]>
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> |
74493f7a | 22 | #include "hash.h" |
a5eb62e3 | 23 | #include "crc32c.h" |
fec577fb CM |
24 | #include "ctree.h" |
25 | #include "disk-io.h" | |
26 | #include "print-tree.h" | |
e089f05c | 27 | #include "transaction.h" |
0b86a832 | 28 | #include "volumes.h" |
925baedd | 29 | #include "locking.h" |
31153d81 | 30 | #include "ref-cache.h" |
fec577fb | 31 | |
31840ae1 ZY |
32 | #define PENDING_EXTENT_INSERT 0 |
33 | #define PENDING_EXTENT_DELETE 1 | |
34 | #define PENDING_BACKREF_UPDATE 2 | |
35 | ||
36 | struct pending_extent_op { | |
37 | int type; | |
38 | u64 bytenr; | |
39 | u64 num_bytes; | |
40 | u64 parent; | |
41 | u64 orig_parent; | |
42 | u64 generation; | |
43 | u64 orig_generation; | |
44 | int level; | |
f3465ca4 JB |
45 | struct list_head list; |
46 | int del; | |
31840ae1 ZY |
47 | }; |
48 | ||
e089f05c | 49 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
87ef2bb4 | 50 | btrfs_root *extent_root, int all); |
e20d96d6 | 51 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
87ef2bb4 | 52 | btrfs_root *extent_root, int all); |
925baedd CM |
53 | static struct btrfs_block_group_cache * |
54 | __btrfs_find_block_group(struct btrfs_root *root, | |
55 | struct btrfs_block_group_cache *hint, | |
56 | u64 search_start, int data, int owner); | |
f3465ca4 JB |
57 | static int pin_down_bytes(struct btrfs_trans_handle *trans, |
58 | struct btrfs_root *root, | |
59 | u64 bytenr, u64 num_bytes, int is_data); | |
60 | static int update_block_group(struct btrfs_trans_handle *trans, | |
61 | struct btrfs_root *root, | |
62 | u64 bytenr, u64 num_bytes, int alloc, | |
63 | int mark_free); | |
d548ee51 | 64 | |
0f9dd46c JB |
65 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
66 | { | |
67 | return (cache->flags & bits) == bits; | |
68 | } | |
69 | ||
70 | /* | |
71 | * this adds the block group to the fs_info rb tree for the block group | |
72 | * cache | |
73 | */ | |
74 | int btrfs_add_block_group_cache(struct btrfs_fs_info *info, | |
75 | struct btrfs_block_group_cache *block_group) | |
76 | { | |
77 | struct rb_node **p; | |
78 | struct rb_node *parent = NULL; | |
79 | struct btrfs_block_group_cache *cache; | |
80 | ||
81 | spin_lock(&info->block_group_cache_lock); | |
82 | p = &info->block_group_cache_tree.rb_node; | |
83 | ||
84 | while (*p) { | |
85 | parent = *p; | |
86 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
87 | cache_node); | |
88 | if (block_group->key.objectid < cache->key.objectid) { | |
89 | p = &(*p)->rb_left; | |
90 | } else if (block_group->key.objectid > cache->key.objectid) { | |
91 | p = &(*p)->rb_right; | |
92 | } else { | |
93 | spin_unlock(&info->block_group_cache_lock); | |
94 | return -EEXIST; | |
95 | } | |
96 | } | |
97 | ||
98 | rb_link_node(&block_group->cache_node, parent, p); | |
99 | rb_insert_color(&block_group->cache_node, | |
100 | &info->block_group_cache_tree); | |
101 | spin_unlock(&info->block_group_cache_lock); | |
102 | ||
103 | return 0; | |
104 | } | |
105 | ||
106 | /* | |
107 | * This will return the block group at or after bytenr if contains is 0, else | |
108 | * it will return the block group that contains the bytenr | |
109 | */ | |
110 | static struct btrfs_block_group_cache * | |
111 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
112 | int contains) | |
113 | { | |
114 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
115 | struct rb_node *n; | |
116 | u64 end, start; | |
117 | ||
118 | spin_lock(&info->block_group_cache_lock); | |
119 | n = info->block_group_cache_tree.rb_node; | |
120 | ||
121 | while (n) { | |
122 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
123 | cache_node); | |
124 | end = cache->key.objectid + cache->key.offset - 1; | |
125 | start = cache->key.objectid; | |
126 | ||
127 | if (bytenr < start) { | |
128 | if (!contains && (!ret || start < ret->key.objectid)) | |
129 | ret = cache; | |
130 | n = n->rb_left; | |
131 | } else if (bytenr > start) { | |
132 | if (contains && bytenr <= end) { | |
133 | ret = cache; | |
134 | break; | |
135 | } | |
136 | n = n->rb_right; | |
137 | } else { | |
138 | ret = cache; | |
139 | break; | |
140 | } | |
141 | } | |
142 | spin_unlock(&info->block_group_cache_lock); | |
143 | ||
144 | return ret; | |
145 | } | |
146 | ||
147 | /* | |
148 | * this is only called by cache_block_group, since we could have freed extents | |
149 | * we need to check the pinned_extents for any extents that can't be used yet | |
150 | * since their free space will be released as soon as the transaction commits. | |
151 | */ | |
152 | static int add_new_free_space(struct btrfs_block_group_cache *block_group, | |
153 | struct btrfs_fs_info *info, u64 start, u64 end) | |
154 | { | |
155 | u64 extent_start, extent_end, size; | |
156 | int ret; | |
157 | ||
25179201 | 158 | mutex_lock(&info->pinned_mutex); |
0f9dd46c JB |
159 | while (start < end) { |
160 | ret = find_first_extent_bit(&info->pinned_extents, start, | |
161 | &extent_start, &extent_end, | |
162 | EXTENT_DIRTY); | |
163 | if (ret) | |
164 | break; | |
165 | ||
166 | if (extent_start == start) { | |
167 | start = extent_end + 1; | |
168 | } else if (extent_start > start && extent_start < end) { | |
169 | size = extent_start - start; | |
25179201 JB |
170 | ret = btrfs_add_free_space_lock(block_group, start, |
171 | size); | |
0f9dd46c JB |
172 | BUG_ON(ret); |
173 | start = extent_end + 1; | |
174 | } else { | |
175 | break; | |
176 | } | |
177 | } | |
178 | ||
179 | if (start < end) { | |
180 | size = end - start; | |
25179201 | 181 | ret = btrfs_add_free_space_lock(block_group, start, size); |
0f9dd46c JB |
182 | BUG_ON(ret); |
183 | } | |
25179201 | 184 | mutex_unlock(&info->pinned_mutex); |
0f9dd46c JB |
185 | |
186 | return 0; | |
187 | } | |
188 | ||
e37c9e69 CM |
189 | static int cache_block_group(struct btrfs_root *root, |
190 | struct btrfs_block_group_cache *block_group) | |
191 | { | |
192 | struct btrfs_path *path; | |
ef8bbdfe | 193 | int ret = 0; |
e37c9e69 | 194 | struct btrfs_key key; |
5f39d397 | 195 | struct extent_buffer *leaf; |
e37c9e69 | 196 | int slot; |
e37c9e69 | 197 | u64 last = 0; |
7d7d6068 | 198 | u64 first_free; |
e37c9e69 CM |
199 | int found = 0; |
200 | ||
00f5c795 CM |
201 | if (!block_group) |
202 | return 0; | |
203 | ||
e37c9e69 | 204 | root = root->fs_info->extent_root; |
e37c9e69 CM |
205 | |
206 | if (block_group->cached) | |
207 | return 0; | |
f510cfec | 208 | |
e37c9e69 CM |
209 | path = btrfs_alloc_path(); |
210 | if (!path) | |
211 | return -ENOMEM; | |
7d7d6068 | 212 | |
2cc58cf2 | 213 | path->reada = 2; |
5cd57b2c CM |
214 | /* |
215 | * we get into deadlocks with paths held by callers of this function. | |
216 | * since the alloc_mutex is protecting things right now, just | |
217 | * skip the locking here | |
218 | */ | |
219 | path->skip_locking = 1; | |
0f9dd46c JB |
220 | first_free = max_t(u64, block_group->key.objectid, |
221 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
e37c9e69 | 222 | key.objectid = block_group->key.objectid; |
e37c9e69 CM |
223 | key.offset = 0; |
224 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
225 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
226 | if (ret < 0) | |
ef8bbdfe | 227 | goto err; |
0b86a832 | 228 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
d548ee51 | 229 | if (ret < 0) |
ef8bbdfe | 230 | goto err; |
d548ee51 Y |
231 | if (ret == 0) { |
232 | leaf = path->nodes[0]; | |
233 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
234 | if (key.objectid + key.offset > first_free) | |
235 | first_free = key.objectid + key.offset; | |
236 | } | |
e37c9e69 | 237 | while(1) { |
5f39d397 | 238 | leaf = path->nodes[0]; |
e37c9e69 | 239 | slot = path->slots[0]; |
5f39d397 | 240 | if (slot >= btrfs_header_nritems(leaf)) { |
e37c9e69 | 241 | ret = btrfs_next_leaf(root, path); |
54aa1f4d CM |
242 | if (ret < 0) |
243 | goto err; | |
0f9dd46c | 244 | if (ret == 0) |
e37c9e69 | 245 | continue; |
0f9dd46c | 246 | else |
e37c9e69 | 247 | break; |
e37c9e69 | 248 | } |
5f39d397 | 249 | btrfs_item_key_to_cpu(leaf, &key, slot); |
0f9dd46c | 250 | if (key.objectid < block_group->key.objectid) |
7d7d6068 | 251 | goto next; |
0f9dd46c | 252 | |
e37c9e69 | 253 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 254 | block_group->key.offset) |
e37c9e69 | 255 | break; |
7d7d6068 | 256 | |
e37c9e69 CM |
257 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
258 | if (!found) { | |
7d7d6068 | 259 | last = first_free; |
e37c9e69 | 260 | found = 1; |
e37c9e69 | 261 | } |
0f9dd46c JB |
262 | |
263 | add_new_free_space(block_group, root->fs_info, last, | |
264 | key.objectid); | |
265 | ||
7d7d6068 | 266 | last = key.objectid + key.offset; |
e37c9e69 | 267 | } |
7d7d6068 | 268 | next: |
e37c9e69 CM |
269 | path->slots[0]++; |
270 | } | |
271 | ||
7d7d6068 Y |
272 | if (!found) |
273 | last = first_free; | |
0f9dd46c JB |
274 | |
275 | add_new_free_space(block_group, root->fs_info, last, | |
276 | block_group->key.objectid + | |
277 | block_group->key.offset); | |
278 | ||
e37c9e69 | 279 | block_group->cached = 1; |
ef8bbdfe | 280 | ret = 0; |
54aa1f4d | 281 | err: |
e37c9e69 | 282 | btrfs_free_path(path); |
ef8bbdfe | 283 | return ret; |
e37c9e69 CM |
284 | } |
285 | ||
0f9dd46c JB |
286 | /* |
287 | * return the block group that starts at or after bytenr | |
288 | */ | |
0ef3e66b CM |
289 | struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct |
290 | btrfs_fs_info *info, | |
291 | u64 bytenr) | |
292 | { | |
0f9dd46c | 293 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 294 | |
0f9dd46c | 295 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 296 | |
0f9dd46c | 297 | return cache; |
0ef3e66b CM |
298 | } |
299 | ||
0f9dd46c JB |
300 | /* |
301 | * return the block group that contains teh given bytenr | |
302 | */ | |
5276aeda CM |
303 | struct btrfs_block_group_cache *btrfs_lookup_block_group(struct |
304 | btrfs_fs_info *info, | |
db94535d | 305 | u64 bytenr) |
be744175 | 306 | { |
0f9dd46c | 307 | struct btrfs_block_group_cache *cache; |
be744175 | 308 | |
0f9dd46c | 309 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 310 | |
0f9dd46c | 311 | return cache; |
be744175 | 312 | } |
0b86a832 | 313 | |
0f9dd46c JB |
314 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
315 | u64 flags) | |
6324fbf3 | 316 | { |
0f9dd46c JB |
317 | struct list_head *head = &info->space_info; |
318 | struct list_head *cur; | |
319 | struct btrfs_space_info *found; | |
320 | list_for_each(cur, head) { | |
321 | found = list_entry(cur, struct btrfs_space_info, list); | |
322 | if (found->flags == flags) | |
323 | return found; | |
324 | } | |
325 | return NULL; | |
6324fbf3 CM |
326 | } |
327 | ||
80eb234a JB |
328 | static u64 div_factor(u64 num, int factor) |
329 | { | |
330 | if (factor == 10) | |
331 | return num; | |
332 | num *= factor; | |
333 | do_div(num, 10); | |
334 | return num; | |
335 | } | |
336 | ||
925baedd CM |
337 | static struct btrfs_block_group_cache * |
338 | __btrfs_find_block_group(struct btrfs_root *root, | |
339 | struct btrfs_block_group_cache *hint, | |
340 | u64 search_start, int data, int owner) | |
cd1bc465 | 341 | { |
96b5179d | 342 | struct btrfs_block_group_cache *cache; |
31f3c99b | 343 | struct btrfs_block_group_cache *found_group = NULL; |
cd1bc465 CM |
344 | struct btrfs_fs_info *info = root->fs_info; |
345 | u64 used; | |
31f3c99b | 346 | u64 last = 0; |
96b5179d | 347 | u64 free_check; |
31f3c99b | 348 | int full_search = 0; |
bce4eae9 | 349 | int factor = 10; |
0ef3e66b | 350 | int wrapped = 0; |
de428b63 | 351 | |
a236aed1 CM |
352 | if (data & BTRFS_BLOCK_GROUP_METADATA) |
353 | factor = 9; | |
be744175 | 354 | |
0ef3e66b | 355 | if (search_start) { |
be744175 | 356 | struct btrfs_block_group_cache *shint; |
0ef3e66b | 357 | shint = btrfs_lookup_first_block_group(info, search_start); |
2b82032c | 358 | if (shint && block_group_bits(shint, data)) { |
c286ac48 | 359 | spin_lock(&shint->lock); |
be744175 | 360 | used = btrfs_block_group_used(&shint->item); |
e8569813 | 361 | if (used + shint->pinned + shint->reserved < |
324ae4df | 362 | div_factor(shint->key.offset, factor)) { |
c286ac48 | 363 | spin_unlock(&shint->lock); |
be744175 CM |
364 | return shint; |
365 | } | |
c286ac48 | 366 | spin_unlock(&shint->lock); |
be744175 CM |
367 | } |
368 | } | |
2b82032c | 369 | if (hint && block_group_bits(hint, data)) { |
c286ac48 | 370 | spin_lock(&hint->lock); |
31f3c99b | 371 | used = btrfs_block_group_used(&hint->item); |
e8569813 | 372 | if (used + hint->pinned + hint->reserved < |
324ae4df | 373 | div_factor(hint->key.offset, factor)) { |
c286ac48 | 374 | spin_unlock(&hint->lock); |
31f3c99b CM |
375 | return hint; |
376 | } | |
c286ac48 | 377 | spin_unlock(&hint->lock); |
e19caa5f | 378 | last = hint->key.objectid + hint->key.offset; |
31f3c99b | 379 | } else { |
e37c9e69 | 380 | if (hint) |
0ef3e66b | 381 | last = max(hint->key.objectid, search_start); |
e37c9e69 | 382 | else |
0ef3e66b | 383 | last = search_start; |
31f3c99b | 384 | } |
31f3c99b | 385 | again: |
e8569813 ZY |
386 | while (1) { |
387 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
388 | if (!cache) |
389 | break; | |
96b5179d | 390 | |
c286ac48 | 391 | spin_lock(&cache->lock); |
96b5179d CM |
392 | last = cache->key.objectid + cache->key.offset; |
393 | used = btrfs_block_group_used(&cache->item); | |
394 | ||
2b82032c | 395 | if (block_group_bits(cache, data)) { |
0ef3e66b | 396 | free_check = div_factor(cache->key.offset, factor); |
e8569813 ZY |
397 | if (used + cache->pinned + cache->reserved < |
398 | free_check) { | |
8790d502 | 399 | found_group = cache; |
c286ac48 | 400 | spin_unlock(&cache->lock); |
8790d502 CM |
401 | goto found; |
402 | } | |
6324fbf3 | 403 | } |
c286ac48 | 404 | spin_unlock(&cache->lock); |
de428b63 | 405 | cond_resched(); |
cd1bc465 | 406 | } |
0ef3e66b CM |
407 | if (!wrapped) { |
408 | last = search_start; | |
409 | wrapped = 1; | |
410 | goto again; | |
411 | } | |
412 | if (!full_search && factor < 10) { | |
be744175 | 413 | last = search_start; |
31f3c99b | 414 | full_search = 1; |
0ef3e66b | 415 | factor = 10; |
31f3c99b CM |
416 | goto again; |
417 | } | |
be744175 | 418 | found: |
31f3c99b | 419 | return found_group; |
cd1bc465 CM |
420 | } |
421 | ||
925baedd CM |
422 | struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, |
423 | struct btrfs_block_group_cache | |
424 | *hint, u64 search_start, | |
425 | int data, int owner) | |
426 | { | |
427 | ||
428 | struct btrfs_block_group_cache *ret; | |
925baedd | 429 | ret = __btrfs_find_block_group(root, hint, search_start, data, owner); |
925baedd CM |
430 | return ret; |
431 | } | |
0f9dd46c | 432 | |
e02119d5 | 433 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 434 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
435 | { |
436 | int ret; | |
437 | struct btrfs_key key; | |
31840ae1 | 438 | struct btrfs_path *path; |
e02119d5 | 439 | |
31840ae1 ZY |
440 | path = btrfs_alloc_path(); |
441 | BUG_ON(!path); | |
e02119d5 CM |
442 | key.objectid = start; |
443 | key.offset = len; | |
444 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
445 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
446 | 0, 0); | |
31840ae1 | 447 | btrfs_free_path(path); |
7bb86316 CM |
448 | return ret; |
449 | } | |
450 | ||
d8d5f3e1 CM |
451 | /* |
452 | * Back reference rules. Back refs have three main goals: | |
453 | * | |
454 | * 1) differentiate between all holders of references to an extent so that | |
455 | * when a reference is dropped we can make sure it was a valid reference | |
456 | * before freeing the extent. | |
457 | * | |
458 | * 2) Provide enough information to quickly find the holders of an extent | |
459 | * if we notice a given block is corrupted or bad. | |
460 | * | |
461 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
462 | * maintenance. This is actually the same as #2, but with a slightly | |
463 | * different use case. | |
464 | * | |
465 | * File extents can be referenced by: | |
466 | * | |
467 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 468 | * - different files inside a single subvolume |
d8d5f3e1 CM |
469 | * - different offsets inside a file (bookend extents in file.c) |
470 | * | |
471 | * The extent ref structure has fields for: | |
472 | * | |
473 | * - Objectid of the subvolume root | |
474 | * - Generation number of the tree holding the reference | |
475 | * - objectid of the file holding the reference | |
31840ae1 ZY |
476 | * - number of references holding by parent node (alway 1 for tree blocks) |
477 | * | |
478 | * Btree leaf may hold multiple references to a file extent. In most cases, | |
479 | * these references are from same file and the corresponding offsets inside | |
3bb1a1bc | 480 | * the file are close together. |
d8d5f3e1 CM |
481 | * |
482 | * When a file extent is allocated the fields are filled in: | |
3bb1a1bc | 483 | * (root_key.objectid, trans->transid, inode objectid, 1) |
d8d5f3e1 CM |
484 | * |
485 | * When a leaf is cow'd new references are added for every file extent found | |
31840ae1 ZY |
486 | * in the leaf. It looks similar to the create case, but trans->transid will |
487 | * be different when the block is cow'd. | |
d8d5f3e1 | 488 | * |
3bb1a1bc | 489 | * (root_key.objectid, trans->transid, inode objectid, |
31840ae1 | 490 | * number of references in the leaf) |
d8d5f3e1 | 491 | * |
3bb1a1bc YZ |
492 | * When a file extent is removed either during snapshot deletion or |
493 | * file truncation, we find the corresponding back reference and check | |
494 | * the following fields: | |
d8d5f3e1 | 495 | * |
3bb1a1bc YZ |
496 | * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), |
497 | * inode objectid) | |
d8d5f3e1 CM |
498 | * |
499 | * Btree extents can be referenced by: | |
500 | * | |
501 | * - Different subvolumes | |
502 | * - Different generations of the same subvolume | |
503 | * | |
d8d5f3e1 CM |
504 | * When a tree block is created, back references are inserted: |
505 | * | |
3bb1a1bc | 506 | * (root->root_key.objectid, trans->transid, level, 1) |
d8d5f3e1 | 507 | * |
31840ae1 ZY |
508 | * When a tree block is cow'd, new back references are added for all the |
509 | * blocks it points to. If the tree block isn't in reference counted root, | |
510 | * the old back references are removed. These new back references are of | |
511 | * the form (trans->transid will have increased since creation): | |
d8d5f3e1 | 512 | * |
3bb1a1bc | 513 | * (root->root_key.objectid, trans->transid, level, 1) |
d8d5f3e1 | 514 | * |
31840ae1 | 515 | * When a backref is in deleting, the following fields are checked: |
d8d5f3e1 CM |
516 | * |
517 | * if backref was for a tree root: | |
3bb1a1bc | 518 | * (btrfs_header_owner(itself), btrfs_header_generation(itself), level) |
d8d5f3e1 | 519 | * else |
3bb1a1bc | 520 | * (btrfs_header_owner(parent), btrfs_header_generation(parent), level) |
d8d5f3e1 | 521 | * |
31840ae1 | 522 | * Back Reference Key composing: |
d8d5f3e1 | 523 | * |
31840ae1 ZY |
524 | * The key objectid corresponds to the first byte in the extent, the key |
525 | * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first | |
526 | * byte of parent extent. If a extent is tree root, the key offset is set | |
527 | * to the key objectid. | |
d8d5f3e1 | 528 | */ |
31840ae1 ZY |
529 | |
530 | static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, | |
531 | struct btrfs_root *root, | |
3bb1a1bc YZ |
532 | struct btrfs_path *path, |
533 | u64 bytenr, u64 parent, | |
534 | u64 ref_root, u64 ref_generation, | |
535 | u64 owner_objectid, int del) | |
7bb86316 | 536 | { |
7bb86316 | 537 | struct btrfs_key key; |
31840ae1 ZY |
538 | struct btrfs_extent_ref *ref; |
539 | struct extent_buffer *leaf; | |
3bb1a1bc | 540 | u64 ref_objectid; |
74493f7a CM |
541 | int ret; |
542 | ||
31840ae1 ZY |
543 | key.objectid = bytenr; |
544 | key.type = BTRFS_EXTENT_REF_KEY; | |
545 | key.offset = parent; | |
546 | ||
547 | ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1); | |
548 | if (ret < 0) | |
549 | goto out; | |
550 | if (ret > 0) { | |
551 | ret = -ENOENT; | |
552 | goto out; | |
553 | } | |
554 | ||
555 | leaf = path->nodes[0]; | |
556 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); | |
3bb1a1bc | 557 | ref_objectid = btrfs_ref_objectid(leaf, ref); |
31840ae1 | 558 | if (btrfs_ref_root(leaf, ref) != ref_root || |
3bb1a1bc YZ |
559 | btrfs_ref_generation(leaf, ref) != ref_generation || |
560 | (ref_objectid != owner_objectid && | |
561 | ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) { | |
31840ae1 ZY |
562 | ret = -EIO; |
563 | WARN_ON(1); | |
564 | goto out; | |
565 | } | |
566 | ret = 0; | |
567 | out: | |
568 | return ret; | |
569 | } | |
570 | ||
f3465ca4 JB |
571 | /* |
572 | * updates all the backrefs that are pending on update_list for the | |
573 | * extent_root | |
574 | */ | |
575 | static int noinline update_backrefs(struct btrfs_trans_handle *trans, | |
576 | struct btrfs_root *extent_root, | |
577 | struct btrfs_path *path, | |
578 | struct list_head *update_list) | |
579 | { | |
580 | struct btrfs_key key; | |
581 | struct btrfs_extent_ref *ref; | |
582 | struct btrfs_fs_info *info = extent_root->fs_info; | |
583 | struct pending_extent_op *op; | |
584 | struct extent_buffer *leaf; | |
585 | int ret = 0; | |
586 | struct list_head *cur = update_list->next; | |
587 | u64 ref_objectid; | |
588 | u64 ref_root = extent_root->root_key.objectid; | |
589 | ||
590 | op = list_entry(cur, struct pending_extent_op, list); | |
591 | ||
592 | search: | |
593 | key.objectid = op->bytenr; | |
594 | key.type = BTRFS_EXTENT_REF_KEY; | |
595 | key.offset = op->orig_parent; | |
596 | ||
597 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1); | |
598 | BUG_ON(ret); | |
599 | ||
600 | leaf = path->nodes[0]; | |
601 | ||
602 | loop: | |
603 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); | |
604 | ||
605 | ref_objectid = btrfs_ref_objectid(leaf, ref); | |
606 | ||
607 | if (btrfs_ref_root(leaf, ref) != ref_root || | |
608 | btrfs_ref_generation(leaf, ref) != op->orig_generation || | |
609 | (ref_objectid != op->level && | |
610 | ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) { | |
611 | printk(KERN_ERR "couldn't find %Lu, parent %Lu, root %Lu, " | |
612 | "owner %u\n", op->bytenr, op->orig_parent, | |
613 | ref_root, op->level); | |
614 | btrfs_print_leaf(extent_root, leaf); | |
615 | BUG(); | |
616 | } | |
617 | ||
618 | key.objectid = op->bytenr; | |
619 | key.offset = op->parent; | |
620 | key.type = BTRFS_EXTENT_REF_KEY; | |
621 | ret = btrfs_set_item_key_safe(trans, extent_root, path, &key); | |
622 | BUG_ON(ret); | |
623 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); | |
624 | btrfs_set_ref_generation(leaf, ref, op->generation); | |
625 | ||
626 | cur = cur->next; | |
627 | ||
628 | list_del_init(&op->list); | |
629 | unlock_extent(&info->extent_ins, op->bytenr, | |
630 | op->bytenr + op->num_bytes - 1, GFP_NOFS); | |
631 | kfree(op); | |
632 | ||
633 | if (cur == update_list) { | |
634 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
635 | btrfs_release_path(extent_root, path); | |
636 | goto out; | |
637 | } | |
638 | ||
639 | op = list_entry(cur, struct pending_extent_op, list); | |
640 | ||
641 | path->slots[0]++; | |
642 | while (path->slots[0] < btrfs_header_nritems(leaf)) { | |
643 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
644 | if (key.objectid == op->bytenr && | |
645 | key.type == BTRFS_EXTENT_REF_KEY) | |
646 | goto loop; | |
647 | path->slots[0]++; | |
648 | } | |
649 | ||
650 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
651 | btrfs_release_path(extent_root, path); | |
652 | goto search; | |
653 | ||
654 | out: | |
655 | return 0; | |
656 | } | |
657 | ||
658 | static int noinline insert_extents(struct btrfs_trans_handle *trans, | |
659 | struct btrfs_root *extent_root, | |
660 | struct btrfs_path *path, | |
661 | struct list_head *insert_list, int nr) | |
662 | { | |
663 | struct btrfs_key *keys; | |
664 | u32 *data_size; | |
665 | struct pending_extent_op *op; | |
666 | struct extent_buffer *leaf; | |
667 | struct list_head *cur = insert_list->next; | |
668 | struct btrfs_fs_info *info = extent_root->fs_info; | |
669 | u64 ref_root = extent_root->root_key.objectid; | |
670 | int i = 0, last = 0, ret; | |
671 | int total = nr * 2; | |
672 | ||
673 | if (!nr) | |
674 | return 0; | |
675 | ||
676 | keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS); | |
677 | if (!keys) | |
678 | return -ENOMEM; | |
679 | ||
680 | data_size = kzalloc(total * sizeof(u32), GFP_NOFS); | |
681 | if (!data_size) { | |
682 | kfree(keys); | |
683 | return -ENOMEM; | |
684 | } | |
685 | ||
686 | list_for_each_entry(op, insert_list, list) { | |
687 | keys[i].objectid = op->bytenr; | |
688 | keys[i].offset = op->num_bytes; | |
689 | keys[i].type = BTRFS_EXTENT_ITEM_KEY; | |
690 | data_size[i] = sizeof(struct btrfs_extent_item); | |
691 | i++; | |
692 | ||
693 | keys[i].objectid = op->bytenr; | |
694 | keys[i].offset = op->parent; | |
695 | keys[i].type = BTRFS_EXTENT_REF_KEY; | |
696 | data_size[i] = sizeof(struct btrfs_extent_ref); | |
697 | i++; | |
698 | } | |
699 | ||
700 | op = list_entry(cur, struct pending_extent_op, list); | |
701 | i = 0; | |
702 | while (i < total) { | |
703 | int c; | |
704 | ret = btrfs_insert_some_items(trans, extent_root, path, | |
705 | keys+i, data_size+i, total-i); | |
706 | BUG_ON(ret < 0); | |
707 | ||
708 | if (last && ret > 1) | |
709 | BUG(); | |
710 | ||
711 | leaf = path->nodes[0]; | |
712 | for (c = 0; c < ret; c++) { | |
713 | int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY; | |
714 | ||
715 | /* | |
716 | * if the first item we inserted was a backref, then | |
717 | * the EXTENT_ITEM will be the odd c's, else it will | |
718 | * be the even c's | |
719 | */ | |
720 | if ((ref_first && (c % 2)) || | |
721 | (!ref_first && !(c % 2))) { | |
722 | struct btrfs_extent_item *itm; | |
723 | ||
724 | itm = btrfs_item_ptr(leaf, path->slots[0] + c, | |
725 | struct btrfs_extent_item); | |
726 | btrfs_set_extent_refs(path->nodes[0], itm, 1); | |
727 | op->del++; | |
728 | } else { | |
729 | struct btrfs_extent_ref *ref; | |
730 | ||
731 | ref = btrfs_item_ptr(leaf, path->slots[0] + c, | |
732 | struct btrfs_extent_ref); | |
733 | btrfs_set_ref_root(leaf, ref, ref_root); | |
734 | btrfs_set_ref_generation(leaf, ref, | |
735 | op->generation); | |
736 | btrfs_set_ref_objectid(leaf, ref, op->level); | |
737 | btrfs_set_ref_num_refs(leaf, ref, 1); | |
738 | op->del++; | |
739 | } | |
740 | ||
741 | /* | |
742 | * using del to see when its ok to free up the | |
743 | * pending_extent_op. In the case where we insert the | |
744 | * last item on the list in order to help do batching | |
745 | * we need to not free the extent op until we actually | |
746 | * insert the extent_item | |
747 | */ | |
748 | if (op->del == 2) { | |
749 | unlock_extent(&info->extent_ins, op->bytenr, | |
750 | op->bytenr + op->num_bytes - 1, | |
751 | GFP_NOFS); | |
752 | cur = cur->next; | |
753 | list_del_init(&op->list); | |
754 | kfree(op); | |
755 | if (cur != insert_list) | |
756 | op = list_entry(cur, | |
757 | struct pending_extent_op, | |
758 | list); | |
759 | } | |
760 | } | |
761 | btrfs_mark_buffer_dirty(leaf); | |
762 | btrfs_release_path(extent_root, path); | |
763 | ||
764 | /* | |
765 | * Ok backref's and items usually go right next to eachother, | |
766 | * but if we could only insert 1 item that means that we | |
767 | * inserted on the end of a leaf, and we have no idea what may | |
768 | * be on the next leaf so we just play it safe. In order to | |
769 | * try and help this case we insert the last thing on our | |
770 | * insert list so hopefully it will end up being the last | |
771 | * thing on the leaf and everything else will be before it, | |
772 | * which will let us insert a whole bunch of items at the same | |
773 | * time. | |
774 | */ | |
775 | if (ret == 1 && !last && (i + ret < total)) { | |
776 | /* | |
777 | * last: where we will pick up the next time around | |
778 | * i: our current key to insert, will be total - 1 | |
779 | * cur: the current op we are screwing with | |
780 | * op: duh | |
781 | */ | |
782 | last = i + ret; | |
783 | i = total - 1; | |
784 | cur = insert_list->prev; | |
785 | op = list_entry(cur, struct pending_extent_op, list); | |
786 | } else if (last) { | |
787 | /* | |
788 | * ok we successfully inserted the last item on the | |
789 | * list, lets reset everything | |
790 | * | |
791 | * i: our current key to insert, so where we left off | |
792 | * last time | |
793 | * last: done with this | |
794 | * cur: the op we are messing with | |
795 | * op: duh | |
796 | * total: since we inserted the last key, we need to | |
797 | * decrement total so we dont overflow | |
798 | */ | |
799 | i = last; | |
800 | last = 0; | |
f3465ca4 | 801 | total--; |
b4eec2ca LH |
802 | if (i < total) { |
803 | cur = insert_list->next; | |
804 | op = list_entry(cur, struct pending_extent_op, | |
805 | list); | |
806 | } | |
f3465ca4 JB |
807 | } else { |
808 | i += ret; | |
809 | } | |
810 | ||
811 | cond_resched(); | |
812 | } | |
813 | ret = 0; | |
814 | kfree(keys); | |
815 | kfree(data_size); | |
816 | return ret; | |
817 | } | |
818 | ||
31840ae1 ZY |
819 | static int noinline insert_extent_backref(struct btrfs_trans_handle *trans, |
820 | struct btrfs_root *root, | |
821 | struct btrfs_path *path, | |
822 | u64 bytenr, u64 parent, | |
823 | u64 ref_root, u64 ref_generation, | |
3bb1a1bc | 824 | u64 owner_objectid) |
31840ae1 ZY |
825 | { |
826 | struct btrfs_key key; | |
827 | struct extent_buffer *leaf; | |
828 | struct btrfs_extent_ref *ref; | |
829 | u32 num_refs; | |
830 | int ret; | |
74493f7a | 831 | |
74493f7a CM |
832 | key.objectid = bytenr; |
833 | key.type = BTRFS_EXTENT_REF_KEY; | |
31840ae1 | 834 | key.offset = parent; |
74493f7a | 835 | |
31840ae1 ZY |
836 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref)); |
837 | if (ret == 0) { | |
838 | leaf = path->nodes[0]; | |
839 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
840 | struct btrfs_extent_ref); | |
841 | btrfs_set_ref_root(leaf, ref, ref_root); | |
842 | btrfs_set_ref_generation(leaf, ref, ref_generation); | |
843 | btrfs_set_ref_objectid(leaf, ref, owner_objectid); | |
31840ae1 ZY |
844 | btrfs_set_ref_num_refs(leaf, ref, 1); |
845 | } else if (ret == -EEXIST) { | |
846 | u64 existing_owner; | |
847 | BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID); | |
848 | leaf = path->nodes[0]; | |
849 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
850 | struct btrfs_extent_ref); | |
851 | if (btrfs_ref_root(leaf, ref) != ref_root || | |
852 | btrfs_ref_generation(leaf, ref) != ref_generation) { | |
853 | ret = -EIO; | |
854 | WARN_ON(1); | |
7bb86316 | 855 | goto out; |
31840ae1 ZY |
856 | } |
857 | ||
858 | num_refs = btrfs_ref_num_refs(leaf, ref); | |
859 | BUG_ON(num_refs == 0); | |
860 | btrfs_set_ref_num_refs(leaf, ref, num_refs + 1); | |
861 | ||
862 | existing_owner = btrfs_ref_objectid(leaf, ref); | |
3bb1a1bc YZ |
863 | if (existing_owner != owner_objectid && |
864 | existing_owner != BTRFS_MULTIPLE_OBJECTIDS) { | |
31840ae1 ZY |
865 | btrfs_set_ref_objectid(leaf, ref, |
866 | BTRFS_MULTIPLE_OBJECTIDS); | |
31840ae1 ZY |
867 | } |
868 | ret = 0; | |
869 | } else { | |
7bb86316 | 870 | goto out; |
31840ae1 | 871 | } |
7bb86316 CM |
872 | btrfs_mark_buffer_dirty(path->nodes[0]); |
873 | out: | |
874 | btrfs_release_path(root, path); | |
875 | return ret; | |
74493f7a CM |
876 | } |
877 | ||
31840ae1 ZY |
878 | static int noinline remove_extent_backref(struct btrfs_trans_handle *trans, |
879 | struct btrfs_root *root, | |
880 | struct btrfs_path *path) | |
881 | { | |
882 | struct extent_buffer *leaf; | |
883 | struct btrfs_extent_ref *ref; | |
884 | u32 num_refs; | |
885 | int ret = 0; | |
886 | ||
887 | leaf = path->nodes[0]; | |
888 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); | |
889 | num_refs = btrfs_ref_num_refs(leaf, ref); | |
890 | BUG_ON(num_refs == 0); | |
891 | num_refs -= 1; | |
892 | if (num_refs == 0) { | |
893 | ret = btrfs_del_item(trans, root, path); | |
894 | } else { | |
895 | btrfs_set_ref_num_refs(leaf, ref, num_refs); | |
896 | btrfs_mark_buffer_dirty(leaf); | |
897 | } | |
898 | btrfs_release_path(root, path); | |
899 | return ret; | |
900 | } | |
901 | ||
f3465ca4 JB |
902 | static int noinline free_extents(struct btrfs_trans_handle *trans, |
903 | struct btrfs_root *extent_root, | |
904 | struct list_head *del_list) | |
905 | { | |
906 | struct btrfs_fs_info *info = extent_root->fs_info; | |
907 | struct btrfs_path *path; | |
908 | struct btrfs_key key, found_key; | |
909 | struct extent_buffer *leaf; | |
910 | struct list_head *cur; | |
911 | struct pending_extent_op *op; | |
912 | struct btrfs_extent_item *ei; | |
913 | int ret, num_to_del, extent_slot = 0, found_extent = 0; | |
914 | u32 refs; | |
915 | u64 bytes_freed = 0; | |
916 | ||
917 | path = btrfs_alloc_path(); | |
918 | if (!path) | |
919 | return -ENOMEM; | |
920 | path->reada = 1; | |
921 | ||
922 | search: | |
923 | /* search for the backref for the current ref we want to delete */ | |
924 | cur = del_list->next; | |
925 | op = list_entry(cur, struct pending_extent_op, list); | |
926 | ret = lookup_extent_backref(trans, extent_root, path, op->bytenr, | |
927 | op->orig_parent, | |
928 | extent_root->root_key.objectid, | |
929 | op->orig_generation, op->level, 1); | |
930 | if (ret) { | |
931 | printk("Unable to find backref byte nr %Lu root %Lu gen %Lu " | |
932 | "owner %u\n", op->bytenr, | |
933 | extent_root->root_key.objectid, op->orig_generation, | |
934 | op->level); | |
935 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
936 | WARN_ON(1); | |
937 | goto out; | |
938 | } | |
939 | ||
940 | extent_slot = path->slots[0]; | |
941 | num_to_del = 1; | |
942 | found_extent = 0; | |
943 | ||
944 | /* | |
945 | * if we aren't the first item on the leaf we can move back one and see | |
946 | * if our ref is right next to our extent item | |
947 | */ | |
948 | if (likely(extent_slot)) { | |
949 | extent_slot--; | |
950 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
951 | extent_slot); | |
952 | if (found_key.objectid == op->bytenr && | |
953 | found_key.type == BTRFS_EXTENT_ITEM_KEY && | |
954 | found_key.offset == op->num_bytes) { | |
955 | num_to_del++; | |
956 | found_extent = 1; | |
957 | } | |
958 | } | |
959 | ||
960 | /* | |
961 | * if we didn't find the extent we need to delete the backref and then | |
962 | * search for the extent item key so we can update its ref count | |
963 | */ | |
964 | if (!found_extent) { | |
965 | key.objectid = op->bytenr; | |
966 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
967 | key.offset = op->num_bytes; | |
968 | ||
969 | ret = remove_extent_backref(trans, extent_root, path); | |
970 | BUG_ON(ret); | |
971 | btrfs_release_path(extent_root, path); | |
972 | ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); | |
973 | BUG_ON(ret); | |
974 | extent_slot = path->slots[0]; | |
975 | } | |
976 | ||
977 | /* this is where we update the ref count for the extent */ | |
978 | leaf = path->nodes[0]; | |
979 | ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item); | |
980 | refs = btrfs_extent_refs(leaf, ei); | |
981 | BUG_ON(refs == 0); | |
982 | refs--; | |
983 | btrfs_set_extent_refs(leaf, ei, refs); | |
984 | ||
985 | btrfs_mark_buffer_dirty(leaf); | |
986 | ||
987 | /* | |
988 | * This extent needs deleting. The reason cur_slot is extent_slot + | |
989 | * num_to_del is because extent_slot points to the slot where the extent | |
990 | * is, and if the backref was not right next to the extent we will be | |
991 | * deleting at least 1 item, and will want to start searching at the | |
992 | * slot directly next to extent_slot. However if we did find the | |
993 | * backref next to the extent item them we will be deleting at least 2 | |
994 | * items and will want to start searching directly after the ref slot | |
995 | */ | |
996 | if (!refs) { | |
997 | struct list_head *pos, *n, *end; | |
998 | int cur_slot = extent_slot+num_to_del; | |
999 | u64 super_used; | |
1000 | u64 root_used; | |
1001 | ||
1002 | path->slots[0] = extent_slot; | |
1003 | bytes_freed = op->num_bytes; | |
1004 | ||
e3e469f8 JB |
1005 | mutex_lock(&info->pinned_mutex); |
1006 | ret = pin_down_bytes(trans, extent_root, op->bytenr, | |
1007 | op->num_bytes, op->level >= | |
1008 | BTRFS_FIRST_FREE_OBJECTID); | |
1009 | mutex_unlock(&info->pinned_mutex); | |
1010 | BUG_ON(ret < 0); | |
1011 | op->del = ret; | |
1012 | ||
f3465ca4 JB |
1013 | /* |
1014 | * we need to see if we can delete multiple things at once, so | |
1015 | * start looping through the list of extents we are wanting to | |
1016 | * delete and see if their extent/backref's are right next to | |
1017 | * eachother and the extents only have 1 ref | |
1018 | */ | |
1019 | for (pos = cur->next; pos != del_list; pos = pos->next) { | |
1020 | struct pending_extent_op *tmp; | |
1021 | ||
1022 | tmp = list_entry(pos, struct pending_extent_op, list); | |
1023 | ||
1024 | /* we only want to delete extent+ref at this stage */ | |
1025 | if (cur_slot >= btrfs_header_nritems(leaf) - 1) | |
1026 | break; | |
1027 | ||
1028 | btrfs_item_key_to_cpu(leaf, &found_key, cur_slot); | |
1029 | if (found_key.objectid != tmp->bytenr || | |
1030 | found_key.type != BTRFS_EXTENT_ITEM_KEY || | |
1031 | found_key.offset != tmp->num_bytes) | |
1032 | break; | |
1033 | ||
1034 | /* check to make sure this extent only has one ref */ | |
1035 | ei = btrfs_item_ptr(leaf, cur_slot, | |
1036 | struct btrfs_extent_item); | |
1037 | if (btrfs_extent_refs(leaf, ei) != 1) | |
1038 | break; | |
1039 | ||
1040 | btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1); | |
1041 | if (found_key.objectid != tmp->bytenr || | |
1042 | found_key.type != BTRFS_EXTENT_REF_KEY || | |
1043 | found_key.offset != tmp->orig_parent) | |
1044 | break; | |
1045 | ||
1046 | /* | |
1047 | * the ref is right next to the extent, we can set the | |
1048 | * ref count to 0 since we will delete them both now | |
1049 | */ | |
1050 | btrfs_set_extent_refs(leaf, ei, 0); | |
1051 | ||
1052 | /* pin down the bytes for this extent */ | |
1053 | mutex_lock(&info->pinned_mutex); | |
1054 | ret = pin_down_bytes(trans, extent_root, tmp->bytenr, | |
1055 | tmp->num_bytes, tmp->level >= | |
1056 | BTRFS_FIRST_FREE_OBJECTID); | |
1057 | mutex_unlock(&info->pinned_mutex); | |
1058 | BUG_ON(ret < 0); | |
1059 | ||
1060 | /* | |
1061 | * use the del field to tell if we need to go ahead and | |
1062 | * free up the extent when we delete the item or not. | |
1063 | */ | |
1064 | tmp->del = ret; | |
1065 | bytes_freed += tmp->num_bytes; | |
1066 | ||
1067 | num_to_del += 2; | |
1068 | cur_slot += 2; | |
1069 | } | |
1070 | end = pos; | |
1071 | ||
1072 | /* update the free space counters */ | |
1073 | spin_lock_irq(&info->delalloc_lock); | |
1074 | super_used = btrfs_super_bytes_used(&info->super_copy); | |
1075 | btrfs_set_super_bytes_used(&info->super_copy, | |
1076 | super_used - bytes_freed); | |
1077 | spin_unlock_irq(&info->delalloc_lock); | |
1078 | ||
1079 | root_used = btrfs_root_used(&extent_root->root_item); | |
1080 | btrfs_set_root_used(&extent_root->root_item, | |
1081 | root_used - bytes_freed); | |
1082 | ||
1083 | /* delete the items */ | |
1084 | ret = btrfs_del_items(trans, extent_root, path, | |
1085 | path->slots[0], num_to_del); | |
1086 | BUG_ON(ret); | |
1087 | ||
1088 | /* | |
1089 | * loop through the extents we deleted and do the cleanup work | |
1090 | * on them | |
1091 | */ | |
1092 | for (pos = cur, n = pos->next; pos != end; | |
1093 | pos = n, n = pos->next) { | |
1094 | struct pending_extent_op *tmp; | |
1095 | #ifdef BIO_RW_DISCARD | |
1096 | u64 map_length; | |
1097 | struct btrfs_multi_bio *multi = NULL; | |
1098 | #endif | |
1099 | tmp = list_entry(pos, struct pending_extent_op, list); | |
1100 | ||
1101 | /* | |
1102 | * remember tmp->del tells us wether or not we pinned | |
1103 | * down the extent | |
1104 | */ | |
1105 | ret = update_block_group(trans, extent_root, | |
1106 | tmp->bytenr, tmp->num_bytes, 0, | |
1107 | tmp->del); | |
1108 | BUG_ON(ret); | |
1109 | ||
1110 | #ifdef BIO_RW_DISCARD | |
1111 | ret = btrfs_map_block(&info->mapping_tree, READ, | |
1112 | tmp->bytenr, &map_length, &multi, | |
1113 | 0); | |
1114 | if (!ret) { | |
1115 | struct btrfs_bio_stripe *stripe; | |
1116 | int i; | |
1117 | ||
1118 | stripe = multi->stripe; | |
1119 | ||
1120 | if (map_length > tmp->num_bytes) | |
1121 | map_length = tmp->num_bytes; | |
1122 | ||
1123 | for (i = 0; i < multi->num_stripes; | |
1124 | i++, stripe++) | |
1125 | blkdev_issue_discard(stripe->dev->bdev, | |
1126 | stripe->physical >> 9, | |
1127 | map_length >> 9); | |
1128 | kfree(multi); | |
1129 | } | |
1130 | #endif | |
1131 | list_del_init(&tmp->list); | |
1132 | unlock_extent(&info->extent_ins, tmp->bytenr, | |
1133 | tmp->bytenr + tmp->num_bytes - 1, | |
1134 | GFP_NOFS); | |
1135 | kfree(tmp); | |
1136 | } | |
1137 | } else if (refs && found_extent) { | |
1138 | /* | |
1139 | * the ref and extent were right next to eachother, but the | |
1140 | * extent still has a ref, so just free the backref and keep | |
1141 | * going | |
1142 | */ | |
1143 | ret = remove_extent_backref(trans, extent_root, path); | |
1144 | BUG_ON(ret); | |
1145 | ||
1146 | list_del_init(&op->list); | |
1147 | unlock_extent(&info->extent_ins, op->bytenr, | |
1148 | op->bytenr + op->num_bytes - 1, GFP_NOFS); | |
1149 | kfree(op); | |
1150 | } else { | |
1151 | /* | |
1152 | * the extent has multiple refs and the backref we were looking | |
1153 | * for was not right next to it, so just unlock and go next, | |
1154 | * we're good to go | |
1155 | */ | |
1156 | list_del_init(&op->list); | |
1157 | unlock_extent(&info->extent_ins, op->bytenr, | |
1158 | op->bytenr + op->num_bytes - 1, GFP_NOFS); | |
1159 | kfree(op); | |
1160 | } | |
1161 | ||
1162 | btrfs_release_path(extent_root, path); | |
1163 | if (!list_empty(del_list)) | |
1164 | goto search; | |
1165 | ||
1166 | out: | |
1167 | btrfs_free_path(path); | |
1168 | return ret; | |
1169 | } | |
1170 | ||
31840ae1 ZY |
1171 | static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, |
1172 | struct btrfs_root *root, u64 bytenr, | |
1173 | u64 orig_parent, u64 parent, | |
1174 | u64 orig_root, u64 ref_root, | |
1175 | u64 orig_generation, u64 ref_generation, | |
3bb1a1bc | 1176 | u64 owner_objectid) |
31840ae1 ZY |
1177 | { |
1178 | int ret; | |
1179 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
1180 | struct btrfs_path *path; | |
1181 | ||
1182 | if (root == root->fs_info->extent_root) { | |
1183 | struct pending_extent_op *extent_op; | |
1184 | u64 num_bytes; | |
1185 | ||
1186 | BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL); | |
1187 | num_bytes = btrfs_level_size(root, (int)owner_objectid); | |
25179201 | 1188 | mutex_lock(&root->fs_info->extent_ins_mutex); |
31840ae1 | 1189 | if (test_range_bit(&root->fs_info->extent_ins, bytenr, |
25179201 | 1190 | bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) { |
31840ae1 ZY |
1191 | u64 priv; |
1192 | ret = get_state_private(&root->fs_info->extent_ins, | |
1193 | bytenr, &priv); | |
1194 | BUG_ON(ret); | |
1195 | extent_op = (struct pending_extent_op *) | |
1196 | (unsigned long)priv; | |
1197 | BUG_ON(extent_op->parent != orig_parent); | |
1198 | BUG_ON(extent_op->generation != orig_generation); | |
25179201 | 1199 | |
31840ae1 ZY |
1200 | extent_op->parent = parent; |
1201 | extent_op->generation = ref_generation; | |
1202 | } else { | |
1203 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
1204 | BUG_ON(!extent_op); | |
1205 | ||
1206 | extent_op->type = PENDING_BACKREF_UPDATE; | |
1207 | extent_op->bytenr = bytenr; | |
1208 | extent_op->num_bytes = num_bytes; | |
1209 | extent_op->parent = parent; | |
1210 | extent_op->orig_parent = orig_parent; | |
1211 | extent_op->generation = ref_generation; | |
1212 | extent_op->orig_generation = orig_generation; | |
1213 | extent_op->level = (int)owner_objectid; | |
f3465ca4 JB |
1214 | INIT_LIST_HEAD(&extent_op->list); |
1215 | extent_op->del = 0; | |
31840ae1 ZY |
1216 | |
1217 | set_extent_bits(&root->fs_info->extent_ins, | |
1218 | bytenr, bytenr + num_bytes - 1, | |
25179201 | 1219 | EXTENT_WRITEBACK, GFP_NOFS); |
31840ae1 ZY |
1220 | set_state_private(&root->fs_info->extent_ins, |
1221 | bytenr, (unsigned long)extent_op); | |
1222 | } | |
25179201 | 1223 | mutex_unlock(&root->fs_info->extent_ins_mutex); |
31840ae1 ZY |
1224 | return 0; |
1225 | } | |
1226 | ||
1227 | path = btrfs_alloc_path(); | |
1228 | if (!path) | |
1229 | return -ENOMEM; | |
1230 | ret = lookup_extent_backref(trans, extent_root, path, | |
1231 | bytenr, orig_parent, orig_root, | |
3bb1a1bc | 1232 | orig_generation, owner_objectid, 1); |
31840ae1 ZY |
1233 | if (ret) |
1234 | goto out; | |
1235 | ret = remove_extent_backref(trans, extent_root, path); | |
1236 | if (ret) | |
1237 | goto out; | |
1238 | ret = insert_extent_backref(trans, extent_root, path, bytenr, | |
1239 | parent, ref_root, ref_generation, | |
3bb1a1bc | 1240 | owner_objectid); |
31840ae1 | 1241 | BUG_ON(ret); |
87ef2bb4 CM |
1242 | finish_current_insert(trans, extent_root, 0); |
1243 | del_pending_extents(trans, extent_root, 0); | |
31840ae1 ZY |
1244 | out: |
1245 | btrfs_free_path(path); | |
1246 | return ret; | |
1247 | } | |
1248 | ||
1249 | int btrfs_update_extent_ref(struct btrfs_trans_handle *trans, | |
1250 | struct btrfs_root *root, u64 bytenr, | |
1251 | u64 orig_parent, u64 parent, | |
1252 | u64 ref_root, u64 ref_generation, | |
3bb1a1bc | 1253 | u64 owner_objectid) |
31840ae1 ZY |
1254 | { |
1255 | int ret; | |
1256 | if (ref_root == BTRFS_TREE_LOG_OBJECTID && | |
1257 | owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
1258 | return 0; | |
31840ae1 ZY |
1259 | ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent, |
1260 | parent, ref_root, ref_root, | |
1261 | ref_generation, ref_generation, | |
3bb1a1bc | 1262 | owner_objectid); |
31840ae1 ZY |
1263 | return ret; |
1264 | } | |
1265 | ||
925baedd | 1266 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
1267 | struct btrfs_root *root, u64 bytenr, |
1268 | u64 orig_parent, u64 parent, | |
1269 | u64 orig_root, u64 ref_root, | |
1270 | u64 orig_generation, u64 ref_generation, | |
3bb1a1bc | 1271 | u64 owner_objectid) |
02217ed2 | 1272 | { |
5caf2a00 | 1273 | struct btrfs_path *path; |
02217ed2 | 1274 | int ret; |
e2fa7227 | 1275 | struct btrfs_key key; |
5f39d397 | 1276 | struct extent_buffer *l; |
234b63a0 | 1277 | struct btrfs_extent_item *item; |
cf27e1ee | 1278 | u32 refs; |
037e6390 | 1279 | |
5caf2a00 | 1280 | path = btrfs_alloc_path(); |
54aa1f4d CM |
1281 | if (!path) |
1282 | return -ENOMEM; | |
26b8003f | 1283 | |
3c12ac72 | 1284 | path->reada = 1; |
db94535d | 1285 | key.objectid = bytenr; |
31840ae1 ZY |
1286 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1287 | key.offset = (u64)-1; | |
1288 | ||
5caf2a00 | 1289 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 1290 | 0, 1); |
54aa1f4d CM |
1291 | if (ret < 0) |
1292 | return ret; | |
31840ae1 ZY |
1293 | BUG_ON(ret == 0 || path->slots[0] == 0); |
1294 | ||
1295 | path->slots[0]--; | |
5f39d397 | 1296 | l = path->nodes[0]; |
31840ae1 ZY |
1297 | |
1298 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
771ed689 CM |
1299 | if (key.objectid != bytenr) { |
1300 | btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]); | |
1301 | printk("wanted %Lu found %Lu\n", bytenr, key.objectid); | |
1302 | BUG(); | |
1303 | } | |
31840ae1 ZY |
1304 | BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY); |
1305 | ||
5caf2a00 | 1306 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 CM |
1307 | refs = btrfs_extent_refs(l, item); |
1308 | btrfs_set_extent_refs(l, item, refs + 1); | |
5caf2a00 | 1309 | btrfs_mark_buffer_dirty(path->nodes[0]); |
a28ec197 | 1310 | |
5caf2a00 | 1311 | btrfs_release_path(root->fs_info->extent_root, path); |
7bb86316 | 1312 | |
3c12ac72 | 1313 | path->reada = 1; |
31840ae1 ZY |
1314 | ret = insert_extent_backref(trans, root->fs_info->extent_root, |
1315 | path, bytenr, parent, | |
1316 | ref_root, ref_generation, | |
3bb1a1bc | 1317 | owner_objectid); |
7bb86316 | 1318 | BUG_ON(ret); |
87ef2bb4 CM |
1319 | finish_current_insert(trans, root->fs_info->extent_root, 0); |
1320 | del_pending_extents(trans, root->fs_info->extent_root, 0); | |
74493f7a CM |
1321 | |
1322 | btrfs_free_path(path); | |
02217ed2 CM |
1323 | return 0; |
1324 | } | |
1325 | ||
925baedd | 1326 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
1327 | struct btrfs_root *root, |
1328 | u64 bytenr, u64 num_bytes, u64 parent, | |
1329 | u64 ref_root, u64 ref_generation, | |
3bb1a1bc | 1330 | u64 owner_objectid) |
925baedd CM |
1331 | { |
1332 | int ret; | |
31840ae1 ZY |
1333 | if (ref_root == BTRFS_TREE_LOG_OBJECTID && |
1334 | owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
1335 | return 0; | |
31840ae1 ZY |
1336 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent, |
1337 | 0, ref_root, 0, ref_generation, | |
3bb1a1bc | 1338 | owner_objectid); |
925baedd CM |
1339 | return ret; |
1340 | } | |
1341 | ||
e9d0b13b CM |
1342 | int btrfs_extent_post_op(struct btrfs_trans_handle *trans, |
1343 | struct btrfs_root *root) | |
1344 | { | |
87ef2bb4 CM |
1345 | finish_current_insert(trans, root->fs_info->extent_root, 1); |
1346 | del_pending_extents(trans, root->fs_info->extent_root, 1); | |
e9d0b13b CM |
1347 | return 0; |
1348 | } | |
1349 | ||
31840ae1 ZY |
1350 | int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans, |
1351 | struct btrfs_root *root, u64 bytenr, | |
1352 | u64 num_bytes, u32 *refs) | |
a28ec197 | 1353 | { |
5caf2a00 | 1354 | struct btrfs_path *path; |
a28ec197 | 1355 | int ret; |
e2fa7227 | 1356 | struct btrfs_key key; |
5f39d397 | 1357 | struct extent_buffer *l; |
234b63a0 | 1358 | struct btrfs_extent_item *item; |
5caf2a00 | 1359 | |
db94535d | 1360 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 1361 | path = btrfs_alloc_path(); |
3c12ac72 | 1362 | path->reada = 1; |
db94535d CM |
1363 | key.objectid = bytenr; |
1364 | key.offset = num_bytes; | |
62e2749e | 1365 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
5caf2a00 | 1366 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 1367 | 0, 0); |
54aa1f4d CM |
1368 | if (ret < 0) |
1369 | goto out; | |
5f39d397 CM |
1370 | if (ret != 0) { |
1371 | btrfs_print_leaf(root, path->nodes[0]); | |
db94535d | 1372 | printk("failed to find block number %Lu\n", bytenr); |
a28ec197 | 1373 | BUG(); |
5f39d397 CM |
1374 | } |
1375 | l = path->nodes[0]; | |
5caf2a00 | 1376 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 | 1377 | *refs = btrfs_extent_refs(l, item); |
54aa1f4d | 1378 | out: |
5caf2a00 | 1379 | btrfs_free_path(path); |
a28ec197 CM |
1380 | return 0; |
1381 | } | |
1382 | ||
80ff3856 YZ |
1383 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, |
1384 | struct btrfs_root *root, u64 bytenr) | |
be20aa9d CM |
1385 | { |
1386 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
1387 | struct btrfs_path *path; | |
f321e491 YZ |
1388 | struct extent_buffer *leaf; |
1389 | struct btrfs_extent_ref *ref_item; | |
1390 | struct btrfs_key key; | |
1391 | struct btrfs_key found_key; | |
80ff3856 YZ |
1392 | u64 ref_root; |
1393 | u64 last_snapshot; | |
be20aa9d CM |
1394 | u32 nritems; |
1395 | int ret; | |
925baedd | 1396 | |
be20aa9d | 1397 | key.objectid = bytenr; |
31840ae1 | 1398 | key.offset = (u64)-1; |
f321e491 | 1399 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 1400 | |
f321e491 | 1401 | path = btrfs_alloc_path(); |
be20aa9d CM |
1402 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
1403 | if (ret < 0) | |
1404 | goto out; | |
1405 | BUG_ON(ret == 0); | |
80ff3856 YZ |
1406 | |
1407 | ret = -ENOENT; | |
1408 | if (path->slots[0] == 0) | |
31840ae1 | 1409 | goto out; |
be20aa9d | 1410 | |
31840ae1 | 1411 | path->slots[0]--; |
f321e491 YZ |
1412 | leaf = path->nodes[0]; |
1413 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
be20aa9d CM |
1414 | |
1415 | if (found_key.objectid != bytenr || | |
80ff3856 | 1416 | found_key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 1417 | goto out; |
f321e491 | 1418 | |
80ff3856 | 1419 | last_snapshot = btrfs_root_last_snapshot(&root->root_item); |
be20aa9d | 1420 | while (1) { |
f321e491 YZ |
1421 | leaf = path->nodes[0]; |
1422 | nritems = btrfs_header_nritems(leaf); | |
be20aa9d CM |
1423 | if (path->slots[0] >= nritems) { |
1424 | ret = btrfs_next_leaf(extent_root, path); | |
f321e491 YZ |
1425 | if (ret < 0) |
1426 | goto out; | |
be20aa9d CM |
1427 | if (ret == 0) |
1428 | continue; | |
1429 | break; | |
1430 | } | |
f321e491 | 1431 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
be20aa9d CM |
1432 | if (found_key.objectid != bytenr) |
1433 | break; | |
bd09835d | 1434 | |
be20aa9d CM |
1435 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
1436 | path->slots[0]++; | |
1437 | continue; | |
1438 | } | |
1439 | ||
f321e491 | 1440 | ref_item = btrfs_item_ptr(leaf, path->slots[0], |
be20aa9d | 1441 | struct btrfs_extent_ref); |
80ff3856 YZ |
1442 | ref_root = btrfs_ref_root(leaf, ref_item); |
1443 | if (ref_root != root->root_key.objectid && | |
1444 | ref_root != BTRFS_TREE_LOG_OBJECTID) { | |
1445 | ret = 1; | |
f321e491 | 1446 | goto out; |
80ff3856 YZ |
1447 | } |
1448 | if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) { | |
f321e491 YZ |
1449 | ret = 1; |
1450 | goto out; | |
1451 | } | |
80ff3856 YZ |
1452 | |
1453 | path->slots[0]++; | |
f321e491 YZ |
1454 | } |
1455 | ret = 0; | |
be20aa9d | 1456 | out: |
80ff3856 | 1457 | btrfs_free_path(path); |
f321e491 | 1458 | return ret; |
be20aa9d | 1459 | } |
c5739bba | 1460 | |
31840ae1 ZY |
1461 | int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1462 | struct extent_buffer *buf, u32 nr_extents) | |
02217ed2 | 1463 | { |
5f39d397 | 1464 | struct btrfs_key key; |
6407bf6d | 1465 | struct btrfs_file_extent_item *fi; |
e4657689 ZY |
1466 | u64 root_gen; |
1467 | u32 nritems; | |
02217ed2 | 1468 | int i; |
db94535d | 1469 | int level; |
31840ae1 | 1470 | int ret = 0; |
e4657689 | 1471 | int shared = 0; |
a28ec197 | 1472 | |
3768f368 | 1473 | if (!root->ref_cows) |
a28ec197 | 1474 | return 0; |
5f39d397 | 1475 | |
e4657689 ZY |
1476 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
1477 | shared = 0; | |
1478 | root_gen = root->root_key.offset; | |
1479 | } else { | |
1480 | shared = 1; | |
1481 | root_gen = trans->transid - 1; | |
1482 | } | |
1483 | ||
db94535d | 1484 | level = btrfs_header_level(buf); |
5f39d397 | 1485 | nritems = btrfs_header_nritems(buf); |
4a096752 | 1486 | |
31840ae1 | 1487 | if (level == 0) { |
31153d81 YZ |
1488 | struct btrfs_leaf_ref *ref; |
1489 | struct btrfs_extent_info *info; | |
1490 | ||
31840ae1 | 1491 | ref = btrfs_alloc_leaf_ref(root, nr_extents); |
31153d81 | 1492 | if (!ref) { |
31840ae1 | 1493 | ret = -ENOMEM; |
31153d81 YZ |
1494 | goto out; |
1495 | } | |
1496 | ||
e4657689 | 1497 | ref->root_gen = root_gen; |
31153d81 YZ |
1498 | ref->bytenr = buf->start; |
1499 | ref->owner = btrfs_header_owner(buf); | |
1500 | ref->generation = btrfs_header_generation(buf); | |
31840ae1 | 1501 | ref->nritems = nr_extents; |
31153d81 | 1502 | info = ref->extents; |
bcc63abb | 1503 | |
31840ae1 | 1504 | for (i = 0; nr_extents > 0 && i < nritems; i++) { |
31153d81 YZ |
1505 | u64 disk_bytenr; |
1506 | btrfs_item_key_to_cpu(buf, &key, i); | |
1507 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
1508 | continue; | |
1509 | fi = btrfs_item_ptr(buf, i, | |
1510 | struct btrfs_file_extent_item); | |
1511 | if (btrfs_file_extent_type(buf, fi) == | |
1512 | BTRFS_FILE_EXTENT_INLINE) | |
1513 | continue; | |
1514 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
1515 | if (disk_bytenr == 0) | |
1516 | continue; | |
1517 | ||
1518 | info->bytenr = disk_bytenr; | |
1519 | info->num_bytes = | |
1520 | btrfs_file_extent_disk_num_bytes(buf, fi); | |
1521 | info->objectid = key.objectid; | |
1522 | info->offset = key.offset; | |
1523 | info++; | |
1524 | } | |
1525 | ||
e4657689 | 1526 | ret = btrfs_add_leaf_ref(root, ref, shared); |
5b84e8d6 YZ |
1527 | if (ret == -EEXIST && shared) { |
1528 | struct btrfs_leaf_ref *old; | |
1529 | old = btrfs_lookup_leaf_ref(root, ref->bytenr); | |
1530 | BUG_ON(!old); | |
1531 | btrfs_remove_leaf_ref(root, old); | |
1532 | btrfs_free_leaf_ref(root, old); | |
1533 | ret = btrfs_add_leaf_ref(root, ref, shared); | |
1534 | } | |
31153d81 | 1535 | WARN_ON(ret); |
bcc63abb | 1536 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
1537 | } |
1538 | out: | |
31840ae1 ZY |
1539 | return ret; |
1540 | } | |
1541 | ||
1542 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
1543 | struct extent_buffer *orig_buf, struct extent_buffer *buf, | |
1544 | u32 *nr_extents) | |
1545 | { | |
1546 | u64 bytenr; | |
1547 | u64 ref_root; | |
1548 | u64 orig_root; | |
1549 | u64 ref_generation; | |
1550 | u64 orig_generation; | |
1551 | u32 nritems; | |
1552 | u32 nr_file_extents = 0; | |
1553 | struct btrfs_key key; | |
1554 | struct btrfs_file_extent_item *fi; | |
1555 | int i; | |
1556 | int level; | |
1557 | int ret = 0; | |
1558 | int faili = 0; | |
1559 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, | |
3bb1a1bc | 1560 | u64, u64, u64, u64, u64, u64, u64, u64); |
31840ae1 ZY |
1561 | |
1562 | ref_root = btrfs_header_owner(buf); | |
1563 | ref_generation = btrfs_header_generation(buf); | |
1564 | orig_root = btrfs_header_owner(orig_buf); | |
1565 | orig_generation = btrfs_header_generation(orig_buf); | |
1566 | ||
1567 | nritems = btrfs_header_nritems(buf); | |
1568 | level = btrfs_header_level(buf); | |
1569 | ||
1570 | if (root->ref_cows) { | |
1571 | process_func = __btrfs_inc_extent_ref; | |
1572 | } else { | |
1573 | if (level == 0 && | |
1574 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
1575 | goto out; | |
1576 | if (level != 0 && | |
1577 | root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) | |
1578 | goto out; | |
1579 | process_func = __btrfs_update_extent_ref; | |
1580 | } | |
1581 | ||
1582 | for (i = 0; i < nritems; i++) { | |
1583 | cond_resched(); | |
db94535d | 1584 | if (level == 0) { |
5f39d397 CM |
1585 | btrfs_item_key_to_cpu(buf, &key, i); |
1586 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
54aa1f4d | 1587 | continue; |
5f39d397 | 1588 | fi = btrfs_item_ptr(buf, i, |
54aa1f4d | 1589 | struct btrfs_file_extent_item); |
5f39d397 | 1590 | if (btrfs_file_extent_type(buf, fi) == |
54aa1f4d CM |
1591 | BTRFS_FILE_EXTENT_INLINE) |
1592 | continue; | |
31840ae1 ZY |
1593 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
1594 | if (bytenr == 0) | |
54aa1f4d | 1595 | continue; |
31840ae1 ZY |
1596 | |
1597 | nr_file_extents++; | |
1598 | ||
31840ae1 ZY |
1599 | ret = process_func(trans, root, bytenr, |
1600 | orig_buf->start, buf->start, | |
1601 | orig_root, ref_root, | |
1602 | orig_generation, ref_generation, | |
3bb1a1bc | 1603 | key.objectid); |
31840ae1 ZY |
1604 | |
1605 | if (ret) { | |
1606 | faili = i; | |
1607 | WARN_ON(1); | |
1608 | goto fail; | |
1609 | } | |
54aa1f4d | 1610 | } else { |
db94535d | 1611 | bytenr = btrfs_node_blockptr(buf, i); |
31840ae1 ZY |
1612 | ret = process_func(trans, root, bytenr, |
1613 | orig_buf->start, buf->start, | |
1614 | orig_root, ref_root, | |
1615 | orig_generation, ref_generation, | |
3bb1a1bc | 1616 | level - 1); |
31840ae1 ZY |
1617 | if (ret) { |
1618 | faili = i; | |
1619 | WARN_ON(1); | |
1620 | goto fail; | |
1621 | } | |
54aa1f4d CM |
1622 | } |
1623 | } | |
31840ae1 ZY |
1624 | out: |
1625 | if (nr_extents) { | |
1626 | if (level == 0) | |
1627 | *nr_extents = nr_file_extents; | |
1628 | else | |
1629 | *nr_extents = nritems; | |
1630 | } | |
1631 | return 0; | |
1632 | fail: | |
1633 | WARN_ON(1); | |
54aa1f4d | 1634 | return ret; |
02217ed2 CM |
1635 | } |
1636 | ||
31840ae1 ZY |
1637 | int btrfs_update_ref(struct btrfs_trans_handle *trans, |
1638 | struct btrfs_root *root, struct extent_buffer *orig_buf, | |
1639 | struct extent_buffer *buf, int start_slot, int nr) | |
1640 | ||
1641 | { | |
1642 | u64 bytenr; | |
1643 | u64 ref_root; | |
1644 | u64 orig_root; | |
1645 | u64 ref_generation; | |
1646 | u64 orig_generation; | |
1647 | struct btrfs_key key; | |
1648 | struct btrfs_file_extent_item *fi; | |
1649 | int i; | |
1650 | int ret; | |
1651 | int slot; | |
1652 | int level; | |
1653 | ||
1654 | BUG_ON(start_slot < 0); | |
1655 | BUG_ON(start_slot + nr > btrfs_header_nritems(buf)); | |
1656 | ||
1657 | ref_root = btrfs_header_owner(buf); | |
1658 | ref_generation = btrfs_header_generation(buf); | |
1659 | orig_root = btrfs_header_owner(orig_buf); | |
1660 | orig_generation = btrfs_header_generation(orig_buf); | |
1661 | level = btrfs_header_level(buf); | |
1662 | ||
1663 | if (!root->ref_cows) { | |
1664 | if (level == 0 && | |
1665 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
1666 | return 0; | |
1667 | if (level != 0 && | |
1668 | root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) | |
1669 | return 0; | |
1670 | } | |
1671 | ||
1672 | for (i = 0, slot = start_slot; i < nr; i++, slot++) { | |
1673 | cond_resched(); | |
1674 | if (level == 0) { | |
1675 | btrfs_item_key_to_cpu(buf, &key, slot); | |
1676 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
1677 | continue; | |
1678 | fi = btrfs_item_ptr(buf, slot, | |
1679 | struct btrfs_file_extent_item); | |
1680 | if (btrfs_file_extent_type(buf, fi) == | |
1681 | BTRFS_FILE_EXTENT_INLINE) | |
1682 | continue; | |
1683 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
1684 | if (bytenr == 0) | |
1685 | continue; | |
31840ae1 ZY |
1686 | ret = __btrfs_update_extent_ref(trans, root, bytenr, |
1687 | orig_buf->start, buf->start, | |
1688 | orig_root, ref_root, | |
1689 | orig_generation, ref_generation, | |
3bb1a1bc | 1690 | key.objectid); |
31840ae1 ZY |
1691 | if (ret) |
1692 | goto fail; | |
1693 | } else { | |
1694 | bytenr = btrfs_node_blockptr(buf, slot); | |
31840ae1 ZY |
1695 | ret = __btrfs_update_extent_ref(trans, root, bytenr, |
1696 | orig_buf->start, buf->start, | |
1697 | orig_root, ref_root, | |
1698 | orig_generation, ref_generation, | |
3bb1a1bc | 1699 | level - 1); |
31840ae1 ZY |
1700 | if (ret) |
1701 | goto fail; | |
1702 | } | |
1703 | } | |
1704 | return 0; | |
1705 | fail: | |
1706 | WARN_ON(1); | |
1707 | return -1; | |
1708 | } | |
1709 | ||
9078a3e1 CM |
1710 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
1711 | struct btrfs_root *root, | |
1712 | struct btrfs_path *path, | |
1713 | struct btrfs_block_group_cache *cache) | |
1714 | { | |
1715 | int ret; | |
1716 | int pending_ret; | |
1717 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
5f39d397 CM |
1718 | unsigned long bi; |
1719 | struct extent_buffer *leaf; | |
9078a3e1 | 1720 | |
9078a3e1 | 1721 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
1722 | if (ret < 0) |
1723 | goto fail; | |
9078a3e1 | 1724 | BUG_ON(ret); |
5f39d397 CM |
1725 | |
1726 | leaf = path->nodes[0]; | |
1727 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1728 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
1729 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 1730 | btrfs_release_path(extent_root, path); |
54aa1f4d | 1731 | fail: |
87ef2bb4 CM |
1732 | finish_current_insert(trans, extent_root, 0); |
1733 | pending_ret = del_pending_extents(trans, extent_root, 0); | |
9078a3e1 CM |
1734 | if (ret) |
1735 | return ret; | |
1736 | if (pending_ret) | |
1737 | return pending_ret; | |
1738 | return 0; | |
1739 | ||
1740 | } | |
1741 | ||
96b5179d CM |
1742 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
1743 | struct btrfs_root *root) | |
9078a3e1 | 1744 | { |
0f9dd46c JB |
1745 | struct btrfs_block_group_cache *cache, *entry; |
1746 | struct rb_node *n; | |
9078a3e1 CM |
1747 | int err = 0; |
1748 | int werr = 0; | |
9078a3e1 | 1749 | struct btrfs_path *path; |
96b5179d | 1750 | u64 last = 0; |
9078a3e1 CM |
1751 | |
1752 | path = btrfs_alloc_path(); | |
1753 | if (!path) | |
1754 | return -ENOMEM; | |
1755 | ||
1756 | while(1) { | |
0f9dd46c JB |
1757 | cache = NULL; |
1758 | spin_lock(&root->fs_info->block_group_cache_lock); | |
1759 | for (n = rb_first(&root->fs_info->block_group_cache_tree); | |
1760 | n; n = rb_next(n)) { | |
1761 | entry = rb_entry(n, struct btrfs_block_group_cache, | |
1762 | cache_node); | |
1763 | if (entry->dirty) { | |
1764 | cache = entry; | |
1765 | break; | |
1766 | } | |
1767 | } | |
1768 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
54aa1f4d | 1769 | |
0f9dd46c | 1770 | if (!cache) |
96b5179d | 1771 | break; |
0f9dd46c | 1772 | |
e8569813 | 1773 | cache->dirty = 0; |
0f9dd46c JB |
1774 | last += cache->key.offset; |
1775 | ||
96b5179d CM |
1776 | err = write_one_cache_group(trans, root, |
1777 | path, cache); | |
1778 | /* | |
1779 | * if we fail to write the cache group, we want | |
1780 | * to keep it marked dirty in hopes that a later | |
1781 | * write will work | |
1782 | */ | |
1783 | if (err) { | |
1784 | werr = err; | |
1785 | continue; | |
9078a3e1 CM |
1786 | } |
1787 | } | |
1788 | btrfs_free_path(path); | |
1789 | return werr; | |
1790 | } | |
1791 | ||
593060d7 CM |
1792 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
1793 | u64 total_bytes, u64 bytes_used, | |
1794 | struct btrfs_space_info **space_info) | |
1795 | { | |
1796 | struct btrfs_space_info *found; | |
1797 | ||
1798 | found = __find_space_info(info, flags); | |
1799 | if (found) { | |
25179201 | 1800 | spin_lock(&found->lock); |
593060d7 CM |
1801 | found->total_bytes += total_bytes; |
1802 | found->bytes_used += bytes_used; | |
8f18cf13 | 1803 | found->full = 0; |
25179201 | 1804 | spin_unlock(&found->lock); |
593060d7 CM |
1805 | *space_info = found; |
1806 | return 0; | |
1807 | } | |
c146afad | 1808 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
1809 | if (!found) |
1810 | return -ENOMEM; | |
1811 | ||
1812 | list_add(&found->list, &info->space_info); | |
0f9dd46c | 1813 | INIT_LIST_HEAD(&found->block_groups); |
80eb234a | 1814 | init_rwsem(&found->groups_sem); |
0f9dd46c | 1815 | spin_lock_init(&found->lock); |
593060d7 CM |
1816 | found->flags = flags; |
1817 | found->total_bytes = total_bytes; | |
1818 | found->bytes_used = bytes_used; | |
1819 | found->bytes_pinned = 0; | |
e8569813 | 1820 | found->bytes_reserved = 0; |
c146afad | 1821 | found->bytes_readonly = 0; |
593060d7 | 1822 | found->full = 0; |
0ef3e66b | 1823 | found->force_alloc = 0; |
593060d7 CM |
1824 | *space_info = found; |
1825 | return 0; | |
1826 | } | |
1827 | ||
8790d502 CM |
1828 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
1829 | { | |
1830 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 1831 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 1832 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 1833 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
1834 | if (extra_flags) { |
1835 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1836 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1837 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1838 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1839 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1840 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1841 | } | |
1842 | } | |
593060d7 | 1843 | |
c146afad YZ |
1844 | static void set_block_group_readonly(struct btrfs_block_group_cache *cache) |
1845 | { | |
1846 | spin_lock(&cache->space_info->lock); | |
1847 | spin_lock(&cache->lock); | |
1848 | if (!cache->ro) { | |
1849 | cache->space_info->bytes_readonly += cache->key.offset - | |
1850 | btrfs_block_group_used(&cache->item); | |
1851 | cache->ro = 1; | |
1852 | } | |
1853 | spin_unlock(&cache->lock); | |
1854 | spin_unlock(&cache->space_info->lock); | |
1855 | } | |
1856 | ||
2b82032c | 1857 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 1858 | { |
2b82032c | 1859 | u64 num_devices = root->fs_info->fs_devices->rw_devices; |
a061fc8d CM |
1860 | |
1861 | if (num_devices == 1) | |
1862 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
1863 | if (num_devices < 4) | |
1864 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
1865 | ||
ec44a35c CM |
1866 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
1867 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 1868 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 1869 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 1870 | } |
ec44a35c CM |
1871 | |
1872 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 1873 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 1874 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 1875 | } |
ec44a35c CM |
1876 | |
1877 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
1878 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
1879 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
1880 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
1881 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
1882 | return flags; | |
1883 | } | |
1884 | ||
6324fbf3 CM |
1885 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
1886 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 1887 | u64 flags, int force) |
6324fbf3 CM |
1888 | { |
1889 | struct btrfs_space_info *space_info; | |
1890 | u64 thresh; | |
c146afad YZ |
1891 | int ret = 0; |
1892 | ||
1893 | mutex_lock(&extent_root->fs_info->chunk_mutex); | |
6324fbf3 | 1894 | |
2b82032c | 1895 | flags = btrfs_reduce_alloc_profile(extent_root, flags); |
ec44a35c | 1896 | |
6324fbf3 | 1897 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
1898 | if (!space_info) { |
1899 | ret = update_space_info(extent_root->fs_info, flags, | |
1900 | 0, 0, &space_info); | |
1901 | BUG_ON(ret); | |
1902 | } | |
6324fbf3 CM |
1903 | BUG_ON(!space_info); |
1904 | ||
25179201 | 1905 | spin_lock(&space_info->lock); |
0ef3e66b CM |
1906 | if (space_info->force_alloc) { |
1907 | force = 1; | |
1908 | space_info->force_alloc = 0; | |
1909 | } | |
25179201 JB |
1910 | if (space_info->full) { |
1911 | spin_unlock(&space_info->lock); | |
925baedd | 1912 | goto out; |
25179201 | 1913 | } |
6324fbf3 | 1914 | |
c146afad YZ |
1915 | thresh = space_info->total_bytes - space_info->bytes_readonly; |
1916 | thresh = div_factor(thresh, 6); | |
0ef3e66b | 1917 | if (!force && |
e8569813 | 1918 | (space_info->bytes_used + space_info->bytes_pinned + |
25179201 JB |
1919 | space_info->bytes_reserved + alloc_bytes) < thresh) { |
1920 | spin_unlock(&space_info->lock); | |
925baedd | 1921 | goto out; |
25179201 | 1922 | } |
25179201 JB |
1923 | spin_unlock(&space_info->lock); |
1924 | ||
2b82032c | 1925 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
25179201 | 1926 | if (ret) { |
6324fbf3 CM |
1927 | printk("space info full %Lu\n", flags); |
1928 | space_info->full = 1; | |
6324fbf3 | 1929 | } |
a74a4b97 | 1930 | out: |
c146afad | 1931 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 1932 | return ret; |
6324fbf3 CM |
1933 | } |
1934 | ||
9078a3e1 CM |
1935 | static int update_block_group(struct btrfs_trans_handle *trans, |
1936 | struct btrfs_root *root, | |
db94535d | 1937 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 1938 | int mark_free) |
9078a3e1 CM |
1939 | { |
1940 | struct btrfs_block_group_cache *cache; | |
1941 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 1942 | u64 total = num_bytes; |
9078a3e1 | 1943 | u64 old_val; |
db94535d | 1944 | u64 byte_in_group; |
3e1ad54f | 1945 | |
9078a3e1 | 1946 | while(total) { |
db94535d | 1947 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 1948 | if (!cache) |
9078a3e1 | 1949 | return -1; |
db94535d CM |
1950 | byte_in_group = bytenr - cache->key.objectid; |
1951 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 1952 | |
25179201 | 1953 | spin_lock(&cache->space_info->lock); |
c286ac48 | 1954 | spin_lock(&cache->lock); |
0f9dd46c | 1955 | cache->dirty = 1; |
9078a3e1 | 1956 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 1957 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 1958 | if (alloc) { |
db94535d | 1959 | old_val += num_bytes; |
6324fbf3 | 1960 | cache->space_info->bytes_used += num_bytes; |
c146afad YZ |
1961 | if (cache->ro) { |
1962 | cache->space_info->bytes_readonly -= num_bytes; | |
1963 | WARN_ON(1); | |
1964 | } | |
c286ac48 CM |
1965 | btrfs_set_block_group_used(&cache->item, old_val); |
1966 | spin_unlock(&cache->lock); | |
25179201 | 1967 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 1968 | } else { |
db94535d | 1969 | old_val -= num_bytes; |
6324fbf3 | 1970 | cache->space_info->bytes_used -= num_bytes; |
c146afad YZ |
1971 | if (cache->ro) |
1972 | cache->space_info->bytes_readonly += num_bytes; | |
c286ac48 CM |
1973 | btrfs_set_block_group_used(&cache->item, old_val); |
1974 | spin_unlock(&cache->lock); | |
25179201 | 1975 | spin_unlock(&cache->space_info->lock); |
f510cfec | 1976 | if (mark_free) { |
0f9dd46c JB |
1977 | int ret; |
1978 | ret = btrfs_add_free_space(cache, bytenr, | |
1979 | num_bytes); | |
1980 | if (ret) | |
1981 | return -1; | |
e37c9e69 | 1982 | } |
cd1bc465 | 1983 | } |
db94535d CM |
1984 | total -= num_bytes; |
1985 | bytenr += num_bytes; | |
9078a3e1 CM |
1986 | } |
1987 | return 0; | |
1988 | } | |
6324fbf3 | 1989 | |
a061fc8d CM |
1990 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
1991 | { | |
0f9dd46c JB |
1992 | struct btrfs_block_group_cache *cache; |
1993 | ||
1994 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
1995 | if (!cache) | |
a061fc8d | 1996 | return 0; |
0f9dd46c JB |
1997 | |
1998 | return cache->key.objectid; | |
a061fc8d CM |
1999 | } |
2000 | ||
e02119d5 | 2001 | int btrfs_update_pinned_extents(struct btrfs_root *root, |
324ae4df Y |
2002 | u64 bytenr, u64 num, int pin) |
2003 | { | |
2004 | u64 len; | |
2005 | struct btrfs_block_group_cache *cache; | |
2006 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2007 | ||
25179201 | 2008 | WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex)); |
324ae4df Y |
2009 | if (pin) { |
2010 | set_extent_dirty(&fs_info->pinned_extents, | |
2011 | bytenr, bytenr + num - 1, GFP_NOFS); | |
2012 | } else { | |
2013 | clear_extent_dirty(&fs_info->pinned_extents, | |
2014 | bytenr, bytenr + num - 1, GFP_NOFS); | |
2015 | } | |
2016 | while (num > 0) { | |
2017 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
e8569813 ZY |
2018 | BUG_ON(!cache); |
2019 | len = min(num, cache->key.offset - | |
2020 | (bytenr - cache->key.objectid)); | |
324ae4df | 2021 | if (pin) { |
25179201 | 2022 | spin_lock(&cache->space_info->lock); |
e8569813 ZY |
2023 | spin_lock(&cache->lock); |
2024 | cache->pinned += len; | |
2025 | cache->space_info->bytes_pinned += len; | |
2026 | spin_unlock(&cache->lock); | |
25179201 | 2027 | spin_unlock(&cache->space_info->lock); |
324ae4df Y |
2028 | fs_info->total_pinned += len; |
2029 | } else { | |
25179201 | 2030 | spin_lock(&cache->space_info->lock); |
e8569813 ZY |
2031 | spin_lock(&cache->lock); |
2032 | cache->pinned -= len; | |
2033 | cache->space_info->bytes_pinned -= len; | |
2034 | spin_unlock(&cache->lock); | |
25179201 | 2035 | spin_unlock(&cache->space_info->lock); |
324ae4df | 2036 | fs_info->total_pinned -= len; |
07103a3c JB |
2037 | if (cache->cached) |
2038 | btrfs_add_free_space(cache, bytenr, len); | |
324ae4df Y |
2039 | } |
2040 | bytenr += len; | |
2041 | num -= len; | |
2042 | } | |
2043 | return 0; | |
2044 | } | |
9078a3e1 | 2045 | |
e8569813 ZY |
2046 | static int update_reserved_extents(struct btrfs_root *root, |
2047 | u64 bytenr, u64 num, int reserve) | |
2048 | { | |
2049 | u64 len; | |
2050 | struct btrfs_block_group_cache *cache; | |
2051 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2052 | ||
e8569813 ZY |
2053 | while (num > 0) { |
2054 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
2055 | BUG_ON(!cache); | |
2056 | len = min(num, cache->key.offset - | |
2057 | (bytenr - cache->key.objectid)); | |
25179201 JB |
2058 | |
2059 | spin_lock(&cache->space_info->lock); | |
2060 | spin_lock(&cache->lock); | |
e8569813 | 2061 | if (reserve) { |
e8569813 ZY |
2062 | cache->reserved += len; |
2063 | cache->space_info->bytes_reserved += len; | |
e8569813 | 2064 | } else { |
e8569813 ZY |
2065 | cache->reserved -= len; |
2066 | cache->space_info->bytes_reserved -= len; | |
e8569813 | 2067 | } |
25179201 JB |
2068 | spin_unlock(&cache->lock); |
2069 | spin_unlock(&cache->space_info->lock); | |
e8569813 ZY |
2070 | bytenr += len; |
2071 | num -= len; | |
2072 | } | |
2073 | return 0; | |
2074 | } | |
2075 | ||
d1310b2e | 2076 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) |
ccd467d6 | 2077 | { |
ccd467d6 | 2078 | u64 last = 0; |
1a5bc167 CM |
2079 | u64 start; |
2080 | u64 end; | |
d1310b2e | 2081 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; |
ccd467d6 | 2082 | int ret; |
ccd467d6 | 2083 | |
25179201 | 2084 | mutex_lock(&root->fs_info->pinned_mutex); |
ccd467d6 | 2085 | while(1) { |
1a5bc167 CM |
2086 | ret = find_first_extent_bit(pinned_extents, last, |
2087 | &start, &end, EXTENT_DIRTY); | |
2088 | if (ret) | |
ccd467d6 | 2089 | break; |
1a5bc167 CM |
2090 | set_extent_dirty(copy, start, end, GFP_NOFS); |
2091 | last = end + 1; | |
ccd467d6 | 2092 | } |
25179201 | 2093 | mutex_unlock(&root->fs_info->pinned_mutex); |
ccd467d6 CM |
2094 | return 0; |
2095 | } | |
2096 | ||
2097 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
2098 | struct btrfs_root *root, | |
d1310b2e | 2099 | struct extent_io_tree *unpin) |
a28ec197 | 2100 | { |
1a5bc167 CM |
2101 | u64 start; |
2102 | u64 end; | |
a28ec197 | 2103 | int ret; |
a28ec197 | 2104 | |
25179201 | 2105 | mutex_lock(&root->fs_info->pinned_mutex); |
a28ec197 | 2106 | while(1) { |
1a5bc167 CM |
2107 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
2108 | EXTENT_DIRTY); | |
2109 | if (ret) | |
a28ec197 | 2110 | break; |
e02119d5 | 2111 | btrfs_update_pinned_extents(root, start, end + 1 - start, 0); |
1a5bc167 | 2112 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
c286ac48 | 2113 | if (need_resched()) { |
25179201 | 2114 | mutex_unlock(&root->fs_info->pinned_mutex); |
c286ac48 | 2115 | cond_resched(); |
25179201 | 2116 | mutex_lock(&root->fs_info->pinned_mutex); |
c286ac48 | 2117 | } |
a28ec197 | 2118 | } |
25179201 | 2119 | mutex_unlock(&root->fs_info->pinned_mutex); |
a28ec197 CM |
2120 | return 0; |
2121 | } | |
2122 | ||
98ed5174 | 2123 | static int finish_current_insert(struct btrfs_trans_handle *trans, |
87ef2bb4 | 2124 | struct btrfs_root *extent_root, int all) |
037e6390 | 2125 | { |
7bb86316 CM |
2126 | u64 start; |
2127 | u64 end; | |
31840ae1 | 2128 | u64 priv; |
25179201 | 2129 | u64 search = 0; |
f3465ca4 | 2130 | u64 skipped = 0; |
7bb86316 CM |
2131 | struct btrfs_fs_info *info = extent_root->fs_info; |
2132 | struct btrfs_path *path; | |
f3465ca4 JB |
2133 | struct pending_extent_op *extent_op, *tmp; |
2134 | struct list_head insert_list, update_list; | |
037e6390 | 2135 | int ret; |
f3465ca4 | 2136 | int num_inserts = 0, max_inserts; |
037e6390 | 2137 | |
7bb86316 | 2138 | path = btrfs_alloc_path(); |
f3465ca4 JB |
2139 | INIT_LIST_HEAD(&insert_list); |
2140 | INIT_LIST_HEAD(&update_list); | |
037e6390 | 2141 | |
f3465ca4 JB |
2142 | max_inserts = extent_root->leafsize / |
2143 | (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) + | |
2144 | sizeof(struct btrfs_extent_ref) + | |
2145 | sizeof(struct btrfs_extent_item)); | |
2146 | again: | |
2147 | mutex_lock(&info->extent_ins_mutex); | |
2148 | while (1) { | |
25179201 JB |
2149 | ret = find_first_extent_bit(&info->extent_ins, search, &start, |
2150 | &end, EXTENT_WRITEBACK); | |
2151 | if (ret) { | |
f3465ca4 JB |
2152 | if (skipped && all && !num_inserts) { |
2153 | skipped = 0; | |
b4eec2ca | 2154 | search = 0; |
25179201 JB |
2155 | continue; |
2156 | } | |
f3465ca4 | 2157 | mutex_unlock(&info->extent_ins_mutex); |
26b8003f | 2158 | break; |
25179201 JB |
2159 | } |
2160 | ||
2161 | ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS); | |
2162 | if (!ret) { | |
f3465ca4 | 2163 | skipped = 1; |
87ef2bb4 | 2164 | search = end + 1; |
f3465ca4 JB |
2165 | if (need_resched()) { |
2166 | mutex_unlock(&info->extent_ins_mutex); | |
2167 | cond_resched(); | |
2168 | mutex_lock(&info->extent_ins_mutex); | |
2169 | } | |
25179201 JB |
2170 | continue; |
2171 | } | |
26b8003f | 2172 | |
31840ae1 ZY |
2173 | ret = get_state_private(&info->extent_ins, start, &priv); |
2174 | BUG_ON(ret); | |
f3465ca4 | 2175 | extent_op = (struct pending_extent_op *)(unsigned long) priv; |
25179201 | 2176 | |
31840ae1 | 2177 | if (extent_op->type == PENDING_EXTENT_INSERT) { |
f3465ca4 JB |
2178 | num_inserts++; |
2179 | list_add_tail(&extent_op->list, &insert_list); | |
2180 | search = end + 1; | |
2181 | if (num_inserts == max_inserts) { | |
2182 | mutex_unlock(&info->extent_ins_mutex); | |
2183 | break; | |
2184 | } | |
2185 | } else if (extent_op->type == PENDING_BACKREF_UPDATE) { | |
2186 | list_add_tail(&extent_op->list, &update_list); | |
2187 | search = end + 1; | |
2188 | } else { | |
2189 | BUG(); | |
2190 | } | |
2191 | } | |
c286ac48 | 2192 | |
f3465ca4 | 2193 | /* |
b4eec2ca | 2194 | * process the update list, clear the writeback bit for it, and if |
f3465ca4 JB |
2195 | * somebody marked this thing for deletion then just unlock it and be |
2196 | * done, the free_extents will handle it | |
2197 | */ | |
2198 | mutex_lock(&info->extent_ins_mutex); | |
2199 | list_for_each_entry_safe(extent_op, tmp, &update_list, list) { | |
2200 | clear_extent_bits(&info->extent_ins, extent_op->bytenr, | |
2201 | extent_op->bytenr + extent_op->num_bytes - 1, | |
2202 | EXTENT_WRITEBACK, GFP_NOFS); | |
2203 | if (extent_op->del) { | |
2204 | list_del_init(&extent_op->list); | |
2205 | unlock_extent(&info->extent_ins, extent_op->bytenr, | |
2206 | extent_op->bytenr + extent_op->num_bytes | |
2207 | - 1, GFP_NOFS); | |
2208 | kfree(extent_op); | |
2209 | } | |
2210 | } | |
2211 | mutex_unlock(&info->extent_ins_mutex); | |
c286ac48 | 2212 | |
f3465ca4 JB |
2213 | /* |
2214 | * still have things left on the update list, go ahead an update | |
2215 | * everything | |
2216 | */ | |
2217 | if (!list_empty(&update_list)) { | |
2218 | ret = update_backrefs(trans, extent_root, path, &update_list); | |
2219 | BUG_ON(ret); | |
2220 | } | |
c286ac48 | 2221 | |
f3465ca4 JB |
2222 | /* |
2223 | * if no inserts need to be done, but we skipped some extents and we | |
2224 | * need to make sure everything is cleaned then reset everything and | |
2225 | * go back to the beginning | |
2226 | */ | |
2227 | if (!num_inserts && all && skipped) { | |
2228 | search = 0; | |
2229 | skipped = 0; | |
2230 | INIT_LIST_HEAD(&update_list); | |
2231 | INIT_LIST_HEAD(&insert_list); | |
2232 | goto again; | |
2233 | } else if (!num_inserts) { | |
2234 | goto out; | |
2235 | } | |
31840ae1 | 2236 | |
f3465ca4 JB |
2237 | /* |
2238 | * process the insert extents list. Again if we are deleting this | |
2239 | * extent, then just unlock it, pin down the bytes if need be, and be | |
2240 | * done with it. Saves us from having to actually insert the extent | |
2241 | * into the tree and then subsequently come along and delete it | |
2242 | */ | |
2243 | mutex_lock(&info->extent_ins_mutex); | |
2244 | list_for_each_entry_safe(extent_op, tmp, &insert_list, list) { | |
2245 | clear_extent_bits(&info->extent_ins, extent_op->bytenr, | |
2246 | extent_op->bytenr + extent_op->num_bytes - 1, | |
2247 | EXTENT_WRITEBACK, GFP_NOFS); | |
2248 | if (extent_op->del) { | |
2249 | list_del_init(&extent_op->list); | |
2250 | unlock_extent(&info->extent_ins, extent_op->bytenr, | |
2251 | extent_op->bytenr + extent_op->num_bytes | |
2252 | - 1, GFP_NOFS); | |
2253 | ||
2254 | mutex_lock(&extent_root->fs_info->pinned_mutex); | |
2255 | ret = pin_down_bytes(trans, extent_root, | |
2256 | extent_op->bytenr, | |
2257 | extent_op->num_bytes, 0); | |
2258 | mutex_unlock(&extent_root->fs_info->pinned_mutex); | |
2259 | ||
2260 | ret = update_block_group(trans, extent_root, | |
2261 | extent_op->bytenr, | |
2262 | extent_op->num_bytes, | |
2263 | 0, ret > 0); | |
2264 | BUG_ON(ret); | |
2265 | kfree(extent_op); | |
2266 | num_inserts--; | |
d8d5f3e1 | 2267 | } |
f3465ca4 JB |
2268 | } |
2269 | mutex_unlock(&info->extent_ins_mutex); | |
31840ae1 | 2270 | |
f3465ca4 JB |
2271 | ret = insert_extents(trans, extent_root, path, &insert_list, |
2272 | num_inserts); | |
2273 | BUG_ON(ret); | |
2274 | ||
2275 | /* | |
2276 | * if we broke out of the loop in order to insert stuff because we hit | |
2277 | * the maximum number of inserts at a time we can handle, then loop | |
2278 | * back and pick up where we left off | |
2279 | */ | |
2280 | if (num_inserts == max_inserts) { | |
2281 | INIT_LIST_HEAD(&insert_list); | |
2282 | INIT_LIST_HEAD(&update_list); | |
2283 | num_inserts = 0; | |
2284 | goto again; | |
2285 | } | |
2286 | ||
2287 | /* | |
2288 | * again, if we need to make absolutely sure there are no more pending | |
2289 | * extent operations left and we know that we skipped some, go back to | |
2290 | * the beginning and do it all again | |
2291 | */ | |
2292 | if (all && skipped) { | |
2293 | INIT_LIST_HEAD(&insert_list); | |
2294 | INIT_LIST_HEAD(&update_list); | |
2295 | search = 0; | |
2296 | skipped = 0; | |
2297 | num_inserts = 0; | |
2298 | goto again; | |
037e6390 | 2299 | } |
f3465ca4 | 2300 | out: |
7bb86316 | 2301 | btrfs_free_path(path); |
037e6390 CM |
2302 | return 0; |
2303 | } | |
2304 | ||
31840ae1 ZY |
2305 | static int pin_down_bytes(struct btrfs_trans_handle *trans, |
2306 | struct btrfs_root *root, | |
2307 | u64 bytenr, u64 num_bytes, int is_data) | |
e20d96d6 | 2308 | { |
1a5bc167 | 2309 | int err = 0; |
31840ae1 | 2310 | struct extent_buffer *buf; |
8ef97622 | 2311 | |
31840ae1 ZY |
2312 | if (is_data) |
2313 | goto pinit; | |
2314 | ||
2315 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); | |
2316 | if (!buf) | |
2317 | goto pinit; | |
2318 | ||
2319 | /* we can reuse a block if it hasn't been written | |
2320 | * and it is from this transaction. We can't | |
2321 | * reuse anything from the tree log root because | |
2322 | * it has tiny sub-transactions. | |
2323 | */ | |
2324 | if (btrfs_buffer_uptodate(buf, 0) && | |
2325 | btrfs_try_tree_lock(buf)) { | |
2326 | u64 header_owner = btrfs_header_owner(buf); | |
2327 | u64 header_transid = btrfs_header_generation(buf); | |
2328 | if (header_owner != BTRFS_TREE_LOG_OBJECTID && | |
1a40e23b | 2329 | header_owner != BTRFS_TREE_RELOC_OBJECTID && |
31840ae1 ZY |
2330 | header_transid == trans->transid && |
2331 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
2332 | clean_tree_block(NULL, root, buf); | |
2333 | btrfs_tree_unlock(buf); | |
5f39d397 | 2334 | free_extent_buffer(buf); |
31840ae1 | 2335 | return 1; |
8ef97622 | 2336 | } |
31840ae1 | 2337 | btrfs_tree_unlock(buf); |
f4b9aa8d | 2338 | } |
31840ae1 ZY |
2339 | free_extent_buffer(buf); |
2340 | pinit: | |
2341 | btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); | |
2342 | ||
be744175 | 2343 | BUG_ON(err < 0); |
e20d96d6 CM |
2344 | return 0; |
2345 | } | |
2346 | ||
fec577fb | 2347 | /* |
a28ec197 | 2348 | * remove an extent from the root, returns 0 on success |
fec577fb | 2349 | */ |
31840ae1 ZY |
2350 | static int __free_extent(struct btrfs_trans_handle *trans, |
2351 | struct btrfs_root *root, | |
2352 | u64 bytenr, u64 num_bytes, u64 parent, | |
7bb86316 | 2353 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 2354 | u64 owner_objectid, int pin, int mark_free) |
a28ec197 | 2355 | { |
5caf2a00 | 2356 | struct btrfs_path *path; |
e2fa7227 | 2357 | struct btrfs_key key; |
1261ec42 CM |
2358 | struct btrfs_fs_info *info = root->fs_info; |
2359 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 2360 | struct extent_buffer *leaf; |
a28ec197 | 2361 | int ret; |
952fccac CM |
2362 | int extent_slot = 0; |
2363 | int found_extent = 0; | |
2364 | int num_to_del = 1; | |
234b63a0 | 2365 | struct btrfs_extent_item *ei; |
cf27e1ee | 2366 | u32 refs; |
037e6390 | 2367 | |
db94535d | 2368 | key.objectid = bytenr; |
62e2749e | 2369 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 2370 | key.offset = num_bytes; |
5caf2a00 | 2371 | path = btrfs_alloc_path(); |
54aa1f4d CM |
2372 | if (!path) |
2373 | return -ENOMEM; | |
5f26f772 | 2374 | |
3c12ac72 | 2375 | path->reada = 1; |
3bb1a1bc YZ |
2376 | ret = lookup_extent_backref(trans, extent_root, path, |
2377 | bytenr, parent, root_objectid, | |
2378 | ref_generation, owner_objectid, 1); | |
7bb86316 | 2379 | if (ret == 0) { |
952fccac CM |
2380 | struct btrfs_key found_key; |
2381 | extent_slot = path->slots[0]; | |
2382 | while(extent_slot > 0) { | |
2383 | extent_slot--; | |
2384 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2385 | extent_slot); | |
2386 | if (found_key.objectid != bytenr) | |
2387 | break; | |
2388 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && | |
2389 | found_key.offset == num_bytes) { | |
2390 | found_extent = 1; | |
2391 | break; | |
2392 | } | |
2393 | if (path->slots[0] - extent_slot > 5) | |
2394 | break; | |
2395 | } | |
31840ae1 ZY |
2396 | if (!found_extent) { |
2397 | ret = remove_extent_backref(trans, extent_root, path); | |
2398 | BUG_ON(ret); | |
2399 | btrfs_release_path(extent_root, path); | |
2400 | ret = btrfs_search_slot(trans, extent_root, | |
2401 | &key, path, -1, 1); | |
f3465ca4 JB |
2402 | if (ret) { |
2403 | printk(KERN_ERR "umm, got %d back from search" | |
2404 | ", was looking for %Lu\n", ret, | |
2405 | bytenr); | |
2406 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
2407 | } | |
31840ae1 ZY |
2408 | BUG_ON(ret); |
2409 | extent_slot = path->slots[0]; | |
2410 | } | |
7bb86316 CM |
2411 | } else { |
2412 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
2413 | WARN_ON(1); | |
2414 | printk("Unable to find ref byte nr %Lu root %Lu " | |
3bb1a1bc YZ |
2415 | "gen %Lu owner %Lu\n", bytenr, |
2416 | root_objectid, ref_generation, owner_objectid); | |
7bb86316 | 2417 | } |
5f39d397 CM |
2418 | |
2419 | leaf = path->nodes[0]; | |
952fccac | 2420 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 2421 | struct btrfs_extent_item); |
5f39d397 CM |
2422 | refs = btrfs_extent_refs(leaf, ei); |
2423 | BUG_ON(refs == 0); | |
2424 | refs -= 1; | |
2425 | btrfs_set_extent_refs(leaf, ei, refs); | |
952fccac | 2426 | |
5f39d397 CM |
2427 | btrfs_mark_buffer_dirty(leaf); |
2428 | ||
952fccac | 2429 | if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { |
31840ae1 ZY |
2430 | struct btrfs_extent_ref *ref; |
2431 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
2432 | struct btrfs_extent_ref); | |
2433 | BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1); | |
952fccac CM |
2434 | /* if the back ref and the extent are next to each other |
2435 | * they get deleted below in one shot | |
2436 | */ | |
2437 | path->slots[0] = extent_slot; | |
2438 | num_to_del = 2; | |
2439 | } else if (found_extent) { | |
2440 | /* otherwise delete the extent back ref */ | |
31840ae1 | 2441 | ret = remove_extent_backref(trans, extent_root, path); |
952fccac CM |
2442 | BUG_ON(ret); |
2443 | /* if refs are 0, we need to setup the path for deletion */ | |
2444 | if (refs == 0) { | |
2445 | btrfs_release_path(extent_root, path); | |
2446 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
2447 | -1, 1); | |
952fccac CM |
2448 | BUG_ON(ret); |
2449 | } | |
2450 | } | |
2451 | ||
cf27e1ee | 2452 | if (refs == 0) { |
db94535d CM |
2453 | u64 super_used; |
2454 | u64 root_used; | |
21af804c DW |
2455 | #ifdef BIO_RW_DISCARD |
2456 | u64 map_length = num_bytes; | |
2457 | struct btrfs_multi_bio *multi = NULL; | |
2458 | #endif | |
78fae27e CM |
2459 | |
2460 | if (pin) { | |
25179201 | 2461 | mutex_lock(&root->fs_info->pinned_mutex); |
31840ae1 ZY |
2462 | ret = pin_down_bytes(trans, root, bytenr, num_bytes, |
2463 | owner_objectid >= BTRFS_FIRST_FREE_OBJECTID); | |
25179201 | 2464 | mutex_unlock(&root->fs_info->pinned_mutex); |
c549228f Y |
2465 | if (ret > 0) |
2466 | mark_free = 1; | |
2467 | BUG_ON(ret < 0); | |
78fae27e CM |
2468 | } |
2469 | ||
58176a96 | 2470 | /* block accounting for super block */ |
a2135011 | 2471 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
2472 | super_used = btrfs_super_bytes_used(&info->super_copy); |
2473 | btrfs_set_super_bytes_used(&info->super_copy, | |
2474 | super_used - num_bytes); | |
a2135011 | 2475 | spin_unlock_irq(&info->delalloc_lock); |
58176a96 JB |
2476 | |
2477 | /* block accounting for root item */ | |
db94535d | 2478 | root_used = btrfs_root_used(&root->root_item); |
5f39d397 | 2479 | btrfs_set_root_used(&root->root_item, |
db94535d | 2480 | root_used - num_bytes); |
952fccac CM |
2481 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
2482 | num_to_del); | |
31840ae1 | 2483 | BUG_ON(ret); |
25179201 | 2484 | btrfs_release_path(extent_root, path); |
db94535d | 2485 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
0b86a832 | 2486 | mark_free); |
9078a3e1 | 2487 | BUG_ON(ret); |
21af804c DW |
2488 | |
2489 | #ifdef BIO_RW_DISCARD | |
2490 | /* Tell the block device(s) that the sectors can be discarded */ | |
2491 | ret = btrfs_map_block(&root->fs_info->mapping_tree, READ, | |
2492 | bytenr, &map_length, &multi, 0); | |
2493 | if (!ret) { | |
2494 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
2495 | int i; | |
2496 | ||
2497 | if (map_length > num_bytes) | |
2498 | map_length = num_bytes; | |
2499 | ||
2500 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
2501 | blkdev_issue_discard(stripe->dev->bdev, | |
2502 | stripe->physical >> 9, | |
2503 | map_length >> 9); | |
2504 | } | |
2505 | kfree(multi); | |
2506 | } | |
2507 | #endif | |
a28ec197 | 2508 | } |
5caf2a00 | 2509 | btrfs_free_path(path); |
87ef2bb4 | 2510 | finish_current_insert(trans, extent_root, 0); |
a28ec197 CM |
2511 | return ret; |
2512 | } | |
2513 | ||
a28ec197 CM |
2514 | /* |
2515 | * find all the blocks marked as pending in the radix tree and remove | |
2516 | * them from the extent map | |
2517 | */ | |
e089f05c | 2518 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
87ef2bb4 | 2519 | btrfs_root *extent_root, int all) |
a28ec197 CM |
2520 | { |
2521 | int ret; | |
e20d96d6 | 2522 | int err = 0; |
1a5bc167 CM |
2523 | u64 start; |
2524 | u64 end; | |
31840ae1 | 2525 | u64 priv; |
25179201 | 2526 | u64 search = 0; |
f3465ca4 | 2527 | int nr = 0, skipped = 0; |
d1310b2e | 2528 | struct extent_io_tree *pending_del; |
31840ae1 ZY |
2529 | struct extent_io_tree *extent_ins; |
2530 | struct pending_extent_op *extent_op; | |
25179201 | 2531 | struct btrfs_fs_info *info = extent_root->fs_info; |
f3465ca4 | 2532 | struct list_head delete_list; |
8ef97622 | 2533 | |
f3465ca4 | 2534 | INIT_LIST_HEAD(&delete_list); |
31840ae1 | 2535 | extent_ins = &extent_root->fs_info->extent_ins; |
1a5bc167 | 2536 | pending_del = &extent_root->fs_info->pending_del; |
a28ec197 | 2537 | |
f3465ca4 JB |
2538 | again: |
2539 | mutex_lock(&info->extent_ins_mutex); | |
a28ec197 | 2540 | while(1) { |
25179201 JB |
2541 | ret = find_first_extent_bit(pending_del, search, &start, &end, |
2542 | EXTENT_WRITEBACK); | |
2543 | if (ret) { | |
f3465ca4 | 2544 | if (all && skipped && !nr) { |
25179201 JB |
2545 | search = 0; |
2546 | continue; | |
2547 | } | |
f3465ca4 | 2548 | mutex_unlock(&info->extent_ins_mutex); |
a28ec197 | 2549 | break; |
25179201 JB |
2550 | } |
2551 | ||
2552 | ret = try_lock_extent(extent_ins, start, end, GFP_NOFS); | |
2553 | if (!ret) { | |
2554 | search = end+1; | |
f3465ca4 JB |
2555 | skipped = 1; |
2556 | ||
2557 | if (need_resched()) { | |
2558 | mutex_unlock(&info->extent_ins_mutex); | |
2559 | cond_resched(); | |
2560 | mutex_lock(&info->extent_ins_mutex); | |
2561 | } | |
2562 | ||
25179201 JB |
2563 | continue; |
2564 | } | |
2565 | BUG_ON(ret < 0); | |
31840ae1 ZY |
2566 | |
2567 | ret = get_state_private(pending_del, start, &priv); | |
2568 | BUG_ON(ret); | |
2569 | extent_op = (struct pending_extent_op *)(unsigned long)priv; | |
2570 | ||
25179201 | 2571 | clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK, |
1a5bc167 | 2572 | GFP_NOFS); |
31840ae1 | 2573 | if (!test_range_bit(extent_ins, start, end, |
25179201 | 2574 | EXTENT_WRITEBACK, 0)) { |
f3465ca4 JB |
2575 | list_add_tail(&extent_op->list, &delete_list); |
2576 | nr++; | |
c286ac48 | 2577 | } else { |
31840ae1 | 2578 | kfree(extent_op); |
25179201 JB |
2579 | |
2580 | ret = get_state_private(&info->extent_ins, start, | |
2581 | &priv); | |
31840ae1 ZY |
2582 | BUG_ON(ret); |
2583 | extent_op = (struct pending_extent_op *) | |
25179201 JB |
2584 | (unsigned long)priv; |
2585 | ||
2586 | clear_extent_bits(&info->extent_ins, start, end, | |
2587 | EXTENT_WRITEBACK, GFP_NOFS); | |
31840ae1 | 2588 | |
f3465ca4 JB |
2589 | if (extent_op->type == PENDING_BACKREF_UPDATE) { |
2590 | list_add_tail(&extent_op->list, &delete_list); | |
2591 | search = end + 1; | |
2592 | nr++; | |
2593 | continue; | |
2594 | } | |
31840ae1 | 2595 | |
25179201 JB |
2596 | mutex_lock(&extent_root->fs_info->pinned_mutex); |
2597 | ret = pin_down_bytes(trans, extent_root, start, | |
2598 | end + 1 - start, 0); | |
2599 | mutex_unlock(&extent_root->fs_info->pinned_mutex); | |
2600 | ||
31840ae1 | 2601 | ret = update_block_group(trans, extent_root, start, |
25179201 JB |
2602 | end + 1 - start, 0, ret > 0); |
2603 | ||
f3465ca4 | 2604 | unlock_extent(extent_ins, start, end, GFP_NOFS); |
31840ae1 ZY |
2605 | BUG_ON(ret); |
2606 | kfree(extent_op); | |
c286ac48 | 2607 | } |
1a5bc167 CM |
2608 | if (ret) |
2609 | err = ret; | |
c286ac48 | 2610 | |
f3465ca4 JB |
2611 | search = end + 1; |
2612 | ||
2613 | if (need_resched()) { | |
2614 | mutex_unlock(&info->extent_ins_mutex); | |
2615 | cond_resched(); | |
2616 | mutex_lock(&info->extent_ins_mutex); | |
2617 | } | |
2618 | } | |
2619 | ||
2620 | if (nr) { | |
2621 | ret = free_extents(trans, extent_root, &delete_list); | |
2622 | BUG_ON(ret); | |
2623 | } | |
2624 | ||
2625 | if (all && skipped) { | |
2626 | INIT_LIST_HEAD(&delete_list); | |
2627 | search = 0; | |
2628 | nr = 0; | |
2629 | goto again; | |
fec577fb | 2630 | } |
f3465ca4 | 2631 | |
e20d96d6 | 2632 | return err; |
fec577fb CM |
2633 | } |
2634 | ||
2635 | /* | |
2636 | * remove an extent from the root, returns 0 on success | |
2637 | */ | |
925baedd | 2638 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
2639 | struct btrfs_root *root, |
2640 | u64 bytenr, u64 num_bytes, u64 parent, | |
2641 | u64 root_objectid, u64 ref_generation, | |
3bb1a1bc | 2642 | u64 owner_objectid, int pin) |
fec577fb | 2643 | { |
9f5fae2f | 2644 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
fec577fb CM |
2645 | int pending_ret; |
2646 | int ret; | |
a28ec197 | 2647 | |
db94535d | 2648 | WARN_ON(num_bytes < root->sectorsize); |
fec577fb | 2649 | if (root == extent_root) { |
f3465ca4 JB |
2650 | struct pending_extent_op *extent_op = NULL; |
2651 | ||
2652 | mutex_lock(&root->fs_info->extent_ins_mutex); | |
2653 | if (test_range_bit(&root->fs_info->extent_ins, bytenr, | |
2654 | bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) { | |
2655 | u64 priv; | |
2656 | ret = get_state_private(&root->fs_info->extent_ins, | |
2657 | bytenr, &priv); | |
2658 | BUG_ON(ret); | |
2659 | extent_op = (struct pending_extent_op *) | |
2660 | (unsigned long)priv; | |
2661 | ||
2662 | extent_op->del = 1; | |
2663 | if (extent_op->type == PENDING_EXTENT_INSERT) { | |
2664 | mutex_unlock(&root->fs_info->extent_ins_mutex); | |
2665 | return 0; | |
2666 | } | |
2667 | } | |
2668 | ||
2669 | if (extent_op) { | |
2670 | ref_generation = extent_op->orig_generation; | |
2671 | parent = extent_op->orig_parent; | |
2672 | } | |
31840ae1 ZY |
2673 | |
2674 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2675 | BUG_ON(!extent_op); | |
2676 | ||
2677 | extent_op->type = PENDING_EXTENT_DELETE; | |
2678 | extent_op->bytenr = bytenr; | |
2679 | extent_op->num_bytes = num_bytes; | |
2680 | extent_op->parent = parent; | |
2681 | extent_op->orig_parent = parent; | |
2682 | extent_op->generation = ref_generation; | |
2683 | extent_op->orig_generation = ref_generation; | |
2684 | extent_op->level = (int)owner_objectid; | |
f3465ca4 JB |
2685 | INIT_LIST_HEAD(&extent_op->list); |
2686 | extent_op->del = 0; | |
31840ae1 ZY |
2687 | |
2688 | set_extent_bits(&root->fs_info->pending_del, | |
2689 | bytenr, bytenr + num_bytes - 1, | |
25179201 | 2690 | EXTENT_WRITEBACK, GFP_NOFS); |
31840ae1 ZY |
2691 | set_state_private(&root->fs_info->pending_del, |
2692 | bytenr, (unsigned long)extent_op); | |
25179201 | 2693 | mutex_unlock(&root->fs_info->extent_ins_mutex); |
fec577fb CM |
2694 | return 0; |
2695 | } | |
4bef0848 | 2696 | /* if metadata always pin */ |
d00aff00 CM |
2697 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
2698 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { | |
0f9dd46c JB |
2699 | struct btrfs_block_group_cache *cache; |
2700 | ||
d00aff00 | 2701 | /* btrfs_free_reserved_extent */ |
0f9dd46c JB |
2702 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
2703 | BUG_ON(!cache); | |
2704 | btrfs_add_free_space(cache, bytenr, num_bytes); | |
e8569813 | 2705 | update_reserved_extents(root, bytenr, num_bytes, 0); |
d00aff00 CM |
2706 | return 0; |
2707 | } | |
4bef0848 | 2708 | pin = 1; |
d00aff00 | 2709 | } |
4bef0848 CM |
2710 | |
2711 | /* if data pin when any transaction has committed this */ | |
2712 | if (ref_generation != trans->transid) | |
2713 | pin = 1; | |
2714 | ||
31840ae1 | 2715 | ret = __free_extent(trans, root, bytenr, num_bytes, parent, |
3bb1a1bc YZ |
2716 | root_objectid, ref_generation, |
2717 | owner_objectid, pin, pin == 0); | |
ee6e6504 | 2718 | |
87ef2bb4 CM |
2719 | finish_current_insert(trans, root->fs_info->extent_root, 0); |
2720 | pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0); | |
fec577fb CM |
2721 | return ret ? ret : pending_ret; |
2722 | } | |
2723 | ||
925baedd | 2724 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
2725 | struct btrfs_root *root, |
2726 | u64 bytenr, u64 num_bytes, u64 parent, | |
2727 | u64 root_objectid, u64 ref_generation, | |
3bb1a1bc | 2728 | u64 owner_objectid, int pin) |
925baedd CM |
2729 | { |
2730 | int ret; | |
2731 | ||
31840ae1 | 2732 | ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent, |
925baedd | 2733 | root_objectid, ref_generation, |
3bb1a1bc | 2734 | owner_objectid, pin); |
925baedd CM |
2735 | return ret; |
2736 | } | |
2737 | ||
87ee04eb CM |
2738 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
2739 | { | |
2740 | u64 mask = ((u64)root->stripesize - 1); | |
2741 | u64 ret = (val + mask) & ~mask; | |
2742 | return ret; | |
2743 | } | |
2744 | ||
fec577fb CM |
2745 | /* |
2746 | * walks the btree of allocated extents and find a hole of a given size. | |
2747 | * The key ins is changed to record the hole: | |
2748 | * ins->objectid == block start | |
62e2749e | 2749 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
2750 | * ins->offset == number of blocks |
2751 | * Any available blocks before search_start are skipped. | |
2752 | */ | |
98ed5174 CM |
2753 | static int noinline find_free_extent(struct btrfs_trans_handle *trans, |
2754 | struct btrfs_root *orig_root, | |
2755 | u64 num_bytes, u64 empty_size, | |
2756 | u64 search_start, u64 search_end, | |
2757 | u64 hint_byte, struct btrfs_key *ins, | |
2758 | u64 exclude_start, u64 exclude_nr, | |
2759 | int data) | |
fec577fb | 2760 | { |
80eb234a | 2761 | int ret = 0; |
9f5fae2f | 2762 | struct btrfs_root * root = orig_root->fs_info->extent_root; |
db94535d | 2763 | u64 total_needed = num_bytes; |
239b14b3 | 2764 | u64 *last_ptr = NULL; |
4366211c | 2765 | u64 last_wanted = 0; |
80eb234a | 2766 | struct btrfs_block_group_cache *block_group = NULL; |
0ef3e66b | 2767 | int chunk_alloc_done = 0; |
239b14b3 | 2768 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 2769 | int allowed_chunk_alloc = 0; |
80eb234a JB |
2770 | struct list_head *head = NULL, *cur = NULL; |
2771 | int loop = 0; | |
f5a31e16 | 2772 | int extra_loop = 0; |
80eb234a | 2773 | struct btrfs_space_info *space_info; |
fec577fb | 2774 | |
db94535d | 2775 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 2776 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
2777 | ins->objectid = 0; |
2778 | ins->offset = 0; | |
b1a4d965 | 2779 | |
0ef3e66b CM |
2780 | if (orig_root->ref_cows || empty_size) |
2781 | allowed_chunk_alloc = 1; | |
2782 | ||
239b14b3 CM |
2783 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
2784 | last_ptr = &root->fs_info->last_alloc; | |
4366211c | 2785 | empty_cluster = 64 * 1024; |
239b14b3 CM |
2786 | } |
2787 | ||
0f9dd46c | 2788 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) |
239b14b3 | 2789 | last_ptr = &root->fs_info->last_data_alloc; |
0f9dd46c | 2790 | |
239b14b3 | 2791 | if (last_ptr) { |
4366211c | 2792 | if (*last_ptr) { |
239b14b3 | 2793 | hint_byte = *last_ptr; |
4366211c CM |
2794 | last_wanted = *last_ptr; |
2795 | } else | |
239b14b3 | 2796 | empty_size += empty_cluster; |
4366211c CM |
2797 | } else { |
2798 | empty_cluster = 0; | |
239b14b3 | 2799 | } |
a061fc8d | 2800 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 2801 | search_start = max(search_start, hint_byte); |
0b86a832 | 2802 | |
42e70e7a | 2803 | if (last_wanted && search_start != last_wanted) { |
4366211c | 2804 | last_wanted = 0; |
42e70e7a CM |
2805 | empty_size += empty_cluster; |
2806 | } | |
4366211c | 2807 | |
42e70e7a | 2808 | total_needed += empty_size; |
80eb234a | 2809 | block_group = btrfs_lookup_block_group(root->fs_info, search_start); |
d899e052 YZ |
2810 | if (!block_group) |
2811 | block_group = btrfs_lookup_first_block_group(root->fs_info, | |
2812 | search_start); | |
80eb234a | 2813 | space_info = __find_space_info(root->fs_info, data); |
0f9dd46c | 2814 | |
80eb234a JB |
2815 | down_read(&space_info->groups_sem); |
2816 | while (1) { | |
2817 | struct btrfs_free_space *free_space; | |
0f9dd46c | 2818 | /* |
80eb234a JB |
2819 | * the only way this happens if our hint points to a block |
2820 | * group thats not of the proper type, while looping this | |
2821 | * should never happen | |
0f9dd46c | 2822 | */ |
8a1413a2 CM |
2823 | if (empty_size) |
2824 | extra_loop = 1; | |
2825 | ||
42e70e7a CM |
2826 | if (!block_group) |
2827 | goto new_group_no_lock; | |
2828 | ||
25179201 | 2829 | mutex_lock(&block_group->alloc_mutex); |
80eb234a | 2830 | if (unlikely(!block_group_bits(block_group, data))) |
0f9dd46c | 2831 | goto new_group; |
0f9dd46c | 2832 | |
80eb234a | 2833 | ret = cache_block_group(root, block_group); |
25179201 JB |
2834 | if (ret) { |
2835 | mutex_unlock(&block_group->alloc_mutex); | |
80eb234a | 2836 | break; |
25179201 | 2837 | } |
0f9dd46c | 2838 | |
80eb234a JB |
2839 | if (block_group->ro) |
2840 | goto new_group; | |
0f9dd46c | 2841 | |
80eb234a JB |
2842 | free_space = btrfs_find_free_space(block_group, search_start, |
2843 | total_needed); | |
2844 | if (free_space) { | |
2845 | u64 start = block_group->key.objectid; | |
2846 | u64 end = block_group->key.objectid + | |
0f9dd46c | 2847 | block_group->key.offset; |
239b14b3 | 2848 | |
80eb234a | 2849 | search_start = stripe_align(root, free_space->offset); |
0f9dd46c | 2850 | |
80eb234a JB |
2851 | /* move on to the next group */ |
2852 | if (search_start + num_bytes >= search_end) | |
2853 | goto new_group; | |
e37c9e69 | 2854 | |
80eb234a JB |
2855 | /* move on to the next group */ |
2856 | if (search_start + num_bytes > end) | |
2857 | goto new_group; | |
2858 | ||
4366211c | 2859 | if (last_wanted && search_start != last_wanted) { |
3b7885bf | 2860 | total_needed += empty_cluster; |
8a1413a2 | 2861 | empty_size += empty_cluster; |
4366211c | 2862 | last_wanted = 0; |
3b7885bf CM |
2863 | /* |
2864 | * if search_start is still in this block group | |
2865 | * then we just re-search this block group | |
2866 | */ | |
2867 | if (search_start >= start && | |
2868 | search_start < end) { | |
2869 | mutex_unlock(&block_group->alloc_mutex); | |
2870 | continue; | |
2871 | } | |
2872 | ||
2873 | /* else we go to the next block group */ | |
2874 | goto new_group; | |
2875 | } | |
2876 | ||
80eb234a JB |
2877 | if (exclude_nr > 0 && |
2878 | (search_start + num_bytes > exclude_start && | |
2879 | search_start < exclude_start + exclude_nr)) { | |
2880 | search_start = exclude_start + exclude_nr; | |
2881 | /* | |
2882 | * if search_start is still in this block group | |
2883 | * then we just re-search this block group | |
2884 | */ | |
2885 | if (search_start >= start && | |
25179201 JB |
2886 | search_start < end) { |
2887 | mutex_unlock(&block_group->alloc_mutex); | |
4366211c | 2888 | last_wanted = 0; |
80eb234a | 2889 | continue; |
25179201 | 2890 | } |
80eb234a JB |
2891 | |
2892 | /* else we go to the next block group */ | |
2893 | goto new_group; | |
2894 | } | |
2895 | ||
2896 | ins->objectid = search_start; | |
2897 | ins->offset = num_bytes; | |
25179201 JB |
2898 | |
2899 | btrfs_remove_free_space_lock(block_group, search_start, | |
2900 | num_bytes); | |
80eb234a | 2901 | /* we are all good, lets return */ |
25179201 | 2902 | mutex_unlock(&block_group->alloc_mutex); |
80eb234a | 2903 | break; |
0f9dd46c | 2904 | } |
80eb234a | 2905 | new_group: |
42e70e7a CM |
2906 | mutex_unlock(&block_group->alloc_mutex); |
2907 | new_group_no_lock: | |
f5a31e16 CM |
2908 | /* don't try to compare new allocations against the |
2909 | * last allocation any more | |
2910 | */ | |
4366211c | 2911 | last_wanted = 0; |
f5a31e16 | 2912 | |
80eb234a JB |
2913 | /* |
2914 | * Here's how this works. | |
2915 | * loop == 0: we were searching a block group via a hint | |
2916 | * and didn't find anything, so we start at | |
2917 | * the head of the block groups and keep searching | |
2918 | * loop == 1: we're searching through all of the block groups | |
2919 | * if we hit the head again we have searched | |
2920 | * all of the block groups for this space and we | |
2921 | * need to try and allocate, if we cant error out. | |
2922 | * loop == 2: we allocated more space and are looping through | |
2923 | * all of the block groups again. | |
2924 | */ | |
2925 | if (loop == 0) { | |
2926 | head = &space_info->block_groups; | |
2927 | cur = head->next; | |
80eb234a JB |
2928 | loop++; |
2929 | } else if (loop == 1 && cur == head) { | |
f5a31e16 CM |
2930 | int keep_going; |
2931 | ||
2932 | /* at this point we give up on the empty_size | |
2933 | * allocations and just try to allocate the min | |
2934 | * space. | |
2935 | * | |
2936 | * The extra_loop field was set if an empty_size | |
2937 | * allocation was attempted above, and if this | |
2938 | * is try we need to try the loop again without | |
2939 | * the additional empty_size. | |
2940 | */ | |
5b7c3fcc CM |
2941 | total_needed -= empty_size; |
2942 | empty_size = 0; | |
f5a31e16 CM |
2943 | keep_going = extra_loop; |
2944 | loop++; | |
42e70e7a | 2945 | |
80eb234a JB |
2946 | if (allowed_chunk_alloc && !chunk_alloc_done) { |
2947 | up_read(&space_info->groups_sem); | |
2948 | ret = do_chunk_alloc(trans, root, num_bytes + | |
2949 | 2 * 1024 * 1024, data, 1); | |
80eb234a | 2950 | down_read(&space_info->groups_sem); |
2ed6d664 CM |
2951 | if (ret < 0) |
2952 | goto loop_check; | |
80eb234a | 2953 | head = &space_info->block_groups; |
f5a31e16 CM |
2954 | /* |
2955 | * we've allocated a new chunk, keep | |
2956 | * trying | |
2957 | */ | |
2958 | keep_going = 1; | |
80eb234a JB |
2959 | chunk_alloc_done = 1; |
2960 | } else if (!allowed_chunk_alloc) { | |
2961 | space_info->force_alloc = 1; | |
f5a31e16 | 2962 | } |
2ed6d664 | 2963 | loop_check: |
f5a31e16 CM |
2964 | if (keep_going) { |
2965 | cur = head->next; | |
2966 | extra_loop = 0; | |
80eb234a JB |
2967 | } else { |
2968 | break; | |
2969 | } | |
2970 | } else if (cur == head) { | |
2971 | break; | |
0f9dd46c | 2972 | } |
0b86a832 | 2973 | |
80eb234a JB |
2974 | block_group = list_entry(cur, struct btrfs_block_group_cache, |
2975 | list); | |
2976 | search_start = block_group->key.objectid; | |
2977 | cur = cur->next; | |
f2654de4 | 2978 | } |
0b86a832 | 2979 | |
80eb234a JB |
2980 | /* we found what we needed */ |
2981 | if (ins->objectid) { | |
2982 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) | |
2983 | trans->block_group = block_group; | |
0f9dd46c | 2984 | |
80eb234a JB |
2985 | if (last_ptr) |
2986 | *last_ptr = ins->objectid + ins->offset; | |
2987 | ret = 0; | |
2988 | } else if (!ret) { | |
4ce4cb52 JB |
2989 | printk(KERN_ERR "we were searching for %Lu bytes, num_bytes %Lu," |
2990 | " loop %d, allowed_alloc %d\n", total_needed, num_bytes, | |
2991 | loop, allowed_chunk_alloc); | |
80eb234a | 2992 | ret = -ENOSPC; |
be744175 | 2993 | } |
be744175 | 2994 | |
80eb234a | 2995 | up_read(&space_info->groups_sem); |
0f70abe2 | 2996 | return ret; |
fec577fb | 2997 | } |
ec44a35c | 2998 | |
0f9dd46c JB |
2999 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes) |
3000 | { | |
3001 | struct btrfs_block_group_cache *cache; | |
3002 | struct list_head *l; | |
3003 | ||
3004 | printk(KERN_INFO "space_info has %Lu free, is %sfull\n", | |
e8569813 ZY |
3005 | info->total_bytes - info->bytes_used - info->bytes_pinned - |
3006 | info->bytes_reserved, (info->full) ? "" : "not "); | |
0f9dd46c | 3007 | |
80eb234a | 3008 | down_read(&info->groups_sem); |
0f9dd46c JB |
3009 | list_for_each(l, &info->block_groups) { |
3010 | cache = list_entry(l, struct btrfs_block_group_cache, list); | |
3011 | spin_lock(&cache->lock); | |
3012 | printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used " | |
e8569813 | 3013 | "%Lu pinned %Lu reserved\n", |
0f9dd46c | 3014 | cache->key.objectid, cache->key.offset, |
e8569813 ZY |
3015 | btrfs_block_group_used(&cache->item), |
3016 | cache->pinned, cache->reserved); | |
0f9dd46c JB |
3017 | btrfs_dump_free_space(cache, bytes); |
3018 | spin_unlock(&cache->lock); | |
3019 | } | |
80eb234a | 3020 | up_read(&info->groups_sem); |
0f9dd46c | 3021 | } |
e8569813 | 3022 | |
e6dcd2dc CM |
3023 | static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
3024 | struct btrfs_root *root, | |
3025 | u64 num_bytes, u64 min_alloc_size, | |
3026 | u64 empty_size, u64 hint_byte, | |
3027 | u64 search_end, struct btrfs_key *ins, | |
3028 | u64 data) | |
fec577fb CM |
3029 | { |
3030 | int ret; | |
fbdc762b | 3031 | u64 search_start = 0; |
8790d502 | 3032 | u64 alloc_profile; |
1261ec42 | 3033 | struct btrfs_fs_info *info = root->fs_info; |
925baedd | 3034 | |
6324fbf3 | 3035 | if (data) { |
8790d502 CM |
3036 | alloc_profile = info->avail_data_alloc_bits & |
3037 | info->data_alloc_profile; | |
3038 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
6324fbf3 | 3039 | } else if (root == root->fs_info->chunk_root) { |
8790d502 CM |
3040 | alloc_profile = info->avail_system_alloc_bits & |
3041 | info->system_alloc_profile; | |
3042 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
6324fbf3 | 3043 | } else { |
8790d502 CM |
3044 | alloc_profile = info->avail_metadata_alloc_bits & |
3045 | info->metadata_alloc_profile; | |
3046 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
6324fbf3 | 3047 | } |
98d20f67 | 3048 | again: |
2b82032c | 3049 | data = btrfs_reduce_alloc_profile(root, data); |
0ef3e66b CM |
3050 | /* |
3051 | * the only place that sets empty_size is btrfs_realloc_node, which | |
3052 | * is not called recursively on allocations | |
3053 | */ | |
3054 | if (empty_size || root->ref_cows) { | |
593060d7 | 3055 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 3056 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b CM |
3057 | 2 * 1024 * 1024, |
3058 | BTRFS_BLOCK_GROUP_METADATA | | |
3059 | (info->metadata_alloc_profile & | |
3060 | info->avail_metadata_alloc_bits), 0); | |
6324fbf3 CM |
3061 | } |
3062 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
0ef3e66b | 3063 | num_bytes + 2 * 1024 * 1024, data, 0); |
6324fbf3 | 3064 | } |
0b86a832 | 3065 | |
db94535d CM |
3066 | WARN_ON(num_bytes < root->sectorsize); |
3067 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
3068 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
3069 | trans->alloc_exclude_start, |
3070 | trans->alloc_exclude_nr, data); | |
3b951516 | 3071 | |
98d20f67 CM |
3072 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
3073 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 3074 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 3075 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b CM |
3076 | do_chunk_alloc(trans, root->fs_info->extent_root, |
3077 | num_bytes, data, 1); | |
98d20f67 CM |
3078 | goto again; |
3079 | } | |
ec44a35c | 3080 | if (ret) { |
0f9dd46c JB |
3081 | struct btrfs_space_info *sinfo; |
3082 | ||
3083 | sinfo = __find_space_info(root->fs_info, data); | |
3084 | printk("allocation failed flags %Lu, wanted %Lu\n", | |
3085 | data, num_bytes); | |
3086 | dump_space_info(sinfo, num_bytes); | |
925baedd | 3087 | BUG(); |
925baedd | 3088 | } |
0f9dd46c JB |
3089 | |
3090 | return ret; | |
e6dcd2dc CM |
3091 | } |
3092 | ||
65b51a00 CM |
3093 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
3094 | { | |
0f9dd46c JB |
3095 | struct btrfs_block_group_cache *cache; |
3096 | ||
0f9dd46c JB |
3097 | cache = btrfs_lookup_block_group(root->fs_info, start); |
3098 | if (!cache) { | |
3099 | printk(KERN_ERR "Unable to find block group for %Lu\n", start); | |
0f9dd46c JB |
3100 | return -ENOSPC; |
3101 | } | |
3102 | btrfs_add_free_space(cache, start, len); | |
1a40e23b | 3103 | update_reserved_extents(root, start, len, 0); |
65b51a00 CM |
3104 | return 0; |
3105 | } | |
3106 | ||
e6dcd2dc CM |
3107 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
3108 | struct btrfs_root *root, | |
3109 | u64 num_bytes, u64 min_alloc_size, | |
3110 | u64 empty_size, u64 hint_byte, | |
3111 | u64 search_end, struct btrfs_key *ins, | |
3112 | u64 data) | |
3113 | { | |
3114 | int ret; | |
e6dcd2dc CM |
3115 | ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, |
3116 | empty_size, hint_byte, search_end, ins, | |
3117 | data); | |
e8569813 | 3118 | update_reserved_extents(root, ins->objectid, ins->offset, 1); |
e6dcd2dc CM |
3119 | return ret; |
3120 | } | |
3121 | ||
3122 | static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
31840ae1 | 3123 | struct btrfs_root *root, u64 parent, |
e6dcd2dc | 3124 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 3125 | u64 owner, struct btrfs_key *ins) |
e6dcd2dc CM |
3126 | { |
3127 | int ret; | |
3128 | int pending_ret; | |
3129 | u64 super_used; | |
3130 | u64 root_used; | |
3131 | u64 num_bytes = ins->offset; | |
3132 | u32 sizes[2]; | |
3133 | struct btrfs_fs_info *info = root->fs_info; | |
3134 | struct btrfs_root *extent_root = info->extent_root; | |
3135 | struct btrfs_extent_item *extent_item; | |
3136 | struct btrfs_extent_ref *ref; | |
3137 | struct btrfs_path *path; | |
3138 | struct btrfs_key keys[2]; | |
fec577fb | 3139 | |
31840ae1 ZY |
3140 | if (parent == 0) |
3141 | parent = ins->objectid; | |
3142 | ||
58176a96 | 3143 | /* block accounting for super block */ |
a2135011 | 3144 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
3145 | super_used = btrfs_super_bytes_used(&info->super_copy); |
3146 | btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); | |
a2135011 | 3147 | spin_unlock_irq(&info->delalloc_lock); |
26b8003f | 3148 | |
58176a96 | 3149 | /* block accounting for root item */ |
db94535d CM |
3150 | root_used = btrfs_root_used(&root->root_item); |
3151 | btrfs_set_root_used(&root->root_item, root_used + num_bytes); | |
58176a96 | 3152 | |
26b8003f | 3153 | if (root == extent_root) { |
31840ae1 ZY |
3154 | struct pending_extent_op *extent_op; |
3155 | ||
3156 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
3157 | BUG_ON(!extent_op); | |
3158 | ||
3159 | extent_op->type = PENDING_EXTENT_INSERT; | |
3160 | extent_op->bytenr = ins->objectid; | |
3161 | extent_op->num_bytes = ins->offset; | |
3162 | extent_op->parent = parent; | |
3163 | extent_op->orig_parent = 0; | |
3164 | extent_op->generation = ref_generation; | |
3165 | extent_op->orig_generation = 0; | |
3166 | extent_op->level = (int)owner; | |
f3465ca4 JB |
3167 | INIT_LIST_HEAD(&extent_op->list); |
3168 | extent_op->del = 0; | |
31840ae1 | 3169 | |
25179201 | 3170 | mutex_lock(&root->fs_info->extent_ins_mutex); |
1a5bc167 CM |
3171 | set_extent_bits(&root->fs_info->extent_ins, ins->objectid, |
3172 | ins->objectid + ins->offset - 1, | |
25179201 | 3173 | EXTENT_WRITEBACK, GFP_NOFS); |
31840ae1 ZY |
3174 | set_state_private(&root->fs_info->extent_ins, |
3175 | ins->objectid, (unsigned long)extent_op); | |
25179201 | 3176 | mutex_unlock(&root->fs_info->extent_ins_mutex); |
26b8003f CM |
3177 | goto update_block; |
3178 | } | |
3179 | ||
47e4bb98 | 3180 | memcpy(&keys[0], ins, sizeof(*ins)); |
47e4bb98 CM |
3181 | keys[1].objectid = ins->objectid; |
3182 | keys[1].type = BTRFS_EXTENT_REF_KEY; | |
31840ae1 | 3183 | keys[1].offset = parent; |
47e4bb98 CM |
3184 | sizes[0] = sizeof(*extent_item); |
3185 | sizes[1] = sizeof(*ref); | |
7bb86316 CM |
3186 | |
3187 | path = btrfs_alloc_path(); | |
3188 | BUG_ON(!path); | |
47e4bb98 CM |
3189 | |
3190 | ret = btrfs_insert_empty_items(trans, extent_root, path, keys, | |
3191 | sizes, 2); | |
ccd467d6 | 3192 | BUG_ON(ret); |
0f9dd46c | 3193 | |
47e4bb98 CM |
3194 | extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
3195 | struct btrfs_extent_item); | |
3196 | btrfs_set_extent_refs(path->nodes[0], extent_item, 1); | |
3197 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
3198 | struct btrfs_extent_ref); | |
3199 | ||
3200 | btrfs_set_ref_root(path->nodes[0], ref, root_objectid); | |
3201 | btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); | |
3202 | btrfs_set_ref_objectid(path->nodes[0], ref, owner); | |
31840ae1 | 3203 | btrfs_set_ref_num_refs(path->nodes[0], ref, 1); |
47e4bb98 CM |
3204 | |
3205 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
3206 | ||
3207 | trans->alloc_exclude_start = 0; | |
3208 | trans->alloc_exclude_nr = 0; | |
7bb86316 | 3209 | btrfs_free_path(path); |
87ef2bb4 CM |
3210 | finish_current_insert(trans, extent_root, 0); |
3211 | pending_ret = del_pending_extents(trans, extent_root, 0); | |
f510cfec | 3212 | |
925baedd CM |
3213 | if (ret) |
3214 | goto out; | |
e37c9e69 | 3215 | if (pending_ret) { |
925baedd CM |
3216 | ret = pending_ret; |
3217 | goto out; | |
e37c9e69 | 3218 | } |
26b8003f CM |
3219 | |
3220 | update_block: | |
0b86a832 | 3221 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0); |
f5947066 CM |
3222 | if (ret) { |
3223 | printk("update block group failed for %Lu %Lu\n", | |
3224 | ins->objectid, ins->offset); | |
3225 | BUG(); | |
3226 | } | |
925baedd | 3227 | out: |
e6dcd2dc CM |
3228 | return ret; |
3229 | } | |
3230 | ||
3231 | int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
31840ae1 | 3232 | struct btrfs_root *root, u64 parent, |
e6dcd2dc | 3233 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 3234 | u64 owner, struct btrfs_key *ins) |
e6dcd2dc CM |
3235 | { |
3236 | int ret; | |
1c2308f8 CM |
3237 | |
3238 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) | |
3239 | return 0; | |
3bb1a1bc YZ |
3240 | ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, |
3241 | ref_generation, owner, ins); | |
e8569813 | 3242 | update_reserved_extents(root, ins->objectid, ins->offset, 0); |
e6dcd2dc CM |
3243 | return ret; |
3244 | } | |
e02119d5 CM |
3245 | |
3246 | /* | |
3247 | * this is used by the tree logging recovery code. It records that | |
3248 | * an extent has been allocated and makes sure to clear the free | |
3249 | * space cache bits as well | |
3250 | */ | |
3251 | int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans, | |
31840ae1 | 3252 | struct btrfs_root *root, u64 parent, |
e02119d5 | 3253 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 3254 | u64 owner, struct btrfs_key *ins) |
e02119d5 CM |
3255 | { |
3256 | int ret; | |
3257 | struct btrfs_block_group_cache *block_group; | |
3258 | ||
e02119d5 | 3259 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
25179201 | 3260 | mutex_lock(&block_group->alloc_mutex); |
e02119d5 CM |
3261 | cache_block_group(root, block_group); |
3262 | ||
25179201 JB |
3263 | ret = btrfs_remove_free_space_lock(block_group, ins->objectid, |
3264 | ins->offset); | |
3265 | mutex_unlock(&block_group->alloc_mutex); | |
0f9dd46c | 3266 | BUG_ON(ret); |
3bb1a1bc YZ |
3267 | ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, |
3268 | ref_generation, owner, ins); | |
e02119d5 CM |
3269 | return ret; |
3270 | } | |
3271 | ||
e6dcd2dc CM |
3272 | /* |
3273 | * finds a free extent and does all the dirty work required for allocation | |
3274 | * returns the key for the extent through ins, and a tree buffer for | |
3275 | * the first block of the extent through buf. | |
3276 | * | |
3277 | * returns 0 if everything worked, non-zero otherwise. | |
3278 | */ | |
3279 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, | |
3280 | struct btrfs_root *root, | |
31840ae1 | 3281 | u64 num_bytes, u64 parent, u64 min_alloc_size, |
e6dcd2dc | 3282 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 3283 | u64 owner_objectid, u64 empty_size, u64 hint_byte, |
e6dcd2dc CM |
3284 | u64 search_end, struct btrfs_key *ins, u64 data) |
3285 | { | |
3286 | int ret; | |
3287 | ||
e6dcd2dc CM |
3288 | ret = __btrfs_reserve_extent(trans, root, num_bytes, |
3289 | min_alloc_size, empty_size, hint_byte, | |
3290 | search_end, ins, data); | |
3291 | BUG_ON(ret); | |
d00aff00 | 3292 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { |
31840ae1 ZY |
3293 | ret = __btrfs_alloc_reserved_extent(trans, root, parent, |
3294 | root_objectid, ref_generation, | |
3bb1a1bc | 3295 | owner_objectid, ins); |
d00aff00 | 3296 | BUG_ON(ret); |
e6dcd2dc | 3297 | |
e8569813 ZY |
3298 | } else { |
3299 | update_reserved_extents(root, ins->objectid, ins->offset, 1); | |
d00aff00 | 3300 | } |
925baedd | 3301 | return ret; |
fec577fb | 3302 | } |
65b51a00 CM |
3303 | |
3304 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, | |
3305 | struct btrfs_root *root, | |
3306 | u64 bytenr, u32 blocksize) | |
3307 | { | |
3308 | struct extent_buffer *buf; | |
3309 | ||
3310 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
3311 | if (!buf) | |
3312 | return ERR_PTR(-ENOMEM); | |
3313 | btrfs_set_header_generation(buf, trans->transid); | |
3314 | btrfs_tree_lock(buf); | |
3315 | clean_tree_block(trans, root, buf); | |
3316 | btrfs_set_buffer_uptodate(buf); | |
d0c803c4 CM |
3317 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
3318 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
3319 | buf->start + buf->len - 1, GFP_NOFS); | |
3320 | } else { | |
3321 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 3322 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 3323 | } |
65b51a00 CM |
3324 | trans->blocks_used++; |
3325 | return buf; | |
3326 | } | |
3327 | ||
fec577fb CM |
3328 | /* |
3329 | * helper function to allocate a block for a given tree | |
3330 | * returns the tree buffer or NULL. | |
3331 | */ | |
5f39d397 | 3332 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
7bb86316 | 3333 | struct btrfs_root *root, |
31840ae1 | 3334 | u32 blocksize, u64 parent, |
7bb86316 CM |
3335 | u64 root_objectid, |
3336 | u64 ref_generation, | |
7bb86316 CM |
3337 | int level, |
3338 | u64 hint, | |
5f39d397 | 3339 | u64 empty_size) |
fec577fb | 3340 | { |
e2fa7227 | 3341 | struct btrfs_key ins; |
fec577fb | 3342 | int ret; |
5f39d397 | 3343 | struct extent_buffer *buf; |
fec577fb | 3344 | |
31840ae1 | 3345 | ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize, |
3bb1a1bc | 3346 | root_objectid, ref_generation, level, |
31840ae1 | 3347 | empty_size, hint, (u64)-1, &ins, 0); |
fec577fb | 3348 | if (ret) { |
54aa1f4d CM |
3349 | BUG_ON(ret > 0); |
3350 | return ERR_PTR(ret); | |
fec577fb | 3351 | } |
55c69072 | 3352 | |
65b51a00 | 3353 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize); |
fec577fb CM |
3354 | return buf; |
3355 | } | |
a28ec197 | 3356 | |
e02119d5 CM |
3357 | int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, |
3358 | struct btrfs_root *root, struct extent_buffer *leaf) | |
6407bf6d | 3359 | { |
7bb86316 CM |
3360 | u64 leaf_owner; |
3361 | u64 leaf_generation; | |
5f39d397 | 3362 | struct btrfs_key key; |
6407bf6d CM |
3363 | struct btrfs_file_extent_item *fi; |
3364 | int i; | |
3365 | int nritems; | |
3366 | int ret; | |
3367 | ||
5f39d397 CM |
3368 | BUG_ON(!btrfs_is_leaf(leaf)); |
3369 | nritems = btrfs_header_nritems(leaf); | |
7bb86316 CM |
3370 | leaf_owner = btrfs_header_owner(leaf); |
3371 | leaf_generation = btrfs_header_generation(leaf); | |
3372 | ||
6407bf6d | 3373 | for (i = 0; i < nritems; i++) { |
db94535d | 3374 | u64 disk_bytenr; |
e34a5b4f | 3375 | cond_resched(); |
5f39d397 CM |
3376 | |
3377 | btrfs_item_key_to_cpu(leaf, &key, i); | |
3378 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d CM |
3379 | continue; |
3380 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
5f39d397 CM |
3381 | if (btrfs_file_extent_type(leaf, fi) == |
3382 | BTRFS_FILE_EXTENT_INLINE) | |
236454df | 3383 | continue; |
6407bf6d CM |
3384 | /* |
3385 | * FIXME make sure to insert a trans record that | |
3386 | * repeats the snapshot del on crash | |
3387 | */ | |
db94535d CM |
3388 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
3389 | if (disk_bytenr == 0) | |
3a686375 | 3390 | continue; |
4a096752 | 3391 | |
925baedd | 3392 | ret = __btrfs_free_extent(trans, root, disk_bytenr, |
7bb86316 | 3393 | btrfs_file_extent_disk_num_bytes(leaf, fi), |
31840ae1 | 3394 | leaf->start, leaf_owner, leaf_generation, |
3bb1a1bc | 3395 | key.objectid, 0); |
31840ae1 | 3396 | BUG_ON(ret); |
2dd3e67b CM |
3397 | |
3398 | atomic_inc(&root->fs_info->throttle_gen); | |
3399 | wake_up(&root->fs_info->transaction_throttle); | |
3400 | cond_resched(); | |
6407bf6d CM |
3401 | } |
3402 | return 0; | |
3403 | } | |
3404 | ||
e02119d5 CM |
3405 | static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans, |
3406 | struct btrfs_root *root, | |
3407 | struct btrfs_leaf_ref *ref) | |
31153d81 YZ |
3408 | { |
3409 | int i; | |
3410 | int ret; | |
3411 | struct btrfs_extent_info *info = ref->extents; | |
3412 | ||
31153d81 | 3413 | for (i = 0; i < ref->nritems; i++) { |
31840ae1 ZY |
3414 | ret = __btrfs_free_extent(trans, root, info->bytenr, |
3415 | info->num_bytes, ref->bytenr, | |
3416 | ref->owner, ref->generation, | |
3bb1a1bc | 3417 | info->objectid, 0); |
2dd3e67b CM |
3418 | |
3419 | atomic_inc(&root->fs_info->throttle_gen); | |
3420 | wake_up(&root->fs_info->transaction_throttle); | |
3421 | cond_resched(); | |
3422 | ||
31153d81 YZ |
3423 | BUG_ON(ret); |
3424 | info++; | |
3425 | } | |
31153d81 YZ |
3426 | |
3427 | return 0; | |
3428 | } | |
3429 | ||
333db94c CM |
3430 | int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, |
3431 | u32 *refs) | |
3432 | { | |
017e5369 | 3433 | int ret; |
f87f057b | 3434 | |
31840ae1 | 3435 | ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs); |
f87f057b CM |
3436 | BUG_ON(ret); |
3437 | ||
3438 | #if 0 // some debugging code in case we see problems here | |
3439 | /* if the refs count is one, it won't get increased again. But | |
3440 | * if the ref count is > 1, someone may be decreasing it at | |
3441 | * the same time we are. | |
3442 | */ | |
3443 | if (*refs != 1) { | |
3444 | struct extent_buffer *eb = NULL; | |
3445 | eb = btrfs_find_create_tree_block(root, start, len); | |
3446 | if (eb) | |
3447 | btrfs_tree_lock(eb); | |
3448 | ||
3449 | mutex_lock(&root->fs_info->alloc_mutex); | |
3450 | ret = lookup_extent_ref(NULL, root, start, len, refs); | |
3451 | BUG_ON(ret); | |
3452 | mutex_unlock(&root->fs_info->alloc_mutex); | |
3453 | ||
3454 | if (eb) { | |
3455 | btrfs_tree_unlock(eb); | |
3456 | free_extent_buffer(eb); | |
3457 | } | |
3458 | if (*refs == 1) { | |
3459 | printk("block %llu went down to one during drop_snap\n", | |
3460 | (unsigned long long)start); | |
3461 | } | |
3462 | ||
3463 | } | |
3464 | #endif | |
3465 | ||
e7a84565 | 3466 | cond_resched(); |
017e5369 | 3467 | return ret; |
333db94c CM |
3468 | } |
3469 | ||
9aca1d51 CM |
3470 | /* |
3471 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
3472 | * counts as it goes. | |
3473 | */ | |
98ed5174 CM |
3474 | static int noinline walk_down_tree(struct btrfs_trans_handle *trans, |
3475 | struct btrfs_root *root, | |
3476 | struct btrfs_path *path, int *level) | |
20524f02 | 3477 | { |
7bb86316 CM |
3478 | u64 root_owner; |
3479 | u64 root_gen; | |
3480 | u64 bytenr; | |
ca7a79ad | 3481 | u64 ptr_gen; |
5f39d397 CM |
3482 | struct extent_buffer *next; |
3483 | struct extent_buffer *cur; | |
7bb86316 | 3484 | struct extent_buffer *parent; |
31153d81 | 3485 | struct btrfs_leaf_ref *ref; |
db94535d | 3486 | u32 blocksize; |
20524f02 CM |
3487 | int ret; |
3488 | u32 refs; | |
3489 | ||
5caf2a00 CM |
3490 | WARN_ON(*level < 0); |
3491 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
333db94c | 3492 | ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start, |
db94535d | 3493 | path->nodes[*level]->len, &refs); |
20524f02 CM |
3494 | BUG_ON(ret); |
3495 | if (refs > 1) | |
3496 | goto out; | |
e011599b | 3497 | |
9aca1d51 CM |
3498 | /* |
3499 | * walk down to the last node level and free all the leaves | |
3500 | */ | |
6407bf6d | 3501 | while(*level >= 0) { |
5caf2a00 CM |
3502 | WARN_ON(*level < 0); |
3503 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
20524f02 | 3504 | cur = path->nodes[*level]; |
e011599b | 3505 | |
5f39d397 | 3506 | if (btrfs_header_level(cur) != *level) |
2c90e5d6 | 3507 | WARN_ON(1); |
e011599b | 3508 | |
7518a238 | 3509 | if (path->slots[*level] >= |
5f39d397 | 3510 | btrfs_header_nritems(cur)) |
20524f02 | 3511 | break; |
6407bf6d | 3512 | if (*level == 0) { |
e02119d5 | 3513 | ret = btrfs_drop_leaf_ref(trans, root, cur); |
6407bf6d CM |
3514 | BUG_ON(ret); |
3515 | break; | |
3516 | } | |
db94535d | 3517 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
ca7a79ad | 3518 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); |
db94535d | 3519 | blocksize = btrfs_level_size(root, *level - 1); |
925baedd | 3520 | |
333db94c | 3521 | ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); |
6407bf6d CM |
3522 | BUG_ON(ret); |
3523 | if (refs != 1) { | |
7bb86316 CM |
3524 | parent = path->nodes[*level]; |
3525 | root_owner = btrfs_header_owner(parent); | |
3526 | root_gen = btrfs_header_generation(parent); | |
20524f02 | 3527 | path->slots[*level]++; |
f87f057b | 3528 | |
925baedd | 3529 | ret = __btrfs_free_extent(trans, root, bytenr, |
31840ae1 | 3530 | blocksize, parent->start, |
3bb1a1bc YZ |
3531 | root_owner, root_gen, |
3532 | *level - 1, 1); | |
20524f02 | 3533 | BUG_ON(ret); |
18e35e0a CM |
3534 | |
3535 | atomic_inc(&root->fs_info->throttle_gen); | |
3536 | wake_up(&root->fs_info->transaction_throttle); | |
2dd3e67b | 3537 | cond_resched(); |
18e35e0a | 3538 | |
20524f02 CM |
3539 | continue; |
3540 | } | |
f87f057b CM |
3541 | /* |
3542 | * at this point, we have a single ref, and since the | |
3543 | * only place referencing this extent is a dead root | |
3544 | * the reference count should never go higher. | |
3545 | * So, we don't need to check it again | |
3546 | */ | |
31153d81 | 3547 | if (*level == 1) { |
017e5369 | 3548 | ref = btrfs_lookup_leaf_ref(root, bytenr); |
1a40e23b ZY |
3549 | if (ref && ref->generation != ptr_gen) { |
3550 | btrfs_free_leaf_ref(root, ref); | |
3551 | ref = NULL; | |
3552 | } | |
31153d81 | 3553 | if (ref) { |
e02119d5 | 3554 | ret = cache_drop_leaf_ref(trans, root, ref); |
31153d81 YZ |
3555 | BUG_ON(ret); |
3556 | btrfs_remove_leaf_ref(root, ref); | |
bcc63abb | 3557 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
3558 | *level = 0; |
3559 | break; | |
3560 | } | |
3bb1a1bc | 3561 | if (printk_ratelimit()) { |
37d1aeee CM |
3562 | printk("leaf ref miss for bytenr %llu\n", |
3563 | (unsigned long long)bytenr); | |
3bb1a1bc | 3564 | } |
31153d81 | 3565 | } |
db94535d | 3566 | next = btrfs_find_tree_block(root, bytenr, blocksize); |
1259ab75 | 3567 | if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) { |
5f39d397 | 3568 | free_extent_buffer(next); |
333db94c | 3569 | |
ca7a79ad CM |
3570 | next = read_tree_block(root, bytenr, blocksize, |
3571 | ptr_gen); | |
e7a84565 | 3572 | cond_resched(); |
f87f057b CM |
3573 | #if 0 |
3574 | /* | |
3575 | * this is a debugging check and can go away | |
3576 | * the ref should never go all the way down to 1 | |
3577 | * at this point | |
3578 | */ | |
e6dcd2dc CM |
3579 | ret = lookup_extent_ref(NULL, root, bytenr, blocksize, |
3580 | &refs); | |
e9d0b13b | 3581 | BUG_ON(ret); |
f87f057b CM |
3582 | WARN_ON(refs != 1); |
3583 | #endif | |
e9d0b13b | 3584 | } |
5caf2a00 | 3585 | WARN_ON(*level <= 0); |
83e15a28 | 3586 | if (path->nodes[*level-1]) |
5f39d397 | 3587 | free_extent_buffer(path->nodes[*level-1]); |
20524f02 | 3588 | path->nodes[*level-1] = next; |
5f39d397 | 3589 | *level = btrfs_header_level(next); |
20524f02 | 3590 | path->slots[*level] = 0; |
2dd3e67b | 3591 | cond_resched(); |
20524f02 CM |
3592 | } |
3593 | out: | |
5caf2a00 CM |
3594 | WARN_ON(*level < 0); |
3595 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
7bb86316 CM |
3596 | |
3597 | if (path->nodes[*level] == root->node) { | |
7bb86316 | 3598 | parent = path->nodes[*level]; |
31153d81 | 3599 | bytenr = path->nodes[*level]->start; |
7bb86316 CM |
3600 | } else { |
3601 | parent = path->nodes[*level + 1]; | |
31153d81 | 3602 | bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]); |
7bb86316 CM |
3603 | } |
3604 | ||
31153d81 YZ |
3605 | blocksize = btrfs_level_size(root, *level); |
3606 | root_owner = btrfs_header_owner(parent); | |
7bb86316 | 3607 | root_gen = btrfs_header_generation(parent); |
31153d81 YZ |
3608 | |
3609 | ret = __btrfs_free_extent(trans, root, bytenr, blocksize, | |
31840ae1 | 3610 | parent->start, root_owner, root_gen, |
3bb1a1bc | 3611 | *level, 1); |
5f39d397 | 3612 | free_extent_buffer(path->nodes[*level]); |
20524f02 CM |
3613 | path->nodes[*level] = NULL; |
3614 | *level += 1; | |
3615 | BUG_ON(ret); | |
f87f057b | 3616 | |
e7a84565 | 3617 | cond_resched(); |
20524f02 CM |
3618 | return 0; |
3619 | } | |
3620 | ||
f82d02d9 YZ |
3621 | /* |
3622 | * helper function for drop_subtree, this function is similar to | |
3623 | * walk_down_tree. The main difference is that it checks reference | |
3624 | * counts while tree blocks are locked. | |
3625 | */ | |
3626 | static int noinline walk_down_subtree(struct btrfs_trans_handle *trans, | |
3627 | struct btrfs_root *root, | |
3628 | struct btrfs_path *path, int *level) | |
3629 | { | |
3630 | struct extent_buffer *next; | |
3631 | struct extent_buffer *cur; | |
3632 | struct extent_buffer *parent; | |
3633 | u64 bytenr; | |
3634 | u64 ptr_gen; | |
3635 | u32 blocksize; | |
3636 | u32 refs; | |
3637 | int ret; | |
3638 | ||
3639 | cur = path->nodes[*level]; | |
3640 | ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len, | |
3641 | &refs); | |
3642 | BUG_ON(ret); | |
3643 | if (refs > 1) | |
3644 | goto out; | |
3645 | ||
3646 | while (*level >= 0) { | |
3647 | cur = path->nodes[*level]; | |
3648 | if (*level == 0) { | |
3649 | ret = btrfs_drop_leaf_ref(trans, root, cur); | |
3650 | BUG_ON(ret); | |
3651 | clean_tree_block(trans, root, cur); | |
3652 | break; | |
3653 | } | |
3654 | if (path->slots[*level] >= btrfs_header_nritems(cur)) { | |
3655 | clean_tree_block(trans, root, cur); | |
3656 | break; | |
3657 | } | |
3658 | ||
3659 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
3660 | blocksize = btrfs_level_size(root, *level - 1); | |
3661 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
3662 | ||
3663 | next = read_tree_block(root, bytenr, blocksize, ptr_gen); | |
3664 | btrfs_tree_lock(next); | |
3665 | ||
3666 | ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize, | |
3667 | &refs); | |
3668 | BUG_ON(ret); | |
3669 | if (refs > 1) { | |
3670 | parent = path->nodes[*level]; | |
3671 | ret = btrfs_free_extent(trans, root, bytenr, | |
3672 | blocksize, parent->start, | |
3673 | btrfs_header_owner(parent), | |
3674 | btrfs_header_generation(parent), | |
3675 | *level - 1, 1); | |
3676 | BUG_ON(ret); | |
3677 | path->slots[*level]++; | |
3678 | btrfs_tree_unlock(next); | |
3679 | free_extent_buffer(next); | |
3680 | continue; | |
3681 | } | |
3682 | ||
3683 | *level = btrfs_header_level(next); | |
3684 | path->nodes[*level] = next; | |
3685 | path->slots[*level] = 0; | |
3686 | path->locks[*level] = 1; | |
3687 | cond_resched(); | |
3688 | } | |
3689 | out: | |
3690 | parent = path->nodes[*level + 1]; | |
3691 | bytenr = path->nodes[*level]->start; | |
3692 | blocksize = path->nodes[*level]->len; | |
3693 | ||
3694 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, | |
3695 | parent->start, btrfs_header_owner(parent), | |
3696 | btrfs_header_generation(parent), *level, 1); | |
3697 | BUG_ON(ret); | |
3698 | ||
3699 | if (path->locks[*level]) { | |
3700 | btrfs_tree_unlock(path->nodes[*level]); | |
3701 | path->locks[*level] = 0; | |
3702 | } | |
3703 | free_extent_buffer(path->nodes[*level]); | |
3704 | path->nodes[*level] = NULL; | |
3705 | *level += 1; | |
3706 | cond_resched(); | |
3707 | return 0; | |
3708 | } | |
3709 | ||
9aca1d51 CM |
3710 | /* |
3711 | * helper for dropping snapshots. This walks back up the tree in the path | |
3712 | * to find the first node higher up where we haven't yet gone through | |
3713 | * all the slots | |
3714 | */ | |
98ed5174 CM |
3715 | static int noinline walk_up_tree(struct btrfs_trans_handle *trans, |
3716 | struct btrfs_root *root, | |
f82d02d9 YZ |
3717 | struct btrfs_path *path, |
3718 | int *level, int max_level) | |
20524f02 | 3719 | { |
7bb86316 CM |
3720 | u64 root_owner; |
3721 | u64 root_gen; | |
3722 | struct btrfs_root_item *root_item = &root->root_item; | |
20524f02 CM |
3723 | int i; |
3724 | int slot; | |
3725 | int ret; | |
9f3a7427 | 3726 | |
f82d02d9 | 3727 | for (i = *level; i < max_level && path->nodes[i]; i++) { |
20524f02 | 3728 | slot = path->slots[i]; |
5f39d397 CM |
3729 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { |
3730 | struct extent_buffer *node; | |
3731 | struct btrfs_disk_key disk_key; | |
3732 | node = path->nodes[i]; | |
20524f02 CM |
3733 | path->slots[i]++; |
3734 | *level = i; | |
9f3a7427 | 3735 | WARN_ON(*level == 0); |
5f39d397 | 3736 | btrfs_node_key(node, &disk_key, path->slots[i]); |
9f3a7427 | 3737 | memcpy(&root_item->drop_progress, |
5f39d397 | 3738 | &disk_key, sizeof(disk_key)); |
9f3a7427 | 3739 | root_item->drop_level = i; |
20524f02 CM |
3740 | return 0; |
3741 | } else { | |
31840ae1 ZY |
3742 | struct extent_buffer *parent; |
3743 | if (path->nodes[*level] == root->node) | |
3744 | parent = path->nodes[*level]; | |
3745 | else | |
3746 | parent = path->nodes[*level + 1]; | |
3747 | ||
3748 | root_owner = btrfs_header_owner(parent); | |
3749 | root_gen = btrfs_header_generation(parent); | |
f82d02d9 YZ |
3750 | |
3751 | clean_tree_block(trans, root, path->nodes[*level]); | |
e089f05c | 3752 | ret = btrfs_free_extent(trans, root, |
db94535d | 3753 | path->nodes[*level]->start, |
7bb86316 | 3754 | path->nodes[*level]->len, |
3bb1a1bc YZ |
3755 | parent->start, root_owner, |
3756 | root_gen, *level, 1); | |
6407bf6d | 3757 | BUG_ON(ret); |
f82d02d9 YZ |
3758 | if (path->locks[*level]) { |
3759 | btrfs_tree_unlock(path->nodes[*level]); | |
3760 | path->locks[*level] = 0; | |
3761 | } | |
5f39d397 | 3762 | free_extent_buffer(path->nodes[*level]); |
83e15a28 | 3763 | path->nodes[*level] = NULL; |
20524f02 | 3764 | *level = i + 1; |
20524f02 CM |
3765 | } |
3766 | } | |
3767 | return 1; | |
3768 | } | |
3769 | ||
9aca1d51 CM |
3770 | /* |
3771 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
3772 | * the tree freeing any blocks that have a ref count of zero after being | |
3773 | * decremented. | |
3774 | */ | |
e089f05c | 3775 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
9f3a7427 | 3776 | *root) |
20524f02 | 3777 | { |
3768f368 | 3778 | int ret = 0; |
9aca1d51 | 3779 | int wret; |
20524f02 | 3780 | int level; |
5caf2a00 | 3781 | struct btrfs_path *path; |
20524f02 CM |
3782 | int i; |
3783 | int orig_level; | |
9f3a7427 | 3784 | struct btrfs_root_item *root_item = &root->root_item; |
20524f02 | 3785 | |
a2135011 | 3786 | WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); |
5caf2a00 CM |
3787 | path = btrfs_alloc_path(); |
3788 | BUG_ON(!path); | |
20524f02 | 3789 | |
5f39d397 | 3790 | level = btrfs_header_level(root->node); |
20524f02 | 3791 | orig_level = level; |
9f3a7427 CM |
3792 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
3793 | path->nodes[level] = root->node; | |
f510cfec | 3794 | extent_buffer_get(root->node); |
9f3a7427 CM |
3795 | path->slots[level] = 0; |
3796 | } else { | |
3797 | struct btrfs_key key; | |
5f39d397 CM |
3798 | struct btrfs_disk_key found_key; |
3799 | struct extent_buffer *node; | |
6702ed49 | 3800 | |
9f3a7427 | 3801 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
6702ed49 CM |
3802 | level = root_item->drop_level; |
3803 | path->lowest_level = level; | |
9f3a7427 | 3804 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6702ed49 | 3805 | if (wret < 0) { |
9f3a7427 CM |
3806 | ret = wret; |
3807 | goto out; | |
3808 | } | |
5f39d397 CM |
3809 | node = path->nodes[level]; |
3810 | btrfs_node_key(node, &found_key, path->slots[level]); | |
3811 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, | |
3812 | sizeof(found_key))); | |
7d9eb12c CM |
3813 | /* |
3814 | * unlock our path, this is safe because only this | |
3815 | * function is allowed to delete this snapshot | |
3816 | */ | |
925baedd CM |
3817 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3818 | if (path->nodes[i] && path->locks[i]) { | |
3819 | path->locks[i] = 0; | |
3820 | btrfs_tree_unlock(path->nodes[i]); | |
3821 | } | |
3822 | } | |
9f3a7427 | 3823 | } |
20524f02 | 3824 | while(1) { |
5caf2a00 | 3825 | wret = walk_down_tree(trans, root, path, &level); |
9aca1d51 | 3826 | if (wret > 0) |
20524f02 | 3827 | break; |
9aca1d51 CM |
3828 | if (wret < 0) |
3829 | ret = wret; | |
3830 | ||
f82d02d9 YZ |
3831 | wret = walk_up_tree(trans, root, path, &level, |
3832 | BTRFS_MAX_LEVEL); | |
9aca1d51 | 3833 | if (wret > 0) |
20524f02 | 3834 | break; |
9aca1d51 CM |
3835 | if (wret < 0) |
3836 | ret = wret; | |
e7a84565 CM |
3837 | if (trans->transaction->in_commit) { |
3838 | ret = -EAGAIN; | |
3839 | break; | |
3840 | } | |
18e35e0a | 3841 | atomic_inc(&root->fs_info->throttle_gen); |
017e5369 | 3842 | wake_up(&root->fs_info->transaction_throttle); |
20524f02 | 3843 | } |
83e15a28 | 3844 | for (i = 0; i <= orig_level; i++) { |
5caf2a00 | 3845 | if (path->nodes[i]) { |
5f39d397 | 3846 | free_extent_buffer(path->nodes[i]); |
0f82731f | 3847 | path->nodes[i] = NULL; |
83e15a28 | 3848 | } |
20524f02 | 3849 | } |
9f3a7427 | 3850 | out: |
5caf2a00 | 3851 | btrfs_free_path(path); |
9aca1d51 | 3852 | return ret; |
20524f02 | 3853 | } |
9078a3e1 | 3854 | |
f82d02d9 YZ |
3855 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
3856 | struct btrfs_root *root, | |
3857 | struct extent_buffer *node, | |
3858 | struct extent_buffer *parent) | |
3859 | { | |
3860 | struct btrfs_path *path; | |
3861 | int level; | |
3862 | int parent_level; | |
3863 | int ret = 0; | |
3864 | int wret; | |
3865 | ||
3866 | path = btrfs_alloc_path(); | |
3867 | BUG_ON(!path); | |
3868 | ||
3869 | BUG_ON(!btrfs_tree_locked(parent)); | |
3870 | parent_level = btrfs_header_level(parent); | |
3871 | extent_buffer_get(parent); | |
3872 | path->nodes[parent_level] = parent; | |
3873 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
3874 | ||
3875 | BUG_ON(!btrfs_tree_locked(node)); | |
3876 | level = btrfs_header_level(node); | |
3877 | extent_buffer_get(node); | |
3878 | path->nodes[level] = node; | |
3879 | path->slots[level] = 0; | |
3880 | ||
3881 | while (1) { | |
3882 | wret = walk_down_subtree(trans, root, path, &level); | |
3883 | if (wret < 0) | |
3884 | ret = wret; | |
3885 | if (wret != 0) | |
3886 | break; | |
3887 | ||
3888 | wret = walk_up_tree(trans, root, path, &level, parent_level); | |
3889 | if (wret < 0) | |
3890 | ret = wret; | |
3891 | if (wret != 0) | |
3892 | break; | |
3893 | } | |
3894 | ||
3895 | btrfs_free_path(path); | |
3896 | return ret; | |
3897 | } | |
3898 | ||
8e7bf94f CM |
3899 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
3900 | unsigned long nr) | |
3901 | { | |
3902 | return min(last, start + nr - 1); | |
3903 | } | |
3904 | ||
98ed5174 CM |
3905 | static int noinline relocate_inode_pages(struct inode *inode, u64 start, |
3906 | u64 len) | |
edbd8d4e CM |
3907 | { |
3908 | u64 page_start; | |
3909 | u64 page_end; | |
1a40e23b | 3910 | unsigned long first_index; |
edbd8d4e | 3911 | unsigned long last_index; |
edbd8d4e CM |
3912 | unsigned long i; |
3913 | struct page *page; | |
d1310b2e | 3914 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 3915 | struct file_ra_state *ra; |
3eaa2885 | 3916 | struct btrfs_ordered_extent *ordered; |
1a40e23b ZY |
3917 | unsigned int total_read = 0; |
3918 | unsigned int total_dirty = 0; | |
3919 | int ret = 0; | |
4313b399 CM |
3920 | |
3921 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
3922 | |
3923 | mutex_lock(&inode->i_mutex); | |
1a40e23b | 3924 | first_index = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
3925 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
3926 | ||
1a40e23b ZY |
3927 | /* make sure the dirty trick played by the caller work */ |
3928 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
3929 | first_index, last_index); | |
3930 | if (ret) | |
3931 | goto out_unlock; | |
8e7bf94f | 3932 | |
4313b399 | 3933 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 3934 | |
1a40e23b ZY |
3935 | for (i = first_index ; i <= last_index; i++) { |
3936 | if (total_read % ra->ra_pages == 0) { | |
8e7bf94f | 3937 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, |
1a40e23b | 3938 | calc_ra(i, last_index, ra->ra_pages)); |
8e7bf94f CM |
3939 | } |
3940 | total_read++; | |
3eaa2885 CM |
3941 | again: |
3942 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
1a40e23b | 3943 | BUG_ON(1); |
edbd8d4e | 3944 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 3945 | if (!page) { |
1a40e23b | 3946 | ret = -ENOMEM; |
edbd8d4e | 3947 | goto out_unlock; |
a061fc8d | 3948 | } |
edbd8d4e CM |
3949 | if (!PageUptodate(page)) { |
3950 | btrfs_readpage(NULL, page); | |
3951 | lock_page(page); | |
3952 | if (!PageUptodate(page)) { | |
3953 | unlock_page(page); | |
3954 | page_cache_release(page); | |
1a40e23b | 3955 | ret = -EIO; |
edbd8d4e CM |
3956 | goto out_unlock; |
3957 | } | |
3958 | } | |
ec44a35c | 3959 | wait_on_page_writeback(page); |
3eaa2885 | 3960 | |
edbd8d4e CM |
3961 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
3962 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
d1310b2e | 3963 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 3964 | |
3eaa2885 CM |
3965 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
3966 | if (ordered) { | |
3967 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3968 | unlock_page(page); | |
3969 | page_cache_release(page); | |
3970 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3971 | btrfs_put_ordered_extent(ordered); | |
3972 | goto again; | |
3973 | } | |
3974 | set_page_extent_mapped(page); | |
3975 | ||
ea8c2819 | 3976 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
1a40e23b ZY |
3977 | if (i == first_index) |
3978 | set_extent_bits(io_tree, page_start, page_end, | |
3979 | EXTENT_BOUNDARY, GFP_NOFS); | |
3980 | ||
a061fc8d | 3981 | set_page_dirty(page); |
1a40e23b | 3982 | total_dirty++; |
edbd8d4e | 3983 | |
d1310b2e | 3984 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
3985 | unlock_page(page); |
3986 | page_cache_release(page); | |
3987 | } | |
3988 | ||
3989 | out_unlock: | |
ec44a35c | 3990 | kfree(ra); |
edbd8d4e | 3991 | mutex_unlock(&inode->i_mutex); |
1a40e23b ZY |
3992 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty); |
3993 | return ret; | |
edbd8d4e CM |
3994 | } |
3995 | ||
1a40e23b ZY |
3996 | static int noinline relocate_data_extent(struct inode *reloc_inode, |
3997 | struct btrfs_key *extent_key, | |
3998 | u64 offset) | |
3999 | { | |
4000 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
4001 | struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree; | |
4002 | struct extent_map *em; | |
6643558d YZ |
4003 | u64 start = extent_key->objectid - offset; |
4004 | u64 end = start + extent_key->offset - 1; | |
bf4ef679 | 4005 | |
1a40e23b ZY |
4006 | em = alloc_extent_map(GFP_NOFS); |
4007 | BUG_ON(!em || IS_ERR(em)); | |
bf4ef679 | 4008 | |
6643558d | 4009 | em->start = start; |
1a40e23b | 4010 | em->len = extent_key->offset; |
c8b97818 | 4011 | em->block_len = extent_key->offset; |
1a40e23b ZY |
4012 | em->block_start = extent_key->objectid; |
4013 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
4014 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
4015 | ||
4016 | /* setup extent map to cheat btrfs_readpage */ | |
6643558d | 4017 | lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
1a40e23b ZY |
4018 | while (1) { |
4019 | int ret; | |
4020 | spin_lock(&em_tree->lock); | |
4021 | ret = add_extent_mapping(em_tree, em); | |
4022 | spin_unlock(&em_tree->lock); | |
4023 | if (ret != -EEXIST) { | |
4024 | free_extent_map(em); | |
bf4ef679 CM |
4025 | break; |
4026 | } | |
6643558d | 4027 | btrfs_drop_extent_cache(reloc_inode, start, end, 0); |
bf4ef679 | 4028 | } |
6643558d | 4029 | unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
bf4ef679 | 4030 | |
6643558d | 4031 | return relocate_inode_pages(reloc_inode, start, extent_key->offset); |
1a40e23b | 4032 | } |
edbd8d4e | 4033 | |
1a40e23b ZY |
4034 | struct btrfs_ref_path { |
4035 | u64 extent_start; | |
4036 | u64 nodes[BTRFS_MAX_LEVEL]; | |
4037 | u64 root_objectid; | |
4038 | u64 root_generation; | |
4039 | u64 owner_objectid; | |
1a40e23b ZY |
4040 | u32 num_refs; |
4041 | int lowest_level; | |
4042 | int current_level; | |
f82d02d9 YZ |
4043 | int shared_level; |
4044 | ||
4045 | struct btrfs_key node_keys[BTRFS_MAX_LEVEL]; | |
4046 | u64 new_nodes[BTRFS_MAX_LEVEL]; | |
1a40e23b | 4047 | }; |
7d9eb12c | 4048 | |
1a40e23b | 4049 | struct disk_extent { |
c8b97818 | 4050 | u64 ram_bytes; |
1a40e23b ZY |
4051 | u64 disk_bytenr; |
4052 | u64 disk_num_bytes; | |
4053 | u64 offset; | |
4054 | u64 num_bytes; | |
c8b97818 CM |
4055 | u8 compression; |
4056 | u8 encryption; | |
4057 | u16 other_encoding; | |
1a40e23b | 4058 | }; |
4313b399 | 4059 | |
1a40e23b ZY |
4060 | static int is_cowonly_root(u64 root_objectid) |
4061 | { | |
4062 | if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || | |
4063 | root_objectid == BTRFS_EXTENT_TREE_OBJECTID || | |
4064 | root_objectid == BTRFS_CHUNK_TREE_OBJECTID || | |
4065 | root_objectid == BTRFS_DEV_TREE_OBJECTID || | |
4066 | root_objectid == BTRFS_TREE_LOG_OBJECTID) | |
4067 | return 1; | |
4068 | return 0; | |
4069 | } | |
edbd8d4e | 4070 | |
1a40e23b ZY |
4071 | static int noinline __next_ref_path(struct btrfs_trans_handle *trans, |
4072 | struct btrfs_root *extent_root, | |
4073 | struct btrfs_ref_path *ref_path, | |
4074 | int first_time) | |
4075 | { | |
4076 | struct extent_buffer *leaf; | |
4077 | struct btrfs_path *path; | |
4078 | struct btrfs_extent_ref *ref; | |
4079 | struct btrfs_key key; | |
4080 | struct btrfs_key found_key; | |
4081 | u64 bytenr; | |
4082 | u32 nritems; | |
4083 | int level; | |
4084 | int ret = 1; | |
edbd8d4e | 4085 | |
1a40e23b ZY |
4086 | path = btrfs_alloc_path(); |
4087 | if (!path) | |
4088 | return -ENOMEM; | |
bf4ef679 | 4089 | |
1a40e23b ZY |
4090 | if (first_time) { |
4091 | ref_path->lowest_level = -1; | |
4092 | ref_path->current_level = -1; | |
f82d02d9 | 4093 | ref_path->shared_level = -1; |
1a40e23b ZY |
4094 | goto walk_up; |
4095 | } | |
4096 | walk_down: | |
4097 | level = ref_path->current_level - 1; | |
4098 | while (level >= -1) { | |
4099 | u64 parent; | |
4100 | if (level < ref_path->lowest_level) | |
4101 | break; | |
bf4ef679 | 4102 | |
1a40e23b ZY |
4103 | if (level >= 0) { |
4104 | bytenr = ref_path->nodes[level]; | |
4105 | } else { | |
4106 | bytenr = ref_path->extent_start; | |
4107 | } | |
4108 | BUG_ON(bytenr == 0); | |
bf4ef679 | 4109 | |
1a40e23b ZY |
4110 | parent = ref_path->nodes[level + 1]; |
4111 | ref_path->nodes[level + 1] = 0; | |
4112 | ref_path->current_level = level; | |
4113 | BUG_ON(parent == 0); | |
0ef3e66b | 4114 | |
1a40e23b ZY |
4115 | key.objectid = bytenr; |
4116 | key.offset = parent + 1; | |
4117 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 4118 | |
1a40e23b ZY |
4119 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
4120 | if (ret < 0) | |
edbd8d4e | 4121 | goto out; |
1a40e23b | 4122 | BUG_ON(ret == 0); |
7d9eb12c | 4123 | |
1a40e23b ZY |
4124 | leaf = path->nodes[0]; |
4125 | nritems = btrfs_header_nritems(leaf); | |
4126 | if (path->slots[0] >= nritems) { | |
4127 | ret = btrfs_next_leaf(extent_root, path); | |
4128 | if (ret < 0) | |
4129 | goto out; | |
4130 | if (ret > 0) | |
4131 | goto next; | |
4132 | leaf = path->nodes[0]; | |
4133 | } | |
0ef3e66b | 4134 | |
1a40e23b ZY |
4135 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4136 | if (found_key.objectid == bytenr && | |
f82d02d9 YZ |
4137 | found_key.type == BTRFS_EXTENT_REF_KEY) { |
4138 | if (level < ref_path->shared_level) | |
4139 | ref_path->shared_level = level; | |
1a40e23b | 4140 | goto found; |
f82d02d9 | 4141 | } |
1a40e23b ZY |
4142 | next: |
4143 | level--; | |
4144 | btrfs_release_path(extent_root, path); | |
d899e052 | 4145 | cond_resched(); |
1a40e23b ZY |
4146 | } |
4147 | /* reached lowest level */ | |
4148 | ret = 1; | |
4149 | goto out; | |
4150 | walk_up: | |
4151 | level = ref_path->current_level; | |
4152 | while (level < BTRFS_MAX_LEVEL - 1) { | |
4153 | u64 ref_objectid; | |
4154 | if (level >= 0) { | |
4155 | bytenr = ref_path->nodes[level]; | |
4156 | } else { | |
4157 | bytenr = ref_path->extent_start; | |
4158 | } | |
4159 | BUG_ON(bytenr == 0); | |
edbd8d4e | 4160 | |
1a40e23b ZY |
4161 | key.objectid = bytenr; |
4162 | key.offset = 0; | |
4163 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 4164 | |
1a40e23b ZY |
4165 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
4166 | if (ret < 0) | |
4167 | goto out; | |
edbd8d4e | 4168 | |
1a40e23b ZY |
4169 | leaf = path->nodes[0]; |
4170 | nritems = btrfs_header_nritems(leaf); | |
4171 | if (path->slots[0] >= nritems) { | |
4172 | ret = btrfs_next_leaf(extent_root, path); | |
4173 | if (ret < 0) | |
4174 | goto out; | |
4175 | if (ret > 0) { | |
4176 | /* the extent was freed by someone */ | |
4177 | if (ref_path->lowest_level == level) | |
4178 | goto out; | |
4179 | btrfs_release_path(extent_root, path); | |
4180 | goto walk_down; | |
4181 | } | |
4182 | leaf = path->nodes[0]; | |
4183 | } | |
edbd8d4e | 4184 | |
1a40e23b ZY |
4185 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
4186 | if (found_key.objectid != bytenr || | |
4187 | found_key.type != BTRFS_EXTENT_REF_KEY) { | |
4188 | /* the extent was freed by someone */ | |
4189 | if (ref_path->lowest_level == level) { | |
4190 | ret = 1; | |
4191 | goto out; | |
4192 | } | |
4193 | btrfs_release_path(extent_root, path); | |
4194 | goto walk_down; | |
4195 | } | |
4196 | found: | |
4197 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
4198 | struct btrfs_extent_ref); | |
4199 | ref_objectid = btrfs_ref_objectid(leaf, ref); | |
4200 | if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
4201 | if (first_time) { | |
4202 | level = (int)ref_objectid; | |
4203 | BUG_ON(level >= BTRFS_MAX_LEVEL); | |
4204 | ref_path->lowest_level = level; | |
4205 | ref_path->current_level = level; | |
4206 | ref_path->nodes[level] = bytenr; | |
4207 | } else { | |
4208 | WARN_ON(ref_objectid != level); | |
4209 | } | |
4210 | } else { | |
4211 | WARN_ON(level != -1); | |
4212 | } | |
4213 | first_time = 0; | |
bf4ef679 | 4214 | |
1a40e23b ZY |
4215 | if (ref_path->lowest_level == level) { |
4216 | ref_path->owner_objectid = ref_objectid; | |
1a40e23b ZY |
4217 | ref_path->num_refs = btrfs_ref_num_refs(leaf, ref); |
4218 | } | |
bf4ef679 | 4219 | |
7d9eb12c | 4220 | /* |
1a40e23b ZY |
4221 | * the block is tree root or the block isn't in reference |
4222 | * counted tree. | |
7d9eb12c | 4223 | */ |
1a40e23b ZY |
4224 | if (found_key.objectid == found_key.offset || |
4225 | is_cowonly_root(btrfs_ref_root(leaf, ref))) { | |
4226 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
4227 | ref_path->root_generation = | |
4228 | btrfs_ref_generation(leaf, ref); | |
4229 | if (level < 0) { | |
4230 | /* special reference from the tree log */ | |
4231 | ref_path->nodes[0] = found_key.offset; | |
4232 | ref_path->current_level = 0; | |
4233 | } | |
4234 | ret = 0; | |
4235 | goto out; | |
4236 | } | |
7d9eb12c | 4237 | |
1a40e23b ZY |
4238 | level++; |
4239 | BUG_ON(ref_path->nodes[level] != 0); | |
4240 | ref_path->nodes[level] = found_key.offset; | |
4241 | ref_path->current_level = level; | |
bf4ef679 | 4242 | |
1a40e23b ZY |
4243 | /* |
4244 | * the reference was created in the running transaction, | |
4245 | * no need to continue walking up. | |
4246 | */ | |
4247 | if (btrfs_ref_generation(leaf, ref) == trans->transid) { | |
4248 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
4249 | ref_path->root_generation = | |
4250 | btrfs_ref_generation(leaf, ref); | |
4251 | ret = 0; | |
4252 | goto out; | |
7d9eb12c CM |
4253 | } |
4254 | ||
1a40e23b | 4255 | btrfs_release_path(extent_root, path); |
d899e052 | 4256 | cond_resched(); |
7d9eb12c | 4257 | } |
1a40e23b ZY |
4258 | /* reached max tree level, but no tree root found. */ |
4259 | BUG(); | |
edbd8d4e | 4260 | out: |
1a40e23b ZY |
4261 | btrfs_free_path(path); |
4262 | return ret; | |
edbd8d4e CM |
4263 | } |
4264 | ||
1a40e23b ZY |
4265 | static int btrfs_first_ref_path(struct btrfs_trans_handle *trans, |
4266 | struct btrfs_root *extent_root, | |
4267 | struct btrfs_ref_path *ref_path, | |
4268 | u64 extent_start) | |
a061fc8d | 4269 | { |
1a40e23b ZY |
4270 | memset(ref_path, 0, sizeof(*ref_path)); |
4271 | ref_path->extent_start = extent_start; | |
a061fc8d | 4272 | |
1a40e23b | 4273 | return __next_ref_path(trans, extent_root, ref_path, 1); |
a061fc8d CM |
4274 | } |
4275 | ||
1a40e23b ZY |
4276 | static int btrfs_next_ref_path(struct btrfs_trans_handle *trans, |
4277 | struct btrfs_root *extent_root, | |
4278 | struct btrfs_ref_path *ref_path) | |
edbd8d4e | 4279 | { |
1a40e23b ZY |
4280 | return __next_ref_path(trans, extent_root, ref_path, 0); |
4281 | } | |
4282 | ||
4283 | static int noinline get_new_locations(struct inode *reloc_inode, | |
4284 | struct btrfs_key *extent_key, | |
4285 | u64 offset, int no_fragment, | |
4286 | struct disk_extent **extents, | |
4287 | int *nr_extents) | |
4288 | { | |
4289 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
4290 | struct btrfs_path *path; | |
4291 | struct btrfs_file_extent_item *fi; | |
edbd8d4e | 4292 | struct extent_buffer *leaf; |
1a40e23b ZY |
4293 | struct disk_extent *exts = *extents; |
4294 | struct btrfs_key found_key; | |
4295 | u64 cur_pos; | |
4296 | u64 last_byte; | |
edbd8d4e | 4297 | u32 nritems; |
1a40e23b ZY |
4298 | int nr = 0; |
4299 | int max = *nr_extents; | |
4300 | int ret; | |
edbd8d4e | 4301 | |
1a40e23b ZY |
4302 | WARN_ON(!no_fragment && *extents); |
4303 | if (!exts) { | |
4304 | max = 1; | |
4305 | exts = kmalloc(sizeof(*exts) * max, GFP_NOFS); | |
4306 | if (!exts) | |
4307 | return -ENOMEM; | |
a061fc8d | 4308 | } |
edbd8d4e | 4309 | |
1a40e23b ZY |
4310 | path = btrfs_alloc_path(); |
4311 | BUG_ON(!path); | |
edbd8d4e | 4312 | |
1a40e23b ZY |
4313 | cur_pos = extent_key->objectid - offset; |
4314 | last_byte = extent_key->objectid + extent_key->offset; | |
4315 | ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino, | |
4316 | cur_pos, 0); | |
4317 | if (ret < 0) | |
4318 | goto out; | |
4319 | if (ret > 0) { | |
4320 | ret = -ENOENT; | |
4321 | goto out; | |
4322 | } | |
edbd8d4e | 4323 | |
1a40e23b | 4324 | while (1) { |
edbd8d4e CM |
4325 | leaf = path->nodes[0]; |
4326 | nritems = btrfs_header_nritems(leaf); | |
1a40e23b ZY |
4327 | if (path->slots[0] >= nritems) { |
4328 | ret = btrfs_next_leaf(root, path); | |
a061fc8d CM |
4329 | if (ret < 0) |
4330 | goto out; | |
1a40e23b ZY |
4331 | if (ret > 0) |
4332 | break; | |
bf4ef679 | 4333 | leaf = path->nodes[0]; |
a061fc8d | 4334 | } |
edbd8d4e CM |
4335 | |
4336 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b ZY |
4337 | if (found_key.offset != cur_pos || |
4338 | found_key.type != BTRFS_EXTENT_DATA_KEY || | |
4339 | found_key.objectid != reloc_inode->i_ino) | |
edbd8d4e CM |
4340 | break; |
4341 | ||
1a40e23b ZY |
4342 | fi = btrfs_item_ptr(leaf, path->slots[0], |
4343 | struct btrfs_file_extent_item); | |
4344 | if (btrfs_file_extent_type(leaf, fi) != | |
4345 | BTRFS_FILE_EXTENT_REG || | |
4346 | btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
edbd8d4e | 4347 | break; |
1a40e23b ZY |
4348 | |
4349 | if (nr == max) { | |
4350 | struct disk_extent *old = exts; | |
4351 | max *= 2; | |
4352 | exts = kzalloc(sizeof(*exts) * max, GFP_NOFS); | |
4353 | memcpy(exts, old, sizeof(*exts) * nr); | |
4354 | if (old != *extents) | |
4355 | kfree(old); | |
a061fc8d | 4356 | } |
edbd8d4e | 4357 | |
1a40e23b ZY |
4358 | exts[nr].disk_bytenr = |
4359 | btrfs_file_extent_disk_bytenr(leaf, fi); | |
4360 | exts[nr].disk_num_bytes = | |
4361 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
4362 | exts[nr].offset = btrfs_file_extent_offset(leaf, fi); | |
4363 | exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
c8b97818 CM |
4364 | exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
4365 | exts[nr].compression = btrfs_file_extent_compression(leaf, fi); | |
4366 | exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi); | |
4367 | exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf, | |
4368 | fi); | |
d899e052 YZ |
4369 | BUG_ON(exts[nr].offset > 0); |
4370 | BUG_ON(exts[nr].compression || exts[nr].encryption); | |
4371 | BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes); | |
edbd8d4e | 4372 | |
1a40e23b ZY |
4373 | cur_pos += exts[nr].num_bytes; |
4374 | nr++; | |
4375 | ||
4376 | if (cur_pos + offset >= last_byte) | |
4377 | break; | |
4378 | ||
4379 | if (no_fragment) { | |
4380 | ret = 1; | |
edbd8d4e | 4381 | goto out; |
1a40e23b ZY |
4382 | } |
4383 | path->slots[0]++; | |
4384 | } | |
4385 | ||
4386 | WARN_ON(cur_pos + offset > last_byte); | |
4387 | if (cur_pos + offset < last_byte) { | |
4388 | ret = -ENOENT; | |
4389 | goto out; | |
edbd8d4e CM |
4390 | } |
4391 | ret = 0; | |
4392 | out: | |
1a40e23b ZY |
4393 | btrfs_free_path(path); |
4394 | if (ret) { | |
4395 | if (exts != *extents) | |
4396 | kfree(exts); | |
4397 | } else { | |
4398 | *extents = exts; | |
4399 | *nr_extents = nr; | |
4400 | } | |
4401 | return ret; | |
4402 | } | |
4403 | ||
4404 | static int noinline replace_one_extent(struct btrfs_trans_handle *trans, | |
4405 | struct btrfs_root *root, | |
4406 | struct btrfs_path *path, | |
4407 | struct btrfs_key *extent_key, | |
4408 | struct btrfs_key *leaf_key, | |
4409 | struct btrfs_ref_path *ref_path, | |
4410 | struct disk_extent *new_extents, | |
4411 | int nr_extents) | |
4412 | { | |
4413 | struct extent_buffer *leaf; | |
4414 | struct btrfs_file_extent_item *fi; | |
4415 | struct inode *inode = NULL; | |
4416 | struct btrfs_key key; | |
4417 | u64 lock_start = 0; | |
4418 | u64 lock_end = 0; | |
4419 | u64 num_bytes; | |
4420 | u64 ext_offset; | |
4421 | u64 first_pos; | |
4422 | u32 nritems; | |
3bb1a1bc | 4423 | int nr_scaned = 0; |
1a40e23b | 4424 | int extent_locked = 0; |
d899e052 | 4425 | int extent_type; |
1a40e23b ZY |
4426 | int ret; |
4427 | ||
3bb1a1bc YZ |
4428 | memcpy(&key, leaf_key, sizeof(key)); |
4429 | first_pos = INT_LIMIT(loff_t) - extent_key->offset; | |
1a40e23b | 4430 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { |
3bb1a1bc YZ |
4431 | if (key.objectid < ref_path->owner_objectid || |
4432 | (key.objectid == ref_path->owner_objectid && | |
4433 | key.type < BTRFS_EXTENT_DATA_KEY)) { | |
4434 | key.objectid = ref_path->owner_objectid; | |
4435 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4436 | key.offset = 0; | |
4437 | } | |
1a40e23b ZY |
4438 | } |
4439 | ||
4440 | while (1) { | |
4441 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
4442 | if (ret < 0) | |
4443 | goto out; | |
4444 | ||
4445 | leaf = path->nodes[0]; | |
4446 | nritems = btrfs_header_nritems(leaf); | |
4447 | next: | |
4448 | if (extent_locked && ret > 0) { | |
4449 | /* | |
4450 | * the file extent item was modified by someone | |
4451 | * before the extent got locked. | |
4452 | */ | |
1a40e23b ZY |
4453 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
4454 | lock_end, GFP_NOFS); | |
4455 | extent_locked = 0; | |
4456 | } | |
4457 | ||
4458 | if (path->slots[0] >= nritems) { | |
3bb1a1bc | 4459 | if (++nr_scaned > 2) |
1a40e23b ZY |
4460 | break; |
4461 | ||
4462 | BUG_ON(extent_locked); | |
4463 | ret = btrfs_next_leaf(root, path); | |
4464 | if (ret < 0) | |
4465 | goto out; | |
4466 | if (ret > 0) | |
4467 | break; | |
4468 | leaf = path->nodes[0]; | |
4469 | nritems = btrfs_header_nritems(leaf); | |
4470 | } | |
4471 | ||
4472 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
4473 | ||
4474 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { | |
4475 | if ((key.objectid > ref_path->owner_objectid) || | |
4476 | (key.objectid == ref_path->owner_objectid && | |
4477 | key.type > BTRFS_EXTENT_DATA_KEY) || | |
4478 | (key.offset >= first_pos + extent_key->offset)) | |
4479 | break; | |
4480 | } | |
4481 | ||
4482 | if (inode && key.objectid != inode->i_ino) { | |
4483 | BUG_ON(extent_locked); | |
4484 | btrfs_release_path(root, path); | |
4485 | mutex_unlock(&inode->i_mutex); | |
4486 | iput(inode); | |
4487 | inode = NULL; | |
4488 | continue; | |
4489 | } | |
4490 | ||
4491 | if (key.type != BTRFS_EXTENT_DATA_KEY) { | |
4492 | path->slots[0]++; | |
4493 | ret = 1; | |
4494 | goto next; | |
4495 | } | |
4496 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4497 | struct btrfs_file_extent_item); | |
d899e052 YZ |
4498 | extent_type = btrfs_file_extent_type(leaf, fi); |
4499 | if ((extent_type != BTRFS_FILE_EXTENT_REG && | |
4500 | extent_type != BTRFS_FILE_EXTENT_PREALLOC) || | |
1a40e23b ZY |
4501 | (btrfs_file_extent_disk_bytenr(leaf, fi) != |
4502 | extent_key->objectid)) { | |
4503 | path->slots[0]++; | |
4504 | ret = 1; | |
4505 | goto next; | |
4506 | } | |
4507 | ||
4508 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
4509 | ext_offset = btrfs_file_extent_offset(leaf, fi); | |
4510 | ||
4511 | if (first_pos > key.offset - ext_offset) | |
4512 | first_pos = key.offset - ext_offset; | |
4513 | ||
4514 | if (!extent_locked) { | |
4515 | lock_start = key.offset; | |
4516 | lock_end = lock_start + num_bytes - 1; | |
4517 | } else { | |
6643558d YZ |
4518 | if (lock_start > key.offset || |
4519 | lock_end + 1 < key.offset + num_bytes) { | |
4520 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
4521 | lock_start, lock_end, GFP_NOFS); | |
4522 | extent_locked = 0; | |
4523 | } | |
1a40e23b ZY |
4524 | } |
4525 | ||
4526 | if (!inode) { | |
4527 | btrfs_release_path(root, path); | |
4528 | ||
4529 | inode = btrfs_iget_locked(root->fs_info->sb, | |
4530 | key.objectid, root); | |
4531 | if (inode->i_state & I_NEW) { | |
4532 | BTRFS_I(inode)->root = root; | |
4533 | BTRFS_I(inode)->location.objectid = | |
4534 | key.objectid; | |
4535 | BTRFS_I(inode)->location.type = | |
4536 | BTRFS_INODE_ITEM_KEY; | |
4537 | BTRFS_I(inode)->location.offset = 0; | |
4538 | btrfs_read_locked_inode(inode); | |
4539 | unlock_new_inode(inode); | |
4540 | } | |
4541 | /* | |
4542 | * some code call btrfs_commit_transaction while | |
4543 | * holding the i_mutex, so we can't use mutex_lock | |
4544 | * here. | |
4545 | */ | |
4546 | if (is_bad_inode(inode) || | |
4547 | !mutex_trylock(&inode->i_mutex)) { | |
4548 | iput(inode); | |
4549 | inode = NULL; | |
4550 | key.offset = (u64)-1; | |
4551 | goto skip; | |
4552 | } | |
4553 | } | |
4554 | ||
4555 | if (!extent_locked) { | |
4556 | struct btrfs_ordered_extent *ordered; | |
4557 | ||
4558 | btrfs_release_path(root, path); | |
4559 | ||
4560 | lock_extent(&BTRFS_I(inode)->io_tree, lock_start, | |
4561 | lock_end, GFP_NOFS); | |
4562 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
4563 | lock_end); | |
4564 | if (ordered && | |
4565 | ordered->file_offset <= lock_end && | |
4566 | ordered->file_offset + ordered->len > lock_start) { | |
4567 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
4568 | lock_start, lock_end, GFP_NOFS); | |
4569 | btrfs_start_ordered_extent(inode, ordered, 1); | |
4570 | btrfs_put_ordered_extent(ordered); | |
4571 | key.offset += num_bytes; | |
4572 | goto skip; | |
4573 | } | |
4574 | if (ordered) | |
4575 | btrfs_put_ordered_extent(ordered); | |
4576 | ||
1a40e23b ZY |
4577 | extent_locked = 1; |
4578 | continue; | |
4579 | } | |
4580 | ||
4581 | if (nr_extents == 1) { | |
4582 | /* update extent pointer in place */ | |
1a40e23b ZY |
4583 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
4584 | new_extents[0].disk_bytenr); | |
4585 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
4586 | new_extents[0].disk_num_bytes); | |
1a40e23b ZY |
4587 | btrfs_mark_buffer_dirty(leaf); |
4588 | ||
4589 | btrfs_drop_extent_cache(inode, key.offset, | |
4590 | key.offset + num_bytes - 1, 0); | |
4591 | ||
4592 | ret = btrfs_inc_extent_ref(trans, root, | |
4593 | new_extents[0].disk_bytenr, | |
4594 | new_extents[0].disk_num_bytes, | |
4595 | leaf->start, | |
4596 | root->root_key.objectid, | |
4597 | trans->transid, | |
3bb1a1bc | 4598 | key.objectid); |
1a40e23b ZY |
4599 | BUG_ON(ret); |
4600 | ||
4601 | ret = btrfs_free_extent(trans, root, | |
4602 | extent_key->objectid, | |
4603 | extent_key->offset, | |
4604 | leaf->start, | |
4605 | btrfs_header_owner(leaf), | |
4606 | btrfs_header_generation(leaf), | |
3bb1a1bc | 4607 | key.objectid, 0); |
1a40e23b ZY |
4608 | BUG_ON(ret); |
4609 | ||
4610 | btrfs_release_path(root, path); | |
4611 | key.offset += num_bytes; | |
4612 | } else { | |
d899e052 YZ |
4613 | BUG_ON(1); |
4614 | #if 0 | |
1a40e23b ZY |
4615 | u64 alloc_hint; |
4616 | u64 extent_len; | |
4617 | int i; | |
4618 | /* | |
4619 | * drop old extent pointer at first, then insert the | |
4620 | * new pointers one bye one | |
4621 | */ | |
4622 | btrfs_release_path(root, path); | |
4623 | ret = btrfs_drop_extents(trans, root, inode, key.offset, | |
4624 | key.offset + num_bytes, | |
4625 | key.offset, &alloc_hint); | |
4626 | BUG_ON(ret); | |
4627 | ||
4628 | for (i = 0; i < nr_extents; i++) { | |
4629 | if (ext_offset >= new_extents[i].num_bytes) { | |
4630 | ext_offset -= new_extents[i].num_bytes; | |
4631 | continue; | |
4632 | } | |
4633 | extent_len = min(new_extents[i].num_bytes - | |
4634 | ext_offset, num_bytes); | |
4635 | ||
4636 | ret = btrfs_insert_empty_item(trans, root, | |
4637 | path, &key, | |
4638 | sizeof(*fi)); | |
4639 | BUG_ON(ret); | |
4640 | ||
4641 | leaf = path->nodes[0]; | |
4642 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4643 | struct btrfs_file_extent_item); | |
4644 | btrfs_set_file_extent_generation(leaf, fi, | |
4645 | trans->transid); | |
4646 | btrfs_set_file_extent_type(leaf, fi, | |
4647 | BTRFS_FILE_EXTENT_REG); | |
4648 | btrfs_set_file_extent_disk_bytenr(leaf, fi, | |
4649 | new_extents[i].disk_bytenr); | |
4650 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
4651 | new_extents[i].disk_num_bytes); | |
c8b97818 CM |
4652 | btrfs_set_file_extent_ram_bytes(leaf, fi, |
4653 | new_extents[i].ram_bytes); | |
4654 | ||
4655 | btrfs_set_file_extent_compression(leaf, fi, | |
4656 | new_extents[i].compression); | |
4657 | btrfs_set_file_extent_encryption(leaf, fi, | |
4658 | new_extents[i].encryption); | |
4659 | btrfs_set_file_extent_other_encoding(leaf, fi, | |
4660 | new_extents[i].other_encoding); | |
4661 | ||
1a40e23b ZY |
4662 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4663 | extent_len); | |
4664 | ext_offset += new_extents[i].offset; | |
4665 | btrfs_set_file_extent_offset(leaf, fi, | |
4666 | ext_offset); | |
4667 | btrfs_mark_buffer_dirty(leaf); | |
4668 | ||
4669 | btrfs_drop_extent_cache(inode, key.offset, | |
4670 | key.offset + extent_len - 1, 0); | |
4671 | ||
4672 | ret = btrfs_inc_extent_ref(trans, root, | |
4673 | new_extents[i].disk_bytenr, | |
4674 | new_extents[i].disk_num_bytes, | |
4675 | leaf->start, | |
4676 | root->root_key.objectid, | |
3bb1a1bc | 4677 | trans->transid, key.objectid); |
1a40e23b ZY |
4678 | BUG_ON(ret); |
4679 | btrfs_release_path(root, path); | |
4680 | ||
a76a3cd4 | 4681 | inode_add_bytes(inode, extent_len); |
1a40e23b ZY |
4682 | |
4683 | ext_offset = 0; | |
4684 | num_bytes -= extent_len; | |
4685 | key.offset += extent_len; | |
4686 | ||
4687 | if (num_bytes == 0) | |
4688 | break; | |
4689 | } | |
4690 | BUG_ON(i >= nr_extents); | |
d899e052 | 4691 | #endif |
1a40e23b ZY |
4692 | } |
4693 | ||
4694 | if (extent_locked) { | |
1a40e23b ZY |
4695 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
4696 | lock_end, GFP_NOFS); | |
4697 | extent_locked = 0; | |
4698 | } | |
4699 | skip: | |
4700 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS && | |
4701 | key.offset >= first_pos + extent_key->offset) | |
4702 | break; | |
4703 | ||
4704 | cond_resched(); | |
4705 | } | |
4706 | ret = 0; | |
4707 | out: | |
4708 | btrfs_release_path(root, path); | |
4709 | if (inode) { | |
4710 | mutex_unlock(&inode->i_mutex); | |
4711 | if (extent_locked) { | |
1a40e23b ZY |
4712 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
4713 | lock_end, GFP_NOFS); | |
4714 | } | |
4715 | iput(inode); | |
4716 | } | |
4717 | return ret; | |
4718 | } | |
4719 | ||
1a40e23b ZY |
4720 | int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, |
4721 | struct btrfs_root *root, | |
4722 | struct extent_buffer *buf, u64 orig_start) | |
4723 | { | |
4724 | int level; | |
4725 | int ret; | |
4726 | ||
4727 | BUG_ON(btrfs_header_generation(buf) != trans->transid); | |
4728 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
4729 | ||
4730 | level = btrfs_header_level(buf); | |
4731 | if (level == 0) { | |
4732 | struct btrfs_leaf_ref *ref; | |
4733 | struct btrfs_leaf_ref *orig_ref; | |
4734 | ||
4735 | orig_ref = btrfs_lookup_leaf_ref(root, orig_start); | |
4736 | if (!orig_ref) | |
4737 | return -ENOENT; | |
4738 | ||
4739 | ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems); | |
4740 | if (!ref) { | |
4741 | btrfs_free_leaf_ref(root, orig_ref); | |
4742 | return -ENOMEM; | |
4743 | } | |
4744 | ||
4745 | ref->nritems = orig_ref->nritems; | |
4746 | memcpy(ref->extents, orig_ref->extents, | |
4747 | sizeof(ref->extents[0]) * ref->nritems); | |
4748 | ||
4749 | btrfs_free_leaf_ref(root, orig_ref); | |
4750 | ||
4751 | ref->root_gen = trans->transid; | |
4752 | ref->bytenr = buf->start; | |
4753 | ref->owner = btrfs_header_owner(buf); | |
4754 | ref->generation = btrfs_header_generation(buf); | |
4755 | ret = btrfs_add_leaf_ref(root, ref, 0); | |
4756 | WARN_ON(ret); | |
4757 | btrfs_free_leaf_ref(root, ref); | |
4758 | } | |
4759 | return 0; | |
4760 | } | |
4761 | ||
4762 | static int noinline invalidate_extent_cache(struct btrfs_root *root, | |
4763 | struct extent_buffer *leaf, | |
4764 | struct btrfs_block_group_cache *group, | |
4765 | struct btrfs_root *target_root) | |
4766 | { | |
4767 | struct btrfs_key key; | |
4768 | struct inode *inode = NULL; | |
4769 | struct btrfs_file_extent_item *fi; | |
4770 | u64 num_bytes; | |
4771 | u64 skip_objectid = 0; | |
4772 | u32 nritems; | |
4773 | u32 i; | |
4774 | ||
4775 | nritems = btrfs_header_nritems(leaf); | |
4776 | for (i = 0; i < nritems; i++) { | |
4777 | btrfs_item_key_to_cpu(leaf, &key, i); | |
4778 | if (key.objectid == skip_objectid || | |
4779 | key.type != BTRFS_EXTENT_DATA_KEY) | |
4780 | continue; | |
4781 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
4782 | if (btrfs_file_extent_type(leaf, fi) == | |
4783 | BTRFS_FILE_EXTENT_INLINE) | |
4784 | continue; | |
4785 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
4786 | continue; | |
4787 | if (!inode || inode->i_ino != key.objectid) { | |
4788 | iput(inode); | |
4789 | inode = btrfs_ilookup(target_root->fs_info->sb, | |
4790 | key.objectid, target_root, 1); | |
4791 | } | |
4792 | if (!inode) { | |
4793 | skip_objectid = key.objectid; | |
4794 | continue; | |
4795 | } | |
4796 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
4797 | ||
4798 | lock_extent(&BTRFS_I(inode)->io_tree, key.offset, | |
4799 | key.offset + num_bytes - 1, GFP_NOFS); | |
1a40e23b ZY |
4800 | btrfs_drop_extent_cache(inode, key.offset, |
4801 | key.offset + num_bytes - 1, 1); | |
1a40e23b ZY |
4802 | unlock_extent(&BTRFS_I(inode)->io_tree, key.offset, |
4803 | key.offset + num_bytes - 1, GFP_NOFS); | |
4804 | cond_resched(); | |
4805 | } | |
4806 | iput(inode); | |
4807 | return 0; | |
4808 | } | |
4809 | ||
4810 | static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans, | |
4811 | struct btrfs_root *root, | |
4812 | struct extent_buffer *leaf, | |
4813 | struct btrfs_block_group_cache *group, | |
4814 | struct inode *reloc_inode) | |
4815 | { | |
4816 | struct btrfs_key key; | |
4817 | struct btrfs_key extent_key; | |
4818 | struct btrfs_file_extent_item *fi; | |
4819 | struct btrfs_leaf_ref *ref; | |
4820 | struct disk_extent *new_extent; | |
4821 | u64 bytenr; | |
4822 | u64 num_bytes; | |
4823 | u32 nritems; | |
4824 | u32 i; | |
4825 | int ext_index; | |
4826 | int nr_extent; | |
4827 | int ret; | |
4828 | ||
4829 | new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS); | |
4830 | BUG_ON(!new_extent); | |
4831 | ||
4832 | ref = btrfs_lookup_leaf_ref(root, leaf->start); | |
4833 | BUG_ON(!ref); | |
4834 | ||
4835 | ext_index = -1; | |
4836 | nritems = btrfs_header_nritems(leaf); | |
4837 | for (i = 0; i < nritems; i++) { | |
4838 | btrfs_item_key_to_cpu(leaf, &key, i); | |
4839 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
4840 | continue; | |
4841 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
4842 | if (btrfs_file_extent_type(leaf, fi) == | |
4843 | BTRFS_FILE_EXTENT_INLINE) | |
4844 | continue; | |
4845 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
4846 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
4847 | if (bytenr == 0) | |
4848 | continue; | |
4849 | ||
4850 | ext_index++; | |
4851 | if (bytenr >= group->key.objectid + group->key.offset || | |
4852 | bytenr + num_bytes <= group->key.objectid) | |
4853 | continue; | |
4854 | ||
4855 | extent_key.objectid = bytenr; | |
4856 | extent_key.offset = num_bytes; | |
4857 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
4858 | nr_extent = 1; | |
4859 | ret = get_new_locations(reloc_inode, &extent_key, | |
4860 | group->key.objectid, 1, | |
4861 | &new_extent, &nr_extent); | |
4862 | if (ret > 0) | |
4863 | continue; | |
4864 | BUG_ON(ret < 0); | |
4865 | ||
4866 | BUG_ON(ref->extents[ext_index].bytenr != bytenr); | |
4867 | BUG_ON(ref->extents[ext_index].num_bytes != num_bytes); | |
4868 | ref->extents[ext_index].bytenr = new_extent->disk_bytenr; | |
4869 | ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes; | |
4870 | ||
1a40e23b ZY |
4871 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
4872 | new_extent->disk_bytenr); | |
4873 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
4874 | new_extent->disk_num_bytes); | |
1a40e23b ZY |
4875 | btrfs_mark_buffer_dirty(leaf); |
4876 | ||
4877 | ret = btrfs_inc_extent_ref(trans, root, | |
4878 | new_extent->disk_bytenr, | |
4879 | new_extent->disk_num_bytes, | |
4880 | leaf->start, | |
4881 | root->root_key.objectid, | |
3bb1a1bc | 4882 | trans->transid, key.objectid); |
1a40e23b ZY |
4883 | BUG_ON(ret); |
4884 | ret = btrfs_free_extent(trans, root, | |
4885 | bytenr, num_bytes, leaf->start, | |
4886 | btrfs_header_owner(leaf), | |
4887 | btrfs_header_generation(leaf), | |
3bb1a1bc | 4888 | key.objectid, 0); |
1a40e23b ZY |
4889 | BUG_ON(ret); |
4890 | cond_resched(); | |
4891 | } | |
4892 | kfree(new_extent); | |
4893 | BUG_ON(ext_index + 1 != ref->nritems); | |
4894 | btrfs_free_leaf_ref(root, ref); | |
4895 | return 0; | |
4896 | } | |
4897 | ||
f82d02d9 YZ |
4898 | int btrfs_free_reloc_root(struct btrfs_trans_handle *trans, |
4899 | struct btrfs_root *root) | |
1a40e23b ZY |
4900 | { |
4901 | struct btrfs_root *reloc_root; | |
f82d02d9 | 4902 | int ret; |
1a40e23b ZY |
4903 | |
4904 | if (root->reloc_root) { | |
4905 | reloc_root = root->reloc_root; | |
4906 | root->reloc_root = NULL; | |
4907 | list_add(&reloc_root->dead_list, | |
4908 | &root->fs_info->dead_reloc_roots); | |
f82d02d9 YZ |
4909 | |
4910 | btrfs_set_root_bytenr(&reloc_root->root_item, | |
4911 | reloc_root->node->start); | |
4912 | btrfs_set_root_level(&root->root_item, | |
4913 | btrfs_header_level(reloc_root->node)); | |
4914 | memset(&reloc_root->root_item.drop_progress, 0, | |
4915 | sizeof(struct btrfs_disk_key)); | |
4916 | reloc_root->root_item.drop_level = 0; | |
4917 | ||
4918 | ret = btrfs_update_root(trans, root->fs_info->tree_root, | |
4919 | &reloc_root->root_key, | |
4920 | &reloc_root->root_item); | |
4921 | BUG_ON(ret); | |
1a40e23b ZY |
4922 | } |
4923 | return 0; | |
4924 | } | |
4925 | ||
4926 | int btrfs_drop_dead_reloc_roots(struct btrfs_root *root) | |
4927 | { | |
4928 | struct btrfs_trans_handle *trans; | |
4929 | struct btrfs_root *reloc_root; | |
4930 | struct btrfs_root *prev_root = NULL; | |
4931 | struct list_head dead_roots; | |
4932 | int ret; | |
4933 | unsigned long nr; | |
4934 | ||
4935 | INIT_LIST_HEAD(&dead_roots); | |
4936 | list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots); | |
4937 | ||
4938 | while (!list_empty(&dead_roots)) { | |
4939 | reloc_root = list_entry(dead_roots.prev, | |
4940 | struct btrfs_root, dead_list); | |
4941 | list_del_init(&reloc_root->dead_list); | |
4942 | ||
4943 | BUG_ON(reloc_root->commit_root != NULL); | |
4944 | while (1) { | |
4945 | trans = btrfs_join_transaction(root, 1); | |
4946 | BUG_ON(!trans); | |
4947 | ||
4948 | mutex_lock(&root->fs_info->drop_mutex); | |
4949 | ret = btrfs_drop_snapshot(trans, reloc_root); | |
4950 | if (ret != -EAGAIN) | |
4951 | break; | |
4952 | mutex_unlock(&root->fs_info->drop_mutex); | |
4953 | ||
4954 | nr = trans->blocks_used; | |
4955 | ret = btrfs_end_transaction(trans, root); | |
4956 | BUG_ON(ret); | |
4957 | btrfs_btree_balance_dirty(root, nr); | |
4958 | } | |
4959 | ||
4960 | free_extent_buffer(reloc_root->node); | |
4961 | ||
4962 | ret = btrfs_del_root(trans, root->fs_info->tree_root, | |
4963 | &reloc_root->root_key); | |
4964 | BUG_ON(ret); | |
4965 | mutex_unlock(&root->fs_info->drop_mutex); | |
4966 | ||
4967 | nr = trans->blocks_used; | |
4968 | ret = btrfs_end_transaction(trans, root); | |
4969 | BUG_ON(ret); | |
4970 | btrfs_btree_balance_dirty(root, nr); | |
4971 | ||
4972 | kfree(prev_root); | |
4973 | prev_root = reloc_root; | |
4974 | } | |
4975 | if (prev_root) { | |
4976 | btrfs_remove_leaf_refs(prev_root, (u64)-1, 0); | |
4977 | kfree(prev_root); | |
4978 | } | |
4979 | return 0; | |
4980 | } | |
4981 | ||
4982 | int btrfs_add_dead_reloc_root(struct btrfs_root *root) | |
4983 | { | |
4984 | list_add(&root->dead_list, &root->fs_info->dead_reloc_roots); | |
4985 | return 0; | |
4986 | } | |
4987 | ||
4988 | int btrfs_cleanup_reloc_trees(struct btrfs_root *root) | |
4989 | { | |
4990 | struct btrfs_root *reloc_root; | |
4991 | struct btrfs_trans_handle *trans; | |
4992 | struct btrfs_key location; | |
4993 | int found; | |
4994 | int ret; | |
4995 | ||
4996 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
4997 | ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL); | |
4998 | BUG_ON(ret); | |
4999 | found = !list_empty(&root->fs_info->dead_reloc_roots); | |
5000 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
5001 | ||
5002 | if (found) { | |
5003 | trans = btrfs_start_transaction(root, 1); | |
5004 | BUG_ON(!trans); | |
5005 | ret = btrfs_commit_transaction(trans, root); | |
5006 | BUG_ON(ret); | |
5007 | } | |
5008 | ||
5009 | location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
5010 | location.offset = (u64)-1; | |
5011 | location.type = BTRFS_ROOT_ITEM_KEY; | |
5012 | ||
5013 | reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location); | |
5014 | BUG_ON(!reloc_root); | |
5015 | btrfs_orphan_cleanup(reloc_root); | |
5016 | return 0; | |
5017 | } | |
5018 | ||
5019 | static int noinline init_reloc_tree(struct btrfs_trans_handle *trans, | |
5020 | struct btrfs_root *root) | |
5021 | { | |
5022 | struct btrfs_root *reloc_root; | |
5023 | struct extent_buffer *eb; | |
5024 | struct btrfs_root_item *root_item; | |
5025 | struct btrfs_key root_key; | |
5026 | int ret; | |
5027 | ||
5028 | BUG_ON(!root->ref_cows); | |
5029 | if (root->reloc_root) | |
5030 | return 0; | |
5031 | ||
5032 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); | |
5033 | BUG_ON(!root_item); | |
5034 | ||
5035 | ret = btrfs_copy_root(trans, root, root->commit_root, | |
5036 | &eb, BTRFS_TREE_RELOC_OBJECTID); | |
5037 | BUG_ON(ret); | |
5038 | ||
5039 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
5040 | root_key.offset = root->root_key.objectid; | |
5041 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
5042 | ||
5043 | memcpy(root_item, &root->root_item, sizeof(root_item)); | |
5044 | btrfs_set_root_refs(root_item, 0); | |
5045 | btrfs_set_root_bytenr(root_item, eb->start); | |
5046 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
84234f3a | 5047 | btrfs_set_root_generation(root_item, trans->transid); |
1a40e23b ZY |
5048 | |
5049 | btrfs_tree_unlock(eb); | |
5050 | free_extent_buffer(eb); | |
5051 | ||
5052 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, | |
5053 | &root_key, root_item); | |
5054 | BUG_ON(ret); | |
5055 | kfree(root_item); | |
5056 | ||
5057 | reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root, | |
5058 | &root_key); | |
5059 | BUG_ON(!reloc_root); | |
5060 | reloc_root->last_trans = trans->transid; | |
5061 | reloc_root->commit_root = NULL; | |
5062 | reloc_root->ref_tree = &root->fs_info->reloc_ref_tree; | |
5063 | ||
5064 | root->reloc_root = reloc_root; | |
5065 | return 0; | |
5066 | } | |
5067 | ||
5068 | /* | |
5069 | * Core function of space balance. | |
5070 | * | |
5071 | * The idea is using reloc trees to relocate tree blocks in reference | |
f82d02d9 YZ |
5072 | * counted roots. There is one reloc tree for each subvol, and all |
5073 | * reloc trees share same root key objectid. Reloc trees are snapshots | |
5074 | * of the latest committed roots of subvols (root->commit_root). | |
5075 | * | |
5076 | * To relocate a tree block referenced by a subvol, there are two steps. | |
5077 | * COW the block through subvol's reloc tree, then update block pointer | |
5078 | * in the subvol to point to the new block. Since all reloc trees share | |
5079 | * same root key objectid, doing special handing for tree blocks owned | |
5080 | * by them is easy. Once a tree block has been COWed in one reloc tree, | |
5081 | * we can use the resulting new block directly when the same block is | |
5082 | * required to COW again through other reloc trees. By this way, relocated | |
5083 | * tree blocks are shared between reloc trees, so they are also shared | |
5084 | * between subvols. | |
1a40e23b ZY |
5085 | */ |
5086 | static int noinline relocate_one_path(struct btrfs_trans_handle *trans, | |
5087 | struct btrfs_root *root, | |
5088 | struct btrfs_path *path, | |
5089 | struct btrfs_key *first_key, | |
5090 | struct btrfs_ref_path *ref_path, | |
5091 | struct btrfs_block_group_cache *group, | |
5092 | struct inode *reloc_inode) | |
5093 | { | |
5094 | struct btrfs_root *reloc_root; | |
5095 | struct extent_buffer *eb = NULL; | |
5096 | struct btrfs_key *keys; | |
5097 | u64 *nodes; | |
5098 | int level; | |
f82d02d9 | 5099 | int shared_level; |
1a40e23b | 5100 | int lowest_level = 0; |
1a40e23b ZY |
5101 | int ret; |
5102 | ||
5103 | if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
5104 | lowest_level = ref_path->owner_objectid; | |
5105 | ||
f82d02d9 | 5106 | if (!root->ref_cows) { |
1a40e23b ZY |
5107 | path->lowest_level = lowest_level; |
5108 | ret = btrfs_search_slot(trans, root, first_key, path, 0, 1); | |
5109 | BUG_ON(ret < 0); | |
5110 | path->lowest_level = 0; | |
5111 | btrfs_release_path(root, path); | |
5112 | return 0; | |
5113 | } | |
5114 | ||
1a40e23b ZY |
5115 | mutex_lock(&root->fs_info->tree_reloc_mutex); |
5116 | ret = init_reloc_tree(trans, root); | |
5117 | BUG_ON(ret); | |
5118 | reloc_root = root->reloc_root; | |
5119 | ||
f82d02d9 YZ |
5120 | shared_level = ref_path->shared_level; |
5121 | ref_path->shared_level = BTRFS_MAX_LEVEL - 1; | |
1a40e23b | 5122 | |
f82d02d9 YZ |
5123 | keys = ref_path->node_keys; |
5124 | nodes = ref_path->new_nodes; | |
5125 | memset(&keys[shared_level + 1], 0, | |
5126 | sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
5127 | memset(&nodes[shared_level + 1], 0, | |
5128 | sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
1a40e23b | 5129 | |
f82d02d9 YZ |
5130 | if (nodes[lowest_level] == 0) { |
5131 | path->lowest_level = lowest_level; | |
5132 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, | |
5133 | 0, 1); | |
5134 | BUG_ON(ret); | |
5135 | for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) { | |
5136 | eb = path->nodes[level]; | |
5137 | if (!eb || eb == reloc_root->node) | |
5138 | break; | |
5139 | nodes[level] = eb->start; | |
5140 | if (level == 0) | |
5141 | btrfs_item_key_to_cpu(eb, &keys[level], 0); | |
5142 | else | |
5143 | btrfs_node_key_to_cpu(eb, &keys[level], 0); | |
5144 | } | |
2b82032c YZ |
5145 | if (nodes[0] && |
5146 | ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
5147 | eb = path->nodes[0]; |
5148 | ret = replace_extents_in_leaf(trans, reloc_root, eb, | |
5149 | group, reloc_inode); | |
5150 | BUG_ON(ret); | |
5151 | } | |
5152 | btrfs_release_path(reloc_root, path); | |
5153 | } else { | |
1a40e23b | 5154 | ret = btrfs_merge_path(trans, reloc_root, keys, nodes, |
f82d02d9 | 5155 | lowest_level); |
1a40e23b ZY |
5156 | BUG_ON(ret); |
5157 | } | |
5158 | ||
1a40e23b ZY |
5159 | /* |
5160 | * replace tree blocks in the fs tree with tree blocks in | |
5161 | * the reloc tree. | |
5162 | */ | |
5163 | ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level); | |
5164 | BUG_ON(ret < 0); | |
5165 | ||
5166 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
5167 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, |
5168 | 0, 0); | |
5169 | BUG_ON(ret); | |
5170 | extent_buffer_get(path->nodes[0]); | |
5171 | eb = path->nodes[0]; | |
5172 | btrfs_release_path(reloc_root, path); | |
1a40e23b ZY |
5173 | ret = invalidate_extent_cache(reloc_root, eb, group, root); |
5174 | BUG_ON(ret); | |
5175 | free_extent_buffer(eb); | |
5176 | } | |
1a40e23b | 5177 | |
f82d02d9 | 5178 | mutex_unlock(&root->fs_info->tree_reloc_mutex); |
1a40e23b | 5179 | path->lowest_level = 0; |
1a40e23b ZY |
5180 | return 0; |
5181 | } | |
5182 | ||
5183 | static int noinline relocate_tree_block(struct btrfs_trans_handle *trans, | |
5184 | struct btrfs_root *root, | |
5185 | struct btrfs_path *path, | |
5186 | struct btrfs_key *first_key, | |
5187 | struct btrfs_ref_path *ref_path) | |
5188 | { | |
5189 | int ret; | |
1a40e23b ZY |
5190 | |
5191 | ret = relocate_one_path(trans, root, path, first_key, | |
5192 | ref_path, NULL, NULL); | |
5193 | BUG_ON(ret); | |
5194 | ||
5195 | if (root == root->fs_info->extent_root) | |
5196 | btrfs_extent_post_op(trans, root); | |
1a40e23b ZY |
5197 | |
5198 | return 0; | |
5199 | } | |
5200 | ||
5201 | static int noinline del_extent_zero(struct btrfs_trans_handle *trans, | |
5202 | struct btrfs_root *extent_root, | |
5203 | struct btrfs_path *path, | |
5204 | struct btrfs_key *extent_key) | |
5205 | { | |
5206 | int ret; | |
5207 | ||
1a40e23b ZY |
5208 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); |
5209 | if (ret) | |
5210 | goto out; | |
5211 | ret = btrfs_del_item(trans, extent_root, path); | |
5212 | out: | |
5213 | btrfs_release_path(extent_root, path); | |
1a40e23b ZY |
5214 | return ret; |
5215 | } | |
5216 | ||
5217 | static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info, | |
5218 | struct btrfs_ref_path *ref_path) | |
5219 | { | |
5220 | struct btrfs_key root_key; | |
5221 | ||
5222 | root_key.objectid = ref_path->root_objectid; | |
5223 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
5224 | if (is_cowonly_root(ref_path->root_objectid)) | |
5225 | root_key.offset = 0; | |
5226 | else | |
5227 | root_key.offset = (u64)-1; | |
5228 | ||
5229 | return btrfs_read_fs_root_no_name(fs_info, &root_key); | |
5230 | } | |
5231 | ||
5232 | static int noinline relocate_one_extent(struct btrfs_root *extent_root, | |
5233 | struct btrfs_path *path, | |
5234 | struct btrfs_key *extent_key, | |
5235 | struct btrfs_block_group_cache *group, | |
5236 | struct inode *reloc_inode, int pass) | |
5237 | { | |
5238 | struct btrfs_trans_handle *trans; | |
5239 | struct btrfs_root *found_root; | |
5240 | struct btrfs_ref_path *ref_path = NULL; | |
5241 | struct disk_extent *new_extents = NULL; | |
5242 | int nr_extents = 0; | |
5243 | int loops; | |
5244 | int ret; | |
5245 | int level; | |
5246 | struct btrfs_key first_key; | |
5247 | u64 prev_block = 0; | |
5248 | ||
1a40e23b ZY |
5249 | |
5250 | trans = btrfs_start_transaction(extent_root, 1); | |
5251 | BUG_ON(!trans); | |
5252 | ||
5253 | if (extent_key->objectid == 0) { | |
5254 | ret = del_extent_zero(trans, extent_root, path, extent_key); | |
5255 | goto out; | |
5256 | } | |
5257 | ||
5258 | ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS); | |
5259 | if (!ref_path) { | |
5260 | ret = -ENOMEM; | |
5261 | goto out; | |
5262 | } | |
5263 | ||
5264 | for (loops = 0; ; loops++) { | |
5265 | if (loops == 0) { | |
5266 | ret = btrfs_first_ref_path(trans, extent_root, ref_path, | |
5267 | extent_key->objectid); | |
5268 | } else { | |
5269 | ret = btrfs_next_ref_path(trans, extent_root, ref_path); | |
5270 | } | |
5271 | if (ret < 0) | |
5272 | goto out; | |
5273 | if (ret > 0) | |
5274 | break; | |
5275 | ||
5276 | if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID || | |
5277 | ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
5278 | continue; | |
5279 | ||
5280 | found_root = read_ref_root(extent_root->fs_info, ref_path); | |
5281 | BUG_ON(!found_root); | |
5282 | /* | |
5283 | * for reference counted tree, only process reference paths | |
5284 | * rooted at the latest committed root. | |
5285 | */ | |
5286 | if (found_root->ref_cows && | |
5287 | ref_path->root_generation != found_root->root_key.offset) | |
5288 | continue; | |
5289 | ||
5290 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
5291 | if (pass == 0) { | |
5292 | /* | |
5293 | * copy data extents to new locations | |
5294 | */ | |
5295 | u64 group_start = group->key.objectid; | |
5296 | ret = relocate_data_extent(reloc_inode, | |
5297 | extent_key, | |
5298 | group_start); | |
5299 | if (ret < 0) | |
5300 | goto out; | |
5301 | break; | |
5302 | } | |
5303 | level = 0; | |
5304 | } else { | |
5305 | level = ref_path->owner_objectid; | |
5306 | } | |
5307 | ||
5308 | if (prev_block != ref_path->nodes[level]) { | |
5309 | struct extent_buffer *eb; | |
5310 | u64 block_start = ref_path->nodes[level]; | |
5311 | u64 block_size = btrfs_level_size(found_root, level); | |
5312 | ||
5313 | eb = read_tree_block(found_root, block_start, | |
5314 | block_size, 0); | |
5315 | btrfs_tree_lock(eb); | |
5316 | BUG_ON(level != btrfs_header_level(eb)); | |
5317 | ||
5318 | if (level == 0) | |
5319 | btrfs_item_key_to_cpu(eb, &first_key, 0); | |
5320 | else | |
5321 | btrfs_node_key_to_cpu(eb, &first_key, 0); | |
5322 | ||
5323 | btrfs_tree_unlock(eb); | |
5324 | free_extent_buffer(eb); | |
5325 | prev_block = block_start; | |
5326 | } | |
5327 | ||
5328 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID && | |
5329 | pass >= 2) { | |
5330 | /* | |
5331 | * use fallback method to process the remaining | |
5332 | * references. | |
5333 | */ | |
5334 | if (!new_extents) { | |
5335 | u64 group_start = group->key.objectid; | |
d899e052 YZ |
5336 | new_extents = kmalloc(sizeof(*new_extents), |
5337 | GFP_NOFS); | |
5338 | nr_extents = 1; | |
1a40e23b ZY |
5339 | ret = get_new_locations(reloc_inode, |
5340 | extent_key, | |
d899e052 | 5341 | group_start, 1, |
1a40e23b ZY |
5342 | &new_extents, |
5343 | &nr_extents); | |
d899e052 | 5344 | if (ret) |
1a40e23b ZY |
5345 | goto out; |
5346 | } | |
5347 | btrfs_record_root_in_trans(found_root); | |
5348 | ret = replace_one_extent(trans, found_root, | |
5349 | path, extent_key, | |
5350 | &first_key, ref_path, | |
5351 | new_extents, nr_extents); | |
5352 | if (ret < 0) | |
5353 | goto out; | |
5354 | continue; | |
5355 | } | |
5356 | ||
5357 | btrfs_record_root_in_trans(found_root); | |
5358 | if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
5359 | ret = relocate_tree_block(trans, found_root, path, | |
5360 | &first_key, ref_path); | |
5361 | } else { | |
5362 | /* | |
5363 | * try to update data extent references while | |
5364 | * keeping metadata shared between snapshots. | |
5365 | */ | |
5366 | ret = relocate_one_path(trans, found_root, path, | |
5367 | &first_key, ref_path, | |
5368 | group, reloc_inode); | |
5369 | } | |
5370 | if (ret < 0) | |
5371 | goto out; | |
5372 | } | |
5373 | ret = 0; | |
5374 | out: | |
5375 | btrfs_end_transaction(trans, extent_root); | |
5376 | kfree(new_extents); | |
5377 | kfree(ref_path); | |
1a40e23b ZY |
5378 | return ret; |
5379 | } | |
5380 | ||
ec44a35c CM |
5381 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
5382 | { | |
5383 | u64 num_devices; | |
5384 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
5385 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
5386 | ||
2b82032c | 5387 | num_devices = root->fs_info->fs_devices->rw_devices; |
ec44a35c CM |
5388 | if (num_devices == 1) { |
5389 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
5390 | stripped = flags & ~stripped; | |
5391 | ||
5392 | /* turn raid0 into single device chunks */ | |
5393 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
5394 | return stripped; | |
5395 | ||
5396 | /* turn mirroring into duplication */ | |
5397 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
5398 | BTRFS_BLOCK_GROUP_RAID10)) | |
5399 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
5400 | return flags; | |
5401 | } else { | |
5402 | /* they already had raid on here, just return */ | |
ec44a35c CM |
5403 | if (flags & stripped) |
5404 | return flags; | |
5405 | ||
5406 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
5407 | stripped = flags & ~stripped; | |
5408 | ||
5409 | /* switch duplicated blocks with raid1 */ | |
5410 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
5411 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
5412 | ||
5413 | /* turn single device chunks into raid0 */ | |
5414 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
5415 | } | |
5416 | return flags; | |
5417 | } | |
5418 | ||
0ef3e66b CM |
5419 | int __alloc_chunk_for_shrink(struct btrfs_root *root, |
5420 | struct btrfs_block_group_cache *shrink_block_group, | |
5421 | int force) | |
5422 | { | |
5423 | struct btrfs_trans_handle *trans; | |
5424 | u64 new_alloc_flags; | |
5425 | u64 calc; | |
5426 | ||
c286ac48 | 5427 | spin_lock(&shrink_block_group->lock); |
0ef3e66b | 5428 | if (btrfs_block_group_used(&shrink_block_group->item) > 0) { |
c286ac48 | 5429 | spin_unlock(&shrink_block_group->lock); |
c286ac48 | 5430 | |
0ef3e66b | 5431 | trans = btrfs_start_transaction(root, 1); |
c286ac48 | 5432 | spin_lock(&shrink_block_group->lock); |
7d9eb12c | 5433 | |
0ef3e66b CM |
5434 | new_alloc_flags = update_block_group_flags(root, |
5435 | shrink_block_group->flags); | |
5436 | if (new_alloc_flags != shrink_block_group->flags) { | |
5437 | calc = | |
5438 | btrfs_block_group_used(&shrink_block_group->item); | |
5439 | } else { | |
5440 | calc = shrink_block_group->key.offset; | |
5441 | } | |
c286ac48 CM |
5442 | spin_unlock(&shrink_block_group->lock); |
5443 | ||
0ef3e66b CM |
5444 | do_chunk_alloc(trans, root->fs_info->extent_root, |
5445 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
7d9eb12c | 5446 | |
0ef3e66b | 5447 | btrfs_end_transaction(trans, root); |
c286ac48 CM |
5448 | } else |
5449 | spin_unlock(&shrink_block_group->lock); | |
0ef3e66b CM |
5450 | return 0; |
5451 | } | |
5452 | ||
1a40e23b ZY |
5453 | static int __insert_orphan_inode(struct btrfs_trans_handle *trans, |
5454 | struct btrfs_root *root, | |
5455 | u64 objectid, u64 size) | |
5456 | { | |
5457 | struct btrfs_path *path; | |
5458 | struct btrfs_inode_item *item; | |
5459 | struct extent_buffer *leaf; | |
5460 | int ret; | |
5461 | ||
5462 | path = btrfs_alloc_path(); | |
5463 | if (!path) | |
5464 | return -ENOMEM; | |
5465 | ||
5466 | ret = btrfs_insert_empty_inode(trans, root, path, objectid); | |
5467 | if (ret) | |
5468 | goto out; | |
5469 | ||
5470 | leaf = path->nodes[0]; | |
5471 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); | |
5472 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
5473 | btrfs_set_inode_generation(leaf, item, 1); | |
5474 | btrfs_set_inode_size(leaf, item, size); | |
5475 | btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); | |
d899e052 YZ |
5476 | btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM | |
5477 | BTRFS_INODE_NOCOMPRESS); | |
1a40e23b ZY |
5478 | btrfs_mark_buffer_dirty(leaf); |
5479 | btrfs_release_path(root, path); | |
5480 | out: | |
5481 | btrfs_free_path(path); | |
5482 | return ret; | |
5483 | } | |
5484 | ||
5485 | static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info, | |
5486 | struct btrfs_block_group_cache *group) | |
5487 | { | |
5488 | struct inode *inode = NULL; | |
5489 | struct btrfs_trans_handle *trans; | |
5490 | struct btrfs_root *root; | |
5491 | struct btrfs_key root_key; | |
5492 | u64 objectid = BTRFS_FIRST_FREE_OBJECTID; | |
5493 | int err = 0; | |
5494 | ||
5495 | root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
5496 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
5497 | root_key.offset = (u64)-1; | |
5498 | root = btrfs_read_fs_root_no_name(fs_info, &root_key); | |
5499 | if (IS_ERR(root)) | |
5500 | return ERR_CAST(root); | |
5501 | ||
5502 | trans = btrfs_start_transaction(root, 1); | |
5503 | BUG_ON(!trans); | |
5504 | ||
5505 | err = btrfs_find_free_objectid(trans, root, objectid, &objectid); | |
5506 | if (err) | |
5507 | goto out; | |
5508 | ||
5509 | err = __insert_orphan_inode(trans, root, objectid, group->key.offset); | |
5510 | BUG_ON(err); | |
5511 | ||
5512 | err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0, | |
c8b97818 CM |
5513 | group->key.offset, 0, group->key.offset, |
5514 | 0, 0, 0); | |
1a40e23b ZY |
5515 | BUG_ON(err); |
5516 | ||
5517 | inode = btrfs_iget_locked(root->fs_info->sb, objectid, root); | |
5518 | if (inode->i_state & I_NEW) { | |
5519 | BTRFS_I(inode)->root = root; | |
5520 | BTRFS_I(inode)->location.objectid = objectid; | |
5521 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
5522 | BTRFS_I(inode)->location.offset = 0; | |
5523 | btrfs_read_locked_inode(inode); | |
5524 | unlock_new_inode(inode); | |
5525 | BUG_ON(is_bad_inode(inode)); | |
5526 | } else { | |
5527 | BUG_ON(1); | |
5528 | } | |
5529 | ||
5530 | err = btrfs_orphan_add(trans, inode); | |
5531 | out: | |
5532 | btrfs_end_transaction(trans, root); | |
5533 | if (err) { | |
5534 | if (inode) | |
5535 | iput(inode); | |
5536 | inode = ERR_PTR(err); | |
5537 | } | |
5538 | return inode; | |
5539 | } | |
5540 | ||
5541 | int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) | |
edbd8d4e CM |
5542 | { |
5543 | struct btrfs_trans_handle *trans; | |
edbd8d4e | 5544 | struct btrfs_path *path; |
1a40e23b ZY |
5545 | struct btrfs_fs_info *info = root->fs_info; |
5546 | struct extent_buffer *leaf; | |
5547 | struct inode *reloc_inode; | |
5548 | struct btrfs_block_group_cache *block_group; | |
5549 | struct btrfs_key key; | |
d899e052 | 5550 | u64 skipped; |
edbd8d4e CM |
5551 | u64 cur_byte; |
5552 | u64 total_found; | |
edbd8d4e CM |
5553 | u32 nritems; |
5554 | int ret; | |
a061fc8d | 5555 | int progress; |
1a40e23b | 5556 | int pass = 0; |
edbd8d4e | 5557 | |
1a40e23b ZY |
5558 | root = root->fs_info->extent_root; |
5559 | ||
5560 | block_group = btrfs_lookup_block_group(info, group_start); | |
5561 | BUG_ON(!block_group); | |
8f18cf13 | 5562 | |
1a40e23b ZY |
5563 | printk("btrfs relocating block group %llu flags %llu\n", |
5564 | (unsigned long long)block_group->key.objectid, | |
5565 | (unsigned long long)block_group->flags); | |
8f18cf13 | 5566 | |
edbd8d4e | 5567 | path = btrfs_alloc_path(); |
1a40e23b | 5568 | BUG_ON(!path); |
edbd8d4e | 5569 | |
1a40e23b ZY |
5570 | reloc_inode = create_reloc_inode(info, block_group); |
5571 | BUG_ON(IS_ERR(reloc_inode)); | |
323da79c | 5572 | |
1a40e23b | 5573 | __alloc_chunk_for_shrink(root, block_group, 1); |
c146afad | 5574 | set_block_group_readonly(block_group); |
323da79c | 5575 | |
1a40e23b ZY |
5576 | btrfs_start_delalloc_inodes(info->tree_root); |
5577 | btrfs_wait_ordered_extents(info->tree_root, 0); | |
5578 | again: | |
d899e052 | 5579 | skipped = 0; |
edbd8d4e | 5580 | total_found = 0; |
a061fc8d | 5581 | progress = 0; |
1a40e23b | 5582 | key.objectid = block_group->key.objectid; |
edbd8d4e CM |
5583 | key.offset = 0; |
5584 | key.type = 0; | |
73e48b27 | 5585 | cur_byte = key.objectid; |
4313b399 | 5586 | |
1a40e23b ZY |
5587 | trans = btrfs_start_transaction(info->tree_root, 1); |
5588 | btrfs_commit_transaction(trans, info->tree_root); | |
ea8c2819 | 5589 | |
1a40e23b ZY |
5590 | mutex_lock(&root->fs_info->cleaner_mutex); |
5591 | btrfs_clean_old_snapshots(info->tree_root); | |
5592 | btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); | |
5593 | mutex_unlock(&root->fs_info->cleaner_mutex); | |
ea8c2819 | 5594 | |
73e48b27 | 5595 | while(1) { |
edbd8d4e CM |
5596 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5597 | if (ret < 0) | |
5598 | goto out; | |
7d9eb12c | 5599 | next: |
edbd8d4e | 5600 | leaf = path->nodes[0]; |
73e48b27 | 5601 | nritems = btrfs_header_nritems(leaf); |
73e48b27 Y |
5602 | if (path->slots[0] >= nritems) { |
5603 | ret = btrfs_next_leaf(root, path); | |
5604 | if (ret < 0) | |
5605 | goto out; | |
5606 | if (ret == 1) { | |
5607 | ret = 0; | |
5608 | break; | |
edbd8d4e | 5609 | } |
73e48b27 Y |
5610 | leaf = path->nodes[0]; |
5611 | nritems = btrfs_header_nritems(leaf); | |
edbd8d4e | 5612 | } |
73e48b27 | 5613 | |
1a40e23b | 5614 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
725c8463 | 5615 | |
1a40e23b ZY |
5616 | if (key.objectid >= block_group->key.objectid + |
5617 | block_group->key.offset) | |
8f18cf13 CM |
5618 | break; |
5619 | ||
725c8463 | 5620 | if (progress && need_resched()) { |
725c8463 | 5621 | btrfs_release_path(root, path); |
1a40e23b | 5622 | cond_resched(); |
725c8463 | 5623 | progress = 0; |
1a40e23b | 5624 | continue; |
725c8463 CM |
5625 | } |
5626 | progress = 1; | |
5627 | ||
1a40e23b ZY |
5628 | if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY || |
5629 | key.objectid + key.offset <= cur_byte) { | |
edbd8d4e | 5630 | path->slots[0]++; |
edbd8d4e CM |
5631 | goto next; |
5632 | } | |
73e48b27 | 5633 | |
edbd8d4e | 5634 | total_found++; |
1a40e23b | 5635 | cur_byte = key.objectid + key.offset; |
edbd8d4e | 5636 | btrfs_release_path(root, path); |
edbd8d4e | 5637 | |
1a40e23b ZY |
5638 | __alloc_chunk_for_shrink(root, block_group, 0); |
5639 | ret = relocate_one_extent(root, path, &key, block_group, | |
5640 | reloc_inode, pass); | |
5641 | BUG_ON(ret < 0); | |
d899e052 YZ |
5642 | if (ret > 0) |
5643 | skipped++; | |
edbd8d4e | 5644 | |
1a40e23b ZY |
5645 | key.objectid = cur_byte; |
5646 | key.type = 0; | |
5647 | key.offset = 0; | |
edbd8d4e CM |
5648 | } |
5649 | ||
1a40e23b | 5650 | btrfs_release_path(root, path); |
7d9eb12c | 5651 | |
1a40e23b ZY |
5652 | if (pass == 0) { |
5653 | btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1); | |
5654 | invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1); | |
5655 | WARN_ON(reloc_inode->i_mapping->nrpages); | |
8e8a1e31 | 5656 | } |
73e48b27 | 5657 | |
1a40e23b ZY |
5658 | if (total_found > 0) { |
5659 | printk("btrfs found %llu extents in pass %d\n", | |
5660 | (unsigned long long)total_found, pass); | |
5661 | pass++; | |
d899e052 YZ |
5662 | if (total_found == skipped && pass > 2) { |
5663 | iput(reloc_inode); | |
5664 | reloc_inode = create_reloc_inode(info, block_group); | |
5665 | pass = 0; | |
5666 | } | |
1a40e23b | 5667 | goto again; |
0f9dd46c | 5668 | } |
0ef3e66b | 5669 | |
1a40e23b ZY |
5670 | /* delete reloc_inode */ |
5671 | iput(reloc_inode); | |
5672 | ||
5673 | /* unpin extents in this range */ | |
5674 | trans = btrfs_start_transaction(info->tree_root, 1); | |
5675 | btrfs_commit_transaction(trans, info->tree_root); | |
0ef3e66b | 5676 | |
1a40e23b ZY |
5677 | spin_lock(&block_group->lock); |
5678 | WARN_ON(block_group->pinned > 0); | |
5679 | WARN_ON(block_group->reserved > 0); | |
5680 | WARN_ON(btrfs_block_group_used(&block_group->item) > 0); | |
5681 | spin_unlock(&block_group->lock); | |
5682 | ret = 0; | |
edbd8d4e | 5683 | out: |
1a40e23b | 5684 | btrfs_free_path(path); |
edbd8d4e CM |
5685 | return ret; |
5686 | } | |
5687 | ||
0b86a832 CM |
5688 | int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, |
5689 | struct btrfs_key *key) | |
5690 | { | |
925baedd | 5691 | int ret = 0; |
0b86a832 CM |
5692 | struct btrfs_key found_key; |
5693 | struct extent_buffer *leaf; | |
5694 | int slot; | |
edbd8d4e | 5695 | |
0b86a832 CM |
5696 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
5697 | if (ret < 0) | |
925baedd CM |
5698 | goto out; |
5699 | ||
0b86a832 CM |
5700 | while(1) { |
5701 | slot = path->slots[0]; | |
edbd8d4e | 5702 | leaf = path->nodes[0]; |
0b86a832 CM |
5703 | if (slot >= btrfs_header_nritems(leaf)) { |
5704 | ret = btrfs_next_leaf(root, path); | |
5705 | if (ret == 0) | |
5706 | continue; | |
5707 | if (ret < 0) | |
925baedd | 5708 | goto out; |
0b86a832 | 5709 | break; |
edbd8d4e | 5710 | } |
0b86a832 | 5711 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 5712 | |
0b86a832 | 5713 | if (found_key.objectid >= key->objectid && |
925baedd CM |
5714 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5715 | ret = 0; | |
5716 | goto out; | |
5717 | } | |
0b86a832 | 5718 | path->slots[0]++; |
edbd8d4e | 5719 | } |
0b86a832 | 5720 | ret = -ENOENT; |
925baedd | 5721 | out: |
0b86a832 | 5722 | return ret; |
edbd8d4e CM |
5723 | } |
5724 | ||
1a40e23b ZY |
5725 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
5726 | { | |
5727 | struct btrfs_block_group_cache *block_group; | |
5728 | struct rb_node *n; | |
5729 | ||
1a40e23b ZY |
5730 | spin_lock(&info->block_group_cache_lock); |
5731 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
5732 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
5733 | cache_node); | |
1a40e23b ZY |
5734 | rb_erase(&block_group->cache_node, |
5735 | &info->block_group_cache_tree); | |
d899e052 YZ |
5736 | spin_unlock(&info->block_group_cache_lock); |
5737 | ||
5738 | btrfs_remove_free_space_cache(block_group); | |
80eb234a | 5739 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 5740 | list_del(&block_group->list); |
80eb234a | 5741 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 5742 | kfree(block_group); |
d899e052 YZ |
5743 | |
5744 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
5745 | } |
5746 | spin_unlock(&info->block_group_cache_lock); | |
1a40e23b ZY |
5747 | return 0; |
5748 | } | |
5749 | ||
9078a3e1 CM |
5750 | int btrfs_read_block_groups(struct btrfs_root *root) |
5751 | { | |
5752 | struct btrfs_path *path; | |
5753 | int ret; | |
9078a3e1 | 5754 | struct btrfs_block_group_cache *cache; |
be744175 | 5755 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 5756 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
5757 | struct btrfs_key key; |
5758 | struct btrfs_key found_key; | |
5f39d397 | 5759 | struct extent_buffer *leaf; |
96b5179d | 5760 | |
be744175 | 5761 | root = info->extent_root; |
9078a3e1 | 5762 | key.objectid = 0; |
0b86a832 | 5763 | key.offset = 0; |
9078a3e1 | 5764 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
5765 | path = btrfs_alloc_path(); |
5766 | if (!path) | |
5767 | return -ENOMEM; | |
5768 | ||
5769 | while(1) { | |
0b86a832 CM |
5770 | ret = find_first_block_group(root, path, &key); |
5771 | if (ret > 0) { | |
5772 | ret = 0; | |
5773 | goto error; | |
9078a3e1 | 5774 | } |
0b86a832 CM |
5775 | if (ret != 0) |
5776 | goto error; | |
5777 | ||
5f39d397 CM |
5778 | leaf = path->nodes[0]; |
5779 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 5780 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 5781 | if (!cache) { |
0b86a832 | 5782 | ret = -ENOMEM; |
9078a3e1 CM |
5783 | break; |
5784 | } | |
3e1ad54f | 5785 | |
c286ac48 | 5786 | spin_lock_init(&cache->lock); |
25179201 | 5787 | mutex_init(&cache->alloc_mutex); |
0f9dd46c | 5788 | INIT_LIST_HEAD(&cache->list); |
5f39d397 CM |
5789 | read_extent_buffer(leaf, &cache->item, |
5790 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
5791 | sizeof(cache->item)); | |
9078a3e1 | 5792 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 5793 | |
9078a3e1 CM |
5794 | key.objectid = found_key.objectid + found_key.offset; |
5795 | btrfs_release_path(root, path); | |
0b86a832 | 5796 | cache->flags = btrfs_block_group_flags(&cache->item); |
96b5179d | 5797 | |
6324fbf3 CM |
5798 | ret = update_space_info(info, cache->flags, found_key.offset, |
5799 | btrfs_block_group_used(&cache->item), | |
5800 | &space_info); | |
5801 | BUG_ON(ret); | |
5802 | cache->space_info = space_info; | |
80eb234a JB |
5803 | down_write(&space_info->groups_sem); |
5804 | list_add_tail(&cache->list, &space_info->block_groups); | |
5805 | up_write(&space_info->groups_sem); | |
0f9dd46c JB |
5806 | |
5807 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
5808 | BUG_ON(ret); | |
75ccf47d CM |
5809 | |
5810 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c YZ |
5811 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
5812 | set_block_group_readonly(cache); | |
9078a3e1 | 5813 | } |
0b86a832 CM |
5814 | ret = 0; |
5815 | error: | |
9078a3e1 | 5816 | btrfs_free_path(path); |
0b86a832 | 5817 | return ret; |
9078a3e1 | 5818 | } |
6324fbf3 CM |
5819 | |
5820 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
5821 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 5822 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
5823 | u64 size) |
5824 | { | |
5825 | int ret; | |
6324fbf3 CM |
5826 | struct btrfs_root *extent_root; |
5827 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
5828 | |
5829 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 5830 | |
e02119d5 CM |
5831 | root->fs_info->last_trans_new_blockgroup = trans->transid; |
5832 | ||
8f18cf13 | 5833 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
5834 | if (!cache) |
5835 | return -ENOMEM; | |
5836 | ||
e17cade2 | 5837 | cache->key.objectid = chunk_offset; |
6324fbf3 | 5838 | cache->key.offset = size; |
c286ac48 | 5839 | spin_lock_init(&cache->lock); |
25179201 | 5840 | mutex_init(&cache->alloc_mutex); |
0f9dd46c | 5841 | INIT_LIST_HEAD(&cache->list); |
6324fbf3 | 5842 | btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
0ef3e66b | 5843 | |
6324fbf3 | 5844 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
5845 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
5846 | cache->flags = type; | |
5847 | btrfs_set_block_group_flags(&cache->item, type); | |
5848 | ||
5849 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, | |
5850 | &cache->space_info); | |
5851 | BUG_ON(ret); | |
80eb234a JB |
5852 | down_write(&cache->space_info->groups_sem); |
5853 | list_add_tail(&cache->list, &cache->space_info->block_groups); | |
5854 | up_write(&cache->space_info->groups_sem); | |
6324fbf3 | 5855 | |
0f9dd46c JB |
5856 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
5857 | BUG_ON(ret); | |
c286ac48 | 5858 | |
6324fbf3 CM |
5859 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
5860 | sizeof(cache->item)); | |
5861 | BUG_ON(ret); | |
5862 | ||
87ef2bb4 CM |
5863 | finish_current_insert(trans, extent_root, 0); |
5864 | ret = del_pending_extents(trans, extent_root, 0); | |
6324fbf3 | 5865 | BUG_ON(ret); |
d18a2c44 | 5866 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 5867 | |
6324fbf3 CM |
5868 | return 0; |
5869 | } | |
1a40e23b ZY |
5870 | |
5871 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
5872 | struct btrfs_root *root, u64 group_start) | |
5873 | { | |
5874 | struct btrfs_path *path; | |
5875 | struct btrfs_block_group_cache *block_group; | |
5876 | struct btrfs_key key; | |
5877 | int ret; | |
5878 | ||
1a40e23b ZY |
5879 | root = root->fs_info->extent_root; |
5880 | ||
5881 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
5882 | BUG_ON(!block_group); | |
c146afad | 5883 | BUG_ON(!block_group->ro); |
1a40e23b ZY |
5884 | |
5885 | memcpy(&key, &block_group->key, sizeof(key)); | |
5886 | ||
5887 | path = btrfs_alloc_path(); | |
5888 | BUG_ON(!path); | |
5889 | ||
5890 | btrfs_remove_free_space_cache(block_group); | |
5891 | rb_erase(&block_group->cache_node, | |
5892 | &root->fs_info->block_group_cache_tree); | |
80eb234a | 5893 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 5894 | list_del(&block_group->list); |
80eb234a | 5895 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 5896 | |
c146afad YZ |
5897 | spin_lock(&block_group->space_info->lock); |
5898 | block_group->space_info->total_bytes -= block_group->key.offset; | |
5899 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
5900 | spin_unlock(&block_group->space_info->lock); | |
2b82032c | 5901 | block_group->space_info->full = 0; |
c146afad | 5902 | |
1a40e23b ZY |
5903 | /* |
5904 | memset(shrink_block_group, 0, sizeof(*shrink_block_group)); | |
5905 | kfree(shrink_block_group); | |
5906 | */ | |
5907 | ||
5908 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5909 | if (ret > 0) | |
5910 | ret = -EIO; | |
5911 | if (ret < 0) | |
5912 | goto out; | |
5913 | ||
5914 | ret = btrfs_del_item(trans, root, path); | |
5915 | out: | |
5916 | btrfs_free_path(path); | |
5917 | return ret; | |
5918 | } |