]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
edbd8d4e | 19 | #include <linux/pagemap.h> |
ec44a35c | 20 | #include <linux/writeback.h> |
21af804c | 21 | #include <linux/blkdev.h> |
74493f7a | 22 | #include "hash.h" |
a5eb62e3 | 23 | #include "crc32c.h" |
fec577fb CM |
24 | #include "ctree.h" |
25 | #include "disk-io.h" | |
26 | #include "print-tree.h" | |
e089f05c | 27 | #include "transaction.h" |
0b86a832 | 28 | #include "volumes.h" |
925baedd | 29 | #include "locking.h" |
31153d81 | 30 | #include "ref-cache.h" |
fec577fb | 31 | |
0b86a832 | 32 | #define BLOCK_GROUP_DATA EXTENT_WRITEBACK |
96b5179d | 33 | #define BLOCK_GROUP_METADATA EXTENT_UPTODATE |
0b86a832 CM |
34 | #define BLOCK_GROUP_SYSTEM EXTENT_NEW |
35 | ||
96b5179d CM |
36 | #define BLOCK_GROUP_DIRTY EXTENT_DIRTY |
37 | ||
e089f05c CM |
38 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
39 | btrfs_root *extent_root); | |
e20d96d6 CM |
40 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
41 | btrfs_root *extent_root); | |
925baedd CM |
42 | static struct btrfs_block_group_cache * |
43 | __btrfs_find_block_group(struct btrfs_root *root, | |
44 | struct btrfs_block_group_cache *hint, | |
45 | u64 search_start, int data, int owner); | |
d548ee51 | 46 | |
925baedd CM |
47 | void maybe_lock_mutex(struct btrfs_root *root) |
48 | { | |
49 | if (root != root->fs_info->extent_root && | |
50 | root != root->fs_info->chunk_root && | |
51 | root != root->fs_info->dev_root) { | |
52 | mutex_lock(&root->fs_info->alloc_mutex); | |
53 | } | |
54 | } | |
55 | ||
56 | void maybe_unlock_mutex(struct btrfs_root *root) | |
57 | { | |
58 | if (root != root->fs_info->extent_root && | |
59 | root != root->fs_info->chunk_root && | |
60 | root != root->fs_info->dev_root) { | |
61 | mutex_unlock(&root->fs_info->alloc_mutex); | |
62 | } | |
63 | } | |
fec577fb | 64 | |
e37c9e69 CM |
65 | static int cache_block_group(struct btrfs_root *root, |
66 | struct btrfs_block_group_cache *block_group) | |
67 | { | |
68 | struct btrfs_path *path; | |
69 | int ret; | |
70 | struct btrfs_key key; | |
5f39d397 | 71 | struct extent_buffer *leaf; |
d1310b2e | 72 | struct extent_io_tree *free_space_cache; |
e37c9e69 | 73 | int slot; |
e37c9e69 CM |
74 | u64 last = 0; |
75 | u64 hole_size; | |
7d7d6068 | 76 | u64 first_free; |
e37c9e69 CM |
77 | int found = 0; |
78 | ||
00f5c795 CM |
79 | if (!block_group) |
80 | return 0; | |
81 | ||
e37c9e69 | 82 | root = root->fs_info->extent_root; |
f510cfec | 83 | free_space_cache = &root->fs_info->free_space_cache; |
e37c9e69 CM |
84 | |
85 | if (block_group->cached) | |
86 | return 0; | |
f510cfec | 87 | |
e37c9e69 CM |
88 | path = btrfs_alloc_path(); |
89 | if (!path) | |
90 | return -ENOMEM; | |
7d7d6068 | 91 | |
2cc58cf2 | 92 | path->reada = 2; |
5cd57b2c CM |
93 | /* |
94 | * we get into deadlocks with paths held by callers of this function. | |
95 | * since the alloc_mutex is protecting things right now, just | |
96 | * skip the locking here | |
97 | */ | |
98 | path->skip_locking = 1; | |
7d7d6068 | 99 | first_free = block_group->key.objectid; |
e37c9e69 | 100 | key.objectid = block_group->key.objectid; |
e37c9e69 CM |
101 | key.offset = 0; |
102 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
103 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
104 | if (ret < 0) | |
105 | return ret; | |
0b86a832 | 106 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
d548ee51 Y |
107 | if (ret < 0) |
108 | return ret; | |
109 | if (ret == 0) { | |
110 | leaf = path->nodes[0]; | |
111 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
112 | if (key.objectid + key.offset > first_free) | |
113 | first_free = key.objectid + key.offset; | |
114 | } | |
e37c9e69 | 115 | while(1) { |
5f39d397 | 116 | leaf = path->nodes[0]; |
e37c9e69 | 117 | slot = path->slots[0]; |
5f39d397 | 118 | if (slot >= btrfs_header_nritems(leaf)) { |
e37c9e69 | 119 | ret = btrfs_next_leaf(root, path); |
54aa1f4d CM |
120 | if (ret < 0) |
121 | goto err; | |
de428b63 | 122 | if (ret == 0) { |
e37c9e69 | 123 | continue; |
de428b63 | 124 | } else { |
e37c9e69 CM |
125 | break; |
126 | } | |
127 | } | |
5f39d397 | 128 | btrfs_item_key_to_cpu(leaf, &key, slot); |
7d7d6068 | 129 | if (key.objectid < block_group->key.objectid) { |
7d7d6068 Y |
130 | goto next; |
131 | } | |
e37c9e69 CM |
132 | if (key.objectid >= block_group->key.objectid + |
133 | block_group->key.offset) { | |
e37c9e69 CM |
134 | break; |
135 | } | |
7d7d6068 | 136 | |
e37c9e69 CM |
137 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
138 | if (!found) { | |
7d7d6068 | 139 | last = first_free; |
e37c9e69 | 140 | found = 1; |
e37c9e69 | 141 | } |
f510cfec CM |
142 | if (key.objectid > last) { |
143 | hole_size = key.objectid - last; | |
144 | set_extent_dirty(free_space_cache, last, | |
145 | last + hole_size - 1, | |
146 | GFP_NOFS); | |
7d7d6068 Y |
147 | } |
148 | last = key.objectid + key.offset; | |
e37c9e69 | 149 | } |
7d7d6068 | 150 | next: |
e37c9e69 CM |
151 | path->slots[0]++; |
152 | } | |
153 | ||
7d7d6068 Y |
154 | if (!found) |
155 | last = first_free; | |
156 | if (block_group->key.objectid + | |
157 | block_group->key.offset > last) { | |
158 | hole_size = block_group->key.objectid + | |
159 | block_group->key.offset - last; | |
f510cfec CM |
160 | set_extent_dirty(free_space_cache, last, |
161 | last + hole_size - 1, GFP_NOFS); | |
7d7d6068 | 162 | } |
e37c9e69 | 163 | block_group->cached = 1; |
54aa1f4d | 164 | err: |
e37c9e69 CM |
165 | btrfs_free_path(path); |
166 | return 0; | |
167 | } | |
168 | ||
0ef3e66b CM |
169 | struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct |
170 | btrfs_fs_info *info, | |
171 | u64 bytenr) | |
172 | { | |
173 | struct extent_io_tree *block_group_cache; | |
174 | struct btrfs_block_group_cache *block_group = NULL; | |
175 | u64 ptr; | |
176 | u64 start; | |
177 | u64 end; | |
178 | int ret; | |
179 | ||
180 | bytenr = max_t(u64, bytenr, | |
181 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
182 | block_group_cache = &info->block_group_cache; | |
183 | ret = find_first_extent_bit(block_group_cache, | |
184 | bytenr, &start, &end, | |
185 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | | |
186 | BLOCK_GROUP_SYSTEM); | |
187 | if (ret) { | |
188 | return NULL; | |
189 | } | |
190 | ret = get_state_private(block_group_cache, start, &ptr); | |
191 | if (ret) | |
192 | return NULL; | |
193 | ||
194 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; | |
195 | return block_group; | |
196 | } | |
197 | ||
5276aeda CM |
198 | struct btrfs_block_group_cache *btrfs_lookup_block_group(struct |
199 | btrfs_fs_info *info, | |
db94535d | 200 | u64 bytenr) |
be744175 | 201 | { |
d1310b2e | 202 | struct extent_io_tree *block_group_cache; |
96b5179d CM |
203 | struct btrfs_block_group_cache *block_group = NULL; |
204 | u64 ptr; | |
205 | u64 start; | |
206 | u64 end; | |
be744175 CM |
207 | int ret; |
208 | ||
a061fc8d CM |
209 | bytenr = max_t(u64, bytenr, |
210 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
96b5179d CM |
211 | block_group_cache = &info->block_group_cache; |
212 | ret = find_first_extent_bit(block_group_cache, | |
db94535d | 213 | bytenr, &start, &end, |
0b86a832 CM |
214 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | |
215 | BLOCK_GROUP_SYSTEM); | |
be744175 | 216 | if (ret) { |
96b5179d | 217 | return NULL; |
be744175 | 218 | } |
96b5179d CM |
219 | ret = get_state_private(block_group_cache, start, &ptr); |
220 | if (ret) | |
221 | return NULL; | |
222 | ||
ae2f5411 | 223 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
5cf66426 | 224 | if (block_group->key.objectid <= bytenr && bytenr < |
96b5179d CM |
225 | block_group->key.objectid + block_group->key.offset) |
226 | return block_group; | |
be744175 CM |
227 | return NULL; |
228 | } | |
0b86a832 CM |
229 | |
230 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) | |
231 | { | |
593060d7 | 232 | return (cache->flags & bits) == bits; |
0b86a832 CM |
233 | } |
234 | ||
235 | static int noinline find_search_start(struct btrfs_root *root, | |
98ed5174 | 236 | struct btrfs_block_group_cache **cache_ret, |
0ef3e66b | 237 | u64 *start_ret, u64 num, int data) |
e37c9e69 | 238 | { |
e37c9e69 CM |
239 | int ret; |
240 | struct btrfs_block_group_cache *cache = *cache_ret; | |
d7fc640e | 241 | struct extent_io_tree *free_space_cache; |
7d1660d4 | 242 | struct extent_state *state; |
e19caa5f | 243 | u64 last; |
f510cfec | 244 | u64 start = 0; |
257d0ce3 | 245 | u64 cache_miss = 0; |
c31f8830 | 246 | u64 total_fs_bytes; |
0b86a832 | 247 | u64 search_start = *start_ret; |
f84a8b36 | 248 | int wrapped = 0; |
e37c9e69 | 249 | |
7d9eb12c | 250 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
c31f8830 | 251 | total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); |
d7fc640e CM |
252 | free_space_cache = &root->fs_info->free_space_cache; |
253 | ||
0ef3e66b CM |
254 | if (!cache) |
255 | goto out; | |
256 | ||
e37c9e69 | 257 | again: |
54aa1f4d | 258 | ret = cache_block_group(root, cache); |
0ef3e66b | 259 | if (ret) { |
54aa1f4d | 260 | goto out; |
0ef3e66b | 261 | } |
f84a8b36 | 262 | |
e19caa5f | 263 | last = max(search_start, cache->key.objectid); |
0ef3e66b | 264 | if (!block_group_bits(cache, data) || cache->ro) |
0b86a832 | 265 | goto new_group; |
e19caa5f | 266 | |
7d1660d4 CM |
267 | spin_lock_irq(&free_space_cache->lock); |
268 | state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY); | |
e37c9e69 | 269 | while(1) { |
7d1660d4 | 270 | if (!state) { |
257d0ce3 CM |
271 | if (!cache_miss) |
272 | cache_miss = last; | |
7d1660d4 | 273 | spin_unlock_irq(&free_space_cache->lock); |
e19caa5f CM |
274 | goto new_group; |
275 | } | |
f510cfec | 276 | |
7d1660d4 CM |
277 | start = max(last, state->start); |
278 | last = state->end + 1; | |
257d0ce3 | 279 | if (last - start < num) { |
7d1660d4 CM |
280 | do { |
281 | state = extent_state_next(state); | |
282 | } while(state && !(state->state & EXTENT_DIRTY)); | |
f510cfec | 283 | continue; |
257d0ce3 | 284 | } |
7d1660d4 | 285 | spin_unlock_irq(&free_space_cache->lock); |
0ef3e66b | 286 | if (cache->ro) { |
8f18cf13 | 287 | goto new_group; |
0ef3e66b | 288 | } |
0b86a832 | 289 | if (start + num > cache->key.objectid + cache->key.offset) |
e37c9e69 | 290 | goto new_group; |
8790d502 | 291 | if (!block_group_bits(cache, data)) { |
611f0e00 | 292 | printk("block group bits don't match %Lu %d\n", cache->flags, data); |
8790d502 | 293 | } |
0b86a832 CM |
294 | *start_ret = start; |
295 | return 0; | |
8790d502 CM |
296 | } |
297 | out: | |
1a2b2ac7 CM |
298 | cache = btrfs_lookup_block_group(root->fs_info, search_start); |
299 | if (!cache) { | |
0b86a832 | 300 | printk("Unable to find block group for %Lu\n", search_start); |
1a2b2ac7 | 301 | WARN_ON(1); |
1a2b2ac7 | 302 | } |
0b86a832 | 303 | return -ENOSPC; |
e37c9e69 CM |
304 | |
305 | new_group: | |
e19caa5f | 306 | last = cache->key.objectid + cache->key.offset; |
f84a8b36 | 307 | wrapped: |
0ef3e66b | 308 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
c31f8830 | 309 | if (!cache || cache->key.objectid >= total_fs_bytes) { |
0e4de584 | 310 | no_cache: |
f84a8b36 CM |
311 | if (!wrapped) { |
312 | wrapped = 1; | |
313 | last = search_start; | |
f84a8b36 CM |
314 | goto wrapped; |
315 | } | |
1a2b2ac7 | 316 | goto out; |
e37c9e69 | 317 | } |
257d0ce3 CM |
318 | if (cache_miss && !cache->cached) { |
319 | cache_block_group(root, cache); | |
320 | last = cache_miss; | |
0ef3e66b | 321 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
257d0ce3 | 322 | } |
0ef3e66b | 323 | cache_miss = 0; |
c286ac48 | 324 | cache = btrfs_find_block_group(root, cache, last, data, 0); |
0e4de584 CM |
325 | if (!cache) |
326 | goto no_cache; | |
e37c9e69 CM |
327 | *cache_ret = cache; |
328 | goto again; | |
329 | } | |
330 | ||
84f54cfa CM |
331 | static u64 div_factor(u64 num, int factor) |
332 | { | |
257d0ce3 CM |
333 | if (factor == 10) |
334 | return num; | |
84f54cfa CM |
335 | num *= factor; |
336 | do_div(num, 10); | |
337 | return num; | |
338 | } | |
339 | ||
6324fbf3 CM |
340 | static int block_group_state_bits(u64 flags) |
341 | { | |
342 | int bits = 0; | |
343 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
344 | bits |= BLOCK_GROUP_DATA; | |
345 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
346 | bits |= BLOCK_GROUP_METADATA; | |
347 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
348 | bits |= BLOCK_GROUP_SYSTEM; | |
349 | return bits; | |
350 | } | |
351 | ||
925baedd CM |
352 | static struct btrfs_block_group_cache * |
353 | __btrfs_find_block_group(struct btrfs_root *root, | |
354 | struct btrfs_block_group_cache *hint, | |
355 | u64 search_start, int data, int owner) | |
cd1bc465 | 356 | { |
96b5179d | 357 | struct btrfs_block_group_cache *cache; |
d1310b2e | 358 | struct extent_io_tree *block_group_cache; |
31f3c99b | 359 | struct btrfs_block_group_cache *found_group = NULL; |
cd1bc465 CM |
360 | struct btrfs_fs_info *info = root->fs_info; |
361 | u64 used; | |
31f3c99b | 362 | u64 last = 0; |
96b5179d CM |
363 | u64 start; |
364 | u64 end; | |
365 | u64 free_check; | |
366 | u64 ptr; | |
367 | int bit; | |
cd1bc465 | 368 | int ret; |
31f3c99b | 369 | int full_search = 0; |
bce4eae9 | 370 | int factor = 10; |
0ef3e66b | 371 | int wrapped = 0; |
de428b63 | 372 | |
96b5179d CM |
373 | block_group_cache = &info->block_group_cache; |
374 | ||
a236aed1 CM |
375 | if (data & BTRFS_BLOCK_GROUP_METADATA) |
376 | factor = 9; | |
be744175 | 377 | |
6324fbf3 | 378 | bit = block_group_state_bits(data); |
be744175 | 379 | |
0ef3e66b | 380 | if (search_start) { |
be744175 | 381 | struct btrfs_block_group_cache *shint; |
0ef3e66b | 382 | shint = btrfs_lookup_first_block_group(info, search_start); |
8f18cf13 | 383 | if (shint && block_group_bits(shint, data) && !shint->ro) { |
c286ac48 | 384 | spin_lock(&shint->lock); |
be744175 | 385 | used = btrfs_block_group_used(&shint->item); |
324ae4df Y |
386 | if (used + shint->pinned < |
387 | div_factor(shint->key.offset, factor)) { | |
c286ac48 | 388 | spin_unlock(&shint->lock); |
be744175 CM |
389 | return shint; |
390 | } | |
c286ac48 | 391 | spin_unlock(&shint->lock); |
be744175 CM |
392 | } |
393 | } | |
0ef3e66b | 394 | if (hint && !hint->ro && block_group_bits(hint, data)) { |
c286ac48 | 395 | spin_lock(&hint->lock); |
31f3c99b | 396 | used = btrfs_block_group_used(&hint->item); |
324ae4df Y |
397 | if (used + hint->pinned < |
398 | div_factor(hint->key.offset, factor)) { | |
c286ac48 | 399 | spin_unlock(&hint->lock); |
31f3c99b CM |
400 | return hint; |
401 | } | |
c286ac48 | 402 | spin_unlock(&hint->lock); |
e19caa5f | 403 | last = hint->key.objectid + hint->key.offset; |
31f3c99b | 404 | } else { |
e37c9e69 | 405 | if (hint) |
0ef3e66b | 406 | last = max(hint->key.objectid, search_start); |
e37c9e69 | 407 | else |
0ef3e66b | 408 | last = search_start; |
31f3c99b | 409 | } |
31f3c99b | 410 | again: |
cd1bc465 | 411 | while(1) { |
96b5179d CM |
412 | ret = find_first_extent_bit(block_group_cache, last, |
413 | &start, &end, bit); | |
414 | if (ret) | |
cd1bc465 | 415 | break; |
96b5179d CM |
416 | |
417 | ret = get_state_private(block_group_cache, start, &ptr); | |
0ef3e66b CM |
418 | if (ret) { |
419 | last = end + 1; | |
420 | continue; | |
421 | } | |
96b5179d | 422 | |
ae2f5411 | 423 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
c286ac48 | 424 | spin_lock(&cache->lock); |
96b5179d CM |
425 | last = cache->key.objectid + cache->key.offset; |
426 | used = btrfs_block_group_used(&cache->item); | |
427 | ||
8f18cf13 | 428 | if (!cache->ro && block_group_bits(cache, data)) { |
0ef3e66b | 429 | free_check = div_factor(cache->key.offset, factor); |
8790d502 CM |
430 | if (used + cache->pinned < free_check) { |
431 | found_group = cache; | |
c286ac48 | 432 | spin_unlock(&cache->lock); |
8790d502 CM |
433 | goto found; |
434 | } | |
6324fbf3 | 435 | } |
c286ac48 | 436 | spin_unlock(&cache->lock); |
de428b63 | 437 | cond_resched(); |
cd1bc465 | 438 | } |
0ef3e66b CM |
439 | if (!wrapped) { |
440 | last = search_start; | |
441 | wrapped = 1; | |
442 | goto again; | |
443 | } | |
444 | if (!full_search && factor < 10) { | |
be744175 | 445 | last = search_start; |
31f3c99b | 446 | full_search = 1; |
0ef3e66b | 447 | factor = 10; |
31f3c99b CM |
448 | goto again; |
449 | } | |
be744175 | 450 | found: |
31f3c99b | 451 | return found_group; |
cd1bc465 CM |
452 | } |
453 | ||
925baedd CM |
454 | struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, |
455 | struct btrfs_block_group_cache | |
456 | *hint, u64 search_start, | |
457 | int data, int owner) | |
458 | { | |
459 | ||
460 | struct btrfs_block_group_cache *ret; | |
925baedd | 461 | ret = __btrfs_find_block_group(root, hint, search_start, data, owner); |
925baedd CM |
462 | return ret; |
463 | } | |
7bb86316 | 464 | static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation, |
74493f7a CM |
465 | u64 owner, u64 owner_offset) |
466 | { | |
467 | u32 high_crc = ~(u32)0; | |
468 | u32 low_crc = ~(u32)0; | |
469 | __le64 lenum; | |
74493f7a | 470 | lenum = cpu_to_le64(root_objectid); |
a5eb62e3 | 471 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
7bb86316 | 472 | lenum = cpu_to_le64(ref_generation); |
a5eb62e3 | 473 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d CM |
474 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) { |
475 | lenum = cpu_to_le64(owner); | |
a5eb62e3 | 476 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d | 477 | lenum = cpu_to_le64(owner_offset); |
a5eb62e3 | 478 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d | 479 | } |
74493f7a CM |
480 | return ((u64)high_crc << 32) | (u64)low_crc; |
481 | } | |
482 | ||
7bb86316 CM |
483 | static int match_extent_ref(struct extent_buffer *leaf, |
484 | struct btrfs_extent_ref *disk_ref, | |
485 | struct btrfs_extent_ref *cpu_ref) | |
486 | { | |
487 | int ret; | |
488 | int len; | |
489 | ||
490 | if (cpu_ref->objectid) | |
491 | len = sizeof(*cpu_ref); | |
492 | else | |
493 | len = 2 * sizeof(u64); | |
494 | ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref, | |
495 | len); | |
496 | return ret == 0; | |
497 | } | |
498 | ||
98ed5174 CM |
499 | static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, |
500 | struct btrfs_root *root, | |
501 | struct btrfs_path *path, u64 bytenr, | |
502 | u64 root_objectid, | |
503 | u64 ref_generation, u64 owner, | |
504 | u64 owner_offset, int del) | |
74493f7a CM |
505 | { |
506 | u64 hash; | |
507 | struct btrfs_key key; | |
7bb86316 | 508 | struct btrfs_key found_key; |
74493f7a | 509 | struct btrfs_extent_ref ref; |
7bb86316 CM |
510 | struct extent_buffer *leaf; |
511 | struct btrfs_extent_ref *disk_ref; | |
512 | int ret; | |
513 | int ret2; | |
514 | ||
515 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
516 | btrfs_set_stack_ref_generation(&ref, ref_generation); | |
517 | btrfs_set_stack_ref_objectid(&ref, owner); | |
518 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
519 | ||
520 | hash = hash_extent_ref(root_objectid, ref_generation, owner, | |
521 | owner_offset); | |
522 | key.offset = hash; | |
523 | key.objectid = bytenr; | |
524 | key.type = BTRFS_EXTENT_REF_KEY; | |
525 | ||
526 | while (1) { | |
527 | ret = btrfs_search_slot(trans, root, &key, path, | |
528 | del ? -1 : 0, del); | |
529 | if (ret < 0) | |
530 | goto out; | |
531 | leaf = path->nodes[0]; | |
532 | if (ret != 0) { | |
533 | u32 nritems = btrfs_header_nritems(leaf); | |
534 | if (path->slots[0] >= nritems) { | |
535 | ret2 = btrfs_next_leaf(root, path); | |
536 | if (ret2) | |
537 | goto out; | |
538 | leaf = path->nodes[0]; | |
539 | } | |
540 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
541 | if (found_key.objectid != bytenr || | |
542 | found_key.type != BTRFS_EXTENT_REF_KEY) | |
543 | goto out; | |
544 | key.offset = found_key.offset; | |
545 | if (del) { | |
546 | btrfs_release_path(root, path); | |
547 | continue; | |
548 | } | |
549 | } | |
550 | disk_ref = btrfs_item_ptr(path->nodes[0], | |
551 | path->slots[0], | |
552 | struct btrfs_extent_ref); | |
553 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) { | |
554 | ret = 0; | |
555 | goto out; | |
556 | } | |
557 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
558 | key.offset = found_key.offset + 1; | |
559 | btrfs_release_path(root, path); | |
560 | } | |
561 | out: | |
562 | return ret; | |
563 | } | |
564 | ||
d8d5f3e1 CM |
565 | /* |
566 | * Back reference rules. Back refs have three main goals: | |
567 | * | |
568 | * 1) differentiate between all holders of references to an extent so that | |
569 | * when a reference is dropped we can make sure it was a valid reference | |
570 | * before freeing the extent. | |
571 | * | |
572 | * 2) Provide enough information to quickly find the holders of an extent | |
573 | * if we notice a given block is corrupted or bad. | |
574 | * | |
575 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
576 | * maintenance. This is actually the same as #2, but with a slightly | |
577 | * different use case. | |
578 | * | |
579 | * File extents can be referenced by: | |
580 | * | |
581 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
582 | * - different files inside a single subvolume (in theory, not implemented yet) | |
583 | * - different offsets inside a file (bookend extents in file.c) | |
584 | * | |
585 | * The extent ref structure has fields for: | |
586 | * | |
587 | * - Objectid of the subvolume root | |
588 | * - Generation number of the tree holding the reference | |
589 | * - objectid of the file holding the reference | |
590 | * - offset in the file corresponding to the key holding the reference | |
591 | * | |
592 | * When a file extent is allocated the fields are filled in: | |
593 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
594 | * | |
595 | * When a leaf is cow'd new references are added for every file extent found | |
596 | * in the leaf. It looks the same as the create case, but trans->transid | |
597 | * will be different when the block is cow'd. | |
598 | * | |
599 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
600 | * | |
601 | * When a file extent is removed either during snapshot deletion or file | |
602 | * truncation, the corresponding back reference is found | |
603 | * by searching for: | |
604 | * | |
605 | * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), | |
606 | * inode objectid, offset in file) | |
607 | * | |
608 | * Btree extents can be referenced by: | |
609 | * | |
610 | * - Different subvolumes | |
611 | * - Different generations of the same subvolume | |
612 | * | |
613 | * Storing sufficient information for a full reverse mapping of a btree | |
614 | * block would require storing the lowest key of the block in the backref, | |
615 | * and it would require updating that lowest key either before write out or | |
616 | * every time it changed. Instead, the objectid of the lowest key is stored | |
617 | * along with the level of the tree block. This provides a hint | |
618 | * about where in the btree the block can be found. Searches through the | |
619 | * btree only need to look for a pointer to that block, so they stop one | |
620 | * level higher than the level recorded in the backref. | |
621 | * | |
622 | * Some btrees do not do reference counting on their extents. These | |
623 | * include the extent tree and the tree of tree roots. Backrefs for these | |
624 | * trees always have a generation of zero. | |
625 | * | |
626 | * When a tree block is created, back references are inserted: | |
627 | * | |
f6dbff55 | 628 | * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid) |
d8d5f3e1 CM |
629 | * |
630 | * When a tree block is cow'd in a reference counted root, | |
631 | * new back references are added for all the blocks it points to. | |
632 | * These are of the form (trans->transid will have increased since creation): | |
633 | * | |
f6dbff55 | 634 | * (root->root_key.objectid, trans->transid, level, lowest_key_objectid) |
d8d5f3e1 CM |
635 | * |
636 | * Because the lowest_key_objectid and the level are just hints | |
637 | * they are not used when backrefs are deleted. When a backref is deleted: | |
638 | * | |
639 | * if backref was for a tree root: | |
640 | * root_objectid = root->root_key.objectid | |
641 | * else | |
642 | * root_objectid = btrfs_header_owner(parent) | |
643 | * | |
644 | * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0) | |
645 | * | |
646 | * Back Reference Key hashing: | |
647 | * | |
648 | * Back references have four fields, each 64 bits long. Unfortunately, | |
649 | * This is hashed into a single 64 bit number and placed into the key offset. | |
650 | * The key objectid corresponds to the first byte in the extent, and the | |
651 | * key type is set to BTRFS_EXTENT_REF_KEY | |
652 | */ | |
7bb86316 CM |
653 | int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans, |
654 | struct btrfs_root *root, | |
655 | struct btrfs_path *path, u64 bytenr, | |
656 | u64 root_objectid, u64 ref_generation, | |
657 | u64 owner, u64 owner_offset) | |
658 | { | |
659 | u64 hash; | |
660 | struct btrfs_key key; | |
661 | struct btrfs_extent_ref ref; | |
662 | struct btrfs_extent_ref *disk_ref; | |
74493f7a CM |
663 | int ret; |
664 | ||
665 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
7bb86316 | 666 | btrfs_set_stack_ref_generation(&ref, ref_generation); |
74493f7a CM |
667 | btrfs_set_stack_ref_objectid(&ref, owner); |
668 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
669 | ||
7bb86316 CM |
670 | hash = hash_extent_ref(root_objectid, ref_generation, owner, |
671 | owner_offset); | |
74493f7a CM |
672 | key.offset = hash; |
673 | key.objectid = bytenr; | |
674 | key.type = BTRFS_EXTENT_REF_KEY; | |
675 | ||
676 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref)); | |
677 | while (ret == -EEXIST) { | |
7bb86316 CM |
678 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
679 | struct btrfs_extent_ref); | |
680 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) | |
681 | goto out; | |
682 | key.offset++; | |
683 | btrfs_release_path(root, path); | |
684 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
685 | sizeof(ref)); | |
74493f7a | 686 | } |
7bb86316 CM |
687 | if (ret) |
688 | goto out; | |
689 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
690 | struct btrfs_extent_ref); | |
691 | write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref, | |
692 | sizeof(ref)); | |
693 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
694 | out: | |
695 | btrfs_release_path(root, path); | |
696 | return ret; | |
74493f7a CM |
697 | } |
698 | ||
925baedd | 699 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
b18c6685 | 700 | struct btrfs_root *root, |
74493f7a | 701 | u64 bytenr, u64 num_bytes, |
7bb86316 | 702 | u64 root_objectid, u64 ref_generation, |
74493f7a | 703 | u64 owner, u64 owner_offset) |
02217ed2 | 704 | { |
5caf2a00 | 705 | struct btrfs_path *path; |
02217ed2 | 706 | int ret; |
e2fa7227 | 707 | struct btrfs_key key; |
5f39d397 | 708 | struct extent_buffer *l; |
234b63a0 | 709 | struct btrfs_extent_item *item; |
cf27e1ee | 710 | u32 refs; |
037e6390 | 711 | |
db94535d | 712 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 713 | path = btrfs_alloc_path(); |
54aa1f4d CM |
714 | if (!path) |
715 | return -ENOMEM; | |
26b8003f | 716 | |
3c12ac72 | 717 | path->reada = 1; |
db94535d | 718 | key.objectid = bytenr; |
62e2749e | 719 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 720 | key.offset = num_bytes; |
5caf2a00 | 721 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 722 | 0, 1); |
54aa1f4d CM |
723 | if (ret < 0) |
724 | return ret; | |
a429e513 | 725 | if (ret != 0) { |
a28ec197 | 726 | BUG(); |
a429e513 | 727 | } |
02217ed2 | 728 | BUG_ON(ret != 0); |
5f39d397 | 729 | l = path->nodes[0]; |
5caf2a00 | 730 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 CM |
731 | refs = btrfs_extent_refs(l, item); |
732 | btrfs_set_extent_refs(l, item, refs + 1); | |
5caf2a00 | 733 | btrfs_mark_buffer_dirty(path->nodes[0]); |
a28ec197 | 734 | |
5caf2a00 | 735 | btrfs_release_path(root->fs_info->extent_root, path); |
7bb86316 | 736 | |
3c12ac72 | 737 | path->reada = 1; |
7bb86316 CM |
738 | ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root, |
739 | path, bytenr, root_objectid, | |
740 | ref_generation, owner, owner_offset); | |
741 | BUG_ON(ret); | |
9f5fae2f | 742 | finish_current_insert(trans, root->fs_info->extent_root); |
e20d96d6 | 743 | del_pending_extents(trans, root->fs_info->extent_root); |
74493f7a CM |
744 | |
745 | btrfs_free_path(path); | |
02217ed2 CM |
746 | return 0; |
747 | } | |
748 | ||
925baedd CM |
749 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
750 | struct btrfs_root *root, | |
751 | u64 bytenr, u64 num_bytes, | |
752 | u64 root_objectid, u64 ref_generation, | |
753 | u64 owner, u64 owner_offset) | |
754 | { | |
755 | int ret; | |
756 | ||
757 | mutex_lock(&root->fs_info->alloc_mutex); | |
758 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, | |
759 | root_objectid, ref_generation, | |
760 | owner, owner_offset); | |
761 | mutex_unlock(&root->fs_info->alloc_mutex); | |
762 | return ret; | |
763 | } | |
764 | ||
e9d0b13b CM |
765 | int btrfs_extent_post_op(struct btrfs_trans_handle *trans, |
766 | struct btrfs_root *root) | |
767 | { | |
768 | finish_current_insert(trans, root->fs_info->extent_root); | |
769 | del_pending_extents(trans, root->fs_info->extent_root); | |
770 | return 0; | |
771 | } | |
772 | ||
b18c6685 | 773 | static int lookup_extent_ref(struct btrfs_trans_handle *trans, |
db94535d CM |
774 | struct btrfs_root *root, u64 bytenr, |
775 | u64 num_bytes, u32 *refs) | |
a28ec197 | 776 | { |
5caf2a00 | 777 | struct btrfs_path *path; |
a28ec197 | 778 | int ret; |
e2fa7227 | 779 | struct btrfs_key key; |
5f39d397 | 780 | struct extent_buffer *l; |
234b63a0 | 781 | struct btrfs_extent_item *item; |
5caf2a00 | 782 | |
db94535d | 783 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 784 | path = btrfs_alloc_path(); |
3c12ac72 | 785 | path->reada = 1; |
db94535d CM |
786 | key.objectid = bytenr; |
787 | key.offset = num_bytes; | |
62e2749e | 788 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
5caf2a00 | 789 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 790 | 0, 0); |
54aa1f4d CM |
791 | if (ret < 0) |
792 | goto out; | |
5f39d397 CM |
793 | if (ret != 0) { |
794 | btrfs_print_leaf(root, path->nodes[0]); | |
db94535d | 795 | printk("failed to find block number %Lu\n", bytenr); |
a28ec197 | 796 | BUG(); |
5f39d397 CM |
797 | } |
798 | l = path->nodes[0]; | |
5caf2a00 | 799 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 | 800 | *refs = btrfs_extent_refs(l, item); |
54aa1f4d | 801 | out: |
5caf2a00 | 802 | btrfs_free_path(path); |
a28ec197 CM |
803 | return 0; |
804 | } | |
805 | ||
f321e491 YZ |
806 | |
807 | static int get_reference_status(struct btrfs_root *root, u64 bytenr, | |
808 | u64 parent_gen, u64 ref_objectid, | |
809 | u64 *min_generation, u32 *ref_count) | |
be20aa9d CM |
810 | { |
811 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
812 | struct btrfs_path *path; | |
f321e491 YZ |
813 | struct extent_buffer *leaf; |
814 | struct btrfs_extent_ref *ref_item; | |
815 | struct btrfs_key key; | |
816 | struct btrfs_key found_key; | |
56b453c9 | 817 | u64 root_objectid = root->root_key.objectid; |
f321e491 | 818 | u64 ref_generation; |
be20aa9d CM |
819 | u32 nritems; |
820 | int ret; | |
925baedd | 821 | |
be20aa9d CM |
822 | key.objectid = bytenr; |
823 | key.offset = 0; | |
f321e491 | 824 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 825 | |
f321e491 YZ |
826 | path = btrfs_alloc_path(); |
827 | mutex_lock(&root->fs_info->alloc_mutex); | |
be20aa9d CM |
828 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
829 | if (ret < 0) | |
830 | goto out; | |
831 | BUG_ON(ret == 0); | |
832 | ||
f321e491 YZ |
833 | leaf = path->nodes[0]; |
834 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
be20aa9d CM |
835 | |
836 | if (found_key.objectid != bytenr || | |
837 | found_key.type != BTRFS_EXTENT_ITEM_KEY) { | |
f321e491 | 838 | ret = 1; |
be20aa9d CM |
839 | goto out; |
840 | } | |
841 | ||
f321e491 YZ |
842 | *ref_count = 0; |
843 | *min_generation = (u64)-1; | |
844 | ||
be20aa9d | 845 | while (1) { |
f321e491 YZ |
846 | leaf = path->nodes[0]; |
847 | nritems = btrfs_header_nritems(leaf); | |
be20aa9d CM |
848 | if (path->slots[0] >= nritems) { |
849 | ret = btrfs_next_leaf(extent_root, path); | |
f321e491 YZ |
850 | if (ret < 0) |
851 | goto out; | |
be20aa9d CM |
852 | if (ret == 0) |
853 | continue; | |
854 | break; | |
855 | } | |
f321e491 | 856 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
be20aa9d CM |
857 | if (found_key.objectid != bytenr) |
858 | break; | |
bd09835d | 859 | |
be20aa9d CM |
860 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
861 | path->slots[0]++; | |
862 | continue; | |
863 | } | |
864 | ||
f321e491 | 865 | ref_item = btrfs_item_ptr(leaf, path->slots[0], |
be20aa9d | 866 | struct btrfs_extent_ref); |
f321e491 YZ |
867 | ref_generation = btrfs_ref_generation(leaf, ref_item); |
868 | /* | |
869 | * For (parent_gen > 0 && parent_gen > ref_gen): | |
870 | * | |
bcc63abb Y |
871 | * we reach here through the oldest root, therefore |
872 | * all other reference from same snapshot should have | |
f321e491 YZ |
873 | * a larger generation. |
874 | */ | |
875 | if ((root_objectid != btrfs_ref_root(leaf, ref_item)) || | |
876 | (parent_gen > 0 && parent_gen > ref_generation) || | |
877 | (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID && | |
878 | ref_objectid != btrfs_ref_objectid(leaf, ref_item))) { | |
879 | if (ref_count) | |
880 | *ref_count = 2; | |
881 | break; | |
a68d5933 | 882 | } |
f321e491 YZ |
883 | |
884 | *ref_count = 1; | |
885 | if (*min_generation > ref_generation) | |
886 | *min_generation = ref_generation; | |
887 | ||
be20aa9d CM |
888 | path->slots[0]++; |
889 | } | |
f321e491 YZ |
890 | ret = 0; |
891 | out: | |
892 | mutex_unlock(&root->fs_info->alloc_mutex); | |
893 | btrfs_free_path(path); | |
894 | return ret; | |
895 | } | |
896 | ||
7ea394f1 YZ |
897 | int btrfs_cross_ref_exists(struct btrfs_trans_handle *trans, |
898 | struct btrfs_root *root, | |
f321e491 YZ |
899 | struct btrfs_key *key, u64 bytenr) |
900 | { | |
f321e491 YZ |
901 | struct btrfs_root *old_root; |
902 | struct btrfs_path *path = NULL; | |
903 | struct extent_buffer *eb; | |
904 | struct btrfs_file_extent_item *item; | |
905 | u64 ref_generation; | |
906 | u64 min_generation; | |
907 | u64 extent_start; | |
908 | u32 ref_count; | |
909 | int level; | |
910 | int ret; | |
911 | ||
7ea394f1 | 912 | BUG_ON(trans == NULL); |
f321e491 YZ |
913 | BUG_ON(key->type != BTRFS_EXTENT_DATA_KEY); |
914 | ret = get_reference_status(root, bytenr, 0, key->objectid, | |
915 | &min_generation, &ref_count); | |
916 | if (ret) | |
917 | return ret; | |
918 | ||
919 | if (ref_count != 1) | |
920 | return 1; | |
921 | ||
f321e491 YZ |
922 | old_root = root->dirty_root->root; |
923 | ref_generation = old_root->root_key.offset; | |
924 | ||
925 | /* all references are created in running transaction */ | |
926 | if (min_generation > ref_generation) { | |
927 | ret = 0; | |
bbaf549e CM |
928 | goto out; |
929 | } | |
f321e491 YZ |
930 | |
931 | path = btrfs_alloc_path(); | |
932 | if (!path) { | |
933 | ret = -ENOMEM; | |
be20aa9d CM |
934 | goto out; |
935 | } | |
f321e491 YZ |
936 | |
937 | path->skip_locking = 1; | |
938 | /* if no item found, the extent is referenced by other snapshot */ | |
939 | ret = btrfs_search_slot(NULL, old_root, key, path, 0, 0); | |
940 | if (ret) | |
be20aa9d | 941 | goto out; |
be20aa9d | 942 | |
f321e491 YZ |
943 | eb = path->nodes[0]; |
944 | item = btrfs_item_ptr(eb, path->slots[0], | |
945 | struct btrfs_file_extent_item); | |
946 | if (btrfs_file_extent_type(eb, item) != BTRFS_FILE_EXTENT_REG || | |
947 | btrfs_file_extent_disk_bytenr(eb, item) != bytenr) { | |
948 | ret = 1; | |
949 | goto out; | |
950 | } | |
951 | ||
952 | for (level = BTRFS_MAX_LEVEL - 1; level >= -1; level--) { | |
953 | if (level >= 0) { | |
954 | eb = path->nodes[level]; | |
955 | if (!eb) | |
956 | continue; | |
957 | extent_start = eb->start; | |
bcc63abb | 958 | } else |
f321e491 YZ |
959 | extent_start = bytenr; |
960 | ||
961 | ret = get_reference_status(root, extent_start, ref_generation, | |
962 | 0, &min_generation, &ref_count); | |
963 | if (ret) | |
964 | goto out; | |
965 | ||
966 | if (ref_count != 1) { | |
967 | ret = 1; | |
968 | goto out; | |
969 | } | |
970 | if (level >= 0) | |
971 | ref_generation = btrfs_header_generation(eb); | |
972 | } | |
973 | ret = 0; | |
be20aa9d | 974 | out: |
f321e491 YZ |
975 | if (path) |
976 | btrfs_free_path(path); | |
f321e491 | 977 | return ret; |
be20aa9d | 978 | } |
c5739bba | 979 | |
e089f05c | 980 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
31153d81 | 981 | struct extent_buffer *buf, int cache_ref) |
02217ed2 | 982 | { |
db94535d | 983 | u64 bytenr; |
5f39d397 CM |
984 | u32 nritems; |
985 | struct btrfs_key key; | |
6407bf6d | 986 | struct btrfs_file_extent_item *fi; |
02217ed2 | 987 | int i; |
db94535d | 988 | int level; |
6407bf6d | 989 | int ret; |
54aa1f4d | 990 | int faili; |
31153d81 | 991 | int nr_file_extents = 0; |
a28ec197 | 992 | |
3768f368 | 993 | if (!root->ref_cows) |
a28ec197 | 994 | return 0; |
5f39d397 | 995 | |
db94535d | 996 | level = btrfs_header_level(buf); |
5f39d397 CM |
997 | nritems = btrfs_header_nritems(buf); |
998 | for (i = 0; i < nritems; i++) { | |
e34a5b4f | 999 | cond_resched(); |
db94535d CM |
1000 | if (level == 0) { |
1001 | u64 disk_bytenr; | |
5f39d397 CM |
1002 | btrfs_item_key_to_cpu(buf, &key, i); |
1003 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d | 1004 | continue; |
5f39d397 | 1005 | fi = btrfs_item_ptr(buf, i, |
6407bf6d | 1006 | struct btrfs_file_extent_item); |
5f39d397 | 1007 | if (btrfs_file_extent_type(buf, fi) == |
236454df CM |
1008 | BTRFS_FILE_EXTENT_INLINE) |
1009 | continue; | |
db94535d CM |
1010 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
1011 | if (disk_bytenr == 0) | |
3a686375 | 1012 | continue; |
4a096752 | 1013 | |
31153d81 YZ |
1014 | if (buf != root->commit_root) |
1015 | nr_file_extents++; | |
1016 | ||
4a096752 | 1017 | mutex_lock(&root->fs_info->alloc_mutex); |
925baedd | 1018 | ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr, |
7bb86316 CM |
1019 | btrfs_file_extent_disk_num_bytes(buf, fi), |
1020 | root->root_key.objectid, trans->transid, | |
1021 | key.objectid, key.offset); | |
4a096752 | 1022 | mutex_unlock(&root->fs_info->alloc_mutex); |
54aa1f4d CM |
1023 | if (ret) { |
1024 | faili = i; | |
4a096752 | 1025 | WARN_ON(1); |
54aa1f4d CM |
1026 | goto fail; |
1027 | } | |
6407bf6d | 1028 | } else { |
db94535d | 1029 | bytenr = btrfs_node_blockptr(buf, i); |
6caab489 | 1030 | btrfs_node_key_to_cpu(buf, &key, i); |
4a096752 CM |
1031 | |
1032 | mutex_lock(&root->fs_info->alloc_mutex); | |
925baedd | 1033 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, |
7bb86316 CM |
1034 | btrfs_level_size(root, level - 1), |
1035 | root->root_key.objectid, | |
f6dbff55 CM |
1036 | trans->transid, |
1037 | level - 1, key.objectid); | |
4a096752 | 1038 | mutex_unlock(&root->fs_info->alloc_mutex); |
54aa1f4d CM |
1039 | if (ret) { |
1040 | faili = i; | |
4a096752 | 1041 | WARN_ON(1); |
54aa1f4d CM |
1042 | goto fail; |
1043 | } | |
6407bf6d | 1044 | } |
02217ed2 | 1045 | } |
31153d81 YZ |
1046 | /* cache orignal leaf block's references */ |
1047 | if (level == 0 && cache_ref && buf != root->commit_root) { | |
1048 | struct btrfs_leaf_ref *ref; | |
1049 | struct btrfs_extent_info *info; | |
1050 | ||
bcc63abb | 1051 | ref = btrfs_alloc_leaf_ref(root, nr_file_extents); |
31153d81 YZ |
1052 | if (!ref) { |
1053 | WARN_ON(1); | |
1054 | goto out; | |
1055 | } | |
1056 | ||
47ac14fa | 1057 | ref->root_gen = root->root_key.offset; |
31153d81 YZ |
1058 | ref->bytenr = buf->start; |
1059 | ref->owner = btrfs_header_owner(buf); | |
1060 | ref->generation = btrfs_header_generation(buf); | |
1061 | ref->nritems = nr_file_extents; | |
1062 | info = ref->extents; | |
bcc63abb | 1063 | |
31153d81 YZ |
1064 | for (i = 0; nr_file_extents > 0 && i < nritems; i++) { |
1065 | u64 disk_bytenr; | |
1066 | btrfs_item_key_to_cpu(buf, &key, i); | |
1067 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
1068 | continue; | |
1069 | fi = btrfs_item_ptr(buf, i, | |
1070 | struct btrfs_file_extent_item); | |
1071 | if (btrfs_file_extent_type(buf, fi) == | |
1072 | BTRFS_FILE_EXTENT_INLINE) | |
1073 | continue; | |
1074 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
1075 | if (disk_bytenr == 0) | |
1076 | continue; | |
1077 | ||
1078 | info->bytenr = disk_bytenr; | |
1079 | info->num_bytes = | |
1080 | btrfs_file_extent_disk_num_bytes(buf, fi); | |
1081 | info->objectid = key.objectid; | |
1082 | info->offset = key.offset; | |
1083 | info++; | |
1084 | } | |
1085 | ||
1086 | BUG_ON(!root->ref_tree); | |
1087 | ret = btrfs_add_leaf_ref(root, ref); | |
1088 | WARN_ON(ret); | |
bcc63abb | 1089 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
1090 | } |
1091 | out: | |
02217ed2 | 1092 | return 0; |
54aa1f4d | 1093 | fail: |
ccd467d6 | 1094 | WARN_ON(1); |
7bb86316 | 1095 | #if 0 |
54aa1f4d | 1096 | for (i =0; i < faili; i++) { |
db94535d CM |
1097 | if (level == 0) { |
1098 | u64 disk_bytenr; | |
5f39d397 CM |
1099 | btrfs_item_key_to_cpu(buf, &key, i); |
1100 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
54aa1f4d | 1101 | continue; |
5f39d397 | 1102 | fi = btrfs_item_ptr(buf, i, |
54aa1f4d | 1103 | struct btrfs_file_extent_item); |
5f39d397 | 1104 | if (btrfs_file_extent_type(buf, fi) == |
54aa1f4d CM |
1105 | BTRFS_FILE_EXTENT_INLINE) |
1106 | continue; | |
db94535d CM |
1107 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
1108 | if (disk_bytenr == 0) | |
54aa1f4d | 1109 | continue; |
db94535d CM |
1110 | err = btrfs_free_extent(trans, root, disk_bytenr, |
1111 | btrfs_file_extent_disk_num_bytes(buf, | |
5f39d397 | 1112 | fi), 0); |
54aa1f4d CM |
1113 | BUG_ON(err); |
1114 | } else { | |
db94535d CM |
1115 | bytenr = btrfs_node_blockptr(buf, i); |
1116 | err = btrfs_free_extent(trans, root, bytenr, | |
1117 | btrfs_level_size(root, level - 1), 0); | |
54aa1f4d CM |
1118 | BUG_ON(err); |
1119 | } | |
1120 | } | |
7bb86316 | 1121 | #endif |
54aa1f4d | 1122 | return ret; |
02217ed2 CM |
1123 | } |
1124 | ||
9078a3e1 CM |
1125 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
1126 | struct btrfs_root *root, | |
1127 | struct btrfs_path *path, | |
1128 | struct btrfs_block_group_cache *cache) | |
1129 | { | |
1130 | int ret; | |
1131 | int pending_ret; | |
1132 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
5f39d397 CM |
1133 | unsigned long bi; |
1134 | struct extent_buffer *leaf; | |
9078a3e1 | 1135 | |
9078a3e1 | 1136 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
1137 | if (ret < 0) |
1138 | goto fail; | |
9078a3e1 | 1139 | BUG_ON(ret); |
5f39d397 CM |
1140 | |
1141 | leaf = path->nodes[0]; | |
1142 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1143 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
1144 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 1145 | btrfs_release_path(extent_root, path); |
54aa1f4d | 1146 | fail: |
9078a3e1 CM |
1147 | finish_current_insert(trans, extent_root); |
1148 | pending_ret = del_pending_extents(trans, extent_root); | |
1149 | if (ret) | |
1150 | return ret; | |
1151 | if (pending_ret) | |
1152 | return pending_ret; | |
1153 | return 0; | |
1154 | ||
1155 | } | |
1156 | ||
96b5179d CM |
1157 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
1158 | struct btrfs_root *root) | |
9078a3e1 | 1159 | { |
d1310b2e | 1160 | struct extent_io_tree *block_group_cache; |
96b5179d | 1161 | struct btrfs_block_group_cache *cache; |
9078a3e1 CM |
1162 | int ret; |
1163 | int err = 0; | |
1164 | int werr = 0; | |
9078a3e1 | 1165 | struct btrfs_path *path; |
96b5179d CM |
1166 | u64 last = 0; |
1167 | u64 start; | |
1168 | u64 end; | |
1169 | u64 ptr; | |
9078a3e1 | 1170 | |
96b5179d | 1171 | block_group_cache = &root->fs_info->block_group_cache; |
9078a3e1 CM |
1172 | path = btrfs_alloc_path(); |
1173 | if (!path) | |
1174 | return -ENOMEM; | |
1175 | ||
925baedd | 1176 | mutex_lock(&root->fs_info->alloc_mutex); |
9078a3e1 | 1177 | while(1) { |
96b5179d CM |
1178 | ret = find_first_extent_bit(block_group_cache, last, |
1179 | &start, &end, BLOCK_GROUP_DIRTY); | |
1180 | if (ret) | |
9078a3e1 | 1181 | break; |
54aa1f4d | 1182 | |
96b5179d CM |
1183 | last = end + 1; |
1184 | ret = get_state_private(block_group_cache, start, &ptr); | |
1185 | if (ret) | |
1186 | break; | |
ae2f5411 | 1187 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
96b5179d CM |
1188 | err = write_one_cache_group(trans, root, |
1189 | path, cache); | |
1190 | /* | |
1191 | * if we fail to write the cache group, we want | |
1192 | * to keep it marked dirty in hopes that a later | |
1193 | * write will work | |
1194 | */ | |
1195 | if (err) { | |
1196 | werr = err; | |
1197 | continue; | |
9078a3e1 | 1198 | } |
96b5179d CM |
1199 | clear_extent_bits(block_group_cache, start, end, |
1200 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 CM |
1201 | } |
1202 | btrfs_free_path(path); | |
925baedd | 1203 | mutex_unlock(&root->fs_info->alloc_mutex); |
9078a3e1 CM |
1204 | return werr; |
1205 | } | |
1206 | ||
6324fbf3 CM |
1207 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
1208 | u64 flags) | |
1209 | { | |
1210 | struct list_head *head = &info->space_info; | |
1211 | struct list_head *cur; | |
1212 | struct btrfs_space_info *found; | |
1213 | list_for_each(cur, head) { | |
1214 | found = list_entry(cur, struct btrfs_space_info, list); | |
1215 | if (found->flags == flags) | |
1216 | return found; | |
1217 | } | |
1218 | return NULL; | |
1219 | ||
1220 | } | |
1221 | ||
593060d7 CM |
1222 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
1223 | u64 total_bytes, u64 bytes_used, | |
1224 | struct btrfs_space_info **space_info) | |
1225 | { | |
1226 | struct btrfs_space_info *found; | |
1227 | ||
1228 | found = __find_space_info(info, flags); | |
1229 | if (found) { | |
1230 | found->total_bytes += total_bytes; | |
1231 | found->bytes_used += bytes_used; | |
8f18cf13 | 1232 | found->full = 0; |
593060d7 CM |
1233 | *space_info = found; |
1234 | return 0; | |
1235 | } | |
1236 | found = kmalloc(sizeof(*found), GFP_NOFS); | |
1237 | if (!found) | |
1238 | return -ENOMEM; | |
1239 | ||
1240 | list_add(&found->list, &info->space_info); | |
1241 | found->flags = flags; | |
1242 | found->total_bytes = total_bytes; | |
1243 | found->bytes_used = bytes_used; | |
1244 | found->bytes_pinned = 0; | |
1245 | found->full = 0; | |
0ef3e66b | 1246 | found->force_alloc = 0; |
593060d7 CM |
1247 | *space_info = found; |
1248 | return 0; | |
1249 | } | |
1250 | ||
8790d502 CM |
1251 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
1252 | { | |
1253 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 1254 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 1255 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 1256 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
1257 | if (extra_flags) { |
1258 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1259 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1260 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1261 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1262 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1263 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1264 | } | |
1265 | } | |
593060d7 | 1266 | |
a061fc8d | 1267 | static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 1268 | { |
a061fc8d CM |
1269 | u64 num_devices = root->fs_info->fs_devices->num_devices; |
1270 | ||
1271 | if (num_devices == 1) | |
1272 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
1273 | if (num_devices < 4) | |
1274 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
1275 | ||
ec44a35c CM |
1276 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
1277 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 1278 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 1279 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 1280 | } |
ec44a35c CM |
1281 | |
1282 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 1283 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 1284 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 1285 | } |
ec44a35c CM |
1286 | |
1287 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
1288 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
1289 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
1290 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
1291 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
1292 | return flags; | |
1293 | } | |
1294 | ||
6324fbf3 CM |
1295 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
1296 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 1297 | u64 flags, int force) |
6324fbf3 CM |
1298 | { |
1299 | struct btrfs_space_info *space_info; | |
1300 | u64 thresh; | |
1301 | u64 start; | |
1302 | u64 num_bytes; | |
1303 | int ret; | |
1304 | ||
a061fc8d | 1305 | flags = reduce_alloc_profile(extent_root, flags); |
ec44a35c | 1306 | |
6324fbf3 | 1307 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
1308 | if (!space_info) { |
1309 | ret = update_space_info(extent_root->fs_info, flags, | |
1310 | 0, 0, &space_info); | |
1311 | BUG_ON(ret); | |
1312 | } | |
6324fbf3 CM |
1313 | BUG_ON(!space_info); |
1314 | ||
0ef3e66b CM |
1315 | if (space_info->force_alloc) { |
1316 | force = 1; | |
1317 | space_info->force_alloc = 0; | |
1318 | } | |
6324fbf3 | 1319 | if (space_info->full) |
925baedd | 1320 | goto out; |
6324fbf3 | 1321 | |
8790d502 | 1322 | thresh = div_factor(space_info->total_bytes, 6); |
0ef3e66b CM |
1323 | if (!force && |
1324 | (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) < | |
6324fbf3 | 1325 | thresh) |
925baedd | 1326 | goto out; |
6324fbf3 | 1327 | |
925baedd | 1328 | mutex_lock(&extent_root->fs_info->chunk_mutex); |
6324fbf3 CM |
1329 | ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags); |
1330 | if (ret == -ENOSPC) { | |
1331 | printk("space info full %Lu\n", flags); | |
1332 | space_info->full = 1; | |
a74a4b97 | 1333 | goto out_unlock; |
6324fbf3 | 1334 | } |
6324fbf3 CM |
1335 | BUG_ON(ret); |
1336 | ||
1337 | ret = btrfs_make_block_group(trans, extent_root, 0, flags, | |
e17cade2 | 1338 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes); |
6324fbf3 | 1339 | BUG_ON(ret); |
a74a4b97 | 1340 | out_unlock: |
333db94c | 1341 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
a74a4b97 | 1342 | out: |
6324fbf3 CM |
1343 | return 0; |
1344 | } | |
1345 | ||
9078a3e1 CM |
1346 | static int update_block_group(struct btrfs_trans_handle *trans, |
1347 | struct btrfs_root *root, | |
db94535d | 1348 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 1349 | int mark_free) |
9078a3e1 CM |
1350 | { |
1351 | struct btrfs_block_group_cache *cache; | |
1352 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 1353 | u64 total = num_bytes; |
9078a3e1 | 1354 | u64 old_val; |
db94535d | 1355 | u64 byte_in_group; |
96b5179d CM |
1356 | u64 start; |
1357 | u64 end; | |
3e1ad54f | 1358 | |
7d9eb12c | 1359 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
9078a3e1 | 1360 | while(total) { |
db94535d | 1361 | cache = btrfs_lookup_block_group(info, bytenr); |
3e1ad54f | 1362 | if (!cache) { |
9078a3e1 | 1363 | return -1; |
cd1bc465 | 1364 | } |
db94535d CM |
1365 | byte_in_group = bytenr - cache->key.objectid; |
1366 | WARN_ON(byte_in_group > cache->key.offset); | |
96b5179d CM |
1367 | start = cache->key.objectid; |
1368 | end = start + cache->key.offset - 1; | |
1369 | set_extent_bits(&info->block_group_cache, start, end, | |
1370 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 | 1371 | |
c286ac48 | 1372 | spin_lock(&cache->lock); |
9078a3e1 | 1373 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 1374 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 1375 | if (alloc) { |
db94535d | 1376 | old_val += num_bytes; |
6324fbf3 | 1377 | cache->space_info->bytes_used += num_bytes; |
c286ac48 CM |
1378 | btrfs_set_block_group_used(&cache->item, old_val); |
1379 | spin_unlock(&cache->lock); | |
cd1bc465 | 1380 | } else { |
db94535d | 1381 | old_val -= num_bytes; |
6324fbf3 | 1382 | cache->space_info->bytes_used -= num_bytes; |
c286ac48 CM |
1383 | btrfs_set_block_group_used(&cache->item, old_val); |
1384 | spin_unlock(&cache->lock); | |
f510cfec CM |
1385 | if (mark_free) { |
1386 | set_extent_dirty(&info->free_space_cache, | |
db94535d | 1387 | bytenr, bytenr + num_bytes - 1, |
f510cfec | 1388 | GFP_NOFS); |
e37c9e69 | 1389 | } |
cd1bc465 | 1390 | } |
db94535d CM |
1391 | total -= num_bytes; |
1392 | bytenr += num_bytes; | |
9078a3e1 CM |
1393 | } |
1394 | return 0; | |
1395 | } | |
6324fbf3 | 1396 | |
a061fc8d CM |
1397 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
1398 | { | |
1399 | u64 start; | |
1400 | u64 end; | |
1401 | int ret; | |
1402 | ret = find_first_extent_bit(&root->fs_info->block_group_cache, | |
1403 | search_start, &start, &end, | |
1404 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | | |
1405 | BLOCK_GROUP_SYSTEM); | |
1406 | if (ret) | |
1407 | return 0; | |
1408 | return start; | |
1409 | } | |
1410 | ||
1411 | ||
324ae4df Y |
1412 | static int update_pinned_extents(struct btrfs_root *root, |
1413 | u64 bytenr, u64 num, int pin) | |
1414 | { | |
1415 | u64 len; | |
1416 | struct btrfs_block_group_cache *cache; | |
1417 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1418 | ||
7d9eb12c | 1419 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
324ae4df Y |
1420 | if (pin) { |
1421 | set_extent_dirty(&fs_info->pinned_extents, | |
1422 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1423 | } else { | |
1424 | clear_extent_dirty(&fs_info->pinned_extents, | |
1425 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1426 | } | |
1427 | while (num > 0) { | |
1428 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
a061fc8d CM |
1429 | if (!cache) { |
1430 | u64 first = first_logical_byte(root, bytenr); | |
1431 | WARN_ON(first < bytenr); | |
1432 | len = min(first - bytenr, num); | |
1433 | } else { | |
1434 | len = min(num, cache->key.offset - | |
1435 | (bytenr - cache->key.objectid)); | |
1436 | } | |
324ae4df | 1437 | if (pin) { |
a061fc8d | 1438 | if (cache) { |
c286ac48 | 1439 | spin_lock(&cache->lock); |
a061fc8d CM |
1440 | cache->pinned += len; |
1441 | cache->space_info->bytes_pinned += len; | |
c286ac48 | 1442 | spin_unlock(&cache->lock); |
a061fc8d | 1443 | } |
324ae4df Y |
1444 | fs_info->total_pinned += len; |
1445 | } else { | |
a061fc8d | 1446 | if (cache) { |
c286ac48 | 1447 | spin_lock(&cache->lock); |
a061fc8d CM |
1448 | cache->pinned -= len; |
1449 | cache->space_info->bytes_pinned -= len; | |
c286ac48 | 1450 | spin_unlock(&cache->lock); |
a061fc8d | 1451 | } |
324ae4df Y |
1452 | fs_info->total_pinned -= len; |
1453 | } | |
1454 | bytenr += len; | |
1455 | num -= len; | |
1456 | } | |
1457 | return 0; | |
1458 | } | |
9078a3e1 | 1459 | |
d1310b2e | 1460 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) |
ccd467d6 | 1461 | { |
ccd467d6 | 1462 | u64 last = 0; |
1a5bc167 CM |
1463 | u64 start; |
1464 | u64 end; | |
d1310b2e | 1465 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; |
ccd467d6 | 1466 | int ret; |
ccd467d6 CM |
1467 | |
1468 | while(1) { | |
1a5bc167 CM |
1469 | ret = find_first_extent_bit(pinned_extents, last, |
1470 | &start, &end, EXTENT_DIRTY); | |
1471 | if (ret) | |
ccd467d6 | 1472 | break; |
1a5bc167 CM |
1473 | set_extent_dirty(copy, start, end, GFP_NOFS); |
1474 | last = end + 1; | |
ccd467d6 CM |
1475 | } |
1476 | return 0; | |
1477 | } | |
1478 | ||
1479 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
1480 | struct btrfs_root *root, | |
d1310b2e | 1481 | struct extent_io_tree *unpin) |
a28ec197 | 1482 | { |
1a5bc167 CM |
1483 | u64 start; |
1484 | u64 end; | |
a28ec197 | 1485 | int ret; |
d1310b2e | 1486 | struct extent_io_tree *free_space_cache; |
f510cfec | 1487 | free_space_cache = &root->fs_info->free_space_cache; |
a28ec197 | 1488 | |
925baedd | 1489 | mutex_lock(&root->fs_info->alloc_mutex); |
a28ec197 | 1490 | while(1) { |
1a5bc167 CM |
1491 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
1492 | EXTENT_DIRTY); | |
1493 | if (ret) | |
a28ec197 | 1494 | break; |
324ae4df | 1495 | update_pinned_extents(root, start, end + 1 - start, 0); |
1a5bc167 CM |
1496 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
1497 | set_extent_dirty(free_space_cache, start, end, GFP_NOFS); | |
c286ac48 CM |
1498 | if (need_resched()) { |
1499 | mutex_unlock(&root->fs_info->alloc_mutex); | |
1500 | cond_resched(); | |
1501 | mutex_lock(&root->fs_info->alloc_mutex); | |
1502 | } | |
a28ec197 | 1503 | } |
925baedd | 1504 | mutex_unlock(&root->fs_info->alloc_mutex); |
a28ec197 CM |
1505 | return 0; |
1506 | } | |
1507 | ||
98ed5174 CM |
1508 | static int finish_current_insert(struct btrfs_trans_handle *trans, |
1509 | struct btrfs_root *extent_root) | |
037e6390 | 1510 | { |
7bb86316 CM |
1511 | u64 start; |
1512 | u64 end; | |
1513 | struct btrfs_fs_info *info = extent_root->fs_info; | |
d8d5f3e1 | 1514 | struct extent_buffer *eb; |
7bb86316 | 1515 | struct btrfs_path *path; |
e2fa7227 | 1516 | struct btrfs_key ins; |
d8d5f3e1 | 1517 | struct btrfs_disk_key first; |
234b63a0 | 1518 | struct btrfs_extent_item extent_item; |
037e6390 | 1519 | int ret; |
d8d5f3e1 | 1520 | int level; |
1a5bc167 | 1521 | int err = 0; |
037e6390 | 1522 | |
7d9eb12c | 1523 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); |
5f39d397 | 1524 | btrfs_set_stack_extent_refs(&extent_item, 1); |
62e2749e | 1525 | btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY); |
7bb86316 | 1526 | path = btrfs_alloc_path(); |
037e6390 | 1527 | |
26b8003f | 1528 | while(1) { |
1a5bc167 CM |
1529 | ret = find_first_extent_bit(&info->extent_ins, 0, &start, |
1530 | &end, EXTENT_LOCKED); | |
1531 | if (ret) | |
26b8003f CM |
1532 | break; |
1533 | ||
1a5bc167 CM |
1534 | ins.objectid = start; |
1535 | ins.offset = end + 1 - start; | |
1536 | err = btrfs_insert_item(trans, extent_root, &ins, | |
1537 | &extent_item, sizeof(extent_item)); | |
1538 | clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED, | |
1539 | GFP_NOFS); | |
c286ac48 CM |
1540 | |
1541 | eb = btrfs_find_tree_block(extent_root, ins.objectid, | |
1542 | ins.offset); | |
1543 | ||
1544 | if (!btrfs_buffer_uptodate(eb, trans->transid)) { | |
1545 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
1546 | btrfs_read_buffer(eb, trans->transid); | |
1547 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
1548 | } | |
1549 | ||
925baedd | 1550 | btrfs_tree_lock(eb); |
d8d5f3e1 CM |
1551 | level = btrfs_header_level(eb); |
1552 | if (level == 0) { | |
1553 | btrfs_item_key(eb, &first, 0); | |
1554 | } else { | |
1555 | btrfs_node_key(eb, &first, 0); | |
1556 | } | |
925baedd CM |
1557 | btrfs_tree_unlock(eb); |
1558 | free_extent_buffer(eb); | |
1559 | /* | |
1560 | * the first key is just a hint, so the race we've created | |
1561 | * against reading it is fine | |
1562 | */ | |
7bb86316 CM |
1563 | err = btrfs_insert_extent_backref(trans, extent_root, path, |
1564 | start, extent_root->root_key.objectid, | |
f6dbff55 CM |
1565 | 0, level, |
1566 | btrfs_disk_key_objectid(&first)); | |
7bb86316 | 1567 | BUG_ON(err); |
c286ac48 CM |
1568 | if (need_resched()) { |
1569 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
1570 | cond_resched(); | |
1571 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
1572 | } | |
037e6390 | 1573 | } |
7bb86316 | 1574 | btrfs_free_path(path); |
037e6390 CM |
1575 | return 0; |
1576 | } | |
1577 | ||
db94535d CM |
1578 | static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes, |
1579 | int pending) | |
e20d96d6 | 1580 | { |
1a5bc167 | 1581 | int err = 0; |
8ef97622 | 1582 | |
7d9eb12c | 1583 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
f4b9aa8d | 1584 | if (!pending) { |
925baedd | 1585 | struct extent_buffer *buf; |
db94535d | 1586 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); |
5f39d397 | 1587 | if (buf) { |
974e35a8 Y |
1588 | if (btrfs_buffer_uptodate(buf, 0) && |
1589 | btrfs_try_tree_lock(buf)) { | |
2c90e5d6 CM |
1590 | u64 transid = |
1591 | root->fs_info->running_transaction->transid; | |
dc17ff8f CM |
1592 | u64 header_transid = |
1593 | btrfs_header_generation(buf); | |
6bc34676 CM |
1594 | if (header_transid == transid && |
1595 | !btrfs_header_flag(buf, | |
1596 | BTRFS_HEADER_FLAG_WRITTEN)) { | |
55c69072 | 1597 | clean_tree_block(NULL, root, buf); |
925baedd | 1598 | btrfs_tree_unlock(buf); |
5f39d397 | 1599 | free_extent_buffer(buf); |
c549228f | 1600 | return 1; |
2c90e5d6 | 1601 | } |
925baedd | 1602 | btrfs_tree_unlock(buf); |
f4b9aa8d | 1603 | } |
5f39d397 | 1604 | free_extent_buffer(buf); |
8ef97622 | 1605 | } |
324ae4df | 1606 | update_pinned_extents(root, bytenr, num_bytes, 1); |
f4b9aa8d | 1607 | } else { |
1a5bc167 | 1608 | set_extent_bits(&root->fs_info->pending_del, |
db94535d CM |
1609 | bytenr, bytenr + num_bytes - 1, |
1610 | EXTENT_LOCKED, GFP_NOFS); | |
f4b9aa8d | 1611 | } |
be744175 | 1612 | BUG_ON(err < 0); |
e20d96d6 CM |
1613 | return 0; |
1614 | } | |
1615 | ||
fec577fb | 1616 | /* |
a28ec197 | 1617 | * remove an extent from the root, returns 0 on success |
fec577fb | 1618 | */ |
e089f05c | 1619 | static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
7bb86316 CM |
1620 | *root, u64 bytenr, u64 num_bytes, |
1621 | u64 root_objectid, u64 ref_generation, | |
1622 | u64 owner_objectid, u64 owner_offset, int pin, | |
e37c9e69 | 1623 | int mark_free) |
a28ec197 | 1624 | { |
5caf2a00 | 1625 | struct btrfs_path *path; |
e2fa7227 | 1626 | struct btrfs_key key; |
1261ec42 CM |
1627 | struct btrfs_fs_info *info = root->fs_info; |
1628 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 1629 | struct extent_buffer *leaf; |
a28ec197 | 1630 | int ret; |
952fccac CM |
1631 | int extent_slot = 0; |
1632 | int found_extent = 0; | |
1633 | int num_to_del = 1; | |
234b63a0 | 1634 | struct btrfs_extent_item *ei; |
cf27e1ee | 1635 | u32 refs; |
037e6390 | 1636 | |
7d9eb12c | 1637 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
db94535d | 1638 | key.objectid = bytenr; |
62e2749e | 1639 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 1640 | key.offset = num_bytes; |
5caf2a00 | 1641 | path = btrfs_alloc_path(); |
54aa1f4d CM |
1642 | if (!path) |
1643 | return -ENOMEM; | |
5f26f772 | 1644 | |
3c12ac72 | 1645 | path->reada = 1; |
7bb86316 CM |
1646 | ret = lookup_extent_backref(trans, extent_root, path, |
1647 | bytenr, root_objectid, | |
1648 | ref_generation, | |
1649 | owner_objectid, owner_offset, 1); | |
1650 | if (ret == 0) { | |
952fccac CM |
1651 | struct btrfs_key found_key; |
1652 | extent_slot = path->slots[0]; | |
1653 | while(extent_slot > 0) { | |
1654 | extent_slot--; | |
1655 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1656 | extent_slot); | |
1657 | if (found_key.objectid != bytenr) | |
1658 | break; | |
1659 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && | |
1660 | found_key.offset == num_bytes) { | |
1661 | found_extent = 1; | |
1662 | break; | |
1663 | } | |
1664 | if (path->slots[0] - extent_slot > 5) | |
1665 | break; | |
1666 | } | |
1667 | if (!found_extent) | |
1668 | ret = btrfs_del_item(trans, extent_root, path); | |
7bb86316 CM |
1669 | } else { |
1670 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
1671 | WARN_ON(1); | |
1672 | printk("Unable to find ref byte nr %Lu root %Lu " | |
1673 | " gen %Lu owner %Lu offset %Lu\n", bytenr, | |
1674 | root_objectid, ref_generation, owner_objectid, | |
1675 | owner_offset); | |
1676 | } | |
952fccac CM |
1677 | if (!found_extent) { |
1678 | btrfs_release_path(extent_root, path); | |
1679 | ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); | |
1680 | if (ret < 0) | |
1681 | return ret; | |
1682 | BUG_ON(ret); | |
1683 | extent_slot = path->slots[0]; | |
1684 | } | |
5f39d397 CM |
1685 | |
1686 | leaf = path->nodes[0]; | |
952fccac | 1687 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 1688 | struct btrfs_extent_item); |
5f39d397 CM |
1689 | refs = btrfs_extent_refs(leaf, ei); |
1690 | BUG_ON(refs == 0); | |
1691 | refs -= 1; | |
1692 | btrfs_set_extent_refs(leaf, ei, refs); | |
952fccac | 1693 | |
5f39d397 CM |
1694 | btrfs_mark_buffer_dirty(leaf); |
1695 | ||
952fccac CM |
1696 | if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { |
1697 | /* if the back ref and the extent are next to each other | |
1698 | * they get deleted below in one shot | |
1699 | */ | |
1700 | path->slots[0] = extent_slot; | |
1701 | num_to_del = 2; | |
1702 | } else if (found_extent) { | |
1703 | /* otherwise delete the extent back ref */ | |
1704 | ret = btrfs_del_item(trans, extent_root, path); | |
1705 | BUG_ON(ret); | |
1706 | /* if refs are 0, we need to setup the path for deletion */ | |
1707 | if (refs == 0) { | |
1708 | btrfs_release_path(extent_root, path); | |
1709 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
1710 | -1, 1); | |
1711 | if (ret < 0) | |
1712 | return ret; | |
1713 | BUG_ON(ret); | |
1714 | } | |
1715 | } | |
1716 | ||
cf27e1ee | 1717 | if (refs == 0) { |
db94535d CM |
1718 | u64 super_used; |
1719 | u64 root_used; | |
21af804c DW |
1720 | #ifdef BIO_RW_DISCARD |
1721 | u64 map_length = num_bytes; | |
1722 | struct btrfs_multi_bio *multi = NULL; | |
1723 | #endif | |
78fae27e CM |
1724 | |
1725 | if (pin) { | |
db94535d | 1726 | ret = pin_down_bytes(root, bytenr, num_bytes, 0); |
c549228f Y |
1727 | if (ret > 0) |
1728 | mark_free = 1; | |
1729 | BUG_ON(ret < 0); | |
78fae27e CM |
1730 | } |
1731 | ||
58176a96 | 1732 | /* block accounting for super block */ |
a2135011 | 1733 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
1734 | super_used = btrfs_super_bytes_used(&info->super_copy); |
1735 | btrfs_set_super_bytes_used(&info->super_copy, | |
1736 | super_used - num_bytes); | |
a2135011 | 1737 | spin_unlock_irq(&info->delalloc_lock); |
58176a96 JB |
1738 | |
1739 | /* block accounting for root item */ | |
db94535d | 1740 | root_used = btrfs_root_used(&root->root_item); |
5f39d397 | 1741 | btrfs_set_root_used(&root->root_item, |
db94535d | 1742 | root_used - num_bytes); |
952fccac CM |
1743 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
1744 | num_to_del); | |
54aa1f4d CM |
1745 | if (ret) { |
1746 | return ret; | |
1747 | } | |
db94535d | 1748 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
0b86a832 | 1749 | mark_free); |
9078a3e1 | 1750 | BUG_ON(ret); |
21af804c DW |
1751 | |
1752 | #ifdef BIO_RW_DISCARD | |
1753 | /* Tell the block device(s) that the sectors can be discarded */ | |
1754 | ret = btrfs_map_block(&root->fs_info->mapping_tree, READ, | |
1755 | bytenr, &map_length, &multi, 0); | |
1756 | if (!ret) { | |
1757 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
1758 | int i; | |
1759 | ||
1760 | if (map_length > num_bytes) | |
1761 | map_length = num_bytes; | |
1762 | ||
1763 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
1764 | blkdev_issue_discard(stripe->dev->bdev, | |
1765 | stripe->physical >> 9, | |
1766 | map_length >> 9); | |
1767 | } | |
1768 | kfree(multi); | |
1769 | } | |
1770 | #endif | |
a28ec197 | 1771 | } |
5caf2a00 | 1772 | btrfs_free_path(path); |
e089f05c | 1773 | finish_current_insert(trans, extent_root); |
a28ec197 CM |
1774 | return ret; |
1775 | } | |
1776 | ||
a28ec197 CM |
1777 | /* |
1778 | * find all the blocks marked as pending in the radix tree and remove | |
1779 | * them from the extent map | |
1780 | */ | |
e089f05c CM |
1781 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
1782 | btrfs_root *extent_root) | |
a28ec197 CM |
1783 | { |
1784 | int ret; | |
e20d96d6 | 1785 | int err = 0; |
1a5bc167 CM |
1786 | u64 start; |
1787 | u64 end; | |
d1310b2e CM |
1788 | struct extent_io_tree *pending_del; |
1789 | struct extent_io_tree *pinned_extents; | |
8ef97622 | 1790 | |
7d9eb12c | 1791 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); |
1a5bc167 CM |
1792 | pending_del = &extent_root->fs_info->pending_del; |
1793 | pinned_extents = &extent_root->fs_info->pinned_extents; | |
a28ec197 CM |
1794 | |
1795 | while(1) { | |
1a5bc167 CM |
1796 | ret = find_first_extent_bit(pending_del, 0, &start, &end, |
1797 | EXTENT_LOCKED); | |
1798 | if (ret) | |
a28ec197 | 1799 | break; |
1a5bc167 CM |
1800 | clear_extent_bits(pending_del, start, end, EXTENT_LOCKED, |
1801 | GFP_NOFS); | |
c286ac48 CM |
1802 | if (!test_range_bit(&extent_root->fs_info->extent_ins, |
1803 | start, end, EXTENT_LOCKED, 0)) { | |
1804 | update_pinned_extents(extent_root, start, | |
1805 | end + 1 - start, 1); | |
1806 | ret = __free_extent(trans, extent_root, | |
1807 | start, end + 1 - start, | |
1808 | extent_root->root_key.objectid, | |
1809 | 0, 0, 0, 0, 0); | |
1810 | } else { | |
1811 | clear_extent_bits(&extent_root->fs_info->extent_ins, | |
1812 | start, end, EXTENT_LOCKED, GFP_NOFS); | |
1813 | } | |
1a5bc167 CM |
1814 | if (ret) |
1815 | err = ret; | |
c286ac48 CM |
1816 | |
1817 | if (need_resched()) { | |
1818 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
1819 | cond_resched(); | |
1820 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
1821 | } | |
fec577fb | 1822 | } |
e20d96d6 | 1823 | return err; |
fec577fb CM |
1824 | } |
1825 | ||
1826 | /* | |
1827 | * remove an extent from the root, returns 0 on success | |
1828 | */ | |
925baedd CM |
1829 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
1830 | struct btrfs_root *root, u64 bytenr, | |
1831 | u64 num_bytes, u64 root_objectid, | |
1832 | u64 ref_generation, u64 owner_objectid, | |
1833 | u64 owner_offset, int pin) | |
fec577fb | 1834 | { |
9f5fae2f | 1835 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
fec577fb CM |
1836 | int pending_ret; |
1837 | int ret; | |
a28ec197 | 1838 | |
db94535d | 1839 | WARN_ON(num_bytes < root->sectorsize); |
7bb86316 CM |
1840 | if (!root->ref_cows) |
1841 | ref_generation = 0; | |
1842 | ||
fec577fb | 1843 | if (root == extent_root) { |
db94535d | 1844 | pin_down_bytes(root, bytenr, num_bytes, 1); |
fec577fb CM |
1845 | return 0; |
1846 | } | |
7bb86316 CM |
1847 | ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid, |
1848 | ref_generation, owner_objectid, owner_offset, | |
1849 | pin, pin == 0); | |
ee6e6504 CM |
1850 | |
1851 | finish_current_insert(trans, root->fs_info->extent_root); | |
e20d96d6 | 1852 | pending_ret = del_pending_extents(trans, root->fs_info->extent_root); |
fec577fb CM |
1853 | return ret ? ret : pending_ret; |
1854 | } | |
1855 | ||
925baedd CM |
1856 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
1857 | struct btrfs_root *root, u64 bytenr, | |
1858 | u64 num_bytes, u64 root_objectid, | |
1859 | u64 ref_generation, u64 owner_objectid, | |
1860 | u64 owner_offset, int pin) | |
1861 | { | |
1862 | int ret; | |
1863 | ||
1864 | maybe_lock_mutex(root); | |
1865 | ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1866 | root_objectid, ref_generation, | |
1867 | owner_objectid, owner_offset, pin); | |
1868 | maybe_unlock_mutex(root); | |
1869 | return ret; | |
1870 | } | |
1871 | ||
87ee04eb CM |
1872 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
1873 | { | |
1874 | u64 mask = ((u64)root->stripesize - 1); | |
1875 | u64 ret = (val + mask) & ~mask; | |
1876 | return ret; | |
1877 | } | |
1878 | ||
fec577fb CM |
1879 | /* |
1880 | * walks the btree of allocated extents and find a hole of a given size. | |
1881 | * The key ins is changed to record the hole: | |
1882 | * ins->objectid == block start | |
62e2749e | 1883 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
1884 | * ins->offset == number of blocks |
1885 | * Any available blocks before search_start are skipped. | |
1886 | */ | |
98ed5174 CM |
1887 | static int noinline find_free_extent(struct btrfs_trans_handle *trans, |
1888 | struct btrfs_root *orig_root, | |
1889 | u64 num_bytes, u64 empty_size, | |
1890 | u64 search_start, u64 search_end, | |
1891 | u64 hint_byte, struct btrfs_key *ins, | |
1892 | u64 exclude_start, u64 exclude_nr, | |
1893 | int data) | |
fec577fb | 1894 | { |
87ee04eb | 1895 | int ret; |
a061fc8d | 1896 | u64 orig_search_start; |
9f5fae2f | 1897 | struct btrfs_root * root = orig_root->fs_info->extent_root; |
f2458e1d | 1898 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 1899 | u64 total_needed = num_bytes; |
239b14b3 | 1900 | u64 *last_ptr = NULL; |
be08c1b9 | 1901 | struct btrfs_block_group_cache *block_group; |
be744175 | 1902 | int full_scan = 0; |
fbdc762b | 1903 | int wrapped = 0; |
0ef3e66b | 1904 | int chunk_alloc_done = 0; |
239b14b3 | 1905 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 1906 | int allowed_chunk_alloc = 0; |
fec577fb | 1907 | |
db94535d | 1908 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 CM |
1909 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
1910 | ||
0ef3e66b CM |
1911 | if (orig_root->ref_cows || empty_size) |
1912 | allowed_chunk_alloc = 1; | |
1913 | ||
239b14b3 CM |
1914 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
1915 | last_ptr = &root->fs_info->last_alloc; | |
8790d502 | 1916 | empty_cluster = 256 * 1024; |
239b14b3 CM |
1917 | } |
1918 | ||
1919 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { | |
1920 | last_ptr = &root->fs_info->last_data_alloc; | |
1921 | } | |
1922 | ||
1923 | if (last_ptr) { | |
1924 | if (*last_ptr) | |
1925 | hint_byte = *last_ptr; | |
1926 | else { | |
1927 | empty_size += empty_cluster; | |
1928 | } | |
1929 | } | |
1930 | ||
a061fc8d CM |
1931 | search_start = max(search_start, first_logical_byte(root, 0)); |
1932 | orig_search_start = search_start; | |
1933 | ||
7f93bf8d CM |
1934 | if (search_end == (u64)-1) |
1935 | search_end = btrfs_super_total_bytes(&info->super_copy); | |
0b86a832 | 1936 | |
db94535d | 1937 | if (hint_byte) { |
0ef3e66b | 1938 | block_group = btrfs_lookup_first_block_group(info, hint_byte); |
1a2b2ac7 CM |
1939 | if (!block_group) |
1940 | hint_byte = search_start; | |
c286ac48 | 1941 | block_group = btrfs_find_block_group(root, block_group, |
db94535d | 1942 | hint_byte, data, 1); |
239b14b3 CM |
1943 | if (last_ptr && *last_ptr == 0 && block_group) |
1944 | hint_byte = block_group->key.objectid; | |
be744175 | 1945 | } else { |
c286ac48 | 1946 | block_group = btrfs_find_block_group(root, |
1a2b2ac7 CM |
1947 | trans->block_group, |
1948 | search_start, data, 1); | |
be744175 | 1949 | } |
239b14b3 | 1950 | search_start = max(search_start, hint_byte); |
be744175 | 1951 | |
6702ed49 | 1952 | total_needed += empty_size; |
0b86a832 | 1953 | |
be744175 | 1954 | check_failed: |
70b043f0 | 1955 | if (!block_group) { |
0ef3e66b CM |
1956 | block_group = btrfs_lookup_first_block_group(info, |
1957 | search_start); | |
70b043f0 | 1958 | if (!block_group) |
0ef3e66b | 1959 | block_group = btrfs_lookup_first_block_group(info, |
70b043f0 CM |
1960 | orig_search_start); |
1961 | } | |
0ef3e66b CM |
1962 | if (full_scan && !chunk_alloc_done) { |
1963 | if (allowed_chunk_alloc) { | |
1964 | do_chunk_alloc(trans, root, | |
1965 | num_bytes + 2 * 1024 * 1024, data, 1); | |
1966 | allowed_chunk_alloc = 0; | |
1967 | } else if (block_group && block_group_bits(block_group, data)) { | |
1968 | block_group->space_info->force_alloc = 1; | |
1969 | } | |
1970 | chunk_alloc_done = 1; | |
1971 | } | |
0b86a832 CM |
1972 | ret = find_search_start(root, &block_group, &search_start, |
1973 | total_needed, data); | |
239b14b3 CM |
1974 | if (ret == -ENOSPC && last_ptr && *last_ptr) { |
1975 | *last_ptr = 0; | |
0ef3e66b CM |
1976 | block_group = btrfs_lookup_first_block_group(info, |
1977 | orig_search_start); | |
239b14b3 CM |
1978 | search_start = orig_search_start; |
1979 | ret = find_search_start(root, &block_group, &search_start, | |
1980 | total_needed, data); | |
1981 | } | |
1982 | if (ret == -ENOSPC) | |
1983 | goto enospc; | |
0b86a832 | 1984 | if (ret) |
d548ee51 | 1985 | goto error; |
e19caa5f | 1986 | |
239b14b3 CM |
1987 | if (last_ptr && *last_ptr && search_start != *last_ptr) { |
1988 | *last_ptr = 0; | |
1989 | if (!empty_size) { | |
1990 | empty_size += empty_cluster; | |
1991 | total_needed += empty_size; | |
1992 | } | |
0ef3e66b | 1993 | block_group = btrfs_lookup_first_block_group(info, |
239b14b3 CM |
1994 | orig_search_start); |
1995 | search_start = orig_search_start; | |
1996 | ret = find_search_start(root, &block_group, | |
1997 | &search_start, total_needed, data); | |
1998 | if (ret == -ENOSPC) | |
1999 | goto enospc; | |
2000 | if (ret) | |
2001 | goto error; | |
2002 | } | |
2003 | ||
0b86a832 CM |
2004 | search_start = stripe_align(root, search_start); |
2005 | ins->objectid = search_start; | |
2006 | ins->offset = num_bytes; | |
e37c9e69 | 2007 | |
db94535d | 2008 | if (ins->objectid + num_bytes >= search_end) |
cf67582b | 2009 | goto enospc; |
0b86a832 CM |
2010 | |
2011 | if (ins->objectid + num_bytes > | |
2012 | block_group->key.objectid + block_group->key.offset) { | |
e19caa5f CM |
2013 | search_start = block_group->key.objectid + |
2014 | block_group->key.offset; | |
2015 | goto new_group; | |
2016 | } | |
0b86a832 | 2017 | |
1a5bc167 | 2018 | if (test_range_bit(&info->extent_ins, ins->objectid, |
db94535d CM |
2019 | ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) { |
2020 | search_start = ins->objectid + num_bytes; | |
1a5bc167 CM |
2021 | goto new_group; |
2022 | } | |
0b86a832 | 2023 | |
1a5bc167 | 2024 | if (test_range_bit(&info->pinned_extents, ins->objectid, |
db94535d CM |
2025 | ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) { |
2026 | search_start = ins->objectid + num_bytes; | |
1a5bc167 | 2027 | goto new_group; |
fec577fb | 2028 | } |
0b86a832 | 2029 | |
db94535d | 2030 | if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start && |
f2654de4 CM |
2031 | ins->objectid < exclude_start + exclude_nr)) { |
2032 | search_start = exclude_start + exclude_nr; | |
2033 | goto new_group; | |
2034 | } | |
0b86a832 | 2035 | |
6324fbf3 | 2036 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) { |
5276aeda | 2037 | block_group = btrfs_lookup_block_group(info, ins->objectid); |
26b8003f CM |
2038 | if (block_group) |
2039 | trans->block_group = block_group; | |
f2458e1d | 2040 | } |
db94535d | 2041 | ins->offset = num_bytes; |
239b14b3 CM |
2042 | if (last_ptr) { |
2043 | *last_ptr = ins->objectid + ins->offset; | |
2044 | if (*last_ptr == | |
2045 | btrfs_super_total_bytes(&root->fs_info->super_copy)) { | |
2046 | *last_ptr = 0; | |
2047 | } | |
2048 | } | |
fec577fb | 2049 | return 0; |
be744175 CM |
2050 | |
2051 | new_group: | |
db94535d | 2052 | if (search_start + num_bytes >= search_end) { |
cf67582b | 2053 | enospc: |
be744175 | 2054 | search_start = orig_search_start; |
fbdc762b CM |
2055 | if (full_scan) { |
2056 | ret = -ENOSPC; | |
2057 | goto error; | |
2058 | } | |
6702ed49 CM |
2059 | if (wrapped) { |
2060 | if (!full_scan) | |
2061 | total_needed -= empty_size; | |
fbdc762b | 2062 | full_scan = 1; |
6702ed49 | 2063 | } else |
fbdc762b | 2064 | wrapped = 1; |
be744175 | 2065 | } |
0ef3e66b | 2066 | block_group = btrfs_lookup_first_block_group(info, search_start); |
fbdc762b | 2067 | cond_resched(); |
c286ac48 | 2068 | block_group = btrfs_find_block_group(root, block_group, |
1a2b2ac7 | 2069 | search_start, data, 0); |
be744175 CM |
2070 | goto check_failed; |
2071 | ||
0f70abe2 | 2072 | error: |
0f70abe2 | 2073 | return ret; |
fec577fb | 2074 | } |
ec44a35c | 2075 | |
e6dcd2dc CM |
2076 | static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
2077 | struct btrfs_root *root, | |
2078 | u64 num_bytes, u64 min_alloc_size, | |
2079 | u64 empty_size, u64 hint_byte, | |
2080 | u64 search_end, struct btrfs_key *ins, | |
2081 | u64 data) | |
fec577fb CM |
2082 | { |
2083 | int ret; | |
fbdc762b | 2084 | u64 search_start = 0; |
8790d502 | 2085 | u64 alloc_profile; |
1261ec42 | 2086 | struct btrfs_fs_info *info = root->fs_info; |
925baedd | 2087 | |
6324fbf3 | 2088 | if (data) { |
8790d502 CM |
2089 | alloc_profile = info->avail_data_alloc_bits & |
2090 | info->data_alloc_profile; | |
2091 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
6324fbf3 | 2092 | } else if (root == root->fs_info->chunk_root) { |
8790d502 CM |
2093 | alloc_profile = info->avail_system_alloc_bits & |
2094 | info->system_alloc_profile; | |
2095 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
6324fbf3 | 2096 | } else { |
8790d502 CM |
2097 | alloc_profile = info->avail_metadata_alloc_bits & |
2098 | info->metadata_alloc_profile; | |
2099 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
6324fbf3 | 2100 | } |
98d20f67 | 2101 | again: |
a061fc8d | 2102 | data = reduce_alloc_profile(root, data); |
0ef3e66b CM |
2103 | /* |
2104 | * the only place that sets empty_size is btrfs_realloc_node, which | |
2105 | * is not called recursively on allocations | |
2106 | */ | |
2107 | if (empty_size || root->ref_cows) { | |
593060d7 | 2108 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 2109 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b CM |
2110 | 2 * 1024 * 1024, |
2111 | BTRFS_BLOCK_GROUP_METADATA | | |
2112 | (info->metadata_alloc_profile & | |
2113 | info->avail_metadata_alloc_bits), 0); | |
6324fbf3 CM |
2114 | BUG_ON(ret); |
2115 | } | |
2116 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
0ef3e66b | 2117 | num_bytes + 2 * 1024 * 1024, data, 0); |
6324fbf3 CM |
2118 | BUG_ON(ret); |
2119 | } | |
0b86a832 | 2120 | |
db94535d CM |
2121 | WARN_ON(num_bytes < root->sectorsize); |
2122 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
2123 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
2124 | trans->alloc_exclude_start, |
2125 | trans->alloc_exclude_nr, data); | |
3b951516 | 2126 | |
98d20f67 CM |
2127 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
2128 | num_bytes = num_bytes >> 1; | |
2129 | num_bytes = max(num_bytes, min_alloc_size); | |
0ef3e66b CM |
2130 | do_chunk_alloc(trans, root->fs_info->extent_root, |
2131 | num_bytes, data, 1); | |
98d20f67 CM |
2132 | goto again; |
2133 | } | |
ec44a35c CM |
2134 | if (ret) { |
2135 | printk("allocation failed flags %Lu\n", data); | |
925baedd | 2136 | BUG(); |
925baedd | 2137 | } |
e6dcd2dc CM |
2138 | clear_extent_dirty(&root->fs_info->free_space_cache, |
2139 | ins->objectid, ins->objectid + ins->offset - 1, | |
2140 | GFP_NOFS); | |
2141 | return 0; | |
2142 | } | |
2143 | ||
65b51a00 CM |
2144 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
2145 | { | |
2146 | maybe_lock_mutex(root); | |
2147 | set_extent_dirty(&root->fs_info->free_space_cache, | |
2148 | start, start + len - 1, GFP_NOFS); | |
2149 | maybe_unlock_mutex(root); | |
2150 | return 0; | |
2151 | } | |
2152 | ||
e6dcd2dc CM |
2153 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
2154 | struct btrfs_root *root, | |
2155 | u64 num_bytes, u64 min_alloc_size, | |
2156 | u64 empty_size, u64 hint_byte, | |
2157 | u64 search_end, struct btrfs_key *ins, | |
2158 | u64 data) | |
2159 | { | |
2160 | int ret; | |
2161 | maybe_lock_mutex(root); | |
2162 | ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, | |
2163 | empty_size, hint_byte, search_end, ins, | |
2164 | data); | |
2165 | maybe_unlock_mutex(root); | |
2166 | return ret; | |
2167 | } | |
2168 | ||
2169 | static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
2170 | struct btrfs_root *root, | |
2171 | u64 root_objectid, u64 ref_generation, | |
2172 | u64 owner, u64 owner_offset, | |
2173 | struct btrfs_key *ins) | |
2174 | { | |
2175 | int ret; | |
2176 | int pending_ret; | |
2177 | u64 super_used; | |
2178 | u64 root_used; | |
2179 | u64 num_bytes = ins->offset; | |
2180 | u32 sizes[2]; | |
2181 | struct btrfs_fs_info *info = root->fs_info; | |
2182 | struct btrfs_root *extent_root = info->extent_root; | |
2183 | struct btrfs_extent_item *extent_item; | |
2184 | struct btrfs_extent_ref *ref; | |
2185 | struct btrfs_path *path; | |
2186 | struct btrfs_key keys[2]; | |
fec577fb | 2187 | |
58176a96 | 2188 | /* block accounting for super block */ |
a2135011 | 2189 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
2190 | super_used = btrfs_super_bytes_used(&info->super_copy); |
2191 | btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); | |
a2135011 | 2192 | spin_unlock_irq(&info->delalloc_lock); |
26b8003f | 2193 | |
58176a96 | 2194 | /* block accounting for root item */ |
db94535d CM |
2195 | root_used = btrfs_root_used(&root->root_item); |
2196 | btrfs_set_root_used(&root->root_item, root_used + num_bytes); | |
58176a96 | 2197 | |
26b8003f | 2198 | if (root == extent_root) { |
1a5bc167 CM |
2199 | set_extent_bits(&root->fs_info->extent_ins, ins->objectid, |
2200 | ins->objectid + ins->offset - 1, | |
2201 | EXTENT_LOCKED, GFP_NOFS); | |
26b8003f CM |
2202 | goto update_block; |
2203 | } | |
2204 | ||
47e4bb98 CM |
2205 | memcpy(&keys[0], ins, sizeof(*ins)); |
2206 | keys[1].offset = hash_extent_ref(root_objectid, ref_generation, | |
2207 | owner, owner_offset); | |
2208 | keys[1].objectid = ins->objectid; | |
2209 | keys[1].type = BTRFS_EXTENT_REF_KEY; | |
2210 | sizes[0] = sizeof(*extent_item); | |
2211 | sizes[1] = sizeof(*ref); | |
7bb86316 CM |
2212 | |
2213 | path = btrfs_alloc_path(); | |
2214 | BUG_ON(!path); | |
47e4bb98 CM |
2215 | |
2216 | ret = btrfs_insert_empty_items(trans, extent_root, path, keys, | |
2217 | sizes, 2); | |
26b8003f | 2218 | |
ccd467d6 | 2219 | BUG_ON(ret); |
47e4bb98 CM |
2220 | extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2221 | struct btrfs_extent_item); | |
2222 | btrfs_set_extent_refs(path->nodes[0], extent_item, 1); | |
2223 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
2224 | struct btrfs_extent_ref); | |
2225 | ||
2226 | btrfs_set_ref_root(path->nodes[0], ref, root_objectid); | |
2227 | btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); | |
2228 | btrfs_set_ref_objectid(path->nodes[0], ref, owner); | |
2229 | btrfs_set_ref_offset(path->nodes[0], ref, owner_offset); | |
2230 | ||
2231 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
2232 | ||
2233 | trans->alloc_exclude_start = 0; | |
2234 | trans->alloc_exclude_nr = 0; | |
7bb86316 | 2235 | btrfs_free_path(path); |
e089f05c | 2236 | finish_current_insert(trans, extent_root); |
e20d96d6 | 2237 | pending_ret = del_pending_extents(trans, extent_root); |
f510cfec | 2238 | |
925baedd CM |
2239 | if (ret) |
2240 | goto out; | |
e37c9e69 | 2241 | if (pending_ret) { |
925baedd CM |
2242 | ret = pending_ret; |
2243 | goto out; | |
e37c9e69 | 2244 | } |
26b8003f CM |
2245 | |
2246 | update_block: | |
0b86a832 | 2247 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0); |
f5947066 CM |
2248 | if (ret) { |
2249 | printk("update block group failed for %Lu %Lu\n", | |
2250 | ins->objectid, ins->offset); | |
2251 | BUG(); | |
2252 | } | |
925baedd | 2253 | out: |
e6dcd2dc CM |
2254 | return ret; |
2255 | } | |
2256 | ||
2257 | int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
2258 | struct btrfs_root *root, | |
2259 | u64 root_objectid, u64 ref_generation, | |
2260 | u64 owner, u64 owner_offset, | |
2261 | struct btrfs_key *ins) | |
2262 | { | |
2263 | int ret; | |
2264 | maybe_lock_mutex(root); | |
2265 | ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid, | |
2266 | ref_generation, owner, | |
2267 | owner_offset, ins); | |
2268 | maybe_unlock_mutex(root); | |
2269 | return ret; | |
2270 | } | |
2271 | /* | |
2272 | * finds a free extent and does all the dirty work required for allocation | |
2273 | * returns the key for the extent through ins, and a tree buffer for | |
2274 | * the first block of the extent through buf. | |
2275 | * | |
2276 | * returns 0 if everything worked, non-zero otherwise. | |
2277 | */ | |
2278 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, | |
2279 | struct btrfs_root *root, | |
2280 | u64 num_bytes, u64 min_alloc_size, | |
2281 | u64 root_objectid, u64 ref_generation, | |
2282 | u64 owner, u64 owner_offset, | |
2283 | u64 empty_size, u64 hint_byte, | |
2284 | u64 search_end, struct btrfs_key *ins, u64 data) | |
2285 | { | |
2286 | int ret; | |
2287 | ||
2288 | maybe_lock_mutex(root); | |
2289 | ||
2290 | ret = __btrfs_reserve_extent(trans, root, num_bytes, | |
2291 | min_alloc_size, empty_size, hint_byte, | |
2292 | search_end, ins, data); | |
2293 | BUG_ON(ret); | |
2294 | ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid, | |
2295 | ref_generation, owner, | |
2296 | owner_offset, ins); | |
2297 | BUG_ON(ret); | |
2298 | ||
925baedd CM |
2299 | maybe_unlock_mutex(root); |
2300 | return ret; | |
fec577fb | 2301 | } |
65b51a00 CM |
2302 | |
2303 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, | |
2304 | struct btrfs_root *root, | |
2305 | u64 bytenr, u32 blocksize) | |
2306 | { | |
2307 | struct extent_buffer *buf; | |
2308 | ||
2309 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
2310 | if (!buf) | |
2311 | return ERR_PTR(-ENOMEM); | |
2312 | btrfs_set_header_generation(buf, trans->transid); | |
2313 | btrfs_tree_lock(buf); | |
2314 | clean_tree_block(trans, root, buf); | |
2315 | btrfs_set_buffer_uptodate(buf); | |
2316 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
2317 | buf->start + buf->len - 1, GFP_NOFS); | |
2318 | trans->blocks_used++; | |
2319 | return buf; | |
2320 | } | |
2321 | ||
fec577fb CM |
2322 | /* |
2323 | * helper function to allocate a block for a given tree | |
2324 | * returns the tree buffer or NULL. | |
2325 | */ | |
5f39d397 | 2326 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
7bb86316 CM |
2327 | struct btrfs_root *root, |
2328 | u32 blocksize, | |
2329 | u64 root_objectid, | |
2330 | u64 ref_generation, | |
2331 | u64 first_objectid, | |
2332 | int level, | |
2333 | u64 hint, | |
5f39d397 | 2334 | u64 empty_size) |
fec577fb | 2335 | { |
e2fa7227 | 2336 | struct btrfs_key ins; |
fec577fb | 2337 | int ret; |
5f39d397 | 2338 | struct extent_buffer *buf; |
fec577fb | 2339 | |
98d20f67 | 2340 | ret = btrfs_alloc_extent(trans, root, blocksize, blocksize, |
7bb86316 | 2341 | root_objectid, ref_generation, |
f6dbff55 | 2342 | level, first_objectid, empty_size, hint, |
db94535d | 2343 | (u64)-1, &ins, 0); |
fec577fb | 2344 | if (ret) { |
54aa1f4d CM |
2345 | BUG_ON(ret > 0); |
2346 | return ERR_PTR(ret); | |
fec577fb | 2347 | } |
55c69072 | 2348 | |
65b51a00 | 2349 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize); |
fec577fb CM |
2350 | return buf; |
2351 | } | |
a28ec197 | 2352 | |
31153d81 | 2353 | static int noinline drop_leaf_ref_no_cache(struct btrfs_trans_handle *trans, |
bcc63abb | 2354 | struct btrfs_root *root, |
31153d81 | 2355 | struct extent_buffer *leaf) |
6407bf6d | 2356 | { |
7bb86316 CM |
2357 | u64 leaf_owner; |
2358 | u64 leaf_generation; | |
5f39d397 | 2359 | struct btrfs_key key; |
6407bf6d CM |
2360 | struct btrfs_file_extent_item *fi; |
2361 | int i; | |
2362 | int nritems; | |
2363 | int ret; | |
2364 | ||
5f39d397 CM |
2365 | BUG_ON(!btrfs_is_leaf(leaf)); |
2366 | nritems = btrfs_header_nritems(leaf); | |
7bb86316 CM |
2367 | leaf_owner = btrfs_header_owner(leaf); |
2368 | leaf_generation = btrfs_header_generation(leaf); | |
2369 | ||
6407bf6d | 2370 | for (i = 0; i < nritems; i++) { |
db94535d | 2371 | u64 disk_bytenr; |
e34a5b4f | 2372 | cond_resched(); |
5f39d397 CM |
2373 | |
2374 | btrfs_item_key_to_cpu(leaf, &key, i); | |
2375 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d CM |
2376 | continue; |
2377 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
5f39d397 CM |
2378 | if (btrfs_file_extent_type(leaf, fi) == |
2379 | BTRFS_FILE_EXTENT_INLINE) | |
236454df | 2380 | continue; |
6407bf6d CM |
2381 | /* |
2382 | * FIXME make sure to insert a trans record that | |
2383 | * repeats the snapshot del on crash | |
2384 | */ | |
db94535d CM |
2385 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
2386 | if (disk_bytenr == 0) | |
3a686375 | 2387 | continue; |
4a096752 CM |
2388 | |
2389 | mutex_lock(&root->fs_info->alloc_mutex); | |
925baedd | 2390 | ret = __btrfs_free_extent(trans, root, disk_bytenr, |
7bb86316 CM |
2391 | btrfs_file_extent_disk_num_bytes(leaf, fi), |
2392 | leaf_owner, leaf_generation, | |
2393 | key.objectid, key.offset, 0); | |
4a096752 | 2394 | mutex_unlock(&root->fs_info->alloc_mutex); |
2dd3e67b CM |
2395 | |
2396 | atomic_inc(&root->fs_info->throttle_gen); | |
2397 | wake_up(&root->fs_info->transaction_throttle); | |
2398 | cond_resched(); | |
2399 | ||
6407bf6d CM |
2400 | BUG_ON(ret); |
2401 | } | |
2402 | return 0; | |
2403 | } | |
2404 | ||
31153d81 | 2405 | static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans, |
bcc63abb | 2406 | struct btrfs_root *root, |
31153d81 YZ |
2407 | struct btrfs_leaf_ref *ref) |
2408 | { | |
2409 | int i; | |
2410 | int ret; | |
2411 | struct btrfs_extent_info *info = ref->extents; | |
2412 | ||
31153d81 YZ |
2413 | for (i = 0; i < ref->nritems; i++) { |
2414 | mutex_lock(&root->fs_info->alloc_mutex); | |
2415 | ret = __btrfs_free_extent(trans, root, | |
2416 | info->bytenr, info->num_bytes, | |
2417 | ref->owner, ref->generation, | |
2418 | info->objectid, info->offset, 0); | |
2419 | mutex_unlock(&root->fs_info->alloc_mutex); | |
2dd3e67b CM |
2420 | |
2421 | atomic_inc(&root->fs_info->throttle_gen); | |
2422 | wake_up(&root->fs_info->transaction_throttle); | |
2423 | cond_resched(); | |
2424 | ||
31153d81 YZ |
2425 | BUG_ON(ret); |
2426 | info++; | |
2427 | } | |
31153d81 YZ |
2428 | |
2429 | return 0; | |
2430 | } | |
2431 | ||
333db94c CM |
2432 | int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, |
2433 | u32 *refs) | |
2434 | { | |
017e5369 | 2435 | int ret; |
f87f057b | 2436 | |
017e5369 | 2437 | ret = lookup_extent_ref(NULL, root, start, len, refs); |
f87f057b CM |
2438 | BUG_ON(ret); |
2439 | ||
2440 | #if 0 // some debugging code in case we see problems here | |
2441 | /* if the refs count is one, it won't get increased again. But | |
2442 | * if the ref count is > 1, someone may be decreasing it at | |
2443 | * the same time we are. | |
2444 | */ | |
2445 | if (*refs != 1) { | |
2446 | struct extent_buffer *eb = NULL; | |
2447 | eb = btrfs_find_create_tree_block(root, start, len); | |
2448 | if (eb) | |
2449 | btrfs_tree_lock(eb); | |
2450 | ||
2451 | mutex_lock(&root->fs_info->alloc_mutex); | |
2452 | ret = lookup_extent_ref(NULL, root, start, len, refs); | |
2453 | BUG_ON(ret); | |
2454 | mutex_unlock(&root->fs_info->alloc_mutex); | |
2455 | ||
2456 | if (eb) { | |
2457 | btrfs_tree_unlock(eb); | |
2458 | free_extent_buffer(eb); | |
2459 | } | |
2460 | if (*refs == 1) { | |
2461 | printk("block %llu went down to one during drop_snap\n", | |
2462 | (unsigned long long)start); | |
2463 | } | |
2464 | ||
2465 | } | |
2466 | #endif | |
2467 | ||
e7a84565 | 2468 | cond_resched(); |
017e5369 | 2469 | return ret; |
333db94c CM |
2470 | } |
2471 | ||
9aca1d51 CM |
2472 | /* |
2473 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
2474 | * counts as it goes. | |
2475 | */ | |
98ed5174 CM |
2476 | static int noinline walk_down_tree(struct btrfs_trans_handle *trans, |
2477 | struct btrfs_root *root, | |
2478 | struct btrfs_path *path, int *level) | |
20524f02 | 2479 | { |
7bb86316 CM |
2480 | u64 root_owner; |
2481 | u64 root_gen; | |
2482 | u64 bytenr; | |
ca7a79ad | 2483 | u64 ptr_gen; |
5f39d397 CM |
2484 | struct extent_buffer *next; |
2485 | struct extent_buffer *cur; | |
7bb86316 | 2486 | struct extent_buffer *parent; |
31153d81 | 2487 | struct btrfs_leaf_ref *ref; |
db94535d | 2488 | u32 blocksize; |
20524f02 CM |
2489 | int ret; |
2490 | u32 refs; | |
2491 | ||
5caf2a00 CM |
2492 | WARN_ON(*level < 0); |
2493 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
333db94c | 2494 | ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start, |
db94535d | 2495 | path->nodes[*level]->len, &refs); |
20524f02 CM |
2496 | BUG_ON(ret); |
2497 | if (refs > 1) | |
2498 | goto out; | |
e011599b | 2499 | |
9aca1d51 CM |
2500 | /* |
2501 | * walk down to the last node level and free all the leaves | |
2502 | */ | |
6407bf6d | 2503 | while(*level >= 0) { |
5caf2a00 CM |
2504 | WARN_ON(*level < 0); |
2505 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
20524f02 | 2506 | cur = path->nodes[*level]; |
e011599b | 2507 | |
5f39d397 | 2508 | if (btrfs_header_level(cur) != *level) |
2c90e5d6 | 2509 | WARN_ON(1); |
e011599b | 2510 | |
7518a238 | 2511 | if (path->slots[*level] >= |
5f39d397 | 2512 | btrfs_header_nritems(cur)) |
20524f02 | 2513 | break; |
6407bf6d | 2514 | if (*level == 0) { |
31153d81 | 2515 | ret = drop_leaf_ref_no_cache(trans, root, cur); |
6407bf6d CM |
2516 | BUG_ON(ret); |
2517 | break; | |
2518 | } | |
db94535d | 2519 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
ca7a79ad | 2520 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); |
db94535d | 2521 | blocksize = btrfs_level_size(root, *level - 1); |
925baedd | 2522 | |
333db94c | 2523 | ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); |
6407bf6d CM |
2524 | BUG_ON(ret); |
2525 | if (refs != 1) { | |
7bb86316 CM |
2526 | parent = path->nodes[*level]; |
2527 | root_owner = btrfs_header_owner(parent); | |
2528 | root_gen = btrfs_header_generation(parent); | |
20524f02 | 2529 | path->slots[*level]++; |
f87f057b CM |
2530 | |
2531 | mutex_lock(&root->fs_info->alloc_mutex); | |
925baedd | 2532 | ret = __btrfs_free_extent(trans, root, bytenr, |
7bb86316 CM |
2533 | blocksize, root_owner, |
2534 | root_gen, 0, 0, 1); | |
20524f02 | 2535 | BUG_ON(ret); |
f87f057b | 2536 | mutex_unlock(&root->fs_info->alloc_mutex); |
18e35e0a CM |
2537 | |
2538 | atomic_inc(&root->fs_info->throttle_gen); | |
2539 | wake_up(&root->fs_info->transaction_throttle); | |
2dd3e67b | 2540 | cond_resched(); |
18e35e0a | 2541 | |
20524f02 CM |
2542 | continue; |
2543 | } | |
f87f057b CM |
2544 | /* |
2545 | * at this point, we have a single ref, and since the | |
2546 | * only place referencing this extent is a dead root | |
2547 | * the reference count should never go higher. | |
2548 | * So, we don't need to check it again | |
2549 | */ | |
31153d81 YZ |
2550 | if (*level == 1) { |
2551 | struct btrfs_key key; | |
2552 | btrfs_node_key_to_cpu(cur, &key, path->slots[*level]); | |
017e5369 | 2553 | ref = btrfs_lookup_leaf_ref(root, bytenr); |
31153d81 YZ |
2554 | if (ref) { |
2555 | ret = drop_leaf_ref(trans, root, ref); | |
2556 | BUG_ON(ret); | |
2557 | btrfs_remove_leaf_ref(root, ref); | |
bcc63abb | 2558 | btrfs_free_leaf_ref(root, ref); |
31153d81 YZ |
2559 | *level = 0; |
2560 | break; | |
2561 | } | |
37d1aeee CM |
2562 | if (printk_ratelimit()) |
2563 | printk("leaf ref miss for bytenr %llu\n", | |
2564 | (unsigned long long)bytenr); | |
31153d81 | 2565 | } |
db94535d | 2566 | next = btrfs_find_tree_block(root, bytenr, blocksize); |
1259ab75 | 2567 | if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) { |
5f39d397 | 2568 | free_extent_buffer(next); |
333db94c | 2569 | |
ca7a79ad CM |
2570 | next = read_tree_block(root, bytenr, blocksize, |
2571 | ptr_gen); | |
e7a84565 | 2572 | cond_resched(); |
f87f057b CM |
2573 | #if 0 |
2574 | /* | |
2575 | * this is a debugging check and can go away | |
2576 | * the ref should never go all the way down to 1 | |
2577 | * at this point | |
2578 | */ | |
e6dcd2dc CM |
2579 | ret = lookup_extent_ref(NULL, root, bytenr, blocksize, |
2580 | &refs); | |
e9d0b13b | 2581 | BUG_ON(ret); |
f87f057b CM |
2582 | WARN_ON(refs != 1); |
2583 | #endif | |
e9d0b13b | 2584 | } |
5caf2a00 | 2585 | WARN_ON(*level <= 0); |
83e15a28 | 2586 | if (path->nodes[*level-1]) |
5f39d397 | 2587 | free_extent_buffer(path->nodes[*level-1]); |
20524f02 | 2588 | path->nodes[*level-1] = next; |
5f39d397 | 2589 | *level = btrfs_header_level(next); |
20524f02 | 2590 | path->slots[*level] = 0; |
2dd3e67b | 2591 | cond_resched(); |
20524f02 CM |
2592 | } |
2593 | out: | |
5caf2a00 CM |
2594 | WARN_ON(*level < 0); |
2595 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
7bb86316 CM |
2596 | |
2597 | if (path->nodes[*level] == root->node) { | |
7bb86316 | 2598 | parent = path->nodes[*level]; |
31153d81 | 2599 | bytenr = path->nodes[*level]->start; |
7bb86316 CM |
2600 | } else { |
2601 | parent = path->nodes[*level + 1]; | |
31153d81 | 2602 | bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]); |
7bb86316 CM |
2603 | } |
2604 | ||
31153d81 YZ |
2605 | blocksize = btrfs_level_size(root, *level); |
2606 | root_owner = btrfs_header_owner(parent); | |
7bb86316 | 2607 | root_gen = btrfs_header_generation(parent); |
31153d81 | 2608 | |
f87f057b | 2609 | mutex_lock(&root->fs_info->alloc_mutex); |
31153d81 YZ |
2610 | ret = __btrfs_free_extent(trans, root, bytenr, blocksize, |
2611 | root_owner, root_gen, 0, 0, 1); | |
5f39d397 | 2612 | free_extent_buffer(path->nodes[*level]); |
20524f02 CM |
2613 | path->nodes[*level] = NULL; |
2614 | *level += 1; | |
2615 | BUG_ON(ret); | |
925baedd | 2616 | mutex_unlock(&root->fs_info->alloc_mutex); |
f87f057b | 2617 | |
e7a84565 | 2618 | cond_resched(); |
20524f02 CM |
2619 | return 0; |
2620 | } | |
2621 | ||
9aca1d51 CM |
2622 | /* |
2623 | * helper for dropping snapshots. This walks back up the tree in the path | |
2624 | * to find the first node higher up where we haven't yet gone through | |
2625 | * all the slots | |
2626 | */ | |
98ed5174 CM |
2627 | static int noinline walk_up_tree(struct btrfs_trans_handle *trans, |
2628 | struct btrfs_root *root, | |
2629 | struct btrfs_path *path, int *level) | |
20524f02 | 2630 | { |
7bb86316 CM |
2631 | u64 root_owner; |
2632 | u64 root_gen; | |
2633 | struct btrfs_root_item *root_item = &root->root_item; | |
20524f02 CM |
2634 | int i; |
2635 | int slot; | |
2636 | int ret; | |
9f3a7427 | 2637 | |
234b63a0 | 2638 | for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
20524f02 | 2639 | slot = path->slots[i]; |
5f39d397 CM |
2640 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { |
2641 | struct extent_buffer *node; | |
2642 | struct btrfs_disk_key disk_key; | |
2643 | node = path->nodes[i]; | |
20524f02 CM |
2644 | path->slots[i]++; |
2645 | *level = i; | |
9f3a7427 | 2646 | WARN_ON(*level == 0); |
5f39d397 | 2647 | btrfs_node_key(node, &disk_key, path->slots[i]); |
9f3a7427 | 2648 | memcpy(&root_item->drop_progress, |
5f39d397 | 2649 | &disk_key, sizeof(disk_key)); |
9f3a7427 | 2650 | root_item->drop_level = i; |
20524f02 CM |
2651 | return 0; |
2652 | } else { | |
7bb86316 CM |
2653 | if (path->nodes[*level] == root->node) { |
2654 | root_owner = root->root_key.objectid; | |
2655 | root_gen = | |
2656 | btrfs_header_generation(path->nodes[*level]); | |
2657 | } else { | |
2658 | struct extent_buffer *node; | |
2659 | node = path->nodes[*level + 1]; | |
2660 | root_owner = btrfs_header_owner(node); | |
2661 | root_gen = btrfs_header_generation(node); | |
2662 | } | |
e089f05c | 2663 | ret = btrfs_free_extent(trans, root, |
db94535d | 2664 | path->nodes[*level]->start, |
7bb86316 CM |
2665 | path->nodes[*level]->len, |
2666 | root_owner, root_gen, 0, 0, 1); | |
6407bf6d | 2667 | BUG_ON(ret); |
5f39d397 | 2668 | free_extent_buffer(path->nodes[*level]); |
83e15a28 | 2669 | path->nodes[*level] = NULL; |
20524f02 | 2670 | *level = i + 1; |
20524f02 CM |
2671 | } |
2672 | } | |
2673 | return 1; | |
2674 | } | |
2675 | ||
9aca1d51 CM |
2676 | /* |
2677 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2678 | * the tree freeing any blocks that have a ref count of zero after being | |
2679 | * decremented. | |
2680 | */ | |
e089f05c | 2681 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
9f3a7427 | 2682 | *root) |
20524f02 | 2683 | { |
3768f368 | 2684 | int ret = 0; |
9aca1d51 | 2685 | int wret; |
20524f02 | 2686 | int level; |
5caf2a00 | 2687 | struct btrfs_path *path; |
20524f02 CM |
2688 | int i; |
2689 | int orig_level; | |
9f3a7427 | 2690 | struct btrfs_root_item *root_item = &root->root_item; |
20524f02 | 2691 | |
a2135011 | 2692 | WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); |
5caf2a00 CM |
2693 | path = btrfs_alloc_path(); |
2694 | BUG_ON(!path); | |
20524f02 | 2695 | |
5f39d397 | 2696 | level = btrfs_header_level(root->node); |
20524f02 | 2697 | orig_level = level; |
9f3a7427 CM |
2698 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2699 | path->nodes[level] = root->node; | |
f510cfec | 2700 | extent_buffer_get(root->node); |
9f3a7427 CM |
2701 | path->slots[level] = 0; |
2702 | } else { | |
2703 | struct btrfs_key key; | |
5f39d397 CM |
2704 | struct btrfs_disk_key found_key; |
2705 | struct extent_buffer *node; | |
6702ed49 | 2706 | |
9f3a7427 | 2707 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
6702ed49 CM |
2708 | level = root_item->drop_level; |
2709 | path->lowest_level = level; | |
9f3a7427 | 2710 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6702ed49 | 2711 | if (wret < 0) { |
9f3a7427 CM |
2712 | ret = wret; |
2713 | goto out; | |
2714 | } | |
5f39d397 CM |
2715 | node = path->nodes[level]; |
2716 | btrfs_node_key(node, &found_key, path->slots[level]); | |
2717 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, | |
2718 | sizeof(found_key))); | |
7d9eb12c CM |
2719 | /* |
2720 | * unlock our path, this is safe because only this | |
2721 | * function is allowed to delete this snapshot | |
2722 | */ | |
925baedd CM |
2723 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
2724 | if (path->nodes[i] && path->locks[i]) { | |
2725 | path->locks[i] = 0; | |
2726 | btrfs_tree_unlock(path->nodes[i]); | |
2727 | } | |
2728 | } | |
9f3a7427 | 2729 | } |
20524f02 | 2730 | while(1) { |
5caf2a00 | 2731 | wret = walk_down_tree(trans, root, path, &level); |
9aca1d51 | 2732 | if (wret > 0) |
20524f02 | 2733 | break; |
9aca1d51 CM |
2734 | if (wret < 0) |
2735 | ret = wret; | |
2736 | ||
5caf2a00 | 2737 | wret = walk_up_tree(trans, root, path, &level); |
9aca1d51 | 2738 | if (wret > 0) |
20524f02 | 2739 | break; |
9aca1d51 CM |
2740 | if (wret < 0) |
2741 | ret = wret; | |
e7a84565 CM |
2742 | if (trans->transaction->in_commit) { |
2743 | ret = -EAGAIN; | |
2744 | break; | |
2745 | } | |
18e35e0a | 2746 | atomic_inc(&root->fs_info->throttle_gen); |
017e5369 | 2747 | wake_up(&root->fs_info->transaction_throttle); |
20524f02 | 2748 | } |
83e15a28 | 2749 | for (i = 0; i <= orig_level; i++) { |
5caf2a00 | 2750 | if (path->nodes[i]) { |
5f39d397 | 2751 | free_extent_buffer(path->nodes[i]); |
0f82731f | 2752 | path->nodes[i] = NULL; |
83e15a28 | 2753 | } |
20524f02 | 2754 | } |
9f3a7427 | 2755 | out: |
5caf2a00 | 2756 | btrfs_free_path(path); |
9aca1d51 | 2757 | return ret; |
20524f02 | 2758 | } |
9078a3e1 | 2759 | |
96b5179d | 2760 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9078a3e1 | 2761 | { |
96b5179d CM |
2762 | u64 start; |
2763 | u64 end; | |
b97f9203 | 2764 | u64 ptr; |
9078a3e1 | 2765 | int ret; |
925baedd CM |
2766 | |
2767 | mutex_lock(&info->alloc_mutex); | |
9078a3e1 | 2768 | while(1) { |
96b5179d CM |
2769 | ret = find_first_extent_bit(&info->block_group_cache, 0, |
2770 | &start, &end, (unsigned int)-1); | |
2771 | if (ret) | |
9078a3e1 | 2772 | break; |
b97f9203 Y |
2773 | ret = get_state_private(&info->block_group_cache, start, &ptr); |
2774 | if (!ret) | |
2775 | kfree((void *)(unsigned long)ptr); | |
96b5179d CM |
2776 | clear_extent_bits(&info->block_group_cache, start, |
2777 | end, (unsigned int)-1, GFP_NOFS); | |
9078a3e1 | 2778 | } |
e37c9e69 | 2779 | while(1) { |
f510cfec CM |
2780 | ret = find_first_extent_bit(&info->free_space_cache, 0, |
2781 | &start, &end, EXTENT_DIRTY); | |
2782 | if (ret) | |
e37c9e69 | 2783 | break; |
f510cfec CM |
2784 | clear_extent_dirty(&info->free_space_cache, start, |
2785 | end, GFP_NOFS); | |
e37c9e69 | 2786 | } |
925baedd | 2787 | mutex_unlock(&info->alloc_mutex); |
be744175 CM |
2788 | return 0; |
2789 | } | |
2790 | ||
8e7bf94f CM |
2791 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
2792 | unsigned long nr) | |
2793 | { | |
2794 | return min(last, start + nr - 1); | |
2795 | } | |
2796 | ||
98ed5174 CM |
2797 | static int noinline relocate_inode_pages(struct inode *inode, u64 start, |
2798 | u64 len) | |
edbd8d4e CM |
2799 | { |
2800 | u64 page_start; | |
2801 | u64 page_end; | |
edbd8d4e | 2802 | unsigned long last_index; |
edbd8d4e CM |
2803 | unsigned long i; |
2804 | struct page *page; | |
d1310b2e | 2805 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 2806 | struct file_ra_state *ra; |
8e7bf94f CM |
2807 | unsigned long total_read = 0; |
2808 | unsigned long ra_pages; | |
3eaa2885 | 2809 | struct btrfs_ordered_extent *ordered; |
a061fc8d | 2810 | struct btrfs_trans_handle *trans; |
4313b399 CM |
2811 | |
2812 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
2813 | |
2814 | mutex_lock(&inode->i_mutex); | |
4313b399 | 2815 | i = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
2816 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
2817 | ||
8e7bf94f CM |
2818 | ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages; |
2819 | ||
4313b399 | 2820 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 2821 | |
4313b399 | 2822 | for (; i <= last_index; i++) { |
8e7bf94f CM |
2823 | if (total_read % ra_pages == 0) { |
2824 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, | |
2825 | calc_ra(i, last_index, ra_pages)); | |
2826 | } | |
2827 | total_read++; | |
3eaa2885 CM |
2828 | again: |
2829 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
a061fc8d | 2830 | goto truncate_racing; |
edbd8d4e | 2831 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 2832 | if (!page) { |
edbd8d4e | 2833 | goto out_unlock; |
a061fc8d | 2834 | } |
edbd8d4e CM |
2835 | if (!PageUptodate(page)) { |
2836 | btrfs_readpage(NULL, page); | |
2837 | lock_page(page); | |
2838 | if (!PageUptodate(page)) { | |
2839 | unlock_page(page); | |
2840 | page_cache_release(page); | |
2841 | goto out_unlock; | |
2842 | } | |
2843 | } | |
ec44a35c | 2844 | wait_on_page_writeback(page); |
3eaa2885 | 2845 | |
edbd8d4e CM |
2846 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
2847 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
d1310b2e | 2848 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 2849 | |
3eaa2885 CM |
2850 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
2851 | if (ordered) { | |
2852 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2853 | unlock_page(page); | |
2854 | page_cache_release(page); | |
2855 | btrfs_start_ordered_extent(inode, ordered, 1); | |
2856 | btrfs_put_ordered_extent(ordered); | |
2857 | goto again; | |
2858 | } | |
2859 | set_page_extent_mapped(page); | |
2860 | ||
f87f057b CM |
2861 | /* |
2862 | * make sure page_mkwrite is called for this page if userland | |
2863 | * wants to change it from mmap | |
2864 | */ | |
2865 | clear_page_dirty_for_io(page); | |
3eaa2885 | 2866 | |
ea8c2819 | 2867 | btrfs_set_extent_delalloc(inode, page_start, page_end); |
a061fc8d | 2868 | set_page_dirty(page); |
edbd8d4e | 2869 | |
d1310b2e | 2870 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
2871 | unlock_page(page); |
2872 | page_cache_release(page); | |
2873 | } | |
2874 | ||
2875 | out_unlock: | |
3eaa2885 CM |
2876 | /* we have to start the IO in order to get the ordered extents |
2877 | * instantiated. This allows the relocation to code to wait | |
2878 | * for all the ordered extents to hit the disk. | |
2879 | * | |
2880 | * Otherwise, it would constantly loop over the same extents | |
2881 | * because the old ones don't get deleted until the IO is | |
2882 | * started | |
2883 | */ | |
2884 | btrfs_fdatawrite_range(inode->i_mapping, start, start + len - 1, | |
2885 | WB_SYNC_NONE); | |
ec44a35c | 2886 | kfree(ra); |
a061fc8d CM |
2887 | trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1); |
2888 | if (trans) { | |
a061fc8d CM |
2889 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); |
2890 | mark_inode_dirty(inode); | |
2891 | } | |
edbd8d4e CM |
2892 | mutex_unlock(&inode->i_mutex); |
2893 | return 0; | |
a061fc8d CM |
2894 | |
2895 | truncate_racing: | |
2896 | vmtruncate(inode, inode->i_size); | |
2897 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, | |
2898 | total_read); | |
2899 | goto out_unlock; | |
edbd8d4e CM |
2900 | } |
2901 | ||
bf4ef679 CM |
2902 | /* |
2903 | * The back references tell us which tree holds a ref on a block, | |
2904 | * but it is possible for the tree root field in the reference to | |
2905 | * reflect the original root before a snapshot was made. In this | |
2906 | * case we should search through all the children of a given root | |
2907 | * to find potential holders of references on a block. | |
2908 | * | |
2909 | * Instead, we do something a little less fancy and just search | |
2910 | * all the roots for a given key/block combination. | |
2911 | */ | |
2912 | static int find_root_for_ref(struct btrfs_root *root, | |
2913 | struct btrfs_path *path, | |
2914 | struct btrfs_key *key0, | |
2915 | int level, | |
2916 | int file_key, | |
2917 | struct btrfs_root **found_root, | |
2918 | u64 bytenr) | |
2919 | { | |
2920 | struct btrfs_key root_location; | |
2921 | struct btrfs_root *cur_root = *found_root; | |
2922 | struct btrfs_file_extent_item *file_extent; | |
2923 | u64 root_search_start = BTRFS_FS_TREE_OBJECTID; | |
2924 | u64 found_bytenr; | |
2925 | int ret; | |
bf4ef679 CM |
2926 | |
2927 | root_location.offset = (u64)-1; | |
2928 | root_location.type = BTRFS_ROOT_ITEM_KEY; | |
2929 | path->lowest_level = level; | |
2930 | path->reada = 0; | |
2931 | while(1) { | |
2932 | ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0); | |
2933 | found_bytenr = 0; | |
2934 | if (ret == 0 && file_key) { | |
2935 | struct extent_buffer *leaf = path->nodes[0]; | |
2936 | file_extent = btrfs_item_ptr(leaf, path->slots[0], | |
2937 | struct btrfs_file_extent_item); | |
2938 | if (btrfs_file_extent_type(leaf, file_extent) == | |
2939 | BTRFS_FILE_EXTENT_REG) { | |
2940 | found_bytenr = | |
2941 | btrfs_file_extent_disk_bytenr(leaf, | |
2942 | file_extent); | |
2943 | } | |
323da79c | 2944 | } else if (!file_key) { |
bf4ef679 CM |
2945 | if (path->nodes[level]) |
2946 | found_bytenr = path->nodes[level]->start; | |
2947 | } | |
2948 | ||
bf4ef679 CM |
2949 | btrfs_release_path(cur_root, path); |
2950 | ||
2951 | if (found_bytenr == bytenr) { | |
2952 | *found_root = cur_root; | |
2953 | ret = 0; | |
2954 | goto out; | |
2955 | } | |
2956 | ret = btrfs_search_root(root->fs_info->tree_root, | |
2957 | root_search_start, &root_search_start); | |
2958 | if (ret) | |
2959 | break; | |
2960 | ||
2961 | root_location.objectid = root_search_start; | |
2962 | cur_root = btrfs_read_fs_root_no_name(root->fs_info, | |
2963 | &root_location); | |
2964 | if (!cur_root) { | |
2965 | ret = 1; | |
2966 | break; | |
2967 | } | |
2968 | } | |
2969 | out: | |
2970 | path->lowest_level = 0; | |
2971 | return ret; | |
2972 | } | |
2973 | ||
4313b399 CM |
2974 | /* |
2975 | * note, this releases the path | |
2976 | */ | |
98ed5174 | 2977 | static int noinline relocate_one_reference(struct btrfs_root *extent_root, |
edbd8d4e | 2978 | struct btrfs_path *path, |
0ef3e66b CM |
2979 | struct btrfs_key *extent_key, |
2980 | u64 *last_file_objectid, | |
2981 | u64 *last_file_offset, | |
2982 | u64 *last_file_root, | |
2983 | u64 last_extent) | |
edbd8d4e CM |
2984 | { |
2985 | struct inode *inode; | |
2986 | struct btrfs_root *found_root; | |
bf4ef679 CM |
2987 | struct btrfs_key root_location; |
2988 | struct btrfs_key found_key; | |
4313b399 CM |
2989 | struct btrfs_extent_ref *ref; |
2990 | u64 ref_root; | |
2991 | u64 ref_gen; | |
2992 | u64 ref_objectid; | |
2993 | u64 ref_offset; | |
edbd8d4e | 2994 | int ret; |
bf4ef679 | 2995 | int level; |
edbd8d4e | 2996 | |
7d9eb12c CM |
2997 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); |
2998 | ||
4313b399 CM |
2999 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
3000 | struct btrfs_extent_ref); | |
3001 | ref_root = btrfs_ref_root(path->nodes[0], ref); | |
3002 | ref_gen = btrfs_ref_generation(path->nodes[0], ref); | |
3003 | ref_objectid = btrfs_ref_objectid(path->nodes[0], ref); | |
3004 | ref_offset = btrfs_ref_offset(path->nodes[0], ref); | |
3005 | btrfs_release_path(extent_root, path); | |
3006 | ||
bf4ef679 | 3007 | root_location.objectid = ref_root; |
edbd8d4e | 3008 | if (ref_gen == 0) |
bf4ef679 | 3009 | root_location.offset = 0; |
edbd8d4e | 3010 | else |
bf4ef679 CM |
3011 | root_location.offset = (u64)-1; |
3012 | root_location.type = BTRFS_ROOT_ITEM_KEY; | |
edbd8d4e CM |
3013 | |
3014 | found_root = btrfs_read_fs_root_no_name(extent_root->fs_info, | |
bf4ef679 | 3015 | &root_location); |
edbd8d4e | 3016 | BUG_ON(!found_root); |
7d9eb12c | 3017 | mutex_unlock(&extent_root->fs_info->alloc_mutex); |
edbd8d4e CM |
3018 | |
3019 | if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
bf4ef679 CM |
3020 | found_key.objectid = ref_objectid; |
3021 | found_key.type = BTRFS_EXTENT_DATA_KEY; | |
3022 | found_key.offset = ref_offset; | |
3023 | level = 0; | |
3024 | ||
0ef3e66b CM |
3025 | if (last_extent == extent_key->objectid && |
3026 | *last_file_objectid == ref_objectid && | |
3027 | *last_file_offset == ref_offset && | |
3028 | *last_file_root == ref_root) | |
3029 | goto out; | |
3030 | ||
bf4ef679 CM |
3031 | ret = find_root_for_ref(extent_root, path, &found_key, |
3032 | level, 1, &found_root, | |
3033 | extent_key->objectid); | |
3034 | ||
3035 | if (ret) | |
3036 | goto out; | |
3037 | ||
0ef3e66b CM |
3038 | if (last_extent == extent_key->objectid && |
3039 | *last_file_objectid == ref_objectid && | |
3040 | *last_file_offset == ref_offset && | |
3041 | *last_file_root == ref_root) | |
3042 | goto out; | |
3043 | ||
edbd8d4e CM |
3044 | inode = btrfs_iget_locked(extent_root->fs_info->sb, |
3045 | ref_objectid, found_root); | |
3046 | if (inode->i_state & I_NEW) { | |
3047 | /* the inode and parent dir are two different roots */ | |
3048 | BTRFS_I(inode)->root = found_root; | |
3049 | BTRFS_I(inode)->location.objectid = ref_objectid; | |
3050 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
3051 | BTRFS_I(inode)->location.offset = 0; | |
3052 | btrfs_read_locked_inode(inode); | |
3053 | unlock_new_inode(inode); | |
3054 | ||
3055 | } | |
3056 | /* this can happen if the reference is not against | |
3057 | * the latest version of the tree root | |
3058 | */ | |
7d9eb12c | 3059 | if (is_bad_inode(inode)) |
edbd8d4e | 3060 | goto out; |
7d9eb12c | 3061 | |
0ef3e66b CM |
3062 | *last_file_objectid = inode->i_ino; |
3063 | *last_file_root = found_root->root_key.objectid; | |
3064 | *last_file_offset = ref_offset; | |
3065 | ||
edbd8d4e | 3066 | relocate_inode_pages(inode, ref_offset, extent_key->offset); |
edbd8d4e | 3067 | iput(inode); |
edbd8d4e CM |
3068 | } else { |
3069 | struct btrfs_trans_handle *trans; | |
edbd8d4e | 3070 | struct extent_buffer *eb; |
7d9eb12c | 3071 | int needs_lock = 0; |
edbd8d4e | 3072 | |
edbd8d4e | 3073 | eb = read_tree_block(found_root, extent_key->objectid, |
ca7a79ad | 3074 | extent_key->offset, 0); |
925baedd | 3075 | btrfs_tree_lock(eb); |
edbd8d4e CM |
3076 | level = btrfs_header_level(eb); |
3077 | ||
3078 | if (level == 0) | |
3079 | btrfs_item_key_to_cpu(eb, &found_key, 0); | |
3080 | else | |
3081 | btrfs_node_key_to_cpu(eb, &found_key, 0); | |
3082 | ||
925baedd | 3083 | btrfs_tree_unlock(eb); |
edbd8d4e CM |
3084 | free_extent_buffer(eb); |
3085 | ||
bf4ef679 CM |
3086 | ret = find_root_for_ref(extent_root, path, &found_key, |
3087 | level, 0, &found_root, | |
3088 | extent_key->objectid); | |
3089 | ||
3090 | if (ret) | |
3091 | goto out; | |
3092 | ||
7d9eb12c CM |
3093 | /* |
3094 | * right here almost anything could happen to our key, | |
3095 | * but that's ok. The cow below will either relocate it | |
3096 | * or someone else will have relocated it. Either way, | |
3097 | * it is in a different spot than it was before and | |
3098 | * we're happy. | |
3099 | */ | |
3100 | ||
bf4ef679 CM |
3101 | trans = btrfs_start_transaction(found_root, 1); |
3102 | ||
7d9eb12c CM |
3103 | if (found_root == extent_root->fs_info->extent_root || |
3104 | found_root == extent_root->fs_info->chunk_root || | |
3105 | found_root == extent_root->fs_info->dev_root) { | |
3106 | needs_lock = 1; | |
3107 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
3108 | } | |
3109 | ||
edbd8d4e | 3110 | path->lowest_level = level; |
8f662a76 | 3111 | path->reada = 2; |
edbd8d4e CM |
3112 | ret = btrfs_search_slot(trans, found_root, &found_key, path, |
3113 | 0, 1); | |
3114 | path->lowest_level = 0; | |
edbd8d4e | 3115 | btrfs_release_path(found_root, path); |
7d9eb12c | 3116 | |
0ef3e66b CM |
3117 | if (found_root == found_root->fs_info->extent_root) |
3118 | btrfs_extent_post_op(trans, found_root); | |
7d9eb12c CM |
3119 | if (needs_lock) |
3120 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
3121 | ||
edbd8d4e | 3122 | btrfs_end_transaction(trans, found_root); |
edbd8d4e | 3123 | |
7d9eb12c | 3124 | } |
edbd8d4e | 3125 | out: |
7d9eb12c | 3126 | mutex_lock(&extent_root->fs_info->alloc_mutex); |
edbd8d4e CM |
3127 | return 0; |
3128 | } | |
3129 | ||
a061fc8d CM |
3130 | static int noinline del_extent_zero(struct btrfs_root *extent_root, |
3131 | struct btrfs_path *path, | |
3132 | struct btrfs_key *extent_key) | |
3133 | { | |
3134 | int ret; | |
3135 | struct btrfs_trans_handle *trans; | |
3136 | ||
3137 | trans = btrfs_start_transaction(extent_root, 1); | |
3138 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); | |
3139 | if (ret > 0) { | |
3140 | ret = -EIO; | |
3141 | goto out; | |
3142 | } | |
3143 | if (ret < 0) | |
3144 | goto out; | |
3145 | ret = btrfs_del_item(trans, extent_root, path); | |
3146 | out: | |
3147 | btrfs_end_transaction(trans, extent_root); | |
3148 | return ret; | |
3149 | } | |
3150 | ||
98ed5174 CM |
3151 | static int noinline relocate_one_extent(struct btrfs_root *extent_root, |
3152 | struct btrfs_path *path, | |
3153 | struct btrfs_key *extent_key) | |
edbd8d4e CM |
3154 | { |
3155 | struct btrfs_key key; | |
3156 | struct btrfs_key found_key; | |
edbd8d4e | 3157 | struct extent_buffer *leaf; |
0ef3e66b CM |
3158 | u64 last_file_objectid = 0; |
3159 | u64 last_file_root = 0; | |
3160 | u64 last_file_offset = (u64)-1; | |
3161 | u64 last_extent = 0; | |
edbd8d4e CM |
3162 | u32 nritems; |
3163 | u32 item_size; | |
3164 | int ret = 0; | |
3165 | ||
a061fc8d CM |
3166 | if (extent_key->objectid == 0) { |
3167 | ret = del_extent_zero(extent_root, path, extent_key); | |
3168 | goto out; | |
3169 | } | |
edbd8d4e CM |
3170 | key.objectid = extent_key->objectid; |
3171 | key.type = BTRFS_EXTENT_REF_KEY; | |
3172 | key.offset = 0; | |
3173 | ||
3174 | while(1) { | |
3175 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
3176 | ||
edbd8d4e CM |
3177 | if (ret < 0) |
3178 | goto out; | |
3179 | ||
3180 | ret = 0; | |
3181 | leaf = path->nodes[0]; | |
3182 | nritems = btrfs_header_nritems(leaf); | |
a061fc8d CM |
3183 | if (path->slots[0] == nritems) { |
3184 | ret = btrfs_next_leaf(extent_root, path); | |
3185 | if (ret > 0) { | |
3186 | ret = 0; | |
3187 | goto out; | |
3188 | } | |
3189 | if (ret < 0) | |
3190 | goto out; | |
bf4ef679 | 3191 | leaf = path->nodes[0]; |
a061fc8d | 3192 | } |
edbd8d4e CM |
3193 | |
3194 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
a061fc8d | 3195 | if (found_key.objectid != extent_key->objectid) { |
edbd8d4e | 3196 | break; |
a061fc8d | 3197 | } |
edbd8d4e | 3198 | |
a061fc8d | 3199 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
edbd8d4e | 3200 | break; |
a061fc8d | 3201 | } |
edbd8d4e CM |
3202 | |
3203 | key.offset = found_key.offset + 1; | |
3204 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3205 | ||
0ef3e66b CM |
3206 | ret = relocate_one_reference(extent_root, path, extent_key, |
3207 | &last_file_objectid, | |
3208 | &last_file_offset, | |
3209 | &last_file_root, last_extent); | |
edbd8d4e CM |
3210 | if (ret) |
3211 | goto out; | |
0ef3e66b | 3212 | last_extent = extent_key->objectid; |
edbd8d4e CM |
3213 | } |
3214 | ret = 0; | |
3215 | out: | |
3216 | btrfs_release_path(extent_root, path); | |
3217 | return ret; | |
3218 | } | |
3219 | ||
ec44a35c CM |
3220 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
3221 | { | |
3222 | u64 num_devices; | |
3223 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
3224 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
3225 | ||
a061fc8d | 3226 | num_devices = root->fs_info->fs_devices->num_devices; |
ec44a35c CM |
3227 | if (num_devices == 1) { |
3228 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
3229 | stripped = flags & ~stripped; | |
3230 | ||
3231 | /* turn raid0 into single device chunks */ | |
3232 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
3233 | return stripped; | |
3234 | ||
3235 | /* turn mirroring into duplication */ | |
3236 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
3237 | BTRFS_BLOCK_GROUP_RAID10)) | |
3238 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
3239 | return flags; | |
3240 | } else { | |
3241 | /* they already had raid on here, just return */ | |
ec44a35c CM |
3242 | if (flags & stripped) |
3243 | return flags; | |
3244 | ||
3245 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
3246 | stripped = flags & ~stripped; | |
3247 | ||
3248 | /* switch duplicated blocks with raid1 */ | |
3249 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
3250 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
3251 | ||
3252 | /* turn single device chunks into raid0 */ | |
3253 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
3254 | } | |
3255 | return flags; | |
3256 | } | |
3257 | ||
0ef3e66b CM |
3258 | int __alloc_chunk_for_shrink(struct btrfs_root *root, |
3259 | struct btrfs_block_group_cache *shrink_block_group, | |
3260 | int force) | |
3261 | { | |
3262 | struct btrfs_trans_handle *trans; | |
3263 | u64 new_alloc_flags; | |
3264 | u64 calc; | |
3265 | ||
c286ac48 | 3266 | spin_lock(&shrink_block_group->lock); |
0ef3e66b | 3267 | if (btrfs_block_group_used(&shrink_block_group->item) > 0) { |
c286ac48 | 3268 | spin_unlock(&shrink_block_group->lock); |
7d9eb12c | 3269 | mutex_unlock(&root->fs_info->alloc_mutex); |
c286ac48 | 3270 | |
0ef3e66b | 3271 | trans = btrfs_start_transaction(root, 1); |
7d9eb12c | 3272 | mutex_lock(&root->fs_info->alloc_mutex); |
c286ac48 | 3273 | spin_lock(&shrink_block_group->lock); |
7d9eb12c | 3274 | |
0ef3e66b CM |
3275 | new_alloc_flags = update_block_group_flags(root, |
3276 | shrink_block_group->flags); | |
3277 | if (new_alloc_flags != shrink_block_group->flags) { | |
3278 | calc = | |
3279 | btrfs_block_group_used(&shrink_block_group->item); | |
3280 | } else { | |
3281 | calc = shrink_block_group->key.offset; | |
3282 | } | |
c286ac48 CM |
3283 | spin_unlock(&shrink_block_group->lock); |
3284 | ||
0ef3e66b CM |
3285 | do_chunk_alloc(trans, root->fs_info->extent_root, |
3286 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
7d9eb12c CM |
3287 | |
3288 | mutex_unlock(&root->fs_info->alloc_mutex); | |
0ef3e66b | 3289 | btrfs_end_transaction(trans, root); |
7d9eb12c | 3290 | mutex_lock(&root->fs_info->alloc_mutex); |
c286ac48 CM |
3291 | } else |
3292 | spin_unlock(&shrink_block_group->lock); | |
0ef3e66b CM |
3293 | return 0; |
3294 | } | |
3295 | ||
8f18cf13 | 3296 | int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start) |
edbd8d4e CM |
3297 | { |
3298 | struct btrfs_trans_handle *trans; | |
3299 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
3300 | struct btrfs_path *path; | |
3301 | u64 cur_byte; | |
3302 | u64 total_found; | |
8f18cf13 CM |
3303 | u64 shrink_last_byte; |
3304 | struct btrfs_block_group_cache *shrink_block_group; | |
edbd8d4e | 3305 | struct btrfs_fs_info *info = root->fs_info; |
edbd8d4e | 3306 | struct btrfs_key key; |
73e48b27 | 3307 | struct btrfs_key found_key; |
edbd8d4e CM |
3308 | struct extent_buffer *leaf; |
3309 | u32 nritems; | |
3310 | int ret; | |
a061fc8d | 3311 | int progress; |
edbd8d4e | 3312 | |
925baedd | 3313 | mutex_lock(&root->fs_info->alloc_mutex); |
8f18cf13 CM |
3314 | shrink_block_group = btrfs_lookup_block_group(root->fs_info, |
3315 | shrink_start); | |
3316 | BUG_ON(!shrink_block_group); | |
3317 | ||
0ef3e66b CM |
3318 | shrink_last_byte = shrink_block_group->key.objectid + |
3319 | shrink_block_group->key.offset; | |
8f18cf13 CM |
3320 | |
3321 | shrink_block_group->space_info->total_bytes -= | |
3322 | shrink_block_group->key.offset; | |
edbd8d4e CM |
3323 | path = btrfs_alloc_path(); |
3324 | root = root->fs_info->extent_root; | |
8f662a76 | 3325 | path->reada = 2; |
edbd8d4e | 3326 | |
323da79c CM |
3327 | printk("btrfs relocating block group %llu flags %llu\n", |
3328 | (unsigned long long)shrink_start, | |
3329 | (unsigned long long)shrink_block_group->flags); | |
3330 | ||
0ef3e66b CM |
3331 | __alloc_chunk_for_shrink(root, shrink_block_group, 1); |
3332 | ||
edbd8d4e | 3333 | again: |
323da79c | 3334 | |
8f18cf13 CM |
3335 | shrink_block_group->ro = 1; |
3336 | ||
edbd8d4e | 3337 | total_found = 0; |
a061fc8d | 3338 | progress = 0; |
8f18cf13 | 3339 | key.objectid = shrink_start; |
edbd8d4e CM |
3340 | key.offset = 0; |
3341 | key.type = 0; | |
73e48b27 | 3342 | cur_byte = key.objectid; |
4313b399 | 3343 | |
ea8c2819 CM |
3344 | mutex_unlock(&root->fs_info->alloc_mutex); |
3345 | ||
3346 | btrfs_start_delalloc_inodes(root); | |
7ea394f1 | 3347 | btrfs_wait_ordered_extents(tree_root, 0); |
ea8c2819 CM |
3348 | |
3349 | mutex_lock(&root->fs_info->alloc_mutex); | |
3350 | ||
73e48b27 Y |
3351 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3352 | if (ret < 0) | |
3353 | goto out; | |
3354 | ||
0b86a832 | 3355 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
73e48b27 Y |
3356 | if (ret < 0) |
3357 | goto out; | |
8f18cf13 | 3358 | |
73e48b27 Y |
3359 | if (ret == 0) { |
3360 | leaf = path->nodes[0]; | |
3361 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 CM |
3362 | if (found_key.objectid + found_key.offset > shrink_start && |
3363 | found_key.objectid < shrink_last_byte) { | |
73e48b27 Y |
3364 | cur_byte = found_key.objectid; |
3365 | key.objectid = cur_byte; | |
3366 | } | |
3367 | } | |
3368 | btrfs_release_path(root, path); | |
3369 | ||
3370 | while(1) { | |
edbd8d4e CM |
3371 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3372 | if (ret < 0) | |
3373 | goto out; | |
73e48b27 | 3374 | |
7d9eb12c | 3375 | next: |
edbd8d4e | 3376 | leaf = path->nodes[0]; |
73e48b27 | 3377 | nritems = btrfs_header_nritems(leaf); |
73e48b27 Y |
3378 | if (path->slots[0] >= nritems) { |
3379 | ret = btrfs_next_leaf(root, path); | |
3380 | if (ret < 0) | |
3381 | goto out; | |
3382 | if (ret == 1) { | |
3383 | ret = 0; | |
3384 | break; | |
edbd8d4e | 3385 | } |
73e48b27 Y |
3386 | leaf = path->nodes[0]; |
3387 | nritems = btrfs_header_nritems(leaf); | |
edbd8d4e | 3388 | } |
73e48b27 Y |
3389 | |
3390 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
725c8463 | 3391 | |
8f18cf13 CM |
3392 | if (found_key.objectid >= shrink_last_byte) |
3393 | break; | |
3394 | ||
725c8463 CM |
3395 | if (progress && need_resched()) { |
3396 | memcpy(&key, &found_key, sizeof(key)); | |
725c8463 | 3397 | cond_resched(); |
725c8463 CM |
3398 | btrfs_release_path(root, path); |
3399 | btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3400 | progress = 0; | |
3401 | goto next; | |
3402 | } | |
3403 | progress = 1; | |
3404 | ||
73e48b27 Y |
3405 | if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY || |
3406 | found_key.objectid + found_key.offset <= cur_byte) { | |
0ef3e66b CM |
3407 | memcpy(&key, &found_key, sizeof(key)); |
3408 | key.offset++; | |
edbd8d4e | 3409 | path->slots[0]++; |
edbd8d4e CM |
3410 | goto next; |
3411 | } | |
73e48b27 | 3412 | |
edbd8d4e CM |
3413 | total_found++; |
3414 | cur_byte = found_key.objectid + found_key.offset; | |
3415 | key.objectid = cur_byte; | |
3416 | btrfs_release_path(root, path); | |
3417 | ret = relocate_one_extent(root, path, &found_key); | |
0ef3e66b | 3418 | __alloc_chunk_for_shrink(root, shrink_block_group, 0); |
edbd8d4e CM |
3419 | } |
3420 | ||
3421 | btrfs_release_path(root, path); | |
3422 | ||
3423 | if (total_found > 0) { | |
323da79c CM |
3424 | printk("btrfs relocate found %llu last extent was %llu\n", |
3425 | (unsigned long long)total_found, | |
3426 | (unsigned long long)found_key.objectid); | |
7d9eb12c | 3427 | mutex_unlock(&root->fs_info->alloc_mutex); |
edbd8d4e CM |
3428 | trans = btrfs_start_transaction(tree_root, 1); |
3429 | btrfs_commit_transaction(trans, tree_root); | |
3430 | ||
edbd8d4e | 3431 | btrfs_clean_old_snapshots(tree_root); |
edbd8d4e | 3432 | |
ea8c2819 | 3433 | btrfs_start_delalloc_inodes(root); |
7ea394f1 | 3434 | btrfs_wait_ordered_extents(tree_root, 0); |
3eaa2885 | 3435 | |
edbd8d4e CM |
3436 | trans = btrfs_start_transaction(tree_root, 1); |
3437 | btrfs_commit_transaction(trans, tree_root); | |
7d9eb12c | 3438 | mutex_lock(&root->fs_info->alloc_mutex); |
edbd8d4e CM |
3439 | goto again; |
3440 | } | |
3441 | ||
8f18cf13 CM |
3442 | /* |
3443 | * we've freed all the extents, now remove the block | |
3444 | * group item from the tree | |
3445 | */ | |
7d9eb12c CM |
3446 | mutex_unlock(&root->fs_info->alloc_mutex); |
3447 | ||
edbd8d4e | 3448 | trans = btrfs_start_transaction(root, 1); |
c286ac48 | 3449 | |
7d9eb12c | 3450 | mutex_lock(&root->fs_info->alloc_mutex); |
8f18cf13 | 3451 | memcpy(&key, &shrink_block_group->key, sizeof(key)); |
1372f8e6 | 3452 | |
8f18cf13 CM |
3453 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
3454 | if (ret > 0) | |
3455 | ret = -EIO; | |
8e8a1e31 JB |
3456 | if (ret < 0) { |
3457 | btrfs_end_transaction(trans, root); | |
8f18cf13 | 3458 | goto out; |
8e8a1e31 | 3459 | } |
73e48b27 | 3460 | |
0ef3e66b CM |
3461 | clear_extent_bits(&info->block_group_cache, key.objectid, |
3462 | key.objectid + key.offset - 1, | |
8f18cf13 | 3463 | (unsigned int)-1, GFP_NOFS); |
edbd8d4e | 3464 | |
0ef3e66b CM |
3465 | |
3466 | clear_extent_bits(&info->free_space_cache, | |
3467 | key.objectid, key.objectid + key.offset - 1, | |
3468 | (unsigned int)-1, GFP_NOFS); | |
3469 | ||
d7a029a8 | 3470 | /* |
0ef3e66b CM |
3471 | memset(shrink_block_group, 0, sizeof(*shrink_block_group)); |
3472 | kfree(shrink_block_group); | |
d7a029a8 | 3473 | */ |
0ef3e66b | 3474 | |
8f18cf13 | 3475 | btrfs_del_item(trans, root, path); |
7d9eb12c CM |
3476 | btrfs_release_path(root, path); |
3477 | mutex_unlock(&root->fs_info->alloc_mutex); | |
edbd8d4e | 3478 | btrfs_commit_transaction(trans, root); |
0ef3e66b | 3479 | |
7d9eb12c CM |
3480 | mutex_lock(&root->fs_info->alloc_mutex); |
3481 | ||
0ef3e66b CM |
3482 | /* the code to unpin extents might set a few bits in the free |
3483 | * space cache for this range again | |
3484 | */ | |
3485 | clear_extent_bits(&info->free_space_cache, | |
3486 | key.objectid, key.objectid + key.offset - 1, | |
3487 | (unsigned int)-1, GFP_NOFS); | |
edbd8d4e CM |
3488 | out: |
3489 | btrfs_free_path(path); | |
925baedd | 3490 | mutex_unlock(&root->fs_info->alloc_mutex); |
edbd8d4e CM |
3491 | return ret; |
3492 | } | |
3493 | ||
0b86a832 CM |
3494 | int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, |
3495 | struct btrfs_key *key) | |
3496 | { | |
925baedd | 3497 | int ret = 0; |
0b86a832 CM |
3498 | struct btrfs_key found_key; |
3499 | struct extent_buffer *leaf; | |
3500 | int slot; | |
edbd8d4e | 3501 | |
0b86a832 CM |
3502 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
3503 | if (ret < 0) | |
925baedd CM |
3504 | goto out; |
3505 | ||
0b86a832 CM |
3506 | while(1) { |
3507 | slot = path->slots[0]; | |
edbd8d4e | 3508 | leaf = path->nodes[0]; |
0b86a832 CM |
3509 | if (slot >= btrfs_header_nritems(leaf)) { |
3510 | ret = btrfs_next_leaf(root, path); | |
3511 | if (ret == 0) | |
3512 | continue; | |
3513 | if (ret < 0) | |
925baedd | 3514 | goto out; |
0b86a832 | 3515 | break; |
edbd8d4e | 3516 | } |
0b86a832 | 3517 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 3518 | |
0b86a832 | 3519 | if (found_key.objectid >= key->objectid && |
925baedd CM |
3520 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
3521 | ret = 0; | |
3522 | goto out; | |
3523 | } | |
0b86a832 | 3524 | path->slots[0]++; |
edbd8d4e | 3525 | } |
0b86a832 | 3526 | ret = -ENOENT; |
925baedd | 3527 | out: |
0b86a832 | 3528 | return ret; |
edbd8d4e CM |
3529 | } |
3530 | ||
9078a3e1 CM |
3531 | int btrfs_read_block_groups(struct btrfs_root *root) |
3532 | { | |
3533 | struct btrfs_path *path; | |
3534 | int ret; | |
96b5179d | 3535 | int bit; |
9078a3e1 | 3536 | struct btrfs_block_group_cache *cache; |
be744175 | 3537 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 3538 | struct btrfs_space_info *space_info; |
d1310b2e | 3539 | struct extent_io_tree *block_group_cache; |
9078a3e1 CM |
3540 | struct btrfs_key key; |
3541 | struct btrfs_key found_key; | |
5f39d397 | 3542 | struct extent_buffer *leaf; |
96b5179d CM |
3543 | |
3544 | block_group_cache = &info->block_group_cache; | |
be744175 | 3545 | root = info->extent_root; |
9078a3e1 | 3546 | key.objectid = 0; |
0b86a832 | 3547 | key.offset = 0; |
9078a3e1 | 3548 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
3549 | path = btrfs_alloc_path(); |
3550 | if (!path) | |
3551 | return -ENOMEM; | |
3552 | ||
925baedd | 3553 | mutex_lock(&root->fs_info->alloc_mutex); |
9078a3e1 | 3554 | while(1) { |
0b86a832 CM |
3555 | ret = find_first_block_group(root, path, &key); |
3556 | if (ret > 0) { | |
3557 | ret = 0; | |
3558 | goto error; | |
9078a3e1 | 3559 | } |
0b86a832 CM |
3560 | if (ret != 0) |
3561 | goto error; | |
3562 | ||
5f39d397 CM |
3563 | leaf = path->nodes[0]; |
3564 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 3565 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 3566 | if (!cache) { |
0b86a832 | 3567 | ret = -ENOMEM; |
9078a3e1 CM |
3568 | break; |
3569 | } | |
3e1ad54f | 3570 | |
c286ac48 | 3571 | spin_lock_init(&cache->lock); |
5f39d397 CM |
3572 | read_extent_buffer(leaf, &cache->item, |
3573 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3574 | sizeof(cache->item)); | |
9078a3e1 | 3575 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 3576 | |
9078a3e1 CM |
3577 | key.objectid = found_key.objectid + found_key.offset; |
3578 | btrfs_release_path(root, path); | |
0b86a832 CM |
3579 | cache->flags = btrfs_block_group_flags(&cache->item); |
3580 | bit = 0; | |
3581 | if (cache->flags & BTRFS_BLOCK_GROUP_DATA) { | |
96b5179d | 3582 | bit = BLOCK_GROUP_DATA; |
0b86a832 CM |
3583 | } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
3584 | bit = BLOCK_GROUP_SYSTEM; | |
3585 | } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
96b5179d | 3586 | bit = BLOCK_GROUP_METADATA; |
31f3c99b | 3587 | } |
8790d502 | 3588 | set_avail_alloc_bits(info, cache->flags); |
96b5179d | 3589 | |
6324fbf3 CM |
3590 | ret = update_space_info(info, cache->flags, found_key.offset, |
3591 | btrfs_block_group_used(&cache->item), | |
3592 | &space_info); | |
3593 | BUG_ON(ret); | |
3594 | cache->space_info = space_info; | |
3595 | ||
96b5179d CM |
3596 | /* use EXTENT_LOCKED to prevent merging */ |
3597 | set_extent_bits(block_group_cache, found_key.objectid, | |
3598 | found_key.objectid + found_key.offset - 1, | |
c286ac48 | 3599 | EXTENT_LOCKED, GFP_NOFS); |
96b5179d | 3600 | set_state_private(block_group_cache, found_key.objectid, |
ae2f5411 | 3601 | (unsigned long)cache); |
c286ac48 CM |
3602 | set_extent_bits(block_group_cache, found_key.objectid, |
3603 | found_key.objectid + found_key.offset - 1, | |
3604 | bit | EXTENT_LOCKED, GFP_NOFS); | |
9078a3e1 | 3605 | if (key.objectid >= |
db94535d | 3606 | btrfs_super_total_bytes(&info->super_copy)) |
9078a3e1 CM |
3607 | break; |
3608 | } | |
0b86a832 CM |
3609 | ret = 0; |
3610 | error: | |
9078a3e1 | 3611 | btrfs_free_path(path); |
925baedd | 3612 | mutex_unlock(&root->fs_info->alloc_mutex); |
0b86a832 | 3613 | return ret; |
9078a3e1 | 3614 | } |
6324fbf3 CM |
3615 | |
3616 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
3617 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 3618 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
3619 | u64 size) |
3620 | { | |
3621 | int ret; | |
3622 | int bit = 0; | |
3623 | struct btrfs_root *extent_root; | |
3624 | struct btrfs_block_group_cache *cache; | |
3625 | struct extent_io_tree *block_group_cache; | |
3626 | ||
7d9eb12c | 3627 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
6324fbf3 CM |
3628 | extent_root = root->fs_info->extent_root; |
3629 | block_group_cache = &root->fs_info->block_group_cache; | |
3630 | ||
8f18cf13 | 3631 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
6324fbf3 | 3632 | BUG_ON(!cache); |
e17cade2 | 3633 | cache->key.objectid = chunk_offset; |
6324fbf3 | 3634 | cache->key.offset = size; |
c286ac48 | 3635 | spin_lock_init(&cache->lock); |
6324fbf3 | 3636 | btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
0ef3e66b | 3637 | |
6324fbf3 | 3638 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
3639 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
3640 | cache->flags = type; | |
3641 | btrfs_set_block_group_flags(&cache->item, type); | |
3642 | ||
3643 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, | |
3644 | &cache->space_info); | |
3645 | BUG_ON(ret); | |
3646 | ||
d18a2c44 | 3647 | bit = block_group_state_bits(type); |
e17cade2 CM |
3648 | set_extent_bits(block_group_cache, chunk_offset, |
3649 | chunk_offset + size - 1, | |
c286ac48 | 3650 | EXTENT_LOCKED, GFP_NOFS); |
e17cade2 CM |
3651 | set_state_private(block_group_cache, chunk_offset, |
3652 | (unsigned long)cache); | |
c286ac48 CM |
3653 | set_extent_bits(block_group_cache, chunk_offset, |
3654 | chunk_offset + size - 1, | |
3655 | bit | EXTENT_LOCKED, GFP_NOFS); | |
3656 | ||
6324fbf3 CM |
3657 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
3658 | sizeof(cache->item)); | |
3659 | BUG_ON(ret); | |
3660 | ||
3661 | finish_current_insert(trans, extent_root); | |
3662 | ret = del_pending_extents(trans, extent_root); | |
3663 | BUG_ON(ret); | |
d18a2c44 | 3664 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 3665 | |
6324fbf3 CM |
3666 | return 0; |
3667 | } |