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; | |
45 | }; | |
46 | ||
e089f05c | 47 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
87ef2bb4 | 48 | btrfs_root *extent_root, int all); |
e20d96d6 | 49 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
87ef2bb4 | 50 | btrfs_root *extent_root, int all); |
925baedd CM |
51 | static struct btrfs_block_group_cache * |
52 | __btrfs_find_block_group(struct btrfs_root *root, | |
53 | struct btrfs_block_group_cache *hint, | |
54 | u64 search_start, int data, int owner); | |
d548ee51 | 55 | |
0f9dd46c JB |
56 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
57 | { | |
58 | return (cache->flags & bits) == bits; | |
59 | } | |
60 | ||
61 | /* | |
62 | * this adds the block group to the fs_info rb tree for the block group | |
63 | * cache | |
64 | */ | |
65 | int btrfs_add_block_group_cache(struct btrfs_fs_info *info, | |
66 | struct btrfs_block_group_cache *block_group) | |
67 | { | |
68 | struct rb_node **p; | |
69 | struct rb_node *parent = NULL; | |
70 | struct btrfs_block_group_cache *cache; | |
71 | ||
72 | spin_lock(&info->block_group_cache_lock); | |
73 | p = &info->block_group_cache_tree.rb_node; | |
74 | ||
75 | while (*p) { | |
76 | parent = *p; | |
77 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
78 | cache_node); | |
79 | if (block_group->key.objectid < cache->key.objectid) { | |
80 | p = &(*p)->rb_left; | |
81 | } else if (block_group->key.objectid > cache->key.objectid) { | |
82 | p = &(*p)->rb_right; | |
83 | } else { | |
84 | spin_unlock(&info->block_group_cache_lock); | |
85 | return -EEXIST; | |
86 | } | |
87 | } | |
88 | ||
89 | rb_link_node(&block_group->cache_node, parent, p); | |
90 | rb_insert_color(&block_group->cache_node, | |
91 | &info->block_group_cache_tree); | |
92 | spin_unlock(&info->block_group_cache_lock); | |
93 | ||
94 | return 0; | |
95 | } | |
96 | ||
97 | /* | |
98 | * This will return the block group at or after bytenr if contains is 0, else | |
99 | * it will return the block group that contains the bytenr | |
100 | */ | |
101 | static struct btrfs_block_group_cache * | |
102 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
103 | int contains) | |
104 | { | |
105 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
106 | struct rb_node *n; | |
107 | u64 end, start; | |
108 | ||
109 | spin_lock(&info->block_group_cache_lock); | |
110 | n = info->block_group_cache_tree.rb_node; | |
111 | ||
112 | while (n) { | |
113 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
114 | cache_node); | |
115 | end = cache->key.objectid + cache->key.offset - 1; | |
116 | start = cache->key.objectid; | |
117 | ||
118 | if (bytenr < start) { | |
119 | if (!contains && (!ret || start < ret->key.objectid)) | |
120 | ret = cache; | |
121 | n = n->rb_left; | |
122 | } else if (bytenr > start) { | |
123 | if (contains && bytenr <= end) { | |
124 | ret = cache; | |
125 | break; | |
126 | } | |
127 | n = n->rb_right; | |
128 | } else { | |
129 | ret = cache; | |
130 | break; | |
131 | } | |
132 | } | |
133 | spin_unlock(&info->block_group_cache_lock); | |
134 | ||
135 | return ret; | |
136 | } | |
137 | ||
138 | /* | |
139 | * this is only called by cache_block_group, since we could have freed extents | |
140 | * we need to check the pinned_extents for any extents that can't be used yet | |
141 | * since their free space will be released as soon as the transaction commits. | |
142 | */ | |
143 | static int add_new_free_space(struct btrfs_block_group_cache *block_group, | |
144 | struct btrfs_fs_info *info, u64 start, u64 end) | |
145 | { | |
146 | u64 extent_start, extent_end, size; | |
147 | int ret; | |
148 | ||
25179201 | 149 | mutex_lock(&info->pinned_mutex); |
0f9dd46c JB |
150 | while (start < end) { |
151 | ret = find_first_extent_bit(&info->pinned_extents, start, | |
152 | &extent_start, &extent_end, | |
153 | EXTENT_DIRTY); | |
154 | if (ret) | |
155 | break; | |
156 | ||
157 | if (extent_start == start) { | |
158 | start = extent_end + 1; | |
159 | } else if (extent_start > start && extent_start < end) { | |
160 | size = extent_start - start; | |
25179201 JB |
161 | ret = btrfs_add_free_space_lock(block_group, start, |
162 | size); | |
0f9dd46c JB |
163 | BUG_ON(ret); |
164 | start = extent_end + 1; | |
165 | } else { | |
166 | break; | |
167 | } | |
168 | } | |
169 | ||
170 | if (start < end) { | |
171 | size = end - start; | |
25179201 | 172 | ret = btrfs_add_free_space_lock(block_group, start, size); |
0f9dd46c JB |
173 | BUG_ON(ret); |
174 | } | |
25179201 | 175 | mutex_unlock(&info->pinned_mutex); |
0f9dd46c JB |
176 | |
177 | return 0; | |
178 | } | |
179 | ||
e37c9e69 CM |
180 | static int cache_block_group(struct btrfs_root *root, |
181 | struct btrfs_block_group_cache *block_group) | |
182 | { | |
183 | struct btrfs_path *path; | |
ef8bbdfe | 184 | int ret = 0; |
e37c9e69 | 185 | struct btrfs_key key; |
5f39d397 | 186 | struct extent_buffer *leaf; |
e37c9e69 | 187 | int slot; |
e37c9e69 | 188 | u64 last = 0; |
7d7d6068 | 189 | u64 first_free; |
e37c9e69 CM |
190 | int found = 0; |
191 | ||
00f5c795 CM |
192 | if (!block_group) |
193 | return 0; | |
194 | ||
e37c9e69 | 195 | root = root->fs_info->extent_root; |
e37c9e69 CM |
196 | |
197 | if (block_group->cached) | |
198 | return 0; | |
f510cfec | 199 | |
e37c9e69 CM |
200 | path = btrfs_alloc_path(); |
201 | if (!path) | |
202 | return -ENOMEM; | |
7d7d6068 | 203 | |
2cc58cf2 | 204 | path->reada = 2; |
5cd57b2c CM |
205 | /* |
206 | * we get into deadlocks with paths held by callers of this function. | |
207 | * since the alloc_mutex is protecting things right now, just | |
208 | * skip the locking here | |
209 | */ | |
210 | path->skip_locking = 1; | |
0f9dd46c JB |
211 | first_free = max_t(u64, block_group->key.objectid, |
212 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
e37c9e69 | 213 | key.objectid = block_group->key.objectid; |
e37c9e69 CM |
214 | key.offset = 0; |
215 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
216 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
217 | if (ret < 0) | |
ef8bbdfe | 218 | goto err; |
0b86a832 | 219 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
d548ee51 | 220 | if (ret < 0) |
ef8bbdfe | 221 | goto err; |
d548ee51 Y |
222 | if (ret == 0) { |
223 | leaf = path->nodes[0]; | |
224 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
225 | if (key.objectid + key.offset > first_free) | |
226 | first_free = key.objectid + key.offset; | |
227 | } | |
e37c9e69 | 228 | while(1) { |
5f39d397 | 229 | leaf = path->nodes[0]; |
e37c9e69 | 230 | slot = path->slots[0]; |
5f39d397 | 231 | if (slot >= btrfs_header_nritems(leaf)) { |
e37c9e69 | 232 | ret = btrfs_next_leaf(root, path); |
54aa1f4d CM |
233 | if (ret < 0) |
234 | goto err; | |
0f9dd46c | 235 | if (ret == 0) |
e37c9e69 | 236 | continue; |
0f9dd46c | 237 | else |
e37c9e69 | 238 | break; |
e37c9e69 | 239 | } |
5f39d397 | 240 | btrfs_item_key_to_cpu(leaf, &key, slot); |
0f9dd46c | 241 | if (key.objectid < block_group->key.objectid) |
7d7d6068 | 242 | goto next; |
0f9dd46c | 243 | |
e37c9e69 | 244 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 245 | block_group->key.offset) |
e37c9e69 | 246 | break; |
7d7d6068 | 247 | |
e37c9e69 CM |
248 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
249 | if (!found) { | |
7d7d6068 | 250 | last = first_free; |
e37c9e69 | 251 | found = 1; |
e37c9e69 | 252 | } |
0f9dd46c JB |
253 | |
254 | add_new_free_space(block_group, root->fs_info, last, | |
255 | key.objectid); | |
256 | ||
7d7d6068 | 257 | last = key.objectid + key.offset; |
e37c9e69 | 258 | } |
7d7d6068 | 259 | next: |
e37c9e69 CM |
260 | path->slots[0]++; |
261 | } | |
262 | ||
7d7d6068 Y |
263 | if (!found) |
264 | last = first_free; | |
0f9dd46c JB |
265 | |
266 | add_new_free_space(block_group, root->fs_info, last, | |
267 | block_group->key.objectid + | |
268 | block_group->key.offset); | |
269 | ||
e37c9e69 | 270 | block_group->cached = 1; |
ef8bbdfe | 271 | ret = 0; |
54aa1f4d | 272 | err: |
e37c9e69 | 273 | btrfs_free_path(path); |
ef8bbdfe | 274 | return ret; |
e37c9e69 CM |
275 | } |
276 | ||
0f9dd46c JB |
277 | /* |
278 | * return the block group that starts at or after bytenr | |
279 | */ | |
0ef3e66b CM |
280 | struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct |
281 | btrfs_fs_info *info, | |
282 | u64 bytenr) | |
283 | { | |
0f9dd46c | 284 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 285 | |
0f9dd46c | 286 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 287 | |
0f9dd46c | 288 | return cache; |
0ef3e66b CM |
289 | } |
290 | ||
0f9dd46c JB |
291 | /* |
292 | * return the block group that contains teh given bytenr | |
293 | */ | |
5276aeda CM |
294 | struct btrfs_block_group_cache *btrfs_lookup_block_group(struct |
295 | btrfs_fs_info *info, | |
db94535d | 296 | u64 bytenr) |
be744175 | 297 | { |
0f9dd46c | 298 | struct btrfs_block_group_cache *cache; |
be744175 | 299 | |
0f9dd46c | 300 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 301 | |
0f9dd46c | 302 | return cache; |
be744175 | 303 | } |
0b86a832 | 304 | |
0f9dd46c JB |
305 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
306 | u64 flags) | |
6324fbf3 | 307 | { |
0f9dd46c JB |
308 | struct list_head *head = &info->space_info; |
309 | struct list_head *cur; | |
310 | struct btrfs_space_info *found; | |
311 | list_for_each(cur, head) { | |
312 | found = list_entry(cur, struct btrfs_space_info, list); | |
313 | if (found->flags == flags) | |
314 | return found; | |
315 | } | |
316 | return NULL; | |
6324fbf3 CM |
317 | } |
318 | ||
80eb234a JB |
319 | static u64 div_factor(u64 num, int factor) |
320 | { | |
321 | if (factor == 10) | |
322 | return num; | |
323 | num *= factor; | |
324 | do_div(num, 10); | |
325 | return num; | |
326 | } | |
327 | ||
925baedd CM |
328 | static struct btrfs_block_group_cache * |
329 | __btrfs_find_block_group(struct btrfs_root *root, | |
330 | struct btrfs_block_group_cache *hint, | |
331 | u64 search_start, int data, int owner) | |
cd1bc465 | 332 | { |
96b5179d | 333 | struct btrfs_block_group_cache *cache; |
31f3c99b | 334 | struct btrfs_block_group_cache *found_group = NULL; |
cd1bc465 CM |
335 | struct btrfs_fs_info *info = root->fs_info; |
336 | u64 used; | |
31f3c99b | 337 | u64 last = 0; |
96b5179d | 338 | u64 free_check; |
31f3c99b | 339 | int full_search = 0; |
bce4eae9 | 340 | int factor = 10; |
0ef3e66b | 341 | int wrapped = 0; |
de428b63 | 342 | |
a236aed1 CM |
343 | if (data & BTRFS_BLOCK_GROUP_METADATA) |
344 | factor = 9; | |
be744175 | 345 | |
0ef3e66b | 346 | if (search_start) { |
be744175 | 347 | struct btrfs_block_group_cache *shint; |
0ef3e66b | 348 | shint = btrfs_lookup_first_block_group(info, search_start); |
8f18cf13 | 349 | if (shint && block_group_bits(shint, data) && !shint->ro) { |
c286ac48 | 350 | spin_lock(&shint->lock); |
be744175 | 351 | used = btrfs_block_group_used(&shint->item); |
e8569813 | 352 | if (used + shint->pinned + shint->reserved < |
324ae4df | 353 | div_factor(shint->key.offset, factor)) { |
c286ac48 | 354 | spin_unlock(&shint->lock); |
be744175 CM |
355 | return shint; |
356 | } | |
c286ac48 | 357 | spin_unlock(&shint->lock); |
be744175 CM |
358 | } |
359 | } | |
0ef3e66b | 360 | if (hint && !hint->ro && block_group_bits(hint, data)) { |
c286ac48 | 361 | spin_lock(&hint->lock); |
31f3c99b | 362 | used = btrfs_block_group_used(&hint->item); |
e8569813 | 363 | if (used + hint->pinned + hint->reserved < |
324ae4df | 364 | div_factor(hint->key.offset, factor)) { |
c286ac48 | 365 | spin_unlock(&hint->lock); |
31f3c99b CM |
366 | return hint; |
367 | } | |
c286ac48 | 368 | spin_unlock(&hint->lock); |
e19caa5f | 369 | last = hint->key.objectid + hint->key.offset; |
31f3c99b | 370 | } else { |
e37c9e69 | 371 | if (hint) |
0ef3e66b | 372 | last = max(hint->key.objectid, search_start); |
e37c9e69 | 373 | else |
0ef3e66b | 374 | last = search_start; |
31f3c99b | 375 | } |
31f3c99b | 376 | again: |
e8569813 ZY |
377 | while (1) { |
378 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
379 | if (!cache) |
380 | break; | |
96b5179d | 381 | |
c286ac48 | 382 | spin_lock(&cache->lock); |
96b5179d CM |
383 | last = cache->key.objectid + cache->key.offset; |
384 | used = btrfs_block_group_used(&cache->item); | |
385 | ||
8f18cf13 | 386 | if (!cache->ro && block_group_bits(cache, data)) { |
0ef3e66b | 387 | free_check = div_factor(cache->key.offset, factor); |
e8569813 ZY |
388 | if (used + cache->pinned + cache->reserved < |
389 | free_check) { | |
8790d502 | 390 | found_group = cache; |
c286ac48 | 391 | spin_unlock(&cache->lock); |
8790d502 CM |
392 | goto found; |
393 | } | |
6324fbf3 | 394 | } |
c286ac48 | 395 | spin_unlock(&cache->lock); |
de428b63 | 396 | cond_resched(); |
cd1bc465 | 397 | } |
0ef3e66b CM |
398 | if (!wrapped) { |
399 | last = search_start; | |
400 | wrapped = 1; | |
401 | goto again; | |
402 | } | |
403 | if (!full_search && factor < 10) { | |
be744175 | 404 | last = search_start; |
31f3c99b | 405 | full_search = 1; |
0ef3e66b | 406 | factor = 10; |
31f3c99b CM |
407 | goto again; |
408 | } | |
be744175 | 409 | found: |
31f3c99b | 410 | return found_group; |
cd1bc465 CM |
411 | } |
412 | ||
925baedd CM |
413 | struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, |
414 | struct btrfs_block_group_cache | |
415 | *hint, u64 search_start, | |
416 | int data, int owner) | |
417 | { | |
418 | ||
419 | struct btrfs_block_group_cache *ret; | |
925baedd | 420 | ret = __btrfs_find_block_group(root, hint, search_start, data, owner); |
925baedd CM |
421 | return ret; |
422 | } | |
0f9dd46c | 423 | |
e02119d5 | 424 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 425 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
426 | { |
427 | int ret; | |
428 | struct btrfs_key key; | |
31840ae1 | 429 | struct btrfs_path *path; |
e02119d5 | 430 | |
31840ae1 ZY |
431 | path = btrfs_alloc_path(); |
432 | BUG_ON(!path); | |
e02119d5 CM |
433 | key.objectid = start; |
434 | key.offset = len; | |
435 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
436 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
437 | 0, 0); | |
31840ae1 | 438 | btrfs_free_path(path); |
7bb86316 CM |
439 | return ret; |
440 | } | |
441 | ||
d8d5f3e1 CM |
442 | /* |
443 | * Back reference rules. Back refs have three main goals: | |
444 | * | |
445 | * 1) differentiate between all holders of references to an extent so that | |
446 | * when a reference is dropped we can make sure it was a valid reference | |
447 | * before freeing the extent. | |
448 | * | |
449 | * 2) Provide enough information to quickly find the holders of an extent | |
450 | * if we notice a given block is corrupted or bad. | |
451 | * | |
452 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
453 | * maintenance. This is actually the same as #2, but with a slightly | |
454 | * different use case. | |
455 | * | |
456 | * File extents can be referenced by: | |
457 | * | |
458 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 459 | * - different files inside a single subvolume |
d8d5f3e1 CM |
460 | * - different offsets inside a file (bookend extents in file.c) |
461 | * | |
462 | * The extent ref structure has fields for: | |
463 | * | |
464 | * - Objectid of the subvolume root | |
465 | * - Generation number of the tree holding the reference | |
466 | * - objectid of the file holding the reference | |
31840ae1 ZY |
467 | * - number of references holding by parent node (alway 1 for tree blocks) |
468 | * | |
469 | * Btree leaf may hold multiple references to a file extent. In most cases, | |
470 | * these references are from same file and the corresponding offsets inside | |
3bb1a1bc | 471 | * the file are close together. |
d8d5f3e1 CM |
472 | * |
473 | * When a file extent is allocated the fields are filled in: | |
3bb1a1bc | 474 | * (root_key.objectid, trans->transid, inode objectid, 1) |
d8d5f3e1 CM |
475 | * |
476 | * When a leaf is cow'd new references are added for every file extent found | |
31840ae1 ZY |
477 | * in the leaf. It looks similar to the create case, but trans->transid will |
478 | * be different when the block is cow'd. | |
d8d5f3e1 | 479 | * |
3bb1a1bc | 480 | * (root_key.objectid, trans->transid, inode objectid, |
31840ae1 | 481 | * number of references in the leaf) |
d8d5f3e1 | 482 | * |
3bb1a1bc YZ |
483 | * When a file extent is removed either during snapshot deletion or |
484 | * file truncation, we find the corresponding back reference and check | |
485 | * the following fields: | |
d8d5f3e1 | 486 | * |
3bb1a1bc YZ |
487 | * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), |
488 | * inode objectid) | |
d8d5f3e1 CM |
489 | * |
490 | * Btree extents can be referenced by: | |
491 | * | |
492 | * - Different subvolumes | |
493 | * - Different generations of the same subvolume | |
494 | * | |
d8d5f3e1 CM |
495 | * When a tree block is created, back references are inserted: |
496 | * | |
3bb1a1bc | 497 | * (root->root_key.objectid, trans->transid, level, 1) |
d8d5f3e1 | 498 | * |
31840ae1 ZY |
499 | * When a tree block is cow'd, new back references are added for all the |
500 | * blocks it points to. If the tree block isn't in reference counted root, | |
501 | * the old back references are removed. These new back references are of | |
502 | * the form (trans->transid will have increased since creation): | |
d8d5f3e1 | 503 | * |
3bb1a1bc | 504 | * (root->root_key.objectid, trans->transid, level, 1) |
d8d5f3e1 | 505 | * |
31840ae1 | 506 | * When a backref is in deleting, the following fields are checked: |
d8d5f3e1 CM |
507 | * |
508 | * if backref was for a tree root: | |
3bb1a1bc | 509 | * (btrfs_header_owner(itself), btrfs_header_generation(itself), level) |
d8d5f3e1 | 510 | * else |
3bb1a1bc | 511 | * (btrfs_header_owner(parent), btrfs_header_generation(parent), level) |
d8d5f3e1 | 512 | * |
31840ae1 | 513 | * Back Reference Key composing: |
d8d5f3e1 | 514 | * |
31840ae1 ZY |
515 | * The key objectid corresponds to the first byte in the extent, the key |
516 | * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first | |
517 | * byte of parent extent. If a extent is tree root, the key offset is set | |
518 | * to the key objectid. | |
d8d5f3e1 | 519 | */ |
31840ae1 ZY |
520 | |
521 | static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, | |
522 | struct btrfs_root *root, | |
3bb1a1bc YZ |
523 | struct btrfs_path *path, |
524 | u64 bytenr, u64 parent, | |
525 | u64 ref_root, u64 ref_generation, | |
526 | u64 owner_objectid, int del) | |
7bb86316 | 527 | { |
7bb86316 | 528 | struct btrfs_key key; |
31840ae1 ZY |
529 | struct btrfs_extent_ref *ref; |
530 | struct extent_buffer *leaf; | |
3bb1a1bc | 531 | u64 ref_objectid; |
74493f7a CM |
532 | int ret; |
533 | ||
31840ae1 ZY |
534 | key.objectid = bytenr; |
535 | key.type = BTRFS_EXTENT_REF_KEY; | |
536 | key.offset = parent; | |
537 | ||
538 | ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1); | |
539 | if (ret < 0) | |
540 | goto out; | |
541 | if (ret > 0) { | |
542 | ret = -ENOENT; | |
543 | goto out; | |
544 | } | |
545 | ||
546 | leaf = path->nodes[0]; | |
547 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); | |
3bb1a1bc | 548 | ref_objectid = btrfs_ref_objectid(leaf, ref); |
31840ae1 | 549 | if (btrfs_ref_root(leaf, ref) != ref_root || |
3bb1a1bc YZ |
550 | btrfs_ref_generation(leaf, ref) != ref_generation || |
551 | (ref_objectid != owner_objectid && | |
552 | ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) { | |
31840ae1 ZY |
553 | ret = -EIO; |
554 | WARN_ON(1); | |
555 | goto out; | |
556 | } | |
557 | ret = 0; | |
558 | out: | |
559 | return ret; | |
560 | } | |
561 | ||
562 | static int noinline insert_extent_backref(struct btrfs_trans_handle *trans, | |
563 | struct btrfs_root *root, | |
564 | struct btrfs_path *path, | |
565 | u64 bytenr, u64 parent, | |
566 | u64 ref_root, u64 ref_generation, | |
3bb1a1bc | 567 | u64 owner_objectid) |
31840ae1 ZY |
568 | { |
569 | struct btrfs_key key; | |
570 | struct extent_buffer *leaf; | |
571 | struct btrfs_extent_ref *ref; | |
572 | u32 num_refs; | |
573 | int ret; | |
74493f7a | 574 | |
74493f7a CM |
575 | key.objectid = bytenr; |
576 | key.type = BTRFS_EXTENT_REF_KEY; | |
31840ae1 | 577 | key.offset = parent; |
74493f7a | 578 | |
31840ae1 ZY |
579 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref)); |
580 | if (ret == 0) { | |
581 | leaf = path->nodes[0]; | |
582 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
583 | struct btrfs_extent_ref); | |
584 | btrfs_set_ref_root(leaf, ref, ref_root); | |
585 | btrfs_set_ref_generation(leaf, ref, ref_generation); | |
586 | btrfs_set_ref_objectid(leaf, ref, owner_objectid); | |
31840ae1 ZY |
587 | btrfs_set_ref_num_refs(leaf, ref, 1); |
588 | } else if (ret == -EEXIST) { | |
589 | u64 existing_owner; | |
590 | BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID); | |
591 | leaf = path->nodes[0]; | |
592 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
593 | struct btrfs_extent_ref); | |
594 | if (btrfs_ref_root(leaf, ref) != ref_root || | |
595 | btrfs_ref_generation(leaf, ref) != ref_generation) { | |
596 | ret = -EIO; | |
597 | WARN_ON(1); | |
7bb86316 | 598 | goto out; |
31840ae1 ZY |
599 | } |
600 | ||
601 | num_refs = btrfs_ref_num_refs(leaf, ref); | |
602 | BUG_ON(num_refs == 0); | |
603 | btrfs_set_ref_num_refs(leaf, ref, num_refs + 1); | |
604 | ||
605 | existing_owner = btrfs_ref_objectid(leaf, ref); | |
3bb1a1bc YZ |
606 | if (existing_owner != owner_objectid && |
607 | existing_owner != BTRFS_MULTIPLE_OBJECTIDS) { | |
31840ae1 ZY |
608 | btrfs_set_ref_objectid(leaf, ref, |
609 | BTRFS_MULTIPLE_OBJECTIDS); | |
31840ae1 ZY |
610 | } |
611 | ret = 0; | |
612 | } else { | |
7bb86316 | 613 | goto out; |
31840ae1 | 614 | } |
7bb86316 CM |
615 | btrfs_mark_buffer_dirty(path->nodes[0]); |
616 | out: | |
617 | btrfs_release_path(root, path); | |
618 | return ret; | |
74493f7a CM |
619 | } |
620 | ||
31840ae1 ZY |
621 | static int noinline remove_extent_backref(struct btrfs_trans_handle *trans, |
622 | struct btrfs_root *root, | |
623 | struct btrfs_path *path) | |
624 | { | |
625 | struct extent_buffer *leaf; | |
626 | struct btrfs_extent_ref *ref; | |
627 | u32 num_refs; | |
628 | int ret = 0; | |
629 | ||
630 | leaf = path->nodes[0]; | |
631 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); | |
632 | num_refs = btrfs_ref_num_refs(leaf, ref); | |
633 | BUG_ON(num_refs == 0); | |
634 | num_refs -= 1; | |
635 | if (num_refs == 0) { | |
636 | ret = btrfs_del_item(trans, root, path); | |
637 | } else { | |
638 | btrfs_set_ref_num_refs(leaf, ref, num_refs); | |
639 | btrfs_mark_buffer_dirty(leaf); | |
640 | } | |
641 | btrfs_release_path(root, path); | |
642 | return ret; | |
643 | } | |
644 | ||
645 | static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, | |
646 | struct btrfs_root *root, u64 bytenr, | |
647 | u64 orig_parent, u64 parent, | |
648 | u64 orig_root, u64 ref_root, | |
649 | u64 orig_generation, u64 ref_generation, | |
3bb1a1bc | 650 | u64 owner_objectid) |
31840ae1 ZY |
651 | { |
652 | int ret; | |
653 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
654 | struct btrfs_path *path; | |
655 | ||
656 | if (root == root->fs_info->extent_root) { | |
657 | struct pending_extent_op *extent_op; | |
658 | u64 num_bytes; | |
659 | ||
660 | BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL); | |
661 | num_bytes = btrfs_level_size(root, (int)owner_objectid); | |
25179201 | 662 | mutex_lock(&root->fs_info->extent_ins_mutex); |
31840ae1 | 663 | if (test_range_bit(&root->fs_info->extent_ins, bytenr, |
25179201 | 664 | bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) { |
31840ae1 ZY |
665 | u64 priv; |
666 | ret = get_state_private(&root->fs_info->extent_ins, | |
667 | bytenr, &priv); | |
668 | BUG_ON(ret); | |
669 | extent_op = (struct pending_extent_op *) | |
670 | (unsigned long)priv; | |
671 | BUG_ON(extent_op->parent != orig_parent); | |
672 | BUG_ON(extent_op->generation != orig_generation); | |
25179201 | 673 | |
31840ae1 ZY |
674 | extent_op->parent = parent; |
675 | extent_op->generation = ref_generation; | |
676 | } else { | |
677 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
678 | BUG_ON(!extent_op); | |
679 | ||
680 | extent_op->type = PENDING_BACKREF_UPDATE; | |
681 | extent_op->bytenr = bytenr; | |
682 | extent_op->num_bytes = num_bytes; | |
683 | extent_op->parent = parent; | |
684 | extent_op->orig_parent = orig_parent; | |
685 | extent_op->generation = ref_generation; | |
686 | extent_op->orig_generation = orig_generation; | |
687 | extent_op->level = (int)owner_objectid; | |
688 | ||
689 | set_extent_bits(&root->fs_info->extent_ins, | |
690 | bytenr, bytenr + num_bytes - 1, | |
25179201 | 691 | EXTENT_WRITEBACK, GFP_NOFS); |
31840ae1 ZY |
692 | set_state_private(&root->fs_info->extent_ins, |
693 | bytenr, (unsigned long)extent_op); | |
694 | } | |
25179201 | 695 | mutex_unlock(&root->fs_info->extent_ins_mutex); |
31840ae1 ZY |
696 | return 0; |
697 | } | |
698 | ||
699 | path = btrfs_alloc_path(); | |
700 | if (!path) | |
701 | return -ENOMEM; | |
702 | ret = lookup_extent_backref(trans, extent_root, path, | |
703 | bytenr, orig_parent, orig_root, | |
3bb1a1bc | 704 | orig_generation, owner_objectid, 1); |
31840ae1 ZY |
705 | if (ret) |
706 | goto out; | |
707 | ret = remove_extent_backref(trans, extent_root, path); | |
708 | if (ret) | |
709 | goto out; | |
710 | ret = insert_extent_backref(trans, extent_root, path, bytenr, | |
711 | parent, ref_root, ref_generation, | |
3bb1a1bc | 712 | owner_objectid); |
31840ae1 | 713 | BUG_ON(ret); |
87ef2bb4 CM |
714 | finish_current_insert(trans, extent_root, 0); |
715 | del_pending_extents(trans, extent_root, 0); | |
31840ae1 ZY |
716 | out: |
717 | btrfs_free_path(path); | |
718 | return ret; | |
719 | } | |
720 | ||
721 | int btrfs_update_extent_ref(struct btrfs_trans_handle *trans, | |
722 | struct btrfs_root *root, u64 bytenr, | |
723 | u64 orig_parent, u64 parent, | |
724 | u64 ref_root, u64 ref_generation, | |
3bb1a1bc | 725 | u64 owner_objectid) |
31840ae1 ZY |
726 | { |
727 | int ret; | |
728 | if (ref_root == BTRFS_TREE_LOG_OBJECTID && | |
729 | owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
730 | return 0; | |
31840ae1 ZY |
731 | ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent, |
732 | parent, ref_root, ref_root, | |
733 | ref_generation, ref_generation, | |
3bb1a1bc | 734 | owner_objectid); |
31840ae1 ZY |
735 | return ret; |
736 | } | |
737 | ||
925baedd | 738 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
739 | struct btrfs_root *root, u64 bytenr, |
740 | u64 orig_parent, u64 parent, | |
741 | u64 orig_root, u64 ref_root, | |
742 | u64 orig_generation, u64 ref_generation, | |
3bb1a1bc | 743 | u64 owner_objectid) |
02217ed2 | 744 | { |
5caf2a00 | 745 | struct btrfs_path *path; |
02217ed2 | 746 | int ret; |
e2fa7227 | 747 | struct btrfs_key key; |
5f39d397 | 748 | struct extent_buffer *l; |
234b63a0 | 749 | struct btrfs_extent_item *item; |
cf27e1ee | 750 | u32 refs; |
037e6390 | 751 | |
5caf2a00 | 752 | path = btrfs_alloc_path(); |
54aa1f4d CM |
753 | if (!path) |
754 | return -ENOMEM; | |
26b8003f | 755 | |
3c12ac72 | 756 | path->reada = 1; |
db94535d | 757 | key.objectid = bytenr; |
31840ae1 ZY |
758 | key.type = BTRFS_EXTENT_ITEM_KEY; |
759 | key.offset = (u64)-1; | |
760 | ||
5caf2a00 | 761 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 762 | 0, 1); |
54aa1f4d CM |
763 | if (ret < 0) |
764 | return ret; | |
31840ae1 ZY |
765 | BUG_ON(ret == 0 || path->slots[0] == 0); |
766 | ||
767 | path->slots[0]--; | |
5f39d397 | 768 | l = path->nodes[0]; |
31840ae1 ZY |
769 | |
770 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
771ed689 CM |
771 | if (key.objectid != bytenr) { |
772 | btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]); | |
773 | printk("wanted %Lu found %Lu\n", bytenr, key.objectid); | |
774 | BUG(); | |
775 | } | |
31840ae1 ZY |
776 | BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY); |
777 | ||
5caf2a00 | 778 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 CM |
779 | refs = btrfs_extent_refs(l, item); |
780 | btrfs_set_extent_refs(l, item, refs + 1); | |
5caf2a00 | 781 | btrfs_mark_buffer_dirty(path->nodes[0]); |
a28ec197 | 782 | |
5caf2a00 | 783 | btrfs_release_path(root->fs_info->extent_root, path); |
7bb86316 | 784 | |
3c12ac72 | 785 | path->reada = 1; |
31840ae1 ZY |
786 | ret = insert_extent_backref(trans, root->fs_info->extent_root, |
787 | path, bytenr, parent, | |
788 | ref_root, ref_generation, | |
3bb1a1bc | 789 | owner_objectid); |
7bb86316 | 790 | BUG_ON(ret); |
87ef2bb4 CM |
791 | finish_current_insert(trans, root->fs_info->extent_root, 0); |
792 | del_pending_extents(trans, root->fs_info->extent_root, 0); | |
74493f7a CM |
793 | |
794 | btrfs_free_path(path); | |
02217ed2 CM |
795 | return 0; |
796 | } | |
797 | ||
925baedd | 798 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
799 | struct btrfs_root *root, |
800 | u64 bytenr, u64 num_bytes, u64 parent, | |
801 | u64 ref_root, u64 ref_generation, | |
3bb1a1bc | 802 | u64 owner_objectid) |
925baedd CM |
803 | { |
804 | int ret; | |
31840ae1 ZY |
805 | if (ref_root == BTRFS_TREE_LOG_OBJECTID && |
806 | owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
807 | return 0; | |
31840ae1 ZY |
808 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent, |
809 | 0, ref_root, 0, ref_generation, | |
3bb1a1bc | 810 | owner_objectid); |
925baedd CM |
811 | return ret; |
812 | } | |
813 | ||
e9d0b13b CM |
814 | int btrfs_extent_post_op(struct btrfs_trans_handle *trans, |
815 | struct btrfs_root *root) | |
816 | { | |
87ef2bb4 CM |
817 | finish_current_insert(trans, root->fs_info->extent_root, 1); |
818 | del_pending_extents(trans, root->fs_info->extent_root, 1); | |
e9d0b13b CM |
819 | return 0; |
820 | } | |
821 | ||
31840ae1 ZY |
822 | int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans, |
823 | struct btrfs_root *root, u64 bytenr, | |
824 | u64 num_bytes, u32 *refs) | |
a28ec197 | 825 | { |
5caf2a00 | 826 | struct btrfs_path *path; |
a28ec197 | 827 | int ret; |
e2fa7227 | 828 | struct btrfs_key key; |
5f39d397 | 829 | struct extent_buffer *l; |
234b63a0 | 830 | struct btrfs_extent_item *item; |
5caf2a00 | 831 | |
db94535d | 832 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 833 | path = btrfs_alloc_path(); |
3c12ac72 | 834 | path->reada = 1; |
db94535d CM |
835 | key.objectid = bytenr; |
836 | key.offset = num_bytes; | |
62e2749e | 837 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
5caf2a00 | 838 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 839 | 0, 0); |
54aa1f4d CM |
840 | if (ret < 0) |
841 | goto out; | |
5f39d397 CM |
842 | if (ret != 0) { |
843 | btrfs_print_leaf(root, path->nodes[0]); | |
db94535d | 844 | printk("failed to find block number %Lu\n", bytenr); |
a28ec197 | 845 | BUG(); |
5f39d397 CM |
846 | } |
847 | l = path->nodes[0]; | |
5caf2a00 | 848 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 | 849 | *refs = btrfs_extent_refs(l, item); |
54aa1f4d | 850 | out: |
5caf2a00 | 851 | btrfs_free_path(path); |
a28ec197 CM |
852 | return 0; |
853 | } | |
854 | ||
80ff3856 YZ |
855 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, |
856 | struct btrfs_root *root, u64 bytenr) | |
be20aa9d CM |
857 | { |
858 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
859 | struct btrfs_path *path; | |
f321e491 YZ |
860 | struct extent_buffer *leaf; |
861 | struct btrfs_extent_ref *ref_item; | |
862 | struct btrfs_key key; | |
863 | struct btrfs_key found_key; | |
80ff3856 YZ |
864 | u64 ref_root; |
865 | u64 last_snapshot; | |
be20aa9d CM |
866 | u32 nritems; |
867 | int ret; | |
925baedd | 868 | |
be20aa9d | 869 | key.objectid = bytenr; |
31840ae1 | 870 | key.offset = (u64)-1; |
f321e491 | 871 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 872 | |
f321e491 | 873 | path = btrfs_alloc_path(); |
be20aa9d CM |
874 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
875 | if (ret < 0) | |
876 | goto out; | |
877 | BUG_ON(ret == 0); | |
80ff3856 YZ |
878 | |
879 | ret = -ENOENT; | |
880 | if (path->slots[0] == 0) | |
31840ae1 | 881 | goto out; |
be20aa9d | 882 | |
31840ae1 | 883 | path->slots[0]--; |
f321e491 YZ |
884 | leaf = path->nodes[0]; |
885 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
be20aa9d CM |
886 | |
887 | if (found_key.objectid != bytenr || | |
80ff3856 | 888 | found_key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 889 | goto out; |
f321e491 | 890 | |
80ff3856 | 891 | last_snapshot = btrfs_root_last_snapshot(&root->root_item); |
be20aa9d | 892 | while (1) { |
f321e491 YZ |
893 | leaf = path->nodes[0]; |
894 | nritems = btrfs_header_nritems(leaf); | |
be20aa9d CM |
895 | if (path->slots[0] >= nritems) { |
896 | ret = btrfs_next_leaf(extent_root, path); | |
f321e491 YZ |
897 | if (ret < 0) |
898 | goto out; | |
be20aa9d CM |
899 | if (ret == 0) |
900 | continue; | |
901 | break; | |
902 | } | |
f321e491 | 903 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
be20aa9d CM |
904 | if (found_key.objectid != bytenr) |
905 | break; | |
bd09835d | 906 | |
be20aa9d CM |
907 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
908 | path->slots[0]++; | |
909 | continue; | |
910 | } | |
911 | ||
f321e491 | 912 | ref_item = btrfs_item_ptr(leaf, path->slots[0], |
be20aa9d | 913 | struct btrfs_extent_ref); |
80ff3856 YZ |
914 | ref_root = btrfs_ref_root(leaf, ref_item); |
915 | if (ref_root != root->root_key.objectid && | |
916 | ref_root != BTRFS_TREE_LOG_OBJECTID) { | |
917 | ret = 1; | |
f321e491 | 918 | goto out; |
80ff3856 YZ |
919 | } |
920 | if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) { | |
f321e491 YZ |
921 | ret = 1; |
922 | goto out; | |
923 | } | |
80ff3856 YZ |
924 | |
925 | path->slots[0]++; | |
f321e491 YZ |
926 | } |
927 | ret = 0; | |
be20aa9d | 928 | out: |
80ff3856 | 929 | btrfs_free_path(path); |
f321e491 | 930 | return ret; |
be20aa9d | 931 | } |
c5739bba | 932 | |
31840ae1 ZY |
933 | int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
934 | struct extent_buffer *buf, u32 nr_extents) | |
02217ed2 | 935 | { |
5f39d397 | 936 | struct btrfs_key key; |
6407bf6d | 937 | struct btrfs_file_extent_item *fi; |
e4657689 ZY |
938 | u64 root_gen; |
939 | u32 nritems; | |
02217ed2 | 940 | int i; |
db94535d | 941 | int level; |
31840ae1 | 942 | int ret = 0; |
e4657689 | 943 | int shared = 0; |
a28ec197 | 944 | |
3768f368 | 945 | if (!root->ref_cows) |
a28ec197 | 946 | return 0; |
5f39d397 | 947 | |
e4657689 ZY |
948 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
949 | shared = 0; | |
950 | root_gen = root->root_key.offset; | |
951 | } else { | |
952 | shared = 1; | |
953 | root_gen = trans->transid - 1; | |
954 | } | |
955 | ||
db94535d | 956 | level = btrfs_header_level(buf); |
5f39d397 | 957 | nritems = btrfs_header_nritems(buf); |
4a096752 | 958 | |
31840ae1 | 959 | if (level == 0) { |
31153d81 YZ |
960 | struct btrfs_leaf_ref *ref; |
961 | struct btrfs_extent_info *info; | |
962 | ||
31840ae1 | 963 | ref = btrfs_alloc_leaf_ref(root, nr_extents); |
31153d81 | 964 | if (!ref) { |
31840ae1 | 965 | ret = -ENOMEM; |
31153d81 YZ |
966 | goto out; |
967 | } | |
968 | ||
e4657689 | 969 | ref->root_gen = root_gen; |
31153d81 YZ |
970 | ref->bytenr = buf->start; |
971 | ref->owner = btrfs_header_owner(buf); | |
972 | ref->generation = btrfs_header_generation(buf); | |
31840ae1 | 973 | ref->nritems = nr_extents; |
31153d81 | 974 | info = ref->extents; |
bcc63abb | 975 | |
31840ae1 | 976 | for (i = 0; nr_extents > 0 && i < nritems; i++) { |
31153d81 YZ |
977 | u64 disk_bytenr; |
978 | btrfs_item_key_to_cpu(buf, &key, i); | |
979 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
980 | continue; | |
981 | fi = btrfs_item_ptr(buf, i, | |
982 | struct btrfs_file_extent_item); | |
983 | if (btrfs_file_extent_type(buf, fi) == | |
984 | BTRFS_FILE_EXTENT_INLINE) | |
985 | continue; | |
986 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
987 | if (disk_bytenr == 0) | |
988 | continue; | |
989 | ||
990 | info->bytenr = disk_bytenr; | |
991 | info->num_bytes = | |
992 | btrfs_file_extent_disk_num_bytes(buf, fi); | |
993 | info->objectid = key.objectid; | |
994 | info->offset = key.offset; | |
995 | info++; | |
996 | } | |
997 | ||
e4657689 | 998 | ret = btrfs_add_leaf_ref(root, ref, shared); |
5b84e8d6 YZ |
999 | if (ret == -EEXIST && shared) { |
1000 | struct btrfs_leaf_ref *old; | |
1001 | old = btrfs_lookup_leaf_ref(root, ref->bytenr); | |
1002 | BUG_ON(!old); | |
1003 | btrfs_remove_leaf_ref(root, old); | |
1004 | btrfs_free_leaf_ref(root, old); | |
1005 | ret = btrfs_add_leaf_ref(root, ref, shared); | |
1006 | } | |
31153d81 | 1007 | WARN_ON(ret); |
bcc63abb | 1008 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
1009 | } |
1010 | out: | |
31840ae1 ZY |
1011 | return ret; |
1012 | } | |
1013 | ||
1014 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
1015 | struct extent_buffer *orig_buf, struct extent_buffer *buf, | |
1016 | u32 *nr_extents) | |
1017 | { | |
1018 | u64 bytenr; | |
1019 | u64 ref_root; | |
1020 | u64 orig_root; | |
1021 | u64 ref_generation; | |
1022 | u64 orig_generation; | |
1023 | u32 nritems; | |
1024 | u32 nr_file_extents = 0; | |
1025 | struct btrfs_key key; | |
1026 | struct btrfs_file_extent_item *fi; | |
1027 | int i; | |
1028 | int level; | |
1029 | int ret = 0; | |
1030 | int faili = 0; | |
1031 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, | |
3bb1a1bc | 1032 | u64, u64, u64, u64, u64, u64, u64, u64); |
31840ae1 ZY |
1033 | |
1034 | ref_root = btrfs_header_owner(buf); | |
1035 | ref_generation = btrfs_header_generation(buf); | |
1036 | orig_root = btrfs_header_owner(orig_buf); | |
1037 | orig_generation = btrfs_header_generation(orig_buf); | |
1038 | ||
1039 | nritems = btrfs_header_nritems(buf); | |
1040 | level = btrfs_header_level(buf); | |
1041 | ||
1042 | if (root->ref_cows) { | |
1043 | process_func = __btrfs_inc_extent_ref; | |
1044 | } else { | |
1045 | if (level == 0 && | |
1046 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
1047 | goto out; | |
1048 | if (level != 0 && | |
1049 | root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) | |
1050 | goto out; | |
1051 | process_func = __btrfs_update_extent_ref; | |
1052 | } | |
1053 | ||
1054 | for (i = 0; i < nritems; i++) { | |
1055 | cond_resched(); | |
db94535d | 1056 | if (level == 0) { |
5f39d397 CM |
1057 | btrfs_item_key_to_cpu(buf, &key, i); |
1058 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
54aa1f4d | 1059 | continue; |
5f39d397 | 1060 | fi = btrfs_item_ptr(buf, i, |
54aa1f4d | 1061 | struct btrfs_file_extent_item); |
5f39d397 | 1062 | if (btrfs_file_extent_type(buf, fi) == |
54aa1f4d CM |
1063 | BTRFS_FILE_EXTENT_INLINE) |
1064 | continue; | |
31840ae1 ZY |
1065 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
1066 | if (bytenr == 0) | |
54aa1f4d | 1067 | continue; |
31840ae1 ZY |
1068 | |
1069 | nr_file_extents++; | |
1070 | ||
31840ae1 ZY |
1071 | ret = process_func(trans, root, bytenr, |
1072 | orig_buf->start, buf->start, | |
1073 | orig_root, ref_root, | |
1074 | orig_generation, ref_generation, | |
3bb1a1bc | 1075 | key.objectid); |
31840ae1 ZY |
1076 | |
1077 | if (ret) { | |
1078 | faili = i; | |
1079 | WARN_ON(1); | |
1080 | goto fail; | |
1081 | } | |
54aa1f4d | 1082 | } else { |
db94535d | 1083 | bytenr = btrfs_node_blockptr(buf, i); |
31840ae1 ZY |
1084 | ret = process_func(trans, root, bytenr, |
1085 | orig_buf->start, buf->start, | |
1086 | orig_root, ref_root, | |
1087 | orig_generation, ref_generation, | |
3bb1a1bc | 1088 | level - 1); |
31840ae1 ZY |
1089 | if (ret) { |
1090 | faili = i; | |
1091 | WARN_ON(1); | |
1092 | goto fail; | |
1093 | } | |
54aa1f4d CM |
1094 | } |
1095 | } | |
31840ae1 ZY |
1096 | out: |
1097 | if (nr_extents) { | |
1098 | if (level == 0) | |
1099 | *nr_extents = nr_file_extents; | |
1100 | else | |
1101 | *nr_extents = nritems; | |
1102 | } | |
1103 | return 0; | |
1104 | fail: | |
1105 | WARN_ON(1); | |
54aa1f4d | 1106 | return ret; |
02217ed2 CM |
1107 | } |
1108 | ||
31840ae1 ZY |
1109 | int btrfs_update_ref(struct btrfs_trans_handle *trans, |
1110 | struct btrfs_root *root, struct extent_buffer *orig_buf, | |
1111 | struct extent_buffer *buf, int start_slot, int nr) | |
1112 | ||
1113 | { | |
1114 | u64 bytenr; | |
1115 | u64 ref_root; | |
1116 | u64 orig_root; | |
1117 | u64 ref_generation; | |
1118 | u64 orig_generation; | |
1119 | struct btrfs_key key; | |
1120 | struct btrfs_file_extent_item *fi; | |
1121 | int i; | |
1122 | int ret; | |
1123 | int slot; | |
1124 | int level; | |
1125 | ||
1126 | BUG_ON(start_slot < 0); | |
1127 | BUG_ON(start_slot + nr > btrfs_header_nritems(buf)); | |
1128 | ||
1129 | ref_root = btrfs_header_owner(buf); | |
1130 | ref_generation = btrfs_header_generation(buf); | |
1131 | orig_root = btrfs_header_owner(orig_buf); | |
1132 | orig_generation = btrfs_header_generation(orig_buf); | |
1133 | level = btrfs_header_level(buf); | |
1134 | ||
1135 | if (!root->ref_cows) { | |
1136 | if (level == 0 && | |
1137 | root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) | |
1138 | return 0; | |
1139 | if (level != 0 && | |
1140 | root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) | |
1141 | return 0; | |
1142 | } | |
1143 | ||
1144 | for (i = 0, slot = start_slot; i < nr; i++, slot++) { | |
1145 | cond_resched(); | |
1146 | if (level == 0) { | |
1147 | btrfs_item_key_to_cpu(buf, &key, slot); | |
1148 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
1149 | continue; | |
1150 | fi = btrfs_item_ptr(buf, slot, | |
1151 | struct btrfs_file_extent_item); | |
1152 | if (btrfs_file_extent_type(buf, fi) == | |
1153 | BTRFS_FILE_EXTENT_INLINE) | |
1154 | continue; | |
1155 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
1156 | if (bytenr == 0) | |
1157 | continue; | |
31840ae1 ZY |
1158 | ret = __btrfs_update_extent_ref(trans, root, bytenr, |
1159 | orig_buf->start, buf->start, | |
1160 | orig_root, ref_root, | |
1161 | orig_generation, ref_generation, | |
3bb1a1bc | 1162 | key.objectid); |
31840ae1 ZY |
1163 | if (ret) |
1164 | goto fail; | |
1165 | } else { | |
1166 | bytenr = btrfs_node_blockptr(buf, slot); | |
31840ae1 ZY |
1167 | ret = __btrfs_update_extent_ref(trans, root, bytenr, |
1168 | orig_buf->start, buf->start, | |
1169 | orig_root, ref_root, | |
1170 | orig_generation, ref_generation, | |
3bb1a1bc | 1171 | level - 1); |
31840ae1 ZY |
1172 | if (ret) |
1173 | goto fail; | |
1174 | } | |
1175 | } | |
1176 | return 0; | |
1177 | fail: | |
1178 | WARN_ON(1); | |
1179 | return -1; | |
1180 | } | |
1181 | ||
9078a3e1 CM |
1182 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
1183 | struct btrfs_root *root, | |
1184 | struct btrfs_path *path, | |
1185 | struct btrfs_block_group_cache *cache) | |
1186 | { | |
1187 | int ret; | |
1188 | int pending_ret; | |
1189 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
5f39d397 CM |
1190 | unsigned long bi; |
1191 | struct extent_buffer *leaf; | |
9078a3e1 | 1192 | |
9078a3e1 | 1193 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
1194 | if (ret < 0) |
1195 | goto fail; | |
9078a3e1 | 1196 | BUG_ON(ret); |
5f39d397 CM |
1197 | |
1198 | leaf = path->nodes[0]; | |
1199 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1200 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
1201 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 1202 | btrfs_release_path(extent_root, path); |
54aa1f4d | 1203 | fail: |
87ef2bb4 CM |
1204 | finish_current_insert(trans, extent_root, 0); |
1205 | pending_ret = del_pending_extents(trans, extent_root, 0); | |
9078a3e1 CM |
1206 | if (ret) |
1207 | return ret; | |
1208 | if (pending_ret) | |
1209 | return pending_ret; | |
1210 | return 0; | |
1211 | ||
1212 | } | |
1213 | ||
96b5179d CM |
1214 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
1215 | struct btrfs_root *root) | |
9078a3e1 | 1216 | { |
0f9dd46c JB |
1217 | struct btrfs_block_group_cache *cache, *entry; |
1218 | struct rb_node *n; | |
9078a3e1 CM |
1219 | int err = 0; |
1220 | int werr = 0; | |
9078a3e1 | 1221 | struct btrfs_path *path; |
96b5179d | 1222 | u64 last = 0; |
9078a3e1 CM |
1223 | |
1224 | path = btrfs_alloc_path(); | |
1225 | if (!path) | |
1226 | return -ENOMEM; | |
1227 | ||
1228 | while(1) { | |
0f9dd46c JB |
1229 | cache = NULL; |
1230 | spin_lock(&root->fs_info->block_group_cache_lock); | |
1231 | for (n = rb_first(&root->fs_info->block_group_cache_tree); | |
1232 | n; n = rb_next(n)) { | |
1233 | entry = rb_entry(n, struct btrfs_block_group_cache, | |
1234 | cache_node); | |
1235 | if (entry->dirty) { | |
1236 | cache = entry; | |
1237 | break; | |
1238 | } | |
1239 | } | |
1240 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
54aa1f4d | 1241 | |
0f9dd46c | 1242 | if (!cache) |
96b5179d | 1243 | break; |
0f9dd46c | 1244 | |
e8569813 | 1245 | cache->dirty = 0; |
0f9dd46c JB |
1246 | last += cache->key.offset; |
1247 | ||
96b5179d CM |
1248 | err = write_one_cache_group(trans, root, |
1249 | path, cache); | |
1250 | /* | |
1251 | * if we fail to write the cache group, we want | |
1252 | * to keep it marked dirty in hopes that a later | |
1253 | * write will work | |
1254 | */ | |
1255 | if (err) { | |
1256 | werr = err; | |
1257 | continue; | |
9078a3e1 CM |
1258 | } |
1259 | } | |
1260 | btrfs_free_path(path); | |
1261 | return werr; | |
1262 | } | |
1263 | ||
593060d7 CM |
1264 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
1265 | u64 total_bytes, u64 bytes_used, | |
1266 | struct btrfs_space_info **space_info) | |
1267 | { | |
1268 | struct btrfs_space_info *found; | |
1269 | ||
1270 | found = __find_space_info(info, flags); | |
1271 | if (found) { | |
25179201 | 1272 | spin_lock(&found->lock); |
593060d7 CM |
1273 | found->total_bytes += total_bytes; |
1274 | found->bytes_used += bytes_used; | |
8f18cf13 | 1275 | found->full = 0; |
25179201 | 1276 | spin_unlock(&found->lock); |
593060d7 CM |
1277 | *space_info = found; |
1278 | return 0; | |
1279 | } | |
1280 | found = kmalloc(sizeof(*found), GFP_NOFS); | |
1281 | if (!found) | |
1282 | return -ENOMEM; | |
1283 | ||
1284 | list_add(&found->list, &info->space_info); | |
0f9dd46c | 1285 | INIT_LIST_HEAD(&found->block_groups); |
80eb234a | 1286 | init_rwsem(&found->groups_sem); |
0f9dd46c | 1287 | spin_lock_init(&found->lock); |
593060d7 CM |
1288 | found->flags = flags; |
1289 | found->total_bytes = total_bytes; | |
1290 | found->bytes_used = bytes_used; | |
1291 | found->bytes_pinned = 0; | |
e8569813 | 1292 | found->bytes_reserved = 0; |
593060d7 | 1293 | found->full = 0; |
0ef3e66b | 1294 | found->force_alloc = 0; |
593060d7 CM |
1295 | *space_info = found; |
1296 | return 0; | |
1297 | } | |
1298 | ||
8790d502 CM |
1299 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
1300 | { | |
1301 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 1302 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 1303 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 1304 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
1305 | if (extra_flags) { |
1306 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1307 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1308 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1309 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1310 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1311 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1312 | } | |
1313 | } | |
593060d7 | 1314 | |
a061fc8d | 1315 | static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 1316 | { |
a061fc8d CM |
1317 | u64 num_devices = root->fs_info->fs_devices->num_devices; |
1318 | ||
1319 | if (num_devices == 1) | |
1320 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
1321 | if (num_devices < 4) | |
1322 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
1323 | ||
ec44a35c CM |
1324 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
1325 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 1326 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 1327 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 1328 | } |
ec44a35c CM |
1329 | |
1330 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 1331 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 1332 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 1333 | } |
ec44a35c CM |
1334 | |
1335 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
1336 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
1337 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
1338 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
1339 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
1340 | return flags; | |
1341 | } | |
1342 | ||
6324fbf3 CM |
1343 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
1344 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 1345 | u64 flags, int force) |
6324fbf3 CM |
1346 | { |
1347 | struct btrfs_space_info *space_info; | |
1348 | u64 thresh; | |
1349 | u64 start; | |
1350 | u64 num_bytes; | |
cf749823 | 1351 | int ret = 0, waited = 0; |
6324fbf3 | 1352 | |
a061fc8d | 1353 | flags = reduce_alloc_profile(extent_root, flags); |
ec44a35c | 1354 | |
6324fbf3 | 1355 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
1356 | if (!space_info) { |
1357 | ret = update_space_info(extent_root->fs_info, flags, | |
1358 | 0, 0, &space_info); | |
1359 | BUG_ON(ret); | |
1360 | } | |
6324fbf3 CM |
1361 | BUG_ON(!space_info); |
1362 | ||
25179201 | 1363 | spin_lock(&space_info->lock); |
0ef3e66b CM |
1364 | if (space_info->force_alloc) { |
1365 | force = 1; | |
1366 | space_info->force_alloc = 0; | |
1367 | } | |
25179201 JB |
1368 | if (space_info->full) { |
1369 | spin_unlock(&space_info->lock); | |
925baedd | 1370 | goto out; |
25179201 | 1371 | } |
6324fbf3 | 1372 | |
8790d502 | 1373 | thresh = div_factor(space_info->total_bytes, 6); |
0ef3e66b | 1374 | if (!force && |
e8569813 | 1375 | (space_info->bytes_used + space_info->bytes_pinned + |
25179201 JB |
1376 | space_info->bytes_reserved + alloc_bytes) < thresh) { |
1377 | spin_unlock(&space_info->lock); | |
925baedd | 1378 | goto out; |
25179201 | 1379 | } |
6324fbf3 | 1380 | |
25179201 JB |
1381 | spin_unlock(&space_info->lock); |
1382 | ||
1383 | ret = mutex_trylock(&extent_root->fs_info->chunk_mutex); | |
1384 | if (!ret && !force) { | |
1385 | goto out; | |
1386 | } else if (!ret) { | |
1387 | mutex_lock(&extent_root->fs_info->chunk_mutex); | |
cf749823 JB |
1388 | waited = 1; |
1389 | } | |
1390 | ||
25179201 JB |
1391 | if (waited) { |
1392 | spin_lock(&space_info->lock); | |
1393 | if (space_info->full) { | |
1394 | spin_unlock(&space_info->lock); | |
1395 | goto out_unlock; | |
1396 | } | |
1397 | spin_unlock(&space_info->lock); | |
1398 | } | |
cf749823 | 1399 | |
6324fbf3 | 1400 | ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags); |
25179201 | 1401 | if (ret) { |
6324fbf3 CM |
1402 | printk("space info full %Lu\n", flags); |
1403 | space_info->full = 1; | |
a74a4b97 | 1404 | goto out_unlock; |
6324fbf3 | 1405 | } |
6324fbf3 CM |
1406 | |
1407 | ret = btrfs_make_block_group(trans, extent_root, 0, flags, | |
e17cade2 | 1408 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes); |
6324fbf3 | 1409 | BUG_ON(ret); |
a74a4b97 | 1410 | out_unlock: |
333db94c | 1411 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
a74a4b97 | 1412 | out: |
0f9dd46c | 1413 | return ret; |
6324fbf3 CM |
1414 | } |
1415 | ||
9078a3e1 CM |
1416 | static int update_block_group(struct btrfs_trans_handle *trans, |
1417 | struct btrfs_root *root, | |
db94535d | 1418 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 1419 | int mark_free) |
9078a3e1 CM |
1420 | { |
1421 | struct btrfs_block_group_cache *cache; | |
1422 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 1423 | u64 total = num_bytes; |
9078a3e1 | 1424 | u64 old_val; |
db94535d | 1425 | u64 byte_in_group; |
3e1ad54f | 1426 | |
9078a3e1 | 1427 | while(total) { |
db94535d | 1428 | cache = btrfs_lookup_block_group(info, bytenr); |
3e1ad54f | 1429 | if (!cache) { |
9078a3e1 | 1430 | return -1; |
cd1bc465 | 1431 | } |
db94535d CM |
1432 | byte_in_group = bytenr - cache->key.objectid; |
1433 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 1434 | |
25179201 | 1435 | spin_lock(&cache->space_info->lock); |
c286ac48 | 1436 | spin_lock(&cache->lock); |
0f9dd46c | 1437 | cache->dirty = 1; |
9078a3e1 | 1438 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 1439 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 1440 | if (alloc) { |
db94535d | 1441 | old_val += num_bytes; |
6324fbf3 | 1442 | cache->space_info->bytes_used += num_bytes; |
c286ac48 CM |
1443 | btrfs_set_block_group_used(&cache->item, old_val); |
1444 | spin_unlock(&cache->lock); | |
25179201 | 1445 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 1446 | } else { |
db94535d | 1447 | old_val -= num_bytes; |
6324fbf3 | 1448 | cache->space_info->bytes_used -= num_bytes; |
c286ac48 CM |
1449 | btrfs_set_block_group_used(&cache->item, old_val); |
1450 | spin_unlock(&cache->lock); | |
25179201 | 1451 | spin_unlock(&cache->space_info->lock); |
f510cfec | 1452 | if (mark_free) { |
0f9dd46c JB |
1453 | int ret; |
1454 | ret = btrfs_add_free_space(cache, bytenr, | |
1455 | num_bytes); | |
1456 | if (ret) | |
1457 | return -1; | |
e37c9e69 | 1458 | } |
cd1bc465 | 1459 | } |
db94535d CM |
1460 | total -= num_bytes; |
1461 | bytenr += num_bytes; | |
9078a3e1 CM |
1462 | } |
1463 | return 0; | |
1464 | } | |
6324fbf3 | 1465 | |
a061fc8d CM |
1466 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
1467 | { | |
0f9dd46c JB |
1468 | struct btrfs_block_group_cache *cache; |
1469 | ||
1470 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
1471 | if (!cache) | |
a061fc8d | 1472 | return 0; |
0f9dd46c JB |
1473 | |
1474 | return cache->key.objectid; | |
a061fc8d CM |
1475 | } |
1476 | ||
e02119d5 | 1477 | int btrfs_update_pinned_extents(struct btrfs_root *root, |
324ae4df Y |
1478 | u64 bytenr, u64 num, int pin) |
1479 | { | |
1480 | u64 len; | |
1481 | struct btrfs_block_group_cache *cache; | |
1482 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1483 | ||
25179201 | 1484 | WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex)); |
324ae4df Y |
1485 | if (pin) { |
1486 | set_extent_dirty(&fs_info->pinned_extents, | |
1487 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1488 | } else { | |
1489 | clear_extent_dirty(&fs_info->pinned_extents, | |
1490 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1491 | } | |
1492 | while (num > 0) { | |
1493 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
e8569813 ZY |
1494 | BUG_ON(!cache); |
1495 | len = min(num, cache->key.offset - | |
1496 | (bytenr - cache->key.objectid)); | |
324ae4df | 1497 | if (pin) { |
25179201 | 1498 | spin_lock(&cache->space_info->lock); |
e8569813 ZY |
1499 | spin_lock(&cache->lock); |
1500 | cache->pinned += len; | |
1501 | cache->space_info->bytes_pinned += len; | |
1502 | spin_unlock(&cache->lock); | |
25179201 | 1503 | spin_unlock(&cache->space_info->lock); |
324ae4df Y |
1504 | fs_info->total_pinned += len; |
1505 | } else { | |
25179201 | 1506 | spin_lock(&cache->space_info->lock); |
e8569813 ZY |
1507 | spin_lock(&cache->lock); |
1508 | cache->pinned -= len; | |
1509 | cache->space_info->bytes_pinned -= len; | |
1510 | spin_unlock(&cache->lock); | |
25179201 | 1511 | spin_unlock(&cache->space_info->lock); |
324ae4df Y |
1512 | fs_info->total_pinned -= len; |
1513 | } | |
1514 | bytenr += len; | |
1515 | num -= len; | |
1516 | } | |
1517 | return 0; | |
1518 | } | |
9078a3e1 | 1519 | |
e8569813 ZY |
1520 | static int update_reserved_extents(struct btrfs_root *root, |
1521 | u64 bytenr, u64 num, int reserve) | |
1522 | { | |
1523 | u64 len; | |
1524 | struct btrfs_block_group_cache *cache; | |
1525 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1526 | ||
e8569813 ZY |
1527 | while (num > 0) { |
1528 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
1529 | BUG_ON(!cache); | |
1530 | len = min(num, cache->key.offset - | |
1531 | (bytenr - cache->key.objectid)); | |
25179201 JB |
1532 | |
1533 | spin_lock(&cache->space_info->lock); | |
1534 | spin_lock(&cache->lock); | |
e8569813 | 1535 | if (reserve) { |
e8569813 ZY |
1536 | cache->reserved += len; |
1537 | cache->space_info->bytes_reserved += len; | |
e8569813 | 1538 | } else { |
e8569813 ZY |
1539 | cache->reserved -= len; |
1540 | cache->space_info->bytes_reserved -= len; | |
e8569813 | 1541 | } |
25179201 JB |
1542 | spin_unlock(&cache->lock); |
1543 | spin_unlock(&cache->space_info->lock); | |
e8569813 ZY |
1544 | bytenr += len; |
1545 | num -= len; | |
1546 | } | |
1547 | return 0; | |
1548 | } | |
1549 | ||
d1310b2e | 1550 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) |
ccd467d6 | 1551 | { |
ccd467d6 | 1552 | u64 last = 0; |
1a5bc167 CM |
1553 | u64 start; |
1554 | u64 end; | |
d1310b2e | 1555 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; |
ccd467d6 | 1556 | int ret; |
ccd467d6 | 1557 | |
25179201 | 1558 | mutex_lock(&root->fs_info->pinned_mutex); |
ccd467d6 | 1559 | while(1) { |
1a5bc167 CM |
1560 | ret = find_first_extent_bit(pinned_extents, last, |
1561 | &start, &end, EXTENT_DIRTY); | |
1562 | if (ret) | |
ccd467d6 | 1563 | break; |
1a5bc167 CM |
1564 | set_extent_dirty(copy, start, end, GFP_NOFS); |
1565 | last = end + 1; | |
ccd467d6 | 1566 | } |
25179201 | 1567 | mutex_unlock(&root->fs_info->pinned_mutex); |
ccd467d6 CM |
1568 | return 0; |
1569 | } | |
1570 | ||
1571 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
1572 | struct btrfs_root *root, | |
d1310b2e | 1573 | struct extent_io_tree *unpin) |
a28ec197 | 1574 | { |
1a5bc167 CM |
1575 | u64 start; |
1576 | u64 end; | |
a28ec197 | 1577 | int ret; |
0f9dd46c | 1578 | struct btrfs_block_group_cache *cache; |
a28ec197 | 1579 | |
25179201 | 1580 | mutex_lock(&root->fs_info->pinned_mutex); |
a28ec197 | 1581 | while(1) { |
1a5bc167 CM |
1582 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
1583 | EXTENT_DIRTY); | |
1584 | if (ret) | |
a28ec197 | 1585 | break; |
e02119d5 | 1586 | btrfs_update_pinned_extents(root, start, end + 1 - start, 0); |
1a5bc167 | 1587 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
0f9dd46c JB |
1588 | cache = btrfs_lookup_block_group(root->fs_info, start); |
1589 | if (cache->cached) | |
1590 | btrfs_add_free_space(cache, start, end - start + 1); | |
c286ac48 | 1591 | if (need_resched()) { |
25179201 | 1592 | mutex_unlock(&root->fs_info->pinned_mutex); |
c286ac48 | 1593 | cond_resched(); |
25179201 | 1594 | mutex_lock(&root->fs_info->pinned_mutex); |
c286ac48 | 1595 | } |
a28ec197 | 1596 | } |
25179201 | 1597 | mutex_unlock(&root->fs_info->pinned_mutex); |
a28ec197 CM |
1598 | return 0; |
1599 | } | |
1600 | ||
98ed5174 | 1601 | static int finish_current_insert(struct btrfs_trans_handle *trans, |
87ef2bb4 | 1602 | struct btrfs_root *extent_root, int all) |
037e6390 | 1603 | { |
7bb86316 CM |
1604 | u64 start; |
1605 | u64 end; | |
31840ae1 | 1606 | u64 priv; |
25179201 | 1607 | u64 search = 0; |
7bb86316 CM |
1608 | struct btrfs_fs_info *info = extent_root->fs_info; |
1609 | struct btrfs_path *path; | |
31840ae1 ZY |
1610 | struct btrfs_extent_ref *ref; |
1611 | struct pending_extent_op *extent_op; | |
1612 | struct btrfs_key key; | |
234b63a0 | 1613 | struct btrfs_extent_item extent_item; |
037e6390 | 1614 | int ret; |
1a5bc167 | 1615 | int err = 0; |
037e6390 | 1616 | |
5f39d397 | 1617 | btrfs_set_stack_extent_refs(&extent_item, 1); |
7bb86316 | 1618 | path = btrfs_alloc_path(); |
037e6390 | 1619 | |
26b8003f | 1620 | while(1) { |
25179201 JB |
1621 | mutex_lock(&info->extent_ins_mutex); |
1622 | ret = find_first_extent_bit(&info->extent_ins, search, &start, | |
1623 | &end, EXTENT_WRITEBACK); | |
1624 | if (ret) { | |
1625 | mutex_unlock(&info->extent_ins_mutex); | |
87ef2bb4 | 1626 | if (search && all) { |
25179201 JB |
1627 | search = 0; |
1628 | continue; | |
1629 | } | |
26b8003f | 1630 | break; |
25179201 JB |
1631 | } |
1632 | ||
1633 | ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS); | |
1634 | if (!ret) { | |
87ef2bb4 | 1635 | search = end + 1; |
25179201 JB |
1636 | mutex_unlock(&info->extent_ins_mutex); |
1637 | cond_resched(); | |
1638 | continue; | |
1639 | } | |
1640 | BUG_ON(ret < 0); | |
26b8003f | 1641 | |
31840ae1 ZY |
1642 | ret = get_state_private(&info->extent_ins, start, &priv); |
1643 | BUG_ON(ret); | |
1644 | extent_op = (struct pending_extent_op *)(unsigned long)priv; | |
1645 | ||
25179201 JB |
1646 | mutex_unlock(&info->extent_ins_mutex); |
1647 | ||
31840ae1 ZY |
1648 | if (extent_op->type == PENDING_EXTENT_INSERT) { |
1649 | key.objectid = start; | |
1650 | key.offset = end + 1 - start; | |
1651 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1652 | err = btrfs_insert_item(trans, extent_root, &key, | |
1a5bc167 | 1653 | &extent_item, sizeof(extent_item)); |
31840ae1 | 1654 | BUG_ON(err); |
c286ac48 | 1655 | |
25179201 | 1656 | mutex_lock(&info->extent_ins_mutex); |
31840ae1 | 1657 | clear_extent_bits(&info->extent_ins, start, end, |
25179201 JB |
1658 | EXTENT_WRITEBACK, GFP_NOFS); |
1659 | mutex_unlock(&info->extent_ins_mutex); | |
c286ac48 | 1660 | |
31840ae1 ZY |
1661 | err = insert_extent_backref(trans, extent_root, path, |
1662 | start, extent_op->parent, | |
1663 | extent_root->root_key.objectid, | |
1664 | extent_op->generation, | |
3bb1a1bc | 1665 | extent_op->level); |
31840ae1 ZY |
1666 | BUG_ON(err); |
1667 | } else if (extent_op->type == PENDING_BACKREF_UPDATE) { | |
1668 | err = lookup_extent_backref(trans, extent_root, path, | |
1669 | start, extent_op->orig_parent, | |
1670 | extent_root->root_key.objectid, | |
3bb1a1bc YZ |
1671 | extent_op->orig_generation, |
1672 | extent_op->level, 0); | |
31840ae1 | 1673 | BUG_ON(err); |
c286ac48 | 1674 | |
25179201 | 1675 | mutex_lock(&info->extent_ins_mutex); |
31840ae1 | 1676 | clear_extent_bits(&info->extent_ins, start, end, |
25179201 JB |
1677 | EXTENT_WRITEBACK, GFP_NOFS); |
1678 | mutex_unlock(&info->extent_ins_mutex); | |
31840ae1 ZY |
1679 | |
1680 | key.objectid = start; | |
1681 | key.offset = extent_op->parent; | |
1682 | key.type = BTRFS_EXTENT_REF_KEY; | |
1683 | err = btrfs_set_item_key_safe(trans, extent_root, path, | |
1684 | &key); | |
1685 | BUG_ON(err); | |
1686 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1687 | struct btrfs_extent_ref); | |
1688 | btrfs_set_ref_generation(path->nodes[0], ref, | |
1689 | extent_op->generation); | |
1690 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
1691 | btrfs_release_path(extent_root, path); | |
d8d5f3e1 | 1692 | } else { |
31840ae1 | 1693 | BUG_ON(1); |
d8d5f3e1 | 1694 | } |
31840ae1 | 1695 | kfree(extent_op); |
25179201 | 1696 | unlock_extent(&info->extent_ins, start, end, GFP_NOFS); |
87ef2bb4 CM |
1697 | if (all) |
1698 | search = 0; | |
1699 | else | |
1700 | search = end + 1; | |
31840ae1 | 1701 | |
25179201 | 1702 | cond_resched(); |
037e6390 | 1703 | } |
7bb86316 | 1704 | btrfs_free_path(path); |
037e6390 CM |
1705 | return 0; |
1706 | } | |
1707 | ||
31840ae1 ZY |
1708 | static int pin_down_bytes(struct btrfs_trans_handle *trans, |
1709 | struct btrfs_root *root, | |
1710 | u64 bytenr, u64 num_bytes, int is_data) | |
e20d96d6 | 1711 | { |
1a5bc167 | 1712 | int err = 0; |
31840ae1 | 1713 | struct extent_buffer *buf; |
8ef97622 | 1714 | |
31840ae1 ZY |
1715 | if (is_data) |
1716 | goto pinit; | |
1717 | ||
1718 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); | |
1719 | if (!buf) | |
1720 | goto pinit; | |
1721 | ||
1722 | /* we can reuse a block if it hasn't been written | |
1723 | * and it is from this transaction. We can't | |
1724 | * reuse anything from the tree log root because | |
1725 | * it has tiny sub-transactions. | |
1726 | */ | |
1727 | if (btrfs_buffer_uptodate(buf, 0) && | |
1728 | btrfs_try_tree_lock(buf)) { | |
1729 | u64 header_owner = btrfs_header_owner(buf); | |
1730 | u64 header_transid = btrfs_header_generation(buf); | |
1731 | if (header_owner != BTRFS_TREE_LOG_OBJECTID && | |
1a40e23b | 1732 | header_owner != BTRFS_TREE_RELOC_OBJECTID && |
31840ae1 ZY |
1733 | header_transid == trans->transid && |
1734 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
1735 | clean_tree_block(NULL, root, buf); | |
1736 | btrfs_tree_unlock(buf); | |
5f39d397 | 1737 | free_extent_buffer(buf); |
31840ae1 | 1738 | return 1; |
8ef97622 | 1739 | } |
31840ae1 | 1740 | btrfs_tree_unlock(buf); |
f4b9aa8d | 1741 | } |
31840ae1 ZY |
1742 | free_extent_buffer(buf); |
1743 | pinit: | |
1744 | btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); | |
1745 | ||
be744175 | 1746 | BUG_ON(err < 0); |
e20d96d6 CM |
1747 | return 0; |
1748 | } | |
1749 | ||
fec577fb | 1750 | /* |
a28ec197 | 1751 | * remove an extent from the root, returns 0 on success |
fec577fb | 1752 | */ |
31840ae1 ZY |
1753 | static int __free_extent(struct btrfs_trans_handle *trans, |
1754 | struct btrfs_root *root, | |
1755 | u64 bytenr, u64 num_bytes, u64 parent, | |
7bb86316 | 1756 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 1757 | u64 owner_objectid, int pin, int mark_free) |
a28ec197 | 1758 | { |
5caf2a00 | 1759 | struct btrfs_path *path; |
e2fa7227 | 1760 | struct btrfs_key key; |
1261ec42 CM |
1761 | struct btrfs_fs_info *info = root->fs_info; |
1762 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 1763 | struct extent_buffer *leaf; |
a28ec197 | 1764 | int ret; |
952fccac CM |
1765 | int extent_slot = 0; |
1766 | int found_extent = 0; | |
1767 | int num_to_del = 1; | |
234b63a0 | 1768 | struct btrfs_extent_item *ei; |
cf27e1ee | 1769 | u32 refs; |
037e6390 | 1770 | |
db94535d | 1771 | key.objectid = bytenr; |
62e2749e | 1772 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 1773 | key.offset = num_bytes; |
5caf2a00 | 1774 | path = btrfs_alloc_path(); |
54aa1f4d CM |
1775 | if (!path) |
1776 | return -ENOMEM; | |
5f26f772 | 1777 | |
3c12ac72 | 1778 | path->reada = 1; |
3bb1a1bc YZ |
1779 | ret = lookup_extent_backref(trans, extent_root, path, |
1780 | bytenr, parent, root_objectid, | |
1781 | ref_generation, owner_objectid, 1); | |
7bb86316 | 1782 | if (ret == 0) { |
952fccac CM |
1783 | struct btrfs_key found_key; |
1784 | extent_slot = path->slots[0]; | |
1785 | while(extent_slot > 0) { | |
1786 | extent_slot--; | |
1787 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1788 | extent_slot); | |
1789 | if (found_key.objectid != bytenr) | |
1790 | break; | |
1791 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && | |
1792 | found_key.offset == num_bytes) { | |
1793 | found_extent = 1; | |
1794 | break; | |
1795 | } | |
1796 | if (path->slots[0] - extent_slot > 5) | |
1797 | break; | |
1798 | } | |
31840ae1 ZY |
1799 | if (!found_extent) { |
1800 | ret = remove_extent_backref(trans, extent_root, path); | |
1801 | BUG_ON(ret); | |
1802 | btrfs_release_path(extent_root, path); | |
1803 | ret = btrfs_search_slot(trans, extent_root, | |
1804 | &key, path, -1, 1); | |
1805 | BUG_ON(ret); | |
1806 | extent_slot = path->slots[0]; | |
1807 | } | |
7bb86316 CM |
1808 | } else { |
1809 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
1810 | WARN_ON(1); | |
1811 | printk("Unable to find ref byte nr %Lu root %Lu " | |
3bb1a1bc YZ |
1812 | "gen %Lu owner %Lu\n", bytenr, |
1813 | root_objectid, ref_generation, owner_objectid); | |
7bb86316 | 1814 | } |
5f39d397 CM |
1815 | |
1816 | leaf = path->nodes[0]; | |
952fccac | 1817 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 1818 | struct btrfs_extent_item); |
5f39d397 CM |
1819 | refs = btrfs_extent_refs(leaf, ei); |
1820 | BUG_ON(refs == 0); | |
1821 | refs -= 1; | |
1822 | btrfs_set_extent_refs(leaf, ei, refs); | |
952fccac | 1823 | |
5f39d397 CM |
1824 | btrfs_mark_buffer_dirty(leaf); |
1825 | ||
952fccac | 1826 | if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { |
31840ae1 ZY |
1827 | struct btrfs_extent_ref *ref; |
1828 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1829 | struct btrfs_extent_ref); | |
1830 | BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1); | |
952fccac CM |
1831 | /* if the back ref and the extent are next to each other |
1832 | * they get deleted below in one shot | |
1833 | */ | |
1834 | path->slots[0] = extent_slot; | |
1835 | num_to_del = 2; | |
1836 | } else if (found_extent) { | |
1837 | /* otherwise delete the extent back ref */ | |
31840ae1 | 1838 | ret = remove_extent_backref(trans, extent_root, path); |
952fccac CM |
1839 | BUG_ON(ret); |
1840 | /* if refs are 0, we need to setup the path for deletion */ | |
1841 | if (refs == 0) { | |
1842 | btrfs_release_path(extent_root, path); | |
1843 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
1844 | -1, 1); | |
952fccac CM |
1845 | BUG_ON(ret); |
1846 | } | |
1847 | } | |
1848 | ||
cf27e1ee | 1849 | if (refs == 0) { |
db94535d CM |
1850 | u64 super_used; |
1851 | u64 root_used; | |
21af804c DW |
1852 | #ifdef BIO_RW_DISCARD |
1853 | u64 map_length = num_bytes; | |
1854 | struct btrfs_multi_bio *multi = NULL; | |
1855 | #endif | |
78fae27e CM |
1856 | |
1857 | if (pin) { | |
25179201 | 1858 | mutex_lock(&root->fs_info->pinned_mutex); |
31840ae1 ZY |
1859 | ret = pin_down_bytes(trans, root, bytenr, num_bytes, |
1860 | owner_objectid >= BTRFS_FIRST_FREE_OBJECTID); | |
25179201 | 1861 | mutex_unlock(&root->fs_info->pinned_mutex); |
c549228f Y |
1862 | if (ret > 0) |
1863 | mark_free = 1; | |
1864 | BUG_ON(ret < 0); | |
78fae27e CM |
1865 | } |
1866 | ||
58176a96 | 1867 | /* block accounting for super block */ |
a2135011 | 1868 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
1869 | super_used = btrfs_super_bytes_used(&info->super_copy); |
1870 | btrfs_set_super_bytes_used(&info->super_copy, | |
1871 | super_used - num_bytes); | |
a2135011 | 1872 | spin_unlock_irq(&info->delalloc_lock); |
58176a96 JB |
1873 | |
1874 | /* block accounting for root item */ | |
db94535d | 1875 | root_used = btrfs_root_used(&root->root_item); |
5f39d397 | 1876 | btrfs_set_root_used(&root->root_item, |
db94535d | 1877 | root_used - num_bytes); |
952fccac CM |
1878 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
1879 | num_to_del); | |
31840ae1 | 1880 | BUG_ON(ret); |
25179201 | 1881 | btrfs_release_path(extent_root, path); |
db94535d | 1882 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
0b86a832 | 1883 | mark_free); |
9078a3e1 | 1884 | BUG_ON(ret); |
21af804c DW |
1885 | |
1886 | #ifdef BIO_RW_DISCARD | |
1887 | /* Tell the block device(s) that the sectors can be discarded */ | |
1888 | ret = btrfs_map_block(&root->fs_info->mapping_tree, READ, | |
1889 | bytenr, &map_length, &multi, 0); | |
1890 | if (!ret) { | |
1891 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
1892 | int i; | |
1893 | ||
1894 | if (map_length > num_bytes) | |
1895 | map_length = num_bytes; | |
1896 | ||
1897 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
1898 | blkdev_issue_discard(stripe->dev->bdev, | |
1899 | stripe->physical >> 9, | |
1900 | map_length >> 9); | |
1901 | } | |
1902 | kfree(multi); | |
1903 | } | |
1904 | #endif | |
a28ec197 | 1905 | } |
5caf2a00 | 1906 | btrfs_free_path(path); |
87ef2bb4 | 1907 | finish_current_insert(trans, extent_root, 0); |
a28ec197 CM |
1908 | return ret; |
1909 | } | |
1910 | ||
a28ec197 CM |
1911 | /* |
1912 | * find all the blocks marked as pending in the radix tree and remove | |
1913 | * them from the extent map | |
1914 | */ | |
e089f05c | 1915 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
87ef2bb4 | 1916 | btrfs_root *extent_root, int all) |
a28ec197 CM |
1917 | { |
1918 | int ret; | |
e20d96d6 | 1919 | int err = 0; |
1a5bc167 CM |
1920 | u64 start; |
1921 | u64 end; | |
31840ae1 | 1922 | u64 priv; |
25179201 | 1923 | u64 search = 0; |
d1310b2e | 1924 | struct extent_io_tree *pending_del; |
31840ae1 ZY |
1925 | struct extent_io_tree *extent_ins; |
1926 | struct pending_extent_op *extent_op; | |
25179201 | 1927 | struct btrfs_fs_info *info = extent_root->fs_info; |
8ef97622 | 1928 | |
31840ae1 | 1929 | extent_ins = &extent_root->fs_info->extent_ins; |
1a5bc167 | 1930 | pending_del = &extent_root->fs_info->pending_del; |
a28ec197 CM |
1931 | |
1932 | while(1) { | |
25179201 JB |
1933 | mutex_lock(&info->extent_ins_mutex); |
1934 | ret = find_first_extent_bit(pending_del, search, &start, &end, | |
1935 | EXTENT_WRITEBACK); | |
1936 | if (ret) { | |
1937 | mutex_unlock(&info->extent_ins_mutex); | |
87ef2bb4 | 1938 | if (all && search) { |
25179201 JB |
1939 | search = 0; |
1940 | continue; | |
1941 | } | |
a28ec197 | 1942 | break; |
25179201 JB |
1943 | } |
1944 | ||
1945 | ret = try_lock_extent(extent_ins, start, end, GFP_NOFS); | |
1946 | if (!ret) { | |
1947 | search = end+1; | |
1948 | mutex_unlock(&info->extent_ins_mutex); | |
1949 | cond_resched(); | |
1950 | continue; | |
1951 | } | |
1952 | BUG_ON(ret < 0); | |
31840ae1 ZY |
1953 | |
1954 | ret = get_state_private(pending_del, start, &priv); | |
1955 | BUG_ON(ret); | |
1956 | extent_op = (struct pending_extent_op *)(unsigned long)priv; | |
1957 | ||
25179201 | 1958 | clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK, |
1a5bc167 | 1959 | GFP_NOFS); |
31840ae1 | 1960 | if (!test_range_bit(extent_ins, start, end, |
25179201 JB |
1961 | EXTENT_WRITEBACK, 0)) { |
1962 | mutex_unlock(&info->extent_ins_mutex); | |
31840ae1 | 1963 | free_extent: |
c286ac48 | 1964 | ret = __free_extent(trans, extent_root, |
31840ae1 ZY |
1965 | start, end + 1 - start, |
1966 | extent_op->orig_parent, | |
1967 | extent_root->root_key.objectid, | |
1968 | extent_op->orig_generation, | |
25179201 | 1969 | extent_op->level, 1, 0); |
31840ae1 | 1970 | kfree(extent_op); |
c286ac48 | 1971 | } else { |
31840ae1 | 1972 | kfree(extent_op); |
25179201 JB |
1973 | |
1974 | ret = get_state_private(&info->extent_ins, start, | |
1975 | &priv); | |
31840ae1 ZY |
1976 | BUG_ON(ret); |
1977 | extent_op = (struct pending_extent_op *) | |
25179201 JB |
1978 | (unsigned long)priv; |
1979 | ||
1980 | clear_extent_bits(&info->extent_ins, start, end, | |
1981 | EXTENT_WRITEBACK, GFP_NOFS); | |
31840ae1 | 1982 | |
25179201 | 1983 | mutex_unlock(&info->extent_ins_mutex); |
31840ae1 ZY |
1984 | |
1985 | if (extent_op->type == PENDING_BACKREF_UPDATE) | |
1986 | goto free_extent; | |
1987 | ||
25179201 JB |
1988 | mutex_lock(&extent_root->fs_info->pinned_mutex); |
1989 | ret = pin_down_bytes(trans, extent_root, start, | |
1990 | end + 1 - start, 0); | |
1991 | mutex_unlock(&extent_root->fs_info->pinned_mutex); | |
1992 | ||
31840ae1 | 1993 | ret = update_block_group(trans, extent_root, start, |
25179201 JB |
1994 | end + 1 - start, 0, ret > 0); |
1995 | ||
31840ae1 ZY |
1996 | BUG_ON(ret); |
1997 | kfree(extent_op); | |
c286ac48 | 1998 | } |
1a5bc167 CM |
1999 | if (ret) |
2000 | err = ret; | |
25179201 | 2001 | unlock_extent(extent_ins, start, end, GFP_NOFS); |
c286ac48 | 2002 | |
87ef2bb4 CM |
2003 | if (all) |
2004 | search = 0; | |
2005 | else | |
2006 | search = end + 1; | |
25179201 | 2007 | cond_resched(); |
fec577fb | 2008 | } |
e20d96d6 | 2009 | return err; |
fec577fb CM |
2010 | } |
2011 | ||
2012 | /* | |
2013 | * remove an extent from the root, returns 0 on success | |
2014 | */ | |
925baedd | 2015 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
2016 | struct btrfs_root *root, |
2017 | u64 bytenr, u64 num_bytes, u64 parent, | |
2018 | u64 root_objectid, u64 ref_generation, | |
3bb1a1bc | 2019 | u64 owner_objectid, int pin) |
fec577fb | 2020 | { |
9f5fae2f | 2021 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
fec577fb CM |
2022 | int pending_ret; |
2023 | int ret; | |
a28ec197 | 2024 | |
db94535d | 2025 | WARN_ON(num_bytes < root->sectorsize); |
fec577fb | 2026 | if (root == extent_root) { |
31840ae1 ZY |
2027 | struct pending_extent_op *extent_op; |
2028 | ||
2029 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2030 | BUG_ON(!extent_op); | |
2031 | ||
2032 | extent_op->type = PENDING_EXTENT_DELETE; | |
2033 | extent_op->bytenr = bytenr; | |
2034 | extent_op->num_bytes = num_bytes; | |
2035 | extent_op->parent = parent; | |
2036 | extent_op->orig_parent = parent; | |
2037 | extent_op->generation = ref_generation; | |
2038 | extent_op->orig_generation = ref_generation; | |
2039 | extent_op->level = (int)owner_objectid; | |
2040 | ||
25179201 | 2041 | mutex_lock(&root->fs_info->extent_ins_mutex); |
31840ae1 ZY |
2042 | set_extent_bits(&root->fs_info->pending_del, |
2043 | bytenr, bytenr + num_bytes - 1, | |
25179201 | 2044 | EXTENT_WRITEBACK, GFP_NOFS); |
31840ae1 ZY |
2045 | set_state_private(&root->fs_info->pending_del, |
2046 | bytenr, (unsigned long)extent_op); | |
25179201 | 2047 | mutex_unlock(&root->fs_info->extent_ins_mutex); |
fec577fb CM |
2048 | return 0; |
2049 | } | |
4bef0848 | 2050 | /* if metadata always pin */ |
d00aff00 CM |
2051 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
2052 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { | |
0f9dd46c JB |
2053 | struct btrfs_block_group_cache *cache; |
2054 | ||
d00aff00 | 2055 | /* btrfs_free_reserved_extent */ |
0f9dd46c JB |
2056 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
2057 | BUG_ON(!cache); | |
2058 | btrfs_add_free_space(cache, bytenr, num_bytes); | |
e8569813 | 2059 | update_reserved_extents(root, bytenr, num_bytes, 0); |
d00aff00 CM |
2060 | return 0; |
2061 | } | |
4bef0848 | 2062 | pin = 1; |
d00aff00 | 2063 | } |
4bef0848 CM |
2064 | |
2065 | /* if data pin when any transaction has committed this */ | |
2066 | if (ref_generation != trans->transid) | |
2067 | pin = 1; | |
2068 | ||
31840ae1 | 2069 | ret = __free_extent(trans, root, bytenr, num_bytes, parent, |
3bb1a1bc YZ |
2070 | root_objectid, ref_generation, |
2071 | owner_objectid, pin, pin == 0); | |
ee6e6504 | 2072 | |
87ef2bb4 CM |
2073 | finish_current_insert(trans, root->fs_info->extent_root, 0); |
2074 | pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0); | |
fec577fb CM |
2075 | return ret ? ret : pending_ret; |
2076 | } | |
2077 | ||
925baedd | 2078 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
2079 | struct btrfs_root *root, |
2080 | u64 bytenr, u64 num_bytes, u64 parent, | |
2081 | u64 root_objectid, u64 ref_generation, | |
3bb1a1bc | 2082 | u64 owner_objectid, int pin) |
925baedd CM |
2083 | { |
2084 | int ret; | |
2085 | ||
31840ae1 | 2086 | ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent, |
925baedd | 2087 | root_objectid, ref_generation, |
3bb1a1bc | 2088 | owner_objectid, pin); |
925baedd CM |
2089 | return ret; |
2090 | } | |
2091 | ||
87ee04eb CM |
2092 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
2093 | { | |
2094 | u64 mask = ((u64)root->stripesize - 1); | |
2095 | u64 ret = (val + mask) & ~mask; | |
2096 | return ret; | |
2097 | } | |
2098 | ||
fec577fb CM |
2099 | /* |
2100 | * walks the btree of allocated extents and find a hole of a given size. | |
2101 | * The key ins is changed to record the hole: | |
2102 | * ins->objectid == block start | |
62e2749e | 2103 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
2104 | * ins->offset == number of blocks |
2105 | * Any available blocks before search_start are skipped. | |
2106 | */ | |
98ed5174 CM |
2107 | static int noinline find_free_extent(struct btrfs_trans_handle *trans, |
2108 | struct btrfs_root *orig_root, | |
2109 | u64 num_bytes, u64 empty_size, | |
2110 | u64 search_start, u64 search_end, | |
2111 | u64 hint_byte, struct btrfs_key *ins, | |
2112 | u64 exclude_start, u64 exclude_nr, | |
2113 | int data) | |
fec577fb | 2114 | { |
80eb234a | 2115 | int ret = 0; |
9f5fae2f | 2116 | struct btrfs_root * root = orig_root->fs_info->extent_root; |
db94535d | 2117 | u64 total_needed = num_bytes; |
239b14b3 | 2118 | u64 *last_ptr = NULL; |
4366211c | 2119 | u64 last_wanted = 0; |
80eb234a | 2120 | struct btrfs_block_group_cache *block_group = NULL; |
0ef3e66b | 2121 | int chunk_alloc_done = 0; |
239b14b3 | 2122 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 2123 | int allowed_chunk_alloc = 0; |
80eb234a JB |
2124 | struct list_head *head = NULL, *cur = NULL; |
2125 | int loop = 0; | |
f5a31e16 | 2126 | int extra_loop = 0; |
80eb234a | 2127 | struct btrfs_space_info *space_info; |
fec577fb | 2128 | |
db94535d | 2129 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 2130 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
2131 | ins->objectid = 0; |
2132 | ins->offset = 0; | |
b1a4d965 | 2133 | |
0ef3e66b CM |
2134 | if (orig_root->ref_cows || empty_size) |
2135 | allowed_chunk_alloc = 1; | |
2136 | ||
239b14b3 CM |
2137 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
2138 | last_ptr = &root->fs_info->last_alloc; | |
4366211c | 2139 | empty_cluster = 64 * 1024; |
239b14b3 CM |
2140 | } |
2141 | ||
0f9dd46c | 2142 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) |
239b14b3 | 2143 | last_ptr = &root->fs_info->last_data_alloc; |
0f9dd46c | 2144 | |
239b14b3 | 2145 | if (last_ptr) { |
4366211c | 2146 | if (*last_ptr) { |
239b14b3 | 2147 | hint_byte = *last_ptr; |
4366211c CM |
2148 | last_wanted = *last_ptr; |
2149 | } else | |
239b14b3 | 2150 | empty_size += empty_cluster; |
4366211c CM |
2151 | } else { |
2152 | empty_cluster = 0; | |
239b14b3 | 2153 | } |
a061fc8d | 2154 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 2155 | search_start = max(search_start, hint_byte); |
0b86a832 | 2156 | |
42e70e7a | 2157 | if (last_wanted && search_start != last_wanted) { |
4366211c | 2158 | last_wanted = 0; |
42e70e7a CM |
2159 | empty_size += empty_cluster; |
2160 | } | |
4366211c | 2161 | |
42e70e7a | 2162 | total_needed += empty_size; |
80eb234a | 2163 | block_group = btrfs_lookup_block_group(root->fs_info, search_start); |
d899e052 YZ |
2164 | if (!block_group) |
2165 | block_group = btrfs_lookup_first_block_group(root->fs_info, | |
2166 | search_start); | |
80eb234a | 2167 | space_info = __find_space_info(root->fs_info, data); |
0f9dd46c | 2168 | |
80eb234a JB |
2169 | down_read(&space_info->groups_sem); |
2170 | while (1) { | |
2171 | struct btrfs_free_space *free_space; | |
0f9dd46c | 2172 | /* |
80eb234a JB |
2173 | * the only way this happens if our hint points to a block |
2174 | * group thats not of the proper type, while looping this | |
2175 | * should never happen | |
0f9dd46c | 2176 | */ |
8a1413a2 CM |
2177 | if (empty_size) |
2178 | extra_loop = 1; | |
2179 | ||
42e70e7a CM |
2180 | if (!block_group) |
2181 | goto new_group_no_lock; | |
2182 | ||
25179201 | 2183 | mutex_lock(&block_group->alloc_mutex); |
80eb234a | 2184 | if (unlikely(!block_group_bits(block_group, data))) |
0f9dd46c | 2185 | goto new_group; |
0f9dd46c | 2186 | |
80eb234a | 2187 | ret = cache_block_group(root, block_group); |
25179201 JB |
2188 | if (ret) { |
2189 | mutex_unlock(&block_group->alloc_mutex); | |
80eb234a | 2190 | break; |
25179201 | 2191 | } |
0f9dd46c | 2192 | |
80eb234a JB |
2193 | if (block_group->ro) |
2194 | goto new_group; | |
0f9dd46c | 2195 | |
80eb234a JB |
2196 | free_space = btrfs_find_free_space(block_group, search_start, |
2197 | total_needed); | |
2198 | if (free_space) { | |
2199 | u64 start = block_group->key.objectid; | |
2200 | u64 end = block_group->key.objectid + | |
0f9dd46c | 2201 | block_group->key.offset; |
239b14b3 | 2202 | |
80eb234a | 2203 | search_start = stripe_align(root, free_space->offset); |
0f9dd46c | 2204 | |
80eb234a JB |
2205 | /* move on to the next group */ |
2206 | if (search_start + num_bytes >= search_end) | |
2207 | goto new_group; | |
e37c9e69 | 2208 | |
80eb234a JB |
2209 | /* move on to the next group */ |
2210 | if (search_start + num_bytes > end) | |
2211 | goto new_group; | |
2212 | ||
4366211c | 2213 | if (last_wanted && search_start != last_wanted) { |
3b7885bf | 2214 | total_needed += empty_cluster; |
8a1413a2 | 2215 | empty_size += empty_cluster; |
4366211c | 2216 | last_wanted = 0; |
3b7885bf CM |
2217 | /* |
2218 | * if search_start is still in this block group | |
2219 | * then we just re-search this block group | |
2220 | */ | |
2221 | if (search_start >= start && | |
2222 | search_start < end) { | |
2223 | mutex_unlock(&block_group->alloc_mutex); | |
2224 | continue; | |
2225 | } | |
2226 | ||
2227 | /* else we go to the next block group */ | |
2228 | goto new_group; | |
2229 | } | |
2230 | ||
80eb234a JB |
2231 | if (exclude_nr > 0 && |
2232 | (search_start + num_bytes > exclude_start && | |
2233 | search_start < exclude_start + exclude_nr)) { | |
2234 | search_start = exclude_start + exclude_nr; | |
2235 | /* | |
2236 | * if search_start is still in this block group | |
2237 | * then we just re-search this block group | |
2238 | */ | |
2239 | if (search_start >= start && | |
25179201 JB |
2240 | search_start < end) { |
2241 | mutex_unlock(&block_group->alloc_mutex); | |
4366211c | 2242 | last_wanted = 0; |
80eb234a | 2243 | continue; |
25179201 | 2244 | } |
80eb234a JB |
2245 | |
2246 | /* else we go to the next block group */ | |
2247 | goto new_group; | |
2248 | } | |
2249 | ||
2250 | ins->objectid = search_start; | |
2251 | ins->offset = num_bytes; | |
25179201 JB |
2252 | |
2253 | btrfs_remove_free_space_lock(block_group, search_start, | |
2254 | num_bytes); | |
80eb234a | 2255 | /* we are all good, lets return */ |
25179201 | 2256 | mutex_unlock(&block_group->alloc_mutex); |
80eb234a | 2257 | break; |
0f9dd46c | 2258 | } |
80eb234a | 2259 | new_group: |
42e70e7a CM |
2260 | mutex_unlock(&block_group->alloc_mutex); |
2261 | new_group_no_lock: | |
f5a31e16 CM |
2262 | /* don't try to compare new allocations against the |
2263 | * last allocation any more | |
2264 | */ | |
4366211c | 2265 | last_wanted = 0; |
f5a31e16 | 2266 | |
80eb234a JB |
2267 | /* |
2268 | * Here's how this works. | |
2269 | * loop == 0: we were searching a block group via a hint | |
2270 | * and didn't find anything, so we start at | |
2271 | * the head of the block groups and keep searching | |
2272 | * loop == 1: we're searching through all of the block groups | |
2273 | * if we hit the head again we have searched | |
2274 | * all of the block groups for this space and we | |
2275 | * need to try and allocate, if we cant error out. | |
2276 | * loop == 2: we allocated more space and are looping through | |
2277 | * all of the block groups again. | |
2278 | */ | |
2279 | if (loop == 0) { | |
2280 | head = &space_info->block_groups; | |
2281 | cur = head->next; | |
80eb234a JB |
2282 | loop++; |
2283 | } else if (loop == 1 && cur == head) { | |
f5a31e16 CM |
2284 | int keep_going; |
2285 | ||
2286 | /* at this point we give up on the empty_size | |
2287 | * allocations and just try to allocate the min | |
2288 | * space. | |
2289 | * | |
2290 | * The extra_loop field was set if an empty_size | |
2291 | * allocation was attempted above, and if this | |
2292 | * is try we need to try the loop again without | |
2293 | * the additional empty_size. | |
2294 | */ | |
5b7c3fcc CM |
2295 | total_needed -= empty_size; |
2296 | empty_size = 0; | |
f5a31e16 CM |
2297 | keep_going = extra_loop; |
2298 | loop++; | |
42e70e7a | 2299 | |
80eb234a JB |
2300 | if (allowed_chunk_alloc && !chunk_alloc_done) { |
2301 | up_read(&space_info->groups_sem); | |
2302 | ret = do_chunk_alloc(trans, root, num_bytes + | |
2303 | 2 * 1024 * 1024, data, 1); | |
2304 | if (ret < 0) | |
2305 | break; | |
2306 | down_read(&space_info->groups_sem); | |
80eb234a | 2307 | head = &space_info->block_groups; |
f5a31e16 CM |
2308 | /* |
2309 | * we've allocated a new chunk, keep | |
2310 | * trying | |
2311 | */ | |
2312 | keep_going = 1; | |
80eb234a JB |
2313 | chunk_alloc_done = 1; |
2314 | } else if (!allowed_chunk_alloc) { | |
2315 | space_info->force_alloc = 1; | |
f5a31e16 CM |
2316 | } |
2317 | if (keep_going) { | |
2318 | cur = head->next; | |
2319 | extra_loop = 0; | |
80eb234a JB |
2320 | } else { |
2321 | break; | |
2322 | } | |
2323 | } else if (cur == head) { | |
2324 | break; | |
0f9dd46c | 2325 | } |
0b86a832 | 2326 | |
80eb234a JB |
2327 | block_group = list_entry(cur, struct btrfs_block_group_cache, |
2328 | list); | |
2329 | search_start = block_group->key.objectid; | |
2330 | cur = cur->next; | |
f2654de4 | 2331 | } |
0b86a832 | 2332 | |
80eb234a JB |
2333 | /* we found what we needed */ |
2334 | if (ins->objectid) { | |
2335 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) | |
2336 | trans->block_group = block_group; | |
0f9dd46c | 2337 | |
80eb234a JB |
2338 | if (last_ptr) |
2339 | *last_ptr = ins->objectid + ins->offset; | |
2340 | ret = 0; | |
2341 | } else if (!ret) { | |
2342 | ret = -ENOSPC; | |
be744175 | 2343 | } |
be744175 | 2344 | |
80eb234a | 2345 | up_read(&space_info->groups_sem); |
0f70abe2 | 2346 | return ret; |
fec577fb | 2347 | } |
ec44a35c | 2348 | |
0f9dd46c JB |
2349 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes) |
2350 | { | |
2351 | struct btrfs_block_group_cache *cache; | |
2352 | struct list_head *l; | |
2353 | ||
2354 | printk(KERN_INFO "space_info has %Lu free, is %sfull\n", | |
e8569813 ZY |
2355 | info->total_bytes - info->bytes_used - info->bytes_pinned - |
2356 | info->bytes_reserved, (info->full) ? "" : "not "); | |
0f9dd46c | 2357 | |
80eb234a | 2358 | down_read(&info->groups_sem); |
0f9dd46c JB |
2359 | list_for_each(l, &info->block_groups) { |
2360 | cache = list_entry(l, struct btrfs_block_group_cache, list); | |
2361 | spin_lock(&cache->lock); | |
2362 | printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used " | |
e8569813 | 2363 | "%Lu pinned %Lu reserved\n", |
0f9dd46c | 2364 | cache->key.objectid, cache->key.offset, |
e8569813 ZY |
2365 | btrfs_block_group_used(&cache->item), |
2366 | cache->pinned, cache->reserved); | |
0f9dd46c JB |
2367 | btrfs_dump_free_space(cache, bytes); |
2368 | spin_unlock(&cache->lock); | |
2369 | } | |
80eb234a | 2370 | up_read(&info->groups_sem); |
0f9dd46c | 2371 | } |
e8569813 | 2372 | |
e6dcd2dc CM |
2373 | static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
2374 | struct btrfs_root *root, | |
2375 | u64 num_bytes, u64 min_alloc_size, | |
2376 | u64 empty_size, u64 hint_byte, | |
2377 | u64 search_end, struct btrfs_key *ins, | |
2378 | u64 data) | |
fec577fb CM |
2379 | { |
2380 | int ret; | |
fbdc762b | 2381 | u64 search_start = 0; |
8790d502 | 2382 | u64 alloc_profile; |
1261ec42 | 2383 | struct btrfs_fs_info *info = root->fs_info; |
925baedd | 2384 | |
6324fbf3 | 2385 | if (data) { |
8790d502 CM |
2386 | alloc_profile = info->avail_data_alloc_bits & |
2387 | info->data_alloc_profile; | |
2388 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
6324fbf3 | 2389 | } else if (root == root->fs_info->chunk_root) { |
8790d502 CM |
2390 | alloc_profile = info->avail_system_alloc_bits & |
2391 | info->system_alloc_profile; | |
2392 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
6324fbf3 | 2393 | } else { |
8790d502 CM |
2394 | alloc_profile = info->avail_metadata_alloc_bits & |
2395 | info->metadata_alloc_profile; | |
2396 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
6324fbf3 | 2397 | } |
98d20f67 | 2398 | again: |
a061fc8d | 2399 | data = reduce_alloc_profile(root, data); |
0ef3e66b CM |
2400 | /* |
2401 | * the only place that sets empty_size is btrfs_realloc_node, which | |
2402 | * is not called recursively on allocations | |
2403 | */ | |
2404 | if (empty_size || root->ref_cows) { | |
593060d7 | 2405 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 2406 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b CM |
2407 | 2 * 1024 * 1024, |
2408 | BTRFS_BLOCK_GROUP_METADATA | | |
2409 | (info->metadata_alloc_profile & | |
2410 | info->avail_metadata_alloc_bits), 0); | |
6324fbf3 CM |
2411 | } |
2412 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
0ef3e66b | 2413 | num_bytes + 2 * 1024 * 1024, data, 0); |
6324fbf3 | 2414 | } |
0b86a832 | 2415 | |
db94535d CM |
2416 | WARN_ON(num_bytes < root->sectorsize); |
2417 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
2418 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
2419 | trans->alloc_exclude_start, |
2420 | trans->alloc_exclude_nr, data); | |
3b951516 | 2421 | |
98d20f67 CM |
2422 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
2423 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 2424 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 2425 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b CM |
2426 | do_chunk_alloc(trans, root->fs_info->extent_root, |
2427 | num_bytes, data, 1); | |
98d20f67 CM |
2428 | goto again; |
2429 | } | |
ec44a35c | 2430 | if (ret) { |
0f9dd46c JB |
2431 | struct btrfs_space_info *sinfo; |
2432 | ||
2433 | sinfo = __find_space_info(root->fs_info, data); | |
2434 | printk("allocation failed flags %Lu, wanted %Lu\n", | |
2435 | data, num_bytes); | |
2436 | dump_space_info(sinfo, num_bytes); | |
925baedd | 2437 | BUG(); |
925baedd | 2438 | } |
0f9dd46c JB |
2439 | |
2440 | return ret; | |
e6dcd2dc CM |
2441 | } |
2442 | ||
65b51a00 CM |
2443 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
2444 | { | |
0f9dd46c JB |
2445 | struct btrfs_block_group_cache *cache; |
2446 | ||
0f9dd46c JB |
2447 | cache = btrfs_lookup_block_group(root->fs_info, start); |
2448 | if (!cache) { | |
2449 | printk(KERN_ERR "Unable to find block group for %Lu\n", start); | |
0f9dd46c JB |
2450 | return -ENOSPC; |
2451 | } | |
2452 | btrfs_add_free_space(cache, start, len); | |
1a40e23b | 2453 | update_reserved_extents(root, start, len, 0); |
65b51a00 CM |
2454 | return 0; |
2455 | } | |
2456 | ||
e6dcd2dc CM |
2457 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
2458 | struct btrfs_root *root, | |
2459 | u64 num_bytes, u64 min_alloc_size, | |
2460 | u64 empty_size, u64 hint_byte, | |
2461 | u64 search_end, struct btrfs_key *ins, | |
2462 | u64 data) | |
2463 | { | |
2464 | int ret; | |
e6dcd2dc CM |
2465 | ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, |
2466 | empty_size, hint_byte, search_end, ins, | |
2467 | data); | |
e8569813 | 2468 | update_reserved_extents(root, ins->objectid, ins->offset, 1); |
e6dcd2dc CM |
2469 | return ret; |
2470 | } | |
2471 | ||
2472 | static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
31840ae1 | 2473 | struct btrfs_root *root, u64 parent, |
e6dcd2dc | 2474 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 2475 | u64 owner, struct btrfs_key *ins) |
e6dcd2dc CM |
2476 | { |
2477 | int ret; | |
2478 | int pending_ret; | |
2479 | u64 super_used; | |
2480 | u64 root_used; | |
2481 | u64 num_bytes = ins->offset; | |
2482 | u32 sizes[2]; | |
2483 | struct btrfs_fs_info *info = root->fs_info; | |
2484 | struct btrfs_root *extent_root = info->extent_root; | |
2485 | struct btrfs_extent_item *extent_item; | |
2486 | struct btrfs_extent_ref *ref; | |
2487 | struct btrfs_path *path; | |
2488 | struct btrfs_key keys[2]; | |
fec577fb | 2489 | |
31840ae1 ZY |
2490 | if (parent == 0) |
2491 | parent = ins->objectid; | |
2492 | ||
58176a96 | 2493 | /* block accounting for super block */ |
a2135011 | 2494 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
2495 | super_used = btrfs_super_bytes_used(&info->super_copy); |
2496 | btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); | |
a2135011 | 2497 | spin_unlock_irq(&info->delalloc_lock); |
26b8003f | 2498 | |
58176a96 | 2499 | /* block accounting for root item */ |
db94535d CM |
2500 | root_used = btrfs_root_used(&root->root_item); |
2501 | btrfs_set_root_used(&root->root_item, root_used + num_bytes); | |
58176a96 | 2502 | |
26b8003f | 2503 | if (root == extent_root) { |
31840ae1 ZY |
2504 | struct pending_extent_op *extent_op; |
2505 | ||
2506 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2507 | BUG_ON(!extent_op); | |
2508 | ||
2509 | extent_op->type = PENDING_EXTENT_INSERT; | |
2510 | extent_op->bytenr = ins->objectid; | |
2511 | extent_op->num_bytes = ins->offset; | |
2512 | extent_op->parent = parent; | |
2513 | extent_op->orig_parent = 0; | |
2514 | extent_op->generation = ref_generation; | |
2515 | extent_op->orig_generation = 0; | |
2516 | extent_op->level = (int)owner; | |
2517 | ||
25179201 | 2518 | mutex_lock(&root->fs_info->extent_ins_mutex); |
1a5bc167 CM |
2519 | set_extent_bits(&root->fs_info->extent_ins, ins->objectid, |
2520 | ins->objectid + ins->offset - 1, | |
25179201 | 2521 | EXTENT_WRITEBACK, GFP_NOFS); |
31840ae1 ZY |
2522 | set_state_private(&root->fs_info->extent_ins, |
2523 | ins->objectid, (unsigned long)extent_op); | |
25179201 | 2524 | mutex_unlock(&root->fs_info->extent_ins_mutex); |
26b8003f CM |
2525 | goto update_block; |
2526 | } | |
2527 | ||
47e4bb98 | 2528 | memcpy(&keys[0], ins, sizeof(*ins)); |
47e4bb98 CM |
2529 | keys[1].objectid = ins->objectid; |
2530 | keys[1].type = BTRFS_EXTENT_REF_KEY; | |
31840ae1 | 2531 | keys[1].offset = parent; |
47e4bb98 CM |
2532 | sizes[0] = sizeof(*extent_item); |
2533 | sizes[1] = sizeof(*ref); | |
7bb86316 CM |
2534 | |
2535 | path = btrfs_alloc_path(); | |
2536 | BUG_ON(!path); | |
47e4bb98 CM |
2537 | |
2538 | ret = btrfs_insert_empty_items(trans, extent_root, path, keys, | |
2539 | sizes, 2); | |
ccd467d6 | 2540 | BUG_ON(ret); |
0f9dd46c | 2541 | |
47e4bb98 CM |
2542 | extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2543 | struct btrfs_extent_item); | |
2544 | btrfs_set_extent_refs(path->nodes[0], extent_item, 1); | |
2545 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
2546 | struct btrfs_extent_ref); | |
2547 | ||
2548 | btrfs_set_ref_root(path->nodes[0], ref, root_objectid); | |
2549 | btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); | |
2550 | btrfs_set_ref_objectid(path->nodes[0], ref, owner); | |
31840ae1 | 2551 | btrfs_set_ref_num_refs(path->nodes[0], ref, 1); |
47e4bb98 CM |
2552 | |
2553 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
2554 | ||
2555 | trans->alloc_exclude_start = 0; | |
2556 | trans->alloc_exclude_nr = 0; | |
7bb86316 | 2557 | btrfs_free_path(path); |
87ef2bb4 CM |
2558 | finish_current_insert(trans, extent_root, 0); |
2559 | pending_ret = del_pending_extents(trans, extent_root, 0); | |
f510cfec | 2560 | |
925baedd CM |
2561 | if (ret) |
2562 | goto out; | |
e37c9e69 | 2563 | if (pending_ret) { |
925baedd CM |
2564 | ret = pending_ret; |
2565 | goto out; | |
e37c9e69 | 2566 | } |
26b8003f CM |
2567 | |
2568 | update_block: | |
0b86a832 | 2569 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0); |
f5947066 CM |
2570 | if (ret) { |
2571 | printk("update block group failed for %Lu %Lu\n", | |
2572 | ins->objectid, ins->offset); | |
2573 | BUG(); | |
2574 | } | |
925baedd | 2575 | out: |
e6dcd2dc CM |
2576 | return ret; |
2577 | } | |
2578 | ||
2579 | int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
31840ae1 | 2580 | struct btrfs_root *root, u64 parent, |
e6dcd2dc | 2581 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 2582 | u64 owner, struct btrfs_key *ins) |
e6dcd2dc CM |
2583 | { |
2584 | int ret; | |
1c2308f8 CM |
2585 | |
2586 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) | |
2587 | return 0; | |
3bb1a1bc YZ |
2588 | ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, |
2589 | ref_generation, owner, ins); | |
e8569813 | 2590 | update_reserved_extents(root, ins->objectid, ins->offset, 0); |
e6dcd2dc CM |
2591 | return ret; |
2592 | } | |
e02119d5 CM |
2593 | |
2594 | /* | |
2595 | * this is used by the tree logging recovery code. It records that | |
2596 | * an extent has been allocated and makes sure to clear the free | |
2597 | * space cache bits as well | |
2598 | */ | |
2599 | int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans, | |
31840ae1 | 2600 | struct btrfs_root *root, u64 parent, |
e02119d5 | 2601 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 2602 | u64 owner, struct btrfs_key *ins) |
e02119d5 CM |
2603 | { |
2604 | int ret; | |
2605 | struct btrfs_block_group_cache *block_group; | |
2606 | ||
e02119d5 | 2607 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
25179201 | 2608 | mutex_lock(&block_group->alloc_mutex); |
e02119d5 CM |
2609 | cache_block_group(root, block_group); |
2610 | ||
25179201 JB |
2611 | ret = btrfs_remove_free_space_lock(block_group, ins->objectid, |
2612 | ins->offset); | |
2613 | mutex_unlock(&block_group->alloc_mutex); | |
0f9dd46c | 2614 | BUG_ON(ret); |
3bb1a1bc YZ |
2615 | ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, |
2616 | ref_generation, owner, ins); | |
e02119d5 CM |
2617 | return ret; |
2618 | } | |
2619 | ||
e6dcd2dc CM |
2620 | /* |
2621 | * finds a free extent and does all the dirty work required for allocation | |
2622 | * returns the key for the extent through ins, and a tree buffer for | |
2623 | * the first block of the extent through buf. | |
2624 | * | |
2625 | * returns 0 if everything worked, non-zero otherwise. | |
2626 | */ | |
2627 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, | |
2628 | struct btrfs_root *root, | |
31840ae1 | 2629 | u64 num_bytes, u64 parent, u64 min_alloc_size, |
e6dcd2dc | 2630 | u64 root_objectid, u64 ref_generation, |
3bb1a1bc | 2631 | u64 owner_objectid, u64 empty_size, u64 hint_byte, |
e6dcd2dc CM |
2632 | u64 search_end, struct btrfs_key *ins, u64 data) |
2633 | { | |
2634 | int ret; | |
2635 | ||
e6dcd2dc CM |
2636 | ret = __btrfs_reserve_extent(trans, root, num_bytes, |
2637 | min_alloc_size, empty_size, hint_byte, | |
2638 | search_end, ins, data); | |
2639 | BUG_ON(ret); | |
d00aff00 | 2640 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { |
31840ae1 ZY |
2641 | ret = __btrfs_alloc_reserved_extent(trans, root, parent, |
2642 | root_objectid, ref_generation, | |
3bb1a1bc | 2643 | owner_objectid, ins); |
d00aff00 | 2644 | BUG_ON(ret); |
e6dcd2dc | 2645 | |
e8569813 ZY |
2646 | } else { |
2647 | update_reserved_extents(root, ins->objectid, ins->offset, 1); | |
d00aff00 | 2648 | } |
925baedd | 2649 | return ret; |
fec577fb | 2650 | } |
65b51a00 CM |
2651 | |
2652 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, | |
2653 | struct btrfs_root *root, | |
2654 | u64 bytenr, u32 blocksize) | |
2655 | { | |
2656 | struct extent_buffer *buf; | |
2657 | ||
2658 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
2659 | if (!buf) | |
2660 | return ERR_PTR(-ENOMEM); | |
2661 | btrfs_set_header_generation(buf, trans->transid); | |
2662 | btrfs_tree_lock(buf); | |
2663 | clean_tree_block(trans, root, buf); | |
2664 | btrfs_set_buffer_uptodate(buf); | |
d0c803c4 CM |
2665 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
2666 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
2667 | buf->start + buf->len - 1, GFP_NOFS); | |
2668 | } else { | |
2669 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 2670 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 2671 | } |
65b51a00 CM |
2672 | trans->blocks_used++; |
2673 | return buf; | |
2674 | } | |
2675 | ||
fec577fb CM |
2676 | /* |
2677 | * helper function to allocate a block for a given tree | |
2678 | * returns the tree buffer or NULL. | |
2679 | */ | |
5f39d397 | 2680 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
7bb86316 | 2681 | struct btrfs_root *root, |
31840ae1 | 2682 | u32 blocksize, u64 parent, |
7bb86316 CM |
2683 | u64 root_objectid, |
2684 | u64 ref_generation, | |
7bb86316 CM |
2685 | int level, |
2686 | u64 hint, | |
5f39d397 | 2687 | u64 empty_size) |
fec577fb | 2688 | { |
e2fa7227 | 2689 | struct btrfs_key ins; |
fec577fb | 2690 | int ret; |
5f39d397 | 2691 | struct extent_buffer *buf; |
fec577fb | 2692 | |
31840ae1 | 2693 | ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize, |
3bb1a1bc | 2694 | root_objectid, ref_generation, level, |
31840ae1 | 2695 | empty_size, hint, (u64)-1, &ins, 0); |
fec577fb | 2696 | if (ret) { |
54aa1f4d CM |
2697 | BUG_ON(ret > 0); |
2698 | return ERR_PTR(ret); | |
fec577fb | 2699 | } |
55c69072 | 2700 | |
65b51a00 | 2701 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize); |
fec577fb CM |
2702 | return buf; |
2703 | } | |
a28ec197 | 2704 | |
e02119d5 CM |
2705 | int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, |
2706 | struct btrfs_root *root, struct extent_buffer *leaf) | |
6407bf6d | 2707 | { |
7bb86316 CM |
2708 | u64 leaf_owner; |
2709 | u64 leaf_generation; | |
5f39d397 | 2710 | struct btrfs_key key; |
6407bf6d CM |
2711 | struct btrfs_file_extent_item *fi; |
2712 | int i; | |
2713 | int nritems; | |
2714 | int ret; | |
2715 | ||
5f39d397 CM |
2716 | BUG_ON(!btrfs_is_leaf(leaf)); |
2717 | nritems = btrfs_header_nritems(leaf); | |
7bb86316 CM |
2718 | leaf_owner = btrfs_header_owner(leaf); |
2719 | leaf_generation = btrfs_header_generation(leaf); | |
2720 | ||
6407bf6d | 2721 | for (i = 0; i < nritems; i++) { |
db94535d | 2722 | u64 disk_bytenr; |
e34a5b4f | 2723 | cond_resched(); |
5f39d397 CM |
2724 | |
2725 | btrfs_item_key_to_cpu(leaf, &key, i); | |
2726 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d CM |
2727 | continue; |
2728 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
5f39d397 CM |
2729 | if (btrfs_file_extent_type(leaf, fi) == |
2730 | BTRFS_FILE_EXTENT_INLINE) | |
236454df | 2731 | continue; |
6407bf6d CM |
2732 | /* |
2733 | * FIXME make sure to insert a trans record that | |
2734 | * repeats the snapshot del on crash | |
2735 | */ | |
db94535d CM |
2736 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
2737 | if (disk_bytenr == 0) | |
3a686375 | 2738 | continue; |
4a096752 | 2739 | |
925baedd | 2740 | ret = __btrfs_free_extent(trans, root, disk_bytenr, |
7bb86316 | 2741 | btrfs_file_extent_disk_num_bytes(leaf, fi), |
31840ae1 | 2742 | leaf->start, leaf_owner, leaf_generation, |
3bb1a1bc | 2743 | key.objectid, 0); |
31840ae1 | 2744 | BUG_ON(ret); |
2dd3e67b CM |
2745 | |
2746 | atomic_inc(&root->fs_info->throttle_gen); | |
2747 | wake_up(&root->fs_info->transaction_throttle); | |
2748 | cond_resched(); | |
6407bf6d CM |
2749 | } |
2750 | return 0; | |
2751 | } | |
2752 | ||
e02119d5 CM |
2753 | static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans, |
2754 | struct btrfs_root *root, | |
2755 | struct btrfs_leaf_ref *ref) | |
31153d81 YZ |
2756 | { |
2757 | int i; | |
2758 | int ret; | |
2759 | struct btrfs_extent_info *info = ref->extents; | |
2760 | ||
31153d81 | 2761 | for (i = 0; i < ref->nritems; i++) { |
31840ae1 ZY |
2762 | ret = __btrfs_free_extent(trans, root, info->bytenr, |
2763 | info->num_bytes, ref->bytenr, | |
2764 | ref->owner, ref->generation, | |
3bb1a1bc | 2765 | info->objectid, 0); |
2dd3e67b CM |
2766 | |
2767 | atomic_inc(&root->fs_info->throttle_gen); | |
2768 | wake_up(&root->fs_info->transaction_throttle); | |
2769 | cond_resched(); | |
2770 | ||
31153d81 YZ |
2771 | BUG_ON(ret); |
2772 | info++; | |
2773 | } | |
31153d81 YZ |
2774 | |
2775 | return 0; | |
2776 | } | |
2777 | ||
333db94c CM |
2778 | int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, |
2779 | u32 *refs) | |
2780 | { | |
017e5369 | 2781 | int ret; |
f87f057b | 2782 | |
31840ae1 | 2783 | ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs); |
f87f057b CM |
2784 | BUG_ON(ret); |
2785 | ||
2786 | #if 0 // some debugging code in case we see problems here | |
2787 | /* if the refs count is one, it won't get increased again. But | |
2788 | * if the ref count is > 1, someone may be decreasing it at | |
2789 | * the same time we are. | |
2790 | */ | |
2791 | if (*refs != 1) { | |
2792 | struct extent_buffer *eb = NULL; | |
2793 | eb = btrfs_find_create_tree_block(root, start, len); | |
2794 | if (eb) | |
2795 | btrfs_tree_lock(eb); | |
2796 | ||
2797 | mutex_lock(&root->fs_info->alloc_mutex); | |
2798 | ret = lookup_extent_ref(NULL, root, start, len, refs); | |
2799 | BUG_ON(ret); | |
2800 | mutex_unlock(&root->fs_info->alloc_mutex); | |
2801 | ||
2802 | if (eb) { | |
2803 | btrfs_tree_unlock(eb); | |
2804 | free_extent_buffer(eb); | |
2805 | } | |
2806 | if (*refs == 1) { | |
2807 | printk("block %llu went down to one during drop_snap\n", | |
2808 | (unsigned long long)start); | |
2809 | } | |
2810 | ||
2811 | } | |
2812 | #endif | |
2813 | ||
e7a84565 | 2814 | cond_resched(); |
017e5369 | 2815 | return ret; |
333db94c CM |
2816 | } |
2817 | ||
9aca1d51 CM |
2818 | /* |
2819 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
2820 | * counts as it goes. | |
2821 | */ | |
98ed5174 CM |
2822 | static int noinline walk_down_tree(struct btrfs_trans_handle *trans, |
2823 | struct btrfs_root *root, | |
2824 | struct btrfs_path *path, int *level) | |
20524f02 | 2825 | { |
7bb86316 CM |
2826 | u64 root_owner; |
2827 | u64 root_gen; | |
2828 | u64 bytenr; | |
ca7a79ad | 2829 | u64 ptr_gen; |
5f39d397 CM |
2830 | struct extent_buffer *next; |
2831 | struct extent_buffer *cur; | |
7bb86316 | 2832 | struct extent_buffer *parent; |
31153d81 | 2833 | struct btrfs_leaf_ref *ref; |
db94535d | 2834 | u32 blocksize; |
20524f02 CM |
2835 | int ret; |
2836 | u32 refs; | |
2837 | ||
5caf2a00 CM |
2838 | WARN_ON(*level < 0); |
2839 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
333db94c | 2840 | ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start, |
db94535d | 2841 | path->nodes[*level]->len, &refs); |
20524f02 CM |
2842 | BUG_ON(ret); |
2843 | if (refs > 1) | |
2844 | goto out; | |
e011599b | 2845 | |
9aca1d51 CM |
2846 | /* |
2847 | * walk down to the last node level and free all the leaves | |
2848 | */ | |
6407bf6d | 2849 | while(*level >= 0) { |
5caf2a00 CM |
2850 | WARN_ON(*level < 0); |
2851 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
20524f02 | 2852 | cur = path->nodes[*level]; |
e011599b | 2853 | |
5f39d397 | 2854 | if (btrfs_header_level(cur) != *level) |
2c90e5d6 | 2855 | WARN_ON(1); |
e011599b | 2856 | |
7518a238 | 2857 | if (path->slots[*level] >= |
5f39d397 | 2858 | btrfs_header_nritems(cur)) |
20524f02 | 2859 | break; |
6407bf6d | 2860 | if (*level == 0) { |
e02119d5 | 2861 | ret = btrfs_drop_leaf_ref(trans, root, cur); |
6407bf6d CM |
2862 | BUG_ON(ret); |
2863 | break; | |
2864 | } | |
db94535d | 2865 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
ca7a79ad | 2866 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); |
db94535d | 2867 | blocksize = btrfs_level_size(root, *level - 1); |
925baedd | 2868 | |
333db94c | 2869 | ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); |
6407bf6d CM |
2870 | BUG_ON(ret); |
2871 | if (refs != 1) { | |
7bb86316 CM |
2872 | parent = path->nodes[*level]; |
2873 | root_owner = btrfs_header_owner(parent); | |
2874 | root_gen = btrfs_header_generation(parent); | |
20524f02 | 2875 | path->slots[*level]++; |
f87f057b | 2876 | |
925baedd | 2877 | ret = __btrfs_free_extent(trans, root, bytenr, |
31840ae1 | 2878 | blocksize, parent->start, |
3bb1a1bc YZ |
2879 | root_owner, root_gen, |
2880 | *level - 1, 1); | |
20524f02 | 2881 | BUG_ON(ret); |
18e35e0a CM |
2882 | |
2883 | atomic_inc(&root->fs_info->throttle_gen); | |
2884 | wake_up(&root->fs_info->transaction_throttle); | |
2dd3e67b | 2885 | cond_resched(); |
18e35e0a | 2886 | |
20524f02 CM |
2887 | continue; |
2888 | } | |
f87f057b CM |
2889 | /* |
2890 | * at this point, we have a single ref, and since the | |
2891 | * only place referencing this extent is a dead root | |
2892 | * the reference count should never go higher. | |
2893 | * So, we don't need to check it again | |
2894 | */ | |
31153d81 | 2895 | if (*level == 1) { |
017e5369 | 2896 | ref = btrfs_lookup_leaf_ref(root, bytenr); |
1a40e23b ZY |
2897 | if (ref && ref->generation != ptr_gen) { |
2898 | btrfs_free_leaf_ref(root, ref); | |
2899 | ref = NULL; | |
2900 | } | |
31153d81 | 2901 | if (ref) { |
e02119d5 | 2902 | ret = cache_drop_leaf_ref(trans, root, ref); |
31153d81 YZ |
2903 | BUG_ON(ret); |
2904 | btrfs_remove_leaf_ref(root, ref); | |
bcc63abb | 2905 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
2906 | *level = 0; |
2907 | break; | |
2908 | } | |
3bb1a1bc | 2909 | if (printk_ratelimit()) { |
37d1aeee CM |
2910 | printk("leaf ref miss for bytenr %llu\n", |
2911 | (unsigned long long)bytenr); | |
3bb1a1bc | 2912 | } |
31153d81 | 2913 | } |
db94535d | 2914 | next = btrfs_find_tree_block(root, bytenr, blocksize); |
1259ab75 | 2915 | if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) { |
5f39d397 | 2916 | free_extent_buffer(next); |
333db94c | 2917 | |
ca7a79ad CM |
2918 | next = read_tree_block(root, bytenr, blocksize, |
2919 | ptr_gen); | |
e7a84565 | 2920 | cond_resched(); |
f87f057b CM |
2921 | #if 0 |
2922 | /* | |
2923 | * this is a debugging check and can go away | |
2924 | * the ref should never go all the way down to 1 | |
2925 | * at this point | |
2926 | */ | |
e6dcd2dc CM |
2927 | ret = lookup_extent_ref(NULL, root, bytenr, blocksize, |
2928 | &refs); | |
e9d0b13b | 2929 | BUG_ON(ret); |
f87f057b CM |
2930 | WARN_ON(refs != 1); |
2931 | #endif | |
e9d0b13b | 2932 | } |
5caf2a00 | 2933 | WARN_ON(*level <= 0); |
83e15a28 | 2934 | if (path->nodes[*level-1]) |
5f39d397 | 2935 | free_extent_buffer(path->nodes[*level-1]); |
20524f02 | 2936 | path->nodes[*level-1] = next; |
5f39d397 | 2937 | *level = btrfs_header_level(next); |
20524f02 | 2938 | path->slots[*level] = 0; |
2dd3e67b | 2939 | cond_resched(); |
20524f02 CM |
2940 | } |
2941 | out: | |
5caf2a00 CM |
2942 | WARN_ON(*level < 0); |
2943 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
7bb86316 CM |
2944 | |
2945 | if (path->nodes[*level] == root->node) { | |
7bb86316 | 2946 | parent = path->nodes[*level]; |
31153d81 | 2947 | bytenr = path->nodes[*level]->start; |
7bb86316 CM |
2948 | } else { |
2949 | parent = path->nodes[*level + 1]; | |
31153d81 | 2950 | bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]); |
7bb86316 CM |
2951 | } |
2952 | ||
31153d81 YZ |
2953 | blocksize = btrfs_level_size(root, *level); |
2954 | root_owner = btrfs_header_owner(parent); | |
7bb86316 | 2955 | root_gen = btrfs_header_generation(parent); |
31153d81 YZ |
2956 | |
2957 | ret = __btrfs_free_extent(trans, root, bytenr, blocksize, | |
31840ae1 | 2958 | parent->start, root_owner, root_gen, |
3bb1a1bc | 2959 | *level, 1); |
5f39d397 | 2960 | free_extent_buffer(path->nodes[*level]); |
20524f02 CM |
2961 | path->nodes[*level] = NULL; |
2962 | *level += 1; | |
2963 | BUG_ON(ret); | |
f87f057b | 2964 | |
e7a84565 | 2965 | cond_resched(); |
20524f02 CM |
2966 | return 0; |
2967 | } | |
2968 | ||
f82d02d9 YZ |
2969 | /* |
2970 | * helper function for drop_subtree, this function is similar to | |
2971 | * walk_down_tree. The main difference is that it checks reference | |
2972 | * counts while tree blocks are locked. | |
2973 | */ | |
2974 | static int noinline walk_down_subtree(struct btrfs_trans_handle *trans, | |
2975 | struct btrfs_root *root, | |
2976 | struct btrfs_path *path, int *level) | |
2977 | { | |
2978 | struct extent_buffer *next; | |
2979 | struct extent_buffer *cur; | |
2980 | struct extent_buffer *parent; | |
2981 | u64 bytenr; | |
2982 | u64 ptr_gen; | |
2983 | u32 blocksize; | |
2984 | u32 refs; | |
2985 | int ret; | |
2986 | ||
2987 | cur = path->nodes[*level]; | |
2988 | ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len, | |
2989 | &refs); | |
2990 | BUG_ON(ret); | |
2991 | if (refs > 1) | |
2992 | goto out; | |
2993 | ||
2994 | while (*level >= 0) { | |
2995 | cur = path->nodes[*level]; | |
2996 | if (*level == 0) { | |
2997 | ret = btrfs_drop_leaf_ref(trans, root, cur); | |
2998 | BUG_ON(ret); | |
2999 | clean_tree_block(trans, root, cur); | |
3000 | break; | |
3001 | } | |
3002 | if (path->slots[*level] >= btrfs_header_nritems(cur)) { | |
3003 | clean_tree_block(trans, root, cur); | |
3004 | break; | |
3005 | } | |
3006 | ||
3007 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
3008 | blocksize = btrfs_level_size(root, *level - 1); | |
3009 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
3010 | ||
3011 | next = read_tree_block(root, bytenr, blocksize, ptr_gen); | |
3012 | btrfs_tree_lock(next); | |
3013 | ||
3014 | ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize, | |
3015 | &refs); | |
3016 | BUG_ON(ret); | |
3017 | if (refs > 1) { | |
3018 | parent = path->nodes[*level]; | |
3019 | ret = btrfs_free_extent(trans, root, bytenr, | |
3020 | blocksize, parent->start, | |
3021 | btrfs_header_owner(parent), | |
3022 | btrfs_header_generation(parent), | |
3023 | *level - 1, 1); | |
3024 | BUG_ON(ret); | |
3025 | path->slots[*level]++; | |
3026 | btrfs_tree_unlock(next); | |
3027 | free_extent_buffer(next); | |
3028 | continue; | |
3029 | } | |
3030 | ||
3031 | *level = btrfs_header_level(next); | |
3032 | path->nodes[*level] = next; | |
3033 | path->slots[*level] = 0; | |
3034 | path->locks[*level] = 1; | |
3035 | cond_resched(); | |
3036 | } | |
3037 | out: | |
3038 | parent = path->nodes[*level + 1]; | |
3039 | bytenr = path->nodes[*level]->start; | |
3040 | blocksize = path->nodes[*level]->len; | |
3041 | ||
3042 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, | |
3043 | parent->start, btrfs_header_owner(parent), | |
3044 | btrfs_header_generation(parent), *level, 1); | |
3045 | BUG_ON(ret); | |
3046 | ||
3047 | if (path->locks[*level]) { | |
3048 | btrfs_tree_unlock(path->nodes[*level]); | |
3049 | path->locks[*level] = 0; | |
3050 | } | |
3051 | free_extent_buffer(path->nodes[*level]); | |
3052 | path->nodes[*level] = NULL; | |
3053 | *level += 1; | |
3054 | cond_resched(); | |
3055 | return 0; | |
3056 | } | |
3057 | ||
9aca1d51 CM |
3058 | /* |
3059 | * helper for dropping snapshots. This walks back up the tree in the path | |
3060 | * to find the first node higher up where we haven't yet gone through | |
3061 | * all the slots | |
3062 | */ | |
98ed5174 CM |
3063 | static int noinline walk_up_tree(struct btrfs_trans_handle *trans, |
3064 | struct btrfs_root *root, | |
f82d02d9 YZ |
3065 | struct btrfs_path *path, |
3066 | int *level, int max_level) | |
20524f02 | 3067 | { |
7bb86316 CM |
3068 | u64 root_owner; |
3069 | u64 root_gen; | |
3070 | struct btrfs_root_item *root_item = &root->root_item; | |
20524f02 CM |
3071 | int i; |
3072 | int slot; | |
3073 | int ret; | |
9f3a7427 | 3074 | |
f82d02d9 | 3075 | for (i = *level; i < max_level && path->nodes[i]; i++) { |
20524f02 | 3076 | slot = path->slots[i]; |
5f39d397 CM |
3077 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { |
3078 | struct extent_buffer *node; | |
3079 | struct btrfs_disk_key disk_key; | |
3080 | node = path->nodes[i]; | |
20524f02 CM |
3081 | path->slots[i]++; |
3082 | *level = i; | |
9f3a7427 | 3083 | WARN_ON(*level == 0); |
5f39d397 | 3084 | btrfs_node_key(node, &disk_key, path->slots[i]); |
9f3a7427 | 3085 | memcpy(&root_item->drop_progress, |
5f39d397 | 3086 | &disk_key, sizeof(disk_key)); |
9f3a7427 | 3087 | root_item->drop_level = i; |
20524f02 CM |
3088 | return 0; |
3089 | } else { | |
31840ae1 ZY |
3090 | struct extent_buffer *parent; |
3091 | if (path->nodes[*level] == root->node) | |
3092 | parent = path->nodes[*level]; | |
3093 | else | |
3094 | parent = path->nodes[*level + 1]; | |
3095 | ||
3096 | root_owner = btrfs_header_owner(parent); | |
3097 | root_gen = btrfs_header_generation(parent); | |
f82d02d9 YZ |
3098 | |
3099 | clean_tree_block(trans, root, path->nodes[*level]); | |
e089f05c | 3100 | ret = btrfs_free_extent(trans, root, |
db94535d | 3101 | path->nodes[*level]->start, |
7bb86316 | 3102 | path->nodes[*level]->len, |
3bb1a1bc YZ |
3103 | parent->start, root_owner, |
3104 | root_gen, *level, 1); | |
6407bf6d | 3105 | BUG_ON(ret); |
f82d02d9 YZ |
3106 | if (path->locks[*level]) { |
3107 | btrfs_tree_unlock(path->nodes[*level]); | |
3108 | path->locks[*level] = 0; | |
3109 | } | |
5f39d397 | 3110 | free_extent_buffer(path->nodes[*level]); |
83e15a28 | 3111 | path->nodes[*level] = NULL; |
20524f02 | 3112 | *level = i + 1; |
20524f02 CM |
3113 | } |
3114 | } | |
3115 | return 1; | |
3116 | } | |
3117 | ||
9aca1d51 CM |
3118 | /* |
3119 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
3120 | * the tree freeing any blocks that have a ref count of zero after being | |
3121 | * decremented. | |
3122 | */ | |
e089f05c | 3123 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
9f3a7427 | 3124 | *root) |
20524f02 | 3125 | { |
3768f368 | 3126 | int ret = 0; |
9aca1d51 | 3127 | int wret; |
20524f02 | 3128 | int level; |
5caf2a00 | 3129 | struct btrfs_path *path; |
20524f02 CM |
3130 | int i; |
3131 | int orig_level; | |
9f3a7427 | 3132 | struct btrfs_root_item *root_item = &root->root_item; |
20524f02 | 3133 | |
a2135011 | 3134 | WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); |
5caf2a00 CM |
3135 | path = btrfs_alloc_path(); |
3136 | BUG_ON(!path); | |
20524f02 | 3137 | |
5f39d397 | 3138 | level = btrfs_header_level(root->node); |
20524f02 | 3139 | orig_level = level; |
9f3a7427 CM |
3140 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
3141 | path->nodes[level] = root->node; | |
f510cfec | 3142 | extent_buffer_get(root->node); |
9f3a7427 CM |
3143 | path->slots[level] = 0; |
3144 | } else { | |
3145 | struct btrfs_key key; | |
5f39d397 CM |
3146 | struct btrfs_disk_key found_key; |
3147 | struct extent_buffer *node; | |
6702ed49 | 3148 | |
9f3a7427 | 3149 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
6702ed49 CM |
3150 | level = root_item->drop_level; |
3151 | path->lowest_level = level; | |
9f3a7427 | 3152 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6702ed49 | 3153 | if (wret < 0) { |
9f3a7427 CM |
3154 | ret = wret; |
3155 | goto out; | |
3156 | } | |
5f39d397 CM |
3157 | node = path->nodes[level]; |
3158 | btrfs_node_key(node, &found_key, path->slots[level]); | |
3159 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, | |
3160 | sizeof(found_key))); | |
7d9eb12c CM |
3161 | /* |
3162 | * unlock our path, this is safe because only this | |
3163 | * function is allowed to delete this snapshot | |
3164 | */ | |
925baedd CM |
3165 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3166 | if (path->nodes[i] && path->locks[i]) { | |
3167 | path->locks[i] = 0; | |
3168 | btrfs_tree_unlock(path->nodes[i]); | |
3169 | } | |
3170 | } | |
9f3a7427 | 3171 | } |
20524f02 | 3172 | while(1) { |
5caf2a00 | 3173 | wret = walk_down_tree(trans, root, path, &level); |
9aca1d51 | 3174 | if (wret > 0) |
20524f02 | 3175 | break; |
9aca1d51 CM |
3176 | if (wret < 0) |
3177 | ret = wret; | |
3178 | ||
f82d02d9 YZ |
3179 | wret = walk_up_tree(trans, root, path, &level, |
3180 | BTRFS_MAX_LEVEL); | |
9aca1d51 | 3181 | if (wret > 0) |
20524f02 | 3182 | break; |
9aca1d51 CM |
3183 | if (wret < 0) |
3184 | ret = wret; | |
e7a84565 CM |
3185 | if (trans->transaction->in_commit) { |
3186 | ret = -EAGAIN; | |
3187 | break; | |
3188 | } | |
18e35e0a | 3189 | atomic_inc(&root->fs_info->throttle_gen); |
017e5369 | 3190 | wake_up(&root->fs_info->transaction_throttle); |
20524f02 | 3191 | } |
83e15a28 | 3192 | for (i = 0; i <= orig_level; i++) { |
5caf2a00 | 3193 | if (path->nodes[i]) { |
5f39d397 | 3194 | free_extent_buffer(path->nodes[i]); |
0f82731f | 3195 | path->nodes[i] = NULL; |
83e15a28 | 3196 | } |
20524f02 | 3197 | } |
9f3a7427 | 3198 | out: |
5caf2a00 | 3199 | btrfs_free_path(path); |
9aca1d51 | 3200 | return ret; |
20524f02 | 3201 | } |
9078a3e1 | 3202 | |
f82d02d9 YZ |
3203 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
3204 | struct btrfs_root *root, | |
3205 | struct extent_buffer *node, | |
3206 | struct extent_buffer *parent) | |
3207 | { | |
3208 | struct btrfs_path *path; | |
3209 | int level; | |
3210 | int parent_level; | |
3211 | int ret = 0; | |
3212 | int wret; | |
3213 | ||
3214 | path = btrfs_alloc_path(); | |
3215 | BUG_ON(!path); | |
3216 | ||
3217 | BUG_ON(!btrfs_tree_locked(parent)); | |
3218 | parent_level = btrfs_header_level(parent); | |
3219 | extent_buffer_get(parent); | |
3220 | path->nodes[parent_level] = parent; | |
3221 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
3222 | ||
3223 | BUG_ON(!btrfs_tree_locked(node)); | |
3224 | level = btrfs_header_level(node); | |
3225 | extent_buffer_get(node); | |
3226 | path->nodes[level] = node; | |
3227 | path->slots[level] = 0; | |
3228 | ||
3229 | while (1) { | |
3230 | wret = walk_down_subtree(trans, root, path, &level); | |
3231 | if (wret < 0) | |
3232 | ret = wret; | |
3233 | if (wret != 0) | |
3234 | break; | |
3235 | ||
3236 | wret = walk_up_tree(trans, root, path, &level, parent_level); | |
3237 | if (wret < 0) | |
3238 | ret = wret; | |
3239 | if (wret != 0) | |
3240 | break; | |
3241 | } | |
3242 | ||
3243 | btrfs_free_path(path); | |
3244 | return ret; | |
3245 | } | |
3246 | ||
8e7bf94f CM |
3247 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
3248 | unsigned long nr) | |
3249 | { | |
3250 | return min(last, start + nr - 1); | |
3251 | } | |
3252 | ||
98ed5174 CM |
3253 | static int noinline relocate_inode_pages(struct inode *inode, u64 start, |
3254 | u64 len) | |
edbd8d4e CM |
3255 | { |
3256 | u64 page_start; | |
3257 | u64 page_end; | |
1a40e23b | 3258 | unsigned long first_index; |
edbd8d4e | 3259 | unsigned long last_index; |
edbd8d4e CM |
3260 | unsigned long i; |
3261 | struct page *page; | |
d1310b2e | 3262 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 3263 | struct file_ra_state *ra; |
3eaa2885 | 3264 | struct btrfs_ordered_extent *ordered; |
1a40e23b ZY |
3265 | unsigned int total_read = 0; |
3266 | unsigned int total_dirty = 0; | |
3267 | int ret = 0; | |
4313b399 CM |
3268 | |
3269 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
3270 | |
3271 | mutex_lock(&inode->i_mutex); | |
1a40e23b | 3272 | first_index = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
3273 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
3274 | ||
1a40e23b ZY |
3275 | /* make sure the dirty trick played by the caller work */ |
3276 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
3277 | first_index, last_index); | |
3278 | if (ret) | |
3279 | goto out_unlock; | |
8e7bf94f | 3280 | |
4313b399 | 3281 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 3282 | |
1a40e23b ZY |
3283 | for (i = first_index ; i <= last_index; i++) { |
3284 | if (total_read % ra->ra_pages == 0) { | |
8e7bf94f | 3285 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, |
1a40e23b | 3286 | calc_ra(i, last_index, ra->ra_pages)); |
8e7bf94f CM |
3287 | } |
3288 | total_read++; | |
3eaa2885 CM |
3289 | again: |
3290 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
1a40e23b | 3291 | BUG_ON(1); |
edbd8d4e | 3292 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 3293 | if (!page) { |
1a40e23b | 3294 | ret = -ENOMEM; |
edbd8d4e | 3295 | goto out_unlock; |
a061fc8d | 3296 | } |
edbd8d4e CM |
3297 | if (!PageUptodate(page)) { |
3298 | btrfs_readpage(NULL, page); | |
3299 | lock_page(page); | |
3300 | if (!PageUptodate(page)) { | |
3301 | unlock_page(page); | |
3302 | page_cache_release(page); | |
1a40e23b | 3303 | ret = -EIO; |
edbd8d4e CM |
3304 | goto out_unlock; |
3305 | } | |
3306 | } | |
ec44a35c | 3307 | wait_on_page_writeback(page); |
3eaa2885 | 3308 | |
edbd8d4e CM |
3309 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
3310 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
d1310b2e | 3311 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 3312 | |
3eaa2885 CM |
3313 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
3314 | if (ordered) { | |
3315 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
3316 | unlock_page(page); | |
3317 | page_cache_release(page); | |
3318 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3319 | btrfs_put_ordered_extent(ordered); | |
3320 | goto again; | |
3321 | } | |
3322 | set_page_extent_mapped(page); | |
3323 | ||
ea8c2819 | 3324 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
1a40e23b ZY |
3325 | if (i == first_index) |
3326 | set_extent_bits(io_tree, page_start, page_end, | |
3327 | EXTENT_BOUNDARY, GFP_NOFS); | |
3328 | ||
a061fc8d | 3329 | set_page_dirty(page); |
1a40e23b | 3330 | total_dirty++; |
edbd8d4e | 3331 | |
d1310b2e | 3332 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
3333 | unlock_page(page); |
3334 | page_cache_release(page); | |
3335 | } | |
3336 | ||
3337 | out_unlock: | |
ec44a35c | 3338 | kfree(ra); |
edbd8d4e | 3339 | mutex_unlock(&inode->i_mutex); |
1a40e23b ZY |
3340 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty); |
3341 | return ret; | |
edbd8d4e CM |
3342 | } |
3343 | ||
1a40e23b ZY |
3344 | static int noinline relocate_data_extent(struct inode *reloc_inode, |
3345 | struct btrfs_key *extent_key, | |
3346 | u64 offset) | |
3347 | { | |
3348 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
3349 | struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree; | |
3350 | struct extent_map *em; | |
6643558d YZ |
3351 | u64 start = extent_key->objectid - offset; |
3352 | u64 end = start + extent_key->offset - 1; | |
bf4ef679 | 3353 | |
1a40e23b ZY |
3354 | em = alloc_extent_map(GFP_NOFS); |
3355 | BUG_ON(!em || IS_ERR(em)); | |
bf4ef679 | 3356 | |
6643558d | 3357 | em->start = start; |
1a40e23b | 3358 | em->len = extent_key->offset; |
c8b97818 | 3359 | em->block_len = extent_key->offset; |
1a40e23b ZY |
3360 | em->block_start = extent_key->objectid; |
3361 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
3362 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
3363 | ||
3364 | /* setup extent map to cheat btrfs_readpage */ | |
6643558d | 3365 | lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
1a40e23b ZY |
3366 | while (1) { |
3367 | int ret; | |
3368 | spin_lock(&em_tree->lock); | |
3369 | ret = add_extent_mapping(em_tree, em); | |
3370 | spin_unlock(&em_tree->lock); | |
3371 | if (ret != -EEXIST) { | |
3372 | free_extent_map(em); | |
bf4ef679 CM |
3373 | break; |
3374 | } | |
6643558d | 3375 | btrfs_drop_extent_cache(reloc_inode, start, end, 0); |
bf4ef679 | 3376 | } |
6643558d | 3377 | unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS); |
bf4ef679 | 3378 | |
6643558d | 3379 | return relocate_inode_pages(reloc_inode, start, extent_key->offset); |
1a40e23b | 3380 | } |
edbd8d4e | 3381 | |
1a40e23b ZY |
3382 | struct btrfs_ref_path { |
3383 | u64 extent_start; | |
3384 | u64 nodes[BTRFS_MAX_LEVEL]; | |
3385 | u64 root_objectid; | |
3386 | u64 root_generation; | |
3387 | u64 owner_objectid; | |
1a40e23b ZY |
3388 | u32 num_refs; |
3389 | int lowest_level; | |
3390 | int current_level; | |
f82d02d9 YZ |
3391 | int shared_level; |
3392 | ||
3393 | struct btrfs_key node_keys[BTRFS_MAX_LEVEL]; | |
3394 | u64 new_nodes[BTRFS_MAX_LEVEL]; | |
1a40e23b | 3395 | }; |
7d9eb12c | 3396 | |
1a40e23b | 3397 | struct disk_extent { |
c8b97818 | 3398 | u64 ram_bytes; |
1a40e23b ZY |
3399 | u64 disk_bytenr; |
3400 | u64 disk_num_bytes; | |
3401 | u64 offset; | |
3402 | u64 num_bytes; | |
c8b97818 CM |
3403 | u8 compression; |
3404 | u8 encryption; | |
3405 | u16 other_encoding; | |
1a40e23b | 3406 | }; |
4313b399 | 3407 | |
1a40e23b ZY |
3408 | static int is_cowonly_root(u64 root_objectid) |
3409 | { | |
3410 | if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || | |
3411 | root_objectid == BTRFS_EXTENT_TREE_OBJECTID || | |
3412 | root_objectid == BTRFS_CHUNK_TREE_OBJECTID || | |
3413 | root_objectid == BTRFS_DEV_TREE_OBJECTID || | |
3414 | root_objectid == BTRFS_TREE_LOG_OBJECTID) | |
3415 | return 1; | |
3416 | return 0; | |
3417 | } | |
edbd8d4e | 3418 | |
1a40e23b ZY |
3419 | static int noinline __next_ref_path(struct btrfs_trans_handle *trans, |
3420 | struct btrfs_root *extent_root, | |
3421 | struct btrfs_ref_path *ref_path, | |
3422 | int first_time) | |
3423 | { | |
3424 | struct extent_buffer *leaf; | |
3425 | struct btrfs_path *path; | |
3426 | struct btrfs_extent_ref *ref; | |
3427 | struct btrfs_key key; | |
3428 | struct btrfs_key found_key; | |
3429 | u64 bytenr; | |
3430 | u32 nritems; | |
3431 | int level; | |
3432 | int ret = 1; | |
edbd8d4e | 3433 | |
1a40e23b ZY |
3434 | path = btrfs_alloc_path(); |
3435 | if (!path) | |
3436 | return -ENOMEM; | |
bf4ef679 | 3437 | |
1a40e23b ZY |
3438 | if (first_time) { |
3439 | ref_path->lowest_level = -1; | |
3440 | ref_path->current_level = -1; | |
f82d02d9 | 3441 | ref_path->shared_level = -1; |
1a40e23b ZY |
3442 | goto walk_up; |
3443 | } | |
3444 | walk_down: | |
3445 | level = ref_path->current_level - 1; | |
3446 | while (level >= -1) { | |
3447 | u64 parent; | |
3448 | if (level < ref_path->lowest_level) | |
3449 | break; | |
bf4ef679 | 3450 | |
1a40e23b ZY |
3451 | if (level >= 0) { |
3452 | bytenr = ref_path->nodes[level]; | |
3453 | } else { | |
3454 | bytenr = ref_path->extent_start; | |
3455 | } | |
3456 | BUG_ON(bytenr == 0); | |
bf4ef679 | 3457 | |
1a40e23b ZY |
3458 | parent = ref_path->nodes[level + 1]; |
3459 | ref_path->nodes[level + 1] = 0; | |
3460 | ref_path->current_level = level; | |
3461 | BUG_ON(parent == 0); | |
0ef3e66b | 3462 | |
1a40e23b ZY |
3463 | key.objectid = bytenr; |
3464 | key.offset = parent + 1; | |
3465 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 3466 | |
1a40e23b ZY |
3467 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
3468 | if (ret < 0) | |
edbd8d4e | 3469 | goto out; |
1a40e23b | 3470 | BUG_ON(ret == 0); |
7d9eb12c | 3471 | |
1a40e23b ZY |
3472 | leaf = path->nodes[0]; |
3473 | nritems = btrfs_header_nritems(leaf); | |
3474 | if (path->slots[0] >= nritems) { | |
3475 | ret = btrfs_next_leaf(extent_root, path); | |
3476 | if (ret < 0) | |
3477 | goto out; | |
3478 | if (ret > 0) | |
3479 | goto next; | |
3480 | leaf = path->nodes[0]; | |
3481 | } | |
0ef3e66b | 3482 | |
1a40e23b ZY |
3483 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3484 | if (found_key.objectid == bytenr && | |
f82d02d9 YZ |
3485 | found_key.type == BTRFS_EXTENT_REF_KEY) { |
3486 | if (level < ref_path->shared_level) | |
3487 | ref_path->shared_level = level; | |
1a40e23b | 3488 | goto found; |
f82d02d9 | 3489 | } |
1a40e23b ZY |
3490 | next: |
3491 | level--; | |
3492 | btrfs_release_path(extent_root, path); | |
d899e052 | 3493 | cond_resched(); |
1a40e23b ZY |
3494 | } |
3495 | /* reached lowest level */ | |
3496 | ret = 1; | |
3497 | goto out; | |
3498 | walk_up: | |
3499 | level = ref_path->current_level; | |
3500 | while (level < BTRFS_MAX_LEVEL - 1) { | |
3501 | u64 ref_objectid; | |
3502 | if (level >= 0) { | |
3503 | bytenr = ref_path->nodes[level]; | |
3504 | } else { | |
3505 | bytenr = ref_path->extent_start; | |
3506 | } | |
3507 | BUG_ON(bytenr == 0); | |
edbd8d4e | 3508 | |
1a40e23b ZY |
3509 | key.objectid = bytenr; |
3510 | key.offset = 0; | |
3511 | key.type = BTRFS_EXTENT_REF_KEY; | |
edbd8d4e | 3512 | |
1a40e23b ZY |
3513 | ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0); |
3514 | if (ret < 0) | |
3515 | goto out; | |
edbd8d4e | 3516 | |
1a40e23b ZY |
3517 | leaf = path->nodes[0]; |
3518 | nritems = btrfs_header_nritems(leaf); | |
3519 | if (path->slots[0] >= nritems) { | |
3520 | ret = btrfs_next_leaf(extent_root, path); | |
3521 | if (ret < 0) | |
3522 | goto out; | |
3523 | if (ret > 0) { | |
3524 | /* the extent was freed by someone */ | |
3525 | if (ref_path->lowest_level == level) | |
3526 | goto out; | |
3527 | btrfs_release_path(extent_root, path); | |
3528 | goto walk_down; | |
3529 | } | |
3530 | leaf = path->nodes[0]; | |
3531 | } | |
edbd8d4e | 3532 | |
1a40e23b ZY |
3533 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3534 | if (found_key.objectid != bytenr || | |
3535 | found_key.type != BTRFS_EXTENT_REF_KEY) { | |
3536 | /* the extent was freed by someone */ | |
3537 | if (ref_path->lowest_level == level) { | |
3538 | ret = 1; | |
3539 | goto out; | |
3540 | } | |
3541 | btrfs_release_path(extent_root, path); | |
3542 | goto walk_down; | |
3543 | } | |
3544 | found: | |
3545 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
3546 | struct btrfs_extent_ref); | |
3547 | ref_objectid = btrfs_ref_objectid(leaf, ref); | |
3548 | if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
3549 | if (first_time) { | |
3550 | level = (int)ref_objectid; | |
3551 | BUG_ON(level >= BTRFS_MAX_LEVEL); | |
3552 | ref_path->lowest_level = level; | |
3553 | ref_path->current_level = level; | |
3554 | ref_path->nodes[level] = bytenr; | |
3555 | } else { | |
3556 | WARN_ON(ref_objectid != level); | |
3557 | } | |
3558 | } else { | |
3559 | WARN_ON(level != -1); | |
3560 | } | |
3561 | first_time = 0; | |
bf4ef679 | 3562 | |
1a40e23b ZY |
3563 | if (ref_path->lowest_level == level) { |
3564 | ref_path->owner_objectid = ref_objectid; | |
1a40e23b ZY |
3565 | ref_path->num_refs = btrfs_ref_num_refs(leaf, ref); |
3566 | } | |
bf4ef679 | 3567 | |
7d9eb12c | 3568 | /* |
1a40e23b ZY |
3569 | * the block is tree root or the block isn't in reference |
3570 | * counted tree. | |
7d9eb12c | 3571 | */ |
1a40e23b ZY |
3572 | if (found_key.objectid == found_key.offset || |
3573 | is_cowonly_root(btrfs_ref_root(leaf, ref))) { | |
3574 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
3575 | ref_path->root_generation = | |
3576 | btrfs_ref_generation(leaf, ref); | |
3577 | if (level < 0) { | |
3578 | /* special reference from the tree log */ | |
3579 | ref_path->nodes[0] = found_key.offset; | |
3580 | ref_path->current_level = 0; | |
3581 | } | |
3582 | ret = 0; | |
3583 | goto out; | |
3584 | } | |
7d9eb12c | 3585 | |
1a40e23b ZY |
3586 | level++; |
3587 | BUG_ON(ref_path->nodes[level] != 0); | |
3588 | ref_path->nodes[level] = found_key.offset; | |
3589 | ref_path->current_level = level; | |
bf4ef679 | 3590 | |
1a40e23b ZY |
3591 | /* |
3592 | * the reference was created in the running transaction, | |
3593 | * no need to continue walking up. | |
3594 | */ | |
3595 | if (btrfs_ref_generation(leaf, ref) == trans->transid) { | |
3596 | ref_path->root_objectid = btrfs_ref_root(leaf, ref); | |
3597 | ref_path->root_generation = | |
3598 | btrfs_ref_generation(leaf, ref); | |
3599 | ret = 0; | |
3600 | goto out; | |
7d9eb12c CM |
3601 | } |
3602 | ||
1a40e23b | 3603 | btrfs_release_path(extent_root, path); |
d899e052 | 3604 | cond_resched(); |
7d9eb12c | 3605 | } |
1a40e23b ZY |
3606 | /* reached max tree level, but no tree root found. */ |
3607 | BUG(); | |
edbd8d4e | 3608 | out: |
1a40e23b ZY |
3609 | btrfs_free_path(path); |
3610 | return ret; | |
edbd8d4e CM |
3611 | } |
3612 | ||
1a40e23b ZY |
3613 | static int btrfs_first_ref_path(struct btrfs_trans_handle *trans, |
3614 | struct btrfs_root *extent_root, | |
3615 | struct btrfs_ref_path *ref_path, | |
3616 | u64 extent_start) | |
a061fc8d | 3617 | { |
1a40e23b ZY |
3618 | memset(ref_path, 0, sizeof(*ref_path)); |
3619 | ref_path->extent_start = extent_start; | |
a061fc8d | 3620 | |
1a40e23b | 3621 | return __next_ref_path(trans, extent_root, ref_path, 1); |
a061fc8d CM |
3622 | } |
3623 | ||
1a40e23b ZY |
3624 | static int btrfs_next_ref_path(struct btrfs_trans_handle *trans, |
3625 | struct btrfs_root *extent_root, | |
3626 | struct btrfs_ref_path *ref_path) | |
edbd8d4e | 3627 | { |
1a40e23b ZY |
3628 | return __next_ref_path(trans, extent_root, ref_path, 0); |
3629 | } | |
3630 | ||
3631 | static int noinline get_new_locations(struct inode *reloc_inode, | |
3632 | struct btrfs_key *extent_key, | |
3633 | u64 offset, int no_fragment, | |
3634 | struct disk_extent **extents, | |
3635 | int *nr_extents) | |
3636 | { | |
3637 | struct btrfs_root *root = BTRFS_I(reloc_inode)->root; | |
3638 | struct btrfs_path *path; | |
3639 | struct btrfs_file_extent_item *fi; | |
edbd8d4e | 3640 | struct extent_buffer *leaf; |
1a40e23b ZY |
3641 | struct disk_extent *exts = *extents; |
3642 | struct btrfs_key found_key; | |
3643 | u64 cur_pos; | |
3644 | u64 last_byte; | |
edbd8d4e | 3645 | u32 nritems; |
1a40e23b ZY |
3646 | int nr = 0; |
3647 | int max = *nr_extents; | |
3648 | int ret; | |
edbd8d4e | 3649 | |
1a40e23b ZY |
3650 | WARN_ON(!no_fragment && *extents); |
3651 | if (!exts) { | |
3652 | max = 1; | |
3653 | exts = kmalloc(sizeof(*exts) * max, GFP_NOFS); | |
3654 | if (!exts) | |
3655 | return -ENOMEM; | |
a061fc8d | 3656 | } |
edbd8d4e | 3657 | |
1a40e23b ZY |
3658 | path = btrfs_alloc_path(); |
3659 | BUG_ON(!path); | |
edbd8d4e | 3660 | |
1a40e23b ZY |
3661 | cur_pos = extent_key->objectid - offset; |
3662 | last_byte = extent_key->objectid + extent_key->offset; | |
3663 | ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino, | |
3664 | cur_pos, 0); | |
3665 | if (ret < 0) | |
3666 | goto out; | |
3667 | if (ret > 0) { | |
3668 | ret = -ENOENT; | |
3669 | goto out; | |
3670 | } | |
edbd8d4e | 3671 | |
1a40e23b | 3672 | while (1) { |
edbd8d4e CM |
3673 | leaf = path->nodes[0]; |
3674 | nritems = btrfs_header_nritems(leaf); | |
1a40e23b ZY |
3675 | if (path->slots[0] >= nritems) { |
3676 | ret = btrfs_next_leaf(root, path); | |
a061fc8d CM |
3677 | if (ret < 0) |
3678 | goto out; | |
1a40e23b ZY |
3679 | if (ret > 0) |
3680 | break; | |
bf4ef679 | 3681 | leaf = path->nodes[0]; |
a061fc8d | 3682 | } |
edbd8d4e CM |
3683 | |
3684 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b ZY |
3685 | if (found_key.offset != cur_pos || |
3686 | found_key.type != BTRFS_EXTENT_DATA_KEY || | |
3687 | found_key.objectid != reloc_inode->i_ino) | |
edbd8d4e CM |
3688 | break; |
3689 | ||
1a40e23b ZY |
3690 | fi = btrfs_item_ptr(leaf, path->slots[0], |
3691 | struct btrfs_file_extent_item); | |
3692 | if (btrfs_file_extent_type(leaf, fi) != | |
3693 | BTRFS_FILE_EXTENT_REG || | |
3694 | btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
edbd8d4e | 3695 | break; |
1a40e23b ZY |
3696 | |
3697 | if (nr == max) { | |
3698 | struct disk_extent *old = exts; | |
3699 | max *= 2; | |
3700 | exts = kzalloc(sizeof(*exts) * max, GFP_NOFS); | |
3701 | memcpy(exts, old, sizeof(*exts) * nr); | |
3702 | if (old != *extents) | |
3703 | kfree(old); | |
a061fc8d | 3704 | } |
edbd8d4e | 3705 | |
1a40e23b ZY |
3706 | exts[nr].disk_bytenr = |
3707 | btrfs_file_extent_disk_bytenr(leaf, fi); | |
3708 | exts[nr].disk_num_bytes = | |
3709 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
3710 | exts[nr].offset = btrfs_file_extent_offset(leaf, fi); | |
3711 | exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
c8b97818 CM |
3712 | exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
3713 | exts[nr].compression = btrfs_file_extent_compression(leaf, fi); | |
3714 | exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi); | |
3715 | exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf, | |
3716 | fi); | |
d899e052 YZ |
3717 | BUG_ON(exts[nr].offset > 0); |
3718 | BUG_ON(exts[nr].compression || exts[nr].encryption); | |
3719 | BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes); | |
edbd8d4e | 3720 | |
1a40e23b ZY |
3721 | cur_pos += exts[nr].num_bytes; |
3722 | nr++; | |
3723 | ||
3724 | if (cur_pos + offset >= last_byte) | |
3725 | break; | |
3726 | ||
3727 | if (no_fragment) { | |
3728 | ret = 1; | |
edbd8d4e | 3729 | goto out; |
1a40e23b ZY |
3730 | } |
3731 | path->slots[0]++; | |
3732 | } | |
3733 | ||
3734 | WARN_ON(cur_pos + offset > last_byte); | |
3735 | if (cur_pos + offset < last_byte) { | |
3736 | ret = -ENOENT; | |
3737 | goto out; | |
edbd8d4e CM |
3738 | } |
3739 | ret = 0; | |
3740 | out: | |
1a40e23b ZY |
3741 | btrfs_free_path(path); |
3742 | if (ret) { | |
3743 | if (exts != *extents) | |
3744 | kfree(exts); | |
3745 | } else { | |
3746 | *extents = exts; | |
3747 | *nr_extents = nr; | |
3748 | } | |
3749 | return ret; | |
3750 | } | |
3751 | ||
3752 | static int noinline replace_one_extent(struct btrfs_trans_handle *trans, | |
3753 | struct btrfs_root *root, | |
3754 | struct btrfs_path *path, | |
3755 | struct btrfs_key *extent_key, | |
3756 | struct btrfs_key *leaf_key, | |
3757 | struct btrfs_ref_path *ref_path, | |
3758 | struct disk_extent *new_extents, | |
3759 | int nr_extents) | |
3760 | { | |
3761 | struct extent_buffer *leaf; | |
3762 | struct btrfs_file_extent_item *fi; | |
3763 | struct inode *inode = NULL; | |
3764 | struct btrfs_key key; | |
3765 | u64 lock_start = 0; | |
3766 | u64 lock_end = 0; | |
3767 | u64 num_bytes; | |
3768 | u64 ext_offset; | |
3769 | u64 first_pos; | |
3770 | u32 nritems; | |
3bb1a1bc | 3771 | int nr_scaned = 0; |
1a40e23b | 3772 | int extent_locked = 0; |
d899e052 | 3773 | int extent_type; |
1a40e23b ZY |
3774 | int ret; |
3775 | ||
3bb1a1bc YZ |
3776 | memcpy(&key, leaf_key, sizeof(key)); |
3777 | first_pos = INT_LIMIT(loff_t) - extent_key->offset; | |
1a40e23b | 3778 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { |
3bb1a1bc YZ |
3779 | if (key.objectid < ref_path->owner_objectid || |
3780 | (key.objectid == ref_path->owner_objectid && | |
3781 | key.type < BTRFS_EXTENT_DATA_KEY)) { | |
3782 | key.objectid = ref_path->owner_objectid; | |
3783 | key.type = BTRFS_EXTENT_DATA_KEY; | |
3784 | key.offset = 0; | |
3785 | } | |
1a40e23b ZY |
3786 | } |
3787 | ||
3788 | while (1) { | |
3789 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
3790 | if (ret < 0) | |
3791 | goto out; | |
3792 | ||
3793 | leaf = path->nodes[0]; | |
3794 | nritems = btrfs_header_nritems(leaf); | |
3795 | next: | |
3796 | if (extent_locked && ret > 0) { | |
3797 | /* | |
3798 | * the file extent item was modified by someone | |
3799 | * before the extent got locked. | |
3800 | */ | |
1a40e23b ZY |
3801 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
3802 | lock_end, GFP_NOFS); | |
3803 | extent_locked = 0; | |
3804 | } | |
3805 | ||
3806 | if (path->slots[0] >= nritems) { | |
3bb1a1bc | 3807 | if (++nr_scaned > 2) |
1a40e23b ZY |
3808 | break; |
3809 | ||
3810 | BUG_ON(extent_locked); | |
3811 | ret = btrfs_next_leaf(root, path); | |
3812 | if (ret < 0) | |
3813 | goto out; | |
3814 | if (ret > 0) | |
3815 | break; | |
3816 | leaf = path->nodes[0]; | |
3817 | nritems = btrfs_header_nritems(leaf); | |
3818 | } | |
3819 | ||
3820 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3821 | ||
3822 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { | |
3823 | if ((key.objectid > ref_path->owner_objectid) || | |
3824 | (key.objectid == ref_path->owner_objectid && | |
3825 | key.type > BTRFS_EXTENT_DATA_KEY) || | |
3826 | (key.offset >= first_pos + extent_key->offset)) | |
3827 | break; | |
3828 | } | |
3829 | ||
3830 | if (inode && key.objectid != inode->i_ino) { | |
3831 | BUG_ON(extent_locked); | |
3832 | btrfs_release_path(root, path); | |
3833 | mutex_unlock(&inode->i_mutex); | |
3834 | iput(inode); | |
3835 | inode = NULL; | |
3836 | continue; | |
3837 | } | |
3838 | ||
3839 | if (key.type != BTRFS_EXTENT_DATA_KEY) { | |
3840 | path->slots[0]++; | |
3841 | ret = 1; | |
3842 | goto next; | |
3843 | } | |
3844 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3845 | struct btrfs_file_extent_item); | |
d899e052 YZ |
3846 | extent_type = btrfs_file_extent_type(leaf, fi); |
3847 | if ((extent_type != BTRFS_FILE_EXTENT_REG && | |
3848 | extent_type != BTRFS_FILE_EXTENT_PREALLOC) || | |
1a40e23b ZY |
3849 | (btrfs_file_extent_disk_bytenr(leaf, fi) != |
3850 | extent_key->objectid)) { | |
3851 | path->slots[0]++; | |
3852 | ret = 1; | |
3853 | goto next; | |
3854 | } | |
3855 | ||
3856 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
3857 | ext_offset = btrfs_file_extent_offset(leaf, fi); | |
3858 | ||
3859 | if (first_pos > key.offset - ext_offset) | |
3860 | first_pos = key.offset - ext_offset; | |
3861 | ||
3862 | if (!extent_locked) { | |
3863 | lock_start = key.offset; | |
3864 | lock_end = lock_start + num_bytes - 1; | |
3865 | } else { | |
6643558d YZ |
3866 | if (lock_start > key.offset || |
3867 | lock_end + 1 < key.offset + num_bytes) { | |
3868 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
3869 | lock_start, lock_end, GFP_NOFS); | |
3870 | extent_locked = 0; | |
3871 | } | |
1a40e23b ZY |
3872 | } |
3873 | ||
3874 | if (!inode) { | |
3875 | btrfs_release_path(root, path); | |
3876 | ||
3877 | inode = btrfs_iget_locked(root->fs_info->sb, | |
3878 | key.objectid, root); | |
3879 | if (inode->i_state & I_NEW) { | |
3880 | BTRFS_I(inode)->root = root; | |
3881 | BTRFS_I(inode)->location.objectid = | |
3882 | key.objectid; | |
3883 | BTRFS_I(inode)->location.type = | |
3884 | BTRFS_INODE_ITEM_KEY; | |
3885 | BTRFS_I(inode)->location.offset = 0; | |
3886 | btrfs_read_locked_inode(inode); | |
3887 | unlock_new_inode(inode); | |
3888 | } | |
3889 | /* | |
3890 | * some code call btrfs_commit_transaction while | |
3891 | * holding the i_mutex, so we can't use mutex_lock | |
3892 | * here. | |
3893 | */ | |
3894 | if (is_bad_inode(inode) || | |
3895 | !mutex_trylock(&inode->i_mutex)) { | |
3896 | iput(inode); | |
3897 | inode = NULL; | |
3898 | key.offset = (u64)-1; | |
3899 | goto skip; | |
3900 | } | |
3901 | } | |
3902 | ||
3903 | if (!extent_locked) { | |
3904 | struct btrfs_ordered_extent *ordered; | |
3905 | ||
3906 | btrfs_release_path(root, path); | |
3907 | ||
3908 | lock_extent(&BTRFS_I(inode)->io_tree, lock_start, | |
3909 | lock_end, GFP_NOFS); | |
3910 | ordered = btrfs_lookup_first_ordered_extent(inode, | |
3911 | lock_end); | |
3912 | if (ordered && | |
3913 | ordered->file_offset <= lock_end && | |
3914 | ordered->file_offset + ordered->len > lock_start) { | |
3915 | unlock_extent(&BTRFS_I(inode)->io_tree, | |
3916 | lock_start, lock_end, GFP_NOFS); | |
3917 | btrfs_start_ordered_extent(inode, ordered, 1); | |
3918 | btrfs_put_ordered_extent(ordered); | |
3919 | key.offset += num_bytes; | |
3920 | goto skip; | |
3921 | } | |
3922 | if (ordered) | |
3923 | btrfs_put_ordered_extent(ordered); | |
3924 | ||
1a40e23b ZY |
3925 | extent_locked = 1; |
3926 | continue; | |
3927 | } | |
3928 | ||
3929 | if (nr_extents == 1) { | |
3930 | /* update extent pointer in place */ | |
1a40e23b ZY |
3931 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
3932 | new_extents[0].disk_bytenr); | |
3933 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
3934 | new_extents[0].disk_num_bytes); | |
1a40e23b ZY |
3935 | btrfs_mark_buffer_dirty(leaf); |
3936 | ||
3937 | btrfs_drop_extent_cache(inode, key.offset, | |
3938 | key.offset + num_bytes - 1, 0); | |
3939 | ||
3940 | ret = btrfs_inc_extent_ref(trans, root, | |
3941 | new_extents[0].disk_bytenr, | |
3942 | new_extents[0].disk_num_bytes, | |
3943 | leaf->start, | |
3944 | root->root_key.objectid, | |
3945 | trans->transid, | |
3bb1a1bc | 3946 | key.objectid); |
1a40e23b ZY |
3947 | BUG_ON(ret); |
3948 | ||
3949 | ret = btrfs_free_extent(trans, root, | |
3950 | extent_key->objectid, | |
3951 | extent_key->offset, | |
3952 | leaf->start, | |
3953 | btrfs_header_owner(leaf), | |
3954 | btrfs_header_generation(leaf), | |
3bb1a1bc | 3955 | key.objectid, 0); |
1a40e23b ZY |
3956 | BUG_ON(ret); |
3957 | ||
3958 | btrfs_release_path(root, path); | |
3959 | key.offset += num_bytes; | |
3960 | } else { | |
d899e052 YZ |
3961 | BUG_ON(1); |
3962 | #if 0 | |
1a40e23b ZY |
3963 | u64 alloc_hint; |
3964 | u64 extent_len; | |
3965 | int i; | |
3966 | /* | |
3967 | * drop old extent pointer at first, then insert the | |
3968 | * new pointers one bye one | |
3969 | */ | |
3970 | btrfs_release_path(root, path); | |
3971 | ret = btrfs_drop_extents(trans, root, inode, key.offset, | |
3972 | key.offset + num_bytes, | |
3973 | key.offset, &alloc_hint); | |
3974 | BUG_ON(ret); | |
3975 | ||
3976 | for (i = 0; i < nr_extents; i++) { | |
3977 | if (ext_offset >= new_extents[i].num_bytes) { | |
3978 | ext_offset -= new_extents[i].num_bytes; | |
3979 | continue; | |
3980 | } | |
3981 | extent_len = min(new_extents[i].num_bytes - | |
3982 | ext_offset, num_bytes); | |
3983 | ||
3984 | ret = btrfs_insert_empty_item(trans, root, | |
3985 | path, &key, | |
3986 | sizeof(*fi)); | |
3987 | BUG_ON(ret); | |
3988 | ||
3989 | leaf = path->nodes[0]; | |
3990 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3991 | struct btrfs_file_extent_item); | |
3992 | btrfs_set_file_extent_generation(leaf, fi, | |
3993 | trans->transid); | |
3994 | btrfs_set_file_extent_type(leaf, fi, | |
3995 | BTRFS_FILE_EXTENT_REG); | |
3996 | btrfs_set_file_extent_disk_bytenr(leaf, fi, | |
3997 | new_extents[i].disk_bytenr); | |
3998 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
3999 | new_extents[i].disk_num_bytes); | |
c8b97818 CM |
4000 | btrfs_set_file_extent_ram_bytes(leaf, fi, |
4001 | new_extents[i].ram_bytes); | |
4002 | ||
4003 | btrfs_set_file_extent_compression(leaf, fi, | |
4004 | new_extents[i].compression); | |
4005 | btrfs_set_file_extent_encryption(leaf, fi, | |
4006 | new_extents[i].encryption); | |
4007 | btrfs_set_file_extent_other_encoding(leaf, fi, | |
4008 | new_extents[i].other_encoding); | |
4009 | ||
1a40e23b ZY |
4010 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4011 | extent_len); | |
4012 | ext_offset += new_extents[i].offset; | |
4013 | btrfs_set_file_extent_offset(leaf, fi, | |
4014 | ext_offset); | |
4015 | btrfs_mark_buffer_dirty(leaf); | |
4016 | ||
4017 | btrfs_drop_extent_cache(inode, key.offset, | |
4018 | key.offset + extent_len - 1, 0); | |
4019 | ||
4020 | ret = btrfs_inc_extent_ref(trans, root, | |
4021 | new_extents[i].disk_bytenr, | |
4022 | new_extents[i].disk_num_bytes, | |
4023 | leaf->start, | |
4024 | root->root_key.objectid, | |
3bb1a1bc | 4025 | trans->transid, key.objectid); |
1a40e23b ZY |
4026 | BUG_ON(ret); |
4027 | btrfs_release_path(root, path); | |
4028 | ||
a76a3cd4 | 4029 | inode_add_bytes(inode, extent_len); |
1a40e23b ZY |
4030 | |
4031 | ext_offset = 0; | |
4032 | num_bytes -= extent_len; | |
4033 | key.offset += extent_len; | |
4034 | ||
4035 | if (num_bytes == 0) | |
4036 | break; | |
4037 | } | |
4038 | BUG_ON(i >= nr_extents); | |
d899e052 | 4039 | #endif |
1a40e23b ZY |
4040 | } |
4041 | ||
4042 | if (extent_locked) { | |
1a40e23b ZY |
4043 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
4044 | lock_end, GFP_NOFS); | |
4045 | extent_locked = 0; | |
4046 | } | |
4047 | skip: | |
4048 | if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS && | |
4049 | key.offset >= first_pos + extent_key->offset) | |
4050 | break; | |
4051 | ||
4052 | cond_resched(); | |
4053 | } | |
4054 | ret = 0; | |
4055 | out: | |
4056 | btrfs_release_path(root, path); | |
4057 | if (inode) { | |
4058 | mutex_unlock(&inode->i_mutex); | |
4059 | if (extent_locked) { | |
1a40e23b ZY |
4060 | unlock_extent(&BTRFS_I(inode)->io_tree, lock_start, |
4061 | lock_end, GFP_NOFS); | |
4062 | } | |
4063 | iput(inode); | |
4064 | } | |
4065 | return ret; | |
4066 | } | |
4067 | ||
1a40e23b ZY |
4068 | int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, |
4069 | struct btrfs_root *root, | |
4070 | struct extent_buffer *buf, u64 orig_start) | |
4071 | { | |
4072 | int level; | |
4073 | int ret; | |
4074 | ||
4075 | BUG_ON(btrfs_header_generation(buf) != trans->transid); | |
4076 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
4077 | ||
4078 | level = btrfs_header_level(buf); | |
4079 | if (level == 0) { | |
4080 | struct btrfs_leaf_ref *ref; | |
4081 | struct btrfs_leaf_ref *orig_ref; | |
4082 | ||
4083 | orig_ref = btrfs_lookup_leaf_ref(root, orig_start); | |
4084 | if (!orig_ref) | |
4085 | return -ENOENT; | |
4086 | ||
4087 | ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems); | |
4088 | if (!ref) { | |
4089 | btrfs_free_leaf_ref(root, orig_ref); | |
4090 | return -ENOMEM; | |
4091 | } | |
4092 | ||
4093 | ref->nritems = orig_ref->nritems; | |
4094 | memcpy(ref->extents, orig_ref->extents, | |
4095 | sizeof(ref->extents[0]) * ref->nritems); | |
4096 | ||
4097 | btrfs_free_leaf_ref(root, orig_ref); | |
4098 | ||
4099 | ref->root_gen = trans->transid; | |
4100 | ref->bytenr = buf->start; | |
4101 | ref->owner = btrfs_header_owner(buf); | |
4102 | ref->generation = btrfs_header_generation(buf); | |
4103 | ret = btrfs_add_leaf_ref(root, ref, 0); | |
4104 | WARN_ON(ret); | |
4105 | btrfs_free_leaf_ref(root, ref); | |
4106 | } | |
4107 | return 0; | |
4108 | } | |
4109 | ||
4110 | static int noinline invalidate_extent_cache(struct btrfs_root *root, | |
4111 | struct extent_buffer *leaf, | |
4112 | struct btrfs_block_group_cache *group, | |
4113 | struct btrfs_root *target_root) | |
4114 | { | |
4115 | struct btrfs_key key; | |
4116 | struct inode *inode = NULL; | |
4117 | struct btrfs_file_extent_item *fi; | |
4118 | u64 num_bytes; | |
4119 | u64 skip_objectid = 0; | |
4120 | u32 nritems; | |
4121 | u32 i; | |
4122 | ||
4123 | nritems = btrfs_header_nritems(leaf); | |
4124 | for (i = 0; i < nritems; i++) { | |
4125 | btrfs_item_key_to_cpu(leaf, &key, i); | |
4126 | if (key.objectid == skip_objectid || | |
4127 | key.type != BTRFS_EXTENT_DATA_KEY) | |
4128 | continue; | |
4129 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
4130 | if (btrfs_file_extent_type(leaf, fi) == | |
4131 | BTRFS_FILE_EXTENT_INLINE) | |
4132 | continue; | |
4133 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) | |
4134 | continue; | |
4135 | if (!inode || inode->i_ino != key.objectid) { | |
4136 | iput(inode); | |
4137 | inode = btrfs_ilookup(target_root->fs_info->sb, | |
4138 | key.objectid, target_root, 1); | |
4139 | } | |
4140 | if (!inode) { | |
4141 | skip_objectid = key.objectid; | |
4142 | continue; | |
4143 | } | |
4144 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi); | |
4145 | ||
4146 | lock_extent(&BTRFS_I(inode)->io_tree, key.offset, | |
4147 | key.offset + num_bytes - 1, GFP_NOFS); | |
1a40e23b ZY |
4148 | btrfs_drop_extent_cache(inode, key.offset, |
4149 | key.offset + num_bytes - 1, 1); | |
1a40e23b ZY |
4150 | unlock_extent(&BTRFS_I(inode)->io_tree, key.offset, |
4151 | key.offset + num_bytes - 1, GFP_NOFS); | |
4152 | cond_resched(); | |
4153 | } | |
4154 | iput(inode); | |
4155 | return 0; | |
4156 | } | |
4157 | ||
4158 | static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans, | |
4159 | struct btrfs_root *root, | |
4160 | struct extent_buffer *leaf, | |
4161 | struct btrfs_block_group_cache *group, | |
4162 | struct inode *reloc_inode) | |
4163 | { | |
4164 | struct btrfs_key key; | |
4165 | struct btrfs_key extent_key; | |
4166 | struct btrfs_file_extent_item *fi; | |
4167 | struct btrfs_leaf_ref *ref; | |
4168 | struct disk_extent *new_extent; | |
4169 | u64 bytenr; | |
4170 | u64 num_bytes; | |
4171 | u32 nritems; | |
4172 | u32 i; | |
4173 | int ext_index; | |
4174 | int nr_extent; | |
4175 | int ret; | |
4176 | ||
4177 | new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS); | |
4178 | BUG_ON(!new_extent); | |
4179 | ||
4180 | ref = btrfs_lookup_leaf_ref(root, leaf->start); | |
4181 | BUG_ON(!ref); | |
4182 | ||
4183 | ext_index = -1; | |
4184 | nritems = btrfs_header_nritems(leaf); | |
4185 | for (i = 0; i < nritems; i++) { | |
4186 | btrfs_item_key_to_cpu(leaf, &key, i); | |
4187 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
4188 | continue; | |
4189 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
4190 | if (btrfs_file_extent_type(leaf, fi) == | |
4191 | BTRFS_FILE_EXTENT_INLINE) | |
4192 | continue; | |
4193 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
4194 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
4195 | if (bytenr == 0) | |
4196 | continue; | |
4197 | ||
4198 | ext_index++; | |
4199 | if (bytenr >= group->key.objectid + group->key.offset || | |
4200 | bytenr + num_bytes <= group->key.objectid) | |
4201 | continue; | |
4202 | ||
4203 | extent_key.objectid = bytenr; | |
4204 | extent_key.offset = num_bytes; | |
4205 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
4206 | nr_extent = 1; | |
4207 | ret = get_new_locations(reloc_inode, &extent_key, | |
4208 | group->key.objectid, 1, | |
4209 | &new_extent, &nr_extent); | |
4210 | if (ret > 0) | |
4211 | continue; | |
4212 | BUG_ON(ret < 0); | |
4213 | ||
4214 | BUG_ON(ref->extents[ext_index].bytenr != bytenr); | |
4215 | BUG_ON(ref->extents[ext_index].num_bytes != num_bytes); | |
4216 | ref->extents[ext_index].bytenr = new_extent->disk_bytenr; | |
4217 | ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes; | |
4218 | ||
1a40e23b ZY |
4219 | btrfs_set_file_extent_disk_bytenr(leaf, fi, |
4220 | new_extent->disk_bytenr); | |
4221 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, | |
4222 | new_extent->disk_num_bytes); | |
1a40e23b ZY |
4223 | btrfs_mark_buffer_dirty(leaf); |
4224 | ||
4225 | ret = btrfs_inc_extent_ref(trans, root, | |
4226 | new_extent->disk_bytenr, | |
4227 | new_extent->disk_num_bytes, | |
4228 | leaf->start, | |
4229 | root->root_key.objectid, | |
3bb1a1bc | 4230 | trans->transid, key.objectid); |
1a40e23b ZY |
4231 | BUG_ON(ret); |
4232 | ret = btrfs_free_extent(trans, root, | |
4233 | bytenr, num_bytes, leaf->start, | |
4234 | btrfs_header_owner(leaf), | |
4235 | btrfs_header_generation(leaf), | |
3bb1a1bc | 4236 | key.objectid, 0); |
1a40e23b ZY |
4237 | BUG_ON(ret); |
4238 | cond_resched(); | |
4239 | } | |
4240 | kfree(new_extent); | |
4241 | BUG_ON(ext_index + 1 != ref->nritems); | |
4242 | btrfs_free_leaf_ref(root, ref); | |
4243 | return 0; | |
4244 | } | |
4245 | ||
f82d02d9 YZ |
4246 | int btrfs_free_reloc_root(struct btrfs_trans_handle *trans, |
4247 | struct btrfs_root *root) | |
1a40e23b ZY |
4248 | { |
4249 | struct btrfs_root *reloc_root; | |
f82d02d9 | 4250 | int ret; |
1a40e23b ZY |
4251 | |
4252 | if (root->reloc_root) { | |
4253 | reloc_root = root->reloc_root; | |
4254 | root->reloc_root = NULL; | |
4255 | list_add(&reloc_root->dead_list, | |
4256 | &root->fs_info->dead_reloc_roots); | |
f82d02d9 YZ |
4257 | |
4258 | btrfs_set_root_bytenr(&reloc_root->root_item, | |
4259 | reloc_root->node->start); | |
4260 | btrfs_set_root_level(&root->root_item, | |
4261 | btrfs_header_level(reloc_root->node)); | |
4262 | memset(&reloc_root->root_item.drop_progress, 0, | |
4263 | sizeof(struct btrfs_disk_key)); | |
4264 | reloc_root->root_item.drop_level = 0; | |
4265 | ||
4266 | ret = btrfs_update_root(trans, root->fs_info->tree_root, | |
4267 | &reloc_root->root_key, | |
4268 | &reloc_root->root_item); | |
4269 | BUG_ON(ret); | |
1a40e23b ZY |
4270 | } |
4271 | return 0; | |
4272 | } | |
4273 | ||
4274 | int btrfs_drop_dead_reloc_roots(struct btrfs_root *root) | |
4275 | { | |
4276 | struct btrfs_trans_handle *trans; | |
4277 | struct btrfs_root *reloc_root; | |
4278 | struct btrfs_root *prev_root = NULL; | |
4279 | struct list_head dead_roots; | |
4280 | int ret; | |
4281 | unsigned long nr; | |
4282 | ||
4283 | INIT_LIST_HEAD(&dead_roots); | |
4284 | list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots); | |
4285 | ||
4286 | while (!list_empty(&dead_roots)) { | |
4287 | reloc_root = list_entry(dead_roots.prev, | |
4288 | struct btrfs_root, dead_list); | |
4289 | list_del_init(&reloc_root->dead_list); | |
4290 | ||
4291 | BUG_ON(reloc_root->commit_root != NULL); | |
4292 | while (1) { | |
4293 | trans = btrfs_join_transaction(root, 1); | |
4294 | BUG_ON(!trans); | |
4295 | ||
4296 | mutex_lock(&root->fs_info->drop_mutex); | |
4297 | ret = btrfs_drop_snapshot(trans, reloc_root); | |
4298 | if (ret != -EAGAIN) | |
4299 | break; | |
4300 | mutex_unlock(&root->fs_info->drop_mutex); | |
4301 | ||
4302 | nr = trans->blocks_used; | |
4303 | ret = btrfs_end_transaction(trans, root); | |
4304 | BUG_ON(ret); | |
4305 | btrfs_btree_balance_dirty(root, nr); | |
4306 | } | |
4307 | ||
4308 | free_extent_buffer(reloc_root->node); | |
4309 | ||
4310 | ret = btrfs_del_root(trans, root->fs_info->tree_root, | |
4311 | &reloc_root->root_key); | |
4312 | BUG_ON(ret); | |
4313 | mutex_unlock(&root->fs_info->drop_mutex); | |
4314 | ||
4315 | nr = trans->blocks_used; | |
4316 | ret = btrfs_end_transaction(trans, root); | |
4317 | BUG_ON(ret); | |
4318 | btrfs_btree_balance_dirty(root, nr); | |
4319 | ||
4320 | kfree(prev_root); | |
4321 | prev_root = reloc_root; | |
4322 | } | |
4323 | if (prev_root) { | |
4324 | btrfs_remove_leaf_refs(prev_root, (u64)-1, 0); | |
4325 | kfree(prev_root); | |
4326 | } | |
4327 | return 0; | |
4328 | } | |
4329 | ||
4330 | int btrfs_add_dead_reloc_root(struct btrfs_root *root) | |
4331 | { | |
4332 | list_add(&root->dead_list, &root->fs_info->dead_reloc_roots); | |
4333 | return 0; | |
4334 | } | |
4335 | ||
4336 | int btrfs_cleanup_reloc_trees(struct btrfs_root *root) | |
4337 | { | |
4338 | struct btrfs_root *reloc_root; | |
4339 | struct btrfs_trans_handle *trans; | |
4340 | struct btrfs_key location; | |
4341 | int found; | |
4342 | int ret; | |
4343 | ||
4344 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
4345 | ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL); | |
4346 | BUG_ON(ret); | |
4347 | found = !list_empty(&root->fs_info->dead_reloc_roots); | |
4348 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
4349 | ||
4350 | if (found) { | |
4351 | trans = btrfs_start_transaction(root, 1); | |
4352 | BUG_ON(!trans); | |
4353 | ret = btrfs_commit_transaction(trans, root); | |
4354 | BUG_ON(ret); | |
4355 | } | |
4356 | ||
4357 | location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
4358 | location.offset = (u64)-1; | |
4359 | location.type = BTRFS_ROOT_ITEM_KEY; | |
4360 | ||
4361 | reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location); | |
4362 | BUG_ON(!reloc_root); | |
4363 | btrfs_orphan_cleanup(reloc_root); | |
4364 | return 0; | |
4365 | } | |
4366 | ||
4367 | static int noinline init_reloc_tree(struct btrfs_trans_handle *trans, | |
4368 | struct btrfs_root *root) | |
4369 | { | |
4370 | struct btrfs_root *reloc_root; | |
4371 | struct extent_buffer *eb; | |
4372 | struct btrfs_root_item *root_item; | |
4373 | struct btrfs_key root_key; | |
4374 | int ret; | |
4375 | ||
4376 | BUG_ON(!root->ref_cows); | |
4377 | if (root->reloc_root) | |
4378 | return 0; | |
4379 | ||
4380 | root_item = kmalloc(sizeof(*root_item), GFP_NOFS); | |
4381 | BUG_ON(!root_item); | |
4382 | ||
4383 | ret = btrfs_copy_root(trans, root, root->commit_root, | |
4384 | &eb, BTRFS_TREE_RELOC_OBJECTID); | |
4385 | BUG_ON(ret); | |
4386 | ||
4387 | root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; | |
4388 | root_key.offset = root->root_key.objectid; | |
4389 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
4390 | ||
4391 | memcpy(root_item, &root->root_item, sizeof(root_item)); | |
4392 | btrfs_set_root_refs(root_item, 0); | |
4393 | btrfs_set_root_bytenr(root_item, eb->start); | |
4394 | btrfs_set_root_level(root_item, btrfs_header_level(eb)); | |
84234f3a | 4395 | btrfs_set_root_generation(root_item, trans->transid); |
1a40e23b ZY |
4396 | |
4397 | btrfs_tree_unlock(eb); | |
4398 | free_extent_buffer(eb); | |
4399 | ||
4400 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, | |
4401 | &root_key, root_item); | |
4402 | BUG_ON(ret); | |
4403 | kfree(root_item); | |
4404 | ||
4405 | reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root, | |
4406 | &root_key); | |
4407 | BUG_ON(!reloc_root); | |
4408 | reloc_root->last_trans = trans->transid; | |
4409 | reloc_root->commit_root = NULL; | |
4410 | reloc_root->ref_tree = &root->fs_info->reloc_ref_tree; | |
4411 | ||
4412 | root->reloc_root = reloc_root; | |
4413 | return 0; | |
4414 | } | |
4415 | ||
4416 | /* | |
4417 | * Core function of space balance. | |
4418 | * | |
4419 | * The idea is using reloc trees to relocate tree blocks in reference | |
f82d02d9 YZ |
4420 | * counted roots. There is one reloc tree for each subvol, and all |
4421 | * reloc trees share same root key objectid. Reloc trees are snapshots | |
4422 | * of the latest committed roots of subvols (root->commit_root). | |
4423 | * | |
4424 | * To relocate a tree block referenced by a subvol, there are two steps. | |
4425 | * COW the block through subvol's reloc tree, then update block pointer | |
4426 | * in the subvol to point to the new block. Since all reloc trees share | |
4427 | * same root key objectid, doing special handing for tree blocks owned | |
4428 | * by them is easy. Once a tree block has been COWed in one reloc tree, | |
4429 | * we can use the resulting new block directly when the same block is | |
4430 | * required to COW again through other reloc trees. By this way, relocated | |
4431 | * tree blocks are shared between reloc trees, so they are also shared | |
4432 | * between subvols. | |
1a40e23b ZY |
4433 | */ |
4434 | static int noinline relocate_one_path(struct btrfs_trans_handle *trans, | |
4435 | struct btrfs_root *root, | |
4436 | struct btrfs_path *path, | |
4437 | struct btrfs_key *first_key, | |
4438 | struct btrfs_ref_path *ref_path, | |
4439 | struct btrfs_block_group_cache *group, | |
4440 | struct inode *reloc_inode) | |
4441 | { | |
4442 | struct btrfs_root *reloc_root; | |
4443 | struct extent_buffer *eb = NULL; | |
4444 | struct btrfs_key *keys; | |
4445 | u64 *nodes; | |
4446 | int level; | |
f82d02d9 | 4447 | int shared_level; |
1a40e23b | 4448 | int lowest_level = 0; |
1a40e23b ZY |
4449 | int ret; |
4450 | ||
4451 | if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) | |
4452 | lowest_level = ref_path->owner_objectid; | |
4453 | ||
f82d02d9 | 4454 | if (!root->ref_cows) { |
1a40e23b ZY |
4455 | path->lowest_level = lowest_level; |
4456 | ret = btrfs_search_slot(trans, root, first_key, path, 0, 1); | |
4457 | BUG_ON(ret < 0); | |
4458 | path->lowest_level = 0; | |
4459 | btrfs_release_path(root, path); | |
4460 | return 0; | |
4461 | } | |
4462 | ||
1a40e23b ZY |
4463 | mutex_lock(&root->fs_info->tree_reloc_mutex); |
4464 | ret = init_reloc_tree(trans, root); | |
4465 | BUG_ON(ret); | |
4466 | reloc_root = root->reloc_root; | |
4467 | ||
f82d02d9 YZ |
4468 | shared_level = ref_path->shared_level; |
4469 | ref_path->shared_level = BTRFS_MAX_LEVEL - 1; | |
1a40e23b | 4470 | |
f82d02d9 YZ |
4471 | keys = ref_path->node_keys; |
4472 | nodes = ref_path->new_nodes; | |
4473 | memset(&keys[shared_level + 1], 0, | |
4474 | sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
4475 | memset(&nodes[shared_level + 1], 0, | |
4476 | sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1)); | |
1a40e23b | 4477 | |
f82d02d9 YZ |
4478 | if (nodes[lowest_level] == 0) { |
4479 | path->lowest_level = lowest_level; | |
4480 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, | |
4481 | 0, 1); | |
4482 | BUG_ON(ret); | |
4483 | for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) { | |
4484 | eb = path->nodes[level]; | |
4485 | if (!eb || eb == reloc_root->node) | |
4486 | break; | |
4487 | nodes[level] = eb->start; | |
4488 | if (level == 0) | |
4489 | btrfs_item_key_to_cpu(eb, &keys[level], 0); | |
4490 | else | |
4491 | btrfs_node_key_to_cpu(eb, &keys[level], 0); | |
4492 | } | |
4493 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
4494 | eb = path->nodes[0]; | |
4495 | ret = replace_extents_in_leaf(trans, reloc_root, eb, | |
4496 | group, reloc_inode); | |
4497 | BUG_ON(ret); | |
4498 | } | |
4499 | btrfs_release_path(reloc_root, path); | |
4500 | } else { | |
1a40e23b | 4501 | ret = btrfs_merge_path(trans, reloc_root, keys, nodes, |
f82d02d9 | 4502 | lowest_level); |
1a40e23b ZY |
4503 | BUG_ON(ret); |
4504 | } | |
4505 | ||
1a40e23b ZY |
4506 | /* |
4507 | * replace tree blocks in the fs tree with tree blocks in | |
4508 | * the reloc tree. | |
4509 | */ | |
4510 | ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level); | |
4511 | BUG_ON(ret < 0); | |
4512 | ||
4513 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
f82d02d9 YZ |
4514 | ret = btrfs_search_slot(trans, reloc_root, first_key, path, |
4515 | 0, 0); | |
4516 | BUG_ON(ret); | |
4517 | extent_buffer_get(path->nodes[0]); | |
4518 | eb = path->nodes[0]; | |
4519 | btrfs_release_path(reloc_root, path); | |
1a40e23b ZY |
4520 | ret = invalidate_extent_cache(reloc_root, eb, group, root); |
4521 | BUG_ON(ret); | |
4522 | free_extent_buffer(eb); | |
4523 | } | |
1a40e23b | 4524 | |
f82d02d9 | 4525 | mutex_unlock(&root->fs_info->tree_reloc_mutex); |
1a40e23b | 4526 | path->lowest_level = 0; |
1a40e23b ZY |
4527 | return 0; |
4528 | } | |
4529 | ||
4530 | static int noinline relocate_tree_block(struct btrfs_trans_handle *trans, | |
4531 | struct btrfs_root *root, | |
4532 | struct btrfs_path *path, | |
4533 | struct btrfs_key *first_key, | |
4534 | struct btrfs_ref_path *ref_path) | |
4535 | { | |
4536 | int ret; | |
1a40e23b ZY |
4537 | |
4538 | ret = relocate_one_path(trans, root, path, first_key, | |
4539 | ref_path, NULL, NULL); | |
4540 | BUG_ON(ret); | |
4541 | ||
4542 | if (root == root->fs_info->extent_root) | |
4543 | btrfs_extent_post_op(trans, root); | |
1a40e23b ZY |
4544 | |
4545 | return 0; | |
4546 | } | |
4547 | ||
4548 | static int noinline del_extent_zero(struct btrfs_trans_handle *trans, | |
4549 | struct btrfs_root *extent_root, | |
4550 | struct btrfs_path *path, | |
4551 | struct btrfs_key *extent_key) | |
4552 | { | |
4553 | int ret; | |
4554 | ||
1a40e23b ZY |
4555 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); |
4556 | if (ret) | |
4557 | goto out; | |
4558 | ret = btrfs_del_item(trans, extent_root, path); | |
4559 | out: | |
4560 | btrfs_release_path(extent_root, path); | |
1a40e23b ZY |
4561 | return ret; |
4562 | } | |
4563 | ||
4564 | static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info, | |
4565 | struct btrfs_ref_path *ref_path) | |
4566 | { | |
4567 | struct btrfs_key root_key; | |
4568 | ||
4569 | root_key.objectid = ref_path->root_objectid; | |
4570 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
4571 | if (is_cowonly_root(ref_path->root_objectid)) | |
4572 | root_key.offset = 0; | |
4573 | else | |
4574 | root_key.offset = (u64)-1; | |
4575 | ||
4576 | return btrfs_read_fs_root_no_name(fs_info, &root_key); | |
4577 | } | |
4578 | ||
4579 | static int noinline relocate_one_extent(struct btrfs_root *extent_root, | |
4580 | struct btrfs_path *path, | |
4581 | struct btrfs_key *extent_key, | |
4582 | struct btrfs_block_group_cache *group, | |
4583 | struct inode *reloc_inode, int pass) | |
4584 | { | |
4585 | struct btrfs_trans_handle *trans; | |
4586 | struct btrfs_root *found_root; | |
4587 | struct btrfs_ref_path *ref_path = NULL; | |
4588 | struct disk_extent *new_extents = NULL; | |
4589 | int nr_extents = 0; | |
4590 | int loops; | |
4591 | int ret; | |
4592 | int level; | |
4593 | struct btrfs_key first_key; | |
4594 | u64 prev_block = 0; | |
4595 | ||
1a40e23b ZY |
4596 | |
4597 | trans = btrfs_start_transaction(extent_root, 1); | |
4598 | BUG_ON(!trans); | |
4599 | ||
4600 | if (extent_key->objectid == 0) { | |
4601 | ret = del_extent_zero(trans, extent_root, path, extent_key); | |
4602 | goto out; | |
4603 | } | |
4604 | ||
4605 | ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS); | |
4606 | if (!ref_path) { | |
4607 | ret = -ENOMEM; | |
4608 | goto out; | |
4609 | } | |
4610 | ||
4611 | for (loops = 0; ; loops++) { | |
4612 | if (loops == 0) { | |
4613 | ret = btrfs_first_ref_path(trans, extent_root, ref_path, | |
4614 | extent_key->objectid); | |
4615 | } else { | |
4616 | ret = btrfs_next_ref_path(trans, extent_root, ref_path); | |
4617 | } | |
4618 | if (ret < 0) | |
4619 | goto out; | |
4620 | if (ret > 0) | |
4621 | break; | |
4622 | ||
4623 | if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID || | |
4624 | ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
4625 | continue; | |
4626 | ||
4627 | found_root = read_ref_root(extent_root->fs_info, ref_path); | |
4628 | BUG_ON(!found_root); | |
4629 | /* | |
4630 | * for reference counted tree, only process reference paths | |
4631 | * rooted at the latest committed root. | |
4632 | */ | |
4633 | if (found_root->ref_cows && | |
4634 | ref_path->root_generation != found_root->root_key.offset) | |
4635 | continue; | |
4636 | ||
4637 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
4638 | if (pass == 0) { | |
4639 | /* | |
4640 | * copy data extents to new locations | |
4641 | */ | |
4642 | u64 group_start = group->key.objectid; | |
4643 | ret = relocate_data_extent(reloc_inode, | |
4644 | extent_key, | |
4645 | group_start); | |
4646 | if (ret < 0) | |
4647 | goto out; | |
4648 | break; | |
4649 | } | |
4650 | level = 0; | |
4651 | } else { | |
4652 | level = ref_path->owner_objectid; | |
4653 | } | |
4654 | ||
4655 | if (prev_block != ref_path->nodes[level]) { | |
4656 | struct extent_buffer *eb; | |
4657 | u64 block_start = ref_path->nodes[level]; | |
4658 | u64 block_size = btrfs_level_size(found_root, level); | |
4659 | ||
4660 | eb = read_tree_block(found_root, block_start, | |
4661 | block_size, 0); | |
4662 | btrfs_tree_lock(eb); | |
4663 | BUG_ON(level != btrfs_header_level(eb)); | |
4664 | ||
4665 | if (level == 0) | |
4666 | btrfs_item_key_to_cpu(eb, &first_key, 0); | |
4667 | else | |
4668 | btrfs_node_key_to_cpu(eb, &first_key, 0); | |
4669 | ||
4670 | btrfs_tree_unlock(eb); | |
4671 | free_extent_buffer(eb); | |
4672 | prev_block = block_start; | |
4673 | } | |
4674 | ||
4675 | if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID && | |
4676 | pass >= 2) { | |
4677 | /* | |
4678 | * use fallback method to process the remaining | |
4679 | * references. | |
4680 | */ | |
4681 | if (!new_extents) { | |
4682 | u64 group_start = group->key.objectid; | |
d899e052 YZ |
4683 | new_extents = kmalloc(sizeof(*new_extents), |
4684 | GFP_NOFS); | |
4685 | nr_extents = 1; | |
1a40e23b ZY |
4686 | ret = get_new_locations(reloc_inode, |
4687 | extent_key, | |
d899e052 | 4688 | group_start, 1, |
1a40e23b ZY |
4689 | &new_extents, |
4690 | &nr_extents); | |
d899e052 | 4691 | if (ret) |
1a40e23b ZY |
4692 | goto out; |
4693 | } | |
4694 | btrfs_record_root_in_trans(found_root); | |
4695 | ret = replace_one_extent(trans, found_root, | |
4696 | path, extent_key, | |
4697 | &first_key, ref_path, | |
4698 | new_extents, nr_extents); | |
4699 | if (ret < 0) | |
4700 | goto out; | |
4701 | continue; | |
4702 | } | |
4703 | ||
4704 | btrfs_record_root_in_trans(found_root); | |
4705 | if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { | |
4706 | ret = relocate_tree_block(trans, found_root, path, | |
4707 | &first_key, ref_path); | |
4708 | } else { | |
4709 | /* | |
4710 | * try to update data extent references while | |
4711 | * keeping metadata shared between snapshots. | |
4712 | */ | |
4713 | ret = relocate_one_path(trans, found_root, path, | |
4714 | &first_key, ref_path, | |
4715 | group, reloc_inode); | |
4716 | } | |
4717 | if (ret < 0) | |
4718 | goto out; | |
4719 | } | |
4720 | ret = 0; | |
4721 | out: | |
4722 | btrfs_end_transaction(trans, extent_root); | |
4723 | kfree(new_extents); | |
4724 | kfree(ref_path); | |
1a40e23b ZY |
4725 | return ret; |
4726 | } | |
4727 | ||
ec44a35c CM |
4728 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
4729 | { | |
4730 | u64 num_devices; | |
4731 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
4732 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
4733 | ||
a061fc8d | 4734 | num_devices = root->fs_info->fs_devices->num_devices; |
ec44a35c CM |
4735 | if (num_devices == 1) { |
4736 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
4737 | stripped = flags & ~stripped; | |
4738 | ||
4739 | /* turn raid0 into single device chunks */ | |
4740 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
4741 | return stripped; | |
4742 | ||
4743 | /* turn mirroring into duplication */ | |
4744 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
4745 | BTRFS_BLOCK_GROUP_RAID10)) | |
4746 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
4747 | return flags; | |
4748 | } else { | |
4749 | /* they already had raid on here, just return */ | |
ec44a35c CM |
4750 | if (flags & stripped) |
4751 | return flags; | |
4752 | ||
4753 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
4754 | stripped = flags & ~stripped; | |
4755 | ||
4756 | /* switch duplicated blocks with raid1 */ | |
4757 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
4758 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
4759 | ||
4760 | /* turn single device chunks into raid0 */ | |
4761 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
4762 | } | |
4763 | return flags; | |
4764 | } | |
4765 | ||
0ef3e66b CM |
4766 | int __alloc_chunk_for_shrink(struct btrfs_root *root, |
4767 | struct btrfs_block_group_cache *shrink_block_group, | |
4768 | int force) | |
4769 | { | |
4770 | struct btrfs_trans_handle *trans; | |
4771 | u64 new_alloc_flags; | |
4772 | u64 calc; | |
4773 | ||
c286ac48 | 4774 | spin_lock(&shrink_block_group->lock); |
0ef3e66b | 4775 | if (btrfs_block_group_used(&shrink_block_group->item) > 0) { |
c286ac48 | 4776 | spin_unlock(&shrink_block_group->lock); |
c286ac48 | 4777 | |
0ef3e66b | 4778 | trans = btrfs_start_transaction(root, 1); |
c286ac48 | 4779 | spin_lock(&shrink_block_group->lock); |
7d9eb12c | 4780 | |
0ef3e66b CM |
4781 | new_alloc_flags = update_block_group_flags(root, |
4782 | shrink_block_group->flags); | |
4783 | if (new_alloc_flags != shrink_block_group->flags) { | |
4784 | calc = | |
4785 | btrfs_block_group_used(&shrink_block_group->item); | |
4786 | } else { | |
4787 | calc = shrink_block_group->key.offset; | |
4788 | } | |
c286ac48 CM |
4789 | spin_unlock(&shrink_block_group->lock); |
4790 | ||
0ef3e66b CM |
4791 | do_chunk_alloc(trans, root->fs_info->extent_root, |
4792 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
7d9eb12c | 4793 | |
0ef3e66b | 4794 | btrfs_end_transaction(trans, root); |
c286ac48 CM |
4795 | } else |
4796 | spin_unlock(&shrink_block_group->lock); | |
0ef3e66b CM |
4797 | return 0; |
4798 | } | |
4799 | ||
1a40e23b ZY |
4800 | static int __insert_orphan_inode(struct btrfs_trans_handle *trans, |
4801 | struct btrfs_root *root, | |
4802 | u64 objectid, u64 size) | |
4803 | { | |
4804 | struct btrfs_path *path; | |
4805 | struct btrfs_inode_item *item; | |
4806 | struct extent_buffer *leaf; | |
4807 | int ret; | |
4808 | ||
4809 | path = btrfs_alloc_path(); | |
4810 | if (!path) | |
4811 | return -ENOMEM; | |
4812 | ||
4813 | ret = btrfs_insert_empty_inode(trans, root, path, objectid); | |
4814 | if (ret) | |
4815 | goto out; | |
4816 | ||
4817 | leaf = path->nodes[0]; | |
4818 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); | |
4819 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
4820 | btrfs_set_inode_generation(leaf, item, 1); | |
4821 | btrfs_set_inode_size(leaf, item, size); | |
4822 | btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); | |
d899e052 YZ |
4823 | btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM | |
4824 | BTRFS_INODE_NOCOMPRESS); | |
1a40e23b ZY |
4825 | btrfs_mark_buffer_dirty(leaf); |
4826 | btrfs_release_path(root, path); | |
4827 | out: | |
4828 | btrfs_free_path(path); | |
4829 | return ret; | |
4830 | } | |
4831 | ||
4832 | static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info, | |
4833 | struct btrfs_block_group_cache *group) | |
4834 | { | |
4835 | struct inode *inode = NULL; | |
4836 | struct btrfs_trans_handle *trans; | |
4837 | struct btrfs_root *root; | |
4838 | struct btrfs_key root_key; | |
4839 | u64 objectid = BTRFS_FIRST_FREE_OBJECTID; | |
4840 | int err = 0; | |
4841 | ||
4842 | root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID; | |
4843 | root_key.type = BTRFS_ROOT_ITEM_KEY; | |
4844 | root_key.offset = (u64)-1; | |
4845 | root = btrfs_read_fs_root_no_name(fs_info, &root_key); | |
4846 | if (IS_ERR(root)) | |
4847 | return ERR_CAST(root); | |
4848 | ||
4849 | trans = btrfs_start_transaction(root, 1); | |
4850 | BUG_ON(!trans); | |
4851 | ||
4852 | err = btrfs_find_free_objectid(trans, root, objectid, &objectid); | |
4853 | if (err) | |
4854 | goto out; | |
4855 | ||
4856 | err = __insert_orphan_inode(trans, root, objectid, group->key.offset); | |
4857 | BUG_ON(err); | |
4858 | ||
4859 | err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0, | |
c8b97818 CM |
4860 | group->key.offset, 0, group->key.offset, |
4861 | 0, 0, 0); | |
1a40e23b ZY |
4862 | BUG_ON(err); |
4863 | ||
4864 | inode = btrfs_iget_locked(root->fs_info->sb, objectid, root); | |
4865 | if (inode->i_state & I_NEW) { | |
4866 | BTRFS_I(inode)->root = root; | |
4867 | BTRFS_I(inode)->location.objectid = objectid; | |
4868 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
4869 | BTRFS_I(inode)->location.offset = 0; | |
4870 | btrfs_read_locked_inode(inode); | |
4871 | unlock_new_inode(inode); | |
4872 | BUG_ON(is_bad_inode(inode)); | |
4873 | } else { | |
4874 | BUG_ON(1); | |
4875 | } | |
4876 | ||
4877 | err = btrfs_orphan_add(trans, inode); | |
4878 | out: | |
4879 | btrfs_end_transaction(trans, root); | |
4880 | if (err) { | |
4881 | if (inode) | |
4882 | iput(inode); | |
4883 | inode = ERR_PTR(err); | |
4884 | } | |
4885 | return inode; | |
4886 | } | |
4887 | ||
4888 | int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) | |
edbd8d4e CM |
4889 | { |
4890 | struct btrfs_trans_handle *trans; | |
edbd8d4e | 4891 | struct btrfs_path *path; |
1a40e23b ZY |
4892 | struct btrfs_fs_info *info = root->fs_info; |
4893 | struct extent_buffer *leaf; | |
4894 | struct inode *reloc_inode; | |
4895 | struct btrfs_block_group_cache *block_group; | |
4896 | struct btrfs_key key; | |
d899e052 | 4897 | u64 skipped; |
edbd8d4e CM |
4898 | u64 cur_byte; |
4899 | u64 total_found; | |
edbd8d4e CM |
4900 | u32 nritems; |
4901 | int ret; | |
a061fc8d | 4902 | int progress; |
1a40e23b | 4903 | int pass = 0; |
edbd8d4e | 4904 | |
1a40e23b ZY |
4905 | root = root->fs_info->extent_root; |
4906 | ||
4907 | block_group = btrfs_lookup_block_group(info, group_start); | |
4908 | BUG_ON(!block_group); | |
8f18cf13 | 4909 | |
1a40e23b ZY |
4910 | printk("btrfs relocating block group %llu flags %llu\n", |
4911 | (unsigned long long)block_group->key.objectid, | |
4912 | (unsigned long long)block_group->flags); | |
8f18cf13 | 4913 | |
edbd8d4e | 4914 | path = btrfs_alloc_path(); |
1a40e23b | 4915 | BUG_ON(!path); |
edbd8d4e | 4916 | |
1a40e23b ZY |
4917 | reloc_inode = create_reloc_inode(info, block_group); |
4918 | BUG_ON(IS_ERR(reloc_inode)); | |
323da79c | 4919 | |
1a40e23b ZY |
4920 | __alloc_chunk_for_shrink(root, block_group, 1); |
4921 | block_group->ro = 1; | |
4922 | block_group->space_info->total_bytes -= block_group->key.offset; | |
323da79c | 4923 | |
1a40e23b ZY |
4924 | btrfs_start_delalloc_inodes(info->tree_root); |
4925 | btrfs_wait_ordered_extents(info->tree_root, 0); | |
4926 | again: | |
d899e052 | 4927 | skipped = 0; |
edbd8d4e | 4928 | total_found = 0; |
a061fc8d | 4929 | progress = 0; |
1a40e23b | 4930 | key.objectid = block_group->key.objectid; |
edbd8d4e CM |
4931 | key.offset = 0; |
4932 | key.type = 0; | |
73e48b27 | 4933 | cur_byte = key.objectid; |
4313b399 | 4934 | |
1a40e23b ZY |
4935 | trans = btrfs_start_transaction(info->tree_root, 1); |
4936 | btrfs_commit_transaction(trans, info->tree_root); | |
ea8c2819 | 4937 | |
1a40e23b ZY |
4938 | mutex_lock(&root->fs_info->cleaner_mutex); |
4939 | btrfs_clean_old_snapshots(info->tree_root); | |
4940 | btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); | |
4941 | mutex_unlock(&root->fs_info->cleaner_mutex); | |
ea8c2819 | 4942 | |
73e48b27 | 4943 | while(1) { |
edbd8d4e CM |
4944 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4945 | if (ret < 0) | |
4946 | goto out; | |
7d9eb12c | 4947 | next: |
edbd8d4e | 4948 | leaf = path->nodes[0]; |
73e48b27 | 4949 | nritems = btrfs_header_nritems(leaf); |
73e48b27 Y |
4950 | if (path->slots[0] >= nritems) { |
4951 | ret = btrfs_next_leaf(root, path); | |
4952 | if (ret < 0) | |
4953 | goto out; | |
4954 | if (ret == 1) { | |
4955 | ret = 0; | |
4956 | break; | |
edbd8d4e | 4957 | } |
73e48b27 Y |
4958 | leaf = path->nodes[0]; |
4959 | nritems = btrfs_header_nritems(leaf); | |
edbd8d4e | 4960 | } |
73e48b27 | 4961 | |
1a40e23b | 4962 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
725c8463 | 4963 | |
1a40e23b ZY |
4964 | if (key.objectid >= block_group->key.objectid + |
4965 | block_group->key.offset) | |
8f18cf13 CM |
4966 | break; |
4967 | ||
725c8463 | 4968 | if (progress && need_resched()) { |
725c8463 | 4969 | btrfs_release_path(root, path); |
1a40e23b | 4970 | cond_resched(); |
725c8463 | 4971 | progress = 0; |
1a40e23b | 4972 | continue; |
725c8463 CM |
4973 | } |
4974 | progress = 1; | |
4975 | ||
1a40e23b ZY |
4976 | if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY || |
4977 | key.objectid + key.offset <= cur_byte) { | |
edbd8d4e | 4978 | path->slots[0]++; |
edbd8d4e CM |
4979 | goto next; |
4980 | } | |
73e48b27 | 4981 | |
edbd8d4e | 4982 | total_found++; |
1a40e23b | 4983 | cur_byte = key.objectid + key.offset; |
edbd8d4e | 4984 | btrfs_release_path(root, path); |
edbd8d4e | 4985 | |
1a40e23b ZY |
4986 | __alloc_chunk_for_shrink(root, block_group, 0); |
4987 | ret = relocate_one_extent(root, path, &key, block_group, | |
4988 | reloc_inode, pass); | |
4989 | BUG_ON(ret < 0); | |
d899e052 YZ |
4990 | if (ret > 0) |
4991 | skipped++; | |
edbd8d4e | 4992 | |
1a40e23b ZY |
4993 | key.objectid = cur_byte; |
4994 | key.type = 0; | |
4995 | key.offset = 0; | |
edbd8d4e CM |
4996 | } |
4997 | ||
1a40e23b | 4998 | btrfs_release_path(root, path); |
7d9eb12c | 4999 | |
1a40e23b ZY |
5000 | if (pass == 0) { |
5001 | btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1); | |
5002 | invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1); | |
5003 | WARN_ON(reloc_inode->i_mapping->nrpages); | |
8e8a1e31 | 5004 | } |
73e48b27 | 5005 | |
1a40e23b ZY |
5006 | if (total_found > 0) { |
5007 | printk("btrfs found %llu extents in pass %d\n", | |
5008 | (unsigned long long)total_found, pass); | |
5009 | pass++; | |
d899e052 YZ |
5010 | if (total_found == skipped && pass > 2) { |
5011 | iput(reloc_inode); | |
5012 | reloc_inode = create_reloc_inode(info, block_group); | |
5013 | pass = 0; | |
5014 | } | |
1a40e23b | 5015 | goto again; |
0f9dd46c | 5016 | } |
0ef3e66b | 5017 | |
1a40e23b ZY |
5018 | /* delete reloc_inode */ |
5019 | iput(reloc_inode); | |
5020 | ||
5021 | /* unpin extents in this range */ | |
5022 | trans = btrfs_start_transaction(info->tree_root, 1); | |
5023 | btrfs_commit_transaction(trans, info->tree_root); | |
0ef3e66b | 5024 | |
1a40e23b ZY |
5025 | spin_lock(&block_group->lock); |
5026 | WARN_ON(block_group->pinned > 0); | |
5027 | WARN_ON(block_group->reserved > 0); | |
5028 | WARN_ON(btrfs_block_group_used(&block_group->item) > 0); | |
5029 | spin_unlock(&block_group->lock); | |
5030 | ret = 0; | |
edbd8d4e | 5031 | out: |
1a40e23b | 5032 | btrfs_free_path(path); |
edbd8d4e CM |
5033 | return ret; |
5034 | } | |
5035 | ||
0b86a832 CM |
5036 | int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, |
5037 | struct btrfs_key *key) | |
5038 | { | |
925baedd | 5039 | int ret = 0; |
0b86a832 CM |
5040 | struct btrfs_key found_key; |
5041 | struct extent_buffer *leaf; | |
5042 | int slot; | |
edbd8d4e | 5043 | |
0b86a832 CM |
5044 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
5045 | if (ret < 0) | |
925baedd CM |
5046 | goto out; |
5047 | ||
0b86a832 CM |
5048 | while(1) { |
5049 | slot = path->slots[0]; | |
edbd8d4e | 5050 | leaf = path->nodes[0]; |
0b86a832 CM |
5051 | if (slot >= btrfs_header_nritems(leaf)) { |
5052 | ret = btrfs_next_leaf(root, path); | |
5053 | if (ret == 0) | |
5054 | continue; | |
5055 | if (ret < 0) | |
925baedd | 5056 | goto out; |
0b86a832 | 5057 | break; |
edbd8d4e | 5058 | } |
0b86a832 | 5059 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 5060 | |
0b86a832 | 5061 | if (found_key.objectid >= key->objectid && |
925baedd CM |
5062 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5063 | ret = 0; | |
5064 | goto out; | |
5065 | } | |
0b86a832 | 5066 | path->slots[0]++; |
edbd8d4e | 5067 | } |
0b86a832 | 5068 | ret = -ENOENT; |
925baedd | 5069 | out: |
0b86a832 | 5070 | return ret; |
edbd8d4e CM |
5071 | } |
5072 | ||
1a40e23b ZY |
5073 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
5074 | { | |
5075 | struct btrfs_block_group_cache *block_group; | |
5076 | struct rb_node *n; | |
5077 | ||
1a40e23b ZY |
5078 | spin_lock(&info->block_group_cache_lock); |
5079 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
5080 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
5081 | cache_node); | |
1a40e23b ZY |
5082 | rb_erase(&block_group->cache_node, |
5083 | &info->block_group_cache_tree); | |
d899e052 YZ |
5084 | spin_unlock(&info->block_group_cache_lock); |
5085 | ||
5086 | btrfs_remove_free_space_cache(block_group); | |
80eb234a | 5087 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 5088 | list_del(&block_group->list); |
80eb234a | 5089 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 5090 | kfree(block_group); |
d899e052 YZ |
5091 | |
5092 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
5093 | } |
5094 | spin_unlock(&info->block_group_cache_lock); | |
1a40e23b ZY |
5095 | return 0; |
5096 | } | |
5097 | ||
9078a3e1 CM |
5098 | int btrfs_read_block_groups(struct btrfs_root *root) |
5099 | { | |
5100 | struct btrfs_path *path; | |
5101 | int ret; | |
9078a3e1 | 5102 | struct btrfs_block_group_cache *cache; |
be744175 | 5103 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 5104 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
5105 | struct btrfs_key key; |
5106 | struct btrfs_key found_key; | |
5f39d397 | 5107 | struct extent_buffer *leaf; |
96b5179d | 5108 | |
be744175 | 5109 | root = info->extent_root; |
9078a3e1 | 5110 | key.objectid = 0; |
0b86a832 | 5111 | key.offset = 0; |
9078a3e1 | 5112 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
5113 | path = btrfs_alloc_path(); |
5114 | if (!path) | |
5115 | return -ENOMEM; | |
5116 | ||
5117 | while(1) { | |
0b86a832 CM |
5118 | ret = find_first_block_group(root, path, &key); |
5119 | if (ret > 0) { | |
5120 | ret = 0; | |
5121 | goto error; | |
9078a3e1 | 5122 | } |
0b86a832 CM |
5123 | if (ret != 0) |
5124 | goto error; | |
5125 | ||
5f39d397 CM |
5126 | leaf = path->nodes[0]; |
5127 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 5128 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 5129 | if (!cache) { |
0b86a832 | 5130 | ret = -ENOMEM; |
9078a3e1 CM |
5131 | break; |
5132 | } | |
3e1ad54f | 5133 | |
c286ac48 | 5134 | spin_lock_init(&cache->lock); |
25179201 | 5135 | mutex_init(&cache->alloc_mutex); |
0f9dd46c | 5136 | INIT_LIST_HEAD(&cache->list); |
5f39d397 CM |
5137 | read_extent_buffer(leaf, &cache->item, |
5138 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
5139 | sizeof(cache->item)); | |
9078a3e1 | 5140 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 5141 | |
9078a3e1 CM |
5142 | key.objectid = found_key.objectid + found_key.offset; |
5143 | btrfs_release_path(root, path); | |
0b86a832 | 5144 | cache->flags = btrfs_block_group_flags(&cache->item); |
96b5179d | 5145 | |
6324fbf3 CM |
5146 | ret = update_space_info(info, cache->flags, found_key.offset, |
5147 | btrfs_block_group_used(&cache->item), | |
5148 | &space_info); | |
5149 | BUG_ON(ret); | |
5150 | cache->space_info = space_info; | |
80eb234a JB |
5151 | down_write(&space_info->groups_sem); |
5152 | list_add_tail(&cache->list, &space_info->block_groups); | |
5153 | up_write(&space_info->groups_sem); | |
0f9dd46c JB |
5154 | |
5155 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
5156 | BUG_ON(ret); | |
75ccf47d CM |
5157 | |
5158 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
9078a3e1 | 5159 | } |
0b86a832 CM |
5160 | ret = 0; |
5161 | error: | |
9078a3e1 | 5162 | btrfs_free_path(path); |
0b86a832 | 5163 | return ret; |
9078a3e1 | 5164 | } |
6324fbf3 CM |
5165 | |
5166 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
5167 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 5168 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
5169 | u64 size) |
5170 | { | |
5171 | int ret; | |
6324fbf3 CM |
5172 | struct btrfs_root *extent_root; |
5173 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
5174 | |
5175 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 5176 | |
e02119d5 CM |
5177 | root->fs_info->last_trans_new_blockgroup = trans->transid; |
5178 | ||
8f18cf13 | 5179 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
5180 | if (!cache) |
5181 | return -ENOMEM; | |
5182 | ||
e17cade2 | 5183 | cache->key.objectid = chunk_offset; |
6324fbf3 | 5184 | cache->key.offset = size; |
c286ac48 | 5185 | spin_lock_init(&cache->lock); |
25179201 | 5186 | mutex_init(&cache->alloc_mutex); |
0f9dd46c | 5187 | INIT_LIST_HEAD(&cache->list); |
6324fbf3 | 5188 | btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
0ef3e66b | 5189 | |
6324fbf3 | 5190 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
5191 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
5192 | cache->flags = type; | |
5193 | btrfs_set_block_group_flags(&cache->item, type); | |
5194 | ||
5195 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, | |
5196 | &cache->space_info); | |
5197 | BUG_ON(ret); | |
80eb234a JB |
5198 | down_write(&cache->space_info->groups_sem); |
5199 | list_add_tail(&cache->list, &cache->space_info->block_groups); | |
5200 | up_write(&cache->space_info->groups_sem); | |
6324fbf3 | 5201 | |
0f9dd46c JB |
5202 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
5203 | BUG_ON(ret); | |
c286ac48 | 5204 | |
6324fbf3 CM |
5205 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
5206 | sizeof(cache->item)); | |
5207 | BUG_ON(ret); | |
5208 | ||
87ef2bb4 CM |
5209 | finish_current_insert(trans, extent_root, 0); |
5210 | ret = del_pending_extents(trans, extent_root, 0); | |
6324fbf3 | 5211 | BUG_ON(ret); |
d18a2c44 | 5212 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 5213 | |
6324fbf3 CM |
5214 | return 0; |
5215 | } | |
1a40e23b ZY |
5216 | |
5217 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
5218 | struct btrfs_root *root, u64 group_start) | |
5219 | { | |
5220 | struct btrfs_path *path; | |
5221 | struct btrfs_block_group_cache *block_group; | |
5222 | struct btrfs_key key; | |
5223 | int ret; | |
5224 | ||
1a40e23b ZY |
5225 | root = root->fs_info->extent_root; |
5226 | ||
5227 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
5228 | BUG_ON(!block_group); | |
5229 | ||
5230 | memcpy(&key, &block_group->key, sizeof(key)); | |
5231 | ||
5232 | path = btrfs_alloc_path(); | |
5233 | BUG_ON(!path); | |
5234 | ||
5235 | btrfs_remove_free_space_cache(block_group); | |
5236 | rb_erase(&block_group->cache_node, | |
5237 | &root->fs_info->block_group_cache_tree); | |
80eb234a | 5238 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 5239 | list_del(&block_group->list); |
80eb234a | 5240 | up_write(&block_group->space_info->groups_sem); |
1a40e23b ZY |
5241 | |
5242 | /* | |
5243 | memset(shrink_block_group, 0, sizeof(*shrink_block_group)); | |
5244 | kfree(shrink_block_group); | |
5245 | */ | |
5246 | ||
5247 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5248 | if (ret > 0) | |
5249 | ret = -EIO; | |
5250 | if (ret < 0) | |
5251 | goto out; | |
5252 | ||
5253 | ret = btrfs_del_item(trans, root, path); | |
5254 | out: | |
5255 | btrfs_free_path(path); | |
5256 | return ret; | |
5257 | } |