group_used(&block_group->item) < | |
7ce618db | 7298 | space_info->total_bytes)) { |
ba1bf481 JB |
7299 | spin_unlock(&space_info->lock); |
7300 | goto out; | |
17d217fe | 7301 | } |
ba1bf481 | 7302 | spin_unlock(&space_info->lock); |
ea8c2819 | 7303 | |
ba1bf481 JB |
7304 | /* |
7305 | * ok we don't have enough space, but maybe we have free space on our | |
7306 | * devices to allocate new chunks for relocation, so loop through our | |
7307 | * alloc devices and guess if we have enough space. However, if we | |
7308 | * were marked as full, then we know there aren't enough chunks, and we | |
7309 | * can just return. | |
7310 | */ | |
7311 | ret = -1; | |
7312 | if (full) | |
7313 | goto out; | |
ea8c2819 | 7314 | |
ba1bf481 JB |
7315 | mutex_lock(&root->fs_info->chunk_mutex); |
7316 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7317 | u64 min_free = btrfs_block_group_used(&block_group->item); | |
7318 | u64 dev_offset, max_avail; | |
56bec294 | 7319 | |
ba1bf481 JB |
7320 | /* |
7321 | * check to make sure we can actually find a chunk with enough | |
7322 | * space to fit our block group in. | |
7323 | */ | |
7324 | if (device->total_bytes > device->bytes_used + min_free) { | |
7325 | ret = find_free_dev_extent(NULL, device, min_free, | |
7326 | &dev_offset, &max_avail); | |
7327 | if (!ret) | |
73e48b27 | 7328 | break; |
ba1bf481 | 7329 | ret = -1; |
725c8463 | 7330 | } |
edbd8d4e | 7331 | } |
ba1bf481 | 7332 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7333 | out: |
ba1bf481 | 7334 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7335 | return ret; |
7336 | } | |
7337 | ||
b2950863 CH |
7338 | static int find_first_block_group(struct btrfs_root *root, |
7339 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7340 | { |
925baedd | 7341 | int ret = 0; |
0b86a832 CM |
7342 | struct btrfs_key found_key; |
7343 | struct extent_buffer *leaf; | |
7344 | int slot; | |
edbd8d4e | 7345 | |
0b86a832 CM |
7346 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7347 | if (ret < 0) | |
925baedd CM |
7348 | goto out; |
7349 | ||
d397712b | 7350 | while (1) { |
0b86a832 | 7351 | slot = path->slots[0]; |
edbd8d4e | 7352 | leaf = path->nodes[0]; |
0b86a832 CM |
7353 | if (slot >= btrfs_header_nritems(leaf)) { |
7354 | ret = btrfs_next_leaf(root, path); | |
7355 | if (ret == 0) | |
7356 | continue; | |
7357 | if (ret < 0) | |
925baedd | 7358 | goto out; |
0b86a832 | 7359 | break; |
edbd8d4e | 7360 | } |
0b86a832 | 7361 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7362 | |
0b86a832 | 7363 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7364 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7365 | ret = 0; | |
7366 | goto out; | |
7367 | } | |
0b86a832 | 7368 | path->slots[0]++; |
edbd8d4e | 7369 | } |
0b86a832 | 7370 | ret = -ENOENT; |
925baedd | 7371 | out: |
0b86a832 | 7372 | return ret; |
edbd8d4e CM |
7373 | } |
7374 | ||
1a40e23b ZY |
7375 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7376 | { | |
7377 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7378 | struct btrfs_space_info *space_info; |
11833d66 | 7379 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7380 | struct rb_node *n; |
7381 | ||
11833d66 YZ |
7382 | down_write(&info->extent_commit_sem); |
7383 | while (!list_empty(&info->caching_block_groups)) { | |
7384 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7385 | struct btrfs_caching_control, list); | |
7386 | list_del(&caching_ctl->list); | |
7387 | put_caching_control(caching_ctl); | |
7388 | } | |
7389 | up_write(&info->extent_commit_sem); | |
7390 | ||
1a40e23b ZY |
7391 | spin_lock(&info->block_group_cache_lock); |
7392 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7393 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7394 | cache_node); | |
1a40e23b ZY |
7395 | rb_erase(&block_group->cache_node, |
7396 | &info->block_group_cache_tree); | |
d899e052 YZ |
7397 | spin_unlock(&info->block_group_cache_lock); |
7398 | ||
80eb234a | 7399 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7400 | list_del(&block_group->list); |
80eb234a | 7401 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7402 | |
817d52f8 | 7403 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7404 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
7405 | |
7406 | btrfs_remove_free_space_cache(block_group); | |
7407 | ||
d2fb3437 | 7408 | WARN_ON(atomic_read(&block_group->count) != 1); |
1a40e23b | 7409 | kfree(block_group); |
d899e052 YZ |
7410 | |
7411 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7412 | } |
7413 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7414 | |
7415 | /* now that all the block groups are freed, go through and | |
7416 | * free all the space_info structs. This is only called during | |
7417 | * the final stages of unmount, and so we know nobody is | |
7418 | * using them. We call synchronize_rcu() once before we start, | |
7419 | * just to be on the safe side. | |
7420 | */ | |
7421 | synchronize_rcu(); | |
7422 | ||
7423 | while(!list_empty(&info->space_info)) { | |
7424 | space_info = list_entry(info->space_info.next, | |
7425 | struct btrfs_space_info, | |
7426 | list); | |
7427 | ||
7428 | list_del(&space_info->list); | |
7429 | kfree(space_info); | |
7430 | } | |
1a40e23b ZY |
7431 | return 0; |
7432 | } | |
7433 | ||
9078a3e1 CM |
7434 | int btrfs_read_block_groups(struct btrfs_root *root) |
7435 | { | |
7436 | struct btrfs_path *path; | |
7437 | int ret; | |
9078a3e1 | 7438 | struct btrfs_block_group_cache *cache; |
be744175 | 7439 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7440 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7441 | struct btrfs_key key; |
7442 | struct btrfs_key found_key; | |
5f39d397 | 7443 | struct extent_buffer *leaf; |
96b5179d | 7444 | |
be744175 | 7445 | root = info->extent_root; |
9078a3e1 | 7446 | key.objectid = 0; |
0b86a832 | 7447 | key.offset = 0; |
9078a3e1 | 7448 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7449 | path = btrfs_alloc_path(); |
7450 | if (!path) | |
7451 | return -ENOMEM; | |
7452 | ||
d397712b | 7453 | while (1) { |
0b86a832 CM |
7454 | ret = find_first_block_group(root, path, &key); |
7455 | if (ret > 0) { | |
7456 | ret = 0; | |
7457 | goto error; | |
9078a3e1 | 7458 | } |
0b86a832 CM |
7459 | if (ret != 0) |
7460 | goto error; | |
7461 | ||
5f39d397 CM |
7462 | leaf = path->nodes[0]; |
7463 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7464 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7465 | if (!cache) { |
0b86a832 | 7466 | ret = -ENOMEM; |
9078a3e1 CM |
7467 | break; |
7468 | } | |
3e1ad54f | 7469 | |
d2fb3437 | 7470 | atomic_set(&cache->count, 1); |
c286ac48 | 7471 | spin_lock_init(&cache->lock); |
6226cb0a | 7472 | spin_lock_init(&cache->tree_lock); |
817d52f8 | 7473 | cache->fs_info = info; |
0f9dd46c | 7474 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7475 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 JB |
7476 | |
7477 | /* | |
7478 | * we only want to have 32k of ram per block group for keeping | |
7479 | * track of free space, and if we pass 1/2 of that we want to | |
7480 | * start converting things over to using bitmaps | |
7481 | */ | |
7482 | cache->extents_thresh = ((1024 * 32) / 2) / | |
7483 | sizeof(struct btrfs_free_space); | |
7484 | ||
5f39d397 CM |
7485 | read_extent_buffer(leaf, &cache->item, |
7486 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7487 | sizeof(cache->item)); | |
9078a3e1 | 7488 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7489 | |
9078a3e1 CM |
7490 | key.objectid = found_key.objectid + found_key.offset; |
7491 | btrfs_release_path(root, path); | |
0b86a832 | 7492 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7493 | cache->sectorsize = root->sectorsize; |
7494 | ||
817d52f8 JB |
7495 | /* |
7496 | * check for two cases, either we are full, and therefore | |
7497 | * don't need to bother with the caching work since we won't | |
7498 | * find any space, or we are empty, and we can just add all | |
7499 | * the space in and be done with it. This saves us _alot_ of | |
7500 | * time, particularly in the full case. | |
7501 | */ | |
7502 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
1b2da372 | 7503 | exclude_super_stripes(root, cache); |
11833d66 | 7504 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7505 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7506 | free_excluded_extents(root, cache); |
817d52f8 | 7507 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 YZ |
7508 | exclude_super_stripes(root, cache); |
7509 | cache->last_byte_to_unpin = (u64)-1; | |
817d52f8 JB |
7510 | cache->cached = BTRFS_CACHE_FINISHED; |
7511 | add_new_free_space(cache, root->fs_info, | |
7512 | found_key.objectid, | |
7513 | found_key.objectid + | |
7514 | found_key.offset); | |
11833d66 | 7515 | free_excluded_extents(root, cache); |
817d52f8 | 7516 | } |
96b5179d | 7517 | |
6324fbf3 CM |
7518 | ret = update_space_info(info, cache->flags, found_key.offset, |
7519 | btrfs_block_group_used(&cache->item), | |
7520 | &space_info); | |
7521 | BUG_ON(ret); | |
7522 | cache->space_info = space_info; | |
1b2da372 JB |
7523 | spin_lock(&cache->space_info->lock); |
7524 | cache->space_info->bytes_super += cache->bytes_super; | |
7525 | spin_unlock(&cache->space_info->lock); | |
7526 | ||
80eb234a JB |
7527 | down_write(&space_info->groups_sem); |
7528 | list_add_tail(&cache->list, &space_info->block_groups); | |
7529 | up_write(&space_info->groups_sem); | |
0f9dd46c JB |
7530 | |
7531 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
7532 | BUG_ON(ret); | |
75ccf47d CM |
7533 | |
7534 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c YZ |
7535 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
7536 | set_block_group_readonly(cache); | |
9078a3e1 | 7537 | } |
0b86a832 CM |
7538 | ret = 0; |
7539 | error: | |
9078a3e1 | 7540 | btrfs_free_path(path); |
0b86a832 | 7541 | return ret; |
9078a3e1 | 7542 | } |
6324fbf3 CM |
7543 | |
7544 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7545 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7546 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7547 | u64 size) |
7548 | { | |
7549 | int ret; | |
6324fbf3 CM |
7550 | struct btrfs_root *extent_root; |
7551 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7552 | |
7553 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7554 | |
12fcfd22 | 7555 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7556 | |
8f18cf13 | 7557 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7558 | if (!cache) |
7559 | return -ENOMEM; | |
7560 | ||
e17cade2 | 7561 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7562 | cache->key.offset = size; |
d2fb3437 | 7563 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 JB |
7564 | cache->sectorsize = root->sectorsize; |
7565 | ||
7566 | /* | |
7567 | * we only want to have 32k of ram per block group for keeping track | |
7568 | * of free space, and if we pass 1/2 of that we want to start | |
7569 | * converting things over to using bitmaps | |
7570 | */ | |
7571 | cache->extents_thresh = ((1024 * 32) / 2) / | |
7572 | sizeof(struct btrfs_free_space); | |
d2fb3437 | 7573 | atomic_set(&cache->count, 1); |
c286ac48 | 7574 | spin_lock_init(&cache->lock); |
6226cb0a | 7575 | spin_lock_init(&cache->tree_lock); |
0f9dd46c | 7576 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7577 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7578 | |
6324fbf3 | 7579 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7580 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7581 | cache->flags = type; | |
7582 | btrfs_set_block_group_flags(&cache->item, type); | |
7583 | ||
11833d66 | 7584 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7585 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7586 | exclude_super_stripes(root, cache); |
96303081 | 7587 | |
817d52f8 JB |
7588 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7589 | chunk_offset + size); | |
7590 | ||
11833d66 YZ |
7591 | free_excluded_extents(root, cache); |
7592 | ||
6324fbf3 CM |
7593 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7594 | &cache->space_info); | |
7595 | BUG_ON(ret); | |
1b2da372 JB |
7596 | |
7597 | spin_lock(&cache->space_info->lock); | |
7598 | cache->space_info->bytes_super += cache->bytes_super; | |
7599 | spin_unlock(&cache->space_info->lock); | |
7600 | ||
80eb234a JB |
7601 | down_write(&cache->space_info->groups_sem); |
7602 | list_add_tail(&cache->list, &cache->space_info->block_groups); | |
7603 | up_write(&cache->space_info->groups_sem); | |
6324fbf3 | 7604 | |
0f9dd46c JB |
7605 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
7606 | BUG_ON(ret); | |
c286ac48 | 7607 | |
6324fbf3 CM |
7608 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7609 | sizeof(cache->item)); | |
7610 | BUG_ON(ret); | |
7611 | ||
d18a2c44 | 7612 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7613 | |
6324fbf3 CM |
7614 | return 0; |
7615 | } | |
1a40e23b ZY |
7616 | |
7617 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
7618 | struct btrfs_root *root, u64 group_start) | |
7619 | { | |
7620 | struct btrfs_path *path; | |
7621 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7622 | struct btrfs_free_cluster *cluster; |
1a40e23b ZY |
7623 | struct btrfs_key key; |
7624 | int ret; | |
7625 | ||
1a40e23b ZY |
7626 | root = root->fs_info->extent_root; |
7627 | ||
7628 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7629 | BUG_ON(!block_group); | |
c146afad | 7630 | BUG_ON(!block_group->ro); |
1a40e23b ZY |
7631 | |
7632 | memcpy(&key, &block_group->key, sizeof(key)); | |
7633 | ||
44fb5511 CM |
7634 | /* make sure this block group isn't part of an allocation cluster */ |
7635 | cluster = &root->fs_info->data_alloc_cluster; | |
7636 | spin_lock(&cluster->refill_lock); | |
7637 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7638 | spin_unlock(&cluster->refill_lock); | |
7639 | ||
7640 | /* | |
7641 | * make sure this block group isn't part of a metadata | |
7642 | * allocation cluster | |
7643 | */ | |
7644 | cluster = &root->fs_info->meta_alloc_cluster; | |
7645 | spin_lock(&cluster->refill_lock); | |
7646 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7647 | spin_unlock(&cluster->refill_lock); | |
7648 | ||
1a40e23b ZY |
7649 | path = btrfs_alloc_path(); |
7650 | BUG_ON(!path); | |
7651 | ||
3dfdb934 | 7652 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
7653 | rb_erase(&block_group->cache_node, |
7654 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 7655 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 7656 | |
80eb234a | 7657 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
7658 | /* |
7659 | * we must use list_del_init so people can check to see if they | |
7660 | * are still on the list after taking the semaphore | |
7661 | */ | |
7662 | list_del_init(&block_group->list); | |
80eb234a | 7663 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 7664 | |
817d52f8 | 7665 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7666 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
7667 | |
7668 | btrfs_remove_free_space_cache(block_group); | |
7669 | ||
c146afad YZ |
7670 | spin_lock(&block_group->space_info->lock); |
7671 | block_group->space_info->total_bytes -= block_group->key.offset; | |
7672 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
7673 | spin_unlock(&block_group->space_info->lock); | |
283bb197 CM |
7674 | |
7675 | btrfs_clear_space_info_full(root->fs_info); | |
c146afad | 7676 | |
fa9c0d79 CM |
7677 | btrfs_put_block_group(block_group); |
7678 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
7679 | |
7680 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
7681 | if (ret > 0) | |
7682 | ret = -EIO; | |
7683 | if (ret < 0) | |
7684 | goto out; | |
7685 | ||
7686 | ret = btrfs_del_item(trans, root, path); | |
7687 | out: | |
7688 | btrfs_free_path(path); | |
7689 | return ret; | |
7690 | } |