]>
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> |
74493f7a | 21 | #include "hash.h" |
a5eb62e3 | 22 | #include "crc32c.h" |
fec577fb CM |
23 | #include "ctree.h" |
24 | #include "disk-io.h" | |
25 | #include "print-tree.h" | |
e089f05c | 26 | #include "transaction.h" |
0b86a832 | 27 | #include "volumes.h" |
925baedd | 28 | #include "locking.h" |
fec577fb | 29 | |
0b86a832 | 30 | #define BLOCK_GROUP_DATA EXTENT_WRITEBACK |
96b5179d | 31 | #define BLOCK_GROUP_METADATA EXTENT_UPTODATE |
0b86a832 CM |
32 | #define BLOCK_GROUP_SYSTEM EXTENT_NEW |
33 | ||
96b5179d CM |
34 | #define BLOCK_GROUP_DIRTY EXTENT_DIRTY |
35 | ||
e089f05c CM |
36 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
37 | btrfs_root *extent_root); | |
e20d96d6 CM |
38 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
39 | btrfs_root *extent_root); | |
925baedd CM |
40 | static struct btrfs_block_group_cache * |
41 | __btrfs_find_block_group(struct btrfs_root *root, | |
42 | struct btrfs_block_group_cache *hint, | |
43 | u64 search_start, int data, int owner); | |
d548ee51 | 44 | |
925baedd CM |
45 | void maybe_lock_mutex(struct btrfs_root *root) |
46 | { | |
47 | if (root != root->fs_info->extent_root && | |
48 | root != root->fs_info->chunk_root && | |
49 | root != root->fs_info->dev_root) { | |
50 | mutex_lock(&root->fs_info->alloc_mutex); | |
51 | } | |
52 | } | |
53 | ||
54 | void maybe_unlock_mutex(struct btrfs_root *root) | |
55 | { | |
56 | if (root != root->fs_info->extent_root && | |
57 | root != root->fs_info->chunk_root && | |
58 | root != root->fs_info->dev_root) { | |
59 | mutex_unlock(&root->fs_info->alloc_mutex); | |
60 | } | |
61 | } | |
fec577fb | 62 | |
e37c9e69 CM |
63 | static int cache_block_group(struct btrfs_root *root, |
64 | struct btrfs_block_group_cache *block_group) | |
65 | { | |
66 | struct btrfs_path *path; | |
67 | int ret; | |
68 | struct btrfs_key key; | |
5f39d397 | 69 | struct extent_buffer *leaf; |
d1310b2e | 70 | struct extent_io_tree *free_space_cache; |
e37c9e69 | 71 | int slot; |
e37c9e69 CM |
72 | u64 last = 0; |
73 | u64 hole_size; | |
7d7d6068 | 74 | u64 first_free; |
e37c9e69 CM |
75 | int found = 0; |
76 | ||
00f5c795 CM |
77 | if (!block_group) |
78 | return 0; | |
79 | ||
e37c9e69 | 80 | root = root->fs_info->extent_root; |
f510cfec | 81 | free_space_cache = &root->fs_info->free_space_cache; |
e37c9e69 CM |
82 | |
83 | if (block_group->cached) | |
84 | return 0; | |
f510cfec | 85 | |
e37c9e69 CM |
86 | path = btrfs_alloc_path(); |
87 | if (!path) | |
88 | return -ENOMEM; | |
7d7d6068 | 89 | |
2cc58cf2 | 90 | path->reada = 2; |
5cd57b2c CM |
91 | /* |
92 | * we get into deadlocks with paths held by callers of this function. | |
93 | * since the alloc_mutex is protecting things right now, just | |
94 | * skip the locking here | |
95 | */ | |
96 | path->skip_locking = 1; | |
7d7d6068 | 97 | first_free = block_group->key.objectid; |
e37c9e69 | 98 | key.objectid = block_group->key.objectid; |
e37c9e69 CM |
99 | key.offset = 0; |
100 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
101 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
102 | if (ret < 0) | |
103 | return ret; | |
0b86a832 | 104 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
d548ee51 Y |
105 | if (ret < 0) |
106 | return ret; | |
107 | if (ret == 0) { | |
108 | leaf = path->nodes[0]; | |
109 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
110 | if (key.objectid + key.offset > first_free) | |
111 | first_free = key.objectid + key.offset; | |
112 | } | |
e37c9e69 | 113 | while(1) { |
5f39d397 | 114 | leaf = path->nodes[0]; |
e37c9e69 | 115 | slot = path->slots[0]; |
5f39d397 | 116 | if (slot >= btrfs_header_nritems(leaf)) { |
e37c9e69 | 117 | ret = btrfs_next_leaf(root, path); |
54aa1f4d CM |
118 | if (ret < 0) |
119 | goto err; | |
de428b63 | 120 | if (ret == 0) { |
e37c9e69 | 121 | continue; |
de428b63 | 122 | } else { |
e37c9e69 CM |
123 | break; |
124 | } | |
125 | } | |
5f39d397 | 126 | btrfs_item_key_to_cpu(leaf, &key, slot); |
7d7d6068 | 127 | if (key.objectid < block_group->key.objectid) { |
7d7d6068 Y |
128 | goto next; |
129 | } | |
e37c9e69 CM |
130 | if (key.objectid >= block_group->key.objectid + |
131 | block_group->key.offset) { | |
e37c9e69 CM |
132 | break; |
133 | } | |
7d7d6068 | 134 | |
e37c9e69 CM |
135 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
136 | if (!found) { | |
7d7d6068 | 137 | last = first_free; |
e37c9e69 | 138 | found = 1; |
e37c9e69 | 139 | } |
f510cfec CM |
140 | if (key.objectid > last) { |
141 | hole_size = key.objectid - last; | |
142 | set_extent_dirty(free_space_cache, last, | |
143 | last + hole_size - 1, | |
144 | GFP_NOFS); | |
7d7d6068 Y |
145 | } |
146 | last = key.objectid + key.offset; | |
e37c9e69 | 147 | } |
7d7d6068 | 148 | next: |
e37c9e69 CM |
149 | path->slots[0]++; |
150 | } | |
151 | ||
7d7d6068 Y |
152 | if (!found) |
153 | last = first_free; | |
154 | if (block_group->key.objectid + | |
155 | block_group->key.offset > last) { | |
156 | hole_size = block_group->key.objectid + | |
157 | block_group->key.offset - last; | |
f510cfec CM |
158 | set_extent_dirty(free_space_cache, last, |
159 | last + hole_size - 1, GFP_NOFS); | |
7d7d6068 | 160 | } |
e37c9e69 | 161 | block_group->cached = 1; |
54aa1f4d | 162 | err: |
e37c9e69 CM |
163 | btrfs_free_path(path); |
164 | return 0; | |
165 | } | |
166 | ||
0ef3e66b CM |
167 | struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct |
168 | btrfs_fs_info *info, | |
169 | u64 bytenr) | |
170 | { | |
171 | struct extent_io_tree *block_group_cache; | |
172 | struct btrfs_block_group_cache *block_group = NULL; | |
173 | u64 ptr; | |
174 | u64 start; | |
175 | u64 end; | |
176 | int ret; | |
177 | ||
178 | bytenr = max_t(u64, bytenr, | |
179 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
180 | block_group_cache = &info->block_group_cache; | |
181 | ret = find_first_extent_bit(block_group_cache, | |
182 | bytenr, &start, &end, | |
183 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | | |
184 | BLOCK_GROUP_SYSTEM); | |
185 | if (ret) { | |
186 | return NULL; | |
187 | } | |
188 | ret = get_state_private(block_group_cache, start, &ptr); | |
189 | if (ret) | |
190 | return NULL; | |
191 | ||
192 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; | |
193 | return block_group; | |
194 | } | |
195 | ||
5276aeda CM |
196 | struct btrfs_block_group_cache *btrfs_lookup_block_group(struct |
197 | btrfs_fs_info *info, | |
db94535d | 198 | u64 bytenr) |
be744175 | 199 | { |
d1310b2e | 200 | struct extent_io_tree *block_group_cache; |
96b5179d CM |
201 | struct btrfs_block_group_cache *block_group = NULL; |
202 | u64 ptr; | |
203 | u64 start; | |
204 | u64 end; | |
be744175 CM |
205 | int ret; |
206 | ||
a061fc8d CM |
207 | bytenr = max_t(u64, bytenr, |
208 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
96b5179d CM |
209 | block_group_cache = &info->block_group_cache; |
210 | ret = find_first_extent_bit(block_group_cache, | |
db94535d | 211 | bytenr, &start, &end, |
0b86a832 CM |
212 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | |
213 | BLOCK_GROUP_SYSTEM); | |
be744175 | 214 | if (ret) { |
96b5179d | 215 | return NULL; |
be744175 | 216 | } |
96b5179d CM |
217 | ret = get_state_private(block_group_cache, start, &ptr); |
218 | if (ret) | |
219 | return NULL; | |
220 | ||
ae2f5411 | 221 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
5cf66426 | 222 | if (block_group->key.objectid <= bytenr && bytenr < |
96b5179d CM |
223 | block_group->key.objectid + block_group->key.offset) |
224 | return block_group; | |
be744175 CM |
225 | return NULL; |
226 | } | |
0b86a832 CM |
227 | |
228 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) | |
229 | { | |
593060d7 | 230 | return (cache->flags & bits) == bits; |
0b86a832 CM |
231 | } |
232 | ||
233 | static int noinline find_search_start(struct btrfs_root *root, | |
98ed5174 | 234 | struct btrfs_block_group_cache **cache_ret, |
0ef3e66b | 235 | u64 *start_ret, u64 num, int data) |
e37c9e69 | 236 | { |
e37c9e69 CM |
237 | int ret; |
238 | struct btrfs_block_group_cache *cache = *cache_ret; | |
d7fc640e | 239 | struct extent_io_tree *free_space_cache; |
7d1660d4 | 240 | struct extent_state *state; |
e19caa5f | 241 | u64 last; |
f510cfec | 242 | u64 start = 0; |
257d0ce3 | 243 | u64 cache_miss = 0; |
c31f8830 | 244 | u64 total_fs_bytes; |
0b86a832 | 245 | u64 search_start = *start_ret; |
f84a8b36 | 246 | int wrapped = 0; |
e37c9e69 | 247 | |
7d9eb12c | 248 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
c31f8830 | 249 | total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); |
d7fc640e CM |
250 | free_space_cache = &root->fs_info->free_space_cache; |
251 | ||
0ef3e66b CM |
252 | if (!cache) |
253 | goto out; | |
254 | ||
e37c9e69 | 255 | again: |
54aa1f4d | 256 | ret = cache_block_group(root, cache); |
0ef3e66b | 257 | if (ret) { |
54aa1f4d | 258 | goto out; |
0ef3e66b | 259 | } |
f84a8b36 | 260 | |
e19caa5f | 261 | last = max(search_start, cache->key.objectid); |
0ef3e66b | 262 | if (!block_group_bits(cache, data) || cache->ro) |
0b86a832 | 263 | goto new_group; |
e19caa5f | 264 | |
7d1660d4 CM |
265 | spin_lock_irq(&free_space_cache->lock); |
266 | state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY); | |
e37c9e69 | 267 | while(1) { |
7d1660d4 | 268 | if (!state) { |
257d0ce3 CM |
269 | if (!cache_miss) |
270 | cache_miss = last; | |
7d1660d4 | 271 | spin_unlock_irq(&free_space_cache->lock); |
e19caa5f CM |
272 | goto new_group; |
273 | } | |
f510cfec | 274 | |
7d1660d4 CM |
275 | start = max(last, state->start); |
276 | last = state->end + 1; | |
257d0ce3 | 277 | if (last - start < num) { |
7d1660d4 CM |
278 | do { |
279 | state = extent_state_next(state); | |
280 | } while(state && !(state->state & EXTENT_DIRTY)); | |
f510cfec | 281 | continue; |
257d0ce3 | 282 | } |
7d1660d4 | 283 | spin_unlock_irq(&free_space_cache->lock); |
0ef3e66b | 284 | if (cache->ro) { |
8f18cf13 | 285 | goto new_group; |
0ef3e66b | 286 | } |
0b86a832 | 287 | if (start + num > cache->key.objectid + cache->key.offset) |
e37c9e69 | 288 | goto new_group; |
8790d502 | 289 | if (!block_group_bits(cache, data)) { |
611f0e00 | 290 | printk("block group bits don't match %Lu %d\n", cache->flags, data); |
8790d502 | 291 | } |
0b86a832 CM |
292 | *start_ret = start; |
293 | return 0; | |
8790d502 CM |
294 | } |
295 | out: | |
1a2b2ac7 CM |
296 | cache = btrfs_lookup_block_group(root->fs_info, search_start); |
297 | if (!cache) { | |
0b86a832 | 298 | printk("Unable to find block group for %Lu\n", search_start); |
1a2b2ac7 | 299 | WARN_ON(1); |
1a2b2ac7 | 300 | } |
0b86a832 | 301 | return -ENOSPC; |
e37c9e69 CM |
302 | |
303 | new_group: | |
e19caa5f | 304 | last = cache->key.objectid + cache->key.offset; |
f84a8b36 | 305 | wrapped: |
0ef3e66b | 306 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
c31f8830 | 307 | if (!cache || cache->key.objectid >= total_fs_bytes) { |
0e4de584 | 308 | no_cache: |
f84a8b36 CM |
309 | if (!wrapped) { |
310 | wrapped = 1; | |
311 | last = search_start; | |
f84a8b36 CM |
312 | goto wrapped; |
313 | } | |
1a2b2ac7 | 314 | goto out; |
e37c9e69 | 315 | } |
257d0ce3 CM |
316 | if (cache_miss && !cache->cached) { |
317 | cache_block_group(root, cache); | |
318 | last = cache_miss; | |
0ef3e66b | 319 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
257d0ce3 | 320 | } |
0ef3e66b | 321 | cache_miss = 0; |
c286ac48 | 322 | cache = btrfs_find_block_group(root, cache, last, data, 0); |
0e4de584 CM |
323 | if (!cache) |
324 | goto no_cache; | |
e37c9e69 CM |
325 | *cache_ret = cache; |
326 | goto again; | |
327 | } | |
328 | ||
84f54cfa CM |
329 | static u64 div_factor(u64 num, int factor) |
330 | { | |
257d0ce3 CM |
331 | if (factor == 10) |
332 | return num; | |
84f54cfa CM |
333 | num *= factor; |
334 | do_div(num, 10); | |
335 | return num; | |
336 | } | |
337 | ||
6324fbf3 CM |
338 | static int block_group_state_bits(u64 flags) |
339 | { | |
340 | int bits = 0; | |
341 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
342 | bits |= BLOCK_GROUP_DATA; | |
343 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
344 | bits |= BLOCK_GROUP_METADATA; | |
345 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
346 | bits |= BLOCK_GROUP_SYSTEM; | |
347 | return bits; | |
348 | } | |
349 | ||
925baedd CM |
350 | static struct btrfs_block_group_cache * |
351 | __btrfs_find_block_group(struct btrfs_root *root, | |
352 | struct btrfs_block_group_cache *hint, | |
353 | u64 search_start, int data, int owner) | |
cd1bc465 | 354 | { |
96b5179d | 355 | struct btrfs_block_group_cache *cache; |
d1310b2e | 356 | struct extent_io_tree *block_group_cache; |
31f3c99b | 357 | struct btrfs_block_group_cache *found_group = NULL; |
cd1bc465 CM |
358 | struct btrfs_fs_info *info = root->fs_info; |
359 | u64 used; | |
31f3c99b | 360 | u64 last = 0; |
96b5179d CM |
361 | u64 start; |
362 | u64 end; | |
363 | u64 free_check; | |
364 | u64 ptr; | |
365 | int bit; | |
cd1bc465 | 366 | int ret; |
31f3c99b | 367 | int full_search = 0; |
bce4eae9 | 368 | int factor = 10; |
0ef3e66b | 369 | int wrapped = 0; |
de428b63 | 370 | |
96b5179d CM |
371 | block_group_cache = &info->block_group_cache; |
372 | ||
a236aed1 CM |
373 | if (data & BTRFS_BLOCK_GROUP_METADATA) |
374 | factor = 9; | |
be744175 | 375 | |
6324fbf3 | 376 | bit = block_group_state_bits(data); |
be744175 | 377 | |
0ef3e66b | 378 | if (search_start) { |
be744175 | 379 | struct btrfs_block_group_cache *shint; |
0ef3e66b | 380 | shint = btrfs_lookup_first_block_group(info, search_start); |
8f18cf13 | 381 | if (shint && block_group_bits(shint, data) && !shint->ro) { |
c286ac48 | 382 | spin_lock(&shint->lock); |
be744175 | 383 | used = btrfs_block_group_used(&shint->item); |
324ae4df Y |
384 | if (used + shint->pinned < |
385 | div_factor(shint->key.offset, factor)) { | |
c286ac48 | 386 | spin_unlock(&shint->lock); |
be744175 CM |
387 | return shint; |
388 | } | |
c286ac48 | 389 | spin_unlock(&shint->lock); |
be744175 CM |
390 | } |
391 | } | |
0ef3e66b | 392 | if (hint && !hint->ro && block_group_bits(hint, data)) { |
c286ac48 | 393 | spin_lock(&hint->lock); |
31f3c99b | 394 | used = btrfs_block_group_used(&hint->item); |
324ae4df Y |
395 | if (used + hint->pinned < |
396 | div_factor(hint->key.offset, factor)) { | |
c286ac48 | 397 | spin_unlock(&hint->lock); |
31f3c99b CM |
398 | return hint; |
399 | } | |
c286ac48 | 400 | spin_unlock(&hint->lock); |
e19caa5f | 401 | last = hint->key.objectid + hint->key.offset; |
31f3c99b | 402 | } else { |
e37c9e69 | 403 | if (hint) |
0ef3e66b | 404 | last = max(hint->key.objectid, search_start); |
e37c9e69 | 405 | else |
0ef3e66b | 406 | last = search_start; |
31f3c99b | 407 | } |
31f3c99b | 408 | again: |
cd1bc465 | 409 | while(1) { |
96b5179d CM |
410 | ret = find_first_extent_bit(block_group_cache, last, |
411 | &start, &end, bit); | |
412 | if (ret) | |
cd1bc465 | 413 | break; |
96b5179d CM |
414 | |
415 | ret = get_state_private(block_group_cache, start, &ptr); | |
0ef3e66b CM |
416 | if (ret) { |
417 | last = end + 1; | |
418 | continue; | |
419 | } | |
96b5179d | 420 | |
ae2f5411 | 421 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
c286ac48 | 422 | spin_lock(&cache->lock); |
96b5179d CM |
423 | last = cache->key.objectid + cache->key.offset; |
424 | used = btrfs_block_group_used(&cache->item); | |
425 | ||
8f18cf13 | 426 | if (!cache->ro && block_group_bits(cache, data)) { |
0ef3e66b | 427 | free_check = div_factor(cache->key.offset, factor); |
8790d502 CM |
428 | if (used + cache->pinned < free_check) { |
429 | found_group = cache; | |
c286ac48 | 430 | spin_unlock(&cache->lock); |
8790d502 CM |
431 | goto found; |
432 | } | |
6324fbf3 | 433 | } |
c286ac48 | 434 | spin_unlock(&cache->lock); |
de428b63 | 435 | cond_resched(); |
cd1bc465 | 436 | } |
0ef3e66b CM |
437 | if (!wrapped) { |
438 | last = search_start; | |
439 | wrapped = 1; | |
440 | goto again; | |
441 | } | |
442 | if (!full_search && factor < 10) { | |
be744175 | 443 | last = search_start; |
31f3c99b | 444 | full_search = 1; |
0ef3e66b | 445 | factor = 10; |
31f3c99b CM |
446 | goto again; |
447 | } | |
be744175 | 448 | found: |
31f3c99b | 449 | return found_group; |
cd1bc465 CM |
450 | } |
451 | ||
925baedd CM |
452 | struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, |
453 | struct btrfs_block_group_cache | |
454 | *hint, u64 search_start, | |
455 | int data, int owner) | |
456 | { | |
457 | ||
458 | struct btrfs_block_group_cache *ret; | |
925baedd | 459 | ret = __btrfs_find_block_group(root, hint, search_start, data, owner); |
925baedd CM |
460 | return ret; |
461 | } | |
7bb86316 | 462 | static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation, |
74493f7a CM |
463 | u64 owner, u64 owner_offset) |
464 | { | |
465 | u32 high_crc = ~(u32)0; | |
466 | u32 low_crc = ~(u32)0; | |
467 | __le64 lenum; | |
74493f7a | 468 | lenum = cpu_to_le64(root_objectid); |
a5eb62e3 | 469 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
7bb86316 | 470 | lenum = cpu_to_le64(ref_generation); |
a5eb62e3 | 471 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d CM |
472 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) { |
473 | lenum = cpu_to_le64(owner); | |
a5eb62e3 | 474 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d | 475 | lenum = cpu_to_le64(owner_offset); |
a5eb62e3 | 476 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d | 477 | } |
74493f7a CM |
478 | return ((u64)high_crc << 32) | (u64)low_crc; |
479 | } | |
480 | ||
7bb86316 CM |
481 | static int match_extent_ref(struct extent_buffer *leaf, |
482 | struct btrfs_extent_ref *disk_ref, | |
483 | struct btrfs_extent_ref *cpu_ref) | |
484 | { | |
485 | int ret; | |
486 | int len; | |
487 | ||
488 | if (cpu_ref->objectid) | |
489 | len = sizeof(*cpu_ref); | |
490 | else | |
491 | len = 2 * sizeof(u64); | |
492 | ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref, | |
493 | len); | |
494 | return ret == 0; | |
495 | } | |
496 | ||
98ed5174 CM |
497 | static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, |
498 | struct btrfs_root *root, | |
499 | struct btrfs_path *path, u64 bytenr, | |
500 | u64 root_objectid, | |
501 | u64 ref_generation, u64 owner, | |
502 | u64 owner_offset, int del) | |
74493f7a CM |
503 | { |
504 | u64 hash; | |
505 | struct btrfs_key key; | |
7bb86316 | 506 | struct btrfs_key found_key; |
74493f7a | 507 | struct btrfs_extent_ref ref; |
7bb86316 CM |
508 | struct extent_buffer *leaf; |
509 | struct btrfs_extent_ref *disk_ref; | |
510 | int ret; | |
511 | int ret2; | |
512 | ||
513 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
514 | btrfs_set_stack_ref_generation(&ref, ref_generation); | |
515 | btrfs_set_stack_ref_objectid(&ref, owner); | |
516 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
517 | ||
518 | hash = hash_extent_ref(root_objectid, ref_generation, owner, | |
519 | owner_offset); | |
520 | key.offset = hash; | |
521 | key.objectid = bytenr; | |
522 | key.type = BTRFS_EXTENT_REF_KEY; | |
523 | ||
524 | while (1) { | |
525 | ret = btrfs_search_slot(trans, root, &key, path, | |
526 | del ? -1 : 0, del); | |
527 | if (ret < 0) | |
528 | goto out; | |
529 | leaf = path->nodes[0]; | |
530 | if (ret != 0) { | |
531 | u32 nritems = btrfs_header_nritems(leaf); | |
532 | if (path->slots[0] >= nritems) { | |
533 | ret2 = btrfs_next_leaf(root, path); | |
534 | if (ret2) | |
535 | goto out; | |
536 | leaf = path->nodes[0]; | |
537 | } | |
538 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
539 | if (found_key.objectid != bytenr || | |
540 | found_key.type != BTRFS_EXTENT_REF_KEY) | |
541 | goto out; | |
542 | key.offset = found_key.offset; | |
543 | if (del) { | |
544 | btrfs_release_path(root, path); | |
545 | continue; | |
546 | } | |
547 | } | |
548 | disk_ref = btrfs_item_ptr(path->nodes[0], | |
549 | path->slots[0], | |
550 | struct btrfs_extent_ref); | |
551 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) { | |
552 | ret = 0; | |
553 | goto out; | |
554 | } | |
555 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
556 | key.offset = found_key.offset + 1; | |
557 | btrfs_release_path(root, path); | |
558 | } | |
559 | out: | |
560 | return ret; | |
561 | } | |
562 | ||
d8d5f3e1 CM |
563 | /* |
564 | * Back reference rules. Back refs have three main goals: | |
565 | * | |
566 | * 1) differentiate between all holders of references to an extent so that | |
567 | * when a reference is dropped we can make sure it was a valid reference | |
568 | * before freeing the extent. | |
569 | * | |
570 | * 2) Provide enough information to quickly find the holders of an extent | |
571 | * if we notice a given block is corrupted or bad. | |
572 | * | |
573 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
574 | * maintenance. This is actually the same as #2, but with a slightly | |
575 | * different use case. | |
576 | * | |
577 | * File extents can be referenced by: | |
578 | * | |
579 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
580 | * - different files inside a single subvolume (in theory, not implemented yet) | |
581 | * - different offsets inside a file (bookend extents in file.c) | |
582 | * | |
583 | * The extent ref structure has fields for: | |
584 | * | |
585 | * - Objectid of the subvolume root | |
586 | * - Generation number of the tree holding the reference | |
587 | * - objectid of the file holding the reference | |
588 | * - offset in the file corresponding to the key holding the reference | |
589 | * | |
590 | * When a file extent is allocated the fields are filled in: | |
591 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
592 | * | |
593 | * When a leaf is cow'd new references are added for every file extent found | |
594 | * in the leaf. It looks the same as the create case, but trans->transid | |
595 | * will be different when the block is cow'd. | |
596 | * | |
597 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
598 | * | |
599 | * When a file extent is removed either during snapshot deletion or file | |
600 | * truncation, the corresponding back reference is found | |
601 | * by searching for: | |
602 | * | |
603 | * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), | |
604 | * inode objectid, offset in file) | |
605 | * | |
606 | * Btree extents can be referenced by: | |
607 | * | |
608 | * - Different subvolumes | |
609 | * - Different generations of the same subvolume | |
610 | * | |
611 | * Storing sufficient information for a full reverse mapping of a btree | |
612 | * block would require storing the lowest key of the block in the backref, | |
613 | * and it would require updating that lowest key either before write out or | |
614 | * every time it changed. Instead, the objectid of the lowest key is stored | |
615 | * along with the level of the tree block. This provides a hint | |
616 | * about where in the btree the block can be found. Searches through the | |
617 | * btree only need to look for a pointer to that block, so they stop one | |
618 | * level higher than the level recorded in the backref. | |
619 | * | |
620 | * Some btrees do not do reference counting on their extents. These | |
621 | * include the extent tree and the tree of tree roots. Backrefs for these | |
622 | * trees always have a generation of zero. | |
623 | * | |
624 | * When a tree block is created, back references are inserted: | |
625 | * | |
f6dbff55 | 626 | * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid) |
d8d5f3e1 CM |
627 | * |
628 | * When a tree block is cow'd in a reference counted root, | |
629 | * new back references are added for all the blocks it points to. | |
630 | * These are of the form (trans->transid will have increased since creation): | |
631 | * | |
f6dbff55 | 632 | * (root->root_key.objectid, trans->transid, level, lowest_key_objectid) |
d8d5f3e1 CM |
633 | * |
634 | * Because the lowest_key_objectid and the level are just hints | |
635 | * they are not used when backrefs are deleted. When a backref is deleted: | |
636 | * | |
637 | * if backref was for a tree root: | |
638 | * root_objectid = root->root_key.objectid | |
639 | * else | |
640 | * root_objectid = btrfs_header_owner(parent) | |
641 | * | |
642 | * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0) | |
643 | * | |
644 | * Back Reference Key hashing: | |
645 | * | |
646 | * Back references have four fields, each 64 bits long. Unfortunately, | |
647 | * This is hashed into a single 64 bit number and placed into the key offset. | |
648 | * The key objectid corresponds to the first byte in the extent, and the | |
649 | * key type is set to BTRFS_EXTENT_REF_KEY | |
650 | */ | |
7bb86316 CM |
651 | int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans, |
652 | struct btrfs_root *root, | |
653 | struct btrfs_path *path, u64 bytenr, | |
654 | u64 root_objectid, u64 ref_generation, | |
655 | u64 owner, u64 owner_offset) | |
656 | { | |
657 | u64 hash; | |
658 | struct btrfs_key key; | |
659 | struct btrfs_extent_ref ref; | |
660 | struct btrfs_extent_ref *disk_ref; | |
74493f7a CM |
661 | int ret; |
662 | ||
663 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
7bb86316 | 664 | btrfs_set_stack_ref_generation(&ref, ref_generation); |
74493f7a CM |
665 | btrfs_set_stack_ref_objectid(&ref, owner); |
666 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
667 | ||
7bb86316 CM |
668 | hash = hash_extent_ref(root_objectid, ref_generation, owner, |
669 | owner_offset); | |
74493f7a CM |
670 | key.offset = hash; |
671 | key.objectid = bytenr; | |
672 | key.type = BTRFS_EXTENT_REF_KEY; | |
673 | ||
674 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref)); | |
675 | while (ret == -EEXIST) { | |
7bb86316 CM |
676 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
677 | struct btrfs_extent_ref); | |
678 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) | |
679 | goto out; | |
680 | key.offset++; | |
681 | btrfs_release_path(root, path); | |
682 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
683 | sizeof(ref)); | |
74493f7a | 684 | } |
7bb86316 CM |
685 | if (ret) |
686 | goto out; | |
687 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
688 | struct btrfs_extent_ref); | |
689 | write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref, | |
690 | sizeof(ref)); | |
691 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
692 | out: | |
693 | btrfs_release_path(root, path); | |
694 | return ret; | |
74493f7a CM |
695 | } |
696 | ||
925baedd | 697 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
b18c6685 | 698 | struct btrfs_root *root, |
74493f7a | 699 | u64 bytenr, u64 num_bytes, |
7bb86316 | 700 | u64 root_objectid, u64 ref_generation, |
74493f7a | 701 | u64 owner, u64 owner_offset) |
02217ed2 | 702 | { |
5caf2a00 | 703 | struct btrfs_path *path; |
02217ed2 | 704 | int ret; |
e2fa7227 | 705 | struct btrfs_key key; |
5f39d397 | 706 | struct extent_buffer *l; |
234b63a0 | 707 | struct btrfs_extent_item *item; |
cf27e1ee | 708 | u32 refs; |
037e6390 | 709 | |
db94535d | 710 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 711 | path = btrfs_alloc_path(); |
54aa1f4d CM |
712 | if (!path) |
713 | return -ENOMEM; | |
26b8003f | 714 | |
3c12ac72 | 715 | path->reada = 1; |
db94535d | 716 | key.objectid = bytenr; |
62e2749e | 717 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 718 | key.offset = num_bytes; |
5caf2a00 | 719 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 720 | 0, 1); |
54aa1f4d CM |
721 | if (ret < 0) |
722 | return ret; | |
a429e513 | 723 | if (ret != 0) { |
a28ec197 | 724 | BUG(); |
a429e513 | 725 | } |
02217ed2 | 726 | BUG_ON(ret != 0); |
5f39d397 | 727 | l = path->nodes[0]; |
5caf2a00 | 728 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 CM |
729 | refs = btrfs_extent_refs(l, item); |
730 | btrfs_set_extent_refs(l, item, refs + 1); | |
5caf2a00 | 731 | btrfs_mark_buffer_dirty(path->nodes[0]); |
a28ec197 | 732 | |
5caf2a00 | 733 | btrfs_release_path(root->fs_info->extent_root, path); |
7bb86316 | 734 | |
3c12ac72 | 735 | path->reada = 1; |
7bb86316 CM |
736 | ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root, |
737 | path, bytenr, root_objectid, | |
738 | ref_generation, owner, owner_offset); | |
739 | BUG_ON(ret); | |
9f5fae2f | 740 | finish_current_insert(trans, root->fs_info->extent_root); |
e20d96d6 | 741 | del_pending_extents(trans, root->fs_info->extent_root); |
74493f7a CM |
742 | |
743 | btrfs_free_path(path); | |
02217ed2 CM |
744 | return 0; |
745 | } | |
746 | ||
925baedd CM |
747 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
748 | struct btrfs_root *root, | |
749 | u64 bytenr, u64 num_bytes, | |
750 | u64 root_objectid, u64 ref_generation, | |
751 | u64 owner, u64 owner_offset) | |
752 | { | |
753 | int ret; | |
754 | ||
755 | mutex_lock(&root->fs_info->alloc_mutex); | |
756 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, | |
757 | root_objectid, ref_generation, | |
758 | owner, owner_offset); | |
759 | mutex_unlock(&root->fs_info->alloc_mutex); | |
760 | return ret; | |
761 | } | |
762 | ||
e9d0b13b CM |
763 | int btrfs_extent_post_op(struct btrfs_trans_handle *trans, |
764 | struct btrfs_root *root) | |
765 | { | |
766 | finish_current_insert(trans, root->fs_info->extent_root); | |
767 | del_pending_extents(trans, root->fs_info->extent_root); | |
768 | return 0; | |
769 | } | |
770 | ||
b18c6685 | 771 | static int lookup_extent_ref(struct btrfs_trans_handle *trans, |
db94535d CM |
772 | struct btrfs_root *root, u64 bytenr, |
773 | u64 num_bytes, u32 *refs) | |
a28ec197 | 774 | { |
5caf2a00 | 775 | struct btrfs_path *path; |
a28ec197 | 776 | int ret; |
e2fa7227 | 777 | struct btrfs_key key; |
5f39d397 | 778 | struct extent_buffer *l; |
234b63a0 | 779 | struct btrfs_extent_item *item; |
5caf2a00 | 780 | |
db94535d | 781 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 782 | path = btrfs_alloc_path(); |
3c12ac72 | 783 | path->reada = 1; |
db94535d CM |
784 | key.objectid = bytenr; |
785 | key.offset = num_bytes; | |
62e2749e | 786 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
5caf2a00 | 787 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 788 | 0, 0); |
54aa1f4d CM |
789 | if (ret < 0) |
790 | goto out; | |
5f39d397 CM |
791 | if (ret != 0) { |
792 | btrfs_print_leaf(root, path->nodes[0]); | |
db94535d | 793 | printk("failed to find block number %Lu\n", bytenr); |
a28ec197 | 794 | BUG(); |
5f39d397 CM |
795 | } |
796 | l = path->nodes[0]; | |
5caf2a00 | 797 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 | 798 | *refs = btrfs_extent_refs(l, item); |
54aa1f4d | 799 | out: |
5caf2a00 | 800 | btrfs_free_path(path); |
a28ec197 CM |
801 | return 0; |
802 | } | |
803 | ||
be20aa9d CM |
804 | u32 btrfs_count_snapshots_in_path(struct btrfs_root *root, |
805 | struct btrfs_path *count_path, | |
a68d5933 | 806 | u64 expected_owner, |
be20aa9d CM |
807 | u64 first_extent) |
808 | { | |
809 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
810 | struct btrfs_path *path; | |
811 | u64 bytenr; | |
812 | u64 found_objectid; | |
a68d5933 | 813 | u64 found_owner; |
56b453c9 | 814 | u64 root_objectid = root->root_key.objectid; |
be20aa9d | 815 | u32 total_count = 0; |
bbaf549e | 816 | u32 extent_refs; |
be20aa9d | 817 | u32 cur_count; |
be20aa9d CM |
818 | u32 nritems; |
819 | int ret; | |
820 | struct btrfs_key key; | |
821 | struct btrfs_key found_key; | |
822 | struct extent_buffer *l; | |
823 | struct btrfs_extent_item *item; | |
824 | struct btrfs_extent_ref *ref_item; | |
825 | int level = -1; | |
826 | ||
925baedd CM |
827 | /* FIXME, needs locking */ |
828 | BUG(); | |
829 | ||
830 | mutex_lock(&root->fs_info->alloc_mutex); | |
be20aa9d CM |
831 | path = btrfs_alloc_path(); |
832 | again: | |
833 | if (level == -1) | |
834 | bytenr = first_extent; | |
835 | else | |
836 | bytenr = count_path->nodes[level]->start; | |
837 | ||
838 | cur_count = 0; | |
839 | key.objectid = bytenr; | |
840 | key.offset = 0; | |
841 | ||
842 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
843 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
844 | if (ret < 0) | |
845 | goto out; | |
846 | BUG_ON(ret == 0); | |
847 | ||
848 | l = path->nodes[0]; | |
849 | btrfs_item_key_to_cpu(l, &found_key, path->slots[0]); | |
850 | ||
851 | if (found_key.objectid != bytenr || | |
852 | found_key.type != BTRFS_EXTENT_ITEM_KEY) { | |
853 | goto out; | |
854 | } | |
855 | ||
856 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); | |
bbaf549e | 857 | extent_refs = btrfs_extent_refs(l, item); |
be20aa9d | 858 | while (1) { |
bd09835d | 859 | l = path->nodes[0]; |
be20aa9d CM |
860 | nritems = btrfs_header_nritems(l); |
861 | if (path->slots[0] >= nritems) { | |
862 | ret = btrfs_next_leaf(extent_root, path); | |
863 | if (ret == 0) | |
864 | continue; | |
865 | break; | |
866 | } | |
867 | btrfs_item_key_to_cpu(l, &found_key, path->slots[0]); | |
868 | if (found_key.objectid != bytenr) | |
869 | break; | |
bd09835d | 870 | |
be20aa9d CM |
871 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
872 | path->slots[0]++; | |
873 | continue; | |
874 | } | |
875 | ||
876 | cur_count++; | |
877 | ref_item = btrfs_item_ptr(l, path->slots[0], | |
878 | struct btrfs_extent_ref); | |
879 | found_objectid = btrfs_ref_root(l, ref_item); | |
880 | ||
56b453c9 CM |
881 | if (found_objectid != root_objectid) { |
882 | total_count = 2; | |
4313b399 | 883 | goto out; |
56b453c9 | 884 | } |
a68d5933 CM |
885 | if (level == -1) { |
886 | found_owner = btrfs_ref_objectid(l, ref_item); | |
887 | if (found_owner != expected_owner) { | |
888 | total_count = 2; | |
889 | goto out; | |
890 | } | |
bbaf549e CM |
891 | /* |
892 | * nasty. we don't count a reference held by | |
893 | * the running transaction. This allows nodatacow | |
894 | * to avoid cow most of the time | |
895 | */ | |
896 | if (found_owner >= BTRFS_FIRST_FREE_OBJECTID && | |
897 | btrfs_ref_generation(l, ref_item) == | |
898 | root->fs_info->generation) { | |
899 | extent_refs--; | |
900 | } | |
a68d5933 | 901 | } |
56b453c9 | 902 | total_count = 1; |
be20aa9d CM |
903 | path->slots[0]++; |
904 | } | |
bbaf549e CM |
905 | /* |
906 | * if there is more than one reference against a data extent, | |
907 | * we have to assume the other ref is another snapshot | |
908 | */ | |
909 | if (level == -1 && extent_refs > 1) { | |
910 | total_count = 2; | |
911 | goto out; | |
912 | } | |
be20aa9d CM |
913 | if (cur_count == 0) { |
914 | total_count = 0; | |
915 | goto out; | |
916 | } | |
be20aa9d CM |
917 | if (level >= 0 && root->node == count_path->nodes[level]) |
918 | goto out; | |
919 | level++; | |
920 | btrfs_release_path(root, path); | |
921 | goto again; | |
922 | ||
923 | out: | |
924 | btrfs_free_path(path); | |
925baedd | 925 | mutex_unlock(&root->fs_info->alloc_mutex); |
be20aa9d | 926 | return total_count; |
be20aa9d | 927 | } |
c5739bba | 928 | |
e089f05c | 929 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5f39d397 | 930 | struct extent_buffer *buf) |
02217ed2 | 931 | { |
db94535d | 932 | u64 bytenr; |
5f39d397 CM |
933 | u32 nritems; |
934 | struct btrfs_key key; | |
6407bf6d | 935 | struct btrfs_file_extent_item *fi; |
02217ed2 | 936 | int i; |
db94535d | 937 | int level; |
6407bf6d | 938 | int ret; |
54aa1f4d | 939 | int faili; |
a28ec197 | 940 | |
3768f368 | 941 | if (!root->ref_cows) |
a28ec197 | 942 | return 0; |
5f39d397 | 943 | |
db94535d | 944 | level = btrfs_header_level(buf); |
5f39d397 CM |
945 | nritems = btrfs_header_nritems(buf); |
946 | for (i = 0; i < nritems; i++) { | |
e34a5b4f | 947 | cond_resched(); |
db94535d CM |
948 | if (level == 0) { |
949 | u64 disk_bytenr; | |
5f39d397 CM |
950 | btrfs_item_key_to_cpu(buf, &key, i); |
951 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d | 952 | continue; |
5f39d397 | 953 | fi = btrfs_item_ptr(buf, i, |
6407bf6d | 954 | struct btrfs_file_extent_item); |
5f39d397 | 955 | if (btrfs_file_extent_type(buf, fi) == |
236454df CM |
956 | BTRFS_FILE_EXTENT_INLINE) |
957 | continue; | |
db94535d CM |
958 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
959 | if (disk_bytenr == 0) | |
3a686375 | 960 | continue; |
4a096752 CM |
961 | |
962 | mutex_lock(&root->fs_info->alloc_mutex); | |
925baedd | 963 | ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr, |
7bb86316 CM |
964 | btrfs_file_extent_disk_num_bytes(buf, fi), |
965 | root->root_key.objectid, trans->transid, | |
966 | key.objectid, key.offset); | |
4a096752 | 967 | mutex_unlock(&root->fs_info->alloc_mutex); |
54aa1f4d CM |
968 | if (ret) { |
969 | faili = i; | |
4a096752 | 970 | WARN_ON(1); |
54aa1f4d CM |
971 | goto fail; |
972 | } | |
6407bf6d | 973 | } else { |
db94535d | 974 | bytenr = btrfs_node_blockptr(buf, i); |
6caab489 | 975 | btrfs_node_key_to_cpu(buf, &key, i); |
4a096752 CM |
976 | |
977 | mutex_lock(&root->fs_info->alloc_mutex); | |
925baedd | 978 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, |
7bb86316 CM |
979 | btrfs_level_size(root, level - 1), |
980 | root->root_key.objectid, | |
f6dbff55 CM |
981 | trans->transid, |
982 | level - 1, key.objectid); | |
4a096752 | 983 | mutex_unlock(&root->fs_info->alloc_mutex); |
54aa1f4d CM |
984 | if (ret) { |
985 | faili = i; | |
4a096752 | 986 | WARN_ON(1); |
54aa1f4d CM |
987 | goto fail; |
988 | } | |
6407bf6d | 989 | } |
02217ed2 CM |
990 | } |
991 | return 0; | |
54aa1f4d | 992 | fail: |
ccd467d6 | 993 | WARN_ON(1); |
7bb86316 | 994 | #if 0 |
54aa1f4d | 995 | for (i =0; i < faili; i++) { |
db94535d CM |
996 | if (level == 0) { |
997 | u64 disk_bytenr; | |
5f39d397 CM |
998 | btrfs_item_key_to_cpu(buf, &key, i); |
999 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
54aa1f4d | 1000 | continue; |
5f39d397 | 1001 | fi = btrfs_item_ptr(buf, i, |
54aa1f4d | 1002 | struct btrfs_file_extent_item); |
5f39d397 | 1003 | if (btrfs_file_extent_type(buf, fi) == |
54aa1f4d CM |
1004 | BTRFS_FILE_EXTENT_INLINE) |
1005 | continue; | |
db94535d CM |
1006 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
1007 | if (disk_bytenr == 0) | |
54aa1f4d | 1008 | continue; |
db94535d CM |
1009 | err = btrfs_free_extent(trans, root, disk_bytenr, |
1010 | btrfs_file_extent_disk_num_bytes(buf, | |
5f39d397 | 1011 | fi), 0); |
54aa1f4d CM |
1012 | BUG_ON(err); |
1013 | } else { | |
db94535d CM |
1014 | bytenr = btrfs_node_blockptr(buf, i); |
1015 | err = btrfs_free_extent(trans, root, bytenr, | |
1016 | btrfs_level_size(root, level - 1), 0); | |
54aa1f4d CM |
1017 | BUG_ON(err); |
1018 | } | |
1019 | } | |
7bb86316 | 1020 | #endif |
54aa1f4d | 1021 | return ret; |
02217ed2 CM |
1022 | } |
1023 | ||
9078a3e1 CM |
1024 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
1025 | struct btrfs_root *root, | |
1026 | struct btrfs_path *path, | |
1027 | struct btrfs_block_group_cache *cache) | |
1028 | { | |
1029 | int ret; | |
1030 | int pending_ret; | |
1031 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
5f39d397 CM |
1032 | unsigned long bi; |
1033 | struct extent_buffer *leaf; | |
9078a3e1 | 1034 | |
9078a3e1 | 1035 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
1036 | if (ret < 0) |
1037 | goto fail; | |
9078a3e1 | 1038 | BUG_ON(ret); |
5f39d397 CM |
1039 | |
1040 | leaf = path->nodes[0]; | |
1041 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1042 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
1043 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 1044 | btrfs_release_path(extent_root, path); |
54aa1f4d | 1045 | fail: |
9078a3e1 CM |
1046 | finish_current_insert(trans, extent_root); |
1047 | pending_ret = del_pending_extents(trans, extent_root); | |
1048 | if (ret) | |
1049 | return ret; | |
1050 | if (pending_ret) | |
1051 | return pending_ret; | |
1052 | return 0; | |
1053 | ||
1054 | } | |
1055 | ||
96b5179d CM |
1056 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
1057 | struct btrfs_root *root) | |
9078a3e1 | 1058 | { |
d1310b2e | 1059 | struct extent_io_tree *block_group_cache; |
96b5179d | 1060 | struct btrfs_block_group_cache *cache; |
9078a3e1 CM |
1061 | int ret; |
1062 | int err = 0; | |
1063 | int werr = 0; | |
9078a3e1 | 1064 | struct btrfs_path *path; |
96b5179d CM |
1065 | u64 last = 0; |
1066 | u64 start; | |
1067 | u64 end; | |
1068 | u64 ptr; | |
9078a3e1 | 1069 | |
96b5179d | 1070 | block_group_cache = &root->fs_info->block_group_cache; |
9078a3e1 CM |
1071 | path = btrfs_alloc_path(); |
1072 | if (!path) | |
1073 | return -ENOMEM; | |
1074 | ||
925baedd | 1075 | mutex_lock(&root->fs_info->alloc_mutex); |
9078a3e1 | 1076 | while(1) { |
96b5179d CM |
1077 | ret = find_first_extent_bit(block_group_cache, last, |
1078 | &start, &end, BLOCK_GROUP_DIRTY); | |
1079 | if (ret) | |
9078a3e1 | 1080 | break; |
54aa1f4d | 1081 | |
96b5179d CM |
1082 | last = end + 1; |
1083 | ret = get_state_private(block_group_cache, start, &ptr); | |
1084 | if (ret) | |
1085 | break; | |
ae2f5411 | 1086 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
96b5179d CM |
1087 | err = write_one_cache_group(trans, root, |
1088 | path, cache); | |
1089 | /* | |
1090 | * if we fail to write the cache group, we want | |
1091 | * to keep it marked dirty in hopes that a later | |
1092 | * write will work | |
1093 | */ | |
1094 | if (err) { | |
1095 | werr = err; | |
1096 | continue; | |
9078a3e1 | 1097 | } |
96b5179d CM |
1098 | clear_extent_bits(block_group_cache, start, end, |
1099 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 CM |
1100 | } |
1101 | btrfs_free_path(path); | |
925baedd | 1102 | mutex_unlock(&root->fs_info->alloc_mutex); |
9078a3e1 CM |
1103 | return werr; |
1104 | } | |
1105 | ||
6324fbf3 CM |
1106 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
1107 | u64 flags) | |
1108 | { | |
1109 | struct list_head *head = &info->space_info; | |
1110 | struct list_head *cur; | |
1111 | struct btrfs_space_info *found; | |
1112 | list_for_each(cur, head) { | |
1113 | found = list_entry(cur, struct btrfs_space_info, list); | |
1114 | if (found->flags == flags) | |
1115 | return found; | |
1116 | } | |
1117 | return NULL; | |
1118 | ||
1119 | } | |
1120 | ||
593060d7 CM |
1121 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
1122 | u64 total_bytes, u64 bytes_used, | |
1123 | struct btrfs_space_info **space_info) | |
1124 | { | |
1125 | struct btrfs_space_info *found; | |
1126 | ||
1127 | found = __find_space_info(info, flags); | |
1128 | if (found) { | |
1129 | found->total_bytes += total_bytes; | |
1130 | found->bytes_used += bytes_used; | |
8f18cf13 | 1131 | found->full = 0; |
593060d7 CM |
1132 | WARN_ON(found->total_bytes < found->bytes_used); |
1133 | *space_info = found; | |
1134 | return 0; | |
1135 | } | |
1136 | found = kmalloc(sizeof(*found), GFP_NOFS); | |
1137 | if (!found) | |
1138 | return -ENOMEM; | |
1139 | ||
1140 | list_add(&found->list, &info->space_info); | |
1141 | found->flags = flags; | |
1142 | found->total_bytes = total_bytes; | |
1143 | found->bytes_used = bytes_used; | |
1144 | found->bytes_pinned = 0; | |
1145 | found->full = 0; | |
0ef3e66b | 1146 | found->force_alloc = 0; |
593060d7 CM |
1147 | *space_info = found; |
1148 | return 0; | |
1149 | } | |
1150 | ||
8790d502 CM |
1151 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
1152 | { | |
1153 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 1154 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 1155 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 1156 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
1157 | if (extra_flags) { |
1158 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1159 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1160 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1161 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1162 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1163 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1164 | } | |
1165 | } | |
593060d7 | 1166 | |
a061fc8d | 1167 | static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 1168 | { |
a061fc8d CM |
1169 | u64 num_devices = root->fs_info->fs_devices->num_devices; |
1170 | ||
1171 | if (num_devices == 1) | |
1172 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
1173 | if (num_devices < 4) | |
1174 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
1175 | ||
ec44a35c CM |
1176 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
1177 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 1178 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 1179 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 1180 | } |
ec44a35c CM |
1181 | |
1182 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 1183 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 1184 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 1185 | } |
ec44a35c CM |
1186 | |
1187 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
1188 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
1189 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
1190 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
1191 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
1192 | return flags; | |
1193 | } | |
1194 | ||
6324fbf3 CM |
1195 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
1196 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 1197 | u64 flags, int force) |
6324fbf3 CM |
1198 | { |
1199 | struct btrfs_space_info *space_info; | |
1200 | u64 thresh; | |
1201 | u64 start; | |
1202 | u64 num_bytes; | |
1203 | int ret; | |
1204 | ||
a061fc8d | 1205 | flags = reduce_alloc_profile(extent_root, flags); |
ec44a35c | 1206 | |
6324fbf3 | 1207 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
1208 | if (!space_info) { |
1209 | ret = update_space_info(extent_root->fs_info, flags, | |
1210 | 0, 0, &space_info); | |
1211 | BUG_ON(ret); | |
1212 | } | |
6324fbf3 CM |
1213 | BUG_ON(!space_info); |
1214 | ||
0ef3e66b CM |
1215 | if (space_info->force_alloc) { |
1216 | force = 1; | |
1217 | space_info->force_alloc = 0; | |
1218 | } | |
6324fbf3 | 1219 | if (space_info->full) |
925baedd | 1220 | goto out; |
6324fbf3 | 1221 | |
8790d502 | 1222 | thresh = div_factor(space_info->total_bytes, 6); |
0ef3e66b CM |
1223 | if (!force && |
1224 | (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) < | |
6324fbf3 | 1225 | thresh) |
925baedd | 1226 | goto out; |
6324fbf3 | 1227 | |
925baedd | 1228 | mutex_lock(&extent_root->fs_info->chunk_mutex); |
6324fbf3 CM |
1229 | ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags); |
1230 | if (ret == -ENOSPC) { | |
1231 | printk("space info full %Lu\n", flags); | |
1232 | space_info->full = 1; | |
a74a4b97 | 1233 | goto out_unlock; |
6324fbf3 | 1234 | } |
6324fbf3 CM |
1235 | BUG_ON(ret); |
1236 | ||
1237 | ret = btrfs_make_block_group(trans, extent_root, 0, flags, | |
e17cade2 | 1238 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes); |
6324fbf3 | 1239 | BUG_ON(ret); |
a74a4b97 | 1240 | out_unlock: |
333db94c | 1241 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
a74a4b97 | 1242 | out: |
6324fbf3 CM |
1243 | return 0; |
1244 | } | |
1245 | ||
9078a3e1 CM |
1246 | static int update_block_group(struct btrfs_trans_handle *trans, |
1247 | struct btrfs_root *root, | |
db94535d | 1248 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 1249 | int mark_free) |
9078a3e1 CM |
1250 | { |
1251 | struct btrfs_block_group_cache *cache; | |
1252 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 1253 | u64 total = num_bytes; |
9078a3e1 | 1254 | u64 old_val; |
db94535d | 1255 | u64 byte_in_group; |
96b5179d CM |
1256 | u64 start; |
1257 | u64 end; | |
3e1ad54f | 1258 | |
7d9eb12c | 1259 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
9078a3e1 | 1260 | while(total) { |
db94535d | 1261 | cache = btrfs_lookup_block_group(info, bytenr); |
3e1ad54f | 1262 | if (!cache) { |
9078a3e1 | 1263 | return -1; |
cd1bc465 | 1264 | } |
db94535d CM |
1265 | byte_in_group = bytenr - cache->key.objectid; |
1266 | WARN_ON(byte_in_group > cache->key.offset); | |
96b5179d CM |
1267 | start = cache->key.objectid; |
1268 | end = start + cache->key.offset - 1; | |
1269 | set_extent_bits(&info->block_group_cache, start, end, | |
1270 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 | 1271 | |
c286ac48 | 1272 | spin_lock(&cache->lock); |
9078a3e1 | 1273 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 1274 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 1275 | if (alloc) { |
db94535d | 1276 | old_val += num_bytes; |
6324fbf3 | 1277 | cache->space_info->bytes_used += num_bytes; |
c286ac48 CM |
1278 | btrfs_set_block_group_used(&cache->item, old_val); |
1279 | spin_unlock(&cache->lock); | |
cd1bc465 | 1280 | } else { |
db94535d | 1281 | old_val -= num_bytes; |
6324fbf3 | 1282 | cache->space_info->bytes_used -= num_bytes; |
c286ac48 CM |
1283 | btrfs_set_block_group_used(&cache->item, old_val); |
1284 | spin_unlock(&cache->lock); | |
f510cfec CM |
1285 | if (mark_free) { |
1286 | set_extent_dirty(&info->free_space_cache, | |
db94535d | 1287 | bytenr, bytenr + num_bytes - 1, |
f510cfec | 1288 | GFP_NOFS); |
e37c9e69 | 1289 | } |
cd1bc465 | 1290 | } |
db94535d CM |
1291 | total -= num_bytes; |
1292 | bytenr += num_bytes; | |
9078a3e1 CM |
1293 | } |
1294 | return 0; | |
1295 | } | |
6324fbf3 | 1296 | |
a061fc8d CM |
1297 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
1298 | { | |
1299 | u64 start; | |
1300 | u64 end; | |
1301 | int ret; | |
1302 | ret = find_first_extent_bit(&root->fs_info->block_group_cache, | |
1303 | search_start, &start, &end, | |
1304 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | | |
1305 | BLOCK_GROUP_SYSTEM); | |
1306 | if (ret) | |
1307 | return 0; | |
1308 | return start; | |
1309 | } | |
1310 | ||
1311 | ||
324ae4df Y |
1312 | static int update_pinned_extents(struct btrfs_root *root, |
1313 | u64 bytenr, u64 num, int pin) | |
1314 | { | |
1315 | u64 len; | |
1316 | struct btrfs_block_group_cache *cache; | |
1317 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1318 | ||
7d9eb12c | 1319 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
324ae4df Y |
1320 | if (pin) { |
1321 | set_extent_dirty(&fs_info->pinned_extents, | |
1322 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1323 | } else { | |
1324 | clear_extent_dirty(&fs_info->pinned_extents, | |
1325 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1326 | } | |
1327 | while (num > 0) { | |
1328 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
a061fc8d CM |
1329 | if (!cache) { |
1330 | u64 first = first_logical_byte(root, bytenr); | |
1331 | WARN_ON(first < bytenr); | |
1332 | len = min(first - bytenr, num); | |
1333 | } else { | |
1334 | len = min(num, cache->key.offset - | |
1335 | (bytenr - cache->key.objectid)); | |
1336 | } | |
324ae4df | 1337 | if (pin) { |
a061fc8d | 1338 | if (cache) { |
c286ac48 | 1339 | spin_lock(&cache->lock); |
a061fc8d CM |
1340 | cache->pinned += len; |
1341 | cache->space_info->bytes_pinned += len; | |
c286ac48 | 1342 | spin_unlock(&cache->lock); |
a061fc8d | 1343 | } |
324ae4df Y |
1344 | fs_info->total_pinned += len; |
1345 | } else { | |
a061fc8d | 1346 | if (cache) { |
c286ac48 | 1347 | spin_lock(&cache->lock); |
a061fc8d CM |
1348 | cache->pinned -= len; |
1349 | cache->space_info->bytes_pinned -= len; | |
c286ac48 | 1350 | spin_unlock(&cache->lock); |
a061fc8d | 1351 | } |
324ae4df Y |
1352 | fs_info->total_pinned -= len; |
1353 | } | |
1354 | bytenr += len; | |
1355 | num -= len; | |
1356 | } | |
1357 | return 0; | |
1358 | } | |
9078a3e1 | 1359 | |
d1310b2e | 1360 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) |
ccd467d6 | 1361 | { |
ccd467d6 | 1362 | u64 last = 0; |
1a5bc167 CM |
1363 | u64 start; |
1364 | u64 end; | |
d1310b2e | 1365 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; |
ccd467d6 | 1366 | int ret; |
ccd467d6 CM |
1367 | |
1368 | while(1) { | |
1a5bc167 CM |
1369 | ret = find_first_extent_bit(pinned_extents, last, |
1370 | &start, &end, EXTENT_DIRTY); | |
1371 | if (ret) | |
ccd467d6 | 1372 | break; |
1a5bc167 CM |
1373 | set_extent_dirty(copy, start, end, GFP_NOFS); |
1374 | last = end + 1; | |
ccd467d6 CM |
1375 | } |
1376 | return 0; | |
1377 | } | |
1378 | ||
1379 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
1380 | struct btrfs_root *root, | |
d1310b2e | 1381 | struct extent_io_tree *unpin) |
a28ec197 | 1382 | { |
1a5bc167 CM |
1383 | u64 start; |
1384 | u64 end; | |
a28ec197 | 1385 | int ret; |
d1310b2e | 1386 | struct extent_io_tree *free_space_cache; |
f510cfec | 1387 | free_space_cache = &root->fs_info->free_space_cache; |
a28ec197 | 1388 | |
925baedd | 1389 | mutex_lock(&root->fs_info->alloc_mutex); |
a28ec197 | 1390 | while(1) { |
1a5bc167 CM |
1391 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
1392 | EXTENT_DIRTY); | |
1393 | if (ret) | |
a28ec197 | 1394 | break; |
324ae4df | 1395 | update_pinned_extents(root, start, end + 1 - start, 0); |
1a5bc167 CM |
1396 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
1397 | set_extent_dirty(free_space_cache, start, end, GFP_NOFS); | |
c286ac48 CM |
1398 | if (need_resched()) { |
1399 | mutex_unlock(&root->fs_info->alloc_mutex); | |
1400 | cond_resched(); | |
1401 | mutex_lock(&root->fs_info->alloc_mutex); | |
1402 | } | |
a28ec197 | 1403 | } |
925baedd | 1404 | mutex_unlock(&root->fs_info->alloc_mutex); |
a28ec197 CM |
1405 | return 0; |
1406 | } | |
1407 | ||
98ed5174 CM |
1408 | static int finish_current_insert(struct btrfs_trans_handle *trans, |
1409 | struct btrfs_root *extent_root) | |
037e6390 | 1410 | { |
7bb86316 CM |
1411 | u64 start; |
1412 | u64 end; | |
1413 | struct btrfs_fs_info *info = extent_root->fs_info; | |
d8d5f3e1 | 1414 | struct extent_buffer *eb; |
7bb86316 | 1415 | struct btrfs_path *path; |
e2fa7227 | 1416 | struct btrfs_key ins; |
d8d5f3e1 | 1417 | struct btrfs_disk_key first; |
234b63a0 | 1418 | struct btrfs_extent_item extent_item; |
037e6390 | 1419 | int ret; |
d8d5f3e1 | 1420 | int level; |
1a5bc167 | 1421 | int err = 0; |
037e6390 | 1422 | |
7d9eb12c | 1423 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); |
5f39d397 | 1424 | btrfs_set_stack_extent_refs(&extent_item, 1); |
62e2749e | 1425 | btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY); |
7bb86316 | 1426 | path = btrfs_alloc_path(); |
037e6390 | 1427 | |
26b8003f | 1428 | while(1) { |
1a5bc167 CM |
1429 | ret = find_first_extent_bit(&info->extent_ins, 0, &start, |
1430 | &end, EXTENT_LOCKED); | |
1431 | if (ret) | |
26b8003f CM |
1432 | break; |
1433 | ||
1a5bc167 CM |
1434 | ins.objectid = start; |
1435 | ins.offset = end + 1 - start; | |
1436 | err = btrfs_insert_item(trans, extent_root, &ins, | |
1437 | &extent_item, sizeof(extent_item)); | |
1438 | clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED, | |
1439 | GFP_NOFS); | |
c286ac48 CM |
1440 | |
1441 | eb = btrfs_find_tree_block(extent_root, ins.objectid, | |
1442 | ins.offset); | |
1443 | ||
1444 | if (!btrfs_buffer_uptodate(eb, trans->transid)) { | |
1445 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
1446 | btrfs_read_buffer(eb, trans->transid); | |
1447 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
1448 | } | |
1449 | ||
925baedd | 1450 | btrfs_tree_lock(eb); |
d8d5f3e1 CM |
1451 | level = btrfs_header_level(eb); |
1452 | if (level == 0) { | |
1453 | btrfs_item_key(eb, &first, 0); | |
1454 | } else { | |
1455 | btrfs_node_key(eb, &first, 0); | |
1456 | } | |
925baedd CM |
1457 | btrfs_tree_unlock(eb); |
1458 | free_extent_buffer(eb); | |
1459 | /* | |
1460 | * the first key is just a hint, so the race we've created | |
1461 | * against reading it is fine | |
1462 | */ | |
7bb86316 CM |
1463 | err = btrfs_insert_extent_backref(trans, extent_root, path, |
1464 | start, extent_root->root_key.objectid, | |
f6dbff55 CM |
1465 | 0, level, |
1466 | btrfs_disk_key_objectid(&first)); | |
7bb86316 | 1467 | BUG_ON(err); |
c286ac48 CM |
1468 | if (need_resched()) { |
1469 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
1470 | cond_resched(); | |
1471 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
1472 | } | |
037e6390 | 1473 | } |
7bb86316 | 1474 | btrfs_free_path(path); |
037e6390 CM |
1475 | return 0; |
1476 | } | |
1477 | ||
db94535d CM |
1478 | static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes, |
1479 | int pending) | |
e20d96d6 | 1480 | { |
1a5bc167 | 1481 | int err = 0; |
8ef97622 | 1482 | |
7d9eb12c | 1483 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
f4b9aa8d | 1484 | if (!pending) { |
925baedd | 1485 | struct extent_buffer *buf; |
db94535d | 1486 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); |
5f39d397 | 1487 | if (buf) { |
a61e6f29 | 1488 | if (btrfs_try_tree_lock(buf) && |
925baedd | 1489 | btrfs_buffer_uptodate(buf, 0)) { |
2c90e5d6 CM |
1490 | u64 transid = |
1491 | root->fs_info->running_transaction->transid; | |
dc17ff8f CM |
1492 | u64 header_transid = |
1493 | btrfs_header_generation(buf); | |
6bc34676 CM |
1494 | if (header_transid == transid && |
1495 | !btrfs_header_flag(buf, | |
1496 | BTRFS_HEADER_FLAG_WRITTEN)) { | |
55c69072 | 1497 | clean_tree_block(NULL, root, buf); |
925baedd | 1498 | btrfs_tree_unlock(buf); |
5f39d397 | 1499 | free_extent_buffer(buf); |
c549228f | 1500 | return 1; |
2c90e5d6 | 1501 | } |
925baedd | 1502 | btrfs_tree_unlock(buf); |
f4b9aa8d | 1503 | } |
5f39d397 | 1504 | free_extent_buffer(buf); |
8ef97622 | 1505 | } |
324ae4df | 1506 | update_pinned_extents(root, bytenr, num_bytes, 1); |
f4b9aa8d | 1507 | } else { |
1a5bc167 | 1508 | set_extent_bits(&root->fs_info->pending_del, |
db94535d CM |
1509 | bytenr, bytenr + num_bytes - 1, |
1510 | EXTENT_LOCKED, GFP_NOFS); | |
f4b9aa8d | 1511 | } |
be744175 | 1512 | BUG_ON(err < 0); |
e20d96d6 CM |
1513 | return 0; |
1514 | } | |
1515 | ||
fec577fb | 1516 | /* |
a28ec197 | 1517 | * remove an extent from the root, returns 0 on success |
fec577fb | 1518 | */ |
e089f05c | 1519 | static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
7bb86316 CM |
1520 | *root, u64 bytenr, u64 num_bytes, |
1521 | u64 root_objectid, u64 ref_generation, | |
1522 | u64 owner_objectid, u64 owner_offset, int pin, | |
e37c9e69 | 1523 | int mark_free) |
a28ec197 | 1524 | { |
5caf2a00 | 1525 | struct btrfs_path *path; |
e2fa7227 | 1526 | struct btrfs_key key; |
1261ec42 CM |
1527 | struct btrfs_fs_info *info = root->fs_info; |
1528 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 1529 | struct extent_buffer *leaf; |
a28ec197 | 1530 | int ret; |
952fccac CM |
1531 | int extent_slot = 0; |
1532 | int found_extent = 0; | |
1533 | int num_to_del = 1; | |
234b63a0 | 1534 | struct btrfs_extent_item *ei; |
cf27e1ee | 1535 | u32 refs; |
037e6390 | 1536 | |
7d9eb12c | 1537 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
db94535d | 1538 | key.objectid = bytenr; |
62e2749e | 1539 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 1540 | key.offset = num_bytes; |
5caf2a00 | 1541 | path = btrfs_alloc_path(); |
54aa1f4d CM |
1542 | if (!path) |
1543 | return -ENOMEM; | |
5f26f772 | 1544 | |
3c12ac72 | 1545 | path->reada = 1; |
7bb86316 CM |
1546 | ret = lookup_extent_backref(trans, extent_root, path, |
1547 | bytenr, root_objectid, | |
1548 | ref_generation, | |
1549 | owner_objectid, owner_offset, 1); | |
1550 | if (ret == 0) { | |
952fccac CM |
1551 | struct btrfs_key found_key; |
1552 | extent_slot = path->slots[0]; | |
1553 | while(extent_slot > 0) { | |
1554 | extent_slot--; | |
1555 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1556 | extent_slot); | |
1557 | if (found_key.objectid != bytenr) | |
1558 | break; | |
1559 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && | |
1560 | found_key.offset == num_bytes) { | |
1561 | found_extent = 1; | |
1562 | break; | |
1563 | } | |
1564 | if (path->slots[0] - extent_slot > 5) | |
1565 | break; | |
1566 | } | |
1567 | if (!found_extent) | |
1568 | ret = btrfs_del_item(trans, extent_root, path); | |
7bb86316 CM |
1569 | } else { |
1570 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
1571 | WARN_ON(1); | |
1572 | printk("Unable to find ref byte nr %Lu root %Lu " | |
1573 | " gen %Lu owner %Lu offset %Lu\n", bytenr, | |
1574 | root_objectid, ref_generation, owner_objectid, | |
1575 | owner_offset); | |
1576 | } | |
952fccac CM |
1577 | if (!found_extent) { |
1578 | btrfs_release_path(extent_root, path); | |
1579 | ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); | |
1580 | if (ret < 0) | |
1581 | return ret; | |
1582 | BUG_ON(ret); | |
1583 | extent_slot = path->slots[0]; | |
1584 | } | |
5f39d397 CM |
1585 | |
1586 | leaf = path->nodes[0]; | |
952fccac | 1587 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 1588 | struct btrfs_extent_item); |
5f39d397 CM |
1589 | refs = btrfs_extent_refs(leaf, ei); |
1590 | BUG_ON(refs == 0); | |
1591 | refs -= 1; | |
1592 | btrfs_set_extent_refs(leaf, ei, refs); | |
952fccac | 1593 | |
5f39d397 CM |
1594 | btrfs_mark_buffer_dirty(leaf); |
1595 | ||
952fccac CM |
1596 | if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { |
1597 | /* if the back ref and the extent are next to each other | |
1598 | * they get deleted below in one shot | |
1599 | */ | |
1600 | path->slots[0] = extent_slot; | |
1601 | num_to_del = 2; | |
1602 | } else if (found_extent) { | |
1603 | /* otherwise delete the extent back ref */ | |
1604 | ret = btrfs_del_item(trans, extent_root, path); | |
1605 | BUG_ON(ret); | |
1606 | /* if refs are 0, we need to setup the path for deletion */ | |
1607 | if (refs == 0) { | |
1608 | btrfs_release_path(extent_root, path); | |
1609 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
1610 | -1, 1); | |
1611 | if (ret < 0) | |
1612 | return ret; | |
1613 | BUG_ON(ret); | |
1614 | } | |
1615 | } | |
1616 | ||
cf27e1ee | 1617 | if (refs == 0) { |
db94535d CM |
1618 | u64 super_used; |
1619 | u64 root_used; | |
78fae27e CM |
1620 | |
1621 | if (pin) { | |
db94535d | 1622 | ret = pin_down_bytes(root, bytenr, num_bytes, 0); |
c549228f Y |
1623 | if (ret > 0) |
1624 | mark_free = 1; | |
1625 | BUG_ON(ret < 0); | |
78fae27e CM |
1626 | } |
1627 | ||
58176a96 | 1628 | /* block accounting for super block */ |
a2135011 | 1629 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
1630 | super_used = btrfs_super_bytes_used(&info->super_copy); |
1631 | btrfs_set_super_bytes_used(&info->super_copy, | |
1632 | super_used - num_bytes); | |
a2135011 | 1633 | spin_unlock_irq(&info->delalloc_lock); |
58176a96 JB |
1634 | |
1635 | /* block accounting for root item */ | |
db94535d | 1636 | root_used = btrfs_root_used(&root->root_item); |
5f39d397 | 1637 | btrfs_set_root_used(&root->root_item, |
db94535d | 1638 | root_used - num_bytes); |
952fccac CM |
1639 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
1640 | num_to_del); | |
54aa1f4d CM |
1641 | if (ret) { |
1642 | return ret; | |
1643 | } | |
db94535d | 1644 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
0b86a832 | 1645 | mark_free); |
9078a3e1 | 1646 | BUG_ON(ret); |
a28ec197 | 1647 | } |
5caf2a00 | 1648 | btrfs_free_path(path); |
e089f05c | 1649 | finish_current_insert(trans, extent_root); |
a28ec197 CM |
1650 | return ret; |
1651 | } | |
1652 | ||
a28ec197 CM |
1653 | /* |
1654 | * find all the blocks marked as pending in the radix tree and remove | |
1655 | * them from the extent map | |
1656 | */ | |
e089f05c CM |
1657 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
1658 | btrfs_root *extent_root) | |
a28ec197 CM |
1659 | { |
1660 | int ret; | |
e20d96d6 | 1661 | int err = 0; |
1a5bc167 CM |
1662 | u64 start; |
1663 | u64 end; | |
d1310b2e CM |
1664 | struct extent_io_tree *pending_del; |
1665 | struct extent_io_tree *pinned_extents; | |
8ef97622 | 1666 | |
7d9eb12c | 1667 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); |
1a5bc167 CM |
1668 | pending_del = &extent_root->fs_info->pending_del; |
1669 | pinned_extents = &extent_root->fs_info->pinned_extents; | |
a28ec197 CM |
1670 | |
1671 | while(1) { | |
1a5bc167 CM |
1672 | ret = find_first_extent_bit(pending_del, 0, &start, &end, |
1673 | EXTENT_LOCKED); | |
1674 | if (ret) | |
a28ec197 | 1675 | break; |
1a5bc167 CM |
1676 | clear_extent_bits(pending_del, start, end, EXTENT_LOCKED, |
1677 | GFP_NOFS); | |
c286ac48 CM |
1678 | if (!test_range_bit(&extent_root->fs_info->extent_ins, |
1679 | start, end, EXTENT_LOCKED, 0)) { | |
1680 | update_pinned_extents(extent_root, start, | |
1681 | end + 1 - start, 1); | |
1682 | ret = __free_extent(trans, extent_root, | |
1683 | start, end + 1 - start, | |
1684 | extent_root->root_key.objectid, | |
1685 | 0, 0, 0, 0, 0); | |
1686 | } else { | |
1687 | clear_extent_bits(&extent_root->fs_info->extent_ins, | |
1688 | start, end, EXTENT_LOCKED, GFP_NOFS); | |
1689 | } | |
1a5bc167 CM |
1690 | if (ret) |
1691 | err = ret; | |
c286ac48 CM |
1692 | |
1693 | if (need_resched()) { | |
1694 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
1695 | cond_resched(); | |
1696 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
1697 | } | |
fec577fb | 1698 | } |
e20d96d6 | 1699 | return err; |
fec577fb CM |
1700 | } |
1701 | ||
1702 | /* | |
1703 | * remove an extent from the root, returns 0 on success | |
1704 | */ | |
925baedd CM |
1705 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
1706 | struct btrfs_root *root, u64 bytenr, | |
1707 | u64 num_bytes, u64 root_objectid, | |
1708 | u64 ref_generation, u64 owner_objectid, | |
1709 | u64 owner_offset, int pin) | |
fec577fb | 1710 | { |
9f5fae2f | 1711 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
fec577fb CM |
1712 | int pending_ret; |
1713 | int ret; | |
a28ec197 | 1714 | |
db94535d | 1715 | WARN_ON(num_bytes < root->sectorsize); |
7bb86316 CM |
1716 | if (!root->ref_cows) |
1717 | ref_generation = 0; | |
1718 | ||
fec577fb | 1719 | if (root == extent_root) { |
db94535d | 1720 | pin_down_bytes(root, bytenr, num_bytes, 1); |
fec577fb CM |
1721 | return 0; |
1722 | } | |
7bb86316 CM |
1723 | ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid, |
1724 | ref_generation, owner_objectid, owner_offset, | |
1725 | pin, pin == 0); | |
ee6e6504 CM |
1726 | |
1727 | finish_current_insert(trans, root->fs_info->extent_root); | |
e20d96d6 | 1728 | pending_ret = del_pending_extents(trans, root->fs_info->extent_root); |
fec577fb CM |
1729 | return ret ? ret : pending_ret; |
1730 | } | |
1731 | ||
925baedd CM |
1732 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
1733 | struct btrfs_root *root, u64 bytenr, | |
1734 | u64 num_bytes, u64 root_objectid, | |
1735 | u64 ref_generation, u64 owner_objectid, | |
1736 | u64 owner_offset, int pin) | |
1737 | { | |
1738 | int ret; | |
1739 | ||
1740 | maybe_lock_mutex(root); | |
1741 | ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1742 | root_objectid, ref_generation, | |
1743 | owner_objectid, owner_offset, pin); | |
1744 | maybe_unlock_mutex(root); | |
1745 | return ret; | |
1746 | } | |
1747 | ||
87ee04eb CM |
1748 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
1749 | { | |
1750 | u64 mask = ((u64)root->stripesize - 1); | |
1751 | u64 ret = (val + mask) & ~mask; | |
1752 | return ret; | |
1753 | } | |
1754 | ||
fec577fb CM |
1755 | /* |
1756 | * walks the btree of allocated extents and find a hole of a given size. | |
1757 | * The key ins is changed to record the hole: | |
1758 | * ins->objectid == block start | |
62e2749e | 1759 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
1760 | * ins->offset == number of blocks |
1761 | * Any available blocks before search_start are skipped. | |
1762 | */ | |
98ed5174 CM |
1763 | static int noinline find_free_extent(struct btrfs_trans_handle *trans, |
1764 | struct btrfs_root *orig_root, | |
1765 | u64 num_bytes, u64 empty_size, | |
1766 | u64 search_start, u64 search_end, | |
1767 | u64 hint_byte, struct btrfs_key *ins, | |
1768 | u64 exclude_start, u64 exclude_nr, | |
1769 | int data) | |
fec577fb | 1770 | { |
87ee04eb | 1771 | int ret; |
a061fc8d | 1772 | u64 orig_search_start; |
9f5fae2f | 1773 | struct btrfs_root * root = orig_root->fs_info->extent_root; |
f2458e1d | 1774 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 1775 | u64 total_needed = num_bytes; |
239b14b3 | 1776 | u64 *last_ptr = NULL; |
be08c1b9 | 1777 | struct btrfs_block_group_cache *block_group; |
be744175 | 1778 | int full_scan = 0; |
fbdc762b | 1779 | int wrapped = 0; |
0ef3e66b | 1780 | int chunk_alloc_done = 0; |
239b14b3 | 1781 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 1782 | int allowed_chunk_alloc = 0; |
fec577fb | 1783 | |
db94535d | 1784 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 CM |
1785 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
1786 | ||
0ef3e66b CM |
1787 | if (orig_root->ref_cows || empty_size) |
1788 | allowed_chunk_alloc = 1; | |
1789 | ||
239b14b3 CM |
1790 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
1791 | last_ptr = &root->fs_info->last_alloc; | |
8790d502 | 1792 | empty_cluster = 256 * 1024; |
239b14b3 CM |
1793 | } |
1794 | ||
1795 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { | |
1796 | last_ptr = &root->fs_info->last_data_alloc; | |
1797 | } | |
1798 | ||
1799 | if (last_ptr) { | |
1800 | if (*last_ptr) | |
1801 | hint_byte = *last_ptr; | |
1802 | else { | |
1803 | empty_size += empty_cluster; | |
1804 | } | |
1805 | } | |
1806 | ||
a061fc8d CM |
1807 | search_start = max(search_start, first_logical_byte(root, 0)); |
1808 | orig_search_start = search_start; | |
1809 | ||
7f93bf8d CM |
1810 | if (search_end == (u64)-1) |
1811 | search_end = btrfs_super_total_bytes(&info->super_copy); | |
0b86a832 | 1812 | |
db94535d | 1813 | if (hint_byte) { |
0ef3e66b | 1814 | block_group = btrfs_lookup_first_block_group(info, hint_byte); |
1a2b2ac7 CM |
1815 | if (!block_group) |
1816 | hint_byte = search_start; | |
c286ac48 | 1817 | block_group = btrfs_find_block_group(root, block_group, |
db94535d | 1818 | hint_byte, data, 1); |
239b14b3 CM |
1819 | if (last_ptr && *last_ptr == 0 && block_group) |
1820 | hint_byte = block_group->key.objectid; | |
be744175 | 1821 | } else { |
c286ac48 | 1822 | block_group = btrfs_find_block_group(root, |
1a2b2ac7 CM |
1823 | trans->block_group, |
1824 | search_start, data, 1); | |
be744175 | 1825 | } |
239b14b3 | 1826 | search_start = max(search_start, hint_byte); |
be744175 | 1827 | |
6702ed49 | 1828 | total_needed += empty_size; |
0b86a832 | 1829 | |
be744175 | 1830 | check_failed: |
70b043f0 | 1831 | if (!block_group) { |
0ef3e66b CM |
1832 | block_group = btrfs_lookup_first_block_group(info, |
1833 | search_start); | |
70b043f0 | 1834 | if (!block_group) |
0ef3e66b | 1835 | block_group = btrfs_lookup_first_block_group(info, |
70b043f0 CM |
1836 | orig_search_start); |
1837 | } | |
0ef3e66b CM |
1838 | if (full_scan && !chunk_alloc_done) { |
1839 | if (allowed_chunk_alloc) { | |
1840 | do_chunk_alloc(trans, root, | |
1841 | num_bytes + 2 * 1024 * 1024, data, 1); | |
1842 | allowed_chunk_alloc = 0; | |
1843 | } else if (block_group && block_group_bits(block_group, data)) { | |
1844 | block_group->space_info->force_alloc = 1; | |
1845 | } | |
1846 | chunk_alloc_done = 1; | |
1847 | } | |
0b86a832 CM |
1848 | ret = find_search_start(root, &block_group, &search_start, |
1849 | total_needed, data); | |
239b14b3 CM |
1850 | if (ret == -ENOSPC && last_ptr && *last_ptr) { |
1851 | *last_ptr = 0; | |
0ef3e66b CM |
1852 | block_group = btrfs_lookup_first_block_group(info, |
1853 | orig_search_start); | |
239b14b3 CM |
1854 | search_start = orig_search_start; |
1855 | ret = find_search_start(root, &block_group, &search_start, | |
1856 | total_needed, data); | |
1857 | } | |
1858 | if (ret == -ENOSPC) | |
1859 | goto enospc; | |
0b86a832 | 1860 | if (ret) |
d548ee51 | 1861 | goto error; |
e19caa5f | 1862 | |
239b14b3 CM |
1863 | if (last_ptr && *last_ptr && search_start != *last_ptr) { |
1864 | *last_ptr = 0; | |
1865 | if (!empty_size) { | |
1866 | empty_size += empty_cluster; | |
1867 | total_needed += empty_size; | |
1868 | } | |
0ef3e66b | 1869 | block_group = btrfs_lookup_first_block_group(info, |
239b14b3 CM |
1870 | orig_search_start); |
1871 | search_start = orig_search_start; | |
1872 | ret = find_search_start(root, &block_group, | |
1873 | &search_start, total_needed, data); | |
1874 | if (ret == -ENOSPC) | |
1875 | goto enospc; | |
1876 | if (ret) | |
1877 | goto error; | |
1878 | } | |
1879 | ||
0b86a832 CM |
1880 | search_start = stripe_align(root, search_start); |
1881 | ins->objectid = search_start; | |
1882 | ins->offset = num_bytes; | |
e37c9e69 | 1883 | |
db94535d | 1884 | if (ins->objectid + num_bytes >= search_end) |
cf67582b | 1885 | goto enospc; |
0b86a832 CM |
1886 | |
1887 | if (ins->objectid + num_bytes > | |
1888 | block_group->key.objectid + block_group->key.offset) { | |
e19caa5f CM |
1889 | search_start = block_group->key.objectid + |
1890 | block_group->key.offset; | |
1891 | goto new_group; | |
1892 | } | |
0b86a832 | 1893 | |
1a5bc167 | 1894 | if (test_range_bit(&info->extent_ins, ins->objectid, |
db94535d CM |
1895 | ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) { |
1896 | search_start = ins->objectid + num_bytes; | |
1a5bc167 CM |
1897 | goto new_group; |
1898 | } | |
0b86a832 | 1899 | |
1a5bc167 | 1900 | if (test_range_bit(&info->pinned_extents, ins->objectid, |
db94535d CM |
1901 | ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) { |
1902 | search_start = ins->objectid + num_bytes; | |
1a5bc167 | 1903 | goto new_group; |
fec577fb | 1904 | } |
0b86a832 | 1905 | |
db94535d | 1906 | if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start && |
f2654de4 CM |
1907 | ins->objectid < exclude_start + exclude_nr)) { |
1908 | search_start = exclude_start + exclude_nr; | |
1909 | goto new_group; | |
1910 | } | |
0b86a832 | 1911 | |
6324fbf3 | 1912 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) { |
5276aeda | 1913 | block_group = btrfs_lookup_block_group(info, ins->objectid); |
26b8003f CM |
1914 | if (block_group) |
1915 | trans->block_group = block_group; | |
f2458e1d | 1916 | } |
db94535d | 1917 | ins->offset = num_bytes; |
239b14b3 CM |
1918 | if (last_ptr) { |
1919 | *last_ptr = ins->objectid + ins->offset; | |
1920 | if (*last_ptr == | |
1921 | btrfs_super_total_bytes(&root->fs_info->super_copy)) { | |
1922 | *last_ptr = 0; | |
1923 | } | |
1924 | } | |
fec577fb | 1925 | return 0; |
be744175 CM |
1926 | |
1927 | new_group: | |
db94535d | 1928 | if (search_start + num_bytes >= search_end) { |
cf67582b | 1929 | enospc: |
be744175 | 1930 | search_start = orig_search_start; |
fbdc762b CM |
1931 | if (full_scan) { |
1932 | ret = -ENOSPC; | |
1933 | goto error; | |
1934 | } | |
6702ed49 CM |
1935 | if (wrapped) { |
1936 | if (!full_scan) | |
1937 | total_needed -= empty_size; | |
fbdc762b | 1938 | full_scan = 1; |
6702ed49 | 1939 | } else |
fbdc762b | 1940 | wrapped = 1; |
be744175 | 1941 | } |
0ef3e66b | 1942 | block_group = btrfs_lookup_first_block_group(info, search_start); |
fbdc762b | 1943 | cond_resched(); |
c286ac48 | 1944 | block_group = btrfs_find_block_group(root, block_group, |
1a2b2ac7 | 1945 | search_start, data, 0); |
be744175 CM |
1946 | goto check_failed; |
1947 | ||
0f70abe2 | 1948 | error: |
0f70abe2 | 1949 | return ret; |
fec577fb | 1950 | } |
ec44a35c | 1951 | |
e6dcd2dc CM |
1952 | static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
1953 | struct btrfs_root *root, | |
1954 | u64 num_bytes, u64 min_alloc_size, | |
1955 | u64 empty_size, u64 hint_byte, | |
1956 | u64 search_end, struct btrfs_key *ins, | |
1957 | u64 data) | |
fec577fb CM |
1958 | { |
1959 | int ret; | |
fbdc762b | 1960 | u64 search_start = 0; |
8790d502 | 1961 | u64 alloc_profile; |
1261ec42 | 1962 | struct btrfs_fs_info *info = root->fs_info; |
925baedd | 1963 | |
6324fbf3 | 1964 | if (data) { |
8790d502 CM |
1965 | alloc_profile = info->avail_data_alloc_bits & |
1966 | info->data_alloc_profile; | |
1967 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
6324fbf3 | 1968 | } else if (root == root->fs_info->chunk_root) { |
8790d502 CM |
1969 | alloc_profile = info->avail_system_alloc_bits & |
1970 | info->system_alloc_profile; | |
1971 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
6324fbf3 | 1972 | } else { |
8790d502 CM |
1973 | alloc_profile = info->avail_metadata_alloc_bits & |
1974 | info->metadata_alloc_profile; | |
1975 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
6324fbf3 | 1976 | } |
98d20f67 | 1977 | again: |
a061fc8d | 1978 | data = reduce_alloc_profile(root, data); |
0ef3e66b CM |
1979 | /* |
1980 | * the only place that sets empty_size is btrfs_realloc_node, which | |
1981 | * is not called recursively on allocations | |
1982 | */ | |
1983 | if (empty_size || root->ref_cows) { | |
593060d7 | 1984 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 1985 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b CM |
1986 | 2 * 1024 * 1024, |
1987 | BTRFS_BLOCK_GROUP_METADATA | | |
1988 | (info->metadata_alloc_profile & | |
1989 | info->avail_metadata_alloc_bits), 0); | |
6324fbf3 CM |
1990 | BUG_ON(ret); |
1991 | } | |
1992 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
0ef3e66b | 1993 | num_bytes + 2 * 1024 * 1024, data, 0); |
6324fbf3 CM |
1994 | BUG_ON(ret); |
1995 | } | |
0b86a832 | 1996 | |
db94535d CM |
1997 | WARN_ON(num_bytes < root->sectorsize); |
1998 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
1999 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
2000 | trans->alloc_exclude_start, |
2001 | trans->alloc_exclude_nr, data); | |
3b951516 | 2002 | |
98d20f67 CM |
2003 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
2004 | num_bytes = num_bytes >> 1; | |
2005 | num_bytes = max(num_bytes, min_alloc_size); | |
0ef3e66b CM |
2006 | do_chunk_alloc(trans, root->fs_info->extent_root, |
2007 | num_bytes, data, 1); | |
98d20f67 CM |
2008 | goto again; |
2009 | } | |
ec44a35c CM |
2010 | if (ret) { |
2011 | printk("allocation failed flags %Lu\n", data); | |
925baedd | 2012 | BUG(); |
925baedd | 2013 | } |
e6dcd2dc CM |
2014 | clear_extent_dirty(&root->fs_info->free_space_cache, |
2015 | ins->objectid, ins->objectid + ins->offset - 1, | |
2016 | GFP_NOFS); | |
2017 | return 0; | |
2018 | } | |
2019 | ||
2020 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, | |
2021 | struct btrfs_root *root, | |
2022 | u64 num_bytes, u64 min_alloc_size, | |
2023 | u64 empty_size, u64 hint_byte, | |
2024 | u64 search_end, struct btrfs_key *ins, | |
2025 | u64 data) | |
2026 | { | |
2027 | int ret; | |
2028 | maybe_lock_mutex(root); | |
2029 | ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, | |
2030 | empty_size, hint_byte, search_end, ins, | |
2031 | data); | |
2032 | maybe_unlock_mutex(root); | |
2033 | return ret; | |
2034 | } | |
2035 | ||
2036 | static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
2037 | struct btrfs_root *root, | |
2038 | u64 root_objectid, u64 ref_generation, | |
2039 | u64 owner, u64 owner_offset, | |
2040 | struct btrfs_key *ins) | |
2041 | { | |
2042 | int ret; | |
2043 | int pending_ret; | |
2044 | u64 super_used; | |
2045 | u64 root_used; | |
2046 | u64 num_bytes = ins->offset; | |
2047 | u32 sizes[2]; | |
2048 | struct btrfs_fs_info *info = root->fs_info; | |
2049 | struct btrfs_root *extent_root = info->extent_root; | |
2050 | struct btrfs_extent_item *extent_item; | |
2051 | struct btrfs_extent_ref *ref; | |
2052 | struct btrfs_path *path; | |
2053 | struct btrfs_key keys[2]; | |
fec577fb | 2054 | |
58176a96 | 2055 | /* block accounting for super block */ |
a2135011 | 2056 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
2057 | super_used = btrfs_super_bytes_used(&info->super_copy); |
2058 | btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); | |
a2135011 | 2059 | spin_unlock_irq(&info->delalloc_lock); |
26b8003f | 2060 | |
58176a96 | 2061 | /* block accounting for root item */ |
db94535d CM |
2062 | root_used = btrfs_root_used(&root->root_item); |
2063 | btrfs_set_root_used(&root->root_item, root_used + num_bytes); | |
58176a96 | 2064 | |
26b8003f | 2065 | if (root == extent_root) { |
1a5bc167 CM |
2066 | set_extent_bits(&root->fs_info->extent_ins, ins->objectid, |
2067 | ins->objectid + ins->offset - 1, | |
2068 | EXTENT_LOCKED, GFP_NOFS); | |
26b8003f CM |
2069 | goto update_block; |
2070 | } | |
2071 | ||
47e4bb98 CM |
2072 | memcpy(&keys[0], ins, sizeof(*ins)); |
2073 | keys[1].offset = hash_extent_ref(root_objectid, ref_generation, | |
2074 | owner, owner_offset); | |
2075 | keys[1].objectid = ins->objectid; | |
2076 | keys[1].type = BTRFS_EXTENT_REF_KEY; | |
2077 | sizes[0] = sizeof(*extent_item); | |
2078 | sizes[1] = sizeof(*ref); | |
7bb86316 CM |
2079 | |
2080 | path = btrfs_alloc_path(); | |
2081 | BUG_ON(!path); | |
47e4bb98 CM |
2082 | |
2083 | ret = btrfs_insert_empty_items(trans, extent_root, path, keys, | |
2084 | sizes, 2); | |
26b8003f | 2085 | |
ccd467d6 | 2086 | BUG_ON(ret); |
47e4bb98 CM |
2087 | extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2088 | struct btrfs_extent_item); | |
2089 | btrfs_set_extent_refs(path->nodes[0], extent_item, 1); | |
2090 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
2091 | struct btrfs_extent_ref); | |
2092 | ||
2093 | btrfs_set_ref_root(path->nodes[0], ref, root_objectid); | |
2094 | btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); | |
2095 | btrfs_set_ref_objectid(path->nodes[0], ref, owner); | |
2096 | btrfs_set_ref_offset(path->nodes[0], ref, owner_offset); | |
2097 | ||
2098 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
2099 | ||
2100 | trans->alloc_exclude_start = 0; | |
2101 | trans->alloc_exclude_nr = 0; | |
7bb86316 | 2102 | btrfs_free_path(path); |
e089f05c | 2103 | finish_current_insert(trans, extent_root); |
e20d96d6 | 2104 | pending_ret = del_pending_extents(trans, extent_root); |
f510cfec | 2105 | |
925baedd CM |
2106 | if (ret) |
2107 | goto out; | |
e37c9e69 | 2108 | if (pending_ret) { |
925baedd CM |
2109 | ret = pending_ret; |
2110 | goto out; | |
e37c9e69 | 2111 | } |
26b8003f CM |
2112 | |
2113 | update_block: | |
0b86a832 | 2114 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0); |
f5947066 CM |
2115 | if (ret) { |
2116 | printk("update block group failed for %Lu %Lu\n", | |
2117 | ins->objectid, ins->offset); | |
2118 | BUG(); | |
2119 | } | |
925baedd | 2120 | out: |
e6dcd2dc CM |
2121 | return ret; |
2122 | } | |
2123 | ||
2124 | int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
2125 | struct btrfs_root *root, | |
2126 | u64 root_objectid, u64 ref_generation, | |
2127 | u64 owner, u64 owner_offset, | |
2128 | struct btrfs_key *ins) | |
2129 | { | |
2130 | int ret; | |
2131 | maybe_lock_mutex(root); | |
2132 | ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid, | |
2133 | ref_generation, owner, | |
2134 | owner_offset, ins); | |
2135 | maybe_unlock_mutex(root); | |
2136 | return ret; | |
2137 | } | |
2138 | /* | |
2139 | * finds a free extent and does all the dirty work required for allocation | |
2140 | * returns the key for the extent through ins, and a tree buffer for | |
2141 | * the first block of the extent through buf. | |
2142 | * | |
2143 | * returns 0 if everything worked, non-zero otherwise. | |
2144 | */ | |
2145 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, | |
2146 | struct btrfs_root *root, | |
2147 | u64 num_bytes, u64 min_alloc_size, | |
2148 | u64 root_objectid, u64 ref_generation, | |
2149 | u64 owner, u64 owner_offset, | |
2150 | u64 empty_size, u64 hint_byte, | |
2151 | u64 search_end, struct btrfs_key *ins, u64 data) | |
2152 | { | |
2153 | int ret; | |
2154 | ||
2155 | maybe_lock_mutex(root); | |
2156 | ||
2157 | ret = __btrfs_reserve_extent(trans, root, num_bytes, | |
2158 | min_alloc_size, empty_size, hint_byte, | |
2159 | search_end, ins, data); | |
2160 | BUG_ON(ret); | |
2161 | ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid, | |
2162 | ref_generation, owner, | |
2163 | owner_offset, ins); | |
2164 | BUG_ON(ret); | |
2165 | ||
925baedd CM |
2166 | maybe_unlock_mutex(root); |
2167 | return ret; | |
fec577fb | 2168 | } |
fec577fb CM |
2169 | /* |
2170 | * helper function to allocate a block for a given tree | |
2171 | * returns the tree buffer or NULL. | |
2172 | */ | |
5f39d397 | 2173 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
7bb86316 CM |
2174 | struct btrfs_root *root, |
2175 | u32 blocksize, | |
2176 | u64 root_objectid, | |
2177 | u64 ref_generation, | |
2178 | u64 first_objectid, | |
2179 | int level, | |
2180 | u64 hint, | |
5f39d397 | 2181 | u64 empty_size) |
fec577fb | 2182 | { |
e2fa7227 | 2183 | struct btrfs_key ins; |
fec577fb | 2184 | int ret; |
5f39d397 | 2185 | struct extent_buffer *buf; |
fec577fb | 2186 | |
98d20f67 | 2187 | ret = btrfs_alloc_extent(trans, root, blocksize, blocksize, |
7bb86316 | 2188 | root_objectid, ref_generation, |
f6dbff55 | 2189 | level, first_objectid, empty_size, hint, |
db94535d | 2190 | (u64)-1, &ins, 0); |
fec577fb | 2191 | if (ret) { |
54aa1f4d CM |
2192 | BUG_ON(ret > 0); |
2193 | return ERR_PTR(ret); | |
fec577fb | 2194 | } |
db94535d | 2195 | buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize); |
54aa1f4d | 2196 | if (!buf) { |
7bb86316 CM |
2197 | btrfs_free_extent(trans, root, ins.objectid, blocksize, |
2198 | root->root_key.objectid, ref_generation, | |
2199 | 0, 0, 0); | |
54aa1f4d CM |
2200 | return ERR_PTR(-ENOMEM); |
2201 | } | |
55c69072 | 2202 | btrfs_set_header_generation(buf, trans->transid); |
925baedd | 2203 | btrfs_tree_lock(buf); |
55c69072 | 2204 | clean_tree_block(trans, root, buf); |
5f39d397 | 2205 | btrfs_set_buffer_uptodate(buf); |
55c69072 CM |
2206 | |
2207 | if (PageDirty(buf->first_page)) { | |
2208 | printk("page %lu dirty\n", buf->first_page->index); | |
2209 | WARN_ON(1); | |
2210 | } | |
2211 | ||
5f39d397 CM |
2212 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
2213 | buf->start + buf->len - 1, GFP_NOFS); | |
d3c2fdcf | 2214 | trans->blocks_used++; |
fec577fb CM |
2215 | return buf; |
2216 | } | |
a28ec197 | 2217 | |
98ed5174 CM |
2218 | static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans, |
2219 | struct btrfs_root *root, | |
2220 | struct extent_buffer *leaf) | |
6407bf6d | 2221 | { |
7bb86316 CM |
2222 | u64 leaf_owner; |
2223 | u64 leaf_generation; | |
5f39d397 | 2224 | struct btrfs_key key; |
6407bf6d CM |
2225 | struct btrfs_file_extent_item *fi; |
2226 | int i; | |
2227 | int nritems; | |
2228 | int ret; | |
2229 | ||
5f39d397 CM |
2230 | BUG_ON(!btrfs_is_leaf(leaf)); |
2231 | nritems = btrfs_header_nritems(leaf); | |
7bb86316 CM |
2232 | leaf_owner = btrfs_header_owner(leaf); |
2233 | leaf_generation = btrfs_header_generation(leaf); | |
2234 | ||
4a096752 CM |
2235 | mutex_unlock(&root->fs_info->alloc_mutex); |
2236 | ||
6407bf6d | 2237 | for (i = 0; i < nritems; i++) { |
db94535d | 2238 | u64 disk_bytenr; |
e34a5b4f | 2239 | cond_resched(); |
5f39d397 CM |
2240 | |
2241 | btrfs_item_key_to_cpu(leaf, &key, i); | |
2242 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d CM |
2243 | continue; |
2244 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
5f39d397 CM |
2245 | if (btrfs_file_extent_type(leaf, fi) == |
2246 | BTRFS_FILE_EXTENT_INLINE) | |
236454df | 2247 | continue; |
6407bf6d CM |
2248 | /* |
2249 | * FIXME make sure to insert a trans record that | |
2250 | * repeats the snapshot del on crash | |
2251 | */ | |
db94535d CM |
2252 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
2253 | if (disk_bytenr == 0) | |
3a686375 | 2254 | continue; |
4a096752 CM |
2255 | |
2256 | mutex_lock(&root->fs_info->alloc_mutex); | |
925baedd | 2257 | ret = __btrfs_free_extent(trans, root, disk_bytenr, |
7bb86316 CM |
2258 | btrfs_file_extent_disk_num_bytes(leaf, fi), |
2259 | leaf_owner, leaf_generation, | |
2260 | key.objectid, key.offset, 0); | |
4a096752 | 2261 | mutex_unlock(&root->fs_info->alloc_mutex); |
6407bf6d CM |
2262 | BUG_ON(ret); |
2263 | } | |
4a096752 CM |
2264 | |
2265 | mutex_lock(&root->fs_info->alloc_mutex); | |
6407bf6d CM |
2266 | return 0; |
2267 | } | |
2268 | ||
98ed5174 | 2269 | static void noinline reada_walk_down(struct btrfs_root *root, |
bea495e5 CM |
2270 | struct extent_buffer *node, |
2271 | int slot) | |
e011599b | 2272 | { |
db94535d | 2273 | u64 bytenr; |
bea495e5 CM |
2274 | u64 last = 0; |
2275 | u32 nritems; | |
e011599b | 2276 | u32 refs; |
db94535d | 2277 | u32 blocksize; |
bea495e5 CM |
2278 | int ret; |
2279 | int i; | |
2280 | int level; | |
2281 | int skipped = 0; | |
e011599b | 2282 | |
5f39d397 | 2283 | nritems = btrfs_header_nritems(node); |
db94535d | 2284 | level = btrfs_header_level(node); |
bea495e5 CM |
2285 | if (level) |
2286 | return; | |
2287 | ||
2288 | for (i = slot; i < nritems && skipped < 32; i++) { | |
db94535d | 2289 | bytenr = btrfs_node_blockptr(node, i); |
bea495e5 CM |
2290 | if (last && ((bytenr > last && bytenr - last > 32 * 1024) || |
2291 | (last > bytenr && last - bytenr > 32 * 1024))) { | |
2292 | skipped++; | |
e011599b | 2293 | continue; |
bea495e5 CM |
2294 | } |
2295 | blocksize = btrfs_level_size(root, level - 1); | |
2296 | if (i != slot) { | |
2297 | ret = lookup_extent_ref(NULL, root, bytenr, | |
2298 | blocksize, &refs); | |
2299 | BUG_ON(ret); | |
2300 | if (refs != 1) { | |
2301 | skipped++; | |
2302 | continue; | |
2303 | } | |
2304 | } | |
ca7a79ad CM |
2305 | ret = readahead_tree_block(root, bytenr, blocksize, |
2306 | btrfs_node_ptr_generation(node, i)); | |
bea495e5 | 2307 | last = bytenr + blocksize; |
409eb95d | 2308 | cond_resched(); |
e011599b CM |
2309 | if (ret) |
2310 | break; | |
2311 | } | |
2312 | } | |
2313 | ||
333db94c CM |
2314 | /* |
2315 | * we want to avoid as much random IO as we can with the alloc mutex | |
2316 | * held, so drop the lock and do the lookup, then do it again with the | |
2317 | * lock held. | |
2318 | */ | |
2319 | int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, | |
2320 | u32 *refs) | |
2321 | { | |
2322 | mutex_unlock(&root->fs_info->alloc_mutex); | |
2323 | lookup_extent_ref(NULL, root, start, len, refs); | |
e7a84565 | 2324 | cond_resched(); |
333db94c CM |
2325 | mutex_lock(&root->fs_info->alloc_mutex); |
2326 | return lookup_extent_ref(NULL, root, start, len, refs); | |
2327 | } | |
2328 | ||
9aca1d51 CM |
2329 | /* |
2330 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
2331 | * counts as it goes. | |
2332 | */ | |
98ed5174 CM |
2333 | static int noinline walk_down_tree(struct btrfs_trans_handle *trans, |
2334 | struct btrfs_root *root, | |
2335 | struct btrfs_path *path, int *level) | |
20524f02 | 2336 | { |
7bb86316 CM |
2337 | u64 root_owner; |
2338 | u64 root_gen; | |
2339 | u64 bytenr; | |
ca7a79ad | 2340 | u64 ptr_gen; |
5f39d397 CM |
2341 | struct extent_buffer *next; |
2342 | struct extent_buffer *cur; | |
7bb86316 | 2343 | struct extent_buffer *parent; |
db94535d | 2344 | u32 blocksize; |
20524f02 CM |
2345 | int ret; |
2346 | u32 refs; | |
2347 | ||
925baedd CM |
2348 | mutex_lock(&root->fs_info->alloc_mutex); |
2349 | ||
5caf2a00 CM |
2350 | WARN_ON(*level < 0); |
2351 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
333db94c | 2352 | ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start, |
db94535d | 2353 | path->nodes[*level]->len, &refs); |
20524f02 CM |
2354 | BUG_ON(ret); |
2355 | if (refs > 1) | |
2356 | goto out; | |
e011599b | 2357 | |
9aca1d51 CM |
2358 | /* |
2359 | * walk down to the last node level and free all the leaves | |
2360 | */ | |
6407bf6d | 2361 | while(*level >= 0) { |
5caf2a00 CM |
2362 | WARN_ON(*level < 0); |
2363 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
20524f02 | 2364 | cur = path->nodes[*level]; |
e011599b | 2365 | |
5f39d397 | 2366 | if (btrfs_header_level(cur) != *level) |
2c90e5d6 | 2367 | WARN_ON(1); |
e011599b | 2368 | |
7518a238 | 2369 | if (path->slots[*level] >= |
5f39d397 | 2370 | btrfs_header_nritems(cur)) |
20524f02 | 2371 | break; |
6407bf6d CM |
2372 | if (*level == 0) { |
2373 | ret = drop_leaf_ref(trans, root, cur); | |
2374 | BUG_ON(ret); | |
2375 | break; | |
2376 | } | |
db94535d | 2377 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
ca7a79ad | 2378 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); |
db94535d | 2379 | blocksize = btrfs_level_size(root, *level - 1); |
925baedd | 2380 | |
333db94c | 2381 | ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); |
6407bf6d CM |
2382 | BUG_ON(ret); |
2383 | if (refs != 1) { | |
7bb86316 CM |
2384 | parent = path->nodes[*level]; |
2385 | root_owner = btrfs_header_owner(parent); | |
2386 | root_gen = btrfs_header_generation(parent); | |
20524f02 | 2387 | path->slots[*level]++; |
925baedd | 2388 | ret = __btrfs_free_extent(trans, root, bytenr, |
7bb86316 CM |
2389 | blocksize, root_owner, |
2390 | root_gen, 0, 0, 1); | |
20524f02 CM |
2391 | BUG_ON(ret); |
2392 | continue; | |
2393 | } | |
db94535d | 2394 | next = btrfs_find_tree_block(root, bytenr, blocksize); |
1259ab75 | 2395 | if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) { |
5f39d397 | 2396 | free_extent_buffer(next); |
333db94c CM |
2397 | mutex_unlock(&root->fs_info->alloc_mutex); |
2398 | ||
a74a4b97 CM |
2399 | if (path->slots[*level] == 0) |
2400 | reada_walk_down(root, cur, path->slots[*level]); | |
8790d502 | 2401 | |
ca7a79ad CM |
2402 | next = read_tree_block(root, bytenr, blocksize, |
2403 | ptr_gen); | |
e7a84565 | 2404 | cond_resched(); |
925baedd | 2405 | mutex_lock(&root->fs_info->alloc_mutex); |
e9d0b13b | 2406 | |
8790d502 | 2407 | /* we've dropped the lock, double check */ |
e6dcd2dc CM |
2408 | ret = lookup_extent_ref(NULL, root, bytenr, blocksize, |
2409 | &refs); | |
e9d0b13b CM |
2410 | BUG_ON(ret); |
2411 | if (refs != 1) { | |
7bb86316 CM |
2412 | parent = path->nodes[*level]; |
2413 | root_owner = btrfs_header_owner(parent); | |
2414 | root_gen = btrfs_header_generation(parent); | |
2415 | ||
e9d0b13b | 2416 | path->slots[*level]++; |
5f39d397 | 2417 | free_extent_buffer(next); |
925baedd | 2418 | ret = __btrfs_free_extent(trans, root, bytenr, |
7bb86316 CM |
2419 | blocksize, |
2420 | root_owner, | |
2421 | root_gen, 0, 0, 1); | |
e9d0b13b CM |
2422 | BUG_ON(ret); |
2423 | continue; | |
2424 | } | |
2425 | } | |
5caf2a00 | 2426 | WARN_ON(*level <= 0); |
83e15a28 | 2427 | if (path->nodes[*level-1]) |
5f39d397 | 2428 | free_extent_buffer(path->nodes[*level-1]); |
20524f02 | 2429 | path->nodes[*level-1] = next; |
5f39d397 | 2430 | *level = btrfs_header_level(next); |
20524f02 CM |
2431 | path->slots[*level] = 0; |
2432 | } | |
2433 | out: | |
5caf2a00 CM |
2434 | WARN_ON(*level < 0); |
2435 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
7bb86316 CM |
2436 | |
2437 | if (path->nodes[*level] == root->node) { | |
2438 | root_owner = root->root_key.objectid; | |
2439 | parent = path->nodes[*level]; | |
2440 | } else { | |
2441 | parent = path->nodes[*level + 1]; | |
2442 | root_owner = btrfs_header_owner(parent); | |
2443 | } | |
2444 | ||
2445 | root_gen = btrfs_header_generation(parent); | |
925baedd | 2446 | ret = __btrfs_free_extent(trans, root, path->nodes[*level]->start, |
7bb86316 CM |
2447 | path->nodes[*level]->len, |
2448 | root_owner, root_gen, 0, 0, 1); | |
5f39d397 | 2449 | free_extent_buffer(path->nodes[*level]); |
20524f02 CM |
2450 | path->nodes[*level] = NULL; |
2451 | *level += 1; | |
2452 | BUG_ON(ret); | |
925baedd | 2453 | mutex_unlock(&root->fs_info->alloc_mutex); |
e7a84565 | 2454 | cond_resched(); |
20524f02 CM |
2455 | return 0; |
2456 | } | |
2457 | ||
9aca1d51 CM |
2458 | /* |
2459 | * helper for dropping snapshots. This walks back up the tree in the path | |
2460 | * to find the first node higher up where we haven't yet gone through | |
2461 | * all the slots | |
2462 | */ | |
98ed5174 CM |
2463 | static int noinline walk_up_tree(struct btrfs_trans_handle *trans, |
2464 | struct btrfs_root *root, | |
2465 | struct btrfs_path *path, int *level) | |
20524f02 | 2466 | { |
7bb86316 CM |
2467 | u64 root_owner; |
2468 | u64 root_gen; | |
2469 | struct btrfs_root_item *root_item = &root->root_item; | |
20524f02 CM |
2470 | int i; |
2471 | int slot; | |
2472 | int ret; | |
9f3a7427 | 2473 | |
234b63a0 | 2474 | for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
20524f02 | 2475 | slot = path->slots[i]; |
5f39d397 CM |
2476 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { |
2477 | struct extent_buffer *node; | |
2478 | struct btrfs_disk_key disk_key; | |
2479 | node = path->nodes[i]; | |
20524f02 CM |
2480 | path->slots[i]++; |
2481 | *level = i; | |
9f3a7427 | 2482 | WARN_ON(*level == 0); |
5f39d397 | 2483 | btrfs_node_key(node, &disk_key, path->slots[i]); |
9f3a7427 | 2484 | memcpy(&root_item->drop_progress, |
5f39d397 | 2485 | &disk_key, sizeof(disk_key)); |
9f3a7427 | 2486 | root_item->drop_level = i; |
20524f02 CM |
2487 | return 0; |
2488 | } else { | |
7bb86316 CM |
2489 | if (path->nodes[*level] == root->node) { |
2490 | root_owner = root->root_key.objectid; | |
2491 | root_gen = | |
2492 | btrfs_header_generation(path->nodes[*level]); | |
2493 | } else { | |
2494 | struct extent_buffer *node; | |
2495 | node = path->nodes[*level + 1]; | |
2496 | root_owner = btrfs_header_owner(node); | |
2497 | root_gen = btrfs_header_generation(node); | |
2498 | } | |
e089f05c | 2499 | ret = btrfs_free_extent(trans, root, |
db94535d | 2500 | path->nodes[*level]->start, |
7bb86316 CM |
2501 | path->nodes[*level]->len, |
2502 | root_owner, root_gen, 0, 0, 1); | |
6407bf6d | 2503 | BUG_ON(ret); |
5f39d397 | 2504 | free_extent_buffer(path->nodes[*level]); |
83e15a28 | 2505 | path->nodes[*level] = NULL; |
20524f02 | 2506 | *level = i + 1; |
20524f02 CM |
2507 | } |
2508 | } | |
2509 | return 1; | |
2510 | } | |
2511 | ||
9aca1d51 CM |
2512 | /* |
2513 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2514 | * the tree freeing any blocks that have a ref count of zero after being | |
2515 | * decremented. | |
2516 | */ | |
e089f05c | 2517 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
9f3a7427 | 2518 | *root) |
20524f02 | 2519 | { |
3768f368 | 2520 | int ret = 0; |
9aca1d51 | 2521 | int wret; |
20524f02 | 2522 | int level; |
5caf2a00 | 2523 | struct btrfs_path *path; |
20524f02 CM |
2524 | int i; |
2525 | int orig_level; | |
9f3a7427 | 2526 | struct btrfs_root_item *root_item = &root->root_item; |
20524f02 | 2527 | |
a2135011 | 2528 | WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); |
5caf2a00 CM |
2529 | path = btrfs_alloc_path(); |
2530 | BUG_ON(!path); | |
20524f02 | 2531 | |
5f39d397 | 2532 | level = btrfs_header_level(root->node); |
20524f02 | 2533 | orig_level = level; |
9f3a7427 CM |
2534 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2535 | path->nodes[level] = root->node; | |
f510cfec | 2536 | extent_buffer_get(root->node); |
9f3a7427 CM |
2537 | path->slots[level] = 0; |
2538 | } else { | |
2539 | struct btrfs_key key; | |
5f39d397 CM |
2540 | struct btrfs_disk_key found_key; |
2541 | struct extent_buffer *node; | |
6702ed49 | 2542 | |
9f3a7427 | 2543 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
6702ed49 CM |
2544 | level = root_item->drop_level; |
2545 | path->lowest_level = level; | |
9f3a7427 | 2546 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6702ed49 | 2547 | if (wret < 0) { |
9f3a7427 CM |
2548 | ret = wret; |
2549 | goto out; | |
2550 | } | |
5f39d397 CM |
2551 | node = path->nodes[level]; |
2552 | btrfs_node_key(node, &found_key, path->slots[level]); | |
2553 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, | |
2554 | sizeof(found_key))); | |
7d9eb12c CM |
2555 | /* |
2556 | * unlock our path, this is safe because only this | |
2557 | * function is allowed to delete this snapshot | |
2558 | */ | |
925baedd CM |
2559 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
2560 | if (path->nodes[i] && path->locks[i]) { | |
2561 | path->locks[i] = 0; | |
2562 | btrfs_tree_unlock(path->nodes[i]); | |
2563 | } | |
2564 | } | |
9f3a7427 | 2565 | } |
20524f02 | 2566 | while(1) { |
5caf2a00 | 2567 | wret = walk_down_tree(trans, root, path, &level); |
9aca1d51 | 2568 | if (wret > 0) |
20524f02 | 2569 | break; |
9aca1d51 CM |
2570 | if (wret < 0) |
2571 | ret = wret; | |
2572 | ||
5caf2a00 | 2573 | wret = walk_up_tree(trans, root, path, &level); |
9aca1d51 | 2574 | if (wret > 0) |
20524f02 | 2575 | break; |
9aca1d51 CM |
2576 | if (wret < 0) |
2577 | ret = wret; | |
e7a84565 CM |
2578 | if (trans->transaction->in_commit) { |
2579 | ret = -EAGAIN; | |
2580 | break; | |
2581 | } | |
20524f02 | 2582 | } |
83e15a28 | 2583 | for (i = 0; i <= orig_level; i++) { |
5caf2a00 | 2584 | if (path->nodes[i]) { |
5f39d397 | 2585 | free_extent_buffer(path->nodes[i]); |
0f82731f | 2586 | path->nodes[i] = NULL; |
83e15a28 | 2587 | } |
20524f02 | 2588 | } |
9f3a7427 | 2589 | out: |
5caf2a00 | 2590 | btrfs_free_path(path); |
9aca1d51 | 2591 | return ret; |
20524f02 | 2592 | } |
9078a3e1 | 2593 | |
96b5179d | 2594 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9078a3e1 | 2595 | { |
96b5179d CM |
2596 | u64 start; |
2597 | u64 end; | |
b97f9203 | 2598 | u64 ptr; |
9078a3e1 | 2599 | int ret; |
925baedd CM |
2600 | |
2601 | mutex_lock(&info->alloc_mutex); | |
9078a3e1 | 2602 | while(1) { |
96b5179d CM |
2603 | ret = find_first_extent_bit(&info->block_group_cache, 0, |
2604 | &start, &end, (unsigned int)-1); | |
2605 | if (ret) | |
9078a3e1 | 2606 | break; |
b97f9203 Y |
2607 | ret = get_state_private(&info->block_group_cache, start, &ptr); |
2608 | if (!ret) | |
2609 | kfree((void *)(unsigned long)ptr); | |
96b5179d CM |
2610 | clear_extent_bits(&info->block_group_cache, start, |
2611 | end, (unsigned int)-1, GFP_NOFS); | |
9078a3e1 | 2612 | } |
e37c9e69 | 2613 | while(1) { |
f510cfec CM |
2614 | ret = find_first_extent_bit(&info->free_space_cache, 0, |
2615 | &start, &end, EXTENT_DIRTY); | |
2616 | if (ret) | |
e37c9e69 | 2617 | break; |
f510cfec CM |
2618 | clear_extent_dirty(&info->free_space_cache, start, |
2619 | end, GFP_NOFS); | |
e37c9e69 | 2620 | } |
925baedd | 2621 | mutex_unlock(&info->alloc_mutex); |
be744175 CM |
2622 | return 0; |
2623 | } | |
2624 | ||
8e7bf94f CM |
2625 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
2626 | unsigned long nr) | |
2627 | { | |
2628 | return min(last, start + nr - 1); | |
2629 | } | |
2630 | ||
98ed5174 CM |
2631 | static int noinline relocate_inode_pages(struct inode *inode, u64 start, |
2632 | u64 len) | |
edbd8d4e CM |
2633 | { |
2634 | u64 page_start; | |
2635 | u64 page_end; | |
edbd8d4e | 2636 | unsigned long last_index; |
edbd8d4e CM |
2637 | unsigned long i; |
2638 | struct page *page; | |
d1310b2e | 2639 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 2640 | struct file_ra_state *ra; |
8e7bf94f CM |
2641 | unsigned long total_read = 0; |
2642 | unsigned long ra_pages; | |
3eaa2885 | 2643 | struct btrfs_ordered_extent *ordered; |
a061fc8d | 2644 | struct btrfs_trans_handle *trans; |
4313b399 CM |
2645 | |
2646 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
2647 | |
2648 | mutex_lock(&inode->i_mutex); | |
4313b399 | 2649 | i = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
2650 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
2651 | ||
8e7bf94f CM |
2652 | ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages; |
2653 | ||
4313b399 | 2654 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 2655 | |
4313b399 | 2656 | for (; i <= last_index; i++) { |
8e7bf94f CM |
2657 | if (total_read % ra_pages == 0) { |
2658 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, | |
2659 | calc_ra(i, last_index, ra_pages)); | |
2660 | } | |
2661 | total_read++; | |
3eaa2885 CM |
2662 | again: |
2663 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
a061fc8d | 2664 | goto truncate_racing; |
edbd8d4e | 2665 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 2666 | if (!page) { |
edbd8d4e | 2667 | goto out_unlock; |
a061fc8d | 2668 | } |
edbd8d4e CM |
2669 | if (!PageUptodate(page)) { |
2670 | btrfs_readpage(NULL, page); | |
2671 | lock_page(page); | |
2672 | if (!PageUptodate(page)) { | |
2673 | unlock_page(page); | |
2674 | page_cache_release(page); | |
2675 | goto out_unlock; | |
2676 | } | |
2677 | } | |
ec44a35c | 2678 | wait_on_page_writeback(page); |
3eaa2885 | 2679 | |
edbd8d4e CM |
2680 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
2681 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
d1310b2e | 2682 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 2683 | |
3eaa2885 CM |
2684 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
2685 | if (ordered) { | |
2686 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2687 | unlock_page(page); | |
2688 | page_cache_release(page); | |
2689 | btrfs_start_ordered_extent(inode, ordered, 1); | |
2690 | btrfs_put_ordered_extent(ordered); | |
2691 | goto again; | |
2692 | } | |
2693 | set_page_extent_mapped(page); | |
2694 | ||
2695 | ||
d1310b2e | 2696 | set_extent_delalloc(io_tree, page_start, |
edbd8d4e | 2697 | page_end, GFP_NOFS); |
a061fc8d | 2698 | set_page_dirty(page); |
edbd8d4e | 2699 | |
d1310b2e | 2700 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
2701 | unlock_page(page); |
2702 | page_cache_release(page); | |
2703 | } | |
2704 | ||
2705 | out_unlock: | |
3eaa2885 CM |
2706 | /* we have to start the IO in order to get the ordered extents |
2707 | * instantiated. This allows the relocation to code to wait | |
2708 | * for all the ordered extents to hit the disk. | |
2709 | * | |
2710 | * Otherwise, it would constantly loop over the same extents | |
2711 | * because the old ones don't get deleted until the IO is | |
2712 | * started | |
2713 | */ | |
2714 | btrfs_fdatawrite_range(inode->i_mapping, start, start + len - 1, | |
2715 | WB_SYNC_NONE); | |
ec44a35c | 2716 | kfree(ra); |
a061fc8d CM |
2717 | trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1); |
2718 | if (trans) { | |
a061fc8d CM |
2719 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); |
2720 | mark_inode_dirty(inode); | |
2721 | } | |
edbd8d4e CM |
2722 | mutex_unlock(&inode->i_mutex); |
2723 | return 0; | |
a061fc8d CM |
2724 | |
2725 | truncate_racing: | |
2726 | vmtruncate(inode, inode->i_size); | |
2727 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, | |
2728 | total_read); | |
2729 | goto out_unlock; | |
edbd8d4e CM |
2730 | } |
2731 | ||
bf4ef679 CM |
2732 | /* |
2733 | * The back references tell us which tree holds a ref on a block, | |
2734 | * but it is possible for the tree root field in the reference to | |
2735 | * reflect the original root before a snapshot was made. In this | |
2736 | * case we should search through all the children of a given root | |
2737 | * to find potential holders of references on a block. | |
2738 | * | |
2739 | * Instead, we do something a little less fancy and just search | |
2740 | * all the roots for a given key/block combination. | |
2741 | */ | |
2742 | static int find_root_for_ref(struct btrfs_root *root, | |
2743 | struct btrfs_path *path, | |
2744 | struct btrfs_key *key0, | |
2745 | int level, | |
2746 | int file_key, | |
2747 | struct btrfs_root **found_root, | |
2748 | u64 bytenr) | |
2749 | { | |
2750 | struct btrfs_key root_location; | |
2751 | struct btrfs_root *cur_root = *found_root; | |
2752 | struct btrfs_file_extent_item *file_extent; | |
2753 | u64 root_search_start = BTRFS_FS_TREE_OBJECTID; | |
2754 | u64 found_bytenr; | |
2755 | int ret; | |
bf4ef679 CM |
2756 | |
2757 | root_location.offset = (u64)-1; | |
2758 | root_location.type = BTRFS_ROOT_ITEM_KEY; | |
2759 | path->lowest_level = level; | |
2760 | path->reada = 0; | |
2761 | while(1) { | |
2762 | ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0); | |
2763 | found_bytenr = 0; | |
2764 | if (ret == 0 && file_key) { | |
2765 | struct extent_buffer *leaf = path->nodes[0]; | |
2766 | file_extent = btrfs_item_ptr(leaf, path->slots[0], | |
2767 | struct btrfs_file_extent_item); | |
2768 | if (btrfs_file_extent_type(leaf, file_extent) == | |
2769 | BTRFS_FILE_EXTENT_REG) { | |
2770 | found_bytenr = | |
2771 | btrfs_file_extent_disk_bytenr(leaf, | |
2772 | file_extent); | |
2773 | } | |
323da79c | 2774 | } else if (!file_key) { |
bf4ef679 CM |
2775 | if (path->nodes[level]) |
2776 | found_bytenr = path->nodes[level]->start; | |
2777 | } | |
2778 | ||
bf4ef679 CM |
2779 | btrfs_release_path(cur_root, path); |
2780 | ||
2781 | if (found_bytenr == bytenr) { | |
2782 | *found_root = cur_root; | |
2783 | ret = 0; | |
2784 | goto out; | |
2785 | } | |
2786 | ret = btrfs_search_root(root->fs_info->tree_root, | |
2787 | root_search_start, &root_search_start); | |
2788 | if (ret) | |
2789 | break; | |
2790 | ||
2791 | root_location.objectid = root_search_start; | |
2792 | cur_root = btrfs_read_fs_root_no_name(root->fs_info, | |
2793 | &root_location); | |
2794 | if (!cur_root) { | |
2795 | ret = 1; | |
2796 | break; | |
2797 | } | |
2798 | } | |
2799 | out: | |
2800 | path->lowest_level = 0; | |
2801 | return ret; | |
2802 | } | |
2803 | ||
4313b399 CM |
2804 | /* |
2805 | * note, this releases the path | |
2806 | */ | |
98ed5174 | 2807 | static int noinline relocate_one_reference(struct btrfs_root *extent_root, |
edbd8d4e | 2808 | struct btrfs_path *path, |
0ef3e66b CM |
2809 | struct btrfs_key *extent_key, |
2810 | u64 *last_file_objectid, | |
2811 | u64 *last_file_offset, | |
2812 | u64 *last_file_root, | |
2813 | u64 last_extent) | |
edbd8d4e CM |
2814 | { |
2815 | struct inode *inode; | |
2816 | struct btrfs_root *found_root; | |
bf4ef679 CM |
2817 | struct btrfs_key root_location; |
2818 | struct btrfs_key found_key; | |
4313b399 CM |
2819 | struct btrfs_extent_ref *ref; |
2820 | u64 ref_root; | |
2821 | u64 ref_gen; | |
2822 | u64 ref_objectid; | |
2823 | u64 ref_offset; | |
edbd8d4e | 2824 | int ret; |
bf4ef679 | 2825 | int level; |
edbd8d4e | 2826 | |
7d9eb12c CM |
2827 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); |
2828 | ||
4313b399 CM |
2829 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2830 | struct btrfs_extent_ref); | |
2831 | ref_root = btrfs_ref_root(path->nodes[0], ref); | |
2832 | ref_gen = btrfs_ref_generation(path->nodes[0], ref); | |
2833 | ref_objectid = btrfs_ref_objectid(path->nodes[0], ref); | |
2834 | ref_offset = btrfs_ref_offset(path->nodes[0], ref); | |
2835 | btrfs_release_path(extent_root, path); | |
2836 | ||
bf4ef679 | 2837 | root_location.objectid = ref_root; |
edbd8d4e | 2838 | if (ref_gen == 0) |
bf4ef679 | 2839 | root_location.offset = 0; |
edbd8d4e | 2840 | else |
bf4ef679 CM |
2841 | root_location.offset = (u64)-1; |
2842 | root_location.type = BTRFS_ROOT_ITEM_KEY; | |
edbd8d4e CM |
2843 | |
2844 | found_root = btrfs_read_fs_root_no_name(extent_root->fs_info, | |
bf4ef679 | 2845 | &root_location); |
edbd8d4e | 2846 | BUG_ON(!found_root); |
7d9eb12c | 2847 | mutex_unlock(&extent_root->fs_info->alloc_mutex); |
edbd8d4e CM |
2848 | |
2849 | if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
bf4ef679 CM |
2850 | found_key.objectid = ref_objectid; |
2851 | found_key.type = BTRFS_EXTENT_DATA_KEY; | |
2852 | found_key.offset = ref_offset; | |
2853 | level = 0; | |
2854 | ||
0ef3e66b CM |
2855 | if (last_extent == extent_key->objectid && |
2856 | *last_file_objectid == ref_objectid && | |
2857 | *last_file_offset == ref_offset && | |
2858 | *last_file_root == ref_root) | |
2859 | goto out; | |
2860 | ||
bf4ef679 CM |
2861 | ret = find_root_for_ref(extent_root, path, &found_key, |
2862 | level, 1, &found_root, | |
2863 | extent_key->objectid); | |
2864 | ||
2865 | if (ret) | |
2866 | goto out; | |
2867 | ||
0ef3e66b CM |
2868 | if (last_extent == extent_key->objectid && |
2869 | *last_file_objectid == ref_objectid && | |
2870 | *last_file_offset == ref_offset && | |
2871 | *last_file_root == ref_root) | |
2872 | goto out; | |
2873 | ||
edbd8d4e CM |
2874 | inode = btrfs_iget_locked(extent_root->fs_info->sb, |
2875 | ref_objectid, found_root); | |
2876 | if (inode->i_state & I_NEW) { | |
2877 | /* the inode and parent dir are two different roots */ | |
2878 | BTRFS_I(inode)->root = found_root; | |
2879 | BTRFS_I(inode)->location.objectid = ref_objectid; | |
2880 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
2881 | BTRFS_I(inode)->location.offset = 0; | |
2882 | btrfs_read_locked_inode(inode); | |
2883 | unlock_new_inode(inode); | |
2884 | ||
2885 | } | |
2886 | /* this can happen if the reference is not against | |
2887 | * the latest version of the tree root | |
2888 | */ | |
7d9eb12c | 2889 | if (is_bad_inode(inode)) |
edbd8d4e | 2890 | goto out; |
7d9eb12c | 2891 | |
0ef3e66b CM |
2892 | *last_file_objectid = inode->i_ino; |
2893 | *last_file_root = found_root->root_key.objectid; | |
2894 | *last_file_offset = ref_offset; | |
2895 | ||
edbd8d4e | 2896 | relocate_inode_pages(inode, ref_offset, extent_key->offset); |
edbd8d4e | 2897 | iput(inode); |
edbd8d4e CM |
2898 | } else { |
2899 | struct btrfs_trans_handle *trans; | |
edbd8d4e | 2900 | struct extent_buffer *eb; |
7d9eb12c | 2901 | int needs_lock = 0; |
edbd8d4e | 2902 | |
edbd8d4e | 2903 | eb = read_tree_block(found_root, extent_key->objectid, |
ca7a79ad | 2904 | extent_key->offset, 0); |
925baedd | 2905 | btrfs_tree_lock(eb); |
edbd8d4e CM |
2906 | level = btrfs_header_level(eb); |
2907 | ||
2908 | if (level == 0) | |
2909 | btrfs_item_key_to_cpu(eb, &found_key, 0); | |
2910 | else | |
2911 | btrfs_node_key_to_cpu(eb, &found_key, 0); | |
2912 | ||
925baedd | 2913 | btrfs_tree_unlock(eb); |
edbd8d4e CM |
2914 | free_extent_buffer(eb); |
2915 | ||
bf4ef679 CM |
2916 | ret = find_root_for_ref(extent_root, path, &found_key, |
2917 | level, 0, &found_root, | |
2918 | extent_key->objectid); | |
2919 | ||
2920 | if (ret) | |
2921 | goto out; | |
2922 | ||
7d9eb12c CM |
2923 | /* |
2924 | * right here almost anything could happen to our key, | |
2925 | * but that's ok. The cow below will either relocate it | |
2926 | * or someone else will have relocated it. Either way, | |
2927 | * it is in a different spot than it was before and | |
2928 | * we're happy. | |
2929 | */ | |
2930 | ||
bf4ef679 CM |
2931 | trans = btrfs_start_transaction(found_root, 1); |
2932 | ||
7d9eb12c CM |
2933 | if (found_root == extent_root->fs_info->extent_root || |
2934 | found_root == extent_root->fs_info->chunk_root || | |
2935 | found_root == extent_root->fs_info->dev_root) { | |
2936 | needs_lock = 1; | |
2937 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
2938 | } | |
2939 | ||
edbd8d4e | 2940 | path->lowest_level = level; |
8f662a76 | 2941 | path->reada = 2; |
edbd8d4e CM |
2942 | ret = btrfs_search_slot(trans, found_root, &found_key, path, |
2943 | 0, 1); | |
2944 | path->lowest_level = 0; | |
edbd8d4e | 2945 | btrfs_release_path(found_root, path); |
7d9eb12c | 2946 | |
0ef3e66b CM |
2947 | if (found_root == found_root->fs_info->extent_root) |
2948 | btrfs_extent_post_op(trans, found_root); | |
7d9eb12c CM |
2949 | if (needs_lock) |
2950 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
2951 | ||
edbd8d4e | 2952 | btrfs_end_transaction(trans, found_root); |
edbd8d4e | 2953 | |
7d9eb12c | 2954 | } |
edbd8d4e | 2955 | out: |
7d9eb12c | 2956 | mutex_lock(&extent_root->fs_info->alloc_mutex); |
edbd8d4e CM |
2957 | return 0; |
2958 | } | |
2959 | ||
a061fc8d CM |
2960 | static int noinline del_extent_zero(struct btrfs_root *extent_root, |
2961 | struct btrfs_path *path, | |
2962 | struct btrfs_key *extent_key) | |
2963 | { | |
2964 | int ret; | |
2965 | struct btrfs_trans_handle *trans; | |
2966 | ||
2967 | trans = btrfs_start_transaction(extent_root, 1); | |
2968 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); | |
2969 | if (ret > 0) { | |
2970 | ret = -EIO; | |
2971 | goto out; | |
2972 | } | |
2973 | if (ret < 0) | |
2974 | goto out; | |
2975 | ret = btrfs_del_item(trans, extent_root, path); | |
2976 | out: | |
2977 | btrfs_end_transaction(trans, extent_root); | |
2978 | return ret; | |
2979 | } | |
2980 | ||
98ed5174 CM |
2981 | static int noinline relocate_one_extent(struct btrfs_root *extent_root, |
2982 | struct btrfs_path *path, | |
2983 | struct btrfs_key *extent_key) | |
edbd8d4e CM |
2984 | { |
2985 | struct btrfs_key key; | |
2986 | struct btrfs_key found_key; | |
edbd8d4e | 2987 | struct extent_buffer *leaf; |
0ef3e66b CM |
2988 | u64 last_file_objectid = 0; |
2989 | u64 last_file_root = 0; | |
2990 | u64 last_file_offset = (u64)-1; | |
2991 | u64 last_extent = 0; | |
edbd8d4e CM |
2992 | u32 nritems; |
2993 | u32 item_size; | |
2994 | int ret = 0; | |
2995 | ||
a061fc8d CM |
2996 | if (extent_key->objectid == 0) { |
2997 | ret = del_extent_zero(extent_root, path, extent_key); | |
2998 | goto out; | |
2999 | } | |
edbd8d4e CM |
3000 | key.objectid = extent_key->objectid; |
3001 | key.type = BTRFS_EXTENT_REF_KEY; | |
3002 | key.offset = 0; | |
3003 | ||
3004 | while(1) { | |
3005 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
3006 | ||
edbd8d4e CM |
3007 | if (ret < 0) |
3008 | goto out; | |
3009 | ||
3010 | ret = 0; | |
3011 | leaf = path->nodes[0]; | |
3012 | nritems = btrfs_header_nritems(leaf); | |
a061fc8d CM |
3013 | if (path->slots[0] == nritems) { |
3014 | ret = btrfs_next_leaf(extent_root, path); | |
3015 | if (ret > 0) { | |
3016 | ret = 0; | |
3017 | goto out; | |
3018 | } | |
3019 | if (ret < 0) | |
3020 | goto out; | |
bf4ef679 | 3021 | leaf = path->nodes[0]; |
a061fc8d | 3022 | } |
edbd8d4e CM |
3023 | |
3024 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
a061fc8d | 3025 | if (found_key.objectid != extent_key->objectid) { |
edbd8d4e | 3026 | break; |
a061fc8d | 3027 | } |
edbd8d4e | 3028 | |
a061fc8d | 3029 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
edbd8d4e | 3030 | break; |
a061fc8d | 3031 | } |
edbd8d4e CM |
3032 | |
3033 | key.offset = found_key.offset + 1; | |
3034 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3035 | ||
0ef3e66b CM |
3036 | ret = relocate_one_reference(extent_root, path, extent_key, |
3037 | &last_file_objectid, | |
3038 | &last_file_offset, | |
3039 | &last_file_root, last_extent); | |
edbd8d4e CM |
3040 | if (ret) |
3041 | goto out; | |
0ef3e66b | 3042 | last_extent = extent_key->objectid; |
edbd8d4e CM |
3043 | } |
3044 | ret = 0; | |
3045 | out: | |
3046 | btrfs_release_path(extent_root, path); | |
3047 | return ret; | |
3048 | } | |
3049 | ||
ec44a35c CM |
3050 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
3051 | { | |
3052 | u64 num_devices; | |
3053 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
3054 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
3055 | ||
a061fc8d | 3056 | num_devices = root->fs_info->fs_devices->num_devices; |
ec44a35c CM |
3057 | if (num_devices == 1) { |
3058 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
3059 | stripped = flags & ~stripped; | |
3060 | ||
3061 | /* turn raid0 into single device chunks */ | |
3062 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
3063 | return stripped; | |
3064 | ||
3065 | /* turn mirroring into duplication */ | |
3066 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
3067 | BTRFS_BLOCK_GROUP_RAID10)) | |
3068 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
3069 | return flags; | |
3070 | } else { | |
3071 | /* they already had raid on here, just return */ | |
ec44a35c CM |
3072 | if (flags & stripped) |
3073 | return flags; | |
3074 | ||
3075 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
3076 | stripped = flags & ~stripped; | |
3077 | ||
3078 | /* switch duplicated blocks with raid1 */ | |
3079 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
3080 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
3081 | ||
3082 | /* turn single device chunks into raid0 */ | |
3083 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
3084 | } | |
3085 | return flags; | |
3086 | } | |
3087 | ||
0ef3e66b CM |
3088 | int __alloc_chunk_for_shrink(struct btrfs_root *root, |
3089 | struct btrfs_block_group_cache *shrink_block_group, | |
3090 | int force) | |
3091 | { | |
3092 | struct btrfs_trans_handle *trans; | |
3093 | u64 new_alloc_flags; | |
3094 | u64 calc; | |
3095 | ||
c286ac48 | 3096 | spin_lock(&shrink_block_group->lock); |
0ef3e66b | 3097 | if (btrfs_block_group_used(&shrink_block_group->item) > 0) { |
c286ac48 | 3098 | spin_unlock(&shrink_block_group->lock); |
7d9eb12c | 3099 | mutex_unlock(&root->fs_info->alloc_mutex); |
c286ac48 | 3100 | |
0ef3e66b | 3101 | trans = btrfs_start_transaction(root, 1); |
7d9eb12c | 3102 | mutex_lock(&root->fs_info->alloc_mutex); |
c286ac48 | 3103 | spin_lock(&shrink_block_group->lock); |
7d9eb12c | 3104 | |
0ef3e66b CM |
3105 | new_alloc_flags = update_block_group_flags(root, |
3106 | shrink_block_group->flags); | |
3107 | if (new_alloc_flags != shrink_block_group->flags) { | |
3108 | calc = | |
3109 | btrfs_block_group_used(&shrink_block_group->item); | |
3110 | } else { | |
3111 | calc = shrink_block_group->key.offset; | |
3112 | } | |
c286ac48 CM |
3113 | spin_unlock(&shrink_block_group->lock); |
3114 | ||
0ef3e66b CM |
3115 | do_chunk_alloc(trans, root->fs_info->extent_root, |
3116 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
7d9eb12c CM |
3117 | |
3118 | mutex_unlock(&root->fs_info->alloc_mutex); | |
0ef3e66b | 3119 | btrfs_end_transaction(trans, root); |
7d9eb12c | 3120 | mutex_lock(&root->fs_info->alloc_mutex); |
c286ac48 CM |
3121 | } else |
3122 | spin_unlock(&shrink_block_group->lock); | |
0ef3e66b CM |
3123 | return 0; |
3124 | } | |
3125 | ||
8f18cf13 | 3126 | int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start) |
edbd8d4e CM |
3127 | { |
3128 | struct btrfs_trans_handle *trans; | |
3129 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
3130 | struct btrfs_path *path; | |
3131 | u64 cur_byte; | |
3132 | u64 total_found; | |
8f18cf13 CM |
3133 | u64 shrink_last_byte; |
3134 | struct btrfs_block_group_cache *shrink_block_group; | |
edbd8d4e | 3135 | struct btrfs_fs_info *info = root->fs_info; |
edbd8d4e | 3136 | struct btrfs_key key; |
73e48b27 | 3137 | struct btrfs_key found_key; |
edbd8d4e CM |
3138 | struct extent_buffer *leaf; |
3139 | u32 nritems; | |
3140 | int ret; | |
a061fc8d | 3141 | int progress; |
edbd8d4e | 3142 | |
925baedd | 3143 | mutex_lock(&root->fs_info->alloc_mutex); |
8f18cf13 CM |
3144 | shrink_block_group = btrfs_lookup_block_group(root->fs_info, |
3145 | shrink_start); | |
3146 | BUG_ON(!shrink_block_group); | |
3147 | ||
0ef3e66b CM |
3148 | shrink_last_byte = shrink_block_group->key.objectid + |
3149 | shrink_block_group->key.offset; | |
8f18cf13 CM |
3150 | |
3151 | shrink_block_group->space_info->total_bytes -= | |
3152 | shrink_block_group->key.offset; | |
edbd8d4e CM |
3153 | path = btrfs_alloc_path(); |
3154 | root = root->fs_info->extent_root; | |
8f662a76 | 3155 | path->reada = 2; |
edbd8d4e | 3156 | |
323da79c CM |
3157 | printk("btrfs relocating block group %llu flags %llu\n", |
3158 | (unsigned long long)shrink_start, | |
3159 | (unsigned long long)shrink_block_group->flags); | |
3160 | ||
0ef3e66b CM |
3161 | __alloc_chunk_for_shrink(root, shrink_block_group, 1); |
3162 | ||
edbd8d4e | 3163 | again: |
323da79c | 3164 | |
8f18cf13 CM |
3165 | shrink_block_group->ro = 1; |
3166 | ||
edbd8d4e | 3167 | total_found = 0; |
a061fc8d | 3168 | progress = 0; |
8f18cf13 | 3169 | key.objectid = shrink_start; |
edbd8d4e CM |
3170 | key.offset = 0; |
3171 | key.type = 0; | |
73e48b27 | 3172 | cur_byte = key.objectid; |
4313b399 | 3173 | |
73e48b27 Y |
3174 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3175 | if (ret < 0) | |
3176 | goto out; | |
3177 | ||
0b86a832 | 3178 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
73e48b27 Y |
3179 | if (ret < 0) |
3180 | goto out; | |
8f18cf13 | 3181 | |
73e48b27 Y |
3182 | if (ret == 0) { |
3183 | leaf = path->nodes[0]; | |
3184 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 CM |
3185 | if (found_key.objectid + found_key.offset > shrink_start && |
3186 | found_key.objectid < shrink_last_byte) { | |
73e48b27 Y |
3187 | cur_byte = found_key.objectid; |
3188 | key.objectid = cur_byte; | |
3189 | } | |
3190 | } | |
3191 | btrfs_release_path(root, path); | |
3192 | ||
3193 | while(1) { | |
edbd8d4e CM |
3194 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3195 | if (ret < 0) | |
3196 | goto out; | |
73e48b27 | 3197 | |
7d9eb12c | 3198 | next: |
edbd8d4e | 3199 | leaf = path->nodes[0]; |
73e48b27 | 3200 | nritems = btrfs_header_nritems(leaf); |
73e48b27 Y |
3201 | if (path->slots[0] >= nritems) { |
3202 | ret = btrfs_next_leaf(root, path); | |
3203 | if (ret < 0) | |
3204 | goto out; | |
3205 | if (ret == 1) { | |
3206 | ret = 0; | |
3207 | break; | |
edbd8d4e | 3208 | } |
73e48b27 Y |
3209 | leaf = path->nodes[0]; |
3210 | nritems = btrfs_header_nritems(leaf); | |
edbd8d4e | 3211 | } |
73e48b27 Y |
3212 | |
3213 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
725c8463 | 3214 | |
8f18cf13 CM |
3215 | if (found_key.objectid >= shrink_last_byte) |
3216 | break; | |
3217 | ||
725c8463 CM |
3218 | if (progress && need_resched()) { |
3219 | memcpy(&key, &found_key, sizeof(key)); | |
725c8463 | 3220 | cond_resched(); |
725c8463 CM |
3221 | btrfs_release_path(root, path); |
3222 | btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3223 | progress = 0; | |
3224 | goto next; | |
3225 | } | |
3226 | progress = 1; | |
3227 | ||
73e48b27 Y |
3228 | if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY || |
3229 | found_key.objectid + found_key.offset <= cur_byte) { | |
0ef3e66b CM |
3230 | memcpy(&key, &found_key, sizeof(key)); |
3231 | key.offset++; | |
edbd8d4e | 3232 | path->slots[0]++; |
edbd8d4e CM |
3233 | goto next; |
3234 | } | |
73e48b27 | 3235 | |
edbd8d4e CM |
3236 | total_found++; |
3237 | cur_byte = found_key.objectid + found_key.offset; | |
3238 | key.objectid = cur_byte; | |
3239 | btrfs_release_path(root, path); | |
3240 | ret = relocate_one_extent(root, path, &found_key); | |
0ef3e66b | 3241 | __alloc_chunk_for_shrink(root, shrink_block_group, 0); |
edbd8d4e CM |
3242 | } |
3243 | ||
3244 | btrfs_release_path(root, path); | |
3245 | ||
3246 | if (total_found > 0) { | |
323da79c CM |
3247 | printk("btrfs relocate found %llu last extent was %llu\n", |
3248 | (unsigned long long)total_found, | |
3249 | (unsigned long long)found_key.objectid); | |
7d9eb12c | 3250 | mutex_unlock(&root->fs_info->alloc_mutex); |
edbd8d4e CM |
3251 | trans = btrfs_start_transaction(tree_root, 1); |
3252 | btrfs_commit_transaction(trans, tree_root); | |
3253 | ||
edbd8d4e | 3254 | btrfs_clean_old_snapshots(tree_root); |
edbd8d4e | 3255 | |
3eaa2885 CM |
3256 | btrfs_wait_ordered_extents(tree_root); |
3257 | ||
edbd8d4e CM |
3258 | trans = btrfs_start_transaction(tree_root, 1); |
3259 | btrfs_commit_transaction(trans, tree_root); | |
7d9eb12c | 3260 | mutex_lock(&root->fs_info->alloc_mutex); |
edbd8d4e CM |
3261 | goto again; |
3262 | } | |
3263 | ||
8f18cf13 CM |
3264 | /* |
3265 | * we've freed all the extents, now remove the block | |
3266 | * group item from the tree | |
3267 | */ | |
7d9eb12c CM |
3268 | mutex_unlock(&root->fs_info->alloc_mutex); |
3269 | ||
edbd8d4e | 3270 | trans = btrfs_start_transaction(root, 1); |
c286ac48 | 3271 | |
7d9eb12c | 3272 | mutex_lock(&root->fs_info->alloc_mutex); |
8f18cf13 | 3273 | memcpy(&key, &shrink_block_group->key, sizeof(key)); |
1372f8e6 | 3274 | |
8f18cf13 CM |
3275 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
3276 | if (ret > 0) | |
3277 | ret = -EIO; | |
3278 | if (ret < 0) | |
3279 | goto out; | |
73e48b27 | 3280 | |
0ef3e66b CM |
3281 | clear_extent_bits(&info->block_group_cache, key.objectid, |
3282 | key.objectid + key.offset - 1, | |
8f18cf13 | 3283 | (unsigned int)-1, GFP_NOFS); |
edbd8d4e | 3284 | |
0ef3e66b CM |
3285 | |
3286 | clear_extent_bits(&info->free_space_cache, | |
3287 | key.objectid, key.objectid + key.offset - 1, | |
3288 | (unsigned int)-1, GFP_NOFS); | |
3289 | ||
3290 | memset(shrink_block_group, 0, sizeof(*shrink_block_group)); | |
3291 | kfree(shrink_block_group); | |
3292 | ||
8f18cf13 | 3293 | btrfs_del_item(trans, root, path); |
7d9eb12c CM |
3294 | btrfs_release_path(root, path); |
3295 | mutex_unlock(&root->fs_info->alloc_mutex); | |
edbd8d4e | 3296 | btrfs_commit_transaction(trans, root); |
0ef3e66b | 3297 | |
7d9eb12c CM |
3298 | mutex_lock(&root->fs_info->alloc_mutex); |
3299 | ||
0ef3e66b CM |
3300 | /* the code to unpin extents might set a few bits in the free |
3301 | * space cache for this range again | |
3302 | */ | |
3303 | clear_extent_bits(&info->free_space_cache, | |
3304 | key.objectid, key.objectid + key.offset - 1, | |
3305 | (unsigned int)-1, GFP_NOFS); | |
edbd8d4e CM |
3306 | out: |
3307 | btrfs_free_path(path); | |
925baedd | 3308 | mutex_unlock(&root->fs_info->alloc_mutex); |
edbd8d4e CM |
3309 | return ret; |
3310 | } | |
3311 | ||
0b86a832 CM |
3312 | int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, |
3313 | struct btrfs_key *key) | |
3314 | { | |
925baedd | 3315 | int ret = 0; |
0b86a832 CM |
3316 | struct btrfs_key found_key; |
3317 | struct extent_buffer *leaf; | |
3318 | int slot; | |
edbd8d4e | 3319 | |
0b86a832 CM |
3320 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
3321 | if (ret < 0) | |
925baedd CM |
3322 | goto out; |
3323 | ||
0b86a832 CM |
3324 | while(1) { |
3325 | slot = path->slots[0]; | |
edbd8d4e | 3326 | leaf = path->nodes[0]; |
0b86a832 CM |
3327 | if (slot >= btrfs_header_nritems(leaf)) { |
3328 | ret = btrfs_next_leaf(root, path); | |
3329 | if (ret == 0) | |
3330 | continue; | |
3331 | if (ret < 0) | |
925baedd | 3332 | goto out; |
0b86a832 | 3333 | break; |
edbd8d4e | 3334 | } |
0b86a832 | 3335 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 3336 | |
0b86a832 | 3337 | if (found_key.objectid >= key->objectid && |
925baedd CM |
3338 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
3339 | ret = 0; | |
3340 | goto out; | |
3341 | } | |
0b86a832 | 3342 | path->slots[0]++; |
edbd8d4e | 3343 | } |
0b86a832 | 3344 | ret = -ENOENT; |
925baedd | 3345 | out: |
0b86a832 | 3346 | return ret; |
edbd8d4e CM |
3347 | } |
3348 | ||
9078a3e1 CM |
3349 | int btrfs_read_block_groups(struct btrfs_root *root) |
3350 | { | |
3351 | struct btrfs_path *path; | |
3352 | int ret; | |
96b5179d | 3353 | int bit; |
9078a3e1 | 3354 | struct btrfs_block_group_cache *cache; |
be744175 | 3355 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 3356 | struct btrfs_space_info *space_info; |
d1310b2e | 3357 | struct extent_io_tree *block_group_cache; |
9078a3e1 CM |
3358 | struct btrfs_key key; |
3359 | struct btrfs_key found_key; | |
5f39d397 | 3360 | struct extent_buffer *leaf; |
96b5179d CM |
3361 | |
3362 | block_group_cache = &info->block_group_cache; | |
be744175 | 3363 | root = info->extent_root; |
9078a3e1 | 3364 | key.objectid = 0; |
0b86a832 | 3365 | key.offset = 0; |
9078a3e1 | 3366 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
3367 | path = btrfs_alloc_path(); |
3368 | if (!path) | |
3369 | return -ENOMEM; | |
3370 | ||
925baedd | 3371 | mutex_lock(&root->fs_info->alloc_mutex); |
9078a3e1 | 3372 | while(1) { |
0b86a832 CM |
3373 | ret = find_first_block_group(root, path, &key); |
3374 | if (ret > 0) { | |
3375 | ret = 0; | |
3376 | goto error; | |
9078a3e1 | 3377 | } |
0b86a832 CM |
3378 | if (ret != 0) |
3379 | goto error; | |
3380 | ||
5f39d397 CM |
3381 | leaf = path->nodes[0]; |
3382 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 3383 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 3384 | if (!cache) { |
0b86a832 | 3385 | ret = -ENOMEM; |
9078a3e1 CM |
3386 | break; |
3387 | } | |
3e1ad54f | 3388 | |
c286ac48 | 3389 | spin_lock_init(&cache->lock); |
5f39d397 CM |
3390 | read_extent_buffer(leaf, &cache->item, |
3391 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3392 | sizeof(cache->item)); | |
9078a3e1 | 3393 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 3394 | |
9078a3e1 CM |
3395 | key.objectid = found_key.objectid + found_key.offset; |
3396 | btrfs_release_path(root, path); | |
0b86a832 CM |
3397 | cache->flags = btrfs_block_group_flags(&cache->item); |
3398 | bit = 0; | |
3399 | if (cache->flags & BTRFS_BLOCK_GROUP_DATA) { | |
96b5179d | 3400 | bit = BLOCK_GROUP_DATA; |
0b86a832 CM |
3401 | } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
3402 | bit = BLOCK_GROUP_SYSTEM; | |
3403 | } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
96b5179d | 3404 | bit = BLOCK_GROUP_METADATA; |
31f3c99b | 3405 | } |
8790d502 | 3406 | set_avail_alloc_bits(info, cache->flags); |
96b5179d | 3407 | |
6324fbf3 CM |
3408 | ret = update_space_info(info, cache->flags, found_key.offset, |
3409 | btrfs_block_group_used(&cache->item), | |
3410 | &space_info); | |
3411 | BUG_ON(ret); | |
3412 | cache->space_info = space_info; | |
3413 | ||
96b5179d CM |
3414 | /* use EXTENT_LOCKED to prevent merging */ |
3415 | set_extent_bits(block_group_cache, found_key.objectid, | |
3416 | found_key.objectid + found_key.offset - 1, | |
c286ac48 | 3417 | EXTENT_LOCKED, GFP_NOFS); |
96b5179d | 3418 | set_state_private(block_group_cache, found_key.objectid, |
ae2f5411 | 3419 | (unsigned long)cache); |
c286ac48 CM |
3420 | set_extent_bits(block_group_cache, found_key.objectid, |
3421 | found_key.objectid + found_key.offset - 1, | |
3422 | bit | EXTENT_LOCKED, GFP_NOFS); | |
9078a3e1 | 3423 | if (key.objectid >= |
db94535d | 3424 | btrfs_super_total_bytes(&info->super_copy)) |
9078a3e1 CM |
3425 | break; |
3426 | } | |
0b86a832 CM |
3427 | ret = 0; |
3428 | error: | |
9078a3e1 | 3429 | btrfs_free_path(path); |
925baedd | 3430 | mutex_unlock(&root->fs_info->alloc_mutex); |
0b86a832 | 3431 | return ret; |
9078a3e1 | 3432 | } |
6324fbf3 CM |
3433 | |
3434 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
3435 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 3436 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
3437 | u64 size) |
3438 | { | |
3439 | int ret; | |
3440 | int bit = 0; | |
3441 | struct btrfs_root *extent_root; | |
3442 | struct btrfs_block_group_cache *cache; | |
3443 | struct extent_io_tree *block_group_cache; | |
3444 | ||
7d9eb12c | 3445 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
6324fbf3 CM |
3446 | extent_root = root->fs_info->extent_root; |
3447 | block_group_cache = &root->fs_info->block_group_cache; | |
3448 | ||
8f18cf13 | 3449 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
6324fbf3 | 3450 | BUG_ON(!cache); |
e17cade2 | 3451 | cache->key.objectid = chunk_offset; |
6324fbf3 | 3452 | cache->key.offset = size; |
c286ac48 | 3453 | spin_lock_init(&cache->lock); |
6324fbf3 | 3454 | btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
0ef3e66b | 3455 | |
6324fbf3 | 3456 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
3457 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
3458 | cache->flags = type; | |
3459 | btrfs_set_block_group_flags(&cache->item, type); | |
3460 | ||
3461 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, | |
3462 | &cache->space_info); | |
3463 | BUG_ON(ret); | |
3464 | ||
d18a2c44 | 3465 | bit = block_group_state_bits(type); |
e17cade2 CM |
3466 | set_extent_bits(block_group_cache, chunk_offset, |
3467 | chunk_offset + size - 1, | |
c286ac48 | 3468 | EXTENT_LOCKED, GFP_NOFS); |
e17cade2 CM |
3469 | set_state_private(block_group_cache, chunk_offset, |
3470 | (unsigned long)cache); | |
c286ac48 CM |
3471 | set_extent_bits(block_group_cache, chunk_offset, |
3472 | chunk_offset + size - 1, | |
3473 | bit | EXTENT_LOCKED, GFP_NOFS); | |
3474 | ||
6324fbf3 CM |
3475 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
3476 | sizeof(cache->item)); | |
3477 | BUG_ON(ret); | |
3478 | ||
3479 | finish_current_insert(trans, extent_root); | |
3480 | ret = del_pending_extents(trans, extent_root); | |
3481 | BUG_ON(ret); | |
d18a2c44 | 3482 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 3483 | |
6324fbf3 CM |
3484 | return 0; |
3485 | } |